drivers/net/e1000e/netdev.c

   1 /*******************************************************************************
   2
   3   Intel PRO/1000 Linux driver
   4   Copyright(c) 1999 - 2008 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 #ifdef DEBUG
  69 /**
  70  * e1000_get_hw_dev_name - return device name string
  71  * used by hardware layer to print debugging information
  72  **/
  73 char *e1000e_get_hw_dev_name(struct e1000_hw *hw)
  74 {
  75         return hw->adapter->netdev->name;
  76 }
  77 #endif
  78
  79 /**
  80  * e1000_desc_unused - calculate if we have unused descriptors
  81  **/
  82 static int e1000_desc_unused(struct e1000_ring *ring)
  83 {
  84         if (ring->next_to_clean > ring->next_to_use)
  85                 return ring->next_to_clean - ring->next_to_use - 1;
  86
  87         return ring->count + ring->next_to_clean - ring->next_to_use - 1;
  88 }
  89
  90 /**
  91  * e1000_receive_skb - helper function to handle Rx indications
  92  * @adapter: board private structure
  93  * @status: descriptor status field as written by hardware
  94  * @vlan: descriptor vlan field as written by hardware (no le/be conversion)
  95  * @skb: pointer to sk_buff to be indicated to stack
  96  **/
  97 static void e1000_receive_skb(struct e1000_adapter *adapter,
  98                               struct net_device *netdev,
  99                               struct sk_buff *skb,
 100                               u8 status, __le16 vlan)
 101 {
 102         skb->protocol = eth_type_trans(skb, netdev);
 103
 104         if (adapter->vlgrp && (status & E1000_RXD_STAT_VP))
 105                 vlan_gro_receive(&adapter->napi, adapter->vlgrp,
 106                                  le16_to_cpu(vlan), skb);
 107         else
 108                 napi_gro_receive(&adapter->napi, skb);
 109 }
 110
 111 /**
 112  * e1000_rx_checksum - Receive Checksum Offload for 82543
 113  * @adapter:     board private structure
 114  * @status_err:  receive descriptor status and error fields
 115  * @csum:       receive descriptor csum field
 116  * @sk_buff:     socket buffer with received data
 117  **/
 118 static void e1000_rx_checksum(struct e1000_adapter *adapter, u32 status_err,
 119                               u32 csum, struct sk_buff *skb)
 120 {
 121         u16 status = (u16)status_err;
 122         u8 errors = (u8)(status_err >> 24);
 123         skb->ip_summed = CHECKSUM_NONE;
 124
 125         /* Ignore Checksum bit is set */
 126         if (status & E1000_RXD_STAT_IXSM)
 127                 return;
 128         /* TCP/UDP checksum error bit is set */
 129         if (errors & E1000_RXD_ERR_TCPE) {
 130                 /* let the stack verify checksum errors */
 131                 adapter->hw_csum_err++;
 132                 return;
 133         }
 134
 135         /* TCP/UDP Checksum has not been calculated */
 136         if (!(status & (E1000_RXD_STAT_TCPCS | E1000_RXD_STAT_UDPCS)))
 137                 return;
 138
 139         /* It must be a TCP or UDP packet with a valid checksum */
 140         if (status & E1000_RXD_STAT_TCPCS) {
 141                 /* TCP checksum is good */
 142                 skb->ip_summed = CHECKSUM_UNNECESSARY;
 143         } else {
 144                 /*
 145                  * IP fragment with UDP payload
 146                  * Hardware complements the payload checksum, so we undo it
 147                  * and then put the value in host order for further stack use.
 148                  */
 149                 __sum16 sum = (__force __sum16)htons(csum);
 150                 skb->csum = csum_unfold(~sum);
 151                 skb->ip_summed = CHECKSUM_COMPLETE;
 152         }
 153         adapter->hw_csum_good++;
 154 }
 155
 156 /**
 157  * e1000_alloc_rx_buffers - Replace used receive buffers; legacy & extended
 158  * @adapter: address of board private structure
 159  **/
 160 static void e1000_alloc_rx_buffers(struct e1000_adapter *adapter,
 161                                    int cleaned_count)
 162 {
 163         struct net_device *netdev = adapter->netdev;
 164         struct pci_dev *pdev = adapter->pdev;
 165         struct e1000_ring *rx_ring = adapter->rx_ring;
 166         struct e1000_rx_desc *rx_desc;
 167         struct e1000_buffer *buffer_info;
 168         struct sk_buff *skb;
 169         unsigned int i;
 170         unsigned int bufsz = adapter->rx_buffer_len;
 171
 172         i = rx_ring->next_to_use;
 173         buffer_info = &rx_ring->buffer_info[i];
 174
 175         while (cleaned_count--) {
 176                 skb = buffer_info->skb;
 177                 if (skb) {
 178                         skb_trim(skb, 0);
 179                         goto map_skb;
 180                 }
 181
 182                 skb = netdev_alloc_skb_ip_align(netdev, bufsz);
 183                 if (!skb) {
 184                         /* Better luck next round */
 185                         adapter->alloc_rx_buff_failed++;
 186                         break;
 187                 }
 188
 189                 buffer_info->skb = skb;
 190 map_skb:
 191                 buffer_info->dma = pci_map_single(pdev, skb->data,
 192                                                   adapter->rx_buffer_len,
 193                                                   PCI_DMA_FROMDEVICE);
 194                 if (pci_dma_mapping_error(pdev, buffer_info->dma)) {
 195                         dev_err(&pdev->dev, "RX DMA map failed\n");
 196                         adapter->rx_dma_failed++;
 197                         break;
 198                 }
 199
 200                 rx_desc = E1000_RX_DESC(*rx_ring, i);
 201                 rx_desc->buffer_addr = cpu_to_le64(buffer_info->dma);
 202
 203                 i++;
 204                 if (i == rx_ring->count)
 205                         i = 0;
 206                 buffer_info = &rx_ring->buffer_info[i];
 207         }
 208
 209         if (rx_ring->next_to_use != i) {
 210                 rx_ring->next_to_use = i;
 211                 if (i-- == 0)
 212                         i = (rx_ring->count - 1);
 213
 214                 /*
 215                  * Force memory writes to complete before letting h/w
 216                  * know there are new descriptors to fetch.  (Only
 217                  * applicable for weak-ordered memory model archs,
 218                  * such as IA-64).
 219                  */
 220                 wmb();
 221                 writel(i, adapter->hw.hw_addr + rx_ring->tail);
 222         }
 223 }
 224
 225 /**
 226  * e1000_alloc_rx_buffers_ps - Replace used receive buffers; packet split
 227  * @adapter: address of board private structure
 228  **/
 229 static void e1000_alloc_rx_buffers_ps(struct e1000_adapter *adapter,
 230                                       int cleaned_count)
 231 {
 232         struct net_device *netdev = adapter->netdev;
 233         struct pci_dev *pdev = adapter->pdev;
 234         union e1000_rx_desc_packet_split *rx_desc;
 235         struct e1000_ring *rx_ring = adapter->rx_ring;
 236         struct e1000_buffer *buffer_info;
 237         struct e1000_ps_page *ps_page;
 238         struct sk_buff *skb;
 239         unsigned int i, j;
 240
 241         i = rx_ring->next_to_use;
 242         buffer_info = &rx_ring->buffer_info[i];
 243
 244         while (cleaned_count--) {
 245                 rx_desc = E1000_RX_DESC_PS(*rx_ring, i);
 246
 247                 for (j = 0; j < PS_PAGE_BUFFERS; j++) {
 248                         ps_page = &buffer_info->ps_pages[j];
 249                         if (j >= adapter->rx_ps_pages) {
 250                                 /* all unused desc entries get hw null ptr */
 251                                 rx_desc->read.buffer_addr[j+1] = ~cpu_to_le64(0);
 252                                 continue;
 253                         }
 254                         if (!ps_page->page) {
 255                                 ps_page->page = alloc_page(GFP_ATOMIC);
 256                                 if (!ps_page->page) {
 257                                         adapter->alloc_rx_buff_failed++;
 258                                         goto no_buffers;
 259                                 }
 260                                 ps_page->dma = pci_map_page(pdev,
 261                                                    ps_page->page,
 262                                                    0, PAGE_SIZE,
 263                                                    PCI_DMA_FROMDEVICE);
 264                                 if (pci_dma_mapping_error(pdev, ps_page->dma)) {
 265                                         dev_err(&adapter->pdev->dev,
 266                                           "RX DMA page map failed\n");
 267                                         adapter->rx_dma_failed++;
 268                                         goto no_buffers;
 269                                 }
 270                         }
 271                         /*
 272                          * Refresh the desc even if buffer_addrs
 273                          * didn't change because each write-back
 274                          * erases this info.
 275                          */
 276                         rx_desc->read.buffer_addr[j+1] =
 277                              cpu_to_le64(ps_page->dma);
 278                 }
 279
 280                 skb = netdev_alloc_skb_ip_align(netdev,
 281                                                 adapter->rx_ps_bsize0);
 282
 283                 if (!skb) {
 284                         adapter->alloc_rx_buff_failed++;
 285                         break;
 286                 }
 287
 288                 buffer_info->skb = skb;
 289                 buffer_info->dma = pci_map_single(pdev, skb->data,
 290                                                   adapter->rx_ps_bsize0,
 291                                                   PCI_DMA_FROMDEVICE);
 292                 if (pci_dma_mapping_error(pdev, buffer_info->dma)) {
 293                         dev_err(&pdev->dev, "RX DMA map failed\n");
 294                         adapter->rx_dma_failed++;
 295                         /* cleanup skb */
 296                         dev_kfree_skb_any(skb);
 297                         buffer_info->skb = NULL;
 298                         break;
 299                 }
 300
 301                 rx_desc->read.buffer_addr[0] = cpu_to_le64(buffer_info->dma);
 302
 303                 i++;
 304                 if (i == rx_ring->count)
 305                         i = 0;
 306                 buffer_info = &rx_ring->buffer_info[i];
 307         }
 308
 309 no_buffers:
 310         if (rx_ring->next_to_use != i) {
 311                 rx_ring->next_to_use = i;
 312
 313                 if (!(i--))
 314                         i = (rx_ring->count - 1);
 315
 316                 /*
 317                  * Force memory writes to complete before letting h/w
 318                  * know there are new descriptors to fetch.  (Only
 319                  * applicable for weak-ordered memory model archs,
 320                  * such as IA-64).
 321                  */
 322                 wmb();
 323                 /*
 324                  * Hardware increments by 16 bytes, but packet split
 325                  * descriptors are 32 bytes...so we increment tail
 326                  * twice as much.
