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 "1.9.5" 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, "RDLEN"},
 114         {E1000_RDH, "RDH"},
 115         {E1000_RDT, "RDT"},
 116         {E1000_RDTR, "RDTR"},
 117         {E1000_RXDCTL(0), "RXDCTL"},
 118         {E1000_ERT, "ERT"},
 119         {E1000_RDBAL, "RDBAL"},
 120         {E1000_RDBAH, "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, "TDBAL"},
 130         {E1000_TDBAH, "TDBAH"},
 131         {E1000_TDLEN, "TDLEN"},
 132         {E1000_TDH, "TDH"},
 133         {E1000_TDT, "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 /**
 542  * e1000e_update_tail_wa - helper function for e1000e_update_[rt]dt_wa()
 543  * @hw: pointer to the HW structure
 544  * @tail: address of tail descriptor register
 545  * @i: value to write to tail descriptor register
 546  *
 547  * When updating the tail register, the ME could be accessing Host CSR
 548  * registers at the same time.  Normally, this is handled in h/w by an
 549  * arbiter but on some parts there is a bug that acknowledges Host accesses
 550  * later than it should which could result in the descriptor register to
 551  * have an incorrect value.  Workaround this by checking the FWSM register
 552  * which has bit 24 set while ME is accessing Host CSR registers, wait
 553  * if it is set and try again a number of times.
 554  **/
 555 static inline s32 e1000e_update_tail_wa(struct e1000_hw *hw, void __iomem *tail,
 556                                         unsigned int i)
 557 {
 558         unsigned int j = 0;
 559
 560         while ((j++ < E1000_ICH_FWSM_PCIM2PCI_COUNT) &&
 561                (er32(FWSM) & E1000_ICH_FWSM_PCIM2PCI))
 562                 udelay(50);
 563
 564         writel(i, tail);
 565
 566         if ((j == E1000_ICH_FWSM_PCIM2PCI_COUNT) && (i != readl(tail)))
 567                 return E1000_ERR_SWFW_SYNC;
 568
 569         return 0;
 570 }
 571
 572 static void e1000e_update_rdt_wa(struct e1000_ring *rx_ring, unsigned int i)
 573 {
 574         struct e1000_adapter *adapter = rx_ring->adapter;
 575         struct e1000_hw *hw = &adapter->hw;
 576
 577         if (e1000e_update_tail_wa(hw, rx_ring->tail, i)) {
 578                 u32 rctl = er32(RCTL);
 579                 ew32(RCTL, rctl & ~E1000_RCTL_EN);
 580                 e_err("ME firmware caused invalid RDT - resetting\n");
 581                 schedule_work(&adapter->reset_task);
 582         }
 583 }
 584
 585 static void e1000e_update_tdt_wa(struct e1000_ring *tx_ring, unsigned int i)
 586 {
 587         struct e1000_adapter *adapter = tx_ring->adapter;
 588         struct e1000_hw *hw = &adapter->hw;
 589
 590         if (e1000e_update_tail_wa(hw, tx_ring->tail, i)) {
 591                 u32 tctl = er32(TCTL);
 592                 ew32(TCTL, tctl & ~E1000_TCTL_EN);
 593                 e_err("ME firmware caused invalid TDT - resetting\n");
 594                 schedule_work(&adapter->reset_task);
 595         }
 596 }
 597
 598 /**
 599  * e1000_alloc_rx_buffers - Replace used receive buffers
 600  * @rx_ring: Rx descriptor ring
 601  **/
 602 static void e1000_alloc_rx_buffers(struct e1000_ring *rx_ring,
 603                                    int cleaned_count, gfp_t gfp)
 604 {
 605         struct e1000_adapter *adapter = rx_ring->adapter;
 606         struct net_device *netdev = adapter->netdev;
 607         struct pci_dev *pdev = adapter->pdev;
 608         union e1000_rx_desc_extended *rx_desc;
 609         struct e1000_buffer *buffer_info;
 610         struct sk_buff *skb;
 611         unsigned int i;
 612         unsigned int bufsz = adapter->rx_buffer_len;
 613
 614         i = rx_ring->next_to_use;
 615         buffer_info = &rx_ring->buffer_info[i];
 616
 617         while (cleaned_count--) {
 618                 skb = buffer_info->skb;
 619                 if (skb) {
 620                         skb_trim(skb, 0);
 621                         goto map_skb;
 622                 }
 623
 624                 skb = __netdev_alloc_skb_ip_align(netdev, bufsz, gfp);
 625                 if (!skb) {
 626                         /* Better luck next round */
 627                         adapter->alloc_rx_buff_failed++;
 628                         break;
 629                 }
 630
 631                 buffer_info->skb = skb;
 632 map_skb:
 633                 buffer_info->dma = dma_map_single(&pdev->dev, skb->data,
 634                                                   adapter->rx_buffer_len,
 635                                                   DMA_FROM_DEVICE);
 636                 if (dma_mapping_error(&pdev->dev, buffer_info->dma)) {
 637                         dev_err(&pdev->dev, "Rx DMA map failed\n");
 638                         adapter->rx_dma_failed++;
 639                         break;
 640                 }
 641
 642                 rx_desc = E1000_RX_DESC_EXT(*rx_ring, i);
 643                 rx_desc->read.buffer_addr = cpu_to_le64(buffer_info->dma);
 644
 645                 if (unlikely(!(i & (E1000_RX_BUFFER_WRITE - 1)))) {
 646                         /*
 647                          * Force memory writes to complete before letting h/w
 648                          * know there are new descriptors to fetch.  (Only
 649                          * applicable for weak-ordered memory model archs,
 650                          * such as IA-64).
 651                          */
 652                         wmb();
 653                         if (adapter->flags2 & FLAG2_PCIM2PCI_ARBITER_WA)
 654                                 e1000e_update_rdt_wa(rx_ring, i);
 655                         else
 656                                 writel(i, rx_ring->tail);
 657                 }
 658                 i++;
 659                 if (i == rx_ring->count)
 660                         i = 0;
 661                 buffer_info = &rx_ring->buffer_info[i];
 662         }
 663
 664         rx_ring->next_to_use = i;
 665 }
 666
 667 /**
 668  * e1000_alloc_rx_buffers_ps - Replace used receive buffers; packet split
 669  * @rx_ring: Rx descriptor ring
 670  **/
 671 static void e1000_alloc_rx_buffers_ps(struct e1000_ring *rx_ring,
 672                                       int cleaned_count, gfp_t gfp)
 673 {
 674         struct e1000_adapter *adapter = rx_ring->adapter;
 675         struct net_device *netdev = adapter->netdev;
 676         struct pci_dev *pdev = adapter->pdev;
 677         union e1000_rx_desc_packet_split *rx_desc;
 678         struct e1000_buffer *buffer_info;
 679         struct e1000_ps_page *ps_page;
 680         struct sk_buff *skb;
 681         unsigned int i, j;
 682
 683         i = rx_ring->next_to_use;
 684         buffer_info = &rx_ring->buffer_info[i];
 685
 686         while (cleaned_count--) {
 687                 rx_desc = E1000_RX_DESC_PS(*rx_ring, i);
 688
 689                 for (j = 0; j < PS_PAGE_BUFFERS; j++) {
 690                         ps_page = &buffer_info->ps_pages[j];
 691                         if (j >= adapter->rx_ps_pages) {
 692                                 /* all unused desc entries get hw null ptr */
 693                                 rx_desc->read.buffer_addr[j + 1] =
 694                                     ~cpu_to_le64(0);
 695                                 continue;
 696                         }
 697                         if (!ps_page->page) {
 698                                 ps_page->page = alloc_page(gfp);
 699                                 if (!ps_page->page) {
 700                                         adapter->alloc_rx_buff_failed++;
 701                                         goto no_buffers;
 702                                 }
 703                                 ps_page->dma = dma_map_page(&pdev->dev,
 704                                                             ps_page->page,
 705                                                             0, PAGE_SIZE,
 706                                                             DMA_FROM_DEVICE);
 707                                 if (dma_mapping_error(&pdev->dev,
 708                                                       ps_page->dma)) {
 709                                         dev_err(&adapter->pdev->dev,
 710                                                 "Rx DMA page map failed\n");
 711                                         adapter->rx_dma_failed++;
 712                                         goto no_buffers;
 713                                 }
 714                         }
 715                         /*
 716                          * Refresh the desc even if buffer_addrs
 717                          * didn't change because each write-back
 718                          * erases this info.
 719                          */
 720                         rx_desc->read.buffer_addr[j + 1] =
 721                             cpu_to_le64(ps_page->dma);
 722                 }
 723
 724                 skb = __netdev_alloc_skb_ip_align(netdev,
 725                                                   adapter->rx_ps_bsize0,
 726                                                   gfp);
 727
 728                 if (!skb) {
 729                         adapter->alloc_rx_buff_failed++;
 730                         break;
 731                 }
 732
 733                 buffer_info->skb = skb;
 734                 buffer_info->dma = dma_map_single(&pdev->dev, skb->data,
 735                                                   adapter->rx_ps_bsize0,
 736                                                   DMA_FROM_DEVICE);
 737                 if (dma_mapping_error(&pdev->dev, buffer_info->dma)) {
 738                         dev_err(&pdev->dev, "Rx DMA map failed\n");
 739                         adapter->rx_dma_failed++;
 740                         /* cleanup skb */
 741                         dev_kfree_skb_any(skb);
 742                         buffer_info->skb = NULL;
 743                         break;
 744                 }
 745
 746                 rx_desc->read.buffer_addr[0] = cpu_to_le64(buffer_info->dma);
 747
 748                 if (unlikely(!(i & (E1000_RX_BUFFER_WRITE - 1)))) {
 749                         /*
 750                          * Force memory writes to complete before letting h/w
 751                          * know there are new descriptors to fetch.  (Only
 752                          * applicable for weak-ordered memory model archs,
 753                          * such as IA-64).
 754                          */
 755                         wmb();
 756                         if (adapter->flags2 & FLAG2_PCIM2PCI_ARBITER_WA)
 757                                 e1000e_update_rdt_wa(rx_ring, i << 1);
 758                         else
 759                                 writel(i << 1, rx_ring->tail);
 760                 }
 761
 762                 i++;
 763                 if (i == rx_ring->count)
 764                         i = 0;
 765                 buffer_info = &rx_ring->buffer_info[i];
 766         }
 767
 768 no_buffers:
 769         rx_ring->next_to_use = i;
 770 }
 771
 772 /**
 773  * e1000_alloc_jumbo_rx_buffers - Replace used jumbo receive buffers
 774  * @rx_ring: Rx descriptor ring
 775  * @cleaned_count: number of buffers to allocate this pass
 776  **/
 777
 778 static void e1000_alloc_jumbo_rx_buffers(struct e1000_ring *rx_ring,
 779                                          int cleaned_count, gfp_t gfp)
 780 {
 781         struct e1000_adapter *adapter = rx_ring->adapter;
 782         struct net_device *netdev = adapter->netdev;
 783         struct pci_dev *pdev = adapter->pdev;
 784         union e1000_rx_desc_extended *rx_desc;
 785         struct e1000_buffer *buffer_info;
 786         struct sk_buff *skb;
 787         unsigned int i;
 788         unsigned int bufsz = 256 - 16 /* for skb_reserve */;
 789
 790         i = rx_ring->next_to_use;
 791         buffer_info = &rx_ring->buffer_info[i];
 792
 793         while (cleaned_count--) {
 794                 skb = buffer_info->skb;
 795                 if (skb) {
 796                         skb_trim(skb, 0);
 797                         goto check_page;
 798                 }
 799
 800                 skb = __netdev_alloc_skb_ip_align(netdev, bufsz, gfp);
 801                 if (unlikely(!skb)) {
 802                         /* Better luck next round */
 803                         adapter->alloc_rx_buff_failed++;
 804                         break;
 805                 }
 806
 807                 buffer_info->skb = skb;
 808 check_page:
 809                 /* allocate a new page if necessary */
 810                 if (!buffer_info->page) {
 811                         buffer_info->page = alloc_page(gfp);
 812                         if (unlikely(!buffer_info->page)) {
 813                                 adapter->alloc_rx_buff_failed++;
 814                                 break;
 815                         }
 816                 }
 817
 818                 if (!buffer_info->dma)
 819                         buffer_info->dma = dma_map_page(&pdev->dev,
 820                                                         buffer_info->page, 0,
 821                                                         PAGE_SIZE,
 822                                                         DMA_FROM_DEVICE);
 823
 824                 rx_desc = E1000_RX_DESC_EXT(*rx_ring, i);
 825                 rx_desc->read.buffer_addr = cpu_to_le64(buffer_info->dma);
 826
 827                 if (unlikely(++i == rx_ring->count))
 828                         i = 0;
 829                 buffer_info = &rx_ring->buffer_info[i];
 830         }
 831
 832         if (likely(rx_ring->next_to_use != i)) {
 833                 rx_ring->next_to_use = i;
 834                 if (unlikely(i-- == 0))
 835                         i = (rx_ring->count - 1);
 836
 837                 /* Force memory writes to complete before letting h/w
 838                  * know there are new descriptors to fetch.  (Only
 839                  * applicable for weak-ordered memory model archs,
 840                  * such as IA-64). */
 841                 wmb();
 842                 if (adapter->flags2 & FLAG2_PCIM2PCI_ARBITER_WA)
 843                         e1000e_update_rdt_wa(rx_ring, i);
 844                 else
 845                         writel(i, rx_ring->tail);
 846         }
 847 }
 848
 849 static inline void e1000_rx_hash(struct net_device *netdev, __le32 rss,
 850                                  struct sk_buff *skb)
 851 {
 852         if (netdev->features & NETIF_F_RXHASH)
 853                 skb->rxhash = le32_to_cpu(rss);
 854 }
 855
 856 /**
 857  * e1000_clean_rx_irq - Send received data up the network stack
 858  * @rx_ring: Rx descriptor ring
 859  *
 860  * the return value indicates whether actual cleaning was done, there
 861  * is no guarantee that everything was cleaned
 862  **/
 863 static bool e1000_clean_rx_irq(struct e1000_ring *rx_ring, int *work_done,
 864                                int work_to_do)
 865 {
 866         struct e1000_adapter *adapter = rx_ring->adapter;
 867         struct net_device *netdev = adapter->netdev;
 868         struct pci_dev *pdev = adapter->pdev;
 869         struct e1000_hw *hw = &adapter->hw;
 870         union e1000_rx_desc_extended *rx_desc, *next_rxd;
 871         struct e1000_buffer *buffer_info, *next_buffer;
 872         u32 length, staterr;
 873         unsigned int i;
 874         int cleaned_count = 0;
 875         bool cleaned = false;
 876         unsigned int total_rx_bytes = 0, total_rx_packets = 0;
 877
 878         i = rx_ring->next_to_clean;
 879         rx_desc = E1000_RX_DESC_EXT(*rx_ring, i);
 880         staterr = le32_to_cpu(rx_desc->wb.upper.status_error);
 881         buffer_info = &rx_ring->buffer_info[i];
 882
 883         while (staterr & E1000_RXD_STAT_DD) {
 884                 struct sk_buff *skb;
 885
 886                 if (*work_done >= work_to_do)
 887                         break;
 888                 (*work_done)++;
 889                 rmb();  /* read descriptor and rx_buffer_info after status DD */
 890
 891                 skb = buffer_info->skb;
 892                 buffer_info->skb = NULL;
 893
 894                 prefetch(skb->data - NET_IP_ALIGN);
 895
 896                 i++;
 897                 if (i == rx_ring->count)
 898                         i = 0;
 899                 next_rxd = E1000_RX_DESC_EXT(*rx_ring, i);
 900                 prefetch(next_rxd);
 901
 902                 next_buffer = &rx_ring->buffer_info[i];
 903
 904                 cleaned = true;
 905                 cleaned_count++;
 906                 dma_unmap_single(&pdev->dev,
 907                                  buffer_info->dma,
 908                                  adapter->rx_buffer_len,
 909                                  DMA_FROM_DEVICE);
 910                 buffer_info->dma = 0;
 911
 912                 length = le16_to_cpu(rx_desc->wb.upper.length);
 913
 914                 /*
 915                  * !EOP means multiple descriptors were used to store a single
 916                  * packet, if that's the case we need to toss it.  In fact, we
 917                  * need to toss every packet with the EOP bit clear and the
 918                  * next frame that _does_ have the EOP bit set, as it is by
 919                  * definition only a frame fragment
 920                  */
 921                 if (unlikely(!(staterr & E1000_RXD_STAT_EOP)))
 922                         adapter->flags2 |= FLAG2_IS_DISCARDING;
 923
 924                 if (adapter->flags2 & FLAG2_IS_DISCARDING) {
 925                         /* All receives must fit into a single buffer */
 926                         e_dbg("Receive packet consumed multiple buffers\n");
 927                         /* recycle */
 928                         buffer_info->skb = skb;
 929                         if (staterr & E1000_RXD_STAT_EOP)
 930                                 adapter->flags2 &= ~FLAG2_IS_DISCARDING;
 931                         goto next_desc;
 932                 }
 933
 934                 if (unlikely((staterr & E1000_RXDEXT_ERR_FRAME_ERR_MASK) &&
 935                              !(netdev->features & NETIF_F_RXALL))) {
 936                         /* recycle */
 937                         buffer_info->skb = skb;
 938                         goto next_desc;
 939                 }
 940
 941                 /* adjust length to remove Ethernet CRC */
 942                 if (!(adapter->flags2 & FLAG2_CRC_STRIPPING)) {
 943                         /* If configured to store CRC, don't subtract FCS,
 944                          * but keep the FCS bytes out of the total_rx_bytes
 945                          * counter
 946                          */
 947                         if (netdev->features & NETIF_F_RXFCS)
 948                                 total_rx_bytes -= 4;
 949                         else
 950                                 length -= 4;
 951                 }
 952
 953                 total_rx_bytes += length;
 954                 total_rx_packets++;
 955
 956                 /*
 957                  * code added for copybreak, this should improve
 958                  * performance for small packets with large amounts
 959                  * of reassembly being done in the stack
 960                  */
 961                 if (length < copybreak) {
 962                         struct sk_buff *new_skb =
 963                             netdev_alloc_skb_ip_align(netdev, length);
 964                         if (new_skb) {
 965                                 skb_copy_to_linear_data_offset(new_skb,
 966                                                                -NET_IP_ALIGN,
 967                                                                (skb->data -
 968                                                                 NET_IP_ALIGN),
 969                                                                (length +
 970                                                                 NET_IP_ALIGN));
 971                                 /* save the skb in buffer_info as good */
 972                                 buffer_info->skb = skb;
 973                                 skb = new_skb;
 974                         }
 975                         /* else just continue with the old one */
 976                 }
 977                 /* end copybreak code */
 978                 skb_put(skb, length);
 979
 980                 /* Receive Checksum Offload */
 981                 e1000_rx_checksum(adapter, staterr,
 982                                   rx_desc->wb.lower.hi_dword.csum_ip.csum, skb);
 983
 984                 e1000_rx_hash(netdev, rx_desc->wb.lower.hi_dword.rss, skb);
 985
 986                 e1000_receive_skb(adapter, netdev, skb, staterr,
 987                                   rx_desc->wb.upper.vlan);
 988
 989 next_desc:
 990                 rx_desc->wb.upper.status_error &= cpu_to_le32(~0xFF);
 991
 992                 /* return some buffers to hardware, one at a time is too slow */
 993                 if (cleaned_count >= E1000_RX_BUFFER_WRITE) {
 994                         adapter->alloc_rx_buf(rx_ring, cleaned_count,
 995                                               GFP_ATOMIC);
 996                         cleaned_count = 0;
 997                 }
 998
 999                 /* use prefetched values */
1000                 rx_desc = next_rxd;
1001                 buffer_info = next_buffer;
1002
1003                 staterr = le32_to_cpu(rx_desc->wb.upper.status_error);
1004         }
1005         rx_ring->next_to_clean = i;
1006
1007         cleaned_count = e1000_desc_unused(rx_ring);
1008         if (cleaned_count)
1009                 adapter->alloc_rx_buf(rx_ring, cleaned_count, GFP_ATOMIC);
1010
1011         adapter->total_rx_bytes += total_rx_bytes;
1012         adapter->total_rx_packets += total_rx_packets;
1013         return cleaned;
1014 }
1015
1016 static void e1000_put_txbuf(struct e1000_ring *tx_ring,
1017                             struct e1000_buffer *buffer_info)
1018 {
1019         struct e1000_adapter *adapter = tx_ring->adapter;
1020
1021         if (buffer_info->dma) {
1022                 if (buffer_info->mapped_as_page)
1023                         dma_unmap_page(&adapter->pdev->dev, buffer_info->dma,
1024                                        buffer_info->length, DMA_TO_DEVICE);
1025                 else
1026                         dma_unmap_single(&adapter->pdev->dev, buffer_info->dma,
1027                                          buffer_info->length, DMA_TO_DEVICE);
1028                 buffer_info->dma = 0;
1029         }
1030         if (buffer_info->skb) {
1031                 dev_kfree_skb_any(buffer_info->skb);
1032                 buffer_info->skb = NULL;
1033         }
1034         buffer_info->time_stamp = 0;
1035 }
1036
1037 static void e1000_print_hw_hang(struct work_struct *work)
1038 {
1039         struct e1000_adapter *adapter = container_of(work,
1040                                                      struct e1000_adapter,
1041                                                      print_hang_task);
1042         struct net_device *netdev = adapter->netdev;
1043         struct e1000_ring *tx_ring = adapter->tx_ring;
1044         unsigned int i = tx_ring->next_to_clean;
1045         unsigned int eop = tx_ring->buffer_info[i].next_to_watch;
1046         struct e1000_tx_desc *eop_desc = E1000_TX_DESC(*tx_ring, eop);
1047         struct e1000_hw *hw = &adapter->hw;
1048         u16 phy_status, phy_1000t_status, phy_ext_status;
1049         u16 pci_status;
1050
1051         if (test_bit(__E1000_DOWN, &adapter->state))
1052                 return;
1053
1054         if (!adapter->tx_hang_recheck &&
1055             (adapter->flags2 & FLAG2_DMA_BURST)) {
1056                 /* May be block on write-back, flush and detect again
1057                  * flush pending descriptor writebacks to memory
1058                  */
1059                 ew32(TIDV, adapter->tx_int_delay | E1000_TIDV_FPD);
1060                 /* execute the writes immediately */
1061                 e1e_flush();
1062                 adapter->tx_hang_recheck = true;
1063                 return;
1064         }
1065         /* Real hang detected */
1066         adapter->tx_hang_recheck = false;
1067         netif_stop_queue(netdev);
1068
1069         e1e_rphy(hw, PHY_STATUS, &phy_status);
1070         e1e_rphy(hw, PHY_1000T_STATUS, &phy_1000t_status);
1071         e1e_rphy(hw, PHY_EXT_STATUS, &phy_ext_status);
1072
1073         pci_read_config_word(adapter->pdev, PCI_STATUS, &pci_status);
1074
1075         /* detected Hardware unit hang */
1076         e_err("Detected Hardware Unit Hang:\n"
1077               "  TDH                  <%x>\n"
1078               "  TDT                  <%x>\n"
1079               "  next_to_use          <%x>\n"
1080               "  next_to_clean        <%x>\n"
1081               "buffer_info[next_to_clean]:\n"
1082               "  time_stamp           <%lx>\n"
1083               "  next_to_watch        <%x>\n"
1084               "  jiffies              <%lx>\n"
1085               "  next_to_watch.status <%x>\n"
1086               "MAC Status             <%x>\n"
1087               "PHY Status             <%x>\n"
1088               "PHY 1000BASE-T Status  <%x>\n"
1089               "PHY Extended Status    <%x>\n"
1090               "PCI Status             <%x>\n",
1091               readl(tx_ring->head),
1092               readl(tx_ring->tail),
1093               tx_ring->next_to_use,
1094               tx_ring->next_to_clean,
1095               tx_ring->buffer_info[eop].time_stamp,
1096               eop,
1097               jiffies,
1098               eop_desc->upper.fields.status,
1099               er32(STATUS),
1100               phy_status,
1101               phy_1000t_status,
1102               phy_ext_status,
1103               pci_status);
1104 }
1105
1106 /**
1107  * e1000_clean_tx_irq - Reclaim resources after transmit completes
1108  * @tx_ring: Tx descriptor ring
1109  *
1110  * the return value indicates whether actual cleaning was done, there
1111  * is no guarantee that everything was cleaned
1112  **/
1113 static bool e1000_clean_tx_irq(struct e1000_ring *tx_ring)
1114 {
1115         struct e1000_adapter *adapter = tx_ring->adapter;
1116         struct net_device *netdev = adapter->netdev;
1117         struct e1000_hw *hw = &adapter->hw;
1118         struct e1000_tx_desc *tx_desc, *eop_desc;
1119         struct e1000_buffer *buffer_info;
1120         unsigned int i, eop;
1121         unsigned int count = 0;
1122         unsigned int total_tx_bytes = 0, total_tx_packets = 0;
1123         unsigned int bytes_compl = 0, pkts_compl = 0;
1124
1125         i = tx_ring->next_to_clean;
1126         eop = tx_ring->buffer_info[i].next_to_watch;
1127         eop_desc = E1000_TX_DESC(*tx_ring, eop);
1128
1129         while ((eop_desc->upper.data & cpu_to_le32(E1000_TXD_STAT_DD)) &&
1130                (count < tx_ring->count)) {
1131                 bool cleaned = false;
1132                 rmb(); /* read buffer_info after eop_desc */
1133                 for (; !cleaned; count++) {
1134                         tx_desc = E1000_TX_DESC(*tx_ring, i);
1135                         buffer_info = &tx_ring->buffer_info[i];
1136                         cleaned = (i == eop);
1137
1138                         if (cleaned) {
1139                                 total_tx_packets += buffer_info->segs;
1140                                 total_tx_bytes += buffer_info->bytecount;
1141                                 if (buffer_info->skb) {
1142                                         bytes_compl += buffer_info->skb->len;
1143                                         pkts_compl++;
1144                                 }
1145                         }
1146
1147                         e1000_put_txbuf(tx_ring, buffer_info);
1148                         tx_desc->upper.data = 0;
1149
1150                         i++;
1151                         if (i == tx_ring->count)
1152                                 i = 0;
1153                 }
1154
1155                 if (i == tx_ring->next_to_use)
1156                         break;
1157                 eop = tx_ring->buffer_info[i].next_to_watch;
1158                 eop_desc = E1000_TX_DESC(*tx_ring, eop);
1159         }
1160
1161         tx_ring->next_to_clean = i;
1162
1163         netdev_completed_queue(netdev, pkts_compl, bytes_compl);
1164
1165 #define TX_WAKE_THRESHOLD 32
1166         if (count && netif_carrier_ok(netdev) &&
1167             e1000_desc_unused(tx_ring) >= TX_WAKE_THRESHOLD) {
1168                 /* Make sure that anybody stopping the queue after this
1169                  * sees the new next_to_clean.
