drivers/net/e1000e/netdev.c

   1 /*******************************************************************************
   2
   3   Intel PRO/1000 Linux driver
   4   Copyright(c) 1999 - 2011 Intel Corporation.
   5
   6   This program is free software; you can redistribute it and/or modify it
   7   under the terms and conditions of the GNU General Public License,
   8   version 2, as published by the Free Software Foundation.
   9
  10   This program is distributed in the hope it will be useful, but WITHOUT
  11   ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
  12   FITNESS FOR A PARTICULAR PURPOSE.  See the GNU General Public License for
  13   more details.
  14
  15   You should have received a copy of the GNU General Public License along with
  16   this program; if not, write to the Free Software Foundation, Inc.,
  17   51 Franklin St - Fifth Floor, Boston, MA 02110-1301 USA.
  18
  19   The full GNU General Public License is included in this distribution in
  20   the file called "COPYING".
  21
  22   Contact Information:
  23   Linux NICS <linux.nics@intel.com>
  24   e1000-devel Mailing List <e1000-devel@lists.sourceforge.net>
  25   Intel Corporation, 5200 N.E. Elam Young Parkway, Hillsboro, OR 97124-6497
  26
  27 *******************************************************************************/
  28
  29 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
  30
  31 #include <linux/module.h>
  32 #include <linux/types.h>
  33 #include <linux/init.h>
  34 #include <linux/interrupt.h>
  35 #include <linux/pci.h>
  36 #include <linux/vmalloc.h>
  37 #include <linux/pagemap.h>
  38 #include <linux/delay.h>
  39 #include <linux/netdevice.h>
  40 #include <linux/tcp.h>
  41 #include <linux/ipv6.h>
  42 #include <linux/slab.h>
  43 #include <net/checksum.h>
  44 #include <net/ip6_checksum.h>
  45 #include <linux/mii.h>
  46 #include <linux/ethtool.h>
  47 #include <linux/if_vlan.h>
  48 #include <linux/cpu.h>
  49 #include <linux/smp.h>
  50 #include <linux/pm_qos_params.h>
  51 #include <linux/pm_runtime.h>
  52 #include <linux/aer.h>
  53 #include <linux/prefetch.h>
  54
  55 #include "e1000.h"
  56
  57 #define DRV_EXTRAVERSION "-k"
  58
  59 #define DRV_VERSION "1.3.16" DRV_EXTRAVERSION
  60 char e1000e_driver_name[] = "e1000e";
  61 const char e1000e_driver_version[] = DRV_VERSION;
  62
  63 static void e1000e_disable_aspm(struct pci_dev *pdev, u16 state);
  64
  65 static const struct e1000_info *e1000_info_tbl[] = {
  66         [board_82571]           = &e1000_82571_info,
  67         [board_82572]           = &e1000_82572_info,
  68         [board_82573]           = &e1000_82573_info,
  69         [board_82574]           = &e1000_82574_info,
  70         [board_82583]           = &e1000_82583_info,
  71         [board_80003es2lan]     = &e1000_es2_info,
  72         [board_ich8lan]         = &e1000_ich8_info,
  73         [board_ich9lan]         = &e1000_ich9_info,
  74         [board_ich10lan]        = &e1000_ich10_info,
  75         [board_pchlan]          = &e1000_pch_info,
  76         [board_pch2lan]         = &e1000_pch2_info,
  77 };
  78
  79 struct e1000_reg_info {
  80         u32 ofs;
  81         char *name;
  82 };
  83
  84 #define E1000_RDFH      0x02410 /* Rx Data FIFO Head - RW */
  85 #define E1000_RDFT      0x02418 /* Rx Data FIFO Tail - RW */
  86 #define E1000_RDFHS     0x02420 /* Rx Data FIFO Head Saved - RW */
  87 #define E1000_RDFTS     0x02428 /* Rx Data FIFO Tail Saved - RW */
  88 #define E1000_RDFPC     0x02430 /* Rx Data FIFO Packet Count - RW */
  89
  90 #define E1000_TDFH      0x03410 /* Tx Data FIFO Head - RW */
  91 #define E1000_TDFT      0x03418 /* Tx Data FIFO Tail - RW */
  92 #define E1000_TDFHS     0x03420 /* Tx Data FIFO Head Saved - RW */
  93 #define E1000_TDFTS     0x03428 /* Tx Data FIFO Tail Saved - RW */
  94 #define E1000_TDFPC     0x03430 /* Tx Data FIFO Packet Count - RW */
  95
  96 static const struct e1000_reg_info e1000_reg_info_tbl[] = {
  97
  98         /* General Registers */
  99         {E1000_CTRL, "CTRL"},
 100         {E1000_STATUS, "STATUS"},
 101         {E1000_CTRL_EXT, "CTRL_EXT"},
 102
 103         /* Interrupt Registers */
 104         {E1000_ICR, "ICR"},
 105
 106         /* Rx Registers */
 107         {E1000_RCTL, "RCTL"},
 108         {E1000_RDLEN, "RDLEN"},
 109         {E1000_RDH, "RDH"},
 110         {E1000_RDT, "RDT"},
 111         {E1000_RDTR, "RDTR"},
 112         {E1000_RXDCTL(0), "RXDCTL"},
 113         {E1000_ERT, "ERT"},
 114         {E1000_RDBAL, "RDBAL"},
 115         {E1000_RDBAH, "RDBAH"},
 116         {E1000_RDFH, "RDFH"},
 117         {E1000_RDFT, "RDFT"},
 118         {E1000_RDFHS, "RDFHS"},
 119         {E1000_RDFTS, "RDFTS"},
 120         {E1000_RDFPC, "RDFPC"},
 121
 122         /* Tx Registers */
 123         {E1000_TCTL, "TCTL"},
 124         {E1000_TDBAL, "TDBAL"},
 125         {E1000_TDBAH, "TDBAH"},
 126         {E1000_TDLEN, "TDLEN"},
 127         {E1000_TDH, "TDH"},
 128         {E1000_TDT, "TDT"},
 129         {E1000_TIDV, "TIDV"},
 130         {E1000_TXDCTL(0), "TXDCTL"},
 131         {E1000_TADV, "TADV"},
 132         {E1000_TARC(0), "TARC"},
 133         {E1000_TDFH, "TDFH"},
 134         {E1000_TDFT, "TDFT"},
 135         {E1000_TDFHS, "TDFHS"},
 136         {E1000_TDFTS, "TDFTS"},
 137         {E1000_TDFPC, "TDFPC"},
 138
 139         /* List Terminator */
 140         {}
 141 };
 142
 143 /*
 144  * e1000_regdump - register printout routine
 145  */
 146 static void e1000_regdump(struct e1000_hw *hw, struct e1000_reg_info *reginfo)
 147 {
 148         int n = 0;
 149         char rname[16];
 150         u32 regs[8];
 151
 152         switch (reginfo->ofs) {
 153         case E1000_RXDCTL(0):
 154                 for (n = 0; n < 2; n++)
 155                         regs[n] = __er32(hw, E1000_RXDCTL(n));
 156                 break;
 157         case E1000_TXDCTL(0):
 158                 for (n = 0; n < 2; n++)
 159                         regs[n] = __er32(hw, E1000_TXDCTL(n));
 160                 break;
 161         case E1000_TARC(0):
 162                 for (n = 0; n < 2; n++)
 163                         regs[n] = __er32(hw, E1000_TARC(n));
 164                 break;
 165         default:
 166                 printk(KERN_INFO "%-15s %08x\n",
 167                        reginfo->name, __er32(hw, reginfo->ofs));
 168                 return;
 169         }
 170
 171         snprintf(rname, 16, "%s%s", reginfo->name, "[0-1]");
 172         printk(KERN_INFO "%-15s ", rname);
 173         for (n = 0; n < 2; n++)
 174                 printk(KERN_CONT "%08x ", regs[n]);
 175         printk(KERN_CONT "\n");
 176 }
 177
 178 /*
 179  * e1000e_dump - Print registers, Tx-ring and Rx-ring
 180  */
 181 static void e1000e_dump(struct e1000_adapter *adapter)
 182 {
 183         struct net_device *netdev = adapter->netdev;
 184         struct e1000_hw *hw = &adapter->hw;
 185         struct e1000_reg_info *reginfo;
 186         struct e1000_ring *tx_ring = adapter->tx_ring;
 187         struct e1000_tx_desc *tx_desc;
 188         struct my_u0 {
 189                 u64 a;
 190                 u64 b;
 191         } *u0;
 192         struct e1000_buffer *buffer_info;
 193         struct e1000_ring *rx_ring = adapter->rx_ring;
 194         union e1000_rx_desc_packet_split *rx_desc_ps;
 195         struct e1000_rx_desc *rx_desc;
 196         struct my_u1 {
 197                 u64 a;
 198                 u64 b;
 199                 u64 c;
 200                 u64 d;
 201         } *u1;
 202         u32 staterr;
 203         int i = 0;
 204
 205         if (!netif_msg_hw(adapter))
 206                 return;
 207
 208         /* Print netdevice Info */
 209         if (netdev) {
 210                 dev_info(&adapter->pdev->dev, "Net device Info\n");
 211                 printk(KERN_INFO "Device Name     state            "
 212                        "trans_start      last_rx\n");
 213                 printk(KERN_INFO "%-15s %016lX %016lX %016lX\n",
 214                        netdev->name, netdev->state, netdev->trans_start,
 215                        netdev->last_rx);
 216         }
 217
 218         /* Print Registers */
 219         dev_info(&adapter->pdev->dev, "Register Dump\n");
 220         printk(KERN_INFO " Register Name   Value\n");
 221         for (reginfo = (struct e1000_reg_info *)e1000_reg_info_tbl;
 222              reginfo->name; reginfo++) {
 223                 e1000_regdump(hw, reginfo);
 224         }
 225
 226         /* Print Tx Ring Summary */
 227         if (!netdev || !netif_running(netdev))
 228                 goto exit;
 229
 230         dev_info(&adapter->pdev->dev, "Tx Ring Summary\n");
 231         printk(KERN_INFO "Queue [NTU] [NTC] [bi(ntc)->dma  ]"
 232                " leng ntw timestamp\n");
 233         buffer_info = &tx_ring->buffer_info[tx_ring->next_to_clean];
 234         printk(KERN_INFO " %5d %5X %5X %016llX %04X %3X %016llX\n",
 235                0, tx_ring->next_to_use, tx_ring->next_to_clean,
 236                (unsigned long long)buffer_info->dma,
 237                buffer_info->length,
 238                buffer_info->next_to_watch,
 239                (unsigned long long)buffer_info->time_stamp);
 240
 241         /* Print Tx Ring */
 242         if (!netif_msg_tx_done(adapter))
 243                 goto rx_ring_summary;
 244
 245         dev_info(&adapter->pdev->dev, "Tx Ring Dump\n");
 246
 247         /* Transmit Descriptor Formats - DEXT[29] is 0 (Legacy) or 1 (Extended)
 248          *
 249          * Legacy Transmit Descriptor
 250          *   +--------------------------------------------------------------+
 251          * 0 |         Buffer Address [63:0] (Reserved on Write Back)       |
 252          *   +--------------------------------------------------------------+
 253          * 8 | Special  |    CSS     | Status |  CMD    |  CSO   |  Length  |
 254          *   +--------------------------------------------------------------+
 255          *   63       48 47        36 35    32 31     24 23    16 15        0
 256          *
 257          * Extended Context Descriptor (DTYP=0x0) for TSO or checksum offload
 258          *   63      48 47    40 39       32 31             16 15    8 7      0
 259          *   +----------------------------------------------------------------+
 260          * 0 |  TUCSE  | TUCS0  |   TUCSS   |     IPCSE       | IPCS0 | IPCSS |
 261          *   +----------------------------------------------------------------+
 262          * 8 |   MSS   | HDRLEN | RSV | STA | TUCMD | DTYP |      PAYLEN      |
 263          *   +----------------------------------------------------------------+
 264          *   63      48 47    40 39 36 35 32 31   24 23  20 19                0
 265          *
 266          * Extended Data Descriptor (DTYP=0x1)
 267          *   +----------------------------------------------------------------+
 268          * 0 |                     Buffer Address [63:0]                      |
 269          *   +----------------------------------------------------------------+
 270          * 8 | VLAN tag |  POPTS  | Rsvd | Status | Command | DTYP |  DTALEN  |
 271          *   +----------------------------------------------------------------+
 272          *   63       48 47     40 39  36 35    32 31     24 23  20 19        0
 273          */
 274         printk(KERN_INFO "Tl[desc]     [address 63:0  ] [SpeCssSCmCsLen]"
 275                " [bi->dma       ] leng  ntw timestamp        bi->skb "
 276                "<-- Legacy format\n");
 277         printk(KERN_INFO "Tc[desc]     [Ce CoCsIpceCoS] [MssHlRSCm0Plen]"
 278                " [bi->dma       ] leng  ntw timestamp        bi->skb "
 279                "<-- Ext Context format\n");
 280         printk(KERN_INFO "Td[desc]     [address 63:0  ] [VlaPoRSCm1Dlen]"
 281                " [bi->dma       ] leng  ntw timestamp        bi->skb "
 282                "<-- Ext Data format\n");
 283         for (i = 0; tx_ring->desc && (i < tx_ring->count); i++) {
 284                 tx_desc = E1000_TX_DESC(*tx_ring, i);
 285                 buffer_info = &tx_ring->buffer_info[i];
 286                 u0 = (struct my_u0 *)tx_desc;
 287                 printk(KERN_INFO "T%c[0x%03X]    %016llX %016llX %016llX "
 288                        "%04X  %3X %016llX %p",
 289                        (!(le64_to_cpu(u0->b) & (1 << 29)) ? 'l' :
 290                         ((le64_to_cpu(u0->b) & (1 << 20)) ? 'd' : 'c')), i,
 291                        (unsigned long long)le64_to_cpu(u0->a),
 292                        (unsigned long long)le64_to_cpu(u0->b),
 293                        (unsigned long long)buffer_info->dma,
 294                        buffer_info->length, buffer_info->next_to_watch,
 295                        (unsigned long long)buffer_info->time_stamp,
 296                        buffer_info->skb);
 297                 if (i == tx_ring->next_to_use && i == tx_ring->next_to_clean)
 298                         printk(KERN_CONT " NTC/U\n");
 299                 else if (i == tx_ring->next_to_use)
 300                         printk(KERN_CONT " NTU\n");
 301                 else if (i == tx_ring->next_to_clean)
 302                         printk(KERN_CONT " NTC\n");
 303                 else
 304                         printk(KERN_CONT "\n");
 305
 306                 if (netif_msg_pktdata(adapter) && buffer_info->dma != 0)
 307                         print_hex_dump(KERN_INFO, "", DUMP_PREFIX_ADDRESS,
 308                                        16, 1, phys_to_virt(buffer_info->dma),
 309                                        buffer_info->length, true);
 310         }
 311
 312         /* Print Rx Ring Summary */
 313 rx_ring_summary:
 314         dev_info(&adapter->pdev->dev, "Rx Ring Summary\n");
 315         printk(KERN_INFO "Queue [NTU] [NTC]\n");
 316         printk(KERN_INFO " %5d %5X %5X\n", 0,
 317                rx_ring->next_to_use, rx_ring->next_to_clean);
 318
 319         /* Print Rx Ring */
 320         if (!netif_msg_rx_status(adapter))
 321                 goto exit;
 322
 323         dev_info(&adapter->pdev->dev, "Rx Ring Dump\n");
 324         switch (adapter->rx_ps_pages) {
 325         case 1:
 326         case 2:
 327         case 3:
 328                 /* [Extended] Packet Split Receive Descriptor Format
 329                  *
 330                  *    +-----------------------------------------------------+
 331                  *  0 |                Buffer Address 0 [63:0]              |
 332                  *    +-----------------------------------------------------+
 333                  *  8 |                Buffer Address 1 [63:0]              |
 334                  *    +-----------------------------------------------------+
 335                  * 16 |                Buffer Address 2 [63:0]              |
 336                  *    +-----------------------------------------------------+
 337                  * 24 |                Buffer Address 3 [63:0]              |
 338                  *    +-----------------------------------------------------+
 339                  */
 340                 printk(KERN_INFO "R  [desc]      [buffer 0 63:0 ] "
 341                        "[buffer 1 63:0 ] "
 342                        "[buffer 2 63:0 ] [buffer 3 63:0 ] [bi->dma       ] "
 343                        "[bi->skb] <-- Ext Pkt Split format\n");
 344                 /* [Extended] Receive Descriptor (Write-Back) Format
 345                  *
 346                  *   63       48 47    32 31     13 12    8 7    4 3        0
 347                  *   +------------------------------------------------------+
 348                  * 0 | Packet   | IP     |  Rsvd   | MRQ   | Rsvd | MRQ RSS |
 349                  *   | Checksum | Ident  |         | Queue |      |  Type   |
 350                  *   +------------------------------------------------------+
 351                  * 8 | VLAN Tag | Length | Extended Error | Extended Status |
 352                  *   +------------------------------------------------------+
 353                  *   63       48 47    32 31            20 19               0
 354                  */
 355                 printk(KERN_INFO "RWB[desc]      [ck ipid mrqhsh] "
 356                        "[vl   l0 ee  es] "
 357                        "[ l3  l2  l1 hs] [reserved      ] ---------------- "
 358                        "[bi->skb] <-- Ext Rx Write-Back format\n");
 359                 for (i = 0; i < rx_ring->count; i++) {
 360                         buffer_info = &rx_ring->buffer_info[i];
 361                         rx_desc_ps = E1000_RX_DESC_PS(*rx_ring, i);
 362                         u1 = (struct my_u1 *)rx_desc_ps;
 363                         staterr =
 364                             le32_to_cpu(rx_desc_ps->wb.middle.status_error);
 365                         if (staterr & E1000_RXD_STAT_DD) {
 366                                 /* Descriptor Done */
 367                                 printk(KERN_INFO "RWB[0x%03X]     %016llX "
 368                                        "%016llX %016llX %016llX "
 369                                        "---------------- %p", i,
 370                                        (unsigned long long)le64_to_cpu(u1->a),
 371                                        (unsigned long long)le64_to_cpu(u1->b),
 372                                        (unsigned long long)le64_to_cpu(u1->c),
 373                                        (unsigned long long)le64_to_cpu(u1->d),
 374                                        buffer_info->skb);
 375                         } else {
 376                                 printk(KERN_INFO "R  [0x%03X]     %016llX "
 377                                        "%016llX %016llX %016llX %016llX %p", i,
 378                                        (unsigned long long)le64_to_cpu(u1->a),
 379                                        (unsigned long long)le64_to_cpu(u1->b),
 380                                        (unsigned long long)le64_to_cpu(u1->c),
 381                                        (unsigned long long)le64_to_cpu(u1->d),
 382                                        (unsigned long long)buffer_info->dma,
 383                                        buffer_info->skb);
 384
 385                                 if (netif_msg_pktdata(adapter))
 386                                         print_hex_dump(KERN_INFO, "",
 387                                                 DUMP_PREFIX_ADDRESS, 16, 1,
 388                                                 phys_to_virt(buffer_info->dma),
 389                                                 adapter->rx_ps_bsize0, true);
 390                         }
 391
 392                         if (i == rx_ring->next_to_use)
 393                                 printk(KERN_CONT " NTU\n");
 394                         else if (i == rx_ring->next_to_clean)
 395                                 printk(KERN_CONT " NTC\n");
 396                         else
 397                                 printk(KERN_CONT "\n");
 398                 }
 399                 break;
 400         default:
 401         case 0:
 402                 /* Legacy Receive Descriptor Format
 403                  *
 404                  * +-----------------------------------------------------+
 405                  * |                Buffer Address [63:0]                |
 406                  * +-----------------------------------------------------+
 407                  * | VLAN Tag | Errors | Status 0 | Packet csum | Length |
 408                  * +-----------------------------------------------------+
 409                  * 63       48 47    40 39      32 31         16 15      0
 410                  */
 411                 printk(KERN_INFO "Rl[desc]     [address 63:0  ] "
 412                        "[vl er S cks ln] [bi->dma       ] [bi->skb] "
 413                        "<-- Legacy format\n");
 414                 for (i = 0; rx_ring->desc && (i < rx_ring->count); i++) {
 415                         rx_desc = E1000_RX_DESC(*rx_ring, i);
 416                         buffer_info = &rx_ring->buffer_info[i];
 417                         u0 = (struct my_u0 *)rx_desc;
 418                         printk(KERN_INFO "Rl[0x%03X]    %016llX %016llX "
 419                                "%016llX %p", i,
 420                                (unsigned long long)le64_to_cpu(u0->a),
 421                                (unsigned long long)le64_to_cpu(u0->b),
 422                                (unsigned long long)buffer_info->dma,
 423                                buffer_info->skb);
 424                         if (i == rx_ring->next_to_use)
 425                                 printk(KERN_CONT " NTU\n");
 426                         else if (i == rx_ring->next_to_clean)
 427                                 printk(KERN_CONT " NTC\n");
 428                         else
 429                                 printk(KERN_CONT "\n");
 430
 431                         if (netif_msg_pktdata(adapter))
 432                                 print_hex_dump(KERN_INFO, "",
 433                                                DUMP_PREFIX_ADDRESS,
 434                                                16, 1,
 435                                                phys_to_virt(buffer_info->dma),
 436                                                adapter->rx_buffer_len, true);
 437                 }
 438         }
 439
 440 exit:
 441         return;
 442 }
 443
 444 /**
 445  * e1000_desc_unused - calculate if we have unused descriptors
 446  **/
 447 static int e1000_desc_unused(struct e1000_ring *ring)
 448 {
 449         if (ring->next_to_clean > ring->next_to_use)
 450                 return ring->next_to_clean - ring->next_to_use - 1;
 451
 452         return ring->count + ring->next_to_clean - ring->next_to_use - 1;
 453 }
 454
 455 /**
 456  * e1000_receive_skb - helper function to handle Rx indications
 457  * @adapter: board private structure
 458  * @status: descriptor status field as written by hardware
 459  * @vlan: descriptor vlan field as written by hardware (no le/be conversion)
 460  * @skb: pointer to sk_buff to be indicated to stack
 461  **/
 462 static void e1000_receive_skb(struct e1000_adapter *adapter,
 463                               struct net_device *netdev, struct sk_buff *skb,
 464                               u8 status, __le16 vlan)
 465 {
 466         u16 tag = le16_to_cpu(vlan);
 467         skb->protocol = eth_type_trans(skb, netdev);
 468
 469         if (status & E1000_RXD_STAT_VP)
 470                 __vlan_hwaccel_put_tag(skb, tag);
 471
 472         napi_gro_receive(&adapter->napi, skb);
 473 }
 474
 475 /**
 476  * e1000_rx_checksum - Receive Checksum Offload
 477  * @adapter:     board private structure
 478  * @status_err:  receive descriptor status and error fields
 479  * @csum:       receive descriptor csum field
 480  * @sk_buff:     socket buffer with received data
 481  **/
 482 static void e1000_rx_checksum(struct e1000_adapter *adapter, u32 status_err,
 483                               u32 csum, struct sk_buff *skb)
 484 {
 485         u16 status = (u16)status_err;
 486         u8 errors = (u8)(status_err >> 24);
 487
 488         skb_checksum_none_assert(skb);
 489
 490         /* Ignore Checksum bit is set */
 491         if (status & E1000_RXD_STAT_IXSM)
 492                 return;
 493         /* TCP/UDP checksum error bit is set */
 494         if (errors & E1000_RXD_ERR_TCPE) {
 495                 /* let the stack verify checksum errors */
 496                 adapter->hw_csum_err++;
 497                 return;
 498         }
 499
 500         /* TCP/UDP Checksum has not been calculated */
 501         if (!(status & (E1000_RXD_STAT_TCPCS | E1000_RXD_STAT_UDPCS)))
 502                 return;
 503
 504         /* It must be a TCP or UDP packet with a valid checksum */
 505         if (status & E1000_RXD_STAT_TCPCS) {
 506                 /* TCP checksum is good */
 507                 skb->ip_summed = CHECKSUM_UNNECESSARY;
 508         } else {
 509                 /*
 510                  * IP fragment with UDP payload
 511                  * Hardware complements the payload checksum, so we undo it
 512                  * and then put the value in host order for further stack use.
 513                  */
 514                 __sum16 sum = (__force __sum16)htons(csum);
 515                 skb->csum = csum_unfold(~sum);
 516                 skb->ip_summed = CHECKSUM_COMPLETE;
 517         }
 518         adapter->hw_csum_good++;
 519 }
 520
 521 /**
 522  * e1000_alloc_rx_buffers - Replace used receive buffers; legacy & extended
 523  * @adapter: address of board private structure
 524  **/
 525 static void e1000_alloc_rx_buffers(struct e1000_adapter *adapter,
 526                                    int cleaned_count)
 527 {
 528         struct net_device *netdev = adapter->netdev;
 529         struct pci_dev *pdev = adapter->pdev;
 530         struct e1000_ring *rx_ring = adapter->rx_ring;
 531         struct e1000_rx_desc *rx_desc;
 532         struct e1000_buffer *buffer_info;
 533         struct sk_buff *skb;
 534         unsigned int i;
 535         unsigned int bufsz = adapter->rx_buffer_len;
 536
 537         i = rx_ring->next_to_use;
 538         buffer_info = &rx_ring->buffer_info[i];
 539
 540         while (cleaned_count--) {
 541                 skb = buffer_info->skb;
 542                 if (skb) {
 543                         skb_trim(skb, 0);
 544                         goto map_skb;
 545                 }
 546
 547                 skb = netdev_alloc_skb_ip_align(netdev, bufsz);
 548                 if (!skb) {
 549                         /* Better luck next round */
 550                         adapter->alloc_rx_buff_failed++;
 551                         break;
 552                 }
 553
 554                 buffer_info->skb = skb;
 555 map_skb:
 556                 buffer_info->dma = dma_map_single(&pdev->dev, skb->data,
 557                                                   adapter->rx_buffer_len,
 558                                                   DMA_FROM_DEVICE);
 559                 if (dma_mapping_error(&pdev->dev, buffer_info->dma)) {
 560                         dev_err(&pdev->dev, "Rx DMA map failed\n");
 561                         adapter->rx_dma_failed++;
 562                         break;
 563                 }
 564
 565                 rx_desc = E1000_RX_DESC(*rx_ring, i);
 566                 rx_desc->buffer_addr = cpu_to_le64(buffer_info->dma);
 567
 568                 if (unlikely(!(i & (E1000_RX_BUFFER_WRITE - 1)))) {
 569                         /*
 570                          * Force memory writes to complete before letting h/w
 571                          * know there are new descriptors to fetch.  (Only
 572                          * applicable for weak-ordered memory model archs,
 573                          * such as IA-64).
 574                          */
 575                         wmb();
 576                         writel(i, adapter->hw.hw_addr + rx_ring->tail);
 577                 }
 578                 i++;
 579                 if (i == rx_ring->count)
 580                         i = 0;
 581                 buffer_info = &rx_ring->buffer_info[i];
 582         }
 583
 584         rx_ring->next_to_use = i;
 585 }
 586
 587 /**
 588  * e1000_alloc_rx_buffers_ps - Replace used receive buffers; packet split
 589  * @adapter: address of board private structure
 590  **/
 591 static void e1000_alloc_rx_buffers_ps(struct e1000_adapter *adapter,
 592                                       int cleaned_count)
 593 {
 594         struct net_device *netdev = adapter->netdev;
 595         struct pci_dev *pdev = adapter->pdev;
 596         union e1000_rx_desc_packet_split *rx_desc;
 597         struct e1000_ring *rx_ring = adapter->rx_ring;
 598         struct e1000_buffer *buffer_info;
 599         struct e1000_ps_page *ps_page;
 600         struct sk_buff *skb;
 601         unsigned int i, j;
 602
 603         i = rx_ring->next_to_use;
 604         buffer_info = &rx_ring->buffer_info[i];
 605
 606         while (cleaned_count--) {
 607                 rx_desc = E1000_RX_DESC_PS(*rx_ring, i);
 608
 609                 for (j = 0; j < PS_PAGE_BUFFERS; j++) {
 610                         ps_page = &buffer_info->ps_pages[j];
 611                         if (j >= adapter->rx_ps_pages) {
 612                                 /* all unused desc entries get hw null ptr */
 613                                 rx_desc->read.buffer_addr[j + 1] =
 614                                     ~cpu_to_le64(0);
 615                                 continue;
 616                         }
 617                         if (!ps_page->page) {
 618                                 ps_page->page = alloc_page(GFP_ATOMIC);
 619                                 if (!ps_page->page) {
 620                                         adapter->alloc_rx_buff_failed++;
 621                                         goto no_buffers;
 622                                 }
 623                                 ps_page->dma = dma_map_page(&pdev->dev,
 624                                                             ps_page->page,
 625                                                             0, PAGE_SIZE,
 626                                                             DMA_FROM_DEVICE);
 627                                 if (dma_mapping_error(&pdev->dev,
 628                                                       ps_page->dma)) {
 629                                         dev_err(&adapter->pdev->dev,
 630                                                 "Rx DMA page map failed\n");
 631                                         adapter->rx_dma_failed++;
 632                                         goto no_buffers;
 633                                 }
 634                         }
 635                         /*
 636                          * Refresh the desc even if buffer_addrs
 637                          * didn't change because each write-back
 638                          * erases this info.
 639                          */
 640                         rx_desc->read.buffer_addr[j + 1] =
 641                             cpu_to_le64(ps_page->dma);
 642                 }
 643
 644                 skb = netdev_alloc_skb_ip_align(netdev,
 645                                                 adapter->rx_ps_bsize0);
 646
 647                 if (!skb) {
 648                         adapter->alloc_rx_buff_failed++;
 649                         break;
 650                 }
 651
 652                 buffer_info->skb = skb;
 653                 buffer_info->dma = dma_map_single(&pdev->dev, skb->data,
 654                                                   adapter->rx_ps_bsize0,
 655                                                   DMA_FROM_DEVICE);
 656                 if (dma_mapping_error(&pdev->dev, buffer_info->dma)) {
 657                         dev_err(&pdev->dev, "Rx DMA map failed\n");
 658                         adapter->rx_dma_failed++;
 659                         /* cleanup skb */
 660                         dev_kfree_skb_any(skb);
 661                         buffer_info->skb = NULL;
 662                         break;
 663                 }
 664
 665                 rx_desc->read.buffer_addr[0] = cpu_to_le64(buffer_info->dma);
 666
 667                 if (unlikely(!(i & (E1000_RX_BUFFER_WRITE - 1)))) {
 668                         /*
 669                          * Force memory writes to complete before letting h/w
 670                          * know there are new descriptors to fetch.  (Only
 671                          * applicable for weak-ordered memory model archs,
 672                          * such as IA-64).
