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                 pr_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         pr_info("%-15s %08x %08x\n", rname, regs[0], regs[1]);
 173 }
 174
 175 /*
 176  * e1000e_dump - Print registers, Tx-ring and Rx-ring
 177  */
 178 static void e1000e_dump(struct e1000_adapter *adapter)
 179 {
 180         struct net_device *netdev = adapter->netdev;
 181         struct e1000_hw *hw = &adapter->hw;
 182         struct e1000_reg_info *reginfo;
 183         struct e1000_ring *tx_ring = adapter->tx_ring;
 184         struct e1000_tx_desc *tx_desc;
 185         struct my_u0 {
 186                 u64 a;
 187                 u64 b;
 188         } *u0;
 189         struct e1000_buffer *buffer_info;
 190         struct e1000_ring *rx_ring = adapter->rx_ring;
 191         union e1000_rx_desc_packet_split *rx_desc_ps;
 192         union e1000_rx_desc_extended *rx_desc;
 193         struct my_u1 {
 194                 u64 a;
 195                 u64 b;
 196                 u64 c;
 197                 u64 d;
 198         } *u1;
 199         u32 staterr;
 200         int i = 0;
 201
 202         if (!netif_msg_hw(adapter))
 203                 return;
 204
 205         /* Print netdevice Info */
 206         if (netdev) {
 207                 dev_info(&adapter->pdev->dev, "Net device Info\n");
 208                 pr_info("Device Name     state            trans_start      last_rx\n");
 209                 pr_info("%-15s %016lX %016lX %016lX\n",
 210                         netdev->name, netdev->state, netdev->trans_start,
 211                         netdev->last_rx);
 212         }
 213
 214         /* Print Registers */
 215         dev_info(&adapter->pdev->dev, "Register Dump\n");
 216         pr_info(" Register Name   Value\n");
 217         for (reginfo = (struct e1000_reg_info *)e1000_reg_info_tbl;
 218              reginfo->name; reginfo++) {
 219                 e1000_regdump(hw, reginfo);
 220         }
 221
 222         /* Print Tx Ring Summary */
 223         if (!netdev || !netif_running(netdev))
 224                 goto exit;
 225
 226         dev_info(&adapter->pdev->dev, "Tx Ring Summary\n");
 227         pr_info("Queue [NTU] [NTC] [bi(ntc)->dma  ] leng ntw timestamp\n");
 228         buffer_info = &tx_ring->buffer_info[tx_ring->next_to_clean];
 229         pr_info(" %5d %5X %5X %016llX %04X %3X %016llX\n",
 230                 0, tx_ring->next_to_use, tx_ring->next_to_clean,
 231                 (unsigned long long)buffer_info->dma,
 232                 buffer_info->length,
 233                 buffer_info->next_to_watch,
 234                 (unsigned long long)buffer_info->time_stamp);
 235
 236         /* Print Tx Ring */
 237         if (!netif_msg_tx_done(adapter))
 238                 goto rx_ring_summary;
 239
 240         dev_info(&adapter->pdev->dev, "Tx Ring Dump\n");
 241
 242         /* Transmit Descriptor Formats - DEXT[29] is 0 (Legacy) or 1 (Extended)
 243          *
 244          * Legacy Transmit Descriptor
 245          *   +--------------------------------------------------------------+
 246          * 0 |         Buffer Address [63:0] (Reserved on Write Back)       |
 247          *   +--------------------------------------------------------------+
 248          * 8 | Special  |    CSS     | Status |  CMD    |  CSO   |  Length  |
 249          *   +--------------------------------------------------------------+
 250          *   63       48 47        36 35    32 31     24 23    16 15        0
 251          *
 252          * Extended Context Descriptor (DTYP=0x0) for TSO or checksum offload
 253          *   63      48 47    40 39       32 31             16 15    8 7      0
 254          *   +----------------------------------------------------------------+
 255          * 0 |  TUCSE  | TUCS0  |   TUCSS   |     IPCSE       | IPCS0 | IPCSS |
 256          *   +----------------------------------------------------------------+
 257          * 8 |   MSS   | HDRLEN | RSV | STA | TUCMD | DTYP |      PAYLEN      |
 258          *   +----------------------------------------------------------------+
 259          *   63      48 47    40 39 36 35 32 31   24 23  20 19                0
 260          *
 261          * Extended Data Descriptor (DTYP=0x1)
 262          *   +----------------------------------------------------------------+
 263          * 0 |                     Buffer Address [63:0]                      |
 264          *   +----------------------------------------------------------------+
 265          * 8 | VLAN tag |  POPTS  | Rsvd | Status | Command | DTYP |  DTALEN  |
 266          *   +----------------------------------------------------------------+
 267          *   63       48 47     40 39  36 35    32 31     24 23  20 19        0
 268          */
 269         pr_info("Tl[desc]     [address 63:0  ] [SpeCssSCmCsLen] [bi->dma       ] leng  ntw timestamp        bi->skb <-- Legacy format\n");
 270         pr_info("Tc[desc]     [Ce CoCsIpceCoS] [MssHlRSCm0Plen] [bi->dma       ] leng  ntw timestamp        bi->skb <-- Ext Context format\n");
 271         pr_info("Td[desc]     [address 63:0  ] [VlaPoRSCm1Dlen] [bi->dma       ] leng  ntw timestamp        bi->skb <-- Ext Data format\n");
 272         for (i = 0; tx_ring->desc && (i < tx_ring->count); i++) {
 273                 const char *next_desc;
 274                 tx_desc = E1000_TX_DESC(*tx_ring, i);
 275                 buffer_info = &tx_ring->buffer_info[i];
 276                 u0 = (struct my_u0 *)tx_desc;
 277                 if (i == tx_ring->next_to_use && i == tx_ring->next_to_clean)
 278                         next_desc = " NTC/U";
 279                 else if (i == tx_ring->next_to_use)
 280                         next_desc = " NTU";
 281                 else if (i == tx_ring->next_to_clean)
 282                         next_desc = " NTC";
 283                 else
 284                         next_desc = "";
 285                 pr_info("T%c[0x%03X]    %016llX %016llX %016llX %04X  %3X %016llX %p%s\n",
 286                         (!(le64_to_cpu(u0->b) & (1 << 29)) ? 'l' :
 287                          ((le64_to_cpu(u0->b) & (1 << 20)) ? 'd' : 'c')),
 288                         i,
 289                         (unsigned long long)le64_to_cpu(u0->a),
 290                         (unsigned long long)le64_to_cpu(u0->b),
 291                         (unsigned long long)buffer_info->dma,
 292                         buffer_info->length, buffer_info->next_to_watch,
 293                         (unsigned long long)buffer_info->time_stamp,
 294                         buffer_info->skb, next_desc);
 295
 296                 if (netif_msg_pktdata(adapter) && buffer_info->dma != 0)
 297                         print_hex_dump(KERN_INFO, "", DUMP_PREFIX_ADDRESS,
 298                                        16, 1, phys_to_virt(buffer_info->dma),
 299                                        buffer_info->length, true);
 300         }
 301
 302         /* Print Rx Ring Summary */
 303 rx_ring_summary:
 304         dev_info(&adapter->pdev->dev, "Rx Ring Summary\n");
 305         pr_info("Queue [NTU] [NTC]\n");
 306         pr_info(" %5d %5X %5X\n",
 307                 0, rx_ring->next_to_use, rx_ring->next_to_clean);
 308
 309         /* Print Rx Ring */
 310         if (!netif_msg_rx_status(adapter))
 311                 goto exit;
 312
 313         dev_info(&adapter->pdev->dev, "Rx Ring Dump\n");
 314         switch (adapter->rx_ps_pages) {
 315         case 1:
 316         case 2:
 317         case 3:
 318                 /* [Extended] Packet Split Receive Descriptor Format
 319                  *
 320                  *    +-----------------------------------------------------+
 321                  *  0 |                Buffer Address 0 [63:0]              |
 322                  *    +-----------------------------------------------------+
 323                  *  8 |                Buffer Address 1 [63:0]              |
 324                  *    +-----------------------------------------------------+
 325                  * 16 |                Buffer Address 2 [63:0]              |
 326                  *    +-----------------------------------------------------+
 327                  * 24 |                Buffer Address 3 [63:0]              |
 328                  *    +-----------------------------------------------------+
 329                  */
 330                 pr_info("R  [desc]      [buffer 0 63:0 ] [buffer 1 63:0 ] [buffer 2 63:0 ] [buffer 3 63:0 ] [bi->dma       ] [bi->skb] <-- Ext Pkt Split format\n");
 331                 /* [Extended] Receive Descriptor (Write-Back) Format
 332                  *
 333                  *   63       48 47    32 31     13 12    8 7    4 3        0
 334                  *   +------------------------------------------------------+
 335                  * 0 | Packet   | IP     |  Rsvd   | MRQ   | Rsvd | MRQ RSS |
 336                  *   | Checksum | Ident  |         | Queue |      |  Type   |
 337                  *   +------------------------------------------------------+
 338                  * 8 | VLAN Tag | Length | Extended Error | Extended Status |
 339                  *   +------------------------------------------------------+
 340                  *   63       48 47    32 31            20 19               0
 341                  */
 342                 pr_info("RWB[desc]      [ck ipid mrqhsh] [vl   l0 ee  es] [ l3  l2  l1 hs] [reserved      ] ---------------- [bi->skb] <-- Ext Rx Write-Back format\n");
 343                 for (i = 0; i < rx_ring->count; i++) {
 344                         const char *next_desc;
 345                         buffer_info = &rx_ring->buffer_info[i];
 346                         rx_desc_ps = E1000_RX_DESC_PS(*rx_ring, i);
 347                         u1 = (struct my_u1 *)rx_desc_ps;
 348                         staterr =
 349                             le32_to_cpu(rx_desc_ps->wb.middle.status_error);
 350
 351                         if (i == rx_ring->next_to_use)
 352                                 next_desc = " NTU";
 353                         else if (i == rx_ring->next_to_clean)
 354                                 next_desc = " NTC";
 355                         else
 356                                 next_desc = "";
 357
 358                         if (staterr & E1000_RXD_STAT_DD) {
 359                                 /* Descriptor Done */
 360                                 pr_info("%s[0x%03X]     %016llX %016llX %016llX %016llX ---------------- %p%s\n",
 361                                         "RWB", i,
 362                                         (unsigned long long)le64_to_cpu(u1->a),
 363                                         (unsigned long long)le64_to_cpu(u1->b),
 364                                         (unsigned long long)le64_to_cpu(u1->c),
 365                                         (unsigned long long)le64_to_cpu(u1->d),
 366                                         buffer_info->skb, next_desc);
 367                         } else {
 368                                 pr_info("%s[0x%03X]     %016llX %016llX %016llX %016llX %016llX %p%s\n",
 369                                         "R  ", 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                                         (unsigned long long)buffer_info->dma,
 375                                         buffer_info->skb, next_desc);
 376
 377                                 if (netif_msg_pktdata(adapter))
 378                                         print_hex_dump(KERN_INFO, "",
 379                                                 DUMP_PREFIX_ADDRESS, 16, 1,
 380                                                 phys_to_virt(buffer_info->dma),
 381                                                 adapter->rx_ps_bsize0, true);
 382                         }
 383                 }
 384                 break;
 385         default:
 386         case 0:
 387                 /* Extended Receive Descriptor (Read) Format
 388                  *
 389                  *   +-----------------------------------------------------+
 390                  * 0 |                Buffer Address [63:0]                |
 391                  *   +-----------------------------------------------------+
 392                  * 8 |                      Reserved                       |
 393                  *   +-----------------------------------------------------+
 394                  */
 395                 pr_info("R  [desc]      [buf addr 63:0 ] [reserved 63:0 ] [bi->dma       ] [bi->skb] <-- Ext (Read) format\n");
 396                 /* Extended Receive Descriptor (Write-Back) Format
 397                  *
 398                  *   63       48 47    32 31    24 23            4 3        0
 399                  *   +------------------------------------------------------+
 400                  *   |     RSS Hash      |        |               |         |
 401                  * 0 +-------------------+  Rsvd  |   Reserved    | MRQ RSS |
 402                  *   | Packet   | IP     |        |               |  Type   |
 403                  *   | Checksum | Ident  |        |               |         |
 404                  *   +------------------------------------------------------+
 405                  * 8 | VLAN Tag | Length | Extended Error | Extended Status |
 406                  *   +------------------------------------------------------+
 407                  *   63       48 47    32 31            20 19               0
 408                  */
 409                 pr_info("RWB[desc]      [cs ipid    mrq] [vt   ln xe  xs] [bi->skb] <-- Ext (Write-Back) format\n");
 410
 411                 for (i = 0; i < rx_ring->count; i++) {
 412                         const char *next_desc;
 413
 414                         buffer_info = &rx_ring->buffer_info[i];
 415                         rx_desc = E1000_RX_DESC_EXT(*rx_ring, i);
 416                         u1 = (struct my_u1 *)rx_desc;
 417                         staterr = le32_to_cpu(rx_desc->wb.upper.status_error);
 418
 419                         if (i == rx_ring->next_to_use)
 420                                 next_desc = " NTU";
 421                         else if (i == rx_ring->next_to_clean)
 422                                 next_desc = " NTC";
 423                         else
 424                                 next_desc = "";
 425
 426                         if (staterr & E1000_RXD_STAT_DD) {
 427                                 /* Descriptor Done */
 428                                 pr_info("%s[0x%03X]     %016llX %016llX ---------------- %p%s\n",
 429                                         "RWB", i,
 430                                         (unsigned long long)le64_to_cpu(u1->a),
 431                                         (unsigned long long)le64_to_cpu(u1->b),
 432                                         buffer_info->skb, next_desc);
 433                         } else {
 434                                 pr_info("%s[0x%03X]     %016llX %016llX %016llX %p%s\n",
 435                                         "R  ", i,
 436                                         (unsigned long long)le64_to_cpu(u1->a),
 437                                         (unsigned long long)le64_to_cpu(u1->b),
 438                                         (unsigned long long)buffer_info->dma,
 439                                         buffer_info->skb, next_desc);
 440
 441                                 if (netif_msg_pktdata(adapter))
 442                                         print_hex_dump(KERN_INFO, "",
 443                                                        DUMP_PREFIX_ADDRESS, 16,
 444                                                        1,
 445                                                        phys_to_virt
 446                                                        (buffer_info->dma),
 447                                                        adapter->rx_buffer_len,
 448                                                        true);
 449                         }
 450                 }
 451         }
 452
 453 exit:
 454         return;
 455 }
 456
 457 /**
 458  * e1000_desc_unused - calculate if we have unused descriptors
 459  **/
 460 static int e1000_desc_unused(struct e1000_ring *ring)
 461 {
 462         if (ring->next_to_clean > ring->next_to_use)
 463                 return ring->next_to_clean - ring->next_to_use - 1;
 464
 465         return ring->count + ring->next_to_clean - ring->next_to_use - 1;
 466 }
 467
 468 /**
 469  * e1000_receive_skb - helper function to handle Rx indications
 470  * @adapter: board private structure
 471  * @status: descriptor status field as written by hardware
 472  * @vlan: descriptor vlan field as written by hardware (no le/be conversion)
 473  * @skb: pointer to sk_buff to be indicated to stack
 474  **/
 475 static void e1000_receive_skb(struct e1000_adapter *adapter,
 476                               struct net_device *netdev, struct sk_buff *skb,
 477                               u8 status, __le16 vlan)
 478 {
 479         u16 tag = le16_to_cpu(vlan);
 480         skb->protocol = eth_type_trans(skb, netdev);
 481
 482         if (status & E1000_RXD_STAT_VP)
 483                 __vlan_hwaccel_put_tag(skb, tag);
 484
 485         napi_gro_receive(&adapter->napi, skb);
 486 }
 487
 488 /**
 489  * e1000_rx_checksum - Receive Checksum Offload
 490  * @adapter:     board private structure
 491  * @status_err:  receive descriptor status and error fields
 492  * @csum:       receive descriptor csum field
 493  * @sk_buff:     socket buffer with received data
 494  **/
 495 static void e1000_rx_checksum(struct e1000_adapter *adapter, u32 status_err,
 496                               u32 csum, struct sk_buff *skb)
 497 {
 498         u16 status = (u16)status_err;
 499         u8 errors = (u8)(status_err >> 24);
 500
 501         skb_checksum_none_assert(skb);
 502
 503         /* Ignore Checksum bit is set */
 504         if (status & E1000_RXD_STAT_IXSM)
 505                 return;
 506         /* TCP/UDP checksum error bit is set */
 507         if (errors & E1000_RXD_ERR_TCPE) {
 508                 /* let the stack verify checksum errors */
 509                 adapter->hw_csum_err++;
 510                 return;
 511         }
 512
 513         /* TCP/UDP Checksum has not been calculated */
 514         if (!(status & (E1000_RXD_STAT_TCPCS | E1000_RXD_STAT_UDPCS)))
 515                 return;
 516
 517         /* It must be a TCP or UDP packet with a valid checksum */
 518         if (status & E1000_RXD_STAT_TCPCS) {
 519                 /* TCP checksum is good */
 520                 skb->ip_summed = CHECKSUM_UNNECESSARY;
 521         } else {
 522                 /*
 523                  * IP fragment with UDP payload
 524                  * Hardware complements the payload checksum, so we undo it
 525                  * and then put the value in host order for further stack use.
 526                  */
 527                 __sum16 sum = (__force __sum16)htons(csum);
 528                 skb->csum = csum_unfold(~sum);
 529                 skb->ip_summed = CHECKSUM_COMPLETE;
 530         }
 531         adapter->hw_csum_good++;
 532 }
 533
 534 /**
 535  * e1000e_update_tail_wa - helper function for e1000e_update_[rt]dt_wa()
 536  * @hw: pointer to the HW structure
 537  * @tail: address of tail descriptor register
 538  * @i: value to write to tail descriptor register
 539  *
 540  * When updating the tail register, the ME could be accessing Host CSR
 541  * registers at the same time.  Normally, this is handled in h/w by an
 542  * arbiter but on some parts there is a bug that acknowledges Host accesses
 543  * later than it should which could result in the descriptor register to
 544  * have an incorrect value.  Workaround this by checking the FWSM register
 545  * which has bit 24 set while ME is accessing Host CSR registers, wait
 546  * if it is set and try again a number of times.
 547  **/
 548 static inline s32 e1000e_update_tail_wa(struct e1000_hw *hw, u8 __iomem * tail,
 549                                         unsigned int i)
 550 {
 551         unsigned int j = 0;
 552
 553         while ((j++ < E1000_ICH_FWSM_PCIM2PCI_COUNT) &&
 554                (er32(FWSM) & E1000_ICH_FWSM_PCIM2PCI))
 555                 udelay(50);
 556
 557         writel(i, tail);
 558
 559         if ((j == E1000_ICH_FWSM_PCIM2PCI_COUNT) && (i != readl(tail)))
 560                 return E1000_ERR_SWFW_SYNC;
 561
 562         return 0;
 563 }
 564
 565 static void e1000e_update_rdt_wa(struct e1000_adapter *adapter, unsigned int i)
 566 {
 567         u8 __iomem *tail = (adapter->hw.hw_addr + adapter->rx_ring->tail);
 568         struct e1000_hw *hw = &adapter->hw;
 569
 570         if (e1000e_update_tail_wa(hw, tail, i)) {
 571                 u32 rctl = er32(RCTL);
 572                 ew32(RCTL, rctl & ~E1000_RCTL_EN);
 573                 e_err("ME firmware caused invalid RDT - resetting\n");
 574                 schedule_work(&adapter->reset_task);
 575         }
 576 }
 577
 578 static void e1000e_update_tdt_wa(struct e1000_adapter *adapter, unsigned int i)
 579 {
 580         u8 __iomem *tail = (adapter->hw.hw_addr + adapter->tx_ring->tail);
 581         struct e1000_hw *hw = &adapter->hw;
 582
 583         if (e1000e_update_tail_wa(hw, tail, i)) {
 584                 u32 tctl = er32(TCTL);
 585                 ew32(TCTL, tctl & ~E1000_TCTL_EN);
 586                 e_err("ME firmware caused invalid TDT - resetting\n");
 587                 schedule_work(&adapter->reset_task);
 588         }
 589 }
 590
 591 /**
 592  * e1000_alloc_rx_buffers - Replace used receive buffers
 593  * @adapter: address of board private structure
 594  **/
 595 static void e1000_alloc_rx_buffers(struct e1000_adapter *adapter,
 596                                    int cleaned_count, gfp_t gfp)
 597 {
 598         struct net_device *netdev = adapter->netdev;
 599         struct pci_dev *pdev = adapter->pdev;
 600         struct e1000_ring *rx_ring = adapter->rx_ring;
 601         union e1000_rx_desc_extended *rx_desc;
 602         struct e1000_buffer *buffer_info;
 603         struct sk_buff *skb;
 604         unsigned int i;
 605         unsigned int bufsz = adapter->rx_buffer_len;
 606
 607         i = rx_ring->next_to_use;
 608         buffer_info = &rx_ring->buffer_info[i];
 609
 610         while (cleaned_count--) {
 611                 skb = buffer_info->skb;
 612                 if (skb) {
 613                         skb_trim(skb, 0);
 614                         goto map_skb;
 615                 }
 616
 617                 skb = __netdev_alloc_skb_ip_align(netdev, bufsz, gfp);
 618                 if (!skb) {
 619                         /* Better luck next round */
 620                         adapter->alloc_rx_buff_failed++;
 621                         break;
 622                 }
 623
 624                 buffer_info->skb = skb;
 625 map_skb:
 626                 buffer_info->dma = dma_map_single(&pdev->dev, skb->data,
 627                                                   adapter->rx_buffer_len,
 628                                                   DMA_FROM_DEVICE);
 629                 if (dma_mapping_error(&pdev->dev, buffer_info->dma)) {
 630                         dev_err(&pdev->dev, "Rx DMA map failed\n");
 631                         adapter->rx_dma_failed++;
 632                         break;
 633                 }
 634
 635                 rx_desc = E1000_RX_DESC_EXT(*rx_ring, i);
 636                 rx_desc->read.buffer_addr = cpu_to_le64(buffer_info->dma);
 637
 638                 if (unlikely(!(i & (E1000_RX_BUFFER_WRITE - 1)))) {
 639                         /*
 640                          * Force memory writes to complete before letting h/w
 641                          * know there are new descriptors to fetch.  (Only
 642                          * applicable for weak-ordered memory model archs,
 643                          * such as IA-64).
 644                          */
 645                         wmb();
 646                         if (adapter->flags2 & FLAG2_PCIM2PCI_ARBITER_WA)
 647                                 e1000e_update_rdt_wa(adapter, i);
 648                         else
 649                                 writel(i, adapter->hw.hw_addr + rx_ring->tail);
 650                 }
 651                 i++;
 652                 if (i == rx_ring->count)
 653                         i = 0;
 654                 buffer_info = &rx_ring->buffer_info[i];
 655         }
 656
 657         rx_ring->next_to_use = i;
 658 }
 659
 660 /**
 661  * e1000_alloc_rx_buffers_ps - Replace used receive buffers; packet split
 662  * @adapter: address of board private structure
 663  **/
 664 static void e1000_alloc_rx_buffers_ps(struct e1000_adapter *adapter,
 665                                       int cleaned_count, gfp_t gfp)
 666 {
 667         struct net_device *netdev = adapter->netdev;
 668         struct pci_dev *pdev = adapter->pdev;
 669         union e1000_rx_desc_packet_split *rx_desc;
 670         struct e1000_ring *rx_ring = adapter->rx_ring;
 671         struct e1000_buffer *buffer_info;
 672         struct e1000_ps_page *ps_page;
 673         struct sk_buff *skb;
 674         unsigned int i, j;
 675
 676         i = rx_ring->next_to_use;
 677         buffer_info = &rx_ring->buffer_info[i];
 678
 679         while (cleaned_count--) {
 680                 rx_desc = E1000_RX_DESC_PS(*rx_ring, i);
 681
 682                 for (j = 0; j < PS_PAGE_BUFFERS; j++) {
 683                         ps_page = &buffer_info->ps_pages[j];
 684                         if (j >= adapter->rx_ps_pages) {
 685                                 /* all unused desc entries get hw null ptr */
 686                                 rx_desc->read.buffer_addr[j + 1] =
 687                                     ~cpu_to_le64(0);
 688                                 continue;
 689                         }
 690                         if (!ps_page->page) {
 691                                 ps_page->page = alloc_page(gfp);
 692                                 if (!ps_page->page) {
 693                                         adapter->alloc_rx_buff_failed++;
 694                                         goto no_buffers;
 695                                 }
 696                                 ps_page->dma = dma_map_page(&pdev->dev,
 697                                                             ps_page->page,
 698                                                             0, PAGE_SIZE,
 699                                                             DMA_FROM_DEVICE);
 700                                 if (dma_mapping_error(&pdev->dev,
 701                                                       ps_page->dma)) {
 702                                         dev_err(&adapter->pdev->dev,
 703                                                 "Rx DMA page map failed\n");
 704                                         adapter->rx_dma_failed++;
 705                                         goto no_buffers;
 706                                 }
 707                         }
 708                         /*
 709                          * Refresh the desc even if buffer_addrs
 710                          * didn't change because each write-back
 711                          * erases this info.
 712                          */
 713                         rx_desc->read.buffer_addr[j + 1] =
 714                             cpu_to_le64(ps_page->dma);
 715                 }
 716
 717                 skb = __netdev_alloc_skb_ip_align(netdev,
 718                                                   adapter->rx_ps_bsize0,
 719                                                   gfp);
 720
 721                 if (!skb) {
 722                         adapter->alloc_rx_buff_failed++;
 723                         break;
 724                 }
 725
 726                 buffer_info->skb = skb;
 727                 buffer_info->dma = dma_map_single(&pdev->dev, skb->data,
 728                                                   adapter->rx_ps_bsize0,
 729                                                   DMA_FROM_DEVICE);
 730                 if (dma_mapping_error(&pdev->dev, buffer_info->dma)) {
 731                         dev_err(&pdev->dev, "Rx DMA map failed\n");
 732                         adapter->rx_dma_failed++;
 733                         /* cleanup skb */
 734                         dev_kfree_skb_any(skb);
 735                         buffer_info->skb = NULL;
 736                         break;
 737                 }
 738
 739                 rx_desc->read.buffer_addr[0] = cpu_to_le64(buffer_info->dma);
 740
 741                 if (unlikely(!(i & (E1000_RX_BUFFER_WRITE - 1)))) {
 742                         /*
 743                          * Force memory writes to complete before letting h/w
 744                          * know there are new descriptors to fetch.  (Only
 745                          * applicable for weak-ordered memory model archs,
 746                          * such as IA-64).
 747                          */
 748                         wmb();
 749                         if (adapter->flags2 & FLAG2_PCIM2PCI_ARBITER_WA)
 750                                 e1000e_update_rdt_wa(adapter, i << 1);
 751                         else
 752                                 writel(i << 1,
 753                                        adapter->hw.hw_addr + rx_ring->tail);
 754                 }
 755
 756                 i++;
 757                 if (i == rx_ring->count)
 758                         i = 0;
 759                 buffer_info = &rx_ring->buffer_info[i];
 760         }
 761
 762 no_buffers:
 763         rx_ring->next_to_use = i;
 764 }
 765
 766 /**
 767  * e1000_alloc_jumbo_rx_buffers - Replace used jumbo receive buffers
 768  * @adapter: address of board private structure
 769  * @cleaned_count: number of buffers to allocate this pass
 770  **/
 771
 772 static void e1000_alloc_jumbo_rx_buffers(struct e1000_adapter *adapter,
 773                                          int cleaned_count, gfp_t gfp)
 774 {
 775         struct net_device *netdev = adapter->netdev;
 776         struct pci_dev *pdev = adapter->pdev;
 777         union e1000_rx_desc_extended *rx_desc;
 778         struct e1000_ring *rx_ring = adapter->rx_ring;
 779         struct e1000_buffer *buffer_info;
 780         struct sk_buff *skb;
 781         unsigned int i;
 782         unsigned int bufsz = 256 - 16 /* for skb_reserve */;
 783
 784         i = rx_ring->next_to_use;
 785         buffer_info = &rx_ring->buffer_info[i];
 786
 787         while (cleaned_count--) {
 788                 skb = buffer_info->skb;
 789                 if (skb) {
 790                         skb_trim(skb, 0);
 791                         goto check_page;
 792                 }
 793
 794                 skb = __netdev_alloc_skb_ip_align(netdev, bufsz, gfp);
 795                 if (unlikely(!skb)) {
 796                         /* Better luck next round */
 797                         adapter->alloc_rx_buff_failed++;
 798                         break;
 799                 }
 800
 801                 buffer_info->skb = skb;
 802 check_page:
 803                 /* allocate a new page if necessary */
 804                 if (!buffer_info->page) {
 805                         buffer_info->page = alloc_page(gfp);
 806                         if (unlikely(!buffer_info->page)) {
 807                                 adapter->alloc_rx_buff_failed++;
 808                                 break;
 809                         }
 810                 }
 811
 812                 if (!buffer_info->dma)
 813                         buffer_info->dma = dma_map_page(&pdev->dev,
 814                                                         buffer_info->page, 0,
 815                                                         PAGE_SIZE,
 816                                                         DMA_FROM_DEVICE);
 817
 818                 rx_desc = E1000_RX_DESC_EXT(*rx_ring, i);
 819                 rx_desc->read.buffer_addr = cpu_to_le64(buffer_info->dma);
 820
 821                 if (unlikely(++i == rx_ring->count))
 822                         i = 0;
 823                 buffer_info = &rx_ring->buffer_info[i];
 824         }
 825
 826         if (likely(rx_ring->next_to_use != i)) {
 827                 rx_ring->next_to_use = i;
 828                 if (unlikely(i-- == 0))
 829                         i = (rx_ring->count - 1);
 830
 831                 /* Force memory writes to complete before letting h/w
 832                  * know there are new descriptors to fetch.  (Only
 833                  * applicable for weak-ordered memory model archs,
 834                  * such as IA-64). */
 835                 wmb();
 836                 if (adapter->flags2 & FLAG2_PCIM2PCI_ARBITER_WA)
 837                         e1000e_update_rdt_wa(adapter, i);
 838                 else
 839                         writel(i, adapter->hw.hw_addr + rx_ring->tail);
 840         }
 841 }
 842
 843 /**
 844  * e1000_clean_rx_irq - Send received data up the network stack; legacy
 845  * @adapter: board private structure
 846  *
 847  * the return value indicates whether actual cleaning was done, there
 848  * is no guarantee that everything was cleaned
 849  **/
 850 static bool e1000_clean_rx_irq(struct e1000_adapter *adapter,
 851                                int *work_done, int work_to_do)
 852 {
 853         struct net_device *netdev = adapter->netdev;
 854         struct pci_dev *pdev = adapter->pdev;
 855         struct e1000_hw *hw = &adapter->hw;
 856         struct e1000_ring *rx_ring = adapter->rx_ring;
 857         union e1000_rx_desc_extended *rx_desc, *next_rxd;
 858         struct e1000_buffer *buffer_info, *next_buffer;
 859         u32 length, staterr;
 860         unsigned int i;
 861         int cleaned_count = 0;
 862         bool cleaned = false;
 863         unsigned int total_rx_bytes = 0, total_rx_packets = 0;
 864
 865         i = rx_ring->next_to_clean;
 866         rx_desc = E1000_RX_DESC_EXT(*rx_ring, i);
 867         staterr = le32_to_cpu(rx_desc->wb.upper.status_error);
 868         buffer_info = &rx_ring->buffer_info[i];
 869
 870         while (staterr & E1000_RXD_STAT_DD) {
 871                 struct sk_buff *skb;
 872
 873                 if (*work_done >= work_to_do)
 874                         break;
 875                 (*work_done)++;
 876                 rmb();  /* read descriptor and rx_buffer_info after status DD */
 877
 878                 skb = buffer_info->skb;
 879                 buffer_info->skb = NULL;
 880
 881                 prefetch(skb->data - NET_IP_ALIGN);
 882
 883                 i++;
 884                 if (i == rx_ring->count)
 885                         i = 0;
 886                 next_rxd = E1000_RX_DESC_EXT(*rx_ring, i);
 887                 prefetch(next_rxd);
 888
 889                 next_buffer = &rx_ring->buffer_info[i];
 890
 891                 cleaned = true;
 892                 cleaned_count++;
 893                 dma_unmap_single(&pdev->dev,
 894                                  buffer_info->dma,
 895                                  adapter->rx_buffer_len,
 896                                  DMA_FROM_DEVICE);
 897                 buffer_info->dma = 0;
 898
 899                 length = le16_to_cpu(rx_desc->wb.upper.length);
 900
 901                 /*
 902                  * !EOP means multiple descriptors were used to store a single
 903                  * packet, if that's the case we need to toss it.  In fact, we
 904                  * need to toss every packet with the EOP bit clear and the
 905                  * next frame that _does_ have the EOP bit set, as it is by
 906                  * definition only a frame fragment
 907                  */
 908                 if (unlikely(!(staterr & E1000_RXD_STAT_EOP)))
 909                         adapter->flags2 |= FLAG2_IS_DISCARDING;
 910
 911                 if (adapter->flags2 & FLAG2_IS_DISCARDING) {
 912                         /* All receives must fit into a single buffer */
 913                         e_dbg("Receive packet consumed multiple buffers\n");
 914                         /* recycle */
 915                         buffer_info->skb = skb;
 916                         if (staterr & E1000_RXD_STAT_EOP)
 917                                 adapter->flags2 &= ~FLAG2_IS_DISCARDING;
 918                         goto next_desc;
 919                 }
 920
 921                 if (staterr & E1000_RXDEXT_ERR_FRAME_ERR_MASK) {
 922                         /* recycle */
 923                         buffer_info->skb = skb;
 924                         goto next_desc;
 925                 }
 926
 927                 /* adjust length to remove Ethernet CRC */
 928                 if (!(adapter->flags2 & FLAG2_CRC_STRIPPING))
 929                         length -= 4;
 930
 931                 total_rx_bytes += length;
 932                 total_rx_packets++;
 933
 934                 /*
 935                  * code added for copybreak, this should improve
 936                  * performance for small packets with large amounts
 937                  * of reassembly being done in the stack
 938                  */
 939                 if (length < copybreak) {
 940                         struct sk_buff *new_skb =
 941                             netdev_alloc_skb_ip_align(netdev, length);
 942                         if (new_skb) {
 943                                 skb_copy_to_linear_data_offset(new_skb,
 944                                                                -NET_IP_ALIGN,
 945                                                                (skb->data -
 946                                                                 NET_IP_ALIGN),
 947                                                                (length +
 948                                                                 NET_IP_ALIGN));
 949                                 /* save the skb in buffer_info as good */
 950                                 buffer_info->skb = skb;
 951                                 skb = new_skb;
 952                         }
 953                         /* else just continue with the old one */
 954                 }
 955                 /* end copybreak code */
 956                 skb_put(skb, length);
 957
 958                 /* Receive Checksum Offload */
 959                 e1000_rx_checksum(adapter, staterr,
 960                                   le16_to_cpu(rx_desc->wb.lower.hi_dword.
