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

   1 /*******************************************************************************
   2
   3   Intel PRO/1000 Linux driver
   4   Copyright(c) 1999 - 2012 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 "1.10.6" DRV_EXTRAVERSION
  60 char e1000e_driver_name[] = "e1000e";
  61 const char e1000e_driver_version[] = DRV_VERSION;
  62
  63 #define DEFAULT_MSG_ENABLE (NETIF_MSG_DRV|NETIF_MSG_PROBE|NETIF_MSG_LINK)
  64 static int debug = -1;
  65 module_param(debug, int, 0);
  66 MODULE_PARM_DESC(debug, "Debug level (0=none,...,16=all)");
  67
  68 static void e1000e_disable_aspm(struct pci_dev *pdev, u16 state);
  69
  70 static const struct e1000_info *e1000_info_tbl[] = {
  71         [board_82571]           = &e1000_82571_info,
  72         [board_82572]           = &e1000_82572_info,
  73         [board_82573]           = &e1000_82573_info,
  74         [board_82574]           = &e1000_82574_info,
  75         [board_82583]           = &e1000_82583_info,
  76         [board_80003es2lan]     = &e1000_es2_info,
  77         [board_ich8lan]         = &e1000_ich8_info,
  78         [board_ich9lan]         = &e1000_ich9_info,
  79         [board_ich10lan]        = &e1000_ich10_info,
  80         [board_pchlan]          = &e1000_pch_info,
  81         [board_pch2lan]         = &e1000_pch2_info,
  82 };
  83
  84 struct e1000_reg_info {
  85         u32 ofs;
  86         char *name;
  87 };
  88
  89 #define E1000_RDFH      0x02410 /* Rx Data FIFO Head - RW */
  90 #define E1000_RDFT      0x02418 /* Rx Data FIFO Tail - RW */
  91 #define E1000_RDFHS     0x02420 /* Rx Data FIFO Head Saved - RW */
  92 #define E1000_RDFTS     0x02428 /* Rx Data FIFO Tail Saved - RW */
  93 #define E1000_RDFPC     0x02430 /* Rx Data FIFO Packet Count - RW */
  94
  95 #define E1000_TDFH      0x03410 /* Tx Data FIFO Head - RW */
  96 #define E1000_TDFT      0x03418 /* Tx Data FIFO Tail - RW */
  97 #define E1000_TDFHS     0x03420 /* Tx Data FIFO Head Saved - RW */
  98 #define E1000_TDFTS     0x03428 /* Tx Data FIFO Tail Saved - RW */
  99 #define E1000_TDFPC     0x03430 /* Tx Data FIFO Packet Count - RW */
 100
 101 static const struct e1000_reg_info e1000_reg_info_tbl[] = {
 102
 103         /* General Registers */
 104         {E1000_CTRL, "CTRL"},
 105         {E1000_STATUS, "STATUS"},
 106         {E1000_CTRL_EXT, "CTRL_EXT"},
 107
 108         /* Interrupt Registers */
 109         {E1000_ICR, "ICR"},
 110
 111         /* Rx Registers */
 112         {E1000_RCTL, "RCTL"},
 113         {E1000_RDLEN(0), "RDLEN"},
 114         {E1000_RDH(0), "RDH"},
 115         {E1000_RDT(0), "RDT"},
 116         {E1000_RDTR, "RDTR"},
 117         {E1000_RXDCTL(0), "RXDCTL"},
 118         {E1000_ERT, "ERT"},
 119         {E1000_RDBAL(0), "RDBAL"},
 120         {E1000_RDBAH(0), "RDBAH"},
 121         {E1000_RDFH, "RDFH"},
 122         {E1000_RDFT, "RDFT"},
 123         {E1000_RDFHS, "RDFHS"},
 124         {E1000_RDFTS, "RDFTS"},
 125         {E1000_RDFPC, "RDFPC"},
 126
 127         /* Tx Registers */
 128         {E1000_TCTL, "TCTL"},
 129         {E1000_TDBAL(0), "TDBAL"},
 130         {E1000_TDBAH(0), "TDBAH"},
 131         {E1000_TDLEN(0), "TDLEN"},
 132         {E1000_TDH(0), "TDH"},
 133         {E1000_TDT(0), "TDT"},
 134         {E1000_TIDV, "TIDV"},
 135         {E1000_TXDCTL(0), "TXDCTL"},
 136         {E1000_TADV, "TADV"},
 137         {E1000_TARC(0), "TARC"},
 138         {E1000_TDFH, "TDFH"},
 139         {E1000_TDFT, "TDFT"},
 140         {E1000_TDFHS, "TDFHS"},
 141         {E1000_TDFTS, "TDFTS"},
 142         {E1000_TDFPC, "TDFPC"},
 143
 144         /* List Terminator */
 145         {0, NULL}
 146 };
 147
 148 /*
 149  * e1000_regdump - register printout routine
 150  */
 151 static void e1000_regdump(struct e1000_hw *hw, struct e1000_reg_info *reginfo)
 152 {
 153         int n = 0;
 154         char rname[16];
 155         u32 regs[8];
 156
 157         switch (reginfo->ofs) {
 158         case E1000_RXDCTL(0):
 159                 for (n = 0; n < 2; n++)
 160                         regs[n] = __er32(hw, E1000_RXDCTL(n));
 161                 break;
 162         case E1000_TXDCTL(0):
 163                 for (n = 0; n < 2; n++)
 164                         regs[n] = __er32(hw, E1000_TXDCTL(n));
 165                 break;
 166         case E1000_TARC(0):
 167                 for (n = 0; n < 2; n++)
 168                         regs[n] = __er32(hw, E1000_TARC(n));
 169                 break;
 170         default:
 171                 pr_info("%-15s %08x\n",
 172                         reginfo->name, __er32(hw, reginfo->ofs));
 173                 return;
 174         }
 175
 176         snprintf(rname, 16, "%s%s", reginfo->name, "[0-1]");
 177         pr_info("%-15s %08x %08x\n", rname, regs[0], regs[1]);
 178 }
 179
 180 /*
 181  * e1000e_dump - Print registers, Tx-ring and Rx-ring
 182  */
 183 static void e1000e_dump(struct e1000_adapter *adapter)
 184 {
 185         struct net_device *netdev = adapter->netdev;
 186         struct e1000_hw *hw = &adapter->hw;
 187         struct e1000_reg_info *reginfo;
 188         struct e1000_ring *tx_ring = adapter->tx_ring;
 189         struct e1000_tx_desc *tx_desc;
 190         struct my_u0 {
 191                 __le64 a;
 192                 __le64 b;
 193         } *u0;
 194         struct e1000_buffer *buffer_info;
 195         struct e1000_ring *rx_ring = adapter->rx_ring;
 196         union e1000_rx_desc_packet_split *rx_desc_ps;
 197         union e1000_rx_desc_extended *rx_desc;
 198         struct my_u1 {
 199                 __le64 a;
 200                 __le64 b;
 201                 __le64 c;
 202                 __le64 d;
 203         } *u1;
 204         u32 staterr;
 205         int i = 0;
 206
 207         if (!netif_msg_hw(adapter))
 208                 return;
 209
 210         /* Print netdevice Info */
 211         if (netdev) {
 212                 dev_info(&adapter->pdev->dev, "Net device Info\n");
 213                 pr_info("Device Name     state            trans_start      last_rx\n");
 214                 pr_info("%-15s %016lX %016lX %016lX\n",
 215                         netdev->name, netdev->state, netdev->trans_start,
 216                         netdev->last_rx);
 217         }
 218
 219         /* Print Registers */
 220         dev_info(&adapter->pdev->dev, "Register Dump\n");
 221         pr_info(" Register Name   Value\n");
 222         for (reginfo = (struct e1000_reg_info *)e1000_reg_info_tbl;
 223              reginfo->name; reginfo++) {
 224                 e1000_regdump(hw, reginfo);
 225         }
 226
 227         /* Print Tx Ring Summary */
 228         if (!netdev || !netif_running(netdev))
 229                 return;
 230
 231         dev_info(&adapter->pdev->dev, "Tx Ring Summary\n");
 232         pr_info("Queue [NTU] [NTC] [bi(ntc)->dma  ] leng ntw timestamp\n");
 233         buffer_info = &tx_ring->buffer_info[tx_ring->next_to_clean];
 234         pr_info(" %5d %5X %5X %016llX %04X %3X %016llX\n",
 235                 0, tx_ring->next_to_use, tx_ring->next_to_clean,
 236                 (unsigned long long)buffer_info->dma,
 237                 buffer_info->length,
 238                 buffer_info->next_to_watch,
 239                 (unsigned long long)buffer_info->time_stamp);
 240
 241         /* Print Tx Ring */
 242         if (!netif_msg_tx_done(adapter))
 243                 goto rx_ring_summary;
 244
 245         dev_info(&adapter->pdev->dev, "Tx Ring Dump\n");
 246
 247         /* Transmit Descriptor Formats - DEXT[29] is 0 (Legacy) or 1 (Extended)
 248          *
 249          * Legacy Transmit Descriptor
 250          *   +--------------------------------------------------------------+
 251          * 0 |         Buffer Address [63:0] (Reserved on Write Back)       |
 252          *   +--------------------------------------------------------------+
 253          * 8 | Special  |    CSS     | Status |  CMD    |  CSO   |  Length  |
 254          *   +--------------------------------------------------------------+
 255          *   63       48 47        36 35    32 31     24 23    16 15        0
 256          *
 257          * Extended Context Descriptor (DTYP=0x0) for TSO or checksum offload
 258          *   63      48 47    40 39       32 31             16 15    8 7      0
 259          *   +----------------------------------------------------------------+
 260          * 0 |  TUCSE  | TUCS0  |   TUCSS   |     IPCSE       | IPCS0 | IPCSS |
 261          *   +----------------------------------------------------------------+
 262          * 8 |   MSS   | HDRLEN | RSV | STA | TUCMD | DTYP |      PAYLEN      |
 263          *   +----------------------------------------------------------------+
 264          *   63      48 47    40 39 36 35 32 31   24 23  20 19                0
 265          *
 266          * Extended Data Descriptor (DTYP=0x1)
 267          *   +----------------------------------------------------------------+
 268          * 0 |                     Buffer Address [63:0]                      |
 269          *   +----------------------------------------------------------------+
 270          * 8 | VLAN tag |  POPTS  | Rsvd | Status | Command | DTYP |  DTALEN  |
 271          *   +----------------------------------------------------------------+
 272          *   63       48 47     40 39  36 35    32 31     24 23  20 19        0
 273          */
 274         pr_info("Tl[desc]     [address 63:0  ] [SpeCssSCmCsLen] [bi->dma       ] leng  ntw timestamp        bi->skb <-- Legacy format\n");
 275         pr_info("Tc[desc]     [Ce CoCsIpceCoS] [MssHlRSCm0Plen] [bi->dma       ] leng  ntw timestamp        bi->skb <-- Ext Context format\n");
 276         pr_info("Td[desc]     [address 63:0  ] [VlaPoRSCm1Dlen] [bi->dma       ] leng  ntw timestamp        bi->skb <-- Ext Data format\n");
 277         for (i = 0; tx_ring->desc && (i < tx_ring->count); i++) {
 278                 const char *next_desc;
 279                 tx_desc = E1000_TX_DESC(*tx_ring, i);
 280                 buffer_info = &tx_ring->buffer_info[i];
 281                 u0 = (struct my_u0 *)tx_desc;
 282                 if (i == tx_ring->next_to_use && i == tx_ring->next_to_clean)
 283                         next_desc = " NTC/U";
 284                 else if (i == tx_ring->next_to_use)
 285                         next_desc = " NTU";
 286                 else if (i == tx_ring->next_to_clean)
 287                         next_desc = " NTC";
 288                 else
 289                         next_desc = "";
 290                 pr_info("T%c[0x%03X]    %016llX %016llX %016llX %04X  %3X %016llX %p%s\n",
 291                         (!(le64_to_cpu(u0->b) & (1 << 29)) ? 'l' :
 292                          ((le64_to_cpu(u0->b) & (1 << 20)) ? 'd' : 'c')),
 293                         i,
 294                         (unsigned long long)le64_to_cpu(u0->a),
 295                         (unsigned long long)le64_to_cpu(u0->b),
 296                         (unsigned long long)buffer_info->dma,
 297                         buffer_info->length, buffer_info->next_to_watch,
 298                         (unsigned long long)buffer_info->time_stamp,
 299                         buffer_info->skb, next_desc);
 300
 301                 if (netif_msg_pktdata(adapter) && buffer_info->dma != 0)
 302                         print_hex_dump(KERN_INFO, "", DUMP_PREFIX_ADDRESS,
 303                                        16, 1, phys_to_virt(buffer_info->dma),
 304                                        buffer_info->length, true);
 305         }
 306
 307         /* Print Rx Ring Summary */
 308 rx_ring_summary:
 309         dev_info(&adapter->pdev->dev, "Rx Ring Summary\n");
 310         pr_info("Queue [NTU] [NTC]\n");
 311         pr_info(" %5d %5X %5X\n",
 312                 0, rx_ring->next_to_use, rx_ring->next_to_clean);
 313
 314         /* Print Rx Ring */
 315         if (!netif_msg_rx_status(adapter))
 316                 return;
 317
 318         dev_info(&adapter->pdev->dev, "Rx Ring Dump\n");
 319         switch (adapter->rx_ps_pages) {
 320         case 1:
 321         case 2:
 322         case 3:
 323                 /* [Extended] Packet Split Receive Descriptor Format
 324                  *
 325                  *    +-----------------------------------------------------+
 326                  *  0 |                Buffer Address 0 [63:0]              |
 327                  *    +-----------------------------------------------------+
 328                  *  8 |                Buffer Address 1 [63:0]              |
 329                  *    +-----------------------------------------------------+
 330                  * 16 |                Buffer Address 2 [63:0]              |
 331                  *    +-----------------------------------------------------+
 332                  * 24 |                Buffer Address 3 [63:0]              |
 333                  *    +-----------------------------------------------------+
 334                  */
 335                 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");
 336                 /* [Extended] Receive Descriptor (Write-Back) Format
 337                  *
 338                  *   63       48 47    32 31     13 12    8 7    4 3        0
 339                  *   +------------------------------------------------------+
 340                  * 0 | Packet   | IP     |  Rsvd   | MRQ   | Rsvd | MRQ RSS |
 341                  *   | Checksum | Ident  |         | Queue |      |  Type   |
 342                  *   +------------------------------------------------------+
 343                  * 8 | VLAN Tag | Length | Extended Error | Extended Status |
 344                  *   +------------------------------------------------------+
 345                  *   63       48 47    32 31            20 19               0
 346                  */
 347                 pr_info("RWB[desc]      [ck ipid mrqhsh] [vl   l0 ee  es] [ l3  l2  l1 hs] [reserved      ] ---------------- [bi->skb] <-- Ext Rx Write-Back format\n");
 348                 for (i = 0; i < rx_ring->count; i++) {
 349                         const char *next_desc;
 350                         buffer_info = &rx_ring->buffer_info[i];
 351                         rx_desc_ps = E1000_RX_DESC_PS(*rx_ring, i);
 352                         u1 = (struct my_u1 *)rx_desc_ps;
 353                         staterr =
 354                             le32_to_cpu(rx_desc_ps->wb.middle.status_error);
 355
 356                         if (i == rx_ring->next_to_use)
 357                                 next_desc = " NTU";
 358                         else if (i == rx_ring->next_to_clean)
 359                                 next_desc = " NTC";
 360                         else
 361                                 next_desc = "";
 362
 363                         if (staterr & E1000_RXD_STAT_DD) {
 364                                 /* Descriptor Done */
 365                                 pr_info("%s[0x%03X]     %016llX %016llX %016llX %016llX ---------------- %p%s\n",
 366                                         "RWB", i,
 367                                         (unsigned long long)le64_to_cpu(u1->a),
 368                                         (unsigned long long)le64_to_cpu(u1->b),
 369                                         (unsigned long long)le64_to_cpu(u1->c),
 370                                         (unsigned long long)le64_to_cpu(u1->d),
 371                                         buffer_info->skb, next_desc);
 372                         } else {
 373                                 pr_info("%s[0x%03X]     %016llX %016llX %016llX %016llX %016llX %p%s\n",
 374                                         "R  ", i,
 375                                         (unsigned long long)le64_to_cpu(u1->a),
 376                                         (unsigned long long)le64_to_cpu(u1->b),
 377                                         (unsigned long long)le64_to_cpu(u1->c),
 378                                         (unsigned long long)le64_to_cpu(u1->d),
 379                                         (unsigned long long)buffer_info->dma,
 380                                         buffer_info->skb, next_desc);
 381
 382                                 if (netif_msg_pktdata(adapter))
 383                                         print_hex_dump(KERN_INFO, "",
 384                                                 DUMP_PREFIX_ADDRESS, 16, 1,
 385                                                 phys_to_virt(buffer_info->dma),
 386                                                 adapter->rx_ps_bsize0, true);
 387                         }
 388                 }
 389                 break;
 390         default:
 391         case 0:
 392                 /* Extended Receive Descriptor (Read) Format
 393                  *
 394                  *   +-----------------------------------------------------+
 395                  * 0 |                Buffer Address [63:0]                |
 396                  *   +-----------------------------------------------------+
 397                  * 8 |                      Reserved                       |
 398                  *   +-----------------------------------------------------+
 399                  */
 400                 pr_info("R  [desc]      [buf addr 63:0 ] [reserved 63:0 ] [bi->dma       ] [bi->skb] <-- Ext (Read) format\n");
 401                 /* Extended Receive Descriptor (Write-Back) Format
 402                  *
 403                  *   63       48 47    32 31    24 23            4 3        0
 404                  *   +------------------------------------------------------+
 405                  *   |     RSS Hash      |        |               |         |
 406                  * 0 +-------------------+  Rsvd  |   Reserved    | MRQ RSS |
 407                  *   | Packet   | IP     |        |               |  Type   |
 408                  *   | Checksum | Ident  |        |               |         |
 409                  *   +------------------------------------------------------+
 410                  * 8 | VLAN Tag | Length | Extended Error | Extended Status |
 411                  *   +------------------------------------------------------+
 412                  *   63       48 47    32 31            20 19               0
 413                  */
 414                 pr_info("RWB[desc]      [cs ipid    mrq] [vt   ln xe  xs] [bi->skb] <-- Ext (Write-Back) format\n");
 415
 416                 for (i = 0; i < rx_ring->count; i++) {
 417                         const char *next_desc;
 418
 419                         buffer_info = &rx_ring->buffer_info[i];
 420                         rx_desc = E1000_RX_DESC_EXT(*rx_ring, i);
 421                         u1 = (struct my_u1 *)rx_desc;
 422                         staterr = le32_to_cpu(rx_desc->wb.upper.status_error);
 423
 424                         if (i == rx_ring->next_to_use)
 425                                 next_desc = " NTU";
 426                         else if (i == rx_ring->next_to_clean)
 427                                 next_desc = " NTC";
 428                         else
 429                                 next_desc = "";
 430
 431                         if (staterr & E1000_RXD_STAT_DD) {
 432                                 /* Descriptor Done */
 433                                 pr_info("%s[0x%03X]     %016llX %016llX ---------------- %p%s\n",
 434                                         "RWB", i,
 435                                         (unsigned long long)le64_to_cpu(u1->a),
 436                                         (unsigned long long)le64_to_cpu(u1->b),
 437                                         buffer_info->skb, next_desc);
 438                         } else {
 439                                 pr_info("%s[0x%03X]     %016llX %016llX %016llX %p%s\n",
 440                                         "R  ", i,
 441                                         (unsigned long long)le64_to_cpu(u1->a),
 442                                         (unsigned long long)le64_to_cpu(u1->b),
 443                                         (unsigned long long)buffer_info->dma,
 444                                         buffer_info->skb, next_desc);
 445
 446                                 if (netif_msg_pktdata(adapter))
 447                                         print_hex_dump(KERN_INFO, "",
 448                                                        DUMP_PREFIX_ADDRESS, 16,
 449                                                        1,
 450                                                        phys_to_virt
 451                                                        (buffer_info->dma),
 452                                                        adapter->rx_buffer_len,
 453                                                        true);
 454                         }
 455                 }
 456         }
 457 }
 458
 459 /**
 460  * e1000_desc_unused - calculate if we have unused descriptors
 461  **/
 462 static int e1000_desc_unused(struct e1000_ring *ring)
 463 {
 464         if (ring->next_to_clean > ring->next_to_use)
 465                 return ring->next_to_clean - ring->next_to_use - 1;
 466
 467         return ring->count + ring->next_to_clean - ring->next_to_use - 1;
 468 }
 469
 470 /**
 471  * e1000_receive_skb - helper function to handle Rx indications
 472  * @adapter: board private structure
 473  * @status: descriptor status field as written by hardware
 474  * @vlan: descriptor vlan field as written by hardware (no le/be conversion)
 475  * @skb: pointer to sk_buff to be indicated to stack
 476  **/
 477 static void e1000_receive_skb(struct e1000_adapter *adapter,
 478                               struct net_device *netdev, struct sk_buff *skb,
 479                               u8 status, __le16 vlan)
 480 {
 481         u16 tag = le16_to_cpu(vlan);
 482         skb->protocol = eth_type_trans(skb, netdev);
 483
 484         if (status & E1000_RXD_STAT_VP)
 485                 __vlan_hwaccel_put_tag(skb, tag);
 486
 487         napi_gro_receive(&adapter->napi, skb);
 488 }
 489
 490 /**
 491  * e1000_rx_checksum - Receive Checksum Offload
 492  * @adapter: board private structure
 493  * @status_err: receive descriptor status and error fields
 494  * @csum: receive descriptor csum field
 495  * @sk_buff: socket buffer with received data
 496  **/
 497 static void e1000_rx_checksum(struct e1000_adapter *adapter, u32 status_err,
 498                               __le16 csum, struct sk_buff *skb)
 499 {
 500         u16 status = (u16)status_err;
 501         u8 errors = (u8)(status_err >> 24);
 502
 503         skb_checksum_none_assert(skb);
 504
 505         /* Rx checksum disabled */
 506         if (!(adapter->netdev->features & NETIF_F_RXCSUM))
 507                 return;
 508
 509         /* Ignore Checksum bit is set */
 510         if (status & E1000_RXD_STAT_IXSM)
 511                 return;
 512
 513         /* TCP/UDP checksum error bit is set */
 514         if (errors & E1000_RXD_ERR_TCPE) {
 515                 /* let the stack verify checksum errors */
 516                 adapter->hw_csum_err++;
 517                 return;
 518         }
 519
 520         /* TCP/UDP Checksum has not been calculated */
 521         if (!(status & (E1000_RXD_STAT_TCPCS | E1000_RXD_STAT_UDPCS)))
 522                 return;
 523
 524         /* It must be a TCP or UDP packet with a valid checksum */
 525         if (status & E1000_RXD_STAT_TCPCS) {
 526                 /* TCP checksum is good */
 527                 skb->ip_summed = CHECKSUM_UNNECESSARY;
 528         } else {
 529                 /*
 530                  * IP fragment with UDP payload
 531                  * Hardware complements the payload checksum, so we undo it
 532                  * and then put the value in host order for further stack use.
 533                  */
 534                 __sum16 sum = (__force __sum16)swab16((__force u16)csum);
 535                 skb->csum = csum_unfold(~sum);
 536                 skb->ip_summed = CHECKSUM_COMPLETE;
 537         }
 538         adapter->hw_csum_good++;
 539 }
 540
 541 static void e1000e_update_rdt_wa(struct e1000_ring *rx_ring, unsigned int i)
 542 {
 543         struct e1000_adapter *adapter = rx_ring->adapter;
 544         struct e1000_hw *hw = &adapter->hw;
 545         s32 ret_val = __ew32_prepare(hw);
 546
 547         writel(i, rx_ring->tail);
 548
 549         if (unlikely(!ret_val && (i != readl(rx_ring->tail)))) {
 550                 u32 rctl = er32(RCTL);
 551                 ew32(RCTL, rctl & ~E1000_RCTL_EN);
 552                 e_err("ME firmware caused invalid RDT - resetting\n");
 553                 schedule_work(&adapter->reset_task);
 554         }
 555 }
 556
 557 static void e1000e_update_tdt_wa(struct e1000_ring *tx_ring, unsigned int i)
 558 {
 559         struct e1000_adapter *adapter = tx_ring->adapter;
 560         struct e1000_hw *hw = &adapter->hw;
 561         s32 ret_val = __ew32_prepare(hw);
 562
 563         writel(i, tx_ring->tail);
 564
 565         if (unlikely(!ret_val && (i != readl(tx_ring->tail)))) {
 566                 u32 tctl = er32(TCTL);
 567                 ew32(TCTL, tctl & ~E1000_TCTL_EN);
 568                 e_err("ME firmware caused invalid TDT - resetting\n");
 569                 schedule_work(&adapter->reset_task);
 570         }
 571 }
 572
 573 /**
 574  * e1000_alloc_rx_buffers - Replace used receive buffers
 575  * @rx_ring: Rx descriptor ring
 576  **/
 577 static void e1000_alloc_rx_buffers(struct e1000_ring *rx_ring,
 578                                    int cleaned_count, gfp_t gfp)
 579 {
 580         struct e1000_adapter *adapter = rx_ring->adapter;
 581         struct net_device *netdev = adapter->netdev;
 582         struct pci_dev *pdev = adapter->pdev;
 583         union e1000_rx_desc_extended *rx_desc;
 584         struct e1000_buffer *buffer_info;
 585         struct sk_buff *skb;
 586         unsigned int i;
 587         unsigned int bufsz = adapter->rx_buffer_len;
 588
 589         i = rx_ring->next_to_use;
 590         buffer_info = &rx_ring->buffer_info[i];
 591
 592         while (cleaned_count--) {
 593                 skb = buffer_info->skb;
 594                 if (skb) {
 595                         skb_trim(skb, 0);
 596                         goto map_skb;
 597                 }
 598
 599                 skb = __netdev_alloc_skb_ip_align(netdev, bufsz, gfp);
 600                 if (!skb) {
 601                         /* Better luck next round */
 602                         adapter->alloc_rx_buff_failed++;
 603                         break;
 604                 }
 605
 606                 buffer_info->skb = skb;
 607 map_skb:
 608                 buffer_info->dma = dma_map_single(&pdev->dev, skb->data,
 609                                                   adapter->rx_buffer_len,
 610                                                   DMA_FROM_DEVICE);
 611                 if (dma_mapping_error(&pdev->dev, buffer_info->dma)) {
 612                         dev_err(&pdev->dev, "Rx DMA map failed\n");
 613                         adapter->rx_dma_failed++;
 614                         break;
 615                 }
 616
 617                 rx_desc = E1000_RX_DESC_EXT(*rx_ring, i);
 618                 rx_desc->read.buffer_addr = cpu_to_le64(buffer_info->dma);
 619
 620                 if (unlikely(!(i & (E1000_RX_BUFFER_WRITE - 1)))) {
 621                         /*
 622                          * Force memory writes to complete before letting h/w
 623                          * know there are new descriptors to fetch.  (Only
 624                          * applicable for weak-ordered memory model archs,
 625                          * such as IA-64).
 626                          */
 627                         wmb();
 628                         if (adapter->flags2 & FLAG2_PCIM2PCI_ARBITER_WA)
 629                                 e1000e_update_rdt_wa(rx_ring, i);
 630                         else
 631                                 writel(i, rx_ring->tail);
 632                 }
 633                 i++;
 634                 if (i == rx_ring->count)
 635                         i = 0;
 636                 buffer_info = &rx_ring->buffer_info[i];
 637         }
 638
 639         rx_ring->next_to_use = i;
 640 }
 641
 642 /**
 643  * e1000_alloc_rx_buffers_ps - Replace used receive buffers; packet split
 644  * @rx_ring: Rx descriptor ring
 645  **/
 646 static void e1000_alloc_rx_buffers_ps(struct e1000_ring *rx_ring,
 647                                       int cleaned_count, gfp_t gfp)
 648 {
 649         struct e1000_adapter *adapter = rx_ring->adapter;
 650         struct net_device *netdev = adapter->netdev;
 651         struct pci_dev *pdev = adapter->pdev;
 652         union e1000_rx_desc_packet_split *rx_desc;
 653         struct e1000_buffer *buffer_info;
 654         struct e1000_ps_page *ps_page;
 655         struct sk_buff *skb;
 656         unsigned int i, j;
 657
 658         i = rx_ring->next_to_use;
 659         buffer_info = &rx_ring->buffer_info[i];
 660
 661         while (cleaned_count--) {
 662                 rx_desc = E1000_RX_DESC_PS(*rx_ring, i);
 663
 664                 for (j = 0; j < PS_PAGE_BUFFERS; j++) {
 665                         ps_page = &buffer_info->ps_pages[j];
 666                         if (j >= adapter->rx_ps_pages) {
 667                                 /* all unused desc entries get hw null ptr */
 668                                 rx_desc->read.buffer_addr[j + 1] =
 669                                     ~cpu_to_le64(0);
 670                                 continue;
 671                         }
 672                         if (!ps_page->page) {
 673                                 ps_page->page = alloc_page(gfp);
 674                                 if (!ps_page->page) {
 675                                         adapter->alloc_rx_buff_failed++;
 676                                         goto no_buffers;
 677                                 }
 678                                 ps_page->dma = dma_map_page(&pdev->dev,
 679                                                             ps_page->page,
 680                                                             0, PAGE_SIZE,
 681                                                             DMA_FROM_DEVICE);
 682                                 if (dma_mapping_error(&pdev->dev,
 683                                                       ps_page->dma)) {
 684                                         dev_err(&adapter->pdev->dev,
 685                                                 "Rx DMA page map failed\n");
 686                                         adapter->rx_dma_failed++;
 687                                         goto no_buffers;
 688                                 }
 689                         }
 690                         /*
 691                          * Refresh the desc even if buffer_addrs
 692                          * didn't change because each write-back
 693                          * erases this info.
 694                          */
 695                         rx_desc->read.buffer_addr[j + 1] =
 696                             cpu_to_le64(ps_page->dma);
 697                 }
 698
 699                 skb = __netdev_alloc_skb_ip_align(netdev,
 700                                                   adapter->rx_ps_bsize0,
 701                                                   gfp);
 702
 703                 if (!skb) {
 704                         adapter->alloc_rx_buff_failed++;
 705                         break;
 706                 }
 707
 708                 buffer_info->skb = skb;
 709                 buffer_info->dma = dma_map_single(&pdev->dev, skb->data,
 710                                                   adapter->rx_ps_bsize0,
 711                                                   DMA_FROM_DEVICE);
 712                 if (dma_mapping_error(&pdev->dev, buffer_info->dma)) {
 713                         dev_err(&pdev->dev, "Rx DMA map failed\n");
 714                         adapter->rx_dma_failed++;
 715                         /* cleanup skb */
 716                         dev_kfree_skb_any(skb);
 717                         buffer_info->skb = NULL;
 718                         break;
 719                 }
 720
 721                 rx_desc->read.buffer_addr[0] = cpu_to_le64(buffer_info->dma);
 722
 723                 if (unlikely(!(i & (E1000_RX_BUFFER_WRITE - 1)))) {
 724                         /*
 725                          * Force memory writes to complete before letting h/w
 726                          * know there are new descriptors to fetch.  (Only
 727                          * applicable for weak-ordered memory model archs,
 728                          * such as IA-64).
