drivers/net/e1000e/netdev.c

   1 /*******************************************************************************
   2
   3   Intel PRO/1000 Linux driver
   4   Copyright(c) 1999 - 2011 Intel Corporation.
   5
   6   This program is free software; you can redistribute it and/or modify it
   7   under the terms and conditions of the GNU General Public License,
   8   version 2, as published by the Free Software Foundation.
   9
  10   This program is distributed in the hope it will be useful, but WITHOUT
  11   ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
  12   FITNESS FOR A PARTICULAR PURPOSE.  See the GNU General Public License for
  13   more details.
  14
  15   You should have received a copy of the GNU General Public License along with
  16   this program; if not, write to the Free Software Foundation, Inc.,
  17   51 Franklin St - Fifth Floor, Boston, MA 02110-1301 USA.
  18
  19   The full GNU General Public License is included in this distribution in
  20   the file called "COPYING".
  21
  22   Contact Information:
  23   Linux NICS <linux.nics@intel.com>
  24   e1000-devel Mailing List <e1000-devel@lists.sourceforge.net>
  25   Intel Corporation, 5200 N.E. Elam Young Parkway, Hillsboro, OR 97124-6497
  26
  27 *******************************************************************************/
  28
  29 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
  30
  31 #include <linux/module.h>
  32 #include <linux/types.h>
  33 #include <linux/init.h>
  34 #include <linux/pci.h>
  35 #include <linux/vmalloc.h>
  36 #include <linux/pagemap.h>
  37 #include <linux/delay.h>
  38 #include <linux/netdevice.h>
  39 #include <linux/tcp.h>
  40 #include <linux/ipv6.h>
  41 #include <linux/slab.h>
  42 #include <net/checksum.h>
  43 #include <net/ip6_checksum.h>
  44 #include <linux/mii.h>
  45 #include <linux/ethtool.h>
  46 #include <linux/if_vlan.h>
  47 #include <linux/cpu.h>
  48 #include <linux/smp.h>
  49 #include <linux/pm_qos_params.h>
  50 #include <linux/pm_runtime.h>
  51 #include <linux/aer.h>
  52
  53 #include "e1000.h"
  54
  55 #define DRV_EXTRAVERSION "-k2"
  56
  57 #define DRV_VERSION "1.2.20" DRV_EXTRAVERSION
  58 char e1000e_driver_name[] = "e1000e";
  59 const char e1000e_driver_version[] = DRV_VERSION;
  60
  61 static const struct e1000_info *e1000_info_tbl[] = {
  62         [board_82571]           = &e1000_82571_info,
  63         [board_82572]           = &e1000_82572_info,
  64         [board_82573]           = &e1000_82573_info,
  65         [board_82574]           = &e1000_82574_info,
  66         [board_82583]           = &e1000_82583_info,
  67         [board_80003es2lan]     = &e1000_es2_info,
  68         [board_ich8lan]         = &e1000_ich8_info,
  69         [board_ich9lan]         = &e1000_ich9_info,
  70         [board_ich10lan]        = &e1000_ich10_info,
  71         [board_pchlan]          = &e1000_pch_info,
  72         [board_pch2lan]         = &e1000_pch2_info,
  73 };
  74
  75 struct e1000_reg_info {
  76         u32 ofs;
  77         char *name;
  78 };
  79
  80 #define E1000_RDFH      0x02410 /* Rx Data FIFO Head - RW */
  81 #define E1000_RDFT      0x02418 /* Rx Data FIFO Tail - RW */
  82 #define E1000_RDFHS     0x02420 /* Rx Data FIFO Head Saved - RW */
  83 #define E1000_RDFTS     0x02428 /* Rx Data FIFO Tail Saved - RW */
  84 #define E1000_RDFPC     0x02430 /* Rx Data FIFO Packet Count - RW */
  85
  86 #define E1000_TDFH      0x03410 /* Tx Data FIFO Head - RW */
  87 #define E1000_TDFT      0x03418 /* Tx Data FIFO Tail - RW */
  88 #define E1000_TDFHS     0x03420 /* Tx Data FIFO Head Saved - RW */
  89 #define E1000_TDFTS     0x03428 /* Tx Data FIFO Tail Saved - RW */
  90 #define E1000_TDFPC     0x03430 /* Tx Data FIFO Packet Count - RW */
  91
  92 static const struct e1000_reg_info e1000_reg_info_tbl[] = {
  93
  94         /* General Registers */
  95         {E1000_CTRL, "CTRL"},
  96         {E1000_STATUS, "STATUS"},
  97         {E1000_CTRL_EXT, "CTRL_EXT"},
  98
  99         /* Interrupt Registers */
 100         {E1000_ICR, "ICR"},
 101
 102         /* Rx Registers */
 103         {E1000_RCTL, "RCTL"},
 104         {E1000_RDLEN, "RDLEN"},
 105         {E1000_RDH, "RDH"},
 106         {E1000_RDT, "RDT"},
 107         {E1000_RDTR, "RDTR"},
 108         {E1000_RXDCTL(0), "RXDCTL"},
 109         {E1000_ERT, "ERT"},
 110         {E1000_RDBAL, "RDBAL"},
 111         {E1000_RDBAH, "RDBAH"},
 112         {E1000_RDFH, "RDFH"},
 113         {E1000_RDFT, "RDFT"},
 114         {E1000_RDFHS, "RDFHS"},
 115         {E1000_RDFTS, "RDFTS"},
 116         {E1000_RDFPC, "RDFPC"},
 117
 118         /* Tx Registers */
 119         {E1000_TCTL, "TCTL"},
 120         {E1000_TDBAL, "TDBAL"},
 121         {E1000_TDBAH, "TDBAH"},
 122         {E1000_TDLEN, "TDLEN"},
 123         {E1000_TDH, "TDH"},
 124         {E1000_TDT, "TDT"},
 125         {E1000_TIDV, "TIDV"},
 126         {E1000_TXDCTL(0), "TXDCTL"},
 127         {E1000_TADV, "TADV"},
 128         {E1000_TARC(0), "TARC"},
 129         {E1000_TDFH, "TDFH"},
 130         {E1000_TDFT, "TDFT"},
 131         {E1000_TDFHS, "TDFHS"},
 132         {E1000_TDFTS, "TDFTS"},
 133         {E1000_TDFPC, "TDFPC"},
 134
 135         /* List Terminator */
 136         {}
 137 };
 138
 139 /*
 140  * e1000_regdump - register printout routine
 141  */
 142 static void e1000_regdump(struct e1000_hw *hw, struct e1000_reg_info *reginfo)
 143 {
 144         int n = 0;
 145         char rname[16];
 146         u32 regs[8];
 147
 148         switch (reginfo->ofs) {
 149         case E1000_RXDCTL(0):
 150                 for (n = 0; n < 2; n++)
 151                         regs[n] = __er32(hw, E1000_RXDCTL(n));
 152                 break;
 153         case E1000_TXDCTL(0):
 154                 for (n = 0; n < 2; n++)
 155                         regs[n] = __er32(hw, E1000_TXDCTL(n));
 156                 break;
 157         case E1000_TARC(0):
 158                 for (n = 0; n < 2; n++)
 159                         regs[n] = __er32(hw, E1000_TARC(n));
 160                 break;
 161         default:
 162                 printk(KERN_INFO "%-15s %08x\n",
 163                        reginfo->name, __er32(hw, reginfo->ofs));
 164                 return;
 165         }
 166
 167         snprintf(rname, 16, "%s%s", reginfo->name, "[0-1]");
 168         printk(KERN_INFO "%-15s ", rname);
 169         for (n = 0; n < 2; n++)
 170                 printk(KERN_CONT "%08x ", regs[n]);
 171         printk(KERN_CONT "\n");
 172 }
 173
 174 /*
 175  * e1000e_dump - Print registers, Tx-ring and Rx-ring
 176  */
 177 static void e1000e_dump(struct e1000_adapter *adapter)
 178 {
 179         struct net_device *netdev = adapter->netdev;
 180         struct e1000_hw *hw = &adapter->hw;
 181         struct e1000_reg_info *reginfo;
 182         struct e1000_ring *tx_ring = adapter->tx_ring;
 183         struct e1000_tx_desc *tx_desc;
 184         struct my_u0 {
 185                 u64 a;
 186                 u64 b;
 187         } *u0;
 188         struct e1000_buffer *buffer_info;
 189         struct e1000_ring *rx_ring = adapter->rx_ring;
 190         union e1000_rx_desc_packet_split *rx_desc_ps;
 191         struct e1000_rx_desc *rx_desc;
 192         struct my_u1 {
 193                 u64 a;
 194                 u64 b;
 195                 u64 c;
 196                 u64 d;
 197         } *u1;
 198         u32 staterr;
 199         int i = 0;
 200
 201         if (!netif_msg_hw(adapter))
 202                 return;
 203
 204         /* Print netdevice Info */
 205         if (netdev) {
 206                 dev_info(&adapter->pdev->dev, "Net device Info\n");
 207                 printk(KERN_INFO "Device Name     state            "
 208                        "trans_start      last_rx\n");
 209                 printk(KERN_INFO "%-15s %016lX %016lX %016lX\n",
 210                        netdev->name, netdev->state, netdev->trans_start,
 211                        netdev->last_rx);
 212         }
 213
 214         /* Print Registers */
 215         dev_info(&adapter->pdev->dev, "Register Dump\n");
 216         printk(KERN_INFO " Register Name   Value\n");
 217         for (reginfo = (struct e1000_reg_info *)e1000_reg_info_tbl;
 218              reginfo->name; reginfo++) {
 219                 e1000_regdump(hw, reginfo);
 220         }
 221
 222         /* Print Tx Ring Summary */
 223         if (!netdev || !netif_running(netdev))
 224                 goto exit;
 225
 226         dev_info(&adapter->pdev->dev, "Tx Ring Summary\n");
 227         printk(KERN_INFO "Queue [NTU] [NTC] [bi(ntc)->dma  ]"
 228                " leng ntw timestamp\n");
 229         buffer_info = &tx_ring->buffer_info[tx_ring->next_to_clean];
 230         printk(KERN_INFO " %5d %5X %5X %016llX %04X %3X %016llX\n",
 231                0, tx_ring->next_to_use, tx_ring->next_to_clean,
 232                (unsigned long long)buffer_info->dma,
 233                buffer_info->length,
 234                buffer_info->next_to_watch,
 235                (unsigned long long)buffer_info->time_stamp);
 236
 237         /* Print Tx Ring */
 238         if (!netif_msg_tx_done(adapter))
 239                 goto rx_ring_summary;
 240
 241         dev_info(&adapter->pdev->dev, "Tx Ring Dump\n");
 242
 243         /* Transmit Descriptor Formats - DEXT[29] is 0 (Legacy) or 1 (Extended)
 244          *
 245          * Legacy Transmit Descriptor
 246          *   +--------------------------------------------------------------+
 247          * 0 |         Buffer Address [63:0] (Reserved on Write Back)       |
 248          *   +--------------------------------------------------------------+
 249          * 8 | Special  |    CSS     | Status |  CMD    |  CSO   |  Length  |
 250          *   +--------------------------------------------------------------+
 251          *   63       48 47        36 35    32 31     24 23    16 15        0
 252          *
 253          * Extended Context Descriptor (DTYP=0x0) for TSO or checksum offload
 254          *   63      48 47    40 39       32 31             16 15    8 7      0
 255          *   +----------------------------------------------------------------+
 256          * 0 |  TUCSE  | TUCS0  |   TUCSS   |     IPCSE       | IPCS0 | IPCSS |
 257          *   +----------------------------------------------------------------+
 258          * 8 |   MSS   | HDRLEN | RSV | STA | TUCMD | DTYP |      PAYLEN      |
 259          *   +----------------------------------------------------------------+
 260          *   63      48 47    40 39 36 35 32 31   24 23  20 19                0
 261          *
 262          * Extended Data Descriptor (DTYP=0x1)
 263          *   +----------------------------------------------------------------+
 264          * 0 |                     Buffer Address [63:0]                      |
 265          *   +----------------------------------------------------------------+
 266          * 8 | VLAN tag |  POPTS  | Rsvd | Status | Command | DTYP |  DTALEN  |
 267          *   +----------------------------------------------------------------+
 268          *   63       48 47     40 39  36 35    32 31     24 23  20 19        0
 269          */
 270         printk(KERN_INFO "Tl[desc]     [address 63:0  ] [SpeCssSCmCsLen]"
 271                " [bi->dma       ] leng  ntw timestamp        bi->skb "
 272                "<-- Legacy format\n");
 273         printk(KERN_INFO "Tc[desc]     [Ce CoCsIpceCoS] [MssHlRSCm0Plen]"
 274                " [bi->dma       ] leng  ntw timestamp        bi->skb "
 275                "<-- Ext Context format\n");
 276         printk(KERN_INFO "Td[desc]     [address 63:0  ] [VlaPoRSCm1Dlen]"
 277                " [bi->dma       ] leng  ntw timestamp        bi->skb "
 278                "<-- Ext Data format\n");
 279         for (i = 0; tx_ring->desc && (i < tx_ring->count); i++) {
 280                 tx_desc = E1000_TX_DESC(*tx_ring, i);
 281                 buffer_info = &tx_ring->buffer_info[i];
 282                 u0 = (struct my_u0 *)tx_desc;
 283                 printk(KERN_INFO "T%c[0x%03X]    %016llX %016llX %016llX "
 284                        "%04X  %3X %016llX %p",
 285                        (!(le64_to_cpu(u0->b) & (1 << 29)) ? 'l' :
 286                         ((le64_to_cpu(u0->b) & (1 << 20)) ? 'd' : 'c')), i,
 287                        (unsigned long long)le64_to_cpu(u0->a),
 288                        (unsigned long long)le64_to_cpu(u0->b),
 289                        (unsigned long long)buffer_info->dma,
 290                        buffer_info->length, buffer_info->next_to_watch,
 291                        (unsigned long long)buffer_info->time_stamp,
 292                        buffer_info->skb);
 293                 if (i == tx_ring->next_to_use && i == tx_ring->next_to_clean)
 294                         printk(KERN_CONT " NTC/U\n");
 295                 else if (i == tx_ring->next_to_use)
 296                         printk(KERN_CONT " NTU\n");
 297                 else if (i == tx_ring->next_to_clean)
 298                         printk(KERN_CONT " NTC\n");
 299                 else
 300                         printk(KERN_CONT "\n");
 301
 302                 if (netif_msg_pktdata(adapter) && buffer_info->dma != 0)
 303                         print_hex_dump(KERN_INFO, "", DUMP_PREFIX_ADDRESS,
 304                                        16, 1, phys_to_virt(buffer_info->dma),
 305                                        buffer_info->length, true);
 306         }
 307
 308         /* Print Rx Ring Summary */
 309 rx_ring_summary:
 310         dev_info(&adapter->pdev->dev, "Rx Ring Summary\n");
 311         printk(KERN_INFO "Queue [NTU] [NTC]\n");
 312         printk(KERN_INFO " %5d %5X %5X\n", 0,
 313                rx_ring->next_to_use, rx_ring->next_to_clean);
 314
 315         /* Print Rx Ring */
 316         if (!netif_msg_rx_status(adapter))
 317                 goto exit;
 318
 319         dev_info(&adapter->pdev->dev, "Rx Ring Dump\n");
 320         switch (adapter->rx_ps_pages) {
 321         case 1:
 322         case 2:
 323         case 3:
 324                 /* [Extended] Packet Split Receive Descriptor Format
 325                  *
 326                  *    +-----------------------------------------------------+
 327                  *  0 |                Buffer Address 0 [63:0]              |
 328                  *    +-----------------------------------------------------+
 329                  *  8 |                Buffer Address 1 [63:0]              |
 330                  *    +-----------------------------------------------------+
 331                  * 16 |                Buffer Address 2 [63:0]              |
 332                  *    +-----------------------------------------------------+
 333                  * 24 |                Buffer Address 3 [63:0]              |
 334                  *    +-----------------------------------------------------+
 335                  */
 336                 printk(KERN_INFO "R  [desc]      [buffer 0 63:0 ] "
 337                        "[buffer 1 63:0 ] "
 338                        "[buffer 2 63:0 ] [buffer 3 63:0 ] [bi->dma       ] "
 339                        "[bi->skb] <-- Ext Pkt Split format\n");
 340                 /* [Extended] Receive Descriptor (Write-Back) Format
 341                  *
 342                  *   63       48 47    32 31     13 12    8 7    4 3        0
 343                  *   +------------------------------------------------------+
 344                  * 0 | Packet   | IP     |  Rsvd   | MRQ   | Rsvd | MRQ RSS |
 345                  *   | Checksum | Ident  |         | Queue |      |  Type   |
 346                  *   +------------------------------------------------------+
 347                  * 8 | VLAN Tag | Length | Extended Error | Extended Status |
 348                  *   +------------------------------------------------------+
 349                  *   63       48 47    32 31            20 19               0
 350                  */
 351                 printk(KERN_INFO "RWB[desc]      [ck ipid mrqhsh] "
 352                        "[vl   l0 ee  es] "
 353                        "[ l3  l2  l1 hs] [reserved      ] ---------------- "
 354                        "[bi->skb] <-- Ext Rx Write-Back format\n");
 355                 for (i = 0; i < rx_ring->count; i++) {
 356                         buffer_info = &rx_ring->buffer_info[i];
 357                         rx_desc_ps = E1000_RX_DESC_PS(*rx_ring, i);
 358                         u1 = (struct my_u1 *)rx_desc_ps;
 359                         staterr =
 360                             le32_to_cpu(rx_desc_ps->wb.middle.status_error);
 361                         if (staterr & E1000_RXD_STAT_DD) {
 362                                 /* Descriptor Done */
 363                                 printk(KERN_INFO "RWB[0x%03X]     %016llX "
 364                                        "%016llX %016llX %016llX "
 365                                        "---------------- %p", i,
 366                                        (unsigned long long)le64_to_cpu(u1->a),
 367                                        (unsigned long long)le64_to_cpu(u1->b),
 368                                        (unsigned long long)le64_to_cpu(u1->c),
 369                                        (unsigned long long)le64_to_cpu(u1->d),
 370                                        buffer_info->skb);
 371                         } else {
 372                                 printk(KERN_INFO "R  [0x%03X]     %016llX "
 373                                        "%016llX %016llX %016llX %016llX %p", i,
 374                                        (unsigned long long)le64_to_cpu(u1->a),
 375                                        (unsigned long long)le64_to_cpu(u1->b),
 376                                        (unsigned long long)le64_to_cpu(u1->c),
 377                                        (unsigned long long)le64_to_cpu(u1->d),
 378                                        (unsigned long long)buffer_info->dma,
 379                                        buffer_info->skb);
 380
 381                                 if (netif_msg_pktdata(adapter))
 382                                         print_hex_dump(KERN_INFO, "",
 383                                                 DUMP_PREFIX_ADDRESS, 16, 1,
 384                                                 phys_to_virt(buffer_info->dma),
 385                                                 adapter->rx_ps_bsize0, true);
 386                         }
 387
 388                         if (i == rx_ring->next_to_use)
 389                                 printk(KERN_CONT " NTU\n");
 390                         else if (i == rx_ring->next_to_clean)
 391                                 printk(KERN_CONT " NTC\n");
 392                         else
 393                                 printk(KERN_CONT "\n");
 394                 }
 395                 break;
 396         default:
 397         case 0:
 398                 /* Legacy Receive Descriptor Format
 399                  *
 400                  * +-----------------------------------------------------+
 401                  * |                Buffer Address [63:0]                |
 402                  * +-----------------------------------------------------+
 403                  * | VLAN Tag | Errors | Status 0 | Packet csum | Length |
 404                  * +-----------------------------------------------------+
 405                  * 63       48 47    40 39      32 31         16 15      0
 406                  */
 407                 printk(KERN_INFO "Rl[desc]     [address 63:0  ] "
 408                        "[vl er S cks ln] [bi->dma       ] [bi->skb] "
 409                        "<-- Legacy format\n");
 410                 for (i = 0; rx_ring->desc && (i < rx_ring->count); i++) {
 411                         rx_desc = E1000_RX_DESC(*rx_ring, i);
 412                         buffer_info = &rx_ring->buffer_info[i];
 413                         u0 = (struct my_u0 *)rx_desc;
 414                         printk(KERN_INFO "Rl[0x%03X]    %016llX %016llX "
 415                                "%016llX %p", i,
 416                                (unsigned long long)le64_to_cpu(u0->a),
 417                                (unsigned long long)le64_to_cpu(u0->b),
 418                                (unsigned long long)buffer_info->dma,
 419                                buffer_info->skb);
 420                         if (i == rx_ring->next_to_use)
 421                                 printk(KERN_CONT " NTU\n");
 422                         else if (i == rx_ring->next_to_clean)
 423                                 printk(KERN_CONT " NTC\n");
 424                         else
 425                                 printk(KERN_CONT "\n");
 426
 427                         if (netif_msg_pktdata(adapter))
 428                                 print_hex_dump(KERN_INFO, "",
 429                                                DUMP_PREFIX_ADDRESS,
 430                                                16, 1,
 431                                                phys_to_virt(buffer_info->dma),
 432                                                adapter->rx_buffer_len, true);
 433                 }
 434         }
 435
 436 exit:
 437         return;
 438 }
 439
 440 /**
 441  * e1000_desc_unused - calculate if we have unused descriptors
 442  **/
 443 static int e1000_desc_unused(struct e1000_ring *ring)
 444 {
 445         if (ring->next_to_clean > ring->next_to_use)
 446                 return ring->next_to_clean - ring->next_to_use - 1;
 447
 448         return ring->count + ring->next_to_clean - ring->next_to_use - 1;
 449 }
 450
 451 /**
 452  * e1000_receive_skb - helper function to handle Rx indications
 453  * @adapter: board private structure
 454  * @status: descriptor status field as written by hardware
 455  * @vlan: descriptor vlan field as written by hardware (no le/be conversion)
 456  * @skb: pointer to sk_buff to be indicated to stack
 457  **/
 458 static void e1000_receive_skb(struct e1000_adapter *adapter,
 459                               struct net_device *netdev, struct sk_buff *skb,
 460                               u8 status, __le16 vlan)
 461 {
 462         skb->protocol = eth_type_trans(skb, netdev);
 463
 464         if (adapter->vlgrp && (status & E1000_RXD_STAT_VP))
 465                 vlan_gro_receive(&adapter->napi, adapter->vlgrp,
 466                                  le16_to_cpu(vlan), skb);
 467         else
 468                 napi_gro_receive(&adapter->napi, skb);
 469 }
 470
 471 /**
 472  * e1000_rx_checksum - Receive Checksum Offload
 473  * @adapter:     board private structure
 474  * @status_err:  receive descriptor status and error fields
 475  * @csum:       receive descriptor csum field
 476  * @sk_buff:     socket buffer with received data
 477  **/
 478 static void e1000_rx_checksum(struct e1000_adapter *adapter, u32 status_err,
 479                               u32 csum, struct sk_buff *skb)
 480 {
 481         u16 status = (u16)status_err;
 482         u8 errors = (u8)(status_err >> 24);
 483
 484         skb_checksum_none_assert(skb);
 485
 486         /* Ignore Checksum bit is set */
 487         if (status & E1000_RXD_STAT_IXSM)
 488                 return;
 489         /* TCP/UDP checksum error bit is set */
 490         if (errors & E1000_RXD_ERR_TCPE) {
 491                 /* let the stack verify checksum errors */
 492                 adapter->hw_csum_err++;
 493                 return;
 494         }
 495
 496         /* TCP/UDP Checksum has not been calculated */
 497         if (!(status & (E1000_RXD_STAT_TCPCS | E1000_RXD_STAT_UDPCS)))
 498                 return;
 499
 500         /* It must be a TCP or UDP packet with a valid checksum */
 501         if (status & E1000_RXD_STAT_TCPCS) {
 502                 /* TCP checksum is good */
 503                 skb->ip_summed = CHECKSUM_UNNECESSARY;
 504         } else {
 505                 /*
 506                  * IP fragment with UDP payload
 507                  * Hardware complements the payload checksum, so we undo it
 508                  * and then put the value in host order for further stack use.
 509                  */
 510                 __sum16 sum = (__force __sum16)htons(csum);
 511                 skb->csum = csum_unfold(~sum);
 512                 skb->ip_summed = CHECKSUM_COMPLETE;
 513         }
 514         adapter->hw_csum_good++;
 515 }
 516
 517 /**
 518  * e1000_alloc_rx_buffers - Replace used receive buffers; legacy & extended
 519  * @adapter: address of board private structure
 520  **/
 521 static void e1000_alloc_rx_buffers(struct e1000_adapter *adapter,
 522                                    int cleaned_count)
 523 {
 524         struct net_device *netdev = adapter->netdev;
 525         struct pci_dev *pdev = adapter->pdev;
 526         struct e1000_ring *rx_ring = adapter->rx_ring;
 527         struct e1000_rx_desc *rx_desc;
 528         struct e1000_buffer *buffer_info;
 529         struct sk_buff *skb;
 530         unsigned int i;
 531         unsigned int bufsz = adapter->rx_buffer_len;
 532
 533         i = rx_ring->next_to_use;
 534         buffer_info = &rx_ring->buffer_info[i];
 535
 536         while (cleaned_count--) {
 537                 skb = buffer_info->skb;
 538                 if (skb) {
 539                         skb_trim(skb, 0);
 540                         goto map_skb;
 541                 }
 542
 543                 skb = netdev_alloc_skb_ip_align(netdev, bufsz);
 544                 if (!skb) {
 545                         /* Better luck next round */
 546                         adapter->alloc_rx_buff_failed++;
 547                         break;
 548                 }
 549
 550                 buffer_info->skb = skb;
 551 map_skb:
 552                 buffer_info->dma = dma_map_single(&pdev->dev, skb->data,
 553                                                   adapter->rx_buffer_len,
 554                                                   DMA_FROM_DEVICE);
 555                 if (dma_mapping_error(&pdev->dev, buffer_info->dma)) {
 556                         dev_err(&pdev->dev, "Rx DMA map failed\n");
 557                         adapter->rx_dma_failed++;
 558                         break;
 559                 }
 560
 561                 rx_desc = E1000_RX_DESC(*rx_ring, i);
 562                 rx_desc->buffer_addr = cpu_to_le64(buffer_info->dma);
 563
 564                 if (unlikely(!(i & (E1000_RX_BUFFER_WRITE - 1)))) {
 565                         /*
 566                          * Force memory writes to complete before letting h/w
 567                          * know there are new descriptors to fetch.  (Only
 568                          * applicable for weak-ordered memory model archs,
 569                          * such as IA-64).
 570                          */
 571                         wmb();
 572                         writel(i, adapter->hw.hw_addr + rx_ring->tail);
 573                 }
 574                 i++;
 575                 if (i == rx_ring->count)
 576                         i = 0;
 577                 buffer_info = &rx_ring->buffer_info[i];
 578         }
 579
 580         rx_ring->next_to_use = i;
 581 }
 582
 583 /**
 584  * e1000_alloc_rx_buffers_ps - Replace used receive buffers; packet split
 585  * @adapter: address of board private structure
 586  **/
 587 static void e1000_alloc_rx_buffers_ps(struct e1000_adapter *adapter,
 588                                       int cleaned_count)
 589 {
 590         struct net_device *netdev = adapter->netdev;
 591         struct pci_dev *pdev = adapter->pdev;
 592         union e1000_rx_desc_packet_split *rx_desc;
 593         struct e1000_ring *rx_ring = adapter->rx_ring;
 594         struct e1000_buffer *buffer_info;
 595         struct e1000_ps_page *ps_page;
 596         struct sk_buff *skb;
 597         unsigned int i, j;
 598
 599         i = rx_ring->next_to_use;
 600         buffer_info = &rx_ring->buffer_info[i];
 601
 602         while (cleaned_count--) {
 603                 rx_desc = E1000_RX_DESC_PS(*rx_ring, i);
 604
 605                 for (j = 0; j < PS_PAGE_BUFFERS; j++) {
 606                         ps_page = &buffer_info->ps_pages[j];
 607                         if (j >= adapter->rx_ps_pages) {
 608                                 /* all unused desc entries get hw null ptr */
 609                                 rx_desc->read.buffer_addr[j + 1] =
 610                                     ~cpu_to_le64(0);
 611                                 continue;
 612                         }
 613                         if (!ps_page->page) {
 614                                 ps_page->page = alloc_page(GFP_ATOMIC);
 615                                 if (!ps_page->page) {
 616                                         adapter->alloc_rx_buff_failed++;
 617                                         goto no_buffers;
 618                                 }
 619                                 ps_page->dma = dma_map_page(&pdev->dev,
 620                                                             ps_page->page,
 621                                                             0, PAGE_SIZE,
 622                                                             DMA_FROM_DEVICE);
 623                                 if (dma_mapping_error(&pdev->dev,
 624                                                       ps_page->dma)) {
 625                                         dev_err(&adapter->pdev->dev,
 626                                                 "Rx DMA page map failed\n");
 627                                         adapter->rx_dma_failed++;
 628                                         goto no_buffers;
 629                                 }
 630                         }
 631                         /*
 632                          * Refresh the desc even if buffer_addrs
 633                          * didn't change because each write-back
 634                          * erases this info.
 635                          */
 636                         rx_desc->read.buffer_addr[j + 1] =
 637                             cpu_to_le64(ps_page->dma);
 638                 }
 639
 640                 skb = netdev_alloc_skb_ip_align(netdev,
 641                                                 adapter->rx_ps_bsize0);
 642
 643                 if (!skb) {
 644                         adapter->alloc_rx_buff_failed++;
 645                         break;
 646                 }
 647
 648                 buffer_info->skb = skb;
 649                 buffer_info->dma = dma_map_single(&pdev->dev, skb->data,
 650                                                   adapter->rx_ps_bsize0,
 651                                                   DMA_FROM_DEVICE);
 652                 if (dma_mapping_error(&pdev->dev, buffer_info->dma)) {
 653                         dev_err(&pdev->dev, "Rx DMA map failed\n");
 654                         adapter->rx_dma_failed++;
 655                         /* cleanup skb */
 656                         dev_kfree_skb_any(skb);
 657                         buffer_info->skb = NULL;
 658                         break;
 659                 }
 660
 661                 rx_desc->read.buffer_addr[0] = cpu_to_le64(buffer_info->dma);
 662
 663                 if (unlikely(!(i & (E1000_RX_BUFFER_WRITE - 1)))) {
 664                         /*
 665                          * Force memory writes to complete before letting h/w
 666                          * know there are new descriptors to fetch.  (Only
 667                          * applicable for weak-ordered memory model archs,
 668                          * such as IA-64).