 961                                               csum_ip.csum), skb);
 962
 963                 e1000_receive_skb(adapter, netdev, skb, staterr,
 964                                   rx_desc->wb.upper.vlan);
 965
 966 next_desc:
 967                 rx_desc->wb.upper.status_error &= cpu_to_le32(~0xFF);
 968
 969                 /* return some buffers to hardware, one at a time is too slow */
 970                 if (cleaned_count >= E1000_RX_BUFFER_WRITE) {
 971                         adapter->alloc_rx_buf(adapter, cleaned_count,
 972                                               GFP_ATOMIC);
 973                         cleaned_count = 0;
 974                 }
 975
 976                 /* use prefetched values */
 977                 rx_desc = next_rxd;
 978                 buffer_info = next_buffer;
 979
 980                 staterr = le32_to_cpu(rx_desc->wb.upper.status_error);
 981         }
 982         rx_ring->next_to_clean = i;
 983
 984         cleaned_count = e1000_desc_unused(rx_ring);
 985         if (cleaned_count)
 986                 adapter->alloc_rx_buf(adapter, cleaned_count, GFP_ATOMIC);
 987
 988         adapter->total_rx_bytes += total_rx_bytes;
 989         adapter->total_rx_packets += total_rx_packets;
 990         return cleaned;
 991 }
 992
 993 static void e1000_put_txbuf(struct e1000_adapter *adapter,
 994                              struct e1000_buffer *buffer_info)
 995 {
 996         if (buffer_info->dma) {
 997                 if (buffer_info->mapped_as_page)
 998                         dma_unmap_page(&adapter->pdev->dev, buffer_info->dma,
 999                                        buffer_info->length, DMA_TO_DEVICE);
1000                 else
1001                         dma_unmap_single(&adapter->pdev->dev, buffer_info->dma,
1002                                          buffer_info->length, DMA_TO_DEVICE);
1003                 buffer_info->dma = 0;
1004         }
1005         if (buffer_info->skb) {
1006                 dev_kfree_skb_any(buffer_info->skb);
1007                 buffer_info->skb = NULL;
1008         }
1009         buffer_info->time_stamp = 0;
1010 }
1011
1012 static void e1000_print_hw_hang(struct work_struct *work)
1013 {
1014         struct e1000_adapter *adapter = container_of(work,
1015                                                      struct e1000_adapter,
1016                                                      print_hang_task);
1017         struct net_device *netdev = adapter->netdev;
1018         struct e1000_ring *tx_ring = adapter->tx_ring;
1019         unsigned int i = tx_ring->next_to_clean;
1020         unsigned int eop = tx_ring->buffer_info[i].next_to_watch;
1021         struct e1000_tx_desc *eop_desc = E1000_TX_DESC(*tx_ring, eop);
1022         struct e1000_hw *hw = &adapter->hw;
1023         u16 phy_status, phy_1000t_status, phy_ext_status;
1024         u16 pci_status;
1025
1026         if (test_bit(__E1000_DOWN, &adapter->state))
1027                 return;
1028
1029         if (!adapter->tx_hang_recheck &&
1030             (adapter->flags2 & FLAG2_DMA_BURST)) {
1031                 /* May be block on write-back, flush and detect again
1032                  * flush pending descriptor writebacks to memory
1033                  */
1034                 ew32(TIDV, adapter->tx_int_delay | E1000_TIDV_FPD);
1035                 /* execute the writes immediately */
1036                 e1e_flush();
1037                 adapter->tx_hang_recheck = true;
1038                 return;
1039         }
1040         /* Real hang detected */
1041         adapter->tx_hang_recheck = false;
1042         netif_stop_queue(netdev);
1043
1044         e1e_rphy(hw, PHY_STATUS, &phy_status);
1045         e1e_rphy(hw, PHY_1000T_STATUS, &phy_1000t_status);
1046         e1e_rphy(hw, PHY_EXT_STATUS, &phy_ext_status);
1047
1048         pci_read_config_word(adapter->pdev, PCI_STATUS, &pci_status);
1049
1050         /* detected Hardware unit hang */
1051         e_err("Detected Hardware Unit Hang:\n"
1052               "  TDH                  <%x>\n"
1053               "  TDT                  <%x>\n"
1054               "  next_to_use          <%x>\n"
1055               "  next_to_clean        <%x>\n"
1056               "buffer_info[next_to_clean]:\n"
1057               "  time_stamp           <%lx>\n"
1058               "  next_to_watch        <%x>\n"
1059               "  jiffies              <%lx>\n"
1060               "  next_to_watch.status <%x>\n"
1061               "MAC Status             <%x>\n"
1062               "PHY Status             <%x>\n"
1063               "PHY 1000BASE-T Status  <%x>\n"
1064               "PHY Extended Status    <%x>\n"
1065               "PCI Status             <%x>\n",
1066               readl(adapter->hw.hw_addr + tx_ring->head),
1067               readl(adapter->hw.hw_addr + tx_ring->tail),
1068               tx_ring->next_to_use,
1069               tx_ring->next_to_clean,
1070               tx_ring->buffer_info[eop].time_stamp,
1071               eop,
1072               jiffies,
1073               eop_desc->upper.fields.status,
1074               er32(STATUS),
1075               phy_status,
1076               phy_1000t_status,
1077               phy_ext_status,
1078               pci_status);
1079 }
1080
1081 /**
1082  * e1000_clean_tx_irq - Reclaim resources after transmit completes
1083  * @adapter: board private structure
1084  *
1085  * the return value indicates whether actual cleaning was done, there
1086  * is no guarantee that everything was cleaned
1087  **/
1088 static bool e1000_clean_tx_irq(struct e1000_adapter *adapter)
1089 {
1090         struct net_device *netdev = adapter->netdev;
1091         struct e1000_hw *hw = &adapter->hw;
1092         struct e1000_ring *tx_ring = adapter->tx_ring;
1093         struct e1000_tx_desc *tx_desc, *eop_desc;
1094         struct e1000_buffer *buffer_info;
1095         unsigned int i, eop;
1096         unsigned int count = 0;
1097         unsigned int total_tx_bytes = 0, total_tx_packets = 0;
1098         unsigned int bytes_compl = 0, pkts_compl = 0;
1099
1100         i = tx_ring->next_to_clean;
1101         eop = tx_ring->buffer_info[i].next_to_watch;
1102         eop_desc = E1000_TX_DESC(*tx_ring, eop);
1103
1104         while ((eop_desc->upper.data & cpu_to_le32(E1000_TXD_STAT_DD)) &&
1105                (count < tx_ring->count)) {
1106                 bool cleaned = false;
1107                 rmb(); /* read buffer_info after eop_desc */
1108                 for (; !cleaned; count++) {
1109                         tx_desc = E1000_TX_DESC(*tx_ring, i);
1110                         buffer_info = &tx_ring->buffer_info[i];
1111                         cleaned = (i == eop);
1112
1113                         if (cleaned) {
1114                                 total_tx_packets += buffer_info->segs;
1115                                 total_tx_bytes += buffer_info->bytecount;
1116                                 if (buffer_info->skb) {
1117                                         bytes_compl += buffer_info->skb->len;
1118                                         pkts_compl++;
1119                                 }
1120                         }
1121
1122                         e1000_put_txbuf(adapter, buffer_info);
1123                         tx_desc->upper.data = 0;
1124
1125                         i++;
1126                         if (i == tx_ring->count)
1127                                 i = 0;
1128                 }
1129
1130                 if (i == tx_ring->next_to_use)
1131                         break;
1132                 eop = tx_ring->buffer_info[i].next_to_watch;
1133                 eop_desc = E1000_TX_DESC(*tx_ring, eop);
1134         }
1135
1136         tx_ring->next_to_clean = i;
1137
1138         netdev_completed_queue(netdev, pkts_compl, bytes_compl);
1139
1140 #define TX_WAKE_THRESHOLD 32
1141         if (count && netif_carrier_ok(netdev) &&
1142             e1000_desc_unused(tx_ring) >= TX_WAKE_THRESHOLD) {
1143                 /* Make sure that anybody stopping the queue after this
1144                  * sees the new next_to_clean.
1145                  */
1146                 smp_mb();
1147
1148                 if (netif_queue_stopped(netdev) &&
1149                     !(test_bit(__E1000_DOWN, &adapter->state))) {
1150                         netif_wake_queue(netdev);
1151                         ++adapter->restart_queue;
1152                 }
1153         }
1154
1155         if (adapter->detect_tx_hung) {
1156                 /*
1157                  * Detect a transmit hang in hardware, this serializes the
1158                  * check with the clearing of time_stamp and movement of i
1159                  */
1160                 adapter->detect_tx_hung = false;
1161                 if (tx_ring->buffer_info[i].time_stamp &&
1162                     time_after(jiffies, tx_ring->buffer_info[i].time_stamp
1163                                + (adapter->tx_timeout_factor * HZ)) &&
1164                     !(er32(STATUS) & E1000_STATUS_TXOFF))
1165                         schedule_work(&adapter->print_hang_task);
1166                 else
1167                         adapter->tx_hang_recheck = false;
1168         }
1169         adapter->total_tx_bytes += total_tx_bytes;
1170         adapter->total_tx_packets += total_tx_packets;
1171         return count < tx_ring->count;
1172 }
1173
1174 /**
1175  * e1000_clean_rx_irq_ps - Send received data up the network stack; packet split
1176  * @adapter: board private structure
1177  *
1178  * the return value indicates whether actual cleaning was done, there
1179  * is no guarantee that everything was cleaned
1180  **/
1181 static bool e1000_clean_rx_irq_ps(struct e1000_adapter *adapter,
1182                                   int *work_done, int work_to_do)
1183 {
1184         struct e1000_hw *hw = &adapter->hw;
1185         union e1000_rx_desc_packet_split *rx_desc, *next_rxd;
1186         struct net_device *netdev = adapter->netdev;
1187         struct pci_dev *pdev = adapter->pdev;
1188         struct e1000_ring *rx_ring = adapter->rx_ring;
1189         struct e1000_buffer *buffer_info, *next_buffer;
1190         struct e1000_ps_page *ps_page;
1191         struct sk_buff *skb;
1192         unsigned int i, j;
1193         u32 length, staterr;
1194         int cleaned_count = 0;
1195         bool cleaned = false;
1196         unsigned int total_rx_bytes = 0, total_rx_packets = 0;
1197
1198         i = rx_ring->next_to_clean;
1199         rx_desc = E1000_RX_DESC_PS(*rx_ring, i);
1200         staterr = le32_to_cpu(rx_desc->wb.middle.status_error);
1201         buffer_info = &rx_ring->buffer_info[i];
1202
1203         while (staterr & E1000_RXD_STAT_DD) {
1204                 if (*work_done >= work_to_do)
1205                         break;
1206                 (*work_done)++;
1207                 skb = buffer_info->skb;
1208                 rmb();  /* read descriptor and rx_buffer_info after status DD */
1209
1210                 /* in the packet split case this is header only */
1211                 prefetch(skb->data - NET_IP_ALIGN);
1212
1213                 i++;
1214                 if (i == rx_ring->count)
1215                         i = 0;
1216                 next_rxd = E1000_RX_DESC_PS(*rx_ring, i);
1217                 prefetch(next_rxd);
1218
1219                 next_buffer = &rx_ring->buffer_info[i];
1220
1221                 cleaned = true;
1222                 cleaned_count++;
1223                 dma_unmap_single(&pdev->dev, buffer_info->dma,
1224                                  adapter->rx_ps_bsize0, DMA_FROM_DEVICE);
1225                 buffer_info->dma = 0;
1226
1227                 /* see !EOP comment in other Rx routine */
1228                 if (!(staterr & E1000_RXD_STAT_EOP))
1229                         adapter->flags2 |= FLAG2_IS_DISCARDING;
1230
1231                 if (adapter->flags2 & FLAG2_IS_DISCARDING) {
1232                         e_dbg("Packet Split buffers didn't pick up the full packet\n");
1233                         dev_kfree_skb_irq(skb);
1234                         if (staterr & E1000_RXD_STAT_EOP)
1235                                 adapter->flags2 &= ~FLAG2_IS_DISCARDING;
1236                         goto next_desc;
1237                 }
1238
1239                 if (staterr & E1000_RXDEXT_ERR_FRAME_ERR_MASK) {
1240                         dev_kfree_skb_irq(skb);
1241                         goto next_desc;
1242                 }
1243
1244                 length = le16_to_cpu(rx_desc->wb.middle.length0);
1245
1246                 if (!length) {
1247                         e_dbg("Last part of the packet spanning multiple descriptors\n");
1248                         dev_kfree_skb_irq(skb);
1249                         goto next_desc;
1250                 }
1251
1252                 /* Good Receive */
1253                 skb_put(skb, length);
1254
1255                 {
1256                 /*
1257                  * this looks ugly, but it seems compiler issues make it
1258                  * more efficient than reusing j
1259                  */
1260                 int l1 = le16_to_cpu(rx_desc->wb.upper.length[0]);
1261
1262                 /*
1263                  * page alloc/put takes too long and effects small packet
1264                  * throughput, so unsplit small packets and save the alloc/put
1265                  * only valid in softirq (napi) context to call kmap_*
1266                  */
1267                 if (l1 && (l1 <= copybreak) &&
1268                     ((length + l1) <= adapter->rx_ps_bsize0)) {
1269                         u8 *vaddr;
1270
1271                         ps_page = &buffer_info->ps_pages[0];
1272
1273                         /*
1274                          * there is no documentation about how to call
1275                          * kmap_atomic, so we can't hold the mapping
1276                          * very long
1277                          */
1278                         dma_sync_single_for_cpu(&pdev->dev, ps_page->dma,
1279                                                 PAGE_SIZE, DMA_FROM_DEVICE);
1280                         vaddr = kmap_atomic(ps_page->page);
1281                         memcpy(skb_tail_pointer(skb), vaddr, l1);
1282                         kunmap_atomic(vaddr);
1283                         dma_sync_single_for_device(&pdev->dev, ps_page->dma,
1284                                                    PAGE_SIZE, DMA_FROM_DEVICE);
1285
1286                         /* remove the CRC */
1287                         if (!(adapter->flags2 & FLAG2_CRC_STRIPPING))
1288                                 l1 -= 4;
1289
1290                         skb_put(skb, l1);
1291                         goto copydone;
1292                 } /* if */
1293                 }
1294
1295                 for (j = 0; j < PS_PAGE_BUFFERS; j++) {
1296                         length = le16_to_cpu(rx_desc->wb.upper.length[j]);
1297                         if (!length)
1298                                 break;
1299
1300                         ps_page = &buffer_info->ps_pages[j];
1301                         dma_unmap_page(&pdev->dev, ps_page->dma, PAGE_SIZE,
1302                                        DMA_FROM_DEVICE);
1303                         ps_page->dma = 0;
1304                         skb_fill_page_desc(skb, j, ps_page->page, 0, length);
1305                         ps_page->page = NULL;
1306                         skb->len += length;
1307                         skb->data_len += length;
1308                         skb->truesize += PAGE_SIZE;
1309                 }
1310
1311                 /* strip the ethernet crc, problem is we're using pages now so
1312                  * this whole operation can get a little cpu intensive
1313                  */
1314                 if (!(adapter->flags2 & FLAG2_CRC_STRIPPING))
1315                         pskb_trim(skb, skb->len - 4);
1316
1317 copydone:
1318                 total_rx_bytes += skb->len;
1319                 total_rx_packets++;
1320
1321                 e1000_rx_checksum(adapter, staterr, le16_to_cpu(
1322                         rx_desc->wb.lower.hi_dword.csum_ip.csum), skb);
1323
1324                 if (rx_desc->wb.upper.header_status &
1325                            cpu_to_le16(E1000_RXDPS_HDRSTAT_HDRSP))
1326                         adapter->rx_hdr_split++;
1327
1328                 e1000_receive_skb(adapter, netdev, skb,
1329                                   staterr, rx_desc->wb.middle.vlan);
1330
1331 next_desc:
1332                 rx_desc->wb.middle.status_error &= cpu_to_le32(~0xFF);
1333                 buffer_info->skb = NULL;
1334
1335                 /* return some buffers to hardware, one at a time is too slow */
1336                 if (cleaned_count >= E1000_RX_BUFFER_WRITE) {
1337                         adapter->alloc_rx_buf(adapter, cleaned_count,
1338                                               GFP_ATOMIC);
1339                         cleaned_count = 0;
1340                 }
1341
1342                 /* use prefetched values */
1343                 rx_desc = next_rxd;
1344                 buffer_info = next_buffer;
1345
1346                 staterr = le32_to_cpu(rx_desc->wb.middle.status_error);
1347         }
1348         rx_ring->next_to_clean = i;
1349
1350         cleaned_count = e1000_desc_unused(rx_ring);
1351         if (cleaned_count)
1352                 adapter->alloc_rx_buf(adapter, cleaned_count, GFP_ATOMIC);
1353
1354         adapter->total_rx_bytes += total_rx_bytes;
1355         adapter->total_rx_packets += total_rx_packets;
1356         return cleaned;
1357 }
1358
1359 /**
1360  * e1000_consume_page - helper function
1361  **/
1362 static void e1000_consume_page(struct e1000_buffer *bi, struct sk_buff *skb,
1363                                u16 length)
1364 {
1365         bi->page = NULL;
1366         skb->len += length;
1367         skb->data_len += length;
1368         skb->truesize += PAGE_SIZE;
1369 }
1370
1371 /**
1372  * e1000_clean_jumbo_rx_irq - Send received data up the network stack; legacy
1373  * @adapter: board private structure
1374  *
1375  * the return value indicates whether actual cleaning was done, there
1376  * is no guarantee that everything was cleaned
1377  **/
1378
1379 static bool e1000_clean_jumbo_rx_irq(struct e1000_adapter *adapter,
1380                                      int *work_done, int work_to_do)
1381 {
1382         struct net_device *netdev = adapter->netdev;
1383         struct pci_dev *pdev = adapter->pdev;
1384         struct e1000_ring *rx_ring = adapter->rx_ring;
1385         union e1000_rx_desc_extended *rx_desc, *next_rxd;
1386         struct e1000_buffer *buffer_info, *next_buffer;
1387         u32 length, staterr;
1388         unsigned int i;
1389         int cleaned_count = 0;
1390         bool cleaned = false;
1391         unsigned int total_rx_bytes=0, total_rx_packets=0;
1392
1393         i = rx_ring->next_to_clean;
1394         rx_desc = E1000_RX_DESC_EXT(*rx_ring, i);
1395         staterr = le32_to_cpu(rx_desc->wb.upper.status_error);
1396         buffer_info = &rx_ring->buffer_info[i];
1397
1398         while (staterr & E1000_RXD_STAT_DD) {
1399                 struct sk_buff *skb;
1400
1401                 if (*work_done >= work_to_do)
1402                         break;
1403                 (*work_done)++;
1404                 rmb();  /* read descriptor and rx_buffer_info after status DD */
1405
1406                 skb = buffer_info->skb;
1407                 buffer_info->skb = NULL;
1408
1409                 ++i;
1410                 if (i == rx_ring->count)
1411                         i = 0;
1412                 next_rxd = E1000_RX_DESC_EXT(*rx_ring, i);
1413                 prefetch(next_rxd);
1414
1415                 next_buffer = &rx_ring->buffer_info[i];
1416
1417                 cleaned = true;
1418                 cleaned_count++;
1419                 dma_unmap_page(&pdev->dev, buffer_info->dma, PAGE_SIZE,
1420                                DMA_FROM_DEVICE);
1421                 buffer_info->dma = 0;
1422
1423                 length = le16_to_cpu(rx_desc->wb.upper.length);
1424
1425                 /* errors is only valid for DD + EOP descriptors */
1426                 if (unlikely((staterr & E1000_RXD_STAT_EOP) &&
1427                              (staterr & E1000_RXDEXT_ERR_FRAME_ERR_MASK))) {
1428                         /* recycle both page and skb */
1429                         buffer_info->skb = skb;
1430                         /* an error means any chain goes out the window too */
1431                         if (rx_ring->rx_skb_top)
1432                                 dev_kfree_skb_irq(rx_ring->rx_skb_top);
1433                         rx_ring->rx_skb_top = NULL;
1434                         goto next_desc;
1435                 }
1436
1437 #define rxtop (rx_ring->rx_skb_top)
1438                 if (!(staterr & E1000_RXD_STAT_EOP)) {
1439                         /* this descriptor is only the beginning (or middle) */
1440                         if (!rxtop) {
1441                                 /* this is the beginning of a chain */
1442                                 rxtop = skb;
1443                                 skb_fill_page_desc(rxtop, 0, buffer_info->page,
1444                                                    0, length);
1445                         } else {
1446                                 /* this is the middle of a chain */
1447                                 skb_fill_page_desc(rxtop,
1448                                     skb_shinfo(rxtop)->nr_frags,
1449                                     buffer_info->page, 0, length);
1450                                 /* re-use the skb, only consumed the page */
1451                                 buffer_info->skb = skb;
1452                         }
1453                         e1000_consume_page(buffer_info, rxtop, length);
1454                         goto next_desc;
1455                 } else {
1456                         if (rxtop) {
1457                                 /* end of the chain */
1458                                 skb_fill_page_desc(rxtop,
1459                                     skb_shinfo(rxtop)->nr_frags,
1460                                     buffer_info->page, 0, length);
1461                                 /* re-use the current skb, we only consumed the
1462                                  * page */
1463                                 buffer_info->skb = skb;
1464                                 skb = rxtop;
1465                                 rxtop = NULL;
1466                                 e1000_consume_page(buffer_info, skb, length);
1467                         } else {
1468                                 /* no chain, got EOP, this buf is the packet
1469                                  * copybreak to save the put_page/alloc_page */
1470                                 if (length <= copybreak &&
1471                                     skb_tailroom(skb) >= length) {
1472                                         u8 *vaddr;
1473                                         vaddr = kmap_atomic(buffer_info->page);
1474                                         memcpy(skb_tail_pointer(skb), vaddr,
1475                                                length);
1476                                         kunmap_atomic(vaddr);
1477                                         /* re-use the page, so don't erase
1478                                          * buffer_info->page */
1479                                         skb_put(skb, length);
1480                                 } else {
1481                                         skb_fill_page_desc(skb, 0,
1482                                                            buffer_info->page, 0,
1483                                                            length);
1484                                         e1000_consume_page(buffer_info, skb,
1485                                                            length);
1486                                 }
1487                         }
1488                 }
1489
1490                 /* Receive Checksum Offload XXX recompute due to CRC strip? */
1491                 e1000_rx_checksum(adapter, staterr,
1492                                   le16_to_cpu(rx_desc->wb.lower.hi_dword.
1493                                               csum_ip.csum), skb);
1494
1495                 /* probably a little skewed due to removing CRC */
1496                 total_rx_bytes += skb->len;
1497                 total_rx_packets++;
1498
1499                 /* eth type trans needs skb->data to point to something */
1500                 if (!pskb_may_pull(skb, ETH_HLEN)) {
1501                         e_err("pskb_may_pull failed.\n");
1502                         dev_kfree_skb_irq(skb);
1503                         goto next_desc;
1504                 }
1505
1506                 e1000_receive_skb(adapter, netdev, skb, staterr,
1507                                   rx_desc->wb.upper.vlan);
1508
1509 next_desc:
1510                 rx_desc->wb.upper.status_error &= cpu_to_le32(~0xFF);
1511
1512                 /* return some buffers to hardware, one at a time is too slow */
1513                 if (unlikely(cleaned_count >= E1000_RX_BUFFER_WRITE)) {
1514                         adapter->alloc_rx_buf(adapter, cleaned_count,
1515                                               GFP_ATOMIC);
1516                         cleaned_count = 0;
1517                 }
1518
1519                 /* use prefetched values */
1520                 rx_desc = next_rxd;
1521                 buffer_info = next_buffer;
1522
1523                 staterr = le32_to_cpu(rx_desc->wb.upper.status_error);
1524         }
1525         rx_ring->next_to_clean = i;
1526
1527         cleaned_count = e1000_desc_unused(rx_ring);
1528         if (cleaned_count)
1529                 adapter->alloc_rx_buf(adapter, cleaned_count, GFP_ATOMIC);
1530
1531         adapter->total_rx_bytes += total_rx_bytes;
1532         adapter->total_rx_packets += total_rx_packets;
1533         return cleaned;
1534 }
1535
1536 /**
1537  * e1000_clean_rx_ring - Free Rx Buffers per Queue
1538  * @adapter: board private structure
1539  **/
1540 static void e1000_clean_rx_ring(struct e1000_adapter *adapter)
1541 {
1542         struct e1000_ring *rx_ring = adapter->rx_ring;
1543         struct e1000_buffer *buffer_info;
1544         struct e1000_ps_page *ps_page;
1545         struct pci_dev *pdev = adapter->pdev;
1546         unsigned int i, j;
1547
1548         /* Free all the Rx ring sk_buffs */
1549         for (i = 0; i < rx_ring->count; i++) {
1550                 buffer_info = &rx_ring->buffer_info[i];
1551                 if (buffer_info->dma) {
1552                         if (adapter->clean_rx == e1000_clean_rx_irq)
1553                                 dma_unmap_single(&pdev->dev, buffer_info->dma,
1554                                                  adapter->rx_buffer_len,
1555                                                  DMA_FROM_DEVICE);
1556                         else if (adapter->clean_rx == e1000_clean_jumbo_rx_irq)
1557                                 dma_unmap_page(&pdev->dev, buffer_info->dma,
1558                                                PAGE_SIZE,
1559                                                DMA_FROM_DEVICE);
1560                         else if (adapter->clean_rx == e1000_clean_rx_irq_ps)
1561                                 dma_unmap_single(&pdev->dev, buffer_info->dma,
1562                                                  adapter->rx_ps_bsize0,
1563                                                  DMA_FROM_DEVICE);
1564                         buffer_info->dma = 0;
1565                 }
1566
1567                 if (buffer_info->page) {
1568                         put_page(buffer_info->page);
1569                         buffer_info->page = NULL;
1570                 }
1571
1572                 if (buffer_info->skb) {
1573                         dev_kfree_skb(buffer_info->skb);
1574                         buffer_info->skb = NULL;
1575                 }
1576
1577                 for (j = 0; j < PS_PAGE_BUFFERS; j++) {
1578                         ps_page = &buffer_info->ps_pages[j];
1579                         if (!ps_page->page)
1580                                 break;
1581                         dma_unmap_page(&pdev->dev, ps_page->dma, PAGE_SIZE,
1582                                        DMA_FROM_DEVICE);
1583                         ps_page->dma = 0;
1584                         put_page(ps_page->page);
1585                         ps_page->page = NULL;
1586                 }
1587         }
1588
1589         /* there also may be some cached data from a chained receive */
1590         if (rx_ring->rx_skb_top) {
1591                 dev_kfree_skb(rx_ring->rx_skb_top);
1592                 rx_ring->rx_skb_top = NULL;
1593         }
1594
1595         /* Zero out the descriptor ring */
1596         memset(rx_ring->desc, 0, rx_ring->size);
1597
1598         rx_ring->next_to_clean = 0;
1599         rx_ring->next_to_use = 0;
1600         adapter->flags2 &= ~FLAG2_IS_DISCARDING;
1601
1602         writel(0, adapter->hw.hw_addr + rx_ring->head);
1603         writel(0, adapter->hw.hw_addr + rx_ring->tail);
1604 }
1605
1606 static void e1000e_downshift_workaround(struct work_struct *work)
1607 {
1608         struct e1000_adapter *adapter = container_of(work,
1609                                         struct e1000_adapter, downshift_task);
1610
1611         if (test_bit(__E1000_DOWN, &adapter->state))
1612                 return;
1613
1614         e1000e_gig_downshift_workaround_ich8lan(&adapter->hw);
1615 }
1616
1617 /**
1618  * e1000_intr_msi - Interrupt Handler
1619  * @irq: interrupt number
1620  * @data: pointer to a network interface device structure
1621  **/
1622 static irqreturn_t e1000_intr_msi(int irq, void *data)
1623 {
1624         struct net_device *netdev = data;
1625         struct e1000_adapter *adapter = netdev_priv(netdev);
1626         struct e1000_hw *hw = &adapter->hw;
1627         u32 icr = er32(ICR);
1628
1629         /*
1630          * read ICR disables interrupts using IAM
1631          */
1632
1633         if (icr & E1000_ICR_LSC) {
1634                 hw->mac.get_link_status = 1;
1635                 /*
1636                  * ICH8 workaround-- Call gig speed drop workaround on cable
1637                  * disconnect (LSC) before accessing any PHY registers
1638                  */
1639                 if ((adapter->flags & FLAG_LSC_GIG_SPEED_DROP) &&
1640                     (!(er32(STATUS) & E1000_STATUS_LU)))
1641                         schedule_work(&adapter->downshift_task);
1642
1643                 /*
1644                  * 80003ES2LAN workaround-- For packet buffer work-around on
1645                  * link down event; disable receives here in the ISR and reset
1646                  * adapter in watchdog
1647                  */
1648                 if (netif_carrier_ok(netdev) &&
1649                     adapter->flags & FLAG_RX_NEEDS_RESTART) {
1650                         /* disable receives */
1651                         u32 rctl = er32(RCTL);
1652                         ew32(RCTL, rctl & ~E1000_RCTL_EN);
1653                         adapter->flags |= FLAG_RX_RESTART_NOW;
1654                 }
1655                 /* guard against interrupt when we're going down */
1656                 if (!test_bit(__E1000_DOWN, &adapter->state))
1657                         mod_timer(&adapter->watchdog_timer, jiffies + 1);
1658         }
1659
1660         if (napi_schedule_prep(&adapter->napi)) {
1661                 adapter->total_tx_bytes = 0;
1662                 adapter->total_tx_packets = 0;
1663                 adapter->total_rx_bytes = 0;
1664                 adapter->total_rx_packets = 0;
1665                 __napi_schedule(&adapter->napi);
1666         }
1667
1668         return IRQ_HANDLED;
1669 }
1670
1671 /**
1672  * e1000_intr - Interrupt Handler
1673  * @irq: interrupt number
1674  * @data: pointer to a network interface device structure
1675  **/
1676 static irqreturn_t e1000_intr(int irq, void *data)
1677 {
1678         struct net_device *netdev = data;
1679         struct e1000_adapter *adapter = netdev_priv(netdev);
1680         struct e1000_hw *hw = &adapter->hw;
1681         u32 rctl, icr = er32(ICR);
1682
1683         if (!icr || test_bit(__E1000_DOWN, &adapter->state))
1684                 return IRQ_NONE;  /* Not our interrupt */
1685
1686         /*
1687          * IMS will not auto-mask if INT_ASSERTED is not set, and if it is
1688          * not set, then the adapter didn't send an interrupt
1689          */
1690         if (!(icr & E1000_ICR_INT_ASSERTED))
1691                 return IRQ_NONE;
1692
1693         /*
1694          * Interrupt Auto-Mask...upon reading ICR,
1695          * interrupts are masked.  No need for the
1696          * IMC write
1697          */
1698
1699         if (icr & E1000_ICR_LSC) {
1700                 hw->mac.get_link_status = 1;
1701                 /*
1702                  * ICH8 workaround-- Call gig speed drop workaround on cable
1703                  * disconnect (LSC) before accessing any PHY registers
1704                  */
1705                 if ((adapter->flags & FLAG_LSC_GIG_SPEED_DROP) &&
1706                     (!(er32(STATUS) & E1000_STATUS_LU)))
1707                         schedule_work(&adapter->downshift_task);
1708
1709                 /*
1710                  * 80003ES2LAN workaround--
1711                  * For packet buffer work-around on link down event;
1712                  * disable receives here in the ISR and
1713                  * reset adapter in watchdog
1714                  */
1715                 if (netif_carrier_ok(netdev) &&
1716                     (adapter->flags & FLAG_RX_NEEDS_RESTART)) {
1717                         /* disable receives */
1718                         rctl = er32(RCTL);
1719                         ew32(RCTL, rctl & ~E1000_RCTL_EN);
1720                         adapter->flags |= FLAG_RX_RESTART_NOW;
1721                 }
1722                 /* guard against interrupt when we're going down */
1723                 if (!test_bit(__E1000_DOWN, &adapter->state))
1724                         mod_timer(&adapter->watchdog_timer, jiffies + 1);
1725         }
1726
1727         if (napi_schedule_prep(&adapter->napi)) {
1728                 adapter->total_tx_bytes = 0;
1729                 adapter->total_tx_packets = 0;
1730                 adapter->total_rx_bytes = 0;
1731                 adapter->total_rx_packets = 0;
1732                 __napi_schedule(&adapter->napi);
1733         }
1734
1735         return IRQ_HANDLED;
1736 }
1737
1738 static irqreturn_t e1000_msix_other(int irq, void *data)
1739 {
1740         struct net_device *netdev = data;
1741         struct e1000_adapter *adapter = netdev_priv(netdev);
1742         struct e1000_hw *hw = &adapter->hw;
1743         u32 icr = er32(ICR);
1744
1745         if (!(icr & E1000_ICR_INT_ASSERTED)) {
1746                 if (!test_bit(__E1000_DOWN, &adapter->state))
1747                         ew32(IMS, E1000_IMS_OTHER);
1748                 return IRQ_NONE;
1749         }
1750
1751         if (icr & adapter->eiac_mask)
1752                 ew32(ICS, (icr & adapter->eiac_mask));
1753
1754         if (icr & E1000_ICR_OTHER) {
1755                 if (!(icr & E1000_ICR_LSC))
1756                         goto no_link_interrupt;
1757                 hw->mac.get_link_status = 1;
1758                 /* guard against interrupt when we're going down */
1759                 if (!test_bit(__E1000_DOWN, &adapter->state))
1760                         mod_timer(&adapter->watchdog_timer, jiffies + 1);
1761         }
1762
1763 no_link_interrupt:
1764         if (!test_bit(__E1000_DOWN, &adapter->state))
1765                 ew32(IMS, E1000_IMS_LSC | E1000_IMS_OTHER);
1766
1767         return IRQ_HANDLED;
1768 }
1769
1770
1771 static irqreturn_t e1000_intr_msix_tx(int irq, void *data)
1772 {
1773         struct net_device *netdev = data;
1774         struct e1000_adapter *adapter = netdev_priv(netdev);
1775         struct e1000_hw *hw = &adapter->hw;
1776         struct e1000_ring *tx_ring = adapter->tx_ring;
1777
1778
1779         adapter->total_tx_bytes = 0;
1780         adapter->total_tx_packets = 0;
1781
1782         if (!e1000_clean_tx_irq(adapter))
1783                 /* Ring was not completely cleaned, so fire another interrupt */
1784                 ew32(ICS, tx_ring->ims_val);
1785
1786         return IRQ_HANDLED;
1787 }
1788
1789 static irqreturn_t e1000_intr_msix_rx(int irq, void *data)
1790 {
1791         struct net_device *netdev = data;
1792         struct e1000_adapter *adapter = netdev_priv(netdev);
1793
1794         /* Write the ITR value calculated at the end of the
1795          * previous interrupt.
