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 "1.3.10" 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         return cleaned;
 904 }
 905
 906 static void e1000_put_txbuf(struct e1000_adapter *adapter,
 907                              struct e1000_buffer *buffer_info)
 908 {
 909         if (buffer_info->dma) {
 910                 if (buffer_info->mapped_as_page)
 911                         dma_unmap_page(&adapter->pdev->dev, buffer_info->dma,
 912                                        buffer_info->length, DMA_TO_DEVICE);
 913                 else
 914                         dma_unmap_single(&adapter->pdev->dev, buffer_info->dma,
 915                                          buffer_info->length, DMA_TO_DEVICE);
 916                 buffer_info->dma = 0;
 917         }
 918         if (buffer_info->skb) {
 919                 dev_kfree_skb_any(buffer_info->skb);
 920                 buffer_info->skb = NULL;
 921         }
 922         buffer_info->time_stamp = 0;
 923 }
 924
 925 static void e1000_print_hw_hang(struct work_struct *work)
 926 {
 927         struct e1000_adapter *adapter = container_of(work,
 928                                                      struct e1000_adapter,
 929                                                      print_hang_task);
 930         struct e1000_ring *tx_ring = adapter->tx_ring;
 931         unsigned int i = tx_ring->next_to_clean;
 932         unsigned int eop = tx_ring->buffer_info[i].next_to_watch;
 933         struct e1000_tx_desc *eop_desc = E1000_TX_DESC(*tx_ring, eop);
 934         struct e1000_hw *hw = &adapter->hw;
 935         u16 phy_status, phy_1000t_status, phy_ext_status;
 936         u16 pci_status;
 937
 938         if (test_bit(__E1000_DOWN, &adapter->state))
 939                 return;
 940
 941         e1e_rphy(hw, PHY_STATUS, &phy_status);
 942         e1e_rphy(hw, PHY_1000T_STATUS, &phy_1000t_status);
 943         e1e_rphy(hw, PHY_EXT_STATUS, &phy_ext_status);
 944
 945         pci_read_config_word(adapter->pdev, PCI_STATUS, &pci_status);
 946
 947         /* detected Hardware unit hang */
 948         e_err("Detected Hardware Unit Hang:\n"
 949               "  TDH                  <%x>\n"
 950               "  TDT                  <%x>\n"
 951               "  next_to_use          <%x>\n"
 952               "  next_to_clean        <%x>\n"
 953               "buffer_info[next_to_clean]:\n"
 954               "  time_stamp           <%lx>\n"
 955               "  next_to_watch        <%x>\n"
 956               "  jiffies              <%lx>\n"
 957               "  next_to_watch.status <%x>\n"
 958               "MAC Status             <%x>\n"
 959               "PHY Status             <%x>\n"
 960               "PHY 1000BASE-T Status  <%x>\n"
 961               "PHY Extended Status    <%x>\n"
 962               "PCI Status             <%x>\n",
 963               readl(adapter->hw.hw_addr + tx_ring->head),
 964               readl(adapter->hw.hw_addr + tx_ring->tail),
 965               tx_ring->next_to_use,
 966               tx_ring->next_to_clean,
 967               tx_ring->buffer_info[eop].time_stamp,
 968               eop,
 969               jiffies,
 970               eop_desc->upper.fields.status,
 971               er32(STATUS),
 972               phy_status,
 973               phy_1000t_status,
 974               phy_ext_status,
 975               pci_status);
 976 }
 977
 978 /**
 979  * e1000_clean_tx_irq - Reclaim resources after transmit completes
 980  * @adapter: board private structure
 981  *
 982  * the return value indicates whether actual cleaning was done, there
 983  * is no guarantee that everything was cleaned
 984  **/
 985 static bool e1000_clean_tx_irq(struct e1000_adapter *adapter)
 986 {
 987         struct net_device *netdev = adapter->netdev;
 988         struct e1000_hw *hw = &adapter->hw;
 989         struct e1000_ring *tx_ring = adapter->tx_ring;
 990         struct e1000_tx_desc *tx_desc, *eop_desc;
 991         struct e1000_buffer *buffer_info;
 992         unsigned int i, eop;
 993         unsigned int count = 0;
 994         unsigned int total_tx_bytes = 0, total_tx_packets = 0;
 995
 996         i = tx_ring->next_to_clean;
 997         eop = tx_ring->buffer_info[i].next_to_watch;
 998         eop_desc = E1000_TX_DESC(*tx_ring, eop);
 999
1000         while ((eop_desc->upper.data & cpu_to_le32(E1000_TXD_STAT_DD)) &&
1001                (count < tx_ring->count)) {
1002                 bool cleaned = false;
1003                 rmb(); /* read buffer_info after eop_desc */
1004                 for (; !cleaned; count++) {
1005                         tx_desc = E1000_TX_DESC(*tx_ring, i);
1006                         buffer_info = &tx_ring->buffer_info[i];
1007                         cleaned = (i == eop);
1008
1009                         if (cleaned) {
1010                                 total_tx_packets += buffer_info->segs;
1011                                 total_tx_bytes += buffer_info->bytecount;
1012                         }
1013
1014                         e1000_put_txbuf(adapter, buffer_info);
1015                         tx_desc->upper.data = 0;
1016
1017                         i++;
1018                         if (i == tx_ring->count)
1019                                 i = 0;
1020                 }
1021
1022                 if (i == tx_ring->next_to_use)
1023                         break;
1024                 eop = tx_ring->buffer_info[i].next_to_watch;
1025                 eop_desc = E1000_TX_DESC(*tx_ring, eop);
1026         }
1027
1028         tx_ring->next_to_clean = i;
1029
1030 #define TX_WAKE_THRESHOLD 32
1031         if (count && netif_carrier_ok(netdev) &&
1032             e1000_desc_unused(tx_ring) >= TX_WAKE_THRESHOLD) {
1033                 /* Make sure that anybody stopping the queue after this
1034                  * sees the new next_to_clean.
1035                  */
1036                 smp_mb();
1037
1038                 if (netif_queue_stopped(netdev) &&
1039                     !(test_bit(__E1000_DOWN, &adapter->state))) {
1040                         netif_wake_queue(netdev);
1041                         ++adapter->restart_queue;
1042                 }
1043         }
1044
1045         if (adapter->detect_tx_hung) {
1046                 /*
1047                  * Detect a transmit hang in hardware, this serializes the
1048                  * check with the clearing of time_stamp and movement of i
1049                  */
1050                 adapter->detect_tx_hung = 0;
1051                 if (tx_ring->buffer_info[i].time_stamp &&
1052                     time_after(jiffies, tx_ring->buffer_info[i].time_stamp
1053                                + (adapter->tx_timeout_factor * HZ)) &&
1054                     !(er32(STATUS) & E1000_STATUS_TXOFF)) {
1055                         schedule_work(&adapter->print_hang_task);
1056                         netif_stop_queue(netdev);
1057                 }
1058         }
1059         adapter->total_tx_bytes += total_tx_bytes;
1060         adapter->total_tx_packets += total_tx_packets;
1061         return count < tx_ring->count;
1062 }
1063
1064 /**
1065  * e1000_clean_rx_irq_ps - Send received data up the network stack; packet split
1066  * @adapter: board private structure
1067  *
1068  * the return value indicates whether actual cleaning was done, there
1069  * is no guarantee that everything was cleaned
1070  **/
1071 static bool e1000_clean_rx_irq_ps(struct e1000_adapter *adapter,
1072                                   int *work_done, int work_to_do)
1073 {
1074         struct e1000_hw *hw = &adapter->hw;
1075         union e1000_rx_desc_packet_split *rx_desc, *next_rxd;
1076         struct net_device *netdev = adapter->netdev;
1077         struct pci_dev *pdev = adapter->pdev;
1078         struct e1000_ring *rx_ring = adapter->rx_ring;
1079         struct e1000_buffer *buffer_info, *next_buffer;
1080         struct e1000_ps_page *ps_page;
1081         struct sk_buff *skb;
1082         unsigned int i, j;
1083         u32 length, staterr;
1084         int cleaned_count = 0;
1085         bool cleaned = 0;
1086         unsigned int total_rx_bytes = 0, total_rx_packets = 0;
1087
1088         i = rx_ring->next_to_clean;
1089         rx_desc = E1000_RX_DESC_PS(*rx_ring, i);
1090         staterr = le32_to_cpu(rx_desc->wb.middle.status_error);
1091         buffer_info = &rx_ring->buffer_info[i];
1092
1093         while (staterr & E1000_RXD_STAT_DD) {
1094                 if (*work_done >= work_to_do)
1095                         break;
1096                 (*work_done)++;
1097                 skb = buffer_info->skb;
1098                 rmb();  /* read descriptor and rx_buffer_info after status DD */
1099
1100                 /* in the packet split case this is header only */
1101                 prefetch(skb->data - NET_IP_ALIGN);
1102
1103                 i++;
1104                 if (i == rx_ring->count)
1105                         i = 0;
1106                 next_rxd = E1000_RX_DESC_PS(*rx_ring, i);
1107                 prefetch(next_rxd);
1108
1109                 next_buffer = &rx_ring->buffer_info[i];
1110
1111                 cleaned = 1;
1112                 cleaned_count++;
1113                 dma_unmap_single(&pdev->dev, buffer_info->dma,
1114                                  adapter->rx_ps_bsize0, DMA_FROM_DEVICE);
1115                 buffer_info->dma = 0;
1116
1117                 /* see !EOP comment in other Rx routine */
1118                 if (!(staterr & E1000_RXD_STAT_EOP))
1119                         adapter->flags2 |= FLAG2_IS_DISCARDING;
1120
1121                 if (adapter->flags2 & FLAG2_IS_DISCARDING) {
1122                         e_dbg("Packet Split buffers didn't pick up the full "
1123                               "packet\n");
1124                         dev_kfree_skb_irq(skb);
1125                         if (staterr & E1000_RXD_STAT_EOP)
1126                                 adapter->flags2 &= ~FLAG2_IS_DISCARDING;
1127                         goto next_desc;
1128                 }
1129
1130                 if (staterr & E1000_RXDEXT_ERR_FRAME_ERR_MASK) {
1131                         dev_kfree_skb_irq(skb);
1132                         goto next_desc;
1133                 }
1134
1135                 length = le16_to_cpu(rx_desc->wb.middle.length0);
1136
1137                 if (!length) {
1138                         e_dbg("Last part of the packet spanning multiple "
1139                               "descriptors\n");
1140                         dev_kfree_skb_irq(skb);
1141                         goto next_desc;
1142                 }
1143
1144                 /* Good Receive */
1145                 skb_put(skb, length);
1146
1147                 {
1148                 /*
1149                  * this looks ugly, but it seems compiler issues make it
1150                  * more efficient than reusing j
1151                  */
1152                 int l1 = le16_to_cpu(rx_desc->wb.upper.length[0]);
1153
1154                 /*
1155                  * page alloc/put takes too long and effects small packet
1156                  * throughput, so unsplit small packets and save the alloc/put
1157                  * only valid in softirq (napi) context to call kmap_*
1158                  */
1159                 if (l1 && (l1 <= copybreak) &&
1160                     ((length + l1) <= adapter->rx_ps_bsize0)) {
1161                         u8 *vaddr;
1162
1163                         ps_page = &buffer_info->ps_pages[0];
1164
1165                         /*
1166                          * there is no documentation about how to call
1167                          * kmap_atomic, so we can't hold the mapping
1168                          * very long
1169                          */
1170                         dma_sync_single_for_cpu(&pdev->dev, ps_page->dma,
1171                                                 PAGE_SIZE, DMA_FROM_DEVICE);
1172                         vaddr = kmap_atomic(ps_page->page, KM_SKB_DATA_SOFTIRQ);
1173                         memcpy(skb_tail_pointer(skb), vaddr, l1);
1174                         kunmap_atomic(vaddr, KM_SKB_DATA_SOFTIRQ);
1175                         dma_sync_single_for_device(&pdev->dev, ps_page->dma,
1176                                                    PAGE_SIZE, DMA_FROM_DEVICE);
1177
1178                         /* remove the CRC */
1179                         if (!(adapter->flags2 & FLAG2_CRC_STRIPPING))
1180                                 l1 -= 4;
1181
1182                         skb_put(skb, l1);
1183                         goto copydone;
1184                 } /* if */
1185                 }
1186
1187                 for (j = 0; j < PS_PAGE_BUFFERS; j++) {
1188                         length = le16_to_cpu(rx_desc->wb.upper.length[j]);
1189                         if (!length)
1190                                 break;
1191
1192                         ps_page = &buffer_info->ps_pages[j];
1193                         dma_unmap_page(&pdev->dev, ps_page->dma, PAGE_SIZE,
1194                                        DMA_FROM_DEVICE);
1195                         ps_page->dma = 0;
1196                         skb_fill_page_desc(skb, j, ps_page->page, 0, length);
1197                         ps_page->page = NULL;
1198                         skb->len += length;
1199                         skb->data_len += length;
1200                         skb->truesize += length;
1201                 }
1202
1203                 /* strip the ethernet crc, problem is we're using pages now so
1204                  * this whole operation can get a little cpu intensive
1205                  */
1206                 if (!(adapter->flags2 & FLAG2_CRC_STRIPPING))
1207                         pskb_trim(skb, skb->len - 4);
1208
1209 copydone:
1210                 total_rx_bytes += skb->len;
1211                 total_rx_packets++;
1212
1213                 e1000_rx_checksum(adapter, staterr, le16_to_cpu(
1214                         rx_desc->wb.lower.hi_dword.csum_ip.csum), skb);
1215
1216                 if (rx_desc->wb.upper.header_status &
1217                            cpu_to_le16(E1000_RXDPS_HDRSTAT_HDRSP))
1218                         adapter->rx_hdr_split++;
1219
1220                 e1000_receive_skb(adapter, netdev, skb,
1221                                   staterr, rx_desc->wb.middle.vlan);
1222
1223 next_desc:
1224                 rx_desc->wb.middle.status_error &= cpu_to_le32(~0xFF);
1225                 buffer_info->skb = NULL;
1226
1227                 /* return some buffers to hardware, one at a time is too slow */
1228                 if (cleaned_count >= E1000_RX_BUFFER_WRITE) {
1229                         adapter->alloc_rx_buf(adapter, cleaned_count);
1230                         cleaned_count = 0;
1231                 }
1232
1233                 /* use prefetched values */
1234                 rx_desc = next_rxd;
1235                 buffer_info = next_buffer;
1236
1237                 staterr = le32_to_cpu(rx_desc->wb.middle.status_error);
1238         }
1239         rx_ring->next_to_clean = i;
1240
1241         cleaned_count = e1000_desc_unused(rx_ring);
1242         if (cleaned_count)
1243                 adapter->alloc_rx_buf(adapter, cleaned_count);
1244
1245         adapter->total_rx_bytes += total_rx_bytes;
1246         adapter->total_rx_packets += total_rx_packets;
1247         return cleaned;
1248 }
1249
1250 /**
1251  * e1000_consume_page - helper function
1252  **/
1253 static void e1000_consume_page(struct e1000_buffer *bi, struct sk_buff *skb,
1254                                u16 length)
1255 {
1256         bi->page = NULL;
1257         skb->len += length;
1258         skb->data_len += length;
1259         skb->truesize += length;
1260 }
1261
1262 /**
1263  * e1000_clean_jumbo_rx_irq - Send received data up the network stack; legacy
1264  * @adapter: board private structure
1265  *
1266  * the return value indicates whether actual cleaning was done, there
1267  * is no guarantee that everything was cleaned
1268  **/
1269
1270 static bool e1000_clean_jumbo_rx_irq(struct e1000_adapter *adapter,
1271                                      int *work_done, int work_to_do)
1272 {
1273         struct net_device *netdev = adapter->netdev;
1274         struct pci_dev *pdev = adapter->pdev;
1275         struct e1000_ring *rx_ring = adapter->rx_ring;
1276         struct e1000_rx_desc *rx_desc, *next_rxd;
1277         struct e1000_buffer *buffer_info, *next_buffer;
1278         u32 length;
1279         unsigned int i;
1280         int cleaned_count = 0;
1281         bool cleaned = false;
1282         unsigned int total_rx_bytes=0, total_rx_packets=0;
1283
1284         i = rx_ring->next_to_clean;
1285         rx_desc = E1000_RX_DESC(*rx_ring, i);
1286         buffer_info = &rx_ring->buffer_info[i];
1287
1288         while (rx_desc->status & E1000_RXD_STAT_DD) {
1289                 struct sk_buff *skb;
1290                 u8 status;
1291
1292                 if (*work_done >= work_to_do)
1293                         break;
1294                 (*work_done)++;
1295                 rmb();  /* read descriptor and rx_buffer_info after status DD */
1296
1297                 status = rx_desc->status;
1298                 skb = buffer_info->skb;
1299                 buffer_info->skb = NULL;
1300
1301                 ++i;
1302                 if (i == rx_ring->count)
1303                         i = 0;
1304                 next_rxd = E1000_RX_DESC(*rx_ring, i);
1305                 prefetch(next_rxd);
1306
1307                 next_buffer = &rx_ring->buffer_info[i];
1308
1309                 cleaned = true;
1310                 cleaned_count++;
1311                 dma_unmap_page(&pdev->dev, buffer_info->dma, PAGE_SIZE,
1312                                DMA_FROM_DEVICE);
1313                 buffer_info->dma = 0;
1314
1315                 length = le16_to_cpu(rx_desc->length);
1316
1317                 /* errors is only valid for DD + EOP descriptors */
1318                 if (unlikely((status & E1000_RXD_STAT_EOP) &&
1319                     (rx_desc->errors & E1000_RXD_ERR_FRAME_ERR_MASK))) {
1320                                 /* recycle both page and skb */
1321                                 buffer_info->skb = skb;
1322                                 /* an error means any chain goes out the window
1323                                  * too */
1324                                 if (rx_ring->rx_skb_top)
1325                                         dev_kfree_skb_irq(rx_ring->rx_skb_top);
1326                                 rx_ring->rx_skb_top = NULL;
1327                                 goto next_desc;
1328                 }
1329
1330 #define rxtop (rx_ring->rx_skb_top)
1331                 if (!(status & E1000_RXD_STAT_EOP)) {
1332                         /* this descriptor is only the beginning (or middle) */
1333                         if (!rxtop) {
1334                                 /* this is the beginning of a chain */
1335                                 rxtop = skb;
1336                                 skb_fill_page_desc(rxtop, 0, buffer_info->page,
1337                                                    0, length);
1338                         } else {
1339                                 /* this is the middle of a chain */
1340                                 skb_fill_page_desc(rxtop,
1341                                     skb_shinfo(rxtop)->nr_frags,
1342                                     buffer_info->page, 0, length);
1343                                 /* re-use the skb, only consumed the page */
1344                                 buffer_info->skb = skb;
1345                         }
1346                         e1000_consume_page(buffer_info, rxtop, length);
1347                         goto next_desc;
1348                 } else {
1349                         if (rxtop) {
1350                                 /* end of the chain */
1351                                 skb_fill_page_desc(rxtop,
1352                                     skb_shinfo(rxtop)->nr_frags,
1353                                     buffer_info->page, 0, length);
1354                                 /* re-use the current skb, we only consumed the
1355                                  * page */
1356                                 buffer_info->skb = skb;
1357                                 skb = rxtop;
1358                                 rxtop = NULL;
1359                                 e1000_consume_page(buffer_info, skb, length);
1360                         } else {
1361                                 /* no chain, got EOP, this buf is the packet
1362                                  * copybreak to save the put_page/alloc_page */
1363                                 if (length <= copybreak &&
1364                                     skb_tailroom(skb) >= length) {
1365                                         u8 *vaddr;
1366                                         vaddr = kmap_atomic(buffer_info->page,
1367                                                            KM_SKB_DATA_SOFTIRQ);
1368                                         memcpy(skb_tail_pointer(skb), vaddr,
1369                                                length);
1370                                         kunmap_atomic(vaddr,
1371                                                       KM_SKB_DATA_SOFTIRQ);
1372                                         /* re-use the page, so don't erase
1373                                          * buffer_info->page */
1374                                         skb_put(skb, length);
1375                                 } else {
1376                                         skb_fill_page_desc(skb, 0,
1377                                                            buffer_info->page, 0,
1378                                                            length);
1379                                         e1000_consume_page(buffer_info, skb,
1380                                                            length);
1381                                 }
1382                         }
1383                 }
1384
1385                 /* Receive Checksum Offload XXX recompute due to CRC strip? */
1386                 e1000_rx_checksum(adapter,
1387                                   (u32)(status) |
1388                                   ((u32)(rx_desc->errors) << 24),
1389                                   le16_to_cpu(rx_desc->csum), skb);
1390
1391                 /* probably a little skewed due to removing CRC */
1392                 total_rx_bytes += skb->len;
1393                 total_rx_packets++;
1394
1395                 /* eth type trans needs skb->data to point to something */
1396                 if (!pskb_may_pull(skb, ETH_HLEN)) {
1397                         e_err("pskb_may_pull failed.\n");
1398                         dev_kfree_skb_irq(skb);
1399                         goto next_desc;
1400                 }
1401
1402                 e1000_receive_skb(adapter, netdev, skb, status,
1403                                   rx_desc->special);
1404
1405 next_desc:
1406                 rx_desc->status = 0;
1407
1408                 /* return some buffers to hardware, one at a time is too slow */
1409                 if (unlikely(cleaned_count >= E1000_RX_BUFFER_WRITE)) {
1410                         adapter->alloc_rx_buf(adapter, cleaned_count);
1411                         cleaned_count = 0;
1412                 }
1413
1414                 /* use prefetched values */
1415                 rx_desc = next_rxd;
1416                 buffer_info = next_buffer;
1417         }
1418         rx_ring->next_to_clean = i;
1419
1420         cleaned_count = e1000_desc_unused(rx_ring);
1421         if (cleaned_count)
1422                 adapter->alloc_rx_buf(adapter, cleaned_count);
1423
1424         adapter->total_rx_bytes += total_rx_bytes;
1425         adapter->total_rx_packets += total_rx_packets;
1426         return cleaned;
1427 }
1428
1429 /**
1430  * e1000_clean_rx_ring - Free Rx Buffers per Queue
1431  * @adapter: board private structure
1432  **/
1433 static void e1000_clean_rx_ring(struct e1000_adapter *adapter)
1434 {
1435         struct e1000_ring *rx_ring = adapter->rx_ring;
1436         struct e1000_buffer *buffer_info;
1437         struct e1000_ps_page *ps_page;
1438         struct pci_dev *pdev = adapter->pdev;
1439         unsigned int i, j;
1440
1441         /* Free all the Rx ring sk_buffs */
1442         for (i = 0; i < rx_ring->count; i++) {
1443                 buffer_info = &rx_ring->buffer_info[i];
1444                 if (buffer_info->dma) {
1445                         if (adapter->clean_rx == e1000_clean_rx_irq)
1446                                 dma_unmap_single(&pdev->dev, buffer_info->dma,
1447                                                  adapter->rx_buffer_len,
1448                                                  DMA_FROM_DEVICE);
1449                         else if (adapter->clean_rx == e1000_clean_jumbo_rx_irq)
1450                                 dma_unmap_page(&pdev->dev, buffer_info->dma,
1451                                                PAGE_SIZE,
1452                                                DMA_FROM_DEVICE);
1453                         else if (adapter->clean_rx == e1000_clean_rx_irq_ps)
1454                                 dma_unmap_single(&pdev->dev, buffer_info->dma,
1455                                                  adapter->rx_ps_bsize0,
1456                                                  DMA_FROM_DEVICE);
1457                         buffer_info->dma = 0;
1458                 }
1459
1460                 if (buffer_info->page) {
1461                         put_page(buffer_info->page);
1462                         buffer_info->page = NULL;
1463                 }
1464
1465                 if (buffer_info->skb) {
1466                         dev_kfree_skb(buffer_info->skb);
1467                         buffer_info->skb = NULL;
1468                 }
1469
1470                 for (j = 0; j < PS_PAGE_BUFFERS; j++) {
1471                         ps_page = &buffer_info->ps_pages[j];
1472                         if (!ps_page->page)
1473                                 break;
1474                         dma_unmap_page(&pdev->dev, ps_page->dma, PAGE_SIZE,
1475                                        DMA_FROM_DEVICE);
1476                         ps_page->dma = 0;
1477                         put_page(ps_page->page);
1478                         ps_page->page = NULL;
1479                 }
1480         }
1481
1482         /* there also may be some cached data from a chained receive */
1483         if (rx_ring->rx_skb_top) {
1484                 dev_kfree_skb(rx_ring->rx_skb_top);
1485                 rx_ring->rx_skb_top = NULL;
1486         }
1487
1488         /* Zero out the descriptor ring */
1489         memset(rx_ring->desc, 0, rx_ring->size);
1490
1491         rx_ring->next_to_clean = 0;
1492         rx_ring->next_to_use = 0;
1493         adapter->flags2 &= ~FLAG2_IS_DISCARDING;
1494
1495         writel(0, adapter->hw.hw_addr + rx_ring->head);
1496         writel(0, adapter->hw.hw_addr + rx_ring->tail);
1497 }
1498
1499 static void e1000e_downshift_workaround(struct work_struct *work)
1500 {
1501         struct e1000_adapter *adapter = container_of(work,
1502                                         struct e1000_adapter, downshift_task);
1503
1504         if (test_bit(__E1000_DOWN, &adapter->state))
1505                 return;
1506
1507         e1000e_gig_downshift_workaround_ich8lan(&adapter->hw);
1508 }
1509
1510 /**
1511  * e1000_intr_msi - Interrupt Handler
1512  * @irq: interrupt number
1513  * @data: pointer to a network interface device structure
1514  **/
1515 static irqreturn_t e1000_intr_msi(int irq, void *data)
1516 {
1517         struct net_device *netdev = data;
1518         struct e1000_adapter *adapter = netdev_priv(netdev);
1519         struct e1000_hw *hw = &adapter->hw;
1520         u32 icr = er32(ICR);
1521
1522         /*
1523          * read ICR disables interrupts using IAM
1524          */
1525
1526         if (icr & E1000_ICR_LSC) {
1527                 hw->mac.get_link_status = 1;
1528                 /*
1529                  * ICH8 workaround-- Call gig speed drop workaround on cable
1530                  * disconnect (LSC) before accessing any PHY registers
1531                  */
1532                 if ((adapter->flags & FLAG_LSC_GIG_SPEED_DROP) &&
1533                     (!(er32(STATUS) & E1000_STATUS_LU)))
1534                         schedule_work(&adapter->downshift_task);
1535
1536                 /*
1537                  * 80003ES2LAN workaround-- For packet buffer work-around on
1538                  * link down event; disable receives here in the ISR and reset
1539                  * adapter in watchdog
1540                  */
1541                 if (netif_carrier_ok(netdev) &&
1542                     adapter->flags & FLAG_RX_NEEDS_RESTART) {
1543                         /* disable receives */
1544                         u32 rctl = er32(RCTL);
1545                         ew32(RCTL, rctl & ~E1000_RCTL_EN);
1546                         adapter->flags |= FLAG_RX_RESTART_NOW;
1547                 }
1548                 /* guard against interrupt when we're going down */
1549                 if (!test_bit(__E1000_DOWN, &adapter->state))
1550                         mod_timer(&adapter->watchdog_timer, jiffies + 1);
1551         }
1552
1553         if (napi_schedule_prep(&adapter->napi)) {
1554                 adapter->total_tx_bytes = 0;
1555                 adapter->total_tx_packets = 0;
1556                 adapter->total_rx_bytes = 0;
1557                 adapter->total_rx_packets = 0;
1558                 __napi_schedule(&adapter->napi);
1559         }
1560
1561         return IRQ_HANDLED;
1562 }
1563
1564 /**
1565  * e1000_intr - Interrupt Handler
1566  * @irq: interrupt number
1567  * @data: pointer to a network interface device structure
1568  **/
1569 static irqreturn_t e1000_intr(int irq, void *data)
1570 {
1571         struct net_device *netdev = data;
1572         struct e1000_adapter *adapter = netdev_priv(netdev);
1573         struct e1000_hw *hw = &adapter->hw;
1574         u32 rctl, icr = er32(ICR);
1575
1576         if (!icr || test_bit(__E1000_DOWN, &adapter->state))
1577                 return IRQ_NONE;  /* Not our interrupt */
1578
1579         /*
1580          * IMS will not auto-mask if INT_ASSERTED is not set, and if it is
1581          * not set, then the adapter didn't send an interrupt
1582          */
1583         if (!(icr & E1000_ICR_INT_ASSERTED))
1584                 return IRQ_NONE;
1585
1586         /*
1587          * Interrupt Auto-Mask...upon reading ICR,
1588          * interrupts are masked.  No need for the
1589          * IMC write
1590          */
1591
1592         if (icr & E1000_ICR_LSC) {
1593                 hw->mac.get_link_status = 1;
1594                 /*
1595                  * ICH8 workaround-- Call gig speed drop workaround on cable
1596                  * disconnect (LSC) before accessing any PHY registers
1597                  */
1598                 if ((adapter->flags & FLAG_LSC_GIG_SPEED_DROP) &&
1599                     (!(er32(STATUS) & E1000_STATUS_LU)))
1600                         schedule_work(&adapter->downshift_task);
1601
1602                 /*
1603                  * 80003ES2LAN workaround--
1604                  * For packet buffer work-around on link down event;
1605                  * disable receives here in the ISR and
1606                  * reset adapter in watchdog
1607                  */
1608                 if (netif_carrier_ok(netdev) &&
1609                     (adapter->flags & FLAG_RX_NEEDS_RESTART)) {
1610                         /* disable receives */
1611                         rctl = er32(RCTL);
1612                         ew32(RCTL, rctl & ~E1000_RCTL_EN);
1613                         adapter->flags |= FLAG_RX_RESTART_NOW;
1614                 }
1615                 /* guard against interrupt when we're going down */
1616                 if (!test_bit(__E1000_DOWN, &adapter->state))
1617                         mod_timer(&adapter->watchdog_timer, jiffies + 1);
1618         }
1619
1620         if (napi_schedule_prep(&adapter->napi)) {
1621                 adapter->total_tx_bytes = 0;
1622                 adapter->total_tx_packets = 0;
1623                 adapter->total_rx_bytes = 0;
1624                 adapter->total_rx_packets = 0;
1625                 __napi_schedule(&adapter->napi);
1626         }
1627
1628         return IRQ_HANDLED;
1629 }
1630
1631 static irqreturn_t e1000_msix_other(int irq, void *data)
1632 {
1633         struct net_device *netdev = data;
1634         struct e1000_adapter *adapter = netdev_priv(netdev);
1635         struct e1000_hw *hw = &adapter->hw;
1636         u32 icr = er32(ICR);
1637
1638         if (!(icr & E1000_ICR_INT_ASSERTED)) {
1639                 if (!test_bit(__E1000_DOWN, &adapter->state))
1640                         ew32(IMS, E1000_IMS_OTHER);
1641                 return IRQ_NONE;
1642         }
1643
1644         if (icr & adapter->eiac_mask)
1645                 ew32(ICS, (icr & adapter->eiac_mask));
1646
1647         if (icr & E1000_ICR_OTHER) {
1648                 if (!(icr & E1000_ICR_LSC))
1649                         goto no_link_interrupt;
1650                 hw->mac.get_link_status = 1;
1651                 /* guard against interrupt when we're going down */
1652                 if (!test_bit(__E1000_DOWN, &adapter->state))
1653                         mod_timer(&adapter->watchdog_timer, jiffies + 1);
1654         }
1655
1656 no_link_interrupt:
1657         if (!test_bit(__E1000_DOWN, &adapter->state))
1658                 ew32(IMS, E1000_IMS_LSC | E1000_IMS_OTHER);
1659
1660         return IRQ_HANDLED;
1661 }
1662
1663
1664 static irqreturn_t e1000_intr_msix_tx(int irq, void *data)
1665 {
1666         struct net_device *netdev = data;
1667         struct e1000_adapter *adapter = netdev_priv(netdev);
1668         struct e1000_hw *hw = &adapter->hw;
1669         struct e1000_ring *tx_ring = adapter->tx_ring;
1670
1671
1672         adapter->total_tx_bytes = 0;
1673         adapter->total_tx_packets = 0;
1674
1675         if (!e1000_clean_tx_irq(adapter))
1676                 /* Ring was not completely cleaned, so fire another interrupt */
1677                 ew32(ICS, tx_ring->ims_val);
1678
1679         return IRQ_HANDLED;
1680 }
1681
1682 static irqreturn_t e1000_intr_msix_rx(int irq, void *data)
1683 {
1684         struct net_device *netdev = data;
1685         struct e1000_adapter *adapter = netdev_priv(netdev);
1686
1687         /* Write the ITR value calculated at the end of the
1688          * previous interrupt.
