drivers/net/e1000e/netdev.c

   1 /*******************************************************************************
   2
   3   Intel PRO/1000 Linux driver
   4   Copyright(c) 1999 - 2009 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 #include <linux/module.h>
  30 #include <linux/types.h>
  31 #include <linux/init.h>
  32 #include <linux/pci.h>
  33 #include <linux/vmalloc.h>
  34 #include <linux/pagemap.h>
  35 #include <linux/delay.h>
  36 #include <linux/netdevice.h>
  37 #include <linux/tcp.h>
  38 #include <linux/ipv6.h>
  39 #include <net/checksum.h>
  40 #include <net/ip6_checksum.h>
  41 #include <linux/mii.h>
  42 #include <linux/ethtool.h>
  43 #include <linux/if_vlan.h>
  44 #include <linux/cpu.h>
  45 #include <linux/smp.h>
  46 #include <linux/pm_qos_params.h>
  47 #include <linux/aer.h>
  48
  49 #include "e1000.h"
  50
  51 #define DRV_VERSION "1.0.2-k2"
  52 char e1000e_driver_name[] = "e1000e";
  53 const char e1000e_driver_version[] = DRV_VERSION;
  54
  55 static const struct e1000_info *e1000_info_tbl[] = {
  56         [board_82571]           = &e1000_82571_info,
  57         [board_82572]           = &e1000_82572_info,
  58         [board_82573]           = &e1000_82573_info,
  59         [board_82574]           = &e1000_82574_info,
  60         [board_82583]           = &e1000_82583_info,
  61         [board_80003es2lan]     = &e1000_es2_info,
  62         [board_ich8lan]         = &e1000_ich8_info,
  63         [board_ich9lan]         = &e1000_ich9_info,
  64         [board_ich10lan]        = &e1000_ich10_info,
  65         [board_pchlan]          = &e1000_pch_info,
  66 };
  67
  68 /**
  69  * e1000_desc_unused - calculate if we have unused descriptors
  70  **/
  71 static int e1000_desc_unused(struct e1000_ring *ring)
  72 {
  73         if (ring->next_to_clean > ring->next_to_use)
  74                 return ring->next_to_clean - ring->next_to_use - 1;
  75
  76         return ring->count + ring->next_to_clean - ring->next_to_use - 1;
  77 }
  78
  79 /**
  80  * e1000_receive_skb - helper function to handle Rx indications
  81  * @adapter: board private structure
  82  * @status: descriptor status field as written by hardware
  83  * @vlan: descriptor vlan field as written by hardware (no le/be conversion)
  84  * @skb: pointer to sk_buff to be indicated to stack
  85  **/
  86 static void e1000_receive_skb(struct e1000_adapter *adapter,
  87                               struct net_device *netdev,
  88                               struct sk_buff *skb,
  89                               u8 status, __le16 vlan)
  90 {
  91         skb->protocol = eth_type_trans(skb, netdev);
  92
  93         if (adapter->vlgrp && (status & E1000_RXD_STAT_VP))
  94                 vlan_gro_receive(&adapter->napi, adapter->vlgrp,
  95                                  le16_to_cpu(vlan), skb);
  96         else
  97                 napi_gro_receive(&adapter->napi, skb);
  98 }
  99
 100 /**
 101  * e1000_rx_checksum - Receive Checksum Offload for 82543
 102  * @adapter:     board private structure
 103  * @status_err:  receive descriptor status and error fields
 104  * @csum:       receive descriptor csum field
 105  * @sk_buff:     socket buffer with received data
 106  **/
 107 static void e1000_rx_checksum(struct e1000_adapter *adapter, u32 status_err,
 108                               u32 csum, struct sk_buff *skb)
 109 {
 110         u16 status = (u16)status_err;
 111         u8 errors = (u8)(status_err >> 24);
 112         skb->ip_summed = CHECKSUM_NONE;
 113
 114         /* Ignore Checksum bit is set */
 115         if (status & E1000_RXD_STAT_IXSM)
 116                 return;
 117         /* TCP/UDP checksum error bit is set */
 118         if (errors & E1000_RXD_ERR_TCPE) {
 119                 /* let the stack verify checksum errors */
 120                 adapter->hw_csum_err++;
 121                 return;
 122         }
 123
 124         /* TCP/UDP Checksum has not been calculated */
 125         if (!(status & (E1000_RXD_STAT_TCPCS | E1000_RXD_STAT_UDPCS)))
 126                 return;
 127
 128         /* It must be a TCP or UDP packet with a valid checksum */
 129         if (status & E1000_RXD_STAT_TCPCS) {
 130                 /* TCP checksum is good */
 131                 skb->ip_summed = CHECKSUM_UNNECESSARY;
 132         } else {
 133                 /*
 134                  * IP fragment with UDP payload
 135                  * Hardware complements the payload checksum, so we undo it
 136                  * and then put the value in host order for further stack use.
 137                  */
 138                 __sum16 sum = (__force __sum16)htons(csum);
 139                 skb->csum = csum_unfold(~sum);
 140                 skb->ip_summed = CHECKSUM_COMPLETE;
 141         }
 142         adapter->hw_csum_good++;
 143 }
 144
 145 /**
 146  * e1000_alloc_rx_buffers - Replace used receive buffers; legacy & extended
 147  * @adapter: address of board private structure
 148  **/
 149 static void e1000_alloc_rx_buffers(struct e1000_adapter *adapter,
 150                                    int cleaned_count)
 151 {
 152         struct net_device *netdev = adapter->netdev;
 153         struct pci_dev *pdev = adapter->pdev;
 154         struct e1000_ring *rx_ring = adapter->rx_ring;
 155         struct e1000_rx_desc *rx_desc;
 156         struct e1000_buffer *buffer_info;
 157         struct sk_buff *skb;
 158         unsigned int i;
 159         unsigned int bufsz = adapter->rx_buffer_len;
 160
 161         i = rx_ring->next_to_use;
 162         buffer_info = &rx_ring->buffer_info[i];
 163
 164         while (cleaned_count--) {
 165                 skb = buffer_info->skb;
 166                 if (skb) {
 167                         skb_trim(skb, 0);
 168                         goto map_skb;
 169                 }
 170
 171                 skb = netdev_alloc_skb_ip_align(netdev, bufsz);
 172                 if (!skb) {
 173                         /* Better luck next round */
 174                         adapter->alloc_rx_buff_failed++;
 175                         break;
 176                 }
 177
 178                 buffer_info->skb = skb;
 179 map_skb:
 180                 buffer_info->dma = pci_map_single(pdev, skb->data,
 181                                                   adapter->rx_buffer_len,
 182                                                   PCI_DMA_FROMDEVICE);
 183                 if (pci_dma_mapping_error(pdev, buffer_info->dma)) {
 184                         dev_err(&pdev->dev, "RX DMA map failed\n");
 185                         adapter->rx_dma_failed++;
 186                         break;
 187                 }
 188
 189                 rx_desc = E1000_RX_DESC(*rx_ring, i);
 190                 rx_desc->buffer_addr = cpu_to_le64(buffer_info->dma);
 191
 192                 i++;
 193                 if (i == rx_ring->count)
 194                         i = 0;
 195                 buffer_info = &rx_ring->buffer_info[i];
 196         }
 197
 198         if (rx_ring->next_to_use != i) {
 199                 rx_ring->next_to_use = i;
 200                 if (i-- == 0)
 201                         i = (rx_ring->count - 1);
 202
 203                 /*
 204                  * Force memory writes to complete before letting h/w
 205                  * know there are new descriptors to fetch.  (Only
 206                  * applicable for weak-ordered memory model archs,
 207                  * such as IA-64).
 208                  */
 209                 wmb();
 210                 writel(i, adapter->hw.hw_addr + rx_ring->tail);
 211         }
 212 }
 213
 214 /**
 215  * e1000_alloc_rx_buffers_ps - Replace used receive buffers; packet split
 216  * @adapter: address of board private structure
 217  **/
 218 static void e1000_alloc_rx_buffers_ps(struct e1000_adapter *adapter,
 219                                       int cleaned_count)
 220 {
 221         struct net_device *netdev = adapter->netdev;
 222         struct pci_dev *pdev = adapter->pdev;
 223         union e1000_rx_desc_packet_split *rx_desc;
 224         struct e1000_ring *rx_ring = adapter->rx_ring;
 225         struct e1000_buffer *buffer_info;
 226         struct e1000_ps_page *ps_page;
 227         struct sk_buff *skb;
 228         unsigned int i, j;
 229
 230         i = rx_ring->next_to_use;
 231         buffer_info = &rx_ring->buffer_info[i];
 232
 233         while (cleaned_count--) {
 234                 rx_desc = E1000_RX_DESC_PS(*rx_ring, i);
 235
 236                 for (j = 0; j < PS_PAGE_BUFFERS; j++) {
 237                         ps_page = &buffer_info->ps_pages[j];
 238                         if (j >= adapter->rx_ps_pages) {
 239                                 /* all unused desc entries get hw null ptr */
 240                                 rx_desc->read.buffer_addr[j+1] = ~cpu_to_le64(0);
 241                                 continue;
 242                         }
 243                         if (!ps_page->page) {
 244                                 ps_page->page = alloc_page(GFP_ATOMIC);
 245                                 if (!ps_page->page) {
 246                                         adapter->alloc_rx_buff_failed++;
 247                                         goto no_buffers;
 248                                 }
 249                                 ps_page->dma = pci_map_page(pdev,
 250                                                    ps_page->page,
 251                                                    0, PAGE_SIZE,
 252                                                    PCI_DMA_FROMDEVICE);
 253                                 if (pci_dma_mapping_error(pdev, ps_page->dma)) {
 254                                         dev_err(&adapter->pdev->dev,
 255                                           "RX DMA page map failed\n");
 256                                         adapter->rx_dma_failed++;
 257                                         goto no_buffers;
 258                                 }
 259                         }
 260                         /*
 261                          * Refresh the desc even if buffer_addrs
 262                          * didn't change because each write-back
 263                          * erases this info.
 264                          */
 265                         rx_desc->read.buffer_addr[j+1] =
 266                              cpu_to_le64(ps_page->dma);
 267                 }
 268
 269                 skb = netdev_alloc_skb_ip_align(netdev,
 270                                                 adapter->rx_ps_bsize0);
 271
 272                 if (!skb) {
 273                         adapter->alloc_rx_buff_failed++;
 274                         break;
 275                 }
 276
 277                 buffer_info->skb = skb;
 278                 buffer_info->dma = pci_map_single(pdev, skb->data,
 279                                                   adapter->rx_ps_bsize0,
 280                                                   PCI_DMA_FROMDEVICE);
 281                 if (pci_dma_mapping_error(pdev, buffer_info->dma)) {
 282                         dev_err(&pdev->dev, "RX DMA map failed\n");
 283                         adapter->rx_dma_failed++;
 284                         /* cleanup skb */
 285                         dev_kfree_skb_any(skb);
 286                         buffer_info->skb = NULL;
 287                         break;
 288                 }
 289
 290                 rx_desc->read.buffer_addr[0] = cpu_to_le64(buffer_info->dma);
 291
 292                 i++;
 293                 if (i == rx_ring->count)
 294                         i = 0;
 295                 buffer_info = &rx_ring->buffer_info[i];
 296         }
 297
 298 no_buffers:
 299         if (rx_ring->next_to_use != i) {
 300                 rx_ring->next_to_use = i;
 301
 302                 if (!(i--))
 303                         i = (rx_ring->count - 1);
 304
 305                 /*
 306                  * Force memory writes to complete before letting h/w
 307                  * know there are new descriptors to fetch.  (Only
 308                  * applicable for weak-ordered memory model archs,
 309                  * such as IA-64).
 310                  */
 311                 wmb();
 312                 /*
 313                  * Hardware increments by 16 bytes, but packet split
 314                  * descriptors are 32 bytes...so we increment tail
 315                  * twice as much.
 316                  */
 317                 writel(i<<1, adapter->hw.hw_addr + rx_ring->tail);
 318         }
 319 }
 320
 321 /**
 322  * e1000_alloc_jumbo_rx_buffers - Replace used jumbo receive buffers
 323  * @adapter: address of board private structure
 324  * @cleaned_count: number of buffers to allocate this pass
 325  **/
 326
 327 static void e1000_alloc_jumbo_rx_buffers(struct e1000_adapter *adapter,
 328                                          int cleaned_count)
 329 {
 330         struct net_device *netdev = adapter->netdev;
 331         struct pci_dev *pdev = adapter->pdev;
 332         struct e1000_rx_desc *rx_desc;
 333         struct e1000_ring *rx_ring = adapter->rx_ring;
 334         struct e1000_buffer *buffer_info;
 335         struct sk_buff *skb;
 336         unsigned int i;
 337         unsigned int bufsz = 256 - 16 /* for skb_reserve */;
 338
 339         i = rx_ring->next_to_use;
 340         buffer_info = &rx_ring->buffer_info[i];
 341
 342         while (cleaned_count--) {
 343                 skb = buffer_info->skb;
 344                 if (skb) {
 345                         skb_trim(skb, 0);
 346                         goto check_page;
 347                 }
 348
 349                 skb = netdev_alloc_skb_ip_align(netdev, bufsz);
 350                 if (unlikely(!skb)) {
 351                         /* Better luck next round */
 352                         adapter->alloc_rx_buff_failed++;
 353                         break;
 354                 }
 355
 356                 buffer_info->skb = skb;
 357 check_page:
 358                 /* allocate a new page if necessary */
 359                 if (!buffer_info->page) {
 360                         buffer_info->page = alloc_page(GFP_ATOMIC);
 361                         if (unlikely(!buffer_info->page)) {
 362                                 adapter->alloc_rx_buff_failed++;
 363                                 break;
 364                         }
 365                 }
 366
 367                 if (!buffer_info->dma)
 368                         buffer_info->dma = pci_map_page(pdev,
 369                                                         buffer_info->page, 0,
 370                                                         PAGE_SIZE,
 371                                                         PCI_DMA_FROMDEVICE);
 372
 373                 rx_desc = E1000_RX_DESC(*rx_ring, i);
 374                 rx_desc->buffer_addr = cpu_to_le64(buffer_info->dma);
 375
 376                 if (unlikely(++i == rx_ring->count))
 377                         i = 0;
 378                 buffer_info = &rx_ring->buffer_info[i];
 379         }
 380
 381         if (likely(rx_ring->next_to_use != i)) {
 382                 rx_ring->next_to_use = i;
 383                 if (unlikely(i-- == 0))
 384                         i = (rx_ring->count - 1);
 385
 386                 /* Force memory writes to complete before letting h/w
 387                  * know there are new descriptors to fetch.  (Only
 388                  * applicable for weak-ordered memory model archs,
 389                  * such as IA-64). */
 390                 wmb();
 391                 writel(i, adapter->hw.hw_addr + rx_ring->tail);
 392         }
 393 }
 394
 395 /**
 396  * e1000_clean_rx_irq - Send received data up the network stack; legacy
 397  * @adapter: board private structure
 398  *
 399  * the return value indicates whether actual cleaning was done, there
 400  * is no guarantee that everything was cleaned
 401  **/
 402 static bool e1000_clean_rx_irq(struct e1000_adapter *adapter,
 403                                int *work_done, int work_to_do)
 404 {
 405         struct net_device *netdev = adapter->netdev;
 406         struct pci_dev *pdev = adapter->pdev;
 407         struct e1000_hw *hw = &adapter->hw;
 408         struct e1000_ring *rx_ring = adapter->rx_ring;
 409         struct e1000_rx_desc *rx_desc, *next_rxd;
 410         struct e1000_buffer *buffer_info, *next_buffer;
 411         u32 length;
 412         unsigned int i;
 413         int cleaned_count = 0;
 414         bool cleaned = 0;
 415         unsigned int total_rx_bytes = 0, total_rx_packets = 0;
 416
 417         i = rx_ring->next_to_clean;
 418         rx_desc = E1000_RX_DESC(*rx_ring, i);
 419         buffer_info = &rx_ring->buffer_info[i];
 420
 421         while (rx_desc->status & E1000_RXD_STAT_DD) {
 422                 struct sk_buff *skb;
 423                 u8 status;
 424
 425                 if (*work_done >= work_to_do)
 426                         break;
 427                 (*work_done)++;
 428
 429                 status = rx_desc->status;
 430                 skb = buffer_info->skb;
 431                 buffer_info->skb = NULL;
 432
 433                 prefetch(skb->data - NET_IP_ALIGN);
 434
 435                 i++;
 436                 if (i == rx_ring->count)
 437                         i = 0;
 438                 next_rxd = E1000_RX_DESC(*rx_ring, i);
 439                 prefetch(next_rxd);
 440
 441                 next_buffer = &rx_ring->buffer_info[i];
 442
 443                 cleaned = 1;
 444                 cleaned_count++;
 445                 pci_unmap_single(pdev,
 446                                  buffer_info->dma,
 447                                  adapter->rx_buffer_len,
 448                                  PCI_DMA_FROMDEVICE);
 449                 buffer_info->dma = 0;
 450
 451                 length = le16_to_cpu(rx_desc->length);
 452
 453                 /* !EOP means multiple descriptors were used to store a single
 454                  * packet, also make sure the frame isn't just CRC only */
 455                 if (!(status & E1000_RXD_STAT_EOP) || (length <= 4)) {
 456                         /* All receives must fit into a single buffer */
 457                         e_dbg("Receive packet consumed multiple buffers\n");
 458                         /* recycle */
 459                         buffer_info->skb = skb;
 460                         goto next_desc;
 461                 }
 462
 463                 if (rx_desc->errors & E1000_RXD_ERR_FRAME_ERR_MASK) {
 464                         /* recycle */
 465                         buffer_info->skb = skb;
 466                         goto next_desc;
 467                 }
 468
 469                 /* adjust length to remove Ethernet CRC */
 470                 if (!(adapter->flags2 & FLAG2_CRC_STRIPPING))
 471                         length -= 4;
 472
 473                 total_rx_bytes += length;
 474                 total_rx_packets++;
 475
 476                 /*
 477                  * code added for copybreak, this should improve
 478                  * performance for small packets with large amounts
 479                  * of reassembly being done in the stack
 480                  */
 481                 if (length < copybreak) {
 482                         struct sk_buff *new_skb =
 483                             netdev_alloc_skb_ip_align(netdev, length);
 484                         if (new_skb) {
 485                                 skb_copy_to_linear_data_offset(new_skb,
 486                                                                -NET_IP_ALIGN,
 487                                                                (skb->data -
 488                                                                 NET_IP_ALIGN),
 489                                                                (length +
 490                                                                 NET_IP_ALIGN));
 491                                 /* save the skb in buffer_info as good */
 492                                 buffer_info->skb = skb;
 493                                 skb = new_skb;
 494                         }
 495                         /* else just continue with the old one */
 496                 }
 497                 /* end copybreak code */
 498                 skb_put(skb, length);
 499
 500                 /* Receive Checksum Offload */
 501                 e1000_rx_checksum(adapter,
 502                                   (u32)(status) |
 503                                   ((u32)(rx_desc->errors) << 24),
 504                                   le16_to_cpu(rx_desc->csum), skb);
 505
 506                 e1000_receive_skb(adapter, netdev, skb,status,rx_desc->special);
 507
 508 next_desc:
 509                 rx_desc->status = 0;
 510
 511                 /* return some buffers to hardware, one at a time is too slow */
 512                 if (cleaned_count >= E1000_RX_BUFFER_WRITE) {
 513                         adapter->alloc_rx_buf(adapter, cleaned_count);
 514                         cleaned_count = 0;
 515                 }
 516
 517                 /* use prefetched values */
 518                 rx_desc = next_rxd;
 519                 buffer_info = next_buffer;
 520         }
 521         rx_ring->next_to_clean = i;
 522
 523         cleaned_count = e1000_desc_unused(rx_ring);
 524         if (cleaned_count)
 525                 adapter->alloc_rx_buf(adapter, cleaned_count);
 526
 527         adapter->total_rx_bytes += total_rx_bytes;
 528         adapter->total_rx_packets += total_rx_packets;
 529         netdev->stats.rx_bytes += total_rx_bytes;
 530         netdev->stats.rx_packets += total_rx_packets;
 531         return cleaned;
 532 }
 533
 534 static void e1000_put_txbuf(struct e1000_adapter *adapter,
 535                              struct e1000_buffer *buffer_info)
 536 {
 537         buffer_info->dma = 0;
 538         if (buffer_info->skb) {
 539                 skb_dma_unmap(&adapter->pdev->dev, buffer_info->skb,
 540                               DMA_TO_DEVICE);
 541                 dev_kfree_skb_any(buffer_info->skb);
 542                 buffer_info->skb = NULL;
 543         }
 544         buffer_info->time_stamp = 0;
 545 }
 546
 547 static void e1000_print_tx_hang(struct e1000_adapter *adapter)
 548 {
 549         struct e1000_ring *tx_ring = adapter->tx_ring;
 550         unsigned int i = tx_ring->next_to_clean;
 551         unsigned int eop = tx_ring->buffer_info[i].next_to_watch;
 552         struct e1000_tx_desc *eop_desc = E1000_TX_DESC(*tx_ring, eop);
 553
 554         /* detected Tx unit hang */
 555         e_err("Detected Tx Unit Hang:\n"
 556               "  TDH                  <%x>\n"
 557               "  TDT                  <%x>\n"
 558               "  next_to_use          <%x>\n"
 559               "  next_to_clean        <%x>\n"
 560               "buffer_info[next_to_clean]:\n"
 561               "  time_stamp           <%lx>\n"
 562               "  next_to_watch        <%x>\n"
 563               "  jiffies              <%lx>\n"
 564               "  next_to_watch.status <%x>\n",
 565               readl(adapter->hw.hw_addr + tx_ring->head),
 566               readl(adapter->hw.hw_addr + tx_ring->tail),
 567               tx_ring->next_to_use,
 568               tx_ring->next_to_clean,
 569               tx_ring->buffer_info[eop].time_stamp,
 570               eop,
 571               jiffies,
 572               eop_desc->upper.fields.status);
 573 }
 574
 575 /**
 576  * e1000_clean_tx_irq - Reclaim resources after transmit completes
 577  * @adapter: board private structure
 578  *
 579  * the return value indicates whether actual cleaning was done, there
 580  * is no guarantee that everything was cleaned
 581  **/
 582 static bool e1000_clean_tx_irq(struct e1000_adapter *adapter)
 583 {
 584         struct net_device *netdev = adapter->netdev;
 585         struct e1000_hw *hw = &adapter->hw;
 586         struct e1000_ring *tx_ring = adapter->tx_ring;
 587         struct e1000_tx_desc *tx_desc, *eop_desc;
 588         struct e1000_buffer *buffer_info;
 589         unsigned int i, eop;
 590         unsigned int count = 0;
 591         unsigned int total_tx_bytes = 0, total_tx_packets = 0;
 592
 593         i = tx_ring->next_to_clean;
 594         eop = tx_ring->buffer_info[i].next_to_watch;
 595         eop_desc = E1000_TX_DESC(*tx_ring, eop);
 596
 597         while ((eop_desc->upper.data & cpu_to_le32(E1000_TXD_STAT_DD)) &&
 598                (count < tx_ring->count)) {
 599                 bool cleaned = false;
 600                 for (; !cleaned; count++) {
 601                         tx_desc = E1000_TX_DESC(*tx_ring, i);
 602                         buffer_info = &tx_ring->buffer_info[i];
 603                         cleaned = (i == eop);
 604
 605                         if (cleaned) {
 606                                 struct sk_buff *skb = buffer_info->skb;
 607                                 unsigned int segs, bytecount;
 608                                 segs = skb_shinfo(skb)->gso_segs ?: 1;
 609                                 /* multiply data chunks by size of headers */
 610                                 bytecount = ((segs - 1) * skb_headlen(skb)) +
 611                                             skb->len;
 612                                 total_tx_packets += segs;
 613                                 total_tx_bytes += bytecount;
 614                         }
 615
 616                         e1000_put_txbuf(adapter, buffer_info);
 617                         tx_desc->upper.data = 0;
 618
 619                         i++;
 620                         if (i == tx_ring->count)
 621                                 i = 0;
 622                 }
 623
 624                 eop = tx_ring->buffer_info[i].next_to_watch;
 625                 eop_desc = E1000_TX_DESC(*tx_ring, eop);
 626         }
 627
 628         tx_ring->next_to_clean = i;
 629
 630 #define TX_WAKE_THRESHOLD 32
 631         if (count && netif_carrier_ok(netdev) &&
 632             e1000_desc_unused(tx_ring) >= TX_WAKE_THRESHOLD) {
 633                 /* Make sure that anybody stopping the queue after this
 634                  * sees the new next_to_clean.
