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

   1 /*******************************************************************************
   2
   3   Intel PRO/1000 Linux driver
   4   Copyright(c) 1999 - 2012 Intel Corporation.
   5
   6   This program is free software; you can redistribute it and/or modify it
   7   under the terms and conditions of the GNU General Public License,
   8   version 2, as published by the Free Software Foundation.
   9
  10   This program is distributed in the hope it will be useful, but WITHOUT
  11   ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
  12   FITNESS FOR A PARTICULAR PURPOSE.  See the GNU General Public License for
  13   more details.
  14
  15   You should have received a copy of the GNU General Public License along with
  16   this program; if not, write to the Free Software Foundation, Inc.,
  17   51 Franklin St - Fifth Floor, Boston, MA 02110-1301 USA.
  18
  19   The full GNU General Public License is included in this distribution in
  20   the file called "COPYING".
  21
  22   Contact Information:
  23   Linux NICS <linux.nics@intel.com>
  24   e1000-devel Mailing List <e1000-devel@lists.sourceforge.net>
  25   Intel Corporation, 5200 N.E. Elam Young Parkway, Hillsboro, OR 97124-6497
  26
  27 *******************************************************************************/
  28
  29 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
  30
  31 #include <linux/module.h>
  32 #include <linux/types.h>
  33 #include <linux/init.h>
  34 #include <linux/pci.h>
  35 #include <linux/vmalloc.h>
  36 #include <linux/pagemap.h>
  37 #include <linux/delay.h>
  38 #include <linux/netdevice.h>
  39 #include <linux/interrupt.h>
  40 #include <linux/tcp.h>
  41 #include <linux/ipv6.h>
  42 #include <linux/slab.h>
  43 #include <net/checksum.h>
  44 #include <net/ip6_checksum.h>
  45 #include <linux/mii.h>
  46 #include <linux/ethtool.h>
  47 #include <linux/if_vlan.h>
  48 #include <linux/cpu.h>
  49 #include <linux/smp.h>
  50 #include <linux/pm_qos.h>
  51 #include <linux/pm_runtime.h>
  52 #include <linux/aer.h>
  53 #include <linux/prefetch.h>
  54
  55 #include "e1000.h"
  56
  57 #define DRV_EXTRAVERSION "-k"
  58
  59 #define DRV_VERSION "1.10.6" DRV_EXTRAVERSION
  60 char e1000e_driver_name[] = "e1000e";
  61 const char e1000e_driver_version[] = DRV_VERSION;
  62
  63 #define DEFAULT_MSG_ENABLE (NETIF_MSG_DRV|NETIF_MSG_PROBE|NETIF_MSG_LINK)
  64 static int debug = -1;
  65 module_param(debug, int, 0);
  66 MODULE_PARM_DESC(debug, "Debug level (0=none,...,16=all)");
  67
  68 static void e1000e_disable_aspm(struct pci_dev *pdev, u16 state);
  69
  70 static const struct e1000_info *e1000_info_tbl[] = {
  71         [board_82571]           = &e1000_82571_info,
  72         [board_82572]           = &e1000_82572_info,
  73         [board_82573]           = &e1000_82573_info,
  74         [board_82574]           = &e1000_82574_info,
  75         [board_82583]           = &e1000_82583_info,
  76         [board_80003es2lan]     = &e1000_es2_info,
  77         [board_ich8lan]         = &e1000_ich8_info,
  78         [board_ich9lan]         = &e1000_ich9_info,
  79         [board_ich10lan]        = &e1000_ich10_info,
  80         [board_pchlan]          = &e1000_pch_info,
  81         [board_pch2lan]         = &e1000_pch2_info,
  82 };
  83
  84 struct e1000_reg_info {
  85         u32 ofs;
  86         char *name;
  87 };
  88
  89 #define E1000_RDFH      0x02410 /* Rx Data FIFO Head - RW */
  90 #define E1000_RDFT      0x02418 /* Rx Data FIFO Tail - RW */
  91 #define E1000_RDFHS     0x02420 /* Rx Data FIFO Head Saved - RW */
  92 #define E1000_RDFTS     0x02428 /* Rx Data FIFO Tail Saved - RW */
  93 #define E1000_RDFPC     0x02430 /* Rx Data FIFO Packet Count - RW */
  94
  95 #define E1000_TDFH      0x03410 /* Tx Data FIFO Head - RW */
  96 #define E1000_TDFT      0x03418 /* Tx Data FIFO Tail - RW */
  97 #define E1000_TDFHS     0x03420 /* Tx Data FIFO Head Saved - RW */
  98 #define E1000_TDFTS     0x03428 /* Tx Data FIFO Tail Saved - RW */
  99 #define E1000_TDFPC     0x03430 /* Tx Data FIFO Packet Count - RW */
 100
 101 static const struct e1000_reg_info e1000_reg_info_tbl[] = {
 102
 103         /* General Registers */
 104         {E1000_CTRL, "CTRL"},
 105         {E1000_STATUS, "STATUS"},
 106         {E1000_CTRL_EXT, "CTRL_EXT"},
 107
 108         /* Interrupt Registers */
 109         {E1000_ICR, "ICR"},
 110
 111         /* Rx Registers */
 112         {E1000_RCTL, "RCTL"},
 113         {E1000_RDLEN(0), "RDLEN"},
 114         {E1000_RDH(0), "RDH"},
 115         {E1000_RDT(0), "RDT"},
 116         {E1000_RDTR, "RDTR"},
 117         {E1000_RXDCTL(0), "RXDCTL"},
 118         {E1000_ERT, "ERT"},
 119         {E1000_RDBAL(0), "RDBAL"},
 120         {E1000_RDBAH(0), "RDBAH"},
 121         {E1000_RDFH, "RDFH"},
 122         {E1000_RDFT, "RDFT"},
 123         {E1000_RDFHS, "RDFHS"},
 124         {E1000_RDFTS, "RDFTS"},
 125         {E1000_RDFPC, "RDFPC"},
 126
 127         /* Tx Registers */
 128         {E1000_TCTL, "TCTL"},
 129         {E1000_TDBAL(0), "TDBAL"},
 130         {E1000_TDBAH(0), "TDBAH"},
 131         {E1000_TDLEN(0), "TDLEN"},
 132         {E1000_TDH(0), "TDH"},
 133         {E1000_TDT(0), "TDT"},
 134         {E1000_TIDV, "TIDV"},
 135         {E1000_TXDCTL(0), "TXDCTL"},
 136         {E1000_TADV, "TADV"},
 137         {E1000_TARC(0), "TARC"},
 138         {E1000_TDFH, "TDFH"},
 139         {E1000_TDFT, "TDFT"},
 140         {E1000_TDFHS, "TDFHS"},
 141         {E1000_TDFTS, "TDFTS"},
 142         {E1000_TDFPC, "TDFPC"},
 143
 144         /* List Terminator */
 145         {0, NULL}
 146 };
 147
 148 /*
 149  * e1000_regdump - register printout routine
 150  */
 151 static void e1000_regdump(struct e1000_hw *hw, struct e1000_reg_info *reginfo)
 152 {
 153         int n = 0;
 154         char rname[16];
 155         u32 regs[8];
 156
 157         switch (reginfo->ofs) {
 158         case E1000_RXDCTL(0):
 159                 for (n = 0; n < 2; n++)
 160                         regs[n] = __er32(hw, E1000_RXDCTL(n));
 161                 break;
 162         case E1000_TXDCTL(0):
 163                 for (n = 0; n < 2; n++)
 164                         regs[n] = __er32(hw, E1000_TXDCTL(n));
 165                 break;
 166         case E1000_TARC(0):
 167                 for (n = 0; n < 2; n++)
 168                         regs[n] = __er32(hw, E1000_TARC(n));
 169                 break;
 170         default:
 171                 pr_info("%-15s %08x\n",
 172                         reginfo->name, __er32(hw, reginfo->ofs));
 173                 return;
 174         }
 175
 176         snprintf(rname, 16, "%s%s", reginfo->name, "[0-1]");
 177         pr_info("%-15s %08x %08x\n", rname, regs[0], regs[1]);
 178 }
 179
 180 /*
 181  * e1000e_dump - Print registers, Tx-ring and Rx-ring
 182  */
 183 static void e1000e_dump(struct e1000_adapter *adapter)
 184 {
 185         struct net_device *netdev = adapter->netdev;
 186         struct e1000_hw *hw = &adapter->hw;
 187         struct e1000_reg_info *reginfo;
 188         struct e1000_ring *tx_ring = adapter->tx_ring;
 189         struct e1000_tx_desc *tx_desc;
 190         struct my_u0 {
 191                 __le64 a;
 192                 __le64 b;
 193         } *u0;
 194         struct e1000_buffer *buffer_info;
 195         struct e1000_ring *rx_ring = adapter->rx_ring;
 196         union e1000_rx_desc_packet_split *rx_desc_ps;
 197         union e1000_rx_desc_extended *rx_desc;
 198         struct my_u1 {
 199                 __le64 a;
 200                 __le64 b;
 201                 __le64 c;
 202                 __le64 d;
 203         } *u1;
 204         u32 staterr;
 205         int i = 0;
 206
 207         if (!netif_msg_hw(adapter))
 208                 return;
 209
 210         /* Print netdevice Info */
 211         if (netdev) {
 212                 dev_info(&adapter->pdev->dev, "Net device Info\n");
 213                 pr_info("Device Name     state            trans_start      last_rx\n");
 214                 pr_info("%-15s %016lX %016lX %016lX\n",
 215                         netdev->name, netdev->state, netdev->trans_start,
 216                         netdev->last_rx);
 217         }
 218
 219         /* Print Registers */
 220         dev_info(&adapter->pdev->dev, "Register Dump\n");
 221         pr_info(" Register Name   Value\n");
 222         for (reginfo = (struct e1000_reg_info *)e1000_reg_info_tbl;
 223              reginfo->name; reginfo++) {
 224                 e1000_regdump(hw, reginfo);
 225         }
 226
 227         /* Print Tx Ring Summary */
 228         if (!netdev || !netif_running(netdev))
 229                 return;
 230
 231         dev_info(&adapter->pdev->dev, "Tx Ring Summary\n");
 232         pr_info("Queue [NTU] [NTC] [bi(ntc)->dma  ] leng ntw timestamp\n");
 233         buffer_info = &tx_ring->buffer_info[tx_ring->next_to_clean];
 234         pr_info(" %5d %5X %5X %016llX %04X %3X %016llX\n",
 235                 0, tx_ring->next_to_use, tx_ring->next_to_clean,
 236                 (unsigned long long)buffer_info->dma,
 237                 buffer_info->length,
 238                 buffer_info->next_to_watch,
 239                 (unsigned long long)buffer_info->time_stamp);
 240
 241         /* Print Tx Ring */
 242         if (!netif_msg_tx_done(adapter))
 243                 goto rx_ring_summary;
 244
 245         dev_info(&adapter->pdev->dev, "Tx Ring Dump\n");
 246
 247         /* Transmit Descriptor Formats - DEXT[29] is 0 (Legacy) or 1 (Extended)
 248          *
 249          * Legacy Transmit Descriptor
 250          *   +--------------------------------------------------------------+
 251          * 0 |         Buffer Address [63:0] (Reserved on Write Back)       |
 252          *   +--------------------------------------------------------------+
 253          * 8 | Special  |    CSS     | Status |  CMD    |  CSO   |  Length  |
 254          *   +--------------------------------------------------------------+
 255          *   63       48 47        36 35    32 31     24 23    16 15        0
 256          *
 257          * Extended Context Descriptor (DTYP=0x0) for TSO or checksum offload
 258          *   63      48 47    40 39       32 31             16 15    8 7      0
 259          *   +----------------------------------------------------------------+
 260          * 0 |  TUCSE  | TUCS0  |   TUCSS   |     IPCSE       | IPCS0 | IPCSS |
 261          *   +----------------------------------------------------------------+
 262          * 8 |   MSS   | HDRLEN | RSV | STA | TUCMD | DTYP |      PAYLEN      |
 263          *   +----------------------------------------------------------------+
 264          *   63      48 47    40 39 36 35 32 31   24 23  20 19                0
 265          *
 266          * Extended Data Descriptor (DTYP=0x1)
 267          *   +----------------------------------------------------------------+
 268          * 0 |                     Buffer Address [63:0]                      |
 269          *   +----------------------------------------------------------------+
 270          * 8 | VLAN tag |  POPTS  | Rsvd | Status | Command | DTYP |  DTALEN  |
 271          *   +----------------------------------------------------------------+
 272          *   63       48 47     40 39  36 35    32 31     24 23  20 19        0
 273          */
 274         pr_info("Tl[desc]     [address 63:0  ] [SpeCssSCmCsLen] [bi->dma       ] leng  ntw timestamp        bi->skb <-- Legacy format\n");
 275         pr_info("Tc[desc]     [Ce CoCsIpceCoS] [MssHlRSCm0Plen] [bi->dma       ] leng  ntw timestamp        bi->skb <-- Ext Context format\n");
 276         pr_info("Td[desc]     [address 63:0  ] [VlaPoRSCm1Dlen] [bi->dma       ] leng  ntw timestamp        bi->skb <-- Ext Data format\n");
 277         for (i = 0; tx_ring->desc && (i < tx_ring->count); i++) {
 278                 const char *next_desc;
 279                 tx_desc = E1000_TX_DESC(*tx_ring, i);
 280                 buffer_info = &tx_ring->buffer_info[i];
 281                 u0 = (struct my_u0 *)tx_desc;
 282                 if (i == tx_ring->next_to_use && i == tx_ring->next_to_clean)
 283                         next_desc = " NTC/U";
 284                 else if (i == tx_ring->next_to_use)
 285                         next_desc = " NTU";
 286                 else if (i == tx_ring->next_to_clean)
 287                         next_desc = " NTC";
 288                 else
 289                         next_desc = "";
 290                 pr_info("T%c[0x%03X]    %016llX %016llX %016llX %04X  %3X %016llX %p%s\n",
 291                         (!(le64_to_cpu(u0->b) & (1 << 29)) ? 'l' :
 292                          ((le64_to_cpu(u0->b) & (1 << 20)) ? 'd' : 'c')),
 293                         i,
 294                         (unsigned long long)le64_to_cpu(u0->a),
 295                         (unsigned long long)le64_to_cpu(u0->b),
 296                         (unsigned long long)buffer_info->dma,
 297                         buffer_info->length, buffer_info->next_to_watch,
 298                         (unsigned long long)buffer_info->time_stamp,
 299                         buffer_info->skb, next_desc);
 300
 301                 if (netif_msg_pktdata(adapter) && buffer_info->dma != 0)
 302                         print_hex_dump(KERN_INFO, "", DUMP_PREFIX_ADDRESS,
 303                                        16, 1, phys_to_virt(buffer_info->dma),
 304                                        buffer_info->length, true);
 305         }
 306
 307         /* Print Rx Ring Summary */
 308 rx_ring_summary:
 309         dev_info(&adapter->pdev->dev, "Rx Ring Summary\n");
 310         pr_info("Queue [NTU] [NTC]\n");
 311         pr_info(" %5d %5X %5X\n",
 312                 0, rx_ring->next_to_use, rx_ring->next_to_clean);
 313
 314         /* Print Rx Ring */
 315         if (!netif_msg_rx_status(adapter))
 316                 return;
 317
 318         dev_info(&adapter->pdev->dev, "Rx Ring Dump\n");
 319         switch (adapter->rx_ps_pages) {
 320         case 1:
 321         case 2:
 322         case 3:
 323                 /* [Extended] Packet Split Receive Descriptor Format
 324                  *
 325                  *    +-----------------------------------------------------+
 326                  *  0 |                Buffer Address 0 [63:0]              |
 327                  *    +-----------------------------------------------------+
 328                  *  8 |                Buffer Address 1 [63:0]              |
 329                  *    +-----------------------------------------------------+
 330                  * 16 |                Buffer Address 2 [63:0]              |
 331                  *    +-----------------------------------------------------+
 332                  * 24 |                Buffer Address 3 [63:0]              |
 333                  *    +-----------------------------------------------------+
 334                  */
 335                 pr_info("R  [desc]      [buffer 0 63:0 ] [buffer 1 63:0 ] [buffer 2 63:0 ] [buffer 3 63:0 ] [bi->dma       ] [bi->skb] <-- Ext Pkt Split format\n");
 336                 /* [Extended] Receive Descriptor (Write-Back) Format
 337                  *
 338                  *   63       48 47    32 31     13 12    8 7    4 3        0
 339                  *   +------------------------------------------------------+
 340                  * 0 | Packet   | IP     |  Rsvd   | MRQ   | Rsvd | MRQ RSS |
 341                  *   | Checksum | Ident  |         | Queue |      |  Type   |
 342                  *   +------------------------------------------------------+
 343                  * 8 | VLAN Tag | Length | Extended Error | Extended Status |
 344                  *   +------------------------------------------------------+
 345                  *   63       48 47    32 31            20 19               0
 346                  */
 347                 pr_info("RWB[desc]      [ck ipid mrqhsh] [vl   l0 ee  es] [ l3  l2  l1 hs] [reserved      ] ---------------- [bi->skb] <-- Ext Rx Write-Back format\n");
 348                 for (i = 0; i < rx_ring->count; i++) {
 349                         const char *next_desc;
 350                         buffer_info = &rx_ring->buffer_info[i];
 351                         rx_desc_ps = E1000_RX_DESC_PS(*rx_ring, i);
 352                         u1 = (struct my_u1 *)rx_desc_ps;
 353                         staterr =
 354                             le32_to_cpu(rx_desc_ps->wb.middle.status_error);
 355
 356                         if (i == rx_ring->next_to_use)
 357                                 next_desc = " NTU";
 358                         else if (i == rx_ring->next_to_clean)
 359                                 next_desc = " NTC";
 360                         else
 361                                 next_desc = "";
 362
 363                         if (staterr & E1000_RXD_STAT_DD) {
 364                                 /* Descriptor Done */
 365                                 pr_info("%s[0x%03X]     %016llX %016llX %016llX %016llX ---------------- %p%s\n",
 366                                         "RWB", i,
 367                                         (unsigned long long)le64_to_cpu(u1->a),
 368                                         (unsigned long long)le64_to_cpu(u1->b),
 369                                         (unsigned long long)le64_to_cpu(u1->c),
 370                                         (unsigned long long)le64_to_cpu(u1->d),
 371                                         buffer_info->skb, next_desc);
 372                         } else {
 373                                 pr_info("%s[0x%03X]     %016llX %016llX %016llX %016llX %016llX %p%s\n",
 374                                         "R  ", i,
 375                                         (unsigned long long)le64_to_cpu(u1->a),
 376                                         (unsigned long long)le64_to_cpu(u1->b),
 377                                         (unsigned long long)le64_to_cpu(u1->c),
 378                                         (unsigned long long)le64_to_cpu(u1->d),
 379                                         (unsigned long long)buffer_info->dma,
 380                                         buffer_info->skb, next_desc);
 381
 382                                 if (netif_msg_pktdata(adapter))
 383                                         print_hex_dump(KERN_INFO, "",
 384                                                 DUMP_PREFIX_ADDRESS, 16, 1,
 385                                                 phys_to_virt(buffer_info->dma),
 386                                                 adapter->rx_ps_bsize0, true);
 387                         }
 388                 }
 389                 break;
 390         default:
 391         case 0:
 392                 /* Extended Receive Descriptor (Read) Format
 393                  *
 394                  *   +-----------------------------------------------------+
 395                  * 0 |                Buffer Address [63:0]                |
 396                  *   +-----------------------------------------------------+
 397                  * 8 |                      Reserved                       |
 398                  *   +-----------------------------------------------------+
 399                  */
 400                 pr_info("R  [desc]      [buf addr 63:0 ] [reserved 63:0 ] [bi->dma       ] [bi->skb] <-- Ext (Read) format\n");
 401                 /* Extended Receive Descriptor (Write-Back) Format
 402                  *
 403                  *   63       48 47    32 31    24 23            4 3        0
 404                  *   +------------------------------------------------------+
 405                  *   |     RSS Hash      |        |               |         |
 406                  * 0 +-------------------+  Rsvd  |   Reserved    | MRQ RSS |
 407                  *   | Packet   | IP     |        |               |  Type   |
 408                  *   | Checksum | Ident  |        |               |         |
 409                  *   +------------------------------------------------------+
 410                  * 8 | VLAN Tag | Length | Extended Error | Extended Status |
 411                  *   +------------------------------------------------------+
 412                  *   63       48 47    32 31            20 19               0
 413                  */
 414                 pr_info("RWB[desc]      [cs ipid    mrq] [vt   ln xe  xs] [bi->skb] <-- Ext (Write-Back) format\n");
 415
 416                 for (i = 0; i < rx_ring->count; i++) {
 417                         const char *next_desc;
 418
 419                         buffer_info = &rx_ring->buffer_info[i];
 420                         rx_desc = E1000_RX_DESC_EXT(*rx_ring, i);
 421                         u1 = (struct my_u1 *)rx_desc;
 422                         staterr = le32_to_cpu(rx_desc->wb.upper.status_error);
 423
 424                         if (i == rx_ring->next_to_use)
 425                                 next_desc = " NTU";
 426                         else if (i == rx_ring->next_to_clean)
 427                                 next_desc = " NTC";
 428                         else
 429                                 next_desc = "";
 430
 431                         if (staterr & E1000_RXD_STAT_DD) {
 432                                 /* Descriptor Done */
 433                                 pr_info("%s[0x%03X]     %016llX %016llX ---------------- %p%s\n",
 434                                         "RWB", i,
 435                                         (unsigned long long)le64_to_cpu(u1->a),
 436                                         (unsigned long long)le64_to_cpu(u1->b),
 437                                         buffer_info->skb, next_desc);
 438                         } else {
 439                                 pr_info("%s[0x%03X]     %016llX %016llX %016llX %p%s\n",
 440                                         "R  ", i,
 441                                         (unsigned long long)le64_to_cpu(u1->a),
 442                                         (unsigned long long)le64_to_cpu(u1->b),
 443                                         (unsigned long long)buffer_info->dma,
 444                                         buffer_info->skb, next_desc);
 445
 446                                 if (netif_msg_pktdata(adapter))
 447                                         print_hex_dump(KERN_INFO, "",
 448                                                        DUMP_PREFIX_ADDRESS, 16,
 449                                                        1,
 450                                                        phys_to_virt
 451                                                        (buffer_info->dma),
 452                                                        adapter->rx_buffer_len,
 453                                                        true);
 454                         }
 455                 }
 456         }
 457 }
 458
 459 /**
 460  * e1000_desc_unused - calculate if we have unused descriptors
 461  **/
 462 static int e1000_desc_unused(struct e1000_ring *ring)
 463 {
 464         if (ring->next_to_clean > ring->next_to_use)
 465                 return ring->next_to_clean - ring->next_to_use - 1;
 466
 467         return ring->count + ring->next_to_clean - ring->next_to_use - 1;
 468 }
 469
 470 /**
 471  * e1000_receive_skb - helper function to handle Rx indications
 472  * @adapter: board private structure
 473  * @status: descriptor status field as written by hardware
 474  * @vlan: descriptor vlan field as written by hardware (no le/be conversion)
 475  * @skb: pointer to sk_buff to be indicated to stack
 476  **/
 477 static void e1000_receive_skb(struct e1000_adapter *adapter,
 478                               struct net_device *netdev, struct sk_buff *skb,
 479                               u8 status, __le16 vlan)
 480 {
 481         u16 tag = le16_to_cpu(vlan);
 482         skb->protocol = eth_type_trans(skb, netdev);
 483
 484         if (status & E1000_RXD_STAT_VP)
 485                 __vlan_hwaccel_put_tag(skb, tag);
 486
 487         napi_gro_receive(&adapter->napi, skb);
 488 }
 489
 490 /**
 491  * e1000_rx_checksum - Receive Checksum Offload
 492  * @adapter: board private structure
 493  * @status_err: receive descriptor status and error fields
 494  * @csum: receive descriptor csum field
 495  * @sk_buff: socket buffer with received data
 496  **/
 497 static void e1000_rx_checksum(struct e1000_adapter *adapter, u32 status_err,
 498                               __le16 csum, struct sk_buff *skb)
 499 {
 500         u16 status = (u16)status_err;
 501         u8 errors = (u8)(status_err >> 24);
 502
 503         skb_checksum_none_assert(skb);
 504
 505         /* Rx checksum disabled */
 506         if (!(adapter->netdev->features & NETIF_F_RXCSUM))
 507                 return;
 508
 509         /* Ignore Checksum bit is set */
 510         if (status & E1000_RXD_STAT_IXSM)
 511                 return;
 512
 513         /* TCP/UDP checksum error bit is set */
 514         if (errors & E1000_RXD_ERR_TCPE) {
 515                 /* let the stack verify checksum errors */
 516                 adapter->hw_csum_err++;
 517                 return;
 518         }
 519
 520         /* TCP/UDP Checksum has not been calculated */
 521         if (!(status & (E1000_RXD_STAT_TCPCS | E1000_RXD_STAT_UDPCS)))
 522                 return;
 523
 524         /* It must be a TCP or UDP packet with a valid checksum */
 525         if (status & E1000_RXD_STAT_TCPCS) {
 526                 /* TCP checksum is good */
 527                 skb->ip_summed = CHECKSUM_UNNECESSARY;
 528         } else {
 529                 /*
 530                  * IP fragment with UDP payload
 531                  * Hardware complements the payload checksum, so we undo it
 532                  * and then put the value in host order for further stack use.
 533                  */
 534                 __sum16 sum = (__force __sum16)swab16((__force u16)csum);
 535                 skb->csum = csum_unfold(~sum);
 536                 skb->ip_summed = CHECKSUM_COMPLETE;
 537         }
 538         adapter->hw_csum_good++;
 539 }
 540
 541 /**
 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                 /*
1057                  * May be block on write-back, flush and detect again
1058                  * flush pending descriptor writebacks to memory
1059                  */
1060                 ew32(TIDV, adapter->tx_int_delay | E1000_TIDV_FPD);
1061                 /* execute the writes immediately */
1062                 e1e_flush();
1063                 /*
1064                  * Due to rare timing issues, write to TIDV again to ensure
1065                  * the write is successful
1066                  */
1067                 ew32(TIDV, adapter->tx_int_delay | E1000_TIDV_FPD);
1068                 /* execute the writes immediately */
1069                 e1e_flush();
1070                 adapter->tx_hang_recheck = true;
1071                 return;
1072         }
1073         /* Real hang detected */
1074         adapter->tx_hang_recheck = false;
1075         netif_stop_queue(netdev);
1076
1077         e1e_rphy(hw, PHY_STATUS, &phy_status);
1078         e1e_rphy(hw, PHY_1000T_STATUS, &phy_1000t_status);
1079         e1e_rphy(hw, PHY_EXT_STATUS, &phy_ext_status);
1080
1081         pci_read_config_word(adapter->pdev, PCI_STATUS, &pci_status);
1082
1083         /* detected Hardware unit hang */
1084         e_err("Detected Hardware Unit Hang:\n"
1085               "  TDH                  <%x>\n"
1086               "  TDT                  <%x>\n"
1087               "  next_to_use          <%x>\n"
1088               "  next_to_clean        <%x>\n"
1089               "buffer_info[next_to_clean]:\n"
1090               "  time_stamp           <%lx>\n"
1091               "  next_to_watch        <%x>\n"
1092               "  jiffies              <%lx>\n"
1093               "  next_to_watch.status <%x>\n"
1094               "MAC Status             <%x>\n"
1095               "PHY Status             <%x>\n"
1096               "PHY 1000BASE-T Status  <%x>\n"
1097               "PHY Extended Status    <%x>\n"
1098               "PCI Status             <%x>\n",
1099               readl(tx_ring->head),
1100               readl(tx_ring->tail),
1101               tx_ring->next_to_use,
1102               tx_ring->next_to_clean,
1103               tx_ring->buffer_info[eop].time_stamp,
1104               eop,
1105               jiffies,
1106               eop_desc->upper.fields.status,
1107               er32(STATUS),
1108               phy_status,
1109               phy_1000t_status,
1110               phy_ext_status,
1111               pci_status);
1112 }
1113
1114 /**
1115  * e1000_clean_tx_irq - Reclaim resources after transmit completes
1116  * @tx_ring: Tx descriptor ring
1117  *
1118  * the return value indicates whether actual cleaning was done, there
1119  * is no guarantee that everything was cleaned
1120  **/
1121 static bool e1000_clean_tx_irq(struct e1000_ring *tx_ring)
1122 {
1123         struct e1000_adapter *adapter = tx_ring->adapter;
1124         struct net_device *netdev = adapter->netdev;
1125         struct e1000_hw *hw = &adapter->hw;
1126         struct e1000_tx_desc *tx_desc, *eop_desc;
1127         struct e1000_buffer *buffer_info;
1128         unsigned int i, eop;
1129         unsigned int count = 0;
1130         unsigned int total_tx_bytes = 0, total_tx_packets = 0;
1131         unsigned int bytes_compl = 0, pkts_compl = 0;
1132
1133         i = tx_ring->next_to_clean;
1134         eop = tx_ring->buffer_info[i].next_to_watch;
1135         eop_desc = E1000_TX_DESC(*tx_ring, eop);
1136
1137         while ((eop_desc->upper.data & cpu_to_le32(E1000_TXD_STAT_DD)) &&
1138                (count < tx_ring->count)) {
1139                 bool cleaned = false;
1140                 rmb(); /* read buffer_info after eop_desc */
1141                 for (; !cleaned; count++) {
1142                         tx_desc = E1000_TX_DESC(*tx_ring, i);
1143                         buffer_info = &tx_ring->buffer_info[i];
1144                         cleaned = (i == eop);
1145
1146                         if (cleaned) {
1147                                 total_tx_packets += buffer_info->segs;
1148                                 total_tx_bytes += buffer_info->bytecount;
1149                                 if (buffer_info->skb) {
1150                                         bytes_compl += buffer_info->skb->len;
1151                                         pkts_compl++;
1152                                 }
1153                         }
1154
1155                         e1000_put_txbuf(tx_ring, buffer_info);
1156                         tx_desc->upper.data = 0;
1157
1158                         i++;
1159                         if (i == tx_ring->count)
1160                                 i = 0;
1161                 }
1162
1163                 if (i == tx_ring->next_to_use)
1164                         break;
1165                 eop = tx_ring->buffer_info[i].next_to_watch;
1166                 eop_desc = E1000_TX_DESC(*tx_ring, eop);
1167         }
1168
1169         tx_ring->next_to_clean = i;
1170
1171         netdev_completed_queue(netdev, pkts_compl, bytes_compl);
1172
1173 #define TX_WAKE_THRESHOLD 32
1174         if (count && netif_carrier_ok(netdev) &&
1175             e1000_desc_unused(tx_ring) >= TX_WAKE_THRESHOLD) {
1176                 /* Make sure that anybody stopping the queue after this
1177                  * sees the new next_to_clean.