1689          */
1690         if (adapter->rx_ring->set_itr) {
1691                 writel(1000000000 / (adapter->rx_ring->itr_val * 256),
1692                        adapter->hw.hw_addr + adapter->rx_ring->itr_register);
1693                 adapter->rx_ring->set_itr = 0;
1694         }
1695
1696         if (napi_schedule_prep(&adapter->napi)) {
1697                 adapter->total_rx_bytes = 0;
1698                 adapter->total_rx_packets = 0;
1699                 __napi_schedule(&adapter->napi);
1700         }
1701         return IRQ_HANDLED;
1702 }
1703
1704 /**
1705  * e1000_configure_msix - Configure MSI-X hardware
1706  *
1707  * e1000_configure_msix sets up the hardware to properly
1708  * generate MSI-X interrupts.
1709  **/
1710 static void e1000_configure_msix(struct e1000_adapter *adapter)
1711 {
1712         struct e1000_hw *hw = &adapter->hw;
1713         struct e1000_ring *rx_ring = adapter->rx_ring;
1714         struct e1000_ring *tx_ring = adapter->tx_ring;
1715         int vector = 0;
1716         u32 ctrl_ext, ivar = 0;
1717
1718         adapter->eiac_mask = 0;
1719
1720         /* Workaround issue with spurious interrupts on 82574 in MSI-X mode */
1721         if (hw->mac.type == e1000_82574) {
1722                 u32 rfctl = er32(RFCTL);
1723                 rfctl |= E1000_RFCTL_ACK_DIS;
1724                 ew32(RFCTL, rfctl);
1725         }
1726
1727 #define E1000_IVAR_INT_ALLOC_VALID      0x8
1728         /* Configure Rx vector */
1729         rx_ring->ims_val = E1000_IMS_RXQ0;
1730         adapter->eiac_mask |= rx_ring->ims_val;
1731         if (rx_ring->itr_val)
1732                 writel(1000000000 / (rx_ring->itr_val * 256),
1733                        hw->hw_addr + rx_ring->itr_register);
1734         else
1735                 writel(1, hw->hw_addr + rx_ring->itr_register);
1736         ivar = E1000_IVAR_INT_ALLOC_VALID | vector;
1737
1738         /* Configure Tx vector */
1739         tx_ring->ims_val = E1000_IMS_TXQ0;
1740         vector++;
1741         if (tx_ring->itr_val)
1742                 writel(1000000000 / (tx_ring->itr_val * 256),
1743                        hw->hw_addr + tx_ring->itr_register);
1744         else
1745                 writel(1, hw->hw_addr + tx_ring->itr_register);
1746         adapter->eiac_mask |= tx_ring->ims_val;
1747         ivar |= ((E1000_IVAR_INT_ALLOC_VALID | vector) << 8);
1748
1749         /* set vector for Other Causes, e.g. link changes */
1750         vector++;
1751         ivar |= ((E1000_IVAR_INT_ALLOC_VALID | vector) << 16);
1752         if (rx_ring->itr_val)
1753                 writel(1000000000 / (rx_ring->itr_val * 256),
1754                        hw->hw_addr + E1000_EITR_82574(vector));
1755         else
1756                 writel(1, hw->hw_addr + E1000_EITR_82574(vector));
1757
1758         /* Cause Tx interrupts on every write back */
1759         ivar |= (1 << 31);
1760
1761         ew32(IVAR, ivar);
1762
1763         /* enable MSI-X PBA support */
1764         ctrl_ext = er32(CTRL_EXT);
1765         ctrl_ext |= E1000_CTRL_EXT_PBA_CLR;
1766
1767         /* Auto-Mask Other interrupts upon ICR read */
1768 #define E1000_EIAC_MASK_82574   0x01F00000
1769         ew32(IAM, ~E1000_EIAC_MASK_82574 | E1000_IMS_OTHER);
1770         ctrl_ext |= E1000_CTRL_EXT_EIAME;
1771         ew32(CTRL_EXT, ctrl_ext);
1772         e1e_flush();
1773 }
1774
1775 void e1000e_reset_interrupt_capability(struct e1000_adapter *adapter)
1776 {
1777         if (adapter->msix_entries) {
1778                 pci_disable_msix(adapter->pdev);
1779                 kfree(adapter->msix_entries);
1780                 adapter->msix_entries = NULL;
1781         } else if (adapter->flags & FLAG_MSI_ENABLED) {
1782                 pci_disable_msi(adapter->pdev);
1783                 adapter->flags &= ~FLAG_MSI_ENABLED;
1784         }
1785 }
1786
1787 /**
1788  * e1000e_set_interrupt_capability - set MSI or MSI-X if supported
1789  *
1790  * Attempt to configure interrupts using the best available
1791  * capabilities of the hardware and kernel.
1792  **/
1793 void e1000e_set_interrupt_capability(struct e1000_adapter *adapter)
1794 {
1795         int err;
1796         int i;
1797
1798         switch (adapter->int_mode) {
1799         case E1000E_INT_MODE_MSIX:
1800                 if (adapter->flags & FLAG_HAS_MSIX) {
1801                         adapter->num_vectors = 3; /* RxQ0, TxQ0 and other */
1802                         adapter->msix_entries = kcalloc(adapter->num_vectors,
1803                                                       sizeof(struct msix_entry),
1804                                                       GFP_KERNEL);
1805                         if (adapter->msix_entries) {
1806                                 for (i = 0; i < adapter->num_vectors; i++)
1807                                         adapter->msix_entries[i].entry = i;
1808
1809                                 err = pci_enable_msix(adapter->pdev,
1810                                                       adapter->msix_entries,
1811                                                       adapter->num_vectors);
1812                                 if (err == 0)
1813                                         return;
1814                         }
1815                         /* MSI-X failed, so fall through and try MSI */
1816                         e_err("Failed to initialize MSI-X interrupts.  "
1817                               "Falling back to MSI interrupts.\n");
1818                         e1000e_reset_interrupt_capability(adapter);
1819                 }
1820                 adapter->int_mode = E1000E_INT_MODE_MSI;
1821                 /* Fall through */
1822         case E1000E_INT_MODE_MSI:
1823                 if (!pci_enable_msi(adapter->pdev)) {
1824                         adapter->flags |= FLAG_MSI_ENABLED;
1825                 } else {
1826                         adapter->int_mode = E1000E_INT_MODE_LEGACY;
1827                         e_err("Failed to initialize MSI interrupts.  Falling "
1828                               "back to legacy interrupts.\n");
1829                 }
1830                 /* Fall through */
1831         case E1000E_INT_MODE_LEGACY:
1832                 /* Don't do anything; this is the system default */
1833                 break;
1834         }
1835
1836         /* store the number of vectors being used */
1837         adapter->num_vectors = 1;
1838 }
1839
1840 /**
1841  * e1000_request_msix - Initialize MSI-X interrupts
1842  *
1843  * e1000_request_msix allocates MSI-X vectors and requests interrupts from the
1844  * kernel.
1845  **/
1846 static int e1000_request_msix(struct e1000_adapter *adapter)
1847 {
1848         struct net_device *netdev = adapter->netdev;
1849         int err = 0, vector = 0;
1850
1851         if (strlen(netdev->name) < (IFNAMSIZ - 5))
1852                 snprintf(adapter->rx_ring->name,
1853                          sizeof(adapter->rx_ring->name) - 1,
1854                          "%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                 snprintf(adapter->tx_ring->name,
1868                          sizeof(adapter->tx_ring->name) - 1,
1869                          "%s-tx-0", netdev->name);
1870         else
1871                 memcpy(adapter->tx_ring->name, netdev->name, IFNAMSIZ);
1872         err = request_irq(adapter->msix_entries[vector].vector,
1873                           e1000_intr_msix_tx, 0, adapter->tx_ring->name,
1874                           netdev);
1875         if (err)
1876                 goto out;
1877         adapter->tx_ring->itr_register = E1000_EITR_82574(vector);
1878         adapter->tx_ring->itr_val = adapter->itr;
1879         vector++;
1880
1881         err = request_irq(adapter->msix_entries[vector].vector,
1882                           e1000_msix_other, 0, netdev->name, netdev);
1883         if (err)
1884                 goto out;
1885
1886         e1000_configure_msix(adapter);
1887         return 0;
1888 out:
1889         return err;
1890 }
1891
1892 /**
1893  * e1000_request_irq - initialize interrupts
1894  *
1895  * Attempts to configure interrupts using the best available
1896  * capabilities of the hardware and kernel.
1897  **/
1898 static int e1000_request_irq(struct e1000_adapter *adapter)
1899 {
1900         struct net_device *netdev = adapter->netdev;
1901         int err;
1902
1903         if (adapter->msix_entries) {
1904                 err = e1000_request_msix(adapter);
1905                 if (!err)
1906                         return err;
1907                 /* fall back to MSI */
1908                 e1000e_reset_interrupt_capability(adapter);
1909                 adapter->int_mode = E1000E_INT_MODE_MSI;
1910                 e1000e_set_interrupt_capability(adapter);
1911         }
1912         if (adapter->flags & FLAG_MSI_ENABLED) {
1913                 err = request_irq(adapter->pdev->irq, e1000_intr_msi, 0,
1914                                   netdev->name, netdev);
1915                 if (!err)
1916                         return err;
1917
1918                 /* fall back to legacy interrupt */
1919                 e1000e_reset_interrupt_capability(adapter);
1920                 adapter->int_mode = E1000E_INT_MODE_LEGACY;
1921         }
1922
1923         err = request_irq(adapter->pdev->irq, e1000_intr, IRQF_SHARED,
1924                           netdev->name, netdev);
1925         if (err)
1926                 e_err("Unable to allocate interrupt, Error: %d\n", err);
1927
1928         return err;
1929 }
1930
1931 static void e1000_free_irq(struct e1000_adapter *adapter)
1932 {
1933         struct net_device *netdev = adapter->netdev;
1934
1935         if (adapter->msix_entries) {
1936                 int vector = 0;
1937
1938                 free_irq(adapter->msix_entries[vector].vector, netdev);
1939                 vector++;
1940
1941                 free_irq(adapter->msix_entries[vector].vector, netdev);
1942                 vector++;
1943
1944                 /* Other Causes interrupt vector */
1945                 free_irq(adapter->msix_entries[vector].vector, netdev);
1946                 return;
1947         }
1948
1949         free_irq(adapter->pdev->irq, netdev);
1950 }
1951
1952 /**
1953  * e1000_irq_disable - Mask off interrupt generation on the NIC
1954  **/
1955 static void e1000_irq_disable(struct e1000_adapter *adapter)
1956 {
1957         struct e1000_hw *hw = &adapter->hw;
1958
1959         ew32(IMC, ~0);
1960         if (adapter->msix_entries)
1961                 ew32(EIAC_82574, 0);
1962         e1e_flush();
1963
1964         if (adapter->msix_entries) {
1965                 int i;
1966                 for (i = 0; i < adapter->num_vectors; i++)
1967                         synchronize_irq(adapter->msix_entries[i].vector);
1968         } else {
1969                 synchronize_irq(adapter->pdev->irq);
1970         }
1971 }
1972
1973 /**
1974  * e1000_irq_enable - Enable default interrupt generation settings
1975  **/
1976 static void e1000_irq_enable(struct e1000_adapter *adapter)
1977 {
1978         struct e1000_hw *hw = &adapter->hw;
1979
1980         if (adapter->msix_entries) {
1981                 ew32(EIAC_82574, adapter->eiac_mask & E1000_EIAC_MASK_82574);
1982                 ew32(IMS, adapter->eiac_mask | E1000_IMS_OTHER | E1000_IMS_LSC);
1983         } else {
1984                 ew32(IMS, IMS_ENABLE_MASK);
1985         }
1986         e1e_flush();
1987 }
1988
1989 /**
1990  * e1000e_get_hw_control - get control of the h/w from f/w
1991  * @adapter: address of board private structure
1992  *
1993  * e1000e_get_hw_control sets {CTRL_EXT|SWSM}:DRV_LOAD bit.
1994  * For ASF and Pass Through versions of f/w this means that
1995  * the driver is loaded. For AMT version (only with 82573)
1996  * of the f/w this means that the network i/f is open.
1997  **/
1998 void e1000e_get_hw_control(struct e1000_adapter *adapter)
1999 {
2000         struct e1000_hw *hw = &adapter->hw;
2001         u32 ctrl_ext;
2002         u32 swsm;
2003
2004         /* Let firmware know the driver has taken over */
2005         if (adapter->flags & FLAG_HAS_SWSM_ON_LOAD) {
2006                 swsm = er32(SWSM);
2007                 ew32(SWSM, swsm | E1000_SWSM_DRV_LOAD);
2008         } else if (adapter->flags & FLAG_HAS_CTRLEXT_ON_LOAD) {
2009                 ctrl_ext = er32(CTRL_EXT);
2010                 ew32(CTRL_EXT, ctrl_ext | E1000_CTRL_EXT_DRV_LOAD);
2011         }
2012 }
2013
2014 /**
2015  * e1000e_release_hw_control - release control of the h/w to f/w
2016  * @adapter: address of board private structure
2017  *
2018  * e1000e_release_hw_control resets {CTRL_EXT|SWSM}:DRV_LOAD bit.
2019  * For ASF and Pass Through versions of f/w this means that the
2020  * driver is no longer loaded. For AMT version (only with 82573) i
2021  * of the f/w this means that the network i/f is closed.
2022  *
2023  **/
2024 void e1000e_release_hw_control(struct e1000_adapter *adapter)
2025 {
2026         struct e1000_hw *hw = &adapter->hw;
2027         u32 ctrl_ext;
2028         u32 swsm;
2029
2030         /* Let firmware taken over control of h/w */
2031         if (adapter->flags & FLAG_HAS_SWSM_ON_LOAD) {
2032                 swsm = er32(SWSM);
2033                 ew32(SWSM, swsm & ~E1000_SWSM_DRV_LOAD);
2034         } else if (adapter->flags & FLAG_HAS_CTRLEXT_ON_LOAD) {
2035                 ctrl_ext = er32(CTRL_EXT);
2036                 ew32(CTRL_EXT, ctrl_ext & ~E1000_CTRL_EXT_DRV_LOAD);
2037         }
2038 }
2039
2040 /**
2041  * @e1000_alloc_ring - allocate memory for a ring structure
2042  **/
2043 static int e1000_alloc_ring_dma(struct e1000_adapter *adapter,
2044                                 struct e1000_ring *ring)
2045 {
2046         struct pci_dev *pdev = adapter->pdev;
2047
2048         ring->desc = dma_alloc_coherent(&pdev->dev, ring->size, &ring->dma,
2049                                         GFP_KERNEL);
2050         if (!ring->desc)
2051                 return -ENOMEM;
2052
2053         return 0;
2054 }
2055
2056 /**
2057  * e1000e_setup_tx_resources - allocate Tx resources (Descriptors)
2058  * @adapter: board private structure
2059  *
2060  * Return 0 on success, negative on failure
2061  **/
2062 int e1000e_setup_tx_resources(struct e1000_adapter *adapter)
2063 {
2064         struct e1000_ring *tx_ring = adapter->tx_ring;
2065         int err = -ENOMEM, size;
2066
2067         size = sizeof(struct e1000_buffer) * tx_ring->count;
2068         tx_ring->buffer_info = vzalloc(size);
2069         if (!tx_ring->buffer_info)
2070                 goto err;
2071
2072         /* round up to nearest 4K */
2073         tx_ring->size = tx_ring->count * sizeof(struct e1000_tx_desc);
2074         tx_ring->size = ALIGN(tx_ring->size, 4096);
2075
2076         err = e1000_alloc_ring_dma(adapter, tx_ring);
2077         if (err)
2078                 goto err;
2079
2080         tx_ring->next_to_use = 0;
2081         tx_ring->next_to_clean = 0;
2082
2083         return 0;
2084 err:
2085         vfree(tx_ring->buffer_info);
2086         e_err("Unable to allocate memory for the transmit descriptor ring\n");
2087         return err;
2088 }
2089
2090 /**
2091  * e1000e_setup_rx_resources - allocate Rx resources (Descriptors)
2092  * @adapter: board private structure
2093  *
2094  * Returns 0 on success, negative on failure
2095  **/
2096 int e1000e_setup_rx_resources(struct e1000_adapter *adapter)
2097 {
2098         struct e1000_ring *rx_ring = adapter->rx_ring;
2099         struct e1000_buffer *buffer_info;
2100         int i, size, desc_len, err = -ENOMEM;
2101
2102         size = sizeof(struct e1000_buffer) * rx_ring->count;
2103         rx_ring->buffer_info = vzalloc(size);
2104         if (!rx_ring->buffer_info)
2105                 goto err;
2106
2107         for (i = 0; i < rx_ring->count; i++) {
2108                 buffer_info = &rx_ring->buffer_info[i];
2109                 buffer_info->ps_pages = kcalloc(PS_PAGE_BUFFERS,
2110                                                 sizeof(struct e1000_ps_page),
2111                                                 GFP_KERNEL);
2112                 if (!buffer_info->ps_pages)
2113                         goto err_pages;
2114         }
2115
2116         desc_len = sizeof(union e1000_rx_desc_packet_split);
2117
2118         /* Round up to nearest 4K */
2119         rx_ring->size = rx_ring->count * desc_len;
2120         rx_ring->size = ALIGN(rx_ring->size, 4096);
2121
2122         err = e1000_alloc_ring_dma(adapter, rx_ring);
2123         if (err)
2124                 goto err_pages;
2125
2126         rx_ring->next_to_clean = 0;
2127         rx_ring->next_to_use = 0;
2128         rx_ring->rx_skb_top = NULL;
2129
2130         return 0;
2131
2132 err_pages:
2133         for (i = 0; i < rx_ring->count; i++) {
2134                 buffer_info = &rx_ring->buffer_info[i];
2135                 kfree(buffer_info->ps_pages);
2136         }
2137 err:
2138         vfree(rx_ring->buffer_info);
2139         e_err("Unable to allocate memory for the receive descriptor ring\n");
2140         return err;
2141 }
2142
2143 /**
2144  * e1000_clean_tx_ring - Free Tx Buffers
2145  * @adapter: board private structure
2146  **/
2147 static void e1000_clean_tx_ring(struct e1000_adapter *adapter)
2148 {
2149         struct e1000_ring *tx_ring = adapter->tx_ring;
2150         struct e1000_buffer *buffer_info;
2151         unsigned long size;
2152         unsigned int i;
2153
2154         for (i = 0; i < tx_ring->count; i++) {
2155                 buffer_info = &tx_ring->buffer_info[i];
2156                 e1000_put_txbuf(adapter, buffer_info);
2157         }
2158
2159         size = sizeof(struct e1000_buffer) * tx_ring->count;
2160         memset(tx_ring->buffer_info, 0, size);
2161
2162         memset(tx_ring->desc, 0, tx_ring->size);
2163
2164         tx_ring->next_to_use = 0;
2165         tx_ring->next_to_clean = 0;
2166
2167         writel(0, adapter->hw.hw_addr + tx_ring->head);
2168         writel(0, adapter->hw.hw_addr + tx_ring->tail);
2169 }
2170
2171 /**
2172  * e1000e_free_tx_resources - Free Tx Resources per Queue
2173  * @adapter: board private structure
2174  *
2175  * Free all transmit software resources
2176  **/
2177 void e1000e_free_tx_resources(struct e1000_adapter *adapter)
2178 {
2179         struct pci_dev *pdev = adapter->pdev;
2180         struct e1000_ring *tx_ring = adapter->tx_ring;
2181
2182         e1000_clean_tx_ring(adapter);
2183
2184         vfree(tx_ring->buffer_info);
2185         tx_ring->buffer_info = NULL;
2186
2187         dma_free_coherent(&pdev->dev, tx_ring->size, tx_ring->desc,
2188                           tx_ring->dma);
2189         tx_ring->desc = NULL;
2190 }
2191
2192 /**
2193  * e1000e_free_rx_resources - Free Rx Resources
2194  * @adapter: board private structure
2195  *
2196  * Free all receive software resources
2197  **/
2198
2199 void e1000e_free_rx_resources(struct e1000_adapter *adapter)
2200 {
2201         struct pci_dev *pdev = adapter->pdev;
2202         struct e1000_ring *rx_ring = adapter->rx_ring;
2203         int i;
2204
2205         e1000_clean_rx_ring(adapter);
2206
2207         for (i = 0; i < rx_ring->count; i++)
2208                 kfree(rx_ring->buffer_info[i].ps_pages);
2209
2210         vfree(rx_ring->buffer_info);
2211         rx_ring->buffer_info = NULL;
2212
2213         dma_free_coherent(&pdev->dev, rx_ring->size, rx_ring->desc,
2214                           rx_ring->dma);
2215         rx_ring->desc = NULL;
2216 }
2217
2218 /**
2219  * e1000_update_itr - update the dynamic ITR value based on statistics
2220  * @adapter: pointer to adapter
2221  * @itr_setting: current adapter->itr
2222  * @packets: the number of packets during this measurement interval
2223  * @bytes: the number of bytes during this measurement interval
2224  *
2225  *      Stores a new ITR value based on packets and byte
2226  *      counts during the last interrupt.  The advantage of per interrupt
2227  *      computation is faster updates and more accurate ITR for the current
2228  *      traffic pattern.  Constants in this function were computed
2229  *      based on theoretical maximum wire speed and thresholds were set based
2230  *      on testing data as well as attempting to minimize response time
2231  *      while increasing bulk throughput.  This functionality is controlled
2232  *      by the InterruptThrottleRate module parameter.
2233  **/
2234 static unsigned int e1000_update_itr(struct e1000_adapter *adapter,
2235                                      u16 itr_setting, int packets,
2236                                      int bytes)
2237 {
2238         unsigned int retval = itr_setting;
2239
2240         if (packets == 0)
2241                 goto update_itr_done;
2242
2243         switch (itr_setting) {
2244         case lowest_latency:
2245                 /* handle TSO and jumbo frames */
2246                 if (bytes/packets > 8000)
2247                         retval = bulk_latency;
2248                 else if ((packets < 5) && (bytes > 512))
2249                         retval = low_latency;
2250                 break;
2251         case low_latency:  /* 50 usec aka 20000 ints/s */
2252                 if (bytes > 10000) {
2253                         /* this if handles the TSO accounting */
2254                         if (bytes/packets > 8000)
2255                                 retval = bulk_latency;
2256                         else if ((packets < 10) || ((bytes/packets) > 1200))
2257                                 retval = bulk_latency;
2258                         else if ((packets > 35))
2259                                 retval = lowest_latency;
2260                 } else if (bytes/packets > 2000) {
2261                         retval = bulk_latency;
2262                 } else if (packets <= 2 && bytes < 512) {
2263                         retval = lowest_latency;
2264                 }
2265                 break;
2266         case bulk_latency: /* 250 usec aka 4000 ints/s */
2267                 if (bytes > 25000) {
2268                         if (packets > 35)
2269                                 retval = low_latency;
2270                 } else if (bytes < 6000) {
2271                         retval = low_latency;
2272                 }
2273                 break;
2274         }
2275
2276 update_itr_done:
2277         return retval;
2278 }
2279
2280 static void e1000_set_itr(struct e1000_adapter *adapter)
2281 {
2282         struct e1000_hw *hw = &adapter->hw;
2283         u16 current_itr;
2284         u32 new_itr = adapter->itr;
2285
2286         /* for non-gigabit speeds, just fix the interrupt rate at 4000 */
2287         if (adapter->link_speed != SPEED_1000) {
2288                 current_itr = 0;
2289                 new_itr = 4000;
2290                 goto set_itr_now;
2291         }
2292
2293         if (adapter->flags2 & FLAG2_DISABLE_AIM) {
2294                 new_itr = 0;
2295                 goto set_itr_now;
2296         }
2297
2298         adapter->tx_itr = e1000_update_itr(adapter,
2299                                     adapter->tx_itr,
2300                                     adapter->total_tx_packets,
2301                                     adapter->total_tx_bytes);
2302         /* conservative mode (itr 3) eliminates the lowest_latency setting */
2303         if (adapter->itr_setting == 3 && adapter->tx_itr == lowest_latency)
2304                 adapter->tx_itr = low_latency;
2305
2306         adapter->rx_itr = e1000_update_itr(adapter,
2307                                     adapter->rx_itr,
2308                                     adapter->total_rx_packets,
2309                                     adapter->total_rx_bytes);
2310         /* conservative mode (itr 3) eliminates the lowest_latency setting */
2311         if (adapter->itr_setting == 3 && adapter->rx_itr == lowest_latency)
2312                 adapter->rx_itr = low_latency;
2313
2314         current_itr = max(adapter->rx_itr, adapter->tx_itr);
2315
2316         switch (current_itr) {
2317         /* counts and packets in update_itr are dependent on these numbers */
2318         case lowest_latency:
2319                 new_itr = 70000;
2320                 break;
2321         case low_latency:
2322                 new_itr = 20000; /* aka hwitr = ~200 */
2323                 break;
2324         case bulk_latency:
2325                 new_itr = 4000;
2326                 break;
2327         default:
2328                 break;
2329         }
2330
2331 set_itr_now:
2332         if (new_itr != adapter->itr) {
2333                 /*
2334                  * this attempts to bias the interrupt rate towards Bulk
2335                  * by adding intermediate steps when interrupt rate is
2336                  * increasing
2337                  */
2338                 new_itr = new_itr > adapter->itr ?