1796          */
1797         if (adapter->rx_ring->set_itr) {
1798                 writel(1000000000 / (adapter->rx_ring->itr_val * 256),
1799                        adapter->hw.hw_addr + adapter->rx_ring->itr_register);
1800                 adapter->rx_ring->set_itr = 0;
1801         }
1802
1803         if (napi_schedule_prep(&adapter->napi)) {
1804                 adapter->total_rx_bytes = 0;
1805                 adapter->total_rx_packets = 0;
1806                 __napi_schedule(&adapter->napi);
1807         }
1808         return IRQ_HANDLED;
1809 }
1810
1811 /**
1812  * e1000_configure_msix - Configure MSI-X hardware
1813  *
1814  * e1000_configure_msix sets up the hardware to properly
1815  * generate MSI-X interrupts.
1816  **/
1817 static void e1000_configure_msix(struct e1000_adapter *adapter)
1818 {
1819         struct e1000_hw *hw = &adapter->hw;
1820         struct e1000_ring *rx_ring = adapter->rx_ring;
1821         struct e1000_ring *tx_ring = adapter->tx_ring;
1822         int vector = 0;
1823         u32 ctrl_ext, ivar = 0;
1824
1825         adapter->eiac_mask = 0;
1826
1827         /* Workaround issue with spurious interrupts on 82574 in MSI-X mode */
1828         if (hw->mac.type == e1000_82574) {
1829                 u32 rfctl = er32(RFCTL);
1830                 rfctl |= E1000_RFCTL_ACK_DIS;
1831                 ew32(RFCTL, rfctl);
1832         }
1833
1834 #define E1000_IVAR_INT_ALLOC_VALID      0x8
1835         /* Configure Rx vector */
1836         rx_ring->ims_val = E1000_IMS_RXQ0;
1837         adapter->eiac_mask |= rx_ring->ims_val;
1838         if (rx_ring->itr_val)
1839                 writel(1000000000 / (rx_ring->itr_val * 256),
1840                        hw->hw_addr + rx_ring->itr_register);
1841         else
1842                 writel(1, hw->hw_addr + rx_ring->itr_register);
1843         ivar = E1000_IVAR_INT_ALLOC_VALID | vector;
1844
1845         /* Configure Tx vector */
1846         tx_ring->ims_val = E1000_IMS_TXQ0;
1847         vector++;
1848         if (tx_ring->itr_val)
1849                 writel(1000000000 / (tx_ring->itr_val * 256),
1850                        hw->hw_addr + tx_ring->itr_register);
1851         else
1852                 writel(1, hw->hw_addr + tx_ring->itr_register);
1853         adapter->eiac_mask |= tx_ring->ims_val;
1854         ivar |= ((E1000_IVAR_INT_ALLOC_VALID | vector) << 8);
1855
1856         /* set vector for Other Causes, e.g. link changes */
1857         vector++;
1858         ivar |= ((E1000_IVAR_INT_ALLOC_VALID | vector) << 16);
1859         if (rx_ring->itr_val)
1860                 writel(1000000000 / (rx_ring->itr_val * 256),
1861                        hw->hw_addr + E1000_EITR_82574(vector));
1862         else
1863                 writel(1, hw->hw_addr + E1000_EITR_82574(vector));
1864
1865         /* Cause Tx interrupts on every write back */
1866         ivar |= (1 << 31);
1867
1868         ew32(IVAR, ivar);
1869
1870         /* enable MSI-X PBA support */
1871         ctrl_ext = er32(CTRL_EXT);
1872         ctrl_ext |= E1000_CTRL_EXT_PBA_CLR;
1873
1874         /* Auto-Mask Other interrupts upon ICR read */
1875 #define E1000_EIAC_MASK_82574   0x01F00000
1876         ew32(IAM, ~E1000_EIAC_MASK_82574 | E1000_IMS_OTHER);
1877         ctrl_ext |= E1000_CTRL_EXT_EIAME;
1878         ew32(CTRL_EXT, ctrl_ext);
1879         e1e_flush();
1880 }
1881
1882 void e1000e_reset_interrupt_capability(struct e1000_adapter *adapter)
1883 {
1884         if (adapter->msix_entries) {
1885                 pci_disable_msix(adapter->pdev);
1886                 kfree(adapter->msix_entries);
1887                 adapter->msix_entries = NULL;
1888         } else if (adapter->flags & FLAG_MSI_ENABLED) {
1889                 pci_disable_msi(adapter->pdev);
1890                 adapter->flags &= ~FLAG_MSI_ENABLED;
1891         }
1892 }
1893
1894 /**
1895  * e1000e_set_interrupt_capability - set MSI or MSI-X if supported
1896  *
1897  * Attempt to configure interrupts using the best available
1898  * capabilities of the hardware and kernel.
1899  **/
1900 void e1000e_set_interrupt_capability(struct e1000_adapter *adapter)
1901 {
1902         int err;
1903         int i;
1904
1905         switch (adapter->int_mode) {
1906         case E1000E_INT_MODE_MSIX:
1907                 if (adapter->flags & FLAG_HAS_MSIX) {
1908                         adapter->num_vectors = 3; /* RxQ0, TxQ0 and other */
1909                         adapter->msix_entries = kcalloc(adapter->num_vectors,
1910                                                       sizeof(struct msix_entry),
1911                                                       GFP_KERNEL);
1912                         if (adapter->msix_entries) {
1913                                 for (i = 0; i < adapter->num_vectors; i++)
1914                                         adapter->msix_entries[i].entry = i;
1915
1916                                 err = pci_enable_msix(adapter->pdev,
1917                                                       adapter->msix_entries,
1918                                                       adapter->num_vectors);
1919                                 if (err == 0)
1920                                         return;
1921                         }
1922                         /* MSI-X failed, so fall through and try MSI */
1923                         e_err("Failed to initialize MSI-X interrupts.  Falling back to MSI interrupts.\n");
1924                         e1000e_reset_interrupt_capability(adapter);
1925                 }
1926                 adapter->int_mode = E1000E_INT_MODE_MSI;
1927                 /* Fall through */
1928         case E1000E_INT_MODE_MSI:
1929                 if (!pci_enable_msi(adapter->pdev)) {
1930                         adapter->flags |= FLAG_MSI_ENABLED;
1931                 } else {
1932                         adapter->int_mode = E1000E_INT_MODE_LEGACY;
1933                         e_err("Failed to initialize MSI interrupts.  Falling back to legacy interrupts.\n");
1934                 }
1935                 /* Fall through */
1936         case E1000E_INT_MODE_LEGACY:
1937                 /* Don't do anything; this is the system default */
1938                 break;
1939         }
1940
1941         /* store the number of vectors being used */
1942         adapter->num_vectors = 1;
1943 }
1944
1945 /**
1946  * e1000_request_msix - Initialize MSI-X interrupts
1947  *
1948  * e1000_request_msix allocates MSI-X vectors and requests interrupts from the
1949  * kernel.
1950  **/
1951 static int e1000_request_msix(struct e1000_adapter *adapter)
1952 {
1953         struct net_device *netdev = adapter->netdev;
1954         int err = 0, vector = 0;
1955
1956         if (strlen(netdev->name) < (IFNAMSIZ - 5))
1957                 snprintf(adapter->rx_ring->name,
1958                          sizeof(adapter->rx_ring->name) - 1,
1959                          "%s-rx-0", netdev->name);
1960         else
1961                 memcpy(adapter->rx_ring->name, netdev->name, IFNAMSIZ);
1962         err = request_irq(adapter->msix_entries[vector].vector,
1963                           e1000_intr_msix_rx, 0, adapter->rx_ring->name,
1964                           netdev);
1965         if (err)
1966                 goto out;
1967         adapter->rx_ring->itr_register = E1000_EITR_82574(vector);
1968         adapter->rx_ring->itr_val = adapter->itr;
1969         vector++;
1970
1971         if (strlen(netdev->name) < (IFNAMSIZ - 5))
1972                 snprintf(adapter->tx_ring->name,
1973                          sizeof(adapter->tx_ring->name) - 1,
1974                          "%s-tx-0", netdev->name);
1975         else
1976                 memcpy(adapter->tx_ring->name, netdev->name, IFNAMSIZ);
1977         err = request_irq(adapter->msix_entries[vector].vector,
1978                           e1000_intr_msix_tx, 0, adapter->tx_ring->name,
1979                           netdev);
1980         if (err)
1981                 goto out;
1982         adapter->tx_ring->itr_register = E1000_EITR_82574(vector);
1983         adapter->tx_ring->itr_val = adapter->itr;
1984         vector++;
1985
1986         err = request_irq(adapter->msix_entries[vector].vector,
1987                           e1000_msix_other, 0, netdev->name, netdev);
1988         if (err)
1989                 goto out;
1990
1991         e1000_configure_msix(adapter);
1992         return 0;
1993 out:
1994         return err;
1995 }
1996
1997 /**
1998  * e1000_request_irq - initialize interrupts
1999  *
2000  * Attempts to configure interrupts using the best available
2001  * capabilities of the hardware and kernel.
2002  **/
2003 static int e1000_request_irq(struct e1000_adapter *adapter)
2004 {
2005         struct net_device *netdev = adapter->netdev;
2006         int err;
2007
2008         if (adapter->msix_entries) {
2009                 err = e1000_request_msix(adapter);
2010                 if (!err)
2011                         return err;
2012                 /* fall back to MSI */
2013                 e1000e_reset_interrupt_capability(adapter);
2014                 adapter->int_mode = E1000E_INT_MODE_MSI;
2015                 e1000e_set_interrupt_capability(adapter);
2016         }
2017         if (adapter->flags & FLAG_MSI_ENABLED) {
2018                 err = request_irq(adapter->pdev->irq, e1000_intr_msi, 0,
2019                                   netdev->name, netdev);
2020                 if (!err)
2021                         return err;
2022
2023                 /* fall back to legacy interrupt */
2024                 e1000e_reset_interrupt_capability(adapter);
2025                 adapter->int_mode = E1000E_INT_MODE_LEGACY;
2026         }
2027
2028         err = request_irq(adapter->pdev->irq, e1000_intr, IRQF_SHARED,
2029                           netdev->name, netdev);
2030         if (err)
2031                 e_err("Unable to allocate interrupt, Error: %d\n", err);
2032
2033         return err;
2034 }
2035
2036 static void e1000_free_irq(struct e1000_adapter *adapter)
2037 {
2038         struct net_device *netdev = adapter->netdev;
2039
2040         if (adapter->msix_entries) {
2041                 int vector = 0;
2042
2043                 free_irq(adapter->msix_entries[vector].vector, netdev);
2044                 vector++;
2045
2046                 free_irq(adapter->msix_entries[vector].vector, netdev);
2047                 vector++;
2048
2049                 /* Other Causes interrupt vector */
2050                 free_irq(adapter->msix_entries[vector].vector, netdev);
2051                 return;
2052         }
2053
2054         free_irq(adapter->pdev->irq, netdev);
2055 }
2056
2057 /**
2058  * e1000_irq_disable - Mask off interrupt generation on the NIC
2059  **/
2060 static void e1000_irq_disable(struct e1000_adapter *adapter)
2061 {
2062         struct e1000_hw *hw = &adapter->hw;
2063
2064         ew32(IMC, ~0);
2065         if (adapter->msix_entries)
2066                 ew32(EIAC_82574, 0);
2067         e1e_flush();
2068
2069         if (adapter->msix_entries) {
2070                 int i;
2071                 for (i = 0; i < adapter->num_vectors; i++)
2072                         synchronize_irq(adapter->msix_entries[i].vector);
2073         } else {
2074                 synchronize_irq(adapter->pdev->irq);
2075         }
2076 }
2077
2078 /**
2079  * e1000_irq_enable - Enable default interrupt generation settings
2080  **/
2081 static void e1000_irq_enable(struct e1000_adapter *adapter)
2082 {
2083         struct e1000_hw *hw = &adapter->hw;
2084
2085         if (adapter->msix_entries) {
2086                 ew32(EIAC_82574, adapter->eiac_mask & E1000_EIAC_MASK_82574);
2087                 ew32(IMS, adapter->eiac_mask | E1000_IMS_OTHER | E1000_IMS_LSC);
2088         } else {
2089                 ew32(IMS, IMS_ENABLE_MASK);
2090         }
2091         e1e_flush();
2092 }
2093
2094 /**
2095  * e1000e_get_hw_control - get control of the h/w from f/w
2096  * @adapter: address of board private structure
2097  *
2098  * e1000e_get_hw_control sets {CTRL_EXT|SWSM}:DRV_LOAD bit.
2099  * For ASF and Pass Through versions of f/w this means that
2100  * the driver is loaded. For AMT version (only with 82573)
2101  * of the f/w this means that the network i/f is open.
2102  **/
2103 void e1000e_get_hw_control(struct e1000_adapter *adapter)
2104 {
2105         struct e1000_hw *hw = &adapter->hw;
2106         u32 ctrl_ext;
2107         u32 swsm;
2108
2109         /* Let firmware know the driver has taken over */
2110         if (adapter->flags & FLAG_HAS_SWSM_ON_LOAD) {
2111                 swsm = er32(SWSM);
2112                 ew32(SWSM, swsm | E1000_SWSM_DRV_LOAD);
2113         } else if (adapter->flags & FLAG_HAS_CTRLEXT_ON_LOAD) {
2114                 ctrl_ext = er32(CTRL_EXT);
2115                 ew32(CTRL_EXT, ctrl_ext | E1000_CTRL_EXT_DRV_LOAD);
2116         }
2117 }
2118
2119 /**
2120  * e1000e_release_hw_control - release control of the h/w to f/w
2121  * @adapter: address of board private structure
2122  *
2123  * e1000e_release_hw_control resets {CTRL_EXT|SWSM}:DRV_LOAD bit.
2124  * For ASF and Pass Through versions of f/w this means that the
2125  * driver is no longer loaded. For AMT version (only with 82573) i
2126  * of the f/w this means that the network i/f is closed.
2127  *
2128  **/
2129 void e1000e_release_hw_control(struct e1000_adapter *adapter)
2130 {
2131         struct e1000_hw *hw = &adapter->hw;
2132         u32 ctrl_ext;
2133         u32 swsm;
2134
2135         /* Let firmware taken over control of h/w */
2136         if (adapter->flags & FLAG_HAS_SWSM_ON_LOAD) {
2137                 swsm = er32(SWSM);
2138                 ew32(SWSM, swsm & ~E1000_SWSM_DRV_LOAD);
2139         } else if (adapter->flags & FLAG_HAS_CTRLEXT_ON_LOAD) {
2140                 ctrl_ext = er32(CTRL_EXT);
2141                 ew32(CTRL_EXT, ctrl_ext & ~E1000_CTRL_EXT_DRV_LOAD);
2142         }
2143 }
2144
2145 /**
2146  * @e1000_alloc_ring - allocate memory for a ring structure
2147  **/
2148 static int e1000_alloc_ring_dma(struct e1000_adapter *adapter,
2149                                 struct e1000_ring *ring)
2150 {
2151         struct pci_dev *pdev = adapter->pdev;
2152
2153         ring->desc = dma_alloc_coherent(&pdev->dev, ring->size, &ring->dma,
2154                                         GFP_KERNEL);
2155         if (!ring->desc)
2156                 return -ENOMEM;
2157
2158         return 0;
2159 }
2160
2161 /**
2162  * e1000e_setup_tx_resources - allocate Tx resources (Descriptors)
2163  * @adapter: board private structure
2164  *
2165  * Return 0 on success, negative on failure
2166  **/
2167 int e1000e_setup_tx_resources(struct e1000_adapter *adapter)
2168 {
2169         struct e1000_ring *tx_ring = adapter->tx_ring;
2170         int err = -ENOMEM, size;
2171
2172         size = sizeof(struct e1000_buffer) * tx_ring->count;
2173         tx_ring->buffer_info = vzalloc(size);
2174         if (!tx_ring->buffer_info)
2175                 goto err;
2176
2177         /* round up to nearest 4K */
2178         tx_ring->size = tx_ring->count * sizeof(struct e1000_tx_desc);
2179         tx_ring->size = ALIGN(tx_ring->size, 4096);
2180
2181         err = e1000_alloc_ring_dma(adapter, tx_ring);
2182         if (err)
2183                 goto err;
2184
2185         tx_ring->next_to_use = 0;
2186         tx_ring->next_to_clean = 0;
2187
2188         return 0;
2189 err:
2190         vfree(tx_ring->buffer_info);
2191         e_err("Unable to allocate memory for the transmit descriptor ring\n");
2192         return err;
2193 }
2194
2195 /**
2196  * e1000e_setup_rx_resources - allocate Rx resources (Descriptors)
2197  * @adapter: board private structure
2198  *
2199  * Returns 0 on success, negative on failure
2200  **/
2201 int e1000e_setup_rx_resources(struct e1000_adapter *adapter)
2202 {
2203         struct e1000_ring *rx_ring = adapter->rx_ring;
2204         struct e1000_buffer *buffer_info;
2205         int i, size, desc_len, err = -ENOMEM;
2206
2207         size = sizeof(struct e1000_buffer) * rx_ring->count;
2208         rx_ring->buffer_info = vzalloc(size);
2209         if (!rx_ring->buffer_info)
2210                 goto err;
2211
2212         for (i = 0; i < rx_ring->count; i++) {
2213                 buffer_info = &rx_ring->buffer_info[i];
2214                 buffer_info->ps_pages = kcalloc(PS_PAGE_BUFFERS,
2215                                                 sizeof(struct e1000_ps_page),
2216                                                 GFP_KERNEL);
2217                 if (!buffer_info->ps_pages)
2218                         goto err_pages;
2219         }
2220
2221         desc_len = sizeof(union e1000_rx_desc_packet_split);
2222
2223         /* Round up to nearest 4K */
2224         rx_ring->size = rx_ring->count * desc_len;
2225         rx_ring->size = ALIGN(rx_ring->size, 4096);
2226
2227         err = e1000_alloc_ring_dma(adapter, rx_ring);
2228         if (err)
2229                 goto err_pages;
2230
2231         rx_ring->next_to_clean = 0;
2232         rx_ring->next_to_use = 0;
2233         rx_ring->rx_skb_top = NULL;
2234
2235         return 0;
2236
2237 err_pages:
2238         for (i = 0; i < rx_ring->count; i++) {
2239                 buffer_info = &rx_ring->buffer_info[i];
2240                 kfree(buffer_info->ps_pages);
2241         }
2242 err:
2243         vfree(rx_ring->buffer_info);
2244         e_err("Unable to allocate memory for the receive descriptor ring\n");
2245         return err;
2246 }
2247
2248 /**
2249  * e1000_clean_tx_ring - Free Tx Buffers
2250  * @adapter: board private structure
2251  **/
2252 static void e1000_clean_tx_ring(struct e1000_adapter *adapter)
2253 {
2254         struct e1000_ring *tx_ring = adapter->tx_ring;
2255         struct e1000_buffer *buffer_info;
2256         unsigned long size;
2257         unsigned int i;
2258
2259         for (i = 0; i < tx_ring->count; i++) {
2260                 buffer_info = &tx_ring->buffer_info[i];
2261                 e1000_put_txbuf(adapter, buffer_info);
2262         }
2263
2264         netdev_reset_queue(adapter->netdev);
2265         size = sizeof(struct e1000_buffer) * tx_ring->count;
2266         memset(tx_ring->buffer_info, 0, size);
2267
2268         memset(tx_ring->desc, 0, tx_ring->size);
2269
2270         tx_ring->next_to_use = 0;
2271         tx_ring->next_to_clean = 0;
2272
2273         writel(0, adapter->hw.hw_addr + tx_ring->head);
2274         writel(0, adapter->hw.hw_addr + tx_ring->tail);
2275 }
2276
2277 /**
2278  * e1000e_free_tx_resources - Free Tx Resources per Queue
2279  * @adapter: board private structure
2280  *
2281  * Free all transmit software resources
2282  **/
2283 void e1000e_free_tx_resources(struct e1000_adapter *adapter)
2284 {
2285         struct pci_dev *pdev = adapter->pdev;
2286         struct e1000_ring *tx_ring = adapter->tx_ring;
2287
2288         e1000_clean_tx_ring(adapter);
2289
2290         vfree(tx_ring->buffer_info);
2291         tx_ring->buffer_info = NULL;
2292
2293         dma_free_coherent(&pdev->dev, tx_ring->size, tx_ring->desc,
2294                           tx_ring->dma);
2295         tx_ring->desc = NULL;
2296 }
2297
2298 /**
2299  * e1000e_free_rx_resources - Free Rx Resources
2300  * @adapter: board private structure
2301  *
2302  * Free all receive software resources
2303  **/
2304
2305 void e1000e_free_rx_resources(struct e1000_adapter *adapter)
2306 {
2307         struct pci_dev *pdev = adapter->pdev;
2308         struct e1000_ring *rx_ring = adapter->rx_ring;
2309         int i;
2310
2311         e1000_clean_rx_ring(adapter);
2312
2313         for (i = 0; i < rx_ring->count; i++)
2314                 kfree(rx_ring->buffer_info[i].ps_pages);
2315
2316         vfree(rx_ring->buffer_info);
2317         rx_ring->buffer_info = NULL;
2318
2319         dma_free_coherent(&pdev->dev, rx_ring->size, rx_ring->desc,
2320                           rx_ring->dma);
2321         rx_ring->desc = NULL;
2322 }
2323
2324 /**
2325  * e1000_update_itr - update the dynamic ITR value based on statistics
2326  * @adapter: pointer to adapter
2327  * @itr_setting: current adapter->itr
2328  * @packets: the number of packets during this measurement interval
2329  * @bytes: the number of bytes during this measurement interval
2330  *
2331  *      Stores a new ITR value based on packets and byte
2332  *      counts during the last interrupt.  The advantage of per interrupt
2333  *      computation is faster updates and more accurate ITR for the current
2334  *      traffic pattern.  Constants in this function were computed
2335  *      based on theoretical maximum wire speed and thresholds were set based
2336  *      on testing data as well as attempting to minimize response time
2337  *      while increasing bulk throughput.  This functionality is controlled
2338  *      by the InterruptThrottleRate module parameter.
2339  **/
2340 static unsigned int e1000_update_itr(struct e1000_adapter *adapter,
2341                                      u16 itr_setting, int packets,
2342                                      int bytes)
2343 {
2344         unsigned int retval = itr_setting;
2345
2346         if (packets == 0)
2347                 goto update_itr_done;
2348
2349         switch (itr_setting) {
2350         case lowest_latency:
2351                 /* handle TSO and jumbo frames */
2352                 if (bytes/packets > 8000)
2353                         retval = bulk_latency;
2354                 else if ((packets < 5) && (bytes > 512))
2355                         retval = low_latency;
2356                 break;
2357         case low_latency:  /* 50 usec aka 20000 ints/s */
2358                 if (bytes > 10000) {
2359                         /* this if handles the TSO accounting */
2360                         if (bytes/packets > 8000)
2361                                 retval = bulk_latency;
2362                         else if ((packets < 10) || ((bytes/packets) > 1200))
2363                                 retval = bulk_latency;
2364                         else if ((packets > 35))
2365                                 retval = lowest_latency;
2366                 } else if (bytes/packets > 2000) {
2367                         retval = bulk_latency;
2368                 } else if (packets <= 2 && bytes < 512) {
2369                         retval = lowest_latency;
2370                 }
2371                 break;
2372         case bulk_latency: /* 250 usec aka 4000 ints/s */
2373                 if (bytes > 25000) {
2374                         if (packets > 35)
2375                                 retval = low_latency;
2376                 } else if (bytes < 6000) {
2377                         retval = low_latency;
2378                 }
2379                 break;
2380         }
2381
2382 update_itr_done:
2383         return retval;
2384 }
2385
2386 static void e1000_set_itr(struct e1000_adapter *adapter)
2387 {
2388         struct e1000_hw *hw = &adapter->hw;
2389         u16 current_itr;
2390         u32 new_itr = adapter->itr;
2391
2392         /* for non-gigabit speeds, just fix the interrupt rate at 4000 */
2393         if (adapter->link_speed != SPEED_1000) {
2394                 current_itr = 0;
2395                 new_itr = 4000;
2396                 goto set_itr_now;
2397         }
2398
2399         if (adapter->flags2 & FLAG2_DISABLE_AIM) {
2400                 new_itr = 0;
2401                 goto set_itr_now;
2402         }
2403
2404         adapter->tx_itr = e1000_update_itr(adapter,
2405                                     adapter->tx_itr,
2406                                     adapter->total_tx_packets,
2407                                     adapter->total_tx_bytes);
2408         /* conservative mode (itr 3) eliminates the lowest_latency setting */
2409         if (adapter->itr_setting == 3 && adapter->tx_itr == lowest_latency)
2410                 adapter->tx_itr = low_latency;
2411
2412         adapter->rx_itr = e1000_update_itr(adapter,
2413                                     adapter->rx_itr,
2414                                     adapter->total_rx_packets,
2415                                     adapter->total_rx_bytes);
2416         /* conservative mode (itr 3) eliminates the lowest_latency setting */
2417         if (adapter->itr_setting == 3 && adapter->rx_itr == lowest_latency)
2418                 adapter->rx_itr = low_latency;
2419
2420         current_itr = max(adapter->rx_itr, adapter->tx_itr);
2421
2422         switch (current_itr) {
2423         /* counts and packets in update_itr are dependent on these numbers */
2424         case lowest_latency:
2425                 new_itr = 70000;
2426                 break;
2427         case low_latency:
2428                 new_itr = 20000; /* aka hwitr = ~200 */
2429                 break;
2430         case bulk_latency:
2431                 new_itr = 4000;
2432                 break;
2433         default:
2434                 break;
2435         }
2436
2437 set_itr_now:
2438         if (new_itr != adapter->itr) {
2439                 /*
2440                  * this attempts to bias the interrupt rate towards Bulk
2441                  * by adding intermediate steps when interrupt rate is
2442                  * increasing
2443                  */
2444                 new_itr = new_itr > adapter->itr ?