1178                  */
1179                 smp_mb();
1180
1181                 if (netif_queue_stopped(netdev) &&
1182                     !(test_bit(__E1000_DOWN, &adapter->state))) {
1183                         netif_wake_queue(netdev);
1184                         ++adapter->restart_queue;
1185                 }
1186         }
1187
1188         if (adapter->detect_tx_hung) {
1189                 /*
1190                  * Detect a transmit hang in hardware, this serializes the
1191                  * check with the clearing of time_stamp and movement of i
1192                  */
1193                 adapter->detect_tx_hung = false;
1194                 if (tx_ring->buffer_info[i].time_stamp &&
1195                     time_after(jiffies, tx_ring->buffer_info[i].time_stamp
1196                                + (adapter->tx_timeout_factor * HZ)) &&
1197                     !(er32(STATUS) & E1000_STATUS_TXOFF))
1198                         schedule_work(&adapter->print_hang_task);
1199                 else
1200                         adapter->tx_hang_recheck = false;
1201         }
1202         adapter->total_tx_bytes += total_tx_bytes;
1203         adapter->total_tx_packets += total_tx_packets;
1204         return count < tx_ring->count;
1205 }
1206
1207 /**
1208  * e1000_clean_rx_irq_ps - Send received data up the network stack; packet split
1209  * @rx_ring: Rx descriptor ring
1210  *
1211  * the return value indicates whether actual cleaning was done, there
1212  * is no guarantee that everything was cleaned
1213  **/
1214 static bool e1000_clean_rx_irq_ps(struct e1000_ring *rx_ring, int *work_done,
1215                                   int work_to_do)
1216 {
1217         struct e1000_adapter *adapter = rx_ring->adapter;
1218         struct e1000_hw *hw = &adapter->hw;
1219         union e1000_rx_desc_packet_split *rx_desc, *next_rxd;
1220         struct net_device *netdev = adapter->netdev;
1221         struct pci_dev *pdev = adapter->pdev;
1222         struct e1000_buffer *buffer_info, *next_buffer;
1223         struct e1000_ps_page *ps_page;
1224         struct sk_buff *skb;
1225         unsigned int i, j;
1226         u32 length, staterr;
1227         int cleaned_count = 0;
1228         bool cleaned = false;
1229         unsigned int total_rx_bytes = 0, total_rx_packets = 0;
1230
1231         i = rx_ring->next_to_clean;
1232         rx_desc = E1000_RX_DESC_PS(*rx_ring, i);
1233         staterr = le32_to_cpu(rx_desc->wb.middle.status_error);
1234         buffer_info = &rx_ring->buffer_info[i];
1235
1236         while (staterr & E1000_RXD_STAT_DD) {
1237                 if (*work_done >= work_to_do)
1238                         break;
1239                 (*work_done)++;
1240                 skb = buffer_info->skb;
1241                 rmb();  /* read descriptor and rx_buffer_info after status DD */
1242
1243                 /* in the packet split case this is header only */
1244                 prefetch(skb->data - NET_IP_ALIGN);
1245
1246                 i++;
1247                 if (i == rx_ring->count)
1248                         i = 0;
1249                 next_rxd = E1000_RX_DESC_PS(*rx_ring, i);
1250                 prefetch(next_rxd);
1251
1252                 next_buffer = &rx_ring->buffer_info[i];
1253
1254                 cleaned = true;
1255                 cleaned_count++;
1256                 dma_unmap_single(&pdev->dev, buffer_info->dma,
1257                                  adapter->rx_ps_bsize0, DMA_FROM_DEVICE);
1258                 buffer_info->dma = 0;
1259
1260                 /* see !EOP comment in other Rx routine */
1261                 if (!(staterr & E1000_RXD_STAT_EOP))
1262                         adapter->flags2 |= FLAG2_IS_DISCARDING;
1263
1264                 if (adapter->flags2 & FLAG2_IS_DISCARDING) {
1265                         e_dbg("Packet Split buffers didn't pick up the full packet\n");
1266                         dev_kfree_skb_irq(skb);
1267                         if (staterr & E1000_RXD_STAT_EOP)
1268                                 adapter->flags2 &= ~FLAG2_IS_DISCARDING;
1269                         goto next_desc;
1270                 }
1271
1272                 if (unlikely((staterr & E1000_RXDEXT_ERR_FRAME_ERR_MASK) &&
1273                              !(netdev->features & NETIF_F_RXALL))) {
1274                         dev_kfree_skb_irq(skb);
1275                         goto next_desc;
1276                 }
1277
1278                 length = le16_to_cpu(rx_desc->wb.middle.length0);
1279
1280                 if (!length) {
1281                         e_dbg("Last part of the packet spanning multiple descriptors\n");
1282                         dev_kfree_skb_irq(skb);
1283                         goto next_desc;
1284                 }
1285
1286                 /* Good Receive */
1287                 skb_put(skb, length);
1288
1289                 {
1290                         /*
1291                          * this looks ugly, but it seems compiler issues make
1292                          * it more efficient than reusing j
1293                          */
1294                         int l1 = le16_to_cpu(rx_desc->wb.upper.length[0]);
1295
1296                         /*
1297                          * page alloc/put takes too long and effects small
1298                          * packet throughput, so unsplit small packets and
1299                          * save the alloc/put only valid in softirq (napi)
1300                          * context to call kmap_*
1301                          */
1302                         if (l1 && (l1 <= copybreak) &&
1303                             ((length + l1) <= adapter->rx_ps_bsize0)) {
1304                                 u8 *vaddr;
1305
1306                                 ps_page = &buffer_info->ps_pages[0];
1307
1308                                 /*
1309                                  * there is no documentation about how to call
1310                                  * kmap_atomic, so we can't hold the mapping
1311                                  * very long
1312                                  */
1313                                 dma_sync_single_for_cpu(&pdev->dev,
1314                                                         ps_page->dma,
1315                                                         PAGE_SIZE,
1316                                                         DMA_FROM_DEVICE);
1317                                 vaddr = kmap_atomic(ps_page->page);
1318                                 memcpy(skb_tail_pointer(skb), vaddr, l1);
1319                                 kunmap_atomic(vaddr);
1320                                 dma_sync_single_for_device(&pdev->dev,
1321                                                            ps_page->dma,
1322                                                            PAGE_SIZE,
1323                                                            DMA_FROM_DEVICE);
1324
1325                                 /* remove the CRC */
1326                                 if (!(adapter->flags2 & FLAG2_CRC_STRIPPING)) {
1327                                         if (!(netdev->features & NETIF_F_RXFCS))
1328                                                 l1 -= 4;
1329                                 }
1330
1331                                 skb_put(skb, l1);
1332                                 goto copydone;
1333                         } /* if */
1334                 }
1335
1336                 for (j = 0; j < PS_PAGE_BUFFERS; j++) {
1337                         length = le16_to_cpu(rx_desc->wb.upper.length[j]);
1338                         if (!length)
1339                                 break;
1340
1341                         ps_page = &buffer_info->ps_pages[j];
1342                         dma_unmap_page(&pdev->dev, ps_page->dma, PAGE_SIZE,
1343                                        DMA_FROM_DEVICE);
1344                         ps_page->dma = 0;
1345                         skb_fill_page_desc(skb, j, ps_page->page, 0, length);
1346                         ps_page->page = NULL;
1347                         skb->len += length;
1348                         skb->data_len += length;
1349                         skb->truesize += PAGE_SIZE;
1350                 }
1351
1352                 /* strip the ethernet crc, problem is we're using pages now so
1353                  * this whole operation can get a little cpu intensive
1354                  */
1355                 if (!(adapter->flags2 & FLAG2_CRC_STRIPPING)) {
1356                         if (!(netdev->features & NETIF_F_RXFCS))
1357                                 pskb_trim(skb, skb->len - 4);
1358                 }
1359
1360 copydone:
1361                 total_rx_bytes += skb->len;
1362                 total_rx_packets++;
1363
1364                 e1000_rx_checksum(adapter, staterr,
1365                                   rx_desc->wb.lower.hi_dword.csum_ip.csum, skb);
1366
1367                 e1000_rx_hash(netdev, rx_desc->wb.lower.hi_dword.rss, skb);
1368
1369                 if (rx_desc->wb.upper.header_status &
1370                            cpu_to_le16(E1000_RXDPS_HDRSTAT_HDRSP))
1371                         adapter->rx_hdr_split++;
1372
1373                 e1000_receive_skb(adapter, netdev, skb,
1374                                   staterr, rx_desc->wb.middle.vlan);
1375
1376 next_desc:
1377                 rx_desc->wb.middle.status_error &= cpu_to_le32(~0xFF);
1378                 buffer_info->skb = NULL;
1379
1380                 /* return some buffers to hardware, one at a time is too slow */
1381                 if (cleaned_count >= E1000_RX_BUFFER_WRITE) {
1382                         adapter->alloc_rx_buf(rx_ring, cleaned_count,
1383                                               GFP_ATOMIC);
1384                         cleaned_count = 0;
1385                 }
1386
1387                 /* use prefetched values */
1388                 rx_desc = next_rxd;
1389                 buffer_info = next_buffer;
1390
1391                 staterr = le32_to_cpu(rx_desc->wb.middle.status_error);
1392         }
1393         rx_ring->next_to_clean = i;
1394
1395         cleaned_count = e1000_desc_unused(rx_ring);
1396         if (cleaned_count)
1397                 adapter->alloc_rx_buf(rx_ring, cleaned_count, GFP_ATOMIC);
1398
1399         adapter->total_rx_bytes += total_rx_bytes;
1400         adapter->total_rx_packets += total_rx_packets;
1401         return cleaned;
1402 }
1403
1404 /**
1405  * e1000_consume_page - helper function
1406  **/
1407 static void e1000_consume_page(struct e1000_buffer *bi, struct sk_buff *skb,
1408                                u16 length)
1409 {
1410         bi->page = NULL;
1411         skb->len += length;
1412         skb->data_len += length;
1413         skb->truesize += PAGE_SIZE;
1414 }
1415
1416 /**
1417  * e1000_clean_jumbo_rx_irq - Send received data up the network stack; legacy
1418  * @adapter: board private structure
1419  *
1420  * the return value indicates whether actual cleaning was done, there
1421  * is no guarantee that everything was cleaned
1422  **/
1423 static bool e1000_clean_jumbo_rx_irq(struct e1000_ring *rx_ring, int *work_done,
1424                                      int work_to_do)
1425 {
1426         struct e1000_adapter *adapter = rx_ring->adapter;
1427         struct net_device *netdev = adapter->netdev;
1428         struct pci_dev *pdev = adapter->pdev;
1429         union e1000_rx_desc_extended *rx_desc, *next_rxd;
1430         struct e1000_buffer *buffer_info, *next_buffer;
1431         u32 length, staterr;
1432         unsigned int i;
1433         int cleaned_count = 0;
1434         bool cleaned = false;
1435         unsigned int total_rx_bytes=0, total_rx_packets=0;
1436
1437         i = rx_ring->next_to_clean;
1438         rx_desc = E1000_RX_DESC_EXT(*rx_ring, i);
1439         staterr = le32_to_cpu(rx_desc->wb.upper.status_error);
1440         buffer_info = &rx_ring->buffer_info[i];
1441
1442         while (staterr & E1000_RXD_STAT_DD) {
1443                 struct sk_buff *skb;
1444
1445                 if (*work_done >= work_to_do)
1446                         break;
1447                 (*work_done)++;
1448                 rmb();  /* read descriptor and rx_buffer_info after status DD */
1449
1450                 skb = buffer_info->skb;
1451                 buffer_info->skb = NULL;
1452
1453                 ++i;
1454                 if (i == rx_ring->count)
1455                         i = 0;
1456                 next_rxd = E1000_RX_DESC_EXT(*rx_ring, i);
1457                 prefetch(next_rxd);
1458
1459                 next_buffer = &rx_ring->buffer_info[i];
1460
1461                 cleaned = true;
1462                 cleaned_count++;
1463                 dma_unmap_page(&pdev->dev, buffer_info->dma, PAGE_SIZE,
1464                                DMA_FROM_DEVICE);
1465                 buffer_info->dma = 0;
1466
1467                 length = le16_to_cpu(rx_desc->wb.upper.length);
1468
1469                 /* errors is only valid for DD + EOP descriptors */
1470                 if (unlikely((staterr & E1000_RXD_STAT_EOP) &&
1471                              ((staterr & E1000_RXDEXT_ERR_FRAME_ERR_MASK) &&
1472                               !(netdev->features & NETIF_F_RXALL)))) {
1473                         /* recycle both page and skb */
1474                         buffer_info->skb = skb;
1475                         /* an error means any chain goes out the window too */
1476                         if (rx_ring->rx_skb_top)
1477                                 dev_kfree_skb_irq(rx_ring->rx_skb_top);
1478                         rx_ring->rx_skb_top = NULL;
1479                         goto next_desc;
1480                 }
1481
1482 #define rxtop (rx_ring->rx_skb_top)
1483                 if (!(staterr & E1000_RXD_STAT_EOP)) {
1484                         /* this descriptor is only the beginning (or middle) */
1485                         if (!rxtop) {
1486                                 /* this is the beginning of a chain */
1487                                 rxtop = skb;
1488                                 skb_fill_page_desc(rxtop, 0, buffer_info->page,
1489                                                    0, length);
1490                         } else {
1491                                 /* this is the middle of a chain */
1492                                 skb_fill_page_desc(rxtop,
1493                                     skb_shinfo(rxtop)->nr_frags,
1494                                     buffer_info->page, 0, length);
1495                                 /* re-use the skb, only consumed the page */
1496                                 buffer_info->skb = skb;
1497                         }
1498                         e1000_consume_page(buffer_info, rxtop, length);
1499                         goto next_desc;
1500                 } else {
1501                         if (rxtop) {
1502                                 /* end of the chain */
1503                                 skb_fill_page_desc(rxtop,
1504                                     skb_shinfo(rxtop)->nr_frags,
1505                                     buffer_info->page, 0, length);
1506                                 /* re-use the current skb, we only consumed the
1507                                  * page */
1508                                 buffer_info->skb = skb;
1509                                 skb = rxtop;
1510                                 rxtop = NULL;
1511                                 e1000_consume_page(buffer_info, skb, length);
1512                         } else {
1513                                 /* no chain, got EOP, this buf is the packet
1514                                  * copybreak to save the put_page/alloc_page */
1515                                 if (length <= copybreak &&
1516                                     skb_tailroom(skb) >= length) {
1517                                         u8 *vaddr;
1518                                         vaddr = kmap_atomic(buffer_info->page);
1519                                         memcpy(skb_tail_pointer(skb), vaddr,
1520                                                length);
1521                                         kunmap_atomic(vaddr);
1522                                         /* re-use the page, so don't erase
1523                                          * buffer_info->page */
1524                                         skb_put(skb, length);
1525                                 } else {
1526                                         skb_fill_page_desc(skb, 0,
1527                                                            buffer_info->page, 0,
1528                                                            length);
1529                                         e1000_consume_page(buffer_info, skb,
1530                                                            length);
1531                                 }
1532                         }
1533                 }
1534
1535                 /* Receive Checksum Offload XXX recompute due to CRC strip? */
1536                 e1000_rx_checksum(adapter, staterr,
1537                                   rx_desc->wb.lower.hi_dword.csum_ip.csum, skb);
1538
1539                 e1000_rx_hash(netdev, rx_desc->wb.lower.hi_dword.rss, skb);
1540
1541                 /* probably a little skewed due to removing CRC */
1542                 total_rx_bytes += skb->len;
1543                 total_rx_packets++;
1544
1545                 /* eth type trans needs skb->data to point to something */
1546                 if (!pskb_may_pull(skb, ETH_HLEN)) {
1547                         e_err("pskb_may_pull failed.\n");
1548                         dev_kfree_skb_irq(skb);
1549                         goto next_desc;
1550                 }
1551
1552                 e1000_receive_skb(adapter, netdev, skb, staterr,
1553                                   rx_desc->wb.upper.vlan);
1554
1555 next_desc:
1556                 rx_desc->wb.upper.status_error &= cpu_to_le32(~0xFF);
1557
1558                 /* return some buffers to hardware, one at a time is too slow */
1559                 if (unlikely(cleaned_count >= E1000_RX_BUFFER_WRITE)) {
1560                         adapter->alloc_rx_buf(rx_ring, cleaned_count,
1561                                               GFP_ATOMIC);
1562                         cleaned_count = 0;
1563                 }
1564
1565                 /* use prefetched values */
1566                 rx_desc = next_rxd;
1567                 buffer_info = next_buffer;
1568
1569                 staterr = le32_to_cpu(rx_desc->wb.upper.status_error);
1570         }
1571         rx_ring->next_to_clean = i;
1572
1573         cleaned_count = e1000_desc_unused(rx_ring);
1574         if (cleaned_count)
1575                 adapter->alloc_rx_buf(rx_ring, cleaned_count, GFP_ATOMIC);
1576
1577         adapter->total_rx_bytes += total_rx_bytes;
1578         adapter->total_rx_packets += total_rx_packets;
1579         return cleaned;
1580 }
1581
1582 /**
1583  * e1000_clean_rx_ring - Free Rx Buffers per Queue
1584  * @rx_ring: Rx descriptor ring
1585  **/
1586 static void e1000_clean_rx_ring(struct e1000_ring *rx_ring)
1587 {
1588         struct e1000_adapter *adapter = rx_ring->adapter;
1589         struct e1000_buffer *buffer_info;
1590         struct e1000_ps_page *ps_page;
1591         struct pci_dev *pdev = adapter->pdev;
1592         unsigned int i, j;
1593
1594         /* Free all the Rx ring sk_buffs */
1595         for (i = 0; i < rx_ring->count; i++) {
1596                 buffer_info = &rx_ring->buffer_info[i];
1597                 if (buffer_info->dma) {
1598                         if (adapter->clean_rx == e1000_clean_rx_irq)
1599                                 dma_unmap_single(&pdev->dev, buffer_info->dma,
1600                                                  adapter->rx_buffer_len,
1601                                                  DMA_FROM_DEVICE);
1602                         else if (adapter->clean_rx == e1000_clean_jumbo_rx_irq)
1603                                 dma_unmap_page(&pdev->dev, buffer_info->dma,
1604                                                PAGE_SIZE,
1605                                                DMA_FROM_DEVICE);
1606                         else if (adapter->clean_rx == e1000_clean_rx_irq_ps)
1607                                 dma_unmap_single(&pdev->dev, buffer_info->dma,
1608                                                  adapter->rx_ps_bsize0,
1609                                                  DMA_FROM_DEVICE);
1610                         buffer_info->dma = 0;
1611                 }
1612
1613                 if (buffer_info->page) {
1614                         put_page(buffer_info->page);
1615                         buffer_info->page = NULL;
1616                 }
1617
1618                 if (buffer_info->skb) {
1619                         dev_kfree_skb(buffer_info->skb);
1620                         buffer_info->skb = NULL;
1621                 }
1622
1623                 for (j = 0; j < PS_PAGE_BUFFERS; j++) {
1624                         ps_page = &buffer_info->ps_pages[j];
1625                         if (!ps_page->page)
1626                                 break;
1627                         dma_unmap_page(&pdev->dev, ps_page->dma, PAGE_SIZE,
1628                                        DMA_FROM_DEVICE);
1629                         ps_page->dma = 0;
1630                         put_page(ps_page->page);
1631                         ps_page->page = NULL;
1632                 }
1633         }
1634
1635         /* there also may be some cached data from a chained receive */
1636         if (rx_ring->rx_skb_top) {
1637                 dev_kfree_skb(rx_ring->rx_skb_top);
1638                 rx_ring->rx_skb_top = NULL;
1639         }
1640
1641         /* Zero out the descriptor ring */
1642         memset(rx_ring->desc, 0, rx_ring->size);
1643
1644         rx_ring->next_to_clean = 0;
1645         rx_ring->next_to_use = 0;
1646         adapter->flags2 &= ~FLAG2_IS_DISCARDING;
1647
1648         writel(0, rx_ring->head);
1649         writel(0, rx_ring->tail);
1650 }
1651
1652 static void e1000e_downshift_workaround(struct work_struct *work)
1653 {
1654         struct e1000_adapter *adapter = container_of(work,
1655                                         struct e1000_adapter, downshift_task);
1656
1657         if (test_bit(__E1000_DOWN, &adapter->state))
1658                 return;
1659
1660         e1000e_gig_downshift_workaround_ich8lan(&adapter->hw);
1661 }
1662
1663 /**
1664  * e1000_intr_msi - Interrupt Handler
1665  * @irq: interrupt number
1666  * @data: pointer to a network interface device structure
1667  **/
1668 static irqreturn_t e1000_intr_msi(int irq, void *data)
1669 {
1670         struct net_device *netdev = data;
1671         struct e1000_adapter *adapter = netdev_priv(netdev);
1672         struct e1000_hw *hw = &adapter->hw;
1673         u32 icr = er32(ICR);
1674
1675         /*
1676          * read ICR disables interrupts using IAM
1677          */
1678
1679         if (icr & E1000_ICR_LSC) {
1680                 hw->mac.get_link_status = true;
1681                 /*
1682                  * ICH8 workaround-- Call gig speed drop workaround on cable
1683                  * disconnect (LSC) before accessing any PHY registers
1684                  */
1685                 if ((adapter->flags & FLAG_LSC_GIG_SPEED_DROP) &&
1686                     (!(er32(STATUS) & E1000_STATUS_LU)))
1687                         schedule_work(&adapter->downshift_task);
1688
1689                 /*
1690                  * 80003ES2LAN workaround-- For packet buffer work-around on
1691                  * link down event; disable receives here in the ISR and reset
1692                  * adapter in watchdog
1693                  */
1694                 if (netif_carrier_ok(netdev) &&
1695                     adapter->flags & FLAG_RX_NEEDS_RESTART) {
1696                         /* disable receives */
1697                         u32 rctl = er32(RCTL);
1698                         ew32(RCTL, rctl & ~E1000_RCTL_EN);
1699                         adapter->flags |= FLAG_RX_RESTART_NOW;
1700                 }
1701                 /* guard against interrupt when we're going down */
1702                 if (!test_bit(__E1000_DOWN, &adapter->state))
1703                         mod_timer(&adapter->watchdog_timer, jiffies + 1);
1704         }
1705
1706         if (napi_schedule_prep(&adapter->napi)) {
1707                 adapter->total_tx_bytes = 0;
1708                 adapter->total_tx_packets = 0;
1709                 adapter->total_rx_bytes = 0;
1710                 adapter->total_rx_packets = 0;
1711                 __napi_schedule(&adapter->napi);
1712         }
1713
1714         return IRQ_HANDLED;
1715 }
1716
1717 /**
1718  * e1000_intr - Interrupt Handler
1719  * @irq: interrupt number
1720  * @data: pointer to a network interface device structure
1721  **/
1722 static irqreturn_t e1000_intr(int irq, void *data)
1723 {
1724         struct net_device *netdev = data;
1725         struct e1000_adapter *adapter = netdev_priv(netdev);
1726         struct e1000_hw *hw = &adapter->hw;
1727         u32 rctl, icr = er32(ICR);
1728
1729         if (!icr || test_bit(__E1000_DOWN, &adapter->state))
1730                 return IRQ_NONE;  /* Not our interrupt */
1731
1732         /*
1733          * IMS will not auto-mask if INT_ASSERTED is not set, and if it is
1734          * not set, then the adapter didn't send an interrupt
1735          */
1736         if (!(icr & E1000_ICR_INT_ASSERTED))
1737                 return IRQ_NONE;
1738
1739         /*
1740          * Interrupt Auto-Mask...upon reading ICR,
1741          * interrupts are masked.  No need for the
1742          * IMC write
1743          */
1744
1745         if (icr & E1000_ICR_LSC) {
1746                 hw->mac.get_link_status = true;
1747                 /*
1748                  * ICH8 workaround-- Call gig speed drop workaround on cable
1749                  * disconnect (LSC) before accessing any PHY registers
1750                  */
1751                 if ((adapter->flags & FLAG_LSC_GIG_SPEED_DROP) &&
1752                     (!(er32(STATUS) & E1000_STATUS_LU)))
1753                         schedule_work(&adapter->downshift_task);
1754
1755                 /*
1756                  * 80003ES2LAN workaround--
1757                  * For packet buffer work-around on link down event;
1758                  * disable receives here in the ISR and
1759                  * reset adapter in watchdog
1760                  */
1761                 if (netif_carrier_ok(netdev) &&
1762                     (adapter->flags & FLAG_RX_NEEDS_RESTART)) {
1763                         /* disable receives */
1764                         rctl = er32(RCTL);
1765                         ew32(RCTL, rctl & ~E1000_RCTL_EN);
1766                         adapter->flags |= FLAG_RX_RESTART_NOW;
1767                 }
1768                 /* guard against interrupt when we're going down */
1769                 if (!test_bit(__E1000_DOWN, &adapter->state))
1770                         mod_timer(&adapter->watchdog_timer, jiffies + 1);
1771         }
1772
1773         if (napi_schedule_prep(&adapter->napi)) {
1774                 adapter->total_tx_bytes = 0;
1775                 adapter->total_tx_packets = 0;
1776                 adapter->total_rx_bytes = 0;
1777                 adapter->total_rx_packets = 0;
1778                 __napi_schedule(&adapter->napi);
1779         }
1780
1781         return IRQ_HANDLED;
1782 }
1783
1784 static irqreturn_t e1000_msix_other(int irq, void *data)
1785 {
1786         struct net_device *netdev = data;
1787         struct e1000_adapter *adapter = netdev_priv(netdev);
1788         struct e1000_hw *hw = &adapter->hw;
1789         u32 icr = er32(ICR);
1790
1791         if (!(icr & E1000_ICR_INT_ASSERTED)) {
1792                 if (!test_bit(__E1000_DOWN, &adapter->state))
1793                         ew32(IMS, E1000_IMS_OTHER);
1794                 return IRQ_NONE;
1795         }
1796
1797         if (icr & adapter->eiac_mask)
1798                 ew32(ICS, (icr & adapter->eiac_mask));
1799
1800         if (icr & E1000_ICR_OTHER) {
1801                 if (!(icr & E1000_ICR_LSC))
1802                         goto no_link_interrupt;
1803                 hw->mac.get_link_status = true;
1804                 /* guard against interrupt when we're going down */
1805                 if (!test_bit(__E1000_DOWN, &adapter->state))
1806                         mod_timer(&adapter->watchdog_timer, jiffies + 1);
1807         }
1808
1809 no_link_interrupt:
1810         if (!test_bit(__E1000_DOWN, &adapter->state))
1811                 ew32(IMS, E1000_IMS_LSC | E1000_IMS_OTHER);
1812
1813         return IRQ_HANDLED;
1814 }
1815
1816
1817 static irqreturn_t e1000_intr_msix_tx(int irq, void *data)
1818 {
1819         struct net_device *netdev = data;
1820         struct e1000_adapter *adapter = netdev_priv(netdev);
1821         struct e1000_hw *hw = &adapter->hw;
1822         struct e1000_ring *tx_ring = adapter->tx_ring;
1823
1824
1825         adapter->total_tx_bytes = 0;
1826         adapter->total_tx_packets = 0;
1827
1828         if (!e1000_clean_tx_irq(tx_ring))
1829                 /* Ring was not completely cleaned, so fire another interrupt */
1830                 ew32(ICS, tx_ring->ims_val);
1831
1832         return IRQ_HANDLED;
1833 }
1834
1835 static irqreturn_t e1000_intr_msix_rx(int irq, void *data)
1836 {
1837         struct net_device *netdev = data;
1838         struct e1000_adapter *adapter = netdev_priv(netdev);
1839         struct e1000_ring *rx_ring = adapter->rx_ring;
1840
1841         /* Write the ITR value calculated at the end of the
1842          * previous interrupt.
1843          */
1844         if (rx_ring->set_itr) {
1845                 writel(1000000000 / (rx_ring->itr_val * 256),
1846                        rx_ring->itr_register);
1847                 rx_ring->set_itr = 0;
1848         }
1849
1850         if (napi_schedule_prep(&adapter->napi)) {
1851                 adapter->total_rx_bytes = 0;
1852                 adapter->total_rx_packets = 0;
1853                 __napi_schedule(&adapter->napi);
1854         }
1855         return IRQ_HANDLED;
1856 }
1857
1858 /**
1859  * e1000_configure_msix - Configure MSI-X hardware
1860  *
1861  * e1000_configure_msix sets up the hardware to properly
1862  * generate MSI-X interrupts.
1863  **/
1864 static void e1000_configure_msix(struct e1000_adapter *adapter)
1865 {
1866         struct e1000_hw *hw = &adapter->hw;
1867         struct e1000_ring *rx_ring = adapter->rx_ring;
1868         struct e1000_ring *tx_ring = adapter->tx_ring;
1869         int vector = 0;
1870         u32 ctrl_ext, ivar = 0;
1871
1872         adapter->eiac_mask = 0;
1873
1874         /* Workaround issue with spurious interrupts on 82574 in MSI-X mode */
1875         if (hw->mac.type == e1000_82574) {
1876                 u32 rfctl = er32(RFCTL);
1877                 rfctl |= E1000_RFCTL_ACK_DIS;
1878                 ew32(RFCTL, rfctl);
1879         }
1880
1881 #define E1000_IVAR_INT_ALLOC_VALID      0x8
1882         /* Configure Rx vector */
1883         rx_ring->ims_val = E1000_IMS_RXQ0;
1884         adapter->eiac_mask |= rx_ring->ims_val;
1885         if (rx_ring->itr_val)
1886                 writel(1000000000 / (rx_ring->itr_val * 256),
1887                        rx_ring->itr_register);
1888         else
1889                 writel(1, rx_ring->itr_register);
1890         ivar = E1000_IVAR_INT_ALLOC_VALID | vector;
1891
1892         /* Configure Tx vector */
1893         tx_ring->ims_val = E1000_IMS_TXQ0;
1894         vector++;
1895         if (tx_ring->itr_val)
1896                 writel(1000000000 / (tx_ring->itr_val * 256),
1897                        tx_ring->itr_register);
1898         else
1899                 writel(1, tx_ring->itr_register);
1900         adapter->eiac_mask |= tx_ring->ims_val;
1901         ivar |= ((E1000_IVAR_INT_ALLOC_VALID | vector) << 8);
1902
1903         /* set vector for Other Causes, e.g. link changes */
1904         vector++;
1905         ivar |= ((E1000_IVAR_INT_ALLOC_VALID | vector) << 16);
1906         if (rx_ring->itr_val)
1907                 writel(1000000000 / (rx_ring->itr_val * 256),
1908                        hw->hw_addr + E1000_EITR_82574(vector));
1909         else
1910                 writel(1, hw->hw_addr + E1000_EITR_82574(vector));
1911
1912         /* Cause Tx interrupts on every write back */
1913         ivar |= (1 << 31);
1914
1915         ew32(IVAR, ivar);
1916
1917         /* enable MSI-X PBA support */
1918         ctrl_ext = er32(CTRL_EXT);
1919         ctrl_ext |= E1000_CTRL_EXT_PBA_CLR;
1920
1921         /* Auto-Mask Other interrupts upon ICR read */
1922 #define E1000_EIAC_MASK_82574   0x01F00000
1923         ew32(IAM, ~E1000_EIAC_MASK_82574 | E1000_IMS_OTHER);
1924         ctrl_ext |= E1000_CTRL_EXT_EIAME;
1925         ew32(CTRL_EXT, ctrl_ext);
1926         e1e_flush();
1927 }
1928
1929 void e1000e_reset_interrupt_capability(struct e1000_adapter *adapter)
1930 {
1931         if (adapter->msix_entries) {
1932                 pci_disable_msix(adapter->pdev);
1933                 kfree(adapter->msix_entries);
1934                 adapter->msix_entries = NULL;
1935         } else if (adapter->flags & FLAG_MSI_ENABLED) {
1936                 pci_disable_msi(adapter->pdev);
1937                 adapter->flags &= ~FLAG_MSI_ENABLED;
1938         }
1939 }
1940
1941 /**
1942  * e1000e_set_interrupt_capability - set MSI or MSI-X if supported
1943  *
1944  * Attempt to configure interrupts using the best available
1945  * capabilities of the hardware and kernel.
1946  **/
1947 void e1000e_set_interrupt_capability(struct e1000_adapter *adapter)
1948 {
1949         int err;
1950         int i;
1951
1952         switch (adapter->int_mode) {
1953         case E1000E_INT_MODE_MSIX:
1954                 if (adapter->flags & FLAG_HAS_MSIX) {
1955                         adapter->num_vectors = 3; /* RxQ0, TxQ0 and other */
1956                         adapter->msix_entries = kcalloc(adapter->num_vectors,
1957                                                       sizeof(struct msix_entry),
1958                                                       GFP_KERNEL);
1959                         if (adapter->msix_entries) {
1960                                 for (i = 0; i < adapter->num_vectors; i++)
1961                                         adapter->msix_entries[i].entry = i;
1962
1963                                 err = pci_enable_msix(adapter->pdev,
1964                                                       adapter->msix_entries,
1965                                                       adapter->num_vectors);
1966                                 if (err == 0)
1967                                         return;
1968                         }
1969                         /* MSI-X failed, so fall through and try MSI */
1970                         e_err("Failed to initialize MSI-X interrupts.  Falling back to MSI interrupts.\n");
1971                         e1000e_reset_interrupt_capability(adapter);
1972                 }
1973                 adapter->int_mode = E1000E_INT_MODE_MSI;
1974                 /* Fall through */
1975         case E1000E_INT_MODE_MSI:
1976                 if (!pci_enable_msi(adapter->pdev)) {
1977                         adapter->flags |= FLAG_MSI_ENABLED;
1978                 } else {
1979                         adapter->int_mode = E1000E_INT_MODE_LEGACY;
1980                         e_err("Failed to initialize MSI interrupts.  Falling back to legacy interrupts.\n");
1981                 }
1982                 /* Fall through */
1983         case E1000E_INT_MODE_LEGACY:
1984                 /* Don't do anything; this is the system default */
1985                 break;
1986         }
1987
1988         /* store the number of vectors being used */
1989         adapter->num_vectors = 1;
1990 }
1991
1992 /**
1993  * e1000_request_msix - Initialize MSI-X interrupts
1994  *
1995  * e1000_request_msix allocates MSI-X vectors and requests interrupts from the
1996  * kernel.
