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

   1 /* Intel PRO/1000 Linux driver
   2  * Copyright(c) 1999 - 2014 Intel Corporation.
   3  *
   4  * This program is free software; you can redistribute it and/or modify it
   5  * under the terms and conditions of the GNU General Public License,
   6  * version 2, as published by the Free Software Foundation.
   7  *
   8  * This program is distributed in the hope it will be useful, but WITHOUT
   9  * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
  10  * FITNESS FOR A PARTICULAR PURPOSE.  See the GNU General Public License for
  11  * more details.
  12  *
  13  * The full GNU General Public License is included in this distribution in
  14  * the file called "COPYING".
  15  *
  16  * Contact Information:
  17  * Linux NICS <linux.nics@intel.com>
  18  * e1000-devel Mailing List <e1000-devel@lists.sourceforge.net>
  19  * Intel Corporation, 5200 N.E. Elam Young Parkway, Hillsboro, OR 97124-6497
  20  */
  21
  22 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
  23
  24 #include <linux/module.h>
  25 #include <linux/types.h>
  26 #include <linux/init.h>
  27 #include <linux/pci.h>
  28 #include <linux/vmalloc.h>
  29 #include <linux/pagemap.h>
  30 #include <linux/delay.h>
  31 #include <linux/netdevice.h>
  32 #include <linux/interrupt.h>
  33 #include <linux/tcp.h>
  34 #include <linux/ipv6.h>
  35 #include <linux/slab.h>
  36 #include <net/checksum.h>
  37 #include <net/ip6_checksum.h>
  38 #include <linux/ethtool.h>
  39 #include <linux/if_vlan.h>
  40 #include <linux/cpu.h>
  41 #include <linux/smp.h>
  42 #include <linux/pm_qos.h>
  43 #include <linux/pm_runtime.h>
  44 #include <linux/aer.h>
  45 #include <linux/prefetch.h>
  46
  47 #include "e1000.h"
  48
  49 #define DRV_EXTRAVERSION "-k"
  50
  51 #define DRV_VERSION "2.3.2" DRV_EXTRAVERSION
  52 char e1000e_driver_name[] = "e1000e";
  53 const char e1000e_driver_version[] = DRV_VERSION;
  54
  55 #define DEFAULT_MSG_ENABLE (NETIF_MSG_DRV|NETIF_MSG_PROBE|NETIF_MSG_LINK)
  56 static int debug = -1;
  57 module_param(debug, int, 0);
  58 MODULE_PARM_DESC(debug, "Debug level (0=none,...,16=all)");
  59
  60 static const struct e1000_info *e1000_info_tbl[] = {
  61         [board_82571]           = &e1000_82571_info,
  62         [board_82572]           = &e1000_82572_info,
  63         [board_82573]           = &e1000_82573_info,
  64         [board_82574]           = &e1000_82574_info,
  65         [board_82583]           = &e1000_82583_info,
  66         [board_80003es2lan]     = &e1000_es2_info,
  67         [board_ich8lan]         = &e1000_ich8_info,
  68         [board_ich9lan]         = &e1000_ich9_info,
  69         [board_ich10lan]        = &e1000_ich10_info,
  70         [board_pchlan]          = &e1000_pch_info,
  71         [board_pch2lan]         = &e1000_pch2_info,
  72         [board_pch_lpt]         = &e1000_pch_lpt_info,
  73 };
  74
  75 struct e1000_reg_info {
  76         u32 ofs;
  77         char *name;
  78 };
  79
  80 static const struct e1000_reg_info e1000_reg_info_tbl[] = {
  81         /* General Registers */
  82         {E1000_CTRL, "CTRL"},
  83         {E1000_STATUS, "STATUS"},
  84         {E1000_CTRL_EXT, "CTRL_EXT"},
  85
  86         /* Interrupt Registers */
  87         {E1000_ICR, "ICR"},
  88
  89         /* Rx Registers */
  90         {E1000_RCTL, "RCTL"},
  91         {E1000_RDLEN(0), "RDLEN"},
  92         {E1000_RDH(0), "RDH"},
  93         {E1000_RDT(0), "RDT"},
  94         {E1000_RDTR, "RDTR"},
  95         {E1000_RXDCTL(0), "RXDCTL"},
  96         {E1000_ERT, "ERT"},
  97         {E1000_RDBAL(0), "RDBAL"},
  98         {E1000_RDBAH(0), "RDBAH"},
  99         {E1000_RDFH, "RDFH"},
 100         {E1000_RDFT, "RDFT"},
 101         {E1000_RDFHS, "RDFHS"},
 102         {E1000_RDFTS, "RDFTS"},
 103         {E1000_RDFPC, "RDFPC"},
 104
 105         /* Tx Registers */
 106         {E1000_TCTL, "TCTL"},
 107         {E1000_TDBAL(0), "TDBAL"},
 108         {E1000_TDBAH(0), "TDBAH"},
 109         {E1000_TDLEN(0), "TDLEN"},
 110         {E1000_TDH(0), "TDH"},
 111         {E1000_TDT(0), "TDT"},
 112         {E1000_TIDV, "TIDV"},
 113         {E1000_TXDCTL(0), "TXDCTL"},
 114         {E1000_TADV, "TADV"},
 115         {E1000_TARC(0), "TARC"},
 116         {E1000_TDFH, "TDFH"},
 117         {E1000_TDFT, "TDFT"},
 118         {E1000_TDFHS, "TDFHS"},
 119         {E1000_TDFTS, "TDFTS"},
 120         {E1000_TDFPC, "TDFPC"},
 121
 122         /* List Terminator */
 123         {0, NULL}
 124 };
 125
 126 /**
 127  * e1000_regdump - register printout routine
 128  * @hw: pointer to the HW structure
 129  * @reginfo: pointer to the register info table
 130  **/
 131 static void e1000_regdump(struct e1000_hw *hw, struct e1000_reg_info *reginfo)
 132 {
 133         int n = 0;
 134         char rname[16];
 135         u32 regs[8];
 136
 137         switch (reginfo->ofs) {
 138         case E1000_RXDCTL(0):
 139                 for (n = 0; n < 2; n++)
 140                         regs[n] = __er32(hw, E1000_RXDCTL(n));
 141                 break;
 142         case E1000_TXDCTL(0):
 143                 for (n = 0; n < 2; n++)
 144                         regs[n] = __er32(hw, E1000_TXDCTL(n));
 145                 break;
 146         case E1000_TARC(0):
 147                 for (n = 0; n < 2; n++)
 148                         regs[n] = __er32(hw, E1000_TARC(n));
 149                 break;
 150         default:
 151                 pr_info("%-15s %08x\n",
 152                         reginfo->name, __er32(hw, reginfo->ofs));
 153                 return;
 154         }
 155
 156         snprintf(rname, 16, "%s%s", reginfo->name, "[0-1]");
 157         pr_info("%-15s %08x %08x\n", rname, regs[0], regs[1]);
 158 }
 159
 160 static void e1000e_dump_ps_pages(struct e1000_adapter *adapter,
 161                                  struct e1000_buffer *bi)
 162 {
 163         int i;
 164         struct e1000_ps_page *ps_page;
 165
 166         for (i = 0; i < adapter->rx_ps_pages; i++) {
 167                 ps_page = &bi->ps_pages[i];
 168
 169                 if (ps_page->page) {
 170                         pr_info("packet dump for ps_page %d:\n", i);
 171                         print_hex_dump(KERN_INFO, "", DUMP_PREFIX_ADDRESS,
 172                                        16, 1, page_address(ps_page->page),
 173                                        PAGE_SIZE, true);
 174                 }
 175         }
 176 }
 177
 178 /**
 179  * e1000e_dump - Print registers, Tx-ring and Rx-ring
 180  * @adapter: board private structure
 181  **/
 182 static void e1000e_dump(struct e1000_adapter *adapter)
 183 {
 184         struct net_device *netdev = adapter->netdev;
 185         struct e1000_hw *hw = &adapter->hw;
 186         struct e1000_reg_info *reginfo;
 187         struct e1000_ring *tx_ring = adapter->tx_ring;
 188         struct e1000_tx_desc *tx_desc;
 189         struct my_u0 {
 190                 __le64 a;
 191                 __le64 b;
 192         } *u0;
 193         struct e1000_buffer *buffer_info;
 194         struct e1000_ring *rx_ring = adapter->rx_ring;
 195         union e1000_rx_desc_packet_split *rx_desc_ps;
 196         union e1000_rx_desc_extended *rx_desc;
 197         struct my_u1 {
 198                 __le64 a;
 199                 __le64 b;
 200                 __le64 c;
 201                 __le64 d;
 202         } *u1;
 203         u32 staterr;
 204         int i = 0;
 205
 206         if (!netif_msg_hw(adapter))
 207                 return;
 208
 209         /* Print netdevice Info */
 210         if (netdev) {
 211                 dev_info(&adapter->pdev->dev, "Net device Info\n");
 212                 pr_info("Device Name     state            trans_start      last_rx\n");
 213                 pr_info("%-15s %016lX %016lX %016lX\n", netdev->name,
 214                         netdev->state, netdev->trans_start, netdev->last_rx);
 215         }
 216
 217         /* Print Registers */
 218         dev_info(&adapter->pdev->dev, "Register Dump\n");
 219         pr_info(" Register Name   Value\n");
 220         for (reginfo = (struct e1000_reg_info *)e1000_reg_info_tbl;
 221              reginfo->name; reginfo++) {
 222                 e1000_regdump(hw, reginfo);
 223         }
 224
 225         /* Print Tx Ring Summary */
 226         if (!netdev || !netif_running(netdev))
 227                 return;
 228
 229         dev_info(&adapter->pdev->dev, "Tx Ring Summary\n");
 230         pr_info("Queue [NTU] [NTC] [bi(ntc)->dma  ] leng ntw timestamp\n");
 231         buffer_info = &tx_ring->buffer_info[tx_ring->next_to_clean];
 232         pr_info(" %5d %5X %5X %016llX %04X %3X %016llX\n",
 233                 0, tx_ring->next_to_use, tx_ring->next_to_clean,
 234                 (unsigned long long)buffer_info->dma,
 235                 buffer_info->length,
 236                 buffer_info->next_to_watch,
 237                 (unsigned long long)buffer_info->time_stamp);
 238
 239         /* Print Tx Ring */
 240         if (!netif_msg_tx_done(adapter))
 241                 goto rx_ring_summary;
 242
 243         dev_info(&adapter->pdev->dev, "Tx Ring Dump\n");
 244
 245         /* Transmit Descriptor Formats - DEXT[29] is 0 (Legacy) or 1 (Extended)
 246          *
 247          * Legacy Transmit Descriptor
 248          *   +--------------------------------------------------------------+
 249          * 0 |         Buffer Address [63:0] (Reserved on Write Back)       |
 250          *   +--------------------------------------------------------------+
 251          * 8 | Special  |    CSS     | Status |  CMD    |  CSO   |  Length  |
 252          *   +--------------------------------------------------------------+
 253          *   63       48 47        36 35    32 31     24 23    16 15        0
 254          *
 255          * Extended Context Descriptor (DTYP=0x0) for TSO or checksum offload
 256          *   63      48 47    40 39       32 31             16 15    8 7      0
 257          *   +----------------------------------------------------------------+
 258          * 0 |  TUCSE  | TUCS0  |   TUCSS   |     IPCSE       | IPCS0 | IPCSS |
 259          *   +----------------------------------------------------------------+
 260          * 8 |   MSS   | HDRLEN | RSV | STA | TUCMD | DTYP |      PAYLEN      |
 261          *   +----------------------------------------------------------------+
 262          *   63      48 47    40 39 36 35 32 31   24 23  20 19                0
 263          *
 264          * Extended Data Descriptor (DTYP=0x1)
 265          *   +----------------------------------------------------------------+
 266          * 0 |                     Buffer Address [63:0]                      |
 267          *   +----------------------------------------------------------------+
 268          * 8 | VLAN tag |  POPTS  | Rsvd | Status | Command | DTYP |  DTALEN  |
 269          *   +----------------------------------------------------------------+
 270          *   63       48 47     40 39  36 35    32 31     24 23  20 19        0
 271          */
 272         pr_info("Tl[desc]     [address 63:0  ] [SpeCssSCmCsLen] [bi->dma       ] leng  ntw timestamp        bi->skb <-- Legacy format\n");
 273         pr_info("Tc[desc]     [Ce CoCsIpceCoS] [MssHlRSCm0Plen] [bi->dma       ] leng  ntw timestamp        bi->skb <-- Ext Context format\n");
 274         pr_info("Td[desc]     [address 63:0  ] [VlaPoRSCm1Dlen] [bi->dma       ] leng  ntw timestamp        bi->skb <-- Ext Data format\n");
 275         for (i = 0; tx_ring->desc && (i < tx_ring->count); i++) {
 276                 const char *next_desc;
 277                 tx_desc = E1000_TX_DESC(*tx_ring, i);
 278                 buffer_info = &tx_ring->buffer_info[i];
 279                 u0 = (struct my_u0 *)tx_desc;
 280                 if (i == tx_ring->next_to_use && i == tx_ring->next_to_clean)
 281                         next_desc = " NTC/U";
 282                 else if (i == tx_ring->next_to_use)
 283                         next_desc = " NTU";
 284                 else if (i == tx_ring->next_to_clean)
 285                         next_desc = " NTC";
 286                 else
 287                         next_desc = "";
 288                 pr_info("T%c[0x%03X]    %016llX %016llX %016llX %04X  %3X %016llX %p%s\n",
 289                         (!(le64_to_cpu(u0->b) & (1 << 29)) ? 'l' :
 290                          ((le64_to_cpu(u0->b) & (1 << 20)) ? 'd' : 'c')),
 291                         i,
 292                         (unsigned long long)le64_to_cpu(u0->a),
 293                         (unsigned long long)le64_to_cpu(u0->b),
 294                         (unsigned long long)buffer_info->dma,
 295                         buffer_info->length, buffer_info->next_to_watch,
 296                         (unsigned long long)buffer_info->time_stamp,
 297                         buffer_info->skb, next_desc);
 298
 299                 if (netif_msg_pktdata(adapter) && buffer_info->skb)
 300                         print_hex_dump(KERN_INFO, "", DUMP_PREFIX_ADDRESS,
 301                                        16, 1, buffer_info->skb->data,
 302                                        buffer_info->skb->len, true);
 303         }
 304
 305         /* Print Rx Ring Summary */
 306 rx_ring_summary:
 307         dev_info(&adapter->pdev->dev, "Rx Ring Summary\n");
 308         pr_info("Queue [NTU] [NTC]\n");
 309         pr_info(" %5d %5X %5X\n",
 310                 0, rx_ring->next_to_use, rx_ring->next_to_clean);
 311
 312         /* Print Rx Ring */
 313         if (!netif_msg_rx_status(adapter))
 314                 return;
 315
 316         dev_info(&adapter->pdev->dev, "Rx Ring Dump\n");
 317         switch (adapter->rx_ps_pages) {
 318         case 1:
 319         case 2:
 320         case 3:
 321                 /* [Extended] Packet Split Receive Descriptor Format
 322                  *
 323                  *    +-----------------------------------------------------+
 324                  *  0 |                Buffer Address 0 [63:0]              |
 325                  *    +-----------------------------------------------------+
 326                  *  8 |                Buffer Address 1 [63:0]              |
 327                  *    +-----------------------------------------------------+
 328                  * 16 |                Buffer Address 2 [63:0]              |
 329                  *    +-----------------------------------------------------+
 330                  * 24 |                Buffer Address 3 [63:0]              |
 331                  *    +-----------------------------------------------------+
 332                  */
 333                 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");
 334                 /* [Extended] Receive Descriptor (Write-Back) Format
 335                  *
 336                  *   63       48 47    32 31     13 12    8 7    4 3        0
 337                  *   +------------------------------------------------------+
 338                  * 0 | Packet   | IP     |  Rsvd   | MRQ   | Rsvd | MRQ RSS |
 339                  *   | Checksum | Ident  |         | Queue |      |  Type   |
 340                  *   +------------------------------------------------------+
 341                  * 8 | VLAN Tag | Length | Extended Error | Extended Status |
 342                  *   +------------------------------------------------------+
 343                  *   63       48 47    32 31            20 19               0
 344                  */
 345                 pr_info("RWB[desc]      [ck ipid mrqhsh] [vl   l0 ee  es] [ l3  l2  l1 hs] [reserved      ] ---------------- [bi->skb] <-- Ext Rx Write-Back format\n");
 346                 for (i = 0; i < rx_ring->count; i++) {
 347                         const char *next_desc;
 348                         buffer_info = &rx_ring->buffer_info[i];
 349                         rx_desc_ps = E1000_RX_DESC_PS(*rx_ring, i);
 350                         u1 = (struct my_u1 *)rx_desc_ps;
 351                         staterr =
 352                             le32_to_cpu(rx_desc_ps->wb.middle.status_error);
 353
 354                         if (i == rx_ring->next_to_use)
 355                                 next_desc = " NTU";
 356                         else if (i == rx_ring->next_to_clean)
 357                                 next_desc = " NTC";
 358                         else
 359                                 next_desc = "";
 360
 361                         if (staterr & E1000_RXD_STAT_DD) {
 362                                 /* Descriptor Done */
 363                                 pr_info("%s[0x%03X]     %016llX %016llX %016llX %016llX ---------------- %p%s\n",
 364                                         "RWB", i,
 365                                         (unsigned long long)le64_to_cpu(u1->a),
 366                                         (unsigned long long)le64_to_cpu(u1->b),
 367                                         (unsigned long long)le64_to_cpu(u1->c),
 368                                         (unsigned long long)le64_to_cpu(u1->d),
 369                                         buffer_info->skb, next_desc);
 370                         } else {
 371                                 pr_info("%s[0x%03X]     %016llX %016llX %016llX %016llX %016llX %p%s\n",
 372                                         "R  ", i,
 373                                         (unsigned long long)le64_to_cpu(u1->a),
 374                                         (unsigned long long)le64_to_cpu(u1->b),
 375                                         (unsigned long long)le64_to_cpu(u1->c),
 376                                         (unsigned long long)le64_to_cpu(u1->d),
 377                                         (unsigned long long)buffer_info->dma,
 378                                         buffer_info->skb, next_desc);
 379
 380                                 if (netif_msg_pktdata(adapter))
 381                                         e1000e_dump_ps_pages(adapter,
 382                                                              buffer_info);
 383                         }
 384                 }
 385                 break;
 386         default:
 387         case 0:
 388                 /* Extended Receive Descriptor (Read) Format
 389                  *
 390                  *   +-----------------------------------------------------+
 391                  * 0 |                Buffer Address [63:0]                |
 392                  *   +-----------------------------------------------------+
 393                  * 8 |                      Reserved                       |
 394                  *   +-----------------------------------------------------+
 395                  */
 396                 pr_info("R  [desc]      [buf addr 63:0 ] [reserved 63:0 ] [bi->dma       ] [bi->skb] <-- Ext (Read) format\n");
 397                 /* Extended Receive Descriptor (Write-Back) Format
 398                  *
 399                  *   63       48 47    32 31    24 23            4 3        0
 400                  *   +------------------------------------------------------+
 401                  *   |     RSS Hash      |        |               |         |
 402                  * 0 +-------------------+  Rsvd  |   Reserved    | MRQ RSS |
 403                  *   | Packet   | IP     |        |               |  Type   |
 404                  *   | Checksum | Ident  |        |               |         |
 405                  *   +------------------------------------------------------+
 406                  * 8 | VLAN Tag | Length | Extended Error | Extended Status |
 407                  *   +------------------------------------------------------+
 408                  *   63       48 47    32 31            20 19               0
 409                  */
 410                 pr_info("RWB[desc]      [cs ipid    mrq] [vt   ln xe  xs] [bi->skb] <-- Ext (Write-Back) format\n");
 411
 412                 for (i = 0; i < rx_ring->count; i++) {
 413                         const char *next_desc;
 414
 415                         buffer_info = &rx_ring->buffer_info[i];
 416                         rx_desc = E1000_RX_DESC_EXT(*rx_ring, i);
 417                         u1 = (struct my_u1 *)rx_desc;
 418                         staterr = le32_to_cpu(rx_desc->wb.upper.status_error);
 419
 420                         if (i == rx_ring->next_to_use)
 421                                 next_desc = " NTU";
 422                         else if (i == rx_ring->next_to_clean)
 423                                 next_desc = " NTC";
 424                         else
 425                                 next_desc = "";
 426
 427                         if (staterr & E1000_RXD_STAT_DD) {
 428                                 /* Descriptor Done */
 429                                 pr_info("%s[0x%03X]     %016llX %016llX ---------------- %p%s\n",
 430                                         "RWB", i,
 431                                         (unsigned long long)le64_to_cpu(u1->a),
 432                                         (unsigned long long)le64_to_cpu(u1->b),
 433                                         buffer_info->skb, next_desc);
 434                         } else {
 435                                 pr_info("%s[0x%03X]     %016llX %016llX %016llX %p%s\n",
 436                                         "R  ", i,
 437                                         (unsigned long long)le64_to_cpu(u1->a),
 438                                         (unsigned long long)le64_to_cpu(u1->b),
 439                                         (unsigned long long)buffer_info->dma,
 440                                         buffer_info->skb, next_desc);
 441
 442                                 if (netif_msg_pktdata(adapter) &&
 443                                     buffer_info->skb)
 444                                         print_hex_dump(KERN_INFO, "",
 445                                                        DUMP_PREFIX_ADDRESS, 16,
 446                                                        1,
 447                                                        buffer_info->skb->data,
 448                                                        adapter->rx_buffer_len,
 449                                                        true);
 450                         }
 451                 }
 452         }
 453 }
 454
 455 /**
 456  * e1000_desc_unused - calculate if we have unused descriptors
 457  **/
 458 static int e1000_desc_unused(struct e1000_ring *ring)
 459 {
 460         if (ring->next_to_clean > ring->next_to_use)
 461                 return ring->next_to_clean - ring->next_to_use - 1;
 462
 463         return ring->count + ring->next_to_clean - ring->next_to_use - 1;
 464 }
 465
 466 /**
 467  * e1000e_systim_to_hwtstamp - convert system time value to hw time stamp
 468  * @adapter: board private structure
 469  * @hwtstamps: time stamp structure to update
 470  * @systim: unsigned 64bit system time value.
 471  *
 472  * Convert the system time value stored in the RX/TXSTMP registers into a
 473  * hwtstamp which can be used by the upper level time stamping functions.
 474  *
 475  * The 'systim_lock' spinlock is used to protect the consistency of the
 476  * system time value. This is needed because reading the 64 bit time
 477  * value involves reading two 32 bit registers. The first read latches the
 478  * value.
 479  **/
 480 static void e1000e_systim_to_hwtstamp(struct e1000_adapter *adapter,
 481                                       struct skb_shared_hwtstamps *hwtstamps,
 482                                       u64 systim)
 483 {
 484         u64 ns;
 485         unsigned long flags;
 486
 487         spin_lock_irqsave(&adapter->systim_lock, flags);
 488         ns = timecounter_cyc2time(&adapter->tc, systim);
 489         spin_unlock_irqrestore(&adapter->systim_lock, flags);
 490
 491         memset(hwtstamps, 0, sizeof(*hwtstamps));
 492         hwtstamps->hwtstamp = ns_to_ktime(ns);
 493 }
 494
 495 /**
 496  * e1000e_rx_hwtstamp - utility function which checks for Rx time stamp
 497  * @adapter: board private structure
 498  * @status: descriptor extended error and status field
 499  * @skb: particular skb to include time stamp
 500  *
 501  * If the time stamp is valid, convert it into the timecounter ns value
 502  * and store that result into the shhwtstamps structure which is passed
 503  * up the network stack.
 504  **/
 505 static void e1000e_rx_hwtstamp(struct e1000_adapter *adapter, u32 status,
 506                                struct sk_buff *skb)
 507 {
 508         struct e1000_hw *hw = &adapter->hw;
 509         u64 rxstmp;
 510
 511         if (!(adapter->flags & FLAG_HAS_HW_TIMESTAMP) ||
 512             !(status & E1000_RXDEXT_STATERR_TST) ||
 513             !(er32(TSYNCRXCTL) & E1000_TSYNCRXCTL_VALID))
 514                 return;
 515
 516         /* The Rx time stamp registers contain the time stamp.  No other
 517          * received packet will be time stamped until the Rx time stamp
 518          * registers are read.  Because only one packet can be time stamped
 519          * at a time, the register values must belong to this packet and
 520          * therefore none of the other additional attributes need to be
 521          * compared.
 522          */
 523         rxstmp = (u64)er32(RXSTMPL);
 524         rxstmp |= (u64)er32(RXSTMPH) << 32;
 525         e1000e_systim_to_hwtstamp(adapter, skb_hwtstamps(skb), rxstmp);
 526
 527         adapter->flags2 &= ~FLAG2_CHECK_RX_HWTSTAMP;
 528 }
 529
 530 /**
 531  * e1000_receive_skb - helper function to handle Rx indications
 532  * @adapter: board private structure
 533  * @staterr: descriptor extended error and status field as written by hardware
 534  * @vlan: descriptor vlan field as written by hardware (no le/be conversion)
 535  * @skb: pointer to sk_buff to be indicated to stack
 536  **/
 537 static void e1000_receive_skb(struct e1000_adapter *adapter,
 538                               struct net_device *netdev, struct sk_buff *skb,
 539                               u32 staterr, __le16 vlan)
 540 {
 541         u16 tag = le16_to_cpu(vlan);
 542
 543         e1000e_rx_hwtstamp(adapter, staterr, skb);
 544
 545         skb->protocol = eth_type_trans(skb, netdev);
 546
 547         if (staterr & E1000_RXD_STAT_VP)
 548                 __vlan_hwaccel_put_tag(skb, htons(ETH_P_8021Q), tag);
 549
 550         napi_gro_receive(&adapter->napi, skb);
 551 }
 552
 553 /**
 554  * e1000_rx_checksum - Receive Checksum Offload
 555  * @adapter: board private structure
 556  * @status_err: receive descriptor status and error fields
 557  * @csum: receive descriptor csum field
 558  * @sk_buff: socket buffer with received data
 559  **/
 560 static void e1000_rx_checksum(struct e1000_adapter *adapter, u32 status_err,
 561                               struct sk_buff *skb)
 562 {
 563         u16 status = (u16)status_err;
 564         u8 errors = (u8)(status_err >> 24);
 565
 566         skb_checksum_none_assert(skb);
 567
 568         /* Rx checksum disabled */
 569         if (!(adapter->netdev->features & NETIF_F_RXCSUM))
 570                 return;
 571
 572         /* Ignore Checksum bit is set */
 573         if (status & E1000_RXD_STAT_IXSM)
 574                 return;
 575
 576         /* TCP/UDP checksum error bit or IP checksum error bit is set */
 577         if (errors & (E1000_RXD_ERR_TCPE | E1000_RXD_ERR_IPE)) {
 578                 /* let the stack verify checksum errors */
 579                 adapter->hw_csum_err++;
 580                 return;
 581         }
 582
 583         /* TCP/UDP Checksum has not been calculated */
 584         if (!(status & (E1000_RXD_STAT_TCPCS | E1000_RXD_STAT_UDPCS)))
 585                 return;
 586
 587         /* It must be a TCP or UDP packet with a valid checksum */
 588         skb->ip_summed = CHECKSUM_UNNECESSARY;
 589         adapter->hw_csum_good++;
 590 }
 591
 592 static void e1000e_update_rdt_wa(struct e1000_ring *rx_ring, unsigned int i)
 593 {
 594         struct e1000_adapter *adapter = rx_ring->adapter;
 595         struct e1000_hw *hw = &adapter->hw;
 596         s32 ret_val = __ew32_prepare(hw);
 597
 598         writel(i, rx_ring->tail);
 599
 600         if (unlikely(!ret_val && (i != readl(rx_ring->tail)))) {
 601                 u32 rctl = er32(RCTL);
 602                 ew32(RCTL, rctl & ~E1000_RCTL_EN);
 603                 e_err("ME firmware caused invalid RDT - resetting\n");
 604                 schedule_work(&adapter->reset_task);
 605         }
 606 }
 607
 608 static void e1000e_update_tdt_wa(struct e1000_ring *tx_ring, unsigned int i)
 609 {
 610         struct e1000_adapter *adapter = tx_ring->adapter;
 611         struct e1000_hw *hw = &adapter->hw;
 612         s32 ret_val = __ew32_prepare(hw);
 613
 614         writel(i, tx_ring->tail);
 615
 616         if (unlikely(!ret_val && (i != readl(tx_ring->tail)))) {
 617                 u32 tctl = er32(TCTL);
 618                 ew32(TCTL, tctl & ~E1000_TCTL_EN);
 619                 e_err("ME firmware caused invalid TDT - resetting\n");
 620                 schedule_work(&adapter->reset_task);
 621         }
 622 }
 623
 624 /**
 625  * e1000_alloc_rx_buffers - Replace used receive buffers
 626  * @rx_ring: Rx descriptor ring
 627  **/
 628 static void e1000_alloc_rx_buffers(struct e1000_ring *rx_ring,
 629                                    int cleaned_count, gfp_t gfp)
 630 {
 631         struct e1000_adapter *adapter = rx_ring->adapter;
 632         struct net_device *netdev = adapter->netdev;
 633         struct pci_dev *pdev = adapter->pdev;
 634         union e1000_rx_desc_extended *rx_desc;
 635         struct e1000_buffer *buffer_info;
 636         struct sk_buff *skb;
 637         unsigned int i;
 638         unsigned int bufsz = adapter->rx_buffer_len;
 639
 640         i = rx_ring->next_to_use;
 641         buffer_info = &rx_ring->buffer_info[i];
 642
 643         while (cleaned_count--) {
 644                 skb = buffer_info->skb;
 645                 if (skb) {
 646                         skb_trim(skb, 0);
 647                         goto map_skb;
 648                 }
 649
 650                 skb = __netdev_alloc_skb_ip_align(netdev, bufsz, gfp);
 651                 if (!skb) {
 652                         /* Better luck next round */
 653                         adapter->alloc_rx_buff_failed++;
 654                         break;
 655                 }
 656
 657                 buffer_info->skb = skb;
 658 map_skb:
 659                 buffer_info->dma = dma_map_single(&pdev->dev, skb->data,
 660                                                   adapter->rx_buffer_len,
 661                                                   DMA_FROM_DEVICE);
 662                 if (dma_mapping_error(&pdev->dev, buffer_info->dma)) {
 663                         dev_err(&pdev->dev, "Rx DMA map failed\n");
 664                         adapter->rx_dma_failed++;
 665                         break;
 666                 }
 667
 668                 rx_desc = E1000_RX_DESC_EXT(*rx_ring, i);
 669                 rx_desc->read.buffer_addr = cpu_to_le64(buffer_info->dma);
 670
 671                 if (unlikely(!(i & (E1000_RX_BUFFER_WRITE - 1)))) {
 672                         /* Force memory writes to complete before letting h/w
 673                          * know there are new descriptors to fetch.  (Only
 674                          * applicable for weak-ordered memory model archs,
 675                          * such as IA-64).
 676                          */
 677                         wmb();
 678                         if (adapter->flags2 & FLAG2_PCIM2PCI_ARBITER_WA)
 679                                 e1000e_update_rdt_wa(rx_ring, i);
 680                         else
 681                                 writel(i, rx_ring->tail);
 682                 }
 683                 i++;
 684                 if (i == rx_ring->count)
 685                         i = 0;
 686                 buffer_info = &rx_ring->buffer_info[i];
 687         }
 688
 689         rx_ring->next_to_use = i;
 690 }
 691
 692 /**
 693  * e1000_alloc_rx_buffers_ps - Replace used receive buffers; packet split
 694  * @rx_ring: Rx descriptor ring
 695  **/
 696 static void e1000_alloc_rx_buffers_ps(struct e1000_ring *rx_ring,
 697                                       int cleaned_count, gfp_t gfp)
 698 {
 699         struct e1000_adapter *adapter = rx_ring->adapter;
 700         struct net_device *netdev = adapter->netdev;
 701         struct pci_dev *pdev = adapter->pdev;
 702         union e1000_rx_desc_packet_split *rx_desc;
 703         struct e1000_buffer *buffer_info;
 704         struct e1000_ps_page *ps_page;
 705         struct sk_buff *skb;
 706         unsigned int i, j;
 707
 708         i = rx_ring->next_to_use;
 709         buffer_info = &rx_ring->buffer_info[i];
 710
 711         while (cleaned_count--) {
 712                 rx_desc = E1000_RX_DESC_PS(*rx_ring, i);
 713
 714                 for (j = 0; j < PS_PAGE_BUFFERS; j++) {
 715                         ps_page = &buffer_info->ps_pages[j];
 716                         if (j >= adapter->rx_ps_pages) {
 717                                 /* all unused desc entries get hw null ptr */
 718                                 rx_desc->read.buffer_addr[j + 1] =
 719                                     ~cpu_to_le64(0);
 720                                 continue;
 721                         }
 722                         if (!ps_page->page) {
 723                                 ps_page->page = alloc_page(gfp);
 724                                 if (!ps_page->page) {
 725                                         adapter->alloc_rx_buff_failed++;
 726                                         goto no_buffers;
 727                                 }
 728                                 ps_page->dma = dma_map_page(&pdev->dev,
 729                                                             ps_page->page,
 730                                                             0, PAGE_SIZE,
 731                                                             DMA_FROM_DEVICE);
 732                                 if (dma_mapping_error(&pdev->dev,
 733                                                       ps_page->dma)) {
 734                                         dev_err(&adapter->pdev->dev,
 735                                                 "Rx DMA page map failed\n");
 736                                         adapter->rx_dma_failed++;
 737                                         goto no_buffers;
 738                                 }
 739                         }
 740                         /* Refresh the desc even if buffer_addrs
 741                          * didn't change because each write-back
 742                          * erases this info.
 743                          */
 744                         rx_desc->read.buffer_addr[j + 1] =
 745                             cpu_to_le64(ps_page->dma);
 746                 }
 747
 748                 skb = __netdev_alloc_skb_ip_align(netdev, adapter->rx_ps_bsize0,
 749                                                   gfp);
 750
 751                 if (!skb) {
 752                         adapter->alloc_rx_buff_failed++;
 753                         break;
 754                 }
 755
 756                 buffer_info->skb = skb;
 757                 buffer_info->dma = dma_map_single(&pdev->dev, skb->data,
 758                                                   adapter->rx_ps_bsize0,
 759                                                   DMA_FROM_DEVICE);
 760                 if (dma_mapping_error(&pdev->dev, buffer_info->dma)) {
 761                         dev_err(&pdev->dev, "Rx DMA map failed\n");
 762                         adapter->rx_dma_failed++;
 763                         /* cleanup skb */
 764                         dev_kfree_skb_any(skb);
 765                         buffer_info->skb = NULL;
 766                         break;
 767                 }
 768
 769                 rx_desc->read.buffer_addr[0] = cpu_to_le64(buffer_info->dma);
 770
 771                 if (unlikely(!(i & (E1000_RX_BUFFER_WRITE - 1)))) {
 772                         /* Force memory writes to complete before letting h/w
 773                          * know there are new descriptors to fetch.  (Only
 774                          * applicable for weak-ordered memory model archs,
 775                          * such as IA-64).
 776                          */
 777                         wmb();
 778                         if (adapter->flags2 & FLAG2_PCIM2PCI_ARBITER_WA)
 779                                 e1000e_update_rdt_wa(rx_ring, i << 1);
 780                         else
 781                                 writel(i << 1, rx_ring->tail);
 782                 }
 783
 784                 i++;
 785                 if (i == rx_ring->count)
 786                         i = 0;
 787                 buffer_info = &rx_ring->buffer_info[i];
 788         }
 789
 790 no_buffers:
 791         rx_ring->next_to_use = i;
 792 }
 793
 794 /**
 795  * e1000_alloc_jumbo_rx_buffers - Replace used jumbo receive buffers
 796  * @rx_ring: Rx descriptor ring
 797  * @cleaned_count: number of buffers to allocate this pass
 798  **/
 799
 800 static void e1000_alloc_jumbo_rx_buffers(struct e1000_ring *rx_ring,
 801                                          int cleaned_count, gfp_t gfp)
 802 {
 803         struct e1000_adapter *adapter = rx_ring->adapter;
 804         struct net_device *netdev = adapter->netdev;
 805         struct pci_dev *pdev = adapter->pdev;
 806         union e1000_rx_desc_extended *rx_desc;
 807         struct e1000_buffer *buffer_info;
 808         struct sk_buff *skb;
 809         unsigned int i;
 810         unsigned int bufsz = 256 - 16;  /* for skb_reserve */
 811
 812         i = rx_ring->next_to_use;
 813         buffer_info = &rx_ring->buffer_info[i];
 814
 815         while (cleaned_count--) {
 816                 skb = buffer_info->skb;
 817                 if (skb) {
 818                         skb_trim(skb, 0);
 819                         goto check_page;
 820                 }
 821
 822                 skb = __netdev_alloc_skb_ip_align(netdev, bufsz, gfp);
 823                 if (unlikely(!skb)) {
 824                         /* Better luck next round */
 825                         adapter->alloc_rx_buff_failed++;
 826                         break;
 827                 }
 828
 829                 buffer_info->skb = skb;
 830 check_page:
 831                 /* allocate a new page if necessary */
 832                 if (!buffer_info->page) {
 833                         buffer_info->page = alloc_page(gfp);
 834                         if (unlikely(!buffer_info->page)) {
 835                                 adapter->alloc_rx_buff_failed++;
 836                                 break;
 837                         }
 838                 }
 839
 840                 if (!buffer_info->dma) {
 841                         buffer_info->dma = dma_map_page(&pdev->dev,
 842                                                         buffer_info->page, 0,
 843                                                         PAGE_SIZE,
 844                                                         DMA_FROM_DEVICE);
 845                         if (dma_mapping_error(&pdev->dev, buffer_info->dma)) {
 846                                 adapter->alloc_rx_buff_failed++;
 847                                 break;
 848                         }
 849                 }
 850
 851                 rx_desc = E1000_RX_DESC_EXT(*rx_ring, i);
 852                 rx_desc->read.buffer_addr = cpu_to_le64(buffer_info->dma);
 853
 854                 if (unlikely(++i == rx_ring->count))
 855                         i = 0;
 856                 buffer_info = &rx_ring->buffer_info[i];
 857         }
 858
 859         if (likely(rx_ring->next_to_use != i)) {
 860                 rx_ring->next_to_use = i;
 861                 if (unlikely(i-- == 0))
 862                         i = (rx_ring->count - 1);
 863
 864                 /* Force memory writes to complete before letting h/w
 865                  * know there are new descriptors to fetch.  (Only
 866                  * applicable for weak-ordered memory model archs,
 867                  * such as IA-64).
