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

   1 /*******************************************************************************
   2
   3   Intel PRO/1000 Linux driver
   4   Copyright(c) 1999 - 2012 Intel Corporation.
   5
   6   This program is free software; you can redistribute it and/or modify it
   7   under the terms and conditions of the GNU General Public License,
   8   version 2, as published by the Free Software Foundation.
   9
  10   This program is distributed in the hope it will be useful, but WITHOUT
  11   ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
  12   FITNESS FOR A PARTICULAR PURPOSE.  See the GNU General Public License for
  13   more details.
  14
  15   You should have received a copy of the GNU General Public License along with
  16   this program; if not, write to the Free Software Foundation, Inc.,
  17   51 Franklin St - Fifth Floor, Boston, MA 02110-1301 USA.
  18
  19   The full GNU General Public License is included in this distribution in
  20   the file called "COPYING".
  21
  22   Contact Information:
  23   Linux NICS <linux.nics@intel.com>
  24   e1000-devel Mailing List <e1000-devel@lists.sourceforge.net>
  25   Intel Corporation, 5200 N.E. Elam Young Parkway, Hillsboro, OR 97124-6497
  26
  27 *******************************************************************************/
  28
  29 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
  30
  31 #include <linux/module.h>
  32 #include <linux/types.h>
  33 #include <linux/init.h>
  34 #include <linux/pci.h>
  35 #include <linux/vmalloc.h>
  36 #include <linux/pagemap.h>
  37 #include <linux/delay.h>
  38 #include <linux/netdevice.h>
  39 #include <linux/interrupt.h>
  40 #include <linux/tcp.h>
  41 #include <linux/ipv6.h>
  42 #include <linux/slab.h>
  43 #include <net/checksum.h>
  44 #include <net/ip6_checksum.h>
  45 #include <linux/mii.h>
  46 #include <linux/ethtool.h>
  47 #include <linux/if_vlan.h>
  48 #include <linux/cpu.h>
  49 #include <linux/smp.h>
  50 #include <linux/pm_qos.h>
  51 #include <linux/pm_runtime.h>
  52 #include <linux/aer.h>
  53 #include <linux/prefetch.h>
  54
  55 #include "e1000.h"
  56
  57 #define DRV_EXTRAVERSION "-k"
  58
  59 #define DRV_VERSION "2.0.0" DRV_EXTRAVERSION
  60 char e1000e_driver_name[] = "e1000e";
  61 const char e1000e_driver_version[] = DRV_VERSION;
  62
  63 #define DEFAULT_MSG_ENABLE (NETIF_MSG_DRV|NETIF_MSG_PROBE|NETIF_MSG_LINK)
  64 static int debug = -1;
  65 module_param(debug, int, 0);
  66 MODULE_PARM_DESC(debug, "Debug level (0=none,...,16=all)");
  67
  68 static void e1000e_disable_aspm(struct pci_dev *pdev, u16 state);
  69
  70 static const struct e1000_info *e1000_info_tbl[] = {
  71         [board_82571]           = &e1000_82571_info,
  72         [board_82572]           = &e1000_82572_info,
  73         [board_82573]           = &e1000_82573_info,
  74         [board_82574]           = &e1000_82574_info,
  75         [board_82583]           = &e1000_82583_info,
  76         [board_80003es2lan]     = &e1000_es2_info,
  77         [board_ich8lan]         = &e1000_ich8_info,
  78         [board_ich9lan]         = &e1000_ich9_info,
  79         [board_ich10lan]        = &e1000_ich10_info,
  80         [board_pchlan]          = &e1000_pch_info,
  81         [board_pch2lan]         = &e1000_pch2_info,
  82         [board_pch_lpt]         = &e1000_pch_lpt_info,
  83 };
  84
  85 struct e1000_reg_info {
  86         u32 ofs;
  87         char *name;
  88 };
  89
  90 #define E1000_RDFH      0x02410 /* Rx Data FIFO Head - RW */
  91 #define E1000_RDFT      0x02418 /* Rx Data FIFO Tail - RW */
  92 #define E1000_RDFHS     0x02420 /* Rx Data FIFO Head Saved - RW */
  93 #define E1000_RDFTS     0x02428 /* Rx Data FIFO Tail Saved - RW */
  94 #define E1000_RDFPC     0x02430 /* Rx Data FIFO Packet Count - RW */
  95
  96 #define E1000_TDFH      0x03410 /* Tx Data FIFO Head - RW */
  97 #define E1000_TDFT      0x03418 /* Tx Data FIFO Tail - RW */
  98 #define E1000_TDFHS     0x03420 /* Tx Data FIFO Head Saved - RW */
  99 #define E1000_TDFTS     0x03428 /* Tx Data FIFO Tail Saved - RW */
 100 #define E1000_TDFPC     0x03430 /* Tx Data FIFO Packet Count - RW */
 101
 102 static const struct e1000_reg_info e1000_reg_info_tbl[] = {
 103
 104         /* General Registers */
 105         {E1000_CTRL, "CTRL"},
 106         {E1000_STATUS, "STATUS"},
 107         {E1000_CTRL_EXT, "CTRL_EXT"},
 108
 109         /* Interrupt Registers */
 110         {E1000_ICR, "ICR"},
 111
 112         /* Rx Registers */
 113         {E1000_RCTL, "RCTL"},
 114         {E1000_RDLEN(0), "RDLEN"},
 115         {E1000_RDH(0), "RDH"},
 116         {E1000_RDT(0), "RDT"},
 117         {E1000_RDTR, "RDTR"},
 118         {E1000_RXDCTL(0), "RXDCTL"},
 119         {E1000_ERT, "ERT"},
 120         {E1000_RDBAL(0), "RDBAL"},
 121         {E1000_RDBAH(0), "RDBAH"},
 122         {E1000_RDFH, "RDFH"},
 123         {E1000_RDFT, "RDFT"},
 124         {E1000_RDFHS, "RDFHS"},
 125         {E1000_RDFTS, "RDFTS"},
 126         {E1000_RDFPC, "RDFPC"},
 127
 128         /* Tx Registers */
 129         {E1000_TCTL, "TCTL"},
 130         {E1000_TDBAL(0), "TDBAL"},
 131         {E1000_TDBAH(0), "TDBAH"},
 132         {E1000_TDLEN(0), "TDLEN"},
 133         {E1000_TDH(0), "TDH"},
 134         {E1000_TDT(0), "TDT"},
 135         {E1000_TIDV, "TIDV"},
 136         {E1000_TXDCTL(0), "TXDCTL"},
 137         {E1000_TADV, "TADV"},
 138         {E1000_TARC(0), "TARC"},
 139         {E1000_TDFH, "TDFH"},
 140         {E1000_TDFT, "TDFT"},
 141         {E1000_TDFHS, "TDFHS"},
 142         {E1000_TDFTS, "TDFTS"},
 143         {E1000_TDFPC, "TDFPC"},
 144
 145         /* List Terminator */
 146         {0, NULL}
 147 };
 148
 149 /*
 150  * e1000_regdump - register printout routine
 151  */
 152 static void e1000_regdump(struct e1000_hw *hw, struct e1000_reg_info *reginfo)
 153 {
 154         int n = 0;
 155         char rname[16];
 156         u32 regs[8];
 157
 158         switch (reginfo->ofs) {
 159         case E1000_RXDCTL(0):
 160                 for (n = 0; n < 2; n++)
 161                         regs[n] = __er32(hw, E1000_RXDCTL(n));
 162                 break;
 163         case E1000_TXDCTL(0):
 164                 for (n = 0; n < 2; n++)
 165                         regs[n] = __er32(hw, E1000_TXDCTL(n));
 166                 break;
 167         case E1000_TARC(0):
 168                 for (n = 0; n < 2; n++)
 169                         regs[n] = __er32(hw, E1000_TARC(n));
 170                 break;
 171         default:
 172                 pr_info("%-15s %08x\n",
 173                         reginfo->name, __er32(hw, reginfo->ofs));
 174                 return;
 175         }
 176
 177         snprintf(rname, 16, "%s%s", reginfo->name, "[0-1]");
 178         pr_info("%-15s %08x %08x\n", rname, regs[0], regs[1]);
 179 }
 180
 181 /*
 182  * e1000e_dump - Print registers, Tx-ring and Rx-ring
 183  */
 184 static void e1000e_dump(struct e1000_adapter *adapter)
 185 {
 186         struct net_device *netdev = adapter->netdev;
 187         struct e1000_hw *hw = &adapter->hw;
 188         struct e1000_reg_info *reginfo;
 189         struct e1000_ring *tx_ring = adapter->tx_ring;
 190         struct e1000_tx_desc *tx_desc;
 191         struct my_u0 {
 192                 __le64 a;
 193                 __le64 b;
 194         } *u0;
 195         struct e1000_buffer *buffer_info;
 196         struct e1000_ring *rx_ring = adapter->rx_ring;
 197         union e1000_rx_desc_packet_split *rx_desc_ps;
 198         union e1000_rx_desc_extended *rx_desc;
 199         struct my_u1 {
 200                 __le64 a;
 201                 __le64 b;
 202                 __le64 c;
 203                 __le64 d;
 204         } *u1;
 205         u32 staterr;
 206         int i = 0;
 207
 208         if (!netif_msg_hw(adapter))
 209                 return;
 210
 211         /* Print netdevice Info */
 212         if (netdev) {
 213                 dev_info(&adapter->pdev->dev, "Net device Info\n");
 214                 pr_info("Device Name     state            trans_start      last_rx\n");
 215                 pr_info("%-15s %016lX %016lX %016lX\n",
 216                         netdev->name, netdev->state, netdev->trans_start,
 217                         netdev->last_rx);
 218         }
 219
 220         /* Print Registers */
 221         dev_info(&adapter->pdev->dev, "Register Dump\n");
 222         pr_info(" Register Name   Value\n");
 223         for (reginfo = (struct e1000_reg_info *)e1000_reg_info_tbl;
 224              reginfo->name; reginfo++) {
 225                 e1000_regdump(hw, reginfo);
 226         }
 227
 228         /* Print Tx Ring Summary */
 229         if (!netdev || !netif_running(netdev))
 230                 return;
 231
 232         dev_info(&adapter->pdev->dev, "Tx Ring Summary\n");
 233         pr_info("Queue [NTU] [NTC] [bi(ntc)->dma  ] leng ntw timestamp\n");
 234         buffer_info = &tx_ring->buffer_info[tx_ring->next_to_clean];
 235         pr_info(" %5d %5X %5X %016llX %04X %3X %016llX\n",
 236                 0, tx_ring->next_to_use, tx_ring->next_to_clean,
 237                 (unsigned long long)buffer_info->dma,
 238                 buffer_info->length,
 239                 buffer_info->next_to_watch,
 240                 (unsigned long long)buffer_info->time_stamp);
 241
 242         /* Print Tx Ring */
 243         if (!netif_msg_tx_done(adapter))
 244                 goto rx_ring_summary;
 245
 246         dev_info(&adapter->pdev->dev, "Tx Ring Dump\n");
 247
 248         /* Transmit Descriptor Formats - DEXT[29] is 0 (Legacy) or 1 (Extended)
 249          *
 250          * Legacy Transmit Descriptor
 251          *   +--------------------------------------------------------------+
 252          * 0 |         Buffer Address [63:0] (Reserved on Write Back)       |
 253          *   +--------------------------------------------------------------+
 254          * 8 | Special  |    CSS     | Status |  CMD    |  CSO   |  Length  |
 255          *   +--------------------------------------------------------------+
 256          *   63       48 47        36 35    32 31     24 23    16 15        0
 257          *
 258          * Extended Context Descriptor (DTYP=0x0) for TSO or checksum offload
 259          *   63      48 47    40 39       32 31             16 15    8 7      0
 260          *   +----------------------------------------------------------------+
 261          * 0 |  TUCSE  | TUCS0  |   TUCSS   |     IPCSE       | IPCS0 | IPCSS |
 262          *   +----------------------------------------------------------------+
 263          * 8 |   MSS   | HDRLEN | RSV | STA | TUCMD | DTYP |      PAYLEN      |
 264          *   +----------------------------------------------------------------+
 265          *   63      48 47    40 39 36 35 32 31   24 23  20 19                0
 266          *
 267          * Extended Data Descriptor (DTYP=0x1)
 268          *   +----------------------------------------------------------------+
 269          * 0 |                     Buffer Address [63:0]                      |
 270          *   +----------------------------------------------------------------+
 271          * 8 | VLAN tag |  POPTS  | Rsvd | Status | Command | DTYP |  DTALEN  |
 272          *   +----------------------------------------------------------------+
 273          *   63       48 47     40 39  36 35    32 31     24 23  20 19        0
 274          */
 275         pr_info("Tl[desc]     [address 63:0  ] [SpeCssSCmCsLen] [bi->dma       ] leng  ntw timestamp        bi->skb <-- Legacy format\n");
 276         pr_info("Tc[desc]     [Ce CoCsIpceCoS] [MssHlRSCm0Plen] [bi->dma       ] leng  ntw timestamp        bi->skb <-- Ext Context format\n");
 277         pr_info("Td[desc]     [address 63:0  ] [VlaPoRSCm1Dlen] [bi->dma       ] leng  ntw timestamp        bi->skb <-- Ext Data format\n");
 278         for (i = 0; tx_ring->desc && (i < tx_ring->count); i++) {
 279                 const char *next_desc;
 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                 if (i == tx_ring->next_to_use && i == tx_ring->next_to_clean)
 284                         next_desc = " NTC/U";
 285                 else if (i == tx_ring->next_to_use)
 286                         next_desc = " NTU";
 287                 else if (i == tx_ring->next_to_clean)
 288                         next_desc = " NTC";
 289                 else
 290                         next_desc = "";
 291                 pr_info("T%c[0x%03X]    %016llX %016llX %016llX %04X  %3X %016llX %p%s\n",
 292                         (!(le64_to_cpu(u0->b) & (1 << 29)) ? 'l' :
 293                          ((le64_to_cpu(u0->b) & (1 << 20)) ? 'd' : 'c')),
 294                         i,
 295                         (unsigned long long)le64_to_cpu(u0->a),
 296                         (unsigned long long)le64_to_cpu(u0->b),
 297                         (unsigned long long)buffer_info->dma,
 298                         buffer_info->length, buffer_info->next_to_watch,
 299                         (unsigned long long)buffer_info->time_stamp,
 300                         buffer_info->skb, next_desc);
 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         pr_info("Queue [NTU] [NTC]\n");
 312         pr_info(" %5d %5X %5X\n",
 313                 0, rx_ring->next_to_use, rx_ring->next_to_clean);
 314
 315         /* Print Rx Ring */
 316         if (!netif_msg_rx_status(adapter))
 317                 return;
 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                 pr_info("R  [desc]      [buffer 0 63:0 ] [buffer 1 63:0 ] [buffer 2 63:0 ] [buffer 3 63:0 ] [bi->dma       ] [bi->skb] <-- Ext Pkt Split format\n");
 337                 /* [Extended] Receive Descriptor (Write-Back) Format
 338                  *
 339                  *   63       48 47    32 31     13 12    8 7    4 3        0
 340                  *   +------------------------------------------------------+
 341                  * 0 | Packet   | IP     |  Rsvd   | MRQ   | Rsvd | MRQ RSS |
 342                  *   | Checksum | Ident  |         | Queue |      |  Type   |
 343                  *   +------------------------------------------------------+
 344                  * 8 | VLAN Tag | Length | Extended Error | Extended Status |
 345                  *   +------------------------------------------------------+
 346                  *   63       48 47    32 31            20 19               0
 347                  */
 348                 pr_info("RWB[desc]      [ck ipid mrqhsh] [vl   l0 ee  es] [ l3  l2  l1 hs] [reserved      ] ---------------- [bi->skb] <-- Ext Rx Write-Back format\n");
 349                 for (i = 0; i < rx_ring->count; i++) {
 350                         const char *next_desc;
 351                         buffer_info = &rx_ring->buffer_info[i];
 352                         rx_desc_ps = E1000_RX_DESC_PS(*rx_ring, i);
 353                         u1 = (struct my_u1 *)rx_desc_ps;
 354                         staterr =
 355                             le32_to_cpu(rx_desc_ps->wb.middle.status_error);
 356
 357                         if (i == rx_ring->next_to_use)
 358                                 next_desc = " NTU";
 359                         else if (i == rx_ring->next_to_clean)
 360                                 next_desc = " NTC";
 361                         else
 362                                 next_desc = "";
 363
 364                         if (staterr & E1000_RXD_STAT_DD) {
 365                                 /* Descriptor Done */
 366                                 pr_info("%s[0x%03X]     %016llX %016llX %016llX %016llX ---------------- %p%s\n",
 367                                         "RWB", i,
 368                                         (unsigned long long)le64_to_cpu(u1->a),
 369                                         (unsigned long long)le64_to_cpu(u1->b),
 370                                         (unsigned long long)le64_to_cpu(u1->c),
 371                                         (unsigned long long)le64_to_cpu(u1->d),
 372                                         buffer_info->skb, next_desc);
 373                         } else {
 374                                 pr_info("%s[0x%03X]     %016llX %016llX %016llX %016llX %016llX %p%s\n",
 375                                         "R  ", i,
 376                                         (unsigned long long)le64_to_cpu(u1->a),
 377                                         (unsigned long long)le64_to_cpu(u1->b),
 378                                         (unsigned long long)le64_to_cpu(u1->c),
 379                                         (unsigned long long)le64_to_cpu(u1->d),
 380                                         (unsigned long long)buffer_info->dma,
 381                                         buffer_info->skb, next_desc);
 382
 383                                 if (netif_msg_pktdata(adapter))
 384                                         print_hex_dump(KERN_INFO, "",
 385                                                 DUMP_PREFIX_ADDRESS, 16, 1,
 386                                                 phys_to_virt(buffer_info->dma),
 387                                                 adapter->rx_ps_bsize0, true);
 388                         }
 389                 }
 390                 break;
 391         default:
 392         case 0:
 393                 /* Extended Receive Descriptor (Read) Format
 394                  *
 395                  *   +-----------------------------------------------------+
 396                  * 0 |                Buffer Address [63:0]                |
 397                  *   +-----------------------------------------------------+
 398                  * 8 |                      Reserved                       |
 399                  *   +-----------------------------------------------------+
 400                  */
 401                 pr_info("R  [desc]      [buf addr 63:0 ] [reserved 63:0 ] [bi->dma       ] [bi->skb] <-- Ext (Read) format\n");
 402                 /* Extended Receive Descriptor (Write-Back) Format
 403                  *
 404                  *   63       48 47    32 31    24 23            4 3        0
 405                  *   +------------------------------------------------------+
 406                  *   |     RSS Hash      |        |               |         |
 407                  * 0 +-------------------+  Rsvd  |   Reserved    | MRQ RSS |
 408                  *   | Packet   | IP     |        |               |  Type   |
 409                  *   | Checksum | Ident  |        |               |         |
 410                  *   +------------------------------------------------------+
 411                  * 8 | VLAN Tag | Length | Extended Error | Extended Status |
 412                  *   +------------------------------------------------------+
 413                  *   63       48 47    32 31            20 19               0
 414                  */
 415                 pr_info("RWB[desc]      [cs ipid    mrq] [vt   ln xe  xs] [bi->skb] <-- Ext (Write-Back) format\n");
 416
 417                 for (i = 0; i < rx_ring->count; i++) {
 418                         const char *next_desc;
 419
 420                         buffer_info = &rx_ring->buffer_info[i];
 421                         rx_desc = E1000_RX_DESC_EXT(*rx_ring, i);
 422                         u1 = (struct my_u1 *)rx_desc;
 423                         staterr = le32_to_cpu(rx_desc->wb.upper.status_error);
 424
 425                         if (i == rx_ring->next_to_use)
 426                                 next_desc = " NTU";
 427                         else if (i == rx_ring->next_to_clean)
 428                                 next_desc = " NTC";
 429                         else
 430                                 next_desc = "";
 431
 432                         if (staterr & E1000_RXD_STAT_DD) {
 433                                 /* Descriptor Done */
 434                                 pr_info("%s[0x%03X]     %016llX %016llX ---------------- %p%s\n",
 435                                         "RWB", i,
 436                                         (unsigned long long)le64_to_cpu(u1->a),
 437                                         (unsigned long long)le64_to_cpu(u1->b),
 438                                         buffer_info->skb, next_desc);
 439                         } else {
 440                                 pr_info("%s[0x%03X]     %016llX %016llX %016llX %p%s\n",
 441                                         "R  ", i,
 442                                         (unsigned long long)le64_to_cpu(u1->a),
 443                                         (unsigned long long)le64_to_cpu(u1->b),
 444                                         (unsigned long long)buffer_info->dma,
 445                                         buffer_info->skb, next_desc);
 446
 447                                 if (netif_msg_pktdata(adapter))
 448                                         print_hex_dump(KERN_INFO, "",
 449                                                        DUMP_PREFIX_ADDRESS, 16,
 450                                                        1,
 451                                                        phys_to_virt
 452                                                        (buffer_info->dma),
 453                                                        adapter->rx_buffer_len,
 454                                                        true);
 455                         }
 456                 }
 457         }
 458 }
 459
 460 /**
 461  * e1000_desc_unused - calculate if we have unused descriptors
 462  **/
 463 static int e1000_desc_unused(struct e1000_ring *ring)
 464 {
 465         if (ring->next_to_clean > ring->next_to_use)
 466                 return ring->next_to_clean - ring->next_to_use - 1;
 467
 468         return ring->count + ring->next_to_clean - ring->next_to_use - 1;
 469 }
 470
 471 /**
 472  * e1000_receive_skb - helper function to handle Rx indications
 473  * @adapter: board private structure
 474  * @status: descriptor status field as written by hardware
 475  * @vlan: descriptor vlan field as written by hardware (no le/be conversion)
 476  * @skb: pointer to sk_buff to be indicated to stack
 477  **/
 478 static void e1000_receive_skb(struct e1000_adapter *adapter,
 479                               struct net_device *netdev, struct sk_buff *skb,
 480                               u8 status, __le16 vlan)
 481 {
 482         u16 tag = le16_to_cpu(vlan);
 483         skb->protocol = eth_type_trans(skb, netdev);
 484
 485         if (status & E1000_RXD_STAT_VP)
 486                 __vlan_hwaccel_put_tag(skb, tag);
 487
 488         napi_gro_receive(&adapter->napi, skb);
 489 }
 490
 491 /**
 492  * e1000_rx_checksum - Receive Checksum Offload
 493  * @adapter: board private structure
 494  * @status_err: receive descriptor status and error fields
 495  * @csum: receive descriptor csum field
 496  * @sk_buff: socket buffer with received data
 497  **/
 498 static void e1000_rx_checksum(struct e1000_adapter *adapter, u32 status_err,
 499                               __le16 csum, struct sk_buff *skb)
 500 {
 501         u16 status = (u16)status_err;
 502         u8 errors = (u8)(status_err >> 24);
 503
 504         skb_checksum_none_assert(skb);
 505
 506         /* Rx checksum disabled */
 507         if (!(adapter->netdev->features & NETIF_F_RXCSUM))
 508                 return;
 509
 510         /* Ignore Checksum bit is set */
 511         if (status & E1000_RXD_STAT_IXSM)
 512                 return;
 513
 514         /* TCP/UDP checksum error bit is set */
 515         if (errors & E1000_RXD_ERR_TCPE) {
 516                 /* let the stack verify checksum errors */
 517                 adapter->hw_csum_err++;
 518                 return;
 519         }
 520
 521         /* TCP/UDP Checksum has not been calculated */
 522         if (!(status & (E1000_RXD_STAT_TCPCS | E1000_RXD_STAT_UDPCS)))
 523                 return;
 524
 525         /* It must be a TCP or UDP packet with a valid checksum */
 526         if (status & E1000_RXD_STAT_TCPCS) {
 527                 /* TCP checksum is good */
 528                 skb->ip_summed = CHECKSUM_UNNECESSARY;
 529         } else {
 530                 /*
 531                  * IP fragment with UDP payload
 532                  * Hardware complements the payload checksum, so we undo it
 533                  * and then put the value in host order for further stack use.
 534                  */
 535                 __sum16 sum = (__force __sum16)swab16((__force u16)csum);
 536                 skb->csum = csum_unfold(~sum);
 537                 skb->ip_summed = CHECKSUM_COMPLETE;
 538         }
 539         adapter->hw_csum_good++;
 540 }
 541
 542 static void e1000e_update_rdt_wa(struct e1000_ring *rx_ring, unsigned int i)
 543 {
 544         struct e1000_adapter *adapter = rx_ring->adapter;
 545         struct e1000_hw *hw = &adapter->hw;
 546         s32 ret_val = __ew32_prepare(hw);
 547
 548         writel(i, rx_ring->tail);
 549
 550         if (unlikely(!ret_val && (i != readl(rx_ring->tail)))) {
 551                 u32 rctl = er32(RCTL);
 552                 ew32(RCTL, rctl & ~E1000_RCTL_EN);
 553                 e_err("ME firmware caused invalid RDT - resetting\n");
 554                 schedule_work(&adapter->reset_task);
 555         }
 556 }
 557
 558 static void e1000e_update_tdt_wa(struct e1000_ring *tx_ring, unsigned int i)
 559 {
 560         struct e1000_adapter *adapter = tx_ring->adapter;
 561         struct e1000_hw *hw = &adapter->hw;
 562         s32 ret_val = __ew32_prepare(hw);
 563
 564         writel(i, tx_ring->tail);
 565
 566         if (unlikely(!ret_val && (i != readl(tx_ring->tail)))) {
 567                 u32 tctl = er32(TCTL);
 568                 ew32(TCTL, tctl & ~E1000_TCTL_EN);
 569                 e_err("ME firmware caused invalid TDT - resetting\n");
 570                 schedule_work(&adapter->reset_task);
 571         }
 572 }
 573
 574 /**
 575  * e1000_alloc_rx_buffers - Replace used receive buffers
 576  * @rx_ring: Rx descriptor ring
 577  **/
 578 static void e1000_alloc_rx_buffers(struct e1000_ring *rx_ring,
 579                                    int cleaned_count, gfp_t gfp)
 580 {
 581         struct e1000_adapter *adapter = rx_ring->adapter;
 582         struct net_device *netdev = adapter->netdev;
 583         struct pci_dev *pdev = adapter->pdev;
 584         union e1000_rx_desc_extended *rx_desc;
 585         struct e1000_buffer *buffer_info;
 586         struct sk_buff *skb;
 587         unsigned int i;
 588         unsigned int bufsz = adapter->rx_buffer_len;
 589
 590         i = rx_ring->next_to_use;
 591         buffer_info = &rx_ring->buffer_info[i];
 592
 593         while (cleaned_count--) {
 594                 skb = buffer_info->skb;
 595                 if (skb) {
 596                         skb_trim(skb, 0);
 597                         goto map_skb;
 598                 }
 599
 600                 skb = __netdev_alloc_skb_ip_align(netdev, bufsz, gfp);
 601                 if (!skb) {
 602                         /* Better luck next round */
 603                         adapter->alloc_rx_buff_failed++;
 604                         break;
 605                 }
 606
 607                 buffer_info->skb = skb;
 608 map_skb:
 609                 buffer_info->dma = dma_map_single(&pdev->dev, skb->data,
 610                                                   adapter->rx_buffer_len,
 611                                                   DMA_FROM_DEVICE);
 612                 if (dma_mapping_error(&pdev->dev, buffer_info->dma)) {
 613                         dev_err(&pdev->dev, "Rx DMA map failed\n");
 614                         adapter->rx_dma_failed++;
 615                         break;
 616                 }
 617
 618                 rx_desc = E1000_RX_DESC_EXT(*rx_ring, i);
 619                 rx_desc->read.buffer_addr = cpu_to_le64(buffer_info->dma);
 620
 621                 if (unlikely(!(i & (E1000_RX_BUFFER_WRITE - 1)))) {
 622                         /*
 623                          * Force memory writes to complete before letting h/w
 624                          * know there are new descriptors to fetch.  (Only
 625                          * applicable for weak-ordered memory model archs,
 626                          * such as IA-64).
 627                          */
 628                         wmb();
 629                         if (adapter->flags2 & FLAG2_PCIM2PCI_ARBITER_WA)
 630                                 e1000e_update_rdt_wa(rx_ring, i);
 631                         else
 632                                 writel(i, rx_ring->tail);
 633                 }
 634                 i++;
 635                 if (i == rx_ring->count)
 636                         i = 0;
 637                 buffer_info = &rx_ring->buffer_info[i];
 638         }
 639
 640         rx_ring->next_to_use = i;
 641 }
 642
 643 /**
 644  * e1000_alloc_rx_buffers_ps - Replace used receive buffers; packet split
 645  * @rx_ring: Rx descriptor ring
 646  **/
 647 static void e1000_alloc_rx_buffers_ps(struct e1000_ring *rx_ring,
 648                                       int cleaned_count, gfp_t gfp)
 649 {
 650         struct e1000_adapter *adapter = rx_ring->adapter;
 651         struct net_device *netdev = adapter->netdev;
 652         struct pci_dev *pdev = adapter->pdev;
 653         union e1000_rx_desc_packet_split *rx_desc;
 654         struct e1000_buffer *buffer_info;
 655         struct e1000_ps_page *ps_page;
 656         struct sk_buff *skb;
 657         unsigned int i, j;
 658
 659         i = rx_ring->next_to_use;
 660         buffer_info = &rx_ring->buffer_info[i];
 661
 662         while (cleaned_count--) {
 663                 rx_desc = E1000_RX_DESC_PS(*rx_ring, i);
 664
 665                 for (j = 0; j < PS_PAGE_BUFFERS; j++) {
 666                         ps_page = &buffer_info->ps_pages[j];
 667                         if (j >= adapter->rx_ps_pages) {
 668                                 /* all unused desc entries get hw null ptr */
 669                                 rx_desc->read.buffer_addr[j + 1] =
 670                                     ~cpu_to_le64(0);
 671                                 continue;
 672                         }
 673                         if (!ps_page->page) {
 674                                 ps_page->page = alloc_page(gfp);
 675                                 if (!ps_page->page) {
 676                                         adapter->alloc_rx_buff_failed++;
 677                                         goto no_buffers;
 678                                 }
 679                                 ps_page->dma = dma_map_page(&pdev->dev,
 680                                                             ps_page->page,
 681                                                             0, PAGE_SIZE,
 682                                                             DMA_FROM_DEVICE);
 683                                 if (dma_mapping_error(&pdev->dev,
 684                                                       ps_page->dma)) {
 685                                         dev_err(&adapter->pdev->dev,
 686                                                 "Rx DMA page map failed\n");
 687                                         adapter->rx_dma_failed++;
 688                                         goto no_buffers;
 689                                 }
 690                         }
 691                         /*
 692                          * Refresh the desc even if buffer_addrs
 693                          * didn't change because each write-back
 694                          * erases this info.
 695                          */
 696                         rx_desc->read.buffer_addr[j + 1] =
 697                             cpu_to_le64(ps_page->dma);
 698                 }
 699
 700                 skb = __netdev_alloc_skb_ip_align(netdev,
 701                                                   adapter->rx_ps_bsize0,
 702                                                   gfp);
 703
 704                 if (!skb) {
 705                         adapter->alloc_rx_buff_failed++;
 706                         break;
 707                 }
 708
 709                 buffer_info->skb = skb;
 710                 buffer_info->dma = dma_map_single(&pdev->dev, skb->data,
 711                                                   adapter->rx_ps_bsize0,
 712                                                   DMA_FROM_DEVICE);
 713                 if (dma_mapping_error(&pdev->dev, buffer_info->dma)) {
 714                         dev_err(&pdev->dev, "Rx DMA map failed\n");
 715                         adapter->rx_dma_failed++;
 716                         /* cleanup skb */
 717                         dev_kfree_skb_any(skb);
 718                         buffer_info->skb = NULL;
 719                         break;
 720                 }
 721
 722                 rx_desc->read.buffer_addr[0] = cpu_to_le64(buffer_info->dma);
 723
 724                 if (unlikely(!(i & (E1000_RX_BUFFER_WRITE - 1)))) {
 725                         /*
 726                          * Force memory writes to complete before letting h/w
 727                          * know there are new descriptors to fetch.  (Only
 728                          * applicable for weak-ordered memory model archs,
 729                          * such as IA-64).