 327                  */
 328                 writel(i<<1, adapter->hw.hw_addr + rx_ring->tail);
 329         }
 330 }
 331
 332 /**
 333  * e1000_alloc_jumbo_rx_buffers - Replace used jumbo receive buffers
 334  * @adapter: address of board private structure
 335  * @cleaned_count: number of buffers to allocate this pass
 336  **/
 337
 338 static void e1000_alloc_jumbo_rx_buffers(struct e1000_adapter *adapter,
 339                                          int cleaned_count)
 340 {
 341         struct net_device *netdev = adapter->netdev;
 342         struct pci_dev *pdev = adapter->pdev;
 343         struct e1000_rx_desc *rx_desc;
 344         struct e1000_ring *rx_ring = adapter->rx_ring;
 345         struct e1000_buffer *buffer_info;
 346         struct sk_buff *skb;
 347         unsigned int i;
 348         unsigned int bufsz = 256 - 16 /* for skb_reserve */;
 349
 350         i = rx_ring->next_to_use;
 351         buffer_info = &rx_ring->buffer_info[i];
 352
 353         while (cleaned_count--) {
 354                 skb = buffer_info->skb;
 355                 if (skb) {
 356                         skb_trim(skb, 0);
 357                         goto check_page;
 358                 }
 359
 360                 skb = netdev_alloc_skb_ip_align(netdev, bufsz);
 361                 if (unlikely(!skb)) {
 362                         /* Better luck next round */
 363                         adapter->alloc_rx_buff_failed++;
 364                         break;
 365                 }
 366
 367                 buffer_info->skb = skb;
 368 check_page:
 369                 /* allocate a new page if necessary */
 370                 if (!buffer_info->page) {
 371                         buffer_info->page = alloc_page(GFP_ATOMIC);
 372                         if (unlikely(!buffer_info->page)) {
 373                                 adapter->alloc_rx_buff_failed++;
 374                                 break;
 375                         }
 376                 }
 377
 378                 if (!buffer_info->dma)
 379                         buffer_info->dma = pci_map_page(pdev,
 380                                                         buffer_info->page, 0,
 381                                                         PAGE_SIZE,
 382                                                         PCI_DMA_FROMDEVICE);
 383
 384                 rx_desc = E1000_RX_DESC(*rx_ring, i);
 385                 rx_desc->buffer_addr = cpu_to_le64(buffer_info->dma);
 386
 387                 if (unlikely(++i == rx_ring->count))
 388                         i = 0;
 389                 buffer_info = &rx_ring->buffer_info[i];
 390         }
 391
 392         if (likely(rx_ring->next_to_use != i)) {
 393                 rx_ring->next_to_use = i;
 394                 if (unlikely(i-- == 0))
 395                         i = (rx_ring->count - 1);
 396
 397                 /* Force memory writes to complete before letting h/w
 398                  * know there are new descriptors to fetch.  (Only
 399                  * applicable for weak-ordered memory model archs,
 400                  * such as IA-64). */
 401                 wmb();
 402                 writel(i, adapter->hw.hw_addr + rx_ring->tail);
 403         }
 404 }
 405
 406 /**
 407  * e1000_clean_rx_irq - Send received data up the network stack; legacy
 408  * @adapter: board private structure
 409  *
 410  * the return value indicates whether actual cleaning was done, there
 411  * is no guarantee that everything was cleaned
 412  **/
 413 static bool e1000_clean_rx_irq(struct e1000_adapter *adapter,
 414                                int *work_done, int work_to_do)
 415 {
 416         struct net_device *netdev = adapter->netdev;
 417         struct pci_dev *pdev = adapter->pdev;
 418         struct e1000_ring *rx_ring = adapter->rx_ring;
 419         struct e1000_rx_desc *rx_desc, *next_rxd;
 420         struct e1000_buffer *buffer_info, *next_buffer;
 421         u32 length;
 422         unsigned int i;
 423         int cleaned_count = 0;
 424         bool cleaned = 0;
 425         unsigned int total_rx_bytes = 0, total_rx_packets = 0;
 426
 427         i = rx_ring->next_to_clean;
 428         rx_desc = E1000_RX_DESC(*rx_ring, i);
 429         buffer_info = &rx_ring->buffer_info[i];
 430
 431         while (rx_desc->status & E1000_RXD_STAT_DD) {
 432                 struct sk_buff *skb;
 433                 u8 status;
 434
 435                 if (*work_done >= work_to_do)
 436                         break;
 437                 (*work_done)++;
 438
 439                 status = rx_desc->status;
 440                 skb = buffer_info->skb;
 441                 buffer_info->skb = NULL;
 442
 443                 prefetch(skb->data - NET_IP_ALIGN);
 444
 445                 i++;
 446                 if (i == rx_ring->count)
 447                         i = 0;
 448                 next_rxd = E1000_RX_DESC(*rx_ring, i);
 449                 prefetch(next_rxd);
 450
 451                 next_buffer = &rx_ring->buffer_info[i];
 452
 453                 cleaned = 1;
 454                 cleaned_count++;
 455                 pci_unmap_single(pdev,
 456                                  buffer_info->dma,
 457                                  adapter->rx_buffer_len,
 458                                  PCI_DMA_FROMDEVICE);
 459                 buffer_info->dma = 0;
 460
 461                 length = le16_to_cpu(rx_desc->length);
 462
 463                 /* !EOP means multiple descriptors were used to store a single
 464                  * packet, also make sure the frame isn't just CRC only */
 465                 if (!(status & E1000_RXD_STAT_EOP) || (length <= 4)) {
 466                         /* All receives must fit into a single buffer */
 467                         e_dbg("%s: Receive packet consumed multiple buffers\n",
 468                               netdev->name);
 469                         /* recycle */
 470                         buffer_info->skb = skb;
 471                         goto next_desc;
 472                 }
 473
 474                 if (rx_desc->errors & E1000_RXD_ERR_FRAME_ERR_MASK) {
 475                         /* recycle */
 476                         buffer_info->skb = skb;
 477                         goto next_desc;
 478                 }
 479
 480                 /* adjust length to remove Ethernet CRC */
 481                 if (!(adapter->flags2 & FLAG2_CRC_STRIPPING))
 482                         length -= 4;
 483
 484                 total_rx_bytes += length;
 485                 total_rx_packets++;
 486
 487                 /*
 488                  * code added for copybreak, this should improve
 489                  * performance for small packets with large amounts
 490                  * of reassembly being done in the stack
 491                  */
 492                 if (length < copybreak) {
 493                         struct sk_buff *new_skb =
 494                             netdev_alloc_skb_ip_align(netdev, length);
 495                         if (new_skb) {
 496                                 skb_copy_to_linear_data_offset(new_skb,
 497                                                                -NET_IP_ALIGN,
 498                                                                (skb->data -
 499                                                                 NET_IP_ALIGN),
 500                                                                (length +
 501                                                                 NET_IP_ALIGN));
 502                                 /* save the skb in buffer_info as good */
 503                                 buffer_info->skb = skb;
 504                                 skb = new_skb;
 505                         }
 506                         /* else just continue with the old one */
 507                 }
 508                 /* end copybreak code */
 509                 skb_put(skb, length);
 510
 511                 /* Receive Checksum Offload */
 512                 e1000_rx_checksum(adapter,
 513                                   (u32)(status) |
 514                                   ((u32)(rx_desc->errors) << 24),
 515                                   le16_to_cpu(rx_desc->csum), skb);
 516
 517                 e1000_receive_skb(adapter, netdev, skb,status,rx_desc->special);
 518
 519 next_desc:
 520                 rx_desc->status = 0;
 521
 522                 /* return some buffers to hardware, one at a time is too slow */
 523                 if (cleaned_count >= E1000_RX_BUFFER_WRITE) {
 524                         adapter->alloc_rx_buf(adapter, cleaned_count);
 525                         cleaned_count = 0;
 526                 }
 527
 528                 /* use prefetched values */
 529                 rx_desc = next_rxd;
 530                 buffer_info = next_buffer;
 531         }
 532         rx_ring->next_to_clean = i;
 533
 534         cleaned_count = e1000_desc_unused(rx_ring);
 535         if (cleaned_count)
 536                 adapter->alloc_rx_buf(adapter, cleaned_count);
 537
 538         adapter->total_rx_bytes += total_rx_bytes;
 539         adapter->total_rx_packets += total_rx_packets;
 540         netdev->stats.rx_bytes += total_rx_bytes;
 541         netdev->stats.rx_packets += total_rx_packets;
 542         return cleaned;
 543 }
 544
 545 static void e1000_put_txbuf(struct e1000_adapter *adapter,
 546                              struct e1000_buffer *buffer_info)
 547 {
 548         buffer_info->dma = 0;
 549         if (buffer_info->skb) {
 550                 skb_dma_unmap(&adapter->pdev->dev, buffer_info->skb,
 551                               DMA_TO_DEVICE);
 552                 dev_kfree_skb_any(buffer_info->skb);
 553                 buffer_info->skb = NULL;
 554         }
 555         buffer_info->time_stamp = 0;
 556 }
 557
 558 static void e1000_print_tx_hang(struct e1000_adapter *adapter)
 559 {
 560         struct e1000_ring *tx_ring = adapter->tx_ring;
 561         unsigned int i = tx_ring->next_to_clean;
 562         unsigned int eop = tx_ring->buffer_info[i].next_to_watch;
 563         struct e1000_tx_desc *eop_desc = E1000_TX_DESC(*tx_ring, eop);
 564
 565         /* detected Tx unit hang */
 566         e_err("Detected Tx Unit Hang:\n"
 567               "  TDH                  <%x>\n"
 568               "  TDT                  <%x>\n"
 569               "  next_to_use          <%x>\n"
 570               "  next_to_clean        <%x>\n"
 571               "buffer_info[next_to_clean]:\n"
 572               "  time_stamp           <%lx>\n"
 573               "  next_to_watch        <%x>\n"
 574               "  jiffies              <%lx>\n"
 575               "  next_to_watch.status <%x>\n",
 576               readl(adapter->hw.hw_addr + tx_ring->head),
 577               readl(adapter->hw.hw_addr + tx_ring->tail),
 578               tx_ring->next_to_use,
 579               tx_ring->next_to_clean,
 580               tx_ring->buffer_info[eop].time_stamp,
 581               eop,
 582               jiffies,
 583               eop_desc->upper.fields.status);
 584 }
 585
 586 /**
 587  * e1000_clean_tx_irq - Reclaim resources after transmit completes
 588  * @adapter: board private structure
 589  *
 590  * the return value indicates whether actual cleaning was done, there
 591  * is no guarantee that everything was cleaned
 592  **/
 593 static bool e1000_clean_tx_irq(struct e1000_adapter *adapter)
 594 {
 595         struct net_device *netdev = adapter->netdev;
 596         struct e1000_hw *hw = &adapter->hw;
 597         struct e1000_ring *tx_ring = adapter->tx_ring;
 598         struct e1000_tx_desc *tx_desc, *eop_desc;
 599         struct e1000_buffer *buffer_info;
 600         unsigned int i, eop;
 601         unsigned int count = 0;
 602         unsigned int total_tx_bytes = 0, total_tx_packets = 0;
 603
 604         i = tx_ring->next_to_clean;
 605         eop = tx_ring->buffer_info[i].next_to_watch;
 606         eop_desc = E1000_TX_DESC(*tx_ring, eop);
 607
 608         while ((eop_desc->upper.data & cpu_to_le32(E1000_TXD_STAT_DD)) &&
 609                (count < tx_ring->count)) {
 610                 bool cleaned = false;
 611                 for (; !cleaned; count++) {
 612                         tx_desc = E1000_TX_DESC(*tx_ring, i);
 613                         buffer_info = &tx_ring->buffer_info[i];
 614                         cleaned = (i == eop);
 615
 616                         if (cleaned) {
 617                                 struct sk_buff *skb = buffer_info->skb;
 618                                 unsigned int segs, bytecount;
 619                                 segs = skb_shinfo(skb)->gso_segs ?: 1;
 620                                 /* multiply data chunks by size of headers */
 621                                 bytecount = ((segs - 1) * skb_headlen(skb)) +
 622                                             skb->len;
 623                                 total_tx_packets += segs;
 624                                 total_tx_bytes += bytecount;
 625                         }
 626
 627                         e1000_put_txbuf(adapter, buffer_info);
 628                         tx_desc->upper.data = 0;
 629
 630                         i++;
 631                         if (i == tx_ring->count)
 632                                 i = 0;
 633                 }
 634
 635                 eop = tx_ring->buffer_info[i].next_to_watch;
 636                 eop_desc = E1000_TX_DESC(*tx_ring, eop);
 637         }
 638
 639         tx_ring->next_to_clean = i;
 640
 641 #define TX_WAKE_THRESHOLD 32
 642         if (count && netif_carrier_ok(netdev) &&
 643             e1000_desc_unused(tx_ring) >= TX_WAKE_THRESHOLD) {
 644                 /* Make sure that anybody stopping the queue after this
 645                  * sees the new next_to_clean.