 729                          */
 730                         wmb();
 731                         if (adapter->flags2 & FLAG2_PCIM2PCI_ARBITER_WA)
 732                                 e1000e_update_rdt_wa(rx_ring, i << 1);
 733                         else
 734                                 writel(i << 1, rx_ring->tail);
 735                 }
 736
 737                 i++;
 738                 if (i == rx_ring->count)
 739                         i = 0;
 740                 buffer_info = &rx_ring->buffer_info[i];
 741         }
 742
 743 no_buffers:
 744         rx_ring->next_to_use = i;
 745 }
 746
 747 /**
 748  * e1000_alloc_jumbo_rx_buffers - Replace used jumbo receive buffers
 749  * @rx_ring: Rx descriptor ring
 750  * @cleaned_count: number of buffers to allocate this pass
 751  **/
 752
 753 static void e1000_alloc_jumbo_rx_buffers(struct e1000_ring *rx_ring,
 754                                          int cleaned_count, gfp_t gfp)
 755 {
 756         struct e1000_adapter *adapter = rx_ring->adapter;
 757         struct net_device *netdev = adapter->netdev;
 758         struct pci_dev *pdev = adapter->pdev;
 759         union e1000_rx_desc_extended *rx_desc;
 760         struct e1000_buffer *buffer_info;
 761         struct sk_buff *skb;
 762         unsigned int i;
 763         unsigned int bufsz = 256 - 16 /* for skb_reserve */;
 764
 765         i = rx_ring->next_to_use;
 766         buffer_info = &rx_ring->buffer_info[i];
 767
 768         while (cleaned_count--) {
 769                 skb = buffer_info->skb;
 770                 if (skb) {
 771                         skb_trim(skb, 0);
 772                         goto check_page;
 773                 }
 774
 775                 skb = __netdev_alloc_skb_ip_align(netdev, bufsz, gfp);
 776                 if (unlikely(!skb)) {
 777                         /* Better luck next round */
 778                         adapter->alloc_rx_buff_failed++;
 779                         break;
 780                 }
 781
 782                 buffer_info->skb = skb;
 783 check_page:
 784                 /* allocate a new page if necessary */
 785                 if (!buffer_info->page) {
 786                         buffer_info->page = alloc_page(gfp);
 787                         if (unlikely(!buffer_info->page)) {
 788                                 adapter->alloc_rx_buff_failed++;
 789                                 break;
 790                         }
 791                 }
 792
 793                 if (!buffer_info->dma)
 794                         buffer_info->dma = dma_map_page(&pdev->dev,
 795                                                         buffer_info->page, 0,
 796                                                         PAGE_SIZE,
 797                                                         DMA_FROM_DEVICE);
 798
 799                 rx_desc = E1000_RX_DESC_EXT(*rx_ring, i);
 800                 rx_desc->read.buffer_addr = cpu_to_le64(buffer_info->dma);
 801
 802                 if (unlikely(++i == rx_ring->count))
 803                         i = 0;
 804                 buffer_info = &rx_ring->buffer_info[i];
 805         }
 806
 807         if (likely(rx_ring->next_to_use != i)) {
 808                 rx_ring->next_to_use = i;
 809                 if (unlikely(i-- == 0))
 810                         i = (rx_ring->count - 1);
 811
 812                 /* Force memory writes to complete before letting h/w
 813                  * know there are new descriptors to fetch.  (Only
 814                  * applicable for weak-ordered memory model archs,
 815                  * such as IA-64). */
 816                 wmb();
 817                 if (adapter->flags2 & FLAG2_PCIM2PCI_ARBITER_WA)
 818                         e1000e_update_rdt_wa(rx_ring, i);
 819                 else
 820                         writel(i, rx_ring->tail);
 821         }
 822 }
 823
 824 static inline void e1000_rx_hash(struct net_device *netdev, __le32 rss,
 825                                  struct sk_buff *skb)
 826 {
 827         if (netdev->features & NETIF_F_RXHASH)
 828                 skb->rxhash = le32_to_cpu(rss);
 829 }
 830
 831 /**
 832  * e1000_clean_rx_irq - Send received data up the network stack
 833  * @rx_ring: Rx descriptor ring
 834  *
 835  * the return value indicates whether actual cleaning was done, there
 836  * is no guarantee that everything was cleaned
 837  **/
 838 static bool e1000_clean_rx_irq(struct e1000_ring *rx_ring, int *work_done,
 839                                int work_to_do)
 840 {
 841         struct e1000_adapter *adapter = rx_ring->adapter;
 842         struct net_device *netdev = adapter->netdev;
 843         struct pci_dev *pdev = adapter->pdev;
 844         struct e1000_hw *hw = &adapter->hw;
 845         union e1000_rx_desc_extended *rx_desc, *next_rxd;
 846         struct e1000_buffer *buffer_info, *next_buffer;
 847         u32 length, staterr;
 848         unsigned int i;
 849         int cleaned_count = 0;
 850         bool cleaned = false;
 851         unsigned int total_rx_bytes = 0, total_rx_packets = 0;
 852
 853         i = rx_ring->next_to_clean;
 854         rx_desc = E1000_RX_DESC_EXT(*rx_ring, i);
 855         staterr = le32_to_cpu(rx_desc->wb.upper.status_error);
 856         buffer_info = &rx_ring->buffer_info[i];
 857
 858         while (staterr & E1000_RXD_STAT_DD) {
 859                 struct sk_buff *skb;
 860
 861                 if (*work_done >= work_to_do)
 862                         break;
 863                 (*work_done)++;
 864                 rmb();  /* read descriptor and rx_buffer_info after status DD */
 865
 866                 skb = buffer_info->skb;
 867                 buffer_info->skb = NULL;
 868
 869                 prefetch(skb->data - NET_IP_ALIGN);
 870
 871                 i++;
 872                 if (i == rx_ring->count)
 873                         i = 0;
 874                 next_rxd = E1000_RX_DESC_EXT(*rx_ring, i);
 875                 prefetch(next_rxd);
 876
 877                 next_buffer = &rx_ring->buffer_info[i];
 878
 879                 cleaned = true;
 880                 cleaned_count++;
 881                 dma_unmap_single(&pdev->dev,
 882                                  buffer_info->dma,
 883                                  adapter->rx_buffer_len,
 884                                  DMA_FROM_DEVICE);
 885                 buffer_info->dma = 0;
 886
 887                 length = le16_to_cpu(rx_desc->wb.upper.length);
 888
 889                 /*
 890                  * !EOP means multiple descriptors were used to store a single
 891                  * packet, if that's the case we need to toss it.  In fact, we
 892                  * need to toss every packet with the EOP bit clear and the
 893                  * next frame that _does_ have the EOP bit set, as it is by
 894                  * definition only a frame fragment
 895                  */
 896                 if (unlikely(!(staterr & E1000_RXD_STAT_EOP)))
 897                         adapter->flags2 |= FLAG2_IS_DISCARDING;
 898
 899                 if (adapter->flags2 & FLAG2_IS_DISCARDING) {
 900                         /* All receives must fit into a single buffer */
 901                         e_dbg("Receive packet consumed multiple buffers\n");
 902                         /* recycle */
 903                         buffer_info->skb = skb;
 904                         if (staterr & E1000_RXD_STAT_EOP)
 905                                 adapter->flags2 &= ~FLAG2_IS_DISCARDING;
 906                         goto next_desc;
 907                 }
 908
 909                 if (unlikely((staterr & E1000_RXDEXT_ERR_FRAME_ERR_MASK) &&
 910                              !(netdev->features & NETIF_F_RXALL))) {
 911                         /* recycle */
 912                         buffer_info->skb = skb;
 913                         goto next_desc;
 914                 }
 915
 916                 /* adjust length to remove Ethernet CRC */
 917                 if (!(adapter->flags2 & FLAG2_CRC_STRIPPING)) {
 918                         /* If configured to store CRC, don't subtract FCS,
 919                          * but keep the FCS bytes out of the total_rx_bytes
 920                          * counter
 921                          */
 922                         if (netdev->features & NETIF_F_RXFCS)
 923                                 total_rx_bytes -= 4;
 924                         else
 925                                 length -= 4;
 926                 }
 927
 928                 total_rx_bytes += length;
 929                 total_rx_packets++;
 930
 931                 /*
 932                  * code added for copybreak, this should improve
 933                  * performance for small packets with large amounts
 934                  * of reassembly being done in the stack
 935                  */
 936                 if (length < copybreak) {
 937                         struct sk_buff *new_skb =
 938                             netdev_alloc_skb_ip_align(netdev, length);
 939                         if (new_skb) {
 940                                 skb_copy_to_linear_data_offset(new_skb,
 941                                                                -NET_IP_ALIGN,
 942                                                                (skb->data -
 943                                                                 NET_IP_ALIGN),
 944                                                                (length +
 945                                                                 NET_IP_ALIGN));
 946                                 /* save the skb in buffer_info as good */
 947                                 buffer_info->skb = skb;
 948                                 skb = new_skb;
 949                         }
 950                         /* else just continue with the old one */
 951                 }
 952                 /* end copybreak code */
 953                 skb_put(skb, length);
 954
 955                 /* Receive Checksum Offload */
 956                 e1000_rx_checksum(adapter, staterr,
 957                                   rx_desc->wb.lower.hi_dword.csum_ip.csum, skb);
 958
 959                 e1000_rx_hash(netdev, rx_desc->wb.lower.hi_dword.rss, skb);
 960
 961                 e1000_receive_skb(adapter, netdev, skb, staterr,
 962                                   rx_desc->wb.upper.vlan);
 963
 964 next_desc:
 965                 rx_desc->wb.upper.status_error &= cpu_to_le32(~0xFF);
 966
 967                 /* return some buffers to hardware, one at a time is too slow */
 968                 if (cleaned_count >= E1000_RX_BUFFER_WRITE) {
 969                         adapter->alloc_rx_buf(rx_ring, cleaned_count,
 970                                               GFP_ATOMIC);
 971                         cleaned_count = 0;
 972                 }
 973
 974                 /* use prefetched values */
 975                 rx_desc = next_rxd;
 976                 buffer_info = next_buffer;
 977
 978                 staterr = le32_to_cpu(rx_desc->wb.upper.status_error);
 979         }
 980         rx_ring->next_to_clean = i;
 981
 982         cleaned_count = e1000_desc_unused(rx_ring);
 983         if (cleaned_count)
 984                 adapter->alloc_rx_buf(rx_ring, cleaned_count, GFP_ATOMIC);
 985
 986         adapter->total_rx_bytes += total_rx_bytes;
 987         adapter->total_rx_packets += total_rx_packets;
 988         return cleaned;
 989 }
 990
 991 static void e1000_put_txbuf(struct e1000_ring *tx_ring,
 992                             struct e1000_buffer *buffer_info)
 993 {
 994         struct e1000_adapter *adapter = tx_ring->adapter;
 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                 /*
1032                  * May be block on write-back, flush and detect again
1033                  * flush pending descriptor writebacks to memory
1034                  */
1035                 ew32(TIDV, adapter->tx_int_delay | E1000_TIDV_FPD);
1036                 /* execute the writes immediately */
1037                 e1e_flush();
1038                 /*
1039                  * Due to rare timing issues, write to TIDV again to ensure
1040                  * the write is successful
1041                  */
1042                 ew32(TIDV, adapter->tx_int_delay | E1000_TIDV_FPD);
1043                 /* execute the writes immediately */
1044                 e1e_flush();
1045                 adapter->tx_hang_recheck = true;
1046                 return;
1047         }
1048         /* Real hang detected */
1049         adapter->tx_hang_recheck = false;
1050         netif_stop_queue(netdev);
1051
1052         e1e_rphy(hw, PHY_STATUS, &phy_status);
1053         e1e_rphy(hw, PHY_1000T_STATUS, &phy_1000t_status);
1054         e1e_rphy(hw, PHY_EXT_STATUS, &phy_ext_status);
1055
1056         pci_read_config_word(adapter->pdev, PCI_STATUS, &pci_status);
1057
1058         /* detected Hardware unit hang */
1059         e_err("Detected Hardware Unit Hang:\n"
1060               "  TDH                  <%x>\n"
1061               "  TDT                  <%x>\n"
1062               "  next_to_use          <%x>\n"
1063               "  next_to_clean        <%x>\n"
1064               "buffer_info[next_to_clean]:\n"
1065               "  time_stamp           <%lx>\n"
1066               "  next_to_watch        <%x>\n"
1067               "  jiffies              <%lx>\n"
1068               "  next_to_watch.status <%x>\n"
1069               "MAC Status             <%x>\n"
1070               "PHY Status             <%x>\n"
1071               "PHY 1000BASE-T Status  <%x>\n"
1072               "PHY Extended Status    <%x>\n"
1073               "PCI Status             <%x>\n",
1074               readl(tx_ring->head),
1075               readl(tx_ring->tail),
1076               tx_ring->next_to_use,
1077               tx_ring->next_to_clean,
1078               tx_ring->buffer_info[eop].time_stamp,
1079               eop,
1080               jiffies,
1081               eop_desc->upper.fields.status,
1082               er32(STATUS),
1083               phy_status,
1084               phy_1000t_status,
1085               phy_ext_status,
1086               pci_status);
1087 }
1088
1089 /**
1090  * e1000_clean_tx_irq - Reclaim resources after transmit completes
1091  * @tx_ring: Tx descriptor ring
1092  *
1093  * the return value indicates whether actual cleaning was done, there
1094  * is no guarantee that everything was cleaned
1095  **/
1096 static bool e1000_clean_tx_irq(struct e1000_ring *tx_ring)
1097 {
1098         struct e1000_adapter *adapter = tx_ring->adapter;
1099         struct net_device *netdev = adapter->netdev;
1100         struct e1000_hw *hw = &adapter->hw;
1101         struct e1000_tx_desc *tx_desc, *eop_desc;
1102         struct e1000_buffer *buffer_info;
1103         unsigned int i, eop;
1104         unsigned int count = 0;
1105         unsigned int total_tx_bytes = 0, total_tx_packets = 0;
1106         unsigned int bytes_compl = 0, pkts_compl = 0;
1107
1108         i = tx_ring->next_to_clean;
1109         eop = tx_ring->buffer_info[i].next_to_watch;
1110         eop_desc = E1000_TX_DESC(*tx_ring, eop);
1111
1112         while ((eop_desc->upper.data & cpu_to_le32(E1000_TXD_STAT_DD)) &&
1113                (count < tx_ring->count)) {
1114                 bool cleaned = false;
1115                 rmb(); /* read buffer_info after eop_desc */
1116                 for (; !cleaned; count++) {
1117                         tx_desc = E1000_TX_DESC(*tx_ring, i);
1118                         buffer_info = &tx_ring->buffer_info[i];
1119                         cleaned = (i == eop);
1120
1121                         if (cleaned) {
1122                                 total_tx_packets += buffer_info->segs;
1123                                 total_tx_bytes += buffer_info->bytecount;
1124                                 if (buffer_info->skb) {
1125                                         bytes_compl += buffer_info->skb->len;
1126                                         pkts_compl++;
1127                                 }
1128                         }
1129
1130                         e1000_put_txbuf(tx_ring, buffer_info);
1131                         tx_desc->upper.data = 0;
1132
1133                         i++;
1134                         if (i == tx_ring->count)
1135                                 i = 0;
1136                 }
1137
1138                 if (i == tx_ring->next_to_use)
1139                         break;
1140                 eop = tx_ring->buffer_info[i].next_to_watch;
1141                 eop_desc = E1000_TX_DESC(*tx_ring, eop);
1142         }
1143
1144         tx_ring->next_to_clean = i;
1145
1146         netdev_completed_queue(netdev, pkts_compl, bytes_compl);
1147
1148 #define TX_WAKE_THRESHOLD 32
1149         if (count && netif_carrier_ok(netdev) &&
1150             e1000_desc_unused(tx_ring) >= TX_WAKE_THRESHOLD) {
1151                 /* Make sure that anybody stopping the queue after this
1152                  * sees the new next_to_clean.
1153                  */
1154                 smp_mb();
1155
1156                 if (netif_queue_stopped(netdev) &&
1157                     !(test_bit(__E1000_DOWN, &adapter->state))) {
1158                         netif_wake_queue(netdev);
1159                         ++adapter->restart_queue;
1160                 }
1161         }
1162
1163         if (adapter->detect_tx_hung) {
1164                 /*
1165                  * Detect a transmit hang in hardware, this serializes the
1166                  * check with the clearing of time_stamp and movement of i
1167                  */
1168                 adapter->detect_tx_hung = false;
1169                 if (tx_ring->buffer_info[i].time_stamp &&
1170                     time_after(jiffies, tx_ring->buffer_info[i].time_stamp
1171                                + (adapter->tx_timeout_factor * HZ)) &&
1172                     !(er32(STATUS) & E1000_STATUS_TXOFF))
1173                         schedule_work(&adapter->print_hang_task);
1174                 else
1175                         adapter->tx_hang_recheck = false;
1176         }
1177         adapter->total_tx_bytes += total_tx_bytes;
1178         adapter->total_tx_packets += total_tx_packets;
1179         return count < tx_ring->count;
1180 }
1181
1182 /**
1183  * e1000_clean_rx_irq_ps - Send received data up the network stack; packet split
1184  * @rx_ring: Rx descriptor ring
1185  *
1186  * the return value indicates whether actual cleaning was done, there
1187  * is no guarantee that everything was cleaned
1188  **/
1189 static bool e1000_clean_rx_irq_ps(struct e1000_ring *rx_ring, int *work_done,
1190                                   int work_to_do)
1191 {
1192         struct e1000_adapter *adapter = rx_ring->adapter;
1193         struct e1000_hw *hw = &adapter->hw;
1194         union e1000_rx_desc_packet_split *rx_desc, *next_rxd;
1195         struct net_device *netdev = adapter->netdev;
1196         struct pci_dev *pdev = adapter->pdev;
1197         struct e1000_buffer *buffer_info, *next_buffer;
1198         struct e1000_ps_page *ps_page;
1199         struct sk_buff *skb;
1200         unsigned int i, j;
1201         u32 length, staterr;
1202         int cleaned_count = 0;
1203         bool cleaned = false;
1204         unsigned int total_rx_bytes = 0, total_rx_packets = 0;
1205
1206         i = rx_ring->next_to_clean;
1207         rx_desc = E1000_RX_DESC_PS(*rx_ring, i);
1208         staterr = le32_to_cpu(rx_desc->wb.middle.status_error);
1209         buffer_info = &rx_ring->buffer_info[i];
1210
1211         while (staterr & E1000_RXD_STAT_DD) {
1212                 if (*work_done >= work_to_do)
1213                         break;
1214                 (*work_done)++;
1215                 skb = buffer_info->skb;
1216                 rmb();  /* read descriptor and rx_buffer_info after status DD */
1217
1218                 /* in the packet split case this is header only */
1219                 prefetch(skb->data - NET_IP_ALIGN);
1220
1221                 i++;
1222                 if (i == rx_ring->count)
1223                         i = 0;
1224                 next_rxd = E1000_RX_DESC_PS(*rx_ring, i);
1225                 prefetch(next_rxd);
1226
1227                 next_buffer = &rx_ring->buffer_info[i];
1228
1229                 cleaned = true;
1230                 cleaned_count++;
1231                 dma_unmap_single(&pdev->dev, buffer_info->dma,
1232                                  adapter->rx_ps_bsize0, DMA_FROM_DEVICE);
1233                 buffer_info->dma = 0;
1234
1235                 /* see !EOP comment in other Rx routine */
1236                 if (!(staterr & E1000_RXD_STAT_EOP))
1237                         adapter->flags2 |= FLAG2_IS_DISCARDING;
1238
1239                 if (adapter->flags2 & FLAG2_IS_DISCARDING) {
1240                         e_dbg("Packet Split buffers didn't pick up the full packet\n");
1241                         dev_kfree_skb_irq(skb);
1242                         if (staterr & E1000_RXD_STAT_EOP)
1243                                 adapter->flags2 &= ~FLAG2_IS_DISCARDING;
1244                         goto next_desc;
1245                 }
1246
1247                 if (unlikely((staterr & E1000_RXDEXT_ERR_FRAME_ERR_MASK) &&
1248                              !(netdev->features & NETIF_F_RXALL))) {
1249                         dev_kfree_skb_irq(skb);
1250                         goto next_desc;
1251                 }
1252
1253                 length = le16_to_cpu(rx_desc->wb.middle.length0);
1254
1255                 if (!length) {
1256                         e_dbg("Last part of the packet spanning multiple descriptors\n");
1257                         dev_kfree_skb_irq(skb);
1258                         goto next_desc;
1259                 }
1260
1261                 /* Good Receive */
1262                 skb_put(skb, length);
1263
1264                 {
1265                         /*
1266                          * this looks ugly, but it seems compiler issues make
1267                          * it more efficient than reusing j
1268                          */
1269                         int l1 = le16_to_cpu(rx_desc->wb.upper.length[0]);
1270
1271                         /*
1272                          * page alloc/put takes too long and effects small
1273                          * packet throughput, so unsplit small packets and
1274                          * save the alloc/put only valid in softirq (napi)
1275                          * context to call kmap_*
1276                          */
1277                         if (l1 && (l1 <= copybreak) &&
1278                             ((length + l1) <= adapter->rx_ps_bsize0)) {
1279                                 u8 *vaddr;
1280
1281                                 ps_page = &buffer_info->ps_pages[0];
1282
1283                                 /*
1284                                  * there is no documentation about how to call
1285                                  * kmap_atomic, so we can't hold the mapping
1286                                  * very long
1287                                  */
1288                                 dma_sync_single_for_cpu(&pdev->dev,
1289                                                         ps_page->dma,
1290                                                         PAGE_SIZE,
1291                                                         DMA_FROM_DEVICE);
1292                                 vaddr = kmap_atomic(ps_page->page);
1293                                 memcpy(skb_tail_pointer(skb), vaddr, l1);
1294                                 kunmap_atomic(vaddr);
1295                                 dma_sync_single_for_device(&pdev->dev,
1296                                                            ps_page->dma,
1297                                                            PAGE_SIZE,
1298                                                            DMA_FROM_DEVICE);
1299
1300                                 /* remove the CRC */
1301                                 if (!(adapter->flags2 & FLAG2_CRC_STRIPPING)) {
1302                                         if (!(netdev->features & NETIF_F_RXFCS))
1303                                                 l1 -= 4;
1304                                 }
1305
1306                                 skb_put(skb, l1);
1307                                 goto copydone;
1308                         } /* if */
1309                 }
1310
1311                 for (j = 0; j < PS_PAGE_BUFFERS; j++) {
1312                         length = le16_to_cpu(rx_desc->wb.upper.length[j]);
1313                         if (!length)
1314                                 break;
1315
1316                         ps_page = &buffer_info->ps_pages[j];
1317                         dma_unmap_page(&pdev->dev, ps_page->dma, PAGE_SIZE,
1318                                        DMA_FROM_DEVICE);
1319                         ps_page->dma = 0;
1320                         skb_fill_page_desc(skb, j, ps_page->page, 0, length);
1321                         ps_page->page = NULL;
1322                         skb->len += length;
1323                         skb->data_len += length;
1324                         skb->truesize += PAGE_SIZE;
1325                 }
1326
1327                 /* strip the ethernet crc, problem is we're using pages now so
1328                  * this whole operation can get a little cpu intensive
1329                  */
1330                 if (!(adapter->flags2 & FLAG2_CRC_STRIPPING)) {
1331                         if (!(netdev->features & NETIF_F_RXFCS))
1332                                 pskb_trim(skb, skb->len - 4);
1333                 }
1334
1335 copydone:
1336                 total_rx_bytes += skb->len;
1337                 total_rx_packets++;
1338
1339                 e1000_rx_checksum(adapter, staterr,
1340                                   rx_desc->wb.lower.hi_dword.csum_ip.csum, skb);
1341
1342                 e1000_rx_hash(netdev, rx_desc->wb.lower.hi_dword.rss, skb);
1343
1344                 if (rx_desc->wb.upper.header_status &
1345                            cpu_to_le16(E1000_RXDPS_HDRSTAT_HDRSP))
1346                         adapter->rx_hdr_split++;
1347
1348                 e1000_receive_skb(adapter, netdev, skb,
1349                                   staterr, rx_desc->wb.middle.vlan);
1350
1351 next_desc:
1352                 rx_desc->wb.middle.status_error &= cpu_to_le32(~0xFF);
1353                 buffer_info->skb = NULL;
1354
1355                 /* return some buffers to hardware, one at a time is too slow */
1356                 if (cleaned_count >= E1000_RX_BUFFER_WRITE) {
1357                         adapter->alloc_rx_buf(rx_ring, cleaned_count,
1358                                               GFP_ATOMIC);
1359                         cleaned_count = 0;
1360                 }
1361
1362                 /* use prefetched values */
1363                 rx_desc = next_rxd;
1364                 buffer_info = next_buffer;
1365
1366                 staterr = le32_to_cpu(rx_desc->wb.middle.status_error);
1367         }
1368         rx_ring->next_to_clean = i;
1369
1370         cleaned_count = e1000_desc_unused(rx_ring);
1371         if (cleaned_count)
1372                 adapter->alloc_rx_buf(rx_ring, cleaned_count, GFP_ATOMIC);
1373
1374         adapter->total_rx_bytes += total_rx_bytes;
1375         adapter->total_rx_packets += total_rx_packets;
1376         return cleaned;
1377 }
1378
1379 /**
1380  * e1000_consume_page - helper function
1381  **/
1382 static void e1000_consume_page(struct e1000_buffer *bi, struct sk_buff *skb,
1383                                u16 length)
1384 {
1385         bi->page = NULL;
1386         skb->len += length;
1387         skb->data_len += length;
1388         skb->truesize += PAGE_SIZE;
1389 }
1390
1391 /**
1392  * e1000_clean_jumbo_rx_irq - Send received data up the network stack; legacy
1393  * @adapter: board private structure
1394  *
1395  * the return value indicates whether actual cleaning was done, there
1396  * is no guarantee that everything was cleaned
1397  **/
1398 static bool e1000_clean_jumbo_rx_irq(struct e1000_ring *rx_ring, int *work_done,
1399                                      int work_to_do)
1400 {
1401         struct e1000_adapter *adapter = rx_ring->adapter;
1402         struct net_device *netdev = adapter->netdev;
1403         struct pci_dev *pdev = adapter->pdev;
1404         union e1000_rx_desc_extended *rx_desc, *next_rxd;
1405         struct e1000_buffer *buffer_info, *next_buffer;
1406         u32 length, staterr;
1407         unsigned int i;
1408         int cleaned_count = 0;
1409         bool cleaned = false;
1410         unsigned int total_rx_bytes=0, total_rx_packets=0;
1411
1412         i = rx_ring->next_to_clean;
1413         rx_desc = E1000_RX_DESC_EXT(*rx_ring, i);
1414         staterr = le32_to_cpu(rx_desc->wb.upper.status_error);
1415         buffer_info = &rx_ring->buffer_info[i];
1416
1417         while (staterr & E1000_RXD_STAT_DD) {
1418                 struct sk_buff *skb;
1419
1420                 if (*work_done >= work_to_do)
1421                         break;
1422                 (*work_done)++;
1423                 rmb();  /* read descriptor and rx_buffer_info after status DD */
1424
1425                 skb = buffer_info->skb;
1426                 buffer_info->skb = NULL;
1427
1428                 ++i;
1429                 if (i == rx_ring->count)
1430                         i = 0;
1431                 next_rxd = E1000_RX_DESC_EXT(*rx_ring, i);
1432                 prefetch(next_rxd);
1433
1434                 next_buffer = &rx_ring->buffer_info[i];
1435
1436                 cleaned = true;
1437                 cleaned_count++;
1438                 dma_unmap_page(&pdev->dev, buffer_info->dma, PAGE_SIZE,
1439                                DMA_FROM_DEVICE);
1440                 buffer_info->dma = 0;
1441
1442                 length = le16_to_cpu(rx_desc->wb.upper.length);
1443
1444                 /* errors is only valid for DD + EOP descriptors */
1445                 if (unlikely((staterr & E1000_RXD_STAT_EOP) &&
1446                              ((staterr & E1000_RXDEXT_ERR_FRAME_ERR_MASK) &&
1447                               !(netdev->features & NETIF_F_RXALL)))) {
1448                         /* recycle both page and skb */
1449                         buffer_info->skb = skb;
1450                         /* an error means any chain goes out the window too */
1451                         if (rx_ring->rx_skb_top)
1452                                 dev_kfree_skb_irq(rx_ring->rx_skb_top);
1453                         rx_ring->rx_skb_top = NULL;
1454                         goto next_desc;
1455                 }
1456
1457 #define rxtop (rx_ring->rx_skb_top)
1458                 if (!(staterr & E1000_RXD_STAT_EOP)) {
1459                         /* this descriptor is only the beginning (or middle) */
1460                         if (!rxtop) {
1461                                 /* this is the beginning of a chain */
1462                                 rxtop = skb;
1463                                 skb_fill_page_desc(rxtop, 0, buffer_info->page,
1464                                                    0, length);
1465                         } else {
1466                                 /* this is the middle of a chain */
1467                                 skb_fill_page_desc(rxtop,
1468                                     skb_shinfo(rxtop)->nr_frags,
1469                                     buffer_info->page, 0, length);
1470                                 /* re-use the skb, only consumed the page */
1471                                 buffer_info->skb = skb;
1472                         }
1473                         e1000_consume_page(buffer_info, rxtop, length);
1474                         goto next_desc;
1475                 } else {
1476                         if (rxtop) {
1477                                 /* end of the chain */
1478                                 skb_fill_page_desc(rxtop,
1479                                     skb_shinfo(rxtop)->nr_frags,
1480                                     buffer_info->page, 0, length);
1481                                 /* re-use the current skb, we only consumed the
1482                                  * page */
1483                                 buffer_info->skb = skb;
1484                                 skb = rxtop;
1485                                 rxtop = NULL;
1486                                 e1000_consume_page(buffer_info, skb, length);
1487                         } else {
1488                                 /* no chain, got EOP, this buf is the packet
1489                                  * copybreak to save the put_page/alloc_page */
1490                                 if (length <= copybreak &&
1491                                     skb_tailroom(skb) >= length) {
1492                                         u8 *vaddr;
1493                                         vaddr = kmap_atomic(buffer_info->page);
1494                                         memcpy(skb_tail_pointer(skb), vaddr,
1495                                                length);
1496                                         kunmap_atomic(vaddr);
1497                                         /* re-use the page, so don't erase
1498                                          * buffer_info->page */
1499                                         skb_put(skb, length);
1500                                 } else {
1501                                         skb_fill_page_desc(skb, 0,
1502                                                            buffer_info->page, 0,
1503                                                            length);
1504                                         e1000_consume_page(buffer_info, skb,
1505                                                            length);
1506                                 }
1507                         }
1508                 }
1509
1510                 /* Receive Checksum Offload XXX recompute due to CRC strip? */
1511                 e1000_rx_checksum(adapter, staterr,
1512                                   rx_desc->wb.lower.hi_dword.csum_ip.csum, skb);
1513
1514                 e1000_rx_hash(netdev, rx_desc->wb.lower.hi_dword.rss, skb);
1515
1516                 /* probably a little skewed due to removing CRC */
1517                 total_rx_bytes += skb->len;
1518                 total_rx_packets++;
1519
1520                 /* eth type trans needs skb->data to point to something */
1521                 if (!pskb_may_pull(skb, ETH_HLEN)) {
1522                         e_err("pskb_may_pull failed.\n");
1523                         dev_kfree_skb_irq(skb);
1524                         goto next_desc;
1525                 }
1526
1527                 e1000_receive_skb(adapter, netdev, skb, staterr,
1528                                   rx_desc->wb.upper.vlan);
1529
1530 next_desc:
1531                 rx_desc->wb.upper.status_error &= cpu_to_le32(~0xFF);
1532
1533                 /* return some buffers to hardware, one at a time is too slow */
1534                 if (unlikely(cleaned_count >= E1000_RX_BUFFER_WRITE)) {
1535                         adapter->alloc_rx_buf(rx_ring, cleaned_count,
1536                                               GFP_ATOMIC);
1537                         cleaned_count = 0;
1538                 }
1539
1540                 /* use prefetched values */
1541                 rx_desc = next_rxd;
1542                 buffer_info = next_buffer;
1543
1544                 staterr = le32_to_cpu(rx_desc->wb.upper.status_error);
1545         }
1546         rx_ring->next_to_clean = i;
1547
1548         cleaned_count = e1000_desc_unused(rx_ring);
1549         if (cleaned_count)
1550                 adapter->alloc_rx_buf(rx_ring, cleaned_count, GFP_ATOMIC);
1551
1552         adapter->total_rx_bytes += total_rx_bytes;
1553         adapter->total_rx_packets += total_rx_packets;
1554         return cleaned;
1555 }
1556
1557 /**
1558  * e1000_clean_rx_ring - Free Rx Buffers per Queue
1559  * @rx_ring: Rx descriptor ring
1560  **/
1561 static void e1000_clean_rx_ring(struct e1000_ring *rx_ring)
1562 {
1563         struct e1000_adapter *adapter = rx_ring->adapter;
1564         struct e1000_buffer *buffer_info;
1565         struct e1000_ps_page *ps_page;
1566         struct pci_dev *pdev = adapter->pdev;
1567         unsigned int i, j;
1568
1569         /* Free all the Rx ring sk_buffs */
1570         for (i = 0; i < rx_ring->count; i++) {
1571                 buffer_info = &rx_ring->buffer_info[i];
1572                 if (buffer_info->dma) {
1573                         if (adapter->clean_rx == e1000_clean_rx_irq)
1574                                 dma_unmap_single(&pdev->dev, buffer_info->dma,
1575                                                  adapter->rx_buffer_len,
1576                                                  DMA_FROM_DEVICE);
1577                         else if (adapter->clean_rx == e1000_clean_jumbo_rx_irq)
1578                                 dma_unmap_page(&pdev->dev, buffer_info->dma,
1579                                                PAGE_SIZE,
1580                                                DMA_FROM_DEVICE);
1581                         else if (adapter->clean_rx == e1000_clean_rx_irq_ps)
1582                                 dma_unmap_single(&pdev->dev, buffer_info->dma,
1583                                                  adapter->rx_ps_bsize0,
1584                                                  DMA_FROM_DEVICE);
1585                         buffer_info->dma = 0;
1586                 }
1587
1588                 if (buffer_info->page) {
1589                         put_page(buffer_info->page);
1590                         buffer_info->page = NULL;
1591                 }
1592
1593                 if (buffer_info->skb) {
1594                         dev_kfree_skb(buffer_info->skb);
1595                         buffer_info->skb = NULL;
1596                 }
1597
1598                 for (j = 0; j < PS_PAGE_BUFFERS; j++) {
1599                         ps_page = &buffer_info->ps_pages[j];
1600                         if (!ps_page->page)
1601                                 break;
1602                         dma_unmap_page(&pdev->dev, ps_page->dma, PAGE_SIZE,
1603                                        DMA_FROM_DEVICE);
1604                         ps_page->dma = 0;
1605                         put_page(ps_page->page);
1606                         ps_page->page = NULL;
1607                 }
1608         }
1609
1610         /* there also may be some cached data from a chained receive */
1611         if (rx_ring->rx_skb_top) {
1612                 dev_kfree_skb(rx_ring->rx_skb_top);
1613                 rx_ring->rx_skb_top = NULL;
1614         }
1615
1616         /* Zero out the descriptor ring */
1617         memset(rx_ring->desc, 0, rx_ring->size);
1618
1619         rx_ring->next_to_clean = 0;
1620         rx_ring->next_to_use = 0;
1621         adapter->flags2 &= ~FLAG2_IS_DISCARDING;
1622
1623         writel(0, rx_ring->head);
1624         if (rx_ring->adapter->flags2 & FLAG2_PCIM2PCI_ARBITER_WA)
1625                 e1000e_update_rdt_wa(rx_ring, 0);
1626         else
1627                 writel(0, rx_ring->tail);
1628 }
1629
1630 static void e1000e_downshift_workaround(struct work_struct *work)
1631 {
1632         struct e1000_adapter *adapter = container_of(work,
1633                                         struct e1000_adapter, downshift_task);
1634
1635         if (test_bit(__E1000_DOWN, &adapter->state))
1636                 return;
1637
1638         e1000e_gig_downshift_workaround_ich8lan(&adapter->hw);
1639 }
1640
1641 /**
1642  * e1000_intr_msi - Interrupt Handler
1643  * @irq: interrupt number
1644  * @data: pointer to a network interface device structure
1645  **/
1646 static irqreturn_t e1000_intr_msi(int irq, void *data)
1647 {
1648         struct net_device *netdev = data;
1649         struct e1000_adapter *adapter = netdev_priv(netdev);
1650         struct e1000_hw *hw = &adapter->hw;
1651         u32 icr = er32(ICR);
1652
1653         /*
1654          * read ICR disables interrupts using IAM
1655          */
1656
1657         if (icr & E1000_ICR_LSC) {
1658                 hw->mac.get_link_status = true;
1659                 /*
1660                  * ICH8 workaround-- Call gig speed drop workaround on cable
1661                  * disconnect (LSC) before accessing any PHY registers
1662                  */
1663                 if ((adapter->flags & FLAG_LSC_GIG_SPEED_DROP) &&
1664                     (!(er32(STATUS) & E1000_STATUS_LU)))
1665                         schedule_work(&adapter->downshift_task);
1666
1667                 /*
1668                  * 80003ES2LAN workaround-- For packet buffer work-around on
1669                  * link down event; disable receives here in the ISR and reset
1670                  * adapter in watchdog
1671                  */
1672                 if (netif_carrier_ok(netdev) &&
1673                     adapter->flags & FLAG_RX_NEEDS_RESTART) {
1674                         /* disable receives */
1675                         u32 rctl = er32(RCTL);
1676                         ew32(RCTL, rctl & ~E1000_RCTL_EN);
1677                         adapter->flags |= FLAG_RX_RESTART_NOW;
1678                 }
1679                 /* guard against interrupt when we're going down */
1680                 if (!test_bit(__E1000_DOWN, &adapter->state))
1681                         mod_timer(&adapter->watchdog_timer, jiffies + 1);
1682         }
1683
1684         if (napi_schedule_prep(&adapter->napi)) {
1685                 adapter->total_tx_bytes = 0;
1686                 adapter->total_tx_packets = 0;
1687                 adapter->total_rx_bytes = 0;
1688                 adapter->total_rx_packets = 0;
1689                 __napi_schedule(&adapter->napi);
1690         }
1691
1692         return IRQ_HANDLED;
1693 }
1694
1695 /**
1696  * e1000_intr - Interrupt Handler
1697  * @irq: interrupt number
1698  * @data: pointer to a network interface device structure
1699  **/
1700 static irqreturn_t e1000_intr(int irq, void *data)
1701 {
1702         struct net_device *netdev = data;
1703         struct e1000_adapter *adapter = netdev_priv(netdev);
1704         struct e1000_hw *hw = &adapter->hw;
1705         u32 rctl, icr = er32(ICR);
1706
1707         if (!icr || test_bit(__E1000_DOWN, &adapter->state))
1708                 return IRQ_NONE;  /* Not our interrupt */
1709
1710         /*
1711          * IMS will not auto-mask if INT_ASSERTED is not set, and if it is
1712          * not set, then the adapter didn't send an interrupt
1713          */
1714         if (!(icr & E1000_ICR_INT_ASSERTED))
1715                 return IRQ_NONE;
1716
1717         /*
1718          * Interrupt Auto-Mask...upon reading ICR,
1719          * interrupts are masked.  No need for the
1720          * IMC write
1721          */
1722
1723         if (icr & E1000_ICR_LSC) {
1724                 hw->mac.get_link_status = true;
1725                 /*
1726                  * ICH8 workaround-- Call gig speed drop workaround on cable
1727                  * disconnect (LSC) before accessing any PHY registers
1728                  */
1729                 if ((adapter->flags & FLAG_LSC_GIG_SPEED_DROP) &&
1730                     (!(er32(STATUS) & E1000_STATUS_LU)))
1731                         schedule_work(&adapter->downshift_task);
1732
1733                 /*
1734                  * 80003ES2LAN workaround--
1735                  * For packet buffer work-around on link down event;
1736                  * disable receives here in the ISR and
1737                  * reset adapter in watchdog
1738                  */
1739                 if (netif_carrier_ok(netdev) &&
1740                     (adapter->flags & FLAG_RX_NEEDS_RESTART)) {
1741                         /* disable receives */
1742                         rctl = er32(RCTL);
1743                         ew32(RCTL, rctl & ~E1000_RCTL_EN);
1744                         adapter->flags |= FLAG_RX_RESTART_NOW;
1745                 }
1746                 /* guard against interrupt when we're going down */
1747                 if (!test_bit(__E1000_DOWN, &adapter->state))
1748                         mod_timer(&adapter->watchdog_timer, jiffies + 1);
1749         }
1750
1751         if (napi_schedule_prep(&adapter->napi)) {
1752                 adapter->total_tx_bytes = 0;
1753                 adapter->total_tx_packets = 0;
1754                 adapter->total_rx_bytes = 0;
1755                 adapter->total_rx_packets = 0;
1756                 __napi_schedule(&adapter->napi);
1757         }
1758
1759         return IRQ_HANDLED;
1760 }
1761
1762 static irqreturn_t e1000_msix_other(int irq, void *data)
1763 {
1764         struct net_device *netdev = data;
1765         struct e1000_adapter *adapter = netdev_priv(netdev);
1766         struct e1000_hw *hw = &adapter->hw;
1767         u32 icr = er32(ICR);
1768
1769         if (!(icr & E1000_ICR_INT_ASSERTED)) {
1770                 if (!test_bit(__E1000_DOWN, &adapter->state))
1771                         ew32(IMS, E1000_IMS_OTHER);
1772                 return IRQ_NONE;
1773         }
1774
1775         if (icr & adapter->eiac_mask)
1776                 ew32(ICS, (icr & adapter->eiac_mask));
1777
1778         if (icr & E1000_ICR_OTHER) {
1779                 if (!(icr & E1000_ICR_LSC))
1780                         goto no_link_interrupt;
1781                 hw->mac.get_link_status = true;
1782                 /* guard against interrupt when we're going down */
1783                 if (!test_bit(__E1000_DOWN, &adapter->state))
1784                         mod_timer(&adapter->watchdog_timer, jiffies + 1);
1785         }
1786
1787 no_link_interrupt:
1788         if (!test_bit(__E1000_DOWN, &adapter->state))
1789                 ew32(IMS, E1000_IMS_LSC | E1000_IMS_OTHER);
1790
1791         return IRQ_HANDLED;
1792 }
1793
1794
1795 static irqreturn_t e1000_intr_msix_tx(int irq, void *data)
1796 {
1797         struct net_device *netdev = data;
1798         struct e1000_adapter *adapter = netdev_priv(netdev);
1799         struct e1000_hw *hw = &adapter->hw;
1800         struct e1000_ring *tx_ring = adapter->tx_ring;
1801
1802
1803         adapter->total_tx_bytes = 0;
1804         adapter->total_tx_packets = 0;
1805
1806         if (!e1000_clean_tx_irq(tx_ring))
1807                 /* Ring was not completely cleaned, so fire another interrupt */
1808                 ew32(ICS, tx_ring->ims_val);
1809
1810         return IRQ_HANDLED;
1811 }
1812
1813 static irqreturn_t e1000_intr_msix_rx(int irq, void *data)
1814 {
1815         struct net_device *netdev = data;
1816         struct e1000_adapter *adapter = netdev_priv(netdev);
1817         struct e1000_ring *rx_ring = adapter->rx_ring;
1818
1819         /* Write the ITR value calculated at the end of the
1820          * previous interrupt.
