Pileus Git - ~andy/linux/blob - drivers/net/e1000e/netdev.c

   1 /*******************************************************************************
   2
   3   Intel PRO/1000 Linux driver
   4   Copyright(c) 1999 - 2011 Intel Corporation.
   5
   6   This program is free software; you can redistribute it and/or modify it
   7   under the terms and conditions of the GNU General Public License,
   8   version 2, as published by the Free Software Foundation.
   9
  10   This program is distributed in the hope it will be useful, but WITHOUT
  11   ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
  12   FITNESS FOR A PARTICULAR PURPOSE.  See the GNU General Public License for
  13   more details.
  14
  15   You should have received a copy of the GNU General Public License along with
  16   this program; if not, write to the Free Software Foundation, Inc.,
  17   51 Franklin St - Fifth Floor, Boston, MA 02110-1301 USA.
  18
  19   The full GNU General Public License is included in this distribution in
  20   the file called "COPYING".
  21
  22   Contact Information:
  23   Linux NICS <linux.nics@intel.com>
  24   e1000-devel Mailing List <e1000-devel@lists.sourceforge.net>
  25   Intel Corporation, 5200 N.E. Elam Young Parkway, Hillsboro, OR 97124-6497
  26
  27 *******************************************************************************/
  28
  29 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
  30
  31 #include <linux/module.h>
  32 #include <linux/types.h>
  33 #include <linux/init.h>
  34 #include <linux/pci.h>
  35 #include <linux/vmalloc.h>
  36 #include <linux/pagemap.h>
  37 #include <linux/delay.h>
  38 #include <linux/netdevice.h>
  39 #include <linux/tcp.h>
  40 #include <linux/ipv6.h>
  41 #include <linux/slab.h>
  42 #include <net/checksum.h>
  43 #include <net/ip6_checksum.h>
  44 #include <linux/mii.h>
  45 #include <linux/ethtool.h>
  46 #include <linux/if_vlan.h>
  47 #include <linux/cpu.h>
  48 #include <linux/smp.h>
  49 #include <linux/pm_qos_params.h>
  50 #include <linux/pm_runtime.h>
  51 #include <linux/aer.h>
  52
  53 #include "e1000.h"
  54
  55 #define DRV_EXTRAVERSION "-k2"
  56
  57 #define DRV_VERSION "1.2.20" DRV_EXTRAVERSION
  58 char e1000e_driver_name[] = "e1000e";
  59 const char e1000e_driver_version[] = DRV_VERSION;
  60
  61 static const struct e1000_info *e1000_info_tbl[] = {
  62         [board_82571]           = &e1000_82571_info,
  63         [board_82572]           = &e1000_82572_info,
  64         [board_82573]           = &e1000_82573_info,
  65         [board_82574]           = &e1000_82574_info,
  66         [board_82583]           = &e1000_82583_info,
  67         [board_80003es2lan]     = &e1000_es2_info,
  68         [board_ich8lan]         = &e1000_ich8_info,
  69         [board_ich9lan]         = &e1000_ich9_info,
  70         [board_ich10lan]        = &e1000_ich10_info,
  71         [board_pchlan]          = &e1000_pch_info,
  72         [board_pch2lan]         = &e1000_pch2_info,
  73 };
  74
  75 struct e1000_reg_info {
  76         u32 ofs;
  77         char *name;
  78 };
  79
  80 #define E1000_RDFH      0x02410 /* Rx Data FIFO Head - RW */
  81 #define E1000_RDFT      0x02418 /* Rx Data FIFO Tail - RW */
  82 #define E1000_RDFHS     0x02420 /* Rx Data FIFO Head Saved - RW */
  83 #define E1000_RDFTS     0x02428 /* Rx Data FIFO Tail Saved - RW */
  84 #define E1000_RDFPC     0x02430 /* Rx Data FIFO Packet Count - RW */
  85
  86 #define E1000_TDFH      0x03410 /* Tx Data FIFO Head - RW */
  87 #define E1000_TDFT      0x03418 /* Tx Data FIFO Tail - RW */
  88 #define E1000_TDFHS     0x03420 /* Tx Data FIFO Head Saved - RW */
  89 #define E1000_TDFTS     0x03428 /* Tx Data FIFO Tail Saved - RW */
  90 #define E1000_TDFPC     0x03430 /* Tx Data FIFO Packet Count - RW */
  91
  92 static const struct e1000_reg_info e1000_reg_info_tbl[] = {
  93
  94         /* General Registers */
  95         {E1000_CTRL, "CTRL"},
  96         {E1000_STATUS, "STATUS"},
  97         {E1000_CTRL_EXT, "CTRL_EXT"},
  98
  99         /* Interrupt Registers */
 100         {E1000_ICR, "ICR"},
 101
 102         /* Rx Registers */
 103         {E1000_RCTL, "RCTL"},
 104         {E1000_RDLEN, "RDLEN"},
 105         {E1000_RDH, "RDH"},
 106         {E1000_RDT, "RDT"},
 107         {E1000_RDTR, "RDTR"},
 108         {E1000_RXDCTL(0), "RXDCTL"},
 109         {E1000_ERT, "ERT"},
 110         {E1000_RDBAL, "RDBAL"},
 111         {E1000_RDBAH, "RDBAH"},
 112         {E1000_RDFH, "RDFH"},
 113         {E1000_RDFT, "RDFT"},
 114         {E1000_RDFHS, "RDFHS"},
 115         {E1000_RDFTS, "RDFTS"},
 116         {E1000_RDFPC, "RDFPC"},
 117
 118         /* Tx Registers */
 119         {E1000_TCTL, "TCTL"},
 120         {E1000_TDBAL, "TDBAL"},
 121         {E1000_TDBAH, "TDBAH"},
 122         {E1000_TDLEN, "TDLEN"},
 123         {E1000_TDH, "TDH"},
 124         {E1000_TDT, "TDT"},
 125         {E1000_TIDV, "TIDV"},
 126         {E1000_TXDCTL(0), "TXDCTL"},
 127         {E1000_TADV, "TADV"},
 128         {E1000_TARC(0), "TARC"},
 129         {E1000_TDFH, "TDFH"},
 130         {E1000_TDFT, "TDFT"},
 131         {E1000_TDFHS, "TDFHS"},
 132         {E1000_TDFTS, "TDFTS"},
 133         {E1000_TDFPC, "TDFPC"},
 134
 135         /* List Terminator */
 136         {}
 137 };
 138
 139 /*
 140  * e1000_regdump - register printout routine
 141  */
 142 static void e1000_regdump(struct e1000_hw *hw, struct e1000_reg_info *reginfo)
 143 {
 144         int n = 0;
 145         char rname[16];
 146         u32 regs[8];
 147
 148         switch (reginfo->ofs) {
 149         case E1000_RXDCTL(0):
 150                 for (n = 0; n < 2; n++)
 151                         regs[n] = __er32(hw, E1000_RXDCTL(n));
 152                 break;
 153         case E1000_TXDCTL(0):
 154                 for (n = 0; n < 2; n++)
 155                         regs[n] = __er32(hw, E1000_TXDCTL(n));
 156                 break;
 157         case E1000_TARC(0):
 158                 for (n = 0; n < 2; n++)
 159                         regs[n] = __er32(hw, E1000_TARC(n));
 160                 break;
 161         default:
 162                 printk(KERN_INFO "%-15s %08x\n",
 163                        reginfo->name, __er32(hw, reginfo->ofs));
 164                 return;
 165         }
 166
 167         snprintf(rname, 16, "%s%s", reginfo->name, "[0-1]");
 168         printk(KERN_INFO "%-15s ", rname);
 169         for (n = 0; n < 2; n++)
 170                 printk(KERN_CONT "%08x ", regs[n]);
 171         printk(KERN_CONT "\n");
 172 }
 173
 174 /*
 175  * e1000e_dump - Print registers, Tx-ring and Rx-ring
 176  */
 177 static void e1000e_dump(struct e1000_adapter *adapter)
 178 {
 179         struct net_device *netdev = adapter->netdev;
 180         struct e1000_hw *hw = &adapter->hw;
 181         struct e1000_reg_info *reginfo;
 182         struct e1000_ring *tx_ring = adapter->tx_ring;
 183         struct e1000_tx_desc *tx_desc;
 184         struct my_u0 {
 185                 u64 a;
 186                 u64 b;
 187         } *u0;
 188         struct e1000_buffer *buffer_info;
 189         struct e1000_ring *rx_ring = adapter->rx_ring;
 190         union e1000_rx_desc_packet_split *rx_desc_ps;
 191         struct e1000_rx_desc *rx_desc;
 192         struct my_u1 {
 193                 u64 a;
 194                 u64 b;
 195                 u64 c;
 196                 u64 d;
 197         } *u1;
 198         u32 staterr;
 199         int i = 0;
 200
 201         if (!netif_msg_hw(adapter))
 202                 return;
 203
 204         /* Print netdevice Info */
 205         if (netdev) {
 206                 dev_info(&adapter->pdev->dev, "Net device Info\n");
 207                 printk(KERN_INFO "Device Name     state            "
 208                        "trans_start      last_rx\n");
 209                 printk(KERN_INFO "%-15s %016lX %016lX %016lX\n",
 210                        netdev->name, netdev->state, netdev->trans_start,
 211                        netdev->last_rx);
 212         }
 213
 214         /* Print Registers */
 215         dev_info(&adapter->pdev->dev, "Register Dump\n");
 216         printk(KERN_INFO " Register Name   Value\n");
 217         for (reginfo = (struct e1000_reg_info *)e1000_reg_info_tbl;
 218              reginfo->name; reginfo++) {
 219                 e1000_regdump(hw, reginfo);
 220         }
 221
 222         /* Print Tx Ring Summary */
 223         if (!netdev || !netif_running(netdev))
 224                 goto exit;
 225
 226         dev_info(&adapter->pdev->dev, "Tx Ring Summary\n");
 227         printk(KERN_INFO "Queue [NTU] [NTC] [bi(ntc)->dma  ]"
 228                " leng ntw timestamp\n");
 229         buffer_info = &tx_ring->buffer_info[tx_ring->next_to_clean];
 230         printk(KERN_INFO " %5d %5X %5X %016llX %04X %3X %016llX\n",
 231                0, tx_ring->next_to_use, tx_ring->next_to_clean,
 232                (unsigned long long)buffer_info->dma,
 233                buffer_info->length,
 234                buffer_info->next_to_watch,
 235                (unsigned long long)buffer_info->time_stamp);
 236
 237         /* Print Tx Ring */
 238         if (!netif_msg_tx_done(adapter))
 239                 goto rx_ring_summary;
 240
 241         dev_info(&adapter->pdev->dev, "Tx Ring Dump\n");
 242
 243         /* Transmit Descriptor Formats - DEXT[29] is 0 (Legacy) or 1 (Extended)
 244          *
 245          * Legacy Transmit Descriptor
 246          *   +--------------------------------------------------------------+
 247          * 0 |         Buffer Address [63:0] (Reserved on Write Back)       |
 248          *   +--------------------------------------------------------------+
 249          * 8 | Special  |    CSS     | Status |  CMD    |  CSO   |  Length  |
 250          *   +--------------------------------------------------------------+
 251          *   63       48 47        36 35    32 31     24 23    16 15        0
 252          *
 253          * Extended Context Descriptor (DTYP=0x0) for TSO or checksum offload
 254          *   63      48 47    40 39       32 31             16 15    8 7      0
 255          *   +----------------------------------------------------------------+
 256          * 0 |  TUCSE  | TUCS0  |   TUCSS   |     IPCSE       | IPCS0 | IPCSS |
 257          *   +----------------------------------------------------------------+
 258          * 8 |   MSS   | HDRLEN | RSV | STA | TUCMD | DTYP |      PAYLEN      |
 259          *   +----------------------------------------------------------------+
 260          *   63      48 47    40 39 36 35 32 31   24 23  20 19                0
 261          *
 262          * Extended Data Descriptor (DTYP=0x1)
 263          *   +----------------------------------------------------------------+
 264          * 0 |                     Buffer Address [63:0]                      |
 265          *   +----------------------------------------------------------------+
 266          * 8 | VLAN tag |  POPTS  | Rsvd | Status | Command | DTYP |  DTALEN  |
 267          *   +----------------------------------------------------------------+
 268          *   63       48 47     40 39  36 35    32 31     24 23  20 19        0
 269          */
 270         printk(KERN_INFO "Tl[desc]     [address 63:0  ] [SpeCssSCmCsLen]"
 271                " [bi->dma       ] leng  ntw timestamp        bi->skb "
 272                "<-- Legacy format\n");
 273         printk(KERN_INFO "Tc[desc]     [Ce CoCsIpceCoS] [MssHlRSCm0Plen]"
 274                " [bi->dma       ] leng  ntw timestamp        bi->skb "
 275                "<-- Ext Context format\n");
 276         printk(KERN_INFO "Td[desc]     [address 63:0  ] [VlaPoRSCm1Dlen]"
 277                " [bi->dma       ] leng  ntw timestamp        bi->skb "
 278                "<-- Ext Data format\n");
 279         for (i = 0; tx_ring->desc && (i < tx_ring->count); i++) {
 280                 tx_desc = E1000_TX_DESC(*tx_ring, i);
 281                 buffer_info = &tx_ring->buffer_info[i];
 282                 u0 = (struct my_u0 *)tx_desc;
 283                 printk(KERN_INFO "T%c[0x%03X]    %016llX %016llX %016llX "
 284                        "%04X  %3X %016llX %p",
 285                        (!(le64_to_cpu(u0->b) & (1 << 29)) ? 'l' :
 286                         ((le64_to_cpu(u0->b) & (1 << 20)) ? 'd' : 'c')), i,
 287                        (unsigned long long)le64_to_cpu(u0->a),
 288                        (unsigned long long)le64_to_cpu(u0->b),
 289                        (unsigned long long)buffer_info->dma,
 290                        buffer_info->length, buffer_info->next_to_watch,
 291                        (unsigned long long)buffer_info->time_stamp,
 292                        buffer_info->skb);
 293                 if (i == tx_ring->next_to_use && i == tx_ring->next_to_clean)
 294                         printk(KERN_CONT " NTC/U\n");
 295                 else if (i == tx_ring->next_to_use)
 296                         printk(KERN_CONT " NTU\n");
 297                 else if (i == tx_ring->next_to_clean)
 298                         printk(KERN_CONT " NTC\n");
 299                 else
 300                         printk(KERN_CONT "\n");
 301
 302                 if (netif_msg_pktdata(adapter) && buffer_info->dma != 0)
 303                         print_hex_dump(KERN_INFO, "", DUMP_PREFIX_ADDRESS,
 304                                        16, 1, phys_to_virt(buffer_info->dma),
 305                                        buffer_info->length, true);
 306         }
 307
 308         /* Print Rx Ring Summary */
 309 rx_ring_summary:
 310         dev_info(&adapter->pdev->dev, "Rx Ring Summary\n");
 311         printk(KERN_INFO "Queue [NTU] [NTC]\n");
 312         printk(KERN_INFO " %5d %5X %5X\n", 0,
 313                rx_ring->next_to_use, rx_ring->next_to_clean);
 314
 315         /* Print Rx Ring */
 316         if (!netif_msg_rx_status(adapter))
 317                 goto exit;
 318
 319         dev_info(&adapter->pdev->dev, "Rx Ring Dump\n");
 320         switch (adapter->rx_ps_pages) {
 321         case 1:
 322         case 2:
 323         case 3:
 324                 /* [Extended] Packet Split Receive Descriptor Format
 325                  *
 326                  *    +-----------------------------------------------------+
 327                  *  0 |                Buffer Address 0 [63:0]              |
 328                  *    +-----------------------------------------------------+
 329                  *  8 |                Buffer Address 1 [63:0]              |
 330                  *    +-----------------------------------------------------+
 331                  * 16 |                Buffer Address 2 [63:0]              |
 332                  *    +-----------------------------------------------------+
 333                  * 24 |                Buffer Address 3 [63:0]              |
 334                  *    +-----------------------------------------------------+
 335                  */
 336                 printk(KERN_INFO "R  [desc]      [buffer 0 63:0 ] "
 337                        "[buffer 1 63:0 ] "
 338                        "[buffer 2 63:0 ] [buffer 3 63:0 ] [bi->dma       ] "
 339                        "[bi->skb] <-- Ext Pkt Split format\n");
 340                 /* [Extended] Receive Descriptor (Write-Back) Format
 341                  *
 342                  *   63       48 47    32 31     13 12    8 7    4 3        0
 343                  *   +------------------------------------------------------+
 344                  * 0 | Packet   | IP     |  Rsvd   | MRQ   | Rsvd | MRQ RSS |
 345                  *   | Checksum | Ident  |         | Queue |      |  Type   |
 346                  *   +------------------------------------------------------+
 347                  * 8 | VLAN Tag | Length | Extended Error | Extended Status |
 348                  *   +------------------------------------------------------+
 349                  *   63       48 47    32 31            20 19               0
 350                  */
 351                 printk(KERN_INFO "RWB[desc]      [ck ipid mrqhsh] "
 352                        "[vl   l0 ee  es] "
 353                        "[ l3  l2  l1 hs] [reserved      ] ---------------- "
 354                        "[bi->skb] <-- Ext Rx Write-Back format\n");
 355                 for (i = 0; i < rx_ring->count; i++) {
 356                         buffer_info = &rx_ring->buffer_info[i];
 357                         rx_desc_ps = E1000_RX_DESC_PS(*rx_ring, i);
 358                         u1 = (struct my_u1 *)rx_desc_ps;
 359                         staterr =
 360                             le32_to_cpu(rx_desc_ps->wb.middle.status_error);
 361                         if (staterr & E1000_RXD_STAT_DD) {
 362                                 /* Descriptor Done */
 363                                 printk(KERN_INFO "RWB[0x%03X]     %016llX "
 364                                        "%016llX %016llX %016llX "
 365                                        "---------------- %p", i,
 366                                        (unsigned long long)le64_to_cpu(u1->a),
 367                                        (unsigned long long)le64_to_cpu(u1->b),
 368                                        (unsigned long long)le64_to_cpu(u1->c),
 369                                        (unsigned long long)le64_to_cpu(u1->d),
 370                                        buffer_info->skb);
 371                         } else {
 372                                 printk(KERN_INFO "R  [0x%03X]     %016llX "
 373                                        "%016llX %016llX %016llX %016llX %p", i,
 374                                        (unsigned long long)le64_to_cpu(u1->a),
 375                                        (unsigned long long)le64_to_cpu(u1->b),
 376                                        (unsigned long long)le64_to_cpu(u1->c),
 377                                        (unsigned long long)le64_to_cpu(u1->d),
 378                                        (unsigned long long)buffer_info->dma,
 379                                        buffer_info->skb);
 380
 381                                 if (netif_msg_pktdata(adapter))
 382                                         print_hex_dump(KERN_INFO, "",
 383                                                 DUMP_PREFIX_ADDRESS, 16, 1,
 384                                                 phys_to_virt(buffer_info->dma),
 385                                                 adapter->rx_ps_bsize0, true);
 386                         }
 387
 388                         if (i == rx_ring->next_to_use)
 389                                 printk(KERN_CONT " NTU\n");
 390                         else if (i == rx_ring->next_to_clean)
 391                                 printk(KERN_CONT " NTC\n");
 392                         else
 393                                 printk(KERN_CONT "\n");
 394                 }
 395                 break;
 396         default:
 397         case 0:
 398                 /* Legacy Receive Descriptor Format
 399                  *
 400                  * +-----------------------------------------------------+
 401                  * |                Buffer Address [63:0]                |
 402                  * +-----------------------------------------------------+
 403                  * | VLAN Tag | Errors | Status 0 | Packet csum | Length |
 404                  * +-----------------------------------------------------+
 405                  * 63       48 47    40 39      32 31         16 15      0
 406                  */
 407                 printk(KERN_INFO "Rl[desc]     [address 63:0  ] "
 408                        "[vl er S cks ln] [bi->dma       ] [bi->skb] "
 409                        "<-- Legacy format\n");
 410                 for (i = 0; rx_ring->desc && (i < rx_ring->count); i++) {
 411                         rx_desc = E1000_RX_DESC(*rx_ring, i);
 412                         buffer_info = &rx_ring->buffer_info[i];
 413                         u0 = (struct my_u0 *)rx_desc;
 414                         printk(KERN_INFO "Rl[0x%03X]    %016llX %016llX "
 415                                "%016llX %p", i,
 416                                (unsigned long long)le64_to_cpu(u0->a),
 417                                (unsigned long long)le64_to_cpu(u0->b),
 418                                (unsigned long long)buffer_info->dma,
 419                                buffer_info->skb);
 420                         if (i == rx_ring->next_to_use)
 421                                 printk(KERN_CONT " NTU\n");
 422                         else if (i == rx_ring->next_to_clean)
 423                                 printk(KERN_CONT " NTC\n");
 424                         else
 425                                 printk(KERN_CONT "\n");
 426
 427                         if (netif_msg_pktdata(adapter))
 428                                 print_hex_dump(KERN_INFO, "",
 429                                                DUMP_PREFIX_ADDRESS,
 430                                                16, 1,
 431                                                phys_to_virt(buffer_info->dma),
 432                                                adapter->rx_buffer_len, true);
 433                 }
 434         }
 435
 436 exit:
 437         return;
 438 }
 439
 440 /**
 441  * e1000_desc_unused - calculate if we have unused descriptors
 442  **/
 443 static int e1000_desc_unused(struct e1000_ring *ring)
 444 {
 445         if (ring->next_to_clean > ring->next_to_use)
 446                 return ring->next_to_clean - ring->next_to_use - 1;
 447
 448         return ring->count + ring->next_to_clean - ring->next_to_use - 1;
 449 }
 450
 451 /**
 452  * e1000_receive_skb - helper function to handle Rx indications
 453  * @adapter: board private structure
 454  * @status: descriptor status field as written by hardware
 455  * @vlan: descriptor vlan field as written by hardware (no le/be conversion)
 456  * @skb: pointer to sk_buff to be indicated to stack
 457  **/
 458 static void e1000_receive_skb(struct e1000_adapter *adapter,
 459                               struct net_device *netdev, struct sk_buff *skb,
 460                               u8 status, __le16 vlan)
 461 {
 462         skb->protocol = eth_type_trans(skb, netdev);
 463
 464         if (adapter->vlgrp && (status & E1000_RXD_STAT_VP))
 465                 vlan_gro_receive(&adapter->napi, adapter->vlgrp,
 466                                  le16_to_cpu(vlan), skb);
 467         else
 468                 napi_gro_receive(&adapter->napi, skb);
 469 }
 470
 471 /**
 472  * e1000_rx_checksum - Receive Checksum Offload
 473  * @adapter:     board private structure
 474  * @status_err:  receive descriptor status and error fields
 475  * @csum:       receive descriptor csum field
 476  * @sk_buff:     socket buffer with received data
 477  **/
 478 static void e1000_rx_checksum(struct e1000_adapter *adapter, u32 status_err,
 479                               u32 csum, struct sk_buff *skb)
 480 {
 481         u16 status = (u16)status_err;
 482         u8 errors = (u8)(status_err >> 24);
 483
 484         skb_checksum_none_assert(skb);
 485
 486         /* Ignore Checksum bit is set */
 487         if (status & E1000_RXD_STAT_IXSM)
 488                 return;
 489         /* TCP/UDP checksum error bit is set */
 490         if (errors & E1000_RXD_ERR_TCPE) {
 491                 /* let the stack verify checksum errors */
 492                 adapter->hw_csum_err++;
 493                 return;
 494         }
 495
 496         /* TCP/UDP Checksum has not been calculated */
 497         if (!(status & (E1000_RXD_STAT_TCPCS | E1000_RXD_STAT_UDPCS)))
 498                 return;
 499
 500         /* It must be a TCP or UDP packet with a valid checksum */
 501         if (status & E1000_RXD_STAT_TCPCS) {
 502                 /* TCP checksum is good */
 503                 skb->ip_summed = CHECKSUM_UNNECESSARY;
 504         } else {
 505                 /*
 506                  * IP fragment with UDP payload
 507                  * Hardware complements the payload checksum, so we undo it
 508                  * and then put the value in host order for further stack use.
 509                  */
 510                 __sum16 sum = (__force __sum16)htons(csum);
 511                 skb->csum = csum_unfold(~sum);
 512                 skb->ip_summed = CHECKSUM_COMPLETE;
 513         }
 514         adapter->hw_csum_good++;
 515 }
 516
 517 /**
 518  * e1000_alloc_rx_buffers - Replace used receive buffers; legacy & extended
 519  * @adapter: address of board private structure
 520  **/
 521 static void e1000_alloc_rx_buffers(struct e1000_adapter *adapter,
 522                                    int cleaned_count)
 523 {
 524         struct net_device *netdev = adapter->netdev;
 525         struct pci_dev *pdev = adapter->pdev;
 526         struct e1000_ring *rx_ring = adapter->rx_ring;
 527         struct e1000_rx_desc *rx_desc;
 528         struct e1000_buffer *buffer_info;
 529         struct sk_buff *skb;
 530         unsigned int i;
 531         unsigned int bufsz = adapter->rx_buffer_len;
 532
 533         i = rx_ring->next_to_use;
 534         buffer_info = &rx_ring->buffer_info[i];
 535
 536         while (cleaned_count--) {
 537                 skb = buffer_info->skb;
 538                 if (skb) {
 539                         skb_trim(skb, 0);
 540                         goto map_skb;
 541                 }
 542
 543                 skb = netdev_alloc_skb_ip_align(netdev, bufsz);
 544                 if (!skb) {
 545                         /* Better luck next round */
 546                         adapter->alloc_rx_buff_failed++;
 547                         break;
 548                 }
 549
 550                 buffer_info->skb = skb;
 551 map_skb:
 552                 buffer_info->dma = dma_map_single(&pdev->dev, skb->data,
 553                                                   adapter->rx_buffer_len,
 554                                                   DMA_FROM_DEVICE);
 555                 if (dma_mapping_error(&pdev->dev, buffer_info->dma)) {
 556                         dev_err(&pdev->dev, "Rx DMA map failed\n");
 557                         adapter->rx_dma_failed++;
 558                         break;
 559                 }
 560
 561                 rx_desc = E1000_RX_DESC(*rx_ring, i);
 562                 rx_desc->buffer_addr = cpu_to_le64(buffer_info->dma);
 563
 564                 if (unlikely(!(i & (E1000_RX_BUFFER_WRITE - 1)))) {
 565                         /*
 566                          * Force memory writes to complete before letting h/w
 567                          * know there are new descriptors to fetch.  (Only
 568                          * applicable for weak-ordered memory model archs,
 569                          * such as IA-64).
 570                          */
 571                         wmb();
 572                         writel(i, adapter->hw.hw_addr + rx_ring->tail);
 573                 }
 574                 i++;
 575                 if (i == rx_ring->count)
 576                         i = 0;
 577                 buffer_info = &rx_ring->buffer_info[i];
 578         }
 579
 580         rx_ring->next_to_use = i;
 581 }
 582
 583 /**
 584  * e1000_alloc_rx_buffers_ps - Replace used receive buffers; packet split
 585  * @adapter: address of board private structure
 586  **/
 587 static void e1000_alloc_rx_buffers_ps(struct e1000_adapter *adapter,
 588                                       int cleaned_count)
 589 {
 590         struct net_device *netdev = adapter->netdev;
 591         struct pci_dev *pdev = adapter->pdev;
 592         union e1000_rx_desc_packet_split *rx_desc;
 593         struct e1000_ring *rx_ring = adapter->rx_ring;
 594         struct e1000_buffer *buffer_info;
 595         struct e1000_ps_page *ps_page;
 596         struct sk_buff *skb;
 597         unsigned int i, j;
 598
 599         i = rx_ring->next_to_use;
 600         buffer_info = &rx_ring->buffer_info[i];
 601
 602         while (cleaned_count--) {
 603                 rx_desc = E1000_RX_DESC_PS(*rx_ring, i);
 604
 605                 for (j = 0; j < PS_PAGE_BUFFERS; j++) {
 606                         ps_page = &buffer_info->ps_pages[j];
 607                         if (j >= adapter->rx_ps_pages) {
 608                                 /* all unused desc entries get hw null ptr */
 609                                 rx_desc->read.buffer_addr[j + 1] =
 610                                     ~cpu_to_le64(0);
 611                                 continue;
 612                         }
 613                         if (!ps_page->page) {
 614                                 ps_page->page = alloc_page(GFP_ATOMIC);
 615                                 if (!ps_page->page) {
 616                                         adapter->alloc_rx_buff_failed++;
 617                                         goto no_buffers;
 618                                 }
 619                                 ps_page->dma = dma_map_page(&pdev->dev,
 620                                                             ps_page->page,
 621                                                             0, PAGE_SIZE,
 622                                                             DMA_FROM_DEVICE);
 623                                 if (dma_mapping_error(&pdev->dev,
 624                                                       ps_page->dma)) {
 625                                         dev_err(&adapter->pdev->dev,
 626                                                 "Rx DMA page map failed\n");
 627                                         adapter->rx_dma_failed++;
 628                                         goto no_buffers;
 629                                 }
 630                         }
 631                         /*
 632                          * Refresh the desc even if buffer_addrs
 633                          * didn't change because each write-back
 634                          * erases this info.
 635                          */
 636                         rx_desc->read.buffer_addr[j + 1] =
 637                             cpu_to_le64(ps_page->dma);
 638                 }
 639
 640                 skb = netdev_alloc_skb_ip_align(netdev,
 641                                                 adapter->rx_ps_bsize0);
 642
 643                 if (!skb) {
 644                         adapter->alloc_rx_buff_failed++;
 645                         break;
 646                 }
 647
 648                 buffer_info->skb = skb;
 649                 buffer_info->dma = dma_map_single(&pdev->dev, skb->data,
 650                                                   adapter->rx_ps_bsize0,
 651                                                   DMA_FROM_DEVICE);
 652                 if (dma_mapping_error(&pdev->dev, buffer_info->dma)) {
 653                         dev_err(&pdev->dev, "Rx DMA map failed\n");
 654                         adapter->rx_dma_failed++;
 655                         /* cleanup skb */
 656                         dev_kfree_skb_any(skb);
 657                         buffer_info->skb = NULL;
 658                         break;
 659                 }
 660
 661                 rx_desc->read.buffer_addr[0] = cpu_to_le64(buffer_info->dma);
 662
 663                 if (unlikely(!(i & (E1000_RX_BUFFER_WRITE - 1)))) {
 664                         /*
 665                          * Force memory writes to complete before letting h/w
 666                          * know there are new descriptors to fetch.  (Only
 667                          * applicable for weak-ordered memory model archs,
 668                          * such as IA-64).