 673                          */
 674                         wmb();
 675                         writel(i << 1, adapter->hw.hw_addr + rx_ring->tail);
 676                 }
 677
 678                 i++;
 679                 if (i == rx_ring->count)
 680                         i = 0;
 681                 buffer_info = &rx_ring->buffer_info[i];
 682         }
 683
 684 no_buffers:
 685         rx_ring->next_to_use = i;
 686 }
 687
 688 /**
 689  * e1000_alloc_jumbo_rx_buffers - Replace used jumbo receive buffers
 690  * @adapter: address of board private structure
 691  * @cleaned_count: number of buffers to allocate this pass
 692  **/
 693
 694 static void e1000_alloc_jumbo_rx_buffers(struct e1000_adapter *adapter,
 695                                          int cleaned_count)
 696 {
 697         struct net_device *netdev = adapter->netdev;
 698         struct pci_dev *pdev = adapter->pdev;
 699         struct e1000_rx_desc *rx_desc;
 700         struct e1000_ring *rx_ring = adapter->rx_ring;
 701         struct e1000_buffer *buffer_info;
 702         struct sk_buff *skb;
 703         unsigned int i;
 704         unsigned int bufsz = 256 - 16 /* for skb_reserve */;
 705
 706         i = rx_ring->next_to_use;
 707         buffer_info = &rx_ring->buffer_info[i];
 708
 709         while (cleaned_count--) {
 710                 skb = buffer_info->skb;
 711                 if (skb) {
 712                         skb_trim(skb, 0);
 713                         goto check_page;
 714                 }
 715
 716                 skb = netdev_alloc_skb_ip_align(netdev, bufsz);
 717                 if (unlikely(!skb)) {
 718                         /* Better luck next round */
 719                         adapter->alloc_rx_buff_failed++;
 720                         break;
 721                 }
 722
 723                 buffer_info->skb = skb;
 724 check_page:
 725                 /* allocate a new page if necessary */
 726                 if (!buffer_info->page) {
 727                         buffer_info->page = alloc_page(GFP_ATOMIC);
 728                         if (unlikely(!buffer_info->page)) {
 729                                 adapter->alloc_rx_buff_failed++;
 730                                 break;
 731                         }
 732                 }
 733
 734                 if (!buffer_info->dma)
 735                         buffer_info->dma = dma_map_page(&pdev->dev,
 736                                                         buffer_info->page, 0,
 737                                                         PAGE_SIZE,
 738                                                         DMA_FROM_DEVICE);
 739
 740                 rx_desc = E1000_RX_DESC(*rx_ring, i);
 741                 rx_desc->buffer_addr = cpu_to_le64(buffer_info->dma);
 742
 743                 if (unlikely(++i == rx_ring->count))
 744                         i = 0;
 745                 buffer_info = &rx_ring->buffer_info[i];
 746         }
 747
 748         if (likely(rx_ring->next_to_use != i)) {
 749                 rx_ring->next_to_use = i;
 750                 if (unlikely(i-- == 0))
 751                         i = (rx_ring->count - 1);
 752
 753                 /* Force memory writes to complete before letting h/w
 754                  * know there are new descriptors to fetch.  (Only
 755                  * applicable for weak-ordered memory model archs,
 756                  * such as IA-64). */
 757                 wmb();
 758                 writel(i, adapter->hw.hw_addr + rx_ring->tail);
 759         }
 760 }
 761
 762 /**
 763  * e1000_clean_rx_irq - Send received data up the network stack; legacy
 764  * @adapter: board private structure
 765  *
 766  * the return value indicates whether actual cleaning was done, there
 767  * is no guarantee that everything was cleaned
 768  **/
 769 static bool e1000_clean_rx_irq(struct e1000_adapter *adapter,
 770                                int *work_done, int work_to_do)
 771 {
 772         struct net_device *netdev = adapter->netdev;
 773         struct pci_dev *pdev = adapter->pdev;
 774         struct e1000_hw *hw = &adapter->hw;
 775         struct e1000_ring *rx_ring = adapter->rx_ring;
 776         struct e1000_rx_desc *rx_desc, *next_rxd;
 777         struct e1000_buffer *buffer_info, *next_buffer;
 778         u32 length;
 779         unsigned int i;
 780         int cleaned_count = 0;
 781         bool cleaned = 0;
 782         unsigned int total_rx_bytes = 0, total_rx_packets = 0;
 783
 784         i = rx_ring->next_to_clean;
 785         rx_desc = E1000_RX_DESC(*rx_ring, i);
 786         buffer_info = &rx_ring->buffer_info[i];
 787
 788         while (rx_desc->status & E1000_RXD_STAT_DD) {
 789                 struct sk_buff *skb;
 790                 u8 status;
 791
 792                 if (*work_done >= work_to_do)
 793                         break;
 794                 (*work_done)++;
 795                 rmb();  /* read descriptor and rx_buffer_info after status DD */
 796
 797                 status = rx_desc->status;
 798                 skb = buffer_info->skb;
 799                 buffer_info->skb = NULL;
 800
 801                 prefetch(skb->data - NET_IP_ALIGN);
 802
 803                 i++;
 804                 if (i == rx_ring->count)
 805                         i = 0;
 806                 next_rxd = E1000_RX_DESC(*rx_ring, i);
 807                 prefetch(next_rxd);
 808
 809                 next_buffer = &rx_ring->buffer_info[i];
 810
 811                 cleaned = 1;
 812                 cleaned_count++;
 813                 dma_unmap_single(&pdev->dev,
 814                                  buffer_info->dma,
 815                                  adapter->rx_buffer_len,
 816                                  DMA_FROM_DEVICE);
 817                 buffer_info->dma = 0;
 818
 819                 length = le16_to_cpu(rx_desc->length);
 820
 821                 /*
 822                  * !EOP means multiple descriptors were used to store a single
 823                  * packet, if that's the case we need to toss it.  In fact, we
 824                  * need to toss every packet with the EOP bit clear and the
 825                  * next frame that _does_ have the EOP bit set, as it is by
 826                  * definition only a frame fragment
 827                  */
 828                 if (unlikely(!(status & E1000_RXD_STAT_EOP)))
 829                         adapter->flags2 |= FLAG2_IS_DISCARDING;
 830
 831                 if (adapter->flags2 & FLAG2_IS_DISCARDING) {
 832                         /* All receives must fit into a single buffer */
 833                         e_dbg("Receive packet consumed multiple buffers\n");
 834                         /* recycle */
 835                         buffer_info->skb = skb;
 836                         if (status & E1000_RXD_STAT_EOP)
 837                                 adapter->flags2 &= ~FLAG2_IS_DISCARDING;
 838                         goto next_desc;
 839                 }
 840
 841                 if (rx_desc->errors & E1000_RXD_ERR_FRAME_ERR_MASK) {
 842                         /* recycle */
 843                         buffer_info->skb = skb;
 844                         goto next_desc;
 845                 }
 846
 847                 /* adjust length to remove Ethernet CRC */
 848                 if (!(adapter->flags2 & FLAG2_CRC_STRIPPING))
 849                         length -= 4;
 850
 851                 total_rx_bytes += length;
 852                 total_rx_packets++;
 853
 854                 /*
 855                  * code added for copybreak, this should improve
 856                  * performance for small packets with large amounts
 857                  * of reassembly being done in the stack
 858                  */
 859                 if (length < copybreak) {
 860                         struct sk_buff *new_skb =
 861                             netdev_alloc_skb_ip_align(netdev, length);
 862                         if (new_skb) {
 863                                 skb_copy_to_linear_data_offset(new_skb,
 864                                                                -NET_IP_ALIGN,
 865                                                                (skb->data -
 866                                                                 NET_IP_ALIGN),
 867                                                                (length +
 868                                                                 NET_IP_ALIGN));
 869                                 /* save the skb in buffer_info as good */
 870                                 buffer_info->skb = skb;
 871                                 skb = new_skb;
 872                         }
 873                         /* else just continue with the old one */
 874                 }
 875                 /* end copybreak code */
 876                 skb_put(skb, length);
 877
 878                 /* Receive Checksum Offload */
 879                 e1000_rx_checksum(adapter,
 880                                   (u32)(status) |
 881                                   ((u32)(rx_desc->errors) << 24),
 882                                   le16_to_cpu(rx_desc->csum), skb);
 883
 884                 e1000_receive_skb(adapter, netdev, skb,status,rx_desc->special);
 885
 886 next_desc:
 887                 rx_desc->status = 0;
 888
 889                 /* return some buffers to hardware, one at a time is too slow */
 890                 if (cleaned_count >= E1000_RX_BUFFER_WRITE) {
 891                         adapter->alloc_rx_buf(adapter, cleaned_count);
 892                         cleaned_count = 0;
 893                 }
 894
 895                 /* use prefetched values */
 896                 rx_desc = next_rxd;
 897                 buffer_info = next_buffer;
 898         }
 899         rx_ring->next_to_clean = i;
 900
 901         cleaned_count = e1000_desc_unused(rx_ring);
 902         if (cleaned_count)
 903                 adapter->alloc_rx_buf(adapter, cleaned_count);
 904
 905         adapter->total_rx_bytes += total_rx_bytes;
 906         adapter->total_rx_packets += total_rx_packets;
 907         return cleaned;
 908 }
 909
 910 static void e1000_put_txbuf(struct e1000_adapter *adapter,
 911                              struct e1000_buffer *buffer_info)
 912 {
 913         if (buffer_info->dma) {
 914                 if (buffer_info->mapped_as_page)
 915                         dma_unmap_page(&adapter->pdev->dev, buffer_info->dma,
 916                                        buffer_info->length, DMA_TO_DEVICE);
 917                 else
 918                         dma_unmap_single(&adapter->pdev->dev, buffer_info->dma,
 919                                          buffer_info->length, DMA_TO_DEVICE);
 920                 buffer_info->dma = 0;
 921         }
 922         if (buffer_info->skb) {
 923                 dev_kfree_skb_any(buffer_info->skb);
 924                 buffer_info->skb = NULL;
 925         }
 926         buffer_info->time_stamp = 0;
 927 }
 928
 929 static void e1000_print_hw_hang(struct work_struct *work)
 930 {
 931         struct e1000_adapter *adapter = container_of(work,
 932                                                      struct e1000_adapter,
 933                                                      print_hang_task);
 934         struct e1000_ring *tx_ring = adapter->tx_ring;
 935         unsigned int i = tx_ring->next_to_clean;
 936         unsigned int eop = tx_ring->buffer_info[i].next_to_watch;
 937         struct e1000_tx_desc *eop_desc = E1000_TX_DESC(*tx_ring, eop);
 938         struct e1000_hw *hw = &adapter->hw;
 939         u16 phy_status, phy_1000t_status, phy_ext_status;
 940         u16 pci_status;
 941
 942         if (test_bit(__E1000_DOWN, &adapter->state))
 943                 return;
 944
 945         e1e_rphy(hw, PHY_STATUS, &phy_status);
 946         e1e_rphy(hw, PHY_1000T_STATUS, &phy_1000t_status);
 947         e1e_rphy(hw, PHY_EXT_STATUS, &phy_ext_status);
 948
 949         pci_read_config_word(adapter->pdev, PCI_STATUS, &pci_status);
 950
 951         /* detected Hardware unit hang */
 952         e_err("Detected Hardware Unit Hang:\n"
 953               "  TDH                  <%x>\n"
 954               "  TDT                  <%x>\n"
 955               "  next_to_use          <%x>\n"
 956               "  next_to_clean        <%x>\n"
 957               "buffer_info[next_to_clean]:\n"
 958               "  time_stamp           <%lx>\n"
 959               "  next_to_watch        <%x>\n"
 960               "  jiffies              <%lx>\n"
 961               "  next_to_watch.status <%x>\n"
 962               "MAC Status             <%x>\n"
 963               "PHY Status             <%x>\n"
 964               "PHY 1000BASE-T Status  <%x>\n"
 965               "PHY Extended Status    <%x>\n"
 966               "PCI Status             <%x>\n",
 967               readl(adapter->hw.hw_addr + tx_ring->head),
 968               readl(adapter->hw.hw_addr + tx_ring->tail),
 969               tx_ring->next_to_use,
 970               tx_ring->next_to_clean,
 971               tx_ring->buffer_info[eop].time_stamp,
 972               eop,
 973               jiffies,
 974               eop_desc->upper.fields.status,
 975               er32(STATUS),
 976               phy_status,
 977               phy_1000t_status,
 978               phy_ext_status,
 979               pci_status);
 980 }
 981
 982 /**
 983  * e1000_clean_tx_irq - Reclaim resources after transmit completes
 984  * @adapter: board private structure
 985  *
 986  * the return value indicates whether actual cleaning was done, there
 987  * is no guarantee that everything was cleaned
 988  **/
 989 static bool e1000_clean_tx_irq(struct e1000_adapter *adapter)
 990 {
 991         struct net_device *netdev = adapter->netdev;
 992         struct e1000_hw *hw = &adapter->hw;
 993         struct e1000_ring *tx_ring = adapter->tx_ring;
 994         struct e1000_tx_desc *tx_desc, *eop_desc;
 995         struct e1000_buffer *buffer_info;
 996         unsigned int i, eop;
 997         unsigned int count = 0;
 998         unsigned int total_tx_bytes = 0, total_tx_packets = 0;
 999
1000         i = tx_ring->next_to_clean;
1001         eop = tx_ring->buffer_info[i].next_to_watch;
1002         eop_desc = E1000_TX_DESC(*tx_ring, eop);
1003
1004         while ((eop_desc->upper.data & cpu_to_le32(E1000_TXD_STAT_DD)) &&
1005                (count < tx_ring->count)) {
1006                 bool cleaned = false;
1007                 rmb(); /* read buffer_info after eop_desc */
1008                 for (; !cleaned; count++) {
1009                         tx_desc = E1000_TX_DESC(*tx_ring, i);
1010                         buffer_info = &tx_ring->buffer_info[i];
1011                         cleaned = (i == eop);
1012
1013                         if (cleaned) {
1014                                 total_tx_packets += buffer_info->segs;
1015                                 total_tx_bytes += buffer_info->bytecount;
1016                         }
1017
1018                         e1000_put_txbuf(adapter, buffer_info);
1019                         tx_desc->upper.data = 0;
1020
1021                         i++;
1022                         if (i == tx_ring->count)
1023                                 i = 0;
1024                 }
1025
1026                 if (i == tx_ring->next_to_use)
1027                         break;
1028                 eop = tx_ring->buffer_info[i].next_to_watch;
1029                 eop_desc = E1000_TX_DESC(*tx_ring, eop);
1030         }
1031
1032         tx_ring->next_to_clean = i;
1033
1034 #define TX_WAKE_THRESHOLD 32
1035         if (count && netif_carrier_ok(netdev) &&
1036             e1000_desc_unused(tx_ring) >= TX_WAKE_THRESHOLD) {
1037                 /* Make sure that anybody stopping the queue after this
1038                  * sees the new next_to_clean.
1039                  */
1040                 smp_mb();
1041
1042                 if (netif_queue_stopped(netdev) &&
1043                     !(test_bit(__E1000_DOWN, &adapter->state))) {
1044                         netif_wake_queue(netdev);
1045                         ++adapter->restart_queue;
1046                 }
1047         }
1048
1049         if (adapter->detect_tx_hung) {
1050                 /*
1051                  * Detect a transmit hang in hardware, this serializes the
1052                  * check with the clearing of time_stamp and movement of i
1053                  */
1054                 adapter->detect_tx_hung = 0;
1055                 if (tx_ring->buffer_info[i].time_stamp &&
1056                     time_after(jiffies, tx_ring->buffer_info[i].time_stamp
1057                                + (adapter->tx_timeout_factor * HZ)) &&
1058                     !(er32(STATUS) & E1000_STATUS_TXOFF)) {
1059                         schedule_work(&adapter->print_hang_task);
1060                         netif_stop_queue(netdev);
1061                 }
1062         }
1063         adapter->total_tx_bytes += total_tx_bytes;
1064         adapter->total_tx_packets += total_tx_packets;
1065         return count < tx_ring->count;
1066 }
1067
1068 /**
1069  * e1000_clean_rx_irq_ps - Send received data up the network stack; packet split
1070  * @adapter: board private structure
1071  *
1072  * the return value indicates whether actual cleaning was done, there
1073  * is no guarantee that everything was cleaned
1074  **/
1075 static bool e1000_clean_rx_irq_ps(struct e1000_adapter *adapter,
1076                                   int *work_done, int work_to_do)
1077 {
1078         struct e1000_hw *hw = &adapter->hw;
1079         union e1000_rx_desc_packet_split *rx_desc, *next_rxd;
1080         struct net_device *netdev = adapter->netdev;
1081         struct pci_dev *pdev = adapter->pdev;
1082         struct e1000_ring *rx_ring = adapter->rx_ring;
1083         struct e1000_buffer *buffer_info, *next_buffer;
1084         struct e1000_ps_page *ps_page;
1085         struct sk_buff *skb;
1086         unsigned int i, j;
1087         u32 length, staterr;
1088         int cleaned_count = 0;
1089         bool cleaned = 0;
1090         unsigned int total_rx_bytes = 0, total_rx_packets = 0;
1091
1092         i = rx_ring->next_to_clean;
1093         rx_desc = E1000_RX_DESC_PS(*rx_ring, i);
1094         staterr = le32_to_cpu(rx_desc->wb.middle.status_error);
1095         buffer_info = &rx_ring->buffer_info[i];
1096
1097         while (staterr & E1000_RXD_STAT_DD) {
1098                 if (*work_done >= work_to_do)
1099                         break;
1100                 (*work_done)++;
1101                 skb = buffer_info->skb;
1102                 rmb();  /* read descriptor and rx_buffer_info after status DD */
1103
1104                 /* in the packet split case this is header only */
1105                 prefetch(skb->data - NET_IP_ALIGN);
1106
1107                 i++;
1108                 if (i == rx_ring->count)
1109                         i = 0;
1110                 next_rxd = E1000_RX_DESC_PS(*rx_ring, i);
1111                 prefetch(next_rxd);
1112
1113                 next_buffer = &rx_ring->buffer_info[i];
1114
1115                 cleaned = 1;
1116                 cleaned_count++;
1117                 dma_unmap_single(&pdev->dev, buffer_info->dma,
1118                                  adapter->rx_ps_bsize0, DMA_FROM_DEVICE);
1119                 buffer_info->dma = 0;
1120
1121                 /* see !EOP comment in other Rx routine */
1122                 if (!(staterr & E1000_RXD_STAT_EOP))
1123                         adapter->flags2 |= FLAG2_IS_DISCARDING;
1124
1125                 if (adapter->flags2 & FLAG2_IS_DISCARDING) {
1126                         e_dbg("Packet Split buffers didn't pick up the full "
1127                               "packet\n");
1128                         dev_kfree_skb_irq(skb);
1129                         if (staterr & E1000_RXD_STAT_EOP)
1130                                 adapter->flags2 &= ~FLAG2_IS_DISCARDING;
1131                         goto next_desc;
1132                 }
1133
1134                 if (staterr & E1000_RXDEXT_ERR_FRAME_ERR_MASK) {
1135                         dev_kfree_skb_irq(skb);
1136                         goto next_desc;
1137                 }
1138
1139                 length = le16_to_cpu(rx_desc->wb.middle.length0);
1140
1141                 if (!length) {
1142                         e_dbg("Last part of the packet spanning multiple "
1143                               "descriptors\n");
1144                         dev_kfree_skb_irq(skb);
1145                         goto next_desc;
1146                 }
1147
1148                 /* Good Receive */
1149                 skb_put(skb, length);
1150
1151                 {
1152                 /*
1153                  * this looks ugly, but it seems compiler issues make it
1154                  * more efficient than reusing j
1155                  */
1156                 int l1 = le16_to_cpu(rx_desc->wb.upper.length[0]);
1157
1158                 /*
1159                  * page alloc/put takes too long and effects small packet
1160                  * throughput, so unsplit small packets and save the alloc/put
1161                  * only valid in softirq (napi) context to call kmap_*
1162                  */
1163                 if (l1 && (l1 <= copybreak) &&
1164                     ((length + l1) <= adapter->rx_ps_bsize0)) {
1165                         u8 *vaddr;
1166
1167                         ps_page = &buffer_info->ps_pages[0];
1168
1169                         /*
1170                          * there is no documentation about how to call
1171                          * kmap_atomic, so we can't hold the mapping
1172                          * very long
1173                          */
1174                         dma_sync_single_for_cpu(&pdev->dev, ps_page->dma,
1175                                                 PAGE_SIZE, DMA_FROM_DEVICE);
1176                         vaddr = kmap_atomic(ps_page->page, KM_SKB_DATA_SOFTIRQ);
1177                         memcpy(skb_tail_pointer(skb), vaddr, l1);
1178                         kunmap_atomic(vaddr, KM_SKB_DATA_SOFTIRQ);
1179                         dma_sync_single_for_device(&pdev->dev, ps_page->dma,
1180                                                    PAGE_SIZE, DMA_FROM_DEVICE);
1181
1182                         /* remove the CRC */
1183                         if (!(adapter->flags2 & FLAG2_CRC_STRIPPING))
1184                                 l1 -= 4;
1185
1186                         skb_put(skb, l1);
1187                         goto copydone;
1188                 } /* if */
1189                 }
1190
1191                 for (j = 0; j < PS_PAGE_BUFFERS; j++) {
1192                         length = le16_to_cpu(rx_desc->wb.upper.length[j]);
1193                         if (!length)
1194                                 break;
1195
1196                         ps_page = &buffer_info->ps_pages[j];
1197                         dma_unmap_page(&pdev->dev, ps_page->dma, PAGE_SIZE,
1198                                        DMA_FROM_DEVICE);
1199                         ps_page->dma = 0;
1200                         skb_fill_page_desc(skb, j, ps_page->page, 0, length);
1201                         ps_page->page = NULL;
1202                         skb->len += length;
1203                         skb->data_len += length;
1204                         skb->truesize += length;
1205                 }
1206
1207                 /* strip the ethernet crc, problem is we're using pages now so
1208                  * this whole operation can get a little cpu intensive
1209                  */
1210                 if (!(adapter->flags2 & FLAG2_CRC_STRIPPING))
1211                         pskb_trim(skb, skb->len - 4);
1212
1213 copydone:
1214                 total_rx_bytes += skb->len;
1215                 total_rx_packets++;
1216
1217                 e1000_rx_checksum(adapter, staterr, le16_to_cpu(
1218                         rx_desc->wb.lower.hi_dword.csum_ip.csum), skb);
1219
1220                 if (rx_desc->wb.upper.header_status &
1221                            cpu_to_le16(E1000_RXDPS_HDRSTAT_HDRSP))
1222                         adapter->rx_hdr_split++;
1223
1224                 e1000_receive_skb(adapter, netdev, skb,
1225                                   staterr, rx_desc->wb.middle.vlan);
1226
1227 next_desc:
1228                 rx_desc->wb.middle.status_error &= cpu_to_le32(~0xFF);
1229                 buffer_info->skb = NULL;
1230
1231                 /* return some buffers to hardware, one at a time is too slow */
1232                 if (cleaned_count >= E1000_RX_BUFFER_WRITE) {
1233                         adapter->alloc_rx_buf(adapter, cleaned_count);
1234                         cleaned_count = 0;
1235                 }
1236
1237                 /* use prefetched values */
1238                 rx_desc = next_rxd;
1239                 buffer_info = next_buffer;
1240
1241                 staterr = le32_to_cpu(rx_desc->wb.middle.status_error);
1242         }
1243         rx_ring->next_to_clean = i;
1244
1245         cleaned_count = e1000_desc_unused(rx_ring);
1246         if (cleaned_count)
1247                 adapter->alloc_rx_buf(adapter, cleaned_count);
1248
1249         adapter->total_rx_bytes += total_rx_bytes;
1250         adapter->total_rx_packets += total_rx_packets;
1251         return cleaned;
1252 }
1253
1254 /**
1255  * e1000_consume_page - helper function
1256  **/
1257 static void e1000_consume_page(struct e1000_buffer *bi, struct sk_buff *skb,
1258                                u16 length)
1259 {
1260         bi->page = NULL;
1261         skb->len += length;
1262         skb->data_len += length;
1263         skb->truesize += length;
1264 }
1265
1266 /**
1267  * e1000_clean_jumbo_rx_irq - Send received data up the network stack; legacy
1268  * @adapter: board private structure
1269  *
1270  * the return value indicates whether actual cleaning was done, there
1271  * is no guarantee that everything was cleaned
1272  **/
1273
1274 static bool e1000_clean_jumbo_rx_irq(struct e1000_adapter *adapter,
1275                                      int *work_done, int work_to_do)
1276 {
1277         struct net_device *netdev = adapter->netdev;
1278         struct pci_dev *pdev = adapter->pdev;
1279         struct e1000_ring *rx_ring = adapter->rx_ring;
1280         struct e1000_rx_desc *rx_desc, *next_rxd;
1281         struct e1000_buffer *buffer_info, *next_buffer;
1282         u32 length;
1283         unsigned int i;
1284         int cleaned_count = 0;
1285         bool cleaned = false;
1286         unsigned int total_rx_bytes=0, total_rx_packets=0;
1287
1288         i = rx_ring->next_to_clean;
1289         rx_desc = E1000_RX_DESC(*rx_ring, i);
1290         buffer_info = &rx_ring->buffer_info[i];
1291
1292         while (rx_desc->status & E1000_RXD_STAT_DD) {
1293                 struct sk_buff *skb;
1294                 u8 status;
1295
1296                 if (*work_done >= work_to_do)
1297                         break;
1298                 (*work_done)++;
1299                 rmb();  /* read descriptor and rx_buffer_info after status DD */
1300
1301                 status = rx_desc->status;
1302                 skb = buffer_info->skb;
1303                 buffer_info->skb = NULL;
1304
1305                 ++i;
1306                 if (i == rx_ring->count)
1307                         i = 0;
1308                 next_rxd = E1000_RX_DESC(*rx_ring, i);
1309                 prefetch(next_rxd);
1310
1311                 next_buffer = &rx_ring->buffer_info[i];
1312
1313                 cleaned = true;
1314                 cleaned_count++;
1315                 dma_unmap_page(&pdev->dev, buffer_info->dma, PAGE_SIZE,
1316                                DMA_FROM_DEVICE);
1317                 buffer_info->dma = 0;
1318
1319                 length = le16_to_cpu(rx_desc->length);
1320
1321                 /* errors is only valid for DD + EOP descriptors */
1322                 if (unlikely((status & E1000_RXD_STAT_EOP) &&
1323                     (rx_desc->errors & E1000_RXD_ERR_FRAME_ERR_MASK))) {
1324                                 /* recycle both page and skb */
1325                                 buffer_info->skb = skb;
1326                                 /* an error means any chain goes out the window
1327                                  * too */
1328                                 if (rx_ring->rx_skb_top)
1329                                         dev_kfree_skb_irq(rx_ring->rx_skb_top);
1330                                 rx_ring->rx_skb_top = NULL;
1331                                 goto next_desc;
1332                 }
1333
1334 #define rxtop (rx_ring->rx_skb_top)
1335                 if (!(status & E1000_RXD_STAT_EOP)) {
1336                         /* this descriptor is only the beginning (or middle) */
1337                         if (!rxtop) {
1338                                 /* this is the beginning of a chain */
1339                                 rxtop = skb;
1340                                 skb_fill_page_desc(rxtop, 0, buffer_info->page,
1341                                                    0, length);
1342                         } else {
1343                                 /* this is the middle of a chain */
1344                                 skb_fill_page_desc(rxtop,
1345                                     skb_shinfo(rxtop)->nr_frags,
1346                                     buffer_info->page, 0, length);
1347                                 /* re-use the skb, only consumed the page */
1348                                 buffer_info->skb = skb;
1349                         }
1350                         e1000_consume_page(buffer_info, rxtop, length);
1351                         goto next_desc;
1352                 } else {
1353                         if (rxtop) {
1354                                 /* end of the chain */
1355                                 skb_fill_page_desc(rxtop,
1356                                     skb_shinfo(rxtop)->nr_frags,
1357                                     buffer_info->page, 0, length);
1358                                 /* re-use the current skb, we only consumed the
1359                                  * page */
1360                                 buffer_info->skb = skb;
1361                                 skb = rxtop;
1362                                 rxtop = NULL;
1363                                 e1000_consume_page(buffer_info, skb, length);
1364                         } else {
1365                                 /* no chain, got EOP, this buf is the packet
1366                                  * copybreak to save the put_page/alloc_page */
1367                                 if (length <= copybreak &&
1368                                     skb_tailroom(skb) >= length) {
1369                                         u8 *vaddr;
1370                                         vaddr = kmap_atomic(buffer_info->page,
1371                                                            KM_SKB_DATA_SOFTIRQ);
1372                                         memcpy(skb_tail_pointer(skb), vaddr,
1373                                                length);
1374                                         kunmap_atomic(vaddr,
1375                                                       KM_SKB_DATA_SOFTIRQ);
1376                                         /* re-use the page, so don't erase
1377                                          * buffer_info->page */
1378                                         skb_put(skb, length);
1379                                 } else {
1380                                         skb_fill_page_desc(skb, 0,
1381                                                            buffer_info->page, 0,
1382                                                            length);
1383                                         e1000_consume_page(buffer_info, skb,
1384                                                            length);
1385                                 }
1386                         }
1387                 }
1388
1389                 /* Receive Checksum Offload XXX recompute due to CRC strip? */
1390                 e1000_rx_checksum(adapter,
1391                                   (u32)(status) |
1392                                   ((u32)(rx_desc->errors) << 24),
1393                                   le16_to_cpu(rx_desc->csum), skb);
1394
1395                 /* probably a little skewed due to removing CRC */
1396                 total_rx_bytes += skb->len;
1397                 total_rx_packets++;
1398
1399                 /* eth type trans needs skb->data to point to something */
1400                 if (!pskb_may_pull(skb, ETH_HLEN)) {
1401                         e_err("pskb_may_pull failed.\n");
1402                         dev_kfree_skb_irq(skb);
1403                         goto next_desc;
1404                 }
1405
1406                 e1000_receive_skb(adapter, netdev, skb, status,
1407                                   rx_desc->special);
1408
1409 next_desc:
1410                 rx_desc->status = 0;
1411
1412                 /* return some buffers to hardware, one at a time is too slow */
1413                 if (unlikely(cleaned_count >= E1000_RX_BUFFER_WRITE)) {
1414                         adapter->alloc_rx_buf(adapter, cleaned_count);
1415                         cleaned_count = 0;
1416                 }
1417
1418                 /* use prefetched values */
1419                 rx_desc = next_rxd;
1420                 buffer_info = next_buffer;
1421         }
1422         rx_ring->next_to_clean = i;
1423
1424         cleaned_count = e1000_desc_unused(rx_ring);
1425         if (cleaned_count)
1426                 adapter->alloc_rx_buf(adapter, cleaned_count);
1427
1428         adapter->total_rx_bytes += total_rx_bytes;
1429         adapter->total_rx_packets += total_rx_packets;
1430         return cleaned;
1431 }
1432
1433 /**
1434  * e1000_clean_rx_ring - Free Rx Buffers per Queue
1435  * @adapter: board private structure
1436  **/
1437 static void e1000_clean_rx_ring(struct e1000_adapter *adapter)
1438 {
1439         struct e1000_ring *rx_ring = adapter->rx_ring;
1440         struct e1000_buffer *buffer_info;
1441         struct e1000_ps_page *ps_page;
1442         struct pci_dev *pdev = adapter->pdev;
1443         unsigned int i, j;
1444
1445         /* Free all the Rx ring sk_buffs */
1446         for (i = 0; i < rx_ring->count; i++) {
1447                 buffer_info = &rx_ring->buffer_info[i];
1448                 if (buffer_info->dma) {
1449                         if (adapter->clean_rx == e1000_clean_rx_irq)
1450                                 dma_unmap_single(&pdev->dev, buffer_info->dma,
1451                                                  adapter->rx_buffer_len,
1452                                                  DMA_FROM_DEVICE);
1453                         else if (adapter->clean_rx == e1000_clean_jumbo_rx_irq)
1454                                 dma_unmap_page(&pdev->dev, buffer_info->dma,
1455                                                PAGE_SIZE,
1456                                                DMA_FROM_DEVICE);
1457                         else if (adapter->clean_rx == e1000_clean_rx_irq_ps)
1458                                 dma_unmap_single(&pdev->dev, buffer_info->dma,
1459                                                  adapter->rx_ps_bsize0,
1460                                                  DMA_FROM_DEVICE);
1461                         buffer_info->dma = 0;
1462                 }
1463
1464                 if (buffer_info->page) {
1465                         put_page(buffer_info->page);
1466                         buffer_info->page = NULL;
1467                 }
1468
1469                 if (buffer_info->skb) {
1470                         dev_kfree_skb(buffer_info->skb);
1471                         buffer_info->skb = NULL;
1472                 }
1473
1474                 for (j = 0; j < PS_PAGE_BUFFERS; j++) {
1475                         ps_page = &buffer_info->ps_pages[j];
1476                         if (!ps_page->page)
1477                                 break;
1478                         dma_unmap_page(&pdev->dev, ps_page->dma, PAGE_SIZE,
1479                                        DMA_FROM_DEVICE);
1480                         ps_page->dma = 0;
1481                         put_page(ps_page->page);
1482                         ps_page->page = NULL;
1483                 }
1484         }
1485
1486         /* there also may be some cached data from a chained receive */
1487         if (rx_ring->rx_skb_top) {
1488                 dev_kfree_skb(rx_ring->rx_skb_top);
1489                 rx_ring->rx_skb_top = NULL;
1490         }
1491
1492         /* Zero out the descriptor ring */
1493         memset(rx_ring->desc, 0, rx_ring->size);
1494
1495         rx_ring->next_to_clean = 0;
1496         rx_ring->next_to_use = 0;
1497         adapter->flags2 &= ~FLAG2_IS_DISCARDING;
1498
1499         writel(0, adapter->hw.hw_addr + rx_ring->head);
1500         writel(0, adapter->hw.hw_addr + rx_ring->tail);
1501 }
1502
1503 static void e1000e_downshift_workaround(struct work_struct *work)
1504 {
1505         struct e1000_adapter *adapter = container_of(work,
1506                                         struct e1000_adapter, downshift_task);
1507
1508         if (test_bit(__E1000_DOWN, &adapter->state))
1509                 return;
1510
1511         e1000e_gig_downshift_workaround_ich8lan(&adapter->hw);
1512 }
1513
1514 /**
1515  * e1000_intr_msi - Interrupt Handler
1516  * @irq: interrupt number
1517  * @data: pointer to a network interface device structure
1518  **/
1519 static irqreturn_t e1000_intr_msi(int irq, void *data)
1520 {
1521         struct net_device *netdev = data;
1522         struct e1000_adapter *adapter = netdev_priv(netdev);
1523         struct e1000_hw *hw = &adapter->hw;
1524         u32 icr = er32(ICR);
1525
1526         /*
1527          * read ICR disables interrupts using IAM
1528          */
1529
1530         if (icr & E1000_ICR_LSC) {
1531                 hw->mac.get_link_status = 1;
1532                 /*
1533                  * ICH8 workaround-- Call gig speed drop workaround on cable
1534                  * disconnect (LSC) before accessing any PHY registers
1535                  */
1536                 if ((adapter->flags & FLAG_LSC_GIG_SPEED_DROP) &&
1537                     (!(er32(STATUS) & E1000_STATUS_LU)))
1538                         schedule_work(&adapter->downshift_task);
1539
1540                 /*
1541                  * 80003ES2LAN workaround-- For packet buffer work-around on
1542                  * link down event; disable receives here in the ISR and reset
1543                  * adapter in watchdog
1544                  */
1545                 if (netif_carrier_ok(netdev) &&
1546                     adapter->flags & FLAG_RX_NEEDS_RESTART) {
1547                         /* disable receives */
1548                         u32 rctl = er32(RCTL);
1549                         ew32(RCTL, rctl & ~E1000_RCTL_EN);
1550                         adapter->flags |= FLAG_RX_RESTART_NOW;
1551                 }
1552                 /* guard against interrupt when we're going down */
1553                 if (!test_bit(__E1000_DOWN, &adapter->state))
1554                         mod_timer(&adapter->watchdog_timer, jiffies + 1);
1555         }
1556
1557         if (napi_schedule_prep(&adapter->napi)) {
1558                 adapter->total_tx_bytes = 0;
1559                 adapter->total_tx_packets = 0;
1560                 adapter->total_rx_bytes = 0;
1561                 adapter->total_rx_packets = 0;
1562                 __napi_schedule(&adapter->napi);
1563         }
1564
1565         return IRQ_HANDLED;
1566 }
1567
1568 /**
1569  * e1000_intr - Interrupt Handler
1570  * @irq: interrupt number
1571  * @data: pointer to a network interface device structure
1572  **/
1573 static irqreturn_t e1000_intr(int irq, void *data)
1574 {
1575         struct net_device *netdev = data;
1576         struct e1000_adapter *adapter = netdev_priv(netdev);
1577         struct e1000_hw *hw = &adapter->hw;
1578         u32 rctl, icr = er32(ICR);
1579
1580         if (!icr || test_bit(__E1000_DOWN, &adapter->state))
1581                 return IRQ_NONE;  /* Not our interrupt */
1582
1583         /*
1584          * IMS will not auto-mask if INT_ASSERTED is not set, and if it is
1585          * not set, then the adapter didn't send an interrupt
1586          */
1587         if (!(icr & E1000_ICR_INT_ASSERTED))
1588                 return IRQ_NONE;
1589
1590         /*
1591          * Interrupt Auto-Mask...upon reading ICR,
1592          * interrupts are masked.  No need for the
1593          * IMC write
1594          */
1595
1596         if (icr & E1000_ICR_LSC) {
1597                 hw->mac.get_link_status = 1;
1598                 /*
1599                  * ICH8 workaround-- Call gig speed drop workaround on cable
1600                  * disconnect (LSC) before accessing any PHY registers
1601                  */
1602                 if ((adapter->flags & FLAG_LSC_GIG_SPEED_DROP) &&
1603                     (!(er32(STATUS) & E1000_STATUS_LU)))
1604                         schedule_work(&adapter->downshift_task);
1605
1606                 /*
1607                  * 80003ES2LAN workaround--
1608                  * For packet buffer work-around on link down event;
1609                  * disable receives here in the ISR and
1610                  * reset adapter in watchdog
1611                  */
1612                 if (netif_carrier_ok(netdev) &&
1613                     (adapter->flags & FLAG_RX_NEEDS_RESTART)) {
1614                         /* disable receives */
1615                         rctl = er32(RCTL);
1616                         ew32(RCTL, rctl & ~E1000_RCTL_EN);
1617                         adapter->flags |= FLAG_RX_RESTART_NOW;
1618                 }
1619                 /* guard against interrupt when we're going down */
1620                 if (!test_bit(__E1000_DOWN, &adapter->state))
1621                         mod_timer(&adapter->watchdog_timer, jiffies + 1);
1622         }
1623
1624         if (napi_schedule_prep(&adapter->napi)) {
1625                 adapter->total_tx_bytes = 0;
1626                 adapter->total_tx_packets = 0;
1627                 adapter->total_rx_bytes = 0;
1628                 adapter->total_rx_packets = 0;
1629                 __napi_schedule(&adapter->napi);
1630         }
1631
1632         return IRQ_HANDLED;
1633 }
1634
1635 static irqreturn_t e1000_msix_other(int irq, void *data)
1636 {
1637         struct net_device *netdev = data;
1638         struct e1000_adapter *adapter = netdev_priv(netdev);
1639         struct e1000_hw *hw = &adapter->hw;
1640         u32 icr = er32(ICR);
1641
1642         if (!(icr & E1000_ICR_INT_ASSERTED)) {
1643                 if (!test_bit(__E1000_DOWN, &adapter->state))
1644                         ew32(IMS, E1000_IMS_OTHER);
1645                 return IRQ_NONE;
1646         }
1647
1648         if (icr & adapter->eiac_mask)
1649                 ew32(ICS, (icr & adapter->eiac_mask));
1650
1651         if (icr & E1000_ICR_OTHER) {
1652                 if (!(icr & E1000_ICR_LSC))
1653                         goto no_link_interrupt;
1654                 hw->mac.get_link_status = 1;
1655                 /* guard against interrupt when we're going down */
1656                 if (!test_bit(__E1000_DOWN, &adapter->state))
1657                         mod_timer(&adapter->watchdog_timer, jiffies + 1);
1658         }
1659
1660 no_link_interrupt:
1661         if (!test_bit(__E1000_DOWN, &adapter->state))
1662                 ew32(IMS, E1000_IMS_LSC | E1000_IMS_OTHER);
1663
1664         return IRQ_HANDLED;
1665 }
1666
1667
1668 static irqreturn_t e1000_intr_msix_tx(int irq, void *data)
1669 {
1670         struct net_device *netdev = data;
1671         struct e1000_adapter *adapter = netdev_priv(netdev);
1672         struct e1000_hw *hw = &adapter->hw;
1673         struct e1000_ring *tx_ring = adapter->tx_ring;
1674
1675
1676         adapter->total_tx_bytes = 0;
1677         adapter->total_tx_packets = 0;
1678
1679         if (!e1000_clean_tx_irq(adapter))
1680                 /* Ring was not completely cleaned, so fire another interrupt */
1681                 ew32(ICS, tx_ring->ims_val);
1682
1683         return IRQ_HANDLED;
1684 }
1685
1686 static irqreturn_t e1000_intr_msix_rx(int irq, void *data)
1687 {
1688         struct net_device *netdev = data;
1689         struct e1000_adapter *adapter = netdev_priv(netdev);
1690
1691         /* Write the ITR value calculated at the end of the
1692          * previous interrupt.