 669                          */
 670                         wmb();
 671                         writel(i << 1, adapter->hw.hw_addr + rx_ring->tail);
 672                 }
 673
 674                 i++;
 675                 if (i == rx_ring->count)
 676                         i = 0;
 677                 buffer_info = &rx_ring->buffer_info[i];
 678         }
 679
 680 no_buffers:
 681         rx_ring->next_to_use = i;
 682 }
 683
 684 /**
 685  * e1000_alloc_jumbo_rx_buffers - Replace used jumbo receive buffers
 686  * @adapter: address of board private structure
 687  * @cleaned_count: number of buffers to allocate this pass
 688  **/
 689
 690 static void e1000_alloc_jumbo_rx_buffers(struct e1000_adapter *adapter,
 691                                          int cleaned_count)
 692 {
 693         struct net_device *netdev = adapter->netdev;
 694         struct pci_dev *pdev = adapter->pdev;
 695         struct e1000_rx_desc *rx_desc;
 696         struct e1000_ring *rx_ring = adapter->rx_ring;
 697         struct e1000_buffer *buffer_info;
 698         struct sk_buff *skb;
 699         unsigned int i;
 700         unsigned int bufsz = 256 - 16 /* for skb_reserve */;
 701
 702         i = rx_ring->next_to_use;
 703         buffer_info = &rx_ring->buffer_info[i];
 704
 705         while (cleaned_count--) {
 706                 skb = buffer_info->skb;
 707                 if (skb) {
 708                         skb_trim(skb, 0);
 709                         goto check_page;
 710                 }
 711
 712                 skb = netdev_alloc_skb_ip_align(netdev, bufsz);
 713                 if (unlikely(!skb)) {
 714                         /* Better luck next round */
 715                         adapter->alloc_rx_buff_failed++;
 716                         break;
 717                 }
 718
 719                 buffer_info->skb = skb;
 720 check_page:
 721                 /* allocate a new page if necessary */
 722                 if (!buffer_info->page) {
 723                         buffer_info->page = alloc_page(GFP_ATOMIC);
 724                         if (unlikely(!buffer_info->page)) {
 725                                 adapter->alloc_rx_buff_failed++;
 726                                 break;
 727                         }
 728                 }
 729
 730                 if (!buffer_info->dma)
 731                         buffer_info->dma = dma_map_page(&pdev->dev,
 732                                                         buffer_info->page, 0,
 733                                                         PAGE_SIZE,
 734                                                         DMA_FROM_DEVICE);
 735
 736                 rx_desc = E1000_RX_DESC(*rx_ring, i);
 737                 rx_desc->buffer_addr = cpu_to_le64(buffer_info->dma);
 738
 739                 if (unlikely(++i == rx_ring->count))
 740                         i = 0;
 741                 buffer_info = &rx_ring->buffer_info[i];
 742         }
 743
 744         if (likely(rx_ring->next_to_use != i)) {
 745                 rx_ring->next_to_use = i;
 746                 if (unlikely(i-- == 0))
 747                         i = (rx_ring->count - 1);
 748
 749                 /* Force memory writes to complete before letting h/w
 750                  * know there are new descriptors to fetch.  (Only
 751                  * applicable for weak-ordered memory model archs,
 752                  * such as IA-64). */
 753                 wmb();
 754                 writel(i, adapter->hw.hw_addr + rx_ring->tail);
 755         }
 756 }
 757
 758 /**
 759  * e1000_clean_rx_irq - Send received data up the network stack; legacy
 760  * @adapter: board private structure
 761  *
 762  * the return value indicates whether actual cleaning was done, there
 763  * is no guarantee that everything was cleaned
 764  **/
 765 static bool e1000_clean_rx_irq(struct e1000_adapter *adapter,
 766                                int *work_done, int work_to_do)
 767 {
 768         struct net_device *netdev = adapter->netdev;
 769         struct pci_dev *pdev = adapter->pdev;
 770         struct e1000_hw *hw = &adapter->hw;
 771         struct e1000_ring *rx_ring = adapter->rx_ring;
 772         struct e1000_rx_desc *rx_desc, *next_rxd;
 773         struct e1000_buffer *buffer_info, *next_buffer;
 774         u32 length;
 775         unsigned int i;
 776         int cleaned_count = 0;
 777         bool cleaned = 0;
 778         unsigned int total_rx_bytes = 0, total_rx_packets = 0;
 779
 780         i = rx_ring->next_to_clean;
 781         rx_desc = E1000_RX_DESC(*rx_ring, i);
 782         buffer_info = &rx_ring->buffer_info[i];
 783
 784         while (rx_desc->status & E1000_RXD_STAT_DD) {
 785                 struct sk_buff *skb;
 786                 u8 status;
 787
 788                 if (*work_done >= work_to_do)
 789                         break;
 790                 (*work_done)++;
 791                 rmb();  /* read descriptor and rx_buffer_info after status DD */
 792
 793                 status = rx_desc->status;
 794                 skb = buffer_info->skb;
 795                 buffer_info->skb = NULL;
 796
 797                 prefetch(skb->data - NET_IP_ALIGN);
 798
 799                 i++;
 800                 if (i == rx_ring->count)
 801                         i = 0;
 802                 next_rxd = E1000_RX_DESC(*rx_ring, i);
 803                 prefetch(next_rxd);
 804
 805                 next_buffer = &rx_ring->buffer_info[i];
 806
 807                 cleaned = 1;
 808                 cleaned_count++;
 809                 dma_unmap_single(&pdev->dev,
 810                                  buffer_info->dma,
 811                                  adapter->rx_buffer_len,
 812                                  DMA_FROM_DEVICE);
 813                 buffer_info->dma = 0;
 814
 815                 length = le16_to_cpu(rx_desc->length);
 816
 817                 /*
 818                  * !EOP means multiple descriptors were used to store a single
 819                  * packet, if that's the case we need to toss it.  In fact, we
 820                  * need to toss every packet with the EOP bit clear and the
 821                  * next frame that _does_ have the EOP bit set, as it is by
 822                  * definition only a frame fragment
 823                  */
 824                 if (unlikely(!(status & E1000_RXD_STAT_EOP)))
 825                         adapter->flags2 |= FLAG2_IS_DISCARDING;
 826
 827                 if (adapter->flags2 & FLAG2_IS_DISCARDING) {
 828                         /* All receives must fit into a single buffer */
 829                         e_dbg("Receive packet consumed multiple buffers\n");
 830                         /* recycle */
 831                         buffer_info->skb = skb;
 832                         if (status & E1000_RXD_STAT_EOP)
 833                                 adapter->flags2 &= ~FLAG2_IS_DISCARDING;
 834                         goto next_desc;
 835                 }
 836
 837                 if (rx_desc->errors & E1000_RXD_ERR_FRAME_ERR_MASK) {
 838                         /* recycle */
 839                         buffer_info->skb = skb;
 840                         goto next_desc;
 841                 }
 842
 843                 /* adjust length to remove Ethernet CRC */
 844                 if (!(adapter->flags2 & FLAG2_CRC_STRIPPING))
 845                         length -= 4;
 846
 847                 total_rx_bytes += length;
 848                 total_rx_packets++;
 849
 850                 /*
 851                  * code added for copybreak, this should improve
 852                  * performance for small packets with large amounts
 853                  * of reassembly being done in the stack
 854                  */
 855                 if (length < copybreak) {
 856                         struct sk_buff *new_skb =
 857                             netdev_alloc_skb_ip_align(netdev, length);
 858                         if (new_skb) {
 859                                 skb_copy_to_linear_data_offset(new_skb,
 860                                                                -NET_IP_ALIGN,
 861                                                                (skb->data -
 862                                                                 NET_IP_ALIGN),
 863                                                                (length +
 864                                                                 NET_IP_ALIGN));
 865                                 /* save the skb in buffer_info as good */
 866                                 buffer_info->skb = skb;
 867                                 skb = new_skb;
 868                         }
 869                         /* else just continue with the old one */
 870                 }
 871                 /* end copybreak code */
 872                 skb_put(skb, length);
 873
 874                 /* Receive Checksum Offload */
 875                 e1000_rx_checksum(adapter,
 876                                   (u32)(status) |
 877                                   ((u32)(rx_desc->errors) << 24),
 878                                   le16_to_cpu(rx_desc->csum), skb);
 879
 880                 e1000_receive_skb(adapter, netdev, skb,status,rx_desc->special);
 881
 882 next_desc:
 883                 rx_desc->status = 0;
 884
 885                 /* return some buffers to hardware, one at a time is too slow */
 886                 if (cleaned_count >= E1000_RX_BUFFER_WRITE) {
 887                         adapter->alloc_rx_buf(adapter, cleaned_count);
 888                         cleaned_count = 0;
 889                 }
 890
 891                 /* use prefetched values */
 892                 rx_desc = next_rxd;
 893                 buffer_info = next_buffer;
 894         }
 895         rx_ring->next_to_clean = i;
 896
 897         cleaned_count = e1000_desc_unused(rx_ring);
 898         if (cleaned_count)
 899                 adapter->alloc_rx_buf(adapter, cleaned_count);
 900
 901         adapter->total_rx_bytes += total_rx_bytes;
 902         adapter->total_rx_packets += total_rx_packets;
 903         netdev->stats.rx_bytes += total_rx_bytes;
 904         netdev->stats.rx_packets += total_rx_packets;
 905         return cleaned;
 906 }
 907
 908 static void e1000_put_txbuf(struct e1000_adapter *adapter,
 909                              struct e1000_buffer *buffer_info)
 910 {
 911         if (buffer_info->dma) {
 912                 if (buffer_info->mapped_as_page)
 913                         dma_unmap_page(&adapter->pdev->dev, buffer_info->dma,
 914                                        buffer_info->length, DMA_TO_DEVICE);
 915                 else
 916                         dma_unmap_single(&adapter->pdev->dev, buffer_info->dma,
 917                                          buffer_info->length, DMA_TO_DEVICE);
 918                 buffer_info->dma = 0;
 919         }
 920         if (buffer_info->skb) {
 921                 dev_kfree_skb_any(buffer_info->skb);
 922                 buffer_info->skb = NULL;
 923         }
 924         buffer_info->time_stamp = 0;
 925 }
 926
 927 static void e1000_print_hw_hang(struct work_struct *work)
 928 {
 929         struct e1000_adapter *adapter = container_of(work,
 930                                                      struct e1000_adapter,
 931                                                      print_hang_task);
 932         struct e1000_ring *tx_ring = adapter->tx_ring;
 933         unsigned int i = tx_ring->next_to_clean;
 934         unsigned int eop = tx_ring->buffer_info[i].next_to_watch;
 935         struct e1000_tx_desc *eop_desc = E1000_TX_DESC(*tx_ring, eop);
 936         struct e1000_hw *hw = &adapter->hw;
 937         u16 phy_status, phy_1000t_status, phy_ext_status;
 938         u16 pci_status;
 939
 940         e1e_rphy(hw, PHY_STATUS, &phy_status);
 941         e1e_rphy(hw, PHY_1000T_STATUS, &phy_1000t_status);
 942         e1e_rphy(hw, PHY_EXT_STATUS, &phy_ext_status);
 943
 944         pci_read_config_word(adapter->pdev, PCI_STATUS, &pci_status);
 945
 946         /* detected Hardware unit hang */
 947         e_err("Detected Hardware Unit Hang:\n"
 948               "  TDH                  <%x>\n"
 949               "  TDT                  <%x>\n"
 950               "  next_to_use          <%x>\n"
 951               "  next_to_clean        <%x>\n"
 952               "buffer_info[next_to_clean]:\n"
 953               "  time_stamp           <%lx>\n"
 954               "  next_to_watch        <%x>\n"
 955               "  jiffies              <%lx>\n"
 956               "  next_to_watch.status <%x>\n"
 957               "MAC Status             <%x>\n"
 958               "PHY Status             <%x>\n"
 959               "PHY 1000BASE-T Status  <%x>\n"
 960               "PHY Extended Status    <%x>\n"
 961               "PCI Status             <%x>\n",
 962               readl(adapter->hw.hw_addr + tx_ring->head),
 963               readl(adapter->hw.hw_addr + tx_ring->tail),
 964               tx_ring->next_to_use,
 965               tx_ring->next_to_clean,
 966               tx_ring->buffer_info[eop].time_stamp,
 967               eop,
 968               jiffies,
 969               eop_desc->upper.fields.status,
 970               er32(STATUS),
 971               phy_status,
 972               phy_1000t_status,
 973               phy_ext_status,
 974               pci_status);
 975 }
 976
 977 /**
 978  * e1000_clean_tx_irq - Reclaim resources after transmit completes
 979  * @adapter: board private structure
 980  *
 981  * the return value indicates whether actual cleaning was done, there
 982  * is no guarantee that everything was cleaned
 983  **/
 984 static bool e1000_clean_tx_irq(struct e1000_adapter *adapter)
 985 {
 986         struct net_device *netdev = adapter->netdev;
 987         struct e1000_hw *hw = &adapter->hw;
 988         struct e1000_ring *tx_ring = adapter->tx_ring;
 989         struct e1000_tx_desc *tx_desc, *eop_desc;
 990         struct e1000_buffer *buffer_info;
 991         unsigned int i, eop;
 992         unsigned int count = 0;
 993         unsigned int total_tx_bytes = 0, total_tx_packets = 0;
 994
 995         i = tx_ring->next_to_clean;
 996         eop = tx_ring->buffer_info[i].next_to_watch;
 997         eop_desc = E1000_TX_DESC(*tx_ring, eop);
 998
 999         while ((eop_desc->upper.data & cpu_to_le32(E1000_TXD_STAT_DD)) &&
1000                (count < tx_ring->count)) {
1001                 bool cleaned = false;
1002                 rmb(); /* read buffer_info after eop_desc */
1003                 for (; !cleaned; count++) {
1004                         tx_desc = E1000_TX_DESC(*tx_ring, i);
1005                         buffer_info = &tx_ring->buffer_info[i];
1006                         cleaned = (i == eop);
1007
1008                         if (cleaned) {
1009                                 total_tx_packets += buffer_info->segs;
1010                                 total_tx_bytes += buffer_info->bytecount;
1011                         }
1012
1013                         e1000_put_txbuf(adapter, buffer_info);
1014                         tx_desc->upper.data = 0;
1015
1016                         i++;
1017                         if (i == tx_ring->count)
1018                                 i = 0;
1019                 }
1020
1021                 if (i == tx_ring->next_to_use)
1022                         break;
1023                 eop = tx_ring->buffer_info[i].next_to_watch;
1024                 eop_desc = E1000_TX_DESC(*tx_ring, eop);
1025         }
1026
1027         tx_ring->next_to_clean = i;
1028
1029 #define TX_WAKE_THRESHOLD 32
1030         if (count && netif_carrier_ok(netdev) &&
1031             e1000_desc_unused(tx_ring) >= TX_WAKE_THRESHOLD) {
1032                 /* Make sure that anybody stopping the queue after this
1033                  * sees the new next_to_clean.
1034                  */
1035                 smp_mb();
1036
1037                 if (netif_queue_stopped(netdev) &&
1038                     !(test_bit(__E1000_DOWN, &adapter->state))) {
1039                         netif_wake_queue(netdev);
1040                         ++adapter->restart_queue;
1041                 }
1042         }
1043
1044         if (adapter->detect_tx_hung) {
1045                 /*
1046                  * Detect a transmit hang in hardware, this serializes the
1047                  * check with the clearing of time_stamp and movement of i
1048                  */
1049                 adapter->detect_tx_hung = 0;
1050                 if (tx_ring->buffer_info[i].time_stamp &&
1051                     time_after(jiffies, tx_ring->buffer_info[i].time_stamp
1052                                + (adapter->tx_timeout_factor * HZ)) &&
1053                     !(er32(STATUS) & E1000_STATUS_TXOFF)) {
1054                         schedule_work(&adapter->print_hang_task);
1055                         netif_stop_queue(netdev);
1056                 }
1057         }
1058         adapter->total_tx_bytes += total_tx_bytes;
1059         adapter->total_tx_packets += total_tx_packets;
1060         netdev->stats.tx_bytes += total_tx_bytes;
1061         netdev->stats.tx_packets += total_tx_packets;
1062         return count < tx_ring->count;
1063 }
1064
1065 /**
1066  * e1000_clean_rx_irq_ps - Send received data up the network stack; packet split
1067  * @adapter: board private structure
1068  *
1069  * the return value indicates whether actual cleaning was done, there
1070  * is no guarantee that everything was cleaned
1071  **/
1072 static bool e1000_clean_rx_irq_ps(struct e1000_adapter *adapter,
1073                                   int *work_done, int work_to_do)
1074 {
1075         struct e1000_hw *hw = &adapter->hw;
1076         union e1000_rx_desc_packet_split *rx_desc, *next_rxd;
1077         struct net_device *netdev = adapter->netdev;
1078         struct pci_dev *pdev = adapter->pdev;
1079         struct e1000_ring *rx_ring = adapter->rx_ring;
1080         struct e1000_buffer *buffer_info, *next_buffer;
1081         struct e1000_ps_page *ps_page;
1082         struct sk_buff *skb;
1083         unsigned int i, j;
1084         u32 length, staterr;
1085         int cleaned_count = 0;
1086         bool cleaned = 0;
1087         unsigned int total_rx_bytes = 0, total_rx_packets = 0;
1088
1089         i = rx_ring->next_to_clean;
1090         rx_desc = E1000_RX_DESC_PS(*rx_ring, i);
1091         staterr = le32_to_cpu(rx_desc->wb.middle.status_error);
1092         buffer_info = &rx_ring->buffer_info[i];
1093
1094         while (staterr & E1000_RXD_STAT_DD) {
1095                 if (*work_done >= work_to_do)
1096                         break;
1097                 (*work_done)++;
1098                 skb = buffer_info->skb;
1099                 rmb();  /* read descriptor and rx_buffer_info after status DD */
1100
1101                 /* in the packet split case this is header only */
1102                 prefetch(skb->data - NET_IP_ALIGN);
1103
1104                 i++;
1105                 if (i == rx_ring->count)
1106                         i = 0;
1107                 next_rxd = E1000_RX_DESC_PS(*rx_ring, i);
1108                 prefetch(next_rxd);
1109
1110                 next_buffer = &rx_ring->buffer_info[i];
1111
1112                 cleaned = 1;
1113                 cleaned_count++;
1114                 dma_unmap_single(&pdev->dev, buffer_info->dma,
1115                                  adapter->rx_ps_bsize0, DMA_FROM_DEVICE);
1116                 buffer_info->dma = 0;
1117
1118                 /* see !EOP comment in other Rx routine */
1119                 if (!(staterr & E1000_RXD_STAT_EOP))
1120                         adapter->flags2 |= FLAG2_IS_DISCARDING;
1121
1122                 if (adapter->flags2 & FLAG2_IS_DISCARDING) {
1123                         e_dbg("Packet Split buffers didn't pick up the full "
1124                               "packet\n");
1125                         dev_kfree_skb_irq(skb);
1126                         if (staterr & E1000_RXD_STAT_EOP)
1127                                 adapter->flags2 &= ~FLAG2_IS_DISCARDING;
1128                         goto next_desc;
1129                 }
1130
1131                 if (staterr & E1000_RXDEXT_ERR_FRAME_ERR_MASK) {
1132                         dev_kfree_skb_irq(skb);
1133                         goto next_desc;
1134                 }
1135
1136                 length = le16_to_cpu(rx_desc->wb.middle.length0);
1137
1138                 if (!length) {
1139                         e_dbg("Last part of the packet spanning multiple "
1140                               "descriptors\n");
1141                         dev_kfree_skb_irq(skb);
1142                         goto next_desc;
1143                 }
1144
1145                 /* Good Receive */
1146                 skb_put(skb, length);
1147
1148                 {
1149                 /*
1150                  * this looks ugly, but it seems compiler issues make it
1151                  * more efficient than reusing j
1152                  */
1153                 int l1 = le16_to_cpu(rx_desc->wb.upper.length[0]);
1154
1155                 /*
1156                  * page alloc/put takes too long and effects small packet
1157                  * throughput, so unsplit small packets and save the alloc/put
1158                  * only valid in softirq (napi) context to call kmap_*
1159                  */
1160                 if (l1 && (l1 <= copybreak) &&
1161                     ((length + l1) <= adapter->rx_ps_bsize0)) {
1162                         u8 *vaddr;
1163
1164                         ps_page = &buffer_info->ps_pages[0];
1165
1166                         /*
1167                          * there is no documentation about how to call
1168                          * kmap_atomic, so we can't hold the mapping
1169                          * very long
1170                          */
1171                         dma_sync_single_for_cpu(&pdev->dev, ps_page->dma,
1172                                                 PAGE_SIZE, DMA_FROM_DEVICE);
1173                         vaddr = kmap_atomic(ps_page->page, KM_SKB_DATA_SOFTIRQ);
1174                         memcpy(skb_tail_pointer(skb), vaddr, l1);
1175                         kunmap_atomic(vaddr, KM_SKB_DATA_SOFTIRQ);
1176                         dma_sync_single_for_device(&pdev->dev, ps_page->dma,
1177                                                    PAGE_SIZE, DMA_FROM_DEVICE);
1178
1179                         /* remove the CRC */
1180                         if (!(adapter->flags2 & FLAG2_CRC_STRIPPING))
1181                                 l1 -= 4;
1182
1183                         skb_put(skb, l1);
1184                         goto copydone;
1185                 } /* if */
1186                 }
1187
1188                 for (j = 0; j < PS_PAGE_BUFFERS; j++) {
1189                         length = le16_to_cpu(rx_desc->wb.upper.length[j]);
1190                         if (!length)
1191                                 break;
1192
1193                         ps_page = &buffer_info->ps_pages[j];
1194                         dma_unmap_page(&pdev->dev, ps_page->dma, PAGE_SIZE,
1195                                        DMA_FROM_DEVICE);
1196                         ps_page->dma = 0;
1197                         skb_fill_page_desc(skb, j, ps_page->page, 0, length);
1198                         ps_page->page = NULL;
1199                         skb->len += length;
1200                         skb->data_len += length;
1201                         skb->truesize += length;
1202                 }
1203
1204                 /* strip the ethernet crc, problem is we're using pages now so
1205                  * this whole operation can get a little cpu intensive
1206                  */
1207                 if (!(adapter->flags2 & FLAG2_CRC_STRIPPING))
1208                         pskb_trim(skb, skb->len - 4);
1209
1210 copydone:
1211                 total_rx_bytes += skb->len;
1212                 total_rx_packets++;
1213
1214                 e1000_rx_checksum(adapter, staterr, le16_to_cpu(
1215                         rx_desc->wb.lower.hi_dword.csum_ip.csum), skb);
1216
1217                 if (rx_desc->wb.upper.header_status &
1218                            cpu_to_le16(E1000_RXDPS_HDRSTAT_HDRSP))
1219                         adapter->rx_hdr_split++;
1220
1221                 e1000_receive_skb(adapter, netdev, skb,
1222                                   staterr, rx_desc->wb.middle.vlan);
1223
1224 next_desc:
1225                 rx_desc->wb.middle.status_error &= cpu_to_le32(~0xFF);
1226                 buffer_info->skb = NULL;
1227
1228                 /* return some buffers to hardware, one at a time is too slow */
1229                 if (cleaned_count >= E1000_RX_BUFFER_WRITE) {
1230                         adapter->alloc_rx_buf(adapter, cleaned_count);
1231                         cleaned_count = 0;
1232                 }
1233
1234                 /* use prefetched values */
1235                 rx_desc = next_rxd;
1236                 buffer_info = next_buffer;
1237
1238                 staterr = le32_to_cpu(rx_desc->wb.middle.status_error);
1239         }
1240         rx_ring->next_to_clean = i;
1241
1242         cleaned_count = e1000_desc_unused(rx_ring);
1243         if (cleaned_count)
1244                 adapter->alloc_rx_buf(adapter, cleaned_count);
1245
1246         adapter->total_rx_bytes += total_rx_bytes;
1247         adapter->total_rx_packets += total_rx_packets;
1248         netdev->stats.rx_bytes += total_rx_bytes;
1249         netdev->stats.rx_packets += total_rx_packets;
1250         return cleaned;
1251 }
1252
1253 /**
1254  * e1000_consume_page - helper function
1255  **/
1256 static void e1000_consume_page(struct e1000_buffer *bi, struct sk_buff *skb,
1257                                u16 length)
1258 {
1259         bi->page = NULL;
1260         skb->len += length;
1261         skb->data_len += length;
1262         skb->truesize += length;
1263 }
1264
1265 /**
1266  * e1000_clean_jumbo_rx_irq - Send received data up the network stack; legacy
1267  * @adapter: board private structure
1268  *
1269  * the return value indicates whether actual cleaning was done, there
1270  * is no guarantee that everything was cleaned
1271  **/
1272
1273 static bool e1000_clean_jumbo_rx_irq(struct e1000_adapter *adapter,
1274                                      int *work_done, int work_to_do)
1275 {
1276         struct net_device *netdev = adapter->netdev;
1277         struct pci_dev *pdev = adapter->pdev;
1278         struct e1000_ring *rx_ring = adapter->rx_ring;
1279         struct e1000_rx_desc *rx_desc, *next_rxd;
1280         struct e1000_buffer *buffer_info, *next_buffer;
1281         u32 length;
1282         unsigned int i;
1283         int cleaned_count = 0;
1284         bool cleaned = false;
1285         unsigned int total_rx_bytes=0, total_rx_packets=0;
1286
1287         i = rx_ring->next_to_clean;
1288         rx_desc = E1000_RX_DESC(*rx_ring, i);
1289         buffer_info = &rx_ring->buffer_info[i];
1290
1291         while (rx_desc->status & E1000_RXD_STAT_DD) {
1292                 struct sk_buff *skb;
1293                 u8 status;
1294
1295                 if (*work_done >= work_to_do)
1296                         break;
1297                 (*work_done)++;
1298                 rmb();  /* read descriptor and rx_buffer_info after status DD */
1299
1300                 status = rx_desc->status;
1301                 skb = buffer_info->skb;
1302                 buffer_info->skb = NULL;
1303
1304                 ++i;
1305                 if (i == rx_ring->count)
1306                         i = 0;
1307                 next_rxd = E1000_RX_DESC(*rx_ring, i);
1308                 prefetch(next_rxd);
1309
1310                 next_buffer = &rx_ring->buffer_info[i];
1311
1312                 cleaned = true;
1313                 cleaned_count++;
1314                 dma_unmap_page(&pdev->dev, buffer_info->dma, PAGE_SIZE,
1315                                DMA_FROM_DEVICE);
1316                 buffer_info->dma = 0;
1317
1318                 length = le16_to_cpu(rx_desc->length);
1319
1320                 /* errors is only valid for DD + EOP descriptors */
1321                 if (unlikely((status & E1000_RXD_STAT_EOP) &&
1322                     (rx_desc->errors & E1000_RXD_ERR_FRAME_ERR_MASK))) {
1323                                 /* recycle both page and skb */
1324                                 buffer_info->skb = skb;
1325                                 /* an error means any chain goes out the window
1326                                  * too */
1327                                 if (rx_ring->rx_skb_top)
1328                                         dev_kfree_skb(rx_ring->rx_skb_top);
1329                                 rx_ring->rx_skb_top = NULL;
1330                                 goto next_desc;
1331                 }
1332
1333 #define rxtop (rx_ring->rx_skb_top)
1334                 if (!(status & E1000_RXD_STAT_EOP)) {
1335                         /* this descriptor is only the beginning (or middle) */
1336                         if (!rxtop) {
1337                                 /* this is the beginning of a chain */
1338                                 rxtop = skb;
1339                                 skb_fill_page_desc(rxtop, 0, buffer_info->page,
1340                                                    0, length);
1341                         } else {
1342                                 /* this is the middle of a chain */
1343                                 skb_fill_page_desc(rxtop,
1344                                     skb_shinfo(rxtop)->nr_frags,
1345                                     buffer_info->page, 0, length);
1346                                 /* re-use the skb, only consumed the page */
1347                                 buffer_info->skb = skb;
1348                         }
1349                         e1000_consume_page(buffer_info, rxtop, length);
1350                         goto next_desc;
1351                 } else {
1352                         if (rxtop) {
1353                                 /* end of the chain */
1354                                 skb_fill_page_desc(rxtop,
1355                                     skb_shinfo(rxtop)->nr_frags,
1356                                     buffer_info->page, 0, length);
1357                                 /* re-use the current skb, we only consumed the
1358                                  * page */
1359                                 buffer_info->skb = skb;
1360                                 skb = rxtop;
1361                                 rxtop = NULL;
1362                                 e1000_consume_page(buffer_info, skb, length);
1363                         } else {
1364                                 /* no chain, got EOP, this buf is the packet
1365                                  * copybreak to save the put_page/alloc_page */
1366                                 if (length <= copybreak &&
1367                                     skb_tailroom(skb) >= length) {
1368                                         u8 *vaddr;
1369                                         vaddr = kmap_atomic(buffer_info->page,
1370                                                            KM_SKB_DATA_SOFTIRQ);
1371                                         memcpy(skb_tail_pointer(skb), vaddr,
1372                                                length);
1373                                         kunmap_atomic(vaddr,
1374                                                       KM_SKB_DATA_SOFTIRQ);
1375                                         /* re-use the page, so don't erase
1376                                          * buffer_info->page */
1377                                         skb_put(skb, length);
1378                                 } else {
1379                                         skb_fill_page_desc(skb, 0,
1380                                                            buffer_info->page, 0,
1381                                                            length);
1382                                         e1000_consume_page(buffer_info, skb,
1383                                                            length);
1384                                 }
1385                         }
1386                 }
1387
1388                 /* Receive Checksum Offload XXX recompute due to CRC strip? */
1389                 e1000_rx_checksum(adapter,
1390                                   (u32)(status) |
1391                                   ((u32)(rx_desc->errors) << 24),
1392                                   le16_to_cpu(rx_desc->csum), skb);
1393
1394                 /* probably a little skewed due to removing CRC */
1395                 total_rx_bytes += skb->len;
1396                 total_rx_packets++;
1397
1398                 /* eth type trans needs skb->data to point to something */
1399                 if (!pskb_may_pull(skb, ETH_HLEN)) {
1400                         e_err("pskb_may_pull failed.\n");
1401                         dev_kfree_skb(skb);
1402                         goto next_desc;
1403                 }
1404
1405                 e1000_receive_skb(adapter, netdev, skb, status,
1406                                   rx_desc->special);
1407
1408 next_desc:
1409                 rx_desc->status = 0;
1410
1411                 /* return some buffers to hardware, one at a time is too slow */
1412                 if (unlikely(cleaned_count >= E1000_RX_BUFFER_WRITE)) {
1413                         adapter->alloc_rx_buf(adapter, cleaned_count);
1414                         cleaned_count = 0;
1415                 }
1416
1417                 /* use prefetched values */
1418                 rx_desc = next_rxd;
1419                 buffer_info = next_buffer;
1420         }
1421         rx_ring->next_to_clean = i;
1422
1423         cleaned_count = e1000_desc_unused(rx_ring);
1424         if (cleaned_count)
1425                 adapter->alloc_rx_buf(adapter, cleaned_count);
1426
1427         adapter->total_rx_bytes += total_rx_bytes;
1428         adapter->total_rx_packets += total_rx_packets;
1429         netdev->stats.rx_bytes += total_rx_bytes;
1430         netdev->stats.rx_packets += total_rx_packets;
1431         return cleaned;
1432 }
1433
1434 /**
1435  * e1000_clean_rx_ring - Free Rx Buffers per Queue
1436  * @adapter: board private structure
1437  **/
1438 static void e1000_clean_rx_ring(struct e1000_adapter *adapter)
1439 {
1440         struct e1000_ring *rx_ring = adapter->rx_ring;
1441         struct e1000_buffer *buffer_info;
1442         struct e1000_ps_page *ps_page;
1443         struct pci_dev *pdev = adapter->pdev;
1444         unsigned int i, j;
1445
1446         /* Free all the Rx ring sk_buffs */
1447         for (i = 0; i < rx_ring->count; i++) {
1448                 buffer_info = &rx_ring->buffer_info[i];
1449                 if (buffer_info->dma) {
1450                         if (adapter->clean_rx == e1000_clean_rx_irq)
1451                                 dma_unmap_single(&pdev->dev, buffer_info->dma,
1452                                                  adapter->rx_buffer_len,
1453                                                  DMA_FROM_DEVICE);
1454                         else if (adapter->clean_rx == e1000_clean_jumbo_rx_irq)
1455                                 dma_unmap_page(&pdev->dev, buffer_info->dma,
1456                                                PAGE_SIZE,
1457                                                DMA_FROM_DEVICE);
1458                         else if (adapter->clean_rx == e1000_clean_rx_irq_ps)
1459                                 dma_unmap_single(&pdev->dev, buffer_info->dma,
1460                                                  adapter->rx_ps_bsize0,
1461                                                  DMA_FROM_DEVICE);
1462                         buffer_info->dma = 0;
1463                 }
1464
1465                 if (buffer_info->page) {
1466                         put_page(buffer_info->page);
1467                         buffer_info->page = NULL;
1468                 }
1469
1470                 if (buffer_info->skb) {
1471                         dev_kfree_skb(buffer_info->skb);
1472                         buffer_info->skb = NULL;
1473                 }
1474
1475                 for (j = 0; j < PS_PAGE_BUFFERS; j++) {
1476                         ps_page = &buffer_info->ps_pages[j];
1477                         if (!ps_page->page)
1478                                 break;
1479                         dma_unmap_page(&pdev->dev, ps_page->dma, PAGE_SIZE,
1480                                        DMA_FROM_DEVICE);
1481                         ps_page->dma = 0;
1482                         put_page(ps_page->page);
1483                         ps_page->page = NULL;
1484                 }
1485         }
1486
1487         /* there also may be some cached data from a chained receive */
1488         if (rx_ring->rx_skb_top) {
1489                 dev_kfree_skb(rx_ring->rx_skb_top);
1490                 rx_ring->rx_skb_top = NULL;
1491         }
1492
1493         /* Zero out the descriptor ring */
1494         memset(rx_ring->desc, 0, rx_ring->size);
1495
1496         rx_ring->next_to_clean = 0;
1497         rx_ring->next_to_use = 0;
1498         adapter->flags2 &= ~FLAG2_IS_DISCARDING;
1499
1500         writel(0, adapter->hw.hw_addr + rx_ring->head);
1501         writel(0, adapter->hw.hw_addr + rx_ring->tail);
1502 }
1503
1504 static void e1000e_downshift_workaround(struct work_struct *work)
1505 {
1506         struct e1000_adapter *adapter = container_of(work,
1507                                         struct e1000_adapter, downshift_task);
1508
1509         e1000e_gig_downshift_workaround_ich8lan(&adapter->hw);
1510 }
1511
1512 /**
1513  * e1000_intr_msi - Interrupt Handler
1514  * @irq: interrupt number
1515  * @data: pointer to a network interface device structure
1516  **/
1517 static irqreturn_t e1000_intr_msi(int irq, void *data)
1518 {
1519         struct net_device *netdev = data;
1520         struct e1000_adapter *adapter = netdev_priv(netdev);
1521         struct e1000_hw *hw = &adapter->hw;
1522         u32 icr = er32(ICR);
1523
1524         /*
1525          * read ICR disables interrupts using IAM
1526          */
1527
1528         if (icr & E1000_ICR_LSC) {
1529                 hw->mac.get_link_status = 1;
1530                 /*
1531                  * ICH8 workaround-- Call gig speed drop workaround on cable
1532                  * disconnect (LSC) before accessing any PHY registers
1533                  */
1534                 if ((adapter->flags & FLAG_LSC_GIG_SPEED_DROP) &&
1535                     (!(er32(STATUS) & E1000_STATUS_LU)))
1536                         schedule_work(&adapter->downshift_task);
1537
1538                 /*
1539                  * 80003ES2LAN workaround-- For packet buffer work-around on
1540                  * link down event; disable receives here in the ISR and reset
1541                  * adapter in watchdog
1542                  */
1543                 if (netif_carrier_ok(netdev) &&
1544                     adapter->flags & FLAG_RX_NEEDS_RESTART) {
1545                         /* disable receives */
1546                         u32 rctl = er32(RCTL);
1547                         ew32(RCTL, rctl & ~E1000_RCTL_EN);
1548                         adapter->flags |= FLAG_RX_RESTART_NOW;
1549                 }
1550                 /* guard against interrupt when we're going down */
1551                 if (!test_bit(__E1000_DOWN, &adapter->state))
1552                         mod_timer(&adapter->watchdog_timer, jiffies + 1);
1553         }
1554
1555         if (napi_schedule_prep(&adapter->napi)) {
1556                 adapter->total_tx_bytes = 0;
1557                 adapter->total_tx_packets = 0;
1558                 adapter->total_rx_bytes = 0;
1559                 adapter->total_rx_packets = 0;
1560                 __napi_schedule(&adapter->napi);
1561         }
1562
1563         return IRQ_HANDLED;
1564 }
1565
1566 /**
1567  * e1000_intr - Interrupt Handler
1568  * @irq: interrupt number
1569  * @data: pointer to a network interface device structure
1570  **/
1571 static irqreturn_t e1000_intr(int irq, void *data)
1572 {
1573         struct net_device *netdev = data;
1574         struct e1000_adapter *adapter = netdev_priv(netdev);
1575         struct e1000_hw *hw = &adapter->hw;
1576         u32 rctl, icr = er32(ICR);
1577
1578         if (!icr || test_bit(__E1000_DOWN, &adapter->state))
1579                 return IRQ_NONE;  /* Not our interrupt */
1580
1581         /*
1582          * IMS will not auto-mask if INT_ASSERTED is not set, and if it is
1583          * not set, then the adapter didn't send an interrupt
1584          */
1585         if (!(icr & E1000_ICR_INT_ASSERTED))
1586                 return IRQ_NONE;
1587
1588         /*
1589          * Interrupt Auto-Mask...upon reading ICR,
1590          * interrupts are masked.  No need for the
1591          * IMC write
1592          */
1593
1594         if (icr & E1000_ICR_LSC) {
1595                 hw->mac.get_link_status = 1;
1596                 /*
1597                  * ICH8 workaround-- Call gig speed drop workaround on cable
1598                  * disconnect (LSC) before accessing any PHY registers
1599                  */
1600                 if ((adapter->flags & FLAG_LSC_GIG_SPEED_DROP) &&
1601                     (!(er32(STATUS) & E1000_STATUS_LU)))
1602                         schedule_work(&adapter->downshift_task);
1603
1604                 /*
1605                  * 80003ES2LAN workaround--
1606                  * For packet buffer work-around on link down event;
1607                  * disable receives here in the ISR and
1608                  * reset adapter in watchdog
1609                  */
1610                 if (netif_carrier_ok(netdev) &&
1611                     (adapter->flags & FLAG_RX_NEEDS_RESTART)) {
1612                         /* disable receives */
1613                         rctl = er32(RCTL);
1614                         ew32(RCTL, rctl & ~E1000_RCTL_EN);
1615                         adapter->flags |= FLAG_RX_RESTART_NOW;
1616                 }
1617                 /* guard against interrupt when we're going down */
1618                 if (!test_bit(__E1000_DOWN, &adapter->state))
1619                         mod_timer(&adapter->watchdog_timer, jiffies + 1);
1620         }
1621
1622         if (napi_schedule_prep(&adapter->napi)) {
1623                 adapter->total_tx_bytes = 0;
1624                 adapter->total_tx_packets = 0;
1625                 adapter->total_rx_bytes = 0;
1626                 adapter->total_rx_packets = 0;
1627                 __napi_schedule(&adapter->napi);
1628         }
1629
1630         return IRQ_HANDLED;
1631 }
1632
1633 static irqreturn_t e1000_msix_other(int irq, void *data)
1634 {
1635         struct net_device *netdev = data;
1636         struct e1000_adapter *adapter = netdev_priv(netdev);
1637         struct e1000_hw *hw = &adapter->hw;
1638         u32 icr = er32(ICR);
1639
1640         if (!(icr & E1000_ICR_INT_ASSERTED)) {
1641                 if (!test_bit(__E1000_DOWN, &adapter->state))
1642                         ew32(IMS, E1000_IMS_OTHER);
1643                 return IRQ_NONE;
1644         }
1645
1646         if (icr & adapter->eiac_mask)
1647                 ew32(ICS, (icr & adapter->eiac_mask));
1648
1649         if (icr & E1000_ICR_OTHER) {
1650                 if (!(icr & E1000_ICR_LSC))
1651                         goto no_link_interrupt;
1652                 hw->mac.get_link_status = 1;
1653                 /* guard against interrupt when we're going down */
1654                 if (!test_bit(__E1000_DOWN, &adapter->state))
1655                         mod_timer(&adapter->watchdog_timer, jiffies + 1);
1656         }
1657
1658 no_link_interrupt:
1659         if (!test_bit(__E1000_DOWN, &adapter->state))
1660                 ew32(IMS, E1000_IMS_LSC | E1000_IMS_OTHER);
1661
1662         return IRQ_HANDLED;
1663 }
1664
1665
1666 static irqreturn_t e1000_intr_msix_tx(int irq, void *data)
1667 {
1668         struct net_device *netdev = data;
1669         struct e1000_adapter *adapter = netdev_priv(netdev);
1670         struct e1000_hw *hw = &adapter->hw;
1671         struct e1000_ring *tx_ring = adapter->tx_ring;
1672
1673
1674         adapter->total_tx_bytes = 0;
1675         adapter->total_tx_packets = 0;
1676
1677         if (!e1000_clean_tx_irq(adapter))
1678                 /* Ring was not completely cleaned, so fire another interrupt */
1679                 ew32(ICS, tx_ring->ims_val);
1680
1681         return IRQ_HANDLED;
1682 }
1683
1684 static irqreturn_t e1000_intr_msix_rx(int irq, void *data)
1685 {
1686         struct net_device *netdev = data;
1687         struct e1000_adapter *adapter = netdev_priv(netdev);
1688
1689         /* Write the ITR value calculated at the end of the
1690          * previous interrupt.