 669                          */
 670                         wmb();
 671                         writel(i << 1, adapter->hw.hw_addr + rx_ring->tail);
 672                 }
 673
 674                 i++;
 675                 if (i == rx_ring->count)
 676                         i = 0;
 677                 buffer_info = &rx_ring->buffer_info[i];
 678         }
 679
 680 no_buffers:
 681         rx_ring->next_to_use = i;
 682 }
 683
 684 /**
 685  * e1000_alloc_jumbo_rx_buffers - Replace used jumbo receive buffers
 686  * @adapter: address of board private structure
 687  * @cleaned_count: number of buffers to allocate this pass
 688  **/
 689
 690 static void e1000_alloc_jumbo_rx_buffers(struct e1000_adapter *adapter,
 691                                          int cleaned_count)
 692 {
 693         struct net_device *netdev = adapter->netdev;
 694         struct pci_dev *pdev = adapter->pdev;
 695         struct e1000_rx_desc *rx_desc;
 696         struct e1000_ring *rx_ring = adapter->rx_ring;
 697         struct e1000_buffer *buffer_info;
 698         struct sk_buff *skb;
 699         unsigned int i;
 700         unsigned int bufsz = 256 - 16 /* for skb_reserve */;
 701
 702         i = rx_ring->next_to_use;
 703         buffer_info = &rx_ring->buffer_info[i];
 704
 705         while (cleaned_count--) {
 706                 skb = buffer_info->skb;
 707                 if (skb) {
 708                         skb_trim(skb, 0);
 709                         goto check_page;
 710                 }
 711
 712                 skb = netdev_alloc_skb_ip_align(netdev, bufsz);
 713                 if (unlikely(!skb)) {
 714                         /* Better luck next round */
 715                         adapter->alloc_rx_buff_failed++;
 716                         break;
 717                 }
 718
 719                 buffer_info->skb = skb;
 720 check_page:
 721                 /* allocate a new page if necessary */
 722                 if (!buffer_info->page) {
 723                         buffer_info->page = alloc_page(GFP_ATOMIC);
 724                         if (unlikely(!buffer_info->page)) {
 725                                 adapter->alloc_rx_buff_failed++;
 726                                 break;
 727                         }
 728                 }
 729
 730                 if (!buffer_info->dma)
 731                         buffer_info->dma = dma_map_page(&pdev->dev,
 732                                                         buffer_info->page, 0,
 733                                                         PAGE_SIZE,
 734                                                         DMA_FROM_DEVICE);
 735
 736                 rx_desc = E1000_RX_DESC(*rx_ring, i);
 737                 rx_desc->buffer_addr = cpu_to_le64(buffer_info->dma);
 738
 739                 if (unlikely(++i == rx_ring->count))
 740                         i = 0;
 741                 buffer_info = &rx_ring->buffer_info[i];
 742         }
 743
 744         if (likely(rx_ring->next_to_use != i)) {
 745                 rx_ring->next_to_use = i;
 746                 if (unlikely(i-- == 0))
 747                         i = (rx_ring->count - 1);
 748
 749                 /* Force memory writes to complete before letting h/w
 750                  * know there are new descriptors to fetch.  (Only
 751                  * applicable for weak-ordered memory model archs,
 752                  * such as IA-64). */
 753                 wmb();
 754                 writel(i, adapter->hw.hw_addr + rx_ring->tail);
 755         }
 756 }
 757
 758 /**
 759  * e1000_clean_rx_irq - Send received data up the network stack; legacy
 760  * @adapter: board private structure
 761  *
 762  * the return value indicates whether actual cleaning was done, there
 763  * is no guarantee that everything was cleaned
 764  **/
 765 static bool e1000_clean_rx_irq(struct e1000_adapter *adapter,
 766                                int *work_done, int work_to_do)
 767 {
 768         struct net_device *netdev = adapter->netdev;
 769         struct pci_dev *pdev = adapter->pdev;
 770         struct e1000_hw *hw = &adapter->hw;
 771         struct e1000_ring *rx_ring = adapter->rx_ring;
 772         struct e1000_rx_desc *rx_desc, *next_rxd;
 773         struct e1000_buffer *buffer_info, *next_buffer;
 774         u32 length;
 775         unsigned int i;
 776         int cleaned_count = 0;
 777         bool cleaned = 0;
 778         unsigned int total_rx_bytes = 0, total_rx_packets = 0;
 779
 780         i = rx_ring->next_to_clean;
 781         rx_desc = E1000_RX_DESC(*rx_ring, i);
 782         buffer_info = &rx_ring->buffer_info[i];
 783
 784         while (rx_desc->status & E1000_RXD_STAT_DD) {
 785                 struct sk_buff *skb;
 786                 u8 status;
 787
 788                 if (*work_done >= work_to_do)
 789                         break;
 790                 (*work_done)++;
 791                 rmb();  /* read descriptor and rx_buffer_info after status DD */
 792
 793                 status = rx_desc->status;
 794                 skb = buffer_info->skb;
 795                 buffer_info->skb = NULL;
 796
 797                 prefetch(skb->data - NET_IP_ALIGN);
 798
 799                 i++;
 800                 if (i == rx_ring->count)
 801                         i = 0;
 802                 next_rxd = E1000_RX_DESC(*rx_ring, i);
 803                 prefetch(next_rxd);
 804
 805                 next_buffer = &rx_ring->buffer_info[i];
 806
 807                 cleaned = 1;
 808                 cleaned_count++;
 809                 dma_unmap_single(&pdev->dev,
 810                                  buffer_info->dma,
 811                                  adapter->rx_buffer_len,
 812                                  DMA_FROM_DEVICE);
 813                 buffer_info->dma = 0;
 814
 815                 length = le16_to_cpu(rx_desc->length);
 816
 817                 /*
 818                  * !EOP means multiple descriptors were used to store a single
 819                  * packet, if that's the case we need to toss it.  In fact, we
 820                  * need to toss every packet with the EOP bit clear and the
 821                  * next frame that _does_ have the EOP bit set, as it is by
 822                  * definition only a frame fragment
 823                  */
 824                 if (unlikely(!(status & E1000_RXD_STAT_EOP)))
 825                         adapter->flags2 |= FLAG2_IS_DISCARDING;
 826
 827                 if (adapter->flags2 & FLAG2_IS_DISCARDING) {
 828                         /* All receives must fit into a single buffer */
 829                         e_dbg("Receive packet consumed multiple buffers\n");
 830                         /* recycle */
 831                         buffer_info->skb = skb;
 832                         if (status & E1000_RXD_STAT_EOP)
 833                                 adapter->flags2 &= ~FLAG2_IS_DISCARDING;
 834                         goto next_desc;
 835                 }
 836
 837                 if (rx_desc->errors & E1000_RXD_ERR_FRAME_ERR_MASK) {
 838                         /* recycle */
 839                         buffer_info->skb = skb;
 840                         goto next_desc;
 841                 }
 842
 843                 /* adjust length to remove Ethernet CRC */
 844                 if (!(adapter->flags2 & FLAG2_CRC_STRIPPING))
 845                         length -= 4;
 846
 847                 total_rx_bytes += length;
 848                 total_rx_packets++;
 849
 850                 /*
 851                  * code added for copybreak, this should improve
 852                  * performance for small packets with large amounts
 853                  * of reassembly being done in the stack
 854                  */
 855                 if (length < copybreak) {
 856                         struct sk_buff *new_skb =
 857                             netdev_alloc_skb_ip_align(netdev, length);
 858                         if (new_skb) {
 859                                 skb_copy_to_linear_data_offset(new_skb,
 860                                                                -NET_IP_ALIGN,
 861                                                                (skb->data -
 862                                                                 NET_IP_ALIGN),
 863                                                                (length +
 864                                                                 NET_IP_ALIGN));
 865                                 /* save the skb in buffer_info as good */
 866                                 buffer_info->skb = skb;
 867                                 skb = new_skb;
 868                         }
 869                         /* else just continue with the old one */
 870                 }
 871                 /* end copybreak code */
 872                 skb_put(skb, length);
 873
 874                 /* Receive Checksum Offload */
 875                 e1000_rx_checksum(adapter,
 876                                   (u32)(status) |
 877                                   ((u32)(rx_desc->errors) << 24),
 878                                   le16_to_cpu(rx_desc->csum), skb);
 879
 880                 e1000_receive_skb(adapter, netdev, skb,status,rx_desc->special);
 881
 882 next_desc:
 883                 rx_desc->status = 0;
 884
 885                 /* return some buffers to hardware, one at a time is too slow */
 886                 if (cleaned_count >= E1000_RX_BUFFER_WRITE) {
 887                         adapter->alloc_rx_buf(adapter, cleaned_count);
 888                         cleaned_count = 0;
 889                 }
 890
 891                 /* use prefetched values */
 892                 rx_desc = next_rxd;
 893                 buffer_info = next_buffer;
 894         }
 895         rx_ring->next_to_clean = i;
 896
 897         cleaned_count = e1000_desc_unused(rx_ring);
 898         if (cleaned_count)
 899                 adapter->alloc_rx_buf(adapter, cleaned_count);
 900
 901         adapter->total_rx_bytes += total_rx_bytes;
 902         adapter->total_rx_packets += total_rx_packets;
 903         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         e1e_rphy(hw, PHY_STATUS, &phy_status);
 939         e1e_rphy(hw, PHY_1000T_STATUS, &phy_1000t_status);
 940         e1e_rphy(hw, PHY_EXT_STATUS, &phy_ext_status);
 941
 942         pci_read_config_word(adapter->pdev, PCI_STATUS, &pci_status);
 943
 944         /* detected Hardware unit hang */
 945         e_err("Detected Hardware Unit Hang:\n"
 946               "  TDH                  <%x>\n"
 947               "  TDT                  <%x>\n"
 948               "  next_to_use          <%x>\n"
 949               "  next_to_clean        <%x>\n"
 950               "buffer_info[next_to_clean]:\n"
 951               "  time_stamp           <%lx>\n"
 952               "  next_to_watch        <%x>\n"
 953               "  jiffies              <%lx>\n"
 954               "  next_to_watch.status <%x>\n"
 955               "MAC Status             <%x>\n"
 956               "PHY Status             <%x>\n"
 957               "PHY 1000BASE-T Status  <%x>\n"
 958               "PHY Extended Status    <%x>\n"
 959               "PCI Status             <%x>\n",
 960               readl(adapter->hw.hw_addr + tx_ring->head),
 961               readl(adapter->hw.hw_addr + tx_ring->tail),
 962               tx_ring->next_to_use,
 963               tx_ring->next_to_clean,
 964               tx_ring->buffer_info[eop].time_stamp,
 965               eop,
 966               jiffies,
 967               eop_desc->upper.fields.status,
 968               er32(STATUS),
 969               phy_status,
 970               phy_1000t_status,
 971               phy_ext_status,
 972               pci_status);
 973 }
 974
 975 /**
 976  * e1000_clean_tx_irq - Reclaim resources after transmit completes
 977  * @adapter: board private structure
 978  *
 979  * the return value indicates whether actual cleaning was done, there
 980  * is no guarantee that everything was cleaned
 981  **/
 982 static bool e1000_clean_tx_irq(struct e1000_adapter *adapter)
 983 {
 984         struct net_device *netdev = adapter->netdev;
 985         struct e1000_hw *hw = &adapter->hw;
 986         struct e1000_ring *tx_ring = adapter->tx_ring;
 987         struct e1000_tx_desc *tx_desc, *eop_desc;
 988         struct e1000_buffer *buffer_info;
 989         unsigned int i, eop;
 990         unsigned int count = 0;
 991         unsigned int total_tx_bytes = 0, total_tx_packets = 0;
 992
 993         i = tx_ring->next_to_clean;
 994         eop = tx_ring->buffer_info[i].next_to_watch;
 995         eop_desc = E1000_TX_DESC(*tx_ring, eop);
 996
 997         while ((eop_desc->upper.data & cpu_to_le32(E1000_TXD_STAT_DD)) &&
 998                (count < tx_ring->count)) {
 999                 bool cleaned = false;
1000                 rmb(); /* read buffer_info after eop_desc */
1001                 for (; !cleaned; count++) {
1002                         tx_desc = E1000_TX_DESC(*tx_ring, i);
1003                         buffer_info = &tx_ring->buffer_info[i];
1004                         cleaned = (i == eop);
1005
1006                         if (cleaned) {
1007                                 total_tx_packets += buffer_info->segs;
1008                                 total_tx_bytes += buffer_info->bytecount;
1009                         }
1010
1011                         e1000_put_txbuf(adapter, buffer_info);
1012                         tx_desc->upper.data = 0;
1013
1014                         i++;
1015                         if (i == tx_ring->count)
1016                                 i = 0;
1017                 }
1018
1019                 if (i == tx_ring->next_to_use)
1020                         break;
1021                 eop = tx_ring->buffer_info[i].next_to_watch;
1022                 eop_desc = E1000_TX_DESC(*tx_ring, eop);
1023         }
1024
1025         tx_ring->next_to_clean = i;
1026
1027 #define TX_WAKE_THRESHOLD 32
1028         if (count && netif_carrier_ok(netdev) &&
1029             e1000_desc_unused(tx_ring) >= TX_WAKE_THRESHOLD) {
1030                 /* Make sure that anybody stopping the queue after this
1031                  * sees the new next_to_clean.
1032                  */
1033                 smp_mb();
1034
1035                 if (netif_queue_stopped(netdev) &&
1036                     !(test_bit(__E1000_DOWN, &adapter->state))) {
1037                         netif_wake_queue(netdev);
1038                         ++adapter->restart_queue;
1039                 }
1040         }
1041
1042         if (adapter->detect_tx_hung) {
1043                 /*
1044                  * Detect a transmit hang in hardware, this serializes the
1045                  * check with the clearing of time_stamp and movement of i
1046                  */
1047                 adapter->detect_tx_hung = 0;
1048                 if (tx_ring->buffer_info[i].time_stamp &&
1049                     time_after(jiffies, tx_ring->buffer_info[i].time_stamp
1050                                + (adapter->tx_timeout_factor * HZ)) &&
1051                     !(er32(STATUS) & E1000_STATUS_TXOFF)) {
1052                         schedule_work(&adapter->print_hang_task);
1053                         netif_stop_queue(netdev);
1054                 }
1055         }
1056         adapter->total_tx_bytes += total_tx_bytes;
1057         adapter->total_tx_packets += total_tx_packets;
1058         return count < tx_ring->count;
1059 }
1060
1061 /**
1062  * e1000_clean_rx_irq_ps - Send received data up the network stack; packet split
1063  * @adapter: board private structure
1064  *
1065  * the return value indicates whether actual cleaning was done, there
1066  * is no guarantee that everything was cleaned
1067  **/
1068 static bool e1000_clean_rx_irq_ps(struct e1000_adapter *adapter,
1069                                   int *work_done, int work_to_do)
1070 {
1071         struct e1000_hw *hw = &adapter->hw;
1072         union e1000_rx_desc_packet_split *rx_desc, *next_rxd;
1073         struct net_device *netdev = adapter->netdev;
1074         struct pci_dev *pdev = adapter->pdev;
1075         struct e1000_ring *rx_ring = adapter->rx_ring;
1076         struct e1000_buffer *buffer_info, *next_buffer;
1077         struct e1000_ps_page *ps_page;
1078         struct sk_buff *skb;
1079         unsigned int i, j;
1080         u32 length, staterr;
1081         int cleaned_count = 0;
1082         bool cleaned = 0;
1083         unsigned int total_rx_bytes = 0, total_rx_packets = 0;
1084
1085         i = rx_ring->next_to_clean;
1086         rx_desc = E1000_RX_DESC_PS(*rx_ring, i);
1087         staterr = le32_to_cpu(rx_desc->wb.middle.status_error);
1088         buffer_info = &rx_ring->buffer_info[i];
1089
1090         while (staterr & E1000_RXD_STAT_DD) {
1091                 if (*work_done >= work_to_do)
1092                         break;
1093                 (*work_done)++;
1094                 skb = buffer_info->skb;
1095                 rmb();  /* read descriptor and rx_buffer_info after status DD */
1096
1097                 /* in the packet split case this is header only */
1098                 prefetch(skb->data - NET_IP_ALIGN);
1099
1100                 i++;
1101                 if (i == rx_ring->count)
1102                         i = 0;
1103                 next_rxd = E1000_RX_DESC_PS(*rx_ring, i);
1104                 prefetch(next_rxd);
1105
1106                 next_buffer = &rx_ring->buffer_info[i];
1107
1108                 cleaned = 1;
1109                 cleaned_count++;
1110                 dma_unmap_single(&pdev->dev, buffer_info->dma,
1111                                  adapter->rx_ps_bsize0, DMA_FROM_DEVICE);
1112                 buffer_info->dma = 0;
1113
1114                 /* see !EOP comment in other Rx routine */
1115                 if (!(staterr & E1000_RXD_STAT_EOP))
1116                         adapter->flags2 |= FLAG2_IS_DISCARDING;
1117
1118                 if (adapter->flags2 & FLAG2_IS_DISCARDING) {
1119                         e_dbg("Packet Split buffers didn't pick up the full "
1120                               "packet\n");
1121                         dev_kfree_skb_irq(skb);
1122                         if (staterr & E1000_RXD_STAT_EOP)
1123                                 adapter->flags2 &= ~FLAG2_IS_DISCARDING;
1124                         goto next_desc;
1125                 }
1126
1127                 if (staterr & E1000_RXDEXT_ERR_FRAME_ERR_MASK) {
1128                         dev_kfree_skb_irq(skb);
1129                         goto next_desc;
1130                 }
1131
1132                 length = le16_to_cpu(rx_desc->wb.middle.length0);
1133
1134                 if (!length) {
1135                         e_dbg("Last part of the packet spanning multiple "
1136                               "descriptors\n");
1137                         dev_kfree_skb_irq(skb);
1138                         goto next_desc;
1139                 }
1140
1141                 /* Good Receive */
1142                 skb_put(skb, length);
1143
1144                 {
1145                 /*
1146                  * this looks ugly, but it seems compiler issues make it
1147                  * more efficient than reusing j
1148                  */
1149                 int l1 = le16_to_cpu(rx_desc->wb.upper.length[0]);
1150
1151                 /*
1152                  * page alloc/put takes too long and effects small packet
1153                  * throughput, so unsplit small packets and save the alloc/put
1154                  * only valid in softirq (napi) context to call kmap_*
1155                  */
1156                 if (l1 && (l1 <= copybreak) &&
1157                     ((length + l1) <= adapter->rx_ps_bsize0)) {
1158                         u8 *vaddr;
1159
1160                         ps_page = &buffer_info->ps_pages[0];
1161
1162                         /*
1163                          * there is no documentation about how to call
1164                          * kmap_atomic, so we can't hold the mapping
1165                          * very long
1166                          */
1167                         dma_sync_single_for_cpu(&pdev->dev, ps_page->dma,
1168                                                 PAGE_SIZE, DMA_FROM_DEVICE);
1169                         vaddr = kmap_atomic(ps_page->page, KM_SKB_DATA_SOFTIRQ);
1170                         memcpy(skb_tail_pointer(skb), vaddr, l1);
1171                         kunmap_atomic(vaddr, KM_SKB_DATA_SOFTIRQ);
1172                         dma_sync_single_for_device(&pdev->dev, ps_page->dma,
1173                                                    PAGE_SIZE, DMA_FROM_DEVICE);
1174
1175                         /* remove the CRC */
1176                         if (!(adapter->flags2 & FLAG2_CRC_STRIPPING))
1177                                 l1 -= 4;
1178
1179                         skb_put(skb, l1);
1180                         goto copydone;
1181                 } /* if */
1182                 }
1183
1184                 for (j = 0; j < PS_PAGE_BUFFERS; j++) {
1185                         length = le16_to_cpu(rx_desc->wb.upper.length[j]);
1186                         if (!length)
1187                                 break;
1188
1189                         ps_page = &buffer_info->ps_pages[j];
1190                         dma_unmap_page(&pdev->dev, ps_page->dma, PAGE_SIZE,
1191                                        DMA_FROM_DEVICE);
1192                         ps_page->dma = 0;
1193                         skb_fill_page_desc(skb, j, ps_page->page, 0, length);
1194                         ps_page->page = NULL;
1195                         skb->len += length;
1196                         skb->data_len += length;
1197                         skb->truesize += length;
1198                 }
1199
1200                 /* strip the ethernet crc, problem is we're using pages now so
1201                  * this whole operation can get a little cpu intensive
1202                  */
1203                 if (!(adapter->flags2 & FLAG2_CRC_STRIPPING))
1204                         pskb_trim(skb, skb->len - 4);
1205
1206 copydone:
1207                 total_rx_bytes += skb->len;
1208                 total_rx_packets++;
1209
1210                 e1000_rx_checksum(adapter, staterr, le16_to_cpu(
1211                         rx_desc->wb.lower.hi_dword.csum_ip.csum), skb);
1212
1213                 if (rx_desc->wb.upper.header_status &
1214                            cpu_to_le16(E1000_RXDPS_HDRSTAT_HDRSP))
1215                         adapter->rx_hdr_split++;
1216
1217                 e1000_receive_skb(adapter, netdev, skb,
1218                                   staterr, rx_desc->wb.middle.vlan);
1219
1220 next_desc:
1221                 rx_desc->wb.middle.status_error &= cpu_to_le32(~0xFF);
1222                 buffer_info->skb = NULL;
1223
1224                 /* return some buffers to hardware, one at a time is too slow */
1225                 if (cleaned_count >= E1000_RX_BUFFER_WRITE) {
1226                         adapter->alloc_rx_buf(adapter, cleaned_count);
1227                         cleaned_count = 0;
1228                 }
1229
1230                 /* use prefetched values */
1231                 rx_desc = next_rxd;
1232                 buffer_info = next_buffer;
1233
1234                 staterr = le32_to_cpu(rx_desc->wb.middle.status_error);
1235         }
1236         rx_ring->next_to_clean = i;
1237
1238         cleaned_count = e1000_desc_unused(rx_ring);
1239         if (cleaned_count)
1240                 adapter->alloc_rx_buf(adapter, cleaned_count);
1241
1242         adapter->total_rx_bytes += total_rx_bytes;
1243         adapter->total_rx_packets += total_rx_packets;
1244         return cleaned;
1245 }
1246
1247 /**
1248  * e1000_consume_page - helper function
1249  **/
1250 static void e1000_consume_page(struct e1000_buffer *bi, struct sk_buff *skb,
1251                                u16 length)
1252 {
1253         bi->page = NULL;
1254         skb->len += length;
1255         skb->data_len += length;
1256         skb->truesize += length;
1257 }
1258
1259 /**
1260  * e1000_clean_jumbo_rx_irq - Send received data up the network stack; legacy
1261  * @adapter: board private structure
1262  *
1263  * the return value indicates whether actual cleaning was done, there
1264  * is no guarantee that everything was cleaned
1265  **/
1266
1267 static bool e1000_clean_jumbo_rx_irq(struct e1000_adapter *adapter,
1268                                      int *work_done, int work_to_do)
1269 {
1270         struct net_device *netdev = adapter->netdev;
1271         struct pci_dev *pdev = adapter->pdev;
1272         struct e1000_ring *rx_ring = adapter->rx_ring;
1273         struct e1000_rx_desc *rx_desc, *next_rxd;
1274         struct e1000_buffer *buffer_info, *next_buffer;
1275         u32 length;
1276         unsigned int i;
1277         int cleaned_count = 0;
1278         bool cleaned = false;
1279         unsigned int total_rx_bytes=0, total_rx_packets=0;
1280
1281         i = rx_ring->next_to_clean;
1282         rx_desc = E1000_RX_DESC(*rx_ring, i);
1283         buffer_info = &rx_ring->buffer_info[i];
1284
1285         while (rx_desc->status & E1000_RXD_STAT_DD) {
1286                 struct sk_buff *skb;
1287                 u8 status;
1288
1289                 if (*work_done >= work_to_do)
1290                         break;
1291                 (*work_done)++;
1292                 rmb();  /* read descriptor and rx_buffer_info after status DD */
1293
1294                 status = rx_desc->status;
1295                 skb = buffer_info->skb;
1296                 buffer_info->skb = NULL;
1297
1298                 ++i;
1299                 if (i == rx_ring->count)
1300                         i = 0;
1301                 next_rxd = E1000_RX_DESC(*rx_ring, i);
1302                 prefetch(next_rxd);
1303
1304                 next_buffer = &rx_ring->buffer_info[i];
1305
1306                 cleaned = true;
1307                 cleaned_count++;
1308                 dma_unmap_page(&pdev->dev, buffer_info->dma, PAGE_SIZE,
1309                                DMA_FROM_DEVICE);
1310                 buffer_info->dma = 0;
1311
1312                 length = le16_to_cpu(rx_desc->length);
1313
1314                 /* errors is only valid for DD + EOP descriptors */
1315                 if (unlikely((status & E1000_RXD_STAT_EOP) &&
1316                     (rx_desc->errors & E1000_RXD_ERR_FRAME_ERR_MASK))) {
1317                                 /* recycle both page and skb */
1318                                 buffer_info->skb = skb;
1319                                 /* an error means any chain goes out the window
1320                                  * too */
1321                                 if (rx_ring->rx_skb_top)
1322                                         dev_kfree_skb(rx_ring->rx_skb_top);
1323                                 rx_ring->rx_skb_top = NULL;
1324                                 goto next_desc;
1325                 }
1326
1327 #define rxtop (rx_ring->rx_skb_top)
1328                 if (!(status & E1000_RXD_STAT_EOP)) {
1329                         /* this descriptor is only the beginning (or middle) */
1330                         if (!rxtop) {
1331                                 /* this is the beginning of a chain */
1332                                 rxtop = skb;
1333                                 skb_fill_page_desc(rxtop, 0, buffer_info->page,
1334                                                    0, length);
1335                         } else {
1336                                 /* this is the middle of a chain */
1337                                 skb_fill_page_desc(rxtop,
1338                                     skb_shinfo(rxtop)->nr_frags,
1339                                     buffer_info->page, 0, length);
1340                                 /* re-use the skb, only consumed the page */
1341                                 buffer_info->skb = skb;
1342                         }
1343                         e1000_consume_page(buffer_info, rxtop, length);
1344                         goto next_desc;
1345                 } else {
1346                         if (rxtop) {
1347                                 /* end of the chain */
1348                                 skb_fill_page_desc(rxtop,
1349                                     skb_shinfo(rxtop)->nr_frags,
1350                                     buffer_info->page, 0, length);
1351                                 /* re-use the current skb, we only consumed the
1352                                  * page */
1353                                 buffer_info->skb = skb;
1354                                 skb = rxtop;
1355                                 rxtop = NULL;
1356                                 e1000_consume_page(buffer_info, skb, length);
1357                         } else {
1358                                 /* no chain, got EOP, this buf is the packet
1359                                  * copybreak to save the put_page/alloc_page */
1360                                 if (length <= copybreak &&
1361                                     skb_tailroom(skb) >= length) {
1362                                         u8 *vaddr;
1363                                         vaddr = kmap_atomic(buffer_info->page,
1364                                                            KM_SKB_DATA_SOFTIRQ);
1365                                         memcpy(skb_tail_pointer(skb), vaddr,
1366                                                length);
1367                                         kunmap_atomic(vaddr,
1368                                                       KM_SKB_DATA_SOFTIRQ);
1369                                         /* re-use the page, so don't erase
1370                                          * buffer_info->page */
1371                                         skb_put(skb, length);
1372                                 } else {
1373                                         skb_fill_page_desc(skb, 0,
1374                                                            buffer_info->page, 0,
1375                                                            length);
1376                                         e1000_consume_page(buffer_info, skb,
1377                                                            length);
1378                                 }
1379                         }
1380                 }
1381
1382                 /* Receive Checksum Offload XXX recompute due to CRC strip? */
1383                 e1000_rx_checksum(adapter,
1384                                   (u32)(status) |
1385                                   ((u32)(rx_desc->errors) << 24),
1386                                   le16_to_cpu(rx_desc->csum), skb);
1387
1388                 /* probably a little skewed due to removing CRC */
1389                 total_rx_bytes += skb->len;
1390                 total_rx_packets++;
1391
1392                 /* eth type trans needs skb->data to point to something */
1393                 if (!pskb_may_pull(skb, ETH_HLEN)) {
1394                         e_err("pskb_may_pull failed.\n");
1395                         dev_kfree_skb(skb);
1396                         goto next_desc;
1397                 }
1398
1399                 e1000_receive_skb(adapter, netdev, skb, status,
1400                                   rx_desc->special);
1401
1402 next_desc:
1403                 rx_desc->status = 0;
1404
1405                 /* return some buffers to hardware, one at a time is too slow */
1406                 if (unlikely(cleaned_count >= E1000_RX_BUFFER_WRITE)) {
1407                         adapter->alloc_rx_buf(adapter, cleaned_count);
1408                         cleaned_count = 0;
1409                 }
1410
1411                 /* use prefetched values */
1412                 rx_desc = next_rxd;
1413                 buffer_info = next_buffer;
1414         }
1415         rx_ring->next_to_clean = i;
1416
1417         cleaned_count = e1000_desc_unused(rx_ring);
1418         if (cleaned_count)
1419                 adapter->alloc_rx_buf(adapter, cleaned_count);
1420
1421         adapter->total_rx_bytes += total_rx_bytes;
1422         adapter->total_rx_packets += total_rx_packets;
1423         return cleaned;
1424 }
1425
1426 /**
1427  * e1000_clean_rx_ring - Free Rx Buffers per Queue
1428  * @adapter: board private structure
1429  **/
1430 static void e1000_clean_rx_ring(struct e1000_adapter *adapter)
1431 {
1432         struct e1000_ring *rx_ring = adapter->rx_ring;
1433         struct e1000_buffer *buffer_info;
1434         struct e1000_ps_page *ps_page;
1435         struct pci_dev *pdev = adapter->pdev;
1436         unsigned int i, j;
1437
1438         /* Free all the Rx ring sk_buffs */
1439         for (i = 0; i < rx_ring->count; i++) {
1440                 buffer_info = &rx_ring->buffer_info[i];
1441                 if (buffer_info->dma) {
1442                         if (adapter->clean_rx == e1000_clean_rx_irq)
1443                                 dma_unmap_single(&pdev->dev, buffer_info->dma,
1444                                                  adapter->rx_buffer_len,
1445                                                  DMA_FROM_DEVICE);
1446                         else if (adapter->clean_rx == e1000_clean_jumbo_rx_irq)
1447                                 dma_unmap_page(&pdev->dev, buffer_info->dma,
1448                                                PAGE_SIZE,
1449                                                DMA_FROM_DEVICE);
1450                         else if (adapter->clean_rx == e1000_clean_rx_irq_ps)
1451                                 dma_unmap_single(&pdev->dev, buffer_info->dma,
1452                                                  adapter->rx_ps_bsize0,
1453                                                  DMA_FROM_DEVICE);
1454                         buffer_info->dma = 0;
1455                 }
1456
1457                 if (buffer_info->page) {
1458                         put_page(buffer_info->page);
1459                         buffer_info->page = NULL;
1460                 }
1461
1462                 if (buffer_info->skb) {
1463                         dev_kfree_skb(buffer_info->skb);
1464                         buffer_info->skb = NULL;
1465                 }
1466
1467                 for (j = 0; j < PS_PAGE_BUFFERS; j++) {
1468                         ps_page = &buffer_info->ps_pages[j];
1469                         if (!ps_page->page)
1470                                 break;
1471                         dma_unmap_page(&pdev->dev, ps_page->dma, PAGE_SIZE,
1472                                        DMA_FROM_DEVICE);
1473                         ps_page->dma = 0;
1474                         put_page(ps_page->page);
1475                         ps_page->page = NULL;
1476                 }
1477         }
1478
1479         /* there also may be some cached data from a chained receive */
1480         if (rx_ring->rx_skb_top) {
1481                 dev_kfree_skb(rx_ring->rx_skb_top);
1482                 rx_ring->rx_skb_top = NULL;
1483         }
1484
1485         /* Zero out the descriptor ring */
1486         memset(rx_ring->desc, 0, rx_ring->size);
1487
1488         rx_ring->next_to_clean = 0;
1489         rx_ring->next_to_use = 0;
1490         adapter->flags2 &= ~FLAG2_IS_DISCARDING;
1491
1492         writel(0, adapter->hw.hw_addr + rx_ring->head);
1493         writel(0, adapter->hw.hw_addr + rx_ring->tail);
1494 }
1495
1496 static void e1000e_downshift_workaround(struct work_struct *work)
1497 {
1498         struct e1000_adapter *adapter = container_of(work,
1499                                         struct e1000_adapter, downshift_task);
1500
1501         e1000e_gig_downshift_workaround_ich8lan(&adapter->hw);
1502 }
1503
1504 /**
1505  * e1000_intr_msi - Interrupt Handler
1506  * @irq: interrupt number
1507  * @data: pointer to a network interface device structure
1508  **/
1509 static irqreturn_t e1000_intr_msi(int irq, void *data)
1510 {
1511         struct net_device *netdev = data;
1512         struct e1000_adapter *adapter = netdev_priv(netdev);
1513         struct e1000_hw *hw = &adapter->hw;
1514         u32 icr = er32(ICR);
1515
1516         /*
1517          * read ICR disables interrupts using IAM
1518          */
1519
1520         if (icr & E1000_ICR_LSC) {
1521                 hw->mac.get_link_status = 1;
1522                 /*
1523                  * ICH8 workaround-- Call gig speed drop workaround on cable
1524                  * disconnect (LSC) before accessing any PHY registers
1525                  */
1526                 if ((adapter->flags & FLAG_LSC_GIG_SPEED_DROP) &&
1527                     (!(er32(STATUS) & E1000_STATUS_LU)))
1528                         schedule_work(&adapter->downshift_task);
1529
1530                 /*
1531                  * 80003ES2LAN workaround-- For packet buffer work-around on
1532                  * link down event; disable receives here in the ISR and reset
1533                  * adapter in watchdog
1534                  */
1535                 if (netif_carrier_ok(netdev) &&
1536                     adapter->flags & FLAG_RX_NEEDS_RESTART) {
1537                         /* disable receives */
1538                         u32 rctl = er32(RCTL);
1539                         ew32(RCTL, rctl & ~E1000_RCTL_EN);
1540                         adapter->flags |= FLAG_RX_RESTART_NOW;
1541                 }
1542                 /* guard against interrupt when we're going down */
1543                 if (!test_bit(__E1000_DOWN, &adapter->state))
1544                         mod_timer(&adapter->watchdog_timer, jiffies + 1);
1545         }
1546
1547         if (napi_schedule_prep(&adapter->napi)) {
1548                 adapter->total_tx_bytes = 0;
1549                 adapter->total_tx_packets = 0;
1550                 adapter->total_rx_bytes = 0;
1551                 adapter->total_rx_packets = 0;
1552                 __napi_schedule(&adapter->napi);
1553         }
1554
1555         return IRQ_HANDLED;
1556 }
1557
1558 /**
1559  * e1000_intr - Interrupt Handler
1560  * @irq: interrupt number
1561  * @data: pointer to a network interface device structure
1562  **/
1563 static irqreturn_t e1000_intr(int irq, void *data)
1564 {
1565         struct net_device *netdev = data;
1566         struct e1000_adapter *adapter = netdev_priv(netdev);
1567         struct e1000_hw *hw = &adapter->hw;
1568         u32 rctl, icr = er32(ICR);
1569
1570         if (!icr || test_bit(__E1000_DOWN, &adapter->state))
1571                 return IRQ_NONE;  /* Not our interrupt */
1572
1573         /*
1574          * IMS will not auto-mask if INT_ASSERTED is not set, and if it is
1575          * not set, then the adapter didn't send an interrupt
1576          */
1577         if (!(icr & E1000_ICR_INT_ASSERTED))
1578                 return IRQ_NONE;
1579
1580         /*
1581          * Interrupt Auto-Mask...upon reading ICR,
1582          * interrupts are masked.  No need for the
1583          * IMC write
1584          */
1585
1586         if (icr & E1000_ICR_LSC) {
1587                 hw->mac.get_link_status = 1;
1588                 /*
1589                  * ICH8 workaround-- Call gig speed drop workaround on cable
1590                  * disconnect (LSC) before accessing any PHY registers
1591                  */
1592                 if ((adapter->flags & FLAG_LSC_GIG_SPEED_DROP) &&
1593                     (!(er32(STATUS) & E1000_STATUS_LU)))
1594                         schedule_work(&adapter->downshift_task);
1595
1596                 /*
1597                  * 80003ES2LAN workaround--
1598                  * For packet buffer work-around on link down event;
1599                  * disable receives here in the ISR and
1600                  * reset adapter in watchdog
1601                  */
1602                 if (netif_carrier_ok(netdev) &&
1603                     (adapter->flags & FLAG_RX_NEEDS_RESTART)) {
1604                         /* disable receives */
1605                         rctl = er32(RCTL);
1606                         ew32(RCTL, rctl & ~E1000_RCTL_EN);
1607                         adapter->flags |= FLAG_RX_RESTART_NOW;
1608                 }
1609                 /* guard against interrupt when we're going down */
1610                 if (!test_bit(__E1000_DOWN, &adapter->state))
1611                         mod_timer(&adapter->watchdog_timer, jiffies + 1);
1612         }
1613
1614         if (napi_schedule_prep(&adapter->napi)) {
1615                 adapter->total_tx_bytes = 0;
1616                 adapter->total_tx_packets = 0;
1617                 adapter->total_rx_bytes = 0;
1618                 adapter->total_rx_packets = 0;
1619                 __napi_schedule(&adapter->napi);
1620         }
1621
1622         return IRQ_HANDLED;
1623 }
1624
1625 static irqreturn_t e1000_msix_other(int irq, void *data)
1626 {
1627         struct net_device *netdev = data;
1628         struct e1000_adapter *adapter = netdev_priv(netdev);
1629         struct e1000_hw *hw = &adapter->hw;
1630         u32 icr = er32(ICR);
1631
1632         if (!(icr & E1000_ICR_INT_ASSERTED)) {
1633                 if (!test_bit(__E1000_DOWN, &adapter->state))
1634                         ew32(IMS, E1000_IMS_OTHER);
1635                 return IRQ_NONE;
1636         }
1637
1638         if (icr & adapter->eiac_mask)
1639                 ew32(ICS, (icr & adapter->eiac_mask));
1640
1641         if (icr & E1000_ICR_OTHER) {
1642                 if (!(icr & E1000_ICR_LSC))
1643                         goto no_link_interrupt;
1644                 hw->mac.get_link_status = 1;
1645                 /* guard against interrupt when we're going down */
1646                 if (!test_bit(__E1000_DOWN, &adapter->state))
1647                         mod_timer(&adapter->watchdog_timer, jiffies + 1);
1648         }
1649
1650 no_link_interrupt:
1651         if (!test_bit(__E1000_DOWN, &adapter->state))
1652                 ew32(IMS, E1000_IMS_LSC | E1000_IMS_OTHER);
1653
1654         return IRQ_HANDLED;
1655 }
1656
1657
1658 static irqreturn_t e1000_intr_msix_tx(int irq, void *data)
1659 {
1660         struct net_device *netdev = data;
1661         struct e1000_adapter *adapter = netdev_priv(netdev);
1662         struct e1000_hw *hw = &adapter->hw;
1663         struct e1000_ring *tx_ring = adapter->tx_ring;
1664
1665
1666         adapter->total_tx_bytes = 0;
1667         adapter->total_tx_packets = 0;
1668
1669         if (!e1000_clean_tx_irq(adapter))
1670                 /* Ring was not completely cleaned, so fire another interrupt */
1671                 ew32(ICS, tx_ring->ims_val);
1672
1673         return IRQ_HANDLED;
1674 }
1675
1676 static irqreturn_t e1000_intr_msix_rx(int irq, void *data)
1677 {
1678         struct net_device *netdev = data;
1679         struct e1000_adapter *adapter = netdev_priv(netdev);
1680
1681         /* Write the ITR value calculated at the end of the
1682          * previous interrupt.