1693          */
1694         if (adapter->rx_ring->set_itr) {
1695                 writel(1000000000 / (adapter->rx_ring->itr_val * 256),
1696                        adapter->hw.hw_addr + adapter->rx_ring->itr_register);
1697                 adapter->rx_ring->set_itr = 0;
1698         }
1699
1700         if (napi_schedule_prep(&adapter->napi)) {
1701                 adapter->total_rx_bytes = 0;
1702                 adapter->total_rx_packets = 0;
1703                 __napi_schedule(&adapter->napi);
1704         }
1705         return IRQ_HANDLED;
1706 }
1707
1708 /**
1709  * e1000_configure_msix - Configure MSI-X hardware
1710  *
1711  * e1000_configure_msix sets up the hardware to properly
1712  * generate MSI-X interrupts.
1713  **/
1714 static void e1000_configure_msix(struct e1000_adapter *adapter)
1715 {
1716         struct e1000_hw *hw = &adapter->hw;
1717         struct e1000_ring *rx_ring = adapter->rx_ring;
1718         struct e1000_ring *tx_ring = adapter->tx_ring;
1719         int vector = 0;
1720         u32 ctrl_ext, ivar = 0;
1721
1722         adapter->eiac_mask = 0;
1723
1724         /* Workaround issue with spurious interrupts on 82574 in MSI-X mode */
1725         if (hw->mac.type == e1000_82574) {
1726                 u32 rfctl = er32(RFCTL);
1727                 rfctl |= E1000_RFCTL_ACK_DIS;
1728                 ew32(RFCTL, rfctl);
1729         }
1730
1731 #define E1000_IVAR_INT_ALLOC_VALID      0x8
1732         /* Configure Rx vector */
1733         rx_ring->ims_val = E1000_IMS_RXQ0;
1734         adapter->eiac_mask |= rx_ring->ims_val;
1735         if (rx_ring->itr_val)
1736                 writel(1000000000 / (rx_ring->itr_val * 256),
1737                        hw->hw_addr + rx_ring->itr_register);
1738         else
1739                 writel(1, hw->hw_addr + rx_ring->itr_register);
1740         ivar = E1000_IVAR_INT_ALLOC_VALID | vector;
1741
1742         /* Configure Tx vector */
1743         tx_ring->ims_val = E1000_IMS_TXQ0;
1744         vector++;
1745         if (tx_ring->itr_val)
1746                 writel(1000000000 / (tx_ring->itr_val * 256),
1747                        hw->hw_addr + tx_ring->itr_register);
1748         else
1749                 writel(1, hw->hw_addr + tx_ring->itr_register);
1750         adapter->eiac_mask |= tx_ring->ims_val;
1751         ivar |= ((E1000_IVAR_INT_ALLOC_VALID | vector) << 8);
1752
1753         /* set vector for Other Causes, e.g. link changes */
1754         vector++;
1755         ivar |= ((E1000_IVAR_INT_ALLOC_VALID | vector) << 16);
1756         if (rx_ring->itr_val)
1757                 writel(1000000000 / (rx_ring->itr_val * 256),
1758                        hw->hw_addr + E1000_EITR_82574(vector));
1759         else
1760                 writel(1, hw->hw_addr + E1000_EITR_82574(vector));
1761
1762         /* Cause Tx interrupts on every write back */
1763         ivar |= (1 << 31);
1764
1765         ew32(IVAR, ivar);
1766
1767         /* enable MSI-X PBA support */
1768         ctrl_ext = er32(CTRL_EXT);
1769         ctrl_ext |= E1000_CTRL_EXT_PBA_CLR;
1770
1771         /* Auto-Mask Other interrupts upon ICR read */
1772 #define E1000_EIAC_MASK_82574   0x01F00000
1773         ew32(IAM, ~E1000_EIAC_MASK_82574 | E1000_IMS_OTHER);
1774         ctrl_ext |= E1000_CTRL_EXT_EIAME;
1775         ew32(CTRL_EXT, ctrl_ext);
1776         e1e_flush();
1777 }
1778
1779 void e1000e_reset_interrupt_capability(struct e1000_adapter *adapter)
1780 {
1781         if (adapter->msix_entries) {
1782                 pci_disable_msix(adapter->pdev);
1783                 kfree(adapter->msix_entries);
1784                 adapter->msix_entries = NULL;
1785         } else if (adapter->flags & FLAG_MSI_ENABLED) {
1786                 pci_disable_msi(adapter->pdev);
1787                 adapter->flags &= ~FLAG_MSI_ENABLED;
1788         }
1789 }
1790
1791 /**
1792  * e1000e_set_interrupt_capability - set MSI or MSI-X if supported
1793  *
1794  * Attempt to configure interrupts using the best available
1795  * capabilities of the hardware and kernel.
1796  **/
1797 void e1000e_set_interrupt_capability(struct e1000_adapter *adapter)
1798 {
1799         int err;
1800         int i;
1801
1802         switch (adapter->int_mode) {
1803         case E1000E_INT_MODE_MSIX:
1804                 if (adapter->flags & FLAG_HAS_MSIX) {
1805                         adapter->num_vectors = 3; /* RxQ0, TxQ0 and other */
1806                         adapter->msix_entries = kcalloc(adapter->num_vectors,
1807                                                       sizeof(struct msix_entry),
1808                                                       GFP_KERNEL);
1809                         if (adapter->msix_entries) {
1810                                 for (i = 0; i < adapter->num_vectors; i++)
1811                                         adapter->msix_entries[i].entry = i;
1812
1813                                 err = pci_enable_msix(adapter->pdev,
1814                                                       adapter->msix_entries,
1815                                                       adapter->num_vectors);
1816                                 if (err == 0)
1817                                         return;
1818                         }
1819                         /* MSI-X failed, so fall through and try MSI */
1820                         e_err("Failed to initialize MSI-X interrupts.  "
1821                               "Falling back to MSI interrupts.\n");
1822                         e1000e_reset_interrupt_capability(adapter);
1823                 }
1824                 adapter->int_mode = E1000E_INT_MODE_MSI;
1825                 /* Fall through */
1826         case E1000E_INT_MODE_MSI:
1827                 if (!pci_enable_msi(adapter->pdev)) {
1828                         adapter->flags |= FLAG_MSI_ENABLED;
1829                 } else {
1830                         adapter->int_mode = E1000E_INT_MODE_LEGACY;
1831                         e_err("Failed to initialize MSI interrupts.  Falling "
1832                               "back to legacy interrupts.\n");
1833                 }
1834                 /* Fall through */
1835         case E1000E_INT_MODE_LEGACY:
1836                 /* Don't do anything; this is the system default */
1837                 break;
1838         }
1839
1840         /* store the number of vectors being used */
1841         adapter->num_vectors = 1;
1842 }
1843
1844 /**
1845  * e1000_request_msix - Initialize MSI-X interrupts
1846  *
1847  * e1000_request_msix allocates MSI-X vectors and requests interrupts from the
1848  * kernel.
1849  **/
1850 static int e1000_request_msix(struct e1000_adapter *adapter)
1851 {
1852         struct net_device *netdev = adapter->netdev;
1853         int err = 0, vector = 0;
1854
1855         if (strlen(netdev->name) < (IFNAMSIZ - 5))
1856                 snprintf(adapter->rx_ring->name,
1857                          sizeof(adapter->rx_ring->name) - 1,
1858                          "%s-rx-0", netdev->name);
1859         else
1860                 memcpy(adapter->rx_ring->name, netdev->name, IFNAMSIZ);
1861         err = request_irq(adapter->msix_entries[vector].vector,
1862                           e1000_intr_msix_rx, 0, adapter->rx_ring->name,
1863                           netdev);
1864         if (err)
1865                 goto out;
1866         adapter->rx_ring->itr_register = E1000_EITR_82574(vector);
1867         adapter->rx_ring->itr_val = adapter->itr;
1868         vector++;
1869
1870         if (strlen(netdev->name) < (IFNAMSIZ - 5))
1871                 snprintf(adapter->tx_ring->name,
1872                          sizeof(adapter->tx_ring->name) - 1,
1873                          "%s-tx-0", netdev->name);
1874         else
1875                 memcpy(adapter->tx_ring->name, netdev->name, IFNAMSIZ);
1876         err = request_irq(adapter->msix_entries[vector].vector,
1877                           e1000_intr_msix_tx, 0, adapter->tx_ring->name,
1878                           netdev);
1879         if (err)
1880                 goto out;
1881         adapter->tx_ring->itr_register = E1000_EITR_82574(vector);
1882         adapter->tx_ring->itr_val = adapter->itr;
1883         vector++;
1884
1885         err = request_irq(adapter->msix_entries[vector].vector,
1886                           e1000_msix_other, 0, netdev->name, netdev);
1887         if (err)
1888                 goto out;
1889
1890         e1000_configure_msix(adapter);
1891         return 0;
1892 out:
1893         return err;
1894 }
1895
1896 /**
1897  * e1000_request_irq - initialize interrupts
1898  *
1899  * Attempts to configure interrupts using the best available
1900  * capabilities of the hardware and kernel.
1901  **/
1902 static int e1000_request_irq(struct e1000_adapter *adapter)
1903 {
1904         struct net_device *netdev = adapter->netdev;
1905         int err;
1906
1907         if (adapter->msix_entries) {
1908                 err = e1000_request_msix(adapter);
1909                 if (!err)
1910                         return err;
1911                 /* fall back to MSI */
1912                 e1000e_reset_interrupt_capability(adapter);
1913                 adapter->int_mode = E1000E_INT_MODE_MSI;
1914                 e1000e_set_interrupt_capability(adapter);
1915         }
1916         if (adapter->flags & FLAG_MSI_ENABLED) {
1917                 err = request_irq(adapter->pdev->irq, e1000_intr_msi, 0,
1918                                   netdev->name, netdev);
1919                 if (!err)
1920                         return err;
1921
1922                 /* fall back to legacy interrupt */
1923                 e1000e_reset_interrupt_capability(adapter);
1924                 adapter->int_mode = E1000E_INT_MODE_LEGACY;
1925         }
1926
1927         err = request_irq(adapter->pdev->irq, e1000_intr, IRQF_SHARED,
1928                           netdev->name, netdev);
1929         if (err)
1930                 e_err("Unable to allocate interrupt, Error: %d\n", err);
1931
1932         return err;
1933 }
1934
1935 static void e1000_free_irq(struct e1000_adapter *adapter)
1936 {
1937         struct net_device *netdev = adapter->netdev;
1938
1939         if (adapter->msix_entries) {
1940                 int vector = 0;
1941
1942                 free_irq(adapter->msix_entries[vector].vector, netdev);
1943                 vector++;
1944
1945                 free_irq(adapter->msix_entries[vector].vector, netdev);
1946                 vector++;
1947
1948                 /* Other Causes interrupt vector */
1949                 free_irq(adapter->msix_entries[vector].vector, netdev);
1950                 return;
1951         }
1952
1953         free_irq(adapter->pdev->irq, netdev);
1954 }
1955
1956 /**
1957  * e1000_irq_disable - Mask off interrupt generation on the NIC
1958  **/
1959 static void e1000_irq_disable(struct e1000_adapter *adapter)
1960 {
1961         struct e1000_hw *hw = &adapter->hw;
1962
1963         ew32(IMC, ~0);
1964         if (adapter->msix_entries)
1965                 ew32(EIAC_82574, 0);
1966         e1e_flush();
1967
1968         if (adapter->msix_entries) {
1969                 int i;
1970                 for (i = 0; i < adapter->num_vectors; i++)
1971                         synchronize_irq(adapter->msix_entries[i].vector);
1972         } else {
1973                 synchronize_irq(adapter->pdev->irq);
1974         }
1975 }
1976
1977 /**
1978  * e1000_irq_enable - Enable default interrupt generation settings
1979  **/
1980 static void e1000_irq_enable(struct e1000_adapter *adapter)
1981 {
1982         struct e1000_hw *hw = &adapter->hw;
1983
1984         if (adapter->msix_entries) {
1985                 ew32(EIAC_82574, adapter->eiac_mask & E1000_EIAC_MASK_82574);
1986                 ew32(IMS, adapter->eiac_mask | E1000_IMS_OTHER | E1000_IMS_LSC);
1987         } else {
1988                 ew32(IMS, IMS_ENABLE_MASK);
1989         }
1990         e1e_flush();
1991 }
1992
1993 /**
1994  * e1000e_get_hw_control - get control of the h/w from f/w
1995  * @adapter: address of board private structure
1996  *
1997  * e1000e_get_hw_control sets {CTRL_EXT|SWSM}:DRV_LOAD bit.
1998  * For ASF and Pass Through versions of f/w this means that
1999  * the driver is loaded. For AMT version (only with 82573)
2000  * of the f/w this means that the network i/f is open.
2001  **/
2002 void e1000e_get_hw_control(struct e1000_adapter *adapter)
2003 {
2004         struct e1000_hw *hw = &adapter->hw;
2005         u32 ctrl_ext;
2006         u32 swsm;
2007
2008         /* Let firmware know the driver has taken over */
2009         if (adapter->flags & FLAG_HAS_SWSM_ON_LOAD) {
2010                 swsm = er32(SWSM);
2011                 ew32(SWSM, swsm | E1000_SWSM_DRV_LOAD);
2012         } else if (adapter->flags & FLAG_HAS_CTRLEXT_ON_LOAD) {
2013                 ctrl_ext = er32(CTRL_EXT);
2014                 ew32(CTRL_EXT, ctrl_ext | E1000_CTRL_EXT_DRV_LOAD);
2015         }
2016 }
2017
2018 /**
2019  * e1000e_release_hw_control - release control of the h/w to f/w
2020  * @adapter: address of board private structure
2021  *
2022  * e1000e_release_hw_control resets {CTRL_EXT|SWSM}:DRV_LOAD bit.
2023  * For ASF and Pass Through versions of f/w this means that the
2024  * driver is no longer loaded. For AMT version (only with 82573) i
2025  * of the f/w this means that the network i/f is closed.
2026  *
2027  **/
2028 void e1000e_release_hw_control(struct e1000_adapter *adapter)
2029 {
2030         struct e1000_hw *hw = &adapter->hw;
2031         u32 ctrl_ext;
2032         u32 swsm;
2033
2034         /* Let firmware taken over control of h/w */
2035         if (adapter->flags & FLAG_HAS_SWSM_ON_LOAD) {
2036                 swsm = er32(SWSM);
2037                 ew32(SWSM, swsm & ~E1000_SWSM_DRV_LOAD);
2038         } else if (adapter->flags & FLAG_HAS_CTRLEXT_ON_LOAD) {
2039                 ctrl_ext = er32(CTRL_EXT);
2040                 ew32(CTRL_EXT, ctrl_ext & ~E1000_CTRL_EXT_DRV_LOAD);
2041         }
2042 }
2043
2044 /**
2045  * @e1000_alloc_ring - allocate memory for a ring structure
2046  **/
2047 static int e1000_alloc_ring_dma(struct e1000_adapter *adapter,
2048                                 struct e1000_ring *ring)
2049 {
2050         struct pci_dev *pdev = adapter->pdev;
2051
2052         ring->desc = dma_alloc_coherent(&pdev->dev, ring->size, &ring->dma,
2053                                         GFP_KERNEL);
2054         if (!ring->desc)
2055                 return -ENOMEM;
2056
2057         return 0;
2058 }
2059
2060 /**
2061  * e1000e_setup_tx_resources - allocate Tx resources (Descriptors)
2062  * @adapter: board private structure
2063  *
2064  * Return 0 on success, negative on failure
2065  **/
2066 int e1000e_setup_tx_resources(struct e1000_adapter *adapter)
2067 {
2068         struct e1000_ring *tx_ring = adapter->tx_ring;
2069         int err = -ENOMEM, size;
2070
2071         size = sizeof(struct e1000_buffer) * tx_ring->count;
2072         tx_ring->buffer_info = vzalloc(size);
2073         if (!tx_ring->buffer_info)
2074                 goto err;
2075
2076         /* round up to nearest 4K */
2077         tx_ring->size = tx_ring->count * sizeof(struct e1000_tx_desc);
2078         tx_ring->size = ALIGN(tx_ring->size, 4096);
2079
2080         err = e1000_alloc_ring_dma(adapter, tx_ring);
2081         if (err)
2082                 goto err;
2083
2084         tx_ring->next_to_use = 0;
2085         tx_ring->next_to_clean = 0;
2086
2087         return 0;
2088 err:
2089         vfree(tx_ring->buffer_info);
2090         e_err("Unable to allocate memory for the transmit descriptor ring\n");
2091         return err;
2092 }
2093
2094 /**
2095  * e1000e_setup_rx_resources - allocate Rx resources (Descriptors)
2096  * @adapter: board private structure
2097  *
2098  * Returns 0 on success, negative on failure
2099  **/
2100 int e1000e_setup_rx_resources(struct e1000_adapter *adapter)
2101 {
2102         struct e1000_ring *rx_ring = adapter->rx_ring;
2103         struct e1000_buffer *buffer_info;
2104         int i, size, desc_len, err = -ENOMEM;
2105
2106         size = sizeof(struct e1000_buffer) * rx_ring->count;
2107         rx_ring->buffer_info = vzalloc(size);
2108         if (!rx_ring->buffer_info)
2109                 goto err;
2110
2111         for (i = 0; i < rx_ring->count; i++) {
2112                 buffer_info = &rx_ring->buffer_info[i];
2113                 buffer_info->ps_pages = kcalloc(PS_PAGE_BUFFERS,
2114                                                 sizeof(struct e1000_ps_page),
2115                                                 GFP_KERNEL);
2116                 if (!buffer_info->ps_pages)
2117                         goto err_pages;
2118         }
2119
2120         desc_len = sizeof(union e1000_rx_desc_packet_split);
2121
2122         /* Round up to nearest 4K */
2123         rx_ring->size = rx_ring->count * desc_len;
2124         rx_ring->size = ALIGN(rx_ring->size, 4096);
2125
2126         err = e1000_alloc_ring_dma(adapter, rx_ring);
2127         if (err)
2128                 goto err_pages;
2129
2130         rx_ring->next_to_clean = 0;
2131         rx_ring->next_to_use = 0;
2132         rx_ring->rx_skb_top = NULL;
2133
2134         return 0;
2135
2136 err_pages:
2137         for (i = 0; i < rx_ring->count; i++) {
2138                 buffer_info = &rx_ring->buffer_info[i];
2139                 kfree(buffer_info->ps_pages);
2140         }
2141 err:
2142         vfree(rx_ring->buffer_info);
2143         e_err("Unable to allocate memory for the receive descriptor ring\n");
2144         return err;
2145 }
2146
2147 /**
2148  * e1000_clean_tx_ring - Free Tx Buffers
2149  * @adapter: board private structure
2150  **/
2151 static void e1000_clean_tx_ring(struct e1000_adapter *adapter)
2152 {
2153         struct e1000_ring *tx_ring = adapter->tx_ring;
2154         struct e1000_buffer *buffer_info;
2155         unsigned long size;
2156         unsigned int i;
2157
2158         for (i = 0; i < tx_ring->count; i++) {
2159                 buffer_info = &tx_ring->buffer_info[i];
2160                 e1000_put_txbuf(adapter, buffer_info);
2161         }
2162
2163         size = sizeof(struct e1000_buffer) * tx_ring->count;
2164         memset(tx_ring->buffer_info, 0, size);
2165
2166         memset(tx_ring->desc, 0, tx_ring->size);
2167
2168         tx_ring->next_to_use = 0;
2169         tx_ring->next_to_clean = 0;
2170
2171         writel(0, adapter->hw.hw_addr + tx_ring->head);
2172         writel(0, adapter->hw.hw_addr + tx_ring->tail);
2173 }
2174
2175 /**
2176  * e1000e_free_tx_resources - Free Tx Resources per Queue
2177  * @adapter: board private structure
2178  *
2179  * Free all transmit software resources
2180  **/
2181 void e1000e_free_tx_resources(struct e1000_adapter *adapter)
2182 {
2183         struct pci_dev *pdev = adapter->pdev;
2184         struct e1000_ring *tx_ring = adapter->tx_ring;
2185
2186         e1000_clean_tx_ring(adapter);
2187
2188         vfree(tx_ring->buffer_info);
2189         tx_ring->buffer_info = NULL;
2190
2191         dma_free_coherent(&pdev->dev, tx_ring->size, tx_ring->desc,
2192                           tx_ring->dma);
2193         tx_ring->desc = NULL;
2194 }
2195
2196 /**
2197  * e1000e_free_rx_resources - Free Rx Resources
2198  * @adapter: board private structure
2199  *
2200  * Free all receive software resources
2201  **/
2202
2203 void e1000e_free_rx_resources(struct e1000_adapter *adapter)
2204 {
2205         struct pci_dev *pdev = adapter->pdev;
2206         struct e1000_ring *rx_ring = adapter->rx_ring;
2207         int i;
2208
2209         e1000_clean_rx_ring(adapter);
2210
2211         for (i = 0; i < rx_ring->count; i++)
2212                 kfree(rx_ring->buffer_info[i].ps_pages);
2213
2214         vfree(rx_ring->buffer_info);
2215         rx_ring->buffer_info = NULL;
2216
2217         dma_free_coherent(&pdev->dev, rx_ring->size, rx_ring->desc,
2218                           rx_ring->dma);
2219         rx_ring->desc = NULL;
2220 }
2221
2222 /**
2223  * e1000_update_itr - update the dynamic ITR value based on statistics
2224  * @adapter: pointer to adapter
2225  * @itr_setting: current adapter->itr
2226  * @packets: the number of packets during this measurement interval
2227  * @bytes: the number of bytes during this measurement interval
2228  *
2229  *      Stores a new ITR value based on packets and byte
2230  *      counts during the last interrupt.  The advantage of per interrupt
2231  *      computation is faster updates and more accurate ITR for the current
2232  *      traffic pattern.  Constants in this function were computed
2233  *      based on theoretical maximum wire speed and thresholds were set based
2234  *      on testing data as well as attempting to minimize response time
2235  *      while increasing bulk throughput.  This functionality is controlled
2236  *      by the InterruptThrottleRate module parameter.
2237  **/
2238 static unsigned int e1000_update_itr(struct e1000_adapter *adapter,
2239                                      u16 itr_setting, int packets,
2240                                      int bytes)
2241 {
2242         unsigned int retval = itr_setting;
2243
2244         if (packets == 0)
2245                 goto update_itr_done;
2246
2247         switch (itr_setting) {
2248         case lowest_latency:
2249                 /* handle TSO and jumbo frames */
2250                 if (bytes/packets > 8000)
2251                         retval = bulk_latency;
2252                 else if ((packets < 5) && (bytes > 512))
2253                         retval = low_latency;
2254                 break;
2255         case low_latency:  /* 50 usec aka 20000 ints/s */
2256                 if (bytes > 10000) {
2257                         /* this if handles the TSO accounting */
2258                         if (bytes/packets > 8000)
2259                                 retval = bulk_latency;
2260                         else if ((packets < 10) || ((bytes/packets) > 1200))
2261                                 retval = bulk_latency;
2262                         else if ((packets > 35))
2263                                 retval = lowest_latency;
2264                 } else if (bytes/packets > 2000) {
2265                         retval = bulk_latency;
2266                 } else if (packets <= 2 && bytes < 512) {
2267                         retval = lowest_latency;
2268                 }
2269                 break;
2270         case bulk_latency: /* 250 usec aka 4000 ints/s */
2271                 if (bytes > 25000) {
2272                         if (packets > 35)
2273                                 retval = low_latency;
2274                 } else if (bytes < 6000) {
2275                         retval = low_latency;
2276                 }
2277                 break;
2278         }
2279
2280 update_itr_done:
2281         return retval;
2282 }
2283
2284 static void e1000_set_itr(struct e1000_adapter *adapter)
2285 {
2286         struct e1000_hw *hw = &adapter->hw;
2287         u16 current_itr;
2288         u32 new_itr = adapter->itr;
2289
2290         /* for non-gigabit speeds, just fix the interrupt rate at 4000 */
2291         if (adapter->link_speed != SPEED_1000) {
2292                 current_itr = 0;
2293                 new_itr = 4000;
2294                 goto set_itr_now;
2295         }
2296
2297         if (adapter->flags2 & FLAG2_DISABLE_AIM) {
2298                 new_itr = 0;
2299                 goto set_itr_now;
2300         }
2301
2302         adapter->tx_itr = e1000_update_itr(adapter,
2303                                     adapter->tx_itr,
2304                                     adapter->total_tx_packets,
2305                                     adapter->total_tx_bytes);
2306         /* conservative mode (itr 3) eliminates the lowest_latency setting */
2307         if (adapter->itr_setting == 3 && adapter->tx_itr == lowest_latency)
2308                 adapter->tx_itr = low_latency;
2309
2310         adapter->rx_itr = e1000_update_itr(adapter,
2311                                     adapter->rx_itr,
2312                                     adapter->total_rx_packets,
2313                                     adapter->total_rx_bytes);
2314         /* conservative mode (itr 3) eliminates the lowest_latency setting */
2315         if (adapter->itr_setting == 3 && adapter->rx_itr == lowest_latency)
2316                 adapter->rx_itr = low_latency;
2317
2318         current_itr = max(adapter->rx_itr, adapter->tx_itr);
2319
2320         switch (current_itr) {
2321         /* counts and packets in update_itr are dependent on these numbers */
2322         case lowest_latency:
2323                 new_itr = 70000;
2324                 break;
2325         case low_latency:
2326                 new_itr = 20000; /* aka hwitr = ~200 */
2327                 break;
2328         case bulk_latency:
2329                 new_itr = 4000;
2330                 break;
2331         default:
2332                 break;
2333         }
2334
2335 set_itr_now:
2336         if (new_itr != adapter->itr) {
2337                 /*
2338                  * this attempts to bias the interrupt rate towards Bulk
2339                  * by adding intermediate steps when interrupt rate is
2340                  * increasing
2341                  */
2342                 new_itr = new_itr > adapter->itr ?