2339                              min(adapter->itr + (new_itr >> 2), new_itr) :
2340                              new_itr;
2341                 adapter->itr = new_itr;
2342                 adapter->rx_ring->itr_val = new_itr;
2343                 if (adapter->msix_entries)
2344                         adapter->rx_ring->set_itr = 1;
2345                 else
2346                         if (new_itr)
2347                                 ew32(ITR, 1000000000 / (new_itr * 256));
2348                         else
2349                                 ew32(ITR, 0);
2350         }
2351 }
2352
2353 /**
2354  * e1000_alloc_queues - Allocate memory for all rings
2355  * @adapter: board private structure to initialize
2356  **/
2357 static int __devinit e1000_alloc_queues(struct e1000_adapter *adapter)
2358 {
2359         adapter->tx_ring = kzalloc(sizeof(struct e1000_ring), GFP_KERNEL);
2360         if (!adapter->tx_ring)
2361                 goto err;
2362
2363         adapter->rx_ring = kzalloc(sizeof(struct e1000_ring), GFP_KERNEL);
2364         if (!adapter->rx_ring)
2365                 goto err;
2366
2367         return 0;
2368 err:
2369         e_err("Unable to allocate memory for queues\n");
2370         kfree(adapter->rx_ring);
2371         kfree(adapter->tx_ring);
2372         return -ENOMEM;
2373 }
2374
2375 /**
2376  * e1000_clean - NAPI Rx polling callback
2377  * @napi: struct associated with this polling callback
2378  * @budget: amount of packets driver is allowed to process this poll
2379  **/
2380 static int e1000_clean(struct napi_struct *napi, int budget)
2381 {
2382         struct e1000_adapter *adapter = container_of(napi, struct e1000_adapter, napi);
2383         struct e1000_hw *hw = &adapter->hw;
2384         struct net_device *poll_dev = adapter->netdev;
2385         int tx_cleaned = 1, work_done = 0;
2386
2387         adapter = netdev_priv(poll_dev);
2388
2389         if (adapter->msix_entries &&
2390             !(adapter->rx_ring->ims_val & adapter->tx_ring->ims_val))
2391                 goto clean_rx;
2392
2393         tx_cleaned = e1000_clean_tx_irq(adapter);
2394
2395 clean_rx:
2396         adapter->clean_rx(adapter, &work_done, budget);
2397
2398         if (!tx_cleaned)
2399                 work_done = budget;
2400
2401         /* If budget not fully consumed, exit the polling mode */
2402         if (work_done < budget) {
2403                 if (adapter->itr_setting & 3)
2404                         e1000_set_itr(adapter);
2405                 napi_complete(napi);
2406                 if (!test_bit(__E1000_DOWN, &adapter->state)) {
2407                         if (adapter->msix_entries)
2408                                 ew32(IMS, adapter->rx_ring->ims_val);
2409                         else
2410                                 e1000_irq_enable(adapter);
2411                 }
2412         }
2413
2414         return work_done;
2415 }
2416
2417 static void e1000_vlan_rx_add_vid(struct net_device *netdev, u16 vid)
2418 {
2419         struct e1000_adapter *adapter = netdev_priv(netdev);
2420         struct e1000_hw *hw = &adapter->hw;
2421         u32 vfta, index;
2422
2423         /* don't update vlan cookie if already programmed */
2424         if ((adapter->hw.mng_cookie.status &
2425              E1000_MNG_DHCP_COOKIE_STATUS_VLAN) &&
2426             (vid == adapter->mng_vlan_id))
2427                 return;
2428
2429         /* add VID to filter table */
2430         if (adapter->flags & FLAG_HAS_HW_VLAN_FILTER) {
2431                 index = (vid >> 5) & 0x7F;
2432                 vfta = E1000_READ_REG_ARRAY(hw, E1000_VFTA, index);
2433                 vfta |= (1 << (vid & 0x1F));
2434                 hw->mac.ops.write_vfta(hw, index, vfta);
2435         }
2436 }
2437
2438 static void e1000_vlan_rx_kill_vid(struct net_device *netdev, u16 vid)
2439 {
2440         struct e1000_adapter *adapter = netdev_priv(netdev);
2441         struct e1000_hw *hw = &adapter->hw;
2442         u32 vfta, index;
2443
2444         if (!test_bit(__E1000_DOWN, &adapter->state))
2445                 e1000_irq_disable(adapter);
2446         vlan_group_set_device(adapter->vlgrp, vid, NULL);
2447
2448         if (!test_bit(__E1000_DOWN, &adapter->state))
2449                 e1000_irq_enable(adapter);
2450
2451         if ((adapter->hw.mng_cookie.status &
2452              E1000_MNG_DHCP_COOKIE_STATUS_VLAN) &&
2453             (vid == adapter->mng_vlan_id)) {
2454                 /* release control to f/w */
2455                 e1000e_release_hw_control(adapter);
2456                 return;
2457         }
2458
2459         /* remove VID from filter table */
2460         if (adapter->flags & FLAG_HAS_HW_VLAN_FILTER) {
2461                 index = (vid >> 5) & 0x7F;
2462                 vfta = E1000_READ_REG_ARRAY(hw, E1000_VFTA, index);
2463                 vfta &= ~(1 << (vid & 0x1F));
2464                 hw->mac.ops.write_vfta(hw, index, vfta);
2465         }
2466 }
2467
2468 static void e1000_update_mng_vlan(struct e1000_adapter *adapter)
2469 {
2470         struct net_device *netdev = adapter->netdev;
2471         u16 vid = adapter->hw.mng_cookie.vlan_id;
2472         u16 old_vid = adapter->mng_vlan_id;
2473
2474         if (!adapter->vlgrp)
2475                 return;
2476
2477         if (!vlan_group_get_device(adapter->vlgrp, vid)) {
2478                 adapter->mng_vlan_id = E1000_MNG_VLAN_NONE;
2479                 if (adapter->hw.mng_cookie.status &
2480                         E1000_MNG_DHCP_COOKIE_STATUS_VLAN) {
2481                         e1000_vlan_rx_add_vid(netdev, vid);
2482                         adapter->mng_vlan_id = vid;
2483                 }
2484
2485                 if ((old_vid != (u16)E1000_MNG_VLAN_NONE) &&
2486                                 (vid != old_vid) &&
2487                     !vlan_group_get_device(adapter->vlgrp, old_vid))
2488                         e1000_vlan_rx_kill_vid(netdev, old_vid);
2489         } else {
2490                 adapter->mng_vlan_id = vid;
2491         }
2492 }
2493
2494
2495 static void e1000_vlan_rx_register(struct net_device *netdev,
2496                                    struct vlan_group *grp)
2497 {
2498         struct e1000_adapter *adapter = netdev_priv(netdev);
2499         struct e1000_hw *hw = &adapter->hw;
2500         u32 ctrl, rctl;
2501
2502         if (!test_bit(__E1000_DOWN, &adapter->state))
2503                 e1000_irq_disable(adapter);
2504         adapter->vlgrp = grp;
2505
2506         if (grp) {
2507                 /* enable VLAN tag insert/strip */
2508                 ctrl = er32(CTRL);
2509                 ctrl |= E1000_CTRL_VME;
2510                 ew32(CTRL, ctrl);
2511
2512                 if (adapter->flags & FLAG_HAS_HW_VLAN_FILTER) {
2513                         /* enable VLAN receive filtering */
2514                         rctl = er32(RCTL);
2515                         rctl &= ~E1000_RCTL_CFIEN;
2516                         ew32(RCTL, rctl);
2517                         e1000_update_mng_vlan(adapter);
2518                 }
2519         } else {
2520                 /* disable VLAN tag insert/strip */
2521                 ctrl = er32(CTRL);
2522                 ctrl &= ~E1000_CTRL_VME;
2523                 ew32(CTRL, ctrl);
2524
2525                 if (adapter->flags & FLAG_HAS_HW_VLAN_FILTER) {
2526                         if (adapter->mng_vlan_id !=
2527                             (u16)E1000_MNG_VLAN_NONE) {
2528                                 e1000_vlan_rx_kill_vid(netdev,
2529                                                        adapter->mng_vlan_id);
2530                                 adapter->mng_vlan_id = E1000_MNG_VLAN_NONE;
2531                         }
2532                 }
2533         }
2534
2535         if (!test_bit(__E1000_DOWN, &adapter->state))
2536                 e1000_irq_enable(adapter);
2537 }
2538
2539 static void e1000_restore_vlan(struct e1000_adapter *adapter)
2540 {
2541         u16 vid;
2542
2543         e1000_vlan_rx_register(adapter->netdev, adapter->vlgrp);
2544
2545         if (!adapter->vlgrp)
2546                 return;
2547
2548         for (vid = 0; vid < VLAN_N_VID; vid++) {
2549                 if (!vlan_group_get_device(adapter->vlgrp, vid))
2550                         continue;
2551                 e1000_vlan_rx_add_vid(adapter->netdev, vid);
2552         }
2553 }
2554
2555 static void e1000_init_manageability_pt(struct e1000_adapter *adapter)
2556 {
2557         struct e1000_hw *hw = &adapter->hw;
2558         u32 manc, manc2h, mdef, i, j;
2559
2560         if (!(adapter->flags & FLAG_MNG_PT_ENABLED))
2561                 return;
2562
2563         manc = er32(MANC);
2564
2565         /*
2566          * enable receiving management packets to the host. this will probably
2567          * generate destination unreachable messages from the host OS, but
2568          * the packets will be handled on SMBUS
2569          */
2570         manc |= E1000_MANC_EN_MNG2HOST;
2571         manc2h = er32(MANC2H);
2572
2573         switch (hw->mac.type) {
2574         default:
2575                 manc2h |= (E1000_MANC2H_PORT_623 | E1000_MANC2H_PORT_664);
2576                 break;
2577         case e1000_82574:
2578         case e1000_82583:
2579                 /*
2580                  * Check if IPMI pass-through decision filter already exists;
2581                  * if so, enable it.
2582                  */
2583                 for (i = 0, j = 0; i < 8; i++) {
2584                         mdef = er32(MDEF(i));
2585
2586                         /* Ignore filters with anything other than IPMI ports */
2587                         if (mdef & ~(E1000_MDEF_PORT_623 | E1000_MDEF_PORT_664))
2588                                 continue;
2589
2590                         /* Enable this decision filter in MANC2H */
2591                         if (mdef)
2592                                 manc2h |= (1 << i);
2593
2594                         j |= mdef;
2595                 }
2596
2597                 if (j == (E1000_MDEF_PORT_623 | E1000_MDEF_PORT_664))
2598                         break;
2599
2600                 /* Create new decision filter in an empty filter */
2601                 for (i = 0, j = 0; i < 8; i++)
2602                         if (er32(MDEF(i)) == 0) {
2603                                 ew32(MDEF(i), (E1000_MDEF_PORT_623 |
2604                                                E1000_MDEF_PORT_664));
2605                                 manc2h |= (1 << 1);
2606                                 j++;
2607                                 break;
2608                         }
2609
2610                 if (!j)
2611                         e_warn("Unable to create IPMI pass-through filter\n");
2612                 break;
2613         }
2614
2615         ew32(MANC2H, manc2h);
2616         ew32(MANC, manc);
2617 }
2618
2619 /**
2620  * e1000_configure_tx - Configure Transmit Unit after Reset
2621  * @adapter: board private structure
2622  *
2623  * Configure the Tx unit of the MAC after a reset.
2624  **/
2625 static void e1000_configure_tx(struct e1000_adapter *adapter)
2626 {
2627         struct e1000_hw *hw = &adapter->hw;
2628         struct e1000_ring *tx_ring = adapter->tx_ring;
2629         u64 tdba;
2630         u32 tdlen, tctl, tipg, tarc;
2631         u32 ipgr1, ipgr2;
2632
2633         /* Setup the HW Tx Head and Tail descriptor pointers */
2634         tdba = tx_ring->dma;
2635         tdlen = tx_ring->count * sizeof(struct e1000_tx_desc);
2636         ew32(TDBAL, (tdba & DMA_BIT_MASK(32)));
2637         ew32(TDBAH, (tdba >> 32));
2638         ew32(TDLEN, tdlen);
2639         ew32(TDH, 0);
2640         ew32(TDT, 0);
2641         tx_ring->head = E1000_TDH;
2642         tx_ring->tail = E1000_TDT;
2643
2644         /* Set the default values for the Tx Inter Packet Gap timer */
2645         tipg = DEFAULT_82543_TIPG_IPGT_COPPER;          /*  8  */
2646         ipgr1 = DEFAULT_82543_TIPG_IPGR1;               /*  8  */
2647         ipgr2 = DEFAULT_82543_TIPG_IPGR2;               /*  6  */
2648
2649         if (adapter->flags & FLAG_TIPG_MEDIUM_FOR_80003ESLAN)
2650                 ipgr2 = DEFAULT_80003ES2LAN_TIPG_IPGR2; /*  7  */
2651
2652         tipg |= ipgr1 << E1000_TIPG_IPGR1_SHIFT;
2653         tipg |= ipgr2 << E1000_TIPG_IPGR2_SHIFT;
2654         ew32(TIPG, tipg);
2655
2656         /* Set the Tx Interrupt Delay register */
2657         ew32(TIDV, adapter->tx_int_delay);
2658         /* Tx irq moderation */
2659         ew32(TADV, adapter->tx_abs_int_delay);
2660
2661         if (adapter->flags2 & FLAG2_DMA_BURST) {
2662                 u32 txdctl = er32(TXDCTL(0));
2663                 txdctl &= ~(E1000_TXDCTL_PTHRESH | E1000_TXDCTL_HTHRESH |
2664                             E1000_TXDCTL_WTHRESH);
2665                 /*
2666                  * set up some performance related parameters to encourage the
2667                  * hardware to use the bus more efficiently in bursts, depends
2668                  * on the tx_int_delay to be enabled,
2669                  * wthresh = 5 ==> burst write a cacheline (64 bytes) at a time
2670                  * hthresh = 1 ==> prefetch when one or more available
2671                  * pthresh = 0x1f ==> prefetch if internal cache 31 or less
2672                  * BEWARE: this seems to work but should be considered first if
2673                  * there are Tx hangs or other Tx related bugs
2674                  */
2675                 txdctl |= E1000_TXDCTL_DMA_BURST_ENABLE;
2676                 ew32(TXDCTL(0), txdctl);
2677                 /* erratum work around: set txdctl the same for both queues */
2678                 ew32(TXDCTL(1), txdctl);
2679         }
2680
2681         /* Program the Transmit Control Register */
2682         tctl = er32(TCTL);
2683         tctl &= ~E1000_TCTL_CT;
2684         tctl |= E1000_TCTL_PSP | E1000_TCTL_RTLC |
2685                 (E1000_COLLISION_THRESHOLD << E1000_CT_SHIFT);
2686
2687         if (adapter->flags & FLAG_TARC_SPEED_MODE_BIT) {
2688                 tarc = er32(TARC(0));
2689                 /*
2690                  * set the speed mode bit, we'll clear it if we're not at
2691                  * gigabit link later
2692                  */
2693 #define SPEED_MODE_BIT (1 << 21)
2694                 tarc |= SPEED_MODE_BIT;
2695                 ew32(TARC(0), tarc);
2696         }
2697
2698         /* errata: program both queues to unweighted RR */
2699         if (adapter->flags & FLAG_TARC_SET_BIT_ZERO) {
2700                 tarc = er32(TARC(0));
2701                 tarc |= 1;
2702                 ew32(TARC(0), tarc);
2703                 tarc = er32(TARC(1));
2704                 tarc |= 1;
2705                 ew32(TARC(1), tarc);
2706         }
2707
2708         /* Setup Transmit Descriptor Settings for eop descriptor */
2709         adapter->txd_cmd = E1000_TXD_CMD_EOP | E1000_TXD_CMD_IFCS;
2710
2711         /* only set IDE if we are delaying interrupts using the timers */
2712         if (adapter->tx_int_delay)
2713                 adapter->txd_cmd |= E1000_TXD_CMD_IDE;
2714
2715         /* enable Report Status bit */
2716         adapter->txd_cmd |= E1000_TXD_CMD_RS;
2717
2718         ew32(TCTL, tctl);
2719
2720         e1000e_config_collision_dist(hw);
2721 }
2722
2723 /**
2724  * e1000_setup_rctl - configure the receive control registers
2725  * @adapter: Board private structure
2726  **/
2727 #define PAGE_USE_COUNT(S) (((S) >> PAGE_SHIFT) + \
2728                            (((S) & (PAGE_SIZE - 1)) ? 1 : 0))
2729 static void e1000_setup_rctl(struct e1000_adapter *adapter)
2730 {
2731         struct e1000_hw *hw = &adapter->hw;
2732         u32 rctl, rfctl;
2733         u32 pages = 0;
2734
2735         /* Workaround Si errata on 82579 - configure jumbo frame flow */
2736         if (hw->mac.type == e1000_pch2lan) {
2737                 s32 ret_val;
2738
2739                 if (adapter->netdev->mtu > ETH_DATA_LEN)
2740                         ret_val = e1000_lv_jumbo_workaround_ich8lan(hw, true);
2741                 else
2742                         ret_val = e1000_lv_jumbo_workaround_ich8lan(hw, false);
2743
2744                 if (ret_val)
2745                         e_dbg("failed to enable jumbo frame workaround mode\n");
2746         }
2747
2748         /* Program MC offset vector base */
2749         rctl = er32(RCTL);
2750         rctl &= ~(3 << E1000_RCTL_MO_SHIFT);
2751         rctl |= E1000_RCTL_EN | E1000_RCTL_BAM |
2752                 E1000_RCTL_LBM_NO | E1000_RCTL_RDMTS_HALF |
2753                 (adapter->hw.mac.mc_filter_type << E1000_RCTL_MO_SHIFT);
2754
2755         /* Do not Store bad packets */
2756         rctl &= ~E1000_RCTL_SBP;
2757
2758         /* Enable Long Packet receive */
2759         if (adapter->netdev->mtu <= ETH_DATA_LEN)
2760                 rctl &= ~E1000_RCTL_LPE;
2761         else
2762                 rctl |= E1000_RCTL_LPE;
2763
2764         /* Some systems expect that the CRC is included in SMBUS traffic. The
2765          * hardware strips the CRC before sending to both SMBUS (BMC) and to
2766          * host memory when this is enabled
2767          */
2768         if (adapter->flags2 & FLAG2_CRC_STRIPPING)
2769                 rctl |= E1000_RCTL_SECRC;
2770
2771         /* Workaround Si errata on 82577 PHY - configure IPG for jumbos */
2772         if ((hw->phy.type == e1000_phy_82577) && (rctl & E1000_RCTL_LPE)) {
2773                 u16 phy_data;
2774
2775                 e1e_rphy(hw, PHY_REG(770, 26), &phy_data);
2776                 phy_data &= 0xfff8;
2777                 phy_data |= (1 << 2);
2778                 e1e_wphy(hw, PHY_REG(770, 26), phy_data);
2779
2780                 e1e_rphy(hw, 22, &phy_data);
2781                 phy_data &= 0x0fff;
2782                 phy_data |= (1 << 14);
2783                 e1e_wphy(hw, 0x10, 0x2823);
2784                 e1e_wphy(hw, 0x11, 0x0003);
2785                 e1e_wphy(hw, 22, phy_data);
2786         }
2787
2788         /* Setup buffer sizes */
2789         rctl &= ~E1000_RCTL_SZ_4096;
2790         rctl |= E1000_RCTL_BSEX;
2791         switch (adapter->rx_buffer_len) {
2792         case 2048:
2793         default:
2794                 rctl |= E1000_RCTL_SZ_2048;
2795                 rctl &= ~E1000_RCTL_BSEX;
2796                 break;
2797         case 4096:
2798                 rctl |= E1000_RCTL_SZ_4096;
2799                 break;
2800         case 8192:
2801                 rctl |= E1000_RCTL_SZ_8192;
2802                 break;
2803         case 16384:
2804                 rctl |= E1000_RCTL_SZ_16384;
2805                 break;
2806         }
2807
2808         /*
2809          * 82571 and greater support packet-split where the protocol
2810          * header is placed in skb->data and the packet data is
2811          * placed in pages hanging off of skb_shinfo(skb)->nr_frags.
2812          * In the case of a non-split, skb->data is linearly filled,
2813          * followed by the page buffers.  Therefore, skb->data is
2814          * sized to hold the largest protocol header.
2815          *
2816          * allocations using alloc_page take too long for regular MTU
2817          * so only enable packet split for jumbo frames
2818          *
2819          * Using pages when the page size is greater than 16k wastes
2820          * a lot of memory, since we allocate 3 pages at all times
2821          * per packet.
2822          */
2823         pages = PAGE_USE_COUNT(adapter->netdev->mtu);
2824         if (!(adapter->flags & FLAG_HAS_ERT) && (pages <= 3) &&
2825             (PAGE_SIZE <= 16384) && (rctl & E1000_RCTL_LPE))
2826                 adapter->rx_ps_pages = pages;
2827         else
2828                 adapter->rx_ps_pages = 0;
2829
2830         if (adapter->rx_ps_pages) {
2831                 u32 psrctl = 0;
2832
2833                 /* Configure extra packet-split registers */
2834                 rfctl = er32(RFCTL);
2835                 rfctl |= E1000_RFCTL_EXTEN;
2836                 /*
2837                  * disable packet split support for IPv6 extension headers,
2838                  * because some malformed IPv6 headers can hang the Rx
2839                  */
2840                 rfctl |= (E1000_RFCTL_IPV6_EX_DIS |
2841                           E1000_RFCTL_NEW_IPV6_EXT_DIS);
2842
2843                 ew32(RFCTL, rfctl);
2844
2845                 /* Enable Packet split descriptors */
2846                 rctl |= E1000_RCTL_DTYP_PS;
2847
2848                 psrctl |= adapter->rx_ps_bsize0 >>
2849                         E1000_PSRCTL_BSIZE0_SHIFT;
2850
2851                 switch (adapter->rx_ps_pages) {
2852                 case 3:
2853                         psrctl |= PAGE_SIZE <<
2854                                 E1000_PSRCTL_BSIZE3_SHIFT;
2855                 case 2:
2856                         psrctl |= PAGE_SIZE <<
2857                                 E1000_PSRCTL_BSIZE2_SHIFT;
2858                 case 1:
2859                         psrctl |= PAGE_SIZE >>
2860                                 E1000_PSRCTL_BSIZE1_SHIFT;
2861                         break;
2862                 }
2863
2864                 ew32(PSRCTL, psrctl);
2865         }
2866
2867         ew32(RCTL, rctl);
2868         /* just started the receive unit, no need to restart */
2869         adapter->flags &= ~FLAG_RX_RESTART_NOW;
2870 }
2871
2872 /**
2873  * e1000_configure_rx - Configure Receive Unit after Reset
2874  * @adapter: board private structure
2875  *
2876  * Configure the Rx unit of the MAC after a reset.
2877  **/
2878 static void e1000_configure_rx(struct e1000_adapter *adapter)
2879 {
2880         struct e1000_hw *hw = &adapter->hw;
2881         struct e1000_ring *rx_ring = adapter->rx_ring;
2882         u64 rdba;
2883         u32 rdlen, rctl, rxcsum, ctrl_ext;
2884
2885         if (adapter->rx_ps_pages) {
2886                 /* this is a 32 byte descriptor */
2887                 rdlen = rx_ring->count *
2888                     sizeof(union e1000_rx_desc_packet_split);
2889                 adapter->clean_rx = e1000_clean_rx_irq_ps;
2890                 adapter->alloc_rx_buf = e1000_alloc_rx_buffers_ps;
2891         } else if (adapter->netdev->mtu > ETH_FRAME_LEN + ETH_FCS_LEN) {
2892                 rdlen = rx_ring->count * sizeof(struct e1000_rx_desc);
2893                 adapter->clean_rx = e1000_clean_jumbo_rx_irq;
2894                 adapter->alloc_rx_buf = e1000_alloc_jumbo_rx_buffers;
2895         } else {
2896                 rdlen = rx_ring->count * sizeof(struct e1000_rx_desc);
2897                 adapter->clean_rx = e1000_clean_rx_irq;
2898                 adapter->alloc_rx_buf = e1000_alloc_rx_buffers;
2899         }
2900
2901         /* disable receives while setting up the descriptors */
2902         rctl = er32(RCTL);
2903         ew32(RCTL, rctl & ~E1000_RCTL_EN);
2904         e1e_flush();
2905         msleep(10);
2906
2907         if (adapter->flags2 & FLAG2_DMA_BURST) {
2908                 /*
2909                  * set the writeback threshold (only takes effect if the RDTR
2910                  * is set). set GRAN=1 and write back up to 0x4 worth, and
2911                  * enable prefetching of 0x20 Rx descriptors
2912                  * granularity = 01
2913                  * wthresh = 04,
2914                  * hthresh = 04,
2915                  * pthresh = 0x20
2916                  */
2917                 ew32(RXDCTL(0), E1000_RXDCTL_DMA_BURST_ENABLE);
2918                 ew32(RXDCTL(1), E1000_RXDCTL_DMA_BURST_ENABLE);
2919
2920                 /*
2921                  * override the delay timers for enabling bursting, only if
2922                  * the value was not set by the user via module options
2923                  */
2924                 if (adapter->rx_int_delay == DEFAULT_RDTR)
2925                         adapter->rx_int_delay = BURST_RDTR;
2926                 if (adapter->rx_abs_int_delay == DEFAULT_RADV)
2927                         adapter->rx_abs_int_delay = BURST_RADV;
2928         }
2929
2930         /* set the Receive Delay Timer Register */
2931         ew32(RDTR, adapter->rx_int_delay);
2932
2933         /* irq moderation */
2934         ew32(RADV, adapter->rx_abs_int_delay);
2935         if ((adapter->itr_setting != 0) && (adapter->itr != 0))
2936                 ew32(ITR, 1000000000 / (adapter->itr * 256));
2937
2938         ctrl_ext = er32(CTRL_EXT);
2939         /* Auto-Mask interrupts upon ICR access */
2940         ctrl_ext |= E1000_CTRL_EXT_IAME;
2941         ew32(IAM, 0xffffffff);
2942         ew32(CTRL_EXT, ctrl_ext);
2943         e1e_flush();
2944
2945         /*
2946          * Setup the HW Rx Head and Tail Descriptor Pointers and
2947          * the Base and Length of the Rx Descriptor Ring
2948          */
2949         rdba = rx_ring->dma;
2950         ew32(RDBAL, (rdba & DMA_BIT_MASK(32)));
2951         ew32(RDBAH, (rdba >> 32));
2952         ew32(RDLEN, rdlen);
2953         ew32(RDH, 0);
2954         ew32(RDT, 0);
2955         rx_ring->head = E1000_RDH;
2956         rx_ring->tail = E1000_RDT;
2957
2958         /* Enable Receive Checksum Offload for TCP and UDP */
2959         rxcsum = er32(RXCSUM);
2960         if (adapter->flags & FLAG_RX_CSUM_ENABLED) {
2961                 rxcsum |= E1000_RXCSUM_TUOFL;
2962
2963                 /*
2964                  * IPv4 payload checksum for UDP fragments must be
2965                  * used in conjunction with packet-split.