 646                  */
 647                 smp_mb();
 648
 649                 if (netif_queue_stopped(netdev) &&
 650                     !(test_bit(__E1000_DOWN, &adapter->state))) {
 651                         netif_wake_queue(netdev);
 652                         ++adapter->restart_queue;
 653                 }
 654         }
 655
 656         if (adapter->detect_tx_hung) {
 657                 /* Detect a transmit hang in hardware, this serializes the
 658                  * check with the clearing of time_stamp and movement of i */
 659                 adapter->detect_tx_hung = 0;
 660                 if (tx_ring->buffer_info[i].time_stamp &&
 661                     time_after(jiffies, tx_ring->buffer_info[i].time_stamp
 662                                + (adapter->tx_timeout_factor * HZ))
 663                     && !(er32(STATUS) & E1000_STATUS_TXOFF)) {
 664                         e1000_print_tx_hang(adapter);
 665                         netif_stop_queue(netdev);
 666                 }
 667         }
 668         adapter->total_tx_bytes += total_tx_bytes;
 669         adapter->total_tx_packets += total_tx_packets;
 670         netdev->stats.tx_bytes += total_tx_bytes;
 671         netdev->stats.tx_packets += total_tx_packets;
 672         return (count < tx_ring->count);
 673 }
 674
 675 /**
 676  * e1000_clean_rx_irq_ps - Send received data up the network stack; packet split
 677  * @adapter: board private structure
 678  *
 679  * the return value indicates whether actual cleaning was done, there
 680  * is no guarantee that everything was cleaned
 681  **/
 682 static bool e1000_clean_rx_irq_ps(struct e1000_adapter *adapter,
 683                                   int *work_done, int work_to_do)
 684 {
 685         union e1000_rx_desc_packet_split *rx_desc, *next_rxd;
 686         struct net_device *netdev = adapter->netdev;
 687         struct pci_dev *pdev = adapter->pdev;
 688         struct e1000_ring *rx_ring = adapter->rx_ring;
 689         struct e1000_buffer *buffer_info, *next_buffer;
 690         struct e1000_ps_page *ps_page;
 691         struct sk_buff *skb;
 692         unsigned int i, j;
 693         u32 length, staterr;
 694         int cleaned_count = 0;
 695         bool cleaned = 0;
 696         unsigned int total_rx_bytes = 0, total_rx_packets = 0;
 697
 698         i = rx_ring->next_to_clean;
 699         rx_desc = E1000_RX_DESC_PS(*rx_ring, i);
 700         staterr = le32_to_cpu(rx_desc->wb.middle.status_error);
 701         buffer_info = &rx_ring->buffer_info[i];
 702
 703         while (staterr & E1000_RXD_STAT_DD) {
 704                 if (*work_done >= work_to_do)
 705                         break;
 706                 (*work_done)++;
 707                 skb = buffer_info->skb;
 708
 709                 /* in the packet split case this is header only */
 710                 prefetch(skb->data - NET_IP_ALIGN);
 711
 712                 i++;
 713                 if (i == rx_ring->count)
 714                         i = 0;
 715                 next_rxd = E1000_RX_DESC_PS(*rx_ring, i);
 716                 prefetch(next_rxd);
 717
 718                 next_buffer = &rx_ring->buffer_info[i];
 719
 720                 cleaned = 1;
 721                 cleaned_count++;
 722                 pci_unmap_single(pdev, buffer_info->dma,
 723                                  adapter->rx_ps_bsize0,
 724                                  PCI_DMA_FROMDEVICE);
 725                 buffer_info->dma = 0;
 726
 727                 if (!(staterr & E1000_RXD_STAT_EOP)) {
 728                         e_dbg("%s: Packet Split buffers didn't pick up the "
 729                               "full packet\n", netdev->name);
 730                         dev_kfree_skb_irq(skb);
 731                         goto next_desc;
 732                 }
 733
 734                 if (staterr & E1000_RXDEXT_ERR_FRAME_ERR_MASK) {
 735                         dev_kfree_skb_irq(skb);
 736                         goto next_desc;
 737                 }
 738
 739                 length = le16_to_cpu(rx_desc->wb.middle.length0);
 740
 741                 if (!length) {
 742                         e_dbg("%s: Last part of the packet spanning multiple "
 743                               "descriptors\n", netdev->name);
 744                         dev_kfree_skb_irq(skb);
 745                         goto next_desc;
 746                 }
 747
 748                 /* Good Receive */
 749                 skb_put(skb, length);
 750
 751                 {
 752                 /*
 753                  * this looks ugly, but it seems compiler issues make it
 754                  * more efficient than reusing j
 755                  */
 756                 int l1 = le16_to_cpu(rx_desc->wb.upper.length[0]);
 757
 758                 /*
 759                  * page alloc/put takes too long and effects small packet
 760                  * throughput, so unsplit small packets and save the alloc/put
 761                  * only valid in softirq (napi) context to call kmap_*
 762                  */
 763                 if (l1 && (l1 <= copybreak) &&
 764                     ((length + l1) <= adapter->rx_ps_bsize0)) {
 765                         u8 *vaddr;
 766
 767                         ps_page = &buffer_info->ps_pages[0];
 768
 769                         /*
 770                          * there is no documentation about how to call
 771                          * kmap_atomic, so we can't hold the mapping
 772                          * very long
 773                          */
 774                         pci_dma_sync_single_for_cpu(pdev, ps_page->dma,
 775                                 PAGE_SIZE, PCI_DMA_FROMDEVICE);
 776                         vaddr = kmap_atomic(ps_page->page, KM_SKB_DATA_SOFTIRQ);
 777                         memcpy(skb_tail_pointer(skb), vaddr, l1);
 778                         kunmap_atomic(vaddr, KM_SKB_DATA_SOFTIRQ);
 779                         pci_dma_sync_single_for_device(pdev, ps_page->dma,
 780                                 PAGE_SIZE, PCI_DMA_FROMDEVICE);
 781
 782                         /* remove the CRC */
 783                         if (!(adapter->flags2 & FLAG2_CRC_STRIPPING))
 784                                 l1 -= 4;
 785
 786                         skb_put(skb, l1);
 787                         goto copydone;
 788                 } /* if */
 789                 }
 790
 791                 for (j = 0; j < PS_PAGE_BUFFERS; j++) {
 792                         length = le16_to_cpu(rx_desc->wb.upper.length[j]);
 793                         if (!length)
 794                                 break;
 795
 796                         ps_page = &buffer_info->ps_pages[j];
 797                         pci_unmap_page(pdev, ps_page->dma, PAGE_SIZE,
 798                                        PCI_DMA_FROMDEVICE);
 799                         ps_page->dma = 0;
 800                         skb_fill_page_desc(skb, j, ps_page->page, 0, length);
 801                         ps_page->page = NULL;
 802                         skb->len += length;
 803                         skb->data_len += length;
 804                         skb->truesize += length;
 805                 }
 806
 807                 /* strip the ethernet crc, problem is we're using pages now so
 808                  * this whole operation can get a little cpu intensive
 809                  */
 810                 if (!(adapter->flags2 & FLAG2_CRC_STRIPPING))
 811                         pskb_trim(skb, skb->len - 4);
 812
 813 copydone:
 814                 total_rx_bytes += skb->len;
 815                 total_rx_packets++;
 816
 817                 e1000_rx_checksum(adapter, staterr, le16_to_cpu(
 818                         rx_desc->wb.lower.hi_dword.csum_ip.csum), skb);
 819
 820                 if (rx_desc->wb.upper.header_status &
 821                            cpu_to_le16(E1000_RXDPS_HDRSTAT_HDRSP))
 822                         adapter->rx_hdr_split++;
 823
 824                 e1000_receive_skb(adapter, netdev, skb,
 825                                   staterr, rx_desc->wb.middle.vlan);
 826
 827 next_desc:
 828                 rx_desc->wb.middle.status_error &= cpu_to_le32(~0xFF);
 829                 buffer_info->skb = NULL;
 830
 831                 /* return some buffers to hardware, one at a time is too slow */
 832                 if (cleaned_count >= E1000_RX_BUFFER_WRITE) {
 833                         adapter->alloc_rx_buf(adapter, cleaned_count);
 834                         cleaned_count = 0;
 835                 }
 836
 837                 /* use prefetched values */
 838                 rx_desc = next_rxd;
 839                 buffer_info = next_buffer;
 840
 841                 staterr = le32_to_cpu(rx_desc->wb.middle.status_error);
 842         }
 843         rx_ring->next_to_clean = i;
 844
 845         cleaned_count = e1000_desc_unused(rx_ring);
 846         if (cleaned_count)
 847                 adapter->alloc_rx_buf(adapter, cleaned_count);
 848
 849         adapter->total_rx_bytes += total_rx_bytes;
 850         adapter->total_rx_packets += total_rx_packets;
 851         netdev->stats.rx_bytes += total_rx_bytes;
 852         netdev->stats.rx_packets += total_rx_packets;
 853         return cleaned;
 854 }
 855
 856 /**
 857  * e1000_consume_page - helper function
 858  **/
 859 static void e1000_consume_page(struct e1000_buffer *bi, struct sk_buff *skb,
 860                                u16 length)
 861 {
 862         bi->page = NULL;
 863         skb->len += length;
 864         skb->data_len += length;
 865         skb->truesize += length;
 866 }
 867
 868 /**
 869  * e1000_clean_jumbo_rx_irq - Send received data up the network stack; legacy
 870  * @adapter: board private structure
 871  *
 872  * the return value indicates whether actual cleaning was done, there
 873  * is no guarantee that everything was cleaned
 874  **/
 875
 876 static bool e1000_clean_jumbo_rx_irq(struct e1000_adapter *adapter,
 877                                      int *work_done, int work_to_do)
 878 {
 879         struct net_device *netdev = adapter->netdev;
 880         struct pci_dev *pdev = adapter->pdev;
 881         struct e1000_ring *rx_ring = adapter->rx_ring;
 882         struct e1000_rx_desc *rx_desc, *next_rxd;
 883         struct e1000_buffer *buffer_info, *next_buffer;
 884         u32 length;
 885         unsigned int i;
 886         int cleaned_count = 0;
 887         bool cleaned = false;
 888         unsigned int total_rx_bytes=0, total_rx_packets=0;
 889
 890         i = rx_ring->next_to_clean;
 891         rx_desc = E1000_RX_DESC(*rx_ring, i);
 892         buffer_info = &rx_ring->buffer_info[i];
 893
 894         while (rx_desc->status & E1000_RXD_STAT_DD) {
 895                 struct sk_buff *skb;
 896                 u8 status;
 897
 898                 if (*work_done >= work_to_do)
 899                         break;
 900                 (*work_done)++;
 901
 902                 status = rx_desc->status;
 903                 skb = buffer_info->skb;
 904                 buffer_info->skb = NULL;
 905
 906                 ++i;
 907                 if (i == rx_ring->count)
 908                         i = 0;
 909                 next_rxd = E1000_RX_DESC(*rx_ring, i);
 910                 prefetch(next_rxd);
 911
 912                 next_buffer = &rx_ring->buffer_info[i];
 913
 914                 cleaned = true;
 915                 cleaned_count++;
 916                 pci_unmap_page(pdev, buffer_info->dma, PAGE_SIZE,
 917                                PCI_DMA_FROMDEVICE);
 918                 buffer_info->dma = 0;
 919
 920                 length = le16_to_cpu(rx_desc->length);
 921
 922                 /* errors is only valid for DD + EOP descriptors */
 923                 if (unlikely((status & E1000_RXD_STAT_EOP) &&
 924                     (rx_desc->errors & E1000_RXD_ERR_FRAME_ERR_MASK))) {
 925                                 /* recycle both page and skb */
 926                                 buffer_info->skb = skb;
 927                                 /* an error means any chain goes out the window
 928                                  * too */
 929                                 if (rx_ring->rx_skb_top)
 930                                         dev_kfree_skb(rx_ring->rx_skb_top);
 931                                 rx_ring->rx_skb_top = NULL;
 932                                 goto next_desc;
 933                 }
 934
 935 #define rxtop rx_ring->rx_skb_top
 936                 if (!(status & E1000_RXD_STAT_EOP)) {
 937                         /* this descriptor is only the beginning (or middle) */
 938                         if (!rxtop) {
 939                                 /* this is the beginning of a chain */
 940                                 rxtop = skb;
 941                                 skb_fill_page_desc(rxtop, 0, buffer_info->page,
 942                                                    0, length);
 943                         } else {
 944                                 /* this is the middle of a chain */
 945                                 skb_fill_page_desc(rxtop,
 946                                     skb_shinfo(rxtop)->nr_frags,
 947                                     buffer_info->page, 0, length);
 948                                 /* re-use the skb, only consumed the page */
 949                                 buffer_info->skb = skb;
 950                         }
 951                         e1000_consume_page(buffer_info, rxtop, length);
 952                         goto next_desc;
 953                 } else {
 954                         if (rxtop) {
 955                                 /* end of the chain */
 956                                 skb_fill_page_desc(rxtop,
 957                                     skb_shinfo(rxtop)->nr_frags,
 958                                     buffer_info->page, 0, length);
 959                                 /* re-use the current skb, we only consumed the
 960                                  * page */
 961                                 buffer_info->skb = skb;
 962                                 skb = rxtop;
 963                                 rxtop = NULL;
 964                                 e1000_consume_page(buffer_info, skb, length);
 965                         } else {
 966                                 /* no chain, got EOP, this buf is the packet
 967                                  * copybreak to save the put_page/alloc_page */
 968                                 if (length <= copybreak &&