2343                              min(adapter->itr + (new_itr >> 2), new_itr) :
2344                              new_itr;
2345                 adapter->itr = new_itr;
2346                 adapter->rx_ring->itr_val = new_itr;
2347                 if (adapter->msix_entries)
2348                         adapter->rx_ring->set_itr = 1;
2349                 else
2350                         if (new_itr)
2351                                 ew32(ITR, 1000000000 / (new_itr * 256));
2352                         else
2353                                 ew32(ITR, 0);
2354         }
2355 }
2356
2357 /**
2358  * e1000_alloc_queues - Allocate memory for all rings
2359  * @adapter: board private structure to initialize
2360  **/
2361 static int __devinit e1000_alloc_queues(struct e1000_adapter *adapter)
2362 {
2363         adapter->tx_ring = kzalloc(sizeof(struct e1000_ring), GFP_KERNEL);
2364         if (!adapter->tx_ring)
2365                 goto err;
2366
2367         adapter->rx_ring = kzalloc(sizeof(struct e1000_ring), GFP_KERNEL);
2368         if (!adapter->rx_ring)
2369                 goto err;
2370
2371         return 0;
2372 err:
2373         e_err("Unable to allocate memory for queues\n");
2374         kfree(adapter->rx_ring);
2375         kfree(adapter->tx_ring);
2376         return -ENOMEM;
2377 }
2378
2379 /**
2380  * e1000_clean - NAPI Rx polling callback
2381  * @napi: struct associated with this polling callback
2382  * @budget: amount of packets driver is allowed to process this poll
2383  **/
2384 static int e1000_clean(struct napi_struct *napi, int budget)
2385 {
2386         struct e1000_adapter *adapter = container_of(napi, struct e1000_adapter, napi);
2387         struct e1000_hw *hw = &adapter->hw;
2388         struct net_device *poll_dev = adapter->netdev;
2389         int tx_cleaned = 1, work_done = 0;
2390
2391         adapter = netdev_priv(poll_dev);
2392
2393         if (adapter->msix_entries &&
2394             !(adapter->rx_ring->ims_val & adapter->tx_ring->ims_val))
2395                 goto clean_rx;
2396
2397         tx_cleaned = e1000_clean_tx_irq(adapter);
2398
2399 clean_rx:
2400         adapter->clean_rx(adapter, &work_done, budget);
2401
2402         if (!tx_cleaned)
2403                 work_done = budget;
2404
2405         /* If budget not fully consumed, exit the polling mode */
2406         if (work_done < budget) {
2407                 if (adapter->itr_setting & 3)
2408                         e1000_set_itr(adapter);
2409                 napi_complete(napi);
2410                 if (!test_bit(__E1000_DOWN, &adapter->state)) {
2411                         if (adapter->msix_entries)
2412                                 ew32(IMS, adapter->rx_ring->ims_val);
2413                         else
2414                                 e1000_irq_enable(adapter);
2415                 }
2416         }
2417
2418         return work_done;
2419 }
2420
2421 static void e1000_vlan_rx_add_vid(struct net_device *netdev, u16 vid)
2422 {
2423         struct e1000_adapter *adapter = netdev_priv(netdev);
2424         struct e1000_hw *hw = &adapter->hw;
2425         u32 vfta, index;
2426
2427         /* don't update vlan cookie if already programmed */
2428         if ((adapter->hw.mng_cookie.status &
2429              E1000_MNG_DHCP_COOKIE_STATUS_VLAN) &&
2430             (vid == adapter->mng_vlan_id))
2431                 return;
2432
2433         /* add VID to filter table */
2434         if (adapter->flags & FLAG_HAS_HW_VLAN_FILTER) {
2435                 index = (vid >> 5) & 0x7F;
2436                 vfta = E1000_READ_REG_ARRAY(hw, E1000_VFTA, index);
2437                 vfta |= (1 << (vid & 0x1F));
2438                 hw->mac.ops.write_vfta(hw, index, vfta);
2439         }
2440
2441         set_bit(vid, adapter->active_vlans);
2442 }
2443
2444 static void e1000_vlan_rx_kill_vid(struct net_device *netdev, u16 vid)
2445 {
2446         struct e1000_adapter *adapter = netdev_priv(netdev);
2447         struct e1000_hw *hw = &adapter->hw;
2448         u32 vfta, index;
2449
2450         if ((adapter->hw.mng_cookie.status &
2451              E1000_MNG_DHCP_COOKIE_STATUS_VLAN) &&
2452             (vid == adapter->mng_vlan_id)) {
2453                 /* release control to f/w */
2454                 e1000e_release_hw_control(adapter);
2455                 return;
2456         }
2457
2458         /* remove VID from filter table */
2459         if (adapter->flags & FLAG_HAS_HW_VLAN_FILTER) {
2460                 index = (vid >> 5) & 0x7F;
2461                 vfta = E1000_READ_REG_ARRAY(hw, E1000_VFTA, index);
2462                 vfta &= ~(1 << (vid & 0x1F));
2463                 hw->mac.ops.write_vfta(hw, index, vfta);
2464         }
2465
2466         clear_bit(vid, adapter->active_vlans);
2467 }
2468
2469 /**
2470  * e1000e_vlan_filter_disable - helper to disable hw VLAN filtering
2471  * @adapter: board private structure to initialize
2472  **/
2473 static void e1000e_vlan_filter_disable(struct e1000_adapter *adapter)
2474 {
2475         struct net_device *netdev = adapter->netdev;
2476         struct e1000_hw *hw = &adapter->hw;
2477         u32 rctl;
2478
2479         if (adapter->flags & FLAG_HAS_HW_VLAN_FILTER) {
2480                 /* disable VLAN receive filtering */
2481                 rctl = er32(RCTL);
2482                 rctl &= ~(E1000_RCTL_VFE | E1000_RCTL_CFIEN);
2483                 ew32(RCTL, rctl);
2484
2485                 if (adapter->mng_vlan_id != (u16)E1000_MNG_VLAN_NONE) {
2486                         e1000_vlan_rx_kill_vid(netdev, adapter->mng_vlan_id);
2487                         adapter->mng_vlan_id = E1000_MNG_VLAN_NONE;
2488                 }
2489         }
2490 }
2491
2492 /**
2493  * e1000e_vlan_filter_enable - helper to enable HW VLAN filtering
2494  * @adapter: board private structure to initialize
2495  **/
2496 static void e1000e_vlan_filter_enable(struct e1000_adapter *adapter)
2497 {
2498         struct e1000_hw *hw = &adapter->hw;
2499         u32 rctl;
2500
2501         if (adapter->flags & FLAG_HAS_HW_VLAN_FILTER) {
2502                 /* enable VLAN receive filtering */
2503                 rctl = er32(RCTL);
2504                 rctl |= E1000_RCTL_VFE;
2505                 rctl &= ~E1000_RCTL_CFIEN;
2506                 ew32(RCTL, rctl);
2507         }
2508 }
2509
2510 /**
2511  * e1000e_vlan_strip_enable - helper to disable HW VLAN stripping
2512  * @adapter: board private structure to initialize
2513  **/
2514 static void e1000e_vlan_strip_disable(struct e1000_adapter *adapter)
2515 {
2516         struct e1000_hw *hw = &adapter->hw;
2517         u32 ctrl;
2518
2519         /* disable VLAN tag insert/strip */
2520         ctrl = er32(CTRL);
2521         ctrl &= ~E1000_CTRL_VME;
2522         ew32(CTRL, ctrl);
2523 }
2524
2525 /**
2526  * e1000e_vlan_strip_enable - helper to enable HW VLAN stripping
2527  * @adapter: board private structure to initialize
2528  **/
2529 static void e1000e_vlan_strip_enable(struct e1000_adapter *adapter)
2530 {
2531         struct e1000_hw *hw = &adapter->hw;
2532         u32 ctrl;
2533
2534         /* enable VLAN tag insert/strip */
2535         ctrl = er32(CTRL);
2536         ctrl |= E1000_CTRL_VME;
2537         ew32(CTRL, ctrl);
2538 }
2539
2540 static void e1000_update_mng_vlan(struct e1000_adapter *adapter)
2541 {
2542         struct net_device *netdev = adapter->netdev;
2543         u16 vid = adapter->hw.mng_cookie.vlan_id;
2544         u16 old_vid = adapter->mng_vlan_id;
2545
2546         if (adapter->hw.mng_cookie.status &
2547             E1000_MNG_DHCP_COOKIE_STATUS_VLAN) {
2548                 e1000_vlan_rx_add_vid(netdev, vid);
2549                 adapter->mng_vlan_id = vid;
2550         }
2551
2552         if ((old_vid != (u16)E1000_MNG_VLAN_NONE) && (vid != old_vid))
2553                 e1000_vlan_rx_kill_vid(netdev, old_vid);
2554 }
2555
2556 static void e1000_restore_vlan(struct e1000_adapter *adapter)
2557 {
2558         u16 vid;
2559
2560         e1000_vlan_rx_add_vid(adapter->netdev, 0);
2561
2562         for_each_set_bit(vid, adapter->active_vlans, VLAN_N_VID)
2563                 e1000_vlan_rx_add_vid(adapter->netdev, vid);
2564 }
2565
2566 static void e1000_init_manageability_pt(struct e1000_adapter *adapter)
2567 {
2568         struct e1000_hw *hw = &adapter->hw;
2569         u32 manc, manc2h, mdef, i, j;
2570
2571         if (!(adapter->flags & FLAG_MNG_PT_ENABLED))
2572                 return;
2573
2574         manc = er32(MANC);
2575
2576         /*
2577          * enable receiving management packets to the host. this will probably
2578          * generate destination unreachable messages from the host OS, but
2579          * the packets will be handled on SMBUS
2580          */
2581         manc |= E1000_MANC_EN_MNG2HOST;
2582         manc2h = er32(MANC2H);
2583
2584         switch (hw->mac.type) {
2585         default:
2586                 manc2h |= (E1000_MANC2H_PORT_623 | E1000_MANC2H_PORT_664);
2587                 break;
2588         case e1000_82574:
2589         case e1000_82583:
2590                 /*
2591                  * Check if IPMI pass-through decision filter already exists;
2592                  * if so, enable it.
2593                  */
2594                 for (i = 0, j = 0; i < 8; i++) {
2595                         mdef = er32(MDEF(i));
2596
2597                         /* Ignore filters with anything other than IPMI ports */
2598                         if (mdef & ~(E1000_MDEF_PORT_623 | E1000_MDEF_PORT_664))
2599                                 continue;
2600
2601                         /* Enable this decision filter in MANC2H */
2602                         if (mdef)
2603                                 manc2h |= (1 << i);
2604
2605                         j |= mdef;
2606                 }
2607
2608                 if (j == (E1000_MDEF_PORT_623 | E1000_MDEF_PORT_664))
2609                         break;
2610
2611                 /* Create new decision filter in an empty filter */
2612                 for (i = 0, j = 0; i < 8; i++)
2613                         if (er32(MDEF(i)) == 0) {
2614                                 ew32(MDEF(i), (E1000_MDEF_PORT_623 |
2615                                                E1000_MDEF_PORT_664));
2616                                 manc2h |= (1 << 1);
2617                                 j++;
2618                                 break;
2619                         }
2620
2621                 if (!j)
2622                         e_warn("Unable to create IPMI pass-through filter\n");
2623                 break;
2624         }
2625
2626         ew32(MANC2H, manc2h);
2627         ew32(MANC, manc);
2628 }
2629
2630 /**
2631  * e1000_configure_tx - Configure Transmit Unit after Reset
2632  * @adapter: board private structure
2633  *
2634  * Configure the Tx unit of the MAC after a reset.
2635  **/
2636 static void e1000_configure_tx(struct e1000_adapter *adapter)
2637 {
2638         struct e1000_hw *hw = &adapter->hw;
2639         struct e1000_ring *tx_ring = adapter->tx_ring;
2640         u64 tdba;
2641         u32 tdlen, tctl, tipg, tarc;
2642         u32 ipgr1, ipgr2;
2643
2644         /* Setup the HW Tx Head and Tail descriptor pointers */
2645         tdba = tx_ring->dma;
2646         tdlen = tx_ring->count * sizeof(struct e1000_tx_desc);
2647         ew32(TDBAL, (tdba & DMA_BIT_MASK(32)));
2648         ew32(TDBAH, (tdba >> 32));
2649         ew32(TDLEN, tdlen);
2650         ew32(TDH, 0);
2651         ew32(TDT, 0);
2652         tx_ring->head = E1000_TDH;
2653         tx_ring->tail = E1000_TDT;
2654
2655         /* Set the default values for the Tx Inter Packet Gap timer */
2656         tipg = DEFAULT_82543_TIPG_IPGT_COPPER;          /*  8  */
2657         ipgr1 = DEFAULT_82543_TIPG_IPGR1;               /*  8  */
2658         ipgr2 = DEFAULT_82543_TIPG_IPGR2;               /*  6  */
2659
2660         if (adapter->flags & FLAG_TIPG_MEDIUM_FOR_80003ESLAN)
2661                 ipgr2 = DEFAULT_80003ES2LAN_TIPG_IPGR2; /*  7  */
2662
2663         tipg |= ipgr1 << E1000_TIPG_IPGR1_SHIFT;
2664         tipg |= ipgr2 << E1000_TIPG_IPGR2_SHIFT;
2665         ew32(TIPG, tipg);
2666
2667         /* Set the Tx Interrupt Delay register */
2668         ew32(TIDV, adapter->tx_int_delay);
2669         /* Tx irq moderation */
2670         ew32(TADV, adapter->tx_abs_int_delay);
2671
2672         if (adapter->flags2 & FLAG2_DMA_BURST) {
2673                 u32 txdctl = er32(TXDCTL(0));
2674                 txdctl &= ~(E1000_TXDCTL_PTHRESH | E1000_TXDCTL_HTHRESH |
2675                             E1000_TXDCTL_WTHRESH);
2676                 /*
2677                  * set up some performance related parameters to encourage the
2678                  * hardware to use the bus more efficiently in bursts, depends
2679                  * on the tx_int_delay to be enabled,
2680                  * wthresh = 5 ==> burst write a cacheline (64 bytes) at a time
2681                  * hthresh = 1 ==> prefetch when one or more available
2682                  * pthresh = 0x1f ==> prefetch if internal cache 31 or less
2683                  * BEWARE: this seems to work but should be considered first if
2684                  * there are Tx hangs or other Tx related bugs
2685                  */
2686                 txdctl |= E1000_TXDCTL_DMA_BURST_ENABLE;
2687                 ew32(TXDCTL(0), txdctl);
2688                 /* erratum work around: set txdctl the same for both queues */
2689                 ew32(TXDCTL(1), txdctl);
2690         }
2691
2692         /* Program the Transmit Control Register */
2693         tctl = er32(TCTL);
2694         tctl &= ~E1000_TCTL_CT;
2695         tctl |= E1000_TCTL_PSP | E1000_TCTL_RTLC |
2696                 (E1000_COLLISION_THRESHOLD << E1000_CT_SHIFT);
2697
2698         if (adapter->flags & FLAG_TARC_SPEED_MODE_BIT) {
2699                 tarc = er32(TARC(0));
2700                 /*
2701                  * set the speed mode bit, we'll clear it if we're not at
2702                  * gigabit link later
2703                  */
2704 #define SPEED_MODE_BIT (1 << 21)
2705                 tarc |= SPEED_MODE_BIT;
2706                 ew32(TARC(0), tarc);
2707         }
2708
2709         /* errata: program both queues to unweighted RR */
2710         if (adapter->flags & FLAG_TARC_SET_BIT_ZERO) {
2711                 tarc = er32(TARC(0));
2712                 tarc |= 1;
2713                 ew32(TARC(0), tarc);
2714                 tarc = er32(TARC(1));
2715                 tarc |= 1;
2716                 ew32(TARC(1), tarc);
2717         }
2718
2719         /* Setup Transmit Descriptor Settings for eop descriptor */
2720         adapter->txd_cmd = E1000_TXD_CMD_EOP | E1000_TXD_CMD_IFCS;
2721
2722         /* only set IDE if we are delaying interrupts using the timers */
2723         if (adapter->tx_int_delay)
2724                 adapter->txd_cmd |= E1000_TXD_CMD_IDE;
2725
2726         /* enable Report Status bit */
2727         adapter->txd_cmd |= E1000_TXD_CMD_RS;
2728
2729         ew32(TCTL, tctl);
2730
2731         e1000e_config_collision_dist(hw);
2732 }
2733
2734 /**
2735  * e1000_setup_rctl - configure the receive control registers
2736  * @adapter: Board private structure
2737  **/
2738 #define PAGE_USE_COUNT(S) (((S) >> PAGE_SHIFT) + \
2739                            (((S) & (PAGE_SIZE - 1)) ? 1 : 0))
2740 static void e1000_setup_rctl(struct e1000_adapter *adapter)
2741 {
2742         struct e1000_hw *hw = &adapter->hw;
2743         u32 rctl, rfctl;
2744         u32 pages = 0;
2745
2746         /* Workaround Si errata on 82579 - configure jumbo frame flow */
2747         if (hw->mac.type == e1000_pch2lan) {
2748                 s32 ret_val;
2749
2750                 if (adapter->netdev->mtu > ETH_DATA_LEN)
2751                         ret_val = e1000_lv_jumbo_workaround_ich8lan(hw, true);
2752                 else
2753                         ret_val = e1000_lv_jumbo_workaround_ich8lan(hw, false);
2754
2755                 if (ret_val)
2756                         e_dbg("failed to enable jumbo frame workaround mode\n");
2757         }
2758
2759         /* Program MC offset vector base */
2760         rctl = er32(RCTL);
2761         rctl &= ~(3 << E1000_RCTL_MO_SHIFT);
2762         rctl |= E1000_RCTL_EN | E1000_RCTL_BAM |
2763                 E1000_RCTL_LBM_NO | E1000_RCTL_RDMTS_HALF |
2764                 (adapter->hw.mac.mc_filter_type << E1000_RCTL_MO_SHIFT);
2765
2766         /* Do not Store bad packets */
2767         rctl &= ~E1000_RCTL_SBP;
2768
2769         /* Enable Long Packet receive */
2770         if (adapter->netdev->mtu <= ETH_DATA_LEN)
2771                 rctl &= ~E1000_RCTL_LPE;
2772         else
2773                 rctl |= E1000_RCTL_LPE;
2774
2775         /* Some systems expect that the CRC is included in SMBUS traffic. The
2776          * hardware strips the CRC before sending to both SMBUS (BMC) and to
2777          * host memory when this is enabled
2778          */
2779         if (adapter->flags2 & FLAG2_CRC_STRIPPING)
2780                 rctl |= E1000_RCTL_SECRC;
2781
2782         /* Workaround Si errata on 82577 PHY - configure IPG for jumbos */
2783         if ((hw->phy.type == e1000_phy_82577) && (rctl & E1000_RCTL_LPE)) {
2784                 u16 phy_data;
2785
2786                 e1e_rphy(hw, PHY_REG(770, 26), &phy_data);
2787                 phy_data &= 0xfff8;
2788                 phy_data |= (1 << 2);
2789                 e1e_wphy(hw, PHY_REG(770, 26), phy_data);
2790
2791                 e1e_rphy(hw, 22, &phy_data);
2792                 phy_data &= 0x0fff;
2793                 phy_data |= (1 << 14);
2794                 e1e_wphy(hw, 0x10, 0x2823);
2795                 e1e_wphy(hw, 0x11, 0x0003);
2796                 e1e_wphy(hw, 22, phy_data);
2797         }
2798
2799         /* Setup buffer sizes */
2800         rctl &= ~E1000_RCTL_SZ_4096;
2801         rctl |= E1000_RCTL_BSEX;
2802         switch (adapter->rx_buffer_len) {
2803         case 2048:
2804         default:
2805                 rctl |= E1000_RCTL_SZ_2048;
2806                 rctl &= ~E1000_RCTL_BSEX;
2807                 break;
2808         case 4096:
2809                 rctl |= E1000_RCTL_SZ_4096;
2810                 break;
2811         case 8192:
2812                 rctl |= E1000_RCTL_SZ_8192;
2813                 break;
2814         case 16384:
2815                 rctl |= E1000_RCTL_SZ_16384;
2816                 break;
2817         }
2818
2819         /*
2820          * 82571 and greater support packet-split where the protocol
2821          * header is placed in skb->data and the packet data is
2822          * placed in pages hanging off of skb_shinfo(skb)->nr_frags.
2823          * In the case of a non-split, skb->data is linearly filled,
2824          * followed by the page buffers.  Therefore, skb->data is
2825          * sized to hold the largest protocol header.
2826          *
2827          * allocations using alloc_page take too long for regular MTU
2828          * so only enable packet split for jumbo frames
2829          *
2830          * Using pages when the page size is greater than 16k wastes
2831          * a lot of memory, since we allocate 3 pages at all times
2832          * per packet.
2833          */
2834         pages = PAGE_USE_COUNT(adapter->netdev->mtu);
2835         if (!(adapter->flags & FLAG_HAS_ERT) && (pages <= 3) &&
2836             (PAGE_SIZE <= 16384) && (rctl & E1000_RCTL_LPE))
2837                 adapter->rx_ps_pages = pages;
2838         else
2839                 adapter->rx_ps_pages = 0;
2840
2841         if (adapter->rx_ps_pages) {
2842                 u32 psrctl = 0;
2843
2844                 /* Configure extra packet-split registers */
2845                 rfctl = er32(RFCTL);
2846                 rfctl |= E1000_RFCTL_EXTEN;
2847                 /*
2848                  * disable packet split support for IPv6 extension headers,
2849                  * because some malformed IPv6 headers can hang the Rx
2850                  */
2851                 rfctl |= (E1000_RFCTL_IPV6_EX_DIS |
2852                           E1000_RFCTL_NEW_IPV6_EXT_DIS);
2853
2854                 ew32(RFCTL, rfctl);
2855
2856                 /* Enable Packet split descriptors */
2857                 rctl |= E1000_RCTL_DTYP_PS;
2858
2859                 psrctl |= adapter->rx_ps_bsize0 >>
2860                         E1000_PSRCTL_BSIZE0_SHIFT;
2861
2862                 switch (adapter->rx_ps_pages) {
2863                 case 3:
2864                         psrctl |= PAGE_SIZE <<
2865                                 E1000_PSRCTL_BSIZE3_SHIFT;
2866                 case 2:
2867                         psrctl |= PAGE_SIZE <<
2868                                 E1000_PSRCTL_BSIZE2_SHIFT;
2869                 case 1:
2870                         psrctl |= PAGE_SIZE >>
2871                                 E1000_PSRCTL_BSIZE1_SHIFT;
2872                         break;
2873                 }
2874
2875                 ew32(PSRCTL, psrctl);
2876         }
2877
2878         ew32(RCTL, rctl);
2879         /* just started the receive unit, no need to restart */
2880         adapter->flags &= ~FLAG_RX_RESTART_NOW;
2881 }
2882
2883 /**
2884  * e1000_configure_rx - Configure Receive Unit after Reset
2885  * @adapter: board private structure
2886  *
2887  * Configure the Rx unit of the MAC after a reset.
2888  **/
2889 static void e1000_configure_rx(struct e1000_adapter *adapter)
2890 {
2891         struct e1000_hw *hw = &adapter->hw;
2892         struct e1000_ring *rx_ring = adapter->rx_ring;
2893         u64 rdba;
2894         u32 rdlen, rctl, rxcsum, ctrl_ext;
2895
2896         if (adapter->rx_ps_pages) {
2897                 /* this is a 32 byte descriptor */
2898                 rdlen = rx_ring->count *
2899                     sizeof(union e1000_rx_desc_packet_split);
2900                 adapter->clean_rx = e1000_clean_rx_irq_ps;
2901                 adapter->alloc_rx_buf = e1000_alloc_rx_buffers_ps;
2902         } else if (adapter->netdev->mtu > ETH_FRAME_LEN + ETH_FCS_LEN) {
2903                 rdlen = rx_ring->count * sizeof(struct e1000_rx_desc);
2904                 adapter->clean_rx = e1000_clean_jumbo_rx_irq;
2905                 adapter->alloc_rx_buf = e1000_alloc_jumbo_rx_buffers;
2906         } else {
2907                 rdlen = rx_ring->count * sizeof(struct e1000_rx_desc);
2908                 adapter->clean_rx = e1000_clean_rx_irq;
2909                 adapter->alloc_rx_buf = e1000_alloc_rx_buffers;
2910         }
2911
2912         /* disable receives while setting up the descriptors */
2913         rctl = er32(RCTL);
2914         ew32(RCTL, rctl & ~E1000_RCTL_EN);
2915         e1e_flush();
2916         usleep_range(10000, 20000);
2917
2918         if (adapter->flags2 & FLAG2_DMA_BURST) {
2919                 /*
2920                  * set the writeback threshold (only takes effect if the RDTR
2921                  * is set). set GRAN=1 and write back up to 0x4 worth, and
2922                  * enable prefetching of 0x20 Rx descriptors
2923                  * granularity = 01
2924                  * wthresh = 04,
2925                  * hthresh = 04,
2926                  * pthresh = 0x20
2927                  */
2928                 ew32(RXDCTL(0), E1000_RXDCTL_DMA_BURST_ENABLE);
2929                 ew32(RXDCTL(1), E1000_RXDCTL_DMA_BURST_ENABLE);
2930
2931                 /*
2932                  * override the delay timers for enabling bursting, only if
2933                  * the value was not set by the user via module options
2934                  */
2935                 if (adapter->rx_int_delay == DEFAULT_RDTR)
2936                         adapter->rx_int_delay = BURST_RDTR;
2937                 if (adapter->rx_abs_int_delay == DEFAULT_RADV)
2938                         adapter->rx_abs_int_delay = BURST_RADV;
2939         }
2940
2941         /* set the Receive Delay Timer Register */
2942         ew32(RDTR, adapter->rx_int_delay);
2943
2944         /* irq moderation */
2945         ew32(RADV, adapter->rx_abs_int_delay);
2946         if ((adapter->itr_setting != 0) && (adapter->itr != 0))
2947                 ew32(ITR, 1000000000 / (adapter->itr * 256));
2948
2949         ctrl_ext = er32(CTRL_EXT);
2950         /* Auto-Mask interrupts upon ICR access */
2951         ctrl_ext |= E1000_CTRL_EXT_IAME;
2952         ew32(IAM, 0xffffffff);
2953         ew32(CTRL_EXT, ctrl_ext);
2954         e1e_flush();
2955
2956         /*
2957          * Setup the HW Rx Head and Tail Descriptor Pointers and
2958          * the Base and Length of the Rx Descriptor Ring
2959          */
2960         rdba = rx_ring->dma;
2961         ew32(RDBAL, (rdba & DMA_BIT_MASK(32)));
2962         ew32(RDBAH, (rdba >> 32));
2963         ew32(RDLEN, rdlen);
2964         ew32(RDH, 0);
2965         ew32(RDT, 0);
2966         rx_ring->head = E1000_RDH;
2967         rx_ring->tail = E1000_RDT;
2968
2969         /* Enable Receive Checksum Offload for TCP and UDP */
2970         rxcsum = er32(RXCSUM);
2971         if (adapter->flags & FLAG_RX_CSUM_ENABLED) {
2972                 rxcsum |= E1000_RXCSUM_TUOFL;
2973
2974                 /*
2975                  * IPv4 payload checksum for UDP fragments must be
2976                  * used in conjunction with packet-split.
2977                  */
2978                 if (adapter->rx_ps_pages)
2979                         rxcsum |= E1000_RXCSUM_IPPCSE;
2980         } else {
2981                 rxcsum &= ~E1000_RXCSUM_TUOFL;
2982                 /* no need to clear IPPCSE as it defaults to 0 */
2983         }
2984         ew32(RXCSUM, rxcsum);
2985
2986         /*
2987          * Enable early receives on supported devices, only takes effect when
2988          * packet size is equal or larger than the specified value (in 8 byte
2989          * units), e.g. using jumbo frames when setting to E1000_ERT_2048
2990          */
2991         if ((adapter->flags & FLAG_HAS_ERT) ||
2992             (adapter->hw.mac.type == e1000_pch2lan)) {
2993                 if (adapter->netdev->mtu > ETH_DATA_LEN) {
2994                         u32 rxdctl = er32(RXDCTL(0));
2995                         ew32(RXDCTL(0), rxdctl | 0x3);
2996                         if (adapter->flags & FLAG_HAS_ERT)
2997                                 ew32(ERT, E1000_ERT_2048 | (1 << 13));
2998                         /*
2999                          * With jumbo frames and early-receive enabled,
3000                          * excessive C-state transition latencies result in
3001                          * dropped transactions.
3002                          */
3003                         pm_qos_update_request(&adapter->netdev->pm_qos_req, 55);
3004                 } else {
3005                         pm_qos_update_request(&adapter->netdev->pm_qos_req,
3006                                               PM_QOS_DEFAULT_VALUE);
3007                 }
3008         }
3009
3010         /* Enable Receives */
3011         ew32(RCTL, rctl);
3012 }
3013
3014 /**
3015  *  e1000_update_mc_addr_list - Update Multicast addresses
3016  *  @hw: pointer to the HW structure
3017  *  @mc_addr_list: array of multicast addresses to program
3018  *  @mc_addr_count: number of multicast addresses to program
3019  *
3020  *  Updates the Multicast Table Array.
3021  *  The caller must have a packed mc_addr_list of multicast addresses.
3022  **/
3023 static void e1000_update_mc_addr_list(struct e1000_hw *hw, u8 *mc_addr_list,
3024                                       u32 mc_addr_count)
3025 {
3026         hw->mac.ops.update_mc_addr_list(hw, mc_addr_list, mc_addr_count);
3027 }
3028
3029 /**
3030  * e1000_set_multi - Multicast and Promiscuous mode set
3031  * @netdev: network interface device structure
3032  *
3033  * The set_multi entry point is called whenever the multicast address
3034  * list or the network interface flags are updated.  This routine is
3035  * responsible for configuring the hardware for proper multicast,
3036  * promiscuous mode, and all-multi behavior.
3037  **/
3038 static void e1000_set_multi(struct net_device *netdev)
3039 {
3040         struct e1000_adapter *adapter = netdev_priv(netdev);
3041         struct e1000_hw *hw = &adapter->hw;
3042         struct netdev_hw_addr *ha;
3043         u8  *mta_list;
3044         u32 rctl;
3045
3046         /* Check for Promiscuous and All Multicast modes */
3047
3048         rctl = er32(RCTL);
3049
3050         if (netdev->flags & IFF_PROMISC) {
3051                 rctl |= (E1000_RCTL_UPE | E1000_RCTL_MPE);
3052                 rctl &= ~E1000_RCTL_VFE;
3053                 /* Do not hardware filter VLANs in promisc mode */
3054                 e1000e_vlan_filter_disable(adapter);
3055         } else {
3056                 if (netdev->flags & IFF_ALLMULTI) {
3057                         rctl |= E1000_RCTL_MPE;
3058                         rctl &= ~E1000_RCTL_UPE;
3059                 } else {
3060                         rctl &= ~(E1000_RCTL_UPE | E1000_RCTL_MPE);
3061                 }
3062                 e1000e_vlan_filter_enable(adapter);
3063         }
3064
3065         ew32(RCTL, rctl);
3066
3067         if (!netdev_mc_empty(netdev)) {
3068                 int i = 0;
3069
3070                 mta_list = kmalloc(netdev_mc_count(netdev) * 6, GFP_ATOMIC);
3071                 if (!mta_list)
3072                         return;
3073
3074                 /* prepare a packed array of only addresses. */
3075                 netdev_for_each_mc_addr(ha, netdev)
3076                         memcpy(mta_list + (i++ * ETH_ALEN), ha->addr, ETH_ALEN);
3077
3078                 e1000_update_mc_addr_list(hw, mta_list, i);
3079                 kfree(mta_list);
3080         } else {
3081                 /*
3082                  * if we're called from probe, we might not have
3083                  * anything to do here, so clear out the list
3084                  */
3085                 e1000_update_mc_addr_list(hw, NULL, 0);
3086         }
3087
3088         if (netdev->features & NETIF_F_HW_VLAN_RX)
3089                 e1000e_vlan_strip_enable(adapter);
3090         else
3091                 e1000e_vlan_strip_disable(adapter);
3092 }
3093
3094 /**
3095  * e1000_configure - configure the hardware for Rx and Tx
3096  * @adapter: private board structure
3097  **/
3098 static void e1000_configure(struct e1000_adapter *adapter)
3099 {
3100         e1000_set_multi(adapter->netdev);
3101
3102         e1000_restore_vlan(adapter);
3103         e1000_init_manageability_pt(adapter);
3104
3105         e1000_configure_tx(adapter);
3106         e1000_setup_rctl(adapter);
3107         e1000_configure_rx(adapter);
3108         adapter->alloc_rx_buf(adapter, e1000_desc_unused(adapter->rx_ring));
3109 }
3110
3111 /**
3112  * e1000e_power_up_phy - restore link in case the phy was powered down
3113  * @adapter: address of board private structure
3114  *
3115  * The phy may be powered down to save power and turn off link when the
3116  * driver is unloaded and wake on lan is not enabled (among others)
3117  * *** this routine MUST be followed by a call to e1000e_reset ***
3118  **/
3119 void e1000e_power_up_phy(struct e1000_adapter *adapter)
3120 {
3121         if (adapter->hw.phy.ops.power_up)
3122                 adapter->hw.phy.ops.power_up(&adapter->hw);
3123
3124         adapter->hw.mac.ops.setup_link(&adapter->hw);
3125 }
3126
3127 /**
3128  * e1000_power_down_phy - Power down the PHY
3129  *
3130  * Power down the PHY so no link is implied when interface is down.
3131  * The PHY cannot be powered down if management or WoL is active.
3132  */
3133 static void e1000_power_down_phy(struct e1000_adapter *adapter)
3134 {
3135         /* WoL is enabled */
3136         if (adapter->wol)
3137                 return;
3138
3139         if (adapter->hw.phy.ops.power_down)
3140                 adapter->hw.phy.ops.power_down(&adapter->hw);
3141 }
3142
3143 /**
3144  * e1000e_reset - bring the hardware into a known good state
3145  *
3146  * This function boots the hardware and enables some settings that
3147  * require a configuration cycle of the hardware - those cannot be
3148  * set/changed during runtime. After reset the device needs to be
3149  * properly configured for Rx, Tx etc.