 730                          */
 731                         wmb();
 732                         if (adapter->flags2 & FLAG2_PCIM2PCI_ARBITER_WA)
 733                                 e1000e_update_rdt_wa(rx_ring, i << 1);
 734                         else
 735                                 writel(i << 1, rx_ring->tail);
 736                 }
 737
 738                 i++;
 739                 if (i == rx_ring->count)
 740                         i = 0;
 741                 buffer_info = &rx_ring->buffer_info[i];
 742         }
 743
 744 no_buffers:
 745         rx_ring->next_to_use = i;
 746 }
 747
 748 /**
 749  * e1000_alloc_jumbo_rx_buffers - Replace used jumbo receive buffers
 750  * @rx_ring: Rx descriptor ring
 751  * @cleaned_count: number of buffers to allocate this pass
 752  **/
 753
 754 static void e1000_alloc_jumbo_rx_buffers(struct e1000_ring *rx_ring,
 755                                          int cleaned_count, gfp_t gfp)
 756 {
 757         struct e1000_adapter *adapter = rx_ring->adapter;
 758         struct net_device *netdev = adapter->netdev;
 759         struct pci_dev *pdev = adapter->pdev;
 760         union e1000_rx_desc_extended *rx_desc;
 761         struct e1000_buffer *buffer_info;
 762         struct sk_buff *skb;
 763         unsigned int i;
 764         unsigned int bufsz = 256 - 16 /* for skb_reserve */;
 765
 766         i = rx_ring->next_to_use;
 767         buffer_info = &rx_ring->buffer_info[i];
 768
 769         while (cleaned_count--) {
 770                 skb = buffer_info->skb;
 771                 if (skb) {
 772                         skb_trim(skb, 0);
 773                         goto check_page;
 774                 }
 775
 776                 skb = __netdev_alloc_skb_ip_align(netdev, bufsz, gfp);
 777                 if (unlikely(!skb)) {
 778                         /* Better luck next round */
 779                         adapter->alloc_rx_buff_failed++;
 780                         break;
 781                 }
 782
 783                 buffer_info->skb = skb;
 784 check_page:
 785                 /* allocate a new page if necessary */
 786                 if (!buffer_info->page) {
 787                         buffer_info->page = alloc_page(gfp);
 788                         if (unlikely(!buffer_info->page)) {
 789                                 adapter->alloc_rx_buff_failed++;
 790                                 break;
 791                         }
 792                 }
 793
 794                 if (!buffer_info->dma)
 795                         buffer_info->dma = dma_map_page(&pdev->dev,
 796                                                         buffer_info->page, 0,
 797                                                         PAGE_SIZE,
 798                                                         DMA_FROM_DEVICE);
 799
 800                 rx_desc = E1000_RX_DESC_EXT(*rx_ring, i);
 801                 rx_desc->read.buffer_addr = cpu_to_le64(buffer_info->dma);
 802
 803                 if (unlikely(++i == rx_ring->count))
 804                         i = 0;
 805                 buffer_info = &rx_ring->buffer_info[i];
 806         }
 807
 808         if (likely(rx_ring->next_to_use != i)) {
 809                 rx_ring->next_to_use = i;
 810                 if (unlikely(i-- == 0))
 811                         i = (rx_ring->count - 1);
 812
 813                 /* Force memory writes to complete before letting h/w
 814                  * know there are new descriptors to fetch.  (Only
 815                  * applicable for weak-ordered memory model archs,
 816                  * such as IA-64). */
 817                 wmb();
 818                 if (adapter->flags2 & FLAG2_PCIM2PCI_ARBITER_WA)
 819                         e1000e_update_rdt_wa(rx_ring, i);
 820                 else
 821                         writel(i, rx_ring->tail);
 822         }
 823 }
 824
 825 static inline void e1000_rx_hash(struct net_device *netdev, __le32 rss,
 826                                  struct sk_buff *skb)
 827 {
 828         if (netdev->features & NETIF_F_RXHASH)
 829                 skb->rxhash = le32_to_cpu(rss);
 830 }
 831
 832 /**
 833  * e1000_clean_rx_irq - Send received data up the network stack
 834  * @rx_ring: Rx descriptor ring
 835  *
 836  * the return value indicates whether actual cleaning was done, there
 837  * is no guarantee that everything was cleaned
 838  **/
 839 static bool e1000_clean_rx_irq(struct e1000_ring *rx_ring, int *work_done,
 840                                int work_to_do)
 841 {
 842         struct e1000_adapter *adapter = rx_ring->adapter;
 843         struct net_device *netdev = adapter->netdev;
 844         struct pci_dev *pdev = adapter->pdev;
 845         struct e1000_hw *hw = &adapter->hw;
 846         union e1000_rx_desc_extended *rx_desc, *next_rxd;
 847         struct e1000_buffer *buffer_info, *next_buffer;
 848         u32 length, staterr;
 849         unsigned int i;
 850         int cleaned_count = 0;
 851         bool cleaned = false;
 852         unsigned int total_rx_bytes = 0, total_rx_packets = 0;
 853
 854         i = rx_ring->next_to_clean;
 855         rx_desc = E1000_RX_DESC_EXT(*rx_ring, i);
 856         staterr = le32_to_cpu(rx_desc->wb.upper.status_error);
 857         buffer_info = &rx_ring->buffer_info[i];
 858
 859         while (staterr & E1000_RXD_STAT_DD) {
 860                 struct sk_buff *skb;
 861
 862                 if (*work_done >= work_to_do)
 863                         break;
 864                 (*work_done)++;
 865                 rmb();  /* read descriptor and rx_buffer_info after status DD */
 866
 867                 skb = buffer_info->skb;
 868                 buffer_info->skb = NULL;
 869
 870                 prefetch(skb->data - NET_IP_ALIGN);
 871
 872                 i++;
 873                 if (i == rx_ring->count)
 874                         i = 0;
 875                 next_rxd = E1000_RX_DESC_EXT(*rx_ring, i);
 876                 prefetch(next_rxd);
 877
 878                 next_buffer = &rx_ring->buffer_info[i];
 879
 880                 cleaned = true;
 881                 cleaned_count++;
 882                 dma_unmap_single(&pdev->dev,
 883                                  buffer_info->dma,
 884                                  adapter->rx_buffer_len,
 885                                  DMA_FROM_DEVICE);
 886                 buffer_info->dma = 0;
 887
 888                 length = le16_to_cpu(rx_desc->wb.upper.length);
 889
 890                 /*
 891                  * !EOP means multiple descriptors were used to store a single
 892                  * packet, if that's the case we need to toss it.  In fact, we
 893                  * need to toss every packet with the EOP bit clear and the
 894                  * next frame that _does_ have the EOP bit set, as it is by
 895                  * definition only a frame fragment
 896                  */
 897                 if (unlikely(!(staterr & E1000_RXD_STAT_EOP)))
 898                         adapter->flags2 |= FLAG2_IS_DISCARDING;
 899
 900                 if (adapter->flags2 & FLAG2_IS_DISCARDING) {
 901                         /* All receives must fit into a single buffer */
 902                         e_dbg("Receive packet consumed multiple buffers\n");
 903                         /* recycle */
 904                         buffer_info->skb = skb;
 905                         if (staterr & E1000_RXD_STAT_EOP)
 906                                 adapter->flags2 &= ~FLAG2_IS_DISCARDING;
 907                         goto next_desc;
 908                 }
 909
 910                 if (unlikely((staterr & E1000_RXDEXT_ERR_FRAME_ERR_MASK) &&
 911                              !(netdev->features & NETIF_F_RXALL))) {
 912                         /* recycle */
 913                         buffer_info->skb = skb;
 914                         goto next_desc;
 915                 }
 916
 917                 /* adjust length to remove Ethernet CRC */
 918                 if (!(adapter->flags2 & FLAG2_CRC_STRIPPING)) {
 919                         /* If configured to store CRC, don't subtract FCS,
 920                          * but keep the FCS bytes out of the total_rx_bytes
 921                          * counter
 922                          */
 923                         if (netdev->features & NETIF_F_RXFCS)
 924                                 total_rx_bytes -= 4;
 925                         else
 926                                 length -= 4;
 927                 }
 928
 929                 total_rx_bytes += length;
 930                 total_rx_packets++;
 931
 932                 /*
 933                  * code added for copybreak, this should improve
 934                  * performance for small packets with large amounts
 935                  * of reassembly being done in the stack
 936                  */
 937                 if (length < copybreak) {
 938                         struct sk_buff *new_skb =
 939                             netdev_alloc_skb_ip_align(netdev, length);
 940                         if (new_skb) {
 941                                 skb_copy_to_linear_data_offset(new_skb,
 942                                                                -NET_IP_ALIGN,
 943                                                                (skb->data -
 944                                                                 NET_IP_ALIGN),
 945                                                                (length +
 946                                                                 NET_IP_ALIGN));
 947                                 /* save the skb in buffer_info as good */
 948                                 buffer_info->skb = skb;
 949                                 skb = new_skb;
 950                         }
 951                         /* else just continue with the old one */
 952                 }
 953                 /* end copybreak code */
 954                 skb_put(skb, length);
 955
 956                 /* Receive Checksum Offload */
 957                 e1000_rx_checksum(adapter, staterr,
 958                                   rx_desc->wb.lower.hi_dword.csum_ip.csum, skb);
 959
 960                 e1000_rx_hash(netdev, rx_desc->wb.lower.hi_dword.rss, skb);
 961
 962                 e1000_receive_skb(adapter, netdev, skb, staterr,
 963                                   rx_desc->wb.upper.vlan);
 964
 965 next_desc:
 966                 rx_desc->wb.upper.status_error &= cpu_to_le32(~0xFF);
 967
 968                 /* return some buffers to hardware, one at a time is too slow */
 969                 if (cleaned_count >= E1000_RX_BUFFER_WRITE) {
 970                         adapter->alloc_rx_buf(rx_ring, cleaned_count,
 971                                               GFP_ATOMIC);
 972                         cleaned_count = 0;
 973                 }
 974
 975                 /* use prefetched values */
 976                 rx_desc = next_rxd;
 977                 buffer_info = next_buffer;
 978
 979                 staterr = le32_to_cpu(rx_desc->wb.upper.status_error);
 980         }
 981         rx_ring->next_to_clean = i;
 982
 983         cleaned_count = e1000_desc_unused(rx_ring);
 984         if (cleaned_count)
 985                 adapter->alloc_rx_buf(rx_ring, cleaned_count, GFP_ATOMIC);
 986
 987         adapter->total_rx_bytes += total_rx_bytes;
 988         adapter->total_rx_packets += total_rx_packets;
 989         return cleaned;
 990 }
 991
 992 static void e1000_put_txbuf(struct e1000_ring *tx_ring,
 993                             struct e1000_buffer *buffer_info)
 994 {
 995         struct e1000_adapter *adapter = tx_ring->adapter;
 996
 997         if (buffer_info->dma) {
 998                 if (buffer_info->mapped_as_page)
 999                         dma_unmap_page(&adapter->pdev->dev, buffer_info->dma,
1000                                        buffer_info->length, DMA_TO_DEVICE);
1001                 else
1002                         dma_unmap_single(&adapter->pdev->dev, buffer_info->dma,
1003                                          buffer_info->length, DMA_TO_DEVICE);
1004                 buffer_info->dma = 0;
1005         }
1006         if (buffer_info->skb) {
1007                 dev_kfree_skb_any(buffer_info->skb);
1008                 buffer_info->skb = NULL;
1009         }
1010         buffer_info->time_stamp = 0;
1011 }
1012
1013 static void e1000_print_hw_hang(struct work_struct *work)
1014 {
1015         struct e1000_adapter *adapter = container_of(work,
1016                                                      struct e1000_adapter,
1017                                                      print_hang_task);
1018         struct net_device *netdev = adapter->netdev;
1019         struct e1000_ring *tx_ring = adapter->tx_ring;
1020         unsigned int i = tx_ring->next_to_clean;
1021         unsigned int eop = tx_ring->buffer_info[i].next_to_watch;
1022         struct e1000_tx_desc *eop_desc = E1000_TX_DESC(*tx_ring, eop);
1023         struct e1000_hw *hw = &adapter->hw;
1024         u16 phy_status, phy_1000t_status, phy_ext_status;
1025         u16 pci_status;
1026
1027         if (test_bit(__E1000_DOWN, &adapter->state))
1028                 return;
1029
1030         if (!adapter->tx_hang_recheck &&
1031             (adapter->flags2 & FLAG2_DMA_BURST)) {
1032                 /*
1033                  * May be block on write-back, flush and detect again
1034                  * flush pending descriptor writebacks to memory
1035                  */
1036                 ew32(TIDV, adapter->tx_int_delay | E1000_TIDV_FPD);
1037                 /* execute the writes immediately */
1038                 e1e_flush();
1039                 /*
1040                  * Due to rare timing issues, write to TIDV again to ensure
1041                  * the write is successful
1042                  */
1043                 ew32(TIDV, adapter->tx_int_delay | E1000_TIDV_FPD);
1044                 /* execute the writes immediately */
1045                 e1e_flush();
1046                 adapter->tx_hang_recheck = true;
1047                 return;
1048         }
1049         /* Real hang detected */
1050         adapter->tx_hang_recheck = false;
1051         netif_stop_queue(netdev);
1052
1053         e1e_rphy(hw, PHY_STATUS, &phy_status);
1054         e1e_rphy(hw, PHY_1000T_STATUS, &phy_1000t_status);
1055         e1e_rphy(hw, PHY_EXT_STATUS, &phy_ext_status);
1056
1057         pci_read_config_word(adapter->pdev, PCI_STATUS, &pci_status);
1058
1059         /* detected Hardware unit hang */
1060         e_err("Detected Hardware Unit Hang:\n"
1061               "  TDH                  <%x>\n"
1062               "  TDT                  <%x>\n"
1063               "  next_to_use          <%x>\n"
1064               "  next_to_clean        <%x>\n"
1065               "buffer_info[next_to_clean]:\n"
1066               "  time_stamp           <%lx>\n"
1067               "  next_to_watch        <%x>\n"
1068               "  jiffies              <%lx>\n"
1069               "  next_to_watch.status <%x>\n"
1070               "MAC Status             <%x>\n"
1071               "PHY Status             <%x>\n"
1072               "PHY 1000BASE-T Status  <%x>\n"
1073               "PHY Extended Status    <%x>\n"
1074               "PCI Status             <%x>\n",
1075               readl(tx_ring->head),
1076               readl(tx_ring->tail),
1077               tx_ring->next_to_use,
1078               tx_ring->next_to_clean,
1079               tx_ring->buffer_info[eop].time_stamp,
1080               eop,
1081               jiffies,
1082               eop_desc->upper.fields.status,
1083               er32(STATUS),
1084               phy_status,
1085               phy_1000t_status,
1086               phy_ext_status,
1087               pci_status);
1088
1089         /* Suggest workaround for known h/w issue */
1090         if ((hw->mac.type == e1000_pchlan) && (er32(CTRL) & E1000_CTRL_TFCE))
1091                 e_err("Try turning off Tx pause (flow control) via ethtool\n");
1092 }
1093
1094 /**
1095  * e1000_clean_tx_irq - Reclaim resources after transmit completes
1096  * @tx_ring: Tx descriptor ring
1097  *
1098  * the return value indicates whether actual cleaning was done, there
1099  * is no guarantee that everything was cleaned
1100  **/
1101 static bool e1000_clean_tx_irq(struct e1000_ring *tx_ring)
1102 {
1103         struct e1000_adapter *adapter = tx_ring->adapter;
1104         struct net_device *netdev = adapter->netdev;
1105         struct e1000_hw *hw = &adapter->hw;
1106         struct e1000_tx_desc *tx_desc, *eop_desc;
1107         struct e1000_buffer *buffer_info;
1108         unsigned int i, eop;
1109         unsigned int count = 0;
1110         unsigned int total_tx_bytes = 0, total_tx_packets = 0;
1111         unsigned int bytes_compl = 0, pkts_compl = 0;
1112
1113         i = tx_ring->next_to_clean;
1114         eop = tx_ring->buffer_info[i].next_to_watch;
1115         eop_desc = E1000_TX_DESC(*tx_ring, eop);
1116
1117         while ((eop_desc->upper.data & cpu_to_le32(E1000_TXD_STAT_DD)) &&
1118                (count < tx_ring->count)) {
1119                 bool cleaned = false;
1120                 rmb(); /* read buffer_info after eop_desc */
1121                 for (; !cleaned; count++) {
1122                         tx_desc = E1000_TX_DESC(*tx_ring, i);
1123                         buffer_info = &tx_ring->buffer_info[i];
1124                         cleaned = (i == eop);
1125
1126                         if (cleaned) {
1127                                 total_tx_packets += buffer_info->segs;
1128                                 total_tx_bytes += buffer_info->bytecount;
1129                                 if (buffer_info->skb) {
1130                                         bytes_compl += buffer_info->skb->len;
1131                                         pkts_compl++;
1132                                 }
1133                         }
1134
1135                         e1000_put_txbuf(tx_ring, buffer_info);
1136                         tx_desc->upper.data = 0;
1137
1138                         i++;
1139                         if (i == tx_ring->count)
1140                                 i = 0;
1141                 }
1142
1143                 if (i == tx_ring->next_to_use)
1144                         break;
1145                 eop = tx_ring->buffer_info[i].next_to_watch;
1146                 eop_desc = E1000_TX_DESC(*tx_ring, eop);
1147         }
1148
1149         tx_ring->next_to_clean = i;
1150
1151         netdev_completed_queue(netdev, pkts_compl, bytes_compl);
1152
1153 #define TX_WAKE_THRESHOLD 32
1154         if (count && netif_carrier_ok(netdev) &&
1155             e1000_desc_unused(tx_ring) >= TX_WAKE_THRESHOLD) {
1156                 /* Make sure that anybody stopping the queue after this
1157                  * sees the new next_to_clean.
1158                  */
1159                 smp_mb();
1160
1161                 if (netif_queue_stopped(netdev) &&
1162                     !(test_bit(__E1000_DOWN, &adapter->state))) {
1163                         netif_wake_queue(netdev);
1164                         ++adapter->restart_queue;
1165                 }
1166         }
1167
1168         if (adapter->detect_tx_hung) {
1169                 /*
1170                  * Detect a transmit hang in hardware, this serializes the
1171                  * check with the clearing of time_stamp and movement of i
1172                  */
1173                 adapter->detect_tx_hung = false;
1174                 if (tx_ring->buffer_info[i].time_stamp &&
1175                     time_after(jiffies, tx_ring->buffer_info[i].time_stamp
1176                                + (adapter->tx_timeout_factor * HZ)) &&
1177                     !(er32(STATUS) & E1000_STATUS_TXOFF))
1178                         schedule_work(&adapter->print_hang_task);
1179                 else
1180                         adapter->tx_hang_recheck = false;
1181         }
1182         adapter->total_tx_bytes += total_tx_bytes;
1183         adapter->total_tx_packets += total_tx_packets;
1184         return count < tx_ring->count;
1185 }
1186
1187 /**
1188  * e1000_clean_rx_irq_ps - Send received data up the network stack; packet split
1189  * @rx_ring: Rx descriptor ring
1190  *
1191  * the return value indicates whether actual cleaning was done, there
1192  * is no guarantee that everything was cleaned
1193  **/
1194 static bool e1000_clean_rx_irq_ps(struct e1000_ring *rx_ring, int *work_done,
1195                                   int work_to_do)
1196 {
1197         struct e1000_adapter *adapter = rx_ring->adapter;
1198         struct e1000_hw *hw = &adapter->hw;
1199         union e1000_rx_desc_packet_split *rx_desc, *next_rxd;
1200         struct net_device *netdev = adapter->netdev;
1201         struct pci_dev *pdev = adapter->pdev;
1202         struct e1000_buffer *buffer_info, *next_buffer;
1203         struct e1000_ps_page *ps_page;
1204         struct sk_buff *skb;
1205         unsigned int i, j;
1206         u32 length, staterr;
1207         int cleaned_count = 0;
1208         bool cleaned = false;
1209         unsigned int total_rx_bytes = 0, total_rx_packets = 0;
1210
1211         i = rx_ring->next_to_clean;
1212         rx_desc = E1000_RX_DESC_PS(*rx_ring, i);
1213         staterr = le32_to_cpu(rx_desc->wb.middle.status_error);
1214         buffer_info = &rx_ring->buffer_info[i];
1215
1216         while (staterr & E1000_RXD_STAT_DD) {
1217                 if (*work_done >= work_to_do)
1218                         break;
1219                 (*work_done)++;
1220                 skb = buffer_info->skb;
1221                 rmb();  /* read descriptor and rx_buffer_info after status DD */
1222
1223                 /* in the packet split case this is header only */
1224                 prefetch(skb->data - NET_IP_ALIGN);
1225
1226                 i++;
1227                 if (i == rx_ring->count)
1228                         i = 0;
1229                 next_rxd = E1000_RX_DESC_PS(*rx_ring, i);
1230                 prefetch(next_rxd);
1231
1232                 next_buffer = &rx_ring->buffer_info[i];
1233
1234                 cleaned = true;
1235                 cleaned_count++;
1236                 dma_unmap_single(&pdev->dev, buffer_info->dma,
1237                                  adapter->rx_ps_bsize0, DMA_FROM_DEVICE);
1238                 buffer_info->dma = 0;
1239
1240                 /* see !EOP comment in other Rx routine */
1241                 if (!(staterr & E1000_RXD_STAT_EOP))
1242                         adapter->flags2 |= FLAG2_IS_DISCARDING;
1243
1244                 if (adapter->flags2 & FLAG2_IS_DISCARDING) {
1245                         e_dbg("Packet Split buffers didn't pick up the full packet\n");
1246                         dev_kfree_skb_irq(skb);
1247                         if (staterr & E1000_RXD_STAT_EOP)
1248                                 adapter->flags2 &= ~FLAG2_IS_DISCARDING;
1249                         goto next_desc;
1250                 }
1251
1252                 if (unlikely((staterr & E1000_RXDEXT_ERR_FRAME_ERR_MASK) &&
1253                              !(netdev->features & NETIF_F_RXALL))) {
1254                         dev_kfree_skb_irq(skb);
1255                         goto next_desc;
1256                 }
1257
1258                 length = le16_to_cpu(rx_desc->wb.middle.length0);
1259
1260                 if (!length) {
1261                         e_dbg("Last part of the packet spanning multiple descriptors\n");
1262                         dev_kfree_skb_irq(skb);
1263                         goto next_desc;
1264                 }
1265
1266                 /* Good Receive */
1267                 skb_put(skb, length);
1268
1269                 {
1270                         /*
1271                          * this looks ugly, but it seems compiler issues make
1272                          * it more efficient than reusing j
1273                          */
1274                         int l1 = le16_to_cpu(rx_desc->wb.upper.length[0]);
1275
1276                         /*
1277                          * page alloc/put takes too long and effects small
1278                          * packet throughput, so unsplit small packets and
1279                          * save the alloc/put only valid in softirq (napi)
1280                          * context to call kmap_*
1281                          */
1282                         if (l1 && (l1 <= copybreak) &&
1283                             ((length + l1) <= adapter->rx_ps_bsize0)) {
1284                                 u8 *vaddr;
1285
1286                                 ps_page = &buffer_info->ps_pages[0];
1287
1288                                 /*
1289                                  * there is no documentation about how to call
1290                                  * kmap_atomic, so we can't hold the mapping
1291                                  * very long
1292                                  */
1293                                 dma_sync_single_for_cpu(&pdev->dev,
1294                                                         ps_page->dma,
1295                                                         PAGE_SIZE,
1296                                                         DMA_FROM_DEVICE);
1297                                 vaddr = kmap_atomic(ps_page->page);
1298                                 memcpy(skb_tail_pointer(skb), vaddr, l1);
1299                                 kunmap_atomic(vaddr);
1300                                 dma_sync_single_for_device(&pdev->dev,
1301                                                            ps_page->dma,
1302                                                            PAGE_SIZE,
1303                                                            DMA_FROM_DEVICE);
1304
1305                                 /* remove the CRC */
1306                                 if (!(adapter->flags2 & FLAG2_CRC_STRIPPING)) {
1307                                         if (!(netdev->features & NETIF_F_RXFCS))
1308                                                 l1 -= 4;
1309                                 }
1310
1311                                 skb_put(skb, l1);
1312                                 goto copydone;
1313                         } /* if */
1314                 }
1315
1316                 for (j = 0; j < PS_PAGE_BUFFERS; j++) {
1317                         length = le16_to_cpu(rx_desc->wb.upper.length[j]);
1318                         if (!length)
1319                                 break;
1320
1321                         ps_page = &buffer_info->ps_pages[j];
1322                         dma_unmap_page(&pdev->dev, ps_page->dma, PAGE_SIZE,
1323                                        DMA_FROM_DEVICE);
1324                         ps_page->dma = 0;
1325                         skb_fill_page_desc(skb, j, ps_page->page, 0, length);
1326                         ps_page->page = NULL;
1327                         skb->len += length;
1328                         skb->data_len += length;
1329                         skb->truesize += PAGE_SIZE;
1330                 }
1331
1332                 /* strip the ethernet crc, problem is we're using pages now so
1333                  * this whole operation can get a little cpu intensive
1334                  */
1335                 if (!(adapter->flags2 & FLAG2_CRC_STRIPPING)) {
1336                         if (!(netdev->features & NETIF_F_RXFCS))
1337                                 pskb_trim(skb, skb->len - 4);
1338                 }
1339
1340 copydone:
1341                 total_rx_bytes += skb->len;
1342                 total_rx_packets++;
1343
1344                 e1000_rx_checksum(adapter, staterr,
1345                                   rx_desc->wb.lower.hi_dword.csum_ip.csum, skb);
1346
1347                 e1000_rx_hash(netdev, rx_desc->wb.lower.hi_dword.rss, skb);
1348
1349                 if (rx_desc->wb.upper.header_status &
1350                            cpu_to_le16(E1000_RXDPS_HDRSTAT_HDRSP))
1351                         adapter->rx_hdr_split++;
1352
1353                 e1000_receive_skb(adapter, netdev, skb,
1354                                   staterr, rx_desc->wb.middle.vlan);
1355
1356 next_desc:
1357                 rx_desc->wb.middle.status_error &= cpu_to_le32(~0xFF);
1358                 buffer_info->skb = NULL;
1359
1360                 /* return some buffers to hardware, one at a time is too slow */
1361                 if (cleaned_count >= E1000_RX_BUFFER_WRITE) {
1362                         adapter->alloc_rx_buf(rx_ring, cleaned_count,
1363                                               GFP_ATOMIC);
1364                         cleaned_count = 0;
1365                 }
1366
1367                 /* use prefetched values */
1368                 rx_desc = next_rxd;
1369                 buffer_info = next_buffer;
1370
1371                 staterr = le32_to_cpu(rx_desc->wb.middle.status_error);
1372         }
1373         rx_ring->next_to_clean = i;
1374
1375         cleaned_count = e1000_desc_unused(rx_ring);
1376         if (cleaned_count)
1377                 adapter->alloc_rx_buf(rx_ring, cleaned_count, GFP_ATOMIC);
1378
1379         adapter->total_rx_bytes += total_rx_bytes;
1380         adapter->total_rx_packets += total_rx_packets;
1381         return cleaned;
1382 }
1383
1384 /**
1385  * e1000_consume_page - helper function
1386  **/
1387 static void e1000_consume_page(struct e1000_buffer *bi, struct sk_buff *skb,
1388                                u16 length)
1389 {
1390         bi->page = NULL;
1391         skb->len += length;
1392         skb->data_len += length;
1393         skb->truesize += PAGE_SIZE;
1394 }
1395
1396 /**
1397  * e1000_clean_jumbo_rx_irq - Send received data up the network stack; legacy
1398  * @adapter: board private structure
1399  *
1400  * the return value indicates whether actual cleaning was done, there
1401  * is no guarantee that everything was cleaned
1402  **/
1403 static bool e1000_clean_jumbo_rx_irq(struct e1000_ring *rx_ring, int *work_done,
1404                                      int work_to_do)
1405 {
1406         struct e1000_adapter *adapter = rx_ring->adapter;
1407         struct net_device *netdev = adapter->netdev;
1408         struct pci_dev *pdev = adapter->pdev;
1409         union e1000_rx_desc_extended *rx_desc, *next_rxd;
1410         struct e1000_buffer *buffer_info, *next_buffer;
1411         u32 length, staterr;
1412         unsigned int i;
1413         int cleaned_count = 0;
1414         bool cleaned = false;
1415         unsigned int total_rx_bytes=0, total_rx_packets=0;
1416
1417         i = rx_ring->next_to_clean;
1418         rx_desc = E1000_RX_DESC_EXT(*rx_ring, i);
1419         staterr = le32_to_cpu(rx_desc->wb.upper.status_error);
1420         buffer_info = &rx_ring->buffer_info[i];
1421
1422         while (staterr & E1000_RXD_STAT_DD) {
1423                 struct sk_buff *skb;
1424
1425                 if (*work_done >= work_to_do)
1426                         break;
1427                 (*work_done)++;
1428                 rmb();  /* read descriptor and rx_buffer_info after status DD */
1429
1430                 skb = buffer_info->skb;
1431                 buffer_info->skb = NULL;
1432
1433                 ++i;
1434                 if (i == rx_ring->count)
1435                         i = 0;
1436                 next_rxd = E1000_RX_DESC_EXT(*rx_ring, i);
1437                 prefetch(next_rxd);
1438
1439                 next_buffer = &rx_ring->buffer_info[i];
1440
1441                 cleaned = true;
1442                 cleaned_count++;
1443                 dma_unmap_page(&pdev->dev, buffer_info->dma, PAGE_SIZE,
1444                                DMA_FROM_DEVICE);
1445                 buffer_info->dma = 0;
1446
1447                 length = le16_to_cpu(rx_desc->wb.upper.length);
1448
1449                 /* errors is only valid for DD + EOP descriptors */
1450                 if (unlikely((staterr & E1000_RXD_STAT_EOP) &&
1451                              ((staterr & E1000_RXDEXT_ERR_FRAME_ERR_MASK) &&
1452                               !(netdev->features & NETIF_F_RXALL)))) {
1453                         /* recycle both page and skb */
1454                         buffer_info->skb = skb;
1455                         /* an error means any chain goes out the window too */
1456                         if (rx_ring->rx_skb_top)
1457                                 dev_kfree_skb_irq(rx_ring->rx_skb_top);
1458                         rx_ring->rx_skb_top = NULL;
1459                         goto next_desc;
1460                 }
1461
1462 #define rxtop (rx_ring->rx_skb_top)
1463                 if (!(staterr & E1000_RXD_STAT_EOP)) {
1464                         /* this descriptor is only the beginning (or middle) */
1465                         if (!rxtop) {
1466                                 /* this is the beginning of a chain */
1467                                 rxtop = skb;
1468                                 skb_fill_page_desc(rxtop, 0, buffer_info->page,
1469                                                    0, length);
1470                         } else {
1471                                 /* this is the middle of a chain */
1472                                 skb_fill_page_desc(rxtop,
1473                                     skb_shinfo(rxtop)->nr_frags,
1474                                     buffer_info->page, 0, length);
1475                                 /* re-use the skb, only consumed the page */
1476                                 buffer_info->skb = skb;
1477                         }
1478                         e1000_consume_page(buffer_info, rxtop, length);
1479                         goto next_desc;
1480                 } else {
1481                         if (rxtop) {
1482                                 /* end of the chain */
1483                                 skb_fill_page_desc(rxtop,
1484                                     skb_shinfo(rxtop)->nr_frags,
1485                                     buffer_info->page, 0, length);
1486                                 /* re-use the current skb, we only consumed the
1487                                  * page */
1488                                 buffer_info->skb = skb;
1489                                 skb = rxtop;
1490                                 rxtop = NULL;
1491                                 e1000_consume_page(buffer_info, skb, length);
1492                         } else {
1493                                 /* no chain, got EOP, this buf is the packet
1494                                  * copybreak to save the put_page/alloc_page */
1495                                 if (length <= copybreak &&
1496                                     skb_tailroom(skb) >= length) {
1497                                         u8 *vaddr;
1498                                         vaddr = kmap_atomic(buffer_info->page);
1499                                         memcpy(skb_tail_pointer(skb), vaddr,
1500                                                length);
1501                                         kunmap_atomic(vaddr);
1502                                         /* re-use the page, so don't erase
1503                                          * buffer_info->page */
1504                                         skb_put(skb, length);
1505                                 } else {
1506                                         skb_fill_page_desc(skb, 0,
1507                                                            buffer_info->page, 0,
1508                                                            length);
1509                                         e1000_consume_page(buffer_info, skb,
1510                                                            length);
1511                                 }
1512                         }
1513                 }
1514
1515                 /* Receive Checksum Offload XXX recompute due to CRC strip? */
1516                 e1000_rx_checksum(adapter, staterr,
1517                                   rx_desc->wb.lower.hi_dword.csum_ip.csum, skb);
1518
1519                 e1000_rx_hash(netdev, rx_desc->wb.lower.hi_dword.rss, skb);
1520
1521                 /* probably a little skewed due to removing CRC */
1522                 total_rx_bytes += skb->len;
1523                 total_rx_packets++;
1524
1525                 /* eth type trans needs skb->data to point to something */
1526                 if (!pskb_may_pull(skb, ETH_HLEN)) {
1527                         e_err("pskb_may_pull failed.\n");
1528                         dev_kfree_skb_irq(skb);
1529                         goto next_desc;
1530                 }
1531
1532                 e1000_receive_skb(adapter, netdev, skb, staterr,
1533                                   rx_desc->wb.upper.vlan);
1534
1535 next_desc:
1536                 rx_desc->wb.upper.status_error &= cpu_to_le32(~0xFF);
1537
1538                 /* return some buffers to hardware, one at a time is too slow */
1539                 if (unlikely(cleaned_count >= E1000_RX_BUFFER_WRITE)) {
1540                         adapter->alloc_rx_buf(rx_ring, cleaned_count,
1541                                               GFP_ATOMIC);
1542                         cleaned_count = 0;
1543                 }
1544
1545                 /* use prefetched values */
1546                 rx_desc = next_rxd;
1547                 buffer_info = next_buffer;
1548
1549                 staterr = le32_to_cpu(rx_desc->wb.upper.status_error);
1550         }
1551         rx_ring->next_to_clean = i;
1552
1553         cleaned_count = e1000_desc_unused(rx_ring);
1554         if (cleaned_count)
1555                 adapter->alloc_rx_buf(rx_ring, cleaned_count, GFP_ATOMIC);
1556
1557         adapter->total_rx_bytes += total_rx_bytes;
1558         adapter->total_rx_packets += total_rx_packets;
1559         return cleaned;
1560 }
1561
1562 /**
1563  * e1000_clean_rx_ring - Free Rx Buffers per Queue
1564  * @rx_ring: Rx descriptor ring
1565  **/
1566 static void e1000_clean_rx_ring(struct e1000_ring *rx_ring)
1567 {
1568         struct e1000_adapter *adapter = rx_ring->adapter;
1569         struct e1000_buffer *buffer_info;
1570         struct e1000_ps_page *ps_page;
1571         struct pci_dev *pdev = adapter->pdev;
1572         unsigned int i, j;
1573
1574         /* Free all the Rx ring sk_buffs */
1575         for (i = 0; i < rx_ring->count; i++) {
1576                 buffer_info = &rx_ring->buffer_info[i];
1577                 if (buffer_info->dma) {
1578                         if (adapter->clean_rx == e1000_clean_rx_irq)
1579                                 dma_unmap_single(&pdev->dev, buffer_info->dma,
1580                                                  adapter->rx_buffer_len,
1581                                                  DMA_FROM_DEVICE);
1582                         else if (adapter->clean_rx == e1000_clean_jumbo_rx_irq)
1583                                 dma_unmap_page(&pdev->dev, buffer_info->dma,
1584                                                PAGE_SIZE,
1585                                                DMA_FROM_DEVICE);
1586                         else if (adapter->clean_rx == e1000_clean_rx_irq_ps)
1587                                 dma_unmap_single(&pdev->dev, buffer_info->dma,
1588                                                  adapter->rx_ps_bsize0,
1589                                                  DMA_FROM_DEVICE);
1590                         buffer_info->dma = 0;
1591                 }
1592
1593                 if (buffer_info->page) {
1594                         put_page(buffer_info->page);
1595                         buffer_info->page = NULL;
1596                 }
1597
1598                 if (buffer_info->skb) {
1599                         dev_kfree_skb(buffer_info->skb);
1600                         buffer_info->skb = NULL;
1601                 }
1602
1603                 for (j = 0; j < PS_PAGE_BUFFERS; j++) {
1604                         ps_page = &buffer_info->ps_pages[j];
1605                         if (!ps_page->page)
1606                                 break;
1607                         dma_unmap_page(&pdev->dev, ps_page->dma, PAGE_SIZE,
1608                                        DMA_FROM_DEVICE);
1609                         ps_page->dma = 0;
1610                         put_page(ps_page->page);
1611                         ps_page->page = NULL;
1612                 }
1613         }
1614
1615         /* there also may be some cached data from a chained receive */
1616         if (rx_ring->rx_skb_top) {
1617                 dev_kfree_skb(rx_ring->rx_skb_top);
1618                 rx_ring->rx_skb_top = NULL;
1619         }
1620
1621         /* Zero out the descriptor ring */
1622         memset(rx_ring->desc, 0, rx_ring->size);
1623
1624         rx_ring->next_to_clean = 0;
1625         rx_ring->next_to_use = 0;
1626         adapter->flags2 &= ~FLAG2_IS_DISCARDING;
1627
1628         writel(0, rx_ring->head);
1629         if (rx_ring->adapter->flags2 & FLAG2_PCIM2PCI_ARBITER_WA)
1630                 e1000e_update_rdt_wa(rx_ring, 0);
1631         else
1632                 writel(0, rx_ring->tail);
1633 }
1634
1635 static void e1000e_downshift_workaround(struct work_struct *work)
1636 {
1637         struct e1000_adapter *adapter = container_of(work,
1638                                         struct e1000_adapter, downshift_task);
1639
1640         if (test_bit(__E1000_DOWN, &adapter->state))
1641                 return;
1642
1643         e1000e_gig_downshift_workaround_ich8lan(&adapter->hw);
1644 }
1645
1646 /**
1647  * e1000_intr_msi - Interrupt Handler
1648  * @irq: interrupt number
1649  * @data: pointer to a network interface device structure
1650  **/
1651 static irqreturn_t e1000_intr_msi(int irq, void *data)
1652 {
1653         struct net_device *netdev = data;
1654         struct e1000_adapter *adapter = netdev_priv(netdev);
1655         struct e1000_hw *hw = &adapter->hw;
1656         u32 icr = er32(ICR);
1657
1658         /*
1659          * read ICR disables interrupts using IAM
1660          */
1661
1662         if (icr & E1000_ICR_LSC) {
1663                 hw->mac.get_link_status = true;
1664                 /*
1665                  * ICH8 workaround-- Call gig speed drop workaround on cable
1666                  * disconnect (LSC) before accessing any PHY registers
1667                  */
1668                 if ((adapter->flags & FLAG_LSC_GIG_SPEED_DROP) &&
1669                     (!(er32(STATUS) & E1000_STATUS_LU)))
1670                         schedule_work(&adapter->downshift_task);
1671
1672                 /*
1673                  * 80003ES2LAN workaround-- For packet buffer work-around on
1674                  * link down event; disable receives here in the ISR and reset
1675                  * adapter in watchdog
1676                  */
1677                 if (netif_carrier_ok(netdev) &&
1678                     adapter->flags & FLAG_RX_NEEDS_RESTART) {
1679                         /* disable receives */
1680                         u32 rctl = er32(RCTL);
1681                         ew32(RCTL, rctl & ~E1000_RCTL_EN);
1682                         adapter->flags |= FLAG_RX_RESTART_NOW;
1683                 }
1684                 /* guard against interrupt when we're going down */
1685                 if (!test_bit(__E1000_DOWN, &adapter->state))
1686                         mod_timer(&adapter->watchdog_timer, jiffies + 1);
1687         }
1688
1689         if (napi_schedule_prep(&adapter->napi)) {
1690                 adapter->total_tx_bytes = 0;
1691                 adapter->total_tx_packets = 0;
1692                 adapter->total_rx_bytes = 0;
1693                 adapter->total_rx_packets = 0;
1694                 __napi_schedule(&adapter->napi);
1695         }
1696
1697         return IRQ_HANDLED;
1698 }
1699
1700 /**
1701  * e1000_intr - Interrupt Handler
1702  * @irq: interrupt number
1703  * @data: pointer to a network interface device structure
1704  **/
1705 static irqreturn_t e1000_intr(int irq, void *data)
1706 {
1707         struct net_device *netdev = data;
1708         struct e1000_adapter *adapter = netdev_priv(netdev);
1709         struct e1000_hw *hw = &adapter->hw;
1710         u32 rctl, icr = er32(ICR);
1711
1712         if (!icr || test_bit(__E1000_DOWN, &adapter->state))
1713                 return IRQ_NONE;  /* Not our interrupt */
1714
1715         /*
1716          * IMS will not auto-mask if INT_ASSERTED is not set, and if it is
1717          * not set, then the adapter didn't send an interrupt
1718          */
1719         if (!(icr & E1000_ICR_INT_ASSERTED))
1720                 return IRQ_NONE;
1721
1722         /*
1723          * Interrupt Auto-Mask...upon reading ICR,
1724          * interrupts are masked.  No need for the
1725          * IMC write
1726          */
1727
1728         if (icr & E1000_ICR_LSC) {
1729                 hw->mac.get_link_status = true;
1730                 /*
1731                  * ICH8 workaround-- Call gig speed drop workaround on cable
1732                  * disconnect (LSC) before accessing any PHY registers
1733                  */
1734                 if ((adapter->flags & FLAG_LSC_GIG_SPEED_DROP) &&
1735                     (!(er32(STATUS) & E1000_STATUS_LU)))
1736                         schedule_work(&adapter->downshift_task);
1737
1738                 /*
1739                  * 80003ES2LAN workaround--
1740                  * For packet buffer work-around on link down event;
1741                  * disable receives here in the ISR and
1742                  * reset adapter in watchdog
1743                  */
1744                 if (netif_carrier_ok(netdev) &&
1745                     (adapter->flags & FLAG_RX_NEEDS_RESTART)) {
1746                         /* disable receives */
1747                         rctl = er32(RCTL);
1748                         ew32(RCTL, rctl & ~E1000_RCTL_EN);
1749                         adapter->flags |= FLAG_RX_RESTART_NOW;
1750                 }
1751                 /* guard against interrupt when we're going down */
1752                 if (!test_bit(__E1000_DOWN, &adapter->state))
1753                         mod_timer(&adapter->watchdog_timer, jiffies + 1);
1754         }
1755
1756         if (napi_schedule_prep(&adapter->napi)) {
1757                 adapter->total_tx_bytes = 0;
1758                 adapter->total_tx_packets = 0;
1759                 adapter->total_rx_bytes = 0;
1760                 adapter->total_rx_packets = 0;
1761                 __napi_schedule(&adapter->napi);
1762         }
1763
1764         return IRQ_HANDLED;
1765 }
1766
1767 static irqreturn_t e1000_msix_other(int irq, void *data)
1768 {
1769         struct net_device *netdev = data;
1770         struct e1000_adapter *adapter = netdev_priv(netdev);
1771         struct e1000_hw *hw = &adapter->hw;
1772         u32 icr = er32(ICR);
1773
1774         if (!(icr & E1000_ICR_INT_ASSERTED)) {
1775                 if (!test_bit(__E1000_DOWN, &adapter->state))
1776                         ew32(IMS, E1000_IMS_OTHER);
1777                 return IRQ_NONE;
1778         }
1779
1780         if (icr & adapter->eiac_mask)
1781                 ew32(ICS, (icr & adapter->eiac_mask));
1782
1783         if (icr & E1000_ICR_OTHER) {
1784                 if (!(icr & E1000_ICR_LSC))
1785                         goto no_link_interrupt;
1786                 hw->mac.get_link_status = true;
1787                 /* guard against interrupt when we're going down */
1788                 if (!test_bit(__E1000_DOWN, &adapter->state))
1789                         mod_timer(&adapter->watchdog_timer, jiffies + 1);
1790         }
1791
1792 no_link_interrupt:
1793         if (!test_bit(__E1000_DOWN, &adapter->state))
1794                 ew32(IMS, E1000_IMS_LSC | E1000_IMS_OTHER);
1795
1796         return IRQ_HANDLED;
1797 }
1798
1799
1800 static irqreturn_t e1000_intr_msix_tx(int irq, void *data)
1801 {
1802         struct net_device *netdev = data;
1803         struct e1000_adapter *adapter = netdev_priv(netdev);
1804         struct e1000_hw *hw = &adapter->hw;
1805         struct e1000_ring *tx_ring = adapter->tx_ring;
1806
1807
1808         adapter->total_tx_bytes = 0;
1809         adapter->total_tx_packets = 0;
1810
1811         if (!e1000_clean_tx_irq(tx_ring))
1812                 /* Ring was not completely cleaned, so fire another interrupt */
1813                 ew32(ICS, tx_ring->ims_val);
1814
1815         return IRQ_HANDLED;
1816 }
1817
1818 static irqreturn_t e1000_intr_msix_rx(int irq, void *data)
1819 {
1820         struct net_device *netdev = data;
1821         struct e1000_adapter *adapter = netdev_priv(netdev);
1822         struct e1000_ring *rx_ring = adapter->rx_ring;
1823
1824         /* Write the ITR value calculated at the end of the
1825          * previous interrupt.