1170                  */
1171                 smp_mb();
1172
1173                 if (netif_queue_stopped(netdev) &&
1174                     !(test_bit(__E1000_DOWN, &adapter->state))) {
1175                         netif_wake_queue(netdev);
1176                         ++adapter->restart_queue;
1177                 }
1178         }
1179
1180         if (adapter->detect_tx_hung) {
1181                 /*
1182                  * Detect a transmit hang in hardware, this serializes the
1183                  * check with the clearing of time_stamp and movement of i
1184                  */
1185                 adapter->detect_tx_hung = false;
1186                 if (tx_ring->buffer_info[i].time_stamp &&
1187                     time_after(jiffies, tx_ring->buffer_info[i].time_stamp
1188                                + (adapter->tx_timeout_factor * HZ)) &&
1189                     !(er32(STATUS) & E1000_STATUS_TXOFF))
1190                         schedule_work(&adapter->print_hang_task);
1191                 else
1192                         adapter->tx_hang_recheck = false;
1193         }
1194         adapter->total_tx_bytes += total_tx_bytes;
1195         adapter->total_tx_packets += total_tx_packets;
1196         return count < tx_ring->count;
1197 }
1198
1199 /**
1200  * e1000_clean_rx_irq_ps - Send received data up the network stack; packet split
1201  * @rx_ring: Rx descriptor ring
1202  *
1203  * the return value indicates whether actual cleaning was done, there
1204  * is no guarantee that everything was cleaned
1205  **/
1206 static bool e1000_clean_rx_irq_ps(struct e1000_ring *rx_ring, int *work_done,
1207                                   int work_to_do)
1208 {
1209         struct e1000_adapter *adapter = rx_ring->adapter;
1210         struct e1000_hw *hw = &adapter->hw;
1211         union e1000_rx_desc_packet_split *rx_desc, *next_rxd;
1212         struct net_device *netdev = adapter->netdev;
1213         struct pci_dev *pdev = adapter->pdev;
1214         struct e1000_buffer *buffer_info, *next_buffer;
1215         struct e1000_ps_page *ps_page;
1216         struct sk_buff *skb;
1217         unsigned int i, j;
1218         u32 length, staterr;
1219         int cleaned_count = 0;
1220         bool cleaned = false;
1221         unsigned int total_rx_bytes = 0, total_rx_packets = 0;
1222
1223         i = rx_ring->next_to_clean;
1224         rx_desc = E1000_RX_DESC_PS(*rx_ring, i);
1225         staterr = le32_to_cpu(rx_desc->wb.middle.status_error);
1226         buffer_info = &rx_ring->buffer_info[i];
1227
1228         while (staterr & E1000_RXD_STAT_DD) {
1229                 if (*work_done >= work_to_do)
1230                         break;
1231                 (*work_done)++;
1232                 skb = buffer_info->skb;
1233                 rmb();  /* read descriptor and rx_buffer_info after status DD */
1234
1235                 /* in the packet split case this is header only */
1236                 prefetch(skb->data - NET_IP_ALIGN);
1237
1238                 i++;
1239                 if (i == rx_ring->count)
1240                         i = 0;
1241                 next_rxd = E1000_RX_DESC_PS(*rx_ring, i);
1242                 prefetch(next_rxd);
1243
1244                 next_buffer = &rx_ring->buffer_info[i];
1245
1246                 cleaned = true;
1247                 cleaned_count++;
1248                 dma_unmap_single(&pdev->dev, buffer_info->dma,
1249                                  adapter->rx_ps_bsize0, DMA_FROM_DEVICE);
1250                 buffer_info->dma = 0;
1251
1252                 /* see !EOP comment in other Rx routine */
1253                 if (!(staterr & E1000_RXD_STAT_EOP))
1254                         adapter->flags2 |= FLAG2_IS_DISCARDING;
1255
1256                 if (adapter->flags2 & FLAG2_IS_DISCARDING) {
1257                         e_dbg("Packet Split buffers didn't pick up the full packet\n");
1258                         dev_kfree_skb_irq(skb);
1259                         if (staterr & E1000_RXD_STAT_EOP)
1260                                 adapter->flags2 &= ~FLAG2_IS_DISCARDING;
1261                         goto next_desc;
1262                 }
1263
1264                 if (unlikely((staterr & E1000_RXDEXT_ERR_FRAME_ERR_MASK) &&
1265                              !(netdev->features & NETIF_F_RXALL))) {
1266                         dev_kfree_skb_irq(skb);
1267                         goto next_desc;
1268                 }
1269
1270                 length = le16_to_cpu(rx_desc->wb.middle.length0);
1271
1272                 if (!length) {
1273                         e_dbg("Last part of the packet spanning multiple descriptors\n");
1274                         dev_kfree_skb_irq(skb);
1275                         goto next_desc;
1276                 }
1277
1278                 /* Good Receive */
1279                 skb_put(skb, length);
1280
1281                 {
1282                         /*
1283                          * this looks ugly, but it seems compiler issues make
1284                          * it more efficient than reusing j
1285                          */
1286                         int l1 = le16_to_cpu(rx_desc->wb.upper.length[0]);
1287
1288                         /*
1289                          * page alloc/put takes too long and effects small
1290                          * packet throughput, so unsplit small packets and
1291                          * save the alloc/put only valid in softirq (napi)
1292                          * context to call kmap_*
1293                          */
1294                         if (l1 && (l1 <= copybreak) &&
1295                             ((length + l1) <= adapter->rx_ps_bsize0)) {
1296                                 u8 *vaddr;
1297
1298                                 ps_page = &buffer_info->ps_pages[0];
1299
1300                                 /*
1301                                  * there is no documentation about how to call
1302                                  * kmap_atomic, so we can't hold the mapping
1303                                  * very long
1304                                  */
1305                                 dma_sync_single_for_cpu(&pdev->dev,
1306                                                         ps_page->dma,
1307                                                         PAGE_SIZE,
1308                                                         DMA_FROM_DEVICE);
1309                                 vaddr = kmap_atomic(ps_page->page);
1310                                 memcpy(skb_tail_pointer(skb), vaddr, l1);
1311                                 kunmap_atomic(vaddr);
1312                                 dma_sync_single_for_device(&pdev->dev,
1313                                                            ps_page->dma,
1314                                                            PAGE_SIZE,
1315                                                            DMA_FROM_DEVICE);
1316
1317                                 /* remove the CRC */
1318                                 if (!(adapter->flags2 & FLAG2_CRC_STRIPPING)) {
1319                                         if (!(netdev->features & NETIF_F_RXFCS))
1320                                                 l1 -= 4;
1321                                 }
1322
1323                                 skb_put(skb, l1);
1324                                 goto copydone;
1325                         } /* if */
1326                 }
1327
1328                 for (j = 0; j < PS_PAGE_BUFFERS; j++) {
1329                         length = le16_to_cpu(rx_desc->wb.upper.length[j]);
1330                         if (!length)
1331                                 break;
1332
1333                         ps_page = &buffer_info->ps_pages[j];
1334                         dma_unmap_page(&pdev->dev, ps_page->dma, PAGE_SIZE,
1335                                        DMA_FROM_DEVICE);
1336                         ps_page->dma = 0;
1337                         skb_fill_page_desc(skb, j, ps_page->page, 0, length);
1338                         ps_page->page = NULL;
1339                         skb->len += length;
1340                         skb->data_len += length;
1341                         skb->truesize += PAGE_SIZE;
1342                 }
1343
1344                 /* strip the ethernet crc, problem is we're using pages now so
1345                  * this whole operation can get a little cpu intensive
1346                  */
1347                 if (!(adapter->flags2 & FLAG2_CRC_STRIPPING)) {
1348                         if (!(netdev->features & NETIF_F_RXFCS))
1349                                 pskb_trim(skb, skb->len - 4);
1350                 }
1351
1352 copydone:
1353                 total_rx_bytes += skb->len;
1354                 total_rx_packets++;
1355
1356                 e1000_rx_checksum(adapter, staterr,
1357                                   rx_desc->wb.lower.hi_dword.csum_ip.csum, skb);
1358
1359                 e1000_rx_hash(netdev, rx_desc->wb.lower.hi_dword.rss, skb);
1360
1361                 if (rx_desc->wb.upper.header_status &
1362                            cpu_to_le16(E1000_RXDPS_HDRSTAT_HDRSP))
1363                         adapter->rx_hdr_split++;
1364
1365                 e1000_receive_skb(adapter, netdev, skb,
1366                                   staterr, rx_desc->wb.middle.vlan);
1367
1368 next_desc:
1369                 rx_desc->wb.middle.status_error &= cpu_to_le32(~0xFF);
1370                 buffer_info->skb = NULL;
1371
1372                 /* return some buffers to hardware, one at a time is too slow */
1373                 if (cleaned_count >= E1000_RX_BUFFER_WRITE) {
1374                         adapter->alloc_rx_buf(rx_ring, cleaned_count,
1375                                               GFP_ATOMIC);
1376                         cleaned_count = 0;
1377                 }
1378
1379                 /* use prefetched values */
1380                 rx_desc = next_rxd;
1381                 buffer_info = next_buffer;
1382
1383                 staterr = le32_to_cpu(rx_desc->wb.middle.status_error);
1384         }
1385         rx_ring->next_to_clean = i;
1386
1387         cleaned_count = e1000_desc_unused(rx_ring);
1388         if (cleaned_count)
1389                 adapter->alloc_rx_buf(rx_ring, cleaned_count, GFP_ATOMIC);
1390
1391         adapter->total_rx_bytes += total_rx_bytes;
1392         adapter->total_rx_packets += total_rx_packets;
1393         return cleaned;
1394 }
1395
1396 /**
1397  * e1000_consume_page - helper function
1398  **/
1399 static void e1000_consume_page(struct e1000_buffer *bi, struct sk_buff *skb,
1400                                u16 length)
1401 {
1402         bi->page = NULL;
1403         skb->len += length;
1404         skb->data_len += length;
1405         skb->truesize += PAGE_SIZE;
1406 }
1407
1408 /**
1409  * e1000_clean_jumbo_rx_irq - Send received data up the network stack; legacy
1410  * @adapter: board private structure
1411  *
1412  * the return value indicates whether actual cleaning was done, there
1413  * is no guarantee that everything was cleaned
1414  **/
1415 static bool e1000_clean_jumbo_rx_irq(struct e1000_ring *rx_ring, int *work_done,
1416                                      int work_to_do)
1417 {
1418         struct e1000_adapter *adapter = rx_ring->adapter;
1419         struct net_device *netdev = adapter->netdev;
1420         struct pci_dev *pdev = adapter->pdev;
1421         union e1000_rx_desc_extended *rx_desc, *next_rxd;
1422         struct e1000_buffer *buffer_info, *next_buffer;
1423         u32 length, staterr;
1424         unsigned int i;
1425         int cleaned_count = 0;
1426         bool cleaned = false;
1427         unsigned int total_rx_bytes=0, total_rx_packets=0;
1428
1429         i = rx_ring->next_to_clean;
1430         rx_desc = E1000_RX_DESC_EXT(*rx_ring, i);
1431         staterr = le32_to_cpu(rx_desc->wb.upper.status_error);
1432         buffer_info = &rx_ring->buffer_info[i];
1433
1434         while (staterr & E1000_RXD_STAT_DD) {
1435                 struct sk_buff *skb;
1436
1437                 if (*work_done >= work_to_do)
1438                         break;
1439                 (*work_done)++;
1440                 rmb();  /* read descriptor and rx_buffer_info after status DD */
1441
1442                 skb = buffer_info->skb;
1443                 buffer_info->skb = NULL;
1444
1445                 ++i;
1446                 if (i == rx_ring->count)
1447                         i = 0;
1448                 next_rxd = E1000_RX_DESC_EXT(*rx_ring, i);
1449                 prefetch(next_rxd);
1450
1451                 next_buffer = &rx_ring->buffer_info[i];
1452
1453                 cleaned = true;
1454                 cleaned_count++;
1455                 dma_unmap_page(&pdev->dev, buffer_info->dma, PAGE_SIZE,
1456                                DMA_FROM_DEVICE);
1457                 buffer_info->dma = 0;
1458
1459                 length = le16_to_cpu(rx_desc->wb.upper.length);
1460
1461                 /* errors is only valid for DD + EOP descriptors */
1462                 if (unlikely((staterr & E1000_RXD_STAT_EOP) &&
1463                              ((staterr & E1000_RXDEXT_ERR_FRAME_ERR_MASK) &&
1464                               !(netdev->features & NETIF_F_RXALL)))) {
1465                         /* recycle both page and skb */
1466                         buffer_info->skb = skb;
1467                         /* an error means any chain goes out the window too */
1468                         if (rx_ring->rx_skb_top)
1469                                 dev_kfree_skb_irq(rx_ring->rx_skb_top);
1470                         rx_ring->rx_skb_top = NULL;
1471                         goto next_desc;
1472                 }
1473
1474 #define rxtop (rx_ring->rx_skb_top)
1475                 if (!(staterr & E1000_RXD_STAT_EOP)) {
1476                         /* this descriptor is only the beginning (or middle) */
1477                         if (!rxtop) {
1478                                 /* this is the beginning of a chain */
1479                                 rxtop = skb;
1480                                 skb_fill_page_desc(rxtop, 0, buffer_info->page,
1481                                                    0, length);
1482                         } else {
1483                                 /* this is the middle of a chain */
1484                                 skb_fill_page_desc(rxtop,
1485                                     skb_shinfo(rxtop)->nr_frags,
1486                                     buffer_info->page, 0, length);
1487                                 /* re-use the skb, only consumed the page */
1488                                 buffer_info->skb = skb;
1489                         }
1490                         e1000_consume_page(buffer_info, rxtop, length);
1491                         goto next_desc;
1492                 } else {
1493                         if (rxtop) {
1494                                 /* end of the chain */
1495                                 skb_fill_page_desc(rxtop,
1496                                     skb_shinfo(rxtop)->nr_frags,
1497                                     buffer_info->page, 0, length);
1498                                 /* re-use the current skb, we only consumed the
1499                                  * page */
1500                                 buffer_info->skb = skb;
1501                                 skb = rxtop;
1502                                 rxtop = NULL;
1503                                 e1000_consume_page(buffer_info, skb, length);
1504                         } else {
1505                                 /* no chain, got EOP, this buf is the packet
1506                                  * copybreak to save the put_page/alloc_page */
1507                                 if (length <= copybreak &&
1508                                     skb_tailroom(skb) >= length) {
1509                                         u8 *vaddr;
1510                                         vaddr = kmap_atomic(buffer_info->page);
1511                                         memcpy(skb_tail_pointer(skb), vaddr,
1512                                                length);
1513                                         kunmap_atomic(vaddr);
1514                                         /* re-use the page, so don't erase
1515                                          * buffer_info->page */
1516                                         skb_put(skb, length);
1517                                 } else {
1518                                         skb_fill_page_desc(skb, 0,
1519                                                            buffer_info->page, 0,
1520                                                            length);
1521                                         e1000_consume_page(buffer_info, skb,
1522                                                            length);
1523                                 }
1524                         }
1525                 }
1526
1527                 /* Receive Checksum Offload XXX recompute due to CRC strip? */
1528                 e1000_rx_checksum(adapter, staterr,
1529                                   rx_desc->wb.lower.hi_dword.csum_ip.csum, skb);
1530
1531                 e1000_rx_hash(netdev, rx_desc->wb.lower.hi_dword.rss, skb);
1532
1533                 /* probably a little skewed due to removing CRC */
1534                 total_rx_bytes += skb->len;
1535                 total_rx_packets++;
1536
1537                 /* eth type trans needs skb->data to point to something */
1538                 if (!pskb_may_pull(skb, ETH_HLEN)) {
1539                         e_err("pskb_may_pull failed.\n");
1540                         dev_kfree_skb_irq(skb);
1541                         goto next_desc;
1542                 }
1543
1544                 e1000_receive_skb(adapter, netdev, skb, staterr,
1545                                   rx_desc->wb.upper.vlan);
1546
1547 next_desc:
1548                 rx_desc->wb.upper.status_error &= cpu_to_le32(~0xFF);
1549
1550                 /* return some buffers to hardware, one at a time is too slow */
1551                 if (unlikely(cleaned_count >= E1000_RX_BUFFER_WRITE)) {
1552                         adapter->alloc_rx_buf(rx_ring, cleaned_count,
1553                                               GFP_ATOMIC);
1554                         cleaned_count = 0;
1555                 }
1556
1557                 /* use prefetched values */
1558                 rx_desc = next_rxd;
1559                 buffer_info = next_buffer;
1560
1561                 staterr = le32_to_cpu(rx_desc->wb.upper.status_error);
1562         }
1563         rx_ring->next_to_clean = i;
1564
1565         cleaned_count = e1000_desc_unused(rx_ring);
1566         if (cleaned_count)
1567                 adapter->alloc_rx_buf(rx_ring, cleaned_count, GFP_ATOMIC);
1568
1569         adapter->total_rx_bytes += total_rx_bytes;
1570         adapter->total_rx_packets += total_rx_packets;
1571         return cleaned;
1572 }
1573
1574 /**
1575  * e1000_clean_rx_ring - Free Rx Buffers per Queue
1576  * @rx_ring: Rx descriptor ring
1577  **/
1578 static void e1000_clean_rx_ring(struct e1000_ring *rx_ring)
1579 {
1580         struct e1000_adapter *adapter = rx_ring->adapter;
1581         struct e1000_buffer *buffer_info;
1582         struct e1000_ps_page *ps_page;
1583         struct pci_dev *pdev = adapter->pdev;
1584         unsigned int i, j;
1585
1586         /* Free all the Rx ring sk_buffs */
1587         for (i = 0; i < rx_ring->count; i++) {
1588                 buffer_info = &rx_ring->buffer_info[i];
1589                 if (buffer_info->dma) {
1590                         if (adapter->clean_rx == e1000_clean_rx_irq)
1591                                 dma_unmap_single(&pdev->dev, buffer_info->dma,
1592                                                  adapter->rx_buffer_len,
1593                                                  DMA_FROM_DEVICE);
1594                         else if (adapter->clean_rx == e1000_clean_jumbo_rx_irq)
1595                                 dma_unmap_page(&pdev->dev, buffer_info->dma,
1596                                                PAGE_SIZE,
1597                                                DMA_FROM_DEVICE);
1598                         else if (adapter->clean_rx == e1000_clean_rx_irq_ps)
1599                                 dma_unmap_single(&pdev->dev, buffer_info->dma,
1600                                                  adapter->rx_ps_bsize0,
1601                                                  DMA_FROM_DEVICE);
1602                         buffer_info->dma = 0;
1603                 }
1604
1605                 if (buffer_info->page) {
1606                         put_page(buffer_info->page);
1607                         buffer_info->page = NULL;
1608                 }
1609
1610                 if (buffer_info->skb) {
1611                         dev_kfree_skb(buffer_info->skb);
1612                         buffer_info->skb = NULL;
1613                 }
1614
1615                 for (j = 0; j < PS_PAGE_BUFFERS; j++) {
1616                         ps_page = &buffer_info->ps_pages[j];
1617                         if (!ps_page->page)
1618                                 break;
1619                         dma_unmap_page(&pdev->dev, ps_page->dma, PAGE_SIZE,
1620                                        DMA_FROM_DEVICE);
1621                         ps_page->dma = 0;
1622                         put_page(ps_page->page);
1623                         ps_page->page = NULL;
1624                 }
1625         }
1626
1627         /* there also may be some cached data from a chained receive */
1628         if (rx_ring->rx_skb_top) {
1629                 dev_kfree_skb(rx_ring->rx_skb_top);
1630                 rx_ring->rx_skb_top = NULL;
1631         }
1632
1633         /* Zero out the descriptor ring */
1634         memset(rx_ring->desc, 0, rx_ring->size);
1635
1636         rx_ring->next_to_clean = 0;
1637         rx_ring->next_to_use = 0;
1638         adapter->flags2 &= ~FLAG2_IS_DISCARDING;
1639
1640         writel(0, rx_ring->head);
1641         writel(0, rx_ring->tail);
1642 }
1643
1644 static void e1000e_downshift_workaround(struct work_struct *work)
1645 {
1646         struct e1000_adapter *adapter = container_of(work,
1647                                         struct e1000_adapter, downshift_task);
1648
1649         if (test_bit(__E1000_DOWN, &adapter->state))
1650                 return;
1651
1652         e1000e_gig_downshift_workaround_ich8lan(&adapter->hw);
1653 }
1654
1655 /**
1656  * e1000_intr_msi - Interrupt Handler
1657  * @irq: interrupt number
1658  * @data: pointer to a network interface device structure
1659  **/
1660 static irqreturn_t e1000_intr_msi(int irq, void *data)
1661 {
1662         struct net_device *netdev = data;
1663         struct e1000_adapter *adapter = netdev_priv(netdev);
1664         struct e1000_hw *hw = &adapter->hw;
1665         u32 icr = er32(ICR);
1666
1667         /*
1668          * read ICR disables interrupts using IAM
1669          */
1670
1671         if (icr & E1000_ICR_LSC) {
1672                 hw->mac.get_link_status = true;
1673                 /*
1674                  * ICH8 workaround-- Call gig speed drop workaround on cable
1675                  * disconnect (LSC) before accessing any PHY registers
1676                  */
1677                 if ((adapter->flags & FLAG_LSC_GIG_SPEED_DROP) &&
1678                     (!(er32(STATUS) & E1000_STATUS_LU)))
1679                         schedule_work(&adapter->downshift_task);
1680
1681                 /*
1682                  * 80003ES2LAN workaround-- For packet buffer work-around on
1683                  * link down event; disable receives here in the ISR and reset
1684                  * adapter in watchdog
1685                  */
1686                 if (netif_carrier_ok(netdev) &&
1687                     adapter->flags & FLAG_RX_NEEDS_RESTART) {
1688                         /* disable receives */
1689                         u32 rctl = er32(RCTL);
1690                         ew32(RCTL, rctl & ~E1000_RCTL_EN);
1691                         adapter->flags |= FLAG_RX_RESTART_NOW;
1692                 }
1693                 /* guard against interrupt when we're going down */
1694                 if (!test_bit(__E1000_DOWN, &adapter->state))
1695                         mod_timer(&adapter->watchdog_timer, jiffies + 1);
1696         }
1697
1698         if (napi_schedule_prep(&adapter->napi)) {
1699                 adapter->total_tx_bytes = 0;
1700                 adapter->total_tx_packets = 0;
1701                 adapter->total_rx_bytes = 0;
1702                 adapter->total_rx_packets = 0;
1703                 __napi_schedule(&adapter->napi);
1704         }
1705
1706         return IRQ_HANDLED;
1707 }
1708
1709 /**
1710  * e1000_intr - Interrupt Handler
1711  * @irq: interrupt number
1712  * @data: pointer to a network interface device structure
1713  **/
1714 static irqreturn_t e1000_intr(int irq, void *data)
1715 {
1716         struct net_device *netdev = data;
1717         struct e1000_adapter *adapter = netdev_priv(netdev);
1718         struct e1000_hw *hw = &adapter->hw;
1719         u32 rctl, icr = er32(ICR);
1720
1721         if (!icr || test_bit(__E1000_DOWN, &adapter->state))
1722                 return IRQ_NONE;  /* Not our interrupt */
1723
1724         /*
1725          * IMS will not auto-mask if INT_ASSERTED is not set, and if it is
1726          * not set, then the adapter didn't send an interrupt
1727          */
1728         if (!(icr & E1000_ICR_INT_ASSERTED))
1729                 return IRQ_NONE;
1730
1731         /*
1732          * Interrupt Auto-Mask...upon reading ICR,
1733          * interrupts are masked.  No need for the
1734          * IMC write
1735          */
1736
1737         if (icr & E1000_ICR_LSC) {
1738                 hw->mac.get_link_status = true;
1739                 /*
1740                  * ICH8 workaround-- Call gig speed drop workaround on cable
1741                  * disconnect (LSC) before accessing any PHY registers
1742                  */
1743                 if ((adapter->flags & FLAG_LSC_GIG_SPEED_DROP) &&
1744                     (!(er32(STATUS) & E1000_STATUS_LU)))
1745                         schedule_work(&adapter->downshift_task);
1746
1747                 /*
1748                  * 80003ES2LAN workaround--
1749                  * For packet buffer work-around on link down event;
1750                  * disable receives here in the ISR and
1751                  * reset adapter in watchdog
1752                  */
1753                 if (netif_carrier_ok(netdev) &&
1754                     (adapter->flags & FLAG_RX_NEEDS_RESTART)) {
1755                         /* disable receives */
1756                         rctl = er32(RCTL);
1757                         ew32(RCTL, rctl & ~E1000_RCTL_EN);
1758                         adapter->flags |= FLAG_RX_RESTART_NOW;
1759                 }
1760                 /* guard against interrupt when we're going down */
1761                 if (!test_bit(__E1000_DOWN, &adapter->state))
1762                         mod_timer(&adapter->watchdog_timer, jiffies + 1);
1763         }
1764
1765         if (napi_schedule_prep(&adapter->napi)) {
1766                 adapter->total_tx_bytes = 0;
1767                 adapter->total_tx_packets = 0;
1768                 adapter->total_rx_bytes = 0;
1769                 adapter->total_rx_packets = 0;
1770                 __napi_schedule(&adapter->napi);
1771         }
1772
1773         return IRQ_HANDLED;
1774 }
1775
1776 static irqreturn_t e1000_msix_other(int irq, void *data)
1777 {
1778         struct net_device *netdev = data;
1779         struct e1000_adapter *adapter = netdev_priv(netdev);
1780         struct e1000_hw *hw = &adapter->hw;
1781         u32 icr = er32(ICR);
1782
1783         if (!(icr & E1000_ICR_INT_ASSERTED)) {
1784                 if (!test_bit(__E1000_DOWN, &adapter->state))
1785                         ew32(IMS, E1000_IMS_OTHER);
1786                 return IRQ_NONE;
1787         }
1788
1789         if (icr & adapter->eiac_mask)
1790                 ew32(ICS, (icr & adapter->eiac_mask));
1791
1792         if (icr & E1000_ICR_OTHER) {
1793                 if (!(icr & E1000_ICR_LSC))
1794                         goto no_link_interrupt;
1795                 hw->mac.get_link_status = true;
1796                 /* guard against interrupt when we're going down */
1797                 if (!test_bit(__E1000_DOWN, &adapter->state))
1798                         mod_timer(&adapter->watchdog_timer, jiffies + 1);
1799         }
1800
1801 no_link_interrupt:
1802         if (!test_bit(__E1000_DOWN, &adapter->state))
1803                 ew32(IMS, E1000_IMS_LSC | E1000_IMS_OTHER);
1804
1805         return IRQ_HANDLED;
1806 }
1807
1808
1809 static irqreturn_t e1000_intr_msix_tx(int irq, void *data)
1810 {
1811         struct net_device *netdev = data;
1812         struct e1000_adapter *adapter = netdev_priv(netdev);
1813         struct e1000_hw *hw = &adapter->hw;
1814         struct e1000_ring *tx_ring = adapter->tx_ring;
1815
1816
1817         adapter->total_tx_bytes = 0;
1818         adapter->total_tx_packets = 0;
1819
1820         if (!e1000_clean_tx_irq(tx_ring))
1821                 /* Ring was not completely cleaned, so fire another interrupt */
1822                 ew32(ICS, tx_ring->ims_val);
1823
1824         return IRQ_HANDLED;
1825 }
1826
1827 static irqreturn_t e1000_intr_msix_rx(int irq, void *data)
1828 {
1829         struct net_device *netdev = data;
1830         struct e1000_adapter *adapter = netdev_priv(netdev);
1831         struct e1000_ring *rx_ring = adapter->rx_ring;
1832
1833         /* Write the ITR value calculated at the end of the
1834          * previous interrupt.
1835          */
1836         if (rx_ring->set_itr) {
1837                 writel(1000000000 / (rx_ring->itr_val * 256),
1838                        rx_ring->itr_register);
1839                 rx_ring->set_itr = 0;
1840         }
1841
1842         if (napi_schedule_prep(&adapter->napi)) {
1843                 adapter->total_rx_bytes = 0;
1844                 adapter->total_rx_packets = 0;
1845                 __napi_schedule(&adapter->napi);
1846         }
1847         return IRQ_HANDLED;
1848 }
1849
1850 /**
1851  * e1000_configure_msix - Configure MSI-X hardware
1852  *
1853  * e1000_configure_msix sets up the hardware to properly
1854  * generate MSI-X interrupts.
1855  **/
1856 static void e1000_configure_msix(struct e1000_adapter *adapter)
1857 {
1858         struct e1000_hw *hw = &adapter->hw;
1859         struct e1000_ring *rx_ring = adapter->rx_ring;
1860         struct e1000_ring *tx_ring = adapter->tx_ring;
1861         int vector = 0;
1862         u32 ctrl_ext, ivar = 0;
1863
1864         adapter->eiac_mask = 0;
1865
1866         /* Workaround issue with spurious interrupts on 82574 in MSI-X mode */
1867         if (hw->mac.type == e1000_82574) {
1868                 u32 rfctl = er32(RFCTL);
1869                 rfctl |= E1000_RFCTL_ACK_DIS;
1870                 ew32(RFCTL, rfctl);
1871         }
1872
1873 #define E1000_IVAR_INT_ALLOC_VALID      0x8
1874         /* Configure Rx vector */
1875         rx_ring->ims_val = E1000_IMS_RXQ0;
1876         adapter->eiac_mask |= rx_ring->ims_val;
1877         if (rx_ring->itr_val)
1878                 writel(1000000000 / (rx_ring->itr_val * 256),
1879                        rx_ring->itr_register);
1880         else
1881                 writel(1, rx_ring->itr_register);
1882         ivar = E1000_IVAR_INT_ALLOC_VALID | vector;
1883
1884         /* Configure Tx vector */
1885         tx_ring->ims_val = E1000_IMS_TXQ0;
1886         vector++;
1887         if (tx_ring->itr_val)
1888                 writel(1000000000 / (tx_ring->itr_val * 256),
1889                        tx_ring->itr_register);
1890         else
1891                 writel(1, tx_ring->itr_register);
1892         adapter->eiac_mask |= tx_ring->ims_val;
1893         ivar |= ((E1000_IVAR_INT_ALLOC_VALID | vector) << 8);
1894
1895         /* set vector for Other Causes, e.g. link changes */
1896         vector++;
1897         ivar |= ((E1000_IVAR_INT_ALLOC_VALID | vector) << 16);
1898         if (rx_ring->itr_val)
1899                 writel(1000000000 / (rx_ring->itr_val * 256),
1900                        hw->hw_addr + E1000_EITR_82574(vector));
1901         else
1902                 writel(1, hw->hw_addr + E1000_EITR_82574(vector));
1903
1904         /* Cause Tx interrupts on every write back */
1905         ivar |= (1 << 31);
1906
1907         ew32(IVAR, ivar);
1908
1909         /* enable MSI-X PBA support */
1910         ctrl_ext = er32(CTRL_EXT);
1911         ctrl_ext |= E1000_CTRL_EXT_PBA_CLR;
1912
1913         /* Auto-Mask Other interrupts upon ICR read */
1914 #define E1000_EIAC_MASK_82574   0x01F00000
1915         ew32(IAM, ~E1000_EIAC_MASK_82574 | E1000_IMS_OTHER);
1916         ctrl_ext |= E1000_CTRL_EXT_EIAME;
1917         ew32(CTRL_EXT, ctrl_ext);
1918         e1e_flush();
1919 }
1920
1921 void e1000e_reset_interrupt_capability(struct e1000_adapter *adapter)
1922 {
1923         if (adapter->msix_entries) {
1924                 pci_disable_msix(adapter->pdev);
1925                 kfree(adapter->msix_entries);
1926                 adapter->msix_entries = NULL;
1927         } else if (adapter->flags & FLAG_MSI_ENABLED) {
1928                 pci_disable_msi(adapter->pdev);
1929                 adapter->flags &= ~FLAG_MSI_ENABLED;
1930         }
1931 }
1932
1933 /**
1934  * e1000e_set_interrupt_capability - set MSI or MSI-X if supported
1935  *
1936  * Attempt to configure interrupts using the best available
1937  * capabilities of the hardware and kernel.
1938  **/
1939 void e1000e_set_interrupt_capability(struct e1000_adapter *adapter)
1940 {
1941         int err;
1942         int i;
1943
1944         switch (adapter->int_mode) {
1945         case E1000E_INT_MODE_MSIX:
1946                 if (adapter->flags & FLAG_HAS_MSIX) {
1947                         adapter->num_vectors = 3; /* RxQ0, TxQ0 and other */
1948                         adapter->msix_entries = kcalloc(adapter->num_vectors,
1949                                                       sizeof(struct msix_entry),
1950                                                       GFP_KERNEL);
1951                         if (adapter->msix_entries) {
1952                                 for (i = 0; i < adapter->num_vectors; i++)
1953                                         adapter->msix_entries[i].entry = i;
1954
1955                                 err = pci_enable_msix(adapter->pdev,
1956                                                       adapter->msix_entries,
1957                                                       adapter->num_vectors);
1958                                 if (err == 0)
1959                                         return;
1960                         }
1961                         /* MSI-X failed, so fall through and try MSI */
1962                         e_err("Failed to initialize MSI-X interrupts.  Falling back to MSI interrupts.\n");
1963                         e1000e_reset_interrupt_capability(adapter);
1964                 }
1965                 adapter->int_mode = E1000E_INT_MODE_MSI;
1966                 /* Fall through */
1967         case E1000E_INT_MODE_MSI:
1968                 if (!pci_enable_msi(adapter->pdev)) {
1969                         adapter->flags |= FLAG_MSI_ENABLED;
1970                 } else {
1971                         adapter->int_mode = E1000E_INT_MODE_LEGACY;
1972                         e_err("Failed to initialize MSI interrupts.  Falling back to legacy interrupts.\n");
1973                 }
1974                 /* Fall through */
1975         case E1000E_INT_MODE_LEGACY:
1976                 /* Don't do anything; this is the system default */
1977                 break;
1978         }
1979
1980         /* store the number of vectors being used */
1981         adapter->num_vectors = 1;
1982 }
1983
1984 /**
1985  * e1000_request_msix - Initialize MSI-X interrupts
1986  *
1987  * e1000_request_msix allocates MSI-X vectors and requests interrupts from the
1988  * kernel.
1989  **/
1990 static int e1000_request_msix(struct e1000_adapter *adapter)
1991 {
1992         struct net_device *netdev = adapter->netdev;
1993         int err = 0, vector = 0;
1994
1995         if (strlen(netdev->name) < (IFNAMSIZ - 5))
1996                 snprintf(adapter->rx_ring->name,
1997                          sizeof(adapter->rx_ring->name) - 1,
1998                          "%s-rx-0", netdev->name);
1999         else
2000                 memcpy(adapter->rx_ring->name, netdev->name, IFNAMSIZ);
2001         err = request_irq(adapter->msix_entries[vector].vector,
2002                           e1000_intr_msix_rx, 0, adapter->rx_ring->name,
2003                           netdev);
2004         if (err)
2005                 return err;
2006         adapter->rx_ring->itr_register = adapter->hw.hw_addr +
2007             E1000_EITR_82574(vector);
2008         adapter->rx_ring->itr_val = adapter->itr;
2009         vector++;
2010
2011         if (strlen(netdev->name) < (IFNAMSIZ - 5))
2012                 snprintf(adapter->tx_ring->name,
2013                          sizeof(adapter->tx_ring->name) - 1,
2014                          "%s-tx-0", netdev->name);
2015         else
2016                 memcpy(adapter->tx_ring->name, netdev->name, IFNAMSIZ);
2017         err = request_irq(adapter->msix_entries[vector].vector,
2018                           e1000_intr_msix_tx, 0, adapter->tx_ring->name,
2019                           netdev);
2020         if (err)
2021                 return err;
2022         adapter->tx_ring->itr_register = adapter->hw.hw_addr +
2023             E1000_EITR_82574(vector);
2024         adapter->tx_ring->itr_val = adapter->itr;
2025         vector++;
2026
2027         err = request_irq(adapter->msix_entries[vector].vector,
2028                           e1000_msix_other, 0, netdev->name, netdev);
2029         if (err)
2030                 return err;
2031
2032         e1000_configure_msix(adapter);
2033
2034         return 0;
2035 }
2036
2037 /**
2038  * e1000_request_irq - initialize interrupts
2039  *
2040  * Attempts to configure interrupts using the best available
2041  * capabilities of the hardware and kernel.