1691          */
1692         if (adapter->rx_ring->set_itr) {
1693                 writel(1000000000 / (adapter->rx_ring->itr_val * 256),
1694                        adapter->hw.hw_addr + adapter->rx_ring->itr_register);
1695                 adapter->rx_ring->set_itr = 0;
1696         }
1697
1698         if (napi_schedule_prep(&adapter->napi)) {
1699                 adapter->total_rx_bytes = 0;
1700                 adapter->total_rx_packets = 0;
1701                 __napi_schedule(&adapter->napi);
1702         }
1703         return IRQ_HANDLED;
1704 }
1705
1706 /**
1707  * e1000_configure_msix - Configure MSI-X hardware
1708  *
1709  * e1000_configure_msix sets up the hardware to properly
1710  * generate MSI-X interrupts.
1711  **/
1712 static void e1000_configure_msix(struct e1000_adapter *adapter)
1713 {
1714         struct e1000_hw *hw = &adapter->hw;
1715         struct e1000_ring *rx_ring = adapter->rx_ring;
1716         struct e1000_ring *tx_ring = adapter->tx_ring;
1717         int vector = 0;
1718         u32 ctrl_ext, ivar = 0;
1719
1720         adapter->eiac_mask = 0;
1721
1722         /* Workaround issue with spurious interrupts on 82574 in MSI-X mode */
1723         if (hw->mac.type == e1000_82574) {
1724                 u32 rfctl = er32(RFCTL);
1725                 rfctl |= E1000_RFCTL_ACK_DIS;
1726                 ew32(RFCTL, rfctl);
1727         }
1728
1729 #define E1000_IVAR_INT_ALLOC_VALID      0x8
1730         /* Configure Rx vector */
1731         rx_ring->ims_val = E1000_IMS_RXQ0;
1732         adapter->eiac_mask |= rx_ring->ims_val;
1733         if (rx_ring->itr_val)
1734                 writel(1000000000 / (rx_ring->itr_val * 256),
1735                        hw->hw_addr + rx_ring->itr_register);
1736         else
1737                 writel(1, hw->hw_addr + rx_ring->itr_register);
1738         ivar = E1000_IVAR_INT_ALLOC_VALID | vector;
1739
1740         /* Configure Tx vector */
1741         tx_ring->ims_val = E1000_IMS_TXQ0;
1742         vector++;
1743         if (tx_ring->itr_val)
1744                 writel(1000000000 / (tx_ring->itr_val * 256),
1745                        hw->hw_addr + tx_ring->itr_register);
1746         else
1747                 writel(1, hw->hw_addr + tx_ring->itr_register);
1748         adapter->eiac_mask |= tx_ring->ims_val;
1749         ivar |= ((E1000_IVAR_INT_ALLOC_VALID | vector) << 8);
1750
1751         /* set vector for Other Causes, e.g. link changes */
1752         vector++;
1753         ivar |= ((E1000_IVAR_INT_ALLOC_VALID | vector) << 16);
1754         if (rx_ring->itr_val)
1755                 writel(1000000000 / (rx_ring->itr_val * 256),
1756                        hw->hw_addr + E1000_EITR_82574(vector));
1757         else
1758                 writel(1, hw->hw_addr + E1000_EITR_82574(vector));
1759
1760         /* Cause Tx interrupts on every write back */
1761         ivar |= (1 << 31);
1762
1763         ew32(IVAR, ivar);
1764
1765         /* enable MSI-X PBA support */
1766         ctrl_ext = er32(CTRL_EXT);
1767         ctrl_ext |= E1000_CTRL_EXT_PBA_CLR;
1768
1769         /* Auto-Mask Other interrupts upon ICR read */
1770 #define E1000_EIAC_MASK_82574   0x01F00000
1771         ew32(IAM, ~E1000_EIAC_MASK_82574 | E1000_IMS_OTHER);
1772         ctrl_ext |= E1000_CTRL_EXT_EIAME;
1773         ew32(CTRL_EXT, ctrl_ext);
1774         e1e_flush();
1775 }
1776
1777 void e1000e_reset_interrupt_capability(struct e1000_adapter *adapter)
1778 {
1779         if (adapter->msix_entries) {
1780                 pci_disable_msix(adapter->pdev);
1781                 kfree(adapter->msix_entries);
1782                 adapter->msix_entries = NULL;
1783         } else if (adapter->flags & FLAG_MSI_ENABLED) {
1784                 pci_disable_msi(adapter->pdev);
1785                 adapter->flags &= ~FLAG_MSI_ENABLED;
1786         }
1787 }
1788
1789 /**
1790  * e1000e_set_interrupt_capability - set MSI or MSI-X if supported
1791  *
1792  * Attempt to configure interrupts using the best available
1793  * capabilities of the hardware and kernel.
1794  **/
1795 void e1000e_set_interrupt_capability(struct e1000_adapter *adapter)
1796 {
1797         int err;
1798         int i;
1799
1800         switch (adapter->int_mode) {
1801         case E1000E_INT_MODE_MSIX:
1802                 if (adapter->flags & FLAG_HAS_MSIX) {
1803                         adapter->num_vectors = 3; /* RxQ0, TxQ0 and other */
1804                         adapter->msix_entries = kcalloc(adapter->num_vectors,
1805                                                       sizeof(struct msix_entry),
1806                                                       GFP_KERNEL);
1807                         if (adapter->msix_entries) {
1808                                 for (i = 0; i < adapter->num_vectors; i++)
1809                                         adapter->msix_entries[i].entry = i;
1810
1811                                 err = pci_enable_msix(adapter->pdev,
1812                                                       adapter->msix_entries,
1813                                                       adapter->num_vectors);
1814                                 if (err == 0)
1815                                         return;
1816                         }
1817                         /* MSI-X failed, so fall through and try MSI */
1818                         e_err("Failed to initialize MSI-X interrupts.  "
1819                               "Falling back to MSI interrupts.\n");
1820                         e1000e_reset_interrupt_capability(adapter);
1821                 }
1822                 adapter->int_mode = E1000E_INT_MODE_MSI;
1823                 /* Fall through */
1824         case E1000E_INT_MODE_MSI:
1825                 if (!pci_enable_msi(adapter->pdev)) {
1826                         adapter->flags |= FLAG_MSI_ENABLED;
1827                 } else {
1828                         adapter->int_mode = E1000E_INT_MODE_LEGACY;
1829                         e_err("Failed to initialize MSI interrupts.  Falling "
1830                               "back to legacy interrupts.\n");
1831                 }
1832                 /* Fall through */
1833         case E1000E_INT_MODE_LEGACY:
1834                 /* Don't do anything; this is the system default */
1835                 break;
1836         }
1837
1838         /* store the number of vectors being used */
1839         adapter->num_vectors = 1;
1840 }
1841
1842 /**
1843  * e1000_request_msix - Initialize MSI-X interrupts
1844  *
1845  * e1000_request_msix allocates MSI-X vectors and requests interrupts from the
1846  * kernel.
1847  **/
1848 static int e1000_request_msix(struct e1000_adapter *adapter)
1849 {
1850         struct net_device *netdev = adapter->netdev;
1851         int err = 0, vector = 0;
1852
1853         if (strlen(netdev->name) < (IFNAMSIZ - 5))
1854                 sprintf(adapter->rx_ring->name, "%s-rx-0", netdev->name);
1855         else
1856                 memcpy(adapter->rx_ring->name, netdev->name, IFNAMSIZ);
1857         err = request_irq(adapter->msix_entries[vector].vector,
1858                           e1000_intr_msix_rx, 0, adapter->rx_ring->name,
1859                           netdev);
1860         if (err)
1861                 goto out;
1862         adapter->rx_ring->itr_register = E1000_EITR_82574(vector);
1863         adapter->rx_ring->itr_val = adapter->itr;
1864         vector++;
1865
1866         if (strlen(netdev->name) < (IFNAMSIZ - 5))
1867                 sprintf(adapter->tx_ring->name, "%s-tx-0", netdev->name);
1868         else
1869                 memcpy(adapter->tx_ring->name, netdev->name, IFNAMSIZ);
1870         err = request_irq(adapter->msix_entries[vector].vector,
1871                           e1000_intr_msix_tx, 0, adapter->tx_ring->name,
1872                           netdev);
1873         if (err)
1874                 goto out;
1875         adapter->tx_ring->itr_register = E1000_EITR_82574(vector);
1876         adapter->tx_ring->itr_val = adapter->itr;
1877         vector++;
1878
1879         err = request_irq(adapter->msix_entries[vector].vector,
1880                           e1000_msix_other, 0, netdev->name, netdev);
1881         if (err)
1882                 goto out;
1883
1884         e1000_configure_msix(adapter);
1885         return 0;
1886 out:
1887         return err;
1888 }
1889
1890 /**
1891  * e1000_request_irq - initialize interrupts
1892  *
1893  * Attempts to configure interrupts using the best available
1894  * capabilities of the hardware and kernel.
1895  **/
1896 static int e1000_request_irq(struct e1000_adapter *adapter)
1897 {
1898         struct net_device *netdev = adapter->netdev;
1899         int err;
1900
1901         if (adapter->msix_entries) {
1902                 err = e1000_request_msix(adapter);
1903                 if (!err)
1904                         return err;
1905                 /* fall back to MSI */
1906                 e1000e_reset_interrupt_capability(adapter);
1907                 adapter->int_mode = E1000E_INT_MODE_MSI;
1908                 e1000e_set_interrupt_capability(adapter);
1909         }
1910         if (adapter->flags & FLAG_MSI_ENABLED) {
1911                 err = request_irq(adapter->pdev->irq, e1000_intr_msi, 0,
1912                                   netdev->name, netdev);
1913                 if (!err)
1914                         return err;
1915
1916                 /* fall back to legacy interrupt */
1917                 e1000e_reset_interrupt_capability(adapter);
1918                 adapter->int_mode = E1000E_INT_MODE_LEGACY;
1919         }
1920
1921         err = request_irq(adapter->pdev->irq, e1000_intr, IRQF_SHARED,
1922                           netdev->name, netdev);
1923         if (err)
1924                 e_err("Unable to allocate interrupt, Error: %d\n", err);
1925
1926         return err;
1927 }
1928
1929 static void e1000_free_irq(struct e1000_adapter *adapter)
1930 {
1931         struct net_device *netdev = adapter->netdev;
1932
1933         if (adapter->msix_entries) {
1934                 int vector = 0;
1935
1936                 free_irq(adapter->msix_entries[vector].vector, netdev);
1937                 vector++;
1938
1939                 free_irq(adapter->msix_entries[vector].vector, netdev);
1940                 vector++;
1941
1942                 /* Other Causes interrupt vector */
1943                 free_irq(adapter->msix_entries[vector].vector, netdev);
1944                 return;
1945         }
1946
1947         free_irq(adapter->pdev->irq, netdev);
1948 }
1949
1950 /**
1951  * e1000_irq_disable - Mask off interrupt generation on the NIC
1952  **/
1953 static void e1000_irq_disable(struct e1000_adapter *adapter)
1954 {
1955         struct e1000_hw *hw = &adapter->hw;
1956
1957         ew32(IMC, ~0);
1958         if (adapter->msix_entries)
1959                 ew32(EIAC_82574, 0);
1960         e1e_flush();
1961
1962         if (adapter->msix_entries) {
1963                 int i;
1964                 for (i = 0; i < adapter->num_vectors; i++)
1965                         synchronize_irq(adapter->msix_entries[i].vector);
1966         } else {
1967                 synchronize_irq(adapter->pdev->irq);
1968         }
1969 }
1970
1971 /**
1972  * e1000_irq_enable - Enable default interrupt generation settings
1973  **/
1974 static void e1000_irq_enable(struct e1000_adapter *adapter)
1975 {
1976         struct e1000_hw *hw = &adapter->hw;
1977
1978         if (adapter->msix_entries) {
1979                 ew32(EIAC_82574, adapter->eiac_mask & E1000_EIAC_MASK_82574);
1980                 ew32(IMS, adapter->eiac_mask | E1000_IMS_OTHER | E1000_IMS_LSC);
1981         } else {
1982                 ew32(IMS, IMS_ENABLE_MASK);
1983         }
1984         e1e_flush();
1985 }
1986
1987 /**
1988  * e1000e_get_hw_control - get control of the h/w from f/w
1989  * @adapter: address of board private structure
1990  *
1991  * e1000e_get_hw_control sets {CTRL_EXT|SWSM}:DRV_LOAD bit.
1992  * For ASF and Pass Through versions of f/w this means that
1993  * the driver is loaded. For AMT version (only with 82573)
1994  * of the f/w this means that the network i/f is open.
1995  **/
1996 void e1000e_get_hw_control(struct e1000_adapter *adapter)
1997 {
1998         struct e1000_hw *hw = &adapter->hw;
1999         u32 ctrl_ext;
2000         u32 swsm;
2001
2002         /* Let firmware know the driver has taken over */
2003         if (adapter->flags & FLAG_HAS_SWSM_ON_LOAD) {
2004                 swsm = er32(SWSM);
2005                 ew32(SWSM, swsm | E1000_SWSM_DRV_LOAD);
2006         } else if (adapter->flags & FLAG_HAS_CTRLEXT_ON_LOAD) {
2007                 ctrl_ext = er32(CTRL_EXT);
2008                 ew32(CTRL_EXT, ctrl_ext | E1000_CTRL_EXT_DRV_LOAD);
2009         }
2010 }
2011
2012 /**
2013  * e1000e_release_hw_control - release control of the h/w to f/w
2014  * @adapter: address of board private structure
2015  *
2016  * e1000e_release_hw_control resets {CTRL_EXT|SWSM}:DRV_LOAD bit.
2017  * For ASF and Pass Through versions of f/w this means that the
2018  * driver is no longer loaded. For AMT version (only with 82573) i
2019  * of the f/w this means that the network i/f is closed.
2020  *
2021  **/
2022 void e1000e_release_hw_control(struct e1000_adapter *adapter)
2023 {
2024         struct e1000_hw *hw = &adapter->hw;
2025         u32 ctrl_ext;
2026         u32 swsm;
2027
2028         /* Let firmware taken over control of h/w */
2029         if (adapter->flags & FLAG_HAS_SWSM_ON_LOAD) {
2030                 swsm = er32(SWSM);
2031                 ew32(SWSM, swsm & ~E1000_SWSM_DRV_LOAD);
2032         } else if (adapter->flags & FLAG_HAS_CTRLEXT_ON_LOAD) {
2033                 ctrl_ext = er32(CTRL_EXT);
2034                 ew32(CTRL_EXT, ctrl_ext & ~E1000_CTRL_EXT_DRV_LOAD);
2035         }
2036 }
2037
2038 /**
2039  * @e1000_alloc_ring - allocate memory for a ring structure
2040  **/
2041 static int e1000_alloc_ring_dma(struct e1000_adapter *adapter,
2042                                 struct e1000_ring *ring)
2043 {
2044         struct pci_dev *pdev = adapter->pdev;
2045
2046         ring->desc = dma_alloc_coherent(&pdev->dev, ring->size, &ring->dma,
2047                                         GFP_KERNEL);
2048         if (!ring->desc)
2049                 return -ENOMEM;
2050
2051         return 0;
2052 }
2053
2054 /**
2055  * e1000e_setup_tx_resources - allocate Tx resources (Descriptors)
2056  * @adapter: board private structure
2057  *
2058  * Return 0 on success, negative on failure
2059  **/
2060 int e1000e_setup_tx_resources(struct e1000_adapter *adapter)
2061 {
2062         struct e1000_ring *tx_ring = adapter->tx_ring;
2063         int err = -ENOMEM, size;
2064
2065         size = sizeof(struct e1000_buffer) * tx_ring->count;
2066         tx_ring->buffer_info = vzalloc(size);
2067         if (!tx_ring->buffer_info)
2068                 goto err;
2069
2070         /* round up to nearest 4K */
2071         tx_ring->size = tx_ring->count * sizeof(struct e1000_tx_desc);
2072         tx_ring->size = ALIGN(tx_ring->size, 4096);
2073
2074         err = e1000_alloc_ring_dma(adapter, tx_ring);
2075         if (err)
2076                 goto err;
2077
2078         tx_ring->next_to_use = 0;
2079         tx_ring->next_to_clean = 0;
2080
2081         return 0;
2082 err:
2083         vfree(tx_ring->buffer_info);
2084         e_err("Unable to allocate memory for the transmit descriptor ring\n");
2085         return err;
2086 }
2087
2088 /**
2089  * e1000e_setup_rx_resources - allocate Rx resources (Descriptors)
2090  * @adapter: board private structure
2091  *
2092  * Returns 0 on success, negative on failure
2093  **/
2094 int e1000e_setup_rx_resources(struct e1000_adapter *adapter)
2095 {
2096         struct e1000_ring *rx_ring = adapter->rx_ring;
2097         struct e1000_buffer *buffer_info;
2098         int i, size, desc_len, err = -ENOMEM;
2099
2100         size = sizeof(struct e1000_buffer) * rx_ring->count;
2101         rx_ring->buffer_info = vzalloc(size);
2102         if (!rx_ring->buffer_info)
2103                 goto err;
2104
2105         for (i = 0; i < rx_ring->count; i++) {
2106                 buffer_info = &rx_ring->buffer_info[i];
2107                 buffer_info->ps_pages = kcalloc(PS_PAGE_BUFFERS,
2108                                                 sizeof(struct e1000_ps_page),
2109                                                 GFP_KERNEL);
2110                 if (!buffer_info->ps_pages)
2111                         goto err_pages;
2112         }
2113
2114         desc_len = sizeof(union e1000_rx_desc_packet_split);
2115
2116         /* Round up to nearest 4K */
2117         rx_ring->size = rx_ring->count * desc_len;
2118         rx_ring->size = ALIGN(rx_ring->size, 4096);
2119
2120         err = e1000_alloc_ring_dma(adapter, rx_ring);
2121         if (err)
2122                 goto err_pages;
2123
2124         rx_ring->next_to_clean = 0;
2125         rx_ring->next_to_use = 0;
2126         rx_ring->rx_skb_top = NULL;
2127
2128         return 0;
2129
2130 err_pages:
2131         for (i = 0; i < rx_ring->count; i++) {
2132                 buffer_info = &rx_ring->buffer_info[i];
2133                 kfree(buffer_info->ps_pages);
2134         }
2135 err:
2136         vfree(rx_ring->buffer_info);
2137         e_err("Unable to allocate memory for the receive descriptor ring\n");
2138         return err;
2139 }
2140
2141 /**
2142  * e1000_clean_tx_ring - Free Tx Buffers
2143  * @adapter: board private structure
2144  **/
2145 static void e1000_clean_tx_ring(struct e1000_adapter *adapter)
2146 {
2147         struct e1000_ring *tx_ring = adapter->tx_ring;
2148         struct e1000_buffer *buffer_info;
2149         unsigned long size;
2150         unsigned int i;
2151
2152         for (i = 0; i < tx_ring->count; i++) {
2153                 buffer_info = &tx_ring->buffer_info[i];
2154                 e1000_put_txbuf(adapter, buffer_info);
2155         }
2156
2157         size = sizeof(struct e1000_buffer) * tx_ring->count;
2158         memset(tx_ring->buffer_info, 0, size);
2159
2160         memset(tx_ring->desc, 0, tx_ring->size);
2161
2162         tx_ring->next_to_use = 0;
2163         tx_ring->next_to_clean = 0;
2164
2165         writel(0, adapter->hw.hw_addr + tx_ring->head);
2166         writel(0, adapter->hw.hw_addr + tx_ring->tail);
2167 }
2168
2169 /**
2170  * e1000e_free_tx_resources - Free Tx Resources per Queue
2171  * @adapter: board private structure
2172  *
2173  * Free all transmit software resources
2174  **/
2175 void e1000e_free_tx_resources(struct e1000_adapter *adapter)
2176 {
2177         struct pci_dev *pdev = adapter->pdev;
2178         struct e1000_ring *tx_ring = adapter->tx_ring;
2179
2180         e1000_clean_tx_ring(adapter);
2181
2182         vfree(tx_ring->buffer_info);
2183         tx_ring->buffer_info = NULL;
2184
2185         dma_free_coherent(&pdev->dev, tx_ring->size, tx_ring->desc,
2186                           tx_ring->dma);
2187         tx_ring->desc = NULL;
2188 }
2189
2190 /**
2191  * e1000e_free_rx_resources - Free Rx Resources
2192  * @adapter: board private structure
2193  *
2194  * Free all receive software resources
2195  **/
2196
2197 void e1000e_free_rx_resources(struct e1000_adapter *adapter)
2198 {
2199         struct pci_dev *pdev = adapter->pdev;
2200         struct e1000_ring *rx_ring = adapter->rx_ring;
2201         int i;
2202
2203         e1000_clean_rx_ring(adapter);
2204
2205         for (i = 0; i < rx_ring->count; i++)
2206                 kfree(rx_ring->buffer_info[i].ps_pages);
2207
2208         vfree(rx_ring->buffer_info);
2209         rx_ring->buffer_info = NULL;
2210
2211         dma_free_coherent(&pdev->dev, rx_ring->size, rx_ring->desc,
2212                           rx_ring->dma);
2213         rx_ring->desc = NULL;
2214 }
2215
2216 /**
2217  * e1000_update_itr - update the dynamic ITR value based on statistics
2218  * @adapter: pointer to adapter
2219  * @itr_setting: current adapter->itr
2220  * @packets: the number of packets during this measurement interval
2221  * @bytes: the number of bytes during this measurement interval
2222  *
2223  *      Stores a new ITR value based on packets and byte
2224  *      counts during the last interrupt.  The advantage of per interrupt
2225  *      computation is faster updates and more accurate ITR for the current
2226  *      traffic pattern.  Constants in this function were computed
2227  *      based on theoretical maximum wire speed and thresholds were set based
2228  *      on testing data as well as attempting to minimize response time
2229  *      while increasing bulk throughput.  This functionality is controlled
2230  *      by the InterruptThrottleRate module parameter.
2231  **/
2232 static unsigned int e1000_update_itr(struct e1000_adapter *adapter,
2233                                      u16 itr_setting, int packets,
2234                                      int bytes)
2235 {
2236         unsigned int retval = itr_setting;
2237
2238         if (packets == 0)
2239                 goto update_itr_done;
2240
2241         switch (itr_setting) {
2242         case lowest_latency:
2243                 /* handle TSO and jumbo frames */
2244                 if (bytes/packets > 8000)
2245                         retval = bulk_latency;
2246                 else if ((packets < 5) && (bytes > 512))
2247                         retval = low_latency;
2248                 break;
2249         case low_latency:  /* 50 usec aka 20000 ints/s */
2250                 if (bytes > 10000) {
2251                         /* this if handles the TSO accounting */
2252                         if (bytes/packets > 8000)
2253                                 retval = bulk_latency;
2254                         else if ((packets < 10) || ((bytes/packets) > 1200))
2255                                 retval = bulk_latency;
2256                         else if ((packets > 35))
2257                                 retval = lowest_latency;
2258                 } else if (bytes/packets > 2000) {
2259                         retval = bulk_latency;
2260                 } else if (packets <= 2 && bytes < 512) {
2261                         retval = lowest_latency;
2262                 }
2263                 break;
2264         case bulk_latency: /* 250 usec aka 4000 ints/s */
2265                 if (bytes > 25000) {
2266                         if (packets > 35)
2267                                 retval = low_latency;
2268                 } else if (bytes < 6000) {
2269                         retval = low_latency;
2270                 }
2271                 break;
2272         }
2273
2274 update_itr_done:
2275         return retval;
2276 }
2277
2278 static void e1000_set_itr(struct e1000_adapter *adapter)
2279 {
2280         struct e1000_hw *hw = &adapter->hw;
2281         u16 current_itr;
2282         u32 new_itr = adapter->itr;
2283
2284         /* for non-gigabit speeds, just fix the interrupt rate at 4000 */
2285         if (adapter->link_speed != SPEED_1000) {
2286                 current_itr = 0;
2287                 new_itr = 4000;
2288                 goto set_itr_now;
2289         }
2290
2291         if (adapter->flags2 & FLAG2_DISABLE_AIM) {
2292                 new_itr = 0;
2293                 goto set_itr_now;
2294         }
2295
2296         adapter->tx_itr = e1000_update_itr(adapter,
2297                                     adapter->tx_itr,
2298                                     adapter->total_tx_packets,
2299                                     adapter->total_tx_bytes);
2300         /* conservative mode (itr 3) eliminates the lowest_latency setting */
2301         if (adapter->itr_setting == 3 && adapter->tx_itr == lowest_latency)
2302                 adapter->tx_itr = low_latency;
2303
2304         adapter->rx_itr = e1000_update_itr(adapter,
2305                                     adapter->rx_itr,
2306                                     adapter->total_rx_packets,
2307                                     adapter->total_rx_bytes);
2308         /* conservative mode (itr 3) eliminates the lowest_latency setting */
2309         if (adapter->itr_setting == 3 && adapter->rx_itr == lowest_latency)
2310                 adapter->rx_itr = low_latency;
2311
2312         current_itr = max(adapter->rx_itr, adapter->tx_itr);
2313
2314         switch (current_itr) {
2315         /* counts and packets in update_itr are dependent on these numbers */
2316         case lowest_latency:
2317                 new_itr = 70000;
2318                 break;
2319         case low_latency:
2320                 new_itr = 20000; /* aka hwitr = ~200 */
2321                 break;
2322         case bulk_latency:
2323                 new_itr = 4000;
2324                 break;
2325         default:
2326                 break;
2327         }
2328
2329 set_itr_now:
2330         if (new_itr != adapter->itr) {
2331                 /*
2332                  * this attempts to bias the interrupt rate towards Bulk
2333                  * by adding intermediate steps when interrupt rate is
2334                  * increasing
2335                  */
2336                 new_itr = new_itr > adapter->itr ?