1821          */
1822         if (rx_ring->set_itr) {
1823                 writel(1000000000 / (rx_ring->itr_val * 256),
1824                        rx_ring->itr_register);
1825                 rx_ring->set_itr = 0;
1826         }
1827
1828         if (napi_schedule_prep(&adapter->napi)) {
1829                 adapter->total_rx_bytes = 0;
1830                 adapter->total_rx_packets = 0;
1831                 __napi_schedule(&adapter->napi);
1832         }
1833         return IRQ_HANDLED;
1834 }
1835
1836 /**
1837  * e1000_configure_msix - Configure MSI-X hardware
1838  *
1839  * e1000_configure_msix sets up the hardware to properly
1840  * generate MSI-X interrupts.
1841  **/
1842 static void e1000_configure_msix(struct e1000_adapter *adapter)
1843 {
1844         struct e1000_hw *hw = &adapter->hw;
1845         struct e1000_ring *rx_ring = adapter->rx_ring;
1846         struct e1000_ring *tx_ring = adapter->tx_ring;
1847         int vector = 0;
1848         u32 ctrl_ext, ivar = 0;
1849
1850         adapter->eiac_mask = 0;
1851
1852         /* Workaround issue with spurious interrupts on 82574 in MSI-X mode */
1853         if (hw->mac.type == e1000_82574) {
1854                 u32 rfctl = er32(RFCTL);
1855                 rfctl |= E1000_RFCTL_ACK_DIS;
1856                 ew32(RFCTL, rfctl);
1857         }
1858
1859 #define E1000_IVAR_INT_ALLOC_VALID      0x8
1860         /* Configure Rx vector */
1861         rx_ring->ims_val = E1000_IMS_RXQ0;
1862         adapter->eiac_mask |= rx_ring->ims_val;
1863         if (rx_ring->itr_val)
1864                 writel(1000000000 / (rx_ring->itr_val * 256),
1865                        rx_ring->itr_register);
1866         else
1867                 writel(1, rx_ring->itr_register);
1868         ivar = E1000_IVAR_INT_ALLOC_VALID | vector;
1869
1870         /* Configure Tx vector */
1871         tx_ring->ims_val = E1000_IMS_TXQ0;
1872         vector++;
1873         if (tx_ring->itr_val)
1874                 writel(1000000000 / (tx_ring->itr_val * 256),
1875                        tx_ring->itr_register);
1876         else
1877                 writel(1, tx_ring->itr_register);
1878         adapter->eiac_mask |= tx_ring->ims_val;
1879         ivar |= ((E1000_IVAR_INT_ALLOC_VALID | vector) << 8);
1880
1881         /* set vector for Other Causes, e.g. link changes */
1882         vector++;
1883         ivar |= ((E1000_IVAR_INT_ALLOC_VALID | vector) << 16);
1884         if (rx_ring->itr_val)
1885                 writel(1000000000 / (rx_ring->itr_val * 256),
1886                        hw->hw_addr + E1000_EITR_82574(vector));
1887         else
1888                 writel(1, hw->hw_addr + E1000_EITR_82574(vector));
1889
1890         /* Cause Tx interrupts on every write back */
1891         ivar |= (1 << 31);
1892
1893         ew32(IVAR, ivar);
1894
1895         /* enable MSI-X PBA support */
1896         ctrl_ext = er32(CTRL_EXT);
1897         ctrl_ext |= E1000_CTRL_EXT_PBA_CLR;
1898
1899         /* Auto-Mask Other interrupts upon ICR read */
1900 #define E1000_EIAC_MASK_82574   0x01F00000
1901         ew32(IAM, ~E1000_EIAC_MASK_82574 | E1000_IMS_OTHER);
1902         ctrl_ext |= E1000_CTRL_EXT_EIAME;
1903         ew32(CTRL_EXT, ctrl_ext);
1904         e1e_flush();
1905 }
1906
1907 void e1000e_reset_interrupt_capability(struct e1000_adapter *adapter)
1908 {
1909         if (adapter->msix_entries) {
1910                 pci_disable_msix(adapter->pdev);
1911                 kfree(adapter->msix_entries);
1912                 adapter->msix_entries = NULL;
1913         } else if (adapter->flags & FLAG_MSI_ENABLED) {
1914                 pci_disable_msi(adapter->pdev);
1915                 adapter->flags &= ~FLAG_MSI_ENABLED;
1916         }
1917 }
1918
1919 /**
1920  * e1000e_set_interrupt_capability - set MSI or MSI-X if supported
1921  *
1922  * Attempt to configure interrupts using the best available
1923  * capabilities of the hardware and kernel.
1924  **/
1925 void e1000e_set_interrupt_capability(struct e1000_adapter *adapter)
1926 {
1927         int err;
1928         int i;
1929
1930         switch (adapter->int_mode) {
1931         case E1000E_INT_MODE_MSIX:
1932                 if (adapter->flags & FLAG_HAS_MSIX) {
1933                         adapter->num_vectors = 3; /* RxQ0, TxQ0 and other */
1934                         adapter->msix_entries = kcalloc(adapter->num_vectors,
1935                                                       sizeof(struct msix_entry),
1936                                                       GFP_KERNEL);
1937                         if (adapter->msix_entries) {
1938                                 for (i = 0; i < adapter->num_vectors; i++)
1939                                         adapter->msix_entries[i].entry = i;
1940
1941                                 err = pci_enable_msix(adapter->pdev,
1942                                                       adapter->msix_entries,
1943                                                       adapter->num_vectors);
1944                                 if (err == 0)
1945                                         return;
1946                         }
1947                         /* MSI-X failed, so fall through and try MSI */
1948                         e_err("Failed to initialize MSI-X interrupts.  Falling back to MSI interrupts.\n");
1949                         e1000e_reset_interrupt_capability(adapter);
1950                 }
1951                 adapter->int_mode = E1000E_INT_MODE_MSI;
1952                 /* Fall through */
1953         case E1000E_INT_MODE_MSI:
1954                 if (!pci_enable_msi(adapter->pdev)) {
1955                         adapter->flags |= FLAG_MSI_ENABLED;
1956                 } else {
1957                         adapter->int_mode = E1000E_INT_MODE_LEGACY;
1958                         e_err("Failed to initialize MSI interrupts.  Falling back to legacy interrupts.\n");
1959                 }
1960                 /* Fall through */
1961         case E1000E_INT_MODE_LEGACY:
1962                 /* Don't do anything; this is the system default */
1963                 break;
1964         }
1965
1966         /* store the number of vectors being used */
1967         adapter->num_vectors = 1;
1968 }
1969
1970 /**
1971  * e1000_request_msix - Initialize MSI-X interrupts
1972  *
1973  * e1000_request_msix allocates MSI-X vectors and requests interrupts from the
1974  * kernel.
1975  **/
1976 static int e1000_request_msix(struct e1000_adapter *adapter)
1977 {
1978         struct net_device *netdev = adapter->netdev;
1979         int err = 0, vector = 0;
1980
1981         if (strlen(netdev->name) < (IFNAMSIZ - 5))
1982                 snprintf(adapter->rx_ring->name,
1983                          sizeof(adapter->rx_ring->name) - 1,
1984                          "%s-rx-0", netdev->name);
1985         else
1986                 memcpy(adapter->rx_ring->name, netdev->name, IFNAMSIZ);
1987         err = request_irq(adapter->msix_entries[vector].vector,
1988                           e1000_intr_msix_rx, 0, adapter->rx_ring->name,
1989                           netdev);
1990         if (err)
1991                 return err;
1992         adapter->rx_ring->itr_register = adapter->hw.hw_addr +
1993             E1000_EITR_82574(vector);
1994         adapter->rx_ring->itr_val = adapter->itr;
1995         vector++;
1996
1997         if (strlen(netdev->name) < (IFNAMSIZ - 5))
1998                 snprintf(adapter->tx_ring->name,
1999                          sizeof(adapter->tx_ring->name) - 1,
2000                          "%s-tx-0", netdev->name);
2001         else
2002                 memcpy(adapter->tx_ring->name, netdev->name, IFNAMSIZ);
2003         err = request_irq(adapter->msix_entries[vector].vector,
2004                           e1000_intr_msix_tx, 0, adapter->tx_ring->name,
2005                           netdev);
2006         if (err)
2007                 return err;
2008         adapter->tx_ring->itr_register = adapter->hw.hw_addr +
2009             E1000_EITR_82574(vector);
2010         adapter->tx_ring->itr_val = adapter->itr;
2011         vector++;
2012
2013         err = request_irq(adapter->msix_entries[vector].vector,
2014                           e1000_msix_other, 0, netdev->name, netdev);
2015         if (err)
2016                 return err;
2017
2018         e1000_configure_msix(adapter);
2019
2020         return 0;
2021 }
2022
2023 /**
2024  * e1000_request_irq - initialize interrupts
2025  *
2026  * Attempts to configure interrupts using the best available
2027  * capabilities of the hardware and kernel.
2028  **/
2029 static int e1000_request_irq(struct e1000_adapter *adapter)
2030 {
2031         struct net_device *netdev = adapter->netdev;
2032         int err;
2033
2034         if (adapter->msix_entries) {
2035                 err = e1000_request_msix(adapter);
2036                 if (!err)
2037                         return err;
2038                 /* fall back to MSI */
2039                 e1000e_reset_interrupt_capability(adapter);
2040                 adapter->int_mode = E1000E_INT_MODE_MSI;
2041                 e1000e_set_interrupt_capability(adapter);
2042         }
2043         if (adapter->flags & FLAG_MSI_ENABLED) {
2044                 err = request_irq(adapter->pdev->irq, e1000_intr_msi, 0,
2045                                   netdev->name, netdev);
2046                 if (!err)
2047                         return err;
2048
2049                 /* fall back to legacy interrupt */
2050                 e1000e_reset_interrupt_capability(adapter);
2051                 adapter->int_mode = E1000E_INT_MODE_LEGACY;
2052         }
2053
2054         err = request_irq(adapter->pdev->irq, e1000_intr, IRQF_SHARED,
2055                           netdev->name, netdev);
2056         if (err)
2057                 e_err("Unable to allocate interrupt, Error: %d\n", err);
2058
2059         return err;
2060 }
2061
2062 static void e1000_free_irq(struct e1000_adapter *adapter)
2063 {
2064         struct net_device *netdev = adapter->netdev;
2065
2066         if (adapter->msix_entries) {
2067                 int vector = 0;
2068
2069                 free_irq(adapter->msix_entries[vector].vector, netdev);
2070                 vector++;
2071
2072                 free_irq(adapter->msix_entries[vector].vector, netdev);
2073                 vector++;
2074
2075                 /* Other Causes interrupt vector */
2076                 free_irq(adapter->msix_entries[vector].vector, netdev);
2077                 return;
2078         }
2079
2080         free_irq(adapter->pdev->irq, netdev);
2081 }
2082
2083 /**
2084  * e1000_irq_disable - Mask off interrupt generation on the NIC
2085  **/
2086 static void e1000_irq_disable(struct e1000_adapter *adapter)
2087 {
2088         struct e1000_hw *hw = &adapter->hw;
2089
2090         ew32(IMC, ~0);
2091         if (adapter->msix_entries)
2092                 ew32(EIAC_82574, 0);
2093         e1e_flush();
2094
2095         if (adapter->msix_entries) {
2096                 int i;
2097                 for (i = 0; i < adapter->num_vectors; i++)
2098                         synchronize_irq(adapter->msix_entries[i].vector);
2099         } else {
2100                 synchronize_irq(adapter->pdev->irq);
2101         }
2102 }
2103
2104 /**
2105  * e1000_irq_enable - Enable default interrupt generation settings
2106  **/
2107 static void e1000_irq_enable(struct e1000_adapter *adapter)
2108 {
2109         struct e1000_hw *hw = &adapter->hw;
2110
2111         if (adapter->msix_entries) {
2112                 ew32(EIAC_82574, adapter->eiac_mask & E1000_EIAC_MASK_82574);
2113                 ew32(IMS, adapter->eiac_mask | E1000_IMS_OTHER | E1000_IMS_LSC);
2114         } else {
2115                 ew32(IMS, IMS_ENABLE_MASK);
2116         }
2117         e1e_flush();
2118 }
2119
2120 /**
2121  * e1000e_get_hw_control - get control of the h/w from f/w
2122  * @adapter: address of board private structure
2123  *
2124  * e1000e_get_hw_control sets {CTRL_EXT|SWSM}:DRV_LOAD bit.
2125  * For ASF and Pass Through versions of f/w this means that
2126  * the driver is loaded. For AMT version (only with 82573)
2127  * of the f/w this means that the network i/f is open.
2128  **/
2129 void e1000e_get_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 know the driver has taken over */
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  * e1000e_release_hw_control - release control of the h/w to f/w
2147  * @adapter: address of board private structure
2148  *
2149  * e1000e_release_hw_control resets {CTRL_EXT|SWSM}:DRV_LOAD bit.
2150  * For ASF and Pass Through versions of f/w this means that the
2151  * driver is no longer loaded. For AMT version (only with 82573) i
2152  * of the f/w this means that the network i/f is closed.
2153  *
2154  **/
2155 void e1000e_release_hw_control(struct e1000_adapter *adapter)
2156 {
2157         struct e1000_hw *hw = &adapter->hw;
2158         u32 ctrl_ext;
2159         u32 swsm;
2160
2161         /* Let firmware taken over control of h/w */
2162         if (adapter->flags & FLAG_HAS_SWSM_ON_LOAD) {
2163                 swsm = er32(SWSM);
2164                 ew32(SWSM, swsm & ~E1000_SWSM_DRV_LOAD);
2165         } else if (adapter->flags & FLAG_HAS_CTRLEXT_ON_LOAD) {
2166                 ctrl_ext = er32(CTRL_EXT);
2167                 ew32(CTRL_EXT, ctrl_ext & ~E1000_CTRL_EXT_DRV_LOAD);
2168         }
2169 }
2170
2171 /**
2172  * @e1000_alloc_ring - allocate memory for a ring structure
2173  **/
2174 static int e1000_alloc_ring_dma(struct e1000_adapter *adapter,
2175                                 struct e1000_ring *ring)
2176 {
2177         struct pci_dev *pdev = adapter->pdev;
2178
2179         ring->desc = dma_alloc_coherent(&pdev->dev, ring->size, &ring->dma,
2180                                         GFP_KERNEL);
2181         if (!ring->desc)
2182                 return -ENOMEM;
2183
2184         return 0;
2185 }
2186
2187 /**
2188  * e1000e_setup_tx_resources - allocate Tx resources (Descriptors)
2189  * @tx_ring: Tx descriptor ring
2190  *
2191  * Return 0 on success, negative on failure
2192  **/
2193 int e1000e_setup_tx_resources(struct e1000_ring *tx_ring)
2194 {
2195         struct e1000_adapter *adapter = tx_ring->adapter;
2196         int err = -ENOMEM, size;
2197
2198         size = sizeof(struct e1000_buffer) * tx_ring->count;
2199         tx_ring->buffer_info = vzalloc(size);
2200         if (!tx_ring->buffer_info)
2201                 goto err;
2202
2203         /* round up to nearest 4K */
2204         tx_ring->size = tx_ring->count * sizeof(struct e1000_tx_desc);
2205         tx_ring->size = ALIGN(tx_ring->size, 4096);
2206
2207         err = e1000_alloc_ring_dma(adapter, tx_ring);
2208         if (err)
2209                 goto err;
2210
2211         tx_ring->next_to_use = 0;
2212         tx_ring->next_to_clean = 0;
2213
2214         return 0;
2215 err:
2216         vfree(tx_ring->buffer_info);
2217         e_err("Unable to allocate memory for the transmit descriptor ring\n");
2218         return err;
2219 }
2220
2221 /**
2222  * e1000e_setup_rx_resources - allocate Rx resources (Descriptors)
2223  * @rx_ring: Rx descriptor ring
2224  *
2225  * Returns 0 on success, negative on failure
2226  **/
2227 int e1000e_setup_rx_resources(struct e1000_ring *rx_ring)
2228 {
2229         struct e1000_adapter *adapter = rx_ring->adapter;
2230         struct e1000_buffer *buffer_info;
2231         int i, size, desc_len, err = -ENOMEM;
2232
2233         size = sizeof(struct e1000_buffer) * rx_ring->count;
2234         rx_ring->buffer_info = vzalloc(size);
2235         if (!rx_ring->buffer_info)
2236                 goto err;
2237
2238         for (i = 0; i < rx_ring->count; i++) {
2239                 buffer_info = &rx_ring->buffer_info[i];
2240                 buffer_info->ps_pages = kcalloc(PS_PAGE_BUFFERS,
2241                                                 sizeof(struct e1000_ps_page),
2242                                                 GFP_KERNEL);
2243                 if (!buffer_info->ps_pages)
2244                         goto err_pages;
2245         }
2246
2247         desc_len = sizeof(union e1000_rx_desc_packet_split);
2248
2249         /* Round up to nearest 4K */
2250         rx_ring->size = rx_ring->count * desc_len;
2251         rx_ring->size = ALIGN(rx_ring->size, 4096);
2252
2253         err = e1000_alloc_ring_dma(adapter, rx_ring);
2254         if (err)
2255                 goto err_pages;
2256
2257         rx_ring->next_to_clean = 0;
2258         rx_ring->next_to_use = 0;
2259         rx_ring->rx_skb_top = NULL;
2260
2261         return 0;
2262
2263 err_pages:
2264         for (i = 0; i < rx_ring->count; i++) {
2265                 buffer_info = &rx_ring->buffer_info[i];
2266                 kfree(buffer_info->ps_pages);
2267         }
2268 err:
2269         vfree(rx_ring->buffer_info);
2270         e_err("Unable to allocate memory for the receive descriptor ring\n");
2271         return err;
2272 }
2273
2274 /**
2275  * e1000_clean_tx_ring - Free Tx Buffers
2276  * @tx_ring: Tx descriptor ring
2277  **/
2278 static void e1000_clean_tx_ring(struct e1000_ring *tx_ring)
2279 {
2280         struct e1000_adapter *adapter = tx_ring->adapter;
2281         struct e1000_buffer *buffer_info;
2282         unsigned long size;
2283         unsigned int i;
2284
2285         for (i = 0; i < tx_ring->count; i++) {
2286                 buffer_info = &tx_ring->buffer_info[i];
2287                 e1000_put_txbuf(tx_ring, buffer_info);
2288         }
2289
2290         netdev_reset_queue(adapter->netdev);
2291         size = sizeof(struct e1000_buffer) * tx_ring->count;
2292         memset(tx_ring->buffer_info, 0, size);
2293
2294         memset(tx_ring->desc, 0, tx_ring->size);
2295
2296         tx_ring->next_to_use = 0;
2297         tx_ring->next_to_clean = 0;
2298
2299         writel(0, tx_ring->head);
2300         if (tx_ring->adapter->flags2 & FLAG2_PCIM2PCI_ARBITER_WA)
2301                 e1000e_update_tdt_wa(tx_ring, 0);
2302         else
2303                 writel(0, tx_ring->tail);
2304 }
2305
2306 /**
2307  * e1000e_free_tx_resources - Free Tx Resources per Queue
2308  * @tx_ring: Tx descriptor ring
2309  *
2310  * Free all transmit software resources
2311  **/
2312 void e1000e_free_tx_resources(struct e1000_ring *tx_ring)
2313 {
2314         struct e1000_adapter *adapter = tx_ring->adapter;
2315         struct pci_dev *pdev = adapter->pdev;
2316
2317         e1000_clean_tx_ring(tx_ring);
2318
2319         vfree(tx_ring->buffer_info);
2320         tx_ring->buffer_info = NULL;
2321
2322         dma_free_coherent(&pdev->dev, tx_ring->size, tx_ring->desc,
2323                           tx_ring->dma);
2324         tx_ring->desc = NULL;
2325 }
2326
2327 /**
2328  * e1000e_free_rx_resources - Free Rx Resources
2329  * @rx_ring: Rx descriptor ring
2330  *
2331  * Free all receive software resources
2332  **/
2333 void e1000e_free_rx_resources(struct e1000_ring *rx_ring)
2334 {
2335         struct e1000_adapter *adapter = rx_ring->adapter;
2336         struct pci_dev *pdev = adapter->pdev;
2337         int i;
2338
2339         e1000_clean_rx_ring(rx_ring);
2340
2341         for (i = 0; i < rx_ring->count; i++)
2342                 kfree(rx_ring->buffer_info[i].ps_pages);
2343
2344         vfree(rx_ring->buffer_info);
2345         rx_ring->buffer_info = NULL;
2346
2347         dma_free_coherent(&pdev->dev, rx_ring->size, rx_ring->desc,
2348                           rx_ring->dma);
2349         rx_ring->desc = NULL;
2350 }
2351
2352 /**
2353  * e1000_update_itr - update the dynamic ITR value based on statistics
2354  * @adapter: pointer to adapter
2355  * @itr_setting: current adapter->itr
2356  * @packets: the number of packets during this measurement interval
2357  * @bytes: the number of bytes during this measurement interval
2358  *
2359  *      Stores a new ITR value based on packets and byte
2360  *      counts during the last interrupt.  The advantage of per interrupt
2361  *      computation is faster updates and more accurate ITR for the current
2362  *      traffic pattern.  Constants in this function were computed
2363  *      based on theoretical maximum wire speed and thresholds were set based
2364  *      on testing data as well as attempting to minimize response time
2365  *      while increasing bulk throughput.  This functionality is controlled
2366  *      by the InterruptThrottleRate module parameter.
2367  **/
2368 static unsigned int e1000_update_itr(struct e1000_adapter *adapter,
2369                                      u16 itr_setting, int packets,
2370                                      int bytes)
2371 {
2372         unsigned int retval = itr_setting;
2373
2374         if (packets == 0)
2375                 return itr_setting;
2376
2377         switch (itr_setting) {
2378         case lowest_latency:
2379                 /* handle TSO and jumbo frames */
2380                 if (bytes/packets > 8000)
2381                         retval = bulk_latency;
2382                 else if ((packets < 5) && (bytes > 512))
2383                         retval = low_latency;
2384                 break;
2385         case low_latency:  /* 50 usec aka 20000 ints/s */
2386                 if (bytes > 10000) {
2387                         /* this if handles the TSO accounting */
2388                         if (bytes/packets > 8000)
2389                                 retval = bulk_latency;
2390                         else if ((packets < 10) || ((bytes/packets) > 1200))
2391                                 retval = bulk_latency;
2392                         else if ((packets > 35))
2393                                 retval = lowest_latency;
2394                 } else if (bytes/packets > 2000) {
2395                         retval = bulk_latency;
2396                 } else if (packets <= 2 && bytes < 512) {
2397                         retval = lowest_latency;
2398                 }
2399                 break;
2400         case bulk_latency: /* 250 usec aka 4000 ints/s */
2401                 if (bytes > 25000) {
2402                         if (packets > 35)
2403                                 retval = low_latency;
2404                 } else if (bytes < 6000) {
2405                         retval = low_latency;
2406                 }
2407                 break;
2408         }
2409
2410         return retval;
2411 }
2412
2413 static void e1000_set_itr(struct e1000_adapter *adapter)
2414 {
2415         struct e1000_hw *hw = &adapter->hw;
2416         u16 current_itr;
2417         u32 new_itr = adapter->itr;
2418
2419         /* for non-gigabit speeds, just fix the interrupt rate at 4000 */
2420         if (adapter->link_speed != SPEED_1000) {
2421                 current_itr = 0;
2422                 new_itr = 4000;
2423                 goto set_itr_now;
2424         }
2425
2426         if (adapter->flags2 & FLAG2_DISABLE_AIM) {
2427                 new_itr = 0;
2428                 goto set_itr_now;
2429         }
2430
2431         adapter->tx_itr = e1000_update_itr(adapter,
2432                                     adapter->tx_itr,
2433                                     adapter->total_tx_packets,
2434                                     adapter->total_tx_bytes);
2435         /* conservative mode (itr 3) eliminates the lowest_latency setting */
2436         if (adapter->itr_setting == 3 && adapter->tx_itr == lowest_latency)
2437                 adapter->tx_itr = low_latency;
2438
2439         adapter->rx_itr = e1000_update_itr(adapter,
2440                                     adapter->rx_itr,
2441                                     adapter->total_rx_packets,
2442                                     adapter->total_rx_bytes);
2443         /* conservative mode (itr 3) eliminates the lowest_latency setting */
2444         if (adapter->itr_setting == 3 && adapter->rx_itr == lowest_latency)
2445                 adapter->rx_itr = low_latency;
2446
2447         current_itr = max(adapter->rx_itr, adapter->tx_itr);
2448
2449         switch (current_itr) {
2450         /* counts and packets in update_itr are dependent on these numbers */
2451         case lowest_latency:
2452                 new_itr = 70000;
2453                 break;
2454         case low_latency:
2455                 new_itr = 20000; /* aka hwitr = ~200 */
2456                 break;
2457         case bulk_latency:
2458                 new_itr = 4000;
2459                 break;
2460         default:
2461                 break;
2462         }
2463
2464 set_itr_now:
2465         if (new_itr != adapter->itr) {
2466                 /*
2467                  * this attempts to bias the interrupt rate towards Bulk
2468                  * by adding intermediate steps when interrupt rate is
2469                  * increasing
2470                  */
2471                 new_itr = new_itr > adapter->itr ?