1826          */
1827         if (rx_ring->set_itr) {
1828                 writel(1000000000 / (rx_ring->itr_val * 256),
1829                        rx_ring->itr_register);
1830                 rx_ring->set_itr = 0;
1831         }
1832
1833         if (napi_schedule_prep(&adapter->napi)) {
1834                 adapter->total_rx_bytes = 0;
1835                 adapter->total_rx_packets = 0;
1836                 __napi_schedule(&adapter->napi);
1837         }
1838         return IRQ_HANDLED;
1839 }
1840
1841 /**
1842  * e1000_configure_msix - Configure MSI-X hardware
1843  *
1844  * e1000_configure_msix sets up the hardware to properly
1845  * generate MSI-X interrupts.
1846  **/
1847 static void e1000_configure_msix(struct e1000_adapter *adapter)
1848 {
1849         struct e1000_hw *hw = &adapter->hw;
1850         struct e1000_ring *rx_ring = adapter->rx_ring;
1851         struct e1000_ring *tx_ring = adapter->tx_ring;
1852         int vector = 0;
1853         u32 ctrl_ext, ivar = 0;
1854
1855         adapter->eiac_mask = 0;
1856
1857         /* Workaround issue with spurious interrupts on 82574 in MSI-X mode */
1858         if (hw->mac.type == e1000_82574) {
1859                 u32 rfctl = er32(RFCTL);
1860                 rfctl |= E1000_RFCTL_ACK_DIS;
1861                 ew32(RFCTL, rfctl);
1862         }
1863
1864 #define E1000_IVAR_INT_ALLOC_VALID      0x8
1865         /* Configure Rx vector */
1866         rx_ring->ims_val = E1000_IMS_RXQ0;
1867         adapter->eiac_mask |= rx_ring->ims_val;
1868         if (rx_ring->itr_val)
1869                 writel(1000000000 / (rx_ring->itr_val * 256),
1870                        rx_ring->itr_register);
1871         else
1872                 writel(1, rx_ring->itr_register);
1873         ivar = E1000_IVAR_INT_ALLOC_VALID | vector;
1874
1875         /* Configure Tx vector */
1876         tx_ring->ims_val = E1000_IMS_TXQ0;
1877         vector++;
1878         if (tx_ring->itr_val)
1879                 writel(1000000000 / (tx_ring->itr_val * 256),
1880                        tx_ring->itr_register);
1881         else
1882                 writel(1, tx_ring->itr_register);
1883         adapter->eiac_mask |= tx_ring->ims_val;
1884         ivar |= ((E1000_IVAR_INT_ALLOC_VALID | vector) << 8);
1885
1886         /* set vector for Other Causes, e.g. link changes */
1887         vector++;
1888         ivar |= ((E1000_IVAR_INT_ALLOC_VALID | vector) << 16);
1889         if (rx_ring->itr_val)
1890                 writel(1000000000 / (rx_ring->itr_val * 256),
1891                        hw->hw_addr + E1000_EITR_82574(vector));
1892         else
1893                 writel(1, hw->hw_addr + E1000_EITR_82574(vector));
1894
1895         /* Cause Tx interrupts on every write back */
1896         ivar |= (1 << 31);
1897
1898         ew32(IVAR, ivar);
1899
1900         /* enable MSI-X PBA support */
1901         ctrl_ext = er32(CTRL_EXT);
1902         ctrl_ext |= E1000_CTRL_EXT_PBA_CLR;
1903
1904         /* Auto-Mask Other interrupts upon ICR read */
1905 #define E1000_EIAC_MASK_82574   0x01F00000
1906         ew32(IAM, ~E1000_EIAC_MASK_82574 | E1000_IMS_OTHER);
1907         ctrl_ext |= E1000_CTRL_EXT_EIAME;
1908         ew32(CTRL_EXT, ctrl_ext);
1909         e1e_flush();
1910 }
1911
1912 void e1000e_reset_interrupt_capability(struct e1000_adapter *adapter)
1913 {
1914         if (adapter->msix_entries) {
1915                 pci_disable_msix(adapter->pdev);
1916                 kfree(adapter->msix_entries);
1917                 adapter->msix_entries = NULL;
1918         } else if (adapter->flags & FLAG_MSI_ENABLED) {
1919                 pci_disable_msi(adapter->pdev);
1920                 adapter->flags &= ~FLAG_MSI_ENABLED;
1921         }
1922 }
1923
1924 /**
1925  * e1000e_set_interrupt_capability - set MSI or MSI-X if supported
1926  *
1927  * Attempt to configure interrupts using the best available
1928  * capabilities of the hardware and kernel.
1929  **/
1930 void e1000e_set_interrupt_capability(struct e1000_adapter *adapter)
1931 {
1932         int err;
1933         int i;
1934
1935         switch (adapter->int_mode) {
1936         case E1000E_INT_MODE_MSIX:
1937                 if (adapter->flags & FLAG_HAS_MSIX) {
1938                         adapter->num_vectors = 3; /* RxQ0, TxQ0 and other */
1939                         adapter->msix_entries = kcalloc(adapter->num_vectors,
1940                                                       sizeof(struct msix_entry),
1941                                                       GFP_KERNEL);
1942                         if (adapter->msix_entries) {
1943                                 for (i = 0; i < adapter->num_vectors; i++)
1944                                         adapter->msix_entries[i].entry = i;
1945
1946                                 err = pci_enable_msix(adapter->pdev,
1947                                                       adapter->msix_entries,
1948                                                       adapter->num_vectors);
1949                                 if (err == 0)
1950                                         return;
1951                         }
1952                         /* MSI-X failed, so fall through and try MSI */
1953                         e_err("Failed to initialize MSI-X interrupts.  Falling back to MSI interrupts.\n");
1954                         e1000e_reset_interrupt_capability(adapter);
1955                 }
1956                 adapter->int_mode = E1000E_INT_MODE_MSI;
1957                 /* Fall through */
1958         case E1000E_INT_MODE_MSI:
1959                 if (!pci_enable_msi(adapter->pdev)) {
1960                         adapter->flags |= FLAG_MSI_ENABLED;
1961                 } else {
1962                         adapter->int_mode = E1000E_INT_MODE_LEGACY;
1963                         e_err("Failed to initialize MSI interrupts.  Falling back to legacy interrupts.\n");
1964                 }
1965                 /* Fall through */
1966         case E1000E_INT_MODE_LEGACY:
1967                 /* Don't do anything; this is the system default */
1968                 break;
1969         }
1970
1971         /* store the number of vectors being used */
1972         adapter->num_vectors = 1;
1973 }
1974
1975 /**
1976  * e1000_request_msix - Initialize MSI-X interrupts
1977  *
1978  * e1000_request_msix allocates MSI-X vectors and requests interrupts from the
1979  * kernel.
1980  **/
1981 static int e1000_request_msix(struct e1000_adapter *adapter)
1982 {
1983         struct net_device *netdev = adapter->netdev;
1984         int err = 0, vector = 0;
1985
1986         if (strlen(netdev->name) < (IFNAMSIZ - 5))
1987                 snprintf(adapter->rx_ring->name,
1988                          sizeof(adapter->rx_ring->name) - 1,
1989                          "%s-rx-0", netdev->name);
1990         else
1991                 memcpy(adapter->rx_ring->name, netdev->name, IFNAMSIZ);
1992         err = request_irq(adapter->msix_entries[vector].vector,
1993                           e1000_intr_msix_rx, 0, adapter->rx_ring->name,
1994                           netdev);
1995         if (err)
1996                 return err;
1997         adapter->rx_ring->itr_register = adapter->hw.hw_addr +
1998             E1000_EITR_82574(vector);
1999         adapter->rx_ring->itr_val = adapter->itr;
2000         vector++;
2001
2002         if (strlen(netdev->name) < (IFNAMSIZ - 5))
2003                 snprintf(adapter->tx_ring->name,
2004                          sizeof(adapter->tx_ring->name) - 1,
2005                          "%s-tx-0", netdev->name);
2006         else
2007                 memcpy(adapter->tx_ring->name, netdev->name, IFNAMSIZ);
2008         err = request_irq(adapter->msix_entries[vector].vector,
2009                           e1000_intr_msix_tx, 0, adapter->tx_ring->name,
2010                           netdev);
2011         if (err)
2012                 return err;
2013         adapter->tx_ring->itr_register = adapter->hw.hw_addr +
2014             E1000_EITR_82574(vector);
2015         adapter->tx_ring->itr_val = adapter->itr;
2016         vector++;
2017
2018         err = request_irq(adapter->msix_entries[vector].vector,
2019                           e1000_msix_other, 0, netdev->name, netdev);
2020         if (err)
2021                 return err;
2022
2023         e1000_configure_msix(adapter);
2024
2025         return 0;
2026 }
2027
2028 /**
2029  * e1000_request_irq - initialize interrupts
2030  *
2031  * Attempts to configure interrupts using the best available
2032  * capabilities of the hardware and kernel.
2033  **/
2034 static int e1000_request_irq(struct e1000_adapter *adapter)
2035 {
2036         struct net_device *netdev = adapter->netdev;
2037         int err;
2038
2039         if (adapter->msix_entries) {
2040                 err = e1000_request_msix(adapter);
2041                 if (!err)
2042                         return err;
2043                 /* fall back to MSI */
2044                 e1000e_reset_interrupt_capability(adapter);
2045                 adapter->int_mode = E1000E_INT_MODE_MSI;
2046                 e1000e_set_interrupt_capability(adapter);
2047         }
2048         if (adapter->flags & FLAG_MSI_ENABLED) {
2049                 err = request_irq(adapter->pdev->irq, e1000_intr_msi, 0,
2050                                   netdev->name, netdev);
2051                 if (!err)
2052                         return err;
2053
2054                 /* fall back to legacy interrupt */
2055                 e1000e_reset_interrupt_capability(adapter);
2056                 adapter->int_mode = E1000E_INT_MODE_LEGACY;
2057         }
2058
2059         err = request_irq(adapter->pdev->irq, e1000_intr, IRQF_SHARED,
2060                           netdev->name, netdev);
2061         if (err)
2062                 e_err("Unable to allocate interrupt, Error: %d\n", err);
2063
2064         return err;
2065 }
2066
2067 static void e1000_free_irq(struct e1000_adapter *adapter)
2068 {
2069         struct net_device *netdev = adapter->netdev;
2070
2071         if (adapter->msix_entries) {
2072                 int vector = 0;
2073
2074                 free_irq(adapter->msix_entries[vector].vector, netdev);
2075                 vector++;
2076
2077                 free_irq(adapter->msix_entries[vector].vector, netdev);
2078                 vector++;
2079
2080                 /* Other Causes interrupt vector */
2081                 free_irq(adapter->msix_entries[vector].vector, netdev);
2082                 return;
2083         }
2084
2085         free_irq(adapter->pdev->irq, netdev);
2086 }
2087
2088 /**
2089  * e1000_irq_disable - Mask off interrupt generation on the NIC
2090  **/
2091 static void e1000_irq_disable(struct e1000_adapter *adapter)
2092 {
2093         struct e1000_hw *hw = &adapter->hw;
2094
2095         ew32(IMC, ~0);
2096         if (adapter->msix_entries)
2097                 ew32(EIAC_82574, 0);
2098         e1e_flush();
2099
2100         if (adapter->msix_entries) {
2101                 int i;
2102                 for (i = 0; i < adapter->num_vectors; i++)
2103                         synchronize_irq(adapter->msix_entries[i].vector);
2104         } else {
2105                 synchronize_irq(adapter->pdev->irq);
2106         }
2107 }
2108
2109 /**
2110  * e1000_irq_enable - Enable default interrupt generation settings
2111  **/
2112 static void e1000_irq_enable(struct e1000_adapter *adapter)
2113 {
2114         struct e1000_hw *hw = &adapter->hw;
2115
2116         if (adapter->msix_entries) {
2117                 ew32(EIAC_82574, adapter->eiac_mask & E1000_EIAC_MASK_82574);
2118                 ew32(IMS, adapter->eiac_mask | E1000_IMS_OTHER | E1000_IMS_LSC);
2119         } else {
2120                 ew32(IMS, IMS_ENABLE_MASK);
2121         }
2122         e1e_flush();
2123 }
2124
2125 /**
2126  * e1000e_get_hw_control - get control of the h/w from f/w
2127  * @adapter: address of board private structure
2128  *
2129  * e1000e_get_hw_control sets {CTRL_EXT|SWSM}:DRV_LOAD bit.
2130  * For ASF and Pass Through versions of f/w this means that
2131  * the driver is loaded. For AMT version (only with 82573)
2132  * of the f/w this means that the network i/f is open.
2133  **/
2134 void e1000e_get_hw_control(struct e1000_adapter *adapter)
2135 {
2136         struct e1000_hw *hw = &adapter->hw;
2137         u32 ctrl_ext;
2138         u32 swsm;
2139
2140         /* Let firmware know the driver has taken over */
2141         if (adapter->flags & FLAG_HAS_SWSM_ON_LOAD) {
2142                 swsm = er32(SWSM);
2143                 ew32(SWSM, swsm | E1000_SWSM_DRV_LOAD);
2144         } else if (adapter->flags & FLAG_HAS_CTRLEXT_ON_LOAD) {
2145                 ctrl_ext = er32(CTRL_EXT);
2146                 ew32(CTRL_EXT, ctrl_ext | E1000_CTRL_EXT_DRV_LOAD);
2147         }
2148 }
2149
2150 /**
2151  * e1000e_release_hw_control - release control of the h/w to f/w
2152  * @adapter: address of board private structure
2153  *
2154  * e1000e_release_hw_control resets {CTRL_EXT|SWSM}:DRV_LOAD bit.
2155  * For ASF and Pass Through versions of f/w this means that the
2156  * driver is no longer loaded. For AMT version (only with 82573) i
2157  * of the f/w this means that the network i/f is closed.
2158  *
2159  **/
2160 void e1000e_release_hw_control(struct e1000_adapter *adapter)
2161 {
2162         struct e1000_hw *hw = &adapter->hw;
2163         u32 ctrl_ext;
2164         u32 swsm;
2165
2166         /* Let firmware taken over control of h/w */
2167         if (adapter->flags & FLAG_HAS_SWSM_ON_LOAD) {
2168                 swsm = er32(SWSM);
2169                 ew32(SWSM, swsm & ~E1000_SWSM_DRV_LOAD);
2170         } else if (adapter->flags & FLAG_HAS_CTRLEXT_ON_LOAD) {
2171                 ctrl_ext = er32(CTRL_EXT);
2172                 ew32(CTRL_EXT, ctrl_ext & ~E1000_CTRL_EXT_DRV_LOAD);
2173         }
2174 }
2175
2176 /**
2177  * @e1000_alloc_ring - allocate memory for a ring structure
2178  **/
2179 static int e1000_alloc_ring_dma(struct e1000_adapter *adapter,
2180                                 struct e1000_ring *ring)
2181 {
2182         struct pci_dev *pdev = adapter->pdev;
2183
2184         ring->desc = dma_alloc_coherent(&pdev->dev, ring->size, &ring->dma,
2185                                         GFP_KERNEL);
2186         if (!ring->desc)
2187                 return -ENOMEM;
2188
2189         return 0;
2190 }
2191
2192 /**
2193  * e1000e_setup_tx_resources - allocate Tx resources (Descriptors)
2194  * @tx_ring: Tx descriptor ring
2195  *
2196  * Return 0 on success, negative on failure
2197  **/
2198 int e1000e_setup_tx_resources(struct e1000_ring *tx_ring)
2199 {
2200         struct e1000_adapter *adapter = tx_ring->adapter;
2201         int err = -ENOMEM, size;
2202
2203         size = sizeof(struct e1000_buffer) * tx_ring->count;
2204         tx_ring->buffer_info = vzalloc(size);
2205         if (!tx_ring->buffer_info)
2206                 goto err;
2207
2208         /* round up to nearest 4K */
2209         tx_ring->size = tx_ring->count * sizeof(struct e1000_tx_desc);
2210         tx_ring->size = ALIGN(tx_ring->size, 4096);
2211
2212         err = e1000_alloc_ring_dma(adapter, tx_ring);
2213         if (err)
2214                 goto err;
2215
2216         tx_ring->next_to_use = 0;
2217         tx_ring->next_to_clean = 0;
2218
2219         return 0;
2220 err:
2221         vfree(tx_ring->buffer_info);
2222         e_err("Unable to allocate memory for the transmit descriptor ring\n");
2223         return err;
2224 }
2225
2226 /**
2227  * e1000e_setup_rx_resources - allocate Rx resources (Descriptors)
2228  * @rx_ring: Rx descriptor ring
2229  *
2230  * Returns 0 on success, negative on failure
2231  **/
2232 int e1000e_setup_rx_resources(struct e1000_ring *rx_ring)
2233 {
2234         struct e1000_adapter *adapter = rx_ring->adapter;
2235         struct e1000_buffer *buffer_info;
2236         int i, size, desc_len, err = -ENOMEM;
2237
2238         size = sizeof(struct e1000_buffer) * rx_ring->count;
2239         rx_ring->buffer_info = vzalloc(size);
2240         if (!rx_ring->buffer_info)
2241                 goto err;
2242
2243         for (i = 0; i < rx_ring->count; i++) {
2244                 buffer_info = &rx_ring->buffer_info[i];
2245                 buffer_info->ps_pages = kcalloc(PS_PAGE_BUFFERS,
2246                                                 sizeof(struct e1000_ps_page),
2247                                                 GFP_KERNEL);
2248                 if (!buffer_info->ps_pages)
2249                         goto err_pages;
2250         }
2251
2252         desc_len = sizeof(union e1000_rx_desc_packet_split);
2253
2254         /* Round up to nearest 4K */
2255         rx_ring->size = rx_ring->count * desc_len;
2256         rx_ring->size = ALIGN(rx_ring->size, 4096);
2257
2258         err = e1000_alloc_ring_dma(adapter, rx_ring);
2259         if (err)
2260                 goto err_pages;
2261
2262         rx_ring->next_to_clean = 0;
2263         rx_ring->next_to_use = 0;
2264         rx_ring->rx_skb_top = NULL;
2265
2266         return 0;
2267
2268 err_pages:
2269         for (i = 0; i < rx_ring->count; i++) {
2270                 buffer_info = &rx_ring->buffer_info[i];
2271                 kfree(buffer_info->ps_pages);
2272         }
2273 err:
2274         vfree(rx_ring->buffer_info);
2275         e_err("Unable to allocate memory for the receive descriptor ring\n");
2276         return err;
2277 }
2278
2279 /**
2280  * e1000_clean_tx_ring - Free Tx Buffers
2281  * @tx_ring: Tx descriptor ring
2282  **/
2283 static void e1000_clean_tx_ring(struct e1000_ring *tx_ring)
2284 {
2285         struct e1000_adapter *adapter = tx_ring->adapter;
2286         struct e1000_buffer *buffer_info;
2287         unsigned long size;
2288         unsigned int i;
2289
2290         for (i = 0; i < tx_ring->count; i++) {
2291                 buffer_info = &tx_ring->buffer_info[i];
2292                 e1000_put_txbuf(tx_ring, buffer_info);
2293         }
2294
2295         netdev_reset_queue(adapter->netdev);
2296         size = sizeof(struct e1000_buffer) * tx_ring->count;
2297         memset(tx_ring->buffer_info, 0, size);
2298
2299         memset(tx_ring->desc, 0, tx_ring->size);
2300
2301         tx_ring->next_to_use = 0;
2302         tx_ring->next_to_clean = 0;
2303
2304         writel(0, tx_ring->head);
2305         if (tx_ring->adapter->flags2 & FLAG2_PCIM2PCI_ARBITER_WA)
2306                 e1000e_update_tdt_wa(tx_ring, 0);
2307         else
2308                 writel(0, tx_ring->tail);
2309 }
2310
2311 /**
2312  * e1000e_free_tx_resources - Free Tx Resources per Queue
2313  * @tx_ring: Tx descriptor ring
2314  *
2315  * Free all transmit software resources
2316  **/
2317 void e1000e_free_tx_resources(struct e1000_ring *tx_ring)
2318 {
2319         struct e1000_adapter *adapter = tx_ring->adapter;
2320         struct pci_dev *pdev = adapter->pdev;
2321
2322         e1000_clean_tx_ring(tx_ring);
2323
2324         vfree(tx_ring->buffer_info);
2325         tx_ring->buffer_info = NULL;
2326
2327         dma_free_coherent(&pdev->dev, tx_ring->size, tx_ring->desc,
2328                           tx_ring->dma);
2329         tx_ring->desc = NULL;
2330 }
2331
2332 /**
2333  * e1000e_free_rx_resources - Free Rx Resources
2334  * @rx_ring: Rx descriptor ring
2335  *
2336  * Free all receive software resources
2337  **/
2338 void e1000e_free_rx_resources(struct e1000_ring *rx_ring)
2339 {
2340         struct e1000_adapter *adapter = rx_ring->adapter;
2341         struct pci_dev *pdev = adapter->pdev;
2342         int i;
2343
2344         e1000_clean_rx_ring(rx_ring);
2345
2346         for (i = 0; i < rx_ring->count; i++)
2347                 kfree(rx_ring->buffer_info[i].ps_pages);
2348
2349         vfree(rx_ring->buffer_info);
2350         rx_ring->buffer_info = NULL;
2351
2352         dma_free_coherent(&pdev->dev, rx_ring->size, rx_ring->desc,
2353                           rx_ring->dma);
2354         rx_ring->desc = NULL;
2355 }
2356
2357 /**
2358  * e1000_update_itr - update the dynamic ITR value based on statistics
2359  * @adapter: pointer to adapter
2360  * @itr_setting: current adapter->itr
2361  * @packets: the number of packets during this measurement interval
2362  * @bytes: the number of bytes during this measurement interval
2363  *
2364  *      Stores a new ITR value based on packets and byte
2365  *      counts during the last interrupt.  The advantage of per interrupt
2366  *      computation is faster updates and more accurate ITR for the current
2367  *      traffic pattern.  Constants in this function were computed
2368  *      based on theoretical maximum wire speed and thresholds were set based
2369  *      on testing data as well as attempting to minimize response time
2370  *      while increasing bulk throughput.  This functionality is controlled
2371  *      by the InterruptThrottleRate module parameter.
2372  **/
2373 static unsigned int e1000_update_itr(struct e1000_adapter *adapter,
2374                                      u16 itr_setting, int packets,
2375                                      int bytes)
2376 {
2377         unsigned int retval = itr_setting;
2378
2379         if (packets == 0)
2380                 return itr_setting;
2381
2382         switch (itr_setting) {
2383         case lowest_latency:
2384                 /* handle TSO and jumbo frames */
2385                 if (bytes/packets > 8000)
2386                         retval = bulk_latency;
2387                 else if ((packets < 5) && (bytes > 512))
2388                         retval = low_latency;
2389                 break;
2390         case low_latency:  /* 50 usec aka 20000 ints/s */
2391                 if (bytes > 10000) {
2392                         /* this if handles the TSO accounting */
2393                         if (bytes/packets > 8000)
2394                                 retval = bulk_latency;
2395                         else if ((packets < 10) || ((bytes/packets) > 1200))
2396                                 retval = bulk_latency;
2397                         else if ((packets > 35))
2398                                 retval = lowest_latency;
2399                 } else if (bytes/packets > 2000) {
2400                         retval = bulk_latency;
2401                 } else if (packets <= 2 && bytes < 512) {
2402                         retval = lowest_latency;
2403                 }
2404                 break;
2405         case bulk_latency: /* 250 usec aka 4000 ints/s */
2406                 if (bytes > 25000) {
2407                         if (packets > 35)
2408                                 retval = low_latency;
2409                 } else if (bytes < 6000) {
2410                         retval = low_latency;
2411                 }
2412                 break;
2413         }
2414
2415         return retval;
2416 }
2417
2418 static void e1000_set_itr(struct e1000_adapter *adapter)
2419 {
2420         struct e1000_hw *hw = &adapter->hw;
2421         u16 current_itr;
2422         u32 new_itr = adapter->itr;
2423
2424         /* for non-gigabit speeds, just fix the interrupt rate at 4000 */
2425         if (adapter->link_speed != SPEED_1000) {
2426                 current_itr = 0;
2427                 new_itr = 4000;
2428                 goto set_itr_now;
2429         }
2430
2431         if (adapter->flags2 & FLAG2_DISABLE_AIM) {
2432                 new_itr = 0;
2433                 goto set_itr_now;
2434         }
2435
2436         adapter->tx_itr = e1000_update_itr(adapter,
2437                                     adapter->tx_itr,
2438                                     adapter->total_tx_packets,
2439                                     adapter->total_tx_bytes);
2440         /* conservative mode (itr 3) eliminates the lowest_latency setting */
2441         if (adapter->itr_setting == 3 && adapter->tx_itr == lowest_latency)
2442                 adapter->tx_itr = low_latency;
2443
2444         adapter->rx_itr = e1000_update_itr(adapter,
2445                                     adapter->rx_itr,
2446                                     adapter->total_rx_packets,
2447                                     adapter->total_rx_bytes);
2448         /* conservative mode (itr 3) eliminates the lowest_latency setting */
2449         if (adapter->itr_setting == 3 && adapter->rx_itr == lowest_latency)
2450                 adapter->rx_itr = low_latency;
2451
2452         current_itr = max(adapter->rx_itr, adapter->tx_itr);
2453
2454         switch (current_itr) {
2455         /* counts and packets in update_itr are dependent on these numbers */
2456         case lowest_latency:
2457                 new_itr = 70000;
2458                 break;
2459         case low_latency:
2460                 new_itr = 20000; /* aka hwitr = ~200 */
2461                 break;
2462         case bulk_latency:
2463                 new_itr = 4000;
2464                 break;
2465         default:
2466                 break;
2467         }
2468
2469 set_itr_now:
2470         if (new_itr != adapter->itr) {
2471                 /*
2472                  * this attempts to bias the interrupt rate towards Bulk
2473                  * by adding intermediate steps when interrupt rate is
2474                  * increasing
2475                  */
2476                 new_itr = new_itr > adapter->itr ?