2337                              min(adapter->itr + (new_itr >> 2), new_itr) :
2338                              new_itr;
2339                 adapter->itr = new_itr;
2340                 adapter->rx_ring->itr_val = new_itr;
2341                 if (adapter->msix_entries)
2342                         adapter->rx_ring->set_itr = 1;
2343                 else
2344                         if (new_itr)
2345                                 ew32(ITR, 1000000000 / (new_itr * 256));
2346                         else
2347                                 ew32(ITR, 0);
2348         }
2349 }
2350
2351 /**
2352  * e1000_alloc_queues - Allocate memory for all rings
2353  * @adapter: board private structure to initialize
2354  **/
2355 static int __devinit e1000_alloc_queues(struct e1000_adapter *adapter)
2356 {
2357         adapter->tx_ring = kzalloc(sizeof(struct e1000_ring), GFP_KERNEL);
2358         if (!adapter->tx_ring)
2359                 goto err;
2360
2361         adapter->rx_ring = kzalloc(sizeof(struct e1000_ring), GFP_KERNEL);
2362         if (!adapter->rx_ring)
2363                 goto err;
2364
2365         return 0;
2366 err:
2367         e_err("Unable to allocate memory for queues\n");
2368         kfree(adapter->rx_ring);
2369         kfree(adapter->tx_ring);
2370         return -ENOMEM;
2371 }
2372
2373 /**
2374  * e1000_clean - NAPI Rx polling callback
2375  * @napi: struct associated with this polling callback
2376  * @budget: amount of packets driver is allowed to process this poll
2377  **/
2378 static int e1000_clean(struct napi_struct *napi, int budget)
2379 {
2380         struct e1000_adapter *adapter = container_of(napi, struct e1000_adapter, napi);
2381         struct e1000_hw *hw = &adapter->hw;
2382         struct net_device *poll_dev = adapter->netdev;
2383         int tx_cleaned = 1, work_done = 0;
2384
2385         adapter = netdev_priv(poll_dev);
2386
2387         if (adapter->msix_entries &&
2388             !(adapter->rx_ring->ims_val & adapter->tx_ring->ims_val))
2389                 goto clean_rx;
2390
2391         tx_cleaned = e1000_clean_tx_irq(adapter);
2392
2393 clean_rx:
2394         adapter->clean_rx(adapter, &work_done, budget);
2395
2396         if (!tx_cleaned)
2397                 work_done = budget;
2398
2399         /* If budget not fully consumed, exit the polling mode */
2400         if (work_done < budget) {
2401                 if (adapter->itr_setting & 3)
2402                         e1000_set_itr(adapter);
2403                 napi_complete(napi);
2404                 if (!test_bit(__E1000_DOWN, &adapter->state)) {
2405                         if (adapter->msix_entries)
2406                                 ew32(IMS, adapter->rx_ring->ims_val);
2407                         else
2408                                 e1000_irq_enable(adapter);
2409                 }
2410         }
2411
2412         return work_done;
2413 }
2414
2415 static void e1000_vlan_rx_add_vid(struct net_device *netdev, u16 vid)
2416 {
2417         struct e1000_adapter *adapter = netdev_priv(netdev);
2418         struct e1000_hw *hw = &adapter->hw;
2419         u32 vfta, index;
2420
2421         /* don't update vlan cookie if already programmed */
2422         if ((adapter->hw.mng_cookie.status &
2423              E1000_MNG_DHCP_COOKIE_STATUS_VLAN) &&
2424             (vid == adapter->mng_vlan_id))
2425                 return;
2426
2427         /* add VID to filter table */
2428         if (adapter->flags & FLAG_HAS_HW_VLAN_FILTER) {
2429                 index = (vid >> 5) & 0x7F;
2430                 vfta = E1000_READ_REG_ARRAY(hw, E1000_VFTA, index);
2431                 vfta |= (1 << (vid & 0x1F));
2432                 hw->mac.ops.write_vfta(hw, index, vfta);
2433         }
2434 }
2435
2436 static void e1000_vlan_rx_kill_vid(struct net_device *netdev, u16 vid)
2437 {
2438         struct e1000_adapter *adapter = netdev_priv(netdev);
2439         struct e1000_hw *hw = &adapter->hw;
2440         u32 vfta, index;
2441
2442         if (!test_bit(__E1000_DOWN, &adapter->state))
2443                 e1000_irq_disable(adapter);
2444         vlan_group_set_device(adapter->vlgrp, vid, NULL);
2445
2446         if (!test_bit(__E1000_DOWN, &adapter->state))
2447                 e1000_irq_enable(adapter);
2448
2449         if ((adapter->hw.mng_cookie.status &
2450              E1000_MNG_DHCP_COOKIE_STATUS_VLAN) &&
2451             (vid == adapter->mng_vlan_id)) {
2452                 /* release control to f/w */
2453                 e1000e_release_hw_control(adapter);
2454                 return;
2455         }
2456
2457         /* remove VID from filter table */
2458         if (adapter->flags & FLAG_HAS_HW_VLAN_FILTER) {
2459                 index = (vid >> 5) & 0x7F;
2460                 vfta = E1000_READ_REG_ARRAY(hw, E1000_VFTA, index);
2461                 vfta &= ~(1 << (vid & 0x1F));
2462                 hw->mac.ops.write_vfta(hw, index, vfta);
2463         }
2464 }
2465
2466 static void e1000_update_mng_vlan(struct e1000_adapter *adapter)
2467 {
2468         struct net_device *netdev = adapter->netdev;
2469         u16 vid = adapter->hw.mng_cookie.vlan_id;
2470         u16 old_vid = adapter->mng_vlan_id;
2471
2472         if (!adapter->vlgrp)
2473                 return;
2474
2475         if (!vlan_group_get_device(adapter->vlgrp, vid)) {
2476                 adapter->mng_vlan_id = E1000_MNG_VLAN_NONE;
2477                 if (adapter->hw.mng_cookie.status &
2478                         E1000_MNG_DHCP_COOKIE_STATUS_VLAN) {
2479                         e1000_vlan_rx_add_vid(netdev, vid);
2480                         adapter->mng_vlan_id = vid;
2481                 }
2482
2483                 if ((old_vid != (u16)E1000_MNG_VLAN_NONE) &&
2484                                 (vid != old_vid) &&
2485                     !vlan_group_get_device(adapter->vlgrp, old_vid))
2486                         e1000_vlan_rx_kill_vid(netdev, old_vid);
2487         } else {
2488                 adapter->mng_vlan_id = vid;
2489         }
2490 }
2491
2492
2493 static void e1000_vlan_rx_register(struct net_device *netdev,
2494                                    struct vlan_group *grp)
2495 {
2496         struct e1000_adapter *adapter = netdev_priv(netdev);
2497         struct e1000_hw *hw = &adapter->hw;
2498         u32 ctrl, rctl;
2499
2500         if (!test_bit(__E1000_DOWN, &adapter->state))
2501                 e1000_irq_disable(adapter);
2502         adapter->vlgrp = grp;
2503
2504         if (grp) {
2505                 /* enable VLAN tag insert/strip */
2506                 ctrl = er32(CTRL);
2507                 ctrl |= E1000_CTRL_VME;
2508                 ew32(CTRL, ctrl);
2509
2510                 if (adapter->flags & FLAG_HAS_HW_VLAN_FILTER) {
2511                         /* enable VLAN receive filtering */
2512                         rctl = er32(RCTL);
2513                         rctl &= ~E1000_RCTL_CFIEN;
2514                         ew32(RCTL, rctl);
2515                         e1000_update_mng_vlan(adapter);
2516                 }
2517         } else {
2518                 /* disable VLAN tag insert/strip */
2519                 ctrl = er32(CTRL);
2520                 ctrl &= ~E1000_CTRL_VME;
2521                 ew32(CTRL, ctrl);
2522
2523                 if (adapter->flags & FLAG_HAS_HW_VLAN_FILTER) {
2524                         if (adapter->mng_vlan_id !=
2525                             (u16)E1000_MNG_VLAN_NONE) {
2526                                 e1000_vlan_rx_kill_vid(netdev,
2527                                                        adapter->mng_vlan_id);
2528                                 adapter->mng_vlan_id = E1000_MNG_VLAN_NONE;
2529                         }
2530                 }
2531         }
2532
2533         if (!test_bit(__E1000_DOWN, &adapter->state))
2534                 e1000_irq_enable(adapter);
2535 }
2536
2537 static void e1000_restore_vlan(struct e1000_adapter *adapter)
2538 {
2539         u16 vid;
2540
2541         e1000_vlan_rx_register(adapter->netdev, adapter->vlgrp);
2542
2543         if (!adapter->vlgrp)
2544                 return;
2545
2546         for (vid = 0; vid < VLAN_N_VID; vid++) {
2547                 if (!vlan_group_get_device(adapter->vlgrp, vid))
2548                         continue;
2549                 e1000_vlan_rx_add_vid(adapter->netdev, vid);
2550         }
2551 }
2552
2553 static void e1000_init_manageability_pt(struct e1000_adapter *adapter)
2554 {
2555         struct e1000_hw *hw = &adapter->hw;
2556         u32 manc, manc2h, mdef, i, j;
2557
2558         if (!(adapter->flags & FLAG_MNG_PT_ENABLED))
2559                 return;
2560
2561         manc = er32(MANC);
2562
2563         /*
2564          * enable receiving management packets to the host. this will probably
2565          * generate destination unreachable messages from the host OS, but
2566          * the packets will be handled on SMBUS
2567          */
2568         manc |= E1000_MANC_EN_MNG2HOST;
2569         manc2h = er32(MANC2H);
2570
2571         switch (hw->mac.type) {
2572         default:
2573                 manc2h |= (E1000_MANC2H_PORT_623 | E1000_MANC2H_PORT_664);
2574                 break;
2575         case e1000_82574:
2576         case e1000_82583:
2577                 /*
2578                  * Check if IPMI pass-through decision filter already exists;
2579                  * if so, enable it.
2580                  */
2581                 for (i = 0, j = 0; i < 8; i++) {
2582                         mdef = er32(MDEF(i));
2583
2584                         /* Ignore filters with anything other than IPMI ports */
2585                         if (mdef & ~(E1000_MDEF_PORT_623 | E1000_MDEF_PORT_664))
2586                                 continue;
2587
2588                         /* Enable this decision filter in MANC2H */
2589                         if (mdef)
2590                                 manc2h |= (1 << i);
2591
2592                         j |= mdef;
2593                 }
2594
2595                 if (j == (E1000_MDEF_PORT_623 | E1000_MDEF_PORT_664))
2596                         break;
2597
2598                 /* Create new decision filter in an empty filter */
2599                 for (i = 0, j = 0; i < 8; i++)
2600                         if (er32(MDEF(i)) == 0) {
2601                                 ew32(MDEF(i), (E1000_MDEF_PORT_623 |
2602                                                E1000_MDEF_PORT_664));
2603                                 manc2h |= (1 << 1);
2604                                 j++;
2605                                 break;
2606                         }
2607
2608                 if (!j)
2609                         e_warn("Unable to create IPMI pass-through filter\n");
2610                 break;
2611         }
2612
2613         ew32(MANC2H, manc2h);
2614         ew32(MANC, manc);
2615 }
2616
2617 /**
2618  * e1000_configure_tx - Configure Transmit Unit after Reset
2619  * @adapter: board private structure
2620  *
2621  * Configure the Tx unit of the MAC after a reset.
2622  **/
2623 static void e1000_configure_tx(struct e1000_adapter *adapter)
2624 {
2625         struct e1000_hw *hw = &adapter->hw;
2626         struct e1000_ring *tx_ring = adapter->tx_ring;
2627         u64 tdba;
2628         u32 tdlen, tctl, tipg, tarc;
2629         u32 ipgr1, ipgr2;
2630
2631         /* Setup the HW Tx Head and Tail descriptor pointers */
2632         tdba = tx_ring->dma;
2633         tdlen = tx_ring->count * sizeof(struct e1000_tx_desc);
2634         ew32(TDBAL, (tdba & DMA_BIT_MASK(32)));
2635         ew32(TDBAH, (tdba >> 32));
2636         ew32(TDLEN, tdlen);
2637         ew32(TDH, 0);
2638         ew32(TDT, 0);
2639         tx_ring->head = E1000_TDH;
2640         tx_ring->tail = E1000_TDT;
2641
2642         /* Set the default values for the Tx Inter Packet Gap timer */
2643         tipg = DEFAULT_82543_TIPG_IPGT_COPPER;          /*  8  */
2644         ipgr1 = DEFAULT_82543_TIPG_IPGR1;               /*  8  */
2645         ipgr2 = DEFAULT_82543_TIPG_IPGR2;               /*  6  */
2646
2647         if (adapter->flags & FLAG_TIPG_MEDIUM_FOR_80003ESLAN)
2648                 ipgr2 = DEFAULT_80003ES2LAN_TIPG_IPGR2; /*  7  */
2649
2650         tipg |= ipgr1 << E1000_TIPG_IPGR1_SHIFT;
2651         tipg |= ipgr2 << E1000_TIPG_IPGR2_SHIFT;
2652         ew32(TIPG, tipg);
2653
2654         /* Set the Tx Interrupt Delay register */
2655         ew32(TIDV, adapter->tx_int_delay);
2656         /* Tx irq moderation */
2657         ew32(TADV, adapter->tx_abs_int_delay);
2658
2659         if (adapter->flags2 & FLAG2_DMA_BURST) {
2660                 u32 txdctl = er32(TXDCTL(0));
2661                 txdctl &= ~(E1000_TXDCTL_PTHRESH | E1000_TXDCTL_HTHRESH |
2662                             E1000_TXDCTL_WTHRESH);
2663                 /*
2664                  * set up some performance related parameters to encourage the
2665                  * hardware to use the bus more efficiently in bursts, depends
2666                  * on the tx_int_delay to be enabled,
2667                  * wthresh = 5 ==> burst write a cacheline (64 bytes) at a time
2668                  * hthresh = 1 ==> prefetch when one or more available
2669                  * pthresh = 0x1f ==> prefetch if internal cache 31 or less
2670                  * BEWARE: this seems to work but should be considered first if
2671                  * there are Tx hangs or other Tx related bugs
2672                  */
2673                 txdctl |= E1000_TXDCTL_DMA_BURST_ENABLE;
2674                 ew32(TXDCTL(0), txdctl);
2675                 /* erratum work around: set txdctl the same for both queues */
2676                 ew32(TXDCTL(1), txdctl);
2677         }
2678
2679         /* Program the Transmit Control Register */
2680         tctl = er32(TCTL);
2681         tctl &= ~E1000_TCTL_CT;
2682         tctl |= E1000_TCTL_PSP | E1000_TCTL_RTLC |
2683                 (E1000_COLLISION_THRESHOLD << E1000_CT_SHIFT);
2684
2685         if (adapter->flags & FLAG_TARC_SPEED_MODE_BIT) {
2686                 tarc = er32(TARC(0));
2687                 /*
2688                  * set the speed mode bit, we'll clear it if we're not at
2689                  * gigabit link later
2690                  */
2691 #define SPEED_MODE_BIT (1 << 21)
2692                 tarc |= SPEED_MODE_BIT;
2693                 ew32(TARC(0), tarc);
2694         }
2695
2696         /* errata: program both queues to unweighted RR */
2697         if (adapter->flags & FLAG_TARC_SET_BIT_ZERO) {
2698                 tarc = er32(TARC(0));
2699                 tarc |= 1;
2700                 ew32(TARC(0), tarc);
2701                 tarc = er32(TARC(1));
2702                 tarc |= 1;
2703                 ew32(TARC(1), tarc);
2704         }
2705
2706         /* Setup Transmit Descriptor Settings for eop descriptor */
2707         adapter->txd_cmd = E1000_TXD_CMD_EOP | E1000_TXD_CMD_IFCS;
2708
2709         /* only set IDE if we are delaying interrupts using the timers */
2710         if (adapter->tx_int_delay)
2711                 adapter->txd_cmd |= E1000_TXD_CMD_IDE;
2712
2713         /* enable Report Status bit */
2714         adapter->txd_cmd |= E1000_TXD_CMD_RS;
2715
2716         ew32(TCTL, tctl);
2717
2718         e1000e_config_collision_dist(hw);
2719 }
2720
2721 /**
2722  * e1000_setup_rctl - configure the receive control registers
2723  * @adapter: Board private structure
2724  **/
2725 #define PAGE_USE_COUNT(S) (((S) >> PAGE_SHIFT) + \
2726                            (((S) & (PAGE_SIZE - 1)) ? 1 : 0))
2727 static void e1000_setup_rctl(struct e1000_adapter *adapter)
2728 {
2729         struct e1000_hw *hw = &adapter->hw;
2730         u32 rctl, rfctl;
2731         u32 psrctl = 0;
2732         u32 pages = 0;
2733
2734         /* Workaround Si errata on 82579 - configure jumbo frame flow */
2735         if (hw->mac.type == e1000_pch2lan) {
2736                 s32 ret_val;
2737
2738                 if (adapter->netdev->mtu > ETH_DATA_LEN)
2739                         ret_val = e1000_lv_jumbo_workaround_ich8lan(hw, true);
2740                 else
2741                         ret_val = e1000_lv_jumbo_workaround_ich8lan(hw, false);
2742
2743                 if (ret_val)
2744                         e_dbg("failed to enable jumbo frame workaround mode\n");
2745         }
2746
2747         /* Program MC offset vector base */
2748         rctl = er32(RCTL);
2749         rctl &= ~(3 << E1000_RCTL_MO_SHIFT);
2750         rctl |= E1000_RCTL_EN | E1000_RCTL_BAM |
2751                 E1000_RCTL_LBM_NO | E1000_RCTL_RDMTS_HALF |
2752                 (adapter->hw.mac.mc_filter_type << E1000_RCTL_MO_SHIFT);
2753
2754         /* Do not Store bad packets */
2755         rctl &= ~E1000_RCTL_SBP;
2756
2757         /* Enable Long Packet receive */
2758         if (adapter->netdev->mtu <= ETH_DATA_LEN)
2759                 rctl &= ~E1000_RCTL_LPE;
2760         else
2761                 rctl |= E1000_RCTL_LPE;
2762
2763         /* Some systems expect that the CRC is included in SMBUS traffic. The
2764          * hardware strips the CRC before sending to both SMBUS (BMC) and to
2765          * host memory when this is enabled
2766          */
2767         if (adapter->flags2 & FLAG2_CRC_STRIPPING)
2768                 rctl |= E1000_RCTL_SECRC;
2769
2770         /* Workaround Si errata on 82577 PHY - configure IPG for jumbos */
2771         if ((hw->phy.type == e1000_phy_82577) && (rctl & E1000_RCTL_LPE)) {
2772                 u16 phy_data;
2773
2774                 e1e_rphy(hw, PHY_REG(770, 26), &phy_data);
2775                 phy_data &= 0xfff8;
2776                 phy_data |= (1 << 2);
2777                 e1e_wphy(hw, PHY_REG(770, 26), phy_data);
2778
2779                 e1e_rphy(hw, 22, &phy_data);
2780                 phy_data &= 0x0fff;
2781                 phy_data |= (1 << 14);
2782                 e1e_wphy(hw, 0x10, 0x2823);
2783                 e1e_wphy(hw, 0x11, 0x0003);
2784                 e1e_wphy(hw, 22, phy_data);
2785         }
2786
2787         /* Setup buffer sizes */
2788         rctl &= ~E1000_RCTL_SZ_4096;
2789         rctl |= E1000_RCTL_BSEX;
2790         switch (adapter->rx_buffer_len) {
2791         case 2048:
2792         default:
2793                 rctl |= E1000_RCTL_SZ_2048;
2794                 rctl &= ~E1000_RCTL_BSEX;
2795                 break;
2796         case 4096:
2797                 rctl |= E1000_RCTL_SZ_4096;
2798                 break;
2799         case 8192:
2800                 rctl |= E1000_RCTL_SZ_8192;
2801                 break;
2802         case 16384:
2803                 rctl |= E1000_RCTL_SZ_16384;
2804                 break;
2805         }
2806
2807         /*
2808          * 82571 and greater support packet-split where the protocol
2809          * header is placed in skb->data and the packet data is
2810          * placed in pages hanging off of skb_shinfo(skb)->nr_frags.
2811          * In the case of a non-split, skb->data is linearly filled,
2812          * followed by the page buffers.  Therefore, skb->data is
2813          * sized to hold the largest protocol header.
2814          *
2815          * allocations using alloc_page take too long for regular MTU
2816          * so only enable packet split for jumbo frames
2817          *
2818          * Using pages when the page size is greater than 16k wastes
2819          * a lot of memory, since we allocate 3 pages at all times
2820          * per packet.
2821          */
2822         pages = PAGE_USE_COUNT(adapter->netdev->mtu);
2823         if (!(adapter->flags & FLAG_HAS_ERT) && (pages <= 3) &&
2824             (PAGE_SIZE <= 16384) && (rctl & E1000_RCTL_LPE))
2825                 adapter->rx_ps_pages = pages;
2826         else
2827                 adapter->rx_ps_pages = 0;
2828
2829         if (adapter->rx_ps_pages) {
2830                 /* Configure extra packet-split registers */
2831                 rfctl = er32(RFCTL);
2832                 rfctl |= E1000_RFCTL_EXTEN;
2833                 /*
2834                  * disable packet split support for IPv6 extension headers,
2835                  * because some malformed IPv6 headers can hang the Rx
2836                  */
2837                 rfctl |= (E1000_RFCTL_IPV6_EX_DIS |
2838                           E1000_RFCTL_NEW_IPV6_EXT_DIS);
2839
2840                 ew32(RFCTL, rfctl);
2841
2842                 /* Enable Packet split descriptors */
2843                 rctl |= E1000_RCTL_DTYP_PS;
2844
2845                 psrctl |= adapter->rx_ps_bsize0 >>
2846                         E1000_PSRCTL_BSIZE0_SHIFT;
2847
2848                 switch (adapter->rx_ps_pages) {
2849                 case 3:
2850                         psrctl |= PAGE_SIZE <<
2851                                 E1000_PSRCTL_BSIZE3_SHIFT;
2852                 case 2:
2853                         psrctl |= PAGE_SIZE <<
2854                                 E1000_PSRCTL_BSIZE2_SHIFT;
2855                 case 1:
2856                         psrctl |= PAGE_SIZE >>
2857                                 E1000_PSRCTL_BSIZE1_SHIFT;
2858                         break;
2859                 }
2860
2861                 ew32(PSRCTL, psrctl);
2862         }
2863
2864         ew32(RCTL, rctl);
2865         /* just started the receive unit, no need to restart */
2866         adapter->flags &= ~FLAG_RX_RESTART_NOW;
2867 }
2868
2869 /**
2870  * e1000_configure_rx - Configure Receive Unit after Reset
2871  * @adapter: board private structure
2872  *
2873  * Configure the Rx unit of the MAC after a reset.
2874  **/
2875 static void e1000_configure_rx(struct e1000_adapter *adapter)
2876 {
2877         struct e1000_hw *hw = &adapter->hw;
2878         struct e1000_ring *rx_ring = adapter->rx_ring;
2879         u64 rdba;
2880         u32 rdlen, rctl, rxcsum, ctrl_ext;
2881
2882         if (adapter->rx_ps_pages) {
2883                 /* this is a 32 byte descriptor */
2884                 rdlen = rx_ring->count *
2885                     sizeof(union e1000_rx_desc_packet_split);
2886                 adapter->clean_rx = e1000_clean_rx_irq_ps;
2887                 adapter->alloc_rx_buf = e1000_alloc_rx_buffers_ps;
2888         } else if (adapter->netdev->mtu > ETH_FRAME_LEN + ETH_FCS_LEN) {
2889                 rdlen = rx_ring->count * sizeof(struct e1000_rx_desc);
2890                 adapter->clean_rx = e1000_clean_jumbo_rx_irq;
2891                 adapter->alloc_rx_buf = e1000_alloc_jumbo_rx_buffers;
2892         } else {
2893                 rdlen = rx_ring->count * sizeof(struct e1000_rx_desc);
2894                 adapter->clean_rx = e1000_clean_rx_irq;
2895                 adapter->alloc_rx_buf = e1000_alloc_rx_buffers;
2896         }
2897
2898         /* disable receives while setting up the descriptors */
2899         rctl = er32(RCTL);
2900         ew32(RCTL, rctl & ~E1000_RCTL_EN);
2901         e1e_flush();
2902         msleep(10);
2903
2904         if (adapter->flags2 & FLAG2_DMA_BURST) {
2905                 /*
2906                  * set the writeback threshold (only takes effect if the RDTR
2907                  * is set). set GRAN=1 and write back up to 0x4 worth, and
2908                  * enable prefetching of 0x20 Rx descriptors
2909                  * granularity = 01
2910                  * wthresh = 04,
2911                  * hthresh = 04,
2912                  * pthresh = 0x20
2913                  */
2914                 ew32(RXDCTL(0), E1000_RXDCTL_DMA_BURST_ENABLE);
2915                 ew32(RXDCTL(1), E1000_RXDCTL_DMA_BURST_ENABLE);
2916
2917                 /*
2918                  * override the delay timers for enabling bursting, only if
2919                  * the value was not set by the user via module options
2920                  */
2921                 if (adapter->rx_int_delay == DEFAULT_RDTR)
2922                         adapter->rx_int_delay = BURST_RDTR;
2923                 if (adapter->rx_abs_int_delay == DEFAULT_RADV)
2924                         adapter->rx_abs_int_delay = BURST_RADV;
2925         }
2926
2927         /* set the Receive Delay Timer Register */
2928         ew32(RDTR, adapter->rx_int_delay);
2929
2930         /* irq moderation */
2931         ew32(RADV, adapter->rx_abs_int_delay);
2932         if ((adapter->itr_setting != 0) && (adapter->itr != 0))
2933                 ew32(ITR, 1000000000 / (adapter->itr * 256));
2934
2935         ctrl_ext = er32(CTRL_EXT);
2936         /* Auto-Mask interrupts upon ICR access */
2937         ctrl_ext |= E1000_CTRL_EXT_IAME;
2938         ew32(IAM, 0xffffffff);
2939         ew32(CTRL_EXT, ctrl_ext);
2940         e1e_flush();
2941
2942         /*
2943          * Setup the HW Rx Head and Tail Descriptor Pointers and
2944          * the Base and Length of the Rx Descriptor Ring
2945          */
2946         rdba = rx_ring->dma;
2947         ew32(RDBAL, (rdba & DMA_BIT_MASK(32)));
2948         ew32(RDBAH, (rdba >> 32));
2949         ew32(RDLEN, rdlen);
2950         ew32(RDH, 0);
2951         ew32(RDT, 0);
2952         rx_ring->head = E1000_RDH;
2953         rx_ring->tail = E1000_RDT;
2954
2955         /* Enable Receive Checksum Offload for TCP and UDP */
2956         rxcsum = er32(RXCSUM);
2957         if (adapter->flags & FLAG_RX_CSUM_ENABLED) {
2958                 rxcsum |= E1000_RXCSUM_TUOFL;
2959
2960                 /*
2961                  * IPv4 payload checksum for UDP fragments must be
2962                  * used in conjunction with packet-split.
2963                  */
2964                 if (adapter->rx_ps_pages)
2965                         rxcsum |= E1000_RXCSUM_IPPCSE;
2966         } else {
2967                 rxcsum &= ~E1000_RXCSUM_TUOFL;
2968                 /* no need to clear IPPCSE as it defaults to 0 */
2969         }
2970         ew32(RXCSUM, rxcsum);
2971
2972         /*
2973          * Enable early receives on supported devices, only takes effect when
2974          * packet size is equal or larger than the specified value (in 8 byte
2975          * units), e.g. using jumbo frames when setting to E1000_ERT_2048
2976          */
2977         if ((adapter->flags & FLAG_HAS_ERT) ||
2978             (adapter->hw.mac.type == e1000_pch2lan)) {
2979                 if (adapter->netdev->mtu > ETH_DATA_LEN) {
2980                         u32 rxdctl = er32(RXDCTL(0));
2981                         ew32(RXDCTL(0), rxdctl | 0x3);
2982                         if (adapter->flags & FLAG_HAS_ERT)
2983                                 ew32(ERT, E1000_ERT_2048 | (1 << 13));
2984                         /*
2985                          * With jumbo frames and early-receive enabled,
2986                          * excessive C-state transition latencies result in
2987                          * dropped transactions.
2988                          */
2989                         pm_qos_update_request(&adapter->netdev->pm_qos_req, 55);
2990                 } else {
2991                         pm_qos_update_request(&adapter->netdev->pm_qos_req,
2992                                               PM_QOS_DEFAULT_VALUE);
2993                 }
2994         }
2995
2996         /* Enable Receives */
2997         ew32(RCTL, rctl);
2998 }
2999
3000 /**
3001  *  e1000_update_mc_addr_list - Update Multicast addresses
3002  *  @hw: pointer to the HW structure
3003  *  @mc_addr_list: array of multicast addresses to program
3004  *  @mc_addr_count: number of multicast addresses to program
3005  *
3006  *  Updates the Multicast Table Array.
3007  *  The caller must have a packed mc_addr_list of multicast addresses.
3008  **/
3009 static void e1000_update_mc_addr_list(struct e1000_hw *hw, u8 *mc_addr_list,
3010                                       u32 mc_addr_count)
3011 {
3012         hw->mac.ops.update_mc_addr_list(hw, mc_addr_list, mc_addr_count);
3013 }
3014
3015 /**
3016  * e1000_set_multi - Multicast and Promiscuous mode set
3017  * @netdev: network interface device structure
3018  *
3019  * The set_multi entry point is called whenever the multicast address
3020  * list or the network interface flags are updated.  This routine is
3021  * responsible for configuring the hardware for proper multicast,
3022  * promiscuous mode, and all-multi behavior.
3023  **/
3024 static void e1000_set_multi(struct net_device *netdev)
3025 {
3026         struct e1000_adapter *adapter = netdev_priv(netdev);
3027         struct e1000_hw *hw = &adapter->hw;
3028         struct netdev_hw_addr *ha;
3029         u8  *mta_list;
3030         u32 rctl;
3031         int i;
3032
3033         /* Check for Promiscuous and All Multicast modes */
3034
3035         rctl = er32(RCTL);
3036
3037         if (netdev->flags & IFF_PROMISC) {
3038                 rctl |= (E1000_RCTL_UPE | E1000_RCTL_MPE);
3039                 rctl &= ~E1000_RCTL_VFE;
3040         } else {
3041                 if (netdev->flags & IFF_ALLMULTI) {
3042                         rctl |= E1000_RCTL_MPE;
3043                         rctl &= ~E1000_RCTL_UPE;
3044                 } else {
3045                         rctl &= ~(E1000_RCTL_UPE | E1000_RCTL_MPE);
3046                 }
3047                 if (adapter->flags & FLAG_HAS_HW_VLAN_FILTER)
3048                         rctl |= E1000_RCTL_VFE;
3049         }
3050
3051         ew32(RCTL, rctl);
3052
3053         if (!netdev_mc_empty(netdev)) {
3054                 mta_list = kmalloc(netdev_mc_count(netdev) * 6, GFP_ATOMIC);
3055                 if (!mta_list)
3056                         return;
3057
3058                 /* prepare a packed array of only addresses. */
3059                 i = 0;
3060                 netdev_for_each_mc_addr(ha, netdev)
3061                         memcpy(mta_list + (i++ * ETH_ALEN), ha->addr, ETH_ALEN);
3062
3063                 e1000_update_mc_addr_list(hw, mta_list, i);
3064                 kfree(mta_list);
3065         } else {
3066                 /*
3067                  * if we're called from probe, we might not have
3068                  * anything to do here, so clear out the list
3069                  */
3070                 e1000_update_mc_addr_list(hw, NULL, 0);
3071         }
3072 }
3073
3074 /**
3075  * e1000_configure - configure the hardware for Rx and Tx
3076  * @adapter: private board structure
3077  **/
3078 static void e1000_configure(struct e1000_adapter *adapter)
3079 {
3080         e1000_set_multi(adapter->netdev);
3081
3082         e1000_restore_vlan(adapter);
3083         e1000_init_manageability_pt(adapter);
3084
3085         e1000_configure_tx(adapter);
3086         e1000_setup_rctl(adapter);
3087         e1000_configure_rx(adapter);
3088         adapter->alloc_rx_buf(adapter, e1000_desc_unused(adapter->rx_ring));
3089 }
3090
3091 /**
3092  * e1000e_power_up_phy - restore link in case the phy was powered down
3093  * @adapter: address of board private structure
3094  *
3095  * The phy may be powered down to save power and turn off link when the
3096  * driver is unloaded and wake on lan is not enabled (among others)
3097  * *** this routine MUST be followed by a call to e1000e_reset ***
3098  **/
3099 void e1000e_power_up_phy(struct e1000_adapter *adapter)
3100 {
3101         if (adapter->hw.phy.ops.power_up)
3102                 adapter->hw.phy.ops.power_up(&adapter->hw);
3103
3104         adapter->hw.mac.ops.setup_link(&adapter->hw);
3105 }
3106
3107 /**
3108  * e1000_power_down_phy - Power down the PHY
3109  *
3110  * Power down the PHY so no link is implied when interface is down.