1997  **/
1998 static int e1000_request_msix(struct e1000_adapter *adapter)
1999 {
2000         struct net_device *netdev = adapter->netdev;
2001         int err = 0, vector = 0;
2002
2003         if (strlen(netdev->name) < (IFNAMSIZ - 5))
2004                 snprintf(adapter->rx_ring->name,
2005                          sizeof(adapter->rx_ring->name) - 1,
2006                          "%s-rx-0", netdev->name);
2007         else
2008                 memcpy(adapter->rx_ring->name, netdev->name, IFNAMSIZ);
2009         err = request_irq(adapter->msix_entries[vector].vector,
2010                           e1000_intr_msix_rx, 0, adapter->rx_ring->name,
2011                           netdev);
2012         if (err)
2013                 return err;
2014         adapter->rx_ring->itr_register = adapter->hw.hw_addr +
2015             E1000_EITR_82574(vector);
2016         adapter->rx_ring->itr_val = adapter->itr;
2017         vector++;
2018
2019         if (strlen(netdev->name) < (IFNAMSIZ - 5))
2020                 snprintf(adapter->tx_ring->name,
2021                          sizeof(adapter->tx_ring->name) - 1,
2022                          "%s-tx-0", netdev->name);
2023         else
2024                 memcpy(adapter->tx_ring->name, netdev->name, IFNAMSIZ);
2025         err = request_irq(adapter->msix_entries[vector].vector,
2026                           e1000_intr_msix_tx, 0, adapter->tx_ring->name,
2027                           netdev);
2028         if (err)
2029                 return err;
2030         adapter->tx_ring->itr_register = adapter->hw.hw_addr +
2031             E1000_EITR_82574(vector);
2032         adapter->tx_ring->itr_val = adapter->itr;
2033         vector++;
2034
2035         err = request_irq(adapter->msix_entries[vector].vector,
2036                           e1000_msix_other, 0, netdev->name, netdev);
2037         if (err)
2038                 return err;
2039
2040         e1000_configure_msix(adapter);
2041
2042         return 0;
2043 }
2044
2045 /**
2046  * e1000_request_irq - initialize interrupts
2047  *
2048  * Attempts to configure interrupts using the best available
2049  * capabilities of the hardware and kernel.
2050  **/
2051 static int e1000_request_irq(struct e1000_adapter *adapter)
2052 {
2053         struct net_device *netdev = adapter->netdev;
2054         int err;
2055
2056         if (adapter->msix_entries) {
2057                 err = e1000_request_msix(adapter);
2058                 if (!err)
2059                         return err;
2060                 /* fall back to MSI */
2061                 e1000e_reset_interrupt_capability(adapter);
2062                 adapter->int_mode = E1000E_INT_MODE_MSI;
2063                 e1000e_set_interrupt_capability(adapter);
2064         }
2065         if (adapter->flags & FLAG_MSI_ENABLED) {
2066                 err = request_irq(adapter->pdev->irq, e1000_intr_msi, 0,
2067                                   netdev->name, netdev);
2068                 if (!err)
2069                         return err;
2070
2071                 /* fall back to legacy interrupt */
2072                 e1000e_reset_interrupt_capability(adapter);
2073                 adapter->int_mode = E1000E_INT_MODE_LEGACY;
2074         }
2075
2076         err = request_irq(adapter->pdev->irq, e1000_intr, IRQF_SHARED,
2077                           netdev->name, netdev);
2078         if (err)
2079                 e_err("Unable to allocate interrupt, Error: %d\n", err);
2080
2081         return err;
2082 }
2083
2084 static void e1000_free_irq(struct e1000_adapter *adapter)
2085 {
2086         struct net_device *netdev = adapter->netdev;
2087
2088         if (adapter->msix_entries) {
2089                 int vector = 0;
2090
2091                 free_irq(adapter->msix_entries[vector].vector, netdev);
2092                 vector++;
2093
2094                 free_irq(adapter->msix_entries[vector].vector, netdev);
2095                 vector++;
2096
2097                 /* Other Causes interrupt vector */
2098                 free_irq(adapter->msix_entries[vector].vector, netdev);
2099                 return;
2100         }
2101
2102         free_irq(adapter->pdev->irq, netdev);
2103 }
2104
2105 /**
2106  * e1000_irq_disable - Mask off interrupt generation on the NIC
2107  **/
2108 static void e1000_irq_disable(struct e1000_adapter *adapter)
2109 {
2110         struct e1000_hw *hw = &adapter->hw;
2111
2112         ew32(IMC, ~0);
2113         if (adapter->msix_entries)
2114                 ew32(EIAC_82574, 0);
2115         e1e_flush();
2116
2117         if (adapter->msix_entries) {
2118                 int i;
2119                 for (i = 0; i < adapter->num_vectors; i++)
2120                         synchronize_irq(adapter->msix_entries[i].vector);
2121         } else {
2122                 synchronize_irq(adapter->pdev->irq);
2123         }
2124 }
2125
2126 /**
2127  * e1000_irq_enable - Enable default interrupt generation settings
2128  **/
2129 static void e1000_irq_enable(struct e1000_adapter *adapter)
2130 {
2131         struct e1000_hw *hw = &adapter->hw;
2132
2133         if (adapter->msix_entries) {
2134                 ew32(EIAC_82574, adapter->eiac_mask & E1000_EIAC_MASK_82574);
2135                 ew32(IMS, adapter->eiac_mask | E1000_IMS_OTHER | E1000_IMS_LSC);
2136         } else {
2137                 ew32(IMS, IMS_ENABLE_MASK);
2138         }
2139         e1e_flush();
2140 }
2141
2142 /**
2143  * e1000e_get_hw_control - get control of the h/w from f/w
2144  * @adapter: address of board private structure
2145  *
2146  * e1000e_get_hw_control sets {CTRL_EXT|SWSM}:DRV_LOAD bit.
2147  * For ASF and Pass Through versions of f/w this means that
2148  * the driver is loaded. For AMT version (only with 82573)
2149  * of the f/w this means that the network i/f is open.
2150  **/
2151 void e1000e_get_hw_control(struct e1000_adapter *adapter)
2152 {
2153         struct e1000_hw *hw = &adapter->hw;
2154         u32 ctrl_ext;
2155         u32 swsm;
2156
2157         /* Let firmware know the driver has taken over */
2158         if (adapter->flags & FLAG_HAS_SWSM_ON_LOAD) {
2159                 swsm = er32(SWSM);
2160                 ew32(SWSM, swsm | E1000_SWSM_DRV_LOAD);
2161         } else if (adapter->flags & FLAG_HAS_CTRLEXT_ON_LOAD) {
2162                 ctrl_ext = er32(CTRL_EXT);
2163                 ew32(CTRL_EXT, ctrl_ext | E1000_CTRL_EXT_DRV_LOAD);
2164         }
2165 }
2166
2167 /**
2168  * e1000e_release_hw_control - release control of the h/w to f/w
2169  * @adapter: address of board private structure
2170  *
2171  * e1000e_release_hw_control resets {CTRL_EXT|SWSM}:DRV_LOAD bit.
2172  * For ASF and Pass Through versions of f/w this means that the
2173  * driver is no longer loaded. For AMT version (only with 82573) i
2174  * of the f/w this means that the network i/f is closed.
2175  *
2176  **/
2177 void e1000e_release_hw_control(struct e1000_adapter *adapter)
2178 {
2179         struct e1000_hw *hw = &adapter->hw;
2180         u32 ctrl_ext;
2181         u32 swsm;
2182
2183         /* Let firmware taken over control of h/w */
2184         if (adapter->flags & FLAG_HAS_SWSM_ON_LOAD) {
2185                 swsm = er32(SWSM);
2186                 ew32(SWSM, swsm & ~E1000_SWSM_DRV_LOAD);
2187         } else if (adapter->flags & FLAG_HAS_CTRLEXT_ON_LOAD) {
2188                 ctrl_ext = er32(CTRL_EXT);
2189                 ew32(CTRL_EXT, ctrl_ext & ~E1000_CTRL_EXT_DRV_LOAD);
2190         }
2191 }
2192
2193 /**
2194  * @e1000_alloc_ring - allocate memory for a ring structure
2195  **/
2196 static int e1000_alloc_ring_dma(struct e1000_adapter *adapter,
2197                                 struct e1000_ring *ring)
2198 {
2199         struct pci_dev *pdev = adapter->pdev;
2200
2201         ring->desc = dma_alloc_coherent(&pdev->dev, ring->size, &ring->dma,
2202                                         GFP_KERNEL);
2203         if (!ring->desc)
2204                 return -ENOMEM;
2205
2206         return 0;
2207 }
2208
2209 /**
2210  * e1000e_setup_tx_resources - allocate Tx resources (Descriptors)
2211  * @tx_ring: Tx descriptor ring
2212  *
2213  * Return 0 on success, negative on failure
2214  **/
2215 int e1000e_setup_tx_resources(struct e1000_ring *tx_ring)
2216 {
2217         struct e1000_adapter *adapter = tx_ring->adapter;
2218         int err = -ENOMEM, size;
2219
2220         size = sizeof(struct e1000_buffer) * tx_ring->count;
2221         tx_ring->buffer_info = vzalloc(size);
2222         if (!tx_ring->buffer_info)
2223                 goto err;
2224
2225         /* round up to nearest 4K */
2226         tx_ring->size = tx_ring->count * sizeof(struct e1000_tx_desc);
2227         tx_ring->size = ALIGN(tx_ring->size, 4096);
2228
2229         err = e1000_alloc_ring_dma(adapter, tx_ring);
2230         if (err)
2231                 goto err;
2232
2233         tx_ring->next_to_use = 0;
2234         tx_ring->next_to_clean = 0;
2235
2236         return 0;
2237 err:
2238         vfree(tx_ring->buffer_info);
2239         e_err("Unable to allocate memory for the transmit descriptor ring\n");
2240         return err;
2241 }
2242
2243 /**
2244  * e1000e_setup_rx_resources - allocate Rx resources (Descriptors)
2245  * @rx_ring: Rx descriptor ring
2246  *
2247  * Returns 0 on success, negative on failure
2248  **/
2249 int e1000e_setup_rx_resources(struct e1000_ring *rx_ring)
2250 {
2251         struct e1000_adapter *adapter = rx_ring->adapter;
2252         struct e1000_buffer *buffer_info;
2253         int i, size, desc_len, err = -ENOMEM;
2254
2255         size = sizeof(struct e1000_buffer) * rx_ring->count;
2256         rx_ring->buffer_info = vzalloc(size);
2257         if (!rx_ring->buffer_info)
2258                 goto err;
2259
2260         for (i = 0; i < rx_ring->count; i++) {
2261                 buffer_info = &rx_ring->buffer_info[i];
2262                 buffer_info->ps_pages = kcalloc(PS_PAGE_BUFFERS,
2263                                                 sizeof(struct e1000_ps_page),
2264                                                 GFP_KERNEL);
2265                 if (!buffer_info->ps_pages)
2266                         goto err_pages;
2267         }
2268
2269         desc_len = sizeof(union e1000_rx_desc_packet_split);
2270
2271         /* Round up to nearest 4K */
2272         rx_ring->size = rx_ring->count * desc_len;
2273         rx_ring->size = ALIGN(rx_ring->size, 4096);
2274
2275         err = e1000_alloc_ring_dma(adapter, rx_ring);
2276         if (err)
2277                 goto err_pages;
2278
2279         rx_ring->next_to_clean = 0;
2280         rx_ring->next_to_use = 0;
2281         rx_ring->rx_skb_top = NULL;
2282
2283         return 0;
2284
2285 err_pages:
2286         for (i = 0; i < rx_ring->count; i++) {
2287                 buffer_info = &rx_ring->buffer_info[i];
2288                 kfree(buffer_info->ps_pages);
2289         }
2290 err:
2291         vfree(rx_ring->buffer_info);
2292         e_err("Unable to allocate memory for the receive descriptor ring\n");
2293         return err;
2294 }
2295
2296 /**
2297  * e1000_clean_tx_ring - Free Tx Buffers
2298  * @tx_ring: Tx descriptor ring
2299  **/
2300 static void e1000_clean_tx_ring(struct e1000_ring *tx_ring)
2301 {
2302         struct e1000_adapter *adapter = tx_ring->adapter;
2303         struct e1000_buffer *buffer_info;
2304         unsigned long size;
2305         unsigned int i;
2306
2307         for (i = 0; i < tx_ring->count; i++) {
2308                 buffer_info = &tx_ring->buffer_info[i];
2309                 e1000_put_txbuf(tx_ring, buffer_info);
2310         }
2311
2312         netdev_reset_queue(adapter->netdev);
2313         size = sizeof(struct e1000_buffer) * tx_ring->count;
2314         memset(tx_ring->buffer_info, 0, size);
2315
2316         memset(tx_ring->desc, 0, tx_ring->size);
2317
2318         tx_ring->next_to_use = 0;
2319         tx_ring->next_to_clean = 0;
2320
2321         writel(0, tx_ring->head);
2322         writel(0, tx_ring->tail);
2323 }
2324
2325 /**
2326  * e1000e_free_tx_resources - Free Tx Resources per Queue
2327  * @tx_ring: Tx descriptor ring
2328  *
2329  * Free all transmit software resources
2330  **/
2331 void e1000e_free_tx_resources(struct e1000_ring *tx_ring)
2332 {
2333         struct e1000_adapter *adapter = tx_ring->adapter;
2334         struct pci_dev *pdev = adapter->pdev;
2335
2336         e1000_clean_tx_ring(tx_ring);
2337
2338         vfree(tx_ring->buffer_info);
2339         tx_ring->buffer_info = NULL;
2340
2341         dma_free_coherent(&pdev->dev, tx_ring->size, tx_ring->desc,
2342                           tx_ring->dma);
2343         tx_ring->desc = NULL;
2344 }
2345
2346 /**
2347  * e1000e_free_rx_resources - Free Rx Resources
2348  * @rx_ring: Rx descriptor ring
2349  *
2350  * Free all receive software resources
2351  **/
2352 void e1000e_free_rx_resources(struct e1000_ring *rx_ring)
2353 {
2354         struct e1000_adapter *adapter = rx_ring->adapter;
2355         struct pci_dev *pdev = adapter->pdev;
2356         int i;
2357
2358         e1000_clean_rx_ring(rx_ring);
2359
2360         for (i = 0; i < rx_ring->count; i++)
2361                 kfree(rx_ring->buffer_info[i].ps_pages);
2362
2363         vfree(rx_ring->buffer_info);
2364         rx_ring->buffer_info = NULL;
2365
2366         dma_free_coherent(&pdev->dev, rx_ring->size, rx_ring->desc,
2367                           rx_ring->dma);
2368         rx_ring->desc = NULL;
2369 }
2370
2371 /**
2372  * e1000_update_itr - update the dynamic ITR value based on statistics
2373  * @adapter: pointer to adapter
2374  * @itr_setting: current adapter->itr
2375  * @packets: the number of packets during this measurement interval
2376  * @bytes: the number of bytes during this measurement interval
2377  *
2378  *      Stores a new ITR value based on packets and byte
2379  *      counts during the last interrupt.  The advantage of per interrupt
2380  *      computation is faster updates and more accurate ITR for the current
2381  *      traffic pattern.  Constants in this function were computed
2382  *      based on theoretical maximum wire speed and thresholds were set based
2383  *      on testing data as well as attempting to minimize response time
2384  *      while increasing bulk throughput.  This functionality is controlled
2385  *      by the InterruptThrottleRate module parameter.
2386  **/
2387 static unsigned int e1000_update_itr(struct e1000_adapter *adapter,
2388                                      u16 itr_setting, int packets,
2389                                      int bytes)
2390 {
2391         unsigned int retval = itr_setting;
2392
2393         if (packets == 0)
2394                 return itr_setting;
2395
2396         switch (itr_setting) {
2397         case lowest_latency:
2398                 /* handle TSO and jumbo frames */
2399                 if (bytes/packets > 8000)
2400                         retval = bulk_latency;
2401                 else if ((packets < 5) && (bytes > 512))
2402                         retval = low_latency;
2403                 break;
2404         case low_latency:  /* 50 usec aka 20000 ints/s */
2405                 if (bytes > 10000) {
2406                         /* this if handles the TSO accounting */
2407                         if (bytes/packets > 8000)
2408                                 retval = bulk_latency;
2409                         else if ((packets < 10) || ((bytes/packets) > 1200))
2410                                 retval = bulk_latency;
2411                         else if ((packets > 35))
2412                                 retval = lowest_latency;
2413                 } else if (bytes/packets > 2000) {
2414                         retval = bulk_latency;
2415                 } else if (packets <= 2 && bytes < 512) {
2416                         retval = lowest_latency;
2417                 }
2418                 break;
2419         case bulk_latency: /* 250 usec aka 4000 ints/s */
2420                 if (bytes > 25000) {
2421                         if (packets > 35)
2422                                 retval = low_latency;
2423                 } else if (bytes < 6000) {
2424                         retval = low_latency;
2425                 }
2426                 break;
2427         }
2428
2429         return retval;
2430 }
2431
2432 static void e1000_set_itr(struct e1000_adapter *adapter)
2433 {
2434         struct e1000_hw *hw = &adapter->hw;
2435         u16 current_itr;
2436         u32 new_itr = adapter->itr;
2437
2438         /* for non-gigabit speeds, just fix the interrupt rate at 4000 */
2439         if (adapter->link_speed != SPEED_1000) {
2440                 current_itr = 0;
2441                 new_itr = 4000;
2442                 goto set_itr_now;
2443         }
2444
2445         if (adapter->flags2 & FLAG2_DISABLE_AIM) {
2446                 new_itr = 0;
2447                 goto set_itr_now;
2448         }
2449
2450         adapter->tx_itr = e1000_update_itr(adapter,
2451                                     adapter->tx_itr,
2452                                     adapter->total_tx_packets,
2453                                     adapter->total_tx_bytes);
2454         /* conservative mode (itr 3) eliminates the lowest_latency setting */
2455         if (adapter->itr_setting == 3 && adapter->tx_itr == lowest_latency)
2456                 adapter->tx_itr = low_latency;
2457
2458         adapter->rx_itr = e1000_update_itr(adapter,
2459                                     adapter->rx_itr,
2460                                     adapter->total_rx_packets,
2461                                     adapter->total_rx_bytes);
2462         /* conservative mode (itr 3) eliminates the lowest_latency setting */
2463         if (adapter->itr_setting == 3 && adapter->rx_itr == lowest_latency)
2464                 adapter->rx_itr = low_latency;
2465
2466         current_itr = max(adapter->rx_itr, adapter->tx_itr);
2467
2468         switch (current_itr) {
2469         /* counts and packets in update_itr are dependent on these numbers */
2470         case lowest_latency:
2471                 new_itr = 70000;
2472                 break;
2473         case low_latency:
2474                 new_itr = 20000; /* aka hwitr = ~200 */
2475                 break;
2476         case bulk_latency:
2477                 new_itr = 4000;
2478                 break;
2479         default:
2480                 break;
2481         }
2482
2483 set_itr_now:
2484         if (new_itr != adapter->itr) {
2485                 /*
2486                  * this attempts to bias the interrupt rate towards Bulk
2487                  * by adding intermediate steps when interrupt rate is
2488                  * increasing
2489                  */
2490                 new_itr = new_itr > adapter->itr ?
2491                              min(adapter->itr + (new_itr >> 2), new_itr) :
2492                              new_itr;
2493                 adapter->itr = new_itr;
2494                 adapter->rx_ring->itr_val = new_itr;
2495                 if (adapter->msix_entries)
2496                         adapter->rx_ring->set_itr = 1;
2497                 else
2498                         if (new_itr)
2499                                 ew32(ITR, 1000000000 / (new_itr * 256));
2500                         else
2501                                 ew32(ITR, 0);
2502         }
2503 }
2504
2505 /**
2506  * e1000_alloc_queues - Allocate memory for all rings
2507  * @adapter: board private structure to initialize
2508  **/
2509 static int __devinit e1000_alloc_queues(struct e1000_adapter *adapter)
2510 {
2511         int size = sizeof(struct e1000_ring);
2512
2513         adapter->tx_ring = kzalloc(size, GFP_KERNEL);
2514         if (!adapter->tx_ring)
2515                 goto err;
2516         adapter->tx_ring->count = adapter->tx_ring_count;
2517         adapter->tx_ring->adapter = adapter;
2518
2519         adapter->rx_ring = kzalloc(size, GFP_KERNEL);
2520         if (!adapter->rx_ring)
2521                 goto err;
2522         adapter->rx_ring->count = adapter->rx_ring_count;
2523         adapter->rx_ring->adapter = adapter;
2524
2525         return 0;
2526 err:
2527         e_err("Unable to allocate memory for queues\n");
2528         kfree(adapter->rx_ring);
2529         kfree(adapter->tx_ring);
2530         return -ENOMEM;
2531 }
2532
2533 /**
2534  * e1000e_poll - NAPI Rx polling callback
2535  * @napi: struct associated with this polling callback
2536  * @weight: number of packets driver is allowed to process this poll
2537  **/
2538 static int e1000e_poll(struct napi_struct *napi, int weight)
2539 {
2540         struct e1000_adapter *adapter = container_of(napi, struct e1000_adapter,
2541                                                      napi);
2542         struct e1000_hw *hw = &adapter->hw;
2543         struct net_device *poll_dev = adapter->netdev;
2544         int tx_cleaned = 1, work_done = 0;
2545
2546         adapter = netdev_priv(poll_dev);
2547
2548         if (!adapter->msix_entries ||
2549             (adapter->rx_ring->ims_val & adapter->tx_ring->ims_val))
2550                 tx_cleaned = e1000_clean_tx_irq(adapter->tx_ring);
2551
2552         adapter->clean_rx(adapter->rx_ring, &work_done, weight);
2553
2554         if (!tx_cleaned)
2555                 work_done = weight;
2556
2557         /* If weight not fully consumed, exit the polling mode */
2558         if (work_done < weight) {
2559                 if (adapter->itr_setting & 3)
2560                         e1000_set_itr(adapter);
2561                 napi_complete(napi);
2562                 if (!test_bit(__E1000_DOWN, &adapter->state)) {
2563                         if (adapter->msix_entries)
2564                                 ew32(IMS, adapter->rx_ring->ims_val);
2565                         else
2566                                 e1000_irq_enable(adapter);
2567                 }
2568         }
2569
2570         return work_done;
2571 }
2572
2573 static int e1000_vlan_rx_add_vid(struct net_device *netdev, u16 vid)
2574 {
2575         struct e1000_adapter *adapter = netdev_priv(netdev);
2576         struct e1000_hw *hw = &adapter->hw;
2577         u32 vfta, index;
2578
2579         /* don't update vlan cookie if already programmed */
2580         if ((adapter->hw.mng_cookie.status &
2581              E1000_MNG_DHCP_COOKIE_STATUS_VLAN) &&
2582             (vid == adapter->mng_vlan_id))
2583                 return 0;
2584
2585         /* add VID to filter table */
2586         if (adapter->flags & FLAG_HAS_HW_VLAN_FILTER) {
2587                 index = (vid >> 5) & 0x7F;
2588                 vfta = E1000_READ_REG_ARRAY(hw, E1000_VFTA, index);
2589                 vfta |= (1 << (vid & 0x1F));
2590                 hw->mac.ops.write_vfta(hw, index, vfta);
2591         }
2592
2593         set_bit(vid, adapter->active_vlans);
2594
2595         return 0;
2596 }
2597
2598 static int e1000_vlan_rx_kill_vid(struct net_device *netdev, u16 vid)
2599 {
2600         struct e1000_adapter *adapter = netdev_priv(netdev);
2601         struct e1000_hw *hw = &adapter->hw;
2602         u32 vfta, index;
2603
2604         if ((adapter->hw.mng_cookie.status &
2605              E1000_MNG_DHCP_COOKIE_STATUS_VLAN) &&
2606             (vid == adapter->mng_vlan_id)) {
2607                 /* release control to f/w */
2608                 e1000e_release_hw_control(adapter);
2609                 return 0;
2610         }
2611
2612         /* remove VID from filter table */
2613         if (adapter->flags & FLAG_HAS_HW_VLAN_FILTER) {
2614                 index = (vid >> 5) & 0x7F;
2615                 vfta = E1000_READ_REG_ARRAY(hw, E1000_VFTA, index);
2616                 vfta &= ~(1 << (vid & 0x1F));
2617                 hw->mac.ops.write_vfta(hw, index, vfta);
2618         }
2619
2620         clear_bit(vid, adapter->active_vlans);
2621
2622         return 0;
2623 }
2624
2625 /**
2626  * e1000e_vlan_filter_disable - helper to disable hw VLAN filtering
2627  * @adapter: board private structure to initialize
2628  **/
2629 static void e1000e_vlan_filter_disable(struct e1000_adapter *adapter)
2630 {
2631         struct net_device *netdev = adapter->netdev;
2632         struct e1000_hw *hw = &adapter->hw;
2633         u32 rctl;
2634
2635         if (adapter->flags & FLAG_HAS_HW_VLAN_FILTER) {
2636                 /* disable VLAN receive filtering */
2637                 rctl = er32(RCTL);
2638                 rctl &= ~(E1000_RCTL_VFE | E1000_RCTL_CFIEN);
2639                 ew32(RCTL, rctl);
2640
2641                 if (adapter->mng_vlan_id != (u16)E1000_MNG_VLAN_NONE) {
2642                         e1000_vlan_rx_kill_vid(netdev, adapter->mng_vlan_id);
2643                         adapter->mng_vlan_id = E1000_MNG_VLAN_NONE;
2644                 }
2645         }
2646 }
2647
2648 /**
2649  * e1000e_vlan_filter_enable - helper to enable HW VLAN filtering
2650  * @adapter: board private structure to initialize
2651  **/
2652 static void e1000e_vlan_filter_enable(struct e1000_adapter *adapter)
2653 {
2654         struct e1000_hw *hw = &adapter->hw;
2655         u32 rctl;
2656
2657         if (adapter->flags & FLAG_HAS_HW_VLAN_FILTER) {
2658                 /* enable VLAN receive filtering */
2659                 rctl = er32(RCTL);
2660                 rctl |= E1000_RCTL_VFE;
2661                 rctl &= ~E1000_RCTL_CFIEN;
2662                 ew32(RCTL, rctl);
2663         }
2664 }
2665
2666 /**
2667  * e1000e_vlan_strip_enable - helper to disable HW VLAN stripping
2668  * @adapter: board private structure to initialize
2669  **/
2670 static void e1000e_vlan_strip_disable(struct e1000_adapter *adapter)
2671 {
2672         struct e1000_hw *hw = &adapter->hw;
2673         u32 ctrl;
2674
2675         /* disable VLAN tag insert/strip */
2676         ctrl = er32(CTRL);
2677         ctrl &= ~E1000_CTRL_VME;
2678         ew32(CTRL, ctrl);
2679 }
2680
2681 /**
2682  * e1000e_vlan_strip_enable - helper to enable HW VLAN stripping
2683  * @adapter: board private structure to initialize
2684  **/
2685 static void e1000e_vlan_strip_enable(struct e1000_adapter *adapter)
2686 {
2687         struct e1000_hw *hw = &adapter->hw;
2688         u32 ctrl;
2689
2690         /* enable VLAN tag insert/strip */
2691         ctrl = er32(CTRL);
2692         ctrl |= E1000_CTRL_VME;
2693         ew32(CTRL, ctrl);
2694 }
2695
2696 static void e1000_update_mng_vlan(struct e1000_adapter *adapter)
2697 {
2698         struct net_device *netdev = adapter->netdev;
2699         u16 vid = adapter->hw.mng_cookie.vlan_id;
2700         u16 old_vid = adapter->mng_vlan_id;
2701
2702         if (adapter->hw.mng_cookie.status &
2703             E1000_MNG_DHCP_COOKIE_STATUS_VLAN) {
2704                 e1000_vlan_rx_add_vid(netdev, vid);
2705                 adapter->mng_vlan_id = vid;
2706         }
2707
2708         if ((old_vid != (u16)E1000_MNG_VLAN_NONE) && (vid != old_vid))
2709                 e1000_vlan_rx_kill_vid(netdev, old_vid);
2710 }
2711
2712 static void e1000_restore_vlan(struct e1000_adapter *adapter)
2713 {
2714         u16 vid;
2715
2716         e1000_vlan_rx_add_vid(adapter->netdev, 0);
2717
2718         for_each_set_bit(vid, adapter->active_vlans, VLAN_N_VID)
2719                 e1000_vlan_rx_add_vid(adapter->netdev, vid);
2720 }
2721
2722 static void e1000_init_manageability_pt(struct e1000_adapter *adapter)
2723 {
2724         struct e1000_hw *hw = &adapter->hw;
2725         u32 manc, manc2h, mdef, i, j;
2726
2727         if (!(adapter->flags & FLAG_MNG_PT_ENABLED))
2728                 return;
2729
2730         manc = er32(MANC);
2731
2732         /*
2733          * enable receiving management packets to the host. this will probably
2734          * generate destination unreachable messages from the host OS, but
2735          * the packets will be handled on SMBUS
2736          */
2737         manc |= E1000_MANC_EN_MNG2HOST;
2738         manc2h = er32(MANC2H);
2739
2740         switch (hw->mac.type) {
2741         default:
2742                 manc2h |= (E1000_MANC2H_PORT_623 | E1000_MANC2H_PORT_664);
2743                 break;
2744         case e1000_82574:
2745         case e1000_82583:
2746                 /*
2747                  * Check if IPMI pass-through decision filter already exists;
2748                  * if so, enable it.
2749                  */
2750                 for (i = 0, j = 0; i < 8; i++) {
2751                         mdef = er32(MDEF(i));
2752
2753                         /* Ignore filters with anything other than IPMI ports */
2754                         if (mdef & ~(E1000_MDEF_PORT_623 | E1000_MDEF_PORT_664))
2755                                 continue;
2756
2757                         /* Enable this decision filter in MANC2H */
2758                         if (mdef)
2759                                 manc2h |= (1 << i);
2760
2761                         j |= mdef;
2762                 }
2763
2764                 if (j == (E1000_MDEF_PORT_623 | E1000_MDEF_PORT_664))
2765                         break;
2766
2767                 /* Create new decision filter in an empty filter */
2768                 for (i = 0, j = 0; i < 8; i++)
2769                         if (er32(MDEF(i)) == 0) {
2770                                 ew32(MDEF(i), (E1000_MDEF_PORT_623 |
2771                                                E1000_MDEF_PORT_664));
2772                                 manc2h |= (1 << 1);
2773                                 j++;
2774                                 break;
2775                         }
2776
2777                 if (!j)
2778                         e_warn("Unable to create IPMI pass-through filter\n");
2779                 break;
2780         }
2781
2782         ew32(MANC2H, manc2h);
2783         ew32(MANC, manc);
2784 }
2785
2786 /**
2787  * e1000_configure_tx - Configure Transmit Unit after Reset
2788  * @adapter: board private structure
2789  *
2790  * Configure the Tx unit of the MAC after a reset.