1683          */
1684         if (adapter->rx_ring->set_itr) {
1685                 writel(1000000000 / (adapter->rx_ring->itr_val * 256),
1686                        adapter->hw.hw_addr + adapter->rx_ring->itr_register);
1687                 adapter->rx_ring->set_itr = 0;
1688         }
1689
1690         if (napi_schedule_prep(&adapter->napi)) {
1691                 adapter->total_rx_bytes = 0;
1692                 adapter->total_rx_packets = 0;
1693                 __napi_schedule(&adapter->napi);
1694         }
1695         return IRQ_HANDLED;
1696 }
1697
1698 /**
1699  * e1000_configure_msix - Configure MSI-X hardware
1700  *
1701  * e1000_configure_msix sets up the hardware to properly
1702  * generate MSI-X interrupts.
1703  **/
1704 static void e1000_configure_msix(struct e1000_adapter *adapter)
1705 {
1706         struct e1000_hw *hw = &adapter->hw;
1707         struct e1000_ring *rx_ring = adapter->rx_ring;
1708         struct e1000_ring *tx_ring = adapter->tx_ring;
1709         int vector = 0;
1710         u32 ctrl_ext, ivar = 0;
1711
1712         adapter->eiac_mask = 0;
1713
1714         /* Workaround issue with spurious interrupts on 82574 in MSI-X mode */
1715         if (hw->mac.type == e1000_82574) {
1716                 u32 rfctl = er32(RFCTL);
1717                 rfctl |= E1000_RFCTL_ACK_DIS;
1718                 ew32(RFCTL, rfctl);
1719         }
1720
1721 #define E1000_IVAR_INT_ALLOC_VALID      0x8
1722         /* Configure Rx vector */
1723         rx_ring->ims_val = E1000_IMS_RXQ0;
1724         adapter->eiac_mask |= rx_ring->ims_val;
1725         if (rx_ring->itr_val)
1726                 writel(1000000000 / (rx_ring->itr_val * 256),
1727                        hw->hw_addr + rx_ring->itr_register);
1728         else
1729                 writel(1, hw->hw_addr + rx_ring->itr_register);
1730         ivar = E1000_IVAR_INT_ALLOC_VALID | vector;
1731
1732         /* Configure Tx vector */
1733         tx_ring->ims_val = E1000_IMS_TXQ0;
1734         vector++;
1735         if (tx_ring->itr_val)
1736                 writel(1000000000 / (tx_ring->itr_val * 256),
1737                        hw->hw_addr + tx_ring->itr_register);
1738         else
1739                 writel(1, hw->hw_addr + tx_ring->itr_register);
1740         adapter->eiac_mask |= tx_ring->ims_val;
1741         ivar |= ((E1000_IVAR_INT_ALLOC_VALID | vector) << 8);
1742
1743         /* set vector for Other Causes, e.g. link changes */
1744         vector++;
1745         ivar |= ((E1000_IVAR_INT_ALLOC_VALID | vector) << 16);
1746         if (rx_ring->itr_val)
1747                 writel(1000000000 / (rx_ring->itr_val * 256),
1748                        hw->hw_addr + E1000_EITR_82574(vector));
1749         else
1750                 writel(1, hw->hw_addr + E1000_EITR_82574(vector));
1751
1752         /* Cause Tx interrupts on every write back */
1753         ivar |= (1 << 31);
1754
1755         ew32(IVAR, ivar);
1756
1757         /* enable MSI-X PBA support */
1758         ctrl_ext = er32(CTRL_EXT);
1759         ctrl_ext |= E1000_CTRL_EXT_PBA_CLR;
1760
1761         /* Auto-Mask Other interrupts upon ICR read */
1762 #define E1000_EIAC_MASK_82574   0x01F00000
1763         ew32(IAM, ~E1000_EIAC_MASK_82574 | E1000_IMS_OTHER);
1764         ctrl_ext |= E1000_CTRL_EXT_EIAME;
1765         ew32(CTRL_EXT, ctrl_ext);
1766         e1e_flush();
1767 }
1768
1769 void e1000e_reset_interrupt_capability(struct e1000_adapter *adapter)
1770 {
1771         if (adapter->msix_entries) {
1772                 pci_disable_msix(adapter->pdev);
1773                 kfree(adapter->msix_entries);
1774                 adapter->msix_entries = NULL;
1775         } else if (adapter->flags & FLAG_MSI_ENABLED) {
1776                 pci_disable_msi(adapter->pdev);
1777                 adapter->flags &= ~FLAG_MSI_ENABLED;
1778         }
1779 }
1780
1781 /**
1782  * e1000e_set_interrupt_capability - set MSI or MSI-X if supported
1783  *
1784  * Attempt to configure interrupts using the best available
1785  * capabilities of the hardware and kernel.
1786  **/
1787 void e1000e_set_interrupt_capability(struct e1000_adapter *adapter)
1788 {
1789         int err;
1790         int i;
1791
1792         switch (adapter->int_mode) {
1793         case E1000E_INT_MODE_MSIX:
1794                 if (adapter->flags & FLAG_HAS_MSIX) {
1795                         adapter->num_vectors = 3; /* RxQ0, TxQ0 and other */
1796                         adapter->msix_entries = kcalloc(adapter->num_vectors,
1797                                                       sizeof(struct msix_entry),
1798                                                       GFP_KERNEL);
1799                         if (adapter->msix_entries) {
1800                                 for (i = 0; i < adapter->num_vectors; i++)
1801                                         adapter->msix_entries[i].entry = i;
1802
1803                                 err = pci_enable_msix(adapter->pdev,
1804                                                       adapter->msix_entries,
1805                                                       adapter->num_vectors);
1806                                 if (err == 0)
1807                                         return;
1808                         }
1809                         /* MSI-X failed, so fall through and try MSI */
1810                         e_err("Failed to initialize MSI-X interrupts.  "
1811                               "Falling back to MSI interrupts.\n");
1812                         e1000e_reset_interrupt_capability(adapter);
1813                 }
1814                 adapter->int_mode = E1000E_INT_MODE_MSI;
1815                 /* Fall through */
1816         case E1000E_INT_MODE_MSI:
1817                 if (!pci_enable_msi(adapter->pdev)) {
1818                         adapter->flags |= FLAG_MSI_ENABLED;
1819                 } else {
1820                         adapter->int_mode = E1000E_INT_MODE_LEGACY;
1821                         e_err("Failed to initialize MSI interrupts.  Falling "
1822                               "back to legacy interrupts.\n");
1823                 }
1824                 /* Fall through */
1825         case E1000E_INT_MODE_LEGACY:
1826                 /* Don't do anything; this is the system default */
1827                 break;
1828         }
1829
1830         /* store the number of vectors being used */
1831         adapter->num_vectors = 1;
1832 }
1833
1834 /**
1835  * e1000_request_msix - Initialize MSI-X interrupts
1836  *
1837  * e1000_request_msix allocates MSI-X vectors and requests interrupts from the
1838  * kernel.
1839  **/
1840 static int e1000_request_msix(struct e1000_adapter *adapter)
1841 {
1842         struct net_device *netdev = adapter->netdev;
1843         int err = 0, vector = 0;
1844
1845         if (strlen(netdev->name) < (IFNAMSIZ - 5))
1846                 sprintf(adapter->rx_ring->name, "%s-rx-0", netdev->name);
1847         else
1848                 memcpy(adapter->rx_ring->name, netdev->name, IFNAMSIZ);
1849         err = request_irq(adapter->msix_entries[vector].vector,
1850                           e1000_intr_msix_rx, 0, adapter->rx_ring->name,
1851                           netdev);
1852         if (err)
1853                 goto out;
1854         adapter->rx_ring->itr_register = E1000_EITR_82574(vector);
1855         adapter->rx_ring->itr_val = adapter->itr;
1856         vector++;
1857
1858         if (strlen(netdev->name) < (IFNAMSIZ - 5))
1859                 sprintf(adapter->tx_ring->name, "%s-tx-0", netdev->name);
1860         else
1861                 memcpy(adapter->tx_ring->name, netdev->name, IFNAMSIZ);
1862         err = request_irq(adapter->msix_entries[vector].vector,
1863                           e1000_intr_msix_tx, 0, adapter->tx_ring->name,
1864                           netdev);
1865         if (err)
1866                 goto out;
1867         adapter->tx_ring->itr_register = E1000_EITR_82574(vector);
1868         adapter->tx_ring->itr_val = adapter->itr;
1869         vector++;
1870
1871         err = request_irq(adapter->msix_entries[vector].vector,
1872                           e1000_msix_other, 0, netdev->name, netdev);
1873         if (err)
1874                 goto out;
1875
1876         e1000_configure_msix(adapter);
1877         return 0;
1878 out:
1879         return err;
1880 }
1881
1882 /**
1883  * e1000_request_irq - initialize interrupts
1884  *
1885  * Attempts to configure interrupts using the best available
1886  * capabilities of the hardware and kernel.
1887  **/
1888 static int e1000_request_irq(struct e1000_adapter *adapter)
1889 {
1890         struct net_device *netdev = adapter->netdev;
1891         int err;
1892
1893         if (adapter->msix_entries) {
1894                 err = e1000_request_msix(adapter);
1895                 if (!err)
1896                         return err;
1897                 /* fall back to MSI */
1898                 e1000e_reset_interrupt_capability(adapter);
1899                 adapter->int_mode = E1000E_INT_MODE_MSI;
1900                 e1000e_set_interrupt_capability(adapter);
1901         }
1902         if (adapter->flags & FLAG_MSI_ENABLED) {
1903                 err = request_irq(adapter->pdev->irq, e1000_intr_msi, 0,
1904                                   netdev->name, netdev);
1905                 if (!err)
1906                         return err;
1907
1908                 /* fall back to legacy interrupt */
1909                 e1000e_reset_interrupt_capability(adapter);
1910                 adapter->int_mode = E1000E_INT_MODE_LEGACY;
1911         }
1912
1913         err = request_irq(adapter->pdev->irq, e1000_intr, IRQF_SHARED,
1914                           netdev->name, netdev);
1915         if (err)
1916                 e_err("Unable to allocate interrupt, Error: %d\n", err);
1917
1918         return err;
1919 }
1920
1921 static void e1000_free_irq(struct e1000_adapter *adapter)
1922 {
1923         struct net_device *netdev = adapter->netdev;
1924
1925         if (adapter->msix_entries) {
1926                 int vector = 0;
1927
1928                 free_irq(adapter->msix_entries[vector].vector, netdev);
1929                 vector++;
1930
1931                 free_irq(adapter->msix_entries[vector].vector, netdev);
1932                 vector++;
1933
1934                 /* Other Causes interrupt vector */
1935                 free_irq(adapter->msix_entries[vector].vector, netdev);
1936                 return;
1937         }
1938
1939         free_irq(adapter->pdev->irq, netdev);
1940 }
1941
1942 /**
1943  * e1000_irq_disable - Mask off interrupt generation on the NIC
1944  **/
1945 static void e1000_irq_disable(struct e1000_adapter *adapter)
1946 {
1947         struct e1000_hw *hw = &adapter->hw;
1948
1949         ew32(IMC, ~0);
1950         if (adapter->msix_entries)
1951                 ew32(EIAC_82574, 0);
1952         e1e_flush();
1953
1954         if (adapter->msix_entries) {
1955                 int i;
1956                 for (i = 0; i < adapter->num_vectors; i++)
1957                         synchronize_irq(adapter->msix_entries[i].vector);
1958         } else {
1959                 synchronize_irq(adapter->pdev->irq);
1960         }
1961 }
1962
1963 /**
1964  * e1000_irq_enable - Enable default interrupt generation settings
1965  **/
1966 static void e1000_irq_enable(struct e1000_adapter *adapter)
1967 {
1968         struct e1000_hw *hw = &adapter->hw;
1969
1970         if (adapter->msix_entries) {
1971                 ew32(EIAC_82574, adapter->eiac_mask & E1000_EIAC_MASK_82574);
1972                 ew32(IMS, adapter->eiac_mask | E1000_IMS_OTHER | E1000_IMS_LSC);
1973         } else {
1974                 ew32(IMS, IMS_ENABLE_MASK);
1975         }
1976         e1e_flush();
1977 }
1978
1979 /**
1980  * e1000e_get_hw_control - get control of the h/w from f/w
1981  * @adapter: address of board private structure
1982  *
1983  * e1000e_get_hw_control sets {CTRL_EXT|SWSM}:DRV_LOAD bit.
1984  * For ASF and Pass Through versions of f/w this means that
1985  * the driver is loaded. For AMT version (only with 82573)
1986  * of the f/w this means that the network i/f is open.
1987  **/
1988 void e1000e_get_hw_control(struct e1000_adapter *adapter)
1989 {
1990         struct e1000_hw *hw = &adapter->hw;
1991         u32 ctrl_ext;
1992         u32 swsm;
1993
1994         /* Let firmware know the driver has taken over */
1995         if (adapter->flags & FLAG_HAS_SWSM_ON_LOAD) {
1996                 swsm = er32(SWSM);
1997                 ew32(SWSM, swsm | E1000_SWSM_DRV_LOAD);
1998         } else if (adapter->flags & FLAG_HAS_CTRLEXT_ON_LOAD) {
1999                 ctrl_ext = er32(CTRL_EXT);
2000                 ew32(CTRL_EXT, ctrl_ext | E1000_CTRL_EXT_DRV_LOAD);
2001         }
2002 }
2003
2004 /**
2005  * e1000e_release_hw_control - release control of the h/w to f/w
2006  * @adapter: address of board private structure
2007  *
2008  * e1000e_release_hw_control resets {CTRL_EXT|SWSM}:DRV_LOAD bit.
2009  * For ASF and Pass Through versions of f/w this means that the
2010  * driver is no longer loaded. For AMT version (only with 82573) i
2011  * of the f/w this means that the network i/f is closed.
2012  *
2013  **/
2014 void e1000e_release_hw_control(struct e1000_adapter *adapter)
2015 {
2016         struct e1000_hw *hw = &adapter->hw;
2017         u32 ctrl_ext;
2018         u32 swsm;
2019
2020         /* Let firmware taken over control of h/w */
2021         if (adapter->flags & FLAG_HAS_SWSM_ON_LOAD) {
2022                 swsm = er32(SWSM);
2023                 ew32(SWSM, swsm & ~E1000_SWSM_DRV_LOAD);
2024         } else if (adapter->flags & FLAG_HAS_CTRLEXT_ON_LOAD) {
2025                 ctrl_ext = er32(CTRL_EXT);
2026                 ew32(CTRL_EXT, ctrl_ext & ~E1000_CTRL_EXT_DRV_LOAD);
2027         }
2028 }
2029
2030 /**
2031  * @e1000_alloc_ring - allocate memory for a ring structure
2032  **/
2033 static int e1000_alloc_ring_dma(struct e1000_adapter *adapter,
2034                                 struct e1000_ring *ring)
2035 {
2036         struct pci_dev *pdev = adapter->pdev;
2037
2038         ring->desc = dma_alloc_coherent(&pdev->dev, ring->size, &ring->dma,
2039                                         GFP_KERNEL);
2040         if (!ring->desc)
2041                 return -ENOMEM;
2042
2043         return 0;
2044 }
2045
2046 /**
2047  * e1000e_setup_tx_resources - allocate Tx resources (Descriptors)
2048  * @adapter: board private structure
2049  *
2050  * Return 0 on success, negative on failure
2051  **/
2052 int e1000e_setup_tx_resources(struct e1000_adapter *adapter)
2053 {
2054         struct e1000_ring *tx_ring = adapter->tx_ring;
2055         int err = -ENOMEM, size;
2056
2057         size = sizeof(struct e1000_buffer) * tx_ring->count;
2058         tx_ring->buffer_info = vzalloc(size);
2059         if (!tx_ring->buffer_info)
2060                 goto err;
2061
2062         /* round up to nearest 4K */
2063         tx_ring->size = tx_ring->count * sizeof(struct e1000_tx_desc);
2064         tx_ring->size = ALIGN(tx_ring->size, 4096);
2065
2066         err = e1000_alloc_ring_dma(adapter, tx_ring);
2067         if (err)
2068                 goto err;
2069
2070         tx_ring->next_to_use = 0;
2071         tx_ring->next_to_clean = 0;
2072
2073         return 0;
2074 err:
2075         vfree(tx_ring->buffer_info);
2076         e_err("Unable to allocate memory for the transmit descriptor ring\n");
2077         return err;
2078 }
2079
2080 /**
2081  * e1000e_setup_rx_resources - allocate Rx resources (Descriptors)
2082  * @adapter: board private structure
2083  *
2084  * Returns 0 on success, negative on failure
2085  **/
2086 int e1000e_setup_rx_resources(struct e1000_adapter *adapter)
2087 {
2088         struct e1000_ring *rx_ring = adapter->rx_ring;
2089         struct e1000_buffer *buffer_info;
2090         int i, size, desc_len, err = -ENOMEM;
2091
2092         size = sizeof(struct e1000_buffer) * rx_ring->count;
2093         rx_ring->buffer_info = vzalloc(size);
2094         if (!rx_ring->buffer_info)
2095                 goto err;
2096
2097         for (i = 0; i < rx_ring->count; i++) {
2098                 buffer_info = &rx_ring->buffer_info[i];
2099                 buffer_info->ps_pages = kcalloc(PS_PAGE_BUFFERS,
2100                                                 sizeof(struct e1000_ps_page),
2101                                                 GFP_KERNEL);
2102                 if (!buffer_info->ps_pages)
2103                         goto err_pages;
2104         }
2105
2106         desc_len = sizeof(union e1000_rx_desc_packet_split);
2107
2108         /* Round up to nearest 4K */
2109         rx_ring->size = rx_ring->count * desc_len;
2110         rx_ring->size = ALIGN(rx_ring->size, 4096);
2111
2112         err = e1000_alloc_ring_dma(adapter, rx_ring);
2113         if (err)
2114                 goto err_pages;
2115
2116         rx_ring->next_to_clean = 0;
2117         rx_ring->next_to_use = 0;
2118         rx_ring->rx_skb_top = NULL;
2119
2120         return 0;
2121
2122 err_pages:
2123         for (i = 0; i < rx_ring->count; i++) {
2124                 buffer_info = &rx_ring->buffer_info[i];
2125                 kfree(buffer_info->ps_pages);
2126         }
2127 err:
2128         vfree(rx_ring->buffer_info);
2129         e_err("Unable to allocate memory for the receive descriptor ring\n");
2130         return err;
2131 }
2132
2133 /**
2134  * e1000_clean_tx_ring - Free Tx Buffers
2135  * @adapter: board private structure
2136  **/
2137 static void e1000_clean_tx_ring(struct e1000_adapter *adapter)
2138 {
2139         struct e1000_ring *tx_ring = adapter->tx_ring;
2140         struct e1000_buffer *buffer_info;
2141         unsigned long size;
2142         unsigned int i;
2143
2144         for (i = 0; i < tx_ring->count; i++) {
2145                 buffer_info = &tx_ring->buffer_info[i];
2146                 e1000_put_txbuf(adapter, buffer_info);
2147         }
2148
2149         size = sizeof(struct e1000_buffer) * tx_ring->count;
2150         memset(tx_ring->buffer_info, 0, size);
2151
2152         memset(tx_ring->desc, 0, tx_ring->size);
2153
2154         tx_ring->next_to_use = 0;
2155         tx_ring->next_to_clean = 0;
2156
2157         writel(0, adapter->hw.hw_addr + tx_ring->head);
2158         writel(0, adapter->hw.hw_addr + tx_ring->tail);
2159 }
2160
2161 /**
2162  * e1000e_free_tx_resources - Free Tx Resources per Queue
2163  * @adapter: board private structure
2164  *
2165  * Free all transmit software resources
2166  **/
2167 void e1000e_free_tx_resources(struct e1000_adapter *adapter)
2168 {
2169         struct pci_dev *pdev = adapter->pdev;
2170         struct e1000_ring *tx_ring = adapter->tx_ring;
2171
2172         e1000_clean_tx_ring(adapter);
2173
2174         vfree(tx_ring->buffer_info);
2175         tx_ring->buffer_info = NULL;
2176
2177         dma_free_coherent(&pdev->dev, tx_ring->size, tx_ring->desc,
2178                           tx_ring->dma);
2179         tx_ring->desc = NULL;
2180 }
2181
2182 /**
2183  * e1000e_free_rx_resources - Free Rx Resources
2184  * @adapter: board private structure
2185  *
2186  * Free all receive software resources
2187  **/
2188
2189 void e1000e_free_rx_resources(struct e1000_adapter *adapter)
2190 {
2191         struct pci_dev *pdev = adapter->pdev;
2192         struct e1000_ring *rx_ring = adapter->rx_ring;
2193         int i;
2194
2195         e1000_clean_rx_ring(adapter);
2196
2197         for (i = 0; i < rx_ring->count; i++)
2198                 kfree(rx_ring->buffer_info[i].ps_pages);
2199
2200         vfree(rx_ring->buffer_info);
2201         rx_ring->buffer_info = NULL;
2202
2203         dma_free_coherent(&pdev->dev, rx_ring->size, rx_ring->desc,
2204                           rx_ring->dma);
2205         rx_ring->desc = NULL;
2206 }
2207
2208 /**
2209  * e1000_update_itr - update the dynamic ITR value based on statistics
2210  * @adapter: pointer to adapter
2211  * @itr_setting: current adapter->itr
2212  * @packets: the number of packets during this measurement interval
2213  * @bytes: the number of bytes during this measurement interval
2214  *
2215  *      Stores a new ITR value based on packets and byte
2216  *      counts during the last interrupt.  The advantage of per interrupt
2217  *      computation is faster updates and more accurate ITR for the current
2218  *      traffic pattern.  Constants in this function were computed
2219  *      based on theoretical maximum wire speed and thresholds were set based
2220  *      on testing data as well as attempting to minimize response time
2221  *      while increasing bulk throughput.  This functionality is controlled
2222  *      by the InterruptThrottleRate module parameter.
2223  **/
2224 static unsigned int e1000_update_itr(struct e1000_adapter *adapter,
2225                                      u16 itr_setting, int packets,
2226                                      int bytes)
2227 {
2228         unsigned int retval = itr_setting;
2229
2230         if (packets == 0)
2231                 goto update_itr_done;
2232
2233         switch (itr_setting) {
2234         case lowest_latency:
2235                 /* handle TSO and jumbo frames */
2236                 if (bytes/packets > 8000)
2237                         retval = bulk_latency;
2238                 else if ((packets < 5) && (bytes > 512))
2239                         retval = low_latency;
2240                 break;
2241         case low_latency:  /* 50 usec aka 20000 ints/s */
2242                 if (bytes > 10000) {
2243                         /* this if handles the TSO accounting */
2244                         if (bytes/packets > 8000)
2245                                 retval = bulk_latency;
2246                         else if ((packets < 10) || ((bytes/packets) > 1200))
2247                                 retval = bulk_latency;
2248                         else if ((packets > 35))
2249                                 retval = lowest_latency;
2250                 } else if (bytes/packets > 2000) {
2251                         retval = bulk_latency;
2252                 } else if (packets <= 2 && bytes < 512) {
2253                         retval = lowest_latency;
2254                 }
2255                 break;
2256         case bulk_latency: /* 250 usec aka 4000 ints/s */
2257                 if (bytes > 25000) {
2258                         if (packets > 35)
2259                                 retval = low_latency;
2260                 } else if (bytes < 6000) {
2261                         retval = low_latency;
2262                 }
2263                 break;
2264         }
2265
2266 update_itr_done:
2267         return retval;
2268 }
2269
2270 static void e1000_set_itr(struct e1000_adapter *adapter)
2271 {
2272         struct e1000_hw *hw = &adapter->hw;
2273         u16 current_itr;
2274         u32 new_itr = adapter->itr;
2275
2276         /* for non-gigabit speeds, just fix the interrupt rate at 4000 */
2277         if (adapter->link_speed != SPEED_1000) {
2278                 current_itr = 0;
2279                 new_itr = 4000;
2280                 goto set_itr_now;
2281         }
2282
2283         if (adapter->flags2 & FLAG2_DISABLE_AIM) {
2284                 new_itr = 0;
2285                 goto set_itr_now;
2286         }
2287
2288         adapter->tx_itr = e1000_update_itr(adapter,
2289                                     adapter->tx_itr,
2290                                     adapter->total_tx_packets,
2291                                     adapter->total_tx_bytes);
2292         /* conservative mode (itr 3) eliminates the lowest_latency setting */
2293         if (adapter->itr_setting == 3 && adapter->tx_itr == lowest_latency)
2294                 adapter->tx_itr = low_latency;
2295
2296         adapter->rx_itr = e1000_update_itr(adapter,
2297                                     adapter->rx_itr,
2298                                     adapter->total_rx_packets,
2299                                     adapter->total_rx_bytes);
2300         /* conservative mode (itr 3) eliminates the lowest_latency setting */
2301         if (adapter->itr_setting == 3 && adapter->rx_itr == lowest_latency)
2302                 adapter->rx_itr = low_latency;
2303
2304         current_itr = max(adapter->rx_itr, adapter->tx_itr);
2305
2306         switch (current_itr) {
2307         /* counts and packets in update_itr are dependent on these numbers */
2308         case lowest_latency:
2309                 new_itr = 70000;
2310                 break;
2311         case low_latency:
2312                 new_itr = 20000; /* aka hwitr = ~200 */
2313                 break;
2314         case bulk_latency:
2315                 new_itr = 4000;
2316                 break;
2317         default:
2318                 break;
2319         }
2320
2321 set_itr_now:
2322         if (new_itr != adapter->itr) {
2323                 /*
2324                  * this attempts to bias the interrupt rate towards Bulk
2325                  * by adding intermediate steps when interrupt rate is
2326                  * increasing
2327                  */
2328                 new_itr = new_itr > adapter->itr ?