2966                  */
2967                 if (adapter->rx_ps_pages)
2968                         rxcsum |= E1000_RXCSUM_IPPCSE;
2969         } else {
2970                 rxcsum &= ~E1000_RXCSUM_TUOFL;
2971                 /* no need to clear IPPCSE as it defaults to 0 */
2972         }
2973         ew32(RXCSUM, rxcsum);
2974
2975         /*
2976          * Enable early receives on supported devices, only takes effect when
2977          * packet size is equal or larger than the specified value (in 8 byte
2978          * units), e.g. using jumbo frames when setting to E1000_ERT_2048
2979          */
2980         if ((adapter->flags & FLAG_HAS_ERT) ||
2981             (adapter->hw.mac.type == e1000_pch2lan)) {
2982                 if (adapter->netdev->mtu > ETH_DATA_LEN) {
2983                         u32 rxdctl = er32(RXDCTL(0));
2984                         ew32(RXDCTL(0), rxdctl | 0x3);
2985                         if (adapter->flags & FLAG_HAS_ERT)
2986                                 ew32(ERT, E1000_ERT_2048 | (1 << 13));
2987                         /*
2988                          * With jumbo frames and early-receive enabled,
2989                          * excessive C-state transition latencies result in
2990                          * dropped transactions.
2991                          */
2992                         pm_qos_update_request(&adapter->netdev->pm_qos_req, 55);
2993                 } else {
2994                         pm_qos_update_request(&adapter->netdev->pm_qos_req,
2995                                               PM_QOS_DEFAULT_VALUE);
2996                 }
2997         }
2998
2999         /* Enable Receives */
3000         ew32(RCTL, rctl);
3001 }
3002
3003 /**
3004  *  e1000_update_mc_addr_list - Update Multicast addresses
3005  *  @hw: pointer to the HW structure
3006  *  @mc_addr_list: array of multicast addresses to program
3007  *  @mc_addr_count: number of multicast addresses to program
3008  *
3009  *  Updates the Multicast Table Array.
3010  *  The caller must have a packed mc_addr_list of multicast addresses.
3011  **/
3012 static void e1000_update_mc_addr_list(struct e1000_hw *hw, u8 *mc_addr_list,
3013                                       u32 mc_addr_count)
3014 {
3015         hw->mac.ops.update_mc_addr_list(hw, mc_addr_list, mc_addr_count);
3016 }
3017
3018 /**
3019  * e1000_set_multi - Multicast and Promiscuous mode set
3020  * @netdev: network interface device structure
3021  *
3022  * The set_multi entry point is called whenever the multicast address
3023  * list or the network interface flags are updated.  This routine is
3024  * responsible for configuring the hardware for proper multicast,
3025  * promiscuous mode, and all-multi behavior.
3026  **/
3027 static void e1000_set_multi(struct net_device *netdev)
3028 {
3029         struct e1000_adapter *adapter = netdev_priv(netdev);
3030         struct e1000_hw *hw = &adapter->hw;
3031         struct netdev_hw_addr *ha;
3032         u8  *mta_list;
3033         u32 rctl;
3034
3035         /* Check for Promiscuous and All Multicast modes */
3036
3037         rctl = er32(RCTL);
3038
3039         if (netdev->flags & IFF_PROMISC) {
3040                 rctl |= (E1000_RCTL_UPE | E1000_RCTL_MPE);
3041                 rctl &= ~E1000_RCTL_VFE;
3042         } else {
3043                 if (netdev->flags & IFF_ALLMULTI) {
3044                         rctl |= E1000_RCTL_MPE;
3045                         rctl &= ~E1000_RCTL_UPE;
3046                 } else {
3047                         rctl &= ~(E1000_RCTL_UPE | E1000_RCTL_MPE);
3048                 }
3049                 if (adapter->flags & FLAG_HAS_HW_VLAN_FILTER)
3050                         rctl |= E1000_RCTL_VFE;
3051         }
3052
3053         ew32(RCTL, rctl);
3054
3055         if (!netdev_mc_empty(netdev)) {
3056                 int i = 0;
3057
3058                 mta_list = kmalloc(netdev_mc_count(netdev) * 6, GFP_ATOMIC);
3059                 if (!mta_list)
3060                         return;
3061
3062                 /* prepare a packed array of only addresses. */
3063                 netdev_for_each_mc_addr(ha, netdev)
3064                         memcpy(mta_list + (i++ * ETH_ALEN), ha->addr, ETH_ALEN);
3065
3066                 e1000_update_mc_addr_list(hw, mta_list, i);
3067                 kfree(mta_list);
3068         } else {
3069                 /*
3070                  * if we're called from probe, we might not have
3071                  * anything to do here, so clear out the list
3072                  */
3073                 e1000_update_mc_addr_list(hw, NULL, 0);
3074         }
3075 }
3076
3077 /**
3078  * e1000_configure - configure the hardware for Rx and Tx
3079  * @adapter: private board structure
3080  **/
3081 static void e1000_configure(struct e1000_adapter *adapter)
3082 {
3083         e1000_set_multi(adapter->netdev);
3084
3085         e1000_restore_vlan(adapter);
3086         e1000_init_manageability_pt(adapter);
3087
3088         e1000_configure_tx(adapter);
3089         e1000_setup_rctl(adapter);
3090         e1000_configure_rx(adapter);
3091         adapter->alloc_rx_buf(adapter, e1000_desc_unused(adapter->rx_ring));
3092 }
3093
3094 /**
3095  * e1000e_power_up_phy - restore link in case the phy was powered down
3096  * @adapter: address of board private structure
3097  *
3098  * The phy may be powered down to save power and turn off link when the
3099  * driver is unloaded and wake on lan is not enabled (among others)
3100  * *** this routine MUST be followed by a call to e1000e_reset ***
3101  **/
3102 void e1000e_power_up_phy(struct e1000_adapter *adapter)
3103 {
3104         if (adapter->hw.phy.ops.power_up)
3105                 adapter->hw.phy.ops.power_up(&adapter->hw);
3106
3107         adapter->hw.mac.ops.setup_link(&adapter->hw);
3108 }
3109
3110 /**
3111  * e1000_power_down_phy - Power down the PHY
3112  *
3113  * Power down the PHY so no link is implied when interface is down.
3114  * The PHY cannot be powered down if management or WoL is active.
3115  */
3116 static void e1000_power_down_phy(struct e1000_adapter *adapter)
3117 {
3118         /* WoL is enabled */
3119         if (adapter->wol)
3120                 return;
3121
3122         if (adapter->hw.phy.ops.power_down)
3123                 adapter->hw.phy.ops.power_down(&adapter->hw);
3124 }
3125
3126 /**
3127  * e1000e_reset - bring the hardware into a known good state
3128  *
3129  * This function boots the hardware and enables some settings that
3130  * require a configuration cycle of the hardware - those cannot be
3131  * set/changed during runtime. After reset the device needs to be
3132  * properly configured for Rx, Tx etc.
3133  */
3134 void e1000e_reset(struct e1000_adapter *adapter)
3135 {
3136         struct e1000_mac_info *mac = &adapter->hw.mac;
3137         struct e1000_fc_info *fc = &adapter->hw.fc;
3138         struct e1000_hw *hw = &adapter->hw;
3139         u32 tx_space, min_tx_space, min_rx_space;
3140         u32 pba = adapter->pba;
3141         u16 hwm;
3142
3143         /* reset Packet Buffer Allocation to default */
3144         ew32(PBA, pba);
3145
3146         if (adapter->max_frame_size > ETH_FRAME_LEN + ETH_FCS_LEN) {
3147                 /*
3148                  * To maintain wire speed transmits, the Tx FIFO should be
3149                  * large enough to accommodate two full transmit packets,
3150                  * rounded up to the next 1KB and expressed in KB.  Likewise,
3151                  * the Rx FIFO should be large enough to accommodate at least
3152                  * one full receive packet and is similarly rounded up and
3153                  * expressed in KB.
3154                  */
3155                 pba = er32(PBA);
3156                 /* upper 16 bits has Tx packet buffer allocation size in KB */
3157                 tx_space = pba >> 16;
3158                 /* lower 16 bits has Rx packet buffer allocation size in KB */
3159                 pba &= 0xffff;
3160                 /*
3161                  * the Tx fifo also stores 16 bytes of information about the Tx
3162                  * but don't include ethernet FCS because hardware appends it
3163                  */
3164                 min_tx_space = (adapter->max_frame_size +
3165                                 sizeof(struct e1000_tx_desc) -
3166                                 ETH_FCS_LEN) * 2;
3167                 min_tx_space = ALIGN(min_tx_space, 1024);
3168                 min_tx_space >>= 10;
3169                 /* software strips receive CRC, so leave room for it */
3170                 min_rx_space = adapter->max_frame_size;
3171                 min_rx_space = ALIGN(min_rx_space, 1024);
3172                 min_rx_space >>= 10;
3173
3174                 /*
3175                  * If current Tx allocation is less than the min Tx FIFO size,
3176                  * and the min Tx FIFO size is less than the current Rx FIFO
3177                  * allocation, take space away from current Rx allocation
3178                  */
3179                 if ((tx_space < min_tx_space) &&
3180                     ((min_tx_space - tx_space) < pba)) {
3181                         pba -= min_tx_space - tx_space;
3182
3183                         /*
3184                          * if short on Rx space, Rx wins and must trump Tx
3185                          * adjustment or use Early Receive if available
3186                          */
3187                         if ((pba < min_rx_space) &&
3188                             (!(adapter->flags & FLAG_HAS_ERT)))
3189                                 /* ERT enabled in e1000_configure_rx */
3190                                 pba = min_rx_space;
3191                 }
3192
3193                 ew32(PBA, pba);
3194         }
3195
3196         /*
3197          * flow control settings
3198          *
3199          * The high water mark must be low enough to fit one full frame
3200          * (or the size used for early receive) above it in the Rx FIFO.
3201          * Set it to the lower of:
3202          * - 90% of the Rx FIFO size, and
3203          * - the full Rx FIFO size minus the early receive size (for parts
3204          *   with ERT support assuming ERT set to E1000_ERT_2048), or
3205          * - the full Rx FIFO size minus one full frame
3206          */
3207         if (adapter->flags & FLAG_DISABLE_FC_PAUSE_TIME)
3208                 fc->pause_time = 0xFFFF;
3209         else
3210                 fc->pause_time = E1000_FC_PAUSE_TIME;
3211         fc->send_xon = 1;
3212         fc->current_mode = fc->requested_mode;
3213
3214         switch (hw->mac.type) {
3215         default:
3216                 if ((adapter->flags & FLAG_HAS_ERT) &&
3217                     (adapter->netdev->mtu > ETH_DATA_LEN))
3218                         hwm = min(((pba << 10) * 9 / 10),
3219                                   ((pba << 10) - (E1000_ERT_2048 << 3)));
3220                 else
3221                         hwm = min(((pba << 10) * 9 / 10),
3222                                   ((pba << 10) - adapter->max_frame_size));
3223
3224                 fc->high_water = hwm & E1000_FCRTH_RTH; /* 8-byte granularity */
3225                 fc->low_water = fc->high_water - 8;
3226                 break;
3227         case e1000_pchlan:
3228                 /*
3229                  * Workaround PCH LOM adapter hangs with certain network
3230                  * loads.  If hangs persist, try disabling Tx flow control.
3231                  */
3232                 if (adapter->netdev->mtu > ETH_DATA_LEN) {
3233                         fc->high_water = 0x3500;
3234                         fc->low_water  = 0x1500;
3235                 } else {
3236                         fc->high_water = 0x5000;
3237                         fc->low_water  = 0x3000;
3238                 }
3239                 fc->refresh_time = 0x1000;
3240                 break;
3241         case e1000_pch2lan:
3242                 fc->high_water = 0x05C20;
3243                 fc->low_water = 0x05048;
3244                 fc->pause_time = 0x0650;
3245                 fc->refresh_time = 0x0400;
3246                 if (adapter->netdev->mtu > ETH_DATA_LEN) {
3247                         pba = 14;
3248                         ew32(PBA, pba);
3249                 }
3250                 break;
3251         }
3252
3253         /*
3254          * Disable Adaptive Interrupt Moderation if 2 full packets cannot
3255          * fit in receive buffer and early-receive not supported.
3256          */
3257         if (adapter->itr_setting & 0x3) {
3258                 if (((adapter->max_frame_size * 2) > (pba << 10)) &&
3259                     !(adapter->flags & FLAG_HAS_ERT)) {
3260                         if (!(adapter->flags2 & FLAG2_DISABLE_AIM)) {
3261                                 dev_info(&adapter->pdev->dev,
3262                                         "Interrupt Throttle Rate turned off\n");
3263                                 adapter->flags2 |= FLAG2_DISABLE_AIM;
3264                                 ew32(ITR, 0);
3265                         }
3266                 } else if (adapter->flags2 & FLAG2_DISABLE_AIM) {
3267                         dev_info(&adapter->pdev->dev,
3268                                  "Interrupt Throttle Rate turned on\n");
3269                         adapter->flags2 &= ~FLAG2_DISABLE_AIM;
3270                         adapter->itr = 20000;
3271                         ew32(ITR, 1000000000 / (adapter->itr * 256));
3272                 }
3273         }
3274
3275         /* Allow time for pending master requests to run */
3276         mac->ops.reset_hw(hw);
3277
3278         /*
3279          * For parts with AMT enabled, let the firmware know
3280          * that the network interface is in control
3281          */
3282         if (adapter->flags & FLAG_HAS_AMT)
3283                 e1000e_get_hw_control(adapter);
3284
3285         ew32(WUC, 0);
3286
3287         if (mac->ops.init_hw(hw))
3288                 e_err("Hardware Error\n");
3289
3290         e1000_update_mng_vlan(adapter);
3291
3292         /* Enable h/w to recognize an 802.1Q VLAN Ethernet packet */
3293         ew32(VET, ETH_P_8021Q);
3294
3295         e1000e_reset_adaptive(hw);
3296
3297         if (!netif_running(adapter->netdev) &&
3298             !test_bit(__E1000_TESTING, &adapter->state)) {
3299                 e1000_power_down_phy(adapter);
3300                 return;
3301         }
3302
3303         e1000_get_phy_info(hw);
3304
3305         if ((adapter->flags & FLAG_HAS_SMART_POWER_DOWN) &&
3306             !(adapter->flags & FLAG_SMART_POWER_DOWN)) {
3307                 u16 phy_data = 0;
3308                 /*
3309                  * speed up time to link by disabling smart power down, ignore
3310                  * the return value of this function because there is nothing
3311                  * different we would do if it failed
3312                  */
3313                 e1e_rphy(hw, IGP02E1000_PHY_POWER_MGMT, &phy_data);
3314                 phy_data &= ~IGP02E1000_PM_SPD;
3315                 e1e_wphy(hw, IGP02E1000_PHY_POWER_MGMT, phy_data);
3316         }
3317 }
3318
3319 int e1000e_up(struct e1000_adapter *adapter)
3320 {
3321         struct e1000_hw *hw = &adapter->hw;
3322
3323         /* hardware has been reset, we need to reload some things */
3324         e1000_configure(adapter);
3325
3326         clear_bit(__E1000_DOWN, &adapter->state);
3327
3328         napi_enable(&adapter->napi);
3329         if (adapter->msix_entries)
3330                 e1000_configure_msix(adapter);
3331         e1000_irq_enable(adapter);
3332
3333         netif_wake_queue(adapter->netdev);
3334
3335         /* fire a link change interrupt to start the watchdog */
3336         if (adapter->msix_entries)
3337                 ew32(ICS, E1000_ICS_LSC | E1000_ICR_OTHER);
3338         else
3339                 ew32(ICS, E1000_ICS_LSC);
3340
3341         return 0;
3342 }
3343
3344 static void e1000e_flush_descriptors(struct e1000_adapter *adapter)
3345 {
3346         struct e1000_hw *hw = &adapter->hw;
3347
3348         if (!(adapter->flags2 & FLAG2_DMA_BURST))
3349                 return;
3350
3351         /* flush pending descriptor writebacks to memory */
3352         ew32(TIDV, adapter->tx_int_delay | E1000_TIDV_FPD);
3353         ew32(RDTR, adapter->rx_int_delay | E1000_RDTR_FPD);
3354
3355         /* execute the writes immediately */
3356         e1e_flush();
3357 }
3358
3359 static void e1000e_update_stats(struct e1000_adapter *adapter);
3360
3361 void e1000e_down(struct e1000_adapter *adapter)
3362 {
3363         struct net_device *netdev = adapter->netdev;
3364         struct e1000_hw *hw = &adapter->hw;
3365         u32 tctl, rctl;
3366
3367         /*
3368          * signal that we're down so the interrupt handler does not
3369          * reschedule our watchdog timer
3370          */
3371         set_bit(__E1000_DOWN, &adapter->state);
3372
3373         /* disable receives in the hardware */
3374         rctl = er32(RCTL);
3375         ew32(RCTL, rctl & ~E1000_RCTL_EN);
3376         /* flush and sleep below */
3377
3378         netif_stop_queue(netdev);
3379
3380         /* disable transmits in the hardware */
3381         tctl = er32(TCTL);
3382         tctl &= ~E1000_TCTL_EN;
3383         ew32(TCTL, tctl);
3384         /* flush both disables and wait for them to finish */
3385         e1e_flush();
3386         msleep(10);
3387
3388         napi_disable(&adapter->napi);
3389         e1000_irq_disable(adapter);
3390
3391         del_timer_sync(&adapter->watchdog_timer);
3392         del_timer_sync(&adapter->phy_info_timer);
3393
3394         netif_carrier_off(netdev);
3395
3396         spin_lock(&adapter->stats64_lock);
3397         e1000e_update_stats(adapter);
3398         spin_unlock(&adapter->stats64_lock);
3399
3400         adapter->link_speed = 0;
3401         adapter->link_duplex = 0;
3402
3403         if (!pci_channel_offline(adapter->pdev))
3404                 e1000e_reset(adapter);
3405
3406         e1000e_flush_descriptors(adapter);
3407
3408         e1000_clean_tx_ring(adapter);
3409         e1000_clean_rx_ring(adapter);
3410
3411         /*
3412          * TODO: for power management, we could drop the link and
3413          * pci_disable_device here.
3414          */
3415 }
3416
3417 void e1000e_reinit_locked(struct e1000_adapter *adapter)
3418 {
3419         might_sleep();
3420         while (test_and_set_bit(__E1000_RESETTING, &adapter->state))
3421                 msleep(1);
3422         e1000e_down(adapter);
3423         e1000e_up(adapter);
3424         clear_bit(__E1000_RESETTING, &adapter->state);
3425 }
3426
3427 /**
3428  * e1000_sw_init - Initialize general software structures (struct e1000_adapter)
3429  * @adapter: board private structure to initialize
3430  *
3431  * e1000_sw_init initializes the Adapter private data structure.
3432  * Fields are initialized based on PCI device information and
3433  * OS network device settings (MTU size).
3434  **/
3435 static int __devinit e1000_sw_init(struct e1000_adapter *adapter)
3436 {
3437         struct net_device *netdev = adapter->netdev;
3438
3439         adapter->rx_buffer_len = ETH_FRAME_LEN + VLAN_HLEN + ETH_FCS_LEN;
3440         adapter->rx_ps_bsize0 = 128;
3441         adapter->max_frame_size = netdev->mtu + ETH_HLEN + ETH_FCS_LEN;
3442         adapter->min_frame_size = ETH_ZLEN + ETH_FCS_LEN;
3443
3444         spin_lock_init(&adapter->stats64_lock);
3445
3446         e1000e_set_interrupt_capability(adapter);
3447
3448         if (e1000_alloc_queues(adapter))
3449                 return -ENOMEM;
3450
3451         /* Explicitly disable IRQ since the NIC can be in any state. */
3452         e1000_irq_disable(adapter);
3453
3454         set_bit(__E1000_DOWN, &adapter->state);
3455         return 0;
3456 }
3457
3458 /**
3459  * e1000_intr_msi_test - Interrupt Handler
3460  * @irq: interrupt number
3461  * @data: pointer to a network interface device structure
3462  **/
3463 static irqreturn_t e1000_intr_msi_test(int irq, void *data)
3464 {
3465         struct net_device *netdev = data;
3466         struct e1000_adapter *adapter = netdev_priv(netdev);
3467         struct e1000_hw *hw = &adapter->hw;
3468         u32 icr = er32(ICR);
3469
3470         e_dbg("icr is %08X\n", icr);
3471         if (icr & E1000_ICR_RXSEQ) {
3472                 adapter->flags &= ~FLAG_MSI_TEST_FAILED;
3473                 wmb();
3474         }
3475
3476         return IRQ_HANDLED;
3477 }
3478
3479 /**
3480  * e1000_test_msi_interrupt - Returns 0 for successful test
3481  * @adapter: board private struct
3482  *
3483  * code flow taken from tg3.c
3484  **/
3485 static int e1000_test_msi_interrupt(struct e1000_adapter *adapter)
3486 {
3487         struct net_device *netdev = adapter->netdev;
3488         struct e1000_hw *hw = &adapter->hw;
3489         int err;
3490
3491         /* poll_enable hasn't been called yet, so don't need disable */
3492         /* clear any pending events */
3493         er32(ICR);
3494
3495         /* free the real vector and request a test handler */
3496         e1000_free_irq(adapter);
3497         e1000e_reset_interrupt_capability(adapter);
3498
3499         /* Assume that the test fails, if it succeeds then the test
3500          * MSI irq handler will unset this flag */
3501         adapter->flags |= FLAG_MSI_TEST_FAILED;
3502
3503         err = pci_enable_msi(adapter->pdev);
3504         if (err)
3505                 goto msi_test_failed;
3506
3507         err = request_irq(adapter->pdev->irq, e1000_intr_msi_test, 0,
3508                           netdev->name, netdev);
3509         if (err) {
3510                 pci_disable_msi(adapter->pdev);
3511                 goto msi_test_failed;
3512         }
3513
3514         wmb();
3515
3516         e1000_irq_enable(adapter);
3517
3518         /* fire an unusual interrupt on the test handler */
3519         ew32(ICS, E1000_ICS_RXSEQ);
3520         e1e_flush();
3521         msleep(50);
3522
3523         e1000_irq_disable(adapter);
3524
3525         rmb();
3526
3527         if (adapter->flags & FLAG_MSI_TEST_FAILED) {
3528                 adapter->int_mode = E1000E_INT_MODE_LEGACY;
3529                 e_info("MSI interrupt test failed, using legacy interrupt.\n");
3530         } else
3531                 e_dbg("MSI interrupt test succeeded!\n");
3532
3533         free_irq(adapter->pdev->irq, netdev);
3534         pci_disable_msi(adapter->pdev);
3535
3536 msi_test_failed:
3537         e1000e_set_interrupt_capability(adapter);
3538         return e1000_request_irq(adapter);
3539 }
3540
3541 /**
3542  * e1000_test_msi - Returns 0 if MSI test succeeds or INTx mode is restored
3543  * @adapter: board private struct
3544  *
3545  * code flow taken from tg3.c, called with e1000 interrupts disabled.
3546  **/
3547 static int e1000_test_msi(struct e1000_adapter *adapter)
3548 {
3549         int err;
3550         u16 pci_cmd;
3551
3552         if (!(adapter->flags & FLAG_MSI_ENABLED))
3553                 return 0;
3554
3555         /* disable SERR in case the MSI write causes a master abort */
3556         pci_read_config_word(adapter->pdev, PCI_COMMAND, &pci_cmd);
3557         if (pci_cmd & PCI_COMMAND_SERR)
3558                 pci_write_config_word(adapter->pdev, PCI_COMMAND,
3559                                       pci_cmd & ~PCI_COMMAND_SERR);
3560
3561         err = e1000_test_msi_interrupt(adapter);
3562
3563         /* re-enable SERR */
3564         if (pci_cmd & PCI_COMMAND_SERR) {
3565                 pci_read_config_word(adapter->pdev, PCI_COMMAND, &pci_cmd);
3566                 pci_cmd |= PCI_COMMAND_SERR;
3567                 pci_write_config_word(adapter->pdev, PCI_COMMAND, pci_cmd);
3568         }
3569
3570         return err;
3571 }
3572
3573 /**
3574  * e1000_open - Called when a network interface is made active
3575  * @netdev: network interface device structure
3576  *
3577  * Returns 0 on success, negative value on failure
3578  *
3579  * The open entry point is called when a network interface is made
3580  * active by the system (IFF_UP).  At this point all resources needed
3581  * for transmit and receive operations are allocated, the interrupt
3582  * handler is registered with the OS, the watchdog timer is started,
3583  * and the stack is notified that the interface is ready.
3584  **/
3585 static int e1000_open(struct net_device *netdev)
3586 {
3587         struct e1000_adapter *adapter = netdev_priv(netdev);
3588         struct e1000_hw *hw = &adapter->hw;
3589         struct pci_dev *pdev = adapter->pdev;
3590         int err;
3591
3592         /* disallow open during test */
3593         if (test_bit(__E1000_TESTING, &adapter->state))
3594                 return -EBUSY;
3595
3596         pm_runtime_get_sync(&pdev->dev);
3597
3598         netif_carrier_off(netdev);
3599
3600         /* allocate transmit descriptors */
3601         err = e1000e_setup_tx_resources(adapter);
3602         if (err)
3603                 goto err_setup_tx;
3604
3605         /* allocate receive descriptors */
3606         err = e1000e_setup_rx_resources(adapter);
3607         if (err)
3608                 goto err_setup_rx;
3609
3610         /*
3611          * If AMT is enabled, let the firmware know that the network
3612          * interface is now open and reset the part to a known state.
3613          */
3614         if (adapter->flags & FLAG_HAS_AMT) {
3615                 e1000e_get_hw_control(adapter);
3616                 e1000e_reset(adapter);
3617         }
3618
3619         e1000e_power_up_phy(adapter);
3620
3621         adapter->mng_vlan_id = E1000_MNG_VLAN_NONE;
3622         if ((adapter->hw.mng_cookie.status &
3623              E1000_MNG_DHCP_COOKIE_STATUS_VLAN))
3624                 e1000_update_mng_vlan(adapter);
3625
3626         /* DMA latency requirement to workaround early-receive/jumbo issue */
3627         if ((adapter->flags & FLAG_HAS_ERT) ||
3628             (adapter->hw.mac.type == e1000_pch2lan))
3629                 pm_qos_add_request(&adapter->netdev->pm_qos_req,
3630                                    PM_QOS_CPU_DMA_LATENCY,
3631                                    PM_QOS_DEFAULT_VALUE);
3632
3633         /*
3634          * before we allocate an interrupt, we must be ready to handle it.
3635          * Setting DEBUG_SHIRQ in the kernel makes it fire an interrupt
3636          * as soon as we call pci_request_irq, so we have to setup our
3637          * clean_rx handler before we do so.