2472                              min(adapter->itr + (new_itr >> 2), new_itr) :
2473                              new_itr;
2474                 adapter->itr = new_itr;
2475                 adapter->rx_ring->itr_val = new_itr;
2476                 if (adapter->msix_entries)
2477                         adapter->rx_ring->set_itr = 1;
2478                 else
2479                         if (new_itr)
2480                                 ew32(ITR, 1000000000 / (new_itr * 256));
2481                         else
2482                                 ew32(ITR, 0);
2483         }
2484 }
2485
2486 /**
2487  * e1000_alloc_queues - Allocate memory for all rings
2488  * @adapter: board private structure to initialize
2489  **/
2490 static int __devinit e1000_alloc_queues(struct e1000_adapter *adapter)
2491 {
2492         int size = sizeof(struct e1000_ring);
2493
2494         adapter->tx_ring = kzalloc(size, GFP_KERNEL);
2495         if (!adapter->tx_ring)
2496                 goto err;
2497         adapter->tx_ring->count = adapter->tx_ring_count;
2498         adapter->tx_ring->adapter = adapter;
2499
2500         adapter->rx_ring = kzalloc(size, GFP_KERNEL);
2501         if (!adapter->rx_ring)
2502                 goto err;
2503         adapter->rx_ring->count = adapter->rx_ring_count;
2504         adapter->rx_ring->adapter = adapter;
2505
2506         return 0;
2507 err:
2508         e_err("Unable to allocate memory for queues\n");
2509         kfree(adapter->rx_ring);
2510         kfree(adapter->tx_ring);
2511         return -ENOMEM;
2512 }
2513
2514 /**
2515  * e1000e_poll - NAPI Rx polling callback
2516  * @napi: struct associated with this polling callback
2517  * @weight: number of packets driver is allowed to process this poll
2518  **/
2519 static int e1000e_poll(struct napi_struct *napi, int weight)
2520 {
2521         struct e1000_adapter *adapter = container_of(napi, struct e1000_adapter,
2522                                                      napi);
2523         struct e1000_hw *hw = &adapter->hw;
2524         struct net_device *poll_dev = adapter->netdev;
2525         int tx_cleaned = 1, work_done = 0;
2526
2527         adapter = netdev_priv(poll_dev);
2528
2529         if (!adapter->msix_entries ||
2530             (adapter->rx_ring->ims_val & adapter->tx_ring->ims_val))
2531                 tx_cleaned = e1000_clean_tx_irq(adapter->tx_ring);
2532
2533         adapter->clean_rx(adapter->rx_ring, &work_done, weight);
2534
2535         if (!tx_cleaned)
2536                 work_done = weight;
2537
2538         /* If weight not fully consumed, exit the polling mode */
2539         if (work_done < weight) {
2540                 if (adapter->itr_setting & 3)
2541                         e1000_set_itr(adapter);
2542                 napi_complete(napi);
2543                 if (!test_bit(__E1000_DOWN, &adapter->state)) {
2544                         if (adapter->msix_entries)
2545                                 ew32(IMS, adapter->rx_ring->ims_val);
2546                         else
2547                                 e1000_irq_enable(adapter);
2548                 }
2549         }
2550
2551         return work_done;
2552 }
2553
2554 static int e1000_vlan_rx_add_vid(struct net_device *netdev, u16 vid)
2555 {
2556         struct e1000_adapter *adapter = netdev_priv(netdev);
2557         struct e1000_hw *hw = &adapter->hw;
2558         u32 vfta, index;
2559
2560         /* don't update vlan cookie if already programmed */
2561         if ((adapter->hw.mng_cookie.status &
2562              E1000_MNG_DHCP_COOKIE_STATUS_VLAN) &&
2563             (vid == adapter->mng_vlan_id))
2564                 return 0;
2565
2566         /* add VID to filter table */
2567         if (adapter->flags & FLAG_HAS_HW_VLAN_FILTER) {
2568                 index = (vid >> 5) & 0x7F;
2569                 vfta = E1000_READ_REG_ARRAY(hw, E1000_VFTA, index);
2570                 vfta |= (1 << (vid & 0x1F));
2571                 hw->mac.ops.write_vfta(hw, index, vfta);
2572         }
2573
2574         set_bit(vid, adapter->active_vlans);
2575
2576         return 0;
2577 }
2578
2579 static int e1000_vlan_rx_kill_vid(struct net_device *netdev, u16 vid)
2580 {
2581         struct e1000_adapter *adapter = netdev_priv(netdev);
2582         struct e1000_hw *hw = &adapter->hw;
2583         u32 vfta, index;
2584
2585         if ((adapter->hw.mng_cookie.status &
2586              E1000_MNG_DHCP_COOKIE_STATUS_VLAN) &&
2587             (vid == adapter->mng_vlan_id)) {
2588                 /* release control to f/w */
2589                 e1000e_release_hw_control(adapter);
2590                 return 0;
2591         }
2592
2593         /* remove VID from filter table */
2594         if (adapter->flags & FLAG_HAS_HW_VLAN_FILTER) {
2595                 index = (vid >> 5) & 0x7F;
2596                 vfta = E1000_READ_REG_ARRAY(hw, E1000_VFTA, index);
2597                 vfta &= ~(1 << (vid & 0x1F));
2598                 hw->mac.ops.write_vfta(hw, index, vfta);
2599         }
2600
2601         clear_bit(vid, adapter->active_vlans);
2602
2603         return 0;
2604 }
2605
2606 /**
2607  * e1000e_vlan_filter_disable - helper to disable hw VLAN filtering
2608  * @adapter: board private structure to initialize
2609  **/
2610 static void e1000e_vlan_filter_disable(struct e1000_adapter *adapter)
2611 {
2612         struct net_device *netdev = adapter->netdev;
2613         struct e1000_hw *hw = &adapter->hw;
2614         u32 rctl;
2615
2616         if (adapter->flags & FLAG_HAS_HW_VLAN_FILTER) {
2617                 /* disable VLAN receive filtering */
2618                 rctl = er32(RCTL);
2619                 rctl &= ~(E1000_RCTL_VFE | E1000_RCTL_CFIEN);
2620                 ew32(RCTL, rctl);
2621
2622                 if (adapter->mng_vlan_id != (u16)E1000_MNG_VLAN_NONE) {
2623                         e1000_vlan_rx_kill_vid(netdev, adapter->mng_vlan_id);
2624                         adapter->mng_vlan_id = E1000_MNG_VLAN_NONE;
2625                 }
2626         }
2627 }
2628
2629 /**
2630  * e1000e_vlan_filter_enable - helper to enable HW VLAN filtering
2631  * @adapter: board private structure to initialize
2632  **/
2633 static void e1000e_vlan_filter_enable(struct e1000_adapter *adapter)
2634 {
2635         struct e1000_hw *hw = &adapter->hw;
2636         u32 rctl;
2637
2638         if (adapter->flags & FLAG_HAS_HW_VLAN_FILTER) {
2639                 /* enable VLAN receive filtering */
2640                 rctl = er32(RCTL);
2641                 rctl |= E1000_RCTL_VFE;
2642                 rctl &= ~E1000_RCTL_CFIEN;
2643                 ew32(RCTL, rctl);
2644         }
2645 }
2646
2647 /**
2648  * e1000e_vlan_strip_enable - helper to disable HW VLAN stripping
2649  * @adapter: board private structure to initialize
2650  **/
2651 static void e1000e_vlan_strip_disable(struct e1000_adapter *adapter)
2652 {
2653         struct e1000_hw *hw = &adapter->hw;
2654         u32 ctrl;
2655
2656         /* disable VLAN tag insert/strip */
2657         ctrl = er32(CTRL);
2658         ctrl &= ~E1000_CTRL_VME;
2659         ew32(CTRL, ctrl);
2660 }
2661
2662 /**
2663  * e1000e_vlan_strip_enable - helper to enable HW VLAN stripping
2664  * @adapter: board private structure to initialize
2665  **/
2666 static void e1000e_vlan_strip_enable(struct e1000_adapter *adapter)
2667 {
2668         struct e1000_hw *hw = &adapter->hw;
2669         u32 ctrl;
2670
2671         /* enable VLAN tag insert/strip */
2672         ctrl = er32(CTRL);
2673         ctrl |= E1000_CTRL_VME;
2674         ew32(CTRL, ctrl);
2675 }
2676
2677 static void e1000_update_mng_vlan(struct e1000_adapter *adapter)
2678 {
2679         struct net_device *netdev = adapter->netdev;
2680         u16 vid = adapter->hw.mng_cookie.vlan_id;
2681         u16 old_vid = adapter->mng_vlan_id;
2682
2683         if (adapter->hw.mng_cookie.status &
2684             E1000_MNG_DHCP_COOKIE_STATUS_VLAN) {
2685                 e1000_vlan_rx_add_vid(netdev, vid);
2686                 adapter->mng_vlan_id = vid;
2687         }
2688
2689         if ((old_vid != (u16)E1000_MNG_VLAN_NONE) && (vid != old_vid))
2690                 e1000_vlan_rx_kill_vid(netdev, old_vid);
2691 }
2692
2693 static void e1000_restore_vlan(struct e1000_adapter *adapter)
2694 {
2695         u16 vid;
2696
2697         e1000_vlan_rx_add_vid(adapter->netdev, 0);
2698
2699         for_each_set_bit(vid, adapter->active_vlans, VLAN_N_VID)
2700                 e1000_vlan_rx_add_vid(adapter->netdev, vid);
2701 }
2702
2703 static void e1000_init_manageability_pt(struct e1000_adapter *adapter)
2704 {
2705         struct e1000_hw *hw = &adapter->hw;
2706         u32 manc, manc2h, mdef, i, j;
2707
2708         if (!(adapter->flags & FLAG_MNG_PT_ENABLED))
2709                 return;
2710
2711         manc = er32(MANC);
2712
2713         /*
2714          * enable receiving management packets to the host. this will probably
2715          * generate destination unreachable messages from the host OS, but
2716          * the packets will be handled on SMBUS
2717          */
2718         manc |= E1000_MANC_EN_MNG2HOST;
2719         manc2h = er32(MANC2H);
2720
2721         switch (hw->mac.type) {
2722         default:
2723                 manc2h |= (E1000_MANC2H_PORT_623 | E1000_MANC2H_PORT_664);
2724                 break;
2725         case e1000_82574:
2726         case e1000_82583:
2727                 /*
2728                  * Check if IPMI pass-through decision filter already exists;
2729                  * if so, enable it.
2730                  */
2731                 for (i = 0, j = 0; i < 8; i++) {
2732                         mdef = er32(MDEF(i));
2733
2734                         /* Ignore filters with anything other than IPMI ports */
2735                         if (mdef & ~(E1000_MDEF_PORT_623 | E1000_MDEF_PORT_664))
2736                                 continue;
2737
2738                         /* Enable this decision filter in MANC2H */
2739                         if (mdef)
2740                                 manc2h |= (1 << i);
2741
2742                         j |= mdef;
2743                 }
2744
2745                 if (j == (E1000_MDEF_PORT_623 | E1000_MDEF_PORT_664))
2746                         break;
2747
2748                 /* Create new decision filter in an empty filter */
2749                 for (i = 0, j = 0; i < 8; i++)
2750                         if (er32(MDEF(i)) == 0) {
2751                                 ew32(MDEF(i), (E1000_MDEF_PORT_623 |
2752                                                E1000_MDEF_PORT_664));
2753                                 manc2h |= (1 << 1);
2754                                 j++;
2755                                 break;
2756                         }
2757
2758                 if (!j)
2759                         e_warn("Unable to create IPMI pass-through filter\n");
2760                 break;
2761         }
2762
2763         ew32(MANC2H, manc2h);
2764         ew32(MANC, manc);
2765 }
2766
2767 /**
2768  * e1000_configure_tx - Configure Transmit Unit after Reset
2769  * @adapter: board private structure
2770  *
2771  * Configure the Tx unit of the MAC after a reset.
2772  **/
2773 static void e1000_configure_tx(struct e1000_adapter *adapter)
2774 {
2775         struct e1000_hw *hw = &adapter->hw;
2776         struct e1000_ring *tx_ring = adapter->tx_ring;
2777         u64 tdba;
2778         u32 tdlen, tarc;
2779
2780         /* Setup the HW Tx Head and Tail descriptor pointers */
2781         tdba = tx_ring->dma;
2782         tdlen = tx_ring->count * sizeof(struct e1000_tx_desc);
2783         ew32(TDBAL(0), (tdba & DMA_BIT_MASK(32)));
2784         ew32(TDBAH(0), (tdba >> 32));
2785         ew32(TDLEN(0), tdlen);
2786         ew32(TDH(0), 0);
2787         ew32(TDT(0), 0);
2788         tx_ring->head = adapter->hw.hw_addr + E1000_TDH(0);
2789         tx_ring->tail = adapter->hw.hw_addr + E1000_TDT(0);
2790
2791         /* Set the Tx Interrupt Delay register */
2792         ew32(TIDV, adapter->tx_int_delay);
2793         /* Tx irq moderation */
2794         ew32(TADV, adapter->tx_abs_int_delay);
2795
2796         if (adapter->flags2 & FLAG2_DMA_BURST) {
2797                 u32 txdctl = er32(TXDCTL(0));
2798                 txdctl &= ~(E1000_TXDCTL_PTHRESH | E1000_TXDCTL_HTHRESH |
2799                             E1000_TXDCTL_WTHRESH);
2800                 /*
2801                  * set up some performance related parameters to encourage the
2802                  * hardware to use the bus more efficiently in bursts, depends
2803                  * on the tx_int_delay to be enabled,
2804                  * wthresh = 5 ==> burst write a cacheline (64 bytes) at a time
2805                  * hthresh = 1 ==> prefetch when one or more available
2806                  * pthresh = 0x1f ==> prefetch if internal cache 31 or less
2807                  * BEWARE: this seems to work but should be considered first if
2808                  * there are Tx hangs or other Tx related bugs
2809                  */
2810                 txdctl |= E1000_TXDCTL_DMA_BURST_ENABLE;
2811                 ew32(TXDCTL(0), txdctl);
2812         }
2813         /* erratum work around: set txdctl the same for both queues */
2814         ew32(TXDCTL(1), er32(TXDCTL(0)));
2815
2816         if (adapter->flags & FLAG_TARC_SPEED_MODE_BIT) {
2817                 tarc = er32(TARC(0));
2818                 /*
2819                  * set the speed mode bit, we'll clear it if we're not at
2820                  * gigabit link later
2821                  */
2822 #define SPEED_MODE_BIT (1 << 21)
2823                 tarc |= SPEED_MODE_BIT;
2824                 ew32(TARC(0), tarc);
2825         }
2826
2827         /* errata: program both queues to unweighted RR */
2828         if (adapter->flags & FLAG_TARC_SET_BIT_ZERO) {
2829                 tarc = er32(TARC(0));
2830                 tarc |= 1;
2831                 ew32(TARC(0), tarc);
2832                 tarc = er32(TARC(1));
2833                 tarc |= 1;
2834                 ew32(TARC(1), tarc);
2835         }
2836
2837         /* Setup Transmit Descriptor Settings for eop descriptor */
2838         adapter->txd_cmd = E1000_TXD_CMD_EOP | E1000_TXD_CMD_IFCS;
2839
2840         /* only set IDE if we are delaying interrupts using the timers */
2841         if (adapter->tx_int_delay)
2842                 adapter->txd_cmd |= E1000_TXD_CMD_IDE;
2843
2844         /* enable Report Status bit */
2845         adapter->txd_cmd |= E1000_TXD_CMD_RS;
2846
2847         hw->mac.ops.config_collision_dist(hw);
2848 }
2849
2850 /**
2851  * e1000_setup_rctl - configure the receive control registers
2852  * @adapter: Board private structure
2853  **/
2854 #define PAGE_USE_COUNT(S) (((S) >> PAGE_SHIFT) + \
2855                            (((S) & (PAGE_SIZE - 1)) ? 1 : 0))
2856 static void e1000_setup_rctl(struct e1000_adapter *adapter)
2857 {
2858         struct e1000_hw *hw = &adapter->hw;
2859         u32 rctl, rfctl;
2860         u32 pages = 0;
2861
2862         /* Workaround Si errata on 82579 - configure jumbo frame flow */
2863         if (hw->mac.type == e1000_pch2lan) {
2864                 s32 ret_val;
2865
2866                 if (adapter->netdev->mtu > ETH_DATA_LEN)
2867                         ret_val = e1000_lv_jumbo_workaround_ich8lan(hw, true);
2868                 else
2869                         ret_val = e1000_lv_jumbo_workaround_ich8lan(hw, false);
2870
2871                 if (ret_val)
2872                         e_dbg("failed to enable jumbo frame workaround mode\n");
2873         }
2874
2875         /* Program MC offset vector base */
2876         rctl = er32(RCTL);
2877         rctl &= ~(3 << E1000_RCTL_MO_SHIFT);
2878         rctl |= E1000_RCTL_EN | E1000_RCTL_BAM |
2879                 E1000_RCTL_LBM_NO | E1000_RCTL_RDMTS_HALF |
2880                 (adapter->hw.mac.mc_filter_type << E1000_RCTL_MO_SHIFT);
2881
2882         /* Do not Store bad packets */
2883         rctl &= ~E1000_RCTL_SBP;
2884
2885         /* Enable Long Packet receive */
2886         if (adapter->netdev->mtu <= ETH_DATA_LEN)
2887                 rctl &= ~E1000_RCTL_LPE;
2888         else
2889                 rctl |= E1000_RCTL_LPE;
2890
2891         /* Some systems expect that the CRC is included in SMBUS traffic. The
2892          * hardware strips the CRC before sending to both SMBUS (BMC) and to
2893          * host memory when this is enabled
2894          */
2895         if (adapter->flags2 & FLAG2_CRC_STRIPPING)
2896                 rctl |= E1000_RCTL_SECRC;
2897
2898         /* Workaround Si errata on 82577 PHY - configure IPG for jumbos */
2899         if ((hw->phy.type == e1000_phy_82577) && (rctl & E1000_RCTL_LPE)) {
2900                 u16 phy_data;
2901
2902                 e1e_rphy(hw, PHY_REG(770, 26), &phy_data);
2903                 phy_data &= 0xfff8;
2904                 phy_data |= (1 << 2);
2905                 e1e_wphy(hw, PHY_REG(770, 26), phy_data);
2906
2907                 e1e_rphy(hw, 22, &phy_data);
2908                 phy_data &= 0x0fff;
2909                 phy_data |= (1 << 14);
2910                 e1e_wphy(hw, 0x10, 0x2823);
2911                 e1e_wphy(hw, 0x11, 0x0003);
2912                 e1e_wphy(hw, 22, phy_data);
2913         }
2914
2915         /* Setup buffer sizes */
2916         rctl &= ~E1000_RCTL_SZ_4096;
2917         rctl |= E1000_RCTL_BSEX;
2918         switch (adapter->rx_buffer_len) {
2919         case 2048:
2920         default:
2921                 rctl |= E1000_RCTL_SZ_2048;
2922                 rctl &= ~E1000_RCTL_BSEX;
2923                 break;
2924         case 4096:
2925                 rctl |= E1000_RCTL_SZ_4096;
2926                 break;
2927         case 8192:
2928                 rctl |= E1000_RCTL_SZ_8192;
2929                 break;
2930         case 16384:
2931                 rctl |= E1000_RCTL_SZ_16384;
2932                 break;
2933         }
2934
2935         /* Enable Extended Status in all Receive Descriptors */
2936         rfctl = er32(RFCTL);
2937         rfctl |= E1000_RFCTL_EXTEN;
2938
2939         /*
2940          * 82571 and greater support packet-split where the protocol
2941          * header is placed in skb->data and the packet data is
2942          * placed in pages hanging off of skb_shinfo(skb)->nr_frags.
2943          * In the case of a non-split, skb->data is linearly filled,
2944          * followed by the page buffers.  Therefore, skb->data is
2945          * sized to hold the largest protocol header.
2946          *
2947          * allocations using alloc_page take too long for regular MTU
2948          * so only enable packet split for jumbo frames
2949          *
2950          * Using pages when the page size is greater than 16k wastes
2951          * a lot of memory, since we allocate 3 pages at all times
2952          * per packet.
2953          */
2954         pages = PAGE_USE_COUNT(adapter->netdev->mtu);
2955         if ((pages <= 3) && (PAGE_SIZE <= 16384) && (rctl & E1000_RCTL_LPE))
2956                 adapter->rx_ps_pages = pages;
2957         else
2958                 adapter->rx_ps_pages = 0;
2959
2960         if (adapter->rx_ps_pages) {
2961                 u32 psrctl = 0;
2962
2963                 /*
2964                  * disable packet split support for IPv6 extension headers,
2965                  * because some malformed IPv6 headers can hang the Rx
2966                  */
2967                 rfctl |= (E1000_RFCTL_IPV6_EX_DIS |
2968                           E1000_RFCTL_NEW_IPV6_EXT_DIS);
2969
2970                 /* Enable Packet split descriptors */
2971                 rctl |= E1000_RCTL_DTYP_PS;
2972
2973                 psrctl |= adapter->rx_ps_bsize0 >>
2974                         E1000_PSRCTL_BSIZE0_SHIFT;
2975
2976                 switch (adapter->rx_ps_pages) {
2977                 case 3:
2978                         psrctl |= PAGE_SIZE <<
2979                                 E1000_PSRCTL_BSIZE3_SHIFT;
2980                 case 2:
2981                         psrctl |= PAGE_SIZE <<
2982                                 E1000_PSRCTL_BSIZE2_SHIFT;
2983                 case 1:
2984                         psrctl |= PAGE_SIZE >>
2985                                 E1000_PSRCTL_BSIZE1_SHIFT;
2986                         break;
2987                 }
2988
2989                 ew32(PSRCTL, psrctl);
2990         }
2991
2992         /* This is useful for sniffing bad packets. */
2993         if (adapter->netdev->features & NETIF_F_RXALL) {
2994                 /* UPE and MPE will be handled by normal PROMISC logic
2995                  * in e1000e_set_rx_mode */
2996                 rctl |= (E1000_RCTL_SBP | /* Receive bad packets */
2997                          E1000_RCTL_BAM | /* RX All Bcast Pkts */
2998                          E1000_RCTL_PMCF); /* RX All MAC Ctrl Pkts */
2999
3000                 rctl &= ~(E1000_RCTL_VFE | /* Disable VLAN filter */
3001                           E1000_RCTL_DPF | /* Allow filtered pause */
3002                           E1000_RCTL_CFIEN); /* Dis VLAN CFIEN Filter */
3003                 /* Do not mess with E1000_CTRL_VME, it affects transmit as well,
3004                  * and that breaks VLANs.
3005                  */
3006         }
3007
3008         ew32(RFCTL, rfctl);
3009         ew32(RCTL, rctl);
3010         /* just started the receive unit, no need to restart */
3011         adapter->flags &= ~FLAG_RX_RESTART_NOW;
3012 }
3013
3014 /**
3015  * e1000_configure_rx - Configure Receive Unit after Reset
3016  * @adapter: board private structure
3017  *
3018  * Configure the Rx unit of the MAC after a reset.
3019  **/
3020 static void e1000_configure_rx(struct e1000_adapter *adapter)
3021 {
3022         struct e1000_hw *hw = &adapter->hw;
3023         struct e1000_ring *rx_ring = adapter->rx_ring;
3024         u64 rdba;
3025         u32 rdlen, rctl, rxcsum, ctrl_ext;
3026
3027         if (adapter->rx_ps_pages) {
3028                 /* this is a 32 byte descriptor */
3029                 rdlen = rx_ring->count *
3030                     sizeof(union e1000_rx_desc_packet_split);
3031                 adapter->clean_rx = e1000_clean_rx_irq_ps;
3032                 adapter->alloc_rx_buf = e1000_alloc_rx_buffers_ps;
3033         } else if (adapter->netdev->mtu > ETH_FRAME_LEN + ETH_FCS_LEN) {
3034                 rdlen = rx_ring->count * sizeof(union e1000_rx_desc_extended);
3035                 adapter->clean_rx = e1000_clean_jumbo_rx_irq;
3036                 adapter->alloc_rx_buf = e1000_alloc_jumbo_rx_buffers;
3037         } else {
3038                 rdlen = rx_ring->count * sizeof(union e1000_rx_desc_extended);
3039                 adapter->clean_rx = e1000_clean_rx_irq;
3040                 adapter->alloc_rx_buf = e1000_alloc_rx_buffers;
3041         }
3042
3043         /* disable receives while setting up the descriptors */
3044         rctl = er32(RCTL);
3045         if (!(adapter->flags2 & FLAG2_NO_DISABLE_RX))
3046                 ew32(RCTL, rctl & ~E1000_RCTL_EN);
3047         e1e_flush();
3048         usleep_range(10000, 20000);
3049
3050         if (adapter->flags2 & FLAG2_DMA_BURST) {
3051                 /*
3052                  * set the writeback threshold (only takes effect if the RDTR
3053                  * is set). set GRAN=1 and write back up to 0x4 worth, and
3054                  * enable prefetching of 0x20 Rx descriptors
3055                  * granularity = 01
3056                  * wthresh = 04,
3057                  * hthresh = 04,
3058                  * pthresh = 0x20
3059                  */
3060                 ew32(RXDCTL(0), E1000_RXDCTL_DMA_BURST_ENABLE);
3061                 ew32(RXDCTL(1), E1000_RXDCTL_DMA_BURST_ENABLE);
3062
3063                 /*
3064                  * override the delay timers for enabling bursting, only if
3065                  * the value was not set by the user via module options
3066                  */
3067                 if (adapter->rx_int_delay == DEFAULT_RDTR)
3068                         adapter->rx_int_delay = BURST_RDTR;
3069                 if (adapter->rx_abs_int_delay == DEFAULT_RADV)
3070                         adapter->rx_abs_int_delay = BURST_RADV;
3071         }
3072
3073         /* set the Receive Delay Timer Register */
3074         ew32(RDTR, adapter->rx_int_delay);
3075
3076         /* irq moderation */
3077         ew32(RADV, adapter->rx_abs_int_delay);
3078         if ((adapter->itr_setting != 0) && (adapter->itr != 0))
3079                 ew32(ITR, 1000000000 / (adapter->itr * 256));
3080
3081         ctrl_ext = er32(CTRL_EXT);
3082         /* Auto-Mask interrupts upon ICR access */
3083         ctrl_ext |= E1000_CTRL_EXT_IAME;
3084         ew32(IAM, 0xffffffff);
3085         ew32(CTRL_EXT, ctrl_ext);
3086         e1e_flush();
3087
3088         /*
3089          * Setup the HW Rx Head and Tail Descriptor Pointers and
3090          * the Base and Length of the Rx Descriptor Ring
3091          */
3092         rdba = rx_ring->dma;
3093         ew32(RDBAL(0), (rdba & DMA_BIT_MASK(32)));
3094         ew32(RDBAH(0), (rdba >> 32));
3095         ew32(RDLEN(0), rdlen);
3096         ew32(RDH(0), 0);
3097         ew32(RDT(0), 0);
3098         rx_ring->head = adapter->hw.hw_addr + E1000_RDH(0);
3099         rx_ring->tail = adapter->hw.hw_addr + E1000_RDT(0);
3100
3101         /* Enable Receive Checksum Offload for TCP and UDP */
3102         rxcsum = er32(RXCSUM);
3103         if (adapter->netdev->features & NETIF_F_RXCSUM) {
3104                 rxcsum |= E1000_RXCSUM_TUOFL;
3105
3106                 /*
3107                  * IPv4 payload checksum for UDP fragments must be
3108                  * used in conjunction with packet-split.
3109                  */
3110                 if (adapter->rx_ps_pages)
3111                         rxcsum |= E1000_RXCSUM_IPPCSE;
3112         } else {
3113                 rxcsum &= ~E1000_RXCSUM_TUOFL;
3114                 /* no need to clear IPPCSE as it defaults to 0 */
3115         }
3116         ew32(RXCSUM, rxcsum);
3117
3118         if (adapter->hw.mac.type == e1000_pch2lan) {
3119                 /*
3120                  * With jumbo frames, excessive C-state transition
3121                  * latencies result in dropped transactions.
3122                  */
3123                 if (adapter->netdev->mtu > ETH_DATA_LEN) {
3124                         u32 rxdctl = er32(RXDCTL(0));
3125                         ew32(RXDCTL(0), rxdctl | 0x3);
3126                         pm_qos_update_request(&adapter->netdev->pm_qos_req, 55);
3127                 } else {
3128                         pm_qos_update_request(&adapter->netdev->pm_qos_req,
3129                                               PM_QOS_DEFAULT_VALUE);
3130                 }
3131         }
3132
3133         /* Enable Receives */
3134         ew32(RCTL, rctl);
3135 }
3136
3137 /**
3138  * e1000e_write_mc_addr_list - write multicast addresses to MTA
3139  * @netdev: network interface device structure
3140  *
3141  * Writes multicast address list to the MTA hash table.
3142  * Returns: -ENOMEM on failure
3143  *                0 on no addresses written
3144  *                X on writing X addresses to MTA
3145  */
3146 static int e1000e_write_mc_addr_list(struct net_device *netdev)
3147 {
3148         struct e1000_adapter *adapter = netdev_priv(netdev);
3149         struct e1000_hw *hw = &adapter->hw;
3150         struct netdev_hw_addr *ha;
3151         u8 *mta_list;
3152         int i;
3153
3154         if (netdev_mc_empty(netdev)) {
3155                 /* nothing to program, so clear mc list */
3156                 hw->mac.ops.update_mc_addr_list(hw, NULL, 0);
3157                 return 0;
3158         }
3159
3160         mta_list = kzalloc(netdev_mc_count(netdev) * ETH_ALEN, GFP_ATOMIC);
3161         if (!mta_list)
3162                 return -ENOMEM;
3163
3164         /* update_mc_addr_list expects a packed array of only addresses. */
3165         i = 0;
3166         netdev_for_each_mc_addr(ha, netdev)
3167                 memcpy(mta_list + (i++ * ETH_ALEN), ha->addr, ETH_ALEN);
3168
3169         hw->mac.ops.update_mc_addr_list(hw, mta_list, i);
3170         kfree(mta_list);
3171
3172         return netdev_mc_count(netdev);
3173 }
3174
3175 /**
3176  * e1000e_write_uc_addr_list - write unicast addresses to RAR table
3177  * @netdev: network interface device structure
3178  *
3179  * Writes unicast address list to the RAR table.
3180  * Returns: -ENOMEM on failure/insufficient address space
3181  *                0 on no addresses written
3182  *                X on writing X addresses to the RAR table
3183  **/
3184 static int e1000e_write_uc_addr_list(struct net_device *netdev)
3185 {
3186         struct e1000_adapter *adapter = netdev_priv(netdev);
3187         struct e1000_hw *hw = &adapter->hw;
3188         unsigned int rar_entries = hw->mac.rar_entry_count;
3189         int count = 0;
3190
3191         /* save a rar entry for our hardware address */
3192         rar_entries--;
3193
3194         /* save a rar entry for the LAA workaround */
3195         if (adapter->flags & FLAG_RESET_OVERWRITES_LAA)
3196                 rar_entries--;
3197
3198         /* return ENOMEM indicating insufficient memory for addresses */
3199         if (netdev_uc_count(netdev) > rar_entries)
3200                 return -ENOMEM;
3201
3202         if (!netdev_uc_empty(netdev) && rar_entries) {
3203                 struct netdev_hw_addr *ha;
3204
3205                 /*
3206                  * write the addresses in reverse order to avoid write
3207                  * combining
3208                  */
3209                 netdev_for_each_uc_addr(ha, netdev) {
3210                         if (!rar_entries)
3211                                 break;
3212                         e1000e_rar_set(hw, ha->addr, rar_entries--);
3213                         count++;
3214                 }
3215         }
3216
3217         /* zero out the remaining RAR entries not used above */
3218         for (; rar_entries > 0; rar_entries--) {
3219                 ew32(RAH(rar_entries), 0);
3220                 ew32(RAL(rar_entries), 0);
3221         }
3222         e1e_flush();
3223
3224         return count;
3225 }
3226
3227 /**
3228  * e1000e_set_rx_mode - secondary unicast, Multicast and Promiscuous mode set
3229  * @netdev: network interface device structure
3230  *
3231  * The ndo_set_rx_mode entry point is called whenever the unicast or multicast
3232  * address list or the network interface flags are updated.  This routine is
3233  * responsible for configuring the hardware for proper unicast, multicast,
3234  * promiscuous mode, and all-multi behavior.
3235  **/
3236 static void e1000e_set_rx_mode(struct net_device *netdev)
3237 {
3238         struct e1000_adapter *adapter = netdev_priv(netdev);
3239         struct e1000_hw *hw = &adapter->hw;
3240         u32 rctl;
3241
3242         /* Check for Promiscuous and All Multicast modes */
3243         rctl = er32(RCTL);
3244
3245         /* clear the affected bits */
3246         rctl &= ~(E1000_RCTL_UPE | E1000_RCTL_MPE);
3247
3248         if (netdev->flags & IFF_PROMISC) {
3249                 rctl |= (E1000_RCTL_UPE | E1000_RCTL_MPE);
3250                 /* Do not hardware filter VLANs in promisc mode */
3251                 e1000e_vlan_filter_disable(adapter);
3252         } else {
3253                 int count;
3254
3255                 if (netdev->flags & IFF_ALLMULTI) {
3256                         rctl |= E1000_RCTL_MPE;
3257                 } else {
3258                         /*
3259                          * Write addresses to the MTA, if the attempt fails
3260                          * then we should just turn on promiscuous mode so
3261                          * that we can at least receive multicast traffic
3262                          */
3263                         count = e1000e_write_mc_addr_list(netdev);
3264                         if (count < 0)
3265                                 rctl |= E1000_RCTL_MPE;
3266                 }
3267                 e1000e_vlan_filter_enable(adapter);
3268                 /*
3269                  * Write addresses to available RAR registers, if there is not
3270                  * sufficient space to store all the addresses then enable
3271                  * unicast promiscuous mode
3272                  */
3273                 count = e1000e_write_uc_addr_list(netdev);
3274                 if (count < 0)
3275                         rctl |= E1000_RCTL_UPE;
3276         }
3277
3278         ew32(RCTL, rctl);
3279
3280         if (netdev->features & NETIF_F_HW_VLAN_RX)
3281                 e1000e_vlan_strip_enable(adapter);
3282         else
3283                 e1000e_vlan_strip_disable(adapter);
3284 }
3285
3286 static void e1000e_setup_rss_hash(struct e1000_adapter *adapter)
3287 {
3288         struct e1000_hw *hw = &adapter->hw;
3289         u32 mrqc, rxcsum;
3290         int i;
3291         static const u32 rsskey[10] = {
3292                 0xda565a6d, 0xc20e5b25, 0x3d256741, 0xb08fa343, 0xcb2bcad0,
3293                 0xb4307bae, 0xa32dcb77, 0x0cf23080, 0x3bb7426a, 0xfa01acbe
3294         };
3295
3296         /* Fill out hash function seed */
3297         for (i = 0; i < 10; i++)
3298                 ew32(RSSRK(i), rsskey[i]);
3299
3300         /* Direct all traffic to queue 0 */
3301         for (i = 0; i < 32; i++)
3302                 ew32(RETA(i), 0);
3303
3304         /*
3305          * Disable raw packet checksumming so that RSS hash is placed in
3306          * descriptor on writeback.
3307          */
3308         rxcsum = er32(RXCSUM);
3309         rxcsum |= E1000_RXCSUM_PCSD;
3310
3311         ew32(RXCSUM, rxcsum);
3312
3313         mrqc = (E1000_MRQC_RSS_FIELD_IPV4 |
3314                 E1000_MRQC_RSS_FIELD_IPV4_TCP |
3315                 E1000_MRQC_RSS_FIELD_IPV6 |
3316                 E1000_MRQC_RSS_FIELD_IPV6_TCP |
3317                 E1000_MRQC_RSS_FIELD_IPV6_TCP_EX);
3318
3319         ew32(MRQC, mrqc);
3320 }
3321
3322 /**
3323  * e1000_configure - configure the hardware for Rx and Tx
3324  * @adapter: private board structure
3325  **/
3326 static void e1000_configure(struct e1000_adapter *adapter)
3327 {
3328         struct e1000_ring *rx_ring = adapter->rx_ring;
3329
3330         e1000e_set_rx_mode(adapter->netdev);
3331
3332         e1000_restore_vlan(adapter);
3333         e1000_init_manageability_pt(adapter);
3334
3335         e1000_configure_tx(adapter);
3336
3337         if (adapter->netdev->features & NETIF_F_RXHASH)
3338                 e1000e_setup_rss_hash(adapter);
3339         e1000_setup_rctl(adapter);
3340         e1000_configure_rx(adapter);
3341         adapter->alloc_rx_buf(rx_ring, e1000_desc_unused(rx_ring), GFP_KERNEL);
3342 }
3343
3344 /**
3345  * e1000e_power_up_phy - restore link in case the phy was powered down
3346  * @adapter: address of board private structure
3347  *
3348  * The phy may be powered down to save power and turn off link when the
3349  * driver is unloaded and wake on lan is not enabled (among others)
3350  * *** this routine MUST be followed by a call to e1000e_reset ***
3351  **/
3352 void e1000e_power_up_phy(struct e1000_adapter *adapter)
3353 {
3354         if (adapter->hw.phy.ops.power_up)
3355                 adapter->hw.phy.ops.power_up(&adapter->hw);
3356
3357         adapter->hw.mac.ops.setup_link(&adapter->hw);
3358 }
3359
3360 /**
3361  * e1000_power_down_phy - Power down the PHY
3362  *
3363  * Power down the PHY so no link is implied when interface is down.