2329                              min(adapter->itr + (new_itr >> 2), new_itr) :
2330                              new_itr;
2331                 adapter->itr = new_itr;
2332                 adapter->rx_ring->itr_val = new_itr;
2333                 if (adapter->msix_entries)
2334                         adapter->rx_ring->set_itr = 1;
2335                 else
2336                         if (new_itr)
2337                                 ew32(ITR, 1000000000 / (new_itr * 256));
2338                         else
2339                                 ew32(ITR, 0);
2340         }
2341 }
2342
2343 /**
2344  * e1000_alloc_queues - Allocate memory for all rings
2345  * @adapter: board private structure to initialize
2346  **/
2347 static int __devinit e1000_alloc_queues(struct e1000_adapter *adapter)
2348 {
2349         adapter->tx_ring = kzalloc(sizeof(struct e1000_ring), GFP_KERNEL);
2350         if (!adapter->tx_ring)
2351                 goto err;
2352
2353         adapter->rx_ring = kzalloc(sizeof(struct e1000_ring), GFP_KERNEL);
2354         if (!adapter->rx_ring)
2355                 goto err;
2356
2357         return 0;
2358 err:
2359         e_err("Unable to allocate memory for queues\n");
2360         kfree(adapter->rx_ring);
2361         kfree(adapter->tx_ring);
2362         return -ENOMEM;
2363 }
2364
2365 /**
2366  * e1000_clean - NAPI Rx polling callback
2367  * @napi: struct associated with this polling callback
2368  * @budget: amount of packets driver is allowed to process this poll
2369  **/
2370 static int e1000_clean(struct napi_struct *napi, int budget)
2371 {
2372         struct e1000_adapter *adapter = container_of(napi, struct e1000_adapter, napi);
2373         struct e1000_hw *hw = &adapter->hw;
2374         struct net_device *poll_dev = adapter->netdev;
2375         int tx_cleaned = 1, work_done = 0;
2376
2377         adapter = netdev_priv(poll_dev);
2378
2379         if (adapter->msix_entries &&
2380             !(adapter->rx_ring->ims_val & adapter->tx_ring->ims_val))
2381                 goto clean_rx;
2382
2383         tx_cleaned = e1000_clean_tx_irq(adapter);
2384
2385 clean_rx:
2386         adapter->clean_rx(adapter, &work_done, budget);
2387
2388         if (!tx_cleaned)
2389                 work_done = budget;
2390
2391         /* If budget not fully consumed, exit the polling mode */
2392         if (work_done < budget) {
2393                 if (adapter->itr_setting & 3)
2394                         e1000_set_itr(adapter);
2395                 napi_complete(napi);
2396                 if (!test_bit(__E1000_DOWN, &adapter->state)) {
2397                         if (adapter->msix_entries)
2398                                 ew32(IMS, adapter->rx_ring->ims_val);
2399                         else
2400                                 e1000_irq_enable(adapter);
2401                 }
2402         }
2403
2404         return work_done;
2405 }
2406
2407 static void e1000_vlan_rx_add_vid(struct net_device *netdev, u16 vid)
2408 {
2409         struct e1000_adapter *adapter = netdev_priv(netdev);
2410         struct e1000_hw *hw = &adapter->hw;
2411         u32 vfta, index;
2412
2413         /* don't update vlan cookie if already programmed */
2414         if ((adapter->hw.mng_cookie.status &
2415              E1000_MNG_DHCP_COOKIE_STATUS_VLAN) &&
2416             (vid == adapter->mng_vlan_id))
2417                 return;
2418
2419         /* add VID to filter table */
2420         if (adapter->flags & FLAG_HAS_HW_VLAN_FILTER) {
2421                 index = (vid >> 5) & 0x7F;
2422                 vfta = E1000_READ_REG_ARRAY(hw, E1000_VFTA, index);
2423                 vfta |= (1 << (vid & 0x1F));
2424                 hw->mac.ops.write_vfta(hw, index, vfta);
2425         }
2426 }
2427
2428 static void e1000_vlan_rx_kill_vid(struct net_device *netdev, u16 vid)
2429 {
2430         struct e1000_adapter *adapter = netdev_priv(netdev);
2431         struct e1000_hw *hw = &adapter->hw;
2432         u32 vfta, index;
2433
2434         if (!test_bit(__E1000_DOWN, &adapter->state))
2435                 e1000_irq_disable(adapter);
2436         vlan_group_set_device(adapter->vlgrp, vid, NULL);
2437
2438         if (!test_bit(__E1000_DOWN, &adapter->state))
2439                 e1000_irq_enable(adapter);
2440
2441         if ((adapter->hw.mng_cookie.status &
2442              E1000_MNG_DHCP_COOKIE_STATUS_VLAN) &&
2443             (vid == adapter->mng_vlan_id)) {
2444                 /* release control to f/w */
2445                 e1000e_release_hw_control(adapter);
2446                 return;
2447         }
2448
2449         /* remove VID from filter table */
2450         if (adapter->flags & FLAG_HAS_HW_VLAN_FILTER) {
2451                 index = (vid >> 5) & 0x7F;
2452                 vfta = E1000_READ_REG_ARRAY(hw, E1000_VFTA, index);
2453                 vfta &= ~(1 << (vid & 0x1F));
2454                 hw->mac.ops.write_vfta(hw, index, vfta);
2455         }
2456 }
2457
2458 static void e1000_update_mng_vlan(struct e1000_adapter *adapter)
2459 {
2460         struct net_device *netdev = adapter->netdev;
2461         u16 vid = adapter->hw.mng_cookie.vlan_id;
2462         u16 old_vid = adapter->mng_vlan_id;
2463
2464         if (!adapter->vlgrp)
2465                 return;
2466
2467         if (!vlan_group_get_device(adapter->vlgrp, vid)) {
2468                 adapter->mng_vlan_id = E1000_MNG_VLAN_NONE;
2469                 if (adapter->hw.mng_cookie.status &
2470                         E1000_MNG_DHCP_COOKIE_STATUS_VLAN) {
2471                         e1000_vlan_rx_add_vid(netdev, vid);
2472                         adapter->mng_vlan_id = vid;
2473                 }
2474
2475                 if ((old_vid != (u16)E1000_MNG_VLAN_NONE) &&
2476                                 (vid != old_vid) &&
2477                     !vlan_group_get_device(adapter->vlgrp, old_vid))
2478                         e1000_vlan_rx_kill_vid(netdev, old_vid);
2479         } else {
2480                 adapter->mng_vlan_id = vid;
2481         }
2482 }
2483
2484
2485 static void e1000_vlan_rx_register(struct net_device *netdev,
2486                                    struct vlan_group *grp)
2487 {
2488         struct e1000_adapter *adapter = netdev_priv(netdev);
2489         struct e1000_hw *hw = &adapter->hw;
2490         u32 ctrl, rctl;
2491
2492         if (!test_bit(__E1000_DOWN, &adapter->state))
2493                 e1000_irq_disable(adapter);
2494         adapter->vlgrp = grp;
2495
2496         if (grp) {
2497                 /* enable VLAN tag insert/strip */
2498                 ctrl = er32(CTRL);
2499                 ctrl |= E1000_CTRL_VME;
2500                 ew32(CTRL, ctrl);
2501
2502                 if (adapter->flags & FLAG_HAS_HW_VLAN_FILTER) {
2503                         /* enable VLAN receive filtering */
2504                         rctl = er32(RCTL);
2505                         rctl &= ~E1000_RCTL_CFIEN;
2506                         ew32(RCTL, rctl);
2507                         e1000_update_mng_vlan(adapter);
2508                 }
2509         } else {
2510                 /* disable VLAN tag insert/strip */
2511                 ctrl = er32(CTRL);
2512                 ctrl &= ~E1000_CTRL_VME;
2513                 ew32(CTRL, ctrl);
2514
2515                 if (adapter->flags & FLAG_HAS_HW_VLAN_FILTER) {
2516                         if (adapter->mng_vlan_id !=
2517                             (u16)E1000_MNG_VLAN_NONE) {
2518                                 e1000_vlan_rx_kill_vid(netdev,
2519                                                        adapter->mng_vlan_id);
2520                                 adapter->mng_vlan_id = E1000_MNG_VLAN_NONE;
2521                         }
2522                 }
2523         }
2524
2525         if (!test_bit(__E1000_DOWN, &adapter->state))
2526                 e1000_irq_enable(adapter);
2527 }
2528
2529 static void e1000_restore_vlan(struct e1000_adapter *adapter)
2530 {
2531         u16 vid;
2532
2533         e1000_vlan_rx_register(adapter->netdev, adapter->vlgrp);
2534
2535         if (!adapter->vlgrp)
2536                 return;
2537
2538         for (vid = 0; vid < VLAN_N_VID; vid++) {
2539                 if (!vlan_group_get_device(adapter->vlgrp, vid))
2540                         continue;
2541                 e1000_vlan_rx_add_vid(adapter->netdev, vid);
2542         }
2543 }
2544
2545 static void e1000_init_manageability_pt(struct e1000_adapter *adapter)
2546 {
2547         struct e1000_hw *hw = &adapter->hw;
2548         u32 manc, manc2h, mdef, i, j;
2549
2550         if (!(adapter->flags & FLAG_MNG_PT_ENABLED))
2551                 return;
2552
2553         manc = er32(MANC);
2554
2555         /*
2556          * enable receiving management packets to the host. this will probably
2557          * generate destination unreachable messages from the host OS, but
2558          * the packets will be handled on SMBUS
2559          */
2560         manc |= E1000_MANC_EN_MNG2HOST;
2561         manc2h = er32(MANC2H);
2562
2563         switch (hw->mac.type) {
2564         default:
2565                 manc2h |= (E1000_MANC2H_PORT_623 | E1000_MANC2H_PORT_664);
2566                 break;
2567         case e1000_82574:
2568         case e1000_82583:
2569                 /*
2570                  * Check if IPMI pass-through decision filter already exists;
2571                  * if so, enable it.
2572                  */
2573                 for (i = 0, j = 0; i < 8; i++) {
2574                         mdef = er32(MDEF(i));
2575
2576                         /* Ignore filters with anything other than IPMI ports */
2577                         if (mdef & ~(E1000_MDEF_PORT_623 | E1000_MDEF_PORT_664))
2578                                 continue;
2579
2580                         /* Enable this decision filter in MANC2H */
2581                         if (mdef)
2582                                 manc2h |= (1 << i);
2583
2584                         j |= mdef;
2585                 }
2586
2587                 if (j == (E1000_MDEF_PORT_623 | E1000_MDEF_PORT_664))
2588                         break;
2589
2590                 /* Create new decision filter in an empty filter */
2591                 for (i = 0, j = 0; i < 8; i++)
2592                         if (er32(MDEF(i)) == 0) {
2593                                 ew32(MDEF(i), (E1000_MDEF_PORT_623 |
2594                                                E1000_MDEF_PORT_664));
2595                                 manc2h |= (1 << 1);
2596                                 j++;
2597                                 break;
2598                         }
2599
2600                 if (!j)
2601                         e_warn("Unable to create IPMI pass-through filter\n");
2602                 break;
2603         }
2604
2605         ew32(MANC2H, manc2h);
2606         ew32(MANC, manc);
2607 }
2608
2609 /**
2610  * e1000_configure_tx - Configure Transmit Unit after Reset
2611  * @adapter: board private structure
2612  *
2613  * Configure the Tx unit of the MAC after a reset.
2614  **/
2615 static void e1000_configure_tx(struct e1000_adapter *adapter)
2616 {
2617         struct e1000_hw *hw = &adapter->hw;
2618         struct e1000_ring *tx_ring = adapter->tx_ring;
2619         u64 tdba;
2620         u32 tdlen, tctl, tipg, tarc;
2621         u32 ipgr1, ipgr2;
2622
2623         /* Setup the HW Tx Head and Tail descriptor pointers */
2624         tdba = tx_ring->dma;
2625         tdlen = tx_ring->count * sizeof(struct e1000_tx_desc);
2626         ew32(TDBAL, (tdba & DMA_BIT_MASK(32)));
2627         ew32(TDBAH, (tdba >> 32));
2628         ew32(TDLEN, tdlen);
2629         ew32(TDH, 0);
2630         ew32(TDT, 0);
2631         tx_ring->head = E1000_TDH;
2632         tx_ring->tail = E1000_TDT;
2633
2634         /* Set the default values for the Tx Inter Packet Gap timer */
2635         tipg = DEFAULT_82543_TIPG_IPGT_COPPER;          /*  8  */
2636         ipgr1 = DEFAULT_82543_TIPG_IPGR1;               /*  8  */
2637         ipgr2 = DEFAULT_82543_TIPG_IPGR2;               /*  6  */
2638
2639         if (adapter->flags & FLAG_TIPG_MEDIUM_FOR_80003ESLAN)
2640                 ipgr2 = DEFAULT_80003ES2LAN_TIPG_IPGR2; /*  7  */
2641
2642         tipg |= ipgr1 << E1000_TIPG_IPGR1_SHIFT;
2643         tipg |= ipgr2 << E1000_TIPG_IPGR2_SHIFT;
2644         ew32(TIPG, tipg);
2645
2646         /* Set the Tx Interrupt Delay register */
2647         ew32(TIDV, adapter->tx_int_delay);
2648         /* Tx irq moderation */
2649         ew32(TADV, adapter->tx_abs_int_delay);
2650
2651         if (adapter->flags2 & FLAG2_DMA_BURST) {
2652                 u32 txdctl = er32(TXDCTL(0));
2653                 txdctl &= ~(E1000_TXDCTL_PTHRESH | E1000_TXDCTL_HTHRESH |
2654                             E1000_TXDCTL_WTHRESH);
2655                 /*
2656                  * set up some performance related parameters to encourage the
2657                  * hardware to use the bus more efficiently in bursts, depends
2658                  * on the tx_int_delay to be enabled,
2659                  * wthresh = 5 ==> burst write a cacheline (64 bytes) at a time
2660                  * hthresh = 1 ==> prefetch when one or more available
2661                  * pthresh = 0x1f ==> prefetch if internal cache 31 or less
2662                  * BEWARE: this seems to work but should be considered first if
2663                  * there are Tx hangs or other Tx related bugs
2664                  */
2665                 txdctl |= E1000_TXDCTL_DMA_BURST_ENABLE;
2666                 ew32(TXDCTL(0), txdctl);
2667                 /* erratum work around: set txdctl the same for both queues */
2668                 ew32(TXDCTL(1), txdctl);
2669         }
2670
2671         /* Program the Transmit Control Register */
2672         tctl = er32(TCTL);
2673         tctl &= ~E1000_TCTL_CT;
2674         tctl |= E1000_TCTL_PSP | E1000_TCTL_RTLC |
2675                 (E1000_COLLISION_THRESHOLD << E1000_CT_SHIFT);
2676
2677         if (adapter->flags & FLAG_TARC_SPEED_MODE_BIT) {
2678                 tarc = er32(TARC(0));
2679                 /*
2680                  * set the speed mode bit, we'll clear it if we're not at
2681                  * gigabit link later
2682                  */
2683 #define SPEED_MODE_BIT (1 << 21)
2684                 tarc |= SPEED_MODE_BIT;
2685                 ew32(TARC(0), tarc);
2686         }
2687
2688         /* errata: program both queues to unweighted RR */
2689         if (adapter->flags & FLAG_TARC_SET_BIT_ZERO) {
2690                 tarc = er32(TARC(0));
2691                 tarc |= 1;
2692                 ew32(TARC(0), tarc);
2693                 tarc = er32(TARC(1));
2694                 tarc |= 1;
2695                 ew32(TARC(1), tarc);
2696         }
2697
2698         /* Setup Transmit Descriptor Settings for eop descriptor */
2699         adapter->txd_cmd = E1000_TXD_CMD_EOP | E1000_TXD_CMD_IFCS;
2700
2701         /* only set IDE if we are delaying interrupts using the timers */
2702         if (adapter->tx_int_delay)
2703                 adapter->txd_cmd |= E1000_TXD_CMD_IDE;
2704
2705         /* enable Report Status bit */
2706         adapter->txd_cmd |= E1000_TXD_CMD_RS;
2707
2708         ew32(TCTL, tctl);
2709
2710         e1000e_config_collision_dist(hw);
2711 }
2712
2713 /**
2714  * e1000_setup_rctl - configure the receive control registers
2715  * @adapter: Board private structure
2716  **/
2717 #define PAGE_USE_COUNT(S) (((S) >> PAGE_SHIFT) + \
2718                            (((S) & (PAGE_SIZE - 1)) ? 1 : 0))
2719 static void e1000_setup_rctl(struct e1000_adapter *adapter)
2720 {
2721         struct e1000_hw *hw = &adapter->hw;
2722         u32 rctl, rfctl;
2723         u32 pages = 0;
2724
2725         /* Workaround Si errata on 82579 - configure jumbo frame flow */
2726         if (hw->mac.type == e1000_pch2lan) {
2727                 s32 ret_val;
2728
2729                 if (adapter->netdev->mtu > ETH_DATA_LEN)
2730                         ret_val = e1000_lv_jumbo_workaround_ich8lan(hw, true);
2731                 else
2732                         ret_val = e1000_lv_jumbo_workaround_ich8lan(hw, false);
2733
2734                 if (ret_val)
2735                         e_dbg("failed to enable jumbo frame workaround mode\n");
2736         }
2737
2738         /* Program MC offset vector base */
2739         rctl = er32(RCTL);
2740         rctl &= ~(3 << E1000_RCTL_MO_SHIFT);
2741         rctl |= E1000_RCTL_EN | E1000_RCTL_BAM |
2742                 E1000_RCTL_LBM_NO | E1000_RCTL_RDMTS_HALF |
2743                 (adapter->hw.mac.mc_filter_type << E1000_RCTL_MO_SHIFT);
2744
2745         /* Do not Store bad packets */
2746         rctl &= ~E1000_RCTL_SBP;
2747
2748         /* Enable Long Packet receive */
2749         if (adapter->netdev->mtu <= ETH_DATA_LEN)
2750                 rctl &= ~E1000_RCTL_LPE;
2751         else
2752                 rctl |= E1000_RCTL_LPE;
2753
2754         /* Some systems expect that the CRC is included in SMBUS traffic. The
2755          * hardware strips the CRC before sending to both SMBUS (BMC) and to
2756          * host memory when this is enabled
2757          */
2758         if (adapter->flags2 & FLAG2_CRC_STRIPPING)
2759                 rctl |= E1000_RCTL_SECRC;
2760
2761         /* Workaround Si errata on 82577 PHY - configure IPG for jumbos */
2762         if ((hw->phy.type == e1000_phy_82577) && (rctl & E1000_RCTL_LPE)) {
2763                 u16 phy_data;
2764
2765                 e1e_rphy(hw, PHY_REG(770, 26), &phy_data);
2766                 phy_data &= 0xfff8;
2767                 phy_data |= (1 << 2);
2768                 e1e_wphy(hw, PHY_REG(770, 26), phy_data);
2769
2770                 e1e_rphy(hw, 22, &phy_data);
2771                 phy_data &= 0x0fff;
2772                 phy_data |= (1 << 14);
2773                 e1e_wphy(hw, 0x10, 0x2823);
2774                 e1e_wphy(hw, 0x11, 0x0003);
2775                 e1e_wphy(hw, 22, phy_data);
2776         }
2777
2778         /* Setup buffer sizes */
2779         rctl &= ~E1000_RCTL_SZ_4096;
2780         rctl |= E1000_RCTL_BSEX;
2781         switch (adapter->rx_buffer_len) {
2782         case 2048:
2783         default:
2784                 rctl |= E1000_RCTL_SZ_2048;
2785                 rctl &= ~E1000_RCTL_BSEX;
2786                 break;
2787         case 4096:
2788                 rctl |= E1000_RCTL_SZ_4096;
2789                 break;
2790         case 8192:
2791                 rctl |= E1000_RCTL_SZ_8192;
2792                 break;
2793         case 16384:
2794                 rctl |= E1000_RCTL_SZ_16384;
2795                 break;
2796         }
2797
2798         /*
2799          * 82571 and greater support packet-split where the protocol
2800          * header is placed in skb->data and the packet data is
2801          * placed in pages hanging off of skb_shinfo(skb)->nr_frags.
2802          * In the case of a non-split, skb->data is linearly filled,
2803          * followed by the page buffers.  Therefore, skb->data is
2804          * sized to hold the largest protocol header.
2805          *
2806          * allocations using alloc_page take too long for regular MTU
2807          * so only enable packet split for jumbo frames
2808          *
2809          * Using pages when the page size is greater than 16k wastes
2810          * a lot of memory, since we allocate 3 pages at all times
2811          * per packet.
2812          */
2813         pages = PAGE_USE_COUNT(adapter->netdev->mtu);
2814         if (!(adapter->flags & FLAG_HAS_ERT) && (pages <= 3) &&
2815             (PAGE_SIZE <= 16384) && (rctl & E1000_RCTL_LPE))
2816                 adapter->rx_ps_pages = pages;
2817         else
2818                 adapter->rx_ps_pages = 0;
2819
2820         if (adapter->rx_ps_pages) {
2821                 u32 psrctl = 0;
2822
2823                 /* Configure extra packet-split registers */
2824                 rfctl = er32(RFCTL);
2825                 rfctl |= E1000_RFCTL_EXTEN;
2826                 /*
2827                  * disable packet split support for IPv6 extension headers,
2828                  * because some malformed IPv6 headers can hang the Rx
2829                  */
2830                 rfctl |= (E1000_RFCTL_IPV6_EX_DIS |
2831                           E1000_RFCTL_NEW_IPV6_EXT_DIS);
2832
2833                 ew32(RFCTL, rfctl);
2834
2835                 /* Enable Packet split descriptors */
2836                 rctl |= E1000_RCTL_DTYP_PS;
2837
2838                 psrctl |= adapter->rx_ps_bsize0 >>
2839                         E1000_PSRCTL_BSIZE0_SHIFT;
2840
2841                 switch (adapter->rx_ps_pages) {
2842                 case 3:
2843                         psrctl |= PAGE_SIZE <<
2844                                 E1000_PSRCTL_BSIZE3_SHIFT;
2845                 case 2:
2846                         psrctl |= PAGE_SIZE <<
2847                                 E1000_PSRCTL_BSIZE2_SHIFT;
2848                 case 1:
2849                         psrctl |= PAGE_SIZE >>
2850                                 E1000_PSRCTL_BSIZE1_SHIFT;
2851                         break;
2852                 }
2853
2854                 ew32(PSRCTL, psrctl);
2855         }
2856
2857         ew32(RCTL, rctl);
2858         /* just started the receive unit, no need to restart */
2859         adapter->flags &= ~FLAG_RX_RESTART_NOW;
2860 }
2861
2862 /**
2863  * e1000_configure_rx - Configure Receive Unit after Reset
2864  * @adapter: board private structure
2865  *
2866  * Configure the Rx unit of the MAC after a reset.
2867  **/
2868 static void e1000_configure_rx(struct e1000_adapter *adapter)
2869 {
2870         struct e1000_hw *hw = &adapter->hw;
2871         struct e1000_ring *rx_ring = adapter->rx_ring;
2872         u64 rdba;
2873         u32 rdlen, rctl, rxcsum, ctrl_ext;
2874
2875         if (adapter->rx_ps_pages) {
2876                 /* this is a 32 byte descriptor */
2877                 rdlen = rx_ring->count *
2878                     sizeof(union e1000_rx_desc_packet_split);
2879                 adapter->clean_rx = e1000_clean_rx_irq_ps;
2880                 adapter->alloc_rx_buf = e1000_alloc_rx_buffers_ps;
2881         } else if (adapter->netdev->mtu > ETH_FRAME_LEN + ETH_FCS_LEN) {
2882                 rdlen = rx_ring->count * sizeof(struct e1000_rx_desc);
2883                 adapter->clean_rx = e1000_clean_jumbo_rx_irq;
2884                 adapter->alloc_rx_buf = e1000_alloc_jumbo_rx_buffers;
2885         } else {
2886                 rdlen = rx_ring->count * sizeof(struct e1000_rx_desc);
2887                 adapter->clean_rx = e1000_clean_rx_irq;
2888                 adapter->alloc_rx_buf = e1000_alloc_rx_buffers;
2889         }
2890
2891         /* disable receives while setting up the descriptors */
2892         rctl = er32(RCTL);
2893         ew32(RCTL, rctl & ~E1000_RCTL_EN);
2894         e1e_flush();
2895         msleep(10);
2896
2897         if (adapter->flags2 & FLAG2_DMA_BURST) {
2898                 /*
2899                  * set the writeback threshold (only takes effect if the RDTR
2900                  * is set). set GRAN=1 and write back up to 0x4 worth, and
2901                  * enable prefetching of 0x20 Rx descriptors
2902                  * granularity = 01
2903                  * wthresh = 04,
2904                  * hthresh = 04,
2905                  * pthresh = 0x20
2906                  */
2907                 ew32(RXDCTL(0), E1000_RXDCTL_DMA_BURST_ENABLE);
2908                 ew32(RXDCTL(1), E1000_RXDCTL_DMA_BURST_ENABLE);
2909
2910                 /*
2911                  * override the delay timers for enabling bursting, only if
2912                  * the value was not set by the user via module options
2913                  */
2914                 if (adapter->rx_int_delay == DEFAULT_RDTR)
2915                         adapter->rx_int_delay = BURST_RDTR;
2916                 if (adapter->rx_abs_int_delay == DEFAULT_RADV)
2917                         adapter->rx_abs_int_delay = BURST_RADV;
2918         }
2919
2920         /* set the Receive Delay Timer Register */
2921         ew32(RDTR, adapter->rx_int_delay);
2922
2923         /* irq moderation */
2924         ew32(RADV, adapter->rx_abs_int_delay);
2925         if ((adapter->itr_setting != 0) && (adapter->itr != 0))
2926                 ew32(ITR, 1000000000 / (adapter->itr * 256));
2927
2928         ctrl_ext = er32(CTRL_EXT);
2929         /* Auto-Mask interrupts upon ICR access */
2930         ctrl_ext |= E1000_CTRL_EXT_IAME;
2931         ew32(IAM, 0xffffffff);
2932         ew32(CTRL_EXT, ctrl_ext);
2933         e1e_flush();
2934
2935         /*
2936          * Setup the HW Rx Head and Tail Descriptor Pointers and
2937          * the Base and Length of the Rx Descriptor Ring
2938          */
2939         rdba = rx_ring->dma;
2940         ew32(RDBAL, (rdba & DMA_BIT_MASK(32)));
2941         ew32(RDBAH, (rdba >> 32));
2942         ew32(RDLEN, rdlen);
2943         ew32(RDH, 0);
2944         ew32(RDT, 0);
2945         rx_ring->head = E1000_RDH;
2946         rx_ring->tail = E1000_RDT;
2947
2948         /* Enable Receive Checksum Offload for TCP and UDP */
2949         rxcsum = er32(RXCSUM);
2950         if (adapter->flags & FLAG_RX_CSUM_ENABLED) {
2951                 rxcsum |= E1000_RXCSUM_TUOFL;
2952
2953                 /*
2954                  * IPv4 payload checksum for UDP fragments must be
2955                  * used in conjunction with packet-split.
2956                  */
2957                 if (adapter->rx_ps_pages)
2958                         rxcsum |= E1000_RXCSUM_IPPCSE;
2959         } else {
2960                 rxcsum &= ~E1000_RXCSUM_TUOFL;
2961                 /* no need to clear IPPCSE as it defaults to 0 */
2962         }
2963         ew32(RXCSUM, rxcsum);
2964
2965         /*
2966          * Enable early receives on supported devices, only takes effect when
2967          * packet size is equal or larger than the specified value (in 8 byte
2968          * units), e.g. using jumbo frames when setting to E1000_ERT_2048
2969          */
2970         if ((adapter->flags & FLAG_HAS_ERT) ||
2971             (adapter->hw.mac.type == e1000_pch2lan)) {
2972                 if (adapter->netdev->mtu > ETH_DATA_LEN) {
2973                         u32 rxdctl = er32(RXDCTL(0));
2974                         ew32(RXDCTL(0), rxdctl | 0x3);
2975                         if (adapter->flags & FLAG_HAS_ERT)
2976                                 ew32(ERT, E1000_ERT_2048 | (1 << 13));
2977                         /*
2978                          * With jumbo frames and early-receive enabled,
2979                          * excessive C-state transition latencies result in
2980                          * dropped transactions.
2981                          */
2982                         pm_qos_update_request(&adapter->netdev->pm_qos_req, 55);
2983                 } else {
2984                         pm_qos_update_request(&adapter->netdev->pm_qos_req,
2985                                               PM_QOS_DEFAULT_VALUE);
2986                 }
2987         }
2988
2989         /* Enable Receives */
2990         ew32(RCTL, rctl);
2991 }
2992
2993 /**
2994  *  e1000_update_mc_addr_list - Update Multicast addresses
2995  *  @hw: pointer to the HW structure
2996  *  @mc_addr_list: array of multicast addresses to program
2997  *  @mc_addr_count: number of multicast addresses to program
2998  *
2999  *  Updates the Multicast Table Array.
3000  *  The caller must have a packed mc_addr_list of multicast addresses.
3001  **/
3002 static void e1000_update_mc_addr_list(struct e1000_hw *hw, u8 *mc_addr_list,
3003                                       u32 mc_addr_count)
3004 {
3005         hw->mac.ops.update_mc_addr_list(hw, mc_addr_list, mc_addr_count);
3006 }
3007
3008 /**
3009  * e1000_set_multi - Multicast and Promiscuous mode set
3010  * @netdev: network interface device structure
3011  *
3012  * The set_multi entry point is called whenever the multicast address
3013  * list or the network interface flags are updated.  This routine is
3014  * responsible for configuring the hardware for proper multicast,
3015  * promiscuous mode, and all-multi behavior.
3016  **/
3017 static void e1000_set_multi(struct net_device *netdev)
3018 {
3019         struct e1000_adapter *adapter = netdev_priv(netdev);
3020         struct e1000_hw *hw = &adapter->hw;
3021         struct netdev_hw_addr *ha;
3022         u8  *mta_list;
3023         u32 rctl;
3024
3025         /* Check for Promiscuous and All Multicast modes */
3026
3027         rctl = er32(RCTL);
3028
3029         if (netdev->flags & IFF_PROMISC) {
3030                 rctl |= (E1000_RCTL_UPE | E1000_RCTL_MPE);
3031                 rctl &= ~E1000_RCTL_VFE;
3032         } else {
3033                 if (netdev->flags & IFF_ALLMULTI) {
3034                         rctl |= E1000_RCTL_MPE;
3035                         rctl &= ~E1000_RCTL_UPE;
3036                 } else {
3037                         rctl &= ~(E1000_RCTL_UPE | E1000_RCTL_MPE);
3038                 }
3039                 if (adapter->flags & FLAG_HAS_HW_VLAN_FILTER)
3040                         rctl |= E1000_RCTL_VFE;
3041         }
3042
3043         ew32(RCTL, rctl);
3044
3045         if (!netdev_mc_empty(netdev)) {
3046                 int i = 0;
3047
3048                 mta_list = kmalloc(netdev_mc_count(netdev) * 6, GFP_ATOMIC);
3049                 if (!mta_list)
3050                         return;
3051
3052                 /* prepare a packed array of only addresses. */
3053                 netdev_for_each_mc_addr(ha, netdev)
3054                         memcpy(mta_list + (i++ * ETH_ALEN), ha->addr, ETH_ALEN);
3055
3056                 e1000_update_mc_addr_list(hw, mta_list, i);
3057                 kfree(mta_list);
3058         } else {
3059                 /*
3060                  * if we're called from probe, we might not have
3061                  * anything to do here, so clear out the list
3062                  */
3063                 e1000_update_mc_addr_list(hw, NULL, 0);
3064         }
3065 }
3066
3067 /**
3068  * e1000_configure - configure the hardware for Rx and Tx
3069  * @adapter: private board structure
3070  **/
3071 static void e1000_configure(struct e1000_adapter *adapter)
3072 {
3073         e1000_set_multi(adapter->netdev);
3074
3075         e1000_restore_vlan(adapter);
3076         e1000_init_manageability_pt(adapter);
3077
3078         e1000_configure_tx(adapter);
3079         e1000_setup_rctl(adapter);
3080         e1000_configure_rx(adapter);
3081         adapter->alloc_rx_buf(adapter, e1000_desc_unused(adapter->rx_ring));
3082 }
3083
3084 /**
3085  * e1000e_power_up_phy - restore link in case the phy was powered down
3086  * @adapter: address of board private structure
3087  *
3088  * The phy may be powered down to save power and turn off link when the
3089  * driver is unloaded and wake on lan is not enabled (among others)
3090  * *** this routine MUST be followed by a call to e1000e_reset ***
3091  **/
3092 void e1000e_power_up_phy(struct e1000_adapter *adapter)
3093 {
3094         if (adapter->hw.phy.ops.power_up)
3095                 adapter->hw.phy.ops.power_up(&adapter->hw);
3096
3097         adapter->hw.mac.ops.setup_link(&adapter->hw);
3098 }
3099
3100 /**
3101  * e1000_power_down_phy - Power down the PHY
3102  *
3103  * Power down the PHY so no link is implied when interface is down.
3104  * The PHY cannot be powered down if management or WoL is active.
3105  */
3106 static void e1000_power_down_phy(struct e1000_adapter *adapter)
3107 {
3108         /* WoL is enabled */
3109         if (adapter->wol)
3110                 return;
3111
3112         if (adapter->hw.phy.ops.power_down)
3113                 adapter->hw.phy.ops.power_down(&adapter->hw);
3114 }
3115
3116 /**
3117  * e1000e_reset - bring the hardware into a known good state
3118  *
3119  * This function boots the hardware and enables some settings that
3120  * require a configuration cycle of the hardware - those cannot be
3121  * set/changed during runtime. After reset the device needs to be
3122  * properly configured for Rx, Tx etc.
3123  */
3124 void e1000e_reset(struct e1000_adapter *adapter)
3125 {
3126         struct e1000_mac_info *mac = &adapter->hw.mac;
3127         struct e1000_fc_info *fc = &adapter->hw.fc;
3128         struct e1000_hw *hw = &adapter->hw;
3129         u32 tx_space, min_tx_space, min_rx_space;
3130         u32 pba = adapter->pba;
3131         u16 hwm;
3132
3133         /* reset Packet Buffer Allocation to default */
3134         ew32(PBA, pba);
3135
3136         if (adapter->max_frame_size > ETH_FRAME_LEN + ETH_FCS_LEN) {
3137                 /*
3138                  * To maintain wire speed transmits, the Tx FIFO should be
3139                  * large enough to accommodate two full transmit packets,
3140                  * rounded up to the next 1KB and expressed in KB.  Likewise,
3141                  * the Rx FIFO should be large enough to accommodate at least
3142                  * one full receive packet and is similarly rounded up and
3143                  * expressed in KB.
