drivers/net/e1000e/netdev.c

   1 /*******************************************************************************
   2
   3   Intel PRO/1000 Linux driver
   4   Copyright(c) 1999 - 2011 Intel Corporation.
   5
   6   This program is free software; you can redistribute it and/or modify it
   7   under the terms and conditions of the GNU General Public License,
   8   version 2, as published by the Free Software Foundation.
   9
  10   This program is distributed in the hope it will be useful, but WITHOUT
  11   ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
  12   FITNESS FOR A PARTICULAR PURPOSE.  See the GNU General Public License for
  13   more details.
  14
  15   You should have received a copy of the GNU General Public License along with
  16   this program; if not, write to the Free Software Foundation, Inc.,
  17   51 Franklin St - Fifth Floor, Boston, MA 02110-1301 USA.
  18
  19   The full GNU General Public License is included in this distribution in
  20   the file called "COPYING".
  21
  22   Contact Information:
  23   Linux NICS <linux.nics@intel.com>
  24   e1000-devel Mailing List <e1000-devel@lists.sourceforge.net>
  25   Intel Corporation, 5200 N.E. Elam Young Parkway, Hillsboro, OR 97124-6497
  26
  27 *******************************************************************************/
  28
  29 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
  30
  31 #include <linux/module.h>
  32 #include <linux/types.h>
  33 #include <linux/init.h>
  34 #include <linux/pci.h>
  35 #include <linux/vmalloc.h>
  36 #include <linux/pagemap.h>
  37 #include <linux/delay.h>
  38 #include <linux/netdevice.h>
  39 #include <linux/interrupt.h>
  40 #include <linux/tcp.h>
  41 #include <linux/ipv6.h>
  42 #include <linux/slab.h>
  43 #include <net/checksum.h>
  44 #include <net/ip6_checksum.h>
  45 #include <linux/mii.h>
  46 #include <linux/ethtool.h>
  47 #include <linux/if_vlan.h>
  48 #include <linux/cpu.h>
  49 #include <linux/smp.h>
  50 #include <linux/pm_qos_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 "1.4.4" 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, gfp_t gfp)
 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, gfp);
 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, gfp_t gfp)
 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);
 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                                                   gfp);
 647
 648                 if (!skb) {
 649                         adapter->alloc_rx_buff_failed++;
 650                         break;
 651                 }
 652
 653                 buffer_info->skb = skb;
 654                 buffer_info->dma = dma_map_single(&pdev->dev, skb->data,
 655                                                   adapter->rx_ps_bsize0,
 656                                                   DMA_FROM_DEVICE);
 657                 if (dma_mapping_error(&pdev->dev, buffer_info->dma)) {
 658                         dev_err(&pdev->dev, "Rx DMA map failed\n");
 659                         adapter->rx_dma_failed++;
 660                         /* cleanup skb */
 661                         dev_kfree_skb_any(skb);
 662                         buffer_info->skb = NULL;
 663                         break;
 664                 }
 665
 666                 rx_desc->read.buffer_addr[0] = cpu_to_le64(buffer_info->dma);
 667
 668                 if (unlikely(!(i & (E1000_RX_BUFFER_WRITE - 1)))) {
 669                         /*
 670                          * Force memory writes to complete before letting h/w
 671                          * know there are new descriptors to fetch.  (Only
 672                          * applicable for weak-ordered memory model archs,
 673                          * such as IA-64).
 674                          */
 675                         wmb();
 676                         writel(i << 1, adapter->hw.hw_addr + rx_ring->tail);
 677                 }
 678
 679                 i++;
 680                 if (i == rx_ring->count)
 681                         i = 0;
 682                 buffer_info = &rx_ring->buffer_info[i];
 683         }
 684
 685 no_buffers:
 686         rx_ring->next_to_use = i;
 687 }
 688
 689 /**
 690  * e1000_alloc_jumbo_rx_buffers - Replace used jumbo receive buffers
 691  * @adapter: address of board private structure
 692  * @cleaned_count: number of buffers to allocate this pass
 693  **/
 694
 695 static void e1000_alloc_jumbo_rx_buffers(struct e1000_adapter *adapter,
 696                                          int cleaned_count, gfp_t gfp)
 697 {
 698         struct net_device *netdev = adapter->netdev;
 699         struct pci_dev *pdev = adapter->pdev;
 700         struct e1000_rx_desc *rx_desc;
 701         struct e1000_ring *rx_ring = adapter->rx_ring;
 702         struct e1000_buffer *buffer_info;
 703         struct sk_buff *skb;
 704         unsigned int i;
 705         unsigned int bufsz = 256 - 16 /* for skb_reserve */;
 706
 707         i = rx_ring->next_to_use;
 708         buffer_info = &rx_ring->buffer_info[i];
 709
 710         while (cleaned_count--) {
 711                 skb = buffer_info->skb;
 712                 if (skb) {
 713                         skb_trim(skb, 0);
 714                         goto check_page;
 715                 }
 716
 717                 skb = __netdev_alloc_skb_ip_align(netdev, bufsz, gfp);
 718                 if (unlikely(!skb)) {
 719                         /* Better luck next round */
 720                         adapter->alloc_rx_buff_failed++;
 721                         break;
 722                 }
 723
 724                 buffer_info->skb = skb;
 725 check_page:
 726                 /* allocate a new page if necessary */
 727                 if (!buffer_info->page) {
 728                         buffer_info->page = alloc_page(gfp);
 729                         if (unlikely(!buffer_info->page)) {
 730                                 adapter->alloc_rx_buff_failed++;
 731                                 break;
 732                         }
 733                 }
 734
 735                 if (!buffer_info->dma)
 736                         buffer_info->dma = dma_map_page(&pdev->dev,
 737                                                         buffer_info->page, 0,
 738                                                         PAGE_SIZE,
 739                                                         DMA_FROM_DEVICE);
 740
 741                 rx_desc = E1000_RX_DESC(*rx_ring, i);
 742                 rx_desc->buffer_addr = cpu_to_le64(buffer_info->dma);
 743
 744                 if (unlikely(++i == rx_ring->count))
 745                         i = 0;
 746                 buffer_info = &rx_ring->buffer_info[i];
 747         }
 748
 749         if (likely(rx_ring->next_to_use != i)) {
 750                 rx_ring->next_to_use = i;
 751                 if (unlikely(i-- == 0))
 752                         i = (rx_ring->count - 1);
 753
 754                 /* Force memory writes to complete before letting h/w
 755                  * know there are new descriptors to fetch.  (Only
 756                  * applicable for weak-ordered memory model archs,
 757                  * such as IA-64). */
 758                 wmb();
 759                 writel(i, adapter->hw.hw_addr + rx_ring->tail);
 760         }
 761 }
 762
 763 /**
 764  * e1000_clean_rx_irq - Send received data up the network stack; legacy
 765  * @adapter: board private structure
 766  *
 767  * the return value indicates whether actual cleaning was done, there
 768  * is no guarantee that everything was cleaned
 769  **/
 770 static bool e1000_clean_rx_irq(struct e1000_adapter *adapter,
 771                                int *work_done, int work_to_do)
 772 {
 773         struct net_device *netdev = adapter->netdev;
 774         struct pci_dev *pdev = adapter->pdev;
 775         struct e1000_hw *hw = &adapter->hw;
 776         struct e1000_ring *rx_ring = adapter->rx_ring;
 777         struct e1000_rx_desc *rx_desc, *next_rxd;
 778         struct e1000_buffer *buffer_info, *next_buffer;
 779         u32 length;
 780         unsigned int i;
 781         int cleaned_count = 0;
 782         bool cleaned = 0;
 783         unsigned int total_rx_bytes = 0, total_rx_packets = 0;
 784
 785         i = rx_ring->next_to_clean;
 786         rx_desc = E1000_RX_DESC(*rx_ring, i);
 787         buffer_info = &rx_ring->buffer_info[i];
 788
 789         while (rx_desc->status & E1000_RXD_STAT_DD) {
 790                 struct sk_buff *skb;
 791                 u8 status;
 792
 793                 if (*work_done >= work_to_do)
 794                         break;
 795                 (*work_done)++;
 796                 rmb();  /* read descriptor and rx_buffer_info after status DD */
 797
 798                 status = rx_desc->status;
 799                 skb = buffer_info->skb;
 800                 buffer_info->skb = NULL;
 801
 802                 prefetch(skb->data - NET_IP_ALIGN);
 803
 804                 i++;
 805                 if (i == rx_ring->count)
 806                         i = 0;
 807                 next_rxd = E1000_RX_DESC(*rx_ring, i);
 808                 prefetch(next_rxd);
 809
 810                 next_buffer = &rx_ring->buffer_info[i];
 811
 812                 cleaned = 1;
 813                 cleaned_count++;
 814                 dma_unmap_single(&pdev->dev,
 815                                  buffer_info->dma,
 816                                  adapter->rx_buffer_len,
 817                                  DMA_FROM_DEVICE);
 818                 buffer_info->dma = 0;
 819
 820                 length = le16_to_cpu(rx_desc->length);
 821
 822                 /*
 823                  * !EOP means multiple descriptors were used to store a single
 824                  * packet, if that's the case we need to toss it.  In fact, we
 825                  * need to toss every packet with the EOP bit clear and the
 826                  * next frame that _does_ have the EOP bit set, as it is by
 827                  * definition only a frame fragment
 828                  */
 829                 if (unlikely(!(status & E1000_RXD_STAT_EOP)))
 830                         adapter->flags2 |= FLAG2_IS_DISCARDING;
 831
 832                 if (adapter->flags2 & FLAG2_IS_DISCARDING) {
 833                         /* All receives must fit into a single buffer */
 834                         e_dbg("Receive packet consumed multiple buffers\n");
 835                         /* recycle */
 836                         buffer_info->skb = skb;
 837                         if (status & E1000_RXD_STAT_EOP)
 838                                 adapter->flags2 &= ~FLAG2_IS_DISCARDING;
 839                         goto next_desc;
 840                 }
 841
 842                 if (rx_desc->errors & E1000_RXD_ERR_FRAME_ERR_MASK) {
 843                         /* recycle */
 844                         buffer_info->skb = skb;
 845                         goto next_desc;
 846                 }
 847
 848                 /* adjust length to remove Ethernet CRC */
 849                 if (!(adapter->flags2 & FLAG2_CRC_STRIPPING))
 850                         length -= 4;
 851
 852                 total_rx_bytes += length;
 853                 total_rx_packets++;
 854
 855                 /*
 856                  * code added for copybreak, this should improve
 857                  * performance for small packets with large amounts
 858                  * of reassembly being done in the stack
 859                  */
 860                 if (length < copybreak) {
 861                         struct sk_buff *new_skb =
 862                             netdev_alloc_skb_ip_align(netdev, length);
 863                         if (new_skb) {
 864                                 skb_copy_to_linear_data_offset(new_skb,
 865                                                                -NET_IP_ALIGN,
 866                                                                (skb->data -
 867                                                                 NET_IP_ALIGN),
 868                                                                (length +
 869                                                                 NET_IP_ALIGN));
 870                                 /* save the skb in buffer_info as good */
 871                                 buffer_info->skb = skb;
 872                                 skb = new_skb;
 873                         }
 874                         /* else just continue with the old one */
 875                 }
 876                 /* end copybreak code */
 877                 skb_put(skb, length);
 878
 879                 /* Receive Checksum Offload */
 880                 e1000_rx_checksum(adapter,
 881                                   (u32)(status) |
 882                                   ((u32)(rx_desc->errors) << 24),
 883                                   le16_to_cpu(rx_desc->csum), skb);
 884
 885                 e1000_receive_skb(adapter, netdev, skb,status,rx_desc->special);
 886
 887 next_desc:
 888                 rx_desc->status = 0;
 889
 890                 /* return some buffers to hardware, one at a time is too slow */
 891                 if (cleaned_count >= E1000_RX_BUFFER_WRITE) {
 892                         adapter->alloc_rx_buf(adapter, cleaned_count,
 893                                               GFP_ATOMIC);
 894                         cleaned_count = 0;
 895                 }
 896
 897                 /* use prefetched values */
 898                 rx_desc = next_rxd;
 899                 buffer_info = next_buffer;
 900         }
 901         rx_ring->next_to_clean = i;
 902
 903         cleaned_count = e1000_desc_unused(rx_ring);
 904         if (cleaned_count)
 905                 adapter->alloc_rx_buf(adapter, cleaned_count, GFP_ATOMIC);
 906
 907         adapter->total_rx_bytes += total_rx_bytes;
 908         adapter->total_rx_packets += total_rx_packets;
 909         return cleaned;
 910 }
 911
 912 static void e1000_put_txbuf(struct e1000_adapter *adapter,
 913                              struct e1000_buffer *buffer_info)
 914 {
 915         if (buffer_info->dma) {
 916                 if (buffer_info->mapped_as_page)
 917                         dma_unmap_page(&adapter->pdev->dev, buffer_info->dma,
 918                                        buffer_info->length, DMA_TO_DEVICE);
 919                 else
 920                         dma_unmap_single(&adapter->pdev->dev, buffer_info->dma,
 921                                          buffer_info->length, DMA_TO_DEVICE);
 922                 buffer_info->dma = 0;
 923         }
 924         if (buffer_info->skb) {
 925                 dev_kfree_skb_any(buffer_info->skb);
 926                 buffer_info->skb = NULL;
 927         }
 928         buffer_info->time_stamp = 0;
 929 }
 930
 931 static void e1000_print_hw_hang(struct work_struct *work)
 932 {
 933         struct e1000_adapter *adapter = container_of(work,
 934                                                      struct e1000_adapter,
 935                                                      print_hang_task);
 936         struct e1000_ring *tx_ring = adapter->tx_ring;
 937         unsigned int i = tx_ring->next_to_clean;
 938         unsigned int eop = tx_ring->buffer_info[i].next_to_watch;
 939         struct e1000_tx_desc *eop_desc = E1000_TX_DESC(*tx_ring, eop);
 940         struct e1000_hw *hw = &adapter->hw;
 941         u16 phy_status, phy_1000t_status, phy_ext_status;
 942         u16 pci_status;
 943
 944         if (test_bit(__E1000_DOWN, &adapter->state))
 945                 return;
 946
 947         e1e_rphy(hw, PHY_STATUS, &phy_status);
 948         e1e_rphy(hw, PHY_1000T_STATUS, &phy_1000t_status);
 949         e1e_rphy(hw, PHY_EXT_STATUS, &phy_ext_status);
 950
 951         pci_read_config_word(adapter->pdev, PCI_STATUS, &pci_status);
 952
 953         /* detected Hardware unit hang */
 954         e_err("Detected Hardware Unit Hang:\n"
 955               "  TDH                  <%x>\n"
 956               "  TDT                  <%x>\n"
 957               "  next_to_use          <%x>\n"
 958               "  next_to_clean        <%x>\n"
 959               "buffer_info[next_to_clean]:\n"
 960               "  time_stamp           <%lx>\n"
 961               "  next_to_watch        <%x>\n"
 962               "  jiffies              <%lx>\n"
 963               "  next_to_watch.status <%x>\n"
 964               "MAC Status             <%x>\n"
 965               "PHY Status             <%x>\n"
 966               "PHY 1000BASE-T Status  <%x>\n"
 967               "PHY Extended Status    <%x>\n"
 968               "PCI Status             <%x>\n",
 969               readl(adapter->hw.hw_addr + tx_ring->head),
 970               readl(adapter->hw.hw_addr + tx_ring->tail),
 971               tx_ring->next_to_use,
 972               tx_ring->next_to_clean,
 973               tx_ring->buffer_info[eop].time_stamp,
 974               eop,
 975               jiffies,
 976               eop_desc->upper.fields.status,
 977               er32(STATUS),
 978               phy_status,
 979               phy_1000t_status,
 980               phy_ext_status,
 981               pci_status);
 982 }
 983
 984 /**
 985  * e1000_clean_tx_irq - Reclaim resources after transmit completes
 986  * @adapter: board private structure
 987  *
 988  * the return value indicates whether actual cleaning was done, there
 989  * is no guarantee that everything was cleaned
 990  **/
 991 static bool e1000_clean_tx_irq(struct e1000_adapter *adapter)
 992 {
 993         struct net_device *netdev = adapter->netdev;
 994         struct e1000_hw *hw = &adapter->hw;
 995         struct e1000_ring *tx_ring = adapter->tx_ring;
 996         struct e1000_tx_desc *tx_desc, *eop_desc;
 997         struct e1000_buffer *buffer_info;
 998         unsigned int i, eop;
 999         unsigned int count = 0;
1000         unsigned int total_tx_bytes = 0, total_tx_packets = 0;
1001
1002         i = tx_ring->next_to_clean;
1003         eop = tx_ring->buffer_info[i].next_to_watch;
1004         eop_desc = E1000_TX_DESC(*tx_ring, eop);
1005
1006         while ((eop_desc->upper.data & cpu_to_le32(E1000_TXD_STAT_DD)) &&
1007                (count < tx_ring->count)) {
1008                 bool cleaned = false;
1009                 rmb(); /* read buffer_info after eop_desc */
1010                 for (; !cleaned; count++) {
1011                         tx_desc = E1000_TX_DESC(*tx_ring, i);
1012                         buffer_info = &tx_ring->buffer_info[i];
1013                         cleaned = (i == eop);
1014
1015                         if (cleaned) {
1016                                 total_tx_packets += buffer_info->segs;
1017                                 total_tx_bytes += buffer_info->bytecount;
1018                         }
1019
1020                         e1000_put_txbuf(adapter, buffer_info);
1021                         tx_desc->upper.data = 0;
1022
1023                         i++;
1024                         if (i == tx_ring->count)
1025                                 i = 0;
1026                 }
1027
1028                 if (i == tx_ring->next_to_use)
1029                         break;
1030                 eop = tx_ring->buffer_info[i].next_to_watch;
1031                 eop_desc = E1000_TX_DESC(*tx_ring, eop);
1032         }
1033
1034         tx_ring->next_to_clean = i;
1035
1036 #define TX_WAKE_THRESHOLD 32
1037         if (count && netif_carrier_ok(netdev) &&
1038             e1000_desc_unused(tx_ring) >= TX_WAKE_THRESHOLD) {
1039                 /* Make sure that anybody stopping the queue after this
1040                  * sees the new next_to_clean.
1041                  */
1042                 smp_mb();
1043
1044                 if (netif_queue_stopped(netdev) &&
1045                     !(test_bit(__E1000_DOWN, &adapter->state))) {
1046                         netif_wake_queue(netdev);
1047                         ++adapter->restart_queue;
1048                 }
1049         }
1050
1051         if (adapter->detect_tx_hung) {
1052                 /*
1053                  * Detect a transmit hang in hardware, this serializes the
1054                  * check with the clearing of time_stamp and movement of i
1055                  */
1056                 adapter->detect_tx_hung = 0;
1057                 if (tx_ring->buffer_info[i].time_stamp &&
1058                     time_after(jiffies, tx_ring->buffer_info[i].time_stamp
1059                                + (adapter->tx_timeout_factor * HZ)) &&
1060                     !(er32(STATUS) & E1000_STATUS_TXOFF)) {
1061                         schedule_work(&adapter->print_hang_task);
1062                         netif_stop_queue(netdev);
1063                 }
1064         }
1065         adapter->total_tx_bytes += total_tx_bytes;
1066         adapter->total_tx_packets += total_tx_packets;
1067         return count < tx_ring->count;
1068 }
1069
1070 /**
1071  * e1000_clean_rx_irq_ps - Send received data up the network stack; packet split
1072  * @adapter: board private structure
1073  *
1074  * the return value indicates whether actual cleaning was done, there
1075  * is no guarantee that everything was cleaned
1076  **/
1077 static bool e1000_clean_rx_irq_ps(struct e1000_adapter *adapter,
1078                                   int *work_done, int work_to_do)
1079 {
1080         struct e1000_hw *hw = &adapter->hw;
1081         union e1000_rx_desc_packet_split *rx_desc, *next_rxd;
1082         struct net_device *netdev = adapter->netdev;
1083         struct pci_dev *pdev = adapter->pdev;
1084         struct e1000_ring *rx_ring = adapter->rx_ring;
1085         struct e1000_buffer *buffer_info, *next_buffer;
1086         struct e1000_ps_page *ps_page;
1087         struct sk_buff *skb;
1088         unsigned int i, j;
1089         u32 length, staterr;
1090         int cleaned_count = 0;
1091         bool cleaned = 0;
1092         unsigned int total_rx_bytes = 0, total_rx_packets = 0;
1093
1094         i = rx_ring->next_to_clean;
1095         rx_desc = E1000_RX_DESC_PS(*rx_ring, i);
1096         staterr = le32_to_cpu(rx_desc->wb.middle.status_error);
1097         buffer_info = &rx_ring->buffer_info[i];
1098
1099         while (staterr & E1000_RXD_STAT_DD) {
1100                 if (*work_done >= work_to_do)
1101                         break;
1102                 (*work_done)++;
1103                 skb = buffer_info->skb;
1104                 rmb();  /* read descriptor and rx_buffer_info after status DD */
1105
1106                 /* in the packet split case this is header only */
1107                 prefetch(skb->data - NET_IP_ALIGN);
1108
1109                 i++;
1110                 if (i == rx_ring->count)
1111                         i = 0;
1112                 next_rxd = E1000_RX_DESC_PS(*rx_ring, i);
1113                 prefetch(next_rxd);
1114
1115                 next_buffer = &rx_ring->buffer_info[i];
1116
1117                 cleaned = 1;
1118                 cleaned_count++;
1119                 dma_unmap_single(&pdev->dev, buffer_info->dma,
1120                                  adapter->rx_ps_bsize0, DMA_FROM_DEVICE);
1121                 buffer_info->dma = 0;
1122
1123                 /* see !EOP comment in other Rx routine */
1124                 if (!(staterr & E1000_RXD_STAT_EOP))
1125                         adapter->flags2 |= FLAG2_IS_DISCARDING;
1126
1127                 if (adapter->flags2 & FLAG2_IS_DISCARDING) {
1128                         e_dbg("Packet Split buffers didn't pick up the full "
1129                               "packet\n");
1130                         dev_kfree_skb_irq(skb);
1131                         if (staterr & E1000_RXD_STAT_EOP)
1132                                 adapter->flags2 &= ~FLAG2_IS_DISCARDING;
1133                         goto next_desc;
1134                 }
1135
1136                 if (staterr & E1000_RXDEXT_ERR_FRAME_ERR_MASK) {
1137                         dev_kfree_skb_irq(skb);
1138                         goto next_desc;
1139                 }
1140
1141                 length = le16_to_cpu(rx_desc->wb.middle.length0);
1142
1143                 if (!length) {
1144                         e_dbg("Last part of the packet spanning multiple "
1145                               "descriptors\n");
1146                         dev_kfree_skb_irq(skb);
1147                         goto next_desc;
1148                 }
1149
1150                 /* Good Receive */
1151                 skb_put(skb, length);
1152
1153                 {
1154                 /*
1155                  * this looks ugly, but it seems compiler issues make it
1156                  * more efficient than reusing j
1157                  */
1158                 int l1 = le16_to_cpu(rx_desc->wb.upper.length[0]);
1159
1160                 /*
1161                  * page alloc/put takes too long and effects small packet
1162                  * throughput, so unsplit small packets and save the alloc/put
1163                  * only valid in softirq (napi) context to call kmap_*
1164                  */
1165                 if (l1 && (l1 <= copybreak) &&
1166                     ((length + l1) <= adapter->rx_ps_bsize0)) {
1167                         u8 *vaddr;
1168
1169                         ps_page = &buffer_info->ps_pages[0];
1170
1171                         /*
1172                          * there is no documentation about how to call
1173                          * kmap_atomic, so we can't hold the mapping
1174                          * very long
1175                          */
1176                         dma_sync_single_for_cpu(&pdev->dev, ps_page->dma,
1177                                                 PAGE_SIZE, DMA_FROM_DEVICE);
1178                         vaddr = kmap_atomic(ps_page->page, KM_SKB_DATA_SOFTIRQ);
1179                         memcpy(skb_tail_pointer(skb), vaddr, l1);
1180                         kunmap_atomic(vaddr, KM_SKB_DATA_SOFTIRQ);
1181                         dma_sync_single_for_device(&pdev->dev, ps_page->dma,
1182                                                    PAGE_SIZE, DMA_FROM_DEVICE);
1183
1184                         /* remove the CRC */
1185                         if (!(adapter->flags2 & FLAG2_CRC_STRIPPING))
1186                                 l1 -= 4;
1187
1188                         skb_put(skb, l1);
1189                         goto copydone;
1190                 } /* if */
1191                 }
1192
1193                 for (j = 0; j < PS_PAGE_BUFFERS; j++) {
1194                         length = le16_to_cpu(rx_desc->wb.upper.length[j]);
1195                         if (!length)
1196                                 break;
1197
1198                         ps_page = &buffer_info->ps_pages[j];
1199                         dma_unmap_page(&pdev->dev, ps_page->dma, PAGE_SIZE,
1200                                        DMA_FROM_DEVICE);
1201                         ps_page->dma = 0;
1202                         skb_fill_page_desc(skb, j, ps_page->page, 0, length);
1203                         ps_page->page = NULL;
1204                         skb->len += length;
1205                         skb->data_len += length;
1206                         skb->truesize += length;
1207                 }
1208
1209                 /* strip the ethernet crc, problem is we're using pages now so
1210                  * this whole operation can get a little cpu intensive
1211                  */
1212                 if (!(adapter->flags2 & FLAG2_CRC_STRIPPING))
1213                         pskb_trim(skb, skb->len - 4);
1214
1215 copydone:
1216                 total_rx_bytes += skb->len;
1217                 total_rx_packets++;
1218
1219                 e1000_rx_checksum(adapter, staterr, le16_to_cpu(
1220                         rx_desc->wb.lower.hi_dword.csum_ip.csum), skb);
1221
1222                 if (rx_desc->wb.upper.header_status &
1223                            cpu_to_le16(E1000_RXDPS_HDRSTAT_HDRSP))
1224                         adapter->rx_hdr_split++;
1225
1226                 e1000_receive_skb(adapter, netdev, skb,
1227                                   staterr, rx_desc->wb.middle.vlan);
1228
1229 next_desc:
1230                 rx_desc->wb.middle.status_error &= cpu_to_le32(~0xFF);
1231                 buffer_info->skb = NULL;
1232
1233                 /* return some buffers to hardware, one at a time is too slow */
1234                 if (cleaned_count >= E1000_RX_BUFFER_WRITE) {
1235                         adapter->alloc_rx_buf(adapter, cleaned_count,
1236                                               GFP_ATOMIC);
1237                         cleaned_count = 0;
1238                 }
1239
1240                 /* use prefetched values */
1241                 rx_desc = next_rxd;
1242                 buffer_info = next_buffer;
1243
1244                 staterr = le32_to_cpu(rx_desc->wb.middle.status_error);
1245         }
1246         rx_ring->next_to_clean = i;
1247
1248         cleaned_count = e1000_desc_unused(rx_ring);
1249         if (cleaned_count)
1250                 adapter->alloc_rx_buf(adapter, cleaned_count, GFP_ATOMIC);
1251
1252         adapter->total_rx_bytes += total_rx_bytes;
1253         adapter->total_rx_packets += total_rx_packets;
1254         return cleaned;
1255 }
1256
1257 /**
1258  * e1000_consume_page - helper function
1259  **/
1260 static void e1000_consume_page(struct e1000_buffer *bi, struct sk_buff *skb,
1261                                u16 length)
1262 {
1263         bi->page = NULL;
1264         skb->len += length;
1265         skb->data_len += length;
1266         skb->truesize += length;
1267 }
1268
1269 /**
1270  * e1000_clean_jumbo_rx_irq - Send received data up the network stack; legacy
1271  * @adapter: board private structure
1272  *
1273  * the return value indicates whether actual cleaning was done, there
1274  * is no guarantee that everything was cleaned
1275  **/
1276
1277 static bool e1000_clean_jumbo_rx_irq(struct e1000_adapter *adapter,
1278                                      int *work_done, int work_to_do)
1279 {
1280         struct net_device *netdev = adapter->netdev;
1281         struct pci_dev *pdev = adapter->pdev;
1282         struct e1000_ring *rx_ring = adapter->rx_ring;
1283         struct e1000_rx_desc *rx_desc, *next_rxd;
1284         struct e1000_buffer *buffer_info, *next_buffer;
1285         u32 length;
1286         unsigned int i;
1287         int cleaned_count = 0;
1288         bool cleaned = false;
1289         unsigned int total_rx_bytes=0, total_rx_packets=0;
1290
1291         i = rx_ring->next_to_clean;
1292         rx_desc = E1000_RX_DESC(*rx_ring, i);
1293         buffer_info = &rx_ring->buffer_info[i];
1294
1295         while (rx_desc->status & E1000_RXD_STAT_DD) {
1296                 struct sk_buff *skb;
1297                 u8 status;
1298
1299                 if (*work_done >= work_to_do)
1300                         break;
1301                 (*work_done)++;
1302                 rmb();  /* read descriptor and rx_buffer_info after status DD */