3364  * The PHY cannot be powered down if management or WoL is active.
3365  */
3366 static void e1000_power_down_phy(struct e1000_adapter *adapter)
3367 {
3368         /* WoL is enabled */
3369         if (adapter->wol)
3370                 return;
3371
3372         if (adapter->hw.phy.ops.power_down)
3373                 adapter->hw.phy.ops.power_down(&adapter->hw);
3374 }
3375
3376 /**
3377  * e1000e_reset - bring the hardware into a known good state
3378  *
3379  * This function boots the hardware and enables some settings that
3380  * require a configuration cycle of the hardware - those cannot be
3381  * set/changed during runtime. After reset the device needs to be
3382  * properly configured for Rx, Tx etc.
3383  */
3384 void e1000e_reset(struct e1000_adapter *adapter)
3385 {
3386         struct e1000_mac_info *mac = &adapter->hw.mac;
3387         struct e1000_fc_info *fc = &adapter->hw.fc;
3388         struct e1000_hw *hw = &adapter->hw;
3389         u32 tx_space, min_tx_space, min_rx_space;
3390         u32 pba = adapter->pba;
3391         u16 hwm;
3392
3393         /* reset Packet Buffer Allocation to default */
3394         ew32(PBA, pba);
3395
3396         if (adapter->max_frame_size > ETH_FRAME_LEN + ETH_FCS_LEN) {
3397                 /*
3398                  * To maintain wire speed transmits, the Tx FIFO should be
3399                  * large enough to accommodate two full transmit packets,
3400                  * rounded up to the next 1KB and expressed in KB.  Likewise,
3401                  * the Rx FIFO should be large enough to accommodate at least
3402                  * one full receive packet and is similarly rounded up and
3403                  * expressed in KB.
3404                  */
3405                 pba = er32(PBA);
3406                 /* upper 16 bits has Tx packet buffer allocation size in KB */
3407                 tx_space = pba >> 16;
3408                 /* lower 16 bits has Rx packet buffer allocation size in KB */
3409                 pba &= 0xffff;
3410                 /*
3411                  * the Tx fifo also stores 16 bytes of information about the Tx
3412                  * but don't include ethernet FCS because hardware appends it
3413                  */
3414                 min_tx_space = (adapter->max_frame_size +
3415                                 sizeof(struct e1000_tx_desc) -
3416                                 ETH_FCS_LEN) * 2;
3417                 min_tx_space = ALIGN(min_tx_space, 1024);
3418                 min_tx_space >>= 10;
3419                 /* software strips receive CRC, so leave room for it */
3420                 min_rx_space = adapter->max_frame_size;
3421                 min_rx_space = ALIGN(min_rx_space, 1024);
3422                 min_rx_space >>= 10;
3423
3424                 /*
3425                  * If current Tx allocation is less than the min Tx FIFO size,
3426                  * and the min Tx FIFO size is less than the current Rx FIFO
3427                  * allocation, take space away from current Rx allocation
3428                  */
3429                 if ((tx_space < min_tx_space) &&
3430                     ((min_tx_space - tx_space) < pba)) {
3431                         pba -= min_tx_space - tx_space;
3432
3433                         /*
3434                          * if short on Rx space, Rx wins and must trump Tx
3435                          * adjustment or use Early Receive if available
3436                          */
3437                         if (pba < min_rx_space)
3438                                 pba = min_rx_space;
3439                 }
3440
3441                 ew32(PBA, pba);
3442         }
3443
3444         /*
3445          * flow control settings
3446          *
3447          * The high water mark must be low enough to fit one full frame
3448          * (or the size used for early receive) above it in the Rx FIFO.
3449          * Set it to the lower of:
3450          * - 90% of the Rx FIFO size, and
3451          * - the full Rx FIFO size minus one full frame
3452          */
3453         if (adapter->flags & FLAG_DISABLE_FC_PAUSE_TIME)
3454                 fc->pause_time = 0xFFFF;
3455         else
3456                 fc->pause_time = E1000_FC_PAUSE_TIME;
3457         fc->send_xon = true;
3458         fc->current_mode = fc->requested_mode;
3459
3460         switch (hw->mac.type) {
3461         case e1000_ich9lan:
3462         case e1000_ich10lan:
3463                 if (adapter->netdev->mtu > ETH_DATA_LEN) {
3464                         pba = 14;
3465                         ew32(PBA, pba);
3466                         fc->high_water = 0x2800;
3467                         fc->low_water = fc->high_water - 8;
3468                         break;
3469                 }
3470                 /* fall-through */
3471         default:
3472                 hwm = min(((pba << 10) * 9 / 10),
3473                           ((pba << 10) - adapter->max_frame_size));
3474
3475                 fc->high_water = hwm & E1000_FCRTH_RTH; /* 8-byte granularity */
3476                 fc->low_water = fc->high_water - 8;
3477                 break;
3478         case e1000_pchlan:
3479                 /*
3480                  * Workaround PCH LOM adapter hangs with certain network
3481                  * loads.  If hangs persist, try disabling Tx flow control.
3482                  */
3483                 if (adapter->netdev->mtu > ETH_DATA_LEN) {
3484                         fc->high_water = 0x3500;
3485                         fc->low_water  = 0x1500;
3486                 } else {
3487                         fc->high_water = 0x5000;
3488                         fc->low_water  = 0x3000;
3489                 }
3490                 fc->refresh_time = 0x1000;
3491                 break;
3492         case e1000_pch2lan:
3493                 fc->high_water = 0x05C20;
3494                 fc->low_water = 0x05048;
3495                 fc->pause_time = 0x0650;
3496                 fc->refresh_time = 0x0400;
3497                 if (adapter->netdev->mtu > ETH_DATA_LEN) {
3498                         pba = 14;
3499                         ew32(PBA, pba);
3500                 }
3501                 break;
3502         }
3503
3504         /*
3505          * Disable Adaptive Interrupt Moderation if 2 full packets cannot
3506          * fit in receive buffer.
3507          */
3508         if (adapter->itr_setting & 0x3) {
3509                 if ((adapter->max_frame_size * 2) > (pba << 10)) {
3510                         if (!(adapter->flags2 & FLAG2_DISABLE_AIM)) {
3511                                 dev_info(&adapter->pdev->dev,
3512                                         "Interrupt Throttle Rate turned off\n");
3513                                 adapter->flags2 |= FLAG2_DISABLE_AIM;
3514                                 ew32(ITR, 0);
3515                         }
3516                 } else if (adapter->flags2 & FLAG2_DISABLE_AIM) {
3517                         dev_info(&adapter->pdev->dev,
3518                                  "Interrupt Throttle Rate turned on\n");
3519                         adapter->flags2 &= ~FLAG2_DISABLE_AIM;
3520                         adapter->itr = 20000;
3521                         ew32(ITR, 1000000000 / (adapter->itr * 256));
3522                 }
3523         }
3524
3525         /* Allow time for pending master requests to run */
3526         mac->ops.reset_hw(hw);
3527
3528         /*
3529          * For parts with AMT enabled, let the firmware know
3530          * that the network interface is in control
3531          */
3532         if (adapter->flags & FLAG_HAS_AMT)
3533                 e1000e_get_hw_control(adapter);
3534
3535         ew32(WUC, 0);
3536
3537         if (mac->ops.init_hw(hw))
3538                 e_err("Hardware Error\n");
3539
3540         e1000_update_mng_vlan(adapter);
3541
3542         /* Enable h/w to recognize an 802.1Q VLAN Ethernet packet */
3543         ew32(VET, ETH_P_8021Q);
3544
3545         e1000e_reset_adaptive(hw);
3546
3547         if (!netif_running(adapter->netdev) &&
3548             !test_bit(__E1000_TESTING, &adapter->state)) {
3549                 e1000_power_down_phy(adapter);
3550                 return;
3551         }
3552
3553         e1000_get_phy_info(hw);
3554
3555         if ((adapter->flags & FLAG_HAS_SMART_POWER_DOWN) &&
3556             !(adapter->flags & FLAG_SMART_POWER_DOWN)) {
3557                 u16 phy_data = 0;
3558                 /*
3559                  * speed up time to link by disabling smart power down, ignore
3560                  * the return value of this function because there is nothing
3561                  * different we would do if it failed
3562                  */
3563                 e1e_rphy(hw, IGP02E1000_PHY_POWER_MGMT, &phy_data);
3564                 phy_data &= ~IGP02E1000_PM_SPD;
3565                 e1e_wphy(hw, IGP02E1000_PHY_POWER_MGMT, phy_data);
3566         }
3567 }
3568
3569 int e1000e_up(struct e1000_adapter *adapter)
3570 {
3571         struct e1000_hw *hw = &adapter->hw;
3572
3573         /* hardware has been reset, we need to reload some things */
3574         e1000_configure(adapter);
3575
3576         clear_bit(__E1000_DOWN, &adapter->state);
3577
3578         if (adapter->msix_entries)
3579                 e1000_configure_msix(adapter);
3580         e1000_irq_enable(adapter);
3581
3582         netif_start_queue(adapter->netdev);
3583
3584         /* fire a link change interrupt to start the watchdog */
3585         if (adapter->msix_entries)
3586                 ew32(ICS, E1000_ICS_LSC | E1000_ICR_OTHER);
3587         else
3588                 ew32(ICS, E1000_ICS_LSC);
3589
3590         return 0;
3591 }
3592
3593 static void e1000e_flush_descriptors(struct e1000_adapter *adapter)
3594 {
3595         struct e1000_hw *hw = &adapter->hw;
3596
3597         if (!(adapter->flags2 & FLAG2_DMA_BURST))
3598                 return;
3599
3600         /* flush pending descriptor writebacks to memory */
3601         ew32(TIDV, adapter->tx_int_delay | E1000_TIDV_FPD);
3602         ew32(RDTR, adapter->rx_int_delay | E1000_RDTR_FPD);
3603
3604         /* execute the writes immediately */
3605         e1e_flush();
3606
3607         /*
3608          * due to rare timing issues, write to TIDV/RDTR again to ensure the
3609          * write is successful
3610          */
3611         ew32(TIDV, adapter->tx_int_delay | E1000_TIDV_FPD);
3612         ew32(RDTR, adapter->rx_int_delay | E1000_RDTR_FPD);
3613
3614         /* execute the writes immediately */
3615         e1e_flush();
3616 }
3617
3618 static void e1000e_update_stats(struct e1000_adapter *adapter);
3619
3620 void e1000e_down(struct e1000_adapter *adapter)
3621 {
3622         struct net_device *netdev = adapter->netdev;
3623         struct e1000_hw *hw = &adapter->hw;
3624         u32 tctl, rctl;
3625
3626         /*
3627          * signal that we're down so the interrupt handler does not
3628          * reschedule our watchdog timer
3629          */
3630         set_bit(__E1000_DOWN, &adapter->state);
3631
3632         /* disable receives in the hardware */
3633         rctl = er32(RCTL);
3634         if (!(adapter->flags2 & FLAG2_NO_DISABLE_RX))
3635                 ew32(RCTL, rctl & ~E1000_RCTL_EN);
3636         /* flush and sleep below */
3637
3638         netif_stop_queue(netdev);
3639
3640         /* disable transmits in the hardware */
3641         tctl = er32(TCTL);
3642         tctl &= ~E1000_TCTL_EN;
3643         ew32(TCTL, tctl);
3644
3645         /* flush both disables and wait for them to finish */
3646         e1e_flush();
3647         usleep_range(10000, 20000);
3648
3649         e1000_irq_disable(adapter);
3650
3651         del_timer_sync(&adapter->watchdog_timer);
3652         del_timer_sync(&adapter->phy_info_timer);
3653
3654         netif_carrier_off(netdev);
3655
3656         spin_lock(&adapter->stats64_lock);
3657         e1000e_update_stats(adapter);
3658         spin_unlock(&adapter->stats64_lock);
3659
3660         e1000e_flush_descriptors(adapter);
3661         e1000_clean_tx_ring(adapter->tx_ring);
3662         e1000_clean_rx_ring(adapter->rx_ring);
3663
3664         adapter->link_speed = 0;
3665         adapter->link_duplex = 0;
3666
3667         if (!pci_channel_offline(adapter->pdev))
3668                 e1000e_reset(adapter);
3669
3670         /*
3671          * TODO: for power management, we could drop the link and
3672          * pci_disable_device here.
3673          */
3674 }
3675
3676 void e1000e_reinit_locked(struct e1000_adapter *adapter)
3677 {
3678         might_sleep();
3679         while (test_and_set_bit(__E1000_RESETTING, &adapter->state))
3680                 usleep_range(1000, 2000);
3681         e1000e_down(adapter);
3682         e1000e_up(adapter);
3683         clear_bit(__E1000_RESETTING, &adapter->state);
3684 }
3685
3686 /**
3687  * e1000_sw_init - Initialize general software structures (struct e1000_adapter)
3688  * @adapter: board private structure to initialize
3689  *
3690  * e1000_sw_init initializes the Adapter private data structure.
3691  * Fields are initialized based on PCI device information and
3692  * OS network device settings (MTU size).
3693  **/
3694 static int __devinit e1000_sw_init(struct e1000_adapter *adapter)
3695 {
3696         struct net_device *netdev = adapter->netdev;
3697
3698         adapter->rx_buffer_len = ETH_FRAME_LEN + VLAN_HLEN + ETH_FCS_LEN;
3699         adapter->rx_ps_bsize0 = 128;
3700         adapter->max_frame_size = netdev->mtu + ETH_HLEN + ETH_FCS_LEN;
3701         adapter->min_frame_size = ETH_ZLEN + ETH_FCS_LEN;
3702         adapter->tx_ring_count = E1000_DEFAULT_TXD;
3703         adapter->rx_ring_count = E1000_DEFAULT_RXD;
3704
3705         spin_lock_init(&adapter->stats64_lock);
3706
3707         e1000e_set_interrupt_capability(adapter);
3708
3709         if (e1000_alloc_queues(adapter))
3710                 return -ENOMEM;
3711
3712         /* Explicitly disable IRQ since the NIC can be in any state. */
3713         e1000_irq_disable(adapter);
3714
3715         set_bit(__E1000_DOWN, &adapter->state);
3716         return 0;
3717 }
3718
3719 /**
3720  * e1000_intr_msi_test - Interrupt Handler
3721  * @irq: interrupt number
3722  * @data: pointer to a network interface device structure
3723  **/
3724 static irqreturn_t e1000_intr_msi_test(int irq, void *data)
3725 {
3726         struct net_device *netdev = data;
3727         struct e1000_adapter *adapter = netdev_priv(netdev);
3728         struct e1000_hw *hw = &adapter->hw;
3729         u32 icr = er32(ICR);
3730
3731         e_dbg("icr is %08X\n", icr);
3732         if (icr & E1000_ICR_RXSEQ) {
3733                 adapter->flags &= ~FLAG_MSI_TEST_FAILED;
3734                 wmb();
3735         }
3736
3737         return IRQ_HANDLED;
3738 }
3739
3740 /**
3741  * e1000_test_msi_interrupt - Returns 0 for successful test
3742  * @adapter: board private struct
3743  *
3744  * code flow taken from tg3.c
3745  **/
3746 static int e1000_test_msi_interrupt(struct e1000_adapter *adapter)
3747 {
3748         struct net_device *netdev = adapter->netdev;
3749         struct e1000_hw *hw = &adapter->hw;
3750         int err;
3751
3752         /* poll_enable hasn't been called yet, so don't need disable */
3753         /* clear any pending events */
3754         er32(ICR);
3755
3756         /* free the real vector and request a test handler */
3757         e1000_free_irq(adapter);
3758         e1000e_reset_interrupt_capability(adapter);
3759
3760         /* Assume that the test fails, if it succeeds then the test
3761          * MSI irq handler will unset this flag */
3762         adapter->flags |= FLAG_MSI_TEST_FAILED;
3763
3764         err = pci_enable_msi(adapter->pdev);
3765         if (err)
3766                 goto msi_test_failed;
3767
3768         err = request_irq(adapter->pdev->irq, e1000_intr_msi_test, 0,
3769                           netdev->name, netdev);
3770         if (err) {
3771                 pci_disable_msi(adapter->pdev);
3772                 goto msi_test_failed;
3773         }
3774
3775         wmb();
3776
3777         e1000_irq_enable(adapter);
3778
3779         /* fire an unusual interrupt on the test handler */
3780         ew32(ICS, E1000_ICS_RXSEQ);
3781         e1e_flush();
3782         msleep(50);
3783
3784         e1000_irq_disable(adapter);
3785
3786         rmb();
3787
3788         if (adapter->flags & FLAG_MSI_TEST_FAILED) {
3789                 adapter->int_mode = E1000E_INT_MODE_LEGACY;
3790                 e_info("MSI interrupt test failed, using legacy interrupt.\n");
3791         } else {
3792                 e_dbg("MSI interrupt test succeeded!\n");
3793         }
3794
3795         free_irq(adapter->pdev->irq, netdev);
3796         pci_disable_msi(adapter->pdev);
3797
3798 msi_test_failed:
3799         e1000e_set_interrupt_capability(adapter);
3800         return e1000_request_irq(adapter);
3801 }
3802
3803 /**
3804  * e1000_test_msi - Returns 0 if MSI test succeeds or INTx mode is restored
3805  * @adapter: board private struct
3806  *
3807  * code flow taken from tg3.c, called with e1000 interrupts disabled.
3808  **/
3809 static int e1000_test_msi(struct e1000_adapter *adapter)
3810 {
3811         int err;
3812         u16 pci_cmd;
3813
3814         if (!(adapter->flags & FLAG_MSI_ENABLED))
3815                 return 0;
3816
3817         /* disable SERR in case the MSI write causes a master abort */
3818         pci_read_config_word(adapter->pdev, PCI_COMMAND, &pci_cmd);
3819         if (pci_cmd & PCI_COMMAND_SERR)
3820                 pci_write_config_word(adapter->pdev, PCI_COMMAND,
3821                                       pci_cmd & ~PCI_COMMAND_SERR);
3822
3823         err = e1000_test_msi_interrupt(adapter);
3824
3825         /* re-enable SERR */
3826         if (pci_cmd & PCI_COMMAND_SERR) {
3827                 pci_read_config_word(adapter->pdev, PCI_COMMAND, &pci_cmd);
3828                 pci_cmd |= PCI_COMMAND_SERR;
3829                 pci_write_config_word(adapter->pdev, PCI_COMMAND, pci_cmd);
3830         }
3831
3832         return err;
3833 }
3834
3835 /**
3836  * e1000_open - Called when a network interface is made active
3837  * @netdev: network interface device structure
3838  *
3839  * Returns 0 on success, negative value on failure
3840  *
3841  * The open entry point is called when a network interface is made
3842  * active by the system (IFF_UP).  At this point all resources needed
3843  * for transmit and receive operations are allocated, the interrupt
3844  * handler is registered with the OS, the watchdog timer is started,
3845  * and the stack is notified that the interface is ready.
3846  **/
3847 static int e1000_open(struct net_device *netdev)
3848 {
3849         struct e1000_adapter *adapter = netdev_priv(netdev);
3850         struct e1000_hw *hw = &adapter->hw;
3851         struct pci_dev *pdev = adapter->pdev;
3852         int err;
3853
3854         /* disallow open during test */
3855         if (test_bit(__E1000_TESTING, &adapter->state))
3856                 return -EBUSY;
3857
3858         pm_runtime_get_sync(&pdev->dev);
3859
3860         netif_carrier_off(netdev);
3861
3862         /* allocate transmit descriptors */
3863         err = e1000e_setup_tx_resources(adapter->tx_ring);
3864         if (err)
3865                 goto err_setup_tx;
3866
3867         /* allocate receive descriptors */
3868         err = e1000e_setup_rx_resources(adapter->rx_ring);
3869         if (err)
3870                 goto err_setup_rx;
3871
3872         /*
3873          * If AMT is enabled, let the firmware know that the network
3874          * interface is now open and reset the part to a known state.
3875          */
3876         if (adapter->flags & FLAG_HAS_AMT) {
3877                 e1000e_get_hw_control(adapter);
3878                 e1000e_reset(adapter);
3879         }
3880
3881         e1000e_power_up_phy(adapter);
3882
3883         adapter->mng_vlan_id = E1000_MNG_VLAN_NONE;
3884         if ((adapter->hw.mng_cookie.status &
3885              E1000_MNG_DHCP_COOKIE_STATUS_VLAN))
3886                 e1000_update_mng_vlan(adapter);
3887
3888         /* DMA latency requirement to workaround jumbo issue */
3889         if (adapter->hw.mac.type == e1000_pch2lan)
3890                 pm_qos_add_request(&adapter->netdev->pm_qos_req,
3891                                    PM_QOS_CPU_DMA_LATENCY,
3892                                    PM_QOS_DEFAULT_VALUE);
3893
3894         /*
3895          * before we allocate an interrupt, we must be ready to handle it.
3896          * Setting DEBUG_SHIRQ in the kernel makes it fire an interrupt
3897          * as soon as we call pci_request_irq, so we have to setup our
3898          * clean_rx handler before we do so.
3899          */
3900         e1000_configure(adapter);
3901
3902         err = e1000_request_irq(adapter);
3903         if (err)
3904                 goto err_req_irq;
3905
3906         /*
3907          * Work around PCIe errata with MSI interrupts causing some chipsets to
3908          * ignore e1000e MSI messages, which means we need to test our MSI
3909          * interrupt now
3910          */
3911         if (adapter->int_mode != E1000E_INT_MODE_LEGACY) {
3912                 err = e1000_test_msi(adapter);
3913                 if (err) {
3914                         e_err("Interrupt allocation failed\n");
3915                         goto err_req_irq;
3916                 }
3917         }
3918
3919         /* From here on the code is the same as e1000e_up() */
3920         clear_bit(__E1000_DOWN, &adapter->state);
3921
3922         napi_enable(&adapter->napi);
3923
3924         e1000_irq_enable(adapter);
3925
3926         adapter->tx_hang_recheck = false;
3927         netif_start_queue(netdev);
3928
3929         adapter->idle_check = true;
3930         pm_runtime_put(&pdev->dev);
3931
3932         /* fire a link status change interrupt to start the watchdog */
3933         if (adapter->msix_entries)
3934                 ew32(ICS, E1000_ICS_LSC | E1000_ICR_OTHER);
3935         else
3936                 ew32(ICS, E1000_ICS_LSC);
3937
3938         return 0;
3939
3940 err_req_irq:
3941         e1000e_release_hw_control(adapter);
3942         e1000_power_down_phy(adapter);
3943         e1000e_free_rx_resources(adapter->rx_ring);
3944 err_setup_rx:
3945         e1000e_free_tx_resources(adapter->tx_ring);
3946 err_setup_tx:
3947         e1000e_reset(adapter);
3948         pm_runtime_put_sync(&pdev->dev);
3949
3950         return err;
3951 }
3952
3953 /**
3954  * e1000_close - Disables a network interface
3955  * @netdev: network interface device structure
3956  *
3957  * Returns 0, this is not allowed to fail
3958  *
3959  * The close entry point is called when an interface is de-activated
3960  * by the OS.  The hardware is still under the drivers control, but
3961  * needs to be disabled.  A global MAC reset is issued to stop the
3962  * hardware, and all transmit and receive resources are freed.
3963  **/
3964 static int e1000_close(struct net_device *netdev)
3965 {
3966         struct e1000_adapter *adapter = netdev_priv(netdev);
3967         struct pci_dev *pdev = adapter->pdev;
3968         int count = E1000_CHECK_RESET_COUNT;
3969
3970         while (test_bit(__E1000_RESETTING, &adapter->state) && count--)
3971                 usleep_range(10000, 20000);
3972
3973         WARN_ON(test_bit(__E1000_RESETTING, &adapter->state));
3974
3975         pm_runtime_get_sync(&pdev->dev);
3976
3977         napi_disable(&adapter->napi);
3978
3979         if (!test_bit(__E1000_DOWN, &adapter->state)) {
3980                 e1000e_down(adapter);
3981                 e1000_free_irq(adapter);
3982         }
3983         e1000_power_down_phy(adapter);
3984
3985         e1000e_free_tx_resources(adapter->tx_ring);
3986         e1000e_free_rx_resources(adapter->rx_ring);
3987
3988         /*
3989          * kill manageability vlan ID if supported, but not if a vlan with
3990          * the same ID is registered on the host OS (let 8021q kill it)
3991          */
3992         if (adapter->hw.mng_cookie.status &
3993             E1000_MNG_DHCP_COOKIE_STATUS_VLAN)
3994                 e1000_vlan_rx_kill_vid(netdev, adapter->mng_vlan_id);
3995
3996         /*
3997          * If AMT is enabled, let the firmware know that the network
3998          * interface is now closed
3999          */
4000         if ((adapter->flags & FLAG_HAS_AMT) &&
4001             !test_bit(__E1000_TESTING, &adapter->state))
4002                 e1000e_release_hw_control(adapter);
4003
4004         if (adapter->hw.mac.type == e1000_pch2lan)
4005                 pm_qos_remove_request(&adapter->netdev->pm_qos_req);
4006
4007         pm_runtime_put_sync(&pdev->dev);
4008
4009         return 0;
4010 }
4011 /**
4012  * e1000_set_mac - Change the Ethernet Address of the NIC
4013  * @netdev: network interface device structure
4014  * @p: pointer to an address structure
4015  *
4016  * Returns 0 on success, negative on failure
4017  **/
4018 static int e1000_set_mac(struct net_device *netdev, void *p)
4019 {
4020         struct e1000_adapter *adapter = netdev_priv(netdev);
4021         struct sockaddr *addr = p;
4022
4023         if (!is_valid_ether_addr(addr->sa_data))
4024                 return -EADDRNOTAVAIL;
4025
4026         memcpy(netdev->dev_addr, addr->sa_data, netdev->addr_len);
4027         memcpy(adapter->hw.mac.addr, addr->sa_data, netdev->addr_len);
4028
4029         e1000e_rar_set(&adapter->hw, adapter->hw.mac.addr, 0);
4030
4031         if (adapter->flags & FLAG_RESET_OVERWRITES_LAA) {
4032                 /* activate the work around */
4033                 e1000e_set_laa_state_82571(&adapter->hw, 1);
4034
4035                 /*
4036                  * Hold a copy of the LAA in RAR[14] This is done so that
4037                  * between the time RAR[0] gets clobbered  and the time it
4038                  * gets fixed (in e1000_watchdog), the actual LAA is in one
4039                  * of the RARs and no incoming packets directed to this port
4040                  * are dropped. Eventually the LAA will be in RAR[0] and
4041                  * RAR[14]
4042                  */
4043                 e1000e_rar_set(&adapter->hw,
4044                               adapter->hw.mac.addr,
4045                               adapter->hw.mac.rar_entry_count - 1);
4046         }
4047
4048         return 0;
4049 }
4050
4051 /**
4052  * e1000e_update_phy_task - work thread to update phy
4053  * @work: pointer to our work struct
4054  *
4055  * this worker thread exists because we must acquire a
4056  * semaphore to read the phy, which we could msleep while
4057  * waiting for it, and we can't msleep in a timer.
4058  **/
4059 static void e1000e_update_phy_task(struct work_struct *work)
4060 {
4061         struct e1000_adapter *adapter = container_of(work,
4062                                         struct e1000_adapter, update_phy_task);
4063
4064         if (test_bit(__E1000_DOWN, &adapter->state))
4065                 return;
4066
4067         e1000_get_phy_info(&adapter->hw);
4068 }
4069
4070 /*
4071  * Need to wait a few seconds after link up to get diagnostic information from
4072  * the phy
4073  */
4074 static void e1000_update_phy_info(unsigned long data)
4075 {
4076         struct e1000_adapter *adapter = (struct e1000_adapter *) data;
4077
4078         if (test_bit(__E1000_DOWN, &adapter->state))
4079                 return;
4080
4081         schedule_work(&adapter->update_phy_task);
4082 }
4083
4084 /**
4085  * e1000e_update_phy_stats - Update the PHY statistics counters
4086  * @adapter: board private structure
4087  *
4088  * Read/clear the upper 16-bit PHY registers and read/accumulate lower
4089  **/
4090 static void e1000e_update_phy_stats(struct e1000_adapter *adapter)
4091 {
4092         struct e1000_hw *hw = &adapter->hw;
4093         s32 ret_val;
4094         u16 phy_data;
4095
4096         ret_val = hw->phy.ops.acquire(hw);
4097         if (ret_val)
4098                 return;
4099
4100         /*
4101          * A page set is expensive so check if already on desired page.
4102          * If not, set to the page with the PHY status registers.
4103          */
4104         hw->phy.addr = 1;
4105         ret_val = e1000e_read_phy_reg_mdic(hw, IGP01E1000_PHY_PAGE_SELECT,
4106                                            &phy_data);
4107         if (ret_val)
4108                 goto release;
4109         if (phy_data != (HV_STATS_PAGE << IGP_PAGE_SHIFT)) {
4110                 ret_val = hw->phy.ops.set_page(hw,
4111                                                HV_STATS_PAGE << IGP_PAGE_SHIFT);
4112                 if (ret_val)
4113                         goto release;
4114         }
4115
4116         /* Single Collision Count */
4117         hw->phy.ops.read_reg_page(hw, HV_SCC_UPPER, &phy_data);
4118         ret_val = hw->phy.ops.read_reg_page(hw, HV_SCC_LOWER, &phy_data);
4119         if (!ret_val)
4120                 adapter->stats.scc += phy_data;
4121
4122         /* Excessive Collision Count */
4123         hw->phy.ops.read_reg_page(hw, HV_ECOL_UPPER, &phy_data);
4124         ret_val = hw->phy.ops.read_reg_page(hw, HV_ECOL_LOWER, &phy_data);
4125         if (!ret_val)
4126                 adapter->stats.ecol += phy_data;
4127
4128         /* Multiple Collision Count */
4129         hw->phy.ops.read_reg_page(hw, HV_MCC_UPPER, &phy_data);
4130         ret_val = hw->phy.ops.read_reg_page(hw, HV_MCC_LOWER, &phy_data);
4131         if (!ret_val)
4132                 adapter->stats.mcc += phy_data;
4133
4134         /* Late Collision Count */
4135         hw->phy.ops.read_reg_page(hw, HV_LATECOL_UPPER, &phy_data);
4136         ret_val = hw->phy.ops.read_reg_page(hw, HV_LATECOL_LOWER, &phy_data);
4137         if (!ret_val)
4138                 adapter->stats.latecol += phy_data;
4139
4140         /* Collision Count - also used for adaptive IFS */
4141         hw->phy.ops.read_reg_page(hw, HV_COLC_UPPER, &phy_data);
4142         ret_val = hw->phy.ops.read_reg_page(hw, HV_COLC_LOWER, &phy_data);
4143         if (!ret_val)
4144                 hw->mac.collision_delta = phy_data;
4145
4146         /* Defer Count */
4147         hw->phy.ops.read_reg_page(hw, HV_DC_UPPER, &phy_data);
4148         ret_val = hw->phy.ops.read_reg_page(hw, HV_DC_LOWER, &phy_data);
4149         if (!ret_val)
4150                 adapter->stats.dc += phy_data;
4151
4152         /* Transmit with no CRS */
4153         hw->phy.ops.read_reg_page(hw, HV_TNCRS_UPPER, &phy_data);
4154         ret_val = hw->phy.ops.read_reg_page(hw, HV_TNCRS_LOWER, &phy_data);
4155         if (!ret_val)
4156                 adapter->stats.tncrs += phy_data;
4157
4158 release:
4159         hw->phy.ops.release(hw);
4160 }
4161
4162 /**
4163  * e1000e_update_stats - Update the board statistics counters
4164  * @adapter: board private structure
4165  **/
4166 static void e1000e_update_stats(struct e1000_adapter *adapter)
4167 {
4168         struct net_device *netdev = adapter->netdev;
4169         struct e1000_hw *hw = &adapter->hw;
4170         struct pci_dev *pdev = adapter->pdev;
4171
4172         /*
4173          * Prevent stats update while adapter is being reset, or if the pci
4174          * connection is down.