3111  * The PHY cannot be powered down if management or WoL is active.
3112  */
3113 static void e1000_power_down_phy(struct e1000_adapter *adapter)
3114 {
3115         /* WoL is enabled */
3116         if (adapter->wol)
3117                 return;
3118
3119         if (adapter->hw.phy.ops.power_down)
3120                 adapter->hw.phy.ops.power_down(&adapter->hw);
3121 }
3122
3123 /**
3124  * e1000e_reset - bring the hardware into a known good state
3125  *
3126  * This function boots the hardware and enables some settings that
3127  * require a configuration cycle of the hardware - those cannot be
3128  * set/changed during runtime. After reset the device needs to be
3129  * properly configured for Rx, Tx etc.
3130  */
3131 void e1000e_reset(struct e1000_adapter *adapter)
3132 {
3133         struct e1000_mac_info *mac = &adapter->hw.mac;
3134         struct e1000_fc_info *fc = &adapter->hw.fc;
3135         struct e1000_hw *hw = &adapter->hw;
3136         u32 tx_space, min_tx_space, min_rx_space;
3137         u32 pba = adapter->pba;
3138         u16 hwm;
3139
3140         /* reset Packet Buffer Allocation to default */
3141         ew32(PBA, pba);
3142
3143         if (adapter->max_frame_size > ETH_FRAME_LEN + ETH_FCS_LEN) {
3144                 /*
3145                  * To maintain wire speed transmits, the Tx FIFO should be
3146                  * large enough to accommodate two full transmit packets,
3147                  * rounded up to the next 1KB and expressed in KB.  Likewise,
3148                  * the Rx FIFO should be large enough to accommodate at least
3149                  * one full receive packet and is similarly rounded up and
3150                  * expressed in KB.
3151                  */
3152                 pba = er32(PBA);
3153                 /* upper 16 bits has Tx packet buffer allocation size in KB */
3154                 tx_space = pba >> 16;
3155                 /* lower 16 bits has Rx packet buffer allocation size in KB */
3156                 pba &= 0xffff;
3157                 /*
3158                  * the Tx fifo also stores 16 bytes of information about the Tx
3159                  * but don't include ethernet FCS because hardware appends it
3160                  */
3161                 min_tx_space = (adapter->max_frame_size +
3162                                 sizeof(struct e1000_tx_desc) -
3163                                 ETH_FCS_LEN) * 2;
3164                 min_tx_space = ALIGN(min_tx_space, 1024);
3165                 min_tx_space >>= 10;
3166                 /* software strips receive CRC, so leave room for it */
3167                 min_rx_space = adapter->max_frame_size;
3168                 min_rx_space = ALIGN(min_rx_space, 1024);
3169                 min_rx_space >>= 10;
3170
3171                 /*
3172                  * If current Tx allocation is less than the min Tx FIFO size,
3173                  * and the min Tx FIFO size is less than the current Rx FIFO
3174                  * allocation, take space away from current Rx allocation
3175                  */
3176                 if ((tx_space < min_tx_space) &&
3177                     ((min_tx_space - tx_space) < pba)) {
3178                         pba -= min_tx_space - tx_space;
3179
3180                         /*
3181                          * if short on Rx space, Rx wins and must trump Tx
3182                          * adjustment or use Early Receive if available
3183                          */
3184                         if ((pba < min_rx_space) &&
3185                             (!(adapter->flags & FLAG_HAS_ERT)))
3186                                 /* ERT enabled in e1000_configure_rx */
3187                                 pba = min_rx_space;
3188                 }
3189
3190                 ew32(PBA, pba);
3191         }
3192
3193         /*
3194          * flow control settings
3195          *
3196          * The high water mark must be low enough to fit one full frame
3197          * (or the size used for early receive) above it in the Rx FIFO.
3198          * Set it to the lower of:
3199          * - 90% of the Rx FIFO size, and
3200          * - the full Rx FIFO size minus the early receive size (for parts
3201          *   with ERT support assuming ERT set to E1000_ERT_2048), or
3202          * - the full Rx FIFO size minus one full frame
3203          */
3204         if (adapter->flags & FLAG_DISABLE_FC_PAUSE_TIME)
3205                 fc->pause_time = 0xFFFF;
3206         else
3207                 fc->pause_time = E1000_FC_PAUSE_TIME;
3208         fc->send_xon = 1;
3209         fc->current_mode = fc->requested_mode;
3210
3211         switch (hw->mac.type) {
3212         default:
3213                 if ((adapter->flags & FLAG_HAS_ERT) &&
3214                     (adapter->netdev->mtu > ETH_DATA_LEN))
3215                         hwm = min(((pba << 10) * 9 / 10),
3216                                   ((pba << 10) - (E1000_ERT_2048 << 3)));
3217                 else
3218                         hwm = min(((pba << 10) * 9 / 10),
3219                                   ((pba << 10) - adapter->max_frame_size));
3220
3221                 fc->high_water = hwm & E1000_FCRTH_RTH; /* 8-byte granularity */
3222                 fc->low_water = fc->high_water - 8;
3223                 break;
3224         case e1000_pchlan:
3225                 /*
3226                  * Workaround PCH LOM adapter hangs with certain network
3227                  * loads.  If hangs persist, try disabling Tx flow control.
3228                  */
3229                 if (adapter->netdev->mtu > ETH_DATA_LEN) {
3230                         fc->high_water = 0x3500;
3231                         fc->low_water  = 0x1500;
3232                 } else {
3233                         fc->high_water = 0x5000;
3234                         fc->low_water  = 0x3000;
3235                 }
3236                 fc->refresh_time = 0x1000;
3237                 break;
3238         case e1000_pch2lan:
3239                 fc->high_water = 0x05C20;
3240                 fc->low_water = 0x05048;
3241                 fc->pause_time = 0x0650;
3242                 fc->refresh_time = 0x0400;
3243                 if (adapter->netdev->mtu > ETH_DATA_LEN) {
3244                         pba = 14;
3245                         ew32(PBA, pba);
3246                 }
3247                 break;
3248         }
3249
3250         /*
3251          * Disable Adaptive Interrupt Moderation if 2 full packets cannot
3252          * fit in receive buffer and early-receive not supported.
3253          */
3254         if (adapter->itr_setting & 0x3) {
3255                 if (((adapter->max_frame_size * 2) > (pba << 10)) &&
3256                     !(adapter->flags & FLAG_HAS_ERT)) {
3257                         if (!(adapter->flags2 & FLAG2_DISABLE_AIM)) {
3258                                 dev_info(&adapter->pdev->dev,
3259                                         "Interrupt Throttle Rate turned off\n");
3260                                 adapter->flags2 |= FLAG2_DISABLE_AIM;
3261                                 ew32(ITR, 0);
3262                         }
3263                 } else if (adapter->flags2 & FLAG2_DISABLE_AIM) {
3264                         dev_info(&adapter->pdev->dev,
3265                                  "Interrupt Throttle Rate turned on\n");
3266                         adapter->flags2 &= ~FLAG2_DISABLE_AIM;
3267                         adapter->itr = 20000;
3268                         ew32(ITR, 1000000000 / (adapter->itr * 256));
3269                 }
3270         }
3271
3272         /* Allow time for pending master requests to run */
3273         mac->ops.reset_hw(hw);
3274
3275         /*
3276          * For parts with AMT enabled, let the firmware know
3277          * that the network interface is in control
3278          */
3279         if (adapter->flags & FLAG_HAS_AMT)
3280                 e1000e_get_hw_control(adapter);
3281
3282         ew32(WUC, 0);
3283
3284         if (mac->ops.init_hw(hw))
3285                 e_err("Hardware Error\n");
3286
3287         e1000_update_mng_vlan(adapter);
3288
3289         /* Enable h/w to recognize an 802.1Q VLAN Ethernet packet */
3290         ew32(VET, ETH_P_8021Q);
3291
3292         e1000e_reset_adaptive(hw);
3293
3294         if (!netif_running(adapter->netdev) &&
3295             !test_bit(__E1000_TESTING, &adapter->state)) {
3296                 e1000_power_down_phy(adapter);
3297                 return;
3298         }
3299
3300         e1000_get_phy_info(hw);
3301
3302         if ((adapter->flags & FLAG_HAS_SMART_POWER_DOWN) &&
3303             !(adapter->flags & FLAG_SMART_POWER_DOWN)) {
3304                 u16 phy_data = 0;
3305                 /*
3306                  * speed up time to link by disabling smart power down, ignore
3307                  * the return value of this function because there is nothing
3308                  * different we would do if it failed
3309                  */
3310                 e1e_rphy(hw, IGP02E1000_PHY_POWER_MGMT, &phy_data);
3311                 phy_data &= ~IGP02E1000_PM_SPD;
3312                 e1e_wphy(hw, IGP02E1000_PHY_POWER_MGMT, phy_data);
3313         }
3314 }
3315
3316 int e1000e_up(struct e1000_adapter *adapter)
3317 {
3318         struct e1000_hw *hw = &adapter->hw;
3319
3320         /* hardware has been reset, we need to reload some things */
3321         e1000_configure(adapter);
3322
3323         clear_bit(__E1000_DOWN, &adapter->state);
3324
3325         napi_enable(&adapter->napi);
3326         if (adapter->msix_entries)
3327                 e1000_configure_msix(adapter);
3328         e1000_irq_enable(adapter);
3329
3330         netif_wake_queue(adapter->netdev);
3331
3332         /* fire a link change interrupt to start the watchdog */
3333         if (adapter->msix_entries)
3334                 ew32(ICS, E1000_ICS_LSC | E1000_ICR_OTHER);
3335         else
3336                 ew32(ICS, E1000_ICS_LSC);
3337
3338         return 0;
3339 }
3340
3341 void e1000e_down(struct e1000_adapter *adapter)
3342 {
3343         struct net_device *netdev = adapter->netdev;
3344         struct e1000_hw *hw = &adapter->hw;
3345         u32 tctl, rctl;
3346
3347         /*
3348          * signal that we're down so the interrupt handler does not
3349          * reschedule our watchdog timer
3350          */
3351         set_bit(__E1000_DOWN, &adapter->state);
3352
3353         /* disable receives in the hardware */
3354         rctl = er32(RCTL);
3355         ew32(RCTL, rctl & ~E1000_RCTL_EN);
3356         /* flush and sleep below */
3357
3358         netif_stop_queue(netdev);
3359
3360         /* disable transmits in the hardware */
3361         tctl = er32(TCTL);
3362         tctl &= ~E1000_TCTL_EN;
3363         ew32(TCTL, tctl);
3364         /* flush both disables and wait for them to finish */
3365         e1e_flush();
3366         msleep(10);
3367
3368         napi_disable(&adapter->napi);
3369         e1000_irq_disable(adapter);
3370
3371         del_timer_sync(&adapter->watchdog_timer);
3372         del_timer_sync(&adapter->phy_info_timer);
3373
3374         netif_carrier_off(netdev);
3375         adapter->link_speed = 0;
3376         adapter->link_duplex = 0;
3377
3378         if (!pci_channel_offline(adapter->pdev))
3379                 e1000e_reset(adapter);
3380         e1000_clean_tx_ring(adapter);
3381         e1000_clean_rx_ring(adapter);
3382
3383         /*
3384          * TODO: for power management, we could drop the link and
3385          * pci_disable_device here.
3386          */
3387 }
3388
3389 void e1000e_reinit_locked(struct e1000_adapter *adapter)
3390 {
3391         might_sleep();
3392         while (test_and_set_bit(__E1000_RESETTING, &adapter->state))
3393                 msleep(1);
3394         e1000e_down(adapter);
3395         e1000e_up(adapter);
3396         clear_bit(__E1000_RESETTING, &adapter->state);
3397 }
3398
3399 /**
3400  * e1000_sw_init - Initialize general software structures (struct e1000_adapter)
3401  * @adapter: board private structure to initialize
3402  *
3403  * e1000_sw_init initializes the Adapter private data structure.
3404  * Fields are initialized based on PCI device information and
3405  * OS network device settings (MTU size).
3406  **/
3407 static int __devinit e1000_sw_init(struct e1000_adapter *adapter)
3408 {
3409         struct net_device *netdev = adapter->netdev;
3410
3411         adapter->rx_buffer_len = ETH_FRAME_LEN + VLAN_HLEN + ETH_FCS_LEN;
3412         adapter->rx_ps_bsize0 = 128;
3413         adapter->max_frame_size = netdev->mtu + ETH_HLEN + ETH_FCS_LEN;
3414         adapter->min_frame_size = ETH_ZLEN + ETH_FCS_LEN;
3415
3416         e1000e_set_interrupt_capability(adapter);
3417
3418         if (e1000_alloc_queues(adapter))
3419                 return -ENOMEM;
3420
3421         /* Explicitly disable IRQ since the NIC can be in any state. */
3422         e1000_irq_disable(adapter);
3423
3424         set_bit(__E1000_DOWN, &adapter->state);
3425         return 0;
3426 }
3427
3428 /**
3429  * e1000_intr_msi_test - Interrupt Handler
3430  * @irq: interrupt number
3431  * @data: pointer to a network interface device structure
3432  **/
3433 static irqreturn_t e1000_intr_msi_test(int irq, void *data)
3434 {
3435         struct net_device *netdev = data;
3436         struct e1000_adapter *adapter = netdev_priv(netdev);
3437         struct e1000_hw *hw = &adapter->hw;
3438         u32 icr = er32(ICR);
3439
3440         e_dbg("icr is %08X\n", icr);
3441         if (icr & E1000_ICR_RXSEQ) {
3442                 adapter->flags &= ~FLAG_MSI_TEST_FAILED;
3443                 wmb();
3444         }
3445
3446         return IRQ_HANDLED;
3447 }
3448
3449 /**
3450  * e1000_test_msi_interrupt - Returns 0 for successful test
3451  * @adapter: board private struct
3452  *
3453  * code flow taken from tg3.c
3454  **/
3455 static int e1000_test_msi_interrupt(struct e1000_adapter *adapter)
3456 {
3457         struct net_device *netdev = adapter->netdev;
3458         struct e1000_hw *hw = &adapter->hw;
3459         int err;
3460
3461         /* poll_enable hasn't been called yet, so don't need disable */
3462         /* clear any pending events */
3463         er32(ICR);
3464
3465         /* free the real vector and request a test handler */
3466         e1000_free_irq(adapter);
3467         e1000e_reset_interrupt_capability(adapter);
3468
3469         /* Assume that the test fails, if it succeeds then the test
3470          * MSI irq handler will unset this flag */
3471         adapter->flags |= FLAG_MSI_TEST_FAILED;
3472
3473         err = pci_enable_msi(adapter->pdev);
3474         if (err)
3475                 goto msi_test_failed;
3476
3477         err = request_irq(adapter->pdev->irq, e1000_intr_msi_test, 0,
3478                           netdev->name, netdev);
3479         if (err) {
3480                 pci_disable_msi(adapter->pdev);
3481                 goto msi_test_failed;
3482         }
3483
3484         wmb();
3485
3486         e1000_irq_enable(adapter);
3487
3488         /* fire an unusual interrupt on the test handler */
3489         ew32(ICS, E1000_ICS_RXSEQ);
3490         e1e_flush();
3491         msleep(50);
3492
3493         e1000_irq_disable(adapter);
3494
3495         rmb();
3496
3497         if (adapter->flags & FLAG_MSI_TEST_FAILED) {
3498                 adapter->int_mode = E1000E_INT_MODE_LEGACY;
3499                 e_info("MSI interrupt test failed, using legacy interrupt.\n");
3500         } else
3501                 e_dbg("MSI interrupt test succeeded!\n");
3502
3503         free_irq(adapter->pdev->irq, netdev);
3504         pci_disable_msi(adapter->pdev);
3505
3506 msi_test_failed:
3507         e1000e_set_interrupt_capability(adapter);
3508         return e1000_request_irq(adapter);
3509 }
3510
3511 /**
3512  * e1000_test_msi - Returns 0 if MSI test succeeds or INTx mode is restored
3513  * @adapter: board private struct
3514  *
3515  * code flow taken from tg3.c, called with e1000 interrupts disabled.
3516  **/
3517 static int e1000_test_msi(struct e1000_adapter *adapter)
3518 {
3519         int err;
3520         u16 pci_cmd;
3521
3522         if (!(adapter->flags & FLAG_MSI_ENABLED))
3523                 return 0;
3524
3525         /* disable SERR in case the MSI write causes a master abort */
3526         pci_read_config_word(adapter->pdev, PCI_COMMAND, &pci_cmd);
3527         if (pci_cmd & PCI_COMMAND_SERR)
3528                 pci_write_config_word(adapter->pdev, PCI_COMMAND,
3529                                       pci_cmd & ~PCI_COMMAND_SERR);
3530
3531         err = e1000_test_msi_interrupt(adapter);
3532
3533         /* re-enable SERR */
3534         if (pci_cmd & PCI_COMMAND_SERR) {
3535                 pci_read_config_word(adapter->pdev, PCI_COMMAND, &pci_cmd);
3536                 pci_cmd |= PCI_COMMAND_SERR;
3537                 pci_write_config_word(adapter->pdev, PCI_COMMAND, pci_cmd);
3538         }
3539
3540         return err;
3541 }
3542
3543 /**
3544  * e1000_open - Called when a network interface is made active
3545  * @netdev: network interface device structure
3546  *
3547  * Returns 0 on success, negative value on failure
3548  *
3549  * The open entry point is called when a network interface is made
3550  * active by the system (IFF_UP).  At this point all resources needed
3551  * for transmit and receive operations are allocated, the interrupt
3552  * handler is registered with the OS, the watchdog timer is started,
3553  * and the stack is notified that the interface is ready.
3554  **/
3555 static int e1000_open(struct net_device *netdev)
3556 {
3557         struct e1000_adapter *adapter = netdev_priv(netdev);
3558         struct e1000_hw *hw = &adapter->hw;
3559         struct pci_dev *pdev = adapter->pdev;
3560         int err;
3561
3562         /* disallow open during test */
3563         if (test_bit(__E1000_TESTING, &adapter->state))
3564                 return -EBUSY;
3565
3566         pm_runtime_get_sync(&pdev->dev);
3567
3568         netif_carrier_off(netdev);
3569
3570         /* allocate transmit descriptors */
3571         err = e1000e_setup_tx_resources(adapter);
3572         if (err)
3573                 goto err_setup_tx;
3574
3575         /* allocate receive descriptors */
3576         err = e1000e_setup_rx_resources(adapter);
3577         if (err)
3578                 goto err_setup_rx;
3579
3580         /*
3581          * If AMT is enabled, let the firmware know that the network
3582          * interface is now open and reset the part to a known state.
3583          */
3584         if (adapter->flags & FLAG_HAS_AMT) {
3585                 e1000e_get_hw_control(adapter);
3586                 e1000e_reset(adapter);
3587         }
3588
3589         e1000e_power_up_phy(adapter);
3590
3591         adapter->mng_vlan_id = E1000_MNG_VLAN_NONE;
3592         if ((adapter->hw.mng_cookie.status &
3593              E1000_MNG_DHCP_COOKIE_STATUS_VLAN))
3594                 e1000_update_mng_vlan(adapter);
3595
3596         /* DMA latency requirement to workaround early-receive/jumbo issue */
3597         if ((adapter->flags & FLAG_HAS_ERT) ||
3598             (adapter->hw.mac.type == e1000_pch2lan))
3599                 pm_qos_add_request(&adapter->netdev->pm_qos_req,
3600                                    PM_QOS_CPU_DMA_LATENCY,
3601                                    PM_QOS_DEFAULT_VALUE);
3602
3603         /*
3604          * before we allocate an interrupt, we must be ready to handle it.
3605          * Setting DEBUG_SHIRQ in the kernel makes it fire an interrupt
3606          * as soon as we call pci_request_irq, so we have to setup our
3607          * clean_rx handler before we do so.
3608          */
3609         e1000_configure(adapter);
3610
3611         err = e1000_request_irq(adapter);
3612         if (err)
3613                 goto err_req_irq;
3614
3615         /*
3616          * Work around PCIe errata with MSI interrupts causing some chipsets to
3617          * ignore e1000e MSI messages, which means we need to test our MSI
3618          * interrupt now
3619          */
3620         if (adapter->int_mode != E1000E_INT_MODE_LEGACY) {
3621                 err = e1000_test_msi(adapter);
3622                 if (err) {
3623                         e_err("Interrupt allocation failed\n");
3624                         goto err_req_irq;
3625                 }
3626         }
3627
3628         /* From here on the code is the same as e1000e_up() */
3629         clear_bit(__E1000_DOWN, &adapter->state);
3630
3631         napi_enable(&adapter->napi);
3632
3633         e1000_irq_enable(adapter);
3634
3635         netif_start_queue(netdev);
3636
3637         adapter->idle_check = true;
3638         pm_runtime_put(&pdev->dev);
3639
3640         /* fire a link status change interrupt to start the watchdog */
3641         if (adapter->msix_entries)
3642                 ew32(ICS, E1000_ICS_LSC | E1000_ICR_OTHER);
3643         else
3644                 ew32(ICS, E1000_ICS_LSC);
3645
3646         return 0;
3647
3648 err_req_irq:
3649         e1000e_release_hw_control(adapter);
3650         e1000_power_down_phy(adapter);
3651         e1000e_free_rx_resources(adapter);
3652 err_setup_rx:
3653         e1000e_free_tx_resources(adapter);
3654 err_setup_tx:
3655         e1000e_reset(adapter);
3656         pm_runtime_put_sync(&pdev->dev);
3657
3658         return err;
3659 }
3660
3661 /**
3662  * e1000_close - Disables a network interface
3663  * @netdev: network interface device structure
3664  *
3665  * Returns 0, this is not allowed to fail
3666  *
3667  * The close entry point is called when an interface is de-activated
3668  * by the OS.  The hardware is still under the drivers control, but
3669  * needs to be disabled.  A global MAC reset is issued to stop the
3670  * hardware, and all transmit and receive resources are freed.
3671  **/
3672 static int e1000_close(struct net_device *netdev)
3673 {
3674         struct e1000_adapter *adapter = netdev_priv(netdev);
3675         struct pci_dev *pdev = adapter->pdev;
3676
3677         WARN_ON(test_bit(__E1000_RESETTING, &adapter->state));
3678
3679         pm_runtime_get_sync(&pdev->dev);
3680
3681         if (!test_bit(__E1000_DOWN, &adapter->state)) {
3682                 e1000e_down(adapter);
3683                 e1000_free_irq(adapter);
3684         }
3685         e1000_power_down_phy(adapter);
3686
3687         e1000e_free_tx_resources(adapter);
3688         e1000e_free_rx_resources(adapter);
3689
3690         /*
3691          * kill manageability vlan ID if supported, but not if a vlan with
3692          * the same ID is registered on the host OS (let 8021q kill it)
3693          */
3694         if ((adapter->hw.mng_cookie.status &
3695                           E1000_MNG_DHCP_COOKIE_STATUS_VLAN) &&
3696              !(adapter->vlgrp &&
3697                vlan_group_get_device(adapter->vlgrp, adapter->mng_vlan_id)))
3698                 e1000_vlan_rx_kill_vid(netdev, adapter->mng_vlan_id);
3699
3700         /*
3701          * If AMT is enabled, let the firmware know that the network
3702          * interface is now closed
3703          */
3704         if ((adapter->flags & FLAG_HAS_AMT) &&
3705             !test_bit(__E1000_TESTING, &adapter->state))
3706                 e1000e_release_hw_control(adapter);
3707
3708         if ((adapter->flags & FLAG_HAS_ERT) ||
3709             (adapter->hw.mac.type == e1000_pch2lan))
3710                 pm_qos_remove_request(&adapter->netdev->pm_qos_req);
3711
3712         pm_runtime_put_sync(&pdev->dev);
3713
3714         return 0;
3715 }
3716 /**
3717  * e1000_set_mac - Change the Ethernet Address of the NIC
3718  * @netdev: network interface device structure
3719  * @p: pointer to an address structure
3720  *
3721  * Returns 0 on success, negative on failure
3722  **/
3723 static int e1000_set_mac(struct net_device *netdev, void *p)
3724 {
3725         struct e1000_adapter *adapter = netdev_priv(netdev);
3726         struct sockaddr *addr = p;
3727
3728         if (!is_valid_ether_addr(addr->sa_data))
3729                 return -EADDRNOTAVAIL;
3730
3731         memcpy(netdev->dev_addr, addr->sa_data, netdev->addr_len);
3732         memcpy(adapter->hw.mac.addr, addr->sa_data, netdev->addr_len);
3733
3734         e1000e_rar_set(&adapter->hw, adapter->hw.mac.addr, 0);
3735
3736         if (adapter->flags & FLAG_RESET_OVERWRITES_LAA) {
3737                 /* activate the work around */
3738                 e1000e_set_laa_state_82571(&adapter->hw, 1);
3739
3740                 /*
3741                  * Hold a copy of the LAA in RAR[14] This is done so that
3742                  * between the time RAR[0] gets clobbered  and the time it
3743                  * gets fixed (in e1000_watchdog), the actual LAA is in one
3744                  * of the RARs and no incoming packets directed to this port
3745                  * are dropped. Eventually the LAA will be in RAR[0] and
3746                  * RAR[14]
3747                  */
3748                 e1000e_rar_set(&adapter->hw,
3749                               adapter->hw.mac.addr,
3750                               adapter->hw.mac.rar_entry_count - 1);
3751         }
3752
3753         return 0;
3754 }
3755
3756 /**
3757  * e1000e_update_phy_task - work thread to update phy
3758  * @work: pointer to our work struct
3759  *
3760  * this worker thread exists because we must acquire a
3761  * semaphore to read the phy, which we could msleep while
3762  * waiting for it, and we can't msleep in a timer.
3763  **/
3764 static void e1000e_update_phy_task(struct work_struct *work)
3765 {
3766         struct e1000_adapter *adapter = container_of(work,
3767                                         struct e1000_adapter, update_phy_task);
3768         e1000_get_phy_info(&adapter->hw);
3769 }
3770
3771 /*
3772  * Need to wait a few seconds after link up to get diagnostic information from
3773  * the phy
3774  */
3775 static void e1000_update_phy_info(unsigned long data)
3776 {
3777         struct e1000_adapter *adapter = (struct e1000_adapter *) data;
3778         schedule_work(&adapter->update_phy_task);
3779 }
3780
3781 /**
3782  * e1000e_update_phy_stats - Update the PHY statistics counters
3783  * @adapter: board private structure
3784  **/
3785 static void e1000e_update_phy_stats(struct e1000_adapter *adapter)
3786 {
3787         struct e1000_hw *hw = &adapter->hw;
3788         s32 ret_val;
3789         u16 phy_data;
3790
3791         ret_val = hw->phy.ops.acquire(hw);
3792         if (ret_val)
3793                 return;
3794
3795         hw->phy.addr = 1;
3796
3797 #define HV_PHY_STATS_PAGE       778
3798         /*
3799          * A page set is expensive so check if already on desired page.
3800          * If not, set to the page with the PHY status registers.
3801          */
3802         ret_val = e1000e_read_phy_reg_mdic(hw, IGP01E1000_PHY_PAGE_SELECT,
3803                                            &phy_data);
3804         if (ret_val)
3805                 goto release;
3806         if (phy_data != (HV_PHY_STATS_PAGE << IGP_PAGE_SHIFT)) {
3807                 ret_val = e1000e_write_phy_reg_mdic(hw,
3808                                                     IGP01E1000_PHY_PAGE_SELECT,
3809                                                     (HV_PHY_STATS_PAGE <<
3810                                                      IGP_PAGE_SHIFT));
3811                 if (ret_val)
3812                         goto release;
3813         }
3814
3815         /* Read/clear the upper 16-bit registers and read/accumulate lower */
3816
3817         /* Single Collision Count */
3818         e1000e_read_phy_reg_mdic(hw, HV_SCC_UPPER & MAX_PHY_REG_ADDRESS,
3819                                  &phy_data);
3820         ret_val = e1000e_read_phy_reg_mdic(hw,
3821                                            HV_SCC_LOWER & MAX_PHY_REG_ADDRESS,
3822                                            &phy_data);
3823         if (!ret_val)
3824                 adapter->stats.scc += phy_data;
3825
3826         /* Excessive Collision Count */
3827         e1000e_read_phy_reg_mdic(hw, HV_ECOL_UPPER & MAX_PHY_REG_ADDRESS,
3828                                  &phy_data);
3829         ret_val = e1000e_read_phy_reg_mdic(hw,
3830                                            HV_ECOL_LOWER & MAX_PHY_REG_ADDRESS,
3831                                            &phy_data);
3832         if (!ret_val)
3833                 adapter->stats.ecol += phy_data;
3834
3835         /* Multiple Collision Count */
3836         e1000e_read_phy_reg_mdic(hw, HV_MCC_UPPER & MAX_PHY_REG_ADDRESS,
3837                                  &phy_data);
3838         ret_val = e1000e_read_phy_reg_mdic(hw,
3839                                            HV_MCC_LOWER & MAX_PHY_REG_ADDRESS,
3840                                            &phy_data);
3841         if (!ret_val)
3842                 adapter->stats.mcc += phy_data;
3843
3844         /* Late Collision Count */
3845         e1000e_read_phy_reg_mdic(hw, HV_LATECOL_UPPER & MAX_PHY_REG_ADDRESS,
3846                                  &phy_data);
3847         ret_val = e1000e_read_phy_reg_mdic(hw,
3848                                            HV_LATECOL_LOWER &
3849                                            MAX_PHY_REG_ADDRESS,
3850                                            &phy_data);
3851         if (!ret_val)
3852                 adapter->stats.latecol += phy_data;
3853
3854         /* Collision Count - also used for adaptive IFS */
3855         e1000e_read_phy_reg_mdic(hw, HV_COLC_UPPER & MAX_PHY_REG_ADDRESS,
3856                                  &phy_data);
3857         ret_val = e1000e_read_phy_reg_mdic(hw,
3858                                            HV_COLC_LOWER & MAX_PHY_REG_ADDRESS,
3859                                            &phy_data);
3860         if (!ret_val)
3861                 hw->mac.collision_delta = phy_data;
3862
3863         /* Defer Count */
3864         e1000e_read_phy_reg_mdic(hw, HV_DC_UPPER & MAX_PHY_REG_ADDRESS,
3865                                  &phy_data);
3866         ret_val = e1000e_read_phy_reg_mdic(hw,
3867                                            HV_DC_LOWER & MAX_PHY_REG_ADDRESS,
3868                                            &phy_data);
3869         if (!ret_val)
3870                 adapter->stats.dc += phy_data;
3871
3872         /* Transmit with no CRS */
3873         e1000e_read_phy_reg_mdic(hw, HV_TNCRS_UPPER & MAX_PHY_REG_ADDRESS,
3874                                  &phy_data);
3875         ret_val = e1000e_read_phy_reg_mdic(hw,
3876                                            HV_TNCRS_LOWER & MAX_PHY_REG_ADDRESS,
3877                                            &phy_data);
3878         if (!ret_val)
3879                 adapter->stats.tncrs += phy_data;
3880
3881 release:
3882         hw->phy.ops.release(hw);
3883 }
3884
3885 /**
3886  * e1000e_update_stats - Update the board statistics counters
3887  * @adapter: board private structure
3888  **/
3889 void e1000e_update_stats(struct e1000_adapter *adapter)
3890 {
3891         struct net_device *netdev = adapter->netdev;
3892         struct e1000_hw *hw = &adapter->hw;
3893         struct pci_dev *pdev = adapter->pdev;
3894
3895         /*
3896          * Prevent stats update while adapter is being reset, or if the pci
3897          * connection is down.