1303
1304                 status = rx_desc->status;
1305                 skb = buffer_info->skb;
1306                 buffer_info->skb = NULL;
1307
1308                 ++i;
1309                 if (i == rx_ring->count)
1310                         i = 0;
1311                 next_rxd = E1000_RX_DESC(*rx_ring, i);
1312                 prefetch(next_rxd);
1313
1314                 next_buffer = &rx_ring->buffer_info[i];
1315
1316                 cleaned = true;
1317                 cleaned_count++;
1318                 dma_unmap_page(&pdev->dev, buffer_info->dma, PAGE_SIZE,
1319                                DMA_FROM_DEVICE);
1320                 buffer_info->dma = 0;
1321
1322                 length = le16_to_cpu(rx_desc->length);
1323
1324                 /* errors is only valid for DD + EOP descriptors */
1325                 if (unlikely((status & E1000_RXD_STAT_EOP) &&
1326                     (rx_desc->errors & E1000_RXD_ERR_FRAME_ERR_MASK))) {
1327                                 /* recycle both page and skb */
1328                                 buffer_info->skb = skb;
1329                                 /* an error means any chain goes out the window
1330                                  * too */
1331                                 if (rx_ring->rx_skb_top)
1332                                         dev_kfree_skb_irq(rx_ring->rx_skb_top);
1333                                 rx_ring->rx_skb_top = NULL;
1334                                 goto next_desc;
1335                 }
1336
1337 #define rxtop (rx_ring->rx_skb_top)
1338                 if (!(status & E1000_RXD_STAT_EOP)) {
1339                         /* this descriptor is only the beginning (or middle) */
1340                         if (!rxtop) {
1341                                 /* this is the beginning of a chain */
1342                                 rxtop = skb;
1343                                 skb_fill_page_desc(rxtop, 0, buffer_info->page,
1344                                                    0, length);
1345                         } else {
1346                                 /* this is the middle of a chain */
1347                                 skb_fill_page_desc(rxtop,
1348                                     skb_shinfo(rxtop)->nr_frags,
1349                                     buffer_info->page, 0, length);
1350                                 /* re-use the skb, only consumed the page */
1351                                 buffer_info->skb = skb;
1352                         }
1353                         e1000_consume_page(buffer_info, rxtop, length);
1354                         goto next_desc;
1355                 } else {
1356                         if (rxtop) {
1357                                 /* end of the chain */
1358                                 skb_fill_page_desc(rxtop,
1359                                     skb_shinfo(rxtop)->nr_frags,
1360                                     buffer_info->page, 0, length);
1361                                 /* re-use the current skb, we only consumed the
1362                                  * page */
1363                                 buffer_info->skb = skb;
1364                                 skb = rxtop;
1365                                 rxtop = NULL;
1366                                 e1000_consume_page(buffer_info, skb, length);
1367                         } else {
1368                                 /* no chain, got EOP, this buf is the packet
1369                                  * copybreak to save the put_page/alloc_page */
1370                                 if (length <= copybreak &&
1371                                     skb_tailroom(skb) >= length) {
1372                                         u8 *vaddr;
1373                                         vaddr = kmap_atomic(buffer_info->page,
1374                                                            KM_SKB_DATA_SOFTIRQ);
1375                                         memcpy(skb_tail_pointer(skb), vaddr,
1376                                                length);
1377                                         kunmap_atomic(vaddr,
1378                                                       KM_SKB_DATA_SOFTIRQ);
1379                                         /* re-use the page, so don't erase
1380                                          * buffer_info->page */
1381                                         skb_put(skb, length);
1382                                 } else {
1383                                         skb_fill_page_desc(skb, 0,
1384                                                            buffer_info->page, 0,
1385                                                            length);
1386                                         e1000_consume_page(buffer_info, skb,
1387                                                            length);
1388                                 }
1389                         }
1390                 }
1391
1392                 /* Receive Checksum Offload XXX recompute due to CRC strip? */
1393                 e1000_rx_checksum(adapter,
1394                                   (u32)(status) |
1395                                   ((u32)(rx_desc->errors) << 24),
1396                                   le16_to_cpu(rx_desc->csum), skb);
1397
1398                 /* probably a little skewed due to removing CRC */
1399                 total_rx_bytes += skb->len;
1400                 total_rx_packets++;
1401
1402                 /* eth type trans needs skb->data to point to something */
1403                 if (!pskb_may_pull(skb, ETH_HLEN)) {
1404                         e_err("pskb_may_pull failed.\n");
1405                         dev_kfree_skb_irq(skb);
1406                         goto next_desc;
1407                 }
1408
1409                 e1000_receive_skb(adapter, netdev, skb, status,
1410                                   rx_desc->special);
1411
1412 next_desc:
1413                 rx_desc->status = 0;
1414
1415                 /* return some buffers to hardware, one at a time is too slow */
1416                 if (unlikely(cleaned_count >= E1000_RX_BUFFER_WRITE)) {
1417                         adapter->alloc_rx_buf(adapter, cleaned_count,
1418                                               GFP_ATOMIC);
1419                         cleaned_count = 0;
1420                 }
1421
1422                 /* use prefetched values */
1423                 rx_desc = next_rxd;
1424                 buffer_info = next_buffer;
1425         }
1426         rx_ring->next_to_clean = i;
1427
1428         cleaned_count = e1000_desc_unused(rx_ring);
1429         if (cleaned_count)
1430                 adapter->alloc_rx_buf(adapter, cleaned_count, GFP_ATOMIC);
1431
1432         adapter->total_rx_bytes += total_rx_bytes;
1433         adapter->total_rx_packets += total_rx_packets;
1434         return cleaned;
1435 }
1436
1437 /**
1438  * e1000_clean_rx_ring - Free Rx Buffers per Queue
1439  * @adapter: board private structure
1440  **/
1441 static void e1000_clean_rx_ring(struct e1000_adapter *adapter)
1442 {
1443         struct e1000_ring *rx_ring = adapter->rx_ring;
1444         struct e1000_buffer *buffer_info;
1445         struct e1000_ps_page *ps_page;
1446         struct pci_dev *pdev = adapter->pdev;
1447         unsigned int i, j;
1448
1449         /* Free all the Rx ring sk_buffs */
1450         for (i = 0; i < rx_ring->count; i++) {
1451                 buffer_info = &rx_ring->buffer_info[i];
1452                 if (buffer_info->dma) {
1453                         if (adapter->clean_rx == e1000_clean_rx_irq)
1454                                 dma_unmap_single(&pdev->dev, buffer_info->dma,
1455                                                  adapter->rx_buffer_len,
1456                                                  DMA_FROM_DEVICE);
1457                         else if (adapter->clean_rx == e1000_clean_jumbo_rx_irq)
1458                                 dma_unmap_page(&pdev->dev, buffer_info->dma,
1459                                                PAGE_SIZE,
1460                                                DMA_FROM_DEVICE);
1461                         else if (adapter->clean_rx == e1000_clean_rx_irq_ps)
1462                                 dma_unmap_single(&pdev->dev, buffer_info->dma,
1463                                                  adapter->rx_ps_bsize0,
1464                                                  DMA_FROM_DEVICE);
1465                         buffer_info->dma = 0;
1466                 }
1467
1468                 if (buffer_info->page) {
1469                         put_page(buffer_info->page);
1470                         buffer_info->page = NULL;
1471                 }
1472
1473                 if (buffer_info->skb) {
1474                         dev_kfree_skb(buffer_info->skb);
1475                         buffer_info->skb = NULL;
1476                 }
1477
1478                 for (j = 0; j < PS_PAGE_BUFFERS; j++) {
1479                         ps_page = &buffer_info->ps_pages[j];
1480                         if (!ps_page->page)
1481                                 break;
1482                         dma_unmap_page(&pdev->dev, ps_page->dma, PAGE_SIZE,
1483                                        DMA_FROM_DEVICE);
1484                         ps_page->dma = 0;
1485                         put_page(ps_page->page);
1486                         ps_page->page = NULL;
1487                 }
1488         }
1489
1490         /* there also may be some cached data from a chained receive */
1491         if (rx_ring->rx_skb_top) {
1492                 dev_kfree_skb(rx_ring->rx_skb_top);
1493                 rx_ring->rx_skb_top = NULL;
1494         }
1495
1496         /* Zero out the descriptor ring */
1497         memset(rx_ring->desc, 0, rx_ring->size);
1498
1499         rx_ring->next_to_clean = 0;
1500         rx_ring->next_to_use = 0;
1501         adapter->flags2 &= ~FLAG2_IS_DISCARDING;
1502
1503         writel(0, adapter->hw.hw_addr + rx_ring->head);
1504         writel(0, adapter->hw.hw_addr + rx_ring->tail);
1505 }
1506
1507 static void e1000e_downshift_workaround(struct work_struct *work)
1508 {
1509         struct e1000_adapter *adapter = container_of(work,
1510                                         struct e1000_adapter, downshift_task);
1511
1512         if (test_bit(__E1000_DOWN, &adapter->state))
1513                 return;
1514
1515         e1000e_gig_downshift_workaround_ich8lan(&adapter->hw);
1516 }
1517
1518 /**
1519  * e1000_intr_msi - Interrupt Handler
1520  * @irq: interrupt number
1521  * @data: pointer to a network interface device structure
1522  **/
1523 static irqreturn_t e1000_intr_msi(int irq, void *data)
1524 {
1525         struct net_device *netdev = data;
1526         struct e1000_adapter *adapter = netdev_priv(netdev);
1527         struct e1000_hw *hw = &adapter->hw;
1528         u32 icr = er32(ICR);
1529
1530         /*
1531          * read ICR disables interrupts using IAM
1532          */
1533
1534         if (icr & E1000_ICR_LSC) {
1535                 hw->mac.get_link_status = 1;
1536                 /*
1537                  * ICH8 workaround-- Call gig speed drop workaround on cable
1538                  * disconnect (LSC) before accessing any PHY registers
1539                  */
1540                 if ((adapter->flags & FLAG_LSC_GIG_SPEED_DROP) &&
1541                     (!(er32(STATUS) & E1000_STATUS_LU)))
1542                         schedule_work(&adapter->downshift_task);
1543
1544                 /*
1545                  * 80003ES2LAN workaround-- For packet buffer work-around on
1546                  * link down event; disable receives here in the ISR and reset
1547                  * adapter in watchdog
1548                  */
1549                 if (netif_carrier_ok(netdev) &&
1550                     adapter->flags & FLAG_RX_NEEDS_RESTART) {
1551                         /* disable receives */
1552                         u32 rctl = er32(RCTL);
1553                         ew32(RCTL, rctl & ~E1000_RCTL_EN);
1554                         adapter->flags |= FLAG_RX_RESTART_NOW;
1555                 }
1556                 /* guard against interrupt when we're going down */
1557                 if (!test_bit(__E1000_DOWN, &adapter->state))
1558                         mod_timer(&adapter->watchdog_timer, jiffies + 1);
1559         }
1560
1561         if (napi_schedule_prep(&adapter->napi)) {
1562                 adapter->total_tx_bytes = 0;
1563                 adapter->total_tx_packets = 0;
1564                 adapter->total_rx_bytes = 0;
1565                 adapter->total_rx_packets = 0;
1566                 __napi_schedule(&adapter->napi);
1567         }
1568
1569         return IRQ_HANDLED;
1570 }
1571
1572 /**
1573  * e1000_intr - Interrupt Handler
1574  * @irq: interrupt number
1575  * @data: pointer to a network interface device structure
1576  **/
1577 static irqreturn_t e1000_intr(int irq, void *data)
1578 {
1579         struct net_device *netdev = data;
1580         struct e1000_adapter *adapter = netdev_priv(netdev);
1581         struct e1000_hw *hw = &adapter->hw;
1582         u32 rctl, icr = er32(ICR);
1583
1584         if (!icr || test_bit(__E1000_DOWN, &adapter->state))
1585                 return IRQ_NONE;  /* Not our interrupt */
1586
1587         /*
1588          * IMS will not auto-mask if INT_ASSERTED is not set, and if it is
1589          * not set, then the adapter didn't send an interrupt
1590          */
1591         if (!(icr & E1000_ICR_INT_ASSERTED))
1592                 return IRQ_NONE;
1593
1594         /*
1595          * Interrupt Auto-Mask...upon reading ICR,
1596          * interrupts are masked.  No need for the
1597          * IMC write
1598          */
1599
1600         if (icr & E1000_ICR_LSC) {
1601                 hw->mac.get_link_status = 1;
1602                 /*
1603                  * ICH8 workaround-- Call gig speed drop workaround on cable
1604                  * disconnect (LSC) before accessing any PHY registers
1605                  */
1606                 if ((adapter->flags & FLAG_LSC_GIG_SPEED_DROP) &&
1607                     (!(er32(STATUS) & E1000_STATUS_LU)))
1608                         schedule_work(&adapter->downshift_task);
1609
1610                 /*
1611                  * 80003ES2LAN workaround--
1612                  * For packet buffer work-around on link down event;
1613                  * disable receives here in the ISR and
1614                  * reset adapter in watchdog
1615                  */
1616                 if (netif_carrier_ok(netdev) &&
1617                     (adapter->flags & FLAG_RX_NEEDS_RESTART)) {
1618                         /* disable receives */
1619                         rctl = er32(RCTL);
1620                         ew32(RCTL, rctl & ~E1000_RCTL_EN);
1621                         adapter->flags |= FLAG_RX_RESTART_NOW;
1622                 }
1623                 /* guard against interrupt when we're going down */
1624                 if (!test_bit(__E1000_DOWN, &adapter->state))
1625                         mod_timer(&adapter->watchdog_timer, jiffies + 1);
1626         }
1627
1628         if (napi_schedule_prep(&adapter->napi)) {
1629                 adapter->total_tx_bytes = 0;
1630                 adapter->total_tx_packets = 0;
1631                 adapter->total_rx_bytes = 0;
1632                 adapter->total_rx_packets = 0;
1633                 __napi_schedule(&adapter->napi);
1634         }
1635
1636         return IRQ_HANDLED;
1637 }
1638
1639 static irqreturn_t e1000_msix_other(int irq, void *data)
1640 {
1641         struct net_device *netdev = data;
1642         struct e1000_adapter *adapter = netdev_priv(netdev);
1643         struct e1000_hw *hw = &adapter->hw;
1644         u32 icr = er32(ICR);
1645
1646         if (!(icr & E1000_ICR_INT_ASSERTED)) {
1647                 if (!test_bit(__E1000_DOWN, &adapter->state))
1648                         ew32(IMS, E1000_IMS_OTHER);
1649                 return IRQ_NONE;
1650         }
1651
1652         if (icr & adapter->eiac_mask)
1653                 ew32(ICS, (icr & adapter->eiac_mask));
1654
1655         if (icr & E1000_ICR_OTHER) {
1656                 if (!(icr & E1000_ICR_LSC))
1657                         goto no_link_interrupt;
1658                 hw->mac.get_link_status = 1;
1659                 /* guard against interrupt when we're going down */
1660                 if (!test_bit(__E1000_DOWN, &adapter->state))
1661                         mod_timer(&adapter->watchdog_timer, jiffies + 1);
1662         }
1663
1664 no_link_interrupt:
1665         if (!test_bit(__E1000_DOWN, &adapter->state))
1666                 ew32(IMS, E1000_IMS_LSC | E1000_IMS_OTHER);
1667
1668         return IRQ_HANDLED;
1669 }
1670
1671
1672 static irqreturn_t e1000_intr_msix_tx(int irq, void *data)
1673 {
1674         struct net_device *netdev = data;
1675         struct e1000_adapter *adapter = netdev_priv(netdev);
1676         struct e1000_hw *hw = &adapter->hw;
1677         struct e1000_ring *tx_ring = adapter->tx_ring;
1678
1679
1680         adapter->total_tx_bytes = 0;
1681         adapter->total_tx_packets = 0;
1682
1683         if (!e1000_clean_tx_irq(adapter))
1684                 /* Ring was not completely cleaned, so fire another interrupt */
1685                 ew32(ICS, tx_ring->ims_val);
1686
1687         return IRQ_HANDLED;
1688 }
1689
1690 static irqreturn_t e1000_intr_msix_rx(int irq, void *data)
1691 {
1692         struct net_device *netdev = data;
1693         struct e1000_adapter *adapter = netdev_priv(netdev);
1694
1695         /* Write the ITR value calculated at the end of the
1696          * previous interrupt.
1697          */
1698         if (adapter->rx_ring->set_itr) {
1699                 writel(1000000000 / (adapter->rx_ring->itr_val * 256),
1700                        adapter->hw.hw_addr + adapter->rx_ring->itr_register);
1701                 adapter->rx_ring->set_itr = 0;
1702         }
1703
1704         if (napi_schedule_prep(&adapter->napi)) {
1705                 adapter->total_rx_bytes = 0;
1706                 adapter->total_rx_packets = 0;
1707                 __napi_schedule(&adapter->napi);
1708         }
1709         return IRQ_HANDLED;
1710 }
1711
1712 /**
1713  * e1000_configure_msix - Configure MSI-X hardware
1714  *
1715  * e1000_configure_msix sets up the hardware to properly
1716  * generate MSI-X interrupts.
1717  **/
1718 static void e1000_configure_msix(struct e1000_adapter *adapter)
1719 {
1720         struct e1000_hw *hw = &adapter->hw;
1721         struct e1000_ring *rx_ring = adapter->rx_ring;
1722         struct e1000_ring *tx_ring = adapter->tx_ring;
1723         int vector = 0;
1724         u32 ctrl_ext, ivar = 0;
1725
1726         adapter->eiac_mask = 0;
1727
1728         /* Workaround issue with spurious interrupts on 82574 in MSI-X mode */
1729         if (hw->mac.type == e1000_82574) {
1730                 u32 rfctl = er32(RFCTL);
1731                 rfctl |= E1000_RFCTL_ACK_DIS;
1732                 ew32(RFCTL, rfctl);
1733         }
1734
1735 #define E1000_IVAR_INT_ALLOC_VALID      0x8
1736         /* Configure Rx vector */
1737         rx_ring->ims_val = E1000_IMS_RXQ0;
1738         adapter->eiac_mask |= rx_ring->ims_val;
1739         if (rx_ring->itr_val)
1740                 writel(1000000000 / (rx_ring->itr_val * 256),
1741                        hw->hw_addr + rx_ring->itr_register);
1742         else
1743                 writel(1, hw->hw_addr + rx_ring->itr_register);
1744         ivar = E1000_IVAR_INT_ALLOC_VALID | vector;
1745
1746         /* Configure Tx vector */
1747         tx_ring->ims_val = E1000_IMS_TXQ0;
1748         vector++;
1749         if (tx_ring->itr_val)
1750                 writel(1000000000 / (tx_ring->itr_val * 256),
1751                        hw->hw_addr + tx_ring->itr_register);
1752         else
1753                 writel(1, hw->hw_addr + tx_ring->itr_register);
1754         adapter->eiac_mask |= tx_ring->ims_val;
1755         ivar |= ((E1000_IVAR_INT_ALLOC_VALID | vector) << 8);
1756
1757         /* set vector for Other Causes, e.g. link changes */
1758         vector++;
1759         ivar |= ((E1000_IVAR_INT_ALLOC_VALID | vector) << 16);
1760         if (rx_ring->itr_val)
1761                 writel(1000000000 / (rx_ring->itr_val * 256),
1762                        hw->hw_addr + E1000_EITR_82574(vector));
1763         else
1764                 writel(1, hw->hw_addr + E1000_EITR_82574(vector));
1765
1766         /* Cause Tx interrupts on every write back */
1767         ivar |= (1 << 31);
1768
1769         ew32(IVAR, ivar);
1770
1771         /* enable MSI-X PBA support */
1772         ctrl_ext = er32(CTRL_EXT);
1773         ctrl_ext |= E1000_CTRL_EXT_PBA_CLR;
1774
1775         /* Auto-Mask Other interrupts upon ICR read */
1776 #define E1000_EIAC_MASK_82574   0x01F00000
1777         ew32(IAM, ~E1000_EIAC_MASK_82574 | E1000_IMS_OTHER);
1778         ctrl_ext |= E1000_CTRL_EXT_EIAME;
1779         ew32(CTRL_EXT, ctrl_ext);
1780         e1e_flush();
1781 }
1782
1783 void e1000e_reset_interrupt_capability(struct e1000_adapter *adapter)
1784 {
1785         if (adapter->msix_entries) {
1786                 pci_disable_msix(adapter->pdev);
1787                 kfree(adapter->msix_entries);
1788                 adapter->msix_entries = NULL;
1789         } else if (adapter->flags & FLAG_MSI_ENABLED) {
1790                 pci_disable_msi(adapter->pdev);
1791                 adapter->flags &= ~FLAG_MSI_ENABLED;
1792         }
1793 }
1794
1795 /**
1796  * e1000e_set_interrupt_capability - set MSI or MSI-X if supported
1797  *
1798  * Attempt to configure interrupts using the best available
1799  * capabilities of the hardware and kernel.
1800  **/
1801 void e1000e_set_interrupt_capability(struct e1000_adapter *adapter)
1802 {
1803         int err;
1804         int i;
1805
1806         switch (adapter->int_mode) {
1807         case E1000E_INT_MODE_MSIX:
1808                 if (adapter->flags & FLAG_HAS_MSIX) {
1809                         adapter->num_vectors = 3; /* RxQ0, TxQ0 and other */
1810                         adapter->msix_entries = kcalloc(adapter->num_vectors,
1811                                                       sizeof(struct msix_entry),
1812                                                       GFP_KERNEL);
1813                         if (adapter->msix_entries) {
1814                                 for (i = 0; i < adapter->num_vectors; i++)
1815                                         adapter->msix_entries[i].entry = i;
1816
1817                                 err = pci_enable_msix(adapter->pdev,
1818                                                       adapter->msix_entries,
1819                                                       adapter->num_vectors);
1820                                 if (err == 0)
1821                                         return;
1822                         }
1823                         /* MSI-X failed, so fall through and try MSI */
1824                         e_err("Failed to initialize MSI-X interrupts.  "
1825                               "Falling back to MSI interrupts.\n");
1826                         e1000e_reset_interrupt_capability(adapter);
1827                 }
1828                 adapter->int_mode = E1000E_INT_MODE_MSI;
1829                 /* Fall through */
1830         case E1000E_INT_MODE_MSI:
1831                 if (!pci_enable_msi(adapter->pdev)) {
1832                         adapter->flags |= FLAG_MSI_ENABLED;
1833                 } else {
1834                         adapter->int_mode = E1000E_INT_MODE_LEGACY;
1835                         e_err("Failed to initialize MSI interrupts.  Falling "
1836                               "back to legacy interrupts.\n");
1837                 }
1838                 /* Fall through */
1839         case E1000E_INT_MODE_LEGACY:
1840                 /* Don't do anything; this is the system default */
1841                 break;
1842         }
1843
1844         /* store the number of vectors being used */
1845         adapter->num_vectors = 1;
1846 }
1847
1848 /**
1849  * e1000_request_msix - Initialize MSI-X interrupts
1850  *
1851  * e1000_request_msix allocates MSI-X vectors and requests interrupts from the
1852  * kernel.
1853  **/
1854 static int e1000_request_msix(struct e1000_adapter *adapter)
1855 {
1856         struct net_device *netdev = adapter->netdev;
1857         int err = 0, vector = 0;
1858
1859         if (strlen(netdev->name) < (IFNAMSIZ - 5))
1860                 snprintf(adapter->rx_ring->name,
1861                          sizeof(adapter->rx_ring->name) - 1,
1862                          "%s-rx-0", netdev->name);
1863         else
1864                 memcpy(adapter->rx_ring->name, netdev->name, IFNAMSIZ);
1865         err = request_irq(adapter->msix_entries[vector].vector,
1866                           e1000_intr_msix_rx, 0, adapter->rx_ring->name,
1867                           netdev);
1868         if (err)
1869                 goto out;
1870         adapter->rx_ring->itr_register = E1000_EITR_82574(vector);
1871         adapter->rx_ring->itr_val = adapter->itr;
1872         vector++;
1873
1874         if (strlen(netdev->name) < (IFNAMSIZ - 5))
1875                 snprintf(adapter->tx_ring->name,
1876                          sizeof(adapter->tx_ring->name) - 1,
1877                          "%s-tx-0", netdev->name);
1878         else
1879                 memcpy(adapter->tx_ring->name, netdev->name, IFNAMSIZ);
1880         err = request_irq(adapter->msix_entries[vector].vector,
1881                           e1000_intr_msix_tx, 0, adapter->tx_ring->name,
1882                           netdev);
1883         if (err)
1884                 goto out;
1885         adapter->tx_ring->itr_register = E1000_EITR_82574(vector);
1886         adapter->tx_ring->itr_val = adapter->itr;
1887         vector++;
1888
1889         err = request_irq(adapter->msix_entries[vector].vector,
1890                           e1000_msix_other, 0, netdev->name, netdev);
1891         if (err)
1892                 goto out;
1893
1894         e1000_configure_msix(adapter);
1895         return 0;
1896 out:
1897         return err;
1898 }
1899
1900 /**
1901  * e1000_request_irq - initialize interrupts
1902  *
1903  * Attempts to configure interrupts using the best available
1904  * capabilities of the hardware and kernel.
1905  **/
1906 static int e1000_request_irq(struct e1000_adapter *adapter)
1907 {
1908         struct net_device *netdev = adapter->netdev;
1909         int err;
1910
1911         if (adapter->msix_entries) {
1912                 err = e1000_request_msix(adapter);
1913                 if (!err)
1914                         return err;
1915                 /* fall back to MSI */
1916                 e1000e_reset_interrupt_capability(adapter);
1917                 adapter->int_mode = E1000E_INT_MODE_MSI;
1918                 e1000e_set_interrupt_capability(adapter);
1919         }
1920         if (adapter->flags & FLAG_MSI_ENABLED) {
1921                 err = request_irq(adapter->pdev->irq, e1000_intr_msi, 0,
1922                                   netdev->name, netdev);
1923                 if (!err)
1924                         return err;
1925
1926                 /* fall back to legacy interrupt */
1927                 e1000e_reset_interrupt_capability(adapter);
1928                 adapter->int_mode = E1000E_INT_MODE_LEGACY;
1929         }
1930
1931         err = request_irq(adapter->pdev->irq, e1000_intr, IRQF_SHARED,
1932                           netdev->name, netdev);
1933         if (err)
1934                 e_err("Unable to allocate interrupt, Error: %d\n", err);
1935
1936         return err;
1937 }
1938
1939 static void e1000_free_irq(struct e1000_adapter *adapter)
1940 {
1941         struct net_device *netdev = adapter->netdev;
1942
1943         if (adapter->msix_entries) {
1944                 int vector = 0;
1945
1946                 free_irq(adapter->msix_entries[vector].vector, netdev);
1947                 vector++;
1948
1949                 free_irq(adapter->msix_entries[vector].vector, netdev);
1950                 vector++;
1951
1952                 /* Other Causes interrupt vector */
1953                 free_irq(adapter->msix_entries[vector].vector, netdev);
1954                 return;
1955         }
1956
1957         free_irq(adapter->pdev->irq, netdev);
1958 }
1959
1960 /**
1961  * e1000_irq_disable - Mask off interrupt generation on the NIC
1962  **/
1963 static void e1000_irq_disable(struct e1000_adapter *adapter)
1964 {
1965         struct e1000_hw *hw = &adapter->hw;
1966
1967         ew32(IMC, ~0);
1968         if (adapter->msix_entries)
1969                 ew32(EIAC_82574, 0);
1970         e1e_flush();
1971
1972         if (adapter->msix_entries) {
1973                 int i;
1974                 for (i = 0; i < adapter->num_vectors; i++)
1975                         synchronize_irq(adapter->msix_entries[i].vector);
1976         } else {
1977                 synchronize_irq(adapter->pdev->irq);
1978         }
1979 }
1980
1981 /**
1982  * e1000_irq_enable - Enable default interrupt generation settings
1983  **/
1984 static void e1000_irq_enable(struct e1000_adapter *adapter)
1985 {
1986         struct e1000_hw *hw = &adapter->hw;
1987
1988         if (adapter->msix_entries) {
1989                 ew32(EIAC_82574, adapter->eiac_mask & E1000_EIAC_MASK_82574);
1990                 ew32(IMS, adapter->eiac_mask | E1000_IMS_OTHER | E1000_IMS_LSC);
1991         } else {
1992                 ew32(IMS, IMS_ENABLE_MASK);
1993         }
1994         e1e_flush();
1995 }
1996
1997 /**
1998  * e1000e_get_hw_control - get control of the h/w from f/w
1999  * @adapter: address of board private structure
2000  *
2001  * e1000e_get_hw_control sets {CTRL_EXT|SWSM}:DRV_LOAD bit.
2002  * For ASF and Pass Through versions of f/w this means that
2003  * the driver is loaded. For AMT version (only with 82573)
2004  * of the f/w this means that the network i/f is open.
2005  **/
2006 void e1000e_get_hw_control(struct e1000_adapter *adapter)
2007 {
2008         struct e1000_hw *hw = &adapter->hw;
2009         u32 ctrl_ext;
2010         u32 swsm;
2011
2012         /* Let firmware know the driver has taken over */
2013         if (adapter->flags & FLAG_HAS_SWSM_ON_LOAD) {
2014                 swsm = er32(SWSM);
2015                 ew32(SWSM, swsm | E1000_SWSM_DRV_LOAD);
2016         } else if (adapter->flags & FLAG_HAS_CTRLEXT_ON_LOAD) {
2017                 ctrl_ext = er32(CTRL_EXT);
2018                 ew32(CTRL_EXT, ctrl_ext | E1000_CTRL_EXT_DRV_LOAD);
2019         }
2020 }
2021
2022 /**
2023  * e1000e_release_hw_control - release control of the h/w to f/w
2024  * @adapter: address of board private structure
2025  *
2026  * e1000e_release_hw_control resets {CTRL_EXT|SWSM}:DRV_LOAD bit.
2027  * For ASF and Pass Through versions of f/w this means that the
2028  * driver is no longer loaded. For AMT version (only with 82573) i
2029  * of the f/w this means that the network i/f is closed.