 635                  */
 636                 smp_mb();
 637
 638                 if (netif_queue_stopped(netdev) &&
 639                     !(test_bit(__E1000_DOWN, &adapter->state))) {
 640                         netif_wake_queue(netdev);
 641                         ++adapter->restart_queue;
 642                 }
 643         }
 644
 645         if (adapter->detect_tx_hung) {
 646                 /* Detect a transmit hang in hardware, this serializes the
 647                  * check with the clearing of time_stamp and movement of i */
 648                 adapter->detect_tx_hung = 0;
 649                 if (tx_ring->buffer_info[i].time_stamp &&
 650                     time_after(jiffies, tx_ring->buffer_info[i].time_stamp
 651                                + (adapter->tx_timeout_factor * HZ))
 652                     && !(er32(STATUS) & E1000_STATUS_TXOFF)) {
 653                         e1000_print_tx_hang(adapter);
 654                         netif_stop_queue(netdev);
 655                 }
 656         }
 657         adapter->total_tx_bytes += total_tx_bytes;
 658         adapter->total_tx_packets += total_tx_packets;
 659         netdev->stats.tx_bytes += total_tx_bytes;
 660         netdev->stats.tx_packets += total_tx_packets;
 661         return (count < tx_ring->count);
 662 }
 663
 664 /**
 665  * e1000_clean_rx_irq_ps - Send received data up the network stack; packet split
 666  * @adapter: board private structure
 667  *
 668  * the return value indicates whether actual cleaning was done, there
 669  * is no guarantee that everything was cleaned
 670  **/
 671 static bool e1000_clean_rx_irq_ps(struct e1000_adapter *adapter,
 672                                   int *work_done, int work_to_do)
 673 {
 674         struct e1000_hw *hw = &adapter->hw;
 675         union e1000_rx_desc_packet_split *rx_desc, *next_rxd;
 676         struct net_device *netdev = adapter->netdev;
 677         struct pci_dev *pdev = adapter->pdev;
 678         struct e1000_ring *rx_ring = adapter->rx_ring;
 679         struct e1000_buffer *buffer_info, *next_buffer;
 680         struct e1000_ps_page *ps_page;
 681         struct sk_buff *skb;
 682         unsigned int i, j;
 683         u32 length, staterr;
 684         int cleaned_count = 0;
 685         bool cleaned = 0;
 686         unsigned int total_rx_bytes = 0, total_rx_packets = 0;
 687
 688         i = rx_ring->next_to_clean;
 689         rx_desc = E1000_RX_DESC_PS(*rx_ring, i);
 690         staterr = le32_to_cpu(rx_desc->wb.middle.status_error);
 691         buffer_info = &rx_ring->buffer_info[i];
 692
 693         while (staterr & E1000_RXD_STAT_DD) {
 694                 if (*work_done >= work_to_do)
 695                         break;
 696                 (*work_done)++;
 697                 skb = buffer_info->skb;
 698
 699                 /* in the packet split case this is header only */
 700                 prefetch(skb->data - NET_IP_ALIGN);
 701
 702                 i++;
 703                 if (i == rx_ring->count)
 704                         i = 0;
 705                 next_rxd = E1000_RX_DESC_PS(*rx_ring, i);
 706                 prefetch(next_rxd);
 707
 708                 next_buffer = &rx_ring->buffer_info[i];
 709
 710                 cleaned = 1;
 711                 cleaned_count++;
 712                 pci_unmap_single(pdev, buffer_info->dma,
 713                                  adapter->rx_ps_bsize0,
 714                                  PCI_DMA_FROMDEVICE);
 715                 buffer_info->dma = 0;
 716
 717                 if (!(staterr & E1000_RXD_STAT_EOP)) {
 718                         e_dbg("Packet Split buffers didn't pick up the full "
 719                               "packet\n");
 720                         dev_kfree_skb_irq(skb);
 721                         goto next_desc;
 722                 }
 723
 724                 if (staterr & E1000_RXDEXT_ERR_FRAME_ERR_MASK) {
 725                         dev_kfree_skb_irq(skb);
 726                         goto next_desc;
 727                 }
 728
 729                 length = le16_to_cpu(rx_desc->wb.middle.length0);
 730
 731                 if (!length) {
 732                         e_dbg("Last part of the packet spanning multiple "
 733                               "descriptors\n");
 734                         dev_kfree_skb_irq(skb);
 735                         goto next_desc;
 736                 }
 737
 738                 /* Good Receive */
 739                 skb_put(skb, length);
 740
 741                 {
 742                 /*
 743                  * this looks ugly, but it seems compiler issues make it
 744                  * more efficient than reusing j
 745                  */
 746                 int l1 = le16_to_cpu(rx_desc->wb.upper.length[0]);
 747
 748                 /*
 749                  * page alloc/put takes too long and effects small packet
 750                  * throughput, so unsplit small packets and save the alloc/put
 751                  * only valid in softirq (napi) context to call kmap_*
 752                  */
 753                 if (l1 && (l1 <= copybreak) &&
 754                     ((length + l1) <= adapter->rx_ps_bsize0)) {
 755                         u8 *vaddr;
 756
 757                         ps_page = &buffer_info->ps_pages[0];
 758
 759                         /*
 760                          * there is no documentation about how to call
 761                          * kmap_atomic, so we can't hold the mapping
 762                          * very long
 763                          */
 764                         pci_dma_sync_single_for_cpu(pdev, ps_page->dma,
 765                                 PAGE_SIZE, PCI_DMA_FROMDEVICE);
 766                         vaddr = kmap_atomic(ps_page->page, KM_SKB_DATA_SOFTIRQ);
 767                         memcpy(skb_tail_pointer(skb), vaddr, l1);
 768                         kunmap_atomic(vaddr, KM_SKB_DATA_SOFTIRQ);
 769                         pci_dma_sync_single_for_device(pdev, ps_page->dma,
 770                                 PAGE_SIZE, PCI_DMA_FROMDEVICE);
 771
 772                         /* remove the CRC */
 773                         if (!(adapter->flags2 & FLAG2_CRC_STRIPPING))
 774                                 l1 -= 4;
 775
 776                         skb_put(skb, l1);
 777                         goto copydone;
 778                 } /* if */
 779                 }
 780
 781                 for (j = 0; j < PS_PAGE_BUFFERS; j++) {
 782                         length = le16_to_cpu(rx_desc->wb.upper.length[j]);
 783                         if (!length)
 784                                 break;
 785
 786                         ps_page = &buffer_info->ps_pages[j];
 787                         pci_unmap_page(pdev, ps_page->dma, PAGE_SIZE,
 788                                        PCI_DMA_FROMDEVICE);
 789                         ps_page->dma = 0;
 790                         skb_fill_page_desc(skb, j, ps_page->page, 0, length);
 791                         ps_page->page = NULL;
 792                         skb->len += length;
 793                         skb->data_len += length;
 794                         skb->truesize += length;
 795                 }
 796
 797                 /* strip the ethernet crc, problem is we're using pages now so
 798                  * this whole operation can get a little cpu intensive
 799                  */
 800                 if (!(adapter->flags2 & FLAG2_CRC_STRIPPING))
 801                         pskb_trim(skb, skb->len - 4);
 802
 803 copydone:
 804                 total_rx_bytes += skb->len;
 805                 total_rx_packets++;
 806
 807                 e1000_rx_checksum(adapter, staterr, le16_to_cpu(
 808                         rx_desc->wb.lower.hi_dword.csum_ip.csum), skb);
 809
 810                 if (rx_desc->wb.upper.header_status &
 811                            cpu_to_le16(E1000_RXDPS_HDRSTAT_HDRSP))
 812                         adapter->rx_hdr_split++;
 813
 814                 e1000_receive_skb(adapter, netdev, skb,
 815                                   staterr, rx_desc->wb.middle.vlan);
 816
 817 next_desc:
 818                 rx_desc->wb.middle.status_error &= cpu_to_le32(~0xFF);
 819                 buffer_info->skb = NULL;
 820
 821                 /* return some buffers to hardware, one at a time is too slow */
 822                 if (cleaned_count >= E1000_RX_BUFFER_WRITE) {
 823                         adapter->alloc_rx_buf(adapter, cleaned_count);
 824                         cleaned_count = 0;
 825                 }
 826
 827                 /* use prefetched values */
 828                 rx_desc = next_rxd;
 829                 buffer_info = next_buffer;
 830
 831                 staterr = le32_to_cpu(rx_desc->wb.middle.status_error);
 832         }
 833         rx_ring->next_to_clean = i;
 834
 835         cleaned_count = e1000_desc_unused(rx_ring);
 836         if (cleaned_count)
 837                 adapter->alloc_rx_buf(adapter, cleaned_count);
 838
 839         adapter->total_rx_bytes += total_rx_bytes;
 840         adapter->total_rx_packets += total_rx_packets;
 841         netdev->stats.rx_bytes += total_rx_bytes;
 842         netdev->stats.rx_packets += total_rx_packets;
 843         return cleaned;
 844 }
 845
 846 /**
 847  * e1000_consume_page - helper function
 848  **/
 849 static void e1000_consume_page(struct e1000_buffer *bi, struct sk_buff *skb,
 850                                u16 length)
 851 {
 852         bi->page = NULL;
 853         skb->len += length;
 854         skb->data_len += length;
 855         skb->truesize += length;
 856 }
 857
 858 /**
 859  * e1000_clean_jumbo_rx_irq - Send received data up the network stack; legacy
 860  * @adapter: board private structure
 861  *
 862  * the return value indicates whether actual cleaning was done, there
 863  * is no guarantee that everything was cleaned
 864  **/
 865
 866 static bool e1000_clean_jumbo_rx_irq(struct e1000_adapter *adapter,
 867                                      int *work_done, int work_to_do)
 868 {
 869         struct net_device *netdev = adapter->netdev;
 870         struct pci_dev *pdev = adapter->pdev;
 871         struct e1000_ring *rx_ring = adapter->rx_ring;
 872         struct e1000_rx_desc *rx_desc, *next_rxd;
 873         struct e1000_buffer *buffer_info, *next_buffer;
 874         u32 length;
 875         unsigned int i;
 876         int cleaned_count = 0;
 877         bool cleaned = false;
 878         unsigned int total_rx_bytes=0, total_rx_packets=0;
 879
 880         i = rx_ring->next_to_clean;
 881         rx_desc = E1000_RX_DESC(*rx_ring, i);
 882         buffer_info = &rx_ring->buffer_info[i];
 883
 884         while (rx_desc->status & E1000_RXD_STAT_DD) {
 885                 struct sk_buff *skb;
 886                 u8 status;
 887
 888                 if (*work_done >= work_to_do)
 889                         break;
 890                 (*work_done)++;
 891
 892                 status = rx_desc->status;
 893                 skb = buffer_info->skb;
 894                 buffer_info->skb = NULL;
 895
 896                 ++i;
 897                 if (i == rx_ring->count)
 898                         i = 0;
 899                 next_rxd = E1000_RX_DESC(*rx_ring, i);
 900                 prefetch(next_rxd);
 901
 902                 next_buffer = &rx_ring->buffer_info[i];
 903
 904                 cleaned = true;
 905                 cleaned_count++;
 906                 pci_unmap_page(pdev, buffer_info->dma, PAGE_SIZE,
 907                                PCI_DMA_FROMDEVICE);
 908                 buffer_info->dma = 0;
 909
 910                 length = le16_to_cpu(rx_desc->length);
 911
 912                 /* errors is only valid for DD + EOP descriptors */
 913                 if (unlikely((status & E1000_RXD_STAT_EOP) &&
 914                     (rx_desc->errors & E1000_RXD_ERR_FRAME_ERR_MASK))) {
 915                                 /* recycle both page and skb */
 916                                 buffer_info->skb = skb;
 917                                 /* an error means any chain goes out the window
 918                                  * too */
 919                                 if (rx_ring->rx_skb_top)
 920                                         dev_kfree_skb(rx_ring->rx_skb_top);
 921                                 rx_ring->rx_skb_top = NULL;
 922                                 goto next_desc;
 923                 }
 924
 925 #define rxtop rx_ring->rx_skb_top
 926                 if (!(status & E1000_RXD_STAT_EOP)) {
 927                         /* this descriptor is only the beginning (or middle) */
 928                         if (!rxtop) {
 929                                 /* this is the beginning of a chain */
 930                                 rxtop = skb;
 931                                 skb_fill_page_desc(rxtop, 0, buffer_info->page,
 932                                                    0, length);
 933                         } else {
 934                                 /* this is the middle of a chain */
 935                                 skb_fill_page_desc(rxtop,
 936                                     skb_shinfo(rxtop)->nr_frags,
 937                                     buffer_info->page, 0, length);
 938                                 /* re-use the skb, only consumed the page */
 939                                 buffer_info->skb = skb;
 940                         }
 941                         e1000_consume_page(buffer_info, rxtop, length);
 942                         goto next_desc;
 943                 } else {
 944                         if (rxtop) {
 945                                 /* end of the chain */
 946                                 skb_fill_page_desc(rxtop,
 947                                     skb_shinfo(rxtop)->nr_frags,
 948                                     buffer_info->page, 0, length);
 949                                 /* re-use the current skb, we only consumed the
 950                                  * page */
 951                                 buffer_info->skb = skb;
 952                                 skb = rxtop;
 953                                 rxtop = NULL;
 954                                 e1000_consume_page(buffer_info, skb, length);
 955                         } else {
 956                                 /* no chain, got EOP, this buf is the packet
 957                                  * copybreak to save the put_page/alloc_page */
 958                                 if (length <= copybreak &&
 959                                     skb_tailroom(skb) >= length) {
 960                                         u8 *vaddr;
 961                                         vaddr = kmap_atomic(buffer_info->page,
 962                                                            KM_SKB_DATA_SOFTIRQ);
 963                                         memcpy(skb_tail_pointer(skb), vaddr,
 964                                                length);
 965                                         kunmap_atomic(vaddr,
 966                                                       KM_SKB_DATA_SOFTIRQ);
 967                                         /* re-use the page, so don't erase
 968                                          * buffer_info->page */
 969                                         skb_put(skb, length);
 970                                 } else {
 971                                         skb_fill_page_desc(skb, 0,
 972                                                            buffer_info->page, 0,
 973                                                            length);
 974                                         e1000_consume_page(buffer_info, skb,
 975                                                            length);
 976                                 }
 977                         }
 978                 }
 979
 980                 /* Receive Checksum Offload XXX recompute due to CRC strip? */
 981                 e1000_rx_checksum(adapter,
 982                                   (u32)(status) |
 983                                   ((u32)(rx_desc->errors) << 24),
 984                                   le16_to_cpu(rx_desc->csum), skb);
 985
 986                 /* probably a little skewed due to removing CRC */
 987                 total_rx_bytes += skb->len;
 988                 total_rx_packets++;
 989
 990                 /* eth type trans needs skb->data to point to something */
 991                 if (!pskb_may_pull(skb, ETH_HLEN)) {
 992                         e_err("pskb_may_pull failed.\n");
 993                         dev_kfree_skb(skb);
 994                         goto next_desc;
 995                 }
 996
 997                 e1000_receive_skb(adapter, netdev, skb, status,
 998                                   rx_desc->special);
 999
1000 next_desc:
1001                 rx_desc->status = 0;
1002
1003                 /* return some buffers to hardware, one at a time is too slow */
1004                 if (unlikely(cleaned_count >= E1000_RX_BUFFER_WRITE)) {
1005                         adapter->alloc_rx_buf(adapter, cleaned_count);
1006                         cleaned_count = 0;
1007                 }
1008
1009                 /* use prefetched values */
1010                 rx_desc = next_rxd;
1011                 buffer_info = next_buffer;
1012         }
1013         rx_ring->next_to_clean = i;
1014
1015         cleaned_count = e1000_desc_unused(rx_ring);
1016         if (cleaned_count)
1017                 adapter->alloc_rx_buf(adapter, cleaned_count);
1018
1019         adapter->total_rx_bytes += total_rx_bytes;
1020         adapter->total_rx_packets += total_rx_packets;
1021         netdev->stats.rx_bytes += total_rx_bytes;
1022         netdev->stats.rx_packets += total_rx_packets;
1023         return cleaned;
1024 }
1025
1026 /**
1027  * e1000_clean_rx_ring - Free Rx Buffers per Queue
1028  * @adapter: board private structure
1029  **/
1030 static void e1000_clean_rx_ring(struct e1000_adapter *adapter)
1031 {
1032         struct e1000_ring *rx_ring = adapter->rx_ring;
1033         struct e1000_buffer *buffer_info;
1034         struct e1000_ps_page *ps_page;
1035         struct pci_dev *pdev = adapter->pdev;
1036         unsigned int i, j;
1037
1038         /* Free all the Rx ring sk_buffs */
1039         for (i = 0; i < rx_ring->count; i++) {
1040                 buffer_info = &rx_ring->buffer_info[i];
1041                 if (buffer_info->dma) {
1042                         if (adapter->clean_rx == e1000_clean_rx_irq)
1043                                 pci_unmap_single(pdev, buffer_info->dma,
1044                                                  adapter->rx_buffer_len,
1045                                                  PCI_DMA_FROMDEVICE);
1046                         else if (adapter->clean_rx == e1000_clean_jumbo_rx_irq)
1047                                 pci_unmap_page(pdev, buffer_info->dma,
1048                                                PAGE_SIZE,
1049                                                PCI_DMA_FROMDEVICE);
1050                         else if (adapter->clean_rx == e1000_clean_rx_irq_ps)
1051                                 pci_unmap_single(pdev, buffer_info->dma,
1052                                                  adapter->rx_ps_bsize0,
1053                                                  PCI_DMA_FROMDEVICE);
1054                         buffer_info->dma = 0;
1055                 }
1056
1057                 if (buffer_info->page) {
1058                         put_page(buffer_info->page);
1059                         buffer_info->page = NULL;
1060                 }
1061
1062                 if (buffer_info->skb) {
1063                         dev_kfree_skb(buffer_info->skb);
1064                         buffer_info->skb = NULL;
1065                 }
1066
1067                 for (j = 0; j < PS_PAGE_BUFFERS; j++) {
1068                         ps_page = &buffer_info->ps_pages[j];
1069                         if (!ps_page->page)
1070                                 break;
1071                         pci_unmap_page(pdev, ps_page->dma, PAGE_SIZE,
1072                                        PCI_DMA_FROMDEVICE);
1073                         ps_page->dma = 0;
1074                         put_page(ps_page->page);
1075                         ps_page->page = NULL;
1076                 }
1077         }
1078
1079         /* there also may be some cached data from a chained receive */
1080         if (rx_ring->rx_skb_top) {
1081                 dev_kfree_skb(rx_ring->rx_skb_top);
1082                 rx_ring->rx_skb_top = NULL;
1083         }
1084
1085         /* Zero out the descriptor ring */
1086         memset(rx_ring->desc, 0, rx_ring->size);
1087
1088         rx_ring->next_to_clean = 0;
1089         rx_ring->next_to_use = 0;
1090
1091         writel(0, adapter->hw.hw_addr + rx_ring->head);
1092         writel(0, adapter->hw.hw_addr + rx_ring->tail);
1093 }
1094
1095 static void e1000e_downshift_workaround(struct work_struct *work)
1096 {
1097         struct e1000_adapter *adapter = container_of(work,
1098                                         struct e1000_adapter, downshift_task);
1099
1100         e1000e_gig_downshift_workaround_ich8lan(&adapter->hw);
1101 }
1102
1103 /**
1104  * e1000_intr_msi - Interrupt Handler
1105  * @irq: interrupt number
1106  * @data: pointer to a network interface device structure
1107  **/
1108 static irqreturn_t e1000_intr_msi(int irq, void *data)
1109 {
1110         struct net_device *netdev = data;
1111         struct e1000_adapter *adapter = netdev_priv(netdev);
1112         struct e1000_hw *hw = &adapter->hw;
1113         u32 icr = er32(ICR);
1114
1115         /*
1116          * read ICR disables interrupts using IAM
1117          */
1118
1119         if (icr & E1000_ICR_LSC) {
1120                 hw->mac.get_link_status = 1;
1121                 /*
1122                  * ICH8 workaround-- Call gig speed drop workaround on cable
1123                  * disconnect (LSC) before accessing any PHY registers
1124                  */
1125                 if ((adapter->flags & FLAG_LSC_GIG_SPEED_DROP) &&
1126                     (!(er32(STATUS) & E1000_STATUS_LU)))
1127                         schedule_work(&adapter->downshift_task);
1128
1129                 /*
1130                  * 80003ES2LAN workaround-- For packet buffer work-around on
1131                  * link down event; disable receives here in the ISR and reset
1132                  * adapter in watchdog
1133                  */
1134                 if (netif_carrier_ok(netdev) &&
1135                     adapter->flags & FLAG_RX_NEEDS_RESTART) {
1136                         /* disable receives */
1137                         u32 rctl = er32(RCTL);
1138                         ew32(RCTL, rctl & ~E1000_RCTL_EN);
1139                         adapter->flags |= FLAG_RX_RESTART_NOW;
1140                 }
1141                 /* guard against interrupt when we're going down */
1142                 if (!test_bit(__E1000_DOWN, &adapter->state))
1143                         mod_timer(&adapter->watchdog_timer, jiffies + 1);
1144         }
1145
1146         if (napi_schedule_prep(&adapter->napi)) {
1147                 adapter->total_tx_bytes = 0;
1148                 adapter->total_tx_packets = 0;
1149                 adapter->total_rx_bytes = 0;
1150                 adapter->total_rx_packets = 0;
1151                 __napi_schedule(&adapter->napi);
1152         }
1153
1154         return IRQ_HANDLED;
1155 }
1156
1157 /**
1158  * e1000_intr - Interrupt Handler
1159  * @irq: interrupt number
1160  * @data: pointer to a network interface device structure
1161  **/
1162 static irqreturn_t e1000_intr(int irq, void *data)
1163 {
1164         struct net_device *netdev = data;
1165         struct e1000_adapter *adapter = netdev_priv(netdev);
1166         struct e1000_hw *hw = &adapter->hw;
1167         u32 rctl, icr = er32(ICR);
1168
1169         if (!icr || test_bit(__E1000_DOWN, &adapter->state))
1170                 return IRQ_NONE;  /* Not our interrupt */
1171
1172         /*
1173          * IMS will not auto-mask if INT_ASSERTED is not set, and if it is
1174          * not set, then the adapter didn't send an interrupt
1175          */
1176         if (!(icr & E1000_ICR_INT_ASSERTED))
1177                 return IRQ_NONE;
1178
1179         /*
1180          * Interrupt Auto-Mask...upon reading ICR,
1181          * interrupts are masked.  No need for the
1182          * IMC write
1183          */
1184
1185         if (icr & E1000_ICR_LSC) {
1186                 hw->mac.get_link_status = 1;
1187                 /*
1188                  * ICH8 workaround-- Call gig speed drop workaround on cable
1189                  * disconnect (LSC) before accessing any PHY registers
1190                  */
1191                 if ((adapter->flags & FLAG_LSC_GIG_SPEED_DROP) &&
1192                     (!(er32(STATUS) & E1000_STATUS_LU)))
1193                         schedule_work(&adapter->downshift_task);
1194
1195                 /*
1196                  * 80003ES2LAN workaround--
1197                  * For packet buffer work-around on link down event;
1198                  * disable receives here in the ISR and
1199                  * reset adapter in watchdog
1200                  */
1201                 if (netif_carrier_ok(netdev) &&
1202                     (adapter->flags & FLAG_RX_NEEDS_RESTART)) {
1203                         /* disable receives */
1204                         rctl = er32(RCTL);
1205                         ew32(RCTL, rctl & ~E1000_RCTL_EN);
1206                         adapter->flags |= FLAG_RX_RESTART_NOW;
1207                 }
1208                 /* guard against interrupt when we're going down */
1209                 if (!test_bit(__E1000_DOWN, &adapter->state))
1210                         mod_timer(&adapter->watchdog_timer, jiffies + 1);
1211         }
1212
1213         if (napi_schedule_prep(&adapter->napi)) {
1214                 adapter->total_tx_bytes = 0;
1215                 adapter->total_tx_packets = 0;
1216                 adapter->total_rx_bytes = 0;
1217                 adapter->total_rx_packets = 0;
1218                 __napi_schedule(&adapter->napi);
1219         }
1220
1221         return IRQ_HANDLED;
1222 }
1223
1224 static irqreturn_t e1000_msix_other(int irq, void *data)
1225 {
1226         struct net_device *netdev = data;
1227         struct e1000_adapter *adapter = netdev_priv(netdev);
1228         struct e1000_hw *hw = &adapter->hw;
1229         u32 icr = er32(ICR);
1230
1231         if (!(icr & E1000_ICR_INT_ASSERTED)) {
1232                 if (!test_bit(__E1000_DOWN, &adapter->state))
1233                         ew32(IMS, E1000_IMS_OTHER);
1234                 return IRQ_NONE;
1235         }
1236
1237         if (icr & adapter->eiac_mask)
1238                 ew32(ICS, (icr & adapter->eiac_mask));
1239
1240         if (icr & E1000_ICR_OTHER) {
1241                 if (!(icr & E1000_ICR_LSC))
1242                         goto no_link_interrupt;
1243                 hw->mac.get_link_status = 1;
1244                 /* guard against interrupt when we're going down */
1245                 if (!test_bit(__E1000_DOWN, &adapter->state))
1246                         mod_timer(&adapter->watchdog_timer, jiffies + 1);
1247         }
1248
1249 no_link_interrupt:
1250         if (!test_bit(__E1000_DOWN, &adapter->state))
1251                 ew32(IMS, E1000_IMS_LSC | E1000_IMS_OTHER);
1252
1253         return IRQ_HANDLED;
1254 }
1255
1256
1257 static irqreturn_t e1000_intr_msix_tx(int irq, void *data)
1258 {
1259         struct net_device *netdev = data;
1260         struct e1000_adapter *adapter = netdev_priv(netdev);
1261         struct e1000_hw *hw = &adapter->hw;
1262         struct e1000_ring *tx_ring = adapter->tx_ring;
1263
1264
1265         adapter->total_tx_bytes = 0;
1266         adapter->total_tx_packets = 0;
1267
1268         if (!e1000_clean_tx_irq(adapter))
1269                 /* Ring was not completely cleaned, so fire another interrupt */
1270                 ew32(ICS, tx_ring->ims_val);
1271
1272         return IRQ_HANDLED;
1273 }
1274
1275 static irqreturn_t e1000_intr_msix_rx(int irq, void *data)
1276 {
1277         struct net_device *netdev = data;
1278         struct e1000_adapter *adapter = netdev_priv(netdev);
1279
1280         /* Write the ITR value calculated at the end of the
1281          * previous interrupt.
1282          */
1283         if (adapter->rx_ring->set_itr) {
1284                 writel(1000000000 / (adapter->rx_ring->itr_val * 256),
1285                        adapter->hw.hw_addr + adapter->rx_ring->itr_register);
1286                 adapter->rx_ring->set_itr = 0;
1287         }
1288
1289         if (napi_schedule_prep(&adapter->napi)) {
1290                 adapter->total_rx_bytes = 0;
1291                 adapter->total_rx_packets = 0;
1292                 __napi_schedule(&adapter->napi);
1293         }
1294         return IRQ_HANDLED;
1295 }
1296
1297 /**
1298  * e1000_configure_msix - Configure MSI-X hardware
1299  *
1300  * e1000_configure_msix sets up the hardware to properly
1301  * generate MSI-X interrupts.
1302  **/
1303 static void e1000_configure_msix(struct e1000_adapter *adapter)
1304 {
1305         struct e1000_hw *hw = &adapter->hw;
1306         struct e1000_ring *rx_ring = adapter->rx_ring;
1307         struct e1000_ring *tx_ring = adapter->tx_ring;
1308         int vector = 0;
1309         u32 ctrl_ext, ivar = 0;
1310
1311         adapter->eiac_mask = 0;
1312
1313         /* Workaround issue with spurious interrupts on 82574 in MSI-X mode */
1314         if (hw->mac.type == e1000_82574) {
1315                 u32 rfctl = er32(RFCTL);
1316                 rfctl |= E1000_RFCTL_ACK_DIS;
1317                 ew32(RFCTL, rfctl);
1318         }
1319
1320 #define E1000_IVAR_INT_ALLOC_VALID      0x8
1321         /* Configure Rx vector */
1322         rx_ring->ims_val = E1000_IMS_RXQ0;
1323         adapter->eiac_mask |= rx_ring->ims_val;
1324         if (rx_ring->itr_val)
1325                 writel(1000000000 / (rx_ring->itr_val * 256),
1326                        hw->hw_addr + rx_ring->itr_register);
1327         else
1328                 writel(1, hw->hw_addr + rx_ring->itr_register);
1329         ivar = E1000_IVAR_INT_ALLOC_VALID | vector;
1330
1331         /* Configure Tx vector */
1332         tx_ring->ims_val = E1000_IMS_TXQ0;
1333         vector++;
1334         if (tx_ring->itr_val)
1335                 writel(1000000000 / (tx_ring->itr_val * 256),
1336                        hw->hw_addr + tx_ring->itr_register);
1337         else
1338                 writel(1, hw->hw_addr + tx_ring->itr_register);
1339         adapter->eiac_mask |= tx_ring->ims_val;
1340         ivar |= ((E1000_IVAR_INT_ALLOC_VALID | vector) << 8);
1341
1342         /* set vector for Other Causes, e.g. link changes */
1343         vector++;
1344         ivar |= ((E1000_IVAR_INT_ALLOC_VALID | vector) << 16);
1345         if (rx_ring->itr_val)
1346                 writel(1000000000 / (rx_ring->itr_val * 256),
1347                        hw->hw_addr + E1000_EITR_82574(vector));
1348         else
1349                 writel(1, hw->hw_addr + E1000_EITR_82574(vector));
1350
1351         /* Cause Tx interrupts on every write back */
1352         ivar |= (1 << 31);
1353
1354         ew32(IVAR, ivar);
1355
1356         /* enable MSI-X PBA support */
1357         ctrl_ext = er32(CTRL_EXT);
1358         ctrl_ext |= E1000_CTRL_EXT_PBA_CLR;
1359
1360         /* Auto-Mask Other interrupts upon ICR read */
1361 #define E1000_EIAC_MASK_82574   0x01F00000
1362         ew32(IAM, ~E1000_EIAC_MASK_82574 | E1000_IMS_OTHER);
1363         ctrl_ext |= E1000_CTRL_EXT_EIAME;
1364         ew32(CTRL_EXT, ctrl_ext);
1365         e1e_flush();
1366 }
1367
1368 void e1000e_reset_interrupt_capability(struct e1000_adapter *adapter)
1369 {
1370         if (adapter->msix_entries) {
1371                 pci_disable_msix(adapter->pdev);
1372                 kfree(adapter->msix_entries);
1373                 adapter->msix_entries = NULL;
1374         } else if (adapter->flags & FLAG_MSI_ENABLED) {
1375                 pci_disable_msi(adapter->pdev);
1376                 adapter->flags &= ~FLAG_MSI_ENABLED;
1377         }
1378
1379         return;
1380 }
1381
1382 /**
1383  * e1000e_set_interrupt_capability - set MSI or MSI-X if supported
1384  *
1385  * Attempt to configure interrupts using the best available
1386  * capabilities of the hardware and kernel.