 969                                     skb_tailroom(skb) >= length) {
 970                                         u8 *vaddr;
 971                                         vaddr = kmap_atomic(buffer_info->page,
 972                                                            KM_SKB_DATA_SOFTIRQ);
 973                                         memcpy(skb_tail_pointer(skb), vaddr,
 974                                                length);
 975                                         kunmap_atomic(vaddr,
 976                                                       KM_SKB_DATA_SOFTIRQ);
 977                                         /* re-use the page, so don't erase
 978                                          * buffer_info->page */
 979                                         skb_put(skb, length);
 980                                 } else {
 981                                         skb_fill_page_desc(skb, 0,
 982                                                            buffer_info->page, 0,
 983                                                            length);
 984                                         e1000_consume_page(buffer_info, skb,
 985                                                            length);
 986                                 }
 987                         }
 988                 }
 989
 990                 /* Receive Checksum Offload XXX recompute due to CRC strip? */
 991                 e1000_rx_checksum(adapter,
 992                                   (u32)(status) |
 993                                   ((u32)(rx_desc->errors) << 24),
 994                                   le16_to_cpu(rx_desc->csum), skb);
 995
 996                 /* probably a little skewed due to removing CRC */
 997                 total_rx_bytes += skb->len;
 998                 total_rx_packets++;
 999
1000                 /* eth type trans needs skb->data to point to something */
1001                 if (!pskb_may_pull(skb, ETH_HLEN)) {
1002                         e_err("pskb_may_pull failed.\n");
1003                         dev_kfree_skb(skb);
1004                         goto next_desc;
1005                 }
1006
1007                 e1000_receive_skb(adapter, netdev, skb, status,
1008                                   rx_desc->special);
1009
1010 next_desc:
1011                 rx_desc->status = 0;
1012
1013                 /* return some buffers to hardware, one at a time is too slow */
1014                 if (unlikely(cleaned_count >= E1000_RX_BUFFER_WRITE)) {
1015                         adapter->alloc_rx_buf(adapter, cleaned_count);
1016                         cleaned_count = 0;
1017                 }
1018
1019                 /* use prefetched values */
1020                 rx_desc = next_rxd;
1021                 buffer_info = next_buffer;
1022         }
1023         rx_ring->next_to_clean = i;
1024
1025         cleaned_count = e1000_desc_unused(rx_ring);
1026         if (cleaned_count)
1027                 adapter->alloc_rx_buf(adapter, cleaned_count);
1028
1029         adapter->total_rx_bytes += total_rx_bytes;
1030         adapter->total_rx_packets += total_rx_packets;
1031         netdev->stats.rx_bytes += total_rx_bytes;
1032         netdev->stats.rx_packets += total_rx_packets;
1033         return cleaned;
1034 }
1035
1036 /**
1037  * e1000_clean_rx_ring - Free Rx Buffers per Queue
1038  * @adapter: board private structure
1039  **/
1040 static void e1000_clean_rx_ring(struct e1000_adapter *adapter)
1041 {
1042         struct e1000_ring *rx_ring = adapter->rx_ring;
1043         struct e1000_buffer *buffer_info;
1044         struct e1000_ps_page *ps_page;
1045         struct pci_dev *pdev = adapter->pdev;
1046         unsigned int i, j;
1047
1048         /* Free all the Rx ring sk_buffs */
1049         for (i = 0; i < rx_ring->count; i++) {
1050                 buffer_info = &rx_ring->buffer_info[i];
1051                 if (buffer_info->dma) {
1052                         if (adapter->clean_rx == e1000_clean_rx_irq)
1053                                 pci_unmap_single(pdev, buffer_info->dma,
1054                                                  adapter->rx_buffer_len,
1055                                                  PCI_DMA_FROMDEVICE);
1056                         else if (adapter->clean_rx == e1000_clean_jumbo_rx_irq)
1057                                 pci_unmap_page(pdev, buffer_info->dma,
1058                                                PAGE_SIZE,
1059                                                PCI_DMA_FROMDEVICE);
1060                         else if (adapter->clean_rx == e1000_clean_rx_irq_ps)
1061                                 pci_unmap_single(pdev, buffer_info->dma,
1062                                                  adapter->rx_ps_bsize0,
1063                                                  PCI_DMA_FROMDEVICE);
1064                         buffer_info->dma = 0;
1065                 }
1066
1067                 if (buffer_info->page) {
1068                         put_page(buffer_info->page);
1069                         buffer_info->page = NULL;
1070                 }
1071
1072                 if (buffer_info->skb) {
1073                         dev_kfree_skb(buffer_info->skb);
1074                         buffer_info->skb = NULL;
1075                 }
1076
1077                 for (j = 0; j < PS_PAGE_BUFFERS; j++) {
1078                         ps_page = &buffer_info->ps_pages[j];
1079                         if (!ps_page->page)
1080                                 break;
1081                         pci_unmap_page(pdev, ps_page->dma, PAGE_SIZE,
1082                                        PCI_DMA_FROMDEVICE);
1083                         ps_page->dma = 0;
1084                         put_page(ps_page->page);
1085                         ps_page->page = NULL;
1086                 }
1087         }
1088
1089         /* there also may be some cached data from a chained receive */
1090         if (rx_ring->rx_skb_top) {
1091                 dev_kfree_skb(rx_ring->rx_skb_top);
1092                 rx_ring->rx_skb_top = NULL;
1093         }
1094
1095         /* Zero out the descriptor ring */
1096         memset(rx_ring->desc, 0, rx_ring->size);
1097
1098         rx_ring->next_to_clean = 0;
1099         rx_ring->next_to_use = 0;
1100
1101         writel(0, adapter->hw.hw_addr + rx_ring->head);
1102         writel(0, adapter->hw.hw_addr + rx_ring->tail);
1103 }
1104
1105 static void e1000e_downshift_workaround(struct work_struct *work)
1106 {
1107         struct e1000_adapter *adapter = container_of(work,
1108                                         struct e1000_adapter, downshift_task);
1109
1110         e1000e_gig_downshift_workaround_ich8lan(&adapter->hw);
1111 }
1112
1113 /**
1114  * e1000_intr_msi - Interrupt Handler
1115  * @irq: interrupt number
1116  * @data: pointer to a network interface device structure
1117  **/
1118 static irqreturn_t e1000_intr_msi(int irq, void *data)
1119 {
1120         struct net_device *netdev = data;
1121         struct e1000_adapter *adapter = netdev_priv(netdev);
1122         struct e1000_hw *hw = &adapter->hw;
1123         u32 icr = er32(ICR);
1124
1125         /*
1126          * read ICR disables interrupts using IAM
1127          */
1128
1129         if (icr & E1000_ICR_LSC) {
1130                 hw->mac.get_link_status = 1;
1131                 /*
1132                  * ICH8 workaround-- Call gig speed drop workaround on cable
1133                  * disconnect (LSC) before accessing any PHY registers
1134                  */
1135                 if ((adapter->flags & FLAG_LSC_GIG_SPEED_DROP) &&
1136                     (!(er32(STATUS) & E1000_STATUS_LU)))
1137                         schedule_work(&adapter->downshift_task);
1138
1139                 /*
1140                  * 80003ES2LAN workaround-- For packet buffer work-around on
1141                  * link down event; disable receives here in the ISR and reset
1142                  * adapter in watchdog
1143                  */
1144                 if (netif_carrier_ok(netdev) &&
1145                     adapter->flags & FLAG_RX_NEEDS_RESTART) {
1146                         /* disable receives */
1147                         u32 rctl = er32(RCTL);
1148                         ew32(RCTL, rctl & ~E1000_RCTL_EN);
1149                         adapter->flags |= FLAG_RX_RESTART_NOW;
1150                 }
1151                 /* guard against interrupt when we're going down */
1152                 if (!test_bit(__E1000_DOWN, &adapter->state))
1153                         mod_timer(&adapter->watchdog_timer, jiffies + 1);
1154         }
1155
1156         if (napi_schedule_prep(&adapter->napi)) {
1157                 adapter->total_tx_bytes = 0;
1158                 adapter->total_tx_packets = 0;
1159                 adapter->total_rx_bytes = 0;
1160                 adapter->total_rx_packets = 0;
1161                 __napi_schedule(&adapter->napi);
1162         }
1163
1164         return IRQ_HANDLED;
1165 }
1166
1167 /**
1168  * e1000_intr - Interrupt Handler
1169  * @irq: interrupt number
1170  * @data: pointer to a network interface device structure
1171  **/
1172 static irqreturn_t e1000_intr(int irq, void *data)
1173 {
1174         struct net_device *netdev = data;
1175         struct e1000_adapter *adapter = netdev_priv(netdev);
1176         struct e1000_hw *hw = &adapter->hw;
1177         u32 rctl, icr = er32(ICR);
1178
1179         if (!icr)
1180                 return IRQ_NONE;  /* Not our interrupt */
1181
1182         /*
1183          * IMS will not auto-mask if INT_ASSERTED is not set, and if it is
1184          * not set, then the adapter didn't send an interrupt
1185          */
1186         if (!(icr & E1000_ICR_INT_ASSERTED))
1187                 return IRQ_NONE;
1188
1189         /*
1190          * Interrupt Auto-Mask...upon reading ICR,
1191          * interrupts are masked.  No need for the
1192          * IMC write
1193          */
1194
1195         if (icr & E1000_ICR_LSC) {
1196                 hw->mac.get_link_status = 1;
1197                 /*
1198                  * ICH8 workaround-- Call gig speed drop workaround on cable
1199                  * disconnect (LSC) before accessing any PHY registers
1200                  */
1201                 if ((adapter->flags & FLAG_LSC_GIG_SPEED_DROP) &&
1202                     (!(er32(STATUS) & E1000_STATUS_LU)))
1203                         schedule_work(&adapter->downshift_task);
1204
1205                 /*
1206                  * 80003ES2LAN workaround--
1207                  * For packet buffer work-around on link down event;
1208                  * disable receives here in the ISR and
1209                  * reset adapter in watchdog
1210                  */
1211                 if (netif_carrier_ok(netdev) &&
1212                     (adapter->flags & FLAG_RX_NEEDS_RESTART)) {
1213                         /* disable receives */
1214                         rctl = er32(RCTL);
1215                         ew32(RCTL, rctl & ~E1000_RCTL_EN);
1216                         adapter->flags |= FLAG_RX_RESTART_NOW;
1217                 }
1218                 /* guard against interrupt when we're going down */
1219                 if (!test_bit(__E1000_DOWN, &adapter->state))
1220                         mod_timer(&adapter->watchdog_timer, jiffies + 1);
1221         }
1222
1223         if (napi_schedule_prep(&adapter->napi)) {
1224                 adapter->total_tx_bytes = 0;
1225                 adapter->total_tx_packets = 0;
1226                 adapter->total_rx_bytes = 0;
1227                 adapter->total_rx_packets = 0;
1228                 __napi_schedule(&adapter->napi);
1229         }
1230
1231         return IRQ_HANDLED;
1232 }
1233
1234 static irqreturn_t e1000_msix_other(int irq, void *data)
1235 {
1236         struct net_device *netdev = data;
1237         struct e1000_adapter *adapter = netdev_priv(netdev);
1238         struct e1000_hw *hw = &adapter->hw;
1239         u32 icr = er32(ICR);
1240
1241         if (!(icr & E1000_ICR_INT_ASSERTED)) {
1242                 if (!test_bit(__E1000_DOWN, &adapter->state))
1243                         ew32(IMS, E1000_IMS_OTHER);
1244                 return IRQ_NONE;
1245         }
1246
1247         if (icr & adapter->eiac_mask)
1248                 ew32(ICS, (icr & adapter->eiac_mask));
1249
1250         if (icr & E1000_ICR_OTHER) {
1251                 if (!(icr & E1000_ICR_LSC))
1252                         goto no_link_interrupt;
1253                 hw->mac.get_link_status = 1;
1254                 /* guard against interrupt when we're going down */
1255                 if (!test_bit(__E1000_DOWN, &adapter->state))
1256                         mod_timer(&adapter->watchdog_timer, jiffies + 1);
1257         }
1258
1259 no_link_interrupt:
1260         if (!test_bit(__E1000_DOWN, &adapter->state))
1261                 ew32(IMS, E1000_IMS_LSC | E1000_IMS_OTHER);
1262
1263         return IRQ_HANDLED;
1264 }
1265
1266
1267 static irqreturn_t e1000_intr_msix_tx(int irq, void *data)
1268 {
1269         struct net_device *netdev = data;
1270         struct e1000_adapter *adapter = netdev_priv(netdev);
1271         struct e1000_hw *hw = &adapter->hw;
1272         struct e1000_ring *tx_ring = adapter->tx_ring;
1273
1274
1275         adapter->total_tx_bytes = 0;
1276         adapter->total_tx_packets = 0;
1277
1278         if (!e1000_clean_tx_irq(adapter))
1279                 /* Ring was not completely cleaned, so fire another interrupt */
1280                 ew32(ICS, tx_ring->ims_val);
1281
1282         return IRQ_HANDLED;
1283 }
1284
1285 static irqreturn_t e1000_intr_msix_rx(int irq, void *data)
1286 {
1287         struct net_device *netdev = data;
1288         struct e1000_adapter *adapter = netdev_priv(netdev);
1289
1290         /* Write the ITR value calculated at the end of the
1291          * previous interrupt.
1292          */
1293         if (adapter->rx_ring->set_itr) {
1294                 writel(1000000000 / (adapter->rx_ring->itr_val * 256),
1295                        adapter->hw.hw_addr + adapter->rx_ring->itr_register);
1296                 adapter->rx_ring->set_itr = 0;
1297         }
1298
1299         if (napi_schedule_prep(&adapter->napi)) {
1300                 adapter->total_rx_bytes = 0;
1301                 adapter->total_rx_packets = 0;
1302                 __napi_schedule(&adapter->napi);
1303         }
1304         return IRQ_HANDLED;
1305 }
1306
1307 /**
1308  * e1000_configure_msix - Configure MSI-X hardware
1309  *
1310  * e1000_configure_msix sets up the hardware to properly
1311  * generate MSI-X interrupts.