2477                              min(adapter->itr + (new_itr >> 2), new_itr) :
2478                              new_itr;
2479                 adapter->itr = new_itr;
2480                 adapter->rx_ring->itr_val = new_itr;
2481                 if (adapter->msix_entries)
2482                         adapter->rx_ring->set_itr = 1;
2483                 else
2484                         if (new_itr)
2485                                 ew32(ITR, 1000000000 / (new_itr * 256));
2486                         else
2487                                 ew32(ITR, 0);
2488         }
2489 }
2490
2491 /**
2492  * e1000_alloc_queues - Allocate memory for all rings
2493  * @adapter: board private structure to initialize
2494  **/
2495 static int __devinit e1000_alloc_queues(struct e1000_adapter *adapter)
2496 {
2497         int size = sizeof(struct e1000_ring);
2498
2499         adapter->tx_ring = kzalloc(size, GFP_KERNEL);
2500         if (!adapter->tx_ring)
2501                 goto err;
2502         adapter->tx_ring->count = adapter->tx_ring_count;
2503         adapter->tx_ring->adapter = adapter;
2504
2505         adapter->rx_ring = kzalloc(size, GFP_KERNEL);
2506         if (!adapter->rx_ring)
2507                 goto err;
2508         adapter->rx_ring->count = adapter->rx_ring_count;
2509         adapter->rx_ring->adapter = adapter;
2510
2511         return 0;
2512 err:
2513         e_err("Unable to allocate memory for queues\n");
2514         kfree(adapter->rx_ring);
2515         kfree(adapter->tx_ring);
2516         return -ENOMEM;
2517 }
2518
2519 /**
2520  * e1000e_poll - NAPI Rx polling callback
2521  * @napi: struct associated with this polling callback
2522  * @weight: number of packets driver is allowed to process this poll
2523  **/
2524 static int e1000e_poll(struct napi_struct *napi, int weight)
2525 {
2526         struct e1000_adapter *adapter = container_of(napi, struct e1000_adapter,
2527                                                      napi);
2528         struct e1000_hw *hw = &adapter->hw;
2529         struct net_device *poll_dev = adapter->netdev;
2530         int tx_cleaned = 1, work_done = 0;
2531
2532         adapter = netdev_priv(poll_dev);
2533
2534         if (!adapter->msix_entries ||
2535             (adapter->rx_ring->ims_val & adapter->tx_ring->ims_val))
2536                 tx_cleaned = e1000_clean_tx_irq(adapter->tx_ring);
2537
2538         adapter->clean_rx(adapter->rx_ring, &work_done, weight);
2539
2540         if (!tx_cleaned)
2541                 work_done = weight;
2542
2543         /* If weight not fully consumed, exit the polling mode */
2544         if (work_done < weight) {
2545                 if (adapter->itr_setting & 3)
2546                         e1000_set_itr(adapter);
2547                 napi_complete(napi);
2548                 if (!test_bit(__E1000_DOWN, &adapter->state)) {
2549                         if (adapter->msix_entries)
2550                                 ew32(IMS, adapter->rx_ring->ims_val);
2551                         else
2552                                 e1000_irq_enable(adapter);
2553                 }
2554         }
2555
2556         return work_done;
2557 }
2558
2559 static int e1000_vlan_rx_add_vid(struct net_device *netdev, u16 vid)
2560 {
2561         struct e1000_adapter *adapter = netdev_priv(netdev);
2562         struct e1000_hw *hw = &adapter->hw;
2563         u32 vfta, index;
2564
2565         /* don't update vlan cookie if already programmed */
2566         if ((adapter->hw.mng_cookie.status &
2567              E1000_MNG_DHCP_COOKIE_STATUS_VLAN) &&
2568             (vid == adapter->mng_vlan_id))
2569                 return 0;
2570
2571         /* add VID to filter table */
2572         if (adapter->flags & FLAG_HAS_HW_VLAN_FILTER) {
2573                 index = (vid >> 5) & 0x7F;
2574                 vfta = E1000_READ_REG_ARRAY(hw, E1000_VFTA, index);
2575                 vfta |= (1 << (vid & 0x1F));
2576                 hw->mac.ops.write_vfta(hw, index, vfta);
2577         }
2578
2579         set_bit(vid, adapter->active_vlans);
2580
2581         return 0;
2582 }
2583
2584 static int e1000_vlan_rx_kill_vid(struct net_device *netdev, u16 vid)
2585 {
2586         struct e1000_adapter *adapter = netdev_priv(netdev);
2587         struct e1000_hw *hw = &adapter->hw;
2588         u32 vfta, index;
2589
2590         if ((adapter->hw.mng_cookie.status &
2591              E1000_MNG_DHCP_COOKIE_STATUS_VLAN) &&
2592             (vid == adapter->mng_vlan_id)) {
2593                 /* release control to f/w */
2594                 e1000e_release_hw_control(adapter);
2595                 return 0;
2596         }
2597
2598         /* remove VID from filter table */
2599         if (adapter->flags & FLAG_HAS_HW_VLAN_FILTER) {
2600                 index = (vid >> 5) & 0x7F;
2601                 vfta = E1000_READ_REG_ARRAY(hw, E1000_VFTA, index);
2602                 vfta &= ~(1 << (vid & 0x1F));
2603                 hw->mac.ops.write_vfta(hw, index, vfta);
2604         }
2605
2606         clear_bit(vid, adapter->active_vlans);
2607
2608         return 0;
2609 }
2610
2611 /**
2612  * e1000e_vlan_filter_disable - helper to disable hw VLAN filtering
2613  * @adapter: board private structure to initialize
2614  **/
2615 static void e1000e_vlan_filter_disable(struct e1000_adapter *adapter)
2616 {
2617         struct net_device *netdev = adapter->netdev;
2618         struct e1000_hw *hw = &adapter->hw;
2619         u32 rctl;
2620
2621         if (adapter->flags & FLAG_HAS_HW_VLAN_FILTER) {
2622                 /* disable VLAN receive filtering */
2623                 rctl = er32(RCTL);
2624                 rctl &= ~(E1000_RCTL_VFE | E1000_RCTL_CFIEN);
2625                 ew32(RCTL, rctl);
2626
2627                 if (adapter->mng_vlan_id != (u16)E1000_MNG_VLAN_NONE) {
2628                         e1000_vlan_rx_kill_vid(netdev, adapter->mng_vlan_id);
2629                         adapter->mng_vlan_id = E1000_MNG_VLAN_NONE;
2630                 }
2631         }
2632 }
2633
2634 /**
2635  * e1000e_vlan_filter_enable - helper to enable HW VLAN filtering
2636  * @adapter: board private structure to initialize
2637  **/
2638 static void e1000e_vlan_filter_enable(struct e1000_adapter *adapter)
2639 {
2640         struct e1000_hw *hw = &adapter->hw;
2641         u32 rctl;
2642
2643         if (adapter->flags & FLAG_HAS_HW_VLAN_FILTER) {
2644                 /* enable VLAN receive filtering */
2645                 rctl = er32(RCTL);
2646                 rctl |= E1000_RCTL_VFE;
2647                 rctl &= ~E1000_RCTL_CFIEN;
2648                 ew32(RCTL, rctl);
2649         }
2650 }
2651
2652 /**
2653  * e1000e_vlan_strip_enable - helper to disable HW VLAN stripping
2654  * @adapter: board private structure to initialize
2655  **/
2656 static void e1000e_vlan_strip_disable(struct e1000_adapter *adapter)
2657 {
2658         struct e1000_hw *hw = &adapter->hw;
2659         u32 ctrl;
2660
2661         /* disable VLAN tag insert/strip */
2662         ctrl = er32(CTRL);
2663         ctrl &= ~E1000_CTRL_VME;
2664         ew32(CTRL, ctrl);
2665 }
2666
2667 /**
2668  * e1000e_vlan_strip_enable - helper to enable HW VLAN stripping
2669  * @adapter: board private structure to initialize
2670  **/
2671 static void e1000e_vlan_strip_enable(struct e1000_adapter *adapter)
2672 {
2673         struct e1000_hw *hw = &adapter->hw;
2674         u32 ctrl;
2675
2676         /* enable VLAN tag insert/strip */
2677         ctrl = er32(CTRL);
2678         ctrl |= E1000_CTRL_VME;
2679         ew32(CTRL, ctrl);
2680 }
2681
2682 static void e1000_update_mng_vlan(struct e1000_adapter *adapter)
2683 {
2684         struct net_device *netdev = adapter->netdev;
2685         u16 vid = adapter->hw.mng_cookie.vlan_id;
2686         u16 old_vid = adapter->mng_vlan_id;
2687
2688         if (adapter->hw.mng_cookie.status &
2689             E1000_MNG_DHCP_COOKIE_STATUS_VLAN) {
2690                 e1000_vlan_rx_add_vid(netdev, vid);
2691                 adapter->mng_vlan_id = vid;
2692         }
2693
2694         if ((old_vid != (u16)E1000_MNG_VLAN_NONE) && (vid != old_vid))
2695                 e1000_vlan_rx_kill_vid(netdev, old_vid);
2696 }
2697
2698 static void e1000_restore_vlan(struct e1000_adapter *adapter)
2699 {
2700         u16 vid;
2701
2702         e1000_vlan_rx_add_vid(adapter->netdev, 0);
2703
2704         for_each_set_bit(vid, adapter->active_vlans, VLAN_N_VID)
2705                 e1000_vlan_rx_add_vid(adapter->netdev, vid);
2706 }
2707
2708 static void e1000_init_manageability_pt(struct e1000_adapter *adapter)
2709 {
2710         struct e1000_hw *hw = &adapter->hw;
2711         u32 manc, manc2h, mdef, i, j;
2712
2713         if (!(adapter->flags & FLAG_MNG_PT_ENABLED))
2714                 return;
2715
2716         manc = er32(MANC);
2717
2718         /*
2719          * enable receiving management packets to the host. this will probably
2720          * generate destination unreachable messages from the host OS, but
2721          * the packets will be handled on SMBUS
2722          */
2723         manc |= E1000_MANC_EN_MNG2HOST;
2724         manc2h = er32(MANC2H);
2725
2726         switch (hw->mac.type) {
2727         default:
2728                 manc2h |= (E1000_MANC2H_PORT_623 | E1000_MANC2H_PORT_664);
2729                 break;
2730         case e1000_82574:
2731         case e1000_82583:
2732                 /*
2733                  * Check if IPMI pass-through decision filter already exists;
2734                  * if so, enable it.
2735                  */
2736                 for (i = 0, j = 0; i < 8; i++) {
2737                         mdef = er32(MDEF(i));
2738
2739                         /* Ignore filters with anything other than IPMI ports */
2740                         if (mdef & ~(E1000_MDEF_PORT_623 | E1000_MDEF_PORT_664))
2741                                 continue;
2742
2743                         /* Enable this decision filter in MANC2H */
2744                         if (mdef)
2745                                 manc2h |= (1 << i);
2746
2747                         j |= mdef;
2748                 }
2749
2750                 if (j == (E1000_MDEF_PORT_623 | E1000_MDEF_PORT_664))
2751                         break;
2752
2753                 /* Create new decision filter in an empty filter */
2754                 for (i = 0, j = 0; i < 8; i++)
2755                         if (er32(MDEF(i)) == 0) {
2756                                 ew32(MDEF(i), (E1000_MDEF_PORT_623 |
2757                                                E1000_MDEF_PORT_664));
2758                                 manc2h |= (1 << 1);
2759                                 j++;
2760                                 break;
2761                         }
2762
2763                 if (!j)
2764                         e_warn("Unable to create IPMI pass-through filter\n");
2765                 break;
2766         }
2767
2768         ew32(MANC2H, manc2h);
2769         ew32(MANC, manc);
2770 }
2771
2772 /**
2773  * e1000_configure_tx - Configure Transmit Unit after Reset
2774  * @adapter: board private structure
2775  *
2776  * Configure the Tx unit of the MAC after a reset.
2777  **/
2778 static void e1000_configure_tx(struct e1000_adapter *adapter)
2779 {
2780         struct e1000_hw *hw = &adapter->hw;
2781         struct e1000_ring *tx_ring = adapter->tx_ring;
2782         u64 tdba;
2783         u32 tdlen, tarc;
2784
2785         /* Setup the HW Tx Head and Tail descriptor pointers */
2786         tdba = tx_ring->dma;
2787         tdlen = tx_ring->count * sizeof(struct e1000_tx_desc);
2788         ew32(TDBAL(0), (tdba & DMA_BIT_MASK(32)));
2789         ew32(TDBAH(0), (tdba >> 32));
2790         ew32(TDLEN(0), tdlen);
2791         ew32(TDH(0), 0);
2792         ew32(TDT(0), 0);
2793         tx_ring->head = adapter->hw.hw_addr + E1000_TDH(0);
2794         tx_ring->tail = adapter->hw.hw_addr + E1000_TDT(0);
2795
2796         /* Set the Tx Interrupt Delay register */
2797         ew32(TIDV, adapter->tx_int_delay);
2798         /* Tx irq moderation */
2799         ew32(TADV, adapter->tx_abs_int_delay);
2800
2801         if (adapter->flags2 & FLAG2_DMA_BURST) {
2802                 u32 txdctl = er32(TXDCTL(0));
2803                 txdctl &= ~(E1000_TXDCTL_PTHRESH | E1000_TXDCTL_HTHRESH |
2804                             E1000_TXDCTL_WTHRESH);
2805                 /*
2806                  * set up some performance related parameters to encourage the
2807                  * hardware to use the bus more efficiently in bursts, depends
2808                  * on the tx_int_delay to be enabled,
2809                  * wthresh = 5 ==> burst write a cacheline (64 bytes) at a time
2810                  * hthresh = 1 ==> prefetch when one or more available
2811                  * pthresh = 0x1f ==> prefetch if internal cache 31 or less
2812                  * BEWARE: this seems to work but should be considered first if
2813                  * there are Tx hangs or other Tx related bugs
2814                  */
2815                 txdctl |= E1000_TXDCTL_DMA_BURST_ENABLE;
2816                 ew32(TXDCTL(0), txdctl);
2817         }
2818         /* erratum work around: set txdctl the same for both queues */
2819         ew32(TXDCTL(1), er32(TXDCTL(0)));
2820
2821         if (adapter->flags & FLAG_TARC_SPEED_MODE_BIT) {
2822                 tarc = er32(TARC(0));
2823                 /*
2824                  * set the speed mode bit, we'll clear it if we're not at
2825                  * gigabit link later
2826                  */
2827 #define SPEED_MODE_BIT (1 << 21)
2828                 tarc |= SPEED_MODE_BIT;
2829                 ew32(TARC(0), tarc);
2830         }
2831
2832         /* errata: program both queues to unweighted RR */
2833         if (adapter->flags & FLAG_TARC_SET_BIT_ZERO) {
2834                 tarc = er32(TARC(0));
2835                 tarc |= 1;
2836                 ew32(TARC(0), tarc);
2837                 tarc = er32(TARC(1));
2838                 tarc |= 1;
2839                 ew32(TARC(1), tarc);
2840         }
2841
2842         /* Setup Transmit Descriptor Settings for eop descriptor */
2843         adapter->txd_cmd = E1000_TXD_CMD_EOP | E1000_TXD_CMD_IFCS;
2844
2845         /* only set IDE if we are delaying interrupts using the timers */
2846         if (adapter->tx_int_delay)
2847                 adapter->txd_cmd |= E1000_TXD_CMD_IDE;
2848
2849         /* enable Report Status bit */
2850         adapter->txd_cmd |= E1000_TXD_CMD_RS;
2851
2852         hw->mac.ops.config_collision_dist(hw);
2853 }
2854
2855 /**
2856  * e1000_setup_rctl - configure the receive control registers
2857  * @adapter: Board private structure
2858  **/
2859 #define PAGE_USE_COUNT(S) (((S) >> PAGE_SHIFT) + \
2860                            (((S) & (PAGE_SIZE - 1)) ? 1 : 0))
2861 static void e1000_setup_rctl(struct e1000_adapter *adapter)
2862 {
2863         struct e1000_hw *hw = &adapter->hw;
2864         u32 rctl, rfctl;
2865         u32 pages = 0;
2866
2867         /* Workaround Si errata on PCHx - configure jumbo frame flow */
2868         if (hw->mac.type >= e1000_pch2lan) {
2869                 s32 ret_val;
2870
2871                 if (adapter->netdev->mtu > ETH_DATA_LEN)
2872                         ret_val = e1000_lv_jumbo_workaround_ich8lan(hw, true);
2873                 else
2874                         ret_val = e1000_lv_jumbo_workaround_ich8lan(hw, false);
2875
2876                 if (ret_val)
2877                         e_dbg("failed to enable jumbo frame workaround mode\n");
2878         }
2879
2880         /* Program MC offset vector base */
2881         rctl = er32(RCTL);
2882         rctl &= ~(3 << E1000_RCTL_MO_SHIFT);
2883         rctl |= E1000_RCTL_EN | E1000_RCTL_BAM |
2884                 E1000_RCTL_LBM_NO | E1000_RCTL_RDMTS_HALF |
2885                 (adapter->hw.mac.mc_filter_type << E1000_RCTL_MO_SHIFT);
2886
2887         /* Do not Store bad packets */
2888         rctl &= ~E1000_RCTL_SBP;
2889
2890         /* Enable Long Packet receive */
2891         if (adapter->netdev->mtu <= ETH_DATA_LEN)
2892                 rctl &= ~E1000_RCTL_LPE;
2893         else
2894                 rctl |= E1000_RCTL_LPE;
2895
2896         /* Some systems expect that the CRC is included in SMBUS traffic. The
2897          * hardware strips the CRC before sending to both SMBUS (BMC) and to
2898          * host memory when this is enabled
2899          */
2900         if (adapter->flags2 & FLAG2_CRC_STRIPPING)
2901                 rctl |= E1000_RCTL_SECRC;
2902
2903         /* Workaround Si errata on 82577 PHY - configure IPG for jumbos */
2904         if ((hw->phy.type == e1000_phy_82577) && (rctl & E1000_RCTL_LPE)) {
2905                 u16 phy_data;
2906
2907                 e1e_rphy(hw, PHY_REG(770, 26), &phy_data);
2908                 phy_data &= 0xfff8;
2909                 phy_data |= (1 << 2);
2910                 e1e_wphy(hw, PHY_REG(770, 26), phy_data);
2911
2912                 e1e_rphy(hw, 22, &phy_data);
2913                 phy_data &= 0x0fff;
2914                 phy_data |= (1 << 14);
2915                 e1e_wphy(hw, 0x10, 0x2823);
2916                 e1e_wphy(hw, 0x11, 0x0003);
2917                 e1e_wphy(hw, 22, phy_data);
2918         }
2919
2920         /* Setup buffer sizes */
2921         rctl &= ~E1000_RCTL_SZ_4096;
2922         rctl |= E1000_RCTL_BSEX;
2923         switch (adapter->rx_buffer_len) {
2924         case 2048:
2925         default:
2926                 rctl |= E1000_RCTL_SZ_2048;
2927                 rctl &= ~E1000_RCTL_BSEX;
2928                 break;
2929         case 4096:
2930                 rctl |= E1000_RCTL_SZ_4096;
2931                 break;
2932         case 8192:
2933                 rctl |= E1000_RCTL_SZ_8192;
2934                 break;
2935         case 16384:
2936                 rctl |= E1000_RCTL_SZ_16384;
2937                 break;
2938         }
2939
2940         /* Enable Extended Status in all Receive Descriptors */
2941         rfctl = er32(RFCTL);
2942         rfctl |= E1000_RFCTL_EXTEN;
2943         ew32(RFCTL, rfctl);
2944
2945         /*
2946          * 82571 and greater support packet-split where the protocol
2947          * header is placed in skb->data and the packet data is
2948          * placed in pages hanging off of skb_shinfo(skb)->nr_frags.
2949          * In the case of a non-split, skb->data is linearly filled,
2950          * followed by the page buffers.  Therefore, skb->data is
2951          * sized to hold the largest protocol header.
2952          *
2953          * allocations using alloc_page take too long for regular MTU
2954          * so only enable packet split for jumbo frames
2955          *
2956          * Using pages when the page size is greater than 16k wastes
2957          * a lot of memory, since we allocate 3 pages at all times
2958          * per packet.
2959          */
2960         pages = PAGE_USE_COUNT(adapter->netdev->mtu);
2961         if ((pages <= 3) && (PAGE_SIZE <= 16384) && (rctl & E1000_RCTL_LPE))
2962                 adapter->rx_ps_pages = pages;
2963         else
2964                 adapter->rx_ps_pages = 0;
2965
2966         if (adapter->rx_ps_pages) {
2967                 u32 psrctl = 0;
2968
2969                 /* Enable Packet split descriptors */
2970                 rctl |= E1000_RCTL_DTYP_PS;
2971
2972                 psrctl |= adapter->rx_ps_bsize0 >>
2973                         E1000_PSRCTL_BSIZE0_SHIFT;
2974
2975                 switch (adapter->rx_ps_pages) {
2976                 case 3:
2977                         psrctl |= PAGE_SIZE <<
2978                                 E1000_PSRCTL_BSIZE3_SHIFT;
2979                 case 2:
2980                         psrctl |= PAGE_SIZE <<
2981                                 E1000_PSRCTL_BSIZE2_SHIFT;
2982                 case 1:
2983                         psrctl |= PAGE_SIZE >>
2984                                 E1000_PSRCTL_BSIZE1_SHIFT;
2985                         break;
2986                 }
2987
2988                 ew32(PSRCTL, psrctl);
2989         }
2990
2991         /* This is useful for sniffing bad packets. */
2992         if (adapter->netdev->features & NETIF_F_RXALL) {
2993                 /* UPE and MPE will be handled by normal PROMISC logic
2994                  * in e1000e_set_rx_mode */
2995                 rctl |= (E1000_RCTL_SBP | /* Receive bad packets */
2996                          E1000_RCTL_BAM | /* RX All Bcast Pkts */
2997                          E1000_RCTL_PMCF); /* RX All MAC Ctrl Pkts */
2998
2999                 rctl &= ~(E1000_RCTL_VFE | /* Disable VLAN filter */
3000                           E1000_RCTL_DPF | /* Allow filtered pause */
3001                           E1000_RCTL_CFIEN); /* Dis VLAN CFIEN Filter */
3002                 /* Do not mess with E1000_CTRL_VME, it affects transmit as well,
3003                  * and that breaks VLANs.
3004                  */
3005         }
3006
3007         ew32(RCTL, rctl);
3008         /* just started the receive unit, no need to restart */
3009         adapter->flags &= ~FLAG_RX_RESTART_NOW;
3010 }
3011
3012 /**
3013  * e1000_configure_rx - Configure Receive Unit after Reset
3014  * @adapter: board private structure
3015  *
3016  * Configure the Rx unit of the MAC after a reset.
3017  **/
3018 static void e1000_configure_rx(struct e1000_adapter *adapter)
3019 {
3020         struct e1000_hw *hw = &adapter->hw;
3021         struct e1000_ring *rx_ring = adapter->rx_ring;
3022         u64 rdba;
3023         u32 rdlen, rctl, rxcsum, ctrl_ext;
3024
3025         if (adapter->rx_ps_pages) {
3026                 /* this is a 32 byte descriptor */
3027                 rdlen = rx_ring->count *
3028                     sizeof(union e1000_rx_desc_packet_split);
3029                 adapter->clean_rx = e1000_clean_rx_irq_ps;
3030                 adapter->alloc_rx_buf = e1000_alloc_rx_buffers_ps;
3031         } else if (adapter->netdev->mtu > ETH_FRAME_LEN + ETH_FCS_LEN) {
3032                 rdlen = rx_ring->count * sizeof(union e1000_rx_desc_extended);
3033                 adapter->clean_rx = e1000_clean_jumbo_rx_irq;
3034                 adapter->alloc_rx_buf = e1000_alloc_jumbo_rx_buffers;
3035         } else {
3036                 rdlen = rx_ring->count * sizeof(union e1000_rx_desc_extended);
3037                 adapter->clean_rx = e1000_clean_rx_irq;
3038                 adapter->alloc_rx_buf = e1000_alloc_rx_buffers;
3039         }
3040
3041         /* disable receives while setting up the descriptors */
3042         rctl = er32(RCTL);
3043         if (!(adapter->flags2 & FLAG2_NO_DISABLE_RX))
3044                 ew32(RCTL, rctl & ~E1000_RCTL_EN);
3045         e1e_flush();
3046         usleep_range(10000, 20000);
3047
3048         if (adapter->flags2 & FLAG2_DMA_BURST) {
3049                 /*
3050                  * set the writeback threshold (only takes effect if the RDTR
3051                  * is set). set GRAN=1 and write back up to 0x4 worth, and
3052                  * enable prefetching of 0x20 Rx descriptors
3053                  * granularity = 01
3054                  * wthresh = 04,
3055                  * hthresh = 04,
3056                  * pthresh = 0x20
3057                  */
3058                 ew32(RXDCTL(0), E1000_RXDCTL_DMA_BURST_ENABLE);
3059                 ew32(RXDCTL(1), E1000_RXDCTL_DMA_BURST_ENABLE);
3060
3061                 /*
3062                  * override the delay timers for enabling bursting, only if
3063                  * the value was not set by the user via module options
3064                  */
3065                 if (adapter->rx_int_delay == DEFAULT_RDTR)
3066                         adapter->rx_int_delay = BURST_RDTR;
3067                 if (adapter->rx_abs_int_delay == DEFAULT_RADV)
3068                         adapter->rx_abs_int_delay = BURST_RADV;
3069         }
3070
3071         /* set the Receive Delay Timer Register */
3072         ew32(RDTR, adapter->rx_int_delay);
3073
3074         /* irq moderation */
3075         ew32(RADV, adapter->rx_abs_int_delay);
3076         if ((adapter->itr_setting != 0) && (adapter->itr != 0))
3077                 ew32(ITR, 1000000000 / (adapter->itr * 256));
3078
3079         ctrl_ext = er32(CTRL_EXT);
3080         /* Auto-Mask interrupts upon ICR access */
3081         ctrl_ext |= E1000_CTRL_EXT_IAME;
3082         ew32(IAM, 0xffffffff);
3083         ew32(CTRL_EXT, ctrl_ext);
3084         e1e_flush();
3085
3086         /*
3087          * Setup the HW Rx Head and Tail Descriptor Pointers and
3088          * the Base and Length of the Rx Descriptor Ring
3089          */
3090         rdba = rx_ring->dma;
3091         ew32(RDBAL(0), (rdba & DMA_BIT_MASK(32)));
3092         ew32(RDBAH(0), (rdba >> 32));
3093         ew32(RDLEN(0), rdlen);
3094         ew32(RDH(0), 0);
3095         ew32(RDT(0), 0);
3096         rx_ring->head = adapter->hw.hw_addr + E1000_RDH(0);
3097         rx_ring->tail = adapter->hw.hw_addr + E1000_RDT(0);
3098
3099         /* Enable Receive Checksum Offload for TCP and UDP */
3100         rxcsum = er32(RXCSUM);
3101         if (adapter->netdev->features & NETIF_F_RXCSUM) {
3102                 rxcsum |= E1000_RXCSUM_TUOFL;
3103
3104                 /*
3105                  * IPv4 payload checksum for UDP fragments must be
3106                  * used in conjunction with packet-split.
3107                  */
3108                 if (adapter->rx_ps_pages)
3109                         rxcsum |= E1000_RXCSUM_IPPCSE;
3110         } else {
3111                 rxcsum &= ~E1000_RXCSUM_TUOFL;
3112                 /* no need to clear IPPCSE as it defaults to 0 */
3113         }
3114         ew32(RXCSUM, rxcsum);
3115
3116         if (adapter->hw.mac.type == e1000_pch2lan) {
3117                 /*
3118                  * With jumbo frames, excessive C-state transition
3119                  * latencies result in dropped transactions.
3120                  */
3121                 if (adapter->netdev->mtu > ETH_DATA_LEN) {
3122                         u32 rxdctl = er32(RXDCTL(0));
3123                         ew32(RXDCTL(0), rxdctl | 0x3);
3124                         pm_qos_update_request(&adapter->netdev->pm_qos_req, 55);
3125                 } else {
3126                         pm_qos_update_request(&adapter->netdev->pm_qos_req,
3127                                               PM_QOS_DEFAULT_VALUE);
3128                 }
3129         }
3130
3131         /* Enable Receives */
3132         ew32(RCTL, rctl);
3133 }
3134
3135 /**
3136  * e1000e_write_mc_addr_list - write multicast addresses to MTA
3137  * @netdev: network interface device structure
3138  *
3139  * Writes multicast address list to the MTA hash table.
3140  * Returns: -ENOMEM on failure
3141  *                0 on no addresses written
3142  *                X on writing X addresses to MTA
3143  */
3144 static int e1000e_write_mc_addr_list(struct net_device *netdev)
3145 {
3146         struct e1000_adapter *adapter = netdev_priv(netdev);
3147         struct e1000_hw *hw = &adapter->hw;
3148         struct netdev_hw_addr *ha;
3149         u8 *mta_list;
3150         int i;
3151
3152         if (netdev_mc_empty(netdev)) {
3153                 /* nothing to program, so clear mc list */
3154                 hw->mac.ops.update_mc_addr_list(hw, NULL, 0);
3155                 return 0;
3156         }
3157
3158         mta_list = kzalloc(netdev_mc_count(netdev) * ETH_ALEN, GFP_ATOMIC);
3159         if (!mta_list)
3160                 return -ENOMEM;
3161
3162         /* update_mc_addr_list expects a packed array of only addresses. */
3163         i = 0;
3164         netdev_for_each_mc_addr(ha, netdev)
3165                 memcpy(mta_list + (i++ * ETH_ALEN), ha->addr, ETH_ALEN);
3166
3167         hw->mac.ops.update_mc_addr_list(hw, mta_list, i);
3168         kfree(mta_list);
3169
3170         return netdev_mc_count(netdev);
3171 }
3172
3173 /**
3174  * e1000e_write_uc_addr_list - write unicast addresses to RAR table
3175  * @netdev: network interface device structure
3176  *
3177  * Writes unicast address list to the RAR table.
3178  * Returns: -ENOMEM on failure/insufficient address space
3179  *                0 on no addresses written
3180  *                X on writing X addresses to the RAR table
3181  **/
3182 static int e1000e_write_uc_addr_list(struct net_device *netdev)
3183 {
3184         struct e1000_adapter *adapter = netdev_priv(netdev);
3185         struct e1000_hw *hw = &adapter->hw;
3186         unsigned int rar_entries = hw->mac.rar_entry_count;
3187         int count = 0;
3188
3189         /* save a rar entry for our hardware address */
3190         rar_entries--;
3191
3192         /* save a rar entry for the LAA workaround */
3193         if (adapter->flags & FLAG_RESET_OVERWRITES_LAA)
3194                 rar_entries--;
3195
3196         /* return ENOMEM indicating insufficient memory for addresses */
3197         if (netdev_uc_count(netdev) > rar_entries)
3198                 return -ENOMEM;
3199
3200         if (!netdev_uc_empty(netdev) && rar_entries) {
3201                 struct netdev_hw_addr *ha;
3202
3203                 /*
3204                  * write the addresses in reverse order to avoid write
3205                  * combining
3206                  */
3207                 netdev_for_each_uc_addr(ha, netdev) {
3208                         if (!rar_entries)
3209                                 break;
3210                         hw->mac.ops.rar_set(hw, ha->addr, rar_entries--);
3211                         count++;
3212                 }
3213         }
3214
3215         /* zero out the remaining RAR entries not used above */
3216         for (; rar_entries > 0; rar_entries--) {
3217                 ew32(RAH(rar_entries), 0);
3218                 ew32(RAL(rar_entries), 0);
3219         }
3220         e1e_flush();
3221
3222         return count;
3223 }
3224
3225 /**
3226  * e1000e_set_rx_mode - secondary unicast, Multicast and Promiscuous mode set
3227  * @netdev: network interface device structure
3228  *
3229  * The ndo_set_rx_mode entry point is called whenever the unicast or multicast
3230  * address list or the network interface flags are updated.  This routine is
3231  * responsible for configuring the hardware for proper unicast, multicast,
3232  * promiscuous mode, and all-multi behavior.
3233  **/
3234 static void e1000e_set_rx_mode(struct net_device *netdev)
3235 {
3236         struct e1000_adapter *adapter = netdev_priv(netdev);
3237         struct e1000_hw *hw = &adapter->hw;
3238         u32 rctl;
3239
3240         /* Check for Promiscuous and All Multicast modes */
3241         rctl = er32(RCTL);
3242
3243         /* clear the affected bits */
3244         rctl &= ~(E1000_RCTL_UPE | E1000_RCTL_MPE);
3245
3246         if (netdev->flags & IFF_PROMISC) {
3247                 rctl |= (E1000_RCTL_UPE | E1000_RCTL_MPE);
3248                 /* Do not hardware filter VLANs in promisc mode */
3249                 e1000e_vlan_filter_disable(adapter);
3250         } else {
3251                 int count;
3252
3253                 if (netdev->flags & IFF_ALLMULTI) {
3254                         rctl |= E1000_RCTL_MPE;
3255                 } else {
3256                         /*
3257                          * Write addresses to the MTA, if the attempt fails
3258                          * then we should just turn on promiscuous mode so
3259                          * that we can at least receive multicast traffic
3260                          */
3261                         count = e1000e_write_mc_addr_list(netdev);
3262                         if (count < 0)
3263                                 rctl |= E1000_RCTL_MPE;
3264                 }
3265                 e1000e_vlan_filter_enable(adapter);
3266                 /*
3267                  * Write addresses to available RAR registers, if there is not
3268                  * sufficient space to store all the addresses then enable
3269                  * unicast promiscuous mode
3270                  */
3271                 count = e1000e_write_uc_addr_list(netdev);
3272                 if (count < 0)
3273                         rctl |= E1000_RCTL_UPE;
3274         }
3275
3276         ew32(RCTL, rctl);
3277
3278         if (netdev->features & NETIF_F_HW_VLAN_RX)
3279                 e1000e_vlan_strip_enable(adapter);
3280         else
3281                 e1000e_vlan_strip_disable(adapter);
3282 }
3283
3284 static void e1000e_setup_rss_hash(struct e1000_adapter *adapter)
3285 {
3286         struct e1000_hw *hw = &adapter->hw;
3287         u32 mrqc, rxcsum;
3288         int i;
3289         static const u32 rsskey[10] = {
3290                 0xda565a6d, 0xc20e5b25, 0x3d256741, 0xb08fa343, 0xcb2bcad0,
3291                 0xb4307bae, 0xa32dcb77, 0x0cf23080, 0x3bb7426a, 0xfa01acbe
3292         };
3293
3294         /* Fill out hash function seed */
3295         for (i = 0; i < 10; i++)
3296                 ew32(RSSRK(i), rsskey[i]);
3297
3298         /* Direct all traffic to queue 0 */
3299         for (i = 0; i < 32; i++)
3300                 ew32(RETA(i), 0);
3301
3302         /*
3303          * Disable raw packet checksumming so that RSS hash is placed in
3304          * descriptor on writeback.