1387  **/
1388 void e1000e_set_interrupt_capability(struct e1000_adapter *adapter)
1389 {
1390         int err;
1391         int numvecs, i;
1392
1393
1394         switch (adapter->int_mode) {
1395         case E1000E_INT_MODE_MSIX:
1396                 if (adapter->flags & FLAG_HAS_MSIX) {
1397                         numvecs = 3; /* RxQ0, TxQ0 and other */
1398                         adapter->msix_entries = kcalloc(numvecs,
1399                                                       sizeof(struct msix_entry),
1400                                                       GFP_KERNEL);
1401                         if (adapter->msix_entries) {
1402                                 for (i = 0; i < numvecs; i++)
1403                                         adapter->msix_entries[i].entry = i;
1404
1405                                 err = pci_enable_msix(adapter->pdev,
1406                                                       adapter->msix_entries,
1407                                                       numvecs);
1408                                 if (err == 0)
1409                                         return;
1410                         }
1411                         /* MSI-X failed, so fall through and try MSI */
1412                         e_err("Failed to initialize MSI-X interrupts.  "
1413                               "Falling back to MSI interrupts.\n");
1414                         e1000e_reset_interrupt_capability(adapter);
1415                 }
1416                 adapter->int_mode = E1000E_INT_MODE_MSI;
1417                 /* Fall through */
1418         case E1000E_INT_MODE_MSI:
1419                 if (!pci_enable_msi(adapter->pdev)) {
1420                         adapter->flags |= FLAG_MSI_ENABLED;
1421                 } else {
1422                         adapter->int_mode = E1000E_INT_MODE_LEGACY;
1423                         e_err("Failed to initialize MSI interrupts.  Falling "
1424                               "back to legacy interrupts.\n");
1425                 }
1426                 /* Fall through */
1427         case E1000E_INT_MODE_LEGACY:
1428                 /* Don't do anything; this is the system default */
1429                 break;
1430         }
1431
1432         return;
1433 }
1434
1435 /**
1436  * e1000_request_msix - Initialize MSI-X interrupts
1437  *
1438  * e1000_request_msix allocates MSI-X vectors and requests interrupts from the
1439  * kernel.
1440  **/
1441 static int e1000_request_msix(struct e1000_adapter *adapter)
1442 {
1443         struct net_device *netdev = adapter->netdev;
1444         int err = 0, vector = 0;
1445
1446         if (strlen(netdev->name) < (IFNAMSIZ - 5))
1447                 sprintf(adapter->rx_ring->name, "%s-rx-0", netdev->name);
1448         else
1449                 memcpy(adapter->rx_ring->name, netdev->name, IFNAMSIZ);
1450         err = request_irq(adapter->msix_entries[vector].vector,
1451                           e1000_intr_msix_rx, 0, adapter->rx_ring->name,
1452                           netdev);
1453         if (err)
1454                 goto out;
1455         adapter->rx_ring->itr_register = E1000_EITR_82574(vector);
1456         adapter->rx_ring->itr_val = adapter->itr;
1457         vector++;
1458
1459         if (strlen(netdev->name) < (IFNAMSIZ - 5))
1460                 sprintf(adapter->tx_ring->name, "%s-tx-0", netdev->name);
1461         else
1462                 memcpy(adapter->tx_ring->name, netdev->name, IFNAMSIZ);
1463         err = request_irq(adapter->msix_entries[vector].vector,
1464                           e1000_intr_msix_tx, 0, adapter->tx_ring->name,
1465                           netdev);
1466         if (err)
1467                 goto out;
1468         adapter->tx_ring->itr_register = E1000_EITR_82574(vector);
1469         adapter->tx_ring->itr_val = adapter->itr;
1470         vector++;
1471
1472         err = request_irq(adapter->msix_entries[vector].vector,
1473                           e1000_msix_other, 0, netdev->name, netdev);
1474         if (err)
1475                 goto out;
1476
1477         e1000_configure_msix(adapter);
1478         return 0;
1479 out:
1480         return err;
1481 }
1482
1483 /**
1484  * e1000_request_irq - initialize interrupts
1485  *
1486  * Attempts to configure interrupts using the best available
1487  * capabilities of the hardware and kernel.
1488  **/
1489 static int e1000_request_irq(struct e1000_adapter *adapter)
1490 {
1491         struct net_device *netdev = adapter->netdev;
1492         int err;
1493
1494         if (adapter->msix_entries) {
1495                 err = e1000_request_msix(adapter);
1496                 if (!err)
1497                         return err;
1498                 /* fall back to MSI */
1499                 e1000e_reset_interrupt_capability(adapter);
1500                 adapter->int_mode = E1000E_INT_MODE_MSI;
1501                 e1000e_set_interrupt_capability(adapter);
1502         }
1503         if (adapter->flags & FLAG_MSI_ENABLED) {
1504                 err = request_irq(adapter->pdev->irq, e1000_intr_msi, 0,
1505                                   netdev->name, netdev);
1506                 if (!err)
1507                         return err;
1508
1509                 /* fall back to legacy interrupt */
1510                 e1000e_reset_interrupt_capability(adapter);
1511                 adapter->int_mode = E1000E_INT_MODE_LEGACY;
1512         }
1513
1514         err = request_irq(adapter->pdev->irq, e1000_intr, IRQF_SHARED,
1515                           netdev->name, netdev);
1516         if (err)
1517                 e_err("Unable to allocate interrupt, Error: %d\n", err);
1518
1519         return err;
1520 }
1521
1522 static void e1000_free_irq(struct e1000_adapter *adapter)
1523 {
1524         struct net_device *netdev = adapter->netdev;
1525
1526         if (adapter->msix_entries) {
1527                 int vector = 0;
1528
1529                 free_irq(adapter->msix_entries[vector].vector, netdev);
1530                 vector++;
1531
1532                 free_irq(adapter->msix_entries[vector].vector, netdev);
1533                 vector++;
1534
1535                 /* Other Causes interrupt vector */
1536                 free_irq(adapter->msix_entries[vector].vector, netdev);
1537                 return;
1538         }
1539
1540         free_irq(adapter->pdev->irq, netdev);
1541 }
1542
1543 /**
1544  * e1000_irq_disable - Mask off interrupt generation on the NIC
1545  **/
1546 static void e1000_irq_disable(struct e1000_adapter *adapter)
1547 {
1548         struct e1000_hw *hw = &adapter->hw;
1549
1550         ew32(IMC, ~0);
1551         if (adapter->msix_entries)
1552                 ew32(EIAC_82574, 0);
1553         e1e_flush();
1554         synchronize_irq(adapter->pdev->irq);
1555 }
1556
1557 /**
1558  * e1000_irq_enable - Enable default interrupt generation settings
1559  **/
1560 static void e1000_irq_enable(struct e1000_adapter *adapter)
1561 {
1562         struct e1000_hw *hw = &adapter->hw;
1563
1564         if (adapter->msix_entries) {
1565                 ew32(EIAC_82574, adapter->eiac_mask & E1000_EIAC_MASK_82574);
1566                 ew32(IMS, adapter->eiac_mask | E1000_IMS_OTHER | E1000_IMS_LSC);
1567         } else {
1568                 ew32(IMS, IMS_ENABLE_MASK);
1569         }
1570         e1e_flush();
1571 }
1572
1573 /**
1574  * e1000_get_hw_control - get control of the h/w from f/w
1575  * @adapter: address of board private structure
1576  *
1577  * e1000_get_hw_control sets {CTRL_EXT|SWSM}:DRV_LOAD bit.
1578  * For ASF and Pass Through versions of f/w this means that
1579  * the driver is loaded. For AMT version (only with 82573)
1580  * of the f/w this means that the network i/f is open.
1581  **/
1582 static void e1000_get_hw_control(struct e1000_adapter *adapter)
1583 {
1584         struct e1000_hw *hw = &adapter->hw;
1585         u32 ctrl_ext;
1586         u32 swsm;
1587
1588         /* Let firmware know the driver has taken over */
1589         if (adapter->flags & FLAG_HAS_SWSM_ON_LOAD) {
1590                 swsm = er32(SWSM);
1591                 ew32(SWSM, swsm | E1000_SWSM_DRV_LOAD);
1592         } else if (adapter->flags & FLAG_HAS_CTRLEXT_ON_LOAD) {
1593                 ctrl_ext = er32(CTRL_EXT);
1594                 ew32(CTRL_EXT, ctrl_ext | E1000_CTRL_EXT_DRV_LOAD);
1595         }
1596 }
1597
1598 /**
1599  * e1000_release_hw_control - release control of the h/w to f/w
1600  * @adapter: address of board private structure
1601  *
1602  * e1000_release_hw_control resets {CTRL_EXT|SWSM}:DRV_LOAD bit.
1603  * For ASF and Pass Through versions of f/w this means that the
1604  * driver is no longer loaded. For AMT version (only with 82573) i
1605  * of the f/w this means that the network i/f is closed.
1606  *
1607  **/
1608 static void e1000_release_hw_control(struct e1000_adapter *adapter)
1609 {
1610         struct e1000_hw *hw = &adapter->hw;
1611         u32 ctrl_ext;
1612         u32 swsm;
1613
1614         /* Let firmware taken over control of h/w */
1615         if (adapter->flags & FLAG_HAS_SWSM_ON_LOAD) {
1616                 swsm = er32(SWSM);
1617                 ew32(SWSM, swsm & ~E1000_SWSM_DRV_LOAD);
1618         } else if (adapter->flags & FLAG_HAS_CTRLEXT_ON_LOAD) {
1619                 ctrl_ext = er32(CTRL_EXT);
1620                 ew32(CTRL_EXT, ctrl_ext & ~E1000_CTRL_EXT_DRV_LOAD);
1621         }
1622 }
1623
1624 /**
1625  * @e1000_alloc_ring - allocate memory for a ring structure
1626  **/
1627 static int e1000_alloc_ring_dma(struct e1000_adapter *adapter,
1628                                 struct e1000_ring *ring)
1629 {
1630         struct pci_dev *pdev = adapter->pdev;
1631
1632         ring->desc = dma_alloc_coherent(&pdev->dev, ring->size, &ring->dma,
1633                                         GFP_KERNEL);
1634         if (!ring->desc)
1635                 return -ENOMEM;
1636
1637         return 0;
1638 }
1639
1640 /**
1641  * e1000e_setup_tx_resources - allocate Tx resources (Descriptors)
1642  * @adapter: board private structure
1643  *
1644  * Return 0 on success, negative on failure
1645  **/
1646 int e1000e_setup_tx_resources(struct e1000_adapter *adapter)
1647 {
1648         struct e1000_ring *tx_ring = adapter->tx_ring;
1649         int err = -ENOMEM, size;
1650
1651         size = sizeof(struct e1000_buffer) * tx_ring->count;
1652         tx_ring->buffer_info = vmalloc(size);
1653         if (!tx_ring->buffer_info)
1654                 goto err;
1655         memset(tx_ring->buffer_info, 0, size);
1656
1657         /* round up to nearest 4K */
1658         tx_ring->size = tx_ring->count * sizeof(struct e1000_tx_desc);
1659         tx_ring->size = ALIGN(tx_ring->size, 4096);
1660
1661         err = e1000_alloc_ring_dma(adapter, tx_ring);
1662         if (err)
1663                 goto err;
1664
1665         tx_ring->next_to_use = 0;
1666         tx_ring->next_to_clean = 0;
1667
1668         return 0;
1669 err:
1670         vfree(tx_ring->buffer_info);
1671         e_err("Unable to allocate memory for the transmit descriptor ring\n");
1672         return err;
1673 }
1674
1675 /**
1676  * e1000e_setup_rx_resources - allocate Rx resources (Descriptors)
1677  * @adapter: board private structure
1678  *
1679  * Returns 0 on success, negative on failure
1680  **/
1681 int e1000e_setup_rx_resources(struct e1000_adapter *adapter)
1682 {
1683         struct e1000_ring *rx_ring = adapter->rx_ring;
1684         struct e1000_buffer *buffer_info;
1685         int i, size, desc_len, err = -ENOMEM;
1686
1687         size = sizeof(struct e1000_buffer) * rx_ring->count;
1688         rx_ring->buffer_info = vmalloc(size);
1689         if (!rx_ring->buffer_info)
1690                 goto err;
1691         memset(rx_ring->buffer_info, 0, size);
1692
1693         for (i = 0; i < rx_ring->count; i++) {
1694                 buffer_info = &rx_ring->buffer_info[i];
1695                 buffer_info->ps_pages = kcalloc(PS_PAGE_BUFFERS,
1696                                                 sizeof(struct e1000_ps_page),
1697                                                 GFP_KERNEL);
1698                 if (!buffer_info->ps_pages)
1699                         goto err_pages;
1700         }
1701
1702         desc_len = sizeof(union e1000_rx_desc_packet_split);
1703
1704         /* Round up to nearest 4K */
1705         rx_ring->size = rx_ring->count * desc_len;
1706         rx_ring->size = ALIGN(rx_ring->size, 4096);
1707
1708         err = e1000_alloc_ring_dma(adapter, rx_ring);
1709         if (err)
1710                 goto err_pages;
1711
1712         rx_ring->next_to_clean = 0;
1713         rx_ring->next_to_use = 0;
1714         rx_ring->rx_skb_top = NULL;
1715
1716         return 0;
1717
1718 err_pages:
1719         for (i = 0; i < rx_ring->count; i++) {
1720                 buffer_info = &rx_ring->buffer_info[i];
1721                 kfree(buffer_info->ps_pages);
1722         }
1723 err:
1724         vfree(rx_ring->buffer_info);
1725         e_err("Unable to allocate memory for the transmit descriptor ring\n");
1726         return err;
1727 }
1728
1729 /**
1730  * e1000_clean_tx_ring - Free Tx Buffers
1731  * @adapter: board private structure
1732  **/
1733 static void e1000_clean_tx_ring(struct e1000_adapter *adapter)
1734 {
1735         struct e1000_ring *tx_ring = adapter->tx_ring;
1736         struct e1000_buffer *buffer_info;
1737         unsigned long size;
1738         unsigned int i;
1739
1740         for (i = 0; i < tx_ring->count; i++) {
1741                 buffer_info = &tx_ring->buffer_info[i];
1742                 e1000_put_txbuf(adapter, buffer_info);
1743         }
1744
1745         size = sizeof(struct e1000_buffer) * tx_ring->count;
1746         memset(tx_ring->buffer_info, 0, size);
1747
1748         memset(tx_ring->desc, 0, tx_ring->size);
1749
1750         tx_ring->next_to_use = 0;
1751         tx_ring->next_to_clean = 0;
1752
1753         writel(0, adapter->hw.hw_addr + tx_ring->head);
1754         writel(0, adapter->hw.hw_addr + tx_ring->tail);
1755 }
1756
1757 /**
1758  * e1000e_free_tx_resources - Free Tx Resources per Queue
1759  * @adapter: board private structure
1760  *
1761  * Free all transmit software resources
1762  **/
1763 void e1000e_free_tx_resources(struct e1000_adapter *adapter)
1764 {
1765         struct pci_dev *pdev = adapter->pdev;
1766         struct e1000_ring *tx_ring = adapter->tx_ring;
1767
1768         e1000_clean_tx_ring(adapter);
1769
1770         vfree(tx_ring->buffer_info);
1771         tx_ring->buffer_info = NULL;
1772
1773         dma_free_coherent(&pdev->dev, tx_ring->size, tx_ring->desc,
1774                           tx_ring->dma);
1775         tx_ring->desc = NULL;
1776 }
1777
1778 /**
1779  * e1000e_free_rx_resources - Free Rx Resources
1780  * @adapter: board private structure
1781  *
1782  * Free all receive software resources
1783  **/
1784
1785 void e1000e_free_rx_resources(struct e1000_adapter *adapter)
1786 {
1787         struct pci_dev *pdev = adapter->pdev;
1788         struct e1000_ring *rx_ring = adapter->rx_ring;
1789         int i;
1790
1791         e1000_clean_rx_ring(adapter);
1792
1793         for (i = 0; i < rx_ring->count; i++) {
1794                 kfree(rx_ring->buffer_info[i].ps_pages);
1795         }
1796
1797         vfree(rx_ring->buffer_info);
1798         rx_ring->buffer_info = NULL;
1799
1800         dma_free_coherent(&pdev->dev, rx_ring->size, rx_ring->desc,
1801                           rx_ring->dma);
1802         rx_ring->desc = NULL;
1803 }
1804
1805 /**
1806  * e1000_update_itr - update the dynamic ITR value based on statistics
1807  * @adapter: pointer to adapter
1808  * @itr_setting: current adapter->itr
1809  * @packets: the number of packets during this measurement interval
1810  * @bytes: the number of bytes during this measurement interval
1811  *
1812  *      Stores a new ITR value based on packets and byte
1813  *      counts during the last interrupt.  The advantage of per interrupt
1814  *      computation is faster updates and more accurate ITR for the current
1815  *      traffic pattern.  Constants in this function were computed
1816  *      based on theoretical maximum wire speed and thresholds were set based
1817  *      on testing data as well as attempting to minimize response time
1818  *      while increasing bulk throughput.  This functionality is controlled
1819  *      by the InterruptThrottleRate module parameter.
1820  **/
1821 static unsigned int e1000_update_itr(struct e1000_adapter *adapter,
1822                                      u16 itr_setting, int packets,
1823                                      int bytes)
1824 {
1825         unsigned int retval = itr_setting;
1826
1827         if (packets == 0)
1828                 goto update_itr_done;
1829
1830         switch (itr_setting) {
1831         case lowest_latency:
1832                 /* handle TSO and jumbo frames */
1833                 if (bytes/packets > 8000)
1834                         retval = bulk_latency;
1835                 else if ((packets < 5) && (bytes > 512)) {
1836                         retval = low_latency;
1837                 }
1838                 break;
1839         case low_latency:  /* 50 usec aka 20000 ints/s */
1840                 if (bytes > 10000) {
1841                         /* this if handles the TSO accounting */
1842                         if (bytes/packets > 8000) {
1843                                 retval = bulk_latency;
1844                         } else if ((packets < 10) || ((bytes/packets) > 1200)) {
1845                                 retval = bulk_latency;
1846                         } else if ((packets > 35)) {
1847                                 retval = lowest_latency;
1848                         }
1849                 } else if (bytes/packets > 2000) {
1850                         retval = bulk_latency;
1851                 } else if (packets <= 2 && bytes < 512) {
1852                         retval = lowest_latency;
1853                 }
1854                 break;
1855         case bulk_latency: /* 250 usec aka 4000 ints/s */
1856                 if (bytes > 25000) {
1857                         if (packets > 35) {
1858                                 retval = low_latency;
1859                         }
1860                 } else if (bytes < 6000) {
1861                         retval = low_latency;
1862                 }
1863                 break;
1864         }
1865
1866 update_itr_done:
1867         return retval;
1868 }
1869
1870 static void e1000_set_itr(struct e1000_adapter *adapter)
1871 {
1872         struct e1000_hw *hw = &adapter->hw;
1873         u16 current_itr;
1874         u32 new_itr = adapter->itr;
1875
1876         /* for non-gigabit speeds, just fix the interrupt rate at 4000 */
1877         if (adapter->link_speed != SPEED_1000) {
1878                 current_itr = 0;
1879                 new_itr = 4000;
1880                 goto set_itr_now;
1881         }
1882
1883         adapter->tx_itr = e1000_update_itr(adapter,
1884                                     adapter->tx_itr,
1885                                     adapter->total_tx_packets,
1886                                     adapter->total_tx_bytes);
1887         /* conservative mode (itr 3) eliminates the lowest_latency setting */
1888         if (adapter->itr_setting == 3 && adapter->tx_itr == lowest_latency)
1889                 adapter->tx_itr = low_latency;
1890
1891         adapter->rx_itr = e1000_update_itr(adapter,
1892                                     adapter->rx_itr,
1893                                     adapter->total_rx_packets,
1894                                     adapter->total_rx_bytes);
1895         /* conservative mode (itr 3) eliminates the lowest_latency setting */
1896         if (adapter->itr_setting == 3 && adapter->rx_itr == lowest_latency)
1897                 adapter->rx_itr = low_latency;
1898
1899         current_itr = max(adapter->rx_itr, adapter->tx_itr);
1900
1901         switch (current_itr) {
1902         /* counts and packets in update_itr are dependent on these numbers */
1903         case lowest_latency:
1904                 new_itr = 70000;
1905                 break;
1906         case low_latency:
1907                 new_itr = 20000; /* aka hwitr = ~200 */
1908                 break;
1909         case bulk_latency:
1910                 new_itr = 4000;
1911                 break;
1912         default:
1913                 break;
1914         }
1915
1916 set_itr_now:
1917         if (new_itr != adapter->itr) {
1918                 /*
1919                  * this attempts to bias the interrupt rate towards Bulk
1920                  * by adding intermediate steps when interrupt rate is
1921                  * increasing
1922                  */
1923                 new_itr = new_itr > adapter->itr ?
1924                              min(adapter->itr + (new_itr >> 2), new_itr) :
1925                              new_itr;
1926                 adapter->itr = new_itr;
1927                 adapter->rx_ring->itr_val = new_itr;
1928                 if (adapter->msix_entries)
1929                         adapter->rx_ring->set_itr = 1;
1930                 else
1931                         ew32(ITR, 1000000000 / (new_itr * 256));
1932         }
1933 }
1934
1935 /**
1936  * e1000_alloc_queues - Allocate memory for all rings
1937  * @adapter: board private structure to initialize
1938  **/
1939 static int __devinit e1000_alloc_queues(struct e1000_adapter *adapter)
1940 {
1941         adapter->tx_ring = kzalloc(sizeof(struct e1000_ring), GFP_KERNEL);
1942         if (!adapter->tx_ring)
1943                 goto err;
1944
1945         adapter->rx_ring = kzalloc(sizeof(struct e1000_ring), GFP_KERNEL);
1946         if (!adapter->rx_ring)
1947                 goto err;
1948
1949         return 0;
1950 err:
1951         e_err("Unable to allocate memory for queues\n");
1952         kfree(adapter->rx_ring);
1953         kfree(adapter->tx_ring);
1954         return -ENOMEM;
1955 }
1956
1957 /**
1958  * e1000_clean - NAPI Rx polling callback
1959  * @napi: struct associated with this polling callback
1960  * @budget: amount of packets driver is allowed to process this poll
1961  **/
1962 static int e1000_clean(struct napi_struct *napi, int budget)
1963 {
1964         struct e1000_adapter *adapter = container_of(napi, struct e1000_adapter, napi);
1965         struct e1000_hw *hw = &adapter->hw;
1966         struct net_device *poll_dev = adapter->netdev;
1967         int tx_cleaned = 1, work_done = 0;
1968
1969         adapter = netdev_priv(poll_dev);
1970
1971         if (adapter->msix_entries &&
1972             !(adapter->rx_ring->ims_val & adapter->tx_ring->ims_val))
1973                 goto clean_rx;
1974
1975         tx_cleaned = e1000_clean_tx_irq(adapter);
1976
1977 clean_rx:
1978         adapter->clean_rx(adapter, &work_done, budget);
1979
1980         if (!tx_cleaned)
1981                 work_done = budget;
1982
1983         /* If budget not fully consumed, exit the polling mode */
1984         if (work_done < budget) {
1985                 if (adapter->itr_setting & 3)
1986                         e1000_set_itr(adapter);
1987                 napi_complete(napi);
1988                 if (!test_bit(__E1000_DOWN, &adapter->state)) {
1989                         if (adapter->msix_entries)
1990                                 ew32(IMS, adapter->rx_ring->ims_val);
1991                         else
1992                                 e1000_irq_enable(adapter);
1993                 }
1994         }
1995
1996         return work_done;
1997 }
1998
1999 static void e1000_vlan_rx_add_vid(struct net_device *netdev, u16 vid)
2000 {
2001         struct e1000_adapter *adapter = netdev_priv(netdev);
2002         struct e1000_hw *hw = &adapter->hw;
2003         u32 vfta, index;
2004
2005         /* don't update vlan cookie if already programmed */
2006         if ((adapter->hw.mng_cookie.status &
2007              E1000_MNG_DHCP_COOKIE_STATUS_VLAN) &&
2008             (vid == adapter->mng_vlan_id))
2009                 return;
2010         /* add VID to filter table */
2011         index = (vid >> 5) & 0x7F;
2012         vfta = E1000_READ_REG_ARRAY(hw, E1000_VFTA, index);
2013         vfta |= (1 << (vid & 0x1F));
2014         e1000e_write_vfta(hw, index, vfta);
2015 }
2016
2017 static void e1000_vlan_rx_kill_vid(struct net_device *netdev, u16 vid)
2018 {
2019         struct e1000_adapter *adapter = netdev_priv(netdev);
2020         struct e1000_hw *hw = &adapter->hw;
2021         u32 vfta, index;
2022
2023         if (!test_bit(__E1000_DOWN, &adapter->state))
2024                 e1000_irq_disable(adapter);
2025         vlan_group_set_device(adapter->vlgrp, vid, NULL);
2026
2027         if (!test_bit(__E1000_DOWN, &adapter->state))
2028                 e1000_irq_enable(adapter);
2029
2030         if ((adapter->hw.mng_cookie.status &
2031              E1000_MNG_DHCP_COOKIE_STATUS_VLAN) &&
2032             (vid == adapter->mng_vlan_id)) {
2033                 /* release control to f/w */
2034                 e1000_release_hw_control(adapter);
2035                 return;
2036         }
2037
2038         /* remove VID from filter table */
2039         index = (vid >> 5) & 0x7F;
2040         vfta = E1000_READ_REG_ARRAY(hw, E1000_VFTA, index);
2041         vfta &= ~(1 << (vid & 0x1F));
2042         e1000e_write_vfta(hw, index, vfta);
2043 }
2044
2045 static void e1000_update_mng_vlan(struct e1000_adapter *adapter)
2046 {
2047         struct net_device *netdev = adapter->netdev;
2048         u16 vid = adapter->hw.mng_cookie.vlan_id;
2049         u16 old_vid = adapter->mng_vlan_id;
2050
2051         if (!adapter->vlgrp)
2052                 return;
2053
2054         if (!vlan_group_get_device(adapter->vlgrp, vid)) {
2055                 adapter->mng_vlan_id = E1000_MNG_VLAN_NONE;
2056                 if (adapter->hw.mng_cookie.status &
2057                         E1000_MNG_DHCP_COOKIE_STATUS_VLAN) {
2058                         e1000_vlan_rx_add_vid(netdev, vid);
2059                         adapter->mng_vlan_id = vid;
2060                 }
2061
2062                 if ((old_vid != (u16)E1000_MNG_VLAN_NONE) &&
2063                                 (vid != old_vid) &&
2064                     !vlan_group_get_device(adapter->vlgrp, old_vid))
2065                         e1000_vlan_rx_kill_vid(netdev, old_vid);
2066         } else {
2067                 adapter->mng_vlan_id = vid;
2068         }
2069 }
2070
2071
2072 static void e1000_vlan_rx_register(struct net_device *netdev,
2073                                    struct vlan_group *grp)
2074 {
2075         struct e1000_adapter *adapter = netdev_priv(netdev);
2076         struct e1000_hw *hw = &adapter->hw;
2077         u32 ctrl, rctl;
2078
2079         if (!test_bit(__E1000_DOWN, &adapter->state))
2080                 e1000_irq_disable(adapter);
2081         adapter->vlgrp = grp;
2082
2083         if (grp) {
2084                 /* enable VLAN tag insert/strip */
2085                 ctrl = er32(CTRL);
2086                 ctrl |= E1000_CTRL_VME;
2087                 ew32(CTRL, ctrl);
2088
2089                 if (adapter->flags & FLAG_HAS_HW_VLAN_FILTER) {
2090                         /* enable VLAN receive filtering */
2091                         rctl = er32(RCTL);
2092                         rctl &= ~E1000_RCTL_CFIEN;
2093                         ew32(RCTL, rctl);
2094                         e1000_update_mng_vlan(adapter);
2095                 }
2096         } else {
2097                 /* disable VLAN tag insert/strip */
2098                 ctrl = er32(CTRL);
2099                 ctrl &= ~E1000_CTRL_VME;
2100                 ew32(CTRL, ctrl);
2101
2102                 if (adapter->flags & FLAG_HAS_HW_VLAN_FILTER) {
2103                         if (adapter->mng_vlan_id !=
2104                             (u16)E1000_MNG_VLAN_NONE) {
2105                                 e1000_vlan_rx_kill_vid(netdev,
2106                                                        adapter->mng_vlan_id);
2107                                 adapter->mng_vlan_id = E1000_MNG_VLAN_NONE;
2108                         }
2109                 }
2110         }
2111
2112         if (!test_bit(__E1000_DOWN, &adapter->state))
2113                 e1000_irq_enable(adapter);
2114 }
2115
2116 static void e1000_restore_vlan(struct e1000_adapter *adapter)
2117 {
2118         u16 vid;
2119
2120         e1000_vlan_rx_register(adapter->netdev, adapter->vlgrp);
2121
2122         if (!adapter->vlgrp)
2123                 return;
2124
2125         for (vid = 0; vid < VLAN_GROUP_ARRAY_LEN; vid++) {
2126                 if (!vlan_group_get_device(adapter->vlgrp, vid))
2127                         continue;
2128                 e1000_vlan_rx_add_vid(adapter->netdev, vid);
2129         }
2130 }
2131
2132 static void e1000_init_manageability(struct e1000_adapter *adapter)
2133 {
2134         struct e1000_hw *hw = &adapter->hw;
2135         u32 manc, manc2h;
2136
2137         if (!(adapter->flags & FLAG_MNG_PT_ENABLED))
2138                 return;
2139
2140         manc = er32(MANC);
2141
2142         /*
2143          * enable receiving management packets to the host. this will probably
2144          * generate destination unreachable messages from the host OS, but
2145          * the packets will be handled on SMBUS
2146          */
2147         manc |= E1000_MANC_EN_MNG2HOST;
2148         manc2h = er32(MANC2H);
2149 #define E1000_MNG2HOST_PORT_623 (1 << 5)
2150 #define E1000_MNG2HOST_PORT_664 (1 << 6)
2151         manc2h |= E1000_MNG2HOST_PORT_623;
2152         manc2h |= E1000_MNG2HOST_PORT_664;
2153         ew32(MANC2H, manc2h);
2154         ew32(MANC, manc);
2155 }
2156
2157 /**
2158  * e1000_configure_tx - Configure 8254x Transmit Unit after Reset
2159  * @adapter: board private structure
2160  *
2161  * Configure the Tx unit of the MAC after a reset.