1312  **/
1313 static void e1000_configure_msix(struct e1000_adapter *adapter)
1314 {
1315         struct e1000_hw *hw = &adapter->hw;
1316         struct e1000_ring *rx_ring = adapter->rx_ring;
1317         struct e1000_ring *tx_ring = adapter->tx_ring;
1318         int vector = 0;
1319         u32 ctrl_ext, ivar = 0;
1320
1321         adapter->eiac_mask = 0;
1322
1323         /* Workaround issue with spurious interrupts on 82574 in MSI-X mode */
1324         if (hw->mac.type == e1000_82574) {
1325                 u32 rfctl = er32(RFCTL);
1326                 rfctl |= E1000_RFCTL_ACK_DIS;
1327                 ew32(RFCTL, rfctl);
1328         }
1329
1330 #define E1000_IVAR_INT_ALLOC_VALID      0x8
1331         /* Configure Rx vector */
1332         rx_ring->ims_val = E1000_IMS_RXQ0;
1333         adapter->eiac_mask |= rx_ring->ims_val;
1334         if (rx_ring->itr_val)
1335                 writel(1000000000 / (rx_ring->itr_val * 256),
1336                        hw->hw_addr + rx_ring->itr_register);
1337         else
1338                 writel(1, hw->hw_addr + rx_ring->itr_register);
1339         ivar = E1000_IVAR_INT_ALLOC_VALID | vector;
1340
1341         /* Configure Tx vector */
1342         tx_ring->ims_val = E1000_IMS_TXQ0;
1343         vector++;
1344         if (tx_ring->itr_val)
1345                 writel(1000000000 / (tx_ring->itr_val * 256),
1346                        hw->hw_addr + tx_ring->itr_register);
1347         else
1348                 writel(1, hw->hw_addr + tx_ring->itr_register);
1349         adapter->eiac_mask |= tx_ring->ims_val;
1350         ivar |= ((E1000_IVAR_INT_ALLOC_VALID | vector) << 8);
1351
1352         /* set vector for Other Causes, e.g. link changes */
1353         vector++;
1354         ivar |= ((E1000_IVAR_INT_ALLOC_VALID | vector) << 16);
1355         if (rx_ring->itr_val)
1356                 writel(1000000000 / (rx_ring->itr_val * 256),
1357                        hw->hw_addr + E1000_EITR_82574(vector));
1358         else
1359                 writel(1, hw->hw_addr + E1000_EITR_82574(vector));
1360
1361         /* Cause Tx interrupts on every write back */
1362         ivar |= (1 << 31);
1363
1364         ew32(IVAR, ivar);
1365
1366         /* enable MSI-X PBA support */
1367         ctrl_ext = er32(CTRL_EXT);
1368         ctrl_ext |= E1000_CTRL_EXT_PBA_CLR;
1369
1370         /* Auto-Mask Other interrupts upon ICR read */
1371 #define E1000_EIAC_MASK_82574   0x01F00000
1372         ew32(IAM, ~E1000_EIAC_MASK_82574 | E1000_IMS_OTHER);
1373         ctrl_ext |= E1000_CTRL_EXT_EIAME;
1374         ew32(CTRL_EXT, ctrl_ext);
1375         e1e_flush();
1376 }
1377
1378 void e1000e_reset_interrupt_capability(struct e1000_adapter *adapter)
1379 {
1380         if (adapter->msix_entries) {
1381                 pci_disable_msix(adapter->pdev);
1382                 kfree(adapter->msix_entries);
1383                 adapter->msix_entries = NULL;
1384         } else if (adapter->flags & FLAG_MSI_ENABLED) {
1385                 pci_disable_msi(adapter->pdev);
1386                 adapter->flags &= ~FLAG_MSI_ENABLED;
1387         }
1388
1389         return;
1390 }
1391
1392 /**
1393  * e1000e_set_interrupt_capability - set MSI or MSI-X if supported
1394  *
1395  * Attempt to configure interrupts using the best available
1396  * capabilities of the hardware and kernel.
1397  **/
1398 void e1000e_set_interrupt_capability(struct e1000_adapter *adapter)
1399 {
1400         int err;
1401         int numvecs, i;
1402
1403
1404         switch (adapter->int_mode) {
1405         case E1000E_INT_MODE_MSIX:
1406                 if (adapter->flags & FLAG_HAS_MSIX) {
1407                         numvecs = 3; /* RxQ0, TxQ0 and other */
1408                         adapter->msix_entries = kcalloc(numvecs,
1409                                                       sizeof(struct msix_entry),
1410                                                       GFP_KERNEL);
1411                         if (adapter->msix_entries) {
1412                                 for (i = 0; i < numvecs; i++)
1413                                         adapter->msix_entries[i].entry = i;
1414
1415                                 err = pci_enable_msix(adapter->pdev,
1416                                                       adapter->msix_entries,
1417                                                       numvecs);
1418                                 if (err == 0)
1419                                         return;
1420                         }
1421                         /* MSI-X failed, so fall through and try MSI */
1422                         e_err("Failed to initialize MSI-X interrupts.  "
1423                               "Falling back to MSI interrupts.\n");
1424                         e1000e_reset_interrupt_capability(adapter);
1425                 }
1426                 adapter->int_mode = E1000E_INT_MODE_MSI;
1427                 /* Fall through */
1428         case E1000E_INT_MODE_MSI:
1429                 if (!pci_enable_msi(adapter->pdev)) {
1430                         adapter->flags |= FLAG_MSI_ENABLED;
1431                 } else {
1432                         adapter->int_mode = E1000E_INT_MODE_LEGACY;
1433                         e_err("Failed to initialize MSI interrupts.  Falling "
1434                               "back to legacy interrupts.\n");
1435                 }
1436                 /* Fall through */
1437         case E1000E_INT_MODE_LEGACY:
1438                 /* Don't do anything; this is the system default */
1439                 break;
1440         }
1441
1442         return;
1443 }
1444
1445 /**
1446  * e1000_request_msix - Initialize MSI-X interrupts
1447  *
1448  * e1000_request_msix allocates MSI-X vectors and requests interrupts from the
1449  * kernel.
1450  **/
1451 static int e1000_request_msix(struct e1000_adapter *adapter)
1452 {
1453         struct net_device *netdev = adapter->netdev;
1454         int err = 0, vector = 0;
1455
1456         if (strlen(netdev->name) < (IFNAMSIZ - 5))
1457                 sprintf(adapter->rx_ring->name, "%s-rx-0", netdev->name);
1458         else
1459                 memcpy(adapter->rx_ring->name, netdev->name, IFNAMSIZ);
1460         err = request_irq(adapter->msix_entries[vector].vector,
1461                           e1000_intr_msix_rx, 0, adapter->rx_ring->name,
1462                           netdev);
1463         if (err)
1464                 goto out;
1465         adapter->rx_ring->itr_register = E1000_EITR_82574(vector);
1466         adapter->rx_ring->itr_val = adapter->itr;
1467         vector++;
1468
1469         if (strlen(netdev->name) < (IFNAMSIZ - 5))
1470                 sprintf(adapter->tx_ring->name, "%s-tx-0", netdev->name);
1471         else
1472                 memcpy(adapter->tx_ring->name, netdev->name, IFNAMSIZ);
1473         err = request_irq(adapter->msix_entries[vector].vector,
1474                           e1000_intr_msix_tx, 0, adapter->tx_ring->name,
1475                           netdev);
1476         if (err)
1477                 goto out;
1478         adapter->tx_ring->itr_register = E1000_EITR_82574(vector);
1479         adapter->tx_ring->itr_val = adapter->itr;
1480         vector++;
1481
1482         err = request_irq(adapter->msix_entries[vector].vector,
1483                           e1000_msix_other, 0, netdev->name, netdev);
1484         if (err)
1485                 goto out;
1486
1487         e1000_configure_msix(adapter);
1488         return 0;
1489 out:
1490         return err;
1491 }
1492
1493 /**
1494  * e1000_request_irq - initialize interrupts
1495  *
1496  * Attempts to configure interrupts using the best available
1497  * capabilities of the hardware and kernel.
1498  **/
1499 static int e1000_request_irq(struct e1000_adapter *adapter)
1500 {
1501         struct net_device *netdev = adapter->netdev;
1502         int err;
1503
1504         if (adapter->msix_entries) {
1505                 err = e1000_request_msix(adapter);
1506                 if (!err)
1507                         return err;
1508                 /* fall back to MSI */
1509                 e1000e_reset_interrupt_capability(adapter);
1510                 adapter->int_mode = E1000E_INT_MODE_MSI;
1511                 e1000e_set_interrupt_capability(adapter);
1512         }
1513         if (adapter->flags & FLAG_MSI_ENABLED) {
1514                 err = request_irq(adapter->pdev->irq, e1000_intr_msi, 0,
1515                                   netdev->name, netdev);
1516                 if (!err)
1517                         return err;
1518
1519                 /* fall back to legacy interrupt */
1520                 e1000e_reset_interrupt_capability(adapter);
1521                 adapter->int_mode = E1000E_INT_MODE_LEGACY;
1522         }
1523
1524         err = request_irq(adapter->pdev->irq, e1000_intr, IRQF_SHARED,
1525                           netdev->name, netdev);
1526         if (err)
1527                 e_err("Unable to allocate interrupt, Error: %d\n", err);
1528
1529         return err;
1530 }
1531
1532 static void e1000_free_irq(struct e1000_adapter *adapter)
1533 {
1534         struct net_device *netdev = adapter->netdev;
1535
1536         if (adapter->msix_entries) {
1537                 int vector = 0;
1538
1539                 free_irq(adapter->msix_entries[vector].vector, netdev);
1540                 vector++;
1541
1542                 free_irq(adapter->msix_entries[vector].vector, netdev);
1543                 vector++;
1544
1545                 /* Other Causes interrupt vector */
1546                 free_irq(adapter->msix_entries[vector].vector, netdev);
1547                 return;
1548         }
1549
1550         free_irq(adapter->pdev->irq, netdev);
1551 }
1552
1553 /**
1554  * e1000_irq_disable - Mask off interrupt generation on the NIC
1555  **/
1556 static void e1000_irq_disable(struct e1000_adapter *adapter)
1557 {
1558         struct e1000_hw *hw = &adapter->hw;
1559
1560         ew32(IMC, ~0);
1561         if (adapter->msix_entries)
1562                 ew32(EIAC_82574, 0);
1563         e1e_flush();
1564         synchronize_irq(adapter->pdev->irq);
1565 }
1566
1567 /**
1568  * e1000_irq_enable - Enable default interrupt generation settings
1569  **/
1570 static void e1000_irq_enable(struct e1000_adapter *adapter)
1571 {
1572         struct e1000_hw *hw = &adapter->hw;
1573
1574         if (adapter->msix_entries) {
1575                 ew32(EIAC_82574, adapter->eiac_mask & E1000_EIAC_MASK_82574);
1576                 ew32(IMS, adapter->eiac_mask | E1000_IMS_OTHER | E1000_IMS_LSC);
1577         } else {
1578                 ew32(IMS, IMS_ENABLE_MASK);
1579         }
1580         e1e_flush();
1581 }
1582
1583 /**
1584  * e1000_get_hw_control - get control of the h/w from f/w
1585  * @adapter: address of board private structure
1586  *
1587  * e1000_get_hw_control sets {CTRL_EXT|SWSM}:DRV_LOAD bit.
1588  * For ASF and Pass Through versions of f/w this means that
1589  * the driver is loaded. For AMT version (only with 82573)
1590  * of the f/w this means that the network i/f is open.
1591  **/
1592 static void e1000_get_hw_control(struct e1000_adapter *adapter)
1593 {
1594         struct e1000_hw *hw = &adapter->hw;
1595         u32 ctrl_ext;
1596         u32 swsm;
1597
1598         /* Let firmware know the driver has taken over */
1599         if (adapter->flags & FLAG_HAS_SWSM_ON_LOAD) {
1600                 swsm = er32(SWSM);
1601                 ew32(SWSM, swsm | E1000_SWSM_DRV_LOAD);
1602         } else if (adapter->flags & FLAG_HAS_CTRLEXT_ON_LOAD) {
1603                 ctrl_ext = er32(CTRL_EXT);
1604                 ew32(CTRL_EXT, ctrl_ext | E1000_CTRL_EXT_DRV_LOAD);
1605         }
1606 }
1607
1608 /**
1609  * e1000_release_hw_control - release control of the h/w to f/w
1610  * @adapter: address of board private structure
1611  *
1612  * e1000_release_hw_control resets {CTRL_EXT|SWSM}:DRV_LOAD bit.
1613  * For ASF and Pass Through versions of f/w this means that the
1614  * driver is no longer loaded. For AMT version (only with 82573) i
1615  * of the f/w this means that the network i/f is closed.