3144                  */
3145                 pba = er32(PBA);
3146                 /* upper 16 bits has Tx packet buffer allocation size in KB */
3147                 tx_space = pba >> 16;
3148                 /* lower 16 bits has Rx packet buffer allocation size in KB */
3149                 pba &= 0xffff;
3150                 /*
3151                  * the Tx fifo also stores 16 bytes of information about the Tx
3152                  * but don't include ethernet FCS because hardware appends it
3153                  */
3154                 min_tx_space = (adapter->max_frame_size +
3155                                 sizeof(struct e1000_tx_desc) -
3156                                 ETH_FCS_LEN) * 2;
3157                 min_tx_space = ALIGN(min_tx_space, 1024);
3158                 min_tx_space >>= 10;
3159                 /* software strips receive CRC, so leave room for it */
3160                 min_rx_space = adapter->max_frame_size;
3161                 min_rx_space = ALIGN(min_rx_space, 1024);
3162                 min_rx_space >>= 10;
3163
3164                 /*
3165                  * If current Tx allocation is less than the min Tx FIFO size,
3166                  * and the min Tx FIFO size is less than the current Rx FIFO
3167                  * allocation, take space away from current Rx allocation
3168                  */
3169                 if ((tx_space < min_tx_space) &&
3170                     ((min_tx_space - tx_space) < pba)) {
3171                         pba -= min_tx_space - tx_space;
3172
3173                         /*
3174                          * if short on Rx space, Rx wins and must trump Tx
3175                          * adjustment or use Early Receive if available
3176                          */
3177                         if ((pba < min_rx_space) &&
3178                             (!(adapter->flags & FLAG_HAS_ERT)))
3179                                 /* ERT enabled in e1000_configure_rx */
3180                                 pba = min_rx_space;
3181                 }
3182
3183                 ew32(PBA, pba);
3184         }
3185
3186         /*
3187          * flow control settings
3188          *
3189          * The high water mark must be low enough to fit one full frame
3190          * (or the size used for early receive) above it in the Rx FIFO.
3191          * Set it to the lower of:
3192          * - 90% of the Rx FIFO size, and
3193          * - the full Rx FIFO size minus the early receive size (for parts
3194          *   with ERT support assuming ERT set to E1000_ERT_2048), or
3195          * - the full Rx FIFO size minus one full frame
3196          */
3197         if (adapter->flags & FLAG_DISABLE_FC_PAUSE_TIME)
3198                 fc->pause_time = 0xFFFF;
3199         else
3200                 fc->pause_time = E1000_FC_PAUSE_TIME;
3201         fc->send_xon = 1;
3202         fc->current_mode = fc->requested_mode;
3203
3204         switch (hw->mac.type) {
3205         default:
3206                 if ((adapter->flags & FLAG_HAS_ERT) &&
3207                     (adapter->netdev->mtu > ETH_DATA_LEN))
3208                         hwm = min(((pba << 10) * 9 / 10),
3209                                   ((pba << 10) - (E1000_ERT_2048 << 3)));
3210                 else
3211                         hwm = min(((pba << 10) * 9 / 10),
3212                                   ((pba << 10) - adapter->max_frame_size));
3213
3214                 fc->high_water = hwm & E1000_FCRTH_RTH; /* 8-byte granularity */
3215                 fc->low_water = fc->high_water - 8;
3216                 break;
3217         case e1000_pchlan:
3218                 /*
3219                  * Workaround PCH LOM adapter hangs with certain network
3220                  * loads.  If hangs persist, try disabling Tx flow control.
3221                  */
3222                 if (adapter->netdev->mtu > ETH_DATA_LEN) {
3223                         fc->high_water = 0x3500;
3224                         fc->low_water  = 0x1500;
3225                 } else {
3226                         fc->high_water = 0x5000;
3227                         fc->low_water  = 0x3000;
3228                 }
3229                 fc->refresh_time = 0x1000;
3230                 break;
3231         case e1000_pch2lan:
3232                 fc->high_water = 0x05C20;
3233                 fc->low_water = 0x05048;
3234                 fc->pause_time = 0x0650;
3235                 fc->refresh_time = 0x0400;
3236                 if (adapter->netdev->mtu > ETH_DATA_LEN) {
3237                         pba = 14;
3238                         ew32(PBA, pba);
3239                 }
3240                 break;
3241         }
3242
3243         /*
3244          * Disable Adaptive Interrupt Moderation if 2 full packets cannot
3245          * fit in receive buffer and early-receive not supported.
3246          */
3247         if (adapter->itr_setting & 0x3) {
3248                 if (((adapter->max_frame_size * 2) > (pba << 10)) &&
3249                     !(adapter->flags & FLAG_HAS_ERT)) {
3250                         if (!(adapter->flags2 & FLAG2_DISABLE_AIM)) {
3251                                 dev_info(&adapter->pdev->dev,
3252                                         "Interrupt Throttle Rate turned off\n");
3253                                 adapter->flags2 |= FLAG2_DISABLE_AIM;
3254                                 ew32(ITR, 0);
3255                         }
3256                 } else if (adapter->flags2 & FLAG2_DISABLE_AIM) {
3257                         dev_info(&adapter->pdev->dev,
3258                                  "Interrupt Throttle Rate turned on\n");
3259                         adapter->flags2 &= ~FLAG2_DISABLE_AIM;
3260                         adapter->itr = 20000;
3261                         ew32(ITR, 1000000000 / (adapter->itr * 256));
3262                 }
3263         }
3264
3265         /* Allow time for pending master requests to run */
3266         mac->ops.reset_hw(hw);
3267
3268         /*
3269          * For parts with AMT enabled, let the firmware know
3270          * that the network interface is in control
3271          */
3272         if (adapter->flags & FLAG_HAS_AMT)
3273                 e1000e_get_hw_control(adapter);
3274
3275         ew32(WUC, 0);
3276
3277         if (mac->ops.init_hw(hw))
3278                 e_err("Hardware Error\n");
3279
3280         e1000_update_mng_vlan(adapter);
3281
3282         /* Enable h/w to recognize an 802.1Q VLAN Ethernet packet */
3283         ew32(VET, ETH_P_8021Q);
3284
3285         e1000e_reset_adaptive(hw);
3286
3287         if (!netif_running(adapter->netdev) &&
3288             !test_bit(__E1000_TESTING, &adapter->state)) {
3289                 e1000_power_down_phy(adapter);
3290                 return;
3291         }
3292
3293         e1000_get_phy_info(hw);
3294
3295         if ((adapter->flags & FLAG_HAS_SMART_POWER_DOWN) &&
3296             !(adapter->flags & FLAG_SMART_POWER_DOWN)) {
3297                 u16 phy_data = 0;
3298                 /*
3299                  * speed up time to link by disabling smart power down, ignore
3300                  * the return value of this function because there is nothing
3301                  * different we would do if it failed
3302                  */
3303                 e1e_rphy(hw, IGP02E1000_PHY_POWER_MGMT, &phy_data);
3304                 phy_data &= ~IGP02E1000_PM_SPD;
3305                 e1e_wphy(hw, IGP02E1000_PHY_POWER_MGMT, phy_data);
3306         }
3307 }
3308
3309 int e1000e_up(struct e1000_adapter *adapter)
3310 {
3311         struct e1000_hw *hw = &adapter->hw;
3312
3313         /* hardware has been reset, we need to reload some things */
3314         e1000_configure(adapter);
3315
3316         clear_bit(__E1000_DOWN, &adapter->state);
3317
3318         napi_enable(&adapter->napi);
3319         if (adapter->msix_entries)
3320                 e1000_configure_msix(adapter);
3321         e1000_irq_enable(adapter);
3322
3323         netif_wake_queue(adapter->netdev);
3324
3325         /* fire a link change interrupt to start the watchdog */
3326         if (adapter->msix_entries)
3327                 ew32(ICS, E1000_ICS_LSC | E1000_ICR_OTHER);
3328         else
3329                 ew32(ICS, E1000_ICS_LSC);
3330
3331         return 0;
3332 }
3333
3334 static void e1000e_update_stats(struct e1000_adapter *adapter);
3335
3336 void e1000e_down(struct e1000_adapter *adapter)
3337 {
3338         struct net_device *netdev = adapter->netdev;
3339         struct e1000_hw *hw = &adapter->hw;
3340         u32 tctl, rctl;
3341
3342         /*
3343          * signal that we're down so the interrupt handler does not
3344          * reschedule our watchdog timer
3345          */
3346         set_bit(__E1000_DOWN, &adapter->state);
3347
3348         /* disable receives in the hardware */
3349         rctl = er32(RCTL);
3350         ew32(RCTL, rctl & ~E1000_RCTL_EN);
3351         /* flush and sleep below */
3352
3353         netif_stop_queue(netdev);
3354
3355         /* disable transmits in the hardware */
3356         tctl = er32(TCTL);
3357         tctl &= ~E1000_TCTL_EN;
3358         ew32(TCTL, tctl);
3359         /* flush both disables and wait for them to finish */
3360         e1e_flush();
3361         msleep(10);
3362
3363         napi_disable(&adapter->napi);
3364         e1000_irq_disable(adapter);
3365
3366         del_timer_sync(&adapter->watchdog_timer);
3367         del_timer_sync(&adapter->phy_info_timer);
3368
3369         netif_carrier_off(netdev);
3370
3371         spin_lock(&adapter->stats64_lock);
3372         e1000e_update_stats(adapter);
3373         spin_unlock(&adapter->stats64_lock);
3374
3375         adapter->link_speed = 0;
3376         adapter->link_duplex = 0;
3377
3378         if (!pci_channel_offline(adapter->pdev))
3379                 e1000e_reset(adapter);
3380         e1000_clean_tx_ring(adapter);
3381         e1000_clean_rx_ring(adapter);
3382
3383         /*
3384          * TODO: for power management, we could drop the link and
3385          * pci_disable_device here.
3386          */
3387 }
3388
3389 void e1000e_reinit_locked(struct e1000_adapter *adapter)
3390 {
3391         might_sleep();
3392         while (test_and_set_bit(__E1000_RESETTING, &adapter->state))
3393                 msleep(1);
3394         e1000e_down(adapter);
3395         e1000e_up(adapter);
3396         clear_bit(__E1000_RESETTING, &adapter->state);
3397 }
3398
3399 /**
3400  * e1000_sw_init - Initialize general software structures (struct e1000_adapter)
3401  * @adapter: board private structure to initialize
3402  *
3403  * e1000_sw_init initializes the Adapter private data structure.
3404  * Fields are initialized based on PCI device information and
3405  * OS network device settings (MTU size).
3406  **/
3407 static int __devinit e1000_sw_init(struct e1000_adapter *adapter)
3408 {
3409         struct net_device *netdev = adapter->netdev;
3410
3411         adapter->rx_buffer_len = ETH_FRAME_LEN + VLAN_HLEN + ETH_FCS_LEN;
3412         adapter->rx_ps_bsize0 = 128;
3413         adapter->max_frame_size = netdev->mtu + ETH_HLEN + ETH_FCS_LEN;
3414         adapter->min_frame_size = ETH_ZLEN + ETH_FCS_LEN;
3415
3416         spin_lock_init(&adapter->stats64_lock);
3417
3418         e1000e_set_interrupt_capability(adapter);
3419
3420         if (e1000_alloc_queues(adapter))
3421                 return -ENOMEM;
3422
3423         /* Explicitly disable IRQ since the NIC can be in any state. */
3424         e1000_irq_disable(adapter);
3425
3426         set_bit(__E1000_DOWN, &adapter->state);
3427         return 0;
3428 }
3429
3430 /**
3431  * e1000_intr_msi_test - Interrupt Handler
3432  * @irq: interrupt number
3433  * @data: pointer to a network interface device structure
3434  **/
3435 static irqreturn_t e1000_intr_msi_test(int irq, void *data)
3436 {
3437         struct net_device *netdev = data;
3438         struct e1000_adapter *adapter = netdev_priv(netdev);
3439         struct e1000_hw *hw = &adapter->hw;
3440         u32 icr = er32(ICR);
3441
3442         e_dbg("icr is %08X\n", icr);
3443         if (icr & E1000_ICR_RXSEQ) {
3444                 adapter->flags &= ~FLAG_MSI_TEST_FAILED;
3445                 wmb();
3446         }
3447
3448         return IRQ_HANDLED;
3449 }
3450
3451 /**
3452  * e1000_test_msi_interrupt - Returns 0 for successful test
3453  * @adapter: board private struct
3454  *
3455  * code flow taken from tg3.c
3456  **/
3457 static int e1000_test_msi_interrupt(struct e1000_adapter *adapter)
3458 {
3459         struct net_device *netdev = adapter->netdev;
3460         struct e1000_hw *hw = &adapter->hw;
3461         int err;
3462
3463         /* poll_enable hasn't been called yet, so don't need disable */
3464         /* clear any pending events */
3465         er32(ICR);
3466
3467         /* free the real vector and request a test handler */
3468         e1000_free_irq(adapter);
3469         e1000e_reset_interrupt_capability(adapter);
3470
3471         /* Assume that the test fails, if it succeeds then the test
3472          * MSI irq handler will unset this flag */
3473         adapter->flags |= FLAG_MSI_TEST_FAILED;
3474
3475         err = pci_enable_msi(adapter->pdev);
3476         if (err)
3477                 goto msi_test_failed;
3478
3479         err = request_irq(adapter->pdev->irq, e1000_intr_msi_test, 0,
3480                           netdev->name, netdev);
3481         if (err) {
3482                 pci_disable_msi(adapter->pdev);
3483                 goto msi_test_failed;
3484         }
3485
3486         wmb();
3487
3488         e1000_irq_enable(adapter);
3489
3490         /* fire an unusual interrupt on the test handler */
3491         ew32(ICS, E1000_ICS_RXSEQ);
3492         e1e_flush();
3493         msleep(50);
3494
3495         e1000_irq_disable(adapter);
3496
3497         rmb();
3498
3499         if (adapter->flags & FLAG_MSI_TEST_FAILED) {
3500                 adapter->int_mode = E1000E_INT_MODE_LEGACY;
3501                 e_info("MSI interrupt test failed, using legacy interrupt.\n");
3502         } else
3503                 e_dbg("MSI interrupt test succeeded!\n");
3504
3505         free_irq(adapter->pdev->irq, netdev);
3506         pci_disable_msi(adapter->pdev);
3507
3508 msi_test_failed:
3509         e1000e_set_interrupt_capability(adapter);
3510         return e1000_request_irq(adapter);
3511 }
3512
3513 /**
3514  * e1000_test_msi - Returns 0 if MSI test succeeds or INTx mode is restored
3515  * @adapter: board private struct
3516  *
3517  * code flow taken from tg3.c, called with e1000 interrupts disabled.
3518  **/
3519 static int e1000_test_msi(struct e1000_adapter *adapter)
3520 {
3521         int err;
3522         u16 pci_cmd;
3523
3524         if (!(adapter->flags & FLAG_MSI_ENABLED))
3525                 return 0;
3526
3527         /* disable SERR in case the MSI write causes a master abort */
3528         pci_read_config_word(adapter->pdev, PCI_COMMAND, &pci_cmd);
3529         if (pci_cmd & PCI_COMMAND_SERR)
3530                 pci_write_config_word(adapter->pdev, PCI_COMMAND,
3531                                       pci_cmd & ~PCI_COMMAND_SERR);
3532
3533         err = e1000_test_msi_interrupt(adapter);
3534
3535         /* re-enable SERR */
3536         if (pci_cmd & PCI_COMMAND_SERR) {
3537                 pci_read_config_word(adapter->pdev, PCI_COMMAND, &pci_cmd);
3538                 pci_cmd |= PCI_COMMAND_SERR;
3539                 pci_write_config_word(adapter->pdev, PCI_COMMAND, pci_cmd);
3540         }
3541
3542         return err;
3543 }
3544
3545 /**
3546  * e1000_open - Called when a network interface is made active
3547  * @netdev: network interface device structure
3548  *
3549  * Returns 0 on success, negative value on failure
3550  *
3551  * The open entry point is called when a network interface is made
3552  * active by the system (IFF_UP).  At this point all resources needed
3553  * for transmit and receive operations are allocated, the interrupt
3554  * handler is registered with the OS, the watchdog timer is started,
3555  * and the stack is notified that the interface is ready.
3556  **/
3557 static int e1000_open(struct net_device *netdev)
3558 {
3559         struct e1000_adapter *adapter = netdev_priv(netdev);
3560         struct e1000_hw *hw = &adapter->hw;
3561         struct pci_dev *pdev = adapter->pdev;
3562         int err;
3563
3564         /* disallow open during test */
3565         if (test_bit(__E1000_TESTING, &adapter->state))
3566                 return -EBUSY;
3567
3568         pm_runtime_get_sync(&pdev->dev);
3569
3570         netif_carrier_off(netdev);
3571
3572         /* allocate transmit descriptors */
3573         err = e1000e_setup_tx_resources(adapter);
3574         if (err)
3575                 goto err_setup_tx;
3576
3577         /* allocate receive descriptors */
3578         err = e1000e_setup_rx_resources(adapter);
3579         if (err)
3580                 goto err_setup_rx;
3581
3582         /*
3583          * If AMT is enabled, let the firmware know that the network
3584          * interface is now open and reset the part to a known state.
3585          */
3586         if (adapter->flags & FLAG_HAS_AMT) {
3587                 e1000e_get_hw_control(adapter);
3588                 e1000e_reset(adapter);
3589         }
3590
3591         e1000e_power_up_phy(adapter);
3592
3593         adapter->mng_vlan_id = E1000_MNG_VLAN_NONE;
3594         if ((adapter->hw.mng_cookie.status &
3595              E1000_MNG_DHCP_COOKIE_STATUS_VLAN))
3596                 e1000_update_mng_vlan(adapter);
3597
3598         /* DMA latency requirement to workaround early-receive/jumbo issue */
3599         if ((adapter->flags & FLAG_HAS_ERT) ||
3600             (adapter->hw.mac.type == e1000_pch2lan))
3601                 pm_qos_add_request(&adapter->netdev->pm_qos_req,
3602                                    PM_QOS_CPU_DMA_LATENCY,
3603                                    PM_QOS_DEFAULT_VALUE);
3604
3605         /*
3606          * before we allocate an interrupt, we must be ready to handle it.
3607          * Setting DEBUG_SHIRQ in the kernel makes it fire an interrupt
3608          * as soon as we call pci_request_irq, so we have to setup our
3609          * clean_rx handler before we do so.
3610          */
3611         e1000_configure(adapter);
3612
3613         err = e1000_request_irq(adapter);
3614         if (err)
3615                 goto err_req_irq;
3616
3617         /*
3618          * Work around PCIe errata with MSI interrupts causing some chipsets to
3619          * ignore e1000e MSI messages, which means we need to test our MSI
3620          * interrupt now
3621          */
3622         if (adapter->int_mode != E1000E_INT_MODE_LEGACY) {
3623                 err = e1000_test_msi(adapter);
3624                 if (err) {
3625                         e_err("Interrupt allocation failed\n");
3626                         goto err_req_irq;
3627                 }
3628         }
3629
3630         /* From here on the code is the same as e1000e_up() */
3631         clear_bit(__E1000_DOWN, &adapter->state);
3632
3633         napi_enable(&adapter->napi);
3634
3635         e1000_irq_enable(adapter);
3636
3637         netif_start_queue(netdev);
3638
3639         adapter->idle_check = true;
3640         pm_runtime_put(&pdev->dev);
3641
3642         /* fire a link status change interrupt to start the watchdog */
3643         if (adapter->msix_entries)
3644                 ew32(ICS, E1000_ICS_LSC | E1000_ICR_OTHER);
3645         else
3646                 ew32(ICS, E1000_ICS_LSC);
3647
3648         return 0;
3649
3650 err_req_irq:
3651         e1000e_release_hw_control(adapter);
3652         e1000_power_down_phy(adapter);
3653         e1000e_free_rx_resources(adapter);
3654 err_setup_rx:
3655         e1000e_free_tx_resources(adapter);
3656 err_setup_tx:
3657         e1000e_reset(adapter);
3658         pm_runtime_put_sync(&pdev->dev);
3659
3660         return err;
3661 }
3662
3663 /**
3664  * e1000_close - Disables a network interface
3665  * @netdev: network interface device structure
3666  *
3667  * Returns 0, this is not allowed to fail
3668  *
3669  * The close entry point is called when an interface is de-activated
3670  * by the OS.  The hardware is still under the drivers control, but
3671  * needs to be disabled.  A global MAC reset is issued to stop the
3672  * hardware, and all transmit and receive resources are freed.
3673  **/
3674 static int e1000_close(struct net_device *netdev)
3675 {
3676         struct e1000_adapter *adapter = netdev_priv(netdev);
3677         struct pci_dev *pdev = adapter->pdev;
3678
3679         WARN_ON(test_bit(__E1000_RESETTING, &adapter->state));
3680
3681         pm_runtime_get_sync(&pdev->dev);
3682
3683         if (!test_bit(__E1000_DOWN, &adapter->state)) {
3684                 e1000e_down(adapter);
3685                 e1000_free_irq(adapter);
3686         }
3687         e1000_power_down_phy(adapter);
3688
3689         e1000e_free_tx_resources(adapter);
3690         e1000e_free_rx_resources(adapter);
3691
3692         /*
3693          * kill manageability vlan ID if supported, but not if a vlan with
3694          * the same ID is registered on the host OS (let 8021q kill it)
3695          */
3696         if ((adapter->hw.mng_cookie.status &
3697                           E1000_MNG_DHCP_COOKIE_STATUS_VLAN) &&
3698              !(adapter->vlgrp &&
3699                vlan_group_get_device(adapter->vlgrp, adapter->mng_vlan_id)))
3700                 e1000_vlan_rx_kill_vid(netdev, adapter->mng_vlan_id);
3701
3702         /*
3703          * If AMT is enabled, let the firmware know that the network
3704          * interface is now closed
3705          */
3706         if ((adapter->flags & FLAG_HAS_AMT) &&
3707             !test_bit(__E1000_TESTING, &adapter->state))
3708                 e1000e_release_hw_control(adapter);
3709
3710         if ((adapter->flags & FLAG_HAS_ERT) ||
3711             (adapter->hw.mac.type == e1000_pch2lan))
3712                 pm_qos_remove_request(&adapter->netdev->pm_qos_req);
3713
3714         pm_runtime_put_sync(&pdev->dev);
3715
3716         return 0;
3717 }
3718 /**
3719  * e1000_set_mac - Change the Ethernet Address of the NIC
3720  * @netdev: network interface device structure
3721  * @p: pointer to an address structure
3722  *
3723  * Returns 0 on success, negative on failure
3724  **/
3725 static int e1000_set_mac(struct net_device *netdev, void *p)
3726 {
3727         struct e1000_adapter *adapter = netdev_priv(netdev);
3728         struct sockaddr *addr = p;
3729
3730         if (!is_valid_ether_addr(addr->sa_data))
3731                 return -EADDRNOTAVAIL;
3732
3733         memcpy(netdev->dev_addr, addr->sa_data, netdev->addr_len);
3734         memcpy(adapter->hw.mac.addr, addr->sa_data, netdev->addr_len);
3735
3736         e1000e_rar_set(&adapter->hw, adapter->hw.mac.addr, 0);
3737
3738         if (adapter->flags & FLAG_RESET_OVERWRITES_LAA) {
3739                 /* activate the work around */
3740                 e1000e_set_laa_state_82571(&adapter->hw, 1);
3741
3742                 /*
3743                  * Hold a copy of the LAA in RAR[14] This is done so that
3744                  * between the time RAR[0] gets clobbered  and the time it
3745                  * gets fixed (in e1000_watchdog), the actual LAA is in one
3746                  * of the RARs and no incoming packets directed to this port
3747                  * are dropped. Eventually the LAA will be in RAR[0] and
3748                  * RAR[14]
3749                  */
3750                 e1000e_rar_set(&adapter->hw,
3751                               adapter->hw.mac.addr,
3752                               adapter->hw.mac.rar_entry_count - 1);
3753         }
3754
3755         return 0;
3756 }
3757
3758 /**
3759  * e1000e_update_phy_task - work thread to update phy
3760  * @work: pointer to our work struct
3761  *
3762  * this worker thread exists because we must acquire a
3763  * semaphore to read the phy, which we could msleep while
3764  * waiting for it, and we can't msleep in a timer.
3765  **/
3766 static void e1000e_update_phy_task(struct work_struct *work)
3767 {
3768         struct e1000_adapter *adapter = container_of(work,
3769                                         struct e1000_adapter, update_phy_task);
3770         e1000_get_phy_info(&adapter->hw);
3771 }
3772
3773 /*
3774  * Need to wait a few seconds after link up to get diagnostic information from
3775  * the phy
3776  */
3777 static void e1000_update_phy_info(unsigned long data)
3778 {
3779         struct e1000_adapter *adapter = (struct e1000_adapter *) data;
3780         schedule_work(&adapter->update_phy_task);
3781 }
3782
3783 /**
3784  * e1000e_update_phy_stats - Update the PHY statistics counters
3785  * @adapter: board private structure
3786  **/
3787 static void e1000e_update_phy_stats(struct e1000_adapter *adapter)
3788 {
3789         struct e1000_hw *hw = &adapter->hw;
3790         s32 ret_val;
3791         u16 phy_data;
3792
3793         ret_val = hw->phy.ops.acquire(hw);
3794         if (ret_val)
3795                 return;
3796
3797         hw->phy.addr = 1;
3798
3799 #define HV_PHY_STATS_PAGE       778
3800         /*
3801          * A page set is expensive so check if already on desired page.
3802          * If not, set to the page with the PHY status registers.
3803          */
3804         ret_val = e1000e_read_phy_reg_mdic(hw, IGP01E1000_PHY_PAGE_SELECT,
3805                                            &phy_data);
3806         if (ret_val)
3807                 goto release;
3808         if (phy_data != (HV_PHY_STATS_PAGE << IGP_PAGE_SHIFT)) {
3809                 ret_val = e1000e_write_phy_reg_mdic(hw,
3810                                                     IGP01E1000_PHY_PAGE_SELECT,
3811                                                     (HV_PHY_STATS_PAGE <<
3812                                                      IGP_PAGE_SHIFT));
3813                 if (ret_val)
3814                         goto release;
3815         }
3816
3817         /* Read/clear the upper 16-bit registers and read/accumulate lower */
3818
3819         /* Single Collision Count */
3820         e1000e_read_phy_reg_mdic(hw, HV_SCC_UPPER & MAX_PHY_REG_ADDRESS,
3821                                  &phy_data);
3822         ret_val = e1000e_read_phy_reg_mdic(hw,
3823                                            HV_SCC_LOWER & MAX_PHY_REG_ADDRESS,
3824                                            &phy_data);
3825         if (!ret_val)
3826                 adapter->stats.scc += phy_data;
3827
3828         /* Excessive Collision Count */
3829         e1000e_read_phy_reg_mdic(hw, HV_ECOL_UPPER & MAX_PHY_REG_ADDRESS,
3830                                  &phy_data);
3831         ret_val = e1000e_read_phy_reg_mdic(hw,
3832                                            HV_ECOL_LOWER & MAX_PHY_REG_ADDRESS,
3833                                            &phy_data);
3834         if (!ret_val)
3835                 adapter->stats.ecol += phy_data;
3836
3837         /* Multiple Collision Count */
3838         e1000e_read_phy_reg_mdic(hw, HV_MCC_UPPER & MAX_PHY_REG_ADDRESS,
3839                                  &phy_data);
3840         ret_val = e1000e_read_phy_reg_mdic(hw,
3841                                            HV_MCC_LOWER & MAX_PHY_REG_ADDRESS,
3842                                            &phy_data);
3843         if (!ret_val)
3844                 adapter->stats.mcc += phy_data;
3845
3846         /* Late Collision Count */
3847         e1000e_read_phy_reg_mdic(hw, HV_LATECOL_UPPER & MAX_PHY_REG_ADDRESS,
3848                                  &phy_data);
3849         ret_val = e1000e_read_phy_reg_mdic(hw,
3850                                            HV_LATECOL_LOWER &
3851                                            MAX_PHY_REG_ADDRESS,
3852                                            &phy_data);
3853         if (!ret_val)
3854                 adapter->stats.latecol += phy_data;
3855
3856         /* Collision Count - also used for adaptive IFS */
3857         e1000e_read_phy_reg_mdic(hw, HV_COLC_UPPER & MAX_PHY_REG_ADDRESS,
3858                                  &phy_data);
3859         ret_val = e1000e_read_phy_reg_mdic(hw,
3860                                            HV_COLC_LOWER & MAX_PHY_REG_ADDRESS,
3861                                            &phy_data);
3862         if (!ret_val)
3863                 hw->mac.collision_delta = phy_data;
3864
3865         /* Defer Count */
3866         e1000e_read_phy_reg_mdic(hw, HV_DC_UPPER & MAX_PHY_REG_ADDRESS,
3867                                  &phy_data);
3868         ret_val = e1000e_read_phy_reg_mdic(hw,
3869                                            HV_DC_LOWER & MAX_PHY_REG_ADDRESS,
3870                                            &phy_data);
3871         if (!ret_val)
3872                 adapter->stats.dc += phy_data;
3873
3874         /* Transmit with no CRS */
3875         e1000e_read_phy_reg_mdic(hw, HV_TNCRS_UPPER & MAX_PHY_REG_ADDRESS,
3876                                  &phy_data);
3877         ret_val = e1000e_read_phy_reg_mdic(hw,
3878                                            HV_TNCRS_LOWER & MAX_PHY_REG_ADDRESS,
3879                                            &phy_data);
3880         if (!ret_val)
3881                 adapter->stats.tncrs += phy_data;
3882
3883 release:
3884         hw->phy.ops.release(hw);
3885 }
3886
3887 /**
3888  * e1000e_update_stats - Update the board statistics counters
3889  * @adapter: board private structure
3890  **/
3891 static void e1000e_update_stats(struct e1000_adapter *adapter)
3892 {
3893         struct net_device *netdev = adapter->netdev;
3894         struct e1000_hw *hw = &adapter->hw;
3895         struct pci_dev *pdev = adapter->pdev;
3896
3897         /*
3898          * Prevent stats update while adapter is being reset, or if the pci
3899          * connection is down.