3898          */
3899         if (adapter->link_speed == 0)
3900                 return;
3901         if (pci_channel_offline(pdev))
3902                 return;
3903
3904         adapter->stats.crcerrs += er32(CRCERRS);
3905         adapter->stats.gprc += er32(GPRC);
3906         adapter->stats.gorc += er32(GORCL);
3907         er32(GORCH); /* Clear gorc */
3908         adapter->stats.bprc += er32(BPRC);
3909         adapter->stats.mprc += er32(MPRC);
3910         adapter->stats.roc += er32(ROC);
3911
3912         adapter->stats.mpc += er32(MPC);
3913
3914         /* Half-duplex statistics */
3915         if (adapter->link_duplex == HALF_DUPLEX) {
3916                 if (adapter->flags2 & FLAG2_HAS_PHY_STATS) {
3917                         e1000e_update_phy_stats(adapter);
3918                 } else {
3919                         adapter->stats.scc += er32(SCC);
3920                         adapter->stats.ecol += er32(ECOL);
3921                         adapter->stats.mcc += er32(MCC);
3922                         adapter->stats.latecol += er32(LATECOL);
3923                         adapter->stats.dc += er32(DC);
3924
3925                         hw->mac.collision_delta = er32(COLC);
3926
3927                         if ((hw->mac.type != e1000_82574) &&
3928                             (hw->mac.type != e1000_82583))
3929                                 adapter->stats.tncrs += er32(TNCRS);
3930                 }
3931                 adapter->stats.colc += hw->mac.collision_delta;
3932         }
3933
3934         adapter->stats.xonrxc += er32(XONRXC);
3935         adapter->stats.xontxc += er32(XONTXC);
3936         adapter->stats.xoffrxc += er32(XOFFRXC);
3937         adapter->stats.xofftxc += er32(XOFFTXC);
3938         adapter->stats.gptc += er32(GPTC);
3939         adapter->stats.gotc += er32(GOTCL);
3940         er32(GOTCH); /* Clear gotc */
3941         adapter->stats.rnbc += er32(RNBC);
3942         adapter->stats.ruc += er32(RUC);
3943
3944         adapter->stats.mptc += er32(MPTC);
3945         adapter->stats.bptc += er32(BPTC);
3946
3947         /* used for adaptive IFS */
3948
3949         hw->mac.tx_packet_delta = er32(TPT);
3950         adapter->stats.tpt += hw->mac.tx_packet_delta;
3951
3952         adapter->stats.algnerrc += er32(ALGNERRC);
3953         adapter->stats.rxerrc += er32(RXERRC);
3954         adapter->stats.cexterr += er32(CEXTERR);
3955         adapter->stats.tsctc += er32(TSCTC);
3956         adapter->stats.tsctfc += er32(TSCTFC);
3957
3958         /* Fill out the OS statistics structure */
3959         netdev->stats.multicast = adapter->stats.mprc;
3960         netdev->stats.collisions = adapter->stats.colc;
3961
3962         /* Rx Errors */
3963
3964         /*
3965          * RLEC on some newer hardware can be incorrect so build
3966          * our own version based on RUC and ROC
3967          */
3968         netdev->stats.rx_errors = adapter->stats.rxerrc +
3969                 adapter->stats.crcerrs + adapter->stats.algnerrc +
3970                 adapter->stats.ruc + adapter->stats.roc +
3971                 adapter->stats.cexterr;
3972         netdev->stats.rx_length_errors = adapter->stats.ruc +
3973                                               adapter->stats.roc;
3974         netdev->stats.rx_crc_errors = adapter->stats.crcerrs;
3975         netdev->stats.rx_frame_errors = adapter->stats.algnerrc;
3976         netdev->stats.rx_missed_errors = adapter->stats.mpc;
3977
3978         /* Tx Errors */
3979         netdev->stats.tx_errors = adapter->stats.ecol +
3980                                        adapter->stats.latecol;
3981         netdev->stats.tx_aborted_errors = adapter->stats.ecol;
3982         netdev->stats.tx_window_errors = adapter->stats.latecol;
3983         netdev->stats.tx_carrier_errors = adapter->stats.tncrs;
3984
3985         /* Tx Dropped needs to be maintained elsewhere */
3986
3987         /* Management Stats */
3988         adapter->stats.mgptc += er32(MGTPTC);
3989         adapter->stats.mgprc += er32(MGTPRC);
3990         adapter->stats.mgpdc += er32(MGTPDC);
3991 }
3992
3993 /**
3994  * e1000_phy_read_status - Update the PHY register status snapshot
3995  * @adapter: board private structure
3996  **/
3997 static void e1000_phy_read_status(struct e1000_adapter *adapter)
3998 {
3999         struct e1000_hw *hw = &adapter->hw;
4000         struct e1000_phy_regs *phy = &adapter->phy_regs;
4001         int ret_val;
4002
4003         if ((er32(STATUS) & E1000_STATUS_LU) &&
4004             (adapter->hw.phy.media_type == e1000_media_type_copper)) {
4005                 ret_val  = e1e_rphy(hw, PHY_CONTROL, &phy->bmcr);
4006                 ret_val |= e1e_rphy(hw, PHY_STATUS, &phy->bmsr);
4007                 ret_val |= e1e_rphy(hw, PHY_AUTONEG_ADV, &phy->advertise);
4008                 ret_val |= e1e_rphy(hw, PHY_LP_ABILITY, &phy->lpa);
4009                 ret_val |= e1e_rphy(hw, PHY_AUTONEG_EXP, &phy->expansion);
4010                 ret_val |= e1e_rphy(hw, PHY_1000T_CTRL, &phy->ctrl1000);
4011                 ret_val |= e1e_rphy(hw, PHY_1000T_STATUS, &phy->stat1000);
4012                 ret_val |= e1e_rphy(hw, PHY_EXT_STATUS, &phy->estatus);
4013                 if (ret_val)
4014                         e_warn("Error reading PHY register\n");
4015         } else {
4016                 /*
4017                  * Do not read PHY registers if link is not up
4018                  * Set values to typical power-on defaults
4019                  */
4020                 phy->bmcr = (BMCR_SPEED1000 | BMCR_ANENABLE | BMCR_FULLDPLX);
4021                 phy->bmsr = (BMSR_100FULL | BMSR_100HALF | BMSR_10FULL |
4022                              BMSR_10HALF | BMSR_ESTATEN | BMSR_ANEGCAPABLE |
4023                              BMSR_ERCAP);
4024                 phy->advertise = (ADVERTISE_PAUSE_ASYM | ADVERTISE_PAUSE_CAP |
4025                                   ADVERTISE_ALL | ADVERTISE_CSMA);
4026                 phy->lpa = 0;
4027                 phy->expansion = EXPANSION_ENABLENPAGE;
4028                 phy->ctrl1000 = ADVERTISE_1000FULL;
4029                 phy->stat1000 = 0;
4030                 phy->estatus = (ESTATUS_1000_TFULL | ESTATUS_1000_THALF);
4031         }
4032 }
4033
4034 static void e1000_print_link_info(struct e1000_adapter *adapter)
4035 {
4036         struct e1000_hw *hw = &adapter->hw;
4037         u32 ctrl = er32(CTRL);
4038
4039         /* Link status message must follow this format for user tools */
4040         printk(KERN_INFO "e1000e: %s NIC Link is Up %d Mbps %s, "
4041                "Flow Control: %s\n",
4042                adapter->netdev->name,
4043                adapter->link_speed,
4044                (adapter->link_duplex == FULL_DUPLEX) ?
4045                "Full Duplex" : "Half Duplex",
4046                ((ctrl & E1000_CTRL_TFCE) && (ctrl & E1000_CTRL_RFCE)) ?
4047                "Rx/Tx" :
4048                ((ctrl & E1000_CTRL_RFCE) ? "Rx" :
4049                 ((ctrl & E1000_CTRL_TFCE) ? "Tx" : "None")));
4050 }
4051
4052 static bool e1000e_has_link(struct e1000_adapter *adapter)
4053 {
4054         struct e1000_hw *hw = &adapter->hw;
4055         bool link_active = 0;
4056         s32 ret_val = 0;
4057
4058         /*
4059          * get_link_status is set on LSC (link status) interrupt or
4060          * Rx sequence error interrupt.  get_link_status will stay
4061          * false until the check_for_link establishes link
4062          * for copper adapters ONLY
4063          */
4064         switch (hw->phy.media_type) {
4065         case e1000_media_type_copper:
4066                 if (hw->mac.get_link_status) {
4067                         ret_val = hw->mac.ops.check_for_link(hw);
4068                         link_active = !hw->mac.get_link_status;
4069                 } else {
4070                         link_active = 1;
4071                 }
4072                 break;
4073         case e1000_media_type_fiber:
4074                 ret_val = hw->mac.ops.check_for_link(hw);
4075                 link_active = !!(er32(STATUS) & E1000_STATUS_LU);
4076                 break;
4077         case e1000_media_type_internal_serdes:
4078                 ret_val = hw->mac.ops.check_for_link(hw);
4079                 link_active = adapter->hw.mac.serdes_has_link;
4080                 break;
4081         default:
4082         case e1000_media_type_unknown:
4083                 break;
4084         }
4085
4086         if ((ret_val == E1000_ERR_PHY) && (hw->phy.type == e1000_phy_igp_3) &&
4087             (er32(CTRL) & E1000_PHY_CTRL_GBE_DISABLE)) {
4088                 /* See e1000_kmrn_lock_loss_workaround_ich8lan() */
4089                 e_info("Gigabit has been disabled, downgrading speed\n");
4090         }
4091
4092         return link_active;
4093 }
4094
4095 static void e1000e_enable_receives(struct e1000_adapter *adapter)
4096 {
4097         /* make sure the receive unit is started */
4098         if ((adapter->flags & FLAG_RX_NEEDS_RESTART) &&
4099             (adapter->flags & FLAG_RX_RESTART_NOW)) {
4100                 struct e1000_hw *hw = &adapter->hw;
4101                 u32 rctl = er32(RCTL);
4102                 ew32(RCTL, rctl | E1000_RCTL_EN);
4103                 adapter->flags &= ~FLAG_RX_RESTART_NOW;
4104         }
4105 }
4106
4107 static void e1000e_check_82574_phy_workaround(struct e1000_adapter *adapter)
4108 {
4109         struct e1000_hw *hw = &adapter->hw;
4110
4111         /*
4112          * With 82574 controllers, PHY needs to be checked periodically
4113          * for hung state and reset, if two calls return true
4114          */
4115         if (e1000_check_phy_82574(hw))
4116                 adapter->phy_hang_count++;
4117         else
4118                 adapter->phy_hang_count = 0;
4119
4120         if (adapter->phy_hang_count > 1) {
4121                 adapter->phy_hang_count = 0;
4122                 schedule_work(&adapter->reset_task);
4123         }
4124 }
4125
4126 /**
4127  * e1000_watchdog - Timer Call-back
4128  * @data: pointer to adapter cast into an unsigned long
4129  **/
4130 static void e1000_watchdog(unsigned long data)
4131 {
4132         struct e1000_adapter *adapter = (struct e1000_adapter *) data;
4133
4134         /* Do the rest outside of interrupt context */
4135         schedule_work(&adapter->watchdog_task);
4136
4137         /* TODO: make this use queue_delayed_work() */
4138 }
4139
4140 static void e1000_watchdog_task(struct work_struct *work)
4141 {
4142         struct e1000_adapter *adapter = container_of(work,
4143                                         struct e1000_adapter, watchdog_task);
4144         struct net_device *netdev = adapter->netdev;
4145         struct e1000_mac_info *mac = &adapter->hw.mac;
4146         struct e1000_phy_info *phy = &adapter->hw.phy;
4147         struct e1000_ring *tx_ring = adapter->tx_ring;
4148         struct e1000_hw *hw = &adapter->hw;
4149         u32 link, tctl;
4150         int tx_pending = 0;
4151
4152         link = e1000e_has_link(adapter);
4153         if ((netif_carrier_ok(netdev)) && link) {
4154                 /* Cancel scheduled suspend requests. */
4155                 pm_runtime_resume(netdev->dev.parent);
4156
4157                 e1000e_enable_receives(adapter);
4158                 goto link_up;
4159         }
4160
4161         if ((e1000e_enable_tx_pkt_filtering(hw)) &&
4162             (adapter->mng_vlan_id != adapter->hw.mng_cookie.vlan_id))
4163                 e1000_update_mng_vlan(adapter);
4164
4165         if (link) {
4166                 if (!netif_carrier_ok(netdev)) {
4167                         bool txb2b = 1;
4168
4169                         /* Cancel scheduled suspend requests. */
4170                         pm_runtime_resume(netdev->dev.parent);
4171
4172                         /* update snapshot of PHY registers on LSC */
4173                         e1000_phy_read_status(adapter);
4174                         mac->ops.get_link_up_info(&adapter->hw,
4175                                                    &adapter->link_speed,
4176                                                    &adapter->link_duplex);
4177                         e1000_print_link_info(adapter);
4178                         /*
4179                          * On supported PHYs, check for duplex mismatch only
4180                          * if link has autonegotiated at 10/100 half
4181                          */
4182                         if ((hw->phy.type == e1000_phy_igp_3 ||
4183                              hw->phy.type == e1000_phy_bm) &&
4184                             (hw->mac.autoneg == true) &&
4185                             (adapter->link_speed == SPEED_10 ||
4186                              adapter->link_speed == SPEED_100) &&
4187                             (adapter->link_duplex == HALF_DUPLEX)) {
4188                                 u16 autoneg_exp;
4189
4190                                 e1e_rphy(hw, PHY_AUTONEG_EXP, &autoneg_exp);
4191
4192                                 if (!(autoneg_exp & NWAY_ER_LP_NWAY_CAPS))
4193                                         e_info("Autonegotiated half duplex but"
4194                                                " link partner cannot autoneg. "
4195                                                " Try forcing full duplex if "
4196                                                "link gets many collisions.\n");
4197                         }
4198
4199                         /* adjust timeout factor according to speed/duplex */
4200                         adapter->tx_timeout_factor = 1;
4201                         switch (adapter->link_speed) {
4202                         case SPEED_10:
4203                                 txb2b = 0;
4204                                 adapter->tx_timeout_factor = 16;
4205                                 break;
4206                         case SPEED_100:
4207                                 txb2b = 0;
4208                                 adapter->tx_timeout_factor = 10;
4209                                 break;
4210                         }
4211
4212                         /*
4213                          * workaround: re-program speed mode bit after
4214                          * link-up event
4215                          */
4216                         if ((adapter->flags & FLAG_TARC_SPEED_MODE_BIT) &&
4217                             !txb2b) {
4218                                 u32 tarc0;
4219                                 tarc0 = er32(TARC(0));
4220                                 tarc0 &= ~SPEED_MODE_BIT;
4221                                 ew32(TARC(0), tarc0);
4222                         }
4223
4224                         /*
4225                          * disable TSO for pcie and 10/100 speeds, to avoid
4226                          * some hardware issues
4227                          */
4228                         if (!(adapter->flags & FLAG_TSO_FORCE)) {
4229                                 switch (adapter->link_speed) {
4230                                 case SPEED_10:
4231                                 case SPEED_100:
4232                                         e_info("10/100 speed: disabling TSO\n");
4233                                         netdev->features &= ~NETIF_F_TSO;
4234                                         netdev->features &= ~NETIF_F_TSO6;
4235                                         break;
4236                                 case SPEED_1000:
4237                                         netdev->features |= NETIF_F_TSO;
4238                                         netdev->features |= NETIF_F_TSO6;
4239                                         break;
4240                                 default:
4241                                         /* oops */
4242                                         break;
4243                                 }
4244                         }
4245
4246                         /*
4247                          * enable transmits in the hardware, need to do this
4248                          * after setting TARC(0)
4249                          */
4250                         tctl = er32(TCTL);
4251                         tctl |= E1000_TCTL_EN;
4252                         ew32(TCTL, tctl);
4253
4254                         /*
4255                          * Perform any post-link-up configuration before
4256                          * reporting link up.
4257                          */
4258                         if (phy->ops.cfg_on_link_up)
4259                                 phy->ops.cfg_on_link_up(hw);
4260
4261                         netif_carrier_on(netdev);
4262
4263                         if (!test_bit(__E1000_DOWN, &adapter->state))
4264                                 mod_timer(&adapter->phy_info_timer,
4265                                           round_jiffies(jiffies + 2 * HZ));
4266                 }
4267         } else {
4268                 if (netif_carrier_ok(netdev)) {
4269                         adapter->link_speed = 0;
4270                         adapter->link_duplex = 0;
4271                         /* Link status message must follow this format */
4272                         printk(KERN_INFO "e1000e: %s NIC Link is Down\n",
4273                                adapter->netdev->name);
4274                         netif_carrier_off(netdev);
4275                         if (!test_bit(__E1000_DOWN, &adapter->state))
4276                                 mod_timer(&adapter->phy_info_timer,
4277                                           round_jiffies(jiffies + 2 * HZ));
4278
4279                         if (adapter->flags & FLAG_RX_NEEDS_RESTART)
4280                                 schedule_work(&adapter->reset_task);
4281                         else
4282                                 pm_schedule_suspend(netdev->dev.parent,
4283                                                         LINK_TIMEOUT);
4284                 }
4285         }
4286
4287 link_up:
4288         e1000e_update_stats(adapter);
4289
4290         mac->tx_packet_delta = adapter->stats.tpt - adapter->tpt_old;
4291         adapter->tpt_old = adapter->stats.tpt;
4292         mac->collision_delta = adapter->stats.colc - adapter->colc_old;
4293         adapter->colc_old = adapter->stats.colc;
4294
4295         adapter->gorc = adapter->stats.gorc - adapter->gorc_old;
4296         adapter->gorc_old = adapter->stats.gorc;
4297         adapter->gotc = adapter->stats.gotc - adapter->gotc_old;
4298         adapter->gotc_old = adapter->stats.gotc;
4299
4300         e1000e_update_adaptive(&adapter->hw);
4301
4302         if (!netif_carrier_ok(netdev)) {
4303                 tx_pending = (e1000_desc_unused(tx_ring) + 1 <
4304                                tx_ring->count);
4305                 if (tx_pending) {
4306                         /*
4307                          * We've lost link, so the controller stops DMA,
4308                          * but we've got queued Tx work that's never going
4309                          * to get done, so reset controller to flush Tx.
4310                          * (Do the reset outside of interrupt context).
4311                          */
4312                         adapter->tx_timeout_count++;
4313                         schedule_work(&adapter->reset_task);
4314                         /* return immediately since reset is imminent */
4315                         return;
4316                 }
4317         }
4318
4319         /* Simple mode for Interrupt Throttle Rate (ITR) */
4320         if (adapter->itr_setting == 4) {
4321                 /*
4322                  * Symmetric Tx/Rx gets a reduced ITR=2000;
4323                  * Total asymmetrical Tx or Rx gets ITR=8000;
4324                  * everyone else is between 2000-8000.
4325                  */
4326                 u32 goc = (adapter->gotc + adapter->gorc) / 10000;
4327                 u32 dif = (adapter->gotc > adapter->gorc ?
4328                             adapter->gotc - adapter->gorc :
4329                             adapter->gorc - adapter->gotc) / 10000;
4330                 u32 itr = goc > 0 ? (dif * 6000 / goc + 2000) : 8000;
4331
4332                 ew32(ITR, 1000000000 / (itr * 256));
4333         }
4334
4335         /* Cause software interrupt to ensure Rx ring is cleaned */
4336         if (adapter->msix_entries)
4337                 ew32(ICS, adapter->rx_ring->ims_val);
4338         else
4339                 ew32(ICS, E1000_ICS_RXDMT0);
4340
4341         /* Force detection of hung controller every watchdog period */
4342         adapter->detect_tx_hung = 1;
4343
4344         /* flush partial descriptors to memory before detecting Tx hang */
4345         if (adapter->flags2 & FLAG2_DMA_BURST) {
4346                 ew32(TIDV, adapter->tx_int_delay | E1000_TIDV_FPD);
4347                 ew32(RDTR, adapter->rx_int_delay | E1000_RDTR_FPD);
4348                 /*
4349                  * no need to flush the writes because the timeout code does
4350                  * an er32 first thing
4351                  */
4352         }
4353
4354         /*
4355          * With 82571 controllers, LAA may be overwritten due to controller
4356          * reset from the other port. Set the appropriate LAA in RAR[0]
4357          */
4358         if (e1000e_get_laa_state_82571(hw))
4359                 e1000e_rar_set(hw, adapter->hw.mac.addr, 0);
4360
4361         if (adapter->flags2 & FLAG2_CHECK_PHY_HANG)
4362                 e1000e_check_82574_phy_workaround(adapter);
4363
4364         /* Reset the timer */
4365         if (!test_bit(__E1000_DOWN, &adapter->state))
4366                 mod_timer(&adapter->watchdog_timer,
4367                           round_jiffies(jiffies + 2 * HZ));
4368 }
4369
4370 #define E1000_TX_FLAGS_CSUM             0x00000001
4371 #define E1000_TX_FLAGS_VLAN             0x00000002
4372 #define E1000_TX_FLAGS_TSO              0x00000004
4373 #define E1000_TX_FLAGS_IPV4             0x00000008
4374 #define E1000_TX_FLAGS_VLAN_MASK        0xffff0000
4375 #define E1000_TX_FLAGS_VLAN_SHIFT       16
4376
4377 static int e1000_tso(struct e1000_adapter *adapter,
4378                      struct sk_buff *skb)
4379 {
4380         struct e1000_ring *tx_ring = adapter->tx_ring;
4381         struct e1000_context_desc *context_desc;
4382         struct e1000_buffer *buffer_info;
4383         unsigned int i;
4384         u32 cmd_length = 0;
4385         u16 ipcse = 0, tucse, mss;
4386         u8 ipcss, ipcso, tucss, tucso, hdr_len;
4387         int err;
4388
4389         if (!skb_is_gso(skb))
4390                 return 0;
4391
4392         if (skb_header_cloned(skb)) {
4393                 err = pskb_expand_head(skb, 0, 0, GFP_ATOMIC);
4394                 if (err)
4395                         return err;
4396         }
4397
4398         hdr_len = skb_transport_offset(skb) + tcp_hdrlen(skb);
4399         mss = skb_shinfo(skb)->gso_size;
4400         if (skb->protocol == htons(ETH_P_IP)) {
4401                 struct iphdr *iph = ip_hdr(skb);
4402                 iph->tot_len = 0;
4403                 iph->check = 0;
4404                 tcp_hdr(skb)->check = ~csum_tcpudp_magic(iph->saddr, iph->daddr,
4405                                                          0, IPPROTO_TCP, 0);
4406                 cmd_length = E1000_TXD_CMD_IP;
4407                 ipcse = skb_transport_offset(skb) - 1;
4408         } else if (skb_is_gso_v6(skb)) {
4409                 ipv6_hdr(skb)->payload_len = 0;
4410                 tcp_hdr(skb)->check = ~csum_ipv6_magic(&ipv6_hdr(skb)->saddr,
4411                                                        &ipv6_hdr(skb)->daddr,
4412                                                        0, IPPROTO_TCP, 0);
4413                 ipcse = 0;
4414         }
4415         ipcss = skb_network_offset(skb);
4416         ipcso = (void *)&(ip_hdr(skb)->check) - (void *)skb->data;
4417         tucss = skb_transport_offset(skb);
4418         tucso = (void *)&(tcp_hdr(skb)->check) - (void *)skb->data;
4419         tucse = 0;
4420
4421         cmd_length |= (E1000_TXD_CMD_DEXT | E1000_TXD_CMD_TSE |
4422                        E1000_TXD_CMD_TCP | (skb->len - (hdr_len)));
4423
4424         i = tx_ring->next_to_use;
4425         context_desc = E1000_CONTEXT_DESC(*tx_ring, i);
4426         buffer_info = &tx_ring->buffer_info[i];
4427
4428         context_desc->lower_setup.ip_fields.ipcss  = ipcss;
4429         context_desc->lower_setup.ip_fields.ipcso  = ipcso;
4430         context_desc->lower_setup.ip_fields.ipcse  = cpu_to_le16(ipcse);
4431         context_desc->upper_setup.tcp_fields.tucss = tucss;
4432         context_desc->upper_setup.tcp_fields.tucso = tucso;
4433         context_desc->upper_setup.tcp_fields.tucse = cpu_to_le16(tucse);
4434         context_desc->tcp_seg_setup.fields.mss     = cpu_to_le16(mss);
4435         context_desc->tcp_seg_setup.fields.hdr_len = hdr_len;
4436         context_desc->cmd_and_length = cpu_to_le32(cmd_length);
4437
4438         buffer_info->time_stamp = jiffies;
4439         buffer_info->next_to_watch = i;
4440
4441         i++;
4442         if (i == tx_ring->count)
4443                 i = 0;
4444         tx_ring->next_to_use = i;
4445
4446         return 1;
4447 }
4448
4449 static bool e1000_tx_csum(struct e1000_adapter *adapter, struct sk_buff *skb)
4450 {
4451         struct e1000_ring *tx_ring = adapter->tx_ring;
4452         struct e1000_context_desc *context_desc;
4453         struct e1000_buffer *buffer_info;
4454         unsigned int i;
4455         u8 css;
4456         u32 cmd_len = E1000_TXD_CMD_DEXT;
4457         __be16 protocol;
4458
4459         if (skb->ip_summed != CHECKSUM_PARTIAL)
4460                 return 0;
4461
4462         if (skb->protocol == cpu_to_be16(ETH_P_8021Q))
4463                 protocol = vlan_eth_hdr(skb)->h_vlan_encapsulated_proto;
4464         else
4465                 protocol = skb->protocol;
4466
4467         switch (protocol) {
4468         case cpu_to_be16(ETH_P_IP):
4469                 if (ip_hdr(skb)->protocol == IPPROTO_TCP)
4470                         cmd_len |= E1000_TXD_CMD_TCP;
4471                 break;
4472         case cpu_to_be16(ETH_P_IPV6):
4473                 /* XXX not handling all IPV6 headers */
4474                 if (ipv6_hdr(skb)->nexthdr == IPPROTO_TCP)
4475                         cmd_len |= E1000_TXD_CMD_TCP;
4476                 break;
4477         default:
4478                 if (unlikely(net_ratelimit()))
4479                         e_warn("checksum_partial proto=%x!\n",
4480                                be16_to_cpu(protocol));
4481                 break;
4482         }
4483
4484         css = skb_checksum_start_offset(skb);
4485
4486         i = tx_ring->next_to_use;
4487         buffer_info = &tx_ring->buffer_info[i];
4488         context_desc = E1000_CONTEXT_DESC(*tx_ring, i);
4489
4490         context_desc->lower_setup.ip_config = 0;
4491         context_desc->upper_setup.tcp_fields.tucss = css;
4492         context_desc->upper_setup.tcp_fields.tucso =
4493                                 css + skb->csum_offset;
4494         context_desc->upper_setup.tcp_fields.tucse = 0;
4495         context_desc->tcp_seg_setup.data = 0;
4496         context_desc->cmd_and_length = cpu_to_le32(cmd_len);
4497
4498         buffer_info->time_stamp = jiffies;
4499         buffer_info->next_to_watch = i;
4500
4501         i++;
4502         if (i == tx_ring->count)
4503                 i = 0;
4504         tx_ring->next_to_use = i;
4505
4506         return 1;
4507 }
4508
4509 #define E1000_MAX_PER_TXD       8192
4510 #define E1000_MAX_TXD_PWR       12
4511
4512 static int e1000_tx_map(struct e1000_adapter *adapter,
4513                         struct sk_buff *skb, unsigned int first,
4514                         unsigned int max_per_txd, unsigned int nr_frags,
4515                         unsigned int mss)
4516 {
4517         struct e1000_ring *tx_ring = adapter->tx_ring;
4518         struct pci_dev *pdev = adapter->pdev;
4519         struct e1000_buffer *buffer_info;
4520         unsigned int len = skb_headlen(skb);
4521         unsigned int offset = 0, size, count = 0, i;
4522         unsigned int f, bytecount, segs;
4523
4524         i = tx_ring->next_to_use;
4525
4526         while (len) {
4527                 buffer_info = &tx_ring->buffer_info[i];
4528                 size = min(len, max_per_txd);
4529
4530                 buffer_info->length = size;
4531                 buffer_info->time_stamp = jiffies;
4532                 buffer_info->next_to_watch = i;
4533                 buffer_info->dma = dma_map_single(&pdev->dev,
4534                                                   skb->data + offset,
4535                                                   size, DMA_TO_DEVICE);
4536                 buffer_info->mapped_as_page = false;
4537                 if (dma_mapping_error(&pdev->dev, buffer_info->dma))
4538                         goto dma_error;
4539
4540                 len -= size;
4541                 offset += size;
4542                 count++;
4543
4544                 if (len) {
4545                         i++;
4546                         if (i == tx_ring->count)
4547                                 i = 0;
4548                 }
4549         }
4550
4551         for (f = 0; f < nr_frags; f++) {
4552                 struct skb_frag_struct *frag;
4553
4554                 frag = &skb_shinfo(skb)->frags[f];
4555                 len = frag->size;
4556                 offset = frag->page_offset;
4557
4558                 while (len) {
4559                         i++;
4560                         if (i == tx_ring->count)
4561                                 i = 0;
4562
4563                         buffer_info = &tx_ring->buffer_info[i];
4564                         size = min(len, max_per_txd);
4565
4566                         buffer_info->length = size;
4567                         buffer_info->time_stamp = jiffies;
4568                         buffer_info->next_to_watch = i;
4569                         buffer_info->dma = dma_map_page(&pdev->dev, frag->page,
4570                                                         offset, size,
4571                                                         DMA_TO_DEVICE);
4572                         buffer_info->mapped_as_page = true;
4573                         if (dma_mapping_error(&pdev->dev, buffer_info->dma))
4574                                 goto dma_error;
4575
4576                         len -= size;
4577                         offset += size;
4578                         count++;
4579                 }
4580         }
4581
4582         segs = skb_shinfo(skb)->gso_segs ? : 1;
4583         /* multiply data chunks by size of headers */
4584         bytecount = ((segs - 1) * skb_headlen(skb)) + skb->len;
4585
4586         tx_ring->buffer_info[i].skb = skb;
4587         tx_ring->buffer_info[i].segs = segs;
4588         tx_ring->buffer_info[i].bytecount = bytecount;
4589         tx_ring->buffer_info[first].next_to_watch = i;
4590
4591         return count;
4592
4593 dma_error:
4594         dev_err(&pdev->dev, "Tx DMA map failed\n");
4595         buffer_info->dma = 0;
4596         if (count)
4597                 count--;
4598
4599         while (count--) {
4600                 if (i == 0)
4601                         i += tx_ring->count;
4602                 i--;
4603                 buffer_info = &tx_ring->buffer_info[i];
4604                 e1000_put_txbuf(adapter, buffer_info);
4605         }
4606
4607         return 0;
4608 }
4609
4610 static void e1000_tx_queue(struct e1000_adapter *adapter,
4611                            int tx_flags, int count)
4612 {
4613         struct e1000_ring *tx_ring = adapter->tx_ring;
4614         struct e1000_tx_desc *tx_desc = NULL;
4615         struct e1000_buffer *buffer_info;
4616         u32 txd_upper = 0, txd_lower = E1000_TXD_CMD_IFCS;
4617         unsigned int i;
4618
4619         if (tx_flags & E1000_TX_FLAGS_TSO) {
4620                 txd_lower |= E1000_TXD_CMD_DEXT | E1000_TXD_DTYP_D |
4621                              E1000_TXD_CMD_TSE;
4622                 txd_upper |= E1000_TXD_POPTS_TXSM << 8;
4623
4624                 if (tx_flags & E1000_TX_FLAGS_IPV4)
4625                         txd_upper |= E1000_TXD_POPTS_IXSM << 8;
4626         }
4627
4628         if (tx_flags & E1000_TX_FLAGS_CSUM) {
4629                 txd_lower |= E1000_TXD_CMD_DEXT | E1000_TXD_DTYP_D;
4630                 txd_upper |= E1000_TXD_POPTS_TXSM << 8;
4631         }
4632
4633         if (tx_flags & E1000_TX_FLAGS_VLAN) {
4634                 txd_lower |= E1000_TXD_CMD_VLE;
4635                 txd_upper |= (tx_flags & E1000_TX_FLAGS_VLAN_MASK);
4636         }
4637
4638         i = tx_ring->next_to_use;
4639
4640         do {
4641                 buffer_info = &tx_ring->buffer_info[i];
4642                 tx_desc = E1000_TX_DESC(*tx_ring, i);
4643                 tx_desc->buffer_addr = cpu_to_le64(buffer_info->dma);
4644                 tx_desc->lower.data =
4645                         cpu_to_le32(txd_lower | buffer_info->length);
4646                 tx_desc->upper.data = cpu_to_le32(txd_upper);
4647
4648                 i++;
4649                 if (i == tx_ring->count)
4650                         i = 0;
4651         } while (--count > 0);
4652
4653         tx_desc->lower.data |= cpu_to_le32(adapter->txd_cmd);
4654
4655         /*
4656          * Force memory writes to complete before letting h/w
4657          * know there are new descriptors to fetch.  (Only
4658          * applicable for weak-ordered memory model archs,
4659          * such as IA-64).