3150  */
3151 void e1000e_reset(struct e1000_adapter *adapter)
3152 {
3153         struct e1000_mac_info *mac = &adapter->hw.mac;
3154         struct e1000_fc_info *fc = &adapter->hw.fc;
3155         struct e1000_hw *hw = &adapter->hw;
3156         u32 tx_space, min_tx_space, min_rx_space;
3157         u32 pba = adapter->pba;
3158         u16 hwm;
3159
3160         /* reset Packet Buffer Allocation to default */
3161         ew32(PBA, pba);
3162
3163         if (adapter->max_frame_size > ETH_FRAME_LEN + ETH_FCS_LEN) {
3164                 /*
3165                  * To maintain wire speed transmits, the Tx FIFO should be
3166                  * large enough to accommodate two full transmit packets,
3167                  * rounded up to the next 1KB and expressed in KB.  Likewise,
3168                  * the Rx FIFO should be large enough to accommodate at least
3169                  * one full receive packet and is similarly rounded up and
3170                  * expressed in KB.
3171                  */
3172                 pba = er32(PBA);
3173                 /* upper 16 bits has Tx packet buffer allocation size in KB */
3174                 tx_space = pba >> 16;
3175                 /* lower 16 bits has Rx packet buffer allocation size in KB */
3176                 pba &= 0xffff;
3177                 /*
3178                  * the Tx fifo also stores 16 bytes of information about the Tx
3179                  * but don't include ethernet FCS because hardware appends it
3180                  */
3181                 min_tx_space = (adapter->max_frame_size +
3182                                 sizeof(struct e1000_tx_desc) -
3183                                 ETH_FCS_LEN) * 2;
3184                 min_tx_space = ALIGN(min_tx_space, 1024);
3185                 min_tx_space >>= 10;
3186                 /* software strips receive CRC, so leave room for it */
3187                 min_rx_space = adapter->max_frame_size;
3188                 min_rx_space = ALIGN(min_rx_space, 1024);
3189                 min_rx_space >>= 10;
3190
3191                 /*
3192                  * If current Tx allocation is less than the min Tx FIFO size,
3193                  * and the min Tx FIFO size is less than the current Rx FIFO
3194                  * allocation, take space away from current Rx allocation
3195                  */
3196                 if ((tx_space < min_tx_space) &&
3197                     ((min_tx_space - tx_space) < pba)) {
3198                         pba -= min_tx_space - tx_space;
3199
3200                         /*
3201                          * if short on Rx space, Rx wins and must trump Tx
3202                          * adjustment or use Early Receive if available
3203                          */
3204                         if ((pba < min_rx_space) &&
3205                             (!(adapter->flags & FLAG_HAS_ERT)))
3206                                 /* ERT enabled in e1000_configure_rx */
3207                                 pba = min_rx_space;
3208                 }
3209
3210                 ew32(PBA, pba);
3211         }
3212
3213         /*
3214          * flow control settings
3215          *
3216          * The high water mark must be low enough to fit one full frame
3217          * (or the size used for early receive) above it in the Rx FIFO.
3218          * Set it to the lower of:
3219          * - 90% of the Rx FIFO size, and
3220          * - the full Rx FIFO size minus the early receive size (for parts
3221          *   with ERT support assuming ERT set to E1000_ERT_2048), or
3222          * - the full Rx FIFO size minus one full frame
3223          */
3224         if (adapter->flags & FLAG_DISABLE_FC_PAUSE_TIME)
3225                 fc->pause_time = 0xFFFF;
3226         else
3227                 fc->pause_time = E1000_FC_PAUSE_TIME;
3228         fc->send_xon = 1;
3229         fc->current_mode = fc->requested_mode;
3230
3231         switch (hw->mac.type) {
3232         default:
3233                 if ((adapter->flags & FLAG_HAS_ERT) &&
3234                     (adapter->netdev->mtu > ETH_DATA_LEN))
3235                         hwm = min(((pba << 10) * 9 / 10),
3236                                   ((pba << 10) - (E1000_ERT_2048 << 3)));
3237                 else
3238                         hwm = min(((pba << 10) * 9 / 10),
3239                                   ((pba << 10) - adapter->max_frame_size));
3240
3241                 fc->high_water = hwm & E1000_FCRTH_RTH; /* 8-byte granularity */
3242                 fc->low_water = fc->high_water - 8;
3243                 break;
3244         case e1000_pchlan:
3245                 /*
3246                  * Workaround PCH LOM adapter hangs with certain network
3247                  * loads.  If hangs persist, try disabling Tx flow control.
3248                  */
3249                 if (adapter->netdev->mtu > ETH_DATA_LEN) {
3250                         fc->high_water = 0x3500;
3251                         fc->low_water  = 0x1500;
3252                 } else {
3253                         fc->high_water = 0x5000;
3254                         fc->low_water  = 0x3000;
3255                 }
3256                 fc->refresh_time = 0x1000;
3257                 break;
3258         case e1000_pch2lan:
3259                 fc->high_water = 0x05C20;
3260                 fc->low_water = 0x05048;
3261                 fc->pause_time = 0x0650;
3262                 fc->refresh_time = 0x0400;
3263                 if (adapter->netdev->mtu > ETH_DATA_LEN) {
3264                         pba = 14;
3265                         ew32(PBA, pba);
3266                 }
3267                 break;
3268         }
3269
3270         /*
3271          * Disable Adaptive Interrupt Moderation if 2 full packets cannot
3272          * fit in receive buffer and early-receive not supported.
3273          */
3274         if (adapter->itr_setting & 0x3) {
3275                 if (((adapter->max_frame_size * 2) > (pba << 10)) &&
3276                     !(adapter->flags & FLAG_HAS_ERT)) {
3277                         if (!(adapter->flags2 & FLAG2_DISABLE_AIM)) {
3278                                 dev_info(&adapter->pdev->dev,
3279                                         "Interrupt Throttle Rate turned off\n");
3280                                 adapter->flags2 |= FLAG2_DISABLE_AIM;
3281                                 ew32(ITR, 0);
3282                         }
3283                 } else if (adapter->flags2 & FLAG2_DISABLE_AIM) {
3284                         dev_info(&adapter->pdev->dev,
3285                                  "Interrupt Throttle Rate turned on\n");
3286                         adapter->flags2 &= ~FLAG2_DISABLE_AIM;
3287                         adapter->itr = 20000;
3288                         ew32(ITR, 1000000000 / (adapter->itr * 256));
3289                 }
3290         }
3291
3292         /* Allow time for pending master requests to run */
3293         mac->ops.reset_hw(hw);
3294
3295         /*
3296          * For parts with AMT enabled, let the firmware know
3297          * that the network interface is in control
3298          */
3299         if (adapter->flags & FLAG_HAS_AMT)
3300                 e1000e_get_hw_control(adapter);
3301
3302         ew32(WUC, 0);
3303
3304         if (mac->ops.init_hw(hw))
3305                 e_err("Hardware Error\n");
3306
3307         e1000_update_mng_vlan(adapter);
3308
3309         /* Enable h/w to recognize an 802.1Q VLAN Ethernet packet */
3310         ew32(VET, ETH_P_8021Q);
3311
3312         e1000e_reset_adaptive(hw);
3313
3314         if (!netif_running(adapter->netdev) &&
3315             !test_bit(__E1000_TESTING, &adapter->state)) {
3316                 e1000_power_down_phy(adapter);
3317                 return;
3318         }
3319
3320         e1000_get_phy_info(hw);
3321
3322         if ((adapter->flags & FLAG_HAS_SMART_POWER_DOWN) &&
3323             !(adapter->flags & FLAG_SMART_POWER_DOWN)) {
3324                 u16 phy_data = 0;
3325                 /*
3326                  * speed up time to link by disabling smart power down, ignore
3327                  * the return value of this function because there is nothing
3328                  * different we would do if it failed
3329                  */
3330                 e1e_rphy(hw, IGP02E1000_PHY_POWER_MGMT, &phy_data);
3331                 phy_data &= ~IGP02E1000_PM_SPD;
3332                 e1e_wphy(hw, IGP02E1000_PHY_POWER_MGMT, phy_data);
3333         }
3334 }
3335
3336 int e1000e_up(struct e1000_adapter *adapter)
3337 {
3338         struct e1000_hw *hw = &adapter->hw;
3339
3340         /* hardware has been reset, we need to reload some things */
3341         e1000_configure(adapter);
3342
3343         clear_bit(__E1000_DOWN, &adapter->state);
3344
3345         napi_enable(&adapter->napi);
3346         if (adapter->msix_entries)
3347                 e1000_configure_msix(adapter);
3348         e1000_irq_enable(adapter);
3349
3350         netif_start_queue(adapter->netdev);
3351
3352         /* fire a link change interrupt to start the watchdog */
3353         if (adapter->msix_entries)
3354                 ew32(ICS, E1000_ICS_LSC | E1000_ICR_OTHER);
3355         else
3356                 ew32(ICS, E1000_ICS_LSC);
3357
3358         return 0;
3359 }
3360
3361 static void e1000e_flush_descriptors(struct e1000_adapter *adapter)
3362 {
3363         struct e1000_hw *hw = &adapter->hw;
3364
3365         if (!(adapter->flags2 & FLAG2_DMA_BURST))
3366                 return;
3367
3368         /* flush pending descriptor writebacks to memory */
3369         ew32(TIDV, adapter->tx_int_delay | E1000_TIDV_FPD);
3370         ew32(RDTR, adapter->rx_int_delay | E1000_RDTR_FPD);
3371
3372         /* execute the writes immediately */
3373         e1e_flush();
3374 }
3375
3376 static void e1000e_update_stats(struct e1000_adapter *adapter);
3377
3378 void e1000e_down(struct e1000_adapter *adapter)
3379 {
3380         struct net_device *netdev = adapter->netdev;
3381         struct e1000_hw *hw = &adapter->hw;
3382         u32 tctl, rctl;
3383
3384         /*
3385          * signal that we're down so the interrupt handler does not
3386          * reschedule our watchdog timer
3387          */
3388         set_bit(__E1000_DOWN, &adapter->state);
3389
3390         /* disable receives in the hardware */
3391         rctl = er32(RCTL);
3392         ew32(RCTL, rctl & ~E1000_RCTL_EN);
3393         /* flush and sleep below */
3394
3395         netif_stop_queue(netdev);
3396
3397         /* disable transmits in the hardware */
3398         tctl = er32(TCTL);
3399         tctl &= ~E1000_TCTL_EN;
3400         ew32(TCTL, tctl);
3401         /* flush both disables and wait for them to finish */
3402         e1e_flush();
3403         usleep_range(10000, 20000);
3404
3405         napi_disable(&adapter->napi);
3406         e1000_irq_disable(adapter);
3407
3408         del_timer_sync(&adapter->watchdog_timer);
3409         del_timer_sync(&adapter->phy_info_timer);
3410
3411         netif_carrier_off(netdev);
3412
3413         spin_lock(&adapter->stats64_lock);
3414         e1000e_update_stats(adapter);
3415         spin_unlock(&adapter->stats64_lock);
3416
3417         e1000e_flush_descriptors(adapter);
3418         e1000_clean_tx_ring(adapter);
3419         e1000_clean_rx_ring(adapter);
3420
3421         adapter->link_speed = 0;
3422         adapter->link_duplex = 0;
3423
3424         if (!pci_channel_offline(adapter->pdev))
3425                 e1000e_reset(adapter);
3426
3427         /*
3428          * TODO: for power management, we could drop the link and
3429          * pci_disable_device here.
3430          */
3431 }
3432
3433 void e1000e_reinit_locked(struct e1000_adapter *adapter)
3434 {
3435         might_sleep();
3436         while (test_and_set_bit(__E1000_RESETTING, &adapter->state))
3437                 usleep_range(1000, 2000);
3438         e1000e_down(adapter);
3439         e1000e_up(adapter);
3440         clear_bit(__E1000_RESETTING, &adapter->state);
3441 }
3442
3443 /**
3444  * e1000_sw_init - Initialize general software structures (struct e1000_adapter)
3445  * @adapter: board private structure to initialize
3446  *
3447  * e1000_sw_init initializes the Adapter private data structure.
3448  * Fields are initialized based on PCI device information and
3449  * OS network device settings (MTU size).
3450  **/
3451 static int __devinit e1000_sw_init(struct e1000_adapter *adapter)
3452 {
3453         struct net_device *netdev = adapter->netdev;
3454
3455         adapter->rx_buffer_len = ETH_FRAME_LEN + VLAN_HLEN + ETH_FCS_LEN;
3456         adapter->rx_ps_bsize0 = 128;
3457         adapter->max_frame_size = netdev->mtu + ETH_HLEN + ETH_FCS_LEN;
3458         adapter->min_frame_size = ETH_ZLEN + ETH_FCS_LEN;
3459
3460         spin_lock_init(&adapter->stats64_lock);
3461
3462         e1000e_set_interrupt_capability(adapter);
3463
3464         if (e1000_alloc_queues(adapter))
3465                 return -ENOMEM;
3466
3467         /* Explicitly disable IRQ since the NIC can be in any state. */
3468         e1000_irq_disable(adapter);
3469
3470         set_bit(__E1000_DOWN, &adapter->state);
3471         return 0;
3472 }
3473
3474 /**
3475  * e1000_intr_msi_test - Interrupt Handler
3476  * @irq: interrupt number
3477  * @data: pointer to a network interface device structure
3478  **/
3479 static irqreturn_t e1000_intr_msi_test(int irq, void *data)
3480 {
3481         struct net_device *netdev = data;
3482         struct e1000_adapter *adapter = netdev_priv(netdev);
3483         struct e1000_hw *hw = &adapter->hw;
3484         u32 icr = er32(ICR);
3485
3486         e_dbg("icr is %08X\n", icr);
3487         if (icr & E1000_ICR_RXSEQ) {
3488                 adapter->flags &= ~FLAG_MSI_TEST_FAILED;
3489                 wmb();
3490         }
3491
3492         return IRQ_HANDLED;
3493 }
3494
3495 /**
3496  * e1000_test_msi_interrupt - Returns 0 for successful test
3497  * @adapter: board private struct
3498  *
3499  * code flow taken from tg3.c
3500  **/
3501 static int e1000_test_msi_interrupt(struct e1000_adapter *adapter)
3502 {
3503         struct net_device *netdev = adapter->netdev;
3504         struct e1000_hw *hw = &adapter->hw;
3505         int err;
3506
3507         /* poll_enable hasn't been called yet, so don't need disable */
3508         /* clear any pending events */
3509         er32(ICR);
3510
3511         /* free the real vector and request a test handler */
3512         e1000_free_irq(adapter);
3513         e1000e_reset_interrupt_capability(adapter);
3514
3515         /* Assume that the test fails, if it succeeds then the test
3516          * MSI irq handler will unset this flag */
3517         adapter->flags |= FLAG_MSI_TEST_FAILED;
3518
3519         err = pci_enable_msi(adapter->pdev);
3520         if (err)
3521                 goto msi_test_failed;
3522
3523         err = request_irq(adapter->pdev->irq, e1000_intr_msi_test, 0,
3524                           netdev->name, netdev);
3525         if (err) {
3526                 pci_disable_msi(adapter->pdev);
3527                 goto msi_test_failed;
3528         }
3529
3530         wmb();
3531
3532         e1000_irq_enable(adapter);
3533
3534         /* fire an unusual interrupt on the test handler */
3535         ew32(ICS, E1000_ICS_RXSEQ);
3536         e1e_flush();
3537         msleep(50);
3538
3539         e1000_irq_disable(adapter);
3540
3541         rmb();
3542
3543         if (adapter->flags & FLAG_MSI_TEST_FAILED) {
3544                 adapter->int_mode = E1000E_INT_MODE_LEGACY;
3545                 e_info("MSI interrupt test failed, using legacy interrupt.\n");
3546         } else
3547                 e_dbg("MSI interrupt test succeeded!\n");
3548
3549         free_irq(adapter->pdev->irq, netdev);
3550         pci_disable_msi(adapter->pdev);
3551
3552 msi_test_failed:
3553         e1000e_set_interrupt_capability(adapter);
3554         return e1000_request_irq(adapter);
3555 }
3556
3557 /**
3558  * e1000_test_msi - Returns 0 if MSI test succeeds or INTx mode is restored
3559  * @adapter: board private struct
3560  *
3561  * code flow taken from tg3.c, called with e1000 interrupts disabled.
3562  **/
3563 static int e1000_test_msi(struct e1000_adapter *adapter)
3564 {
3565         int err;
3566         u16 pci_cmd;
3567
3568         if (!(adapter->flags & FLAG_MSI_ENABLED))
3569                 return 0;
3570
3571         /* disable SERR in case the MSI write causes a master abort */
3572         pci_read_config_word(adapter->pdev, PCI_COMMAND, &pci_cmd);
3573         if (pci_cmd & PCI_COMMAND_SERR)
3574                 pci_write_config_word(adapter->pdev, PCI_COMMAND,
3575                                       pci_cmd & ~PCI_COMMAND_SERR);
3576
3577         err = e1000_test_msi_interrupt(adapter);
3578
3579         /* re-enable SERR */
3580         if (pci_cmd & PCI_COMMAND_SERR) {
3581                 pci_read_config_word(adapter->pdev, PCI_COMMAND, &pci_cmd);
3582                 pci_cmd |= PCI_COMMAND_SERR;
3583                 pci_write_config_word(adapter->pdev, PCI_COMMAND, pci_cmd);
3584         }
3585
3586         return err;
3587 }
3588
3589 /**
3590  * e1000_open - Called when a network interface is made active
3591  * @netdev: network interface device structure
3592  *
3593  * Returns 0 on success, negative value on failure
3594  *
3595  * The open entry point is called when a network interface is made
3596  * active by the system (IFF_UP).  At this point all resources needed
3597  * for transmit and receive operations are allocated, the interrupt
3598  * handler is registered with the OS, the watchdog timer is started,
3599  * and the stack is notified that the interface is ready.
3600  **/
3601 static int e1000_open(struct net_device *netdev)
3602 {
3603         struct e1000_adapter *adapter = netdev_priv(netdev);
3604         struct e1000_hw *hw = &adapter->hw;
3605         struct pci_dev *pdev = adapter->pdev;
3606         int err;
3607
3608         /* disallow open during test */
3609         if (test_bit(__E1000_TESTING, &adapter->state))
3610                 return -EBUSY;
3611
3612         pm_runtime_get_sync(&pdev->dev);
3613
3614         netif_carrier_off(netdev);
3615
3616         /* allocate transmit descriptors */
3617         err = e1000e_setup_tx_resources(adapter);
3618         if (err)
3619                 goto err_setup_tx;
3620
3621         /* allocate receive descriptors */
3622         err = e1000e_setup_rx_resources(adapter);
3623         if (err)
3624                 goto err_setup_rx;
3625
3626         /*
3627          * If AMT is enabled, let the firmware know that the network
3628          * interface is now open and reset the part to a known state.
3629          */
3630         if (adapter->flags & FLAG_HAS_AMT) {
3631                 e1000e_get_hw_control(adapter);
3632                 e1000e_reset(adapter);
3633         }
3634
3635         e1000e_power_up_phy(adapter);
3636
3637         adapter->mng_vlan_id = E1000_MNG_VLAN_NONE;
3638         if ((adapter->hw.mng_cookie.status &
3639              E1000_MNG_DHCP_COOKIE_STATUS_VLAN))
3640                 e1000_update_mng_vlan(adapter);
3641
3642         /* DMA latency requirement to workaround early-receive/jumbo issue */
3643         if ((adapter->flags & FLAG_HAS_ERT) ||
3644             (adapter->hw.mac.type == e1000_pch2lan))
3645                 pm_qos_add_request(&adapter->netdev->pm_qos_req,
3646                                    PM_QOS_CPU_DMA_LATENCY,
3647                                    PM_QOS_DEFAULT_VALUE);
3648
3649         /*
3650          * before we allocate an interrupt, we must be ready to handle it.
3651          * Setting DEBUG_SHIRQ in the kernel makes it fire an interrupt
3652          * as soon as we call pci_request_irq, so we have to setup our
3653          * clean_rx handler before we do so.
3654          */
3655         e1000_configure(adapter);
3656
3657         err = e1000_request_irq(adapter);
3658         if (err)
3659                 goto err_req_irq;
3660
3661         /*
3662          * Work around PCIe errata with MSI interrupts causing some chipsets to
3663          * ignore e1000e MSI messages, which means we need to test our MSI
3664          * interrupt now
3665          */
3666         if (adapter->int_mode != E1000E_INT_MODE_LEGACY) {
3667                 err = e1000_test_msi(adapter);
3668                 if (err) {
3669                         e_err("Interrupt allocation failed\n");
3670                         goto err_req_irq;
3671                 }
3672         }
3673
3674         /* From here on the code is the same as e1000e_up() */
3675         clear_bit(__E1000_DOWN, &adapter->state);
3676
3677         napi_enable(&adapter->napi);
3678
3679         e1000_irq_enable(adapter);
3680
3681         netif_start_queue(netdev);
3682
3683         adapter->idle_check = true;
3684         pm_runtime_put(&pdev->dev);
3685
3686         /* fire a link status change interrupt to start the watchdog */
3687         if (adapter->msix_entries)
3688                 ew32(ICS, E1000_ICS_LSC | E1000_ICR_OTHER);
3689         else
3690                 ew32(ICS, E1000_ICS_LSC);
3691
3692         return 0;
3693
3694 err_req_irq:
3695         e1000e_release_hw_control(adapter);
3696         e1000_power_down_phy(adapter);
3697         e1000e_free_rx_resources(adapter);
3698 err_setup_rx:
3699         e1000e_free_tx_resources(adapter);
3700 err_setup_tx:
3701         e1000e_reset(adapter);
3702         pm_runtime_put_sync(&pdev->dev);
3703
3704         return err;
3705 }
3706
3707 /**
3708  * e1000_close - Disables a network interface
3709  * @netdev: network interface device structure
3710  *
3711  * Returns 0, this is not allowed to fail
3712  *
3713  * The close entry point is called when an interface is de-activated
3714  * by the OS.  The hardware is still under the drivers control, but
3715  * needs to be disabled.  A global MAC reset is issued to stop the
3716  * hardware, and all transmit and receive resources are freed.
3717  **/
3718 static int e1000_close(struct net_device *netdev)
3719 {
3720         struct e1000_adapter *adapter = netdev_priv(netdev);
3721         struct pci_dev *pdev = adapter->pdev;
3722
3723         WARN_ON(test_bit(__E1000_RESETTING, &adapter->state));
3724
3725         pm_runtime_get_sync(&pdev->dev);
3726
3727         if (!test_bit(__E1000_DOWN, &adapter->state)) {
3728                 e1000e_down(adapter);
3729                 e1000_free_irq(adapter);
3730         }
3731         e1000_power_down_phy(adapter);
3732
3733         e1000e_free_tx_resources(adapter);
3734         e1000e_free_rx_resources(adapter);
3735
3736         /*
3737          * kill manageability vlan ID if supported, but not if a vlan with
3738          * the same ID is registered on the host OS (let 8021q kill it)
3739          */
3740         if (adapter->hw.mng_cookie.status &
3741             E1000_MNG_DHCP_COOKIE_STATUS_VLAN)
3742                 e1000_vlan_rx_kill_vid(netdev, adapter->mng_vlan_id);
3743
3744         /*
3745          * If AMT is enabled, let the firmware know that the network
3746          * interface is now closed
3747          */
3748         if ((adapter->flags & FLAG_HAS_AMT) &&
3749             !test_bit(__E1000_TESTING, &adapter->state))
3750                 e1000e_release_hw_control(adapter);
3751
3752         if ((adapter->flags & FLAG_HAS_ERT) ||
3753             (adapter->hw.mac.type == e1000_pch2lan))
3754                 pm_qos_remove_request(&adapter->netdev->pm_qos_req);
3755
3756         pm_runtime_put_sync(&pdev->dev);
3757
3758         return 0;
3759 }
3760 /**
3761  * e1000_set_mac - Change the Ethernet Address of the NIC
3762  * @netdev: network interface device structure
3763  * @p: pointer to an address structure
3764  *
3765  * Returns 0 on success, negative on failure
3766  **/
3767 static int e1000_set_mac(struct net_device *netdev, void *p)
3768 {
3769         struct e1000_adapter *adapter = netdev_priv(netdev);
3770         struct sockaddr *addr = p;
3771
3772         if (!is_valid_ether_addr(addr->sa_data))
3773                 return -EADDRNOTAVAIL;
3774
3775         memcpy(netdev->dev_addr, addr->sa_data, netdev->addr_len);
3776         memcpy(adapter->hw.mac.addr, addr->sa_data, netdev->addr_len);
3777
3778         e1000e_rar_set(&adapter->hw, adapter->hw.mac.addr, 0);
3779
3780         if (adapter->flags & FLAG_RESET_OVERWRITES_LAA) {
3781                 /* activate the work around */
3782                 e1000e_set_laa_state_82571(&adapter->hw, 1);
3783
3784                 /*
3785                  * Hold a copy of the LAA in RAR[14] This is done so that
3786                  * between the time RAR[0] gets clobbered  and the time it
3787                  * gets fixed (in e1000_watchdog), the actual LAA is in one
3788                  * of the RARs and no incoming packets directed to this port
3789                  * are dropped. Eventually the LAA will be in RAR[0] and
3790                  * RAR[14]
3791                  */
3792                 e1000e_rar_set(&adapter->hw,
3793                               adapter->hw.mac.addr,
3794                               adapter->hw.mac.rar_entry_count - 1);
3795         }
3796
3797         return 0;
3798 }
3799
3800 /**
3801  * e1000e_update_phy_task - work thread to update phy
3802  * @work: pointer to our work struct
3803  *
3804  * this worker thread exists because we must acquire a
3805  * semaphore to read the phy, which we could msleep while
3806  * waiting for it, and we can't msleep in a timer.
3807  **/
3808 static void e1000e_update_phy_task(struct work_struct *work)
3809 {
3810         struct e1000_adapter *adapter = container_of(work,
3811                                         struct e1000_adapter, update_phy_task);
3812
3813         if (test_bit(__E1000_DOWN, &adapter->state))
3814                 return;
3815
3816         e1000_get_phy_info(&adapter->hw);
3817 }
3818
3819 /*
3820  * Need to wait a few seconds after link up to get diagnostic information from
3821  * the phy
3822  */
3823 static void e1000_update_phy_info(unsigned long data)
3824 {
3825         struct e1000_adapter *adapter = (struct e1000_adapter *) data;
3826
3827         if (test_bit(__E1000_DOWN, &adapter->state))
3828                 return;
3829
3830         schedule_work(&adapter->update_phy_task);
3831 }
3832
3833 /**
3834  * e1000e_update_phy_stats - Update the PHY statistics counters
3835  * @adapter: board private structure
3836  *
3837  * Read/clear the upper 16-bit PHY registers and read/accumulate lower
3838  **/
3839 static void e1000e_update_phy_stats(struct e1000_adapter *adapter)
3840 {
3841         struct e1000_hw *hw = &adapter->hw;
3842         s32 ret_val;
3843         u16 phy_data;
3844
3845         ret_val = hw->phy.ops.acquire(hw);
3846         if (ret_val)
3847                 return;
3848
3849         /*
3850          * A page set is expensive so check if already on desired page.
3851          * If not, set to the page with the PHY status registers.
3852          */
3853         hw->phy.addr = 1;
3854         ret_val = e1000e_read_phy_reg_mdic(hw, IGP01E1000_PHY_PAGE_SELECT,
3855                                            &phy_data);
3856         if (ret_val)
3857                 goto release;
3858         if (phy_data != (HV_STATS_PAGE << IGP_PAGE_SHIFT)) {
3859                 ret_val = hw->phy.ops.set_page(hw,
3860                                                HV_STATS_PAGE << IGP_PAGE_SHIFT);
3861                 if (ret_val)
3862                         goto release;
3863         }
3864
3865         /* Single Collision Count */
3866         hw->phy.ops.read_reg_page(hw, HV_SCC_UPPER, &phy_data);
3867         ret_val = hw->phy.ops.read_reg_page(hw, HV_SCC_LOWER, &phy_data);
3868         if (!ret_val)
3869                 adapter->stats.scc += phy_data;
3870
3871         /* Excessive Collision Count */
3872         hw->phy.ops.read_reg_page(hw, HV_ECOL_UPPER, &phy_data);
3873         ret_val = hw->phy.ops.read_reg_page(hw, HV_ECOL_LOWER, &phy_data);
3874         if (!ret_val)
3875                 adapter->stats.ecol += phy_data;
3876
3877         /* Multiple Collision Count */
3878         hw->phy.ops.read_reg_page(hw, HV_MCC_UPPER, &phy_data);
3879         ret_val = hw->phy.ops.read_reg_page(hw, HV_MCC_LOWER, &phy_data);
3880         if (!ret_val)
3881                 adapter->stats.mcc += phy_data;
3882
3883         /* Late Collision Count */
3884         hw->phy.ops.read_reg_page(hw, HV_LATECOL_UPPER, &phy_data);
3885         ret_val = hw->phy.ops.read_reg_page(hw, HV_LATECOL_LOWER, &phy_data);
3886         if (!ret_val)
3887                 adapter->stats.latecol += phy_data;
3888
3889         /* Collision Count - also used for adaptive IFS */
3890         hw->phy.ops.read_reg_page(hw, HV_COLC_UPPER, &phy_data);
3891         ret_val = hw->phy.ops.read_reg_page(hw, HV_COLC_LOWER, &phy_data);
3892         if (!ret_val)
3893                 hw->mac.collision_delta = phy_data;
3894
3895         /* Defer Count */
3896         hw->phy.ops.read_reg_page(hw, HV_DC_UPPER, &phy_data);
3897         ret_val = hw->phy.ops.read_reg_page(hw, HV_DC_LOWER, &phy_data);
3898         if (!ret_val)
3899                 adapter->stats.dc += phy_data;
3900
3901         /* Transmit with no CRS */
3902         hw->phy.ops.read_reg_page(hw, HV_TNCRS_UPPER, &phy_data);
3903         ret_val = hw->phy.ops.read_reg_page(hw, HV_TNCRS_LOWER, &phy_data);
3904         if (!ret_val)
3905                 adapter->stats.tncrs += phy_data;
3906
3907 release:
3908         hw->phy.ops.release(hw);
3909 }
3910
3911 /**
3912  * e1000e_update_stats - Update the board statistics counters
3913  * @adapter: board private structure
3914  **/
3915 static void e1000e_update_stats(struct e1000_adapter *adapter)
3916 {
3917         struct net_device *netdev = adapter->netdev;
3918         struct e1000_hw *hw = &adapter->hw;
3919         struct pci_dev *pdev = adapter->pdev;
3920
3921         /*
3922          * Prevent stats update while adapter is being reset, or if the pci
3923          * connection is down.