4175          */
4176         if (adapter->link_speed == 0)
4177                 return;
4178         if (pci_channel_offline(pdev))
4179                 return;
4180
4181         adapter->stats.crcerrs += er32(CRCERRS);
4182         adapter->stats.gprc += er32(GPRC);
4183         adapter->stats.gorc += er32(GORCL);
4184         er32(GORCH); /* Clear gorc */
4185         adapter->stats.bprc += er32(BPRC);
4186         adapter->stats.mprc += er32(MPRC);
4187         adapter->stats.roc += er32(ROC);
4188
4189         adapter->stats.mpc += er32(MPC);
4190
4191         /* Half-duplex statistics */
4192         if (adapter->link_duplex == HALF_DUPLEX) {
4193                 if (adapter->flags2 & FLAG2_HAS_PHY_STATS) {
4194                         e1000e_update_phy_stats(adapter);
4195                 } else {
4196                         adapter->stats.scc += er32(SCC);
4197                         adapter->stats.ecol += er32(ECOL);
4198                         adapter->stats.mcc += er32(MCC);
4199                         adapter->stats.latecol += er32(LATECOL);
4200                         adapter->stats.dc += er32(DC);
4201
4202                         hw->mac.collision_delta = er32(COLC);
4203
4204                         if ((hw->mac.type != e1000_82574) &&
4205                             (hw->mac.type != e1000_82583))
4206                                 adapter->stats.tncrs += er32(TNCRS);
4207                 }
4208                 adapter->stats.colc += hw->mac.collision_delta;
4209         }
4210
4211         adapter->stats.xonrxc += er32(XONRXC);
4212         adapter->stats.xontxc += er32(XONTXC);
4213         adapter->stats.xoffrxc += er32(XOFFRXC);
4214         adapter->stats.xofftxc += er32(XOFFTXC);
4215         adapter->stats.gptc += er32(GPTC);
4216         adapter->stats.gotc += er32(GOTCL);
4217         er32(GOTCH); /* Clear gotc */
4218         adapter->stats.rnbc += er32(RNBC);
4219         adapter->stats.ruc += er32(RUC);
4220
4221         adapter->stats.mptc += er32(MPTC);
4222         adapter->stats.bptc += er32(BPTC);
4223
4224         /* used for adaptive IFS */
4225
4226         hw->mac.tx_packet_delta = er32(TPT);
4227         adapter->stats.tpt += hw->mac.tx_packet_delta;
4228
4229         adapter->stats.algnerrc += er32(ALGNERRC);
4230         adapter->stats.rxerrc += er32(RXERRC);
4231         adapter->stats.cexterr += er32(CEXTERR);
4232         adapter->stats.tsctc += er32(TSCTC);
4233         adapter->stats.tsctfc += er32(TSCTFC);
4234
4235         /* Fill out the OS statistics structure */
4236         netdev->stats.multicast = adapter->stats.mprc;
4237         netdev->stats.collisions = adapter->stats.colc;
4238
4239         /* Rx Errors */
4240
4241         /*
4242          * RLEC on some newer hardware can be incorrect so build
4243          * our own version based on RUC and ROC
4244          */
4245         netdev->stats.rx_errors = adapter->stats.rxerrc +
4246                 adapter->stats.crcerrs + adapter->stats.algnerrc +
4247                 adapter->stats.ruc + adapter->stats.roc +
4248                 adapter->stats.cexterr;
4249         netdev->stats.rx_length_errors = adapter->stats.ruc +
4250                                               adapter->stats.roc;
4251         netdev->stats.rx_crc_errors = adapter->stats.crcerrs;
4252         netdev->stats.rx_frame_errors = adapter->stats.algnerrc;
4253         netdev->stats.rx_missed_errors = adapter->stats.mpc;
4254
4255         /* Tx Errors */
4256         netdev->stats.tx_errors = adapter->stats.ecol +
4257                                        adapter->stats.latecol;
4258         netdev->stats.tx_aborted_errors = adapter->stats.ecol;
4259         netdev->stats.tx_window_errors = adapter->stats.latecol;
4260         netdev->stats.tx_carrier_errors = adapter->stats.tncrs;
4261
4262         /* Tx Dropped needs to be maintained elsewhere */
4263
4264         /* Management Stats */
4265         adapter->stats.mgptc += er32(MGTPTC);
4266         adapter->stats.mgprc += er32(MGTPRC);
4267         adapter->stats.mgpdc += er32(MGTPDC);
4268 }
4269
4270 /**
4271  * e1000_phy_read_status - Update the PHY register status snapshot
4272  * @adapter: board private structure
4273  **/
4274 static void e1000_phy_read_status(struct e1000_adapter *adapter)
4275 {
4276         struct e1000_hw *hw = &adapter->hw;
4277         struct e1000_phy_regs *phy = &adapter->phy_regs;
4278
4279         if ((er32(STATUS) & E1000_STATUS_LU) &&
4280             (adapter->hw.phy.media_type == e1000_media_type_copper)) {
4281                 int ret_val;
4282
4283                 ret_val  = e1e_rphy(hw, PHY_CONTROL, &phy->bmcr);
4284                 ret_val |= e1e_rphy(hw, PHY_STATUS, &phy->bmsr);
4285                 ret_val |= e1e_rphy(hw, PHY_AUTONEG_ADV, &phy->advertise);
4286                 ret_val |= e1e_rphy(hw, PHY_LP_ABILITY, &phy->lpa);
4287                 ret_val |= e1e_rphy(hw, PHY_AUTONEG_EXP, &phy->expansion);
4288                 ret_val |= e1e_rphy(hw, PHY_1000T_CTRL, &phy->ctrl1000);
4289                 ret_val |= e1e_rphy(hw, PHY_1000T_STATUS, &phy->stat1000);
4290                 ret_val |= e1e_rphy(hw, PHY_EXT_STATUS, &phy->estatus);
4291                 if (ret_val)
4292                         e_warn("Error reading PHY register\n");
4293         } else {
4294                 /*
4295                  * Do not read PHY registers if link is not up
4296                  * Set values to typical power-on defaults
4297                  */
4298                 phy->bmcr = (BMCR_SPEED1000 | BMCR_ANENABLE | BMCR_FULLDPLX);
4299                 phy->bmsr = (BMSR_100FULL | BMSR_100HALF | BMSR_10FULL |
4300                              BMSR_10HALF | BMSR_ESTATEN | BMSR_ANEGCAPABLE |
4301                              BMSR_ERCAP);
4302                 phy->advertise = (ADVERTISE_PAUSE_ASYM | ADVERTISE_PAUSE_CAP |
4303                                   ADVERTISE_ALL | ADVERTISE_CSMA);
4304                 phy->lpa = 0;
4305                 phy->expansion = EXPANSION_ENABLENPAGE;
4306                 phy->ctrl1000 = ADVERTISE_1000FULL;
4307                 phy->stat1000 = 0;
4308                 phy->estatus = (ESTATUS_1000_TFULL | ESTATUS_1000_THALF);
4309         }
4310 }
4311
4312 static void e1000_print_link_info(struct e1000_adapter *adapter)
4313 {
4314         struct e1000_hw *hw = &adapter->hw;
4315         u32 ctrl = er32(CTRL);
4316
4317         /* Link status message must follow this format for user tools */
4318         printk(KERN_INFO "e1000e: %s NIC Link is Up %d Mbps %s Duplex, Flow Control: %s\n",
4319                 adapter->netdev->name,
4320                 adapter->link_speed,
4321                 adapter->link_duplex == FULL_DUPLEX ? "Full" : "Half",
4322                 (ctrl & E1000_CTRL_TFCE) && (ctrl & E1000_CTRL_RFCE) ? "Rx/Tx" :
4323                 (ctrl & E1000_CTRL_RFCE) ? "Rx" :
4324                 (ctrl & E1000_CTRL_TFCE) ? "Tx" : "None");
4325 }
4326
4327 static bool e1000e_has_link(struct e1000_adapter *adapter)
4328 {
4329         struct e1000_hw *hw = &adapter->hw;
4330         bool link_active = false;
4331         s32 ret_val = 0;
4332
4333         /*
4334          * get_link_status is set on LSC (link status) interrupt or
4335          * Rx sequence error interrupt.  get_link_status will stay
4336          * false until the check_for_link establishes link
4337          * for copper adapters ONLY
4338          */
4339         switch (hw->phy.media_type) {
4340         case e1000_media_type_copper:
4341                 if (hw->mac.get_link_status) {
4342                         ret_val = hw->mac.ops.check_for_link(hw);
4343                         link_active = !hw->mac.get_link_status;
4344                 } else {
4345                         link_active = true;
4346                 }
4347                 break;
4348         case e1000_media_type_fiber:
4349                 ret_val = hw->mac.ops.check_for_link(hw);
4350                 link_active = !!(er32(STATUS) & E1000_STATUS_LU);
4351                 break;
4352         case e1000_media_type_internal_serdes:
4353                 ret_val = hw->mac.ops.check_for_link(hw);
4354                 link_active = adapter->hw.mac.serdes_has_link;
4355                 break;
4356         default:
4357         case e1000_media_type_unknown:
4358                 break;
4359         }
4360
4361         if ((ret_val == E1000_ERR_PHY) && (hw->phy.type == e1000_phy_igp_3) &&
4362             (er32(CTRL) & E1000_PHY_CTRL_GBE_DISABLE)) {
4363                 /* See e1000_kmrn_lock_loss_workaround_ich8lan() */
4364                 e_info("Gigabit has been disabled, downgrading speed\n");
4365         }
4366
4367         return link_active;
4368 }
4369
4370 static void e1000e_enable_receives(struct e1000_adapter *adapter)
4371 {
4372         /* make sure the receive unit is started */
4373         if ((adapter->flags & FLAG_RX_NEEDS_RESTART) &&
4374             (adapter->flags & FLAG_RX_RESTART_NOW)) {
4375                 struct e1000_hw *hw = &adapter->hw;
4376                 u32 rctl = er32(RCTL);
4377                 ew32(RCTL, rctl | E1000_RCTL_EN);
4378                 adapter->flags &= ~FLAG_RX_RESTART_NOW;
4379         }
4380 }
4381
4382 static void e1000e_check_82574_phy_workaround(struct e1000_adapter *adapter)
4383 {
4384         struct e1000_hw *hw = &adapter->hw;
4385
4386         /*
4387          * With 82574 controllers, PHY needs to be checked periodically
4388          * for hung state and reset, if two calls return true
4389          */
4390         if (e1000_check_phy_82574(hw))
4391                 adapter->phy_hang_count++;
4392         else
4393                 adapter->phy_hang_count = 0;
4394
4395         if (adapter->phy_hang_count > 1) {
4396                 adapter->phy_hang_count = 0;
4397                 schedule_work(&adapter->reset_task);
4398         }
4399 }
4400
4401 /**
4402  * e1000_watchdog - Timer Call-back
4403  * @data: pointer to adapter cast into an unsigned long
4404  **/
4405 static void e1000_watchdog(unsigned long data)
4406 {
4407         struct e1000_adapter *adapter = (struct e1000_adapter *) data;
4408
4409         /* Do the rest outside of interrupt context */
4410         schedule_work(&adapter->watchdog_task);
4411
4412         /* TODO: make this use queue_delayed_work() */
4413 }
4414
4415 static void e1000_watchdog_task(struct work_struct *work)
4416 {
4417         struct e1000_adapter *adapter = container_of(work,
4418                                         struct e1000_adapter, watchdog_task);
4419         struct net_device *netdev = adapter->netdev;
4420         struct e1000_mac_info *mac = &adapter->hw.mac;
4421         struct e1000_phy_info *phy = &adapter->hw.phy;
4422         struct e1000_ring *tx_ring = adapter->tx_ring;
4423         struct e1000_hw *hw = &adapter->hw;
4424         u32 link, tctl;
4425
4426         if (test_bit(__E1000_DOWN, &adapter->state))
4427                 return;
4428
4429         link = e1000e_has_link(adapter);
4430         if ((netif_carrier_ok(netdev)) && link) {
4431                 /* Cancel scheduled suspend requests. */
4432                 pm_runtime_resume(netdev->dev.parent);
4433
4434                 e1000e_enable_receives(adapter);
4435                 goto link_up;
4436         }
4437
4438         if ((e1000e_enable_tx_pkt_filtering(hw)) &&
4439             (adapter->mng_vlan_id != adapter->hw.mng_cookie.vlan_id))
4440                 e1000_update_mng_vlan(adapter);
4441
4442         if (link) {
4443                 if (!netif_carrier_ok(netdev)) {
4444                         bool txb2b = true;
4445
4446                         /* Cancel scheduled suspend requests. */
4447                         pm_runtime_resume(netdev->dev.parent);
4448
4449                         /* update snapshot of PHY registers on LSC */
4450                         e1000_phy_read_status(adapter);
4451                         mac->ops.get_link_up_info(&adapter->hw,
4452                                                    &adapter->link_speed,
4453                                                    &adapter->link_duplex);
4454                         e1000_print_link_info(adapter);
4455                         /*
4456                          * On supported PHYs, check for duplex mismatch only
4457                          * if link has autonegotiated at 10/100 half
4458                          */
4459                         if ((hw->phy.type == e1000_phy_igp_3 ||
4460                              hw->phy.type == e1000_phy_bm) &&
4461                             (hw->mac.autoneg == true) &&
4462                             (adapter->link_speed == SPEED_10 ||
4463                              adapter->link_speed == SPEED_100) &&
4464                             (adapter->link_duplex == HALF_DUPLEX)) {
4465                                 u16 autoneg_exp;
4466
4467                                 e1e_rphy(hw, PHY_AUTONEG_EXP, &autoneg_exp);
4468
4469                                 if (!(autoneg_exp & NWAY_ER_LP_NWAY_CAPS))
4470                                         e_info("Autonegotiated half duplex but link partner cannot autoneg.  Try forcing full duplex if link gets many collisions.\n");
4471                         }
4472
4473                         /* adjust timeout factor according to speed/duplex */
4474                         adapter->tx_timeout_factor = 1;
4475                         switch (adapter->link_speed) {
4476                         case SPEED_10:
4477                                 txb2b = false;
4478                                 adapter->tx_timeout_factor = 16;
4479                                 break;
4480                         case SPEED_100:
4481                                 txb2b = false;
4482                                 adapter->tx_timeout_factor = 10;
4483                                 break;
4484                         }
4485
4486                         /*
4487                          * workaround: re-program speed mode bit after
4488                          * link-up event
4489                          */
4490                         if ((adapter->flags & FLAG_TARC_SPEED_MODE_BIT) &&
4491                             !txb2b) {
4492                                 u32 tarc0;
4493                                 tarc0 = er32(TARC(0));
4494                                 tarc0 &= ~SPEED_MODE_BIT;
4495                                 ew32(TARC(0), tarc0);
4496                         }
4497
4498                         /*
4499                          * disable TSO for pcie and 10/100 speeds, to avoid
4500                          * some hardware issues
4501                          */
4502                         if (!(adapter->flags & FLAG_TSO_FORCE)) {
4503                                 switch (adapter->link_speed) {
4504                                 case SPEED_10:
4505                                 case SPEED_100:
4506                                         e_info("10/100 speed: disabling TSO\n");
4507                                         netdev->features &= ~NETIF_F_TSO;
4508                                         netdev->features &= ~NETIF_F_TSO6;
4509                                         break;
4510                                 case SPEED_1000:
4511                                         netdev->features |= NETIF_F_TSO;
4512                                         netdev->features |= NETIF_F_TSO6;
4513                                         break;
4514                                 default:
4515                                         /* oops */
4516                                         break;
4517                                 }
4518                         }
4519
4520                         /*
4521                          * enable transmits in the hardware, need to do this
4522                          * after setting TARC(0)
4523                          */
4524                         tctl = er32(TCTL);
4525                         tctl |= E1000_TCTL_EN;
4526                         ew32(TCTL, tctl);
4527
4528                         /*
4529                          * Perform any post-link-up configuration before
4530                          * reporting link up.
4531                          */
4532                         if (phy->ops.cfg_on_link_up)
4533                                 phy->ops.cfg_on_link_up(hw);
4534
4535                         netif_carrier_on(netdev);
4536
4537                         if (!test_bit(__E1000_DOWN, &adapter->state))
4538                                 mod_timer(&adapter->phy_info_timer,
4539                                           round_jiffies(jiffies + 2 * HZ));
4540                 }
4541         } else {
4542                 if (netif_carrier_ok(netdev)) {
4543                         adapter->link_speed = 0;
4544                         adapter->link_duplex = 0;
4545                         /* Link status message must follow this format */
4546                         printk(KERN_INFO "e1000e: %s NIC Link is Down\n",
4547                                adapter->netdev->name);
4548                         netif_carrier_off(netdev);
4549                         if (!test_bit(__E1000_DOWN, &adapter->state))
4550                                 mod_timer(&adapter->phy_info_timer,
4551                                           round_jiffies(jiffies + 2 * HZ));
4552
4553                         if (adapter->flags & FLAG_RX_NEEDS_RESTART)
4554                                 schedule_work(&adapter->reset_task);
4555                         else
4556                                 pm_schedule_suspend(netdev->dev.parent,
4557                                                         LINK_TIMEOUT);
4558                 }
4559         }
4560
4561 link_up:
4562         spin_lock(&adapter->stats64_lock);
4563         e1000e_update_stats(adapter);
4564
4565         mac->tx_packet_delta = adapter->stats.tpt - adapter->tpt_old;
4566         adapter->tpt_old = adapter->stats.tpt;
4567         mac->collision_delta = adapter->stats.colc - adapter->colc_old;
4568         adapter->colc_old = adapter->stats.colc;
4569
4570         adapter->gorc = adapter->stats.gorc - adapter->gorc_old;
4571         adapter->gorc_old = adapter->stats.gorc;
4572         adapter->gotc = adapter->stats.gotc - adapter->gotc_old;
4573         adapter->gotc_old = adapter->stats.gotc;
4574         spin_unlock(&adapter->stats64_lock);
4575
4576         e1000e_update_adaptive(&adapter->hw);
4577
4578         if (!netif_carrier_ok(netdev) &&
4579             (e1000_desc_unused(tx_ring) + 1 < tx_ring->count)) {
4580                 /*
4581                  * We've lost link, so the controller stops DMA,
4582                  * but we've got queued Tx work that's never going
4583                  * to get done, so reset controller to flush Tx.
4584                  * (Do the reset outside of interrupt context).
4585                  */
4586                 schedule_work(&adapter->reset_task);
4587                 /* return immediately since reset is imminent */
4588                 return;
4589         }
4590
4591         /* Simple mode for Interrupt Throttle Rate (ITR) */
4592         if (adapter->itr_setting == 4) {
4593                 /*
4594                  * Symmetric Tx/Rx gets a reduced ITR=2000;
4595                  * Total asymmetrical Tx or Rx gets ITR=8000;
4596                  * everyone else is between 2000-8000.
4597                  */
4598                 u32 goc = (adapter->gotc + adapter->gorc) / 10000;
4599                 u32 dif = (adapter->gotc > adapter->gorc ?
4600                             adapter->gotc - adapter->gorc :
4601                             adapter->gorc - adapter->gotc) / 10000;
4602                 u32 itr = goc > 0 ? (dif * 6000 / goc + 2000) : 8000;
4603
4604                 ew32(ITR, 1000000000 / (itr * 256));
4605         }
4606
4607         /* Cause software interrupt to ensure Rx ring is cleaned */
4608         if (adapter->msix_entries)
4609                 ew32(ICS, adapter->rx_ring->ims_val);
4610         else
4611                 ew32(ICS, E1000_ICS_RXDMT0);
4612
4613         /* flush pending descriptors to memory before detecting Tx hang */
4614         e1000e_flush_descriptors(adapter);
4615
4616         /* Force detection of hung controller every watchdog period */
4617         adapter->detect_tx_hung = true;
4618
4619         /*
4620          * With 82571 controllers, LAA may be overwritten due to controller
4621          * reset from the other port. Set the appropriate LAA in RAR[0]
4622          */
4623         if (e1000e_get_laa_state_82571(hw))
4624                 e1000e_rar_set(hw, adapter->hw.mac.addr, 0);
4625
4626         if (adapter->flags2 & FLAG2_CHECK_PHY_HANG)
4627                 e1000e_check_82574_phy_workaround(adapter);
4628
4629         /* Reset the timer */
4630         if (!test_bit(__E1000_DOWN, &adapter->state))
4631                 mod_timer(&adapter->watchdog_timer,
4632                           round_jiffies(jiffies + 2 * HZ));
4633 }
4634
4635 #define E1000_TX_FLAGS_CSUM             0x00000001
4636 #define E1000_TX_FLAGS_VLAN             0x00000002
4637 #define E1000_TX_FLAGS_TSO              0x00000004
4638 #define E1000_TX_FLAGS_IPV4             0x00000008
4639 #define E1000_TX_FLAGS_NO_FCS           0x00000010
4640 #define E1000_TX_FLAGS_VLAN_MASK        0xffff0000
4641 #define E1000_TX_FLAGS_VLAN_SHIFT       16
4642
4643 static int e1000_tso(struct e1000_ring *tx_ring, struct sk_buff *skb)
4644 {
4645         struct e1000_context_desc *context_desc;
4646         struct e1000_buffer *buffer_info;
4647         unsigned int i;
4648         u32 cmd_length = 0;
4649         u16 ipcse = 0, tucse, mss;
4650         u8 ipcss, ipcso, tucss, tucso, hdr_len;
4651
4652         if (!skb_is_gso(skb))
4653                 return 0;
4654
4655         if (skb_header_cloned(skb)) {
4656                 int err = pskb_expand_head(skb, 0, 0, GFP_ATOMIC);
4657
4658                 if (err)
4659                         return err;
4660         }
4661
4662         hdr_len = skb_transport_offset(skb) + tcp_hdrlen(skb);
4663         mss = skb_shinfo(skb)->gso_size;
4664         if (skb->protocol == htons(ETH_P_IP)) {
4665                 struct iphdr *iph = ip_hdr(skb);
4666                 iph->tot_len = 0;
4667                 iph->check = 0;
4668                 tcp_hdr(skb)->check = ~csum_tcpudp_magic(iph->saddr, iph->daddr,
4669                                                          0, IPPROTO_TCP, 0);
4670                 cmd_length = E1000_TXD_CMD_IP;
4671                 ipcse = skb_transport_offset(skb) - 1;
4672         } else if (skb_is_gso_v6(skb)) {
4673                 ipv6_hdr(skb)->payload_len = 0;
4674                 tcp_hdr(skb)->check = ~csum_ipv6_magic(&ipv6_hdr(skb)->saddr,
4675                                                        &ipv6_hdr(skb)->daddr,
4676                                                        0, IPPROTO_TCP, 0);
4677                 ipcse = 0;
4678         }
4679         ipcss = skb_network_offset(skb);
4680         ipcso = (void *)&(ip_hdr(skb)->check) - (void *)skb->data;
4681         tucss = skb_transport_offset(skb);
4682         tucso = (void *)&(tcp_hdr(skb)->check) - (void *)skb->data;
4683         tucse = 0;
4684
4685         cmd_length |= (E1000_TXD_CMD_DEXT | E1000_TXD_CMD_TSE |
4686                        E1000_TXD_CMD_TCP | (skb->len - (hdr_len)));
4687
4688         i = tx_ring->next_to_use;
4689         context_desc = E1000_CONTEXT_DESC(*tx_ring, i);
4690         buffer_info = &tx_ring->buffer_info[i];
4691
4692         context_desc->lower_setup.ip_fields.ipcss  = ipcss;
4693         context_desc->lower_setup.ip_fields.ipcso  = ipcso;
4694         context_desc->lower_setup.ip_fields.ipcse  = cpu_to_le16(ipcse);
4695         context_desc->upper_setup.tcp_fields.tucss = tucss;
4696         context_desc->upper_setup.tcp_fields.tucso = tucso;
4697         context_desc->upper_setup.tcp_fields.tucse = cpu_to_le16(tucse);
4698         context_desc->tcp_seg_setup.fields.mss     = cpu_to_le16(mss);
4699         context_desc->tcp_seg_setup.fields.hdr_len = hdr_len;
4700         context_desc->cmd_and_length = cpu_to_le32(cmd_length);
4701
4702         buffer_info->time_stamp = jiffies;
4703         buffer_info->next_to_watch = i;
4704
4705         i++;
4706         if (i == tx_ring->count)
4707                 i = 0;
4708         tx_ring->next_to_use = i;
4709
4710         return 1;
4711 }
4712
4713 static bool e1000_tx_csum(struct e1000_ring *tx_ring, struct sk_buff *skb)
4714 {
4715         struct e1000_adapter *adapter = tx_ring->adapter;
4716         struct e1000_context_desc *context_desc;
4717         struct e1000_buffer *buffer_info;
4718         unsigned int i;
4719         u8 css;
4720         u32 cmd_len = E1000_TXD_CMD_DEXT;
4721         __be16 protocol;
4722
4723         if (skb->ip_summed != CHECKSUM_PARTIAL)
4724                 return 0;
4725
4726         if (skb->protocol == cpu_to_be16(ETH_P_8021Q))
4727                 protocol = vlan_eth_hdr(skb)->h_vlan_encapsulated_proto;
4728         else
4729                 protocol = skb->protocol;
4730
4731         switch (protocol) {
4732         case cpu_to_be16(ETH_P_IP):
4733                 if (ip_hdr(skb)->protocol == IPPROTO_TCP)
4734                         cmd_len |= E1000_TXD_CMD_TCP;
4735                 break;
4736         case cpu_to_be16(ETH_P_IPV6):
4737                 /* XXX not handling all IPV6 headers */
4738                 if (ipv6_hdr(skb)->nexthdr == IPPROTO_TCP)
4739                         cmd_len |= E1000_TXD_CMD_TCP;
4740                 break;
4741         default:
4742                 if (unlikely(net_ratelimit()))
4743                         e_warn("checksum_partial proto=%x!\n",
4744                                be16_to_cpu(protocol));
4745                 break;
4746         }
4747
4748         css = skb_checksum_start_offset(skb);
4749
4750         i = tx_ring->next_to_use;
4751         buffer_info = &tx_ring->buffer_info[i];
4752         context_desc = E1000_CONTEXT_DESC(*tx_ring, i);
4753
4754         context_desc->lower_setup.ip_config = 0;
4755         context_desc->upper_setup.tcp_fields.tucss = css;
4756         context_desc->upper_setup.tcp_fields.tucso =
4757                                 css + skb->csum_offset;
4758         context_desc->upper_setup.tcp_fields.tucse = 0;
4759         context_desc->tcp_seg_setup.data = 0;
4760         context_desc->cmd_and_length = cpu_to_le32(cmd_len);
4761
4762         buffer_info->time_stamp = jiffies;
4763         buffer_info->next_to_watch = i;
4764
4765         i++;
4766         if (i == tx_ring->count)
4767                 i = 0;
4768         tx_ring->next_to_use = i;
4769
4770         return 1;
4771 }
4772
4773 #define E1000_MAX_PER_TXD       8192
4774 #define E1000_MAX_TXD_PWR       12
4775
4776 static int e1000_tx_map(struct e1000_ring *tx_ring, struct sk_buff *skb,
4777                         unsigned int first, unsigned int max_per_txd,
4778                         unsigned int nr_frags, unsigned int mss)
4779 {
4780         struct e1000_adapter *adapter = tx_ring->adapter;
4781         struct pci_dev *pdev = adapter->pdev;
4782         struct e1000_buffer *buffer_info;
4783         unsigned int len = skb_headlen(skb);
4784         unsigned int offset = 0, size, count = 0, i;
4785         unsigned int f, bytecount, segs;
4786
4787         i = tx_ring->next_to_use;
4788
4789         while (len) {
4790                 buffer_info = &tx_ring->buffer_info[i];
4791                 size = min(len, max_per_txd);
4792
4793                 buffer_info->length = size;
4794                 buffer_info->time_stamp = jiffies;
4795                 buffer_info->next_to_watch = i;
4796                 buffer_info->dma = dma_map_single(&pdev->dev,
4797                                                   skb->data + offset,
4798                                                   size, DMA_TO_DEVICE);
4799                 buffer_info->mapped_as_page = false;
4800                 if (dma_mapping_error(&pdev->dev, buffer_info->dma))
4801                         goto dma_error;
4802
4803                 len -= size;
4804                 offset += size;
4805                 count++;
4806
4807                 if (len) {
4808                         i++;
4809                         if (i == tx_ring->count)
4810                                 i = 0;
4811                 }
4812         }
4813
4814         for (f = 0; f < nr_frags; f++) {
4815                 const struct skb_frag_struct *frag;
4816
4817                 frag = &skb_shinfo(skb)->frags[f];
4818                 len = skb_frag_size(frag);
4819                 offset = 0;
4820
4821                 while (len) {
4822                         i++;
4823                         if (i == tx_ring->count)
4824                                 i = 0;
4825
4826                         buffer_info = &tx_ring->buffer_info[i];
4827                         size = min(len, max_per_txd);
4828
4829                         buffer_info->length = size;
4830                         buffer_info->time_stamp = jiffies;
4831                         buffer_info->next_to_watch = i;
4832                         buffer_info->dma = skb_frag_dma_map(&pdev->dev, frag,
4833                                                 offset, size, DMA_TO_DEVICE);
4834                         buffer_info->mapped_as_page = true;
4835                         if (dma_mapping_error(&pdev->dev, buffer_info->dma))
4836                                 goto dma_error;
4837
4838                         len -= size;
4839                         offset += size;
4840                         count++;
4841                 }
4842         }
4843
4844         segs = skb_shinfo(skb)->gso_segs ? : 1;
4845         /* multiply data chunks by size of headers */
4846         bytecount = ((segs - 1) * skb_headlen(skb)) + skb->len;
4847
4848         tx_ring->buffer_info[i].skb = skb;
4849         tx_ring->buffer_info[i].segs = segs;
4850         tx_ring->buffer_info[i].bytecount = bytecount;
4851         tx_ring->buffer_info[first].next_to_watch = i;
4852
4853         return count;
4854
4855 dma_error:
4856         dev_err(&pdev->dev, "Tx DMA map failed\n");
4857         buffer_info->dma = 0;
4858         if (count)
4859                 count--;
4860
4861         while (count--) {
4862                 if (i == 0)
4863                         i += tx_ring->count;
4864                 i--;
4865                 buffer_info = &tx_ring->buffer_info[i];
4866                 e1000_put_txbuf(tx_ring, buffer_info);
4867         }
4868
4869         return 0;
4870 }
4871
4872 static void e1000_tx_queue(struct e1000_ring *tx_ring, int tx_flags, int count)
4873 {
4874         struct e1000_adapter *adapter = tx_ring->adapter;
4875         struct e1000_tx_desc *tx_desc = NULL;
4876         struct e1000_buffer *buffer_info;
4877         u32 txd_upper = 0, txd_lower = E1000_TXD_CMD_IFCS;
4878         unsigned int i;
4879
4880         if (tx_flags & E1000_TX_FLAGS_TSO) {
4881                 txd_lower |= E1000_TXD_CMD_DEXT | E1000_TXD_DTYP_D |
4882                              E1000_TXD_CMD_TSE;
4883                 txd_upper |= E1000_TXD_POPTS_TXSM << 8;
4884
4885                 if (tx_flags & E1000_TX_FLAGS_IPV4)
4886                         txd_upper |= E1000_TXD_POPTS_IXSM << 8;
4887         }
4888
4889         if (tx_flags & E1000_TX_FLAGS_CSUM) {
4890                 txd_lower |= E1000_TXD_CMD_DEXT | E1000_TXD_DTYP_D;
4891                 txd_upper |= E1000_TXD_POPTS_TXSM << 8;
4892         }
4893
4894         if (tx_flags & E1000_TX_FLAGS_VLAN) {
4895                 txd_lower |= E1000_TXD_CMD_VLE;
4896                 txd_upper |= (tx_flags & E1000_TX_FLAGS_VLAN_MASK);
4897         }
4898
4899         if (unlikely(tx_flags & E1000_TX_FLAGS_NO_FCS))
4900                 txd_lower &= ~(E1000_TXD_CMD_IFCS);
4901
4902         i = tx_ring->next_to_use;
4903
4904         do {
4905                 buffer_info = &tx_ring->buffer_info[i];
4906                 tx_desc = E1000_TX_DESC(*tx_ring, i);
4907                 tx_desc->buffer_addr = cpu_to_le64(buffer_info->dma);
4908                 tx_desc->lower.data =
4909                         cpu_to_le32(txd_lower | buffer_info->length);
4910                 tx_desc->upper.data = cpu_to_le32(txd_upper);
4911
4912                 i++;
4913                 if (i == tx_ring->count)
4914                         i = 0;
4915         } while (--count > 0);
4916
4917         tx_desc->lower.data |= cpu_to_le32(adapter->txd_cmd);
4918
4919         /* txd_cmd re-enables FCS, so we'll re-disable it here as desired. */
4920         if (unlikely(tx_flags & E1000_TX_FLAGS_NO_FCS))
4921                 tx_desc->lower.data &= ~(cpu_to_le32(E1000_TXD_CMD_IFCS));
4922
4923         /*
4924          * Force memory writes to complete before letting h/w
4925          * know there are new descriptors to fetch.  (Only
4926          * applicable for weak-ordered memory model archs,
4927          * such as IA-64).
4928          */
4929         wmb();
4930
4931         tx_ring->next_to_use = i;
4932
4933         if (adapter->flags2 & FLAG2_PCIM2PCI_ARBITER_WA)
4934                 e1000e_update_tdt_wa(tx_ring, i);
4935         else
4936                 writel(i, tx_ring->tail);
4937
4938         /*
4939          * we need this if more than one processor can write to our tail
4940          * at a time, it synchronizes IO on IA64/Altix systems
4941          */
4942         mmiowb();
4943 }
4944
4945 #define MINIMUM_DHCP_PACKET_SIZE 282
4946 static int e1000_transfer_dhcp_info(struct e1000_adapter *adapter,
4947                                     struct sk_buff *skb)
4948 {
4949         struct e1000_hw *hw =  &adapter->hw;
4950         u16 length, offset;
4951
4952         if (vlan_tx_tag_present(skb)) {
4953                 if (!((vlan_tx_tag_get(skb) == adapter->hw.mng_cookie.vlan_id) &&
4954                     (adapter->hw.mng_cookie.status &
4955                         E1000_MNG_DHCP_COOKIE_STATUS_VLAN)))
4956                         return 0;
4957         }
4958
4959         if (skb->len <= MINIMUM_DHCP_PACKET_SIZE)
4960                 return 0;
4961
4962         if (((struct ethhdr *) skb->data)->h_proto != htons(ETH_P_IP))
4963                 return 0;
4964
4965         {
4966                 const struct iphdr *ip = (struct iphdr *)((u8 *)skb->data+14);
4967                 struct udphdr *udp;
4968
4969                 if (ip->protocol != IPPROTO_UDP)
4970                         return 0;
4971
4972                 udp = (struct udphdr *)((u8 *)ip + (ip->ihl << 2));
4973                 if (ntohs(udp->dest) != 67)
4974                         return 0;
4975
4976                 offset = (u8 *)udp + 8 - skb->data;
4977                 length = skb->len - offset;
4978                 return e1000e_mng_write_dhcp_info(hw, (u8 *)udp + 8, length);
4979         }
4980
4981         return 0;
4982 }
4983
4984 static int __e1000_maybe_stop_tx(struct e1000_ring *tx_ring, int size)
4985 {
4986         struct e1000_adapter *adapter = tx_ring->adapter;
4987
4988         netif_stop_queue(adapter->netdev);
4989         /*
4990          * Herbert's original patch had:
4991          *  smp_mb__after_netif_stop_queue();
4992          * but since that doesn't exist yet, just open code it.
4993          */
4994         smp_mb();
4995
4996         /*
4997          * We need to check again in a case another CPU has just
4998          * made room available.