1616  *
1617  **/
1618 static void e1000_release_hw_control(struct e1000_adapter *adapter)
1619 {
1620         struct e1000_hw *hw = &adapter->hw;
1621         u32 ctrl_ext;
1622         u32 swsm;
1623
1624         /* Let firmware taken over control of h/w */
1625         if (adapter->flags & FLAG_HAS_SWSM_ON_LOAD) {
1626                 swsm = er32(SWSM);
1627                 ew32(SWSM, swsm & ~E1000_SWSM_DRV_LOAD);
1628         } else if (adapter->flags & FLAG_HAS_CTRLEXT_ON_LOAD) {
1629                 ctrl_ext = er32(CTRL_EXT);
1630                 ew32(CTRL_EXT, ctrl_ext & ~E1000_CTRL_EXT_DRV_LOAD);
1631         }
1632 }
1633
1634 /**
1635  * @e1000_alloc_ring - allocate memory for a ring structure
1636  **/
1637 static int e1000_alloc_ring_dma(struct e1000_adapter *adapter,
1638                                 struct e1000_ring *ring)
1639 {
1640         struct pci_dev *pdev = adapter->pdev;
1641
1642         ring->desc = dma_alloc_coherent(&pdev->dev, ring->size, &ring->dma,
1643                                         GFP_KERNEL);
1644         if (!ring->desc)
1645                 return -ENOMEM;
1646
1647         return 0;
1648 }
1649
1650 /**
1651  * e1000e_setup_tx_resources - allocate Tx resources (Descriptors)
1652  * @adapter: board private structure
1653  *
1654  * Return 0 on success, negative on failure
1655  **/
1656 int e1000e_setup_tx_resources(struct e1000_adapter *adapter)
1657 {
1658         struct e1000_ring *tx_ring = adapter->tx_ring;
1659         int err = -ENOMEM, size;
1660
1661         size = sizeof(struct e1000_buffer) * tx_ring->count;
1662         tx_ring->buffer_info = vmalloc(size);
1663         if (!tx_ring->buffer_info)
1664                 goto err;
1665         memset(tx_ring->buffer_info, 0, size);
1666
1667         /* round up to nearest 4K */
1668         tx_ring->size = tx_ring->count * sizeof(struct e1000_tx_desc);
1669         tx_ring->size = ALIGN(tx_ring->size, 4096);
1670
1671         err = e1000_alloc_ring_dma(adapter, tx_ring);
1672         if (err)
1673                 goto err;
1674
1675         tx_ring->next_to_use = 0;
1676         tx_ring->next_to_clean = 0;
1677
1678         return 0;
1679 err:
1680         vfree(tx_ring->buffer_info);
1681         e_err("Unable to allocate memory for the transmit descriptor ring\n");
1682         return err;
1683 }
1684
1685 /**
1686  * e1000e_setup_rx_resources - allocate Rx resources (Descriptors)
1687  * @adapter: board private structure
1688  *
1689  * Returns 0 on success, negative on failure
1690  **/
1691 int e1000e_setup_rx_resources(struct e1000_adapter *adapter)
1692 {
1693         struct e1000_ring *rx_ring = adapter->rx_ring;
1694         struct e1000_buffer *buffer_info;
1695         int i, size, desc_len, err = -ENOMEM;
1696
1697         size = sizeof(struct e1000_buffer) * rx_ring->count;
1698         rx_ring->buffer_info = vmalloc(size);
1699         if (!rx_ring->buffer_info)
1700                 goto err;
1701         memset(rx_ring->buffer_info, 0, size);
1702
1703         for (i = 0; i < rx_ring->count; i++) {
1704                 buffer_info = &rx_ring->buffer_info[i];
1705                 buffer_info->ps_pages = kcalloc(PS_PAGE_BUFFERS,
1706                                                 sizeof(struct e1000_ps_page),
1707                                                 GFP_KERNEL);
1708                 if (!buffer_info->ps_pages)
1709                         goto err_pages;
1710         }
1711
1712         desc_len = sizeof(union e1000_rx_desc_packet_split);
1713
1714         /* Round up to nearest 4K */
1715         rx_ring->size = rx_ring->count * desc_len;
1716         rx_ring->size = ALIGN(rx_ring->size, 4096);
1717
1718         err = e1000_alloc_ring_dma(adapter, rx_ring);
1719         if (err)
1720                 goto err_pages;
1721
1722         rx_ring->next_to_clean = 0;
1723         rx_ring->next_to_use = 0;
1724         rx_ring->rx_skb_top = NULL;
1725
1726         return 0;
1727
1728 err_pages:
1729         for (i = 0; i < rx_ring->count; i++) {
1730                 buffer_info = &rx_ring->buffer_info[i];
1731                 kfree(buffer_info->ps_pages);
1732         }
1733 err:
1734         vfree(rx_ring->buffer_info);
1735         e_err("Unable to allocate memory for the transmit descriptor ring\n");
1736         return err;
1737 }
1738
1739 /**
1740  * e1000_clean_tx_ring - Free Tx Buffers
1741  * @adapter: board private structure
1742  **/
1743 static void e1000_clean_tx_ring(struct e1000_adapter *adapter)
1744 {
1745         struct e1000_ring *tx_ring = adapter->tx_ring;
1746         struct e1000_buffer *buffer_info;
1747         unsigned long size;
1748         unsigned int i;
1749
1750         for (i = 0; i < tx_ring->count; i++) {
1751                 buffer_info = &tx_ring->buffer_info[i];
1752                 e1000_put_txbuf(adapter, buffer_info);
1753         }
1754
1755         size = sizeof(struct e1000_buffer) * tx_ring->count;
1756         memset(tx_ring->buffer_info, 0, size);
1757
1758         memset(tx_ring->desc, 0, tx_ring->size);
1759
1760         tx_ring->next_to_use = 0;
1761         tx_ring->next_to_clean = 0;
1762
1763         writel(0, adapter->hw.hw_addr + tx_ring->head);
1764         writel(0, adapter->hw.hw_addr + tx_ring->tail);
1765 }
1766
1767 /**
1768  * e1000e_free_tx_resources - Free Tx Resources per Queue
1769  * @adapter: board private structure
1770  *
1771  * Free all transmit software resources
1772  **/
1773 void e1000e_free_tx_resources(struct e1000_adapter *adapter)
1774 {
1775         struct pci_dev *pdev = adapter->pdev;
1776         struct e1000_ring *tx_ring = adapter->tx_ring;
1777
1778         e1000_clean_tx_ring(adapter);
1779
1780         vfree(tx_ring->buffer_info);
1781         tx_ring->buffer_info = NULL;
1782
1783         dma_free_coherent(&pdev->dev, tx_ring->size, tx_ring->desc,
1784                           tx_ring->dma);
1785         tx_ring->desc = NULL;
1786 }
1787
1788 /**
1789  * e1000e_free_rx_resources - Free Rx Resources
1790  * @adapter: board private structure
1791  *
1792  * Free all receive software resources
1793  **/
1794
1795 void e1000e_free_rx_resources(struct e1000_adapter *adapter)
1796 {
1797         struct pci_dev *pdev = adapter->pdev;
1798         struct e1000_ring *rx_ring = adapter->rx_ring;
1799         int i;
1800
1801         e1000_clean_rx_ring(adapter);
1802
1803         for (i = 0; i < rx_ring->count; i++) {
1804                 kfree(rx_ring->buffer_info[i].ps_pages);
1805         }
1806
1807         vfree(rx_ring->buffer_info);
1808         rx_ring->buffer_info = NULL;
1809
1810         dma_free_coherent(&pdev->dev, rx_ring->size, rx_ring->desc,
1811                           rx_ring->dma);
1812         rx_ring->desc = NULL;
1813 }
1814
1815 /**
1816  * e1000_update_itr - update the dynamic ITR value based on statistics
1817  * @adapter: pointer to adapter
1818  * @itr_setting: current adapter->itr
1819  * @packets: the number of packets during this measurement interval
1820  * @bytes: the number of bytes during this measurement interval
1821  *
1822  *      Stores a new ITR value based on packets and byte
1823  *      counts during the last interrupt.  The advantage of per interrupt
1824  *      computation is faster updates and more accurate ITR for the current
1825  *      traffic pattern.  Constants in this function were computed
1826  *      based on theoretical maximum wire speed and thresholds were set based
1827  *      on testing data as well as attempting to minimize response time
1828  *      while increasing bulk throughput.  This functionality is controlled
1829  *      by the InterruptThrottleRate module parameter.
1830  **/
1831 static unsigned int e1000_update_itr(struct e1000_adapter *adapter,
1832                                      u16 itr_setting, int packets,
1833                                      int bytes)
1834 {
1835         unsigned int retval = itr_setting;
1836
1837         if (packets == 0)
1838                 goto update_itr_done;
1839
1840         switch (itr_setting) {
1841         case lowest_latency:
1842                 /* handle TSO and jumbo frames */
1843                 if (bytes/packets > 8000)
1844                         retval = bulk_latency;
1845                 else if ((packets < 5) && (bytes > 512)) {
1846                         retval = low_latency;
1847                 }
1848                 break;
1849         case low_latency:  /* 50 usec aka 20000 ints/s */
1850                 if (bytes > 10000) {
1851                         /* this if handles the TSO accounting */
1852                         if (bytes/packets > 8000) {
1853                                 retval = bulk_latency;
1854                         } else if ((packets < 10) || ((bytes/packets) > 1200)) {
1855                                 retval = bulk_latency;
1856                         } else if ((packets > 35)) {
1857                                 retval = lowest_latency;
1858                         }
1859                 } else if (bytes/packets > 2000) {
1860                         retval = bulk_latency;
1861                 } else if (packets <= 2 && bytes < 512) {
1862                         retval = lowest_latency;
1863                 }
1864                 break;
1865         case bulk_latency: /* 250 usec aka 4000 ints/s */
1866                 if (bytes > 25000) {
1867                         if (packets > 35) {
1868                                 retval = low_latency;
1869                         }
1870                 } else if (bytes < 6000) {
1871                         retval = low_latency;
1872                 }
1873                 break;
1874         }
1875
1876 update_itr_done:
1877         return retval;
1878 }
1879
1880 static void e1000_set_itr(struct e1000_adapter *adapter)
1881 {
1882         struct e1000_hw *hw = &adapter->hw;
1883         u16 current_itr;
1884         u32 new_itr = adapter->itr;
1885
1886         /* for non-gigabit speeds, just fix the interrupt rate at 4000 */
1887         if (adapter->link_speed != SPEED_1000) {
1888                 current_itr = 0;
1889                 new_itr = 4000;
1890                 goto set_itr_now;
1891         }
1892
1893         adapter->tx_itr = e1000_update_itr(adapter,
1894                                     adapter->tx_itr,
1895                                     adapter->total_tx_packets,
1896                                     adapter->total_tx_bytes);
1897         /* conservative mode (itr 3) eliminates the lowest_latency setting */
1898         if (adapter->itr_setting == 3 && adapter->tx_itr == lowest_latency)
1899                 adapter->tx_itr = low_latency;
1900
1901         adapter->rx_itr = e1000_update_itr(adapter,
1902                                     adapter->rx_itr,
1903                                     adapter->total_rx_packets,
1904                                     adapter->total_rx_bytes);
1905         /* conservative mode (itr 3) eliminates the lowest_latency setting */
1906         if (adapter->itr_setting == 3 && adapter->rx_itr == lowest_latency)
1907                 adapter->rx_itr = low_latency;
1908
1909         current_itr = max(adapter->rx_itr, adapter->tx_itr);
1910
1911         switch (current_itr) {
1912         /* counts and packets in update_itr are dependent on these numbers */
1913         case lowest_latency:
1914                 new_itr = 70000;
1915                 break;
1916         case low_latency:
1917                 new_itr = 20000; /* aka hwitr = ~200 */
1918                 break;
1919         case bulk_latency:
1920                 new_itr = 4000;
1921                 break;
1922         default:
1923                 break;
1924         }
1925
1926 set_itr_now:
1927         if (new_itr != adapter->itr) {
1928                 /*
1929                  * this attempts to bias the interrupt rate towards Bulk
1930                  * by adding intermediate steps when interrupt rate is
1931                  * increasing
1932                  */
1933                 new_itr = new_itr > adapter->itr ?