2030  *
2031  **/
2032 void e1000e_release_hw_control(struct e1000_adapter *adapter)
2033 {
2034         struct e1000_hw *hw = &adapter->hw;
2035         u32 ctrl_ext;
2036         u32 swsm;
2037
2038         /* Let firmware taken over control of h/w */
2039         if (adapter->flags & FLAG_HAS_SWSM_ON_LOAD) {
2040                 swsm = er32(SWSM);
2041                 ew32(SWSM, swsm & ~E1000_SWSM_DRV_LOAD);
2042         } else if (adapter->flags & FLAG_HAS_CTRLEXT_ON_LOAD) {
2043                 ctrl_ext = er32(CTRL_EXT);
2044                 ew32(CTRL_EXT, ctrl_ext & ~E1000_CTRL_EXT_DRV_LOAD);
2045         }
2046 }
2047
2048 /**
2049  * @e1000_alloc_ring - allocate memory for a ring structure
2050  **/
2051 static int e1000_alloc_ring_dma(struct e1000_adapter *adapter,
2052                                 struct e1000_ring *ring)
2053 {
2054         struct pci_dev *pdev = adapter->pdev;
2055
2056         ring->desc = dma_alloc_coherent(&pdev->dev, ring->size, &ring->dma,
2057                                         GFP_KERNEL);
2058         if (!ring->desc)
2059                 return -ENOMEM;
2060
2061         return 0;
2062 }
2063
2064 /**
2065  * e1000e_setup_tx_resources - allocate Tx resources (Descriptors)
2066  * @adapter: board private structure
2067  *
2068  * Return 0 on success, negative on failure
2069  **/
2070 int e1000e_setup_tx_resources(struct e1000_adapter *adapter)
2071 {
2072         struct e1000_ring *tx_ring = adapter->tx_ring;
2073         int err = -ENOMEM, size;
2074
2075         size = sizeof(struct e1000_buffer) * tx_ring->count;
2076         tx_ring->buffer_info = vzalloc(size);
2077         if (!tx_ring->buffer_info)
2078                 goto err;
2079
2080         /* round up to nearest 4K */
2081         tx_ring->size = tx_ring->count * sizeof(struct e1000_tx_desc);
2082         tx_ring->size = ALIGN(tx_ring->size, 4096);
2083
2084         err = e1000_alloc_ring_dma(adapter, tx_ring);
2085         if (err)
2086                 goto err;
2087
2088         tx_ring->next_to_use = 0;
2089         tx_ring->next_to_clean = 0;
2090
2091         return 0;
2092 err:
2093         vfree(tx_ring->buffer_info);
2094         e_err("Unable to allocate memory for the transmit descriptor ring\n");
2095         return err;
2096 }
2097
2098 /**
2099  * e1000e_setup_rx_resources - allocate Rx resources (Descriptors)
2100  * @adapter: board private structure
2101  *
2102  * Returns 0 on success, negative on failure
2103  **/
2104 int e1000e_setup_rx_resources(struct e1000_adapter *adapter)
2105 {
2106         struct e1000_ring *rx_ring = adapter->rx_ring;
2107         struct e1000_buffer *buffer_info;
2108         int i, size, desc_len, err = -ENOMEM;
2109
2110         size = sizeof(struct e1000_buffer) * rx_ring->count;
2111         rx_ring->buffer_info = vzalloc(size);
2112         if (!rx_ring->buffer_info)
2113                 goto err;
2114
2115         for (i = 0; i < rx_ring->count; i++) {
2116                 buffer_info = &rx_ring->buffer_info[i];
2117                 buffer_info->ps_pages = kcalloc(PS_PAGE_BUFFERS,
2118                                                 sizeof(struct e1000_ps_page),
2119                                                 GFP_KERNEL);
2120                 if (!buffer_info->ps_pages)
2121                         goto err_pages;
2122         }
2123
2124         desc_len = sizeof(union e1000_rx_desc_packet_split);
2125
2126         /* Round up to nearest 4K */
2127         rx_ring->size = rx_ring->count * desc_len;
2128         rx_ring->size = ALIGN(rx_ring->size, 4096);
2129
2130         err = e1000_alloc_ring_dma(adapter, rx_ring);
2131         if (err)
2132                 goto err_pages;
2133
2134         rx_ring->next_to_clean = 0;
2135         rx_ring->next_to_use = 0;
2136         rx_ring->rx_skb_top = NULL;
2137
2138         return 0;
2139
2140 err_pages:
2141         for (i = 0; i < rx_ring->count; i++) {
2142                 buffer_info = &rx_ring->buffer_info[i];
2143                 kfree(buffer_info->ps_pages);
2144         }
2145 err:
2146         vfree(rx_ring->buffer_info);
2147         e_err("Unable to allocate memory for the receive descriptor ring\n");
2148         return err;
2149 }
2150
2151 /**
2152  * e1000_clean_tx_ring - Free Tx Buffers
2153  * @adapter: board private structure
2154  **/
2155 static void e1000_clean_tx_ring(struct e1000_adapter *adapter)
2156 {
2157         struct e1000_ring *tx_ring = adapter->tx_ring;
2158         struct e1000_buffer *buffer_info;
2159         unsigned long size;
2160         unsigned int i;
2161
2162         for (i = 0; i < tx_ring->count; i++) {
2163                 buffer_info = &tx_ring->buffer_info[i];
2164                 e1000_put_txbuf(adapter, buffer_info);
2165         }
2166
2167         size = sizeof(struct e1000_buffer) * tx_ring->count;
2168         memset(tx_ring->buffer_info, 0, size);
2169
2170         memset(tx_ring->desc, 0, tx_ring->size);
2171
2172         tx_ring->next_to_use = 0;
2173         tx_ring->next_to_clean = 0;
2174
2175         writel(0, adapter->hw.hw_addr + tx_ring->head);
2176         writel(0, adapter->hw.hw_addr + tx_ring->tail);
2177 }
2178
2179 /**
2180  * e1000e_free_tx_resources - Free Tx Resources per Queue
2181  * @adapter: board private structure
2182  *
2183  * Free all transmit software resources
2184  **/
2185 void e1000e_free_tx_resources(struct e1000_adapter *adapter)
2186 {
2187         struct pci_dev *pdev = adapter->pdev;
2188         struct e1000_ring *tx_ring = adapter->tx_ring;
2189
2190         e1000_clean_tx_ring(adapter);
2191
2192         vfree(tx_ring->buffer_info);
2193         tx_ring->buffer_info = NULL;
2194
2195         dma_free_coherent(&pdev->dev, tx_ring->size, tx_ring->desc,
2196                           tx_ring->dma);
2197         tx_ring->desc = NULL;
2198 }
2199
2200 /**
2201  * e1000e_free_rx_resources - Free Rx Resources
2202  * @adapter: board private structure
2203  *
2204  * Free all receive software resources
2205  **/
2206
2207 void e1000e_free_rx_resources(struct e1000_adapter *adapter)
2208 {
2209         struct pci_dev *pdev = adapter->pdev;
2210         struct e1000_ring *rx_ring = adapter->rx_ring;
2211         int i;
2212
2213         e1000_clean_rx_ring(adapter);
2214
2215         for (i = 0; i < rx_ring->count; i++)
2216                 kfree(rx_ring->buffer_info[i].ps_pages);
2217
2218         vfree(rx_ring->buffer_info);
2219         rx_ring->buffer_info = NULL;
2220
2221         dma_free_coherent(&pdev->dev, rx_ring->size, rx_ring->desc,
2222                           rx_ring->dma);
2223         rx_ring->desc = NULL;
2224 }
2225
2226 /**
2227  * e1000_update_itr - update the dynamic ITR value based on statistics
2228  * @adapter: pointer to adapter
2229  * @itr_setting: current adapter->itr
2230  * @packets: the number of packets during this measurement interval
2231  * @bytes: the number of bytes during this measurement interval
2232  *
2233  *      Stores a new ITR value based on packets and byte
2234  *      counts during the last interrupt.  The advantage of per interrupt
2235  *      computation is faster updates and more accurate ITR for the current
2236  *      traffic pattern.  Constants in this function were computed
2237  *      based on theoretical maximum wire speed and thresholds were set based
2238  *      on testing data as well as attempting to minimize response time
2239  *      while increasing bulk throughput.  This functionality is controlled
2240  *      by the InterruptThrottleRate module parameter.
2241  **/
2242 static unsigned int e1000_update_itr(struct e1000_adapter *adapter,
2243                                      u16 itr_setting, int packets,
2244                                      int bytes)
2245 {
2246         unsigned int retval = itr_setting;
2247
2248         if (packets == 0)
2249                 goto update_itr_done;
2250
2251         switch (itr_setting) {
2252         case lowest_latency:
2253                 /* handle TSO and jumbo frames */
2254                 if (bytes/packets > 8000)
2255                         retval = bulk_latency;
2256                 else if ((packets < 5) && (bytes > 512))
2257                         retval = low_latency;
2258                 break;
2259         case low_latency:  /* 50 usec aka 20000 ints/s */
2260                 if (bytes > 10000) {
2261                         /* this if handles the TSO accounting */
2262                         if (bytes/packets > 8000)
2263                                 retval = bulk_latency;
2264                         else if ((packets < 10) || ((bytes/packets) > 1200))
2265                                 retval = bulk_latency;
2266                         else if ((packets > 35))
2267                                 retval = lowest_latency;
2268                 } else if (bytes/packets > 2000) {
2269                         retval = bulk_latency;
2270                 } else if (packets <= 2 && bytes < 512) {
2271                         retval = lowest_latency;
2272                 }
2273                 break;
2274         case bulk_latency: /* 250 usec aka 4000 ints/s */
2275                 if (bytes > 25000) {
2276                         if (packets > 35)
2277                                 retval = low_latency;
2278                 } else if (bytes < 6000) {
2279                         retval = low_latency;
2280                 }
2281                 break;
2282         }
2283
2284 update_itr_done:
2285         return retval;
2286 }
2287
2288 static void e1000_set_itr(struct e1000_adapter *adapter)
2289 {
2290         struct e1000_hw *hw = &adapter->hw;
2291         u16 current_itr;
2292         u32 new_itr = adapter->itr;
2293
2294         /* for non-gigabit speeds, just fix the interrupt rate at 4000 */
2295         if (adapter->link_speed != SPEED_1000) {
2296                 current_itr = 0;
2297                 new_itr = 4000;
2298                 goto set_itr_now;
2299         }
2300
2301         if (adapter->flags2 & FLAG2_DISABLE_AIM) {
2302                 new_itr = 0;
2303                 goto set_itr_now;
2304         }
2305
2306         adapter->tx_itr = e1000_update_itr(adapter,
2307                                     adapter->tx_itr,
2308                                     adapter->total_tx_packets,
2309                                     adapter->total_tx_bytes);
2310         /* conservative mode (itr 3) eliminates the lowest_latency setting */
2311         if (adapter->itr_setting == 3 && adapter->tx_itr == lowest_latency)
2312                 adapter->tx_itr = low_latency;
2313
2314         adapter->rx_itr = e1000_update_itr(adapter,
2315                                     adapter->rx_itr,
2316                                     adapter->total_rx_packets,
2317                                     adapter->total_rx_bytes);
2318         /* conservative mode (itr 3) eliminates the lowest_latency setting */
2319         if (adapter->itr_setting == 3 && adapter->rx_itr == lowest_latency)
2320                 adapter->rx_itr = low_latency;
2321
2322         current_itr = max(adapter->rx_itr, adapter->tx_itr);
2323
2324         switch (current_itr) {
2325         /* counts and packets in update_itr are dependent on these numbers */
2326         case lowest_latency:
2327                 new_itr = 70000;
2328                 break;
2329         case low_latency:
2330                 new_itr = 20000; /* aka hwitr = ~200 */
2331                 break;
2332         case bulk_latency:
2333                 new_itr = 4000;
2334                 break;
2335         default:
2336                 break;
2337         }
2338
2339 set_itr_now:
2340         if (new_itr != adapter->itr) {
2341                 /*
2342                  * this attempts to bias the interrupt rate towards Bulk
2343                  * by adding intermediate steps when interrupt rate is
2344                  * increasing
2345                  */
2346                 new_itr = new_itr > adapter->itr ?
2347                              min(adapter->itr + (new_itr >> 2), new_itr) :
2348                              new_itr;
2349                 adapter->itr = new_itr;
2350                 adapter->rx_ring->itr_val = new_itr;
2351                 if (adapter->msix_entries)
2352                         adapter->rx_ring->set_itr = 1;
2353                 else
2354                         if (new_itr)
2355                                 ew32(ITR, 1000000000 / (new_itr * 256));
2356                         else
2357                                 ew32(ITR, 0);
2358         }
2359 }
2360
2361 /**
2362  * e1000_alloc_queues - Allocate memory for all rings
2363  * @adapter: board private structure to initialize
2364  **/
2365 static int __devinit e1000_alloc_queues(struct e1000_adapter *adapter)
2366 {
2367         adapter->tx_ring = kzalloc(sizeof(struct e1000_ring), GFP_KERNEL);
2368         if (!adapter->tx_ring)
2369                 goto err;
2370
2371         adapter->rx_ring = kzalloc(sizeof(struct e1000_ring), GFP_KERNEL);
2372         if (!adapter->rx_ring)
2373                 goto err;
2374
2375         return 0;
2376 err:
2377         e_err("Unable to allocate memory for queues\n");
2378         kfree(adapter->rx_ring);
2379         kfree(adapter->tx_ring);
2380         return -ENOMEM;
2381 }
2382
2383 /**
2384  * e1000_clean - NAPI Rx polling callback
2385  * @napi: struct associated with this polling callback
2386  * @budget: amount of packets driver is allowed to process this poll
2387  **/
2388 static int e1000_clean(struct napi_struct *napi, int budget)
2389 {
2390         struct e1000_adapter *adapter = container_of(napi, struct e1000_adapter, napi);
2391         struct e1000_hw *hw = &adapter->hw;
2392         struct net_device *poll_dev = adapter->netdev;
2393         int tx_cleaned = 1, work_done = 0;
2394
2395         adapter = netdev_priv(poll_dev);
2396
2397         if (adapter->msix_entries &&
2398             !(adapter->rx_ring->ims_val & adapter->tx_ring->ims_val))
2399                 goto clean_rx;
2400
2401         tx_cleaned = e1000_clean_tx_irq(adapter);
2402
2403 clean_rx:
2404         adapter->clean_rx(adapter, &work_done, budget);
2405
2406         if (!tx_cleaned)
2407                 work_done = budget;
2408
2409         /* If budget not fully consumed, exit the polling mode */
2410         if (work_done < budget) {
2411                 if (adapter->itr_setting & 3)
2412                         e1000_set_itr(adapter);
2413                 napi_complete(napi);
2414                 if (!test_bit(__E1000_DOWN, &adapter->state)) {
2415                         if (adapter->msix_entries)
2416                                 ew32(IMS, adapter->rx_ring->ims_val);
2417                         else
2418                                 e1000_irq_enable(adapter);
2419                 }
2420         }
2421
2422         return work_done;
2423 }
2424
2425 static void e1000_vlan_rx_add_vid(struct net_device *netdev, u16 vid)
2426 {
2427         struct e1000_adapter *adapter = netdev_priv(netdev);
2428         struct e1000_hw *hw = &adapter->hw;
2429         u32 vfta, index;
2430
2431         /* don't update vlan cookie if already programmed */
2432         if ((adapter->hw.mng_cookie.status &
2433              E1000_MNG_DHCP_COOKIE_STATUS_VLAN) &&
2434             (vid == adapter->mng_vlan_id))
2435                 return;
2436
2437         /* add VID to filter table */
2438         if (adapter->flags & FLAG_HAS_HW_VLAN_FILTER) {
2439                 index = (vid >> 5) & 0x7F;
2440                 vfta = E1000_READ_REG_ARRAY(hw, E1000_VFTA, index);
2441                 vfta |= (1 << (vid & 0x1F));
2442                 hw->mac.ops.write_vfta(hw, index, vfta);
2443         }
2444
2445         set_bit(vid, adapter->active_vlans);
2446 }
2447
2448 static void e1000_vlan_rx_kill_vid(struct net_device *netdev, u16 vid)
2449 {
2450         struct e1000_adapter *adapter = netdev_priv(netdev);
2451         struct e1000_hw *hw = &adapter->hw;
2452         u32 vfta, index;
2453
2454         if ((adapter->hw.mng_cookie.status &
2455              E1000_MNG_DHCP_COOKIE_STATUS_VLAN) &&
2456             (vid == adapter->mng_vlan_id)) {
2457                 /* release control to f/w */
2458                 e1000e_release_hw_control(adapter);
2459                 return;
2460         }
2461
2462         /* remove VID from filter table */
2463         if (adapter->flags & FLAG_HAS_HW_VLAN_FILTER) {
2464                 index = (vid >> 5) & 0x7F;
2465                 vfta = E1000_READ_REG_ARRAY(hw, E1000_VFTA, index);
2466                 vfta &= ~(1 << (vid & 0x1F));
2467                 hw->mac.ops.write_vfta(hw, index, vfta);
2468         }
2469
2470         clear_bit(vid, adapter->active_vlans);
2471 }
2472
2473 /**
2474  * e1000e_vlan_filter_disable - helper to disable hw VLAN filtering
2475  * @adapter: board private structure to initialize
2476  **/
2477 static void e1000e_vlan_filter_disable(struct e1000_adapter *adapter)
2478 {
2479         struct net_device *netdev = adapter->netdev;
2480         struct e1000_hw *hw = &adapter->hw;
2481         u32 rctl;
2482
2483         if (adapter->flags & FLAG_HAS_HW_VLAN_FILTER) {
2484                 /* disable VLAN receive filtering */
2485                 rctl = er32(RCTL);
2486                 rctl &= ~(E1000_RCTL_VFE | E1000_RCTL_CFIEN);
2487                 ew32(RCTL, rctl);
2488
2489                 if (adapter->mng_vlan_id != (u16)E1000_MNG_VLAN_NONE) {
2490                         e1000_vlan_rx_kill_vid(netdev, adapter->mng_vlan_id);
2491                         adapter->mng_vlan_id = E1000_MNG_VLAN_NONE;
2492                 }
2493         }
2494 }
2495
2496 /**
2497  * e1000e_vlan_filter_enable - helper to enable HW VLAN filtering
2498  * @adapter: board private structure to initialize
2499  **/
2500 static void e1000e_vlan_filter_enable(struct e1000_adapter *adapter)
2501 {
2502         struct e1000_hw *hw = &adapter->hw;
2503         u32 rctl;
2504
2505         if (adapter->flags & FLAG_HAS_HW_VLAN_FILTER) {
2506                 /* enable VLAN receive filtering */
2507                 rctl = er32(RCTL);
2508                 rctl |= E1000_RCTL_VFE;
2509                 rctl &= ~E1000_RCTL_CFIEN;
2510                 ew32(RCTL, rctl);
2511         }
2512 }
2513
2514 /**
2515  * e1000e_vlan_strip_enable - helper to disable HW VLAN stripping
2516  * @adapter: board private structure to initialize
2517  **/
2518 static void e1000e_vlan_strip_disable(struct e1000_adapter *adapter)
2519 {
2520         struct e1000_hw *hw = &adapter->hw;
2521         u32 ctrl;
2522
2523         /* disable VLAN tag insert/strip */
2524         ctrl = er32(CTRL);
2525         ctrl &= ~E1000_CTRL_VME;
2526         ew32(CTRL, ctrl);
2527 }
2528
2529 /**
2530  * e1000e_vlan_strip_enable - helper to enable HW VLAN stripping
2531  * @adapter: board private structure to initialize
2532  **/
2533 static void e1000e_vlan_strip_enable(struct e1000_adapter *adapter)
2534 {
2535         struct e1000_hw *hw = &adapter->hw;
2536         u32 ctrl;
2537
2538         /* enable VLAN tag insert/strip */
2539         ctrl = er32(CTRL);
2540         ctrl |= E1000_CTRL_VME;
2541         ew32(CTRL, ctrl);
2542 }
2543
2544 static void e1000_update_mng_vlan(struct e1000_adapter *adapter)
2545 {
2546         struct net_device *netdev = adapter->netdev;
2547         u16 vid = adapter->hw.mng_cookie.vlan_id;
2548         u16 old_vid = adapter->mng_vlan_id;
2549
2550         if (adapter->hw.mng_cookie.status &
2551             E1000_MNG_DHCP_COOKIE_STATUS_VLAN) {
2552                 e1000_vlan_rx_add_vid(netdev, vid);
2553                 adapter->mng_vlan_id = vid;
2554         }
2555
2556         if ((old_vid != (u16)E1000_MNG_VLAN_NONE) && (vid != old_vid))
2557                 e1000_vlan_rx_kill_vid(netdev, old_vid);
2558 }
2559
2560 static void e1000_restore_vlan(struct e1000_adapter *adapter)
2561 {
2562         u16 vid;
2563
2564         e1000_vlan_rx_add_vid(adapter->netdev, 0);
2565
2566         for_each_set_bit(vid, adapter->active_vlans, VLAN_N_VID)
2567                 e1000_vlan_rx_add_vid(adapter->netdev, vid);
2568 }
2569
2570 static void e1000_init_manageability_pt(struct e1000_adapter *adapter)
2571 {
2572         struct e1000_hw *hw = &adapter->hw;
2573         u32 manc, manc2h, mdef, i, j;
2574
2575         if (!(adapter->flags & FLAG_MNG_PT_ENABLED))
2576                 return;
2577
2578         manc = er32(MANC);
2579
2580         /*
2581          * enable receiving management packets to the host. this will probably
2582          * generate destination unreachable messages from the host OS, but
2583          * the packets will be handled on SMBUS
2584          */
2585         manc |= E1000_MANC_EN_MNG2HOST;
2586         manc2h = er32(MANC2H);
2587
2588         switch (hw->mac.type) {
2589         default:
2590                 manc2h |= (E1000_MANC2H_PORT_623 | E1000_MANC2H_PORT_664);
2591                 break;
2592         case e1000_82574:
2593         case e1000_82583:
2594                 /*
2595                  * Check if IPMI pass-through decision filter already exists;
2596                  * if so, enable it.
2597                  */
2598                 for (i = 0, j = 0; i < 8; i++) {
2599                         mdef = er32(MDEF(i));
2600
2601                         /* Ignore filters with anything other than IPMI ports */
2602                         if (mdef & ~(E1000_MDEF_PORT_623 | E1000_MDEF_PORT_664))
2603                                 continue;
2604
2605                         /* Enable this decision filter in MANC2H */
2606                         if (mdef)
2607                                 manc2h |= (1 << i);
2608
2609                         j |= mdef;
2610                 }
2611
2612                 if (j == (E1000_MDEF_PORT_623 | E1000_MDEF_PORT_664))
2613                         break;
2614
2615                 /* Create new decision filter in an empty filter */
2616                 for (i = 0, j = 0; i < 8; i++)
2617                         if (er32(MDEF(i)) == 0) {
2618                                 ew32(MDEF(i), (E1000_MDEF_PORT_623 |
2619                                                E1000_MDEF_PORT_664));
2620                                 manc2h |= (1 << 1);
2621                                 j++;
2622                                 break;
2623                         }
2624
2625                 if (!j)
2626                         e_warn("Unable to create IPMI pass-through filter\n");
2627                 break;
2628         }
2629
2630         ew32(MANC2H, manc2h);
2631         ew32(MANC, manc);
2632 }
2633
2634 /**
2635  * e1000_configure_tx - Configure Transmit Unit after Reset
2636  * @adapter: board private structure
2637  *
2638  * Configure the Tx unit of the MAC after a reset.
2639  **/
2640 static void e1000_configure_tx(struct e1000_adapter *adapter)
2641 {
2642         struct e1000_hw *hw = &adapter->hw;
2643         struct e1000_ring *tx_ring = adapter->tx_ring;
2644         u64 tdba;
2645         u32 tdlen, tctl, tipg, tarc;
2646         u32 ipgr1, ipgr2;
2647
2648         /* Setup the HW Tx Head and Tail descriptor pointers */
2649         tdba = tx_ring->dma;
2650         tdlen = tx_ring->count * sizeof(struct e1000_tx_desc);
2651         ew32(TDBAL, (tdba & DMA_BIT_MASK(32)));
2652         ew32(TDBAH, (tdba >> 32));
2653         ew32(TDLEN, tdlen);
2654         ew32(TDH, 0);
2655         ew32(TDT, 0);
2656         tx_ring->head = E1000_TDH;
2657         tx_ring->tail = E1000_TDT;
2658
2659         /* Set the default values for the Tx Inter Packet Gap timer */
2660         tipg = DEFAULT_82543_TIPG_IPGT_COPPER;          /*  8  */
2661         ipgr1 = DEFAULT_82543_TIPG_IPGR1;               /*  8  */
2662         ipgr2 = DEFAULT_82543_TIPG_IPGR2;               /*  6  */
2663
2664         if (adapter->flags & FLAG_TIPG_MEDIUM_FOR_80003ESLAN)
2665                 ipgr2 = DEFAULT_80003ES2LAN_TIPG_IPGR2; /*  7  */
2666
2667         tipg |= ipgr1 << E1000_TIPG_IPGR1_SHIFT;
2668         tipg |= ipgr2 << E1000_TIPG_IPGR2_SHIFT;
2669         ew32(TIPG, tipg);
2670
2671         /* Set the Tx Interrupt Delay register */
2672         ew32(TIDV, adapter->tx_int_delay);
2673         /* Tx irq moderation */
2674         ew32(TADV, adapter->tx_abs_int_delay);
2675
2676         if (adapter->flags2 & FLAG2_DMA_BURST) {
2677                 u32 txdctl = er32(TXDCTL(0));
2678                 txdctl &= ~(E1000_TXDCTL_PTHRESH | E1000_TXDCTL_HTHRESH |
2679                             E1000_TXDCTL_WTHRESH);
2680                 /*
2681                  * set up some performance related parameters to encourage the
2682                  * hardware to use the bus more efficiently in bursts, depends
2683                  * on the tx_int_delay to be enabled,
2684                  * wthresh = 5 ==> burst write a cacheline (64 bytes) at a time
2685                  * hthresh = 1 ==> prefetch when one or more available
2686                  * pthresh = 0x1f ==> prefetch if internal cache 31 or less
2687                  * BEWARE: this seems to work but should be considered first if
2688                  * there are Tx hangs or other Tx related bugs
2689                  */
2690                 txdctl |= E1000_TXDCTL_DMA_BURST_ENABLE;
2691                 ew32(TXDCTL(0), txdctl);
2692                 /* erratum work around: set txdctl the same for both queues */
2693                 ew32(TXDCTL(1), txdctl);
2694         }
2695
2696         /* Program the Transmit Control Register */
2697         tctl = er32(TCTL);
2698         tctl &= ~E1000_TCTL_CT;
2699         tctl |= E1000_TCTL_PSP | E1000_TCTL_RTLC |
2700                 (E1000_COLLISION_THRESHOLD << E1000_CT_SHIFT);
2701
2702         if (adapter->flags & FLAG_TARC_SPEED_MODE_BIT) {
2703                 tarc = er32(TARC(0));
2704                 /*
2705                  * set the speed mode bit, we'll clear it if we're not at
2706                  * gigabit link later
2707                  */
2708 #define SPEED_MODE_BIT (1 << 21)
2709                 tarc |= SPEED_MODE_BIT;
2710                 ew32(TARC(0), tarc);
2711         }
2712
2713         /* errata: program both queues to unweighted RR */
2714         if (adapter->flags & FLAG_TARC_SET_BIT_ZERO) {
2715                 tarc = er32(TARC(0));
2716                 tarc |= 1;
2717                 ew32(TARC(0), tarc);
2718                 tarc = er32(TARC(1));
2719                 tarc |= 1;
2720                 ew32(TARC(1), tarc);
2721         }
2722
2723         /* Setup Transmit Descriptor Settings for eop descriptor */
2724         adapter->txd_cmd = E1000_TXD_CMD_EOP | E1000_TXD_CMD_IFCS;
2725
2726         /* only set IDE if we are delaying interrupts using the timers */
2727         if (adapter->tx_int_delay)
2728                 adapter->txd_cmd |= E1000_TXD_CMD_IDE;
2729
2730         /* enable Report Status bit */
2731         adapter->txd_cmd |= E1000_TXD_CMD_RS;
2732
2733         ew32(TCTL, tctl);
2734
2735         e1000e_config_collision_dist(hw);
2736 }
2737
2738 /**
2739  * e1000_setup_rctl - configure the receive control registers
2740  * @adapter: Board private structure
2741  **/
2742 #define PAGE_USE_COUNT(S) (((S) >> PAGE_SHIFT) + \
2743                            (((S) & (PAGE_SIZE - 1)) ? 1 : 0))
2744 static void e1000_setup_rctl(struct e1000_adapter *adapter)
2745 {
2746         struct e1000_hw *hw = &adapter->hw;
2747         u32 rctl, rfctl;
2748         u32 pages = 0;
2749
2750         /* Workaround Si errata on 82579 - configure jumbo frame flow */
2751         if (hw->mac.type == e1000_pch2lan) {
2752                 s32 ret_val;
2753
2754                 if (adapter->netdev->mtu > ETH_DATA_LEN)
2755                         ret_val = e1000_lv_jumbo_workaround_ich8lan(hw, true);
2756                 else
2757                         ret_val = e1000_lv_jumbo_workaround_ich8lan(hw, false);
2758
2759                 if (ret_val)
2760                         e_dbg("failed to enable jumbo frame workaround mode\n");
2761         }
2762
2763         /* Program MC offset vector base */
2764         rctl = er32(RCTL);
2765         rctl &= ~(3 << E1000_RCTL_MO_SHIFT);
2766         rctl |= E1000_RCTL_EN | E1000_RCTL_BAM |
2767                 E1000_RCTL_LBM_NO | E1000_RCTL_RDMTS_HALF |
2768                 (adapter->hw.mac.mc_filter_type << E1000_RCTL_MO_SHIFT);
2769
2770         /* Do not Store bad packets */
2771         rctl &= ~E1000_RCTL_SBP;
2772
2773         /* Enable Long Packet receive */
2774         if (adapter->netdev->mtu <= ETH_DATA_LEN)
2775                 rctl &= ~E1000_RCTL_LPE;
2776         else
2777                 rctl |= E1000_RCTL_LPE;
2778
2779         /* Some systems expect that the CRC is included in SMBUS traffic. The
2780          * hardware strips the CRC before sending to both SMBUS (BMC) and to
2781          * host memory when this is enabled
2782          */
2783         if (adapter->flags2 & FLAG2_CRC_STRIPPING)
2784                 rctl |= E1000_RCTL_SECRC;
2785
2786         /* Workaround Si errata on 82577 PHY - configure IPG for jumbos */
2787         if ((hw->phy.type == e1000_phy_82577) && (rctl & E1000_RCTL_LPE)) {
2788                 u16 phy_data;
2789
2790                 e1e_rphy(hw, PHY_REG(770, 26), &phy_data);
2791                 phy_data &= 0xfff8;
2792                 phy_data |= (1 << 2);
2793                 e1e_wphy(hw, PHY_REG(770, 26), phy_data);
2794
2795                 e1e_rphy(hw, 22, &phy_data);
2796                 phy_data &= 0x0fff;
2797                 phy_data |= (1 << 14);
2798                 e1e_wphy(hw, 0x10, 0x2823);
2799                 e1e_wphy(hw, 0x11, 0x0003);
2800                 e1e_wphy(hw, 22, phy_data);
2801         }
2802
2803         /* Setup buffer sizes */
2804         rctl &= ~E1000_RCTL_SZ_4096;
2805         rctl |= E1000_RCTL_BSEX;
2806         switch (adapter->rx_buffer_len) {
2807         case 2048:
2808         default:
2809                 rctl |= E1000_RCTL_SZ_2048;
2810                 rctl &= ~E1000_RCTL_BSEX;
2811                 break;
2812         case 4096:
2813                 rctl |= E1000_RCTL_SZ_4096;
2814                 break;
2815         case 8192:
2816                 rctl |= E1000_RCTL_SZ_8192;
2817                 break;
2818         case 16384:
2819                 rctl |= E1000_RCTL_SZ_16384;
2820                 break;
2821         }
2822
2823         /*
2824          * 82571 and greater support packet-split where the protocol
2825          * header is placed in skb->data and the packet data is
2826          * placed in pages hanging off of skb_shinfo(skb)->nr_frags.