2445                              min(adapter->itr + (new_itr >> 2), new_itr) :
2446                              new_itr;
2447                 adapter->itr = new_itr;
2448                 adapter->rx_ring->itr_val = new_itr;
2449                 if (adapter->msix_entries)
2450                         adapter->rx_ring->set_itr = 1;
2451                 else
2452                         if (new_itr)
2453                                 ew32(ITR, 1000000000 / (new_itr * 256));
2454                         else
2455                                 ew32(ITR, 0);
2456         }
2457 }
2458
2459 /**
2460  * e1000_alloc_queues - Allocate memory for all rings
2461  * @adapter: board private structure to initialize
2462  **/
2463 static int __devinit e1000_alloc_queues(struct e1000_adapter *adapter)
2464 {
2465         adapter->tx_ring = kzalloc(sizeof(struct e1000_ring), GFP_KERNEL);
2466         if (!adapter->tx_ring)
2467                 goto err;
2468
2469         adapter->rx_ring = kzalloc(sizeof(struct e1000_ring), GFP_KERNEL);
2470         if (!adapter->rx_ring)
2471                 goto err;
2472
2473         return 0;
2474 err:
2475         e_err("Unable to allocate memory for queues\n");
2476         kfree(adapter->rx_ring);
2477         kfree(adapter->tx_ring);
2478         return -ENOMEM;
2479 }
2480
2481 /**
2482  * e1000_clean - NAPI Rx polling callback
2483  * @napi: struct associated with this polling callback
2484  * @budget: amount of packets driver is allowed to process this poll
2485  **/
2486 static int e1000_clean(struct napi_struct *napi, int budget)
2487 {
2488         struct e1000_adapter *adapter = container_of(napi, struct e1000_adapter, napi);
2489         struct e1000_hw *hw = &adapter->hw;
2490         struct net_device *poll_dev = adapter->netdev;
2491         int tx_cleaned = 1, work_done = 0;
2492
2493         adapter = netdev_priv(poll_dev);
2494
2495         if (adapter->msix_entries &&
2496             !(adapter->rx_ring->ims_val & adapter->tx_ring->ims_val))
2497                 goto clean_rx;
2498
2499         tx_cleaned = e1000_clean_tx_irq(adapter);
2500
2501 clean_rx:
2502         adapter->clean_rx(adapter, &work_done, budget);
2503
2504         if (!tx_cleaned)
2505                 work_done = budget;
2506
2507         /* If budget not fully consumed, exit the polling mode */
2508         if (work_done < budget) {
2509                 if (adapter->itr_setting & 3)
2510                         e1000_set_itr(adapter);
2511                 napi_complete(napi);
2512                 if (!test_bit(__E1000_DOWN, &adapter->state)) {
2513                         if (adapter->msix_entries)
2514                                 ew32(IMS, adapter->rx_ring->ims_val);
2515                         else
2516                                 e1000_irq_enable(adapter);
2517                 }
2518         }
2519
2520         return work_done;
2521 }
2522
2523 static int e1000_vlan_rx_add_vid(struct net_device *netdev, u16 vid)
2524 {
2525         struct e1000_adapter *adapter = netdev_priv(netdev);
2526         struct e1000_hw *hw = &adapter->hw;
2527         u32 vfta, index;
2528
2529         /* don't update vlan cookie if already programmed */
2530         if ((adapter->hw.mng_cookie.status &
2531              E1000_MNG_DHCP_COOKIE_STATUS_VLAN) &&
2532             (vid == adapter->mng_vlan_id))
2533                 return 0;
2534
2535         /* add VID to filter table */
2536         if (adapter->flags & FLAG_HAS_HW_VLAN_FILTER) {
2537                 index = (vid >> 5) & 0x7F;
2538                 vfta = E1000_READ_REG_ARRAY(hw, E1000_VFTA, index);
2539                 vfta |= (1 << (vid & 0x1F));
2540                 hw->mac.ops.write_vfta(hw, index, vfta);
2541         }
2542
2543         set_bit(vid, adapter->active_vlans);
2544
2545         return 0;
2546 }
2547
2548 static int e1000_vlan_rx_kill_vid(struct net_device *netdev, u16 vid)
2549 {
2550         struct e1000_adapter *adapter = netdev_priv(netdev);
2551         struct e1000_hw *hw = &adapter->hw;
2552         u32 vfta, index;
2553
2554         if ((adapter->hw.mng_cookie.status &
2555              E1000_MNG_DHCP_COOKIE_STATUS_VLAN) &&
2556             (vid == adapter->mng_vlan_id)) {
2557                 /* release control to f/w */
2558                 e1000e_release_hw_control(adapter);
2559                 return 0;
2560         }
2561
2562         /* remove VID from filter table */
2563         if (adapter->flags & FLAG_HAS_HW_VLAN_FILTER) {
2564                 index = (vid >> 5) & 0x7F;
2565                 vfta = E1000_READ_REG_ARRAY(hw, E1000_VFTA, index);
2566                 vfta &= ~(1 << (vid & 0x1F));
2567                 hw->mac.ops.write_vfta(hw, index, vfta);
2568         }
2569
2570         clear_bit(vid, adapter->active_vlans);
2571
2572         return 0;
2573 }
2574
2575 /**
2576  * e1000e_vlan_filter_disable - helper to disable hw VLAN filtering
2577  * @adapter: board private structure to initialize
2578  **/
2579 static void e1000e_vlan_filter_disable(struct e1000_adapter *adapter)
2580 {
2581         struct net_device *netdev = adapter->netdev;
2582         struct e1000_hw *hw = &adapter->hw;
2583         u32 rctl;
2584
2585         if (adapter->flags & FLAG_HAS_HW_VLAN_FILTER) {
2586                 /* disable VLAN receive filtering */
2587                 rctl = er32(RCTL);
2588                 rctl &= ~(E1000_RCTL_VFE | E1000_RCTL_CFIEN);
2589                 ew32(RCTL, rctl);
2590
2591                 if (adapter->mng_vlan_id != (u16)E1000_MNG_VLAN_NONE) {
2592                         e1000_vlan_rx_kill_vid(netdev, adapter->mng_vlan_id);
2593                         adapter->mng_vlan_id = E1000_MNG_VLAN_NONE;
2594                 }
2595         }
2596 }
2597
2598 /**
2599  * e1000e_vlan_filter_enable - helper to enable HW VLAN filtering
2600  * @adapter: board private structure to initialize
2601  **/
2602 static void e1000e_vlan_filter_enable(struct e1000_adapter *adapter)
2603 {
2604         struct e1000_hw *hw = &adapter->hw;
2605         u32 rctl;
2606
2607         if (adapter->flags & FLAG_HAS_HW_VLAN_FILTER) {
2608                 /* enable VLAN receive filtering */
2609                 rctl = er32(RCTL);
2610                 rctl |= E1000_RCTL_VFE;
2611                 rctl &= ~E1000_RCTL_CFIEN;
2612                 ew32(RCTL, rctl);
2613         }
2614 }
2615
2616 /**
2617  * e1000e_vlan_strip_enable - helper to disable HW VLAN stripping
2618  * @adapter: board private structure to initialize
2619  **/
2620 static void e1000e_vlan_strip_disable(struct e1000_adapter *adapter)
2621 {
2622         struct e1000_hw *hw = &adapter->hw;
2623         u32 ctrl;
2624
2625         /* disable VLAN tag insert/strip */
2626         ctrl = er32(CTRL);
2627         ctrl &= ~E1000_CTRL_VME;
2628         ew32(CTRL, ctrl);
2629 }
2630
2631 /**
2632  * e1000e_vlan_strip_enable - helper to enable HW VLAN stripping
2633  * @adapter: board private structure to initialize
2634  **/
2635 static void e1000e_vlan_strip_enable(struct e1000_adapter *adapter)
2636 {
2637         struct e1000_hw *hw = &adapter->hw;
2638         u32 ctrl;
2639
2640         /* enable VLAN tag insert/strip */
2641         ctrl = er32(CTRL);
2642         ctrl |= E1000_CTRL_VME;
2643         ew32(CTRL, ctrl);
2644 }
2645
2646 static void e1000_update_mng_vlan(struct e1000_adapter *adapter)
2647 {
2648         struct net_device *netdev = adapter->netdev;
2649         u16 vid = adapter->hw.mng_cookie.vlan_id;
2650         u16 old_vid = adapter->mng_vlan_id;
2651
2652         if (adapter->hw.mng_cookie.status &
2653             E1000_MNG_DHCP_COOKIE_STATUS_VLAN) {
2654                 e1000_vlan_rx_add_vid(netdev, vid);
2655                 adapter->mng_vlan_id = vid;
2656         }
2657
2658         if ((old_vid != (u16)E1000_MNG_VLAN_NONE) && (vid != old_vid))
2659                 e1000_vlan_rx_kill_vid(netdev, old_vid);
2660 }
2661
2662 static void e1000_restore_vlan(struct e1000_adapter *adapter)
2663 {
2664         u16 vid;
2665
2666         e1000_vlan_rx_add_vid(adapter->netdev, 0);
2667
2668         for_each_set_bit(vid, adapter->active_vlans, VLAN_N_VID)
2669                 e1000_vlan_rx_add_vid(adapter->netdev, vid);
2670 }
2671
2672 static void e1000_init_manageability_pt(struct e1000_adapter *adapter)
2673 {
2674         struct e1000_hw *hw = &adapter->hw;
2675         u32 manc, manc2h, mdef, i, j;
2676
2677         if (!(adapter->flags & FLAG_MNG_PT_ENABLED))
2678                 return;
2679
2680         manc = er32(MANC);
2681
2682         /*
2683          * enable receiving management packets to the host. this will probably
2684          * generate destination unreachable messages from the host OS, but
2685          * the packets will be handled on SMBUS
2686          */
2687         manc |= E1000_MANC_EN_MNG2HOST;
2688         manc2h = er32(MANC2H);
2689
2690         switch (hw->mac.type) {
2691         default:
2692                 manc2h |= (E1000_MANC2H_PORT_623 | E1000_MANC2H_PORT_664);
2693                 break;
2694         case e1000_82574:
2695         case e1000_82583:
2696                 /*
2697                  * Check if IPMI pass-through decision filter already exists;
2698                  * if so, enable it.
2699                  */
2700                 for (i = 0, j = 0; i < 8; i++) {
2701                         mdef = er32(MDEF(i));
2702
2703                         /* Ignore filters with anything other than IPMI ports */
2704                         if (mdef & ~(E1000_MDEF_PORT_623 | E1000_MDEF_PORT_664))
2705                                 continue;
2706
2707                         /* Enable this decision filter in MANC2H */
2708                         if (mdef)
2709                                 manc2h |= (1 << i);
2710
2711                         j |= mdef;
2712                 }
2713
2714                 if (j == (E1000_MDEF_PORT_623 | E1000_MDEF_PORT_664))
2715                         break;
2716
2717                 /* Create new decision filter in an empty filter */
2718                 for (i = 0, j = 0; i < 8; i++)
2719                         if (er32(MDEF(i)) == 0) {
2720                                 ew32(MDEF(i), (E1000_MDEF_PORT_623 |
2721                                                E1000_MDEF_PORT_664));
2722                                 manc2h |= (1 << 1);
2723                                 j++;
2724                                 break;
2725                         }
2726
2727                 if (!j)
2728                         e_warn("Unable to create IPMI pass-through filter\n");
2729                 break;
2730         }
2731
2732         ew32(MANC2H, manc2h);
2733         ew32(MANC, manc);
2734 }
2735
2736 /**
2737  * e1000_configure_tx - Configure Transmit Unit after Reset
2738  * @adapter: board private structure
2739  *
2740  * Configure the Tx unit of the MAC after a reset.
2741  **/
2742 static void e1000_configure_tx(struct e1000_adapter *adapter)
2743 {
2744         struct e1000_hw *hw = &adapter->hw;
2745         struct e1000_ring *tx_ring = adapter->tx_ring;
2746         u64 tdba;
2747         u32 tdlen, tctl, tipg, tarc;
2748         u32 ipgr1, ipgr2;
2749
2750         /* Setup the HW Tx Head and Tail descriptor pointers */
2751         tdba = tx_ring->dma;
2752         tdlen = tx_ring->count * sizeof(struct e1000_tx_desc);
2753         ew32(TDBAL, (tdba & DMA_BIT_MASK(32)));
2754         ew32(TDBAH, (tdba >> 32));
2755         ew32(TDLEN, tdlen);
2756         ew32(TDH, 0);
2757         ew32(TDT, 0);
2758         tx_ring->head = E1000_TDH;
2759         tx_ring->tail = E1000_TDT;
2760
2761         /* Set the default values for the Tx Inter Packet Gap timer */
2762         tipg = DEFAULT_82543_TIPG_IPGT_COPPER;          /*  8  */
2763         ipgr1 = DEFAULT_82543_TIPG_IPGR1;               /*  8  */
2764         ipgr2 = DEFAULT_82543_TIPG_IPGR2;               /*  6  */
2765
2766         if (adapter->flags & FLAG_TIPG_MEDIUM_FOR_80003ESLAN)
2767                 ipgr2 = DEFAULT_80003ES2LAN_TIPG_IPGR2; /*  7  */
2768
2769         tipg |= ipgr1 << E1000_TIPG_IPGR1_SHIFT;
2770         tipg |= ipgr2 << E1000_TIPG_IPGR2_SHIFT;
2771         ew32(TIPG, tipg);
2772
2773         /* Set the Tx Interrupt Delay register */
2774         ew32(TIDV, adapter->tx_int_delay);
2775         /* Tx irq moderation */
2776         ew32(TADV, adapter->tx_abs_int_delay);
2777
2778         if (adapter->flags2 & FLAG2_DMA_BURST) {
2779                 u32 txdctl = er32(TXDCTL(0));
2780                 txdctl &= ~(E1000_TXDCTL_PTHRESH | E1000_TXDCTL_HTHRESH |
2781                             E1000_TXDCTL_WTHRESH);
2782                 /*
2783                  * set up some performance related parameters to encourage the
2784                  * hardware to use the bus more efficiently in bursts, depends
2785                  * on the tx_int_delay to be enabled,
2786                  * wthresh = 5 ==> burst write a cacheline (64 bytes) at a time
2787                  * hthresh = 1 ==> prefetch when one or more available
2788                  * pthresh = 0x1f ==> prefetch if internal cache 31 or less
2789                  * BEWARE: this seems to work but should be considered first if
2790                  * there are Tx hangs or other Tx related bugs
2791                  */
2792                 txdctl |= E1000_TXDCTL_DMA_BURST_ENABLE;
2793                 ew32(TXDCTL(0), txdctl);
2794                 /* erratum work around: set txdctl the same for both queues */
2795                 ew32(TXDCTL(1), txdctl);
2796         }
2797
2798         /* Program the Transmit Control Register */
2799         tctl = er32(TCTL);
2800         tctl &= ~E1000_TCTL_CT;
2801         tctl |= E1000_TCTL_PSP | E1000_TCTL_RTLC |
2802                 (E1000_COLLISION_THRESHOLD << E1000_CT_SHIFT);
2803
2804         if (adapter->flags & FLAG_TARC_SPEED_MODE_BIT) {
2805                 tarc = er32(TARC(0));
2806                 /*
2807                  * set the speed mode bit, we'll clear it if we're not at
2808                  * gigabit link later
2809                  */
2810 #define SPEED_MODE_BIT (1 << 21)
2811                 tarc |= SPEED_MODE_BIT;
2812                 ew32(TARC(0), tarc);
2813         }
2814
2815         /* errata: program both queues to unweighted RR */
2816         if (adapter->flags & FLAG_TARC_SET_BIT_ZERO) {
2817                 tarc = er32(TARC(0));
2818                 tarc |= 1;
2819                 ew32(TARC(0), tarc);
2820                 tarc = er32(TARC(1));
2821                 tarc |= 1;
2822                 ew32(TARC(1), tarc);
2823         }
2824
2825         /* Setup Transmit Descriptor Settings for eop descriptor */
2826         adapter->txd_cmd = E1000_TXD_CMD_EOP | E1000_TXD_CMD_IFCS;
2827
2828         /* only set IDE if we are delaying interrupts using the timers */
2829         if (adapter->tx_int_delay)
2830                 adapter->txd_cmd |= E1000_TXD_CMD_IDE;
2831
2832         /* enable Report Status bit */
2833         adapter->txd_cmd |= E1000_TXD_CMD_RS;
2834
2835         ew32(TCTL, tctl);
2836
2837         e1000e_config_collision_dist(hw);
2838 }
2839
2840 /**
2841  * e1000_setup_rctl - configure the receive control registers
2842  * @adapter: Board private structure
2843  **/
2844 #define PAGE_USE_COUNT(S) (((S) >> PAGE_SHIFT) + \
2845                            (((S) & (PAGE_SIZE - 1)) ? 1 : 0))
2846 static void e1000_setup_rctl(struct e1000_adapter *adapter)
2847 {
2848         struct e1000_hw *hw = &adapter->hw;
2849         u32 rctl, rfctl;
2850         u32 pages = 0;
2851
2852         /* Workaround Si errata on 82579 - configure jumbo frame flow */
2853         if (hw->mac.type == e1000_pch2lan) {
2854                 s32 ret_val;
2855
2856                 if (adapter->netdev->mtu > ETH_DATA_LEN)
2857                         ret_val = e1000_lv_jumbo_workaround_ich8lan(hw, true);
2858                 else
2859                         ret_val = e1000_lv_jumbo_workaround_ich8lan(hw, false);
2860
2861                 if (ret_val)
2862                         e_dbg("failed to enable jumbo frame workaround mode\n");
2863         }
2864
2865         /* Program MC offset vector base */
2866         rctl = er32(RCTL);
2867         rctl &= ~(3 << E1000_RCTL_MO_SHIFT);
2868         rctl |= E1000_RCTL_EN | E1000_RCTL_BAM |
2869                 E1000_RCTL_LBM_NO | E1000_RCTL_RDMTS_HALF |
2870                 (adapter->hw.mac.mc_filter_type << E1000_RCTL_MO_SHIFT);
2871
2872         /* Do not Store bad packets */
2873         rctl &= ~E1000_RCTL_SBP;
2874
2875         /* Enable Long Packet receive */
2876         if (adapter->netdev->mtu <= ETH_DATA_LEN)
2877                 rctl &= ~E1000_RCTL_LPE;
2878         else
2879                 rctl |= E1000_RCTL_LPE;
2880
2881         /* Some systems expect that the CRC is included in SMBUS traffic. The
2882          * hardware strips the CRC before sending to both SMBUS (BMC) and to
2883          * host memory when this is enabled
2884          */
2885         if (adapter->flags2 & FLAG2_CRC_STRIPPING)
2886                 rctl |= E1000_RCTL_SECRC;
2887
2888         /* Workaround Si errata on 82577 PHY - configure IPG for jumbos */
2889         if ((hw->phy.type == e1000_phy_82577) && (rctl & E1000_RCTL_LPE)) {
2890                 u16 phy_data;
2891
2892                 e1e_rphy(hw, PHY_REG(770, 26), &phy_data);
2893                 phy_data &= 0xfff8;
2894                 phy_data |= (1 << 2);
2895                 e1e_wphy(hw, PHY_REG(770, 26), phy_data);
2896
2897                 e1e_rphy(hw, 22, &phy_data);
2898                 phy_data &= 0x0fff;
2899                 phy_data |= (1 << 14);
2900                 e1e_wphy(hw, 0x10, 0x2823);
2901                 e1e_wphy(hw, 0x11, 0x0003);
2902                 e1e_wphy(hw, 22, phy_data);
2903         }
2904
2905         /* Setup buffer sizes */
2906         rctl &= ~E1000_RCTL_SZ_4096;
2907         rctl |= E1000_RCTL_BSEX;
2908         switch (adapter->rx_buffer_len) {
2909         case 2048:
2910         default:
2911                 rctl |= E1000_RCTL_SZ_2048;
2912                 rctl &= ~E1000_RCTL_BSEX;
2913                 break;
2914         case 4096:
2915                 rctl |= E1000_RCTL_SZ_4096;
2916                 break;
2917         case 8192:
2918                 rctl |= E1000_RCTL_SZ_8192;
2919                 break;
2920         case 16384:
2921                 rctl |= E1000_RCTL_SZ_16384;
2922                 break;
2923         }
2924
2925         /* Enable Extended Status in all Receive Descriptors */
2926         rfctl = er32(RFCTL);
2927         rfctl |= E1000_RFCTL_EXTEN;
2928
2929         /*
2930          * 82571 and greater support packet-split where the protocol
2931          * header is placed in skb->data and the packet data is
2932          * placed in pages hanging off of skb_shinfo(skb)->nr_frags.
2933          * In the case of a non-split, skb->data is linearly filled,
2934          * followed by the page buffers.  Therefore, skb->data is
2935          * sized to hold the largest protocol header.
2936          *
2937          * allocations using alloc_page take too long for regular MTU
2938          * so only enable packet split for jumbo frames
2939          *
2940          * Using pages when the page size is greater than 16k wastes
2941          * a lot of memory, since we allocate 3 pages at all times
2942          * per packet.
2943          */
2944         pages = PAGE_USE_COUNT(adapter->netdev->mtu);
2945         if (!(adapter->flags & FLAG_HAS_ERT) && (pages <= 3) &&
2946             (PAGE_SIZE <= 16384) && (rctl & E1000_RCTL_LPE))
2947                 adapter->rx_ps_pages = pages;
2948         else
2949                 adapter->rx_ps_pages = 0;
2950
2951         if (adapter->rx_ps_pages) {
2952                 u32 psrctl = 0;
2953
2954                 /*
2955                  * disable packet split support for IPv6 extension headers,
2956                  * because some malformed IPv6 headers can hang the Rx
2957                  */
2958                 rfctl |= (E1000_RFCTL_IPV6_EX_DIS |
2959                           E1000_RFCTL_NEW_IPV6_EXT_DIS);
2960
2961                 /* Enable Packet split descriptors */
2962                 rctl |= E1000_RCTL_DTYP_PS;
2963
2964                 psrctl |= adapter->rx_ps_bsize0 >>
2965                         E1000_PSRCTL_BSIZE0_SHIFT;
2966
2967                 switch (adapter->rx_ps_pages) {
2968                 case 3:
2969                         psrctl |= PAGE_SIZE <<
2970                                 E1000_PSRCTL_BSIZE3_SHIFT;
2971                 case 2:
2972                         psrctl |= PAGE_SIZE <<
2973                                 E1000_PSRCTL_BSIZE2_SHIFT;
2974                 case 1:
2975                         psrctl |= PAGE_SIZE >>
2976                                 E1000_PSRCTL_BSIZE1_SHIFT;
2977                         break;
2978                 }
2979
2980                 ew32(PSRCTL, psrctl);
2981         }
2982
2983         ew32(RFCTL, rfctl);
2984         ew32(RCTL, rctl);
2985         /* just started the receive unit, no need to restart */
2986         adapter->flags &= ~FLAG_RX_RESTART_NOW;
2987 }
2988
2989 /**
2990  * e1000_configure_rx - Configure Receive Unit after Reset
2991  * @adapter: board private structure
2992  *
2993  * Configure the Rx unit of the MAC after a reset.
2994  **/
2995 static void e1000_configure_rx(struct e1000_adapter *adapter)
2996 {
2997         struct e1000_hw *hw = &adapter->hw;
2998         struct e1000_ring *rx_ring = adapter->rx_ring;
2999         u64 rdba;
3000         u32 rdlen, rctl, rxcsum, ctrl_ext;
3001
3002         if (adapter->rx_ps_pages) {
3003                 /* this is a 32 byte descriptor */
3004                 rdlen = rx_ring->count *
3005                     sizeof(union e1000_rx_desc_packet_split);
3006                 adapter->clean_rx = e1000_clean_rx_irq_ps;
3007                 adapter->alloc_rx_buf = e1000_alloc_rx_buffers_ps;
3008         } else if (adapter->netdev->mtu > ETH_FRAME_LEN + ETH_FCS_LEN) {
3009                 rdlen = rx_ring->count * sizeof(union e1000_rx_desc_extended);
3010                 adapter->clean_rx = e1000_clean_jumbo_rx_irq;
3011                 adapter->alloc_rx_buf = e1000_alloc_jumbo_rx_buffers;
3012         } else {
3013                 rdlen = rx_ring->count * sizeof(union e1000_rx_desc_extended);
3014                 adapter->clean_rx = e1000_clean_rx_irq;
3015                 adapter->alloc_rx_buf = e1000_alloc_rx_buffers;
3016         }
3017
3018         /* disable receives while setting up the descriptors */
3019         rctl = er32(RCTL);
3020         if (!(adapter->flags2 & FLAG2_NO_DISABLE_RX))
3021                 ew32(RCTL, rctl & ~E1000_RCTL_EN);
3022         e1e_flush();
3023         usleep_range(10000, 20000);
3024
3025         if (adapter->flags2 & FLAG2_DMA_BURST) {
3026                 /*
3027                  * set the writeback threshold (only takes effect if the RDTR
3028                  * is set). set GRAN=1 and write back up to 0x4 worth, and
3029                  * enable prefetching of 0x20 Rx descriptors
3030                  * granularity = 01
3031                  * wthresh = 04,
3032                  * hthresh = 04,
3033                  * pthresh = 0x20
3034                  */
3035                 ew32(RXDCTL(0), E1000_RXDCTL_DMA_BURST_ENABLE);
3036                 ew32(RXDCTL(1), E1000_RXDCTL_DMA_BURST_ENABLE);
3037
3038                 /*
3039                  * override the delay timers for enabling bursting, only if
3040                  * the value was not set by the user via module options
3041                  */
3042                 if (adapter->rx_int_delay == DEFAULT_RDTR)
3043                         adapter->rx_int_delay = BURST_RDTR;
3044                 if (adapter->rx_abs_int_delay == DEFAULT_RADV)
3045                         adapter->rx_abs_int_delay = BURST_RADV;
3046         }
3047
3048         /* set the Receive Delay Timer Register */
3049         ew32(RDTR, adapter->rx_int_delay);
3050
3051         /* irq moderation */
3052         ew32(RADV, adapter->rx_abs_int_delay);
3053         if ((adapter->itr_setting != 0) && (adapter->itr != 0))
3054                 ew32(ITR, 1000000000 / (adapter->itr * 256));
3055
3056         ctrl_ext = er32(CTRL_EXT);
3057         /* Auto-Mask interrupts upon ICR access */
3058         ctrl_ext |= E1000_CTRL_EXT_IAME;
3059         ew32(IAM, 0xffffffff);
3060         ew32(CTRL_EXT, ctrl_ext);
3061         e1e_flush();
3062
3063         /*
3064          * Setup the HW Rx Head and Tail Descriptor Pointers and
3065          * the Base and Length of the Rx Descriptor Ring
3066          */
3067         rdba = rx_ring->dma;
3068         ew32(RDBAL, (rdba & DMA_BIT_MASK(32)));
3069         ew32(RDBAH, (rdba >> 32));
3070         ew32(RDLEN, rdlen);
3071         ew32(RDH, 0);
3072         ew32(RDT, 0);
3073         rx_ring->head = E1000_RDH;
3074         rx_ring->tail = E1000_RDT;
3075
3076         /* Enable Receive Checksum Offload for TCP and UDP */
3077         rxcsum = er32(RXCSUM);
3078         if (adapter->netdev->features & NETIF_F_RXCSUM) {
3079                 rxcsum |= E1000_RXCSUM_TUOFL;
3080
3081                 /*
3082                  * IPv4 payload checksum for UDP fragments must be
3083                  * used in conjunction with packet-split.
3084                  */
3085                 if (adapter->rx_ps_pages)
3086                         rxcsum |= E1000_RXCSUM_IPPCSE;
3087         } else {
3088                 rxcsum &= ~E1000_RXCSUM_TUOFL;
3089                 /* no need to clear IPPCSE as it defaults to 0 */
3090         }
3091         ew32(RXCSUM, rxcsum);
3092
3093         /*
3094          * Enable early receives on supported devices, only takes effect when
3095          * packet size is equal or larger than the specified value (in 8 byte
3096          * units), e.g. using jumbo frames when setting to E1000_ERT_2048
3097          */
3098         if ((adapter->flags & FLAG_HAS_ERT) ||
3099             (adapter->hw.mac.type == e1000_pch2lan)) {
3100                 if (adapter->netdev->mtu > ETH_DATA_LEN) {
3101                         u32 rxdctl = er32(RXDCTL(0));
3102                         ew32(RXDCTL(0), rxdctl | 0x3);
3103                         if (adapter->flags & FLAG_HAS_ERT)
3104                                 ew32(ERT, E1000_ERT_2048 | (1 << 13));
3105                         /*
3106                          * With jumbo frames and early-receive enabled,
3107                          * excessive C-state transition latencies result in
3108                          * dropped transactions.
3109                          */
3110                         pm_qos_update_request(&adapter->netdev->pm_qos_req, 55);
3111                 } else {
3112                         pm_qos_update_request(&adapter->netdev->pm_qos_req,
3113                                               PM_QOS_DEFAULT_VALUE);
3114                 }
3115         }
3116
3117         /* Enable Receives */
3118         ew32(RCTL, rctl);
3119 }
3120
3121 /**
3122  * e1000e_write_mc_addr_list - write multicast addresses to MTA
3123  * @netdev: network interface device structure
3124  *
3125  * Writes multicast address list to the MTA hash table.
3126  * Returns: -ENOMEM on failure
3127  *                0 on no addresses written
3128  *                X on writing X addresses to MTA
3129  */
3130 static int e1000e_write_mc_addr_list(struct net_device *netdev)
3131 {
3132         struct e1000_adapter *adapter = netdev_priv(netdev);
3133         struct e1000_hw *hw = &adapter->hw;
3134         struct netdev_hw_addr *ha;
3135         u8 *mta_list;
3136         int i;
3137
3138         if (netdev_mc_empty(netdev)) {
3139                 /* nothing to program, so clear mc list */
3140                 hw->mac.ops.update_mc_addr_list(hw, NULL, 0);
3141                 return 0;
3142         }
3143
3144         mta_list = kzalloc(netdev_mc_count(netdev) * ETH_ALEN, GFP_ATOMIC);
3145         if (!mta_list)
3146                 return -ENOMEM;
3147
3148         /* update_mc_addr_list expects a packed array of only addresses. */
3149         i = 0;
3150         netdev_for_each_mc_addr(ha, netdev)
3151                 memcpy(mta_list + (i++ * ETH_ALEN), ha->addr, ETH_ALEN);
3152
3153         hw->mac.ops.update_mc_addr_list(hw, mta_list, i);
3154         kfree(mta_list);
3155
3156         return netdev_mc_count(netdev);
3157 }
3158
3159 /**
3160  * e1000e_write_uc_addr_list - write unicast addresses to RAR table
3161  * @netdev: network interface device structure
3162  *
3163  * Writes unicast address list to the RAR table.
3164  * Returns: -ENOMEM on failure/insufficient address space
3165  *                0 on no addresses written
3166  *                X on writing X addresses to the RAR table
3167  **/
3168 static int e1000e_write_uc_addr_list(struct net_device *netdev)
3169 {
3170         struct e1000_adapter *adapter = netdev_priv(netdev);
3171         struct e1000_hw *hw = &adapter->hw;
3172         unsigned int rar_entries = hw->mac.rar_entry_count;
3173         int count = 0;
3174
3175         /* save a rar entry for our hardware address */
3176         rar_entries--;
3177
3178         /* save a rar entry for the LAA workaround */
3179         if (adapter->flags & FLAG_RESET_OVERWRITES_LAA)
3180                 rar_entries--;
3181
3182         /* return ENOMEM indicating insufficient memory for addresses */
3183         if (netdev_uc_count(netdev) > rar_entries)
3184                 return -ENOMEM;
3185
3186         if (!netdev_uc_empty(netdev) && rar_entries) {
3187                 struct netdev_hw_addr *ha;
3188
3189                 /*
3190                  * write the addresses in reverse order to avoid write
3191                  * combining
3192                  */
3193                 netdev_for_each_uc_addr(ha, netdev) {
3194                         if (!rar_entries)
3195                                 break;
3196                         e1000e_rar_set(hw, ha->addr, rar_entries--);
3197                         count++;
3198                 }
3199         }
3200
3201         /* zero out the remaining RAR entries not used above */
3202         for (; rar_entries > 0; rar_entries--) {
3203                 ew32(RAH(rar_entries), 0);
3204                 ew32(RAL(rar_entries), 0);
3205         }
3206         e1e_flush();
3207
3208         return count;
3209 }
3210
3211 /**
3212  * e1000e_set_rx_mode - secondary unicast, Multicast and Promiscuous mode set
3213  * @netdev: network interface device structure
3214  *
3215  * The ndo_set_rx_mode entry point is called whenever the unicast or multicast
3216  * address list or the network interface flags are updated.  This routine is
3217  * responsible for configuring the hardware for proper unicast, multicast,
3218  * promiscuous mode, and all-multi behavior.
3219  **/
3220 static void e1000e_set_rx_mode(struct net_device *netdev)
3221 {
3222         struct e1000_adapter *adapter = netdev_priv(netdev);
3223         struct e1000_hw *hw = &adapter->hw;
3224         u32 rctl;
3225
3226         /* Check for Promiscuous and All Multicast modes */
3227         rctl = er32(RCTL);
3228
3229         /* clear the affected bits */
3230         rctl &= ~(E1000_RCTL_UPE | E1000_RCTL_MPE);
3231
3232         if (netdev->flags & IFF_PROMISC) {
3233                 rctl |= (E1000_RCTL_UPE | E1000_RCTL_MPE);
3234                 /* Do not hardware filter VLANs in promisc mode */
3235                 e1000e_vlan_filter_disable(adapter);
3236         } else {
3237                 int count;
3238                 if (netdev->flags & IFF_ALLMULTI) {
3239                         rctl |= E1000_RCTL_MPE;
3240                 } else {
3241                         /*
3242                          * Write addresses to the MTA, if the attempt fails
3243                          * then we should just turn on promiscuous mode so
3244                          * that we can at least receive multicast traffic
3245                          */
3246                         count = e1000e_write_mc_addr_list(netdev);
3247                         if (count < 0)
3248                                 rctl |= E1000_RCTL_MPE;
3249                 }
3250                 e1000e_vlan_filter_enable(adapter);
3251                 /*
3252                  * Write addresses to available RAR registers, if there is not
3253                  * sufficient space to store all the addresses then enable
3254                  * unicast promiscuous mode
3255                  */
3256                 count = e1000e_write_uc_addr_list(netdev);
3257                 if (count < 0)
3258                         rctl |= E1000_RCTL_UPE;
3259         }
3260
3261         ew32(RCTL, rctl);
3262
3263         if (netdev->features & NETIF_F_HW_VLAN_RX)
3264                 e1000e_vlan_strip_enable(adapter);
3265         else
3266                 e1000e_vlan_strip_disable(adapter);
3267 }
3268
3269 /**
3270  * e1000_configure - configure the hardware for Rx and Tx
3271  * @adapter: private board structure
3272  **/
3273 static void e1000_configure(struct e1000_adapter *adapter)
3274 {
3275         e1000e_set_rx_mode(adapter->netdev);
3276
3277         e1000_restore_vlan(adapter);
3278         e1000_init_manageability_pt(adapter);
3279
3280         e1000_configure_tx(adapter);
3281         e1000_setup_rctl(adapter);
3282         e1000_configure_rx(adapter);
3283         adapter->alloc_rx_buf(adapter, e1000_desc_unused(adapter->rx_ring),
3284                               GFP_KERNEL);
3285 }
3286
3287 /**
3288  * e1000e_power_up_phy - restore link in case the phy was powered down
3289  * @adapter: address of board private structure
3290  *
3291  * The phy may be powered down to save power and turn off link when the
3292  * driver is unloaded and wake on lan is not enabled (among others)
3293  * *** this routine MUST be followed by a call to e1000e_reset ***
3294  **/
3295 void e1000e_power_up_phy(struct e1000_adapter *adapter)
3296 {
3297         if (adapter->hw.phy.ops.power_up)
3298                 adapter->hw.phy.ops.power_up(&adapter->hw);
3299
3300         adapter->hw.mac.ops.setup_link(&adapter->hw);
3301 }
3302
3303 /**
3304  * e1000_power_down_phy - Power down the PHY
3305  *
3306  * Power down the PHY so no link is implied when interface is down.
3307  * The PHY cannot be powered down if management or WoL is active.
3308  */
3309 static void e1000_power_down_phy(struct e1000_adapter *adapter)
3310 {
3311         /* WoL is enabled */
3312         if (adapter->wol)
3313                 return;
3314
3315         if (adapter->hw.phy.ops.power_down)
3316                 adapter->hw.phy.ops.power_down(&adapter->hw);
3317 }
3318
3319 /**
3320  * e1000e_reset - bring the hardware into a known good state
3321  *
3322  * This function boots the hardware and enables some settings that
3323  * require a configuration cycle of the hardware - those cannot be
3324  * set/changed during runtime. After reset the device needs to be
3325  * properly configured for Rx, Tx etc.