4660          */
4661         wmb();
4662
4663         tx_ring->next_to_use = i;
4664         writel(i, adapter->hw.hw_addr + tx_ring->tail);
4665         /*
4666          * we need this if more than one processor can write to our tail
4667          * at a time, it synchronizes IO on IA64/Altix systems
4668          */
4669         mmiowb();
4670 }
4671
4672 #define MINIMUM_DHCP_PACKET_SIZE 282
4673 static int e1000_transfer_dhcp_info(struct e1000_adapter *adapter,
4674                                     struct sk_buff *skb)
4675 {
4676         struct e1000_hw *hw =  &adapter->hw;
4677         u16 length, offset;
4678
4679         if (vlan_tx_tag_present(skb)) {
4680                 if (!((vlan_tx_tag_get(skb) == adapter->hw.mng_cookie.vlan_id) &&
4681                     (adapter->hw.mng_cookie.status &
4682                         E1000_MNG_DHCP_COOKIE_STATUS_VLAN)))
4683                         return 0;
4684         }
4685
4686         if (skb->len <= MINIMUM_DHCP_PACKET_SIZE)
4687                 return 0;
4688
4689         if (((struct ethhdr *) skb->data)->h_proto != htons(ETH_P_IP))
4690                 return 0;
4691
4692         {
4693                 const struct iphdr *ip = (struct iphdr *)((u8 *)skb->data+14);
4694                 struct udphdr *udp;
4695
4696                 if (ip->protocol != IPPROTO_UDP)
4697                         return 0;
4698
4699                 udp = (struct udphdr *)((u8 *)ip + (ip->ihl << 2));
4700                 if (ntohs(udp->dest) != 67)
4701                         return 0;
4702
4703                 offset = (u8 *)udp + 8 - skb->data;
4704                 length = skb->len - offset;
4705                 return e1000e_mng_write_dhcp_info(hw, (u8 *)udp + 8, length);
4706         }
4707
4708         return 0;
4709 }
4710
4711 static int __e1000_maybe_stop_tx(struct net_device *netdev, int size)
4712 {
4713         struct e1000_adapter *adapter = netdev_priv(netdev);
4714
4715         netif_stop_queue(netdev);
4716         /*
4717          * Herbert's original patch had:
4718          *  smp_mb__after_netif_stop_queue();
4719          * but since that doesn't exist yet, just open code it.
4720          */
4721         smp_mb();
4722
4723         /*
4724          * We need to check again in a case another CPU has just
4725          * made room available.
4726          */
4727         if (e1000_desc_unused(adapter->tx_ring) < size)
4728                 return -EBUSY;
4729
4730         /* A reprieve! */
4731         netif_start_queue(netdev);
4732         ++adapter->restart_queue;
4733         return 0;
4734 }
4735
4736 static int e1000_maybe_stop_tx(struct net_device *netdev, int size)
4737 {
4738         struct e1000_adapter *adapter = netdev_priv(netdev);
4739
4740         if (e1000_desc_unused(adapter->tx_ring) >= size)
4741                 return 0;
4742         return __e1000_maybe_stop_tx(netdev, size);
4743 }
4744
4745 #define TXD_USE_COUNT(S, X) (((S) >> (X)) + 1 )
4746 static netdev_tx_t e1000_xmit_frame(struct sk_buff *skb,
4747                                     struct net_device *netdev)
4748 {
4749         struct e1000_adapter *adapter = netdev_priv(netdev);
4750         struct e1000_ring *tx_ring = adapter->tx_ring;
4751         unsigned int first;
4752         unsigned int max_per_txd = E1000_MAX_PER_TXD;
4753         unsigned int max_txd_pwr = E1000_MAX_TXD_PWR;
4754         unsigned int tx_flags = 0;
4755         unsigned int len = skb_headlen(skb);
4756         unsigned int nr_frags;
4757         unsigned int mss;
4758         int count = 0;
4759         int tso;
4760         unsigned int f;
4761
4762         if (test_bit(__E1000_DOWN, &adapter->state)) {
4763                 dev_kfree_skb_any(skb);
4764                 return NETDEV_TX_OK;
4765         }
4766
4767         if (skb->len <= 0) {
4768                 dev_kfree_skb_any(skb);
4769                 return NETDEV_TX_OK;
4770         }
4771
4772         mss = skb_shinfo(skb)->gso_size;
4773         /*
4774          * The controller does a simple calculation to
4775          * make sure there is enough room in the FIFO before
4776          * initiating the DMA for each buffer.  The calc is:
4777          * 4 = ceil(buffer len/mss).  To make sure we don't
4778          * overrun the FIFO, adjust the max buffer len if mss
4779          * drops.
4780          */
4781         if (mss) {
4782                 u8 hdr_len;
4783                 max_per_txd = min(mss << 2, max_per_txd);
4784                 max_txd_pwr = fls(max_per_txd) - 1;
4785
4786                 /*
4787                  * TSO Workaround for 82571/2/3 Controllers -- if skb->data
4788                  * points to just header, pull a few bytes of payload from
4789                  * frags into skb->data
4790                  */
4791                 hdr_len = skb_transport_offset(skb) + tcp_hdrlen(skb);
4792                 /*
4793                  * we do this workaround for ES2LAN, but it is un-necessary,
4794                  * avoiding it could save a lot of cycles
4795                  */
4796                 if (skb->data_len && (hdr_len == len)) {
4797                         unsigned int pull_size;
4798
4799                         pull_size = min((unsigned int)4, skb->data_len);
4800                         if (!__pskb_pull_tail(skb, pull_size)) {
4801                                 e_err("__pskb_pull_tail failed.\n");
4802                                 dev_kfree_skb_any(skb);
4803                                 return NETDEV_TX_OK;
4804                         }
4805                         len = skb_headlen(skb);
4806                 }
4807         }
4808
4809         /* reserve a descriptor for the offload context */
4810         if ((mss) || (skb->ip_summed == CHECKSUM_PARTIAL))
4811                 count++;
4812         count++;
4813
4814         count += TXD_USE_COUNT(len, max_txd_pwr);
4815
4816         nr_frags = skb_shinfo(skb)->nr_frags;
4817         for (f = 0; f < nr_frags; f++)
4818                 count += TXD_USE_COUNT(skb_shinfo(skb)->frags[f].size,
4819                                        max_txd_pwr);
4820
4821         if (adapter->hw.mac.tx_pkt_filtering)
4822                 e1000_transfer_dhcp_info(adapter, skb);
4823
4824         /*
4825          * need: count + 2 desc gap to keep tail from touching
4826          * head, otherwise try next time
4827          */
4828         if (e1000_maybe_stop_tx(netdev, count + 2))
4829                 return NETDEV_TX_BUSY;
4830
4831         if (vlan_tx_tag_present(skb)) {
4832                 tx_flags |= E1000_TX_FLAGS_VLAN;
4833                 tx_flags |= (vlan_tx_tag_get(skb) << E1000_TX_FLAGS_VLAN_SHIFT);
4834         }
4835
4836         first = tx_ring->next_to_use;
4837
4838         tso = e1000_tso(adapter, skb);
4839         if (tso < 0) {
4840                 dev_kfree_skb_any(skb);
4841                 return NETDEV_TX_OK;
4842         }
4843
4844         if (tso)
4845                 tx_flags |= E1000_TX_FLAGS_TSO;
4846         else if (e1000_tx_csum(adapter, skb))
4847                 tx_flags |= E1000_TX_FLAGS_CSUM;
4848
4849         /*
4850          * Old method was to assume IPv4 packet by default if TSO was enabled.
4851          * 82571 hardware supports TSO capabilities for IPv6 as well...
4852          * no longer assume, we must.
4853          */
4854         if (skb->protocol == htons(ETH_P_IP))
4855                 tx_flags |= E1000_TX_FLAGS_IPV4;
4856
4857         /* if count is 0 then mapping error has occured */
4858         count = e1000_tx_map(adapter, skb, first, max_per_txd, nr_frags, mss);
4859         if (count) {
4860                 e1000_tx_queue(adapter, tx_flags, count);
4861                 /* Make sure there is space in the ring for the next send. */
4862                 e1000_maybe_stop_tx(netdev, MAX_SKB_FRAGS + 2);
4863
4864         } else {
4865                 dev_kfree_skb_any(skb);
4866                 tx_ring->buffer_info[first].time_stamp = 0;
4867                 tx_ring->next_to_use = first;
4868         }
4869
4870         return NETDEV_TX_OK;
4871 }
4872
4873 /**
4874  * e1000_tx_timeout - Respond to a Tx Hang
4875  * @netdev: network interface device structure
4876  **/
4877 static void e1000_tx_timeout(struct net_device *netdev)
4878 {
4879         struct e1000_adapter *adapter = netdev_priv(netdev);
4880
4881         /* Do the reset outside of interrupt context */
4882         adapter->tx_timeout_count++;
4883         schedule_work(&adapter->reset_task);
4884 }
4885
4886 static void e1000_reset_task(struct work_struct *work)
4887 {
4888         struct e1000_adapter *adapter;
4889         adapter = container_of(work, struct e1000_adapter, reset_task);
4890
4891         if (!((adapter->flags & FLAG_RX_NEEDS_RESTART) &&
4892               (adapter->flags & FLAG_RX_RESTART_NOW))) {
4893                 e1000e_dump(adapter);
4894                 e_err("Reset adapter\n");
4895         }
4896         e1000e_reinit_locked(adapter);
4897 }
4898
4899 /**
4900  * e1000_get_stats - Get System Network Statistics
4901  * @netdev: network interface device structure
4902  *
4903  * Returns the address of the device statistics structure.
4904  * The statistics are actually updated from the timer callback.
4905  **/
4906 static struct net_device_stats *e1000_get_stats(struct net_device *netdev)
4907 {
4908         /* only return the current stats */
4909         return &netdev->stats;
4910 }
4911
4912 /**
4913  * e1000_change_mtu - Change the Maximum Transfer Unit
4914  * @netdev: network interface device structure
4915  * @new_mtu: new value for maximum frame size
4916  *
4917  * Returns 0 on success, negative on failure
4918  **/
4919 static int e1000_change_mtu(struct net_device *netdev, int new_mtu)
4920 {
4921         struct e1000_adapter *adapter = netdev_priv(netdev);
4922         int max_frame = new_mtu + ETH_HLEN + ETH_FCS_LEN;
4923
4924         /* Jumbo frame support */
4925         if ((max_frame > ETH_FRAME_LEN + ETH_FCS_LEN) &&
4926             !(adapter->flags & FLAG_HAS_JUMBO_FRAMES)) {
4927                 e_err("Jumbo Frames not supported.\n");
4928                 return -EINVAL;
4929         }
4930
4931         /* Supported frame sizes */
4932         if ((new_mtu < ETH_ZLEN + ETH_FCS_LEN + VLAN_HLEN) ||
4933             (max_frame > adapter->max_hw_frame_size)) {
4934                 e_err("Unsupported MTU setting\n");
4935                 return -EINVAL;
4936         }
4937
4938         /* Jumbo frame workaround on 82579 requires CRC be stripped */
4939         if ((adapter->hw.mac.type == e1000_pch2lan) &&
4940             !(adapter->flags2 & FLAG2_CRC_STRIPPING) &&
4941             (new_mtu > ETH_DATA_LEN)) {
4942                 e_err("Jumbo Frames not supported on 82579 when CRC "
4943                       "stripping is disabled.\n");
4944                 return -EINVAL;
4945         }
4946
4947         /* 82573 Errata 17 */
4948         if (((adapter->hw.mac.type == e1000_82573) ||
4949              (adapter->hw.mac.type == e1000_82574)) &&
4950             (max_frame > ETH_FRAME_LEN + ETH_FCS_LEN)) {
4951                 adapter->flags2 |= FLAG2_DISABLE_ASPM_L1;
4952                 e1000e_disable_aspm(adapter->pdev, PCIE_LINK_STATE_L1);
4953         }
4954
4955         while (test_and_set_bit(__E1000_RESETTING, &adapter->state))
4956                 msleep(1);
4957         /* e1000e_down -> e1000e_reset dependent on max_frame_size & mtu */
4958         adapter->max_frame_size = max_frame;
4959         e_info("changing MTU from %d to %d\n", netdev->mtu, new_mtu);
4960         netdev->mtu = new_mtu;
4961         if (netif_running(netdev))
4962                 e1000e_down(adapter);
4963
4964         /*
4965          * NOTE: netdev_alloc_skb reserves 16 bytes, and typically NET_IP_ALIGN
4966          * means we reserve 2 more, this pushes us to allocate from the next
4967          * larger slab size.
4968          * i.e. RXBUFFER_2048 --> size-4096 slab
4969          * However with the new *_jumbo_rx* routines, jumbo receives will use
4970          * fragmented skbs
4971          */
4972
4973         if (max_frame <= 2048)
4974                 adapter->rx_buffer_len = 2048;
4975         else
4976                 adapter->rx_buffer_len = 4096;
4977
4978         /* adjust allocation if LPE protects us, and we aren't using SBP */
4979         if ((max_frame == ETH_FRAME_LEN + ETH_FCS_LEN) ||
4980              (max_frame == ETH_FRAME_LEN + VLAN_HLEN + ETH_FCS_LEN))
4981                 adapter->rx_buffer_len = ETH_FRAME_LEN + VLAN_HLEN
4982                                          + ETH_FCS_LEN;
4983
4984         if (netif_running(netdev))
4985                 e1000e_up(adapter);
4986         else
4987                 e1000e_reset(adapter);
4988
4989         clear_bit(__E1000_RESETTING, &adapter->state);
4990
4991         return 0;
4992 }
4993
4994 static int e1000_mii_ioctl(struct net_device *netdev, struct ifreq *ifr,
4995                            int cmd)
4996 {
4997         struct e1000_adapter *adapter = netdev_priv(netdev);
4998         struct mii_ioctl_data *data = if_mii(ifr);
4999
5000         if (adapter->hw.phy.media_type != e1000_media_type_copper)
5001                 return -EOPNOTSUPP;
5002
5003         switch (cmd) {
5004         case SIOCGMIIPHY:
5005                 data->phy_id = adapter->hw.phy.addr;
5006                 break;
5007         case SIOCGMIIREG:
5008                 e1000_phy_read_status(adapter);
5009
5010                 switch (data->reg_num & 0x1F) {
5011                 case MII_BMCR:
5012                         data->val_out = adapter->phy_regs.bmcr;
5013                         break;
5014                 case MII_BMSR:
5015                         data->val_out = adapter->phy_regs.bmsr;
5016                         break;
5017                 case MII_PHYSID1:
5018                         data->val_out = (adapter->hw.phy.id >> 16);
5019                         break;
5020                 case MII_PHYSID2:
5021                         data->val_out = (adapter->hw.phy.id & 0xFFFF);
5022                         break;
5023                 case MII_ADVERTISE:
5024                         data->val_out = adapter->phy_regs.advertise;
5025                         break;
5026                 case MII_LPA:
5027                         data->val_out = adapter->phy_regs.lpa;
5028                         break;
5029                 case MII_EXPANSION:
5030                         data->val_out = adapter->phy_regs.expansion;
5031                         break;
5032                 case MII_CTRL1000:
5033                         data->val_out = adapter->phy_regs.ctrl1000;
5034                         break;
5035                 case MII_STAT1000:
5036                         data->val_out = adapter->phy_regs.stat1000;
5037                         break;
5038                 case MII_ESTATUS:
5039                         data->val_out = adapter->phy_regs.estatus;
5040                         break;
5041                 default:
5042                         return -EIO;
5043                 }
5044                 break;
5045         case SIOCSMIIREG:
5046         default:
5047                 return -EOPNOTSUPP;
5048         }
5049         return 0;
5050 }
5051
5052 static int e1000_ioctl(struct net_device *netdev, struct ifreq *ifr, int cmd)
5053 {
5054         switch (cmd) {
5055         case SIOCGMIIPHY:
5056         case SIOCGMIIREG:
5057         case SIOCSMIIREG:
5058                 return e1000_mii_ioctl(netdev, ifr, cmd);
5059         default:
5060                 return -EOPNOTSUPP;
5061         }
5062 }
5063
5064 static int e1000_init_phy_wakeup(struct e1000_adapter *adapter, u32 wufc)
5065 {
5066         struct e1000_hw *hw = &adapter->hw;
5067         u32 i, mac_reg;
5068         u16 phy_reg;
5069         int retval = 0;
5070
5071         /* copy MAC RARs to PHY RARs */
5072         e1000_copy_rx_addrs_to_phy_ich8lan(hw);
5073
5074         /* copy MAC MTA to PHY MTA */
5075         for (i = 0; i < adapter->hw.mac.mta_reg_count; i++) {
5076                 mac_reg = E1000_READ_REG_ARRAY(hw, E1000_MTA, i);
5077                 e1e_wphy(hw, BM_MTA(i), (u16)(mac_reg & 0xFFFF));
5078                 e1e_wphy(hw, BM_MTA(i) + 1, (u16)((mac_reg >> 16) & 0xFFFF));
5079         }
5080
5081         /* configure PHY Rx Control register */
5082         e1e_rphy(&adapter->hw, BM_RCTL, &phy_reg);
5083         mac_reg = er32(RCTL);
5084         if (mac_reg & E1000_RCTL_UPE)
5085                 phy_reg |= BM_RCTL_UPE;
5086         if (mac_reg & E1000_RCTL_MPE)
5087                 phy_reg |= BM_RCTL_MPE;
5088         phy_reg &= ~(BM_RCTL_MO_MASK);
5089         if (mac_reg & E1000_RCTL_MO_3)
5090                 phy_reg |= (((mac_reg & E1000_RCTL_MO_3) >> E1000_RCTL_MO_SHIFT)
5091                                 << BM_RCTL_MO_SHIFT);
5092         if (mac_reg & E1000_RCTL_BAM)
5093                 phy_reg |= BM_RCTL_BAM;
5094         if (mac_reg & E1000_RCTL_PMCF)
5095                 phy_reg |= BM_RCTL_PMCF;
5096         mac_reg = er32(CTRL);
5097         if (mac_reg & E1000_CTRL_RFCE)
5098                 phy_reg |= BM_RCTL_RFCE;
5099         e1e_wphy(&adapter->hw, BM_RCTL, phy_reg);
5100
5101         /* enable PHY wakeup in MAC register */
5102         ew32(WUFC, wufc);
5103         ew32(WUC, E1000_WUC_PHY_WAKE | E1000_WUC_PME_EN);
5104
5105         /* configure and enable PHY wakeup in PHY registers */
5106         e1e_wphy(&adapter->hw, BM_WUFC, wufc);
5107         e1e_wphy(&adapter->hw, BM_WUC, E1000_WUC_PME_EN);
5108
5109         /* activate PHY wakeup */
5110         retval = hw->phy.ops.acquire(hw);
5111         if (retval) {
5112                 e_err("Could not acquire PHY\n");
5113                 return retval;
5114         }
5115         e1000e_write_phy_reg_mdic(hw, IGP01E1000_PHY_PAGE_SELECT,
5116                                  (BM_WUC_ENABLE_PAGE << IGP_PAGE_SHIFT));
5117         retval = e1000e_read_phy_reg_mdic(hw, BM_WUC_ENABLE_REG, &phy_reg);
5118         if (retval) {
5119                 e_err("Could not read PHY page 769\n");
5120                 goto out;
5121         }
5122         phy_reg |= BM_WUC_ENABLE_BIT | BM_WUC_HOST_WU_BIT;
5123         retval = e1000e_write_phy_reg_mdic(hw, BM_WUC_ENABLE_REG, phy_reg);
5124         if (retval)
5125                 e_err("Could not set PHY Host Wakeup bit\n");
5126 out:
5127         hw->phy.ops.release(hw);
5128
5129         return retval;
5130 }
5131
5132 static int __e1000_shutdown(struct pci_dev *pdev, bool *enable_wake,
5133                             bool runtime)
5134 {
5135         struct net_device *netdev = pci_get_drvdata(pdev);
5136         struct e1000_adapter *adapter = netdev_priv(netdev);
5137         struct e1000_hw *hw = &adapter->hw;
5138         u32 ctrl, ctrl_ext, rctl, status;
5139         /* Runtime suspend should only enable wakeup for link changes */
5140         u32 wufc = runtime ? E1000_WUFC_LNKC : adapter->wol;
5141         int retval = 0;
5142
5143         netif_device_detach(netdev);
5144
5145         if (netif_running(netdev)) {
5146                 WARN_ON(test_bit(__E1000_RESETTING, &adapter->state));
5147                 e1000e_down(adapter);
5148                 e1000_free_irq(adapter);
5149         }
5150         e1000e_reset_interrupt_capability(adapter);
5151
5152         retval = pci_save_state(pdev);
5153         if (retval)
5154                 return retval;
5155
5156         status = er32(STATUS);
5157         if (status & E1000_STATUS_LU)
5158                 wufc &= ~E1000_WUFC_LNKC;
5159
5160         if (wufc) {
5161                 e1000_setup_rctl(adapter);
5162                 e1000_set_multi(netdev);
5163
5164                 /* turn on all-multi mode if wake on multicast is enabled */
5165                 if (wufc & E1000_WUFC_MC) {
5166                         rctl = er32(RCTL);
5167                         rctl |= E1000_RCTL_MPE;
5168                         ew32(RCTL, rctl);
5169                 }
5170
5171                 ctrl = er32(CTRL);
5172                 /* advertise wake from D3Cold */
5173                 #define E1000_CTRL_ADVD3WUC 0x00100000
5174                 /* phy power management enable */
5175                 #define E1000_CTRL_EN_PHY_PWR_MGMT 0x00200000
5176                 ctrl |= E1000_CTRL_ADVD3WUC;
5177                 if (!(adapter->flags2 & FLAG2_HAS_PHY_WAKEUP))
5178                         ctrl |= E1000_CTRL_EN_PHY_PWR_MGMT;
5179                 ew32(CTRL, ctrl);
5180
5181                 if (adapter->hw.phy.media_type == e1000_media_type_fiber ||
5182                     adapter->hw.phy.media_type ==
5183                     e1000_media_type_internal_serdes) {
5184                         /* keep the laser running in D3 */
5185                         ctrl_ext = er32(CTRL_EXT);
5186                         ctrl_ext |= E1000_CTRL_EXT_SDP3_DATA;
5187                         ew32(CTRL_EXT, ctrl_ext);
5188                 }
5189
5190                 if (adapter->flags & FLAG_IS_ICH)
5191                         e1000e_disable_gig_wol_ich8lan(&adapter->hw);
5192
5193                 /* Allow time for pending master requests to run */
5194                 e1000e_disable_pcie_master(&adapter->hw);
5195
5196                 if (adapter->flags2 & FLAG2_HAS_PHY_WAKEUP) {
5197                         /* enable wakeup by the PHY */
5198                         retval = e1000_init_phy_wakeup(adapter, wufc);
5199                         if (retval)
5200                                 return retval;
5201                 } else {
5202                         /* enable wakeup by the MAC */
5203                         ew32(WUFC, wufc);
5204                         ew32(WUC, E1000_WUC_PME_EN);
5205                 }
5206         } else {
5207                 ew32(WUC, 0);
5208                 ew32(WUFC, 0);
5209         }
5210
5211         *enable_wake = !!wufc;
5212
5213         /* make sure adapter isn't asleep if manageability is enabled */
5214         if ((adapter->flags & FLAG_MNG_PT_ENABLED) ||
5215             (hw->mac.ops.check_mng_mode(hw)))
5216                 *enable_wake = true;
5217
5218         if (adapter->hw.phy.type == e1000_phy_igp_3)
5219                 e1000e_igp3_phy_powerdown_workaround_ich8lan(&adapter->hw);
5220
5221         /*
5222          * Release control of h/w to f/w.  If f/w is AMT enabled, this
5223          * would have already happened in close and is redundant.
5224          */
5225         e1000e_release_hw_control(adapter);
5226
5227         pci_disable_device(pdev);
5228
5229         return 0;
5230 }
5231
5232 static void e1000_power_off(struct pci_dev *pdev, bool sleep, bool wake)
5233 {
5234         if (sleep && wake) {
5235                 pci_prepare_to_sleep(pdev);
5236                 return;
5237         }
5238
5239         pci_wake_from_d3(pdev, wake);
5240         pci_set_power_state(pdev, PCI_D3hot);
5241 }
5242
5243 static void e1000_complete_shutdown(struct pci_dev *pdev, bool sleep,
5244                                     bool wake)
5245 {
5246         struct net_device *netdev = pci_get_drvdata(pdev);
5247         struct e1000_adapter *adapter = netdev_priv(netdev);
5248
5249         /*
5250          * The pci-e switch on some quad port adapters will report a
5251          * correctable error when the MAC transitions from D0 to D3.  To
5252          * prevent this we need to mask off the correctable errors on the
5253          * downstream port of the pci-e switch.
5254          */
5255         if (adapter->flags & FLAG_IS_QUAD_PORT) {
5256                 struct pci_dev *us_dev = pdev->bus->self;
5257                 int pos = pci_find_capability(us_dev, PCI_CAP_ID_EXP);
5258                 u16 devctl;
5259
5260                 pci_read_config_word(us_dev, pos + PCI_EXP_DEVCTL, &devctl);
5261                 pci_write_config_word(us_dev, pos + PCI_EXP_DEVCTL,
5262                                       (devctl & ~PCI_EXP_DEVCTL_CERE));
5263
5264                 e1000_power_off(pdev, sleep, wake);
5265
5266                 pci_write_config_word(us_dev, pos + PCI_EXP_DEVCTL, devctl);
5267         } else {
5268                 e1000_power_off(pdev, sleep, wake);
5269         }
5270 }
5271
5272 #ifdef CONFIG_PCIEASPM
5273 static void __e1000e_disable_aspm(struct pci_dev *pdev, u16 state)
5274 {
5275         pci_disable_link_state(pdev, state);
5276 }
5277 #else
5278 static void __e1000e_disable_aspm(struct pci_dev *pdev, u16 state)
5279 {
5280         int pos;
5281         u16 reg16;
5282
5283         /*
5284          * Both device and parent should have the same ASPM setting.
5285          * Disable ASPM in downstream component first and then upstream.
5286          */
5287         pos = pci_pcie_cap(pdev);
5288         pci_read_config_word(pdev, pos + PCI_EXP_LNKCTL, &reg16);
5289         reg16 &= ~state;
5290         pci_write_config_word(pdev, pos + PCI_EXP_LNKCTL, reg16);
5291
5292         if (!pdev->bus->self)
5293                 return;
5294
5295         pos = pci_pcie_cap(pdev->bus->self);
5296         pci_read_config_word(pdev->bus->self, pos + PCI_EXP_LNKCTL, &reg16);
5297         reg16 &= ~state;
5298         pci_write_config_word(pdev->bus->self, pos + PCI_EXP_LNKCTL, reg16);
5299 }
5300 #endif
5301 void e1000e_disable_aspm(struct pci_dev *pdev, u16 state)
5302 {
5303         dev_info(&pdev->dev, "Disabling ASPM %s %s\n",
5304                  (state & PCIE_LINK_STATE_L0S) ? "L0s" : "",
5305                  (state & PCIE_LINK_STATE_L1) ? "L1" : "");
5306
5307         __e1000e_disable_aspm(pdev, state);
5308 }
5309
5310 #ifdef CONFIG_PM_OPS
5311 static bool e1000e_pm_ready(struct e1000_adapter *adapter)
5312 {
5313         return !!adapter->tx_ring->buffer_info;
5314 }
5315
5316 static int __e1000_resume(struct pci_dev *pdev)
5317 {
5318         struct net_device *netdev = pci_get_drvdata(pdev);
5319         struct e1000_adapter *adapter = netdev_priv(netdev);
5320         struct e1000_hw *hw = &adapter->hw;
5321         u32 err;
5322
5323         pci_set_power_state(pdev, PCI_D0);
5324         pci_restore_state(pdev);
5325         pci_save_state(pdev);
5326         if (adapter->flags2 & FLAG2_DISABLE_ASPM_L1)
5327                 e1000e_disable_aspm(pdev, PCIE_LINK_STATE_L1);
5328
5329         e1000e_set_interrupt_capability(adapter);
5330         if (netif_running(netdev)) {
5331                 err = e1000_request_irq(adapter);
5332                 if (err)
5333                         return err;
5334         }
5335
5336         e1000e_power_up_phy(adapter);
5337
5338         /* report the system wakeup cause from S3/S4 */
5339         if (adapter->flags2 & FLAG2_HAS_PHY_WAKEUP) {
5340                 u16 phy_data;
5341
5342                 e1e_rphy(&adapter->hw, BM_WUS, &phy_data);
5343                 if (phy_data) {
5344                         e_info("PHY Wakeup cause - %s\n",
5345                                 phy_data & E1000_WUS_EX ? "Unicast Packet" :
5346                                 phy_data & E1000_WUS_MC ? "Multicast Packet" :
5347                                 phy_data & E1000_WUS_BC ? "Broadcast Packet" :
5348                                 phy_data & E1000_WUS_MAG ? "Magic Packet" :
5349                                 phy_data & E1000_WUS_LNKC ? "Link Status "
5350                                 " Change" : "other");
5351                 }
5352                 e1e_wphy(&adapter->hw, BM_WUS, ~0);
5353         } else {
5354                 u32 wus = er32(WUS);
5355                 if (wus) {
5356                         e_info("MAC Wakeup cause - %s\n",
5357                                 wus & E1000_WUS_EX ? "Unicast Packet" :
5358                                 wus & E1000_WUS_MC ? "Multicast Packet" :
5359                                 wus & E1000_WUS_BC ? "Broadcast Packet" :
5360                                 wus & E1000_WUS_MAG ? "Magic Packet" :
5361                                 wus & E1000_WUS_LNKC ? "Link Status Change" :
5362                                 "other");
5363                 }
5364                 ew32(WUS, ~0);
5365         }
5366
5367         e1000e_reset(adapter);
5368
5369         e1000_init_manageability_pt(adapter);
5370
5371         if (netif_running(netdev))
5372                 e1000e_up(adapter);
5373
5374         netif_device_attach(netdev);
5375
5376         /*
5377          * If the controller has AMT, do not set DRV_LOAD until the interface
5378          * is up.  For all other cases, let the f/w know that the h/w is now
5379          * under the control of the driver.
5380          */
5381         if (!(adapter->flags & FLAG_HAS_AMT))
5382                 e1000e_get_hw_control(adapter);
5383
5384         return 0;
5385 }
5386
5387 #ifdef CONFIG_PM_SLEEP
5388 static int e1000_suspend(struct device *dev)
5389 {
5390         struct pci_dev *pdev = to_pci_dev(dev);
5391         int retval;
5392         bool wake;
5393
5394         retval = __e1000_shutdown(pdev, &wake, false);
5395         if (!retval)
5396                 e1000_complete_shutdown(pdev, true, wake);
5397
5398         return retval;
5399 }
5400
5401 static int e1000_resume(struct device *dev)
5402 {
5403         struct pci_dev *pdev = to_pci_dev(dev);
5404         struct net_device *netdev = pci_get_drvdata(pdev);
5405         struct e1000_adapter *adapter = netdev_priv(netdev);
5406
5407         if (e1000e_pm_ready(adapter))
5408                 adapter->idle_check = true;
5409
5410         return __e1000_resume(pdev);
5411 }
5412 #endif /* CONFIG_PM_SLEEP */
5413
5414 #ifdef CONFIG_PM_RUNTIME
5415 static int e1000_runtime_suspend(struct device *dev)
5416 {
5417         struct pci_dev *pdev = to_pci_dev(dev);
5418         struct net_device *netdev = pci_get_drvdata(pdev);
5419         struct e1000_adapter *adapter = netdev_priv(netdev);
5420
5421         if (e1000e_pm_ready(adapter)) {
5422                 bool wake;
5423
5424                 __e1000_shutdown(pdev, &wake, true);
5425         }
5426
5427         return 0;
5428 }
5429
5430 static int e1000_idle(struct device *dev)
5431 {
5432         struct pci_dev *pdev = to_pci_dev(dev);
5433         struct net_device *netdev = pci_get_drvdata(pdev);
5434         struct e1000_adapter *adapter = netdev_priv(netdev);
5435
5436         if (!e1000e_pm_ready(adapter))
5437                 return 0;
5438
5439         if (adapter->idle_check) {
5440                 adapter->idle_check = false;
5441                 if (!e1000e_has_link(adapter))
5442                         pm_schedule_suspend(dev, MSEC_PER_SEC);
5443         }
5444
5445         return -EBUSY;
5446 }
5447
5448 static int e1000_runtime_resume(struct device *dev)
5449 {
5450         struct pci_dev *pdev = to_pci_dev(dev);
5451         struct net_device *netdev = pci_get_drvdata(pdev);
5452         struct e1000_adapter *adapter = netdev_priv(netdev);
5453
5454         if (!e1000e_pm_ready(adapter))
5455                 return 0;
5456
5457         adapter->idle_check = !dev->power.runtime_auto;
5458         return __e1000_resume(pdev);
5459 }
5460 #endif /* CONFIG_PM_RUNTIME */
5461 #endif /* CONFIG_PM_OPS */
5462
5463 static void e1000_shutdown(struct pci_dev *pdev)
5464 {
5465         bool wake = false;
5466
5467         __e1000_shutdown(pdev, &wake, false);
5468
5469         if (system_state == SYSTEM_POWER_OFF)
5470                 e1000_complete_shutdown(pdev, false, wake);
5471 }
5472
5473 #ifdef CONFIG_NET_POLL_CONTROLLER
5474
5475 static irqreturn_t e1000_intr_msix(int irq, void *data)
5476 {
5477         struct net_device *netdev = data;
5478         struct e1000_adapter *adapter = netdev_priv(netdev);
5479         int vector, msix_irq;
5480
5481         if (adapter->msix_entries) {
5482                 vector = 0;
5483                 msix_irq = adapter->msix_entries[vector].vector;
5484                 disable_irq(msix_irq);
5485                 e1000_intr_msix_rx(msix_irq, netdev);
5486                 enable_irq(msix_irq);
5487
5488                 vector++;
5489                 msix_irq = adapter->msix_entries[vector].vector;
5490                 disable_irq(msix_irq);
5491                 e1000_intr_msix_tx(msix_irq, netdev);
5492                 enable_irq(msix_irq);
5493
5494                 vector++;
5495                 msix_irq = adapter->msix_entries[vector].vector;
5496                 disable_irq(msix_irq);
5497                 e1000_msix_other(msix_irq, netdev);
5498                 enable_irq(msix_irq);
5499         }
5500
5501         return IRQ_HANDLED;
5502 }
5503
5504 /*
5505  * Polling 'interrupt' - used by things like netconsole to send skbs
5506  * without having to re-enable interrupts. It's not called while
5507  * the interrupt routine is executing.