2827          * In the case of a non-split, skb->data is linearly filled,
2828          * followed by the page buffers.  Therefore, skb->data is
2829          * sized to hold the largest protocol header.
2830          *
2831          * allocations using alloc_page take too long for regular MTU
2832          * so only enable packet split for jumbo frames
2833          *
2834          * Using pages when the page size is greater than 16k wastes
2835          * a lot of memory, since we allocate 3 pages at all times
2836          * per packet.
2837          */
2838         pages = PAGE_USE_COUNT(adapter->netdev->mtu);
2839         if (!(adapter->flags & FLAG_HAS_ERT) && (pages <= 3) &&
2840             (PAGE_SIZE <= 16384) && (rctl & E1000_RCTL_LPE))
2841                 adapter->rx_ps_pages = pages;
2842         else
2843                 adapter->rx_ps_pages = 0;
2844
2845         if (adapter->rx_ps_pages) {
2846                 u32 psrctl = 0;
2847
2848                 /* Configure extra packet-split registers */
2849                 rfctl = er32(RFCTL);
2850                 rfctl |= E1000_RFCTL_EXTEN;
2851                 /*
2852                  * disable packet split support for IPv6 extension headers,
2853                  * because some malformed IPv6 headers can hang the Rx
2854                  */
2855                 rfctl |= (E1000_RFCTL_IPV6_EX_DIS |
2856                           E1000_RFCTL_NEW_IPV6_EXT_DIS);
2857
2858                 ew32(RFCTL, rfctl);
2859
2860                 /* Enable Packet split descriptors */
2861                 rctl |= E1000_RCTL_DTYP_PS;
2862
2863                 psrctl |= adapter->rx_ps_bsize0 >>
2864                         E1000_PSRCTL_BSIZE0_SHIFT;
2865
2866                 switch (adapter->rx_ps_pages) {
2867                 case 3:
2868                         psrctl |= PAGE_SIZE <<
2869                                 E1000_PSRCTL_BSIZE3_SHIFT;
2870                 case 2:
2871                         psrctl |= PAGE_SIZE <<
2872                                 E1000_PSRCTL_BSIZE2_SHIFT;
2873                 case 1:
2874                         psrctl |= PAGE_SIZE >>
2875                                 E1000_PSRCTL_BSIZE1_SHIFT;
2876                         break;
2877                 }
2878
2879                 ew32(PSRCTL, psrctl);
2880         }
2881
2882         ew32(RCTL, rctl);
2883         /* just started the receive unit, no need to restart */
2884         adapter->flags &= ~FLAG_RX_RESTART_NOW;
2885 }
2886
2887 /**
2888  * e1000_configure_rx - Configure Receive Unit after Reset
2889  * @adapter: board private structure
2890  *
2891  * Configure the Rx unit of the MAC after a reset.
2892  **/
2893 static void e1000_configure_rx(struct e1000_adapter *adapter)
2894 {
2895         struct e1000_hw *hw = &adapter->hw;
2896         struct e1000_ring *rx_ring = adapter->rx_ring;
2897         u64 rdba;
2898         u32 rdlen, rctl, rxcsum, ctrl_ext;
2899
2900         if (adapter->rx_ps_pages) {
2901                 /* this is a 32 byte descriptor */
2902                 rdlen = rx_ring->count *
2903                     sizeof(union e1000_rx_desc_packet_split);
2904                 adapter->clean_rx = e1000_clean_rx_irq_ps;
2905                 adapter->alloc_rx_buf = e1000_alloc_rx_buffers_ps;
2906         } else if (adapter->netdev->mtu > ETH_FRAME_LEN + ETH_FCS_LEN) {
2907                 rdlen = rx_ring->count * sizeof(struct e1000_rx_desc);
2908                 adapter->clean_rx = e1000_clean_jumbo_rx_irq;
2909                 adapter->alloc_rx_buf = e1000_alloc_jumbo_rx_buffers;
2910         } else {
2911                 rdlen = rx_ring->count * sizeof(struct e1000_rx_desc);
2912                 adapter->clean_rx = e1000_clean_rx_irq;
2913                 adapter->alloc_rx_buf = e1000_alloc_rx_buffers;
2914         }
2915
2916         /* disable receives while setting up the descriptors */
2917         rctl = er32(RCTL);
2918         if (!(adapter->flags2 & FLAG2_NO_DISABLE_RX))
2919                 ew32(RCTL, rctl & ~E1000_RCTL_EN);
2920         e1e_flush();
2921         usleep_range(10000, 20000);
2922
2923         if (adapter->flags2 & FLAG2_DMA_BURST) {
2924                 /*
2925                  * set the writeback threshold (only takes effect if the RDTR
2926                  * is set). set GRAN=1 and write back up to 0x4 worth, and
2927                  * enable prefetching of 0x20 Rx descriptors
2928                  * granularity = 01
2929                  * wthresh = 04,
2930                  * hthresh = 04,
2931                  * pthresh = 0x20
2932                  */
2933                 ew32(RXDCTL(0), E1000_RXDCTL_DMA_BURST_ENABLE);
2934                 ew32(RXDCTL(1), E1000_RXDCTL_DMA_BURST_ENABLE);
2935
2936                 /*
2937                  * override the delay timers for enabling bursting, only if
2938                  * the value was not set by the user via module options
2939                  */
2940                 if (adapter->rx_int_delay == DEFAULT_RDTR)
2941                         adapter->rx_int_delay = BURST_RDTR;
2942                 if (adapter->rx_abs_int_delay == DEFAULT_RADV)
2943                         adapter->rx_abs_int_delay = BURST_RADV;
2944         }
2945
2946         /* set the Receive Delay Timer Register */
2947         ew32(RDTR, adapter->rx_int_delay);
2948
2949         /* irq moderation */
2950         ew32(RADV, adapter->rx_abs_int_delay);
2951         if ((adapter->itr_setting != 0) && (adapter->itr != 0))
2952                 ew32(ITR, 1000000000 / (adapter->itr * 256));
2953
2954         ctrl_ext = er32(CTRL_EXT);
2955         /* Auto-Mask interrupts upon ICR access */
2956         ctrl_ext |= E1000_CTRL_EXT_IAME;
2957         ew32(IAM, 0xffffffff);
2958         ew32(CTRL_EXT, ctrl_ext);
2959         e1e_flush();
2960
2961         /*
2962          * Setup the HW Rx Head and Tail Descriptor Pointers and
2963          * the Base and Length of the Rx Descriptor Ring
2964          */
2965         rdba = rx_ring->dma;
2966         ew32(RDBAL, (rdba & DMA_BIT_MASK(32)));
2967         ew32(RDBAH, (rdba >> 32));
2968         ew32(RDLEN, rdlen);
2969         ew32(RDH, 0);
2970         ew32(RDT, 0);
2971         rx_ring->head = E1000_RDH;
2972         rx_ring->tail = E1000_RDT;
2973
2974         /* Enable Receive Checksum Offload for TCP and UDP */
2975         rxcsum = er32(RXCSUM);
2976         if (adapter->flags & FLAG_RX_CSUM_ENABLED) {
2977                 rxcsum |= E1000_RXCSUM_TUOFL;
2978
2979                 /*
2980                  * IPv4 payload checksum for UDP fragments must be
2981                  * used in conjunction with packet-split.
2982                  */
2983                 if (adapter->rx_ps_pages)
2984                         rxcsum |= E1000_RXCSUM_IPPCSE;
2985         } else {
2986                 rxcsum &= ~E1000_RXCSUM_TUOFL;
2987                 /* no need to clear IPPCSE as it defaults to 0 */
2988         }
2989         ew32(RXCSUM, rxcsum);
2990
2991         /*
2992          * Enable early receives on supported devices, only takes effect when
2993          * packet size is equal or larger than the specified value (in 8 byte
2994          * units), e.g. using jumbo frames when setting to E1000_ERT_2048
2995          */
2996         if ((adapter->flags & FLAG_HAS_ERT) ||
2997             (adapter->hw.mac.type == e1000_pch2lan)) {
2998                 if (adapter->netdev->mtu > ETH_DATA_LEN) {
2999                         u32 rxdctl = er32(RXDCTL(0));
3000                         ew32(RXDCTL(0), rxdctl | 0x3);
3001                         if (adapter->flags & FLAG_HAS_ERT)
3002                                 ew32(ERT, E1000_ERT_2048 | (1 << 13));
3003                         /*
3004                          * With jumbo frames and early-receive enabled,
3005                          * excessive C-state transition latencies result in
3006                          * dropped transactions.
3007                          */
3008                         pm_qos_update_request(&adapter->netdev->pm_qos_req, 55);
3009                 } else {
3010                         pm_qos_update_request(&adapter->netdev->pm_qos_req,
3011                                               PM_QOS_DEFAULT_VALUE);
3012                 }
3013         }
3014
3015         /* Enable Receives */
3016         ew32(RCTL, rctl);
3017 }
3018
3019 /**
3020  *  e1000_update_mc_addr_list - Update Multicast addresses
3021  *  @hw: pointer to the HW structure
3022  *  @mc_addr_list: array of multicast addresses to program
3023  *  @mc_addr_count: number of multicast addresses to program
3024  *
3025  *  Updates the Multicast Table Array.
3026  *  The caller must have a packed mc_addr_list of multicast addresses.
3027  **/
3028 static void e1000_update_mc_addr_list(struct e1000_hw *hw, u8 *mc_addr_list,
3029                                       u32 mc_addr_count)
3030 {
3031         hw->mac.ops.update_mc_addr_list(hw, mc_addr_list, mc_addr_count);
3032 }
3033
3034 /**
3035  * e1000_set_multi - Multicast and Promiscuous mode set
3036  * @netdev: network interface device structure
3037  *
3038  * The set_multi entry point is called whenever the multicast address
3039  * list or the network interface flags are updated.  This routine is
3040  * responsible for configuring the hardware for proper multicast,
3041  * promiscuous mode, and all-multi behavior.
3042  **/
3043 static void e1000_set_multi(struct net_device *netdev)
3044 {
3045         struct e1000_adapter *adapter = netdev_priv(netdev);
3046         struct e1000_hw *hw = &adapter->hw;
3047         struct netdev_hw_addr *ha;
3048         u8  *mta_list;
3049         u32 rctl;
3050
3051         /* Check for Promiscuous and All Multicast modes */
3052
3053         rctl = er32(RCTL);
3054
3055         if (netdev->flags & IFF_PROMISC) {
3056                 rctl |= (E1000_RCTL_UPE | E1000_RCTL_MPE);
3057                 rctl &= ~E1000_RCTL_VFE;
3058                 /* Do not hardware filter VLANs in promisc mode */
3059                 e1000e_vlan_filter_disable(adapter);
3060         } else {
3061                 if (netdev->flags & IFF_ALLMULTI) {
3062                         rctl |= E1000_RCTL_MPE;
3063                         rctl &= ~E1000_RCTL_UPE;
3064                 } else {
3065                         rctl &= ~(E1000_RCTL_UPE | E1000_RCTL_MPE);
3066                 }
3067                 e1000e_vlan_filter_enable(adapter);
3068         }
3069
3070         ew32(RCTL, rctl);
3071
3072         if (!netdev_mc_empty(netdev)) {
3073                 int i = 0;
3074
3075                 mta_list = kmalloc(netdev_mc_count(netdev) * 6, GFP_ATOMIC);
3076                 if (!mta_list)
3077                         return;
3078
3079                 /* prepare a packed array of only addresses. */
3080                 netdev_for_each_mc_addr(ha, netdev)
3081                         memcpy(mta_list + (i++ * ETH_ALEN), ha->addr, ETH_ALEN);
3082
3083                 e1000_update_mc_addr_list(hw, mta_list, i);
3084                 kfree(mta_list);
3085         } else {
3086                 /*
3087                  * if we're called from probe, we might not have
3088                  * anything to do here, so clear out the list
3089                  */
3090                 e1000_update_mc_addr_list(hw, NULL, 0);
3091         }
3092
3093         if (netdev->features & NETIF_F_HW_VLAN_RX)
3094                 e1000e_vlan_strip_enable(adapter);
3095         else
3096                 e1000e_vlan_strip_disable(adapter);
3097 }
3098
3099 /**
3100  * e1000_configure - configure the hardware for Rx and Tx
3101  * @adapter: private board structure
3102  **/
3103 static void e1000_configure(struct e1000_adapter *adapter)
3104 {
3105         e1000_set_multi(adapter->netdev);
3106
3107         e1000_restore_vlan(adapter);
3108         e1000_init_manageability_pt(adapter);
3109
3110         e1000_configure_tx(adapter);
3111         e1000_setup_rctl(adapter);
3112         e1000_configure_rx(adapter);
3113         adapter->alloc_rx_buf(adapter, e1000_desc_unused(adapter->rx_ring),
3114                               GFP_KERNEL);
3115 }
3116
3117 /**
3118  * e1000e_power_up_phy - restore link in case the phy was powered down
3119  * @adapter: address of board private structure
3120  *
3121  * The phy may be powered down to save power and turn off link when the
3122  * driver is unloaded and wake on lan is not enabled (among others)
3123  * *** this routine MUST be followed by a call to e1000e_reset ***
3124  **/
3125 void e1000e_power_up_phy(struct e1000_adapter *adapter)
3126 {
3127         if (adapter->hw.phy.ops.power_up)
3128                 adapter->hw.phy.ops.power_up(&adapter->hw);
3129
3130         adapter->hw.mac.ops.setup_link(&adapter->hw);
3131 }
3132
3133 /**
3134  * e1000_power_down_phy - Power down the PHY
3135  *
3136  * Power down the PHY so no link is implied when interface is down.
3137  * The PHY cannot be powered down if management or WoL is active.
3138  */
3139 static void e1000_power_down_phy(struct e1000_adapter *adapter)
3140 {
3141         /* WoL is enabled */
3142         if (adapter->wol)
3143                 return;
3144
3145         if (adapter->hw.phy.ops.power_down)
3146                 adapter->hw.phy.ops.power_down(&adapter->hw);
3147 }
3148
3149 /**
3150  * e1000e_reset - bring the hardware into a known good state
3151  *
3152  * This function boots the hardware and enables some settings that
3153  * require a configuration cycle of the hardware - those cannot be
3154  * set/changed during runtime. After reset the device needs to be
3155  * properly configured for Rx, Tx etc.
3156  */
3157 void e1000e_reset(struct e1000_adapter *adapter)
3158 {
3159         struct e1000_mac_info *mac = &adapter->hw.mac;
3160         struct e1000_fc_info *fc = &adapter->hw.fc;
3161         struct e1000_hw *hw = &adapter->hw;
3162         u32 tx_space, min_tx_space, min_rx_space;
3163         u32 pba = adapter->pba;
3164         u16 hwm;
3165
3166         /* reset Packet Buffer Allocation to default */
3167         ew32(PBA, pba);
3168
3169         if (adapter->max_frame_size > ETH_FRAME_LEN + ETH_FCS_LEN) {
3170                 /*
3171                  * To maintain wire speed transmits, the Tx FIFO should be
3172                  * large enough to accommodate two full transmit packets,
3173                  * rounded up to the next 1KB and expressed in KB.  Likewise,
3174                  * the Rx FIFO should be large enough to accommodate at least
3175                  * one full receive packet and is similarly rounded up and
3176                  * expressed in KB.
3177                  */
3178                 pba = er32(PBA);
3179                 /* upper 16 bits has Tx packet buffer allocation size in KB */
3180                 tx_space = pba >> 16;
3181                 /* lower 16 bits has Rx packet buffer allocation size in KB */
3182                 pba &= 0xffff;
3183                 /*
3184                  * the Tx fifo also stores 16 bytes of information about the Tx
3185                  * but don't include ethernet FCS because hardware appends it
3186                  */
3187                 min_tx_space = (adapter->max_frame_size +
3188                                 sizeof(struct e1000_tx_desc) -
3189                                 ETH_FCS_LEN) * 2;
3190                 min_tx_space = ALIGN(min_tx_space, 1024);
3191                 min_tx_space >>= 10;
3192                 /* software strips receive CRC, so leave room for it */
3193                 min_rx_space = adapter->max_frame_size;
3194                 min_rx_space = ALIGN(min_rx_space, 1024);
3195                 min_rx_space >>= 10;
3196
3197                 /*
3198                  * If current Tx allocation is less than the min Tx FIFO size,
3199                  * and the min Tx FIFO size is less than the current Rx FIFO
3200                  * allocation, take space away from current Rx allocation
3201                  */
3202                 if ((tx_space < min_tx_space) &&
3203                     ((min_tx_space - tx_space) < pba)) {
3204                         pba -= min_tx_space - tx_space;
3205
3206                         /*
3207                          * if short on Rx space, Rx wins and must trump Tx
3208                          * adjustment or use Early Receive if available
3209                          */
3210                         if ((pba < min_rx_space) &&
3211                             (!(adapter->flags & FLAG_HAS_ERT)))
3212                                 /* ERT enabled in e1000_configure_rx */
3213                                 pba = min_rx_space;
3214                 }
3215
3216                 ew32(PBA, pba);
3217         }
3218
3219         /*
3220          * flow control settings
3221          *
3222          * The high water mark must be low enough to fit one full frame
3223          * (or the size used for early receive) above it in the Rx FIFO.
3224          * Set it to the lower of:
3225          * - 90% of the Rx FIFO size, and
3226          * - the full Rx FIFO size minus the early receive size (for parts
3227          *   with ERT support assuming ERT set to E1000_ERT_2048), or
3228          * - the full Rx FIFO size minus one full frame
3229          */
3230         if (adapter->flags & FLAG_DISABLE_FC_PAUSE_TIME)
3231                 fc->pause_time = 0xFFFF;
3232         else
3233                 fc->pause_time = E1000_FC_PAUSE_TIME;
3234         fc->send_xon = 1;
3235         fc->current_mode = fc->requested_mode;
3236
3237         switch (hw->mac.type) {
3238         default:
3239                 if ((adapter->flags & FLAG_HAS_ERT) &&
3240                     (adapter->netdev->mtu > ETH_DATA_LEN))
3241                         hwm = min(((pba << 10) * 9 / 10),
3242                                   ((pba << 10) - (E1000_ERT_2048 << 3)));
3243                 else
3244                         hwm = min(((pba << 10) * 9 / 10),
3245                                   ((pba << 10) - adapter->max_frame_size));
3246
3247                 fc->high_water = hwm & E1000_FCRTH_RTH; /* 8-byte granularity */
3248                 fc->low_water = fc->high_water - 8;
3249                 break;
3250         case e1000_pchlan:
3251                 /*
3252                  * Workaround PCH LOM adapter hangs with certain network
3253                  * loads.  If hangs persist, try disabling Tx flow control.
3254                  */
3255                 if (adapter->netdev->mtu > ETH_DATA_LEN) {
3256                         fc->high_water = 0x3500;
3257                         fc->low_water  = 0x1500;
3258                 } else {
3259                         fc->high_water = 0x5000;
3260                         fc->low_water  = 0x3000;
3261                 }
3262                 fc->refresh_time = 0x1000;
3263                 break;
3264         case e1000_pch2lan:
3265                 fc->high_water = 0x05C20;
3266                 fc->low_water = 0x05048;
3267                 fc->pause_time = 0x0650;
3268                 fc->refresh_time = 0x0400;
3269                 if (adapter->netdev->mtu > ETH_DATA_LEN) {
3270                         pba = 14;
3271                         ew32(PBA, pba);
3272                 }
3273                 break;
3274         }
3275
3276         /*
3277          * Disable Adaptive Interrupt Moderation if 2 full packets cannot
3278          * fit in receive buffer and early-receive not supported.
3279          */
3280         if (adapter->itr_setting & 0x3) {
3281                 if (((adapter->max_frame_size * 2) > (pba << 10)) &&
3282                     !(adapter->flags & FLAG_HAS_ERT)) {
3283                         if (!(adapter->flags2 & FLAG2_DISABLE_AIM)) {
3284                                 dev_info(&adapter->pdev->dev,
3285                                         "Interrupt Throttle Rate turned off\n");
3286                                 adapter->flags2 |= FLAG2_DISABLE_AIM;
3287                                 ew32(ITR, 0);
3288                         }
3289                 } else if (adapter->flags2 & FLAG2_DISABLE_AIM) {
3290                         dev_info(&adapter->pdev->dev,
3291                                  "Interrupt Throttle Rate turned on\n");
3292                         adapter->flags2 &= ~FLAG2_DISABLE_AIM;
3293                         adapter->itr = 20000;
3294                         ew32(ITR, 1000000000 / (adapter->itr * 256));
3295                 }
3296         }
3297
3298         /* Allow time for pending master requests to run */
3299         mac->ops.reset_hw(hw);
3300
3301         /*
3302          * For parts with AMT enabled, let the firmware know
3303          * that the network interface is in control
3304          */
3305         if (adapter->flags & FLAG_HAS_AMT)
3306                 e1000e_get_hw_control(adapter);
3307
3308         ew32(WUC, 0);
3309
3310         if (mac->ops.init_hw(hw))
3311                 e_err("Hardware Error\n");
3312
3313         e1000_update_mng_vlan(adapter);
3314
3315         /* Enable h/w to recognize an 802.1Q VLAN Ethernet packet */
3316         ew32(VET, ETH_P_8021Q);
3317
3318         e1000e_reset_adaptive(hw);
3319
3320         if (!netif_running(adapter->netdev) &&
3321             !test_bit(__E1000_TESTING, &adapter->state)) {
3322                 e1000_power_down_phy(adapter);
3323                 return;
3324         }
3325
3326         e1000_get_phy_info(hw);
3327
3328         if ((adapter->flags & FLAG_HAS_SMART_POWER_DOWN) &&
3329             !(adapter->flags & FLAG_SMART_POWER_DOWN)) {
3330                 u16 phy_data = 0;
3331                 /*
3332                  * speed up time to link by disabling smart power down, ignore
3333                  * the return value of this function because there is nothing
3334                  * different we would do if it failed
3335                  */
3336                 e1e_rphy(hw, IGP02E1000_PHY_POWER_MGMT, &phy_data);
3337                 phy_data &= ~IGP02E1000_PM_SPD;
3338                 e1e_wphy(hw, IGP02E1000_PHY_POWER_MGMT, phy_data);
3339         }
3340 }
3341
3342 int e1000e_up(struct e1000_adapter *adapter)
3343 {
3344         struct e1000_hw *hw = &adapter->hw;
3345
3346         /* hardware has been reset, we need to reload some things */
3347         e1000_configure(adapter);
3348
3349         clear_bit(__E1000_DOWN, &adapter->state);
3350
3351         napi_enable(&adapter->napi);
3352         if (adapter->msix_entries)
3353                 e1000_configure_msix(adapter);
3354         e1000_irq_enable(adapter);
3355
3356         netif_start_queue(adapter->netdev);
3357
3358         /* fire a link change interrupt to start the watchdog */
3359         if (adapter->msix_entries)
3360                 ew32(ICS, E1000_ICS_LSC | E1000_ICR_OTHER);
3361         else
3362                 ew32(ICS, E1000_ICS_LSC);
3363
3364         return 0;
3365 }
3366
3367 static void e1000e_flush_descriptors(struct e1000_adapter *adapter)
3368 {
3369         struct e1000_hw *hw = &adapter->hw;
3370
3371         if (!(adapter->flags2 & FLAG2_DMA_BURST))
3372                 return;
3373
3374         /* flush pending descriptor writebacks to memory */
3375         ew32(TIDV, adapter->tx_int_delay | E1000_TIDV_FPD);
3376         ew32(RDTR, adapter->rx_int_delay | E1000_RDTR_FPD);
3377
3378         /* execute the writes immediately */
3379         e1e_flush();
3380 }
3381
3382 static void e1000e_update_stats(struct e1000_adapter *adapter);
3383
3384 void e1000e_down(struct e1000_adapter *adapter)
3385 {
3386         struct net_device *netdev = adapter->netdev;
3387         struct e1000_hw *hw = &adapter->hw;
3388         u32 tctl, rctl;
3389
3390         /*
3391          * signal that we're down so the interrupt handler does not
3392          * reschedule our watchdog timer
3393          */
3394         set_bit(__E1000_DOWN, &adapter->state);
3395
3396         /* disable receives in the hardware */
3397         rctl = er32(RCTL);
3398         if (!(adapter->flags2 & FLAG2_NO_DISABLE_RX))
3399                 ew32(RCTL, rctl & ~E1000_RCTL_EN);
3400         /* flush and sleep below */
3401
3402         netif_stop_queue(netdev);
3403
3404         /* disable transmits in the hardware */
3405         tctl = er32(TCTL);
3406         tctl &= ~E1000_TCTL_EN;
3407         ew32(TCTL, tctl);
3408
3409         /* flush both disables and wait for them to finish */
3410         e1e_flush();
3411         usleep_range(10000, 20000);
3412
3413         napi_disable(&adapter->napi);
3414         e1000_irq_disable(adapter);
3415
3416         del_timer_sync(&adapter->watchdog_timer);
3417         del_timer_sync(&adapter->phy_info_timer);
3418
3419         netif_carrier_off(netdev);
3420
3421         spin_lock(&adapter->stats64_lock);
3422         e1000e_update_stats(adapter);
3423         spin_unlock(&adapter->stats64_lock);
3424
3425         e1000e_flush_descriptors(adapter);
3426         e1000_clean_tx_ring(adapter);
3427         e1000_clean_rx_ring(adapter);
3428
3429         adapter->link_speed = 0;
3430         adapter->link_duplex = 0;
3431
3432         if (!pci_channel_offline(adapter->pdev))
3433                 e1000e_reset(adapter);
3434
3435         /*
3436          * TODO: for power management, we could drop the link and
3437          * pci_disable_device here.
3438          */
3439 }
3440
3441 void e1000e_reinit_locked(struct e1000_adapter *adapter)
3442 {
3443         might_sleep();
3444         while (test_and_set_bit(__E1000_RESETTING, &adapter->state))
3445                 usleep_range(1000, 2000);
3446         e1000e_down(adapter);
3447         e1000e_up(adapter);
3448         clear_bit(__E1000_RESETTING, &adapter->state);
3449 }
3450
3451 /**
3452  * e1000_sw_init - Initialize general software structures (struct e1000_adapter)
3453  * @adapter: board private structure to initialize
3454  *
3455  * e1000_sw_init initializes the Adapter private data structure.
3456  * Fields are initialized based on PCI device information and
3457  * OS network device settings (MTU size).