2162  **/
2163 static void e1000_configure_tx(struct e1000_adapter *adapter)
2164 {
2165         struct e1000_hw *hw = &adapter->hw;
2166         struct e1000_ring *tx_ring = adapter->tx_ring;
2167         u64 tdba;
2168         u32 tdlen, tctl, tipg, tarc;
2169         u32 ipgr1, ipgr2;
2170
2171         /* Setup the HW Tx Head and Tail descriptor pointers */
2172         tdba = tx_ring->dma;
2173         tdlen = tx_ring->count * sizeof(struct e1000_tx_desc);
2174         ew32(TDBAL, (tdba & DMA_BIT_MASK(32)));
2175         ew32(TDBAH, (tdba >> 32));
2176         ew32(TDLEN, tdlen);
2177         ew32(TDH, 0);
2178         ew32(TDT, 0);
2179         tx_ring->head = E1000_TDH;
2180         tx_ring->tail = E1000_TDT;
2181
2182         /* Set the default values for the Tx Inter Packet Gap timer */
2183         tipg = DEFAULT_82543_TIPG_IPGT_COPPER;          /*  8  */
2184         ipgr1 = DEFAULT_82543_TIPG_IPGR1;               /*  8  */
2185         ipgr2 = DEFAULT_82543_TIPG_IPGR2;               /*  6  */
2186
2187         if (adapter->flags & FLAG_TIPG_MEDIUM_FOR_80003ESLAN)
2188                 ipgr2 = DEFAULT_80003ES2LAN_TIPG_IPGR2; /*  7  */
2189
2190         tipg |= ipgr1 << E1000_TIPG_IPGR1_SHIFT;
2191         tipg |= ipgr2 << E1000_TIPG_IPGR2_SHIFT;
2192         ew32(TIPG, tipg);
2193
2194         /* Set the Tx Interrupt Delay register */
2195         ew32(TIDV, adapter->tx_int_delay);
2196         /* Tx irq moderation */
2197         ew32(TADV, adapter->tx_abs_int_delay);
2198
2199         /* Program the Transmit Control Register */
2200         tctl = er32(TCTL);
2201         tctl &= ~E1000_TCTL_CT;
2202         tctl |= E1000_TCTL_PSP | E1000_TCTL_RTLC |
2203                 (E1000_COLLISION_THRESHOLD << E1000_CT_SHIFT);
2204
2205         if (adapter->flags & FLAG_TARC_SPEED_MODE_BIT) {
2206                 tarc = er32(TARC(0));
2207                 /*
2208                  * set the speed mode bit, we'll clear it if we're not at
2209                  * gigabit link later
2210                  */
2211 #define SPEED_MODE_BIT (1 << 21)
2212                 tarc |= SPEED_MODE_BIT;
2213                 ew32(TARC(0), tarc);
2214         }
2215
2216         /* errata: program both queues to unweighted RR */
2217         if (adapter->flags & FLAG_TARC_SET_BIT_ZERO) {
2218                 tarc = er32(TARC(0));
2219                 tarc |= 1;
2220                 ew32(TARC(0), tarc);
2221                 tarc = er32(TARC(1));
2222                 tarc |= 1;
2223                 ew32(TARC(1), tarc);
2224         }
2225
2226         /* Setup Transmit Descriptor Settings for eop descriptor */
2227         adapter->txd_cmd = E1000_TXD_CMD_EOP | E1000_TXD_CMD_IFCS;
2228
2229         /* only set IDE if we are delaying interrupts using the timers */
2230         if (adapter->tx_int_delay)
2231                 adapter->txd_cmd |= E1000_TXD_CMD_IDE;
2232
2233         /* enable Report Status bit */
2234         adapter->txd_cmd |= E1000_TXD_CMD_RS;
2235
2236         ew32(TCTL, tctl);
2237
2238         e1000e_config_collision_dist(hw);
2239
2240         adapter->tx_queue_len = adapter->netdev->tx_queue_len;
2241 }
2242
2243 /**
2244  * e1000_setup_rctl - configure the receive control registers
2245  * @adapter: Board private structure
2246  **/
2247 #define PAGE_USE_COUNT(S) (((S) >> PAGE_SHIFT) + \
2248                            (((S) & (PAGE_SIZE - 1)) ? 1 : 0))
2249 static void e1000_setup_rctl(struct e1000_adapter *adapter)
2250 {
2251         struct e1000_hw *hw = &adapter->hw;
2252         u32 rctl, rfctl;
2253         u32 psrctl = 0;
2254         u32 pages = 0;
2255
2256         /* Program MC offset vector base */
2257         rctl = er32(RCTL);
2258         rctl &= ~(3 << E1000_RCTL_MO_SHIFT);
2259         rctl |= E1000_RCTL_EN | E1000_RCTL_BAM |
2260                 E1000_RCTL_LBM_NO | E1000_RCTL_RDMTS_HALF |
2261                 (adapter->hw.mac.mc_filter_type << E1000_RCTL_MO_SHIFT);
2262
2263         /* Do not Store bad packets */
2264         rctl &= ~E1000_RCTL_SBP;
2265
2266         /* Enable Long Packet receive */
2267         if (adapter->netdev->mtu <= ETH_DATA_LEN)
2268                 rctl &= ~E1000_RCTL_LPE;
2269         else
2270                 rctl |= E1000_RCTL_LPE;
2271
2272         /* Some systems expect that the CRC is included in SMBUS traffic. The
2273          * hardware strips the CRC before sending to both SMBUS (BMC) and to
2274          * host memory when this is enabled
2275          */
2276         if (adapter->flags2 & FLAG2_CRC_STRIPPING)
2277                 rctl |= E1000_RCTL_SECRC;
2278
2279         /* Workaround Si errata on 82577 PHY - configure IPG for jumbos */
2280         if ((hw->phy.type == e1000_phy_82577) && (rctl & E1000_RCTL_LPE)) {
2281                 u16 phy_data;
2282
2283                 e1e_rphy(hw, PHY_REG(770, 26), &phy_data);
2284                 phy_data &= 0xfff8;
2285                 phy_data |= (1 << 2);
2286                 e1e_wphy(hw, PHY_REG(770, 26), phy_data);
2287
2288                 e1e_rphy(hw, 22, &phy_data);
2289                 phy_data &= 0x0fff;
2290                 phy_data |= (1 << 14);
2291                 e1e_wphy(hw, 0x10, 0x2823);
2292                 e1e_wphy(hw, 0x11, 0x0003);
2293                 e1e_wphy(hw, 22, phy_data);
2294         }
2295
2296         /* Setup buffer sizes */
2297         rctl &= ~E1000_RCTL_SZ_4096;
2298         rctl |= E1000_RCTL_BSEX;
2299         switch (adapter->rx_buffer_len) {
2300         case 256:
2301                 rctl |= E1000_RCTL_SZ_256;
2302                 rctl &= ~E1000_RCTL_BSEX;
2303                 break;
2304         case 512:
2305                 rctl |= E1000_RCTL_SZ_512;
2306                 rctl &= ~E1000_RCTL_BSEX;
2307                 break;
2308         case 1024:
2309                 rctl |= E1000_RCTL_SZ_1024;
2310                 rctl &= ~E1000_RCTL_BSEX;
2311                 break;
2312         case 2048:
2313         default:
2314                 rctl |= E1000_RCTL_SZ_2048;
2315                 rctl &= ~E1000_RCTL_BSEX;
2316                 break;
2317         case 4096:
2318                 rctl |= E1000_RCTL_SZ_4096;
2319                 break;
2320         case 8192:
2321                 rctl |= E1000_RCTL_SZ_8192;
2322                 break;
2323         case 16384:
2324                 rctl |= E1000_RCTL_SZ_16384;
2325                 break;
2326         }
2327
2328         /*
2329          * 82571 and greater support packet-split where the protocol
2330          * header is placed in skb->data and the packet data is
2331          * placed in pages hanging off of skb_shinfo(skb)->nr_frags.
2332          * In the case of a non-split, skb->data is linearly filled,
2333          * followed by the page buffers.  Therefore, skb->data is
2334          * sized to hold the largest protocol header.
2335          *
2336          * allocations using alloc_page take too long for regular MTU
2337          * so only enable packet split for jumbo frames
2338          *
2339          * Using pages when the page size is greater than 16k wastes
2340          * a lot of memory, since we allocate 3 pages at all times
2341          * per packet.
2342          */
2343         pages = PAGE_USE_COUNT(adapter->netdev->mtu);
2344         if (!(adapter->flags & FLAG_IS_ICH) && (pages <= 3) &&
2345             (PAGE_SIZE <= 16384) && (rctl & E1000_RCTL_LPE))
2346                 adapter->rx_ps_pages = pages;
2347         else
2348                 adapter->rx_ps_pages = 0;
2349
2350         if (adapter->rx_ps_pages) {
2351                 /* Configure extra packet-split registers */
2352                 rfctl = er32(RFCTL);
2353                 rfctl |= E1000_RFCTL_EXTEN;
2354                 /*
2355                  * disable packet split support for IPv6 extension headers,
2356                  * because some malformed IPv6 headers can hang the Rx
2357                  */
2358                 rfctl |= (E1000_RFCTL_IPV6_EX_DIS |
2359                           E1000_RFCTL_NEW_IPV6_EXT_DIS);
2360
2361                 ew32(RFCTL, rfctl);
2362
2363                 /* Enable Packet split descriptors */
2364                 rctl |= E1000_RCTL_DTYP_PS;
2365
2366                 psrctl |= adapter->rx_ps_bsize0 >>
2367                         E1000_PSRCTL_BSIZE0_SHIFT;
2368
2369                 switch (adapter->rx_ps_pages) {
2370                 case 3:
2371                         psrctl |= PAGE_SIZE <<
2372                                 E1000_PSRCTL_BSIZE3_SHIFT;
2373                 case 2:
2374                         psrctl |= PAGE_SIZE <<
2375                                 E1000_PSRCTL_BSIZE2_SHIFT;
2376                 case 1:
2377                         psrctl |= PAGE_SIZE >>
2378                                 E1000_PSRCTL_BSIZE1_SHIFT;
2379                         break;
2380                 }
2381
2382                 ew32(PSRCTL, psrctl);
2383         }
2384
2385         ew32(RCTL, rctl);
2386         /* just started the receive unit, no need to restart */
2387         adapter->flags &= ~FLAG_RX_RESTART_NOW;
2388 }
2389
2390 /**
2391  * e1000_configure_rx - Configure Receive Unit after Reset
2392  * @adapter: board private structure
2393  *
2394  * Configure the Rx unit of the MAC after a reset.
2395  **/
2396 static void e1000_configure_rx(struct e1000_adapter *adapter)
2397 {
2398         struct e1000_hw *hw = &adapter->hw;
2399         struct e1000_ring *rx_ring = adapter->rx_ring;
2400         u64 rdba;
2401         u32 rdlen, rctl, rxcsum, ctrl_ext;
2402
2403         if (adapter->rx_ps_pages) {
2404                 /* this is a 32 byte descriptor */
2405                 rdlen = rx_ring->count *
2406                         sizeof(union e1000_rx_desc_packet_split);
2407                 adapter->clean_rx = e1000_clean_rx_irq_ps;
2408                 adapter->alloc_rx_buf = e1000_alloc_rx_buffers_ps;
2409         } else if (adapter->netdev->mtu > ETH_FRAME_LEN + ETH_FCS_LEN) {
2410                 rdlen = rx_ring->count * sizeof(struct e1000_rx_desc);
2411                 adapter->clean_rx = e1000_clean_jumbo_rx_irq;
2412                 adapter->alloc_rx_buf = e1000_alloc_jumbo_rx_buffers;
2413         } else {
2414                 rdlen = rx_ring->count * sizeof(struct e1000_rx_desc);
2415                 adapter->clean_rx = e1000_clean_rx_irq;
2416                 adapter->alloc_rx_buf = e1000_alloc_rx_buffers;
2417         }
2418
2419         /* disable receives while setting up the descriptors */
2420         rctl = er32(RCTL);
2421         ew32(RCTL, rctl & ~E1000_RCTL_EN);
2422         e1e_flush();
2423         msleep(10);
2424
2425         /* set the Receive Delay Timer Register */
2426         ew32(RDTR, adapter->rx_int_delay);
2427
2428         /* irq moderation */
2429         ew32(RADV, adapter->rx_abs_int_delay);
2430         if (adapter->itr_setting != 0)
2431                 ew32(ITR, 1000000000 / (adapter->itr * 256));
2432
2433         ctrl_ext = er32(CTRL_EXT);
2434         /* Auto-Mask interrupts upon ICR access */
2435         ctrl_ext |= E1000_CTRL_EXT_IAME;
2436         ew32(IAM, 0xffffffff);
2437         ew32(CTRL_EXT, ctrl_ext);
2438         e1e_flush();
2439
2440         /*
2441          * Setup the HW Rx Head and Tail Descriptor Pointers and
2442          * the Base and Length of the Rx Descriptor Ring
2443          */
2444         rdba = rx_ring->dma;
2445         ew32(RDBAL, (rdba & DMA_BIT_MASK(32)));
2446         ew32(RDBAH, (rdba >> 32));
2447         ew32(RDLEN, rdlen);
2448         ew32(RDH, 0);
2449         ew32(RDT, 0);
2450         rx_ring->head = E1000_RDH;
2451         rx_ring->tail = E1000_RDT;
2452
2453         /* Enable Receive Checksum Offload for TCP and UDP */
2454         rxcsum = er32(RXCSUM);
2455         if (adapter->flags & FLAG_RX_CSUM_ENABLED) {
2456                 rxcsum |= E1000_RXCSUM_TUOFL;
2457
2458                 /*
2459                  * IPv4 payload checksum for UDP fragments must be
2460                  * used in conjunction with packet-split.
2461                  */
2462                 if (adapter->rx_ps_pages)
2463                         rxcsum |= E1000_RXCSUM_IPPCSE;
2464         } else {
2465                 rxcsum &= ~E1000_RXCSUM_TUOFL;
2466                 /* no need to clear IPPCSE as it defaults to 0 */
2467         }
2468         ew32(RXCSUM, rxcsum);
2469
2470         /*
2471          * Enable early receives on supported devices, only takes effect when
2472          * packet size is equal or larger than the specified value (in 8 byte
2473          * units), e.g. using jumbo frames when setting to E1000_ERT_2048
2474          */
2475         if ((adapter->flags & FLAG_HAS_ERT) &&
2476             (adapter->netdev->mtu > ETH_DATA_LEN)) {
2477                 u32 rxdctl = er32(RXDCTL(0));
2478                 ew32(RXDCTL(0), rxdctl | 0x3);
2479                 ew32(ERT, E1000_ERT_2048 | (1 << 13));
2480                 /*
2481                  * With jumbo frames and early-receive enabled, excessive
2482                  * C4->C2 latencies result in dropped transactions.
2483                  */
2484                 pm_qos_update_requirement(PM_QOS_CPU_DMA_LATENCY,
2485                                           e1000e_driver_name, 55);
2486         } else {
2487                 pm_qos_update_requirement(PM_QOS_CPU_DMA_LATENCY,
2488                                           e1000e_driver_name,
2489                                           PM_QOS_DEFAULT_VALUE);
2490         }
2491
2492         /* Enable Receives */
2493         ew32(RCTL, rctl);
2494 }
2495
2496 /**
2497  *  e1000_update_mc_addr_list - Update Multicast addresses
2498  *  @hw: pointer to the HW structure
2499  *  @mc_addr_list: array of multicast addresses to program
2500  *  @mc_addr_count: number of multicast addresses to program
2501  *  @rar_used_count: the first RAR register free to program
2502  *  @rar_count: total number of supported Receive Address Registers
2503  *
2504  *  Updates the Receive Address Registers and Multicast Table Array.
2505  *  The caller must have a packed mc_addr_list of multicast addresses.
2506  *  The parameter rar_count will usually be hw->mac.rar_entry_count
2507  *  unless there are workarounds that change this.  Currently no func pointer
2508  *  exists and all implementations are handled in the generic version of this
2509  *  function.
2510  **/
2511 static void e1000_update_mc_addr_list(struct e1000_hw *hw, u8 *mc_addr_list,
2512                                       u32 mc_addr_count, u32 rar_used_count,
2513                                       u32 rar_count)
2514 {
2515         hw->mac.ops.update_mc_addr_list(hw, mc_addr_list, mc_addr_count,
2516                                         rar_used_count, rar_count);
2517 }
2518
2519 /**
2520  * e1000_set_multi - Multicast and Promiscuous mode set
2521  * @netdev: network interface device structure
2522  *
2523  * The set_multi entry point is called whenever the multicast address
2524  * list or the network interface flags are updated.  This routine is
2525  * responsible for configuring the hardware for proper multicast,
2526  * promiscuous mode, and all-multi behavior.
2527  **/
2528 static void e1000_set_multi(struct net_device *netdev)
2529 {
2530         struct e1000_adapter *adapter = netdev_priv(netdev);
2531         struct e1000_hw *hw = &adapter->hw;
2532         struct e1000_mac_info *mac = &hw->mac;
2533         struct dev_mc_list *mc_ptr;
2534         u8  *mta_list;
2535         u32 rctl;
2536         int i;
2537
2538         /* Check for Promiscuous and All Multicast modes */
2539
2540         rctl = er32(RCTL);
2541
2542         if (netdev->flags & IFF_PROMISC) {
2543                 rctl |= (E1000_RCTL_UPE | E1000_RCTL_MPE);
2544                 rctl &= ~E1000_RCTL_VFE;
2545         } else {
2546                 if (netdev->flags & IFF_ALLMULTI) {
2547                         rctl |= E1000_RCTL_MPE;
2548                         rctl &= ~E1000_RCTL_UPE;
2549                 } else {
2550                         rctl &= ~(E1000_RCTL_UPE | E1000_RCTL_MPE);
2551                 }
2552                 if (adapter->flags & FLAG_HAS_HW_VLAN_FILTER)
2553                         rctl |= E1000_RCTL_VFE;
2554         }
2555
2556         ew32(RCTL, rctl);
2557
2558         if (netdev->mc_count) {
2559                 mta_list = kmalloc(netdev->mc_count * 6, GFP_ATOMIC);
2560                 if (!mta_list)
2561                         return;
2562
2563                 /* prepare a packed array of only addresses. */
2564                 mc_ptr = netdev->mc_list;
2565
2566                 for (i = 0; i < netdev->mc_count; i++) {
2567                         if (!mc_ptr)
2568                                 break;
2569                         memcpy(mta_list + (i*ETH_ALEN), mc_ptr->dmi_addr,
2570                                ETH_ALEN);
2571                         mc_ptr = mc_ptr->next;
2572                 }
2573
2574                 e1000_update_mc_addr_list(hw, mta_list, i, 1,
2575                                           mac->rar_entry_count);
2576                 kfree(mta_list);
2577         } else {
2578                 /*
2579                  * if we're called from probe, we might not have
2580                  * anything to do here, so clear out the list
2581                  */
2582                 e1000_update_mc_addr_list(hw, NULL, 0, 1, mac->rar_entry_count);
2583         }
2584 }
2585
2586 /**
2587  * e1000_configure - configure the hardware for Rx and Tx
2588  * @adapter: private board structure
2589  **/
2590 static void e1000_configure(struct e1000_adapter *adapter)
2591 {
2592         e1000_set_multi(adapter->netdev);
2593
2594         e1000_restore_vlan(adapter);
2595         e1000_init_manageability(adapter);
2596
2597         e1000_configure_tx(adapter);
2598         e1000_setup_rctl(adapter);
2599         e1000_configure_rx(adapter);
2600         adapter->alloc_rx_buf(adapter, e1000_desc_unused(adapter->rx_ring));
2601 }
2602
2603 /**
2604  * e1000e_power_up_phy - restore link in case the phy was powered down
2605  * @adapter: address of board private structure
2606  *
2607  * The phy may be powered down to save power and turn off link when the
2608  * driver is unloaded and wake on lan is not enabled (among others)
2609  * *** this routine MUST be followed by a call to e1000e_reset ***
2610  **/
2611 void e1000e_power_up_phy(struct e1000_adapter *adapter)
2612 {
2613         u16 mii_reg = 0;
2614
2615         /* Just clear the power down bit to wake the phy back up */
2616         if (adapter->hw.phy.media_type == e1000_media_type_copper) {
2617                 /*
2618                  * According to the manual, the phy will retain its
2619                  * settings across a power-down/up cycle
2620                  */
2621                 e1e_rphy(&adapter->hw, PHY_CONTROL, &mii_reg);
2622                 mii_reg &= ~MII_CR_POWER_DOWN;
2623                 e1e_wphy(&adapter->hw, PHY_CONTROL, mii_reg);
2624         }
2625
2626         adapter->hw.mac.ops.setup_link(&adapter->hw);
2627 }
2628
2629 /**
2630  * e1000_power_down_phy - Power down the PHY
2631  *
2632  * Power down the PHY so no link is implied when interface is down
2633  * The PHY cannot be powered down is management or WoL is active
2634  */
2635 static void e1000_power_down_phy(struct e1000_adapter *adapter)
2636 {
2637         struct e1000_hw *hw = &adapter->hw;
2638         u16 mii_reg;
2639
2640         /* WoL is enabled */
2641         if (adapter->wol)
2642                 return;
2643
2644         /* non-copper PHY? */
2645         if (adapter->hw.phy.media_type != e1000_media_type_copper)
2646                 return;
2647
2648         /* reset is blocked because of a SoL/IDER session */
2649         if (e1000e_check_mng_mode(hw) || e1000_check_reset_block(hw))
2650                 return;
2651
2652         /* manageability (AMT) is enabled */
2653         if (er32(MANC) & E1000_MANC_SMBUS_EN)
2654                 return;
2655
2656         /* power down the PHY */
2657         e1e_rphy(hw, PHY_CONTROL, &mii_reg);
2658         mii_reg |= MII_CR_POWER_DOWN;
2659         e1e_wphy(hw, PHY_CONTROL, mii_reg);
2660         mdelay(1);
2661 }
2662
2663 /**
2664  * e1000e_reset - bring the hardware into a known good state
2665  *
2666  * This function boots the hardware and enables some settings that
2667  * require a configuration cycle of the hardware - those cannot be
2668  * set/changed during runtime. After reset the device needs to be
2669  * properly configured for Rx, Tx etc.
2670  */
2671 void e1000e_reset(struct e1000_adapter *adapter)
2672 {
2673         struct e1000_mac_info *mac = &adapter->hw.mac;
2674         struct e1000_fc_info *fc = &adapter->hw.fc;
2675         struct e1000_hw *hw = &adapter->hw;
2676         u32 tx_space, min_tx_space, min_rx_space;
2677         u32 pba = adapter->pba;
2678         u16 hwm;
2679
2680         /* reset Packet Buffer Allocation to default */
2681         ew32(PBA, pba);
2682
2683         if (adapter->max_frame_size > ETH_FRAME_LEN + ETH_FCS_LEN) {
2684                 /*
2685                  * To maintain wire speed transmits, the Tx FIFO should be
2686                  * large enough to accommodate two full transmit packets,
2687                  * rounded up to the next 1KB and expressed in KB.  Likewise,
2688                  * the Rx FIFO should be large enough to accommodate at least
2689                  * one full receive packet and is similarly rounded up and
2690                  * expressed in KB.
2691                  */
2692                 pba = er32(PBA);
2693                 /* upper 16 bits has Tx packet buffer allocation size in KB */
2694                 tx_space = pba >> 16;
2695                 /* lower 16 bits has Rx packet buffer allocation size in KB */
2696                 pba &= 0xffff;
2697                 /*
2698                  * the Tx fifo also stores 16 bytes of information about the tx
2699                  * but don't include ethernet FCS because hardware appends it
2700                  */
2701                 min_tx_space = (adapter->max_frame_size +
2702                                 sizeof(struct e1000_tx_desc) -
2703                                 ETH_FCS_LEN) * 2;
2704                 min_tx_space = ALIGN(min_tx_space, 1024);
2705                 min_tx_space >>= 10;
2706                 /* software strips receive CRC, so leave room for it */
2707                 min_rx_space = adapter->max_frame_size;
2708                 min_rx_space = ALIGN(min_rx_space, 1024);
2709                 min_rx_space >>= 10;
2710
2711                 /*
2712                  * If current Tx allocation is less than the min Tx FIFO size,
2713                  * and the min Tx FIFO size is less than the current Rx FIFO
2714                  * allocation, take space away from current Rx allocation
2715                  */
2716                 if ((tx_space < min_tx_space) &&
2717                     ((min_tx_space - tx_space) < pba)) {
2718                         pba -= min_tx_space - tx_space;
2719
2720                         /*
2721                          * if short on Rx space, Rx wins and must trump tx
2722                          * adjustment or use Early Receive if available
2723                          */
2724                         if ((pba < min_rx_space) &&
2725                             (!(adapter->flags & FLAG_HAS_ERT)))
2726                                 /* ERT enabled in e1000_configure_rx */
2727                                 pba = min_rx_space;
2728                 }
2729
2730                 ew32(PBA, pba);
2731         }
2732
2733
2734         /*
2735          * flow control settings
2736          *
2737          * The high water mark must be low enough to fit two full frame
2738          * (or the size used for early receive) above it in the Rx FIFO.