4999          */
5000         if (e1000_desc_unused(tx_ring) < size)
5001                 return -EBUSY;
5002
5003         /* A reprieve! */
5004         netif_start_queue(adapter->netdev);
5005         ++adapter->restart_queue;
5006         return 0;
5007 }
5008
5009 static int e1000_maybe_stop_tx(struct e1000_ring *tx_ring, int size)
5010 {
5011         if (e1000_desc_unused(tx_ring) >= size)
5012                 return 0;
5013         return __e1000_maybe_stop_tx(tx_ring, size);
5014 }
5015
5016 #define TXD_USE_COUNT(S, X) (((S) >> (X)) + 1)
5017 static netdev_tx_t e1000_xmit_frame(struct sk_buff *skb,
5018                                     struct net_device *netdev)
5019 {
5020         struct e1000_adapter *adapter = netdev_priv(netdev);
5021         struct e1000_ring *tx_ring = adapter->tx_ring;
5022         unsigned int first;
5023         unsigned int max_per_txd = E1000_MAX_PER_TXD;
5024         unsigned int max_txd_pwr = E1000_MAX_TXD_PWR;
5025         unsigned int tx_flags = 0;
5026         unsigned int len = skb_headlen(skb);
5027         unsigned int nr_frags;
5028         unsigned int mss;
5029         int count = 0;
5030         int tso;
5031         unsigned int f;
5032
5033         if (test_bit(__E1000_DOWN, &adapter->state)) {
5034                 dev_kfree_skb_any(skb);
5035                 return NETDEV_TX_OK;
5036         }
5037
5038         if (skb->len <= 0) {
5039                 dev_kfree_skb_any(skb);
5040                 return NETDEV_TX_OK;
5041         }
5042
5043         mss = skb_shinfo(skb)->gso_size;
5044         /*
5045          * The controller does a simple calculation to
5046          * make sure there is enough room in the FIFO before
5047          * initiating the DMA for each buffer.  The calc is:
5048          * 4 = ceil(buffer len/mss).  To make sure we don't
5049          * overrun the FIFO, adjust the max buffer len if mss
5050          * drops.
5051          */
5052         if (mss) {
5053                 u8 hdr_len;
5054                 max_per_txd = min(mss << 2, max_per_txd);
5055                 max_txd_pwr = fls(max_per_txd) - 1;
5056
5057                 /*
5058                  * TSO Workaround for 82571/2/3 Controllers -- if skb->data
5059                  * points to just header, pull a few bytes of payload from
5060                  * frags into skb->data
5061                  */
5062                 hdr_len = skb_transport_offset(skb) + tcp_hdrlen(skb);
5063                 /*
5064                  * we do this workaround for ES2LAN, but it is un-necessary,
5065                  * avoiding it could save a lot of cycles
5066                  */
5067                 if (skb->data_len && (hdr_len == len)) {
5068                         unsigned int pull_size;
5069
5070                         pull_size = min_t(unsigned int, 4, skb->data_len);
5071                         if (!__pskb_pull_tail(skb, pull_size)) {
5072                                 e_err("__pskb_pull_tail failed.\n");
5073                                 dev_kfree_skb_any(skb);
5074                                 return NETDEV_TX_OK;
5075                         }
5076                         len = skb_headlen(skb);
5077                 }
5078         }
5079
5080         /* reserve a descriptor for the offload context */
5081         if ((mss) || (skb->ip_summed == CHECKSUM_PARTIAL))
5082                 count++;
5083         count++;
5084
5085         count += TXD_USE_COUNT(len, max_txd_pwr);
5086
5087         nr_frags = skb_shinfo(skb)->nr_frags;
5088         for (f = 0; f < nr_frags; f++)
5089                 count += TXD_USE_COUNT(skb_frag_size(&skb_shinfo(skb)->frags[f]),
5090                                        max_txd_pwr);
5091
5092         if (adapter->hw.mac.tx_pkt_filtering)
5093                 e1000_transfer_dhcp_info(adapter, skb);
5094
5095         /*
5096          * need: count + 2 desc gap to keep tail from touching
5097          * head, otherwise try next time
5098          */
5099         if (e1000_maybe_stop_tx(tx_ring, count + 2))
5100                 return NETDEV_TX_BUSY;
5101
5102         if (vlan_tx_tag_present(skb)) {
5103                 tx_flags |= E1000_TX_FLAGS_VLAN;
5104                 tx_flags |= (vlan_tx_tag_get(skb) << E1000_TX_FLAGS_VLAN_SHIFT);
5105         }
5106
5107         first = tx_ring->next_to_use;
5108
5109         tso = e1000_tso(tx_ring, skb);
5110         if (tso < 0) {
5111                 dev_kfree_skb_any(skb);
5112                 return NETDEV_TX_OK;
5113         }
5114
5115         if (tso)
5116                 tx_flags |= E1000_TX_FLAGS_TSO;
5117         else if (e1000_tx_csum(tx_ring, skb))
5118                 tx_flags |= E1000_TX_FLAGS_CSUM;
5119
5120         /*
5121          * Old method was to assume IPv4 packet by default if TSO was enabled.
5122          * 82571 hardware supports TSO capabilities for IPv6 as well...
5123          * no longer assume, we must.
5124          */
5125         if (skb->protocol == htons(ETH_P_IP))
5126                 tx_flags |= E1000_TX_FLAGS_IPV4;
5127
5128         if (unlikely(skb->no_fcs))
5129                 tx_flags |= E1000_TX_FLAGS_NO_FCS;
5130
5131         /* if count is 0 then mapping error has occurred */
5132         count = e1000_tx_map(tx_ring, skb, first, max_per_txd, nr_frags, mss);
5133         if (count) {
5134                 skb_tx_timestamp(skb);
5135
5136                 netdev_sent_queue(netdev, skb->len);
5137                 e1000_tx_queue(tx_ring, tx_flags, count);
5138                 /* Make sure there is space in the ring for the next send. */
5139                 e1000_maybe_stop_tx(tx_ring, MAX_SKB_FRAGS + 2);
5140
5141         } else {
5142                 dev_kfree_skb_any(skb);
5143                 tx_ring->buffer_info[first].time_stamp = 0;
5144                 tx_ring->next_to_use = first;
5145         }
5146
5147         return NETDEV_TX_OK;
5148 }
5149
5150 /**
5151  * e1000_tx_timeout - Respond to a Tx Hang
5152  * @netdev: network interface device structure
5153  **/
5154 static void e1000_tx_timeout(struct net_device *netdev)
5155 {
5156         struct e1000_adapter *adapter = netdev_priv(netdev);
5157
5158         /* Do the reset outside of interrupt context */
5159         adapter->tx_timeout_count++;
5160         schedule_work(&adapter->reset_task);
5161 }
5162
5163 static void e1000_reset_task(struct work_struct *work)
5164 {
5165         struct e1000_adapter *adapter;
5166         adapter = container_of(work, struct e1000_adapter, reset_task);
5167
5168         /* don't run the task if already down */
5169         if (test_bit(__E1000_DOWN, &adapter->state))
5170                 return;
5171
5172         if (!((adapter->flags & FLAG_RX_NEEDS_RESTART) &&
5173               (adapter->flags & FLAG_RX_RESTART_NOW))) {
5174                 e1000e_dump(adapter);
5175                 e_err("Reset adapter\n");
5176         }
5177         e1000e_reinit_locked(adapter);
5178 }
5179
5180 /**
5181  * e1000_get_stats64 - Get System Network Statistics
5182  * @netdev: network interface device structure
5183  * @stats: rtnl_link_stats64 pointer
5184  *
5185  * Returns the address of the device statistics structure.
5186  **/
5187 struct rtnl_link_stats64 *e1000e_get_stats64(struct net_device *netdev,
5188                                              struct rtnl_link_stats64 *stats)
5189 {
5190         struct e1000_adapter *adapter = netdev_priv(netdev);
5191
5192         memset(stats, 0, sizeof(struct rtnl_link_stats64));
5193         spin_lock(&adapter->stats64_lock);
5194         e1000e_update_stats(adapter);
5195         /* Fill out the OS statistics structure */
5196         stats->rx_bytes = adapter->stats.gorc;
5197         stats->rx_packets = adapter->stats.gprc;
5198         stats->tx_bytes = adapter->stats.gotc;
5199         stats->tx_packets = adapter->stats.gptc;
5200         stats->multicast = adapter->stats.mprc;
5201         stats->collisions = adapter->stats.colc;
5202
5203         /* Rx Errors */
5204
5205         /*
5206          * RLEC on some newer hardware can be incorrect so build
5207          * our own version based on RUC and ROC
5208          */
5209         stats->rx_errors = adapter->stats.rxerrc +
5210                 adapter->stats.crcerrs + adapter->stats.algnerrc +
5211                 adapter->stats.ruc + adapter->stats.roc +
5212                 adapter->stats.cexterr;
5213         stats->rx_length_errors = adapter->stats.ruc +
5214                                               adapter->stats.roc;
5215         stats->rx_crc_errors = adapter->stats.crcerrs;
5216         stats->rx_frame_errors = adapter->stats.algnerrc;
5217         stats->rx_missed_errors = adapter->stats.mpc;
5218
5219         /* Tx Errors */
5220         stats->tx_errors = adapter->stats.ecol +
5221                                        adapter->stats.latecol;
5222         stats->tx_aborted_errors = adapter->stats.ecol;
5223         stats->tx_window_errors = adapter->stats.latecol;
5224         stats->tx_carrier_errors = adapter->stats.tncrs;
5225
5226         /* Tx Dropped needs to be maintained elsewhere */
5227
5228         spin_unlock(&adapter->stats64_lock);
5229         return stats;
5230 }
5231
5232 /**
5233  * e1000_change_mtu - Change the Maximum Transfer Unit
5234  * @netdev: network interface device structure
5235  * @new_mtu: new value for maximum frame size
5236  *
5237  * Returns 0 on success, negative on failure
5238  **/
5239 static int e1000_change_mtu(struct net_device *netdev, int new_mtu)
5240 {
5241         struct e1000_adapter *adapter = netdev_priv(netdev);
5242         int max_frame = new_mtu + ETH_HLEN + ETH_FCS_LEN;
5243
5244         /* Jumbo frame support */
5245         if (max_frame > ETH_FRAME_LEN + ETH_FCS_LEN) {
5246                 if (!(adapter->flags & FLAG_HAS_JUMBO_FRAMES)) {
5247                         e_err("Jumbo Frames not supported.\n");
5248                         return -EINVAL;
5249                 }
5250
5251                 /*
5252                  * IP payload checksum (enabled with jumbos/packet-split when
5253                  * Rx checksum is enabled) and generation of RSS hash is
5254                  * mutually exclusive in the hardware.
5255                  */
5256                 if ((netdev->features & NETIF_F_RXCSUM) &&
5257                     (netdev->features & NETIF_F_RXHASH)) {
5258                         e_err("Jumbo frames cannot be enabled when both receive checksum offload and receive hashing are enabled.  Disable one of the receive offload features before enabling jumbos.\n");
5259                         return -EINVAL;
5260                 }
5261         }
5262
5263         /* Supported frame sizes */
5264         if ((new_mtu < ETH_ZLEN + ETH_FCS_LEN + VLAN_HLEN) ||
5265             (max_frame > adapter->max_hw_frame_size)) {
5266                 e_err("Unsupported MTU setting\n");
5267                 return -EINVAL;
5268         }
5269
5270         /* Jumbo frame workaround on 82579 requires CRC be stripped */
5271         if ((adapter->hw.mac.type == e1000_pch2lan) &&
5272             !(adapter->flags2 & FLAG2_CRC_STRIPPING) &&
5273             (new_mtu > ETH_DATA_LEN)) {
5274                 e_err("Jumbo Frames not supported on 82579 when CRC stripping is disabled.\n");
5275                 return -EINVAL;
5276         }
5277
5278         /* 82573 Errata 17 */
5279         if (((adapter->hw.mac.type == e1000_82573) ||
5280              (adapter->hw.mac.type == e1000_82574)) &&
5281             (max_frame > ETH_FRAME_LEN + ETH_FCS_LEN)) {
5282                 adapter->flags2 |= FLAG2_DISABLE_ASPM_L1;
5283                 e1000e_disable_aspm(adapter->pdev, PCIE_LINK_STATE_L1);
5284         }
5285
5286         while (test_and_set_bit(__E1000_RESETTING, &adapter->state))
5287                 usleep_range(1000, 2000);
5288         /* e1000e_down -> e1000e_reset dependent on max_frame_size & mtu */
5289         adapter->max_frame_size = max_frame;
5290         e_info("changing MTU from %d to %d\n", netdev->mtu, new_mtu);
5291         netdev->mtu = new_mtu;
5292         if (netif_running(netdev))
5293                 e1000e_down(adapter);
5294
5295         /*
5296          * NOTE: netdev_alloc_skb reserves 16 bytes, and typically NET_IP_ALIGN
5297          * means we reserve 2 more, this pushes us to allocate from the next
5298          * larger slab size.
5299          * i.e. RXBUFFER_2048 --> size-4096 slab
5300          * However with the new *_jumbo_rx* routines, jumbo receives will use
5301          * fragmented skbs
5302          */
5303
5304         if (max_frame <= 2048)
5305                 adapter->rx_buffer_len = 2048;
5306         else
5307                 adapter->rx_buffer_len = 4096;
5308
5309         /* adjust allocation if LPE protects us, and we aren't using SBP */
5310         if ((max_frame == ETH_FRAME_LEN + ETH_FCS_LEN) ||
5311              (max_frame == ETH_FRAME_LEN + VLAN_HLEN + ETH_FCS_LEN))
5312                 adapter->rx_buffer_len = ETH_FRAME_LEN + VLAN_HLEN
5313                                          + ETH_FCS_LEN;
5314
5315         if (netif_running(netdev))
5316                 e1000e_up(adapter);
5317         else
5318                 e1000e_reset(adapter);
5319
5320         clear_bit(__E1000_RESETTING, &adapter->state);
5321
5322         return 0;
5323 }
5324
5325 static int e1000_mii_ioctl(struct net_device *netdev, struct ifreq *ifr,
5326                            int cmd)
5327 {
5328         struct e1000_adapter *adapter = netdev_priv(netdev);
5329         struct mii_ioctl_data *data = if_mii(ifr);
5330
5331         if (adapter->hw.phy.media_type != e1000_media_type_copper)
5332                 return -EOPNOTSUPP;
5333
5334         switch (cmd) {
5335         case SIOCGMIIPHY:
5336                 data->phy_id = adapter->hw.phy.addr;
5337                 break;
5338         case SIOCGMIIREG:
5339                 e1000_phy_read_status(adapter);
5340
5341                 switch (data->reg_num & 0x1F) {
5342                 case MII_BMCR:
5343                         data->val_out = adapter->phy_regs.bmcr;
5344                         break;
5345                 case MII_BMSR:
5346                         data->val_out = adapter->phy_regs.bmsr;
5347                         break;
5348                 case MII_PHYSID1:
5349                         data->val_out = (adapter->hw.phy.id >> 16);
5350                         break;
5351                 case MII_PHYSID2:
5352                         data->val_out = (adapter->hw.phy.id & 0xFFFF);
5353                         break;
5354                 case MII_ADVERTISE:
5355                         data->val_out = adapter->phy_regs.advertise;
5356                         break;
5357                 case MII_LPA:
5358                         data->val_out = adapter->phy_regs.lpa;
5359                         break;
5360                 case MII_EXPANSION:
5361                         data->val_out = adapter->phy_regs.expansion;
5362                         break;
5363                 case MII_CTRL1000:
5364                         data->val_out = adapter->phy_regs.ctrl1000;
5365                         break;
5366                 case MII_STAT1000:
5367                         data->val_out = adapter->phy_regs.stat1000;
5368                         break;
5369                 case MII_ESTATUS:
5370                         data->val_out = adapter->phy_regs.estatus;
5371                         break;
5372                 default:
5373                         return -EIO;
5374                 }
5375                 break;
5376         case SIOCSMIIREG:
5377         default:
5378                 return -EOPNOTSUPP;
5379         }
5380         return 0;
5381 }
5382
5383 static int e1000_ioctl(struct net_device *netdev, struct ifreq *ifr, int cmd)
5384 {
5385         switch (cmd) {
5386         case SIOCGMIIPHY:
5387         case SIOCGMIIREG:
5388         case SIOCSMIIREG:
5389                 return e1000_mii_ioctl(netdev, ifr, cmd);
5390         default:
5391                 return -EOPNOTSUPP;
5392         }
5393 }
5394
5395 static int e1000_init_phy_wakeup(struct e1000_adapter *adapter, u32 wufc)
5396 {
5397         struct e1000_hw *hw = &adapter->hw;
5398         u32 i, mac_reg;
5399         u16 phy_reg, wuc_enable;
5400         int retval = 0;
5401
5402         /* copy MAC RARs to PHY RARs */
5403         e1000_copy_rx_addrs_to_phy_ich8lan(hw);
5404
5405         retval = hw->phy.ops.acquire(hw);
5406         if (retval) {
5407                 e_err("Could not acquire PHY\n");
5408                 return retval;
5409         }
5410
5411         /* Enable access to wakeup registers on and set page to BM_WUC_PAGE */
5412         retval = e1000_enable_phy_wakeup_reg_access_bm(hw, &wuc_enable);
5413         if (retval)
5414                 goto release;
5415
5416         /* copy MAC MTA to PHY MTA - only needed for pchlan */
5417         for (i = 0; i < adapter->hw.mac.mta_reg_count; i++) {
5418                 mac_reg = E1000_READ_REG_ARRAY(hw, E1000_MTA, i);
5419                 hw->phy.ops.write_reg_page(hw, BM_MTA(i),
5420                                            (u16)(mac_reg & 0xFFFF));
5421                 hw->phy.ops.write_reg_page(hw, BM_MTA(i) + 1,
5422                                            (u16)((mac_reg >> 16) & 0xFFFF));
5423         }
5424
5425         /* configure PHY Rx Control register */
5426         hw->phy.ops.read_reg_page(&adapter->hw, BM_RCTL, &phy_reg);
5427         mac_reg = er32(RCTL);
5428         if (mac_reg & E1000_RCTL_UPE)
5429                 phy_reg |= BM_RCTL_UPE;
5430         if (mac_reg & E1000_RCTL_MPE)
5431                 phy_reg |= BM_RCTL_MPE;
5432         phy_reg &= ~(BM_RCTL_MO_MASK);
5433         if (mac_reg & E1000_RCTL_MO_3)
5434                 phy_reg |= (((mac_reg & E1000_RCTL_MO_3) >> E1000_RCTL_MO_SHIFT)
5435                                 << BM_RCTL_MO_SHIFT);
5436         if (mac_reg & E1000_RCTL_BAM)
5437                 phy_reg |= BM_RCTL_BAM;
5438         if (mac_reg & E1000_RCTL_PMCF)
5439                 phy_reg |= BM_RCTL_PMCF;
5440         mac_reg = er32(CTRL);
5441         if (mac_reg & E1000_CTRL_RFCE)
5442                 phy_reg |= BM_RCTL_RFCE;
5443         hw->phy.ops.write_reg_page(&adapter->hw, BM_RCTL, phy_reg);
5444
5445         /* enable PHY wakeup in MAC register */
5446         ew32(WUFC, wufc);
5447         ew32(WUC, E1000_WUC_PHY_WAKE | E1000_WUC_PME_EN);
5448
5449         /* configure and enable PHY wakeup in PHY registers */
5450         hw->phy.ops.write_reg_page(&adapter->hw, BM_WUFC, wufc);
5451         hw->phy.ops.write_reg_page(&adapter->hw, BM_WUC, E1000_WUC_PME_EN);
5452
5453         /* activate PHY wakeup */
5454         wuc_enable |= BM_WUC_ENABLE_BIT | BM_WUC_HOST_WU_BIT;
5455         retval = e1000_disable_phy_wakeup_reg_access_bm(hw, &wuc_enable);
5456         if (retval)
5457                 e_err("Could not set PHY Host Wakeup bit\n");
5458 release:
5459         hw->phy.ops.release(hw);
5460
5461         return retval;
5462 }
5463
5464 static int __e1000_shutdown(struct pci_dev *pdev, bool *enable_wake,
5465                             bool runtime)
5466 {
5467         struct net_device *netdev = pci_get_drvdata(pdev);
5468         struct e1000_adapter *adapter = netdev_priv(netdev);
5469         struct e1000_hw *hw = &adapter->hw;
5470         u32 ctrl, ctrl_ext, rctl, status;
5471         /* Runtime suspend should only enable wakeup for link changes */
5472         u32 wufc = runtime ? E1000_WUFC_LNKC : adapter->wol;
5473         int retval = 0;
5474
5475         netif_device_detach(netdev);
5476
5477         if (netif_running(netdev)) {
5478                 int count = E1000_CHECK_RESET_COUNT;
5479
5480                 while (test_bit(__E1000_RESETTING, &adapter->state) && count--)
5481                         usleep_range(10000, 20000);
5482
5483                 WARN_ON(test_bit(__E1000_RESETTING, &adapter->state));
5484                 e1000e_down(adapter);
5485                 e1000_free_irq(adapter);
5486         }
5487         e1000e_reset_interrupt_capability(adapter);
5488
5489         retval = pci_save_state(pdev);
5490         if (retval)
5491                 return retval;
5492
5493         status = er32(STATUS);
5494         if (status & E1000_STATUS_LU)
5495                 wufc &= ~E1000_WUFC_LNKC;
5496
5497         if (wufc) {
5498                 e1000_setup_rctl(adapter);
5499                 e1000e_set_rx_mode(netdev);
5500
5501                 /* turn on all-multi mode if wake on multicast is enabled */
5502                 if (wufc & E1000_WUFC_MC) {
5503                         rctl = er32(RCTL);
5504                         rctl |= E1000_RCTL_MPE;
5505                         ew32(RCTL, rctl);
5506                 }
5507
5508                 ctrl = er32(CTRL);
5509                 /* advertise wake from D3Cold */
5510                 #define E1000_CTRL_ADVD3WUC 0x00100000
5511                 /* phy power management enable */
5512                 #define E1000_CTRL_EN_PHY_PWR_MGMT 0x00200000
5513                 ctrl |= E1000_CTRL_ADVD3WUC;
5514                 if (!(adapter->flags2 & FLAG2_HAS_PHY_WAKEUP))
5515                         ctrl |= E1000_CTRL_EN_PHY_PWR_MGMT;
5516                 ew32(CTRL, ctrl);
5517
5518                 if (adapter->hw.phy.media_type == e1000_media_type_fiber ||
5519                     adapter->hw.phy.media_type ==
5520                     e1000_media_type_internal_serdes) {
5521                         /* keep the laser running in D3 */
5522                         ctrl_ext = er32(CTRL_EXT);
5523                         ctrl_ext |= E1000_CTRL_EXT_SDP3_DATA;
5524                         ew32(CTRL_EXT, ctrl_ext);
5525                 }
5526
5527                 if (adapter->flags & FLAG_IS_ICH)
5528                         e1000_suspend_workarounds_ich8lan(&adapter->hw);
5529
5530                 /* Allow time for pending master requests to run */
5531                 e1000e_disable_pcie_master(&adapter->hw);
5532
5533                 if (adapter->flags2 & FLAG2_HAS_PHY_WAKEUP) {
5534                         /* enable wakeup by the PHY */
5535                         retval = e1000_init_phy_wakeup(adapter, wufc);
5536                         if (retval)
5537                                 return retval;
5538                 } else {
5539                         /* enable wakeup by the MAC */
5540                         ew32(WUFC, wufc);
5541                         ew32(WUC, E1000_WUC_PME_EN);
5542                 }
5543         } else {
5544                 ew32(WUC, 0);
5545                 ew32(WUFC, 0);
5546         }
5547
5548         *enable_wake = !!wufc;
5549
5550         /* make sure adapter isn't asleep if manageability is enabled */
5551         if ((adapter->flags & FLAG_MNG_PT_ENABLED) ||
5552             (hw->mac.ops.check_mng_mode(hw)))
5553                 *enable_wake = true;
5554
5555         if (adapter->hw.phy.type == e1000_phy_igp_3)
5556                 e1000e_igp3_phy_powerdown_workaround_ich8lan(&adapter->hw);
5557
5558         /*
5559          * Release control of h/w to f/w.  If f/w is AMT enabled, this
5560          * would have already happened in close and is redundant.
5561          */
5562         e1000e_release_hw_control(adapter);
5563
5564         pci_disable_device(pdev);
5565
5566         return 0;
5567 }
5568
5569 static void e1000_power_off(struct pci_dev *pdev, bool sleep, bool wake)
5570 {
5571         if (sleep && wake) {
5572                 pci_prepare_to_sleep(pdev);
5573                 return;
5574         }
5575
5576         pci_wake_from_d3(pdev, wake);
5577         pci_set_power_state(pdev, PCI_D3hot);
5578 }
5579
5580 static void e1000_complete_shutdown(struct pci_dev *pdev, bool sleep,
5581                                     bool wake)
5582 {
5583         struct net_device *netdev = pci_get_drvdata(pdev);
5584         struct e1000_adapter *adapter = netdev_priv(netdev);
5585
5586         /*
5587          * The pci-e switch on some quad port adapters will report a
5588          * correctable error when the MAC transitions from D0 to D3.  To
5589          * prevent this we need to mask off the correctable errors on the
5590          * downstream port of the pci-e switch.
5591          */
5592         if (adapter->flags & FLAG_IS_QUAD_PORT) {
5593                 struct pci_dev *us_dev = pdev->bus->self;
5594                 int pos = pci_pcie_cap(us_dev);
5595                 u16 devctl;
5596
5597                 pci_read_config_word(us_dev, pos + PCI_EXP_DEVCTL, &devctl);
5598                 pci_write_config_word(us_dev, pos + PCI_EXP_DEVCTL,
5599                                       (devctl & ~PCI_EXP_DEVCTL_CERE));
5600
5601                 e1000_power_off(pdev, sleep, wake);
5602
5603                 pci_write_config_word(us_dev, pos + PCI_EXP_DEVCTL, devctl);
5604         } else {
5605                 e1000_power_off(pdev, sleep, wake);
5606         }
5607 }
5608
5609 #ifdef CONFIG_PCIEASPM
5610 static void __e1000e_disable_aspm(struct pci_dev *pdev, u16 state)
5611 {
5612         pci_disable_link_state_locked(pdev, state);
5613 }
5614 #else
5615 static void __e1000e_disable_aspm(struct pci_dev *pdev, u16 state)
5616 {
5617         int pos;
5618         u16 reg16;
5619
5620         /*
5621          * Both device and parent should have the same ASPM setting.
5622          * Disable ASPM in downstream component first and then upstream.
5623          */
5624         pos = pci_pcie_cap(pdev);
5625         pci_read_config_word(pdev, pos + PCI_EXP_LNKCTL, &reg16);
5626         reg16 &= ~state;
5627         pci_write_config_word(pdev, pos + PCI_EXP_LNKCTL, reg16);
5628
5629         if (!pdev->bus->self)
5630                 return;
5631
5632         pos = pci_pcie_cap(pdev->bus->self);
5633         pci_read_config_word(pdev->bus->self, pos + PCI_EXP_LNKCTL, &reg16);
5634         reg16 &= ~state;
5635         pci_write_config_word(pdev->bus->self, pos + PCI_EXP_LNKCTL, reg16);
5636 }
5637 #endif
5638 static void e1000e_disable_aspm(struct pci_dev *pdev, u16 state)
5639 {
5640         dev_info(&pdev->dev, "Disabling ASPM %s %s\n",
5641                  (state & PCIE_LINK_STATE_L0S) ? "L0s" : "",
5642                  (state & PCIE_LINK_STATE_L1) ? "L1" : "");
5643
5644         __e1000e_disable_aspm(pdev, state);
5645 }
5646
5647 #ifdef CONFIG_PM
5648 static bool e1000e_pm_ready(struct e1000_adapter *adapter)
5649 {
5650         return !!adapter->tx_ring->buffer_info;
5651 }
5652
5653 static int __e1000_resume(struct pci_dev *pdev)
5654 {
5655         struct net_device *netdev = pci_get_drvdata(pdev);
5656         struct e1000_adapter *adapter = netdev_priv(netdev);
5657         struct e1000_hw *hw = &adapter->hw;
5658         u16 aspm_disable_flag = 0;
5659         u32 err;
5660
5661         if (adapter->flags2 & FLAG2_DISABLE_ASPM_L0S)
5662                 aspm_disable_flag = PCIE_LINK_STATE_L0S;
5663         if (adapter->flags2 & FLAG2_DISABLE_ASPM_L1)
5664                 aspm_disable_flag |= PCIE_LINK_STATE_L1;
5665         if (aspm_disable_flag)
5666                 e1000e_disable_aspm(pdev, aspm_disable_flag);
5667
5668         pci_set_power_state(pdev, PCI_D0);
5669         pci_restore_state(pdev);
5670         pci_save_state(pdev);
5671
5672         e1000e_set_interrupt_capability(adapter);
5673         if (netif_running(netdev)) {
5674                 err = e1000_request_irq(adapter);
5675                 if (err)
5676                         return err;
5677         }
5678
5679         if (hw->mac.type == e1000_pch2lan)
5680                 e1000_resume_workarounds_pchlan(&adapter->hw);
5681
5682         e1000e_power_up_phy(adapter);
5683
5684         /* report the system wakeup cause from S3/S4 */
5685         if (adapter->flags2 & FLAG2_HAS_PHY_WAKEUP) {
5686                 u16 phy_data;
5687
5688                 e1e_rphy(&adapter->hw, BM_WUS, &phy_data);
5689                 if (phy_data) {
5690                         e_info("PHY Wakeup cause - %s\n",
5691                                 phy_data & E1000_WUS_EX ? "Unicast Packet" :
5692                                 phy_data & E1000_WUS_MC ? "Multicast Packet" :
5693                                 phy_data & E1000_WUS_BC ? "Broadcast Packet" :
5694                                 phy_data & E1000_WUS_MAG ? "Magic Packet" :
5695                                 phy_data & E1000_WUS_LNKC ?
5696                                 "Link Status Change" : "other");
5697                 }
5698                 e1e_wphy(&adapter->hw, BM_WUS, ~0);
5699         } else {
5700                 u32 wus = er32(WUS);
5701                 if (wus) {
5702                         e_info("MAC Wakeup cause - %s\n",
5703                                 wus & E1000_WUS_EX ? "Unicast Packet" :
5704                                 wus & E1000_WUS_MC ? "Multicast Packet" :
5705                                 wus & E1000_WUS_BC ? "Broadcast Packet" :
5706                                 wus & E1000_WUS_MAG ? "Magic Packet" :
5707                                 wus & E1000_WUS_LNKC ? "Link Status Change" :
5708                                 "other");
5709                 }
5710                 ew32(WUS, ~0);
5711         }
5712
5713         e1000e_reset(adapter);
5714
5715         e1000_init_manageability_pt(adapter);
5716
5717         if (netif_running(netdev))
5718                 e1000e_up(adapter);
5719
5720         netif_device_attach(netdev);
5721
5722         /*
5723          * If the controller has AMT, do not set DRV_LOAD until the interface
5724          * is up.  For all other cases, let the f/w know that the h/w is now
5725          * under the control of the driver.
5726          */
5727         if (!(adapter->flags & FLAG_HAS_AMT))
5728                 e1000e_get_hw_control(adapter);
5729
5730         return 0;
5731 }
5732
5733 #ifdef CONFIG_PM_SLEEP
5734 static int e1000_suspend(struct device *dev)
5735 {
5736         struct pci_dev *pdev = to_pci_dev(dev);
5737         int retval;
5738         bool wake;
5739
5740         retval = __e1000_shutdown(pdev, &wake, false);
5741         if (!retval)
5742                 e1000_complete_shutdown(pdev, true, wake);
5743
5744         return retval;
5745 }
5746
5747 static int e1000_resume(struct device *dev)
5748 {
5749         struct pci_dev *pdev = to_pci_dev(dev);
5750         struct net_device *netdev = pci_get_drvdata(pdev);
5751         struct e1000_adapter *adapter = netdev_priv(netdev);
5752
5753         if (e1000e_pm_ready(adapter))
5754                 adapter->idle_check = true;
5755
5756         return __e1000_resume(pdev);
5757 }
5758 #endif /* CONFIG_PM_SLEEP */
5759
5760 #ifdef CONFIG_PM_RUNTIME
5761 static int e1000_runtime_suspend(struct device *dev)
5762 {
5763         struct pci_dev *pdev = to_pci_dev(dev);
5764         struct net_device *netdev = pci_get_drvdata(pdev);
5765         struct e1000_adapter *adapter = netdev_priv(netdev);
5766
5767         if (e1000e_pm_ready(adapter)) {
5768                 bool wake;
5769
5770                 __e1000_shutdown(pdev, &wake, true);
5771         }
5772
5773         return 0;
5774 }
5775
5776 static int e1000_idle(struct device *dev)
5777 {
5778         struct pci_dev *pdev = to_pci_dev(dev);
5779         struct net_device *netdev = pci_get_drvdata(pdev);
5780         struct e1000_adapter *adapter = netdev_priv(netdev);
5781
5782         if (!e1000e_pm_ready(adapter))
5783                 return 0;
5784
5785         if (adapter->idle_check) {
5786                 adapter->idle_check = false;
5787                 if (!e1000e_has_link(adapter))
5788                         pm_schedule_suspend(dev, MSEC_PER_SEC);
5789         }
5790
5791         return -EBUSY;
5792 }
5793
5794 static int e1000_runtime_resume(struct device *dev)
5795 {
5796         struct pci_dev *pdev = to_pci_dev(dev);
5797         struct net_device *netdev = pci_get_drvdata(pdev);
5798         struct e1000_adapter *adapter = netdev_priv(netdev);
5799
5800         if (!e1000e_pm_ready(adapter))
5801                 return 0;
5802
5803         adapter->idle_check = !dev->power.runtime_auto;
5804         return __e1000_resume(pdev);
5805 }
5806 #endif /* CONFIG_PM_RUNTIME */
5807 #endif /* CONFIG_PM */
5808
5809 static void e1000_shutdown(struct pci_dev *pdev)
5810 {
5811         bool wake = false;
5812
5813         __e1000_shutdown(pdev, &wake, false);
5814
5815         if (system_state == SYSTEM_POWER_OFF)
5816                 e1000_complete_shutdown(pdev, false, wake);
5817 }
5818
5819 #ifdef CONFIG_NET_POLL_CONTROLLER
5820
5821 static irqreturn_t e1000_intr_msix(int irq, void *data)
5822 {
5823         struct net_device *netdev = data;
5824         struct e1000_adapter *adapter = netdev_priv(netdev);
5825
5826         if (adapter->msix_entries) {
5827                 int vector, msix_irq;
5828
5829                 vector = 0;
5830                 msix_irq = adapter->msix_entries[vector].vector;
5831                 disable_irq(msix_irq);
5832                 e1000_intr_msix_rx(msix_irq, netdev);
5833                 enable_irq(msix_irq);
5834
5835                 vector++;
5836                 msix_irq = adapter->msix_entries[vector].vector;
5837                 disable_irq(msix_irq);
5838                 e1000_intr_msix_tx(msix_irq, netdev);
5839                 enable_irq(msix_irq);
5840
5841                 vector++;
5842                 msix_irq = adapter->msix_entries[vector].vector;
5843                 disable_irq(msix_irq);
5844                 e1000_msix_other(msix_irq, netdev);
5845                 enable_irq(msix_irq);
5846         }
5847
5848         return IRQ_HANDLED;
5849 }
5850
5851 /*
5852  * Polling 'interrupt' - used by things like netconsole to send skbs
5853  * without having to re-enable interrupts. It's not called while
5854  * the interrupt routine is executing.