2042  **/
2043 static int e1000_request_irq(struct e1000_adapter *adapter)
2044 {
2045         struct net_device *netdev = adapter->netdev;
2046         int err;
2047
2048         if (adapter->msix_entries) {
2049                 err = e1000_request_msix(adapter);
2050                 if (!err)
2051                         return err;
2052                 /* fall back to MSI */
2053                 e1000e_reset_interrupt_capability(adapter);
2054                 adapter->int_mode = E1000E_INT_MODE_MSI;
2055                 e1000e_set_interrupt_capability(adapter);
2056         }
2057         if (adapter->flags & FLAG_MSI_ENABLED) {
2058                 err = request_irq(adapter->pdev->irq, e1000_intr_msi, 0,
2059                                   netdev->name, netdev);
2060                 if (!err)
2061                         return err;
2062
2063                 /* fall back to legacy interrupt */
2064                 e1000e_reset_interrupt_capability(adapter);
2065                 adapter->int_mode = E1000E_INT_MODE_LEGACY;
2066         }
2067
2068         err = request_irq(adapter->pdev->irq, e1000_intr, IRQF_SHARED,
2069                           netdev->name, netdev);
2070         if (err)
2071                 e_err("Unable to allocate interrupt, Error: %d\n", err);
2072
2073         return err;
2074 }
2075
2076 static void e1000_free_irq(struct e1000_adapter *adapter)
2077 {
2078         struct net_device *netdev = adapter->netdev;
2079
2080         if (adapter->msix_entries) {
2081                 int vector = 0;
2082
2083                 free_irq(adapter->msix_entries[vector].vector, netdev);
2084                 vector++;
2085
2086                 free_irq(adapter->msix_entries[vector].vector, netdev);
2087                 vector++;
2088
2089                 /* Other Causes interrupt vector */
2090                 free_irq(adapter->msix_entries[vector].vector, netdev);
2091                 return;
2092         }
2093
2094         free_irq(adapter->pdev->irq, netdev);
2095 }
2096
2097 /**
2098  * e1000_irq_disable - Mask off interrupt generation on the NIC
2099  **/
2100 static void e1000_irq_disable(struct e1000_adapter *adapter)
2101 {
2102         struct e1000_hw *hw = &adapter->hw;
2103
2104         ew32(IMC, ~0);
2105         if (adapter->msix_entries)
2106                 ew32(EIAC_82574, 0);
2107         e1e_flush();
2108
2109         if (adapter->msix_entries) {
2110                 int i;
2111                 for (i = 0; i < adapter->num_vectors; i++)
2112                         synchronize_irq(adapter->msix_entries[i].vector);
2113         } else {
2114                 synchronize_irq(adapter->pdev->irq);
2115         }
2116 }
2117
2118 /**
2119  * e1000_irq_enable - Enable default interrupt generation settings
2120  **/
2121 static void e1000_irq_enable(struct e1000_adapter *adapter)
2122 {
2123         struct e1000_hw *hw = &adapter->hw;
2124
2125         if (adapter->msix_entries) {
2126                 ew32(EIAC_82574, adapter->eiac_mask & E1000_EIAC_MASK_82574);
2127                 ew32(IMS, adapter->eiac_mask | E1000_IMS_OTHER | E1000_IMS_LSC);
2128         } else {
2129                 ew32(IMS, IMS_ENABLE_MASK);
2130         }
2131         e1e_flush();
2132 }
2133
2134 /**
2135  * e1000e_get_hw_control - get control of the h/w from f/w
2136  * @adapter: address of board private structure
2137  *
2138  * e1000e_get_hw_control sets {CTRL_EXT|SWSM}:DRV_LOAD bit.
2139  * For ASF and Pass Through versions of f/w this means that
2140  * the driver is loaded. For AMT version (only with 82573)
2141  * of the f/w this means that the network i/f is open.
2142  **/
2143 void e1000e_get_hw_control(struct e1000_adapter *adapter)
2144 {
2145         struct e1000_hw *hw = &adapter->hw;
2146         u32 ctrl_ext;
2147         u32 swsm;
2148
2149         /* Let firmware know the driver has taken over */
2150         if (adapter->flags & FLAG_HAS_SWSM_ON_LOAD) {
2151                 swsm = er32(SWSM);
2152                 ew32(SWSM, swsm | E1000_SWSM_DRV_LOAD);
2153         } else if (adapter->flags & FLAG_HAS_CTRLEXT_ON_LOAD) {
2154                 ctrl_ext = er32(CTRL_EXT);
2155                 ew32(CTRL_EXT, ctrl_ext | E1000_CTRL_EXT_DRV_LOAD);
2156         }
2157 }
2158
2159 /**
2160  * e1000e_release_hw_control - release control of the h/w to f/w
2161  * @adapter: address of board private structure
2162  *
2163  * e1000e_release_hw_control resets {CTRL_EXT|SWSM}:DRV_LOAD bit.
2164  * For ASF and Pass Through versions of f/w this means that the
2165  * driver is no longer loaded. For AMT version (only with 82573) i
2166  * of the f/w this means that the network i/f is closed.
2167  *
2168  **/
2169 void e1000e_release_hw_control(struct e1000_adapter *adapter)
2170 {
2171         struct e1000_hw *hw = &adapter->hw;
2172         u32 ctrl_ext;
2173         u32 swsm;
2174
2175         /* Let firmware taken over control of h/w */
2176         if (adapter->flags & FLAG_HAS_SWSM_ON_LOAD) {
2177                 swsm = er32(SWSM);
2178                 ew32(SWSM, swsm & ~E1000_SWSM_DRV_LOAD);
2179         } else if (adapter->flags & FLAG_HAS_CTRLEXT_ON_LOAD) {
2180                 ctrl_ext = er32(CTRL_EXT);
2181                 ew32(CTRL_EXT, ctrl_ext & ~E1000_CTRL_EXT_DRV_LOAD);
2182         }
2183 }
2184
2185 /**
2186  * @e1000_alloc_ring - allocate memory for a ring structure
2187  **/
2188 static int e1000_alloc_ring_dma(struct e1000_adapter *adapter,
2189                                 struct e1000_ring *ring)
2190 {
2191         struct pci_dev *pdev = adapter->pdev;
2192
2193         ring->desc = dma_alloc_coherent(&pdev->dev, ring->size, &ring->dma,
2194                                         GFP_KERNEL);
2195         if (!ring->desc)
2196                 return -ENOMEM;
2197
2198         return 0;
2199 }
2200
2201 /**
2202  * e1000e_setup_tx_resources - allocate Tx resources (Descriptors)
2203  * @tx_ring: Tx descriptor ring
2204  *
2205  * Return 0 on success, negative on failure
2206  **/
2207 int e1000e_setup_tx_resources(struct e1000_ring *tx_ring)
2208 {
2209         struct e1000_adapter *adapter = tx_ring->adapter;
2210         int err = -ENOMEM, size;
2211
2212         size = sizeof(struct e1000_buffer) * tx_ring->count;
2213         tx_ring->buffer_info = vzalloc(size);
2214         if (!tx_ring->buffer_info)
2215                 goto err;
2216
2217         /* round up to nearest 4K */
2218         tx_ring->size = tx_ring->count * sizeof(struct e1000_tx_desc);
2219         tx_ring->size = ALIGN(tx_ring->size, 4096);
2220
2221         err = e1000_alloc_ring_dma(adapter, tx_ring);
2222         if (err)
2223                 goto err;
2224
2225         tx_ring->next_to_use = 0;
2226         tx_ring->next_to_clean = 0;
2227
2228         return 0;
2229 err:
2230         vfree(tx_ring->buffer_info);
2231         e_err("Unable to allocate memory for the transmit descriptor ring\n");
2232         return err;
2233 }
2234
2235 /**
2236  * e1000e_setup_rx_resources - allocate Rx resources (Descriptors)
2237  * @rx_ring: Rx descriptor ring
2238  *
2239  * Returns 0 on success, negative on failure
2240  **/
2241 int e1000e_setup_rx_resources(struct e1000_ring *rx_ring)
2242 {
2243         struct e1000_adapter *adapter = rx_ring->adapter;
2244         struct e1000_buffer *buffer_info;
2245         int i, size, desc_len, err = -ENOMEM;
2246
2247         size = sizeof(struct e1000_buffer) * rx_ring->count;
2248         rx_ring->buffer_info = vzalloc(size);
2249         if (!rx_ring->buffer_info)
2250                 goto err;
2251
2252         for (i = 0; i < rx_ring->count; i++) {
2253                 buffer_info = &rx_ring->buffer_info[i];
2254                 buffer_info->ps_pages = kcalloc(PS_PAGE_BUFFERS,
2255                                                 sizeof(struct e1000_ps_page),
2256                                                 GFP_KERNEL);
2257                 if (!buffer_info->ps_pages)
2258                         goto err_pages;
2259         }
2260
2261         desc_len = sizeof(union e1000_rx_desc_packet_split);
2262
2263         /* Round up to nearest 4K */
2264         rx_ring->size = rx_ring->count * desc_len;
2265         rx_ring->size = ALIGN(rx_ring->size, 4096);
2266
2267         err = e1000_alloc_ring_dma(adapter, rx_ring);
2268         if (err)
2269                 goto err_pages;
2270
2271         rx_ring->next_to_clean = 0;
2272         rx_ring->next_to_use = 0;
2273         rx_ring->rx_skb_top = NULL;
2274
2275         return 0;
2276
2277 err_pages:
2278         for (i = 0; i < rx_ring->count; i++) {
2279                 buffer_info = &rx_ring->buffer_info[i];
2280                 kfree(buffer_info->ps_pages);
2281         }
2282 err:
2283         vfree(rx_ring->buffer_info);
2284         e_err("Unable to allocate memory for the receive descriptor ring\n");
2285         return err;
2286 }
2287
2288 /**
2289  * e1000_clean_tx_ring - Free Tx Buffers
2290  * @tx_ring: Tx descriptor ring
2291  **/
2292 static void e1000_clean_tx_ring(struct e1000_ring *tx_ring)
2293 {
2294         struct e1000_adapter *adapter = tx_ring->adapter;
2295         struct e1000_buffer *buffer_info;
2296         unsigned long size;
2297         unsigned int i;
2298
2299         for (i = 0; i < tx_ring->count; i++) {
2300                 buffer_info = &tx_ring->buffer_info[i];
2301                 e1000_put_txbuf(tx_ring, buffer_info);
2302         }
2303
2304         netdev_reset_queue(adapter->netdev);
2305         size = sizeof(struct e1000_buffer) * tx_ring->count;
2306         memset(tx_ring->buffer_info, 0, size);
2307
2308         memset(tx_ring->desc, 0, tx_ring->size);
2309
2310         tx_ring->next_to_use = 0;
2311         tx_ring->next_to_clean = 0;
2312
2313         writel(0, tx_ring->head);
2314         writel(0, tx_ring->tail);
2315 }
2316
2317 /**
2318  * e1000e_free_tx_resources - Free Tx Resources per Queue
2319  * @tx_ring: Tx descriptor ring
2320  *
2321  * Free all transmit software resources
2322  **/
2323 void e1000e_free_tx_resources(struct e1000_ring *tx_ring)
2324 {
2325         struct e1000_adapter *adapter = tx_ring->adapter;
2326         struct pci_dev *pdev = adapter->pdev;
2327
2328         e1000_clean_tx_ring(tx_ring);
2329
2330         vfree(tx_ring->buffer_info);
2331         tx_ring->buffer_info = NULL;
2332
2333         dma_free_coherent(&pdev->dev, tx_ring->size, tx_ring->desc,
2334                           tx_ring->dma);
2335         tx_ring->desc = NULL;
2336 }
2337
2338 /**
2339  * e1000e_free_rx_resources - Free Rx Resources
2340  * @rx_ring: Rx descriptor ring
2341  *
2342  * Free all receive software resources
2343  **/
2344 void e1000e_free_rx_resources(struct e1000_ring *rx_ring)
2345 {
2346         struct e1000_adapter *adapter = rx_ring->adapter;
2347         struct pci_dev *pdev = adapter->pdev;
2348         int i;
2349
2350         e1000_clean_rx_ring(rx_ring);
2351
2352         for (i = 0; i < rx_ring->count; i++)
2353                 kfree(rx_ring->buffer_info[i].ps_pages);
2354
2355         vfree(rx_ring->buffer_info);
2356         rx_ring->buffer_info = NULL;
2357
2358         dma_free_coherent(&pdev->dev, rx_ring->size, rx_ring->desc,
2359                           rx_ring->dma);
2360         rx_ring->desc = NULL;
2361 }
2362
2363 /**
2364  * e1000_update_itr - update the dynamic ITR value based on statistics
2365  * @adapter: pointer to adapter
2366  * @itr_setting: current adapter->itr
2367  * @packets: the number of packets during this measurement interval
2368  * @bytes: the number of bytes during this measurement interval
2369  *
2370  *      Stores a new ITR value based on packets and byte
2371  *      counts during the last interrupt.  The advantage of per interrupt
2372  *      computation is faster updates and more accurate ITR for the current
2373  *      traffic pattern.  Constants in this function were computed
2374  *      based on theoretical maximum wire speed and thresholds were set based
2375  *      on testing data as well as attempting to minimize response time
2376  *      while increasing bulk throughput.  This functionality is controlled
2377  *      by the InterruptThrottleRate module parameter.
2378  **/
2379 static unsigned int e1000_update_itr(struct e1000_adapter *adapter,
2380                                      u16 itr_setting, int packets,
2381                                      int bytes)
2382 {
2383         unsigned int retval = itr_setting;
2384
2385         if (packets == 0)
2386                 return itr_setting;
2387
2388         switch (itr_setting) {
2389         case lowest_latency:
2390                 /* handle TSO and jumbo frames */
2391                 if (bytes/packets > 8000)
2392                         retval = bulk_latency;
2393                 else if ((packets < 5) && (bytes > 512))
2394                         retval = low_latency;
2395                 break;
2396         case low_latency:  /* 50 usec aka 20000 ints/s */
2397                 if (bytes > 10000) {
2398                         /* this if handles the TSO accounting */
2399                         if (bytes/packets > 8000)
2400                                 retval = bulk_latency;
2401                         else if ((packets < 10) || ((bytes/packets) > 1200))
2402                                 retval = bulk_latency;
2403                         else if ((packets > 35))
2404                                 retval = lowest_latency;
2405                 } else if (bytes/packets > 2000) {
2406                         retval = bulk_latency;
2407                 } else if (packets <= 2 && bytes < 512) {
2408                         retval = lowest_latency;
2409                 }
2410                 break;
2411         case bulk_latency: /* 250 usec aka 4000 ints/s */
2412                 if (bytes > 25000) {
2413                         if (packets > 35)
2414                                 retval = low_latency;
2415                 } else if (bytes < 6000) {
2416                         retval = low_latency;
2417                 }
2418                 break;
2419         }
2420
2421         return retval;
2422 }
2423
2424 static void e1000_set_itr(struct e1000_adapter *adapter)
2425 {
2426         struct e1000_hw *hw = &adapter->hw;
2427         u16 current_itr;
2428         u32 new_itr = adapter->itr;
2429
2430         /* for non-gigabit speeds, just fix the interrupt rate at 4000 */
2431         if (adapter->link_speed != SPEED_1000) {
2432                 current_itr = 0;
2433                 new_itr = 4000;
2434                 goto set_itr_now;
2435         }
2436
2437         if (adapter->flags2 & FLAG2_DISABLE_AIM) {
2438                 new_itr = 0;
2439                 goto set_itr_now;
2440         }
2441
2442         adapter->tx_itr = e1000_update_itr(adapter,
2443                                     adapter->tx_itr,
2444                                     adapter->total_tx_packets,
2445                                     adapter->total_tx_bytes);
2446         /* conservative mode (itr 3) eliminates the lowest_latency setting */
2447         if (adapter->itr_setting == 3 && adapter->tx_itr == lowest_latency)
2448                 adapter->tx_itr = low_latency;
2449
2450         adapter->rx_itr = e1000_update_itr(adapter,
2451                                     adapter->rx_itr,
2452                                     adapter->total_rx_packets,
2453                                     adapter->total_rx_bytes);
2454         /* conservative mode (itr 3) eliminates the lowest_latency setting */
2455         if (adapter->itr_setting == 3 && adapter->rx_itr == lowest_latency)
2456                 adapter->rx_itr = low_latency;
2457
2458         current_itr = max(adapter->rx_itr, adapter->tx_itr);
2459
2460         switch (current_itr) {
2461         /* counts and packets in update_itr are dependent on these numbers */
2462         case lowest_latency:
2463                 new_itr = 70000;
2464                 break;
2465         case low_latency:
2466                 new_itr = 20000; /* aka hwitr = ~200 */
2467                 break;
2468         case bulk_latency:
2469                 new_itr = 4000;
2470                 break;
2471         default:
2472                 break;
2473         }
2474
2475 set_itr_now:
2476         if (new_itr != adapter->itr) {
2477                 /*
2478                  * this attempts to bias the interrupt rate towards Bulk
2479                  * by adding intermediate steps when interrupt rate is
2480                  * increasing
2481                  */
2482                 new_itr = new_itr > adapter->itr ?
2483                              min(adapter->itr + (new_itr >> 2), new_itr) :
2484                              new_itr;
2485                 adapter->itr = new_itr;
2486                 adapter->rx_ring->itr_val = new_itr;
2487                 if (adapter->msix_entries)
2488                         adapter->rx_ring->set_itr = 1;
2489                 else
2490                         if (new_itr)
2491                                 ew32(ITR, 1000000000 / (new_itr * 256));
2492                         else
2493                                 ew32(ITR, 0);
2494         }
2495 }
2496
2497 /**
2498  * e1000_alloc_queues - Allocate memory for all rings
2499  * @adapter: board private structure to initialize
2500  **/
2501 static int __devinit e1000_alloc_queues(struct e1000_adapter *adapter)
2502 {
2503         int size = sizeof(struct e1000_ring);
2504
2505         adapter->tx_ring = kzalloc(size, GFP_KERNEL);
2506         if (!adapter->tx_ring)
2507                 goto err;
2508         adapter->tx_ring->count = adapter->tx_ring_count;
2509         adapter->tx_ring->adapter = adapter;
2510
2511         adapter->rx_ring = kzalloc(size, GFP_KERNEL);
2512         if (!adapter->rx_ring)
2513                 goto err;
2514         adapter->rx_ring->count = adapter->rx_ring_count;
2515         adapter->rx_ring->adapter = adapter;
2516
2517         return 0;
2518 err:
2519         e_err("Unable to allocate memory for queues\n");
2520         kfree(adapter->rx_ring);
2521         kfree(adapter->tx_ring);
2522         return -ENOMEM;
2523 }
2524
2525 /**
2526  * e1000_clean - NAPI Rx polling callback
2527  * @napi: struct associated with this polling callback
2528  * @budget: amount of packets driver is allowed to process this poll
2529  **/
2530 static int e1000_clean(struct napi_struct *napi, int budget)
2531 {
2532         struct e1000_adapter *adapter = container_of(napi, struct e1000_adapter, napi);
2533         struct e1000_hw *hw = &adapter->hw;
2534         struct net_device *poll_dev = adapter->netdev;
2535         int tx_cleaned = 1, work_done = 0;
2536
2537         adapter = netdev_priv(poll_dev);
2538
2539         if (adapter->msix_entries &&
2540             !(adapter->rx_ring->ims_val & adapter->tx_ring->ims_val))
2541                 goto clean_rx;
2542
2543         tx_cleaned = e1000_clean_tx_irq(adapter->tx_ring);
2544
2545 clean_rx:
2546         adapter->clean_rx(adapter->rx_ring, &work_done, budget);
2547
2548         if (!tx_cleaned)
2549                 work_done = budget;
2550
2551         /* If budget not fully consumed, exit the polling mode */
2552         if (work_done < budget) {
2553                 if (adapter->itr_setting & 3)
2554                         e1000_set_itr(adapter);
2555                 napi_complete(napi);
2556                 if (!test_bit(__E1000_DOWN, &adapter->state)) {
2557                         if (adapter->msix_entries)
2558                                 ew32(IMS, adapter->rx_ring->ims_val);
2559                         else
2560                                 e1000_irq_enable(adapter);
2561                 }
2562         }
2563
2564         return work_done;
2565 }
2566
2567 static int e1000_vlan_rx_add_vid(struct net_device *netdev, u16 vid)
2568 {
2569         struct e1000_adapter *adapter = netdev_priv(netdev);
2570         struct e1000_hw *hw = &adapter->hw;
2571         u32 vfta, index;
2572
2573         /* don't update vlan cookie if already programmed */
2574         if ((adapter->hw.mng_cookie.status &
2575              E1000_MNG_DHCP_COOKIE_STATUS_VLAN) &&
2576             (vid == adapter->mng_vlan_id))
2577                 return 0;
2578
2579         /* add VID to filter table */
2580         if (adapter->flags & FLAG_HAS_HW_VLAN_FILTER) {
2581                 index = (vid >> 5) & 0x7F;
2582                 vfta = E1000_READ_REG_ARRAY(hw, E1000_VFTA, index);
2583                 vfta |= (1 << (vid & 0x1F));
2584                 hw->mac.ops.write_vfta(hw, index, vfta);
2585         }
2586
2587         set_bit(vid, adapter->active_vlans);
2588
2589         return 0;
2590 }
2591
2592 static int e1000_vlan_rx_kill_vid(struct net_device *netdev, u16 vid)
2593 {
2594         struct e1000_adapter *adapter = netdev_priv(netdev);
2595         struct e1000_hw *hw = &adapter->hw;
2596         u32 vfta, index;
2597
2598         if ((adapter->hw.mng_cookie.status &
2599              E1000_MNG_DHCP_COOKIE_STATUS_VLAN) &&
2600             (vid == adapter->mng_vlan_id)) {
2601                 /* release control to f/w */
2602                 e1000e_release_hw_control(adapter);
2603                 return 0;
2604         }
2605
2606         /* remove VID from filter table */
2607         if (adapter->flags & FLAG_HAS_HW_VLAN_FILTER) {
2608                 index = (vid >> 5) & 0x7F;
2609                 vfta = E1000_READ_REG_ARRAY(hw, E1000_VFTA, index);
2610                 vfta &= ~(1 << (vid & 0x1F));
2611                 hw->mac.ops.write_vfta(hw, index, vfta);
2612         }
2613
2614         clear_bit(vid, adapter->active_vlans);
2615
2616         return 0;
2617 }
2618
2619 /**
2620  * e1000e_vlan_filter_disable - helper to disable hw VLAN filtering
2621  * @adapter: board private structure to initialize
2622  **/
2623 static void e1000e_vlan_filter_disable(struct e1000_adapter *adapter)
2624 {
2625         struct net_device *netdev = adapter->netdev;
2626         struct e1000_hw *hw = &adapter->hw;
2627         u32 rctl;
2628
2629         if (adapter->flags & FLAG_HAS_HW_VLAN_FILTER) {
2630                 /* disable VLAN receive filtering */
2631                 rctl = er32(RCTL);
2632                 rctl &= ~(E1000_RCTL_VFE | E1000_RCTL_CFIEN);
2633                 ew32(RCTL, rctl);
2634
2635                 if (adapter->mng_vlan_id != (u16)E1000_MNG_VLAN_NONE) {
2636                         e1000_vlan_rx_kill_vid(netdev, adapter->mng_vlan_id);
2637                         adapter->mng_vlan_id = E1000_MNG_VLAN_NONE;
2638                 }
2639         }
2640 }
2641
2642 /**
2643  * e1000e_vlan_filter_enable - helper to enable HW VLAN filtering
2644  * @adapter: board private structure to initialize
2645  **/
2646 static void e1000e_vlan_filter_enable(struct e1000_adapter *adapter)
2647 {
2648         struct e1000_hw *hw = &adapter->hw;
2649         u32 rctl;
2650
2651         if (adapter->flags & FLAG_HAS_HW_VLAN_FILTER) {
2652                 /* enable VLAN receive filtering */
2653                 rctl = er32(RCTL);
2654                 rctl |= E1000_RCTL_VFE;
2655                 rctl &= ~E1000_RCTL_CFIEN;
2656                 ew32(RCTL, rctl);
2657         }
2658 }
2659
2660 /**
2661  * e1000e_vlan_strip_enable - helper to disable HW VLAN stripping
2662  * @adapter: board private structure to initialize
2663  **/
2664 static void e1000e_vlan_strip_disable(struct e1000_adapter *adapter)
2665 {
2666         struct e1000_hw *hw = &adapter->hw;
2667         u32 ctrl;
2668
2669         /* disable VLAN tag insert/strip */
2670         ctrl = er32(CTRL);
2671         ctrl &= ~E1000_CTRL_VME;
2672         ew32(CTRL, ctrl);
2673 }
2674
2675 /**
2676  * e1000e_vlan_strip_enable - helper to enable HW VLAN stripping
2677  * @adapter: board private structure to initialize
2678  **/
2679 static void e1000e_vlan_strip_enable(struct e1000_adapter *adapter)
2680 {
2681         struct e1000_hw *hw = &adapter->hw;
2682         u32 ctrl;
2683
2684         /* enable VLAN tag insert/strip */
2685         ctrl = er32(CTRL);
2686         ctrl |= E1000_CTRL_VME;
2687         ew32(CTRL, ctrl);
2688 }
2689
2690 static void e1000_update_mng_vlan(struct e1000_adapter *adapter)
2691 {
2692         struct net_device *netdev = adapter->netdev;
2693         u16 vid = adapter->hw.mng_cookie.vlan_id;
2694         u16 old_vid = adapter->mng_vlan_id;
2695
2696         if (adapter->hw.mng_cookie.status &
2697             E1000_MNG_DHCP_COOKIE_STATUS_VLAN) {
2698                 e1000_vlan_rx_add_vid(netdev, vid);
2699                 adapter->mng_vlan_id = vid;
2700         }
2701
2702         if ((old_vid != (u16)E1000_MNG_VLAN_NONE) && (vid != old_vid))
2703                 e1000_vlan_rx_kill_vid(netdev, old_vid);
2704 }
2705
2706 static void e1000_restore_vlan(struct e1000_adapter *adapter)
2707 {
2708         u16 vid;
2709
2710         e1000_vlan_rx_add_vid(adapter->netdev, 0);
2711
2712         for_each_set_bit(vid, adapter->active_vlans, VLAN_N_VID)
2713                 e1000_vlan_rx_add_vid(adapter->netdev, vid);
2714 }
2715
2716 static void e1000_init_manageability_pt(struct e1000_adapter *adapter)
2717 {
2718         struct e1000_hw *hw = &adapter->hw;
2719         u32 manc, manc2h, mdef, i, j;
2720
2721         if (!(adapter->flags & FLAG_MNG_PT_ENABLED))
2722                 return;
2723
2724         manc = er32(MANC);
2725
2726         /*
2727          * enable receiving management packets to the host. this will probably
2728          * generate destination unreachable messages from the host OS, but
2729          * the packets will be handled on SMBUS
2730          */
2731         manc |= E1000_MANC_EN_MNG2HOST;
2732         manc2h = er32(MANC2H);
2733
2734         switch (hw->mac.type) {
2735         default:
2736                 manc2h |= (E1000_MANC2H_PORT_623 | E1000_MANC2H_PORT_664);
2737                 break;
2738         case e1000_82574:
2739         case e1000_82583:
2740                 /*
2741                  * Check if IPMI pass-through decision filter already exists;
2742                  * if so, enable it.
2743                  */
2744                 for (i = 0, j = 0; i < 8; i++) {
2745                         mdef = er32(MDEF(i));
2746
2747                         /* Ignore filters with anything other than IPMI ports */
2748                         if (mdef & ~(E1000_MDEF_PORT_623 | E1000_MDEF_PORT_664))
2749                                 continue;
2750
2751                         /* Enable this decision filter in MANC2H */
2752                         if (mdef)
2753                                 manc2h |= (1 << i);
2754
2755                         j |= mdef;
2756                 }
2757
2758                 if (j == (E1000_MDEF_PORT_623 | E1000_MDEF_PORT_664))
2759                         break;
2760
2761                 /* Create new decision filter in an empty filter */
2762                 for (i = 0, j = 0; i < 8; i++)
2763                         if (er32(MDEF(i)) == 0) {
2764                                 ew32(MDEF(i), (E1000_MDEF_PORT_623 |
2765                                                E1000_MDEF_PORT_664));
2766                                 manc2h |= (1 << 1);
2767                                 j++;
2768                                 break;
2769                         }
2770
2771                 if (!j)
2772                         e_warn("Unable to create IPMI pass-through filter\n");
2773                 break;
2774         }
2775
2776         ew32(MANC2H, manc2h);
2777         ew32(MANC, manc);
2778 }
2779
2780 /**
2781  * e1000_configure_tx - Configure Transmit Unit after Reset
2782  * @adapter: board private structure
2783  *
2784  * Configure the Tx unit of the MAC after a reset.