 868                  */
 869                 wmb();
 870                 if (adapter->flags2 & FLAG2_PCIM2PCI_ARBITER_WA)
 871                         e1000e_update_rdt_wa(rx_ring, i);
 872                 else
 873                         writel(i, rx_ring->tail);
 874         }
 875 }
 876
 877 static inline void e1000_rx_hash(struct net_device *netdev, __le32 rss,
 878                                  struct sk_buff *skb)
 879 {
 880         if (netdev->features & NETIF_F_RXHASH)
 881                 skb->rxhash = le32_to_cpu(rss);
 882 }
 883
 884 /**
 885  * e1000_clean_rx_irq - Send received data up the network stack
 886  * @rx_ring: Rx descriptor ring
 887  *
 888  * the return value indicates whether actual cleaning was done, there
 889  * is no guarantee that everything was cleaned
 890  **/
 891 static bool e1000_clean_rx_irq(struct e1000_ring *rx_ring, int *work_done,
 892                                int work_to_do)
 893 {
 894         struct e1000_adapter *adapter = rx_ring->adapter;
 895         struct net_device *netdev = adapter->netdev;
 896         struct pci_dev *pdev = adapter->pdev;
 897         struct e1000_hw *hw = &adapter->hw;
 898         union e1000_rx_desc_extended *rx_desc, *next_rxd;
 899         struct e1000_buffer *buffer_info, *next_buffer;
 900         u32 length, staterr;
 901         unsigned int i;
 902         int cleaned_count = 0;
 903         bool cleaned = false;
 904         unsigned int total_rx_bytes = 0, total_rx_packets = 0;
 905
 906         i = rx_ring->next_to_clean;
 907         rx_desc = E1000_RX_DESC_EXT(*rx_ring, i);
 908         staterr = le32_to_cpu(rx_desc->wb.upper.status_error);
 909         buffer_info = &rx_ring->buffer_info[i];
 910
 911         while (staterr & E1000_RXD_STAT_DD) {
 912                 struct sk_buff *skb;
 913
 914                 if (*work_done >= work_to_do)
 915                         break;
 916                 (*work_done)++;
 917                 rmb();  /* read descriptor and rx_buffer_info after status DD */
 918
 919                 skb = buffer_info->skb;
 920                 buffer_info->skb = NULL;
 921
 922                 prefetch(skb->data - NET_IP_ALIGN);
 923
 924                 i++;
 925                 if (i == rx_ring->count)
 926                         i = 0;
 927                 next_rxd = E1000_RX_DESC_EXT(*rx_ring, i);
 928                 prefetch(next_rxd);
 929
 930                 next_buffer = &rx_ring->buffer_info[i];
 931
 932                 cleaned = true;
 933                 cleaned_count++;
 934                 dma_unmap_single(&pdev->dev, buffer_info->dma,
 935                                  adapter->rx_buffer_len, DMA_FROM_DEVICE);
 936                 buffer_info->dma = 0;
 937
 938                 length = le16_to_cpu(rx_desc->wb.upper.length);
 939
 940                 /* !EOP means multiple descriptors were used to store a single
 941                  * packet, if that's the case we need to toss it.  In fact, we
 942                  * need to toss every packet with the EOP bit clear and the
 943                  * next frame that _does_ have the EOP bit set, as it is by
 944                  * definition only a frame fragment
 945                  */
 946                 if (unlikely(!(staterr & E1000_RXD_STAT_EOP)))
 947                         adapter->flags2 |= FLAG2_IS_DISCARDING;
 948
 949                 if (adapter->flags2 & FLAG2_IS_DISCARDING) {
 950                         /* All receives must fit into a single buffer */
 951                         e_dbg("Receive packet consumed multiple buffers\n");
 952                         /* recycle */
 953                         buffer_info->skb = skb;
 954                         if (staterr & E1000_RXD_STAT_EOP)
 955                                 adapter->flags2 &= ~FLAG2_IS_DISCARDING;
 956                         goto next_desc;
 957                 }
 958
 959                 if (unlikely((staterr & E1000_RXDEXT_ERR_FRAME_ERR_MASK) &&
 960                              !(netdev->features & NETIF_F_RXALL))) {
 961                         /* recycle */
 962                         buffer_info->skb = skb;
 963                         goto next_desc;
 964                 }
 965
 966                 /* adjust length to remove Ethernet CRC */
 967                 if (!(adapter->flags2 & FLAG2_CRC_STRIPPING)) {
 968                         /* If configured to store CRC, don't subtract FCS,
 969                          * but keep the FCS bytes out of the total_rx_bytes
 970                          * counter
 971                          */
 972                         if (netdev->features & NETIF_F_RXFCS)
 973                                 total_rx_bytes -= 4;
 974                         else
 975                                 length -= 4;
 976                 }
 977
 978                 total_rx_bytes += length;
 979                 total_rx_packets++;
 980
 981                 /* code added for copybreak, this should improve
 982                  * performance for small packets with large amounts
 983                  * of reassembly being done in the stack
 984                  */
 985                 if (length < copybreak) {
 986                         struct sk_buff *new_skb =
 987                             netdev_alloc_skb_ip_align(netdev, length);
 988                         if (new_skb) {
 989                                 skb_copy_to_linear_data_offset(new_skb,
 990                                                                -NET_IP_ALIGN,
 991                                                                (skb->data -
 992                                                                 NET_IP_ALIGN),
 993                                                                (length +
 994                                                                 NET_IP_ALIGN));
 995                                 /* save the skb in buffer_info as good */
 996                                 buffer_info->skb = skb;
 997                                 skb = new_skb;
 998                         }
 999                         /* else just continue with the old one */
1000                 }
1001                 /* end copybreak code */
1002                 skb_put(skb, length);
1003
1004                 /* Receive Checksum Offload */
1005                 e1000_rx_checksum(adapter, staterr, skb);
1006
1007                 e1000_rx_hash(netdev, rx_desc->wb.lower.hi_dword.rss, skb);
1008
1009                 e1000_receive_skb(adapter, netdev, skb, staterr,
1010                                   rx_desc->wb.upper.vlan);
1011
1012 next_desc:
1013                 rx_desc->wb.upper.status_error &= cpu_to_le32(~0xFF);
1014
1015                 /* return some buffers to hardware, one at a time is too slow */
1016                 if (cleaned_count >= E1000_RX_BUFFER_WRITE) {
1017                         adapter->alloc_rx_buf(rx_ring, cleaned_count,
1018                                               GFP_ATOMIC);
1019                         cleaned_count = 0;
1020                 }
1021
1022                 /* use prefetched values */
1023                 rx_desc = next_rxd;
1024                 buffer_info = next_buffer;
1025
1026                 staterr = le32_to_cpu(rx_desc->wb.upper.status_error);
1027         }
1028         rx_ring->next_to_clean = i;
1029
1030         cleaned_count = e1000_desc_unused(rx_ring);
1031         if (cleaned_count)
1032                 adapter->alloc_rx_buf(rx_ring, cleaned_count, GFP_ATOMIC);
1033
1034         adapter->total_rx_bytes += total_rx_bytes;
1035         adapter->total_rx_packets += total_rx_packets;
1036         return cleaned;
1037 }
1038
1039 static void e1000_put_txbuf(struct e1000_ring *tx_ring,
1040                             struct e1000_buffer *buffer_info)
1041 {
1042         struct e1000_adapter *adapter = tx_ring->adapter;
1043
1044         if (buffer_info->dma) {
1045                 if (buffer_info->mapped_as_page)
1046                         dma_unmap_page(&adapter->pdev->dev, buffer_info->dma,
1047                                        buffer_info->length, DMA_TO_DEVICE);
1048                 else
1049                         dma_unmap_single(&adapter->pdev->dev, buffer_info->dma,
1050                                          buffer_info->length, DMA_TO_DEVICE);
1051                 buffer_info->dma = 0;
1052         }
1053         if (buffer_info->skb) {
1054                 dev_kfree_skb_any(buffer_info->skb);
1055                 buffer_info->skb = NULL;
1056         }
1057         buffer_info->time_stamp = 0;
1058 }
1059
1060 static void e1000_print_hw_hang(struct work_struct *work)
1061 {
1062         struct e1000_adapter *adapter = container_of(work,
1063                                                      struct e1000_adapter,
1064                                                      print_hang_task);
1065         struct net_device *netdev = adapter->netdev;
1066         struct e1000_ring *tx_ring = adapter->tx_ring;
1067         unsigned int i = tx_ring->next_to_clean;
1068         unsigned int eop = tx_ring->buffer_info[i].next_to_watch;
1069         struct e1000_tx_desc *eop_desc = E1000_TX_DESC(*tx_ring, eop);
1070         struct e1000_hw *hw = &adapter->hw;
1071         u16 phy_status, phy_1000t_status, phy_ext_status;
1072         u16 pci_status;
1073
1074         if (test_bit(__E1000_DOWN, &adapter->state))
1075                 return;
1076
1077         if (!adapter->tx_hang_recheck && (adapter->flags2 & FLAG2_DMA_BURST)) {
1078                 /* May be block on write-back, flush and detect again
1079                  * flush pending descriptor writebacks to memory
1080                  */
1081                 ew32(TIDV, adapter->tx_int_delay | E1000_TIDV_FPD);
1082                 /* execute the writes immediately */
1083                 e1e_flush();
1084                 /* Due to rare timing issues, write to TIDV again to ensure
1085                  * the write is successful
1086                  */
1087                 ew32(TIDV, adapter->tx_int_delay | E1000_TIDV_FPD);
1088                 /* execute the writes immediately */
1089                 e1e_flush();
1090                 adapter->tx_hang_recheck = true;
1091                 return;
1092         }
1093         /* Real hang detected */
1094         adapter->tx_hang_recheck = false;
1095         netif_stop_queue(netdev);
1096
1097         e1e_rphy(hw, MII_BMSR, &phy_status);
1098         e1e_rphy(hw, MII_STAT1000, &phy_1000t_status);
1099         e1e_rphy(hw, MII_ESTATUS, &phy_ext_status);
1100
1101         pci_read_config_word(adapter->pdev, PCI_STATUS, &pci_status);
1102
1103         /* detected Hardware unit hang */
1104         e_err("Detected Hardware Unit Hang:\n"
1105               "  TDH                  <%x>\n"
1106               "  TDT                  <%x>\n"
1107               "  next_to_use          <%x>\n"
1108               "  next_to_clean        <%x>\n"
1109               "buffer_info[next_to_clean]:\n"
1110               "  time_stamp           <%lx>\n"
1111               "  next_to_watch        <%x>\n"
1112               "  jiffies              <%lx>\n"
1113               "  next_to_watch.status <%x>\n"
1114               "MAC Status             <%x>\n"
1115               "PHY Status             <%x>\n"
1116               "PHY 1000BASE-T Status  <%x>\n"
1117               "PHY Extended Status    <%x>\n"
1118               "PCI Status             <%x>\n",
1119               readl(tx_ring->head), readl(tx_ring->tail), tx_ring->next_to_use,
1120               tx_ring->next_to_clean, tx_ring->buffer_info[eop].time_stamp,
1121               eop, jiffies, eop_desc->upper.fields.status, er32(STATUS),
1122               phy_status, phy_1000t_status, phy_ext_status, pci_status);
1123
1124         /* Suggest workaround for known h/w issue */
1125         if ((hw->mac.type == e1000_pchlan) && (er32(CTRL) & E1000_CTRL_TFCE))
1126                 e_err("Try turning off Tx pause (flow control) via ethtool\n");
1127 }
1128
1129 /**
1130  * e1000e_tx_hwtstamp_work - check for Tx time stamp
1131  * @work: pointer to work struct
1132  *
1133  * This work function polls the TSYNCTXCTL valid bit to determine when a
1134  * timestamp has been taken for the current stored skb.  The timestamp must
1135  * be for this skb because only one such packet is allowed in the queue.
1136  */
1137 static void e1000e_tx_hwtstamp_work(struct work_struct *work)
1138 {
1139         struct e1000_adapter *adapter = container_of(work, struct e1000_adapter,
1140                                                      tx_hwtstamp_work);
1141         struct e1000_hw *hw = &adapter->hw;
1142
1143         if (!adapter->tx_hwtstamp_skb)
1144                 return;
1145
1146         if (er32(TSYNCTXCTL) & E1000_TSYNCTXCTL_VALID) {
1147                 struct skb_shared_hwtstamps shhwtstamps;
1148                 u64 txstmp;
1149
1150                 txstmp = er32(TXSTMPL);
1151                 txstmp |= (u64)er32(TXSTMPH) << 32;
1152
1153                 e1000e_systim_to_hwtstamp(adapter, &shhwtstamps, txstmp);
1154
1155                 skb_tstamp_tx(adapter->tx_hwtstamp_skb, &shhwtstamps);
1156                 dev_kfree_skb_any(adapter->tx_hwtstamp_skb);
1157                 adapter->tx_hwtstamp_skb = NULL;
1158         } else {
1159                 /* reschedule to check later */
1160                 schedule_work(&adapter->tx_hwtstamp_work);
1161         }
1162 }
1163
1164 /**
1165  * e1000_clean_tx_irq - Reclaim resources after transmit completes
1166  * @tx_ring: Tx descriptor ring
1167  *
1168  * the return value indicates whether actual cleaning was done, there
1169  * is no guarantee that everything was cleaned
1170  **/
1171 static bool e1000_clean_tx_irq(struct e1000_ring *tx_ring)
1172 {
1173         struct e1000_adapter *adapter = tx_ring->adapter;
1174         struct net_device *netdev = adapter->netdev;
1175         struct e1000_hw *hw = &adapter->hw;
1176         struct e1000_tx_desc *tx_desc, *eop_desc;
1177         struct e1000_buffer *buffer_info;
1178         unsigned int i, eop;
1179         unsigned int count = 0;
1180         unsigned int total_tx_bytes = 0, total_tx_packets = 0;
1181         unsigned int bytes_compl = 0, pkts_compl = 0;
1182
1183         i = tx_ring->next_to_clean;
1184         eop = tx_ring->buffer_info[i].next_to_watch;
1185         eop_desc = E1000_TX_DESC(*tx_ring, eop);
1186
1187         while ((eop_desc->upper.data & cpu_to_le32(E1000_TXD_STAT_DD)) &&
1188                (count < tx_ring->count)) {
1189                 bool cleaned = false;
1190                 rmb();          /* read buffer_info after eop_desc */
1191                 for (; !cleaned; count++) {
1192                         tx_desc = E1000_TX_DESC(*tx_ring, i);
1193                         buffer_info = &tx_ring->buffer_info[i];
1194                         cleaned = (i == eop);
1195
1196                         if (cleaned) {
1197                                 total_tx_packets += buffer_info->segs;
1198                                 total_tx_bytes += buffer_info->bytecount;
1199                                 if (buffer_info->skb) {
1200                                         bytes_compl += buffer_info->skb->len;
1201                                         pkts_compl++;
1202                                 }
1203                         }
1204
1205                         e1000_put_txbuf(tx_ring, buffer_info);
1206                         tx_desc->upper.data = 0;
1207
1208                         i++;
1209                         if (i == tx_ring->count)
1210                                 i = 0;
1211                 }
1212
1213                 if (i == tx_ring->next_to_use)
1214                         break;
1215                 eop = tx_ring->buffer_info[i].next_to_watch;
1216                 eop_desc = E1000_TX_DESC(*tx_ring, eop);
1217         }
1218
1219         tx_ring->next_to_clean = i;
1220
1221         netdev_completed_queue(netdev, pkts_compl, bytes_compl);
1222
1223 #define TX_WAKE_THRESHOLD 32
1224         if (count && netif_carrier_ok(netdev) &&
1225             e1000_desc_unused(tx_ring) >= TX_WAKE_THRESHOLD) {
1226                 /* Make sure that anybody stopping the queue after this
1227                  * sees the new next_to_clean.
1228                  */
1229                 smp_mb();
1230
1231                 if (netif_queue_stopped(netdev) &&
1232                     !(test_bit(__E1000_DOWN, &adapter->state))) {
1233                         netif_wake_queue(netdev);
1234                         ++adapter->restart_queue;
1235                 }
1236         }
1237
1238         if (adapter->detect_tx_hung) {
1239                 /* Detect a transmit hang in hardware, this serializes the
1240                  * check with the clearing of time_stamp and movement of i
1241                  */
1242                 adapter->detect_tx_hung = false;
1243                 if (tx_ring->buffer_info[i].time_stamp &&
1244                     time_after(jiffies, tx_ring->buffer_info[i].time_stamp
1245                                + (adapter->tx_timeout_factor * HZ)) &&
1246                     !(er32(STATUS) & E1000_STATUS_TXOFF))
1247                         schedule_work(&adapter->print_hang_task);
1248                 else
1249                         adapter->tx_hang_recheck = false;
1250         }
1251         adapter->total_tx_bytes += total_tx_bytes;
1252         adapter->total_tx_packets += total_tx_packets;
1253         return count < tx_ring->count;
1254 }
1255
1256 /**
1257  * e1000_clean_rx_irq_ps - Send received data up the network stack; packet split
1258  * @rx_ring: Rx descriptor ring
1259  *
1260  * the return value indicates whether actual cleaning was done, there
1261  * is no guarantee that everything was cleaned
1262  **/
1263 static bool e1000_clean_rx_irq_ps(struct e1000_ring *rx_ring, int *work_done,
1264                                   int work_to_do)
1265 {
1266         struct e1000_adapter *adapter = rx_ring->adapter;
1267         struct e1000_hw *hw = &adapter->hw;
1268         union e1000_rx_desc_packet_split *rx_desc, *next_rxd;
1269         struct net_device *netdev = adapter->netdev;
1270         struct pci_dev *pdev = adapter->pdev;
1271         struct e1000_buffer *buffer_info, *next_buffer;
1272         struct e1000_ps_page *ps_page;
1273         struct sk_buff *skb;
1274         unsigned int i, j;
1275         u32 length, staterr;
1276         int cleaned_count = 0;
1277         bool cleaned = false;
1278         unsigned int total_rx_bytes = 0, total_rx_packets = 0;
1279
1280         i = rx_ring->next_to_clean;
1281         rx_desc = E1000_RX_DESC_PS(*rx_ring, i);
1282         staterr = le32_to_cpu(rx_desc->wb.middle.status_error);
1283         buffer_info = &rx_ring->buffer_info[i];
1284
1285         while (staterr & E1000_RXD_STAT_DD) {
1286                 if (*work_done >= work_to_do)
1287                         break;
1288                 (*work_done)++;
1289                 skb = buffer_info->skb;
1290                 rmb();  /* read descriptor and rx_buffer_info after status DD */
1291
1292                 /* in the packet split case this is header only */
1293                 prefetch(skb->data - NET_IP_ALIGN);
1294
1295                 i++;
1296                 if (i == rx_ring->count)
1297                         i = 0;
1298                 next_rxd = E1000_RX_DESC_PS(*rx_ring, i);
1299                 prefetch(next_rxd);
1300
1301                 next_buffer = &rx_ring->buffer_info[i];
1302
1303                 cleaned = true;
1304                 cleaned_count++;
1305                 dma_unmap_single(&pdev->dev, buffer_info->dma,
1306                                  adapter->rx_ps_bsize0, DMA_FROM_DEVICE);
1307                 buffer_info->dma = 0;
1308
1309                 /* see !EOP comment in other Rx routine */
1310                 if (!(staterr & E1000_RXD_STAT_EOP))
1311                         adapter->flags2 |= FLAG2_IS_DISCARDING;
1312
1313                 if (adapter->flags2 & FLAG2_IS_DISCARDING) {
1314                         e_dbg("Packet Split buffers didn't pick up the full packet\n");
1315                         dev_kfree_skb_irq(skb);
1316                         if (staterr & E1000_RXD_STAT_EOP)
1317                                 adapter->flags2 &= ~FLAG2_IS_DISCARDING;
1318                         goto next_desc;
1319                 }
1320
1321                 if (unlikely((staterr & E1000_RXDEXT_ERR_FRAME_ERR_MASK) &&
1322                              !(netdev->features & NETIF_F_RXALL))) {
1323                         dev_kfree_skb_irq(skb);
1324                         goto next_desc;
1325                 }
1326
1327                 length = le16_to_cpu(rx_desc->wb.middle.length0);
1328
1329                 if (!length) {
1330                         e_dbg("Last part of the packet spanning multiple descriptors\n");
1331                         dev_kfree_skb_irq(skb);
1332                         goto next_desc;
1333                 }
1334
1335                 /* Good Receive */
1336                 skb_put(skb, length);
1337
1338                 {
1339                         /* this looks ugly, but it seems compiler issues make
1340                          * it more efficient than reusing j
1341                          */
1342                         int l1 = le16_to_cpu(rx_desc->wb.upper.length[0]);
1343
1344                         /* page alloc/put takes too long and effects small
1345                          * packet throughput, so unsplit small packets and
1346                          * save the alloc/put only valid in softirq (napi)
1347                          * context to call kmap_*
1348                          */
1349                         if (l1 && (l1 <= copybreak) &&
1350                             ((length + l1) <= adapter->rx_ps_bsize0)) {
1351                                 u8 *vaddr;
1352
1353                                 ps_page = &buffer_info->ps_pages[0];
1354
1355                                 /* there is no documentation about how to call
1356                                  * kmap_atomic, so we can't hold the mapping
1357                                  * very long
1358                                  */
1359                                 dma_sync_single_for_cpu(&pdev->dev,
1360                                                         ps_page->dma,
1361                                                         PAGE_SIZE,
1362                                                         DMA_FROM_DEVICE);
1363                                 vaddr = kmap_atomic(ps_page->page);
1364                                 memcpy(skb_tail_pointer(skb), vaddr, l1);
1365                                 kunmap_atomic(vaddr);
1366                                 dma_sync_single_for_device(&pdev->dev,
1367                                                            ps_page->dma,
1368                                                            PAGE_SIZE,
1369                                                            DMA_FROM_DEVICE);
1370
1371                                 /* remove the CRC */
1372                                 if (!(adapter->flags2 & FLAG2_CRC_STRIPPING)) {
1373                                         if (!(netdev->features & NETIF_F_RXFCS))
1374                                                 l1 -= 4;
1375                                 }
1376
1377                                 skb_put(skb, l1);
1378                                 goto copydone;
1379                         }       /* if */
1380                 }
1381
1382                 for (j = 0; j < PS_PAGE_BUFFERS; j++) {
1383                         length = le16_to_cpu(rx_desc->wb.upper.length[j]);
1384                         if (!length)
1385                                 break;
1386
1387                         ps_page = &buffer_info->ps_pages[j];
1388                         dma_unmap_page(&pdev->dev, ps_page->dma, PAGE_SIZE,
1389                                        DMA_FROM_DEVICE);
1390                         ps_page->dma = 0;
1391                         skb_fill_page_desc(skb, j, ps_page->page, 0, length);
1392                         ps_page->page = NULL;
1393                         skb->len += length;
1394                         skb->data_len += length;
1395                         skb->truesize += PAGE_SIZE;
1396                 }
1397
1398                 /* strip the ethernet crc, problem is we're using pages now so
1399                  * this whole operation can get a little cpu intensive
1400                  */
1401                 if (!(adapter->flags2 & FLAG2_CRC_STRIPPING)) {
1402                         if (!(netdev->features & NETIF_F_RXFCS))
1403                                 pskb_trim(skb, skb->len - 4);
1404                 }
1405
1406 copydone:
1407                 total_rx_bytes += skb->len;
1408                 total_rx_packets++;
1409
1410                 e1000_rx_checksum(adapter, staterr, skb);
1411
1412                 e1000_rx_hash(netdev, rx_desc->wb.lower.hi_dword.rss, skb);
1413
1414                 if (rx_desc->wb.upper.header_status &
1415                     cpu_to_le16(E1000_RXDPS_HDRSTAT_HDRSP))
1416                         adapter->rx_hdr_split++;
1417
1418                 e1000_receive_skb(adapter, netdev, skb, staterr,
1419                                   rx_desc->wb.middle.vlan);
1420
1421 next_desc:
1422                 rx_desc->wb.middle.status_error &= cpu_to_le32(~0xFF);
1423                 buffer_info->skb = NULL;
1424
1425                 /* return some buffers to hardware, one at a time is too slow */
1426                 if (cleaned_count >= E1000_RX_BUFFER_WRITE) {
1427                         adapter->alloc_rx_buf(rx_ring, cleaned_count,
1428                                               GFP_ATOMIC);
1429                         cleaned_count = 0;
1430                 }
1431
1432                 /* use prefetched values */
1433                 rx_desc = next_rxd;
1434                 buffer_info = next_buffer;
1435
1436                 staterr = le32_to_cpu(rx_desc->wb.middle.status_error);
1437         }
1438         rx_ring->next_to_clean = i;
1439
1440         cleaned_count = e1000_desc_unused(rx_ring);
1441         if (cleaned_count)
1442                 adapter->alloc_rx_buf(rx_ring, cleaned_count, GFP_ATOMIC);
1443
1444         adapter->total_rx_bytes += total_rx_bytes;
1445         adapter->total_rx_packets += total_rx_packets;
1446         return cleaned;
1447 }
1448
1449 /**
1450  * e1000_consume_page - helper function
1451  **/
1452 static void e1000_consume_page(struct e1000_buffer *bi, struct sk_buff *skb,
1453                                u16 length)
1454 {
1455         bi->page = NULL;
1456         skb->len += length;
1457         skb->data_len += length;
1458         skb->truesize += PAGE_SIZE;
1459 }
1460
1461 /**
1462  * e1000_clean_jumbo_rx_irq - Send received data up the network stack; legacy
1463  * @adapter: board private structure
1464  *
1465  * the return value indicates whether actual cleaning was done, there
1466  * is no guarantee that everything was cleaned
1467  **/
1468 static bool e1000_clean_jumbo_rx_irq(struct e1000_ring *rx_ring, int *work_done,
1469                                      int work_to_do)
1470 {
1471         struct e1000_adapter *adapter = rx_ring->adapter;
1472         struct net_device *netdev = adapter->netdev;
1473         struct pci_dev *pdev = adapter->pdev;
1474         union e1000_rx_desc_extended *rx_desc, *next_rxd;
1475         struct e1000_buffer *buffer_info, *next_buffer;
1476         u32 length, staterr;
1477         unsigned int i;
1478         int cleaned_count = 0;
1479         bool cleaned = false;
1480         unsigned int total_rx_bytes = 0, total_rx_packets = 0;
1481         struct skb_shared_info *shinfo;
1482
1483         i = rx_ring->next_to_clean;
1484         rx_desc = E1000_RX_DESC_EXT(*rx_ring, i);
1485         staterr = le32_to_cpu(rx_desc->wb.upper.status_error);
1486         buffer_info = &rx_ring->buffer_info[i];
1487
1488         while (staterr & E1000_RXD_STAT_DD) {
1489                 struct sk_buff *skb;
1490
1491                 if (*work_done >= work_to_do)
1492                         break;
1493                 (*work_done)++;
1494                 rmb();  /* read descriptor and rx_buffer_info after status DD */
1495
1496                 skb = buffer_info->skb;
1497                 buffer_info->skb = NULL;
1498
1499                 ++i;
1500                 if (i == rx_ring->count)
1501                         i = 0;
1502                 next_rxd = E1000_RX_DESC_EXT(*rx_ring, i);
1503                 prefetch(next_rxd);
1504
1505                 next_buffer = &rx_ring->buffer_info[i];
1506
1507                 cleaned = true;
1508                 cleaned_count++;
1509                 dma_unmap_page(&pdev->dev, buffer_info->dma, PAGE_SIZE,
1510                                DMA_FROM_DEVICE);
1511                 buffer_info->dma = 0;
1512
1513                 length = le16_to_cpu(rx_desc->wb.upper.length);
1514
1515                 /* errors is only valid for DD + EOP descriptors */
1516                 if (unlikely((staterr & E1000_RXD_STAT_EOP) &&
1517                              ((staterr & E1000_RXDEXT_ERR_FRAME_ERR_MASK) &&
1518                               !(netdev->features & NETIF_F_RXALL)))) {
1519                         /* recycle both page and skb */
1520                         buffer_info->skb = skb;
1521                         /* an error means any chain goes out the window too */
1522                         if (rx_ring->rx_skb_top)
1523                                 dev_kfree_skb_irq(rx_ring->rx_skb_top);
1524                         rx_ring->rx_skb_top = NULL;
1525                         goto next_desc;
1526                 }
1527 #define rxtop (rx_ring->rx_skb_top)
1528                 if (!(staterr & E1000_RXD_STAT_EOP)) {
1529                         /* this descriptor is only the beginning (or middle) */
1530                         if (!rxtop) {
1531                                 /* this is the beginning of a chain */
1532                                 rxtop = skb;
1533                                 skb_fill_page_desc(rxtop, 0, buffer_info->page,
1534                                                    0, length);
1535                         } else {
1536                                 /* this is the middle of a chain */
1537                                 shinfo = skb_shinfo(rxtop);
1538                                 skb_fill_page_desc(rxtop, shinfo->nr_frags,
1539                                                    buffer_info->page, 0,
1540                                                    length);
1541                                 /* re-use the skb, only consumed the page */
1542                                 buffer_info->skb = skb;
1543                         }
1544                         e1000_consume_page(buffer_info, rxtop, length);
1545                         goto next_desc;
1546                 } else {
1547                         if (rxtop) {
1548                                 /* end of the chain */
1549                                 shinfo = skb_shinfo(rxtop);
1550                                 skb_fill_page_desc(rxtop, shinfo->nr_frags,
1551                                                    buffer_info->page, 0,
1552                                                    length);
1553                                 /* re-use the current skb, we only consumed the
1554                                  * page
1555                                  */
1556                                 buffer_info->skb = skb;
1557                                 skb = rxtop;
1558                                 rxtop = NULL;
1559                                 e1000_consume_page(buffer_info, skb, length);
1560                         } else {
1561                                 /* no chain, got EOP, this buf is the packet
1562                                  * copybreak to save the put_page/alloc_page
1563                                  */
1564                                 if (length <= copybreak &&
1565                                     skb_tailroom(skb) >= length) {
1566                                         u8 *vaddr;
1567                                         vaddr = kmap_atomic(buffer_info->page);
1568                                         memcpy(skb_tail_pointer(skb), vaddr,
1569                                                length);
1570                                         kunmap_atomic(vaddr);
1571                                         /* re-use the page, so don't erase
1572                                          * buffer_info->page
1573                                          */
1574                                         skb_put(skb, length);
1575                                 } else {
1576                                         skb_fill_page_desc(skb, 0,
1577                                                            buffer_info->page, 0,
1578                                                            length);
1579                                         e1000_consume_page(buffer_info, skb,
1580                                                            length);
1581                                 }
1582                         }
1583                 }
1584
1585                 /* Receive Checksum Offload */
1586                 e1000_rx_checksum(adapter, staterr, skb);
1587
1588                 e1000_rx_hash(netdev, rx_desc->wb.lower.hi_dword.rss, skb);
1589
1590                 /* probably a little skewed due to removing CRC */
1591                 total_rx_bytes += skb->len;
1592                 total_rx_packets++;
1593
1594                 /* eth type trans needs skb->data to point to something */
1595                 if (!pskb_may_pull(skb, ETH_HLEN)) {
1596                         e_err("pskb_may_pull failed.\n");
1597                         dev_kfree_skb_irq(skb);
1598                         goto next_desc;
1599                 }
1600
1601                 e1000_receive_skb(adapter, netdev, skb, staterr,
1602                                   rx_desc->wb.upper.vlan);
1603
1604 next_desc:
1605                 rx_desc->wb.upper.status_error &= cpu_to_le32(~0xFF);
1606
1607                 /* return some buffers to hardware, one at a time is too slow */
1608                 if (unlikely(cleaned_count >= E1000_RX_BUFFER_WRITE)) {
1609                         adapter->alloc_rx_buf(rx_ring, cleaned_count,
1610                                               GFP_ATOMIC);
1611                         cleaned_count = 0;
1612                 }
1613
1614                 /* use prefetched values */
1615                 rx_desc = next_rxd;
1616                 buffer_info = next_buffer;
1617
1618                 staterr = le32_to_cpu(rx_desc->wb.upper.status_error);
1619         }
1620         rx_ring->next_to_clean = i;
1621
1622         cleaned_count = e1000_desc_unused(rx_ring);
1623         if (cleaned_count)
1624                 adapter->alloc_rx_buf(rx_ring, cleaned_count, GFP_ATOMIC);
1625
1626         adapter->total_rx_bytes += total_rx_bytes;
1627         adapter->total_rx_packets += total_rx_packets;
1628         return cleaned;
1629 }
1630
1631 /**
1632  * e1000_clean_rx_ring - Free Rx Buffers per Queue
1633  * @rx_ring: Rx descriptor ring
1634  **/
1635 static void e1000_clean_rx_ring(struct e1000_ring *rx_ring)
1636 {
1637         struct e1000_adapter *adapter = rx_ring->adapter;
1638         struct e1000_buffer *buffer_info;
1639         struct e1000_ps_page *ps_page;
1640         struct pci_dev *pdev = adapter->pdev;
1641         unsigned int i, j;
1642
1643         /* Free all the Rx ring sk_buffs */
1644         for (i = 0; i < rx_ring->count; i++) {
1645                 buffer_info = &rx_ring->buffer_info[i];
1646                 if (buffer_info->dma) {
1647                         if (adapter->clean_rx == e1000_clean_rx_irq)
1648                                 dma_unmap_single(&pdev->dev, buffer_info->dma,
1649                                                  adapter->rx_buffer_len,
1650                                                  DMA_FROM_DEVICE);
1651                         else if (adapter->clean_rx == e1000_clean_jumbo_rx_irq)
1652                                 dma_unmap_page(&pdev->dev, buffer_info->dma,
1653                                                PAGE_SIZE, DMA_FROM_DEVICE);
1654                         else if (adapter->clean_rx == e1000_clean_rx_irq_ps)
1655                                 dma_unmap_single(&pdev->dev, buffer_info->dma,
1656                                                  adapter->rx_ps_bsize0,
1657                                                  DMA_FROM_DEVICE);
1658                         buffer_info->dma = 0;
1659                 }
1660
1661                 if (buffer_info->page) {
1662                         put_page(buffer_info->page);
1663                         buffer_info->page = NULL;
1664                 }
1665
1666                 if (buffer_info->skb) {
1667                         dev_kfree_skb(buffer_info->skb);
1668                         buffer_info->skb = NULL;
1669                 }
1670
1671                 for (j = 0; j < PS_PAGE_BUFFERS; j++) {
1672                         ps_page = &buffer_info->ps_pages[j];
1673                         if (!ps_page->page)
1674                                 break;
1675                         dma_unmap_page(&pdev->dev, ps_page->dma, PAGE_SIZE,
1676                                        DMA_FROM_DEVICE);
1677                         ps_page->dma = 0;
1678                         put_page(ps_page->page);
1679                         ps_page->page = NULL;
1680                 }
1681         }
1682
1683         /* there also may be some cached data from a chained receive */
1684         if (rx_ring->rx_skb_top) {
1685                 dev_kfree_skb(rx_ring->rx_skb_top);
1686                 rx_ring->rx_skb_top = NULL;
1687         }
1688
1689         /* Zero out the descriptor ring */
1690         memset(rx_ring->desc, 0, rx_ring->size);
1691
1692         rx_ring->next_to_clean = 0;
1693         rx_ring->next_to_use = 0;
1694         adapter->flags2 &= ~FLAG2_IS_DISCARDING;
1695
1696         writel(0, rx_ring->head);
1697         if (adapter->flags2 & FLAG2_PCIM2PCI_ARBITER_WA)
1698                 e1000e_update_rdt_wa(rx_ring, 0);
1699         else
1700                 writel(0, rx_ring->tail);
1701 }
1702
1703 static void e1000e_downshift_workaround(struct work_struct *work)
1704 {
1705         struct e1000_adapter *adapter = container_of(work,
1706                                                      struct e1000_adapter,
1707                                                      downshift_task);
1708
1709         if (test_bit(__E1000_DOWN, &adapter->state))
1710                 return;
1711
1712         e1000e_gig_downshift_workaround_ich8lan(&adapter->hw);
1713 }
1714
1715 /**
1716  * e1000_intr_msi - Interrupt Handler
1717  * @irq: interrupt number
1718  * @data: pointer to a network interface device structure
1719  **/
1720 static irqreturn_t e1000_intr_msi(int __always_unused irq, void *data)
1721 {
1722         struct net_device *netdev = data;
1723         struct e1000_adapter *adapter = netdev_priv(netdev);
1724         struct e1000_hw *hw = &adapter->hw;
1725         u32 icr = er32(ICR);
1726
1727         /* read ICR disables interrupts using IAM */
1728         if (icr & E1000_ICR_LSC) {
1729                 hw->mac.get_link_status = true;
1730                 /* ICH8 workaround-- Call gig speed drop workaround on cable
1731                  * disconnect (LSC) before accessing any PHY registers
1732                  */
1733                 if ((adapter->flags & FLAG_LSC_GIG_SPEED_DROP) &&
1734                     (!(er32(STATUS) & E1000_STATUS_LU)))
1735                         schedule_work(&adapter->downshift_task);
1736
1737                 /* 80003ES2LAN workaround-- For packet buffer work-around on
1738                  * link down event; disable receives here in the ISR and reset
1739                  * adapter in watchdog
1740                  */
1741                 if (netif_carrier_ok(netdev) &&
1742                     adapter->flags & FLAG_RX_NEEDS_RESTART) {
1743                         /* disable receives */
1744                         u32 rctl = er32(RCTL);
1745                         ew32(RCTL, rctl & ~E1000_RCTL_EN);
1746                         adapter->flags |= FLAG_RESTART_NOW;
1747                 }
1748                 /* guard against interrupt when we're going down */
1749                 if (!test_bit(__E1000_DOWN, &adapter->state))
1750                         mod_timer(&adapter->watchdog_timer, jiffies + 1);
1751         }
1752
1753         /* Reset on uncorrectable ECC error */
1754         if ((icr & E1000_ICR_ECCER) && (hw->mac.type == e1000_pch_lpt)) {
1755                 u32 pbeccsts = er32(PBECCSTS);
1756
1757                 adapter->corr_errors +=
1758                     pbeccsts & E1000_PBECCSTS_CORR_ERR_CNT_MASK;
1759                 adapter->uncorr_errors +=
1760                     (pbeccsts & E1000_PBECCSTS_UNCORR_ERR_CNT_MASK) >>
1761                     E1000_PBECCSTS_UNCORR_ERR_CNT_SHIFT;
1762
1763                 /* Do the reset outside of interrupt context */
1764                 schedule_work(&adapter->reset_task);
1765
1766                 /* return immediately since reset is imminent */
1767                 return IRQ_HANDLED;
1768         }
1769
1770         if (napi_schedule_prep(&adapter->napi)) {
1771                 adapter->total_tx_bytes = 0;
1772                 adapter->total_tx_packets = 0;
1773                 adapter->total_rx_bytes = 0;
1774                 adapter->total_rx_packets = 0;
1775                 __napi_schedule(&adapter->napi);
1776         }
1777
1778         return IRQ_HANDLED;
1779 }
1780
1781 /**
1782  * e1000_intr - Interrupt Handler
1783  * @irq: interrupt number
1784  * @data: pointer to a network interface device structure
1785  **/
1786 static irqreturn_t e1000_intr(int __always_unused irq, void *data)
1787 {
1788         struct net_device *netdev = data;
1789         struct e1000_adapter *adapter = netdev_priv(netdev);
1790         struct e1000_hw *hw = &adapter->hw;
1791         u32 rctl, icr = er32(ICR);
1792
1793         if (!icr || test_bit(__E1000_DOWN, &adapter->state))
1794                 return IRQ_NONE;        /* Not our interrupt */
1795
1796         /* IMS will not auto-mask if INT_ASSERTED is not set, and if it is
1797          * not set, then the adapter didn't send an interrupt
1798          */
1799         if (!(icr & E1000_ICR_INT_ASSERTED))
1800                 return IRQ_NONE;
1801
1802         /* Interrupt Auto-Mask...upon reading ICR,
1803          * interrupts are masked.  No need for the
1804          * IMC write
1805          */
1806
1807         if (icr & E1000_ICR_LSC) {
1808                 hw->mac.get_link_status = true;
1809                 /* ICH8 workaround-- Call gig speed drop workaround on cable
1810                  * disconnect (LSC) before accessing any PHY registers
1811                  */
1812                 if ((adapter->flags & FLAG_LSC_GIG_SPEED_DROP) &&
1813                     (!(er32(STATUS) & E1000_STATUS_LU)))
1814                         schedule_work(&adapter->downshift_task);
1815
1816                 /* 80003ES2LAN workaround--
1817                  * For packet buffer work-around on link down event;
1818                  * disable receives here in the ISR and
1819                  * reset adapter in watchdog
1820                  */
1821                 if (netif_carrier_ok(netdev) &&
1822                     (adapter->flags & FLAG_RX_NEEDS_RESTART)) {
1823                         /* disable receives */
1824                         rctl = er32(RCTL);
1825                         ew32(RCTL, rctl & ~E1000_RCTL_EN);
1826                         adapter->flags |= FLAG_RESTART_NOW;
1827                 }
1828                 /* guard against interrupt when we're going down */
1829                 if (!test_bit(__E1000_DOWN, &adapter->state))
1830                         mod_timer(&adapter->watchdog_timer, jiffies + 1);
1831         }
1832
1833         /* Reset on uncorrectable ECC error */
1834         if ((icr & E1000_ICR_ECCER) && (hw->mac.type == e1000_pch_lpt)) {
1835                 u32 pbeccsts = er32(PBECCSTS);
1836
1837                 adapter->corr_errors +=
1838                     pbeccsts & E1000_PBECCSTS_CORR_ERR_CNT_MASK;
1839                 adapter->uncorr_errors +=
1840                     (pbeccsts & E1000_PBECCSTS_UNCORR_ERR_CNT_MASK) >>
1841                     E1000_PBECCSTS_UNCORR_ERR_CNT_SHIFT;
1842
1843                 /* Do the reset outside of interrupt context */
1844                 schedule_work(&adapter->reset_task);
1845
1846                 /* return immediately since reset is imminent */
1847                 return IRQ_HANDLED;
1848         }
1849
1850         if (napi_schedule_prep(&adapter->napi)) {
1851                 adapter->total_tx_bytes = 0;
1852                 adapter->total_tx_packets = 0;
1853                 adapter->total_rx_bytes = 0;
1854                 adapter->total_rx_packets = 0;
1855                 __napi_schedule(&adapter->napi);
1856         }
1857
1858         return IRQ_HANDLED;
1859 }
1860
1861 static irqreturn_t e1000_msix_other(int __always_unused irq, void *data)
1862 {
1863         struct net_device *netdev = data;
1864         struct e1000_adapter *adapter = netdev_priv(netdev);
1865         struct e1000_hw *hw = &adapter->hw;
1866         u32 icr = er32(ICR);
1867
1868         if (!(icr & E1000_ICR_INT_ASSERTED)) {
1869                 if (!test_bit(__E1000_DOWN, &adapter->state))
1870                         ew32(IMS, E1000_IMS_OTHER);
1871                 return IRQ_NONE;
1872         }
1873
1874         if (icr & adapter->eiac_mask)
1875                 ew32(ICS, (icr & adapter->eiac_mask));
1876
1877         if (icr & E1000_ICR_OTHER) {
1878                 if (!(icr & E1000_ICR_LSC))
1879                         goto no_link_interrupt;
1880                 hw->mac.get_link_status = true;
1881                 /* guard against interrupt when we're going down */
1882                 if (!test_bit(__E1000_DOWN, &adapter->state))
1883                         mod_timer(&adapter->watchdog_timer, jiffies + 1);
1884         }
1885
1886 no_link_interrupt:
1887         if (!test_bit(__E1000_DOWN, &adapter->state))
1888                 ew32(IMS, E1000_IMS_LSC | E1000_IMS_OTHER);
1889
1890         return IRQ_HANDLED;
1891 }
1892
1893 static irqreturn_t e1000_intr_msix_tx(int __always_unused irq, void *data)
1894 {
1895         struct net_device *netdev = data;
1896         struct e1000_adapter *adapter = netdev_priv(netdev);
1897         struct e1000_hw *hw = &adapter->hw;
1898         struct e1000_ring *tx_ring = adapter->tx_ring;
1899
1900         adapter->total_tx_bytes = 0;
1901         adapter->total_tx_packets = 0;
1902
1903         if (!e1000_clean_tx_irq(tx_ring))
1904                 /* Ring was not completely cleaned, so fire another interrupt */
1905                 ew32(ICS, tx_ring->ims_val);
1906
1907         return IRQ_HANDLED;
1908 }
1909
1910 static irqreturn_t e1000_intr_msix_rx(int __always_unused irq, void *data)
1911 {
1912         struct net_device *netdev = data;
1913         struct e1000_adapter *adapter = netdev_priv(netdev);
1914         struct e1000_ring *rx_ring = adapter->rx_ring;
1915
1916         /* Write the ITR value calculated at the end of the
1917          * previous interrupt.