3924          */
3925         if (adapter->link_speed == 0)
3926                 return;
3927         if (pci_channel_offline(pdev))
3928                 return;
3929
3930         adapter->stats.crcerrs += er32(CRCERRS);
3931         adapter->stats.gprc += er32(GPRC);
3932         adapter->stats.gorc += er32(GORCL);
3933         er32(GORCH); /* Clear gorc */
3934         adapter->stats.bprc += er32(BPRC);
3935         adapter->stats.mprc += er32(MPRC);
3936         adapter->stats.roc += er32(ROC);
3937
3938         adapter->stats.mpc += er32(MPC);
3939
3940         /* Half-duplex statistics */
3941         if (adapter->link_duplex == HALF_DUPLEX) {
3942                 if (adapter->flags2 & FLAG2_HAS_PHY_STATS) {
3943                         e1000e_update_phy_stats(adapter);
3944                 } else {
3945                         adapter->stats.scc += er32(SCC);
3946                         adapter->stats.ecol += er32(ECOL);
3947                         adapter->stats.mcc += er32(MCC);
3948                         adapter->stats.latecol += er32(LATECOL);
3949                         adapter->stats.dc += er32(DC);
3950
3951                         hw->mac.collision_delta = er32(COLC);
3952
3953                         if ((hw->mac.type != e1000_82574) &&
3954                             (hw->mac.type != e1000_82583))
3955                                 adapter->stats.tncrs += er32(TNCRS);
3956                 }
3957                 adapter->stats.colc += hw->mac.collision_delta;
3958         }
3959
3960         adapter->stats.xonrxc += er32(XONRXC);
3961         adapter->stats.xontxc += er32(XONTXC);
3962         adapter->stats.xoffrxc += er32(XOFFRXC);
3963         adapter->stats.xofftxc += er32(XOFFTXC);
3964         adapter->stats.gptc += er32(GPTC);
3965         adapter->stats.gotc += er32(GOTCL);
3966         er32(GOTCH); /* Clear gotc */
3967         adapter->stats.rnbc += er32(RNBC);
3968         adapter->stats.ruc += er32(RUC);
3969
3970         adapter->stats.mptc += er32(MPTC);
3971         adapter->stats.bptc += er32(BPTC);
3972
3973         /* used for adaptive IFS */
3974
3975         hw->mac.tx_packet_delta = er32(TPT);
3976         adapter->stats.tpt += hw->mac.tx_packet_delta;
3977
3978         adapter->stats.algnerrc += er32(ALGNERRC);
3979         adapter->stats.rxerrc += er32(RXERRC);
3980         adapter->stats.cexterr += er32(CEXTERR);
3981         adapter->stats.tsctc += er32(TSCTC);
3982         adapter->stats.tsctfc += er32(TSCTFC);
3983
3984         /* Fill out the OS statistics structure */
3985         netdev->stats.multicast = adapter->stats.mprc;
3986         netdev->stats.collisions = adapter->stats.colc;
3987
3988         /* Rx Errors */
3989
3990         /*
3991          * RLEC on some newer hardware can be incorrect so build
3992          * our own version based on RUC and ROC
3993          */
3994         netdev->stats.rx_errors = adapter->stats.rxerrc +
3995                 adapter->stats.crcerrs + adapter->stats.algnerrc +
3996                 adapter->stats.ruc + adapter->stats.roc +
3997                 adapter->stats.cexterr;
3998         netdev->stats.rx_length_errors = adapter->stats.ruc +
3999                                               adapter->stats.roc;
4000         netdev->stats.rx_crc_errors = adapter->stats.crcerrs;
4001         netdev->stats.rx_frame_errors = adapter->stats.algnerrc;
4002         netdev->stats.rx_missed_errors = adapter->stats.mpc;
4003
4004         /* Tx Errors */
4005         netdev->stats.tx_errors = adapter->stats.ecol +
4006                                        adapter->stats.latecol;
4007         netdev->stats.tx_aborted_errors = adapter->stats.ecol;
4008         netdev->stats.tx_window_errors = adapter->stats.latecol;
4009         netdev->stats.tx_carrier_errors = adapter->stats.tncrs;
4010
4011         /* Tx Dropped needs to be maintained elsewhere */
4012
4013         /* Management Stats */
4014         adapter->stats.mgptc += er32(MGTPTC);
4015         adapter->stats.mgprc += er32(MGTPRC);
4016         adapter->stats.mgpdc += er32(MGTPDC);
4017 }
4018
4019 /**
4020  * e1000_phy_read_status - Update the PHY register status snapshot
4021  * @adapter: board private structure
4022  **/
4023 static void e1000_phy_read_status(struct e1000_adapter *adapter)
4024 {
4025         struct e1000_hw *hw = &adapter->hw;
4026         struct e1000_phy_regs *phy = &adapter->phy_regs;
4027
4028         if ((er32(STATUS) & E1000_STATUS_LU) &&
4029             (adapter->hw.phy.media_type == e1000_media_type_copper)) {
4030                 int ret_val;
4031
4032                 ret_val  = e1e_rphy(hw, PHY_CONTROL, &phy->bmcr);
4033                 ret_val |= e1e_rphy(hw, PHY_STATUS, &phy->bmsr);
4034                 ret_val |= e1e_rphy(hw, PHY_AUTONEG_ADV, &phy->advertise);
4035                 ret_val |= e1e_rphy(hw, PHY_LP_ABILITY, &phy->lpa);
4036                 ret_val |= e1e_rphy(hw, PHY_AUTONEG_EXP, &phy->expansion);
4037                 ret_val |= e1e_rphy(hw, PHY_1000T_CTRL, &phy->ctrl1000);
4038                 ret_val |= e1e_rphy(hw, PHY_1000T_STATUS, &phy->stat1000);
4039                 ret_val |= e1e_rphy(hw, PHY_EXT_STATUS, &phy->estatus);
4040                 if (ret_val)
4041                         e_warn("Error reading PHY register\n");
4042         } else {
4043                 /*
4044                  * Do not read PHY registers if link is not up
4045                  * Set values to typical power-on defaults
4046                  */
4047                 phy->bmcr = (BMCR_SPEED1000 | BMCR_ANENABLE | BMCR_FULLDPLX);
4048                 phy->bmsr = (BMSR_100FULL | BMSR_100HALF | BMSR_10FULL |
4049                              BMSR_10HALF | BMSR_ESTATEN | BMSR_ANEGCAPABLE |
4050                              BMSR_ERCAP);
4051                 phy->advertise = (ADVERTISE_PAUSE_ASYM | ADVERTISE_PAUSE_CAP |
4052                                   ADVERTISE_ALL | ADVERTISE_CSMA);
4053                 phy->lpa = 0;
4054                 phy->expansion = EXPANSION_ENABLENPAGE;
4055                 phy->ctrl1000 = ADVERTISE_1000FULL;
4056                 phy->stat1000 = 0;
4057                 phy->estatus = (ESTATUS_1000_TFULL | ESTATUS_1000_THALF);
4058         }
4059 }
4060
4061 static void e1000_print_link_info(struct e1000_adapter *adapter)
4062 {
4063         struct e1000_hw *hw = &adapter->hw;
4064         u32 ctrl = er32(CTRL);
4065
4066         /* Link status message must follow this format for user tools */
4067         printk(KERN_INFO "e1000e: %s NIC Link is Up %d Mbps %s, "
4068                "Flow Control: %s\n",
4069                adapter->netdev->name,
4070                adapter->link_speed,
4071                (adapter->link_duplex == FULL_DUPLEX) ?
4072                "Full Duplex" : "Half Duplex",
4073                ((ctrl & E1000_CTRL_TFCE) && (ctrl & E1000_CTRL_RFCE)) ?
4074                "Rx/Tx" :
4075                ((ctrl & E1000_CTRL_RFCE) ? "Rx" :
4076                 ((ctrl & E1000_CTRL_TFCE) ? "Tx" : "None")));
4077 }
4078
4079 static bool e1000e_has_link(struct e1000_adapter *adapter)
4080 {
4081         struct e1000_hw *hw = &adapter->hw;
4082         bool link_active = 0;
4083         s32 ret_val = 0;
4084
4085         /*
4086          * get_link_status is set on LSC (link status) interrupt or
4087          * Rx sequence error interrupt.  get_link_status will stay
4088          * false until the check_for_link establishes link
4089          * for copper adapters ONLY
4090          */
4091         switch (hw->phy.media_type) {
4092         case e1000_media_type_copper:
4093                 if (hw->mac.get_link_status) {
4094                         ret_val = hw->mac.ops.check_for_link(hw);
4095                         link_active = !hw->mac.get_link_status;
4096                 } else {
4097                         link_active = 1;
4098                 }
4099                 break;
4100         case e1000_media_type_fiber:
4101                 ret_val = hw->mac.ops.check_for_link(hw);
4102                 link_active = !!(er32(STATUS) & E1000_STATUS_LU);
4103                 break;
4104         case e1000_media_type_internal_serdes:
4105                 ret_val = hw->mac.ops.check_for_link(hw);
4106                 link_active = adapter->hw.mac.serdes_has_link;
4107                 break;
4108         default:
4109         case e1000_media_type_unknown:
4110                 break;
4111         }
4112
4113         if ((ret_val == E1000_ERR_PHY) && (hw->phy.type == e1000_phy_igp_3) &&
4114             (er32(CTRL) & E1000_PHY_CTRL_GBE_DISABLE)) {
4115                 /* See e1000_kmrn_lock_loss_workaround_ich8lan() */
4116                 e_info("Gigabit has been disabled, downgrading speed\n");
4117         }
4118
4119         return link_active;
4120 }
4121
4122 static void e1000e_enable_receives(struct e1000_adapter *adapter)
4123 {
4124         /* make sure the receive unit is started */
4125         if ((adapter->flags & FLAG_RX_NEEDS_RESTART) &&
4126             (adapter->flags & FLAG_RX_RESTART_NOW)) {
4127                 struct e1000_hw *hw = &adapter->hw;
4128                 u32 rctl = er32(RCTL);
4129                 ew32(RCTL, rctl | E1000_RCTL_EN);
4130                 adapter->flags &= ~FLAG_RX_RESTART_NOW;
4131         }
4132 }
4133
4134 static void e1000e_check_82574_phy_workaround(struct e1000_adapter *adapter)
4135 {
4136         struct e1000_hw *hw = &adapter->hw;
4137
4138         /*
4139          * With 82574 controllers, PHY needs to be checked periodically
4140          * for hung state and reset, if two calls return true
4141          */
4142         if (e1000_check_phy_82574(hw))
4143                 adapter->phy_hang_count++;
4144         else
4145                 adapter->phy_hang_count = 0;
4146
4147         if (adapter->phy_hang_count > 1) {
4148                 adapter->phy_hang_count = 0;
4149                 schedule_work(&adapter->reset_task);
4150         }
4151 }
4152
4153 /**
4154  * e1000_watchdog - Timer Call-back
4155  * @data: pointer to adapter cast into an unsigned long
4156  **/
4157 static void e1000_watchdog(unsigned long data)
4158 {
4159         struct e1000_adapter *adapter = (struct e1000_adapter *) data;
4160
4161         /* Do the rest outside of interrupt context */
4162         schedule_work(&adapter->watchdog_task);
4163
4164         /* TODO: make this use queue_delayed_work() */
4165 }
4166
4167 static void e1000_watchdog_task(struct work_struct *work)
4168 {
4169         struct e1000_adapter *adapter = container_of(work,
4170                                         struct e1000_adapter, watchdog_task);
4171         struct net_device *netdev = adapter->netdev;
4172         struct e1000_mac_info *mac = &adapter->hw.mac;
4173         struct e1000_phy_info *phy = &adapter->hw.phy;
4174         struct e1000_ring *tx_ring = adapter->tx_ring;
4175         struct e1000_hw *hw = &adapter->hw;
4176         u32 link, tctl;
4177
4178         if (test_bit(__E1000_DOWN, &adapter->state))
4179                 return;
4180
4181         link = e1000e_has_link(adapter);
4182         if ((netif_carrier_ok(netdev)) && link) {
4183                 /* Cancel scheduled suspend requests. */
4184                 pm_runtime_resume(netdev->dev.parent);
4185
4186                 e1000e_enable_receives(adapter);
4187                 goto link_up;
4188         }
4189
4190         if ((e1000e_enable_tx_pkt_filtering(hw)) &&
4191             (adapter->mng_vlan_id != adapter->hw.mng_cookie.vlan_id))
4192                 e1000_update_mng_vlan(adapter);
4193
4194         if (link) {
4195                 if (!netif_carrier_ok(netdev)) {
4196                         bool txb2b = 1;
4197
4198                         /* Cancel scheduled suspend requests. */
4199                         pm_runtime_resume(netdev->dev.parent);
4200
4201                         /* update snapshot of PHY registers on LSC */
4202                         e1000_phy_read_status(adapter);
4203                         mac->ops.get_link_up_info(&adapter->hw,
4204                                                    &adapter->link_speed,
4205                                                    &adapter->link_duplex);
4206                         e1000_print_link_info(adapter);
4207                         /*
4208                          * On supported PHYs, check for duplex mismatch only
4209                          * if link has autonegotiated at 10/100 half
4210                          */
4211                         if ((hw->phy.type == e1000_phy_igp_3 ||
4212                              hw->phy.type == e1000_phy_bm) &&
4213                             (hw->mac.autoneg == true) &&
4214                             (adapter->link_speed == SPEED_10 ||
4215                              adapter->link_speed == SPEED_100) &&
4216                             (adapter->link_duplex == HALF_DUPLEX)) {
4217                                 u16 autoneg_exp;
4218
4219                                 e1e_rphy(hw, PHY_AUTONEG_EXP, &autoneg_exp);
4220
4221                                 if (!(autoneg_exp & NWAY_ER_LP_NWAY_CAPS))
4222                                         e_info("Autonegotiated half duplex but"
4223                                                " link partner cannot autoneg. "
4224                                                " Try forcing full duplex if "
4225                                                "link gets many collisions.\n");
4226                         }
4227
4228                         /* adjust timeout factor according to speed/duplex */
4229                         adapter->tx_timeout_factor = 1;
4230                         switch (adapter->link_speed) {
4231                         case SPEED_10:
4232                                 txb2b = 0;
4233                                 adapter->tx_timeout_factor = 16;
4234                                 break;
4235                         case SPEED_100:
4236                                 txb2b = 0;
4237                                 adapter->tx_timeout_factor = 10;
4238                                 break;
4239                         }
4240
4241                         /*
4242                          * workaround: re-program speed mode bit after
4243                          * link-up event
4244                          */
4245                         if ((adapter->flags & FLAG_TARC_SPEED_MODE_BIT) &&
4246                             !txb2b) {
4247                                 u32 tarc0;
4248                                 tarc0 = er32(TARC(0));
4249                                 tarc0 &= ~SPEED_MODE_BIT;
4250                                 ew32(TARC(0), tarc0);
4251                         }
4252
4253                         /*
4254                          * disable TSO for pcie and 10/100 speeds, to avoid
4255                          * some hardware issues
4256                          */
4257                         if (!(adapter->flags & FLAG_TSO_FORCE)) {
4258                                 switch (adapter->link_speed) {
4259                                 case SPEED_10:
4260                                 case SPEED_100:
4261                                         e_info("10/100 speed: disabling TSO\n");
4262                                         netdev->features &= ~NETIF_F_TSO;
4263                                         netdev->features &= ~NETIF_F_TSO6;
4264                                         break;
4265                                 case SPEED_1000:
4266                                         netdev->features |= NETIF_F_TSO;
4267                                         netdev->features |= NETIF_F_TSO6;
4268                                         break;
4269                                 default:
4270                                         /* oops */
4271                                         break;
4272                                 }
4273                         }
4274
4275                         /*
4276                          * enable transmits in the hardware, need to do this
4277                          * after setting TARC(0)
4278                          */
4279                         tctl = er32(TCTL);
4280                         tctl |= E1000_TCTL_EN;
4281                         ew32(TCTL, tctl);
4282
4283                         /*
4284                          * Perform any post-link-up configuration before
4285                          * reporting link up.
4286                          */
4287                         if (phy->ops.cfg_on_link_up)
4288                                 phy->ops.cfg_on_link_up(hw);
4289
4290                         netif_carrier_on(netdev);
4291
4292                         if (!test_bit(__E1000_DOWN, &adapter->state))
4293                                 mod_timer(&adapter->phy_info_timer,
4294                                           round_jiffies(jiffies + 2 * HZ));
4295                 }
4296         } else {
4297                 if (netif_carrier_ok(netdev)) {
4298                         adapter->link_speed = 0;
4299                         adapter->link_duplex = 0;
4300                         /* Link status message must follow this format */
4301                         printk(KERN_INFO "e1000e: %s NIC Link is Down\n",
4302                                adapter->netdev->name);
4303                         netif_carrier_off(netdev);
4304                         if (!test_bit(__E1000_DOWN, &adapter->state))
4305                                 mod_timer(&adapter->phy_info_timer,
4306                                           round_jiffies(jiffies + 2 * HZ));
4307
4308                         if (adapter->flags & FLAG_RX_NEEDS_RESTART)
4309                                 schedule_work(&adapter->reset_task);
4310                         else
4311                                 pm_schedule_suspend(netdev->dev.parent,
4312                                                         LINK_TIMEOUT);
4313                 }
4314         }
4315
4316 link_up:
4317         spin_lock(&adapter->stats64_lock);
4318         e1000e_update_stats(adapter);
4319
4320         mac->tx_packet_delta = adapter->stats.tpt - adapter->tpt_old;
4321         adapter->tpt_old = adapter->stats.tpt;
4322         mac->collision_delta = adapter->stats.colc - adapter->colc_old;
4323         adapter->colc_old = adapter->stats.colc;
4324
4325         adapter->gorc = adapter->stats.gorc - adapter->gorc_old;
4326         adapter->gorc_old = adapter->stats.gorc;
4327         adapter->gotc = adapter->stats.gotc - adapter->gotc_old;
4328         adapter->gotc_old = adapter->stats.gotc;
4329         spin_unlock(&adapter->stats64_lock);
4330
4331         e1000e_update_adaptive(&adapter->hw);
4332
4333         if (!netif_carrier_ok(netdev) &&
4334             (e1000_desc_unused(tx_ring) + 1 < tx_ring->count)) {
4335                 /*
4336                  * We've lost link, so the controller stops DMA,
4337                  * but we've got queued Tx work that's never going
4338                  * to get done, so reset controller to flush Tx.
4339                  * (Do the reset outside of interrupt context).
4340                  */
4341                 schedule_work(&adapter->reset_task);
4342                 /* return immediately since reset is imminent */
4343                 return;
4344         }
4345
4346         /* Simple mode for Interrupt Throttle Rate (ITR) */
4347         if (adapter->itr_setting == 4) {
4348                 /*
4349                  * Symmetric Tx/Rx gets a reduced ITR=2000;
4350                  * Total asymmetrical Tx or Rx gets ITR=8000;
4351                  * everyone else is between 2000-8000.
4352                  */
4353                 u32 goc = (adapter->gotc + adapter->gorc) / 10000;
4354                 u32 dif = (adapter->gotc > adapter->gorc ?
4355                             adapter->gotc - adapter->gorc :
4356                             adapter->gorc - adapter->gotc) / 10000;
4357                 u32 itr = goc > 0 ? (dif * 6000 / goc + 2000) : 8000;
4358
4359                 ew32(ITR, 1000000000 / (itr * 256));
4360         }
4361
4362         /* Cause software interrupt to ensure Rx ring is cleaned */
4363         if (adapter->msix_entries)
4364                 ew32(ICS, adapter->rx_ring->ims_val);
4365         else
4366                 ew32(ICS, E1000_ICS_RXDMT0);
4367
4368         /* flush pending descriptors to memory before detecting Tx hang */
4369         e1000e_flush_descriptors(adapter);
4370
4371         /* Force detection of hung controller every watchdog period */
4372         adapter->detect_tx_hung = 1;
4373
4374         /*
4375          * With 82571 controllers, LAA may be overwritten due to controller
4376          * reset from the other port. Set the appropriate LAA in RAR[0]
4377          */
4378         if (e1000e_get_laa_state_82571(hw))
4379                 e1000e_rar_set(hw, adapter->hw.mac.addr, 0);
4380
4381         if (adapter->flags2 & FLAG2_CHECK_PHY_HANG)
4382                 e1000e_check_82574_phy_workaround(adapter);
4383
4384         /* Reset the timer */
4385         if (!test_bit(__E1000_DOWN, &adapter->state))
4386                 mod_timer(&adapter->watchdog_timer,
4387                           round_jiffies(jiffies + 2 * HZ));
4388 }
4389
4390 #define E1000_TX_FLAGS_CSUM             0x00000001
4391 #define E1000_TX_FLAGS_VLAN             0x00000002
4392 #define E1000_TX_FLAGS_TSO              0x00000004
4393 #define E1000_TX_FLAGS_IPV4             0x00000008
4394 #define E1000_TX_FLAGS_VLAN_MASK        0xffff0000
4395 #define E1000_TX_FLAGS_VLAN_SHIFT       16
4396
4397 static int e1000_tso(struct e1000_adapter *adapter,
4398                      struct sk_buff *skb)
4399 {
4400         struct e1000_ring *tx_ring = adapter->tx_ring;
4401         struct e1000_context_desc *context_desc;
4402         struct e1000_buffer *buffer_info;
4403         unsigned int i;
4404         u32 cmd_length = 0;
4405         u16 ipcse = 0, tucse, mss;
4406         u8 ipcss, ipcso, tucss, tucso, hdr_len;
4407
4408         if (!skb_is_gso(skb))
4409                 return 0;
4410
4411         if (skb_header_cloned(skb)) {
4412                 int err = pskb_expand_head(skb, 0, 0, GFP_ATOMIC);
4413
4414                 if (err)
4415                         return err;
4416         }
4417
4418         hdr_len = skb_transport_offset(skb) + tcp_hdrlen(skb);
4419         mss = skb_shinfo(skb)->gso_size;
4420         if (skb->protocol == htons(ETH_P_IP)) {
4421                 struct iphdr *iph = ip_hdr(skb);
4422                 iph->tot_len = 0;
4423                 iph->check = 0;
4424                 tcp_hdr(skb)->check = ~csum_tcpudp_magic(iph->saddr, iph->daddr,
4425                                                          0, IPPROTO_TCP, 0);
4426                 cmd_length = E1000_TXD_CMD_IP;
4427                 ipcse = skb_transport_offset(skb) - 1;
4428         } else if (skb_is_gso_v6(skb)) {
4429                 ipv6_hdr(skb)->payload_len = 0;
4430                 tcp_hdr(skb)->check = ~csum_ipv6_magic(&ipv6_hdr(skb)->saddr,
4431                                                        &ipv6_hdr(skb)->daddr,
4432                                                        0, IPPROTO_TCP, 0);
4433                 ipcse = 0;
4434         }
4435         ipcss = skb_network_offset(skb);
4436         ipcso = (void *)&(ip_hdr(skb)->check) - (void *)skb->data;
4437         tucss = skb_transport_offset(skb);
4438         tucso = (void *)&(tcp_hdr(skb)->check) - (void *)skb->data;
4439         tucse = 0;
4440
4441         cmd_length |= (E1000_TXD_CMD_DEXT | E1000_TXD_CMD_TSE |
4442                        E1000_TXD_CMD_TCP | (skb->len - (hdr_len)));
4443
4444         i = tx_ring->next_to_use;
4445         context_desc = E1000_CONTEXT_DESC(*tx_ring, i);
4446         buffer_info = &tx_ring->buffer_info[i];
4447
4448         context_desc->lower_setup.ip_fields.ipcss  = ipcss;
4449         context_desc->lower_setup.ip_fields.ipcso  = ipcso;
4450         context_desc->lower_setup.ip_fields.ipcse  = cpu_to_le16(ipcse);
4451         context_desc->upper_setup.tcp_fields.tucss = tucss;
4452         context_desc->upper_setup.tcp_fields.tucso = tucso;
4453         context_desc->upper_setup.tcp_fields.tucse = cpu_to_le16(tucse);
4454         context_desc->tcp_seg_setup.fields.mss     = cpu_to_le16(mss);
4455         context_desc->tcp_seg_setup.fields.hdr_len = hdr_len;
4456         context_desc->cmd_and_length = cpu_to_le32(cmd_length);
4457
4458         buffer_info->time_stamp = jiffies;
4459         buffer_info->next_to_watch = i;
4460
4461         i++;
4462         if (i == tx_ring->count)
4463                 i = 0;
4464         tx_ring->next_to_use = i;
4465
4466         return 1;
4467 }
4468
4469 static bool e1000_tx_csum(struct e1000_adapter *adapter, struct sk_buff *skb)
4470 {
4471         struct e1000_ring *tx_ring = adapter->tx_ring;
4472         struct e1000_context_desc *context_desc;
4473         struct e1000_buffer *buffer_info;
4474         unsigned int i;
4475         u8 css;
4476         u32 cmd_len = E1000_TXD_CMD_DEXT;
4477         __be16 protocol;
4478
4479         if (skb->ip_summed != CHECKSUM_PARTIAL)
4480                 return 0;
4481
4482         if (skb->protocol == cpu_to_be16(ETH_P_8021Q))
4483                 protocol = vlan_eth_hdr(skb)->h_vlan_encapsulated_proto;
4484         else
4485                 protocol = skb->protocol;
4486
4487         switch (protocol) {
4488         case cpu_to_be16(ETH_P_IP):
4489                 if (ip_hdr(skb)->protocol == IPPROTO_TCP)
4490                         cmd_len |= E1000_TXD_CMD_TCP;
4491                 break;
4492         case cpu_to_be16(ETH_P_IPV6):
4493                 /* XXX not handling all IPV6 headers */
4494                 if (ipv6_hdr(skb)->nexthdr == IPPROTO_TCP)
4495                         cmd_len |= E1000_TXD_CMD_TCP;
4496                 break;
4497         default:
4498                 if (unlikely(net_ratelimit()))
4499                         e_warn("checksum_partial proto=%x!\n",
4500                                be16_to_cpu(protocol));
4501                 break;
4502         }
4503
4504         css = skb_checksum_start_offset(skb);
4505
4506         i = tx_ring->next_to_use;
4507         buffer_info = &tx_ring->buffer_info[i];
4508         context_desc = E1000_CONTEXT_DESC(*tx_ring, i);
4509
4510         context_desc->lower_setup.ip_config = 0;
4511         context_desc->upper_setup.tcp_fields.tucss = css;
4512         context_desc->upper_setup.tcp_fields.tucso =
4513                                 css + skb->csum_offset;
4514         context_desc->upper_setup.tcp_fields.tucse = 0;
4515         context_desc->tcp_seg_setup.data = 0;
4516         context_desc->cmd_and_length = cpu_to_le32(cmd_len);
4517
4518         buffer_info->time_stamp = jiffies;
4519         buffer_info->next_to_watch = i;
4520
4521         i++;
4522         if (i == tx_ring->count)
4523                 i = 0;
4524         tx_ring->next_to_use = i;
4525
4526         return 1;
4527 }
4528
4529 #define E1000_MAX_PER_TXD       8192
4530 #define E1000_MAX_TXD_PWR       12
4531
4532 static int e1000_tx_map(struct e1000_adapter *adapter,
4533                         struct sk_buff *skb, unsigned int first,
4534                         unsigned int max_per_txd, unsigned int nr_frags,
4535                         unsigned int mss)
4536 {
4537         struct e1000_ring *tx_ring = adapter->tx_ring;
4538         struct pci_dev *pdev = adapter->pdev;
4539         struct e1000_buffer *buffer_info;
4540         unsigned int len = skb_headlen(skb);
4541         unsigned int offset = 0, size, count = 0, i;
4542         unsigned int f, bytecount, segs;
4543
4544         i = tx_ring->next_to_use;
4545
4546         while (len) {
4547                 buffer_info = &tx_ring->buffer_info[i];
4548                 size = min(len, max_per_txd);
4549
4550                 buffer_info->length = size;
4551                 buffer_info->time_stamp = jiffies;
4552                 buffer_info->next_to_watch = i;
4553                 buffer_info->dma = dma_map_single(&pdev->dev,
4554                                                   skb->data + offset,
4555                                                   size, DMA_TO_DEVICE);
4556                 buffer_info->mapped_as_page = false;
4557                 if (dma_mapping_error(&pdev->dev, buffer_info->dma))
4558                         goto dma_error;
4559
4560                 len -= size;
4561                 offset += size;
4562                 count++;
4563
4564                 if (len) {
4565                         i++;
4566                         if (i == tx_ring->count)
4567                                 i = 0;
4568                 }
4569         }
4570
4571         for (f = 0; f < nr_frags; f++) {
4572                 struct skb_frag_struct *frag;
4573
4574                 frag = &skb_shinfo(skb)->frags[f];
4575                 len = frag->size;
4576                 offset = frag->page_offset;
4577
4578                 while (len) {
4579                         i++;
4580                         if (i == tx_ring->count)
4581                                 i = 0;
4582
4583                         buffer_info = &tx_ring->buffer_info[i];
4584                         size = min(len, max_per_txd);
4585
4586                         buffer_info->length = size;
4587                         buffer_info->time_stamp = jiffies;
4588                         buffer_info->next_to_watch = i;
4589                         buffer_info->dma = dma_map_page(&pdev->dev, frag->page,
4590                                                         offset, size,
4591                                                         DMA_TO_DEVICE);
4592                         buffer_info->mapped_as_page = true;
4593                         if (dma_mapping_error(&pdev->dev, buffer_info->dma))
4594                                 goto dma_error;
4595
4596                         len -= size;
4597                         offset += size;
4598                         count++;
4599                 }
4600         }
4601
4602         segs = skb_shinfo(skb)->gso_segs ? : 1;
4603         /* multiply data chunks by size of headers */
4604         bytecount = ((segs - 1) * skb_headlen(skb)) + skb->len;
4605
4606         tx_ring->buffer_info[i].skb = skb;
4607         tx_ring->buffer_info[i].segs = segs;
4608         tx_ring->buffer_info[i].bytecount = bytecount;
4609         tx_ring->buffer_info[first].next_to_watch = i;
4610
4611         return count;
4612
4613 dma_error:
4614         dev_err(&pdev->dev, "Tx DMA map failed\n");
4615         buffer_info->dma = 0;
4616         if (count)
4617                 count--;
4618
4619         while (count--) {
4620                 if (i == 0)
4621                         i += tx_ring->count;
4622                 i--;
4623                 buffer_info = &tx_ring->buffer_info[i];
4624                 e1000_put_txbuf(adapter, buffer_info);
4625         }
4626
4627         return 0;
4628 }
4629
4630 static void e1000_tx_queue(struct e1000_adapter *adapter,
4631                            int tx_flags, int count)
4632 {
4633         struct e1000_ring *tx_ring = adapter->tx_ring;
4634         struct e1000_tx_desc *tx_desc = NULL;
4635         struct e1000_buffer *buffer_info;
4636         u32 txd_upper = 0, txd_lower = E1000_TXD_CMD_IFCS;
4637         unsigned int i;
4638
4639         if (tx_flags & E1000_TX_FLAGS_TSO) {
4640                 txd_lower |= E1000_TXD_CMD_DEXT | E1000_TXD_DTYP_D |
4641                              E1000_TXD_CMD_TSE;
4642                 txd_upper |= E1000_TXD_POPTS_TXSM << 8;
4643
4644                 if (tx_flags & E1000_TX_FLAGS_IPV4)
4645                         txd_upper |= E1000_TXD_POPTS_IXSM << 8;
4646         }
4647
4648         if (tx_flags & E1000_TX_FLAGS_CSUM) {
4649                 txd_lower |= E1000_TXD_CMD_DEXT | E1000_TXD_DTYP_D;
4650                 txd_upper |= E1000_TXD_POPTS_TXSM << 8;
4651         }
4652
4653         if (tx_flags & E1000_TX_FLAGS_VLAN) {
4654                 txd_lower |= E1000_TXD_CMD_VLE;
4655                 txd_upper |= (tx_flags & E1000_TX_FLAGS_VLAN_MASK);
4656         }
4657
4658         i = tx_ring->next_to_use;
4659
4660         do {
4661                 buffer_info = &tx_ring->buffer_info[i];
4662                 tx_desc = E1000_TX_DESC(*tx_ring, i);
4663                 tx_desc->buffer_addr = cpu_to_le64(buffer_info->dma);
4664                 tx_desc->lower.data =
4665                         cpu_to_le32(txd_lower | buffer_info->length);
4666                 tx_desc->upper.data = cpu_to_le32(txd_upper);
4667
4668                 i++;
4669                 if (i == tx_ring->count)
4670                         i = 0;
4671         } while (--count > 0);
4672
4673         tx_desc->lower.data |= cpu_to_le32(adapter->txd_cmd);
4674
4675         /*
4676          * Force memory writes to complete before letting h/w
4677          * know there are new descriptors to fetch.  (Only
4678          * applicable for weak-ordered memory model archs,
4679          * such as IA-64).
4680          */
4681         wmb();
4682
4683         tx_ring->next_to_use = i;
4684         writel(i, adapter->hw.hw_addr + tx_ring->tail);
4685         /*
4686          * we need this if more than one processor can write to our tail
4687          * at a time, it synchronizes IO on IA64/Altix systems
4688          */
4689         mmiowb();
4690 }
4691
4692 #define MINIMUM_DHCP_PACKET_SIZE 282
4693 static int e1000_transfer_dhcp_info(struct e1000_adapter *adapter,
4694                                     struct sk_buff *skb)
4695 {
4696         struct e1000_hw *hw =  &adapter->hw;
4697         u16 length, offset;
4698
4699         if (vlan_tx_tag_present(skb)) {
4700                 if (!((vlan_tx_tag_get(skb) == adapter->hw.mng_cookie.vlan_id) &&
4701                     (adapter->hw.mng_cookie.status &
4702                         E1000_MNG_DHCP_COOKIE_STATUS_VLAN)))
4703                         return 0;
4704         }
4705
4706         if (skb->len <= MINIMUM_DHCP_PACKET_SIZE)
4707                 return 0;
4708
4709         if (((struct ethhdr *) skb->data)->h_proto != htons(ETH_P_IP))
4710                 return 0;
4711
4712         {
4713                 const struct iphdr *ip = (struct iphdr *)((u8 *)skb->data+14);
4714                 struct udphdr *udp;
4715
4716                 if (ip->protocol != IPPROTO_UDP)
4717                         return 0;
4718
4719                 udp = (struct udphdr *)((u8 *)ip + (ip->ihl << 2));
4720                 if (ntohs(udp->dest) != 67)
4721                         return 0;
4722
4723                 offset = (u8 *)udp + 8 - skb->data;
4724                 length = skb->len - offset;
4725                 return e1000e_mng_write_dhcp_info(hw, (u8 *)udp + 8, length);
4726         }
4727
4728         return 0;
4729 }
4730
4731 static int __e1000_maybe_stop_tx(struct net_device *netdev, int size)
4732 {
4733         struct e1000_adapter *adapter = netdev_priv(netdev);
4734
4735         netif_stop_queue(netdev);
4736         /*
4737          * Herbert's original patch had:
4738          *  smp_mb__after_netif_stop_queue();
4739          * but since that doesn't exist yet, just open code it.