3458  **/
3459 static int __devinit e1000_sw_init(struct e1000_adapter *adapter)
3460 {
3461         struct net_device *netdev = adapter->netdev;
3462
3463         adapter->rx_buffer_len = ETH_FRAME_LEN + VLAN_HLEN + ETH_FCS_LEN;
3464         adapter->rx_ps_bsize0 = 128;
3465         adapter->max_frame_size = netdev->mtu + ETH_HLEN + ETH_FCS_LEN;
3466         adapter->min_frame_size = ETH_ZLEN + ETH_FCS_LEN;
3467
3468         spin_lock_init(&adapter->stats64_lock);
3469
3470         e1000e_set_interrupt_capability(adapter);
3471
3472         if (e1000_alloc_queues(adapter))
3473                 return -ENOMEM;
3474
3475         /* Explicitly disable IRQ since the NIC can be in any state. */
3476         e1000_irq_disable(adapter);
3477
3478         set_bit(__E1000_DOWN, &adapter->state);
3479         return 0;
3480 }
3481
3482 /**
3483  * e1000_intr_msi_test - Interrupt Handler
3484  * @irq: interrupt number
3485  * @data: pointer to a network interface device structure
3486  **/
3487 static irqreturn_t e1000_intr_msi_test(int irq, void *data)
3488 {
3489         struct net_device *netdev = data;
3490         struct e1000_adapter *adapter = netdev_priv(netdev);
3491         struct e1000_hw *hw = &adapter->hw;
3492         u32 icr = er32(ICR);
3493
3494         e_dbg("icr is %08X\n", icr);
3495         if (icr & E1000_ICR_RXSEQ) {
3496                 adapter->flags &= ~FLAG_MSI_TEST_FAILED;
3497                 wmb();
3498         }
3499
3500         return IRQ_HANDLED;
3501 }
3502
3503 /**
3504  * e1000_test_msi_interrupt - Returns 0 for successful test
3505  * @adapter: board private struct
3506  *
3507  * code flow taken from tg3.c
3508  **/
3509 static int e1000_test_msi_interrupt(struct e1000_adapter *adapter)
3510 {
3511         struct net_device *netdev = adapter->netdev;
3512         struct e1000_hw *hw = &adapter->hw;
3513         int err;
3514
3515         /* poll_enable hasn't been called yet, so don't need disable */
3516         /* clear any pending events */
3517         er32(ICR);
3518
3519         /* free the real vector and request a test handler */
3520         e1000_free_irq(adapter);
3521         e1000e_reset_interrupt_capability(adapter);
3522
3523         /* Assume that the test fails, if it succeeds then the test
3524          * MSI irq handler will unset this flag */
3525         adapter->flags |= FLAG_MSI_TEST_FAILED;
3526
3527         err = pci_enable_msi(adapter->pdev);
3528         if (err)
3529                 goto msi_test_failed;
3530
3531         err = request_irq(adapter->pdev->irq, e1000_intr_msi_test, 0,
3532                           netdev->name, netdev);
3533         if (err) {
3534                 pci_disable_msi(adapter->pdev);
3535                 goto msi_test_failed;
3536         }
3537
3538         wmb();
3539
3540         e1000_irq_enable(adapter);
3541
3542         /* fire an unusual interrupt on the test handler */
3543         ew32(ICS, E1000_ICS_RXSEQ);
3544         e1e_flush();
3545         msleep(50);
3546
3547         e1000_irq_disable(adapter);
3548
3549         rmb();
3550
3551         if (adapter->flags & FLAG_MSI_TEST_FAILED) {
3552                 adapter->int_mode = E1000E_INT_MODE_LEGACY;
3553                 e_info("MSI interrupt test failed, using legacy interrupt.\n");
3554         } else
3555                 e_dbg("MSI interrupt test succeeded!\n");
3556
3557         free_irq(adapter->pdev->irq, netdev);
3558         pci_disable_msi(adapter->pdev);
3559
3560 msi_test_failed:
3561         e1000e_set_interrupt_capability(adapter);
3562         return e1000_request_irq(adapter);
3563 }
3564
3565 /**
3566  * e1000_test_msi - Returns 0 if MSI test succeeds or INTx mode is restored
3567  * @adapter: board private struct
3568  *
3569  * code flow taken from tg3.c, called with e1000 interrupts disabled.
3570  **/
3571 static int e1000_test_msi(struct e1000_adapter *adapter)
3572 {
3573         int err;
3574         u16 pci_cmd;
3575
3576         if (!(adapter->flags & FLAG_MSI_ENABLED))
3577                 return 0;
3578
3579         /* disable SERR in case the MSI write causes a master abort */
3580         pci_read_config_word(adapter->pdev, PCI_COMMAND, &pci_cmd);
3581         if (pci_cmd & PCI_COMMAND_SERR)
3582                 pci_write_config_word(adapter->pdev, PCI_COMMAND,
3583                                       pci_cmd & ~PCI_COMMAND_SERR);
3584
3585         err = e1000_test_msi_interrupt(adapter);
3586
3587         /* re-enable SERR */
3588         if (pci_cmd & PCI_COMMAND_SERR) {
3589                 pci_read_config_word(adapter->pdev, PCI_COMMAND, &pci_cmd);
3590                 pci_cmd |= PCI_COMMAND_SERR;
3591                 pci_write_config_word(adapter->pdev, PCI_COMMAND, pci_cmd);
3592         }
3593
3594         return err;
3595 }
3596
3597 /**
3598  * e1000_open - Called when a network interface is made active
3599  * @netdev: network interface device structure
3600  *
3601  * Returns 0 on success, negative value on failure
3602  *
3603  * The open entry point is called when a network interface is made
3604  * active by the system (IFF_UP).  At this point all resources needed
3605  * for transmit and receive operations are allocated, the interrupt
3606  * handler is registered with the OS, the watchdog timer is started,
3607  * and the stack is notified that the interface is ready.
3608  **/
3609 static int e1000_open(struct net_device *netdev)
3610 {
3611         struct e1000_adapter *adapter = netdev_priv(netdev);
3612         struct e1000_hw *hw = &adapter->hw;
3613         struct pci_dev *pdev = adapter->pdev;
3614         int err;
3615
3616         /* disallow open during test */
3617         if (test_bit(__E1000_TESTING, &adapter->state))
3618                 return -EBUSY;
3619
3620         pm_runtime_get_sync(&pdev->dev);
3621
3622         netif_carrier_off(netdev);
3623
3624         /* allocate transmit descriptors */
3625         err = e1000e_setup_tx_resources(adapter);
3626         if (err)
3627                 goto err_setup_tx;
3628
3629         /* allocate receive descriptors */
3630         err = e1000e_setup_rx_resources(adapter);
3631         if (err)
3632                 goto err_setup_rx;
3633
3634         /*
3635          * If AMT is enabled, let the firmware know that the network
3636          * interface is now open and reset the part to a known state.
3637          */
3638         if (adapter->flags & FLAG_HAS_AMT) {
3639                 e1000e_get_hw_control(adapter);
3640                 e1000e_reset(adapter);
3641         }
3642
3643         e1000e_power_up_phy(adapter);
3644
3645         adapter->mng_vlan_id = E1000_MNG_VLAN_NONE;
3646         if ((adapter->hw.mng_cookie.status &
3647              E1000_MNG_DHCP_COOKIE_STATUS_VLAN))
3648                 e1000_update_mng_vlan(adapter);
3649
3650         /* DMA latency requirement to workaround early-receive/jumbo issue */
3651         if ((adapter->flags & FLAG_HAS_ERT) ||
3652             (adapter->hw.mac.type == e1000_pch2lan))
3653                 pm_qos_add_request(&adapter->netdev->pm_qos_req,
3654                                    PM_QOS_CPU_DMA_LATENCY,
3655                                    PM_QOS_DEFAULT_VALUE);
3656
3657         /*
3658          * before we allocate an interrupt, we must be ready to handle it.
3659          * Setting DEBUG_SHIRQ in the kernel makes it fire an interrupt
3660          * as soon as we call pci_request_irq, so we have to setup our
3661          * clean_rx handler before we do so.
3662          */
3663         e1000_configure(adapter);
3664
3665         err = e1000_request_irq(adapter);
3666         if (err)
3667                 goto err_req_irq;
3668
3669         /*
3670          * Work around PCIe errata with MSI interrupts causing some chipsets to
3671          * ignore e1000e MSI messages, which means we need to test our MSI
3672          * interrupt now
3673          */
3674         if (adapter->int_mode != E1000E_INT_MODE_LEGACY) {
3675                 err = e1000_test_msi(adapter);
3676                 if (err) {
3677                         e_err("Interrupt allocation failed\n");
3678                         goto err_req_irq;
3679                 }
3680         }
3681
3682         /* From here on the code is the same as e1000e_up() */
3683         clear_bit(__E1000_DOWN, &adapter->state);
3684
3685         napi_enable(&adapter->napi);
3686
3687         e1000_irq_enable(adapter);
3688
3689         netif_start_queue(netdev);
3690
3691         adapter->idle_check = true;
3692         pm_runtime_put(&pdev->dev);
3693
3694         /* fire a link status change interrupt to start the watchdog */
3695         if (adapter->msix_entries)
3696                 ew32(ICS, E1000_ICS_LSC | E1000_ICR_OTHER);
3697         else
3698                 ew32(ICS, E1000_ICS_LSC);
3699
3700         return 0;
3701
3702 err_req_irq:
3703         e1000e_release_hw_control(adapter);
3704         e1000_power_down_phy(adapter);
3705         e1000e_free_rx_resources(adapter);
3706 err_setup_rx:
3707         e1000e_free_tx_resources(adapter);
3708 err_setup_tx:
3709         e1000e_reset(adapter);
3710         pm_runtime_put_sync(&pdev->dev);
3711
3712         return err;
3713 }
3714
3715 /**
3716  * e1000_close - Disables a network interface
3717  * @netdev: network interface device structure
3718  *
3719  * Returns 0, this is not allowed to fail
3720  *
3721  * The close entry point is called when an interface is de-activated
3722  * by the OS.  The hardware is still under the drivers control, but
3723  * needs to be disabled.  A global MAC reset is issued to stop the
3724  * hardware, and all transmit and receive resources are freed.
3725  **/
3726 static int e1000_close(struct net_device *netdev)
3727 {
3728         struct e1000_adapter *adapter = netdev_priv(netdev);
3729         struct pci_dev *pdev = adapter->pdev;
3730
3731         WARN_ON(test_bit(__E1000_RESETTING, &adapter->state));
3732
3733         pm_runtime_get_sync(&pdev->dev);
3734
3735         if (!test_bit(__E1000_DOWN, &adapter->state)) {
3736                 e1000e_down(adapter);
3737                 e1000_free_irq(adapter);
3738         }
3739         e1000_power_down_phy(adapter);
3740
3741         e1000e_free_tx_resources(adapter);
3742         e1000e_free_rx_resources(adapter);
3743
3744         /*
3745          * kill manageability vlan ID if supported, but not if a vlan with
3746          * the same ID is registered on the host OS (let 8021q kill it)
3747          */
3748         if (adapter->hw.mng_cookie.status &
3749             E1000_MNG_DHCP_COOKIE_STATUS_VLAN)
3750                 e1000_vlan_rx_kill_vid(netdev, adapter->mng_vlan_id);
3751
3752         /*
3753          * If AMT is enabled, let the firmware know that the network
3754          * interface is now closed
3755          */
3756         if ((adapter->flags & FLAG_HAS_AMT) &&
3757             !test_bit(__E1000_TESTING, &adapter->state))
3758                 e1000e_release_hw_control(adapter);
3759
3760         if ((adapter->flags & FLAG_HAS_ERT) ||
3761             (adapter->hw.mac.type == e1000_pch2lan))
3762                 pm_qos_remove_request(&adapter->netdev->pm_qos_req);
3763
3764         pm_runtime_put_sync(&pdev->dev);
3765
3766         return 0;
3767 }
3768 /**
3769  * e1000_set_mac - Change the Ethernet Address of the NIC
3770  * @netdev: network interface device structure
3771  * @p: pointer to an address structure
3772  *
3773  * Returns 0 on success, negative on failure
3774  **/
3775 static int e1000_set_mac(struct net_device *netdev, void *p)
3776 {
3777         struct e1000_adapter *adapter = netdev_priv(netdev);
3778         struct sockaddr *addr = p;
3779
3780         if (!is_valid_ether_addr(addr->sa_data))
3781                 return -EADDRNOTAVAIL;
3782
3783         memcpy(netdev->dev_addr, addr->sa_data, netdev->addr_len);
3784         memcpy(adapter->hw.mac.addr, addr->sa_data, netdev->addr_len);
3785
3786         e1000e_rar_set(&adapter->hw, adapter->hw.mac.addr, 0);
3787
3788         if (adapter->flags & FLAG_RESET_OVERWRITES_LAA) {
3789                 /* activate the work around */
3790                 e1000e_set_laa_state_82571(&adapter->hw, 1);
3791
3792                 /*
3793                  * Hold a copy of the LAA in RAR[14] This is done so that
3794                  * between the time RAR[0] gets clobbered  and the time it
3795                  * gets fixed (in e1000_watchdog), the actual LAA is in one
3796                  * of the RARs and no incoming packets directed to this port
3797                  * are dropped. Eventually the LAA will be in RAR[0] and
3798                  * RAR[14]
3799                  */
3800                 e1000e_rar_set(&adapter->hw,
3801                               adapter->hw.mac.addr,
3802                               adapter->hw.mac.rar_entry_count - 1);
3803         }
3804
3805         return 0;
3806 }
3807
3808 /**
3809  * e1000e_update_phy_task - work thread to update phy
3810  * @work: pointer to our work struct
3811  *
3812  * this worker thread exists because we must acquire a
3813  * semaphore to read the phy, which we could msleep while
3814  * waiting for it, and we can't msleep in a timer.
3815  **/
3816 static void e1000e_update_phy_task(struct work_struct *work)
3817 {
3818         struct e1000_adapter *adapter = container_of(work,
3819                                         struct e1000_adapter, update_phy_task);
3820
3821         if (test_bit(__E1000_DOWN, &adapter->state))
3822                 return;
3823
3824         e1000_get_phy_info(&adapter->hw);
3825 }
3826
3827 /*
3828  * Need to wait a few seconds after link up to get diagnostic information from
3829  * the phy
3830  */
3831 static void e1000_update_phy_info(unsigned long data)
3832 {
3833         struct e1000_adapter *adapter = (struct e1000_adapter *) data;
3834
3835         if (test_bit(__E1000_DOWN, &adapter->state))
3836                 return;
3837
3838         schedule_work(&adapter->update_phy_task);
3839 }
3840
3841 /**
3842  * e1000e_update_phy_stats - Update the PHY statistics counters
3843  * @adapter: board private structure
3844  *
3845  * Read/clear the upper 16-bit PHY registers and read/accumulate lower
3846  **/
3847 static void e1000e_update_phy_stats(struct e1000_adapter *adapter)
3848 {
3849         struct e1000_hw *hw = &adapter->hw;
3850         s32 ret_val;
3851         u16 phy_data;
3852
3853         ret_val = hw->phy.ops.acquire(hw);
3854         if (ret_val)
3855                 return;
3856
3857         /*
3858          * A page set is expensive so check if already on desired page.
3859          * If not, set to the page with the PHY status registers.
3860          */
3861         hw->phy.addr = 1;
3862         ret_val = e1000e_read_phy_reg_mdic(hw, IGP01E1000_PHY_PAGE_SELECT,
3863                                            &phy_data);
3864         if (ret_val)
3865                 goto release;
3866         if (phy_data != (HV_STATS_PAGE << IGP_PAGE_SHIFT)) {
3867                 ret_val = hw->phy.ops.set_page(hw,
3868                                                HV_STATS_PAGE << IGP_PAGE_SHIFT);
3869                 if (ret_val)
3870                         goto release;
3871         }
3872
3873         /* Single Collision Count */
3874         hw->phy.ops.read_reg_page(hw, HV_SCC_UPPER, &phy_data);
3875         ret_val = hw->phy.ops.read_reg_page(hw, HV_SCC_LOWER, &phy_data);
3876         if (!ret_val)
3877                 adapter->stats.scc += phy_data;
3878
3879         /* Excessive Collision Count */
3880         hw->phy.ops.read_reg_page(hw, HV_ECOL_UPPER, &phy_data);
3881         ret_val = hw->phy.ops.read_reg_page(hw, HV_ECOL_LOWER, &phy_data);
3882         if (!ret_val)
3883                 adapter->stats.ecol += phy_data;
3884
3885         /* Multiple Collision Count */
3886         hw->phy.ops.read_reg_page(hw, HV_MCC_UPPER, &phy_data);
3887         ret_val = hw->phy.ops.read_reg_page(hw, HV_MCC_LOWER, &phy_data);
3888         if (!ret_val)
3889                 adapter->stats.mcc += phy_data;
3890
3891         /* Late Collision Count */
3892         hw->phy.ops.read_reg_page(hw, HV_LATECOL_UPPER, &phy_data);
3893         ret_val = hw->phy.ops.read_reg_page(hw, HV_LATECOL_LOWER, &phy_data);
3894         if (!ret_val)
3895                 adapter->stats.latecol += phy_data;
3896
3897         /* Collision Count - also used for adaptive IFS */
3898         hw->phy.ops.read_reg_page(hw, HV_COLC_UPPER, &phy_data);
3899         ret_val = hw->phy.ops.read_reg_page(hw, HV_COLC_LOWER, &phy_data);
3900         if (!ret_val)
3901                 hw->mac.collision_delta = phy_data;
3902
3903         /* Defer Count */
3904         hw->phy.ops.read_reg_page(hw, HV_DC_UPPER, &phy_data);
3905         ret_val = hw->phy.ops.read_reg_page(hw, HV_DC_LOWER, &phy_data);
3906         if (!ret_val)
3907                 adapter->stats.dc += phy_data;
3908
3909         /* Transmit with no CRS */
3910         hw->phy.ops.read_reg_page(hw, HV_TNCRS_UPPER, &phy_data);
3911         ret_val = hw->phy.ops.read_reg_page(hw, HV_TNCRS_LOWER, &phy_data);
3912         if (!ret_val)
3913                 adapter->stats.tncrs += phy_data;
3914
3915 release:
3916         hw->phy.ops.release(hw);
3917 }
3918
3919 /**
3920  * e1000e_update_stats - Update the board statistics counters
3921  * @adapter: board private structure
3922  **/
3923 static void e1000e_update_stats(struct e1000_adapter *adapter)
3924 {
3925         struct net_device *netdev = adapter->netdev;
3926         struct e1000_hw *hw = &adapter->hw;
3927         struct pci_dev *pdev = adapter->pdev;
3928
3929         /*
3930          * Prevent stats update while adapter is being reset, or if the pci
3931          * connection is down.
3932          */
3933         if (adapter->link_speed == 0)
3934                 return;
3935         if (pci_channel_offline(pdev))
3936                 return;
3937
3938         adapter->stats.crcerrs += er32(CRCERRS);
3939         adapter->stats.gprc += er32(GPRC);
3940         adapter->stats.gorc += er32(GORCL);
3941         er32(GORCH); /* Clear gorc */
3942         adapter->stats.bprc += er32(BPRC);
3943         adapter->stats.mprc += er32(MPRC);
3944         adapter->stats.roc += er32(ROC);
3945
3946         adapter->stats.mpc += er32(MPC);
3947
3948         /* Half-duplex statistics */
3949         if (adapter->link_duplex == HALF_DUPLEX) {
3950                 if (adapter->flags2 & FLAG2_HAS_PHY_STATS) {
3951                         e1000e_update_phy_stats(adapter);
3952                 } else {
3953                         adapter->stats.scc += er32(SCC);
3954                         adapter->stats.ecol += er32(ECOL);
3955                         adapter->stats.mcc += er32(MCC);
3956                         adapter->stats.latecol += er32(LATECOL);
3957                         adapter->stats.dc += er32(DC);
3958
3959                         hw->mac.collision_delta = er32(COLC);
3960
3961                         if ((hw->mac.type != e1000_82574) &&
3962                             (hw->mac.type != e1000_82583))
3963                                 adapter->stats.tncrs += er32(TNCRS);
3964                 }
3965                 adapter->stats.colc += hw->mac.collision_delta;
3966         }
3967
3968         adapter->stats.xonrxc += er32(XONRXC);
3969         adapter->stats.xontxc += er32(XONTXC);
3970         adapter->stats.xoffrxc += er32(XOFFRXC);
3971         adapter->stats.xofftxc += er32(XOFFTXC);
3972         adapter->stats.gptc += er32(GPTC);
3973         adapter->stats.gotc += er32(GOTCL);
3974         er32(GOTCH); /* Clear gotc */
3975         adapter->stats.rnbc += er32(RNBC);
3976         adapter->stats.ruc += er32(RUC);
3977
3978         adapter->stats.mptc += er32(MPTC);
3979         adapter->stats.bptc += er32(BPTC);
3980
3981         /* used for adaptive IFS */
3982
3983         hw->mac.tx_packet_delta = er32(TPT);
3984         adapter->stats.tpt += hw->mac.tx_packet_delta;
3985
3986         adapter->stats.algnerrc += er32(ALGNERRC);
3987         adapter->stats.rxerrc += er32(RXERRC);
3988         adapter->stats.cexterr += er32(CEXTERR);
3989         adapter->stats.tsctc += er32(TSCTC);
3990         adapter->stats.tsctfc += er32(TSCTFC);
3991
3992         /* Fill out the OS statistics structure */
3993         netdev->stats.multicast = adapter->stats.mprc;
3994         netdev->stats.collisions = adapter->stats.colc;
3995
3996         /* Rx Errors */
3997
3998         /*
3999          * RLEC on some newer hardware can be incorrect so build
4000          * our own version based on RUC and ROC
4001          */
4002         netdev->stats.rx_errors = adapter->stats.rxerrc +
4003                 adapter->stats.crcerrs + adapter->stats.algnerrc +
4004                 adapter->stats.ruc + adapter->stats.roc +
4005                 adapter->stats.cexterr;
4006         netdev->stats.rx_length_errors = adapter->stats.ruc +
4007                                               adapter->stats.roc;
4008         netdev->stats.rx_crc_errors = adapter->stats.crcerrs;
4009         netdev->stats.rx_frame_errors = adapter->stats.algnerrc;
4010         netdev->stats.rx_missed_errors = adapter->stats.mpc;
4011
4012         /* Tx Errors */
4013         netdev->stats.tx_errors = adapter->stats.ecol +
4014                                        adapter->stats.latecol;
4015         netdev->stats.tx_aborted_errors = adapter->stats.ecol;
4016         netdev->stats.tx_window_errors = adapter->stats.latecol;
4017         netdev->stats.tx_carrier_errors = adapter->stats.tncrs;
4018
4019         /* Tx Dropped needs to be maintained elsewhere */
4020
4021         /* Management Stats */
4022         adapter->stats.mgptc += er32(MGTPTC);
4023         adapter->stats.mgprc += er32(MGTPRC);
4024         adapter->stats.mgpdc += er32(MGTPDC);
4025 }
4026
4027 /**
4028  * e1000_phy_read_status - Update the PHY register status snapshot
4029  * @adapter: board private structure
4030  **/
4031 static void e1000_phy_read_status(struct e1000_adapter *adapter)
4032 {
4033         struct e1000_hw *hw = &adapter->hw;
4034         struct e1000_phy_regs *phy = &adapter->phy_regs;
4035
4036         if ((er32(STATUS) & E1000_STATUS_LU) &&
4037             (adapter->hw.phy.media_type == e1000_media_type_copper)) {
4038                 int ret_val;
4039
4040                 ret_val  = e1e_rphy(hw, PHY_CONTROL, &phy->bmcr);
4041                 ret_val |= e1e_rphy(hw, PHY_STATUS, &phy->bmsr);
4042                 ret_val |= e1e_rphy(hw, PHY_AUTONEG_ADV, &phy->advertise);
4043                 ret_val |= e1e_rphy(hw, PHY_LP_ABILITY, &phy->lpa);
4044                 ret_val |= e1e_rphy(hw, PHY_AUTONEG_EXP, &phy->expansion);
4045                 ret_val |= e1e_rphy(hw, PHY_1000T_CTRL, &phy->ctrl1000);
4046                 ret_val |= e1e_rphy(hw, PHY_1000T_STATUS, &phy->stat1000);
4047                 ret_val |= e1e_rphy(hw, PHY_EXT_STATUS, &phy->estatus);
4048                 if (ret_val)
4049                         e_warn("Error reading PHY register\n");
4050         } else {
4051                 /*
4052                  * Do not read PHY registers if link is not up
4053                  * Set values to typical power-on defaults
4054                  */
4055                 phy->bmcr = (BMCR_SPEED1000 | BMCR_ANENABLE | BMCR_FULLDPLX);
4056                 phy->bmsr = (BMSR_100FULL | BMSR_100HALF | BMSR_10FULL |
4057                              BMSR_10HALF | BMSR_ESTATEN | BMSR_ANEGCAPABLE |
4058                              BMSR_ERCAP);
4059                 phy->advertise = (ADVERTISE_PAUSE_ASYM | ADVERTISE_PAUSE_CAP |
4060                                   ADVERTISE_ALL | ADVERTISE_CSMA);
4061                 phy->lpa = 0;
4062                 phy->expansion = EXPANSION_ENABLENPAGE;
4063                 phy->ctrl1000 = ADVERTISE_1000FULL;
4064                 phy->stat1000 = 0;
4065                 phy->estatus = (ESTATUS_1000_TFULL | ESTATUS_1000_THALF);
4066         }
4067 }
4068
4069 static void e1000_print_link_info(struct e1000_adapter *adapter)
4070 {
4071         struct e1000_hw *hw = &adapter->hw;
4072         u32 ctrl = er32(CTRL);
4073
4074         /* Link status message must follow this format for user tools */
4075         printk(KERN_INFO "e1000e: %s NIC Link is Up %d Mbps %s, "
4076                "Flow Control: %s\n",
4077                adapter->netdev->name,
4078                adapter->link_speed,
4079                (adapter->link_duplex == FULL_DUPLEX) ?
4080                "Full Duplex" : "Half Duplex",
4081                ((ctrl & E1000_CTRL_TFCE) && (ctrl & E1000_CTRL_RFCE)) ?
4082                "Rx/Tx" :
4083                ((ctrl & E1000_CTRL_RFCE) ? "Rx" :
4084                 ((ctrl & E1000_CTRL_TFCE) ? "Tx" : "None")));
4085 }
4086
4087 static bool e1000e_has_link(struct e1000_adapter *adapter)
4088 {
4089         struct e1000_hw *hw = &adapter->hw;
4090         bool link_active = 0;
4091         s32 ret_val = 0;
4092
4093         /*
4094          * get_link_status is set on LSC (link status) interrupt or
4095          * Rx sequence error interrupt.  get_link_status will stay
4096          * false until the check_for_link establishes link
4097          * for copper adapters ONLY
4098          */
4099         switch (hw->phy.media_type) {
4100         case e1000_media_type_copper:
4101                 if (hw->mac.get_link_status) {
4102                         ret_val = hw->mac.ops.check_for_link(hw);
4103                         link_active = !hw->mac.get_link_status;
4104                 } else {
4105                         link_active = 1;
4106                 }
4107                 break;
4108         case e1000_media_type_fiber:
4109                 ret_val = hw->mac.ops.check_for_link(hw);
4110                 link_active = !!(er32(STATUS) & E1000_STATUS_LU);
4111                 break;
4112         case e1000_media_type_internal_serdes:
4113                 ret_val = hw->mac.ops.check_for_link(hw);
4114                 link_active = adapter->hw.mac.serdes_has_link;
4115                 break;
4116         default:
4117         case e1000_media_type_unknown:
4118                 break;
4119         }
4120
4121         if ((ret_val == E1000_ERR_PHY) && (hw->phy.type == e1000_phy_igp_3) &&
4122             (er32(CTRL) & E1000_PHY_CTRL_GBE_DISABLE)) {
4123                 /* See e1000_kmrn_lock_loss_workaround_ich8lan() */
4124                 e_info("Gigabit has been disabled, downgrading speed\n");
4125         }
4126
4127         return link_active;
4128 }
4129
4130 static void e1000e_enable_receives(struct e1000_adapter *adapter)
4131 {
4132         /* make sure the receive unit is started */
4133         if ((adapter->flags & FLAG_RX_NEEDS_RESTART) &&
4134             (adapter->flags & FLAG_RX_RESTART_NOW)) {
4135                 struct e1000_hw *hw = &adapter->hw;
4136                 u32 rctl = er32(RCTL);
4137                 ew32(RCTL, rctl | E1000_RCTL_EN);
4138                 adapter->flags &= ~FLAG_RX_RESTART_NOW;
4139         }
4140 }
4141
4142 static void e1000e_check_82574_phy_workaround(struct e1000_adapter *adapter)
4143 {
4144         struct e1000_hw *hw = &adapter->hw;
4145
4146         /*
4147          * With 82574 controllers, PHY needs to be checked periodically
4148          * for hung state and reset, if two calls return true
4149          */
4150         if (e1000_check_phy_82574(hw))
4151                 adapter->phy_hang_count++;
4152         else
4153                 adapter->phy_hang_count = 0;
4154
4155         if (adapter->phy_hang_count > 1) {
4156                 adapter->phy_hang_count = 0;
4157                 schedule_work(&adapter->reset_task);
4158         }
4159 }
4160
4161 /**
4162  * e1000_watchdog - Timer Call-back
4163  * @data: pointer to adapter cast into an unsigned long
4164  **/
4165 static void e1000_watchdog(unsigned long data)
4166 {
4167         struct e1000_adapter *adapter = (struct e1000_adapter *) data;
4168
4169         /* Do the rest outside of interrupt context */
4170         schedule_work(&adapter->watchdog_task);
4171
4172         /* TODO: make this use queue_delayed_work() */
4173 }
4174
4175 static void e1000_watchdog_task(struct work_struct *work)
4176 {
4177         struct e1000_adapter *adapter = container_of(work,
4178                                         struct e1000_adapter, watchdog_task);
4179         struct net_device *netdev = adapter->netdev;
4180         struct e1000_mac_info *mac = &adapter->hw.mac;
4181         struct e1000_phy_info *phy = &adapter->hw.phy;
4182         struct e1000_ring *tx_ring = adapter->tx_ring;
4183         struct e1000_hw *hw = &adapter->hw;
4184         u32 link, tctl;
4185
4186         if (test_bit(__E1000_DOWN, &adapter->state))
4187                 return;
4188
4189         link = e1000e_has_link(adapter);
4190         if ((netif_carrier_ok(netdev)) && link) {
4191                 /* Cancel scheduled suspend requests. */
4192                 pm_runtime_resume(netdev->dev.parent);
4193
4194                 e1000e_enable_receives(adapter);
4195                 goto link_up;
4196         }
4197
4198         if ((e1000e_enable_tx_pkt_filtering(hw)) &&
4199             (adapter->mng_vlan_id != adapter->hw.mng_cookie.vlan_id))
4200                 e1000_update_mng_vlan(adapter);
4201
4202         if (link) {
4203                 if (!netif_carrier_ok(netdev)) {
4204                         bool txb2b = 1;
4205
4206                         /* Cancel scheduled suspend requests. */
4207                         pm_runtime_resume(netdev->dev.parent);
4208
4209                         /* update snapshot of PHY registers on LSC */
4210                         e1000_phy_read_status(adapter);
4211                         mac->ops.get_link_up_info(&adapter->hw,
4212                                                    &adapter->link_speed,
4213                                                    &adapter->link_duplex);
4214                         e1000_print_link_info(adapter);
4215                         /*
4216                          * On supported PHYs, check for duplex mismatch only
4217                          * if link has autonegotiated at 10/100 half
4218                          */
4219                         if ((hw->phy.type == e1000_phy_igp_3 ||
4220                              hw->phy.type == e1000_phy_bm) &&
4221                             (hw->mac.autoneg == true) &&
4222                             (adapter->link_speed == SPEED_10 ||
4223                              adapter->link_speed == SPEED_100) &&
4224                             (adapter->link_duplex == HALF_DUPLEX)) {
4225                                 u16 autoneg_exp;
4226
4227                                 e1e_rphy(hw, PHY_AUTONEG_EXP, &autoneg_exp);
4228
4229                                 if (!(autoneg_exp & NWAY_ER_LP_NWAY_CAPS))
4230                                         e_info("Autonegotiated half duplex but"
4231                                                " link partner cannot autoneg. "
4232                                                " Try forcing full duplex if "
4233                                                "link gets many collisions.\n");
4234                         }
4235
4236                         /* adjust timeout factor according to speed/duplex */
4237                         adapter->tx_timeout_factor = 1;
4238                         switch (adapter->link_speed) {
4239                         case SPEED_10:
4240                                 txb2b = 0;
4241                                 adapter->tx_timeout_factor = 16;
4242                                 break;
4243                         case SPEED_100:
4244                                 txb2b = 0;
4245                                 adapter->tx_timeout_factor = 10;
4246                                 break;
4247                         }
4248
4249                         /*
4250                          * workaround: re-program speed mode bit after
4251                          * link-up event
4252                          */
4253                         if ((adapter->flags & FLAG_TARC_SPEED_MODE_BIT) &&
4254                             !txb2b) {
4255                                 u32 tarc0;
4256                                 tarc0 = er32(TARC(0));
4257                                 tarc0 &= ~SPEED_MODE_BIT;
4258                                 ew32(TARC(0), tarc0);
4259                         }
4260
4261                         /*
4262                          * disable TSO for pcie and 10/100 speeds, to avoid
4263                          * some hardware issues
4264                          */
4265                         if (!(adapter->flags & FLAG_TSO_FORCE)) {
4266                                 switch (adapter->link_speed) {
4267                                 case SPEED_10:
4268                                 case SPEED_100:
4269                                         e_info("10/100 speed: disabling TSO\n");
4270                                         netdev->features &= ~NETIF_F_TSO;
4271                                         netdev->features &= ~NETIF_F_TSO6;
4272                                         break;
4273                                 case SPEED_1000:
4274                                         netdev->features |= NETIF_F_TSO;
4275                                         netdev->features |= NETIF_F_TSO6;
4276                                         break;
4277                                 default:
4278                                         /* oops */
4279                                         break;
4280                                 }
4281                         }
4282
4283                         /*
4284                          * enable transmits in the hardware, need to do this
4285                          * after setting TARC(0)
4286                          */
4287                         tctl = er32(TCTL);
4288                         tctl |= E1000_TCTL_EN;
4289                         ew32(TCTL, tctl);
4290
4291                         /*
4292                          * Perform any post-link-up configuration before
4293                          * reporting link up.