2739          * Set it to the lower of:
2740          * - 90% of the Rx FIFO size, and
2741          * - the full Rx FIFO size minus the early receive size (for parts
2742          *   with ERT support assuming ERT set to E1000_ERT_2048), or
2743          * - the full Rx FIFO size minus two full frames
2744          */
2745         if ((adapter->flags & FLAG_HAS_ERT) &&
2746             (adapter->netdev->mtu > ETH_DATA_LEN))
2747                 hwm = min(((pba << 10) * 9 / 10),
2748                           ((pba << 10) - (E1000_ERT_2048 << 3)));
2749         else
2750                 hwm = min(((pba << 10) * 9 / 10),
2751                           ((pba << 10) - (2 * adapter->max_frame_size)));
2752
2753         fc->high_water = hwm & E1000_FCRTH_RTH; /* 8-byte granularity */
2754         fc->low_water = (fc->high_water - (2 * adapter->max_frame_size));
2755         fc->low_water &= E1000_FCRTL_RTL; /* 8-byte granularity */
2756
2757         if (adapter->flags & FLAG_DISABLE_FC_PAUSE_TIME)
2758                 fc->pause_time = 0xFFFF;
2759         else
2760                 fc->pause_time = E1000_FC_PAUSE_TIME;
2761         fc->send_xon = 1;
2762         fc->current_mode = fc->requested_mode;
2763
2764         /* Allow time for pending master requests to run */
2765         mac->ops.reset_hw(hw);
2766
2767         /*
2768          * For parts with AMT enabled, let the firmware know
2769          * that the network interface is in control
2770          */
2771         if (adapter->flags & FLAG_HAS_AMT)
2772                 e1000_get_hw_control(adapter);
2773
2774         ew32(WUC, 0);
2775         if (adapter->flags2 & FLAG2_HAS_PHY_WAKEUP)
2776                 e1e_wphy(&adapter->hw, BM_WUC, 0);
2777
2778         if (mac->ops.init_hw(hw))
2779                 e_err("Hardware Error\n");
2780
2781         e1000_update_mng_vlan(adapter);
2782
2783         /* Enable h/w to recognize an 802.1Q VLAN Ethernet packet */
2784         ew32(VET, ETH_P_8021Q);
2785
2786         e1000e_reset_adaptive(hw);
2787         e1000_get_phy_info(hw);
2788
2789         if ((adapter->flags & FLAG_HAS_SMART_POWER_DOWN) &&
2790             !(adapter->flags & FLAG_SMART_POWER_DOWN)) {
2791                 u16 phy_data = 0;
2792                 /*
2793                  * speed up time to link by disabling smart power down, ignore
2794                  * the return value of this function because there is nothing
2795                  * different we would do if it failed
2796                  */
2797                 e1e_rphy(hw, IGP02E1000_PHY_POWER_MGMT, &phy_data);
2798                 phy_data &= ~IGP02E1000_PM_SPD;
2799                 e1e_wphy(hw, IGP02E1000_PHY_POWER_MGMT, phy_data);
2800         }
2801 }
2802
2803 int e1000e_up(struct e1000_adapter *adapter)
2804 {
2805         struct e1000_hw *hw = &adapter->hw;
2806
2807         /* hardware has been reset, we need to reload some things */
2808         e1000_configure(adapter);
2809
2810         clear_bit(__E1000_DOWN, &adapter->state);
2811
2812         napi_enable(&adapter->napi);
2813         if (adapter->msix_entries)
2814                 e1000_configure_msix(adapter);
2815         e1000_irq_enable(adapter);
2816
2817         netif_wake_queue(adapter->netdev);
2818
2819         /* fire a link change interrupt to start the watchdog */
2820         ew32(ICS, E1000_ICS_LSC);
2821         return 0;
2822 }
2823
2824 void e1000e_down(struct e1000_adapter *adapter)
2825 {
2826         struct net_device *netdev = adapter->netdev;
2827         struct e1000_hw *hw = &adapter->hw;
2828         u32 tctl, rctl;
2829
2830         /*
2831          * signal that we're down so the interrupt handler does not
2832          * reschedule our watchdog timer
2833          */
2834         set_bit(__E1000_DOWN, &adapter->state);
2835
2836         /* disable receives in the hardware */
2837         rctl = er32(RCTL);
2838         ew32(RCTL, rctl & ~E1000_RCTL_EN);
2839         /* flush and sleep below */
2840
2841         netif_stop_queue(netdev);
2842
2843         /* disable transmits in the hardware */
2844         tctl = er32(TCTL);
2845         tctl &= ~E1000_TCTL_EN;
2846         ew32(TCTL, tctl);
2847         /* flush both disables and wait for them to finish */
2848         e1e_flush();
2849         msleep(10);
2850
2851         napi_disable(&adapter->napi);
2852         e1000_irq_disable(adapter);
2853
2854         del_timer_sync(&adapter->watchdog_timer);
2855         del_timer_sync(&adapter->phy_info_timer);
2856
2857         netdev->tx_queue_len = adapter->tx_queue_len;
2858         netif_carrier_off(netdev);
2859         adapter->link_speed = 0;
2860         adapter->link_duplex = 0;
2861
2862         if (!pci_channel_offline(adapter->pdev))
2863                 e1000e_reset(adapter);
2864         e1000_clean_tx_ring(adapter);
2865         e1000_clean_rx_ring(adapter);
2866
2867         /*
2868          * TODO: for power management, we could drop the link and
2869          * pci_disable_device here.
2870          */
2871 }
2872
2873 void e1000e_reinit_locked(struct e1000_adapter *adapter)
2874 {
2875         might_sleep();
2876         while (test_and_set_bit(__E1000_RESETTING, &adapter->state))
2877                 msleep(1);
2878         e1000e_down(adapter);
2879         e1000e_up(adapter);
2880         clear_bit(__E1000_RESETTING, &adapter->state);
2881 }
2882
2883 /**
2884  * e1000_sw_init - Initialize general software structures (struct e1000_adapter)
2885  * @adapter: board private structure to initialize
2886  *
2887  * e1000_sw_init initializes the Adapter private data structure.
2888  * Fields are initialized based on PCI device information and
2889  * OS network device settings (MTU size).
2890  **/
2891 static int __devinit e1000_sw_init(struct e1000_adapter *adapter)
2892 {
2893         struct net_device *netdev = adapter->netdev;
2894
2895         adapter->rx_buffer_len = ETH_FRAME_LEN + VLAN_HLEN + ETH_FCS_LEN;
2896         adapter->rx_ps_bsize0 = 128;
2897         adapter->max_frame_size = netdev->mtu + ETH_HLEN + ETH_FCS_LEN;
2898         adapter->min_frame_size = ETH_ZLEN + ETH_FCS_LEN;
2899
2900         e1000e_set_interrupt_capability(adapter);
2901
2902         if (e1000_alloc_queues(adapter))
2903                 return -ENOMEM;
2904
2905         /* Explicitly disable IRQ since the NIC can be in any state. */
2906         e1000_irq_disable(adapter);
2907
2908         set_bit(__E1000_DOWN, &adapter->state);
2909         return 0;
2910 }
2911
2912 /**
2913  * e1000_intr_msi_test - Interrupt Handler
2914  * @irq: interrupt number
2915  * @data: pointer to a network interface device structure
2916  **/
2917 static irqreturn_t e1000_intr_msi_test(int irq, void *data)
2918 {
2919         struct net_device *netdev = data;
2920         struct e1000_adapter *adapter = netdev_priv(netdev);
2921         struct e1000_hw *hw = &adapter->hw;
2922         u32 icr = er32(ICR);
2923
2924         e_dbg("icr is %08X\n", icr);
2925         if (icr & E1000_ICR_RXSEQ) {
2926                 adapter->flags &= ~FLAG_MSI_TEST_FAILED;
2927                 wmb();
2928         }
2929
2930         return IRQ_HANDLED;
2931 }
2932
2933 /**
2934  * e1000_test_msi_interrupt - Returns 0 for successful test
2935  * @adapter: board private struct
2936  *
2937  * code flow taken from tg3.c
2938  **/
2939 static int e1000_test_msi_interrupt(struct e1000_adapter *adapter)
2940 {
2941         struct net_device *netdev = adapter->netdev;
2942         struct e1000_hw *hw = &adapter->hw;
2943         int err;
2944
2945         /* poll_enable hasn't been called yet, so don't need disable */
2946         /* clear any pending events */
2947         er32(ICR);
2948
2949         /* free the real vector and request a test handler */
2950         e1000_free_irq(adapter);
2951         e1000e_reset_interrupt_capability(adapter);
2952
2953         /* Assume that the test fails, if it succeeds then the test
2954          * MSI irq handler will unset this flag */
2955         adapter->flags |= FLAG_MSI_TEST_FAILED;
2956
2957         err = pci_enable_msi(adapter->pdev);
2958         if (err)
2959                 goto msi_test_failed;
2960
2961         err = request_irq(adapter->pdev->irq, e1000_intr_msi_test, 0,
2962                           netdev->name, netdev);
2963         if (err) {
2964                 pci_disable_msi(adapter->pdev);
2965                 goto msi_test_failed;
2966         }
2967
2968         wmb();
2969
2970         e1000_irq_enable(adapter);
2971
2972         /* fire an unusual interrupt on the test handler */
2973         ew32(ICS, E1000_ICS_RXSEQ);
2974         e1e_flush();
2975         msleep(50);
2976
2977         e1000_irq_disable(adapter);
2978
2979         rmb();
2980
2981         if (adapter->flags & FLAG_MSI_TEST_FAILED) {
2982                 adapter->int_mode = E1000E_INT_MODE_LEGACY;
2983                 err = -EIO;
2984                 e_info("MSI interrupt test failed!\n");
2985         }
2986
2987         free_irq(adapter->pdev->irq, netdev);
2988         pci_disable_msi(adapter->pdev);
2989
2990         if (err == -EIO)
2991                 goto msi_test_failed;
2992
2993         /* okay so the test worked, restore settings */
2994         e_dbg("MSI interrupt test succeeded!\n");
2995 msi_test_failed:
2996         e1000e_set_interrupt_capability(adapter);
2997         e1000_request_irq(adapter);
2998         return err;
2999 }
3000
3001 /**
3002  * e1000_test_msi - Returns 0 if MSI test succeeds or INTx mode is restored
3003  * @adapter: board private struct
3004  *
3005  * code flow taken from tg3.c, called with e1000 interrupts disabled.
3006  **/
3007 static int e1000_test_msi(struct e1000_adapter *adapter)
3008 {
3009         int err;
3010         u16 pci_cmd;
3011
3012         if (!(adapter->flags & FLAG_MSI_ENABLED))
3013                 return 0;
3014
3015         /* disable SERR in case the MSI write causes a master abort */
3016         pci_read_config_word(adapter->pdev, PCI_COMMAND, &pci_cmd);
3017         pci_write_config_word(adapter->pdev, PCI_COMMAND,
3018                               pci_cmd & ~PCI_COMMAND_SERR);
3019
3020         err = e1000_test_msi_interrupt(adapter);
3021
3022         /* restore previous setting of command word */
3023         pci_write_config_word(adapter->pdev, PCI_COMMAND, pci_cmd);
3024
3025         /* success ! */
3026         if (!err)
3027                 return 0;
3028
3029         /* EIO means MSI test failed */
3030         if (err != -EIO)
3031                 return err;
3032
3033         /* back to INTx mode */
3034         e_warn("MSI interrupt test failed, using legacy interrupt.\n");
3035
3036         e1000_free_irq(adapter);
3037
3038         err = e1000_request_irq(adapter);
3039
3040         return err;
3041 }
3042
3043 /**
3044  * e1000_open - Called when a network interface is made active
3045  * @netdev: network interface device structure
3046  *
3047  * Returns 0 on success, negative value on failure
3048  *
3049  * The open entry point is called when a network interface is made
3050  * active by the system (IFF_UP).  At this point all resources needed
3051  * for transmit and receive operations are allocated, the interrupt
3052  * handler is registered with the OS, the watchdog timer is started,
3053  * and the stack is notified that the interface is ready.
3054  **/
3055 static int e1000_open(struct net_device *netdev)
3056 {
3057         struct e1000_adapter *adapter = netdev_priv(netdev);
3058         struct e1000_hw *hw = &adapter->hw;
3059         int err;
3060
3061         /* disallow open during test */
3062         if (test_bit(__E1000_TESTING, &adapter->state))
3063                 return -EBUSY;
3064
3065         netif_carrier_off(netdev);
3066
3067         /* allocate transmit descriptors */
3068         err = e1000e_setup_tx_resources(adapter);
3069         if (err)
3070                 goto err_setup_tx;
3071
3072         /* allocate receive descriptors */
3073         err = e1000e_setup_rx_resources(adapter);
3074         if (err)
3075                 goto err_setup_rx;
3076
3077         e1000e_power_up_phy(adapter);
3078
3079         adapter->mng_vlan_id = E1000_MNG_VLAN_NONE;
3080         if ((adapter->hw.mng_cookie.status &
3081              E1000_MNG_DHCP_COOKIE_STATUS_VLAN))
3082                 e1000_update_mng_vlan(adapter);
3083
3084         /*
3085          * If AMT is enabled, let the firmware know that the network
3086          * interface is now open
3087          */
3088         if (adapter->flags & FLAG_HAS_AMT)
3089                 e1000_get_hw_control(adapter);
3090
3091         /*
3092          * before we allocate an interrupt, we must be ready to handle it.
3093          * Setting DEBUG_SHIRQ in the kernel makes it fire an interrupt
3094          * as soon as we call pci_request_irq, so we have to setup our
3095          * clean_rx handler before we do so.
3096          */
3097         e1000_configure(adapter);
3098
3099         err = e1000_request_irq(adapter);
3100         if (err)
3101                 goto err_req_irq;
3102
3103         /*
3104          * Work around PCIe errata with MSI interrupts causing some chipsets to
3105          * ignore e1000e MSI messages, which means we need to test our MSI
3106          * interrupt now
3107          */
3108         if (adapter->int_mode != E1000E_INT_MODE_LEGACY) {
3109                 err = e1000_test_msi(adapter);
3110                 if (err) {
3111                         e_err("Interrupt allocation failed\n");
3112                         goto err_req_irq;
3113                 }
3114         }
3115
3116         /* From here on the code is the same as e1000e_up() */
3117         clear_bit(__E1000_DOWN, &adapter->state);
3118
3119         napi_enable(&adapter->napi);
3120
3121         e1000_irq_enable(adapter);
3122
3123         netif_start_queue(netdev);
3124
3125         /* fire a link status change interrupt to start the watchdog */
3126         ew32(ICS, E1000_ICS_LSC);
3127
3128         return 0;
3129
3130 err_req_irq:
3131         e1000_release_hw_control(adapter);
3132         e1000_power_down_phy(adapter);
3133         e1000e_free_rx_resources(adapter);
3134 err_setup_rx:
3135         e1000e_free_tx_resources(adapter);
3136 err_setup_tx:
3137         e1000e_reset(adapter);
3138
3139         return err;
3140 }
3141
3142 /**
3143  * e1000_close - Disables a network interface
3144  * @netdev: network interface device structure
3145  *
3146  * Returns 0, this is not allowed to fail
3147  *
3148  * The close entry point is called when an interface is de-activated
3149  * by the OS.  The hardware is still under the drivers control, but
3150  * needs to be disabled.  A global MAC reset is issued to stop the
3151  * hardware, and all transmit and receive resources are freed.
3152  **/
3153 static int e1000_close(struct net_device *netdev)
3154 {
3155         struct e1000_adapter *adapter = netdev_priv(netdev);
3156
3157         WARN_ON(test_bit(__E1000_RESETTING, &adapter->state));
3158         e1000e_down(adapter);
3159         e1000_power_down_phy(adapter);
3160         e1000_free_irq(adapter);
3161
3162         e1000e_free_tx_resources(adapter);
3163         e1000e_free_rx_resources(adapter);
3164
3165         /*
3166          * kill manageability vlan ID if supported, but not if a vlan with
3167          * the same ID is registered on the host OS (let 8021q kill it)
3168          */
3169         if ((adapter->hw.mng_cookie.status &
3170                           E1000_MNG_DHCP_COOKIE_STATUS_VLAN) &&
3171              !(adapter->vlgrp &&
3172                vlan_group_get_device(adapter->vlgrp, adapter->mng_vlan_id)))
3173                 e1000_vlan_rx_kill_vid(netdev, adapter->mng_vlan_id);
3174
3175         /*
3176          * If AMT is enabled, let the firmware know that the network
3177          * interface is now closed
3178          */
3179         if (adapter->flags & FLAG_HAS_AMT)
3180                 e1000_release_hw_control(adapter);
3181
3182         return 0;
3183 }
3184 /**
3185  * e1000_set_mac - Change the Ethernet Address of the NIC
3186  * @netdev: network interface device structure
3187  * @p: pointer to an address structure
3188  *
3189  * Returns 0 on success, negative on failure
3190  **/
3191 static int e1000_set_mac(struct net_device *netdev, void *p)
3192 {
3193         struct e1000_adapter *adapter = netdev_priv(netdev);
3194         struct sockaddr *addr = p;
3195
3196         if (!is_valid_ether_addr(addr->sa_data))
3197                 return -EADDRNOTAVAIL;
3198
3199         memcpy(netdev->dev_addr, addr->sa_data, netdev->addr_len);
3200         memcpy(adapter->hw.mac.addr, addr->sa_data, netdev->addr_len);
3201
3202         e1000e_rar_set(&adapter->hw, adapter->hw.mac.addr, 0);
3203
3204         if (adapter->flags & FLAG_RESET_OVERWRITES_LAA) {
3205                 /* activate the work around */
3206                 e1000e_set_laa_state_82571(&adapter->hw, 1);
3207
3208                 /*
3209                  * Hold a copy of the LAA in RAR[14] This is done so that
3210                  * between the time RAR[0] gets clobbered  and the time it
3211                  * gets fixed (in e1000_watchdog), the actual LAA is in one
3212                  * of the RARs and no incoming packets directed to this port
3213                  * are dropped. Eventually the LAA will be in RAR[0] and
3214                  * RAR[14]
3215                  */
3216                 e1000e_rar_set(&adapter->hw,
3217                               adapter->hw.mac.addr,
3218                               adapter->hw.mac.rar_entry_count - 1);
3219         }
3220
3221         return 0;
3222 }
3223
3224 /**
3225  * e1000e_update_phy_task - work thread to update phy
3226  * @work: pointer to our work struct
3227  *
3228  * this worker thread exists because we must acquire a
3229  * semaphore to read the phy, which we could msleep while
3230  * waiting for it, and we can't msleep in a timer.
3231  **/
3232 static void e1000e_update_phy_task(struct work_struct *work)
3233 {
3234         struct e1000_adapter *adapter = container_of(work,
3235                                         struct e1000_adapter, update_phy_task);
3236         e1000_get_phy_info(&adapter->hw);
3237 }
3238
3239 /*
3240  * Need to wait a few seconds after link up to get diagnostic information from
3241  * the phy
3242  */
3243 static void e1000_update_phy_info(unsigned long data)
3244 {
3245         struct e1000_adapter *adapter = (struct e1000_adapter *) data;
3246         schedule_work(&adapter->update_phy_task);
3247 }
3248
3249 /**
3250  * e1000e_update_stats - Update the board statistics counters
3251  * @adapter: board private structure
3252  **/
3253 void e1000e_update_stats(struct e1000_adapter *adapter)
3254 {
3255         struct net_device *netdev = adapter->netdev;
3256         struct e1000_hw *hw = &adapter->hw;
3257         struct pci_dev *pdev = adapter->pdev;
3258         u16 phy_data;
3259
3260         /*
3261          * Prevent stats update while adapter is being reset, or if the pci
3262          * connection is down.
3263          */
3264         if (adapter->link_speed == 0)
3265                 return;
3266         if (pci_channel_offline(pdev))
3267                 return;
3268
3269         adapter->stats.crcerrs += er32(CRCERRS);
3270         adapter->stats.gprc += er32(GPRC);
3271         adapter->stats.gorc += er32(GORCL);
3272         er32(GORCH); /* Clear gorc */
3273         adapter->stats.bprc += er32(BPRC);
3274         adapter->stats.mprc += er32(MPRC);
3275         adapter->stats.roc += er32(ROC);
3276
3277         adapter->stats.mpc += er32(MPC);
3278         if ((hw->phy.type == e1000_phy_82578) ||
3279             (hw->phy.type == e1000_phy_82577)) {
3280                 e1e_rphy(hw, HV_SCC_UPPER, &phy_data);
3281                 e1e_rphy(hw, HV_SCC_LOWER, &phy_data);
3282                 adapter->stats.scc += phy_data;
3283
3284                 e1e_rphy(hw, HV_ECOL_UPPER, &phy_data);
3285                 e1e_rphy(hw, HV_ECOL_LOWER, &phy_data);
3286                 adapter->stats.ecol += phy_data;
3287
3288                 e1e_rphy(hw, HV_MCC_UPPER, &phy_data);
3289                 e1e_rphy(hw, HV_MCC_LOWER, &phy_data);
3290                 adapter->stats.mcc += phy_data;
3291
3292                 e1e_rphy(hw, HV_LATECOL_UPPER, &phy_data);
3293                 e1e_rphy(hw, HV_LATECOL_LOWER, &phy_data);
3294                 adapter->stats.latecol += phy_data;
3295
3296                 e1e_rphy(hw, HV_DC_UPPER, &phy_data);
3297                 e1e_rphy(hw, HV_DC_LOWER, &phy_data);
3298                 adapter->stats.dc += phy_data;
3299         } else {
3300                 adapter->stats.scc += er32(SCC);
3301                 adapter->stats.ecol += er32(ECOL);
3302                 adapter->stats.mcc += er32(MCC);
3303                 adapter->stats.latecol += er32(LATECOL);
3304                 adapter->stats.dc += er32(DC);
3305         }
3306         adapter->stats.xonrxc += er32(XONRXC);
3307         adapter->stats.xontxc += er32(XONTXC);
3308         adapter->stats.xoffrxc += er32(XOFFRXC);
3309         adapter->stats.xofftxc += er32(XOFFTXC);
3310         adapter->stats.gptc += er32(GPTC);
3311         adapter->stats.gotc += er32(GOTCL);
3312         er32(GOTCH); /* Clear gotc */
3313         adapter->stats.rnbc += er32(RNBC);
3314         adapter->stats.ruc += er32(RUC);
3315
3316         adapter->stats.mptc += er32(MPTC);
3317         adapter->stats.bptc += er32(BPTC);
3318
3319         /* used for adaptive IFS */
3320
3321         hw->mac.tx_packet_delta = er32(TPT);
3322         adapter->stats.tpt += hw->mac.tx_packet_delta;
3323         if ((hw->phy.type == e1000_phy_82578) ||
3324             (hw->phy.type == e1000_phy_82577)) {
3325                 e1e_rphy(hw, HV_COLC_UPPER, &phy_data);
3326                 e1e_rphy(hw, HV_COLC_LOWER, &phy_data);
3327                 hw->mac.collision_delta = phy_data;
3328         } else {
3329                 hw->mac.collision_delta = er32(COLC);
3330         }
3331         adapter->stats.colc += hw->mac.collision_delta;
3332
3333         adapter->stats.algnerrc += er32(ALGNERRC);
3334         adapter->stats.rxerrc += er32(RXERRC);
3335         if ((hw->phy.type == e1000_phy_82578) ||
3336             (hw->phy.type == e1000_phy_82577)) {
3337                 e1e_rphy(hw, HV_TNCRS_UPPER, &phy_data);
3338                 e1e_rphy(hw, HV_TNCRS_LOWER, &phy_data);
3339                 adapter->stats.tncrs += phy_data;
3340         } else {
3341                 if ((hw->mac.type != e1000_82574) &&
3342                     (hw->mac.type != e1000_82583))
3343                         adapter->stats.tncrs += er32(TNCRS);
3344         }
3345         adapter->stats.cexterr += er32(CEXTERR);
3346         adapter->stats.tsctc += er32(TSCTC);
3347         adapter->stats.tsctfc += er32(TSCTFC);
3348
3349         /* Fill out the OS statistics structure */
3350         netdev->stats.multicast = adapter->stats.mprc;
3351         netdev->stats.collisions = adapter->stats.colc;
3352
3353         /* Rx Errors */
3354
3355         /*
3356          * RLEC on some newer hardware can be incorrect so build
3357          * our own version based on RUC and ROC
3358          */
3359         netdev->stats.rx_errors = adapter->stats.rxerrc +
3360                 adapter->stats.crcerrs + adapter->stats.algnerrc +
3361                 adapter->stats.ruc + adapter->stats.roc +
3362                 adapter->stats.cexterr;
3363         netdev->stats.rx_length_errors = adapter->stats.ruc +
3364                                               adapter->stats.roc;
3365         netdev->stats.rx_crc_errors = adapter->stats.crcerrs;
3366         netdev->stats.rx_frame_errors = adapter->stats.algnerrc;
3367         netdev->stats.rx_missed_errors = adapter->stats.mpc;
3368
3369         /* Tx Errors */
3370         netdev->stats.tx_errors = adapter->stats.ecol +
3371                                        adapter->stats.latecol;
3372         netdev->stats.tx_aborted_errors = adapter->stats.ecol;
3373         netdev->stats.tx_window_errors = adapter->stats.latecol;
3374         netdev->stats.tx_carrier_errors = adapter->stats.tncrs;
3375
3376         /* Tx Dropped needs to be maintained elsewhere */
3377
3378         /* Management Stats */
3379         adapter->stats.mgptc += er32(MGTPTC);
3380         adapter->stats.mgprc += er32(MGTPRC);
3381         adapter->stats.mgpdc += er32(MGTPDC);
3382 }
3383
3384 /**
3385  * e1000_phy_read_status - Update the PHY register status snapshot
3386  * @adapter: board private structure
3387  **/
3388 static void e1000_phy_read_status(struct e1000_adapter *adapter)
3389 {
3390         struct e1000_hw *hw = &adapter->hw;
3391         struct e1000_phy_regs *phy = &adapter->phy_regs;
3392         int ret_val;
3393
3394         if ((er32(STATUS) & E1000_STATUS_LU) &&
3395             (adapter->hw.phy.media_type == e1000_media_type_copper)) {
3396                 ret_val  = e1e_rphy(hw, PHY_CONTROL, &phy->bmcr);
3397                 ret_val |= e1e_rphy(hw, PHY_STATUS, &phy->bmsr);
3398                 ret_val |= e1e_rphy(hw, PHY_AUTONEG_ADV, &phy->advertise);
3399                 ret_val |= e1e_rphy(hw, PHY_LP_ABILITY, &phy->lpa);
3400                 ret_val |= e1e_rphy(hw, PHY_AUTONEG_EXP, &phy->expansion);
3401                 ret_val |= e1e_rphy(hw, PHY_1000T_CTRL, &phy->ctrl1000);
3402                 ret_val |= e1e_rphy(hw, PHY_1000T_STATUS, &phy->stat1000);
3403                 ret_val |= e1e_rphy(hw, PHY_EXT_STATUS, &phy->estatus);
3404                 if (ret_val)
3405                         e_warn("Error reading PHY register\n");
3406         } else {
3407                 /*
3408                  * Do not read PHY registers if link is not up
3409                  * Set values to typical power-on defaults
3410                  */
3411                 phy->bmcr = (BMCR_SPEED1000 | BMCR_ANENABLE | BMCR_FULLDPLX);
3412                 phy->bmsr = (BMSR_100FULL | BMSR_100HALF | BMSR_10FULL |
3413                              BMSR_10HALF | BMSR_ESTATEN | BMSR_ANEGCAPABLE |
3414                              BMSR_ERCAP);
3415                 phy->advertise = (ADVERTISE_PAUSE_ASYM | ADVERTISE_PAUSE_CAP |
3416                                   ADVERTISE_ALL | ADVERTISE_CSMA);
3417                 phy->lpa = 0;
3418                 phy->expansion = EXPANSION_ENABLENPAGE;
3419                 phy->ctrl1000 = ADVERTISE_1000FULL;
3420                 phy->stat1000 = 0;
3421                 phy->estatus = (ESTATUS_1000_TFULL | ESTATUS_1000_THALF);
3422         }
3423 }
3424
3425 static void e1000_print_link_info(struct e1000_adapter *adapter)
3426 {
3427         struct e1000_hw *hw = &adapter->hw;
3428         u32 ctrl = er32(CTRL);
3429
3430         /* Link status message must follow this format for user tools */
3431         printk(KERN_INFO "e1000e: %s NIC Link is Up %d Mbps %s, "
3432                "Flow Control: %s\n",
3433                adapter->netdev->name,
3434                adapter->link_speed,
3435                (adapter->link_duplex == FULL_DUPLEX) ?