2791  **/
2792 static void e1000_configure_tx(struct e1000_adapter *adapter)
2793 {
2794         struct e1000_hw *hw = &adapter->hw;
2795         struct e1000_ring *tx_ring = adapter->tx_ring;
2796         u64 tdba;
2797         u32 tdlen, tarc;
2798
2799         /* Setup the HW Tx Head and Tail descriptor pointers */
2800         tdba = tx_ring->dma;
2801         tdlen = tx_ring->count * sizeof(struct e1000_tx_desc);
2802         ew32(TDBAL(0), (tdba & DMA_BIT_MASK(32)));
2803         ew32(TDBAH(0), (tdba >> 32));
2804         ew32(TDLEN(0), tdlen);
2805         ew32(TDH(0), 0);
2806         ew32(TDT(0), 0);
2807         tx_ring->head = adapter->hw.hw_addr + E1000_TDH(0);
2808         tx_ring->tail = adapter->hw.hw_addr + E1000_TDT(0);
2809
2810         /* Set the Tx Interrupt Delay register */
2811         ew32(TIDV, adapter->tx_int_delay);
2812         /* Tx irq moderation */
2813         ew32(TADV, adapter->tx_abs_int_delay);
2814
2815         if (adapter->flags2 & FLAG2_DMA_BURST) {
2816                 u32 txdctl = er32(TXDCTL(0));
2817                 txdctl &= ~(E1000_TXDCTL_PTHRESH | E1000_TXDCTL_HTHRESH |
2818                             E1000_TXDCTL_WTHRESH);
2819                 /*
2820                  * set up some performance related parameters to encourage the
2821                  * hardware to use the bus more efficiently in bursts, depends
2822                  * on the tx_int_delay to be enabled,
2823                  * wthresh = 5 ==> burst write a cacheline (64 bytes) at a time
2824                  * hthresh = 1 ==> prefetch when one or more available
2825                  * pthresh = 0x1f ==> prefetch if internal cache 31 or less
2826                  * BEWARE: this seems to work but should be considered first if
2827                  * there are Tx hangs or other Tx related bugs
2828                  */
2829                 txdctl |= E1000_TXDCTL_DMA_BURST_ENABLE;
2830                 ew32(TXDCTL(0), txdctl);
2831         }
2832         /* erratum work around: set txdctl the same for both queues */
2833         ew32(TXDCTL(1), er32(TXDCTL(0)));
2834
2835         if (adapter->flags & FLAG_TARC_SPEED_MODE_BIT) {
2836                 tarc = er32(TARC(0));
2837                 /*
2838                  * set the speed mode bit, we'll clear it if we're not at
2839                  * gigabit link later
2840                  */
2841 #define SPEED_MODE_BIT (1 << 21)
2842                 tarc |= SPEED_MODE_BIT;
2843                 ew32(TARC(0), tarc);
2844         }
2845
2846         /* errata: program both queues to unweighted RR */
2847         if (adapter->flags & FLAG_TARC_SET_BIT_ZERO) {
2848                 tarc = er32(TARC(0));
2849                 tarc |= 1;
2850                 ew32(TARC(0), tarc);
2851                 tarc = er32(TARC(1));
2852                 tarc |= 1;
2853                 ew32(TARC(1), tarc);
2854         }
2855
2856         /* Setup Transmit Descriptor Settings for eop descriptor */
2857         adapter->txd_cmd = E1000_TXD_CMD_EOP | E1000_TXD_CMD_IFCS;
2858
2859         /* only set IDE if we are delaying interrupts using the timers */
2860         if (adapter->tx_int_delay)
2861                 adapter->txd_cmd |= E1000_TXD_CMD_IDE;
2862
2863         /* enable Report Status bit */
2864         adapter->txd_cmd |= E1000_TXD_CMD_RS;
2865
2866         hw->mac.ops.config_collision_dist(hw);
2867 }
2868
2869 /**
2870  * e1000_setup_rctl - configure the receive control registers
2871  * @adapter: Board private structure
2872  **/
2873 #define PAGE_USE_COUNT(S) (((S) >> PAGE_SHIFT) + \
2874                            (((S) & (PAGE_SIZE - 1)) ? 1 : 0))
2875 static void e1000_setup_rctl(struct e1000_adapter *adapter)
2876 {
2877         struct e1000_hw *hw = &adapter->hw;
2878         u32 rctl, rfctl;
2879         u32 pages = 0;
2880
2881         /* Workaround Si errata on 82579 - configure jumbo frame flow */
2882         if (hw->mac.type == e1000_pch2lan) {
2883                 s32 ret_val;
2884
2885                 if (adapter->netdev->mtu > ETH_DATA_LEN)
2886                         ret_val = e1000_lv_jumbo_workaround_ich8lan(hw, true);
2887                 else
2888                         ret_val = e1000_lv_jumbo_workaround_ich8lan(hw, false);
2889
2890                 if (ret_val)
2891                         e_dbg("failed to enable jumbo frame workaround mode\n");
2892         }
2893
2894         /* Program MC offset vector base */
2895         rctl = er32(RCTL);
2896         rctl &= ~(3 << E1000_RCTL_MO_SHIFT);
2897         rctl |= E1000_RCTL_EN | E1000_RCTL_BAM |
2898                 E1000_RCTL_LBM_NO | E1000_RCTL_RDMTS_HALF |
2899                 (adapter->hw.mac.mc_filter_type << E1000_RCTL_MO_SHIFT);
2900
2901         /* Do not Store bad packets */
2902         rctl &= ~E1000_RCTL_SBP;
2903
2904         /* Enable Long Packet receive */
2905         if (adapter->netdev->mtu <= ETH_DATA_LEN)
2906                 rctl &= ~E1000_RCTL_LPE;
2907         else
2908                 rctl |= E1000_RCTL_LPE;
2909
2910         /* Some systems expect that the CRC is included in SMBUS traffic. The
2911          * hardware strips the CRC before sending to both SMBUS (BMC) and to
2912          * host memory when this is enabled
2913          */
2914         if (adapter->flags2 & FLAG2_CRC_STRIPPING)
2915                 rctl |= E1000_RCTL_SECRC;
2916
2917         /* Workaround Si errata on 82577 PHY - configure IPG for jumbos */
2918         if ((hw->phy.type == e1000_phy_82577) && (rctl & E1000_RCTL_LPE)) {
2919                 u16 phy_data;
2920
2921                 e1e_rphy(hw, PHY_REG(770, 26), &phy_data);
2922                 phy_data &= 0xfff8;
2923                 phy_data |= (1 << 2);
2924                 e1e_wphy(hw, PHY_REG(770, 26), phy_data);
2925
2926                 e1e_rphy(hw, 22, &phy_data);
2927                 phy_data &= 0x0fff;
2928                 phy_data |= (1 << 14);
2929                 e1e_wphy(hw, 0x10, 0x2823);
2930                 e1e_wphy(hw, 0x11, 0x0003);
2931                 e1e_wphy(hw, 22, phy_data);
2932         }
2933
2934         /* Setup buffer sizes */
2935         rctl &= ~E1000_RCTL_SZ_4096;
2936         rctl |= E1000_RCTL_BSEX;
2937         switch (adapter->rx_buffer_len) {
2938         case 2048:
2939         default:
2940                 rctl |= E1000_RCTL_SZ_2048;
2941                 rctl &= ~E1000_RCTL_BSEX;
2942                 break;
2943         case 4096:
2944                 rctl |= E1000_RCTL_SZ_4096;
2945                 break;
2946         case 8192:
2947                 rctl |= E1000_RCTL_SZ_8192;
2948                 break;
2949         case 16384:
2950                 rctl |= E1000_RCTL_SZ_16384;
2951                 break;
2952         }
2953
2954         /* Enable Extended Status in all Receive Descriptors */
2955         rfctl = er32(RFCTL);
2956         rfctl |= E1000_RFCTL_EXTEN;
2957
2958         /*
2959          * 82571 and greater support packet-split where the protocol
2960          * header is placed in skb->data and the packet data is
2961          * placed in pages hanging off of skb_shinfo(skb)->nr_frags.
2962          * In the case of a non-split, skb->data is linearly filled,
2963          * followed by the page buffers.  Therefore, skb->data is
2964          * sized to hold the largest protocol header.
2965          *
2966          * allocations using alloc_page take too long for regular MTU
2967          * so only enable packet split for jumbo frames
2968          *
2969          * Using pages when the page size is greater than 16k wastes
2970          * a lot of memory, since we allocate 3 pages at all times
2971          * per packet.
2972          */
2973         pages = PAGE_USE_COUNT(adapter->netdev->mtu);
2974         if ((pages <= 3) && (PAGE_SIZE <= 16384) && (rctl & E1000_RCTL_LPE))
2975                 adapter->rx_ps_pages = pages;
2976         else
2977                 adapter->rx_ps_pages = 0;
2978
2979         if (adapter->rx_ps_pages) {
2980                 u32 psrctl = 0;
2981
2982                 /*
2983                  * disable packet split support for IPv6 extension headers,
2984                  * because some malformed IPv6 headers can hang the Rx
2985                  */
2986                 rfctl |= (E1000_RFCTL_IPV6_EX_DIS |
2987                           E1000_RFCTL_NEW_IPV6_EXT_DIS);
2988
2989                 /* Enable Packet split descriptors */
2990                 rctl |= E1000_RCTL_DTYP_PS;
2991
2992                 psrctl |= adapter->rx_ps_bsize0 >>
2993                         E1000_PSRCTL_BSIZE0_SHIFT;
2994
2995                 switch (adapter->rx_ps_pages) {
2996                 case 3:
2997                         psrctl |= PAGE_SIZE <<
2998                                 E1000_PSRCTL_BSIZE3_SHIFT;
2999                 case 2:
3000                         psrctl |= PAGE_SIZE <<
3001                                 E1000_PSRCTL_BSIZE2_SHIFT;
3002                 case 1:
3003                         psrctl |= PAGE_SIZE >>
3004                                 E1000_PSRCTL_BSIZE1_SHIFT;
3005                         break;
3006                 }
3007
3008                 ew32(PSRCTL, psrctl);
3009         }
3010
3011         /* This is useful for sniffing bad packets. */
3012         if (adapter->netdev->features & NETIF_F_RXALL) {
3013                 /* UPE and MPE will be handled by normal PROMISC logic
3014                  * in e1000e_set_rx_mode */
3015                 rctl |= (E1000_RCTL_SBP | /* Receive bad packets */
3016                          E1000_RCTL_BAM | /* RX All Bcast Pkts */
3017                          E1000_RCTL_PMCF); /* RX All MAC Ctrl Pkts */
3018
3019                 rctl &= ~(E1000_RCTL_VFE | /* Disable VLAN filter */
3020                           E1000_RCTL_DPF | /* Allow filtered pause */
3021                           E1000_RCTL_CFIEN); /* Dis VLAN CFIEN Filter */
3022                 /* Do not mess with E1000_CTRL_VME, it affects transmit as well,
3023                  * and that breaks VLANs.
3024                  */
3025         }
3026
3027         ew32(RFCTL, rfctl);
3028         ew32(RCTL, rctl);
3029         /* just started the receive unit, no need to restart */
3030         adapter->flags &= ~FLAG_RX_RESTART_NOW;
3031 }
3032
3033 /**
3034  * e1000_configure_rx - Configure Receive Unit after Reset
3035  * @adapter: board private structure
3036  *
3037  * Configure the Rx unit of the MAC after a reset.
3038  **/
3039 static void e1000_configure_rx(struct e1000_adapter *adapter)
3040 {
3041         struct e1000_hw *hw = &adapter->hw;
3042         struct e1000_ring *rx_ring = adapter->rx_ring;
3043         u64 rdba;
3044         u32 rdlen, rctl, rxcsum, ctrl_ext;
3045
3046         if (adapter->rx_ps_pages) {
3047                 /* this is a 32 byte descriptor */
3048                 rdlen = rx_ring->count *
3049                     sizeof(union e1000_rx_desc_packet_split);
3050                 adapter->clean_rx = e1000_clean_rx_irq_ps;
3051                 adapter->alloc_rx_buf = e1000_alloc_rx_buffers_ps;
3052         } else if (adapter->netdev->mtu > ETH_FRAME_LEN + ETH_FCS_LEN) {
3053                 rdlen = rx_ring->count * sizeof(union e1000_rx_desc_extended);
3054                 adapter->clean_rx = e1000_clean_jumbo_rx_irq;
3055                 adapter->alloc_rx_buf = e1000_alloc_jumbo_rx_buffers;
3056         } else {
3057                 rdlen = rx_ring->count * sizeof(union e1000_rx_desc_extended);
3058                 adapter->clean_rx = e1000_clean_rx_irq;
3059                 adapter->alloc_rx_buf = e1000_alloc_rx_buffers;
3060         }
3061
3062         /* disable receives while setting up the descriptors */
3063         rctl = er32(RCTL);
3064         if (!(adapter->flags2 & FLAG2_NO_DISABLE_RX))
3065                 ew32(RCTL, rctl & ~E1000_RCTL_EN);
3066         e1e_flush();
3067         usleep_range(10000, 20000);
3068
3069         if (adapter->flags2 & FLAG2_DMA_BURST) {
3070                 /*
3071                  * set the writeback threshold (only takes effect if the RDTR
3072                  * is set). set GRAN=1 and write back up to 0x4 worth, and
3073                  * enable prefetching of 0x20 Rx descriptors
3074                  * granularity = 01
3075                  * wthresh = 04,
3076                  * hthresh = 04,
3077                  * pthresh = 0x20
3078                  */
3079                 ew32(RXDCTL(0), E1000_RXDCTL_DMA_BURST_ENABLE);
3080                 ew32(RXDCTL(1), E1000_RXDCTL_DMA_BURST_ENABLE);
3081
3082                 /*
3083                  * override the delay timers for enabling bursting, only if
3084                  * the value was not set by the user via module options
3085                  */
3086                 if (adapter->rx_int_delay == DEFAULT_RDTR)
3087                         adapter->rx_int_delay = BURST_RDTR;
3088                 if (adapter->rx_abs_int_delay == DEFAULT_RADV)
3089                         adapter->rx_abs_int_delay = BURST_RADV;
3090         }
3091
3092         /* set the Receive Delay Timer Register */
3093         ew32(RDTR, adapter->rx_int_delay);
3094
3095         /* irq moderation */
3096         ew32(RADV, adapter->rx_abs_int_delay);
3097         if ((adapter->itr_setting != 0) && (adapter->itr != 0))
3098                 ew32(ITR, 1000000000 / (adapter->itr * 256));
3099
3100         ctrl_ext = er32(CTRL_EXT);
3101         /* Auto-Mask interrupts upon ICR access */
3102         ctrl_ext |= E1000_CTRL_EXT_IAME;
3103         ew32(IAM, 0xffffffff);
3104         ew32(CTRL_EXT, ctrl_ext);
3105         e1e_flush();
3106
3107         /*
3108          * Setup the HW Rx Head and Tail Descriptor Pointers and
3109          * the Base and Length of the Rx Descriptor Ring
3110          */
3111         rdba = rx_ring->dma;
3112         ew32(RDBAL(0), (rdba & DMA_BIT_MASK(32)));
3113         ew32(RDBAH(0), (rdba >> 32));
3114         ew32(RDLEN(0), rdlen);
3115         ew32(RDH(0), 0);
3116         ew32(RDT(0), 0);
3117         rx_ring->head = adapter->hw.hw_addr + E1000_RDH(0);
3118         rx_ring->tail = adapter->hw.hw_addr + E1000_RDT(0);
3119
3120         /* Enable Receive Checksum Offload for TCP and UDP */
3121         rxcsum = er32(RXCSUM);
3122         if (adapter->netdev->features & NETIF_F_RXCSUM) {
3123                 rxcsum |= E1000_RXCSUM_TUOFL;
3124
3125                 /*
3126                  * IPv4 payload checksum for UDP fragments must be
3127                  * used in conjunction with packet-split.
3128                  */
3129                 if (adapter->rx_ps_pages)
3130                         rxcsum |= E1000_RXCSUM_IPPCSE;
3131         } else {
3132                 rxcsum &= ~E1000_RXCSUM_TUOFL;
3133                 /* no need to clear IPPCSE as it defaults to 0 */
3134         }
3135         ew32(RXCSUM, rxcsum);
3136
3137         if (adapter->hw.mac.type == e1000_pch2lan) {
3138                 /*
3139                  * With jumbo frames, excessive C-state transition
3140                  * latencies result in dropped transactions.
3141                  */
3142                 if (adapter->netdev->mtu > ETH_DATA_LEN) {
3143                         u32 rxdctl = er32(RXDCTL(0));
3144                         ew32(RXDCTL(0), rxdctl | 0x3);
3145                         pm_qos_update_request(&adapter->netdev->pm_qos_req, 55);
3146                 } else {
3147                         pm_qos_update_request(&adapter->netdev->pm_qos_req,
3148                                               PM_QOS_DEFAULT_VALUE);
3149                 }
3150         }
3151
3152         /* Enable Receives */
3153         ew32(RCTL, rctl);
3154 }
3155
3156 /**
3157  * e1000e_write_mc_addr_list - write multicast addresses to MTA
3158  * @netdev: network interface device structure
3159  *
3160  * Writes multicast address list to the MTA hash table.
3161  * Returns: -ENOMEM on failure
3162  *                0 on no addresses written
3163  *                X on writing X addresses to MTA
3164  */
3165 static int e1000e_write_mc_addr_list(struct net_device *netdev)
3166 {
3167         struct e1000_adapter *adapter = netdev_priv(netdev);
3168         struct e1000_hw *hw = &adapter->hw;
3169         struct netdev_hw_addr *ha;
3170         u8 *mta_list;
3171         int i;
3172
3173         if (netdev_mc_empty(netdev)) {
3174                 /* nothing to program, so clear mc list */
3175                 hw->mac.ops.update_mc_addr_list(hw, NULL, 0);
3176                 return 0;
3177         }
3178
3179         mta_list = kzalloc(netdev_mc_count(netdev) * ETH_ALEN, GFP_ATOMIC);
3180         if (!mta_list)
3181                 return -ENOMEM;
3182
3183         /* update_mc_addr_list expects a packed array of only addresses. */
3184         i = 0;
3185         netdev_for_each_mc_addr(ha, netdev)
3186                 memcpy(mta_list + (i++ * ETH_ALEN), ha->addr, ETH_ALEN);
3187
3188         hw->mac.ops.update_mc_addr_list(hw, mta_list, i);
3189         kfree(mta_list);
3190
3191         return netdev_mc_count(netdev);
3192 }
3193
3194 /**
3195  * e1000e_write_uc_addr_list - write unicast addresses to RAR table
3196  * @netdev: network interface device structure
3197  *
3198  * Writes unicast address list to the RAR table.
3199  * Returns: -ENOMEM on failure/insufficient address space
3200  *                0 on no addresses written
3201  *                X on writing X addresses to the RAR table
3202  **/
3203 static int e1000e_write_uc_addr_list(struct net_device *netdev)
3204 {
3205         struct e1000_adapter *adapter = netdev_priv(netdev);
3206         struct e1000_hw *hw = &adapter->hw;
3207         unsigned int rar_entries = hw->mac.rar_entry_count;
3208         int count = 0;
3209
3210         /* save a rar entry for our hardware address */
3211         rar_entries--;
3212
3213         /* save a rar entry for the LAA workaround */
3214         if (adapter->flags & FLAG_RESET_OVERWRITES_LAA)
3215                 rar_entries--;
3216
3217         /* return ENOMEM indicating insufficient memory for addresses */
3218         if (netdev_uc_count(netdev) > rar_entries)
3219                 return -ENOMEM;
3220
3221         if (!netdev_uc_empty(netdev) && rar_entries) {
3222                 struct netdev_hw_addr *ha;
3223
3224                 /*
3225                  * write the addresses in reverse order to avoid write
3226                  * combining
3227                  */
3228                 netdev_for_each_uc_addr(ha, netdev) {
3229                         if (!rar_entries)
3230                                 break;
3231                         e1000e_rar_set(hw, ha->addr, rar_entries--);
3232                         count++;
3233                 }
3234         }
3235
3236         /* zero out the remaining RAR entries not used above */
3237         for (; rar_entries > 0; rar_entries--) {
3238                 ew32(RAH(rar_entries), 0);
3239                 ew32(RAL(rar_entries), 0);
3240         }
3241         e1e_flush();
3242
3243         return count;
3244 }
3245
3246 /**
3247  * e1000e_set_rx_mode - secondary unicast, Multicast and Promiscuous mode set
3248  * @netdev: network interface device structure
3249  *
3250  * The ndo_set_rx_mode entry point is called whenever the unicast or multicast
3251  * address list or the network interface flags are updated.  This routine is
3252  * responsible for configuring the hardware for proper unicast, multicast,
3253  * promiscuous mode, and all-multi behavior.
3254  **/
3255 static void e1000e_set_rx_mode(struct net_device *netdev)
3256 {
3257         struct e1000_adapter *adapter = netdev_priv(netdev);
3258         struct e1000_hw *hw = &adapter->hw;
3259         u32 rctl;
3260
3261         /* Check for Promiscuous and All Multicast modes */
3262         rctl = er32(RCTL);
3263
3264         /* clear the affected bits */
3265         rctl &= ~(E1000_RCTL_UPE | E1000_RCTL_MPE);
3266
3267         if (netdev->flags & IFF_PROMISC) {
3268                 rctl |= (E1000_RCTL_UPE | E1000_RCTL_MPE);
3269                 /* Do not hardware filter VLANs in promisc mode */
3270                 e1000e_vlan_filter_disable(adapter);
3271         } else {
3272                 int count;
3273
3274                 if (netdev->flags & IFF_ALLMULTI) {
3275                         rctl |= E1000_RCTL_MPE;
3276                 } else {
3277                         /*
3278                          * Write addresses to the MTA, if the attempt fails
3279                          * then we should just turn on promiscuous mode so
3280                          * that we can at least receive multicast traffic
3281                          */
3282                         count = e1000e_write_mc_addr_list(netdev);
3283                         if (count < 0)
3284                                 rctl |= E1000_RCTL_MPE;
3285                 }
3286                 e1000e_vlan_filter_enable(adapter);
3287                 /*
3288                  * Write addresses to available RAR registers, if there is not
3289                  * sufficient space to store all the addresses then enable
3290                  * unicast promiscuous mode
3291                  */
3292                 count = e1000e_write_uc_addr_list(netdev);
3293                 if (count < 0)
3294                         rctl |= E1000_RCTL_UPE;
3295         }
3296
3297         ew32(RCTL, rctl);
3298
3299         if (netdev->features & NETIF_F_HW_VLAN_RX)
3300                 e1000e_vlan_strip_enable(adapter);
3301         else
3302                 e1000e_vlan_strip_disable(adapter);
3303 }
3304
3305 static void e1000e_setup_rss_hash(struct e1000_adapter *adapter)
3306 {
3307         struct e1000_hw *hw = &adapter->hw;
3308         u32 mrqc, rxcsum;
3309         int i;
3310         static const u32 rsskey[10] = {
3311                 0xda565a6d, 0xc20e5b25, 0x3d256741, 0xb08fa343, 0xcb2bcad0,
3312                 0xb4307bae, 0xa32dcb77, 0x0cf23080, 0x3bb7426a, 0xfa01acbe
3313         };
3314
3315         /* Fill out hash function seed */
3316         for (i = 0; i < 10; i++)
3317                 ew32(RSSRK(i), rsskey[i]);
3318
3319         /* Direct all traffic to queue 0 */
3320         for (i = 0; i < 32; i++)
3321                 ew32(RETA(i), 0);
3322
3323         /*
3324          * Disable raw packet checksumming so that RSS hash is placed in
3325          * descriptor on writeback.
3326          */
3327         rxcsum = er32(RXCSUM);
3328         rxcsum |= E1000_RXCSUM_PCSD;
3329
3330         ew32(RXCSUM, rxcsum);
3331
3332         mrqc = (E1000_MRQC_RSS_FIELD_IPV4 |
3333                 E1000_MRQC_RSS_FIELD_IPV4_TCP |
3334                 E1000_MRQC_RSS_FIELD_IPV6 |
3335                 E1000_MRQC_RSS_FIELD_IPV6_TCP |
3336                 E1000_MRQC_RSS_FIELD_IPV6_TCP_EX);
3337
3338         ew32(MRQC, mrqc);
3339 }
3340
3341 /**
3342  * e1000_configure - configure the hardware for Rx and Tx
3343  * @adapter: private board structure
3344  **/
3345 static void e1000_configure(struct e1000_adapter *adapter)
3346 {
3347         struct e1000_ring *rx_ring = adapter->rx_ring;
3348
3349         e1000e_set_rx_mode(adapter->netdev);
3350
3351         e1000_restore_vlan(adapter);
3352         e1000_init_manageability_pt(adapter);
3353
3354         e1000_configure_tx(adapter);
3355
3356         if (adapter->netdev->features & NETIF_F_RXHASH)
3357                 e1000e_setup_rss_hash(adapter);
3358         e1000_setup_rctl(adapter);
3359         e1000_configure_rx(adapter);
3360         adapter->alloc_rx_buf(rx_ring, e1000_desc_unused(rx_ring), GFP_KERNEL);
3361 }
3362
3363 /**
3364  * e1000e_power_up_phy - restore link in case the phy was powered down
3365  * @adapter: address of board private structure
3366  *
3367  * The phy may be powered down to save power and turn off link when the
3368  * driver is unloaded and wake on lan is not enabled (among others)
3369  * *** this routine MUST be followed by a call to e1000e_reset ***
3370  **/
3371 void e1000e_power_up_phy(struct e1000_adapter *adapter)
3372 {
3373         if (adapter->hw.phy.ops.power_up)
3374                 adapter->hw.phy.ops.power_up(&adapter->hw);
3375
3376         adapter->hw.mac.ops.setup_link(&adapter->hw);
3377 }
3378
3379 /**
3380  * e1000_power_down_phy - Power down the PHY
3381  *
3382  * Power down the PHY so no link is implied when interface is down.
3383  * The PHY cannot be powered down if management or WoL is active.
3384  */
3385 static void e1000_power_down_phy(struct e1000_adapter *adapter)
3386 {
3387         /* WoL is enabled */
3388         if (adapter->wol)
3389                 return;
3390
3391         if (adapter->hw.phy.ops.power_down)
3392                 adapter->hw.phy.ops.power_down(&adapter->hw);
3393 }
3394
3395 /**
3396  * e1000e_reset - bring the hardware into a known good state
3397  *
3398  * This function boots the hardware and enables some settings that
3399  * require a configuration cycle of the hardware - those cannot be
3400  * set/changed during runtime. After reset the device needs to be
3401  * properly configured for Rx, Tx etc.
3402  */
3403 void e1000e_reset(struct e1000_adapter *adapter)
3404 {
3405         struct e1000_mac_info *mac = &adapter->hw.mac;
3406         struct e1000_fc_info *fc = &adapter->hw.fc;
3407         struct e1000_hw *hw = &adapter->hw;
3408         u32 tx_space, min_tx_space, min_rx_space;
3409         u32 pba = adapter->pba;
3410         u16 hwm;
3411
3412         /* reset Packet Buffer Allocation to default */
3413         ew32(PBA, pba);
3414
3415         if (adapter->max_frame_size > ETH_FRAME_LEN + ETH_FCS_LEN) {
3416                 /*
3417                  * To maintain wire speed transmits, the Tx FIFO should be
3418                  * large enough to accommodate two full transmit packets,
3419                  * rounded up to the next 1KB and expressed in KB.  Likewise,
3420                  * the Rx FIFO should be large enough to accommodate at least
3421                  * one full receive packet and is similarly rounded up and
3422                  * expressed in KB.
3423                  */
3424                 pba = er32(PBA);
3425                 /* upper 16 bits has Tx packet buffer allocation size in KB */
3426                 tx_space = pba >> 16;
3427                 /* lower 16 bits has Rx packet buffer allocation size in KB */
3428                 pba &= 0xffff;
3429                 /*
3430                  * the Tx fifo also stores 16 bytes of information about the Tx
3431                  * but don't include ethernet FCS because hardware appends it
3432                  */
3433                 min_tx_space = (adapter->max_frame_size +
3434                                 sizeof(struct e1000_tx_desc) -
3435                                 ETH_FCS_LEN) * 2;
3436                 min_tx_space = ALIGN(min_tx_space, 1024);
3437                 min_tx_space >>= 10;
3438                 /* software strips receive CRC, so leave room for it */
3439                 min_rx_space = adapter->max_frame_size;
3440                 min_rx_space = ALIGN(min_rx_space, 1024);
3441                 min_rx_space >>= 10;
3442
3443                 /*
3444                  * If current Tx allocation is less than the min Tx FIFO size,
3445                  * and the min Tx FIFO size is less than the current Rx FIFO
3446                  * allocation, take space away from current Rx allocation
3447                  */
3448                 if ((tx_space < min_tx_space) &&
3449                     ((min_tx_space - tx_space) < pba)) {
3450                         pba -= min_tx_space - tx_space;
3451
3452                         /*
3453                          * if short on Rx space, Rx wins and must trump Tx
3454                          * adjustment or use Early Receive if available
3455                          */
3456                         if (pba < min_rx_space)
3457                                 pba = min_rx_space;
3458                 }
3459
3460                 ew32(PBA, pba);
3461         }
3462
3463         /*
3464          * flow control settings
3465          *
3466          * The high water mark must be low enough to fit one full frame
3467          * (or the size used for early receive) above it in the Rx FIFO.
3468          * Set it to the lower of:
3469          * - 90% of the Rx FIFO size, and
3470          * - the full Rx FIFO size minus one full frame
3471          */
3472         if (adapter->flags & FLAG_DISABLE_FC_PAUSE_TIME)
3473                 fc->pause_time = 0xFFFF;
3474         else
3475                 fc->pause_time = E1000_FC_PAUSE_TIME;
3476         fc->send_xon = true;
3477         fc->current_mode = fc->requested_mode;
3478
3479         switch (hw->mac.type) {
3480         case e1000_ich9lan:
3481         case e1000_ich10lan:
3482                 if (adapter->netdev->mtu > ETH_DATA_LEN) {
3483                         pba = 14;
3484                         ew32(PBA, pba);
3485                         fc->high_water = 0x2800;
3486                         fc->low_water = fc->high_water - 8;
3487                         break;
3488                 }
3489                 /* fall-through */
3490         default:
3491                 hwm = min(((pba << 10) * 9 / 10),
3492                           ((pba << 10) - adapter->max_frame_size));
3493
3494                 fc->high_water = hwm & E1000_FCRTH_RTH; /* 8-byte granularity */
3495                 fc->low_water = fc->high_water - 8;
3496                 break;
3497         case e1000_pchlan:
3498                 /*
3499                  * Workaround PCH LOM adapter hangs with certain network
3500                  * loads.  If hangs persist, try disabling Tx flow control.
3501                  */
3502                 if (adapter->netdev->mtu > ETH_DATA_LEN) {
3503                         fc->high_water = 0x3500;
3504                         fc->low_water  = 0x1500;
3505                 } else {
3506                         fc->high_water = 0x5000;
3507                         fc->low_water  = 0x3000;
3508                 }
3509                 fc->refresh_time = 0x1000;
3510                 break;
3511         case e1000_pch2lan:
3512                 fc->high_water = 0x05C20;
3513                 fc->low_water = 0x05048;
3514                 fc->pause_time = 0x0650;
3515                 fc->refresh_time = 0x0400;
3516                 if (adapter->netdev->mtu > ETH_DATA_LEN) {
3517                         pba = 14;
3518                         ew32(PBA, pba);
3519                 }
3520                 break;
3521         }
3522
3523         /*
3524          * Disable Adaptive Interrupt Moderation if 2 full packets cannot
3525          * fit in receive buffer.