3900          */
3901         if (adapter->link_speed == 0)
3902                 return;
3903         if (pci_channel_offline(pdev))
3904                 return;
3905
3906         adapter->stats.crcerrs += er32(CRCERRS);
3907         adapter->stats.gprc += er32(GPRC);
3908         adapter->stats.gorc += er32(GORCL);
3909         er32(GORCH); /* Clear gorc */
3910         adapter->stats.bprc += er32(BPRC);
3911         adapter->stats.mprc += er32(MPRC);
3912         adapter->stats.roc += er32(ROC);
3913
3914         adapter->stats.mpc += er32(MPC);
3915
3916         /* Half-duplex statistics */
3917         if (adapter->link_duplex == HALF_DUPLEX) {
3918                 if (adapter->flags2 & FLAG2_HAS_PHY_STATS) {
3919                         e1000e_update_phy_stats(adapter);
3920                 } else {
3921                         adapter->stats.scc += er32(SCC);
3922                         adapter->stats.ecol += er32(ECOL);
3923                         adapter->stats.mcc += er32(MCC);
3924                         adapter->stats.latecol += er32(LATECOL);
3925                         adapter->stats.dc += er32(DC);
3926
3927                         hw->mac.collision_delta = er32(COLC);
3928
3929                         if ((hw->mac.type != e1000_82574) &&
3930                             (hw->mac.type != e1000_82583))
3931                                 adapter->stats.tncrs += er32(TNCRS);
3932                 }
3933                 adapter->stats.colc += hw->mac.collision_delta;
3934         }
3935
3936         adapter->stats.xonrxc += er32(XONRXC);
3937         adapter->stats.xontxc += er32(XONTXC);
3938         adapter->stats.xoffrxc += er32(XOFFRXC);
3939         adapter->stats.xofftxc += er32(XOFFTXC);
3940         adapter->stats.gptc += er32(GPTC);
3941         adapter->stats.gotc += er32(GOTCL);
3942         er32(GOTCH); /* Clear gotc */
3943         adapter->stats.rnbc += er32(RNBC);
3944         adapter->stats.ruc += er32(RUC);
3945
3946         adapter->stats.mptc += er32(MPTC);
3947         adapter->stats.bptc += er32(BPTC);
3948
3949         /* used for adaptive IFS */
3950
3951         hw->mac.tx_packet_delta = er32(TPT);
3952         adapter->stats.tpt += hw->mac.tx_packet_delta;
3953
3954         adapter->stats.algnerrc += er32(ALGNERRC);
3955         adapter->stats.rxerrc += er32(RXERRC);
3956         adapter->stats.cexterr += er32(CEXTERR);
3957         adapter->stats.tsctc += er32(TSCTC);
3958         adapter->stats.tsctfc += er32(TSCTFC);
3959
3960         /* Fill out the OS statistics structure */
3961         netdev->stats.multicast = adapter->stats.mprc;
3962         netdev->stats.collisions = adapter->stats.colc;
3963
3964         /* Rx Errors */
3965
3966         /*
3967          * RLEC on some newer hardware can be incorrect so build
3968          * our own version based on RUC and ROC
3969          */
3970         netdev->stats.rx_errors = adapter->stats.rxerrc +
3971                 adapter->stats.crcerrs + adapter->stats.algnerrc +
3972                 adapter->stats.ruc + adapter->stats.roc +
3973                 adapter->stats.cexterr;
3974         netdev->stats.rx_length_errors = adapter->stats.ruc +
3975                                               adapter->stats.roc;
3976         netdev->stats.rx_crc_errors = adapter->stats.crcerrs;
3977         netdev->stats.rx_frame_errors = adapter->stats.algnerrc;
3978         netdev->stats.rx_missed_errors = adapter->stats.mpc;
3979
3980         /* Tx Errors */
3981         netdev->stats.tx_errors = adapter->stats.ecol +
3982                                        adapter->stats.latecol;
3983         netdev->stats.tx_aborted_errors = adapter->stats.ecol;
3984         netdev->stats.tx_window_errors = adapter->stats.latecol;
3985         netdev->stats.tx_carrier_errors = adapter->stats.tncrs;
3986
3987         /* Tx Dropped needs to be maintained elsewhere */
3988
3989         /* Management Stats */
3990         adapter->stats.mgptc += er32(MGTPTC);
3991         adapter->stats.mgprc += er32(MGTPRC);
3992         adapter->stats.mgpdc += er32(MGTPDC);
3993 }
3994
3995 /**
3996  * e1000_phy_read_status - Update the PHY register status snapshot
3997  * @adapter: board private structure
3998  **/
3999 static void e1000_phy_read_status(struct e1000_adapter *adapter)
4000 {
4001         struct e1000_hw *hw = &adapter->hw;
4002         struct e1000_phy_regs *phy = &adapter->phy_regs;
4003
4004         if ((er32(STATUS) & E1000_STATUS_LU) &&
4005             (adapter->hw.phy.media_type == e1000_media_type_copper)) {
4006                 int ret_val;
4007
4008                 ret_val  = e1e_rphy(hw, PHY_CONTROL, &phy->bmcr);
4009                 ret_val |= e1e_rphy(hw, PHY_STATUS, &phy->bmsr);
4010                 ret_val |= e1e_rphy(hw, PHY_AUTONEG_ADV, &phy->advertise);
4011                 ret_val |= e1e_rphy(hw, PHY_LP_ABILITY, &phy->lpa);
4012                 ret_val |= e1e_rphy(hw, PHY_AUTONEG_EXP, &phy->expansion);
4013                 ret_val |= e1e_rphy(hw, PHY_1000T_CTRL, &phy->ctrl1000);
4014                 ret_val |= e1e_rphy(hw, PHY_1000T_STATUS, &phy->stat1000);
4015                 ret_val |= e1e_rphy(hw, PHY_EXT_STATUS, &phy->estatus);
4016                 if (ret_val)
4017                         e_warn("Error reading PHY register\n");
4018         } else {
4019                 /*
4020                  * Do not read PHY registers if link is not up
4021                  * Set values to typical power-on defaults
4022                  */
4023                 phy->bmcr = (BMCR_SPEED1000 | BMCR_ANENABLE | BMCR_FULLDPLX);
4024                 phy->bmsr = (BMSR_100FULL | BMSR_100HALF | BMSR_10FULL |
4025                              BMSR_10HALF | BMSR_ESTATEN | BMSR_ANEGCAPABLE |
4026                              BMSR_ERCAP);
4027                 phy->advertise = (ADVERTISE_PAUSE_ASYM | ADVERTISE_PAUSE_CAP |
4028                                   ADVERTISE_ALL | ADVERTISE_CSMA);
4029                 phy->lpa = 0;
4030                 phy->expansion = EXPANSION_ENABLENPAGE;
4031                 phy->ctrl1000 = ADVERTISE_1000FULL;
4032                 phy->stat1000 = 0;
4033                 phy->estatus = (ESTATUS_1000_TFULL | ESTATUS_1000_THALF);
4034         }
4035 }
4036
4037 static void e1000_print_link_info(struct e1000_adapter *adapter)
4038 {
4039         struct e1000_hw *hw = &adapter->hw;
4040         u32 ctrl = er32(CTRL);
4041
4042         /* Link status message must follow this format for user tools */
4043         printk(KERN_INFO "e1000e: %s NIC Link is Up %d Mbps %s, "
4044                "Flow Control: %s\n",
4045                adapter->netdev->name,
4046                adapter->link_speed,
4047                (adapter->link_duplex == FULL_DUPLEX) ?
4048                "Full Duplex" : "Half Duplex",
4049                ((ctrl & E1000_CTRL_TFCE) && (ctrl & E1000_CTRL_RFCE)) ?
4050                "Rx/Tx" :
4051                ((ctrl & E1000_CTRL_RFCE) ? "Rx" :
4052                 ((ctrl & E1000_CTRL_TFCE) ? "Tx" : "None")));
4053 }
4054
4055 static bool e1000e_has_link(struct e1000_adapter *adapter)
4056 {
4057         struct e1000_hw *hw = &adapter->hw;
4058         bool link_active = 0;
4059         s32 ret_val = 0;
4060
4061         /*
4062          * get_link_status is set on LSC (link status) interrupt or
4063          * Rx sequence error interrupt.  get_link_status will stay
4064          * false until the check_for_link establishes link
4065          * for copper adapters ONLY
4066          */
4067         switch (hw->phy.media_type) {
4068         case e1000_media_type_copper:
4069                 if (hw->mac.get_link_status) {
4070                         ret_val = hw->mac.ops.check_for_link(hw);
4071                         link_active = !hw->mac.get_link_status;
4072                 } else {
4073                         link_active = 1;
4074                 }
4075                 break;
4076         case e1000_media_type_fiber:
4077                 ret_val = hw->mac.ops.check_for_link(hw);
4078                 link_active = !!(er32(STATUS) & E1000_STATUS_LU);
4079                 break;
4080         case e1000_media_type_internal_serdes:
4081                 ret_val = hw->mac.ops.check_for_link(hw);
4082                 link_active = adapter->hw.mac.serdes_has_link;
4083                 break;
4084         default:
4085         case e1000_media_type_unknown:
4086                 break;
4087         }
4088
4089         if ((ret_val == E1000_ERR_PHY) && (hw->phy.type == e1000_phy_igp_3) &&
4090             (er32(CTRL) & E1000_PHY_CTRL_GBE_DISABLE)) {
4091                 /* See e1000_kmrn_lock_loss_workaround_ich8lan() */
4092                 e_info("Gigabit has been disabled, downgrading speed\n");
4093         }
4094
4095         return link_active;
4096 }
4097
4098 static void e1000e_enable_receives(struct e1000_adapter *adapter)
4099 {
4100         /* make sure the receive unit is started */
4101         if ((adapter->flags & FLAG_RX_NEEDS_RESTART) &&
4102             (adapter->flags & FLAG_RX_RESTART_NOW)) {
4103                 struct e1000_hw *hw = &adapter->hw;
4104                 u32 rctl = er32(RCTL);
4105                 ew32(RCTL, rctl | E1000_RCTL_EN);
4106                 adapter->flags &= ~FLAG_RX_RESTART_NOW;
4107         }
4108 }
4109
4110 static void e1000e_check_82574_phy_workaround(struct e1000_adapter *adapter)
4111 {
4112         struct e1000_hw *hw = &adapter->hw;
4113
4114         /*
4115          * With 82574 controllers, PHY needs to be checked periodically
4116          * for hung state and reset, if two calls return true
4117          */
4118         if (e1000_check_phy_82574(hw))
4119                 adapter->phy_hang_count++;
4120         else
4121                 adapter->phy_hang_count = 0;
4122
4123         if (adapter->phy_hang_count > 1) {
4124                 adapter->phy_hang_count = 0;
4125                 schedule_work(&adapter->reset_task);
4126         }
4127 }
4128
4129 /**
4130  * e1000_watchdog - Timer Call-back
4131  * @data: pointer to adapter cast into an unsigned long
4132  **/
4133 static void e1000_watchdog(unsigned long data)
4134 {
4135         struct e1000_adapter *adapter = (struct e1000_adapter *) data;
4136
4137         /* Do the rest outside of interrupt context */
4138         schedule_work(&adapter->watchdog_task);
4139
4140         /* TODO: make this use queue_delayed_work() */
4141 }
4142
4143 static void e1000_watchdog_task(struct work_struct *work)
4144 {
4145         struct e1000_adapter *adapter = container_of(work,
4146                                         struct e1000_adapter, watchdog_task);
4147         struct net_device *netdev = adapter->netdev;
4148         struct e1000_mac_info *mac = &adapter->hw.mac;
4149         struct e1000_phy_info *phy = &adapter->hw.phy;
4150         struct e1000_ring *tx_ring = adapter->tx_ring;
4151         struct e1000_hw *hw = &adapter->hw;
4152         u32 link, tctl;
4153
4154         link = e1000e_has_link(adapter);
4155         if ((netif_carrier_ok(netdev)) && link) {
4156                 /* Cancel scheduled suspend requests. */
4157                 pm_runtime_resume(netdev->dev.parent);
4158
4159                 e1000e_enable_receives(adapter);
4160                 goto link_up;
4161         }
4162
4163         if ((e1000e_enable_tx_pkt_filtering(hw)) &&
4164             (adapter->mng_vlan_id != adapter->hw.mng_cookie.vlan_id))
4165                 e1000_update_mng_vlan(adapter);
4166
4167         if (link) {
4168                 if (!netif_carrier_ok(netdev)) {
4169                         bool txb2b = 1;
4170
4171                         /* Cancel scheduled suspend requests. */
4172                         pm_runtime_resume(netdev->dev.parent);
4173
4174                         /* update snapshot of PHY registers on LSC */
4175                         e1000_phy_read_status(adapter);
4176                         mac->ops.get_link_up_info(&adapter->hw,
4177                                                    &adapter->link_speed,
4178                                                    &adapter->link_duplex);
4179                         e1000_print_link_info(adapter);
4180                         /*
4181                          * On supported PHYs, check for duplex mismatch only
4182                          * if link has autonegotiated at 10/100 half
4183                          */
4184                         if ((hw->phy.type == e1000_phy_igp_3 ||
4185                              hw->phy.type == e1000_phy_bm) &&
4186                             (hw->mac.autoneg == true) &&
4187                             (adapter->link_speed == SPEED_10 ||
4188                              adapter->link_speed == SPEED_100) &&
4189                             (adapter->link_duplex == HALF_DUPLEX)) {
4190                                 u16 autoneg_exp;
4191
4192                                 e1e_rphy(hw, PHY_AUTONEG_EXP, &autoneg_exp);
4193
4194                                 if (!(autoneg_exp & NWAY_ER_LP_NWAY_CAPS))
4195                                         e_info("Autonegotiated half duplex but"
4196                                                " link partner cannot autoneg. "
4197                                                " Try forcing full duplex if "
4198                                                "link gets many collisions.\n");
4199                         }
4200
4201                         /* adjust timeout factor according to speed/duplex */
4202                         adapter->tx_timeout_factor = 1;
4203                         switch (adapter->link_speed) {
4204                         case SPEED_10:
4205                                 txb2b = 0;
4206                                 adapter->tx_timeout_factor = 16;
4207                                 break;
4208                         case SPEED_100:
4209                                 txb2b = 0;
4210                                 adapter->tx_timeout_factor = 10;
4211                                 break;
4212                         }
4213
4214                         /*
4215                          * workaround: re-program speed mode bit after
4216                          * link-up event
4217                          */
4218                         if ((adapter->flags & FLAG_TARC_SPEED_MODE_BIT) &&
4219                             !txb2b) {
4220                                 u32 tarc0;
4221                                 tarc0 = er32(TARC(0));
4222                                 tarc0 &= ~SPEED_MODE_BIT;
4223                                 ew32(TARC(0), tarc0);
4224                         }
4225
4226                         /*
4227                          * disable TSO for pcie and 10/100 speeds, to avoid
4228                          * some hardware issues
4229                          */
4230                         if (!(adapter->flags & FLAG_TSO_FORCE)) {
4231                                 switch (adapter->link_speed) {
4232                                 case SPEED_10:
4233                                 case SPEED_100:
4234                                         e_info("10/100 speed: disabling TSO\n");
4235                                         netdev->features &= ~NETIF_F_TSO;
4236                                         netdev->features &= ~NETIF_F_TSO6;
4237                                         break;
4238                                 case SPEED_1000:
4239                                         netdev->features |= NETIF_F_TSO;
4240                                         netdev->features |= NETIF_F_TSO6;
4241                                         break;
4242                                 default:
4243                                         /* oops */
4244                                         break;
4245                                 }
4246                         }
4247
4248                         /*
4249                          * enable transmits in the hardware, need to do this
4250                          * after setting TARC(0)
4251                          */
4252                         tctl = er32(TCTL);
4253                         tctl |= E1000_TCTL_EN;
4254                         ew32(TCTL, tctl);
4255
4256                         /*
4257                          * Perform any post-link-up configuration before
4258                          * reporting link up.
4259                          */
4260                         if (phy->ops.cfg_on_link_up)
4261                                 phy->ops.cfg_on_link_up(hw);
4262
4263                         netif_carrier_on(netdev);
4264
4265                         if (!test_bit(__E1000_DOWN, &adapter->state))
4266                                 mod_timer(&adapter->phy_info_timer,
4267                                           round_jiffies(jiffies + 2 * HZ));
4268                 }
4269         } else {
4270                 if (netif_carrier_ok(netdev)) {
4271                         adapter->link_speed = 0;
4272                         adapter->link_duplex = 0;
4273                         /* Link status message must follow this format */
4274                         printk(KERN_INFO "e1000e: %s NIC Link is Down\n",
4275                                adapter->netdev->name);
4276                         netif_carrier_off(netdev);
4277                         if (!test_bit(__E1000_DOWN, &adapter->state))
4278                                 mod_timer(&adapter->phy_info_timer,
4279                                           round_jiffies(jiffies + 2 * HZ));
4280
4281                         if (adapter->flags & FLAG_RX_NEEDS_RESTART)
4282                                 schedule_work(&adapter->reset_task);
4283                         else
4284                                 pm_schedule_suspend(netdev->dev.parent,
4285                                                         LINK_TIMEOUT);
4286                 }
4287         }
4288
4289 link_up:
4290         spin_lock(&adapter->stats64_lock);
4291         e1000e_update_stats(adapter);
4292         spin_unlock(&adapter->stats64_lock);
4293
4294         mac->tx_packet_delta = adapter->stats.tpt - adapter->tpt_old;
4295         adapter->tpt_old = adapter->stats.tpt;
4296         mac->collision_delta = adapter->stats.colc - adapter->colc_old;
4297         adapter->colc_old = adapter->stats.colc;
4298
4299         adapter->gorc = adapter->stats.gorc - adapter->gorc_old;
4300         adapter->gorc_old = adapter->stats.gorc;
4301         adapter->gotc = adapter->stats.gotc - adapter->gotc_old;
4302         adapter->gotc_old = adapter->stats.gotc;
4303
4304         e1000e_update_adaptive(&adapter->hw);
4305
4306         if (!netif_carrier_ok(netdev) &&
4307             (e1000_desc_unused(tx_ring) + 1 < tx_ring->count)) {
4308                 /*
4309                  * We've lost link, so the controller stops DMA,
4310                  * but we've got queued Tx work that's never going
4311                  * to get done, so reset controller to flush Tx.
4312                  * (Do the reset outside of interrupt context).
4313                  */
4314                 adapter->tx_timeout_count++;
4315                 schedule_work(&adapter->reset_task);
4316                 /* return immediately since reset is imminent */
4317                 return;
4318         }
4319
4320         /* Simple mode for Interrupt Throttle Rate (ITR) */
4321         if (adapter->itr_setting == 4) {
4322                 /*
4323                  * Symmetric Tx/Rx gets a reduced ITR=2000;
4324                  * Total asymmetrical Tx or Rx gets ITR=8000;
4325                  * everyone else is between 2000-8000.
4326                  */
4327                 u32 goc = (adapter->gotc + adapter->gorc) / 10000;
4328                 u32 dif = (adapter->gotc > adapter->gorc ?
4329                             adapter->gotc - adapter->gorc :
4330                             adapter->gorc - adapter->gotc) / 10000;
4331                 u32 itr = goc > 0 ? (dif * 6000 / goc + 2000) : 8000;
4332
4333                 ew32(ITR, 1000000000 / (itr * 256));
4334         }
4335
4336         /* Cause software interrupt to ensure Rx ring is cleaned */
4337         if (adapter->msix_entries)
4338                 ew32(ICS, adapter->rx_ring->ims_val);
4339         else
4340                 ew32(ICS, E1000_ICS_RXDMT0);
4341
4342         /* Force detection of hung controller every watchdog period */
4343         adapter->detect_tx_hung = 1;
4344
4345         /* flush partial descriptors to memory before detecting Tx hang */
4346         if (adapter->flags2 & FLAG2_DMA_BURST) {
4347                 ew32(TIDV, adapter->tx_int_delay | E1000_TIDV_FPD);
4348                 ew32(RDTR, adapter->rx_int_delay | E1000_RDTR_FPD);
4349                 /*
4350                  * no need to flush the writes because the timeout code does
4351                  * an er32 first thing
4352                  */
4353         }
4354
4355         /*
4356          * With 82571 controllers, LAA may be overwritten due to controller
4357          * reset from the other port. Set the appropriate LAA in RAR[0]
4358          */
4359         if (e1000e_get_laa_state_82571(hw))
4360                 e1000e_rar_set(hw, adapter->hw.mac.addr, 0);
4361
4362         if (adapter->flags2 & FLAG2_CHECK_PHY_HANG)
4363                 e1000e_check_82574_phy_workaround(adapter);
4364
4365         /* Reset the timer */
4366         if (!test_bit(__E1000_DOWN, &adapter->state))
4367                 mod_timer(&adapter->watchdog_timer,
4368                           round_jiffies(jiffies + 2 * HZ));
4369 }
4370
4371 #define E1000_TX_FLAGS_CSUM             0x00000001
4372 #define E1000_TX_FLAGS_VLAN             0x00000002
4373 #define E1000_TX_FLAGS_TSO              0x00000004
4374 #define E1000_TX_FLAGS_IPV4             0x00000008
4375 #define E1000_TX_FLAGS_VLAN_MASK        0xffff0000
4376 #define E1000_TX_FLAGS_VLAN_SHIFT       16
4377
4378 static int e1000_tso(struct e1000_adapter *adapter,
4379                      struct sk_buff *skb)
4380 {
4381         struct e1000_ring *tx_ring = adapter->tx_ring;
4382         struct e1000_context_desc *context_desc;
4383         struct e1000_buffer *buffer_info;
4384         unsigned int i;
4385         u32 cmd_length = 0;
4386         u16 ipcse = 0, tucse, mss;
4387         u8 ipcss, ipcso, tucss, tucso, hdr_len;
4388
4389         if (!skb_is_gso(skb))
4390                 return 0;
4391
4392         if (skb_header_cloned(skb)) {
4393                 int err = pskb_expand_head(skb, 0, 0, GFP_ATOMIC);
4394
4395                 if (err)
4396                         return err;
4397         }
4398
4399         hdr_len = skb_transport_offset(skb) + tcp_hdrlen(skb);
4400         mss = skb_shinfo(skb)->gso_size;
4401         if (skb->protocol == htons(ETH_P_IP)) {
4402                 struct iphdr *iph = ip_hdr(skb);
4403                 iph->tot_len = 0;
4404                 iph->check = 0;
4405                 tcp_hdr(skb)->check = ~csum_tcpudp_magic(iph->saddr, iph->daddr,
4406                                                          0, IPPROTO_TCP, 0);
4407                 cmd_length = E1000_TXD_CMD_IP;
4408                 ipcse = skb_transport_offset(skb) - 1;
4409         } else if (skb_is_gso_v6(skb)) {
4410                 ipv6_hdr(skb)->payload_len = 0;
4411                 tcp_hdr(skb)->check = ~csum_ipv6_magic(&ipv6_hdr(skb)->saddr,
4412                                                        &ipv6_hdr(skb)->daddr,
4413                                                        0, IPPROTO_TCP, 0);
4414                 ipcse = 0;
4415         }
4416         ipcss = skb_network_offset(skb);
4417         ipcso = (void *)&(ip_hdr(skb)->check) - (void *)skb->data;
4418         tucss = skb_transport_offset(skb);
4419         tucso = (void *)&(tcp_hdr(skb)->check) - (void *)skb->data;
4420         tucse = 0;
4421
4422         cmd_length |= (E1000_TXD_CMD_DEXT | E1000_TXD_CMD_TSE |
4423                        E1000_TXD_CMD_TCP | (skb->len - (hdr_len)));
4424
4425         i = tx_ring->next_to_use;
4426         context_desc = E1000_CONTEXT_DESC(*tx_ring, i);
4427         buffer_info = &tx_ring->buffer_info[i];
4428
4429         context_desc->lower_setup.ip_fields.ipcss  = ipcss;
4430         context_desc->lower_setup.ip_fields.ipcso  = ipcso;
4431         context_desc->lower_setup.ip_fields.ipcse  = cpu_to_le16(ipcse);
4432         context_desc->upper_setup.tcp_fields.tucss = tucss;
4433         context_desc->upper_setup.tcp_fields.tucso = tucso;
4434         context_desc->upper_setup.tcp_fields.tucse = cpu_to_le16(tucse);
4435         context_desc->tcp_seg_setup.fields.mss     = cpu_to_le16(mss);
4436         context_desc->tcp_seg_setup.fields.hdr_len = hdr_len;
4437         context_desc->cmd_and_length = cpu_to_le32(cmd_length);
4438
4439         buffer_info->time_stamp = jiffies;
4440         buffer_info->next_to_watch = i;
4441
4442         i++;
4443         if (i == tx_ring->count)
4444                 i = 0;
4445         tx_ring->next_to_use = i;
4446
4447         return 1;
4448 }
4449
4450 static bool e1000_tx_csum(struct e1000_adapter *adapter, struct sk_buff *skb)
4451 {
4452         struct e1000_ring *tx_ring = adapter->tx_ring;
4453         struct e1000_context_desc *context_desc;
4454         struct e1000_buffer *buffer_info;
4455         unsigned int i;
4456         u8 css;
4457         u32 cmd_len = E1000_TXD_CMD_DEXT;
4458         __be16 protocol;
4459
4460         if (skb->ip_summed != CHECKSUM_PARTIAL)
4461                 return 0;
4462
4463         if (skb->protocol == cpu_to_be16(ETH_P_8021Q))
4464                 protocol = vlan_eth_hdr(skb)->h_vlan_encapsulated_proto;
4465         else
4466                 protocol = skb->protocol;
4467
4468         switch (protocol) {
4469         case cpu_to_be16(ETH_P_IP):
4470                 if (ip_hdr(skb)->protocol == IPPROTO_TCP)
4471                         cmd_len |= E1000_TXD_CMD_TCP;
4472                 break;
4473         case cpu_to_be16(ETH_P_IPV6):
4474                 /* XXX not handling all IPV6 headers */
4475                 if (ipv6_hdr(skb)->nexthdr == IPPROTO_TCP)
4476                         cmd_len |= E1000_TXD_CMD_TCP;
4477                 break;
4478         default:
4479                 if (unlikely(net_ratelimit()))
4480                         e_warn("checksum_partial proto=%x!\n",
4481                                be16_to_cpu(protocol));
4482                 break;
4483         }
4484
4485         css = skb_checksum_start_offset(skb);
4486
4487         i = tx_ring->next_to_use;
4488         buffer_info = &tx_ring->buffer_info[i];
4489         context_desc = E1000_CONTEXT_DESC(*tx_ring, i);
4490
4491         context_desc->lower_setup.ip_config = 0;
4492         context_desc->upper_setup.tcp_fields.tucss = css;
4493         context_desc->upper_setup.tcp_fields.tucso =
4494                                 css + skb->csum_offset;
4495         context_desc->upper_setup.tcp_fields.tucse = 0;
4496         context_desc->tcp_seg_setup.data = 0;
4497         context_desc->cmd_and_length = cpu_to_le32(cmd_len);
4498
4499         buffer_info->time_stamp = jiffies;
4500         buffer_info->next_to_watch = i;
4501
4502         i++;
4503         if (i == tx_ring->count)
4504                 i = 0;
4505         tx_ring->next_to_use = i;
4506
4507         return 1;
4508 }
4509
4510 #define E1000_MAX_PER_TXD       8192
4511 #define E1000_MAX_TXD_PWR       12
4512
4513 static int e1000_tx_map(struct e1000_adapter *adapter,
4514                         struct sk_buff *skb, unsigned int first,
4515                         unsigned int max_per_txd, unsigned int nr_frags,
4516                         unsigned int mss)
4517 {
4518         struct e1000_ring *tx_ring = adapter->tx_ring;
4519         struct pci_dev *pdev = adapter->pdev;
4520         struct e1000_buffer *buffer_info;
4521         unsigned int len = skb_headlen(skb);
4522         unsigned int offset = 0, size, count = 0, i;
4523         unsigned int f, bytecount, segs;
4524
4525         i = tx_ring->next_to_use;
4526
4527         while (len) {
4528                 buffer_info = &tx_ring->buffer_info[i];
4529                 size = min(len, max_per_txd);
4530
4531                 buffer_info->length = size;
4532                 buffer_info->time_stamp = jiffies;
4533                 buffer_info->next_to_watch = i;
4534                 buffer_info->dma = dma_map_single(&pdev->dev,
4535                                                   skb->data + offset,
4536                                                   size, DMA_TO_DEVICE);
4537                 buffer_info->mapped_as_page = false;
4538                 if (dma_mapping_error(&pdev->dev, buffer_info->dma))
4539                         goto dma_error;
4540
4541                 len -= size;
4542                 offset += size;
4543                 count++;
4544
4545                 if (len) {
4546                         i++;
4547                         if (i == tx_ring->count)
4548                                 i = 0;
4549                 }
4550         }
4551
4552         for (f = 0; f < nr_frags; f++) {
4553                 struct skb_frag_struct *frag;
4554
4555                 frag = &skb_shinfo(skb)->frags[f];
4556                 len = frag->size;
4557                 offset = frag->page_offset;
4558
4559                 while (len) {
4560                         i++;
4561                         if (i == tx_ring->count)
4562                                 i = 0;
4563
4564                         buffer_info = &tx_ring->buffer_info[i];
4565                         size = min(len, max_per_txd);
4566
4567                         buffer_info->length = size;
4568                         buffer_info->time_stamp = jiffies;
4569                         buffer_info->next_to_watch = i;
4570                         buffer_info->dma = dma_map_page(&pdev->dev, frag->page,
4571                                                         offset, size,
4572                                                         DMA_TO_DEVICE);
4573                         buffer_info->mapped_as_page = true;
4574                         if (dma_mapping_error(&pdev->dev, buffer_info->dma))
4575                                 goto dma_error;
4576
4577                         len -= size;
4578                         offset += size;
4579                         count++;
4580                 }
4581         }
4582
4583         segs = skb_shinfo(skb)->gso_segs ? : 1;
4584         /* multiply data chunks by size of headers */
4585         bytecount = ((segs - 1) * skb_headlen(skb)) + skb->len;
4586
4587         tx_ring->buffer_info[i].skb = skb;
4588         tx_ring->buffer_info[i].segs = segs;
4589         tx_ring->buffer_info[i].bytecount = bytecount;
4590         tx_ring->buffer_info[first].next_to_watch = i;
4591
4592         return count;
4593
4594 dma_error:
4595         dev_err(&pdev->dev, "Tx DMA map failed\n");
4596         buffer_info->dma = 0;
4597         if (count)
4598                 count--;
4599
4600         while (count--) {
4601                 if (i == 0)
4602                         i += tx_ring->count;
4603                 i--;
4604                 buffer_info = &tx_ring->buffer_info[i];
4605                 e1000_put_txbuf(adapter, buffer_info);
4606         }
4607
4608         return 0;
4609 }
4610
4611 static void e1000_tx_queue(struct e1000_adapter *adapter,
4612                            int tx_flags, int count)
4613 {
4614         struct e1000_ring *tx_ring = adapter->tx_ring;
4615         struct e1000_tx_desc *tx_desc = NULL;
4616         struct e1000_buffer *buffer_info;
4617         u32 txd_upper = 0, txd_lower = E1000_TXD_CMD_IFCS;
4618         unsigned int i;
4619
4620         if (tx_flags & E1000_TX_FLAGS_TSO) {
4621                 txd_lower |= E1000_TXD_CMD_DEXT | E1000_TXD_DTYP_D |
4622                              E1000_TXD_CMD_TSE;
4623                 txd_upper |= E1000_TXD_POPTS_TXSM << 8;
4624
4625                 if (tx_flags & E1000_TX_FLAGS_IPV4)
4626                         txd_upper |= E1000_TXD_POPTS_IXSM << 8;
4627         }
4628
4629         if (tx_flags & E1000_TX_FLAGS_CSUM) {
4630                 txd_lower |= E1000_TXD_CMD_DEXT | E1000_TXD_DTYP_D;
4631                 txd_upper |= E1000_TXD_POPTS_TXSM << 8;
4632         }
4633
4634         if (tx_flags & E1000_TX_FLAGS_VLAN) {
4635                 txd_lower |= E1000_TXD_CMD_VLE;
4636                 txd_upper |= (tx_flags & E1000_TX_FLAGS_VLAN_MASK);
4637         }
4638
4639         i = tx_ring->next_to_use;
4640
4641         do {
4642                 buffer_info = &tx_ring->buffer_info[i];
4643                 tx_desc = E1000_TX_DESC(*tx_ring, i);
4644                 tx_desc->buffer_addr = cpu_to_le64(buffer_info->dma);
4645                 tx_desc->lower.data =
4646                         cpu_to_le32(txd_lower | buffer_info->length);
4647                 tx_desc->upper.data = cpu_to_le32(txd_upper);
4648
4649                 i++;
4650                 if (i == tx_ring->count)
4651                         i = 0;
4652         } while (--count > 0);
4653
4654         tx_desc->lower.data |= cpu_to_le32(adapter->txd_cmd);
4655
4656         /*
4657          * Force memory writes to complete before letting h/w
4658          * know there are new descriptors to fetch.  (Only
4659          * applicable for weak-ordered memory model archs,
4660          * such as IA-64).
4661          */
4662         wmb();
4663
4664         tx_ring->next_to_use = i;
4665         writel(i, adapter->hw.hw_addr + tx_ring->tail);
4666         /*
4667          * we need this if more than one processor can write to our tail
4668          * at a time, it synchronizes IO on IA64/Altix systems
4669          */
4670         mmiowb();
4671 }
4672
4673 #define MINIMUM_DHCP_PACKET_SIZE 282
4674 static int e1000_transfer_dhcp_info(struct e1000_adapter *adapter,
4675                                     struct sk_buff *skb)
4676 {
4677         struct e1000_hw *hw =  &adapter->hw;
4678         u16 length, offset;
4679
4680         if (vlan_tx_tag_present(skb)) {
4681                 if (!((vlan_tx_tag_get(skb) == adapter->hw.mng_cookie.vlan_id) &&
4682                     (adapter->hw.mng_cookie.status &
4683                         E1000_MNG_DHCP_COOKIE_STATUS_VLAN)))
4684                         return 0;
4685         }
4686
4687         if (skb->len <= MINIMUM_DHCP_PACKET_SIZE)
4688                 return 0;
4689
4690         if (((struct ethhdr *) skb->data)->h_proto != htons(ETH_P_IP))
4691                 return 0;
4692
4693         {
4694                 const struct iphdr *ip = (struct iphdr *)((u8 *)skb->data+14);
4695                 struct udphdr *udp;
4696
4697                 if (ip->protocol != IPPROTO_UDP)
4698                         return 0;
4699
4700                 udp = (struct udphdr *)((u8 *)ip + (ip->ihl << 2));
4701                 if (ntohs(udp->dest) != 67)
4702                         return 0;
4703
4704                 offset = (u8 *)udp + 8 - skb->data;
4705                 length = skb->len - offset;
4706                 return e1000e_mng_write_dhcp_info(hw, (u8 *)udp + 8, length);
4707         }
4708
4709         return 0;
4710 }
4711
4712 static int __e1000_maybe_stop_tx(struct net_device *netdev, int size)
4713 {
4714         struct e1000_adapter *adapter = netdev_priv(netdev);
4715
4716         netif_stop_queue(netdev);
4717         /*
4718          * Herbert's original patch had:
4719          *  smp_mb__after_netif_stop_queue();
4720          * but since that doesn't exist yet, just open code it.