4294                          */
4295                         if (phy->ops.cfg_on_link_up)
4296                                 phy->ops.cfg_on_link_up(hw);
4297
4298                         netif_carrier_on(netdev);
4299
4300                         if (!test_bit(__E1000_DOWN, &adapter->state))
4301                                 mod_timer(&adapter->phy_info_timer,
4302                                           round_jiffies(jiffies + 2 * HZ));
4303                 }
4304         } else {
4305                 if (netif_carrier_ok(netdev)) {
4306                         adapter->link_speed = 0;
4307                         adapter->link_duplex = 0;
4308                         /* Link status message must follow this format */
4309                         printk(KERN_INFO "e1000e: %s NIC Link is Down\n",
4310                                adapter->netdev->name);
4311                         netif_carrier_off(netdev);
4312                         if (!test_bit(__E1000_DOWN, &adapter->state))
4313                                 mod_timer(&adapter->phy_info_timer,
4314                                           round_jiffies(jiffies + 2 * HZ));
4315
4316                         if (adapter->flags & FLAG_RX_NEEDS_RESTART)
4317                                 schedule_work(&adapter->reset_task);
4318                         else
4319                                 pm_schedule_suspend(netdev->dev.parent,
4320                                                         LINK_TIMEOUT);
4321                 }
4322         }
4323
4324 link_up:
4325         spin_lock(&adapter->stats64_lock);
4326         e1000e_update_stats(adapter);
4327
4328         mac->tx_packet_delta = adapter->stats.tpt - adapter->tpt_old;
4329         adapter->tpt_old = adapter->stats.tpt;
4330         mac->collision_delta = adapter->stats.colc - adapter->colc_old;
4331         adapter->colc_old = adapter->stats.colc;
4332
4333         adapter->gorc = adapter->stats.gorc - adapter->gorc_old;
4334         adapter->gorc_old = adapter->stats.gorc;
4335         adapter->gotc = adapter->stats.gotc - adapter->gotc_old;
4336         adapter->gotc_old = adapter->stats.gotc;
4337         spin_unlock(&adapter->stats64_lock);
4338
4339         e1000e_update_adaptive(&adapter->hw);
4340
4341         if (!netif_carrier_ok(netdev) &&
4342             (e1000_desc_unused(tx_ring) + 1 < tx_ring->count)) {
4343                 /*
4344                  * We've lost link, so the controller stops DMA,
4345                  * but we've got queued Tx work that's never going
4346                  * to get done, so reset controller to flush Tx.
4347                  * (Do the reset outside of interrupt context).
4348                  */
4349                 schedule_work(&adapter->reset_task);
4350                 /* return immediately since reset is imminent */
4351                 return;
4352         }
4353
4354         /* Simple mode for Interrupt Throttle Rate (ITR) */
4355         if (adapter->itr_setting == 4) {
4356                 /*
4357                  * Symmetric Tx/Rx gets a reduced ITR=2000;
4358                  * Total asymmetrical Tx or Rx gets ITR=8000;
4359                  * everyone else is between 2000-8000.
4360                  */
4361                 u32 goc = (adapter->gotc + adapter->gorc) / 10000;
4362                 u32 dif = (adapter->gotc > adapter->gorc ?
4363                             adapter->gotc - adapter->gorc :
4364                             adapter->gorc - adapter->gotc) / 10000;
4365                 u32 itr = goc > 0 ? (dif * 6000 / goc + 2000) : 8000;
4366
4367                 ew32(ITR, 1000000000 / (itr * 256));
4368         }
4369
4370         /* Cause software interrupt to ensure Rx ring is cleaned */
4371         if (adapter->msix_entries)
4372                 ew32(ICS, adapter->rx_ring->ims_val);
4373         else
4374                 ew32(ICS, E1000_ICS_RXDMT0);
4375
4376         /* flush pending descriptors to memory before detecting Tx hang */
4377         e1000e_flush_descriptors(adapter);
4378
4379         /* Force detection of hung controller every watchdog period */
4380         adapter->detect_tx_hung = 1;
4381
4382         /*
4383          * With 82571 controllers, LAA may be overwritten due to controller
4384          * reset from the other port. Set the appropriate LAA in RAR[0]
4385          */
4386         if (e1000e_get_laa_state_82571(hw))
4387                 e1000e_rar_set(hw, adapter->hw.mac.addr, 0);
4388
4389         if (adapter->flags2 & FLAG2_CHECK_PHY_HANG)
4390                 e1000e_check_82574_phy_workaround(adapter);
4391
4392         /* Reset the timer */
4393         if (!test_bit(__E1000_DOWN, &adapter->state))
4394                 mod_timer(&adapter->watchdog_timer,
4395                           round_jiffies(jiffies + 2 * HZ));
4396 }
4397
4398 #define E1000_TX_FLAGS_CSUM             0x00000001
4399 #define E1000_TX_FLAGS_VLAN             0x00000002
4400 #define E1000_TX_FLAGS_TSO              0x00000004
4401 #define E1000_TX_FLAGS_IPV4             0x00000008
4402 #define E1000_TX_FLAGS_VLAN_MASK        0xffff0000
4403 #define E1000_TX_FLAGS_VLAN_SHIFT       16
4404
4405 static int e1000_tso(struct e1000_adapter *adapter,
4406                      struct sk_buff *skb)
4407 {
4408         struct e1000_ring *tx_ring = adapter->tx_ring;
4409         struct e1000_context_desc *context_desc;
4410         struct e1000_buffer *buffer_info;
4411         unsigned int i;
4412         u32 cmd_length = 0;
4413         u16 ipcse = 0, tucse, mss;
4414         u8 ipcss, ipcso, tucss, tucso, hdr_len;
4415
4416         if (!skb_is_gso(skb))
4417                 return 0;
4418
4419         if (skb_header_cloned(skb)) {
4420                 int err = pskb_expand_head(skb, 0, 0, GFP_ATOMIC);
4421
4422                 if (err)
4423                         return err;
4424         }
4425
4426         hdr_len = skb_transport_offset(skb) + tcp_hdrlen(skb);
4427         mss = skb_shinfo(skb)->gso_size;
4428         if (skb->protocol == htons(ETH_P_IP)) {
4429                 struct iphdr *iph = ip_hdr(skb);
4430                 iph->tot_len = 0;
4431                 iph->check = 0;
4432                 tcp_hdr(skb)->check = ~csum_tcpudp_magic(iph->saddr, iph->daddr,
4433                                                          0, IPPROTO_TCP, 0);
4434                 cmd_length = E1000_TXD_CMD_IP;
4435                 ipcse = skb_transport_offset(skb) - 1;
4436         } else if (skb_is_gso_v6(skb)) {
4437                 ipv6_hdr(skb)->payload_len = 0;
4438                 tcp_hdr(skb)->check = ~csum_ipv6_magic(&ipv6_hdr(skb)->saddr,
4439                                                        &ipv6_hdr(skb)->daddr,
4440                                                        0, IPPROTO_TCP, 0);
4441                 ipcse = 0;
4442         }
4443         ipcss = skb_network_offset(skb);
4444         ipcso = (void *)&(ip_hdr(skb)->check) - (void *)skb->data;
4445         tucss = skb_transport_offset(skb);
4446         tucso = (void *)&(tcp_hdr(skb)->check) - (void *)skb->data;
4447         tucse = 0;
4448
4449         cmd_length |= (E1000_TXD_CMD_DEXT | E1000_TXD_CMD_TSE |
4450                        E1000_TXD_CMD_TCP | (skb->len - (hdr_len)));
4451
4452         i = tx_ring->next_to_use;
4453         context_desc = E1000_CONTEXT_DESC(*tx_ring, i);
4454         buffer_info = &tx_ring->buffer_info[i];
4455
4456         context_desc->lower_setup.ip_fields.ipcss  = ipcss;
4457         context_desc->lower_setup.ip_fields.ipcso  = ipcso;
4458         context_desc->lower_setup.ip_fields.ipcse  = cpu_to_le16(ipcse);
4459         context_desc->upper_setup.tcp_fields.tucss = tucss;
4460         context_desc->upper_setup.tcp_fields.tucso = tucso;
4461         context_desc->upper_setup.tcp_fields.tucse = cpu_to_le16(tucse);
4462         context_desc->tcp_seg_setup.fields.mss     = cpu_to_le16(mss);
4463         context_desc->tcp_seg_setup.fields.hdr_len = hdr_len;
4464         context_desc->cmd_and_length = cpu_to_le32(cmd_length);
4465
4466         buffer_info->time_stamp = jiffies;
4467         buffer_info->next_to_watch = i;
4468
4469         i++;
4470         if (i == tx_ring->count)
4471                 i = 0;
4472         tx_ring->next_to_use = i;
4473
4474         return 1;
4475 }
4476
4477 static bool e1000_tx_csum(struct e1000_adapter *adapter, struct sk_buff *skb)
4478 {
4479         struct e1000_ring *tx_ring = adapter->tx_ring;
4480         struct e1000_context_desc *context_desc;
4481         struct e1000_buffer *buffer_info;
4482         unsigned int i;
4483         u8 css;
4484         u32 cmd_len = E1000_TXD_CMD_DEXT;
4485         __be16 protocol;
4486
4487         if (skb->ip_summed != CHECKSUM_PARTIAL)
4488                 return 0;
4489
4490         if (skb->protocol == cpu_to_be16(ETH_P_8021Q))
4491                 protocol = vlan_eth_hdr(skb)->h_vlan_encapsulated_proto;
4492         else
4493                 protocol = skb->protocol;
4494
4495         switch (protocol) {
4496         case cpu_to_be16(ETH_P_IP):
4497                 if (ip_hdr(skb)->protocol == IPPROTO_TCP)
4498                         cmd_len |= E1000_TXD_CMD_TCP;
4499                 break;
4500         case cpu_to_be16(ETH_P_IPV6):
4501                 /* XXX not handling all IPV6 headers */
4502                 if (ipv6_hdr(skb)->nexthdr == IPPROTO_TCP)
4503                         cmd_len |= E1000_TXD_CMD_TCP;
4504                 break;
4505         default:
4506                 if (unlikely(net_ratelimit()))
4507                         e_warn("checksum_partial proto=%x!\n",
4508                                be16_to_cpu(protocol));
4509                 break;
4510         }
4511
4512         css = skb_checksum_start_offset(skb);
4513
4514         i = tx_ring->next_to_use;
4515         buffer_info = &tx_ring->buffer_info[i];
4516         context_desc = E1000_CONTEXT_DESC(*tx_ring, i);
4517
4518         context_desc->lower_setup.ip_config = 0;
4519         context_desc->upper_setup.tcp_fields.tucss = css;
4520         context_desc->upper_setup.tcp_fields.tucso =
4521                                 css + skb->csum_offset;
4522         context_desc->upper_setup.tcp_fields.tucse = 0;
4523         context_desc->tcp_seg_setup.data = 0;
4524         context_desc->cmd_and_length = cpu_to_le32(cmd_len);
4525
4526         buffer_info->time_stamp = jiffies;
4527         buffer_info->next_to_watch = i;
4528
4529         i++;
4530         if (i == tx_ring->count)
4531                 i = 0;
4532         tx_ring->next_to_use = i;
4533
4534         return 1;
4535 }
4536
4537 #define E1000_MAX_PER_TXD       8192
4538 #define E1000_MAX_TXD_PWR       12
4539
4540 static int e1000_tx_map(struct e1000_adapter *adapter,
4541                         struct sk_buff *skb, unsigned int first,
4542                         unsigned int max_per_txd, unsigned int nr_frags,
4543                         unsigned int mss)
4544 {
4545         struct e1000_ring *tx_ring = adapter->tx_ring;
4546         struct pci_dev *pdev = adapter->pdev;
4547         struct e1000_buffer *buffer_info;
4548         unsigned int len = skb_headlen(skb);
4549         unsigned int offset = 0, size, count = 0, i;
4550         unsigned int f, bytecount, segs;
4551
4552         i = tx_ring->next_to_use;
4553
4554         while (len) {
4555                 buffer_info = &tx_ring->buffer_info[i];
4556                 size = min(len, max_per_txd);
4557
4558                 buffer_info->length = size;
4559                 buffer_info->time_stamp = jiffies;
4560                 buffer_info->next_to_watch = i;
4561                 buffer_info->dma = dma_map_single(&pdev->dev,
4562                                                   skb->data + offset,
4563                                                   size, DMA_TO_DEVICE);
4564                 buffer_info->mapped_as_page = false;
4565                 if (dma_mapping_error(&pdev->dev, buffer_info->dma))
4566                         goto dma_error;
4567
4568                 len -= size;
4569                 offset += size;
4570                 count++;
4571
4572                 if (len) {
4573                         i++;
4574                         if (i == tx_ring->count)
4575                                 i = 0;
4576                 }
4577         }
4578
4579         for (f = 0; f < nr_frags; f++) {
4580                 struct skb_frag_struct *frag;
4581
4582                 frag = &skb_shinfo(skb)->frags[f];
4583                 len = frag->size;
4584                 offset = frag->page_offset;
4585
4586                 while (len) {
4587                         i++;
4588                         if (i == tx_ring->count)
4589                                 i = 0;
4590
4591                         buffer_info = &tx_ring->buffer_info[i];
4592                         size = min(len, max_per_txd);
4593
4594                         buffer_info->length = size;
4595                         buffer_info->time_stamp = jiffies;
4596                         buffer_info->next_to_watch = i;
4597                         buffer_info->dma = dma_map_page(&pdev->dev, frag->page,
4598                                                         offset, size,
4599                                                         DMA_TO_DEVICE);
4600                         buffer_info->mapped_as_page = true;
4601                         if (dma_mapping_error(&pdev->dev, buffer_info->dma))
4602                                 goto dma_error;
4603
4604                         len -= size;
4605                         offset += size;
4606                         count++;
4607                 }
4608         }
4609
4610         segs = skb_shinfo(skb)->gso_segs ? : 1;
4611         /* multiply data chunks by size of headers */
4612         bytecount = ((segs - 1) * skb_headlen(skb)) + skb->len;
4613
4614         tx_ring->buffer_info[i].skb = skb;
4615         tx_ring->buffer_info[i].segs = segs;
4616         tx_ring->buffer_info[i].bytecount = bytecount;
4617         tx_ring->buffer_info[first].next_to_watch = i;
4618
4619         return count;
4620
4621 dma_error:
4622         dev_err(&pdev->dev, "Tx DMA map failed\n");
4623         buffer_info->dma = 0;
4624         if (count)
4625                 count--;
4626
4627         while (count--) {
4628                 if (i == 0)
4629                         i += tx_ring->count;
4630                 i--;
4631                 buffer_info = &tx_ring->buffer_info[i];
4632                 e1000_put_txbuf(adapter, buffer_info);
4633         }
4634
4635         return 0;
4636 }
4637
4638 static void e1000_tx_queue(struct e1000_adapter *adapter,
4639                            int tx_flags, int count)
4640 {
4641         struct e1000_ring *tx_ring = adapter->tx_ring;
4642         struct e1000_tx_desc *tx_desc = NULL;
4643         struct e1000_buffer *buffer_info;
4644         u32 txd_upper = 0, txd_lower = E1000_TXD_CMD_IFCS;
4645         unsigned int i;
4646
4647         if (tx_flags & E1000_TX_FLAGS_TSO) {
4648                 txd_lower |= E1000_TXD_CMD_DEXT | E1000_TXD_DTYP_D |
4649                              E1000_TXD_CMD_TSE;
4650                 txd_upper |= E1000_TXD_POPTS_TXSM << 8;
4651
4652                 if (tx_flags & E1000_TX_FLAGS_IPV4)
4653                         txd_upper |= E1000_TXD_POPTS_IXSM << 8;
4654         }
4655
4656         if (tx_flags & E1000_TX_FLAGS_CSUM) {
4657                 txd_lower |= E1000_TXD_CMD_DEXT | E1000_TXD_DTYP_D;
4658                 txd_upper |= E1000_TXD_POPTS_TXSM << 8;
4659         }
4660
4661         if (tx_flags & E1000_TX_FLAGS_VLAN) {
4662                 txd_lower |= E1000_TXD_CMD_VLE;
4663                 txd_upper |= (tx_flags & E1000_TX_FLAGS_VLAN_MASK);
4664         }
4665
4666         i = tx_ring->next_to_use;
4667
4668         do {
4669                 buffer_info = &tx_ring->buffer_info[i];
4670                 tx_desc = E1000_TX_DESC(*tx_ring, i);
4671                 tx_desc->buffer_addr = cpu_to_le64(buffer_info->dma);
4672                 tx_desc->lower.data =
4673                         cpu_to_le32(txd_lower | buffer_info->length);
4674                 tx_desc->upper.data = cpu_to_le32(txd_upper);
4675
4676                 i++;
4677                 if (i == tx_ring->count)
4678                         i = 0;
4679         } while (--count > 0);
4680
4681         tx_desc->lower.data |= cpu_to_le32(adapter->txd_cmd);
4682
4683         /*
4684          * Force memory writes to complete before letting h/w
4685          * know there are new descriptors to fetch.  (Only
4686          * applicable for weak-ordered memory model archs,
4687          * such as IA-64).
4688          */
4689         wmb();
4690
4691         tx_ring->next_to_use = i;
4692         writel(i, adapter->hw.hw_addr + tx_ring->tail);
4693         /*
4694          * we need this if more than one processor can write to our tail
4695          * at a time, it synchronizes IO on IA64/Altix systems
4696          */
4697         mmiowb();
4698 }
4699
4700 #define MINIMUM_DHCP_PACKET_SIZE 282
4701 static int e1000_transfer_dhcp_info(struct e1000_adapter *adapter,
4702                                     struct sk_buff *skb)
4703 {
4704         struct e1000_hw *hw =  &adapter->hw;
4705         u16 length, offset;
4706
4707         if (vlan_tx_tag_present(skb)) {
4708                 if (!((vlan_tx_tag_get(skb) == adapter->hw.mng_cookie.vlan_id) &&
4709                     (adapter->hw.mng_cookie.status &
4710                         E1000_MNG_DHCP_COOKIE_STATUS_VLAN)))
4711                         return 0;
4712         }
4713
4714         if (skb->len <= MINIMUM_DHCP_PACKET_SIZE)
4715                 return 0;
4716
4717         if (((struct ethhdr *) skb->data)->h_proto != htons(ETH_P_IP))
4718                 return 0;
4719
4720         {
4721                 const struct iphdr *ip = (struct iphdr *)((u8 *)skb->data+14);
4722                 struct udphdr *udp;
4723
4724                 if (ip->protocol != IPPROTO_UDP)
4725                         return 0;
4726
4727                 udp = (struct udphdr *)((u8 *)ip + (ip->ihl << 2));
4728                 if (ntohs(udp->dest) != 67)
4729                         return 0;
4730
4731                 offset = (u8 *)udp + 8 - skb->data;
4732                 length = skb->len - offset;
4733                 return e1000e_mng_write_dhcp_info(hw, (u8 *)udp + 8, length);
4734         }
4735
4736         return 0;
4737 }
4738
4739 static int __e1000_maybe_stop_tx(struct net_device *netdev, int size)
4740 {
4741         struct e1000_adapter *adapter = netdev_priv(netdev);
4742
4743         netif_stop_queue(netdev);
4744         /*
4745          * Herbert's original patch had:
4746          *  smp_mb__after_netif_stop_queue();
4747          * but since that doesn't exist yet, just open code it.
4748          */
4749         smp_mb();
4750
4751         /*
4752          * We need to check again in a case another CPU has just
4753          * made room available.
4754          */
4755         if (e1000_desc_unused(adapter->tx_ring) < size)
4756                 return -EBUSY;
4757
4758         /* A reprieve! */
4759         netif_start_queue(netdev);
4760         ++adapter->restart_queue;
4761         return 0;
4762 }
4763
4764 static int e1000_maybe_stop_tx(struct net_device *netdev, int size)
4765 {
4766         struct e1000_adapter *adapter = netdev_priv(netdev);
4767
4768         if (e1000_desc_unused(adapter->tx_ring) >= size)
4769                 return 0;
4770         return __e1000_maybe_stop_tx(netdev, size);
4771 }
4772
4773 #define TXD_USE_COUNT(S, X) (((S) >> (X)) + 1 )
4774 static netdev_tx_t e1000_xmit_frame(struct sk_buff *skb,
4775                                     struct net_device *netdev)
4776 {
4777         struct e1000_adapter *adapter = netdev_priv(netdev);
4778         struct e1000_ring *tx_ring = adapter->tx_ring;
4779         unsigned int first;
4780         unsigned int max_per_txd = E1000_MAX_PER_TXD;
4781         unsigned int max_txd_pwr = E1000_MAX_TXD_PWR;
4782         unsigned int tx_flags = 0;
4783         unsigned int len = skb_headlen(skb);
4784         unsigned int nr_frags;
4785         unsigned int mss;
4786         int count = 0;
4787         int tso;
4788         unsigned int f;
4789
4790         if (test_bit(__E1000_DOWN, &adapter->state)) {
4791                 dev_kfree_skb_any(skb);
4792                 return NETDEV_TX_OK;
4793         }
4794
4795         if (skb->len <= 0) {
4796                 dev_kfree_skb_any(skb);
4797                 return NETDEV_TX_OK;
4798         }
4799
4800         mss = skb_shinfo(skb)->gso_size;
4801         /*
4802          * The controller does a simple calculation to
4803          * make sure there is enough room in the FIFO before
4804          * initiating the DMA for each buffer.  The calc is:
4805          * 4 = ceil(buffer len/mss).  To make sure we don't
4806          * overrun the FIFO, adjust the max buffer len if mss
4807          * drops.
4808          */
4809         if (mss) {
4810                 u8 hdr_len;
4811                 max_per_txd = min(mss << 2, max_per_txd);
4812                 max_txd_pwr = fls(max_per_txd) - 1;
4813
4814                 /*
4815                  * TSO Workaround for 82571/2/3 Controllers -- if skb->data
4816                  * points to just header, pull a few bytes of payload from
4817                  * frags into skb->data
4818                  */
4819                 hdr_len = skb_transport_offset(skb) + tcp_hdrlen(skb);
4820                 /*
4821                  * we do this workaround for ES2LAN, but it is un-necessary,
4822                  * avoiding it could save a lot of cycles
4823                  */
4824                 if (skb->data_len && (hdr_len == len)) {
4825                         unsigned int pull_size;
4826
4827                         pull_size = min((unsigned int)4, skb->data_len);
4828                         if (!__pskb_pull_tail(skb, pull_size)) {
4829                                 e_err("__pskb_pull_tail failed.\n");
4830                                 dev_kfree_skb_any(skb);
4831                                 return NETDEV_TX_OK;
4832                         }
4833                         len = skb_headlen(skb);
4834                 }
4835         }
4836
4837         /* reserve a descriptor for the offload context */
4838         if ((mss) || (skb->ip_summed == CHECKSUM_PARTIAL))
4839                 count++;
4840         count++;
4841
4842         count += TXD_USE_COUNT(len, max_txd_pwr);
4843
4844         nr_frags = skb_shinfo(skb)->nr_frags;
4845         for (f = 0; f < nr_frags; f++)
4846                 count += TXD_USE_COUNT(skb_shinfo(skb)->frags[f].size,
4847                                        max_txd_pwr);
4848
4849         if (adapter->hw.mac.tx_pkt_filtering)
4850                 e1000_transfer_dhcp_info(adapter, skb);
4851
4852         /*
4853          * need: count + 2 desc gap to keep tail from touching
4854          * head, otherwise try next time
4855          */
4856         if (e1000_maybe_stop_tx(netdev, count + 2))
4857                 return NETDEV_TX_BUSY;
4858
4859         if (vlan_tx_tag_present(skb)) {
4860                 tx_flags |= E1000_TX_FLAGS_VLAN;
4861                 tx_flags |= (vlan_tx_tag_get(skb) << E1000_TX_FLAGS_VLAN_SHIFT);
4862         }
4863
4864         first = tx_ring->next_to_use;
4865
4866         tso = e1000_tso(adapter, skb);
4867         if (tso < 0) {
4868                 dev_kfree_skb_any(skb);
4869                 return NETDEV_TX_OK;
4870         }
4871
4872         if (tso)
4873                 tx_flags |= E1000_TX_FLAGS_TSO;
4874         else if (e1000_tx_csum(adapter, skb))
4875                 tx_flags |= E1000_TX_FLAGS_CSUM;
4876
4877         /*
4878          * Old method was to assume IPv4 packet by default if TSO was enabled.
4879          * 82571 hardware supports TSO capabilities for IPv6 as well...
4880          * no longer assume, we must.
4881          */
4882         if (skb->protocol == htons(ETH_P_IP))
4883                 tx_flags |= E1000_TX_FLAGS_IPV4;
4884
4885         /* if count is 0 then mapping error has occurred */
4886         count = e1000_tx_map(adapter, skb, first, max_per_txd, nr_frags, mss);
4887         if (count) {
4888                 e1000_tx_queue(adapter, tx_flags, count);
4889                 /* Make sure there is space in the ring for the next send. */
4890                 e1000_maybe_stop_tx(netdev, MAX_SKB_FRAGS + 2);
4891
4892         } else {
4893                 dev_kfree_skb_any(skb);
4894                 tx_ring->buffer_info[first].time_stamp = 0;
4895                 tx_ring->next_to_use = first;
4896         }
4897
4898         return NETDEV_TX_OK;
4899 }
4900
4901 /**
4902  * e1000_tx_timeout - Respond to a Tx Hang
4903  * @netdev: network interface device structure
4904  **/
4905 static void e1000_tx_timeout(struct net_device *netdev)
4906 {
4907         struct e1000_adapter *adapter = netdev_priv(netdev);
4908
4909         /* Do the reset outside of interrupt context */
4910         adapter->tx_timeout_count++;
4911         schedule_work(&adapter->reset_task);
4912 }
4913
4914 static void e1000_reset_task(struct work_struct *work)
4915 {
4916         struct e1000_adapter *adapter;
4917         adapter = container_of(work, struct e1000_adapter, reset_task);
4918
4919         /* don't run the task if already down */
4920         if (test_bit(__E1000_DOWN, &adapter->state))
4921                 return;
4922
4923         if (!((adapter->flags & FLAG_RX_NEEDS_RESTART) &&
4924               (adapter->flags & FLAG_RX_RESTART_NOW))) {
4925                 e1000e_dump(adapter);
4926                 e_err("Reset adapter\n");
4927         }
4928         e1000e_reinit_locked(adapter);
4929 }
4930
4931 /**
4932  * e1000_get_stats64 - Get System Network Statistics
4933  * @netdev: network interface device structure
4934  * @stats: rtnl_link_stats64 pointer
4935  *
4936  * Returns the address of the device statistics structure.