3305          */
3306         rxcsum = er32(RXCSUM);
3307         rxcsum |= E1000_RXCSUM_PCSD;
3308
3309         ew32(RXCSUM, rxcsum);
3310
3311         mrqc = (E1000_MRQC_RSS_FIELD_IPV4 |
3312                 E1000_MRQC_RSS_FIELD_IPV4_TCP |
3313                 E1000_MRQC_RSS_FIELD_IPV6 |
3314                 E1000_MRQC_RSS_FIELD_IPV6_TCP |
3315                 E1000_MRQC_RSS_FIELD_IPV6_TCP_EX);
3316
3317         ew32(MRQC, mrqc);
3318 }
3319
3320 /**
3321  * e1000_configure - configure the hardware for Rx and Tx
3322  * @adapter: private board structure
3323  **/
3324 static void e1000_configure(struct e1000_adapter *adapter)
3325 {
3326         struct e1000_ring *rx_ring = adapter->rx_ring;
3327
3328         e1000e_set_rx_mode(adapter->netdev);
3329
3330         e1000_restore_vlan(adapter);
3331         e1000_init_manageability_pt(adapter);
3332
3333         e1000_configure_tx(adapter);
3334
3335         if (adapter->netdev->features & NETIF_F_RXHASH)
3336                 e1000e_setup_rss_hash(adapter);
3337         e1000_setup_rctl(adapter);
3338         e1000_configure_rx(adapter);
3339         adapter->alloc_rx_buf(rx_ring, e1000_desc_unused(rx_ring), GFP_KERNEL);
3340 }
3341
3342 /**
3343  * e1000e_power_up_phy - restore link in case the phy was powered down
3344  * @adapter: address of board private structure
3345  *
3346  * The phy may be powered down to save power and turn off link when the
3347  * driver is unloaded and wake on lan is not enabled (among others)
3348  * *** this routine MUST be followed by a call to e1000e_reset ***
3349  **/
3350 void e1000e_power_up_phy(struct e1000_adapter *adapter)
3351 {
3352         if (adapter->hw.phy.ops.power_up)
3353                 adapter->hw.phy.ops.power_up(&adapter->hw);
3354
3355         adapter->hw.mac.ops.setup_link(&adapter->hw);
3356 }
3357
3358 /**
3359  * e1000_power_down_phy - Power down the PHY
3360  *
3361  * Power down the PHY so no link is implied when interface is down.
3362  * The PHY cannot be powered down if management or WoL is active.
3363  */
3364 static void e1000_power_down_phy(struct e1000_adapter *adapter)
3365 {
3366         /* WoL is enabled */
3367         if (adapter->wol)
3368                 return;
3369
3370         if (adapter->hw.phy.ops.power_down)
3371                 adapter->hw.phy.ops.power_down(&adapter->hw);
3372 }
3373
3374 /**
3375  * e1000e_reset - bring the hardware into a known good state
3376  *
3377  * This function boots the hardware and enables some settings that
3378  * require a configuration cycle of the hardware - those cannot be
3379  * set/changed during runtime. After reset the device needs to be
3380  * properly configured for Rx, Tx etc.
3381  */
3382 void e1000e_reset(struct e1000_adapter *adapter)
3383 {
3384         struct e1000_mac_info *mac = &adapter->hw.mac;
3385         struct e1000_fc_info *fc = &adapter->hw.fc;
3386         struct e1000_hw *hw = &adapter->hw;
3387         u32 tx_space, min_tx_space, min_rx_space;
3388         u32 pba = adapter->pba;
3389         u16 hwm;
3390
3391         /* reset Packet Buffer Allocation to default */
3392         ew32(PBA, pba);
3393
3394         if (adapter->max_frame_size > ETH_FRAME_LEN + ETH_FCS_LEN) {
3395                 /*
3396                  * To maintain wire speed transmits, the Tx FIFO should be
3397                  * large enough to accommodate two full transmit packets,
3398                  * rounded up to the next 1KB and expressed in KB.  Likewise,
3399                  * the Rx FIFO should be large enough to accommodate at least
3400                  * one full receive packet and is similarly rounded up and
3401                  * expressed in KB.
3402                  */
3403                 pba = er32(PBA);
3404                 /* upper 16 bits has Tx packet buffer allocation size in KB */
3405                 tx_space = pba >> 16;
3406                 /* lower 16 bits has Rx packet buffer allocation size in KB */
3407                 pba &= 0xffff;
3408                 /*
3409                  * the Tx fifo also stores 16 bytes of information about the Tx
3410                  * but don't include ethernet FCS because hardware appends it
3411                  */
3412                 min_tx_space = (adapter->max_frame_size +
3413                                 sizeof(struct e1000_tx_desc) -
3414                                 ETH_FCS_LEN) * 2;
3415                 min_tx_space = ALIGN(min_tx_space, 1024);
3416                 min_tx_space >>= 10;
3417                 /* software strips receive CRC, so leave room for it */
3418                 min_rx_space = adapter->max_frame_size;
3419                 min_rx_space = ALIGN(min_rx_space, 1024);
3420                 min_rx_space >>= 10;
3421
3422                 /*
3423                  * If current Tx allocation is less than the min Tx FIFO size,
3424                  * and the min Tx FIFO size is less than the current Rx FIFO
3425                  * allocation, take space away from current Rx allocation
3426                  */
3427                 if ((tx_space < min_tx_space) &&
3428                     ((min_tx_space - tx_space) < pba)) {
3429                         pba -= min_tx_space - tx_space;
3430
3431                         /*
3432                          * if short on Rx space, Rx wins and must trump Tx
3433                          * adjustment or use Early Receive if available
3434                          */
3435                         if (pba < min_rx_space)
3436                                 pba = min_rx_space;
3437                 }
3438
3439                 ew32(PBA, pba);
3440         }
3441
3442         /*
3443          * flow control settings
3444          *
3445          * The high water mark must be low enough to fit one full frame
3446          * (or the size used for early receive) above it in the Rx FIFO.
3447          * Set it to the lower of:
3448          * - 90% of the Rx FIFO size, and
3449          * - the full Rx FIFO size minus one full frame
3450          */
3451         if (adapter->flags & FLAG_DISABLE_FC_PAUSE_TIME)
3452                 fc->pause_time = 0xFFFF;
3453         else
3454                 fc->pause_time = E1000_FC_PAUSE_TIME;
3455         fc->send_xon = true;
3456         fc->current_mode = fc->requested_mode;
3457
3458         switch (hw->mac.type) {
3459         case e1000_ich9lan:
3460         case e1000_ich10lan:
3461                 if (adapter->netdev->mtu > ETH_DATA_LEN) {
3462                         pba = 14;
3463                         ew32(PBA, pba);
3464                         fc->high_water = 0x2800;
3465                         fc->low_water = fc->high_water - 8;
3466                         break;
3467                 }
3468                 /* fall-through */
3469         default:
3470                 hwm = min(((pba << 10) * 9 / 10),
3471                           ((pba << 10) - adapter->max_frame_size));
3472
3473                 fc->high_water = hwm & E1000_FCRTH_RTH; /* 8-byte granularity */
3474                 fc->low_water = fc->high_water - 8;
3475                 break;
3476         case e1000_pchlan:
3477                 /*
3478                  * Workaround PCH LOM adapter hangs with certain network
3479                  * loads.  If hangs persist, try disabling Tx flow control.
3480                  */
3481                 if (adapter->netdev->mtu > ETH_DATA_LEN) {
3482                         fc->high_water = 0x3500;
3483                         fc->low_water  = 0x1500;
3484                 } else {
3485                         fc->high_water = 0x5000;
3486                         fc->low_water  = 0x3000;
3487                 }
3488                 fc->refresh_time = 0x1000;
3489                 break;
3490         case e1000_pch2lan:
3491         case e1000_pch_lpt:
3492                 fc->high_water = 0x05C20;
3493                 fc->low_water = 0x05048;
3494                 fc->pause_time = 0x0650;
3495                 fc->refresh_time = 0x0400;
3496                 if (adapter->netdev->mtu > ETH_DATA_LEN) {
3497                         pba = 14;
3498                         ew32(PBA, pba);
3499                 }
3500                 break;
3501         }
3502
3503         /*
3504          * Disable Adaptive Interrupt Moderation if 2 full packets cannot
3505          * fit in receive buffer.
3506          */
3507         if (adapter->itr_setting & 0x3) {
3508                 if ((adapter->max_frame_size * 2) > (pba << 10)) {
3509                         if (!(adapter->flags2 & FLAG2_DISABLE_AIM)) {
3510                                 dev_info(&adapter->pdev->dev,
3511                                         "Interrupt Throttle Rate turned off\n");
3512                                 adapter->flags2 |= FLAG2_DISABLE_AIM;
3513                                 ew32(ITR, 0);
3514                         }
3515                 } else if (adapter->flags2 & FLAG2_DISABLE_AIM) {
3516                         dev_info(&adapter->pdev->dev,
3517                                  "Interrupt Throttle Rate turned on\n");
3518                         adapter->flags2 &= ~FLAG2_DISABLE_AIM;
3519                         adapter->itr = 20000;
3520                         ew32(ITR, 1000000000 / (adapter->itr * 256));
3521                 }
3522         }
3523
3524         /* Allow time for pending master requests to run */
3525         mac->ops.reset_hw(hw);
3526
3527         /*
3528          * For parts with AMT enabled, let the firmware know
3529          * that the network interface is in control
3530          */
3531         if (adapter->flags & FLAG_HAS_AMT)
3532                 e1000e_get_hw_control(adapter);
3533
3534         ew32(WUC, 0);
3535
3536         if (mac->ops.init_hw(hw))
3537                 e_err("Hardware Error\n");
3538
3539         e1000_update_mng_vlan(adapter);
3540
3541         /* Enable h/w to recognize an 802.1Q VLAN Ethernet packet */
3542         ew32(VET, ETH_P_8021Q);
3543
3544         e1000e_reset_adaptive(hw);
3545
3546         if (!netif_running(adapter->netdev) &&
3547             !test_bit(__E1000_TESTING, &adapter->state)) {
3548                 e1000_power_down_phy(adapter);
3549                 return;
3550         }
3551
3552         e1000_get_phy_info(hw);
3553
3554         if ((adapter->flags & FLAG_HAS_SMART_POWER_DOWN) &&
3555             !(adapter->flags & FLAG_SMART_POWER_DOWN)) {
3556                 u16 phy_data = 0;
3557                 /*
3558                  * speed up time to link by disabling smart power down, ignore
3559                  * the return value of this function because there is nothing
3560                  * different we would do if it failed
3561                  */
3562                 e1e_rphy(hw, IGP02E1000_PHY_POWER_MGMT, &phy_data);
3563                 phy_data &= ~IGP02E1000_PM_SPD;
3564                 e1e_wphy(hw, IGP02E1000_PHY_POWER_MGMT, phy_data);
3565         }
3566 }
3567
3568 int e1000e_up(struct e1000_adapter *adapter)
3569 {
3570         struct e1000_hw *hw = &adapter->hw;
3571
3572         /* hardware has been reset, we need to reload some things */
3573         e1000_configure(adapter);
3574
3575         clear_bit(__E1000_DOWN, &adapter->state);
3576
3577         if (adapter->msix_entries)
3578                 e1000_configure_msix(adapter);
3579         e1000_irq_enable(adapter);
3580
3581         netif_start_queue(adapter->netdev);
3582
3583         /* fire a link change interrupt to start the watchdog */
3584         if (adapter->msix_entries)
3585                 ew32(ICS, E1000_ICS_LSC | E1000_ICR_OTHER);
3586         else
3587                 ew32(ICS, E1000_ICS_LSC);
3588
3589         return 0;
3590 }
3591
3592 static void e1000e_flush_descriptors(struct e1000_adapter *adapter)
3593 {
3594         struct e1000_hw *hw = &adapter->hw;
3595
3596         if (!(adapter->flags2 & FLAG2_DMA_BURST))
3597                 return;
3598
3599         /* flush pending descriptor writebacks to memory */
3600         ew32(TIDV, adapter->tx_int_delay | E1000_TIDV_FPD);
3601         ew32(RDTR, adapter->rx_int_delay | E1000_RDTR_FPD);
3602
3603         /* execute the writes immediately */
3604         e1e_flush();
3605
3606         /*
3607          * due to rare timing issues, write to TIDV/RDTR again to ensure the
3608          * write is successful
3609          */
3610         ew32(TIDV, adapter->tx_int_delay | E1000_TIDV_FPD);
3611         ew32(RDTR, adapter->rx_int_delay | E1000_RDTR_FPD);
3612
3613         /* execute the writes immediately */
3614         e1e_flush();
3615 }
3616
3617 static void e1000e_update_stats(struct e1000_adapter *adapter);
3618
3619 void e1000e_down(struct e1000_adapter *adapter)
3620 {
3621         struct net_device *netdev = adapter->netdev;
3622         struct e1000_hw *hw = &adapter->hw;
3623         u32 tctl, rctl;
3624
3625         /*
3626          * signal that we're down so the interrupt handler does not
3627          * reschedule our watchdog timer
3628          */
3629         set_bit(__E1000_DOWN, &adapter->state);
3630
3631         /* disable receives in the hardware */
3632         rctl = er32(RCTL);
3633         if (!(adapter->flags2 & FLAG2_NO_DISABLE_RX))
3634                 ew32(RCTL, rctl & ~E1000_RCTL_EN);
3635         /* flush and sleep below */
3636
3637         netif_stop_queue(netdev);
3638
3639         /* disable transmits in the hardware */
3640         tctl = er32(TCTL);
3641         tctl &= ~E1000_TCTL_EN;
3642         ew32(TCTL, tctl);
3643
3644         /* flush both disables and wait for them to finish */
3645         e1e_flush();
3646         usleep_range(10000, 20000);
3647
3648         e1000_irq_disable(adapter);
3649
3650         del_timer_sync(&adapter->watchdog_timer);
3651         del_timer_sync(&adapter->phy_info_timer);
3652
3653         netif_carrier_off(netdev);
3654
3655         spin_lock(&adapter->stats64_lock);
3656         e1000e_update_stats(adapter);
3657         spin_unlock(&adapter->stats64_lock);
3658
3659         e1000e_flush_descriptors(adapter);
3660         e1000_clean_tx_ring(adapter->tx_ring);
3661         e1000_clean_rx_ring(adapter->rx_ring);
3662
3663         adapter->link_speed = 0;
3664         adapter->link_duplex = 0;
3665
3666         if (!pci_channel_offline(adapter->pdev))
3667                 e1000e_reset(adapter);
3668
3669         /*
3670          * TODO: for power management, we could drop the link and
3671          * pci_disable_device here.
3672          */
3673 }
3674
3675 void e1000e_reinit_locked(struct e1000_adapter *adapter)
3676 {
3677         might_sleep();
3678         while (test_and_set_bit(__E1000_RESETTING, &adapter->state))
3679                 usleep_range(1000, 2000);
3680         e1000e_down(adapter);
3681         e1000e_up(adapter);
3682         clear_bit(__E1000_RESETTING, &adapter->state);
3683 }
3684
3685 /**
3686  * e1000_sw_init - Initialize general software structures (struct e1000_adapter)
3687  * @adapter: board private structure to initialize
3688  *
3689  * e1000_sw_init initializes the Adapter private data structure.
3690  * Fields are initialized based on PCI device information and
3691  * OS network device settings (MTU size).
3692  **/
3693 static int __devinit e1000_sw_init(struct e1000_adapter *adapter)
3694 {
3695         struct net_device *netdev = adapter->netdev;
3696
3697         adapter->rx_buffer_len = ETH_FRAME_LEN + VLAN_HLEN + ETH_FCS_LEN;
3698         adapter->rx_ps_bsize0 = 128;
3699         adapter->max_frame_size = netdev->mtu + ETH_HLEN + ETH_FCS_LEN;
3700         adapter->min_frame_size = ETH_ZLEN + ETH_FCS_LEN;
3701         adapter->tx_ring_count = E1000_DEFAULT_TXD;
3702         adapter->rx_ring_count = E1000_DEFAULT_RXD;
3703
3704         spin_lock_init(&adapter->stats64_lock);
3705
3706         e1000e_set_interrupt_capability(adapter);
3707
3708         if (e1000_alloc_queues(adapter))
3709                 return -ENOMEM;
3710
3711         /* Explicitly disable IRQ since the NIC can be in any state. */
3712         e1000_irq_disable(adapter);
3713
3714         set_bit(__E1000_DOWN, &adapter->state);
3715         return 0;
3716 }
3717
3718 /**
3719  * e1000_intr_msi_test - Interrupt Handler
3720  * @irq: interrupt number
3721  * @data: pointer to a network interface device structure
3722  **/
3723 static irqreturn_t e1000_intr_msi_test(int irq, void *data)
3724 {
3725         struct net_device *netdev = data;
3726         struct e1000_adapter *adapter = netdev_priv(netdev);
3727         struct e1000_hw *hw = &adapter->hw;
3728         u32 icr = er32(ICR);
3729
3730         e_dbg("icr is %08X\n", icr);
3731         if (icr & E1000_ICR_RXSEQ) {
3732                 adapter->flags &= ~FLAG_MSI_TEST_FAILED;
3733                 wmb();
3734         }
3735
3736         return IRQ_HANDLED;
3737 }
3738
3739 /**
3740  * e1000_test_msi_interrupt - Returns 0 for successful test
3741  * @adapter: board private struct
3742  *
3743  * code flow taken from tg3.c
3744  **/
3745 static int e1000_test_msi_interrupt(struct e1000_adapter *adapter)
3746 {
3747         struct net_device *netdev = adapter->netdev;
3748         struct e1000_hw *hw = &adapter->hw;
3749         int err;
3750
3751         /* poll_enable hasn't been called yet, so don't need disable */
3752         /* clear any pending events */
3753         er32(ICR);
3754
3755         /* free the real vector and request a test handler */
3756         e1000_free_irq(adapter);
3757         e1000e_reset_interrupt_capability(adapter);
3758
3759         /* Assume that the test fails, if it succeeds then the test
3760          * MSI irq handler will unset this flag */
3761         adapter->flags |= FLAG_MSI_TEST_FAILED;
3762
3763         err = pci_enable_msi(adapter->pdev);
3764         if (err)
3765                 goto msi_test_failed;
3766
3767         err = request_irq(adapter->pdev->irq, e1000_intr_msi_test, 0,
3768                           netdev->name, netdev);
3769         if (err) {
3770                 pci_disable_msi(adapter->pdev);
3771                 goto msi_test_failed;
3772         }
3773
3774         wmb();
3775
3776         e1000_irq_enable(adapter);
3777
3778         /* fire an unusual interrupt on the test handler */
3779         ew32(ICS, E1000_ICS_RXSEQ);
3780         e1e_flush();
3781         msleep(100);
3782
3783         e1000_irq_disable(adapter);
3784
3785         rmb();
3786
3787         if (adapter->flags & FLAG_MSI_TEST_FAILED) {
3788                 adapter->int_mode = E1000E_INT_MODE_LEGACY;
3789                 e_info("MSI interrupt test failed, using legacy interrupt.\n");
3790         } else {
3791                 e_dbg("MSI interrupt test succeeded!\n");
3792         }
3793
3794         free_irq(adapter->pdev->irq, netdev);
3795         pci_disable_msi(adapter->pdev);
3796
3797 msi_test_failed:
3798         e1000e_set_interrupt_capability(adapter);
3799         return e1000_request_irq(adapter);
3800 }
3801
3802 /**
3803  * e1000_test_msi - Returns 0 if MSI test succeeds or INTx mode is restored
3804  * @adapter: board private struct
3805  *
3806  * code flow taken from tg3.c, called with e1000 interrupts disabled.
3807  **/
3808 static int e1000_test_msi(struct e1000_adapter *adapter)
3809 {
3810         int err;
3811         u16 pci_cmd;
3812
3813         if (!(adapter->flags & FLAG_MSI_ENABLED))
3814                 return 0;
3815
3816         /* disable SERR in case the MSI write causes a master abort */
3817         pci_read_config_word(adapter->pdev, PCI_COMMAND, &pci_cmd);
3818         if (pci_cmd & PCI_COMMAND_SERR)
3819                 pci_write_config_word(adapter->pdev, PCI_COMMAND,
3820                                       pci_cmd & ~PCI_COMMAND_SERR);
3821
3822         err = e1000_test_msi_interrupt(adapter);
3823
3824         /* re-enable SERR */
3825         if (pci_cmd & PCI_COMMAND_SERR) {
3826                 pci_read_config_word(adapter->pdev, PCI_COMMAND, &pci_cmd);
3827                 pci_cmd |= PCI_COMMAND_SERR;
3828                 pci_write_config_word(adapter->pdev, PCI_COMMAND, pci_cmd);
3829         }
3830
3831         return err;
3832 }
3833
3834 /**
3835  * e1000_open - Called when a network interface is made active
3836  * @netdev: network interface device structure
3837  *
3838  * Returns 0 on success, negative value on failure
3839  *
3840  * The open entry point is called when a network interface is made
3841  * active by the system (IFF_UP).  At this point all resources needed
3842  * for transmit and receive operations are allocated, the interrupt
3843  * handler is registered with the OS, the watchdog timer is started,
3844  * and the stack is notified that the interface is ready.
3845  **/
3846 static int e1000_open(struct net_device *netdev)
3847 {
3848         struct e1000_adapter *adapter = netdev_priv(netdev);
3849         struct e1000_hw *hw = &adapter->hw;
3850         struct pci_dev *pdev = adapter->pdev;
3851         int err;
3852
3853         /* disallow open during test */
3854         if (test_bit(__E1000_TESTING, &adapter->state))
3855                 return -EBUSY;
3856
3857         pm_runtime_get_sync(&pdev->dev);
3858
3859         netif_carrier_off(netdev);
3860
3861         /* allocate transmit descriptors */
3862         err = e1000e_setup_tx_resources(adapter->tx_ring);
3863         if (err)
3864                 goto err_setup_tx;
3865
3866         /* allocate receive descriptors */
3867         err = e1000e_setup_rx_resources(adapter->rx_ring);
3868         if (err)
3869                 goto err_setup_rx;
3870
3871         /*
3872          * If AMT is enabled, let the firmware know that the network
3873          * interface is now open and reset the part to a known state.
3874          */
3875         if (adapter->flags & FLAG_HAS_AMT) {
3876                 e1000e_get_hw_control(adapter);
3877                 e1000e_reset(adapter);
3878         }
3879
3880         e1000e_power_up_phy(adapter);
3881
3882         adapter->mng_vlan_id = E1000_MNG_VLAN_NONE;
3883         if ((adapter->hw.mng_cookie.status &
3884              E1000_MNG_DHCP_COOKIE_STATUS_VLAN))
3885                 e1000_update_mng_vlan(adapter);
3886
3887         /* DMA latency requirement to workaround jumbo issue */
3888         if (adapter->hw.mac.type == e1000_pch2lan)
3889                 pm_qos_add_request(&adapter->netdev->pm_qos_req,
3890                                    PM_QOS_CPU_DMA_LATENCY,
3891                                    PM_QOS_DEFAULT_VALUE);
3892
3893         /*
3894          * before we allocate an interrupt, we must be ready to handle it.
3895          * Setting DEBUG_SHIRQ in the kernel makes it fire an interrupt
3896          * as soon as we call pci_request_irq, so we have to setup our
3897          * clean_rx handler before we do so.
3898          */
3899         e1000_configure(adapter);
3900
3901         err = e1000_request_irq(adapter);
3902         if (err)
3903                 goto err_req_irq;
3904
3905         /*
3906          * Work around PCIe errata with MSI interrupts causing some chipsets to
3907          * ignore e1000e MSI messages, which means we need to test our MSI
3908          * interrupt now
3909          */
3910         if (adapter->int_mode != E1000E_INT_MODE_LEGACY) {
3911                 err = e1000_test_msi(adapter);
3912                 if (err) {
3913                         e_err("Interrupt allocation failed\n");
3914                         goto err_req_irq;
3915                 }
3916         }
3917
3918         /* From here on the code is the same as e1000e_up() */
3919         clear_bit(__E1000_DOWN, &adapter->state);
3920
3921         napi_enable(&adapter->napi);
3922
3923         e1000_irq_enable(adapter);
3924
3925         adapter->tx_hang_recheck = false;
3926         netif_start_queue(netdev);
3927
3928         adapter->idle_check = true;
3929         pm_runtime_put(&pdev->dev);
3930
3931         /* fire a link status change interrupt to start the watchdog */
3932         if (adapter->msix_entries)
3933                 ew32(ICS, E1000_ICS_LSC | E1000_ICR_OTHER);
3934         else
3935                 ew32(ICS, E1000_ICS_LSC);
3936
3937         return 0;
3938
3939 err_req_irq:
3940         e1000e_release_hw_control(adapter);
3941         e1000_power_down_phy(adapter);
3942         e1000e_free_rx_resources(adapter->rx_ring);
3943 err_setup_rx:
3944         e1000e_free_tx_resources(adapter->tx_ring);
3945 err_setup_tx:
3946         e1000e_reset(adapter);
3947         pm_runtime_put_sync(&pdev->dev);
3948
3949         return err;
3950 }
3951
3952 /**
3953  * e1000_close - Disables a network interface
3954  * @netdev: network interface device structure
3955  *
3956  * Returns 0, this is not allowed to fail
3957  *
3958  * The close entry point is called when an interface is de-activated
3959  * by the OS.  The hardware is still under the drivers control, but
3960  * needs to be disabled.  A global MAC reset is issued to stop the
3961  * hardware, and all transmit and receive resources are freed.
3962  **/
3963 static int e1000_close(struct net_device *netdev)
3964 {
3965         struct e1000_adapter *adapter = netdev_priv(netdev);
3966         struct pci_dev *pdev = adapter->pdev;
3967         int count = E1000_CHECK_RESET_COUNT;
3968
3969         while (test_bit(__E1000_RESETTING, &adapter->state) && count--)
3970                 usleep_range(10000, 20000);
3971
3972         WARN_ON(test_bit(__E1000_RESETTING, &adapter->state));
3973
3974         pm_runtime_get_sync(&pdev->dev);
3975
3976         napi_disable(&adapter->napi);
3977
3978         if (!test_bit(__E1000_DOWN, &adapter->state)) {
3979                 e1000e_down(adapter);
3980                 e1000_free_irq(adapter);
3981         }
3982         e1000_power_down_phy(adapter);
3983
3984         e1000e_free_tx_resources(adapter->tx_ring);
3985         e1000e_free_rx_resources(adapter->rx_ring);
3986
3987         /*
3988          * kill manageability vlan ID if supported, but not if a vlan with
3989          * the same ID is registered on the host OS (let 8021q kill it)
3990          */
3991         if (adapter->hw.mng_cookie.status &
3992             E1000_MNG_DHCP_COOKIE_STATUS_VLAN)
3993                 e1000_vlan_rx_kill_vid(netdev, adapter->mng_vlan_id);
3994
3995         /*
3996          * If AMT is enabled, let the firmware know that the network
3997          * interface is now closed
3998          */
3999         if ((adapter->flags & FLAG_HAS_AMT) &&
4000             !test_bit(__E1000_TESTING, &adapter->state))
4001                 e1000e_release_hw_control(adapter);
4002
4003         if (adapter->hw.mac.type == e1000_pch2lan)
4004                 pm_qos_remove_request(&adapter->netdev->pm_qos_req);
4005
4006         pm_runtime_put_sync(&pdev->dev);
4007
4008         return 0;
4009 }
4010 /**
4011  * e1000_set_mac - Change the Ethernet Address of the NIC
4012  * @netdev: network interface device structure
4013  * @p: pointer to an address structure
4014  *
4015  * Returns 0 on success, negative on failure
4016  **/
4017 static int e1000_set_mac(struct net_device *netdev, void *p)
4018 {
4019         struct e1000_adapter *adapter = netdev_priv(netdev);
4020         struct e1000_hw *hw = &adapter->hw;
4021         struct sockaddr *addr = p;
4022
4023         if (!is_valid_ether_addr(addr->sa_data))
4024                 return -EADDRNOTAVAIL;
4025
4026         memcpy(netdev->dev_addr, addr->sa_data, netdev->addr_len);
4027         memcpy(adapter->hw.mac.addr, addr->sa_data, netdev->addr_len);
4028
4029         hw->mac.ops.rar_set(&adapter->hw, adapter->hw.mac.addr, 0);
4030
4031         if (adapter->flags & FLAG_RESET_OVERWRITES_LAA) {
4032                 /* activate the work around */
4033                 e1000e_set_laa_state_82571(&adapter->hw, 1);
4034
4035                 /*
4036                  * Hold a copy of the LAA in RAR[14] This is done so that
4037                  * between the time RAR[0] gets clobbered  and the time it
4038                  * gets fixed (in e1000_watchdog), the actual LAA is in one
4039                  * of the RARs and no incoming packets directed to this port
4040                  * are dropped. Eventually the LAA will be in RAR[0] and
4041                  * RAR[14]
4042                  */
4043                 hw->mac.ops.rar_set(&adapter->hw, adapter->hw.mac.addr,
4044                                     adapter->hw.mac.rar_entry_count - 1);
4045         }
4046
4047         return 0;
4048 }
4049
4050 /**
4051  * e1000e_update_phy_task - work thread to update phy
4052  * @work: pointer to our work struct
4053  *
4054  * this worker thread exists because we must acquire a
4055  * semaphore to read the phy, which we could msleep while
4056  * waiting for it, and we can't msleep in a timer.
4057  **/
4058 static void e1000e_update_phy_task(struct work_struct *work)
4059 {
4060         struct e1000_adapter *adapter = container_of(work,
4061                                         struct e1000_adapter, update_phy_task);
4062
4063         if (test_bit(__E1000_DOWN, &adapter->state))
4064                 return;
4065
4066         e1000_get_phy_info(&adapter->hw);
4067 }
4068
4069 /*
4070  * Need to wait a few seconds after link up to get diagnostic information from
4071  * the phy
4072  */
4073 static void e1000_update_phy_info(unsigned long data)
4074 {
4075         struct e1000_adapter *adapter = (struct e1000_adapter *) data;
4076
4077         if (test_bit(__E1000_DOWN, &adapter->state))
4078                 return;
4079
4080         schedule_work(&adapter->update_phy_task);
4081 }
4082
4083 /**
4084  * e1000e_update_phy_stats - Update the PHY statistics counters
4085  * @adapter: board private structure
4086  *
4087  * Read/clear the upper 16-bit PHY registers and read/accumulate lower
4088  **/
4089 static void e1000e_update_phy_stats(struct e1000_adapter *adapter)
4090 {
4091         struct e1000_hw *hw = &adapter->hw;
4092         s32 ret_val;
4093         u16 phy_data;
4094
4095         ret_val = hw->phy.ops.acquire(hw);
4096         if (ret_val)
4097                 return;
4098
4099         /*
4100          * A page set is expensive so check if already on desired page.
4101          * If not, set to the page with the PHY status registers.
4102          */
4103         hw->phy.addr = 1;
4104         ret_val = e1000e_read_phy_reg_mdic(hw, IGP01E1000_PHY_PAGE_SELECT,
4105                                            &phy_data);
4106         if (ret_val)
4107                 goto release;
4108         if (phy_data != (HV_STATS_PAGE << IGP_PAGE_SHIFT)) {
4109                 ret_val = hw->phy.ops.set_page(hw,
4110                                                HV_STATS_PAGE << IGP_PAGE_SHIFT);
4111                 if (ret_val)
4112                         goto release;
4113         }
4114
4115         /* Single Collision Count */
4116         hw->phy.ops.read_reg_page(hw, HV_SCC_UPPER, &phy_data);
4117         ret_val = hw->phy.ops.read_reg_page(hw, HV_SCC_LOWER, &phy_data);
4118         if (!ret_val)
4119                 adapter->stats.scc += phy_data;
4120
4121         /* Excessive Collision Count */
4122         hw->phy.ops.read_reg_page(hw, HV_ECOL_UPPER, &phy_data);
4123         ret_val = hw->phy.ops.read_reg_page(hw, HV_ECOL_LOWER, &phy_data);
4124         if (!ret_val)
4125                 adapter->stats.ecol += phy_data;
4126
4127         /* Multiple Collision Count */
4128         hw->phy.ops.read_reg_page(hw, HV_MCC_UPPER, &phy_data);
4129         ret_val = hw->phy.ops.read_reg_page(hw, HV_MCC_LOWER, &phy_data);
4130         if (!ret_val)
4131                 adapter->stats.mcc += phy_data;
4132
4133         /* Late Collision Count */
4134         hw->phy.ops.read_reg_page(hw, HV_LATECOL_UPPER, &phy_data);
4135         ret_val = hw->phy.ops.read_reg_page(hw, HV_LATECOL_LOWER, &phy_data);
4136         if (!ret_val)
4137                 adapter->stats.latecol += phy_data;
4138
4139         /* Collision Count - also used for adaptive IFS */
4140         hw->phy.ops.read_reg_page(hw, HV_COLC_UPPER, &phy_data);
4141         ret_val = hw->phy.ops.read_reg_page(hw, HV_COLC_LOWER, &phy_data);
4142         if (!ret_val)
4143                 hw->mac.collision_delta = phy_data;
4144
4145         /* Defer Count */
4146         hw->phy.ops.read_reg_page(hw, HV_DC_UPPER, &phy_data);
4147         ret_val = hw->phy.ops.read_reg_page(hw, HV_DC_LOWER, &phy_data);
4148         if (!ret_val)
4149                 adapter->stats.dc += phy_data;
4150
4151         /* Transmit with no CRS */
4152         hw->phy.ops.read_reg_page(hw, HV_TNCRS_UPPER, &phy_data);
4153         ret_val = hw->phy.ops.read_reg_page(hw, HV_TNCRS_LOWER, &phy_data);
4154         if (!ret_val)
4155                 adapter->stats.tncrs += phy_data;
4156
4157 release:
4158         hw->phy.ops.release(hw);
4159 }
4160
4161 /**
4162  * e1000e_update_stats - Update the board statistics counters
4163  * @adapter: board private structure
4164  **/
4165 static void e1000e_update_stats(struct e1000_adapter *adapter)
4166 {
4167         struct net_device *netdev = adapter->netdev;
4168         struct e1000_hw *hw = &adapter->hw;
4169         struct pci_dev *pdev = adapter->pdev;
4170
4171         /*
4172          * Prevent stats update while adapter is being reset, or if the pci
4173          * connection is down.