4740          */
4741         smp_mb();
4742
4743         /*
4744          * We need to check again in a case another CPU has just
4745          * made room available.
4746          */
4747         if (e1000_desc_unused(adapter->tx_ring) < size)
4748                 return -EBUSY;
4749
4750         /* A reprieve! */
4751         netif_start_queue(netdev);
4752         ++adapter->restart_queue;
4753         return 0;
4754 }
4755
4756 static int e1000_maybe_stop_tx(struct net_device *netdev, int size)
4757 {
4758         struct e1000_adapter *adapter = netdev_priv(netdev);
4759
4760         if (e1000_desc_unused(adapter->tx_ring) >= size)
4761                 return 0;
4762         return __e1000_maybe_stop_tx(netdev, size);
4763 }
4764
4765 #define TXD_USE_COUNT(S, X) (((S) >> (X)) + 1 )
4766 static netdev_tx_t e1000_xmit_frame(struct sk_buff *skb,
4767                                     struct net_device *netdev)
4768 {
4769         struct e1000_adapter *adapter = netdev_priv(netdev);
4770         struct e1000_ring *tx_ring = adapter->tx_ring;
4771         unsigned int first;
4772         unsigned int max_per_txd = E1000_MAX_PER_TXD;
4773         unsigned int max_txd_pwr = E1000_MAX_TXD_PWR;
4774         unsigned int tx_flags = 0;
4775         unsigned int len = skb_headlen(skb);
4776         unsigned int nr_frags;
4777         unsigned int mss;
4778         int count = 0;
4779         int tso;
4780         unsigned int f;
4781
4782         if (test_bit(__E1000_DOWN, &adapter->state)) {
4783                 dev_kfree_skb_any(skb);
4784                 return NETDEV_TX_OK;
4785         }
4786
4787         if (skb->len <= 0) {
4788                 dev_kfree_skb_any(skb);
4789                 return NETDEV_TX_OK;
4790         }
4791
4792         mss = skb_shinfo(skb)->gso_size;
4793         /*
4794          * The controller does a simple calculation to
4795          * make sure there is enough room in the FIFO before
4796          * initiating the DMA for each buffer.  The calc is:
4797          * 4 = ceil(buffer len/mss).  To make sure we don't
4798          * overrun the FIFO, adjust the max buffer len if mss
4799          * drops.
4800          */
4801         if (mss) {
4802                 u8 hdr_len;
4803                 max_per_txd = min(mss << 2, max_per_txd);
4804                 max_txd_pwr = fls(max_per_txd) - 1;
4805
4806                 /*
4807                  * TSO Workaround for 82571/2/3 Controllers -- if skb->data
4808                  * points to just header, pull a few bytes of payload from
4809                  * frags into skb->data
4810                  */
4811                 hdr_len = skb_transport_offset(skb) + tcp_hdrlen(skb);
4812                 /*
4813                  * we do this workaround for ES2LAN, but it is un-necessary,
4814                  * avoiding it could save a lot of cycles
4815                  */
4816                 if (skb->data_len && (hdr_len == len)) {
4817                         unsigned int pull_size;
4818
4819                         pull_size = min((unsigned int)4, skb->data_len);
4820                         if (!__pskb_pull_tail(skb, pull_size)) {
4821                                 e_err("__pskb_pull_tail failed.\n");
4822                                 dev_kfree_skb_any(skb);
4823                                 return NETDEV_TX_OK;
4824                         }
4825                         len = skb_headlen(skb);
4826                 }
4827         }
4828
4829         /* reserve a descriptor for the offload context */
4830         if ((mss) || (skb->ip_summed == CHECKSUM_PARTIAL))
4831                 count++;
4832         count++;
4833
4834         count += TXD_USE_COUNT(len, max_txd_pwr);
4835
4836         nr_frags = skb_shinfo(skb)->nr_frags;
4837         for (f = 0; f < nr_frags; f++)
4838                 count += TXD_USE_COUNT(skb_shinfo(skb)->frags[f].size,
4839                                        max_txd_pwr);
4840
4841         if (adapter->hw.mac.tx_pkt_filtering)
4842                 e1000_transfer_dhcp_info(adapter, skb);
4843
4844         /*
4845          * need: count + 2 desc gap to keep tail from touching
4846          * head, otherwise try next time
4847          */
4848         if (e1000_maybe_stop_tx(netdev, count + 2))
4849                 return NETDEV_TX_BUSY;
4850
4851         if (vlan_tx_tag_present(skb)) {
4852                 tx_flags |= E1000_TX_FLAGS_VLAN;
4853                 tx_flags |= (vlan_tx_tag_get(skb) << E1000_TX_FLAGS_VLAN_SHIFT);
4854         }
4855
4856         first = tx_ring->next_to_use;
4857
4858         tso = e1000_tso(adapter, skb);
4859         if (tso < 0) {
4860                 dev_kfree_skb_any(skb);
4861                 return NETDEV_TX_OK;
4862         }
4863
4864         if (tso)
4865                 tx_flags |= E1000_TX_FLAGS_TSO;
4866         else if (e1000_tx_csum(adapter, skb))
4867                 tx_flags |= E1000_TX_FLAGS_CSUM;
4868
4869         /*
4870          * Old method was to assume IPv4 packet by default if TSO was enabled.
4871          * 82571 hardware supports TSO capabilities for IPv6 as well...
4872          * no longer assume, we must.
4873          */
4874         if (skb->protocol == htons(ETH_P_IP))
4875                 tx_flags |= E1000_TX_FLAGS_IPV4;
4876
4877         /* if count is 0 then mapping error has occurred */
4878         count = e1000_tx_map(adapter, skb, first, max_per_txd, nr_frags, mss);
4879         if (count) {
4880                 e1000_tx_queue(adapter, tx_flags, count);
4881                 /* Make sure there is space in the ring for the next send. */
4882                 e1000_maybe_stop_tx(netdev, MAX_SKB_FRAGS + 2);
4883
4884         } else {
4885                 dev_kfree_skb_any(skb);
4886                 tx_ring->buffer_info[first].time_stamp = 0;
4887                 tx_ring->next_to_use = first;
4888         }
4889
4890         return NETDEV_TX_OK;
4891 }
4892
4893 /**
4894  * e1000_tx_timeout - Respond to a Tx Hang
4895  * @netdev: network interface device structure
4896  **/
4897 static void e1000_tx_timeout(struct net_device *netdev)
4898 {
4899         struct e1000_adapter *adapter = netdev_priv(netdev);
4900
4901         /* Do the reset outside of interrupt context */
4902         adapter->tx_timeout_count++;
4903         schedule_work(&adapter->reset_task);
4904 }
4905
4906 static void e1000_reset_task(struct work_struct *work)
4907 {
4908         struct e1000_adapter *adapter;
4909         adapter = container_of(work, struct e1000_adapter, reset_task);
4910
4911         /* don't run the task if already down */
4912         if (test_bit(__E1000_DOWN, &adapter->state))
4913                 return;
4914
4915         if (!((adapter->flags & FLAG_RX_NEEDS_RESTART) &&
4916               (adapter->flags & FLAG_RX_RESTART_NOW))) {
4917                 e1000e_dump(adapter);
4918                 e_err("Reset adapter\n");
4919         }
4920         e1000e_reinit_locked(adapter);
4921 }
4922
4923 /**
4924  * e1000_get_stats64 - Get System Network Statistics
4925  * @netdev: network interface device structure
4926  * @stats: rtnl_link_stats64 pointer
4927  *
4928  * Returns the address of the device statistics structure.
4929  **/
4930 struct rtnl_link_stats64 *e1000e_get_stats64(struct net_device *netdev,
4931                                              struct rtnl_link_stats64 *stats)
4932 {
4933         struct e1000_adapter *adapter = netdev_priv(netdev);
4934
4935         memset(stats, 0, sizeof(struct rtnl_link_stats64));
4936         spin_lock(&adapter->stats64_lock);
4937         e1000e_update_stats(adapter);
4938         /* Fill out the OS statistics structure */
4939         stats->rx_bytes = adapter->stats.gorc;
4940         stats->rx_packets = adapter->stats.gprc;
4941         stats->tx_bytes = adapter->stats.gotc;
4942         stats->tx_packets = adapter->stats.gptc;
4943         stats->multicast = adapter->stats.mprc;
4944         stats->collisions = adapter->stats.colc;
4945
4946         /* Rx Errors */
4947
4948         /*
4949          * RLEC on some newer hardware can be incorrect so build
4950          * our own version based on RUC and ROC
4951          */
4952         stats->rx_errors = adapter->stats.rxerrc +
4953                 adapter->stats.crcerrs + adapter->stats.algnerrc +
4954                 adapter->stats.ruc + adapter->stats.roc +
4955                 adapter->stats.cexterr;
4956         stats->rx_length_errors = adapter->stats.ruc +
4957                                               adapter->stats.roc;
4958         stats->rx_crc_errors = adapter->stats.crcerrs;
4959         stats->rx_frame_errors = adapter->stats.algnerrc;
4960         stats->rx_missed_errors = adapter->stats.mpc;
4961
4962         /* Tx Errors */
4963         stats->tx_errors = adapter->stats.ecol +
4964                                        adapter->stats.latecol;
4965         stats->tx_aborted_errors = adapter->stats.ecol;
4966         stats->tx_window_errors = adapter->stats.latecol;
4967         stats->tx_carrier_errors = adapter->stats.tncrs;
4968
4969         /* Tx Dropped needs to be maintained elsewhere */
4970
4971         spin_unlock(&adapter->stats64_lock);
4972         return stats;
4973 }
4974
4975 /**
4976  * e1000_change_mtu - Change the Maximum Transfer Unit
4977  * @netdev: network interface device structure
4978  * @new_mtu: new value for maximum frame size
4979  *
4980  * Returns 0 on success, negative on failure
4981  **/
4982 static int e1000_change_mtu(struct net_device *netdev, int new_mtu)
4983 {
4984         struct e1000_adapter *adapter = netdev_priv(netdev);
4985         int max_frame = new_mtu + ETH_HLEN + ETH_FCS_LEN;
4986
4987         /* Jumbo frame support */
4988         if ((max_frame > ETH_FRAME_LEN + ETH_FCS_LEN) &&
4989             !(adapter->flags & FLAG_HAS_JUMBO_FRAMES)) {
4990                 e_err("Jumbo Frames not supported.\n");
4991                 return -EINVAL;
4992         }
4993
4994         /* Supported frame sizes */
4995         if ((new_mtu < ETH_ZLEN + ETH_FCS_LEN + VLAN_HLEN) ||
4996             (max_frame > adapter->max_hw_frame_size)) {
4997                 e_err("Unsupported MTU setting\n");
4998                 return -EINVAL;
4999         }
5000
5001         /* Jumbo frame workaround on 82579 requires CRC be stripped */
5002         if ((adapter->hw.mac.type == e1000_pch2lan) &&
5003             !(adapter->flags2 & FLAG2_CRC_STRIPPING) &&
5004             (new_mtu > ETH_DATA_LEN)) {
5005                 e_err("Jumbo Frames not supported on 82579 when CRC "
5006                       "stripping is disabled.\n");
5007                 return -EINVAL;
5008         }
5009
5010         /* 82573 Errata 17 */
5011         if (((adapter->hw.mac.type == e1000_82573) ||
5012              (adapter->hw.mac.type == e1000_82574)) &&
5013             (max_frame > ETH_FRAME_LEN + ETH_FCS_LEN)) {
5014                 adapter->flags2 |= FLAG2_DISABLE_ASPM_L1;
5015                 e1000e_disable_aspm(adapter->pdev, PCIE_LINK_STATE_L1);
5016         }
5017
5018         while (test_and_set_bit(__E1000_RESETTING, &adapter->state))
5019                 usleep_range(1000, 2000);
5020         /* e1000e_down -> e1000e_reset dependent on max_frame_size & mtu */
5021         adapter->max_frame_size = max_frame;
5022         e_info("changing MTU from %d to %d\n", netdev->mtu, new_mtu);
5023         netdev->mtu = new_mtu;
5024         if (netif_running(netdev))
5025                 e1000e_down(adapter);
5026
5027         /*
5028          * NOTE: netdev_alloc_skb reserves 16 bytes, and typically NET_IP_ALIGN
5029          * means we reserve 2 more, this pushes us to allocate from the next
5030          * larger slab size.
5031          * i.e. RXBUFFER_2048 --> size-4096 slab
5032          * However with the new *_jumbo_rx* routines, jumbo receives will use
5033          * fragmented skbs
5034          */
5035
5036         if (max_frame <= 2048)
5037                 adapter->rx_buffer_len = 2048;
5038         else
5039                 adapter->rx_buffer_len = 4096;
5040
5041         /* adjust allocation if LPE protects us, and we aren't using SBP */
5042         if ((max_frame == ETH_FRAME_LEN + ETH_FCS_LEN) ||
5043              (max_frame == ETH_FRAME_LEN + VLAN_HLEN + ETH_FCS_LEN))
5044                 adapter->rx_buffer_len = ETH_FRAME_LEN + VLAN_HLEN
5045                                          + ETH_FCS_LEN;
5046
5047         if (netif_running(netdev))
5048                 e1000e_up(adapter);
5049         else
5050                 e1000e_reset(adapter);
5051
5052         clear_bit(__E1000_RESETTING, &adapter->state);
5053
5054         return 0;
5055 }
5056
5057 static int e1000_mii_ioctl(struct net_device *netdev, struct ifreq *ifr,
5058                            int cmd)
5059 {
5060         struct e1000_adapter *adapter = netdev_priv(netdev);
5061         struct mii_ioctl_data *data = if_mii(ifr);
5062
5063         if (adapter->hw.phy.media_type != e1000_media_type_copper)
5064                 return -EOPNOTSUPP;
5065
5066         switch (cmd) {
5067         case SIOCGMIIPHY:
5068                 data->phy_id = adapter->hw.phy.addr;
5069                 break;
5070         case SIOCGMIIREG:
5071                 e1000_phy_read_status(adapter);
5072
5073                 switch (data->reg_num & 0x1F) {
5074                 case MII_BMCR:
5075                         data->val_out = adapter->phy_regs.bmcr;
5076                         break;
5077                 case MII_BMSR:
5078                         data->val_out = adapter->phy_regs.bmsr;
5079                         break;
5080                 case MII_PHYSID1:
5081                         data->val_out = (adapter->hw.phy.id >> 16);
5082                         break;
5083                 case MII_PHYSID2:
5084                         data->val_out = (adapter->hw.phy.id & 0xFFFF);
5085                         break;
5086                 case MII_ADVERTISE:
5087                         data->val_out = adapter->phy_regs.advertise;
5088                         break;
5089                 case MII_LPA:
5090                         data->val_out = adapter->phy_regs.lpa;
5091                         break;
5092                 case MII_EXPANSION:
5093                         data->val_out = adapter->phy_regs.expansion;
5094                         break;
5095                 case MII_CTRL1000:
5096                         data->val_out = adapter->phy_regs.ctrl1000;
5097                         break;
5098                 case MII_STAT1000:
5099                         data->val_out = adapter->phy_regs.stat1000;
5100                         break;
5101                 case MII_ESTATUS:
5102                         data->val_out = adapter->phy_regs.estatus;
5103                         break;
5104                 default:
5105                         return -EIO;
5106                 }
5107                 break;
5108         case SIOCSMIIREG:
5109         default:
5110                 return -EOPNOTSUPP;
5111         }
5112         return 0;
5113 }
5114
5115 static int e1000_ioctl(struct net_device *netdev, struct ifreq *ifr, int cmd)
5116 {
5117         switch (cmd) {
5118         case SIOCGMIIPHY:
5119         case SIOCGMIIREG:
5120         case SIOCSMIIREG:
5121                 return e1000_mii_ioctl(netdev, ifr, cmd);
5122         default:
5123                 return -EOPNOTSUPP;
5124         }
5125 }
5126
5127 static int e1000_init_phy_wakeup(struct e1000_adapter *adapter, u32 wufc)
5128 {
5129         struct e1000_hw *hw = &adapter->hw;
5130         u32 i, mac_reg;
5131         u16 phy_reg, wuc_enable;
5132         int retval = 0;
5133
5134         /* copy MAC RARs to PHY RARs */
5135         e1000_copy_rx_addrs_to_phy_ich8lan(hw);
5136
5137         retval = hw->phy.ops.acquire(hw);
5138         if (retval) {
5139                 e_err("Could not acquire PHY\n");
5140                 return retval;
5141         }
5142
5143         /* Enable access to wakeup registers on and set page to BM_WUC_PAGE */
5144         retval = e1000_enable_phy_wakeup_reg_access_bm(hw, &wuc_enable);
5145         if (retval)
5146                 goto out;
5147
5148         /* copy MAC MTA to PHY MTA - only needed for pchlan */
5149         for (i = 0; i < adapter->hw.mac.mta_reg_count; i++) {
5150                 mac_reg = E1000_READ_REG_ARRAY(hw, E1000_MTA, i);
5151                 hw->phy.ops.write_reg_page(hw, BM_MTA(i),
5152                                            (u16)(mac_reg & 0xFFFF));
5153                 hw->phy.ops.write_reg_page(hw, BM_MTA(i) + 1,
5154                                            (u16)((mac_reg >> 16) & 0xFFFF));
5155         }
5156
5157         /* configure PHY Rx Control register */
5158         hw->phy.ops.read_reg_page(&adapter->hw, BM_RCTL, &phy_reg);
5159         mac_reg = er32(RCTL);
5160         if (mac_reg & E1000_RCTL_UPE)
5161                 phy_reg |= BM_RCTL_UPE;
5162         if (mac_reg & E1000_RCTL_MPE)
5163                 phy_reg |= BM_RCTL_MPE;
5164         phy_reg &= ~(BM_RCTL_MO_MASK);
5165         if (mac_reg & E1000_RCTL_MO_3)
5166                 phy_reg |= (((mac_reg & E1000_RCTL_MO_3) >> E1000_RCTL_MO_SHIFT)
5167                                 << BM_RCTL_MO_SHIFT);
5168         if (mac_reg & E1000_RCTL_BAM)
5169                 phy_reg |= BM_RCTL_BAM;
5170         if (mac_reg & E1000_RCTL_PMCF)
5171                 phy_reg |= BM_RCTL_PMCF;
5172         mac_reg = er32(CTRL);
5173         if (mac_reg & E1000_CTRL_RFCE)
5174                 phy_reg |= BM_RCTL_RFCE;
5175         hw->phy.ops.write_reg_page(&adapter->hw, BM_RCTL, phy_reg);
5176
5177         /* enable PHY wakeup in MAC register */
5178         ew32(WUFC, wufc);
5179         ew32(WUC, E1000_WUC_PHY_WAKE | E1000_WUC_PME_EN);
5180
5181         /* configure and enable PHY wakeup in PHY registers */
5182         hw->phy.ops.write_reg_page(&adapter->hw, BM_WUFC, wufc);
5183         hw->phy.ops.write_reg_page(&adapter->hw, BM_WUC, E1000_WUC_PME_EN);
5184
5185         /* activate PHY wakeup */
5186         wuc_enable |= BM_WUC_ENABLE_BIT | BM_WUC_HOST_WU_BIT;
5187         retval = e1000_disable_phy_wakeup_reg_access_bm(hw, &wuc_enable);
5188         if (retval)
5189                 e_err("Could not set PHY Host Wakeup bit\n");
5190 out:
5191         hw->phy.ops.release(hw);
5192
5193         return retval;
5194 }
5195
5196 static int __e1000_shutdown(struct pci_dev *pdev, bool *enable_wake,
5197                             bool runtime)
5198 {
5199         struct net_device *netdev = pci_get_drvdata(pdev);
5200         struct e1000_adapter *adapter = netdev_priv(netdev);
5201         struct e1000_hw *hw = &adapter->hw;
5202         u32 ctrl, ctrl_ext, rctl, status;
5203         /* Runtime suspend should only enable wakeup for link changes */
5204         u32 wufc = runtime ? E1000_WUFC_LNKC : adapter->wol;
5205         int retval = 0;
5206
5207         netif_device_detach(netdev);
5208
5209         if (netif_running(netdev)) {
5210                 WARN_ON(test_bit(__E1000_RESETTING, &adapter->state));
5211                 e1000e_down(adapter);
5212                 e1000_free_irq(adapter);
5213         }
5214         e1000e_reset_interrupt_capability(adapter);
5215
5216         retval = pci_save_state(pdev);
5217         if (retval)
5218                 return retval;
5219
5220         status = er32(STATUS);
5221         if (status & E1000_STATUS_LU)
5222                 wufc &= ~E1000_WUFC_LNKC;
5223
5224         if (wufc) {
5225                 e1000_setup_rctl(adapter);
5226                 e1000_set_multi(netdev);
5227
5228                 /* turn on all-multi mode if wake on multicast is enabled */
5229                 if (wufc & E1000_WUFC_MC) {
5230                         rctl = er32(RCTL);
5231                         rctl |= E1000_RCTL_MPE;
5232                         ew32(RCTL, rctl);
5233                 }
5234
5235                 ctrl = er32(CTRL);
5236                 /* advertise wake from D3Cold */
5237                 #define E1000_CTRL_ADVD3WUC 0x00100000
5238                 /* phy power management enable */
5239                 #define E1000_CTRL_EN_PHY_PWR_MGMT 0x00200000
5240                 ctrl |= E1000_CTRL_ADVD3WUC;
5241                 if (!(adapter->flags2 & FLAG2_HAS_PHY_WAKEUP))
5242                         ctrl |= E1000_CTRL_EN_PHY_PWR_MGMT;
5243                 ew32(CTRL, ctrl);
5244
5245                 if (adapter->hw.phy.media_type == e1000_media_type_fiber ||
5246                     adapter->hw.phy.media_type ==
5247                     e1000_media_type_internal_serdes) {
5248                         /* keep the laser running in D3 */
5249                         ctrl_ext = er32(CTRL_EXT);
5250                         ctrl_ext |= E1000_CTRL_EXT_SDP3_DATA;
5251                         ew32(CTRL_EXT, ctrl_ext);
5252                 }
5253
5254                 if (adapter->flags & FLAG_IS_ICH)
5255                         e1000_suspend_workarounds_ich8lan(&adapter->hw);
5256
5257                 /* Allow time for pending master requests to run */
5258                 e1000e_disable_pcie_master(&adapter->hw);
5259
5260                 if (adapter->flags2 & FLAG2_HAS_PHY_WAKEUP) {
5261                         /* enable wakeup by the PHY */
5262                         retval = e1000_init_phy_wakeup(adapter, wufc);
5263                         if (retval)
5264                                 return retval;
5265                 } else {
5266                         /* enable wakeup by the MAC */
5267                         ew32(WUFC, wufc);
5268                         ew32(WUC, E1000_WUC_PME_EN);
5269                 }
5270         } else {
5271                 ew32(WUC, 0);
5272                 ew32(WUFC, 0);
5273         }
5274
5275         *enable_wake = !!wufc;
5276
5277         /* make sure adapter isn't asleep if manageability is enabled */
5278         if ((adapter->flags & FLAG_MNG_PT_ENABLED) ||
5279             (hw->mac.ops.check_mng_mode(hw)))
5280                 *enable_wake = true;
5281
5282         if (adapter->hw.phy.type == e1000_phy_igp_3)
5283                 e1000e_igp3_phy_powerdown_workaround_ich8lan(&adapter->hw);
5284
5285         /*
5286          * Release control of h/w to f/w.  If f/w is AMT enabled, this
5287          * would have already happened in close and is redundant.
5288          */
5289         e1000e_release_hw_control(adapter);
5290
5291         pci_disable_device(pdev);
5292
5293         return 0;
5294 }
5295
5296 static void e1000_power_off(struct pci_dev *pdev, bool sleep, bool wake)
5297 {
5298         if (sleep && wake) {
5299                 pci_prepare_to_sleep(pdev);
5300                 return;
5301         }
5302
5303         pci_wake_from_d3(pdev, wake);
5304         pci_set_power_state(pdev, PCI_D3hot);
5305 }
5306
5307 static void e1000_complete_shutdown(struct pci_dev *pdev, bool sleep,
5308                                     bool wake)
5309 {
5310         struct net_device *netdev = pci_get_drvdata(pdev);
5311         struct e1000_adapter *adapter = netdev_priv(netdev);
5312
5313         /*
5314          * The pci-e switch on some quad port adapters will report a
5315          * correctable error when the MAC transitions from D0 to D3.  To
5316          * prevent this we need to mask off the correctable errors on the
5317          * downstream port of the pci-e switch.
5318          */
5319         if (adapter->flags & FLAG_IS_QUAD_PORT) {
5320                 struct pci_dev *us_dev = pdev->bus->self;
5321                 int pos = pci_pcie_cap(us_dev);
5322                 u16 devctl;
5323
5324                 pci_read_config_word(us_dev, pos + PCI_EXP_DEVCTL, &devctl);
5325                 pci_write_config_word(us_dev, pos + PCI_EXP_DEVCTL,
5326                                       (devctl & ~PCI_EXP_DEVCTL_CERE));
5327
5328                 e1000_power_off(pdev, sleep, wake);
5329
5330                 pci_write_config_word(us_dev, pos + PCI_EXP_DEVCTL, devctl);
5331         } else {
5332                 e1000_power_off(pdev, sleep, wake);
5333         }
5334 }
5335
5336 #ifdef CONFIG_PCIEASPM
5337 static void __e1000e_disable_aspm(struct pci_dev *pdev, u16 state)
5338 {
5339         pci_disable_link_state_locked(pdev, state);
5340 }
5341 #else
5342 static void __e1000e_disable_aspm(struct pci_dev *pdev, u16 state)
5343 {
5344         int pos;
5345         u16 reg16;
5346
5347         /*
5348          * Both device and parent should have the same ASPM setting.
5349          * Disable ASPM in downstream component first and then upstream.
5350          */
5351         pos = pci_pcie_cap(pdev);
5352         pci_read_config_word(pdev, pos + PCI_EXP_LNKCTL, &reg16);
5353         reg16 &= ~state;
5354         pci_write_config_word(pdev, pos + PCI_EXP_LNKCTL, reg16);
5355
5356         if (!pdev->bus->self)
5357                 return;
5358
5359         pos = pci_pcie_cap(pdev->bus->self);
5360         pci_read_config_word(pdev->bus->self, pos + PCI_EXP_LNKCTL, &reg16);
5361         reg16 &= ~state;
5362         pci_write_config_word(pdev->bus->self, pos + PCI_EXP_LNKCTL, reg16);
5363 }
5364 #endif
5365 static void e1000e_disable_aspm(struct pci_dev *pdev, u16 state)
5366 {
5367         dev_info(&pdev->dev, "Disabling ASPM %s %s\n",
5368                  (state & PCIE_LINK_STATE_L0S) ? "L0s" : "",
5369                  (state & PCIE_LINK_STATE_L1) ? "L1" : "");
5370
5371         __e1000e_disable_aspm(pdev, state);
5372 }
5373
5374 #ifdef CONFIG_PM
5375 static bool e1000e_pm_ready(struct e1000_adapter *adapter)
5376 {
5377         return !!adapter->tx_ring->buffer_info;
5378 }
5379
5380 static int __e1000_resume(struct pci_dev *pdev)
5381 {
5382         struct net_device *netdev = pci_get_drvdata(pdev);
5383         struct e1000_adapter *adapter = netdev_priv(netdev);
5384         struct e1000_hw *hw = &adapter->hw;
5385         u16 aspm_disable_flag = 0;
5386         u32 err;
5387
5388         if (adapter->flags2 & FLAG2_DISABLE_ASPM_L0S)
5389                 aspm_disable_flag = PCIE_LINK_STATE_L0S;
5390         if (adapter->flags2 & FLAG2_DISABLE_ASPM_L1)
5391                 aspm_disable_flag |= PCIE_LINK_STATE_L1;
5392         if (aspm_disable_flag)
5393                 e1000e_disable_aspm(pdev, aspm_disable_flag);
5394
5395         pci_set_power_state(pdev, PCI_D0);
5396         pci_restore_state(pdev);
5397         pci_save_state(pdev);
5398
5399         e1000e_set_interrupt_capability(adapter);
5400         if (netif_running(netdev)) {
5401                 err = e1000_request_irq(adapter);
5402                 if (err)
5403                         return err;
5404         }
5405
5406         if (hw->mac.type == e1000_pch2lan)
5407                 e1000_resume_workarounds_pchlan(&adapter->hw);
5408
5409         e1000e_power_up_phy(adapter);
5410
5411         /* report the system wakeup cause from S3/S4 */
5412         if (adapter->flags2 & FLAG2_HAS_PHY_WAKEUP) {
5413                 u16 phy_data;
5414
5415                 e1e_rphy(&adapter->hw, BM_WUS, &phy_data);
5416                 if (phy_data) {
5417                         e_info("PHY Wakeup cause - %s\n",
5418                                 phy_data & E1000_WUS_EX ? "Unicast Packet" :
5419                                 phy_data & E1000_WUS_MC ? "Multicast Packet" :
5420                                 phy_data & E1000_WUS_BC ? "Broadcast Packet" :
5421                                 phy_data & E1000_WUS_MAG ? "Magic Packet" :
5422                                 phy_data & E1000_WUS_LNKC ? "Link Status "
5423                                 " Change" : "other");
5424                 }
5425                 e1e_wphy(&adapter->hw, BM_WUS, ~0);
5426         } else {
5427                 u32 wus = er32(WUS);
5428                 if (wus) {
5429                         e_info("MAC Wakeup cause - %s\n",
5430                                 wus & E1000_WUS_EX ? "Unicast Packet" :
5431                                 wus & E1000_WUS_MC ? "Multicast Packet" :
5432                                 wus & E1000_WUS_BC ? "Broadcast Packet" :
5433                                 wus & E1000_WUS_MAG ? "Magic Packet" :
5434                                 wus & E1000_WUS_LNKC ? "Link Status Change" :
5435                                 "other");
5436                 }
5437                 ew32(WUS, ~0);
5438         }
5439
5440         e1000e_reset(adapter);
5441
5442         e1000_init_manageability_pt(adapter);
5443
5444         if (netif_running(netdev))
5445                 e1000e_up(adapter);
5446
5447         netif_device_attach(netdev);
5448
5449         /*
5450          * If the controller has AMT, do not set DRV_LOAD until the interface
5451          * is up.  For all other cases, let the f/w know that the h/w is now
5452          * under the control of the driver.
5453          */
5454         if (!(adapter->flags & FLAG_HAS_AMT))
5455                 e1000e_get_hw_control(adapter);
5456
5457         return 0;
5458 }
5459
5460 #ifdef CONFIG_PM_SLEEP
5461 static int e1000_suspend(struct device *dev)
5462 {
5463         struct pci_dev *pdev = to_pci_dev(dev);
5464         int retval;
5465         bool wake;
5466
5467         retval = __e1000_shutdown(pdev, &wake, false);
5468         if (!retval)
5469                 e1000_complete_shutdown(pdev, true, wake);
5470
5471         return retval;
5472 }
5473
5474 static int e1000_resume(struct device *dev)
5475 {
5476         struct pci_dev *pdev = to_pci_dev(dev);
5477         struct net_device *netdev = pci_get_drvdata(pdev);
5478         struct e1000_adapter *adapter = netdev_priv(netdev);
5479
5480         if (e1000e_pm_ready(adapter))
5481                 adapter->idle_check = true;
5482
5483         return __e1000_resume(pdev);
5484 }
5485 #endif /* CONFIG_PM_SLEEP */
5486
5487 #ifdef CONFIG_PM_RUNTIME
5488 static int e1000_runtime_suspend(struct device *dev)
5489 {
5490         struct pci_dev *pdev = to_pci_dev(dev);
5491         struct net_device *netdev = pci_get_drvdata(pdev);
5492         struct e1000_adapter *adapter = netdev_priv(netdev);
5493
5494         if (e1000e_pm_ready(adapter)) {
5495                 bool wake;
5496
5497                 __e1000_shutdown(pdev, &wake, true);
5498         }
5499
5500         return 0;
5501 }
5502
5503 static int e1000_idle(struct device *dev)
5504 {
5505         struct pci_dev *pdev = to_pci_dev(dev);
5506         struct net_device *netdev = pci_get_drvdata(pdev);
5507         struct e1000_adapter *adapter = netdev_priv(netdev);
5508
5509         if (!e1000e_pm_ready(adapter))
5510                 return 0;
5511
5512         if (adapter->idle_check) {
5513                 adapter->idle_check = false;
5514                 if (!e1000e_has_link(adapter))
5515                         pm_schedule_suspend(dev, MSEC_PER_SEC);
5516         }
5517
5518         return -EBUSY;
5519 }
5520
5521 static int e1000_runtime_resume(struct device *dev)
5522 {
5523         struct pci_dev *pdev = to_pci_dev(dev);
5524         struct net_device *netdev = pci_get_drvdata(pdev);
5525         struct e1000_adapter *adapter = netdev_priv(netdev);
5526
5527         if (!e1000e_pm_ready(adapter))
5528                 return 0;
5529
5530         adapter->idle_check = !dev->power.runtime_auto;
5531         return __e1000_resume(pdev);
5532 }
5533 #endif /* CONFIG_PM_RUNTIME */
5534 #endif /* CONFIG_PM */
5535
5536 static void e1000_shutdown(struct pci_dev *pdev)
5537 {
5538         bool wake = false;
5539
5540         __e1000_shutdown(pdev, &wake, false);
5541
5542         if (system_state == SYSTEM_POWER_OFF)
5543                 e1000_complete_shutdown(pdev, false, wake);
5544 }
5545
5546 #ifdef CONFIG_NET_POLL_CONTROLLER
5547
5548 static irqreturn_t e1000_intr_msix(int irq, void *data)
5549 {
5550         struct net_device *netdev = data;
5551         struct e1000_adapter *adapter = netdev_priv(netdev);
5552
5553         if (adapter->msix_entries) {
5554                 int vector, msix_irq;
5555
5556                 vector = 0;
5557                 msix_irq = adapter->msix_entries[vector].vector;
5558                 disable_irq(msix_irq);
5559                 e1000_intr_msix_rx(msix_irq, netdev);
5560                 enable_irq(msix_irq);
5561
5562                 vector++;
5563                 msix_irq = adapter->msix_entries[vector].vector;
5564                 disable_irq(msix_irq);
5565                 e1000_intr_msix_tx(msix_irq, netdev);
5566                 enable_irq(msix_irq);
5567
5568                 vector++;
5569                 msix_irq = adapter->msix_entries[vector].vector;
5570                 disable_irq(msix_irq);
5571                 e1000_msix_other(msix_irq, netdev);
5572                 enable_irq(msix_irq);
5573         }
5574
5575         return IRQ_HANDLED;
5576 }
5577
5578 /*
5579  * Polling 'interrupt' - used by things like netconsole to send skbs
5580  * without having to re-enable interrupts. It's not called while
5581  * the interrupt routine is executing.