3436                                 "Full Duplex" : "Half Duplex",
3437                ((ctrl & E1000_CTRL_TFCE) && (ctrl & E1000_CTRL_RFCE)) ?
3438                                 "RX/TX" :
3439                ((ctrl & E1000_CTRL_RFCE) ? "RX" :
3440                ((ctrl & E1000_CTRL_TFCE) ? "TX" : "None" )));
3441 }
3442
3443 bool e1000_has_link(struct e1000_adapter *adapter)
3444 {
3445         struct e1000_hw *hw = &adapter->hw;
3446         bool link_active = 0;
3447         s32 ret_val = 0;
3448
3449         /*
3450          * get_link_status is set on LSC (link status) interrupt or
3451          * Rx sequence error interrupt.  get_link_status will stay
3452          * false until the check_for_link establishes link
3453          * for copper adapters ONLY
3454          */
3455         switch (hw->phy.media_type) {
3456         case e1000_media_type_copper:
3457                 if (hw->mac.get_link_status) {
3458                         ret_val = hw->mac.ops.check_for_link(hw);
3459                         link_active = !hw->mac.get_link_status;
3460                 } else {
3461                         link_active = 1;
3462                 }
3463                 break;
3464         case e1000_media_type_fiber:
3465                 ret_val = hw->mac.ops.check_for_link(hw);
3466                 link_active = !!(er32(STATUS) & E1000_STATUS_LU);
3467                 break;
3468         case e1000_media_type_internal_serdes:
3469                 ret_val = hw->mac.ops.check_for_link(hw);
3470                 link_active = adapter->hw.mac.serdes_has_link;
3471                 break;
3472         default:
3473         case e1000_media_type_unknown:
3474                 break;
3475         }
3476
3477         if ((ret_val == E1000_ERR_PHY) && (hw->phy.type == e1000_phy_igp_3) &&
3478             (er32(CTRL) & E1000_PHY_CTRL_GBE_DISABLE)) {
3479                 /* See e1000_kmrn_lock_loss_workaround_ich8lan() */
3480                 e_info("Gigabit has been disabled, downgrading speed\n");
3481         }
3482
3483         return link_active;
3484 }
3485
3486 static void e1000e_enable_receives(struct e1000_adapter *adapter)
3487 {
3488         /* make sure the receive unit is started */
3489         if ((adapter->flags & FLAG_RX_NEEDS_RESTART) &&
3490             (adapter->flags & FLAG_RX_RESTART_NOW)) {
3491                 struct e1000_hw *hw = &adapter->hw;
3492                 u32 rctl = er32(RCTL);
3493                 ew32(RCTL, rctl | E1000_RCTL_EN);
3494                 adapter->flags &= ~FLAG_RX_RESTART_NOW;
3495         }
3496 }
3497
3498 /**
3499  * e1000_watchdog - Timer Call-back
3500  * @data: pointer to adapter cast into an unsigned long
3501  **/
3502 static void e1000_watchdog(unsigned long data)
3503 {
3504         struct e1000_adapter *adapter = (struct e1000_adapter *) data;
3505
3506         /* Do the rest outside of interrupt context */
3507         schedule_work(&adapter->watchdog_task);
3508
3509         /* TODO: make this use queue_delayed_work() */
3510 }
3511
3512 static void e1000_watchdog_task(struct work_struct *work)
3513 {
3514         struct e1000_adapter *adapter = container_of(work,
3515                                         struct e1000_adapter, watchdog_task);
3516         struct net_device *netdev = adapter->netdev;
3517         struct e1000_mac_info *mac = &adapter->hw.mac;
3518         struct e1000_phy_info *phy = &adapter->hw.phy;
3519         struct e1000_ring *tx_ring = adapter->tx_ring;
3520         struct e1000_hw *hw = &adapter->hw;
3521         u32 link, tctl;
3522         int tx_pending = 0;
3523
3524         link = e1000_has_link(adapter);
3525         if ((netif_carrier_ok(netdev)) && link) {
3526                 e1000e_enable_receives(adapter);
3527                 goto link_up;
3528         }
3529
3530         if ((e1000e_enable_tx_pkt_filtering(hw)) &&
3531             (adapter->mng_vlan_id != adapter->hw.mng_cookie.vlan_id))
3532                 e1000_update_mng_vlan(adapter);
3533
3534         if (link) {
3535                 if (!netif_carrier_ok(netdev)) {
3536                         bool txb2b = 1;
3537                         /* update snapshot of PHY registers on LSC */
3538                         e1000_phy_read_status(adapter);
3539                         mac->ops.get_link_up_info(&adapter->hw,
3540                                                    &adapter->link_speed,
3541                                                    &adapter->link_duplex);
3542                         e1000_print_link_info(adapter);
3543                         /*
3544                          * On supported PHYs, check for duplex mismatch only
3545                          * if link has autonegotiated at 10/100 half
3546                          */
3547                         if ((hw->phy.type == e1000_phy_igp_3 ||
3548                              hw->phy.type == e1000_phy_bm) &&
3549                             (hw->mac.autoneg == true) &&
3550                             (adapter->link_speed == SPEED_10 ||
3551                              adapter->link_speed == SPEED_100) &&
3552                             (adapter->link_duplex == HALF_DUPLEX)) {
3553                                 u16 autoneg_exp;
3554
3555                                 e1e_rphy(hw, PHY_AUTONEG_EXP, &autoneg_exp);
3556
3557                                 if (!(autoneg_exp & NWAY_ER_LP_NWAY_CAPS))
3558                                         e_info("Autonegotiated half duplex but"
3559                                                " link partner cannot autoneg. "
3560                                                " Try forcing full duplex if "
3561                                                "link gets many collisions.\n");
3562                         }
3563
3564                         /*
3565                          * tweak tx_queue_len according to speed/duplex
3566                          * and adjust the timeout factor
3567                          */
3568                         netdev->tx_queue_len = adapter->tx_queue_len;
3569                         adapter->tx_timeout_factor = 1;
3570                         switch (adapter->link_speed) {
3571                         case SPEED_10:
3572                                 txb2b = 0;
3573                                 netdev->tx_queue_len = 10;
3574                                 adapter->tx_timeout_factor = 16;
3575                                 break;
3576                         case SPEED_100:
3577                                 txb2b = 0;
3578                                 netdev->tx_queue_len = 100;
3579                                 /* maybe add some timeout factor ? */
3580                                 break;
3581                         }
3582
3583                         /*
3584                          * workaround: re-program speed mode bit after
3585                          * link-up event
3586                          */
3587                         if ((adapter->flags & FLAG_TARC_SPEED_MODE_BIT) &&
3588                             !txb2b) {
3589                                 u32 tarc0;
3590                                 tarc0 = er32(TARC(0));
3591                                 tarc0 &= ~SPEED_MODE_BIT;
3592                                 ew32(TARC(0), tarc0);
3593                         }
3594
3595                         /*
3596                          * disable TSO for pcie and 10/100 speeds, to avoid
3597                          * some hardware issues
3598                          */
3599                         if (!(adapter->flags & FLAG_TSO_FORCE)) {
3600                                 switch (adapter->link_speed) {
3601                                 case SPEED_10:
3602                                 case SPEED_100:
3603                                         e_info("10/100 speed: disabling TSO\n");
3604                                         netdev->features &= ~NETIF_F_TSO;
3605                                         netdev->features &= ~NETIF_F_TSO6;
3606                                         break;
3607                                 case SPEED_1000:
3608                                         netdev->features |= NETIF_F_TSO;
3609                                         netdev->features |= NETIF_F_TSO6;
3610                                         break;
3611                                 default:
3612                                         /* oops */
3613                                         break;
3614                                 }
3615                         }
3616
3617                         /*
3618                          * enable transmits in the hardware, need to do this
3619                          * after setting TARC(0)
3620                          */
3621                         tctl = er32(TCTL);
3622                         tctl |= E1000_TCTL_EN;
3623                         ew32(TCTL, tctl);
3624
3625                         /*
3626                          * Perform any post-link-up configuration before
3627                          * reporting link up.
3628                          */
3629                         if (phy->ops.cfg_on_link_up)
3630                                 phy->ops.cfg_on_link_up(hw);
3631
3632                         netif_carrier_on(netdev);
3633
3634                         if (!test_bit(__E1000_DOWN, &adapter->state))
3635                                 mod_timer(&adapter->phy_info_timer,
3636                                           round_jiffies(jiffies + 2 * HZ));
3637                 }
3638         } else {
3639                 if (netif_carrier_ok(netdev)) {
3640                         adapter->link_speed = 0;
3641                         adapter->link_duplex = 0;
3642                         /* Link status message must follow this format */
3643                         printk(KERN_INFO "e1000e: %s NIC Link is Down\n",
3644                                adapter->netdev->name);
3645                         netif_carrier_off(netdev);
3646                         if (!test_bit(__E1000_DOWN, &adapter->state))
3647                                 mod_timer(&adapter->phy_info_timer,
3648                                           round_jiffies(jiffies + 2 * HZ));
3649
3650                         if (adapter->flags & FLAG_RX_NEEDS_RESTART)
3651                                 schedule_work(&adapter->reset_task);
3652                 }
3653         }
3654
3655 link_up:
3656         e1000e_update_stats(adapter);
3657
3658         mac->tx_packet_delta = adapter->stats.tpt - adapter->tpt_old;
3659         adapter->tpt_old = adapter->stats.tpt;
3660         mac->collision_delta = adapter->stats.colc - adapter->colc_old;
3661         adapter->colc_old = adapter->stats.colc;
3662
3663         adapter->gorc = adapter->stats.gorc - adapter->gorc_old;
3664         adapter->gorc_old = adapter->stats.gorc;
3665         adapter->gotc = adapter->stats.gotc - adapter->gotc_old;
3666         adapter->gotc_old = adapter->stats.gotc;
3667
3668         e1000e_update_adaptive(&adapter->hw);
3669
3670         if (!netif_carrier_ok(netdev)) {
3671                 tx_pending = (e1000_desc_unused(tx_ring) + 1 <
3672                                tx_ring->count);
3673                 if (tx_pending) {
3674                         /*
3675                          * We've lost link, so the controller stops DMA,
3676                          * but we've got queued Tx work that's never going
3677                          * to get done, so reset controller to flush Tx.
3678                          * (Do the reset outside of interrupt context).
3679                          */
3680                         adapter->tx_timeout_count++;
3681                         schedule_work(&adapter->reset_task);
3682                         /* return immediately since reset is imminent */
3683                         return;
3684                 }
3685         }
3686
3687         /* Cause software interrupt to ensure Rx ring is cleaned */
3688         if (adapter->msix_entries)
3689                 ew32(ICS, adapter->rx_ring->ims_val);
3690         else
3691                 ew32(ICS, E1000_ICS_RXDMT0);
3692
3693         /* Force detection of hung controller every watchdog period */
3694         adapter->detect_tx_hung = 1;
3695
3696         /*
3697          * With 82571 controllers, LAA may be overwritten due to controller
3698          * reset from the other port. Set the appropriate LAA in RAR[0]
3699          */
3700         if (e1000e_get_laa_state_82571(hw))
3701                 e1000e_rar_set(hw, adapter->hw.mac.addr, 0);
3702
3703         /* Reset the timer */
3704         if (!test_bit(__E1000_DOWN, &adapter->state))
3705                 mod_timer(&adapter->watchdog_timer,
3706                           round_jiffies(jiffies + 2 * HZ));
3707 }
3708
3709 #define E1000_TX_FLAGS_CSUM             0x00000001
3710 #define E1000_TX_FLAGS_VLAN             0x00000002
3711 #define E1000_TX_FLAGS_TSO              0x00000004
3712 #define E1000_TX_FLAGS_IPV4             0x00000008
3713 #define E1000_TX_FLAGS_VLAN_MASK        0xffff0000
3714 #define E1000_TX_FLAGS_VLAN_SHIFT       16
3715
3716 static int e1000_tso(struct e1000_adapter *adapter,
3717                      struct sk_buff *skb)
3718 {
3719         struct e1000_ring *tx_ring = adapter->tx_ring;
3720         struct e1000_context_desc *context_desc;
3721         struct e1000_buffer *buffer_info;
3722         unsigned int i;
3723         u32 cmd_length = 0;
3724         u16 ipcse = 0, tucse, mss;
3725         u8 ipcss, ipcso, tucss, tucso, hdr_len;
3726         int err;
3727
3728         if (skb_is_gso(skb)) {
3729                 if (skb_header_cloned(skb)) {
3730                         err = pskb_expand_head(skb, 0, 0, GFP_ATOMIC);
3731                         if (err)
3732                                 return err;
3733                 }
3734
3735                 hdr_len = skb_transport_offset(skb) + tcp_hdrlen(skb);
3736                 mss = skb_shinfo(skb)->gso_size;
3737                 if (skb->protocol == htons(ETH_P_IP)) {
3738                         struct iphdr *iph = ip_hdr(skb);
3739                         iph->tot_len = 0;
3740                         iph->check = 0;
3741                         tcp_hdr(skb)->check = ~csum_tcpudp_magic(iph->saddr,
3742                                                                  iph->daddr, 0,
3743                                                                  IPPROTO_TCP,
3744                                                                  0);
3745                         cmd_length = E1000_TXD_CMD_IP;
3746                         ipcse = skb_transport_offset(skb) - 1;
3747                 } else if (skb_shinfo(skb)->gso_type == SKB_GSO_TCPV6) {
3748                         ipv6_hdr(skb)->payload_len = 0;
3749                         tcp_hdr(skb)->check =
3750                                 ~csum_ipv6_magic(&ipv6_hdr(skb)->saddr,
3751                                                  &ipv6_hdr(skb)->daddr,
3752                                                  0, IPPROTO_TCP, 0);
3753                         ipcse = 0;
3754                 }
3755                 ipcss = skb_network_offset(skb);
3756                 ipcso = (void *)&(ip_hdr(skb)->check) - (void *)skb->data;
3757                 tucss = skb_transport_offset(skb);
3758                 tucso = (void *)&(tcp_hdr(skb)->check) - (void *)skb->data;
3759                 tucse = 0;
3760
3761                 cmd_length |= (E1000_TXD_CMD_DEXT | E1000_TXD_CMD_TSE |
3762                                E1000_TXD_CMD_TCP | (skb->len - (hdr_len)));
3763
3764                 i = tx_ring->next_to_use;
3765                 context_desc = E1000_CONTEXT_DESC(*tx_ring, i);
3766                 buffer_info = &tx_ring->buffer_info[i];
3767
3768                 context_desc->lower_setup.ip_fields.ipcss  = ipcss;
3769                 context_desc->lower_setup.ip_fields.ipcso  = ipcso;
3770                 context_desc->lower_setup.ip_fields.ipcse  = cpu_to_le16(ipcse);
3771                 context_desc->upper_setup.tcp_fields.tucss = tucss;
3772                 context_desc->upper_setup.tcp_fields.tucso = tucso;
3773                 context_desc->upper_setup.tcp_fields.tucse = cpu_to_le16(tucse);
3774                 context_desc->tcp_seg_setup.fields.mss     = cpu_to_le16(mss);
3775                 context_desc->tcp_seg_setup.fields.hdr_len = hdr_len;
3776                 context_desc->cmd_and_length = cpu_to_le32(cmd_length);
3777
3778                 buffer_info->time_stamp = jiffies;
3779                 buffer_info->next_to_watch = i;
3780
3781                 i++;
3782                 if (i == tx_ring->count)
3783                         i = 0;
3784                 tx_ring->next_to_use = i;
3785
3786                 return 1;
3787         }
3788
3789         return 0;
3790 }
3791
3792 static bool e1000_tx_csum(struct e1000_adapter *adapter, struct sk_buff *skb)
3793 {
3794         struct e1000_ring *tx_ring = adapter->tx_ring;
3795         struct e1000_context_desc *context_desc;
3796         struct e1000_buffer *buffer_info;
3797         unsigned int i;
3798         u8 css;
3799         u32 cmd_len = E1000_TXD_CMD_DEXT;
3800         __be16 protocol;
3801
3802         if (skb->ip_summed != CHECKSUM_PARTIAL)
3803                 return 0;
3804
3805         if (skb->protocol == cpu_to_be16(ETH_P_8021Q))
3806                 protocol = vlan_eth_hdr(skb)->h_vlan_encapsulated_proto;
3807         else
3808                 protocol = skb->protocol;
3809
3810         switch (protocol) {
3811         case cpu_to_be16(ETH_P_IP):
3812                 if (ip_hdr(skb)->protocol == IPPROTO_TCP)
3813                         cmd_len |= E1000_TXD_CMD_TCP;
3814                 break;
3815         case cpu_to_be16(ETH_P_IPV6):
3816                 /* XXX not handling all IPV6 headers */
3817                 if (ipv6_hdr(skb)->nexthdr == IPPROTO_TCP)
3818                         cmd_len |= E1000_TXD_CMD_TCP;
3819                 break;
3820         default:
3821                 if (unlikely(net_ratelimit()))
3822                         e_warn("checksum_partial proto=%x!\n",
3823                                be16_to_cpu(protocol));
3824                 break;
3825         }
3826
3827         css = skb_transport_offset(skb);
3828
3829         i = tx_ring->next_to_use;
3830         buffer_info = &tx_ring->buffer_info[i];
3831         context_desc = E1000_CONTEXT_DESC(*tx_ring, i);
3832
3833         context_desc->lower_setup.ip_config = 0;
3834         context_desc->upper_setup.tcp_fields.tucss = css;
3835         context_desc->upper_setup.tcp_fields.tucso =
3836                                 css + skb->csum_offset;
3837         context_desc->upper_setup.tcp_fields.tucse = 0;
3838         context_desc->tcp_seg_setup.data = 0;
3839         context_desc->cmd_and_length = cpu_to_le32(cmd_len);
3840
3841         buffer_info->time_stamp = jiffies;
3842         buffer_info->next_to_watch = i;
3843
3844         i++;
3845         if (i == tx_ring->count)
3846                 i = 0;
3847         tx_ring->next_to_use = i;
3848
3849         return 1;
3850 }
3851
3852 #define E1000_MAX_PER_TXD       8192
3853 #define E1000_MAX_TXD_PWR       12
3854
3855 static int e1000_tx_map(struct e1000_adapter *adapter,
3856                         struct sk_buff *skb, unsigned int first,
3857                         unsigned int max_per_txd, unsigned int nr_frags,
3858                         unsigned int mss)
3859 {
3860         struct e1000_ring *tx_ring = adapter->tx_ring;
3861         struct e1000_buffer *buffer_info;
3862         unsigned int len = skb_headlen(skb);
3863         unsigned int offset, size, count = 0, i;
3864         unsigned int f;
3865         dma_addr_t *map;
3866
3867         i = tx_ring->next_to_use;
3868
3869         if (skb_dma_map(&adapter->pdev->dev, skb, DMA_TO_DEVICE)) {
3870                 dev_err(&adapter->pdev->dev, "TX DMA map failed\n");
3871                 adapter->tx_dma_failed++;
3872                 return 0;
3873         }
3874
3875         map = skb_shinfo(skb)->dma_maps;
3876         offset = 0;
3877
3878         while (len) {
3879                 buffer_info = &tx_ring->buffer_info[i];
3880                 size = min(len, max_per_txd);
3881
3882                 buffer_info->length = size;
3883                 buffer_info->time_stamp = jiffies;
3884                 buffer_info->next_to_watch = i;
3885                 buffer_info->dma = skb_shinfo(skb)->dma_head + offset;
3886                 count++;
3887
3888                 len -= size;
3889                 offset += size;
3890
3891                 if (len) {
3892                         i++;
3893                         if (i == tx_ring->count)
3894                                 i = 0;
3895                 }
3896         }
3897
3898         for (f = 0; f < nr_frags; f++) {
3899                 struct skb_frag_struct *frag;
3900
3901                 frag = &skb_shinfo(skb)->frags[f];
3902                 len = frag->size;
3903                 offset = 0;
3904
3905                 while (len) {
3906                         i++;
3907                         if (i == tx_ring->count)
3908                                 i = 0;
3909
3910                         buffer_info = &tx_ring->buffer_info[i];
3911                         size = min(len, max_per_txd);
3912
3913                         buffer_info->length = size;
3914                         buffer_info->time_stamp = jiffies;
3915                         buffer_info->next_to_watch = i;
3916                         buffer_info->dma = map[f] + offset;
3917
3918                         len -= size;
3919                         offset += size;
3920                         count++;
3921                 }
3922         }
3923
3924         tx_ring->buffer_info[i].skb = skb;
3925         tx_ring->buffer_info[first].next_to_watch = i;
3926
3927         return count;
3928 }
3929
3930 static void e1000_tx_queue(struct e1000_adapter *adapter,
3931                            int tx_flags, int count)
3932 {
3933         struct e1000_ring *tx_ring = adapter->tx_ring;
3934         struct e1000_tx_desc *tx_desc = NULL;
3935         struct e1000_buffer *buffer_info;
3936         u32 txd_upper = 0, txd_lower = E1000_TXD_CMD_IFCS;
3937         unsigned int i;
3938
3939         if (tx_flags & E1000_TX_FLAGS_TSO) {
3940                 txd_lower |= E1000_TXD_CMD_DEXT | E1000_TXD_DTYP_D |
3941                              E1000_TXD_CMD_TSE;
3942                 txd_upper |= E1000_TXD_POPTS_TXSM << 8;
3943
3944                 if (tx_flags & E1000_TX_FLAGS_IPV4)
3945                         txd_upper |= E1000_TXD_POPTS_IXSM << 8;
3946         }
3947
3948         if (tx_flags & E1000_TX_FLAGS_CSUM) {
3949                 txd_lower |= E1000_TXD_CMD_DEXT | E1000_TXD_DTYP_D;
3950                 txd_upper |= E1000_TXD_POPTS_TXSM << 8;
3951         }
3952
3953         if (tx_flags & E1000_TX_FLAGS_VLAN) {
3954                 txd_lower |= E1000_TXD_CMD_VLE;
3955                 txd_upper |= (tx_flags & E1000_TX_FLAGS_VLAN_MASK);
3956         }
3957
3958         i = tx_ring->next_to_use;
3959
3960         while (count--) {
3961                 buffer_info = &tx_ring->buffer_info[i];
3962                 tx_desc = E1000_TX_DESC(*tx_ring, i);
3963                 tx_desc->buffer_addr = cpu_to_le64(buffer_info->dma);
3964                 tx_desc->lower.data =
3965                         cpu_to_le32(txd_lower | buffer_info->length);
3966                 tx_desc->upper.data = cpu_to_le32(txd_upper);
3967
3968                 i++;
3969                 if (i == tx_ring->count)
3970                         i = 0;
3971         }
3972
3973         tx_desc->lower.data |= cpu_to_le32(adapter->txd_cmd);
3974
3975         /*
3976          * Force memory writes to complete before letting h/w
3977          * know there are new descriptors to fetch.  (Only
3978          * applicable for weak-ordered memory model archs,
3979          * such as IA-64).
3980          */
3981         wmb();
3982
3983         tx_ring->next_to_use = i;
3984         writel(i, adapter->hw.hw_addr + tx_ring->tail);
3985         /*
3986          * we need this if more than one processor can write to our tail
3987          * at a time, it synchronizes IO on IA64/Altix systems
3988          */
3989         mmiowb();
3990 }
3991
3992 #define MINIMUM_DHCP_PACKET_SIZE 282
3993 static int e1000_transfer_dhcp_info(struct e1000_adapter *adapter,
3994                                     struct sk_buff *skb)
3995 {
3996         struct e1000_hw *hw =  &adapter->hw;
3997         u16 length, offset;
3998
3999         if (vlan_tx_tag_present(skb)) {
4000                 if (!((vlan_tx_tag_get(skb) == adapter->hw.mng_cookie.vlan_id)
4001                     && (adapter->hw.mng_cookie.status &
4002                         E1000_MNG_DHCP_COOKIE_STATUS_VLAN)))
4003                         return 0;
4004         }
4005
4006         if (skb->len <= MINIMUM_DHCP_PACKET_SIZE)
4007                 return 0;
4008
4009         if (((struct ethhdr *) skb->data)->h_proto != htons(ETH_P_IP))
4010                 return 0;
4011
4012         {
4013                 const struct iphdr *ip = (struct iphdr *)((u8 *)skb->data+14);
4014                 struct udphdr *udp;
4015
4016                 if (ip->protocol != IPPROTO_UDP)
4017                         return 0;
4018
4019                 udp = (struct udphdr *)((u8 *)ip + (ip->ihl << 2));
4020                 if (ntohs(udp->dest) != 67)
4021                         return 0;
4022
4023                 offset = (u8 *)udp + 8 - skb->data;
4024                 length = skb->len - offset;
4025                 return e1000e_mng_write_dhcp_info(hw, (u8 *)udp + 8, length);
4026         }
4027
4028         return 0;
4029 }
4030
4031 static int __e1000_maybe_stop_tx(struct net_device *netdev, int size)
4032 {
4033         struct e1000_adapter *adapter = netdev_priv(netdev);
4034
4035         netif_stop_queue(netdev);
4036         /*
4037          * Herbert's original patch had:
4038          *  smp_mb__after_netif_stop_queue();
4039          * but since that doesn't exist yet, just open code it.
4040          */
4041         smp_mb();
4042
4043         /*
4044          * We need to check again in a case another CPU has just
4045          * made room available.
4046          */
4047         if (e1000_desc_unused(adapter->tx_ring) < size)
4048                 return -EBUSY;
4049
4050         /* A reprieve! */
4051         netif_start_queue(netdev);
4052         ++adapter->restart_queue;
4053         return 0;
4054 }
4055
4056 static int e1000_maybe_stop_tx(struct net_device *netdev, int size)
4057 {
4058         struct e1000_adapter *adapter = netdev_priv(netdev);
4059
4060         if (e1000_desc_unused(adapter->tx_ring) >= size)
4061                 return 0;
4062         return __e1000_maybe_stop_tx(netdev, size);
4063 }
4064
4065 #define TXD_USE_COUNT(S, X) (((S) >> (X)) + 1 )
4066 static netdev_tx_t e1000_xmit_frame(struct sk_buff *skb,
4067                                     struct net_device *netdev)
4068 {
4069         struct e1000_adapter *adapter = netdev_priv(netdev);
4070         struct e1000_ring *tx_ring = adapter->tx_ring;
4071         unsigned int first;
4072         unsigned int max_per_txd = E1000_MAX_PER_TXD;
4073         unsigned int max_txd_pwr = E1000_MAX_TXD_PWR;
4074         unsigned int tx_flags = 0;
4075         unsigned int len = skb->len - skb->data_len;
4076         unsigned int nr_frags;
4077         unsigned int mss;
4078         int count = 0;
4079         int tso;
4080         unsigned int f;
4081
4082         if (test_bit(__E1000_DOWN, &adapter->state)) {
4083                 dev_kfree_skb_any(skb);
4084                 return NETDEV_TX_OK;
4085         }
4086
4087         if (skb->len <= 0) {
4088                 dev_kfree_skb_any(skb);
4089                 return NETDEV_TX_OK;
4090         }
4091
4092         mss = skb_shinfo(skb)->gso_size;
4093         /*
4094          * The controller does a simple calculation to
4095          * make sure there is enough room in the FIFO before
4096          * initiating the DMA for each buffer.  The calc is:
4097          * 4 = ceil(buffer len/mss).  To make sure we don't
4098          * overrun the FIFO, adjust the max buffer len if mss
4099          * drops.