3638          */
3639         e1000_configure(adapter);
3640
3641         err = e1000_request_irq(adapter);
3642         if (err)
3643                 goto err_req_irq;
3644
3645         /*
3646          * Work around PCIe errata with MSI interrupts causing some chipsets to
3647          * ignore e1000e MSI messages, which means we need to test our MSI
3648          * interrupt now
3649          */
3650         if (adapter->int_mode != E1000E_INT_MODE_LEGACY) {
3651                 err = e1000_test_msi(adapter);
3652                 if (err) {
3653                         e_err("Interrupt allocation failed\n");
3654                         goto err_req_irq;
3655                 }
3656         }
3657
3658         /* From here on the code is the same as e1000e_up() */
3659         clear_bit(__E1000_DOWN, &adapter->state);
3660
3661         napi_enable(&adapter->napi);
3662
3663         e1000_irq_enable(adapter);
3664
3665         netif_start_queue(netdev);
3666
3667         adapter->idle_check = true;
3668         pm_runtime_put(&pdev->dev);
3669
3670         /* fire a link status change interrupt to start the watchdog */
3671         if (adapter->msix_entries)
3672                 ew32(ICS, E1000_ICS_LSC | E1000_ICR_OTHER);
3673         else
3674                 ew32(ICS, E1000_ICS_LSC);
3675
3676         return 0;
3677
3678 err_req_irq:
3679         e1000e_release_hw_control(adapter);
3680         e1000_power_down_phy(adapter);
3681         e1000e_free_rx_resources(adapter);
3682 err_setup_rx:
3683         e1000e_free_tx_resources(adapter);
3684 err_setup_tx:
3685         e1000e_reset(adapter);
3686         pm_runtime_put_sync(&pdev->dev);
3687
3688         return err;
3689 }
3690
3691 /**
3692  * e1000_close - Disables a network interface
3693  * @netdev: network interface device structure
3694  *
3695  * Returns 0, this is not allowed to fail
3696  *
3697  * The close entry point is called when an interface is de-activated
3698  * by the OS.  The hardware is still under the drivers control, but
3699  * needs to be disabled.  A global MAC reset is issued to stop the
3700  * hardware, and all transmit and receive resources are freed.
3701  **/
3702 static int e1000_close(struct net_device *netdev)
3703 {
3704         struct e1000_adapter *adapter = netdev_priv(netdev);
3705         struct pci_dev *pdev = adapter->pdev;
3706
3707         WARN_ON(test_bit(__E1000_RESETTING, &adapter->state));
3708
3709         pm_runtime_get_sync(&pdev->dev);
3710
3711         if (!test_bit(__E1000_DOWN, &adapter->state)) {
3712                 e1000e_down(adapter);
3713                 e1000_free_irq(adapter);
3714         }
3715         e1000_power_down_phy(adapter);
3716
3717         e1000e_free_tx_resources(adapter);
3718         e1000e_free_rx_resources(adapter);
3719
3720         /*
3721          * kill manageability vlan ID if supported, but not if a vlan with
3722          * the same ID is registered on the host OS (let 8021q kill it)
3723          */
3724         if ((adapter->hw.mng_cookie.status &
3725                           E1000_MNG_DHCP_COOKIE_STATUS_VLAN) &&
3726              !(adapter->vlgrp &&
3727                vlan_group_get_device(adapter->vlgrp, adapter->mng_vlan_id)))
3728                 e1000_vlan_rx_kill_vid(netdev, adapter->mng_vlan_id);
3729
3730         /*
3731          * If AMT is enabled, let the firmware know that the network
3732          * interface is now closed
3733          */
3734         if ((adapter->flags & FLAG_HAS_AMT) &&
3735             !test_bit(__E1000_TESTING, &adapter->state))
3736                 e1000e_release_hw_control(adapter);
3737
3738         if ((adapter->flags & FLAG_HAS_ERT) ||
3739             (adapter->hw.mac.type == e1000_pch2lan))
3740                 pm_qos_remove_request(&adapter->netdev->pm_qos_req);
3741
3742         pm_runtime_put_sync(&pdev->dev);
3743
3744         return 0;
3745 }
3746 /**
3747  * e1000_set_mac - Change the Ethernet Address of the NIC
3748  * @netdev: network interface device structure
3749  * @p: pointer to an address structure
3750  *
3751  * Returns 0 on success, negative on failure
3752  **/
3753 static int e1000_set_mac(struct net_device *netdev, void *p)
3754 {
3755         struct e1000_adapter *adapter = netdev_priv(netdev);
3756         struct sockaddr *addr = p;
3757
3758         if (!is_valid_ether_addr(addr->sa_data))
3759                 return -EADDRNOTAVAIL;
3760
3761         memcpy(netdev->dev_addr, addr->sa_data, netdev->addr_len);
3762         memcpy(adapter->hw.mac.addr, addr->sa_data, netdev->addr_len);
3763
3764         e1000e_rar_set(&adapter->hw, adapter->hw.mac.addr, 0);
3765
3766         if (adapter->flags & FLAG_RESET_OVERWRITES_LAA) {
3767                 /* activate the work around */
3768                 e1000e_set_laa_state_82571(&adapter->hw, 1);
3769
3770                 /*
3771                  * Hold a copy of the LAA in RAR[14] This is done so that
3772                  * between the time RAR[0] gets clobbered  and the time it
3773                  * gets fixed (in e1000_watchdog), the actual LAA is in one
3774                  * of the RARs and no incoming packets directed to this port
3775                  * are dropped. Eventually the LAA will be in RAR[0] and
3776                  * RAR[14]
3777                  */
3778                 e1000e_rar_set(&adapter->hw,
3779                               adapter->hw.mac.addr,
3780                               adapter->hw.mac.rar_entry_count - 1);
3781         }
3782
3783         return 0;
3784 }
3785
3786 /**
3787  * e1000e_update_phy_task - work thread to update phy
3788  * @work: pointer to our work struct
3789  *
3790  * this worker thread exists because we must acquire a
3791  * semaphore to read the phy, which we could msleep while
3792  * waiting for it, and we can't msleep in a timer.
3793  **/
3794 static void e1000e_update_phy_task(struct work_struct *work)
3795 {
3796         struct e1000_adapter *adapter = container_of(work,
3797                                         struct e1000_adapter, update_phy_task);
3798
3799         if (test_bit(__E1000_DOWN, &adapter->state))
3800                 return;
3801
3802         e1000_get_phy_info(&adapter->hw);
3803 }
3804
3805 /*
3806  * Need to wait a few seconds after link up to get diagnostic information from
3807  * the phy
3808  */
3809 static void e1000_update_phy_info(unsigned long data)
3810 {
3811         struct e1000_adapter *adapter = (struct e1000_adapter *) data;
3812
3813         if (test_bit(__E1000_DOWN, &adapter->state))
3814                 return;
3815
3816         schedule_work(&adapter->update_phy_task);
3817 }
3818
3819 /**
3820  * e1000e_update_phy_stats - Update the PHY statistics counters
3821  * @adapter: board private structure
3822  **/
3823 static void e1000e_update_phy_stats(struct e1000_adapter *adapter)
3824 {
3825         struct e1000_hw *hw = &adapter->hw;
3826         s32 ret_val;
3827         u16 phy_data;
3828
3829         ret_val = hw->phy.ops.acquire(hw);
3830         if (ret_val)
3831                 return;
3832
3833         hw->phy.addr = 1;
3834
3835 #define HV_PHY_STATS_PAGE       778
3836         /*
3837          * A page set is expensive so check if already on desired page.
3838          * If not, set to the page with the PHY status registers.
3839          */
3840         ret_val = e1000e_read_phy_reg_mdic(hw, IGP01E1000_PHY_PAGE_SELECT,
3841                                            &phy_data);
3842         if (ret_val)
3843                 goto release;
3844         if (phy_data != (HV_PHY_STATS_PAGE << IGP_PAGE_SHIFT)) {
3845                 ret_val = e1000e_write_phy_reg_mdic(hw,
3846                                                     IGP01E1000_PHY_PAGE_SELECT,
3847                                                     (HV_PHY_STATS_PAGE <<
3848                                                      IGP_PAGE_SHIFT));
3849                 if (ret_val)
3850                         goto release;
3851         }
3852
3853         /* Read/clear the upper 16-bit registers and read/accumulate lower */
3854
3855         /* Single Collision Count */
3856         e1000e_read_phy_reg_mdic(hw, HV_SCC_UPPER & MAX_PHY_REG_ADDRESS,
3857                                  &phy_data);
3858         ret_val = e1000e_read_phy_reg_mdic(hw,
3859                                            HV_SCC_LOWER & MAX_PHY_REG_ADDRESS,
3860                                            &phy_data);
3861         if (!ret_val)
3862                 adapter->stats.scc += phy_data;
3863
3864         /* Excessive Collision Count */
3865         e1000e_read_phy_reg_mdic(hw, HV_ECOL_UPPER & MAX_PHY_REG_ADDRESS,
3866                                  &phy_data);
3867         ret_val = e1000e_read_phy_reg_mdic(hw,
3868                                            HV_ECOL_LOWER & MAX_PHY_REG_ADDRESS,
3869                                            &phy_data);
3870         if (!ret_val)
3871                 adapter->stats.ecol += phy_data;
3872
3873         /* Multiple Collision Count */
3874         e1000e_read_phy_reg_mdic(hw, HV_MCC_UPPER & MAX_PHY_REG_ADDRESS,
3875                                  &phy_data);
3876         ret_val = e1000e_read_phy_reg_mdic(hw,
3877                                            HV_MCC_LOWER & MAX_PHY_REG_ADDRESS,
3878                                            &phy_data);
3879         if (!ret_val)
3880                 adapter->stats.mcc += phy_data;
3881
3882         /* Late Collision Count */
3883         e1000e_read_phy_reg_mdic(hw, HV_LATECOL_UPPER & MAX_PHY_REG_ADDRESS,
3884                                  &phy_data);
3885         ret_val = e1000e_read_phy_reg_mdic(hw,
3886                                            HV_LATECOL_LOWER &
3887                                            MAX_PHY_REG_ADDRESS,
3888                                            &phy_data);
3889         if (!ret_val)
3890                 adapter->stats.latecol += phy_data;
3891
3892         /* Collision Count - also used for adaptive IFS */
3893         e1000e_read_phy_reg_mdic(hw, HV_COLC_UPPER & MAX_PHY_REG_ADDRESS,
3894                                  &phy_data);
3895         ret_val = e1000e_read_phy_reg_mdic(hw,
3896                                            HV_COLC_LOWER & MAX_PHY_REG_ADDRESS,
3897                                            &phy_data);
3898         if (!ret_val)
3899                 hw->mac.collision_delta = phy_data;
3900
3901         /* Defer Count */
3902         e1000e_read_phy_reg_mdic(hw, HV_DC_UPPER & MAX_PHY_REG_ADDRESS,
3903                                  &phy_data);
3904         ret_val = e1000e_read_phy_reg_mdic(hw,
3905                                            HV_DC_LOWER & MAX_PHY_REG_ADDRESS,
3906                                            &phy_data);
3907         if (!ret_val)
3908                 adapter->stats.dc += phy_data;
3909
3910         /* Transmit with no CRS */
3911         e1000e_read_phy_reg_mdic(hw, HV_TNCRS_UPPER & MAX_PHY_REG_ADDRESS,
3912                                  &phy_data);
3913         ret_val = e1000e_read_phy_reg_mdic(hw,
3914                                            HV_TNCRS_LOWER & MAX_PHY_REG_ADDRESS,
3915                                            &phy_data);
3916         if (!ret_val)
3917                 adapter->stats.tncrs += phy_data;
3918
3919 release:
3920         hw->phy.ops.release(hw);
3921 }
3922
3923 /**
3924  * e1000e_update_stats - Update the board statistics counters
3925  * @adapter: board private structure
3926  **/
3927 static void e1000e_update_stats(struct e1000_adapter *adapter)
3928 {
3929         struct net_device *netdev = adapter->netdev;
3930         struct e1000_hw *hw = &adapter->hw;
3931         struct pci_dev *pdev = adapter->pdev;
3932
3933         /*
3934          * Prevent stats update while adapter is being reset, or if the pci
3935          * connection is down.
3936          */
3937         if (adapter->link_speed == 0)
3938                 return;
3939         if (pci_channel_offline(pdev))
3940                 return;
3941
3942         adapter->stats.crcerrs += er32(CRCERRS);
3943         adapter->stats.gprc += er32(GPRC);
3944         adapter->stats.gorc += er32(GORCL);
3945         er32(GORCH); /* Clear gorc */
3946         adapter->stats.bprc += er32(BPRC);
3947         adapter->stats.mprc += er32(MPRC);
3948         adapter->stats.roc += er32(ROC);
3949
3950         adapter->stats.mpc += er32(MPC);
3951
3952         /* Half-duplex statistics */
3953         if (adapter->link_duplex == HALF_DUPLEX) {
3954                 if (adapter->flags2 & FLAG2_HAS_PHY_STATS) {
3955                         e1000e_update_phy_stats(adapter);
3956                 } else {
3957                         adapter->stats.scc += er32(SCC);
3958                         adapter->stats.ecol += er32(ECOL);
3959                         adapter->stats.mcc += er32(MCC);
3960                         adapter->stats.latecol += er32(LATECOL);
3961                         adapter->stats.dc += er32(DC);
3962
3963                         hw->mac.collision_delta = er32(COLC);
3964
3965                         if ((hw->mac.type != e1000_82574) &&
3966                             (hw->mac.type != e1000_82583))
3967                                 adapter->stats.tncrs += er32(TNCRS);
3968                 }
3969                 adapter->stats.colc += hw->mac.collision_delta;
3970         }
3971
3972         adapter->stats.xonrxc += er32(XONRXC);
3973         adapter->stats.xontxc += er32(XONTXC);
3974         adapter->stats.xoffrxc += er32(XOFFRXC);
3975         adapter->stats.xofftxc += er32(XOFFTXC);
3976         adapter->stats.gptc += er32(GPTC);
3977         adapter->stats.gotc += er32(GOTCL);
3978         er32(GOTCH); /* Clear gotc */
3979         adapter->stats.rnbc += er32(RNBC);
3980         adapter->stats.ruc += er32(RUC);
3981
3982         adapter->stats.mptc += er32(MPTC);
3983         adapter->stats.bptc += er32(BPTC);
3984
3985         /* used for adaptive IFS */
3986
3987         hw->mac.tx_packet_delta = er32(TPT);
3988         adapter->stats.tpt += hw->mac.tx_packet_delta;
3989
3990         adapter->stats.algnerrc += er32(ALGNERRC);
3991         adapter->stats.rxerrc += er32(RXERRC);
3992         adapter->stats.cexterr += er32(CEXTERR);
3993         adapter->stats.tsctc += er32(TSCTC);
3994         adapter->stats.tsctfc += er32(TSCTFC);
3995
3996         /* Fill out the OS statistics structure */
3997         netdev->stats.multicast = adapter->stats.mprc;
3998         netdev->stats.collisions = adapter->stats.colc;
3999
4000         /* Rx Errors */
4001
4002         /*
4003          * RLEC on some newer hardware can be incorrect so build
4004          * our own version based on RUC and ROC
4005          */
4006         netdev->stats.rx_errors = adapter->stats.rxerrc +
4007                 adapter->stats.crcerrs + adapter->stats.algnerrc +
4008                 adapter->stats.ruc + adapter->stats.roc +
4009                 adapter->stats.cexterr;
4010         netdev->stats.rx_length_errors = adapter->stats.ruc +
4011                                               adapter->stats.roc;
4012         netdev->stats.rx_crc_errors = adapter->stats.crcerrs;
4013         netdev->stats.rx_frame_errors = adapter->stats.algnerrc;
4014         netdev->stats.rx_missed_errors = adapter->stats.mpc;
4015
4016         /* Tx Errors */
4017         netdev->stats.tx_errors = adapter->stats.ecol +
4018                                        adapter->stats.latecol;
4019         netdev->stats.tx_aborted_errors = adapter->stats.ecol;
4020         netdev->stats.tx_window_errors = adapter->stats.latecol;
4021         netdev->stats.tx_carrier_errors = adapter->stats.tncrs;
4022
4023         /* Tx Dropped needs to be maintained elsewhere */
4024
4025         /* Management Stats */
4026         adapter->stats.mgptc += er32(MGTPTC);
4027         adapter->stats.mgprc += er32(MGTPRC);
4028         adapter->stats.mgpdc += er32(MGTPDC);
4029 }
4030
4031 /**
4032  * e1000_phy_read_status - Update the PHY register status snapshot
4033  * @adapter: board private structure
4034  **/
4035 static void e1000_phy_read_status(struct e1000_adapter *adapter)
4036 {
4037         struct e1000_hw *hw = &adapter->hw;
4038         struct e1000_phy_regs *phy = &adapter->phy_regs;
4039
4040         if ((er32(STATUS) & E1000_STATUS_LU) &&
4041             (adapter->hw.phy.media_type == e1000_media_type_copper)) {
4042                 int ret_val;
4043
4044                 ret_val  = e1e_rphy(hw, PHY_CONTROL, &phy->bmcr);
4045                 ret_val |= e1e_rphy(hw, PHY_STATUS, &phy->bmsr);
4046                 ret_val |= e1e_rphy(hw, PHY_AUTONEG_ADV, &phy->advertise);
4047                 ret_val |= e1e_rphy(hw, PHY_LP_ABILITY, &phy->lpa);
4048                 ret_val |= e1e_rphy(hw, PHY_AUTONEG_EXP, &phy->expansion);
4049                 ret_val |= e1e_rphy(hw, PHY_1000T_CTRL, &phy->ctrl1000);
4050                 ret_val |= e1e_rphy(hw, PHY_1000T_STATUS, &phy->stat1000);
4051                 ret_val |= e1e_rphy(hw, PHY_EXT_STATUS, &phy->estatus);
4052                 if (ret_val)
4053                         e_warn("Error reading PHY register\n");
4054         } else {
4055                 /*
4056                  * Do not read PHY registers if link is not up
4057                  * Set values to typical power-on defaults
4058                  */
4059                 phy->bmcr = (BMCR_SPEED1000 | BMCR_ANENABLE | BMCR_FULLDPLX);
4060                 phy->bmsr = (BMSR_100FULL | BMSR_100HALF | BMSR_10FULL |
4061                              BMSR_10HALF | BMSR_ESTATEN | BMSR_ANEGCAPABLE |
4062                              BMSR_ERCAP);
4063                 phy->advertise = (ADVERTISE_PAUSE_ASYM | ADVERTISE_PAUSE_CAP |
4064                                   ADVERTISE_ALL | ADVERTISE_CSMA);
4065                 phy->lpa = 0;
4066                 phy->expansion = EXPANSION_ENABLENPAGE;
4067                 phy->ctrl1000 = ADVERTISE_1000FULL;
4068                 phy->stat1000 = 0;
4069                 phy->estatus = (ESTATUS_1000_TFULL | ESTATUS_1000_THALF);
4070         }
4071 }
4072
4073 static void e1000_print_link_info(struct e1000_adapter *adapter)
4074 {
4075         struct e1000_hw *hw = &adapter->hw;
4076         u32 ctrl = er32(CTRL);
4077
4078         /* Link status message must follow this format for user tools */
4079         printk(KERN_INFO "e1000e: %s NIC Link is Up %d Mbps %s, "
4080                "Flow Control: %s\n",
4081                adapter->netdev->name,
4082                adapter->link_speed,
4083                (adapter->link_duplex == FULL_DUPLEX) ?
4084                "Full Duplex" : "Half Duplex",
4085                ((ctrl & E1000_CTRL_TFCE) && (ctrl & E1000_CTRL_RFCE)) ?
4086                "Rx/Tx" :
4087                ((ctrl & E1000_CTRL_RFCE) ? "Rx" :
4088                 ((ctrl & E1000_CTRL_TFCE) ? "Tx" : "None")));
4089 }
4090
4091 static bool e1000e_has_link(struct e1000_adapter *adapter)
4092 {
4093         struct e1000_hw *hw = &adapter->hw;
4094         bool link_active = 0;
4095         s32 ret_val = 0;
4096
4097         /*
4098          * get_link_status is set on LSC (link status) interrupt or
4099          * Rx sequence error interrupt.  get_link_status will stay
4100          * false until the check_for_link establishes link
4101          * for copper adapters ONLY
4102          */
4103         switch (hw->phy.media_type) {
4104         case e1000_media_type_copper:
4105                 if (hw->mac.get_link_status) {
4106                         ret_val = hw->mac.ops.check_for_link(hw);
4107                         link_active = !hw->mac.get_link_status;
4108                 } else {
4109                         link_active = 1;
4110                 }
4111                 break;
4112         case e1000_media_type_fiber:
4113                 ret_val = hw->mac.ops.check_for_link(hw);
4114                 link_active = !!(er32(STATUS) & E1000_STATUS_LU);
4115                 break;
4116         case e1000_media_type_internal_serdes:
4117                 ret_val = hw->mac.ops.check_for_link(hw);
4118                 link_active = adapter->hw.mac.serdes_has_link;
4119                 break;
4120         default:
4121         case e1000_media_type_unknown:
4122                 break;
4123         }
4124
4125         if ((ret_val == E1000_ERR_PHY) && (hw->phy.type == e1000_phy_igp_3) &&
4126             (er32(CTRL) & E1000_PHY_CTRL_GBE_DISABLE)) {
4127                 /* See e1000_kmrn_lock_loss_workaround_ich8lan() */
4128                 e_info("Gigabit has been disabled, downgrading speed\n");
4129         }
4130
4131         return link_active;
4132 }
4133
4134 static void e1000e_enable_receives(struct e1000_adapter *adapter)
4135 {
4136         /* make sure the receive unit is started */
4137         if ((adapter->flags & FLAG_RX_NEEDS_RESTART) &&
4138             (adapter->flags & FLAG_RX_RESTART_NOW)) {
4139                 struct e1000_hw *hw = &adapter->hw;
4140                 u32 rctl = er32(RCTL);
4141                 ew32(RCTL, rctl | E1000_RCTL_EN);
4142                 adapter->flags &= ~FLAG_RX_RESTART_NOW;
4143         }
4144 }
4145
4146 static void e1000e_check_82574_phy_workaround(struct e1000_adapter *adapter)
4147 {
4148         struct e1000_hw *hw = &adapter->hw;
4149
4150         /*
4151          * With 82574 controllers, PHY needs to be checked periodically
4152          * for hung state and reset, if two calls return true
4153          */
4154         if (e1000_check_phy_82574(hw))
4155                 adapter->phy_hang_count++;
4156         else
4157                 adapter->phy_hang_count = 0;
4158
4159         if (adapter->phy_hang_count > 1) {
4160                 adapter->phy_hang_count = 0;
4161                 schedule_work(&adapter->reset_task);
4162         }
4163 }
4164
4165 /**
4166  * e1000_watchdog - Timer Call-back
4167  * @data: pointer to adapter cast into an unsigned long
4168  **/
4169 static void e1000_watchdog(unsigned long data)
4170 {
4171         struct e1000_adapter *adapter = (struct e1000_adapter *) data;
4172
4173         /* Do the rest outside of interrupt context */
4174         schedule_work(&adapter->watchdog_task);
4175
4176         /* TODO: make this use queue_delayed_work() */
4177 }
4178
4179 static void e1000_watchdog_task(struct work_struct *work)
4180 {
4181         struct e1000_adapter *adapter = container_of(work,
4182                                         struct e1000_adapter, watchdog_task);
4183         struct net_device *netdev = adapter->netdev;
4184         struct e1000_mac_info *mac = &adapter->hw.mac;
4185         struct e1000_phy_info *phy = &adapter->hw.phy;
4186         struct e1000_ring *tx_ring = adapter->tx_ring;
4187         struct e1000_hw *hw = &adapter->hw;
4188         u32 link, tctl;
4189
4190         if (test_bit(__E1000_DOWN, &adapter->state))
4191                 return;
4192
4193         link = e1000e_has_link(adapter);
4194         if ((netif_carrier_ok(netdev)) && link) {
4195                 /* Cancel scheduled suspend requests. */
4196                 pm_runtime_resume(netdev->dev.parent);
4197
4198                 e1000e_enable_receives(adapter);
4199                 goto link_up;
4200         }
4201
4202         if ((e1000e_enable_tx_pkt_filtering(hw)) &&
4203             (adapter->mng_vlan_id != adapter->hw.mng_cookie.vlan_id))
4204                 e1000_update_mng_vlan(adapter);
4205
4206         if (link) {
4207                 if (!netif_carrier_ok(netdev)) {
4208                         bool txb2b = 1;
4209
4210                         /* Cancel scheduled suspend requests. */
4211                         pm_runtime_resume(netdev->dev.parent);
4212
4213                         /* update snapshot of PHY registers on LSC */
4214                         e1000_phy_read_status(adapter);
4215                         mac->ops.get_link_up_info(&adapter->hw,
4216                                                    &adapter->link_speed,
4217                                                    &adapter->link_duplex);
4218                         e1000_print_link_info(adapter);
4219                         /*
4220                          * On supported PHYs, check for duplex mismatch only
4221                          * if link has autonegotiated at 10/100 half
4222                          */
4223                         if ((hw->phy.type == e1000_phy_igp_3 ||
4224                              hw->phy.type == e1000_phy_bm) &&
4225                             (hw->mac.autoneg == true) &&
4226                             (adapter->link_speed == SPEED_10 ||
4227                              adapter->link_speed == SPEED_100) &&
4228                             (adapter->link_duplex == HALF_DUPLEX)) {
4229                                 u16 autoneg_exp;
4230
4231                                 e1e_rphy(hw, PHY_AUTONEG_EXP, &autoneg_exp);
4232
4233                                 if (!(autoneg_exp & NWAY_ER_LP_NWAY_CAPS))
4234                                         e_info("Autonegotiated half duplex but"
4235                                                " link partner cannot autoneg. "
4236                                                " Try forcing full duplex if "
4237                                                "link gets many collisions.\n");
4238                         }
4239
4240                         /* adjust timeout factor according to speed/duplex */
4241                         adapter->tx_timeout_factor = 1;
4242                         switch (adapter->link_speed) {
4243                         case SPEED_10:
4244                                 txb2b = 0;
4245                                 adapter->tx_timeout_factor = 16;
4246                                 break;
4247                         case SPEED_100:
4248                                 txb2b = 0;
4249                                 adapter->tx_timeout_factor = 10;
4250                                 break;
4251                         }
4252
4253                         /*
4254                          * workaround: re-program speed mode bit after
4255                          * link-up event
4256                          */
4257                         if ((adapter->flags & FLAG_TARC_SPEED_MODE_BIT) &&
4258                             !txb2b) {
4259                                 u32 tarc0;
4260                                 tarc0 = er32(TARC(0));
4261                                 tarc0 &= ~SPEED_MODE_BIT;
4262                                 ew32(TARC(0), tarc0);
4263                         }
4264
4265                         /*
4266                          * disable TSO for pcie and 10/100 speeds, to avoid
4267                          * some hardware issues
4268                          */
4269                         if (!(adapter->flags & FLAG_TSO_FORCE)) {
4270                                 switch (adapter->link_speed) {
4271                                 case SPEED_10:
4272                                 case SPEED_100:
4273                                         e_info("10/100 speed: disabling TSO\n");
4274                                         netdev->features &= ~NETIF_F_TSO;
4275                                         netdev->features &= ~NETIF_F_TSO6;
4276                                         break;
4277                                 case SPEED_1000:
4278                                         netdev->features |= NETIF_F_TSO;
4279                                         netdev->features |= NETIF_F_TSO6;
4280                                         break;
4281                                 default:
4282                                         /* oops */
4283                                         break;
4284                                 }
4285                         }
4286
4287                         /*
4288                          * enable transmits in the hardware, need to do this
4289                          * after setting TARC(0)
4290                          */
4291                         tctl = er32(TCTL);
4292                         tctl |= E1000_TCTL_EN;
4293                         ew32(TCTL, tctl);
4294
4295                         /*
4296                          * Perform any post-link-up configuration before
4297                          * reporting link up.