4174          */
4175         if (adapter->link_speed == 0)
4176                 return;
4177         if (pci_channel_offline(pdev))
4178                 return;
4179
4180         adapter->stats.crcerrs += er32(CRCERRS);
4181         adapter->stats.gprc += er32(GPRC);
4182         adapter->stats.gorc += er32(GORCL);
4183         er32(GORCH); /* Clear gorc */
4184         adapter->stats.bprc += er32(BPRC);
4185         adapter->stats.mprc += er32(MPRC);
4186         adapter->stats.roc += er32(ROC);
4187
4188         adapter->stats.mpc += er32(MPC);
4189
4190         /* Half-duplex statistics */
4191         if (adapter->link_duplex == HALF_DUPLEX) {
4192                 if (adapter->flags2 & FLAG2_HAS_PHY_STATS) {
4193                         e1000e_update_phy_stats(adapter);
4194                 } else {
4195                         adapter->stats.scc += er32(SCC);
4196                         adapter->stats.ecol += er32(ECOL);
4197                         adapter->stats.mcc += er32(MCC);
4198                         adapter->stats.latecol += er32(LATECOL);
4199                         adapter->stats.dc += er32(DC);
4200
4201                         hw->mac.collision_delta = er32(COLC);
4202
4203                         if ((hw->mac.type != e1000_82574) &&
4204                             (hw->mac.type != e1000_82583))
4205                                 adapter->stats.tncrs += er32(TNCRS);
4206                 }
4207                 adapter->stats.colc += hw->mac.collision_delta;
4208         }
4209
4210         adapter->stats.xonrxc += er32(XONRXC);
4211         adapter->stats.xontxc += er32(XONTXC);
4212         adapter->stats.xoffrxc += er32(XOFFRXC);
4213         adapter->stats.xofftxc += er32(XOFFTXC);
4214         adapter->stats.gptc += er32(GPTC);
4215         adapter->stats.gotc += er32(GOTCL);
4216         er32(GOTCH); /* Clear gotc */
4217         adapter->stats.rnbc += er32(RNBC);
4218         adapter->stats.ruc += er32(RUC);
4219
4220         adapter->stats.mptc += er32(MPTC);
4221         adapter->stats.bptc += er32(BPTC);
4222
4223         /* used for adaptive IFS */
4224
4225         hw->mac.tx_packet_delta = er32(TPT);
4226         adapter->stats.tpt += hw->mac.tx_packet_delta;
4227
4228         adapter->stats.algnerrc += er32(ALGNERRC);
4229         adapter->stats.rxerrc += er32(RXERRC);
4230         adapter->stats.cexterr += er32(CEXTERR);
4231         adapter->stats.tsctc += er32(TSCTC);
4232         adapter->stats.tsctfc += er32(TSCTFC);
4233
4234         /* Fill out the OS statistics structure */
4235         netdev->stats.multicast = adapter->stats.mprc;
4236         netdev->stats.collisions = adapter->stats.colc;
4237
4238         /* Rx Errors */
4239
4240         /*
4241          * RLEC on some newer hardware can be incorrect so build
4242          * our own version based on RUC and ROC
4243          */
4244         netdev->stats.rx_errors = adapter->stats.rxerrc +
4245                 adapter->stats.crcerrs + adapter->stats.algnerrc +
4246                 adapter->stats.ruc + adapter->stats.roc +
4247                 adapter->stats.cexterr;
4248         netdev->stats.rx_length_errors = adapter->stats.ruc +
4249                                               adapter->stats.roc;
4250         netdev->stats.rx_crc_errors = adapter->stats.crcerrs;
4251         netdev->stats.rx_frame_errors = adapter->stats.algnerrc;
4252         netdev->stats.rx_missed_errors = adapter->stats.mpc;
4253
4254         /* Tx Errors */
4255         netdev->stats.tx_errors = adapter->stats.ecol +
4256                                        adapter->stats.latecol;
4257         netdev->stats.tx_aborted_errors = adapter->stats.ecol;
4258         netdev->stats.tx_window_errors = adapter->stats.latecol;
4259         netdev->stats.tx_carrier_errors = adapter->stats.tncrs;
4260
4261         /* Tx Dropped needs to be maintained elsewhere */
4262
4263         /* Management Stats */
4264         adapter->stats.mgptc += er32(MGTPTC);
4265         adapter->stats.mgprc += er32(MGTPRC);
4266         adapter->stats.mgpdc += er32(MGTPDC);
4267 }
4268
4269 /**
4270  * e1000_phy_read_status - Update the PHY register status snapshot
4271  * @adapter: board private structure
4272  **/
4273 static void e1000_phy_read_status(struct e1000_adapter *adapter)
4274 {
4275         struct e1000_hw *hw = &adapter->hw;
4276         struct e1000_phy_regs *phy = &adapter->phy_regs;
4277
4278         if ((er32(STATUS) & E1000_STATUS_LU) &&
4279             (adapter->hw.phy.media_type == e1000_media_type_copper)) {
4280                 int ret_val;
4281
4282                 ret_val  = e1e_rphy(hw, PHY_CONTROL, &phy->bmcr);
4283                 ret_val |= e1e_rphy(hw, PHY_STATUS, &phy->bmsr);
4284                 ret_val |= e1e_rphy(hw, PHY_AUTONEG_ADV, &phy->advertise);
4285                 ret_val |= e1e_rphy(hw, PHY_LP_ABILITY, &phy->lpa);
4286                 ret_val |= e1e_rphy(hw, PHY_AUTONEG_EXP, &phy->expansion);
4287                 ret_val |= e1e_rphy(hw, PHY_1000T_CTRL, &phy->ctrl1000);
4288                 ret_val |= e1e_rphy(hw, PHY_1000T_STATUS, &phy->stat1000);
4289                 ret_val |= e1e_rphy(hw, PHY_EXT_STATUS, &phy->estatus);
4290                 if (ret_val)
4291                         e_warn("Error reading PHY register\n");
4292         } else {
4293                 /*
4294                  * Do not read PHY registers if link is not up
4295                  * Set values to typical power-on defaults
4296                  */
4297                 phy->bmcr = (BMCR_SPEED1000 | BMCR_ANENABLE | BMCR_FULLDPLX);
4298                 phy->bmsr = (BMSR_100FULL | BMSR_100HALF | BMSR_10FULL |
4299                              BMSR_10HALF | BMSR_ESTATEN | BMSR_ANEGCAPABLE |
4300                              BMSR_ERCAP);
4301                 phy->advertise = (ADVERTISE_PAUSE_ASYM | ADVERTISE_PAUSE_CAP |
4302                                   ADVERTISE_ALL | ADVERTISE_CSMA);
4303                 phy->lpa = 0;
4304                 phy->expansion = EXPANSION_ENABLENPAGE;
4305                 phy->ctrl1000 = ADVERTISE_1000FULL;
4306                 phy->stat1000 = 0;
4307                 phy->estatus = (ESTATUS_1000_TFULL | ESTATUS_1000_THALF);
4308         }
4309 }
4310
4311 static void e1000_print_link_info(struct e1000_adapter *adapter)
4312 {
4313         struct e1000_hw *hw = &adapter->hw;
4314         u32 ctrl = er32(CTRL);
4315
4316         /* Link status message must follow this format for user tools */
4317         printk(KERN_INFO "e1000e: %s NIC Link is Up %d Mbps %s Duplex, Flow Control: %s\n",
4318                 adapter->netdev->name,
4319                 adapter->link_speed,
4320                 adapter->link_duplex == FULL_DUPLEX ? "Full" : "Half",
4321                 (ctrl & E1000_CTRL_TFCE) && (ctrl & E1000_CTRL_RFCE) ? "Rx/Tx" :
4322                 (ctrl & E1000_CTRL_RFCE) ? "Rx" :
4323                 (ctrl & E1000_CTRL_TFCE) ? "Tx" : "None");
4324 }
4325
4326 static bool e1000e_has_link(struct e1000_adapter *adapter)
4327 {
4328         struct e1000_hw *hw = &adapter->hw;
4329         bool link_active = false;
4330         s32 ret_val = 0;
4331
4332         /*
4333          * get_link_status is set on LSC (link status) interrupt or
4334          * Rx sequence error interrupt.  get_link_status will stay
4335          * false until the check_for_link establishes link
4336          * for copper adapters ONLY
4337          */
4338         switch (hw->phy.media_type) {
4339         case e1000_media_type_copper:
4340                 if (hw->mac.get_link_status) {
4341                         ret_val = hw->mac.ops.check_for_link(hw);
4342                         link_active = !hw->mac.get_link_status;
4343                 } else {
4344                         link_active = true;
4345                 }
4346                 break;
4347         case e1000_media_type_fiber:
4348                 ret_val = hw->mac.ops.check_for_link(hw);
4349                 link_active = !!(er32(STATUS) & E1000_STATUS_LU);
4350                 break;
4351         case e1000_media_type_internal_serdes:
4352                 ret_val = hw->mac.ops.check_for_link(hw);
4353                 link_active = adapter->hw.mac.serdes_has_link;
4354                 break;
4355         default:
4356         case e1000_media_type_unknown:
4357                 break;
4358         }
4359
4360         if ((ret_val == E1000_ERR_PHY) && (hw->phy.type == e1000_phy_igp_3) &&
4361             (er32(CTRL) & E1000_PHY_CTRL_GBE_DISABLE)) {
4362                 /* See e1000_kmrn_lock_loss_workaround_ich8lan() */
4363                 e_info("Gigabit has been disabled, downgrading speed\n");
4364         }
4365
4366         return link_active;
4367 }
4368
4369 static void e1000e_enable_receives(struct e1000_adapter *adapter)
4370 {
4371         /* make sure the receive unit is started */
4372         if ((adapter->flags & FLAG_RX_NEEDS_RESTART) &&
4373             (adapter->flags & FLAG_RX_RESTART_NOW)) {
4374                 struct e1000_hw *hw = &adapter->hw;
4375                 u32 rctl = er32(RCTL);
4376                 ew32(RCTL, rctl | E1000_RCTL_EN);
4377                 adapter->flags &= ~FLAG_RX_RESTART_NOW;
4378         }
4379 }
4380
4381 static void e1000e_check_82574_phy_workaround(struct e1000_adapter *adapter)
4382 {
4383         struct e1000_hw *hw = &adapter->hw;
4384
4385         /*
4386          * With 82574 controllers, PHY needs to be checked periodically
4387          * for hung state and reset, if two calls return true
4388          */
4389         if (e1000_check_phy_82574(hw))
4390                 adapter->phy_hang_count++;
4391         else
4392                 adapter->phy_hang_count = 0;
4393
4394         if (adapter->phy_hang_count > 1) {
4395                 adapter->phy_hang_count = 0;
4396                 schedule_work(&adapter->reset_task);
4397         }
4398 }
4399
4400 /**
4401  * e1000_watchdog - Timer Call-back
4402  * @data: pointer to adapter cast into an unsigned long
4403  **/
4404 static void e1000_watchdog(unsigned long data)
4405 {
4406         struct e1000_adapter *adapter = (struct e1000_adapter *) data;
4407
4408         /* Do the rest outside of interrupt context */
4409         schedule_work(&adapter->watchdog_task);
4410
4411         /* TODO: make this use queue_delayed_work() */
4412 }
4413
4414 static void e1000_watchdog_task(struct work_struct *work)
4415 {
4416         struct e1000_adapter *adapter = container_of(work,
4417                                         struct e1000_adapter, watchdog_task);
4418         struct net_device *netdev = adapter->netdev;
4419         struct e1000_mac_info *mac = &adapter->hw.mac;
4420         struct e1000_phy_info *phy = &adapter->hw.phy;
4421         struct e1000_ring *tx_ring = adapter->tx_ring;
4422         struct e1000_hw *hw = &adapter->hw;
4423         u32 link, tctl;
4424
4425         if (test_bit(__E1000_DOWN, &adapter->state))
4426                 return;
4427
4428         link = e1000e_has_link(adapter);
4429         if ((netif_carrier_ok(netdev)) && link) {
4430                 /* Cancel scheduled suspend requests. */
4431                 pm_runtime_resume(netdev->dev.parent);
4432
4433                 e1000e_enable_receives(adapter);
4434                 goto link_up;
4435         }
4436
4437         if ((e1000e_enable_tx_pkt_filtering(hw)) &&
4438             (adapter->mng_vlan_id != adapter->hw.mng_cookie.vlan_id))
4439                 e1000_update_mng_vlan(adapter);
4440
4441         if (link) {
4442                 if (!netif_carrier_ok(netdev)) {
4443                         bool txb2b = true;
4444
4445                         /* Cancel scheduled suspend requests. */
4446                         pm_runtime_resume(netdev->dev.parent);
4447
4448                         /* update snapshot of PHY registers on LSC */
4449                         e1000_phy_read_status(adapter);
4450                         mac->ops.get_link_up_info(&adapter->hw,
4451                                                    &adapter->link_speed,
4452                                                    &adapter->link_duplex);
4453                         e1000_print_link_info(adapter);
4454                         /*
4455                          * On supported PHYs, check for duplex mismatch only
4456                          * if link has autonegotiated at 10/100 half
4457                          */
4458                         if ((hw->phy.type == e1000_phy_igp_3 ||
4459                              hw->phy.type == e1000_phy_bm) &&
4460                             (hw->mac.autoneg == true) &&
4461                             (adapter->link_speed == SPEED_10 ||
4462                              adapter->link_speed == SPEED_100) &&
4463                             (adapter->link_duplex == HALF_DUPLEX)) {
4464                                 u16 autoneg_exp;
4465
4466                                 e1e_rphy(hw, PHY_AUTONEG_EXP, &autoneg_exp);
4467
4468                                 if (!(autoneg_exp & NWAY_ER_LP_NWAY_CAPS))
4469                                         e_info("Autonegotiated half duplex but link partner cannot autoneg.  Try forcing full duplex if link gets many collisions.\n");
4470                         }
4471
4472                         /* adjust timeout factor according to speed/duplex */
4473                         adapter->tx_timeout_factor = 1;
4474                         switch (adapter->link_speed) {
4475                         case SPEED_10:
4476                                 txb2b = false;
4477                                 adapter->tx_timeout_factor = 16;
4478                                 break;
4479                         case SPEED_100:
4480                                 txb2b = false;
4481                                 adapter->tx_timeout_factor = 10;
4482                                 break;
4483                         }
4484
4485                         /*
4486                          * workaround: re-program speed mode bit after
4487                          * link-up event
4488                          */
4489                         if ((adapter->flags & FLAG_TARC_SPEED_MODE_BIT) &&
4490                             !txb2b) {
4491                                 u32 tarc0;
4492                                 tarc0 = er32(TARC(0));
4493                                 tarc0 &= ~SPEED_MODE_BIT;
4494                                 ew32(TARC(0), tarc0);
4495                         }
4496
4497                         /*
4498                          * disable TSO for pcie and 10/100 speeds, to avoid
4499                          * some hardware issues
4500                          */
4501                         if (!(adapter->flags & FLAG_TSO_FORCE)) {
4502                                 switch (adapter->link_speed) {
4503                                 case SPEED_10:
4504                                 case SPEED_100:
4505                                         e_info("10/100 speed: disabling TSO\n");
4506                                         netdev->features &= ~NETIF_F_TSO;
4507                                         netdev->features &= ~NETIF_F_TSO6;
4508                                         break;
4509                                 case SPEED_1000:
4510                                         netdev->features |= NETIF_F_TSO;
4511                                         netdev->features |= NETIF_F_TSO6;
4512                                         break;
4513                                 default:
4514                                         /* oops */
4515                                         break;
4516                                 }
4517                         }
4518
4519                         /*
4520                          * enable transmits in the hardware, need to do this
4521                          * after setting TARC(0)
4522                          */
4523                         tctl = er32(TCTL);
4524                         tctl |= E1000_TCTL_EN;
4525                         ew32(TCTL, tctl);
4526
4527                         /*
4528                          * Perform any post-link-up configuration before
4529                          * reporting link up.
4530                          */
4531                         if (phy->ops.cfg_on_link_up)
4532                                 phy->ops.cfg_on_link_up(hw);
4533
4534                         netif_carrier_on(netdev);
4535
4536                         if (!test_bit(__E1000_DOWN, &adapter->state))
4537                                 mod_timer(&adapter->phy_info_timer,
4538                                           round_jiffies(jiffies + 2 * HZ));
4539                 }
4540         } else {
4541                 if (netif_carrier_ok(netdev)) {
4542                         adapter->link_speed = 0;
4543                         adapter->link_duplex = 0;
4544                         /* Link status message must follow this format */
4545                         printk(KERN_INFO "e1000e: %s NIC Link is Down\n",
4546                                adapter->netdev->name);
4547                         netif_carrier_off(netdev);
4548                         if (!test_bit(__E1000_DOWN, &adapter->state))
4549                                 mod_timer(&adapter->phy_info_timer,
4550                                           round_jiffies(jiffies + 2 * HZ));
4551
4552                         if (adapter->flags & FLAG_RX_NEEDS_RESTART)
4553                                 schedule_work(&adapter->reset_task);
4554                         else
4555                                 pm_schedule_suspend(netdev->dev.parent,
4556                                                         LINK_TIMEOUT);
4557                 }
4558         }
4559
4560 link_up:
4561         spin_lock(&adapter->stats64_lock);
4562         e1000e_update_stats(adapter);
4563
4564         mac->tx_packet_delta = adapter->stats.tpt - adapter->tpt_old;
4565         adapter->tpt_old = adapter->stats.tpt;
4566         mac->collision_delta = adapter->stats.colc - adapter->colc_old;
4567         adapter->colc_old = adapter->stats.colc;
4568
4569         adapter->gorc = adapter->stats.gorc - adapter->gorc_old;
4570         adapter->gorc_old = adapter->stats.gorc;
4571         adapter->gotc = adapter->stats.gotc - adapter->gotc_old;
4572         adapter->gotc_old = adapter->stats.gotc;
4573         spin_unlock(&adapter->stats64_lock);
4574
4575         e1000e_update_adaptive(&adapter->hw);
4576
4577         if (!netif_carrier_ok(netdev) &&
4578             (e1000_desc_unused(tx_ring) + 1 < tx_ring->count)) {
4579                 /*
4580                  * We've lost link, so the controller stops DMA,
4581                  * but we've got queued Tx work that's never going
4582                  * to get done, so reset controller to flush Tx.
4583                  * (Do the reset outside of interrupt context).
4584                  */
4585                 schedule_work(&adapter->reset_task);
4586                 /* return immediately since reset is imminent */
4587                 return;
4588         }
4589
4590         /* Simple mode for Interrupt Throttle Rate (ITR) */
4591         if (adapter->itr_setting == 4) {
4592                 /*
4593                  * Symmetric Tx/Rx gets a reduced ITR=2000;
4594                  * Total asymmetrical Tx or Rx gets ITR=8000;
4595                  * everyone else is between 2000-8000.
4596                  */
4597                 u32 goc = (adapter->gotc + adapter->gorc) / 10000;
4598                 u32 dif = (adapter->gotc > adapter->gorc ?
4599                             adapter->gotc - adapter->gorc :
4600                             adapter->gorc - adapter->gotc) / 10000;
4601                 u32 itr = goc > 0 ? (dif * 6000 / goc + 2000) : 8000;
4602
4603                 ew32(ITR, 1000000000 / (itr * 256));
4604         }
4605
4606         /* Cause software interrupt to ensure Rx ring is cleaned */
4607         if (adapter->msix_entries)
4608                 ew32(ICS, adapter->rx_ring->ims_val);
4609         else
4610                 ew32(ICS, E1000_ICS_RXDMT0);
4611
4612         /* flush pending descriptors to memory before detecting Tx hang */
4613         e1000e_flush_descriptors(adapter);
4614
4615         /* Force detection of hung controller every watchdog period */
4616         adapter->detect_tx_hung = true;
4617
4618         /*
4619          * With 82571 controllers, LAA may be overwritten due to controller
4620          * reset from the other port. Set the appropriate LAA in RAR[0]
4621          */
4622         if (e1000e_get_laa_state_82571(hw))
4623                 hw->mac.ops.rar_set(hw, adapter->hw.mac.addr, 0);
4624
4625         if (adapter->flags2 & FLAG2_CHECK_PHY_HANG)
4626                 e1000e_check_82574_phy_workaround(adapter);
4627
4628         /* Reset the timer */
4629         if (!test_bit(__E1000_DOWN, &adapter->state))
4630                 mod_timer(&adapter->watchdog_timer,
4631                           round_jiffies(jiffies + 2 * HZ));
4632 }
4633
4634 #define E1000_TX_FLAGS_CSUM             0x00000001
4635 #define E1000_TX_FLAGS_VLAN             0x00000002
4636 #define E1000_TX_FLAGS_TSO              0x00000004
4637 #define E1000_TX_FLAGS_IPV4             0x00000008
4638 #define E1000_TX_FLAGS_NO_FCS           0x00000010
4639 #define E1000_TX_FLAGS_VLAN_MASK        0xffff0000
4640 #define E1000_TX_FLAGS_VLAN_SHIFT       16
4641
4642 static int e1000_tso(struct e1000_ring *tx_ring, struct sk_buff *skb)
4643 {
4644         struct e1000_context_desc *context_desc;
4645         struct e1000_buffer *buffer_info;
4646         unsigned int i;
4647         u32 cmd_length = 0;
4648         u16 ipcse = 0, tucse, mss;
4649         u8 ipcss, ipcso, tucss, tucso, hdr_len;
4650
4651         if (!skb_is_gso(skb))
4652                 return 0;
4653
4654         if (skb_header_cloned(skb)) {
4655                 int err = pskb_expand_head(skb, 0, 0, GFP_ATOMIC);
4656
4657                 if (err)
4658                         return err;
4659         }
4660
4661         hdr_len = skb_transport_offset(skb) + tcp_hdrlen(skb);
4662         mss = skb_shinfo(skb)->gso_size;
4663         if (skb->protocol == htons(ETH_P_IP)) {
4664                 struct iphdr *iph = ip_hdr(skb);
4665                 iph->tot_len = 0;
4666                 iph->check = 0;
4667                 tcp_hdr(skb)->check = ~csum_tcpudp_magic(iph->saddr, iph->daddr,
4668                                                          0, IPPROTO_TCP, 0);
4669                 cmd_length = E1000_TXD_CMD_IP;
4670                 ipcse = skb_transport_offset(skb) - 1;
4671         } else if (skb_is_gso_v6(skb)) {
4672                 ipv6_hdr(skb)->payload_len = 0;
4673                 tcp_hdr(skb)->check = ~csum_ipv6_magic(&ipv6_hdr(skb)->saddr,
4674                                                        &ipv6_hdr(skb)->daddr,
4675                                                        0, IPPROTO_TCP, 0);
4676                 ipcse = 0;
4677         }
4678         ipcss = skb_network_offset(skb);
4679         ipcso = (void *)&(ip_hdr(skb)->check) - (void *)skb->data;
4680         tucss = skb_transport_offset(skb);
4681         tucso = (void *)&(tcp_hdr(skb)->check) - (void *)skb->data;
4682         tucse = 0;
4683
4684         cmd_length |= (E1000_TXD_CMD_DEXT | E1000_TXD_CMD_TSE |
4685                        E1000_TXD_CMD_TCP | (skb->len - (hdr_len)));
4686
4687         i = tx_ring->next_to_use;
4688         context_desc = E1000_CONTEXT_DESC(*tx_ring, i);
4689         buffer_info = &tx_ring->buffer_info[i];
4690
4691         context_desc->lower_setup.ip_fields.ipcss  = ipcss;
4692         context_desc->lower_setup.ip_fields.ipcso  = ipcso;
4693         context_desc->lower_setup.ip_fields.ipcse  = cpu_to_le16(ipcse);
4694         context_desc->upper_setup.tcp_fields.tucss = tucss;
4695         context_desc->upper_setup.tcp_fields.tucso = tucso;
4696         context_desc->upper_setup.tcp_fields.tucse = cpu_to_le16(tucse);
4697         context_desc->tcp_seg_setup.fields.mss     = cpu_to_le16(mss);
4698         context_desc->tcp_seg_setup.fields.hdr_len = hdr_len;
4699         context_desc->cmd_and_length = cpu_to_le32(cmd_length);
4700
4701         buffer_info->time_stamp = jiffies;
4702         buffer_info->next_to_watch = i;
4703
4704         i++;
4705         if (i == tx_ring->count)
4706                 i = 0;
4707         tx_ring->next_to_use = i;
4708
4709         return 1;
4710 }
4711
4712 static bool e1000_tx_csum(struct e1000_ring *tx_ring, struct sk_buff *skb)
4713 {
4714         struct e1000_adapter *adapter = tx_ring->adapter;
4715         struct e1000_context_desc *context_desc;
4716         struct e1000_buffer *buffer_info;
4717         unsigned int i;
4718         u8 css;
4719         u32 cmd_len = E1000_TXD_CMD_DEXT;
4720         __be16 protocol;
4721
4722         if (skb->ip_summed != CHECKSUM_PARTIAL)
4723                 return 0;
4724
4725         if (skb->protocol == cpu_to_be16(ETH_P_8021Q))
4726                 protocol = vlan_eth_hdr(skb)->h_vlan_encapsulated_proto;
4727         else
4728                 protocol = skb->protocol;
4729
4730         switch (protocol) {
4731         case cpu_to_be16(ETH_P_IP):
4732                 if (ip_hdr(skb)->protocol == IPPROTO_TCP)
4733                         cmd_len |= E1000_TXD_CMD_TCP;
4734                 break;
4735         case cpu_to_be16(ETH_P_IPV6):
4736                 /* XXX not handling all IPV6 headers */
4737                 if (ipv6_hdr(skb)->nexthdr == IPPROTO_TCP)
4738                         cmd_len |= E1000_TXD_CMD_TCP;
4739                 break;
4740         default:
4741                 if (unlikely(net_ratelimit()))
4742                         e_warn("checksum_partial proto=%x!\n",
4743                                be16_to_cpu(protocol));
4744                 break;
4745         }
4746
4747         css = skb_checksum_start_offset(skb);
4748
4749         i = tx_ring->next_to_use;
4750         buffer_info = &tx_ring->buffer_info[i];
4751         context_desc = E1000_CONTEXT_DESC(*tx_ring, i);
4752
4753         context_desc->lower_setup.ip_config = 0;
4754         context_desc->upper_setup.tcp_fields.tucss = css;
4755         context_desc->upper_setup.tcp_fields.tucso =
4756                                 css + skb->csum_offset;
4757         context_desc->upper_setup.tcp_fields.tucse = 0;
4758         context_desc->tcp_seg_setup.data = 0;
4759         context_desc->cmd_and_length = cpu_to_le32(cmd_len);
4760
4761         buffer_info->time_stamp = jiffies;
4762         buffer_info->next_to_watch = i;
4763
4764         i++;
4765         if (i == tx_ring->count)
4766                 i = 0;
4767         tx_ring->next_to_use = i;
4768
4769         return 1;
4770 }
4771
4772 #define E1000_MAX_PER_TXD       8192
4773 #define E1000_MAX_TXD_PWR       12
4774
4775 static int e1000_tx_map(struct e1000_ring *tx_ring, struct sk_buff *skb,
4776                         unsigned int first, unsigned int max_per_txd,
4777                         unsigned int nr_frags, unsigned int mss)
4778 {
4779         struct e1000_adapter *adapter = tx_ring->adapter;
4780         struct pci_dev *pdev = adapter->pdev;
4781         struct e1000_buffer *buffer_info;
4782         unsigned int len = skb_headlen(skb);
4783         unsigned int offset = 0, size, count = 0, i;
4784         unsigned int f, bytecount, segs;
4785
4786         i = tx_ring->next_to_use;
4787
4788         while (len) {
4789                 buffer_info = &tx_ring->buffer_info[i];
4790                 size = min(len, max_per_txd);
4791
4792                 buffer_info->length = size;
4793                 buffer_info->time_stamp = jiffies;
4794                 buffer_info->next_to_watch = i;
4795                 buffer_info->dma = dma_map_single(&pdev->dev,
4796                                                   skb->data + offset,
4797                                                   size, DMA_TO_DEVICE);
4798                 buffer_info->mapped_as_page = false;
4799                 if (dma_mapping_error(&pdev->dev, buffer_info->dma))
4800                         goto dma_error;
4801
4802                 len -= size;
4803                 offset += size;
4804                 count++;
4805
4806                 if (len) {
4807                         i++;
4808                         if (i == tx_ring->count)
4809                                 i = 0;
4810                 }
4811         }
4812
4813         for (f = 0; f < nr_frags; f++) {
4814                 const struct skb_frag_struct *frag;
4815
4816                 frag = &skb_shinfo(skb)->frags[f];
4817                 len = skb_frag_size(frag);
4818                 offset = 0;
4819
4820                 while (len) {
4821                         i++;
4822                         if (i == tx_ring->count)
4823                                 i = 0;
4824
4825                         buffer_info = &tx_ring->buffer_info[i];
4826                         size = min(len, max_per_txd);
4827
4828                         buffer_info->length = size;
4829                         buffer_info->time_stamp = jiffies;
4830                         buffer_info->next_to_watch = i;
4831                         buffer_info->dma = skb_frag_dma_map(&pdev->dev, frag,
4832                                                 offset, size, DMA_TO_DEVICE);
4833                         buffer_info->mapped_as_page = true;
4834                         if (dma_mapping_error(&pdev->dev, buffer_info->dma))
4835                                 goto dma_error;
4836
4837                         len -= size;
4838                         offset += size;
4839                         count++;
4840                 }
4841         }
4842
4843         segs = skb_shinfo(skb)->gso_segs ? : 1;
4844         /* multiply data chunks by size of headers */
4845         bytecount = ((segs - 1) * skb_headlen(skb)) + skb->len;
4846
4847         tx_ring->buffer_info[i].skb = skb;
4848         tx_ring->buffer_info[i].segs = segs;
4849         tx_ring->buffer_info[i].bytecount = bytecount;
4850         tx_ring->buffer_info[first].next_to_watch = i;
4851
4852         return count;
4853
4854 dma_error:
4855         dev_err(&pdev->dev, "Tx DMA map failed\n");
4856         buffer_info->dma = 0;
4857         if (count)
4858                 count--;
4859
4860         while (count--) {
4861                 if (i == 0)
4862                         i += tx_ring->count;
4863                 i--;
4864                 buffer_info = &tx_ring->buffer_info[i];
4865                 e1000_put_txbuf(tx_ring, buffer_info);
4866         }
4867
4868         return 0;
4869 }
4870
4871 static void e1000_tx_queue(struct e1000_ring *tx_ring, int tx_flags, int count)
4872 {
4873         struct e1000_adapter *adapter = tx_ring->adapter;
4874         struct e1000_tx_desc *tx_desc = NULL;
4875         struct e1000_buffer *buffer_info;
4876         u32 txd_upper = 0, txd_lower = E1000_TXD_CMD_IFCS;
4877         unsigned int i;
4878
4879         if (tx_flags & E1000_TX_FLAGS_TSO) {
4880                 txd_lower |= E1000_TXD_CMD_DEXT | E1000_TXD_DTYP_D |
4881                              E1000_TXD_CMD_TSE;
4882                 txd_upper |= E1000_TXD_POPTS_TXSM << 8;
4883
4884                 if (tx_flags & E1000_TX_FLAGS_IPV4)
4885                         txd_upper |= E1000_TXD_POPTS_IXSM << 8;
4886         }
4887
4888         if (tx_flags & E1000_TX_FLAGS_CSUM) {
4889                 txd_lower |= E1000_TXD_CMD_DEXT | E1000_TXD_DTYP_D;
4890                 txd_upper |= E1000_TXD_POPTS_TXSM << 8;
4891         }
4892
4893         if (tx_flags & E1000_TX_FLAGS_VLAN) {
4894                 txd_lower |= E1000_TXD_CMD_VLE;
4895                 txd_upper |= (tx_flags & E1000_TX_FLAGS_VLAN_MASK);
4896         }
4897
4898         if (unlikely(tx_flags & E1000_TX_FLAGS_NO_FCS))
4899                 txd_lower &= ~(E1000_TXD_CMD_IFCS);
4900
4901         i = tx_ring->next_to_use;
4902
4903         do {
4904                 buffer_info = &tx_ring->buffer_info[i];
4905                 tx_desc = E1000_TX_DESC(*tx_ring, i);
4906                 tx_desc->buffer_addr = cpu_to_le64(buffer_info->dma);
4907                 tx_desc->lower.data =
4908                         cpu_to_le32(txd_lower | buffer_info->length);
4909                 tx_desc->upper.data = cpu_to_le32(txd_upper);
4910
4911                 i++;
4912                 if (i == tx_ring->count)
4913                         i = 0;
4914         } while (--count > 0);
4915
4916         tx_desc->lower.data |= cpu_to_le32(adapter->txd_cmd);
4917
4918         /* txd_cmd re-enables FCS, so we'll re-disable it here as desired. */
4919         if (unlikely(tx_flags & E1000_TX_FLAGS_NO_FCS))
4920                 tx_desc->lower.data &= ~(cpu_to_le32(E1000_TXD_CMD_IFCS));
4921
4922         /*
4923          * Force memory writes to complete before letting h/w
4924          * know there are new descriptors to fetch.  (Only
4925          * applicable for weak-ordered memory model archs,
4926          * such as IA-64).
4927          */
4928         wmb();
4929
4930         tx_ring->next_to_use = i;
4931
4932         if (adapter->flags2 & FLAG2_PCIM2PCI_ARBITER_WA)
4933                 e1000e_update_tdt_wa(tx_ring, i);
4934         else
4935                 writel(i, tx_ring->tail);
4936
4937         /*
4938          * we need this if more than one processor can write to our tail
4939          * at a time, it synchronizes IO on IA64/Altix systems
4940          */
4941         mmiowb();
4942 }
4943
4944 #define MINIMUM_DHCP_PACKET_SIZE 282
4945 static int e1000_transfer_dhcp_info(struct e1000_adapter *adapter,
4946                                     struct sk_buff *skb)
4947 {
4948         struct e1000_hw *hw =  &adapter->hw;
4949         u16 length, offset;
4950
4951         if (vlan_tx_tag_present(skb)) {
4952                 if (!((vlan_tx_tag_get(skb) == adapter->hw.mng_cookie.vlan_id) &&
4953                     (adapter->hw.mng_cookie.status &
4954                         E1000_MNG_DHCP_COOKIE_STATUS_VLAN)))
4955                         return 0;
4956         }
4957
4958         if (skb->len <= MINIMUM_DHCP_PACKET_SIZE)
4959                 return 0;
4960
4961         if (((struct ethhdr *) skb->data)->h_proto != htons(ETH_P_IP))
4962                 return 0;
4963
4964         {
4965                 const struct iphdr *ip = (struct iphdr *)((u8 *)skb->data+14);
4966                 struct udphdr *udp;
4967
4968                 if (ip->protocol != IPPROTO_UDP)
4969                         return 0;
4970
4971                 udp = (struct udphdr *)((u8 *)ip + (ip->ihl << 2));
4972                 if (ntohs(udp->dest) != 67)
4973                         return 0;
4974
4975                 offset = (u8 *)udp + 8 - skb->data;
4976                 length = skb->len - offset;
4977                 return e1000e_mng_write_dhcp_info(hw, (u8 *)udp + 8, length);
4978         }
4979
4980         return 0;
4981 }
4982
4983 static int __e1000_maybe_stop_tx(struct e1000_ring *tx_ring, int size)
4984 {
4985         struct e1000_adapter *adapter = tx_ring->adapter;
4986
4987         netif_stop_queue(adapter->netdev);
4988         /*
4989          * Herbert's original patch had:
4990          *  smp_mb__after_netif_stop_queue();
4991          * but since that doesn't exist yet, just open code it.