1918          */
1919         if (rx_ring->set_itr) {
1920                 writel(1000000000 / (rx_ring->itr_val * 256),
1921                        rx_ring->itr_register);
1922                 rx_ring->set_itr = 0;
1923         }
1924
1925         if (napi_schedule_prep(&adapter->napi)) {
1926                 adapter->total_rx_bytes = 0;
1927                 adapter->total_rx_packets = 0;
1928                 __napi_schedule(&adapter->napi);
1929         }
1930         return IRQ_HANDLED;
1931 }
1932
1933 /**
1934  * e1000_configure_msix - Configure MSI-X hardware
1935  *
1936  * e1000_configure_msix sets up the hardware to properly
1937  * generate MSI-X interrupts.
1938  **/
1939 static void e1000_configure_msix(struct e1000_adapter *adapter)
1940 {
1941         struct e1000_hw *hw = &adapter->hw;
1942         struct e1000_ring *rx_ring = adapter->rx_ring;
1943         struct e1000_ring *tx_ring = adapter->tx_ring;
1944         int vector = 0;
1945         u32 ctrl_ext, ivar = 0;
1946
1947         adapter->eiac_mask = 0;
1948
1949         /* Workaround issue with spurious interrupts on 82574 in MSI-X mode */
1950         if (hw->mac.type == e1000_82574) {
1951                 u32 rfctl = er32(RFCTL);
1952                 rfctl |= E1000_RFCTL_ACK_DIS;
1953                 ew32(RFCTL, rfctl);
1954         }
1955
1956         /* Configure Rx vector */
1957         rx_ring->ims_val = E1000_IMS_RXQ0;
1958         adapter->eiac_mask |= rx_ring->ims_val;
1959         if (rx_ring->itr_val)
1960                 writel(1000000000 / (rx_ring->itr_val * 256),
1961                        rx_ring->itr_register);
1962         else
1963                 writel(1, rx_ring->itr_register);
1964         ivar = E1000_IVAR_INT_ALLOC_VALID | vector;
1965
1966         /* Configure Tx vector */
1967         tx_ring->ims_val = E1000_IMS_TXQ0;
1968         vector++;
1969         if (tx_ring->itr_val)
1970                 writel(1000000000 / (tx_ring->itr_val * 256),
1971                        tx_ring->itr_register);
1972         else
1973                 writel(1, tx_ring->itr_register);
1974         adapter->eiac_mask |= tx_ring->ims_val;
1975         ivar |= ((E1000_IVAR_INT_ALLOC_VALID | vector) << 8);
1976
1977         /* set vector for Other Causes, e.g. link changes */
1978         vector++;
1979         ivar |= ((E1000_IVAR_INT_ALLOC_VALID | vector) << 16);
1980         if (rx_ring->itr_val)
1981                 writel(1000000000 / (rx_ring->itr_val * 256),
1982                        hw->hw_addr + E1000_EITR_82574(vector));
1983         else
1984                 writel(1, hw->hw_addr + E1000_EITR_82574(vector));
1985
1986         /* Cause Tx interrupts on every write back */
1987         ivar |= (1 << 31);
1988
1989         ew32(IVAR, ivar);
1990
1991         /* enable MSI-X PBA support */
1992         ctrl_ext = er32(CTRL_EXT);
1993         ctrl_ext |= E1000_CTRL_EXT_PBA_CLR;
1994
1995         /* Auto-Mask Other interrupts upon ICR read */
1996         ew32(IAM, ~E1000_EIAC_MASK_82574 | E1000_IMS_OTHER);
1997         ctrl_ext |= E1000_CTRL_EXT_EIAME;
1998         ew32(CTRL_EXT, ctrl_ext);
1999         e1e_flush();
2000 }
2001
2002 void e1000e_reset_interrupt_capability(struct e1000_adapter *adapter)
2003 {
2004         if (adapter->msix_entries) {
2005                 pci_disable_msix(adapter->pdev);
2006                 kfree(adapter->msix_entries);
2007                 adapter->msix_entries = NULL;
2008         } else if (adapter->flags & FLAG_MSI_ENABLED) {
2009                 pci_disable_msi(adapter->pdev);
2010                 adapter->flags &= ~FLAG_MSI_ENABLED;
2011         }
2012 }
2013
2014 /**
2015  * e1000e_set_interrupt_capability - set MSI or MSI-X if supported
2016  *
2017  * Attempt to configure interrupts using the best available
2018  * capabilities of the hardware and kernel.
2019  **/
2020 void e1000e_set_interrupt_capability(struct e1000_adapter *adapter)
2021 {
2022         int err;
2023         int i;
2024
2025         switch (adapter->int_mode) {
2026         case E1000E_INT_MODE_MSIX:
2027                 if (adapter->flags & FLAG_HAS_MSIX) {
2028                         adapter->num_vectors = 3; /* RxQ0, TxQ0 and other */
2029                         adapter->msix_entries = kcalloc(adapter->num_vectors,
2030                                                         sizeof(struct
2031                                                                msix_entry),
2032                                                         GFP_KERNEL);
2033                         if (adapter->msix_entries) {
2034                                 struct e1000_adapter *a = adapter;
2035
2036                                 for (i = 0; i < adapter->num_vectors; i++)
2037                                         adapter->msix_entries[i].entry = i;
2038
2039                                 err = pci_enable_msix_range(a->pdev,
2040                                                             a->msix_entries,
2041                                                             a->num_vectors,
2042                                                             a->num_vectors);
2043                                 if (err > 0)
2044                                         return;
2045                         }
2046                         /* MSI-X failed, so fall through and try MSI */
2047                         e_err("Failed to initialize MSI-X interrupts.  Falling back to MSI interrupts.\n");
2048                         e1000e_reset_interrupt_capability(adapter);
2049                 }
2050                 adapter->int_mode = E1000E_INT_MODE_MSI;
2051                 /* Fall through */
2052         case E1000E_INT_MODE_MSI:
2053                 if (!pci_enable_msi(adapter->pdev)) {
2054                         adapter->flags |= FLAG_MSI_ENABLED;
2055                 } else {
2056                         adapter->int_mode = E1000E_INT_MODE_LEGACY;
2057                         e_err("Failed to initialize MSI interrupts.  Falling back to legacy interrupts.\n");
2058                 }
2059                 /* Fall through */
2060         case E1000E_INT_MODE_LEGACY:
2061                 /* Don't do anything; this is the system default */
2062                 break;
2063         }
2064
2065         /* store the number of vectors being used */
2066         adapter->num_vectors = 1;
2067 }
2068
2069 /**
2070  * e1000_request_msix - Initialize MSI-X interrupts
2071  *
2072  * e1000_request_msix allocates MSI-X vectors and requests interrupts from the
2073  * kernel.
2074  **/
2075 static int e1000_request_msix(struct e1000_adapter *adapter)
2076 {
2077         struct net_device *netdev = adapter->netdev;
2078         int err = 0, vector = 0;
2079
2080         if (strlen(netdev->name) < (IFNAMSIZ - 5))
2081                 snprintf(adapter->rx_ring->name,
2082                          sizeof(adapter->rx_ring->name) - 1,
2083                          "%s-rx-0", netdev->name);
2084         else
2085                 memcpy(adapter->rx_ring->name, netdev->name, IFNAMSIZ);
2086         err = request_irq(adapter->msix_entries[vector].vector,
2087                           e1000_intr_msix_rx, 0, adapter->rx_ring->name,
2088                           netdev);
2089         if (err)
2090                 return err;
2091         adapter->rx_ring->itr_register = adapter->hw.hw_addr +
2092             E1000_EITR_82574(vector);
2093         adapter->rx_ring->itr_val = adapter->itr;
2094         vector++;
2095
2096         if (strlen(netdev->name) < (IFNAMSIZ - 5))
2097                 snprintf(adapter->tx_ring->name,
2098                          sizeof(adapter->tx_ring->name) - 1,
2099                          "%s-tx-0", netdev->name);
2100         else
2101                 memcpy(adapter->tx_ring->name, netdev->name, IFNAMSIZ);
2102         err = request_irq(adapter->msix_entries[vector].vector,
2103                           e1000_intr_msix_tx, 0, adapter->tx_ring->name,
2104                           netdev);
2105         if (err)
2106                 return err;
2107         adapter->tx_ring->itr_register = adapter->hw.hw_addr +
2108             E1000_EITR_82574(vector);
2109         adapter->tx_ring->itr_val = adapter->itr;
2110         vector++;
2111
2112         err = request_irq(adapter->msix_entries[vector].vector,
2113                           e1000_msix_other, 0, netdev->name, netdev);
2114         if (err)
2115                 return err;
2116
2117         e1000_configure_msix(adapter);
2118
2119         return 0;
2120 }
2121
2122 /**
2123  * e1000_request_irq - initialize interrupts
2124  *
2125  * Attempts to configure interrupts using the best available
2126  * capabilities of the hardware and kernel.
2127  **/
2128 static int e1000_request_irq(struct e1000_adapter *adapter)
2129 {
2130         struct net_device *netdev = adapter->netdev;
2131         int err;
2132
2133         if (adapter->msix_entries) {
2134                 err = e1000_request_msix(adapter);
2135                 if (!err)
2136                         return err;
2137                 /* fall back to MSI */
2138                 e1000e_reset_interrupt_capability(adapter);
2139                 adapter->int_mode = E1000E_INT_MODE_MSI;
2140                 e1000e_set_interrupt_capability(adapter);
2141         }
2142         if (adapter->flags & FLAG_MSI_ENABLED) {
2143                 err = request_irq(adapter->pdev->irq, e1000_intr_msi, 0,
2144                                   netdev->name, netdev);
2145                 if (!err)
2146                         return err;
2147
2148                 /* fall back to legacy interrupt */
2149                 e1000e_reset_interrupt_capability(adapter);
2150                 adapter->int_mode = E1000E_INT_MODE_LEGACY;
2151         }
2152
2153         err = request_irq(adapter->pdev->irq, e1000_intr, IRQF_SHARED,
2154                           netdev->name, netdev);
2155         if (err)
2156                 e_err("Unable to allocate interrupt, Error: %d\n", err);
2157
2158         return err;
2159 }
2160
2161 static void e1000_free_irq(struct e1000_adapter *adapter)
2162 {
2163         struct net_device *netdev = adapter->netdev;
2164
2165         if (adapter->msix_entries) {
2166                 int vector = 0;
2167
2168                 free_irq(adapter->msix_entries[vector].vector, netdev);
2169                 vector++;
2170
2171                 free_irq(adapter->msix_entries[vector].vector, netdev);
2172                 vector++;
2173
2174                 /* Other Causes interrupt vector */
2175                 free_irq(adapter->msix_entries[vector].vector, netdev);
2176                 return;
2177         }
2178
2179         free_irq(adapter->pdev->irq, netdev);
2180 }
2181
2182 /**
2183  * e1000_irq_disable - Mask off interrupt generation on the NIC
2184  **/
2185 static void e1000_irq_disable(struct e1000_adapter *adapter)
2186 {
2187         struct e1000_hw *hw = &adapter->hw;
2188
2189         ew32(IMC, ~0);
2190         if (adapter->msix_entries)
2191                 ew32(EIAC_82574, 0);
2192         e1e_flush();
2193
2194         if (adapter->msix_entries) {
2195                 int i;
2196                 for (i = 0; i < adapter->num_vectors; i++)
2197                         synchronize_irq(adapter->msix_entries[i].vector);
2198         } else {
2199                 synchronize_irq(adapter->pdev->irq);
2200         }
2201 }
2202
2203 /**
2204  * e1000_irq_enable - Enable default interrupt generation settings
2205  **/
2206 static void e1000_irq_enable(struct e1000_adapter *adapter)
2207 {
2208         struct e1000_hw *hw = &adapter->hw;
2209
2210         if (adapter->msix_entries) {
2211                 ew32(EIAC_82574, adapter->eiac_mask & E1000_EIAC_MASK_82574);
2212                 ew32(IMS, adapter->eiac_mask | E1000_IMS_OTHER | E1000_IMS_LSC);
2213         } else if (hw->mac.type == e1000_pch_lpt) {
2214                 ew32(IMS, IMS_ENABLE_MASK | E1000_IMS_ECCER);
2215         } else {
2216                 ew32(IMS, IMS_ENABLE_MASK);
2217         }
2218         e1e_flush();
2219 }
2220
2221 /**
2222  * e1000e_get_hw_control - get control of the h/w from f/w
2223  * @adapter: address of board private structure
2224  *
2225  * e1000e_get_hw_control sets {CTRL_EXT|SWSM}:DRV_LOAD bit.
2226  * For ASF and Pass Through versions of f/w this means that
2227  * the driver is loaded. For AMT version (only with 82573)
2228  * of the f/w this means that the network i/f is open.
2229  **/
2230 void e1000e_get_hw_control(struct e1000_adapter *adapter)
2231 {
2232         struct e1000_hw *hw = &adapter->hw;
2233         u32 ctrl_ext;
2234         u32 swsm;
2235
2236         /* Let firmware know the driver has taken over */
2237         if (adapter->flags & FLAG_HAS_SWSM_ON_LOAD) {
2238                 swsm = er32(SWSM);
2239                 ew32(SWSM, swsm | E1000_SWSM_DRV_LOAD);
2240         } else if (adapter->flags & FLAG_HAS_CTRLEXT_ON_LOAD) {
2241                 ctrl_ext = er32(CTRL_EXT);
2242                 ew32(CTRL_EXT, ctrl_ext | E1000_CTRL_EXT_DRV_LOAD);
2243         }
2244 }
2245
2246 /**
2247  * e1000e_release_hw_control - release control of the h/w to f/w
2248  * @adapter: address of board private structure
2249  *
2250  * e1000e_release_hw_control resets {CTRL_EXT|SWSM}:DRV_LOAD bit.
2251  * For ASF and Pass Through versions of f/w this means that the
2252  * driver is no longer loaded. For AMT version (only with 82573) i
2253  * of the f/w this means that the network i/f is closed.
2254  *
2255  **/
2256 void e1000e_release_hw_control(struct e1000_adapter *adapter)
2257 {
2258         struct e1000_hw *hw = &adapter->hw;
2259         u32 ctrl_ext;
2260         u32 swsm;
2261
2262         /* Let firmware taken over control of h/w */
2263         if (adapter->flags & FLAG_HAS_SWSM_ON_LOAD) {
2264                 swsm = er32(SWSM);
2265                 ew32(SWSM, swsm & ~E1000_SWSM_DRV_LOAD);
2266         } else if (adapter->flags & FLAG_HAS_CTRLEXT_ON_LOAD) {
2267                 ctrl_ext = er32(CTRL_EXT);
2268                 ew32(CTRL_EXT, ctrl_ext & ~E1000_CTRL_EXT_DRV_LOAD);
2269         }
2270 }
2271
2272 /**
2273  * e1000_alloc_ring_dma - allocate memory for a ring structure
2274  **/
2275 static int e1000_alloc_ring_dma(struct e1000_adapter *adapter,
2276                                 struct e1000_ring *ring)
2277 {
2278         struct pci_dev *pdev = adapter->pdev;
2279
2280         ring->desc = dma_alloc_coherent(&pdev->dev, ring->size, &ring->dma,
2281                                         GFP_KERNEL);
2282         if (!ring->desc)
2283                 return -ENOMEM;
2284
2285         return 0;
2286 }
2287
2288 /**
2289  * e1000e_setup_tx_resources - allocate Tx resources (Descriptors)
2290  * @tx_ring: Tx descriptor ring
2291  *
2292  * Return 0 on success, negative on failure
2293  **/
2294 int e1000e_setup_tx_resources(struct e1000_ring *tx_ring)
2295 {
2296         struct e1000_adapter *adapter = tx_ring->adapter;
2297         int err = -ENOMEM, size;
2298
2299         size = sizeof(struct e1000_buffer) * tx_ring->count;
2300         tx_ring->buffer_info = vzalloc(size);
2301         if (!tx_ring->buffer_info)
2302                 goto err;
2303
2304         /* round up to nearest 4K */
2305         tx_ring->size = tx_ring->count * sizeof(struct e1000_tx_desc);
2306         tx_ring->size = ALIGN(tx_ring->size, 4096);
2307
2308         err = e1000_alloc_ring_dma(adapter, tx_ring);
2309         if (err)
2310                 goto err;
2311
2312         tx_ring->next_to_use = 0;
2313         tx_ring->next_to_clean = 0;
2314
2315         return 0;
2316 err:
2317         vfree(tx_ring->buffer_info);
2318         e_err("Unable to allocate memory for the transmit descriptor ring\n");
2319         return err;
2320 }
2321
2322 /**
2323  * e1000e_setup_rx_resources - allocate Rx resources (Descriptors)
2324  * @rx_ring: Rx descriptor ring
2325  *
2326  * Returns 0 on success, negative on failure
2327  **/
2328 int e1000e_setup_rx_resources(struct e1000_ring *rx_ring)
2329 {
2330         struct e1000_adapter *adapter = rx_ring->adapter;
2331         struct e1000_buffer *buffer_info;
2332         int i, size, desc_len, err = -ENOMEM;
2333
2334         size = sizeof(struct e1000_buffer) * rx_ring->count;
2335         rx_ring->buffer_info = vzalloc(size);
2336         if (!rx_ring->buffer_info)
2337                 goto err;
2338
2339         for (i = 0; i < rx_ring->count; i++) {
2340                 buffer_info = &rx_ring->buffer_info[i];
2341                 buffer_info->ps_pages = kcalloc(PS_PAGE_BUFFERS,
2342                                                 sizeof(struct e1000_ps_page),
2343                                                 GFP_KERNEL);
2344                 if (!buffer_info->ps_pages)
2345                         goto err_pages;
2346         }
2347
2348         desc_len = sizeof(union e1000_rx_desc_packet_split);
2349
2350         /* Round up to nearest 4K */
2351         rx_ring->size = rx_ring->count * desc_len;
2352         rx_ring->size = ALIGN(rx_ring->size, 4096);
2353
2354         err = e1000_alloc_ring_dma(adapter, rx_ring);
2355         if (err)
2356                 goto err_pages;
2357
2358         rx_ring->next_to_clean = 0;
2359         rx_ring->next_to_use = 0;
2360         rx_ring->rx_skb_top = NULL;
2361
2362         return 0;
2363
2364 err_pages:
2365         for (i = 0; i < rx_ring->count; i++) {
2366                 buffer_info = &rx_ring->buffer_info[i];
2367                 kfree(buffer_info->ps_pages);
2368         }
2369 err:
2370         vfree(rx_ring->buffer_info);
2371         e_err("Unable to allocate memory for the receive descriptor ring\n");
2372         return err;
2373 }
2374
2375 /**
2376  * e1000_clean_tx_ring - Free Tx Buffers
2377  * @tx_ring: Tx descriptor ring
2378  **/
2379 static void e1000_clean_tx_ring(struct e1000_ring *tx_ring)
2380 {
2381         struct e1000_adapter *adapter = tx_ring->adapter;
2382         struct e1000_buffer *buffer_info;
2383         unsigned long size;
2384         unsigned int i;
2385
2386         for (i = 0; i < tx_ring->count; i++) {
2387                 buffer_info = &tx_ring->buffer_info[i];
2388                 e1000_put_txbuf(tx_ring, buffer_info);
2389         }
2390
2391         netdev_reset_queue(adapter->netdev);
2392         size = sizeof(struct e1000_buffer) * tx_ring->count;
2393         memset(tx_ring->buffer_info, 0, size);
2394
2395         memset(tx_ring->desc, 0, tx_ring->size);
2396
2397         tx_ring->next_to_use = 0;
2398         tx_ring->next_to_clean = 0;
2399
2400         writel(0, tx_ring->head);
2401         if (adapter->flags2 & FLAG2_PCIM2PCI_ARBITER_WA)
2402                 e1000e_update_tdt_wa(tx_ring, 0);
2403         else
2404                 writel(0, tx_ring->tail);
2405 }
2406
2407 /**
2408  * e1000e_free_tx_resources - Free Tx Resources per Queue
2409  * @tx_ring: Tx descriptor ring
2410  *
2411  * Free all transmit software resources
2412  **/
2413 void e1000e_free_tx_resources(struct e1000_ring *tx_ring)
2414 {
2415         struct e1000_adapter *adapter = tx_ring->adapter;
2416         struct pci_dev *pdev = adapter->pdev;
2417
2418         e1000_clean_tx_ring(tx_ring);
2419
2420         vfree(tx_ring->buffer_info);
2421         tx_ring->buffer_info = NULL;
2422
2423         dma_free_coherent(&pdev->dev, tx_ring->size, tx_ring->desc,
2424                           tx_ring->dma);
2425         tx_ring->desc = NULL;
2426 }
2427
2428 /**
2429  * e1000e_free_rx_resources - Free Rx Resources
2430  * @rx_ring: Rx descriptor ring
2431  *
2432  * Free all receive software resources
2433  **/
2434 void e1000e_free_rx_resources(struct e1000_ring *rx_ring)
2435 {
2436         struct e1000_adapter *adapter = rx_ring->adapter;
2437         struct pci_dev *pdev = adapter->pdev;
2438         int i;
2439
2440         e1000_clean_rx_ring(rx_ring);
2441
2442         for (i = 0; i < rx_ring->count; i++)
2443                 kfree(rx_ring->buffer_info[i].ps_pages);
2444
2445         vfree(rx_ring->buffer_info);
2446         rx_ring->buffer_info = NULL;
2447
2448         dma_free_coherent(&pdev->dev, rx_ring->size, rx_ring->desc,
2449                           rx_ring->dma);
2450         rx_ring->desc = NULL;
2451 }
2452
2453 /**
2454  * e1000_update_itr - update the dynamic ITR value based on statistics
2455  * @adapter: pointer to adapter
2456  * @itr_setting: current adapter->itr
2457  * @packets: the number of packets during this measurement interval
2458  * @bytes: the number of bytes during this measurement interval
2459  *
2460  *      Stores a new ITR value based on packets and byte
2461  *      counts during the last interrupt.  The advantage of per interrupt
2462  *      computation is faster updates and more accurate ITR for the current
2463  *      traffic pattern.  Constants in this function were computed
2464  *      based on theoretical maximum wire speed and thresholds were set based
2465  *      on testing data as well as attempting to minimize response time
2466  *      while increasing bulk throughput.  This functionality is controlled
2467  *      by the InterruptThrottleRate module parameter.
2468  **/
2469 static unsigned int e1000_update_itr(u16 itr_setting, int packets, int bytes)
2470 {
2471         unsigned int retval = itr_setting;
2472
2473         if (packets == 0)
2474                 return itr_setting;
2475
2476         switch (itr_setting) {
2477         case lowest_latency:
2478                 /* handle TSO and jumbo frames */
2479                 if (bytes / packets > 8000)
2480                         retval = bulk_latency;
2481                 else if ((packets < 5) && (bytes > 512))
2482                         retval = low_latency;
2483                 break;
2484         case low_latency:       /* 50 usec aka 20000 ints/s */
2485                 if (bytes > 10000) {
2486                         /* this if handles the TSO accounting */
2487                         if (bytes / packets > 8000)
2488                                 retval = bulk_latency;
2489                         else if ((packets < 10) || ((bytes / packets) > 1200))
2490                                 retval = bulk_latency;
2491                         else if ((packets > 35))
2492                                 retval = lowest_latency;
2493                 } else if (bytes / packets > 2000) {
2494                         retval = bulk_latency;
2495                 } else if (packets <= 2 && bytes < 512) {
2496                         retval = lowest_latency;
2497                 }
2498                 break;
2499         case bulk_latency:      /* 250 usec aka 4000 ints/s */
2500                 if (bytes > 25000) {
2501                         if (packets > 35)
2502                                 retval = low_latency;
2503                 } else if (bytes < 6000) {
2504                         retval = low_latency;
2505                 }
2506                 break;
2507         }
2508
2509         return retval;
2510 }
2511
2512 static void e1000_set_itr(struct e1000_adapter *adapter)
2513 {
2514         u16 current_itr;
2515         u32 new_itr = adapter->itr;
2516
2517         /* for non-gigabit speeds, just fix the interrupt rate at 4000 */
2518         if (adapter->link_speed != SPEED_1000) {
2519                 current_itr = 0;
2520                 new_itr = 4000;
2521                 goto set_itr_now;
2522         }
2523
2524         if (adapter->flags2 & FLAG2_DISABLE_AIM) {
2525                 new_itr = 0;
2526                 goto set_itr_now;
2527         }
2528
2529         adapter->tx_itr = e1000_update_itr(adapter->tx_itr,
2530                                            adapter->total_tx_packets,
2531                                            adapter->total_tx_bytes);
2532         /* conservative mode (itr 3) eliminates the lowest_latency setting */
2533         if (adapter->itr_setting == 3 && adapter->tx_itr == lowest_latency)
2534                 adapter->tx_itr = low_latency;
2535
2536         adapter->rx_itr = e1000_update_itr(adapter->rx_itr,
2537                                            adapter->total_rx_packets,
2538                                            adapter->total_rx_bytes);
2539         /* conservative mode (itr 3) eliminates the lowest_latency setting */
2540         if (adapter->itr_setting == 3 && adapter->rx_itr == lowest_latency)
2541                 adapter->rx_itr = low_latency;
2542
2543         current_itr = max(adapter->rx_itr, adapter->tx_itr);
2544
2545         /* counts and packets in update_itr are dependent on these numbers */
2546         switch (current_itr) {
2547         case lowest_latency:
2548                 new_itr = 70000;
2549                 break;
2550         case low_latency:
2551                 new_itr = 20000;        /* aka hwitr = ~200 */
2552                 break;
2553         case bulk_latency:
2554                 new_itr = 4000;
2555                 break;
2556         default:
2557                 break;
2558         }
2559
2560 set_itr_now:
2561         if (new_itr != adapter->itr) {
2562                 /* this attempts to bias the interrupt rate towards Bulk
2563                  * by adding intermediate steps when interrupt rate is
2564                  * increasing
2565                  */
2566                 new_itr = new_itr > adapter->itr ?
2567                     min(adapter->itr + (new_itr >> 2), new_itr) : new_itr;
2568                 adapter->itr = new_itr;
2569                 adapter->rx_ring->itr_val = new_itr;
2570                 if (adapter->msix_entries)
2571                         adapter->rx_ring->set_itr = 1;
2572                 else
2573                         e1000e_write_itr(adapter, new_itr);
2574         }
2575 }
2576
2577 /**
2578  * e1000e_write_itr - write the ITR value to the appropriate registers
2579  * @adapter: address of board private structure
2580  * @itr: new ITR value to program
2581  *
2582  * e1000e_write_itr determines if the adapter is in MSI-X mode
2583  * and, if so, writes the EITR registers with the ITR value.
2584  * Otherwise, it writes the ITR value into the ITR register.