4298                          */
4299                         if (phy->ops.cfg_on_link_up)
4300                                 phy->ops.cfg_on_link_up(hw);
4301
4302                         netif_carrier_on(netdev);
4303
4304                         if (!test_bit(__E1000_DOWN, &adapter->state))
4305                                 mod_timer(&adapter->phy_info_timer,
4306                                           round_jiffies(jiffies + 2 * HZ));
4307                 }
4308         } else {
4309                 if (netif_carrier_ok(netdev)) {
4310                         adapter->link_speed = 0;
4311                         adapter->link_duplex = 0;
4312                         /* Link status message must follow this format */
4313                         printk(KERN_INFO "e1000e: %s NIC Link is Down\n",
4314                                adapter->netdev->name);
4315                         netif_carrier_off(netdev);
4316                         if (!test_bit(__E1000_DOWN, &adapter->state))
4317                                 mod_timer(&adapter->phy_info_timer,
4318                                           round_jiffies(jiffies + 2 * HZ));
4319
4320                         if (adapter->flags & FLAG_RX_NEEDS_RESTART)
4321                                 schedule_work(&adapter->reset_task);
4322                         else
4323                                 pm_schedule_suspend(netdev->dev.parent,
4324                                                         LINK_TIMEOUT);
4325                 }
4326         }
4327
4328 link_up:
4329         spin_lock(&adapter->stats64_lock);
4330         e1000e_update_stats(adapter);
4331         spin_unlock(&adapter->stats64_lock);
4332
4333         mac->tx_packet_delta = adapter->stats.tpt - adapter->tpt_old;
4334         adapter->tpt_old = adapter->stats.tpt;
4335         mac->collision_delta = adapter->stats.colc - adapter->colc_old;
4336         adapter->colc_old = adapter->stats.colc;
4337
4338         adapter->gorc = adapter->stats.gorc - adapter->gorc_old;
4339         adapter->gorc_old = adapter->stats.gorc;
4340         adapter->gotc = adapter->stats.gotc - adapter->gotc_old;
4341         adapter->gotc_old = adapter->stats.gotc;
4342
4343         e1000e_update_adaptive(&adapter->hw);
4344
4345         if (!netif_carrier_ok(netdev) &&
4346             (e1000_desc_unused(tx_ring) + 1 < tx_ring->count)) {
4347                 /*
4348                  * We've lost link, so the controller stops DMA,
4349                  * but we've got queued Tx work that's never going
4350                  * to get done, so reset controller to flush Tx.
4351                  * (Do the reset outside of interrupt context).
4352                  */
4353                 schedule_work(&adapter->reset_task);
4354                 /* return immediately since reset is imminent */
4355                 return;
4356         }
4357
4358         /* Simple mode for Interrupt Throttle Rate (ITR) */
4359         if (adapter->itr_setting == 4) {
4360                 /*
4361                  * Symmetric Tx/Rx gets a reduced ITR=2000;
4362                  * Total asymmetrical Tx or Rx gets ITR=8000;
4363                  * everyone else is between 2000-8000.
4364                  */
4365                 u32 goc = (adapter->gotc + adapter->gorc) / 10000;
4366                 u32 dif = (adapter->gotc > adapter->gorc ?
4367                             adapter->gotc - adapter->gorc :
4368                             adapter->gorc - adapter->gotc) / 10000;
4369                 u32 itr = goc > 0 ? (dif * 6000 / goc + 2000) : 8000;
4370
4371                 ew32(ITR, 1000000000 / (itr * 256));
4372         }
4373
4374         /* Cause software interrupt to ensure Rx ring is cleaned */
4375         if (adapter->msix_entries)
4376                 ew32(ICS, adapter->rx_ring->ims_val);
4377         else
4378                 ew32(ICS, E1000_ICS_RXDMT0);
4379
4380         /* flush pending descriptors to memory before detecting Tx hang */
4381         e1000e_flush_descriptors(adapter);
4382
4383         /* Force detection of hung controller every watchdog period */
4384         adapter->detect_tx_hung = 1;
4385
4386         /*
4387          * With 82571 controllers, LAA may be overwritten due to controller
4388          * reset from the other port. Set the appropriate LAA in RAR[0]
4389          */
4390         if (e1000e_get_laa_state_82571(hw))
4391                 e1000e_rar_set(hw, adapter->hw.mac.addr, 0);
4392
4393         if (adapter->flags2 & FLAG2_CHECK_PHY_HANG)
4394                 e1000e_check_82574_phy_workaround(adapter);
4395
4396         /* Reset the timer */
4397         if (!test_bit(__E1000_DOWN, &adapter->state))
4398                 mod_timer(&adapter->watchdog_timer,
4399                           round_jiffies(jiffies + 2 * HZ));
4400 }
4401
4402 #define E1000_TX_FLAGS_CSUM             0x00000001
4403 #define E1000_TX_FLAGS_VLAN             0x00000002
4404 #define E1000_TX_FLAGS_TSO              0x00000004
4405 #define E1000_TX_FLAGS_IPV4             0x00000008
4406 #define E1000_TX_FLAGS_VLAN_MASK        0xffff0000
4407 #define E1000_TX_FLAGS_VLAN_SHIFT       16
4408
4409 static int e1000_tso(struct e1000_adapter *adapter,
4410                      struct sk_buff *skb)
4411 {
4412         struct e1000_ring *tx_ring = adapter->tx_ring;
4413         struct e1000_context_desc *context_desc;
4414         struct e1000_buffer *buffer_info;
4415         unsigned int i;
4416         u32 cmd_length = 0;
4417         u16 ipcse = 0, tucse, mss;
4418         u8 ipcss, ipcso, tucss, tucso, hdr_len;
4419
4420         if (!skb_is_gso(skb))
4421                 return 0;
4422
4423         if (skb_header_cloned(skb)) {
4424                 int err = pskb_expand_head(skb, 0, 0, GFP_ATOMIC);
4425
4426                 if (err)
4427                         return err;
4428         }
4429
4430         hdr_len = skb_transport_offset(skb) + tcp_hdrlen(skb);
4431         mss = skb_shinfo(skb)->gso_size;
4432         if (skb->protocol == htons(ETH_P_IP)) {
4433                 struct iphdr *iph = ip_hdr(skb);
4434                 iph->tot_len = 0;
4435                 iph->check = 0;
4436                 tcp_hdr(skb)->check = ~csum_tcpudp_magic(iph->saddr, iph->daddr,
4437                                                          0, IPPROTO_TCP, 0);
4438                 cmd_length = E1000_TXD_CMD_IP;
4439                 ipcse = skb_transport_offset(skb) - 1;
4440         } else if (skb_is_gso_v6(skb)) {
4441                 ipv6_hdr(skb)->payload_len = 0;
4442                 tcp_hdr(skb)->check = ~csum_ipv6_magic(&ipv6_hdr(skb)->saddr,
4443                                                        &ipv6_hdr(skb)->daddr,
4444                                                        0, IPPROTO_TCP, 0);
4445                 ipcse = 0;
4446         }
4447         ipcss = skb_network_offset(skb);
4448         ipcso = (void *)&(ip_hdr(skb)->check) - (void *)skb->data;
4449         tucss = skb_transport_offset(skb);
4450         tucso = (void *)&(tcp_hdr(skb)->check) - (void *)skb->data;
4451         tucse = 0;
4452
4453         cmd_length |= (E1000_TXD_CMD_DEXT | E1000_TXD_CMD_TSE |
4454                        E1000_TXD_CMD_TCP | (skb->len - (hdr_len)));
4455
4456         i = tx_ring->next_to_use;
4457         context_desc = E1000_CONTEXT_DESC(*tx_ring, i);
4458         buffer_info = &tx_ring->buffer_info[i];
4459
4460         context_desc->lower_setup.ip_fields.ipcss  = ipcss;
4461         context_desc->lower_setup.ip_fields.ipcso  = ipcso;
4462         context_desc->lower_setup.ip_fields.ipcse  = cpu_to_le16(ipcse);
4463         context_desc->upper_setup.tcp_fields.tucss = tucss;
4464         context_desc->upper_setup.tcp_fields.tucso = tucso;
4465         context_desc->upper_setup.tcp_fields.tucse = cpu_to_le16(tucse);
4466         context_desc->tcp_seg_setup.fields.mss     = cpu_to_le16(mss);
4467         context_desc->tcp_seg_setup.fields.hdr_len = hdr_len;
4468         context_desc->cmd_and_length = cpu_to_le32(cmd_length);
4469
4470         buffer_info->time_stamp = jiffies;
4471         buffer_info->next_to_watch = i;
4472
4473         i++;
4474         if (i == tx_ring->count)
4475                 i = 0;
4476         tx_ring->next_to_use = i;
4477
4478         return 1;
4479 }
4480
4481 static bool e1000_tx_csum(struct e1000_adapter *adapter, struct sk_buff *skb)
4482 {
4483         struct e1000_ring *tx_ring = adapter->tx_ring;
4484         struct e1000_context_desc *context_desc;
4485         struct e1000_buffer *buffer_info;
4486         unsigned int i;
4487         u8 css;
4488         u32 cmd_len = E1000_TXD_CMD_DEXT;
4489         __be16 protocol;
4490
4491         if (skb->ip_summed != CHECKSUM_PARTIAL)
4492                 return 0;
4493
4494         if (skb->protocol == cpu_to_be16(ETH_P_8021Q))
4495                 protocol = vlan_eth_hdr(skb)->h_vlan_encapsulated_proto;
4496         else
4497                 protocol = skb->protocol;
4498
4499         switch (protocol) {
4500         case cpu_to_be16(ETH_P_IP):
4501                 if (ip_hdr(skb)->protocol == IPPROTO_TCP)
4502                         cmd_len |= E1000_TXD_CMD_TCP;
4503                 break;
4504         case cpu_to_be16(ETH_P_IPV6):
4505                 /* XXX not handling all IPV6 headers */
4506                 if (ipv6_hdr(skb)->nexthdr == IPPROTO_TCP)
4507                         cmd_len |= E1000_TXD_CMD_TCP;
4508                 break;
4509         default:
4510                 if (unlikely(net_ratelimit()))
4511                         e_warn("checksum_partial proto=%x!\n",
4512                                be16_to_cpu(protocol));
4513                 break;
4514         }
4515
4516         css = skb_checksum_start_offset(skb);
4517
4518         i = tx_ring->next_to_use;
4519         buffer_info = &tx_ring->buffer_info[i];
4520         context_desc = E1000_CONTEXT_DESC(*tx_ring, i);
4521
4522         context_desc->lower_setup.ip_config = 0;
4523         context_desc->upper_setup.tcp_fields.tucss = css;
4524         context_desc->upper_setup.tcp_fields.tucso =
4525                                 css + skb->csum_offset;
4526         context_desc->upper_setup.tcp_fields.tucse = 0;
4527         context_desc->tcp_seg_setup.data = 0;
4528         context_desc->cmd_and_length = cpu_to_le32(cmd_len);
4529
4530         buffer_info->time_stamp = jiffies;
4531         buffer_info->next_to_watch = i;
4532
4533         i++;
4534         if (i == tx_ring->count)
4535                 i = 0;
4536         tx_ring->next_to_use = i;
4537
4538         return 1;
4539 }
4540
4541 #define E1000_MAX_PER_TXD       8192
4542 #define E1000_MAX_TXD_PWR       12
4543
4544 static int e1000_tx_map(struct e1000_adapter *adapter,
4545                         struct sk_buff *skb, unsigned int first,
4546                         unsigned int max_per_txd, unsigned int nr_frags,
4547                         unsigned int mss)
4548 {
4549         struct e1000_ring *tx_ring = adapter->tx_ring;
4550         struct pci_dev *pdev = adapter->pdev;
4551         struct e1000_buffer *buffer_info;
4552         unsigned int len = skb_headlen(skb);
4553         unsigned int offset = 0, size, count = 0, i;
4554         unsigned int f, bytecount, segs;
4555
4556         i = tx_ring->next_to_use;
4557
4558         while (len) {
4559                 buffer_info = &tx_ring->buffer_info[i];
4560                 size = min(len, max_per_txd);
4561
4562                 buffer_info->length = size;
4563                 buffer_info->time_stamp = jiffies;
4564                 buffer_info->next_to_watch = i;
4565                 buffer_info->dma = dma_map_single(&pdev->dev,
4566                                                   skb->data + offset,
4567                                                   size, DMA_TO_DEVICE);
4568                 buffer_info->mapped_as_page = false;
4569                 if (dma_mapping_error(&pdev->dev, buffer_info->dma))
4570                         goto dma_error;
4571
4572                 len -= size;
4573                 offset += size;
4574                 count++;
4575
4576                 if (len) {
4577                         i++;
4578                         if (i == tx_ring->count)
4579                                 i = 0;
4580                 }
4581         }
4582
4583         for (f = 0; f < nr_frags; f++) {
4584                 struct skb_frag_struct *frag;
4585
4586                 frag = &skb_shinfo(skb)->frags[f];
4587                 len = frag->size;
4588                 offset = frag->page_offset;
4589
4590                 while (len) {
4591                         i++;
4592                         if (i == tx_ring->count)
4593                                 i = 0;
4594
4595                         buffer_info = &tx_ring->buffer_info[i];
4596                         size = min(len, max_per_txd);
4597
4598                         buffer_info->length = size;
4599                         buffer_info->time_stamp = jiffies;
4600                         buffer_info->next_to_watch = i;
4601                         buffer_info->dma = dma_map_page(&pdev->dev, frag->page,
4602                                                         offset, size,
4603                                                         DMA_TO_DEVICE);
4604                         buffer_info->mapped_as_page = true;
4605                         if (dma_mapping_error(&pdev->dev, buffer_info->dma))
4606                                 goto dma_error;
4607
4608                         len -= size;
4609                         offset += size;
4610                         count++;
4611                 }
4612         }
4613
4614         segs = skb_shinfo(skb)->gso_segs ? : 1;
4615         /* multiply data chunks by size of headers */
4616         bytecount = ((segs - 1) * skb_headlen(skb)) + skb->len;
4617
4618         tx_ring->buffer_info[i].skb = skb;
4619         tx_ring->buffer_info[i].segs = segs;
4620         tx_ring->buffer_info[i].bytecount = bytecount;
4621         tx_ring->buffer_info[first].next_to_watch = i;
4622
4623         return count;
4624
4625 dma_error:
4626         dev_err(&pdev->dev, "Tx DMA map failed\n");
4627         buffer_info->dma = 0;
4628         if (count)
4629                 count--;
4630
4631         while (count--) {
4632                 if (i == 0)
4633                         i += tx_ring->count;
4634                 i--;
4635                 buffer_info = &tx_ring->buffer_info[i];
4636                 e1000_put_txbuf(adapter, buffer_info);
4637         }
4638
4639         return 0;
4640 }
4641
4642 static void e1000_tx_queue(struct e1000_adapter *adapter,
4643                            int tx_flags, int count)
4644 {
4645         struct e1000_ring *tx_ring = adapter->tx_ring;
4646         struct e1000_tx_desc *tx_desc = NULL;
4647         struct e1000_buffer *buffer_info;
4648         u32 txd_upper = 0, txd_lower = E1000_TXD_CMD_IFCS;
4649         unsigned int i;
4650
4651         if (tx_flags & E1000_TX_FLAGS_TSO) {
4652                 txd_lower |= E1000_TXD_CMD_DEXT | E1000_TXD_DTYP_D |
4653                              E1000_TXD_CMD_TSE;
4654                 txd_upper |= E1000_TXD_POPTS_TXSM << 8;
4655
4656                 if (tx_flags & E1000_TX_FLAGS_IPV4)
4657                         txd_upper |= E1000_TXD_POPTS_IXSM << 8;
4658         }
4659
4660         if (tx_flags & E1000_TX_FLAGS_CSUM) {
4661                 txd_lower |= E1000_TXD_CMD_DEXT | E1000_TXD_DTYP_D;
4662                 txd_upper |= E1000_TXD_POPTS_TXSM << 8;
4663         }
4664
4665         if (tx_flags & E1000_TX_FLAGS_VLAN) {
4666                 txd_lower |= E1000_TXD_CMD_VLE;
4667                 txd_upper |= (tx_flags & E1000_TX_FLAGS_VLAN_MASK);
4668         }
4669
4670         i = tx_ring->next_to_use;
4671
4672         do {
4673                 buffer_info = &tx_ring->buffer_info[i];
4674                 tx_desc = E1000_TX_DESC(*tx_ring, i);
4675                 tx_desc->buffer_addr = cpu_to_le64(buffer_info->dma);
4676                 tx_desc->lower.data =
4677                         cpu_to_le32(txd_lower | buffer_info->length);
4678                 tx_desc->upper.data = cpu_to_le32(txd_upper);
4679
4680                 i++;
4681                 if (i == tx_ring->count)
4682                         i = 0;
4683         } while (--count > 0);
4684
4685         tx_desc->lower.data |= cpu_to_le32(adapter->txd_cmd);
4686
4687         /*
4688          * Force memory writes to complete before letting h/w
4689          * know there are new descriptors to fetch.  (Only
4690          * applicable for weak-ordered memory model archs,
4691          * such as IA-64).
4692          */
4693         wmb();
4694
4695         tx_ring->next_to_use = i;
4696         writel(i, adapter->hw.hw_addr + tx_ring->tail);
4697         /*
4698          * we need this if more than one processor can write to our tail
4699          * at a time, it synchronizes IO on IA64/Altix systems
4700          */
4701         mmiowb();
4702 }
4703
4704 #define MINIMUM_DHCP_PACKET_SIZE 282
4705 static int e1000_transfer_dhcp_info(struct e1000_adapter *adapter,
4706                                     struct sk_buff *skb)
4707 {
4708         struct e1000_hw *hw =  &adapter->hw;
4709         u16 length, offset;
4710
4711         if (vlan_tx_tag_present(skb)) {
4712                 if (!((vlan_tx_tag_get(skb) == adapter->hw.mng_cookie.vlan_id) &&
4713                     (adapter->hw.mng_cookie.status &
4714                         E1000_MNG_DHCP_COOKIE_STATUS_VLAN)))
4715                         return 0;
4716         }
4717
4718         if (skb->len <= MINIMUM_DHCP_PACKET_SIZE)
4719                 return 0;
4720
4721         if (((struct ethhdr *) skb->data)->h_proto != htons(ETH_P_IP))
4722                 return 0;
4723
4724         {
4725                 const struct iphdr *ip = (struct iphdr *)((u8 *)skb->data+14);
4726                 struct udphdr *udp;
4727
4728                 if (ip->protocol != IPPROTO_UDP)
4729                         return 0;
4730
4731                 udp = (struct udphdr *)((u8 *)ip + (ip->ihl << 2));
4732                 if (ntohs(udp->dest) != 67)
4733                         return 0;
4734
4735                 offset = (u8 *)udp + 8 - skb->data;
4736                 length = skb->len - offset;
4737                 return e1000e_mng_write_dhcp_info(hw, (u8 *)udp + 8, length);
4738         }
4739
4740         return 0;
4741 }
4742
4743 static int __e1000_maybe_stop_tx(struct net_device *netdev, int size)
4744 {
4745         struct e1000_adapter *adapter = netdev_priv(netdev);
4746
4747         netif_stop_queue(netdev);
4748         /*
4749          * Herbert's original patch had:
4750          *  smp_mb__after_netif_stop_queue();
4751          * but since that doesn't exist yet, just open code it.
4752          */
4753         smp_mb();
4754
4755         /*
4756          * We need to check again in a case another CPU has just
4757          * made room available.
4758          */
4759         if (e1000_desc_unused(adapter->tx_ring) < size)
4760                 return -EBUSY;
4761
4762         /* A reprieve! */
4763         netif_start_queue(netdev);
4764         ++adapter->restart_queue;
4765         return 0;
4766 }
4767
4768 static int e1000_maybe_stop_tx(struct net_device *netdev, int size)
4769 {
4770         struct e1000_adapter *adapter = netdev_priv(netdev);
4771
4772         if (e1000_desc_unused(adapter->tx_ring) >= size)
4773                 return 0;
4774         return __e1000_maybe_stop_tx(netdev, size);
4775 }
4776
4777 #define TXD_USE_COUNT(S, X) (((S) >> (X)) + 1 )
4778 static netdev_tx_t e1000_xmit_frame(struct sk_buff *skb,
4779                                     struct net_device *netdev)
4780 {
4781         struct e1000_adapter *adapter = netdev_priv(netdev);
4782         struct e1000_ring *tx_ring = adapter->tx_ring;
4783         unsigned int first;
4784         unsigned int max_per_txd = E1000_MAX_PER_TXD;
4785         unsigned int max_txd_pwr = E1000_MAX_TXD_PWR;
4786         unsigned int tx_flags = 0;
4787         unsigned int len = skb_headlen(skb);
4788         unsigned int nr_frags;
4789         unsigned int mss;
4790         int count = 0;
4791         int tso;
4792         unsigned int f;
4793
4794         if (test_bit(__E1000_DOWN, &adapter->state)) {
4795                 dev_kfree_skb_any(skb);
4796                 return NETDEV_TX_OK;
4797         }
4798
4799         if (skb->len <= 0) {
4800                 dev_kfree_skb_any(skb);
4801                 return NETDEV_TX_OK;
4802         }
4803
4804         mss = skb_shinfo(skb)->gso_size;
4805         /*
4806          * The controller does a simple calculation to
4807          * make sure there is enough room in the FIFO before
4808          * initiating the DMA for each buffer.  The calc is:
4809          * 4 = ceil(buffer len/mss).  To make sure we don't
4810          * overrun the FIFO, adjust the max buffer len if mss
4811          * drops.
4812          */
4813         if (mss) {
4814                 u8 hdr_len;
4815                 max_per_txd = min(mss << 2, max_per_txd);
4816                 max_txd_pwr = fls(max_per_txd) - 1;
4817
4818                 /*
4819                  * TSO Workaround for 82571/2/3 Controllers -- if skb->data
4820                  * points to just header, pull a few bytes of payload from
4821                  * frags into skb->data
4822                  */
4823                 hdr_len = skb_transport_offset(skb) + tcp_hdrlen(skb);
4824                 /*
4825                  * we do this workaround for ES2LAN, but it is un-necessary,
4826                  * avoiding it could save a lot of cycles
4827                  */
4828                 if (skb->data_len && (hdr_len == len)) {
4829                         unsigned int pull_size;
4830
4831                         pull_size = min((unsigned int)4, skb->data_len);
4832                         if (!__pskb_pull_tail(skb, pull_size)) {
4833                                 e_err("__pskb_pull_tail failed.\n");
4834                                 dev_kfree_skb_any(skb);
4835                                 return NETDEV_TX_OK;
4836                         }
4837                         len = skb_headlen(skb);
4838                 }
4839         }
4840
4841         /* reserve a descriptor for the offload context */
4842         if ((mss) || (skb->ip_summed == CHECKSUM_PARTIAL))
4843                 count++;
4844         count++;
4845
4846         count += TXD_USE_COUNT(len, max_txd_pwr);
4847
4848         nr_frags = skb_shinfo(skb)->nr_frags;
4849         for (f = 0; f < nr_frags; f++)
4850                 count += TXD_USE_COUNT(skb_shinfo(skb)->frags[f].size,
4851                                        max_txd_pwr);
4852
4853         if (adapter->hw.mac.tx_pkt_filtering)
4854                 e1000_transfer_dhcp_info(adapter, skb);
4855
4856         /*
4857          * need: count + 2 desc gap to keep tail from touching
4858          * head, otherwise try next time
4859          */
4860         if (e1000_maybe_stop_tx(netdev, count + 2))
4861                 return NETDEV_TX_BUSY;
4862
4863         if (vlan_tx_tag_present(skb)) {
4864                 tx_flags |= E1000_TX_FLAGS_VLAN;
4865                 tx_flags |= (vlan_tx_tag_get(skb) << E1000_TX_FLAGS_VLAN_SHIFT);
4866         }
4867
4868         first = tx_ring->next_to_use;
4869
4870         tso = e1000_tso(adapter, skb);
4871         if (tso < 0) {
4872                 dev_kfree_skb_any(skb);
4873                 return NETDEV_TX_OK;
4874         }
4875
4876         if (tso)
4877                 tx_flags |= E1000_TX_FLAGS_TSO;
4878         else if (e1000_tx_csum(adapter, skb))
4879                 tx_flags |= E1000_TX_FLAGS_CSUM;
4880
4881         /*
4882          * Old method was to assume IPv4 packet by default if TSO was enabled.
4883          * 82571 hardware supports TSO capabilities for IPv6 as well...
4884          * no longer assume, we must.
4885          */
4886         if (skb->protocol == htons(ETH_P_IP))
4887                 tx_flags |= E1000_TX_FLAGS_IPV4;
4888
4889         /* if count is 0 then mapping error has occurred */
4890         count = e1000_tx_map(adapter, skb, first, max_per_txd, nr_frags, mss);
4891         if (count) {
4892                 e1000_tx_queue(adapter, tx_flags, count);
4893                 /* Make sure there is space in the ring for the next send. */
4894                 e1000_maybe_stop_tx(netdev, MAX_SKB_FRAGS + 2);
4895
4896         } else {
4897                 dev_kfree_skb_any(skb);
4898                 tx_ring->buffer_info[first].time_stamp = 0;
4899                 tx_ring->next_to_use = first;
4900         }
4901
4902         return NETDEV_TX_OK;
4903 }
4904
4905 /**
4906  * e1000_tx_timeout - Respond to a Tx Hang
4907  * @netdev: network interface device structure
4908  **/
4909 static void e1000_tx_timeout(struct net_device *netdev)
4910 {
4911         struct e1000_adapter *adapter = netdev_priv(netdev);
4912
4913         /* Do the reset outside of interrupt context */
4914         adapter->tx_timeout_count++;
4915         schedule_work(&adapter->reset_task);
4916 }
4917
4918 static void e1000_reset_task(struct work_struct *work)
4919 {
4920         struct e1000_adapter *adapter;
4921         adapter = container_of(work, struct e1000_adapter, reset_task);
4922
4923         /* don't run the task if already down */
4924         if (test_bit(__E1000_DOWN, &adapter->state))
4925                 return;
4926
4927         if (!((adapter->flags & FLAG_RX_NEEDS_RESTART) &&
4928               (adapter->flags & FLAG_RX_RESTART_NOW))) {
4929                 e1000e_dump(adapter);
4930                 e_err("Reset adapter\n");
4931         }
4932         e1000e_reinit_locked(adapter);
4933 }
4934
4935 /**
4936  * e1000_get_stats64 - Get System Network Statistics
4937  * @netdev: network interface device structure
4938  * @stats: rtnl_link_stats64 pointer
4939  *
4940  * Returns the address of the device statistics structure.