3326  */
3327 void e1000e_reset(struct e1000_adapter *adapter)
3328 {
3329         struct e1000_mac_info *mac = &adapter->hw.mac;
3330         struct e1000_fc_info *fc = &adapter->hw.fc;
3331         struct e1000_hw *hw = &adapter->hw;
3332         u32 tx_space, min_tx_space, min_rx_space;
3333         u32 pba = adapter->pba;
3334         u16 hwm;
3335
3336         /* reset Packet Buffer Allocation to default */
3337         ew32(PBA, pba);
3338
3339         if (adapter->max_frame_size > ETH_FRAME_LEN + ETH_FCS_LEN) {
3340                 /*
3341                  * To maintain wire speed transmits, the Tx FIFO should be
3342                  * large enough to accommodate two full transmit packets,
3343                  * rounded up to the next 1KB and expressed in KB.  Likewise,
3344                  * the Rx FIFO should be large enough to accommodate at least
3345                  * one full receive packet and is similarly rounded up and
3346                  * expressed in KB.
3347                  */
3348                 pba = er32(PBA);
3349                 /* upper 16 bits has Tx packet buffer allocation size in KB */
3350                 tx_space = pba >> 16;
3351                 /* lower 16 bits has Rx packet buffer allocation size in KB */
3352                 pba &= 0xffff;
3353                 /*
3354                  * the Tx fifo also stores 16 bytes of information about the Tx
3355                  * but don't include ethernet FCS because hardware appends it
3356                  */
3357                 min_tx_space = (adapter->max_frame_size +
3358                                 sizeof(struct e1000_tx_desc) -
3359                                 ETH_FCS_LEN) * 2;
3360                 min_tx_space = ALIGN(min_tx_space, 1024);
3361                 min_tx_space >>= 10;
3362                 /* software strips receive CRC, so leave room for it */
3363                 min_rx_space = adapter->max_frame_size;
3364                 min_rx_space = ALIGN(min_rx_space, 1024);
3365                 min_rx_space >>= 10;
3366
3367                 /*
3368                  * If current Tx allocation is less than the min Tx FIFO size,
3369                  * and the min Tx FIFO size is less than the current Rx FIFO
3370                  * allocation, take space away from current Rx allocation
3371                  */
3372                 if ((tx_space < min_tx_space) &&
3373                     ((min_tx_space - tx_space) < pba)) {
3374                         pba -= min_tx_space - tx_space;
3375
3376                         /*
3377                          * if short on Rx space, Rx wins and must trump Tx
3378                          * adjustment or use Early Receive if available
3379                          */
3380                         if ((pba < min_rx_space) &&
3381                             (!(adapter->flags & FLAG_HAS_ERT)))
3382                                 /* ERT enabled in e1000_configure_rx */
3383                                 pba = min_rx_space;
3384                 }
3385
3386                 ew32(PBA, pba);
3387         }
3388
3389         /*
3390          * flow control settings
3391          *
3392          * The high water mark must be low enough to fit one full frame
3393          * (or the size used for early receive) above it in the Rx FIFO.
3394          * Set it to the lower of:
3395          * - 90% of the Rx FIFO size, and
3396          * - the full Rx FIFO size minus the early receive size (for parts
3397          *   with ERT support assuming ERT set to E1000_ERT_2048), or
3398          * - the full Rx FIFO size minus one full frame
3399          */
3400         if (adapter->flags & FLAG_DISABLE_FC_PAUSE_TIME)
3401                 fc->pause_time = 0xFFFF;
3402         else
3403                 fc->pause_time = E1000_FC_PAUSE_TIME;
3404         fc->send_xon = 1;
3405         fc->current_mode = fc->requested_mode;
3406
3407         switch (hw->mac.type) {
3408         default:
3409                 if ((adapter->flags & FLAG_HAS_ERT) &&
3410                     (adapter->netdev->mtu > ETH_DATA_LEN))
3411                         hwm = min(((pba << 10) * 9 / 10),
3412                                   ((pba << 10) - (E1000_ERT_2048 << 3)));
3413                 else
3414                         hwm = min(((pba << 10) * 9 / 10),
3415                                   ((pba << 10) - adapter->max_frame_size));
3416
3417                 fc->high_water = hwm & E1000_FCRTH_RTH; /* 8-byte granularity */
3418                 fc->low_water = fc->high_water - 8;
3419                 break;
3420         case e1000_pchlan:
3421                 /*
3422                  * Workaround PCH LOM adapter hangs with certain network
3423                  * loads.  If hangs persist, try disabling Tx flow control.
3424                  */
3425                 if (adapter->netdev->mtu > ETH_DATA_LEN) {
3426                         fc->high_water = 0x3500;
3427                         fc->low_water  = 0x1500;
3428                 } else {
3429                         fc->high_water = 0x5000;
3430                         fc->low_water  = 0x3000;
3431                 }
3432                 fc->refresh_time = 0x1000;
3433                 break;
3434         case e1000_pch2lan:
3435                 fc->high_water = 0x05C20;
3436                 fc->low_water = 0x05048;
3437                 fc->pause_time = 0x0650;
3438                 fc->refresh_time = 0x0400;
3439                 if (adapter->netdev->mtu > ETH_DATA_LEN) {
3440                         pba = 14;
3441                         ew32(PBA, pba);
3442                 }
3443                 break;
3444         }
3445
3446         /*
3447          * Disable Adaptive Interrupt Moderation if 2 full packets cannot
3448          * fit in receive buffer and early-receive not supported.
3449          */
3450         if (adapter->itr_setting & 0x3) {
3451                 if (((adapter->max_frame_size * 2) > (pba << 10)) &&
3452                     !(adapter->flags & FLAG_HAS_ERT)) {
3453                         if (!(adapter->flags2 & FLAG2_DISABLE_AIM)) {
3454                                 dev_info(&adapter->pdev->dev,
3455                                         "Interrupt Throttle Rate turned off\n");
3456                                 adapter->flags2 |= FLAG2_DISABLE_AIM;
3457                                 ew32(ITR, 0);
3458                         }
3459                 } else if (adapter->flags2 & FLAG2_DISABLE_AIM) {
3460                         dev_info(&adapter->pdev->dev,
3461                                  "Interrupt Throttle Rate turned on\n");
3462                         adapter->flags2 &= ~FLAG2_DISABLE_AIM;
3463                         adapter->itr = 20000;
3464                         ew32(ITR, 1000000000 / (adapter->itr * 256));
3465                 }
3466         }
3467
3468         /* Allow time for pending master requests to run */
3469         mac->ops.reset_hw(hw);
3470
3471         /*
3472          * For parts with AMT enabled, let the firmware know
3473          * that the network interface is in control
3474          */
3475         if (adapter->flags & FLAG_HAS_AMT)
3476                 e1000e_get_hw_control(adapter);
3477
3478         ew32(WUC, 0);
3479
3480         if (mac->ops.init_hw(hw))
3481                 e_err("Hardware Error\n");
3482
3483         e1000_update_mng_vlan(adapter);
3484
3485         /* Enable h/w to recognize an 802.1Q VLAN Ethernet packet */
3486         ew32(VET, ETH_P_8021Q);
3487
3488         e1000e_reset_adaptive(hw);
3489
3490         if (!netif_running(adapter->netdev) &&
3491             !test_bit(__E1000_TESTING, &adapter->state)) {
3492                 e1000_power_down_phy(adapter);
3493                 return;
3494         }
3495
3496         e1000_get_phy_info(hw);
3497
3498         if ((adapter->flags & FLAG_HAS_SMART_POWER_DOWN) &&
3499             !(adapter->flags & FLAG_SMART_POWER_DOWN)) {
3500                 u16 phy_data = 0;
3501                 /*
3502                  * speed up time to link by disabling smart power down, ignore
3503                  * the return value of this function because there is nothing
3504                  * different we would do if it failed
3505                  */
3506                 e1e_rphy(hw, IGP02E1000_PHY_POWER_MGMT, &phy_data);
3507                 phy_data &= ~IGP02E1000_PM_SPD;
3508                 e1e_wphy(hw, IGP02E1000_PHY_POWER_MGMT, phy_data);
3509         }
3510 }
3511
3512 int e1000e_up(struct e1000_adapter *adapter)
3513 {
3514         struct e1000_hw *hw = &adapter->hw;
3515
3516         /* hardware has been reset, we need to reload some things */
3517         e1000_configure(adapter);
3518
3519         clear_bit(__E1000_DOWN, &adapter->state);
3520
3521         if (adapter->msix_entries)
3522                 e1000_configure_msix(adapter);
3523         e1000_irq_enable(adapter);
3524
3525         netif_start_queue(adapter->netdev);
3526
3527         /* fire a link change interrupt to start the watchdog */
3528         if (adapter->msix_entries)
3529                 ew32(ICS, E1000_ICS_LSC | E1000_ICR_OTHER);
3530         else
3531                 ew32(ICS, E1000_ICS_LSC);
3532
3533         return 0;
3534 }
3535
3536 static void e1000e_flush_descriptors(struct e1000_adapter *adapter)
3537 {
3538         struct e1000_hw *hw = &adapter->hw;
3539
3540         if (!(adapter->flags2 & FLAG2_DMA_BURST))
3541                 return;
3542
3543         /* flush pending descriptor writebacks to memory */
3544         ew32(TIDV, adapter->tx_int_delay | E1000_TIDV_FPD);
3545         ew32(RDTR, adapter->rx_int_delay | E1000_RDTR_FPD);
3546
3547         /* execute the writes immediately */
3548         e1e_flush();
3549 }
3550
3551 static void e1000e_update_stats(struct e1000_adapter *adapter);
3552
3553 void e1000e_down(struct e1000_adapter *adapter)
3554 {
3555         struct net_device *netdev = adapter->netdev;
3556         struct e1000_hw *hw = &adapter->hw;
3557         u32 tctl, rctl;
3558
3559         /*
3560          * signal that we're down so the interrupt handler does not
3561          * reschedule our watchdog timer
3562          */
3563         set_bit(__E1000_DOWN, &adapter->state);
3564
3565         /* disable receives in the hardware */
3566         rctl = er32(RCTL);
3567         if (!(adapter->flags2 & FLAG2_NO_DISABLE_RX))
3568                 ew32(RCTL, rctl & ~E1000_RCTL_EN);
3569         /* flush and sleep below */
3570
3571         netif_stop_queue(netdev);
3572
3573         /* disable transmits in the hardware */
3574         tctl = er32(TCTL);
3575         tctl &= ~E1000_TCTL_EN;
3576         ew32(TCTL, tctl);
3577
3578         /* flush both disables and wait for them to finish */
3579         e1e_flush();
3580         usleep_range(10000, 20000);
3581
3582         e1000_irq_disable(adapter);
3583
3584         del_timer_sync(&adapter->watchdog_timer);
3585         del_timer_sync(&adapter->phy_info_timer);
3586
3587         netif_carrier_off(netdev);
3588
3589         spin_lock(&adapter->stats64_lock);
3590         e1000e_update_stats(adapter);
3591         spin_unlock(&adapter->stats64_lock);
3592
3593         e1000e_flush_descriptors(adapter);
3594         e1000_clean_tx_ring(adapter);
3595         e1000_clean_rx_ring(adapter);
3596
3597         adapter->link_speed = 0;
3598         adapter->link_duplex = 0;
3599
3600         if (!pci_channel_offline(adapter->pdev))
3601                 e1000e_reset(adapter);
3602
3603         /*
3604          * TODO: for power management, we could drop the link and
3605          * pci_disable_device here.
3606          */
3607 }
3608
3609 void e1000e_reinit_locked(struct e1000_adapter *adapter)
3610 {
3611         might_sleep();
3612         while (test_and_set_bit(__E1000_RESETTING, &adapter->state))
3613                 usleep_range(1000, 2000);
3614         e1000e_down(adapter);
3615         e1000e_up(adapter);
3616         clear_bit(__E1000_RESETTING, &adapter->state);
3617 }
3618
3619 /**
3620  * e1000_sw_init - Initialize general software structures (struct e1000_adapter)
3621  * @adapter: board private structure to initialize
3622  *
3623  * e1000_sw_init initializes the Adapter private data structure.
3624  * Fields are initialized based on PCI device information and
3625  * OS network device settings (MTU size).
3626  **/
3627 static int __devinit e1000_sw_init(struct e1000_adapter *adapter)
3628 {
3629         struct net_device *netdev = adapter->netdev;
3630
3631         adapter->rx_buffer_len = ETH_FRAME_LEN + VLAN_HLEN + ETH_FCS_LEN;
3632         adapter->rx_ps_bsize0 = 128;
3633         adapter->max_frame_size = netdev->mtu + ETH_HLEN + ETH_FCS_LEN;
3634         adapter->min_frame_size = ETH_ZLEN + ETH_FCS_LEN;
3635
3636         spin_lock_init(&adapter->stats64_lock);
3637
3638         e1000e_set_interrupt_capability(adapter);
3639
3640         if (e1000_alloc_queues(adapter))
3641                 return -ENOMEM;
3642
3643         /* Explicitly disable IRQ since the NIC can be in any state. */
3644         e1000_irq_disable(adapter);
3645
3646         set_bit(__E1000_DOWN, &adapter->state);
3647         return 0;
3648 }
3649
3650 /**
3651  * e1000_intr_msi_test - Interrupt Handler
3652  * @irq: interrupt number
3653  * @data: pointer to a network interface device structure
3654  **/
3655 static irqreturn_t e1000_intr_msi_test(int irq, void *data)
3656 {
3657         struct net_device *netdev = data;
3658         struct e1000_adapter *adapter = netdev_priv(netdev);
3659         struct e1000_hw *hw = &adapter->hw;
3660         u32 icr = er32(ICR);
3661
3662         e_dbg("icr is %08X\n", icr);
3663         if (icr & E1000_ICR_RXSEQ) {
3664                 adapter->flags &= ~FLAG_MSI_TEST_FAILED;
3665                 wmb();
3666         }
3667
3668         return IRQ_HANDLED;
3669 }
3670
3671 /**
3672  * e1000_test_msi_interrupt - Returns 0 for successful test
3673  * @adapter: board private struct
3674  *
3675  * code flow taken from tg3.c
3676  **/
3677 static int e1000_test_msi_interrupt(struct e1000_adapter *adapter)
3678 {
3679         struct net_device *netdev = adapter->netdev;
3680         struct e1000_hw *hw = &adapter->hw;
3681         int err;
3682
3683         /* poll_enable hasn't been called yet, so don't need disable */
3684         /* clear any pending events */
3685         er32(ICR);
3686
3687         /* free the real vector and request a test handler */
3688         e1000_free_irq(adapter);
3689         e1000e_reset_interrupt_capability(adapter);
3690
3691         /* Assume that the test fails, if it succeeds then the test
3692          * MSI irq handler will unset this flag */
3693         adapter->flags |= FLAG_MSI_TEST_FAILED;
3694
3695         err = pci_enable_msi(adapter->pdev);
3696         if (err)
3697                 goto msi_test_failed;
3698
3699         err = request_irq(adapter->pdev->irq, e1000_intr_msi_test, 0,
3700                           netdev->name, netdev);
3701         if (err) {
3702                 pci_disable_msi(adapter->pdev);
3703                 goto msi_test_failed;
3704         }
3705
3706         wmb();
3707
3708         e1000_irq_enable(adapter);
3709
3710         /* fire an unusual interrupt on the test handler */
3711         ew32(ICS, E1000_ICS_RXSEQ);
3712         e1e_flush();
3713         msleep(50);
3714
3715         e1000_irq_disable(adapter);
3716
3717         rmb();
3718
3719         if (adapter->flags & FLAG_MSI_TEST_FAILED) {
3720                 adapter->int_mode = E1000E_INT_MODE_LEGACY;
3721                 e_info("MSI interrupt test failed, using legacy interrupt.\n");
3722         } else
3723                 e_dbg("MSI interrupt test succeeded!\n");
3724
3725         free_irq(adapter->pdev->irq, netdev);
3726         pci_disable_msi(adapter->pdev);
3727
3728 msi_test_failed:
3729         e1000e_set_interrupt_capability(adapter);
3730         return e1000_request_irq(adapter);
3731 }
3732
3733 /**
3734  * e1000_test_msi - Returns 0 if MSI test succeeds or INTx mode is restored
3735  * @adapter: board private struct
3736  *
3737  * code flow taken from tg3.c, called with e1000 interrupts disabled.
3738  **/
3739 static int e1000_test_msi(struct e1000_adapter *adapter)
3740 {
3741         int err;
3742         u16 pci_cmd;
3743
3744         if (!(adapter->flags & FLAG_MSI_ENABLED))
3745                 return 0;
3746
3747         /* disable SERR in case the MSI write causes a master abort */
3748         pci_read_config_word(adapter->pdev, PCI_COMMAND, &pci_cmd);
3749         if (pci_cmd & PCI_COMMAND_SERR)
3750                 pci_write_config_word(adapter->pdev, PCI_COMMAND,
3751                                       pci_cmd & ~PCI_COMMAND_SERR);
3752
3753         err = e1000_test_msi_interrupt(adapter);
3754
3755         /* re-enable SERR */
3756         if (pci_cmd & PCI_COMMAND_SERR) {
3757                 pci_read_config_word(adapter->pdev, PCI_COMMAND, &pci_cmd);
3758                 pci_cmd |= PCI_COMMAND_SERR;
3759                 pci_write_config_word(adapter->pdev, PCI_COMMAND, pci_cmd);
3760         }
3761
3762         return err;
3763 }
3764
3765 /**
3766  * e1000_open - Called when a network interface is made active
3767  * @netdev: network interface device structure
3768  *
3769  * Returns 0 on success, negative value on failure
3770  *
3771  * The open entry point is called when a network interface is made
3772  * active by the system (IFF_UP).  At this point all resources needed
3773  * for transmit and receive operations are allocated, the interrupt
3774  * handler is registered with the OS, the watchdog timer is started,
3775  * and the stack is notified that the interface is ready.
3776  **/
3777 static int e1000_open(struct net_device *netdev)
3778 {
3779         struct e1000_adapter *adapter = netdev_priv(netdev);
3780         struct e1000_hw *hw = &adapter->hw;
3781         struct pci_dev *pdev = adapter->pdev;
3782         int err;
3783
3784         /* disallow open during test */
3785         if (test_bit(__E1000_TESTING, &adapter->state))
3786                 return -EBUSY;
3787
3788         pm_runtime_get_sync(&pdev->dev);
3789
3790         netif_carrier_off(netdev);
3791
3792         /* allocate transmit descriptors */
3793         err = e1000e_setup_tx_resources(adapter);
3794         if (err)
3795                 goto err_setup_tx;
3796
3797         /* allocate receive descriptors */
3798         err = e1000e_setup_rx_resources(adapter);
3799         if (err)
3800                 goto err_setup_rx;
3801
3802         /*
3803          * If AMT is enabled, let the firmware know that the network
3804          * interface is now open and reset the part to a known state.
3805          */
3806         if (adapter->flags & FLAG_HAS_AMT) {
3807                 e1000e_get_hw_control(adapter);
3808                 e1000e_reset(adapter);
3809         }
3810
3811         e1000e_power_up_phy(adapter);
3812
3813         adapter->mng_vlan_id = E1000_MNG_VLAN_NONE;
3814         if ((adapter->hw.mng_cookie.status &
3815              E1000_MNG_DHCP_COOKIE_STATUS_VLAN))
3816                 e1000_update_mng_vlan(adapter);
3817
3818         /* DMA latency requirement to workaround early-receive/jumbo issue */
3819         if ((adapter->flags & FLAG_HAS_ERT) ||
3820             (adapter->hw.mac.type == e1000_pch2lan))
3821                 pm_qos_add_request(&adapter->netdev->pm_qos_req,
3822                                    PM_QOS_CPU_DMA_LATENCY,
3823                                    PM_QOS_DEFAULT_VALUE);
3824
3825         /*
3826          * before we allocate an interrupt, we must be ready to handle it.
3827          * Setting DEBUG_SHIRQ in the kernel makes it fire an interrupt
3828          * as soon as we call pci_request_irq, so we have to setup our
3829          * clean_rx handler before we do so.
3830          */
3831         e1000_configure(adapter);
3832
3833         err = e1000_request_irq(adapter);
3834         if (err)
3835                 goto err_req_irq;
3836
3837         /*
3838          * Work around PCIe errata with MSI interrupts causing some chipsets to
3839          * ignore e1000e MSI messages, which means we need to test our MSI
3840          * interrupt now
3841          */
3842         if (adapter->int_mode != E1000E_INT_MODE_LEGACY) {
3843                 err = e1000_test_msi(adapter);
3844                 if (err) {
3845                         e_err("Interrupt allocation failed\n");
3846                         goto err_req_irq;
3847                 }
3848         }
3849
3850         /* From here on the code is the same as e1000e_up() */
3851         clear_bit(__E1000_DOWN, &adapter->state);
3852
3853         napi_enable(&adapter->napi);
3854
3855         e1000_irq_enable(adapter);
3856
3857         adapter->tx_hang_recheck = false;
3858         netif_start_queue(netdev);
3859
3860         adapter->idle_check = true;
3861         pm_runtime_put(&pdev->dev);
3862
3863         /* fire a link status change interrupt to start the watchdog */
3864         if (adapter->msix_entries)
3865                 ew32(ICS, E1000_ICS_LSC | E1000_ICR_OTHER);
3866         else
3867                 ew32(ICS, E1000_ICS_LSC);
3868
3869         return 0;
3870
3871 err_req_irq:
3872         e1000e_release_hw_control(adapter);
3873         e1000_power_down_phy(adapter);
3874         e1000e_free_rx_resources(adapter);
3875 err_setup_rx:
3876         e1000e_free_tx_resources(adapter);
3877 err_setup_tx:
3878         e1000e_reset(adapter);
3879         pm_runtime_put_sync(&pdev->dev);
3880
3881         return err;
3882 }
3883
3884 /**
3885  * e1000_close - Disables a network interface
3886  * @netdev: network interface device structure
3887  *
3888  * Returns 0, this is not allowed to fail
3889  *
3890  * The close entry point is called when an interface is de-activated
3891  * by the OS.  The hardware is still under the drivers control, but
3892  * needs to be disabled.  A global MAC reset is issued to stop the
3893  * hardware, and all transmit and receive resources are freed.
3894  **/
3895 static int e1000_close(struct net_device *netdev)
3896 {
3897         struct e1000_adapter *adapter = netdev_priv(netdev);
3898         struct pci_dev *pdev = adapter->pdev;
3899
3900         WARN_ON(test_bit(__E1000_RESETTING, &adapter->state));
3901
3902         pm_runtime_get_sync(&pdev->dev);
3903
3904         napi_disable(&adapter->napi);
3905
3906         if (!test_bit(__E1000_DOWN, &adapter->state)) {
3907                 e1000e_down(adapter);
3908                 e1000_free_irq(adapter);
3909         }
3910         e1000_power_down_phy(adapter);
3911
3912         e1000e_free_tx_resources(adapter);
3913         e1000e_free_rx_resources(adapter);
3914
3915         /*
3916          * kill manageability vlan ID if supported, but not if a vlan with
3917          * the same ID is registered on the host OS (let 8021q kill it)
3918          */
3919         if (adapter->hw.mng_cookie.status &
3920             E1000_MNG_DHCP_COOKIE_STATUS_VLAN)
3921                 e1000_vlan_rx_kill_vid(netdev, adapter->mng_vlan_id);
3922
3923         /*
3924          * If AMT is enabled, let the firmware know that the network
3925          * interface is now closed
3926          */
3927         if ((adapter->flags & FLAG_HAS_AMT) &&
3928             !test_bit(__E1000_TESTING, &adapter->state))
3929                 e1000e_release_hw_control(adapter);
3930
3931         if ((adapter->flags & FLAG_HAS_ERT) ||
3932             (adapter->hw.mac.type == e1000_pch2lan))
3933                 pm_qos_remove_request(&adapter->netdev->pm_qos_req);
3934
3935         pm_runtime_put_sync(&pdev->dev);
3936
3937         return 0;
3938 }
3939 /**
3940  * e1000_set_mac - Change the Ethernet Address of the NIC
3941  * @netdev: network interface device structure
3942  * @p: pointer to an address structure
3943  *
3944  * Returns 0 on success, negative on failure
3945  **/
3946 static int e1000_set_mac(struct net_device *netdev, void *p)
3947 {
3948         struct e1000_adapter *adapter = netdev_priv(netdev);
3949         struct sockaddr *addr = p;
3950
3951         if (!is_valid_ether_addr(addr->sa_data))
3952                 return -EADDRNOTAVAIL;
3953
3954         memcpy(netdev->dev_addr, addr->sa_data, netdev->addr_len);
3955         memcpy(adapter->hw.mac.addr, addr->sa_data, netdev->addr_len);
3956
3957         e1000e_rar_set(&adapter->hw, adapter->hw.mac.addr, 0);
3958
3959         if (adapter->flags & FLAG_RESET_OVERWRITES_LAA) {
3960                 /* activate the work around */
3961                 e1000e_set_laa_state_82571(&adapter->hw, 1);
3962
3963                 /*
3964                  * Hold a copy of the LAA in RAR[14] This is done so that
3965                  * between the time RAR[0] gets clobbered  and the time it
3966                  * gets fixed (in e1000_watchdog), the actual LAA is in one
3967                  * of the RARs and no incoming packets directed to this port
3968                  * are dropped. Eventually the LAA will be in RAR[0] and
3969                  * RAR[14]
3970                  */
3971                 e1000e_rar_set(&adapter->hw,
3972                               adapter->hw.mac.addr,
3973                               adapter->hw.mac.rar_entry_count - 1);
3974         }
3975
3976         return 0;
3977 }
3978
3979 /**
3980  * e1000e_update_phy_task - work thread to update phy
3981  * @work: pointer to our work struct
3982  *
3983  * this worker thread exists because we must acquire a
3984  * semaphore to read the phy, which we could msleep while
3985  * waiting for it, and we can't msleep in a timer.
3986  **/
3987 static void e1000e_update_phy_task(struct work_struct *work)
3988 {
3989         struct e1000_adapter *adapter = container_of(work,
3990                                         struct e1000_adapter, update_phy_task);
3991
3992         if (test_bit(__E1000_DOWN, &adapter->state))
3993                 return;
3994
3995         e1000_get_phy_info(&adapter->hw);
3996 }
3997
3998 /*
3999  * Need to wait a few seconds after link up to get diagnostic information from
4000  * the phy
4001  */
4002 static void e1000_update_phy_info(unsigned long data)
4003 {
4004         struct e1000_adapter *adapter = (struct e1000_adapter *) data;
4005
4006         if (test_bit(__E1000_DOWN, &adapter->state))
4007                 return;
4008
4009         schedule_work(&adapter->update_phy_task);
4010 }
4011
4012 /**
4013  * e1000e_update_phy_stats - Update the PHY statistics counters
4014  * @adapter: board private structure
4015  *
4016  * Read/clear the upper 16-bit PHY registers and read/accumulate lower
4017  **/
4018 static void e1000e_update_phy_stats(struct e1000_adapter *adapter)
4019 {
4020         struct e1000_hw *hw = &adapter->hw;
4021         s32 ret_val;
4022         u16 phy_data;
4023
4024         ret_val = hw->phy.ops.acquire(hw);
4025         if (ret_val)
4026                 return;
4027
4028         /*
4029          * A page set is expensive so check if already on desired page.
4030          * If not, set to the page with the PHY status registers.
4031          */
4032         hw->phy.addr = 1;
4033         ret_val = e1000e_read_phy_reg_mdic(hw, IGP01E1000_PHY_PAGE_SELECT,
4034                                            &phy_data);
4035         if (ret_val)
4036                 goto release;
4037         if (phy_data != (HV_STATS_PAGE << IGP_PAGE_SHIFT)) {
4038                 ret_val = hw->phy.ops.set_page(hw,
4039                                                HV_STATS_PAGE << IGP_PAGE_SHIFT);
4040                 if (ret_val)
4041                         goto release;
4042         }
4043
4044         /* Single Collision Count */
4045         hw->phy.ops.read_reg_page(hw, HV_SCC_UPPER, &phy_data);
4046         ret_val = hw->phy.ops.read_reg_page(hw, HV_SCC_LOWER, &phy_data);
4047         if (!ret_val)
4048                 adapter->stats.scc += phy_data;
4049
4050         /* Excessive Collision Count */
4051         hw->phy.ops.read_reg_page(hw, HV_ECOL_UPPER, &phy_data);
4052         ret_val = hw->phy.ops.read_reg_page(hw, HV_ECOL_LOWER, &phy_data);
4053         if (!ret_val)
4054                 adapter->stats.ecol += phy_data;
4055
4056         /* Multiple Collision Count */
4057         hw->phy.ops.read_reg_page(hw, HV_MCC_UPPER, &phy_data);
4058         ret_val = hw->phy.ops.read_reg_page(hw, HV_MCC_LOWER, &phy_data);
4059         if (!ret_val)
4060                 adapter->stats.mcc += phy_data;
4061
4062         /* Late Collision Count */
4063         hw->phy.ops.read_reg_page(hw, HV_LATECOL_UPPER, &phy_data);
4064         ret_val = hw->phy.ops.read_reg_page(hw, HV_LATECOL_LOWER, &phy_data);
4065         if (!ret_val)
4066                 adapter->stats.latecol += phy_data;
4067
4068         /* Collision Count - also used for adaptive IFS */
4069         hw->phy.ops.read_reg_page(hw, HV_COLC_UPPER, &phy_data);
4070         ret_val = hw->phy.ops.read_reg_page(hw, HV_COLC_LOWER, &phy_data);
4071         if (!ret_val)
4072                 hw->mac.collision_delta = phy_data;
4073
4074         /* Defer Count */
4075         hw->phy.ops.read_reg_page(hw, HV_DC_UPPER, &phy_data);
4076         ret_val = hw->phy.ops.read_reg_page(hw, HV_DC_LOWER, &phy_data);
4077         if (!ret_val)
4078                 adapter->stats.dc += phy_data;
4079
4080         /* Transmit with no CRS */
4081         hw->phy.ops.read_reg_page(hw, HV_TNCRS_UPPER, &phy_data);
4082         ret_val = hw->phy.ops.read_reg_page(hw, HV_TNCRS_LOWER, &phy_data);
4083         if (!ret_val)
4084                 adapter->stats.tncrs += phy_data;
4085
4086 release:
4087         hw->phy.ops.release(hw);
4088 }
4089
4090 /**
4091  * e1000e_update_stats - Update the board statistics counters
4092  * @adapter: board private structure
4093  **/
4094 static void e1000e_update_stats(struct e1000_adapter *adapter)
4095 {
4096         struct net_device *netdev = adapter->netdev;
4097         struct e1000_hw *hw = &adapter->hw;
4098         struct pci_dev *pdev = adapter->pdev;
4099
4100         /*
4101          * Prevent stats update while adapter is being reset, or if the pci
4102          * connection is down.