4992          */
4993         smp_mb();
4994
4995         /*
4996          * We need to check again in a case another CPU has just
4997          * made room available.
4998          */
4999         if (e1000_desc_unused(tx_ring) < size)
5000                 return -EBUSY;
5001
5002         /* A reprieve! */
5003         netif_start_queue(adapter->netdev);
5004         ++adapter->restart_queue;
5005         return 0;
5006 }
5007
5008 static int e1000_maybe_stop_tx(struct e1000_ring *tx_ring, int size)
5009 {
5010         if (e1000_desc_unused(tx_ring) >= size)
5011                 return 0;
5012         return __e1000_maybe_stop_tx(tx_ring, size);
5013 }
5014
5015 #define TXD_USE_COUNT(S, X) (((S) >> (X)) + 1)
5016 static netdev_tx_t e1000_xmit_frame(struct sk_buff *skb,
5017                                     struct net_device *netdev)
5018 {
5019         struct e1000_adapter *adapter = netdev_priv(netdev);
5020         struct e1000_ring *tx_ring = adapter->tx_ring;
5021         unsigned int first;
5022         unsigned int max_per_txd = E1000_MAX_PER_TXD;
5023         unsigned int max_txd_pwr = E1000_MAX_TXD_PWR;
5024         unsigned int tx_flags = 0;
5025         unsigned int len = skb_headlen(skb);
5026         unsigned int nr_frags;
5027         unsigned int mss;
5028         int count = 0;
5029         int tso;
5030         unsigned int f;
5031
5032         if (test_bit(__E1000_DOWN, &adapter->state)) {
5033                 dev_kfree_skb_any(skb);
5034                 return NETDEV_TX_OK;
5035         }
5036
5037         if (skb->len <= 0) {
5038                 dev_kfree_skb_any(skb);
5039                 return NETDEV_TX_OK;
5040         }
5041
5042         mss = skb_shinfo(skb)->gso_size;
5043         /*
5044          * The controller does a simple calculation to
5045          * make sure there is enough room in the FIFO before
5046          * initiating the DMA for each buffer.  The calc is:
5047          * 4 = ceil(buffer len/mss).  To make sure we don't
5048          * overrun the FIFO, adjust the max buffer len if mss
5049          * drops.
5050          */
5051         if (mss) {
5052                 u8 hdr_len;
5053                 max_per_txd = min(mss << 2, max_per_txd);
5054                 max_txd_pwr = fls(max_per_txd) - 1;
5055
5056                 /*
5057                  * TSO Workaround for 82571/2/3 Controllers -- if skb->data
5058                  * points to just header, pull a few bytes of payload from
5059                  * frags into skb->data
5060                  */
5061                 hdr_len = skb_transport_offset(skb) + tcp_hdrlen(skb);
5062                 /*
5063                  * we do this workaround for ES2LAN, but it is un-necessary,
5064                  * avoiding it could save a lot of cycles
5065                  */
5066                 if (skb->data_len && (hdr_len == len)) {
5067                         unsigned int pull_size;
5068
5069                         pull_size = min_t(unsigned int, 4, skb->data_len);
5070                         if (!__pskb_pull_tail(skb, pull_size)) {
5071                                 e_err("__pskb_pull_tail failed.\n");
5072                                 dev_kfree_skb_any(skb);
5073                                 return NETDEV_TX_OK;
5074                         }
5075                         len = skb_headlen(skb);
5076                 }
5077         }
5078
5079         /* reserve a descriptor for the offload context */
5080         if ((mss) || (skb->ip_summed == CHECKSUM_PARTIAL))
5081                 count++;
5082         count++;
5083
5084         count += TXD_USE_COUNT(len, max_txd_pwr);
5085
5086         nr_frags = skb_shinfo(skb)->nr_frags;
5087         for (f = 0; f < nr_frags; f++)
5088                 count += TXD_USE_COUNT(skb_frag_size(&skb_shinfo(skb)->frags[f]),
5089                                        max_txd_pwr);
5090
5091         if (adapter->hw.mac.tx_pkt_filtering)
5092                 e1000_transfer_dhcp_info(adapter, skb);
5093
5094         /*
5095          * need: count + 2 desc gap to keep tail from touching
5096          * head, otherwise try next time
5097          */
5098         if (e1000_maybe_stop_tx(tx_ring, count + 2))
5099                 return NETDEV_TX_BUSY;
5100
5101         if (vlan_tx_tag_present(skb)) {
5102                 tx_flags |= E1000_TX_FLAGS_VLAN;
5103                 tx_flags |= (vlan_tx_tag_get(skb) << E1000_TX_FLAGS_VLAN_SHIFT);
5104         }
5105
5106         first = tx_ring->next_to_use;
5107
5108         tso = e1000_tso(tx_ring, skb);
5109         if (tso < 0) {
5110                 dev_kfree_skb_any(skb);
5111                 return NETDEV_TX_OK;
5112         }
5113
5114         if (tso)
5115                 tx_flags |= E1000_TX_FLAGS_TSO;
5116         else if (e1000_tx_csum(tx_ring, skb))
5117                 tx_flags |= E1000_TX_FLAGS_CSUM;
5118
5119         /*
5120          * Old method was to assume IPv4 packet by default if TSO was enabled.
5121          * 82571 hardware supports TSO capabilities for IPv6 as well...
5122          * no longer assume, we must.
5123          */
5124         if (skb->protocol == htons(ETH_P_IP))
5125                 tx_flags |= E1000_TX_FLAGS_IPV4;
5126
5127         if (unlikely(skb->no_fcs))
5128                 tx_flags |= E1000_TX_FLAGS_NO_FCS;
5129
5130         /* if count is 0 then mapping error has occurred */
5131         count = e1000_tx_map(tx_ring, skb, first, max_per_txd, nr_frags, mss);
5132         if (count) {
5133                 skb_tx_timestamp(skb);
5134
5135                 netdev_sent_queue(netdev, skb->len);
5136                 e1000_tx_queue(tx_ring, tx_flags, count);
5137                 /* Make sure there is space in the ring for the next send. */
5138                 e1000_maybe_stop_tx(tx_ring, MAX_SKB_FRAGS + 2);
5139
5140         } else {
5141                 dev_kfree_skb_any(skb);
5142                 tx_ring->buffer_info[first].time_stamp = 0;
5143                 tx_ring->next_to_use = first;
5144         }
5145
5146         return NETDEV_TX_OK;
5147 }
5148
5149 /**
5150  * e1000_tx_timeout - Respond to a Tx Hang
5151  * @netdev: network interface device structure
5152  **/
5153 static void e1000_tx_timeout(struct net_device *netdev)
5154 {
5155         struct e1000_adapter *adapter = netdev_priv(netdev);
5156
5157         /* Do the reset outside of interrupt context */
5158         adapter->tx_timeout_count++;
5159         schedule_work(&adapter->reset_task);
5160 }
5161
5162 static void e1000_reset_task(struct work_struct *work)
5163 {
5164         struct e1000_adapter *adapter;
5165         adapter = container_of(work, struct e1000_adapter, reset_task);
5166
5167         /* don't run the task if already down */
5168         if (test_bit(__E1000_DOWN, &adapter->state))
5169                 return;
5170
5171         if (!((adapter->flags & FLAG_RX_NEEDS_RESTART) &&
5172               (adapter->flags & FLAG_RX_RESTART_NOW))) {
5173                 e1000e_dump(adapter);
5174                 e_err("Reset adapter\n");
5175         }
5176         e1000e_reinit_locked(adapter);
5177 }
5178
5179 /**
5180  * e1000_get_stats64 - Get System Network Statistics
5181  * @netdev: network interface device structure
5182  * @stats: rtnl_link_stats64 pointer
5183  *
5184  * Returns the address of the device statistics structure.
5185  **/
5186 struct rtnl_link_stats64 *e1000e_get_stats64(struct net_device *netdev,
5187                                              struct rtnl_link_stats64 *stats)
5188 {
5189         struct e1000_adapter *adapter = netdev_priv(netdev);
5190
5191         memset(stats, 0, sizeof(struct rtnl_link_stats64));
5192         spin_lock(&adapter->stats64_lock);
5193         e1000e_update_stats(adapter);
5194         /* Fill out the OS statistics structure */
5195         stats->rx_bytes = adapter->stats.gorc;
5196         stats->rx_packets = adapter->stats.gprc;
5197         stats->tx_bytes = adapter->stats.gotc;
5198         stats->tx_packets = adapter->stats.gptc;
5199         stats->multicast = adapter->stats.mprc;
5200         stats->collisions = adapter->stats.colc;
5201
5202         /* Rx Errors */
5203
5204         /*
5205          * RLEC on some newer hardware can be incorrect so build
5206          * our own version based on RUC and ROC
5207          */
5208         stats->rx_errors = adapter->stats.rxerrc +
5209                 adapter->stats.crcerrs + adapter->stats.algnerrc +
5210                 adapter->stats.ruc + adapter->stats.roc +
5211                 adapter->stats.cexterr;
5212         stats->rx_length_errors = adapter->stats.ruc +
5213                                               adapter->stats.roc;
5214         stats->rx_crc_errors = adapter->stats.crcerrs;
5215         stats->rx_frame_errors = adapter->stats.algnerrc;
5216         stats->rx_missed_errors = adapter->stats.mpc;
5217
5218         /* Tx Errors */
5219         stats->tx_errors = adapter->stats.ecol +
5220                                        adapter->stats.latecol;
5221         stats->tx_aborted_errors = adapter->stats.ecol;
5222         stats->tx_window_errors = adapter->stats.latecol;
5223         stats->tx_carrier_errors = adapter->stats.tncrs;
5224
5225         /* Tx Dropped needs to be maintained elsewhere */
5226
5227         spin_unlock(&adapter->stats64_lock);
5228         return stats;
5229 }
5230
5231 /**
5232  * e1000_change_mtu - Change the Maximum Transfer Unit
5233  * @netdev: network interface device structure
5234  * @new_mtu: new value for maximum frame size
5235  *
5236  * Returns 0 on success, negative on failure
5237  **/
5238 static int e1000_change_mtu(struct net_device *netdev, int new_mtu)
5239 {
5240         struct e1000_adapter *adapter = netdev_priv(netdev);
5241         int max_frame = new_mtu + ETH_HLEN + ETH_FCS_LEN;
5242
5243         /* Jumbo frame support */
5244         if (max_frame > ETH_FRAME_LEN + ETH_FCS_LEN) {
5245                 if (!(adapter->flags & FLAG_HAS_JUMBO_FRAMES)) {
5246                         e_err("Jumbo Frames not supported.\n");
5247                         return -EINVAL;
5248                 }
5249
5250                 /*
5251                  * IP payload checksum (enabled with jumbos/packet-split when
5252                  * Rx checksum is enabled) and generation of RSS hash is
5253                  * mutually exclusive in the hardware.
5254                  */
5255                 if ((netdev->features & NETIF_F_RXCSUM) &&
5256                     (netdev->features & NETIF_F_RXHASH)) {
5257                         e_err("Jumbo frames cannot be enabled when both receive checksum offload and receive hashing are enabled.  Disable one of the receive offload features before enabling jumbos.\n");
5258                         return -EINVAL;
5259                 }
5260         }
5261
5262         /* Supported frame sizes */
5263         if ((new_mtu < ETH_ZLEN + ETH_FCS_LEN + VLAN_HLEN) ||
5264             (max_frame > adapter->max_hw_frame_size)) {
5265                 e_err("Unsupported MTU setting\n");
5266                 return -EINVAL;
5267         }
5268
5269         /* Jumbo frame workaround on 82579 and newer requires CRC be stripped */
5270         if ((adapter->hw.mac.type >= e1000_pch2lan) &&
5271             !(adapter->flags2 & FLAG2_CRC_STRIPPING) &&
5272             (new_mtu > ETH_DATA_LEN)) {
5273                 e_err("Jumbo Frames not supported on this device when CRC stripping is disabled.\n");
5274                 return -EINVAL;
5275         }
5276
5277         while (test_and_set_bit(__E1000_RESETTING, &adapter->state))
5278                 usleep_range(1000, 2000);
5279         /* e1000e_down -> e1000e_reset dependent on max_frame_size & mtu */
5280         adapter->max_frame_size = max_frame;
5281         e_info("changing MTU from %d to %d\n", netdev->mtu, new_mtu);
5282         netdev->mtu = new_mtu;
5283         if (netif_running(netdev))
5284                 e1000e_down(adapter);
5285
5286         /*
5287          * NOTE: netdev_alloc_skb reserves 16 bytes, and typically NET_IP_ALIGN
5288          * means we reserve 2 more, this pushes us to allocate from the next
5289          * larger slab size.
5290          * i.e. RXBUFFER_2048 --> size-4096 slab
5291          * However with the new *_jumbo_rx* routines, jumbo receives will use
5292          * fragmented skbs
5293          */
5294
5295         if (max_frame <= 2048)
5296                 adapter->rx_buffer_len = 2048;
5297         else
5298                 adapter->rx_buffer_len = 4096;
5299
5300         /* adjust allocation if LPE protects us, and we aren't using SBP */
5301         if ((max_frame == ETH_FRAME_LEN + ETH_FCS_LEN) ||
5302              (max_frame == ETH_FRAME_LEN + VLAN_HLEN + ETH_FCS_LEN))
5303                 adapter->rx_buffer_len = ETH_FRAME_LEN + VLAN_HLEN
5304                                          + ETH_FCS_LEN;
5305
5306         if (netif_running(netdev))
5307                 e1000e_up(adapter);
5308         else
5309                 e1000e_reset(adapter);
5310
5311         clear_bit(__E1000_RESETTING, &adapter->state);
5312
5313         return 0;
5314 }
5315
5316 static int e1000_mii_ioctl(struct net_device *netdev, struct ifreq *ifr,
5317                            int cmd)
5318 {
5319         struct e1000_adapter *adapter = netdev_priv(netdev);
5320         struct mii_ioctl_data *data = if_mii(ifr);
5321
5322         if (adapter->hw.phy.media_type != e1000_media_type_copper)
5323                 return -EOPNOTSUPP;
5324
5325         switch (cmd) {
5326         case SIOCGMIIPHY:
5327                 data->phy_id = adapter->hw.phy.addr;
5328                 break;
5329         case SIOCGMIIREG:
5330                 e1000_phy_read_status(adapter);
5331
5332                 switch (data->reg_num & 0x1F) {
5333                 case MII_BMCR:
5334                         data->val_out = adapter->phy_regs.bmcr;
5335                         break;
5336                 case MII_BMSR:
5337                         data->val_out = adapter->phy_regs.bmsr;
5338                         break;
5339                 case MII_PHYSID1:
5340                         data->val_out = (adapter->hw.phy.id >> 16);
5341                         break;
5342                 case MII_PHYSID2:
5343                         data->val_out = (adapter->hw.phy.id & 0xFFFF);
5344                         break;
5345                 case MII_ADVERTISE:
5346                         data->val_out = adapter->phy_regs.advertise;
5347                         break;
5348                 case MII_LPA:
5349                         data->val_out = adapter->phy_regs.lpa;
5350                         break;
5351                 case MII_EXPANSION:
5352                         data->val_out = adapter->phy_regs.expansion;
5353                         break;
5354                 case MII_CTRL1000:
5355                         data->val_out = adapter->phy_regs.ctrl1000;
5356                         break;
5357                 case MII_STAT1000:
5358                         data->val_out = adapter->phy_regs.stat1000;
5359                         break;
5360                 case MII_ESTATUS:
5361                         data->val_out = adapter->phy_regs.estatus;
5362                         break;
5363                 default:
5364                         return -EIO;
5365                 }
5366                 break;
5367         case SIOCSMIIREG:
5368         default:
5369                 return -EOPNOTSUPP;
5370         }
5371         return 0;
5372 }
5373
5374 static int e1000_ioctl(struct net_device *netdev, struct ifreq *ifr, int cmd)
5375 {
5376         switch (cmd) {
5377         case SIOCGMIIPHY:
5378         case SIOCGMIIREG:
5379         case SIOCSMIIREG:
5380                 return e1000_mii_ioctl(netdev, ifr, cmd);
5381         default:
5382                 return -EOPNOTSUPP;
5383         }
5384 }
5385
5386 static int e1000_init_phy_wakeup(struct e1000_adapter *adapter, u32 wufc)
5387 {
5388         struct e1000_hw *hw = &adapter->hw;
5389         u32 i, mac_reg;
5390         u16 phy_reg, wuc_enable;
5391         int retval = 0;
5392
5393         /* copy MAC RARs to PHY RARs */
5394         e1000_copy_rx_addrs_to_phy_ich8lan(hw);
5395
5396         retval = hw->phy.ops.acquire(hw);
5397         if (retval) {
5398                 e_err("Could not acquire PHY\n");
5399                 return retval;
5400         }
5401
5402         /* Enable access to wakeup registers on and set page to BM_WUC_PAGE */
5403         retval = e1000_enable_phy_wakeup_reg_access_bm(hw, &wuc_enable);
5404         if (retval)
5405                 goto release;
5406
5407         /* copy MAC MTA to PHY MTA - only needed for pchlan */
5408         for (i = 0; i < adapter->hw.mac.mta_reg_count; i++) {
5409                 mac_reg = E1000_READ_REG_ARRAY(hw, E1000_MTA, i);
5410                 hw->phy.ops.write_reg_page(hw, BM_MTA(i),
5411                                            (u16)(mac_reg & 0xFFFF));
5412                 hw->phy.ops.write_reg_page(hw, BM_MTA(i) + 1,
5413                                            (u16)((mac_reg >> 16) & 0xFFFF));
5414         }
5415
5416         /* configure PHY Rx Control register */
5417         hw->phy.ops.read_reg_page(&adapter->hw, BM_RCTL, &phy_reg);
5418         mac_reg = er32(RCTL);
5419         if (mac_reg & E1000_RCTL_UPE)
5420                 phy_reg |= BM_RCTL_UPE;
5421         if (mac_reg & E1000_RCTL_MPE)
5422                 phy_reg |= BM_RCTL_MPE;
5423         phy_reg &= ~(BM_RCTL_MO_MASK);
5424         if (mac_reg & E1000_RCTL_MO_3)
5425                 phy_reg |= (((mac_reg & E1000_RCTL_MO_3) >> E1000_RCTL_MO_SHIFT)
5426                                 << BM_RCTL_MO_SHIFT);
5427         if (mac_reg & E1000_RCTL_BAM)
5428                 phy_reg |= BM_RCTL_BAM;
5429         if (mac_reg & E1000_RCTL_PMCF)
5430                 phy_reg |= BM_RCTL_PMCF;
5431         mac_reg = er32(CTRL);
5432         if (mac_reg & E1000_CTRL_RFCE)
5433                 phy_reg |= BM_RCTL_RFCE;
5434         hw->phy.ops.write_reg_page(&adapter->hw, BM_RCTL, phy_reg);
5435
5436         /* enable PHY wakeup in MAC register */
5437         ew32(WUFC, wufc);
5438         ew32(WUC, E1000_WUC_PHY_WAKE | E1000_WUC_PME_EN);
5439
5440         /* configure and enable PHY wakeup in PHY registers */
5441         hw->phy.ops.write_reg_page(&adapter->hw, BM_WUFC, wufc);
5442         hw->phy.ops.write_reg_page(&adapter->hw, BM_WUC, E1000_WUC_PME_EN);
5443
5444         /* activate PHY wakeup */
5445         wuc_enable |= BM_WUC_ENABLE_BIT | BM_WUC_HOST_WU_BIT;
5446         retval = e1000_disable_phy_wakeup_reg_access_bm(hw, &wuc_enable);
5447         if (retval)
5448                 e_err("Could not set PHY Host Wakeup bit\n");
5449 release:
5450         hw->phy.ops.release(hw);
5451
5452         return retval;
5453 }
5454
5455 static int __e1000_shutdown(struct pci_dev *pdev, bool *enable_wake,
5456                             bool runtime)
5457 {
5458         struct net_device *netdev = pci_get_drvdata(pdev);
5459         struct e1000_adapter *adapter = netdev_priv(netdev);
5460         struct e1000_hw *hw = &adapter->hw;
5461         u32 ctrl, ctrl_ext, rctl, status;
5462         /* Runtime suspend should only enable wakeup for link changes */
5463         u32 wufc = runtime ? E1000_WUFC_LNKC : adapter->wol;
5464         int retval = 0;
5465
5466         netif_device_detach(netdev);
5467
5468         if (netif_running(netdev)) {
5469                 int count = E1000_CHECK_RESET_COUNT;
5470
5471                 while (test_bit(__E1000_RESETTING, &adapter->state) && count--)
5472                         usleep_range(10000, 20000);
5473
5474                 WARN_ON(test_bit(__E1000_RESETTING, &adapter->state));
5475                 e1000e_down(adapter);
5476                 e1000_free_irq(adapter);
5477         }
5478         e1000e_reset_interrupt_capability(adapter);
5479
5480         retval = pci_save_state(pdev);
5481         if (retval)
5482                 return retval;
5483
5484         status = er32(STATUS);
5485         if (status & E1000_STATUS_LU)
5486                 wufc &= ~E1000_WUFC_LNKC;
5487
5488         if (wufc) {
5489                 e1000_setup_rctl(adapter);
5490                 e1000e_set_rx_mode(netdev);
5491
5492                 /* turn on all-multi mode if wake on multicast is enabled */
5493                 if (wufc & E1000_WUFC_MC) {
5494                         rctl = er32(RCTL);
5495                         rctl |= E1000_RCTL_MPE;
5496                         ew32(RCTL, rctl);
5497                 }
5498
5499                 ctrl = er32(CTRL);
5500                 /* advertise wake from D3Cold */
5501                 #define E1000_CTRL_ADVD3WUC 0x00100000
5502                 /* phy power management enable */
5503                 #define E1000_CTRL_EN_PHY_PWR_MGMT 0x00200000
5504                 ctrl |= E1000_CTRL_ADVD3WUC;
5505                 if (!(adapter->flags2 & FLAG2_HAS_PHY_WAKEUP))
5506                         ctrl |= E1000_CTRL_EN_PHY_PWR_MGMT;
5507                 ew32(CTRL, ctrl);
5508
5509                 if (adapter->hw.phy.media_type == e1000_media_type_fiber ||
5510                     adapter->hw.phy.media_type ==
5511                     e1000_media_type_internal_serdes) {
5512                         /* keep the laser running in D3 */
5513                         ctrl_ext = er32(CTRL_EXT);
5514                         ctrl_ext |= E1000_CTRL_EXT_SDP3_DATA;
5515                         ew32(CTRL_EXT, ctrl_ext);
5516                 }
5517
5518                 if (adapter->flags & FLAG_IS_ICH)
5519                         e1000_suspend_workarounds_ich8lan(&adapter->hw);
5520
5521                 /* Allow time for pending master requests to run */
5522                 e1000e_disable_pcie_master(&adapter->hw);
5523
5524                 if (adapter->flags2 & FLAG2_HAS_PHY_WAKEUP) {
5525                         /* enable wakeup by the PHY */
5526                         retval = e1000_init_phy_wakeup(adapter, wufc);
5527                         if (retval)
5528                                 return retval;
5529                 } else {
5530                         /* enable wakeup by the MAC */
5531                         ew32(WUFC, wufc);
5532                         ew32(WUC, E1000_WUC_PME_EN);
5533                 }
5534         } else {
5535                 ew32(WUC, 0);
5536                 ew32(WUFC, 0);
5537         }
5538
5539         *enable_wake = !!wufc;
5540
5541         /* make sure adapter isn't asleep if manageability is enabled */
5542         if ((adapter->flags & FLAG_MNG_PT_ENABLED) ||
5543             (hw->mac.ops.check_mng_mode(hw)))
5544                 *enable_wake = true;
5545
5546         if (adapter->hw.phy.type == e1000_phy_igp_3)
5547                 e1000e_igp3_phy_powerdown_workaround_ich8lan(&adapter->hw);
5548
5549         /*
5550          * Release control of h/w to f/w.  If f/w is AMT enabled, this
5551          * would have already happened in close and is redundant.
5552          */
5553         e1000e_release_hw_control(adapter);
5554
5555         pci_disable_device(pdev);
5556
5557         return 0;
5558 }
5559
5560 static void e1000_power_off(struct pci_dev *pdev, bool sleep, bool wake)
5561 {
5562         if (sleep && wake) {
5563                 pci_prepare_to_sleep(pdev);
5564                 return;
5565         }
5566
5567         pci_wake_from_d3(pdev, wake);
5568         pci_set_power_state(pdev, PCI_D3hot);
5569 }
5570
5571 static void e1000_complete_shutdown(struct pci_dev *pdev, bool sleep,
5572                                     bool wake)
5573 {
5574         struct net_device *netdev = pci_get_drvdata(pdev);
5575         struct e1000_adapter *adapter = netdev_priv(netdev);
5576
5577         /*
5578          * The pci-e switch on some quad port adapters will report a
5579          * correctable error when the MAC transitions from D0 to D3.  To
5580          * prevent this we need to mask off the correctable errors on the
5581          * downstream port of the pci-e switch.
5582          */
5583         if (adapter->flags & FLAG_IS_QUAD_PORT) {
5584                 struct pci_dev *us_dev = pdev->bus->self;
5585                 int pos = pci_pcie_cap(us_dev);
5586                 u16 devctl;
5587
5588                 pci_read_config_word(us_dev, pos + PCI_EXP_DEVCTL, &devctl);
5589                 pci_write_config_word(us_dev, pos + PCI_EXP_DEVCTL,
5590                                       (devctl & ~PCI_EXP_DEVCTL_CERE));
5591
5592                 e1000_power_off(pdev, sleep, wake);
5593
5594                 pci_write_config_word(us_dev, pos + PCI_EXP_DEVCTL, devctl);
5595         } else {
5596                 e1000_power_off(pdev, sleep, wake);
5597         }
5598 }
5599
5600 #ifdef CONFIG_PCIEASPM
5601 static void __e1000e_disable_aspm(struct pci_dev *pdev, u16 state)
5602 {
5603         pci_disable_link_state_locked(pdev, state);
5604 }
5605 #else
5606 static void __e1000e_disable_aspm(struct pci_dev *pdev, u16 state)
5607 {
5608         int pos;
5609         u16 reg16;
5610
5611         /*
5612          * Both device and parent should have the same ASPM setting.
5613          * Disable ASPM in downstream component first and then upstream.
5614          */
5615         pos = pci_pcie_cap(pdev);
5616         pci_read_config_word(pdev, pos + PCI_EXP_LNKCTL, &reg16);
5617         reg16 &= ~state;
5618         pci_write_config_word(pdev, pos + PCI_EXP_LNKCTL, reg16);
5619
5620         if (!pdev->bus->self)
5621                 return;
5622
5623         pos = pci_pcie_cap(pdev->bus->self);
5624         pci_read_config_word(pdev->bus->self, pos + PCI_EXP_LNKCTL, &reg16);
5625         reg16 &= ~state;
5626         pci_write_config_word(pdev->bus->self, pos + PCI_EXP_LNKCTL, reg16);
5627 }
5628 #endif
5629 static void e1000e_disable_aspm(struct pci_dev *pdev, u16 state)
5630 {
5631         dev_info(&pdev->dev, "Disabling ASPM %s %s\n",
5632                  (state & PCIE_LINK_STATE_L0S) ? "L0s" : "",
5633                  (state & PCIE_LINK_STATE_L1) ? "L1" : "");
5634
5635         __e1000e_disable_aspm(pdev, state);
5636 }
5637
5638 #ifdef CONFIG_PM
5639 static bool e1000e_pm_ready(struct e1000_adapter *adapter)
5640 {
5641         return !!adapter->tx_ring->buffer_info;
5642 }
5643
5644 static int __e1000_resume(struct pci_dev *pdev)
5645 {
5646         struct net_device *netdev = pci_get_drvdata(pdev);
5647         struct e1000_adapter *adapter = netdev_priv(netdev);
5648         struct e1000_hw *hw = &adapter->hw;
5649         u16 aspm_disable_flag = 0;
5650         u32 err;
5651
5652         if (adapter->flags2 & FLAG2_DISABLE_ASPM_L0S)
5653                 aspm_disable_flag = PCIE_LINK_STATE_L0S;
5654         if (adapter->flags2 & FLAG2_DISABLE_ASPM_L1)
5655                 aspm_disable_flag |= PCIE_LINK_STATE_L1;
5656         if (aspm_disable_flag)
5657                 e1000e_disable_aspm(pdev, aspm_disable_flag);
5658
5659         pci_set_power_state(pdev, PCI_D0);
5660         pci_restore_state(pdev);
5661         pci_save_state(pdev);
5662
5663         e1000e_set_interrupt_capability(adapter);
5664         if (netif_running(netdev)) {
5665                 err = e1000_request_irq(adapter);
5666                 if (err)
5667                         return err;
5668         }
5669
5670         if (hw->mac.type >= e1000_pch2lan)
5671                 e1000_resume_workarounds_pchlan(&adapter->hw);
5672
5673         e1000e_power_up_phy(adapter);
5674
5675         /* report the system wakeup cause from S3/S4 */
5676         if (adapter->flags2 & FLAG2_HAS_PHY_WAKEUP) {
5677                 u16 phy_data;
5678
5679                 e1e_rphy(&adapter->hw, BM_WUS, &phy_data);
5680                 if (phy_data) {
5681                         e_info("PHY Wakeup cause - %s\n",
5682                                 phy_data & E1000_WUS_EX ? "Unicast Packet" :
5683                                 phy_data & E1000_WUS_MC ? "Multicast Packet" :
5684                                 phy_data & E1000_WUS_BC ? "Broadcast Packet" :
5685                                 phy_data & E1000_WUS_MAG ? "Magic Packet" :
5686                                 phy_data & E1000_WUS_LNKC ?
5687                                 "Link Status Change" : "other");
5688                 }
5689                 e1e_wphy(&adapter->hw, BM_WUS, ~0);
5690         } else {
5691                 u32 wus = er32(WUS);
5692                 if (wus) {
5693                         e_info("MAC Wakeup cause - %s\n",
5694                                 wus & E1000_WUS_EX ? "Unicast Packet" :
5695                                 wus & E1000_WUS_MC ? "Multicast Packet" :
5696                                 wus & E1000_WUS_BC ? "Broadcast Packet" :
5697                                 wus & E1000_WUS_MAG ? "Magic Packet" :
5698                                 wus & E1000_WUS_LNKC ? "Link Status Change" :
5699                                 "other");
5700                 }
5701                 ew32(WUS, ~0);
5702         }
5703
5704         e1000e_reset(adapter);
5705
5706         e1000_init_manageability_pt(adapter);
5707
5708         if (netif_running(netdev))
5709                 e1000e_up(adapter);
5710
5711         netif_device_attach(netdev);
5712
5713         /*
5714          * If the controller has AMT, do not set DRV_LOAD until the interface
5715          * is up.  For all other cases, let the f/w know that the h/w is now
5716          * under the control of the driver.
5717          */
5718         if (!(adapter->flags & FLAG_HAS_AMT))
5719                 e1000e_get_hw_control(adapter);
5720
5721         return 0;
5722 }
5723
5724 #ifdef CONFIG_PM_SLEEP
5725 static int e1000_suspend(struct device *dev)
5726 {
5727         struct pci_dev *pdev = to_pci_dev(dev);
5728         int retval;
5729         bool wake;
5730
5731         retval = __e1000_shutdown(pdev, &wake, false);
5732         if (!retval)
5733                 e1000_complete_shutdown(pdev, true, wake);
5734
5735         return retval;
5736 }
5737
5738 static int e1000_resume(struct device *dev)
5739 {
5740         struct pci_dev *pdev = to_pci_dev(dev);
5741         struct net_device *netdev = pci_get_drvdata(pdev);
5742         struct e1000_adapter *adapter = netdev_priv(netdev);
5743
5744         if (e1000e_pm_ready(adapter))
5745                 adapter->idle_check = true;
5746
5747         return __e1000_resume(pdev);
5748 }
5749 #endif /* CONFIG_PM_SLEEP */
5750
5751 #ifdef CONFIG_PM_RUNTIME
5752 static int e1000_runtime_suspend(struct device *dev)
5753 {
5754         struct pci_dev *pdev = to_pci_dev(dev);
5755         struct net_device *netdev = pci_get_drvdata(pdev);
5756         struct e1000_adapter *adapter = netdev_priv(netdev);
5757
5758         if (e1000e_pm_ready(adapter)) {
5759                 bool wake;
5760
5761                 __e1000_shutdown(pdev, &wake, true);
5762         }
5763
5764         return 0;
5765 }
5766
5767 static int e1000_idle(struct device *dev)
5768 {
5769         struct pci_dev *pdev = to_pci_dev(dev);
5770         struct net_device *netdev = pci_get_drvdata(pdev);
5771         struct e1000_adapter *adapter = netdev_priv(netdev);
5772
5773         if (!e1000e_pm_ready(adapter))
5774                 return 0;
5775
5776         if (adapter->idle_check) {
5777                 adapter->idle_check = false;
5778                 if (!e1000e_has_link(adapter))
5779                         pm_schedule_suspend(dev, MSEC_PER_SEC);
5780         }
5781
5782         return -EBUSY;
5783 }
5784
5785 static int e1000_runtime_resume(struct device *dev)
5786 {
5787         struct pci_dev *pdev = to_pci_dev(dev);
5788         struct net_device *netdev = pci_get_drvdata(pdev);
5789         struct e1000_adapter *adapter = netdev_priv(netdev);
5790
5791         if (!e1000e_pm_ready(adapter))
5792                 return 0;
5793
5794         adapter->idle_check = !dev->power.runtime_auto;
5795         return __e1000_resume(pdev);
5796 }
5797 #endif /* CONFIG_PM_RUNTIME */
5798 #endif /* CONFIG_PM */
5799
5800 static void e1000_shutdown(struct pci_dev *pdev)
5801 {
5802         bool wake = false;
5803
5804         __e1000_shutdown(pdev, &wake, false);
5805
5806         if (system_state == SYSTEM_POWER_OFF)
5807                 e1000_complete_shutdown(pdev, false, wake);
5808 }
5809
5810 #ifdef CONFIG_NET_POLL_CONTROLLER
5811
5812 static irqreturn_t e1000_intr_msix(int irq, void *data)
5813 {
5814         struct net_device *netdev = data;
5815         struct e1000_adapter *adapter = netdev_priv(netdev);
5816
5817         if (adapter->msix_entries) {
5818                 int vector, msix_irq;
5819
5820                 vector = 0;
5821                 msix_irq = adapter->msix_entries[vector].vector;
5822                 disable_irq(msix_irq);
5823                 e1000_intr_msix_rx(msix_irq, netdev);
5824                 enable_irq(msix_irq);
5825
5826                 vector++;
5827                 msix_irq = adapter->msix_entries[vector].vector;
5828                 disable_irq(msix_irq);
5829                 e1000_intr_msix_tx(msix_irq, netdev);
5830                 enable_irq(msix_irq);
5831
5832                 vector++;
5833                 msix_irq = adapter->msix_entries[vector].vector;
5834                 disable_irq(msix_irq);
5835                 e1000_msix_other(msix_irq, netdev);
5836                 enable_irq(msix_irq);
5837         }
5838
5839         return IRQ_HANDLED;
5840 }
5841
5842 /*
5843  * Polling 'interrupt' - used by things like netconsole to send skbs
5844  * without having to re-enable interrupts. It's not called while
5845  * the interrupt routine is executing.