4941  **/
4942 struct rtnl_link_stats64 *e1000e_get_stats64(struct net_device *netdev,
4943                                              struct rtnl_link_stats64 *stats)
4944 {
4945         struct e1000_adapter *adapter = netdev_priv(netdev);
4946
4947         memset(stats, 0, sizeof(struct rtnl_link_stats64));
4948         spin_lock(&adapter->stats64_lock);
4949         e1000e_update_stats(adapter);
4950         /* Fill out the OS statistics structure */
4951         stats->rx_bytes = adapter->stats.gorc;
4952         stats->rx_packets = adapter->stats.gprc;
4953         stats->tx_bytes = adapter->stats.gotc;
4954         stats->tx_packets = adapter->stats.gptc;
4955         stats->multicast = adapter->stats.mprc;
4956         stats->collisions = adapter->stats.colc;
4957
4958         /* Rx Errors */
4959
4960         /*
4961          * RLEC on some newer hardware can be incorrect so build
4962          * our own version based on RUC and ROC
4963          */
4964         stats->rx_errors = adapter->stats.rxerrc +
4965                 adapter->stats.crcerrs + adapter->stats.algnerrc +
4966                 adapter->stats.ruc + adapter->stats.roc +
4967                 adapter->stats.cexterr;
4968         stats->rx_length_errors = adapter->stats.ruc +
4969                                               adapter->stats.roc;
4970         stats->rx_crc_errors = adapter->stats.crcerrs;
4971         stats->rx_frame_errors = adapter->stats.algnerrc;
4972         stats->rx_missed_errors = adapter->stats.mpc;
4973
4974         /* Tx Errors */
4975         stats->tx_errors = adapter->stats.ecol +
4976                                        adapter->stats.latecol;
4977         stats->tx_aborted_errors = adapter->stats.ecol;
4978         stats->tx_window_errors = adapter->stats.latecol;
4979         stats->tx_carrier_errors = adapter->stats.tncrs;
4980
4981         /* Tx Dropped needs to be maintained elsewhere */
4982
4983         spin_unlock(&adapter->stats64_lock);
4984         return stats;
4985 }
4986
4987 /**
4988  * e1000_change_mtu - Change the Maximum Transfer Unit
4989  * @netdev: network interface device structure
4990  * @new_mtu: new value for maximum frame size
4991  *
4992  * Returns 0 on success, negative on failure
4993  **/
4994 static int e1000_change_mtu(struct net_device *netdev, int new_mtu)
4995 {
4996         struct e1000_adapter *adapter = netdev_priv(netdev);
4997         int max_frame = new_mtu + ETH_HLEN + ETH_FCS_LEN;
4998
4999         /* Jumbo frame support */
5000         if ((max_frame > ETH_FRAME_LEN + ETH_FCS_LEN) &&
5001             !(adapter->flags & FLAG_HAS_JUMBO_FRAMES)) {
5002                 e_err("Jumbo Frames not supported.\n");
5003                 return -EINVAL;
5004         }
5005
5006         /* Supported frame sizes */
5007         if ((new_mtu < ETH_ZLEN + ETH_FCS_LEN + VLAN_HLEN) ||
5008             (max_frame > adapter->max_hw_frame_size)) {
5009                 e_err("Unsupported MTU setting\n");
5010                 return -EINVAL;
5011         }
5012
5013         /* Jumbo frame workaround on 82579 requires CRC be stripped */
5014         if ((adapter->hw.mac.type == e1000_pch2lan) &&
5015             !(adapter->flags2 & FLAG2_CRC_STRIPPING) &&
5016             (new_mtu > ETH_DATA_LEN)) {
5017                 e_err("Jumbo Frames not supported on 82579 when CRC "
5018                       "stripping is disabled.\n");
5019                 return -EINVAL;
5020         }
5021
5022         /* 82573 Errata 17 */
5023         if (((adapter->hw.mac.type == e1000_82573) ||
5024              (adapter->hw.mac.type == e1000_82574)) &&
5025             (max_frame > ETH_FRAME_LEN + ETH_FCS_LEN)) {
5026                 adapter->flags2 |= FLAG2_DISABLE_ASPM_L1;
5027                 e1000e_disable_aspm(adapter->pdev, PCIE_LINK_STATE_L1);
5028         }
5029
5030         while (test_and_set_bit(__E1000_RESETTING, &adapter->state))
5031                 msleep(1);
5032         /* e1000e_down -> e1000e_reset dependent on max_frame_size & mtu */
5033         adapter->max_frame_size = max_frame;
5034         e_info("changing MTU from %d to %d\n", netdev->mtu, new_mtu);
5035         netdev->mtu = new_mtu;
5036         if (netif_running(netdev))
5037                 e1000e_down(adapter);
5038
5039         /*
5040          * NOTE: netdev_alloc_skb reserves 16 bytes, and typically NET_IP_ALIGN
5041          * means we reserve 2 more, this pushes us to allocate from the next
5042          * larger slab size.
5043          * i.e. RXBUFFER_2048 --> size-4096 slab
5044          * However with the new *_jumbo_rx* routines, jumbo receives will use
5045          * fragmented skbs
5046          */
5047
5048         if (max_frame <= 2048)
5049                 adapter->rx_buffer_len = 2048;
5050         else
5051                 adapter->rx_buffer_len = 4096;
5052
5053         /* adjust allocation if LPE protects us, and we aren't using SBP */
5054         if ((max_frame == ETH_FRAME_LEN + ETH_FCS_LEN) ||
5055              (max_frame == ETH_FRAME_LEN + VLAN_HLEN + ETH_FCS_LEN))
5056                 adapter->rx_buffer_len = ETH_FRAME_LEN + VLAN_HLEN
5057                                          + ETH_FCS_LEN;
5058
5059         if (netif_running(netdev))
5060                 e1000e_up(adapter);
5061         else
5062                 e1000e_reset(adapter);
5063
5064         clear_bit(__E1000_RESETTING, &adapter->state);
5065
5066         return 0;
5067 }
5068
5069 static int e1000_mii_ioctl(struct net_device *netdev, struct ifreq *ifr,
5070                            int cmd)
5071 {
5072         struct e1000_adapter *adapter = netdev_priv(netdev);
5073         struct mii_ioctl_data *data = if_mii(ifr);
5074
5075         if (adapter->hw.phy.media_type != e1000_media_type_copper)
5076                 return -EOPNOTSUPP;
5077
5078         switch (cmd) {
5079         case SIOCGMIIPHY:
5080                 data->phy_id = adapter->hw.phy.addr;
5081                 break;
5082         case SIOCGMIIREG:
5083                 e1000_phy_read_status(adapter);
5084
5085                 switch (data->reg_num & 0x1F) {
5086                 case MII_BMCR:
5087                         data->val_out = adapter->phy_regs.bmcr;
5088                         break;
5089                 case MII_BMSR:
5090                         data->val_out = adapter->phy_regs.bmsr;
5091                         break;
5092                 case MII_PHYSID1:
5093                         data->val_out = (adapter->hw.phy.id >> 16);
5094                         break;
5095                 case MII_PHYSID2:
5096                         data->val_out = (adapter->hw.phy.id & 0xFFFF);
5097                         break;
5098                 case MII_ADVERTISE:
5099                         data->val_out = adapter->phy_regs.advertise;
5100                         break;
5101                 case MII_LPA:
5102                         data->val_out = adapter->phy_regs.lpa;
5103                         break;
5104                 case MII_EXPANSION:
5105                         data->val_out = adapter->phy_regs.expansion;
5106                         break;
5107                 case MII_CTRL1000:
5108                         data->val_out = adapter->phy_regs.ctrl1000;
5109                         break;
5110                 case MII_STAT1000:
5111                         data->val_out = adapter->phy_regs.stat1000;
5112                         break;
5113                 case MII_ESTATUS:
5114                         data->val_out = adapter->phy_regs.estatus;
5115                         break;
5116                 default:
5117                         return -EIO;
5118                 }
5119                 break;
5120         case SIOCSMIIREG:
5121         default:
5122                 return -EOPNOTSUPP;
5123         }
5124         return 0;
5125 }
5126
5127 static int e1000_ioctl(struct net_device *netdev, struct ifreq *ifr, int cmd)
5128 {
5129         switch (cmd) {
5130         case SIOCGMIIPHY:
5131         case SIOCGMIIREG:
5132         case SIOCSMIIREG:
5133                 return e1000_mii_ioctl(netdev, ifr, cmd);
5134         default:
5135                 return -EOPNOTSUPP;
5136         }
5137 }
5138
5139 static int e1000_init_phy_wakeup(struct e1000_adapter *adapter, u32 wufc)
5140 {
5141         struct e1000_hw *hw = &adapter->hw;
5142         u32 i, mac_reg;
5143         u16 phy_reg;
5144         int retval = 0;
5145
5146         /* copy MAC RARs to PHY RARs */
5147         e1000_copy_rx_addrs_to_phy_ich8lan(hw);
5148
5149         /* copy MAC MTA to PHY MTA */
5150         for (i = 0; i < adapter->hw.mac.mta_reg_count; i++) {
5151                 mac_reg = E1000_READ_REG_ARRAY(hw, E1000_MTA, i);
5152                 e1e_wphy(hw, BM_MTA(i), (u16)(mac_reg & 0xFFFF));
5153                 e1e_wphy(hw, BM_MTA(i) + 1, (u16)((mac_reg >> 16) & 0xFFFF));
5154         }
5155
5156         /* configure PHY Rx Control register */
5157         e1e_rphy(&adapter->hw, BM_RCTL, &phy_reg);
5158         mac_reg = er32(RCTL);
5159         if (mac_reg & E1000_RCTL_UPE)
5160                 phy_reg |= BM_RCTL_UPE;
5161         if (mac_reg & E1000_RCTL_MPE)
5162                 phy_reg |= BM_RCTL_MPE;
5163         phy_reg &= ~(BM_RCTL_MO_MASK);
5164         if (mac_reg & E1000_RCTL_MO_3)
5165                 phy_reg |= (((mac_reg & E1000_RCTL_MO_3) >> E1000_RCTL_MO_SHIFT)
5166                                 << BM_RCTL_MO_SHIFT);
5167         if (mac_reg & E1000_RCTL_BAM)
5168                 phy_reg |= BM_RCTL_BAM;
5169         if (mac_reg & E1000_RCTL_PMCF)
5170                 phy_reg |= BM_RCTL_PMCF;
5171         mac_reg = er32(CTRL);
5172         if (mac_reg & E1000_CTRL_RFCE)
5173                 phy_reg |= BM_RCTL_RFCE;
5174         e1e_wphy(&adapter->hw, BM_RCTL, phy_reg);
5175
5176         /* enable PHY wakeup in MAC register */
5177         ew32(WUFC, wufc);
5178         ew32(WUC, E1000_WUC_PHY_WAKE | E1000_WUC_PME_EN);
5179
5180         /* configure and enable PHY wakeup in PHY registers */
5181         e1e_wphy(&adapter->hw, BM_WUFC, wufc);
5182         e1e_wphy(&adapter->hw, BM_WUC, E1000_WUC_PME_EN);
5183
5184         /* activate PHY wakeup */
5185         retval = hw->phy.ops.acquire(hw);
5186         if (retval) {
5187                 e_err("Could not acquire PHY\n");
5188                 return retval;
5189         }
5190         e1000e_write_phy_reg_mdic(hw, IGP01E1000_PHY_PAGE_SELECT,
5191                                  (BM_WUC_ENABLE_PAGE << IGP_PAGE_SHIFT));
5192         retval = e1000e_read_phy_reg_mdic(hw, BM_WUC_ENABLE_REG, &phy_reg);
5193         if (retval) {
5194                 e_err("Could not read PHY page 769\n");
5195                 goto out;
5196         }
5197         phy_reg |= BM_WUC_ENABLE_BIT | BM_WUC_HOST_WU_BIT;
5198         retval = e1000e_write_phy_reg_mdic(hw, BM_WUC_ENABLE_REG, phy_reg);
5199         if (retval)
5200                 e_err("Could not set PHY Host Wakeup bit\n");
5201 out:
5202         hw->phy.ops.release(hw);
5203
5204         return retval;
5205 }
5206
5207 static int __e1000_shutdown(struct pci_dev *pdev, bool *enable_wake,
5208                             bool runtime)
5209 {
5210         struct net_device *netdev = pci_get_drvdata(pdev);
5211         struct e1000_adapter *adapter = netdev_priv(netdev);
5212         struct e1000_hw *hw = &adapter->hw;
5213         u32 ctrl, ctrl_ext, rctl, status;
5214         /* Runtime suspend should only enable wakeup for link changes */
5215         u32 wufc = runtime ? E1000_WUFC_LNKC : adapter->wol;
5216         int retval = 0;
5217
5218         netif_device_detach(netdev);
5219
5220         if (netif_running(netdev)) {
5221                 WARN_ON(test_bit(__E1000_RESETTING, &adapter->state));
5222                 e1000e_down(adapter);
5223                 e1000_free_irq(adapter);
5224         }
5225         e1000e_reset_interrupt_capability(adapter);
5226
5227         retval = pci_save_state(pdev);
5228         if (retval)
5229                 return retval;
5230
5231         status = er32(STATUS);
5232         if (status & E1000_STATUS_LU)
5233                 wufc &= ~E1000_WUFC_LNKC;
5234
5235         if (wufc) {
5236                 e1000_setup_rctl(adapter);
5237                 e1000_set_multi(netdev);
5238
5239                 /* turn on all-multi mode if wake on multicast is enabled */
5240                 if (wufc & E1000_WUFC_MC) {
5241                         rctl = er32(RCTL);
5242                         rctl |= E1000_RCTL_MPE;
5243                         ew32(RCTL, rctl);
5244                 }
5245
5246                 ctrl = er32(CTRL);
5247                 /* advertise wake from D3Cold */
5248                 #define E1000_CTRL_ADVD3WUC 0x00100000
5249                 /* phy power management enable */
5250                 #define E1000_CTRL_EN_PHY_PWR_MGMT 0x00200000
5251                 ctrl |= E1000_CTRL_ADVD3WUC;
5252                 if (!(adapter->flags2 & FLAG2_HAS_PHY_WAKEUP))
5253                         ctrl |= E1000_CTRL_EN_PHY_PWR_MGMT;
5254                 ew32(CTRL, ctrl);
5255
5256                 if (adapter->hw.phy.media_type == e1000_media_type_fiber ||
5257                     adapter->hw.phy.media_type ==
5258                     e1000_media_type_internal_serdes) {
5259                         /* keep the laser running in D3 */
5260                         ctrl_ext = er32(CTRL_EXT);
5261                         ctrl_ext |= E1000_CTRL_EXT_SDP3_DATA;
5262                         ew32(CTRL_EXT, ctrl_ext);
5263                 }
5264
5265                 if (adapter->flags & FLAG_IS_ICH)
5266                         e1000e_disable_gig_wol_ich8lan(&adapter->hw);
5267
5268                 /* Allow time for pending master requests to run */
5269                 e1000e_disable_pcie_master(&adapter->hw);
5270
5271                 if (adapter->flags2 & FLAG2_HAS_PHY_WAKEUP) {
5272                         /* enable wakeup by the PHY */
5273                         retval = e1000_init_phy_wakeup(adapter, wufc);
5274                         if (retval)
5275                                 return retval;
5276                 } else {
5277                         /* enable wakeup by the MAC */
5278                         ew32(WUFC, wufc);
5279                         ew32(WUC, E1000_WUC_PME_EN);
5280                 }
5281         } else {
5282                 ew32(WUC, 0);
5283                 ew32(WUFC, 0);
5284         }
5285
5286         *enable_wake = !!wufc;
5287
5288         /* make sure adapter isn't asleep if manageability is enabled */
5289         if ((adapter->flags & FLAG_MNG_PT_ENABLED) ||
5290             (hw->mac.ops.check_mng_mode(hw)))
5291                 *enable_wake = true;
5292
5293         if (adapter->hw.phy.type == e1000_phy_igp_3)
5294                 e1000e_igp3_phy_powerdown_workaround_ich8lan(&adapter->hw);
5295
5296         /*
5297          * Release control of h/w to f/w.  If f/w is AMT enabled, this
5298          * would have already happened in close and is redundant.
5299          */
5300         e1000e_release_hw_control(adapter);
5301
5302         pci_disable_device(pdev);
5303
5304         return 0;
5305 }
5306
5307 static void e1000_power_off(struct pci_dev *pdev, bool sleep, bool wake)
5308 {
5309         if (sleep && wake) {
5310                 pci_prepare_to_sleep(pdev);
5311                 return;
5312         }
5313
5314         pci_wake_from_d3(pdev, wake);
5315         pci_set_power_state(pdev, PCI_D3hot);
5316 }
5317
5318 static void e1000_complete_shutdown(struct pci_dev *pdev, bool sleep,
5319                                     bool wake)
5320 {
5321         struct net_device *netdev = pci_get_drvdata(pdev);
5322         struct e1000_adapter *adapter = netdev_priv(netdev);
5323
5324         /*
5325          * The pci-e switch on some quad port adapters will report a
5326          * correctable error when the MAC transitions from D0 to D3.  To
5327          * prevent this we need to mask off the correctable errors on the
5328          * downstream port of the pci-e switch.
5329          */
5330         if (adapter->flags & FLAG_IS_QUAD_PORT) {
5331                 struct pci_dev *us_dev = pdev->bus->self;
5332                 int pos = pci_find_capability(us_dev, PCI_CAP_ID_EXP);
5333                 u16 devctl;
5334
5335                 pci_read_config_word(us_dev, pos + PCI_EXP_DEVCTL, &devctl);
5336                 pci_write_config_word(us_dev, pos + PCI_EXP_DEVCTL,
5337                                       (devctl & ~PCI_EXP_DEVCTL_CERE));
5338
5339                 e1000_power_off(pdev, sleep, wake);
5340
5341                 pci_write_config_word(us_dev, pos + PCI_EXP_DEVCTL, devctl);
5342         } else {
5343                 e1000_power_off(pdev, sleep, wake);
5344         }
5345 }
5346
5347 #ifdef CONFIG_PCIEASPM
5348 static void __e1000e_disable_aspm(struct pci_dev *pdev, u16 state)
5349 {
5350         pci_disable_link_state(pdev, state);
5351 }
5352 #else
5353 static void __e1000e_disable_aspm(struct pci_dev *pdev, u16 state)
5354 {
5355         int pos;
5356         u16 reg16;
5357
5358         /*
5359          * Both device and parent should have the same ASPM setting.
5360          * Disable ASPM in downstream component first and then upstream.
5361          */
5362         pos = pci_pcie_cap(pdev);
5363         pci_read_config_word(pdev, pos + PCI_EXP_LNKCTL, &reg16);
5364         reg16 &= ~state;
5365         pci_write_config_word(pdev, pos + PCI_EXP_LNKCTL, reg16);
5366
5367         if (!pdev->bus->self)
5368                 return;
5369
5370         pos = pci_pcie_cap(pdev->bus->self);
5371         pci_read_config_word(pdev->bus->self, pos + PCI_EXP_LNKCTL, &reg16);
5372         reg16 &= ~state;
5373         pci_write_config_word(pdev->bus->self, pos + PCI_EXP_LNKCTL, reg16);
5374 }
5375 #endif
5376 void e1000e_disable_aspm(struct pci_dev *pdev, u16 state)
5377 {
5378         dev_info(&pdev->dev, "Disabling ASPM %s %s\n",
5379                  (state & PCIE_LINK_STATE_L0S) ? "L0s" : "",
5380                  (state & PCIE_LINK_STATE_L1) ? "L1" : "");
5381
5382         __e1000e_disable_aspm(pdev, state);
5383 }
5384
5385 #ifdef CONFIG_PM
5386 static bool e1000e_pm_ready(struct e1000_adapter *adapter)
5387 {
5388         return !!adapter->tx_ring->buffer_info;
5389 }
5390
5391 static int __e1000_resume(struct pci_dev *pdev)
5392 {
5393         struct net_device *netdev = pci_get_drvdata(pdev);
5394         struct e1000_adapter *adapter = netdev_priv(netdev);
5395         struct e1000_hw *hw = &adapter->hw;
5396         u32 err;
5397
5398         pci_set_power_state(pdev, PCI_D0);
5399         pci_restore_state(pdev);
5400         pci_save_state(pdev);
5401         if (adapter->flags2 & FLAG2_DISABLE_ASPM_L1)
5402                 e1000e_disable_aspm(pdev, PCIE_LINK_STATE_L1);
5403
5404         e1000e_set_interrupt_capability(adapter);
5405         if (netif_running(netdev)) {
5406                 err = e1000_request_irq(adapter);
5407                 if (err)
5408                         return err;
5409         }
5410
5411         e1000e_power_up_phy(adapter);
5412
5413         /* report the system wakeup cause from S3/S4 */
5414         if (adapter->flags2 & FLAG2_HAS_PHY_WAKEUP) {
5415                 u16 phy_data;
5416
5417                 e1e_rphy(&adapter->hw, BM_WUS, &phy_data);
5418                 if (phy_data) {
5419                         e_info("PHY Wakeup cause - %s\n",
5420                                 phy_data & E1000_WUS_EX ? "Unicast Packet" :
5421                                 phy_data & E1000_WUS_MC ? "Multicast Packet" :
5422                                 phy_data & E1000_WUS_BC ? "Broadcast Packet" :
5423                                 phy_data & E1000_WUS_MAG ? "Magic Packet" :
5424                                 phy_data & E1000_WUS_LNKC ? "Link Status "
5425                                 " Change" : "other");
5426                 }
5427                 e1e_wphy(&adapter->hw, BM_WUS, ~0);
5428         } else {
5429                 u32 wus = er32(WUS);
5430                 if (wus) {
5431                         e_info("MAC Wakeup cause - %s\n",
5432                                 wus & E1000_WUS_EX ? "Unicast Packet" :
5433                                 wus & E1000_WUS_MC ? "Multicast Packet" :
5434                                 wus & E1000_WUS_BC ? "Broadcast Packet" :
5435                                 wus & E1000_WUS_MAG ? "Magic Packet" :
5436                                 wus & E1000_WUS_LNKC ? "Link Status Change" :
5437                                 "other");
5438                 }
5439                 ew32(WUS, ~0);
5440         }
5441
5442         e1000e_reset(adapter);
5443
5444         e1000_init_manageability_pt(adapter);
5445
5446         if (netif_running(netdev))
5447                 e1000e_up(adapter);
5448
5449         netif_device_attach(netdev);
5450
5451         /*
5452          * If the controller has AMT, do not set DRV_LOAD until the interface
5453          * is up.  For all other cases, let the f/w know that the h/w is now
5454          * under the control of the driver.
5455          */
5456         if (!(adapter->flags & FLAG_HAS_AMT))
5457                 e1000e_get_hw_control(adapter);
5458
5459         return 0;
5460 }
5461
5462 #ifdef CONFIG_PM_SLEEP
5463 static int e1000_suspend(struct device *dev)
5464 {
5465         struct pci_dev *pdev = to_pci_dev(dev);
5466         int retval;
5467         bool wake;
5468
5469         retval = __e1000_shutdown(pdev, &wake, false);
5470         if (!retval)
5471                 e1000_complete_shutdown(pdev, true, wake);
5472
5473         return retval;
5474 }
5475
5476 static int e1000_resume(struct device *dev)
5477 {
5478         struct pci_dev *pdev = to_pci_dev(dev);
5479         struct net_device *netdev = pci_get_drvdata(pdev);
5480         struct e1000_adapter *adapter = netdev_priv(netdev);
5481
5482         if (e1000e_pm_ready(adapter))
5483                 adapter->idle_check = true;
5484
5485         return __e1000_resume(pdev);
5486 }
5487 #endif /* CONFIG_PM_SLEEP */
5488
5489 #ifdef CONFIG_PM_RUNTIME
5490 static int e1000_runtime_suspend(struct device *dev)
5491 {
5492         struct pci_dev *pdev = to_pci_dev(dev);
5493         struct net_device *netdev = pci_get_drvdata(pdev);
5494         struct e1000_adapter *adapter = netdev_priv(netdev);
5495
5496         if (e1000e_pm_ready(adapter)) {
5497                 bool wake;
5498
5499                 __e1000_shutdown(pdev, &wake, true);
5500         }
5501
5502         return 0;
5503 }
5504
5505 static int e1000_idle(struct device *dev)
5506 {
5507         struct pci_dev *pdev = to_pci_dev(dev);
5508         struct net_device *netdev = pci_get_drvdata(pdev);
5509         struct e1000_adapter *adapter = netdev_priv(netdev);
5510
5511         if (!e1000e_pm_ready(adapter))
5512                 return 0;
5513
5514         if (adapter->idle_check) {
5515                 adapter->idle_check = false;
5516                 if (!e1000e_has_link(adapter))
5517                         pm_schedule_suspend(dev, MSEC_PER_SEC);
5518         }
5519
5520         return -EBUSY;
5521 }
5522
5523 static int e1000_runtime_resume(struct device *dev)
5524 {
5525         struct pci_dev *pdev = to_pci_dev(dev);
5526         struct net_device *netdev = pci_get_drvdata(pdev);
5527         struct e1000_adapter *adapter = netdev_priv(netdev);
5528
5529         if (!e1000e_pm_ready(adapter))
5530                 return 0;
5531
5532         adapter->idle_check = !dev->power.runtime_auto;
5533         return __e1000_resume(pdev);
5534 }
5535 #endif /* CONFIG_PM_RUNTIME */
5536 #endif /* CONFIG_PM */
5537
5538 static void e1000_shutdown(struct pci_dev *pdev)
5539 {
5540         bool wake = false;
5541
5542         __e1000_shutdown(pdev, &wake, false);
5543
5544         if (system_state == SYSTEM_POWER_OFF)
5545                 e1000_complete_shutdown(pdev, false, wake);
5546 }
5547
5548 #ifdef CONFIG_NET_POLL_CONTROLLER
5549
5550 static irqreturn_t e1000_intr_msix(int irq, void *data)
5551 {
5552         struct net_device *netdev = data;
5553         struct e1000_adapter *adapter = netdev_priv(netdev);
5554
5555         if (adapter->msix_entries) {
5556                 int vector, msix_irq;
5557
5558                 vector = 0;
5559                 msix_irq = adapter->msix_entries[vector].vector;
5560                 disable_irq(msix_irq);
5561                 e1000_intr_msix_rx(msix_irq, netdev);
5562                 enable_irq(msix_irq);
5563
5564                 vector++;
5565                 msix_irq = adapter->msix_entries[vector].vector;
5566                 disable_irq(msix_irq);
5567                 e1000_intr_msix_tx(msix_irq, netdev);
5568                 enable_irq(msix_irq);
5569
5570                 vector++;
5571                 msix_irq = adapter->msix_entries[vector].vector;
5572                 disable_irq(msix_irq);
5573                 e1000_msix_other(msix_irq, netdev);
5574                 enable_irq(msix_irq);
5575         }
5576
5577         return IRQ_HANDLED;
5578 }
5579
5580 /*
5581  * Polling 'interrupt' - used by things like netconsole to send skbs
5582  * without having to re-enable interrupts. It's not called while
5583  * the interrupt routine is executing.
5584  */
5585 static void e1000_netpoll(struct net_device *netdev)
5586 {
5587         struct e1000_adapter *adapter = netdev_priv(netdev);
5588
5589         switch (adapter->int_mode) {
5590         case E1000E_INT_MODE_MSIX:
5591                 e1000_intr_msix(adapter->pdev->irq, netdev);
5592                 break;
5593         case E1000E_INT_MODE_MSI:
5594                 disable_irq(adapter->pdev->irq);
5595                 e1000_intr_msi(adapter->pdev->irq, netdev);
5596                 enable_irq(adapter->pdev->irq);
5597                 break;
5598         default: /* E1000E_INT_MODE_LEGACY */
5599                 disable_irq(adapter->pdev->irq);
5600                 e1000_intr(adapter->pdev->irq, netdev);
5601                 enable_irq(adapter->pdev->irq);
5602                 break;
5603         }
5604 }
5605 #endif
5606
5607 /**
5608  * e1000_io_error_detected - called when PCI error is detected
5609  * @pdev: Pointer to PCI device
5610  * @state: The current pci connection state
5611  *
5612  * This function is called after a PCI bus error affecting
5613  * this device has been detected.
5614  */
5615 static pci_ers_result_t e1000_io_error_detected(struct pci_dev *pdev,
5616                                                 pci_channel_state_t state)
5617 {
5618         struct net_device *netdev = pci_get_drvdata(pdev);
5619         struct e1000_adapter *adapter = netdev_priv(netdev);
5620
5621         netif_device_detach(netdev);
5622
5623         if (state == pci_channel_io_perm_failure)
5624                 return PCI_ERS_RESULT_DISCONNECT;
5625
5626         if (netif_running(netdev))
5627                 e1000e_down(adapter);
5628         pci_disable_device(pdev);
5629
5630         /* Request a slot slot reset. */
5631         return PCI_ERS_RESULT_NEED_RESET;
5632 }
5633
5634 /**
5635  * e1000_io_slot_reset - called after the pci bus has been reset.
5636  * @pdev: Pointer to PCI device
5637  *
5638  * Restart the card from scratch, as if from a cold-boot. Implementation
5639  * resembles the first-half of the e1000_resume routine.
5640  */
5641 static pci_ers_result_t e1000_io_slot_reset(struct pci_dev *pdev)
5642 {
5643         struct net_device *netdev = pci_get_drvdata(pdev);
5644         struct e1000_adapter *adapter = netdev_priv(netdev);
5645         struct e1000_hw *hw = &adapter->hw;
5646         int err;
5647         pci_ers_result_t result;
5648
5649         if (adapter->flags2 & FLAG2_DISABLE_ASPM_L1)
5650                 e1000e_disable_aspm(pdev, PCIE_LINK_STATE_L1);
5651         err = pci_enable_device_mem(pdev);
5652         if (err) {
5653                 dev_err(&pdev->dev,
5654                         "Cannot re-enable PCI device after reset.\n");
5655                 result = PCI_ERS_RESULT_DISCONNECT;
5656         } else {
5657                 pci_set_master(pdev);
5658                 pdev->state_saved = true;
5659                 pci_restore_state(pdev);
5660
5661                 pci_enable_wake(pdev, PCI_D3hot, 0);
5662                 pci_enable_wake(pdev, PCI_D3cold, 0);
5663
5664                 e1000e_reset(adapter);
5665                 ew32(WUS, ~0);
5666                 result = PCI_ERS_RESULT_RECOVERED;
5667         }
5668
5669         pci_cleanup_aer_uncorrect_error_status(pdev);
5670
5671         return result;
5672 }
5673
5674 /**
5675  * e1000_io_resume - called when traffic can start flowing again.