4103          */
4104         if (adapter->link_speed == 0)
4105                 return;
4106         if (pci_channel_offline(pdev))
4107                 return;
4108
4109         adapter->stats.crcerrs += er32(CRCERRS);
4110         adapter->stats.gprc += er32(GPRC);
4111         adapter->stats.gorc += er32(GORCL);
4112         er32(GORCH); /* Clear gorc */
4113         adapter->stats.bprc += er32(BPRC);
4114         adapter->stats.mprc += er32(MPRC);
4115         adapter->stats.roc += er32(ROC);
4116
4117         adapter->stats.mpc += er32(MPC);
4118
4119         /* Half-duplex statistics */
4120         if (adapter->link_duplex == HALF_DUPLEX) {
4121                 if (adapter->flags2 & FLAG2_HAS_PHY_STATS) {
4122                         e1000e_update_phy_stats(adapter);
4123                 } else {
4124                         adapter->stats.scc += er32(SCC);
4125                         adapter->stats.ecol += er32(ECOL);
4126                         adapter->stats.mcc += er32(MCC);
4127                         adapter->stats.latecol += er32(LATECOL);
4128                         adapter->stats.dc += er32(DC);
4129
4130                         hw->mac.collision_delta = er32(COLC);
4131
4132                         if ((hw->mac.type != e1000_82574) &&
4133                             (hw->mac.type != e1000_82583))
4134                                 adapter->stats.tncrs += er32(TNCRS);
4135                 }
4136                 adapter->stats.colc += hw->mac.collision_delta;
4137         }
4138
4139         adapter->stats.xonrxc += er32(XONRXC);
4140         adapter->stats.xontxc += er32(XONTXC);
4141         adapter->stats.xoffrxc += er32(XOFFRXC);
4142         adapter->stats.xofftxc += er32(XOFFTXC);
4143         adapter->stats.gptc += er32(GPTC);
4144         adapter->stats.gotc += er32(GOTCL);
4145         er32(GOTCH); /* Clear gotc */
4146         adapter->stats.rnbc += er32(RNBC);
4147         adapter->stats.ruc += er32(RUC);
4148
4149         adapter->stats.mptc += er32(MPTC);
4150         adapter->stats.bptc += er32(BPTC);
4151
4152         /* used for adaptive IFS */
4153
4154         hw->mac.tx_packet_delta = er32(TPT);
4155         adapter->stats.tpt += hw->mac.tx_packet_delta;
4156
4157         adapter->stats.algnerrc += er32(ALGNERRC);
4158         adapter->stats.rxerrc += er32(RXERRC);
4159         adapter->stats.cexterr += er32(CEXTERR);
4160         adapter->stats.tsctc += er32(TSCTC);
4161         adapter->stats.tsctfc += er32(TSCTFC);
4162
4163         /* Fill out the OS statistics structure */
4164         netdev->stats.multicast = adapter->stats.mprc;
4165         netdev->stats.collisions = adapter->stats.colc;
4166
4167         /* Rx Errors */
4168
4169         /*
4170          * RLEC on some newer hardware can be incorrect so build
4171          * our own version based on RUC and ROC
4172          */
4173         netdev->stats.rx_errors = adapter->stats.rxerrc +
4174                 adapter->stats.crcerrs + adapter->stats.algnerrc +
4175                 adapter->stats.ruc + adapter->stats.roc +
4176                 adapter->stats.cexterr;
4177         netdev->stats.rx_length_errors = adapter->stats.ruc +
4178                                               adapter->stats.roc;
4179         netdev->stats.rx_crc_errors = adapter->stats.crcerrs;
4180         netdev->stats.rx_frame_errors = adapter->stats.algnerrc;
4181         netdev->stats.rx_missed_errors = adapter->stats.mpc;
4182
4183         /* Tx Errors */
4184         netdev->stats.tx_errors = adapter->stats.ecol +
4185                                        adapter->stats.latecol;
4186         netdev->stats.tx_aborted_errors = adapter->stats.ecol;
4187         netdev->stats.tx_window_errors = adapter->stats.latecol;
4188         netdev->stats.tx_carrier_errors = adapter->stats.tncrs;
4189
4190         /* Tx Dropped needs to be maintained elsewhere */
4191
4192         /* Management Stats */
4193         adapter->stats.mgptc += er32(MGTPTC);
4194         adapter->stats.mgprc += er32(MGTPRC);
4195         adapter->stats.mgpdc += er32(MGTPDC);
4196 }
4197
4198 /**
4199  * e1000_phy_read_status - Update the PHY register status snapshot
4200  * @adapter: board private structure
4201  **/
4202 static void e1000_phy_read_status(struct e1000_adapter *adapter)
4203 {
4204         struct e1000_hw *hw = &adapter->hw;
4205         struct e1000_phy_regs *phy = &adapter->phy_regs;
4206
4207         if ((er32(STATUS) & E1000_STATUS_LU) &&
4208             (adapter->hw.phy.media_type == e1000_media_type_copper)) {
4209                 int ret_val;
4210
4211                 ret_val  = e1e_rphy(hw, PHY_CONTROL, &phy->bmcr);
4212                 ret_val |= e1e_rphy(hw, PHY_STATUS, &phy->bmsr);
4213                 ret_val |= e1e_rphy(hw, PHY_AUTONEG_ADV, &phy->advertise);
4214                 ret_val |= e1e_rphy(hw, PHY_LP_ABILITY, &phy->lpa);
4215                 ret_val |= e1e_rphy(hw, PHY_AUTONEG_EXP, &phy->expansion);
4216                 ret_val |= e1e_rphy(hw, PHY_1000T_CTRL, &phy->ctrl1000);
4217                 ret_val |= e1e_rphy(hw, PHY_1000T_STATUS, &phy->stat1000);
4218                 ret_val |= e1e_rphy(hw, PHY_EXT_STATUS, &phy->estatus);
4219                 if (ret_val)
4220                         e_warn("Error reading PHY register\n");
4221         } else {
4222                 /*
4223                  * Do not read PHY registers if link is not up
4224                  * Set values to typical power-on defaults
4225                  */
4226                 phy->bmcr = (BMCR_SPEED1000 | BMCR_ANENABLE | BMCR_FULLDPLX);
4227                 phy->bmsr = (BMSR_100FULL | BMSR_100HALF | BMSR_10FULL |
4228                              BMSR_10HALF | BMSR_ESTATEN | BMSR_ANEGCAPABLE |
4229                              BMSR_ERCAP);
4230                 phy->advertise = (ADVERTISE_PAUSE_ASYM | ADVERTISE_PAUSE_CAP |
4231                                   ADVERTISE_ALL | ADVERTISE_CSMA);
4232                 phy->lpa = 0;
4233                 phy->expansion = EXPANSION_ENABLENPAGE;
4234                 phy->ctrl1000 = ADVERTISE_1000FULL;
4235                 phy->stat1000 = 0;
4236                 phy->estatus = (ESTATUS_1000_TFULL | ESTATUS_1000_THALF);
4237         }
4238 }
4239
4240 static void e1000_print_link_info(struct e1000_adapter *adapter)
4241 {
4242         struct e1000_hw *hw = &adapter->hw;
4243         u32 ctrl = er32(CTRL);
4244
4245         /* Link status message must follow this format for user tools */
4246         printk(KERN_INFO "e1000e: %s NIC Link is Up %d Mbps %s Duplex, Flow Control: %s\n",
4247                 adapter->netdev->name,
4248                 adapter->link_speed,
4249                 adapter->link_duplex == FULL_DUPLEX ? "Full" : "Half",
4250                 (ctrl & E1000_CTRL_TFCE) && (ctrl & E1000_CTRL_RFCE) ? "Rx/Tx" :
4251                 (ctrl & E1000_CTRL_RFCE) ? "Rx" :
4252                 (ctrl & E1000_CTRL_TFCE) ? "Tx" : "None");
4253 }
4254
4255 static bool e1000e_has_link(struct e1000_adapter *adapter)
4256 {
4257         struct e1000_hw *hw = &adapter->hw;
4258         bool link_active = false;
4259         s32 ret_val = 0;
4260
4261         /*
4262          * get_link_status is set on LSC (link status) interrupt or
4263          * Rx sequence error interrupt.  get_link_status will stay
4264          * false until the check_for_link establishes link
4265          * for copper adapters ONLY
4266          */
4267         switch (hw->phy.media_type) {
4268         case e1000_media_type_copper:
4269                 if (hw->mac.get_link_status) {
4270                         ret_val = hw->mac.ops.check_for_link(hw);
4271                         link_active = !hw->mac.get_link_status;
4272                 } else {
4273                         link_active = true;
4274                 }
4275                 break;
4276         case e1000_media_type_fiber:
4277                 ret_val = hw->mac.ops.check_for_link(hw);
4278                 link_active = !!(er32(STATUS) & E1000_STATUS_LU);
4279                 break;
4280         case e1000_media_type_internal_serdes:
4281                 ret_val = hw->mac.ops.check_for_link(hw);
4282                 link_active = adapter->hw.mac.serdes_has_link;
4283                 break;
4284         default:
4285         case e1000_media_type_unknown:
4286                 break;
4287         }
4288
4289         if ((ret_val == E1000_ERR_PHY) && (hw->phy.type == e1000_phy_igp_3) &&
4290             (er32(CTRL) & E1000_PHY_CTRL_GBE_DISABLE)) {
4291                 /* See e1000_kmrn_lock_loss_workaround_ich8lan() */
4292                 e_info("Gigabit has been disabled, downgrading speed\n");
4293         }
4294
4295         return link_active;
4296 }
4297
4298 static void e1000e_enable_receives(struct e1000_adapter *adapter)
4299 {
4300         /* make sure the receive unit is started */
4301         if ((adapter->flags & FLAG_RX_NEEDS_RESTART) &&
4302             (adapter->flags & FLAG_RX_RESTART_NOW)) {
4303                 struct e1000_hw *hw = &adapter->hw;
4304                 u32 rctl = er32(RCTL);
4305                 ew32(RCTL, rctl | E1000_RCTL_EN);
4306                 adapter->flags &= ~FLAG_RX_RESTART_NOW;
4307         }
4308 }
4309
4310 static void e1000e_check_82574_phy_workaround(struct e1000_adapter *adapter)
4311 {
4312         struct e1000_hw *hw = &adapter->hw;
4313
4314         /*
4315          * With 82574 controllers, PHY needs to be checked periodically
4316          * for hung state and reset, if two calls return true
4317          */
4318         if (e1000_check_phy_82574(hw))
4319                 adapter->phy_hang_count++;
4320         else
4321                 adapter->phy_hang_count = 0;
4322
4323         if (adapter->phy_hang_count > 1) {
4324                 adapter->phy_hang_count = 0;
4325                 schedule_work(&adapter->reset_task);
4326         }
4327 }
4328
4329 /**
4330  * e1000_watchdog - Timer Call-back
4331  * @data: pointer to adapter cast into an unsigned long
4332  **/
4333 static void e1000_watchdog(unsigned long data)
4334 {
4335         struct e1000_adapter *adapter = (struct e1000_adapter *) data;
4336
4337         /* Do the rest outside of interrupt context */
4338         schedule_work(&adapter->watchdog_task);
4339
4340         /* TODO: make this use queue_delayed_work() */
4341 }
4342
4343 static void e1000_watchdog_task(struct work_struct *work)
4344 {
4345         struct e1000_adapter *adapter = container_of(work,
4346                                         struct e1000_adapter, watchdog_task);
4347         struct net_device *netdev = adapter->netdev;
4348         struct e1000_mac_info *mac = &adapter->hw.mac;
4349         struct e1000_phy_info *phy = &adapter->hw.phy;
4350         struct e1000_ring *tx_ring = adapter->tx_ring;
4351         struct e1000_hw *hw = &adapter->hw;
4352         u32 link, tctl;
4353
4354         if (test_bit(__E1000_DOWN, &adapter->state))
4355                 return;
4356
4357         link = e1000e_has_link(adapter);
4358         if ((netif_carrier_ok(netdev)) && link) {
4359                 /* Cancel scheduled suspend requests. */
4360                 pm_runtime_resume(netdev->dev.parent);
4361
4362                 e1000e_enable_receives(adapter);
4363                 goto link_up;
4364         }
4365
4366         if ((e1000e_enable_tx_pkt_filtering(hw)) &&
4367             (adapter->mng_vlan_id != adapter->hw.mng_cookie.vlan_id))
4368                 e1000_update_mng_vlan(adapter);
4369
4370         if (link) {
4371                 if (!netif_carrier_ok(netdev)) {
4372                         bool txb2b = true;
4373
4374                         /* Cancel scheduled suspend requests. */
4375                         pm_runtime_resume(netdev->dev.parent);
4376
4377                         /* update snapshot of PHY registers on LSC */
4378                         e1000_phy_read_status(adapter);
4379                         mac->ops.get_link_up_info(&adapter->hw,
4380                                                    &adapter->link_speed,
4381                                                    &adapter->link_duplex);
4382                         e1000_print_link_info(adapter);
4383                         /*
4384                          * On supported PHYs, check for duplex mismatch only
4385                          * if link has autonegotiated at 10/100 half
4386                          */
4387                         if ((hw->phy.type == e1000_phy_igp_3 ||
4388                              hw->phy.type == e1000_phy_bm) &&
4389                             (hw->mac.autoneg == true) &&
4390                             (adapter->link_speed == SPEED_10 ||
4391                              adapter->link_speed == SPEED_100) &&
4392                             (adapter->link_duplex == HALF_DUPLEX)) {
4393                                 u16 autoneg_exp;
4394
4395                                 e1e_rphy(hw, PHY_AUTONEG_EXP, &autoneg_exp);
4396
4397                                 if (!(autoneg_exp & NWAY_ER_LP_NWAY_CAPS))
4398                                         e_info("Autonegotiated half duplex but link partner cannot autoneg.  Try forcing full duplex if link gets many collisions.\n");
4399                         }
4400
4401                         /* adjust timeout factor according to speed/duplex */
4402                         adapter->tx_timeout_factor = 1;
4403                         switch (adapter->link_speed) {
4404                         case SPEED_10:
4405                                 txb2b = false;
4406                                 adapter->tx_timeout_factor = 16;
4407                                 break;
4408                         case SPEED_100:
4409                                 txb2b = false;
4410                                 adapter->tx_timeout_factor = 10;
4411                                 break;
4412                         }
4413
4414                         /*
4415                          * workaround: re-program speed mode bit after
4416                          * link-up event
4417                          */
4418                         if ((adapter->flags & FLAG_TARC_SPEED_MODE_BIT) &&
4419                             !txb2b) {
4420                                 u32 tarc0;
4421                                 tarc0 = er32(TARC(0));
4422                                 tarc0 &= ~SPEED_MODE_BIT;
4423                                 ew32(TARC(0), tarc0);
4424                         }
4425
4426                         /*
4427                          * disable TSO for pcie and 10/100 speeds, to avoid
4428                          * some hardware issues
4429                          */
4430                         if (!(adapter->flags & FLAG_TSO_FORCE)) {
4431                                 switch (adapter->link_speed) {
4432                                 case SPEED_10:
4433                                 case SPEED_100:
4434                                         e_info("10/100 speed: disabling TSO\n");
4435                                         netdev->features &= ~NETIF_F_TSO;
4436                                         netdev->features &= ~NETIF_F_TSO6;
4437                                         break;
4438                                 case SPEED_1000:
4439                                         netdev->features |= NETIF_F_TSO;
4440                                         netdev->features |= NETIF_F_TSO6;
4441                                         break;
4442                                 default:
4443                                         /* oops */
4444                                         break;
4445                                 }
4446                         }
4447
4448                         /*
4449                          * enable transmits in the hardware, need to do this
4450                          * after setting TARC(0)
4451                          */
4452                         tctl = er32(TCTL);
4453                         tctl |= E1000_TCTL_EN;
4454                         ew32(TCTL, tctl);
4455
4456                         /*
4457                          * Perform any post-link-up configuration before
4458                          * reporting link up.
4459                          */
4460                         if (phy->ops.cfg_on_link_up)
4461                                 phy->ops.cfg_on_link_up(hw);
4462
4463                         netif_carrier_on(netdev);
4464
4465                         if (!test_bit(__E1000_DOWN, &adapter->state))
4466                                 mod_timer(&adapter->phy_info_timer,
4467                                           round_jiffies(jiffies + 2 * HZ));
4468                 }
4469         } else {
4470                 if (netif_carrier_ok(netdev)) {
4471                         adapter->link_speed = 0;
4472                         adapter->link_duplex = 0;
4473                         /* Link status message must follow this format */
4474                         printk(KERN_INFO "e1000e: %s NIC Link is Down\n",
4475                                adapter->netdev->name);
4476                         netif_carrier_off(netdev);
4477                         if (!test_bit(__E1000_DOWN, &adapter->state))
4478                                 mod_timer(&adapter->phy_info_timer,
4479                                           round_jiffies(jiffies + 2 * HZ));
4480
4481                         if (adapter->flags & FLAG_RX_NEEDS_RESTART)
4482                                 schedule_work(&adapter->reset_task);
4483                         else
4484                                 pm_schedule_suspend(netdev->dev.parent,
4485                                                         LINK_TIMEOUT);
4486                 }
4487         }
4488
4489 link_up:
4490         spin_lock(&adapter->stats64_lock);
4491         e1000e_update_stats(adapter);
4492
4493         mac->tx_packet_delta = adapter->stats.tpt - adapter->tpt_old;
4494         adapter->tpt_old = adapter->stats.tpt;
4495         mac->collision_delta = adapter->stats.colc - adapter->colc_old;
4496         adapter->colc_old = adapter->stats.colc;
4497
4498         adapter->gorc = adapter->stats.gorc - adapter->gorc_old;
4499         adapter->gorc_old = adapter->stats.gorc;
4500         adapter->gotc = adapter->stats.gotc - adapter->gotc_old;
4501         adapter->gotc_old = adapter->stats.gotc;
4502         spin_unlock(&adapter->stats64_lock);
4503
4504         e1000e_update_adaptive(&adapter->hw);
4505
4506         if (!netif_carrier_ok(netdev) &&
4507             (e1000_desc_unused(tx_ring) + 1 < tx_ring->count)) {
4508                 /*
4509                  * We've lost link, so the controller stops DMA,
4510                  * but we've got queued Tx work that's never going
4511                  * to get done, so reset controller to flush Tx.
4512                  * (Do the reset outside of interrupt context).
4513                  */
4514                 schedule_work(&adapter->reset_task);
4515                 /* return immediately since reset is imminent */
4516                 return;
4517         }
4518
4519         /* Simple mode for Interrupt Throttle Rate (ITR) */
4520         if (adapter->itr_setting == 4) {
4521                 /*
4522                  * Symmetric Tx/Rx gets a reduced ITR=2000;
4523                  * Total asymmetrical Tx or Rx gets ITR=8000;
4524                  * everyone else is between 2000-8000.
4525                  */
4526                 u32 goc = (adapter->gotc + adapter->gorc) / 10000;
4527                 u32 dif = (adapter->gotc > adapter->gorc ?
4528                             adapter->gotc - adapter->gorc :
4529                             adapter->gorc - adapter->gotc) / 10000;
4530                 u32 itr = goc > 0 ? (dif * 6000 / goc + 2000) : 8000;
4531
4532                 ew32(ITR, 1000000000 / (itr * 256));
4533         }
4534
4535         /* Cause software interrupt to ensure Rx ring is cleaned */
4536         if (adapter->msix_entries)
4537                 ew32(ICS, adapter->rx_ring->ims_val);
4538         else
4539                 ew32(ICS, E1000_ICS_RXDMT0);
4540
4541         /* flush pending descriptors to memory before detecting Tx hang */
4542         e1000e_flush_descriptors(adapter);
4543
4544         /* Force detection of hung controller every watchdog period */
4545         adapter->detect_tx_hung = true;
4546
4547         /*
4548          * With 82571 controllers, LAA may be overwritten due to controller
4549          * reset from the other port. Set the appropriate LAA in RAR[0]
4550          */
4551         if (e1000e_get_laa_state_82571(hw))
4552                 e1000e_rar_set(hw, adapter->hw.mac.addr, 0);
4553
4554         if (adapter->flags2 & FLAG2_CHECK_PHY_HANG)
4555                 e1000e_check_82574_phy_workaround(adapter);
4556
4557         /* Reset the timer */
4558         if (!test_bit(__E1000_DOWN, &adapter->state))
4559                 mod_timer(&adapter->watchdog_timer,
4560                           round_jiffies(jiffies + 2 * HZ));
4561 }
4562
4563 #define E1000_TX_FLAGS_CSUM             0x00000001
4564 #define E1000_TX_FLAGS_VLAN             0x00000002
4565 #define E1000_TX_FLAGS_TSO              0x00000004
4566 #define E1000_TX_FLAGS_IPV4             0x00000008
4567 #define E1000_TX_FLAGS_VLAN_MASK        0xffff0000
4568 #define E1000_TX_FLAGS_VLAN_SHIFT       16
4569
4570 static int e1000_tso(struct e1000_adapter *adapter,
4571                      struct sk_buff *skb)
4572 {
4573         struct e1000_ring *tx_ring = adapter->tx_ring;
4574         struct e1000_context_desc *context_desc;
4575         struct e1000_buffer *buffer_info;
4576         unsigned int i;
4577         u32 cmd_length = 0;
4578         u16 ipcse = 0, tucse, mss;
4579         u8 ipcss, ipcso, tucss, tucso, hdr_len;
4580
4581         if (!skb_is_gso(skb))
4582                 return 0;
4583
4584         if (skb_header_cloned(skb)) {
4585                 int err = pskb_expand_head(skb, 0, 0, GFP_ATOMIC);
4586
4587                 if (err)
4588                         return err;
4589         }
4590
4591         hdr_len = skb_transport_offset(skb) + tcp_hdrlen(skb);
4592         mss = skb_shinfo(skb)->gso_size;
4593         if (skb->protocol == htons(ETH_P_IP)) {
4594                 struct iphdr *iph = ip_hdr(skb);
4595                 iph->tot_len = 0;
4596                 iph->check = 0;
4597                 tcp_hdr(skb)->check = ~csum_tcpudp_magic(iph->saddr, iph->daddr,
4598                                                          0, IPPROTO_TCP, 0);
4599                 cmd_length = E1000_TXD_CMD_IP;
4600                 ipcse = skb_transport_offset(skb) - 1;
4601         } else if (skb_is_gso_v6(skb)) {
4602                 ipv6_hdr(skb)->payload_len = 0;
4603                 tcp_hdr(skb)->check = ~csum_ipv6_magic(&ipv6_hdr(skb)->saddr,
4604                                                        &ipv6_hdr(skb)->daddr,
4605                                                        0, IPPROTO_TCP, 0);
4606                 ipcse = 0;
4607         }
4608         ipcss = skb_network_offset(skb);
4609         ipcso = (void *)&(ip_hdr(skb)->check) - (void *)skb->data;
4610         tucss = skb_transport_offset(skb);
4611         tucso = (void *)&(tcp_hdr(skb)->check) - (void *)skb->data;
4612         tucse = 0;
4613
4614         cmd_length |= (E1000_TXD_CMD_DEXT | E1000_TXD_CMD_TSE |
4615                        E1000_TXD_CMD_TCP | (skb->len - (hdr_len)));
4616
4617         i = tx_ring->next_to_use;
4618         context_desc = E1000_CONTEXT_DESC(*tx_ring, i);
4619         buffer_info = &tx_ring->buffer_info[i];
4620
4621         context_desc->lower_setup.ip_fields.ipcss  = ipcss;
4622         context_desc->lower_setup.ip_fields.ipcso  = ipcso;
4623         context_desc->lower_setup.ip_fields.ipcse  = cpu_to_le16(ipcse);
4624         context_desc->upper_setup.tcp_fields.tucss = tucss;
4625         context_desc->upper_setup.tcp_fields.tucso = tucso;
4626         context_desc->upper_setup.tcp_fields.tucse = cpu_to_le16(tucse);
4627         context_desc->tcp_seg_setup.fields.mss     = cpu_to_le16(mss);
4628         context_desc->tcp_seg_setup.fields.hdr_len = hdr_len;
4629         context_desc->cmd_and_length = cpu_to_le32(cmd_length);
4630
4631         buffer_info->time_stamp = jiffies;
4632         buffer_info->next_to_watch = i;
4633
4634         i++;
4635         if (i == tx_ring->count)
4636                 i = 0;
4637         tx_ring->next_to_use = i;
4638
4639         return 1;
4640 }
4641
4642 static bool e1000_tx_csum(struct e1000_adapter *adapter, struct sk_buff *skb)
4643 {
4644         struct e1000_ring *tx_ring = adapter->tx_ring;
4645         struct e1000_context_desc *context_desc;
4646         struct e1000_buffer *buffer_info;
4647         unsigned int i;
4648         u8 css;
4649         u32 cmd_len = E1000_TXD_CMD_DEXT;
4650         __be16 protocol;
4651
4652         if (skb->ip_summed != CHECKSUM_PARTIAL)
4653                 return 0;
4654
4655         if (skb->protocol == cpu_to_be16(ETH_P_8021Q))
4656                 protocol = vlan_eth_hdr(skb)->h_vlan_encapsulated_proto;
4657         else
4658                 protocol = skb->protocol;
4659
4660         switch (protocol) {
4661         case cpu_to_be16(ETH_P_IP):
4662                 if (ip_hdr(skb)->protocol == IPPROTO_TCP)
4663                         cmd_len |= E1000_TXD_CMD_TCP;
4664                 break;
4665         case cpu_to_be16(ETH_P_IPV6):
4666                 /* XXX not handling all IPV6 headers */
4667                 if (ipv6_hdr(skb)->nexthdr == IPPROTO_TCP)
4668                         cmd_len |= E1000_TXD_CMD_TCP;
4669                 break;
4670         default:
4671                 if (unlikely(net_ratelimit()))
4672                         e_warn("checksum_partial proto=%x!\n",
4673                                be16_to_cpu(protocol));
4674                 break;
4675         }
4676
4677         css = skb_checksum_start_offset(skb);
4678
4679         i = tx_ring->next_to_use;
4680         buffer_info = &tx_ring->buffer_info[i];
4681         context_desc = E1000_CONTEXT_DESC(*tx_ring, i);
4682
4683         context_desc->lower_setup.ip_config = 0;
4684         context_desc->upper_setup.tcp_fields.tucss = css;
4685         context_desc->upper_setup.tcp_fields.tucso =
4686                                 css + skb->csum_offset;
4687         context_desc->upper_setup.tcp_fields.tucse = 0;
4688         context_desc->tcp_seg_setup.data = 0;
4689         context_desc->cmd_and_length = cpu_to_le32(cmd_len);
4690
4691         buffer_info->time_stamp = jiffies;
4692         buffer_info->next_to_watch = i;
4693
4694         i++;
4695         if (i == tx_ring->count)
4696                 i = 0;
4697         tx_ring->next_to_use = i;
4698
4699         return 1;
4700 }
4701
4702 #define E1000_MAX_PER_TXD       8192
4703 #define E1000_MAX_TXD_PWR       12
4704
4705 static int e1000_tx_map(struct e1000_adapter *adapter,
4706                         struct sk_buff *skb, unsigned int first,
4707                         unsigned int max_per_txd, unsigned int nr_frags,
4708                         unsigned int mss)
4709 {
4710         struct e1000_ring *tx_ring = adapter->tx_ring;
4711         struct pci_dev *pdev = adapter->pdev;
4712         struct e1000_buffer *buffer_info;
4713         unsigned int len = skb_headlen(skb);
4714         unsigned int offset = 0, size, count = 0, i;
4715         unsigned int f, bytecount, segs;
4716
4717         i = tx_ring->next_to_use;
4718
4719         while (len) {
4720                 buffer_info = &tx_ring->buffer_info[i];
4721                 size = min(len, max_per_txd);
4722
4723                 buffer_info->length = size;
4724                 buffer_info->time_stamp = jiffies;
4725                 buffer_info->next_to_watch = i;
4726                 buffer_info->dma = dma_map_single(&pdev->dev,
4727                                                   skb->data + offset,
4728                                                   size, DMA_TO_DEVICE);
4729                 buffer_info->mapped_as_page = false;
4730                 if (dma_mapping_error(&pdev->dev, buffer_info->dma))
4731                         goto dma_error;
4732
4733                 len -= size;
4734                 offset += size;
4735                 count++;
4736
4737                 if (len) {
4738                         i++;
4739                         if (i == tx_ring->count)
4740                                 i = 0;
4741                 }
4742         }
4743
4744         for (f = 0; f < nr_frags; f++) {
4745                 const struct skb_frag_struct *frag;
4746
4747                 frag = &skb_shinfo(skb)->frags[f];
4748                 len = skb_frag_size(frag);
4749                 offset = 0;
4750
4751                 while (len) {
4752                         i++;
4753                         if (i == tx_ring->count)
4754                                 i = 0;
4755
4756                         buffer_info = &tx_ring->buffer_info[i];
4757                         size = min(len, max_per_txd);
4758
4759                         buffer_info->length = size;
4760                         buffer_info->time_stamp = jiffies;
4761                         buffer_info->next_to_watch = i;
4762                         buffer_info->dma = skb_frag_dma_map(&pdev->dev, frag,
4763                                                 offset, size, DMA_TO_DEVICE);
4764                         buffer_info->mapped_as_page = true;
4765                         if (dma_mapping_error(&pdev->dev, buffer_info->dma))
4766                                 goto dma_error;
4767
4768                         len -= size;
4769                         offset += size;
4770                         count++;
4771                 }
4772         }
4773
4774         segs = skb_shinfo(skb)->gso_segs ? : 1;
4775         /* multiply data chunks by size of headers */
4776         bytecount = ((segs - 1) * skb_headlen(skb)) + skb->len;
4777
4778         tx_ring->buffer_info[i].skb = skb;
4779         tx_ring->buffer_info[i].segs = segs;
4780         tx_ring->buffer_info[i].bytecount = bytecount;
4781         tx_ring->buffer_info[first].next_to_watch = i;
4782
4783         return count;
4784
4785 dma_error:
4786         dev_err(&pdev->dev, "Tx DMA map failed\n");
4787         buffer_info->dma = 0;
4788         if (count)
4789                 count--;
4790
4791         while (count--) {
4792                 if (i == 0)
4793                         i += tx_ring->count;
4794                 i--;
4795                 buffer_info = &tx_ring->buffer_info[i];
4796                 e1000_put_txbuf(adapter, buffer_info);
4797         }
4798
4799         return 0;
4800 }
4801
4802 static void e1000_tx_queue(struct e1000_adapter *adapter,
4803                            int tx_flags, int count)
4804 {
4805         struct e1000_ring *tx_ring = adapter->tx_ring;
4806         struct e1000_tx_desc *tx_desc = NULL;
4807         struct e1000_buffer *buffer_info;
4808         u32 txd_upper = 0, txd_lower = E1000_TXD_CMD_IFCS;
4809         unsigned int i;
4810
4811         if (tx_flags & E1000_TX_FLAGS_TSO) {
4812                 txd_lower |= E1000_TXD_CMD_DEXT | E1000_TXD_DTYP_D |
4813                              E1000_TXD_CMD_TSE;
4814                 txd_upper |= E1000_TXD_POPTS_TXSM << 8;
4815
4816                 if (tx_flags & E1000_TX_FLAGS_IPV4)
4817                         txd_upper |= E1000_TXD_POPTS_IXSM << 8;
4818         }
4819
4820         if (tx_flags & E1000_TX_FLAGS_CSUM) {
4821                 txd_lower |= E1000_TXD_CMD_DEXT | E1000_TXD_DTYP_D;
4822                 txd_upper |= E1000_TXD_POPTS_TXSM << 8;
4823         }
4824
4825         if (tx_flags & E1000_TX_FLAGS_VLAN) {
4826                 txd_lower |= E1000_TXD_CMD_VLE;
4827                 txd_upper |= (tx_flags & E1000_TX_FLAGS_VLAN_MASK);
4828         }
4829
4830         i = tx_ring->next_to_use;
4831
4832         do {
4833                 buffer_info = &tx_ring->buffer_info[i];
4834                 tx_desc = E1000_TX_DESC(*tx_ring, i);
4835                 tx_desc->buffer_addr = cpu_to_le64(buffer_info->dma);
4836                 tx_desc->lower.data =
4837                         cpu_to_le32(txd_lower | buffer_info->length);
4838                 tx_desc->upper.data = cpu_to_le32(txd_upper);
4839
4840                 i++;
4841                 if (i == tx_ring->count)
4842                         i = 0;
4843         } while (--count > 0);
4844
4845         tx_desc->lower.data |= cpu_to_le32(adapter->txd_cmd);
4846
4847         /*
4848          * Force memory writes to complete before letting h/w
4849          * know there are new descriptors to fetch.  (Only
4850          * applicable for weak-ordered memory model archs,
4851          * such as IA-64).
4852          */
4853         wmb();
4854
4855         tx_ring->next_to_use = i;
4856
4857         if (adapter->flags2 & FLAG2_PCIM2PCI_ARBITER_WA)
4858                 e1000e_update_tdt_wa(adapter, i);
4859         else
4860                 writel(i, adapter->hw.hw_addr + tx_ring->tail);
4861
4862         /*
4863          * we need this if more than one processor can write to our tail
4864          * at a time, it synchronizes IO on IA64/Altix systems
4865          */
4866         mmiowb();
4867 }
4868
4869 #define MINIMUM_DHCP_PACKET_SIZE 282
4870 static int e1000_transfer_dhcp_info(struct e1000_adapter *adapter,
4871                                     struct sk_buff *skb)
4872 {
4873         struct e1000_hw *hw =  &adapter->hw;
4874         u16 length, offset;
4875
4876         if (vlan_tx_tag_present(skb)) {
4877                 if (!((vlan_tx_tag_get(skb) == adapter->hw.mng_cookie.vlan_id) &&
4878                     (adapter->hw.mng_cookie.status &
4879                         E1000_MNG_DHCP_COOKIE_STATUS_VLAN)))
4880                         return 0;
4881         }
4882
4883         if (skb->len <= MINIMUM_DHCP_PACKET_SIZE)
4884                 return 0;
4885
4886         if (((struct ethhdr *) skb->data)->h_proto != htons(ETH_P_IP))
4887                 return 0;
4888
4889         {
4890                 const struct iphdr *ip = (struct iphdr *)((u8 *)skb->data+14);
4891                 struct udphdr *udp;
4892
4893                 if (ip->protocol != IPPROTO_UDP)
4894                         return 0;
4895
4896                 udp = (struct udphdr *)((u8 *)ip + (ip->ihl << 2));
4897                 if (ntohs(udp->dest) != 67)
4898                         return 0;
4899
4900                 offset = (u8 *)udp + 8 - skb->data;
4901                 length = skb->len - offset;
4902                 return e1000e_mng_write_dhcp_info(hw, (u8 *)udp + 8, length);
4903         }
4904
4905         return 0;
4906 }
4907
4908 static int __e1000_maybe_stop_tx(struct net_device *netdev, int size)
4909 {
4910         struct e1000_adapter *adapter = netdev_priv(netdev);
4911
4912         netif_stop_queue(netdev);
4913         /*
4914          * Herbert's original patch had:
4915          *  smp_mb__after_netif_stop_queue();
4916          * but since that doesn't exist yet, just open code it.
4917          */
4918         smp_mb();
4919
4920         /*
4921          * We need to check again in a case another CPU has just
4922          * made room available.