4100          */
4101         if (mss) {
4102                 u8 hdr_len;
4103                 max_per_txd = min(mss << 2, max_per_txd);
4104                 max_txd_pwr = fls(max_per_txd) - 1;
4105
4106                 /*
4107                  * TSO Workaround for 82571/2/3 Controllers -- if skb->data
4108                  * points to just header, pull a few bytes of payload from
4109                  * frags into skb->data
4110                  */
4111                 hdr_len = skb_transport_offset(skb) + tcp_hdrlen(skb);
4112                 /*
4113                  * we do this workaround for ES2LAN, but it is un-necessary,
4114                  * avoiding it could save a lot of cycles
4115                  */
4116                 if (skb->data_len && (hdr_len == len)) {
4117                         unsigned int pull_size;
4118
4119                         pull_size = min((unsigned int)4, skb->data_len);
4120                         if (!__pskb_pull_tail(skb, pull_size)) {
4121                                 e_err("__pskb_pull_tail failed.\n");
4122                                 dev_kfree_skb_any(skb);
4123                                 return NETDEV_TX_OK;
4124                         }
4125                         len = skb->len - skb->data_len;
4126                 }
4127         }
4128
4129         /* reserve a descriptor for the offload context */
4130         if ((mss) || (skb->ip_summed == CHECKSUM_PARTIAL))
4131                 count++;
4132         count++;
4133
4134         count += TXD_USE_COUNT(len, max_txd_pwr);
4135
4136         nr_frags = skb_shinfo(skb)->nr_frags;
4137         for (f = 0; f < nr_frags; f++)
4138                 count += TXD_USE_COUNT(skb_shinfo(skb)->frags[f].size,
4139                                        max_txd_pwr);
4140
4141         if (adapter->hw.mac.tx_pkt_filtering)
4142                 e1000_transfer_dhcp_info(adapter, skb);
4143
4144         /*
4145          * need: count + 2 desc gap to keep tail from touching
4146          * head, otherwise try next time
4147          */
4148         if (e1000_maybe_stop_tx(netdev, count + 2))
4149                 return NETDEV_TX_BUSY;
4150
4151         if (adapter->vlgrp && vlan_tx_tag_present(skb)) {
4152                 tx_flags |= E1000_TX_FLAGS_VLAN;
4153                 tx_flags |= (vlan_tx_tag_get(skb) << E1000_TX_FLAGS_VLAN_SHIFT);
4154         }
4155
4156         first = tx_ring->next_to_use;
4157
4158         tso = e1000_tso(adapter, skb);
4159         if (tso < 0) {
4160                 dev_kfree_skb_any(skb);
4161                 return NETDEV_TX_OK;
4162         }
4163
4164         if (tso)
4165                 tx_flags |= E1000_TX_FLAGS_TSO;
4166         else if (e1000_tx_csum(adapter, skb))
4167                 tx_flags |= E1000_TX_FLAGS_CSUM;
4168
4169         /*
4170          * Old method was to assume IPv4 packet by default if TSO was enabled.
4171          * 82571 hardware supports TSO capabilities for IPv6 as well...
4172          * no longer assume, we must.
4173          */
4174         if (skb->protocol == htons(ETH_P_IP))
4175                 tx_flags |= E1000_TX_FLAGS_IPV4;
4176
4177         /* if count is 0 then mapping error has occured */
4178         count = e1000_tx_map(adapter, skb, first, max_per_txd, nr_frags, mss);
4179         if (count) {
4180                 e1000_tx_queue(adapter, tx_flags, count);
4181                 /* Make sure there is space in the ring for the next send. */
4182                 e1000_maybe_stop_tx(netdev, MAX_SKB_FRAGS + 2);
4183
4184         } else {
4185                 dev_kfree_skb_any(skb);
4186                 tx_ring->buffer_info[first].time_stamp = 0;
4187                 tx_ring->next_to_use = first;
4188         }
4189
4190         return NETDEV_TX_OK;
4191 }
4192
4193 /**
4194  * e1000_tx_timeout - Respond to a Tx Hang
4195  * @netdev: network interface device structure
4196  **/
4197 static void e1000_tx_timeout(struct net_device *netdev)
4198 {
4199         struct e1000_adapter *adapter = netdev_priv(netdev);
4200
4201         /* Do the reset outside of interrupt context */
4202         adapter->tx_timeout_count++;
4203         schedule_work(&adapter->reset_task);
4204 }
4205
4206 static void e1000_reset_task(struct work_struct *work)
4207 {
4208         struct e1000_adapter *adapter;
4209         adapter = container_of(work, struct e1000_adapter, reset_task);
4210
4211         e1000e_reinit_locked(adapter);
4212 }
4213
4214 /**
4215  * e1000_get_stats - Get System Network Statistics
4216  * @netdev: network interface device structure
4217  *
4218  * Returns the address of the device statistics structure.
4219  * The statistics are actually updated from the timer callback.
4220  **/
4221 static struct net_device_stats *e1000_get_stats(struct net_device *netdev)
4222 {
4223         /* only return the current stats */
4224         return &netdev->stats;
4225 }
4226
4227 /**
4228  * e1000_change_mtu - Change the Maximum Transfer Unit
4229  * @netdev: network interface device structure
4230  * @new_mtu: new value for maximum frame size
4231  *
4232  * Returns 0 on success, negative on failure
4233  **/
4234 static int e1000_change_mtu(struct net_device *netdev, int new_mtu)
4235 {
4236         struct e1000_adapter *adapter = netdev_priv(netdev);
4237         int max_frame = new_mtu + ETH_HLEN + ETH_FCS_LEN;
4238
4239         /* Jumbo frame support */
4240         if ((max_frame > ETH_FRAME_LEN + ETH_FCS_LEN) &&
4241             !(adapter->flags & FLAG_HAS_JUMBO_FRAMES)) {
4242                 e_err("Jumbo Frames not supported.\n");
4243                 return -EINVAL;
4244         }
4245
4246         /* Supported frame sizes */
4247         if ((new_mtu < ETH_ZLEN + ETH_FCS_LEN + VLAN_HLEN) ||
4248             (max_frame > adapter->max_hw_frame_size)) {
4249                 e_err("Unsupported MTU setting\n");
4250                 return -EINVAL;
4251         }
4252
4253         while (test_and_set_bit(__E1000_RESETTING, &adapter->state))
4254                 msleep(1);
4255         /* e1000e_down has a dependency on max_frame_size */
4256         adapter->max_frame_size = max_frame;
4257         if (netif_running(netdev))
4258                 e1000e_down(adapter);
4259
4260         /*
4261          * NOTE: netdev_alloc_skb reserves 16 bytes, and typically NET_IP_ALIGN
4262          * means we reserve 2 more, this pushes us to allocate from the next
4263          * larger slab size.
4264          * i.e. RXBUFFER_2048 --> size-4096 slab
4265          * However with the new *_jumbo_rx* routines, jumbo receives will use
4266          * fragmented skbs
4267          */
4268
4269         if (max_frame <= 256)
4270                 adapter->rx_buffer_len = 256;
4271         else if (max_frame <= 512)
4272                 adapter->rx_buffer_len = 512;
4273         else if (max_frame <= 1024)
4274                 adapter->rx_buffer_len = 1024;
4275         else if (max_frame <= 2048)
4276                 adapter->rx_buffer_len = 2048;
4277         else
4278                 adapter->rx_buffer_len = 4096;
4279
4280         /* adjust allocation if LPE protects us, and we aren't using SBP */
4281         if ((max_frame == ETH_FRAME_LEN + ETH_FCS_LEN) ||
4282              (max_frame == ETH_FRAME_LEN + VLAN_HLEN + ETH_FCS_LEN))
4283                 adapter->rx_buffer_len = ETH_FRAME_LEN + VLAN_HLEN
4284                                          + ETH_FCS_LEN;
4285
4286         e_info("changing MTU from %d to %d\n", netdev->mtu, new_mtu);
4287         netdev->mtu = new_mtu;
4288
4289         if (netif_running(netdev))
4290                 e1000e_up(adapter);
4291         else
4292                 e1000e_reset(adapter);
4293
4294         clear_bit(__E1000_RESETTING, &adapter->state);
4295
4296         return 0;
4297 }
4298
4299 static int e1000_mii_ioctl(struct net_device *netdev, struct ifreq *ifr,
4300                            int cmd)
4301 {
4302         struct e1000_adapter *adapter = netdev_priv(netdev);
4303         struct mii_ioctl_data *data = if_mii(ifr);
4304
4305         if (adapter->hw.phy.media_type != e1000_media_type_copper)
4306                 return -EOPNOTSUPP;
4307
4308         switch (cmd) {
4309         case SIOCGMIIPHY:
4310                 data->phy_id = adapter->hw.phy.addr;
4311                 break;
4312         case SIOCGMIIREG:
4313                 e1000_phy_read_status(adapter);
4314
4315                 switch (data->reg_num & 0x1F) {
4316                 case MII_BMCR:
4317                         data->val_out = adapter->phy_regs.bmcr;
4318                         break;
4319                 case MII_BMSR:
4320                         data->val_out = adapter->phy_regs.bmsr;
4321                         break;
4322                 case MII_PHYSID1:
4323                         data->val_out = (adapter->hw.phy.id >> 16);
4324                         break;
4325                 case MII_PHYSID2:
4326                         data->val_out = (adapter->hw.phy.id & 0xFFFF);
4327                         break;
4328                 case MII_ADVERTISE:
4329                         data->val_out = adapter->phy_regs.advertise;
4330                         break;
4331                 case MII_LPA:
4332                         data->val_out = adapter->phy_regs.lpa;
4333                         break;
4334                 case MII_EXPANSION:
4335                         data->val_out = adapter->phy_regs.expansion;
4336                         break;
4337                 case MII_CTRL1000:
4338                         data->val_out = adapter->phy_regs.ctrl1000;
4339                         break;
4340                 case MII_STAT1000:
4341                         data->val_out = adapter->phy_regs.stat1000;
4342                         break;
4343                 case MII_ESTATUS:
4344                         data->val_out = adapter->phy_regs.estatus;
4345                         break;
4346                 default:
4347                         return -EIO;
4348                 }
4349                 break;
4350         case SIOCSMIIREG:
4351         default:
4352                 return -EOPNOTSUPP;
4353         }
4354         return 0;
4355 }
4356
4357 static int e1000_ioctl(struct net_device *netdev, struct ifreq *ifr, int cmd)
4358 {
4359         switch (cmd) {
4360         case SIOCGMIIPHY:
4361         case SIOCGMIIREG:
4362         case SIOCSMIIREG:
4363                 return e1000_mii_ioctl(netdev, ifr, cmd);
4364         default:
4365                 return -EOPNOTSUPP;
4366         }
4367 }
4368
4369 static int e1000_init_phy_wakeup(struct e1000_adapter *adapter, u32 wufc)
4370 {
4371         struct e1000_hw *hw = &adapter->hw;
4372         u32 i, mac_reg;
4373         u16 phy_reg;
4374         int retval = 0;
4375
4376         /* copy MAC RARs to PHY RARs */
4377         for (i = 0; i < adapter->hw.mac.rar_entry_count; i++) {
4378                 mac_reg = er32(RAL(i));
4379                 e1e_wphy(hw, BM_RAR_L(i), (u16)(mac_reg & 0xFFFF));
4380                 e1e_wphy(hw, BM_RAR_M(i), (u16)((mac_reg >> 16) & 0xFFFF));
4381                 mac_reg = er32(RAH(i));
4382                 e1e_wphy(hw, BM_RAR_H(i), (u16)(mac_reg & 0xFFFF));
4383                 e1e_wphy(hw, BM_RAR_CTRL(i), (u16)((mac_reg >> 16) & 0xFFFF));
4384         }
4385
4386         /* copy MAC MTA to PHY MTA */
4387         for (i = 0; i < adapter->hw.mac.mta_reg_count; i++) {
4388                 mac_reg = E1000_READ_REG_ARRAY(hw, E1000_MTA, i);
4389                 e1e_wphy(hw, BM_MTA(i), (u16)(mac_reg & 0xFFFF));
4390                 e1e_wphy(hw, BM_MTA(i) + 1, (u16)((mac_reg >> 16) & 0xFFFF));
4391         }
4392
4393         /* configure PHY Rx Control register */
4394         e1e_rphy(&adapter->hw, BM_RCTL, &phy_reg);
4395         mac_reg = er32(RCTL);
4396         if (mac_reg & E1000_RCTL_UPE)
4397                 phy_reg |= BM_RCTL_UPE;
4398         if (mac_reg & E1000_RCTL_MPE)
4399                 phy_reg |= BM_RCTL_MPE;
4400         phy_reg &= ~(BM_RCTL_MO_MASK);
4401         if (mac_reg & E1000_RCTL_MO_3)
4402                 phy_reg |= (((mac_reg & E1000_RCTL_MO_3) >> E1000_RCTL_MO_SHIFT)
4403                                 << BM_RCTL_MO_SHIFT);
4404         if (mac_reg & E1000_RCTL_BAM)
4405                 phy_reg |= BM_RCTL_BAM;
4406         if (mac_reg & E1000_RCTL_PMCF)
4407                 phy_reg |= BM_RCTL_PMCF;
4408         mac_reg = er32(CTRL);
4409         if (mac_reg & E1000_CTRL_RFCE)
4410                 phy_reg |= BM_RCTL_RFCE;
4411         e1e_wphy(&adapter->hw, BM_RCTL, phy_reg);
4412
4413         /* enable PHY wakeup in MAC register */
4414         ew32(WUFC, wufc);
4415         ew32(WUC, E1000_WUC_PHY_WAKE | E1000_WUC_PME_EN);
4416
4417         /* configure and enable PHY wakeup in PHY registers */
4418         e1e_wphy(&adapter->hw, BM_WUFC, wufc);
4419         e1e_wphy(&adapter->hw, BM_WUC, E1000_WUC_PME_EN);
4420
4421         /* activate PHY wakeup */
4422         retval = hw->phy.ops.acquire(hw);
4423         if (retval) {
4424                 e_err("Could not acquire PHY\n");
4425                 return retval;
4426         }
4427         e1000e_write_phy_reg_mdic(hw, IGP01E1000_PHY_PAGE_SELECT,
4428                                  (BM_WUC_ENABLE_PAGE << IGP_PAGE_SHIFT));
4429         retval = e1000e_read_phy_reg_mdic(hw, BM_WUC_ENABLE_REG, &phy_reg);
4430         if (retval) {
4431                 e_err("Could not read PHY page 769\n");
4432                 goto out;
4433         }
4434         phy_reg |= BM_WUC_ENABLE_BIT | BM_WUC_HOST_WU_BIT;
4435         retval = e1000e_write_phy_reg_mdic(hw, BM_WUC_ENABLE_REG, phy_reg);
4436         if (retval)
4437                 e_err("Could not set PHY Host Wakeup bit\n");
4438 out:
4439         hw->phy.ops.release(hw);
4440
4441         return retval;
4442 }
4443
4444 static int __e1000_shutdown(struct pci_dev *pdev, bool *enable_wake)
4445 {
4446         struct net_device *netdev = pci_get_drvdata(pdev);
4447         struct e1000_adapter *adapter = netdev_priv(netdev);
4448         struct e1000_hw *hw = &adapter->hw;
4449         u32 ctrl, ctrl_ext, rctl, status;
4450         u32 wufc = adapter->wol;
4451         int retval = 0;
4452
4453         netif_device_detach(netdev);
4454
4455         if (netif_running(netdev)) {
4456                 WARN_ON(test_bit(__E1000_RESETTING, &adapter->state));
4457                 e1000e_down(adapter);
4458                 e1000_free_irq(adapter);
4459         }
4460         e1000e_reset_interrupt_capability(adapter);
4461
4462         retval = pci_save_state(pdev);
4463         if (retval)
4464                 return retval;
4465
4466         status = er32(STATUS);
4467         if (status & E1000_STATUS_LU)
4468                 wufc &= ~E1000_WUFC_LNKC;
4469
4470         if (wufc) {
4471                 e1000_setup_rctl(adapter);
4472                 e1000_set_multi(netdev);
4473
4474                 /* turn on all-multi mode if wake on multicast is enabled */
4475                 if (wufc & E1000_WUFC_MC) {
4476                         rctl = er32(RCTL);
4477                         rctl |= E1000_RCTL_MPE;
4478                         ew32(RCTL, rctl);
4479                 }
4480
4481                 ctrl = er32(CTRL);
4482                 /* advertise wake from D3Cold */
4483                 #define E1000_CTRL_ADVD3WUC 0x00100000
4484                 /* phy power management enable */
4485                 #define E1000_CTRL_EN_PHY_PWR_MGMT 0x00200000
4486                 ctrl |= E1000_CTRL_ADVD3WUC;
4487                 if (!(adapter->flags2 & FLAG2_HAS_PHY_WAKEUP))
4488                         ctrl |= E1000_CTRL_EN_PHY_PWR_MGMT;
4489                 ew32(CTRL, ctrl);
4490
4491                 if (adapter->hw.phy.media_type == e1000_media_type_fiber ||
4492                     adapter->hw.phy.media_type ==
4493                     e1000_media_type_internal_serdes) {
4494                         /* keep the laser running in D3 */
4495                         ctrl_ext = er32(CTRL_EXT);
4496                         ctrl_ext |= E1000_CTRL_EXT_SDP7_DATA;
4497                         ew32(CTRL_EXT, ctrl_ext);
4498                 }
4499
4500                 if (adapter->flags & FLAG_IS_ICH)
4501                         e1000e_disable_gig_wol_ich8lan(&adapter->hw);
4502
4503                 /* Allow time for pending master requests to run */
4504                 e1000e_disable_pcie_master(&adapter->hw);
4505
4506                 if (adapter->flags2 & FLAG2_HAS_PHY_WAKEUP) {
4507                         /* enable wakeup by the PHY */
4508                         retval = e1000_init_phy_wakeup(adapter, wufc);
4509                         if (retval)
4510                                 return retval;
4511                 } else {
4512                         /* enable wakeup by the MAC */
4513                         ew32(WUFC, wufc);
4514                         ew32(WUC, E1000_WUC_PME_EN);
4515                 }
4516         } else {
4517                 ew32(WUC, 0);
4518                 ew32(WUFC, 0);
4519         }
4520
4521         *enable_wake = !!wufc;
4522
4523         /* make sure adapter isn't asleep if manageability is enabled */
4524         if ((adapter->flags & FLAG_MNG_PT_ENABLED) ||
4525             (hw->mac.ops.check_mng_mode(hw)))
4526                 *enable_wake = true;
4527
4528         if (adapter->hw.phy.type == e1000_phy_igp_3)
4529                 e1000e_igp3_phy_powerdown_workaround_ich8lan(&adapter->hw);
4530
4531         /*
4532          * Release control of h/w to f/w.  If f/w is AMT enabled, this
4533          * would have already happened in close and is redundant.
4534          */
4535         e1000_release_hw_control(adapter);
4536
4537         pci_disable_device(pdev);
4538
4539         return 0;
4540 }
4541
4542 static void e1000_power_off(struct pci_dev *pdev, bool sleep, bool wake)
4543 {
4544         if (sleep && wake) {
4545                 pci_prepare_to_sleep(pdev);
4546                 return;
4547         }
4548
4549         pci_wake_from_d3(pdev, wake);
4550         pci_set_power_state(pdev, PCI_D3hot);
4551 }
4552
4553 static void e1000_complete_shutdown(struct pci_dev *pdev, bool sleep,
4554                                     bool wake)
4555 {
4556         struct net_device *netdev = pci_get_drvdata(pdev);
4557         struct e1000_adapter *adapter = netdev_priv(netdev);
4558
4559         /*
4560          * The pci-e switch on some quad port adapters will report a
4561          * correctable error when the MAC transitions from D0 to D3.  To
4562          * prevent this we need to mask off the correctable errors on the
4563          * downstream port of the pci-e switch.
4564          */
4565         if (adapter->flags & FLAG_IS_QUAD_PORT) {
4566                 struct pci_dev *us_dev = pdev->bus->self;
4567                 int pos = pci_find_capability(us_dev, PCI_CAP_ID_EXP);
4568                 u16 devctl;
4569
4570                 pci_read_config_word(us_dev, pos + PCI_EXP_DEVCTL, &devctl);
4571                 pci_write_config_word(us_dev, pos + PCI_EXP_DEVCTL,
4572                                       (devctl & ~PCI_EXP_DEVCTL_CERE));
4573
4574                 e1000_power_off(pdev, sleep, wake);
4575
4576                 pci_write_config_word(us_dev, pos + PCI_EXP_DEVCTL, devctl);
4577         } else {
4578                 e1000_power_off(pdev, sleep, wake);
4579         }
4580 }
4581
4582 static void e1000e_disable_l1aspm(struct pci_dev *pdev)
4583 {
4584         int pos;
4585         u16 val;
4586
4587         /*
4588          * 82573 workaround - disable L1 ASPM on mobile chipsets
4589          *
4590          * L1 ASPM on various mobile (ich7) chipsets do not behave properly
4591          * resulting in lost data or garbage information on the pci-e link
4592          * level. This could result in (false) bad EEPROM checksum errors,
4593          * long ping times (up to 2s) or even a system freeze/hang.
4594          *
4595          * Unfortunately this feature saves about 1W power consumption when
4596          * active.
4597          */
4598         pos = pci_find_capability(pdev, PCI_CAP_ID_EXP);
4599         pci_read_config_word(pdev, pos + PCI_EXP_LNKCTL, &val);
4600         if (val & 0x2) {
4601                 dev_warn(&pdev->dev, "Disabling L1 ASPM\n");
4602                 val &= ~0x2;
4603                 pci_write_config_word(pdev, pos + PCI_EXP_LNKCTL, val);
4604         }
4605 }
4606
4607 #ifdef CONFIG_PM
4608 static int e1000_suspend(struct pci_dev *pdev, pm_message_t state)
4609 {
4610         int retval;
4611         bool wake;
4612
4613         retval = __e1000_shutdown(pdev, &wake);
4614         if (!retval)
4615                 e1000_complete_shutdown(pdev, true, wake);
4616
4617         return retval;
4618 }
4619
4620 static int e1000_resume(struct pci_dev *pdev)
4621 {
4622         struct net_device *netdev = pci_get_drvdata(pdev);
4623         struct e1000_adapter *adapter = netdev_priv(netdev);
4624         struct e1000_hw *hw = &adapter->hw;
4625         u32 err;
4626
4627         pci_set_power_state(pdev, PCI_D0);
4628         pci_restore_state(pdev);
4629         e1000e_disable_l1aspm(pdev);
4630
4631         err = pci_enable_device_mem(pdev);
4632         if (err) {
4633                 dev_err(&pdev->dev,
4634                         "Cannot enable PCI device from suspend\n");
4635                 return err;
4636         }
4637
4638         pci_set_master(pdev);
4639
4640         pci_enable_wake(pdev, PCI_D3hot, 0);
4641         pci_enable_wake(pdev, PCI_D3cold, 0);
4642
4643         e1000e_set_interrupt_capability(adapter);
4644         if (netif_running(netdev)) {
4645                 err = e1000_request_irq(adapter);
4646                 if (err)
4647                         return err;
4648         }
4649
4650         e1000e_power_up_phy(adapter);
4651
4652         /* report the system wakeup cause from S3/S4 */
4653         if (adapter->flags2 & FLAG2_HAS_PHY_WAKEUP) {
4654                 u16 phy_data;
4655
4656                 e1e_rphy(&adapter->hw, BM_WUS, &phy_data);
4657                 if (phy_data) {
4658                         e_info("PHY Wakeup cause - %s\n",
4659                                 phy_data & E1000_WUS_EX ? "Unicast Packet" :
4660                                 phy_data & E1000_WUS_MC ? "Multicast Packet" :
4661                                 phy_data & E1000_WUS_BC ? "Broadcast Packet" :
4662                                 phy_data & E1000_WUS_MAG ? "Magic Packet" :
4663                                 phy_data & E1000_WUS_LNKC ? "Link Status "
4664                                 " Change" : "other");
4665                 }
4666                 e1e_wphy(&adapter->hw, BM_WUS, ~0);
4667         } else {
4668                 u32 wus = er32(WUS);
4669                 if (wus) {
4670                         e_info("MAC Wakeup cause - %s\n",
4671                                 wus & E1000_WUS_EX ? "Unicast Packet" :
4672                                 wus & E1000_WUS_MC ? "Multicast Packet" :
4673                                 wus & E1000_WUS_BC ? "Broadcast Packet" :
4674                                 wus & E1000_WUS_MAG ? "Magic Packet" :
4675                                 wus & E1000_WUS_LNKC ? "Link Status Change" :
4676                                 "other");
4677                 }
4678                 ew32(WUS, ~0);
4679         }
4680
4681         e1000e_reset(adapter);
4682
4683         e1000_init_manageability(adapter);
4684
4685         if (netif_running(netdev))
4686                 e1000e_up(adapter);
4687
4688         netif_device_attach(netdev);
4689
4690         /*
4691          * If the controller has AMT, do not set DRV_LOAD until the interface
4692          * is up.  For all other cases, let the f/w know that the h/w is now
4693          * under the control of the driver.
4694          */
4695         if (!(adapter->flags & FLAG_HAS_AMT))
4696                 e1000_get_hw_control(adapter);
4697
4698         return 0;
4699 }
4700 #endif
4701
4702 static void e1000_shutdown(struct pci_dev *pdev)
4703 {
4704         bool wake = false;
4705
4706         __e1000_shutdown(pdev, &wake);
4707
4708         if (system_state == SYSTEM_POWER_OFF)
4709                 e1000_complete_shutdown(pdev, false, wake);
4710 }
4711
4712 #ifdef CONFIG_NET_POLL_CONTROLLER
4713 /*
4714  * Polling 'interrupt' - used by things like netconsole to send skbs
4715  * without having to re-enable interrupts. It's not called while
4716  * the interrupt routine is executing.
4717  */
4718 static void e1000_netpoll(struct net_device *netdev)
4719 {
4720         struct e1000_adapter *adapter = netdev_priv(netdev);
4721
4722         disable_irq(adapter->pdev->irq);
4723         e1000_intr(adapter->pdev->irq, netdev);
4724
4725         enable_irq(adapter->pdev->irq);
4726 }
4727 #endif
4728
4729 /**
4730  * e1000_io_error_detected - called when PCI error is detected
4731  * @pdev: Pointer to PCI device
4732  * @state: The current pci connection state
4733  *
4734  * This function is called after a PCI bus error affecting
4735  * this device has been detected.
4736  */
4737 static pci_ers_result_t e1000_io_error_detected(struct pci_dev *pdev,
4738                                                 pci_channel_state_t state)
4739 {
4740         struct net_device *netdev = pci_get_drvdata(pdev);
4741         struct e1000_adapter *adapter = netdev_priv(netdev);
4742
4743         netif_device_detach(netdev);
4744
4745         if (state == pci_channel_io_perm_failure)
4746                 return PCI_ERS_RESULT_DISCONNECT;
4747
4748         if (netif_running(netdev))
4749                 e1000e_down(adapter);
4750         pci_disable_device(pdev);
4751
4752         /* Request a slot slot reset. */
4753         return PCI_ERS_RESULT_NEED_RESET;
4754 }
4755
4756 /**
4757  * e1000_io_slot_reset - called after the pci bus has been reset.