2585  **/
2586 void e1000e_write_itr(struct e1000_adapter *adapter, u32 itr)
2587 {
2588         struct e1000_hw *hw = &adapter->hw;
2589         u32 new_itr = itr ? 1000000000 / (itr * 256) : 0;
2590
2591         if (adapter->msix_entries) {
2592                 int vector;
2593
2594                 for (vector = 0; vector < adapter->num_vectors; vector++)
2595                         writel(new_itr, hw->hw_addr + E1000_EITR_82574(vector));
2596         } else {
2597                 ew32(ITR, new_itr);
2598         }
2599 }
2600
2601 /**
2602  * e1000_alloc_queues - Allocate memory for all rings
2603  * @adapter: board private structure to initialize
2604  **/
2605 static int e1000_alloc_queues(struct e1000_adapter *adapter)
2606 {
2607         int size = sizeof(struct e1000_ring);
2608
2609         adapter->tx_ring = kzalloc(size, GFP_KERNEL);
2610         if (!adapter->tx_ring)
2611                 goto err;
2612         adapter->tx_ring->count = adapter->tx_ring_count;
2613         adapter->tx_ring->adapter = adapter;
2614
2615         adapter->rx_ring = kzalloc(size, GFP_KERNEL);
2616         if (!adapter->rx_ring)
2617                 goto err;
2618         adapter->rx_ring->count = adapter->rx_ring_count;
2619         adapter->rx_ring->adapter = adapter;
2620
2621         return 0;
2622 err:
2623         e_err("Unable to allocate memory for queues\n");
2624         kfree(adapter->rx_ring);
2625         kfree(adapter->tx_ring);
2626         return -ENOMEM;
2627 }
2628
2629 /**
2630  * e1000e_poll - NAPI Rx polling callback
2631  * @napi: struct associated with this polling callback
2632  * @weight: number of packets driver is allowed to process this poll
2633  **/
2634 static int e1000e_poll(struct napi_struct *napi, int weight)
2635 {
2636         struct e1000_adapter *adapter = container_of(napi, struct e1000_adapter,
2637                                                      napi);
2638         struct e1000_hw *hw = &adapter->hw;
2639         struct net_device *poll_dev = adapter->netdev;
2640         int tx_cleaned = 1, work_done = 0;
2641
2642         adapter = netdev_priv(poll_dev);
2643
2644         if (!adapter->msix_entries ||
2645             (adapter->rx_ring->ims_val & adapter->tx_ring->ims_val))
2646                 tx_cleaned = e1000_clean_tx_irq(adapter->tx_ring);
2647
2648         adapter->clean_rx(adapter->rx_ring, &work_done, weight);
2649
2650         if (!tx_cleaned)
2651                 work_done = weight;
2652
2653         /* If weight not fully consumed, exit the polling mode */
2654         if (work_done < weight) {
2655                 if (adapter->itr_setting & 3)
2656                         e1000_set_itr(adapter);
2657                 napi_complete(napi);
2658                 if (!test_bit(__E1000_DOWN, &adapter->state)) {
2659                         if (adapter->msix_entries)
2660                                 ew32(IMS, adapter->rx_ring->ims_val);
2661                         else
2662                                 e1000_irq_enable(adapter);
2663                 }
2664         }
2665
2666         return work_done;
2667 }
2668
2669 static int e1000_vlan_rx_add_vid(struct net_device *netdev,
2670                                  __always_unused __be16 proto, u16 vid)
2671 {
2672         struct e1000_adapter *adapter = netdev_priv(netdev);
2673         struct e1000_hw *hw = &adapter->hw;
2674         u32 vfta, index;
2675
2676         /* don't update vlan cookie if already programmed */
2677         if ((adapter->hw.mng_cookie.status &
2678              E1000_MNG_DHCP_COOKIE_STATUS_VLAN) &&
2679             (vid == adapter->mng_vlan_id))
2680                 return 0;
2681
2682         /* add VID to filter table */
2683         if (adapter->flags & FLAG_HAS_HW_VLAN_FILTER) {
2684                 index = (vid >> 5) & 0x7F;
2685                 vfta = E1000_READ_REG_ARRAY(hw, E1000_VFTA, index);
2686                 vfta |= (1 << (vid & 0x1F));
2687                 hw->mac.ops.write_vfta(hw, index, vfta);
2688         }
2689
2690         set_bit(vid, adapter->active_vlans);
2691
2692         return 0;
2693 }
2694
2695 static int e1000_vlan_rx_kill_vid(struct net_device *netdev,
2696                                   __always_unused __be16 proto, u16 vid)
2697 {
2698         struct e1000_adapter *adapter = netdev_priv(netdev);
2699         struct e1000_hw *hw = &adapter->hw;
2700         u32 vfta, index;
2701
2702         if ((adapter->hw.mng_cookie.status &
2703              E1000_MNG_DHCP_COOKIE_STATUS_VLAN) &&
2704             (vid == adapter->mng_vlan_id)) {
2705                 /* release control to f/w */
2706                 e1000e_release_hw_control(adapter);
2707                 return 0;
2708         }
2709
2710         /* remove VID from filter table */
2711         if (adapter->flags & FLAG_HAS_HW_VLAN_FILTER) {
2712                 index = (vid >> 5) & 0x7F;
2713                 vfta = E1000_READ_REG_ARRAY(hw, E1000_VFTA, index);
2714                 vfta &= ~(1 << (vid & 0x1F));
2715                 hw->mac.ops.write_vfta(hw, index, vfta);
2716         }
2717
2718         clear_bit(vid, adapter->active_vlans);
2719
2720         return 0;
2721 }
2722
2723 /**
2724  * e1000e_vlan_filter_disable - helper to disable hw VLAN filtering
2725  * @adapter: board private structure to initialize
2726  **/
2727 static void e1000e_vlan_filter_disable(struct e1000_adapter *adapter)
2728 {
2729         struct net_device *netdev = adapter->netdev;
2730         struct e1000_hw *hw = &adapter->hw;
2731         u32 rctl;
2732
2733         if (adapter->flags & FLAG_HAS_HW_VLAN_FILTER) {
2734                 /* disable VLAN receive filtering */
2735                 rctl = er32(RCTL);
2736                 rctl &= ~(E1000_RCTL_VFE | E1000_RCTL_CFIEN);
2737                 ew32(RCTL, rctl);
2738
2739                 if (adapter->mng_vlan_id != (u16)E1000_MNG_VLAN_NONE) {
2740                         e1000_vlan_rx_kill_vid(netdev, htons(ETH_P_8021Q),
2741                                                adapter->mng_vlan_id);
2742                         adapter->mng_vlan_id = E1000_MNG_VLAN_NONE;
2743                 }
2744         }
2745 }
2746
2747 /**
2748  * e1000e_vlan_filter_enable - helper to enable HW VLAN filtering
2749  * @adapter: board private structure to initialize
2750  **/
2751 static void e1000e_vlan_filter_enable(struct e1000_adapter *adapter)
2752 {
2753         struct e1000_hw *hw = &adapter->hw;
2754         u32 rctl;
2755
2756         if (adapter->flags & FLAG_HAS_HW_VLAN_FILTER) {
2757                 /* enable VLAN receive filtering */
2758                 rctl = er32(RCTL);
2759                 rctl |= E1000_RCTL_VFE;
2760                 rctl &= ~E1000_RCTL_CFIEN;
2761                 ew32(RCTL, rctl);
2762         }
2763 }
2764
2765 /**
2766  * e1000e_vlan_strip_enable - helper to disable HW VLAN stripping
2767  * @adapter: board private structure to initialize
2768  **/
2769 static void e1000e_vlan_strip_disable(struct e1000_adapter *adapter)
2770 {
2771         struct e1000_hw *hw = &adapter->hw;
2772         u32 ctrl;
2773
2774         /* disable VLAN tag insert/strip */
2775         ctrl = er32(CTRL);
2776         ctrl &= ~E1000_CTRL_VME;
2777         ew32(CTRL, ctrl);
2778 }
2779
2780 /**
2781  * e1000e_vlan_strip_enable - helper to enable HW VLAN stripping
2782  * @adapter: board private structure to initialize
2783  **/
2784 static void e1000e_vlan_strip_enable(struct e1000_adapter *adapter)
2785 {
2786         struct e1000_hw *hw = &adapter->hw;
2787         u32 ctrl;
2788
2789         /* enable VLAN tag insert/strip */
2790         ctrl = er32(CTRL);
2791         ctrl |= E1000_CTRL_VME;
2792         ew32(CTRL, ctrl);
2793 }
2794
2795 static void e1000_update_mng_vlan(struct e1000_adapter *adapter)
2796 {
2797         struct net_device *netdev = adapter->netdev;
2798         u16 vid = adapter->hw.mng_cookie.vlan_id;
2799         u16 old_vid = adapter->mng_vlan_id;
2800
2801         if (adapter->hw.mng_cookie.status & E1000_MNG_DHCP_COOKIE_STATUS_VLAN) {
2802                 e1000_vlan_rx_add_vid(netdev, htons(ETH_P_8021Q), vid);
2803                 adapter->mng_vlan_id = vid;
2804         }
2805
2806         if ((old_vid != (u16)E1000_MNG_VLAN_NONE) && (vid != old_vid))
2807                 e1000_vlan_rx_kill_vid(netdev, htons(ETH_P_8021Q), old_vid);
2808 }
2809
2810 static void e1000_restore_vlan(struct e1000_adapter *adapter)
2811 {
2812         u16 vid;
2813
2814         e1000_vlan_rx_add_vid(adapter->netdev, htons(ETH_P_8021Q), 0);
2815
2816         for_each_set_bit(vid, adapter->active_vlans, VLAN_N_VID)
2817             e1000_vlan_rx_add_vid(adapter->netdev, htons(ETH_P_8021Q), vid);
2818 }
2819
2820 static void e1000_init_manageability_pt(struct e1000_adapter *adapter)
2821 {
2822         struct e1000_hw *hw = &adapter->hw;
2823         u32 manc, manc2h, mdef, i, j;
2824
2825         if (!(adapter->flags & FLAG_MNG_PT_ENABLED))
2826                 return;
2827
2828         manc = er32(MANC);
2829
2830         /* enable receiving management packets to the host. this will probably
2831          * generate destination unreachable messages from the host OS, but
2832          * the packets will be handled on SMBUS
2833          */
2834         manc |= E1000_MANC_EN_MNG2HOST;
2835         manc2h = er32(MANC2H);
2836
2837         switch (hw->mac.type) {
2838         default:
2839                 manc2h |= (E1000_MANC2H_PORT_623 | E1000_MANC2H_PORT_664);
2840                 break;
2841         case e1000_82574:
2842         case e1000_82583:
2843                 /* Check if IPMI pass-through decision filter already exists;
2844                  * if so, enable it.
2845                  */
2846                 for (i = 0, j = 0; i < 8; i++) {
2847                         mdef = er32(MDEF(i));
2848
2849                         /* Ignore filters with anything other than IPMI ports */
2850                         if (mdef & ~(E1000_MDEF_PORT_623 | E1000_MDEF_PORT_664))
2851                                 continue;
2852
2853                         /* Enable this decision filter in MANC2H */
2854                         if (mdef)
2855                                 manc2h |= (1 << i);
2856
2857                         j |= mdef;
2858                 }
2859
2860                 if (j == (E1000_MDEF_PORT_623 | E1000_MDEF_PORT_664))
2861                         break;
2862
2863                 /* Create new decision filter in an empty filter */
2864                 for (i = 0, j = 0; i < 8; i++)
2865                         if (er32(MDEF(i)) == 0) {
2866                                 ew32(MDEF(i), (E1000_MDEF_PORT_623 |
2867                                                E1000_MDEF_PORT_664));
2868                                 manc2h |= (1 << 1);
2869                                 j++;
2870                                 break;
2871                         }
2872
2873                 if (!j)
2874                         e_warn("Unable to create IPMI pass-through filter\n");
2875                 break;
2876         }
2877
2878         ew32(MANC2H, manc2h);
2879         ew32(MANC, manc);
2880 }
2881
2882 /**
2883  * e1000_configure_tx - Configure Transmit Unit after Reset
2884  * @adapter: board private structure
2885  *
2886  * Configure the Tx unit of the MAC after a reset.
2887  **/
2888 static void e1000_configure_tx(struct e1000_adapter *adapter)
2889 {
2890         struct e1000_hw *hw = &adapter->hw;
2891         struct e1000_ring *tx_ring = adapter->tx_ring;
2892         u64 tdba;
2893         u32 tdlen, tarc;
2894
2895         /* Setup the HW Tx Head and Tail descriptor pointers */
2896         tdba = tx_ring->dma;
2897         tdlen = tx_ring->count * sizeof(struct e1000_tx_desc);
2898         ew32(TDBAL(0), (tdba & DMA_BIT_MASK(32)));
2899         ew32(TDBAH(0), (tdba >> 32));
2900         ew32(TDLEN(0), tdlen);
2901         ew32(TDH(0), 0);
2902         ew32(TDT(0), 0);
2903         tx_ring->head = adapter->hw.hw_addr + E1000_TDH(0);
2904         tx_ring->tail = adapter->hw.hw_addr + E1000_TDT(0);
2905
2906         /* Set the Tx Interrupt Delay register */
2907         ew32(TIDV, adapter->tx_int_delay);
2908         /* Tx irq moderation */
2909         ew32(TADV, adapter->tx_abs_int_delay);
2910
2911         if (adapter->flags2 & FLAG2_DMA_BURST) {
2912                 u32 txdctl = er32(TXDCTL(0));
2913                 txdctl &= ~(E1000_TXDCTL_PTHRESH | E1000_TXDCTL_HTHRESH |
2914                             E1000_TXDCTL_WTHRESH);
2915                 /* set up some performance related parameters to encourage the
2916                  * hardware to use the bus more efficiently in bursts, depends
2917                  * on the tx_int_delay to be enabled,
2918                  * wthresh = 1 ==> burst write is disabled to avoid Tx stalls
2919                  * hthresh = 1 ==> prefetch when one or more available
2920                  * pthresh = 0x1f ==> prefetch if internal cache 31 or less
2921                  * BEWARE: this seems to work but should be considered first if
2922                  * there are Tx hangs or other Tx related bugs
2923                  */
2924                 txdctl |= E1000_TXDCTL_DMA_BURST_ENABLE;
2925                 ew32(TXDCTL(0), txdctl);
2926         }
2927         /* erratum work around: set txdctl the same for both queues */
2928         ew32(TXDCTL(1), er32(TXDCTL(0)));
2929
2930         if (adapter->flags & FLAG_TARC_SPEED_MODE_BIT) {
2931                 tarc = er32(TARC(0));
2932                 /* set the speed mode bit, we'll clear it if we're not at
2933                  * gigabit link later
2934                  */
2935 #define SPEED_MODE_BIT (1 << 21)
2936                 tarc |= SPEED_MODE_BIT;
2937                 ew32(TARC(0), tarc);
2938         }
2939
2940         /* errata: program both queues to unweighted RR */
2941         if (adapter->flags & FLAG_TARC_SET_BIT_ZERO) {
2942                 tarc = er32(TARC(0));
2943                 tarc |= 1;
2944                 ew32(TARC(0), tarc);
2945                 tarc = er32(TARC(1));
2946                 tarc |= 1;
2947                 ew32(TARC(1), tarc);
2948         }
2949
2950         /* Setup Transmit Descriptor Settings for eop descriptor */
2951         adapter->txd_cmd = E1000_TXD_CMD_EOP | E1000_TXD_CMD_IFCS;
2952
2953         /* only set IDE if we are delaying interrupts using the timers */
2954         if (adapter->tx_int_delay)
2955                 adapter->txd_cmd |= E1000_TXD_CMD_IDE;
2956
2957         /* enable Report Status bit */
2958         adapter->txd_cmd |= E1000_TXD_CMD_RS;
2959
2960         hw->mac.ops.config_collision_dist(hw);
2961 }
2962
2963 /**
2964  * e1000_setup_rctl - configure the receive control registers
2965  * @adapter: Board private structure
2966  **/
2967 #define PAGE_USE_COUNT(S) (((S) >> PAGE_SHIFT) + \
2968                            (((S) & (PAGE_SIZE - 1)) ? 1 : 0))
2969 static void e1000_setup_rctl(struct e1000_adapter *adapter)
2970 {
2971         struct e1000_hw *hw = &adapter->hw;
2972         u32 rctl, rfctl;
2973         u32 pages = 0;
2974
2975         /* Workaround Si errata on PCHx - configure jumbo frame flow */
2976         if ((hw->mac.type >= e1000_pch2lan) &&
2977             (adapter->netdev->mtu > ETH_DATA_LEN) &&
2978             e1000_lv_jumbo_workaround_ich8lan(hw, true))
2979                 e_dbg("failed to enable jumbo frame workaround mode\n");
2980
2981         /* Program MC offset vector base */
2982         rctl = er32(RCTL);
2983         rctl &= ~(3 << E1000_RCTL_MO_SHIFT);
2984         rctl |= E1000_RCTL_EN | E1000_RCTL_BAM |
2985             E1000_RCTL_LBM_NO | E1000_RCTL_RDMTS_HALF |
2986             (adapter->hw.mac.mc_filter_type << E1000_RCTL_MO_SHIFT);
2987
2988         /* Do not Store bad packets */
2989         rctl &= ~E1000_RCTL_SBP;
2990
2991         /* Enable Long Packet receive */
2992         if (adapter->netdev->mtu <= ETH_DATA_LEN)
2993                 rctl &= ~E1000_RCTL_LPE;
2994         else
2995                 rctl |= E1000_RCTL_LPE;
2996
2997         /* Some systems expect that the CRC is included in SMBUS traffic. The
2998          * hardware strips the CRC before sending to both SMBUS (BMC) and to
2999          * host memory when this is enabled
3000          */
3001         if (adapter->flags2 & FLAG2_CRC_STRIPPING)
3002                 rctl |= E1000_RCTL_SECRC;
3003
3004         /* Workaround Si errata on 82577 PHY - configure IPG for jumbos */
3005         if ((hw->phy.type == e1000_phy_82577) && (rctl & E1000_RCTL_LPE)) {
3006                 u16 phy_data;
3007
3008                 e1e_rphy(hw, PHY_REG(770, 26), &phy_data);
3009                 phy_data &= 0xfff8;
3010                 phy_data |= (1 << 2);
3011                 e1e_wphy(hw, PHY_REG(770, 26), phy_data);
3012
3013                 e1e_rphy(hw, 22, &phy_data);
3014                 phy_data &= 0x0fff;
3015                 phy_data |= (1 << 14);
3016                 e1e_wphy(hw, 0x10, 0x2823);
3017                 e1e_wphy(hw, 0x11, 0x0003);
3018                 e1e_wphy(hw, 22, phy_data);
3019         }
3020
3021         /* Setup buffer sizes */
3022         rctl &= ~E1000_RCTL_SZ_4096;
3023         rctl |= E1000_RCTL_BSEX;
3024         switch (adapter->rx_buffer_len) {
3025         case 2048:
3026         default:
3027                 rctl |= E1000_RCTL_SZ_2048;
3028                 rctl &= ~E1000_RCTL_BSEX;
3029                 break;
3030         case 4096:
3031                 rctl |= E1000_RCTL_SZ_4096;
3032                 break;
3033         case 8192:
3034                 rctl |= E1000_RCTL_SZ_8192;
3035                 break;
3036         case 16384:
3037                 rctl |= E1000_RCTL_SZ_16384;
3038                 break;
3039         }
3040
3041         /* Enable Extended Status in all Receive Descriptors */
3042         rfctl = er32(RFCTL);
3043         rfctl |= E1000_RFCTL_EXTEN;
3044         ew32(RFCTL, rfctl);
3045
3046         /* 82571 and greater support packet-split where the protocol
3047          * header is placed in skb->data and the packet data is
3048          * placed in pages hanging off of skb_shinfo(skb)->nr_frags.
3049          * In the case of a non-split, skb->data is linearly filled,
3050          * followed by the page buffers.  Therefore, skb->data is
3051          * sized to hold the largest protocol header.
3052          *
3053          * allocations using alloc_page take too long for regular MTU
3054          * so only enable packet split for jumbo frames
3055          *
3056          * Using pages when the page size is greater than 16k wastes
3057          * a lot of memory, since we allocate 3 pages at all times
3058          * per packet.
3059          */
3060         pages = PAGE_USE_COUNT(adapter->netdev->mtu);
3061         if ((pages <= 3) && (PAGE_SIZE <= 16384) && (rctl & E1000_RCTL_LPE))
3062                 adapter->rx_ps_pages = pages;
3063         else
3064                 adapter->rx_ps_pages = 0;
3065
3066         if (adapter->rx_ps_pages) {
3067                 u32 psrctl = 0;
3068
3069                 /* Enable Packet split descriptors */
3070                 rctl |= E1000_RCTL_DTYP_PS;
3071
3072                 psrctl |= adapter->rx_ps_bsize0 >> E1000_PSRCTL_BSIZE0_SHIFT;
3073
3074                 switch (adapter->rx_ps_pages) {
3075                 case 3:
3076                         psrctl |= PAGE_SIZE << E1000_PSRCTL_BSIZE3_SHIFT;
3077                         /* fall-through */
3078                 case 2:
3079                         psrctl |= PAGE_SIZE << E1000_PSRCTL_BSIZE2_SHIFT;
3080                         /* fall-through */
3081                 case 1:
3082                         psrctl |= PAGE_SIZE >> E1000_PSRCTL_BSIZE1_SHIFT;
3083                         break;
3084                 }
3085
3086                 ew32(PSRCTL, psrctl);
3087         }
3088
3089         /* This is useful for sniffing bad packets. */
3090         if (adapter->netdev->features & NETIF_F_RXALL) {
3091                 /* UPE and MPE will be handled by normal PROMISC logic
3092                  * in e1000e_set_rx_mode
3093                  */
3094                 rctl |= (E1000_RCTL_SBP |       /* Receive bad packets */
3095                          E1000_RCTL_BAM |       /* RX All Bcast Pkts */
3096                          E1000_RCTL_PMCF);      /* RX All MAC Ctrl Pkts */
3097
3098                 rctl &= ~(E1000_RCTL_VFE |      /* Disable VLAN filter */
3099                           E1000_RCTL_DPF |      /* Allow filtered pause */
3100                           E1000_RCTL_CFIEN);    /* Dis VLAN CFIEN Filter */
3101                 /* Do not mess with E1000_CTRL_VME, it affects transmit as well,
3102                  * and that breaks VLANs.
3103                  */
3104         }
3105
3106         ew32(RCTL, rctl);
3107         /* just started the receive unit, no need to restart */
3108         adapter->flags &= ~FLAG_RESTART_NOW;
3109 }
3110
3111 /**
3112  * e1000_configure_rx - Configure Receive Unit after Reset
3113  * @adapter: board private structure
3114  *
3115  * Configure the Rx unit of the MAC after a reset.
3116  **/
3117 static void e1000_configure_rx(struct e1000_adapter *adapter)
3118 {
3119         struct e1000_hw *hw = &adapter->hw;
3120         struct e1000_ring *rx_ring = adapter->rx_ring;
3121         u64 rdba;
3122         u32 rdlen, rctl, rxcsum, ctrl_ext;
3123
3124         if (adapter->rx_ps_pages) {
3125                 /* this is a 32 byte descriptor */
3126                 rdlen = rx_ring->count *
3127                     sizeof(union e1000_rx_desc_packet_split);
3128                 adapter->clean_rx = e1000_clean_rx_irq_ps;
3129                 adapter->alloc_rx_buf = e1000_alloc_rx_buffers_ps;
3130         } else if (adapter->netdev->mtu > ETH_FRAME_LEN + ETH_FCS_LEN) {
3131                 rdlen = rx_ring->count * sizeof(union e1000_rx_desc_extended);
3132                 adapter->clean_rx = e1000_clean_jumbo_rx_irq;
3133                 adapter->alloc_rx_buf = e1000_alloc_jumbo_rx_buffers;
3134         } else {
3135                 rdlen = rx_ring->count * sizeof(union e1000_rx_desc_extended);
3136                 adapter->clean_rx = e1000_clean_rx_irq;
3137                 adapter->alloc_rx_buf = e1000_alloc_rx_buffers;
3138         }
3139
3140         /* disable receives while setting up the descriptors */
3141         rctl = er32(RCTL);
3142         if (!(adapter->flags2 & FLAG2_NO_DISABLE_RX))
3143                 ew32(RCTL, rctl & ~E1000_RCTL_EN);
3144         e1e_flush();
3145         usleep_range(10000, 20000);
3146
3147         if (adapter->flags2 & FLAG2_DMA_BURST) {
3148                 /* set the writeback threshold (only takes effect if the RDTR
3149                  * is set). set GRAN=1 and write back up to 0x4 worth, and
3150                  * enable prefetching of 0x20 Rx descriptors
3151                  * granularity = 01
3152                  * wthresh = 04,
3153                  * hthresh = 04,
3154                  * pthresh = 0x20
3155                  */
3156                 ew32(RXDCTL(0), E1000_RXDCTL_DMA_BURST_ENABLE);
3157                 ew32(RXDCTL(1), E1000_RXDCTL_DMA_BURST_ENABLE);
3158
3159                 /* override the delay timers for enabling bursting, only if
3160                  * the value was not set by the user via module options
3161                  */
3162                 if (adapter->rx_int_delay == DEFAULT_RDTR)
3163                         adapter->rx_int_delay = BURST_RDTR;
3164                 if (adapter->rx_abs_int_delay == DEFAULT_RADV)
3165                         adapter->rx_abs_int_delay = BURST_RADV;
3166         }
3167
3168         /* set the Receive Delay Timer Register */
3169         ew32(RDTR, adapter->rx_int_delay);
3170
3171         /* irq moderation */
3172         ew32(RADV, adapter->rx_abs_int_delay);
3173         if ((adapter->itr_setting != 0) && (adapter->itr != 0))
3174                 e1000e_write_itr(adapter, adapter->itr);
3175
3176         ctrl_ext = er32(CTRL_EXT);
3177         /* Auto-Mask interrupts upon ICR access */
3178         ctrl_ext |= E1000_CTRL_EXT_IAME;
3179         ew32(IAM, 0xffffffff);
3180         ew32(CTRL_EXT, ctrl_ext);
3181         e1e_flush();
3182
3183         /* Setup the HW Rx Head and Tail Descriptor Pointers and
3184          * the Base and Length of the Rx Descriptor Ring
3185          */
3186         rdba = rx_ring->dma;
3187         ew32(RDBAL(0), (rdba & DMA_BIT_MASK(32)));
3188         ew32(RDBAH(0), (rdba >> 32));
3189         ew32(RDLEN(0), rdlen);
3190         ew32(RDH(0), 0);
3191         ew32(RDT(0), 0);
3192         rx_ring->head = adapter->hw.hw_addr + E1000_RDH(0);
3193         rx_ring->tail = adapter->hw.hw_addr + E1000_RDT(0);
3194
3195         /* Enable Receive Checksum Offload for TCP and UDP */
3196         rxcsum = er32(RXCSUM);
3197         if (adapter->netdev->features & NETIF_F_RXCSUM)
3198                 rxcsum |= E1000_RXCSUM_TUOFL;
3199         else
3200                 rxcsum &= ~E1000_RXCSUM_TUOFL;
3201         ew32(RXCSUM, rxcsum);
3202
3203         /* With jumbo frames, excessive C-state transition latencies result
3204          * in dropped transactions.
3205          */
3206         if (adapter->netdev->mtu > ETH_DATA_LEN) {
3207                 u32 lat =
3208                     ((er32(PBA) & E1000_PBA_RXA_MASK) * 1024 -
3209                      adapter->max_frame_size) * 8 / 1000;
3210
3211                 if (adapter->flags & FLAG_IS_ICH) {
3212                         u32 rxdctl = er32(RXDCTL(0));
3213                         ew32(RXDCTL(0), rxdctl | 0x3);
3214                 }
3215
3216                 pm_qos_update_request(&adapter->netdev->pm_qos_req, lat);
3217         } else {
3218                 pm_qos_update_request(&adapter->netdev->pm_qos_req,
3219                                       PM_QOS_DEFAULT_VALUE);
3220         }
3221
3222         /* Enable Receives */
3223         ew32(RCTL, rctl);
3224 }
3225
3226 /**
3227  * e1000e_write_mc_addr_list - write multicast addresses to MTA
3228  * @netdev: network interface device structure
3229  *
3230  * Writes multicast address list to the MTA hash table.
3231  * Returns: -ENOMEM on failure
3232  *                0 on no addresses written
3233  *                X on writing X addresses to MTA
3234  */
3235 static int e1000e_write_mc_addr_list(struct net_device *netdev)
3236 {
3237         struct e1000_adapter *adapter = netdev_priv(netdev);
3238         struct e1000_hw *hw = &adapter->hw;
3239         struct netdev_hw_addr *ha;
3240         u8 *mta_list;
3241         int i;
3242
3243         if (netdev_mc_empty(netdev)) {
3244                 /* nothing to program, so clear mc list */
3245                 hw->mac.ops.update_mc_addr_list(hw, NULL, 0);
3246                 return 0;
3247         }
3248
3249         mta_list = kzalloc(netdev_mc_count(netdev) * ETH_ALEN, GFP_ATOMIC);
3250         if (!mta_list)
3251                 return -ENOMEM;
3252
3253         /* update_mc_addr_list expects a packed array of only addresses. */
3254         i = 0;
3255         netdev_for_each_mc_addr(ha, netdev)
3256             memcpy(mta_list + (i++ * ETH_ALEN), ha->addr, ETH_ALEN);
3257
3258         hw->mac.ops.update_mc_addr_list(hw, mta_list, i);
3259         kfree(mta_list);
3260
3261         return netdev_mc_count(netdev);
3262 }
3263
3264 /**
3265  * e1000e_write_uc_addr_list - write unicast addresses to RAR table
3266  * @netdev: network interface device structure
3267  *
3268  * Writes unicast address list to the RAR table.
3269  * Returns: -ENOMEM on failure/insufficient address space
3270  *                0 on no addresses written
3271  *                X on writing X addresses to the RAR table
3272  **/
3273 static int e1000e_write_uc_addr_list(struct net_device *netdev)
3274 {
3275         struct e1000_adapter *adapter = netdev_priv(netdev);
3276         struct e1000_hw *hw = &adapter->hw;
3277         unsigned int rar_entries = hw->mac.rar_entry_count;
3278         int count = 0;
3279
3280         /* save a rar entry for our hardware address */
3281         rar_entries--;
3282
3283         /* save a rar entry for the LAA workaround */
3284         if (adapter->flags & FLAG_RESET_OVERWRITES_LAA)
3285                 rar_entries--;
3286
3287         /* return ENOMEM indicating insufficient memory for addresses */
3288         if (netdev_uc_count(netdev) > rar_entries)
3289                 return -ENOMEM;
3290
3291         if (!netdev_uc_empty(netdev) && rar_entries) {
3292                 struct netdev_hw_addr *ha;
3293
3294                 /* write the addresses in reverse order to avoid write
3295                  * combining
3296                  */
3297                 netdev_for_each_uc_addr(ha, netdev) {
3298                         if (!rar_entries)
3299                                 break;
3300                         hw->mac.ops.rar_set(hw, ha->addr, rar_entries--);
3301                         count++;
3302                 }
3303         }
3304
3305         /* zero out the remaining RAR entries not used above */
3306         for (; rar_entries > 0; rar_entries--) {
3307                 ew32(RAH(rar_entries), 0);
3308                 ew32(RAL(rar_entries), 0);
3309         }
3310         e1e_flush();
3311
3312         return count;
3313 }
3314
3315 /**
3316  * e1000e_set_rx_mode - secondary unicast, Multicast and Promiscuous mode set
3317  * @netdev: network interface device structure
3318  *
3319  * The ndo_set_rx_mode entry point is called whenever the unicast or multicast
3320  * address list or the network interface flags are updated.  This routine is
3321  * responsible for configuring the hardware for proper unicast, multicast,
3322  * promiscuous mode, and all-multi behavior.
3323  **/
3324 static void e1000e_set_rx_mode(struct net_device *netdev)
3325 {
3326         struct e1000_adapter *adapter = netdev_priv(netdev);
3327         struct e1000_hw *hw = &adapter->hw;
3328         u32 rctl;
3329
3330         /* Check for Promiscuous and All Multicast modes */
3331         rctl = er32(RCTL);
3332
3333         /* clear the affected bits */
3334         rctl &= ~(E1000_RCTL_UPE | E1000_RCTL_MPE);
3335
3336         if (netdev->flags & IFF_PROMISC) {
3337                 rctl |= (E1000_RCTL_UPE | E1000_RCTL_MPE);
3338                 /* Do not hardware filter VLANs in promisc mode */
3339                 e1000e_vlan_filter_disable(adapter);
3340         } else {
3341                 int count;
3342
3343                 if (netdev->flags & IFF_ALLMULTI) {
3344                         rctl |= E1000_RCTL_MPE;
3345                 } else {
3346                         /* Write addresses to the MTA, if the attempt fails
3347                          * then we should just turn on promiscuous mode so
3348                          * that we can at least receive multicast traffic
3349                          */
3350                         count = e1000e_write_mc_addr_list(netdev);
3351                         if (count < 0)
3352                                 rctl |= E1000_RCTL_MPE;
3353                 }
3354                 e1000e_vlan_filter_enable(adapter);
3355                 /* Write addresses to available RAR registers, if there is not
3356                  * sufficient space to store all the addresses then enable
3357                  * unicast promiscuous mode
3358                  */
3359                 count = e1000e_write_uc_addr_list(netdev);
3360                 if (count < 0)
3361                         rctl |= E1000_RCTL_UPE;
3362         }
3363
3364         ew32(RCTL, rctl);
3365
3366         if (netdev->features & NETIF_F_HW_VLAN_CTAG_RX)
3367                 e1000e_vlan_strip_enable(adapter);
3368         else
3369                 e1000e_vlan_strip_disable(adapter);
3370 }
3371
3372 static void e1000e_setup_rss_hash(struct e1000_adapter *adapter)
3373 {
3374         struct e1000_hw *hw = &adapter->hw;
3375         u32 mrqc, rxcsum;
3376         int i;
3377         static const u32 rsskey[10] = {
3378                 0xda565a6d, 0xc20e5b25, 0x3d256741, 0xb08fa343, 0xcb2bcad0,
3379                 0xb4307bae, 0xa32dcb77, 0x0cf23080, 0x3bb7426a, 0xfa01acbe
3380         };
3381
3382         /* Fill out hash function seed */
3383         for (i = 0; i < 10; i++)
3384                 ew32(RSSRK(i), rsskey[i]);
3385
3386         /* Direct all traffic to queue 0 */
3387         for (i = 0; i < 32; i++)
3388                 ew32(RETA(i), 0);
3389
3390         /* Disable raw packet checksumming so that RSS hash is placed in
3391          * descriptor on writeback.
3392          */
3393         rxcsum = er32(RXCSUM);
3394         rxcsum |= E1000_RXCSUM_PCSD;
3395
3396         ew32(RXCSUM, rxcsum);
3397
3398         mrqc = (E1000_MRQC_RSS_FIELD_IPV4 |
3399                 E1000_MRQC_RSS_FIELD_IPV4_TCP |
3400                 E1000_MRQC_RSS_FIELD_IPV6 |
3401                 E1000_MRQC_RSS_FIELD_IPV6_TCP |
3402                 E1000_MRQC_RSS_FIELD_IPV6_TCP_EX);
3403
3404         ew32(MRQC, mrqc);
3405 }
3406
3407 /**
3408  * e1000e_get_base_timinca - get default SYSTIM time increment attributes
3409  * @adapter: board private structure
3410  * @timinca: pointer to returned time increment attributes
3411  *
3412  * Get attributes for incrementing the System Time Register SYSTIML/H at
3413  * the default base frequency, and set the cyclecounter shift value.
3414  **/
3415 s32 e1000e_get_base_timinca(struct e1000_adapter *adapter, u32 *timinca)
3416 {
3417         struct e1000_hw *hw = &adapter->hw;
3418         u32 incvalue, incperiod, shift;
3419
3420         /* Make sure clock is enabled on I217 before checking the frequency */
3421         if ((hw->mac.type == e1000_pch_lpt) &&
3422             !(er32(TSYNCTXCTL) & E1000_TSYNCTXCTL_ENABLED) &&
3423             !(er32(TSYNCRXCTL) & E1000_TSYNCRXCTL_ENABLED)) {
3424                 u32 fextnvm7 = er32(FEXTNVM7);
3425
3426                 if (!(fextnvm7 & (1 << 0))) {
3427                         ew32(FEXTNVM7, fextnvm7 | (1 << 0));
3428                         e1e_flush();
3429                 }
3430         }
3431
3432         switch (hw->mac.type) {
3433         case e1000_pch2lan:
3434         case e1000_pch_lpt:
3435                 /* On I217, the clock frequency is 25MHz or 96MHz as
3436                  * indicated by the System Clock Frequency Indication
3437                  */
3438                 if ((hw->mac.type != e1000_pch_lpt) ||
3439                     (er32(TSYNCRXCTL) & E1000_TSYNCRXCTL_SYSCFI)) {
3440                         /* Stable 96MHz frequency */
3441                         incperiod = INCPERIOD_96MHz;
3442                         incvalue = INCVALUE_96MHz;
3443                         shift = INCVALUE_SHIFT_96MHz;
3444                         adapter->cc.shift = shift + INCPERIOD_SHIFT_96MHz;
3445                         break;
3446                 }
3447                 /* fall-through */
3448         case e1000_82574:
3449         case e1000_82583:
3450                 /* Stable 25MHz frequency */
3451                 incperiod = INCPERIOD_25MHz;
3452                 incvalue = INCVALUE_25MHz;
3453                 shift = INCVALUE_SHIFT_25MHz;
3454                 adapter->cc.shift = shift;
3455                 break;
3456         default:
3457                 return -EINVAL;
3458         }
3459
3460         *timinca = ((incperiod << E1000_TIMINCA_INCPERIOD_SHIFT) |
3461                     ((incvalue << shift) & E1000_TIMINCA_INCVALUE_MASK));
3462
3463         return 0;
3464 }
3465
3466 /**
3467  * e1000e_config_hwtstamp - configure the hwtstamp registers and enable/disable
3468  * @adapter: board private structure
3469  *
3470  * Outgoing time stamping can be enabled and disabled. Play nice and
3471  * disable it when requested, although it shouldn't cause any overhead
3472  * when no packet needs it. At most one packet in the queue may be
3473  * marked for time stamping, otherwise it would be impossible to tell
3474  * for sure to which packet the hardware time stamp belongs.
3475  *
3476  * Incoming time stamping has to be configured via the hardware filters.
3477  * Not all combinations are supported, in particular event type has to be
3478  * specified. Matching the kind of event packet is not supported, with the
3479  * exception of "all V2 events regardless of level 2 or 4".
3480  **/
3481 static int e1000e_config_hwtstamp(struct e1000_adapter *adapter,
3482                                   struct hwtstamp_config *config)
3483 {
3484         struct e1000_hw *hw = &adapter->hw;
3485         u32 tsync_tx_ctl = E1000_TSYNCTXCTL_ENABLED;
3486         u32 tsync_rx_ctl = E1000_TSYNCRXCTL_ENABLED;
3487         u32 rxmtrl = 0;
3488         u16 rxudp = 0;
3489         bool is_l4 = false;
3490         bool is_l2 = false;
3491         u32 regval;
3492         s32 ret_val;
3493
3494         if (!(adapter->flags & FLAG_HAS_HW_TIMESTAMP))
3495                 return -EINVAL;
3496
3497         /* flags reserved for future extensions - must be zero */
3498         if (config->flags)
3499                 return -EINVAL;
3500
3501         switch (config->tx_type) {
3502         case HWTSTAMP_TX_OFF:
3503                 tsync_tx_ctl = 0;
3504                 break;
3505         case HWTSTAMP_TX_ON:
3506                 break;
3507         default:
3508                 return -ERANGE;
3509         }
3510
3511         switch (config->rx_filter) {
3512         case HWTSTAMP_FILTER_NONE:
3513                 tsync_rx_ctl = 0;
3514                 break;
3515         case HWTSTAMP_FILTER_PTP_V1_L4_SYNC:
3516                 tsync_rx_ctl |= E1000_TSYNCRXCTL_TYPE_L4_V1;
3517                 rxmtrl = E1000_RXMTRL_PTP_V1_SYNC_MESSAGE;
3518                 is_l4 = true;
3519                 break;
3520         case HWTSTAMP_FILTER_PTP_V1_L4_DELAY_REQ:
3521                 tsync_rx_ctl |= E1000_TSYNCRXCTL_TYPE_L4_V1;
3522                 rxmtrl = E1000_RXMTRL_PTP_V1_DELAY_REQ_MESSAGE;
3523                 is_l4 = true;
3524                 break;
3525         case HWTSTAMP_FILTER_PTP_V2_L2_SYNC:
3526                 /* Also time stamps V2 L2 Path Delay Request/Response */
3527                 tsync_rx_ctl |= E1000_TSYNCRXCTL_TYPE_L2_V2;
3528                 rxmtrl = E1000_RXMTRL_PTP_V2_SYNC_MESSAGE;
3529                 is_l2 = true;
3530                 break;
3531         case HWTSTAMP_FILTER_PTP_V2_L2_DELAY_REQ:
3532                 /* Also time stamps V2 L2 Path Delay Request/Response. */
3533                 tsync_rx_ctl |= E1000_TSYNCRXCTL_TYPE_L2_V2;
3534                 rxmtrl = E1000_RXMTRL_PTP_V2_DELAY_REQ_MESSAGE;
3535                 is_l2 = true;
3536                 break;
3537         case HWTSTAMP_FILTER_PTP_V2_L4_SYNC:
3538                 /* Hardware cannot filter just V2 L4 Sync messages;
3539                  * fall-through to V2 (both L2 and L4) Sync.
3540                  */
3541         case HWTSTAMP_FILTER_PTP_V2_SYNC:
3542                 /* Also time stamps V2 Path Delay Request/Response. */
3543                 tsync_rx_ctl |= E1000_TSYNCRXCTL_TYPE_L2_L4_V2;
3544                 rxmtrl = E1000_RXMTRL_PTP_V2_SYNC_MESSAGE;
3545                 is_l2 = true;
3546                 is_l4 = true;
3547                 break;
3548         case HWTSTAMP_FILTER_PTP_V2_L4_DELAY_REQ:
3549                 /* Hardware cannot filter just V2 L4 Delay Request messages;
3550                  * fall-through to V2 (both L2 and L4) Delay Request.