5508  */
5509 static void e1000_netpoll(struct net_device *netdev)
5510 {
5511         struct e1000_adapter *adapter = netdev_priv(netdev);
5512
5513         switch (adapter->int_mode) {
5514         case E1000E_INT_MODE_MSIX:
5515                 e1000_intr_msix(adapter->pdev->irq, netdev);
5516                 break;
5517         case E1000E_INT_MODE_MSI:
5518                 disable_irq(adapter->pdev->irq);
5519                 e1000_intr_msi(adapter->pdev->irq, netdev);
5520                 enable_irq(adapter->pdev->irq);
5521                 break;
5522         default: /* E1000E_INT_MODE_LEGACY */
5523                 disable_irq(adapter->pdev->irq);
5524                 e1000_intr(adapter->pdev->irq, netdev);
5525                 enable_irq(adapter->pdev->irq);
5526                 break;
5527         }
5528 }
5529 #endif
5530
5531 /**
5532  * e1000_io_error_detected - called when PCI error is detected
5533  * @pdev: Pointer to PCI device
5534  * @state: The current pci connection state
5535  *
5536  * This function is called after a PCI bus error affecting
5537  * this device has been detected.
5538  */
5539 static pci_ers_result_t e1000_io_error_detected(struct pci_dev *pdev,
5540                                                 pci_channel_state_t state)
5541 {
5542         struct net_device *netdev = pci_get_drvdata(pdev);
5543         struct e1000_adapter *adapter = netdev_priv(netdev);
5544
5545         netif_device_detach(netdev);
5546
5547         if (state == pci_channel_io_perm_failure)
5548                 return PCI_ERS_RESULT_DISCONNECT;
5549
5550         if (netif_running(netdev))
5551                 e1000e_down(adapter);
5552         pci_disable_device(pdev);
5553
5554         /* Request a slot slot reset. */
5555         return PCI_ERS_RESULT_NEED_RESET;
5556 }
5557
5558 /**
5559  * e1000_io_slot_reset - called after the pci bus has been reset.
5560  * @pdev: Pointer to PCI device
5561  *
5562  * Restart the card from scratch, as if from a cold-boot. Implementation
5563  * resembles the first-half of the e1000_resume routine.
5564  */
5565 static pci_ers_result_t e1000_io_slot_reset(struct pci_dev *pdev)
5566 {
5567         struct net_device *netdev = pci_get_drvdata(pdev);
5568         struct e1000_adapter *adapter = netdev_priv(netdev);
5569         struct e1000_hw *hw = &adapter->hw;
5570         int err;
5571         pci_ers_result_t result;
5572
5573         if (adapter->flags2 & FLAG2_DISABLE_ASPM_L1)
5574                 e1000e_disable_aspm(pdev, PCIE_LINK_STATE_L1);
5575         err = pci_enable_device_mem(pdev);
5576         if (err) {
5577                 dev_err(&pdev->dev,
5578                         "Cannot re-enable PCI device after reset.\n");
5579                 result = PCI_ERS_RESULT_DISCONNECT;
5580         } else {
5581                 pci_set_master(pdev);
5582                 pdev->state_saved = true;
5583                 pci_restore_state(pdev);
5584
5585                 pci_enable_wake(pdev, PCI_D3hot, 0);
5586                 pci_enable_wake(pdev, PCI_D3cold, 0);
5587
5588                 e1000e_reset(adapter);
5589                 ew32(WUS, ~0);
5590                 result = PCI_ERS_RESULT_RECOVERED;
5591         }
5592
5593         pci_cleanup_aer_uncorrect_error_status(pdev);
5594
5595         return result;
5596 }
5597
5598 /**
5599  * e1000_io_resume - called when traffic can start flowing again.
5600  * @pdev: Pointer to PCI device
5601  *
5602  * This callback is called when the error recovery driver tells us that
5603  * its OK to resume normal operation. Implementation resembles the
5604  * second-half of the e1000_resume routine.
5605  */
5606 static void e1000_io_resume(struct pci_dev *pdev)
5607 {
5608         struct net_device *netdev = pci_get_drvdata(pdev);
5609         struct e1000_adapter *adapter = netdev_priv(netdev);
5610
5611         e1000_init_manageability_pt(adapter);
5612
5613         if (netif_running(netdev)) {
5614                 if (e1000e_up(adapter)) {
5615                         dev_err(&pdev->dev,
5616                                 "can't bring device back up after reset\n");
5617                         return;
5618                 }
5619         }
5620
5621         netif_device_attach(netdev);
5622
5623         /*
5624          * If the controller has AMT, do not set DRV_LOAD until the interface
5625          * is up.  For all other cases, let the f/w know that the h/w is now
5626          * under the control of the driver.
5627          */
5628         if (!(adapter->flags & FLAG_HAS_AMT))
5629                 e1000e_get_hw_control(adapter);
5630
5631 }
5632
5633 static void e1000_print_device_info(struct e1000_adapter *adapter)
5634 {
5635         struct e1000_hw *hw = &adapter->hw;
5636         struct net_device *netdev = adapter->netdev;
5637         u32 ret_val;
5638         u8 pba_str[E1000_PBANUM_LENGTH];
5639
5640         /* print bus type/speed/width info */
5641         e_info("(PCI Express:2.5GB/s:%s) %pM\n",
5642                /* bus width */
5643                ((hw->bus.width == e1000_bus_width_pcie_x4) ? "Width x4" :
5644                 "Width x1"),
5645                /* MAC address */
5646                netdev->dev_addr);
5647         e_info("Intel(R) PRO/%s Network Connection\n",
5648                (hw->phy.type == e1000_phy_ife) ? "10/100" : "1000");
5649         ret_val = e1000_read_pba_string_generic(hw, pba_str,
5650                                                 E1000_PBANUM_LENGTH);
5651         if (ret_val)
5652                 strncpy((char *)pba_str, "Unknown", sizeof(pba_str) - 1);
5653         e_info("MAC: %d, PHY: %d, PBA No: %s\n",
5654                hw->mac.type, hw->phy.type, pba_str);
5655 }
5656
5657 static void e1000_eeprom_checks(struct e1000_adapter *adapter)
5658 {
5659         struct e1000_hw *hw = &adapter->hw;
5660         int ret_val;
5661         u16 buf = 0;
5662
5663         if (hw->mac.type != e1000_82573)
5664                 return;
5665
5666         ret_val = e1000_read_nvm(hw, NVM_INIT_CONTROL2_REG, 1, &buf);
5667         if (!ret_val && (!(le16_to_cpu(buf) & (1 << 0)))) {
5668                 /* Deep Smart Power Down (DSPD) */
5669                 dev_warn(&adapter->pdev->dev,
5670                          "Warning: detected DSPD enabled in EEPROM\n");
5671         }
5672 }
5673
5674 static const struct net_device_ops e1000e_netdev_ops = {
5675         .ndo_open               = e1000_open,
5676         .ndo_stop               = e1000_close,
5677         .ndo_start_xmit         = e1000_xmit_frame,
5678         .ndo_get_stats          = e1000_get_stats,
5679         .ndo_set_multicast_list = e1000_set_multi,
5680         .ndo_set_mac_address    = e1000_set_mac,
5681         .ndo_change_mtu         = e1000_change_mtu,
5682         .ndo_do_ioctl           = e1000_ioctl,
5683         .ndo_tx_timeout         = e1000_tx_timeout,
5684         .ndo_validate_addr      = eth_validate_addr,
5685
5686         .ndo_vlan_rx_register   = e1000_vlan_rx_register,
5687         .ndo_vlan_rx_add_vid    = e1000_vlan_rx_add_vid,
5688         .ndo_vlan_rx_kill_vid   = e1000_vlan_rx_kill_vid,
5689 #ifdef CONFIG_NET_POLL_CONTROLLER
5690         .ndo_poll_controller    = e1000_netpoll,
5691 #endif
5692 };
5693
5694 /**
5695  * e1000_probe - Device Initialization Routine
5696  * @pdev: PCI device information struct
5697  * @ent: entry in e1000_pci_tbl
5698  *
5699  * Returns 0 on success, negative on failure
5700  *
5701  * e1000_probe initializes an adapter identified by a pci_dev structure.
5702  * The OS initialization, configuring of the adapter private structure,
5703  * and a hardware reset occur.
5704  **/
5705 static int __devinit e1000_probe(struct pci_dev *pdev,
5706                                  const struct pci_device_id *ent)
5707 {
5708         struct net_device *netdev;
5709         struct e1000_adapter *adapter;
5710         struct e1000_hw *hw;
5711         const struct e1000_info *ei = e1000_info_tbl[ent->driver_data];
5712         resource_size_t mmio_start, mmio_len;
5713         resource_size_t flash_start, flash_len;
5714
5715         static int cards_found;
5716         int i, err, pci_using_dac;
5717         u16 eeprom_data = 0;
5718         u16 eeprom_apme_mask = E1000_EEPROM_APME;
5719
5720         if (ei->flags2 & FLAG2_DISABLE_ASPM_L1)
5721                 e1000e_disable_aspm(pdev, PCIE_LINK_STATE_L1);
5722
5723         err = pci_enable_device_mem(pdev);
5724         if (err)
5725                 return err;
5726
5727         pci_using_dac = 0;
5728         err = dma_set_mask(&pdev->dev, DMA_BIT_MASK(64));
5729         if (!err) {
5730                 err = dma_set_coherent_mask(&pdev->dev, DMA_BIT_MASK(64));
5731                 if (!err)
5732                         pci_using_dac = 1;
5733         } else {
5734                 err = dma_set_mask(&pdev->dev, DMA_BIT_MASK(32));
5735                 if (err) {
5736                         err = dma_set_coherent_mask(&pdev->dev,
5737                                                     DMA_BIT_MASK(32));
5738                         if (err) {
5739                                 dev_err(&pdev->dev, "No usable DMA "
5740                                         "configuration, aborting\n");
5741                                 goto err_dma;
5742                         }
5743                 }
5744         }
5745
5746         err = pci_request_selected_regions_exclusive(pdev,
5747                                           pci_select_bars(pdev, IORESOURCE_MEM),
5748                                           e1000e_driver_name);
5749         if (err)
5750                 goto err_pci_reg;
5751
5752         /* AER (Advanced Error Reporting) hooks */
5753         pci_enable_pcie_error_reporting(pdev);
5754
5755         pci_set_master(pdev);
5756         /* PCI config space info */
5757         err = pci_save_state(pdev);
5758         if (err)
5759                 goto err_alloc_etherdev;
5760
5761         err = -ENOMEM;
5762         netdev = alloc_etherdev(sizeof(struct e1000_adapter));
5763         if (!netdev)
5764                 goto err_alloc_etherdev;
5765
5766         SET_NETDEV_DEV(netdev, &pdev->dev);
5767
5768         netdev->irq = pdev->irq;
5769
5770         pci_set_drvdata(pdev, netdev);
5771         adapter = netdev_priv(netdev);
5772         hw = &adapter->hw;
5773         adapter->netdev = netdev;
5774         adapter->pdev = pdev;
5775         adapter->ei = ei;
5776         adapter->pba = ei->pba;
5777         adapter->flags = ei->flags;
5778         adapter->flags2 = ei->flags2;
5779         adapter->hw.adapter = adapter;
5780         adapter->hw.mac.type = ei->mac;
5781         adapter->max_hw_frame_size = ei->max_hw_frame_size;
5782         adapter->msg_enable = (1 << NETIF_MSG_DRV | NETIF_MSG_PROBE) - 1;
5783
5784         mmio_start = pci_resource_start(pdev, 0);
5785         mmio_len = pci_resource_len(pdev, 0);
5786
5787         err = -EIO;
5788         adapter->hw.hw_addr = ioremap(mmio_start, mmio_len);
5789         if (!adapter->hw.hw_addr)
5790                 goto err_ioremap;
5791
5792         if ((adapter->flags & FLAG_HAS_FLASH) &&
5793             (pci_resource_flags(pdev, 1) & IORESOURCE_MEM)) {
5794                 flash_start = pci_resource_start(pdev, 1);
5795                 flash_len = pci_resource_len(pdev, 1);
5796                 adapter->hw.flash_address = ioremap(flash_start, flash_len);
5797                 if (!adapter->hw.flash_address)
5798                         goto err_flashmap;
5799         }
5800
5801         /* construct the net_device struct */
5802         netdev->netdev_ops              = &e1000e_netdev_ops;
5803         e1000e_set_ethtool_ops(netdev);
5804         netdev->watchdog_timeo          = 5 * HZ;
5805         netif_napi_add(netdev, &adapter->napi, e1000_clean, 64);
5806         strncpy(netdev->name, pci_name(pdev), sizeof(netdev->name) - 1);
5807
5808         netdev->mem_start = mmio_start;
5809         netdev->mem_end = mmio_start + mmio_len;
5810
5811         adapter->bd_number = cards_found++;
5812
5813         e1000e_check_options(adapter);
5814
5815         /* setup adapter struct */
5816         err = e1000_sw_init(adapter);
5817         if (err)
5818                 goto err_sw_init;
5819
5820         memcpy(&hw->mac.ops, ei->mac_ops, sizeof(hw->mac.ops));
5821         memcpy(&hw->nvm.ops, ei->nvm_ops, sizeof(hw->nvm.ops));
5822         memcpy(&hw->phy.ops, ei->phy_ops, sizeof(hw->phy.ops));
5823
5824         err = ei->get_variants(adapter);
5825         if (err)
5826                 goto err_hw_init;
5827
5828         if ((adapter->flags & FLAG_IS_ICH) &&
5829             (adapter->flags & FLAG_READ_ONLY_NVM))
5830                 e1000e_write_protect_nvm_ich8lan(&adapter->hw);
5831
5832         hw->mac.ops.get_bus_info(&adapter->hw);
5833
5834         adapter->hw.phy.autoneg_wait_to_complete = 0;
5835
5836         /* Copper options */
5837         if (adapter->hw.phy.media_type == e1000_media_type_copper) {
5838                 adapter->hw.phy.mdix = AUTO_ALL_MODES;
5839                 adapter->hw.phy.disable_polarity_correction = 0;
5840                 adapter->hw.phy.ms_type = e1000_ms_hw_default;
5841         }
5842
5843         if (e1000_check_reset_block(&adapter->hw))
5844                 e_info("PHY reset is blocked due to SOL/IDER session.\n");
5845
5846         netdev->features = NETIF_F_SG |
5847                            NETIF_F_HW_CSUM |
5848                            NETIF_F_HW_VLAN_TX |
5849                            NETIF_F_HW_VLAN_RX;
5850
5851         if (adapter->flags & FLAG_HAS_HW_VLAN_FILTER)
5852                 netdev->features |= NETIF_F_HW_VLAN_FILTER;
5853
5854         netdev->features |= NETIF_F_TSO;
5855         netdev->features |= NETIF_F_TSO6;
5856
5857         netdev->vlan_features |= NETIF_F_TSO;
5858         netdev->vlan_features |= NETIF_F_TSO6;
5859         netdev->vlan_features |= NETIF_F_HW_CSUM;
5860         netdev->vlan_features |= NETIF_F_SG;
5861
5862         if (pci_using_dac) {
5863                 netdev->features |= NETIF_F_HIGHDMA;
5864                 netdev->vlan_features |= NETIF_F_HIGHDMA;
5865         }
5866
5867         if (e1000e_enable_mng_pass_thru(&adapter->hw))
5868                 adapter->flags |= FLAG_MNG_PT_ENABLED;
5869
5870         /*
5871          * before reading the NVM, reset the controller to
5872          * put the device in a known good starting state
5873          */
5874         adapter->hw.mac.ops.reset_hw(&adapter->hw);
5875
5876         /*
5877          * systems with ASPM and others may see the checksum fail on the first
5878          * attempt. Let's give it a few tries
5879          */
5880         for (i = 0;; i++) {
5881                 if (e1000_validate_nvm_checksum(&adapter->hw) >= 0)
5882                         break;
5883                 if (i == 2) {
5884                         e_err("The NVM Checksum Is Not Valid\n");
5885                         err = -EIO;
5886                         goto err_eeprom;
5887                 }
5888         }
5889
5890         e1000_eeprom_checks(adapter);
5891
5892         /* copy the MAC address */
5893         if (e1000e_read_mac_addr(&adapter->hw))
5894                 e_err("NVM Read Error while reading MAC address\n");
5895
5896         memcpy(netdev->dev_addr, adapter->hw.mac.addr, netdev->addr_len);
5897         memcpy(netdev->perm_addr, adapter->hw.mac.addr, netdev->addr_len);
5898
5899         if (!is_valid_ether_addr(netdev->perm_addr)) {
5900                 e_err("Invalid MAC Address: %pM\n", netdev->perm_addr);
5901                 err = -EIO;
5902                 goto err_eeprom;
5903         }
5904
5905         init_timer(&adapter->watchdog_timer);
5906         adapter->watchdog_timer.function = e1000_watchdog;
5907         adapter->watchdog_timer.data = (unsigned long) adapter;
5908
5909         init_timer(&adapter->phy_info_timer);
5910         adapter->phy_info_timer.function = e1000_update_phy_info;
5911         adapter->phy_info_timer.data = (unsigned long) adapter;
5912
5913         INIT_WORK(&adapter->reset_task, e1000_reset_task);
5914         INIT_WORK(&adapter->watchdog_task, e1000_watchdog_task);
5915         INIT_WORK(&adapter->downshift_task, e1000e_downshift_workaround);
5916         INIT_WORK(&adapter->update_phy_task, e1000e_update_phy_task);
5917         INIT_WORK(&adapter->print_hang_task, e1000_print_hw_hang);
5918         INIT_WORK(&adapter->led_blink_task, e1000e_led_blink_task);
5919
5920         /* Initialize link parameters. User can change them with ethtool */
5921         adapter->hw.mac.autoneg = 1;
5922         adapter->fc_autoneg = 1;
5923         adapter->hw.fc.requested_mode = e1000_fc_default;
5924         adapter->hw.fc.current_mode = e1000_fc_default;
5925         adapter->hw.phy.autoneg_advertised = 0x2f;
5926
5927         /* ring size defaults */
5928         adapter->rx_ring->count = 256;
5929         adapter->tx_ring->count = 256;
5930
5931         /*
5932          * Initial Wake on LAN setting - If APM wake is enabled in
5933          * the EEPROM, enable the ACPI Magic Packet filter
5934          */
5935         if (adapter->flags & FLAG_APME_IN_WUC) {
5936                 /* APME bit in EEPROM is mapped to WUC.APME */
5937                 eeprom_data = er32(WUC);
5938                 eeprom_apme_mask = E1000_WUC_APME;
5939                 if (eeprom_data & E1000_WUC_PHY_WAKE)
5940                         adapter->flags2 |= FLAG2_HAS_PHY_WAKEUP;
5941         } else if (adapter->flags & FLAG_APME_IN_CTRL3) {
5942                 if (adapter->flags & FLAG_APME_CHECK_PORT_B &&
5943                     (adapter->hw.bus.func == 1))
5944                         e1000_read_nvm(&adapter->hw,
5945                                 NVM_INIT_CONTROL3_PORT_B, 1, &eeprom_data);
5946                 else
5947                         e1000_read_nvm(&adapter->hw,
5948                                 NVM_INIT_CONTROL3_PORT_A, 1, &eeprom_data);
5949         }
5950
5951         /* fetch WoL from EEPROM */
5952         if (eeprom_data & eeprom_apme_mask)
5953                 adapter->eeprom_wol |= E1000_WUFC_MAG;
5954
5955         /*
5956          * now that we have the eeprom settings, apply the special cases
5957          * where the eeprom may be wrong or the board simply won't support
5958          * wake on lan on a particular port
5959          */
5960         if (!(adapter->flags & FLAG_HAS_WOL))
5961                 adapter->eeprom_wol = 0;
5962
5963         /* initialize the wol settings based on the eeprom settings */
5964         adapter->wol = adapter->eeprom_wol;
5965         device_set_wakeup_enable(&adapter->pdev->dev, adapter->wol);
5966
5967         /* save off EEPROM version number */
5968         e1000_read_nvm(&adapter->hw, 5, 1, &adapter->eeprom_vers);
5969
5970         /* reset the hardware with the new settings */
5971         e1000e_reset(adapter);
5972
5973         /*
5974          * If the controller has AMT, do not set DRV_LOAD until the interface
5975          * is up.  For all other cases, let the f/w know that the h/w is now
5976          * under the control of the driver.
5977          */
5978         if (!(adapter->flags & FLAG_HAS_AMT))
5979                 e1000e_get_hw_control(adapter);
5980
5981         strncpy(netdev->name, "eth%d", sizeof(netdev->name) - 1);
5982         err = register_netdev(netdev);
5983         if (err)
5984                 goto err_register;
5985
5986         /* carrier off reporting is important to ethtool even BEFORE open */
5987         netif_carrier_off(netdev);
5988
5989         e1000_print_device_info(adapter);
5990
5991         if (pci_dev_run_wake(pdev))
5992                 pm_runtime_put_noidle(&pdev->dev);
5993
5994         return 0;
5995
5996 err_register:
5997         if (!(adapter->flags & FLAG_HAS_AMT))
5998                 e1000e_release_hw_control(adapter);
5999 err_eeprom:
6000         if (!e1000_check_reset_block(&adapter->hw))
6001                 e1000_phy_hw_reset(&adapter->hw);
6002 err_hw_init:
6003         kfree(adapter->tx_ring);
6004         kfree(adapter->rx_ring);
6005 err_sw_init:
6006         if (adapter->hw.flash_address)
6007                 iounmap(adapter->hw.flash_address);
6008         e1000e_reset_interrupt_capability(adapter);
6009 err_flashmap:
6010         iounmap(adapter->hw.hw_addr);
6011 err_ioremap:
6012         free_netdev(netdev);
6013 err_alloc_etherdev:
6014         pci_release_selected_regions(pdev,
6015                                      pci_select_bars(pdev, IORESOURCE_MEM));
6016 err_pci_reg:
6017 err_dma:
6018         pci_disable_device(pdev);
6019         return err;
6020 }
6021
6022 /**
6023  * e1000_remove - Device Removal Routine
6024  * @pdev: PCI device information struct
6025  *
6026  * e1000_remove is called by the PCI subsystem to alert the driver
6027  * that it should release a PCI device.  The could be caused by a
6028  * Hot-Plug event, or because the driver is going to be removed from
6029  * memory.
6030  **/
6031 static void __devexit e1000_remove(struct pci_dev *pdev)
6032 {
6033         struct net_device *netdev = pci_get_drvdata(pdev);
6034         struct e1000_adapter *adapter = netdev_priv(netdev);
6035         bool down = test_bit(__E1000_DOWN, &adapter->state);
6036
6037         /*
6038          * The timers may be rescheduled, so explicitly disable them
6039          * from being rescheduled.
6040          */
6041         if (!down)
6042                 set_bit(__E1000_DOWN, &adapter->state);
6043         del_timer_sync(&adapter->watchdog_timer);
6044         del_timer_sync(&adapter->phy_info_timer);
6045
6046         cancel_work_sync(&adapter->reset_task);
6047         cancel_work_sync(&adapter->watchdog_task);
6048         cancel_work_sync(&adapter->downshift_task);
6049         cancel_work_sync(&adapter->update_phy_task);
6050         cancel_work_sync(&adapter->led_blink_task);
6051         cancel_work_sync(&adapter->print_hang_task);
6052
6053         if (!(netdev->flags & IFF_UP))
6054                 e1000_power_down_phy(adapter);
6055
6056         /* Don't lie to e1000_close() down the road. */
6057         if (!down)
6058                 clear_bit(__E1000_DOWN, &adapter->state);
6059         unregister_netdev(netdev);
6060
6061         if (pci_dev_run_wake(pdev))
6062                 pm_runtime_get_noresume(&pdev->dev);
6063
6064         /*
6065          * Release control of h/w to f/w.  If f/w is AMT enabled, this
6066          * would have already happened in close and is redundant.
6067          */
6068         e1000e_release_hw_control(adapter);
6069
6070         e1000e_reset_interrupt_capability(adapter);
6071         kfree(adapter->tx_ring);
6072         kfree(adapter->rx_ring);
6073
6074         iounmap(adapter->hw.hw_addr);
6075         if (adapter->hw.flash_address)
6076                 iounmap(adapter->hw.flash_address);
6077         pci_release_selected_regions(pdev,
6078                                      pci_select_bars(pdev, IORESOURCE_MEM));
6079
6080         free_netdev(netdev);
6081
6082         /* AER disable */
6083         pci_disable_pcie_error_reporting(pdev);
6084
6085         pci_disable_device(pdev);
6086 }
6087
6088 /* PCI Error Recovery (ERS) */
6089 static struct pci_error_handlers e1000_err_handler = {
6090         .error_detected = e1000_io_error_detected,
6091         .slot_reset = e1000_io_slot_reset,
6092         .resume = e1000_io_resume,
6093 };
6094
6095 static DEFINE_PCI_DEVICE_TABLE(e1000_pci_tbl) = {
6096         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82571EB_COPPER), board_82571 },
6097         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82571EB_FIBER), board_82571 },
6098         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82571EB_QUAD_COPPER), board_82571 },
6099         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82571EB_QUAD_COPPER_LP), board_82571 },
6100         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82571EB_QUAD_FIBER), board_82571 },
6101         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82571EB_SERDES), board_82571 },
6102         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82571EB_SERDES_DUAL), board_82571 },
6103         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82571EB_SERDES_QUAD), board_82571 },
6104         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82571PT_QUAD_COPPER), board_82571 },
6105
6106         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82572EI), board_82572 },
6107         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82572EI_COPPER), board_82572 },
6108         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82572EI_FIBER), board_82572 },
6109         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82572EI_SERDES), board_82572 },
6110
6111         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82573E), board_82573 },
6112         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82573E_IAMT), board_82573 },
6113         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82573L), board_82573 },
6114
6115         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82574L), board_82574 },
6116         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82574LA), board_82574 },
6117         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82583V), board_82583 },
6118
6119         { PCI_VDEVICE(INTEL, E1000_DEV_ID_80003ES2LAN_COPPER_DPT),
6120           board_80003es2lan },
6121         { PCI_VDEVICE(INTEL, E1000_DEV_ID_80003ES2LAN_COPPER_SPT),
6122           board_80003es2lan },
6123         { PCI_VDEVICE(INTEL, E1000_DEV_ID_80003ES2LAN_SERDES_DPT),
6124           board_80003es2lan },
6125         { PCI_VDEVICE(INTEL, E1000_DEV_ID_80003ES2LAN_SERDES_SPT),
6126           board_80003es2lan },
6127
6128         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH8_IFE), board_ich8lan },
6129         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH8_IFE_G), board_ich8lan },
6130         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH8_IFE_GT), board_ich8lan },
6131         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH8_IGP_AMT), board_ich8lan },
6132         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH8_IGP_C), board_ich8lan },
6133         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH8_IGP_M), board_ich8lan },
6134         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH8_IGP_M_AMT), board_ich8lan },
6135         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH8_82567V_3), board_ich8lan },
6136
6137         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH9_IFE), board_ich9lan },
6138         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH9_IFE_G), board_ich9lan },
6139         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH9_IFE_GT), board_ich9lan },
6140         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH9_IGP_AMT), board_ich9lan },
6141         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH9_IGP_C), board_ich9lan },
6142         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH9_BM), board_ich9lan },
6143         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH9_IGP_M), board_ich9lan },
6144         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH9_IGP_M_AMT), board_ich9lan },
6145         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH9_IGP_M_V), board_ich9lan },
6146
6147         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH10_R_BM_LM), board_ich9lan },
6148         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH10_R_BM_LF), board_ich9lan },
6149         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH10_R_BM_V), board_ich9lan },
6150
6151         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH10_D_BM_LM), board_ich10lan },
6152         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH10_D_BM_LF), board_ich10lan },
6153         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH10_D_BM_V), board_ich10lan },
6154
6155         { PCI_VDEVICE(INTEL, E1000_DEV_ID_PCH_M_HV_LM), board_pchlan },
6156         { PCI_VDEVICE(INTEL, E1000_DEV_ID_PCH_M_HV_LC), board_pchlan },
6157         { PCI_VDEVICE(INTEL, E1000_DEV_ID_PCH_D_HV_DM), board_pchlan },
6158         { PCI_VDEVICE(INTEL, E1000_DEV_ID_PCH_D_HV_DC), board_pchlan },
6159
6160         { PCI_VDEVICE(INTEL, E1000_DEV_ID_PCH2_LV_LM), board_pch2lan },
6161         { PCI_VDEVICE(INTEL, E1000_DEV_ID_PCH2_LV_V), board_pch2lan },
6162
6163         { }     /* terminate list */
6164 };
6165 MODULE_DEVICE_TABLE(pci, e1000_pci_tbl);
6166
6167 #ifdef CONFIG_PM_OPS
6168 static const struct dev_pm_ops e1000_pm_ops = {
6169         SET_SYSTEM_SLEEP_PM_OPS(e1000_suspend, e1000_resume)
6170         SET_RUNTIME_PM_OPS(e1000_runtime_suspend,
6171                                 e1000_runtime_resume, e1000_idle)
6172 };
6173 #endif
6174
6175 /* PCI Device API Driver */
6176 static struct pci_driver e1000_driver = {
6177         .name     = e1000e_driver_name,
6178         .id_table = e1000_pci_tbl,
6179         .probe    = e1000_probe,
6180         .remove   = __devexit_p(e1000_remove),
6181 #ifdef CONFIG_PM_OPS
6182         .driver.pm = &e1000_pm_ops,
6183 #endif
6184         .shutdown = e1000_shutdown,
6185         .err_handler = &e1000_err_handler
6186 };
6187
6188 /**
6189  * e1000_init_module - Driver Registration Routine
6190  *
6191  * e1000_init_module is the first routine called when the driver is
6192  * loaded. All it does is register with the PCI subsystem.
6193  **/
6194 static int __init e1000_init_module(void)
6195 {
6196         int ret;
6197         pr_info("Intel(R) PRO/1000 Network Driver - %s\n",
6198                 e1000e_driver_version);
6199         pr_info("Copyright(c) 1999 - 2011 Intel Corporation.\n");
6200         ret = pci_register_driver(&e1000_driver);
6201
6202         return ret;
6203 }
6204 module_init(e1000_init_module);
6205
6206 /**
6207  * e1000_exit_module - Driver Exit Cleanup Routine
6208  *
6209  * e1000_exit_module is called just before the driver is removed
6210  * from memory.
6211  **/
6212 static void __exit e1000_exit_module(void)
6213 {
6214         pci_unregister_driver(&e1000_driver);
6215 }
6216 module_exit(e1000_exit_module);
6217
6218
6219 MODULE_AUTHOR("Intel Corporation, <linux.nics@intel.com>");
6220 MODULE_DESCRIPTION("Intel(R) PRO/1000 Network Driver");
6221 MODULE_LICENSE("GPL");
6222 MODULE_VERSION(DRV_VERSION);
6223
6224 /* e1000_main.c */