2785  **/
2786 static void e1000_configure_tx(struct e1000_adapter *adapter)
2787 {
2788         struct e1000_hw *hw = &adapter->hw;
2789         struct e1000_ring *tx_ring = adapter->tx_ring;
2790         u64 tdba;
2791         u32 tdlen, tarc;
2792
2793         /* Setup the HW Tx Head and Tail descriptor pointers */
2794         tdba = tx_ring->dma;
2795         tdlen = tx_ring->count * sizeof(struct e1000_tx_desc);
2796         ew32(TDBAL, (tdba & DMA_BIT_MASK(32)));
2797         ew32(TDBAH, (tdba >> 32));
2798         ew32(TDLEN, tdlen);
2799         ew32(TDH, 0);
2800         ew32(TDT, 0);
2801         tx_ring->head = adapter->hw.hw_addr + E1000_TDH;
2802         tx_ring->tail = adapter->hw.hw_addr + E1000_TDT;
2803
2804         /* Set the Tx Interrupt Delay register */
2805         ew32(TIDV, adapter->tx_int_delay);
2806         /* Tx irq moderation */
2807         ew32(TADV, adapter->tx_abs_int_delay);
2808
2809         if (adapter->flags2 & FLAG2_DMA_BURST) {
2810                 u32 txdctl = er32(TXDCTL(0));
2811                 txdctl &= ~(E1000_TXDCTL_PTHRESH | E1000_TXDCTL_HTHRESH |
2812                             E1000_TXDCTL_WTHRESH);
2813                 /*
2814                  * set up some performance related parameters to encourage the
2815                  * hardware to use the bus more efficiently in bursts, depends
2816                  * on the tx_int_delay to be enabled,
2817                  * wthresh = 5 ==> burst write a cacheline (64 bytes) at a time
2818                  * hthresh = 1 ==> prefetch when one or more available
2819                  * pthresh = 0x1f ==> prefetch if internal cache 31 or less
2820                  * BEWARE: this seems to work but should be considered first if
2821                  * there are Tx hangs or other Tx related bugs
2822                  */
2823                 txdctl |= E1000_TXDCTL_DMA_BURST_ENABLE;
2824                 ew32(TXDCTL(0), txdctl);
2825         }
2826         /* erratum work around: set txdctl the same for both queues */
2827         ew32(TXDCTL(1), er32(TXDCTL(0)));
2828
2829         if (adapter->flags & FLAG_TARC_SPEED_MODE_BIT) {
2830                 tarc = er32(TARC(0));
2831                 /*
2832                  * set the speed mode bit, we'll clear it if we're not at
2833                  * gigabit link later
2834                  */
2835 #define SPEED_MODE_BIT (1 << 21)
2836                 tarc |= SPEED_MODE_BIT;
2837                 ew32(TARC(0), tarc);
2838         }
2839
2840         /* errata: program both queues to unweighted RR */
2841         if (adapter->flags & FLAG_TARC_SET_BIT_ZERO) {
2842                 tarc = er32(TARC(0));
2843                 tarc |= 1;
2844                 ew32(TARC(0), tarc);
2845                 tarc = er32(TARC(1));
2846                 tarc |= 1;
2847                 ew32(TARC(1), tarc);
2848         }
2849
2850         /* Setup Transmit Descriptor Settings for eop descriptor */
2851         adapter->txd_cmd = E1000_TXD_CMD_EOP | E1000_TXD_CMD_IFCS;
2852
2853         /* only set IDE if we are delaying interrupts using the timers */
2854         if (adapter->tx_int_delay)
2855                 adapter->txd_cmd |= E1000_TXD_CMD_IDE;
2856
2857         /* enable Report Status bit */
2858         adapter->txd_cmd |= E1000_TXD_CMD_RS;
2859
2860         hw->mac.ops.config_collision_dist(hw);
2861 }
2862
2863 /**
2864  * e1000_setup_rctl - configure the receive control registers
2865  * @adapter: Board private structure
2866  **/
2867 #define PAGE_USE_COUNT(S) (((S) >> PAGE_SHIFT) + \
2868                            (((S) & (PAGE_SIZE - 1)) ? 1 : 0))
2869 static void e1000_setup_rctl(struct e1000_adapter *adapter)
2870 {
2871         struct e1000_hw *hw = &adapter->hw;
2872         u32 rctl, rfctl;
2873         u32 pages = 0;
2874
2875         /* Workaround Si errata on 82579 - configure jumbo frame flow */
2876         if (hw->mac.type == e1000_pch2lan) {
2877                 s32 ret_val;
2878
2879                 if (adapter->netdev->mtu > ETH_DATA_LEN)
2880                         ret_val = e1000_lv_jumbo_workaround_ich8lan(hw, true);
2881                 else
2882                         ret_val = e1000_lv_jumbo_workaround_ich8lan(hw, false);
2883
2884                 if (ret_val)
2885                         e_dbg("failed to enable jumbo frame workaround mode\n");
2886         }
2887
2888         /* Program MC offset vector base */
2889         rctl = er32(RCTL);
2890         rctl &= ~(3 << E1000_RCTL_MO_SHIFT);
2891         rctl |= E1000_RCTL_EN | E1000_RCTL_BAM |
2892                 E1000_RCTL_LBM_NO | E1000_RCTL_RDMTS_HALF |
2893                 (adapter->hw.mac.mc_filter_type << E1000_RCTL_MO_SHIFT);
2894
2895         /* Do not Store bad packets */
2896         rctl &= ~E1000_RCTL_SBP;
2897
2898         /* Enable Long Packet receive */
2899         if (adapter->netdev->mtu <= ETH_DATA_LEN)
2900                 rctl &= ~E1000_RCTL_LPE;
2901         else
2902                 rctl |= E1000_RCTL_LPE;
2903
2904         /* Some systems expect that the CRC is included in SMBUS traffic. The
2905          * hardware strips the CRC before sending to both SMBUS (BMC) and to
2906          * host memory when this is enabled
2907          */
2908         if (adapter->flags2 & FLAG2_CRC_STRIPPING)
2909                 rctl |= E1000_RCTL_SECRC;
2910
2911         /* Workaround Si errata on 82577 PHY - configure IPG for jumbos */
2912         if ((hw->phy.type == e1000_phy_82577) && (rctl & E1000_RCTL_LPE)) {
2913                 u16 phy_data;
2914
2915                 e1e_rphy(hw, PHY_REG(770, 26), &phy_data);
2916                 phy_data &= 0xfff8;
2917                 phy_data |= (1 << 2);
2918                 e1e_wphy(hw, PHY_REG(770, 26), phy_data);
2919
2920                 e1e_rphy(hw, 22, &phy_data);
2921                 phy_data &= 0x0fff;
2922                 phy_data |= (1 << 14);
2923                 e1e_wphy(hw, 0x10, 0x2823);
2924                 e1e_wphy(hw, 0x11, 0x0003);
2925                 e1e_wphy(hw, 22, phy_data);
2926         }
2927
2928         /* Setup buffer sizes */
2929         rctl &= ~E1000_RCTL_SZ_4096;
2930         rctl |= E1000_RCTL_BSEX;
2931         switch (adapter->rx_buffer_len) {
2932         case 2048:
2933         default:
2934                 rctl |= E1000_RCTL_SZ_2048;
2935                 rctl &= ~E1000_RCTL_BSEX;
2936                 break;
2937         case 4096:
2938                 rctl |= E1000_RCTL_SZ_4096;
2939                 break;
2940         case 8192:
2941                 rctl |= E1000_RCTL_SZ_8192;
2942                 break;
2943         case 16384:
2944                 rctl |= E1000_RCTL_SZ_16384;
2945                 break;
2946         }
2947
2948         /* Enable Extended Status in all Receive Descriptors */
2949         rfctl = er32(RFCTL);
2950         rfctl |= E1000_RFCTL_EXTEN;
2951
2952         /*
2953          * 82571 and greater support packet-split where the protocol
2954          * header is placed in skb->data and the packet data is
2955          * placed in pages hanging off of skb_shinfo(skb)->nr_frags.
2956          * In the case of a non-split, skb->data is linearly filled,
2957          * followed by the page buffers.  Therefore, skb->data is
2958          * sized to hold the largest protocol header.
2959          *
2960          * allocations using alloc_page take too long for regular MTU
2961          * so only enable packet split for jumbo frames
2962          *
2963          * Using pages when the page size is greater than 16k wastes
2964          * a lot of memory, since we allocate 3 pages at all times
2965          * per packet.
2966          */
2967         pages = PAGE_USE_COUNT(adapter->netdev->mtu);
2968         if ((pages <= 3) && (PAGE_SIZE <= 16384) && (rctl & E1000_RCTL_LPE))
2969                 adapter->rx_ps_pages = pages;
2970         else
2971                 adapter->rx_ps_pages = 0;
2972
2973         if (adapter->rx_ps_pages) {
2974                 u32 psrctl = 0;
2975
2976                 /*
2977                  * disable packet split support for IPv6 extension headers,
2978                  * because some malformed IPv6 headers can hang the Rx
2979                  */
2980                 rfctl |= (E1000_RFCTL_IPV6_EX_DIS |
2981                           E1000_RFCTL_NEW_IPV6_EXT_DIS);
2982
2983                 /* Enable Packet split descriptors */
2984                 rctl |= E1000_RCTL_DTYP_PS;
2985
2986                 psrctl |= adapter->rx_ps_bsize0 >>
2987                         E1000_PSRCTL_BSIZE0_SHIFT;
2988
2989                 switch (adapter->rx_ps_pages) {
2990                 case 3:
2991                         psrctl |= PAGE_SIZE <<
2992                                 E1000_PSRCTL_BSIZE3_SHIFT;
2993                 case 2:
2994                         psrctl |= PAGE_SIZE <<
2995                                 E1000_PSRCTL_BSIZE2_SHIFT;
2996                 case 1:
2997                         psrctl |= PAGE_SIZE >>
2998                                 E1000_PSRCTL_BSIZE1_SHIFT;
2999                         break;
3000                 }
3001
3002                 ew32(PSRCTL, psrctl);
3003         }
3004
3005         /* This is useful for sniffing bad packets. */
3006         if (adapter->netdev->features & NETIF_F_RXALL) {
3007                 /* UPE and MPE will be handled by normal PROMISC logic
3008                  * in e1000e_set_rx_mode */
3009                 rctl |= (E1000_RCTL_SBP | /* Receive bad packets */
3010                          E1000_RCTL_BAM | /* RX All Bcast Pkts */
3011                          E1000_RCTL_PMCF); /* RX All MAC Ctrl Pkts */
3012
3013                 rctl &= ~(E1000_RCTL_VFE | /* Disable VLAN filter */
3014                           E1000_RCTL_DPF | /* Allow filtered pause */
3015                           E1000_RCTL_CFIEN); /* Dis VLAN CFIEN Filter */
3016                 /* Do not mess with E1000_CTRL_VME, it affects transmit as well,
3017                  * and that breaks VLANs.
3018                  */
3019         }
3020
3021         ew32(RFCTL, rfctl);
3022         ew32(RCTL, rctl);
3023         /* just started the receive unit, no need to restart */
3024         adapter->flags &= ~FLAG_RX_RESTART_NOW;
3025 }
3026
3027 /**
3028  * e1000_configure_rx - Configure Receive Unit after Reset
3029  * @adapter: board private structure
3030  *
3031  * Configure the Rx unit of the MAC after a reset.
3032  **/
3033 static void e1000_configure_rx(struct e1000_adapter *adapter)
3034 {
3035         struct e1000_hw *hw = &adapter->hw;
3036         struct e1000_ring *rx_ring = adapter->rx_ring;
3037         u64 rdba;
3038         u32 rdlen, rctl, rxcsum, ctrl_ext;
3039
3040         if (adapter->rx_ps_pages) {
3041                 /* this is a 32 byte descriptor */
3042                 rdlen = rx_ring->count *
3043                     sizeof(union e1000_rx_desc_packet_split);
3044                 adapter->clean_rx = e1000_clean_rx_irq_ps;
3045                 adapter->alloc_rx_buf = e1000_alloc_rx_buffers_ps;
3046         } else if (adapter->netdev->mtu > ETH_FRAME_LEN + ETH_FCS_LEN) {
3047                 rdlen = rx_ring->count * sizeof(union e1000_rx_desc_extended);
3048                 adapter->clean_rx = e1000_clean_jumbo_rx_irq;
3049                 adapter->alloc_rx_buf = e1000_alloc_jumbo_rx_buffers;
3050         } else {
3051                 rdlen = rx_ring->count * sizeof(union e1000_rx_desc_extended);
3052                 adapter->clean_rx = e1000_clean_rx_irq;
3053                 adapter->alloc_rx_buf = e1000_alloc_rx_buffers;
3054         }
3055
3056         /* disable receives while setting up the descriptors */
3057         rctl = er32(RCTL);
3058         if (!(adapter->flags2 & FLAG2_NO_DISABLE_RX))
3059                 ew32(RCTL, rctl & ~E1000_RCTL_EN);
3060         e1e_flush();
3061         usleep_range(10000, 20000);
3062
3063         if (adapter->flags2 & FLAG2_DMA_BURST) {
3064                 /*
3065                  * set the writeback threshold (only takes effect if the RDTR
3066                  * is set). set GRAN=1 and write back up to 0x4 worth, and
3067                  * enable prefetching of 0x20 Rx descriptors
3068                  * granularity = 01
3069                  * wthresh = 04,
3070                  * hthresh = 04,
3071                  * pthresh = 0x20
3072                  */
3073                 ew32(RXDCTL(0), E1000_RXDCTL_DMA_BURST_ENABLE);
3074                 ew32(RXDCTL(1), E1000_RXDCTL_DMA_BURST_ENABLE);
3075
3076                 /*
3077                  * override the delay timers for enabling bursting, only if
3078                  * the value was not set by the user via module options
3079                  */
3080                 if (adapter->rx_int_delay == DEFAULT_RDTR)
3081                         adapter->rx_int_delay = BURST_RDTR;
3082                 if (adapter->rx_abs_int_delay == DEFAULT_RADV)
3083                         adapter->rx_abs_int_delay = BURST_RADV;
3084         }
3085
3086         /* set the Receive Delay Timer Register */
3087         ew32(RDTR, adapter->rx_int_delay);
3088
3089         /* irq moderation */
3090         ew32(RADV, adapter->rx_abs_int_delay);
3091         if ((adapter->itr_setting != 0) && (adapter->itr != 0))
3092                 ew32(ITR, 1000000000 / (adapter->itr * 256));
3093
3094         ctrl_ext = er32(CTRL_EXT);
3095         /* Auto-Mask interrupts upon ICR access */
3096         ctrl_ext |= E1000_CTRL_EXT_IAME;
3097         ew32(IAM, 0xffffffff);
3098         ew32(CTRL_EXT, ctrl_ext);
3099         e1e_flush();
3100
3101         /*
3102          * Setup the HW Rx Head and Tail Descriptor Pointers and
3103          * the Base and Length of the Rx Descriptor Ring
3104          */
3105         rdba = rx_ring->dma;
3106         ew32(RDBAL, (rdba & DMA_BIT_MASK(32)));
3107         ew32(RDBAH, (rdba >> 32));
3108         ew32(RDLEN, rdlen);
3109         ew32(RDH, 0);
3110         ew32(RDT, 0);
3111         rx_ring->head = adapter->hw.hw_addr + E1000_RDH;
3112         rx_ring->tail = adapter->hw.hw_addr + E1000_RDT;
3113
3114         /* Enable Receive Checksum Offload for TCP and UDP */
3115         rxcsum = er32(RXCSUM);
3116         if (adapter->netdev->features & NETIF_F_RXCSUM) {
3117                 rxcsum |= E1000_RXCSUM_TUOFL;
3118
3119                 /*
3120                  * IPv4 payload checksum for UDP fragments must be
3121                  * used in conjunction with packet-split.
3122                  */
3123                 if (adapter->rx_ps_pages)
3124                         rxcsum |= E1000_RXCSUM_IPPCSE;
3125         } else {
3126                 rxcsum &= ~E1000_RXCSUM_TUOFL;
3127                 /* no need to clear IPPCSE as it defaults to 0 */
3128         }
3129         ew32(RXCSUM, rxcsum);
3130
3131         if (adapter->hw.mac.type == e1000_pch2lan) {
3132                 /*
3133                  * With jumbo frames, excessive C-state transition
3134                  * latencies result in dropped transactions.
3135                  */
3136                 if (adapter->netdev->mtu > ETH_DATA_LEN) {
3137                         u32 rxdctl = er32(RXDCTL(0));
3138                         ew32(RXDCTL(0), rxdctl | 0x3);
3139                         pm_qos_update_request(&adapter->netdev->pm_qos_req, 55);
3140                 } else {
3141                         pm_qos_update_request(&adapter->netdev->pm_qos_req,
3142                                               PM_QOS_DEFAULT_VALUE);
3143                 }
3144         }
3145
3146         /* Enable Receives */
3147         ew32(RCTL, rctl);
3148 }
3149
3150 /**
3151  * e1000e_write_mc_addr_list - write multicast addresses to MTA
3152  * @netdev: network interface device structure
3153  *
3154  * Writes multicast address list to the MTA hash table.
3155  * Returns: -ENOMEM on failure
3156  *                0 on no addresses written
3157  *                X on writing X addresses to MTA
3158  */
3159 static int e1000e_write_mc_addr_list(struct net_device *netdev)
3160 {
3161         struct e1000_adapter *adapter = netdev_priv(netdev);
3162         struct e1000_hw *hw = &adapter->hw;
3163         struct netdev_hw_addr *ha;
3164         u8 *mta_list;
3165         int i;
3166
3167         if (netdev_mc_empty(netdev)) {
3168                 /* nothing to program, so clear mc list */
3169                 hw->mac.ops.update_mc_addr_list(hw, NULL, 0);
3170                 return 0;
3171         }
3172
3173         mta_list = kzalloc(netdev_mc_count(netdev) * ETH_ALEN, GFP_ATOMIC);
3174         if (!mta_list)
3175                 return -ENOMEM;
3176
3177         /* update_mc_addr_list expects a packed array of only addresses. */
3178         i = 0;
3179         netdev_for_each_mc_addr(ha, netdev)
3180                 memcpy(mta_list + (i++ * ETH_ALEN), ha->addr, ETH_ALEN);
3181
3182         hw->mac.ops.update_mc_addr_list(hw, mta_list, i);
3183         kfree(mta_list);
3184
3185         return netdev_mc_count(netdev);
3186 }
3187
3188 /**
3189  * e1000e_write_uc_addr_list - write unicast addresses to RAR table
3190  * @netdev: network interface device structure
3191  *
3192  * Writes unicast address list to the RAR table.
3193  * Returns: -ENOMEM on failure/insufficient address space
3194  *                0 on no addresses written
3195  *                X on writing X addresses to the RAR table
3196  **/
3197 static int e1000e_write_uc_addr_list(struct net_device *netdev)
3198 {
3199         struct e1000_adapter *adapter = netdev_priv(netdev);
3200         struct e1000_hw *hw = &adapter->hw;
3201         unsigned int rar_entries = hw->mac.rar_entry_count;
3202         int count = 0;
3203
3204         /* save a rar entry for our hardware address */
3205         rar_entries--;
3206
3207         /* save a rar entry for the LAA workaround */
3208         if (adapter->flags & FLAG_RESET_OVERWRITES_LAA)
3209                 rar_entries--;
3210
3211         /* return ENOMEM indicating insufficient memory for addresses */
3212         if (netdev_uc_count(netdev) > rar_entries)
3213                 return -ENOMEM;
3214
3215         if (!netdev_uc_empty(netdev) && rar_entries) {
3216                 struct netdev_hw_addr *ha;
3217
3218                 /*
3219                  * write the addresses in reverse order to avoid write
3220                  * combining
3221                  */
3222                 netdev_for_each_uc_addr(ha, netdev) {
3223                         if (!rar_entries)
3224                                 break;
3225                         e1000e_rar_set(hw, ha->addr, rar_entries--);
3226                         count++;
3227                 }
3228         }
3229
3230         /* zero out the remaining RAR entries not used above */
3231         for (; rar_entries > 0; rar_entries--) {
3232                 ew32(RAH(rar_entries), 0);
3233                 ew32(RAL(rar_entries), 0);
3234         }
3235         e1e_flush();
3236
3237         return count;
3238 }
3239
3240 /**
3241  * e1000e_set_rx_mode - secondary unicast, Multicast and Promiscuous mode set
3242  * @netdev: network interface device structure
3243  *
3244  * The ndo_set_rx_mode entry point is called whenever the unicast or multicast
3245  * address list or the network interface flags are updated.  This routine is
3246  * responsible for configuring the hardware for proper unicast, multicast,
3247  * promiscuous mode, and all-multi behavior.
3248  **/
3249 static void e1000e_set_rx_mode(struct net_device *netdev)
3250 {
3251         struct e1000_adapter *adapter = netdev_priv(netdev);
3252         struct e1000_hw *hw = &adapter->hw;
3253         u32 rctl;
3254
3255         /* Check for Promiscuous and All Multicast modes */
3256         rctl = er32(RCTL);
3257
3258         /* clear the affected bits */
3259         rctl &= ~(E1000_RCTL_UPE | E1000_RCTL_MPE);
3260
3261         if (netdev->flags & IFF_PROMISC) {
3262                 rctl |= (E1000_RCTL_UPE | E1000_RCTL_MPE);
3263                 /* Do not hardware filter VLANs in promisc mode */
3264                 e1000e_vlan_filter_disable(adapter);
3265         } else {
3266                 int count;
3267
3268                 if (netdev->flags & IFF_ALLMULTI) {
3269                         rctl |= E1000_RCTL_MPE;
3270                 } else {
3271                         /*
3272                          * Write addresses to the MTA, if the attempt fails
3273                          * then we should just turn on promiscuous mode so
3274                          * that we can at least receive multicast traffic
3275                          */
3276                         count = e1000e_write_mc_addr_list(netdev);
3277                         if (count < 0)
3278                                 rctl |= E1000_RCTL_MPE;
3279                 }
3280                 e1000e_vlan_filter_enable(adapter);
3281                 /*
3282                  * Write addresses to available RAR registers, if there is not
3283                  * sufficient space to store all the addresses then enable
3284                  * unicast promiscuous mode
3285                  */
3286                 count = e1000e_write_uc_addr_list(netdev);
3287                 if (count < 0)
3288                         rctl |= E1000_RCTL_UPE;
3289         }
3290
3291         ew32(RCTL, rctl);
3292
3293         if (netdev->features & NETIF_F_HW_VLAN_RX)
3294                 e1000e_vlan_strip_enable(adapter);
3295         else
3296                 e1000e_vlan_strip_disable(adapter);
3297 }
3298
3299 static void e1000e_setup_rss_hash(struct e1000_adapter *adapter)
3300 {
3301         struct e1000_hw *hw = &adapter->hw;
3302         u32 mrqc, rxcsum;
3303         int i;
3304         static const u32 rsskey[10] = {
3305                 0xda565a6d, 0xc20e5b25, 0x3d256741, 0xb08fa343, 0xcb2bcad0,
3306                 0xb4307bae, 0xa32dcb77, 0x0cf23080, 0x3bb7426a, 0xfa01acbe
3307         };
3308
3309         /* Fill out hash function seed */
3310         for (i = 0; i < 10; i++)
3311                 ew32(RSSRK(i), rsskey[i]);
3312
3313         /* Direct all traffic to queue 0 */
3314         for (i = 0; i < 32; i++)
3315                 ew32(RETA(i), 0);
3316
3317         /*
3318          * Disable raw packet checksumming so that RSS hash is placed in
3319          * descriptor on writeback.
3320          */
3321         rxcsum = er32(RXCSUM);
3322         rxcsum |= E1000_RXCSUM_PCSD;
3323
3324         ew32(RXCSUM, rxcsum);
3325
3326         mrqc = (E1000_MRQC_RSS_FIELD_IPV4 |
3327                 E1000_MRQC_RSS_FIELD_IPV4_TCP |
3328                 E1000_MRQC_RSS_FIELD_IPV6 |
3329                 E1000_MRQC_RSS_FIELD_IPV6_TCP |
3330                 E1000_MRQC_RSS_FIELD_IPV6_TCP_EX);
3331
3332         ew32(MRQC, mrqc);
3333 }
3334
3335 /**
3336  * e1000_configure - configure the hardware for Rx and Tx
3337  * @adapter: private board structure
3338  **/
3339 static void e1000_configure(struct e1000_adapter *adapter)
3340 {
3341         struct e1000_ring *rx_ring = adapter->rx_ring;
3342
3343         e1000e_set_rx_mode(adapter->netdev);
3344
3345         e1000_restore_vlan(adapter);
3346         e1000_init_manageability_pt(adapter);
3347
3348         e1000_configure_tx(adapter);
3349
3350         if (adapter->netdev->features & NETIF_F_RXHASH)
3351                 e1000e_setup_rss_hash(adapter);
3352         e1000_setup_rctl(adapter);
3353         e1000_configure_rx(adapter);
3354         adapter->alloc_rx_buf(rx_ring, e1000_desc_unused(rx_ring), GFP_KERNEL);
3355 }
3356
3357 /**
3358  * e1000e_power_up_phy - restore link in case the phy was powered down
3359  * @adapter: address of board private structure
3360  *
3361  * The phy may be powered down to save power and turn off link when the
3362  * driver is unloaded and wake on lan is not enabled (among others)
3363  * *** this routine MUST be followed by a call to e1000e_reset ***
3364  **/
3365 void e1000e_power_up_phy(struct e1000_adapter *adapter)
3366 {
3367         if (adapter->hw.phy.ops.power_up)
3368                 adapter->hw.phy.ops.power_up(&adapter->hw);
3369
3370         adapter->hw.mac.ops.setup_link(&adapter->hw);
3371 }
3372
3373 /**
3374  * e1000_power_down_phy - Power down the PHY
3375  *
3376  * Power down the PHY so no link is implied when interface is down.
3377  * The PHY cannot be powered down if management or WoL is active.
3378  */
3379 static void e1000_power_down_phy(struct e1000_adapter *adapter)
3380 {
3381         /* WoL is enabled */
3382         if (adapter->wol)
3383                 return;
3384
3385         if (adapter->hw.phy.ops.power_down)
3386                 adapter->hw.phy.ops.power_down(&adapter->hw);
3387 }
3388
3389 /**
3390  * e1000e_reset - bring the hardware into a known good state
3391  *
3392  * This function boots the hardware and enables some settings that
3393  * require a configuration cycle of the hardware - those cannot be
3394  * set/changed during runtime. After reset the device needs to be
3395  * properly configured for Rx, Tx etc.
3396  */
3397 void e1000e_reset(struct e1000_adapter *adapter)
3398 {
3399         struct e1000_mac_info *mac = &adapter->hw.mac;
3400         struct e1000_fc_info *fc = &adapter->hw.fc;
3401         struct e1000_hw *hw = &adapter->hw;
3402         u32 tx_space, min_tx_space, min_rx_space;
3403         u32 pba = adapter->pba;
3404         u16 hwm;
3405
3406         /* reset Packet Buffer Allocation to default */
3407         ew32(PBA, pba);
3408
3409         if (adapter->max_frame_size > ETH_FRAME_LEN + ETH_FCS_LEN) {
3410                 /*
3411                  * To maintain wire speed transmits, the Tx FIFO should be
3412                  * large enough to accommodate two full transmit packets,
3413                  * rounded up to the next 1KB and expressed in KB.  Likewise,
3414                  * the Rx FIFO should be large enough to accommodate at least
3415                  * one full receive packet and is similarly rounded up and
3416                  * expressed in KB.
3417                  */
3418                 pba = er32(PBA);
3419                 /* upper 16 bits has Tx packet buffer allocation size in KB */
3420                 tx_space = pba >> 16;
3421                 /* lower 16 bits has Rx packet buffer allocation size in KB */
3422                 pba &= 0xffff;
3423                 /*
3424                  * the Tx fifo also stores 16 bytes of information about the Tx
3425                  * but don't include ethernet FCS because hardware appends it
3426                  */
3427                 min_tx_space = (adapter->max_frame_size +
3428                                 sizeof(struct e1000_tx_desc) -
3429                                 ETH_FCS_LEN) * 2;
3430                 min_tx_space = ALIGN(min_tx_space, 1024);
3431                 min_tx_space >>= 10;
3432                 /* software strips receive CRC, so leave room for it */
3433                 min_rx_space = adapter->max_frame_size;
3434                 min_rx_space = ALIGN(min_rx_space, 1024);
3435                 min_rx_space >>= 10;
3436
3437                 /*
3438                  * If current Tx allocation is less than the min Tx FIFO size,
3439                  * and the min Tx FIFO size is less than the current Rx FIFO
3440                  * allocation, take space away from current Rx allocation
3441                  */
3442                 if ((tx_space < min_tx_space) &&
3443                     ((min_tx_space - tx_space) < pba)) {
3444                         pba -= min_tx_space - tx_space;
3445
3446                         /*
3447                          * if short on Rx space, Rx wins and must trump Tx
3448                          * adjustment or use Early Receive if available
3449                          */
3450                         if (pba < min_rx_space)
3451                                 pba = min_rx_space;
3452                 }
3453
3454                 ew32(PBA, pba);
3455         }
3456
3457         /*
3458          * flow control settings
3459          *
3460          * The high water mark must be low enough to fit one full frame
3461          * (or the size used for early receive) above it in the Rx FIFO.
3462          * Set it to the lower of:
3463          * - 90% of the Rx FIFO size, and
3464          * - the full Rx FIFO size minus one full frame
3465          */
3466         if (adapter->flags & FLAG_DISABLE_FC_PAUSE_TIME)
3467                 fc->pause_time = 0xFFFF;
3468         else
3469                 fc->pause_time = E1000_FC_PAUSE_TIME;
3470         fc->send_xon = true;
3471         fc->current_mode = fc->requested_mode;
3472
3473         switch (hw->mac.type) {
3474         case e1000_ich9lan:
3475         case e1000_ich10lan:
3476                 if (adapter->netdev->mtu > ETH_DATA_LEN) {
3477                         pba = 14;
3478                         ew32(PBA, pba);
3479                         fc->high_water = 0x2800;
3480                         fc->low_water = fc->high_water - 8;
3481                         break;
3482                 }
3483                 /* fall-through */
3484         default:
3485                 hwm = min(((pba << 10) * 9 / 10),
3486                           ((pba << 10) - adapter->max_frame_size));
3487
3488                 fc->high_water = hwm & E1000_FCRTH_RTH; /* 8-byte granularity */
3489                 fc->low_water = fc->high_water - 8;
3490                 break;
3491         case e1000_pchlan:
3492                 /*
3493                  * Workaround PCH LOM adapter hangs with certain network
3494                  * loads.  If hangs persist, try disabling Tx flow control.
3495                  */
3496                 if (adapter->netdev->mtu > ETH_DATA_LEN) {
3497                         fc->high_water = 0x3500;
3498                         fc->low_water  = 0x1500;
3499                 } else {
3500                         fc->high_water = 0x5000;
3501                         fc->low_water  = 0x3000;
3502                 }
3503                 fc->refresh_time = 0x1000;
3504                 break;
3505         case e1000_pch2lan:
3506                 fc->high_water = 0x05C20;
3507                 fc->low_water = 0x05048;
3508                 fc->pause_time = 0x0650;
3509                 fc->refresh_time = 0x0400;
3510                 if (adapter->netdev->mtu > ETH_DATA_LEN) {
3511                         pba = 14;
3512                         ew32(PBA, pba);
3513                 }
3514                 break;
3515         }
3516
3517         /*
3518          * Disable Adaptive Interrupt Moderation if 2 full packets cannot
3519          * fit in receive buffer.
3520          */
3521         if (adapter->itr_setting & 0x3) {
3522                 if ((adapter->max_frame_size * 2) > (pba << 10)) {
3523                         if (!(adapter->flags2 & FLAG2_DISABLE_AIM)) {
3524                                 dev_info(&adapter->pdev->dev,
3525                                         "Interrupt Throttle Rate turned off\n");
3526                                 adapter->flags2 |= FLAG2_DISABLE_AIM;
3527                                 ew32(ITR, 0);
3528                         }
3529                 } else if (adapter->flags2 & FLAG2_DISABLE_AIM) {
3530                         dev_info(&adapter->pdev->dev,
3531                                  "Interrupt Throttle Rate turned on\n");
3532                         adapter->flags2 &= ~FLAG2_DISABLE_AIM;
3533                         adapter->itr = 20000;
3534                         ew32(ITR, 1000000000 / (adapter->itr * 256));
3535                 }
3536         }
3537
3538         /* Allow time for pending master requests to run */
3539         mac->ops.reset_hw(hw);
3540
3541         /*
3542          * For parts with AMT enabled, let the firmware know
3543          * that the network interface is in control
3544          */
3545         if (adapter->flags & FLAG_HAS_AMT)
3546                 e1000e_get_hw_control(adapter);
3547
3548         ew32(WUC, 0);
3549
3550         if (mac->ops.init_hw(hw))
3551                 e_err("Hardware Error\n");
3552
3553         e1000_update_mng_vlan(adapter);
3554
3555         /* Enable h/w to recognize an 802.1Q VLAN Ethernet packet */
3556         ew32(VET, ETH_P_8021Q);
3557
3558         e1000e_reset_adaptive(hw);
3559
3560         if (!netif_running(adapter->netdev) &&
3561             !test_bit(__E1000_TESTING, &adapter->state)) {
3562                 e1000_power_down_phy(adapter);
3563                 return;
3564         }
3565
3566         e1000_get_phy_info(hw);
3567
3568         if ((adapter->flags & FLAG_HAS_SMART_POWER_DOWN) &&
3569             !(adapter->flags & FLAG_SMART_POWER_DOWN)) {
3570                 u16 phy_data = 0;
3571                 /*
3572                  * speed up time to link by disabling smart power down, ignore
3573                  * the return value of this function because there is nothing
3574                  * different we would do if it failed
3575                  */
3576                 e1e_rphy(hw, IGP02E1000_PHY_POWER_MGMT, &phy_data);
3577                 phy_data &= ~IGP02E1000_PM_SPD;
3578                 e1e_wphy(hw, IGP02E1000_PHY_POWER_MGMT, phy_data);
3579         }
3580 }
3581
3582 int e1000e_up(struct e1000_adapter *adapter)
3583 {
3584         struct e1000_hw *hw = &adapter->hw;
3585
3586         /* hardware has been reset, we need to reload some things */
3587         e1000_configure(adapter);
3588
3589         clear_bit(__E1000_DOWN, &adapter->state);
3590
3591         if (adapter->msix_entries)
3592                 e1000_configure_msix(adapter);
3593         e1000_irq_enable(adapter);
3594
3595         netif_start_queue(adapter->netdev);
3596
3597         /* fire a link change interrupt to start the watchdog */
3598         if (adapter->msix_entries)
3599                 ew32(ICS, E1000_ICS_LSC | E1000_ICR_OTHER);
3600         else
3601                 ew32(ICS, E1000_ICS_LSC);
3602
3603         return 0;
3604 }
3605
3606 static void e1000e_flush_descriptors(struct e1000_adapter *adapter)
3607 {
3608         struct e1000_hw *hw = &adapter->hw;
3609
3610         if (!(adapter->flags2 & FLAG2_DMA_BURST))
3611                 return;
3612
3613         /* flush pending descriptor writebacks to memory */
3614         ew32(TIDV, adapter->tx_int_delay | E1000_TIDV_FPD);
3615         ew32(RDTR, adapter->rx_int_delay | E1000_RDTR_FPD);
3616
3617         /* execute the writes immediately */
3618         e1e_flush();
3619 }
3620
3621 static void e1000e_update_stats(struct e1000_adapter *adapter);
3622
3623 void e1000e_down(struct e1000_adapter *adapter)
3624 {
3625         struct net_device *netdev = adapter->netdev;
3626         struct e1000_hw *hw = &adapter->hw;
3627         u32 tctl, rctl;
3628
3629         /*
3630          * signal that we're down so the interrupt handler does not
3631          * reschedule our watchdog timer
3632          */
3633         set_bit(__E1000_DOWN, &adapter->state);
3634
3635         /* disable receives in the hardware */
3636         rctl = er32(RCTL);
3637         if (!(adapter->flags2 & FLAG2_NO_DISABLE_RX))
3638                 ew32(RCTL, rctl & ~E1000_RCTL_EN);
3639         /* flush and sleep below */
3640
3641         netif_stop_queue(netdev);
3642
3643         /* disable transmits in the hardware */
3644         tctl = er32(TCTL);
3645         tctl &= ~E1000_TCTL_EN;
3646         ew32(TCTL, tctl);
3647
3648         /* flush both disables and wait for them to finish */
3649         e1e_flush();
3650         usleep_range(10000, 20000);
3651
3652         e1000_irq_disable(adapter);
3653
3654         del_timer_sync(&adapter->watchdog_timer);
3655         del_timer_sync(&adapter->phy_info_timer);
3656
3657         netif_carrier_off(netdev);
3658
3659         spin_lock(&adapter->stats64_lock);
3660         e1000e_update_stats(adapter);
3661         spin_unlock(&adapter->stats64_lock);
3662
3663         e1000e_flush_descriptors(adapter);
3664         e1000_clean_tx_ring(adapter->tx_ring);
3665         e1000_clean_rx_ring(adapter->rx_ring);
3666
3667         adapter->link_speed = 0;
3668         adapter->link_duplex = 0;
3669
3670         if (!pci_channel_offline(adapter->pdev))
3671                 e1000e_reset(adapter);
3672
3673         /*
3674          * TODO: for power management, we could drop the link and
3675          * pci_disable_device here.