4937  **/
4938 struct rtnl_link_stats64 *e1000e_get_stats64(struct net_device *netdev,
4939                                              struct rtnl_link_stats64 *stats)
4940 {
4941         struct e1000_adapter *adapter = netdev_priv(netdev);
4942
4943         memset(stats, 0, sizeof(struct rtnl_link_stats64));
4944         spin_lock(&adapter->stats64_lock);
4945         e1000e_update_stats(adapter);
4946         /* Fill out the OS statistics structure */
4947         stats->rx_bytes = adapter->stats.gorc;
4948         stats->rx_packets = adapter->stats.gprc;
4949         stats->tx_bytes = adapter->stats.gotc;
4950         stats->tx_packets = adapter->stats.gptc;
4951         stats->multicast = adapter->stats.mprc;
4952         stats->collisions = adapter->stats.colc;
4953
4954         /* Rx Errors */
4955
4956         /*
4957          * RLEC on some newer hardware can be incorrect so build
4958          * our own version based on RUC and ROC
4959          */
4960         stats->rx_errors = adapter->stats.rxerrc +
4961                 adapter->stats.crcerrs + adapter->stats.algnerrc +
4962                 adapter->stats.ruc + adapter->stats.roc +
4963                 adapter->stats.cexterr;
4964         stats->rx_length_errors = adapter->stats.ruc +
4965                                               adapter->stats.roc;
4966         stats->rx_crc_errors = adapter->stats.crcerrs;
4967         stats->rx_frame_errors = adapter->stats.algnerrc;
4968         stats->rx_missed_errors = adapter->stats.mpc;
4969
4970         /* Tx Errors */
4971         stats->tx_errors = adapter->stats.ecol +
4972                                        adapter->stats.latecol;
4973         stats->tx_aborted_errors = adapter->stats.ecol;
4974         stats->tx_window_errors = adapter->stats.latecol;
4975         stats->tx_carrier_errors = adapter->stats.tncrs;
4976
4977         /* Tx Dropped needs to be maintained elsewhere */
4978
4979         spin_unlock(&adapter->stats64_lock);
4980         return stats;
4981 }
4982
4983 /**
4984  * e1000_change_mtu - Change the Maximum Transfer Unit
4985  * @netdev: network interface device structure
4986  * @new_mtu: new value for maximum frame size
4987  *
4988  * Returns 0 on success, negative on failure
4989  **/
4990 static int e1000_change_mtu(struct net_device *netdev, int new_mtu)
4991 {
4992         struct e1000_adapter *adapter = netdev_priv(netdev);
4993         int max_frame = new_mtu + ETH_HLEN + ETH_FCS_LEN;
4994
4995         /* Jumbo frame support */
4996         if ((max_frame > ETH_FRAME_LEN + ETH_FCS_LEN) &&
4997             !(adapter->flags & FLAG_HAS_JUMBO_FRAMES)) {
4998                 e_err("Jumbo Frames not supported.\n");
4999                 return -EINVAL;
5000         }
5001
5002         /* Supported frame sizes */
5003         if ((new_mtu < ETH_ZLEN + ETH_FCS_LEN + VLAN_HLEN) ||
5004             (max_frame > adapter->max_hw_frame_size)) {
5005                 e_err("Unsupported MTU setting\n");
5006                 return -EINVAL;
5007         }
5008
5009         /* Jumbo frame workaround on 82579 requires CRC be stripped */
5010         if ((adapter->hw.mac.type == e1000_pch2lan) &&
5011             !(adapter->flags2 & FLAG2_CRC_STRIPPING) &&
5012             (new_mtu > ETH_DATA_LEN)) {
5013                 e_err("Jumbo Frames not supported on 82579 when CRC "
5014                       "stripping is disabled.\n");
5015                 return -EINVAL;
5016         }
5017
5018         /* 82573 Errata 17 */
5019         if (((adapter->hw.mac.type == e1000_82573) ||
5020              (adapter->hw.mac.type == e1000_82574)) &&
5021             (max_frame > ETH_FRAME_LEN + ETH_FCS_LEN)) {
5022                 adapter->flags2 |= FLAG2_DISABLE_ASPM_L1;
5023                 e1000e_disable_aspm(adapter->pdev, PCIE_LINK_STATE_L1);
5024         }
5025
5026         while (test_and_set_bit(__E1000_RESETTING, &adapter->state))
5027                 usleep_range(1000, 2000);
5028         /* e1000e_down -> e1000e_reset dependent on max_frame_size & mtu */
5029         adapter->max_frame_size = max_frame;
5030         e_info("changing MTU from %d to %d\n", netdev->mtu, new_mtu);
5031         netdev->mtu = new_mtu;
5032         if (netif_running(netdev))
5033                 e1000e_down(adapter);
5034
5035         /*
5036          * NOTE: netdev_alloc_skb reserves 16 bytes, and typically NET_IP_ALIGN
5037          * means we reserve 2 more, this pushes us to allocate from the next
5038          * larger slab size.
5039          * i.e. RXBUFFER_2048 --> size-4096 slab
5040          * However with the new *_jumbo_rx* routines, jumbo receives will use
5041          * fragmented skbs
5042          */
5043
5044         if (max_frame <= 2048)
5045                 adapter->rx_buffer_len = 2048;
5046         else
5047                 adapter->rx_buffer_len = 4096;
5048
5049         /* adjust allocation if LPE protects us, and we aren't using SBP */
5050         if ((max_frame == ETH_FRAME_LEN + ETH_FCS_LEN) ||
5051              (max_frame == ETH_FRAME_LEN + VLAN_HLEN + ETH_FCS_LEN))
5052                 adapter->rx_buffer_len = ETH_FRAME_LEN + VLAN_HLEN
5053                                          + ETH_FCS_LEN;
5054
5055         if (netif_running(netdev))
5056                 e1000e_up(adapter);
5057         else
5058                 e1000e_reset(adapter);
5059
5060         clear_bit(__E1000_RESETTING, &adapter->state);
5061
5062         return 0;
5063 }
5064
5065 static int e1000_mii_ioctl(struct net_device *netdev, struct ifreq *ifr,
5066                            int cmd)
5067 {
5068         struct e1000_adapter *adapter = netdev_priv(netdev);
5069         struct mii_ioctl_data *data = if_mii(ifr);
5070
5071         if (adapter->hw.phy.media_type != e1000_media_type_copper)
5072                 return -EOPNOTSUPP;
5073
5074         switch (cmd) {
5075         case SIOCGMIIPHY:
5076                 data->phy_id = adapter->hw.phy.addr;
5077                 break;
5078         case SIOCGMIIREG:
5079                 e1000_phy_read_status(adapter);
5080
5081                 switch (data->reg_num & 0x1F) {
5082                 case MII_BMCR:
5083                         data->val_out = adapter->phy_regs.bmcr;
5084                         break;
5085                 case MII_BMSR:
5086                         data->val_out = adapter->phy_regs.bmsr;
5087                         break;
5088                 case MII_PHYSID1:
5089                         data->val_out = (adapter->hw.phy.id >> 16);
5090                         break;
5091                 case MII_PHYSID2:
5092                         data->val_out = (adapter->hw.phy.id & 0xFFFF);
5093                         break;
5094                 case MII_ADVERTISE:
5095                         data->val_out = adapter->phy_regs.advertise;
5096                         break;
5097                 case MII_LPA:
5098                         data->val_out = adapter->phy_regs.lpa;
5099                         break;
5100                 case MII_EXPANSION:
5101                         data->val_out = adapter->phy_regs.expansion;
5102                         break;
5103                 case MII_CTRL1000:
5104                         data->val_out = adapter->phy_regs.ctrl1000;
5105                         break;
5106                 case MII_STAT1000:
5107                         data->val_out = adapter->phy_regs.stat1000;
5108                         break;
5109                 case MII_ESTATUS:
5110                         data->val_out = adapter->phy_regs.estatus;
5111                         break;
5112                 default:
5113                         return -EIO;
5114                 }
5115                 break;
5116         case SIOCSMIIREG:
5117         default:
5118                 return -EOPNOTSUPP;
5119         }
5120         return 0;
5121 }
5122
5123 static int e1000_ioctl(struct net_device *netdev, struct ifreq *ifr, int cmd)
5124 {
5125         switch (cmd) {
5126         case SIOCGMIIPHY:
5127         case SIOCGMIIREG:
5128         case SIOCSMIIREG:
5129                 return e1000_mii_ioctl(netdev, ifr, cmd);
5130         default:
5131                 return -EOPNOTSUPP;
5132         }
5133 }
5134
5135 static int e1000_init_phy_wakeup(struct e1000_adapter *adapter, u32 wufc)
5136 {
5137         struct e1000_hw *hw = &adapter->hw;
5138         u32 i, mac_reg;
5139         u16 phy_reg, wuc_enable;
5140         int retval = 0;
5141
5142         /* copy MAC RARs to PHY RARs */
5143         e1000_copy_rx_addrs_to_phy_ich8lan(hw);
5144
5145         retval = hw->phy.ops.acquire(hw);
5146         if (retval) {
5147                 e_err("Could not acquire PHY\n");
5148                 return retval;
5149         }
5150
5151         /* Enable access to wakeup registers on and set page to BM_WUC_PAGE */
5152         retval = e1000_enable_phy_wakeup_reg_access_bm(hw, &wuc_enable);
5153         if (retval)
5154                 goto out;
5155
5156         /* copy MAC MTA to PHY MTA - only needed for pchlan */
5157         for (i = 0; i < adapter->hw.mac.mta_reg_count; i++) {
5158                 mac_reg = E1000_READ_REG_ARRAY(hw, E1000_MTA, i);
5159                 hw->phy.ops.write_reg_page(hw, BM_MTA(i),
5160                                            (u16)(mac_reg & 0xFFFF));
5161                 hw->phy.ops.write_reg_page(hw, BM_MTA(i) + 1,
5162                                            (u16)((mac_reg >> 16) & 0xFFFF));
5163         }
5164
5165         /* configure PHY Rx Control register */
5166         hw->phy.ops.read_reg_page(&adapter->hw, BM_RCTL, &phy_reg);
5167         mac_reg = er32(RCTL);
5168         if (mac_reg & E1000_RCTL_UPE)
5169                 phy_reg |= BM_RCTL_UPE;
5170         if (mac_reg & E1000_RCTL_MPE)
5171                 phy_reg |= BM_RCTL_MPE;
5172         phy_reg &= ~(BM_RCTL_MO_MASK);
5173         if (mac_reg & E1000_RCTL_MO_3)
5174                 phy_reg |= (((mac_reg & E1000_RCTL_MO_3) >> E1000_RCTL_MO_SHIFT)
5175                                 << BM_RCTL_MO_SHIFT);
5176         if (mac_reg & E1000_RCTL_BAM)
5177                 phy_reg |= BM_RCTL_BAM;
5178         if (mac_reg & E1000_RCTL_PMCF)
5179                 phy_reg |= BM_RCTL_PMCF;
5180         mac_reg = er32(CTRL);
5181         if (mac_reg & E1000_CTRL_RFCE)
5182                 phy_reg |= BM_RCTL_RFCE;
5183         hw->phy.ops.write_reg_page(&adapter->hw, BM_RCTL, phy_reg);
5184
5185         /* enable PHY wakeup in MAC register */
5186         ew32(WUFC, wufc);
5187         ew32(WUC, E1000_WUC_PHY_WAKE | E1000_WUC_PME_EN);
5188
5189         /* configure and enable PHY wakeup in PHY registers */
5190         hw->phy.ops.write_reg_page(&adapter->hw, BM_WUFC, wufc);
5191         hw->phy.ops.write_reg_page(&adapter->hw, BM_WUC, E1000_WUC_PME_EN);
5192
5193         /* activate PHY wakeup */
5194         wuc_enable |= BM_WUC_ENABLE_BIT | BM_WUC_HOST_WU_BIT;
5195         retval = e1000_disable_phy_wakeup_reg_access_bm(hw, &wuc_enable);
5196         if (retval)
5197                 e_err("Could not set PHY Host Wakeup bit\n");
5198 out:
5199         hw->phy.ops.release(hw);
5200
5201         return retval;
5202 }
5203
5204 static int __e1000_shutdown(struct pci_dev *pdev, bool *enable_wake,
5205                             bool runtime)
5206 {
5207         struct net_device *netdev = pci_get_drvdata(pdev);
5208         struct e1000_adapter *adapter = netdev_priv(netdev);
5209         struct e1000_hw *hw = &adapter->hw;
5210         u32 ctrl, ctrl_ext, rctl, status;
5211         /* Runtime suspend should only enable wakeup for link changes */
5212         u32 wufc = runtime ? E1000_WUFC_LNKC : adapter->wol;
5213         int retval = 0;
5214
5215         netif_device_detach(netdev);
5216
5217         if (netif_running(netdev)) {
5218                 WARN_ON(test_bit(__E1000_RESETTING, &adapter->state));
5219                 e1000e_down(adapter);
5220                 e1000_free_irq(adapter);
5221         }
5222         e1000e_reset_interrupt_capability(adapter);
5223
5224         retval = pci_save_state(pdev);
5225         if (retval)
5226                 return retval;
5227
5228         status = er32(STATUS);
5229         if (status & E1000_STATUS_LU)
5230                 wufc &= ~E1000_WUFC_LNKC;
5231
5232         if (wufc) {
5233                 e1000_setup_rctl(adapter);
5234                 e1000_set_multi(netdev);
5235
5236                 /* turn on all-multi mode if wake on multicast is enabled */
5237                 if (wufc & E1000_WUFC_MC) {
5238                         rctl = er32(RCTL);
5239                         rctl |= E1000_RCTL_MPE;
5240                         ew32(RCTL, rctl);
5241                 }
5242
5243                 ctrl = er32(CTRL);
5244                 /* advertise wake from D3Cold */
5245                 #define E1000_CTRL_ADVD3WUC 0x00100000
5246                 /* phy power management enable */
5247                 #define E1000_CTRL_EN_PHY_PWR_MGMT 0x00200000
5248                 ctrl |= E1000_CTRL_ADVD3WUC;
5249                 if (!(adapter->flags2 & FLAG2_HAS_PHY_WAKEUP))
5250                         ctrl |= E1000_CTRL_EN_PHY_PWR_MGMT;
5251                 ew32(CTRL, ctrl);
5252
5253                 if (adapter->hw.phy.media_type == e1000_media_type_fiber ||
5254                     adapter->hw.phy.media_type ==
5255                     e1000_media_type_internal_serdes) {
5256                         /* keep the laser running in D3 */
5257                         ctrl_ext = er32(CTRL_EXT);
5258                         ctrl_ext |= E1000_CTRL_EXT_SDP3_DATA;
5259                         ew32(CTRL_EXT, ctrl_ext);
5260                 }
5261
5262                 if (adapter->flags & FLAG_IS_ICH)
5263                         e1000_suspend_workarounds_ich8lan(&adapter->hw);
5264
5265                 /* Allow time for pending master requests to run */
5266                 e1000e_disable_pcie_master(&adapter->hw);
5267
5268                 if (adapter->flags2 & FLAG2_HAS_PHY_WAKEUP) {
5269                         /* enable wakeup by the PHY */
5270                         retval = e1000_init_phy_wakeup(adapter, wufc);
5271                         if (retval)
5272                                 return retval;
5273                 } else {
5274                         /* enable wakeup by the MAC */
5275                         ew32(WUFC, wufc);
5276                         ew32(WUC, E1000_WUC_PME_EN);
5277                 }
5278         } else {
5279                 ew32(WUC, 0);
5280                 ew32(WUFC, 0);
5281         }
5282
5283         *enable_wake = !!wufc;
5284
5285         /* make sure adapter isn't asleep if manageability is enabled */
5286         if ((adapter->flags & FLAG_MNG_PT_ENABLED) ||
5287             (hw->mac.ops.check_mng_mode(hw)))
5288                 *enable_wake = true;
5289
5290         if (adapter->hw.phy.type == e1000_phy_igp_3)
5291                 e1000e_igp3_phy_powerdown_workaround_ich8lan(&adapter->hw);
5292
5293         /*
5294          * Release control of h/w to f/w.  If f/w is AMT enabled, this
5295          * would have already happened in close and is redundant.
5296          */
5297         e1000e_release_hw_control(adapter);
5298
5299         pci_disable_device(pdev);
5300
5301         return 0;
5302 }
5303
5304 static void e1000_power_off(struct pci_dev *pdev, bool sleep, bool wake)
5305 {
5306         if (sleep && wake) {
5307                 pci_prepare_to_sleep(pdev);
5308                 return;
5309         }
5310
5311         pci_wake_from_d3(pdev, wake);
5312         pci_set_power_state(pdev, PCI_D3hot);
5313 }
5314
5315 static void e1000_complete_shutdown(struct pci_dev *pdev, bool sleep,
5316                                     bool wake)
5317 {
5318         struct net_device *netdev = pci_get_drvdata(pdev);
5319         struct e1000_adapter *adapter = netdev_priv(netdev);
5320
5321         /*
5322          * The pci-e switch on some quad port adapters will report a
5323          * correctable error when the MAC transitions from D0 to D3.  To
5324          * prevent this we need to mask off the correctable errors on the
5325          * downstream port of the pci-e switch.
5326          */
5327         if (adapter->flags & FLAG_IS_QUAD_PORT) {
5328                 struct pci_dev *us_dev = pdev->bus->self;
5329                 int pos = pci_pcie_cap(us_dev);
5330                 u16 devctl;
5331
5332                 pci_read_config_word(us_dev, pos + PCI_EXP_DEVCTL, &devctl);
5333                 pci_write_config_word(us_dev, pos + PCI_EXP_DEVCTL,
5334                                       (devctl & ~PCI_EXP_DEVCTL_CERE));
5335
5336                 e1000_power_off(pdev, sleep, wake);
5337
5338                 pci_write_config_word(us_dev, pos + PCI_EXP_DEVCTL, devctl);
5339         } else {
5340                 e1000_power_off(pdev, sleep, wake);
5341         }
5342 }
5343
5344 #ifdef CONFIG_PCIEASPM
5345 static void __e1000e_disable_aspm(struct pci_dev *pdev, u16 state)
5346 {
5347         pci_disable_link_state_locked(pdev, state);
5348 }
5349 #else
5350 static void __e1000e_disable_aspm(struct pci_dev *pdev, u16 state)
5351 {
5352         int pos;
5353         u16 reg16;
5354
5355         /*
5356          * Both device and parent should have the same ASPM setting.
5357          * Disable ASPM in downstream component first and then upstream.
5358          */
5359         pos = pci_pcie_cap(pdev);
5360         pci_read_config_word(pdev, pos + PCI_EXP_LNKCTL, &reg16);
5361         reg16 &= ~state;
5362         pci_write_config_word(pdev, pos + PCI_EXP_LNKCTL, reg16);
5363
5364         if (!pdev->bus->self)
5365                 return;
5366
5367         pos = pci_pcie_cap(pdev->bus->self);
5368         pci_read_config_word(pdev->bus->self, pos + PCI_EXP_LNKCTL, &reg16);
5369         reg16 &= ~state;
5370         pci_write_config_word(pdev->bus->self, pos + PCI_EXP_LNKCTL, reg16);
5371 }
5372 #endif
5373 static void e1000e_disable_aspm(struct pci_dev *pdev, u16 state)
5374 {
5375         dev_info(&pdev->dev, "Disabling ASPM %s %s\n",
5376                  (state & PCIE_LINK_STATE_L0S) ? "L0s" : "",
5377                  (state & PCIE_LINK_STATE_L1) ? "L1" : "");
5378
5379         __e1000e_disable_aspm(pdev, state);
5380 }
5381
5382 #ifdef CONFIG_PM
5383 static bool e1000e_pm_ready(struct e1000_adapter *adapter)
5384 {
5385         return !!adapter->tx_ring->buffer_info;
5386 }
5387
5388 static int __e1000_resume(struct pci_dev *pdev)
5389 {
5390         struct net_device *netdev = pci_get_drvdata(pdev);
5391         struct e1000_adapter *adapter = netdev_priv(netdev);
5392         struct e1000_hw *hw = &adapter->hw;
5393         u16 aspm_disable_flag = 0;
5394         u32 err;
5395
5396         if (adapter->flags2 & FLAG2_DISABLE_ASPM_L0S)
5397                 aspm_disable_flag = PCIE_LINK_STATE_L0S;
5398         if (adapter->flags2 & FLAG2_DISABLE_ASPM_L1)
5399                 aspm_disable_flag |= PCIE_LINK_STATE_L1;
5400         if (aspm_disable_flag)
5401                 e1000e_disable_aspm(pdev, aspm_disable_flag);
5402
5403         pci_set_power_state(pdev, PCI_D0);
5404         pci_restore_state(pdev);
5405         pci_save_state(pdev);
5406
5407         e1000e_set_interrupt_capability(adapter);
5408         if (netif_running(netdev)) {
5409                 err = e1000_request_irq(adapter);
5410                 if (err)
5411                         return err;
5412         }
5413
5414         if (hw->mac.type == e1000_pch2lan)
5415                 e1000_resume_workarounds_pchlan(&adapter->hw);
5416
5417         e1000e_power_up_phy(adapter);
5418
5419         /* report the system wakeup cause from S3/S4 */
5420         if (adapter->flags2 & FLAG2_HAS_PHY_WAKEUP) {
5421                 u16 phy_data;
5422
5423                 e1e_rphy(&adapter->hw, BM_WUS, &phy_data);
5424                 if (phy_data) {
5425                         e_info("PHY Wakeup cause - %s\n",
5426                                 phy_data & E1000_WUS_EX ? "Unicast Packet" :
5427                                 phy_data & E1000_WUS_MC ? "Multicast Packet" :
5428                                 phy_data & E1000_WUS_BC ? "Broadcast Packet" :
5429                                 phy_data & E1000_WUS_MAG ? "Magic Packet" :
5430                                 phy_data & E1000_WUS_LNKC ? "Link Status "
5431                                 " Change" : "other");
5432                 }
5433                 e1e_wphy(&adapter->hw, BM_WUS, ~0);
5434         } else {
5435                 u32 wus = er32(WUS);
5436                 if (wus) {
5437                         e_info("MAC Wakeup cause - %s\n",
5438                                 wus & E1000_WUS_EX ? "Unicast Packet" :
5439                                 wus & E1000_WUS_MC ? "Multicast Packet" :
5440                                 wus & E1000_WUS_BC ? "Broadcast Packet" :
5441                                 wus & E1000_WUS_MAG ? "Magic Packet" :
5442                                 wus & E1000_WUS_LNKC ? "Link Status Change" :
5443                                 "other");
5444                 }
5445                 ew32(WUS, ~0);
5446         }
5447
5448         e1000e_reset(adapter);
5449
5450         e1000_init_manageability_pt(adapter);
5451
5452         if (netif_running(netdev))
5453                 e1000e_up(adapter);
5454
5455         netif_device_attach(netdev);
5456
5457         /*
5458          * If the controller has AMT, do not set DRV_LOAD until the interface
5459          * is up.  For all other cases, let the f/w know that the h/w is now
5460          * under the control of the driver.
5461          */
5462         if (!(adapter->flags & FLAG_HAS_AMT))
5463                 e1000e_get_hw_control(adapter);
5464
5465         return 0;
5466 }
5467
5468 #ifdef CONFIG_PM_SLEEP
5469 static int e1000_suspend(struct device *dev)
5470 {
5471         struct pci_dev *pdev = to_pci_dev(dev);
5472         int retval;
5473         bool wake;
5474
5475         retval = __e1000_shutdown(pdev, &wake, false);
5476         if (!retval)
5477                 e1000_complete_shutdown(pdev, true, wake);
5478
5479         return retval;
5480 }
5481
5482 static int e1000_resume(struct device *dev)
5483 {
5484         struct pci_dev *pdev = to_pci_dev(dev);
5485         struct net_device *netdev = pci_get_drvdata(pdev);
5486         struct e1000_adapter *adapter = netdev_priv(netdev);
5487
5488         if (e1000e_pm_ready(adapter))
5489                 adapter->idle_check = true;
5490
5491         return __e1000_resume(pdev);
5492 }
5493 #endif /* CONFIG_PM_SLEEP */
5494
5495 #ifdef CONFIG_PM_RUNTIME
5496 static int e1000_runtime_suspend(struct device *dev)
5497 {
5498         struct pci_dev *pdev = to_pci_dev(dev);
5499         struct net_device *netdev = pci_get_drvdata(pdev);
5500         struct e1000_adapter *adapter = netdev_priv(netdev);
5501
5502         if (e1000e_pm_ready(adapter)) {
5503                 bool wake;
5504
5505                 __e1000_shutdown(pdev, &wake, true);
5506         }
5507
5508         return 0;
5509 }
5510
5511 static int e1000_idle(struct device *dev)
5512 {
5513         struct pci_dev *pdev = to_pci_dev(dev);
5514         struct net_device *netdev = pci_get_drvdata(pdev);
5515         struct e1000_adapter *adapter = netdev_priv(netdev);
5516
5517         if (!e1000e_pm_ready(adapter))
5518                 return 0;
5519
5520         if (adapter->idle_check) {
5521                 adapter->idle_check = false;
5522                 if (!e1000e_has_link(adapter))
5523                         pm_schedule_suspend(dev, MSEC_PER_SEC);
5524         }
5525
5526         return -EBUSY;
5527 }
5528
5529 static int e1000_runtime_resume(struct device *dev)
5530 {
5531         struct pci_dev *pdev = to_pci_dev(dev);
5532         struct net_device *netdev = pci_get_drvdata(pdev);
5533         struct e1000_adapter *adapter = netdev_priv(netdev);
5534
5535         if (!e1000e_pm_ready(adapter))
5536                 return 0;
5537
5538         adapter->idle_check = !dev->power.runtime_auto;
5539         return __e1000_resume(pdev);
5540 }
5541 #endif /* CONFIG_PM_RUNTIME */
5542 #endif /* CONFIG_PM */
5543
5544 static void e1000_shutdown(struct pci_dev *pdev)
5545 {
5546         bool wake = false;
5547
5548         __e1000_shutdown(pdev, &wake, false);
5549
5550         if (system_state == SYSTEM_POWER_OFF)
5551                 e1000_complete_shutdown(pdev, false, wake);
5552 }
5553
5554 #ifdef CONFIG_NET_POLL_CONTROLLER
5555
5556 static irqreturn_t e1000_intr_msix(int irq, void *data)
5557 {
5558         struct net_device *netdev = data;
5559         struct e1000_adapter *adapter = netdev_priv(netdev);
5560
5561         if (adapter->msix_entries) {
5562                 int vector, msix_irq;
5563
5564                 vector = 0;
5565                 msix_irq = adapter->msix_entries[vector].vector;
5566                 disable_irq(msix_irq);
5567                 e1000_intr_msix_rx(msix_irq, netdev);
5568                 enable_irq(msix_irq);
5569
5570                 vector++;
5571                 msix_irq = adapter->msix_entries[vector].vector;
5572                 disable_irq(msix_irq);
5573                 e1000_intr_msix_tx(msix_irq, netdev);
5574                 enable_irq(msix_irq);
5575
5576                 vector++;
5577                 msix_irq = adapter->msix_entries[vector].vector;
5578                 disable_irq(msix_irq);
5579                 e1000_msix_other(msix_irq, netdev);
5580                 enable_irq(msix_irq);
5581         }
5582
5583         return IRQ_HANDLED;
5584 }
5585
5586 /*
5587  * Polling 'interrupt' - used by things like netconsole to send skbs
5588  * without having to re-enable interrupts. It's not called while
5589  * the interrupt routine is executing.
5590  */
5591 static void e1000_netpoll(struct net_device *netdev)
5592 {
5593         struct e1000_adapter *adapter = netdev_priv(netdev);
5594
5595         switch (adapter->int_mode) {
5596         case E1000E_INT_MODE_MSIX:
5597                 e1000_intr_msix(adapter->pdev->irq, netdev);
5598                 break;
5599         case E1000E_INT_MODE_MSI:
5600                 disable_irq(adapter->pdev->irq);
5601                 e1000_intr_msi(adapter->pdev->irq, netdev);
5602                 enable_irq(adapter->pdev->irq);
5603                 break;
5604         default: /* E1000E_INT_MODE_LEGACY */
5605                 disable_irq(adapter->pdev->irq);
5606                 e1000_intr(adapter->pdev->irq, netdev);
5607                 enable_irq(adapter->pdev->irq);
5608                 break;
5609         }
5610 }
5611 #endif
5612
5613 /**
5614  * e1000_io_error_detected - called when PCI error is detected
5615  * @pdev: Pointer to PCI device
5616  * @state: The current pci connection state
5617  *
5618  * This function is called after a PCI bus error affecting
5619  * this device has been detected.
5620  */
5621 static pci_ers_result_t e1000_io_error_detected(struct pci_dev *pdev,
5622                                                 pci_channel_state_t state)
5623 {
5624         struct net_device *netdev = pci_get_drvdata(pdev);
5625         struct e1000_adapter *adapter = netdev_priv(netdev);
5626
5627         netif_device_detach(netdev);
5628
5629         if (state == pci_channel_io_perm_failure)
5630                 return PCI_ERS_RESULT_DISCONNECT;
5631
5632         if (netif_running(netdev))
5633                 e1000e_down(adapter);
5634         pci_disable_device(pdev);
5635
5636         /* Request a slot slot reset. */
5637         return PCI_ERS_RESULT_NEED_RESET;
5638 }
5639
5640 /**
5641  * e1000_io_slot_reset - called after the pci bus has been reset.