5582  */
5583 static void e1000_netpoll(struct net_device *netdev)
5584 {
5585         struct e1000_adapter *adapter = netdev_priv(netdev);
5586
5587         switch (adapter->int_mode) {
5588         case E1000E_INT_MODE_MSIX:
5589                 e1000_intr_msix(adapter->pdev->irq, netdev);
5590                 break;
5591         case E1000E_INT_MODE_MSI:
5592                 disable_irq(adapter->pdev->irq);
5593                 e1000_intr_msi(adapter->pdev->irq, netdev);
5594                 enable_irq(adapter->pdev->irq);
5595                 break;
5596         default: /* E1000E_INT_MODE_LEGACY */
5597                 disable_irq(adapter->pdev->irq);
5598                 e1000_intr(adapter->pdev->irq, netdev);
5599                 enable_irq(adapter->pdev->irq);
5600                 break;
5601         }
5602 }
5603 #endif
5604
5605 /**
5606  * e1000_io_error_detected - called when PCI error is detected
5607  * @pdev: Pointer to PCI device
5608  * @state: The current pci connection state
5609  *
5610  * This function is called after a PCI bus error affecting
5611  * this device has been detected.
5612  */
5613 static pci_ers_result_t e1000_io_error_detected(struct pci_dev *pdev,
5614                                                 pci_channel_state_t state)
5615 {
5616         struct net_device *netdev = pci_get_drvdata(pdev);
5617         struct e1000_adapter *adapter = netdev_priv(netdev);
5618
5619         netif_device_detach(netdev);
5620
5621         if (state == pci_channel_io_perm_failure)
5622                 return PCI_ERS_RESULT_DISCONNECT;
5623
5624         if (netif_running(netdev))
5625                 e1000e_down(adapter);
5626         pci_disable_device(pdev);
5627
5628         /* Request a slot slot reset. */
5629         return PCI_ERS_RESULT_NEED_RESET;
5630 }
5631
5632 /**
5633  * e1000_io_slot_reset - called after the pci bus has been reset.
5634  * @pdev: Pointer to PCI device
5635  *
5636  * Restart the card from scratch, as if from a cold-boot. Implementation
5637  * resembles the first-half of the e1000_resume routine.
5638  */
5639 static pci_ers_result_t e1000_io_slot_reset(struct pci_dev *pdev)
5640 {
5641         struct net_device *netdev = pci_get_drvdata(pdev);
5642         struct e1000_adapter *adapter = netdev_priv(netdev);
5643         struct e1000_hw *hw = &adapter->hw;
5644         u16 aspm_disable_flag = 0;
5645         int err;
5646         pci_ers_result_t result;
5647
5648         if (adapter->flags2 & FLAG2_DISABLE_ASPM_L0S)
5649                 aspm_disable_flag = PCIE_LINK_STATE_L0S;
5650         if (adapter->flags2 & FLAG2_DISABLE_ASPM_L1)
5651                 aspm_disable_flag |= PCIE_LINK_STATE_L1;
5652         if (aspm_disable_flag)
5653                 e1000e_disable_aspm(pdev, aspm_disable_flag);
5654
5655         err = pci_enable_device_mem(pdev);
5656         if (err) {
5657                 dev_err(&pdev->dev,
5658                         "Cannot re-enable PCI device after reset.\n");
5659                 result = PCI_ERS_RESULT_DISCONNECT;
5660         } else {
5661                 pci_set_master(pdev);
5662                 pdev->state_saved = true;
5663                 pci_restore_state(pdev);
5664
5665                 pci_enable_wake(pdev, PCI_D3hot, 0);
5666                 pci_enable_wake(pdev, PCI_D3cold, 0);
5667
5668                 e1000e_reset(adapter);
5669                 ew32(WUS, ~0);
5670                 result = PCI_ERS_RESULT_RECOVERED;
5671         }
5672
5673         pci_cleanup_aer_uncorrect_error_status(pdev);
5674
5675         return result;
5676 }
5677
5678 /**
5679  * e1000_io_resume - called when traffic can start flowing again.
5680  * @pdev: Pointer to PCI device
5681  *
5682  * This callback is called when the error recovery driver tells us that
5683  * its OK to resume normal operation. Implementation resembles the
5684  * second-half of the e1000_resume routine.
5685  */
5686 static void e1000_io_resume(struct pci_dev *pdev)
5687 {
5688         struct net_device *netdev = pci_get_drvdata(pdev);
5689         struct e1000_adapter *adapter = netdev_priv(netdev);
5690
5691         e1000_init_manageability_pt(adapter);
5692
5693         if (netif_running(netdev)) {
5694                 if (e1000e_up(adapter)) {
5695                         dev_err(&pdev->dev,
5696                                 "can't bring device back up after reset\n");
5697                         return;
5698                 }
5699         }
5700
5701         netif_device_attach(netdev);
5702
5703         /*
5704          * If the controller has AMT, do not set DRV_LOAD until the interface
5705          * is up.  For all other cases, let the f/w know that the h/w is now
5706          * under the control of the driver.
5707          */
5708         if (!(adapter->flags & FLAG_HAS_AMT))
5709                 e1000e_get_hw_control(adapter);
5710
5711 }
5712
5713 static void e1000_print_device_info(struct e1000_adapter *adapter)
5714 {
5715         struct e1000_hw *hw = &adapter->hw;
5716         struct net_device *netdev = adapter->netdev;
5717         u32 ret_val;
5718         u8 pba_str[E1000_PBANUM_LENGTH];
5719
5720         /* print bus type/speed/width info */
5721         e_info("(PCI Express:2.5GT/s:%s) %pM\n",
5722                /* bus width */
5723                ((hw->bus.width == e1000_bus_width_pcie_x4) ? "Width x4" :
5724                 "Width x1"),
5725                /* MAC address */
5726                netdev->dev_addr);
5727         e_info("Intel(R) PRO/%s Network Connection\n",
5728                (hw->phy.type == e1000_phy_ife) ? "10/100" : "1000");
5729         ret_val = e1000_read_pba_string_generic(hw, pba_str,
5730                                                 E1000_PBANUM_LENGTH);
5731         if (ret_val)
5732                 strncpy((char *)pba_str, "Unknown", sizeof(pba_str) - 1);
5733         e_info("MAC: %d, PHY: %d, PBA No: %s\n",
5734                hw->mac.type, hw->phy.type, pba_str);
5735 }
5736
5737 static void e1000_eeprom_checks(struct e1000_adapter *adapter)
5738 {
5739         struct e1000_hw *hw = &adapter->hw;
5740         int ret_val;
5741         u16 buf = 0;
5742
5743         if (hw->mac.type != e1000_82573)
5744                 return;
5745
5746         ret_val = e1000_read_nvm(hw, NVM_INIT_CONTROL2_REG, 1, &buf);
5747         if (!ret_val && (!(le16_to_cpu(buf) & (1 << 0)))) {
5748                 /* Deep Smart Power Down (DSPD) */
5749                 dev_warn(&adapter->pdev->dev,
5750                          "Warning: detected DSPD enabled in EEPROM\n");
5751         }
5752 }
5753
5754 static const struct net_device_ops e1000e_netdev_ops = {
5755         .ndo_open               = e1000_open,
5756         .ndo_stop               = e1000_close,
5757         .ndo_start_xmit         = e1000_xmit_frame,
5758         .ndo_get_stats64        = e1000e_get_stats64,
5759         .ndo_set_multicast_list = e1000_set_multi,
5760         .ndo_set_mac_address    = e1000_set_mac,
5761         .ndo_change_mtu         = e1000_change_mtu,
5762         .ndo_do_ioctl           = e1000_ioctl,
5763         .ndo_tx_timeout         = e1000_tx_timeout,
5764         .ndo_validate_addr      = eth_validate_addr,
5765
5766         .ndo_vlan_rx_add_vid    = e1000_vlan_rx_add_vid,
5767         .ndo_vlan_rx_kill_vid   = e1000_vlan_rx_kill_vid,
5768 #ifdef CONFIG_NET_POLL_CONTROLLER
5769         .ndo_poll_controller    = e1000_netpoll,
5770 #endif
5771 };
5772
5773 /**
5774  * e1000_probe - Device Initialization Routine
5775  * @pdev: PCI device information struct
5776  * @ent: entry in e1000_pci_tbl
5777  *
5778  * Returns 0 on success, negative on failure
5779  *
5780  * e1000_probe initializes an adapter identified by a pci_dev structure.
5781  * The OS initialization, configuring of the adapter private structure,
5782  * and a hardware reset occur.
5783  **/
5784 static int __devinit e1000_probe(struct pci_dev *pdev,
5785                                  const struct pci_device_id *ent)
5786 {
5787         struct net_device *netdev;
5788         struct e1000_adapter *adapter;
5789         struct e1000_hw *hw;
5790         const struct e1000_info *ei = e1000_info_tbl[ent->driver_data];
5791         resource_size_t mmio_start, mmio_len;
5792         resource_size_t flash_start, flash_len;
5793
5794         static int cards_found;
5795         u16 aspm_disable_flag = 0;
5796         int i, err, pci_using_dac;
5797         u16 eeprom_data = 0;
5798         u16 eeprom_apme_mask = E1000_EEPROM_APME;
5799
5800         if (ei->flags2 & FLAG2_DISABLE_ASPM_L0S)
5801                 aspm_disable_flag = PCIE_LINK_STATE_L0S;
5802         if (ei->flags2 & FLAG2_DISABLE_ASPM_L1)
5803                 aspm_disable_flag |= PCIE_LINK_STATE_L1;
5804         if (aspm_disable_flag)
5805                 e1000e_disable_aspm(pdev, aspm_disable_flag);
5806
5807         err = pci_enable_device_mem(pdev);
5808         if (err)
5809                 return err;
5810
5811         pci_using_dac = 0;
5812         err = dma_set_mask(&pdev->dev, DMA_BIT_MASK(64));
5813         if (!err) {
5814                 err = dma_set_coherent_mask(&pdev->dev, DMA_BIT_MASK(64));
5815                 if (!err)
5816                         pci_using_dac = 1;
5817         } else {
5818                 err = dma_set_mask(&pdev->dev, DMA_BIT_MASK(32));
5819                 if (err) {
5820                         err = dma_set_coherent_mask(&pdev->dev,
5821                                                     DMA_BIT_MASK(32));
5822                         if (err) {
5823                                 dev_err(&pdev->dev, "No usable DMA "
5824                                         "configuration, aborting\n");
5825                                 goto err_dma;
5826                         }
5827                 }
5828         }
5829
5830         err = pci_request_selected_regions_exclusive(pdev,
5831                                           pci_select_bars(pdev, IORESOURCE_MEM),
5832                                           e1000e_driver_name);
5833         if (err)
5834                 goto err_pci_reg;
5835
5836         /* AER (Advanced Error Reporting) hooks */
5837         pci_enable_pcie_error_reporting(pdev);
5838
5839         pci_set_master(pdev);
5840         /* PCI config space info */
5841         err = pci_save_state(pdev);
5842         if (err)
5843                 goto err_alloc_etherdev;
5844
5845         err = -ENOMEM;
5846         netdev = alloc_etherdev(sizeof(struct e1000_adapter));
5847         if (!netdev)
5848                 goto err_alloc_etherdev;
5849
5850         SET_NETDEV_DEV(netdev, &pdev->dev);
5851
5852         netdev->irq = pdev->irq;
5853
5854         pci_set_drvdata(pdev, netdev);
5855         adapter = netdev_priv(netdev);
5856         hw = &adapter->hw;
5857         adapter->netdev = netdev;
5858         adapter->pdev = pdev;
5859         adapter->ei = ei;
5860         adapter->pba = ei->pba;
5861         adapter->flags = ei->flags;
5862         adapter->flags2 = ei->flags2;
5863         adapter->hw.adapter = adapter;
5864         adapter->hw.mac.type = ei->mac;
5865         adapter->max_hw_frame_size = ei->max_hw_frame_size;
5866         adapter->msg_enable = (1 << NETIF_MSG_DRV | NETIF_MSG_PROBE) - 1;
5867
5868         mmio_start = pci_resource_start(pdev, 0);
5869         mmio_len = pci_resource_len(pdev, 0);
5870
5871         err = -EIO;
5872         adapter->hw.hw_addr = ioremap(mmio_start, mmio_len);
5873         if (!adapter->hw.hw_addr)
5874                 goto err_ioremap;
5875
5876         if ((adapter->flags & FLAG_HAS_FLASH) &&
5877             (pci_resource_flags(pdev, 1) & IORESOURCE_MEM)) {
5878                 flash_start = pci_resource_start(pdev, 1);
5879                 flash_len = pci_resource_len(pdev, 1);
5880                 adapter->hw.flash_address = ioremap(flash_start, flash_len);
5881                 if (!adapter->hw.flash_address)
5882                         goto err_flashmap;
5883         }
5884
5885         /* construct the net_device struct */
5886         netdev->netdev_ops              = &e1000e_netdev_ops;
5887         e1000e_set_ethtool_ops(netdev);
5888         netdev->watchdog_timeo          = 5 * HZ;
5889         netif_napi_add(netdev, &adapter->napi, e1000_clean, 64);
5890         strncpy(netdev->name, pci_name(pdev), sizeof(netdev->name) - 1);
5891
5892         netdev->mem_start = mmio_start;
5893         netdev->mem_end = mmio_start + mmio_len;
5894
5895         adapter->bd_number = cards_found++;
5896
5897         e1000e_check_options(adapter);
5898
5899         /* setup adapter struct */
5900         err = e1000_sw_init(adapter);
5901         if (err)
5902                 goto err_sw_init;
5903
5904         memcpy(&hw->mac.ops, ei->mac_ops, sizeof(hw->mac.ops));
5905         memcpy(&hw->nvm.ops, ei->nvm_ops, sizeof(hw->nvm.ops));
5906         memcpy(&hw->phy.ops, ei->phy_ops, sizeof(hw->phy.ops));
5907
5908         err = ei->get_variants(adapter);
5909         if (err)
5910                 goto err_hw_init;
5911
5912         if ((adapter->flags & FLAG_IS_ICH) &&
5913             (adapter->flags & FLAG_READ_ONLY_NVM))
5914                 e1000e_write_protect_nvm_ich8lan(&adapter->hw);
5915
5916         hw->mac.ops.get_bus_info(&adapter->hw);
5917
5918         adapter->hw.phy.autoneg_wait_to_complete = 0;
5919
5920         /* Copper options */
5921         if (adapter->hw.phy.media_type == e1000_media_type_copper) {
5922                 adapter->hw.phy.mdix = AUTO_ALL_MODES;
5923                 adapter->hw.phy.disable_polarity_correction = 0;
5924                 adapter->hw.phy.ms_type = e1000_ms_hw_default;
5925         }
5926
5927         if (e1000_check_reset_block(&adapter->hw))
5928                 e_info("PHY reset is blocked due to SOL/IDER session.\n");
5929
5930         netdev->features = NETIF_F_SG |
5931                            NETIF_F_HW_CSUM |
5932                            NETIF_F_HW_VLAN_TX |
5933                            NETIF_F_HW_VLAN_RX;
5934
5935         if (adapter->flags & FLAG_HAS_HW_VLAN_FILTER)
5936                 netdev->features |= NETIF_F_HW_VLAN_FILTER;
5937
5938         netdev->features |= NETIF_F_TSO;
5939         netdev->features |= NETIF_F_TSO6;
5940
5941         netdev->vlan_features |= NETIF_F_TSO;
5942         netdev->vlan_features |= NETIF_F_TSO6;
5943         netdev->vlan_features |= NETIF_F_HW_CSUM;
5944         netdev->vlan_features |= NETIF_F_SG;
5945
5946         if (pci_using_dac) {
5947                 netdev->features |= NETIF_F_HIGHDMA;
5948                 netdev->vlan_features |= NETIF_F_HIGHDMA;
5949         }
5950
5951         if (e1000e_enable_mng_pass_thru(&adapter->hw))
5952                 adapter->flags |= FLAG_MNG_PT_ENABLED;
5953
5954         /*
5955          * before reading the NVM, reset the controller to
5956          * put the device in a known good starting state
5957          */
5958         adapter->hw.mac.ops.reset_hw(&adapter->hw);
5959
5960         /*
5961          * systems with ASPM and others may see the checksum fail on the first
5962          * attempt. Let's give it a few tries
5963          */
5964         for (i = 0;; i++) {
5965                 if (e1000_validate_nvm_checksum(&adapter->hw) >= 0)
5966                         break;
5967                 if (i == 2) {
5968                         e_err("The NVM Checksum Is Not Valid\n");
5969                         err = -EIO;
5970                         goto err_eeprom;
5971                 }
5972         }
5973
5974         e1000_eeprom_checks(adapter);
5975
5976         /* copy the MAC address */
5977         if (e1000e_read_mac_addr(&adapter->hw))
5978                 e_err("NVM Read Error while reading MAC address\n");
5979
5980         memcpy(netdev->dev_addr, adapter->hw.mac.addr, netdev->addr_len);
5981         memcpy(netdev->perm_addr, adapter->hw.mac.addr, netdev->addr_len);
5982
5983         if (!is_valid_ether_addr(netdev->perm_addr)) {
5984                 e_err("Invalid MAC Address: %pM\n", netdev->perm_addr);
5985                 err = -EIO;
5986                 goto err_eeprom;
5987         }
5988
5989         init_timer(&adapter->watchdog_timer);
5990         adapter->watchdog_timer.function = e1000_watchdog;
5991         adapter->watchdog_timer.data = (unsigned long) adapter;
5992
5993         init_timer(&adapter->phy_info_timer);
5994         adapter->phy_info_timer.function = e1000_update_phy_info;
5995         adapter->phy_info_timer.data = (unsigned long) adapter;
5996
5997         INIT_WORK(&adapter->reset_task, e1000_reset_task);
5998         INIT_WORK(&adapter->watchdog_task, e1000_watchdog_task);
5999         INIT_WORK(&adapter->downshift_task, e1000e_downshift_workaround);
6000         INIT_WORK(&adapter->update_phy_task, e1000e_update_phy_task);
6001         INIT_WORK(&adapter->print_hang_task, e1000_print_hw_hang);
6002
6003         /* Initialize link parameters. User can change them with ethtool */
6004         adapter->hw.mac.autoneg = 1;
6005         adapter->fc_autoneg = 1;
6006         adapter->hw.fc.requested_mode = e1000_fc_default;
6007         adapter->hw.fc.current_mode = e1000_fc_default;
6008         adapter->hw.phy.autoneg_advertised = 0x2f;
6009
6010         /* ring size defaults */
6011         adapter->rx_ring->count = 256;
6012         adapter->tx_ring->count = 256;
6013
6014         /*
6015          * Initial Wake on LAN setting - If APM wake is enabled in
6016          * the EEPROM, enable the ACPI Magic Packet filter
6017          */
6018         if (adapter->flags & FLAG_APME_IN_WUC) {
6019                 /* APME bit in EEPROM is mapped to WUC.APME */
6020                 eeprom_data = er32(WUC);
6021                 eeprom_apme_mask = E1000_WUC_APME;
6022                 if ((hw->mac.type > e1000_ich10lan) &&
6023                     (eeprom_data & E1000_WUC_PHY_WAKE))
6024                         adapter->flags2 |= FLAG2_HAS_PHY_WAKEUP;
6025         } else if (adapter->flags & FLAG_APME_IN_CTRL3) {
6026                 if (adapter->flags & FLAG_APME_CHECK_PORT_B &&
6027                     (adapter->hw.bus.func == 1))
6028                         e1000_read_nvm(&adapter->hw,
6029                                 NVM_INIT_CONTROL3_PORT_B, 1, &eeprom_data);
6030                 else
6031                         e1000_read_nvm(&adapter->hw,
6032                                 NVM_INIT_CONTROL3_PORT_A, 1, &eeprom_data);
6033         }
6034
6035         /* fetch WoL from EEPROM */
6036         if (eeprom_data & eeprom_apme_mask)
6037                 adapter->eeprom_wol |= E1000_WUFC_MAG;
6038
6039         /*
6040          * now that we have the eeprom settings, apply the special cases
6041          * where the eeprom may be wrong or the board simply won't support
6042          * wake on lan on a particular port
6043          */
6044         if (!(adapter->flags & FLAG_HAS_WOL))
6045                 adapter->eeprom_wol = 0;
6046
6047         /* initialize the wol settings based on the eeprom settings */
6048         adapter->wol = adapter->eeprom_wol;
6049         device_set_wakeup_enable(&adapter->pdev->dev, adapter->wol);
6050
6051         /* save off EEPROM version number */
6052         e1000_read_nvm(&adapter->hw, 5, 1, &adapter->eeprom_vers);
6053
6054         /* reset the hardware with the new settings */
6055         e1000e_reset(adapter);
6056
6057         /*
6058          * If the controller has AMT, do not set DRV_LOAD until the interface
6059          * is up.  For all other cases, let the f/w know that the h/w is now
6060          * under the control of the driver.
6061          */
6062         if (!(adapter->flags & FLAG_HAS_AMT))
6063                 e1000e_get_hw_control(adapter);
6064
6065         strncpy(netdev->name, "eth%d", sizeof(netdev->name) - 1);
6066         err = register_netdev(netdev);
6067         if (err)
6068                 goto err_register;
6069
6070         /* carrier off reporting is important to ethtool even BEFORE open */
6071         netif_carrier_off(netdev);
6072
6073         e1000_print_device_info(adapter);
6074
6075         if (pci_dev_run_wake(pdev))
6076                 pm_runtime_put_noidle(&pdev->dev);
6077
6078         return 0;
6079
6080 err_register:
6081         if (!(adapter->flags & FLAG_HAS_AMT))
6082                 e1000e_release_hw_control(adapter);
6083 err_eeprom:
6084         if (!e1000_check_reset_block(&adapter->hw))
6085                 e1000_phy_hw_reset(&adapter->hw);
6086 err_hw_init:
6087         kfree(adapter->tx_ring);
6088         kfree(adapter->rx_ring);
6089 err_sw_init:
6090         if (adapter->hw.flash_address)
6091                 iounmap(adapter->hw.flash_address);
6092         e1000e_reset_interrupt_capability(adapter);
6093 err_flashmap:
6094         iounmap(adapter->hw.hw_addr);
6095 err_ioremap:
6096         free_netdev(netdev);
6097 err_alloc_etherdev:
6098         pci_release_selected_regions(pdev,
6099                                      pci_select_bars(pdev, IORESOURCE_MEM));
6100 err_pci_reg:
6101 err_dma:
6102         pci_disable_device(pdev);
6103         return err;
6104 }
6105
6106 /**
6107  * e1000_remove - Device Removal Routine
6108  * @pdev: PCI device information struct
6109  *
6110  * e1000_remove is called by the PCI subsystem to alert the driver
6111  * that it should release a PCI device.  The could be caused by a
6112  * Hot-Plug event, or because the driver is going to be removed from
6113  * memory.
6114  **/
6115 static void __devexit e1000_remove(struct pci_dev *pdev)
6116 {
6117         struct net_device *netdev = pci_get_drvdata(pdev);
6118         struct e1000_adapter *adapter = netdev_priv(netdev);
6119         bool down = test_bit(__E1000_DOWN, &adapter->state);
6120
6121         /*
6122          * The timers may be rescheduled, so explicitly disable them
6123          * from being rescheduled.
6124          */
6125         if (!down)
6126                 set_bit(__E1000_DOWN, &adapter->state);
6127         del_timer_sync(&adapter->watchdog_timer);
6128         del_timer_sync(&adapter->phy_info_timer);
6129
6130         cancel_work_sync(&adapter->reset_task);
6131         cancel_work_sync(&adapter->watchdog_task);
6132         cancel_work_sync(&adapter->downshift_task);
6133         cancel_work_sync(&adapter->update_phy_task);
6134         cancel_work_sync(&adapter->print_hang_task);
6135
6136         if (!(netdev->flags & IFF_UP))
6137                 e1000_power_down_phy(adapter);
6138
6139         /* Don't lie to e1000_close() down the road. */
6140         if (!down)
6141                 clear_bit(__E1000_DOWN, &adapter->state);
6142         unregister_netdev(netdev);
6143
6144         if (pci_dev_run_wake(pdev))
6145                 pm_runtime_get_noresume(&pdev->dev);
6146
6147         /*
6148          * Release control of h/w to f/w.  If f/w is AMT enabled, this
6149          * would have already happened in close and is redundant.
6150          */
6151         e1000e_release_hw_control(adapter);
6152
6153         e1000e_reset_interrupt_capability(adapter);
6154         kfree(adapter->tx_ring);
6155         kfree(adapter->rx_ring);
6156
6157         iounmap(adapter->hw.hw_addr);
6158         if (adapter->hw.flash_address)
6159                 iounmap(adapter->hw.flash_address);
6160         pci_release_selected_regions(pdev,
6161                                      pci_select_bars(pdev, IORESOURCE_MEM));
6162
6163         free_netdev(netdev);
6164
6165         /* AER disable */
6166         pci_disable_pcie_error_reporting(pdev);
6167
6168         pci_disable_device(pdev);
6169 }
6170
6171 /* PCI Error Recovery (ERS) */
6172 static struct pci_error_handlers e1000_err_handler = {
6173         .error_detected = e1000_io_error_detected,
6174         .slot_reset = e1000_io_slot_reset,
6175         .resume = e1000_io_resume,
6176 };
6177
6178 static DEFINE_PCI_DEVICE_TABLE(e1000_pci_tbl) = {
6179         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82571EB_COPPER), board_82571 },
6180         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82571EB_FIBER), board_82571 },
6181         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82571EB_QUAD_COPPER), board_82571 },
6182         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82571EB_QUAD_COPPER_LP), board_82571 },
6183         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82571EB_QUAD_FIBER), board_82571 },
6184         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82571EB_SERDES), board_82571 },
6185         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82571EB_SERDES_DUAL), board_82571 },
6186         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82571EB_SERDES_QUAD), board_82571 },
6187         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82571PT_QUAD_COPPER), board_82571 },
6188
6189         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82572EI), board_82572 },
6190         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82572EI_COPPER), board_82572 },
6191         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82572EI_FIBER), board_82572 },
6192         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82572EI_SERDES), board_82572 },
6193
6194         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82573E), board_82573 },
6195         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82573E_IAMT), board_82573 },
6196         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82573L), board_82573 },
6197
6198         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82574L), board_82574 },
6199         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82574LA), board_82574 },
6200         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82583V), board_82583 },
6201
6202         { PCI_VDEVICE(INTEL, E1000_DEV_ID_80003ES2LAN_COPPER_DPT),
6203           board_80003es2lan },
6204         { PCI_VDEVICE(INTEL, E1000_DEV_ID_80003ES2LAN_COPPER_SPT),
6205           board_80003es2lan },
6206         { PCI_VDEVICE(INTEL, E1000_DEV_ID_80003ES2LAN_SERDES_DPT),
6207           board_80003es2lan },
6208         { PCI_VDEVICE(INTEL, E1000_DEV_ID_80003ES2LAN_SERDES_SPT),
6209           board_80003es2lan },
6210
6211         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH8_IFE), board_ich8lan },
6212         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH8_IFE_G), board_ich8lan },
6213         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH8_IFE_GT), board_ich8lan },
6214         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH8_IGP_AMT), board_ich8lan },
6215         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH8_IGP_C), board_ich8lan },
6216         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH8_IGP_M), board_ich8lan },
6217         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH8_IGP_M_AMT), board_ich8lan },
6218         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH8_82567V_3), board_ich8lan },
6219
6220         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH9_IFE), board_ich9lan },
6221         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH9_IFE_G), board_ich9lan },
6222         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH9_IFE_GT), board_ich9lan },
6223         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH9_IGP_AMT), board_ich9lan },
6224         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH9_IGP_C), board_ich9lan },
6225         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH9_BM), board_ich9lan },
6226         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH9_IGP_M), board_ich9lan },
6227         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH9_IGP_M_AMT), board_ich9lan },
6228         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH9_IGP_M_V), board_ich9lan },
6229
6230         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH10_R_BM_LM), board_ich9lan },
6231         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH10_R_BM_LF), board_ich9lan },
6232         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH10_R_BM_V), board_ich9lan },
6233
6234         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH10_D_BM_LM), board_ich10lan },
6235         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH10_D_BM_LF), board_ich10lan },
6236         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH10_D_BM_V), board_ich10lan },
6237
6238         { PCI_VDEVICE(INTEL, E1000_DEV_ID_PCH_M_HV_LM), board_pchlan },
6239         { PCI_VDEVICE(INTEL, E1000_DEV_ID_PCH_M_HV_LC), board_pchlan },
6240         { PCI_VDEVICE(INTEL, E1000_DEV_ID_PCH_D_HV_DM), board_pchlan },
6241         { PCI_VDEVICE(INTEL, E1000_DEV_ID_PCH_D_HV_DC), board_pchlan },
6242
6243         { PCI_VDEVICE(INTEL, E1000_DEV_ID_PCH2_LV_LM), board_pch2lan },
6244         { PCI_VDEVICE(INTEL, E1000_DEV_ID_PCH2_LV_V), board_pch2lan },
6245
6246         { }     /* terminate list */
6247 };
6248 MODULE_DEVICE_TABLE(pci, e1000_pci_tbl);
6249
6250 #ifdef CONFIG_PM
6251 static const struct dev_pm_ops e1000_pm_ops = {
6252         SET_SYSTEM_SLEEP_PM_OPS(e1000_suspend, e1000_resume)
6253         SET_RUNTIME_PM_OPS(e1000_runtime_suspend,
6254                                 e1000_runtime_resume, e1000_idle)
6255 };
6256 #endif
6257
6258 /* PCI Device API Driver */
6259 static struct pci_driver e1000_driver = {
6260         .name     = e1000e_driver_name,
6261         .id_table = e1000_pci_tbl,
6262         .probe    = e1000_probe,
6263         .remove   = __devexit_p(e1000_remove),
6264 #ifdef CONFIG_PM
6265         .driver.pm = &e1000_pm_ops,
6266 #endif
6267         .shutdown = e1000_shutdown,
6268         .err_handler = &e1000_err_handler
6269 };
6270
6271 /**
6272  * e1000_init_module - Driver Registration Routine
6273  *
6274  * e1000_init_module is the first routine called when the driver is
6275  * loaded. All it does is register with the PCI subsystem.
6276  **/
6277 static int __init e1000_init_module(void)
6278 {
6279         int ret;
6280         pr_info("Intel(R) PRO/1000 Network Driver - %s\n",
6281                 e1000e_driver_version);
6282         pr_info("Copyright(c) 1999 - 2011 Intel Corporation.\n");
6283         ret = pci_register_driver(&e1000_driver);
6284
6285         return ret;
6286 }
6287 module_init(e1000_init_module);
6288
6289 /**
6290  * e1000_exit_module - Driver Exit Cleanup Routine
6291  *
6292  * e1000_exit_module is called just before the driver is removed
6293  * from memory.
6294  **/
6295 static void __exit e1000_exit_module(void)
6296 {
6297         pci_unregister_driver(&e1000_driver);
6298 }
6299 module_exit(e1000_exit_module);
6300
6301
6302 MODULE_AUTHOR("Intel Corporation, <linux.nics@intel.com>");
6303 MODULE_DESCRIPTION("Intel(R) PRO/1000 Network Driver");
6304 MODULE_LICENSE("GPL");
6305 MODULE_VERSION(DRV_VERSION);
6306
6307 /* e1000_main.c */