3676          */
3677 }
3678
3679 void e1000e_reinit_locked(struct e1000_adapter *adapter)
3680 {
3681         might_sleep();
3682         while (test_and_set_bit(__E1000_RESETTING, &adapter->state))
3683                 usleep_range(1000, 2000);
3684         e1000e_down(adapter);
3685         e1000e_up(adapter);
3686         clear_bit(__E1000_RESETTING, &adapter->state);
3687 }
3688
3689 /**
3690  * e1000_sw_init - Initialize general software structures (struct e1000_adapter)
3691  * @adapter: board private structure to initialize
3692  *
3693  * e1000_sw_init initializes the Adapter private data structure.
3694  * Fields are initialized based on PCI device information and
3695  * OS network device settings (MTU size).
3696  **/
3697 static int __devinit e1000_sw_init(struct e1000_adapter *adapter)
3698 {
3699         struct net_device *netdev = adapter->netdev;
3700
3701         adapter->rx_buffer_len = ETH_FRAME_LEN + VLAN_HLEN + ETH_FCS_LEN;
3702         adapter->rx_ps_bsize0 = 128;
3703         adapter->max_frame_size = netdev->mtu + ETH_HLEN + ETH_FCS_LEN;
3704         adapter->min_frame_size = ETH_ZLEN + ETH_FCS_LEN;
3705         adapter->tx_ring_count = E1000_DEFAULT_TXD;
3706         adapter->rx_ring_count = E1000_DEFAULT_RXD;
3707
3708         spin_lock_init(&adapter->stats64_lock);
3709
3710         e1000e_set_interrupt_capability(adapter);
3711
3712         if (e1000_alloc_queues(adapter))
3713                 return -ENOMEM;
3714
3715         /* Explicitly disable IRQ since the NIC can be in any state. */
3716         e1000_irq_disable(adapter);
3717
3718         set_bit(__E1000_DOWN, &adapter->state);
3719         return 0;
3720 }
3721
3722 /**
3723  * e1000_intr_msi_test - Interrupt Handler
3724  * @irq: interrupt number
3725  * @data: pointer to a network interface device structure
3726  **/
3727 static irqreturn_t e1000_intr_msi_test(int irq, void *data)
3728 {
3729         struct net_device *netdev = data;
3730         struct e1000_adapter *adapter = netdev_priv(netdev);
3731         struct e1000_hw *hw = &adapter->hw;
3732         u32 icr = er32(ICR);
3733
3734         e_dbg("icr is %08X\n", icr);
3735         if (icr & E1000_ICR_RXSEQ) {
3736                 adapter->flags &= ~FLAG_MSI_TEST_FAILED;
3737                 wmb();
3738         }
3739
3740         return IRQ_HANDLED;
3741 }
3742
3743 /**
3744  * e1000_test_msi_interrupt - Returns 0 for successful test
3745  * @adapter: board private struct
3746  *
3747  * code flow taken from tg3.c
3748  **/
3749 static int e1000_test_msi_interrupt(struct e1000_adapter *adapter)
3750 {
3751         struct net_device *netdev = adapter->netdev;
3752         struct e1000_hw *hw = &adapter->hw;
3753         int err;
3754
3755         /* poll_enable hasn't been called yet, so don't need disable */
3756         /* clear any pending events */
3757         er32(ICR);
3758
3759         /* free the real vector and request a test handler */
3760         e1000_free_irq(adapter);
3761         e1000e_reset_interrupt_capability(adapter);
3762
3763         /* Assume that the test fails, if it succeeds then the test
3764          * MSI irq handler will unset this flag */
3765         adapter->flags |= FLAG_MSI_TEST_FAILED;
3766
3767         err = pci_enable_msi(adapter->pdev);
3768         if (err)
3769                 goto msi_test_failed;
3770
3771         err = request_irq(adapter->pdev->irq, e1000_intr_msi_test, 0,
3772                           netdev->name, netdev);
3773         if (err) {
3774                 pci_disable_msi(adapter->pdev);
3775                 goto msi_test_failed;
3776         }
3777
3778         wmb();
3779
3780         e1000_irq_enable(adapter);
3781
3782         /* fire an unusual interrupt on the test handler */
3783         ew32(ICS, E1000_ICS_RXSEQ);
3784         e1e_flush();
3785         msleep(50);
3786
3787         e1000_irq_disable(adapter);
3788
3789         rmb();
3790
3791         if (adapter->flags & FLAG_MSI_TEST_FAILED) {
3792                 adapter->int_mode = E1000E_INT_MODE_LEGACY;
3793                 e_info("MSI interrupt test failed, using legacy interrupt.\n");
3794         } else {
3795                 e_dbg("MSI interrupt test succeeded!\n");
3796         }
3797
3798         free_irq(adapter->pdev->irq, netdev);
3799         pci_disable_msi(adapter->pdev);
3800
3801 msi_test_failed:
3802         e1000e_set_interrupt_capability(adapter);
3803         return e1000_request_irq(adapter);
3804 }
3805
3806 /**
3807  * e1000_test_msi - Returns 0 if MSI test succeeds or INTx mode is restored
3808  * @adapter: board private struct
3809  *
3810  * code flow taken from tg3.c, called with e1000 interrupts disabled.
3811  **/
3812 static int e1000_test_msi(struct e1000_adapter *adapter)
3813 {
3814         int err;
3815         u16 pci_cmd;
3816
3817         if (!(adapter->flags & FLAG_MSI_ENABLED))
3818                 return 0;
3819
3820         /* disable SERR in case the MSI write causes a master abort */
3821         pci_read_config_word(adapter->pdev, PCI_COMMAND, &pci_cmd);
3822         if (pci_cmd & PCI_COMMAND_SERR)
3823                 pci_write_config_word(adapter->pdev, PCI_COMMAND,
3824                                       pci_cmd & ~PCI_COMMAND_SERR);
3825
3826         err = e1000_test_msi_interrupt(adapter);
3827
3828         /* re-enable SERR */
3829         if (pci_cmd & PCI_COMMAND_SERR) {
3830                 pci_read_config_word(adapter->pdev, PCI_COMMAND, &pci_cmd);
3831                 pci_cmd |= PCI_COMMAND_SERR;
3832                 pci_write_config_word(adapter->pdev, PCI_COMMAND, pci_cmd);
3833         }
3834
3835         return err;
3836 }
3837
3838 /**
3839  * e1000_open - Called when a network interface is made active
3840  * @netdev: network interface device structure
3841  *
3842  * Returns 0 on success, negative value on failure
3843  *
3844  * The open entry point is called when a network interface is made
3845  * active by the system (IFF_UP).  At this point all resources needed
3846  * for transmit and receive operations are allocated, the interrupt
3847  * handler is registered with the OS, the watchdog timer is started,
3848  * and the stack is notified that the interface is ready.
3849  **/
3850 static int e1000_open(struct net_device *netdev)
3851 {
3852         struct e1000_adapter *adapter = netdev_priv(netdev);
3853         struct e1000_hw *hw = &adapter->hw;
3854         struct pci_dev *pdev = adapter->pdev;
3855         int err;
3856
3857         /* disallow open during test */
3858         if (test_bit(__E1000_TESTING, &adapter->state))
3859                 return -EBUSY;
3860
3861         pm_runtime_get_sync(&pdev->dev);
3862
3863         netif_carrier_off(netdev);
3864
3865         /* allocate transmit descriptors */
3866         err = e1000e_setup_tx_resources(adapter->tx_ring);
3867         if (err)
3868                 goto err_setup_tx;
3869
3870         /* allocate receive descriptors */
3871         err = e1000e_setup_rx_resources(adapter->rx_ring);
3872         if (err)
3873                 goto err_setup_rx;
3874
3875         /*
3876          * If AMT is enabled, let the firmware know that the network
3877          * interface is now open and reset the part to a known state.
3878          */
3879         if (adapter->flags & FLAG_HAS_AMT) {
3880                 e1000e_get_hw_control(adapter);
3881                 e1000e_reset(adapter);
3882         }
3883
3884         e1000e_power_up_phy(adapter);
3885
3886         adapter->mng_vlan_id = E1000_MNG_VLAN_NONE;
3887         if ((adapter->hw.mng_cookie.status &
3888              E1000_MNG_DHCP_COOKIE_STATUS_VLAN))
3889                 e1000_update_mng_vlan(adapter);
3890
3891         /* DMA latency requirement to workaround jumbo issue */
3892         if (adapter->hw.mac.type == e1000_pch2lan)
3893                 pm_qos_add_request(&adapter->netdev->pm_qos_req,
3894                                    PM_QOS_CPU_DMA_LATENCY,
3895                                    PM_QOS_DEFAULT_VALUE);
3896
3897         /*
3898          * before we allocate an interrupt, we must be ready to handle it.
3899          * Setting DEBUG_SHIRQ in the kernel makes it fire an interrupt
3900          * as soon as we call pci_request_irq, so we have to setup our
3901          * clean_rx handler before we do so.
3902          */
3903         e1000_configure(adapter);
3904
3905         err = e1000_request_irq(adapter);
3906         if (err)
3907                 goto err_req_irq;
3908
3909         /*
3910          * Work around PCIe errata with MSI interrupts causing some chipsets to
3911          * ignore e1000e MSI messages, which means we need to test our MSI
3912          * interrupt now
3913          */
3914         if (adapter->int_mode != E1000E_INT_MODE_LEGACY) {
3915                 err = e1000_test_msi(adapter);
3916                 if (err) {
3917                         e_err("Interrupt allocation failed\n");
3918                         goto err_req_irq;
3919                 }
3920         }
3921
3922         /* From here on the code is the same as e1000e_up() */
3923         clear_bit(__E1000_DOWN, &adapter->state);
3924
3925         napi_enable(&adapter->napi);
3926
3927         e1000_irq_enable(adapter);
3928
3929         adapter->tx_hang_recheck = false;
3930         netif_start_queue(netdev);
3931
3932         adapter->idle_check = true;
3933         pm_runtime_put(&pdev->dev);
3934
3935         /* fire a link status change interrupt to start the watchdog */
3936         if (adapter->msix_entries)
3937                 ew32(ICS, E1000_ICS_LSC | E1000_ICR_OTHER);
3938         else
3939                 ew32(ICS, E1000_ICS_LSC);
3940
3941         return 0;
3942
3943 err_req_irq:
3944         e1000e_release_hw_control(adapter);
3945         e1000_power_down_phy(adapter);
3946         e1000e_free_rx_resources(adapter->rx_ring);
3947 err_setup_rx:
3948         e1000e_free_tx_resources(adapter->tx_ring);
3949 err_setup_tx:
3950         e1000e_reset(adapter);
3951         pm_runtime_put_sync(&pdev->dev);
3952
3953         return err;
3954 }
3955
3956 /**
3957  * e1000_close - Disables a network interface
3958  * @netdev: network interface device structure
3959  *
3960  * Returns 0, this is not allowed to fail
3961  *
3962  * The close entry point is called when an interface is de-activated
3963  * by the OS.  The hardware is still under the drivers control, but
3964  * needs to be disabled.  A global MAC reset is issued to stop the
3965  * hardware, and all transmit and receive resources are freed.
3966  **/
3967 static int e1000_close(struct net_device *netdev)
3968 {
3969         struct e1000_adapter *adapter = netdev_priv(netdev);
3970         struct pci_dev *pdev = adapter->pdev;
3971
3972         WARN_ON(test_bit(__E1000_RESETTING, &adapter->state));
3973
3974         pm_runtime_get_sync(&pdev->dev);
3975
3976         napi_disable(&adapter->napi);
3977
3978         if (!test_bit(__E1000_DOWN, &adapter->state)) {
3979                 e1000e_down(adapter);
3980                 e1000_free_irq(adapter);
3981         }
3982         e1000_power_down_phy(adapter);
3983
3984         e1000e_free_tx_resources(adapter->tx_ring);
3985         e1000e_free_rx_resources(adapter->rx_ring);
3986
3987         /*
3988          * kill manageability vlan ID if supported, but not if a vlan with
3989          * the same ID is registered on the host OS (let 8021q kill it)
3990          */
3991         if (adapter->hw.mng_cookie.status &
3992             E1000_MNG_DHCP_COOKIE_STATUS_VLAN)
3993                 e1000_vlan_rx_kill_vid(netdev, adapter->mng_vlan_id);
3994
3995         /*
3996          * If AMT is enabled, let the firmware know that the network
3997          * interface is now closed
3998          */
3999         if ((adapter->flags & FLAG_HAS_AMT) &&
4000             !test_bit(__E1000_TESTING, &adapter->state))
4001                 e1000e_release_hw_control(adapter);
4002
4003         if (adapter->hw.mac.type == e1000_pch2lan)
4004                 pm_qos_remove_request(&adapter->netdev->pm_qos_req);
4005
4006         pm_runtime_put_sync(&pdev->dev);
4007
4008         return 0;
4009 }
4010 /**
4011  * e1000_set_mac - Change the Ethernet Address of the NIC
4012  * @netdev: network interface device structure
4013  * @p: pointer to an address structure
4014  *
4015  * Returns 0 on success, negative on failure
4016  **/
4017 static int e1000_set_mac(struct net_device *netdev, void *p)
4018 {
4019         struct e1000_adapter *adapter = netdev_priv(netdev);
4020         struct sockaddr *addr = p;
4021
4022         if (!is_valid_ether_addr(addr->sa_data))
4023                 return -EADDRNOTAVAIL;
4024
4025         memcpy(netdev->dev_addr, addr->sa_data, netdev->addr_len);
4026         memcpy(adapter->hw.mac.addr, addr->sa_data, netdev->addr_len);
4027
4028         e1000e_rar_set(&adapter->hw, adapter->hw.mac.addr, 0);
4029
4030         if (adapter->flags & FLAG_RESET_OVERWRITES_LAA) {
4031                 /* activate the work around */
4032                 e1000e_set_laa_state_82571(&adapter->hw, 1);
4033
4034                 /*
4035                  * Hold a copy of the LAA in RAR[14] This is done so that
4036                  * between the time RAR[0] gets clobbered  and the time it
4037                  * gets fixed (in e1000_watchdog), the actual LAA is in one
4038                  * of the RARs and no incoming packets directed to this port
4039                  * are dropped. Eventually the LAA will be in RAR[0] and
4040                  * RAR[14]
4041                  */
4042                 e1000e_rar_set(&adapter->hw,
4043                               adapter->hw.mac.addr,
4044                               adapter->hw.mac.rar_entry_count - 1);
4045         }
4046
4047         return 0;
4048 }
4049
4050 /**
4051  * e1000e_update_phy_task - work thread to update phy
4052  * @work: pointer to our work struct
4053  *
4054  * this worker thread exists because we must acquire a
4055  * semaphore to read the phy, which we could msleep while
4056  * waiting for it, and we can't msleep in a timer.
4057  **/
4058 static void e1000e_update_phy_task(struct work_struct *work)
4059 {
4060         struct e1000_adapter *adapter = container_of(work,
4061                                         struct e1000_adapter, update_phy_task);
4062
4063         if (test_bit(__E1000_DOWN, &adapter->state))
4064                 return;
4065
4066         e1000_get_phy_info(&adapter->hw);
4067 }
4068
4069 /*
4070  * Need to wait a few seconds after link up to get diagnostic information from
4071  * the phy
4072  */
4073 static void e1000_update_phy_info(unsigned long data)
4074 {
4075         struct e1000_adapter *adapter = (struct e1000_adapter *) data;
4076
4077         if (test_bit(__E1000_DOWN, &adapter->state))
4078                 return;
4079
4080         schedule_work(&adapter->update_phy_task);
4081 }
4082
4083 /**
4084  * e1000e_update_phy_stats - Update the PHY statistics counters
4085  * @adapter: board private structure
4086  *
4087  * Read/clear the upper 16-bit PHY registers and read/accumulate lower
4088  **/
4089 static void e1000e_update_phy_stats(struct e1000_adapter *adapter)
4090 {
4091         struct e1000_hw *hw = &adapter->hw;
4092         s32 ret_val;
4093         u16 phy_data;
4094
4095         ret_val = hw->phy.ops.acquire(hw);
4096         if (ret_val)
4097                 return;
4098
4099         /*
4100          * A page set is expensive so check if already on desired page.
4101          * If not, set to the page with the PHY status registers.
4102          */
4103         hw->phy.addr = 1;
4104         ret_val = e1000e_read_phy_reg_mdic(hw, IGP01E1000_PHY_PAGE_SELECT,
4105                                            &phy_data);
4106         if (ret_val)
4107                 goto release;
4108         if (phy_data != (HV_STATS_PAGE << IGP_PAGE_SHIFT)) {
4109                 ret_val = hw->phy.ops.set_page(hw,
4110                                                HV_STATS_PAGE << IGP_PAGE_SHIFT);
4111                 if (ret_val)
4112                         goto release;
4113         }
4114
4115         /* Single Collision Count */
4116         hw->phy.ops.read_reg_page(hw, HV_SCC_UPPER, &phy_data);
4117         ret_val = hw->phy.ops.read_reg_page(hw, HV_SCC_LOWER, &phy_data);
4118         if (!ret_val)
4119                 adapter->stats.scc += phy_data;
4120
4121         /* Excessive Collision Count */
4122         hw->phy.ops.read_reg_page(hw, HV_ECOL_UPPER, &phy_data);
4123         ret_val = hw->phy.ops.read_reg_page(hw, HV_ECOL_LOWER, &phy_data);
4124         if (!ret_val)
4125                 adapter->stats.ecol += phy_data;
4126
4127         /* Multiple Collision Count */
4128         hw->phy.ops.read_reg_page(hw, HV_MCC_UPPER, &phy_data);
4129         ret_val = hw->phy.ops.read_reg_page(hw, HV_MCC_LOWER, &phy_data);
4130         if (!ret_val)
4131                 adapter->stats.mcc += phy_data;
4132
4133         /* Late Collision Count */
4134         hw->phy.ops.read_reg_page(hw, HV_LATECOL_UPPER, &phy_data);
4135         ret_val = hw->phy.ops.read_reg_page(hw, HV_LATECOL_LOWER, &phy_data);
4136         if (!ret_val)
4137                 adapter->stats.latecol += phy_data;
4138
4139         /* Collision Count - also used for adaptive IFS */
4140         hw->phy.ops.read_reg_page(hw, HV_COLC_UPPER, &phy_data);
4141         ret_val = hw->phy.ops.read_reg_page(hw, HV_COLC_LOWER, &phy_data);
4142         if (!ret_val)
4143                 hw->mac.collision_delta = phy_data;
4144
4145         /* Defer Count */
4146         hw->phy.ops.read_reg_page(hw, HV_DC_UPPER, &phy_data);
4147         ret_val = hw->phy.ops.read_reg_page(hw, HV_DC_LOWER, &phy_data);
4148         if (!ret_val)
4149                 adapter->stats.dc += phy_data;
4150
4151         /* Transmit with no CRS */
4152         hw->phy.ops.read_reg_page(hw, HV_TNCRS_UPPER, &phy_data);
4153         ret_val = hw->phy.ops.read_reg_page(hw, HV_TNCRS_LOWER, &phy_data);
4154         if (!ret_val)
4155                 adapter->stats.tncrs += phy_data;
4156
4157 release:
4158         hw->phy.ops.release(hw);
4159 }
4160
4161 /**
4162  * e1000e_update_stats - Update the board statistics counters
4163  * @adapter: board private structure
4164  **/
4165 static void e1000e_update_stats(struct e1000_adapter *adapter)
4166 {
4167         struct net_device *netdev = adapter->netdev;
4168         struct e1000_hw *hw = &adapter->hw;
4169         struct pci_dev *pdev = adapter->pdev;
4170
4171         /*
4172          * Prevent stats update while adapter is being reset, or if the pci
4173          * connection is down.
4174          */
4175         if (adapter->link_speed == 0)
4176                 return;
4177         if (pci_channel_offline(pdev))
4178                 return;
4179
4180         adapter->stats.crcerrs += er32(CRCERRS);
4181         adapter->stats.gprc += er32(GPRC);
4182         adapter->stats.gorc += er32(GORCL);
4183         er32(GORCH); /* Clear gorc */
4184         adapter->stats.bprc += er32(BPRC);
4185         adapter->stats.mprc += er32(MPRC);
4186         adapter->stats.roc += er32(ROC);
4187
4188         adapter->stats.mpc += er32(MPC);
4189
4190         /* Half-duplex statistics */
4191         if (adapter->link_duplex == HALF_DUPLEX) {
4192                 if (adapter->flags2 & FLAG2_HAS_PHY_STATS) {
4193                         e1000e_update_phy_stats(adapter);
4194                 } else {
4195                         adapter->stats.scc += er32(SCC);
4196                         adapter->stats.ecol += er32(ECOL);
4197                         adapter->stats.mcc += er32(MCC);
4198                         adapter->stats.latecol += er32(LATECOL);
4199                         adapter->stats.dc += er32(DC);
4200
4201                         hw->mac.collision_delta = er32(COLC);
4202
4203                         if ((hw->mac.type != e1000_82574) &&
4204                             (hw->mac.type != e1000_82583))
4205                                 adapter->stats.tncrs += er32(TNCRS);
4206                 }
4207                 adapter->stats.colc += hw->mac.collision_delta;
4208         }
4209
4210         adapter->stats.xonrxc += er32(XONRXC);
4211         adapter->stats.xontxc += er32(XONTXC);
4212         adapter->stats.xoffrxc += er32(XOFFRXC);
4213         adapter->stats.xofftxc += er32(XOFFTXC);
4214         adapter->stats.gptc += er32(GPTC);
4215         adapter->stats.gotc += er32(GOTCL);
4216         er32(GOTCH); /* Clear gotc */
4217         adapter->stats.rnbc += er32(RNBC);
4218         adapter->stats.ruc += er32(RUC);
4219
4220         adapter->stats.mptc += er32(MPTC);
4221         adapter->stats.bptc += er32(BPTC);
4222
4223         /* used for adaptive IFS */
4224
4225         hw->mac.tx_packet_delta = er32(TPT);
4226         adapter->stats.tpt += hw->mac.tx_packet_delta;
4227
4228         adapter->stats.algnerrc += er32(ALGNERRC);
4229         adapter->stats.rxerrc += er32(RXERRC);
4230         adapter->stats.cexterr += er32(CEXTERR);
4231         adapter->stats.tsctc += er32(TSCTC);
4232         adapter->stats.tsctfc += er32(TSCTFC);
4233
4234         /* Fill out the OS statistics structure */
4235         netdev->stats.multicast = adapter->stats.mprc;
4236         netdev->stats.collisions = adapter->stats.colc;
4237
4238         /* Rx Errors */
4239
4240         /*
4241          * RLEC on some newer hardware can be incorrect so build
4242          * our own version based on RUC and ROC
4243          */
4244         netdev->stats.rx_errors = adapter->stats.rxerrc +
4245                 adapter->stats.crcerrs + adapter->stats.algnerrc +
4246                 adapter->stats.ruc + adapter->stats.roc +
4247                 adapter->stats.cexterr;
4248         netdev->stats.rx_length_errors = adapter->stats.ruc +
4249                                               adapter->stats.roc;
4250         netdev->stats.rx_crc_errors = adapter->stats.crcerrs;
4251         netdev->stats.rx_frame_errors = adapter->stats.algnerrc;
4252         netdev->stats.rx_missed_errors = adapter->stats.mpc;
4253
4254         /* Tx Errors */
4255         netdev->stats.tx_errors = adapter->stats.ecol +
4256                                        adapter->stats.latecol;
4257         netdev->stats.tx_aborted_errors = adapter->stats.ecol;
4258         netdev->stats.tx_window_errors = adapter->stats.latecol;
4259         netdev->stats.tx_carrier_errors = adapter->stats.tncrs;
4260
4261         /* Tx Dropped needs to be maintained elsewhere */
4262
4263         /* Management Stats */
4264         adapter->stats.mgptc += er32(MGTPTC);
4265         adapter->stats.mgprc += er32(MGTPRC);
4266         adapter->stats.mgpdc += er32(MGTPDC);
4267 }
4268
4269 /**
4270  * e1000_phy_read_status - Update the PHY register status snapshot
4271  * @adapter: board private structure
4272  **/
4273 static void e1000_phy_read_status(struct e1000_adapter *adapter)
4274 {
4275         struct e1000_hw *hw = &adapter->hw;
4276         struct e1000_phy_regs *phy = &adapter->phy_regs;
4277
4278         if ((er32(STATUS) & E1000_STATUS_LU) &&
4279             (adapter->hw.phy.media_type == e1000_media_type_copper)) {
4280                 int ret_val;
4281
4282                 ret_val  = e1e_rphy(hw, PHY_CONTROL, &phy->bmcr);
4283                 ret_val |= e1e_rphy(hw, PHY_STATUS, &phy->bmsr);
4284                 ret_val |= e1e_rphy(hw, PHY_AUTONEG_ADV, &phy->advertise);
4285                 ret_val |= e1e_rphy(hw, PHY_LP_ABILITY, &phy->lpa);
4286                 ret_val |= e1e_rphy(hw, PHY_AUTONEG_EXP, &phy->expansion);
4287                 ret_val |= e1e_rphy(hw, PHY_1000T_CTRL, &phy->ctrl1000);
4288                 ret_val |= e1e_rphy(hw, PHY_1000T_STATUS, &phy->stat1000);
4289                 ret_val |= e1e_rphy(hw, PHY_EXT_STATUS, &phy->estatus);
4290                 if (ret_val)
4291                         e_warn("Error reading PHY register\n");
4292         } else {
4293                 /*
4294                  * Do not read PHY registers if link is not up
4295                  * Set values to typical power-on defaults
4296                  */
4297                 phy->bmcr = (BMCR_SPEED1000 | BMCR_ANENABLE | BMCR_FULLDPLX);
4298                 phy->bmsr = (BMSR_100FULL | BMSR_100HALF | BMSR_10FULL |
4299                              BMSR_10HALF | BMSR_ESTATEN | BMSR_ANEGCAPABLE |
4300                              BMSR_ERCAP);
4301                 phy->advertise = (ADVERTISE_PAUSE_ASYM | ADVERTISE_PAUSE_CAP |
4302                                   ADVERTISE_ALL | ADVERTISE_CSMA);
4303                 phy->lpa = 0;
4304                 phy->expansion = EXPANSION_ENABLENPAGE;
4305                 phy->ctrl1000 = ADVERTISE_1000FULL;
4306                 phy->stat1000 = 0;
4307                 phy->estatus = (ESTATUS_1000_TFULL | ESTATUS_1000_THALF);
4308         }
4309 }
4310
4311 static void e1000_print_link_info(struct e1000_adapter *adapter)
4312 {
4313         struct e1000_hw *hw = &adapter->hw;
4314         u32 ctrl = er32(CTRL);
4315
4316         /* Link status message must follow this format for user tools */
4317         printk(KERN_INFO "e1000e: %s NIC Link is Up %d Mbps %s Duplex, Flow Control: %s\n",
4318                 adapter->netdev->name,
4319                 adapter->link_speed,
4320                 adapter->link_duplex == FULL_DUPLEX ? "Full" : "Half",
4321                 (ctrl & E1000_CTRL_TFCE) && (ctrl & E1000_CTRL_RFCE) ? "Rx/Tx" :
4322                 (ctrl & E1000_CTRL_RFCE) ? "Rx" :
4323                 (ctrl & E1000_CTRL_TFCE) ? "Tx" : "None");
4324 }
4325
4326 static bool e1000e_has_link(struct e1000_adapter *adapter)
4327 {
4328         struct e1000_hw *hw = &adapter->hw;
4329         bool link_active = false;
4330         s32 ret_val = 0;
4331
4332         /*
4333          * get_link_status is set on LSC (link status) interrupt or
4334          * Rx sequence error interrupt.  get_link_status will stay
4335          * false until the check_for_link establishes link
4336          * for copper adapters ONLY
4337          */
4338         switch (hw->phy.media_type) {
4339         case e1000_media_type_copper:
4340                 if (hw->mac.get_link_status) {
4341                         ret_val = hw->mac.ops.check_for_link(hw);
4342                         link_active = !hw->mac.get_link_status;
4343                 } else {
4344                         link_active = true;
4345                 }
4346                 break;
4347         case e1000_media_type_fiber:
4348                 ret_val = hw->mac.ops.check_for_link(hw);
4349                 link_active = !!(er32(STATUS) & E1000_STATUS_LU);
4350                 break;
4351         case e1000_media_type_internal_serdes:
4352                 ret_val = hw->mac.ops.check_for_link(hw);
4353                 link_active = adapter->hw.mac.serdes_has_link;
4354                 break;
4355         default:
4356         case e1000_media_type_unknown:
4357                 break;
4358         }
4359
4360         if ((ret_val == E1000_ERR_PHY) && (hw->phy.type == e1000_phy_igp_3) &&
4361             (er32(CTRL) & E1000_PHY_CTRL_GBE_DISABLE)) {
4362                 /* See e1000_kmrn_lock_loss_workaround_ich8lan() */
4363                 e_info("Gigabit has been disabled, downgrading speed\n");
4364         }
4365
4366         return link_active;
4367 }
4368
4369 static void e1000e_enable_receives(struct e1000_adapter *adapter)
4370 {
4371         /* make sure the receive unit is started */
4372         if ((adapter->flags & FLAG_RX_NEEDS_RESTART) &&
4373             (adapter->flags & FLAG_RX_RESTART_NOW)) {
4374                 struct e1000_hw *hw = &adapter->hw;
4375                 u32 rctl = er32(RCTL);
4376                 ew32(RCTL, rctl | E1000_RCTL_EN);
4377                 adapter->flags &= ~FLAG_RX_RESTART_NOW;
4378         }
4379 }
4380
4381 static void e1000e_check_82574_phy_workaround(struct e1000_adapter *adapter)
4382 {
4383         struct e1000_hw *hw = &adapter->hw;
4384
4385         /*
4386          * With 82574 controllers, PHY needs to be checked periodically
4387          * for hung state and reset, if two calls return true
4388          */
4389         if (e1000_check_phy_82574(hw))
4390                 adapter->phy_hang_count++;
4391         else
4392                 adapter->phy_hang_count = 0;
4393
4394         if (adapter->phy_hang_count > 1) {
4395                 adapter->phy_hang_count = 0;
4396                 schedule_work(&adapter->reset_task);
4397         }
4398 }
4399
4400 /**
4401  * e1000_watchdog - Timer Call-back
4402  * @data: pointer to adapter cast into an unsigned long
4403  **/
4404 static void e1000_watchdog(unsigned long data)
4405 {
4406         struct e1000_adapter *adapter = (struct e1000_adapter *) data;
4407
4408         /* Do the rest outside of interrupt context */
4409         schedule_work(&adapter->watchdog_task);
4410
4411         /* TODO: make this use queue_delayed_work() */
4412 }
4413
4414 static void e1000_watchdog_task(struct work_struct *work)
4415 {
4416         struct e1000_adapter *adapter = container_of(work,
4417                                         struct e1000_adapter, watchdog_task);
4418         struct net_device *netdev = adapter->netdev;
4419         struct e1000_mac_info *mac = &adapter->hw.mac;
4420         struct e1000_phy_info *phy = &adapter->hw.phy;
4421         struct e1000_ring *tx_ring = adapter->tx_ring;
4422         struct e1000_hw *hw = &adapter->hw;
4423         u32 link, tctl;
4424
4425         if (test_bit(__E1000_DOWN, &adapter->state))
4426                 return;
4427
4428         link = e1000e_has_link(adapter);
4429         if ((netif_carrier_ok(netdev)) && link) {
4430                 /* Cancel scheduled suspend requests. */
4431                 pm_runtime_resume(netdev->dev.parent);
4432
4433                 e1000e_enable_receives(adapter);
4434                 goto link_up;
4435         }
4436
4437         if ((e1000e_enable_tx_pkt_filtering(hw)) &&
4438             (adapter->mng_vlan_id != adapter->hw.mng_cookie.vlan_id))
4439                 e1000_update_mng_vlan(adapter);
4440
4441         if (link) {
4442                 if (!netif_carrier_ok(netdev)) {
4443                         bool txb2b = true;
4444
4445                         /* Cancel scheduled suspend requests. */
4446                         pm_runtime_resume(netdev->dev.parent);
4447
4448                         /* update snapshot of PHY registers on LSC */
4449                         e1000_phy_read_status(adapter);
4450                         mac->ops.get_link_up_info(&adapter->hw,
4451                                                    &adapter->link_speed,
4452                                                    &adapter->link_duplex);
4453                         e1000_print_link_info(adapter);
4454                         /*
4455                          * On supported PHYs, check for duplex mismatch only
4456                          * if link has autonegotiated at 10/100 half
4457                          */
4458                         if ((hw->phy.type == e1000_phy_igp_3 ||
4459                              hw->phy.type == e1000_phy_bm) &&
4460                             (hw->mac.autoneg == true) &&
4461                             (adapter->link_speed == SPEED_10 ||
4462                              adapter->link_speed == SPEED_100) &&
4463                             (adapter->link_duplex == HALF_DUPLEX)) {
4464                                 u16 autoneg_exp;
4465
4466                                 e1e_rphy(hw, PHY_AUTONEG_EXP, &autoneg_exp);
4467
4468                                 if (!(autoneg_exp & NWAY_ER_LP_NWAY_CAPS))
4469                                         e_info("Autonegotiated half duplex but link partner cannot autoneg.  Try forcing full duplex if link gets many collisions.\n");
4470                         }
4471
4472                         /* adjust timeout factor according to speed/duplex */
4473                         adapter->tx_timeout_factor = 1;
4474                         switch (adapter->link_speed) {
4475                         case SPEED_10:
4476                                 txb2b = false;
4477                                 adapter->tx_timeout_factor = 16;
4478                                 break;
4479                         case SPEED_100:
4480                                 txb2b = false;
4481                                 adapter->tx_timeout_factor = 10;
4482                                 break;
4483                         }
4484
4485                         /*
4486                          * workaround: re-program speed mode bit after
4487                          * link-up event
4488                          */
4489                         if ((adapter->flags & FLAG_TARC_SPEED_MODE_BIT) &&
4490                             !txb2b) {
4491                                 u32 tarc0;
4492                                 tarc0 = er32(TARC(0));
4493                                 tarc0 &= ~SPEED_MODE_BIT;
4494                                 ew32(TARC(0), tarc0);
4495                         }
4496
4497                         /*
4498                          * disable TSO for pcie and 10/100 speeds, to avoid
4499                          * some hardware issues
4500                          */
4501                         if (!(adapter->flags & FLAG_TSO_FORCE)) {
4502                                 switch (adapter->link_speed) {
4503                                 case SPEED_10:
4504                                 case SPEED_100:
4505                                         e_info("10/100 speed: disabling TSO\n");
4506                                         netdev->features &= ~NETIF_F_TSO;
4507                                         netdev->features &= ~NETIF_F_TSO6;
4508                                         break;
4509                                 case SPEED_1000:
4510                                         netdev->features |= NETIF_F_TSO;
4511                                         netdev->features |= NETIF_F_TSO6;
4512                                         break;
4513                                 default:
4514                                         /* oops */
4515                                         break;
4516                                 }
4517                         }
4518
4519                         /*
4520                          * enable transmits in the hardware, need to do this
4521                          * after setting TARC(0)
4522                          */
4523                         tctl = er32(TCTL);
4524                         tctl |= E1000_TCTL_EN;
4525                         ew32(TCTL, tctl);
4526
4527                         /*
4528                          * Perform any post-link-up configuration before
4529                          * reporting link up.