1934                              min(adapter->itr + (new_itr >> 2), new_itr) :
1935                              new_itr;
1936                 adapter->itr = new_itr;
1937                 adapter->rx_ring->itr_val = new_itr;
1938                 if (adapter->msix_entries)
1939                         adapter->rx_ring->set_itr = 1;
1940                 else
1941                         ew32(ITR, 1000000000 / (new_itr * 256));
1942         }
1943 }
1944
1945 /**
1946  * e1000_alloc_queues - Allocate memory for all rings
1947  * @adapter: board private structure to initialize
1948  **/
1949 static int __devinit e1000_alloc_queues(struct e1000_adapter *adapter)
1950 {
1951         adapter->tx_ring = kzalloc(sizeof(struct e1000_ring), GFP_KERNEL);
1952         if (!adapter->tx_ring)
1953                 goto err;
1954
1955         adapter->rx_ring = kzalloc(sizeof(struct e1000_ring), GFP_KERNEL);
1956         if (!adapter->rx_ring)
1957                 goto err;
1958
1959         return 0;
1960 err:
1961         e_err("Unable to allocate memory for queues\n");
1962         kfree(adapter->rx_ring);
1963         kfree(adapter->tx_ring);
1964         return -ENOMEM;
1965 }
1966
1967 /**
1968  * e1000_clean - NAPI Rx polling callback
1969  * @napi: struct associated with this polling callback
1970  * @budget: amount of packets driver is allowed to process this poll
1971  **/
1972 static int e1000_clean(struct napi_struct *napi, int budget)
1973 {
1974         struct e1000_adapter *adapter = container_of(napi, struct e1000_adapter, napi);
1975         struct e1000_hw *hw = &adapter->hw;
1976         struct net_device *poll_dev = adapter->netdev;
1977         int tx_cleaned = 1, work_done = 0;
1978
1979         adapter = netdev_priv(poll_dev);
1980
1981         if (adapter->msix_entries &&
1982             !(adapter->rx_ring->ims_val & adapter->tx_ring->ims_val))
1983                 goto clean_rx;
1984
1985         tx_cleaned = e1000_clean_tx_irq(adapter);
1986
1987 clean_rx:
1988         adapter->clean_rx(adapter, &work_done, budget);
1989
1990         if (!tx_cleaned)
1991                 work_done = budget;
1992
1993         /* If budget not fully consumed, exit the polling mode */
1994         if (work_done < budget) {
1995                 if (adapter->itr_setting & 3)
1996                         e1000_set_itr(adapter);
1997                 napi_complete(napi);
1998                 if (!test_bit(__E1000_DOWN, &adapter->state)) {
1999                         if (adapter->msix_entries)
2000                                 ew32(IMS, adapter->rx_ring->ims_val);
2001                         else
2002                                 e1000_irq_enable(adapter);
2003                 }
2004         }
2005
2006         return work_done;
2007 }
2008
2009 static void e1000_vlan_rx_add_vid(struct net_device *netdev, u16 vid)
2010 {
2011         struct e1000_adapter *adapter = netdev_priv(netdev);
2012         struct e1000_hw *hw = &adapter->hw;
2013         u32 vfta, index;
2014
2015         /* don't update vlan cookie if already programmed */
2016         if ((adapter->hw.mng_cookie.status &
2017              E1000_MNG_DHCP_COOKIE_STATUS_VLAN) &&
2018             (vid == adapter->mng_vlan_id))
2019                 return;
2020         /* add VID to filter table */
2021         index = (vid >> 5) & 0x7F;
2022         vfta = E1000_READ_REG_ARRAY(hw, E1000_VFTA, index);
2023         vfta |= (1 << (vid & 0x1F));
2024         e1000e_write_vfta(hw, index, vfta);
2025 }
2026
2027 static void e1000_vlan_rx_kill_vid(struct net_device *netdev, u16 vid)
2028 {
2029         struct e1000_adapter *adapter = netdev_priv(netdev);
2030         struct e1000_hw *hw = &adapter->hw;
2031         u32 vfta, index;
2032
2033         if (!test_bit(__E1000_DOWN, &adapter->state))
2034                 e1000_irq_disable(adapter);
2035         vlan_group_set_device(adapter->vlgrp, vid, NULL);
2036
2037         if (!test_bit(__E1000_DOWN, &adapter->state))
2038                 e1000_irq_enable(adapter);
2039
2040         if ((adapter->hw.mng_cookie.status &
2041              E1000_MNG_DHCP_COOKIE_STATUS_VLAN) &&
2042             (vid == adapter->mng_vlan_id)) {
2043                 /* release control to f/w */
2044                 e1000_release_hw_control(adapter);
2045                 return;
2046         }
2047
2048         /* remove VID from filter table */
2049         index = (vid >> 5) & 0x7F;
2050         vfta = E1000_READ_REG_ARRAY(hw, E1000_VFTA, index);
2051         vfta &= ~(1 << (vid & 0x1F));
2052         e1000e_write_vfta(hw, index, vfta);
2053 }
2054
2055 static void e1000_update_mng_vlan(struct e1000_adapter *adapter)
2056 {
2057         struct net_device *netdev = adapter->netdev;
2058         u16 vid = adapter->hw.mng_cookie.vlan_id;
2059         u16 old_vid = adapter->mng_vlan_id;
2060
2061         if (!adapter->vlgrp)
2062                 return;
2063
2064         if (!vlan_group_get_device(adapter->vlgrp, vid)) {
2065                 adapter->mng_vlan_id = E1000_MNG_VLAN_NONE;
2066                 if (adapter->hw.mng_cookie.status &
2067                         E1000_MNG_DHCP_COOKIE_STATUS_VLAN) {
2068                         e1000_vlan_rx_add_vid(netdev, vid);
2069                         adapter->mng_vlan_id = vid;
2070                 }
2071
2072                 if ((old_vid != (u16)E1000_MNG_VLAN_NONE) &&
2073                                 (vid != old_vid) &&
2074                     !vlan_group_get_device(adapter->vlgrp, old_vid))
2075                         e1000_vlan_rx_kill_vid(netdev, old_vid);
2076         } else {
2077                 adapter->mng_vlan_id = vid;
2078         }
2079 }
2080
2081
2082 static void e1000_vlan_rx_register(struct net_device *netdev,
2083                                    struct vlan_group *grp)
2084 {
2085         struct e1000_adapter *adapter = netdev_priv(netdev);
2086         struct e1000_hw *hw = &adapter->hw;
2087         u32 ctrl, rctl;
2088
2089         if (!test_bit(__E1000_DOWN, &adapter->state))
2090                 e1000_irq_disable(adapter);
2091         adapter->vlgrp = grp;
2092
2093         if (grp) {
2094                 /* enable VLAN tag insert/strip */
2095                 ctrl = er32(CTRL);
2096                 ctrl |= E1000_CTRL_VME;
2097                 ew32(CTRL, ctrl);
2098
2099                 if (adapter->flags & FLAG_HAS_HW_VLAN_FILTER) {
2100                         /* enable VLAN receive filtering */
2101                         rctl = er32(RCTL);
2102                         rctl &= ~E1000_RCTL_CFIEN;
2103                         ew32(RCTL, rctl);
2104                         e1000_update_mng_vlan(adapter);
2105                 }
2106         } else {
2107                 /* disable VLAN tag insert/strip */
2108                 ctrl = er32(CTRL);
2109                 ctrl &= ~E1000_CTRL_VME;
2110                 ew32(CTRL, ctrl);
2111
2112                 if (adapter->flags & FLAG_HAS_HW_VLAN_FILTER) {
2113                         if (adapter->mng_vlan_id !=
2114                             (u16)E1000_MNG_VLAN_NONE) {
2115                                 e1000_vlan_rx_kill_vid(netdev,
2116                                                        adapter->mng_vlan_id);
2117                                 adapter->mng_vlan_id = E1000_MNG_VLAN_NONE;
2118                         }
2119                 }
2120         }
2121
2122         if (!test_bit(__E1000_DOWN, &adapter->state))
2123                 e1000_irq_enable(adapter);
2124 }
2125
2126 static void e1000_restore_vlan(struct e1000_adapter *adapter)
2127 {
2128         u16 vid;
2129
2130         e1000_vlan_rx_register(adapter->netdev, adapter->vlgrp);
2131
2132         if (!adapter->vlgrp)
2133                 return;
2134
2135         for (vid = 0; vid < VLAN_GROUP_ARRAY_LEN; vid++) {
2136                 if (!vlan_group_get_device(adapter->vlgrp, vid))
2137                         continue;
2138                 e1000_vlan_rx_add_vid(adapter->netdev, vid);
2139         }
2140 }
2141
2142 static void e1000_init_manageability(struct e1000_adapter *adapter)
2143 {
2144         struct e1000_hw *hw = &adapter->hw;
2145         u32 manc, manc2h;
2146
2147         if (!(adapter->flags & FLAG_MNG_PT_ENABLED))
2148                 return;
2149
2150         manc = er32(MANC);
2151
2152         /*
2153          * enable receiving management packets to the host. this will probably
2154          * generate destination unreachable messages from the host OS, but
2155          * the packets will be handled on SMBUS
2156          */
2157         manc |= E1000_MANC_EN_MNG2HOST;
2158         manc2h = er32(MANC2H);
2159 #define E1000_MNG2HOST_PORT_623 (1 << 5)
2160 #define E1000_MNG2HOST_PORT_664 (1 << 6)
2161         manc2h |= E1000_MNG2HOST_PORT_623;
2162         manc2h |= E1000_MNG2HOST_PORT_664;
2163         ew32(MANC2H, manc2h);
2164         ew32(MANC, manc);
2165 }
2166
2167 /**
2168  * e1000_configure_tx - Configure 8254x Transmit Unit after Reset
2169  * @adapter: board private structure
2170  *
2171  * Configure the Tx unit of the MAC after a reset.
2172  **/
2173 static void e1000_configure_tx(struct e1000_adapter *adapter)
2174 {
2175         struct e1000_hw *hw = &adapter->hw;
2176         struct e1000_ring *tx_ring = adapter->tx_ring;
2177         u64 tdba;
2178         u32 tdlen, tctl, tipg, tarc;
2179         u32 ipgr1, ipgr2;
2180
2181         /* Setup the HW Tx Head and Tail descriptor pointers */
2182         tdba = tx_ring->dma;
2183         tdlen = tx_ring->count * sizeof(struct e1000_tx_desc);
2184         ew32(TDBAL, (tdba & DMA_BIT_MASK(32)));
2185         ew32(TDBAH, (tdba >> 32));
2186         ew32(TDLEN, tdlen);
2187         ew32(TDH, 0);
2188         ew32(TDT, 0);
2189         tx_ring->head = E1000_TDH;
2190         tx_ring->tail = E1000_TDT;
2191
2192         /* Set the default values for the Tx Inter Packet Gap timer */
2193         tipg = DEFAULT_82543_TIPG_IPGT_COPPER;          /*  8  */
2194         ipgr1 = DEFAULT_82543_TIPG_IPGR1;               /*  8  */
2195         ipgr2 = DEFAULT_82543_TIPG_IPGR2;               /*  6  */
2196
2197         if (adapter->flags & FLAG_TIPG_MEDIUM_FOR_80003ESLAN)
2198                 ipgr2 = DEFAULT_80003ES2LAN_TIPG_IPGR2; /*  7  */
2199
2200         tipg |= ipgr1 << E1000_TIPG_IPGR1_SHIFT;
2201         tipg |= ipgr2 << E1000_TIPG_IPGR2_SHIFT;
2202         ew32(TIPG, tipg);
2203
2204         /* Set the Tx Interrupt Delay register */
2205         ew32(TIDV, adapter->tx_int_delay);
2206         /* Tx irq moderation */
2207         ew32(TADV, adapter->tx_abs_int_delay);
2208
2209         /* Program the Transmit Control Register */
2210         tctl = er32(TCTL);
2211         tctl &= ~E1000_TCTL_CT;
2212         tctl |= E1000_TCTL_PSP | E1000_TCTL_RTLC |
2213                 (E1000_COLLISION_THRESHOLD << E1000_CT_SHIFT);
2214
2215         if (adapter->flags & FLAG_TARC_SPEED_MODE_BIT) {
2216                 tarc = er32(TARC(0));
2217                 /*
2218                  * set the speed mode bit, we'll clear it if we're not at
2219                  * gigabit link later
2220                  */
2221 #define SPEED_MODE_BIT (1 << 21)
2222                 tarc |= SPEED_MODE_BIT;
2223                 ew32(TARC(0), tarc);
2224         }
2225
2226         /* errata: program both queues to unweighted RR */
2227         if (adapter->flags & FLAG_TARC_SET_BIT_ZERO) {
2228                 tarc = er32(TARC(0));
2229                 tarc |= 1;
2230                 ew32(TARC(0), tarc);
2231                 tarc = er32(TARC(1));
2232                 tarc |= 1;
2233                 ew32(TARC(1), tarc);
2234         }
2235
2236         /* Setup Transmit Descriptor Settings for eop descriptor */
2237         adapter->txd_cmd = E1000_TXD_CMD_EOP | E1000_TXD_CMD_IFCS;
2238
2239         /* only set IDE if we are delaying interrupts using the timers */
2240         if (adapter->tx_int_delay)
2241                 adapter->txd_cmd |= E1000_TXD_CMD_IDE;
2242
2243         /* enable Report Status bit */
2244         adapter->txd_cmd |= E1000_TXD_CMD_RS;
2245
2246         ew32(TCTL, tctl);
2247
2248         e1000e_config_collision_dist(hw);
2249
2250         adapter->tx_queue_len = adapter->netdev->tx_queue_len;
2251 }
2252
2253 /**
2254  * e1000_setup_rctl - configure the receive control registers
2255  * @adapter: Board private structure
2256  **/
2257 #define PAGE_USE_COUNT(S) (((S) >> PAGE_SHIFT) + \
2258                            (((S) & (PAGE_SIZE - 1)) ? 1 : 0))
2259 static void e1000_setup_rctl(struct e1000_adapter *adapter)
2260 {
2261         struct e1000_hw *hw = &adapter->hw;
2262         u32 rctl, rfctl;
2263         u32 psrctl = 0;
2264         u32 pages = 0;
2265
2266         /* Program MC offset vector base */
2267         rctl = er32(RCTL);
2268         rctl &= ~(3 << E1000_RCTL_MO_SHIFT);
2269         rctl |= E1000_RCTL_EN | E1000_RCTL_BAM |
2270                 E1000_RCTL_LBM_NO | E1000_RCTL_RDMTS_HALF |
2271                 (adapter->hw.mac.mc_filter_type << E1000_RCTL_MO_SHIFT);
2272
2273         /* Do not Store bad packets */
2274         rctl &= ~E1000_RCTL_SBP;
2275
2276         /* Enable Long Packet receive */
2277         if (adapter->netdev->mtu <= ETH_DATA_LEN)
2278                 rctl &= ~E1000_RCTL_LPE;
2279         else
2280                 rctl |= E1000_RCTL_LPE;
2281
2282         /* Some systems expect that the CRC is included in SMBUS traffic. The
2283          * hardware strips the CRC before sending to both SMBUS (BMC) and to
2284          * host memory when this is enabled
2285          */
2286         if (adapter->flags2 & FLAG2_CRC_STRIPPING)
2287                 rctl |= E1000_RCTL_SECRC;
2288
2289         /* Workaround Si errata on 82577 PHY - configure IPG for jumbos */
2290         if ((hw->phy.type == e1000_phy_82577) && (rctl & E1000_RCTL_LPE)) {
2291                 u16 phy_data;
2292
2293                 e1e_rphy(hw, PHY_REG(770, 26), &phy_data);
2294                 phy_data &= 0xfff8;
2295                 phy_data |= (1 << 2);
2296                 e1e_wphy(hw, PHY_REG(770, 26), phy_data);
2297
2298                 e1e_rphy(hw, 22, &phy_data);
2299                 phy_data &= 0x0fff;
2300                 phy_data |= (1 << 14);
2301                 e1e_wphy(hw, 0x10, 0x2823);
2302                 e1e_wphy(hw, 0x11, 0x0003);
2303                 e1e_wphy(hw, 22, phy_data);
2304         }
2305
2306         /* Setup buffer sizes */
2307         rctl &= ~E1000_RCTL_SZ_4096;
2308         rctl |= E1000_RCTL_BSEX;
2309         switch (adapter->rx_buffer_len) {
2310         case 256:
2311                 rctl |= E1000_RCTL_SZ_256;
2312                 rctl &= ~E1000_RCTL_BSEX;
2313                 break;
2314         case 512:
2315                 rctl |= E1000_RCTL_SZ_512;
2316                 rctl &= ~E1000_RCTL_BSEX;
2317                 break;
2318         case 1024:
2319                 rctl |= E1000_RCTL_SZ_1024;
2320                 rctl &= ~E1000_RCTL_BSEX;
2321                 break;
2322         case 2048:
2323         default:
2324                 rctl |= E1000_RCTL_SZ_2048;
2325                 rctl &= ~E1000_RCTL_BSEX;
2326                 break;
2327         case 4096:
2328                 rctl |= E1000_RCTL_SZ_4096;
2329                 break;
2330         case 8192:
2331                 rctl |= E1000_RCTL_SZ_8192;
2332                 break;
2333         case 16384:
2334                 rctl |= E1000_RCTL_SZ_16384;
2335                 break;
2336         }
2337
2338         /*
2339          * 82571 and greater support packet-split where the protocol
2340          * header is placed in skb->data and the packet data is
2341          * placed in pages hanging off of skb_shinfo(skb)->nr_frags.