4758  * @pdev: Pointer to PCI device
4759  *
4760  * Restart the card from scratch, as if from a cold-boot. Implementation
4761  * resembles the first-half of the e1000_resume routine.
4762  */
4763 static pci_ers_result_t e1000_io_slot_reset(struct pci_dev *pdev)
4764 {
4765         struct net_device *netdev = pci_get_drvdata(pdev);
4766         struct e1000_adapter *adapter = netdev_priv(netdev);
4767         struct e1000_hw *hw = &adapter->hw;
4768         int err;
4769         pci_ers_result_t result;
4770
4771         e1000e_disable_l1aspm(pdev);
4772         err = pci_enable_device_mem(pdev);
4773         if (err) {
4774                 dev_err(&pdev->dev,
4775                         "Cannot re-enable PCI device after reset.\n");
4776                 result = PCI_ERS_RESULT_DISCONNECT;
4777         } else {
4778                 pci_set_master(pdev);
4779                 pci_restore_state(pdev);
4780
4781                 pci_enable_wake(pdev, PCI_D3hot, 0);
4782                 pci_enable_wake(pdev, PCI_D3cold, 0);
4783
4784                 e1000e_reset(adapter);
4785                 ew32(WUS, ~0);
4786                 result = PCI_ERS_RESULT_RECOVERED;
4787         }
4788
4789         pci_cleanup_aer_uncorrect_error_status(pdev);
4790
4791         return result;
4792 }
4793
4794 /**
4795  * e1000_io_resume - called when traffic can start flowing again.
4796  * @pdev: Pointer to PCI device
4797  *
4798  * This callback is called when the error recovery driver tells us that
4799  * its OK to resume normal operation. Implementation resembles the
4800  * second-half of the e1000_resume routine.
4801  */
4802 static void e1000_io_resume(struct pci_dev *pdev)
4803 {
4804         struct net_device *netdev = pci_get_drvdata(pdev);
4805         struct e1000_adapter *adapter = netdev_priv(netdev);
4806
4807         e1000_init_manageability(adapter);
4808
4809         if (netif_running(netdev)) {
4810                 if (e1000e_up(adapter)) {
4811                         dev_err(&pdev->dev,
4812                                 "can't bring device back up after reset\n");
4813                         return;
4814                 }
4815         }
4816
4817         netif_device_attach(netdev);
4818
4819         /*
4820          * If the controller has AMT, do not set DRV_LOAD until the interface
4821          * is up.  For all other cases, let the f/w know that the h/w is now
4822          * under the control of the driver.
4823          */
4824         if (!(adapter->flags & FLAG_HAS_AMT))
4825                 e1000_get_hw_control(adapter);
4826
4827 }
4828
4829 static void e1000_print_device_info(struct e1000_adapter *adapter)
4830 {
4831         struct e1000_hw *hw = &adapter->hw;
4832         struct net_device *netdev = adapter->netdev;
4833         u32 pba_num;
4834
4835         /* print bus type/speed/width info */
4836         e_info("(PCI Express:2.5GB/s:%s) %pM\n",
4837                /* bus width */
4838                ((hw->bus.width == e1000_bus_width_pcie_x4) ? "Width x4" :
4839                 "Width x1"),
4840                /* MAC address */
4841                netdev->dev_addr);
4842         e_info("Intel(R) PRO/%s Network Connection\n",
4843                (hw->phy.type == e1000_phy_ife) ? "10/100" : "1000");
4844         e1000e_read_pba_num(hw, &pba_num);
4845         e_info("MAC: %d, PHY: %d, PBA No: %06x-%03x\n",
4846                hw->mac.type, hw->phy.type, (pba_num >> 8), (pba_num & 0xff));
4847 }
4848
4849 static void e1000_eeprom_checks(struct e1000_adapter *adapter)
4850 {
4851         struct e1000_hw *hw = &adapter->hw;
4852         int ret_val;
4853         u16 buf = 0;
4854
4855         if (hw->mac.type != e1000_82573)
4856                 return;
4857
4858         ret_val = e1000_read_nvm(hw, NVM_INIT_CONTROL2_REG, 1, &buf);
4859         if (!ret_val && (!(le16_to_cpu(buf) & (1 << 0)))) {
4860                 /* Deep Smart Power Down (DSPD) */
4861                 dev_warn(&adapter->pdev->dev,
4862                          "Warning: detected DSPD enabled in EEPROM\n");
4863         }
4864
4865         ret_val = e1000_read_nvm(hw, NVM_INIT_3GIO_3, 1, &buf);
4866         if (!ret_val && (le16_to_cpu(buf) & (3 << 2))) {
4867                 /* ASPM enable */
4868                 dev_warn(&adapter->pdev->dev,
4869                          "Warning: detected ASPM enabled in EEPROM\n");
4870         }
4871 }
4872
4873 static const struct net_device_ops e1000e_netdev_ops = {
4874         .ndo_open               = e1000_open,
4875         .ndo_stop               = e1000_close,
4876         .ndo_start_xmit         = e1000_xmit_frame,
4877         .ndo_get_stats          = e1000_get_stats,
4878         .ndo_set_multicast_list = e1000_set_multi,
4879         .ndo_set_mac_address    = e1000_set_mac,
4880         .ndo_change_mtu         = e1000_change_mtu,
4881         .ndo_do_ioctl           = e1000_ioctl,
4882         .ndo_tx_timeout         = e1000_tx_timeout,
4883         .ndo_validate_addr      = eth_validate_addr,
4884
4885         .ndo_vlan_rx_register   = e1000_vlan_rx_register,
4886         .ndo_vlan_rx_add_vid    = e1000_vlan_rx_add_vid,
4887         .ndo_vlan_rx_kill_vid   = e1000_vlan_rx_kill_vid,
4888 #ifdef CONFIG_NET_POLL_CONTROLLER
4889         .ndo_poll_controller    = e1000_netpoll,
4890 #endif
4891 };
4892
4893 /**
4894  * e1000_probe - Device Initialization Routine
4895  * @pdev: PCI device information struct
4896  * @ent: entry in e1000_pci_tbl
4897  *
4898  * Returns 0 on success, negative on failure
4899  *
4900  * e1000_probe initializes an adapter identified by a pci_dev structure.
4901  * The OS initialization, configuring of the adapter private structure,
4902  * and a hardware reset occur.
4903  **/
4904 static int __devinit e1000_probe(struct pci_dev *pdev,
4905                                  const struct pci_device_id *ent)
4906 {
4907         struct net_device *netdev;
4908         struct e1000_adapter *adapter;
4909         struct e1000_hw *hw;
4910         const struct e1000_info *ei = e1000_info_tbl[ent->driver_data];
4911         resource_size_t mmio_start, mmio_len;
4912         resource_size_t flash_start, flash_len;
4913
4914         static int cards_found;
4915         int i, err, pci_using_dac;
4916         u16 eeprom_data = 0;
4917         u16 eeprom_apme_mask = E1000_EEPROM_APME;
4918
4919         e1000e_disable_l1aspm(pdev);
4920
4921         err = pci_enable_device_mem(pdev);
4922         if (err)
4923                 return err;
4924
4925         pci_using_dac = 0;
4926         err = pci_set_dma_mask(pdev, DMA_BIT_MASK(64));
4927         if (!err) {
4928                 err = pci_set_consistent_dma_mask(pdev, DMA_BIT_MASK(64));
4929                 if (!err)
4930                         pci_using_dac = 1;
4931         } else {
4932                 err = pci_set_dma_mask(pdev, DMA_BIT_MASK(32));
4933                 if (err) {
4934                         err = pci_set_consistent_dma_mask(pdev,
4935                                                           DMA_BIT_MASK(32));
4936                         if (err) {
4937                                 dev_err(&pdev->dev, "No usable DMA "
4938                                         "configuration, aborting\n");
4939                                 goto err_dma;
4940                         }
4941                 }
4942         }
4943
4944         err = pci_request_selected_regions_exclusive(pdev,
4945                                           pci_select_bars(pdev, IORESOURCE_MEM),
4946                                           e1000e_driver_name);
4947         if (err)
4948                 goto err_pci_reg;
4949
4950         /* AER (Advanced Error Reporting) hooks */
4951         pci_enable_pcie_error_reporting(pdev);
4952
4953         pci_set_master(pdev);
4954         /* PCI config space info */
4955         err = pci_save_state(pdev);
4956         if (err)
4957                 goto err_alloc_etherdev;
4958
4959         err = -ENOMEM;
4960         netdev = alloc_etherdev(sizeof(struct e1000_adapter));
4961         if (!netdev)
4962                 goto err_alloc_etherdev;
4963
4964         SET_NETDEV_DEV(netdev, &pdev->dev);
4965
4966         pci_set_drvdata(pdev, netdev);
4967         adapter = netdev_priv(netdev);
4968         hw = &adapter->hw;
4969         adapter->netdev = netdev;
4970         adapter->pdev = pdev;
4971         adapter->ei = ei;
4972         adapter->pba = ei->pba;
4973         adapter->flags = ei->flags;
4974         adapter->flags2 = ei->flags2;
4975         adapter->hw.adapter = adapter;
4976         adapter->hw.mac.type = ei->mac;
4977         adapter->max_hw_frame_size = ei->max_hw_frame_size;
4978         adapter->msg_enable = (1 << NETIF_MSG_DRV | NETIF_MSG_PROBE) - 1;
4979
4980         mmio_start = pci_resource_start(pdev, 0);
4981         mmio_len = pci_resource_len(pdev, 0);
4982
4983         err = -EIO;
4984         adapter->hw.hw_addr = ioremap(mmio_start, mmio_len);
4985         if (!adapter->hw.hw_addr)
4986                 goto err_ioremap;
4987
4988         if ((adapter->flags & FLAG_HAS_FLASH) &&
4989             (pci_resource_flags(pdev, 1) & IORESOURCE_MEM)) {
4990                 flash_start = pci_resource_start(pdev, 1);
4991                 flash_len = pci_resource_len(pdev, 1);
4992                 adapter->hw.flash_address = ioremap(flash_start, flash_len);
4993                 if (!adapter->hw.flash_address)
4994                         goto err_flashmap;
4995         }
4996
4997         /* construct the net_device struct */
4998         netdev->netdev_ops              = &e1000e_netdev_ops;
4999         e1000e_set_ethtool_ops(netdev);
5000         netdev->watchdog_timeo          = 5 * HZ;
5001         netif_napi_add(netdev, &adapter->napi, e1000_clean, 64);
5002         strncpy(netdev->name, pci_name(pdev), sizeof(netdev->name) - 1);
5003
5004         netdev->mem_start = mmio_start;
5005         netdev->mem_end = mmio_start + mmio_len;
5006
5007         adapter->bd_number = cards_found++;
5008
5009         e1000e_check_options(adapter);
5010
5011         /* setup adapter struct */
5012         err = e1000_sw_init(adapter);
5013         if (err)
5014                 goto err_sw_init;
5015
5016         err = -EIO;
5017
5018         memcpy(&hw->mac.ops, ei->mac_ops, sizeof(hw->mac.ops));
5019         memcpy(&hw->nvm.ops, ei->nvm_ops, sizeof(hw->nvm.ops));
5020         memcpy(&hw->phy.ops, ei->phy_ops, sizeof(hw->phy.ops));
5021
5022         err = ei->get_variants(adapter);
5023         if (err)
5024                 goto err_hw_init;
5025
5026         if ((adapter->flags & FLAG_IS_ICH) &&
5027             (adapter->flags & FLAG_READ_ONLY_NVM))
5028                 e1000e_write_protect_nvm_ich8lan(&adapter->hw);
5029
5030         hw->mac.ops.get_bus_info(&adapter->hw);
5031
5032         adapter->hw.phy.autoneg_wait_to_complete = 0;
5033
5034         /* Copper options */
5035         if (adapter->hw.phy.media_type == e1000_media_type_copper) {
5036                 adapter->hw.phy.mdix = AUTO_ALL_MODES;
5037                 adapter->hw.phy.disable_polarity_correction = 0;
5038                 adapter->hw.phy.ms_type = e1000_ms_hw_default;
5039         }
5040
5041         if (e1000_check_reset_block(&adapter->hw))
5042                 e_info("PHY reset is blocked due to SOL/IDER session.\n");
5043
5044         netdev->features = NETIF_F_SG |
5045                            NETIF_F_HW_CSUM |
5046                            NETIF_F_HW_VLAN_TX |
5047                            NETIF_F_HW_VLAN_RX;
5048
5049         if (adapter->flags & FLAG_HAS_HW_VLAN_FILTER)
5050                 netdev->features |= NETIF_F_HW_VLAN_FILTER;
5051
5052         netdev->features |= NETIF_F_TSO;
5053         netdev->features |= NETIF_F_TSO6;
5054
5055         netdev->vlan_features |= NETIF_F_TSO;
5056         netdev->vlan_features |= NETIF_F_TSO6;
5057         netdev->vlan_features |= NETIF_F_HW_CSUM;
5058         netdev->vlan_features |= NETIF_F_SG;
5059
5060         if (pci_using_dac)
5061                 netdev->features |= NETIF_F_HIGHDMA;
5062
5063         if (e1000e_enable_mng_pass_thru(&adapter->hw))
5064                 adapter->flags |= FLAG_MNG_PT_ENABLED;
5065
5066         /*
5067          * before reading the NVM, reset the controller to
5068          * put the device in a known good starting state
5069          */
5070         adapter->hw.mac.ops.reset_hw(&adapter->hw);
5071
5072         /*
5073          * systems with ASPM and others may see the checksum fail on the first
5074          * attempt. Let's give it a few tries
5075          */
5076         for (i = 0;; i++) {
5077                 if (e1000_validate_nvm_checksum(&adapter->hw) >= 0)
5078                         break;
5079                 if (i == 2) {
5080                         e_err("The NVM Checksum Is Not Valid\n");
5081                         err = -EIO;
5082                         goto err_eeprom;
5083                 }
5084         }
5085
5086         e1000_eeprom_checks(adapter);
5087
5088         /* copy the MAC address out of the NVM */
5089         if (e1000e_read_mac_addr(&adapter->hw))
5090                 e_err("NVM Read Error while reading MAC address\n");
5091
5092         memcpy(netdev->dev_addr, adapter->hw.mac.addr, netdev->addr_len);
5093         memcpy(netdev->perm_addr, adapter->hw.mac.addr, netdev->addr_len);
5094
5095         if (!is_valid_ether_addr(netdev->perm_addr)) {
5096                 e_err("Invalid MAC Address: %pM\n", netdev->perm_addr);
5097                 err = -EIO;
5098                 goto err_eeprom;
5099         }
5100
5101         init_timer(&adapter->watchdog_timer);
5102         adapter->watchdog_timer.function = &e1000_watchdog;
5103         adapter->watchdog_timer.data = (unsigned long) adapter;
5104
5105         init_timer(&adapter->phy_info_timer);
5106         adapter->phy_info_timer.function = &e1000_update_phy_info;
5107         adapter->phy_info_timer.data = (unsigned long) adapter;
5108
5109         INIT_WORK(&adapter->reset_task, e1000_reset_task);
5110         INIT_WORK(&adapter->watchdog_task, e1000_watchdog_task);
5111         INIT_WORK(&adapter->downshift_task, e1000e_downshift_workaround);
5112         INIT_WORK(&adapter->update_phy_task, e1000e_update_phy_task);
5113
5114         /* Initialize link parameters. User can change them with ethtool */
5115         adapter->hw.mac.autoneg = 1;
5116         adapter->fc_autoneg = 1;
5117         adapter->hw.fc.requested_mode = e1000_fc_default;
5118         adapter->hw.fc.current_mode = e1000_fc_default;
5119         adapter->hw.phy.autoneg_advertised = 0x2f;
5120
5121         /* ring size defaults */
5122         adapter->rx_ring->count = 256;
5123         adapter->tx_ring->count = 256;
5124
5125         /*
5126          * Initial Wake on LAN setting - If APM wake is enabled in
5127          * the EEPROM, enable the ACPI Magic Packet filter
5128          */
5129         if (adapter->flags & FLAG_APME_IN_WUC) {
5130                 /* APME bit in EEPROM is mapped to WUC.APME */
5131                 eeprom_data = er32(WUC);
5132                 eeprom_apme_mask = E1000_WUC_APME;
5133                 if (eeprom_data & E1000_WUC_PHY_WAKE)
5134                         adapter->flags2 |= FLAG2_HAS_PHY_WAKEUP;
5135         } else if (adapter->flags & FLAG_APME_IN_CTRL3) {
5136                 if (adapter->flags & FLAG_APME_CHECK_PORT_B &&
5137                     (adapter->hw.bus.func == 1))
5138                         e1000_read_nvm(&adapter->hw,
5139                                 NVM_INIT_CONTROL3_PORT_B, 1, &eeprom_data);
5140                 else
5141                         e1000_read_nvm(&adapter->hw,
5142                                 NVM_INIT_CONTROL3_PORT_A, 1, &eeprom_data);
5143         }
5144
5145         /* fetch WoL from EEPROM */
5146         if (eeprom_data & eeprom_apme_mask)
5147                 adapter->eeprom_wol |= E1000_WUFC_MAG;
5148
5149         /*
5150          * now that we have the eeprom settings, apply the special cases
5151          * where the eeprom may be wrong or the board simply won't support
5152          * wake on lan on a particular port
5153          */
5154         if (!(adapter->flags & FLAG_HAS_WOL))
5155                 adapter->eeprom_wol = 0;
5156
5157         /* initialize the wol settings based on the eeprom settings */
5158         adapter->wol = adapter->eeprom_wol;
5159         device_set_wakeup_enable(&adapter->pdev->dev, adapter->wol);
5160
5161         /* save off EEPROM version number */
5162         e1000_read_nvm(&adapter->hw, 5, 1, &adapter->eeprom_vers);
5163
5164         /* reset the hardware with the new settings */
5165         e1000e_reset(adapter);
5166
5167         /*
5168          * If the controller has AMT, do not set DRV_LOAD until the interface
5169          * is up.  For all other cases, let the f/w know that the h/w is now
5170          * under the control of the driver.
5171          */
5172         if (!(adapter->flags & FLAG_HAS_AMT))
5173                 e1000_get_hw_control(adapter);
5174
5175         strcpy(netdev->name, "eth%d");
5176         err = register_netdev(netdev);
5177         if (err)
5178                 goto err_register;
5179
5180         /* carrier off reporting is important to ethtool even BEFORE open */
5181         netif_carrier_off(netdev);
5182
5183         e1000_print_device_info(adapter);
5184
5185         return 0;
5186
5187 err_register:
5188         if (!(adapter->flags & FLAG_HAS_AMT))
5189                 e1000_release_hw_control(adapter);
5190 err_eeprom:
5191         if (!e1000_check_reset_block(&adapter->hw))
5192                 e1000_phy_hw_reset(&adapter->hw);
5193 err_hw_init:
5194
5195         kfree(adapter->tx_ring);
5196         kfree(adapter->rx_ring);
5197 err_sw_init:
5198         if (adapter->hw.flash_address)
5199                 iounmap(adapter->hw.flash_address);
5200         e1000e_reset_interrupt_capability(adapter);
5201 err_flashmap:
5202         iounmap(adapter->hw.hw_addr);
5203 err_ioremap:
5204         free_netdev(netdev);
5205 err_alloc_etherdev:
5206         pci_release_selected_regions(pdev,
5207                                      pci_select_bars(pdev, IORESOURCE_MEM));
5208 err_pci_reg:
5209 err_dma:
5210         pci_disable_device(pdev);
5211         return err;
5212 }
5213
5214 /**
5215  * e1000_remove - Device Removal Routine
5216  * @pdev: PCI device information struct
5217  *
5218  * e1000_remove is called by the PCI subsystem to alert the driver
5219  * that it should release a PCI device.  The could be caused by a
5220  * Hot-Plug event, or because the driver is going to be removed from
5221  * memory.
5222  **/
5223 static void __devexit e1000_remove(struct pci_dev *pdev)
5224 {
5225         struct net_device *netdev = pci_get_drvdata(pdev);
5226         struct e1000_adapter *adapter = netdev_priv(netdev);
5227
5228         /*
5229          * flush_scheduled work may reschedule our watchdog task, so
5230          * explicitly disable watchdog tasks from being rescheduled
5231          */
5232         set_bit(__E1000_DOWN, &adapter->state);
5233         del_timer_sync(&adapter->watchdog_timer);
5234         del_timer_sync(&adapter->phy_info_timer);
5235
5236         flush_scheduled_work();
5237
5238         /*
5239          * Release control of h/w to f/w.  If f/w is AMT enabled, this
5240          * would have already happened in close and is redundant.
5241          */
5242         e1000_release_hw_control(adapter);
5243
5244         unregister_netdev(netdev);
5245
5246         if (!e1000_check_reset_block(&adapter->hw))
5247                 e1000_phy_hw_reset(&adapter->hw);
5248
5249         e1000e_reset_interrupt_capability(adapter);
5250         kfree(adapter->tx_ring);
5251         kfree(adapter->rx_ring);
5252
5253         iounmap(adapter->hw.hw_addr);
5254         if (adapter->hw.flash_address)
5255                 iounmap(adapter->hw.flash_address);
5256         pci_release_selected_regions(pdev,
5257                                      pci_select_bars(pdev, IORESOURCE_MEM));
5258
5259         free_netdev(netdev);
5260
5261         /* AER disable */
5262         pci_disable_pcie_error_reporting(pdev);
5263
5264         pci_disable_device(pdev);
5265 }
5266
5267 /* PCI Error Recovery (ERS) */
5268 static struct pci_error_handlers e1000_err_handler = {
5269         .error_detected = e1000_io_error_detected,
5270         .slot_reset = e1000_io_slot_reset,
5271         .resume = e1000_io_resume,
5272 };
5273
5274 static struct pci_device_id e1000_pci_tbl[] = {
5275         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82571EB_COPPER), board_82571 },
5276         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82571EB_FIBER), board_82571 },
5277         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82571EB_QUAD_COPPER), board_82571 },
5278         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82571EB_QUAD_COPPER_LP), board_82571 },
5279         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82571EB_QUAD_FIBER), board_82571 },
5280         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82571EB_SERDES), board_82571 },
5281         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82571EB_SERDES_DUAL), board_82571 },
5282         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82571EB_SERDES_QUAD), board_82571 },
5283         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82571PT_QUAD_COPPER), board_82571 },
5284
5285         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82572EI), board_82572 },
5286         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82572EI_COPPER), board_82572 },
5287         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82572EI_FIBER), board_82572 },
5288         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82572EI_SERDES), board_82572 },
5289
5290         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82573E), board_82573 },
5291         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82573E_IAMT), board_82573 },
5292         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82573L), board_82573 },
5293
5294         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82574L), board_82574 },
5295         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82574LA), board_82574 },
5296         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82583V), board_82583 },
5297
5298         { PCI_VDEVICE(INTEL, E1000_DEV_ID_80003ES2LAN_COPPER_DPT),
5299           board_80003es2lan },
5300         { PCI_VDEVICE(INTEL, E1000_DEV_ID_80003ES2LAN_COPPER_SPT),
5301           board_80003es2lan },
5302         { PCI_VDEVICE(INTEL, E1000_DEV_ID_80003ES2LAN_SERDES_DPT),
5303           board_80003es2lan },
5304         { PCI_VDEVICE(INTEL, E1000_DEV_ID_80003ES2LAN_SERDES_SPT),
5305           board_80003es2lan },
5306
5307         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH8_IFE), board_ich8lan },
5308         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH8_IFE_G), board_ich8lan },
5309         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH8_IFE_GT), board_ich8lan },
5310         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH8_IGP_AMT), board_ich8lan },
5311         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH8_IGP_C), board_ich8lan },
5312         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH8_IGP_M), board_ich8lan },
5313         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH8_IGP_M_AMT), board_ich8lan },
5314
5315         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH9_IFE), board_ich9lan },
5316         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH9_IFE_G), board_ich9lan },
5317         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH9_IFE_GT), board_ich9lan },
5318         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH9_IGP_AMT), board_ich9lan },
5319         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH9_IGP_C), board_ich9lan },
5320         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH9_BM), board_ich9lan },
5321         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH9_IGP_M), board_ich9lan },
5322         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH9_IGP_M_AMT), board_ich9lan },
5323         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH9_IGP_M_V), board_ich9lan },
5324
5325         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH10_R_BM_LM), board_ich9lan },
5326         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH10_R_BM_LF), board_ich9lan },
5327         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH10_R_BM_V), board_ich9lan },
5328
5329         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH10_D_BM_LM), board_ich10lan },
5330         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH10_D_BM_LF), board_ich10lan },
5331
5332         { PCI_VDEVICE(INTEL, E1000_DEV_ID_PCH_M_HV_LM), board_pchlan },
5333         { PCI_VDEVICE(INTEL, E1000_DEV_ID_PCH_M_HV_LC), board_pchlan },
5334         { PCI_VDEVICE(INTEL, E1000_DEV_ID_PCH_D_HV_DM), board_pchlan },
5335         { PCI_VDEVICE(INTEL, E1000_DEV_ID_PCH_D_HV_DC), board_pchlan },
5336
5337         { }     /* terminate list */
5338 };
5339 MODULE_DEVICE_TABLE(pci, e1000_pci_tbl);
5340
5341 /* PCI Device API Driver */
5342 static struct pci_driver e1000_driver = {
5343         .name     = e1000e_driver_name,
5344         .id_table = e1000_pci_tbl,
5345         .probe    = e1000_probe,
5346         .remove   = __devexit_p(e1000_remove),
5347 #ifdef CONFIG_PM
5348         /* Power Management Hooks */
5349         .suspend  = e1000_suspend,
5350         .resume   = e1000_resume,
5351 #endif
5352         .shutdown = e1000_shutdown,
5353         .err_handler = &e1000_err_handler
5354 };
5355
5356 /**
5357  * e1000_init_module - Driver Registration Routine
5358  *
5359  * e1000_init_module is the first routine called when the driver is
5360  * loaded. All it does is register with the PCI subsystem.
5361  **/
5362 static int __init e1000_init_module(void)
5363 {
5364         int ret;
5365         printk(KERN_INFO "%s: Intel(R) PRO/1000 Network Driver - %s\n",
5366                e1000e_driver_name, e1000e_driver_version);
5367         printk(KERN_INFO "%s: Copyright (c) 1999 - 2009 Intel Corporation.\n",
5368                e1000e_driver_name);
5369         ret = pci_register_driver(&e1000_driver);
5370         pm_qos_add_requirement(PM_QOS_CPU_DMA_LATENCY, e1000e_driver_name,
5371                                PM_QOS_DEFAULT_VALUE);
5372
5373         return ret;
5374 }
5375 module_init(e1000_init_module);
5376
5377 /**
5378  * e1000_exit_module - Driver Exit Cleanup Routine
5379  *
5380  * e1000_exit_module is called just before the driver is removed
5381  * from memory.
5382  **/
5383 static void __exit e1000_exit_module(void)
5384 {
5385         pci_unregister_driver(&e1000_driver);
5386         pm_qos_remove_requirement(PM_QOS_CPU_DMA_LATENCY, e1000e_driver_name);
5387 }
5388 module_exit(e1000_exit_module);
5389
5390
5391 MODULE_AUTHOR("Intel Corporation, <linux.nics@intel.com>");
5392 MODULE_DESCRIPTION("Intel(R) PRO/1000 Network Driver");
5393 MODULE_LICENSE("GPL");
5394 MODULE_VERSION(DRV_VERSION);
5395
5396 /* e1000_main.c */