4721          */
4722         smp_mb();
4723
4724         /*
4725          * We need to check again in a case another CPU has just
4726          * made room available.
4727          */
4728         if (e1000_desc_unused(adapter->tx_ring) < size)
4729                 return -EBUSY;
4730
4731         /* A reprieve! */
4732         netif_start_queue(netdev);
4733         ++adapter->restart_queue;
4734         return 0;
4735 }
4736
4737 static int e1000_maybe_stop_tx(struct net_device *netdev, int size)
4738 {
4739         struct e1000_adapter *adapter = netdev_priv(netdev);
4740
4741         if (e1000_desc_unused(adapter->tx_ring) >= size)
4742                 return 0;
4743         return __e1000_maybe_stop_tx(netdev, size);
4744 }
4745
4746 #define TXD_USE_COUNT(S, X) (((S) >> (X)) + 1 )
4747 static netdev_tx_t e1000_xmit_frame(struct sk_buff *skb,
4748                                     struct net_device *netdev)
4749 {
4750         struct e1000_adapter *adapter = netdev_priv(netdev);
4751         struct e1000_ring *tx_ring = adapter->tx_ring;
4752         unsigned int first;
4753         unsigned int max_per_txd = E1000_MAX_PER_TXD;
4754         unsigned int max_txd_pwr = E1000_MAX_TXD_PWR;
4755         unsigned int tx_flags = 0;
4756         unsigned int len = skb_headlen(skb);
4757         unsigned int nr_frags;
4758         unsigned int mss;
4759         int count = 0;
4760         int tso;
4761         unsigned int f;
4762
4763         if (test_bit(__E1000_DOWN, &adapter->state)) {
4764                 dev_kfree_skb_any(skb);
4765                 return NETDEV_TX_OK;
4766         }
4767
4768         if (skb->len <= 0) {
4769                 dev_kfree_skb_any(skb);
4770                 return NETDEV_TX_OK;
4771         }
4772
4773         mss = skb_shinfo(skb)->gso_size;
4774         /*
4775          * The controller does a simple calculation to
4776          * make sure there is enough room in the FIFO before
4777          * initiating the DMA for each buffer.  The calc is:
4778          * 4 = ceil(buffer len/mss).  To make sure we don't
4779          * overrun the FIFO, adjust the max buffer len if mss
4780          * drops.
4781          */
4782         if (mss) {
4783                 u8 hdr_len;
4784                 max_per_txd = min(mss << 2, max_per_txd);
4785                 max_txd_pwr = fls(max_per_txd) - 1;
4786
4787                 /*
4788                  * TSO Workaround for 82571/2/3 Controllers -- if skb->data
4789                  * points to just header, pull a few bytes of payload from
4790                  * frags into skb->data
4791                  */
4792                 hdr_len = skb_transport_offset(skb) + tcp_hdrlen(skb);
4793                 /*
4794                  * we do this workaround for ES2LAN, but it is un-necessary,
4795                  * avoiding it could save a lot of cycles
4796                  */
4797                 if (skb->data_len && (hdr_len == len)) {
4798                         unsigned int pull_size;
4799
4800                         pull_size = min((unsigned int)4, skb->data_len);
4801                         if (!__pskb_pull_tail(skb, pull_size)) {
4802                                 e_err("__pskb_pull_tail failed.\n");
4803                                 dev_kfree_skb_any(skb);
4804                                 return NETDEV_TX_OK;
4805                         }
4806                         len = skb_headlen(skb);
4807                 }
4808         }
4809
4810         /* reserve a descriptor for the offload context */
4811         if ((mss) || (skb->ip_summed == CHECKSUM_PARTIAL))
4812                 count++;
4813         count++;
4814
4815         count += TXD_USE_COUNT(len, max_txd_pwr);
4816
4817         nr_frags = skb_shinfo(skb)->nr_frags;
4818         for (f = 0; f < nr_frags; f++)
4819                 count += TXD_USE_COUNT(skb_shinfo(skb)->frags[f].size,
4820                                        max_txd_pwr);
4821
4822         if (adapter->hw.mac.tx_pkt_filtering)
4823                 e1000_transfer_dhcp_info(adapter, skb);
4824
4825         /*
4826          * need: count + 2 desc gap to keep tail from touching
4827          * head, otherwise try next time
4828          */
4829         if (e1000_maybe_stop_tx(netdev, count + 2))
4830                 return NETDEV_TX_BUSY;
4831
4832         if (vlan_tx_tag_present(skb)) {
4833                 tx_flags |= E1000_TX_FLAGS_VLAN;
4834                 tx_flags |= (vlan_tx_tag_get(skb) << E1000_TX_FLAGS_VLAN_SHIFT);
4835         }
4836
4837         first = tx_ring->next_to_use;
4838
4839         tso = e1000_tso(adapter, skb);
4840         if (tso < 0) {
4841                 dev_kfree_skb_any(skb);
4842                 return NETDEV_TX_OK;
4843         }
4844
4845         if (tso)
4846                 tx_flags |= E1000_TX_FLAGS_TSO;
4847         else if (e1000_tx_csum(adapter, skb))
4848                 tx_flags |= E1000_TX_FLAGS_CSUM;
4849
4850         /*
4851          * Old method was to assume IPv4 packet by default if TSO was enabled.
4852          * 82571 hardware supports TSO capabilities for IPv6 as well...
4853          * no longer assume, we must.
4854          */
4855         if (skb->protocol == htons(ETH_P_IP))
4856                 tx_flags |= E1000_TX_FLAGS_IPV4;
4857
4858         /* if count is 0 then mapping error has occured */
4859         count = e1000_tx_map(adapter, skb, first, max_per_txd, nr_frags, mss);
4860         if (count) {
4861                 e1000_tx_queue(adapter, tx_flags, count);
4862                 /* Make sure there is space in the ring for the next send. */
4863                 e1000_maybe_stop_tx(netdev, MAX_SKB_FRAGS + 2);
4864
4865         } else {
4866                 dev_kfree_skb_any(skb);
4867                 tx_ring->buffer_info[first].time_stamp = 0;
4868                 tx_ring->next_to_use = first;
4869         }
4870
4871         return NETDEV_TX_OK;
4872 }
4873
4874 /**
4875  * e1000_tx_timeout - Respond to a Tx Hang
4876  * @netdev: network interface device structure
4877  **/
4878 static void e1000_tx_timeout(struct net_device *netdev)
4879 {
4880         struct e1000_adapter *adapter = netdev_priv(netdev);
4881
4882         /* Do the reset outside of interrupt context */
4883         adapter->tx_timeout_count++;
4884         schedule_work(&adapter->reset_task);
4885 }
4886
4887 static void e1000_reset_task(struct work_struct *work)
4888 {
4889         struct e1000_adapter *adapter;
4890         adapter = container_of(work, struct e1000_adapter, reset_task);
4891
4892         if (!((adapter->flags & FLAG_RX_NEEDS_RESTART) &&
4893               (adapter->flags & FLAG_RX_RESTART_NOW))) {
4894                 e1000e_dump(adapter);
4895                 e_err("Reset adapter\n");
4896         }
4897         e1000e_reinit_locked(adapter);
4898 }
4899
4900 /**
4901  * e1000_get_stats64 - Get System Network Statistics
4902  * @netdev: network interface device structure
4903  * @stats: rtnl_link_stats64 pointer
4904  *
4905  * Returns the address of the device statistics structure.
4906  **/
4907 struct rtnl_link_stats64 *e1000e_get_stats64(struct net_device *netdev,
4908                                              struct rtnl_link_stats64 *stats)
4909 {
4910         struct e1000_adapter *adapter = netdev_priv(netdev);
4911
4912         memset(stats, 0, sizeof(struct rtnl_link_stats64));
4913         spin_lock(&adapter->stats64_lock);
4914         e1000e_update_stats(adapter);
4915         /* Fill out the OS statistics structure */
4916         stats->rx_bytes = adapter->stats.gorc;
4917         stats->rx_packets = adapter->stats.gprc;
4918         stats->tx_bytes = adapter->stats.gotc;
4919         stats->tx_packets = adapter->stats.gptc;
4920         stats->multicast = adapter->stats.mprc;
4921         stats->collisions = adapter->stats.colc;
4922
4923         /* Rx Errors */
4924
4925         /*
4926          * RLEC on some newer hardware can be incorrect so build
4927          * our own version based on RUC and ROC
4928          */
4929         stats->rx_errors = adapter->stats.rxerrc +
4930                 adapter->stats.crcerrs + adapter->stats.algnerrc +
4931                 adapter->stats.ruc + adapter->stats.roc +
4932                 adapter->stats.cexterr;
4933         stats->rx_length_errors = adapter->stats.ruc +
4934                                               adapter->stats.roc;
4935         stats->rx_crc_errors = adapter->stats.crcerrs;
4936         stats->rx_frame_errors = adapter->stats.algnerrc;
4937         stats->rx_missed_errors = adapter->stats.mpc;
4938
4939         /* Tx Errors */
4940         stats->tx_errors = adapter->stats.ecol +
4941                                        adapter->stats.latecol;
4942         stats->tx_aborted_errors = adapter->stats.ecol;
4943         stats->tx_window_errors = adapter->stats.latecol;
4944         stats->tx_carrier_errors = adapter->stats.tncrs;
4945
4946         /* Tx Dropped needs to be maintained elsewhere */
4947
4948         spin_unlock(&adapter->stats64_lock);
4949         return stats;
4950 }
4951
4952 /**
4953  * e1000_change_mtu - Change the Maximum Transfer Unit
4954  * @netdev: network interface device structure
4955  * @new_mtu: new value for maximum frame size
4956  *
4957  * Returns 0 on success, negative on failure
4958  **/
4959 static int e1000_change_mtu(struct net_device *netdev, int new_mtu)
4960 {
4961         struct e1000_adapter *adapter = netdev_priv(netdev);
4962         int max_frame = new_mtu + ETH_HLEN + ETH_FCS_LEN;
4963
4964         /* Jumbo frame support */
4965         if ((max_frame > ETH_FRAME_LEN + ETH_FCS_LEN) &&
4966             !(adapter->flags & FLAG_HAS_JUMBO_FRAMES)) {
4967                 e_err("Jumbo Frames not supported.\n");
4968                 return -EINVAL;
4969         }
4970
4971         /* Supported frame sizes */
4972         if ((new_mtu < ETH_ZLEN + ETH_FCS_LEN + VLAN_HLEN) ||
4973             (max_frame > adapter->max_hw_frame_size)) {
4974                 e_err("Unsupported MTU setting\n");
4975                 return -EINVAL;
4976         }
4977
4978         /* Jumbo frame workaround on 82579 requires CRC be stripped */
4979         if ((adapter->hw.mac.type == e1000_pch2lan) &&
4980             !(adapter->flags2 & FLAG2_CRC_STRIPPING) &&
4981             (new_mtu > ETH_DATA_LEN)) {
4982                 e_err("Jumbo Frames not supported on 82579 when CRC "
4983                       "stripping is disabled.\n");
4984                 return -EINVAL;
4985         }
4986
4987         /* 82573 Errata 17 */
4988         if (((adapter->hw.mac.type == e1000_82573) ||
4989              (adapter->hw.mac.type == e1000_82574)) &&
4990             (max_frame > ETH_FRAME_LEN + ETH_FCS_LEN)) {
4991                 adapter->flags2 |= FLAG2_DISABLE_ASPM_L1;
4992                 e1000e_disable_aspm(adapter->pdev, PCIE_LINK_STATE_L1);
4993         }
4994
4995         while (test_and_set_bit(__E1000_RESETTING, &adapter->state))
4996                 msleep(1);
4997         /* e1000e_down -> e1000e_reset dependent on max_frame_size & mtu */
4998         adapter->max_frame_size = max_frame;
4999         e_info("changing MTU from %d to %d\n", netdev->mtu, new_mtu);
5000         netdev->mtu = new_mtu;
5001         if (netif_running(netdev))
5002                 e1000e_down(adapter);
5003
5004         /*
5005          * NOTE: netdev_alloc_skb reserves 16 bytes, and typically NET_IP_ALIGN
5006          * means we reserve 2 more, this pushes us to allocate from the next
5007          * larger slab size.
5008          * i.e. RXBUFFER_2048 --> size-4096 slab
5009          * However with the new *_jumbo_rx* routines, jumbo receives will use
5010          * fragmented skbs
5011          */
5012
5013         if (max_frame <= 2048)
5014                 adapter->rx_buffer_len = 2048;
5015         else
5016                 adapter->rx_buffer_len = 4096;
5017
5018         /* adjust allocation if LPE protects us, and we aren't using SBP */
5019         if ((max_frame == ETH_FRAME_LEN + ETH_FCS_LEN) ||
5020              (max_frame == ETH_FRAME_LEN + VLAN_HLEN + ETH_FCS_LEN))
5021                 adapter->rx_buffer_len = ETH_FRAME_LEN + VLAN_HLEN
5022                                          + ETH_FCS_LEN;
5023
5024         if (netif_running(netdev))
5025                 e1000e_up(adapter);
5026         else
5027                 e1000e_reset(adapter);
5028
5029         clear_bit(__E1000_RESETTING, &adapter->state);
5030
5031         return 0;
5032 }
5033
5034 static int e1000_mii_ioctl(struct net_device *netdev, struct ifreq *ifr,
5035                            int cmd)
5036 {
5037         struct e1000_adapter *adapter = netdev_priv(netdev);
5038         struct mii_ioctl_data *data = if_mii(ifr);
5039
5040         if (adapter->hw.phy.media_type != e1000_media_type_copper)
5041                 return -EOPNOTSUPP;
5042
5043         switch (cmd) {
5044         case SIOCGMIIPHY:
5045                 data->phy_id = adapter->hw.phy.addr;
5046                 break;
5047         case SIOCGMIIREG:
5048                 e1000_phy_read_status(adapter);
5049
5050                 switch (data->reg_num & 0x1F) {
5051                 case MII_BMCR:
5052                         data->val_out = adapter->phy_regs.bmcr;
5053                         break;
5054                 case MII_BMSR:
5055                         data->val_out = adapter->phy_regs.bmsr;
5056                         break;
5057                 case MII_PHYSID1:
5058                         data->val_out = (adapter->hw.phy.id >> 16);
5059                         break;
5060                 case MII_PHYSID2:
5061                         data->val_out = (adapter->hw.phy.id & 0xFFFF);
5062                         break;
5063                 case MII_ADVERTISE:
5064                         data->val_out = adapter->phy_regs.advertise;
5065                         break;
5066                 case MII_LPA:
5067                         data->val_out = adapter->phy_regs.lpa;
5068                         break;
5069                 case MII_EXPANSION:
5070                         data->val_out = adapter->phy_regs.expansion;
5071                         break;
5072                 case MII_CTRL1000:
5073                         data->val_out = adapter->phy_regs.ctrl1000;
5074                         break;
5075                 case MII_STAT1000:
5076                         data->val_out = adapter->phy_regs.stat1000;
5077                         break;
5078                 case MII_ESTATUS:
5079                         data->val_out = adapter->phy_regs.estatus;
5080                         break;
5081                 default:
5082                         return -EIO;
5083                 }
5084                 break;
5085         case SIOCSMIIREG:
5086         default:
5087                 return -EOPNOTSUPP;
5088         }
5089         return 0;
5090 }
5091
5092 static int e1000_ioctl(struct net_device *netdev, struct ifreq *ifr, int cmd)
5093 {
5094         switch (cmd) {
5095         case SIOCGMIIPHY:
5096         case SIOCGMIIREG:
5097         case SIOCSMIIREG:
5098                 return e1000_mii_ioctl(netdev, ifr, cmd);
5099         default:
5100                 return -EOPNOTSUPP;
5101         }
5102 }
5103
5104 static int e1000_init_phy_wakeup(struct e1000_adapter *adapter, u32 wufc)
5105 {
5106         struct e1000_hw *hw = &adapter->hw;
5107         u32 i, mac_reg;
5108         u16 phy_reg;
5109         int retval = 0;
5110
5111         /* copy MAC RARs to PHY RARs */
5112         e1000_copy_rx_addrs_to_phy_ich8lan(hw);
5113
5114         /* copy MAC MTA to PHY MTA */
5115         for (i = 0; i < adapter->hw.mac.mta_reg_count; i++) {
5116                 mac_reg = E1000_READ_REG_ARRAY(hw, E1000_MTA, i);
5117                 e1e_wphy(hw, BM_MTA(i), (u16)(mac_reg & 0xFFFF));
5118                 e1e_wphy(hw, BM_MTA(i) + 1, (u16)((mac_reg >> 16) & 0xFFFF));
5119         }
5120
5121         /* configure PHY Rx Control register */
5122         e1e_rphy(&adapter->hw, BM_RCTL, &phy_reg);
5123         mac_reg = er32(RCTL);
5124         if (mac_reg & E1000_RCTL_UPE)
5125                 phy_reg |= BM_RCTL_UPE;
5126         if (mac_reg & E1000_RCTL_MPE)
5127                 phy_reg |= BM_RCTL_MPE;
5128         phy_reg &= ~(BM_RCTL_MO_MASK);
5129         if (mac_reg & E1000_RCTL_MO_3)
5130                 phy_reg |= (((mac_reg & E1000_RCTL_MO_3) >> E1000_RCTL_MO_SHIFT)
5131                                 << BM_RCTL_MO_SHIFT);
5132         if (mac_reg & E1000_RCTL_BAM)
5133                 phy_reg |= BM_RCTL_BAM;
5134         if (mac_reg & E1000_RCTL_PMCF)
5135                 phy_reg |= BM_RCTL_PMCF;
5136         mac_reg = er32(CTRL);
5137         if (mac_reg & E1000_CTRL_RFCE)
5138                 phy_reg |= BM_RCTL_RFCE;
5139         e1e_wphy(&adapter->hw, BM_RCTL, phy_reg);
5140
5141         /* enable PHY wakeup in MAC register */
5142         ew32(WUFC, wufc);
5143         ew32(WUC, E1000_WUC_PHY_WAKE | E1000_WUC_PME_EN);
5144
5145         /* configure and enable PHY wakeup in PHY registers */
5146         e1e_wphy(&adapter->hw, BM_WUFC, wufc);
5147         e1e_wphy(&adapter->hw, BM_WUC, E1000_WUC_PME_EN);
5148
5149         /* activate PHY wakeup */
5150         retval = hw->phy.ops.acquire(hw);
5151         if (retval) {
5152                 e_err("Could not acquire PHY\n");
5153                 return retval;
5154         }
5155         e1000e_write_phy_reg_mdic(hw, IGP01E1000_PHY_PAGE_SELECT,
5156                                  (BM_WUC_ENABLE_PAGE << IGP_PAGE_SHIFT));
5157         retval = e1000e_read_phy_reg_mdic(hw, BM_WUC_ENABLE_REG, &phy_reg);
5158         if (retval) {
5159                 e_err("Could not read PHY page 769\n");
5160                 goto out;
5161         }
5162         phy_reg |= BM_WUC_ENABLE_BIT | BM_WUC_HOST_WU_BIT;
5163         retval = e1000e_write_phy_reg_mdic(hw, BM_WUC_ENABLE_REG, phy_reg);
5164         if (retval)
5165                 e_err("Could not set PHY Host Wakeup bit\n");
5166 out:
5167         hw->phy.ops.release(hw);
5168
5169         return retval;
5170 }
5171
5172 static int __e1000_shutdown(struct pci_dev *pdev, bool *enable_wake,
5173                             bool runtime)
5174 {
5175         struct net_device *netdev = pci_get_drvdata(pdev);
5176         struct e1000_adapter *adapter = netdev_priv(netdev);
5177         struct e1000_hw *hw = &adapter->hw;
5178         u32 ctrl, ctrl_ext, rctl, status;
5179         /* Runtime suspend should only enable wakeup for link changes */
5180         u32 wufc = runtime ? E1000_WUFC_LNKC : adapter->wol;
5181         int retval = 0;
5182
5183         netif_device_detach(netdev);
5184
5185         if (netif_running(netdev)) {
5186                 WARN_ON(test_bit(__E1000_RESETTING, &adapter->state));
5187                 e1000e_down(adapter);
5188                 e1000_free_irq(adapter);
5189         }
5190         e1000e_reset_interrupt_capability(adapter);
5191
5192         retval = pci_save_state(pdev);
5193         if (retval)
5194                 return retval;
5195
5196         status = er32(STATUS);
5197         if (status & E1000_STATUS_LU)
5198                 wufc &= ~E1000_WUFC_LNKC;
5199
5200         if (wufc) {
5201                 e1000_setup_rctl(adapter);
5202                 e1000_set_multi(netdev);
5203
5204                 /* turn on all-multi mode if wake on multicast is enabled */
5205                 if (wufc & E1000_WUFC_MC) {
5206                         rctl = er32(RCTL);
5207                         rctl |= E1000_RCTL_MPE;
5208                         ew32(RCTL, rctl);
5209                 }
5210
5211                 ctrl = er32(CTRL);
5212                 /* advertise wake from D3Cold */
5213                 #define E1000_CTRL_ADVD3WUC 0x00100000
5214                 /* phy power management enable */
5215                 #define E1000_CTRL_EN_PHY_PWR_MGMT 0x00200000
5216                 ctrl |= E1000_CTRL_ADVD3WUC;
5217                 if (!(adapter->flags2 & FLAG2_HAS_PHY_WAKEUP))
5218                         ctrl |= E1000_CTRL_EN_PHY_PWR_MGMT;
5219                 ew32(CTRL, ctrl);
5220
5221                 if (adapter->hw.phy.media_type == e1000_media_type_fiber ||
5222                     adapter->hw.phy.media_type ==
5223                     e1000_media_type_internal_serdes) {
5224                         /* keep the laser running in D3 */
5225                         ctrl_ext = er32(CTRL_EXT);
5226                         ctrl_ext |= E1000_CTRL_EXT_SDP3_DATA;
5227                         ew32(CTRL_EXT, ctrl_ext);
5228                 }
5229
5230                 if (adapter->flags & FLAG_IS_ICH)
5231                         e1000e_disable_gig_wol_ich8lan(&adapter->hw);
5232
5233                 /* Allow time for pending master requests to run */
5234                 e1000e_disable_pcie_master(&adapter->hw);
5235
5236                 if (adapter->flags2 & FLAG2_HAS_PHY_WAKEUP) {
5237                         /* enable wakeup by the PHY */
5238                         retval = e1000_init_phy_wakeup(adapter, wufc);
5239                         if (retval)
5240                                 return retval;
5241                 } else {
5242                         /* enable wakeup by the MAC */
5243                         ew32(WUFC, wufc);
5244                         ew32(WUC, E1000_WUC_PME_EN);
5245                 }
5246         } else {
5247                 ew32(WUC, 0);
5248                 ew32(WUFC, 0);
5249         }
5250
5251         *enable_wake = !!wufc;
5252
5253         /* make sure adapter isn't asleep if manageability is enabled */
5254         if ((adapter->flags & FLAG_MNG_PT_ENABLED) ||
5255             (hw->mac.ops.check_mng_mode(hw)))
5256                 *enable_wake = true;
5257
5258         if (adapter->hw.phy.type == e1000_phy_igp_3)
5259                 e1000e_igp3_phy_powerdown_workaround_ich8lan(&adapter->hw);
5260
5261         /*
5262          * Release control of h/w to f/w.  If f/w is AMT enabled, this
5263          * would have already happened in close and is redundant.
5264          */
5265         e1000e_release_hw_control(adapter);
5266
5267         pci_disable_device(pdev);
5268
5269         return 0;
5270 }
5271
5272 static void e1000_power_off(struct pci_dev *pdev, bool sleep, bool wake)
5273 {
5274         if (sleep && wake) {
5275                 pci_prepare_to_sleep(pdev);
5276                 return;
5277         }
5278
5279         pci_wake_from_d3(pdev, wake);
5280         pci_set_power_state(pdev, PCI_D3hot);
5281 }
5282
5283 static void e1000_complete_shutdown(struct pci_dev *pdev, bool sleep,
5284                                     bool wake)
5285 {
5286         struct net_device *netdev = pci_get_drvdata(pdev);
5287         struct e1000_adapter *adapter = netdev_priv(netdev);
5288
5289         /*
5290          * The pci-e switch on some quad port adapters will report a
5291          * correctable error when the MAC transitions from D0 to D3.  To
5292          * prevent this we need to mask off the correctable errors on the
5293          * downstream port of the pci-e switch.
5294          */
5295         if (adapter->flags & FLAG_IS_QUAD_PORT) {
5296                 struct pci_dev *us_dev = pdev->bus->self;
5297                 int pos = pci_find_capability(us_dev, PCI_CAP_ID_EXP);
5298                 u16 devctl;
5299
5300                 pci_read_config_word(us_dev, pos + PCI_EXP_DEVCTL, &devctl);
5301                 pci_write_config_word(us_dev, pos + PCI_EXP_DEVCTL,
5302                                       (devctl & ~PCI_EXP_DEVCTL_CERE));
5303
5304                 e1000_power_off(pdev, sleep, wake);
5305
5306                 pci_write_config_word(us_dev, pos + PCI_EXP_DEVCTL, devctl);
5307         } else {
5308                 e1000_power_off(pdev, sleep, wake);
5309         }
5310 }
5311
5312 #ifdef CONFIG_PCIEASPM
5313 static void __e1000e_disable_aspm(struct pci_dev *pdev, u16 state)
5314 {
5315         pci_disable_link_state(pdev, state);
5316 }
5317 #else
5318 static void __e1000e_disable_aspm(struct pci_dev *pdev, u16 state)
5319 {
5320         int pos;
5321         u16 reg16;
5322
5323         /*
5324          * Both device and parent should have the same ASPM setting.
5325          * Disable ASPM in downstream component first and then upstream.
5326          */
5327         pos = pci_pcie_cap(pdev);
5328         pci_read_config_word(pdev, pos + PCI_EXP_LNKCTL, &reg16);
5329         reg16 &= ~state;
5330         pci_write_config_word(pdev, pos + PCI_EXP_LNKCTL, reg16);
5331
5332         if (!pdev->bus->self)
5333                 return;
5334
5335         pos = pci_pcie_cap(pdev->bus->self);
5336         pci_read_config_word(pdev->bus->self, pos + PCI_EXP_LNKCTL, &reg16);
5337         reg16 &= ~state;
5338         pci_write_config_word(pdev->bus->self, pos + PCI_EXP_LNKCTL, reg16);
5339 }
5340 #endif
5341 void e1000e_disable_aspm(struct pci_dev *pdev, u16 state)
5342 {
5343         dev_info(&pdev->dev, "Disabling ASPM %s %s\n",
5344                  (state & PCIE_LINK_STATE_L0S) ? "L0s" : "",
5345                  (state & PCIE_LINK_STATE_L1) ? "L1" : "");
5346
5347         __e1000e_disable_aspm(pdev, state);
5348 }
5349
5350 #ifdef CONFIG_PM_OPS
5351 static bool e1000e_pm_ready(struct e1000_adapter *adapter)
5352 {
5353         return !!adapter->tx_ring->buffer_info;
5354 }
5355
5356 static int __e1000_resume(struct pci_dev *pdev)
5357 {
5358         struct net_device *netdev = pci_get_drvdata(pdev);
5359         struct e1000_adapter *adapter = netdev_priv(netdev);
5360         struct e1000_hw *hw = &adapter->hw;
5361         u32 err;
5362
5363         pci_set_power_state(pdev, PCI_D0);
5364         pci_restore_state(pdev);
5365         pci_save_state(pdev);
5366         if (adapter->flags2 & FLAG2_DISABLE_ASPM_L1)
5367                 e1000e_disable_aspm(pdev, PCIE_LINK_STATE_L1);
5368
5369         e1000e_set_interrupt_capability(adapter);
5370         if (netif_running(netdev)) {
5371                 err = e1000_request_irq(adapter);
5372                 if (err)
5373                         return err;
5374         }
5375
5376         e1000e_power_up_phy(adapter);
5377
5378         /* report the system wakeup cause from S3/S4 */
5379         if (adapter->flags2 & FLAG2_HAS_PHY_WAKEUP) {
5380                 u16 phy_data;
5381
5382                 e1e_rphy(&adapter->hw, BM_WUS, &phy_data);
5383                 if (phy_data) {
5384                         e_info("PHY Wakeup cause - %s\n",
5385                                 phy_data & E1000_WUS_EX ? "Unicast Packet" :
5386                                 phy_data & E1000_WUS_MC ? "Multicast Packet" :
5387                                 phy_data & E1000_WUS_BC ? "Broadcast Packet" :
5388                                 phy_data & E1000_WUS_MAG ? "Magic Packet" :
5389                                 phy_data & E1000_WUS_LNKC ? "Link Status "
5390                                 " Change" : "other");
5391                 }
5392                 e1e_wphy(&adapter->hw, BM_WUS, ~0);
5393         } else {
5394                 u32 wus = er32(WUS);
5395                 if (wus) {
5396                         e_info("MAC Wakeup cause - %s\n",
5397                                 wus & E1000_WUS_EX ? "Unicast Packet" :
5398                                 wus & E1000_WUS_MC ? "Multicast Packet" :
5399                                 wus & E1000_WUS_BC ? "Broadcast Packet" :
5400                                 wus & E1000_WUS_MAG ? "Magic Packet" :
5401                                 wus & E1000_WUS_LNKC ? "Link Status Change" :
5402                                 "other");
5403                 }
5404                 ew32(WUS, ~0);
5405         }
5406
5407         e1000e_reset(adapter);
5408
5409         e1000_init_manageability_pt(adapter);
5410
5411         if (netif_running(netdev))
5412                 e1000e_up(adapter);
5413
5414         netif_device_attach(netdev);
5415
5416         /*
5417          * If the controller has AMT, do not set DRV_LOAD until the interface
5418          * is up.  For all other cases, let the f/w know that the h/w is now
5419          * under the control of the driver.
5420          */
5421         if (!(adapter->flags & FLAG_HAS_AMT))
5422                 e1000e_get_hw_control(adapter);
5423
5424         return 0;
5425 }
5426
5427 #ifdef CONFIG_PM_SLEEP
5428 static int e1000_suspend(struct device *dev)
5429 {
5430         struct pci_dev *pdev = to_pci_dev(dev);
5431         int retval;
5432         bool wake;
5433
5434         retval = __e1000_shutdown(pdev, &wake, false);
5435         if (!retval)
5436                 e1000_complete_shutdown(pdev, true, wake);
5437
5438         return retval;
5439 }
5440
5441 static int e1000_resume(struct device *dev)
5442 {
5443         struct pci_dev *pdev = to_pci_dev(dev);
5444         struct net_device *netdev = pci_get_drvdata(pdev);
5445         struct e1000_adapter *adapter = netdev_priv(netdev);
5446
5447         if (e1000e_pm_ready(adapter))
5448                 adapter->idle_check = true;
5449
5450         return __e1000_resume(pdev);
5451 }
5452 #endif /* CONFIG_PM_SLEEP */
5453
5454 #ifdef CONFIG_PM_RUNTIME
5455 static int e1000_runtime_suspend(struct device *dev)
5456 {
5457         struct pci_dev *pdev = to_pci_dev(dev);
5458         struct net_device *netdev = pci_get_drvdata(pdev);
5459         struct e1000_adapter *adapter = netdev_priv(netdev);
5460
5461         if (e1000e_pm_ready(adapter)) {
5462                 bool wake;
5463
5464                 __e1000_shutdown(pdev, &wake, true);
5465         }
5466
5467         return 0;
5468 }
5469
5470 static int e1000_idle(struct device *dev)
5471 {
5472         struct pci_dev *pdev = to_pci_dev(dev);
5473         struct net_device *netdev = pci_get_drvdata(pdev);
5474         struct e1000_adapter *adapter = netdev_priv(netdev);
5475
5476         if (!e1000e_pm_ready(adapter))
5477                 return 0;
5478
5479         if (adapter->idle_check) {
5480                 adapter->idle_check = false;
5481                 if (!e1000e_has_link(adapter))
5482                         pm_schedule_suspend(dev, MSEC_PER_SEC);
5483         }
5484
5485         return -EBUSY;
5486 }
5487
5488 static int e1000_runtime_resume(struct device *dev)
5489 {
5490         struct pci_dev *pdev = to_pci_dev(dev);
5491         struct net_device *netdev = pci_get_drvdata(pdev);
5492         struct e1000_adapter *adapter = netdev_priv(netdev);
5493
5494         if (!e1000e_pm_ready(adapter))
5495                 return 0;
5496
5497         adapter->idle_check = !dev->power.runtime_auto;
5498         return __e1000_resume(pdev);
5499 }
5500 #endif /* CONFIG_PM_RUNTIME */
5501 #endif /* CONFIG_PM_OPS */
5502
5503 static void e1000_shutdown(struct pci_dev *pdev)
5504 {
5505         bool wake = false;
5506
5507         __e1000_shutdown(pdev, &wake, false);
5508
5509         if (system_state == SYSTEM_POWER_OFF)
5510                 e1000_complete_shutdown(pdev, false, wake);
5511 }
5512
5513 #ifdef CONFIG_NET_POLL_CONTROLLER
5514
5515 static irqreturn_t e1000_intr_msix(int irq, void *data)
5516 {
5517         struct net_device *netdev = data;
5518         struct e1000_adapter *adapter = netdev_priv(netdev);
5519
5520         if (adapter->msix_entries) {
5521                 int vector, msix_irq;
5522
5523                 vector = 0;
5524                 msix_irq = adapter->msix_entries[vector].vector;
5525                 disable_irq(msix_irq);
5526                 e1000_intr_msix_rx(msix_irq, netdev);
5527                 enable_irq(msix_irq);
5528
5529                 vector++;
5530                 msix_irq = adapter->msix_entries[vector].vector;
5531                 disable_irq(msix_irq);
5532                 e1000_intr_msix_tx(msix_irq, netdev);
5533                 enable_irq(msix_irq);
5534
5535                 vector++;
5536                 msix_irq = adapter->msix_entries[vector].vector;
5537                 disable_irq(msix_irq);
5538                 e1000_msix_other(msix_irq, netdev);
5539                 enable_irq(msix_irq);
5540         }
5541
5542         return IRQ_HANDLED;
5543 }
5544
5545 /*
5546  * Polling 'interrupt' - used by things like netconsole to send skbs
5547  * without having to re-enable interrupts. It's not called while
5548  * the interrupt routine is executing.