3551                  */
3552         case HWTSTAMP_FILTER_PTP_V2_DELAY_REQ:
3553                 /* Also time stamps V2 Path Delay Request/Response. */
3554                 tsync_rx_ctl |= E1000_TSYNCRXCTL_TYPE_L2_L4_V2;
3555                 rxmtrl = E1000_RXMTRL_PTP_V2_DELAY_REQ_MESSAGE;
3556                 is_l2 = true;
3557                 is_l4 = true;
3558                 break;
3559         case HWTSTAMP_FILTER_PTP_V2_L4_EVENT:
3560         case HWTSTAMP_FILTER_PTP_V2_L2_EVENT:
3561                 /* Hardware cannot filter just V2 L4 or L2 Event messages;
3562                  * fall-through to all V2 (both L2 and L4) Events.
3563                  */
3564         case HWTSTAMP_FILTER_PTP_V2_EVENT:
3565                 tsync_rx_ctl |= E1000_TSYNCRXCTL_TYPE_EVENT_V2;
3566                 config->rx_filter = HWTSTAMP_FILTER_PTP_V2_EVENT;
3567                 is_l2 = true;
3568                 is_l4 = true;
3569                 break;
3570         case HWTSTAMP_FILTER_PTP_V1_L4_EVENT:
3571                 /* For V1, the hardware can only filter Sync messages or
3572                  * Delay Request messages but not both so fall-through to
3573                  * time stamp all packets.
3574                  */
3575         case HWTSTAMP_FILTER_ALL:
3576                 is_l2 = true;
3577                 is_l4 = true;
3578                 tsync_rx_ctl |= E1000_TSYNCRXCTL_TYPE_ALL;
3579                 config->rx_filter = HWTSTAMP_FILTER_ALL;
3580                 break;
3581         default:
3582                 return -ERANGE;
3583         }
3584
3585         adapter->hwtstamp_config = *config;
3586
3587         /* enable/disable Tx h/w time stamping */
3588         regval = er32(TSYNCTXCTL);
3589         regval &= ~E1000_TSYNCTXCTL_ENABLED;
3590         regval |= tsync_tx_ctl;
3591         ew32(TSYNCTXCTL, regval);
3592         if ((er32(TSYNCTXCTL) & E1000_TSYNCTXCTL_ENABLED) !=
3593             (regval & E1000_TSYNCTXCTL_ENABLED)) {
3594                 e_err("Timesync Tx Control register not set as expected\n");
3595                 return -EAGAIN;
3596         }
3597
3598         /* enable/disable Rx h/w time stamping */
3599         regval = er32(TSYNCRXCTL);
3600         regval &= ~(E1000_TSYNCRXCTL_ENABLED | E1000_TSYNCRXCTL_TYPE_MASK);
3601         regval |= tsync_rx_ctl;
3602         ew32(TSYNCRXCTL, regval);
3603         if ((er32(TSYNCRXCTL) & (E1000_TSYNCRXCTL_ENABLED |
3604                                  E1000_TSYNCRXCTL_TYPE_MASK)) !=
3605             (regval & (E1000_TSYNCRXCTL_ENABLED |
3606                        E1000_TSYNCRXCTL_TYPE_MASK))) {
3607                 e_err("Timesync Rx Control register not set as expected\n");
3608                 return -EAGAIN;
3609         }
3610
3611         /* L2: define ethertype filter for time stamped packets */
3612         if (is_l2)
3613                 rxmtrl |= ETH_P_1588;
3614
3615         /* define which PTP packets get time stamped */
3616         ew32(RXMTRL, rxmtrl);
3617
3618         /* Filter by destination port */
3619         if (is_l4) {
3620                 rxudp = PTP_EV_PORT;
3621                 cpu_to_be16s(&rxudp);
3622         }
3623         ew32(RXUDP, rxudp);
3624
3625         e1e_flush();
3626
3627         /* Clear TSYNCRXCTL_VALID & TSYNCTXCTL_VALID bit */
3628         er32(RXSTMPH);
3629         er32(TXSTMPH);
3630
3631         /* Get and set the System Time Register SYSTIM base frequency */
3632         ret_val = e1000e_get_base_timinca(adapter, &regval);
3633         if (ret_val)
3634                 return ret_val;
3635         ew32(TIMINCA, regval);
3636
3637         /* reset the ns time counter */
3638         timecounter_init(&adapter->tc, &adapter->cc,
3639                          ktime_to_ns(ktime_get_real()));
3640
3641         return 0;
3642 }
3643
3644 /**
3645  * e1000_configure - configure the hardware for Rx and Tx
3646  * @adapter: private board structure
3647  **/
3648 static void e1000_configure(struct e1000_adapter *adapter)
3649 {
3650         struct e1000_ring *rx_ring = adapter->rx_ring;
3651
3652         e1000e_set_rx_mode(adapter->netdev);
3653
3654         e1000_restore_vlan(adapter);
3655         e1000_init_manageability_pt(adapter);
3656
3657         e1000_configure_tx(adapter);
3658
3659         if (adapter->netdev->features & NETIF_F_RXHASH)
3660                 e1000e_setup_rss_hash(adapter);
3661         e1000_setup_rctl(adapter);
3662         e1000_configure_rx(adapter);
3663         adapter->alloc_rx_buf(rx_ring, e1000_desc_unused(rx_ring), GFP_KERNEL);
3664 }
3665
3666 /**
3667  * e1000e_power_up_phy - restore link in case the phy was powered down
3668  * @adapter: address of board private structure
3669  *
3670  * The phy may be powered down to save power and turn off link when the
3671  * driver is unloaded and wake on lan is not enabled (among others)
3672  * *** this routine MUST be followed by a call to e1000e_reset ***
3673  **/
3674 void e1000e_power_up_phy(struct e1000_adapter *adapter)
3675 {
3676         if (adapter->hw.phy.ops.power_up)
3677                 adapter->hw.phy.ops.power_up(&adapter->hw);
3678
3679         adapter->hw.mac.ops.setup_link(&adapter->hw);
3680 }
3681
3682 /**
3683  * e1000_power_down_phy - Power down the PHY
3684  *
3685  * Power down the PHY so no link is implied when interface is down.
3686  * The PHY cannot be powered down if management or WoL is active.
3687  */
3688 static void e1000_power_down_phy(struct e1000_adapter *adapter)
3689 {
3690         if (adapter->hw.phy.ops.power_down)
3691                 adapter->hw.phy.ops.power_down(&adapter->hw);
3692 }
3693
3694 /**
3695  * e1000e_reset - bring the hardware into a known good state
3696  *
3697  * This function boots the hardware and enables some settings that
3698  * require a configuration cycle of the hardware - those cannot be
3699  * set/changed during runtime. After reset the device needs to be
3700  * properly configured for Rx, Tx etc.
3701  */
3702 void e1000e_reset(struct e1000_adapter *adapter)
3703 {
3704         struct e1000_mac_info *mac = &adapter->hw.mac;
3705         struct e1000_fc_info *fc = &adapter->hw.fc;
3706         struct e1000_hw *hw = &adapter->hw;
3707         u32 tx_space, min_tx_space, min_rx_space;
3708         u32 pba = adapter->pba;
3709         u16 hwm;
3710
3711         /* reset Packet Buffer Allocation to default */
3712         ew32(PBA, pba);
3713
3714         if (adapter->max_frame_size > ETH_FRAME_LEN + ETH_FCS_LEN) {
3715                 /* To maintain wire speed transmits, the Tx FIFO should be
3716                  * large enough to accommodate two full transmit packets,
3717                  * rounded up to the next 1KB and expressed in KB.  Likewise,
3718                  * the Rx FIFO should be large enough to accommodate at least
3719                  * one full receive packet and is similarly rounded up and
3720                  * expressed in KB.
3721                  */
3722                 pba = er32(PBA);
3723                 /* upper 16 bits has Tx packet buffer allocation size in KB */
3724                 tx_space = pba >> 16;
3725                 /* lower 16 bits has Rx packet buffer allocation size in KB */
3726                 pba &= 0xffff;
3727                 /* the Tx fifo also stores 16 bytes of information about the Tx
3728                  * but don't include ethernet FCS because hardware appends it
3729                  */
3730                 min_tx_space = (adapter->max_frame_size +
3731                                 sizeof(struct e1000_tx_desc) - ETH_FCS_LEN) * 2;
3732                 min_tx_space = ALIGN(min_tx_space, 1024);
3733                 min_tx_space >>= 10;
3734                 /* software strips receive CRC, so leave room for it */
3735                 min_rx_space = adapter->max_frame_size;
3736                 min_rx_space = ALIGN(min_rx_space, 1024);
3737                 min_rx_space >>= 10;
3738
3739                 /* If current Tx allocation is less than the min Tx FIFO size,
3740                  * and the min Tx FIFO size is less than the current Rx FIFO
3741                  * allocation, take space away from current Rx allocation
3742                  */
3743                 if ((tx_space < min_tx_space) &&
3744                     ((min_tx_space - tx_space) < pba)) {
3745                         pba -= min_tx_space - tx_space;
3746
3747                         /* if short on Rx space, Rx wins and must trump Tx
3748                          * adjustment
3749                          */
3750                         if (pba < min_rx_space)
3751                                 pba = min_rx_space;
3752                 }
3753
3754                 ew32(PBA, pba);
3755         }
3756
3757         /* flow control settings
3758          *
3759          * The high water mark must be low enough to fit one full frame
3760          * (or the size used for early receive) above it in the Rx FIFO.
3761          * Set it to the lower of:
3762          * - 90% of the Rx FIFO size, and
3763          * - the full Rx FIFO size minus one full frame
3764          */
3765         if (adapter->flags & FLAG_DISABLE_FC_PAUSE_TIME)
3766                 fc->pause_time = 0xFFFF;
3767         else
3768                 fc->pause_time = E1000_FC_PAUSE_TIME;
3769         fc->send_xon = true;
3770         fc->current_mode = fc->requested_mode;
3771
3772         switch (hw->mac.type) {
3773         case e1000_ich9lan:
3774         case e1000_ich10lan:
3775                 if (adapter->netdev->mtu > ETH_DATA_LEN) {
3776                         pba = 14;
3777                         ew32(PBA, pba);
3778                         fc->high_water = 0x2800;
3779                         fc->low_water = fc->high_water - 8;
3780                         break;
3781                 }
3782                 /* fall-through */
3783         default:
3784                 hwm = min(((pba << 10) * 9 / 10),
3785                           ((pba << 10) - adapter->max_frame_size));
3786
3787                 fc->high_water = hwm & E1000_FCRTH_RTH; /* 8-byte granularity */
3788                 fc->low_water = fc->high_water - 8;
3789                 break;
3790         case e1000_pchlan:
3791                 /* Workaround PCH LOM adapter hangs with certain network
3792                  * loads.  If hangs persist, try disabling Tx flow control.
3793                  */
3794                 if (adapter->netdev->mtu > ETH_DATA_LEN) {
3795                         fc->high_water = 0x3500;
3796                         fc->low_water = 0x1500;
3797                 } else {
3798                         fc->high_water = 0x5000;
3799                         fc->low_water = 0x3000;
3800                 }
3801                 fc->refresh_time = 0x1000;
3802                 break;
3803         case e1000_pch2lan:
3804         case e1000_pch_lpt:
3805                 fc->refresh_time = 0x0400;
3806
3807                 if (adapter->netdev->mtu <= ETH_DATA_LEN) {
3808                         fc->high_water = 0x05C20;
3809                         fc->low_water = 0x05048;
3810                         fc->pause_time = 0x0650;
3811                         break;
3812                 }
3813
3814                 pba = 14;
3815                 ew32(PBA, pba);
3816                 fc->high_water = ((pba << 10) * 9 / 10) & E1000_FCRTH_RTH;
3817                 fc->low_water = ((pba << 10) * 8 / 10) & E1000_FCRTL_RTL;
3818                 break;
3819         }
3820
3821         /* Alignment of Tx data is on an arbitrary byte boundary with the
3822          * maximum size per Tx descriptor limited only to the transmit
3823          * allocation of the packet buffer minus 96 bytes with an upper
3824          * limit of 24KB due to receive synchronization limitations.
3825          */
3826         adapter->tx_fifo_limit = min_t(u32, ((er32(PBA) >> 16) << 10) - 96,
3827                                        24 << 10);
3828
3829         /* Disable Adaptive Interrupt Moderation if 2 full packets cannot
3830          * fit in receive buffer.
3831          */
3832         if (adapter->itr_setting & 0x3) {
3833                 if ((adapter->max_frame_size * 2) > (pba << 10)) {
3834                         if (!(adapter->flags2 & FLAG2_DISABLE_AIM)) {
3835                                 dev_info(&adapter->pdev->dev,
3836                                          "Interrupt Throttle Rate off\n");
3837                                 adapter->flags2 |= FLAG2_DISABLE_AIM;
3838                                 e1000e_write_itr(adapter, 0);
3839                         }
3840                 } else if (adapter->flags2 & FLAG2_DISABLE_AIM) {
3841                         dev_info(&adapter->pdev->dev,
3842                                  "Interrupt Throttle Rate on\n");
3843                         adapter->flags2 &= ~FLAG2_DISABLE_AIM;
3844                         adapter->itr = 20000;
3845                         e1000e_write_itr(adapter, adapter->itr);
3846                 }
3847         }
3848
3849         /* Allow time for pending master requests to run */
3850         mac->ops.reset_hw(hw);
3851
3852         /* For parts with AMT enabled, let the firmware know
3853          * that the network interface is in control
3854          */
3855         if (adapter->flags & FLAG_HAS_AMT)
3856                 e1000e_get_hw_control(adapter);
3857
3858         ew32(WUC, 0);
3859
3860         if (mac->ops.init_hw(hw))
3861                 e_err("Hardware Error\n");
3862
3863         e1000_update_mng_vlan(adapter);
3864
3865         /* Enable h/w to recognize an 802.1Q VLAN Ethernet packet */
3866         ew32(VET, ETH_P_8021Q);
3867
3868         e1000e_reset_adaptive(hw);
3869
3870         /* initialize systim and reset the ns time counter */
3871         e1000e_config_hwtstamp(adapter, &adapter->hwtstamp_config);
3872
3873         /* Set EEE advertisement as appropriate */
3874         if (adapter->flags2 & FLAG2_HAS_EEE) {
3875                 s32 ret_val;
3876                 u16 adv_addr;
3877
3878                 switch (hw->phy.type) {
3879                 case e1000_phy_82579:
3880                         adv_addr = I82579_EEE_ADVERTISEMENT;
3881                         break;
3882                 case e1000_phy_i217:
3883                         adv_addr = I217_EEE_ADVERTISEMENT;
3884                         break;
3885                 default:
3886                         dev_err(&adapter->pdev->dev,
3887                                 "Invalid PHY type setting EEE advertisement\n");
3888                         return;
3889                 }
3890
3891                 ret_val = hw->phy.ops.acquire(hw);
3892                 if (ret_val) {
3893                         dev_err(&adapter->pdev->dev,
3894                                 "EEE advertisement - unable to acquire PHY\n");
3895                         return;
3896                 }
3897
3898                 e1000_write_emi_reg_locked(hw, adv_addr,
3899                                            hw->dev_spec.ich8lan.eee_disable ?
3900                                            0 : adapter->eee_advert);
3901
3902                 hw->phy.ops.release(hw);
3903         }
3904
3905         if (!netif_running(adapter->netdev) &&
3906             !test_bit(__E1000_TESTING, &adapter->state))
3907                 e1000_power_down_phy(adapter);
3908
3909         e1000_get_phy_info(hw);
3910
3911         if ((adapter->flags & FLAG_HAS_SMART_POWER_DOWN) &&
3912             !(adapter->flags & FLAG_SMART_POWER_DOWN)) {
3913                 u16 phy_data = 0;
3914                 /* speed up time to link by disabling smart power down, ignore
3915                  * the return value of this function because there is nothing
3916                  * different we would do if it failed
3917                  */
3918                 e1e_rphy(hw, IGP02E1000_PHY_POWER_MGMT, &phy_data);
3919                 phy_data &= ~IGP02E1000_PM_SPD;
3920                 e1e_wphy(hw, IGP02E1000_PHY_POWER_MGMT, phy_data);
3921         }
3922 }
3923
3924 int e1000e_up(struct e1000_adapter *adapter)
3925 {
3926         struct e1000_hw *hw = &adapter->hw;
3927
3928         /* hardware has been reset, we need to reload some things */
3929         e1000_configure(adapter);
3930
3931         clear_bit(__E1000_DOWN, &adapter->state);
3932
3933         if (adapter->msix_entries)
3934                 e1000_configure_msix(adapter);
3935         e1000_irq_enable(adapter);
3936
3937         netif_start_queue(adapter->netdev);
3938
3939         /* fire a link change interrupt to start the watchdog */
3940         if (adapter->msix_entries)
3941                 ew32(ICS, E1000_ICS_LSC | E1000_ICR_OTHER);
3942         else
3943                 ew32(ICS, E1000_ICS_LSC);
3944
3945         return 0;
3946 }
3947
3948 static void e1000e_flush_descriptors(struct e1000_adapter *adapter)
3949 {
3950         struct e1000_hw *hw = &adapter->hw;
3951
3952         if (!(adapter->flags2 & FLAG2_DMA_BURST))
3953                 return;
3954
3955         /* flush pending descriptor writebacks to memory */
3956         ew32(TIDV, adapter->tx_int_delay | E1000_TIDV_FPD);
3957         ew32(RDTR, adapter->rx_int_delay | E1000_RDTR_FPD);
3958
3959         /* execute the writes immediately */
3960         e1e_flush();
3961
3962         /* due to rare timing issues, write to TIDV/RDTR again to ensure the
3963          * write is successful
3964          */
3965         ew32(TIDV, adapter->tx_int_delay | E1000_TIDV_FPD);
3966         ew32(RDTR, adapter->rx_int_delay | E1000_RDTR_FPD);
3967
3968         /* execute the writes immediately */
3969         e1e_flush();
3970 }
3971
3972 static void e1000e_update_stats(struct e1000_adapter *adapter);
3973
3974 /**
3975  * e1000e_down - quiesce the device and optionally reset the hardware
3976  * @adapter: board private structure
3977  * @reset: boolean flag to reset the hardware or not
3978  */
3979 void e1000e_down(struct e1000_adapter *adapter, bool reset)
3980 {
3981         struct net_device *netdev = adapter->netdev;
3982         struct e1000_hw *hw = &adapter->hw;
3983         u32 tctl, rctl;
3984
3985         /* signal that we're down so the interrupt handler does not
3986          * reschedule our watchdog timer
3987          */
3988         set_bit(__E1000_DOWN, &adapter->state);
3989
3990         /* disable receives in the hardware */
3991         rctl = er32(RCTL);
3992         if (!(adapter->flags2 & FLAG2_NO_DISABLE_RX))
3993                 ew32(RCTL, rctl & ~E1000_RCTL_EN);
3994         /* flush and sleep below */
3995
3996         netif_stop_queue(netdev);
3997
3998         /* disable transmits in the hardware */
3999         tctl = er32(TCTL);
4000         tctl &= ~E1000_TCTL_EN;
4001         ew32(TCTL, tctl);
4002
4003         /* flush both disables and wait for them to finish */
4004         e1e_flush();
4005         usleep_range(10000, 20000);
4006
4007         e1000_irq_disable(adapter);
4008
4009         napi_synchronize(&adapter->napi);
4010
4011         del_timer_sync(&adapter->watchdog_timer);
4012         del_timer_sync(&adapter->phy_info_timer);
4013
4014         netif_carrier_off(netdev);
4015
4016         spin_lock(&adapter->stats64_lock);
4017         e1000e_update_stats(adapter);
4018         spin_unlock(&adapter->stats64_lock);
4019
4020         e1000e_flush_descriptors(adapter);
4021         e1000_clean_tx_ring(adapter->tx_ring);
4022         e1000_clean_rx_ring(adapter->rx_ring);
4023
4024         adapter->link_speed = 0;
4025         adapter->link_duplex = 0;
4026
4027         /* Disable Si errata workaround on PCHx for jumbo frame flow */
4028         if ((hw->mac.type >= e1000_pch2lan) &&
4029             (adapter->netdev->mtu > ETH_DATA_LEN) &&
4030             e1000_lv_jumbo_workaround_ich8lan(hw, false))
4031                 e_dbg("failed to disable jumbo frame workaround mode\n");
4032
4033         if (reset && !pci_channel_offline(adapter->pdev))
4034                 e1000e_reset(adapter);
4035 }
4036
4037 void e1000e_reinit_locked(struct e1000_adapter *adapter)
4038 {
4039         might_sleep();
4040         while (test_and_set_bit(__E1000_RESETTING, &adapter->state))
4041                 usleep_range(1000, 2000);
4042         e1000e_down(adapter, true);
4043         e1000e_up(adapter);
4044         clear_bit(__E1000_RESETTING, &adapter->state);
4045 }
4046
4047 /**
4048  * e1000e_cyclecounter_read - read raw cycle counter (used by time counter)
4049  * @cc: cyclecounter structure
4050  **/
4051 static cycle_t e1000e_cyclecounter_read(const struct cyclecounter *cc)
4052 {
4053         struct e1000_adapter *adapter = container_of(cc, struct e1000_adapter,
4054                                                      cc);
4055         struct e1000_hw *hw = &adapter->hw;
4056         cycle_t systim;
4057
4058         /* latch SYSTIMH on read of SYSTIML */
4059         systim = (cycle_t)er32(SYSTIML);
4060         systim |= (cycle_t)er32(SYSTIMH) << 32;
4061
4062         return systim;
4063 }
4064
4065 /**
4066  * e1000_sw_init - Initialize general software structures (struct e1000_adapter)
4067  * @adapter: board private structure to initialize
4068  *
4069  * e1000_sw_init initializes the Adapter private data structure.
4070  * Fields are initialized based on PCI device information and
4071  * OS network device settings (MTU size).
4072  **/
4073 static int e1000_sw_init(struct e1000_adapter *adapter)
4074 {
4075         struct net_device *netdev = adapter->netdev;
4076
4077         adapter->rx_buffer_len = ETH_FRAME_LEN + VLAN_HLEN + ETH_FCS_LEN;
4078         adapter->rx_ps_bsize0 = 128;
4079         adapter->max_frame_size = netdev->mtu + ETH_HLEN + ETH_FCS_LEN;
4080         adapter->min_frame_size = ETH_ZLEN + ETH_FCS_LEN;
4081         adapter->tx_ring_count = E1000_DEFAULT_TXD;
4082         adapter->rx_ring_count = E1000_DEFAULT_RXD;
4083
4084         spin_lock_init(&adapter->stats64_lock);
4085
4086         e1000e_set_interrupt_capability(adapter);
4087
4088         if (e1000_alloc_queues(adapter))
4089                 return -ENOMEM;
4090
4091         /* Setup hardware time stamping cyclecounter */
4092         if (adapter->flags & FLAG_HAS_HW_TIMESTAMP) {
4093                 adapter->cc.read = e1000e_cyclecounter_read;
4094                 adapter->cc.mask = CLOCKSOURCE_MASK(64);
4095                 adapter->cc.mult = 1;
4096                 /* cc.shift set in e1000e_get_base_tininca() */
4097
4098                 spin_lock_init(&adapter->systim_lock);
4099                 INIT_WORK(&adapter->tx_hwtstamp_work, e1000e_tx_hwtstamp_work);
4100         }
4101
4102         /* Explicitly disable IRQ since the NIC can be in any state. */
4103         e1000_irq_disable(adapter);
4104
4105         set_bit(__E1000_DOWN, &adapter->state);
4106         return 0;
4107 }
4108
4109 /**
4110  * e1000_intr_msi_test - Interrupt Handler
4111  * @irq: interrupt number
4112  * @data: pointer to a network interface device structure
4113  **/
4114 static irqreturn_t e1000_intr_msi_test(int __always_unused irq, void *data)
4115 {
4116         struct net_device *netdev = data;
4117         struct e1000_adapter *adapter = netdev_priv(netdev);
4118         struct e1000_hw *hw = &adapter->hw;
4119         u32 icr = er32(ICR);
4120
4121         e_dbg("icr is %08X\n", icr);
4122         if (icr & E1000_ICR_RXSEQ) {
4123                 adapter->flags &= ~FLAG_MSI_TEST_FAILED;
4124                 /* Force memory writes to complete before acknowledging the
4125                  * interrupt is handled.
4126                  */
4127                 wmb();
4128         }
4129
4130         return IRQ_HANDLED;
4131 }
4132
4133 /**
4134  * e1000_test_msi_interrupt - Returns 0 for successful test
4135  * @adapter: board private struct
4136  *
4137  * code flow taken from tg3.c
4138  **/
4139 static int e1000_test_msi_interrupt(struct e1000_adapter *adapter)
4140 {
4141         struct net_device *netdev = adapter->netdev;
4142         struct e1000_hw *hw = &adapter->hw;
4143         int err;
4144
4145         /* poll_enable hasn't been called yet, so don't need disable */
4146         /* clear any pending events */
4147         er32(ICR);
4148
4149         /* free the real vector and request a test handler */
4150         e1000_free_irq(adapter);
4151         e1000e_reset_interrupt_capability(adapter);
4152
4153         /* Assume that the test fails, if it succeeds then the test
4154          * MSI irq handler will unset this flag
4155          */
4156         adapter->flags |= FLAG_MSI_TEST_FAILED;
4157
4158         err = pci_enable_msi(adapter->pdev);
4159         if (err)
4160                 goto msi_test_failed;
4161
4162         err = request_irq(adapter->pdev->irq, e1000_intr_msi_test, 0,
4163                           netdev->name, netdev);
4164         if (err) {
4165                 pci_disable_msi(adapter->pdev);
4166                 goto msi_test_failed;
4167         }
4168
4169         /* Force memory writes to complete before enabling and firing an
4170          * interrupt.
4171          */
4172         wmb();
4173
4174         e1000_irq_enable(adapter);
4175
4176         /* fire an unusual interrupt on the test handler */
4177         ew32(ICS, E1000_ICS_RXSEQ);
4178         e1e_flush();
4179         msleep(100);
4180
4181         e1000_irq_disable(adapter);
4182
4183         rmb();                  /* read flags after interrupt has been fired */
4184
4185         if (adapter->flags & FLAG_MSI_TEST_FAILED) {
4186                 adapter->int_mode = E1000E_INT_MODE_LEGACY;
4187                 e_info("MSI interrupt test failed, using legacy interrupt.\n");
4188         } else {
4189                 e_dbg("MSI interrupt test succeeded!\n");
4190         }
4191
4192         free_irq(adapter->pdev->irq, netdev);
4193         pci_disable_msi(adapter->pdev);
4194
4195 msi_test_failed:
4196         e1000e_set_interrupt_capability(adapter);
4197         return e1000_request_irq(adapter);
4198 }
4199
4200 /**
4201  * e1000_test_msi - Returns 0 if MSI test succeeds or INTx mode is restored
4202  * @adapter: board private struct
4203  *
4204  * code flow taken from tg3.c, called with e1000 interrupts disabled.
4205  **/
4206 static int e1000_test_msi(struct e1000_adapter *adapter)
4207 {
4208         int err;
4209         u16 pci_cmd;
4210
4211         if (!(adapter->flags & FLAG_MSI_ENABLED))
4212                 return 0;
4213
4214         /* disable SERR in case the MSI write causes a master abort */
4215         pci_read_config_word(adapter->pdev, PCI_COMMAND, &pci_cmd);
4216         if (pci_cmd & PCI_COMMAND_SERR)
4217                 pci_write_config_word(adapter->pdev, PCI_COMMAND,
4218                                       pci_cmd & ~PCI_COMMAND_SERR);
4219
4220         err = e1000_test_msi_interrupt(adapter);
4221
4222         /* re-enable SERR */
4223         if (pci_cmd & PCI_COMMAND_SERR) {
4224                 pci_read_config_word(adapter->pdev, PCI_COMMAND, &pci_cmd);
4225                 pci_cmd |= PCI_COMMAND_SERR;
4226                 pci_write_config_word(adapter->pdev, PCI_COMMAND, pci_cmd);
4227         }
4228
4229         return err;
4230 }
4231
4232 /**
4233  * e1000_open - Called when a network interface is made active
4234  * @netdev: network interface device structure
4235  *
4236  * Returns 0 on success, negative value on failure
4237  *
4238  * The open entry point is called when a network interface is made
4239  * active by the system (IFF_UP).  At this point all resources needed
4240  * for transmit and receive operations are allocated, the interrupt
4241  * handler is registered with the OS, the watchdog timer is started,
4242  * and the stack is notified that the interface is ready.
4243  **/
4244 static int e1000_open(struct net_device *netdev)
4245 {
4246         struct e1000_adapter *adapter = netdev_priv(netdev);
4247         struct e1000_hw *hw = &adapter->hw;
4248         struct pci_dev *pdev = adapter->pdev;
4249         int err;
4250
4251         /* disallow open during test */
4252         if (test_bit(__E1000_TESTING, &adapter->state))
4253                 return -EBUSY;
4254
4255         pm_runtime_get_sync(&pdev->dev);
4256
4257         netif_carrier_off(netdev);
4258
4259         /* allocate transmit descriptors */
4260         err = e1000e_setup_tx_resources(adapter->tx_ring);
4261         if (err)
4262                 goto err_setup_tx;
4263
4264         /* allocate receive descriptors */
4265         err = e1000e_setup_rx_resources(adapter->rx_ring);
4266         if (err)
4267                 goto err_setup_rx;
4268
4269         /* If AMT is enabled, let the firmware know that the network
4270          * interface is now open and reset the part to a known state.
4271          */
4272         if (adapter->flags & FLAG_HAS_AMT) {
4273                 e1000e_get_hw_control(adapter);
4274                 e1000e_reset(adapter);
4275         }
4276
4277         e1000e_power_up_phy(adapter);
4278
4279         adapter->mng_vlan_id = E1000_MNG_VLAN_NONE;
4280         if ((adapter->hw.mng_cookie.status & E1000_MNG_DHCP_COOKIE_STATUS_VLAN))
4281                 e1000_update_mng_vlan(adapter);
4282
4283         /* DMA latency requirement to workaround jumbo issue */
4284         pm_qos_add_request(&adapter->netdev->pm_qos_req, PM_QOS_CPU_DMA_LATENCY,
4285                            PM_QOS_DEFAULT_VALUE);
4286
4287         /* before we allocate an interrupt, we must be ready to handle it.
4288          * Setting DEBUG_SHIRQ in the kernel makes it fire an interrupt
4289          * as soon as we call pci_request_irq, so we have to setup our
4290          * clean_rx handler before we do so.
4291          */
4292         e1000_configure(adapter);
4293
4294         err = e1000_request_irq(adapter);
4295         if (err)
4296                 goto err_req_irq;
4297
4298         /* Work around PCIe errata with MSI interrupts causing some chipsets to
4299          * ignore e1000e MSI messages, which means we need to test our MSI
4300          * interrupt now
4301          */
4302         if (adapter->int_mode != E1000E_INT_MODE_LEGACY) {
4303                 err = e1000_test_msi(adapter);
4304                 if (err) {
4305                         e_err("Interrupt allocation failed\n");
4306                         goto err_req_irq;
4307                 }
4308         }
4309
4310         /* From here on the code is the same as e1000e_up() */
4311         clear_bit(__E1000_DOWN, &adapter->state);
4312
4313         napi_enable(&adapter->napi);
4314
4315         e1000_irq_enable(adapter);
4316
4317         adapter->tx_hang_recheck = false;
4318         netif_start_queue(netdev);
4319
4320         adapter->idle_check = true;
4321         hw->mac.get_link_status = true;
4322         pm_runtime_put(&pdev->dev);
4323
4324         /* fire a link status change interrupt to start the watchdog */
4325         if (adapter->msix_entries)
4326                 ew32(ICS, E1000_ICS_LSC | E1000_ICR_OTHER);
4327         else
4328                 ew32(ICS, E1000_ICS_LSC);
4329
4330         return 0;
4331
4332 err_req_irq:
4333         e1000e_release_hw_control(adapter);
4334         e1000_power_down_phy(adapter);
4335         e1000e_free_rx_resources(adapter->rx_ring);
4336 err_setup_rx:
4337         e1000e_free_tx_resources(adapter->tx_ring);
4338 err_setup_tx:
4339         e1000e_reset(adapter);
4340         pm_runtime_put_sync(&pdev->dev);
4341
4342         return err;
4343 }
4344
4345 /**
4346  * e1000_close - Disables a network interface
4347  * @netdev: network interface device structure
4348  *
4349  * Returns 0, this is not allowed to fail
4350  *
4351  * The close entry point is called when an interface is de-activated
4352  * by the OS.  The hardware is still under the drivers control, but
4353  * needs to be disabled.  A global MAC reset is issued to stop the
4354  * hardware, and all transmit and receive resources are freed.
4355  **/
4356 static int e1000_close(struct net_device *netdev)
4357 {
4358         struct e1000_adapter *adapter = netdev_priv(netdev);
4359         struct pci_dev *pdev = adapter->pdev;
4360         int count = E1000_CHECK_RESET_COUNT;
4361
4362         while (test_bit(__E1000_RESETTING, &adapter->state) && count--)
4363                 usleep_range(10000, 20000);
4364
4365         WARN_ON(test_bit(__E1000_RESETTING, &adapter->state));
4366
4367         pm_runtime_get_sync(&pdev->dev);
4368
4369         if (!test_bit(__E1000_DOWN, &adapter->state)) {
4370                 e1000e_down(adapter, true);
4371                 e1000_free_irq(adapter);
4372         }
4373
4374         napi_disable(&adapter->napi);
4375
4376         e1000e_free_tx_resources(adapter->tx_ring);
4377         e1000e_free_rx_resources(adapter->rx_ring);
4378
4379         /* kill manageability vlan ID if supported, but not if a vlan with
4380          * the same ID is registered on the host OS (let 8021q kill it)
4381          */
4382         if (adapter->hw.mng_cookie.status & E1000_MNG_DHCP_COOKIE_STATUS_VLAN)
4383                 e1000_vlan_rx_kill_vid(netdev, htons(ETH_P_8021Q),
4384                                        adapter->mng_vlan_id);
4385
4386         /* If AMT is enabled, let the firmware know that the network
4387          * interface is now closed
4388          */
4389         if ((adapter->flags & FLAG_HAS_AMT) &&
4390             !test_bit(__E1000_TESTING, &adapter->state))
4391                 e1000e_release_hw_control(adapter);
4392
4393         pm_qos_remove_request(&adapter->netdev->pm_qos_req);
4394
4395         pm_runtime_put_sync(&pdev->dev);
4396
4397         return 0;
4398 }
4399
4400 /**
4401  * e1000_set_mac - Change the Ethernet Address of the NIC
4402  * @netdev: network interface device structure
4403  * @p: pointer to an address structure
4404  *
4405  * Returns 0 on success, negative on failure
4406  **/
4407 static int e1000_set_mac(struct net_device *netdev, void *p)
4408 {
4409         struct e1000_adapter *adapter = netdev_priv(netdev);
4410         struct e1000_hw *hw = &adapter->hw;
4411         struct sockaddr *addr = p;
4412
4413         if (!is_valid_ether_addr(addr->sa_data))
4414                 return -EADDRNOTAVAIL;
4415
4416         memcpy(netdev->dev_addr, addr->sa_data, netdev->addr_len);
4417         memcpy(adapter->hw.mac.addr, addr->sa_data, netdev->addr_len);
4418
4419         hw->mac.ops.rar_set(&adapter->hw, adapter->hw.mac.addr, 0);
4420
4421         if (adapter->flags & FLAG_RESET_OVERWRITES_LAA) {
4422                 /* activate the work around */
4423                 e1000e_set_laa_state_82571(&adapter->hw, 1);
4424
4425                 /* Hold a copy of the LAA in RAR[14] This is done so that
4426                  * between the time RAR[0] gets clobbered  and the time it
4427                  * gets fixed (in e1000_watchdog), the actual LAA is in one
4428                  * of the RARs and no incoming packets directed to this port
4429                  * are dropped. Eventually the LAA will be in RAR[0] and
4430                  * RAR[14]
4431                  */
4432                 hw->mac.ops.rar_set(&adapter->hw, adapter->hw.mac.addr,
4433                                     adapter->hw.mac.rar_entry_count - 1);
4434         }
4435
4436         return 0;
4437 }
4438
4439 /**
4440  * e1000e_update_phy_task - work thread to update phy
4441  * @work: pointer to our work struct
4442  *
4443  * this worker thread exists because we must acquire a
4444  * semaphore to read the phy, which we could msleep while
4445  * waiting for it, and we can't msleep in a timer.
4446  **/
4447 static void e1000e_update_phy_task(struct work_struct *work)
4448 {
4449         struct e1000_adapter *adapter = container_of(work,
4450                                                      struct e1000_adapter,
4451                                                      update_phy_task);
4452         struct e1000_hw *hw = &adapter->hw;
4453
4454         if (test_bit(__E1000_DOWN, &adapter->state))
4455                 return;
4456
4457         e1000_get_phy_info(hw);
4458
4459         /* Enable EEE on 82579 after link up */
4460         if (hw->phy.type == e1000_phy_82579)
4461                 e1000_set_eee_pchlan(hw);
4462 }
4463
4464 /**
4465  * e1000_update_phy_info - timre call-back to update PHY info
4466  * @data: pointer to adapter cast into an unsigned long
4467  *
4468  * Need to wait a few seconds after link up to get diagnostic information from
4469  * the phy
4470  **/
4471 static void e1000_update_phy_info(unsigned long data)
4472 {
4473         struct e1000_adapter *adapter = (struct e1000_adapter *)data;
4474
4475         if (test_bit(__E1000_DOWN, &adapter->state))
4476                 return;
4477
4478         schedule_work(&adapter->update_phy_task);
4479 }
4480
4481 /**
4482  * e1000e_update_phy_stats - Update the PHY statistics counters
4483  * @adapter: board private structure
4484  *
4485  * Read/clear the upper 16-bit PHY registers and read/accumulate lower
4486  **/
4487 static void e1000e_update_phy_stats(struct e1000_adapter *adapter)
4488 {
4489         struct e1000_hw *hw = &adapter->hw;
4490         s32 ret_val;
4491         u16 phy_data;
4492
4493         ret_val = hw->phy.ops.acquire(hw);
4494         if (ret_val)
4495                 return;
4496
4497         /* A page set is expensive so check if already on desired page.