5855  */
5856 static void e1000_netpoll(struct net_device *netdev)
5857 {
5858         struct e1000_adapter *adapter = netdev_priv(netdev);
5859
5860         switch (adapter->int_mode) {
5861         case E1000E_INT_MODE_MSIX:
5862                 e1000_intr_msix(adapter->pdev->irq, netdev);
5863                 break;
5864         case E1000E_INT_MODE_MSI:
5865                 disable_irq(adapter->pdev->irq);
5866                 e1000_intr_msi(adapter->pdev->irq, netdev);
5867                 enable_irq(adapter->pdev->irq);
5868                 break;
5869         default: /* E1000E_INT_MODE_LEGACY */
5870                 disable_irq(adapter->pdev->irq);
5871                 e1000_intr(adapter->pdev->irq, netdev);
5872                 enable_irq(adapter->pdev->irq);
5873                 break;
5874         }
5875 }
5876 #endif
5877
5878 /**
5879  * e1000_io_error_detected - called when PCI error is detected
5880  * @pdev: Pointer to PCI device
5881  * @state: The current pci connection state
5882  *
5883  * This function is called after a PCI bus error affecting
5884  * this device has been detected.
5885  */
5886 static pci_ers_result_t e1000_io_error_detected(struct pci_dev *pdev,
5887                                                 pci_channel_state_t state)
5888 {
5889         struct net_device *netdev = pci_get_drvdata(pdev);
5890         struct e1000_adapter *adapter = netdev_priv(netdev);
5891
5892         netif_device_detach(netdev);
5893
5894         if (state == pci_channel_io_perm_failure)
5895                 return PCI_ERS_RESULT_DISCONNECT;
5896
5897         if (netif_running(netdev))
5898                 e1000e_down(adapter);
5899         pci_disable_device(pdev);
5900
5901         /* Request a slot slot reset. */
5902         return PCI_ERS_RESULT_NEED_RESET;
5903 }
5904
5905 /**
5906  * e1000_io_slot_reset - called after the pci bus has been reset.
5907  * @pdev: Pointer to PCI device
5908  *
5909  * Restart the card from scratch, as if from a cold-boot. Implementation
5910  * resembles the first-half of the e1000_resume routine.
5911  */
5912 static pci_ers_result_t e1000_io_slot_reset(struct pci_dev *pdev)
5913 {
5914         struct net_device *netdev = pci_get_drvdata(pdev);
5915         struct e1000_adapter *adapter = netdev_priv(netdev);
5916         struct e1000_hw *hw = &adapter->hw;
5917         u16 aspm_disable_flag = 0;
5918         int err;
5919         pci_ers_result_t result;
5920
5921         if (adapter->flags2 & FLAG2_DISABLE_ASPM_L0S)
5922                 aspm_disable_flag = PCIE_LINK_STATE_L0S;
5923         if (adapter->flags2 & FLAG2_DISABLE_ASPM_L1)
5924                 aspm_disable_flag |= PCIE_LINK_STATE_L1;
5925         if (aspm_disable_flag)
5926                 e1000e_disable_aspm(pdev, aspm_disable_flag);
5927
5928         err = pci_enable_device_mem(pdev);
5929         if (err) {
5930                 dev_err(&pdev->dev,
5931                         "Cannot re-enable PCI device after reset.\n");
5932                 result = PCI_ERS_RESULT_DISCONNECT;
5933         } else {
5934                 pci_set_master(pdev);
5935                 pdev->state_saved = true;
5936                 pci_restore_state(pdev);
5937
5938                 pci_enable_wake(pdev, PCI_D3hot, 0);
5939                 pci_enable_wake(pdev, PCI_D3cold, 0);
5940
5941                 e1000e_reset(adapter);
5942                 ew32(WUS, ~0);
5943                 result = PCI_ERS_RESULT_RECOVERED;
5944         }
5945
5946         pci_cleanup_aer_uncorrect_error_status(pdev);
5947
5948         return result;
5949 }
5950
5951 /**
5952  * e1000_io_resume - called when traffic can start flowing again.
5953  * @pdev: Pointer to PCI device
5954  *
5955  * This callback is called when the error recovery driver tells us that
5956  * its OK to resume normal operation. Implementation resembles the
5957  * second-half of the e1000_resume routine.
5958  */
5959 static void e1000_io_resume(struct pci_dev *pdev)
5960 {
5961         struct net_device *netdev = pci_get_drvdata(pdev);
5962         struct e1000_adapter *adapter = netdev_priv(netdev);
5963
5964         e1000_init_manageability_pt(adapter);
5965
5966         if (netif_running(netdev)) {
5967                 if (e1000e_up(adapter)) {
5968                         dev_err(&pdev->dev,
5969                                 "can't bring device back up after reset\n");
5970                         return;
5971                 }
5972         }
5973
5974         netif_device_attach(netdev);
5975
5976         /*
5977          * If the controller has AMT, do not set DRV_LOAD until the interface
5978          * is up.  For all other cases, let the f/w know that the h/w is now
5979          * under the control of the driver.
5980          */
5981         if (!(adapter->flags & FLAG_HAS_AMT))
5982                 e1000e_get_hw_control(adapter);
5983
5984 }
5985
5986 static void e1000_print_device_info(struct e1000_adapter *adapter)
5987 {
5988         struct e1000_hw *hw = &adapter->hw;
5989         struct net_device *netdev = adapter->netdev;
5990         u32 ret_val;
5991         u8 pba_str[E1000_PBANUM_LENGTH];
5992
5993         /* print bus type/speed/width info */
5994         e_info("(PCI Express:2.5GT/s:%s) %pM\n",
5995                /* bus width */
5996                ((hw->bus.width == e1000_bus_width_pcie_x4) ? "Width x4" :
5997                 "Width x1"),
5998                /* MAC address */
5999                netdev->dev_addr);
6000         e_info("Intel(R) PRO/%s Network Connection\n",
6001                (hw->phy.type == e1000_phy_ife) ? "10/100" : "1000");
6002         ret_val = e1000_read_pba_string_generic(hw, pba_str,
6003                                                 E1000_PBANUM_LENGTH);
6004         if (ret_val)
6005                 strlcpy((char *)pba_str, "Unknown", sizeof(pba_str));
6006         e_info("MAC: %d, PHY: %d, PBA No: %s\n",
6007                hw->mac.type, hw->phy.type, pba_str);
6008 }
6009
6010 static void e1000_eeprom_checks(struct e1000_adapter *adapter)
6011 {
6012         struct e1000_hw *hw = &adapter->hw;
6013         int ret_val;
6014         u16 buf = 0;
6015
6016         if (hw->mac.type != e1000_82573)
6017                 return;
6018
6019         ret_val = e1000_read_nvm(hw, NVM_INIT_CONTROL2_REG, 1, &buf);
6020         le16_to_cpus(&buf);
6021         if (!ret_val && (!(buf & (1 << 0)))) {
6022                 /* Deep Smart Power Down (DSPD) */
6023                 dev_warn(&adapter->pdev->dev,
6024                          "Warning: detected DSPD enabled in EEPROM\n");
6025         }
6026 }
6027
6028 static int e1000_set_features(struct net_device *netdev,
6029                               netdev_features_t features)
6030 {
6031         struct e1000_adapter *adapter = netdev_priv(netdev);
6032         netdev_features_t changed = features ^ netdev->features;
6033
6034         if (changed & (NETIF_F_TSO | NETIF_F_TSO6))
6035                 adapter->flags |= FLAG_TSO_FORCE;
6036
6037         if (!(changed & (NETIF_F_HW_VLAN_RX | NETIF_F_HW_VLAN_TX |
6038                          NETIF_F_RXCSUM | NETIF_F_RXHASH | NETIF_F_RXFCS |
6039                          NETIF_F_RXALL)))
6040                 return 0;
6041
6042         /*
6043          * IP payload checksum (enabled with jumbos/packet-split when Rx
6044          * checksum is enabled) and generation of RSS hash is mutually
6045          * exclusive in the hardware.
6046          */
6047         if (adapter->rx_ps_pages &&
6048             (features & NETIF_F_RXCSUM) && (features & NETIF_F_RXHASH)) {
6049                 e_err("Enabling both receive checksum offload and receive hashing is not possible with jumbo frames.  Disable jumbos or enable only one of the receive offload features.\n");
6050                 return -EINVAL;
6051         }
6052
6053         if (changed & NETIF_F_RXFCS) {
6054                 if (features & NETIF_F_RXFCS) {
6055                         adapter->flags2 &= ~FLAG2_CRC_STRIPPING;
6056                 } else {
6057                         /* We need to take it back to defaults, which might mean
6058                          * stripping is still disabled at the adapter level.
6059                          */
6060                         if (adapter->flags2 & FLAG2_DFLT_CRC_STRIPPING)
6061                                 adapter->flags2 |= FLAG2_CRC_STRIPPING;
6062                         else
6063                                 adapter->flags2 &= ~FLAG2_CRC_STRIPPING;
6064                 }
6065         }
6066
6067         netdev->features = features;
6068
6069         if (netif_running(netdev))
6070                 e1000e_reinit_locked(adapter);
6071         else
6072                 e1000e_reset(adapter);
6073
6074         return 0;
6075 }
6076
6077 static const struct net_device_ops e1000e_netdev_ops = {
6078         .ndo_open               = e1000_open,
6079         .ndo_stop               = e1000_close,
6080         .ndo_start_xmit         = e1000_xmit_frame,
6081         .ndo_get_stats64        = e1000e_get_stats64,
6082         .ndo_set_rx_mode        = e1000e_set_rx_mode,
6083         .ndo_set_mac_address    = e1000_set_mac,
6084         .ndo_change_mtu         = e1000_change_mtu,
6085         .ndo_do_ioctl           = e1000_ioctl,
6086         .ndo_tx_timeout         = e1000_tx_timeout,
6087         .ndo_validate_addr      = eth_validate_addr,
6088
6089         .ndo_vlan_rx_add_vid    = e1000_vlan_rx_add_vid,
6090         .ndo_vlan_rx_kill_vid   = e1000_vlan_rx_kill_vid,
6091 #ifdef CONFIG_NET_POLL_CONTROLLER
6092         .ndo_poll_controller    = e1000_netpoll,
6093 #endif
6094         .ndo_set_features = e1000_set_features,
6095 };
6096
6097 /**
6098  * e1000_probe - Device Initialization Routine
6099  * @pdev: PCI device information struct
6100  * @ent: entry in e1000_pci_tbl
6101  *
6102  * Returns 0 on success, negative on failure
6103  *
6104  * e1000_probe initializes an adapter identified by a pci_dev structure.
6105  * The OS initialization, configuring of the adapter private structure,
6106  * and a hardware reset occur.
6107  **/
6108 static int __devinit e1000_probe(struct pci_dev *pdev,
6109                                  const struct pci_device_id *ent)
6110 {
6111         struct net_device *netdev;
6112         struct e1000_adapter *adapter;
6113         struct e1000_hw *hw;
6114         const struct e1000_info *ei = e1000_info_tbl[ent->driver_data];
6115         resource_size_t mmio_start, mmio_len;
6116         resource_size_t flash_start, flash_len;
6117         static int cards_found;
6118         u16 aspm_disable_flag = 0;
6119         int i, err, pci_using_dac;
6120         u16 eeprom_data = 0;
6121         u16 eeprom_apme_mask = E1000_EEPROM_APME;
6122
6123         if (ei->flags2 & FLAG2_DISABLE_ASPM_L0S)
6124                 aspm_disable_flag = PCIE_LINK_STATE_L0S;
6125         if (ei->flags2 & FLAG2_DISABLE_ASPM_L1)
6126                 aspm_disable_flag |= PCIE_LINK_STATE_L1;
6127         if (aspm_disable_flag)
6128                 e1000e_disable_aspm(pdev, aspm_disable_flag);
6129
6130         err = pci_enable_device_mem(pdev);
6131         if (err)
6132                 return err;
6133
6134         pci_using_dac = 0;
6135         err = dma_set_mask(&pdev->dev, DMA_BIT_MASK(64));
6136         if (!err) {
6137                 err = dma_set_coherent_mask(&pdev->dev, DMA_BIT_MASK(64));
6138                 if (!err)
6139                         pci_using_dac = 1;
6140         } else {
6141                 err = dma_set_mask(&pdev->dev, DMA_BIT_MASK(32));
6142                 if (err) {
6143                         err = dma_set_coherent_mask(&pdev->dev,
6144                                                     DMA_BIT_MASK(32));
6145                         if (err) {
6146                                 dev_err(&pdev->dev, "No usable DMA configuration, aborting\n");
6147                                 goto err_dma;
6148                         }
6149                 }
6150         }
6151
6152         err = pci_request_selected_regions_exclusive(pdev,
6153                                           pci_select_bars(pdev, IORESOURCE_MEM),
6154                                           e1000e_driver_name);
6155         if (err)
6156                 goto err_pci_reg;
6157
6158         /* AER (Advanced Error Reporting) hooks */
6159         pci_enable_pcie_error_reporting(pdev);
6160
6161         pci_set_master(pdev);
6162         /* PCI config space info */
6163         err = pci_save_state(pdev);
6164         if (err)
6165                 goto err_alloc_etherdev;
6166
6167         err = -ENOMEM;
6168         netdev = alloc_etherdev(sizeof(struct e1000_adapter));
6169         if (!netdev)
6170                 goto err_alloc_etherdev;
6171
6172         SET_NETDEV_DEV(netdev, &pdev->dev);
6173
6174         netdev->irq = pdev->irq;
6175
6176         pci_set_drvdata(pdev, netdev);
6177         adapter = netdev_priv(netdev);
6178         hw = &adapter->hw;
6179         adapter->netdev = netdev;
6180         adapter->pdev = pdev;
6181         adapter->ei = ei;
6182         adapter->pba = ei->pba;
6183         adapter->flags = ei->flags;
6184         adapter->flags2 = ei->flags2;
6185         adapter->hw.adapter = adapter;
6186         adapter->hw.mac.type = ei->mac;
6187         adapter->max_hw_frame_size = ei->max_hw_frame_size;
6188         adapter->msg_enable = netif_msg_init(debug, DEFAULT_MSG_ENABLE);
6189
6190         mmio_start = pci_resource_start(pdev, 0);
6191         mmio_len = pci_resource_len(pdev, 0);
6192
6193         err = -EIO;
6194         adapter->hw.hw_addr = ioremap(mmio_start, mmio_len);
6195         if (!adapter->hw.hw_addr)
6196                 goto err_ioremap;
6197
6198         if ((adapter->flags & FLAG_HAS_FLASH) &&
6199             (pci_resource_flags(pdev, 1) & IORESOURCE_MEM)) {
6200                 flash_start = pci_resource_start(pdev, 1);
6201                 flash_len = pci_resource_len(pdev, 1);
6202                 adapter->hw.flash_address = ioremap(flash_start, flash_len);
6203                 if (!adapter->hw.flash_address)
6204                         goto err_flashmap;
6205         }
6206
6207         /* construct the net_device struct */
6208         netdev->netdev_ops              = &e1000e_netdev_ops;
6209         e1000e_set_ethtool_ops(netdev);
6210         netdev->watchdog_timeo          = 5 * HZ;
6211         netif_napi_add(netdev, &adapter->napi, e1000e_poll, 64);
6212         strlcpy(netdev->name, pci_name(pdev), sizeof(netdev->name));
6213
6214         netdev->mem_start = mmio_start;
6215         netdev->mem_end = mmio_start + mmio_len;
6216
6217         adapter->bd_number = cards_found++;
6218
6219         e1000e_check_options(adapter);
6220
6221         /* setup adapter struct */
6222         err = e1000_sw_init(adapter);
6223         if (err)
6224                 goto err_sw_init;
6225
6226         memcpy(&hw->mac.ops, ei->mac_ops, sizeof(hw->mac.ops));
6227         memcpy(&hw->nvm.ops, ei->nvm_ops, sizeof(hw->nvm.ops));
6228         memcpy(&hw->phy.ops, ei->phy_ops, sizeof(hw->phy.ops));
6229
6230         err = ei->get_variants(adapter);
6231         if (err)
6232                 goto err_hw_init;
6233
6234         if ((adapter->flags & FLAG_IS_ICH) &&
6235             (adapter->flags & FLAG_READ_ONLY_NVM))
6236                 e1000e_write_protect_nvm_ich8lan(&adapter->hw);
6237
6238         hw->mac.ops.get_bus_info(&adapter->hw);
6239
6240         adapter->hw.phy.autoneg_wait_to_complete = 0;
6241
6242         /* Copper options */
6243         if (adapter->hw.phy.media_type == e1000_media_type_copper) {
6244                 adapter->hw.phy.mdix = AUTO_ALL_MODES;
6245                 adapter->hw.phy.disable_polarity_correction = 0;
6246                 adapter->hw.phy.ms_type = e1000_ms_hw_default;
6247         }
6248
6249         if (hw->phy.ops.check_reset_block(hw))
6250                 e_info("PHY reset is blocked due to SOL/IDER session.\n");
6251
6252         /* Set initial default active device features */
6253         netdev->features = (NETIF_F_SG |
6254                             NETIF_F_HW_VLAN_RX |
6255                             NETIF_F_HW_VLAN_TX |
6256                             NETIF_F_TSO |
6257                             NETIF_F_TSO6 |
6258                             NETIF_F_RXHASH |
6259                             NETIF_F_RXCSUM |
6260                             NETIF_F_HW_CSUM);
6261
6262         /* Set user-changeable features (subset of all device features) */
6263         netdev->hw_features = netdev->features;
6264         netdev->hw_features |= NETIF_F_RXFCS;
6265         netdev->priv_flags |= IFF_SUPP_NOFCS;
6266         netdev->hw_features |= NETIF_F_RXALL;
6267
6268         if (adapter->flags & FLAG_HAS_HW_VLAN_FILTER)
6269                 netdev->features |= NETIF_F_HW_VLAN_FILTER;
6270
6271         netdev->vlan_features |= (NETIF_F_SG |
6272                                   NETIF_F_TSO |
6273                                   NETIF_F_TSO6 |
6274                                   NETIF_F_HW_CSUM);
6275
6276         netdev->priv_flags |= IFF_UNICAST_FLT;
6277
6278         if (pci_using_dac) {
6279                 netdev->features |= NETIF_F_HIGHDMA;
6280                 netdev->vlan_features |= NETIF_F_HIGHDMA;
6281         }
6282
6283         if (e1000e_enable_mng_pass_thru(&adapter->hw))
6284                 adapter->flags |= FLAG_MNG_PT_ENABLED;
6285
6286         /*
6287          * before reading the NVM, reset the controller to
6288          * put the device in a known good starting state
6289          */
6290         adapter->hw.mac.ops.reset_hw(&adapter->hw);
6291
6292         /*
6293          * systems with ASPM and others may see the checksum fail on the first
6294          * attempt. Let's give it a few tries
6295          */
6296         for (i = 0;; i++) {
6297                 if (e1000_validate_nvm_checksum(&adapter->hw) >= 0)
6298                         break;
6299                 if (i == 2) {
6300                         e_err("The NVM Checksum Is Not Valid\n");
6301                         err = -EIO;
6302                         goto err_eeprom;
6303                 }
6304         }
6305
6306         e1000_eeprom_checks(adapter);
6307
6308         /* copy the MAC address */
6309         if (e1000e_read_mac_addr(&adapter->hw))
6310                 e_err("NVM Read Error while reading MAC address\n");
6311
6312         memcpy(netdev->dev_addr, adapter->hw.mac.addr, netdev->addr_len);
6313         memcpy(netdev->perm_addr, adapter->hw.mac.addr, netdev->addr_len);
6314
6315         if (!is_valid_ether_addr(netdev->perm_addr)) {
6316                 e_err("Invalid MAC Address: %pM\n", netdev->perm_addr);
6317                 err = -EIO;
6318                 goto err_eeprom;
6319         }
6320
6321         init_timer(&adapter->watchdog_timer);
6322         adapter->watchdog_timer.function = e1000_watchdog;
6323         adapter->watchdog_timer.data = (unsigned long) adapter;
6324
6325         init_timer(&adapter->phy_info_timer);
6326         adapter->phy_info_timer.function = e1000_update_phy_info;
6327         adapter->phy_info_timer.data = (unsigned long) adapter;
6328
6329         INIT_WORK(&adapter->reset_task, e1000_reset_task);
6330         INIT_WORK(&adapter->watchdog_task, e1000_watchdog_task);
6331         INIT_WORK(&adapter->downshift_task, e1000e_downshift_workaround);
6332         INIT_WORK(&adapter->update_phy_task, e1000e_update_phy_task);
6333         INIT_WORK(&adapter->print_hang_task, e1000_print_hw_hang);
6334
6335         /* Initialize link parameters. User can change them with ethtool */
6336         adapter->hw.mac.autoneg = 1;
6337         adapter->fc_autoneg = true;
6338         adapter->hw.fc.requested_mode = e1000_fc_default;
6339         adapter->hw.fc.current_mode = e1000_fc_default;
6340         adapter->hw.phy.autoneg_advertised = 0x2f;
6341
6342         /* ring size defaults */
6343         adapter->rx_ring->count = 256;
6344         adapter->tx_ring->count = 256;
6345
6346         /*
6347          * Initial Wake on LAN setting - If APM wake is enabled in
6348          * the EEPROM, enable the ACPI Magic Packet filter
6349          */
6350         if (adapter->flags & FLAG_APME_IN_WUC) {
6351                 /* APME bit in EEPROM is mapped to WUC.APME */
6352                 eeprom_data = er32(WUC);
6353                 eeprom_apme_mask = E1000_WUC_APME;
6354                 if ((hw->mac.type > e1000_ich10lan) &&
6355                     (eeprom_data & E1000_WUC_PHY_WAKE))
6356                         adapter->flags2 |= FLAG2_HAS_PHY_WAKEUP;
6357         } else if (adapter->flags & FLAG_APME_IN_CTRL3) {
6358                 if (adapter->flags & FLAG_APME_CHECK_PORT_B &&
6359                     (adapter->hw.bus.func == 1))
6360                         e1000_read_nvm(&adapter->hw, NVM_INIT_CONTROL3_PORT_B,
6361                                        1, &eeprom_data);
6362                 else
6363                         e1000_read_nvm(&adapter->hw, NVM_INIT_CONTROL3_PORT_A,
6364                                        1, &eeprom_data);
6365         }
6366
6367         /* fetch WoL from EEPROM */
6368         if (eeprom_data & eeprom_apme_mask)
6369                 adapter->eeprom_wol |= E1000_WUFC_MAG;
6370
6371         /*
6372          * now that we have the eeprom settings, apply the special cases
6373          * where the eeprom may be wrong or the board simply won't support
6374          * wake on lan on a particular port
6375          */
6376         if (!(adapter->flags & FLAG_HAS_WOL))
6377                 adapter->eeprom_wol = 0;
6378
6379         /* initialize the wol settings based on the eeprom settings */
6380         adapter->wol = adapter->eeprom_wol;
6381         device_set_wakeup_enable(&adapter->pdev->dev, adapter->wol);
6382
6383         /* save off EEPROM version number */
6384         e1000_read_nvm(&adapter->hw, 5, 1, &adapter->eeprom_vers);
6385
6386         /* reset the hardware with the new settings */
6387         e1000e_reset(adapter);
6388
6389         /*
6390          * If the controller has AMT, do not set DRV_LOAD until the interface
6391          * is up.  For all other cases, let the f/w know that the h/w is now
6392          * under the control of the driver.
6393          */
6394         if (!(adapter->flags & FLAG_HAS_AMT))
6395                 e1000e_get_hw_control(adapter);
6396
6397         strlcpy(netdev->name, "eth%d", sizeof(netdev->name));
6398         err = register_netdev(netdev);
6399         if (err)
6400                 goto err_register;
6401
6402         /* carrier off reporting is important to ethtool even BEFORE open */
6403         netif_carrier_off(netdev);
6404
6405         e1000_print_device_info(adapter);
6406
6407         if (pci_dev_run_wake(pdev))
6408                 pm_runtime_put_noidle(&pdev->dev);
6409
6410         return 0;
6411
6412 err_register:
6413         if (!(adapter->flags & FLAG_HAS_AMT))
6414                 e1000e_release_hw_control(adapter);
6415 err_eeprom:
6416         if (!hw->phy.ops.check_reset_block(hw))
6417                 e1000_phy_hw_reset(&adapter->hw);
6418 err_hw_init:
6419         kfree(adapter->tx_ring);
6420         kfree(adapter->rx_ring);
6421 err_sw_init:
6422         if (adapter->hw.flash_address)
6423                 iounmap(adapter->hw.flash_address);
6424         e1000e_reset_interrupt_capability(adapter);
6425 err_flashmap:
6426         iounmap(adapter->hw.hw_addr);
6427 err_ioremap:
6428         free_netdev(netdev);
6429 err_alloc_etherdev:
6430         pci_release_selected_regions(pdev,
6431                                      pci_select_bars(pdev, IORESOURCE_MEM));
6432 err_pci_reg:
6433 err_dma:
6434         pci_disable_device(pdev);
6435         return err;
6436 }
6437
6438 /**
6439  * e1000_remove - Device Removal Routine
6440  * @pdev: PCI device information struct
6441  *
6442  * e1000_remove is called by the PCI subsystem to alert the driver
6443  * that it should release a PCI device.  The could be caused by a
6444  * Hot-Plug event, or because the driver is going to be removed from
6445  * memory.
6446  **/
6447 static void __devexit e1000_remove(struct pci_dev *pdev)
6448 {
6449         struct net_device *netdev = pci_get_drvdata(pdev);
6450         struct e1000_adapter *adapter = netdev_priv(netdev);
6451         bool down = test_bit(__E1000_DOWN, &adapter->state);
6452
6453         /*
6454          * The timers may be rescheduled, so explicitly disable them
6455          * from being rescheduled.
6456          */
6457         if (!down)
6458                 set_bit(__E1000_DOWN, &adapter->state);
6459         del_timer_sync(&adapter->watchdog_timer);
6460         del_timer_sync(&adapter->phy_info_timer);
6461
6462         cancel_work_sync(&adapter->reset_task);
6463         cancel_work_sync(&adapter->watchdog_task);
6464         cancel_work_sync(&adapter->downshift_task);
6465         cancel_work_sync(&adapter->update_phy_task);
6466         cancel_work_sync(&adapter->print_hang_task);
6467
6468         if (!(netdev->flags & IFF_UP))
6469                 e1000_power_down_phy(adapter);
6470
6471         /* Don't lie to e1000_close() down the road. */
6472         if (!down)
6473                 clear_bit(__E1000_DOWN, &adapter->state);
6474         unregister_netdev(netdev);
6475
6476         if (pci_dev_run_wake(pdev))
6477                 pm_runtime_get_noresume(&pdev->dev);
6478
6479         /*
6480          * Release control of h/w to f/w.  If f/w is AMT enabled, this
6481          * would have already happened in close and is redundant.
6482          */
6483         e1000e_release_hw_control(adapter);
6484
6485         e1000e_reset_interrupt_capability(adapter);
6486         kfree(adapter->tx_ring);
6487         kfree(adapter->rx_ring);
6488
6489         iounmap(adapter->hw.hw_addr);
6490         if (adapter->hw.flash_address)
6491                 iounmap(adapter->hw.flash_address);
6492         pci_release_selected_regions(pdev,
6493                                      pci_select_bars(pdev, IORESOURCE_MEM));
6494
6495         free_netdev(netdev);
6496
6497         /* AER disable */
6498         pci_disable_pcie_error_reporting(pdev);
6499
6500         pci_disable_device(pdev);
6501 }
6502
6503 /* PCI Error Recovery (ERS) */
6504 static struct pci_error_handlers e1000_err_handler = {
6505         .error_detected = e1000_io_error_detected,
6506         .slot_reset = e1000_io_slot_reset,
6507         .resume = e1000_io_resume,
6508 };
6509
6510 static DEFINE_PCI_DEVICE_TABLE(e1000_pci_tbl) = {
6511         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82571EB_COPPER), board_82571 },
6512         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82571EB_FIBER), board_82571 },
6513         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82571EB_QUAD_COPPER), board_82571 },
6514         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82571EB_QUAD_COPPER_LP), board_82571 },
6515         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82571EB_QUAD_FIBER), board_82571 },
6516         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82571EB_SERDES), board_82571 },
6517         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82571EB_SERDES_DUAL), board_82571 },
6518         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82571EB_SERDES_QUAD), board_82571 },
6519         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82571PT_QUAD_COPPER), board_82571 },
6520
6521         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82572EI), board_82572 },
6522         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82572EI_COPPER), board_82572 },
6523         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82572EI_FIBER), board_82572 },
6524         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82572EI_SERDES), board_82572 },
6525
6526         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82573E), board_82573 },
6527         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82573E_IAMT), board_82573 },
6528         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82573L), board_82573 },
6529
6530         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82574L), board_82574 },
6531         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82574LA), board_82574 },
6532         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82583V), board_82583 },
6533
6534         { PCI_VDEVICE(INTEL, E1000_DEV_ID_80003ES2LAN_COPPER_DPT),
6535           board_80003es2lan },
6536         { PCI_VDEVICE(INTEL, E1000_DEV_ID_80003ES2LAN_COPPER_SPT),
6537           board_80003es2lan },
6538         { PCI_VDEVICE(INTEL, E1000_DEV_ID_80003ES2LAN_SERDES_DPT),
6539           board_80003es2lan },
6540         { PCI_VDEVICE(INTEL, E1000_DEV_ID_80003ES2LAN_SERDES_SPT),
6541           board_80003es2lan },
6542
6543         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH8_IFE), board_ich8lan },
6544         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH8_IFE_G), board_ich8lan },
6545         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH8_IFE_GT), board_ich8lan },
6546         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH8_IGP_AMT), board_ich8lan },
6547         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH8_IGP_C), board_ich8lan },
6548         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH8_IGP_M), board_ich8lan },
6549         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH8_IGP_M_AMT), board_ich8lan },
6550         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH8_82567V_3), board_ich8lan },
6551
6552         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH9_IFE), board_ich9lan },
6553         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH9_IFE_G), board_ich9lan },
6554         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH9_IFE_GT), board_ich9lan },
6555         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH9_IGP_AMT), board_ich9lan },
6556         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH9_IGP_C), board_ich9lan },
6557         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH9_BM), board_ich9lan },
6558         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH9_IGP_M), board_ich9lan },
6559         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH9_IGP_M_AMT), board_ich9lan },
6560         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH9_IGP_M_V), board_ich9lan },
6561
6562         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH10_R_BM_LM), board_ich9lan },
6563         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH10_R_BM_LF), board_ich9lan },
6564         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH10_R_BM_V), board_ich9lan },
6565
6566         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH10_D_BM_LM), board_ich10lan },
6567         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH10_D_BM_LF), board_ich10lan },
6568         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH10_D_BM_V), board_ich10lan },
6569
6570         { PCI_VDEVICE(INTEL, E1000_DEV_ID_PCH_M_HV_LM), board_pchlan },
6571         { PCI_VDEVICE(INTEL, E1000_DEV_ID_PCH_M_HV_LC), board_pchlan },
6572         { PCI_VDEVICE(INTEL, E1000_DEV_ID_PCH_D_HV_DM), board_pchlan },
6573         { PCI_VDEVICE(INTEL, E1000_DEV_ID_PCH_D_HV_DC), board_pchlan },
6574
6575         { PCI_VDEVICE(INTEL, E1000_DEV_ID_PCH2_LV_LM), board_pch2lan },
6576         { PCI_VDEVICE(INTEL, E1000_DEV_ID_PCH2_LV_V), board_pch2lan },
6577
6578         { 0, 0, 0, 0, 0, 0, 0 } /* terminate list */
6579 };
6580 MODULE_DEVICE_TABLE(pci, e1000_pci_tbl);
6581
6582 #ifdef CONFIG_PM
6583 static const struct dev_pm_ops e1000_pm_ops = {
6584         SET_SYSTEM_SLEEP_PM_OPS(e1000_suspend, e1000_resume)
6585         SET_RUNTIME_PM_OPS(e1000_runtime_suspend,
6586                                 e1000_runtime_resume, e1000_idle)
6587 };
6588 #endif
6589
6590 /* PCI Device API Driver */
6591 static struct pci_driver e1000_driver = {
6592         .name     = e1000e_driver_name,
6593         .id_table = e1000_pci_tbl,
6594         .probe    = e1000_probe,
6595         .remove   = __devexit_p(e1000_remove),
6596 #ifdef CONFIG_PM
6597         .driver   = {
6598                 .pm = &e1000_pm_ops,
6599         },
6600 #endif
6601         .shutdown = e1000_shutdown,
6602         .err_handler = &e1000_err_handler
6603 };
6604
6605 /**
6606  * e1000_init_module - Driver Registration Routine
6607  *
6608  * e1000_init_module is the first routine called when the driver is
6609  * loaded. All it does is register with the PCI subsystem.
6610  **/
6611 static int __init e1000_init_module(void)
6612 {
6613         int ret;
6614         pr_info("Intel(R) PRO/1000 Network Driver - %s\n",
6615                 e1000e_driver_version);
6616         pr_info("Copyright(c) 1999 - 2012 Intel Corporation.\n");
6617         ret = pci_register_driver(&e1000_driver);
6618
6619         return ret;
6620 }
6621 module_init(e1000_init_module);
6622
6623 /**
6624  * e1000_exit_module - Driver Exit Cleanup Routine
6625  *
6626  * e1000_exit_module is called just before the driver is removed
6627  * from memory.
6628  **/
6629 static void __exit e1000_exit_module(void)
6630 {
6631         pci_unregister_driver(&e1000_driver);
6632 }
6633 module_exit(e1000_exit_module);
6634
6635
6636 MODULE_AUTHOR("Intel Corporation, <linux.nics@intel.com>");
6637 MODULE_DESCRIPTION("Intel(R) PRO/1000 Network Driver");
6638 MODULE_LICENSE("GPL");
6639 MODULE_VERSION(DRV_VERSION);
6640
6641 /* netdev.c */