3526          */
3527         if (adapter->itr_setting & 0x3) {
3528                 if ((adapter->max_frame_size * 2) > (pba << 10)) {
3529                         if (!(adapter->flags2 & FLAG2_DISABLE_AIM)) {
3530                                 dev_info(&adapter->pdev->dev,
3531                                         "Interrupt Throttle Rate turned off\n");
3532                                 adapter->flags2 |= FLAG2_DISABLE_AIM;
3533                                 ew32(ITR, 0);
3534                         }
3535                 } else if (adapter->flags2 & FLAG2_DISABLE_AIM) {
3536                         dev_info(&adapter->pdev->dev,
3537                                  "Interrupt Throttle Rate turned on\n");
3538                         adapter->flags2 &= ~FLAG2_DISABLE_AIM;
3539                         adapter->itr = 20000;
3540                         ew32(ITR, 1000000000 / (adapter->itr * 256));
3541                 }
3542         }
3543
3544         /* Allow time for pending master requests to run */
3545         mac->ops.reset_hw(hw);
3546
3547         /*
3548          * For parts with AMT enabled, let the firmware know
3549          * that the network interface is in control
3550          */
3551         if (adapter->flags & FLAG_HAS_AMT)
3552                 e1000e_get_hw_control(adapter);
3553
3554         ew32(WUC, 0);
3555
3556         if (mac->ops.init_hw(hw))
3557                 e_err("Hardware Error\n");
3558
3559         e1000_update_mng_vlan(adapter);
3560
3561         /* Enable h/w to recognize an 802.1Q VLAN Ethernet packet */
3562         ew32(VET, ETH_P_8021Q);
3563
3564         e1000e_reset_adaptive(hw);
3565
3566         if (!netif_running(adapter->netdev) &&
3567             !test_bit(__E1000_TESTING, &adapter->state)) {
3568                 e1000_power_down_phy(adapter);
3569                 return;
3570         }
3571
3572         e1000_get_phy_info(hw);
3573
3574         if ((adapter->flags & FLAG_HAS_SMART_POWER_DOWN) &&
3575             !(adapter->flags & FLAG_SMART_POWER_DOWN)) {
3576                 u16 phy_data = 0;
3577                 /*
3578                  * speed up time to link by disabling smart power down, ignore
3579                  * the return value of this function because there is nothing
3580                  * different we would do if it failed
3581                  */
3582                 e1e_rphy(hw, IGP02E1000_PHY_POWER_MGMT, &phy_data);
3583                 phy_data &= ~IGP02E1000_PM_SPD;
3584                 e1e_wphy(hw, IGP02E1000_PHY_POWER_MGMT, phy_data);
3585         }
3586 }
3587
3588 int e1000e_up(struct e1000_adapter *adapter)
3589 {
3590         struct e1000_hw *hw = &adapter->hw;
3591
3592         /* hardware has been reset, we need to reload some things */
3593         e1000_configure(adapter);
3594
3595         clear_bit(__E1000_DOWN, &adapter->state);
3596
3597         if (adapter->msix_entries)
3598                 e1000_configure_msix(adapter);
3599         e1000_irq_enable(adapter);
3600
3601         netif_start_queue(adapter->netdev);
3602
3603         /* fire a link change interrupt to start the watchdog */
3604         if (adapter->msix_entries)
3605                 ew32(ICS, E1000_ICS_LSC | E1000_ICR_OTHER);
3606         else
3607                 ew32(ICS, E1000_ICS_LSC);
3608
3609         return 0;
3610 }
3611
3612 static void e1000e_flush_descriptors(struct e1000_adapter *adapter)
3613 {
3614         struct e1000_hw *hw = &adapter->hw;
3615
3616         if (!(adapter->flags2 & FLAG2_DMA_BURST))
3617                 return;
3618
3619         /* flush pending descriptor writebacks to memory */
3620         ew32(TIDV, adapter->tx_int_delay | E1000_TIDV_FPD);
3621         ew32(RDTR, adapter->rx_int_delay | E1000_RDTR_FPD);
3622
3623         /* execute the writes immediately */
3624         e1e_flush();
3625
3626         /*
3627          * due to rare timing issues, write to TIDV/RDTR again to ensure the
3628          * write is successful
3629          */
3630         ew32(TIDV, adapter->tx_int_delay | E1000_TIDV_FPD);
3631         ew32(RDTR, adapter->rx_int_delay | E1000_RDTR_FPD);
3632
3633         /* execute the writes immediately */
3634         e1e_flush();
3635 }
3636
3637 static void e1000e_update_stats(struct e1000_adapter *adapter);
3638
3639 void e1000e_down(struct e1000_adapter *adapter)
3640 {
3641         struct net_device *netdev = adapter->netdev;
3642         struct e1000_hw *hw = &adapter->hw;
3643         u32 tctl, rctl;
3644
3645         /*
3646          * signal that we're down so the interrupt handler does not
3647          * reschedule our watchdog timer
3648          */
3649         set_bit(__E1000_DOWN, &adapter->state);
3650
3651         /* disable receives in the hardware */
3652         rctl = er32(RCTL);
3653         if (!(adapter->flags2 & FLAG2_NO_DISABLE_RX))
3654                 ew32(RCTL, rctl & ~E1000_RCTL_EN);
3655         /* flush and sleep below */
3656
3657         netif_stop_queue(netdev);
3658
3659         /* disable transmits in the hardware */
3660         tctl = er32(TCTL);
3661         tctl &= ~E1000_TCTL_EN;
3662         ew32(TCTL, tctl);
3663
3664         /* flush both disables and wait for them to finish */
3665         e1e_flush();
3666         usleep_range(10000, 20000);
3667
3668         e1000_irq_disable(adapter);
3669
3670         del_timer_sync(&adapter->watchdog_timer);
3671         del_timer_sync(&adapter->phy_info_timer);
3672
3673         netif_carrier_off(netdev);
3674
3675         spin_lock(&adapter->stats64_lock);
3676         e1000e_update_stats(adapter);
3677         spin_unlock(&adapter->stats64_lock);
3678
3679         e1000e_flush_descriptors(adapter);
3680         e1000_clean_tx_ring(adapter->tx_ring);
3681         e1000_clean_rx_ring(adapter->rx_ring);
3682
3683         adapter->link_speed = 0;
3684         adapter->link_duplex = 0;
3685
3686         if (!pci_channel_offline(adapter->pdev))
3687                 e1000e_reset(adapter);
3688
3689         /*
3690          * TODO: for power management, we could drop the link and
3691          * pci_disable_device here.
3692          */
3693 }
3694
3695 void e1000e_reinit_locked(struct e1000_adapter *adapter)
3696 {
3697         might_sleep();
3698         while (test_and_set_bit(__E1000_RESETTING, &adapter->state))
3699                 usleep_range(1000, 2000);
3700         e1000e_down(adapter);
3701         e1000e_up(adapter);
3702         clear_bit(__E1000_RESETTING, &adapter->state);
3703 }
3704
3705 /**
3706  * e1000_sw_init - Initialize general software structures (struct e1000_adapter)
3707  * @adapter: board private structure to initialize
3708  *
3709  * e1000_sw_init initializes the Adapter private data structure.
3710  * Fields are initialized based on PCI device information and
3711  * OS network device settings (MTU size).
3712  **/
3713 static int __devinit e1000_sw_init(struct e1000_adapter *adapter)
3714 {
3715         struct net_device *netdev = adapter->netdev;
3716
3717         adapter->rx_buffer_len = ETH_FRAME_LEN + VLAN_HLEN + ETH_FCS_LEN;
3718         adapter->rx_ps_bsize0 = 128;
3719         adapter->max_frame_size = netdev->mtu + ETH_HLEN + ETH_FCS_LEN;
3720         adapter->min_frame_size = ETH_ZLEN + ETH_FCS_LEN;
3721         adapter->tx_ring_count = E1000_DEFAULT_TXD;
3722         adapter->rx_ring_count = E1000_DEFAULT_RXD;
3723
3724         spin_lock_init(&adapter->stats64_lock);
3725
3726         e1000e_set_interrupt_capability(adapter);
3727
3728         if (e1000_alloc_queues(adapter))
3729                 return -ENOMEM;
3730
3731         /* Explicitly disable IRQ since the NIC can be in any state. */
3732         e1000_irq_disable(adapter);
3733
3734         set_bit(__E1000_DOWN, &adapter->state);
3735         return 0;
3736 }
3737
3738 /**
3739  * e1000_intr_msi_test - Interrupt Handler
3740  * @irq: interrupt number
3741  * @data: pointer to a network interface device structure
3742  **/
3743 static irqreturn_t e1000_intr_msi_test(int irq, void *data)
3744 {
3745         struct net_device *netdev = data;
3746         struct e1000_adapter *adapter = netdev_priv(netdev);
3747         struct e1000_hw *hw = &adapter->hw;
3748         u32 icr = er32(ICR);
3749
3750         e_dbg("icr is %08X\n", icr);
3751         if (icr & E1000_ICR_RXSEQ) {
3752                 adapter->flags &= ~FLAG_MSI_TEST_FAILED;
3753                 wmb();
3754         }
3755
3756         return IRQ_HANDLED;
3757 }
3758
3759 /**
3760  * e1000_test_msi_interrupt - Returns 0 for successful test
3761  * @adapter: board private struct
3762  *
3763  * code flow taken from tg3.c
3764  **/
3765 static int e1000_test_msi_interrupt(struct e1000_adapter *adapter)
3766 {
3767         struct net_device *netdev = adapter->netdev;
3768         struct e1000_hw *hw = &adapter->hw;
3769         int err;
3770
3771         /* poll_enable hasn't been called yet, so don't need disable */
3772         /* clear any pending events */
3773         er32(ICR);
3774
3775         /* free the real vector and request a test handler */
3776         e1000_free_irq(adapter);
3777         e1000e_reset_interrupt_capability(adapter);
3778
3779         /* Assume that the test fails, if it succeeds then the test
3780          * MSI irq handler will unset this flag */
3781         adapter->flags |= FLAG_MSI_TEST_FAILED;
3782
3783         err = pci_enable_msi(adapter->pdev);
3784         if (err)
3785                 goto msi_test_failed;
3786
3787         err = request_irq(adapter->pdev->irq, e1000_intr_msi_test, 0,
3788                           netdev->name, netdev);
3789         if (err) {
3790                 pci_disable_msi(adapter->pdev);
3791                 goto msi_test_failed;
3792         }
3793
3794         wmb();
3795
3796         e1000_irq_enable(adapter);
3797
3798         /* fire an unusual interrupt on the test handler */
3799         ew32(ICS, E1000_ICS_RXSEQ);
3800         e1e_flush();
3801         msleep(50);
3802
3803         e1000_irq_disable(adapter);
3804
3805         rmb();
3806
3807         if (adapter->flags & FLAG_MSI_TEST_FAILED) {
3808                 adapter->int_mode = E1000E_INT_MODE_LEGACY;
3809                 e_info("MSI interrupt test failed, using legacy interrupt.\n");
3810         } else {
3811                 e_dbg("MSI interrupt test succeeded!\n");
3812         }
3813
3814         free_irq(adapter->pdev->irq, netdev);
3815         pci_disable_msi(adapter->pdev);
3816
3817 msi_test_failed:
3818         e1000e_set_interrupt_capability(adapter);
3819         return e1000_request_irq(adapter);
3820 }
3821
3822 /**
3823  * e1000_test_msi - Returns 0 if MSI test succeeds or INTx mode is restored
3824  * @adapter: board private struct
3825  *
3826  * code flow taken from tg3.c, called with e1000 interrupts disabled.
3827  **/
3828 static int e1000_test_msi(struct e1000_adapter *adapter)
3829 {
3830         int err;
3831         u16 pci_cmd;
3832
3833         if (!(adapter->flags & FLAG_MSI_ENABLED))
3834                 return 0;
3835
3836         /* disable SERR in case the MSI write causes a master abort */
3837         pci_read_config_word(adapter->pdev, PCI_COMMAND, &pci_cmd);
3838         if (pci_cmd & PCI_COMMAND_SERR)
3839                 pci_write_config_word(adapter->pdev, PCI_COMMAND,
3840                                       pci_cmd & ~PCI_COMMAND_SERR);
3841
3842         err = e1000_test_msi_interrupt(adapter);
3843
3844         /* re-enable SERR */
3845         if (pci_cmd & PCI_COMMAND_SERR) {
3846                 pci_read_config_word(adapter->pdev, PCI_COMMAND, &pci_cmd);
3847                 pci_cmd |= PCI_COMMAND_SERR;
3848                 pci_write_config_word(adapter->pdev, PCI_COMMAND, pci_cmd);
3849         }
3850
3851         return err;
3852 }
3853
3854 /**
3855  * e1000_open - Called when a network interface is made active
3856  * @netdev: network interface device structure
3857  *
3858  * Returns 0 on success, negative value on failure
3859  *
3860  * The open entry point is called when a network interface is made
3861  * active by the system (IFF_UP).  At this point all resources needed
3862  * for transmit and receive operations are allocated, the interrupt
3863  * handler is registered with the OS, the watchdog timer is started,
3864  * and the stack is notified that the interface is ready.
3865  **/
3866 static int e1000_open(struct net_device *netdev)
3867 {
3868         struct e1000_adapter *adapter = netdev_priv(netdev);
3869         struct e1000_hw *hw = &adapter->hw;
3870         struct pci_dev *pdev = adapter->pdev;
3871         int err;
3872
3873         /* disallow open during test */
3874         if (test_bit(__E1000_TESTING, &adapter->state))
3875                 return -EBUSY;
3876
3877         pm_runtime_get_sync(&pdev->dev);
3878
3879         netif_carrier_off(netdev);
3880
3881         /* allocate transmit descriptors */
3882         err = e1000e_setup_tx_resources(adapter->tx_ring);
3883         if (err)
3884                 goto err_setup_tx;
3885
3886         /* allocate receive descriptors */
3887         err = e1000e_setup_rx_resources(adapter->rx_ring);
3888         if (err)
3889                 goto err_setup_rx;
3890
3891         /*
3892          * If AMT is enabled, let the firmware know that the network
3893          * interface is now open and reset the part to a known state.
3894          */
3895         if (adapter->flags & FLAG_HAS_AMT) {
3896                 e1000e_get_hw_control(adapter);
3897                 e1000e_reset(adapter);
3898         }
3899
3900         e1000e_power_up_phy(adapter);
3901
3902         adapter->mng_vlan_id = E1000_MNG_VLAN_NONE;
3903         if ((adapter->hw.mng_cookie.status &
3904              E1000_MNG_DHCP_COOKIE_STATUS_VLAN))
3905                 e1000_update_mng_vlan(adapter);
3906
3907         /* DMA latency requirement to workaround jumbo issue */
3908         if (adapter->hw.mac.type == e1000_pch2lan)
3909                 pm_qos_add_request(&adapter->netdev->pm_qos_req,
3910                                    PM_QOS_CPU_DMA_LATENCY,
3911                                    PM_QOS_DEFAULT_VALUE);
3912
3913         /*
3914          * before we allocate an interrupt, we must be ready to handle it.
3915          * Setting DEBUG_SHIRQ in the kernel makes it fire an interrupt
3916          * as soon as we call pci_request_irq, so we have to setup our
3917          * clean_rx handler before we do so.
3918          */
3919         e1000_configure(adapter);
3920
3921         err = e1000_request_irq(adapter);
3922         if (err)
3923                 goto err_req_irq;
3924
3925         /*
3926          * Work around PCIe errata with MSI interrupts causing some chipsets to
3927          * ignore e1000e MSI messages, which means we need to test our MSI
3928          * interrupt now
3929          */
3930         if (adapter->int_mode != E1000E_INT_MODE_LEGACY) {
3931                 err = e1000_test_msi(adapter);
3932                 if (err) {
3933                         e_err("Interrupt allocation failed\n");
3934                         goto err_req_irq;
3935                 }
3936         }
3937
3938         /* From here on the code is the same as e1000e_up() */
3939         clear_bit(__E1000_DOWN, &adapter->state);
3940
3941         napi_enable(&adapter->napi);
3942
3943         e1000_irq_enable(adapter);
3944
3945         adapter->tx_hang_recheck = false;
3946         netif_start_queue(netdev);
3947
3948         adapter->idle_check = true;
3949         pm_runtime_put(&pdev->dev);
3950
3951         /* fire a link status change interrupt to start the watchdog */
3952         if (adapter->msix_entries)
3953                 ew32(ICS, E1000_ICS_LSC | E1000_ICR_OTHER);
3954         else
3955                 ew32(ICS, E1000_ICS_LSC);
3956
3957         return 0;
3958
3959 err_req_irq:
3960         e1000e_release_hw_control(adapter);
3961         e1000_power_down_phy(adapter);
3962         e1000e_free_rx_resources(adapter->rx_ring);
3963 err_setup_rx:
3964         e1000e_free_tx_resources(adapter->tx_ring);
3965 err_setup_tx:
3966         e1000e_reset(adapter);
3967         pm_runtime_put_sync(&pdev->dev);
3968
3969         return err;
3970 }
3971
3972 /**
3973  * e1000_close - Disables a network interface
3974  * @netdev: network interface device structure
3975  *
3976  * Returns 0, this is not allowed to fail
3977  *
3978  * The close entry point is called when an interface is de-activated
3979  * by the OS.  The hardware is still under the drivers control, but
3980  * needs to be disabled.  A global MAC reset is issued to stop the
3981  * hardware, and all transmit and receive resources are freed.
3982  **/
3983 static int e1000_close(struct net_device *netdev)
3984 {
3985         struct e1000_adapter *adapter = netdev_priv(netdev);
3986         struct pci_dev *pdev = adapter->pdev;
3987         int count = E1000_CHECK_RESET_COUNT;
3988
3989         while (test_bit(__E1000_RESETTING, &adapter->state) && count--)
3990                 usleep_range(10000, 20000);
3991
3992         WARN_ON(test_bit(__E1000_RESETTING, &adapter->state));
3993
3994         pm_runtime_get_sync(&pdev->dev);
3995
3996         napi_disable(&adapter->napi);
3997
3998         if (!test_bit(__E1000_DOWN, &adapter->state)) {
3999                 e1000e_down(adapter);
4000                 e1000_free_irq(adapter);
4001         }
4002         e1000_power_down_phy(adapter);
4003
4004         e1000e_free_tx_resources(adapter->tx_ring);
4005         e1000e_free_rx_resources(adapter->rx_ring);
4006
4007         /*
4008          * kill manageability vlan ID if supported, but not if a vlan with
4009          * the same ID is registered on the host OS (let 8021q kill it)
4010          */
4011         if (adapter->hw.mng_cookie.status &
4012             E1000_MNG_DHCP_COOKIE_STATUS_VLAN)
4013                 e1000_vlan_rx_kill_vid(netdev, adapter->mng_vlan_id);
4014
4015         /*
4016          * If AMT is enabled, let the firmware know that the network
4017          * interface is now closed
4018          */
4019         if ((adapter->flags & FLAG_HAS_AMT) &&
4020             !test_bit(__E1000_TESTING, &adapter->state))
4021                 e1000e_release_hw_control(adapter);
4022
4023         if (adapter->hw.mac.type == e1000_pch2lan)
4024                 pm_qos_remove_request(&adapter->netdev->pm_qos_req);
4025
4026         pm_runtime_put_sync(&pdev->dev);
4027
4028         return 0;
4029 }
4030 /**
4031  * e1000_set_mac - Change the Ethernet Address of the NIC
4032  * @netdev: network interface device structure
4033  * @p: pointer to an address structure
4034  *
4035  * Returns 0 on success, negative on failure
4036  **/
4037 static int e1000_set_mac(struct net_device *netdev, void *p)
4038 {
4039         struct e1000_adapter *adapter = netdev_priv(netdev);
4040         struct sockaddr *addr = p;
4041
4042         if (!is_valid_ether_addr(addr->sa_data))
4043                 return -EADDRNOTAVAIL;
4044
4045         memcpy(netdev->dev_addr, addr->sa_data, netdev->addr_len);
4046         memcpy(adapter->hw.mac.addr, addr->sa_data, netdev->addr_len);
4047
4048         e1000e_rar_set(&adapter->hw, adapter->hw.mac.addr, 0);
4049
4050         if (adapter->flags & FLAG_RESET_OVERWRITES_LAA) {
4051                 /* activate the work around */
4052                 e1000e_set_laa_state_82571(&adapter->hw, 1);
4053
4054                 /*
4055                  * Hold a copy of the LAA in RAR[14] This is done so that
4056                  * between the time RAR[0] gets clobbered  and the time it
4057                  * gets fixed (in e1000_watchdog), the actual LAA is in one
4058                  * of the RARs and no incoming packets directed to this port
4059                  * are dropped. Eventually the LAA will be in RAR[0] and
4060                  * RAR[14]
4061                  */
4062                 e1000e_rar_set(&adapter->hw,
4063                               adapter->hw.mac.addr,
4064                               adapter->hw.mac.rar_entry_count - 1);
4065         }
4066
4067         return 0;
4068 }
4069
4070 /**
4071  * e1000e_update_phy_task - work thread to update phy
4072  * @work: pointer to our work struct
4073  *
4074  * this worker thread exists because we must acquire a
4075  * semaphore to read the phy, which we could msleep while
4076  * waiting for it, and we can't msleep in a timer.
4077  **/
4078 static void e1000e_update_phy_task(struct work_struct *work)
4079 {
4080         struct e1000_adapter *adapter = container_of(work,
4081                                         struct e1000_adapter, update_phy_task);
4082
4083         if (test_bit(__E1000_DOWN, &adapter->state))
4084                 return;
4085
4086         e1000_get_phy_info(&adapter->hw);
4087 }
4088
4089 /*
4090  * Need to wait a few seconds after link up to get diagnostic information from
4091  * the phy
4092  */
4093 static void e1000_update_phy_info(unsigned long data)
4094 {
4095         struct e1000_adapter *adapter = (struct e1000_adapter *) data;
4096
4097         if (test_bit(__E1000_DOWN, &adapter->state))
4098                 return;
4099
4100         schedule_work(&adapter->update_phy_task);
4101 }
4102
4103 /**
4104  * e1000e_update_phy_stats - Update the PHY statistics counters
4105  * @adapter: board private structure
4106  *
4107  * Read/clear the upper 16-bit PHY registers and read/accumulate lower
4108  **/
4109 static void e1000e_update_phy_stats(struct e1000_adapter *adapter)
4110 {
4111         struct e1000_hw *hw = &adapter->hw;
4112         s32 ret_val;
4113         u16 phy_data;
4114
4115         ret_val = hw->phy.ops.acquire(hw);
4116         if (ret_val)
4117                 return;
4118
4119         /*
4120          * A page set is expensive so check if already on desired page.
4121          * If not, set to the page with the PHY status registers.
4122          */
4123         hw->phy.addr = 1;
4124         ret_val = e1000e_read_phy_reg_mdic(hw, IGP01E1000_PHY_PAGE_SELECT,
4125                                            &phy_data);
4126         if (ret_val)
4127                 goto release;
4128         if (phy_data != (HV_STATS_PAGE << IGP_PAGE_SHIFT)) {
4129                 ret_val = hw->phy.ops.set_page(hw,
4130                                                HV_STATS_PAGE << IGP_PAGE_SHIFT);
4131                 if (ret_val)
4132                         goto release;
4133         }
4134
4135         /* Single Collision Count */
4136         hw->phy.ops.read_reg_page(hw, HV_SCC_UPPER, &phy_data);
4137         ret_val = hw->phy.ops.read_reg_page(hw, HV_SCC_LOWER, &phy_data);
4138         if (!ret_val)
4139                 adapter->stats.scc += phy_data;
4140
4141         /* Excessive Collision Count */
4142         hw->phy.ops.read_reg_page(hw, HV_ECOL_UPPER, &phy_data);
4143         ret_val = hw->phy.ops.read_reg_page(hw, HV_ECOL_LOWER, &phy_data);
4144         if (!ret_val)
4145                 adapter->stats.ecol += phy_data;
4146
4147         /* Multiple Collision Count */
4148         hw->phy.ops.read_reg_page(hw, HV_MCC_UPPER, &phy_data);
4149         ret_val = hw->phy.ops.read_reg_page(hw, HV_MCC_LOWER, &phy_data);
4150         if (!ret_val)
4151                 adapter->stats.mcc += phy_data;
4152
4153         /* Late Collision Count */
4154         hw->phy.ops.read_reg_page(hw, HV_LATECOL_UPPER, &phy_data);
4155         ret_val = hw->phy.ops.read_reg_page(hw, HV_LATECOL_LOWER, &phy_data);
4156         if (!ret_val)
4157                 adapter->stats.latecol += phy_data;
4158
4159         /* Collision Count - also used for adaptive IFS */
4160         hw->phy.ops.read_reg_page(hw, HV_COLC_UPPER, &phy_data);
4161         ret_val = hw->phy.ops.read_reg_page(hw, HV_COLC_LOWER, &phy_data);
4162         if (!ret_val)
4163                 hw->mac.collision_delta = phy_data;
4164
4165         /* Defer Count */
4166         hw->phy.ops.read_reg_page(hw, HV_DC_UPPER, &phy_data);
4167         ret_val = hw->phy.ops.read_reg_page(hw, HV_DC_LOWER, &phy_data);
4168         if (!ret_val)
4169                 adapter->stats.dc += phy_data;
4170
4171         /* Transmit with no CRS */
4172         hw->phy.ops.read_reg_page(hw, HV_TNCRS_UPPER, &phy_data);
4173         ret_val = hw->phy.ops.read_reg_page(hw, HV_TNCRS_LOWER, &phy_data);
4174         if (!ret_val)
4175                 adapter->stats.tncrs += phy_data;
4176
4177 release:
4178         hw->phy.ops.release(hw);
4179 }
4180
4181 /**
4182  * e1000e_update_stats - Update the board statistics counters
4183  * @adapter: board private structure
4184  **/
4185 static void e1000e_update_stats(struct e1000_adapter *adapter)
4186 {
4187         struct net_device *netdev = adapter->netdev;
4188         struct e1000_hw *hw = &adapter->hw;
4189         struct pci_dev *pdev = adapter->pdev;
4190
4191         /*
4192          * Prevent stats update while adapter is being reset, or if the pci
4193          * connection is down.
4194          */
4195         if (adapter->link_speed == 0)
4196                 return;
4197         if (pci_channel_offline(pdev))
4198                 return;
4199
4200         adapter->stats.crcerrs += er32(CRCERRS);
4201         adapter->stats.gprc += er32(GPRC);
4202         adapter->stats.gorc += er32(GORCL);
4203         er32(GORCH); /* Clear gorc */
4204         adapter->stats.bprc += er32(BPRC);
4205         adapter->stats.mprc += er32(MPRC);
4206         adapter->stats.roc += er32(ROC);
4207
4208         adapter->stats.mpc += er32(MPC);
4209
4210         /* Half-duplex statistics */
4211         if (adapter->link_duplex == HALF_DUPLEX) {
4212                 if (adapter->flags2 & FLAG2_HAS_PHY_STATS) {
4213                         e1000e_update_phy_stats(adapter);
4214                 } else {
4215                         adapter->stats.scc += er32(SCC);
4216                         adapter->stats.ecol += er32(ECOL);
4217                         adapter->stats.mcc += er32(MCC);
4218                         adapter->stats.latecol += er32(LATECOL);
4219                         adapter->stats.dc += er32(DC);
4220
4221                         hw->mac.collision_delta = er32(COLC);
4222
4223                         if ((hw->mac.type != e1000_82574) &&
4224                             (hw->mac.type != e1000_82583))
4225                                 adapter->stats.tncrs += er32(TNCRS);
4226                 }
4227                 adapter->stats.colc += hw->mac.collision_delta;
4228         }
4229
4230         adapter->stats.xonrxc += er32(XONRXC);
4231         adapter->stats.xontxc += er32(XONTXC);
4232         adapter->stats.xoffrxc += er32(XOFFRXC);
4233         adapter->stats.xofftxc += er32(XOFFTXC);
4234         adapter->stats.gptc += er32(GPTC);
4235         adapter->stats.gotc += er32(GOTCL);
4236         er32(GOTCH); /* Clear gotc */
4237         adapter->stats.rnbc += er32(RNBC);
4238         adapter->stats.ruc += er32(RUC);
4239
4240         adapter->stats.mptc += er32(MPTC);
4241         adapter->stats.bptc += er32(BPTC);
4242
4243         /* used for adaptive IFS */
4244
4245         hw->mac.tx_packet_delta = er32(TPT);
4246         adapter->stats.tpt += hw->mac.tx_packet_delta;
4247
4248         adapter->stats.algnerrc += er32(ALGNERRC);
4249         adapter->stats.rxerrc += er32(RXERRC);
4250         adapter->stats.cexterr += er32(CEXTERR);
4251         adapter->stats.tsctc += er32(TSCTC);
4252         adapter->stats.tsctfc += er32(TSCTFC);
4253
4254         /* Fill out the OS statistics structure */
4255         netdev->stats.multicast = adapter->stats.mprc;
4256         netdev->stats.collisions = adapter->stats.colc;
4257
4258         /* Rx Errors */
4259
4260         /*
4261          * RLEC on some newer hardware can be incorrect so build
4262          * our own version based on RUC and ROC
4263          */
4264         netdev->stats.rx_errors = adapter->stats.rxerrc +
4265                 adapter->stats.crcerrs + adapter->stats.algnerrc +
4266                 adapter->stats.ruc + adapter->stats.roc +
4267                 adapter->stats.cexterr;
4268         netdev->stats.rx_length_errors = adapter->stats.ruc +
4269                                               adapter->stats.roc;
4270         netdev->stats.rx_crc_errors = adapter->stats.crcerrs;
4271         netdev->stats.rx_frame_errors = adapter->stats.algnerrc;
4272         netdev->stats.rx_missed_errors = adapter->stats.mpc;
4273
4274         /* Tx Errors */
4275         netdev->stats.tx_errors = adapter->stats.ecol +
4276                                        adapter->stats.latecol;
4277         netdev->stats.tx_aborted_errors = adapter->stats.ecol;
4278         netdev->stats.tx_window_errors = adapter->stats.latecol;
4279         netdev->stats.tx_carrier_errors = adapter->stats.tncrs;
4280
4281         /* Tx Dropped needs to be maintained elsewhere */
4282
4283         /* Management Stats */
4284         adapter->stats.mgptc += er32(MGTPTC);
4285         adapter->stats.mgprc += er32(MGTPRC);
4286         adapter->stats.mgpdc += er32(MGTPDC);
4287 }
4288
4289 /**
4290  * e1000_phy_read_status - Update the PHY register status snapshot
4291  * @adapter: board private structure
4292  **/
4293 static void e1000_phy_read_status(struct e1000_adapter *adapter)
4294 {
4295         struct e1000_hw *hw = &adapter->hw;
4296         struct e1000_phy_regs *phy = &adapter->phy_regs;
4297
4298         if ((er32(STATUS) & E1000_STATUS_LU) &&
4299             (adapter->hw.phy.media_type == e1000_media_type_copper)) {
4300                 int ret_val;
4301
4302                 ret_val  = e1e_rphy(hw, PHY_CONTROL, &phy->bmcr);
4303                 ret_val |= e1e_rphy(hw, PHY_STATUS, &phy->bmsr);
4304                 ret_val |= e1e_rphy(hw, PHY_AUTONEG_ADV, &phy->advertise);
4305                 ret_val |= e1e_rphy(hw, PHY_LP_ABILITY, &phy->lpa);
4306                 ret_val |= e1e_rphy(hw, PHY_AUTONEG_EXP, &phy->expansion);
4307                 ret_val |= e1e_rphy(hw, PHY_1000T_CTRL, &phy->ctrl1000);
4308                 ret_val |= e1e_rphy(hw, PHY_1000T_STATUS, &phy->stat1000);
4309                 ret_val |= e1e_rphy(hw, PHY_EXT_STATUS, &phy->estatus);
4310                 if (ret_val)
4311                         e_warn("Error reading PHY register\n");
4312         } else {
4313                 /*
4314                  * Do not read PHY registers if link is not up
4315                  * Set values to typical power-on defaults
4316                  */
4317                 phy->bmcr = (BMCR_SPEED1000 | BMCR_ANENABLE | BMCR_FULLDPLX);
4318                 phy->bmsr = (BMSR_100FULL | BMSR_100HALF | BMSR_10FULL |
4319                              BMSR_10HALF | BMSR_ESTATEN | BMSR_ANEGCAPABLE |
4320                              BMSR_ERCAP);
4321                 phy->advertise = (ADVERTISE_PAUSE_ASYM | ADVERTISE_PAUSE_CAP |
4322                                   ADVERTISE_ALL | ADVERTISE_CSMA);
4323                 phy->lpa = 0;
4324                 phy->expansion = EXPANSION_ENABLENPAGE;
4325                 phy->ctrl1000 = ADVERTISE_1000FULL;
4326                 phy->stat1000 = 0;
4327                 phy->estatus = (ESTATUS_1000_TFULL | ESTATUS_1000_THALF);
4328         }
4329 }
4330
4331 static void e1000_print_link_info(struct e1000_adapter *adapter)
4332 {
4333         struct e1000_hw *hw = &adapter->hw;
4334         u32 ctrl = er32(CTRL);
4335
4336         /* Link status message must follow this format for user tools */
4337         printk(KERN_INFO "e1000e: %s NIC Link is Up %d Mbps %s Duplex, Flow Control: %s\n",
4338                 adapter->netdev->name,
4339                 adapter->link_speed,
4340                 adapter->link_duplex == FULL_DUPLEX ? "Full" : "Half",
4341                 (ctrl & E1000_CTRL_TFCE) && (ctrl & E1000_CTRL_RFCE) ? "Rx/Tx" :
4342                 (ctrl & E1000_CTRL_RFCE) ? "Rx" :
4343                 (ctrl & E1000_CTRL_TFCE) ? "Tx" : "None");
4344 }
4345
4346 static bool e1000e_has_link(struct e1000_adapter *adapter)
4347 {
4348         struct e1000_hw *hw = &adapter->hw;
4349         bool link_active = false;
4350         s32 ret_val = 0;
4351
4352         /*
4353          * get_link_status is set on LSC (link status) interrupt or
4354          * Rx sequence error interrupt.  get_link_status will stay
4355          * false until the check_for_link establishes link
4356          * for copper adapters ONLY
4357          */
4358         switch (hw->phy.media_type) {
4359         case e1000_media_type_copper:
4360                 if (hw->mac.get_link_status) {
4361                         ret_val = hw->mac.ops.check_for_link(hw);
4362                         link_active = !hw->mac.get_link_status;
4363                 } else {
4364                         link_active = true;
4365                 }
4366                 break;
4367         case e1000_media_type_fiber:
4368                 ret_val = hw->mac.ops.check_for_link(hw);
4369                 link_active = !!(er32(STATUS) & E1000_STATUS_LU);
4370                 break;
4371         case e1000_media_type_internal_serdes:
4372                 ret_val = hw->mac.ops.check_for_link(hw);
4373                 link_active = adapter->hw.mac.serdes_has_link;
4374                 break;
4375         default:
4376         case e1000_media_type_unknown:
4377                 break;
4378         }
4379
4380         if ((ret_val == E1000_ERR_PHY) && (hw->phy.type == e1000_phy_igp_3) &&
4381             (er32(CTRL) & E1000_PHY_CTRL_GBE_DISABLE)) {
4382                 /* See e1000_kmrn_lock_loss_workaround_ich8lan() */
4383                 e_info("Gigabit has been disabled, downgrading speed\n");
4384         }
4385
4386         return link_active;
4387 }
4388
4389 static void e1000e_enable_receives(struct e1000_adapter *adapter)
4390 {
4391         /* make sure the receive unit is started */
4392         if ((adapter->flags & FLAG_RX_NEEDS_RESTART) &&
4393             (adapter->flags & FLAG_RX_RESTART_NOW)) {
4394                 struct e1000_hw *hw = &adapter->hw;
4395                 u32 rctl = er32(RCTL);
4396                 ew32(RCTL, rctl | E1000_RCTL_EN);
4397                 adapter->flags &= ~FLAG_RX_RESTART_NOW;
4398         }
4399 }
4400
4401 static void e1000e_check_82574_phy_workaround(struct e1000_adapter *adapter)
4402 {
4403         struct e1000_hw *hw = &adapter->hw;
4404
4405         /*
4406          * With 82574 controllers, PHY needs to be checked periodically
4407          * for hung state and reset, if two calls return true
4408          */
4409         if (e1000_check_phy_82574(hw))
4410                 adapter->phy_hang_count++;
4411         else
4412                 adapter->phy_hang_count = 0;
4413
4414         if (adapter->phy_hang_count > 1) {
4415                 adapter->phy_hang_count = 0;
4416                 schedule_work(&adapter->reset_task);
4417         }
4418 }
4419
4420 /**
4421  * e1000_watchdog - Timer Call-back
4422  * @data: pointer to adapter cast into an unsigned long
4423  **/
4424 static void e1000_watchdog(unsigned long data)
4425 {
4426         struct e1000_adapter *adapter = (struct e1000_adapter *) data;
4427
4428         /* Do the rest outside of interrupt context */
4429         schedule_work(&adapter->watchdog_task);
4430
4431         /* TODO: make this use queue_delayed_work() */
4432 }
4433
4434 static void e1000_watchdog_task(struct work_struct *work)
4435 {
4436         struct e1000_adapter *adapter = container_of(work,
4437                                         struct e1000_adapter, watchdog_task);
4438         struct net_device *netdev = adapter->netdev;
4439         struct e1000_mac_info *mac = &adapter->hw.mac;
4440         struct e1000_phy_info *phy = &adapter->hw.phy;
4441         struct e1000_ring *tx_ring = adapter->tx_ring;
4442         struct e1000_hw *hw = &adapter->hw;
4443         u32 link, tctl;
4444
4445         if (test_bit(__E1000_DOWN, &adapter->state))
4446                 return;
4447
4448         link = e1000e_has_link(adapter);
4449         if ((netif_carrier_ok(netdev)) && link) {
4450                 /* Cancel scheduled suspend requests. */
4451                 pm_runtime_resume(netdev->dev.parent);
4452
4453                 e1000e_enable_receives(adapter);
4454                 goto link_up;
4455         }
4456
4457         if ((e1000e_enable_tx_pkt_filtering(hw)) &&
4458             (adapter->mng_vlan_id != adapter->hw.mng_cookie.vlan_id))
4459                 e1000_update_mng_vlan(adapter);
4460
4461         if (link) {
4462                 if (!netif_carrier_ok(netdev)) {
4463                         bool txb2b = true;
4464
4465                         /* Cancel scheduled suspend requests. */
4466                         pm_runtime_resume(netdev->dev.parent);
4467
4468                         /* update snapshot of PHY registers on LSC */
4469                         e1000_phy_read_status(adapter);
4470                         mac->ops.get_link_up_info(&adapter->hw,
4471                                                    &adapter->link_speed,
4472                                                    &adapter->link_duplex);
4473                         e1000_print_link_info(adapter);
4474                         /*
4475                          * On supported PHYs, check for duplex mismatch only
4476                          * if link has autonegotiated at 10/100 half
4477                          */
4478                         if ((hw->phy.type == e1000_phy_igp_3 ||
4479                              hw->phy.type == e1000_phy_bm) &&
4480                             (hw->mac.autoneg == true) &&
4481                             (adapter->link_speed == SPEED_10 ||
4482                              adapter->link_speed == SPEED_100) &&
4483                             (adapter->link_duplex == HALF_DUPLEX)) {
4484                                 u16 autoneg_exp;
4485
4486                                 e1e_rphy(hw, PHY_AUTONEG_EXP, &autoneg_exp);
4487
4488                                 if (!(autoneg_exp & NWAY_ER_LP_NWAY_CAPS))
4489                                         e_info("Autonegotiated half duplex but link partner cannot autoneg.  Try forcing full duplex if link gets many collisions.\n");
4490                         }
4491
4492                         /* adjust timeout factor according to speed/duplex */
4493                         adapter->tx_timeout_factor = 1;
4494                         switch (adapter->link_speed) {
4495                         case SPEED_10:
4496                                 txb2b = false;
4497                                 adapter->tx_timeout_factor = 16;
4498                                 break;
4499                         case SPEED_100:
4500                                 txb2b = false;
4501                                 adapter->tx_timeout_factor = 10;
4502                                 break;
4503                         }
4504
4505                         /*
4506                          * workaround: re-program speed mode bit after
4507                          * link-up event
4508                          */
4509                         if ((adapter->flags & FLAG_TARC_SPEED_MODE_BIT) &&
4510                             !txb2b) {
4511                                 u32 tarc0;
4512                                 tarc0 = er32(TARC(0));
4513                                 tarc0 &= ~SPEED_MODE_BIT;
4514                                 ew32(TARC(0), tarc0);
4515                         }
4516
4517                         /*
4518                          * disable TSO for pcie and 10/100 speeds, to avoid
4519                          * some hardware issues
4520                          */
4521                         if (!(adapter->flags & FLAG_TSO_FORCE)) {
4522                                 switch (adapter->link_speed) {
4523                                 case SPEED_10:
4524                                 case SPEED_100:
4525                                         e_info("10/100 speed: disabling TSO\n");
4526                                         netdev->features &= ~NETIF_F_TSO;
4527                                         netdev->features &= ~NETIF_F_TSO6;
4528                                         break;
4529                                 case SPEED_1000:
4530                                         netdev->features |= NETIF_F_TSO;
4531                                         netdev->features |= NETIF_F_TSO6;
4532                                         break;
4533                                 default:
4534                                         /* oops */
4535                                         break;
4536                                 }
4537                         }
4538
4539                         /*
4540                          * enable transmits in the hardware, need to do this
4541                          * after setting TARC(0)
4542                          */
4543                         tctl = er32(TCTL);
4544                         tctl |= E1000_TCTL_EN;
4545                         ew32(TCTL, tctl);
4546
4547                         /*
4548                          * Perform any post-link-up configuration before
4549                          * reporting link up.