4498          * If not, set to the page with the PHY status registers.
4499          */
4500         hw->phy.addr = 1;
4501         ret_val = e1000e_read_phy_reg_mdic(hw, IGP01E1000_PHY_PAGE_SELECT,
4502                                            &phy_data);
4503         if (ret_val)
4504                 goto release;
4505         if (phy_data != (HV_STATS_PAGE << IGP_PAGE_SHIFT)) {
4506                 ret_val = hw->phy.ops.set_page(hw,
4507                                                HV_STATS_PAGE << IGP_PAGE_SHIFT);
4508                 if (ret_val)
4509                         goto release;
4510         }
4511
4512         /* Single Collision Count */
4513         hw->phy.ops.read_reg_page(hw, HV_SCC_UPPER, &phy_data);
4514         ret_val = hw->phy.ops.read_reg_page(hw, HV_SCC_LOWER, &phy_data);
4515         if (!ret_val)
4516                 adapter->stats.scc += phy_data;
4517
4518         /* Excessive Collision Count */
4519         hw->phy.ops.read_reg_page(hw, HV_ECOL_UPPER, &phy_data);
4520         ret_val = hw->phy.ops.read_reg_page(hw, HV_ECOL_LOWER, &phy_data);
4521         if (!ret_val)
4522                 adapter->stats.ecol += phy_data;
4523
4524         /* Multiple Collision Count */
4525         hw->phy.ops.read_reg_page(hw, HV_MCC_UPPER, &phy_data);
4526         ret_val = hw->phy.ops.read_reg_page(hw, HV_MCC_LOWER, &phy_data);
4527         if (!ret_val)
4528                 adapter->stats.mcc += phy_data;
4529
4530         /* Late Collision Count */
4531         hw->phy.ops.read_reg_page(hw, HV_LATECOL_UPPER, &phy_data);
4532         ret_val = hw->phy.ops.read_reg_page(hw, HV_LATECOL_LOWER, &phy_data);
4533         if (!ret_val)
4534                 adapter->stats.latecol += phy_data;
4535
4536         /* Collision Count - also used for adaptive IFS */
4537         hw->phy.ops.read_reg_page(hw, HV_COLC_UPPER, &phy_data);
4538         ret_val = hw->phy.ops.read_reg_page(hw, HV_COLC_LOWER, &phy_data);
4539         if (!ret_val)
4540                 hw->mac.collision_delta = phy_data;
4541
4542         /* Defer Count */
4543         hw->phy.ops.read_reg_page(hw, HV_DC_UPPER, &phy_data);
4544         ret_val = hw->phy.ops.read_reg_page(hw, HV_DC_LOWER, &phy_data);
4545         if (!ret_val)
4546                 adapter->stats.dc += phy_data;
4547
4548         /* Transmit with no CRS */
4549         hw->phy.ops.read_reg_page(hw, HV_TNCRS_UPPER, &phy_data);
4550         ret_val = hw->phy.ops.read_reg_page(hw, HV_TNCRS_LOWER, &phy_data);
4551         if (!ret_val)
4552                 adapter->stats.tncrs += phy_data;
4553
4554 release:
4555         hw->phy.ops.release(hw);
4556 }
4557
4558 /**
4559  * e1000e_update_stats - Update the board statistics counters
4560  * @adapter: board private structure
4561  **/
4562 static void e1000e_update_stats(struct e1000_adapter *adapter)
4563 {
4564         struct net_device *netdev = adapter->netdev;
4565         struct e1000_hw *hw = &adapter->hw;
4566         struct pci_dev *pdev = adapter->pdev;
4567
4568         /* Prevent stats update while adapter is being reset, or if the pci
4569          * connection is down.
4570          */
4571         if (adapter->link_speed == 0)
4572                 return;
4573         if (pci_channel_offline(pdev))
4574                 return;
4575
4576         adapter->stats.crcerrs += er32(CRCERRS);
4577         adapter->stats.gprc += er32(GPRC);
4578         adapter->stats.gorc += er32(GORCL);
4579         er32(GORCH);            /* Clear gorc */
4580         adapter->stats.bprc += er32(BPRC);
4581         adapter->stats.mprc += er32(MPRC);
4582         adapter->stats.roc += er32(ROC);
4583
4584         adapter->stats.mpc += er32(MPC);
4585
4586         /* Half-duplex statistics */
4587         if (adapter->link_duplex == HALF_DUPLEX) {
4588                 if (adapter->flags2 & FLAG2_HAS_PHY_STATS) {
4589                         e1000e_update_phy_stats(adapter);
4590                 } else {
4591                         adapter->stats.scc += er32(SCC);
4592                         adapter->stats.ecol += er32(ECOL);
4593                         adapter->stats.mcc += er32(MCC);
4594                         adapter->stats.latecol += er32(LATECOL);
4595                         adapter->stats.dc += er32(DC);
4596
4597                         hw->mac.collision_delta = er32(COLC);
4598
4599                         if ((hw->mac.type != e1000_82574) &&
4600                             (hw->mac.type != e1000_82583))
4601                                 adapter->stats.tncrs += er32(TNCRS);
4602                 }
4603                 adapter->stats.colc += hw->mac.collision_delta;
4604         }
4605
4606         adapter->stats.xonrxc += er32(XONRXC);
4607         adapter->stats.xontxc += er32(XONTXC);
4608         adapter->stats.xoffrxc += er32(XOFFRXC);
4609         adapter->stats.xofftxc += er32(XOFFTXC);
4610         adapter->stats.gptc += er32(GPTC);
4611         adapter->stats.gotc += er32(GOTCL);
4612         er32(GOTCH);            /* Clear gotc */
4613         adapter->stats.rnbc += er32(RNBC);
4614         adapter->stats.ruc += er32(RUC);
4615
4616         adapter->stats.mptc += er32(MPTC);
4617         adapter->stats.bptc += er32(BPTC);
4618
4619         /* used for adaptive IFS */
4620
4621         hw->mac.tx_packet_delta = er32(TPT);
4622         adapter->stats.tpt += hw->mac.tx_packet_delta;
4623
4624         adapter->stats.algnerrc += er32(ALGNERRC);
4625         adapter->stats.rxerrc += er32(RXERRC);
4626         adapter->stats.cexterr += er32(CEXTERR);
4627         adapter->stats.tsctc += er32(TSCTC);
4628         adapter->stats.tsctfc += er32(TSCTFC);
4629
4630         /* Fill out the OS statistics structure */
4631         netdev->stats.multicast = adapter->stats.mprc;
4632         netdev->stats.collisions = adapter->stats.colc;
4633
4634         /* Rx Errors */
4635
4636         /* RLEC on some newer hardware can be incorrect so build
4637          * our own version based on RUC and ROC
4638          */
4639         netdev->stats.rx_errors = adapter->stats.rxerrc +
4640             adapter->stats.crcerrs + adapter->stats.algnerrc +
4641             adapter->stats.ruc + adapter->stats.roc + adapter->stats.cexterr;
4642         netdev->stats.rx_length_errors = adapter->stats.ruc +
4643             adapter->stats.roc;
4644         netdev->stats.rx_crc_errors = adapter->stats.crcerrs;
4645         netdev->stats.rx_frame_errors = adapter->stats.algnerrc;
4646         netdev->stats.rx_missed_errors = adapter->stats.mpc;
4647
4648         /* Tx Errors */
4649         netdev->stats.tx_errors = adapter->stats.ecol + adapter->stats.latecol;
4650         netdev->stats.tx_aborted_errors = adapter->stats.ecol;
4651         netdev->stats.tx_window_errors = adapter->stats.latecol;
4652         netdev->stats.tx_carrier_errors = adapter->stats.tncrs;
4653
4654         /* Tx Dropped needs to be maintained elsewhere */
4655
4656         /* Management Stats */
4657         adapter->stats.mgptc += er32(MGTPTC);
4658         adapter->stats.mgprc += er32(MGTPRC);
4659         adapter->stats.mgpdc += er32(MGTPDC);
4660
4661         /* Correctable ECC Errors */
4662         if (hw->mac.type == e1000_pch_lpt) {
4663                 u32 pbeccsts = er32(PBECCSTS);
4664                 adapter->corr_errors +=
4665                     pbeccsts & E1000_PBECCSTS_CORR_ERR_CNT_MASK;
4666                 adapter->uncorr_errors +=
4667                     (pbeccsts & E1000_PBECCSTS_UNCORR_ERR_CNT_MASK) >>
4668                     E1000_PBECCSTS_UNCORR_ERR_CNT_SHIFT;
4669         }
4670 }
4671
4672 /**
4673  * e1000_phy_read_status - Update the PHY register status snapshot
4674  * @adapter: board private structure
4675  **/
4676 static void e1000_phy_read_status(struct e1000_adapter *adapter)
4677 {
4678         struct e1000_hw *hw = &adapter->hw;
4679         struct e1000_phy_regs *phy = &adapter->phy_regs;
4680
4681         if (!pm_runtime_suspended((&adapter->pdev->dev)->parent) &&
4682             (er32(STATUS) & E1000_STATUS_LU) &&
4683             (adapter->hw.phy.media_type == e1000_media_type_copper)) {
4684                 int ret_val;
4685
4686                 ret_val = e1e_rphy(hw, MII_BMCR, &phy->bmcr);
4687                 ret_val |= e1e_rphy(hw, MII_BMSR, &phy->bmsr);
4688                 ret_val |= e1e_rphy(hw, MII_ADVERTISE, &phy->advertise);
4689                 ret_val |= e1e_rphy(hw, MII_LPA, &phy->lpa);
4690                 ret_val |= e1e_rphy(hw, MII_EXPANSION, &phy->expansion);
4691                 ret_val |= e1e_rphy(hw, MII_CTRL1000, &phy->ctrl1000);
4692                 ret_val |= e1e_rphy(hw, MII_STAT1000, &phy->stat1000);
4693                 ret_val |= e1e_rphy(hw, MII_ESTATUS, &phy->estatus);
4694                 if (ret_val)
4695                         e_warn("Error reading PHY register\n");
4696         } else {
4697                 /* Do not read PHY registers if link is not up
4698                  * Set values to typical power-on defaults
4699                  */
4700                 phy->bmcr = (BMCR_SPEED1000 | BMCR_ANENABLE | BMCR_FULLDPLX);
4701                 phy->bmsr = (BMSR_100FULL | BMSR_100HALF | BMSR_10FULL |
4702                              BMSR_10HALF | BMSR_ESTATEN | BMSR_ANEGCAPABLE |
4703                              BMSR_ERCAP);
4704                 phy->advertise = (ADVERTISE_PAUSE_ASYM | ADVERTISE_PAUSE_CAP |
4705                                   ADVERTISE_ALL | ADVERTISE_CSMA);
4706                 phy->lpa = 0;
4707                 phy->expansion = EXPANSION_ENABLENPAGE;
4708                 phy->ctrl1000 = ADVERTISE_1000FULL;
4709                 phy->stat1000 = 0;
4710                 phy->estatus = (ESTATUS_1000_TFULL | ESTATUS_1000_THALF);
4711         }
4712 }
4713
4714 static void e1000_print_link_info(struct e1000_adapter *adapter)
4715 {
4716         struct e1000_hw *hw = &adapter->hw;
4717         u32 ctrl = er32(CTRL);
4718
4719         /* Link status message must follow this format for user tools */
4720         pr_info("%s NIC Link is Up %d Mbps %s Duplex, Flow Control: %s\n",
4721                 adapter->netdev->name, adapter->link_speed,
4722                 adapter->link_duplex == FULL_DUPLEX ? "Full" : "Half",
4723                 (ctrl & E1000_CTRL_TFCE) && (ctrl & E1000_CTRL_RFCE) ? "Rx/Tx" :
4724                 (ctrl & E1000_CTRL_RFCE) ? "Rx" :
4725                 (ctrl & E1000_CTRL_TFCE) ? "Tx" : "None");
4726 }
4727
4728 static bool e1000e_has_link(struct e1000_adapter *adapter)
4729 {
4730         struct e1000_hw *hw = &adapter->hw;
4731         bool link_active = false;
4732         s32 ret_val = 0;
4733
4734         /* get_link_status is set on LSC (link status) interrupt or
4735          * Rx sequence error interrupt.  get_link_status will stay
4736          * false until the check_for_link establishes link
4737          * for copper adapters ONLY
4738          */
4739         switch (hw->phy.media_type) {
4740         case e1000_media_type_copper:
4741                 if (hw->mac.get_link_status) {
4742                         ret_val = hw->mac.ops.check_for_link(hw);
4743                         link_active = !hw->mac.get_link_status;
4744                 } else {
4745                         link_active = true;
4746                 }
4747                 break;
4748         case e1000_media_type_fiber:
4749                 ret_val = hw->mac.ops.check_for_link(hw);
4750                 link_active = !!(er32(STATUS) & E1000_STATUS_LU);
4751                 break;
4752         case e1000_media_type_internal_serdes:
4753                 ret_val = hw->mac.ops.check_for_link(hw);
4754                 link_active = adapter->hw.mac.serdes_has_link;
4755                 break;
4756         default:
4757         case e1000_media_type_unknown:
4758                 break;
4759         }
4760
4761         if ((ret_val == E1000_ERR_PHY) && (hw->phy.type == e1000_phy_igp_3) &&
4762             (er32(CTRL) & E1000_PHY_CTRL_GBE_DISABLE)) {
4763                 /* See e1000_kmrn_lock_loss_workaround_ich8lan() */
4764                 e_info("Gigabit has been disabled, downgrading speed\n");
4765         }
4766
4767         return link_active;
4768 }
4769
4770 static void e1000e_enable_receives(struct e1000_adapter *adapter)
4771 {
4772         /* make sure the receive unit is started */
4773         if ((adapter->flags & FLAG_RX_NEEDS_RESTART) &&
4774             (adapter->flags & FLAG_RESTART_NOW)) {
4775                 struct e1000_hw *hw = &adapter->hw;
4776                 u32 rctl = er32(RCTL);
4777                 ew32(RCTL, rctl | E1000_RCTL_EN);
4778                 adapter->flags &= ~FLAG_RESTART_NOW;
4779         }
4780 }
4781
4782 static void e1000e_check_82574_phy_workaround(struct e1000_adapter *adapter)
4783 {
4784         struct e1000_hw *hw = &adapter->hw;
4785
4786         /* With 82574 controllers, PHY needs to be checked periodically
4787          * for hung state and reset, if two calls return true
4788          */
4789         if (e1000_check_phy_82574(hw))
4790                 adapter->phy_hang_count++;
4791         else
4792                 adapter->phy_hang_count = 0;
4793
4794         if (adapter->phy_hang_count > 1) {
4795                 adapter->phy_hang_count = 0;
4796                 schedule_work(&adapter->reset_task);
4797         }
4798 }
4799
4800 /**
4801  * e1000_watchdog - Timer Call-back
4802  * @data: pointer to adapter cast into an unsigned long
4803  **/
4804 static void e1000_watchdog(unsigned long data)
4805 {
4806         struct e1000_adapter *adapter = (struct e1000_adapter *)data;
4807
4808         /* Do the rest outside of interrupt context */
4809         schedule_work(&adapter->watchdog_task);
4810
4811         /* TODO: make this use queue_delayed_work() */
4812 }
4813
4814 static void e1000_watchdog_task(struct work_struct *work)
4815 {
4816         struct e1000_adapter *adapter = container_of(work,
4817                                                      struct e1000_adapter,
4818                                                      watchdog_task);
4819         struct net_device *netdev = adapter->netdev;
4820         struct e1000_mac_info *mac = &adapter->hw.mac;
4821         struct e1000_phy_info *phy = &adapter->hw.phy;
4822         struct e1000_ring *tx_ring = adapter->tx_ring;
4823         struct e1000_hw *hw = &adapter->hw;
4824         u32 link, tctl;
4825
4826         if (test_bit(__E1000_DOWN, &adapter->state))
4827                 return;
4828
4829         link = e1000e_has_link(adapter);
4830         if ((netif_carrier_ok(netdev)) && link) {
4831                 /* Cancel scheduled suspend requests. */
4832                 pm_runtime_resume(netdev->dev.parent);
4833
4834                 e1000e_enable_receives(adapter);
4835                 goto link_up;
4836         }
4837
4838         if ((e1000e_enable_tx_pkt_filtering(hw)) &&
4839             (adapter->mng_vlan_id != adapter->hw.mng_cookie.vlan_id))
4840                 e1000_update_mng_vlan(adapter);
4841
4842         if (link) {
4843                 if (!netif_carrier_ok(netdev)) {
4844                         bool txb2b = true;
4845
4846                         /* Cancel scheduled suspend requests. */
4847                         pm_runtime_resume(netdev->dev.parent);
4848
4849                         /* update snapshot of PHY registers on LSC */
4850                         e1000_phy_read_status(adapter);
4851                         mac->ops.get_link_up_info(&adapter->hw,
4852                                                   &adapter->link_speed,
4853                                                   &adapter->link_duplex);
4854                         e1000_print_link_info(adapter);
4855
4856                         /* check if SmartSpeed worked */
4857                         e1000e_check_downshift(hw);
4858                         if (phy->speed_downgraded)
4859                                 netdev_warn(netdev,
4860                                             "Link Speed was downgraded by SmartSpeed\n");
4861
4862                         /* On supported PHYs, check for duplex mismatch only
4863                          * if link has autonegotiated at 10/100 half
4864                          */
4865                         if ((hw->phy.type == e1000_phy_igp_3 ||
4866                              hw->phy.type == e1000_phy_bm) &&
4867                             hw->mac.autoneg &&
4868                             (adapter->link_speed == SPEED_10 ||
4869                              adapter->link_speed == SPEED_100) &&
4870                             (adapter->link_duplex == HALF_DUPLEX)) {
4871                                 u16 autoneg_exp;
4872
4873                                 e1e_rphy(hw, MII_EXPANSION, &autoneg_exp);
4874
4875                                 if (!(autoneg_exp & EXPANSION_NWAY))
4876                                         e_info("Autonegotiated half duplex but link partner cannot autoneg.  Try forcing full duplex if link gets many collisions.\n");
4877                         }
4878
4879                         /* adjust timeout factor according to speed/duplex */
4880                         adapter->tx_timeout_factor = 1;
4881                         switch (adapter->link_speed) {
4882                         case SPEED_10:
4883                                 txb2b = false;
4884                                 adapter->tx_timeout_factor = 16;
4885                                 break;
4886                         case SPEED_100:
4887                                 txb2b = false;
4888                                 adapter->tx_timeout_factor = 10;
4889                                 break;
4890                         }
4891
4892                         /* workaround: re-program speed mode bit after
4893                          * link-up event
4894                          */
4895                         if ((adapter->flags & FLAG_TARC_SPEED_MODE_BIT) &&
4896                             !txb2b) {
4897                                 u32 tarc0;
4898                                 tarc0 = er32(TARC(0));
4899                                 tarc0 &= ~SPEED_MODE_BIT;
4900                                 ew32(TARC(0), tarc0);
4901                         }
4902
4903                         /* disable TSO for pcie and 10/100 speeds, to avoid
4904                          * some hardware issues
4905                          */
4906                         if (!(adapter->flags & FLAG_TSO_FORCE)) {
4907                                 switch (adapter->link_speed) {
4908                                 case SPEED_10:
4909                                 case SPEED_100:
4910                                         e_info("10/100 speed: disabling TSO\n");
4911                                         netdev->features &= ~NETIF_F_TSO;
4912                                         netdev->features &= ~NETIF_F_TSO6;
4913                                         break;
4914                                 case SPEED_1000:
4915                                         netdev->features |= NETIF_F_TSO;
4916                                         netdev->features |= NETIF_F_TSO6;
4917                                         break;
4918                                 default:
4919                                         /* oops */
4920                                         break;
4921                                 }
4922                         }
4923
4924                         /* enable transmits in the hardware, need to do this
4925                          * after setting TARC(0)
4926                          */
4927                         tctl = er32(TCTL);
4928                         tctl |= E1000_TCTL_EN;
4929                         ew32(TCTL, tctl);
4930
4931                         /* Perform any post-link-up configuration before
4932                          * reporting link up.
4933                          */
4934                         if (phy->ops.cfg_on_link_up)
4935                                 phy->ops.cfg_on_link_up(hw);
4936
4937                         netif_carrier_on(netdev);
4938
4939                         if (!test_bit(__E1000_DOWN, &adapter->state))
4940                                 mod_timer(&adapter->phy_info_timer,
4941                                           round_jiffies(jiffies + 2 * HZ));
4942                 }
4943         } else {
4944                 if (netif_carrier_ok(netdev)) {
4945                         adapter->link_speed = 0;
4946                         adapter->link_duplex = 0;
4947                         /* Link status message must follow this format */
4948                         pr_info("%s NIC Link is Down\n", adapter->netdev->name);
4949                         netif_carrier_off(netdev);
4950                         if (!test_bit(__E1000_DOWN, &adapter->state))
4951                                 mod_timer(&adapter->phy_info_timer,
4952                                           round_jiffies(jiffies + 2 * HZ));
4953
4954                         /* The link is lost so the controller stops DMA.
4955                          * If there is queued Tx work that cannot be done
4956                          * or if on an 8000ES2LAN which requires a Rx packet
4957                          * buffer work-around on link down event, reset the
4958                          * controller to flush the Tx/Rx packet buffers.
4959                          * (Do the reset outside of interrupt context).
4960                          */
4961                         if ((adapter->flags & FLAG_RX_NEEDS_RESTART) ||
4962                             (e1000_desc_unused(tx_ring) + 1 < tx_ring->count))
4963                                 adapter->flags |= FLAG_RESTART_NOW;
4964                         else
4965                                 pm_schedule_suspend(netdev->dev.parent,
4966                                                     LINK_TIMEOUT);
4967                 }
4968         }
4969
4970 link_up:
4971         spin_lock(&adapter->stats64_lock);
4972         e1000e_update_stats(adapter);
4973
4974         mac->tx_packet_delta = adapter->stats.tpt - adapter->tpt_old;
4975         adapter->tpt_old = adapter->stats.tpt;
4976         mac->collision_delta = adapter->stats.colc - adapter->colc_old;
4977         adapter->colc_old = adapter->stats.colc;
4978
4979         adapter->gorc = adapter->stats.gorc - adapter->gorc_old;
4980         adapter->gorc_old = adapter->stats.gorc;
4981         adapter->gotc = adapter->stats.gotc - adapter->gotc_old;
4982         adapter->gotc_old = adapter->stats.gotc;
4983         spin_unlock(&adapter->stats64_lock);
4984
4985         if (adapter->flags & FLAG_RESTART_NOW) {
4986                 schedule_work(&adapter->reset_task);
4987                 /* return immediately since reset is imminent */
4988                 return;
4989         }
4990
4991         e1000e_update_adaptive(&adapter->hw);
4992
4993         /* Simple mode for Interrupt Throttle Rate (ITR) */
4994         if (adapter->itr_setting == 4) {
4995                 /* Symmetric Tx/Rx gets a reduced ITR=2000;
4996                  * Total asymmetrical Tx or Rx gets ITR=8000;
4997                  * everyone else is between 2000-8000.
4998                  */
4999                 u32 goc = (adapter->gotc + adapter->gorc) / 10000;
5000                 u32 dif = (adapter->gotc > adapter->gorc ?
5001                            adapter->gotc - adapter->gorc :
5002                            adapter->gorc - adapter->gotc) / 10000;
5003                 u32 itr = goc > 0 ? (dif * 6000 / goc + 2000) : 8000;
5004
5005                 e1000e_write_itr(adapter, itr);
5006         }
5007
5008         /* Cause software interrupt to ensure Rx ring is cleaned */
5009         if (adapter->msix_entries)
5010                 ew32(ICS, adapter->rx_ring->ims_val);
5011         else
5012                 ew32(ICS, E1000_ICS_RXDMT0);
5013
5014         /* flush pending descriptors to memory before detecting Tx hang */
5015         e1000e_flush_descriptors(adapter);
5016
5017         /* Force detection of hung controller every watchdog period */
5018         adapter->detect_tx_hung = true;
5019
5020         /* With 82571 controllers, LAA may be overwritten due to controller
5021          * reset from the other port. Set the appropriate LAA in RAR[0]
5022          */
5023         if (e1000e_get_laa_state_82571(hw))
5024                 hw->mac.ops.rar_set(hw, adapter->hw.mac.addr, 0);
5025
5026         if (adapter->flags2 & FLAG2_CHECK_PHY_HANG)
5027                 e1000e_check_82574_phy_workaround(adapter);
5028
5029         /* Clear valid timestamp stuck in RXSTMPL/H due to a Rx error */
5030         if (adapter->hwtstamp_config.rx_filter != HWTSTAMP_FILTER_NONE) {
5031                 if ((adapter->flags2 & FLAG2_CHECK_RX_HWTSTAMP) &&
5032                     (er32(TSYNCRXCTL) & E1000_TSYNCRXCTL_VALID)) {
5033                         er32(RXSTMPH);
5034                         adapter->rx_hwtstamp_cleared++;
5035                 } else {
5036                         adapter->flags2 |= FLAG2_CHECK_RX_HWTSTAMP;
5037                 }
5038         }
5039
5040         /* Reset the timer */
5041         if (!test_bit(__E1000_DOWN, &adapter->state))
5042                 mod_timer(&adapter->watchdog_timer,
5043                           round_jiffies(jiffies + 2 * HZ));
5044 }
5045
5046 #define E1000_TX_FLAGS_CSUM             0x00000001
5047 #define E1000_TX_FLAGS_VLAN             0x00000002
5048 #define E1000_TX_FLAGS_TSO              0x00000004
5049 #define E1000_TX_FLAGS_IPV4             0x00000008
5050 #define E1000_TX_FLAGS_NO_FCS           0x00000010
5051 #define E1000_TX_FLAGS_HWTSTAMP         0x00000020
5052 #define E1000_TX_FLAGS_VLAN_MASK        0xffff0000
5053 #define E1000_TX_FLAGS_VLAN_SHIFT       16
5054
5055 static int e1000_tso(struct e1000_ring *tx_ring, struct sk_buff *skb)
5056 {
5057         struct e1000_context_desc *context_desc;
5058         struct e1000_buffer *buffer_info;
5059         unsigned int i;
5060         u32 cmd_length = 0;
5061         u16 ipcse = 0, mss;
5062         u8 ipcss, ipcso, tucss, tucso, hdr_len;
5063
5064         if (!skb_is_gso(skb))
5065                 return 0;
5066
5067         if (skb_header_cloned(skb)) {
5068                 int err = pskb_expand_head(skb, 0, 0, GFP_ATOMIC);
5069
5070                 if (err)
5071                         return err;
5072         }
5073
5074         hdr_len = skb_transport_offset(skb) + tcp_hdrlen(skb);
5075         mss = skb_shinfo(skb)->gso_size;
5076         if (skb->protocol == htons(ETH_P_IP)) {
5077                 struct iphdr *iph = ip_hdr(skb);
5078                 iph->tot_len = 0;
5079                 iph->check = 0;
5080                 tcp_hdr(skb)->check = ~csum_tcpudp_magic(iph->saddr, iph->daddr,
5081                                                          0, IPPROTO_TCP, 0);
5082                 cmd_length = E1000_TXD_CMD_IP;
5083                 ipcse = skb_transport_offset(skb) - 1;
5084         } else if (skb_is_gso_v6(skb)) {
5085                 ipv6_hdr(skb)->payload_len = 0;
5086                 tcp_hdr(skb)->check = ~csum_ipv6_magic(&ipv6_hdr(skb)->saddr,
5087                                                        &ipv6_hdr(skb)->daddr,
5088                                                        0, IPPROTO_TCP, 0);
5089                 ipcse = 0;
5090         }
5091         ipcss = skb_network_offset(skb);
5092         ipcso = (void *)&(ip_hdr(skb)->check) - (void *)skb->data;
5093         tucss = skb_transport_offset(skb);
5094         tucso = (void *)&(tcp_hdr(skb)->check) - (void *)skb->data;
5095
5096         cmd_length |= (E1000_TXD_CMD_DEXT | E1000_TXD_CMD_TSE |
5097                        E1000_TXD_CMD_TCP | (skb->len - (hdr_len)));
5098
5099         i = tx_ring->next_to_use;
5100         context_desc = E1000_CONTEXT_DESC(*tx_ring, i);
5101         buffer_info = &tx_ring->buffer_info[i];
5102
5103         context_desc->lower_setup.ip_fields.ipcss = ipcss;
5104         context_desc->lower_setup.ip_fields.ipcso = ipcso;
5105         context_desc->lower_setup.ip_fields.ipcse = cpu_to_le16(ipcse);
5106         context_desc->upper_setup.tcp_fields.tucss = tucss;
5107         context_desc->upper_setup.tcp_fields.tucso = tucso;
5108         context_desc->upper_setup.tcp_fields.tucse = 0;
5109         context_desc->tcp_seg_setup.fields.mss = cpu_to_le16(mss);
5110         context_desc->tcp_seg_setup.fields.hdr_len = hdr_len;
5111         context_desc->cmd_and_length = cpu_to_le32(cmd_length);
5112
5113         buffer_info->time_stamp = jiffies;
5114         buffer_info->next_to_watch = i;
5115
5116         i++;
5117         if (i == tx_ring->count)
5118                 i = 0;
5119         tx_ring->next_to_use = i;
5120
5121         return 1;
5122 }
5123
5124 static bool e1000_tx_csum(struct e1000_ring *tx_ring, struct sk_buff *skb)
5125 {
5126         struct e1000_adapter *adapter = tx_ring->adapter;
5127         struct e1000_context_desc *context_desc;
5128         struct e1000_buffer *buffer_info;
5129         unsigned int i;
5130         u8 css;
5131         u32 cmd_len = E1000_TXD_CMD_DEXT;
5132         __be16 protocol;
5133
5134         if (skb->ip_summed != CHECKSUM_PARTIAL)
5135                 return 0;
5136
5137         if (skb->protocol == cpu_to_be16(ETH_P_8021Q))
5138                 protocol = vlan_eth_hdr(skb)->h_vlan_encapsulated_proto;
5139         else
5140                 protocol = skb->protocol;
5141
5142         switch (protocol) {
5143         case cpu_to_be16(ETH_P_IP):
5144                 if (ip_hdr(skb)->protocol == IPPROTO_TCP)
5145                         cmd_len |= E1000_TXD_CMD_TCP;
5146                 break;
5147         case cpu_to_be16(ETH_P_IPV6):
5148                 /* XXX not handling all IPV6 headers */
5149                 if (ipv6_hdr(skb)->nexthdr == IPPROTO_TCP)
5150                         cmd_len |= E1000_TXD_CMD_TCP;
5151                 break;
5152         default:
5153                 if (unlikely(net_ratelimit()))
5154                         e_warn("checksum_partial proto=%x!\n",
5155                                be16_to_cpu(protocol));
5156                 break;
5157         }
5158
5159         css = skb_checksum_start_offset(skb);
5160
5161         i = tx_ring->next_to_use;
5162         buffer_info = &tx_ring->buffer_info[i];
5163         context_desc = E1000_CONTEXT_DESC(*tx_ring, i);
5164
5165         context_desc->lower_setup.ip_config = 0;
5166         context_desc->upper_setup.tcp_fields.tucss = css;
5167         context_desc->upper_setup.tcp_fields.tucso = css + skb->csum_offset;
5168         context_desc->upper_setup.tcp_fields.tucse = 0;
5169         context_desc->tcp_seg_setup.data = 0;
5170         context_desc->cmd_and_length = cpu_to_le32(cmd_len);
5171
5172         buffer_info->time_stamp = jiffies;
5173         buffer_info->next_to_watch = i;
5174
5175         i++;
5176         if (i == tx_ring->count)
5177                 i = 0;
5178         tx_ring->next_to_use = i;
5179
5180         return 1;
5181 }
5182
5183 static int e1000_tx_map(struct e1000_ring *tx_ring, struct sk_buff *skb,
5184                         unsigned int first, unsigned int max_per_txd,
5185                         unsigned int nr_frags)
5186 {
5187         struct e1000_adapter *adapter = tx_ring->adapter;
5188         struct pci_dev *pdev = adapter->pdev;
5189         struct e1000_buffer *buffer_info;
5190         unsigned int len = skb_headlen(skb);
5191         unsigned int offset = 0, size, count = 0, i;
5192         unsigned int f, bytecount, segs;
5193
5194         i = tx_ring->next_to_use;
5195
5196         while (len) {
5197                 buffer_info = &tx_ring->buffer_info[i];
5198                 size = min(len, max_per_txd);
5199
5200                 buffer_info->length = size;
5201                 buffer_info->time_stamp = jiffies;
5202                 buffer_info->next_to_watch = i;
5203                 buffer_info->dma = dma_map_single(&pdev->dev,
5204                                                   skb->data + offset,
5205                                                   size, DMA_TO_DEVICE);
5206                 buffer_info->mapped_as_page = false;
5207                 if (dma_mapping_error(&pdev->dev, buffer_info->dma))
5208                         goto dma_error;
5209
5210                 len -= size;
5211                 offset += size;
5212                 count++;
5213
5214                 if (len) {
5215                         i++;
5216                         if (i == tx_ring->count)
5217                                 i = 0;
5218                 }
5219         }
5220
5221         for (f = 0; f < nr_frags; f++) {
5222                 const struct skb_frag_struct *frag;
5223
5224                 frag = &skb_shinfo(skb)->frags[f];
5225                 len = skb_frag_size(frag);
5226                 offset = 0;
5227
5228                 while (len) {
5229                         i++;
5230                         if (i == tx_ring->count)
5231                                 i = 0;
5232
5233                         buffer_info = &tx_ring->buffer_info[i];
5234                         size = min(len, max_per_txd);
5235
5236                         buffer_info->length = size;
5237                         buffer_info->time_stamp = jiffies;
5238                         buffer_info->next_to_watch = i;
5239                         buffer_info->dma = skb_frag_dma_map(&pdev->dev, frag,
5240                                                             offset, size,
5241                                                             DMA_TO_DEVICE);
5242                         buffer_info->mapped_as_page = true;
5243                         if (dma_mapping_error(&pdev->dev, buffer_info->dma))
5244                                 goto dma_error;
5245
5246                         len -= size;
5247                         offset += size;
5248                         count++;
5249                 }
5250         }
5251
5252         segs = skb_shinfo(skb)->gso_segs ? : 1;
5253         /* multiply data chunks by size of headers */
5254         bytecount = ((segs - 1) * skb_headlen(skb)) + skb->len;
5255
5256         tx_ring->buffer_info[i].skb = skb;
5257         tx_ring->buffer_info[i].segs = segs;
5258         tx_ring->buffer_info[i].bytecount = bytecount;
5259         tx_ring->buffer_info[first].next_to_watch = i;
5260
5261         return count;
5262
5263 dma_error:
5264         dev_err(&pdev->dev, "Tx DMA map failed\n");
5265         buffer_info->dma = 0;
5266         if (count)
5267                 count--;
5268
5269         while (count--) {
5270                 if (i == 0)
5271                         i += tx_ring->count;
5272                 i--;
5273                 buffer_info = &tx_ring->buffer_info[i];
5274                 e1000_put_txbuf(tx_ring, buffer_info);
5275         }
5276
5277         return 0;
5278 }
5279
5280 static void e1000_tx_queue(struct e1000_ring *tx_ring, int tx_flags, int count)
5281 {
5282         struct e1000_adapter *adapter = tx_ring->adapter;
5283         struct e1000_tx_desc *tx_desc = NULL;
5284         struct e1000_buffer *buffer_info;
5285         u32 txd_upper = 0, txd_lower = E1000_TXD_CMD_IFCS;
5286         unsigned int i;
5287
5288         if (tx_flags & E1000_TX_FLAGS_TSO) {
5289                 txd_lower |= E1000_TXD_CMD_DEXT | E1000_TXD_DTYP_D |
5290                     E1000_TXD_CMD_TSE;
5291                 txd_upper |= E1000_TXD_POPTS_TXSM << 8;
5292
5293                 if (tx_flags & E1000_TX_FLAGS_IPV4)
5294                         txd_upper |= E1000_TXD_POPTS_IXSM << 8;
5295         }
5296
5297         if (tx_flags & E1000_TX_FLAGS_CSUM) {
5298                 txd_lower |= E1000_TXD_CMD_DEXT | E1000_TXD_DTYP_D;
5299                 txd_upper |= E1000_TXD_POPTS_TXSM << 8;
5300         }
5301
5302         if (tx_flags & E1000_TX_FLAGS_VLAN) {
5303                 txd_lower |= E1000_TXD_CMD_VLE;
5304                 txd_upper |= (tx_flags & E1000_TX_FLAGS_VLAN_MASK);
5305         }
5306
5307         if (unlikely(tx_flags & E1000_TX_FLAGS_NO_FCS))
5308                 txd_lower &= ~(E1000_TXD_CMD_IFCS);
5309
5310         if (unlikely(tx_flags & E1000_TX_FLAGS_HWTSTAMP)) {
5311                 txd_lower |= E1000_TXD_CMD_DEXT | E1000_TXD_DTYP_D;
5312                 txd_upper |= E1000_TXD_EXTCMD_TSTAMP;
5313         }
5314
5315         i = tx_ring->next_to_use;
5316
5317         do {
5318                 buffer_info = &tx_ring->buffer_info[i];
5319                 tx_desc = E1000_TX_DESC(*tx_ring, i);
5320                 tx_desc->buffer_addr = cpu_to_le64(buffer_info->dma);
5321                 tx_desc->lower.data = cpu_to_le32(txd_lower |
5322                                                   buffer_info->length);
5323                 tx_desc->upper.data = cpu_to_le32(txd_upper);
5324
5325                 i++;
5326                 if (i == tx_ring->count)
5327                         i = 0;
5328         } while (--count > 0);
5329
5330         tx_desc->lower.data |= cpu_to_le32(adapter->txd_cmd);
5331
5332         /* txd_cmd re-enables FCS, so we'll re-disable it here as desired. */
5333         if (unlikely(tx_flags & E1000_TX_FLAGS_NO_FCS))
5334                 tx_desc->lower.data &= ~(cpu_to_le32(E1000_TXD_CMD_IFCS));
5335
5336         /* Force memory writes to complete before letting h/w
5337          * know there are new descriptors to fetch.  (Only
5338          * applicable for weak-ordered memory model archs,
5339          * such as IA-64).