4923          */
4924         if (e1000_desc_unused(adapter->tx_ring) < size)
4925                 return -EBUSY;
4926
4927         /* A reprieve! */
4928         netif_start_queue(netdev);
4929         ++adapter->restart_queue;
4930         return 0;
4931 }
4932
4933 static int e1000_maybe_stop_tx(struct net_device *netdev, int size)
4934 {
4935         struct e1000_adapter *adapter = netdev_priv(netdev);
4936
4937         if (e1000_desc_unused(adapter->tx_ring) >= size)
4938                 return 0;
4939         return __e1000_maybe_stop_tx(netdev, size);
4940 }
4941
4942 #define TXD_USE_COUNT(S, X) (((S) >> (X)) + 1 )
4943 static netdev_tx_t e1000_xmit_frame(struct sk_buff *skb,
4944                                     struct net_device *netdev)
4945 {
4946         struct e1000_adapter *adapter = netdev_priv(netdev);
4947         struct e1000_ring *tx_ring = adapter->tx_ring;
4948         unsigned int first;
4949         unsigned int max_per_txd = E1000_MAX_PER_TXD;
4950         unsigned int max_txd_pwr = E1000_MAX_TXD_PWR;
4951         unsigned int tx_flags = 0;
4952         unsigned int len = skb_headlen(skb);
4953         unsigned int nr_frags;
4954         unsigned int mss;
4955         int count = 0;
4956         int tso;
4957         unsigned int f;
4958
4959         if (test_bit(__E1000_DOWN, &adapter->state)) {
4960                 dev_kfree_skb_any(skb);
4961                 return NETDEV_TX_OK;
4962         }
4963
4964         if (skb->len <= 0) {
4965                 dev_kfree_skb_any(skb);
4966                 return NETDEV_TX_OK;
4967         }
4968
4969         mss = skb_shinfo(skb)->gso_size;
4970         /*
4971          * The controller does a simple calculation to
4972          * make sure there is enough room in the FIFO before
4973          * initiating the DMA for each buffer.  The calc is:
4974          * 4 = ceil(buffer len/mss).  To make sure we don't
4975          * overrun the FIFO, adjust the max buffer len if mss
4976          * drops.
4977          */
4978         if (mss) {
4979                 u8 hdr_len;
4980                 max_per_txd = min(mss << 2, max_per_txd);
4981                 max_txd_pwr = fls(max_per_txd) - 1;
4982
4983                 /*
4984                  * TSO Workaround for 82571/2/3 Controllers -- if skb->data
4985                  * points to just header, pull a few bytes of payload from
4986                  * frags into skb->data
4987                  */
4988                 hdr_len = skb_transport_offset(skb) + tcp_hdrlen(skb);
4989                 /*
4990                  * we do this workaround for ES2LAN, but it is un-necessary,
4991                  * avoiding it could save a lot of cycles
4992                  */
4993                 if (skb->data_len && (hdr_len == len)) {
4994                         unsigned int pull_size;
4995
4996                         pull_size = min((unsigned int)4, skb->data_len);
4997                         if (!__pskb_pull_tail(skb, pull_size)) {
4998                                 e_err("__pskb_pull_tail failed.\n");
4999                                 dev_kfree_skb_any(skb);
5000                                 return NETDEV_TX_OK;
5001                         }
5002                         len = skb_headlen(skb);
5003                 }
5004         }
5005
5006         /* reserve a descriptor for the offload context */
5007         if ((mss) || (skb->ip_summed == CHECKSUM_PARTIAL))
5008                 count++;
5009         count++;
5010
5011         count += TXD_USE_COUNT(len, max_txd_pwr);
5012
5013         nr_frags = skb_shinfo(skb)->nr_frags;
5014         for (f = 0; f < nr_frags; f++)
5015                 count += TXD_USE_COUNT(skb_frag_size(&skb_shinfo(skb)->frags[f]),
5016                                        max_txd_pwr);
5017
5018         if (adapter->hw.mac.tx_pkt_filtering)
5019                 e1000_transfer_dhcp_info(adapter, skb);
5020
5021         /*
5022          * need: count + 2 desc gap to keep tail from touching
5023          * head, otherwise try next time
5024          */
5025         if (e1000_maybe_stop_tx(netdev, count + 2))
5026                 return NETDEV_TX_BUSY;
5027
5028         if (vlan_tx_tag_present(skb)) {
5029                 tx_flags |= E1000_TX_FLAGS_VLAN;
5030                 tx_flags |= (vlan_tx_tag_get(skb) << E1000_TX_FLAGS_VLAN_SHIFT);
5031         }
5032
5033         first = tx_ring->next_to_use;
5034
5035         tso = e1000_tso(adapter, skb);
5036         if (tso < 0) {
5037                 dev_kfree_skb_any(skb);
5038                 return NETDEV_TX_OK;
5039         }
5040
5041         if (tso)
5042                 tx_flags |= E1000_TX_FLAGS_TSO;
5043         else if (e1000_tx_csum(adapter, skb))
5044                 tx_flags |= E1000_TX_FLAGS_CSUM;
5045
5046         /*
5047          * Old method was to assume IPv4 packet by default if TSO was enabled.
5048          * 82571 hardware supports TSO capabilities for IPv6 as well...
5049          * no longer assume, we must.
5050          */
5051         if (skb->protocol == htons(ETH_P_IP))
5052                 tx_flags |= E1000_TX_FLAGS_IPV4;
5053
5054         /* if count is 0 then mapping error has occurred */
5055         count = e1000_tx_map(adapter, skb, first, max_per_txd, nr_frags, mss);
5056         if (count) {
5057                 netdev_sent_queue(netdev, skb->len);
5058                 e1000_tx_queue(adapter, tx_flags, count);
5059                 /* Make sure there is space in the ring for the next send. */
5060                 e1000_maybe_stop_tx(netdev, MAX_SKB_FRAGS + 2);
5061
5062         } else {
5063                 dev_kfree_skb_any(skb);
5064                 tx_ring->buffer_info[first].time_stamp = 0;
5065                 tx_ring->next_to_use = first;
5066         }
5067
5068         return NETDEV_TX_OK;
5069 }
5070
5071 /**
5072  * e1000_tx_timeout - Respond to a Tx Hang
5073  * @netdev: network interface device structure
5074  **/
5075 static void e1000_tx_timeout(struct net_device *netdev)
5076 {
5077         struct e1000_adapter *adapter = netdev_priv(netdev);
5078
5079         /* Do the reset outside of interrupt context */
5080         adapter->tx_timeout_count++;
5081         schedule_work(&adapter->reset_task);
5082 }
5083
5084 static void e1000_reset_task(struct work_struct *work)
5085 {
5086         struct e1000_adapter *adapter;
5087         adapter = container_of(work, struct e1000_adapter, reset_task);
5088
5089         /* don't run the task if already down */
5090         if (test_bit(__E1000_DOWN, &adapter->state))
5091                 return;
5092
5093         if (!((adapter->flags & FLAG_RX_NEEDS_RESTART) &&
5094               (adapter->flags & FLAG_RX_RESTART_NOW))) {
5095                 e1000e_dump(adapter);
5096                 e_err("Reset adapter\n");
5097         }
5098         e1000e_reinit_locked(adapter);
5099 }
5100
5101 /**
5102  * e1000_get_stats64 - Get System Network Statistics
5103  * @netdev: network interface device structure
5104  * @stats: rtnl_link_stats64 pointer
5105  *
5106  * Returns the address of the device statistics structure.
5107  **/
5108 struct rtnl_link_stats64 *e1000e_get_stats64(struct net_device *netdev,
5109                                              struct rtnl_link_stats64 *stats)
5110 {
5111         struct e1000_adapter *adapter = netdev_priv(netdev);
5112
5113         memset(stats, 0, sizeof(struct rtnl_link_stats64));
5114         spin_lock(&adapter->stats64_lock);
5115         e1000e_update_stats(adapter);
5116         /* Fill out the OS statistics structure */
5117         stats->rx_bytes = adapter->stats.gorc;
5118         stats->rx_packets = adapter->stats.gprc;
5119         stats->tx_bytes = adapter->stats.gotc;
5120         stats->tx_packets = adapter->stats.gptc;
5121         stats->multicast = adapter->stats.mprc;
5122         stats->collisions = adapter->stats.colc;
5123
5124         /* Rx Errors */
5125
5126         /*
5127          * RLEC on some newer hardware can be incorrect so build
5128          * our own version based on RUC and ROC
5129          */
5130         stats->rx_errors = adapter->stats.rxerrc +
5131                 adapter->stats.crcerrs + adapter->stats.algnerrc +
5132                 adapter->stats.ruc + adapter->stats.roc +
5133                 adapter->stats.cexterr;
5134         stats->rx_length_errors = adapter->stats.ruc +
5135                                               adapter->stats.roc;
5136         stats->rx_crc_errors = adapter->stats.crcerrs;
5137         stats->rx_frame_errors = adapter->stats.algnerrc;
5138         stats->rx_missed_errors = adapter->stats.mpc;
5139
5140         /* Tx Errors */
5141         stats->tx_errors = adapter->stats.ecol +
5142                                        adapter->stats.latecol;
5143         stats->tx_aborted_errors = adapter->stats.ecol;
5144         stats->tx_window_errors = adapter->stats.latecol;
5145         stats->tx_carrier_errors = adapter->stats.tncrs;
5146
5147         /* Tx Dropped needs to be maintained elsewhere */
5148
5149         spin_unlock(&adapter->stats64_lock);
5150         return stats;
5151 }
5152
5153 /**
5154  * e1000_change_mtu - Change the Maximum Transfer Unit
5155  * @netdev: network interface device structure
5156  * @new_mtu: new value for maximum frame size
5157  *
5158  * Returns 0 on success, negative on failure
5159  **/
5160 static int e1000_change_mtu(struct net_device *netdev, int new_mtu)
5161 {
5162         struct e1000_adapter *adapter = netdev_priv(netdev);
5163         int max_frame = new_mtu + ETH_HLEN + ETH_FCS_LEN;
5164
5165         /* Jumbo frame support */
5166         if ((max_frame > ETH_FRAME_LEN + ETH_FCS_LEN) &&
5167             !(adapter->flags & FLAG_HAS_JUMBO_FRAMES)) {
5168                 e_err("Jumbo Frames not supported.\n");
5169                 return -EINVAL;
5170         }
5171
5172         /* Supported frame sizes */
5173         if ((new_mtu < ETH_ZLEN + ETH_FCS_LEN + VLAN_HLEN) ||
5174             (max_frame > adapter->max_hw_frame_size)) {
5175                 e_err("Unsupported MTU setting\n");
5176                 return -EINVAL;
5177         }
5178
5179         /* Jumbo frame workaround on 82579 requires CRC be stripped */
5180         if ((adapter->hw.mac.type == e1000_pch2lan) &&
5181             !(adapter->flags2 & FLAG2_CRC_STRIPPING) &&
5182             (new_mtu > ETH_DATA_LEN)) {
5183                 e_err("Jumbo Frames not supported on 82579 when CRC stripping is disabled.\n");
5184                 return -EINVAL;
5185         }
5186
5187         /* 82573 Errata 17 */
5188         if (((adapter->hw.mac.type == e1000_82573) ||
5189              (adapter->hw.mac.type == e1000_82574)) &&
5190             (max_frame > ETH_FRAME_LEN + ETH_FCS_LEN)) {
5191                 adapter->flags2 |= FLAG2_DISABLE_ASPM_L1;
5192                 e1000e_disable_aspm(adapter->pdev, PCIE_LINK_STATE_L1);
5193         }
5194
5195         while (test_and_set_bit(__E1000_RESETTING, &adapter->state))
5196                 usleep_range(1000, 2000);
5197         /* e1000e_down -> e1000e_reset dependent on max_frame_size & mtu */
5198         adapter->max_frame_size = max_frame;
5199         e_info("changing MTU from %d to %d\n", netdev->mtu, new_mtu);
5200         netdev->mtu = new_mtu;
5201         if (netif_running(netdev))
5202                 e1000e_down(adapter);
5203
5204         /*
5205          * NOTE: netdev_alloc_skb reserves 16 bytes, and typically NET_IP_ALIGN
5206          * means we reserve 2 more, this pushes us to allocate from the next
5207          * larger slab size.
5208          * i.e. RXBUFFER_2048 --> size-4096 slab
5209          * However with the new *_jumbo_rx* routines, jumbo receives will use
5210          * fragmented skbs
5211          */
5212
5213         if (max_frame <= 2048)
5214                 adapter->rx_buffer_len = 2048;
5215         else
5216                 adapter->rx_buffer_len = 4096;
5217
5218         /* adjust allocation if LPE protects us, and we aren't using SBP */
5219         if ((max_frame == ETH_FRAME_LEN + ETH_FCS_LEN) ||
5220              (max_frame == ETH_FRAME_LEN + VLAN_HLEN + ETH_FCS_LEN))
5221                 adapter->rx_buffer_len = ETH_FRAME_LEN + VLAN_HLEN
5222                                          + ETH_FCS_LEN;
5223
5224         if (netif_running(netdev))
5225                 e1000e_up(adapter);
5226         else
5227                 e1000e_reset(adapter);
5228
5229         clear_bit(__E1000_RESETTING, &adapter->state);
5230
5231         return 0;
5232 }
5233
5234 static int e1000_mii_ioctl(struct net_device *netdev, struct ifreq *ifr,
5235                            int cmd)
5236 {
5237         struct e1000_adapter *adapter = netdev_priv(netdev);
5238         struct mii_ioctl_data *data = if_mii(ifr);
5239
5240         if (adapter->hw.phy.media_type != e1000_media_type_copper)
5241                 return -EOPNOTSUPP;
5242
5243         switch (cmd) {
5244         case SIOCGMIIPHY:
5245                 data->phy_id = adapter->hw.phy.addr;
5246                 break;
5247         case SIOCGMIIREG:
5248                 e1000_phy_read_status(adapter);
5249
5250                 switch (data->reg_num & 0x1F) {
5251                 case MII_BMCR:
5252                         data->val_out = adapter->phy_regs.bmcr;
5253                         break;
5254                 case MII_BMSR:
5255                         data->val_out = adapter->phy_regs.bmsr;
5256                         break;
5257                 case MII_PHYSID1:
5258                         data->val_out = (adapter->hw.phy.id >> 16);
5259                         break;
5260                 case MII_PHYSID2:
5261                         data->val_out = (adapter->hw.phy.id & 0xFFFF);
5262                         break;
5263                 case MII_ADVERTISE:
5264                         data->val_out = adapter->phy_regs.advertise;
5265                         break;
5266                 case MII_LPA:
5267                         data->val_out = adapter->phy_regs.lpa;
5268                         break;
5269                 case MII_EXPANSION:
5270                         data->val_out = adapter->phy_regs.expansion;
5271                         break;
5272                 case MII_CTRL1000:
5273                         data->val_out = adapter->phy_regs.ctrl1000;
5274                         break;
5275                 case MII_STAT1000:
5276                         data->val_out = adapter->phy_regs.stat1000;
5277                         break;
5278                 case MII_ESTATUS:
5279                         data->val_out = adapter->phy_regs.estatus;
5280                         break;
5281                 default:
5282                         return -EIO;
5283                 }
5284                 break;
5285         case SIOCSMIIREG:
5286         default:
5287                 return -EOPNOTSUPP;
5288         }
5289         return 0;
5290 }
5291
5292 static int e1000_ioctl(struct net_device *netdev, struct ifreq *ifr, int cmd)
5293 {
5294         switch (cmd) {
5295         case SIOCGMIIPHY:
5296         case SIOCGMIIREG:
5297         case SIOCSMIIREG:
5298                 return e1000_mii_ioctl(netdev, ifr, cmd);
5299         default:
5300                 return -EOPNOTSUPP;
5301         }
5302 }
5303
5304 static int e1000_init_phy_wakeup(struct e1000_adapter *adapter, u32 wufc)
5305 {
5306         struct e1000_hw *hw = &adapter->hw;
5307         u32 i, mac_reg;
5308         u16 phy_reg, wuc_enable;
5309         int retval = 0;
5310
5311         /* copy MAC RARs to PHY RARs */
5312         e1000_copy_rx_addrs_to_phy_ich8lan(hw);
5313
5314         retval = hw->phy.ops.acquire(hw);
5315         if (retval) {
5316                 e_err("Could not acquire PHY\n");
5317                 return retval;
5318         }
5319
5320         /* Enable access to wakeup registers on and set page to BM_WUC_PAGE */
5321         retval = e1000_enable_phy_wakeup_reg_access_bm(hw, &wuc_enable);
5322         if (retval)
5323                 goto out;
5324
5325         /* copy MAC MTA to PHY MTA - only needed for pchlan */
5326         for (i = 0; i < adapter->hw.mac.mta_reg_count; i++) {
5327                 mac_reg = E1000_READ_REG_ARRAY(hw, E1000_MTA, i);
5328                 hw->phy.ops.write_reg_page(hw, BM_MTA(i),
5329                                            (u16)(mac_reg & 0xFFFF));
5330                 hw->phy.ops.write_reg_page(hw, BM_MTA(i) + 1,
5331                                            (u16)((mac_reg >> 16) & 0xFFFF));
5332         }
5333
5334         /* configure PHY Rx Control register */
5335         hw->phy.ops.read_reg_page(&adapter->hw, BM_RCTL, &phy_reg);
5336         mac_reg = er32(RCTL);
5337         if (mac_reg & E1000_RCTL_UPE)
5338                 phy_reg |= BM_RCTL_UPE;
5339         if (mac_reg & E1000_RCTL_MPE)
5340                 phy_reg |= BM_RCTL_MPE;
5341         phy_reg &= ~(BM_RCTL_MO_MASK);
5342         if (mac_reg & E1000_RCTL_MO_3)
5343                 phy_reg |= (((mac_reg & E1000_RCTL_MO_3) >> E1000_RCTL_MO_SHIFT)
5344                                 << BM_RCTL_MO_SHIFT);
5345         if (mac_reg & E1000_RCTL_BAM)
5346                 phy_reg |= BM_RCTL_BAM;
5347         if (mac_reg & E1000_RCTL_PMCF)
5348                 phy_reg |= BM_RCTL_PMCF;
5349         mac_reg = er32(CTRL);
5350         if (mac_reg & E1000_CTRL_RFCE)
5351                 phy_reg |= BM_RCTL_RFCE;
5352         hw->phy.ops.write_reg_page(&adapter->hw, BM_RCTL, phy_reg);
5353
5354         /* enable PHY wakeup in MAC register */
5355         ew32(WUFC, wufc);
5356         ew32(WUC, E1000_WUC_PHY_WAKE | E1000_WUC_PME_EN);
5357
5358         /* configure and enable PHY wakeup in PHY registers */
5359         hw->phy.ops.write_reg_page(&adapter->hw, BM_WUFC, wufc);
5360         hw->phy.ops.write_reg_page(&adapter->hw, BM_WUC, E1000_WUC_PME_EN);
5361
5362         /* activate PHY wakeup */
5363         wuc_enable |= BM_WUC_ENABLE_BIT | BM_WUC_HOST_WU_BIT;
5364         retval = e1000_disable_phy_wakeup_reg_access_bm(hw, &wuc_enable);
5365         if (retval)
5366                 e_err("Could not set PHY Host Wakeup bit\n");
5367 out:
5368         hw->phy.ops.release(hw);
5369
5370         return retval;
5371 }
5372
5373 static int __e1000_shutdown(struct pci_dev *pdev, bool *enable_wake,
5374                             bool runtime)
5375 {
5376         struct net_device *netdev = pci_get_drvdata(pdev);
5377         struct e1000_adapter *adapter = netdev_priv(netdev);
5378         struct e1000_hw *hw = &adapter->hw;
5379         u32 ctrl, ctrl_ext, rctl, status;
5380         /* Runtime suspend should only enable wakeup for link changes */
5381         u32 wufc = runtime ? E1000_WUFC_LNKC : adapter->wol;
5382         int retval = 0;
5383
5384         netif_device_detach(netdev);
5385
5386         if (netif_running(netdev)) {
5387                 WARN_ON(test_bit(__E1000_RESETTING, &adapter->state));
5388                 e1000e_down(adapter);
5389                 e1000_free_irq(adapter);
5390         }
5391         e1000e_reset_interrupt_capability(adapter);
5392
5393         retval = pci_save_state(pdev);
5394         if (retval)
5395                 return retval;
5396
5397         status = er32(STATUS);
5398         if (status & E1000_STATUS_LU)
5399                 wufc &= ~E1000_WUFC_LNKC;
5400
5401         if (wufc) {
5402                 e1000_setup_rctl(adapter);
5403                 e1000e_set_rx_mode(netdev);
5404
5405                 /* turn on all-multi mode if wake on multicast is enabled */
5406                 if (wufc & E1000_WUFC_MC) {
5407                         rctl = er32(RCTL);
5408                         rctl |= E1000_RCTL_MPE;
5409                         ew32(RCTL, rctl);
5410                 }
5411
5412                 ctrl = er32(CTRL);
5413                 /* advertise wake from D3Cold */
5414                 #define E1000_CTRL_ADVD3WUC 0x00100000
5415                 /* phy power management enable */
5416                 #define E1000_CTRL_EN_PHY_PWR_MGMT 0x00200000
5417                 ctrl |= E1000_CTRL_ADVD3WUC;
5418                 if (!(adapter->flags2 & FLAG2_HAS_PHY_WAKEUP))
5419                         ctrl |= E1000_CTRL_EN_PHY_PWR_MGMT;
5420                 ew32(CTRL, ctrl);
5421
5422                 if (adapter->hw.phy.media_type == e1000_media_type_fiber ||
5423                     adapter->hw.phy.media_type ==
5424                     e1000_media_type_internal_serdes) {
5425                         /* keep the laser running in D3 */
5426                         ctrl_ext = er32(CTRL_EXT);
5427                         ctrl_ext |= E1000_CTRL_EXT_SDP3_DATA;
5428                         ew32(CTRL_EXT, ctrl_ext);
5429                 }
5430
5431                 if (adapter->flags & FLAG_IS_ICH)
5432                         e1000_suspend_workarounds_ich8lan(&adapter->hw);
5433
5434                 /* Allow time for pending master requests to run */
5435                 e1000e_disable_pcie_master(&adapter->hw);
5436
5437                 if (adapter->flags2 & FLAG2_HAS_PHY_WAKEUP) {
5438                         /* enable wakeup by the PHY */
5439                         retval = e1000_init_phy_wakeup(adapter, wufc);
5440                         if (retval)
5441                                 return retval;
5442                 } else {
5443                         /* enable wakeup by the MAC */
5444                         ew32(WUFC, wufc);
5445                         ew32(WUC, E1000_WUC_PME_EN);
5446                 }
5447         } else {
5448                 ew32(WUC, 0);
5449                 ew32(WUFC, 0);
5450         }
5451
5452         *enable_wake = !!wufc;
5453
5454         /* make sure adapter isn't asleep if manageability is enabled */
5455         if ((adapter->flags & FLAG_MNG_PT_ENABLED) ||
5456             (hw->mac.ops.check_mng_mode(hw)))
5457                 *enable_wake = true;
5458
5459         if (adapter->hw.phy.type == e1000_phy_igp_3)
5460                 e1000e_igp3_phy_powerdown_workaround_ich8lan(&adapter->hw);
5461
5462         /*
5463          * Release control of h/w to f/w.  If f/w is AMT enabled, this
5464          * would have already happened in close and is redundant.
5465          */
5466         e1000e_release_hw_control(adapter);
5467
5468         pci_disable_device(pdev);
5469
5470         return 0;
5471 }
5472
5473 static void e1000_power_off(struct pci_dev *pdev, bool sleep, bool wake)
5474 {
5475         if (sleep && wake) {
5476                 pci_prepare_to_sleep(pdev);
5477                 return;
5478         }
5479
5480         pci_wake_from_d3(pdev, wake);
5481         pci_set_power_state(pdev, PCI_D3hot);
5482 }
5483
5484 static void e1000_complete_shutdown(struct pci_dev *pdev, bool sleep,
5485                                     bool wake)
5486 {
5487         struct net_device *netdev = pci_get_drvdata(pdev);
5488         struct e1000_adapter *adapter = netdev_priv(netdev);
5489
5490         /*
5491          * The pci-e switch on some quad port adapters will report a
5492          * correctable error when the MAC transitions from D0 to D3.  To
5493          * prevent this we need to mask off the correctable errors on the
5494          * downstream port of the pci-e switch.
5495          */
5496         if (adapter->flags & FLAG_IS_QUAD_PORT) {
5497                 struct pci_dev *us_dev = pdev->bus->self;
5498                 int pos = pci_pcie_cap(us_dev);
5499                 u16 devctl;
5500
5501                 pci_read_config_word(us_dev, pos + PCI_EXP_DEVCTL, &devctl);
5502                 pci_write_config_word(us_dev, pos + PCI_EXP_DEVCTL,
5503                                       (devctl & ~PCI_EXP_DEVCTL_CERE));
5504
5505                 e1000_power_off(pdev, sleep, wake);
5506
5507                 pci_write_config_word(us_dev, pos + PCI_EXP_DEVCTL, devctl);
5508         } else {
5509                 e1000_power_off(pdev, sleep, wake);
5510         }
5511 }
5512
5513 #ifdef CONFIG_PCIEASPM
5514 static void __e1000e_disable_aspm(struct pci_dev *pdev, u16 state)
5515 {
5516         pci_disable_link_state_locked(pdev, state);
5517 }
5518 #else
5519 static void __e1000e_disable_aspm(struct pci_dev *pdev, u16 state)
5520 {
5521         int pos;
5522         u16 reg16;
5523
5524         /*
5525          * Both device and parent should have the same ASPM setting.
5526          * Disable ASPM in downstream component first and then upstream.
5527          */
5528         pos = pci_pcie_cap(pdev);
5529         pci_read_config_word(pdev, pos + PCI_EXP_LNKCTL, &reg16);
5530         reg16 &= ~state;
5531         pci_write_config_word(pdev, pos + PCI_EXP_LNKCTL, reg16);
5532
5533         if (!pdev->bus->self)
5534                 return;
5535
5536         pos = pci_pcie_cap(pdev->bus->self);
5537         pci_read_config_word(pdev->bus->self, pos + PCI_EXP_LNKCTL, &reg16);
5538         reg16 &= ~state;
5539         pci_write_config_word(pdev->bus->self, pos + PCI_EXP_LNKCTL, reg16);
5540 }
5541 #endif
5542 static void e1000e_disable_aspm(struct pci_dev *pdev, u16 state)
5543 {
5544         dev_info(&pdev->dev, "Disabling ASPM %s %s\n",
5545                  (state & PCIE_LINK_STATE_L0S) ? "L0s" : "",
5546                  (state & PCIE_LINK_STATE_L1) ? "L1" : "");
5547
5548         __e1000e_disable_aspm(pdev, state);
5549 }
5550
5551 #ifdef CONFIG_PM
5552 static bool e1000e_pm_ready(struct e1000_adapter *adapter)
5553 {
5554         return !!adapter->tx_ring->buffer_info;
5555 }
5556
5557 static int __e1000_resume(struct pci_dev *pdev)
5558 {
5559         struct net_device *netdev = pci_get_drvdata(pdev);
5560         struct e1000_adapter *adapter = netdev_priv(netdev);
5561         struct e1000_hw *hw = &adapter->hw;
5562         u16 aspm_disable_flag = 0;
5563         u32 err;
5564
5565         if (adapter->flags2 & FLAG2_DISABLE_ASPM_L0S)
5566                 aspm_disable_flag = PCIE_LINK_STATE_L0S;
5567         if (adapter->flags2 & FLAG2_DISABLE_ASPM_L1)
5568                 aspm_disable_flag |= PCIE_LINK_STATE_L1;
5569         if (aspm_disable_flag)
5570                 e1000e_disable_aspm(pdev, aspm_disable_flag);
5571
5572         pci_set_power_state(pdev, PCI_D0);
5573         pci_restore_state(pdev);
5574         pci_save_state(pdev);
5575
5576         e1000e_set_interrupt_capability(adapter);
5577         if (netif_running(netdev)) {
5578                 err = e1000_request_irq(adapter);
5579                 if (err)
5580                         return err;
5581         }
5582
5583         if (hw->mac.type == e1000_pch2lan)
5584                 e1000_resume_workarounds_pchlan(&adapter->hw);
5585
5586         e1000e_power_up_phy(adapter);
5587
5588         /* report the system wakeup cause from S3/S4 */
5589         if (adapter->flags2 & FLAG2_HAS_PHY_WAKEUP) {
5590                 u16 phy_data;
5591
5592                 e1e_rphy(&adapter->hw, BM_WUS, &phy_data);
5593                 if (phy_data) {
5594                         e_info("PHY Wakeup cause - %s\n",
5595                                 phy_data & E1000_WUS_EX ? "Unicast Packet" :
5596                                 phy_data & E1000_WUS_MC ? "Multicast Packet" :
5597                                 phy_data & E1000_WUS_BC ? "Broadcast Packet" :
5598                                 phy_data & E1000_WUS_MAG ? "Magic Packet" :
5599                                 phy_data & E1000_WUS_LNKC ?
5600                                 "Link Status Change" : "other");
5601                 }
5602                 e1e_wphy(&adapter->hw, BM_WUS, ~0);
5603         } else {
5604                 u32 wus = er32(WUS);
5605                 if (wus) {
5606                         e_info("MAC Wakeup cause - %s\n",
5607                                 wus & E1000_WUS_EX ? "Unicast Packet" :
5608                                 wus & E1000_WUS_MC ? "Multicast Packet" :
5609                                 wus & E1000_WUS_BC ? "Broadcast Packet" :
5610                                 wus & E1000_WUS_MAG ? "Magic Packet" :
5611                                 wus & E1000_WUS_LNKC ? "Link Status Change" :
5612                                 "other");
5613                 }
5614                 ew32(WUS, ~0);
5615         }
5616
5617         e1000e_reset(adapter);
5618
5619         e1000_init_manageability_pt(adapter);
5620
5621         if (netif_running(netdev))
5622                 e1000e_up(adapter);
5623
5624         netif_device_attach(netdev);
5625
5626         /*
5627          * If the controller has AMT, do not set DRV_LOAD until the interface
5628          * is up.  For all other cases, let the f/w know that the h/w is now
5629          * under the control of the driver.
5630          */
5631         if (!(adapter->flags & FLAG_HAS_AMT))
5632                 e1000e_get_hw_control(adapter);
5633
5634         return 0;
5635 }
5636
5637 #ifdef CONFIG_PM_SLEEP
5638 static int e1000_suspend(struct device *dev)
5639 {
5640         struct pci_dev *pdev = to_pci_dev(dev);
5641         int retval;
5642         bool wake;
5643
5644         retval = __e1000_shutdown(pdev, &wake, false);
5645         if (!retval)
5646                 e1000_complete_shutdown(pdev, true, wake);
5647
5648         return retval;
5649 }
5650
5651 static int e1000_resume(struct device *dev)
5652 {
5653         struct pci_dev *pdev = to_pci_dev(dev);
5654         struct net_device *netdev = pci_get_drvdata(pdev);
5655         struct e1000_adapter *adapter = netdev_priv(netdev);
5656
5657         if (e1000e_pm_ready(adapter))
5658                 adapter->idle_check = true;
5659
5660         return __e1000_resume(pdev);
5661 }
5662 #endif /* CONFIG_PM_SLEEP */
5663
5664 #ifdef CONFIG_PM_RUNTIME
5665 static int e1000_runtime_suspend(struct device *dev)
5666 {
5667         struct pci_dev *pdev = to_pci_dev(dev);
5668         struct net_device *netdev = pci_get_drvdata(pdev);
5669         struct e1000_adapter *adapter = netdev_priv(netdev);
5670
5671         if (e1000e_pm_ready(adapter)) {
5672                 bool wake;
5673
5674                 __e1000_shutdown(pdev, &wake, true);
5675         }
5676
5677         return 0;
5678 }
5679
5680 static int e1000_idle(struct device *dev)
5681 {
5682         struct pci_dev *pdev = to_pci_dev(dev);
5683         struct net_device *netdev = pci_get_drvdata(pdev);
5684         struct e1000_adapter *adapter = netdev_priv(netdev);
5685
5686         if (!e1000e_pm_ready(adapter))
5687                 return 0;
5688
5689         if (adapter->idle_check) {
5690                 adapter->idle_check = false;
5691                 if (!e1000e_has_link(adapter))
5692                         pm_schedule_suspend(dev, MSEC_PER_SEC);
5693         }
5694
5695         return -EBUSY;
5696 }
5697
5698 static int e1000_runtime_resume(struct device *dev)
5699 {
5700         struct pci_dev *pdev = to_pci_dev(dev);
5701         struct net_device *netdev = pci_get_drvdata(pdev);
5702         struct e1000_adapter *adapter = netdev_priv(netdev);
5703
5704         if (!e1000e_pm_ready(adapter))
5705                 return 0;
5706
5707         adapter->idle_check = !dev->power.runtime_auto;
5708         return __e1000_resume(pdev);
5709 }
5710 #endif /* CONFIG_PM_RUNTIME */
5711 #endif /* CONFIG_PM */
5712
5713 static void e1000_shutdown(struct pci_dev *pdev)
5714 {
5715         bool wake = false;
5716
5717         __e1000_shutdown(pdev, &wake, false);
5718
5719         if (system_state == SYSTEM_POWER_OFF)
5720                 e1000_complete_shutdown(pdev, false, wake);
5721 }
5722
5723 #ifdef CONFIG_NET_POLL_CONTROLLER
5724
5725 static irqreturn_t e1000_intr_msix(int irq, void *data)
5726 {
5727         struct net_device *netdev = data;
5728         struct e1000_adapter *adapter = netdev_priv(netdev);
5729
5730         if (adapter->msix_entries) {
5731                 int vector, msix_irq;
5732
5733                 vector = 0;
5734                 msix_irq = adapter->msix_entries[vector].vector;
5735                 disable_irq(msix_irq);
5736                 e1000_intr_msix_rx(msix_irq, netdev);
5737                 enable_irq(msix_irq);
5738
5739                 vector++;
5740                 msix_irq = adapter->msix_entries[vector].vector;
5741                 disable_irq(msix_irq);
5742                 e1000_intr_msix_tx(msix_irq, netdev);
5743                 enable_irq(msix_irq);
5744
5745                 vector++;
5746                 msix_irq = adapter->msix_entries[vector].vector;
5747                 disable_irq(msix_irq);
5748                 e1000_msix_other(msix_irq, netdev);
5749                 enable_irq(msix_irq);
5750         }
5751
5752         return IRQ_HANDLED;
5753 }
5754
5755 /*
5756  * Polling 'interrupt' - used by things like netconsole to send skbs
5757  * without having to re-enable interrupts. It's not called while
5758  * the interrupt routine is executing.