4550                          */
4551                         if (phy->ops.cfg_on_link_up)
4552                                 phy->ops.cfg_on_link_up(hw);
4553
4554                         netif_carrier_on(netdev);
4555
4556                         if (!test_bit(__E1000_DOWN, &adapter->state))
4557                                 mod_timer(&adapter->phy_info_timer,
4558                                           round_jiffies(jiffies + 2 * HZ));
4559                 }
4560         } else {
4561                 if (netif_carrier_ok(netdev)) {
4562                         adapter->link_speed = 0;
4563                         adapter->link_duplex = 0;
4564                         /* Link status message must follow this format */
4565                         printk(KERN_INFO "e1000e: %s NIC Link is Down\n",
4566                                adapter->netdev->name);
4567                         netif_carrier_off(netdev);
4568                         if (!test_bit(__E1000_DOWN, &adapter->state))
4569                                 mod_timer(&adapter->phy_info_timer,
4570                                           round_jiffies(jiffies + 2 * HZ));
4571
4572                         if (adapter->flags & FLAG_RX_NEEDS_RESTART)
4573                                 schedule_work(&adapter->reset_task);
4574                         else
4575                                 pm_schedule_suspend(netdev->dev.parent,
4576                                                         LINK_TIMEOUT);
4577                 }
4578         }
4579
4580 link_up:
4581         spin_lock(&adapter->stats64_lock);
4582         e1000e_update_stats(adapter);
4583
4584         mac->tx_packet_delta = adapter->stats.tpt - adapter->tpt_old;
4585         adapter->tpt_old = adapter->stats.tpt;
4586         mac->collision_delta = adapter->stats.colc - adapter->colc_old;
4587         adapter->colc_old = adapter->stats.colc;
4588
4589         adapter->gorc = adapter->stats.gorc - adapter->gorc_old;
4590         adapter->gorc_old = adapter->stats.gorc;
4591         adapter->gotc = adapter->stats.gotc - adapter->gotc_old;
4592         adapter->gotc_old = adapter->stats.gotc;
4593         spin_unlock(&adapter->stats64_lock);
4594
4595         e1000e_update_adaptive(&adapter->hw);
4596
4597         if (!netif_carrier_ok(netdev) &&
4598             (e1000_desc_unused(tx_ring) + 1 < tx_ring->count)) {
4599                 /*
4600                  * We've lost link, so the controller stops DMA,
4601                  * but we've got queued Tx work that's never going
4602                  * to get done, so reset controller to flush Tx.
4603                  * (Do the reset outside of interrupt context).
4604                  */
4605                 schedule_work(&adapter->reset_task);
4606                 /* return immediately since reset is imminent */
4607                 return;
4608         }
4609
4610         /* Simple mode for Interrupt Throttle Rate (ITR) */
4611         if (adapter->itr_setting == 4) {
4612                 /*
4613                  * Symmetric Tx/Rx gets a reduced ITR=2000;
4614                  * Total asymmetrical Tx or Rx gets ITR=8000;
4615                  * everyone else is between 2000-8000.
4616                  */
4617                 u32 goc = (adapter->gotc + adapter->gorc) / 10000;
4618                 u32 dif = (adapter->gotc > adapter->gorc ?
4619                             adapter->gotc - adapter->gorc :
4620                             adapter->gorc - adapter->gotc) / 10000;
4621                 u32 itr = goc > 0 ? (dif * 6000 / goc + 2000) : 8000;
4622
4623                 ew32(ITR, 1000000000 / (itr * 256));
4624         }
4625
4626         /* Cause software interrupt to ensure Rx ring is cleaned */
4627         if (adapter->msix_entries)
4628                 ew32(ICS, adapter->rx_ring->ims_val);
4629         else
4630                 ew32(ICS, E1000_ICS_RXDMT0);
4631
4632         /* flush pending descriptors to memory before detecting Tx hang */
4633         e1000e_flush_descriptors(adapter);
4634
4635         /* Force detection of hung controller every watchdog period */
4636         adapter->detect_tx_hung = true;
4637
4638         /*
4639          * With 82571 controllers, LAA may be overwritten due to controller
4640          * reset from the other port. Set the appropriate LAA in RAR[0]
4641          */
4642         if (e1000e_get_laa_state_82571(hw))
4643                 e1000e_rar_set(hw, adapter->hw.mac.addr, 0);
4644
4645         if (adapter->flags2 & FLAG2_CHECK_PHY_HANG)
4646                 e1000e_check_82574_phy_workaround(adapter);
4647
4648         /* Reset the timer */
4649         if (!test_bit(__E1000_DOWN, &adapter->state))
4650                 mod_timer(&adapter->watchdog_timer,
4651                           round_jiffies(jiffies + 2 * HZ));
4652 }
4653
4654 #define E1000_TX_FLAGS_CSUM             0x00000001
4655 #define E1000_TX_FLAGS_VLAN             0x00000002
4656 #define E1000_TX_FLAGS_TSO              0x00000004
4657 #define E1000_TX_FLAGS_IPV4             0x00000008
4658 #define E1000_TX_FLAGS_NO_FCS           0x00000010
4659 #define E1000_TX_FLAGS_VLAN_MASK        0xffff0000
4660 #define E1000_TX_FLAGS_VLAN_SHIFT       16
4661
4662 static int e1000_tso(struct e1000_ring *tx_ring, struct sk_buff *skb)
4663 {
4664         struct e1000_context_desc *context_desc;
4665         struct e1000_buffer *buffer_info;
4666         unsigned int i;
4667         u32 cmd_length = 0;
4668         u16 ipcse = 0, tucse, mss;
4669         u8 ipcss, ipcso, tucss, tucso, hdr_len;
4670
4671         if (!skb_is_gso(skb))
4672                 return 0;
4673
4674         if (skb_header_cloned(skb)) {
4675                 int err = pskb_expand_head(skb, 0, 0, GFP_ATOMIC);
4676
4677                 if (err)
4678                         return err;
4679         }
4680
4681         hdr_len = skb_transport_offset(skb) + tcp_hdrlen(skb);
4682         mss = skb_shinfo(skb)->gso_size;
4683         if (skb->protocol == htons(ETH_P_IP)) {
4684                 struct iphdr *iph = ip_hdr(skb);
4685                 iph->tot_len = 0;
4686                 iph->check = 0;
4687                 tcp_hdr(skb)->check = ~csum_tcpudp_magic(iph->saddr, iph->daddr,
4688                                                          0, IPPROTO_TCP, 0);
4689                 cmd_length = E1000_TXD_CMD_IP;
4690                 ipcse = skb_transport_offset(skb) - 1;
4691         } else if (skb_is_gso_v6(skb)) {
4692                 ipv6_hdr(skb)->payload_len = 0;
4693                 tcp_hdr(skb)->check = ~csum_ipv6_magic(&ipv6_hdr(skb)->saddr,
4694                                                        &ipv6_hdr(skb)->daddr,
4695                                                        0, IPPROTO_TCP, 0);
4696                 ipcse = 0;
4697         }
4698         ipcss = skb_network_offset(skb);
4699         ipcso = (void *)&(ip_hdr(skb)->check) - (void *)skb->data;
4700         tucss = skb_transport_offset(skb);
4701         tucso = (void *)&(tcp_hdr(skb)->check) - (void *)skb->data;
4702         tucse = 0;
4703
4704         cmd_length |= (E1000_TXD_CMD_DEXT | E1000_TXD_CMD_TSE |
4705                        E1000_TXD_CMD_TCP | (skb->len - (hdr_len)));
4706
4707         i = tx_ring->next_to_use;
4708         context_desc = E1000_CONTEXT_DESC(*tx_ring, i);
4709         buffer_info = &tx_ring->buffer_info[i];
4710
4711         context_desc->lower_setup.ip_fields.ipcss  = ipcss;
4712         context_desc->lower_setup.ip_fields.ipcso  = ipcso;
4713         context_desc->lower_setup.ip_fields.ipcse  = cpu_to_le16(ipcse);
4714         context_desc->upper_setup.tcp_fields.tucss = tucss;
4715         context_desc->upper_setup.tcp_fields.tucso = tucso;
4716         context_desc->upper_setup.tcp_fields.tucse = cpu_to_le16(tucse);
4717         context_desc->tcp_seg_setup.fields.mss     = cpu_to_le16(mss);
4718         context_desc->tcp_seg_setup.fields.hdr_len = hdr_len;
4719         context_desc->cmd_and_length = cpu_to_le32(cmd_length);
4720
4721         buffer_info->time_stamp = jiffies;
4722         buffer_info->next_to_watch = i;
4723
4724         i++;
4725         if (i == tx_ring->count)
4726                 i = 0;
4727         tx_ring->next_to_use = i;
4728
4729         return 1;
4730 }
4731
4732 static bool e1000_tx_csum(struct e1000_ring *tx_ring, struct sk_buff *skb)
4733 {
4734         struct e1000_adapter *adapter = tx_ring->adapter;
4735         struct e1000_context_desc *context_desc;
4736         struct e1000_buffer *buffer_info;
4737         unsigned int i;
4738         u8 css;
4739         u32 cmd_len = E1000_TXD_CMD_DEXT;
4740         __be16 protocol;
4741
4742         if (skb->ip_summed != CHECKSUM_PARTIAL)
4743                 return 0;
4744
4745         if (skb->protocol == cpu_to_be16(ETH_P_8021Q))
4746                 protocol = vlan_eth_hdr(skb)->h_vlan_encapsulated_proto;
4747         else
4748                 protocol = skb->protocol;
4749
4750         switch (protocol) {
4751         case cpu_to_be16(ETH_P_IP):
4752                 if (ip_hdr(skb)->protocol == IPPROTO_TCP)
4753                         cmd_len |= E1000_TXD_CMD_TCP;
4754                 break;
4755         case cpu_to_be16(ETH_P_IPV6):
4756                 /* XXX not handling all IPV6 headers */
4757                 if (ipv6_hdr(skb)->nexthdr == IPPROTO_TCP)
4758                         cmd_len |= E1000_TXD_CMD_TCP;
4759                 break;
4760         default:
4761                 if (unlikely(net_ratelimit()))
4762                         e_warn("checksum_partial proto=%x!\n",
4763                                be16_to_cpu(protocol));
4764                 break;
4765         }
4766
4767         css = skb_checksum_start_offset(skb);
4768
4769         i = tx_ring->next_to_use;
4770         buffer_info = &tx_ring->buffer_info[i];
4771         context_desc = E1000_CONTEXT_DESC(*tx_ring, i);
4772
4773         context_desc->lower_setup.ip_config = 0;
4774         context_desc->upper_setup.tcp_fields.tucss = css;
4775         context_desc->upper_setup.tcp_fields.tucso =
4776                                 css + skb->csum_offset;
4777         context_desc->upper_setup.tcp_fields.tucse = 0;
4778         context_desc->tcp_seg_setup.data = 0;
4779         context_desc->cmd_and_length = cpu_to_le32(cmd_len);
4780
4781         buffer_info->time_stamp = jiffies;
4782         buffer_info->next_to_watch = i;
4783
4784         i++;
4785         if (i == tx_ring->count)
4786                 i = 0;
4787         tx_ring->next_to_use = i;
4788
4789         return 1;
4790 }
4791
4792 #define E1000_MAX_PER_TXD       8192
4793 #define E1000_MAX_TXD_PWR       12
4794
4795 static int e1000_tx_map(struct e1000_ring *tx_ring, struct sk_buff *skb,
4796                         unsigned int first, unsigned int max_per_txd,
4797                         unsigned int nr_frags, unsigned int mss)
4798 {
4799         struct e1000_adapter *adapter = tx_ring->adapter;
4800         struct pci_dev *pdev = adapter->pdev;
4801         struct e1000_buffer *buffer_info;
4802         unsigned int len = skb_headlen(skb);
4803         unsigned int offset = 0, size, count = 0, i;
4804         unsigned int f, bytecount, segs;
4805
4806         i = tx_ring->next_to_use;
4807
4808         while (len) {
4809                 buffer_info = &tx_ring->buffer_info[i];
4810                 size = min(len, max_per_txd);
4811
4812                 buffer_info->length = size;
4813                 buffer_info->time_stamp = jiffies;
4814                 buffer_info->next_to_watch = i;
4815                 buffer_info->dma = dma_map_single(&pdev->dev,
4816                                                   skb->data + offset,
4817                                                   size, DMA_TO_DEVICE);
4818                 buffer_info->mapped_as_page = false;
4819                 if (dma_mapping_error(&pdev->dev, buffer_info->dma))
4820                         goto dma_error;
4821
4822                 len -= size;
4823                 offset += size;
4824                 count++;
4825
4826                 if (len) {
4827                         i++;
4828                         if (i == tx_ring->count)
4829                                 i = 0;
4830                 }
4831         }
4832
4833         for (f = 0; f < nr_frags; f++) {
4834                 const struct skb_frag_struct *frag;
4835
4836                 frag = &skb_shinfo(skb)->frags[f];
4837                 len = skb_frag_size(frag);
4838                 offset = 0;
4839
4840                 while (len) {
4841                         i++;
4842                         if (i == tx_ring->count)
4843                                 i = 0;
4844
4845                         buffer_info = &tx_ring->buffer_info[i];
4846                         size = min(len, max_per_txd);
4847
4848                         buffer_info->length = size;
4849                         buffer_info->time_stamp = jiffies;
4850                         buffer_info->next_to_watch = i;
4851                         buffer_info->dma = skb_frag_dma_map(&pdev->dev, frag,
4852                                                 offset, size, DMA_TO_DEVICE);
4853                         buffer_info->mapped_as_page = true;
4854                         if (dma_mapping_error(&pdev->dev, buffer_info->dma))
4855                                 goto dma_error;
4856
4857                         len -= size;
4858                         offset += size;
4859                         count++;
4860                 }
4861         }
4862
4863         segs = skb_shinfo(skb)->gso_segs ? : 1;
4864         /* multiply data chunks by size of headers */
4865         bytecount = ((segs - 1) * skb_headlen(skb)) + skb->len;
4866
4867         tx_ring->buffer_info[i].skb = skb;
4868         tx_ring->buffer_info[i].segs = segs;
4869         tx_ring->buffer_info[i].bytecount = bytecount;
4870         tx_ring->buffer_info[first].next_to_watch = i;
4871
4872         return count;
4873
4874 dma_error:
4875         dev_err(&pdev->dev, "Tx DMA map failed\n");
4876         buffer_info->dma = 0;
4877         if (count)
4878                 count--;
4879
4880         while (count--) {
4881                 if (i == 0)
4882                         i += tx_ring->count;
4883                 i--;
4884                 buffer_info = &tx_ring->buffer_info[i];
4885                 e1000_put_txbuf(tx_ring, buffer_info);
4886         }
4887
4888         return 0;
4889 }
4890
4891 static void e1000_tx_queue(struct e1000_ring *tx_ring, int tx_flags, int count)
4892 {
4893         struct e1000_adapter *adapter = tx_ring->adapter;
4894         struct e1000_tx_desc *tx_desc = NULL;
4895         struct e1000_buffer *buffer_info;
4896         u32 txd_upper = 0, txd_lower = E1000_TXD_CMD_IFCS;
4897         unsigned int i;
4898
4899         if (tx_flags & E1000_TX_FLAGS_TSO) {
4900                 txd_lower |= E1000_TXD_CMD_DEXT | E1000_TXD_DTYP_D |
4901                              E1000_TXD_CMD_TSE;
4902                 txd_upper |= E1000_TXD_POPTS_TXSM << 8;
4903
4904                 if (tx_flags & E1000_TX_FLAGS_IPV4)
4905                         txd_upper |= E1000_TXD_POPTS_IXSM << 8;
4906         }
4907
4908         if (tx_flags & E1000_TX_FLAGS_CSUM) {
4909                 txd_lower |= E1000_TXD_CMD_DEXT | E1000_TXD_DTYP_D;
4910                 txd_upper |= E1000_TXD_POPTS_TXSM << 8;
4911         }
4912
4913         if (tx_flags & E1000_TX_FLAGS_VLAN) {
4914                 txd_lower |= E1000_TXD_CMD_VLE;
4915                 txd_upper |= (tx_flags & E1000_TX_FLAGS_VLAN_MASK);
4916         }
4917
4918         if (unlikely(tx_flags & E1000_TX_FLAGS_NO_FCS))
4919                 txd_lower &= ~(E1000_TXD_CMD_IFCS);
4920
4921         i = tx_ring->next_to_use;
4922
4923         do {
4924                 buffer_info = &tx_ring->buffer_info[i];
4925                 tx_desc = E1000_TX_DESC(*tx_ring, i);
4926                 tx_desc->buffer_addr = cpu_to_le64(buffer_info->dma);
4927                 tx_desc->lower.data =
4928                         cpu_to_le32(txd_lower | buffer_info->length);
4929                 tx_desc->upper.data = cpu_to_le32(txd_upper);
4930
4931                 i++;
4932                 if (i == tx_ring->count)
4933                         i = 0;
4934         } while (--count > 0);
4935
4936         tx_desc->lower.data |= cpu_to_le32(adapter->txd_cmd);
4937
4938         /* txd_cmd re-enables FCS, so we'll re-disable it here as desired. */
4939         if (unlikely(tx_flags & E1000_TX_FLAGS_NO_FCS))
4940                 tx_desc->lower.data &= ~(cpu_to_le32(E1000_TXD_CMD_IFCS));
4941
4942         /*
4943          * Force memory writes to complete before letting h/w
4944          * know there are new descriptors to fetch.  (Only
4945          * applicable for weak-ordered memory model archs,
4946          * such as IA-64).
4947          */
4948         wmb();
4949
4950         tx_ring->next_to_use = i;
4951
4952         if (adapter->flags2 & FLAG2_PCIM2PCI_ARBITER_WA)
4953                 e1000e_update_tdt_wa(tx_ring, i);
4954         else
4955                 writel(i, tx_ring->tail);
4956
4957         /*
4958          * we need this if more than one processor can write to our tail
4959          * at a time, it synchronizes IO on IA64/Altix systems
4960          */
4961         mmiowb();
4962 }
4963
4964 #define MINIMUM_DHCP_PACKET_SIZE 282
4965 static int e1000_transfer_dhcp_info(struct e1000_adapter *adapter,
4966                                     struct sk_buff *skb)
4967 {
4968         struct e1000_hw *hw =  &adapter->hw;
4969         u16 length, offset;
4970
4971         if (vlan_tx_tag_present(skb)) {
4972                 if (!((vlan_tx_tag_get(skb) == adapter->hw.mng_cookie.vlan_id) &&
4973                     (adapter->hw.mng_cookie.status &
4974                         E1000_MNG_DHCP_COOKIE_STATUS_VLAN)))
4975                         return 0;
4976         }
4977
4978         if (skb->len <= MINIMUM_DHCP_PACKET_SIZE)
4979                 return 0;
4980
4981         if (((struct ethhdr *) skb->data)->h_proto != htons(ETH_P_IP))
4982                 return 0;
4983
4984         {
4985                 const struct iphdr *ip = (struct iphdr *)((u8 *)skb->data+14);
4986                 struct udphdr *udp;
4987
4988                 if (ip->protocol != IPPROTO_UDP)
4989                         return 0;
4990
4991                 udp = (struct udphdr *)((u8 *)ip + (ip->ihl << 2));
4992                 if (ntohs(udp->dest) != 67)
4993                         return 0;
4994
4995                 offset = (u8 *)udp + 8 - skb->data;
4996                 length = skb->len - offset;
4997                 return e1000e_mng_write_dhcp_info(hw, (u8 *)udp + 8, length);
4998         }
4999
5000         return 0;
5001 }
5002
5003 static int __e1000_maybe_stop_tx(struct e1000_ring *tx_ring, int size)
5004 {
5005         struct e1000_adapter *adapter = tx_ring->adapter;
5006
5007         netif_stop_queue(adapter->netdev);
5008         /*
5009          * Herbert's original patch had:
5010          *  smp_mb__after_netif_stop_queue();
5011          * but since that doesn't exist yet, just open code it.
5012          */
5013         smp_mb();
5014
5015         /*
5016          * We need to check again in a case another CPU has just
5017          * made room available.
5018          */
5019         if (e1000_desc_unused(tx_ring) < size)
5020                 return -EBUSY;
5021
5022         /* A reprieve! */
5023         netif_start_queue(adapter->netdev);
5024         ++adapter->restart_queue;
5025         return 0;
5026 }
5027
5028 static int e1000_maybe_stop_tx(struct e1000_ring *tx_ring, int size)
5029 {
5030         if (e1000_desc_unused(tx_ring) >= size)
5031                 return 0;
5032         return __e1000_maybe_stop_tx(tx_ring, size);
5033 }
5034
5035 #define TXD_USE_COUNT(S, X) (((S) >> (X)) + 1)
5036 static netdev_tx_t e1000_xmit_frame(struct sk_buff *skb,
5037                                     struct net_device *netdev)
5038 {
5039         struct e1000_adapter *adapter = netdev_priv(netdev);
5040         struct e1000_ring *tx_ring = adapter->tx_ring;
5041         unsigned int first;
5042         unsigned int max_per_txd = E1000_MAX_PER_TXD;
5043         unsigned int max_txd_pwr = E1000_MAX_TXD_PWR;
5044         unsigned int tx_flags = 0;
5045         unsigned int len = skb_headlen(skb);
5046         unsigned int nr_frags;
5047         unsigned int mss;
5048         int count = 0;
5049         int tso;
5050         unsigned int f;
5051
5052         if (test_bit(__E1000_DOWN, &adapter->state)) {
5053                 dev_kfree_skb_any(skb);
5054                 return NETDEV_TX_OK;
5055         }
5056
5057         if (skb->len <= 0) {
5058                 dev_kfree_skb_any(skb);
5059                 return NETDEV_TX_OK;
5060         }
5061
5062         mss = skb_shinfo(skb)->gso_size;
5063         /*
5064          * The controller does a simple calculation to
5065          * make sure there is enough room in the FIFO before
5066          * initiating the DMA for each buffer.  The calc is:
5067          * 4 = ceil(buffer len/mss).  To make sure we don't
5068          * overrun the FIFO, adjust the max buffer len if mss
5069          * drops.
5070          */
5071         if (mss) {
5072                 u8 hdr_len;
5073                 max_per_txd = min(mss << 2, max_per_txd);
5074                 max_txd_pwr = fls(max_per_txd) - 1;
5075
5076                 /*
5077                  * TSO Workaround for 82571/2/3 Controllers -- if skb->data
5078                  * points to just header, pull a few bytes of payload from
5079                  * frags into skb->data
5080                  */
5081                 hdr_len = skb_transport_offset(skb) + tcp_hdrlen(skb);
5082                 /*
5083                  * we do this workaround for ES2LAN, but it is un-necessary,
5084                  * avoiding it could save a lot of cycles
5085                  */
5086                 if (skb->data_len && (hdr_len == len)) {
5087                         unsigned int pull_size;
5088
5089                         pull_size = min_t(unsigned int, 4, skb->data_len);
5090                         if (!__pskb_pull_tail(skb, pull_size)) {
5091                                 e_err("__pskb_pull_tail failed.\n");
5092                                 dev_kfree_skb_any(skb);
5093                                 return NETDEV_TX_OK;
5094                         }
5095                         len = skb_headlen(skb);
5096                 }
5097         }
5098
5099         /* reserve a descriptor for the offload context */
5100         if ((mss) || (skb->ip_summed == CHECKSUM_PARTIAL))
5101                 count++;
5102         count++;
5103
5104         count += TXD_USE_COUNT(len, max_txd_pwr);
5105
5106         nr_frags = skb_shinfo(skb)->nr_frags;
5107         for (f = 0; f < nr_frags; f++)
5108                 count += TXD_USE_COUNT(skb_frag_size(&skb_shinfo(skb)->frags[f]),
5109                                        max_txd_pwr);
5110
5111         if (adapter->hw.mac.tx_pkt_filtering)
5112                 e1000_transfer_dhcp_info(adapter, skb);
5113
5114         /*
5115          * need: count + 2 desc gap to keep tail from touching
5116          * head, otherwise try next time
5117          */
5118         if (e1000_maybe_stop_tx(tx_ring, count + 2))
5119                 return NETDEV_TX_BUSY;
5120
5121         if (vlan_tx_tag_present(skb)) {
5122                 tx_flags |= E1000_TX_FLAGS_VLAN;
5123                 tx_flags |= (vlan_tx_tag_get(skb) << E1000_TX_FLAGS_VLAN_SHIFT);
5124         }
5125
5126         first = tx_ring->next_to_use;
5127
5128         tso = e1000_tso(tx_ring, skb);
5129         if (tso < 0) {
5130                 dev_kfree_skb_any(skb);
5131                 return NETDEV_TX_OK;
5132         }
5133
5134         if (tso)
5135                 tx_flags |= E1000_TX_FLAGS_TSO;
5136         else if (e1000_tx_csum(tx_ring, skb))
5137                 tx_flags |= E1000_TX_FLAGS_CSUM;
5138
5139         /*
5140          * Old method was to assume IPv4 packet by default if TSO was enabled.