5340          */
5341         wmb();
5342
5343         tx_ring->next_to_use = i;
5344
5345         if (adapter->flags2 & FLAG2_PCIM2PCI_ARBITER_WA)
5346                 e1000e_update_tdt_wa(tx_ring, i);
5347         else
5348                 writel(i, tx_ring->tail);
5349
5350         /* we need this if more than one processor can write to our tail
5351          * at a time, it synchronizes IO on IA64/Altix systems
5352          */
5353         mmiowb();
5354 }
5355
5356 #define MINIMUM_DHCP_PACKET_SIZE 282
5357 static int e1000_transfer_dhcp_info(struct e1000_adapter *adapter,
5358                                     struct sk_buff *skb)
5359 {
5360         struct e1000_hw *hw = &adapter->hw;
5361         u16 length, offset;
5362
5363         if (vlan_tx_tag_present(skb) &&
5364             !((vlan_tx_tag_get(skb) == adapter->hw.mng_cookie.vlan_id) &&
5365               (adapter->hw.mng_cookie.status &
5366                E1000_MNG_DHCP_COOKIE_STATUS_VLAN)))
5367                 return 0;
5368
5369         if (skb->len <= MINIMUM_DHCP_PACKET_SIZE)
5370                 return 0;
5371
5372         if (((struct ethhdr *)skb->data)->h_proto != htons(ETH_P_IP))
5373                 return 0;
5374
5375         {
5376                 const struct iphdr *ip = (struct iphdr *)((u8 *)skb->data + 14);
5377                 struct udphdr *udp;
5378
5379                 if (ip->protocol != IPPROTO_UDP)
5380                         return 0;
5381
5382                 udp = (struct udphdr *)((u8 *)ip + (ip->ihl << 2));
5383                 if (ntohs(udp->dest) != 67)
5384                         return 0;
5385
5386                 offset = (u8 *)udp + 8 - skb->data;
5387                 length = skb->len - offset;
5388                 return e1000e_mng_write_dhcp_info(hw, (u8 *)udp + 8, length);
5389         }
5390
5391         return 0;
5392 }
5393
5394 static int __e1000_maybe_stop_tx(struct e1000_ring *tx_ring, int size)
5395 {
5396         struct e1000_adapter *adapter = tx_ring->adapter;
5397
5398         netif_stop_queue(adapter->netdev);
5399         /* Herbert's original patch had:
5400          *  smp_mb__after_netif_stop_queue();
5401          * but since that doesn't exist yet, just open code it.
5402          */
5403         smp_mb();
5404
5405         /* We need to check again in a case another CPU has just
5406          * made room available.
5407          */
5408         if (e1000_desc_unused(tx_ring) < size)
5409                 return -EBUSY;
5410
5411         /* A reprieve! */
5412         netif_start_queue(adapter->netdev);
5413         ++adapter->restart_queue;
5414         return 0;
5415 }
5416
5417 static int e1000_maybe_stop_tx(struct e1000_ring *tx_ring, int size)
5418 {
5419         BUG_ON(size > tx_ring->count);
5420
5421         if (e1000_desc_unused(tx_ring) >= size)
5422                 return 0;
5423         return __e1000_maybe_stop_tx(tx_ring, size);
5424 }
5425
5426 static netdev_tx_t e1000_xmit_frame(struct sk_buff *skb,
5427                                     struct net_device *netdev)
5428 {
5429         struct e1000_adapter *adapter = netdev_priv(netdev);
5430         struct e1000_ring *tx_ring = adapter->tx_ring;
5431         unsigned int first;
5432         unsigned int tx_flags = 0;
5433         unsigned int len = skb_headlen(skb);
5434         unsigned int nr_frags;
5435         unsigned int mss;
5436         int count = 0;
5437         int tso;
5438         unsigned int f;
5439
5440         if (test_bit(__E1000_DOWN, &adapter->state)) {
5441                 dev_kfree_skb_any(skb);
5442                 return NETDEV_TX_OK;
5443         }
5444
5445         if (skb->len <= 0) {
5446                 dev_kfree_skb_any(skb);
5447                 return NETDEV_TX_OK;
5448         }
5449
5450         /* The minimum packet size with TCTL.PSP set is 17 bytes so
5451          * pad skb in order to meet this minimum size requirement
5452          */
5453         if (unlikely(skb->len < 17)) {
5454                 if (skb_pad(skb, 17 - skb->len))
5455                         return NETDEV_TX_OK;
5456                 skb->len = 17;
5457                 skb_set_tail_pointer(skb, 17);
5458         }
5459
5460         mss = skb_shinfo(skb)->gso_size;
5461         if (mss) {
5462                 u8 hdr_len;
5463
5464                 /* TSO Workaround for 82571/2/3 Controllers -- if skb->data
5465                  * points to just header, pull a few bytes of payload from
5466                  * frags into skb->data
5467                  */
5468                 hdr_len = skb_transport_offset(skb) + tcp_hdrlen(skb);
5469                 /* we do this workaround for ES2LAN, but it is un-necessary,
5470                  * avoiding it could save a lot of cycles
5471                  */
5472                 if (skb->data_len && (hdr_len == len)) {
5473                         unsigned int pull_size;
5474
5475                         pull_size = min_t(unsigned int, 4, skb->data_len);
5476                         if (!__pskb_pull_tail(skb, pull_size)) {
5477                                 e_err("__pskb_pull_tail failed.\n");
5478                                 dev_kfree_skb_any(skb);
5479                                 return NETDEV_TX_OK;
5480                         }
5481                         len = skb_headlen(skb);
5482                 }
5483         }
5484
5485         /* reserve a descriptor for the offload context */
5486         if ((mss) || (skb->ip_summed == CHECKSUM_PARTIAL))
5487                 count++;
5488         count++;
5489
5490         count += DIV_ROUND_UP(len, adapter->tx_fifo_limit);
5491
5492         nr_frags = skb_shinfo(skb)->nr_frags;
5493         for (f = 0; f < nr_frags; f++)
5494                 count += DIV_ROUND_UP(skb_frag_size(&skb_shinfo(skb)->frags[f]),
5495                                       adapter->tx_fifo_limit);
5496
5497         if (adapter->hw.mac.tx_pkt_filtering)
5498                 e1000_transfer_dhcp_info(adapter, skb);
5499
5500         /* need: count + 2 desc gap to keep tail from touching
5501          * head, otherwise try next time
5502          */
5503         if (e1000_maybe_stop_tx(tx_ring, count + 2))
5504                 return NETDEV_TX_BUSY;
5505
5506         if (vlan_tx_tag_present(skb)) {
5507                 tx_flags |= E1000_TX_FLAGS_VLAN;
5508                 tx_flags |= (vlan_tx_tag_get(skb) << E1000_TX_FLAGS_VLAN_SHIFT);
5509         }
5510
5511         first = tx_ring->next_to_use;
5512
5513         tso = e1000_tso(tx_ring, skb);
5514         if (tso < 0) {
5515                 dev_kfree_skb_any(skb);
5516                 return NETDEV_TX_OK;
5517         }
5518
5519         if (tso)
5520                 tx_flags |= E1000_TX_FLAGS_TSO;
5521         else if (e1000_tx_csum(tx_ring, skb))
5522                 tx_flags |= E1000_TX_FLAGS_CSUM;
5523
5524         /* Old method was to assume IPv4 packet by default if TSO was enabled.
5525          * 82571 hardware supports TSO capabilities for IPv6 as well...
5526          * no longer assume, we must.
5527          */
5528         if (skb->protocol == htons(ETH_P_IP))
5529                 tx_flags |= E1000_TX_FLAGS_IPV4;
5530
5531         if (unlikely(skb->no_fcs))
5532                 tx_flags |= E1000_TX_FLAGS_NO_FCS;
5533
5534         /* if count is 0 then mapping error has occurred */
5535         count = e1000_tx_map(tx_ring, skb, first, adapter->tx_fifo_limit,
5536                              nr_frags);
5537         if (count) {
5538                 if (unlikely((skb_shinfo(skb)->tx_flags & SKBTX_HW_TSTAMP) &&
5539                              !adapter->tx_hwtstamp_skb)) {
5540                         skb_shinfo(skb)->tx_flags |= SKBTX_IN_PROGRESS;
5541                         tx_flags |= E1000_TX_FLAGS_HWTSTAMP;
5542                         adapter->tx_hwtstamp_skb = skb_get(skb);
5543                         schedule_work(&adapter->tx_hwtstamp_work);
5544                 } else {
5545                         skb_tx_timestamp(skb);
5546                 }
5547
5548                 netdev_sent_queue(netdev, skb->len);
5549                 e1000_tx_queue(tx_ring, tx_flags, count);
5550                 /* Make sure there is space in the ring for the next send. */
5551                 e1000_maybe_stop_tx(tx_ring,
5552                                     (MAX_SKB_FRAGS *
5553                                      DIV_ROUND_UP(PAGE_SIZE,
5554                                                   adapter->tx_fifo_limit) + 2));
5555         } else {
5556                 dev_kfree_skb_any(skb);
5557                 tx_ring->buffer_info[first].time_stamp = 0;
5558                 tx_ring->next_to_use = first;
5559         }
5560
5561         return NETDEV_TX_OK;
5562 }
5563
5564 /**
5565  * e1000_tx_timeout - Respond to a Tx Hang
5566  * @netdev: network interface device structure
5567  **/
5568 static void e1000_tx_timeout(struct net_device *netdev)
5569 {
5570         struct e1000_adapter *adapter = netdev_priv(netdev);
5571
5572         /* Do the reset outside of interrupt context */
5573         adapter->tx_timeout_count++;
5574         schedule_work(&adapter->reset_task);
5575 }
5576
5577 static void e1000_reset_task(struct work_struct *work)
5578 {
5579         struct e1000_adapter *adapter;
5580         adapter = container_of(work, struct e1000_adapter, reset_task);
5581
5582         /* don't run the task if already down */
5583         if (test_bit(__E1000_DOWN, &adapter->state))
5584                 return;
5585
5586         if (!(adapter->flags & FLAG_RESTART_NOW)) {
5587                 e1000e_dump(adapter);
5588                 e_err("Reset adapter unexpectedly\n");
5589         }
5590         e1000e_reinit_locked(adapter);
5591 }
5592
5593 /**
5594  * e1000_get_stats64 - Get System Network Statistics
5595  * @netdev: network interface device structure
5596  * @stats: rtnl_link_stats64 pointer
5597  *
5598  * Returns the address of the device statistics structure.
5599  **/
5600 struct rtnl_link_stats64 *e1000e_get_stats64(struct net_device *netdev,
5601                                              struct rtnl_link_stats64 *stats)
5602 {
5603         struct e1000_adapter *adapter = netdev_priv(netdev);
5604
5605         memset(stats, 0, sizeof(struct rtnl_link_stats64));
5606         spin_lock(&adapter->stats64_lock);
5607         e1000e_update_stats(adapter);
5608         /* Fill out the OS statistics structure */
5609         stats->rx_bytes = adapter->stats.gorc;
5610         stats->rx_packets = adapter->stats.gprc;
5611         stats->tx_bytes = adapter->stats.gotc;
5612         stats->tx_packets = adapter->stats.gptc;
5613         stats->multicast = adapter->stats.mprc;
5614         stats->collisions = adapter->stats.colc;
5615
5616         /* Rx Errors */
5617
5618         /* RLEC on some newer hardware can be incorrect so build
5619          * our own version based on RUC and ROC
5620          */
5621         stats->rx_errors = adapter->stats.rxerrc +
5622             adapter->stats.crcerrs + adapter->stats.algnerrc +
5623             adapter->stats.ruc + adapter->stats.roc + adapter->stats.cexterr;
5624         stats->rx_length_errors = adapter->stats.ruc + adapter->stats.roc;
5625         stats->rx_crc_errors = adapter->stats.crcerrs;
5626         stats->rx_frame_errors = adapter->stats.algnerrc;
5627         stats->rx_missed_errors = adapter->stats.mpc;
5628
5629         /* Tx Errors */
5630         stats->tx_errors = adapter->stats.ecol + adapter->stats.latecol;
5631         stats->tx_aborted_errors = adapter->stats.ecol;
5632         stats->tx_window_errors = adapter->stats.latecol;
5633         stats->tx_carrier_errors = adapter->stats.tncrs;
5634
5635         /* Tx Dropped needs to be maintained elsewhere */
5636
5637         spin_unlock(&adapter->stats64_lock);
5638         return stats;
5639 }
5640
5641 /**
5642  * e1000_change_mtu - Change the Maximum Transfer Unit
5643  * @netdev: network interface device structure
5644  * @new_mtu: new value for maximum frame size
5645  *
5646  * Returns 0 on success, negative on failure
5647  **/
5648 static int e1000_change_mtu(struct net_device *netdev, int new_mtu)
5649 {
5650         struct e1000_adapter *adapter = netdev_priv(netdev);
5651         int max_frame = new_mtu + ETH_HLEN + ETH_FCS_LEN;
5652
5653         /* Jumbo frame support */
5654         if ((max_frame > ETH_FRAME_LEN + ETH_FCS_LEN) &&
5655             !(adapter->flags & FLAG_HAS_JUMBO_FRAMES)) {
5656                 e_err("Jumbo Frames not supported.\n");
5657                 return -EINVAL;
5658         }
5659
5660         /* Supported frame sizes */
5661         if ((new_mtu < ETH_ZLEN + ETH_FCS_LEN + VLAN_HLEN) ||
5662             (max_frame > adapter->max_hw_frame_size)) {
5663                 e_err("Unsupported MTU setting\n");
5664                 return -EINVAL;
5665         }
5666
5667         /* Jumbo frame workaround on 82579 and newer requires CRC be stripped */
5668         if ((adapter->hw.mac.type >= e1000_pch2lan) &&
5669             !(adapter->flags2 & FLAG2_CRC_STRIPPING) &&
5670             (new_mtu > ETH_DATA_LEN)) {
5671                 e_err("Jumbo Frames not supported on this device when CRC stripping is disabled.\n");
5672                 return -EINVAL;
5673         }
5674
5675         while (test_and_set_bit(__E1000_RESETTING, &adapter->state))
5676                 usleep_range(1000, 2000);
5677         /* e1000e_down -> e1000e_reset dependent on max_frame_size & mtu */
5678         adapter->max_frame_size = max_frame;
5679         e_info("changing MTU from %d to %d\n", netdev->mtu, new_mtu);
5680         netdev->mtu = new_mtu;
5681         if (netif_running(netdev))
5682                 e1000e_down(adapter, true);
5683
5684         /* NOTE: netdev_alloc_skb reserves 16 bytes, and typically NET_IP_ALIGN
5685          * means we reserve 2 more, this pushes us to allocate from the next
5686          * larger slab size.
5687          * i.e. RXBUFFER_2048 --> size-4096 slab
5688          * However with the new *_jumbo_rx* routines, jumbo receives will use
5689          * fragmented skbs
5690          */
5691
5692         if (max_frame <= 2048)
5693                 adapter->rx_buffer_len = 2048;
5694         else
5695                 adapter->rx_buffer_len = 4096;
5696
5697         /* adjust allocation if LPE protects us, and we aren't using SBP */
5698         if ((max_frame == ETH_FRAME_LEN + ETH_FCS_LEN) ||
5699             (max_frame == ETH_FRAME_LEN + VLAN_HLEN + ETH_FCS_LEN))
5700                 adapter->rx_buffer_len = ETH_FRAME_LEN + VLAN_HLEN
5701                     + ETH_FCS_LEN;
5702
5703         if (netif_running(netdev))
5704                 e1000e_up(adapter);
5705         else
5706                 e1000e_reset(adapter);
5707
5708         clear_bit(__E1000_RESETTING, &adapter->state);
5709
5710         return 0;
5711 }
5712
5713 static int e1000_mii_ioctl(struct net_device *netdev, struct ifreq *ifr,
5714                            int cmd)
5715 {
5716         struct e1000_adapter *adapter = netdev_priv(netdev);
5717         struct mii_ioctl_data *data = if_mii(ifr);
5718
5719         if (adapter->hw.phy.media_type != e1000_media_type_copper)
5720                 return -EOPNOTSUPP;
5721
5722         switch (cmd) {
5723         case SIOCGMIIPHY:
5724                 data->phy_id = adapter->hw.phy.addr;
5725                 break;
5726         case SIOCGMIIREG:
5727                 e1000_phy_read_status(adapter);
5728
5729                 switch (data->reg_num & 0x1F) {
5730                 case MII_BMCR:
5731                         data->val_out = adapter->phy_regs.bmcr;
5732                         break;
5733                 case MII_BMSR:
5734                         data->val_out = adapter->phy_regs.bmsr;
5735                         break;
5736                 case MII_PHYSID1:
5737                         data->val_out = (adapter->hw.phy.id >> 16);
5738                         break;
5739                 case MII_PHYSID2:
5740                         data->val_out = (adapter->hw.phy.id & 0xFFFF);
5741                         break;
5742                 case MII_ADVERTISE:
5743                         data->val_out = adapter->phy_regs.advertise;
5744                         break;
5745                 case MII_LPA:
5746                         data->val_out = adapter->phy_regs.lpa;
5747                         break;
5748                 case MII_EXPANSION:
5749                         data->val_out = adapter->phy_regs.expansion;
5750                         break;
5751                 case MII_CTRL1000:
5752                         data->val_out = adapter->phy_regs.ctrl1000;
5753                         break;
5754                 case MII_STAT1000:
5755                         data->val_out = adapter->phy_regs.stat1000;
5756                         break;
5757                 case MII_ESTATUS:
5758                         data->val_out = adapter->phy_regs.estatus;
5759                         break;
5760                 default:
5761                         return -EIO;
5762                 }
5763                 break;
5764         case SIOCSMIIREG:
5765         default:
5766                 return -EOPNOTSUPP;
5767         }
5768         return 0;
5769 }
5770
5771 /**
5772  * e1000e_hwtstamp_ioctl - control hardware time stamping
5773  * @netdev: network interface device structure
5774  * @ifreq: interface request
5775  *
5776  * Outgoing time stamping can be enabled and disabled. Play nice and
5777  * disable it when requested, although it shouldn't cause any overhead
5778  * when no packet needs it. At most one packet in the queue may be
5779  * marked for time stamping, otherwise it would be impossible to tell
5780  * for sure to which packet the hardware time stamp belongs.
5781  *
5782  * Incoming time stamping has to be configured via the hardware filters.
5783  * Not all combinations are supported, in particular event type has to be
5784  * specified. Matching the kind of event packet is not supported, with the
5785  * exception of "all V2 events regardless of level 2 or 4".
5786  **/
5787 static int e1000e_hwtstamp_set(struct net_device *netdev, struct ifreq *ifr)
5788 {
5789         struct e1000_adapter *adapter = netdev_priv(netdev);
5790         struct hwtstamp_config config;
5791         int ret_val;
5792
5793         if (copy_from_user(&config, ifr->ifr_data, sizeof(config)))
5794                 return -EFAULT;
5795
5796         ret_val = e1000e_config_hwtstamp(adapter, &config);
5797         if (ret_val)
5798                 return ret_val;
5799
5800         switch (config.rx_filter) {
5801         case HWTSTAMP_FILTER_PTP_V2_L4_SYNC:
5802         case HWTSTAMP_FILTER_PTP_V2_L2_SYNC:
5803         case HWTSTAMP_FILTER_PTP_V2_SYNC:
5804         case HWTSTAMP_FILTER_PTP_V2_L4_DELAY_REQ:
5805         case HWTSTAMP_FILTER_PTP_V2_L2_DELAY_REQ:
5806         case HWTSTAMP_FILTER_PTP_V2_DELAY_REQ:
5807                 /* With V2 type filters which specify a Sync or Delay Request,
5808                  * Path Delay Request/Response messages are also time stamped
5809                  * by hardware so notify the caller the requested packets plus
5810                  * some others are time stamped.
5811                  */
5812                 config.rx_filter = HWTSTAMP_FILTER_SOME;
5813                 break;
5814         default:
5815                 break;
5816         }
5817
5818         return copy_to_user(ifr->ifr_data, &config,
5819                             sizeof(config)) ? -EFAULT : 0;
5820 }
5821
5822 static int e1000e_hwtstamp_get(struct net_device *netdev, struct ifreq *ifr)
5823 {
5824         struct e1000_adapter *adapter = netdev_priv(netdev);
5825
5826         return copy_to_user(ifr->ifr_data, &adapter->hwtstamp_config,
5827                             sizeof(adapter->hwtstamp_config)) ? -EFAULT : 0;
5828 }
5829
5830 static int e1000_ioctl(struct net_device *netdev, struct ifreq *ifr, int cmd)
5831 {
5832         switch (cmd) {
5833         case SIOCGMIIPHY:
5834         case SIOCGMIIREG:
5835         case SIOCSMIIREG:
5836                 return e1000_mii_ioctl(netdev, ifr, cmd);
5837         case SIOCSHWTSTAMP:
5838                 return e1000e_hwtstamp_set(netdev, ifr);
5839         case SIOCGHWTSTAMP:
5840                 return e1000e_hwtstamp_get(netdev, ifr);
5841         default:
5842                 return -EOPNOTSUPP;
5843         }
5844 }
5845
5846 static int e1000_init_phy_wakeup(struct e1000_adapter *adapter, u32 wufc)
5847 {
5848         struct e1000_hw *hw = &adapter->hw;
5849         u32 i, mac_reg;
5850         u16 phy_reg, wuc_enable;
5851         int retval;
5852
5853         /* copy MAC RARs to PHY RARs */
5854         e1000_copy_rx_addrs_to_phy_ich8lan(hw);
5855
5856         retval = hw->phy.ops.acquire(hw);
5857         if (retval) {
5858                 e_err("Could not acquire PHY\n");
5859                 return retval;
5860         }
5861
5862         /* Enable access to wakeup registers on and set page to BM_WUC_PAGE */
5863         retval = e1000_enable_phy_wakeup_reg_access_bm(hw, &wuc_enable);
5864         if (retval)
5865                 goto release;
5866
5867         /* copy MAC MTA to PHY MTA - only needed for pchlan */
5868         for (i = 0; i < adapter->hw.mac.mta_reg_count; i++) {
5869                 mac_reg = E1000_READ_REG_ARRAY(hw, E1000_MTA, i);
5870                 hw->phy.ops.write_reg_page(hw, BM_MTA(i),
5871                                            (u16)(mac_reg & 0xFFFF));
5872                 hw->phy.ops.write_reg_page(hw, BM_MTA(i) + 1,
5873                                            (u16)((mac_reg >> 16) & 0xFFFF));
5874         }
5875
5876         /* configure PHY Rx Control register */
5877         hw->phy.ops.read_reg_page(&adapter->hw, BM_RCTL, &phy_reg);
5878         mac_reg = er32(RCTL);
5879         if (mac_reg & E1000_RCTL_UPE)
5880                 phy_reg |= BM_RCTL_UPE;
5881         if (mac_reg & E1000_RCTL_MPE)
5882                 phy_reg |= BM_RCTL_MPE;
5883         phy_reg &= ~(BM_RCTL_MO_MASK);
5884         if (mac_reg & E1000_RCTL_MO_3)
5885                 phy_reg |= (((mac_reg & E1000_RCTL_MO_3) >> E1000_RCTL_MO_SHIFT)
5886                             << BM_RCTL_MO_SHIFT);
5887         if (mac_reg & E1000_RCTL_BAM)
5888                 phy_reg |= BM_RCTL_BAM;
5889         if (mac_reg & E1000_RCTL_PMCF)
5890                 phy_reg |= BM_RCTL_PMCF;
5891         mac_reg = er32(CTRL);
5892         if (mac_reg & E1000_CTRL_RFCE)
5893                 phy_reg |= BM_RCTL_RFCE;
5894         hw->phy.ops.write_reg_page(&adapter->hw, BM_RCTL, phy_reg);
5895
5896         /* enable PHY wakeup in MAC register */
5897         ew32(WUFC, wufc);
5898         ew32(WUC, E1000_WUC_PHY_WAKE | E1000_WUC_PME_EN);
5899
5900         /* configure and enable PHY wakeup in PHY registers */
5901         hw->phy.ops.write_reg_page(&adapter->hw, BM_WUFC, wufc);
5902         hw->phy.ops.write_reg_page(&adapter->hw, BM_WUC, E1000_WUC_PME_EN);
5903
5904         /* activate PHY wakeup */
5905         wuc_enable |= BM_WUC_ENABLE_BIT | BM_WUC_HOST_WU_BIT;
5906         retval = e1000_disable_phy_wakeup_reg_access_bm(hw, &wuc_enable);
5907         if (retval)
5908                 e_err("Could not set PHY Host Wakeup bit\n");
5909 release:
5910         hw->phy.ops.release(hw);
5911
5912         return retval;
5913 }
5914
5915 static int e1000e_pm_freeze(struct device *dev)
5916 {
5917         struct net_device *netdev = pci_get_drvdata(to_pci_dev(dev));
5918         struct e1000_adapter *adapter = netdev_priv(netdev);
5919
5920         netif_device_detach(netdev);
5921
5922         if (netif_running(netdev)) {
5923                 int count = E1000_CHECK_RESET_COUNT;
5924
5925                 while (test_bit(__E1000_RESETTING, &adapter->state) && count--)
5926                         usleep_range(10000, 20000);
5927
5928                 WARN_ON(test_bit(__E1000_RESETTING, &adapter->state));
5929
5930                 /* Quiesce the device without resetting the hardware */
5931                 e1000e_down(adapter, false);
5932                 e1000_free_irq(adapter);
5933         }
5934         e1000e_reset_interrupt_capability(adapter);
5935
5936         /* Allow time for pending master requests to run */
5937         e1000e_disable_pcie_master(&adapter->hw);
5938
5939         return 0;
5940 }
5941
5942 static int __e1000_shutdown(struct pci_dev *pdev, bool runtime)
5943 {
5944         struct net_device *netdev = pci_get_drvdata(pdev);
5945         struct e1000_adapter *adapter = netdev_priv(netdev);
5946         struct e1000_hw *hw = &adapter->hw;
5947         u32 ctrl, ctrl_ext, rctl, status;
5948         /* Runtime suspend should only enable wakeup for link changes */
5949         u32 wufc = runtime ? E1000_WUFC_LNKC : adapter->wol;
5950         int retval = 0;
5951
5952         status = er32(STATUS);
5953         if (status & E1000_STATUS_LU)
5954                 wufc &= ~E1000_WUFC_LNKC;
5955
5956         if (wufc) {
5957                 e1000_setup_rctl(adapter);
5958                 e1000e_set_rx_mode(netdev);
5959
5960                 /* turn on all-multi mode if wake on multicast is enabled */
5961                 if (wufc & E1000_WUFC_MC) {
5962                         rctl = er32(RCTL);
5963                         rctl |= E1000_RCTL_MPE;
5964                         ew32(RCTL, rctl);
5965                 }
5966
5967                 ctrl = er32(CTRL);
5968                 ctrl |= E1000_CTRL_ADVD3WUC;
5969                 if (!(adapter->flags2 & FLAG2_HAS_PHY_WAKEUP))
5970                         ctrl |= E1000_CTRL_EN_PHY_PWR_MGMT;
5971                 ew32(CTRL, ctrl);
5972
5973                 if (adapter->hw.phy.media_type == e1000_media_type_fiber ||
5974                     adapter->hw.phy.media_type ==
5975                     e1000_media_type_internal_serdes) {
5976                         /* keep the laser running in D3 */
5977                         ctrl_ext = er32(CTRL_EXT);
5978                         ctrl_ext |= E1000_CTRL_EXT_SDP3_DATA;
5979                         ew32(CTRL_EXT, ctrl_ext);
5980                 }
5981
5982                 if (adapter->flags & FLAG_IS_ICH)
5983                         e1000_suspend_workarounds_ich8lan(&adapter->hw);
5984
5985                 if (adapter->flags2 & FLAG2_HAS_PHY_WAKEUP) {
5986                         /* enable wakeup by the PHY */
5987                         retval = e1000_init_phy_wakeup(adapter, wufc);
5988                         if (retval)
5989                                 return retval;
5990                 } else {
5991                         /* enable wakeup by the MAC */
5992                         ew32(WUFC, wufc);
5993                         ew32(WUC, E1000_WUC_PME_EN);
5994                 }
5995         } else {
5996                 ew32(WUC, 0);
5997                 ew32(WUFC, 0);
5998
5999                 e1000_power_down_phy(adapter);
6000         }
6001
6002         if (adapter->hw.phy.type == e1000_phy_igp_3)
6003                 e1000e_igp3_phy_powerdown_workaround_ich8lan(&adapter->hw);
6004
6005         /* Release control of h/w to f/w.  If f/w is AMT enabled, this
6006          * would have already happened in close and is redundant.
6007          */
6008         e1000e_release_hw_control(adapter);
6009
6010         pci_clear_master(pdev);
6011
6012         /* The pci-e switch on some quad port adapters will report a
6013          * correctable error when the MAC transitions from D0 to D3.  To
6014          * prevent this we need to mask off the correctable errors on the
6015          * downstream port of the pci-e switch.
6016          *
6017          * We don't have the associated upstream bridge while assigning
6018          * the PCI device into guest. For example, the KVM on power is
6019          * one of the cases.
6020          */
6021         if (adapter->flags & FLAG_IS_QUAD_PORT) {
6022                 struct pci_dev *us_dev = pdev->bus->self;
6023                 u16 devctl;
6024
6025                 if (!us_dev)
6026                         return 0;
6027
6028                 pcie_capability_read_word(us_dev, PCI_EXP_DEVCTL, &devctl);
6029                 pcie_capability_write_word(us_dev, PCI_EXP_DEVCTL,
6030                                            (devctl & ~PCI_EXP_DEVCTL_CERE));
6031
6032                 pci_save_state(pdev);
6033                 pci_prepare_to_sleep(pdev);
6034
6035                 pcie_capability_write_word(us_dev, PCI_EXP_DEVCTL, devctl);
6036         }
6037
6038         return 0;
6039 }
6040
6041 /**
6042  * e1000e_disable_aspm - Disable ASPM states
6043  * @pdev: pointer to PCI device struct
6044  * @state: bit-mask of ASPM states to disable
6045  *
6046  * Some devices *must* have certain ASPM states disabled per hardware errata.
6047  **/
6048 static void e1000e_disable_aspm(struct pci_dev *pdev, u16 state)
6049 {
6050         struct pci_dev *parent = pdev->bus->self;
6051         u16 aspm_dis_mask = 0;
6052         u16 pdev_aspmc, parent_aspmc;
6053
6054         switch (state) {
6055         case PCIE_LINK_STATE_L0S:
6056         case PCIE_LINK_STATE_L0S | PCIE_LINK_STATE_L1:
6057                 aspm_dis_mask |= PCI_EXP_LNKCTL_ASPM_L0S;
6058                 /* fall-through - can't have L1 without L0s */
6059         case PCIE_LINK_STATE_L1:
6060                 aspm_dis_mask |= PCI_EXP_LNKCTL_ASPM_L1;
6061                 break;
6062         default:
6063                 return;
6064         }
6065
6066         pcie_capability_read_word(pdev, PCI_EXP_LNKCTL, &pdev_aspmc);
6067         pdev_aspmc &= PCI_EXP_LNKCTL_ASPMC;
6068
6069         if (parent) {
6070                 pcie_capability_read_word(parent, PCI_EXP_LNKCTL,
6071                                           &parent_aspmc);
6072                 parent_aspmc &= PCI_EXP_LNKCTL_ASPMC;
6073         }
6074
6075         /* Nothing to do if the ASPM states to be disabled already are */
6076         if (!(pdev_aspmc & aspm_dis_mask) &&
6077             (!parent || !(parent_aspmc & aspm_dis_mask)))
6078                 return;
6079
6080         dev_info(&pdev->dev, "Disabling ASPM %s %s\n",
6081                  (aspm_dis_mask & pdev_aspmc & PCI_EXP_LNKCTL_ASPM_L0S) ?
6082                  "L0s" : "",
6083                  (aspm_dis_mask & pdev_aspmc & PCI_EXP_LNKCTL_ASPM_L1) ?
6084                  "L1" : "");
6085
6086 #ifdef CONFIG_PCIEASPM
6087         pci_disable_link_state_locked(pdev, state);
6088
6089         /* Double-check ASPM control.  If not disabled by the above, the
6090          * BIOS is preventing that from happening (or CONFIG_PCIEASPM is
6091          * not enabled); override by writing PCI config space directly.
6092          */
6093         pcie_capability_read_word(pdev, PCI_EXP_LNKCTL, &pdev_aspmc);
6094         pdev_aspmc &= PCI_EXP_LNKCTL_ASPMC;
6095
6096         if (!(aspm_dis_mask & pdev_aspmc))
6097                 return;
6098 #endif
6099
6100         /* Both device and parent should have the same ASPM setting.
6101          * Disable ASPM in downstream component first and then upstream.
6102          */
6103         pcie_capability_clear_word(pdev, PCI_EXP_LNKCTL, aspm_dis_mask);
6104
6105         if (parent)
6106                 pcie_capability_clear_word(parent, PCI_EXP_LNKCTL,
6107                                            aspm_dis_mask);
6108 }
6109
6110 #ifdef CONFIG_PM
6111 static bool e1000e_pm_ready(struct e1000_adapter *adapter)
6112 {
6113         return !!adapter->tx_ring->buffer_info;
6114 }
6115
6116 static int __e1000_resume(struct pci_dev *pdev)
6117 {
6118         struct net_device *netdev = pci_get_drvdata(pdev);
6119         struct e1000_adapter *adapter = netdev_priv(netdev);
6120         struct e1000_hw *hw = &adapter->hw;
6121         u16 aspm_disable_flag = 0;
6122
6123         if (adapter->flags2 & FLAG2_DISABLE_ASPM_L0S)
6124                 aspm_disable_flag = PCIE_LINK_STATE_L0S;
6125         if (adapter->flags2 & FLAG2_DISABLE_ASPM_L1)
6126                 aspm_disable_flag |= PCIE_LINK_STATE_L1;
6127         if (aspm_disable_flag)
6128                 e1000e_disable_aspm(pdev, aspm_disable_flag);
6129
6130         pci_set_master(pdev);
6131
6132         if (hw->mac.type >= e1000_pch2lan)
6133                 e1000_resume_workarounds_pchlan(&adapter->hw);
6134
6135         e1000e_power_up_phy(adapter);
6136
6137         /* report the system wakeup cause from S3/S4 */
6138         if (adapter->flags2 & FLAG2_HAS_PHY_WAKEUP) {
6139                 u16 phy_data;
6140
6141                 e1e_rphy(&adapter->hw, BM_WUS, &phy_data);
6142                 if (phy_data) {
6143                         e_info("PHY Wakeup cause - %s\n",
6144                                phy_data & E1000_WUS_EX ? "Unicast Packet" :
6145                                phy_data & E1000_WUS_MC ? "Multicast Packet" :
6146                                phy_data & E1000_WUS_BC ? "Broadcast Packet" :
6147                                phy_data & E1000_WUS_MAG ? "Magic Packet" :
6148                                phy_data & E1000_WUS_LNKC ?
6149                                "Link Status Change" : "other");
6150                 }
6151                 e1e_wphy(&adapter->hw, BM_WUS, ~0);
6152         } else {
6153                 u32 wus = er32(WUS);
6154                 if (wus) {
6155                         e_info("MAC Wakeup cause - %s\n",
6156                                wus & E1000_WUS_EX ? "Unicast Packet" :
6157                                wus & E1000_WUS_MC ? "Multicast Packet" :
6158                                wus & E1000_WUS_BC ? "Broadcast Packet" :
6159                                wus & E1000_WUS_MAG ? "Magic Packet" :
6160                                wus & E1000_WUS_LNKC ? "Link Status Change" :
6161                                "other");
6162                 }
6163                 ew32(WUS, ~0);
6164         }
6165
6166         e1000e_reset(adapter);
6167
6168         e1000_init_manageability_pt(adapter);
6169
6170         if (netif_running(netdev))
6171                 e1000e_up(adapter);
6172
6173         netif_device_attach(netdev);
6174
6175         /* If the controller has AMT, do not set DRV_LOAD until the interface
6176          * is up.  For all other cases, let the f/w know that the h/w is now
6177          * under the control of the driver.