5642  * @pdev: Pointer to PCI device
5643  *
5644  * Restart the card from scratch, as if from a cold-boot. Implementation
5645  * resembles the first-half of the e1000_resume routine.
5646  */
5647 static pci_ers_result_t e1000_io_slot_reset(struct pci_dev *pdev)
5648 {
5649         struct net_device *netdev = pci_get_drvdata(pdev);
5650         struct e1000_adapter *adapter = netdev_priv(netdev);
5651         struct e1000_hw *hw = &adapter->hw;
5652         u16 aspm_disable_flag = 0;
5653         int err;
5654         pci_ers_result_t result;
5655
5656         if (adapter->flags2 & FLAG2_DISABLE_ASPM_L0S)
5657                 aspm_disable_flag = PCIE_LINK_STATE_L0S;
5658         if (adapter->flags2 & FLAG2_DISABLE_ASPM_L1)
5659                 aspm_disable_flag |= PCIE_LINK_STATE_L1;
5660         if (aspm_disable_flag)
5661                 e1000e_disable_aspm(pdev, aspm_disable_flag);
5662
5663         err = pci_enable_device_mem(pdev);
5664         if (err) {
5665                 dev_err(&pdev->dev,
5666                         "Cannot re-enable PCI device after reset.\n");
5667                 result = PCI_ERS_RESULT_DISCONNECT;
5668         } else {
5669                 pci_set_master(pdev);
5670                 pdev->state_saved = true;
5671                 pci_restore_state(pdev);
5672
5673                 pci_enable_wake(pdev, PCI_D3hot, 0);
5674                 pci_enable_wake(pdev, PCI_D3cold, 0);
5675
5676                 e1000e_reset(adapter);
5677                 ew32(WUS, ~0);
5678                 result = PCI_ERS_RESULT_RECOVERED;
5679         }
5680
5681         pci_cleanup_aer_uncorrect_error_status(pdev);
5682
5683         return result;
5684 }
5685
5686 /**
5687  * e1000_io_resume - called when traffic can start flowing again.
5688  * @pdev: Pointer to PCI device
5689  *
5690  * This callback is called when the error recovery driver tells us that
5691  * its OK to resume normal operation. Implementation resembles the
5692  * second-half of the e1000_resume routine.
5693  */
5694 static void e1000_io_resume(struct pci_dev *pdev)
5695 {
5696         struct net_device *netdev = pci_get_drvdata(pdev);
5697         struct e1000_adapter *adapter = netdev_priv(netdev);
5698
5699         e1000_init_manageability_pt(adapter);
5700
5701         if (netif_running(netdev)) {
5702                 if (e1000e_up(adapter)) {
5703                         dev_err(&pdev->dev,
5704                                 "can't bring device back up after reset\n");
5705                         return;
5706                 }
5707         }
5708
5709         netif_device_attach(netdev);
5710
5711         /*
5712          * If the controller has AMT, do not set DRV_LOAD until the interface
5713          * is up.  For all other cases, let the f/w know that the h/w is now
5714          * under the control of the driver.
5715          */
5716         if (!(adapter->flags & FLAG_HAS_AMT))
5717                 e1000e_get_hw_control(adapter);
5718
5719 }
5720
5721 static void e1000_print_device_info(struct e1000_adapter *adapter)
5722 {
5723         struct e1000_hw *hw = &adapter->hw;
5724         struct net_device *netdev = adapter->netdev;
5725         u32 ret_val;
5726         u8 pba_str[E1000_PBANUM_LENGTH];
5727
5728         /* print bus type/speed/width info */
5729         e_info("(PCI Express:2.5GT/s:%s) %pM\n",
5730                /* bus width */
5731                ((hw->bus.width == e1000_bus_width_pcie_x4) ? "Width x4" :
5732                 "Width x1"),
5733                /* MAC address */
5734                netdev->dev_addr);
5735         e_info("Intel(R) PRO/%s Network Connection\n",
5736                (hw->phy.type == e1000_phy_ife) ? "10/100" : "1000");
5737         ret_val = e1000_read_pba_string_generic(hw, pba_str,
5738                                                 E1000_PBANUM_LENGTH);
5739         if (ret_val)
5740                 strncpy((char *)pba_str, "Unknown", sizeof(pba_str) - 1);
5741         e_info("MAC: %d, PHY: %d, PBA No: %s\n",
5742                hw->mac.type, hw->phy.type, pba_str);
5743 }
5744
5745 static void e1000_eeprom_checks(struct e1000_adapter *adapter)
5746 {
5747         struct e1000_hw *hw = &adapter->hw;
5748         int ret_val;
5749         u16 buf = 0;
5750
5751         if (hw->mac.type != e1000_82573)
5752                 return;
5753
5754         ret_val = e1000_read_nvm(hw, NVM_INIT_CONTROL2_REG, 1, &buf);
5755         if (!ret_val && (!(le16_to_cpu(buf) & (1 << 0)))) {
5756                 /* Deep Smart Power Down (DSPD) */
5757                 dev_warn(&adapter->pdev->dev,
5758                          "Warning: detected DSPD enabled in EEPROM\n");
5759         }
5760 }
5761
5762 static const struct net_device_ops e1000e_netdev_ops = {
5763         .ndo_open               = e1000_open,
5764         .ndo_stop               = e1000_close,
5765         .ndo_start_xmit         = e1000_xmit_frame,
5766         .ndo_get_stats64        = e1000e_get_stats64,
5767         .ndo_set_multicast_list = e1000_set_multi,
5768         .ndo_set_mac_address    = e1000_set_mac,
5769         .ndo_change_mtu         = e1000_change_mtu,
5770         .ndo_do_ioctl           = e1000_ioctl,
5771         .ndo_tx_timeout         = e1000_tx_timeout,
5772         .ndo_validate_addr      = eth_validate_addr,
5773
5774         .ndo_vlan_rx_add_vid    = e1000_vlan_rx_add_vid,
5775         .ndo_vlan_rx_kill_vid   = e1000_vlan_rx_kill_vid,
5776 #ifdef CONFIG_NET_POLL_CONTROLLER
5777         .ndo_poll_controller    = e1000_netpoll,
5778 #endif
5779 };
5780
5781 /**
5782  * e1000_probe - Device Initialization Routine
5783  * @pdev: PCI device information struct
5784  * @ent: entry in e1000_pci_tbl
5785  *
5786  * Returns 0 on success, negative on failure
5787  *
5788  * e1000_probe initializes an adapter identified by a pci_dev structure.
5789  * The OS initialization, configuring of the adapter private structure,
5790  * and a hardware reset occur.
5791  **/
5792 static int __devinit e1000_probe(struct pci_dev *pdev,
5793                                  const struct pci_device_id *ent)
5794 {
5795         struct net_device *netdev;
5796         struct e1000_adapter *adapter;
5797         struct e1000_hw *hw;
5798         const struct e1000_info *ei = e1000_info_tbl[ent->driver_data];
5799         resource_size_t mmio_start, mmio_len;
5800         resource_size_t flash_start, flash_len;
5801
5802         static int cards_found;
5803         u16 aspm_disable_flag = 0;
5804         int i, err, pci_using_dac;
5805         u16 eeprom_data = 0;
5806         u16 eeprom_apme_mask = E1000_EEPROM_APME;
5807
5808         if (ei->flags2 & FLAG2_DISABLE_ASPM_L0S)
5809                 aspm_disable_flag = PCIE_LINK_STATE_L0S;
5810         if (ei->flags2 & FLAG2_DISABLE_ASPM_L1)
5811                 aspm_disable_flag |= PCIE_LINK_STATE_L1;
5812         if (aspm_disable_flag)
5813                 e1000e_disable_aspm(pdev, aspm_disable_flag);
5814
5815         err = pci_enable_device_mem(pdev);
5816         if (err)
5817                 return err;
5818
5819         pci_using_dac = 0;
5820         err = dma_set_mask(&pdev->dev, DMA_BIT_MASK(64));
5821         if (!err) {
5822                 err = dma_set_coherent_mask(&pdev->dev, DMA_BIT_MASK(64));
5823                 if (!err)
5824                         pci_using_dac = 1;
5825         } else {
5826                 err = dma_set_mask(&pdev->dev, DMA_BIT_MASK(32));
5827                 if (err) {
5828                         err = dma_set_coherent_mask(&pdev->dev,
5829                                                     DMA_BIT_MASK(32));
5830                         if (err) {
5831                                 dev_err(&pdev->dev, "No usable DMA "
5832                                         "configuration, aborting\n");
5833                                 goto err_dma;
5834                         }
5835                 }
5836         }
5837
5838         err = pci_request_selected_regions_exclusive(pdev,
5839                                           pci_select_bars(pdev, IORESOURCE_MEM),
5840                                           e1000e_driver_name);
5841         if (err)
5842                 goto err_pci_reg;
5843
5844         /* AER (Advanced Error Reporting) hooks */
5845         pci_enable_pcie_error_reporting(pdev);
5846
5847         pci_set_master(pdev);
5848         /* PCI config space info */
5849         err = pci_save_state(pdev);
5850         if (err)
5851                 goto err_alloc_etherdev;
5852
5853         err = -ENOMEM;
5854         netdev = alloc_etherdev(sizeof(struct e1000_adapter));
5855         if (!netdev)
5856                 goto err_alloc_etherdev;
5857
5858         SET_NETDEV_DEV(netdev, &pdev->dev);
5859
5860         netdev->irq = pdev->irq;
5861
5862         pci_set_drvdata(pdev, netdev);
5863         adapter = netdev_priv(netdev);
5864         hw = &adapter->hw;
5865         adapter->netdev = netdev;
5866         adapter->pdev = pdev;
5867         adapter->ei = ei;
5868         adapter->pba = ei->pba;
5869         adapter->flags = ei->flags;
5870         adapter->flags2 = ei->flags2;
5871         adapter->hw.adapter = adapter;
5872         adapter->hw.mac.type = ei->mac;
5873         adapter->max_hw_frame_size = ei->max_hw_frame_size;
5874         adapter->msg_enable = (1 << NETIF_MSG_DRV | NETIF_MSG_PROBE) - 1;
5875
5876         mmio_start = pci_resource_start(pdev, 0);
5877         mmio_len = pci_resource_len(pdev, 0);
5878
5879         err = -EIO;
5880         adapter->hw.hw_addr = ioremap(mmio_start, mmio_len);
5881         if (!adapter->hw.hw_addr)
5882                 goto err_ioremap;
5883
5884         if ((adapter->flags & FLAG_HAS_FLASH) &&
5885             (pci_resource_flags(pdev, 1) & IORESOURCE_MEM)) {
5886                 flash_start = pci_resource_start(pdev, 1);
5887                 flash_len = pci_resource_len(pdev, 1);
5888                 adapter->hw.flash_address = ioremap(flash_start, flash_len);
5889                 if (!adapter->hw.flash_address)
5890                         goto err_flashmap;
5891         }
5892
5893         /* construct the net_device struct */
5894         netdev->netdev_ops              = &e1000e_netdev_ops;
5895         e1000e_set_ethtool_ops(netdev);
5896         netdev->watchdog_timeo          = 5 * HZ;
5897         netif_napi_add(netdev, &adapter->napi, e1000_clean, 64);
5898         strncpy(netdev->name, pci_name(pdev), sizeof(netdev->name) - 1);
5899
5900         netdev->mem_start = mmio_start;
5901         netdev->mem_end = mmio_start + mmio_len;
5902
5903         adapter->bd_number = cards_found++;
5904
5905         e1000e_check_options(adapter);
5906
5907         /* setup adapter struct */
5908         err = e1000_sw_init(adapter);
5909         if (err)
5910                 goto err_sw_init;
5911
5912         memcpy(&hw->mac.ops, ei->mac_ops, sizeof(hw->mac.ops));
5913         memcpy(&hw->nvm.ops, ei->nvm_ops, sizeof(hw->nvm.ops));
5914         memcpy(&hw->phy.ops, ei->phy_ops, sizeof(hw->phy.ops));
5915
5916         err = ei->get_variants(adapter);
5917         if (err)
5918                 goto err_hw_init;
5919
5920         if ((adapter->flags & FLAG_IS_ICH) &&
5921             (adapter->flags & FLAG_READ_ONLY_NVM))
5922                 e1000e_write_protect_nvm_ich8lan(&adapter->hw);
5923
5924         hw->mac.ops.get_bus_info(&adapter->hw);
5925
5926         adapter->hw.phy.autoneg_wait_to_complete = 0;
5927
5928         /* Copper options */
5929         if (adapter->hw.phy.media_type == e1000_media_type_copper) {
5930                 adapter->hw.phy.mdix = AUTO_ALL_MODES;
5931                 adapter->hw.phy.disable_polarity_correction = 0;
5932                 adapter->hw.phy.ms_type = e1000_ms_hw_default;
5933         }
5934
5935         if (e1000_check_reset_block(&adapter->hw))
5936                 e_info("PHY reset is blocked due to SOL/IDER session.\n");
5937
5938         netdev->features = NETIF_F_SG |
5939                            NETIF_F_HW_CSUM |
5940                            NETIF_F_HW_VLAN_TX |
5941                            NETIF_F_HW_VLAN_RX;
5942
5943         if (adapter->flags & FLAG_HAS_HW_VLAN_FILTER)
5944                 netdev->features |= NETIF_F_HW_VLAN_FILTER;
5945
5946         netdev->features |= NETIF_F_TSO;
5947         netdev->features |= NETIF_F_TSO6;
5948
5949         netdev->vlan_features |= NETIF_F_TSO;
5950         netdev->vlan_features |= NETIF_F_TSO6;
5951         netdev->vlan_features |= NETIF_F_HW_CSUM;
5952         netdev->vlan_features |= NETIF_F_SG;
5953
5954         if (pci_using_dac) {
5955                 netdev->features |= NETIF_F_HIGHDMA;
5956                 netdev->vlan_features |= NETIF_F_HIGHDMA;
5957         }
5958
5959         if (e1000e_enable_mng_pass_thru(&adapter->hw))
5960                 adapter->flags |= FLAG_MNG_PT_ENABLED;
5961
5962         /*
5963          * before reading the NVM, reset the controller to
5964          * put the device in a known good starting state
5965          */
5966         adapter->hw.mac.ops.reset_hw(&adapter->hw);
5967
5968         /*
5969          * systems with ASPM and others may see the checksum fail on the first
5970          * attempt. Let's give it a few tries
5971          */
5972         for (i = 0;; i++) {
5973                 if (e1000_validate_nvm_checksum(&adapter->hw) >= 0)
5974                         break;
5975                 if (i == 2) {
5976                         e_err("The NVM Checksum Is Not Valid\n");
5977                         err = -EIO;
5978                         goto err_eeprom;
5979                 }
5980         }
5981
5982         e1000_eeprom_checks(adapter);
5983
5984         /* copy the MAC address */
5985         if (e1000e_read_mac_addr(&adapter->hw))
5986                 e_err("NVM Read Error while reading MAC address\n");
5987
5988         memcpy(netdev->dev_addr, adapter->hw.mac.addr, netdev->addr_len);
5989         memcpy(netdev->perm_addr, adapter->hw.mac.addr, netdev->addr_len);
5990
5991         if (!is_valid_ether_addr(netdev->perm_addr)) {
5992                 e_err("Invalid MAC Address: %pM\n", netdev->perm_addr);
5993                 err = -EIO;
5994                 goto err_eeprom;
5995         }
5996
5997         init_timer(&adapter->watchdog_timer);
5998         adapter->watchdog_timer.function = e1000_watchdog;
5999         adapter->watchdog_timer.data = (unsigned long) adapter;
6000
6001         init_timer(&adapter->phy_info_timer);
6002         adapter->phy_info_timer.function = e1000_update_phy_info;
6003         adapter->phy_info_timer.data = (unsigned long) adapter;
6004
6005         INIT_WORK(&adapter->reset_task, e1000_reset_task);
6006         INIT_WORK(&adapter->watchdog_task, e1000_watchdog_task);
6007         INIT_WORK(&adapter->downshift_task, e1000e_downshift_workaround);
6008         INIT_WORK(&adapter->update_phy_task, e1000e_update_phy_task);
6009         INIT_WORK(&adapter->print_hang_task, e1000_print_hw_hang);
6010
6011         /* Initialize link parameters. User can change them with ethtool */
6012         adapter->hw.mac.autoneg = 1;
6013         adapter->fc_autoneg = 1;
6014         adapter->hw.fc.requested_mode = e1000_fc_default;
6015         adapter->hw.fc.current_mode = e1000_fc_default;
6016         adapter->hw.phy.autoneg_advertised = 0x2f;
6017
6018         /* ring size defaults */
6019         adapter->rx_ring->count = 256;
6020         adapter->tx_ring->count = 256;
6021
6022         /*
6023          * Initial Wake on LAN setting - If APM wake is enabled in
6024          * the EEPROM, enable the ACPI Magic Packet filter
6025          */
6026         if (adapter->flags & FLAG_APME_IN_WUC) {
6027                 /* APME bit in EEPROM is mapped to WUC.APME */
6028                 eeprom_data = er32(WUC);
6029                 eeprom_apme_mask = E1000_WUC_APME;
6030                 if ((hw->mac.type > e1000_ich10lan) &&
6031                     (eeprom_data & E1000_WUC_PHY_WAKE))
6032                         adapter->flags2 |= FLAG2_HAS_PHY_WAKEUP;
6033         } else if (adapter->flags & FLAG_APME_IN_CTRL3) {
6034                 if (adapter->flags & FLAG_APME_CHECK_PORT_B &&
6035                     (adapter->hw.bus.func == 1))
6036                         e1000_read_nvm(&adapter->hw,
6037                                 NVM_INIT_CONTROL3_PORT_B, 1, &eeprom_data);
6038                 else
6039                         e1000_read_nvm(&adapter->hw,
6040                                 NVM_INIT_CONTROL3_PORT_A, 1, &eeprom_data);
6041         }
6042
6043         /* fetch WoL from EEPROM */
6044         if (eeprom_data & eeprom_apme_mask)
6045                 adapter->eeprom_wol |= E1000_WUFC_MAG;
6046
6047         /*
6048          * now that we have the eeprom settings, apply the special cases
6049          * where the eeprom may be wrong or the board simply won't support
6050          * wake on lan on a particular port
6051          */
6052         if (!(adapter->flags & FLAG_HAS_WOL))
6053                 adapter->eeprom_wol = 0;
6054
6055         /* initialize the wol settings based on the eeprom settings */
6056         adapter->wol = adapter->eeprom_wol;
6057         device_set_wakeup_enable(&adapter->pdev->dev, adapter->wol);
6058
6059         /* save off EEPROM version number */
6060         e1000_read_nvm(&adapter->hw, 5, 1, &adapter->eeprom_vers);
6061
6062         /* reset the hardware with the new settings */
6063         e1000e_reset(adapter);
6064
6065         /*
6066          * If the controller has AMT, do not set DRV_LOAD until the interface
6067          * is up.  For all other cases, let the f/w know that the h/w is now
6068          * under the control of the driver.
6069          */
6070         if (!(adapter->flags & FLAG_HAS_AMT))
6071                 e1000e_get_hw_control(adapter);
6072
6073         strncpy(netdev->name, "eth%d", sizeof(netdev->name) - 1);
6074         err = register_netdev(netdev);
6075         if (err)
6076                 goto err_register;
6077
6078         /* carrier off reporting is important to ethtool even BEFORE open */
6079         netif_carrier_off(netdev);
6080
6081         e1000_print_device_info(adapter);
6082
6083         if (pci_dev_run_wake(pdev))
6084                 pm_runtime_put_noidle(&pdev->dev);
6085
6086         return 0;
6087
6088 err_register:
6089         if (!(adapter->flags & FLAG_HAS_AMT))
6090                 e1000e_release_hw_control(adapter);
6091 err_eeprom:
6092         if (!e1000_check_reset_block(&adapter->hw))
6093                 e1000_phy_hw_reset(&adapter->hw);
6094 err_hw_init:
6095         kfree(adapter->tx_ring);
6096         kfree(adapter->rx_ring);
6097 err_sw_init:
6098         if (adapter->hw.flash_address)
6099                 iounmap(adapter->hw.flash_address);
6100         e1000e_reset_interrupt_capability(adapter);
6101 err_flashmap:
6102         iounmap(adapter->hw.hw_addr);
6103 err_ioremap:
6104         free_netdev(netdev);
6105 err_alloc_etherdev:
6106         pci_release_selected_regions(pdev,
6107                                      pci_select_bars(pdev, IORESOURCE_MEM));
6108 err_pci_reg:
6109 err_dma:
6110         pci_disable_device(pdev);
6111         return err;
6112 }
6113
6114 /**
6115  * e1000_remove - Device Removal Routine
6116  * @pdev: PCI device information struct
6117  *
6118  * e1000_remove is called by the PCI subsystem to alert the driver
6119  * that it should release a PCI device.  The could be caused by a
6120  * Hot-Plug event, or because the driver is going to be removed from
6121  * memory.
6122  **/
6123 static void __devexit e1000_remove(struct pci_dev *pdev)
6124 {
6125         struct net_device *netdev = pci_get_drvdata(pdev);
6126         struct e1000_adapter *adapter = netdev_priv(netdev);
6127         bool down = test_bit(__E1000_DOWN, &adapter->state);
6128
6129         /*
6130          * The timers may be rescheduled, so explicitly disable them
6131          * from being rescheduled.
6132          */
6133         if (!down)
6134                 set_bit(__E1000_DOWN, &adapter->state);
6135         del_timer_sync(&adapter->watchdog_timer);
6136         del_timer_sync(&adapter->phy_info_timer);
6137
6138         cancel_work_sync(&adapter->reset_task);
6139         cancel_work_sync(&adapter->watchdog_task);
6140         cancel_work_sync(&adapter->downshift_task);
6141         cancel_work_sync(&adapter->update_phy_task);
6142         cancel_work_sync(&adapter->print_hang_task);
6143
6144         if (!(netdev->flags & IFF_UP))
6145                 e1000_power_down_phy(adapter);
6146
6147         /* Don't lie to e1000_close() down the road. */
6148         if (!down)
6149                 clear_bit(__E1000_DOWN, &adapter->state);
6150         unregister_netdev(netdev);
6151
6152         if (pci_dev_run_wake(pdev))
6153                 pm_runtime_get_noresume(&pdev->dev);
6154
6155         /*
6156          * Release control of h/w to f/w.  If f/w is AMT enabled, this
6157          * would have already happened in close and is redundant.
6158          */
6159         e1000e_release_hw_control(adapter);
6160
6161         e1000e_reset_interrupt_capability(adapter);
6162         kfree(adapter->tx_ring);
6163         kfree(adapter->rx_ring);
6164
6165         iounmap(adapter->hw.hw_addr);
6166         if (adapter->hw.flash_address)
6167                 iounmap(adapter->hw.flash_address);
6168         pci_release_selected_regions(pdev,
6169                                      pci_select_bars(pdev, IORESOURCE_MEM));
6170
6171         free_netdev(netdev);
6172
6173         /* AER disable */
6174         pci_disable_pcie_error_reporting(pdev);
6175
6176         pci_disable_device(pdev);
6177 }
6178
6179 /* PCI Error Recovery (ERS) */
6180 static struct pci_error_handlers e1000_err_handler = {
6181         .error_detected = e1000_io_error_detected,
6182         .slot_reset = e1000_io_slot_reset,
6183         .resume = e1000_io_resume,
6184 };
6185
6186 static DEFINE_PCI_DEVICE_TABLE(e1000_pci_tbl) = {
6187         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82571EB_COPPER), board_82571 },
6188         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82571EB_FIBER), board_82571 },
6189         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82571EB_QUAD_COPPER), board_82571 },
6190         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82571EB_QUAD_COPPER_LP), board_82571 },
6191         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82571EB_QUAD_FIBER), board_82571 },
6192         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82571EB_SERDES), board_82571 },
6193         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82571EB_SERDES_DUAL), board_82571 },
6194         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82571EB_SERDES_QUAD), board_82571 },
6195         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82571PT_QUAD_COPPER), board_82571 },
6196
6197         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82572EI), board_82572 },
6198         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82572EI_COPPER), board_82572 },
6199         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82572EI_FIBER), board_82572 },
6200         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82572EI_SERDES), board_82572 },
6201
6202         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82573E), board_82573 },
6203         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82573E_IAMT), board_82573 },
6204         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82573L), board_82573 },
6205
6206         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82574L), board_82574 },
6207         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82574LA), board_82574 },
6208         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82583V), board_82583 },
6209
6210         { PCI_VDEVICE(INTEL, E1000_DEV_ID_80003ES2LAN_COPPER_DPT),
6211           board_80003es2lan },
6212         { PCI_VDEVICE(INTEL, E1000_DEV_ID_80003ES2LAN_COPPER_SPT),
6213           board_80003es2lan },
6214         { PCI_VDEVICE(INTEL, E1000_DEV_ID_80003ES2LAN_SERDES_DPT),
6215           board_80003es2lan },
6216         { PCI_VDEVICE(INTEL, E1000_DEV_ID_80003ES2LAN_SERDES_SPT),
6217           board_80003es2lan },
6218
6219         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH8_IFE), board_ich8lan },
6220         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH8_IFE_G), board_ich8lan },
6221         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH8_IFE_GT), board_ich8lan },
6222         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH8_IGP_AMT), board_ich8lan },
6223         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH8_IGP_C), board_ich8lan },
6224         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH8_IGP_M), board_ich8lan },
6225         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH8_IGP_M_AMT), board_ich8lan },
6226         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH8_82567V_3), board_ich8lan },
6227
6228         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH9_IFE), board_ich9lan },
6229         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH9_IFE_G), board_ich9lan },
6230         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH9_IFE_GT), board_ich9lan },
6231         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH9_IGP_AMT), board_ich9lan },
6232         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH9_IGP_C), board_ich9lan },
6233         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH9_BM), board_ich9lan },
6234         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH9_IGP_M), board_ich9lan },
6235         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH9_IGP_M_AMT), board_ich9lan },
6236         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH9_IGP_M_V), board_ich9lan },
6237
6238         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH10_R_BM_LM), board_ich9lan },
6239         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH10_R_BM_LF), board_ich9lan },
6240         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH10_R_BM_V), board_ich9lan },
6241
6242         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH10_D_BM_LM), board_ich10lan },
6243         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH10_D_BM_LF), board_ich10lan },
6244         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH10_D_BM_V), board_ich10lan },
6245
6246         { PCI_VDEVICE(INTEL, E1000_DEV_ID_PCH_M_HV_LM), board_pchlan },
6247         { PCI_VDEVICE(INTEL, E1000_DEV_ID_PCH_M_HV_LC), board_pchlan },
6248         { PCI_VDEVICE(INTEL, E1000_DEV_ID_PCH_D_HV_DM), board_pchlan },
6249         { PCI_VDEVICE(INTEL, E1000_DEV_ID_PCH_D_HV_DC), board_pchlan },
6250
6251         { PCI_VDEVICE(INTEL, E1000_DEV_ID_PCH2_LV_LM), board_pch2lan },
6252         { PCI_VDEVICE(INTEL, E1000_DEV_ID_PCH2_LV_V), board_pch2lan },
6253
6254         { }     /* terminate list */
6255 };
6256 MODULE_DEVICE_TABLE(pci, e1000_pci_tbl);
6257
6258 #ifdef CONFIG_PM
6259 static const struct dev_pm_ops e1000_pm_ops = {
6260         SET_SYSTEM_SLEEP_PM_OPS(e1000_suspend, e1000_resume)
6261         SET_RUNTIME_PM_OPS(e1000_runtime_suspend,
6262                                 e1000_runtime_resume, e1000_idle)
6263 };
6264 #endif
6265
6266 /* PCI Device API Driver */
6267 static struct pci_driver e1000_driver = {
6268         .name     = e1000e_driver_name,
6269         .id_table = e1000_pci_tbl,
6270         .probe    = e1000_probe,
6271         .remove   = __devexit_p(e1000_remove),
6272 #ifdef CONFIG_PM
6273         .driver.pm = &e1000_pm_ops,
6274 #endif
6275         .shutdown = e1000_shutdown,
6276         .err_handler = &e1000_err_handler
6277 };
6278
6279 /**
6280  * e1000_init_module - Driver Registration Routine
6281  *
6282  * e1000_init_module is the first routine called when the driver is
6283  * loaded. All it does is register with the PCI subsystem.
6284  **/
6285 static int __init e1000_init_module(void)
6286 {
6287         int ret;
6288         pr_info("Intel(R) PRO/1000 Network Driver - %s\n",
6289                 e1000e_driver_version);
6290         pr_info("Copyright(c) 1999 - 2011 Intel Corporation.\n");
6291         ret = pci_register_driver(&e1000_driver);
6292
6293         return ret;
6294 }
6295 module_init(e1000_init_module);
6296
6297 /**
6298  * e1000_exit_module - Driver Exit Cleanup Routine
6299  *
6300  * e1000_exit_module is called just before the driver is removed
6301  * from memory.
6302  **/
6303 static void __exit e1000_exit_module(void)
6304 {
6305         pci_unregister_driver(&e1000_driver);
6306 }
6307 module_exit(e1000_exit_module);
6308
6309
6310 MODULE_AUTHOR("Intel Corporation, <linux.nics@intel.com>");
6311 MODULE_DESCRIPTION("Intel(R) PRO/1000 Network Driver");
6312 MODULE_LICENSE("GPL");
6313 MODULE_VERSION(DRV_VERSION);
6314
6315 /* e1000_main.c */