5759  */
5760 static void e1000_netpoll(struct net_device *netdev)
5761 {
5762         struct e1000_adapter *adapter = netdev_priv(netdev);
5763
5764         switch (adapter->int_mode) {
5765         case E1000E_INT_MODE_MSIX:
5766                 e1000_intr_msix(adapter->pdev->irq, netdev);
5767                 break;
5768         case E1000E_INT_MODE_MSI:
5769                 disable_irq(adapter->pdev->irq);
5770                 e1000_intr_msi(adapter->pdev->irq, netdev);
5771                 enable_irq(adapter->pdev->irq);
5772                 break;
5773         default: /* E1000E_INT_MODE_LEGACY */
5774                 disable_irq(adapter->pdev->irq);
5775                 e1000_intr(adapter->pdev->irq, netdev);
5776                 enable_irq(adapter->pdev->irq);
5777                 break;
5778         }
5779 }
5780 #endif
5781
5782 /**
5783  * e1000_io_error_detected - called when PCI error is detected
5784  * @pdev: Pointer to PCI device
5785  * @state: The current pci connection state
5786  *
5787  * This function is called after a PCI bus error affecting
5788  * this device has been detected.
5789  */
5790 static pci_ers_result_t e1000_io_error_detected(struct pci_dev *pdev,
5791                                                 pci_channel_state_t state)
5792 {
5793         struct net_device *netdev = pci_get_drvdata(pdev);
5794         struct e1000_adapter *adapter = netdev_priv(netdev);
5795
5796         netif_device_detach(netdev);
5797
5798         if (state == pci_channel_io_perm_failure)
5799                 return PCI_ERS_RESULT_DISCONNECT;
5800
5801         if (netif_running(netdev))
5802                 e1000e_down(adapter);
5803         pci_disable_device(pdev);
5804
5805         /* Request a slot slot reset. */
5806         return PCI_ERS_RESULT_NEED_RESET;
5807 }
5808
5809 /**
5810  * e1000_io_slot_reset - called after the pci bus has been reset.
5811  * @pdev: Pointer to PCI device
5812  *
5813  * Restart the card from scratch, as if from a cold-boot. Implementation
5814  * resembles the first-half of the e1000_resume routine.
5815  */
5816 static pci_ers_result_t e1000_io_slot_reset(struct pci_dev *pdev)
5817 {
5818         struct net_device *netdev = pci_get_drvdata(pdev);
5819         struct e1000_adapter *adapter = netdev_priv(netdev);
5820         struct e1000_hw *hw = &adapter->hw;
5821         u16 aspm_disable_flag = 0;
5822         int err;
5823         pci_ers_result_t result;
5824
5825         if (adapter->flags2 & FLAG2_DISABLE_ASPM_L0S)
5826                 aspm_disable_flag = PCIE_LINK_STATE_L0S;
5827         if (adapter->flags2 & FLAG2_DISABLE_ASPM_L1)
5828                 aspm_disable_flag |= PCIE_LINK_STATE_L1;
5829         if (aspm_disable_flag)
5830                 e1000e_disable_aspm(pdev, aspm_disable_flag);
5831
5832         err = pci_enable_device_mem(pdev);
5833         if (err) {
5834                 dev_err(&pdev->dev,
5835                         "Cannot re-enable PCI device after reset.\n");
5836                 result = PCI_ERS_RESULT_DISCONNECT;
5837         } else {
5838                 pci_set_master(pdev);
5839                 pdev->state_saved = true;
5840                 pci_restore_state(pdev);
5841
5842                 pci_enable_wake(pdev, PCI_D3hot, 0);
5843                 pci_enable_wake(pdev, PCI_D3cold, 0);
5844
5845                 e1000e_reset(adapter);
5846                 ew32(WUS, ~0);
5847                 result = PCI_ERS_RESULT_RECOVERED;
5848         }
5849
5850         pci_cleanup_aer_uncorrect_error_status(pdev);
5851
5852         return result;
5853 }
5854
5855 /**
5856  * e1000_io_resume - called when traffic can start flowing again.
5857  * @pdev: Pointer to PCI device
5858  *
5859  * This callback is called when the error recovery driver tells us that
5860  * its OK to resume normal operation. Implementation resembles the
5861  * second-half of the e1000_resume routine.
5862  */
5863 static void e1000_io_resume(struct pci_dev *pdev)
5864 {
5865         struct net_device *netdev = pci_get_drvdata(pdev);
5866         struct e1000_adapter *adapter = netdev_priv(netdev);
5867
5868         e1000_init_manageability_pt(adapter);
5869
5870         if (netif_running(netdev)) {
5871                 if (e1000e_up(adapter)) {
5872                         dev_err(&pdev->dev,
5873                                 "can't bring device back up after reset\n");
5874                         return;
5875                 }
5876         }
5877
5878         netif_device_attach(netdev);
5879
5880         /*
5881          * If the controller has AMT, do not set DRV_LOAD until the interface
5882          * is up.  For all other cases, let the f/w know that the h/w is now
5883          * under the control of the driver.
5884          */
5885         if (!(adapter->flags & FLAG_HAS_AMT))
5886                 e1000e_get_hw_control(adapter);
5887
5888 }
5889
5890 static void e1000_print_device_info(struct e1000_adapter *adapter)
5891 {
5892         struct e1000_hw *hw = &adapter->hw;
5893         struct net_device *netdev = adapter->netdev;
5894         u32 ret_val;
5895         u8 pba_str[E1000_PBANUM_LENGTH];
5896
5897         /* print bus type/speed/width info */
5898         e_info("(PCI Express:2.5GT/s:%s) %pM\n",
5899                /* bus width */
5900                ((hw->bus.width == e1000_bus_width_pcie_x4) ? "Width x4" :
5901                 "Width x1"),
5902                /* MAC address */
5903                netdev->dev_addr);
5904         e_info("Intel(R) PRO/%s Network Connection\n",
5905                (hw->phy.type == e1000_phy_ife) ? "10/100" : "1000");
5906         ret_val = e1000_read_pba_string_generic(hw, pba_str,
5907                                                 E1000_PBANUM_LENGTH);
5908         if (ret_val)
5909                 strncpy((char *)pba_str, "Unknown", sizeof(pba_str) - 1);
5910         e_info("MAC: %d, PHY: %d, PBA No: %s\n",
5911                hw->mac.type, hw->phy.type, pba_str);
5912 }
5913
5914 static void e1000_eeprom_checks(struct e1000_adapter *adapter)
5915 {
5916         struct e1000_hw *hw = &adapter->hw;
5917         int ret_val;
5918         u16 buf = 0;
5919
5920         if (hw->mac.type != e1000_82573)
5921                 return;
5922
5923         ret_val = e1000_read_nvm(hw, NVM_INIT_CONTROL2_REG, 1, &buf);
5924         if (!ret_val && (!(le16_to_cpu(buf) & (1 << 0)))) {
5925                 /* Deep Smart Power Down (DSPD) */
5926                 dev_warn(&adapter->pdev->dev,
5927                          "Warning: detected DSPD enabled in EEPROM\n");
5928         }
5929 }
5930
5931 static int e1000_set_features(struct net_device *netdev,
5932         netdev_features_t features)
5933 {
5934         struct e1000_adapter *adapter = netdev_priv(netdev);
5935         netdev_features_t changed = features ^ netdev->features;
5936
5937         if (changed & (NETIF_F_TSO | NETIF_F_TSO6))
5938                 adapter->flags |= FLAG_TSO_FORCE;
5939
5940         if (!(changed & (NETIF_F_HW_VLAN_RX | NETIF_F_HW_VLAN_TX |
5941                          NETIF_F_RXCSUM)))
5942                 return 0;
5943
5944         if (netif_running(netdev))
5945                 e1000e_reinit_locked(adapter);
5946         else
5947                 e1000e_reset(adapter);
5948
5949         return 0;
5950 }
5951
5952 static const struct net_device_ops e1000e_netdev_ops = {
5953         .ndo_open               = e1000_open,
5954         .ndo_stop               = e1000_close,
5955         .ndo_start_xmit         = e1000_xmit_frame,
5956         .ndo_get_stats64        = e1000e_get_stats64,
5957         .ndo_set_rx_mode        = e1000e_set_rx_mode,
5958         .ndo_set_mac_address    = e1000_set_mac,
5959         .ndo_change_mtu         = e1000_change_mtu,
5960         .ndo_do_ioctl           = e1000_ioctl,
5961         .ndo_tx_timeout         = e1000_tx_timeout,
5962         .ndo_validate_addr      = eth_validate_addr,
5963
5964         .ndo_vlan_rx_add_vid    = e1000_vlan_rx_add_vid,
5965         .ndo_vlan_rx_kill_vid   = e1000_vlan_rx_kill_vid,
5966 #ifdef CONFIG_NET_POLL_CONTROLLER
5967         .ndo_poll_controller    = e1000_netpoll,
5968 #endif
5969         .ndo_set_features = e1000_set_features,
5970 };
5971
5972 /**
5973  * e1000_probe - Device Initialization Routine
5974  * @pdev: PCI device information struct
5975  * @ent: entry in e1000_pci_tbl
5976  *
5977  * Returns 0 on success, negative on failure
5978  *
5979  * e1000_probe initializes an adapter identified by a pci_dev structure.
5980  * The OS initialization, configuring of the adapter private structure,
5981  * and a hardware reset occur.
5982  **/
5983 static int __devinit e1000_probe(struct pci_dev *pdev,
5984                                  const struct pci_device_id *ent)
5985 {
5986         struct net_device *netdev;
5987         struct e1000_adapter *adapter;
5988         struct e1000_hw *hw;
5989         const struct e1000_info *ei = e1000_info_tbl[ent->driver_data];
5990         resource_size_t mmio_start, mmio_len;
5991         resource_size_t flash_start, flash_len;
5992
5993         static int cards_found;
5994         u16 aspm_disable_flag = 0;
5995         int i, err, pci_using_dac;
5996         u16 eeprom_data = 0;
5997         u16 eeprom_apme_mask = E1000_EEPROM_APME;
5998
5999         if (ei->flags2 & FLAG2_DISABLE_ASPM_L0S)
6000                 aspm_disable_flag = PCIE_LINK_STATE_L0S;
6001         if (ei->flags2 & FLAG2_DISABLE_ASPM_L1)
6002                 aspm_disable_flag |= PCIE_LINK_STATE_L1;
6003         if (aspm_disable_flag)
6004                 e1000e_disable_aspm(pdev, aspm_disable_flag);
6005
6006         err = pci_enable_device_mem(pdev);
6007         if (err)
6008                 return err;
6009
6010         pci_using_dac = 0;
6011         err = dma_set_mask(&pdev->dev, DMA_BIT_MASK(64));
6012         if (!err) {
6013                 err = dma_set_coherent_mask(&pdev->dev, DMA_BIT_MASK(64));
6014                 if (!err)
6015                         pci_using_dac = 1;
6016         } else {
6017                 err = dma_set_mask(&pdev->dev, DMA_BIT_MASK(32));
6018                 if (err) {
6019                         err = dma_set_coherent_mask(&pdev->dev,
6020                                                     DMA_BIT_MASK(32));
6021                         if (err) {
6022                                 dev_err(&pdev->dev, "No usable DMA configuration, aborting\n");
6023                                 goto err_dma;
6024                         }
6025                 }
6026         }
6027
6028         err = pci_request_selected_regions_exclusive(pdev,
6029                                           pci_select_bars(pdev, IORESOURCE_MEM),
6030                                           e1000e_driver_name);
6031         if (err)
6032                 goto err_pci_reg;
6033
6034         /* AER (Advanced Error Reporting) hooks */
6035         pci_enable_pcie_error_reporting(pdev);
6036
6037         pci_set_master(pdev);
6038         /* PCI config space info */
6039         err = pci_save_state(pdev);
6040         if (err)
6041                 goto err_alloc_etherdev;
6042
6043         err = -ENOMEM;
6044         netdev = alloc_etherdev(sizeof(struct e1000_adapter));
6045         if (!netdev)
6046                 goto err_alloc_etherdev;
6047
6048         SET_NETDEV_DEV(netdev, &pdev->dev);
6049
6050         netdev->irq = pdev->irq;
6051
6052         pci_set_drvdata(pdev, netdev);
6053         adapter = netdev_priv(netdev);
6054         hw = &adapter->hw;
6055         adapter->netdev = netdev;
6056         adapter->pdev = pdev;
6057         adapter->ei = ei;
6058         adapter->pba = ei->pba;
6059         adapter->flags = ei->flags;
6060         adapter->flags2 = ei->flags2;
6061         adapter->hw.adapter = adapter;
6062         adapter->hw.mac.type = ei->mac;
6063         adapter->max_hw_frame_size = ei->max_hw_frame_size;
6064         adapter->msg_enable = (1 << NETIF_MSG_DRV | NETIF_MSG_PROBE) - 1;
6065
6066         mmio_start = pci_resource_start(pdev, 0);
6067         mmio_len = pci_resource_len(pdev, 0);
6068
6069         err = -EIO;
6070         adapter->hw.hw_addr = ioremap(mmio_start, mmio_len);
6071         if (!adapter->hw.hw_addr)
6072                 goto err_ioremap;
6073
6074         if ((adapter->flags & FLAG_HAS_FLASH) &&
6075             (pci_resource_flags(pdev, 1) & IORESOURCE_MEM)) {
6076                 flash_start = pci_resource_start(pdev, 1);
6077                 flash_len = pci_resource_len(pdev, 1);
6078                 adapter->hw.flash_address = ioremap(flash_start, flash_len);
6079                 if (!adapter->hw.flash_address)
6080                         goto err_flashmap;
6081         }
6082
6083         /* construct the net_device struct */
6084         netdev->netdev_ops              = &e1000e_netdev_ops;
6085         e1000e_set_ethtool_ops(netdev);
6086         netdev->watchdog_timeo          = 5 * HZ;
6087         netif_napi_add(netdev, &adapter->napi, e1000_clean, 64);
6088         strncpy(netdev->name, pci_name(pdev), sizeof(netdev->name) - 1);
6089
6090         netdev->mem_start = mmio_start;
6091         netdev->mem_end = mmio_start + mmio_len;
6092
6093         adapter->bd_number = cards_found++;
6094
6095         e1000e_check_options(adapter);
6096
6097         /* setup adapter struct */
6098         err = e1000_sw_init(adapter);
6099         if (err)
6100                 goto err_sw_init;
6101
6102         memcpy(&hw->mac.ops, ei->mac_ops, sizeof(hw->mac.ops));
6103         memcpy(&hw->nvm.ops, ei->nvm_ops, sizeof(hw->nvm.ops));
6104         memcpy(&hw->phy.ops, ei->phy_ops, sizeof(hw->phy.ops));
6105
6106         err = ei->get_variants(adapter);
6107         if (err)
6108                 goto err_hw_init;
6109
6110         if ((adapter->flags & FLAG_IS_ICH) &&
6111             (adapter->flags & FLAG_READ_ONLY_NVM))
6112                 e1000e_write_protect_nvm_ich8lan(&adapter->hw);
6113
6114         hw->mac.ops.get_bus_info(&adapter->hw);
6115
6116         adapter->hw.phy.autoneg_wait_to_complete = 0;
6117
6118         /* Copper options */
6119         if (adapter->hw.phy.media_type == e1000_media_type_copper) {
6120                 adapter->hw.phy.mdix = AUTO_ALL_MODES;
6121                 adapter->hw.phy.disable_polarity_correction = 0;
6122                 adapter->hw.phy.ms_type = e1000_ms_hw_default;
6123         }
6124
6125         if (e1000_check_reset_block(&adapter->hw))
6126                 e_info("PHY reset is blocked due to SOL/IDER session.\n");
6127
6128         /* Set initial default active device features */
6129         netdev->features = (NETIF_F_SG |
6130                             NETIF_F_HW_VLAN_RX |
6131                             NETIF_F_HW_VLAN_TX |
6132                             NETIF_F_TSO |
6133                             NETIF_F_TSO6 |
6134                             NETIF_F_RXCSUM |
6135                             NETIF_F_HW_CSUM);
6136
6137         /* Set user-changeable features (subset of all device features) */
6138         netdev->hw_features = netdev->features;
6139
6140         if (adapter->flags & FLAG_HAS_HW_VLAN_FILTER)
6141                 netdev->features |= NETIF_F_HW_VLAN_FILTER;
6142
6143         netdev->vlan_features |= (NETIF_F_SG |
6144                                   NETIF_F_TSO |
6145                                   NETIF_F_TSO6 |
6146                                   NETIF_F_HW_CSUM);
6147
6148         netdev->priv_flags |= IFF_UNICAST_FLT;
6149
6150         if (pci_using_dac) {
6151                 netdev->features |= NETIF_F_HIGHDMA;
6152                 netdev->vlan_features |= NETIF_F_HIGHDMA;
6153         }
6154
6155         if (e1000e_enable_mng_pass_thru(&adapter->hw))
6156                 adapter->flags |= FLAG_MNG_PT_ENABLED;
6157
6158         /*
6159          * before reading the NVM, reset the controller to
6160          * put the device in a known good starting state
6161          */
6162         adapter->hw.mac.ops.reset_hw(&adapter->hw);
6163
6164         /*
6165          * systems with ASPM and others may see the checksum fail on the first
6166          * attempt. Let's give it a few tries
6167          */
6168         for (i = 0;; i++) {
6169                 if (e1000_validate_nvm_checksum(&adapter->hw) >= 0)
6170                         break;
6171                 if (i == 2) {
6172                         e_err("The NVM Checksum Is Not Valid\n");
6173                         err = -EIO;
6174                         goto err_eeprom;
6175                 }
6176         }
6177
6178         e1000_eeprom_checks(adapter);
6179
6180         /* copy the MAC address */
6181         if (e1000e_read_mac_addr(&adapter->hw))
6182                 e_err("NVM Read Error while reading MAC address\n");
6183
6184         memcpy(netdev->dev_addr, adapter->hw.mac.addr, netdev->addr_len);
6185         memcpy(netdev->perm_addr, adapter->hw.mac.addr, netdev->addr_len);
6186
6187         if (!is_valid_ether_addr(netdev->perm_addr)) {
6188                 e_err("Invalid MAC Address: %pM\n", netdev->perm_addr);
6189                 err = -EIO;
6190                 goto err_eeprom;
6191         }
6192
6193         init_timer(&adapter->watchdog_timer);
6194         adapter->watchdog_timer.function = e1000_watchdog;
6195         adapter->watchdog_timer.data = (unsigned long) adapter;
6196
6197         init_timer(&adapter->phy_info_timer);
6198         adapter->phy_info_timer.function = e1000_update_phy_info;
6199         adapter->phy_info_timer.data = (unsigned long) adapter;
6200
6201         INIT_WORK(&adapter->reset_task, e1000_reset_task);
6202         INIT_WORK(&adapter->watchdog_task, e1000_watchdog_task);
6203         INIT_WORK(&adapter->downshift_task, e1000e_downshift_workaround);
6204         INIT_WORK(&adapter->update_phy_task, e1000e_update_phy_task);
6205         INIT_WORK(&adapter->print_hang_task, e1000_print_hw_hang);
6206
6207         /* Initialize link parameters. User can change them with ethtool */
6208         adapter->hw.mac.autoneg = 1;
6209         adapter->fc_autoneg = true;
6210         adapter->hw.fc.requested_mode = e1000_fc_default;
6211         adapter->hw.fc.current_mode = e1000_fc_default;
6212         adapter->hw.phy.autoneg_advertised = 0x2f;
6213
6214         /* ring size defaults */
6215         adapter->rx_ring->count = 256;
6216         adapter->tx_ring->count = 256;
6217
6218         /*
6219          * Initial Wake on LAN setting - If APM wake is enabled in
6220          * the EEPROM, enable the ACPI Magic Packet filter
6221          */
6222         if (adapter->flags & FLAG_APME_IN_WUC) {
6223                 /* APME bit in EEPROM is mapped to WUC.APME */
6224                 eeprom_data = er32(WUC);
6225                 eeprom_apme_mask = E1000_WUC_APME;
6226                 if ((hw->mac.type > e1000_ich10lan) &&
6227                     (eeprom_data & E1000_WUC_PHY_WAKE))
6228                         adapter->flags2 |= FLAG2_HAS_PHY_WAKEUP;
6229         } else if (adapter->flags & FLAG_APME_IN_CTRL3) {
6230                 if (adapter->flags & FLAG_APME_CHECK_PORT_B &&
6231                     (adapter->hw.bus.func == 1))
6232                         e1000_read_nvm(&adapter->hw,
6233                                 NVM_INIT_CONTROL3_PORT_B, 1, &eeprom_data);
6234                 else
6235                         e1000_read_nvm(&adapter->hw,
6236                                 NVM_INIT_CONTROL3_PORT_A, 1, &eeprom_data);
6237         }
6238
6239         /* fetch WoL from EEPROM */
6240         if (eeprom_data & eeprom_apme_mask)
6241                 adapter->eeprom_wol |= E1000_WUFC_MAG;
6242
6243         /*
6244          * now that we have the eeprom settings, apply the special cases
6245          * where the eeprom may be wrong or the board simply won't support
6246          * wake on lan on a particular port
6247          */
6248         if (!(adapter->flags & FLAG_HAS_WOL))
6249                 adapter->eeprom_wol = 0;
6250
6251         /* initialize the wol settings based on the eeprom settings */
6252         adapter->wol = adapter->eeprom_wol;
6253         device_set_wakeup_enable(&adapter->pdev->dev, adapter->wol);
6254
6255         /* save off EEPROM version number */
6256         e1000_read_nvm(&adapter->hw, 5, 1, &adapter->eeprom_vers);
6257
6258         /* reset the hardware with the new settings */
6259         e1000e_reset(adapter);
6260
6261         /*
6262          * If the controller has AMT, do not set DRV_LOAD until the interface
6263          * is up.  For all other cases, let the f/w know that the h/w is now
6264          * under the control of the driver.
6265          */
6266         if (!(adapter->flags & FLAG_HAS_AMT))
6267                 e1000e_get_hw_control(adapter);
6268
6269         strncpy(netdev->name, "eth%d", sizeof(netdev->name) - 1);
6270         err = register_netdev(netdev);
6271         if (err)
6272                 goto err_register;
6273
6274         /* carrier off reporting is important to ethtool even BEFORE open */
6275         netif_carrier_off(netdev);
6276
6277         e1000_print_device_info(adapter);
6278
6279         if (pci_dev_run_wake(pdev))
6280                 pm_runtime_put_noidle(&pdev->dev);
6281
6282         return 0;
6283
6284 err_register:
6285         if (!(adapter->flags & FLAG_HAS_AMT))
6286                 e1000e_release_hw_control(adapter);
6287 err_eeprom:
6288         if (!e1000_check_reset_block(&adapter->hw))
6289                 e1000_phy_hw_reset(&adapter->hw);
6290 err_hw_init:
6291         kfree(adapter->tx_ring);
6292         kfree(adapter->rx_ring);
6293 err_sw_init:
6294         if (adapter->hw.flash_address)
6295                 iounmap(adapter->hw.flash_address);
6296         e1000e_reset_interrupt_capability(adapter);
6297 err_flashmap:
6298         iounmap(adapter->hw.hw_addr);
6299 err_ioremap:
6300         free_netdev(netdev);
6301 err_alloc_etherdev:
6302         pci_release_selected_regions(pdev,
6303                                      pci_select_bars(pdev, IORESOURCE_MEM));
6304 err_pci_reg:
6305 err_dma:
6306         pci_disable_device(pdev);
6307         return err;
6308 }
6309
6310 /**
6311  * e1000_remove - Device Removal Routine
6312  * @pdev: PCI device information struct
6313  *
6314  * e1000_remove is called by the PCI subsystem to alert the driver
6315  * that it should release a PCI device.  The could be caused by a
6316  * Hot-Plug event, or because the driver is going to be removed from
6317  * memory.
6318  **/
6319 static void __devexit e1000_remove(struct pci_dev *pdev)
6320 {
6321         struct net_device *netdev = pci_get_drvdata(pdev);
6322         struct e1000_adapter *adapter = netdev_priv(netdev);
6323         bool down = test_bit(__E1000_DOWN, &adapter->state);
6324
6325         /*
6326          * The timers may be rescheduled, so explicitly disable them
6327          * from being rescheduled.
6328          */
6329         if (!down)
6330                 set_bit(__E1000_DOWN, &adapter->state);
6331         del_timer_sync(&adapter->watchdog_timer);
6332         del_timer_sync(&adapter->phy_info_timer);
6333
6334         cancel_work_sync(&adapter->reset_task);
6335         cancel_work_sync(&adapter->watchdog_task);
6336         cancel_work_sync(&adapter->downshift_task);
6337         cancel_work_sync(&adapter->update_phy_task);
6338         cancel_work_sync(&adapter->print_hang_task);
6339
6340         if (!(netdev->flags & IFF_UP))
6341                 e1000_power_down_phy(adapter);
6342
6343         /* Don't lie to e1000_close() down the road. */
6344         if (!down)
6345                 clear_bit(__E1000_DOWN, &adapter->state);
6346         unregister_netdev(netdev);
6347
6348         if (pci_dev_run_wake(pdev))
6349                 pm_runtime_get_noresume(&pdev->dev);
6350
6351         /*
6352          * Release control of h/w to f/w.  If f/w is AMT enabled, this
6353          * would have already happened in close and is redundant.
6354          */
6355         e1000e_release_hw_control(adapter);
6356
6357         e1000e_reset_interrupt_capability(adapter);
6358         kfree(adapter->tx_ring);
6359         kfree(adapter->rx_ring);
6360
6361         iounmap(adapter->hw.hw_addr);
6362         if (adapter->hw.flash_address)
6363                 iounmap(adapter->hw.flash_address);
6364         pci_release_selected_regions(pdev,
6365                                      pci_select_bars(pdev, IORESOURCE_MEM));
6366
6367         free_netdev(netdev);
6368
6369         /* AER disable */
6370         pci_disable_pcie_error_reporting(pdev);
6371
6372         pci_disable_device(pdev);
6373 }
6374
6375 /* PCI Error Recovery (ERS) */
6376 static struct pci_error_handlers e1000_err_handler = {
6377         .error_detected = e1000_io_error_detected,
6378         .slot_reset = e1000_io_slot_reset,
6379         .resume = e1000_io_resume,
6380 };
6381
6382 static DEFINE_PCI_DEVICE_TABLE(e1000_pci_tbl) = {
6383         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82571EB_COPPER), board_82571 },
6384         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82571EB_FIBER), board_82571 },
6385         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82571EB_QUAD_COPPER), board_82571 },
6386         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82571EB_QUAD_COPPER_LP), board_82571 },
6387         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82571EB_QUAD_FIBER), board_82571 },
6388         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82571EB_SERDES), board_82571 },
6389         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82571EB_SERDES_DUAL), board_82571 },
6390         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82571EB_SERDES_QUAD), board_82571 },
6391         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82571PT_QUAD_COPPER), board_82571 },
6392
6393         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82572EI), board_82572 },
6394         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82572EI_COPPER), board_82572 },
6395         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82572EI_FIBER), board_82572 },
6396         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82572EI_SERDES), board_82572 },
6397
6398         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82573E), board_82573 },
6399         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82573E_IAMT), board_82573 },
6400         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82573L), board_82573 },
6401
6402         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82574L), board_82574 },
6403         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82574LA), board_82574 },
6404         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82583V), board_82583 },
6405
6406         { PCI_VDEVICE(INTEL, E1000_DEV_ID_80003ES2LAN_COPPER_DPT),
6407           board_80003es2lan },
6408         { PCI_VDEVICE(INTEL, E1000_DEV_ID_80003ES2LAN_COPPER_SPT),
6409           board_80003es2lan },
6410         { PCI_VDEVICE(INTEL, E1000_DEV_ID_80003ES2LAN_SERDES_DPT),
6411           board_80003es2lan },
6412         { PCI_VDEVICE(INTEL, E1000_DEV_ID_80003ES2LAN_SERDES_SPT),
6413           board_80003es2lan },
6414
6415         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH8_IFE), board_ich8lan },
6416         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH8_IFE_G), board_ich8lan },
6417         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH8_IFE_GT), board_ich8lan },
6418         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH8_IGP_AMT), board_ich8lan },
6419         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH8_IGP_C), board_ich8lan },
6420         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH8_IGP_M), board_ich8lan },
6421         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH8_IGP_M_AMT), board_ich8lan },
6422         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH8_82567V_3), board_ich8lan },
6423
6424         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH9_IFE), board_ich9lan },
6425         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH9_IFE_G), board_ich9lan },
6426         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH9_IFE_GT), board_ich9lan },
6427         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH9_IGP_AMT), board_ich9lan },
6428         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH9_IGP_C), board_ich9lan },
6429         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH9_BM), board_ich9lan },
6430         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH9_IGP_M), board_ich9lan },
6431         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH9_IGP_M_AMT), board_ich9lan },
6432         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH9_IGP_M_V), board_ich9lan },
6433
6434         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH10_R_BM_LM), board_ich9lan },
6435         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH10_R_BM_LF), board_ich9lan },
6436         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH10_R_BM_V), board_ich9lan },
6437
6438         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH10_D_BM_LM), board_ich10lan },
6439         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH10_D_BM_LF), board_ich10lan },
6440         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH10_D_BM_V), board_ich10lan },
6441
6442         { PCI_VDEVICE(INTEL, E1000_DEV_ID_PCH_M_HV_LM), board_pchlan },
6443         { PCI_VDEVICE(INTEL, E1000_DEV_ID_PCH_M_HV_LC), board_pchlan },
6444         { PCI_VDEVICE(INTEL, E1000_DEV_ID_PCH_D_HV_DM), board_pchlan },
6445         { PCI_VDEVICE(INTEL, E1000_DEV_ID_PCH_D_HV_DC), board_pchlan },
6446
6447         { PCI_VDEVICE(INTEL, E1000_DEV_ID_PCH2_LV_LM), board_pch2lan },
6448         { PCI_VDEVICE(INTEL, E1000_DEV_ID_PCH2_LV_V), board_pch2lan },
6449
6450         { }     /* terminate list */
6451 };
6452 MODULE_DEVICE_TABLE(pci, e1000_pci_tbl);
6453
6454 #ifdef CONFIG_PM
6455 static const struct dev_pm_ops e1000_pm_ops = {
6456         SET_SYSTEM_SLEEP_PM_OPS(e1000_suspend, e1000_resume)
6457         SET_RUNTIME_PM_OPS(e1000_runtime_suspend,
6458                                 e1000_runtime_resume, e1000_idle)
6459 };
6460 #endif
6461
6462 /* PCI Device API Driver */
6463 static struct pci_driver e1000_driver = {
6464         .name     = e1000e_driver_name,
6465         .id_table = e1000_pci_tbl,
6466         .probe    = e1000_probe,
6467         .remove   = __devexit_p(e1000_remove),
6468 #ifdef CONFIG_PM
6469         .driver.pm = &e1000_pm_ops,
6470 #endif
6471         .shutdown = e1000_shutdown,
6472         .err_handler = &e1000_err_handler
6473 };
6474
6475 /**
6476  * e1000_init_module - Driver Registration Routine
6477  *
6478  * e1000_init_module is the first routine called when the driver is
6479  * loaded. All it does is register with the PCI subsystem.
6480  **/
6481 static int __init e1000_init_module(void)
6482 {
6483         int ret;
6484         pr_info("Intel(R) PRO/1000 Network Driver - %s\n",
6485                 e1000e_driver_version);
6486         pr_info("Copyright(c) 1999 - 2011 Intel Corporation.\n");
6487         ret = pci_register_driver(&e1000_driver);
6488
6489         return ret;
6490 }
6491 module_init(e1000_init_module);
6492
6493 /**
6494  * e1000_exit_module - Driver Exit Cleanup Routine
6495  *
6496  * e1000_exit_module is called just before the driver is removed
6497  * from memory.
6498  **/
6499 static void __exit e1000_exit_module(void)
6500 {
6501         pci_unregister_driver(&e1000_driver);
6502 }
6503 module_exit(e1000_exit_module);
6504
6505
6506 MODULE_AUTHOR("Intel Corporation, <linux.nics@intel.com>");
6507 MODULE_DESCRIPTION("Intel(R) PRO/1000 Network Driver");
6508 MODULE_LICENSE("GPL");
6509 MODULE_VERSION(DRV_VERSION);
6510
6511 /* e1000_main.c */