5549  */
5550 static void e1000_netpoll(struct net_device *netdev)
5551 {
5552         struct e1000_adapter *adapter = netdev_priv(netdev);
5553
5554         switch (adapter->int_mode) {
5555         case E1000E_INT_MODE_MSIX:
5556                 e1000_intr_msix(adapter->pdev->irq, netdev);
5557                 break;
5558         case E1000E_INT_MODE_MSI:
5559                 disable_irq(adapter->pdev->irq);
5560                 e1000_intr_msi(adapter->pdev->irq, netdev);
5561                 enable_irq(adapter->pdev->irq);
5562                 break;
5563         default: /* E1000E_INT_MODE_LEGACY */
5564                 disable_irq(adapter->pdev->irq);
5565                 e1000_intr(adapter->pdev->irq, netdev);
5566                 enable_irq(adapter->pdev->irq);
5567                 break;
5568         }
5569 }
5570 #endif
5571
5572 /**
5573  * e1000_io_error_detected - called when PCI error is detected
5574  * @pdev: Pointer to PCI device
5575  * @state: The current pci connection state
5576  *
5577  * This function is called after a PCI bus error affecting
5578  * this device has been detected.
5579  */
5580 static pci_ers_result_t e1000_io_error_detected(struct pci_dev *pdev,
5581                                                 pci_channel_state_t state)
5582 {
5583         struct net_device *netdev = pci_get_drvdata(pdev);
5584         struct e1000_adapter *adapter = netdev_priv(netdev);
5585
5586         netif_device_detach(netdev);
5587
5588         if (state == pci_channel_io_perm_failure)
5589                 return PCI_ERS_RESULT_DISCONNECT;
5590
5591         if (netif_running(netdev))
5592                 e1000e_down(adapter);
5593         pci_disable_device(pdev);
5594
5595         /* Request a slot slot reset. */
5596         return PCI_ERS_RESULT_NEED_RESET;
5597 }
5598
5599 /**
5600  * e1000_io_slot_reset - called after the pci bus has been reset.
5601  * @pdev: Pointer to PCI device
5602  *
5603  * Restart the card from scratch, as if from a cold-boot. Implementation
5604  * resembles the first-half of the e1000_resume routine.
5605  */
5606 static pci_ers_result_t e1000_io_slot_reset(struct pci_dev *pdev)
5607 {
5608         struct net_device *netdev = pci_get_drvdata(pdev);
5609         struct e1000_adapter *adapter = netdev_priv(netdev);
5610         struct e1000_hw *hw = &adapter->hw;
5611         int err;
5612         pci_ers_result_t result;
5613
5614         if (adapter->flags2 & FLAG2_DISABLE_ASPM_L1)
5615                 e1000e_disable_aspm(pdev, PCIE_LINK_STATE_L1);
5616         err = pci_enable_device_mem(pdev);
5617         if (err) {
5618                 dev_err(&pdev->dev,
5619                         "Cannot re-enable PCI device after reset.\n");
5620                 result = PCI_ERS_RESULT_DISCONNECT;
5621         } else {
5622                 pci_set_master(pdev);
5623                 pdev->state_saved = true;
5624                 pci_restore_state(pdev);
5625
5626                 pci_enable_wake(pdev, PCI_D3hot, 0);
5627                 pci_enable_wake(pdev, PCI_D3cold, 0);
5628
5629                 e1000e_reset(adapter);
5630                 ew32(WUS, ~0);
5631                 result = PCI_ERS_RESULT_RECOVERED;
5632         }
5633
5634         pci_cleanup_aer_uncorrect_error_status(pdev);
5635
5636         return result;
5637 }
5638
5639 /**
5640  * e1000_io_resume - called when traffic can start flowing again.
5641  * @pdev: Pointer to PCI device
5642  *
5643  * This callback is called when the error recovery driver tells us that
5644  * its OK to resume normal operation. Implementation resembles the
5645  * second-half of the e1000_resume routine.
5646  */
5647 static void e1000_io_resume(struct pci_dev *pdev)
5648 {
5649         struct net_device *netdev = pci_get_drvdata(pdev);
5650         struct e1000_adapter *adapter = netdev_priv(netdev);
5651
5652         e1000_init_manageability_pt(adapter);
5653
5654         if (netif_running(netdev)) {
5655                 if (e1000e_up(adapter)) {
5656                         dev_err(&pdev->dev,
5657                                 "can't bring device back up after reset\n");
5658                         return;
5659                 }
5660         }
5661
5662         netif_device_attach(netdev);
5663
5664         /*
5665          * If the controller has AMT, do not set DRV_LOAD until the interface
5666          * is up.  For all other cases, let the f/w know that the h/w is now
5667          * under the control of the driver.
5668          */
5669         if (!(adapter->flags & FLAG_HAS_AMT))
5670                 e1000e_get_hw_control(adapter);
5671
5672 }
5673
5674 static void e1000_print_device_info(struct e1000_adapter *adapter)
5675 {
5676         struct e1000_hw *hw = &adapter->hw;
5677         struct net_device *netdev = adapter->netdev;
5678         u32 ret_val;
5679         u8 pba_str[E1000_PBANUM_LENGTH];
5680
5681         /* print bus type/speed/width info */
5682         e_info("(PCI Express:2.5GB/s:%s) %pM\n",
5683                /* bus width */
5684                ((hw->bus.width == e1000_bus_width_pcie_x4) ? "Width x4" :
5685                 "Width x1"),
5686                /* MAC address */
5687                netdev->dev_addr);
5688         e_info("Intel(R) PRO/%s Network Connection\n",
5689                (hw->phy.type == e1000_phy_ife) ? "10/100" : "1000");
5690         ret_val = e1000_read_pba_string_generic(hw, pba_str,
5691                                                 E1000_PBANUM_LENGTH);
5692         if (ret_val)
5693                 strncpy((char *)pba_str, "Unknown", sizeof(pba_str) - 1);
5694         e_info("MAC: %d, PHY: %d, PBA No: %s\n",
5695                hw->mac.type, hw->phy.type, pba_str);
5696 }
5697
5698 static void e1000_eeprom_checks(struct e1000_adapter *adapter)
5699 {
5700         struct e1000_hw *hw = &adapter->hw;
5701         int ret_val;
5702         u16 buf = 0;
5703
5704         if (hw->mac.type != e1000_82573)
5705                 return;
5706
5707         ret_val = e1000_read_nvm(hw, NVM_INIT_CONTROL2_REG, 1, &buf);
5708         if (!ret_val && (!(le16_to_cpu(buf) & (1 << 0)))) {
5709                 /* Deep Smart Power Down (DSPD) */
5710                 dev_warn(&adapter->pdev->dev,
5711                          "Warning: detected DSPD enabled in EEPROM\n");
5712         }
5713 }
5714
5715 static const struct net_device_ops e1000e_netdev_ops = {
5716         .ndo_open               = e1000_open,
5717         .ndo_stop               = e1000_close,
5718         .ndo_start_xmit         = e1000_xmit_frame,
5719         .ndo_get_stats64        = e1000e_get_stats64,
5720         .ndo_set_multicast_list = e1000_set_multi,
5721         .ndo_set_mac_address    = e1000_set_mac,
5722         .ndo_change_mtu         = e1000_change_mtu,
5723         .ndo_do_ioctl           = e1000_ioctl,
5724         .ndo_tx_timeout         = e1000_tx_timeout,
5725         .ndo_validate_addr      = eth_validate_addr,
5726
5727         .ndo_vlan_rx_register   = e1000_vlan_rx_register,
5728         .ndo_vlan_rx_add_vid    = e1000_vlan_rx_add_vid,
5729         .ndo_vlan_rx_kill_vid   = e1000_vlan_rx_kill_vid,
5730 #ifdef CONFIG_NET_POLL_CONTROLLER
5731         .ndo_poll_controller    = e1000_netpoll,
5732 #endif
5733 };
5734
5735 /**
5736  * e1000_probe - Device Initialization Routine
5737  * @pdev: PCI device information struct
5738  * @ent: entry in e1000_pci_tbl
5739  *
5740  * Returns 0 on success, negative on failure
5741  *
5742  * e1000_probe initializes an adapter identified by a pci_dev structure.
5743  * The OS initialization, configuring of the adapter private structure,
5744  * and a hardware reset occur.
5745  **/
5746 static int __devinit e1000_probe(struct pci_dev *pdev,
5747                                  const struct pci_device_id *ent)
5748 {
5749         struct net_device *netdev;
5750         struct e1000_adapter *adapter;
5751         struct e1000_hw *hw;
5752         const struct e1000_info *ei = e1000_info_tbl[ent->driver_data];
5753         resource_size_t mmio_start, mmio_len;
5754         resource_size_t flash_start, flash_len;
5755
5756         static int cards_found;
5757         int i, err, pci_using_dac;
5758         u16 eeprom_data = 0;
5759         u16 eeprom_apme_mask = E1000_EEPROM_APME;
5760
5761         if (ei->flags2 & FLAG2_DISABLE_ASPM_L1)
5762                 e1000e_disable_aspm(pdev, PCIE_LINK_STATE_L1);
5763
5764         err = pci_enable_device_mem(pdev);
5765         if (err)
5766                 return err;
5767
5768         pci_using_dac = 0;
5769         err = dma_set_mask(&pdev->dev, DMA_BIT_MASK(64));
5770         if (!err) {
5771                 err = dma_set_coherent_mask(&pdev->dev, DMA_BIT_MASK(64));
5772                 if (!err)
5773                         pci_using_dac = 1;
5774         } else {
5775                 err = dma_set_mask(&pdev->dev, DMA_BIT_MASK(32));
5776                 if (err) {
5777                         err = dma_set_coherent_mask(&pdev->dev,
5778                                                     DMA_BIT_MASK(32));
5779                         if (err) {
5780                                 dev_err(&pdev->dev, "No usable DMA "
5781                                         "configuration, aborting\n");
5782                                 goto err_dma;
5783                         }
5784                 }
5785         }
5786
5787         err = pci_request_selected_regions_exclusive(pdev,
5788                                           pci_select_bars(pdev, IORESOURCE_MEM),
5789                                           e1000e_driver_name);
5790         if (err)
5791                 goto err_pci_reg;
5792
5793         /* AER (Advanced Error Reporting) hooks */
5794         pci_enable_pcie_error_reporting(pdev);
5795
5796         pci_set_master(pdev);
5797         /* PCI config space info */
5798         err = pci_save_state(pdev);
5799         if (err)
5800                 goto err_alloc_etherdev;
5801
5802         err = -ENOMEM;
5803         netdev = alloc_etherdev(sizeof(struct e1000_adapter));
5804         if (!netdev)
5805                 goto err_alloc_etherdev;
5806
5807         SET_NETDEV_DEV(netdev, &pdev->dev);
5808
5809         netdev->irq = pdev->irq;
5810
5811         pci_set_drvdata(pdev, netdev);
5812         adapter = netdev_priv(netdev);
5813         hw = &adapter->hw;
5814         adapter->netdev = netdev;
5815         adapter->pdev = pdev;
5816         adapter->ei = ei;
5817         adapter->pba = ei->pba;
5818         adapter->flags = ei->flags;
5819         adapter->flags2 = ei->flags2;
5820         adapter->hw.adapter = adapter;
5821         adapter->hw.mac.type = ei->mac;
5822         adapter->max_hw_frame_size = ei->max_hw_frame_size;
5823         adapter->msg_enable = (1 << NETIF_MSG_DRV | NETIF_MSG_PROBE) - 1;
5824
5825         mmio_start = pci_resource_start(pdev, 0);
5826         mmio_len = pci_resource_len(pdev, 0);
5827
5828         err = -EIO;
5829         adapter->hw.hw_addr = ioremap(mmio_start, mmio_len);
5830         if (!adapter->hw.hw_addr)
5831                 goto err_ioremap;
5832
5833         if ((adapter->flags & FLAG_HAS_FLASH) &&
5834             (pci_resource_flags(pdev, 1) & IORESOURCE_MEM)) {
5835                 flash_start = pci_resource_start(pdev, 1);
5836                 flash_len = pci_resource_len(pdev, 1);
5837                 adapter->hw.flash_address = ioremap(flash_start, flash_len);
5838                 if (!adapter->hw.flash_address)
5839                         goto err_flashmap;
5840         }
5841
5842         /* construct the net_device struct */
5843         netdev->netdev_ops              = &e1000e_netdev_ops;
5844         e1000e_set_ethtool_ops(netdev);
5845         netdev->watchdog_timeo          = 5 * HZ;
5846         netif_napi_add(netdev, &adapter->napi, e1000_clean, 64);
5847         strncpy(netdev->name, pci_name(pdev), sizeof(netdev->name) - 1);
5848
5849         netdev->mem_start = mmio_start;
5850         netdev->mem_end = mmio_start + mmio_len;
5851
5852         adapter->bd_number = cards_found++;
5853
5854         e1000e_check_options(adapter);
5855
5856         /* setup adapter struct */
5857         err = e1000_sw_init(adapter);
5858         if (err)
5859                 goto err_sw_init;
5860
5861         memcpy(&hw->mac.ops, ei->mac_ops, sizeof(hw->mac.ops));
5862         memcpy(&hw->nvm.ops, ei->nvm_ops, sizeof(hw->nvm.ops));
5863         memcpy(&hw->phy.ops, ei->phy_ops, sizeof(hw->phy.ops));
5864
5865         err = ei->get_variants(adapter);
5866         if (err)
5867                 goto err_hw_init;
5868
5869         if ((adapter->flags & FLAG_IS_ICH) &&
5870             (adapter->flags & FLAG_READ_ONLY_NVM))
5871                 e1000e_write_protect_nvm_ich8lan(&adapter->hw);
5872
5873         hw->mac.ops.get_bus_info(&adapter->hw);
5874
5875         adapter->hw.phy.autoneg_wait_to_complete = 0;
5876
5877         /* Copper options */
5878         if (adapter->hw.phy.media_type == e1000_media_type_copper) {
5879                 adapter->hw.phy.mdix = AUTO_ALL_MODES;
5880                 adapter->hw.phy.disable_polarity_correction = 0;
5881                 adapter->hw.phy.ms_type = e1000_ms_hw_default;
5882         }
5883
5884         if (e1000_check_reset_block(&adapter->hw))
5885                 e_info("PHY reset is blocked due to SOL/IDER session.\n");
5886
5887         netdev->features = NETIF_F_SG |
5888                            NETIF_F_HW_CSUM |
5889                            NETIF_F_HW_VLAN_TX |
5890                            NETIF_F_HW_VLAN_RX;
5891
5892         if (adapter->flags & FLAG_HAS_HW_VLAN_FILTER)
5893                 netdev->features |= NETIF_F_HW_VLAN_FILTER;
5894
5895         netdev->features |= NETIF_F_TSO;
5896         netdev->features |= NETIF_F_TSO6;
5897
5898         netdev->vlan_features |= NETIF_F_TSO;
5899         netdev->vlan_features |= NETIF_F_TSO6;
5900         netdev->vlan_features |= NETIF_F_HW_CSUM;
5901         netdev->vlan_features |= NETIF_F_SG;
5902
5903         if (pci_using_dac) {
5904                 netdev->features |= NETIF_F_HIGHDMA;
5905                 netdev->vlan_features |= NETIF_F_HIGHDMA;
5906         }
5907
5908         if (e1000e_enable_mng_pass_thru(&adapter->hw))
5909                 adapter->flags |= FLAG_MNG_PT_ENABLED;
5910
5911         /*
5912          * before reading the NVM, reset the controller to
5913          * put the device in a known good starting state
5914          */
5915         adapter->hw.mac.ops.reset_hw(&adapter->hw);
5916
5917         /*
5918          * systems with ASPM and others may see the checksum fail on the first
5919          * attempt. Let's give it a few tries
5920          */
5921         for (i = 0;; i++) {
5922                 if (e1000_validate_nvm_checksum(&adapter->hw) >= 0)
5923                         break;
5924                 if (i == 2) {
5925                         e_err("The NVM Checksum Is Not Valid\n");
5926                         err = -EIO;
5927                         goto err_eeprom;
5928                 }
5929         }
5930
5931         e1000_eeprom_checks(adapter);
5932
5933         /* copy the MAC address */
5934         if (e1000e_read_mac_addr(&adapter->hw))
5935                 e_err("NVM Read Error while reading MAC address\n");
5936
5937         memcpy(netdev->dev_addr, adapter->hw.mac.addr, netdev->addr_len);
5938         memcpy(netdev->perm_addr, adapter->hw.mac.addr, netdev->addr_len);
5939
5940         if (!is_valid_ether_addr(netdev->perm_addr)) {
5941                 e_err("Invalid MAC Address: %pM\n", netdev->perm_addr);
5942                 err = -EIO;
5943                 goto err_eeprom;
5944         }
5945
5946         init_timer(&adapter->watchdog_timer);
5947         adapter->watchdog_timer.function = e1000_watchdog;
5948         adapter->watchdog_timer.data = (unsigned long) adapter;
5949
5950         init_timer(&adapter->phy_info_timer);
5951         adapter->phy_info_timer.function = e1000_update_phy_info;
5952         adapter->phy_info_timer.data = (unsigned long) adapter;
5953
5954         INIT_WORK(&adapter->reset_task, e1000_reset_task);
5955         INIT_WORK(&adapter->watchdog_task, e1000_watchdog_task);
5956         INIT_WORK(&adapter->downshift_task, e1000e_downshift_workaround);
5957         INIT_WORK(&adapter->update_phy_task, e1000e_update_phy_task);
5958         INIT_WORK(&adapter->print_hang_task, e1000_print_hw_hang);
5959         INIT_WORK(&adapter->led_blink_task, e1000e_led_blink_task);
5960
5961         /* Initialize link parameters. User can change them with ethtool */
5962         adapter->hw.mac.autoneg = 1;
5963         adapter->fc_autoneg = 1;
5964         adapter->hw.fc.requested_mode = e1000_fc_default;
5965         adapter->hw.fc.current_mode = e1000_fc_default;
5966         adapter->hw.phy.autoneg_advertised = 0x2f;
5967
5968         /* ring size defaults */
5969         adapter->rx_ring->count = 256;
5970         adapter->tx_ring->count = 256;
5971
5972         /*
5973          * Initial Wake on LAN setting - If APM wake is enabled in
5974          * the EEPROM, enable the ACPI Magic Packet filter
5975          */
5976         if (adapter->flags & FLAG_APME_IN_WUC) {
5977                 /* APME bit in EEPROM is mapped to WUC.APME */
5978                 eeprom_data = er32(WUC);
5979                 eeprom_apme_mask = E1000_WUC_APME;
5980                 if (eeprom_data & E1000_WUC_PHY_WAKE)
5981                         adapter->flags2 |= FLAG2_HAS_PHY_WAKEUP;
5982         } else if (adapter->flags & FLAG_APME_IN_CTRL3) {
5983                 if (adapter->flags & FLAG_APME_CHECK_PORT_B &&
5984                     (adapter->hw.bus.func == 1))
5985                         e1000_read_nvm(&adapter->hw,
5986                                 NVM_INIT_CONTROL3_PORT_B, 1, &eeprom_data);
5987                 else
5988                         e1000_read_nvm(&adapter->hw,
5989                                 NVM_INIT_CONTROL3_PORT_A, 1, &eeprom_data);
5990         }
5991
5992         /* fetch WoL from EEPROM */
5993         if (eeprom_data & eeprom_apme_mask)
5994                 adapter->eeprom_wol |= E1000_WUFC_MAG;
5995
5996         /*
5997          * now that we have the eeprom settings, apply the special cases
5998          * where the eeprom may be wrong or the board simply won't support
5999          * wake on lan on a particular port
6000          */
6001         if (!(adapter->flags & FLAG_HAS_WOL))
6002                 adapter->eeprom_wol = 0;
6003
6004         /* initialize the wol settings based on the eeprom settings */
6005         adapter->wol = adapter->eeprom_wol;
6006         device_set_wakeup_enable(&adapter->pdev->dev, adapter->wol);
6007
6008         /* save off EEPROM version number */
6009         e1000_read_nvm(&adapter->hw, 5, 1, &adapter->eeprom_vers);
6010
6011         /* reset the hardware with the new settings */
6012         e1000e_reset(adapter);
6013
6014         /*
6015          * If the controller has AMT, do not set DRV_LOAD until the interface
6016          * is up.  For all other cases, let the f/w know that the h/w is now
6017          * under the control of the driver.
6018          */
6019         if (!(adapter->flags & FLAG_HAS_AMT))
6020                 e1000e_get_hw_control(adapter);
6021
6022         strncpy(netdev->name, "eth%d", sizeof(netdev->name) - 1);
6023         err = register_netdev(netdev);
6024         if (err)
6025                 goto err_register;
6026
6027         /* carrier off reporting is important to ethtool even BEFORE open */
6028         netif_carrier_off(netdev);
6029
6030         e1000_print_device_info(adapter);
6031
6032         if (pci_dev_run_wake(pdev))
6033                 pm_runtime_put_noidle(&pdev->dev);
6034
6035         return 0;
6036
6037 err_register:
6038         if (!(adapter->flags & FLAG_HAS_AMT))
6039                 e1000e_release_hw_control(adapter);
6040 err_eeprom:
6041         if (!e1000_check_reset_block(&adapter->hw))
6042                 e1000_phy_hw_reset(&adapter->hw);
6043 err_hw_init:
6044         kfree(adapter->tx_ring);
6045         kfree(adapter->rx_ring);
6046 err_sw_init:
6047         if (adapter->hw.flash_address)
6048                 iounmap(adapter->hw.flash_address);
6049         e1000e_reset_interrupt_capability(adapter);
6050 err_flashmap:
6051         iounmap(adapter->hw.hw_addr);
6052 err_ioremap:
6053         free_netdev(netdev);
6054 err_alloc_etherdev:
6055         pci_release_selected_regions(pdev,
6056                                      pci_select_bars(pdev, IORESOURCE_MEM));
6057 err_pci_reg:
6058 err_dma:
6059         pci_disable_device(pdev);
6060         return err;
6061 }
6062
6063 /**
6064  * e1000_remove - Device Removal Routine
6065  * @pdev: PCI device information struct
6066  *
6067  * e1000_remove is called by the PCI subsystem to alert the driver
6068  * that it should release a PCI device.  The could be caused by a
6069  * Hot-Plug event, or because the driver is going to be removed from
6070  * memory.
6071  **/
6072 static void __devexit e1000_remove(struct pci_dev *pdev)
6073 {
6074         struct net_device *netdev = pci_get_drvdata(pdev);
6075         struct e1000_adapter *adapter = netdev_priv(netdev);
6076         bool down = test_bit(__E1000_DOWN, &adapter->state);
6077
6078         /*
6079          * The timers may be rescheduled, so explicitly disable them
6080          * from being rescheduled.
6081          */
6082         if (!down)
6083                 set_bit(__E1000_DOWN, &adapter->state);
6084         del_timer_sync(&adapter->watchdog_timer);
6085         del_timer_sync(&adapter->phy_info_timer);
6086
6087         cancel_work_sync(&adapter->reset_task);
6088         cancel_work_sync(&adapter->watchdog_task);
6089         cancel_work_sync(&adapter->downshift_task);
6090         cancel_work_sync(&adapter->update_phy_task);
6091         cancel_work_sync(&adapter->led_blink_task);
6092         cancel_work_sync(&adapter->print_hang_task);
6093
6094         if (!(netdev->flags & IFF_UP))
6095                 e1000_power_down_phy(adapter);
6096
6097         /* Don't lie to e1000_close() down the road. */
6098         if (!down)
6099                 clear_bit(__E1000_DOWN, &adapter->state);
6100         unregister_netdev(netdev);
6101
6102         if (pci_dev_run_wake(pdev))
6103                 pm_runtime_get_noresume(&pdev->dev);
6104
6105         /*
6106          * Release control of h/w to f/w.  If f/w is AMT enabled, this
6107          * would have already happened in close and is redundant.
6108          */
6109         e1000e_release_hw_control(adapter);
6110
6111         e1000e_reset_interrupt_capability(adapter);
6112         kfree(adapter->tx_ring);
6113         kfree(adapter->rx_ring);
6114
6115         iounmap(adapter->hw.hw_addr);
6116         if (adapter->hw.flash_address)
6117                 iounmap(adapter->hw.flash_address);
6118         pci_release_selected_regions(pdev,
6119                                      pci_select_bars(pdev, IORESOURCE_MEM));
6120
6121         free_netdev(netdev);
6122
6123         /* AER disable */
6124         pci_disable_pcie_error_reporting(pdev);
6125
6126         pci_disable_device(pdev);
6127 }
6128
6129 /* PCI Error Recovery (ERS) */
6130 static struct pci_error_handlers e1000_err_handler = {
6131         .error_detected = e1000_io_error_detected,
6132         .slot_reset = e1000_io_slot_reset,
6133         .resume = e1000_io_resume,
6134 };
6135
6136 static DEFINE_PCI_DEVICE_TABLE(e1000_pci_tbl) = {
6137         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82571EB_COPPER), board_82571 },
6138         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82571EB_FIBER), board_82571 },
6139         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82571EB_QUAD_COPPER), board_82571 },
6140         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82571EB_QUAD_COPPER_LP), board_82571 },
6141         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82571EB_QUAD_FIBER), board_82571 },
6142         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82571EB_SERDES), board_82571 },
6143         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82571EB_SERDES_DUAL), board_82571 },
6144         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82571EB_SERDES_QUAD), board_82571 },
6145         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82571PT_QUAD_COPPER), board_82571 },
6146
6147         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82572EI), board_82572 },
6148         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82572EI_COPPER), board_82572 },
6149         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82572EI_FIBER), board_82572 },
6150         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82572EI_SERDES), board_82572 },
6151
6152         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82573E), board_82573 },
6153         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82573E_IAMT), board_82573 },
6154         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82573L), board_82573 },
6155
6156         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82574L), board_82574 },
6157         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82574LA), board_82574 },
6158         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82583V), board_82583 },
6159
6160         { PCI_VDEVICE(INTEL, E1000_DEV_ID_80003ES2LAN_COPPER_DPT),
6161           board_80003es2lan },
6162         { PCI_VDEVICE(INTEL, E1000_DEV_ID_80003ES2LAN_COPPER_SPT),
6163           board_80003es2lan },
6164         { PCI_VDEVICE(INTEL, E1000_DEV_ID_80003ES2LAN_SERDES_DPT),
6165           board_80003es2lan },
6166         { PCI_VDEVICE(INTEL, E1000_DEV_ID_80003ES2LAN_SERDES_SPT),
6167           board_80003es2lan },
6168
6169         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH8_IFE), board_ich8lan },
6170         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH8_IFE_G), board_ich8lan },
6171         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH8_IFE_GT), board_ich8lan },
6172         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH8_IGP_AMT), board_ich8lan },
6173         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH8_IGP_C), board_ich8lan },
6174         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH8_IGP_M), board_ich8lan },
6175         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH8_IGP_M_AMT), board_ich8lan },
6176         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH8_82567V_3), board_ich8lan },
6177
6178         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH9_IFE), board_ich9lan },
6179         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH9_IFE_G), board_ich9lan },
6180         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH9_IFE_GT), board_ich9lan },
6181         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH9_IGP_AMT), board_ich9lan },
6182         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH9_IGP_C), board_ich9lan },
6183         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH9_BM), board_ich9lan },
6184         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH9_IGP_M), board_ich9lan },
6185         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH9_IGP_M_AMT), board_ich9lan },
6186         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH9_IGP_M_V), board_ich9lan },
6187
6188         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH10_R_BM_LM), board_ich9lan },
6189         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH10_R_BM_LF), board_ich9lan },
6190         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH10_R_BM_V), board_ich9lan },
6191
6192         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH10_D_BM_LM), board_ich10lan },
6193         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH10_D_BM_LF), board_ich10lan },
6194         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH10_D_BM_V), board_ich10lan },
6195
6196         { PCI_VDEVICE(INTEL, E1000_DEV_ID_PCH_M_HV_LM), board_pchlan },
6197         { PCI_VDEVICE(INTEL, E1000_DEV_ID_PCH_M_HV_LC), board_pchlan },
6198         { PCI_VDEVICE(INTEL, E1000_DEV_ID_PCH_D_HV_DM), board_pchlan },
6199         { PCI_VDEVICE(INTEL, E1000_DEV_ID_PCH_D_HV_DC), board_pchlan },
6200
6201         { PCI_VDEVICE(INTEL, E1000_DEV_ID_PCH2_LV_LM), board_pch2lan },
6202         { PCI_VDEVICE(INTEL, E1000_DEV_ID_PCH2_LV_V), board_pch2lan },
6203
6204         { }     /* terminate list */
6205 };
6206 MODULE_DEVICE_TABLE(pci, e1000_pci_tbl);
6207
6208 #ifdef CONFIG_PM_OPS
6209 static const struct dev_pm_ops e1000_pm_ops = {
6210         SET_SYSTEM_SLEEP_PM_OPS(e1000_suspend, e1000_resume)
6211         SET_RUNTIME_PM_OPS(e1000_runtime_suspend,
6212                                 e1000_runtime_resume, e1000_idle)
6213 };
6214 #endif
6215
6216 /* PCI Device API Driver */
6217 static struct pci_driver e1000_driver = {
6218         .name     = e1000e_driver_name,
6219         .id_table = e1000_pci_tbl,
6220         .probe    = e1000_probe,
6221         .remove   = __devexit_p(e1000_remove),
6222 #ifdef CONFIG_PM_OPS
6223         .driver.pm = &e1000_pm_ops,
6224 #endif
6225         .shutdown = e1000_shutdown,
6226         .err_handler = &e1000_err_handler
6227 };
6228
6229 /**
6230  * e1000_init_module - Driver Registration Routine
6231  *
6232  * e1000_init_module is the first routine called when the driver is
6233  * loaded. All it does is register with the PCI subsystem.
6234  **/
6235 static int __init e1000_init_module(void)
6236 {
6237         int ret;
6238         pr_info("Intel(R) PRO/1000 Network Driver - %s\n",
6239                 e1000e_driver_version);
6240         pr_info("Copyright(c) 1999 - 2011 Intel Corporation.\n");
6241         ret = pci_register_driver(&e1000_driver);
6242
6243         return ret;
6244 }
6245 module_init(e1000_init_module);
6246
6247 /**
6248  * e1000_exit_module - Driver Exit Cleanup Routine
6249  *
6250  * e1000_exit_module is called just before the driver is removed
6251  * from memory.
6252  **/
6253 static void __exit e1000_exit_module(void)
6254 {
6255         pci_unregister_driver(&e1000_driver);
6256 }
6257 module_exit(e1000_exit_module);
6258
6259
6260 MODULE_AUTHOR("Intel Corporation, <linux.nics@intel.com>");
6261 MODULE_DESCRIPTION("Intel(R) PRO/1000 Network Driver");
6262 MODULE_LICENSE("GPL");
6263 MODULE_VERSION(DRV_VERSION);
6264
6265 /* e1000_main.c */