4530                          */
4531                         if (phy->ops.cfg_on_link_up)
4532                                 phy->ops.cfg_on_link_up(hw);
4533
4534                         netif_carrier_on(netdev);
4535
4536                         if (!test_bit(__E1000_DOWN, &adapter->state))
4537                                 mod_timer(&adapter->phy_info_timer,
4538                                           round_jiffies(jiffies + 2 * HZ));
4539                 }
4540         } else {
4541                 if (netif_carrier_ok(netdev)) {
4542                         adapter->link_speed = 0;
4543                         adapter->link_duplex = 0;
4544                         /* Link status message must follow this format */
4545                         printk(KERN_INFO "e1000e: %s NIC Link is Down\n",
4546                                adapter->netdev->name);
4547                         netif_carrier_off(netdev);
4548                         if (!test_bit(__E1000_DOWN, &adapter->state))
4549                                 mod_timer(&adapter->phy_info_timer,
4550                                           round_jiffies(jiffies + 2 * HZ));
4551
4552                         if (adapter->flags & FLAG_RX_NEEDS_RESTART)
4553                                 schedule_work(&adapter->reset_task);
4554                         else
4555                                 pm_schedule_suspend(netdev->dev.parent,
4556                                                         LINK_TIMEOUT);
4557                 }
4558         }
4559
4560 link_up:
4561         spin_lock(&adapter->stats64_lock);
4562         e1000e_update_stats(adapter);
4563
4564         mac->tx_packet_delta = adapter->stats.tpt - adapter->tpt_old;
4565         adapter->tpt_old = adapter->stats.tpt;
4566         mac->collision_delta = adapter->stats.colc - adapter->colc_old;
4567         adapter->colc_old = adapter->stats.colc;
4568
4569         adapter->gorc = adapter->stats.gorc - adapter->gorc_old;
4570         adapter->gorc_old = adapter->stats.gorc;
4571         adapter->gotc = adapter->stats.gotc - adapter->gotc_old;
4572         adapter->gotc_old = adapter->stats.gotc;
4573         spin_unlock(&adapter->stats64_lock);
4574
4575         e1000e_update_adaptive(&adapter->hw);
4576
4577         if (!netif_carrier_ok(netdev) &&
4578             (e1000_desc_unused(tx_ring) + 1 < tx_ring->count)) {
4579                 /*
4580                  * We've lost link, so the controller stops DMA,
4581                  * but we've got queued Tx work that's never going
4582                  * to get done, so reset controller to flush Tx.
4583                  * (Do the reset outside of interrupt context).
4584                  */
4585                 schedule_work(&adapter->reset_task);
4586                 /* return immediately since reset is imminent */
4587                 return;
4588         }
4589
4590         /* Simple mode for Interrupt Throttle Rate (ITR) */
4591         if (adapter->itr_setting == 4) {
4592                 /*
4593                  * Symmetric Tx/Rx gets a reduced ITR=2000;
4594                  * Total asymmetrical Tx or Rx gets ITR=8000;
4595                  * everyone else is between 2000-8000.
4596                  */
4597                 u32 goc = (adapter->gotc + adapter->gorc) / 10000;
4598                 u32 dif = (adapter->gotc > adapter->gorc ?
4599                             adapter->gotc - adapter->gorc :
4600                             adapter->gorc - adapter->gotc) / 10000;
4601                 u32 itr = goc > 0 ? (dif * 6000 / goc + 2000) : 8000;
4602
4603                 ew32(ITR, 1000000000 / (itr * 256));
4604         }
4605
4606         /* Cause software interrupt to ensure Rx ring is cleaned */
4607         if (adapter->msix_entries)
4608                 ew32(ICS, adapter->rx_ring->ims_val);
4609         else
4610                 ew32(ICS, E1000_ICS_RXDMT0);
4611
4612         /* flush pending descriptors to memory before detecting Tx hang */
4613         e1000e_flush_descriptors(adapter);
4614
4615         /* Force detection of hung controller every watchdog period */
4616         adapter->detect_tx_hung = true;
4617
4618         /*
4619          * With 82571 controllers, LAA may be overwritten due to controller
4620          * reset from the other port. Set the appropriate LAA in RAR[0]
4621          */
4622         if (e1000e_get_laa_state_82571(hw))
4623                 e1000e_rar_set(hw, adapter->hw.mac.addr, 0);
4624
4625         if (adapter->flags2 & FLAG2_CHECK_PHY_HANG)
4626                 e1000e_check_82574_phy_workaround(adapter);
4627
4628         /* Reset the timer */
4629         if (!test_bit(__E1000_DOWN, &adapter->state))
4630                 mod_timer(&adapter->watchdog_timer,
4631                           round_jiffies(jiffies + 2 * HZ));
4632 }
4633
4634 #define E1000_TX_FLAGS_CSUM             0x00000001
4635 #define E1000_TX_FLAGS_VLAN             0x00000002
4636 #define E1000_TX_FLAGS_TSO              0x00000004
4637 #define E1000_TX_FLAGS_IPV4             0x00000008
4638 #define E1000_TX_FLAGS_NO_FCS           0x00000010
4639 #define E1000_TX_FLAGS_VLAN_MASK        0xffff0000
4640 #define E1000_TX_FLAGS_VLAN_SHIFT       16
4641
4642 static int e1000_tso(struct e1000_ring *tx_ring, struct sk_buff *skb)
4643 {
4644         struct e1000_context_desc *context_desc;
4645         struct e1000_buffer *buffer_info;
4646         unsigned int i;
4647         u32 cmd_length = 0;
4648         u16 ipcse = 0, tucse, mss;
4649         u8 ipcss, ipcso, tucss, tucso, hdr_len;
4650
4651         if (!skb_is_gso(skb))
4652                 return 0;
4653
4654         if (skb_header_cloned(skb)) {
4655                 int err = pskb_expand_head(skb, 0, 0, GFP_ATOMIC);
4656
4657                 if (err)
4658                         return err;
4659         }
4660
4661         hdr_len = skb_transport_offset(skb) + tcp_hdrlen(skb);
4662         mss = skb_shinfo(skb)->gso_size;
4663         if (skb->protocol == htons(ETH_P_IP)) {
4664                 struct iphdr *iph = ip_hdr(skb);
4665                 iph->tot_len = 0;
4666                 iph->check = 0;
4667                 tcp_hdr(skb)->check = ~csum_tcpudp_magic(iph->saddr, iph->daddr,
4668                                                          0, IPPROTO_TCP, 0);
4669                 cmd_length = E1000_TXD_CMD_IP;
4670                 ipcse = skb_transport_offset(skb) - 1;
4671         } else if (skb_is_gso_v6(skb)) {
4672                 ipv6_hdr(skb)->payload_len = 0;
4673                 tcp_hdr(skb)->check = ~csum_ipv6_magic(&ipv6_hdr(skb)->saddr,
4674                                                        &ipv6_hdr(skb)->daddr,
4675                                                        0, IPPROTO_TCP, 0);
4676                 ipcse = 0;
4677         }
4678         ipcss = skb_network_offset(skb);
4679         ipcso = (void *)&(ip_hdr(skb)->check) - (void *)skb->data;
4680         tucss = skb_transport_offset(skb);
4681         tucso = (void *)&(tcp_hdr(skb)->check) - (void *)skb->data;
4682         tucse = 0;
4683
4684         cmd_length |= (E1000_TXD_CMD_DEXT | E1000_TXD_CMD_TSE |
4685                        E1000_TXD_CMD_TCP | (skb->len - (hdr_len)));
4686
4687         i = tx_ring->next_to_use;
4688         context_desc = E1000_CONTEXT_DESC(*tx_ring, i);
4689         buffer_info = &tx_ring->buffer_info[i];
4690
4691         context_desc->lower_setup.ip_fields.ipcss  = ipcss;
4692         context_desc->lower_setup.ip_fields.ipcso  = ipcso;
4693         context_desc->lower_setup.ip_fields.ipcse  = cpu_to_le16(ipcse);
4694         context_desc->upper_setup.tcp_fields.tucss = tucss;
4695         context_desc->upper_setup.tcp_fields.tucso = tucso;
4696         context_desc->upper_setup.tcp_fields.tucse = cpu_to_le16(tucse);
4697         context_desc->tcp_seg_setup.fields.mss     = cpu_to_le16(mss);
4698         context_desc->tcp_seg_setup.fields.hdr_len = hdr_len;
4699         context_desc->cmd_and_length = cpu_to_le32(cmd_length);
4700
4701         buffer_info->time_stamp = jiffies;
4702         buffer_info->next_to_watch = i;
4703
4704         i++;
4705         if (i == tx_ring->count)
4706                 i = 0;
4707         tx_ring->next_to_use = i;
4708
4709         return 1;
4710 }
4711
4712 static bool e1000_tx_csum(struct e1000_ring *tx_ring, struct sk_buff *skb)
4713 {
4714         struct e1000_adapter *adapter = tx_ring->adapter;
4715         struct e1000_context_desc *context_desc;
4716         struct e1000_buffer *buffer_info;
4717         unsigned int i;
4718         u8 css;
4719         u32 cmd_len = E1000_TXD_CMD_DEXT;
4720         __be16 protocol;
4721
4722         if (skb->ip_summed != CHECKSUM_PARTIAL)
4723                 return 0;
4724
4725         if (skb->protocol == cpu_to_be16(ETH_P_8021Q))
4726                 protocol = vlan_eth_hdr(skb)->h_vlan_encapsulated_proto;
4727         else
4728                 protocol = skb->protocol;
4729
4730         switch (protocol) {
4731         case cpu_to_be16(ETH_P_IP):
4732                 if (ip_hdr(skb)->protocol == IPPROTO_TCP)
4733                         cmd_len |= E1000_TXD_CMD_TCP;
4734                 break;
4735         case cpu_to_be16(ETH_P_IPV6):
4736                 /* XXX not handling all IPV6 headers */
4737                 if (ipv6_hdr(skb)->nexthdr == IPPROTO_TCP)
4738                         cmd_len |= E1000_TXD_CMD_TCP;
4739                 break;
4740         default:
4741                 if (unlikely(net_ratelimit()))
4742                         e_warn("checksum_partial proto=%x!\n",
4743                                be16_to_cpu(protocol));
4744                 break;
4745         }
4746
4747         css = skb_checksum_start_offset(skb);
4748
4749         i = tx_ring->next_to_use;
4750         buffer_info = &tx_ring->buffer_info[i];
4751         context_desc = E1000_CONTEXT_DESC(*tx_ring, i);
4752
4753         context_desc->lower_setup.ip_config = 0;
4754         context_desc->upper_setup.tcp_fields.tucss = css;
4755         context_desc->upper_setup.tcp_fields.tucso =
4756                                 css + skb->csum_offset;
4757         context_desc->upper_setup.tcp_fields.tucse = 0;
4758         context_desc->tcp_seg_setup.data = 0;
4759         context_desc->cmd_and_length = cpu_to_le32(cmd_len);
4760
4761         buffer_info->time_stamp = jiffies;
4762         buffer_info->next_to_watch = i;
4763
4764         i++;
4765         if (i == tx_ring->count)
4766                 i = 0;
4767         tx_ring->next_to_use = i;
4768
4769         return 1;
4770 }
4771
4772 #define E1000_MAX_PER_TXD       8192
4773 #define E1000_MAX_TXD_PWR       12
4774
4775 static int e1000_tx_map(struct e1000_ring *tx_ring, struct sk_buff *skb,
4776                         unsigned int first, unsigned int max_per_txd,
4777                         unsigned int nr_frags, unsigned int mss)
4778 {
4779         struct e1000_adapter *adapter = tx_ring->adapter;
4780         struct pci_dev *pdev = adapter->pdev;
4781         struct e1000_buffer *buffer_info;
4782         unsigned int len = skb_headlen(skb);
4783         unsigned int offset = 0, size, count = 0, i;
4784         unsigned int f, bytecount, segs;
4785
4786         i = tx_ring->next_to_use;
4787
4788         while (len) {
4789                 buffer_info = &tx_ring->buffer_info[i];
4790                 size = min(len, max_per_txd);
4791
4792                 buffer_info->length = size;
4793                 buffer_info->time_stamp = jiffies;
4794                 buffer_info->next_to_watch = i;
4795                 buffer_info->dma = dma_map_single(&pdev->dev,
4796                                                   skb->data + offset,
4797                                                   size, DMA_TO_DEVICE);
4798                 buffer_info->mapped_as_page = false;
4799                 if (dma_mapping_error(&pdev->dev, buffer_info->dma))
4800                         goto dma_error;
4801
4802                 len -= size;
4803                 offset += size;
4804                 count++;
4805
4806                 if (len) {
4807                         i++;
4808                         if (i == tx_ring->count)
4809                                 i = 0;
4810                 }
4811         }
4812
4813         for (f = 0; f < nr_frags; f++) {
4814                 const struct skb_frag_struct *frag;
4815
4816                 frag = &skb_shinfo(skb)->frags[f];
4817                 len = skb_frag_size(frag);
4818                 offset = 0;
4819
4820                 while (len) {
4821                         i++;
4822                         if (i == tx_ring->count)
4823                                 i = 0;
4824
4825                         buffer_info = &tx_ring->buffer_info[i];
4826                         size = min(len, max_per_txd);
4827
4828                         buffer_info->length = size;
4829                         buffer_info->time_stamp = jiffies;
4830                         buffer_info->next_to_watch = i;
4831                         buffer_info->dma = skb_frag_dma_map(&pdev->dev, frag,
4832                                                 offset, size, DMA_TO_DEVICE);
4833                         buffer_info->mapped_as_page = true;
4834                         if (dma_mapping_error(&pdev->dev, buffer_info->dma))
4835                                 goto dma_error;
4836
4837                         len -= size;
4838                         offset += size;
4839                         count++;
4840                 }
4841         }
4842
4843         segs = skb_shinfo(skb)->gso_segs ? : 1;
4844         /* multiply data chunks by size of headers */
4845         bytecount = ((segs - 1) * skb_headlen(skb)) + skb->len;
4846
4847         tx_ring->buffer_info[i].skb = skb;
4848         tx_ring->buffer_info[i].segs = segs;
4849         tx_ring->buffer_info[i].bytecount = bytecount;
4850         tx_ring->buffer_info[first].next_to_watch = i;
4851
4852         return count;
4853
4854 dma_error:
4855         dev_err(&pdev->dev, "Tx DMA map failed\n");
4856         buffer_info->dma = 0;
4857         if (count)
4858                 count--;
4859
4860         while (count--) {
4861                 if (i == 0)
4862                         i += tx_ring->count;
4863                 i--;
4864                 buffer_info = &tx_ring->buffer_info[i];
4865                 e1000_put_txbuf(tx_ring, buffer_info);
4866         }
4867
4868         return 0;
4869 }
4870
4871 static void e1000_tx_queue(struct e1000_ring *tx_ring, int tx_flags, int count)
4872 {
4873         struct e1000_adapter *adapter = tx_ring->adapter;
4874         struct e1000_tx_desc *tx_desc = NULL;
4875         struct e1000_buffer *buffer_info;
4876         u32 txd_upper = 0, txd_lower = E1000_TXD_CMD_IFCS;
4877         unsigned int i;
4878
4879         if (tx_flags & E1000_TX_FLAGS_TSO) {
4880                 txd_lower |= E1000_TXD_CMD_DEXT | E1000_TXD_DTYP_D |
4881                              E1000_TXD_CMD_TSE;
4882                 txd_upper |= E1000_TXD_POPTS_TXSM << 8;
4883
4884                 if (tx_flags & E1000_TX_FLAGS_IPV4)
4885                         txd_upper |= E1000_TXD_POPTS_IXSM << 8;
4886         }
4887
4888         if (tx_flags & E1000_TX_FLAGS_CSUM) {
4889                 txd_lower |= E1000_TXD_CMD_DEXT | E1000_TXD_DTYP_D;
4890                 txd_upper |= E1000_TXD_POPTS_TXSM << 8;
4891         }
4892
4893         if (tx_flags & E1000_TX_FLAGS_VLAN) {
4894                 txd_lower |= E1000_TXD_CMD_VLE;
4895                 txd_upper |= (tx_flags & E1000_TX_FLAGS_VLAN_MASK);
4896         }
4897
4898         if (unlikely(tx_flags & E1000_TX_FLAGS_NO_FCS))
4899                 txd_lower &= ~(E1000_TXD_CMD_IFCS);
4900
4901         i = tx_ring->next_to_use;
4902
4903         do {
4904                 buffer_info = &tx_ring->buffer_info[i];
4905                 tx_desc = E1000_TX_DESC(*tx_ring, i);
4906                 tx_desc->buffer_addr = cpu_to_le64(buffer_info->dma);
4907                 tx_desc->lower.data =
4908                         cpu_to_le32(txd_lower | buffer_info->length);
4909                 tx_desc->upper.data = cpu_to_le32(txd_upper);
4910
4911                 i++;
4912                 if (i == tx_ring->count)
4913                         i = 0;
4914         } while (--count > 0);
4915
4916         tx_desc->lower.data |= cpu_to_le32(adapter->txd_cmd);
4917
4918         /* txd_cmd re-enables FCS, so we'll re-disable it here as desired. */
4919         if (unlikely(tx_flags & E1000_TX_FLAGS_NO_FCS))
4920                 tx_desc->lower.data &= ~(cpu_to_le32(E1000_TXD_CMD_IFCS));
4921
4922         /*
4923          * Force memory writes to complete before letting h/w
4924          * know there are new descriptors to fetch.  (Only
4925          * applicable for weak-ordered memory model archs,
4926          * such as IA-64).
4927          */
4928         wmb();
4929
4930         tx_ring->next_to_use = i;
4931
4932         if (adapter->flags2 & FLAG2_PCIM2PCI_ARBITER_WA)
4933                 e1000e_update_tdt_wa(tx_ring, i);
4934         else
4935                 writel(i, tx_ring->tail);
4936
4937         /*
4938          * we need this if more than one processor can write to our tail
4939          * at a time, it synchronizes IO on IA64/Altix systems
4940          */
4941         mmiowb();
4942 }
4943
4944 #define MINIMUM_DHCP_PACKET_SIZE 282
4945 static int e1000_transfer_dhcp_info(struct e1000_adapter *adapter,
4946                                     struct sk_buff *skb)
4947 {
4948         struct e1000_hw *hw =  &adapter->hw;
4949         u16 length, offset;
4950
4951         if (vlan_tx_tag_present(skb)) {
4952                 if (!((vlan_tx_tag_get(skb) == adapter->hw.mng_cookie.vlan_id) &&
4953                     (adapter->hw.mng_cookie.status &
4954                         E1000_MNG_DHCP_COOKIE_STATUS_VLAN)))
4955                         return 0;
4956         }
4957
4958         if (skb->len <= MINIMUM_DHCP_PACKET_SIZE)
4959                 return 0;
4960
4961         if (((struct ethhdr *) skb->data)->h_proto != htons(ETH_P_IP))
4962                 return 0;
4963
4964         {
4965                 const struct iphdr *ip = (struct iphdr *)((u8 *)skb->data+14);
4966                 struct udphdr *udp;
4967
4968                 if (ip->protocol != IPPROTO_UDP)
4969                         return 0;
4970
4971                 udp = (struct udphdr *)((u8 *)ip + (ip->ihl << 2));
4972                 if (ntohs(udp->dest) != 67)
4973                         return 0;
4974
4975                 offset = (u8 *)udp + 8 - skb->data;
4976                 length = skb->len - offset;
4977                 return e1000e_mng_write_dhcp_info(hw, (u8 *)udp + 8, length);
4978         }
4979
4980         return 0;
4981 }
4982
4983 static int __e1000_maybe_stop_tx(struct e1000_ring *tx_ring, int size)
4984 {
4985         struct e1000_adapter *adapter = tx_ring->adapter;
4986
4987         netif_stop_queue(adapter->netdev);
4988         /*
4989          * Herbert's original patch had:
4990          *  smp_mb__after_netif_stop_queue();
4991          * but since that doesn't exist yet, just open code it.
4992          */
4993         smp_mb();
4994
4995         /*
4996          * We need to check again in a case another CPU has just
4997          * made room available.
4998          */
4999         if (e1000_desc_unused(tx_ring) < size)
5000                 return -EBUSY;
5001
5002         /* A reprieve! */
5003         netif_start_queue(adapter->netdev);
5004         ++adapter->restart_queue;
5005         return 0;
5006 }
5007
5008 static int e1000_maybe_stop_tx(struct e1000_ring *tx_ring, int size)
5009 {
5010         if (e1000_desc_unused(tx_ring) >= size)
5011                 return 0;
5012         return __e1000_maybe_stop_tx(tx_ring, size);
5013 }
5014
5015 #define TXD_USE_COUNT(S, X) (((S) >> (X)) + 1)
5016 static netdev_tx_t e1000_xmit_frame(struct sk_buff *skb,
5017                                     struct net_device *netdev)
5018 {
5019         struct e1000_adapter *adapter = netdev_priv(netdev);
5020         struct e1000_ring *tx_ring = adapter->tx_ring;
5021         unsigned int first;
5022         unsigned int max_per_txd = E1000_MAX_PER_TXD;
5023         unsigned int max_txd_pwr = E1000_MAX_TXD_PWR;
5024         unsigned int tx_flags = 0;
5025         unsigned int len = skb_headlen(skb);
5026         unsigned int nr_frags;
5027         unsigned int mss;
5028         int count = 0;
5029         int tso;
5030         unsigned int f;
5031
5032         if (test_bit(__E1000_DOWN, &adapter->state)) {
5033                 dev_kfree_skb_any(skb);
5034                 return NETDEV_TX_OK;
5035         }
5036
5037         if (skb->len <= 0) {
5038                 dev_kfree_skb_any(skb);
5039                 return NETDEV_TX_OK;
5040         }
5041
5042         mss = skb_shinfo(skb)->gso_size;
5043         /*
5044          * The controller does a simple calculation to
5045          * make sure there is enough room in the FIFO before
5046          * initiating the DMA for each buffer.  The calc is:
5047          * 4 = ceil(buffer len/mss).  To make sure we don't
5048          * overrun the FIFO, adjust the max buffer len if mss
5049          * drops.
5050          */
5051         if (mss) {
5052                 u8 hdr_len;
5053                 max_per_txd = min(mss << 2, max_per_txd);
5054                 max_txd_pwr = fls(max_per_txd) - 1;
5055
5056                 /*
5057                  * TSO Workaround for 82571/2/3 Controllers -- if skb->data
5058                  * points to just header, pull a few bytes of payload from
5059                  * frags into skb->data
5060                  */
5061                 hdr_len = skb_transport_offset(skb) + tcp_hdrlen(skb);
5062                 /*
5063                  * we do this workaround for ES2LAN, but it is un-necessary,
5064                  * avoiding it could save a lot of cycles
5065                  */
5066                 if (skb->data_len && (hdr_len == len)) {
5067                         unsigned int pull_size;
5068
5069                         pull_size = min_t(unsigned int, 4, skb->data_len);
5070                         if (!__pskb_pull_tail(skb, pull_size)) {
5071                                 e_err("__pskb_pull_tail failed.\n");
5072                                 dev_kfree_skb_any(skb);
5073                                 return NETDEV_TX_OK;
5074                         }
5075                         len = skb_headlen(skb);
5076                 }
5077         }
5078
5079         /* reserve a descriptor for the offload context */
5080         if ((mss) || (skb->ip_summed == CHECKSUM_PARTIAL))
5081                 count++;
5082         count++;
5083
5084         count += TXD_USE_COUNT(len, max_txd_pwr);
5085
5086         nr_frags = skb_shinfo(skb)->nr_frags;
5087         for (f = 0; f < nr_frags; f++)
5088                 count += TXD_USE_COUNT(skb_frag_size(&skb_shinfo(skb)->frags[f]),
5089                                        max_txd_pwr);
5090
5091         if (adapter->hw.mac.tx_pkt_filtering)
5092                 e1000_transfer_dhcp_info(adapter, skb);
5093
5094         /*
5095          * need: count + 2 desc gap to keep tail from touching
5096          * head, otherwise try next time
5097          */
5098         if (e1000_maybe_stop_tx(tx_ring, count + 2))
5099                 return NETDEV_TX_BUSY;
5100
5101         if (vlan_tx_tag_present(skb)) {
5102                 tx_flags |= E1000_TX_FLAGS_VLAN;
5103                 tx_flags |= (vlan_tx_tag_get(skb) << E1000_TX_FLAGS_VLAN_SHIFT);
5104         }
5105
5106         first = tx_ring->next_to_use;
5107
5108         tso = e1000_tso(tx_ring, skb);
5109         if (tso < 0) {
5110                 dev_kfree_skb_any(skb);
5111                 return NETDEV_TX_OK;
5112         }
5113
5114         if (tso)
5115                 tx_flags |= E1000_TX_FLAGS_TSO;
5116         else if (e1000_tx_csum(tx_ring, skb))
5117                 tx_flags |= E1000_TX_FLAGS_CSUM;
5118
5119         /*
5120          * Old method was to assume IPv4 packet by default if TSO was enabled.
5121          * 82571 hardware supports TSO capabilities for IPv6 as well...
5122          * no longer assume, we must.