5141          * 82571 hardware supports TSO capabilities for IPv6 as well...
5142          * no longer assume, we must.
5143          */
5144         if (skb->protocol == htons(ETH_P_IP))
5145                 tx_flags |= E1000_TX_FLAGS_IPV4;
5146
5147         if (unlikely(skb->no_fcs))
5148                 tx_flags |= E1000_TX_FLAGS_NO_FCS;
5149
5150         /* if count is 0 then mapping error has occurred */
5151         count = e1000_tx_map(tx_ring, skb, first, max_per_txd, nr_frags, mss);
5152         if (count) {
5153                 netdev_sent_queue(netdev, skb->len);
5154                 e1000_tx_queue(tx_ring, tx_flags, count);
5155                 /* Make sure there is space in the ring for the next send. */
5156                 e1000_maybe_stop_tx(tx_ring, MAX_SKB_FRAGS + 2);
5157
5158         } else {
5159                 dev_kfree_skb_any(skb);
5160                 tx_ring->buffer_info[first].time_stamp = 0;
5161                 tx_ring->next_to_use = first;
5162         }
5163
5164         return NETDEV_TX_OK;
5165 }
5166
5167 /**
5168  * e1000_tx_timeout - Respond to a Tx Hang
5169  * @netdev: network interface device structure
5170  **/
5171 static void e1000_tx_timeout(struct net_device *netdev)
5172 {
5173         struct e1000_adapter *adapter = netdev_priv(netdev);
5174
5175         /* Do the reset outside of interrupt context */
5176         adapter->tx_timeout_count++;
5177         schedule_work(&adapter->reset_task);
5178 }
5179
5180 static void e1000_reset_task(struct work_struct *work)
5181 {
5182         struct e1000_adapter *adapter;
5183         adapter = container_of(work, struct e1000_adapter, reset_task);
5184
5185         /* don't run the task if already down */
5186         if (test_bit(__E1000_DOWN, &adapter->state))
5187                 return;
5188
5189         if (!((adapter->flags & FLAG_RX_NEEDS_RESTART) &&
5190               (adapter->flags & FLAG_RX_RESTART_NOW))) {
5191                 e1000e_dump(adapter);
5192                 e_err("Reset adapter\n");
5193         }
5194         e1000e_reinit_locked(adapter);
5195 }
5196
5197 /**
5198  * e1000_get_stats64 - Get System Network Statistics
5199  * @netdev: network interface device structure
5200  * @stats: rtnl_link_stats64 pointer
5201  *
5202  * Returns the address of the device statistics structure.
5203  **/
5204 struct rtnl_link_stats64 *e1000e_get_stats64(struct net_device *netdev,
5205                                              struct rtnl_link_stats64 *stats)
5206 {
5207         struct e1000_adapter *adapter = netdev_priv(netdev);
5208
5209         memset(stats, 0, sizeof(struct rtnl_link_stats64));
5210         spin_lock(&adapter->stats64_lock);
5211         e1000e_update_stats(adapter);
5212         /* Fill out the OS statistics structure */
5213         stats->rx_bytes = adapter->stats.gorc;
5214         stats->rx_packets = adapter->stats.gprc;
5215         stats->tx_bytes = adapter->stats.gotc;
5216         stats->tx_packets = adapter->stats.gptc;
5217         stats->multicast = adapter->stats.mprc;
5218         stats->collisions = adapter->stats.colc;
5219
5220         /* Rx Errors */
5221
5222         /*
5223          * RLEC on some newer hardware can be incorrect so build
5224          * our own version based on RUC and ROC
5225          */
5226         stats->rx_errors = adapter->stats.rxerrc +
5227                 adapter->stats.crcerrs + adapter->stats.algnerrc +
5228                 adapter->stats.ruc + adapter->stats.roc +
5229                 adapter->stats.cexterr;
5230         stats->rx_length_errors = adapter->stats.ruc +
5231                                               adapter->stats.roc;
5232         stats->rx_crc_errors = adapter->stats.crcerrs;
5233         stats->rx_frame_errors = adapter->stats.algnerrc;
5234         stats->rx_missed_errors = adapter->stats.mpc;
5235
5236         /* Tx Errors */
5237         stats->tx_errors = adapter->stats.ecol +
5238                                        adapter->stats.latecol;
5239         stats->tx_aborted_errors = adapter->stats.ecol;
5240         stats->tx_window_errors = adapter->stats.latecol;
5241         stats->tx_carrier_errors = adapter->stats.tncrs;
5242
5243         /* Tx Dropped needs to be maintained elsewhere */
5244
5245         spin_unlock(&adapter->stats64_lock);
5246         return stats;
5247 }
5248
5249 /**
5250  * e1000_change_mtu - Change the Maximum Transfer Unit
5251  * @netdev: network interface device structure
5252  * @new_mtu: new value for maximum frame size
5253  *
5254  * Returns 0 on success, negative on failure
5255  **/
5256 static int e1000_change_mtu(struct net_device *netdev, int new_mtu)
5257 {
5258         struct e1000_adapter *adapter = netdev_priv(netdev);
5259         int max_frame = new_mtu + ETH_HLEN + ETH_FCS_LEN;
5260
5261         /* Jumbo frame support */
5262         if (max_frame > ETH_FRAME_LEN + ETH_FCS_LEN) {
5263                 if (!(adapter->flags & FLAG_HAS_JUMBO_FRAMES)) {
5264                         e_err("Jumbo Frames not supported.\n");
5265                         return -EINVAL;
5266                 }
5267
5268                 /*
5269                  * IP payload checksum (enabled with jumbos/packet-split when
5270                  * Rx checksum is enabled) and generation of RSS hash is
5271                  * mutually exclusive in the hardware.
5272                  */
5273                 if ((netdev->features & NETIF_F_RXCSUM) &&
5274                     (netdev->features & NETIF_F_RXHASH)) {
5275                         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");
5276                         return -EINVAL;
5277                 }
5278         }
5279
5280         /* Supported frame sizes */
5281         if ((new_mtu < ETH_ZLEN + ETH_FCS_LEN + VLAN_HLEN) ||
5282             (max_frame > adapter->max_hw_frame_size)) {
5283                 e_err("Unsupported MTU setting\n");
5284                 return -EINVAL;
5285         }
5286
5287         /* Jumbo frame workaround on 82579 requires CRC be stripped */
5288         if ((adapter->hw.mac.type == e1000_pch2lan) &&
5289             !(adapter->flags2 & FLAG2_CRC_STRIPPING) &&
5290             (new_mtu > ETH_DATA_LEN)) {
5291                 e_err("Jumbo Frames not supported on 82579 when CRC stripping is disabled.\n");
5292                 return -EINVAL;
5293         }
5294
5295         /* 82573 Errata 17 */
5296         if (((adapter->hw.mac.type == e1000_82573) ||
5297              (adapter->hw.mac.type == e1000_82574)) &&
5298             (max_frame > ETH_FRAME_LEN + ETH_FCS_LEN)) {
5299                 adapter->flags2 |= FLAG2_DISABLE_ASPM_L1;
5300                 e1000e_disable_aspm(adapter->pdev, PCIE_LINK_STATE_L1);
5301         }
5302
5303         while (test_and_set_bit(__E1000_RESETTING, &adapter->state))
5304                 usleep_range(1000, 2000);
5305         /* e1000e_down -> e1000e_reset dependent on max_frame_size & mtu */
5306         adapter->max_frame_size = max_frame;
5307         e_info("changing MTU from %d to %d\n", netdev->mtu, new_mtu);
5308         netdev->mtu = new_mtu;
5309         if (netif_running(netdev))
5310                 e1000e_down(adapter);
5311
5312         /*
5313          * NOTE: netdev_alloc_skb reserves 16 bytes, and typically NET_IP_ALIGN
5314          * means we reserve 2 more, this pushes us to allocate from the next
5315          * larger slab size.
5316          * i.e. RXBUFFER_2048 --> size-4096 slab
5317          * However with the new *_jumbo_rx* routines, jumbo receives will use
5318          * fragmented skbs
5319          */
5320
5321         if (max_frame <= 2048)
5322                 adapter->rx_buffer_len = 2048;
5323         else
5324                 adapter->rx_buffer_len = 4096;
5325
5326         /* adjust allocation if LPE protects us, and we aren't using SBP */
5327         if ((max_frame == ETH_FRAME_LEN + ETH_FCS_LEN) ||
5328              (max_frame == ETH_FRAME_LEN + VLAN_HLEN + ETH_FCS_LEN))
5329                 adapter->rx_buffer_len = ETH_FRAME_LEN + VLAN_HLEN
5330                                          + ETH_FCS_LEN;
5331
5332         if (netif_running(netdev))
5333                 e1000e_up(adapter);
5334         else
5335                 e1000e_reset(adapter);
5336
5337         clear_bit(__E1000_RESETTING, &adapter->state);
5338
5339         return 0;
5340 }
5341
5342 static int e1000_mii_ioctl(struct net_device *netdev, struct ifreq *ifr,
5343                            int cmd)
5344 {
5345         struct e1000_adapter *adapter = netdev_priv(netdev);
5346         struct mii_ioctl_data *data = if_mii(ifr);
5347
5348         if (adapter->hw.phy.media_type != e1000_media_type_copper)
5349                 return -EOPNOTSUPP;
5350
5351         switch (cmd) {
5352         case SIOCGMIIPHY:
5353                 data->phy_id = adapter->hw.phy.addr;
5354                 break;
5355         case SIOCGMIIREG:
5356                 e1000_phy_read_status(adapter);
5357
5358                 switch (data->reg_num & 0x1F) {
5359                 case MII_BMCR:
5360                         data->val_out = adapter->phy_regs.bmcr;
5361                         break;
5362                 case MII_BMSR:
5363                         data->val_out = adapter->phy_regs.bmsr;
5364                         break;
5365                 case MII_PHYSID1:
5366                         data->val_out = (adapter->hw.phy.id >> 16);
5367                         break;
5368                 case MII_PHYSID2:
5369                         data->val_out = (adapter->hw.phy.id & 0xFFFF);
5370                         break;
5371                 case MII_ADVERTISE:
5372                         data->val_out = adapter->phy_regs.advertise;
5373                         break;
5374                 case MII_LPA:
5375                         data->val_out = adapter->phy_regs.lpa;
5376                         break;
5377                 case MII_EXPANSION:
5378                         data->val_out = adapter->phy_regs.expansion;
5379                         break;
5380                 case MII_CTRL1000:
5381                         data->val_out = adapter->phy_regs.ctrl1000;
5382                         break;
5383                 case MII_STAT1000:
5384                         data->val_out = adapter->phy_regs.stat1000;
5385                         break;
5386                 case MII_ESTATUS:
5387                         data->val_out = adapter->phy_regs.estatus;
5388                         break;
5389                 default:
5390                         return -EIO;
5391                 }
5392                 break;
5393         case SIOCSMIIREG:
5394         default:
5395                 return -EOPNOTSUPP;
5396         }
5397         return 0;
5398 }
5399
5400 static int e1000_ioctl(struct net_device *netdev, struct ifreq *ifr, int cmd)
5401 {
5402         switch (cmd) {
5403         case SIOCGMIIPHY:
5404         case SIOCGMIIREG:
5405         case SIOCSMIIREG:
5406                 return e1000_mii_ioctl(netdev, ifr, cmd);
5407         default:
5408                 return -EOPNOTSUPP;
5409         }
5410 }
5411
5412 static int e1000_init_phy_wakeup(struct e1000_adapter *adapter, u32 wufc)
5413 {
5414         struct e1000_hw *hw = &adapter->hw;
5415         u32 i, mac_reg;
5416         u16 phy_reg, wuc_enable;
5417         int retval = 0;
5418
5419         /* copy MAC RARs to PHY RARs */
5420         e1000_copy_rx_addrs_to_phy_ich8lan(hw);
5421
5422         retval = hw->phy.ops.acquire(hw);
5423         if (retval) {
5424                 e_err("Could not acquire PHY\n");
5425                 return retval;
5426         }
5427
5428         /* Enable access to wakeup registers on and set page to BM_WUC_PAGE */
5429         retval = e1000_enable_phy_wakeup_reg_access_bm(hw, &wuc_enable);
5430         if (retval)
5431                 goto release;
5432
5433         /* copy MAC MTA to PHY MTA - only needed for pchlan */
5434         for (i = 0; i < adapter->hw.mac.mta_reg_count; i++) {
5435                 mac_reg = E1000_READ_REG_ARRAY(hw, E1000_MTA, i);
5436                 hw->phy.ops.write_reg_page(hw, BM_MTA(i),
5437                                            (u16)(mac_reg & 0xFFFF));
5438                 hw->phy.ops.write_reg_page(hw, BM_MTA(i) + 1,
5439                                            (u16)((mac_reg >> 16) & 0xFFFF));
5440         }
5441
5442         /* configure PHY Rx Control register */
5443         hw->phy.ops.read_reg_page(&adapter->hw, BM_RCTL, &phy_reg);
5444         mac_reg = er32(RCTL);
5445         if (mac_reg & E1000_RCTL_UPE)
5446                 phy_reg |= BM_RCTL_UPE;
5447         if (mac_reg & E1000_RCTL_MPE)
5448                 phy_reg |= BM_RCTL_MPE;
5449         phy_reg &= ~(BM_RCTL_MO_MASK);
5450         if (mac_reg & E1000_RCTL_MO_3)
5451                 phy_reg |= (((mac_reg & E1000_RCTL_MO_3) >> E1000_RCTL_MO_SHIFT)
5452                                 << BM_RCTL_MO_SHIFT);
5453         if (mac_reg & E1000_RCTL_BAM)
5454                 phy_reg |= BM_RCTL_BAM;
5455         if (mac_reg & E1000_RCTL_PMCF)
5456                 phy_reg |= BM_RCTL_PMCF;
5457         mac_reg = er32(CTRL);
5458         if (mac_reg & E1000_CTRL_RFCE)
5459                 phy_reg |= BM_RCTL_RFCE;
5460         hw->phy.ops.write_reg_page(&adapter->hw, BM_RCTL, phy_reg);
5461
5462         /* enable PHY wakeup in MAC register */
5463         ew32(WUFC, wufc);
5464         ew32(WUC, E1000_WUC_PHY_WAKE | E1000_WUC_PME_EN);
5465
5466         /* configure and enable PHY wakeup in PHY registers */
5467         hw->phy.ops.write_reg_page(&adapter->hw, BM_WUFC, wufc);
5468         hw->phy.ops.write_reg_page(&adapter->hw, BM_WUC, E1000_WUC_PME_EN);
5469
5470         /* activate PHY wakeup */
5471         wuc_enable |= BM_WUC_ENABLE_BIT | BM_WUC_HOST_WU_BIT;
5472         retval = e1000_disable_phy_wakeup_reg_access_bm(hw, &wuc_enable);
5473         if (retval)
5474                 e_err("Could not set PHY Host Wakeup bit\n");
5475 release:
5476         hw->phy.ops.release(hw);
5477
5478         return retval;
5479 }
5480
5481 static int __e1000_shutdown(struct pci_dev *pdev, bool *enable_wake,
5482                             bool runtime)
5483 {
5484         struct net_device *netdev = pci_get_drvdata(pdev);
5485         struct e1000_adapter *adapter = netdev_priv(netdev);
5486         struct e1000_hw *hw = &adapter->hw;
5487         u32 ctrl, ctrl_ext, rctl, status;
5488         /* Runtime suspend should only enable wakeup for link changes */
5489         u32 wufc = runtime ? E1000_WUFC_LNKC : adapter->wol;
5490         int retval = 0;
5491
5492         netif_device_detach(netdev);
5493
5494         if (netif_running(netdev)) {
5495                 int count = E1000_CHECK_RESET_COUNT;
5496
5497                 while (test_bit(__E1000_RESETTING, &adapter->state) && count--)
5498                         usleep_range(10000, 20000);
5499
5500                 WARN_ON(test_bit(__E1000_RESETTING, &adapter->state));
5501                 e1000e_down(adapter);
5502                 e1000_free_irq(adapter);
5503         }
5504         e1000e_reset_interrupt_capability(adapter);
5505
5506         retval = pci_save_state(pdev);
5507         if (retval)
5508                 return retval;
5509
5510         status = er32(STATUS);
5511         if (status & E1000_STATUS_LU)
5512                 wufc &= ~E1000_WUFC_LNKC;
5513
5514         if (wufc) {
5515                 e1000_setup_rctl(adapter);
5516                 e1000e_set_rx_mode(netdev);
5517
5518                 /* turn on all-multi mode if wake on multicast is enabled */
5519                 if (wufc & E1000_WUFC_MC) {
5520                         rctl = er32(RCTL);
5521                         rctl |= E1000_RCTL_MPE;
5522                         ew32(RCTL, rctl);
5523                 }
5524
5525                 ctrl = er32(CTRL);
5526                 /* advertise wake from D3Cold */
5527                 #define E1000_CTRL_ADVD3WUC 0x00100000
5528                 /* phy power management enable */
5529                 #define E1000_CTRL_EN_PHY_PWR_MGMT 0x00200000
5530                 ctrl |= E1000_CTRL_ADVD3WUC;
5531                 if (!(adapter->flags2 & FLAG2_HAS_PHY_WAKEUP))
5532                         ctrl |= E1000_CTRL_EN_PHY_PWR_MGMT;
5533                 ew32(CTRL, ctrl);
5534
5535                 if (adapter->hw.phy.media_type == e1000_media_type_fiber ||
5536                     adapter->hw.phy.media_type ==
5537                     e1000_media_type_internal_serdes) {
5538                         /* keep the laser running in D3 */
5539                         ctrl_ext = er32(CTRL_EXT);
5540                         ctrl_ext |= E1000_CTRL_EXT_SDP3_DATA;
5541                         ew32(CTRL_EXT, ctrl_ext);
5542                 }
5543
5544                 if (adapter->flags & FLAG_IS_ICH)
5545                         e1000_suspend_workarounds_ich8lan(&adapter->hw);
5546
5547                 /* Allow time for pending master requests to run */
5548                 e1000e_disable_pcie_master(&adapter->hw);
5549
5550                 if (adapter->flags2 & FLAG2_HAS_PHY_WAKEUP) {
5551                         /* enable wakeup by the PHY */
5552                         retval = e1000_init_phy_wakeup(adapter, wufc);
5553                         if (retval)
5554                                 return retval;
5555                 } else {
5556                         /* enable wakeup by the MAC */
5557                         ew32(WUFC, wufc);
5558                         ew32(WUC, E1000_WUC_PME_EN);
5559                 }
5560         } else {
5561                 ew32(WUC, 0);
5562                 ew32(WUFC, 0);
5563         }
5564
5565         *enable_wake = !!wufc;
5566
5567         /* make sure adapter isn't asleep if manageability is enabled */
5568         if ((adapter->flags & FLAG_MNG_PT_ENABLED) ||
5569             (hw->mac.ops.check_mng_mode(hw)))
5570                 *enable_wake = true;
5571
5572         if (adapter->hw.phy.type == e1000_phy_igp_3)
5573                 e1000e_igp3_phy_powerdown_workaround_ich8lan(&adapter->hw);
5574
5575         /*
5576          * Release control of h/w to f/w.  If f/w is AMT enabled, this
5577          * would have already happened in close and is redundant.
5578          */
5579         e1000e_release_hw_control(adapter);
5580
5581         pci_disable_device(pdev);
5582
5583         return 0;
5584 }
5585
5586 static void e1000_power_off(struct pci_dev *pdev, bool sleep, bool wake)
5587 {
5588         if (sleep && wake) {
5589                 pci_prepare_to_sleep(pdev);
5590                 return;
5591         }
5592
5593         pci_wake_from_d3(pdev, wake);
5594         pci_set_power_state(pdev, PCI_D3hot);
5595 }
5596
5597 static void e1000_complete_shutdown(struct pci_dev *pdev, bool sleep,
5598                                     bool wake)
5599 {
5600         struct net_device *netdev = pci_get_drvdata(pdev);
5601         struct e1000_adapter *adapter = netdev_priv(netdev);
5602
5603         /*
5604          * The pci-e switch on some quad port adapters will report a
5605          * correctable error when the MAC transitions from D0 to D3.  To
5606          * prevent this we need to mask off the correctable errors on the
5607          * downstream port of the pci-e switch.
5608          */
5609         if (adapter->flags & FLAG_IS_QUAD_PORT) {
5610                 struct pci_dev *us_dev = pdev->bus->self;
5611                 int pos = pci_pcie_cap(us_dev);
5612                 u16 devctl;
5613
5614                 pci_read_config_word(us_dev, pos + PCI_EXP_DEVCTL, &devctl);
5615                 pci_write_config_word(us_dev, pos + PCI_EXP_DEVCTL,
5616                                       (devctl & ~PCI_EXP_DEVCTL_CERE));
5617
5618                 e1000_power_off(pdev, sleep, wake);
5619
5620                 pci_write_config_word(us_dev, pos + PCI_EXP_DEVCTL, devctl);
5621         } else {
5622                 e1000_power_off(pdev, sleep, wake);
5623         }
5624 }
5625
5626 #ifdef CONFIG_PCIEASPM
5627 static void __e1000e_disable_aspm(struct pci_dev *pdev, u16 state)
5628 {
5629         pci_disable_link_state_locked(pdev, state);
5630 }
5631 #else
5632 static void __e1000e_disable_aspm(struct pci_dev *pdev, u16 state)
5633 {
5634         int pos;
5635         u16 reg16;
5636
5637         /*
5638          * Both device and parent should have the same ASPM setting.
5639          * Disable ASPM in downstream component first and then upstream.
5640          */
5641         pos = pci_pcie_cap(pdev);
5642         pci_read_config_word(pdev, pos + PCI_EXP_LNKCTL, &reg16);
5643         reg16 &= ~state;
5644         pci_write_config_word(pdev, pos + PCI_EXP_LNKCTL, reg16);
5645
5646         if (!pdev->bus->self)
5647                 return;
5648
5649         pos = pci_pcie_cap(pdev->bus->self);
5650         pci_read_config_word(pdev->bus->self, pos + PCI_EXP_LNKCTL, &reg16);
5651         reg16 &= ~state;
5652         pci_write_config_word(pdev->bus->self, pos + PCI_EXP_LNKCTL, reg16);
5653 }
5654 #endif
5655 static void e1000e_disable_aspm(struct pci_dev *pdev, u16 state)
5656 {
5657         dev_info(&pdev->dev, "Disabling ASPM %s %s\n",
5658                  (state & PCIE_LINK_STATE_L0S) ? "L0s" : "",
5659                  (state & PCIE_LINK_STATE_L1) ? "L1" : "");
5660
5661         __e1000e_disable_aspm(pdev, state);
5662 }
5663
5664 #ifdef CONFIG_PM
5665 static bool e1000e_pm_ready(struct e1000_adapter *adapter)
5666 {
5667         return !!adapter->tx_ring->buffer_info;
5668 }
5669
5670 static int __e1000_resume(struct pci_dev *pdev)
5671 {
5672         struct net_device *netdev = pci_get_drvdata(pdev);
5673         struct e1000_adapter *adapter = netdev_priv(netdev);
5674         struct e1000_hw *hw = &adapter->hw;
5675         u16 aspm_disable_flag = 0;
5676         u32 err;
5677
5678         if (adapter->flags2 & FLAG2_DISABLE_ASPM_L0S)
5679                 aspm_disable_flag = PCIE_LINK_STATE_L0S;
5680         if (adapter->flags2 & FLAG2_DISABLE_ASPM_L1)
5681                 aspm_disable_flag |= PCIE_LINK_STATE_L1;
5682         if (aspm_disable_flag)
5683                 e1000e_disable_aspm(pdev, aspm_disable_flag);
5684
5685         pci_set_power_state(pdev, PCI_D0);
5686         pci_restore_state(pdev);
5687         pci_save_state(pdev);
5688
5689         e1000e_set_interrupt_capability(adapter);
5690         if (netif_running(netdev)) {
5691                 err = e1000_request_irq(adapter);
5692                 if (err)
5693                         return err;
5694         }
5695
5696         if (hw->mac.type == e1000_pch2lan)
5697                 e1000_resume_workarounds_pchlan(&adapter->hw);
5698
5699         e1000e_power_up_phy(adapter);
5700
5701         /* report the system wakeup cause from S3/S4 */
5702         if (adapter->flags2 & FLAG2_HAS_PHY_WAKEUP) {
5703                 u16 phy_data;
5704
5705                 e1e_rphy(&adapter->hw, BM_WUS, &phy_data);
5706                 if (phy_data) {
5707                         e_info("PHY Wakeup cause - %s\n",
5708                                 phy_data & E1000_WUS_EX ? "Unicast Packet" :
5709                                 phy_data & E1000_WUS_MC ? "Multicast Packet" :
5710                                 phy_data & E1000_WUS_BC ? "Broadcast Packet" :
5711                                 phy_data & E1000_WUS_MAG ? "Magic Packet" :
5712                                 phy_data & E1000_WUS_LNKC ?
5713                                 "Link Status Change" : "other");
5714                 }
5715                 e1e_wphy(&adapter->hw, BM_WUS, ~0);
5716         } else {
5717                 u32 wus = er32(WUS);
5718                 if (wus) {
5719                         e_info("MAC Wakeup cause - %s\n",
5720                                 wus & E1000_WUS_EX ? "Unicast Packet" :
5721                                 wus & E1000_WUS_MC ? "Multicast Packet" :
5722                                 wus & E1000_WUS_BC ? "Broadcast Packet" :
5723                                 wus & E1000_WUS_MAG ? "Magic Packet" :
5724                                 wus & E1000_WUS_LNKC ? "Link Status Change" :
5725                                 "other");
5726                 }
5727                 ew32(WUS, ~0);
5728         }
5729
5730         e1000e_reset(adapter);
5731
5732         e1000_init_manageability_pt(adapter);
5733
5734         if (netif_running(netdev))
5735                 e1000e_up(adapter);
5736
5737         netif_device_attach(netdev);
5738
5739         /*
5740          * If the controller has AMT, do not set DRV_LOAD until the interface
5741          * is up.  For all other cases, let the f/w know that the h/w is now
5742          * under the control of the driver.
5743          */
5744         if (!(adapter->flags & FLAG_HAS_AMT))
5745                 e1000e_get_hw_control(adapter);
5746
5747         return 0;
5748 }
5749
5750 #ifdef CONFIG_PM_SLEEP
5751 static int e1000_suspend(struct device *dev)
5752 {
5753         struct pci_dev *pdev = to_pci_dev(dev);
5754         int retval;
5755         bool wake;
5756
5757         retval = __e1000_shutdown(pdev, &wake, false);
5758         if (!retval)
5759                 e1000_complete_shutdown(pdev, true, wake);
5760
5761         return retval;
5762 }
5763
5764 static int e1000_resume(struct device *dev)
5765 {
5766         struct pci_dev *pdev = to_pci_dev(dev);
5767         struct net_device *netdev = pci_get_drvdata(pdev);
5768         struct e1000_adapter *adapter = netdev_priv(netdev);
5769
5770         if (e1000e_pm_ready(adapter))
5771                 adapter->idle_check = true;
5772
5773         return __e1000_resume(pdev);
5774 }
5775 #endif /* CONFIG_PM_SLEEP */
5776
5777 #ifdef CONFIG_PM_RUNTIME
5778 static int e1000_runtime_suspend(struct device *dev)
5779 {
5780         struct pci_dev *pdev = to_pci_dev(dev);
5781         struct net_device *netdev = pci_get_drvdata(pdev);
5782         struct e1000_adapter *adapter = netdev_priv(netdev);
5783
5784         if (e1000e_pm_ready(adapter)) {
5785                 bool wake;
5786
5787                 __e1000_shutdown(pdev, &wake, true);
5788         }
5789
5790         return 0;
5791 }
5792
5793 static int e1000_idle(struct device *dev)
5794 {
5795         struct pci_dev *pdev = to_pci_dev(dev);
5796         struct net_device *netdev = pci_get_drvdata(pdev);
5797         struct e1000_adapter *adapter = netdev_priv(netdev);
5798
5799         if (!e1000e_pm_ready(adapter))
5800                 return 0;
5801
5802         if (adapter->idle_check) {
5803                 adapter->idle_check = false;
5804                 if (!e1000e_has_link(adapter))
5805                         pm_schedule_suspend(dev, MSEC_PER_SEC);
5806         }
5807
5808         return -EBUSY;
5809 }
5810
5811 static int e1000_runtime_resume(struct device *dev)
5812 {
5813         struct pci_dev *pdev = to_pci_dev(dev);
5814         struct net_device *netdev = pci_get_drvdata(pdev);
5815         struct e1000_adapter *adapter = netdev_priv(netdev);
5816
5817         if (!e1000e_pm_ready(adapter))
5818                 return 0;
5819
5820         adapter->idle_check = !dev->power.runtime_auto;
5821         return __e1000_resume(pdev);
5822 }
5823 #endif /* CONFIG_PM_RUNTIME */
5824 #endif /* CONFIG_PM */
5825
5826 static void e1000_shutdown(struct pci_dev *pdev)
5827 {
5828         bool wake = false;
5829
5830         __e1000_shutdown(pdev, &wake, false);
5831
5832         if (system_state == SYSTEM_POWER_OFF)
5833                 e1000_complete_shutdown(pdev, false, wake);
5834 }
5835
5836 #ifdef CONFIG_NET_POLL_CONTROLLER
5837
5838 static irqreturn_t e1000_intr_msix(int irq, void *data)
5839 {
5840         struct net_device *netdev = data;
5841         struct e1000_adapter *adapter = netdev_priv(netdev);
5842
5843         if (adapter->msix_entries) {
5844                 int vector, msix_irq;
5845
5846                 vector = 0;
5847                 msix_irq = adapter->msix_entries[vector].vector;
5848                 disable_irq(msix_irq);
5849                 e1000_intr_msix_rx(msix_irq, netdev);
5850                 enable_irq(msix_irq);
5851
5852                 vector++;
5853                 msix_irq = adapter->msix_entries[vector].vector;
5854                 disable_irq(msix_irq);
5855                 e1000_intr_msix_tx(msix_irq, netdev);
5856                 enable_irq(msix_irq);
5857
5858                 vector++;
5859                 msix_irq = adapter->msix_entries[vector].vector;
5860                 disable_irq(msix_irq);
5861                 e1000_msix_other(msix_irq, netdev);
5862                 enable_irq(msix_irq);
5863         }
5864
5865         return IRQ_HANDLED;
5866 }
5867
5868 /*
5869  * Polling 'interrupt' - used by things like netconsole to send skbs
5870  * without having to re-enable interrupts. It's not called while
5871  * the interrupt routine is executing.
5872  */
5873 static void e1000_netpoll(struct net_device *netdev)
5874 {
5875         struct e1000_adapter *adapter = netdev_priv(netdev);
5876
5877         switch (adapter->int_mode) {
5878         case E1000E_INT_MODE_MSIX:
5879                 e1000_intr_msix(adapter->pdev->irq, netdev);
5880                 break;
5881         case E1000E_INT_MODE_MSI:
5882                 disable_irq(adapter->pdev->irq);
5883                 e1000_intr_msi(adapter->pdev->irq, netdev);
5884                 enable_irq(adapter->pdev->irq);
5885                 break;
5886         default: /* E1000E_INT_MODE_LEGACY */
5887                 disable_irq(adapter->pdev->irq);
5888                 e1000_intr(adapter->pdev->irq, netdev);
5889                 enable_irq(adapter->pdev->irq);
5890                 break;
5891         }
5892 }
5893 #endif
5894
5895 /**
5896  * e1000_io_error_detected - called when PCI error is detected
5897  * @pdev: Pointer to PCI device
5898  * @state: The current pci connection state
5899  *
5900  * This function is called after a PCI bus error affecting
5901  * this device has been detected.