5846  */
5847 static void e1000_netpoll(struct net_device *netdev)
5848 {
5849         struct e1000_adapter *adapter = netdev_priv(netdev);
5850
5851         switch (adapter->int_mode) {
5852         case E1000E_INT_MODE_MSIX:
5853                 e1000_intr_msix(adapter->pdev->irq, netdev);
5854                 break;
5855         case E1000E_INT_MODE_MSI:
5856                 disable_irq(adapter->pdev->irq);
5857                 e1000_intr_msi(adapter->pdev->irq, netdev);
5858                 enable_irq(adapter->pdev->irq);
5859                 break;
5860         default: /* E1000E_INT_MODE_LEGACY */
5861                 disable_irq(adapter->pdev->irq);
5862                 e1000_intr(adapter->pdev->irq, netdev);
5863                 enable_irq(adapter->pdev->irq);
5864                 break;
5865         }
5866 }
5867 #endif
5868
5869 /**
5870  * e1000_io_error_detected - called when PCI error is detected
5871  * @pdev: Pointer to PCI device
5872  * @state: The current pci connection state
5873  *
5874  * This function is called after a PCI bus error affecting
5875  * this device has been detected.
5876  */
5877 static pci_ers_result_t e1000_io_error_detected(struct pci_dev *pdev,
5878                                                 pci_channel_state_t state)
5879 {
5880         struct net_device *netdev = pci_get_drvdata(pdev);
5881         struct e1000_adapter *adapter = netdev_priv(netdev);
5882
5883         netif_device_detach(netdev);
5884
5885         if (state == pci_channel_io_perm_failure)
5886                 return PCI_ERS_RESULT_DISCONNECT;
5887
5888         if (netif_running(netdev))
5889                 e1000e_down(adapter);
5890         pci_disable_device(pdev);
5891
5892         /* Request a slot slot reset. */
5893         return PCI_ERS_RESULT_NEED_RESET;
5894 }
5895
5896 /**
5897  * e1000_io_slot_reset - called after the pci bus has been reset.
5898  * @pdev: Pointer to PCI device
5899  *
5900  * Restart the card from scratch, as if from a cold-boot. Implementation
5901  * resembles the first-half of the e1000_resume routine.
5902  */
5903 static pci_ers_result_t e1000_io_slot_reset(struct pci_dev *pdev)
5904 {
5905         struct net_device *netdev = pci_get_drvdata(pdev);
5906         struct e1000_adapter *adapter = netdev_priv(netdev);
5907         struct e1000_hw *hw = &adapter->hw;
5908         u16 aspm_disable_flag = 0;
5909         int err;
5910         pci_ers_result_t result;
5911
5912         if (adapter->flags2 & FLAG2_DISABLE_ASPM_L0S)
5913                 aspm_disable_flag = PCIE_LINK_STATE_L0S;
5914         if (adapter->flags2 & FLAG2_DISABLE_ASPM_L1)
5915                 aspm_disable_flag |= PCIE_LINK_STATE_L1;
5916         if (aspm_disable_flag)
5917                 e1000e_disable_aspm(pdev, aspm_disable_flag);
5918
5919         err = pci_enable_device_mem(pdev);
5920         if (err) {
5921                 dev_err(&pdev->dev,
5922                         "Cannot re-enable PCI device after reset.\n");
5923                 result = PCI_ERS_RESULT_DISCONNECT;
5924         } else {
5925                 pci_set_master(pdev);
5926                 pdev->state_saved = true;
5927                 pci_restore_state(pdev);
5928
5929                 pci_enable_wake(pdev, PCI_D3hot, 0);
5930                 pci_enable_wake(pdev, PCI_D3cold, 0);
5931
5932                 e1000e_reset(adapter);
5933                 ew32(WUS, ~0);
5934                 result = PCI_ERS_RESULT_RECOVERED;
5935         }
5936
5937         pci_cleanup_aer_uncorrect_error_status(pdev);
5938
5939         return result;
5940 }
5941
5942 /**
5943  * e1000_io_resume - called when traffic can start flowing again.
5944  * @pdev: Pointer to PCI device
5945  *
5946  * This callback is called when the error recovery driver tells us that
5947  * its OK to resume normal operation. Implementation resembles the
5948  * second-half of the e1000_resume routine.
5949  */
5950 static void e1000_io_resume(struct pci_dev *pdev)
5951 {
5952         struct net_device *netdev = pci_get_drvdata(pdev);
5953         struct e1000_adapter *adapter = netdev_priv(netdev);
5954
5955         e1000_init_manageability_pt(adapter);
5956
5957         if (netif_running(netdev)) {
5958                 if (e1000e_up(adapter)) {
5959                         dev_err(&pdev->dev,
5960                                 "can't bring device back up after reset\n");
5961                         return;
5962                 }
5963         }
5964
5965         netif_device_attach(netdev);
5966
5967         /*
5968          * If the controller has AMT, do not set DRV_LOAD until the interface
5969          * is up.  For all other cases, let the f/w know that the h/w is now
5970          * under the control of the driver.
5971          */
5972         if (!(adapter->flags & FLAG_HAS_AMT))
5973                 e1000e_get_hw_control(adapter);
5974
5975 }
5976
5977 static void e1000_print_device_info(struct e1000_adapter *adapter)
5978 {
5979         struct e1000_hw *hw = &adapter->hw;
5980         struct net_device *netdev = adapter->netdev;
5981         u32 ret_val;
5982         u8 pba_str[E1000_PBANUM_LENGTH];
5983
5984         /* print bus type/speed/width info */
5985         e_info("(PCI Express:2.5GT/s:%s) %pM\n",
5986                /* bus width */
5987                ((hw->bus.width == e1000_bus_width_pcie_x4) ? "Width x4" :
5988                 "Width x1"),
5989                /* MAC address */
5990                netdev->dev_addr);
5991         e_info("Intel(R) PRO/%s Network Connection\n",
5992                (hw->phy.type == e1000_phy_ife) ? "10/100" : "1000");
5993         ret_val = e1000_read_pba_string_generic(hw, pba_str,
5994                                                 E1000_PBANUM_LENGTH);
5995         if (ret_val)
5996                 strlcpy((char *)pba_str, "Unknown", sizeof(pba_str));
5997         e_info("MAC: %d, PHY: %d, PBA No: %s\n",
5998                hw->mac.type, hw->phy.type, pba_str);
5999 }
6000
6001 static void e1000_eeprom_checks(struct e1000_adapter *adapter)
6002 {
6003         struct e1000_hw *hw = &adapter->hw;
6004         int ret_val;
6005         u16 buf = 0;
6006
6007         if (hw->mac.type != e1000_82573)
6008                 return;
6009
6010         ret_val = e1000_read_nvm(hw, NVM_INIT_CONTROL2_REG, 1, &buf);
6011         le16_to_cpus(&buf);
6012         if (!ret_val && (!(buf & (1 << 0)))) {
6013                 /* Deep Smart Power Down (DSPD) */
6014                 dev_warn(&adapter->pdev->dev,
6015                          "Warning: detected DSPD enabled in EEPROM\n");
6016         }
6017 }
6018
6019 static int e1000_set_features(struct net_device *netdev,
6020                               netdev_features_t features)
6021 {
6022         struct e1000_adapter *adapter = netdev_priv(netdev);
6023         netdev_features_t changed = features ^ netdev->features;
6024
6025         if (changed & (NETIF_F_TSO | NETIF_F_TSO6))
6026                 adapter->flags |= FLAG_TSO_FORCE;
6027
6028         if (!(changed & (NETIF_F_HW_VLAN_RX | NETIF_F_HW_VLAN_TX |
6029                          NETIF_F_RXCSUM | NETIF_F_RXHASH | NETIF_F_RXFCS |
6030                          NETIF_F_RXALL)))
6031                 return 0;
6032
6033         /*
6034          * IP payload checksum (enabled with jumbos/packet-split when Rx
6035          * checksum is enabled) and generation of RSS hash is mutually
6036          * exclusive in the hardware.
6037          */
6038         if (adapter->rx_ps_pages &&
6039             (features & NETIF_F_RXCSUM) && (features & NETIF_F_RXHASH)) {
6040                 e_err("Enabling both receive checksum offload and receive hashing is not possible with jumbo frames.  Disable jumbos or enable only one of the receive offload features.\n");
6041                 return -EINVAL;
6042         }
6043
6044         if (changed & NETIF_F_RXFCS) {
6045                 if (features & NETIF_F_RXFCS) {
6046                         adapter->flags2 &= ~FLAG2_CRC_STRIPPING;
6047                 } else {
6048                         /* We need to take it back to defaults, which might mean
6049                          * stripping is still disabled at the adapter level.
6050                          */
6051                         if (adapter->flags2 & FLAG2_DFLT_CRC_STRIPPING)
6052                                 adapter->flags2 |= FLAG2_CRC_STRIPPING;
6053                         else
6054                                 adapter->flags2 &= ~FLAG2_CRC_STRIPPING;
6055                 }
6056         }
6057
6058         netdev->features = features;
6059
6060         if (netif_running(netdev))
6061                 e1000e_reinit_locked(adapter);
6062         else
6063                 e1000e_reset(adapter);
6064
6065         return 0;
6066 }
6067
6068 static const struct net_device_ops e1000e_netdev_ops = {
6069         .ndo_open               = e1000_open,
6070         .ndo_stop               = e1000_close,
6071         .ndo_start_xmit         = e1000_xmit_frame,
6072         .ndo_get_stats64        = e1000e_get_stats64,
6073         .ndo_set_rx_mode        = e1000e_set_rx_mode,
6074         .ndo_set_mac_address    = e1000_set_mac,
6075         .ndo_change_mtu         = e1000_change_mtu,
6076         .ndo_do_ioctl           = e1000_ioctl,
6077         .ndo_tx_timeout         = e1000_tx_timeout,
6078         .ndo_validate_addr      = eth_validate_addr,
6079
6080         .ndo_vlan_rx_add_vid    = e1000_vlan_rx_add_vid,
6081         .ndo_vlan_rx_kill_vid   = e1000_vlan_rx_kill_vid,
6082 #ifdef CONFIG_NET_POLL_CONTROLLER
6083         .ndo_poll_controller    = e1000_netpoll,
6084 #endif
6085         .ndo_set_features = e1000_set_features,
6086 };
6087
6088 /**
6089  * e1000_probe - Device Initialization Routine
6090  * @pdev: PCI device information struct
6091  * @ent: entry in e1000_pci_tbl
6092  *
6093  * Returns 0 on success, negative on failure
6094  *
6095  * e1000_probe initializes an adapter identified by a pci_dev structure.
6096  * The OS initialization, configuring of the adapter private structure,
6097  * and a hardware reset occur.
6098  **/
6099 static int __devinit e1000_probe(struct pci_dev *pdev,
6100                                  const struct pci_device_id *ent)
6101 {
6102         struct net_device *netdev;
6103         struct e1000_adapter *adapter;
6104         struct e1000_hw *hw;
6105         const struct e1000_info *ei = e1000_info_tbl[ent->driver_data];
6106         resource_size_t mmio_start, mmio_len;
6107         resource_size_t flash_start, flash_len;
6108         static int cards_found;
6109         u16 aspm_disable_flag = 0;
6110         int i, err, pci_using_dac;
6111         u16 eeprom_data = 0;
6112         u16 eeprom_apme_mask = E1000_EEPROM_APME;
6113
6114         if (ei->flags2 & FLAG2_DISABLE_ASPM_L0S)
6115                 aspm_disable_flag = PCIE_LINK_STATE_L0S;
6116         if (ei->flags2 & FLAG2_DISABLE_ASPM_L1)
6117                 aspm_disable_flag |= PCIE_LINK_STATE_L1;
6118         if (aspm_disable_flag)
6119                 e1000e_disable_aspm(pdev, aspm_disable_flag);
6120
6121         err = pci_enable_device_mem(pdev);
6122         if (err)
6123                 return err;
6124
6125         pci_using_dac = 0;
6126         err = dma_set_mask(&pdev->dev, DMA_BIT_MASK(64));
6127         if (!err) {
6128                 err = dma_set_coherent_mask(&pdev->dev, DMA_BIT_MASK(64));
6129                 if (!err)
6130                         pci_using_dac = 1;
6131         } else {
6132                 err = dma_set_mask(&pdev->dev, DMA_BIT_MASK(32));
6133                 if (err) {
6134                         err = dma_set_coherent_mask(&pdev->dev,
6135                                                     DMA_BIT_MASK(32));
6136                         if (err) {
6137                                 dev_err(&pdev->dev, "No usable DMA configuration, aborting\n");
6138                                 goto err_dma;
6139                         }
6140                 }
6141         }
6142
6143         err = pci_request_selected_regions_exclusive(pdev,
6144                                           pci_select_bars(pdev, IORESOURCE_MEM),
6145                                           e1000e_driver_name);
6146         if (err)
6147                 goto err_pci_reg;
6148
6149         /* AER (Advanced Error Reporting) hooks */
6150         pci_enable_pcie_error_reporting(pdev);
6151
6152         pci_set_master(pdev);
6153         /* PCI config space info */
6154         err = pci_save_state(pdev);
6155         if (err)
6156                 goto err_alloc_etherdev;
6157
6158         err = -ENOMEM;
6159         netdev = alloc_etherdev(sizeof(struct e1000_adapter));
6160         if (!netdev)
6161                 goto err_alloc_etherdev;
6162
6163         SET_NETDEV_DEV(netdev, &pdev->dev);
6164
6165         netdev->irq = pdev->irq;
6166
6167         pci_set_drvdata(pdev, netdev);
6168         adapter = netdev_priv(netdev);
6169         hw = &adapter->hw;
6170         adapter->netdev = netdev;
6171         adapter->pdev = pdev;
6172         adapter->ei = ei;
6173         adapter->pba = ei->pba;
6174         adapter->flags = ei->flags;
6175         adapter->flags2 = ei->flags2;
6176         adapter->hw.adapter = adapter;
6177         adapter->hw.mac.type = ei->mac;
6178         adapter->max_hw_frame_size = ei->max_hw_frame_size;
6179         adapter->msg_enable = netif_msg_init(debug, DEFAULT_MSG_ENABLE);
6180
6181         mmio_start = pci_resource_start(pdev, 0);
6182         mmio_len = pci_resource_len(pdev, 0);
6183
6184         err = -EIO;
6185         adapter->hw.hw_addr = ioremap(mmio_start, mmio_len);
6186         if (!adapter->hw.hw_addr)
6187                 goto err_ioremap;
6188
6189         if ((adapter->flags & FLAG_HAS_FLASH) &&
6190             (pci_resource_flags(pdev, 1) & IORESOURCE_MEM)) {
6191                 flash_start = pci_resource_start(pdev, 1);
6192                 flash_len = pci_resource_len(pdev, 1);
6193                 adapter->hw.flash_address = ioremap(flash_start, flash_len);
6194                 if (!adapter->hw.flash_address)
6195                         goto err_flashmap;
6196         }
6197
6198         /* construct the net_device struct */
6199         netdev->netdev_ops              = &e1000e_netdev_ops;
6200         e1000e_set_ethtool_ops(netdev);
6201         netdev->watchdog_timeo          = 5 * HZ;
6202         netif_napi_add(netdev, &adapter->napi, e1000e_poll, 64);
6203         strlcpy(netdev->name, pci_name(pdev), sizeof(netdev->name));
6204
6205         netdev->mem_start = mmio_start;
6206         netdev->mem_end = mmio_start + mmio_len;
6207
6208         adapter->bd_number = cards_found++;
6209
6210         e1000e_check_options(adapter);
6211
6212         /* setup adapter struct */
6213         err = e1000_sw_init(adapter);
6214         if (err)
6215                 goto err_sw_init;
6216
6217         memcpy(&hw->mac.ops, ei->mac_ops, sizeof(hw->mac.ops));
6218         memcpy(&hw->nvm.ops, ei->nvm_ops, sizeof(hw->nvm.ops));
6219         memcpy(&hw->phy.ops, ei->phy_ops, sizeof(hw->phy.ops));
6220
6221         err = ei->get_variants(adapter);
6222         if (err)
6223                 goto err_hw_init;
6224
6225         if ((adapter->flags & FLAG_IS_ICH) &&
6226             (adapter->flags & FLAG_READ_ONLY_NVM))
6227                 e1000e_write_protect_nvm_ich8lan(&adapter->hw);
6228
6229         hw->mac.ops.get_bus_info(&adapter->hw);
6230
6231         adapter->hw.phy.autoneg_wait_to_complete = 0;
6232
6233         /* Copper options */
6234         if (adapter->hw.phy.media_type == e1000_media_type_copper) {
6235                 adapter->hw.phy.mdix = AUTO_ALL_MODES;
6236                 adapter->hw.phy.disable_polarity_correction = 0;
6237                 adapter->hw.phy.ms_type = e1000_ms_hw_default;
6238         }
6239
6240         if (hw->phy.ops.check_reset_block(hw))
6241                 e_info("PHY reset is blocked due to SOL/IDER session.\n");
6242
6243         /* Set initial default active device features */
6244         netdev->features = (NETIF_F_SG |
6245                             NETIF_F_HW_VLAN_RX |
6246                             NETIF_F_HW_VLAN_TX |
6247                             NETIF_F_TSO |
6248                             NETIF_F_TSO6 |
6249                             NETIF_F_RXHASH |
6250                             NETIF_F_RXCSUM |
6251                             NETIF_F_HW_CSUM);
6252
6253         /* Set user-changeable features (subset of all device features) */
6254         netdev->hw_features = netdev->features;
6255         netdev->hw_features |= NETIF_F_RXFCS;
6256         netdev->priv_flags |= IFF_SUPP_NOFCS;
6257         netdev->hw_features |= NETIF_F_RXALL;
6258
6259         if (adapter->flags & FLAG_HAS_HW_VLAN_FILTER)
6260                 netdev->features |= NETIF_F_HW_VLAN_FILTER;
6261
6262         netdev->vlan_features |= (NETIF_F_SG |
6263                                   NETIF_F_TSO |
6264                                   NETIF_F_TSO6 |
6265                                   NETIF_F_HW_CSUM);
6266
6267         netdev->priv_flags |= IFF_UNICAST_FLT;
6268
6269         if (pci_using_dac) {
6270                 netdev->features |= NETIF_F_HIGHDMA;
6271                 netdev->vlan_features |= NETIF_F_HIGHDMA;
6272         }
6273
6274         if (e1000e_enable_mng_pass_thru(&adapter->hw))
6275                 adapter->flags |= FLAG_MNG_PT_ENABLED;
6276
6277         /*
6278          * before reading the NVM, reset the controller to
6279          * put the device in a known good starting state
6280          */
6281         adapter->hw.mac.ops.reset_hw(&adapter->hw);
6282
6283         /*
6284          * systems with ASPM and others may see the checksum fail on the first
6285          * attempt. Let's give it a few tries
6286          */
6287         for (i = 0;; i++) {
6288                 if (e1000_validate_nvm_checksum(&adapter->hw) >= 0)
6289                         break;
6290                 if (i == 2) {
6291                         e_err("The NVM Checksum Is Not Valid\n");
6292                         err = -EIO;
6293                         goto err_eeprom;
6294                 }
6295         }
6296
6297         e1000_eeprom_checks(adapter);
6298
6299         /* copy the MAC address */
6300         if (e1000e_read_mac_addr(&adapter->hw))
6301                 e_err("NVM Read Error while reading MAC address\n");
6302
6303         memcpy(netdev->dev_addr, adapter->hw.mac.addr, netdev->addr_len);
6304         memcpy(netdev->perm_addr, adapter->hw.mac.addr, netdev->addr_len);
6305
6306         if (!is_valid_ether_addr(netdev->perm_addr)) {
6307                 e_err("Invalid MAC Address: %pM\n", netdev->perm_addr);
6308                 err = -EIO;
6309                 goto err_eeprom;
6310         }
6311
6312         init_timer(&adapter->watchdog_timer);
6313         adapter->watchdog_timer.function = e1000_watchdog;
6314         adapter->watchdog_timer.data = (unsigned long) adapter;
6315
6316         init_timer(&adapter->phy_info_timer);
6317         adapter->phy_info_timer.function = e1000_update_phy_info;
6318         adapter->phy_info_timer.data = (unsigned long) adapter;
6319
6320         INIT_WORK(&adapter->reset_task, e1000_reset_task);
6321         INIT_WORK(&adapter->watchdog_task, e1000_watchdog_task);
6322         INIT_WORK(&adapter->downshift_task, e1000e_downshift_workaround);
6323         INIT_WORK(&adapter->update_phy_task, e1000e_update_phy_task);
6324         INIT_WORK(&adapter->print_hang_task, e1000_print_hw_hang);
6325
6326         /* Initialize link parameters. User can change them with ethtool */
6327         adapter->hw.mac.autoneg = 1;
6328         adapter->fc_autoneg = true;
6329         adapter->hw.fc.requested_mode = e1000_fc_default;
6330         adapter->hw.fc.current_mode = e1000_fc_default;
6331         adapter->hw.phy.autoneg_advertised = 0x2f;
6332
6333         /* ring size defaults */
6334         adapter->rx_ring->count = 256;
6335         adapter->tx_ring->count = 256;
6336
6337         /*
6338          * Initial Wake on LAN setting - If APM wake is enabled in
6339          * the EEPROM, enable the ACPI Magic Packet filter
6340          */
6341         if (adapter->flags & FLAG_APME_IN_WUC) {
6342                 /* APME bit in EEPROM is mapped to WUC.APME */
6343                 eeprom_data = er32(WUC);
6344                 eeprom_apme_mask = E1000_WUC_APME;
6345                 if ((hw->mac.type > e1000_ich10lan) &&
6346                     (eeprom_data & E1000_WUC_PHY_WAKE))
6347                         adapter->flags2 |= FLAG2_HAS_PHY_WAKEUP;
6348         } else if (adapter->flags & FLAG_APME_IN_CTRL3) {
6349                 if (adapter->flags & FLAG_APME_CHECK_PORT_B &&
6350                     (adapter->hw.bus.func == 1))
6351                         e1000_read_nvm(&adapter->hw, NVM_INIT_CONTROL3_PORT_B,
6352                                        1, &eeprom_data);
6353                 else
6354                         e1000_read_nvm(&adapter->hw, NVM_INIT_CONTROL3_PORT_A,
6355                                        1, &eeprom_data);
6356         }
6357
6358         /* fetch WoL from EEPROM */
6359         if (eeprom_data & eeprom_apme_mask)
6360                 adapter->eeprom_wol |= E1000_WUFC_MAG;
6361
6362         /*
6363          * now that we have the eeprom settings, apply the special cases
6364          * where the eeprom may be wrong or the board simply won't support
6365          * wake on lan on a particular port
6366          */
6367         if (!(adapter->flags & FLAG_HAS_WOL))
6368                 adapter->eeprom_wol = 0;
6369
6370         /* initialize the wol settings based on the eeprom settings */
6371         adapter->wol = adapter->eeprom_wol;
6372         device_set_wakeup_enable(&adapter->pdev->dev, adapter->wol);
6373
6374         /* save off EEPROM version number */
6375         e1000_read_nvm(&adapter->hw, 5, 1, &adapter->eeprom_vers);
6376
6377         /* reset the hardware with the new settings */
6378         e1000e_reset(adapter);
6379
6380         /*
6381          * If the controller has AMT, do not set DRV_LOAD until the interface
6382          * is up.  For all other cases, let the f/w know that the h/w is now
6383          * under the control of the driver.
6384          */
6385         if (!(adapter->flags & FLAG_HAS_AMT))
6386                 e1000e_get_hw_control(adapter);
6387
6388         strlcpy(netdev->name, "eth%d", sizeof(netdev->name));
6389         err = register_netdev(netdev);
6390         if (err)
6391                 goto err_register;
6392
6393         /* carrier off reporting is important to ethtool even BEFORE open */
6394         netif_carrier_off(netdev);
6395
6396         e1000_print_device_info(adapter);
6397
6398         if (pci_dev_run_wake(pdev))
6399                 pm_runtime_put_noidle(&pdev->dev);
6400
6401         return 0;
6402
6403 err_register:
6404         if (!(adapter->flags & FLAG_HAS_AMT))
6405                 e1000e_release_hw_control(adapter);
6406 err_eeprom:
6407         if (!hw->phy.ops.check_reset_block(hw))
6408                 e1000_phy_hw_reset(&adapter->hw);
6409 err_hw_init:
6410         kfree(adapter->tx_ring);
6411         kfree(adapter->rx_ring);
6412 err_sw_init:
6413         if (adapter->hw.flash_address)
6414                 iounmap(adapter->hw.flash_address);
6415         e1000e_reset_interrupt_capability(adapter);
6416 err_flashmap:
6417         iounmap(adapter->hw.hw_addr);
6418 err_ioremap:
6419         free_netdev(netdev);
6420 err_alloc_etherdev:
6421         pci_release_selected_regions(pdev,
6422                                      pci_select_bars(pdev, IORESOURCE_MEM));
6423 err_pci_reg:
6424 err_dma:
6425         pci_disable_device(pdev);
6426         return err;
6427 }
6428
6429 /**
6430  * e1000_remove - Device Removal Routine
6431  * @pdev: PCI device information struct
6432  *
6433  * e1000_remove is called by the PCI subsystem to alert the driver
6434  * that it should release a PCI device.  The could be caused by a
6435  * Hot-Plug event, or because the driver is going to be removed from
6436  * memory.
6437  **/
6438 static void __devexit e1000_remove(struct pci_dev *pdev)
6439 {
6440         struct net_device *netdev = pci_get_drvdata(pdev);
6441         struct e1000_adapter *adapter = netdev_priv(netdev);
6442         bool down = test_bit(__E1000_DOWN, &adapter->state);
6443
6444         /*
6445          * The timers may be rescheduled, so explicitly disable them
6446          * from being rescheduled.
6447          */
6448         if (!down)
6449                 set_bit(__E1000_DOWN, &adapter->state);
6450         del_timer_sync(&adapter->watchdog_timer);
6451         del_timer_sync(&adapter->phy_info_timer);
6452
6453         cancel_work_sync(&adapter->reset_task);
6454         cancel_work_sync(&adapter->watchdog_task);
6455         cancel_work_sync(&adapter->downshift_task);
6456         cancel_work_sync(&adapter->update_phy_task);
6457         cancel_work_sync(&adapter->print_hang_task);
6458
6459         if (!(netdev->flags & IFF_UP))
6460                 e1000_power_down_phy(adapter);
6461
6462         /* Don't lie to e1000_close() down the road. */
6463         if (!down)
6464                 clear_bit(__E1000_DOWN, &adapter->state);
6465         unregister_netdev(netdev);
6466
6467         if (pci_dev_run_wake(pdev))
6468                 pm_runtime_get_noresume(&pdev->dev);
6469
6470         /*
6471          * Release control of h/w to f/w.  If f/w is AMT enabled, this
6472          * would have already happened in close and is redundant.
6473          */
6474         e1000e_release_hw_control(adapter);
6475
6476         e1000e_reset_interrupt_capability(adapter);
6477         kfree(adapter->tx_ring);
6478         kfree(adapter->rx_ring);
6479
6480         iounmap(adapter->hw.hw_addr);
6481         if (adapter->hw.flash_address)
6482                 iounmap(adapter->hw.flash_address);
6483         pci_release_selected_regions(pdev,
6484                                      pci_select_bars(pdev, IORESOURCE_MEM));
6485
6486         free_netdev(netdev);
6487
6488         /* AER disable */
6489         pci_disable_pcie_error_reporting(pdev);
6490
6491         pci_disable_device(pdev);
6492 }
6493
6494 /* PCI Error Recovery (ERS) */
6495 static struct pci_error_handlers e1000_err_handler = {
6496         .error_detected = e1000_io_error_detected,
6497         .slot_reset = e1000_io_slot_reset,
6498         .resume = e1000_io_resume,
6499 };
6500
6501 static DEFINE_PCI_DEVICE_TABLE(e1000_pci_tbl) = {
6502         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82571EB_COPPER), board_82571 },
6503         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82571EB_FIBER), board_82571 },
6504         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82571EB_QUAD_COPPER), board_82571 },
6505         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82571EB_QUAD_COPPER_LP), board_82571 },
6506         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82571EB_QUAD_FIBER), board_82571 },
6507         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82571EB_SERDES), board_82571 },
6508         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82571EB_SERDES_DUAL), board_82571 },
6509         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82571EB_SERDES_QUAD), board_82571 },
6510         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82571PT_QUAD_COPPER), board_82571 },
6511
6512         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82572EI), board_82572 },
6513         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82572EI_COPPER), board_82572 },
6514         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82572EI_FIBER), board_82572 },
6515         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82572EI_SERDES), board_82572 },
6516
6517         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82573E), board_82573 },
6518         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82573E_IAMT), board_82573 },
6519         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82573L), board_82573 },
6520
6521         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82574L), board_82574 },
6522         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82574LA), board_82574 },
6523         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82583V), board_82583 },
6524
6525         { PCI_VDEVICE(INTEL, E1000_DEV_ID_80003ES2LAN_COPPER_DPT),
6526           board_80003es2lan },
6527         { PCI_VDEVICE(INTEL, E1000_DEV_ID_80003ES2LAN_COPPER_SPT),
6528           board_80003es2lan },
6529         { PCI_VDEVICE(INTEL, E1000_DEV_ID_80003ES2LAN_SERDES_DPT),
6530           board_80003es2lan },
6531         { PCI_VDEVICE(INTEL, E1000_DEV_ID_80003ES2LAN_SERDES_SPT),
6532           board_80003es2lan },
6533
6534         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH8_IFE), board_ich8lan },
6535         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH8_IFE_G), board_ich8lan },
6536         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH8_IFE_GT), board_ich8lan },
6537         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH8_IGP_AMT), board_ich8lan },
6538         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH8_IGP_C), board_ich8lan },
6539         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH8_IGP_M), board_ich8lan },
6540         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH8_IGP_M_AMT), board_ich8lan },
6541         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH8_82567V_3), board_ich8lan },
6542
6543         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH9_IFE), board_ich9lan },
6544         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH9_IFE_G), board_ich9lan },
6545         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH9_IFE_GT), board_ich9lan },
6546         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH9_IGP_AMT), board_ich9lan },
6547         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH9_IGP_C), board_ich9lan },
6548         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH9_BM), board_ich9lan },
6549         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH9_IGP_M), board_ich9lan },
6550         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH9_IGP_M_AMT), board_ich9lan },
6551         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH9_IGP_M_V), board_ich9lan },
6552
6553         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH10_R_BM_LM), board_ich9lan },
6554         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH10_R_BM_LF), board_ich9lan },
6555         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH10_R_BM_V), board_ich9lan },
6556
6557         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH10_D_BM_LM), board_ich10lan },
6558         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH10_D_BM_LF), board_ich10lan },
6559         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH10_D_BM_V), board_ich10lan },
6560
6561         { PCI_VDEVICE(INTEL, E1000_DEV_ID_PCH_M_HV_LM), board_pchlan },
6562         { PCI_VDEVICE(INTEL, E1000_DEV_ID_PCH_M_HV_LC), board_pchlan },
6563         { PCI_VDEVICE(INTEL, E1000_DEV_ID_PCH_D_HV_DM), board_pchlan },
6564         { PCI_VDEVICE(INTEL, E1000_DEV_ID_PCH_D_HV_DC), board_pchlan },
6565
6566         { PCI_VDEVICE(INTEL, E1000_DEV_ID_PCH2_LV_LM), board_pch2lan },
6567         { PCI_VDEVICE(INTEL, E1000_DEV_ID_PCH2_LV_V), board_pch2lan },
6568
6569         { PCI_VDEVICE(INTEL, E1000_DEV_ID_PCH_LPT_I217_LM), board_pch_lpt },
6570         { PCI_VDEVICE(INTEL, E1000_DEV_ID_PCH_LPT_I217_V), board_pch_lpt },
6571
6572         { 0, 0, 0, 0, 0, 0, 0 } /* terminate list */
6573 };
6574 MODULE_DEVICE_TABLE(pci, e1000_pci_tbl);
6575
6576 #ifdef CONFIG_PM
6577 static const struct dev_pm_ops e1000_pm_ops = {
6578         SET_SYSTEM_SLEEP_PM_OPS(e1000_suspend, e1000_resume)
6579         SET_RUNTIME_PM_OPS(e1000_runtime_suspend,
6580                                 e1000_runtime_resume, e1000_idle)
6581 };
6582 #endif
6583
6584 /* PCI Device API Driver */
6585 static struct pci_driver e1000_driver = {
6586         .name     = e1000e_driver_name,
6587         .id_table = e1000_pci_tbl,
6588         .probe    = e1000_probe,
6589         .remove   = __devexit_p(e1000_remove),
6590 #ifdef CONFIG_PM
6591         .driver   = {
6592                 .pm = &e1000_pm_ops,
6593         },
6594 #endif
6595         .shutdown = e1000_shutdown,
6596         .err_handler = &e1000_err_handler
6597 };
6598
6599 /**
6600  * e1000_init_module - Driver Registration Routine
6601  *
6602  * e1000_init_module is the first routine called when the driver is
6603  * loaded. All it does is register with the PCI subsystem.
6604  **/
6605 static int __init e1000_init_module(void)
6606 {
6607         int ret;
6608         pr_info("Intel(R) PRO/1000 Network Driver - %s\n",
6609                 e1000e_driver_version);
6610         pr_info("Copyright(c) 1999 - 2012 Intel Corporation.\n");
6611         ret = pci_register_driver(&e1000_driver);
6612
6613         return ret;
6614 }
6615 module_init(e1000_init_module);
6616
6617 /**
6618  * e1000_exit_module - Driver Exit Cleanup Routine
6619  *
6620  * e1000_exit_module is called just before the driver is removed
6621  * from memory.
6622  **/
6623 static void __exit e1000_exit_module(void)
6624 {
6625         pci_unregister_driver(&e1000_driver);
6626 }
6627 module_exit(e1000_exit_module);
6628
6629
6630 MODULE_AUTHOR("Intel Corporation, <linux.nics@intel.com>");
6631 MODULE_DESCRIPTION("Intel(R) PRO/1000 Network Driver");
6632 MODULE_LICENSE("GPL");
6633 MODULE_VERSION(DRV_VERSION);
6634
6635 /* netdev.c */