5676  * @pdev: Pointer to PCI device
5677  *
5678  * This callback is called when the error recovery driver tells us that
5679  * its OK to resume normal operation. Implementation resembles the
5680  * second-half of the e1000_resume routine.
5681  */
5682 static void e1000_io_resume(struct pci_dev *pdev)
5683 {
5684         struct net_device *netdev = pci_get_drvdata(pdev);
5685         struct e1000_adapter *adapter = netdev_priv(netdev);
5686
5687         e1000_init_manageability_pt(adapter);
5688
5689         if (netif_running(netdev)) {
5690                 if (e1000e_up(adapter)) {
5691                         dev_err(&pdev->dev,
5692                                 "can't bring device back up after reset\n");
5693                         return;
5694                 }
5695         }
5696
5697         netif_device_attach(netdev);
5698
5699         /*
5700          * If the controller has AMT, do not set DRV_LOAD until the interface
5701          * is up.  For all other cases, let the f/w know that the h/w is now
5702          * under the control of the driver.
5703          */
5704         if (!(adapter->flags & FLAG_HAS_AMT))
5705                 e1000e_get_hw_control(adapter);
5706
5707 }
5708
5709 static void e1000_print_device_info(struct e1000_adapter *adapter)
5710 {
5711         struct e1000_hw *hw = &adapter->hw;
5712         struct net_device *netdev = adapter->netdev;
5713         u32 ret_val;
5714         u8 pba_str[E1000_PBANUM_LENGTH];
5715
5716         /* print bus type/speed/width info */
5717         e_info("(PCI Express:2.5GB/s:%s) %pM\n",
5718                /* bus width */
5719                ((hw->bus.width == e1000_bus_width_pcie_x4) ? "Width x4" :
5720                 "Width x1"),
5721                /* MAC address */
5722                netdev->dev_addr);
5723         e_info("Intel(R) PRO/%s Network Connection\n",
5724                (hw->phy.type == e1000_phy_ife) ? "10/100" : "1000");
5725         ret_val = e1000_read_pba_string_generic(hw, pba_str,
5726                                                 E1000_PBANUM_LENGTH);
5727         if (ret_val)
5728                 strncpy((char *)pba_str, "Unknown", sizeof(pba_str) - 1);
5729         e_info("MAC: %d, PHY: %d, PBA No: %s\n",
5730                hw->mac.type, hw->phy.type, pba_str);
5731 }
5732
5733 static void e1000_eeprom_checks(struct e1000_adapter *adapter)
5734 {
5735         struct e1000_hw *hw = &adapter->hw;
5736         int ret_val;
5737         u16 buf = 0;
5738
5739         if (hw->mac.type != e1000_82573)
5740                 return;
5741
5742         ret_val = e1000_read_nvm(hw, NVM_INIT_CONTROL2_REG, 1, &buf);
5743         if (!ret_val && (!(le16_to_cpu(buf) & (1 << 0)))) {
5744                 /* Deep Smart Power Down (DSPD) */
5745                 dev_warn(&adapter->pdev->dev,
5746                          "Warning: detected DSPD enabled in EEPROM\n");
5747         }
5748 }
5749
5750 static const struct net_device_ops e1000e_netdev_ops = {
5751         .ndo_open               = e1000_open,
5752         .ndo_stop               = e1000_close,
5753         .ndo_start_xmit         = e1000_xmit_frame,
5754         .ndo_get_stats64        = e1000e_get_stats64,
5755         .ndo_set_multicast_list = e1000_set_multi,
5756         .ndo_set_mac_address    = e1000_set_mac,
5757         .ndo_change_mtu         = e1000_change_mtu,
5758         .ndo_do_ioctl           = e1000_ioctl,
5759         .ndo_tx_timeout         = e1000_tx_timeout,
5760         .ndo_validate_addr      = eth_validate_addr,
5761
5762         .ndo_vlan_rx_register   = e1000_vlan_rx_register,
5763         .ndo_vlan_rx_add_vid    = e1000_vlan_rx_add_vid,
5764         .ndo_vlan_rx_kill_vid   = e1000_vlan_rx_kill_vid,
5765 #ifdef CONFIG_NET_POLL_CONTROLLER
5766         .ndo_poll_controller    = e1000_netpoll,
5767 #endif
5768 };
5769
5770 /**
5771  * e1000_probe - Device Initialization Routine
5772  * @pdev: PCI device information struct
5773  * @ent: entry in e1000_pci_tbl
5774  *
5775  * Returns 0 on success, negative on failure
5776  *
5777  * e1000_probe initializes an adapter identified by a pci_dev structure.
5778  * The OS initialization, configuring of the adapter private structure,
5779  * and a hardware reset occur.
5780  **/
5781 static int __devinit e1000_probe(struct pci_dev *pdev,
5782                                  const struct pci_device_id *ent)
5783 {
5784         struct net_device *netdev;
5785         struct e1000_adapter *adapter;
5786         struct e1000_hw *hw;
5787         const struct e1000_info *ei = e1000_info_tbl[ent->driver_data];
5788         resource_size_t mmio_start, mmio_len;
5789         resource_size_t flash_start, flash_len;
5790
5791         static int cards_found;
5792         int i, err, pci_using_dac;
5793         u16 eeprom_data = 0;
5794         u16 eeprom_apme_mask = E1000_EEPROM_APME;
5795
5796         if (ei->flags2 & FLAG2_DISABLE_ASPM_L1)
5797                 e1000e_disable_aspm(pdev, PCIE_LINK_STATE_L1);
5798
5799         err = pci_enable_device_mem(pdev);
5800         if (err)
5801                 return err;
5802
5803         pci_using_dac = 0;
5804         err = dma_set_mask(&pdev->dev, DMA_BIT_MASK(64));
5805         if (!err) {
5806                 err = dma_set_coherent_mask(&pdev->dev, DMA_BIT_MASK(64));
5807                 if (!err)
5808                         pci_using_dac = 1;
5809         } else {
5810                 err = dma_set_mask(&pdev->dev, DMA_BIT_MASK(32));
5811                 if (err) {
5812                         err = dma_set_coherent_mask(&pdev->dev,
5813                                                     DMA_BIT_MASK(32));
5814                         if (err) {
5815                                 dev_err(&pdev->dev, "No usable DMA "
5816                                         "configuration, aborting\n");
5817                                 goto err_dma;
5818                         }
5819                 }
5820         }
5821
5822         err = pci_request_selected_regions_exclusive(pdev,
5823                                           pci_select_bars(pdev, IORESOURCE_MEM),
5824                                           e1000e_driver_name);
5825         if (err)
5826                 goto err_pci_reg;
5827
5828         /* AER (Advanced Error Reporting) hooks */
5829         pci_enable_pcie_error_reporting(pdev);
5830
5831         pci_set_master(pdev);
5832         /* PCI config space info */
5833         err = pci_save_state(pdev);
5834         if (err)
5835                 goto err_alloc_etherdev;
5836
5837         err = -ENOMEM;
5838         netdev = alloc_etherdev(sizeof(struct e1000_adapter));
5839         if (!netdev)
5840                 goto err_alloc_etherdev;
5841
5842         SET_NETDEV_DEV(netdev, &pdev->dev);
5843
5844         netdev->irq = pdev->irq;
5845
5846         pci_set_drvdata(pdev, netdev);
5847         adapter = netdev_priv(netdev);
5848         hw = &adapter->hw;
5849         adapter->netdev = netdev;
5850         adapter->pdev = pdev;
5851         adapter->ei = ei;
5852         adapter->pba = ei->pba;
5853         adapter->flags = ei->flags;
5854         adapter->flags2 = ei->flags2;
5855         adapter->hw.adapter = adapter;
5856         adapter->hw.mac.type = ei->mac;
5857         adapter->max_hw_frame_size = ei->max_hw_frame_size;
5858         adapter->msg_enable = (1 << NETIF_MSG_DRV | NETIF_MSG_PROBE) - 1;
5859
5860         mmio_start = pci_resource_start(pdev, 0);
5861         mmio_len = pci_resource_len(pdev, 0);
5862
5863         err = -EIO;
5864         adapter->hw.hw_addr = ioremap(mmio_start, mmio_len);
5865         if (!adapter->hw.hw_addr)
5866                 goto err_ioremap;
5867
5868         if ((adapter->flags & FLAG_HAS_FLASH) &&
5869             (pci_resource_flags(pdev, 1) & IORESOURCE_MEM)) {
5870                 flash_start = pci_resource_start(pdev, 1);
5871                 flash_len = pci_resource_len(pdev, 1);
5872                 adapter->hw.flash_address = ioremap(flash_start, flash_len);
5873                 if (!adapter->hw.flash_address)
5874                         goto err_flashmap;
5875         }
5876
5877         /* construct the net_device struct */
5878         netdev->netdev_ops              = &e1000e_netdev_ops;
5879         e1000e_set_ethtool_ops(netdev);
5880         netdev->watchdog_timeo          = 5 * HZ;
5881         netif_napi_add(netdev, &adapter->napi, e1000_clean, 64);
5882         strncpy(netdev->name, pci_name(pdev), sizeof(netdev->name) - 1);
5883
5884         netdev->mem_start = mmio_start;
5885         netdev->mem_end = mmio_start + mmio_len;
5886
5887         adapter->bd_number = cards_found++;
5888
5889         e1000e_check_options(adapter);
5890
5891         /* setup adapter struct */
5892         err = e1000_sw_init(adapter);
5893         if (err)
5894                 goto err_sw_init;
5895
5896         memcpy(&hw->mac.ops, ei->mac_ops, sizeof(hw->mac.ops));
5897         memcpy(&hw->nvm.ops, ei->nvm_ops, sizeof(hw->nvm.ops));
5898         memcpy(&hw->phy.ops, ei->phy_ops, sizeof(hw->phy.ops));
5899
5900         err = ei->get_variants(adapter);
5901         if (err)
5902                 goto err_hw_init;
5903
5904         if ((adapter->flags & FLAG_IS_ICH) &&
5905             (adapter->flags & FLAG_READ_ONLY_NVM))
5906                 e1000e_write_protect_nvm_ich8lan(&adapter->hw);
5907
5908         hw->mac.ops.get_bus_info(&adapter->hw);
5909
5910         adapter->hw.phy.autoneg_wait_to_complete = 0;
5911
5912         /* Copper options */
5913         if (adapter->hw.phy.media_type == e1000_media_type_copper) {
5914                 adapter->hw.phy.mdix = AUTO_ALL_MODES;
5915                 adapter->hw.phy.disable_polarity_correction = 0;
5916                 adapter->hw.phy.ms_type = e1000_ms_hw_default;
5917         }
5918
5919         if (e1000_check_reset_block(&adapter->hw))
5920                 e_info("PHY reset is blocked due to SOL/IDER session.\n");
5921
5922         netdev->features = NETIF_F_SG |
5923                            NETIF_F_HW_CSUM |
5924                            NETIF_F_HW_VLAN_TX |
5925                            NETIF_F_HW_VLAN_RX;
5926
5927         if (adapter->flags & FLAG_HAS_HW_VLAN_FILTER)
5928                 netdev->features |= NETIF_F_HW_VLAN_FILTER;
5929
5930         netdev->features |= NETIF_F_TSO;
5931         netdev->features |= NETIF_F_TSO6;
5932
5933         netdev->vlan_features |= NETIF_F_TSO;
5934         netdev->vlan_features |= NETIF_F_TSO6;
5935         netdev->vlan_features |= NETIF_F_HW_CSUM;
5936         netdev->vlan_features |= NETIF_F_SG;
5937
5938         if (pci_using_dac) {
5939                 netdev->features |= NETIF_F_HIGHDMA;
5940                 netdev->vlan_features |= NETIF_F_HIGHDMA;
5941         }
5942
5943         if (e1000e_enable_mng_pass_thru(&adapter->hw))
5944                 adapter->flags |= FLAG_MNG_PT_ENABLED;
5945
5946         /*
5947          * before reading the NVM, reset the controller to
5948          * put the device in a known good starting state
5949          */
5950         adapter->hw.mac.ops.reset_hw(&adapter->hw);
5951
5952         /*
5953          * systems with ASPM and others may see the checksum fail on the first
5954          * attempt. Let's give it a few tries
5955          */
5956         for (i = 0;; i++) {
5957                 if (e1000_validate_nvm_checksum(&adapter->hw) >= 0)
5958                         break;
5959                 if (i == 2) {
5960                         e_err("The NVM Checksum Is Not Valid\n");
5961                         err = -EIO;
5962                         goto err_eeprom;
5963                 }
5964         }
5965
5966         e1000_eeprom_checks(adapter);
5967
5968         /* copy the MAC address */
5969         if (e1000e_read_mac_addr(&adapter->hw))
5970                 e_err("NVM Read Error while reading MAC address\n");
5971
5972         memcpy(netdev->dev_addr, adapter->hw.mac.addr, netdev->addr_len);
5973         memcpy(netdev->perm_addr, adapter->hw.mac.addr, netdev->addr_len);
5974
5975         if (!is_valid_ether_addr(netdev->perm_addr)) {
5976                 e_err("Invalid MAC Address: %pM\n", netdev->perm_addr);
5977                 err = -EIO;
5978                 goto err_eeprom;
5979         }
5980
5981         init_timer(&adapter->watchdog_timer);
5982         adapter->watchdog_timer.function = e1000_watchdog;
5983         adapter->watchdog_timer.data = (unsigned long) adapter;
5984
5985         init_timer(&adapter->phy_info_timer);
5986         adapter->phy_info_timer.function = e1000_update_phy_info;
5987         adapter->phy_info_timer.data = (unsigned long) adapter;
5988
5989         INIT_WORK(&adapter->reset_task, e1000_reset_task);
5990         INIT_WORK(&adapter->watchdog_task, e1000_watchdog_task);
5991         INIT_WORK(&adapter->downshift_task, e1000e_downshift_workaround);
5992         INIT_WORK(&adapter->update_phy_task, e1000e_update_phy_task);
5993         INIT_WORK(&adapter->print_hang_task, e1000_print_hw_hang);
5994         INIT_WORK(&adapter->led_blink_task, e1000e_led_blink_task);
5995
5996         /* Initialize link parameters. User can change them with ethtool */
5997         adapter->hw.mac.autoneg = 1;
5998         adapter->fc_autoneg = 1;
5999         adapter->hw.fc.requested_mode = e1000_fc_default;
6000         adapter->hw.fc.current_mode = e1000_fc_default;
6001         adapter->hw.phy.autoneg_advertised = 0x2f;
6002
6003         /* ring size defaults */
6004         adapter->rx_ring->count = 256;
6005         adapter->tx_ring->count = 256;
6006
6007         /*
6008          * Initial Wake on LAN setting - If APM wake is enabled in
6009          * the EEPROM, enable the ACPI Magic Packet filter
6010          */
6011         if (adapter->flags & FLAG_APME_IN_WUC) {
6012                 /* APME bit in EEPROM is mapped to WUC.APME */
6013                 eeprom_data = er32(WUC);
6014                 eeprom_apme_mask = E1000_WUC_APME;
6015                 if ((hw->mac.type > e1000_ich10lan) &&
6016                     (eeprom_data & E1000_WUC_PHY_WAKE))
6017                         adapter->flags2 |= FLAG2_HAS_PHY_WAKEUP;
6018         } else if (adapter->flags & FLAG_APME_IN_CTRL3) {
6019                 if (adapter->flags & FLAG_APME_CHECK_PORT_B &&
6020                     (adapter->hw.bus.func == 1))
6021                         e1000_read_nvm(&adapter->hw,
6022                                 NVM_INIT_CONTROL3_PORT_B, 1, &eeprom_data);
6023                 else
6024                         e1000_read_nvm(&adapter->hw,
6025                                 NVM_INIT_CONTROL3_PORT_A, 1, &eeprom_data);
6026         }
6027
6028         /* fetch WoL from EEPROM */
6029         if (eeprom_data & eeprom_apme_mask)
6030                 adapter->eeprom_wol |= E1000_WUFC_MAG;
6031
6032         /*
6033          * now that we have the eeprom settings, apply the special cases
6034          * where the eeprom may be wrong or the board simply won't support
6035          * wake on lan on a particular port
6036          */
6037         if (!(adapter->flags & FLAG_HAS_WOL))
6038                 adapter->eeprom_wol = 0;
6039
6040         /* initialize the wol settings based on the eeprom settings */
6041         adapter->wol = adapter->eeprom_wol;
6042         device_set_wakeup_enable(&adapter->pdev->dev, adapter->wol);
6043
6044         /* save off EEPROM version number */
6045         e1000_read_nvm(&adapter->hw, 5, 1, &adapter->eeprom_vers);
6046
6047         /* reset the hardware with the new settings */
6048         e1000e_reset(adapter);
6049
6050         /*
6051          * If the controller has AMT, do not set DRV_LOAD until the interface
6052          * is up.  For all other cases, let the f/w know that the h/w is now
6053          * under the control of the driver.
6054          */
6055         if (!(adapter->flags & FLAG_HAS_AMT))
6056                 e1000e_get_hw_control(adapter);
6057
6058         strncpy(netdev->name, "eth%d", sizeof(netdev->name) - 1);
6059         err = register_netdev(netdev);
6060         if (err)
6061                 goto err_register;
6062
6063         /* carrier off reporting is important to ethtool even BEFORE open */
6064         netif_carrier_off(netdev);
6065
6066         e1000_print_device_info(adapter);
6067
6068         if (pci_dev_run_wake(pdev))
6069                 pm_runtime_put_noidle(&pdev->dev);
6070
6071         return 0;
6072
6073 err_register:
6074         if (!(adapter->flags & FLAG_HAS_AMT))
6075                 e1000e_release_hw_control(adapter);
6076 err_eeprom:
6077         if (!e1000_check_reset_block(&adapter->hw))
6078                 e1000_phy_hw_reset(&adapter->hw);
6079 err_hw_init:
6080         kfree(adapter->tx_ring);
6081         kfree(adapter->rx_ring);
6082 err_sw_init:
6083         if (adapter->hw.flash_address)
6084                 iounmap(adapter->hw.flash_address);
6085         e1000e_reset_interrupt_capability(adapter);
6086 err_flashmap:
6087         iounmap(adapter->hw.hw_addr);
6088 err_ioremap:
6089         free_netdev(netdev);
6090 err_alloc_etherdev:
6091         pci_release_selected_regions(pdev,
6092                                      pci_select_bars(pdev, IORESOURCE_MEM));
6093 err_pci_reg:
6094 err_dma:
6095         pci_disable_device(pdev);
6096         return err;
6097 }
6098
6099 /**
6100  * e1000_remove - Device Removal Routine
6101  * @pdev: PCI device information struct
6102  *
6103  * e1000_remove is called by the PCI subsystem to alert the driver
6104  * that it should release a PCI device.  The could be caused by a
6105  * Hot-Plug event, or because the driver is going to be removed from
6106  * memory.
6107  **/
6108 static void __devexit e1000_remove(struct pci_dev *pdev)
6109 {
6110         struct net_device *netdev = pci_get_drvdata(pdev);
6111         struct e1000_adapter *adapter = netdev_priv(netdev);
6112         bool down = test_bit(__E1000_DOWN, &adapter->state);
6113
6114         /*
6115          * The timers may be rescheduled, so explicitly disable them
6116          * from being rescheduled.
6117          */
6118         if (!down)
6119                 set_bit(__E1000_DOWN, &adapter->state);
6120         del_timer_sync(&adapter->watchdog_timer);
6121         del_timer_sync(&adapter->phy_info_timer);
6122
6123         cancel_work_sync(&adapter->reset_task);
6124         cancel_work_sync(&adapter->watchdog_task);
6125         cancel_work_sync(&adapter->downshift_task);
6126         cancel_work_sync(&adapter->update_phy_task);
6127         cancel_work_sync(&adapter->led_blink_task);
6128         cancel_work_sync(&adapter->print_hang_task);
6129
6130         if (!(netdev->flags & IFF_UP))
6131                 e1000_power_down_phy(adapter);
6132
6133         /* Don't lie to e1000_close() down the road. */
6134         if (!down)
6135                 clear_bit(__E1000_DOWN, &adapter->state);
6136         unregister_netdev(netdev);
6137
6138         if (pci_dev_run_wake(pdev))
6139                 pm_runtime_get_noresume(&pdev->dev);
6140
6141         /*
6142          * Release control of h/w to f/w.  If f/w is AMT enabled, this
6143          * would have already happened in close and is redundant.
6144          */
6145         e1000e_release_hw_control(adapter);
6146
6147         e1000e_reset_interrupt_capability(adapter);
6148         kfree(adapter->tx_ring);
6149         kfree(adapter->rx_ring);
6150
6151         iounmap(adapter->hw.hw_addr);
6152         if (adapter->hw.flash_address)
6153                 iounmap(adapter->hw.flash_address);
6154         pci_release_selected_regions(pdev,
6155                                      pci_select_bars(pdev, IORESOURCE_MEM));
6156
6157         free_netdev(netdev);
6158
6159         /* AER disable */
6160         pci_disable_pcie_error_reporting(pdev);
6161
6162         pci_disable_device(pdev);
6163 }
6164
6165 /* PCI Error Recovery (ERS) */
6166 static struct pci_error_handlers e1000_err_handler = {
6167         .error_detected = e1000_io_error_detected,
6168         .slot_reset = e1000_io_slot_reset,
6169         .resume = e1000_io_resume,
6170 };
6171
6172 static DEFINE_PCI_DEVICE_TABLE(e1000_pci_tbl) = {
6173         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82571EB_COPPER), board_82571 },
6174         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82571EB_FIBER), board_82571 },
6175         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82571EB_QUAD_COPPER), board_82571 },
6176         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82571EB_QUAD_COPPER_LP), board_82571 },
6177         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82571EB_QUAD_FIBER), board_82571 },
6178         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82571EB_SERDES), board_82571 },
6179         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82571EB_SERDES_DUAL), board_82571 },
6180         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82571EB_SERDES_QUAD), board_82571 },
6181         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82571PT_QUAD_COPPER), board_82571 },
6182
6183         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82572EI), board_82572 },
6184         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82572EI_COPPER), board_82572 },
6185         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82572EI_FIBER), board_82572 },
6186         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82572EI_SERDES), board_82572 },
6187
6188         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82573E), board_82573 },
6189         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82573E_IAMT), board_82573 },
6190         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82573L), board_82573 },
6191
6192         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82574L), board_82574 },
6193         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82574LA), board_82574 },
6194         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82583V), board_82583 },
6195
6196         { PCI_VDEVICE(INTEL, E1000_DEV_ID_80003ES2LAN_COPPER_DPT),
6197           board_80003es2lan },
6198         { PCI_VDEVICE(INTEL, E1000_DEV_ID_80003ES2LAN_COPPER_SPT),
6199           board_80003es2lan },
6200         { PCI_VDEVICE(INTEL, E1000_DEV_ID_80003ES2LAN_SERDES_DPT),
6201           board_80003es2lan },
6202         { PCI_VDEVICE(INTEL, E1000_DEV_ID_80003ES2LAN_SERDES_SPT),
6203           board_80003es2lan },
6204
6205         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH8_IFE), board_ich8lan },
6206         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH8_IFE_G), board_ich8lan },
6207         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH8_IFE_GT), board_ich8lan },
6208         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH8_IGP_AMT), board_ich8lan },
6209         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH8_IGP_C), board_ich8lan },
6210         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH8_IGP_M), board_ich8lan },
6211         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH8_IGP_M_AMT), board_ich8lan },
6212         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH8_82567V_3), board_ich8lan },
6213
6214         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH9_IFE), board_ich9lan },
6215         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH9_IFE_G), board_ich9lan },
6216         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH9_IFE_GT), board_ich9lan },
6217         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH9_IGP_AMT), board_ich9lan },
6218         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH9_IGP_C), board_ich9lan },
6219         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH9_BM), board_ich9lan },
6220         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH9_IGP_M), board_ich9lan },
6221         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH9_IGP_M_AMT), board_ich9lan },
6222         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH9_IGP_M_V), board_ich9lan },
6223
6224         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH10_R_BM_LM), board_ich9lan },
6225         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH10_R_BM_LF), board_ich9lan },
6226         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH10_R_BM_V), board_ich9lan },
6227
6228         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH10_D_BM_LM), board_ich10lan },
6229         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH10_D_BM_LF), board_ich10lan },
6230         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH10_D_BM_V), board_ich10lan },
6231
6232         { PCI_VDEVICE(INTEL, E1000_DEV_ID_PCH_M_HV_LM), board_pchlan },
6233         { PCI_VDEVICE(INTEL, E1000_DEV_ID_PCH_M_HV_LC), board_pchlan },
6234         { PCI_VDEVICE(INTEL, E1000_DEV_ID_PCH_D_HV_DM), board_pchlan },
6235         { PCI_VDEVICE(INTEL, E1000_DEV_ID_PCH_D_HV_DC), board_pchlan },
6236
6237         { PCI_VDEVICE(INTEL, E1000_DEV_ID_PCH2_LV_LM), board_pch2lan },
6238         { PCI_VDEVICE(INTEL, E1000_DEV_ID_PCH2_LV_V), board_pch2lan },
6239
6240         { }     /* terminate list */
6241 };
6242 MODULE_DEVICE_TABLE(pci, e1000_pci_tbl);
6243
6244 #ifdef CONFIG_PM
6245 static const struct dev_pm_ops e1000_pm_ops = {
6246         SET_SYSTEM_SLEEP_PM_OPS(e1000_suspend, e1000_resume)
6247         SET_RUNTIME_PM_OPS(e1000_runtime_suspend,
6248                                 e1000_runtime_resume, e1000_idle)
6249 };
6250 #endif
6251
6252 /* PCI Device API Driver */
6253 static struct pci_driver e1000_driver = {
6254         .name     = e1000e_driver_name,
6255         .id_table = e1000_pci_tbl,
6256         .probe    = e1000_probe,
6257         .remove   = __devexit_p(e1000_remove),
6258 #ifdef CONFIG_PM
6259         .driver.pm = &e1000_pm_ops,
6260 #endif
6261         .shutdown = e1000_shutdown,
6262         .err_handler = &e1000_err_handler
6263 };
6264
6265 /**
6266  * e1000_init_module - Driver Registration Routine
6267  *
6268  * e1000_init_module is the first routine called when the driver is
6269  * loaded. All it does is register with the PCI subsystem.
6270  **/
6271 static int __init e1000_init_module(void)
6272 {
6273         int ret;
6274         pr_info("Intel(R) PRO/1000 Network Driver - %s\n",
6275                 e1000e_driver_version);
6276         pr_info("Copyright(c) 1999 - 2011 Intel Corporation.\n");
6277         ret = pci_register_driver(&e1000_driver);
6278
6279         return ret;
6280 }
6281 module_init(e1000_init_module);
6282
6283 /**
6284  * e1000_exit_module - Driver Exit Cleanup Routine
6285  *
6286  * e1000_exit_module is called just before the driver is removed
6287  * from memory.
6288  **/
6289 static void __exit e1000_exit_module(void)
6290 {
6291         pci_unregister_driver(&e1000_driver);
6292 }
6293 module_exit(e1000_exit_module);
6294
6295
6296 MODULE_AUTHOR("Intel Corporation, <linux.nics@intel.com>");
6297 MODULE_DESCRIPTION("Intel(R) PRO/1000 Network Driver");
6298 MODULE_LICENSE("GPL");
6299 MODULE_VERSION(DRV_VERSION);
6300
6301 /* e1000_main.c */