6178          */
6179         if (!(adapter->flags & FLAG_HAS_AMT))
6180                 e1000e_get_hw_control(adapter);
6181
6182         return 0;
6183 }
6184
6185 static int e1000e_pm_thaw(struct device *dev)
6186 {
6187         struct net_device *netdev = pci_get_drvdata(to_pci_dev(dev));
6188         struct e1000_adapter *adapter = netdev_priv(netdev);
6189
6190         e1000e_set_interrupt_capability(adapter);
6191         if (netif_running(netdev)) {
6192                 u32 err = e1000_request_irq(adapter);
6193
6194                 if (err)
6195                         return err;
6196
6197                 e1000e_up(adapter);
6198         }
6199
6200         netif_device_attach(netdev);
6201
6202         return 0;
6203 }
6204
6205 #ifdef CONFIG_PM_SLEEP
6206 static int e1000e_pm_suspend(struct device *dev)
6207 {
6208         struct pci_dev *pdev = to_pci_dev(dev);
6209
6210         e1000e_pm_freeze(dev);
6211
6212         return __e1000_shutdown(pdev, false);
6213 }
6214
6215 static int e1000e_pm_resume(struct device *dev)
6216 {
6217         struct pci_dev *pdev = to_pci_dev(dev);
6218         int rc;
6219
6220         rc = __e1000_resume(pdev);
6221         if (rc)
6222                 return rc;
6223
6224         return e1000e_pm_thaw(dev);
6225 }
6226 #endif /* CONFIG_PM_SLEEP */
6227
6228 #ifdef CONFIG_PM_RUNTIME
6229 static int e1000_runtime_suspend(struct device *dev)
6230 {
6231         struct pci_dev *pdev = to_pci_dev(dev);
6232         struct net_device *netdev = pci_get_drvdata(pdev);
6233         struct e1000_adapter *adapter = netdev_priv(netdev);
6234
6235         if (!e1000e_pm_ready(adapter))
6236                 return 0;
6237
6238         return __e1000_shutdown(pdev, true);
6239 }
6240
6241 static int e1000_idle(struct device *dev)
6242 {
6243         struct pci_dev *pdev = to_pci_dev(dev);
6244         struct net_device *netdev = pci_get_drvdata(pdev);
6245         struct e1000_adapter *adapter = netdev_priv(netdev);
6246
6247         if (!e1000e_pm_ready(adapter))
6248                 return 0;
6249
6250         if (adapter->idle_check) {
6251                 adapter->idle_check = false;
6252                 if (!e1000e_has_link(adapter))
6253                         pm_schedule_suspend(dev, MSEC_PER_SEC);
6254         }
6255
6256         return -EBUSY;
6257 }
6258
6259 static int e1000_runtime_resume(struct device *dev)
6260 {
6261         struct pci_dev *pdev = to_pci_dev(dev);
6262         struct net_device *netdev = pci_get_drvdata(pdev);
6263         struct e1000_adapter *adapter = netdev_priv(netdev);
6264
6265         if (!e1000e_pm_ready(adapter))
6266                 return 0;
6267
6268         adapter->idle_check = !dev->power.runtime_auto;
6269         return __e1000_resume(pdev);
6270 }
6271 #endif /* CONFIG_PM_RUNTIME */
6272 #endif /* CONFIG_PM */
6273
6274 static void e1000_shutdown(struct pci_dev *pdev)
6275 {
6276         e1000e_pm_freeze(&pdev->dev);
6277
6278         __e1000_shutdown(pdev, false);
6279 }
6280
6281 #ifdef CONFIG_NET_POLL_CONTROLLER
6282
6283 static irqreturn_t e1000_intr_msix(int __always_unused irq, void *data)
6284 {
6285         struct net_device *netdev = data;
6286         struct e1000_adapter *adapter = netdev_priv(netdev);
6287
6288         if (adapter->msix_entries) {
6289                 int vector, msix_irq;
6290
6291                 vector = 0;
6292                 msix_irq = adapter->msix_entries[vector].vector;
6293                 disable_irq(msix_irq);
6294                 e1000_intr_msix_rx(msix_irq, netdev);
6295                 enable_irq(msix_irq);
6296
6297                 vector++;
6298                 msix_irq = adapter->msix_entries[vector].vector;
6299                 disable_irq(msix_irq);
6300                 e1000_intr_msix_tx(msix_irq, netdev);
6301                 enable_irq(msix_irq);
6302
6303                 vector++;
6304                 msix_irq = adapter->msix_entries[vector].vector;
6305                 disable_irq(msix_irq);
6306                 e1000_msix_other(msix_irq, netdev);
6307                 enable_irq(msix_irq);
6308         }
6309
6310         return IRQ_HANDLED;
6311 }
6312
6313 /**
6314  * e1000_netpoll
6315  * @netdev: network interface device structure
6316  *
6317  * Polling 'interrupt' - used by things like netconsole to send skbs
6318  * without having to re-enable interrupts. It's not called while
6319  * the interrupt routine is executing.
6320  */
6321 static void e1000_netpoll(struct net_device *netdev)
6322 {
6323         struct e1000_adapter *adapter = netdev_priv(netdev);
6324
6325         switch (adapter->int_mode) {
6326         case E1000E_INT_MODE_MSIX:
6327                 e1000_intr_msix(adapter->pdev->irq, netdev);
6328                 break;
6329         case E1000E_INT_MODE_MSI:
6330                 disable_irq(adapter->pdev->irq);
6331                 e1000_intr_msi(adapter->pdev->irq, netdev);
6332                 enable_irq(adapter->pdev->irq);
6333                 break;
6334         default:                /* E1000E_INT_MODE_LEGACY */
6335                 disable_irq(adapter->pdev->irq);
6336                 e1000_intr(adapter->pdev->irq, netdev);
6337                 enable_irq(adapter->pdev->irq);
6338                 break;
6339         }
6340 }
6341 #endif
6342
6343 /**
6344  * e1000_io_error_detected - called when PCI error is detected
6345  * @pdev: Pointer to PCI device
6346  * @state: The current pci connection state
6347  *
6348  * This function is called after a PCI bus error affecting
6349  * this device has been detected.
6350  */
6351 static pci_ers_result_t e1000_io_error_detected(struct pci_dev *pdev,
6352                                                 pci_channel_state_t state)
6353 {
6354         struct net_device *netdev = pci_get_drvdata(pdev);
6355         struct e1000_adapter *adapter = netdev_priv(netdev);
6356
6357         netif_device_detach(netdev);
6358
6359         if (state == pci_channel_io_perm_failure)
6360                 return PCI_ERS_RESULT_DISCONNECT;
6361
6362         if (netif_running(netdev))
6363                 e1000e_down(adapter, true);
6364         pci_disable_device(pdev);
6365
6366         /* Request a slot slot reset. */
6367         return PCI_ERS_RESULT_NEED_RESET;
6368 }
6369
6370 /**
6371  * e1000_io_slot_reset - called after the pci bus has been reset.
6372  * @pdev: Pointer to PCI device
6373  *
6374  * Restart the card from scratch, as if from a cold-boot. Implementation
6375  * resembles the first-half of the e1000e_pm_resume routine.
6376  */
6377 static pci_ers_result_t e1000_io_slot_reset(struct pci_dev *pdev)
6378 {
6379         struct net_device *netdev = pci_get_drvdata(pdev);
6380         struct e1000_adapter *adapter = netdev_priv(netdev);
6381         struct e1000_hw *hw = &adapter->hw;
6382         u16 aspm_disable_flag = 0;
6383         int err;
6384         pci_ers_result_t result;
6385
6386         if (adapter->flags2 & FLAG2_DISABLE_ASPM_L0S)
6387                 aspm_disable_flag = PCIE_LINK_STATE_L0S;
6388         if (adapter->flags2 & FLAG2_DISABLE_ASPM_L1)
6389                 aspm_disable_flag |= PCIE_LINK_STATE_L1;
6390         if (aspm_disable_flag)
6391                 e1000e_disable_aspm(pdev, aspm_disable_flag);
6392
6393         err = pci_enable_device_mem(pdev);
6394         if (err) {
6395                 dev_err(&pdev->dev,
6396                         "Cannot re-enable PCI device after reset.\n");
6397                 result = PCI_ERS_RESULT_DISCONNECT;
6398         } else {
6399                 pdev->state_saved = true;
6400                 pci_restore_state(pdev);
6401                 pci_set_master(pdev);
6402
6403                 pci_enable_wake(pdev, PCI_D3hot, 0);
6404                 pci_enable_wake(pdev, PCI_D3cold, 0);
6405
6406                 e1000e_reset(adapter);
6407                 ew32(WUS, ~0);
6408                 result = PCI_ERS_RESULT_RECOVERED;
6409         }
6410
6411         pci_cleanup_aer_uncorrect_error_status(pdev);
6412
6413         return result;
6414 }
6415
6416 /**
6417  * e1000_io_resume - called when traffic can start flowing again.
6418  * @pdev: Pointer to PCI device
6419  *
6420  * This callback is called when the error recovery driver tells us that
6421  * its OK to resume normal operation. Implementation resembles the
6422  * second-half of the e1000e_pm_resume routine.
6423  */
6424 static void e1000_io_resume(struct pci_dev *pdev)
6425 {
6426         struct net_device *netdev = pci_get_drvdata(pdev);
6427         struct e1000_adapter *adapter = netdev_priv(netdev);
6428
6429         e1000_init_manageability_pt(adapter);
6430
6431         if (netif_running(netdev)) {
6432                 if (e1000e_up(adapter)) {
6433                         dev_err(&pdev->dev,
6434                                 "can't bring device back up after reset\n");
6435                         return;
6436                 }
6437         }
6438
6439         netif_device_attach(netdev);
6440
6441         /* If the controller has AMT, do not set DRV_LOAD until the interface
6442          * is up.  For all other cases, let the f/w know that the h/w is now
6443          * under the control of the driver.
6444          */
6445         if (!(adapter->flags & FLAG_HAS_AMT))
6446                 e1000e_get_hw_control(adapter);
6447 }
6448
6449 static void e1000_print_device_info(struct e1000_adapter *adapter)
6450 {
6451         struct e1000_hw *hw = &adapter->hw;
6452         struct net_device *netdev = adapter->netdev;
6453         u32 ret_val;
6454         u8 pba_str[E1000_PBANUM_LENGTH];
6455
6456         /* print bus type/speed/width info */
6457         e_info("(PCI Express:2.5GT/s:%s) %pM\n",
6458                /* bus width */
6459                ((hw->bus.width == e1000_bus_width_pcie_x4) ? "Width x4" :
6460                 "Width x1"),
6461                /* MAC address */
6462                netdev->dev_addr);
6463         e_info("Intel(R) PRO/%s Network Connection\n",
6464                (hw->phy.type == e1000_phy_ife) ? "10/100" : "1000");
6465         ret_val = e1000_read_pba_string_generic(hw, pba_str,
6466                                                 E1000_PBANUM_LENGTH);
6467         if (ret_val)
6468                 strlcpy((char *)pba_str, "Unknown", sizeof(pba_str));
6469         e_info("MAC: %d, PHY: %d, PBA No: %s\n",
6470                hw->mac.type, hw->phy.type, pba_str);
6471 }
6472
6473 static void e1000_eeprom_checks(struct e1000_adapter *adapter)
6474 {
6475         struct e1000_hw *hw = &adapter->hw;
6476         int ret_val;
6477         u16 buf = 0;
6478
6479         if (hw->mac.type != e1000_82573)
6480                 return;
6481
6482         ret_val = e1000_read_nvm(hw, NVM_INIT_CONTROL2_REG, 1, &buf);
6483         le16_to_cpus(&buf);
6484         if (!ret_val && (!(buf & (1 << 0)))) {
6485                 /* Deep Smart Power Down (DSPD) */
6486                 dev_warn(&adapter->pdev->dev,
6487                          "Warning: detected DSPD enabled in EEPROM\n");
6488         }
6489 }
6490
6491 static int e1000_set_features(struct net_device *netdev,
6492                               netdev_features_t features)
6493 {
6494         struct e1000_adapter *adapter = netdev_priv(netdev);
6495         netdev_features_t changed = features ^ netdev->features;
6496
6497         if (changed & (NETIF_F_TSO | NETIF_F_TSO6))
6498                 adapter->flags |= FLAG_TSO_FORCE;
6499
6500         if (!(changed & (NETIF_F_HW_VLAN_CTAG_RX | NETIF_F_HW_VLAN_CTAG_TX |
6501                          NETIF_F_RXCSUM | NETIF_F_RXHASH | NETIF_F_RXFCS |
6502                          NETIF_F_RXALL)))
6503                 return 0;
6504
6505         if (changed & NETIF_F_RXFCS) {
6506                 if (features & NETIF_F_RXFCS) {
6507                         adapter->flags2 &= ~FLAG2_CRC_STRIPPING;
6508                 } else {
6509                         /* We need to take it back to defaults, which might mean
6510                          * stripping is still disabled at the adapter level.
6511                          */
6512                         if (adapter->flags2 & FLAG2_DFLT_CRC_STRIPPING)
6513                                 adapter->flags2 |= FLAG2_CRC_STRIPPING;
6514                         else
6515                                 adapter->flags2 &= ~FLAG2_CRC_STRIPPING;
6516                 }
6517         }
6518
6519         netdev->features = features;
6520
6521         if (netif_running(netdev))
6522                 e1000e_reinit_locked(adapter);
6523         else
6524                 e1000e_reset(adapter);
6525
6526         return 0;
6527 }
6528
6529 static const struct net_device_ops e1000e_netdev_ops = {
6530         .ndo_open               = e1000_open,
6531         .ndo_stop               = e1000_close,
6532         .ndo_start_xmit         = e1000_xmit_frame,
6533         .ndo_get_stats64        = e1000e_get_stats64,
6534         .ndo_set_rx_mode        = e1000e_set_rx_mode,
6535         .ndo_set_mac_address    = e1000_set_mac,
6536         .ndo_change_mtu         = e1000_change_mtu,
6537         .ndo_do_ioctl           = e1000_ioctl,
6538         .ndo_tx_timeout         = e1000_tx_timeout,
6539         .ndo_validate_addr      = eth_validate_addr,
6540
6541         .ndo_vlan_rx_add_vid    = e1000_vlan_rx_add_vid,
6542         .ndo_vlan_rx_kill_vid   = e1000_vlan_rx_kill_vid,
6543 #ifdef CONFIG_NET_POLL_CONTROLLER
6544         .ndo_poll_controller    = e1000_netpoll,
6545 #endif
6546         .ndo_set_features = e1000_set_features,
6547 };
6548
6549 /**
6550  * e1000_probe - Device Initialization Routine
6551  * @pdev: PCI device information struct
6552  * @ent: entry in e1000_pci_tbl
6553  *
6554  * Returns 0 on success, negative on failure
6555  *
6556  * e1000_probe initializes an adapter identified by a pci_dev structure.
6557  * The OS initialization, configuring of the adapter private structure,
6558  * and a hardware reset occur.
6559  **/
6560 static int e1000_probe(struct pci_dev *pdev, const struct pci_device_id *ent)
6561 {
6562         struct net_device *netdev;
6563         struct e1000_adapter *adapter;
6564         struct e1000_hw *hw;
6565         const struct e1000_info *ei = e1000_info_tbl[ent->driver_data];
6566         resource_size_t mmio_start, mmio_len;
6567         resource_size_t flash_start, flash_len;
6568         static int cards_found;
6569         u16 aspm_disable_flag = 0;
6570         int bars, i, err, pci_using_dac;
6571         u16 eeprom_data = 0;
6572         u16 eeprom_apme_mask = E1000_EEPROM_APME;
6573
6574         if (ei->flags2 & FLAG2_DISABLE_ASPM_L0S)
6575                 aspm_disable_flag = PCIE_LINK_STATE_L0S;
6576         if (ei->flags2 & FLAG2_DISABLE_ASPM_L1)
6577                 aspm_disable_flag |= PCIE_LINK_STATE_L1;
6578         if (aspm_disable_flag)
6579                 e1000e_disable_aspm(pdev, aspm_disable_flag);
6580
6581         err = pci_enable_device_mem(pdev);
6582         if (err)
6583                 return err;
6584
6585         pci_using_dac = 0;
6586         err = dma_set_mask_and_coherent(&pdev->dev, DMA_BIT_MASK(64));
6587         if (!err) {
6588                 pci_using_dac = 1;
6589         } else {
6590                 err = dma_set_mask_and_coherent(&pdev->dev, DMA_BIT_MASK(32));
6591                 if (err) {
6592                         dev_err(&pdev->dev,
6593                                 "No usable DMA configuration, aborting\n");
6594                         goto err_dma;
6595                 }
6596         }
6597
6598         bars = pci_select_bars(pdev, IORESOURCE_MEM);
6599         err = pci_request_selected_regions_exclusive(pdev, bars,
6600                                                      e1000e_driver_name);
6601         if (err)
6602                 goto err_pci_reg;
6603
6604         /* AER (Advanced Error Reporting) hooks */
6605         pci_enable_pcie_error_reporting(pdev);
6606
6607         pci_set_master(pdev);
6608         /* PCI config space info */
6609         err = pci_save_state(pdev);
6610         if (err)
6611                 goto err_alloc_etherdev;
6612
6613         err = -ENOMEM;
6614         netdev = alloc_etherdev(sizeof(struct e1000_adapter));
6615         if (!netdev)
6616                 goto err_alloc_etherdev;
6617
6618         SET_NETDEV_DEV(netdev, &pdev->dev);
6619
6620         netdev->irq = pdev->irq;
6621
6622         pci_set_drvdata(pdev, netdev);
6623         adapter = netdev_priv(netdev);
6624         hw = &adapter->hw;
6625         adapter->netdev = netdev;
6626         adapter->pdev = pdev;
6627         adapter->ei = ei;
6628         adapter->pba = ei->pba;
6629         adapter->flags = ei->flags;
6630         adapter->flags2 = ei->flags2;
6631         adapter->hw.adapter = adapter;
6632         adapter->hw.mac.type = ei->mac;
6633         adapter->max_hw_frame_size = ei->max_hw_frame_size;
6634         adapter->msg_enable = netif_msg_init(debug, DEFAULT_MSG_ENABLE);
6635
6636         mmio_start = pci_resource_start(pdev, 0);
6637         mmio_len = pci_resource_len(pdev, 0);
6638
6639         err = -EIO;
6640         adapter->hw.hw_addr = ioremap(mmio_start, mmio_len);
6641         if (!adapter->hw.hw_addr)
6642                 goto err_ioremap;
6643
6644         if ((adapter->flags & FLAG_HAS_FLASH) &&
6645             (pci_resource_flags(pdev, 1) & IORESOURCE_MEM)) {
6646                 flash_start = pci_resource_start(pdev, 1);
6647                 flash_len = pci_resource_len(pdev, 1);
6648                 adapter->hw.flash_address = ioremap(flash_start, flash_len);
6649                 if (!adapter->hw.flash_address)
6650                         goto err_flashmap;
6651         }
6652
6653         /* Set default EEE advertisement */
6654         if (adapter->flags2 & FLAG2_HAS_EEE)
6655                 adapter->eee_advert = MDIO_EEE_100TX | MDIO_EEE_1000T;
6656
6657         /* construct the net_device struct */
6658         netdev->netdev_ops = &e1000e_netdev_ops;
6659         e1000e_set_ethtool_ops(netdev);
6660         netdev->watchdog_timeo = 5 * HZ;
6661         netif_napi_add(netdev, &adapter->napi, e1000e_poll, 64);
6662         strlcpy(netdev->name, pci_name(pdev), sizeof(netdev->name));
6663
6664         netdev->mem_start = mmio_start;
6665         netdev->mem_end = mmio_start + mmio_len;
6666
6667         adapter->bd_number = cards_found++;
6668
6669         e1000e_check_options(adapter);
6670
6671         /* setup adapter struct */
6672         err = e1000_sw_init(adapter);
6673         if (err)
6674                 goto err_sw_init;
6675
6676         memcpy(&hw->mac.ops, ei->mac_ops, sizeof(hw->mac.ops));
6677         memcpy(&hw->nvm.ops, ei->nvm_ops, sizeof(hw->nvm.ops));
6678         memcpy(&hw->phy.ops, ei->phy_ops, sizeof(hw->phy.ops));
6679
6680         err = ei->get_variants(adapter);
6681         if (err)
6682                 goto err_hw_init;
6683
6684         if ((adapter->flags & FLAG_IS_ICH) &&
6685             (adapter->flags & FLAG_READ_ONLY_NVM))
6686                 e1000e_write_protect_nvm_ich8lan(&adapter->hw);
6687
6688         hw->mac.ops.get_bus_info(&adapter->hw);
6689
6690         adapter->hw.phy.autoneg_wait_to_complete = 0;
6691
6692         /* Copper options */
6693         if (adapter->hw.phy.media_type == e1000_media_type_copper) {
6694                 adapter->hw.phy.mdix = AUTO_ALL_MODES;
6695                 adapter->hw.phy.disable_polarity_correction = 0;
6696                 adapter->hw.phy.ms_type = e1000_ms_hw_default;
6697         }
6698
6699         if (hw->phy.ops.check_reset_block && hw->phy.ops.check_reset_block(hw))
6700                 dev_info(&pdev->dev,
6701                          "PHY reset is blocked due to SOL/IDER session.\n");
6702
6703         /* Set initial default active device features */
6704         netdev->features = (NETIF_F_SG |
6705                             NETIF_F_HW_VLAN_CTAG_RX |
6706                             NETIF_F_HW_VLAN_CTAG_TX |
6707                             NETIF_F_TSO |
6708                             NETIF_F_TSO6 |
6709                             NETIF_F_RXHASH |
6710                             NETIF_F_RXCSUM |
6711                             NETIF_F_HW_CSUM);
6712
6713         /* Set user-changeable features (subset of all device features) */
6714         netdev->hw_features = netdev->features;
6715         netdev->hw_features |= NETIF_F_RXFCS;
6716         netdev->priv_flags |= IFF_SUPP_NOFCS;
6717         netdev->hw_features |= NETIF_F_RXALL;
6718
6719         if (adapter->flags & FLAG_HAS_HW_VLAN_FILTER)
6720                 netdev->features |= NETIF_F_HW_VLAN_CTAG_FILTER;
6721
6722         netdev->vlan_features |= (NETIF_F_SG |
6723                                   NETIF_F_TSO |
6724                                   NETIF_F_TSO6 |
6725                                   NETIF_F_HW_CSUM);
6726
6727         netdev->priv_flags |= IFF_UNICAST_FLT;
6728
6729         if (pci_using_dac) {
6730                 netdev->features |= NETIF_F_HIGHDMA;
6731                 netdev->vlan_features |= NETIF_F_HIGHDMA;
6732         }
6733
6734         if (e1000e_enable_mng_pass_thru(&adapter->hw))
6735                 adapter->flags |= FLAG_MNG_PT_ENABLED;
6736
6737         /* before reading the NVM, reset the controller to
6738          * put the device in a known good starting state
6739          */
6740         adapter->hw.mac.ops.reset_hw(&adapter->hw);
6741
6742         /* systems with ASPM and others may see the checksum fail on the first
6743          * attempt. Let's give it a few tries
6744          */
6745         for (i = 0;; i++) {
6746                 if (e1000_validate_nvm_checksum(&adapter->hw) >= 0)
6747                         break;
6748                 if (i == 2) {
6749                         dev_err(&pdev->dev, "The NVM Checksum Is Not Valid\n");
6750                         err = -EIO;
6751                         goto err_eeprom;
6752                 }
6753         }
6754
6755         e1000_eeprom_checks(adapter);
6756
6757         /* copy the MAC address */
6758         if (e1000e_read_mac_addr(&adapter->hw))
6759                 dev_err(&pdev->dev,
6760                         "NVM Read Error while reading MAC address\n");
6761
6762         memcpy(netdev->dev_addr, adapter->hw.mac.addr, netdev->addr_len);
6763
6764         if (!is_valid_ether_addr(netdev->dev_addr)) {
6765                 dev_err(&pdev->dev, "Invalid MAC Address: %pM\n",
6766                         netdev->dev_addr);
6767                 err = -EIO;
6768                 goto err_eeprom;
6769         }
6770
6771         init_timer(&adapter->watchdog_timer);
6772         adapter->watchdog_timer.function = e1000_watchdog;
6773         adapter->watchdog_timer.data = (unsigned long)adapter;
6774
6775         init_timer(&adapter->phy_info_timer);
6776         adapter->phy_info_timer.function = e1000_update_phy_info;
6777         adapter->phy_info_timer.data = (unsigned long)adapter;
6778
6779         INIT_WORK(&adapter->reset_task, e1000_reset_task);
6780         INIT_WORK(&adapter->watchdog_task, e1000_watchdog_task);
6781         INIT_WORK(&adapter->downshift_task, e1000e_downshift_workaround);
6782         INIT_WORK(&adapter->update_phy_task, e1000e_update_phy_task);
6783         INIT_WORK(&adapter->print_hang_task, e1000_print_hw_hang);
6784
6785         /* Initialize link parameters. User can change them with ethtool */
6786         adapter->hw.mac.autoneg = 1;
6787         adapter->fc_autoneg = true;
6788         adapter->hw.fc.requested_mode = e1000_fc_default;
6789         adapter->hw.fc.current_mode = e1000_fc_default;
6790         adapter->hw.phy.autoneg_advertised = 0x2f;
6791
6792         /* Initial Wake on LAN setting - If APM wake is enabled in
6793          * the EEPROM, enable the ACPI Magic Packet filter
6794          */
6795         if (adapter->flags & FLAG_APME_IN_WUC) {
6796                 /* APME bit in EEPROM is mapped to WUC.APME */
6797                 eeprom_data = er32(WUC);
6798                 eeprom_apme_mask = E1000_WUC_APME;
6799                 if ((hw->mac.type > e1000_ich10lan) &&
6800                     (eeprom_data & E1000_WUC_PHY_WAKE))
6801                         adapter->flags2 |= FLAG2_HAS_PHY_WAKEUP;
6802         } else if (adapter->flags & FLAG_APME_IN_CTRL3) {
6803                 if (adapter->flags & FLAG_APME_CHECK_PORT_B &&
6804                     (adapter->hw.bus.func == 1))
6805                         e1000_read_nvm(&adapter->hw, NVM_INIT_CONTROL3_PORT_B,
6806                                        1, &eeprom_data);
6807                 else
6808                         e1000_read_nvm(&adapter->hw, NVM_INIT_CONTROL3_PORT_A,
6809                                        1, &eeprom_data);
6810         }
6811
6812         /* fetch WoL from EEPROM */
6813         if (eeprom_data & eeprom_apme_mask)
6814                 adapter->eeprom_wol |= E1000_WUFC_MAG;
6815
6816         /* now that we have the eeprom settings, apply the special cases
6817          * where the eeprom may be wrong or the board simply won't support
6818          * wake on lan on a particular port
6819          */
6820         if (!(adapter->flags & FLAG_HAS_WOL))
6821                 adapter->eeprom_wol = 0;
6822
6823         /* initialize the wol settings based on the eeprom settings */
6824         adapter->wol = adapter->eeprom_wol;
6825
6826         /* make sure adapter isn't asleep if manageability is enabled */
6827         if (adapter->wol || (adapter->flags & FLAG_MNG_PT_ENABLED) ||
6828             (hw->mac.ops.check_mng_mode(hw)))
6829                 device_wakeup_enable(&pdev->dev);
6830
6831         /* save off EEPROM version number */
6832         e1000_read_nvm(&adapter->hw, 5, 1, &adapter->eeprom_vers);
6833
6834         /* reset the hardware with the new settings */
6835         e1000e_reset(adapter);
6836
6837         /* If the controller has AMT, do not set DRV_LOAD until the interface
6838          * is up.  For all other cases, let the f/w know that the h/w is now
6839          * under the control of the driver.
6840          */
6841         if (!(adapter->flags & FLAG_HAS_AMT))
6842                 e1000e_get_hw_control(adapter);
6843
6844         strlcpy(netdev->name, "eth%d", sizeof(netdev->name));
6845         err = register_netdev(netdev);
6846         if (err)
6847                 goto err_register;
6848
6849         /* carrier off reporting is important to ethtool even BEFORE open */
6850         netif_carrier_off(netdev);
6851
6852         /* init PTP hardware clock */
6853         e1000e_ptp_init(adapter);
6854
6855         e1000_print_device_info(adapter);
6856
6857         if (pci_dev_run_wake(pdev))
6858                 pm_runtime_put_noidle(&pdev->dev);
6859
6860         return 0;
6861
6862 err_register:
6863         if (!(adapter->flags & FLAG_HAS_AMT))
6864                 e1000e_release_hw_control(adapter);
6865 err_eeprom:
6866         if (hw->phy.ops.check_reset_block && !hw->phy.ops.check_reset_block(hw))
6867                 e1000_phy_hw_reset(&adapter->hw);
6868 err_hw_init:
6869         kfree(adapter->tx_ring);
6870         kfree(adapter->rx_ring);
6871 err_sw_init:
6872         if (adapter->hw.flash_address)
6873                 iounmap(adapter->hw.flash_address);
6874         e1000e_reset_interrupt_capability(adapter);
6875 err_flashmap:
6876         iounmap(adapter->hw.hw_addr);
6877 err_ioremap:
6878         free_netdev(netdev);
6879 err_alloc_etherdev:
6880         pci_release_selected_regions(pdev,
6881                                      pci_select_bars(pdev, IORESOURCE_MEM));
6882 err_pci_reg:
6883 err_dma:
6884         pci_disable_device(pdev);
6885         return err;
6886 }
6887
6888 /**
6889  * e1000_remove - Device Removal Routine
6890  * @pdev: PCI device information struct
6891  *
6892  * e1000_remove is called by the PCI subsystem to alert the driver
6893  * that it should release a PCI device.  The could be caused by a
6894  * Hot-Plug event, or because the driver is going to be removed from
6895  * memory.
6896  **/
6897 static void e1000_remove(struct pci_dev *pdev)
6898 {
6899         struct net_device *netdev = pci_get_drvdata(pdev);
6900         struct e1000_adapter *adapter = netdev_priv(netdev);
6901         bool down = test_bit(__E1000_DOWN, &adapter->state);
6902
6903         e1000e_ptp_remove(adapter);
6904
6905         /* The timers may be rescheduled, so explicitly disable them
6906          * from being rescheduled.
6907          */
6908         if (!down)
6909                 set_bit(__E1000_DOWN, &adapter->state);
6910         del_timer_sync(&adapter->watchdog_timer);
6911         del_timer_sync(&adapter->phy_info_timer);
6912
6913         cancel_work_sync(&adapter->reset_task);
6914         cancel_work_sync(&adapter->watchdog_task);
6915         cancel_work_sync(&adapter->downshift_task);
6916         cancel_work_sync(&adapter->update_phy_task);
6917         cancel_work_sync(&adapter->print_hang_task);
6918
6919         if (adapter->flags & FLAG_HAS_HW_TIMESTAMP) {
6920                 cancel_work_sync(&adapter->tx_hwtstamp_work);
6921                 if (adapter->tx_hwtstamp_skb) {
6922                         dev_kfree_skb_any(adapter->tx_hwtstamp_skb);
6923                         adapter->tx_hwtstamp_skb = NULL;
6924                 }
6925         }
6926
6927         /* Don't lie to e1000_close() down the road. */
6928         if (!down)
6929                 clear_bit(__E1000_DOWN, &adapter->state);
6930         unregister_netdev(netdev);
6931
6932         if (pci_dev_run_wake(pdev))
6933                 pm_runtime_get_noresume(&pdev->dev);
6934
6935         /* Release control of h/w to f/w.  If f/w is AMT enabled, this
6936          * would have already happened in close and is redundant.
6937          */
6938         e1000e_release_hw_control(adapter);
6939
6940         e1000e_reset_interrupt_capability(adapter);
6941         kfree(adapter->tx_ring);
6942         kfree(adapter->rx_ring);
6943
6944         iounmap(adapter->hw.hw_addr);
6945         if (adapter->hw.flash_address)
6946                 iounmap(adapter->hw.flash_address);
6947         pci_release_selected_regions(pdev,
6948                                      pci_select_bars(pdev, IORESOURCE_MEM));
6949
6950         free_netdev(netdev);
6951
6952         /* AER disable */
6953         pci_disable_pcie_error_reporting(pdev);
6954
6955         pci_disable_device(pdev);
6956 }
6957
6958 /* PCI Error Recovery (ERS) */
6959 static const struct pci_error_handlers e1000_err_handler = {
6960         .error_detected = e1000_io_error_detected,
6961         .slot_reset = e1000_io_slot_reset,
6962         .resume = e1000_io_resume,
6963 };
6964
6965 static DEFINE_PCI_DEVICE_TABLE(e1000_pci_tbl) = {
6966         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82571EB_COPPER), board_82571 },
6967         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82571EB_FIBER), board_82571 },
6968         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82571EB_QUAD_COPPER), board_82571 },
6969         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82571EB_QUAD_COPPER_LP),
6970           board_82571 },
6971         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82571EB_QUAD_FIBER), board_82571 },
6972         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82571EB_SERDES), board_82571 },
6973         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82571EB_SERDES_DUAL), board_82571 },
6974         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82571EB_SERDES_QUAD), board_82571 },
6975         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82571PT_QUAD_COPPER), board_82571 },
6976
6977         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82572EI), board_82572 },
6978         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82572EI_COPPER), board_82572 },
6979         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82572EI_FIBER), board_82572 },
6980         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82572EI_SERDES), board_82572 },
6981
6982         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82573E), board_82573 },
6983         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82573E_IAMT), board_82573 },
6984         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82573L), board_82573 },
6985
6986         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82574L), board_82574 },
6987         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82574LA), board_82574 },
6988         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82583V), board_82583 },
6989
6990         { PCI_VDEVICE(INTEL, E1000_DEV_ID_80003ES2LAN_COPPER_DPT),
6991           board_80003es2lan },
6992         { PCI_VDEVICE(INTEL, E1000_DEV_ID_80003ES2LAN_COPPER_SPT),
6993           board_80003es2lan },
6994         { PCI_VDEVICE(INTEL, E1000_DEV_ID_80003ES2LAN_SERDES_DPT),
6995           board_80003es2lan },
6996         { PCI_VDEVICE(INTEL, E1000_DEV_ID_80003ES2LAN_SERDES_SPT),
6997           board_80003es2lan },
6998
6999         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH8_IFE), board_ich8lan },
7000         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH8_IFE_G), board_ich8lan },
7001         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH8_IFE_GT), board_ich8lan },
7002         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH8_IGP_AMT), board_ich8lan },
7003         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH8_IGP_C), board_ich8lan },
7004         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH8_IGP_M), board_ich8lan },
7005         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH8_IGP_M_AMT), board_ich8lan },
7006         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH8_82567V_3), board_ich8lan },
7007
7008         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH9_IFE), board_ich9lan },
7009         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH9_IFE_G), board_ich9lan },
7010         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH9_IFE_GT), board_ich9lan },
7011         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH9_IGP_AMT), board_ich9lan },
7012         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH9_IGP_C), board_ich9lan },
7013         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH9_BM), board_ich9lan },
7014         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH9_IGP_M), board_ich9lan },
7015         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH9_IGP_M_AMT), board_ich9lan },
7016         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH9_IGP_M_V), board_ich9lan },
7017
7018         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH10_R_BM_LM), board_ich9lan },
7019         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH10_R_BM_LF), board_ich9lan },
7020         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH10_R_BM_V), board_ich9lan },
7021
7022         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH10_D_BM_LM), board_ich10lan },
7023         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH10_D_BM_LF), board_ich10lan },
7024         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH10_D_BM_V), board_ich10lan },
7025
7026         { PCI_VDEVICE(INTEL, E1000_DEV_ID_PCH_M_HV_LM), board_pchlan },
7027         { PCI_VDEVICE(INTEL, E1000_DEV_ID_PCH_M_HV_LC), board_pchlan },
7028         { PCI_VDEVICE(INTEL, E1000_DEV_ID_PCH_D_HV_DM), board_pchlan },
7029         { PCI_VDEVICE(INTEL, E1000_DEV_ID_PCH_D_HV_DC), board_pchlan },
7030
7031         { PCI_VDEVICE(INTEL, E1000_DEV_ID_PCH2_LV_LM), board_pch2lan },
7032         { PCI_VDEVICE(INTEL, E1000_DEV_ID_PCH2_LV_V), board_pch2lan },
7033
7034         { PCI_VDEVICE(INTEL, E1000_DEV_ID_PCH_LPT_I217_LM), board_pch_lpt },
7035         { PCI_VDEVICE(INTEL, E1000_DEV_ID_PCH_LPT_I217_V), board_pch_lpt },
7036         { PCI_VDEVICE(INTEL, E1000_DEV_ID_PCH_LPTLP_I218_LM), board_pch_lpt },
7037         { PCI_VDEVICE(INTEL, E1000_DEV_ID_PCH_LPTLP_I218_V), board_pch_lpt },
7038         { PCI_VDEVICE(INTEL, E1000_DEV_ID_PCH_I218_LM2), board_pch_lpt },
7039         { PCI_VDEVICE(INTEL, E1000_DEV_ID_PCH_I218_V2), board_pch_lpt },
7040         { PCI_VDEVICE(INTEL, E1000_DEV_ID_PCH_I218_LM3), board_pch_lpt },
7041         { PCI_VDEVICE(INTEL, E1000_DEV_ID_PCH_I218_V3), board_pch_lpt },
7042
7043         { 0, 0, 0, 0, 0, 0, 0 } /* terminate list */
7044 };
7045 MODULE_DEVICE_TABLE(pci, e1000_pci_tbl);
7046
7047 static const struct dev_pm_ops e1000_pm_ops = {
7048         .suspend        = e1000e_pm_suspend,
7049         .resume         = e1000e_pm_resume,
7050         .freeze         = e1000e_pm_freeze,
7051         .thaw           = e1000e_pm_thaw,
7052         .poweroff       = e1000e_pm_suspend,
7053         .restore        = e1000e_pm_resume,
7054         SET_RUNTIME_PM_OPS(e1000_runtime_suspend, e1000_runtime_resume,
7055                            e1000_idle)
7056 };
7057
7058 /* PCI Device API Driver */
7059 static struct pci_driver e1000_driver = {
7060         .name     = e1000e_driver_name,
7061         .id_table = e1000_pci_tbl,
7062         .probe    = e1000_probe,
7063         .remove   = e1000_remove,
7064         .driver   = {
7065                 .pm = &e1000_pm_ops,
7066         },
7067         .shutdown = e1000_shutdown,
7068         .err_handler = &e1000_err_handler
7069 };
7070
7071 /**
7072  * e1000_init_module - Driver Registration Routine
7073  *
7074  * e1000_init_module is the first routine called when the driver is
7075  * loaded. All it does is register with the PCI subsystem.
7076  **/
7077 static int __init e1000_init_module(void)
7078 {
7079         int ret;
7080         pr_info("Intel(R) PRO/1000 Network Driver - %s\n",
7081                 e1000e_driver_version);
7082         pr_info("Copyright(c) 1999 - 2014 Intel Corporation.\n");
7083         ret = pci_register_driver(&e1000_driver);
7084
7085         return ret;
7086 }
7087 module_init(e1000_init_module);
7088
7089 /**
7090  * e1000_exit_module - Driver Exit Cleanup Routine
7091  *
7092  * e1000_exit_module is called just before the driver is removed
7093  * from memory.
7094  **/
7095 static void __exit e1000_exit_module(void)
7096 {
7097         pci_unregister_driver(&e1000_driver);
7098 }
7099 module_exit(e1000_exit_module);
7100
7101 MODULE_AUTHOR("Intel Corporation, <linux.nics@intel.com>");
7102 MODULE_DESCRIPTION("Intel(R) PRO/1000 Network Driver");
7103 MODULE_LICENSE("GPL");
7104 MODULE_VERSION(DRV_VERSION);
7105
7106 /* netdev.c */