5902  */
5903 static pci_ers_result_t e1000_io_error_detected(struct pci_dev *pdev,
5904                                                 pci_channel_state_t state)
5905 {
5906         struct net_device *netdev = pci_get_drvdata(pdev);
5907         struct e1000_adapter *adapter = netdev_priv(netdev);
5908
5909         netif_device_detach(netdev);
5910
5911         if (state == pci_channel_io_perm_failure)
5912                 return PCI_ERS_RESULT_DISCONNECT;
5913
5914         if (netif_running(netdev))
5915                 e1000e_down(adapter);
5916         pci_disable_device(pdev);
5917
5918         /* Request a slot slot reset. */
5919         return PCI_ERS_RESULT_NEED_RESET;
5920 }
5921
5922 /**
5923  * e1000_io_slot_reset - called after the pci bus has been reset.
5924  * @pdev: Pointer to PCI device
5925  *
5926  * Restart the card from scratch, as if from a cold-boot. Implementation
5927  * resembles the first-half of the e1000_resume routine.
5928  */
5929 static pci_ers_result_t e1000_io_slot_reset(struct pci_dev *pdev)
5930 {
5931         struct net_device *netdev = pci_get_drvdata(pdev);
5932         struct e1000_adapter *adapter = netdev_priv(netdev);
5933         struct e1000_hw *hw = &adapter->hw;
5934         u16 aspm_disable_flag = 0;
5935         int err;
5936         pci_ers_result_t result;
5937
5938         if (adapter->flags2 & FLAG2_DISABLE_ASPM_L0S)
5939                 aspm_disable_flag = PCIE_LINK_STATE_L0S;
5940         if (adapter->flags2 & FLAG2_DISABLE_ASPM_L1)
5941                 aspm_disable_flag |= PCIE_LINK_STATE_L1;
5942         if (aspm_disable_flag)
5943                 e1000e_disable_aspm(pdev, aspm_disable_flag);
5944
5945         err = pci_enable_device_mem(pdev);
5946         if (err) {
5947                 dev_err(&pdev->dev,
5948                         "Cannot re-enable PCI device after reset.\n");
5949                 result = PCI_ERS_RESULT_DISCONNECT;
5950         } else {
5951                 pci_set_master(pdev);
5952                 pdev->state_saved = true;
5953                 pci_restore_state(pdev);
5954
5955                 pci_enable_wake(pdev, PCI_D3hot, 0);
5956                 pci_enable_wake(pdev, PCI_D3cold, 0);
5957
5958                 e1000e_reset(adapter);
5959                 ew32(WUS, ~0);
5960                 result = PCI_ERS_RESULT_RECOVERED;
5961         }
5962
5963         pci_cleanup_aer_uncorrect_error_status(pdev);
5964
5965         return result;
5966 }
5967
5968 /**
5969  * e1000_io_resume - called when traffic can start flowing again.
5970  * @pdev: Pointer to PCI device
5971  *
5972  * This callback is called when the error recovery driver tells us that
5973  * its OK to resume normal operation. Implementation resembles the
5974  * second-half of the e1000_resume routine.
5975  */
5976 static void e1000_io_resume(struct pci_dev *pdev)
5977 {
5978         struct net_device *netdev = pci_get_drvdata(pdev);
5979         struct e1000_adapter *adapter = netdev_priv(netdev);
5980
5981         e1000_init_manageability_pt(adapter);
5982
5983         if (netif_running(netdev)) {
5984                 if (e1000e_up(adapter)) {
5985                         dev_err(&pdev->dev,
5986                                 "can't bring device back up after reset\n");
5987                         return;
5988                 }
5989         }
5990
5991         netif_device_attach(netdev);
5992
5993         /*
5994          * If the controller has AMT, do not set DRV_LOAD until the interface
5995          * is up.  For all other cases, let the f/w know that the h/w is now
5996          * under the control of the driver.
5997          */
5998         if (!(adapter->flags & FLAG_HAS_AMT))
5999                 e1000e_get_hw_control(adapter);
6000
6001 }
6002
6003 static void e1000_print_device_info(struct e1000_adapter *adapter)
6004 {
6005         struct e1000_hw *hw = &adapter->hw;
6006         struct net_device *netdev = adapter->netdev;
6007         u32 ret_val;
6008         u8 pba_str[E1000_PBANUM_LENGTH];
6009
6010         /* print bus type/speed/width info */
6011         e_info("(PCI Express:2.5GT/s:%s) %pM\n",
6012                /* bus width */
6013                ((hw->bus.width == e1000_bus_width_pcie_x4) ? "Width x4" :
6014                 "Width x1"),
6015                /* MAC address */
6016                netdev->dev_addr);
6017         e_info("Intel(R) PRO/%s Network Connection\n",
6018                (hw->phy.type == e1000_phy_ife) ? "10/100" : "1000");
6019         ret_val = e1000_read_pba_string_generic(hw, pba_str,
6020                                                 E1000_PBANUM_LENGTH);
6021         if (ret_val)
6022                 strlcpy((char *)pba_str, "Unknown", sizeof(pba_str));
6023         e_info("MAC: %d, PHY: %d, PBA No: %s\n",
6024                hw->mac.type, hw->phy.type, pba_str);
6025 }
6026
6027 static void e1000_eeprom_checks(struct e1000_adapter *adapter)
6028 {
6029         struct e1000_hw *hw = &adapter->hw;
6030         int ret_val;
6031         u16 buf = 0;
6032
6033         if (hw->mac.type != e1000_82573)
6034                 return;
6035
6036         ret_val = e1000_read_nvm(hw, NVM_INIT_CONTROL2_REG, 1, &buf);
6037         le16_to_cpus(&buf);
6038         if (!ret_val && (!(buf & (1 << 0)))) {
6039                 /* Deep Smart Power Down (DSPD) */
6040                 dev_warn(&adapter->pdev->dev,
6041                          "Warning: detected DSPD enabled in EEPROM\n");
6042         }
6043 }
6044
6045 static int e1000_set_features(struct net_device *netdev,
6046                               netdev_features_t features)
6047 {
6048         struct e1000_adapter *adapter = netdev_priv(netdev);
6049         netdev_features_t changed = features ^ netdev->features;
6050
6051         if (changed & (NETIF_F_TSO | NETIF_F_TSO6))
6052                 adapter->flags |= FLAG_TSO_FORCE;
6053
6054         if (!(changed & (NETIF_F_HW_VLAN_RX | NETIF_F_HW_VLAN_TX |
6055                          NETIF_F_RXCSUM | NETIF_F_RXHASH | NETIF_F_RXFCS |
6056                          NETIF_F_RXALL)))
6057                 return 0;
6058
6059         /*
6060          * IP payload checksum (enabled with jumbos/packet-split when Rx
6061          * checksum is enabled) and generation of RSS hash is mutually
6062          * exclusive in the hardware.
6063          */
6064         if (adapter->rx_ps_pages &&
6065             (features & NETIF_F_RXCSUM) && (features & NETIF_F_RXHASH)) {
6066                 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");
6067                 return -EINVAL;
6068         }
6069
6070         if (changed & NETIF_F_RXFCS) {
6071                 if (features & NETIF_F_RXFCS) {
6072                         adapter->flags2 &= ~FLAG2_CRC_STRIPPING;
6073                 } else {
6074                         /* We need to take it back to defaults, which might mean
6075                          * stripping is still disabled at the adapter level.
6076                          */
6077                         if (adapter->flags2 & FLAG2_DFLT_CRC_STRIPPING)
6078                                 adapter->flags2 |= FLAG2_CRC_STRIPPING;
6079                         else
6080                                 adapter->flags2 &= ~FLAG2_CRC_STRIPPING;
6081                 }
6082         }
6083
6084         netdev->features = features;
6085
6086         if (netif_running(netdev))
6087                 e1000e_reinit_locked(adapter);
6088         else
6089                 e1000e_reset(adapter);
6090
6091         return 0;
6092 }
6093
6094 static const struct net_device_ops e1000e_netdev_ops = {
6095         .ndo_open               = e1000_open,
6096         .ndo_stop               = e1000_close,
6097         .ndo_start_xmit         = e1000_xmit_frame,
6098         .ndo_get_stats64        = e1000e_get_stats64,
6099         .ndo_set_rx_mode        = e1000e_set_rx_mode,
6100         .ndo_set_mac_address    = e1000_set_mac,
6101         .ndo_change_mtu         = e1000_change_mtu,
6102         .ndo_do_ioctl           = e1000_ioctl,
6103         .ndo_tx_timeout         = e1000_tx_timeout,
6104         .ndo_validate_addr      = eth_validate_addr,
6105
6106         .ndo_vlan_rx_add_vid    = e1000_vlan_rx_add_vid,
6107         .ndo_vlan_rx_kill_vid   = e1000_vlan_rx_kill_vid,
6108 #ifdef CONFIG_NET_POLL_CONTROLLER
6109         .ndo_poll_controller    = e1000_netpoll,
6110 #endif
6111         .ndo_set_features = e1000_set_features,
6112 };
6113
6114 /**
6115  * e1000_probe - Device Initialization Routine
6116  * @pdev: PCI device information struct
6117  * @ent: entry in e1000_pci_tbl
6118  *
6119  * Returns 0 on success, negative on failure
6120  *
6121  * e1000_probe initializes an adapter identified by a pci_dev structure.
6122  * The OS initialization, configuring of the adapter private structure,
6123  * and a hardware reset occur.
6124  **/
6125 static int __devinit e1000_probe(struct pci_dev *pdev,
6126                                  const struct pci_device_id *ent)
6127 {
6128         struct net_device *netdev;
6129         struct e1000_adapter *adapter;
6130         struct e1000_hw *hw;
6131         const struct e1000_info *ei = e1000_info_tbl[ent->driver_data];
6132         resource_size_t mmio_start, mmio_len;
6133         resource_size_t flash_start, flash_len;
6134         static int cards_found;
6135         u16 aspm_disable_flag = 0;
6136         int i, err, pci_using_dac;
6137         u16 eeprom_data = 0;
6138         u16 eeprom_apme_mask = E1000_EEPROM_APME;
6139
6140         if (ei->flags2 & FLAG2_DISABLE_ASPM_L0S)
6141                 aspm_disable_flag = PCIE_LINK_STATE_L0S;
6142         if (ei->flags2 & FLAG2_DISABLE_ASPM_L1)
6143                 aspm_disable_flag |= PCIE_LINK_STATE_L1;
6144         if (aspm_disable_flag)
6145                 e1000e_disable_aspm(pdev, aspm_disable_flag);
6146
6147         err = pci_enable_device_mem(pdev);
6148         if (err)
6149                 return err;
6150
6151         pci_using_dac = 0;
6152         err = dma_set_mask(&pdev->dev, DMA_BIT_MASK(64));
6153         if (!err) {
6154                 err = dma_set_coherent_mask(&pdev->dev, DMA_BIT_MASK(64));
6155                 if (!err)
6156                         pci_using_dac = 1;
6157         } else {
6158                 err = dma_set_mask(&pdev->dev, DMA_BIT_MASK(32));
6159                 if (err) {
6160                         err = dma_set_coherent_mask(&pdev->dev,
6161                                                     DMA_BIT_MASK(32));
6162                         if (err) {
6163                                 dev_err(&pdev->dev, "No usable DMA configuration, aborting\n");
6164                                 goto err_dma;
6165                         }
6166                 }
6167         }
6168
6169         err = pci_request_selected_regions_exclusive(pdev,
6170                                           pci_select_bars(pdev, IORESOURCE_MEM),
6171                                           e1000e_driver_name);
6172         if (err)
6173                 goto err_pci_reg;
6174
6175         /* AER (Advanced Error Reporting) hooks */
6176         pci_enable_pcie_error_reporting(pdev);
6177
6178         pci_set_master(pdev);
6179         /* PCI config space info */
6180         err = pci_save_state(pdev);
6181         if (err)
6182                 goto err_alloc_etherdev;
6183
6184         err = -ENOMEM;
6185         netdev = alloc_etherdev(sizeof(struct e1000_adapter));
6186         if (!netdev)
6187                 goto err_alloc_etherdev;
6188
6189         SET_NETDEV_DEV(netdev, &pdev->dev);
6190
6191         netdev->irq = pdev->irq;
6192
6193         pci_set_drvdata(pdev, netdev);
6194         adapter = netdev_priv(netdev);
6195         hw = &adapter->hw;
6196         adapter->netdev = netdev;
6197         adapter->pdev = pdev;
6198         adapter->ei = ei;
6199         adapter->pba = ei->pba;
6200         adapter->flags = ei->flags;
6201         adapter->flags2 = ei->flags2;
6202         adapter->hw.adapter = adapter;
6203         adapter->hw.mac.type = ei->mac;
6204         adapter->max_hw_frame_size = ei->max_hw_frame_size;
6205         adapter->msg_enable = netif_msg_init(debug, DEFAULT_MSG_ENABLE);
6206
6207         mmio_start = pci_resource_start(pdev, 0);
6208         mmio_len = pci_resource_len(pdev, 0);
6209
6210         err = -EIO;
6211         adapter->hw.hw_addr = ioremap(mmio_start, mmio_len);
6212         if (!adapter->hw.hw_addr)
6213                 goto err_ioremap;
6214
6215         if ((adapter->flags & FLAG_HAS_FLASH) &&
6216             (pci_resource_flags(pdev, 1) & IORESOURCE_MEM)) {
6217                 flash_start = pci_resource_start(pdev, 1);
6218                 flash_len = pci_resource_len(pdev, 1);
6219                 adapter->hw.flash_address = ioremap(flash_start, flash_len);
6220                 if (!adapter->hw.flash_address)
6221                         goto err_flashmap;
6222         }
6223
6224         /* construct the net_device struct */
6225         netdev->netdev_ops              = &e1000e_netdev_ops;
6226         e1000e_set_ethtool_ops(netdev);
6227         netdev->watchdog_timeo          = 5 * HZ;
6228         netif_napi_add(netdev, &adapter->napi, e1000e_poll, 64);
6229         strlcpy(netdev->name, pci_name(pdev), sizeof(netdev->name));
6230
6231         netdev->mem_start = mmio_start;
6232         netdev->mem_end = mmio_start + mmio_len;
6233
6234         adapter->bd_number = cards_found++;
6235
6236         e1000e_check_options(adapter);
6237
6238         /* setup adapter struct */
6239         err = e1000_sw_init(adapter);
6240         if (err)
6241                 goto err_sw_init;
6242
6243         memcpy(&hw->mac.ops, ei->mac_ops, sizeof(hw->mac.ops));
6244         memcpy(&hw->nvm.ops, ei->nvm_ops, sizeof(hw->nvm.ops));
6245         memcpy(&hw->phy.ops, ei->phy_ops, sizeof(hw->phy.ops));
6246
6247         err = ei->get_variants(adapter);
6248         if (err)
6249                 goto err_hw_init;
6250
6251         if ((adapter->flags & FLAG_IS_ICH) &&
6252             (adapter->flags & FLAG_READ_ONLY_NVM))
6253                 e1000e_write_protect_nvm_ich8lan(&adapter->hw);
6254
6255         hw->mac.ops.get_bus_info(&adapter->hw);
6256
6257         adapter->hw.phy.autoneg_wait_to_complete = 0;
6258
6259         /* Copper options */
6260         if (adapter->hw.phy.media_type == e1000_media_type_copper) {
6261                 adapter->hw.phy.mdix = AUTO_ALL_MODES;
6262                 adapter->hw.phy.disable_polarity_correction = 0;
6263                 adapter->hw.phy.ms_type = e1000_ms_hw_default;
6264         }
6265
6266         if (hw->phy.ops.check_reset_block(hw))
6267                 e_info("PHY reset is blocked due to SOL/IDER session.\n");
6268
6269         /* Set initial default active device features */
6270         netdev->features = (NETIF_F_SG |
6271                             NETIF_F_HW_VLAN_RX |
6272                             NETIF_F_HW_VLAN_TX |
6273                             NETIF_F_TSO |
6274                             NETIF_F_TSO6 |
6275                             NETIF_F_RXHASH |
6276                             NETIF_F_RXCSUM |
6277                             NETIF_F_HW_CSUM);
6278
6279         /* Set user-changeable features (subset of all device features) */
6280         netdev->hw_features = netdev->features;
6281         netdev->hw_features |= NETIF_F_RXFCS;
6282         netdev->priv_flags |= IFF_SUPP_NOFCS;
6283         netdev->hw_features |= NETIF_F_RXALL;
6284
6285         if (adapter->flags & FLAG_HAS_HW_VLAN_FILTER)
6286                 netdev->features |= NETIF_F_HW_VLAN_FILTER;
6287
6288         netdev->vlan_features |= (NETIF_F_SG |
6289                                   NETIF_F_TSO |
6290                                   NETIF_F_TSO6 |
6291                                   NETIF_F_HW_CSUM);
6292
6293         netdev->priv_flags |= IFF_UNICAST_FLT;
6294
6295         if (pci_using_dac) {
6296                 netdev->features |= NETIF_F_HIGHDMA;
6297                 netdev->vlan_features |= NETIF_F_HIGHDMA;
6298         }
6299
6300         if (e1000e_enable_mng_pass_thru(&adapter->hw))
6301                 adapter->flags |= FLAG_MNG_PT_ENABLED;
6302
6303         /*
6304          * before reading the NVM, reset the controller to
6305          * put the device in a known good starting state
6306          */
6307         adapter->hw.mac.ops.reset_hw(&adapter->hw);
6308
6309         /*
6310          * systems with ASPM and others may see the checksum fail on the first
6311          * attempt. Let's give it a few tries
6312          */
6313         for (i = 0;; i++) {
6314                 if (e1000_validate_nvm_checksum(&adapter->hw) >= 0)
6315                         break;
6316                 if (i == 2) {
6317                         e_err("The NVM Checksum Is Not Valid\n");
6318                         err = -EIO;
6319                         goto err_eeprom;
6320                 }
6321         }
6322
6323         e1000_eeprom_checks(adapter);
6324
6325         /* copy the MAC address */
6326         if (e1000e_read_mac_addr(&adapter->hw))
6327                 e_err("NVM Read Error while reading MAC address\n");
6328
6329         memcpy(netdev->dev_addr, adapter->hw.mac.addr, netdev->addr_len);
6330         memcpy(netdev->perm_addr, adapter->hw.mac.addr, netdev->addr_len);
6331
6332         if (!is_valid_ether_addr(netdev->perm_addr)) {
6333                 e_err("Invalid MAC Address: %pM\n", netdev->perm_addr);
6334                 err = -EIO;
6335                 goto err_eeprom;
6336         }
6337
6338         init_timer(&adapter->watchdog_timer);
6339         adapter->watchdog_timer.function = e1000_watchdog;
6340         adapter->watchdog_timer.data = (unsigned long) adapter;
6341
6342         init_timer(&adapter->phy_info_timer);
6343         adapter->phy_info_timer.function = e1000_update_phy_info;
6344         adapter->phy_info_timer.data = (unsigned long) adapter;
6345
6346         INIT_WORK(&adapter->reset_task, e1000_reset_task);
6347         INIT_WORK(&adapter->watchdog_task, e1000_watchdog_task);
6348         INIT_WORK(&adapter->downshift_task, e1000e_downshift_workaround);
6349         INIT_WORK(&adapter->update_phy_task, e1000e_update_phy_task);
6350         INIT_WORK(&adapter->print_hang_task, e1000_print_hw_hang);
6351
6352         /* Initialize link parameters. User can change them with ethtool */
6353         adapter->hw.mac.autoneg = 1;
6354         adapter->fc_autoneg = true;
6355         adapter->hw.fc.requested_mode = e1000_fc_default;
6356         adapter->hw.fc.current_mode = e1000_fc_default;
6357         adapter->hw.phy.autoneg_advertised = 0x2f;
6358
6359         /* ring size defaults */
6360         adapter->rx_ring->count = 256;
6361         adapter->tx_ring->count = 256;
6362
6363         /*
6364          * Initial Wake on LAN setting - If APM wake is enabled in
6365          * the EEPROM, enable the ACPI Magic Packet filter
6366          */
6367         if (adapter->flags & FLAG_APME_IN_WUC) {
6368                 /* APME bit in EEPROM is mapped to WUC.APME */
6369                 eeprom_data = er32(WUC);
6370                 eeprom_apme_mask = E1000_WUC_APME;
6371                 if ((hw->mac.type > e1000_ich10lan) &&
6372                     (eeprom_data & E1000_WUC_PHY_WAKE))
6373                         adapter->flags2 |= FLAG2_HAS_PHY_WAKEUP;
6374         } else if (adapter->flags & FLAG_APME_IN_CTRL3) {
6375                 if (adapter->flags & FLAG_APME_CHECK_PORT_B &&
6376                     (adapter->hw.bus.func == 1))
6377                         e1000_read_nvm(&adapter->hw, NVM_INIT_CONTROL3_PORT_B,
6378                                        1, &eeprom_data);
6379                 else
6380                         e1000_read_nvm(&adapter->hw, NVM_INIT_CONTROL3_PORT_A,
6381                                        1, &eeprom_data);
6382         }
6383
6384         /* fetch WoL from EEPROM */
6385         if (eeprom_data & eeprom_apme_mask)
6386                 adapter->eeprom_wol |= E1000_WUFC_MAG;
6387
6388         /*
6389          * now that we have the eeprom settings, apply the special cases
6390          * where the eeprom may be wrong or the board simply won't support
6391          * wake on lan on a particular port
6392          */
6393         if (!(adapter->flags & FLAG_HAS_WOL))
6394                 adapter->eeprom_wol = 0;
6395
6396         /* initialize the wol settings based on the eeprom settings */
6397         adapter->wol = adapter->eeprom_wol;
6398         device_set_wakeup_enable(&adapter->pdev->dev, adapter->wol);
6399
6400         /* save off EEPROM version number */
6401         e1000_read_nvm(&adapter->hw, 5, 1, &adapter->eeprom_vers);
6402
6403         /* reset the hardware with the new settings */
6404         e1000e_reset(adapter);
6405
6406         /*
6407          * If the controller has AMT, do not set DRV_LOAD until the interface
6408          * is up.  For all other cases, let the f/w know that the h/w is now
6409          * under the control of the driver.
6410          */
6411         if (!(adapter->flags & FLAG_HAS_AMT))
6412                 e1000e_get_hw_control(adapter);
6413
6414         strlcpy(netdev->name, "eth%d", sizeof(netdev->name));
6415         err = register_netdev(netdev);
6416         if (err)
6417                 goto err_register;
6418
6419         /* carrier off reporting is important to ethtool even BEFORE open */
6420         netif_carrier_off(netdev);
6421
6422         e1000_print_device_info(adapter);
6423
6424         if (pci_dev_run_wake(pdev))
6425                 pm_runtime_put_noidle(&pdev->dev);
6426
6427         return 0;
6428
6429 err_register:
6430         if (!(adapter->flags & FLAG_HAS_AMT))
6431                 e1000e_release_hw_control(adapter);
6432 err_eeprom:
6433         if (!hw->phy.ops.check_reset_block(hw))
6434                 e1000_phy_hw_reset(&adapter->hw);
6435 err_hw_init:
6436         kfree(adapter->tx_ring);
6437         kfree(adapter->rx_ring);
6438 err_sw_init:
6439         if (adapter->hw.flash_address)
6440                 iounmap(adapter->hw.flash_address);
6441         e1000e_reset_interrupt_capability(adapter);
6442 err_flashmap:
6443         iounmap(adapter->hw.hw_addr);
6444 err_ioremap:
6445         free_netdev(netdev);
6446 err_alloc_etherdev:
6447         pci_release_selected_regions(pdev,
6448                                      pci_select_bars(pdev, IORESOURCE_MEM));
6449 err_pci_reg:
6450 err_dma:
6451         pci_disable_device(pdev);
6452         return err;
6453 }
6454
6455 /**
6456  * e1000_remove - Device Removal Routine
6457  * @pdev: PCI device information struct
6458  *
6459  * e1000_remove is called by the PCI subsystem to alert the driver
6460  * that it should release a PCI device.  The could be caused by a
6461  * Hot-Plug event, or because the driver is going to be removed from
6462  * memory.
6463  **/
6464 static void __devexit e1000_remove(struct pci_dev *pdev)
6465 {
6466         struct net_device *netdev = pci_get_drvdata(pdev);
6467         struct e1000_adapter *adapter = netdev_priv(netdev);
6468         bool down = test_bit(__E1000_DOWN, &adapter->state);
6469
6470         /*
6471          * The timers may be rescheduled, so explicitly disable them
6472          * from being rescheduled.
6473          */
6474         if (!down)
6475                 set_bit(__E1000_DOWN, &adapter->state);
6476         del_timer_sync(&adapter->watchdog_timer);
6477         del_timer_sync(&adapter->phy_info_timer);
6478
6479         cancel_work_sync(&adapter->reset_task);
6480         cancel_work_sync(&adapter->watchdog_task);
6481         cancel_work_sync(&adapter->downshift_task);
6482         cancel_work_sync(&adapter->update_phy_task);
6483         cancel_work_sync(&adapter->print_hang_task);
6484
6485         if (!(netdev->flags & IFF_UP))
6486                 e1000_power_down_phy(adapter);
6487
6488         /* Don't lie to e1000_close() down the road. */
6489         if (!down)
6490                 clear_bit(__E1000_DOWN, &adapter->state);
6491         unregister_netdev(netdev);
6492
6493         if (pci_dev_run_wake(pdev))
6494                 pm_runtime_get_noresume(&pdev->dev);
6495
6496         /*
6497          * Release control of h/w to f/w.  If f/w is AMT enabled, this
6498          * would have already happened in close and is redundant.
6499          */
6500         e1000e_release_hw_control(adapter);
6501
6502         e1000e_reset_interrupt_capability(adapter);
6503         kfree(adapter->tx_ring);
6504         kfree(adapter->rx_ring);
6505
6506         iounmap(adapter->hw.hw_addr);
6507         if (adapter->hw.flash_address)
6508                 iounmap(adapter->hw.flash_address);
6509         pci_release_selected_regions(pdev,
6510                                      pci_select_bars(pdev, IORESOURCE_MEM));
6511
6512         free_netdev(netdev);
6513
6514         /* AER disable */
6515         pci_disable_pcie_error_reporting(pdev);
6516
6517         pci_disable_device(pdev);
6518 }
6519
6520 /* PCI Error Recovery (ERS) */
6521 static struct pci_error_handlers e1000_err_handler = {
6522         .error_detected = e1000_io_error_detected,
6523         .slot_reset = e1000_io_slot_reset,
6524         .resume = e1000_io_resume,
6525 };
6526
6527 static DEFINE_PCI_DEVICE_TABLE(e1000_pci_tbl) = {
6528         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82571EB_COPPER), board_82571 },
6529         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82571EB_FIBER), board_82571 },
6530         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82571EB_QUAD_COPPER), board_82571 },
6531         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82571EB_QUAD_COPPER_LP), board_82571 },
6532         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82571EB_QUAD_FIBER), board_82571 },
6533         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82571EB_SERDES), board_82571 },
6534         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82571EB_SERDES_DUAL), board_82571 },
6535         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82571EB_SERDES_QUAD), board_82571 },
6536         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82571PT_QUAD_COPPER), board_82571 },
6537
6538         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82572EI), board_82572 },
6539         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82572EI_COPPER), board_82572 },
6540         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82572EI_FIBER), board_82572 },
6541         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82572EI_SERDES), board_82572 },
6542
6543         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82573E), board_82573 },
6544         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82573E_IAMT), board_82573 },
6545         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82573L), board_82573 },
6546
6547         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82574L), board_82574 },
6548         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82574LA), board_82574 },
6549         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82583V), board_82583 },
6550
6551         { PCI_VDEVICE(INTEL, E1000_DEV_ID_80003ES2LAN_COPPER_DPT),
6552           board_80003es2lan },
6553         { PCI_VDEVICE(INTEL, E1000_DEV_ID_80003ES2LAN_COPPER_SPT),
6554           board_80003es2lan },
6555         { PCI_VDEVICE(INTEL, E1000_DEV_ID_80003ES2LAN_SERDES_DPT),
6556           board_80003es2lan },
6557         { PCI_VDEVICE(INTEL, E1000_DEV_ID_80003ES2LAN_SERDES_SPT),
6558           board_80003es2lan },
6559
6560         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH8_IFE), board_ich8lan },
6561         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH8_IFE_G), board_ich8lan },
6562         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH8_IFE_GT), board_ich8lan },
6563         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH8_IGP_AMT), board_ich8lan },
6564         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH8_IGP_C), board_ich8lan },
6565         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH8_IGP_M), board_ich8lan },
6566         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH8_IGP_M_AMT), board_ich8lan },
6567         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH8_82567V_3), board_ich8lan },
6568
6569         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH9_IFE), board_ich9lan },
6570         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH9_IFE_G), board_ich9lan },
6571         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH9_IFE_GT), board_ich9lan },
6572         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH9_IGP_AMT), board_ich9lan },
6573         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH9_IGP_C), board_ich9lan },
6574         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH9_BM), board_ich9lan },
6575         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH9_IGP_M), board_ich9lan },
6576         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH9_IGP_M_AMT), board_ich9lan },
6577         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH9_IGP_M_V), board_ich9lan },
6578
6579         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH10_R_BM_LM), board_ich9lan },
6580         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH10_R_BM_LF), board_ich9lan },
6581         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH10_R_BM_V), board_ich9lan },
6582
6583         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH10_D_BM_LM), board_ich10lan },
6584         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH10_D_BM_LF), board_ich10lan },
6585         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH10_D_BM_V), board_ich10lan },
6586
6587         { PCI_VDEVICE(INTEL, E1000_DEV_ID_PCH_M_HV_LM), board_pchlan },
6588         { PCI_VDEVICE(INTEL, E1000_DEV_ID_PCH_M_HV_LC), board_pchlan },
6589         { PCI_VDEVICE(INTEL, E1000_DEV_ID_PCH_D_HV_DM), board_pchlan },
6590         { PCI_VDEVICE(INTEL, E1000_DEV_ID_PCH_D_HV_DC), board_pchlan },
6591
6592         { PCI_VDEVICE(INTEL, E1000_DEV_ID_PCH2_LV_LM), board_pch2lan },
6593         { PCI_VDEVICE(INTEL, E1000_DEV_ID_PCH2_LV_V), board_pch2lan },
6594
6595         { 0, 0, 0, 0, 0, 0, 0 } /* terminate list */
6596 };
6597 MODULE_DEVICE_TABLE(pci, e1000_pci_tbl);
6598
6599 #ifdef CONFIG_PM
6600 static const struct dev_pm_ops e1000_pm_ops = {
6601         SET_SYSTEM_SLEEP_PM_OPS(e1000_suspend, e1000_resume)
6602         SET_RUNTIME_PM_OPS(e1000_runtime_suspend,
6603                                 e1000_runtime_resume, e1000_idle)
6604 };
6605 #endif
6606
6607 /* PCI Device API Driver */
6608 static struct pci_driver e1000_driver = {
6609         .name     = e1000e_driver_name,
6610         .id_table = e1000_pci_tbl,
6611         .probe    = e1000_probe,
6612         .remove   = __devexit_p(e1000_remove),
6613 #ifdef CONFIG_PM
6614         .driver   = {
6615                 .pm = &e1000_pm_ops,
6616         },
6617 #endif
6618         .shutdown = e1000_shutdown,
6619         .err_handler = &e1000_err_handler
6620 };
6621
6622 /**
6623  * e1000_init_module - Driver Registration Routine
6624  *
6625  * e1000_init_module is the first routine called when the driver is
6626  * loaded. All it does is register with the PCI subsystem.
6627  **/
6628 static int __init e1000_init_module(void)
6629 {
6630         int ret;
6631         pr_info("Intel(R) PRO/1000 Network Driver - %s\n",
6632                 e1000e_driver_version);
6633         pr_info("Copyright(c) 1999 - 2012 Intel Corporation.\n");
6634         ret = pci_register_driver(&e1000_driver);
6635
6636         return ret;
6637 }
6638 module_init(e1000_init_module);
6639
6640 /**
6641  * e1000_exit_module - Driver Exit Cleanup Routine
6642  *
6643  * e1000_exit_module is called just before the driver is removed
6644  * from memory.
6645  **/
6646 static void __exit e1000_exit_module(void)
6647 {
6648         pci_unregister_driver(&e1000_driver);
6649 }
6650 module_exit(e1000_exit_module);
6651
6652
6653 MODULE_AUTHOR("Intel Corporation, <linux.nics@intel.com>");
6654 MODULE_DESCRIPTION("Intel(R) PRO/1000 Network Driver");
6655 MODULE_LICENSE("GPL");
6656 MODULE_VERSION(DRV_VERSION);
6657
6658 /* netdev.c */