5123          */
5124         if (skb->protocol == htons(ETH_P_IP))
5125                 tx_flags |= E1000_TX_FLAGS_IPV4;
5126
5127         if (unlikely(skb->no_fcs))
5128                 tx_flags |= E1000_TX_FLAGS_NO_FCS;
5129
5130         /* if count is 0 then mapping error has occurred */
5131         count = e1000_tx_map(tx_ring, skb, first, max_per_txd, nr_frags, mss);
5132         if (count) {
5133                 netdev_sent_queue(netdev, skb->len);
5134                 e1000_tx_queue(tx_ring, tx_flags, count);
5135                 /* Make sure there is space in the ring for the next send. */
5136                 e1000_maybe_stop_tx(tx_ring, MAX_SKB_FRAGS + 2);
5137
5138         } else {
5139                 dev_kfree_skb_any(skb);
5140                 tx_ring->buffer_info[first].time_stamp = 0;
5141                 tx_ring->next_to_use = first;
5142         }
5143
5144         return NETDEV_TX_OK;
5145 }
5146
5147 /**
5148  * e1000_tx_timeout - Respond to a Tx Hang
5149  * @netdev: network interface device structure
5150  **/
5151 static void e1000_tx_timeout(struct net_device *netdev)
5152 {
5153         struct e1000_adapter *adapter = netdev_priv(netdev);
5154
5155         /* Do the reset outside of interrupt context */
5156         adapter->tx_timeout_count++;
5157         schedule_work(&adapter->reset_task);
5158 }
5159
5160 static void e1000_reset_task(struct work_struct *work)
5161 {
5162         struct e1000_adapter *adapter;
5163         adapter = container_of(work, struct e1000_adapter, reset_task);
5164
5165         /* don't run the task if already down */
5166         if (test_bit(__E1000_DOWN, &adapter->state))
5167                 return;
5168
5169         if (!((adapter->flags & FLAG_RX_NEEDS_RESTART) &&
5170               (adapter->flags & FLAG_RX_RESTART_NOW))) {
5171                 e1000e_dump(adapter);
5172                 e_err("Reset adapter\n");
5173         }
5174         e1000e_reinit_locked(adapter);
5175 }
5176
5177 /**
5178  * e1000_get_stats64 - Get System Network Statistics
5179  * @netdev: network interface device structure
5180  * @stats: rtnl_link_stats64 pointer
5181  *
5182  * Returns the address of the device statistics structure.
5183  **/
5184 struct rtnl_link_stats64 *e1000e_get_stats64(struct net_device *netdev,
5185                                              struct rtnl_link_stats64 *stats)
5186 {
5187         struct e1000_adapter *adapter = netdev_priv(netdev);
5188
5189         memset(stats, 0, sizeof(struct rtnl_link_stats64));
5190         spin_lock(&adapter->stats64_lock);
5191         e1000e_update_stats(adapter);
5192         /* Fill out the OS statistics structure */
5193         stats->rx_bytes = adapter->stats.gorc;
5194         stats->rx_packets = adapter->stats.gprc;
5195         stats->tx_bytes = adapter->stats.gotc;
5196         stats->tx_packets = adapter->stats.gptc;
5197         stats->multicast = adapter->stats.mprc;
5198         stats->collisions = adapter->stats.colc;
5199
5200         /* Rx Errors */
5201
5202         /*
5203          * RLEC on some newer hardware can be incorrect so build
5204          * our own version based on RUC and ROC
5205          */
5206         stats->rx_errors = adapter->stats.rxerrc +
5207                 adapter->stats.crcerrs + adapter->stats.algnerrc +
5208                 adapter->stats.ruc + adapter->stats.roc +
5209                 adapter->stats.cexterr;
5210         stats->rx_length_errors = adapter->stats.ruc +
5211                                               adapter->stats.roc;
5212         stats->rx_crc_errors = adapter->stats.crcerrs;
5213         stats->rx_frame_errors = adapter->stats.algnerrc;
5214         stats->rx_missed_errors = adapter->stats.mpc;
5215
5216         /* Tx Errors */
5217         stats->tx_errors = adapter->stats.ecol +
5218                                        adapter->stats.latecol;
5219         stats->tx_aborted_errors = adapter->stats.ecol;
5220         stats->tx_window_errors = adapter->stats.latecol;
5221         stats->tx_carrier_errors = adapter->stats.tncrs;
5222
5223         /* Tx Dropped needs to be maintained elsewhere */
5224
5225         spin_unlock(&adapter->stats64_lock);
5226         return stats;
5227 }
5228
5229 /**
5230  * e1000_change_mtu - Change the Maximum Transfer Unit
5231  * @netdev: network interface device structure
5232  * @new_mtu: new value for maximum frame size
5233  *
5234  * Returns 0 on success, negative on failure
5235  **/
5236 static int e1000_change_mtu(struct net_device *netdev, int new_mtu)
5237 {
5238         struct e1000_adapter *adapter = netdev_priv(netdev);
5239         int max_frame = new_mtu + ETH_HLEN + ETH_FCS_LEN;
5240
5241         /* Jumbo frame support */
5242         if (max_frame > ETH_FRAME_LEN + ETH_FCS_LEN) {
5243                 if (!(adapter->flags & FLAG_HAS_JUMBO_FRAMES)) {
5244                         e_err("Jumbo Frames not supported.\n");
5245                         return -EINVAL;
5246                 }
5247
5248                 /*
5249                  * IP payload checksum (enabled with jumbos/packet-split when
5250                  * Rx checksum is enabled) and generation of RSS hash is
5251                  * mutually exclusive in the hardware.
5252                  */
5253                 if ((netdev->features & NETIF_F_RXCSUM) &&
5254                     (netdev->features & NETIF_F_RXHASH)) {
5255                         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");
5256                         return -EINVAL;
5257                 }
5258         }
5259
5260         /* Supported frame sizes */
5261         if ((new_mtu < ETH_ZLEN + ETH_FCS_LEN + VLAN_HLEN) ||
5262             (max_frame > adapter->max_hw_frame_size)) {
5263                 e_err("Unsupported MTU setting\n");
5264                 return -EINVAL;
5265         }
5266
5267         /* Jumbo frame workaround on 82579 requires CRC be stripped */
5268         if ((adapter->hw.mac.type == e1000_pch2lan) &&
5269             !(adapter->flags2 & FLAG2_CRC_STRIPPING) &&
5270             (new_mtu > ETH_DATA_LEN)) {
5271                 e_err("Jumbo Frames not supported on 82579 when CRC stripping is disabled.\n");
5272                 return -EINVAL;
5273         }
5274
5275         /* 82573 Errata 17 */
5276         if (((adapter->hw.mac.type == e1000_82573) ||
5277              (adapter->hw.mac.type == e1000_82574)) &&
5278             (max_frame > ETH_FRAME_LEN + ETH_FCS_LEN)) {
5279                 adapter->flags2 |= FLAG2_DISABLE_ASPM_L1;
5280                 e1000e_disable_aspm(adapter->pdev, PCIE_LINK_STATE_L1);
5281         }
5282
5283         while (test_and_set_bit(__E1000_RESETTING, &adapter->state))
5284                 usleep_range(1000, 2000);
5285         /* e1000e_down -> e1000e_reset dependent on max_frame_size & mtu */
5286         adapter->max_frame_size = max_frame;
5287         e_info("changing MTU from %d to %d\n", netdev->mtu, new_mtu);
5288         netdev->mtu = new_mtu;
5289         if (netif_running(netdev))
5290                 e1000e_down(adapter);
5291
5292         /*
5293          * NOTE: netdev_alloc_skb reserves 16 bytes, and typically NET_IP_ALIGN
5294          * means we reserve 2 more, this pushes us to allocate from the next
5295          * larger slab size.
5296          * i.e. RXBUFFER_2048 --> size-4096 slab
5297          * However with the new *_jumbo_rx* routines, jumbo receives will use
5298          * fragmented skbs
5299          */
5300
5301         if (max_frame <= 2048)
5302                 adapter->rx_buffer_len = 2048;
5303         else
5304                 adapter->rx_buffer_len = 4096;
5305
5306         /* adjust allocation if LPE protects us, and we aren't using SBP */
5307         if ((max_frame == ETH_FRAME_LEN + ETH_FCS_LEN) ||
5308              (max_frame == ETH_FRAME_LEN + VLAN_HLEN + ETH_FCS_LEN))
5309                 adapter->rx_buffer_len = ETH_FRAME_LEN + VLAN_HLEN
5310                                          + ETH_FCS_LEN;
5311
5312         if (netif_running(netdev))
5313                 e1000e_up(adapter);
5314         else
5315                 e1000e_reset(adapter);
5316
5317         clear_bit(__E1000_RESETTING, &adapter->state);
5318
5319         return 0;
5320 }
5321
5322 static int e1000_mii_ioctl(struct net_device *netdev, struct ifreq *ifr,
5323                            int cmd)
5324 {
5325         struct e1000_adapter *adapter = netdev_priv(netdev);
5326         struct mii_ioctl_data *data = if_mii(ifr);
5327
5328         if (adapter->hw.phy.media_type != e1000_media_type_copper)
5329                 return -EOPNOTSUPP;
5330
5331         switch (cmd) {
5332         case SIOCGMIIPHY:
5333                 data->phy_id = adapter->hw.phy.addr;
5334                 break;
5335         case SIOCGMIIREG:
5336                 e1000_phy_read_status(adapter);
5337
5338                 switch (data->reg_num & 0x1F) {
5339                 case MII_BMCR:
5340                         data->val_out = adapter->phy_regs.bmcr;
5341                         break;
5342                 case MII_BMSR:
5343                         data->val_out = adapter->phy_regs.bmsr;
5344                         break;
5345                 case MII_PHYSID1:
5346                         data->val_out = (adapter->hw.phy.id >> 16);
5347                         break;
5348                 case MII_PHYSID2:
5349                         data->val_out = (adapter->hw.phy.id & 0xFFFF);
5350                         break;
5351                 case MII_ADVERTISE:
5352                         data->val_out = adapter->phy_regs.advertise;
5353                         break;
5354                 case MII_LPA:
5355                         data->val_out = adapter->phy_regs.lpa;
5356                         break;
5357                 case MII_EXPANSION:
5358                         data->val_out = adapter->phy_regs.expansion;
5359                         break;
5360                 case MII_CTRL1000:
5361                         data->val_out = adapter->phy_regs.ctrl1000;
5362                         break;
5363                 case MII_STAT1000:
5364                         data->val_out = adapter->phy_regs.stat1000;
5365                         break;
5366                 case MII_ESTATUS:
5367                         data->val_out = adapter->phy_regs.estatus;
5368                         break;
5369                 default:
5370                         return -EIO;
5371                 }
5372                 break;
5373         case SIOCSMIIREG:
5374         default:
5375                 return -EOPNOTSUPP;
5376         }
5377         return 0;
5378 }
5379
5380 static int e1000_ioctl(struct net_device *netdev, struct ifreq *ifr, int cmd)
5381 {
5382         switch (cmd) {
5383         case SIOCGMIIPHY:
5384         case SIOCGMIIREG:
5385         case SIOCSMIIREG:
5386                 return e1000_mii_ioctl(netdev, ifr, cmd);
5387         default:
5388                 return -EOPNOTSUPP;
5389         }
5390 }
5391
5392 static int e1000_init_phy_wakeup(struct e1000_adapter *adapter, u32 wufc)
5393 {
5394         struct e1000_hw *hw = &adapter->hw;
5395         u32 i, mac_reg;
5396         u16 phy_reg, wuc_enable;
5397         int retval = 0;
5398
5399         /* copy MAC RARs to PHY RARs */
5400         e1000_copy_rx_addrs_to_phy_ich8lan(hw);
5401
5402         retval = hw->phy.ops.acquire(hw);
5403         if (retval) {
5404                 e_err("Could not acquire PHY\n");
5405                 return retval;
5406         }
5407
5408         /* Enable access to wakeup registers on and set page to BM_WUC_PAGE */
5409         retval = e1000_enable_phy_wakeup_reg_access_bm(hw, &wuc_enable);
5410         if (retval)
5411                 goto release;
5412
5413         /* copy MAC MTA to PHY MTA - only needed for pchlan */
5414         for (i = 0; i < adapter->hw.mac.mta_reg_count; i++) {
5415                 mac_reg = E1000_READ_REG_ARRAY(hw, E1000_MTA, i);
5416                 hw->phy.ops.write_reg_page(hw, BM_MTA(i),
5417                                            (u16)(mac_reg & 0xFFFF));
5418                 hw->phy.ops.write_reg_page(hw, BM_MTA(i) + 1,
5419                                            (u16)((mac_reg >> 16) & 0xFFFF));
5420         }
5421
5422         /* configure PHY Rx Control register */
5423         hw->phy.ops.read_reg_page(&adapter->hw, BM_RCTL, &phy_reg);
5424         mac_reg = er32(RCTL);
5425         if (mac_reg & E1000_RCTL_UPE)
5426                 phy_reg |= BM_RCTL_UPE;
5427         if (mac_reg & E1000_RCTL_MPE)
5428                 phy_reg |= BM_RCTL_MPE;
5429         phy_reg &= ~(BM_RCTL_MO_MASK);
5430         if (mac_reg & E1000_RCTL_MO_3)
5431                 phy_reg |= (((mac_reg & E1000_RCTL_MO_3) >> E1000_RCTL_MO_SHIFT)
5432                                 << BM_RCTL_MO_SHIFT);
5433         if (mac_reg & E1000_RCTL_BAM)
5434                 phy_reg |= BM_RCTL_BAM;
5435         if (mac_reg & E1000_RCTL_PMCF)
5436                 phy_reg |= BM_RCTL_PMCF;
5437         mac_reg = er32(CTRL);
5438         if (mac_reg & E1000_CTRL_RFCE)
5439                 phy_reg |= BM_RCTL_RFCE;
5440         hw->phy.ops.write_reg_page(&adapter->hw, BM_RCTL, phy_reg);
5441
5442         /* enable PHY wakeup in MAC register */
5443         ew32(WUFC, wufc);
5444         ew32(WUC, E1000_WUC_PHY_WAKE | E1000_WUC_PME_EN);
5445
5446         /* configure and enable PHY wakeup in PHY registers */
5447         hw->phy.ops.write_reg_page(&adapter->hw, BM_WUFC, wufc);
5448         hw->phy.ops.write_reg_page(&adapter->hw, BM_WUC, E1000_WUC_PME_EN);
5449
5450         /* activate PHY wakeup */
5451         wuc_enable |= BM_WUC_ENABLE_BIT | BM_WUC_HOST_WU_BIT;
5452         retval = e1000_disable_phy_wakeup_reg_access_bm(hw, &wuc_enable);
5453         if (retval)
5454                 e_err("Could not set PHY Host Wakeup bit\n");
5455 release:
5456         hw->phy.ops.release(hw);
5457
5458         return retval;
5459 }
5460
5461 static int __e1000_shutdown(struct pci_dev *pdev, bool *enable_wake,
5462                             bool runtime)
5463 {
5464         struct net_device *netdev = pci_get_drvdata(pdev);
5465         struct e1000_adapter *adapter = netdev_priv(netdev);
5466         struct e1000_hw *hw = &adapter->hw;
5467         u32 ctrl, ctrl_ext, rctl, status;
5468         /* Runtime suspend should only enable wakeup for link changes */
5469         u32 wufc = runtime ? E1000_WUFC_LNKC : adapter->wol;
5470         int retval = 0;
5471
5472         netif_device_detach(netdev);
5473
5474         if (netif_running(netdev)) {
5475                 WARN_ON(test_bit(__E1000_RESETTING, &adapter->state));
5476                 e1000e_down(adapter);
5477                 e1000_free_irq(adapter);
5478         }
5479         e1000e_reset_interrupt_capability(adapter);
5480
5481         retval = pci_save_state(pdev);
5482         if (retval)
5483                 return retval;
5484
5485         status = er32(STATUS);
5486         if (status & E1000_STATUS_LU)
5487                 wufc &= ~E1000_WUFC_LNKC;
5488
5489         if (wufc) {
5490                 e1000_setup_rctl(adapter);
5491                 e1000e_set_rx_mode(netdev);
5492
5493                 /* turn on all-multi mode if wake on multicast is enabled */
5494                 if (wufc & E1000_WUFC_MC) {
5495                         rctl = er32(RCTL);
5496                         rctl |= E1000_RCTL_MPE;
5497                         ew32(RCTL, rctl);
5498                 }
5499
5500                 ctrl = er32(CTRL);
5501                 /* advertise wake from D3Cold */
5502                 #define E1000_CTRL_ADVD3WUC 0x00100000
5503                 /* phy power management enable */
5504                 #define E1000_CTRL_EN_PHY_PWR_MGMT 0x00200000
5505                 ctrl |= E1000_CTRL_ADVD3WUC;
5506                 if (!(adapter->flags2 & FLAG2_HAS_PHY_WAKEUP))
5507                         ctrl |= E1000_CTRL_EN_PHY_PWR_MGMT;
5508                 ew32(CTRL, ctrl);
5509
5510                 if (adapter->hw.phy.media_type == e1000_media_type_fiber ||
5511                     adapter->hw.phy.media_type ==
5512                     e1000_media_type_internal_serdes) {
5513                         /* keep the laser running in D3 */
5514                         ctrl_ext = er32(CTRL_EXT);
5515                         ctrl_ext |= E1000_CTRL_EXT_SDP3_DATA;
5516                         ew32(CTRL_EXT, ctrl_ext);
5517                 }
5518
5519                 if (adapter->flags & FLAG_IS_ICH)
5520                         e1000_suspend_workarounds_ich8lan(&adapter->hw);
5521
5522                 /* Allow time for pending master requests to run */
5523                 e1000e_disable_pcie_master(&adapter->hw);
5524
5525                 if (adapter->flags2 & FLAG2_HAS_PHY_WAKEUP) {
5526                         /* enable wakeup by the PHY */
5527                         retval = e1000_init_phy_wakeup(adapter, wufc);
5528                         if (retval)
5529                                 return retval;
5530                 } else {
5531                         /* enable wakeup by the MAC */
5532                         ew32(WUFC, wufc);
5533                         ew32(WUC, E1000_WUC_PME_EN);
5534                 }
5535         } else {
5536                 ew32(WUC, 0);
5537                 ew32(WUFC, 0);
5538         }
5539
5540         *enable_wake = !!wufc;
5541
5542         /* make sure adapter isn't asleep if manageability is enabled */
5543         if ((adapter->flags & FLAG_MNG_PT_ENABLED) ||
5544             (hw->mac.ops.check_mng_mode(hw)))
5545                 *enable_wake = true;
5546
5547         if (adapter->hw.phy.type == e1000_phy_igp_3)
5548                 e1000e_igp3_phy_powerdown_workaround_ich8lan(&adapter->hw);
5549
5550         /*
5551          * Release control of h/w to f/w.  If f/w is AMT enabled, this
5552          * would have already happened in close and is redundant.
5553          */
5554         e1000e_release_hw_control(adapter);
5555
5556         pci_disable_device(pdev);
5557
5558         return 0;
5559 }
5560
5561 static void e1000_power_off(struct pci_dev *pdev, bool sleep, bool wake)
5562 {
5563         if (sleep && wake) {
5564                 pci_prepare_to_sleep(pdev);
5565                 return;
5566         }
5567
5568         pci_wake_from_d3(pdev, wake);
5569         pci_set_power_state(pdev, PCI_D3hot);
5570 }
5571
5572 static void e1000_complete_shutdown(struct pci_dev *pdev, bool sleep,
5573                                     bool wake)
5574 {
5575         struct net_device *netdev = pci_get_drvdata(pdev);
5576         struct e1000_adapter *adapter = netdev_priv(netdev);
5577
5578         /*
5579          * The pci-e switch on some quad port adapters will report a
5580          * correctable error when the MAC transitions from D0 to D3.  To
5581          * prevent this we need to mask off the correctable errors on the
5582          * downstream port of the pci-e switch.
5583          */
5584         if (adapter->flags & FLAG_IS_QUAD_PORT) {
5585                 struct pci_dev *us_dev = pdev->bus->self;
5586                 int pos = pci_pcie_cap(us_dev);
5587                 u16 devctl;
5588
5589                 pci_read_config_word(us_dev, pos + PCI_EXP_DEVCTL, &devctl);
5590                 pci_write_config_word(us_dev, pos + PCI_EXP_DEVCTL,
5591                                       (devctl & ~PCI_EXP_DEVCTL_CERE));
5592
5593                 e1000_power_off(pdev, sleep, wake);
5594
5595                 pci_write_config_word(us_dev, pos + PCI_EXP_DEVCTL, devctl);
5596         } else {
5597                 e1000_power_off(pdev, sleep, wake);
5598         }
5599 }
5600
5601 #ifdef CONFIG_PCIEASPM
5602 static void __e1000e_disable_aspm(struct pci_dev *pdev, u16 state)
5603 {
5604         pci_disable_link_state_locked(pdev, state);
5605 }
5606 #else
5607 static void __e1000e_disable_aspm(struct pci_dev *pdev, u16 state)
5608 {
5609         int pos;
5610         u16 reg16;
5611
5612         /*
5613          * Both device and parent should have the same ASPM setting.
5614          * Disable ASPM in downstream component first and then upstream.
5615          */
5616         pos = pci_pcie_cap(pdev);
5617         pci_read_config_word(pdev, pos + PCI_EXP_LNKCTL, &reg16);
5618         reg16 &= ~state;
5619         pci_write_config_word(pdev, pos + PCI_EXP_LNKCTL, reg16);
5620
5621         if (!pdev->bus->self)
5622                 return;
5623
5624         pos = pci_pcie_cap(pdev->bus->self);
5625         pci_read_config_word(pdev->bus->self, pos + PCI_EXP_LNKCTL, &reg16);
5626         reg16 &= ~state;
5627         pci_write_config_word(pdev->bus->self, pos + PCI_EXP_LNKCTL, reg16);
5628 }
5629 #endif
5630 static void e1000e_disable_aspm(struct pci_dev *pdev, u16 state)
5631 {
5632         dev_info(&pdev->dev, "Disabling ASPM %s %s\n",
5633                  (state & PCIE_LINK_STATE_L0S) ? "L0s" : "",
5634                  (state & PCIE_LINK_STATE_L1) ? "L1" : "");
5635
5636         __e1000e_disable_aspm(pdev, state);
5637 }
5638
5639 #ifdef CONFIG_PM
5640 static bool e1000e_pm_ready(struct e1000_adapter *adapter)
5641 {
5642         return !!adapter->tx_ring->buffer_info;
5643 }
5644
5645 static int __e1000_resume(struct pci_dev *pdev)
5646 {
5647         struct net_device *netdev = pci_get_drvdata(pdev);
5648         struct e1000_adapter *adapter = netdev_priv(netdev);
5649         struct e1000_hw *hw = &adapter->hw;
5650         u16 aspm_disable_flag = 0;
5651         u32 err;
5652
5653         if (adapter->flags2 & FLAG2_DISABLE_ASPM_L0S)
5654                 aspm_disable_flag = PCIE_LINK_STATE_L0S;
5655         if (adapter->flags2 & FLAG2_DISABLE_ASPM_L1)
5656                 aspm_disable_flag |= PCIE_LINK_STATE_L1;
5657         if (aspm_disable_flag)
5658                 e1000e_disable_aspm(pdev, aspm_disable_flag);
5659
5660         pci_set_power_state(pdev, PCI_D0);
5661         pci_restore_state(pdev);
5662         pci_save_state(pdev);
5663
5664         e1000e_set_interrupt_capability(adapter);
5665         if (netif_running(netdev)) {
5666                 err = e1000_request_irq(adapter);
5667                 if (err)
5668                         return err;
5669         }
5670
5671         if (hw->mac.type == e1000_pch2lan)
5672                 e1000_resume_workarounds_pchlan(&adapter->hw);
5673
5674         e1000e_power_up_phy(adapter);
5675
5676         /* report the system wakeup cause from S3/S4 */
5677         if (adapter->flags2 & FLAG2_HAS_PHY_WAKEUP) {
5678                 u16 phy_data;
5679
5680                 e1e_rphy(&adapter->hw, BM_WUS, &phy_data);
5681                 if (phy_data) {
5682                         e_info("PHY Wakeup cause - %s\n",
5683                                 phy_data & E1000_WUS_EX ? "Unicast Packet" :
5684                                 phy_data & E1000_WUS_MC ? "Multicast Packet" :
5685                                 phy_data & E1000_WUS_BC ? "Broadcast Packet" :
5686                                 phy_data & E1000_WUS_MAG ? "Magic Packet" :
5687                                 phy_data & E1000_WUS_LNKC ?
5688                                 "Link Status Change" : "other");
5689                 }
5690                 e1e_wphy(&adapter->hw, BM_WUS, ~0);
5691         } else {
5692                 u32 wus = er32(WUS);
5693                 if (wus) {
5694                         e_info("MAC Wakeup cause - %s\n",
5695                                 wus & E1000_WUS_EX ? "Unicast Packet" :
5696                                 wus & E1000_WUS_MC ? "Multicast Packet" :
5697                                 wus & E1000_WUS_BC ? "Broadcast Packet" :
5698                                 wus & E1000_WUS_MAG ? "Magic Packet" :
5699                                 wus & E1000_WUS_LNKC ? "Link Status Change" :
5700                                 "other");
5701                 }
5702                 ew32(WUS, ~0);
5703         }
5704
5705         e1000e_reset(adapter);
5706
5707         e1000_init_manageability_pt(adapter);
5708
5709         if (netif_running(netdev))
5710                 e1000e_up(adapter);
5711
5712         netif_device_attach(netdev);
5713
5714         /*
5715          * If the controller has AMT, do not set DRV_LOAD until the interface
5716          * is up.  For all other cases, let the f/w know that the h/w is now
5717          * under the control of the driver.
5718          */
5719         if (!(adapter->flags & FLAG_HAS_AMT))
5720                 e1000e_get_hw_control(adapter);
5721
5722         return 0;
5723 }
5724
5725 #ifdef CONFIG_PM_SLEEP
5726 static int e1000_suspend(struct device *dev)
5727 {
5728         struct pci_dev *pdev = to_pci_dev(dev);
5729         int retval;
5730         bool wake;
5731
5732         retval = __e1000_shutdown(pdev, &wake, false);
5733         if (!retval)
5734                 e1000_complete_shutdown(pdev, true, wake);
5735
5736         return retval;
5737 }
5738
5739 static int e1000_resume(struct device *dev)
5740 {
5741         struct pci_dev *pdev = to_pci_dev(dev);
5742         struct net_device *netdev = pci_get_drvdata(pdev);
5743         struct e1000_adapter *adapter = netdev_priv(netdev);
5744
5745         if (e1000e_pm_ready(adapter))
5746                 adapter->idle_check = true;
5747
5748         return __e1000_resume(pdev);
5749 }
5750 #endif /* CONFIG_PM_SLEEP */
5751
5752 #ifdef CONFIG_PM_RUNTIME
5753 static int e1000_runtime_suspend(struct device *dev)
5754 {
5755         struct pci_dev *pdev = to_pci_dev(dev);
5756         struct net_device *netdev = pci_get_drvdata(pdev);
5757         struct e1000_adapter *adapter = netdev_priv(netdev);
5758
5759         if (e1000e_pm_ready(adapter)) {
5760                 bool wake;
5761
5762                 __e1000_shutdown(pdev, &wake, true);
5763         }
5764
5765         return 0;
5766 }
5767
5768 static int e1000_idle(struct device *dev)
5769 {
5770         struct pci_dev *pdev = to_pci_dev(dev);
5771         struct net_device *netdev = pci_get_drvdata(pdev);
5772         struct e1000_adapter *adapter = netdev_priv(netdev);
5773
5774         if (!e1000e_pm_ready(adapter))
5775                 return 0;
5776
5777         if (adapter->idle_check) {
5778                 adapter->idle_check = false;
5779                 if (!e1000e_has_link(adapter))
5780                         pm_schedule_suspend(dev, MSEC_PER_SEC);
5781         }
5782
5783         return -EBUSY;
5784 }
5785
5786 static int e1000_runtime_resume(struct device *dev)
5787 {
5788         struct pci_dev *pdev = to_pci_dev(dev);
5789         struct net_device *netdev = pci_get_drvdata(pdev);
5790         struct e1000_adapter *adapter = netdev_priv(netdev);
5791
5792         if (!e1000e_pm_ready(adapter))
5793                 return 0;
5794
5795         adapter->idle_check = !dev->power.runtime_auto;
5796         return __e1000_resume(pdev);
5797 }
5798 #endif /* CONFIG_PM_RUNTIME */
5799 #endif /* CONFIG_PM */
5800
5801 static void e1000_shutdown(struct pci_dev *pdev)
5802 {
5803         bool wake = false;
5804
5805         __e1000_shutdown(pdev, &wake, false);
5806
5807         if (system_state == SYSTEM_POWER_OFF)
5808                 e1000_complete_shutdown(pdev, false, wake);
5809 }
5810
5811 #ifdef CONFIG_NET_POLL_CONTROLLER
5812
5813 static irqreturn_t e1000_intr_msix(int irq, void *data)
5814 {
5815         struct net_device *netdev = data;
5816         struct e1000_adapter *adapter = netdev_priv(netdev);
5817
5818         if (adapter->msix_entries) {
5819                 int vector, msix_irq;
5820
5821                 vector = 0;
5822                 msix_irq = adapter->msix_entries[vector].vector;
5823                 disable_irq(msix_irq);
5824                 e1000_intr_msix_rx(msix_irq, netdev);
5825                 enable_irq(msix_irq);
5826
5827                 vector++;
5828                 msix_irq = adapter->msix_entries[vector].vector;
5829                 disable_irq(msix_irq);
5830                 e1000_intr_msix_tx(msix_irq, netdev);
5831                 enable_irq(msix_irq);
5832
5833                 vector++;
5834                 msix_irq = adapter->msix_entries[vector].vector;
5835                 disable_irq(msix_irq);
5836                 e1000_msix_other(msix_irq, netdev);
5837                 enable_irq(msix_irq);
5838         }
5839
5840         return IRQ_HANDLED;
5841 }
5842
5843 /*
5844  * Polling 'interrupt' - used by things like netconsole to send skbs
5845  * without having to re-enable interrupts. It's not called while
5846  * the interrupt routine is executing.
5847  */
5848 static void e1000_netpoll(struct net_device *netdev)
5849 {
5850         struct e1000_adapter *adapter = netdev_priv(netdev);
5851
5852         switch (adapter->int_mode) {
5853         case E1000E_INT_MODE_MSIX:
5854                 e1000_intr_msix(adapter->pdev->irq, netdev);
5855                 break;
5856         case E1000E_INT_MODE_MSI:
5857                 disable_irq(adapter->pdev->irq);
5858                 e1000_intr_msi(adapter->pdev->irq, netdev);
5859                 enable_irq(adapter->pdev->irq);
5860                 break;
5861         default: /* E1000E_INT_MODE_LEGACY */
5862                 disable_irq(adapter->pdev->irq);
5863                 e1000_intr(adapter->pdev->irq, netdev);
5864                 enable_irq(adapter->pdev->irq);
5865                 break;
5866         }
5867 }
5868 #endif
5869
5870 /**
5871  * e1000_io_error_detected - called when PCI error is detected
5872  * @pdev: Pointer to PCI device
5873  * @state: The current pci connection state
5874  *
5875  * This function is called after a PCI bus error affecting
5876  * this device has been detected.
5877  */
5878 static pci_ers_result_t e1000_io_error_detected(struct pci_dev *pdev,
5879                                                 pci_channel_state_t state)
5880 {
5881         struct net_device *netdev = pci_get_drvdata(pdev);
5882         struct e1000_adapter *adapter = netdev_priv(netdev);
5883
5884         netif_device_detach(netdev);
5885
5886         if (state == pci_channel_io_perm_failure)
5887                 return PCI_ERS_RESULT_DISCONNECT;
5888
5889         if (netif_running(netdev))
5890                 e1000e_down(adapter);
5891         pci_disable_device(pdev);
5892
5893         /* Request a slot slot reset. */
5894         return PCI_ERS_RESULT_NEED_RESET;
5895 }
5896
5897 /**
5898  * e1000_io_slot_reset - called after the pci bus has been reset.
5899  * @pdev: Pointer to PCI device
5900  *
5901  * Restart the card from scratch, as if from a cold-boot. Implementation
5902  * resembles the first-half of the e1000_resume routine.
5903  */
5904 static pci_ers_result_t e1000_io_slot_reset(struct pci_dev *pdev)
5905 {
5906         struct net_device *netdev = pci_get_drvdata(pdev);
5907         struct e1000_adapter *adapter = netdev_priv(netdev);
5908         struct e1000_hw *hw = &adapter->hw;
5909         u16 aspm_disable_flag = 0;
5910         int err;
5911         pci_ers_result_t result;
5912
5913         if (adapter->flags2 & FLAG2_DISABLE_ASPM_L0S)
5914                 aspm_disable_flag = PCIE_LINK_STATE_L0S;
5915         if (adapter->flags2 & FLAG2_DISABLE_ASPM_L1)
5916                 aspm_disable_flag |= PCIE_LINK_STATE_L1;
5917         if (aspm_disable_flag)
5918                 e1000e_disable_aspm(pdev, aspm_disable_flag);
5919
5920         err = pci_enable_device_mem(pdev);
5921         if (err) {
5922                 dev_err(&pdev->dev,
5923                         "Cannot re-enable PCI device after reset.\n");
5924                 result = PCI_ERS_RESULT_DISCONNECT;
5925         } else {
5926                 pci_set_master(pdev);
5927                 pdev->state_saved = true;
5928                 pci_restore_state(pdev);
5929
5930                 pci_enable_wake(pdev, PCI_D3hot, 0);
5931                 pci_enable_wake(pdev, PCI_D3cold, 0);
5932
5933                 e1000e_reset(adapter);
5934                 ew32(WUS, ~0);
5935                 result = PCI_ERS_RESULT_RECOVERED;
5936         }
5937
5938         pci_cleanup_aer_uncorrect_error_status(pdev);
5939
5940         return result;
5941 }
5942
5943 /**
5944  * e1000_io_resume - called when traffic can start flowing again.
5945  * @pdev: Pointer to PCI device
5946  *
5947  * This callback is called when the error recovery driver tells us that
5948  * its OK to resume normal operation. Implementation resembles the
5949  * second-half of the e1000_resume routine.
5950  */
5951 static void e1000_io_resume(struct pci_dev *pdev)
5952 {
5953         struct net_device *netdev = pci_get_drvdata(pdev);
5954         struct e1000_adapter *adapter = netdev_priv(netdev);
5955
5956         e1000_init_manageability_pt(adapter);
5957
5958         if (netif_running(netdev)) {
5959                 if (e1000e_up(adapter)) {
5960                         dev_err(&pdev->dev,
5961                                 "can't bring device back up after reset\n");
5962                         return;
5963                 }
5964         }
5965
5966         netif_device_attach(netdev);
5967
5968         /*
5969          * If the controller has AMT, do not set DRV_LOAD until the interface
5970          * is up.  For all other cases, let the f/w know that the h/w is now
5971          * under the control of the driver.
5972          */
5973         if (!(adapter->flags & FLAG_HAS_AMT))
5974                 e1000e_get_hw_control(adapter);
5975
5976 }
5977
5978 static void e1000_print_device_info(struct e1000_adapter *adapter)
5979 {
5980         struct e1000_hw *hw = &adapter->hw;
5981         struct net_device *netdev = adapter->netdev;
5982         u32 ret_val;
5983         u8 pba_str[E1000_PBANUM_LENGTH];
5984
5985         /* print bus type/speed/width info */
5986         e_info("(PCI Express:2.5GT/s:%s) %pM\n",
5987                /* bus width */
5988                ((hw->bus.width == e1000_bus_width_pcie_x4) ? "Width x4" :
5989                 "Width x1"),
5990                /* MAC address */
5991                netdev->dev_addr);
5992         e_info("Intel(R) PRO/%s Network Connection\n",
5993                (hw->phy.type == e1000_phy_ife) ? "10/100" : "1000");
5994         ret_val = e1000_read_pba_string_generic(hw, pba_str,
5995                                                 E1000_PBANUM_LENGTH);
5996         if (ret_val)
5997                 strlcpy((char *)pba_str, "Unknown", sizeof(pba_str));
5998         e_info("MAC: %d, PHY: %d, PBA No: %s\n",
5999                hw->mac.type, hw->phy.type, pba_str);
6000 }
6001
6002 static void e1000_eeprom_checks(struct e1000_adapter *adapter)
6003 {
6004         struct e1000_hw *hw = &adapter->hw;
6005         int ret_val;
6006         u16 buf = 0;
6007
6008         if (hw->mac.type != e1000_82573)
6009                 return;
6010
6011         ret_val = e1000_read_nvm(hw, NVM_INIT_CONTROL2_REG, 1, &buf);
6012         le16_to_cpus(&buf);
6013         if (!ret_val && (!(buf & (1 << 0)))) {
6014                 /* Deep Smart Power Down (DSPD) */
6015                 dev_warn(&adapter->pdev->dev,
6016                          "Warning: detected DSPD enabled in EEPROM\n");
6017         }
6018 }
6019
6020 static int e1000_set_features(struct net_device *netdev,
6021                               netdev_features_t features)
6022 {
6023         struct e1000_adapter *adapter = netdev_priv(netdev);
6024         netdev_features_t changed = features ^ netdev->features;
6025
6026         if (changed & (NETIF_F_TSO | NETIF_F_TSO6))
6027                 adapter->flags |= FLAG_TSO_FORCE;
6028
6029         if (!(changed & (NETIF_F_HW_VLAN_RX | NETIF_F_HW_VLAN_TX |
6030                          NETIF_F_RXCSUM | NETIF_F_RXHASH | NETIF_F_RXFCS |
6031                          NETIF_F_RXALL)))
6032                 return 0;
6033
6034         /*
6035          * IP payload checksum (enabled with jumbos/packet-split when Rx
6036          * checksum is enabled) and generation of RSS hash is mutually
6037          * exclusive in the hardware.
6038          */
6039         if (adapter->rx_ps_pages &&
6040             (features & NETIF_F_RXCSUM) && (features & NETIF_F_RXHASH)) {
6041                 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");
6042                 return -EINVAL;
6043         }
6044
6045         if (changed & NETIF_F_RXFCS) {
6046                 if (features & NETIF_F_RXFCS) {
6047                         adapter->flags2 &= ~FLAG2_CRC_STRIPPING;
6048                 } else {
6049                         /* We need to take it back to defaults, which might mean
6050                          * stripping is still disabled at the adapter level.
6051                          */
6052                         if (adapter->flags2 & FLAG2_DFLT_CRC_STRIPPING)
6053                                 adapter->flags2 |= FLAG2_CRC_STRIPPING;
6054                         else
6055                                 adapter->flags2 &= ~FLAG2_CRC_STRIPPING;
6056                 }
6057         }
6058
6059         netdev->features = features;
6060
6061         if (netif_running(netdev))
6062                 e1000e_reinit_locked(adapter);
6063         else
6064                 e1000e_reset(adapter);
6065
6066         return 0;
6067 }
6068
6069 static const struct net_device_ops e1000e_netdev_ops = {
6070         .ndo_open               = e1000_open,
6071         .ndo_stop               = e1000_close,
6072         .ndo_start_xmit         = e1000_xmit_frame,
6073         .ndo_get_stats64        = e1000e_get_stats64,
6074         .ndo_set_rx_mode        = e1000e_set_rx_mode,
6075         .ndo_set_mac_address    = e1000_set_mac,
6076         .ndo_change_mtu         = e1000_change_mtu,
6077         .ndo_do_ioctl           = e1000_ioctl,
6078         .ndo_tx_timeout         = e1000_tx_timeout,
6079         .ndo_validate_addr      = eth_validate_addr,
6080
6081         .ndo_vlan_rx_add_vid    = e1000_vlan_rx_add_vid,
6082         .ndo_vlan_rx_kill_vid   = e1000_vlan_rx_kill_vid,
6083 #ifdef CONFIG_NET_POLL_CONTROLLER
6084         .ndo_poll_controller    = e1000_netpoll,
6085 #endif
6086         .ndo_set_features = e1000_set_features,
6087 };
6088
6089 /**
6090  * e1000_probe - Device Initialization Routine
6091  * @pdev: PCI device information struct
6092  * @ent: entry in e1000_pci_tbl
6093  *
6094  * Returns 0 on success, negative on failure
6095  *
6096  * e1000_probe initializes an adapter identified by a pci_dev structure.
6097  * The OS initialization, configuring of the adapter private structure,
6098  * and a hardware reset occur.
6099  **/
6100 static int __devinit e1000_probe(struct pci_dev *pdev,
6101                                  const struct pci_device_id *ent)
6102 {
6103         struct net_device *netdev;
6104         struct e1000_adapter *adapter;
6105         struct e1000_hw *hw;
6106         const struct e1000_info *ei = e1000_info_tbl[ent->driver_data];
6107         resource_size_t mmio_start, mmio_len;
6108         resource_size_t flash_start, flash_len;
6109         static int cards_found;
6110         u16 aspm_disable_flag = 0;
6111         int i, err, pci_using_dac;
6112         u16 eeprom_data = 0;
6113         u16 eeprom_apme_mask = E1000_EEPROM_APME;
6114
6115         if (ei->flags2 & FLAG2_DISABLE_ASPM_L0S)
6116                 aspm_disable_flag = PCIE_LINK_STATE_L0S;
6117         if (ei->flags2 & FLAG2_DISABLE_ASPM_L1)
6118                 aspm_disable_flag |= PCIE_LINK_STATE_L1;
6119         if (aspm_disable_flag)
6120                 e1000e_disable_aspm(pdev, aspm_disable_flag);
6121
6122         err = pci_enable_device_mem(pdev);
6123         if (err)
6124                 return err;
6125
6126         pci_using_dac = 0;
6127         err = dma_set_mask(&pdev->dev, DMA_BIT_MASK(64));
6128         if (!err) {
6129                 err = dma_set_coherent_mask(&pdev->dev, DMA_BIT_MASK(64));
6130                 if (!err)
6131                         pci_using_dac = 1;
6132         } else {
6133                 err = dma_set_mask(&pdev->dev, DMA_BIT_MASK(32));
6134                 if (err) {
6135                         err = dma_set_coherent_mask(&pdev->dev,
6136                                                     DMA_BIT_MASK(32));
6137                         if (err) {
6138                                 dev_err(&pdev->dev, "No usable DMA configuration, aborting\n");
6139                                 goto err_dma;
6140                         }
6141                 }
6142         }
6143
6144         err = pci_request_selected_regions_exclusive(pdev,
6145                                           pci_select_bars(pdev, IORESOURCE_MEM),
6146                                           e1000e_driver_name);
6147         if (err)
6148                 goto err_pci_reg;
6149
6150         /* AER (Advanced Error Reporting) hooks */
6151         pci_enable_pcie_error_reporting(pdev);
6152
6153         pci_set_master(pdev);
6154         /* PCI config space info */
6155         err = pci_save_state(pdev);
6156         if (err)
6157                 goto err_alloc_etherdev;
6158
6159         err = -ENOMEM;
6160         netdev = alloc_etherdev(sizeof(struct e1000_adapter));
6161         if (!netdev)
6162                 goto err_alloc_etherdev;
6163
6164         SET_NETDEV_DEV(netdev, &pdev->dev);
6165
6166         netdev->irq = pdev->irq;
6167
6168         pci_set_drvdata(pdev, netdev);
6169         adapter = netdev_priv(netdev);
6170         hw = &adapter->hw;
6171         adapter->netdev = netdev;
6172         adapter->pdev = pdev;
6173         adapter->ei = ei;
6174         adapter->pba = ei->pba;
6175         adapter->flags = ei->flags;
6176         adapter->flags2 = ei->flags2;
6177         adapter->hw.adapter = adapter;
6178         adapter->hw.mac.type = ei->mac;
6179         adapter->max_hw_frame_size = ei->max_hw_frame_size;
6180         adapter->msg_enable = netif_msg_init(debug, DEFAULT_MSG_ENABLE);
6181
6182         mmio_start = pci_resource_start(pdev, 0);
6183         mmio_len = pci_resource_len(pdev, 0);
6184
6185         err = -EIO;
6186         adapter->hw.hw_addr = ioremap(mmio_start, mmio_len);
6187         if (!adapter->hw.hw_addr)
6188                 goto err_ioremap;
6189
6190         if ((adapter->flags & FLAG_HAS_FLASH) &&
6191             (pci_resource_flags(pdev, 1) & IORESOURCE_MEM)) {
6192                 flash_start = pci_resource_start(pdev, 1);
6193                 flash_len = pci_resource_len(pdev, 1);
6194                 adapter->hw.flash_address = ioremap(flash_start, flash_len);
6195                 if (!adapter->hw.flash_address)
6196                         goto err_flashmap;
6197         }
6198
6199         /* construct the net_device struct */
6200         netdev->netdev_ops              = &e1000e_netdev_ops;
6201         e1000e_set_ethtool_ops(netdev);
6202         netdev->watchdog_timeo          = 5 * HZ;
6203         netif_napi_add(netdev, &adapter->napi, e1000_clean, 64);
6204         strlcpy(netdev->name, pci_name(pdev), sizeof(netdev->name));
6205
6206         netdev->mem_start = mmio_start;
6207         netdev->mem_end = mmio_start + mmio_len;
6208
6209         adapter->bd_number = cards_found++;
6210
6211         e1000e_check_options(adapter);
6212
6213         /* setup adapter struct */
6214         err = e1000_sw_init(adapter);
6215         if (err)
6216                 goto err_sw_init;
6217
6218         memcpy(&hw->mac.ops, ei->mac_ops, sizeof(hw->mac.ops));
6219         memcpy(&hw->nvm.ops, ei->nvm_ops, sizeof(hw->nvm.ops));
6220         memcpy(&hw->phy.ops, ei->phy_ops, sizeof(hw->phy.ops));
6221
6222         err = ei->get_variants(adapter);
6223         if (err)
6224                 goto err_hw_init;
6225
6226         if ((adapter->flags & FLAG_IS_ICH) &&
6227             (adapter->flags & FLAG_READ_ONLY_NVM))
6228                 e1000e_write_protect_nvm_ich8lan(&adapter->hw);
6229
6230         hw->mac.ops.get_bus_info(&adapter->hw);
6231
6232         adapter->hw.phy.autoneg_wait_to_complete = 0;
6233
6234         /* Copper options */
6235         if (adapter->hw.phy.media_type == e1000_media_type_copper) {
6236                 adapter->hw.phy.mdix = AUTO_ALL_MODES;
6237                 adapter->hw.phy.disable_polarity_correction = 0;
6238                 adapter->hw.phy.ms_type = e1000_ms_hw_default;
6239         }
6240
6241         if (hw->phy.ops.check_reset_block(hw))
6242                 e_info("PHY reset is blocked due to SOL/IDER session.\n");
6243
6244         /* Set initial default active device features */
6245         netdev->features = (NETIF_F_SG |
6246                             NETIF_F_HW_VLAN_RX |
6247                             NETIF_F_HW_VLAN_TX |
6248                             NETIF_F_TSO |
6249                             NETIF_F_TSO6 |
6250                             NETIF_F_RXHASH |
6251                             NETIF_F_RXCSUM |
6252                             NETIF_F_HW_CSUM);
6253
6254         /* Set user-changeable features (subset of all device features) */
6255         netdev->hw_features = netdev->features;
6256         netdev->hw_features |= NETIF_F_RXFCS;
6257         netdev->priv_flags |= IFF_SUPP_NOFCS;
6258         netdev->hw_features |= NETIF_F_RXALL;
6259
6260         if (adapter->flags & FLAG_HAS_HW_VLAN_FILTER)
6261                 netdev->features |= NETIF_F_HW_VLAN_FILTER;
6262
6263         netdev->vlan_features |= (NETIF_F_SG |
6264                                   NETIF_F_TSO |
6265                                   NETIF_F_TSO6 |
6266                                   NETIF_F_HW_CSUM);
6267
6268         netdev->priv_flags |= IFF_UNICAST_FLT;
6269
6270         if (pci_using_dac) {
6271                 netdev->features |= NETIF_F_HIGHDMA;
6272                 netdev->vlan_features |= NETIF_F_HIGHDMA;
6273         }
6274
6275         if (e1000e_enable_mng_pass_thru(&adapter->hw))
6276                 adapter->flags |= FLAG_MNG_PT_ENABLED;
6277
6278         /*
6279          * before reading the NVM, reset the controller to
6280          * put the device in a known good starting state
6281          */
6282         adapter->hw.mac.ops.reset_hw(&adapter->hw);
6283
6284         /*
6285          * systems with ASPM and others may see the checksum fail on the first
6286          * attempt. Let's give it a few tries
6287          */
6288         for (i = 0;; i++) {
6289                 if (e1000_validate_nvm_checksum(&adapter->hw) >= 0)
6290                         break;
6291                 if (i == 2) {
6292                         e_err("The NVM Checksum Is Not Valid\n");
6293                         err = -EIO;
6294                         goto err_eeprom;
6295                 }
6296         }
6297
6298         e1000_eeprom_checks(adapter);
6299
6300         /* copy the MAC address */
6301         if (e1000e_read_mac_addr(&adapter->hw))
6302                 e_err("NVM Read Error while reading MAC address\n");
6303
6304         memcpy(netdev->dev_addr, adapter->hw.mac.addr, netdev->addr_len);
6305         memcpy(netdev->perm_addr, adapter->hw.mac.addr, netdev->addr_len);
6306
6307         if (!is_valid_ether_addr(netdev->perm_addr)) {
6308                 e_err("Invalid MAC Address: %pM\n", netdev->perm_addr);
6309                 err = -EIO;
6310                 goto err_eeprom;
6311         }
6312
6313         init_timer(&adapter->watchdog_timer);
6314         adapter->watchdog_timer.function = e1000_watchdog;
6315         adapter->watchdog_timer.data = (unsigned long) adapter;
6316
6317         init_timer(&adapter->phy_info_timer);
6318         adapter->phy_info_timer.function = e1000_update_phy_info;
6319         adapter->phy_info_timer.data = (unsigned long) adapter;
6320
6321         INIT_WORK(&adapter->reset_task, e1000_reset_task);
6322         INIT_WORK(&adapter->watchdog_task, e1000_watchdog_task);
6323         INIT_WORK(&adapter->downshift_task, e1000e_downshift_workaround);
6324         INIT_WORK(&adapter->update_phy_task, e1000e_update_phy_task);
6325         INIT_WORK(&adapter->print_hang_task, e1000_print_hw_hang);
6326
6327         /* Initialize link parameters. User can change them with ethtool */
6328         adapter->hw.mac.autoneg = 1;
6329         adapter->fc_autoneg = true;
6330         adapter->hw.fc.requested_mode = e1000_fc_default;
6331         adapter->hw.fc.current_mode = e1000_fc_default;
6332         adapter->hw.phy.autoneg_advertised = 0x2f;
6333
6334         /* ring size defaults */
6335         adapter->rx_ring->count = 256;
6336         adapter->tx_ring->count = 256;
6337
6338         /*
6339          * Initial Wake on LAN setting - If APM wake is enabled in
6340          * the EEPROM, enable the ACPI Magic Packet filter
6341          */
6342         if (adapter->flags & FLAG_APME_IN_WUC) {
6343                 /* APME bit in EEPROM is mapped to WUC.APME */
6344                 eeprom_data = er32(WUC);
6345                 eeprom_apme_mask = E1000_WUC_APME;
6346                 if ((hw->mac.type > e1000_ich10lan) &&
6347                     (eeprom_data & E1000_WUC_PHY_WAKE))
6348                         adapter->flags2 |= FLAG2_HAS_PHY_WAKEUP;
6349         } else if (adapter->flags & FLAG_APME_IN_CTRL3) {
6350                 if (adapter->flags & FLAG_APME_CHECK_PORT_B &&
6351                     (adapter->hw.bus.func == 1))
6352                         e1000_read_nvm(&adapter->hw, NVM_INIT_CONTROL3_PORT_B,
6353                                        1, &eeprom_data);
6354                 else
6355                         e1000_read_nvm(&adapter->hw, NVM_INIT_CONTROL3_PORT_A,
6356                                        1, &eeprom_data);
6357         }
6358
6359         /* fetch WoL from EEPROM */
6360         if (eeprom_data & eeprom_apme_mask)
6361                 adapter->eeprom_wol |= E1000_WUFC_MAG;
6362
6363         /*
6364          * now that we have the eeprom settings, apply the special cases
6365          * where the eeprom may be wrong or the board simply won't support
6366          * wake on lan on a particular port
6367          */
6368         if (!(adapter->flags & FLAG_HAS_WOL))
6369                 adapter->eeprom_wol = 0;
6370
6371         /* initialize the wol settings based on the eeprom settings */
6372         adapter->wol = adapter->eeprom_wol;
6373         device_set_wakeup_enable(&adapter->pdev->dev, adapter->wol);
6374
6375         /* save off EEPROM version number */
6376         e1000_read_nvm(&adapter->hw, 5, 1, &adapter->eeprom_vers);
6377
6378         /* reset the hardware with the new settings */
6379         e1000e_reset(adapter);
6380
6381         /*
6382          * If the controller has AMT, do not set DRV_LOAD until the interface
6383          * is up.  For all other cases, let the f/w know that the h/w is now
6384          * under the control of the driver.
6385          */
6386         if (!(adapter->flags & FLAG_HAS_AMT))
6387                 e1000e_get_hw_control(adapter);
6388
6389         strlcpy(netdev->name, "eth%d", sizeof(netdev->name));
6390         err = register_netdev(netdev);
6391         if (err)
6392                 goto err_register;
6393
6394         /* carrier off reporting is important to ethtool even BEFORE open */
6395         netif_carrier_off(netdev);
6396
6397         e1000_print_device_info(adapter);
6398
6399         if (pci_dev_run_wake(pdev))
6400                 pm_runtime_put_noidle(&pdev->dev);
6401
6402         return 0;
6403
6404 err_register:
6405         if (!(adapter->flags & FLAG_HAS_AMT))
6406                 e1000e_release_hw_control(adapter);
6407 err_eeprom:
6408         if (!hw->phy.ops.check_reset_block(hw))
6409                 e1000_phy_hw_reset(&adapter->hw);
6410 err_hw_init:
6411         kfree(adapter->tx_ring);
6412         kfree(adapter->rx_ring);
6413 err_sw_init:
6414         if (adapter->hw.flash_address)
6415                 iounmap(adapter->hw.flash_address);
6416         e1000e_reset_interrupt_capability(adapter);
6417 err_flashmap:
6418         iounmap(adapter->hw.hw_addr);
6419 err_ioremap:
6420         free_netdev(netdev);
6421 err_alloc_etherdev:
6422         pci_release_selected_regions(pdev,
6423                                      pci_select_bars(pdev, IORESOURCE_MEM));
6424 err_pci_reg:
6425 err_dma:
6426         pci_disable_device(pdev);
6427         return err;
6428 }
6429
6430 /**
6431  * e1000_remove - Device Removal Routine
6432  * @pdev: PCI device information struct
6433  *
6434  * e1000_remove is called by the PCI subsystem to alert the driver
6435  * that it should release a PCI device.  The could be caused by a
6436  * Hot-Plug event, or because the driver is going to be removed from
6437  * memory.
6438  **/
6439 static void __devexit e1000_remove(struct pci_dev *pdev)
6440 {
6441         struct net_device *netdev = pci_get_drvdata(pdev);
6442         struct e1000_adapter *adapter = netdev_priv(netdev);
6443         bool down = test_bit(__E1000_DOWN, &adapter->state);
6444
6445         /*
6446          * The timers may be rescheduled, so explicitly disable them
6447          * from being rescheduled.
6448          */
6449         if (!down)
6450                 set_bit(__E1000_DOWN, &adapter->state);
6451         del_timer_sync(&adapter->watchdog_timer);
6452         del_timer_sync(&adapter->phy_info_timer);
6453
6454         cancel_work_sync(&adapter->reset_task);
6455         cancel_work_sync(&adapter->watchdog_task);
6456         cancel_work_sync(&adapter->downshift_task);
6457         cancel_work_sync(&adapter->update_phy_task);
6458         cancel_work_sync(&adapter->print_hang_task);
6459
6460         if (!(netdev->flags & IFF_UP))
6461                 e1000_power_down_phy(adapter);
6462
6463         /* Don't lie to e1000_close() down the road. */
6464         if (!down)
6465                 clear_bit(__E1000_DOWN, &adapter->state);
6466         unregister_netdev(netdev);
6467
6468         if (pci_dev_run_wake(pdev))
6469                 pm_runtime_get_noresume(&pdev->dev);
6470
6471         /*
6472          * Release control of h/w to f/w.  If f/w is AMT enabled, this
6473          * would have already happened in close and is redundant.
6474          */
6475         e1000e_release_hw_control(adapter);
6476
6477         e1000e_reset_interrupt_capability(adapter);
6478         kfree(adapter->tx_ring);
6479         kfree(adapter->rx_ring);
6480
6481         iounmap(adapter->hw.hw_addr);
6482         if (adapter->hw.flash_address)
6483                 iounmap(adapter->hw.flash_address);
6484         pci_release_selected_regions(pdev,
6485                                      pci_select_bars(pdev, IORESOURCE_MEM));
6486
6487         free_netdev(netdev);
6488
6489         /* AER disable */
6490         pci_disable_pcie_error_reporting(pdev);
6491
6492         pci_disable_device(pdev);
6493 }
6494
6495 /* PCI Error Recovery (ERS) */
6496 static struct pci_error_handlers e1000_err_handler = {
6497         .error_detected = e1000_io_error_detected,
6498         .slot_reset = e1000_io_slot_reset,
6499         .resume = e1000_io_resume,
6500 };
6501
6502 static DEFINE_PCI_DEVICE_TABLE(e1000_pci_tbl) = {
6503         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82571EB_COPPER), board_82571 },
6504         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82571EB_FIBER), board_82571 },
6505         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82571EB_QUAD_COPPER), board_82571 },
6506         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82571EB_QUAD_COPPER_LP), board_82571 },
6507         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82571EB_QUAD_FIBER), board_82571 },
6508         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82571EB_SERDES), board_82571 },
6509         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82571EB_SERDES_DUAL), board_82571 },
6510         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82571EB_SERDES_QUAD), board_82571 },
6511         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82571PT_QUAD_COPPER), board_82571 },
6512
6513         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82572EI), board_82572 },
6514         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82572EI_COPPER), board_82572 },
6515         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82572EI_FIBER), board_82572 },
6516         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82572EI_SERDES), board_82572 },
6517
6518         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82573E), board_82573 },
6519         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82573E_IAMT), board_82573 },
6520         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82573L), board_82573 },
6521
6522         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82574L), board_82574 },
6523         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82574LA), board_82574 },
6524         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82583V), board_82583 },
6525
6526         { PCI_VDEVICE(INTEL, E1000_DEV_ID_80003ES2LAN_COPPER_DPT),
6527           board_80003es2lan },
6528         { PCI_VDEVICE(INTEL, E1000_DEV_ID_80003ES2LAN_COPPER_SPT),
6529           board_80003es2lan },
6530         { PCI_VDEVICE(INTEL, E1000_DEV_ID_80003ES2LAN_SERDES_DPT),
6531           board_80003es2lan },
6532         { PCI_VDEVICE(INTEL, E1000_DEV_ID_80003ES2LAN_SERDES_SPT),
6533           board_80003es2lan },
6534
6535         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH8_IFE), board_ich8lan },
6536         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH8_IFE_G), board_ich8lan },
6537         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH8_IFE_GT), board_ich8lan },
6538         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH8_IGP_AMT), board_ich8lan },
6539         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH8_IGP_C), board_ich8lan },
6540         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH8_IGP_M), board_ich8lan },
6541         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH8_IGP_M_AMT), board_ich8lan },
6542         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH8_82567V_3), board_ich8lan },
6543
6544         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH9_IFE), board_ich9lan },
6545         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH9_IFE_G), board_ich9lan },
6546         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH9_IFE_GT), board_ich9lan },
6547         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH9_IGP_AMT), board_ich9lan },
6548         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH9_IGP_C), board_ich9lan },
6549         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH9_BM), board_ich9lan },
6550         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH9_IGP_M), board_ich9lan },
6551         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH9_IGP_M_AMT), board_ich9lan },
6552         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH9_IGP_M_V), board_ich9lan },
6553
6554         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH10_R_BM_LM), board_ich9lan },
6555         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH10_R_BM_LF), board_ich9lan },
6556         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH10_R_BM_V), board_ich9lan },
6557
6558         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH10_D_BM_LM), board_ich10lan },
6559         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH10_D_BM_LF), board_ich10lan },
6560         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH10_D_BM_V), board_ich10lan },
6561
6562         { PCI_VDEVICE(INTEL, E1000_DEV_ID_PCH_M_HV_LM), board_pchlan },
6563         { PCI_VDEVICE(INTEL, E1000_DEV_ID_PCH_M_HV_LC), board_pchlan },
6564         { PCI_VDEVICE(INTEL, E1000_DEV_ID_PCH_D_HV_DM), board_pchlan },
6565         { PCI_VDEVICE(INTEL, E1000_DEV_ID_PCH_D_HV_DC), board_pchlan },
6566
6567         { PCI_VDEVICE(INTEL, E1000_DEV_ID_PCH2_LV_LM), board_pch2lan },
6568         { PCI_VDEVICE(INTEL, E1000_DEV_ID_PCH2_LV_V), board_pch2lan },
6569
6570         { 0, 0, 0, 0, 0, 0, 0 } /* terminate list */
6571 };
6572 MODULE_DEVICE_TABLE(pci, e1000_pci_tbl);
6573
6574 #ifdef CONFIG_PM
6575 static const struct dev_pm_ops e1000_pm_ops = {
6576         SET_SYSTEM_SLEEP_PM_OPS(e1000_suspend, e1000_resume)
6577         SET_RUNTIME_PM_OPS(e1000_runtime_suspend,
6578                                 e1000_runtime_resume, e1000_idle)
6579 };
6580 #endif
6581
6582 /* PCI Device API Driver */
6583 static struct pci_driver e1000_driver = {
6584         .name     = e1000e_driver_name,
6585         .id_table = e1000_pci_tbl,
6586         .probe    = e1000_probe,
6587         .remove   = __devexit_p(e1000_remove),
6588 #ifdef CONFIG_PM
6589         .driver   = {
6590                 .pm = &e1000_pm_ops,
6591         },
6592 #endif
6593         .shutdown = e1000_shutdown,
6594         .err_handler = &e1000_err_handler
6595 };
6596
6597 /**
6598  * e1000_init_module - Driver Registration Routine
6599  *
6600  * e1000_init_module is the first routine called when the driver is
6601  * loaded. All it does is register with the PCI subsystem.
6602  **/
6603 static int __init e1000_init_module(void)
6604 {
6605         int ret;
6606         pr_info("Intel(R) PRO/1000 Network Driver - %s\n",
6607                 e1000e_driver_version);
6608         pr_info("Copyright(c) 1999 - 2012 Intel Corporation.\n");
6609         ret = pci_register_driver(&e1000_driver);
6610
6611         return ret;
6612 }
6613 module_init(e1000_init_module);
6614
6615 /**
6616  * e1000_exit_module - Driver Exit Cleanup Routine
6617  *
6618  * e1000_exit_module is called just before the driver is removed
6619  * from memory.
6620  **/
6621 static void __exit e1000_exit_module(void)
6622 {
6623         pci_unregister_driver(&e1000_driver);
6624 }
6625 module_exit(e1000_exit_module);
6626
6627
6628 MODULE_AUTHOR("Intel Corporation, <linux.nics@intel.com>");
6629 MODULE_DESCRIPTION("Intel(R) PRO/1000 Network Driver");
6630 MODULE_LICENSE("GPL");
6631 MODULE_VERSION(DRV_VERSION);
6632
6633 /* netdev.c */