2342          * In the case of a non-split, skb->data is linearly filled,
2343          * followed by the page buffers.  Therefore, skb->data is
2344          * sized to hold the largest protocol header.
2345          *
2346          * allocations using alloc_page take too long for regular MTU
2347          * so only enable packet split for jumbo frames
2348          *
2349          * Using pages when the page size is greater than 16k wastes
2350          * a lot of memory, since we allocate 3 pages at all times
2351          * per packet.
2352          */
2353         pages = PAGE_USE_COUNT(adapter->netdev->mtu);
2354         if (!(adapter->flags & FLAG_IS_ICH) && (pages <= 3) &&
2355             (PAGE_SIZE <= 16384) && (rctl & E1000_RCTL_LPE))
2356                 adapter->rx_ps_pages = pages;
2357         else
2358                 adapter->rx_ps_pages = 0;
2359
2360         if (adapter->rx_ps_pages) {
2361                 /* Configure extra packet-split registers */
2362                 rfctl = er32(RFCTL);
2363                 rfctl |= E1000_RFCTL_EXTEN;
2364                 /*
2365                  * disable packet split support for IPv6 extension headers,
2366                  * because some malformed IPv6 headers can hang the Rx
2367                  */
2368                 rfctl |= (E1000_RFCTL_IPV6_EX_DIS |
2369                           E1000_RFCTL_NEW_IPV6_EXT_DIS);
2370
2371                 ew32(RFCTL, rfctl);
2372
2373                 /* Enable Packet split descriptors */
2374                 rctl |= E1000_RCTL_DTYP_PS;
2375
2376                 psrctl |= adapter->rx_ps_bsize0 >>
2377                         E1000_PSRCTL_BSIZE0_SHIFT;
2378
2379                 switch (adapter->rx_ps_pages) {
2380                 case 3:
2381                         psrctl |= PAGE_SIZE <<
2382                                 E1000_PSRCTL_BSIZE3_SHIFT;
2383                 case 2:
2384                         psrctl |= PAGE_SIZE <<
2385                                 E1000_PSRCTL_BSIZE2_SHIFT;
2386                 case 1:
2387                         psrctl |= PAGE_SIZE >>
2388                                 E1000_PSRCTL_BSIZE1_SHIFT;
2389                         break;
2390                 }
2391
2392                 ew32(PSRCTL, psrctl);
2393         }
2394
2395         ew32(RCTL, rctl);
2396         /* just started the receive unit, no need to restart */
2397         adapter->flags &= ~FLAG_RX_RESTART_NOW;
2398 }
2399
2400 /**
2401  * e1000_configure_rx - Configure Receive Unit after Reset
2402  * @adapter: board private structure
2403  *
2404  * Configure the Rx unit of the MAC after a reset.
2405  **/
2406 static void e1000_configure_rx(struct e1000_adapter *adapter)
2407 {
2408         struct e1000_hw *hw = &adapter->hw;
2409         struct e1000_ring *rx_ring = adapter->rx_ring;
2410         u64 rdba;
2411         u32 rdlen, rctl, rxcsum, ctrl_ext;
2412
2413         if (adapter->rx_ps_pages) {
2414                 /* this is a 32 byte descriptor */
2415                 rdlen = rx_ring->count *
2416                         sizeof(union e1000_rx_desc_packet_split);
2417                 adapter->clean_rx = e1000_clean_rx_irq_ps;
2418                 adapter->alloc_rx_buf = e1000_alloc_rx_buffers_ps;
2419         } else if (adapter->netdev->mtu > ETH_FRAME_LEN + ETH_FCS_LEN) {
2420                 rdlen = rx_ring->count * sizeof(struct e1000_rx_desc);
2421                 adapter->clean_rx = e1000_clean_jumbo_rx_irq;
2422                 adapter->alloc_rx_buf = e1000_alloc_jumbo_rx_buffers;
2423         } else {
2424                 rdlen = rx_ring->count * sizeof(struct e1000_rx_desc);
2425                 adapter->clean_rx = e1000_clean_rx_irq;
2426                 adapter->alloc_rx_buf = e1000_alloc_rx_buffers;
2427         }
2428
2429         /* disable receives while setting up the descriptors */
2430         rctl = er32(RCTL);
2431         ew32(RCTL, rctl & ~E1000_RCTL_EN);
2432         e1e_flush();
2433         msleep(10);
2434
2435         /* set the Receive Delay Timer Register */
2436         ew32(RDTR, adapter->rx_int_delay);
2437
2438         /* irq moderation */
2439         ew32(RADV, adapter->rx_abs_int_delay);
2440         if (adapter->itr_setting != 0)
2441                 ew32(ITR, 1000000000 / (adapter->itr * 256));
2442
2443         ctrl_ext = er32(CTRL_EXT);
2444         /* Reset delay timers after every interrupt */
2445         ctrl_ext |= E1000_CTRL_EXT_INT_TIMER_CLR;
2446         /* Auto-Mask interrupts upon ICR access */
2447         ctrl_ext |= E1000_CTRL_EXT_IAME;
2448         ew32(IAM, 0xffffffff);
2449         ew32(CTRL_EXT, ctrl_ext);
2450         e1e_flush();
2451
2452         /*
2453          * Setup the HW Rx Head and Tail Descriptor Pointers and
2454          * the Base and Length of the Rx Descriptor Ring
2455          */
2456         rdba = rx_ring->dma;
2457         ew32(RDBAL, (rdba & DMA_BIT_MASK(32)));
2458         ew32(RDBAH, (rdba >> 32));
2459         ew32(RDLEN, rdlen);
2460         ew32(RDH, 0);
2461         ew32(RDT, 0);
2462         rx_ring->head = E1000_RDH;
2463         rx_ring->tail = E1000_RDT;
2464
2465         /* Enable Receive Checksum Offload for TCP and UDP */
2466         rxcsum = er32(RXCSUM);
2467         if (adapter->flags & FLAG_RX_CSUM_ENABLED) {
2468                 rxcsum |= E1000_RXCSUM_TUOFL;
2469
2470                 /*
2471                  * IPv4 payload checksum for UDP fragments must be
2472                  * used in conjunction with packet-split.
2473                  */
2474                 if (adapter->rx_ps_pages)
2475                         rxcsum |= E1000_RXCSUM_IPPCSE;
2476         } else {
2477                 rxcsum &= ~E1000_RXCSUM_TUOFL;
2478                 /* no need to clear IPPCSE as it defaults to 0 */
2479         }
2480         ew32(RXCSUM, rxcsum);
2481
2482         /*
2483          * Enable early receives on supported devices, only takes effect when
2484          * packet size is equal or larger than the specified value (in 8 byte
2485          * units), e.g. using jumbo frames when setting to E1000_ERT_2048
2486          */
2487         if ((adapter->flags & FLAG_HAS_ERT) &&
2488             (adapter->netdev->mtu > ETH_DATA_LEN)) {
2489                 u32 rxdctl = er32(RXDCTL(0));
2490                 ew32(RXDCTL(0), rxdctl | 0x3);
2491                 ew32(ERT, E1000_ERT_2048 | (1 << 13));
2492                 /*
2493                  * With jumbo frames and early-receive enabled, excessive
2494                  * C4->C2 latencies result in dropped transactions.
2495                  */
2496                 pm_qos_update_requirement(PM_QOS_CPU_DMA_LATENCY,
2497                                           e1000e_driver_name, 55);
2498         } else {
2499                 pm_qos_update_requirement(PM_QOS_CPU_DMA_LATENCY,
2500                                           e1000e_driver_name,
2501                                           PM_QOS_DEFAULT_VALUE);
2502         }
2503
2504         /* Enable Receives */
2505         ew32(RCTL, rctl);
2506 }
2507
2508 /**
2509  *  e1000_update_mc_addr_list - Update Multicast addresses
2510  *  @hw: pointer to the HW structure
2511  *  @mc_addr_list: array of multicast addresses to program
2512  *  @mc_addr_count: number of multicast addresses to program
2513  *  @rar_used_count: the first RAR register free to program
2514  *  @rar_count: total number of supported Receive Address Registers
2515  *
2516  *  Updates the Receive Address Registers and Multicast Table Array.
2517  *  The caller must have a packed mc_addr_list of multicast addresses.
2518  *  The parameter rar_count will usually be hw->mac.rar_entry_count
2519  *  unless there are workarounds that change this.  Currently no func pointer
2520  *  exists and all implementations are handled in the generic version of this
2521  *  function.
2522  **/
2523 static void e1000_update_mc_addr_list(struct e1000_hw *hw, u8 *mc_addr_list,
2524                                       u32 mc_addr_count, u32 rar_used_count,
2525                                       u32 rar_count)
2526 {
2527         hw->mac.ops.update_mc_addr_list(hw, mc_addr_list, mc_addr_count,
2528                                         rar_used_count, rar_count);
2529 }
2530
2531 /**
2532  * e1000_set_multi - Multicast and Promiscuous mode set
2533  * @netdev: network interface device structure
2534  *
2535  * The set_multi entry point is called whenever the multicast address
2536  * list or the network interface flags are updated.  This routine is
2537  * responsible for configuring the hardware for proper multicast,
2538  * promiscuous mode, and all-multi behavior.
2539  **/
2540 static void e1000_set_multi(struct net_device *netdev)
2541 {
2542         struct e1000_adapter *adapter = netdev_priv(netdev);
2543         struct e1000_hw *hw = &adapter->hw;
2544         struct e1000_mac_info *mac = &hw->mac;
2545         struct dev_mc_list *mc_ptr;
2546         u8  *mta_list;
2547         u32 rctl;
2548         int i;
2549
2550         /* Check for Promiscuous and All Multicast modes */
2551
2552         rctl = er32(RCTL);
2553
2554         if (netdev->flags & IFF_PROMISC) {
2555                 rctl |= (E1000_RCTL_UPE | E1000_RCTL_MPE);
2556                 rctl &= ~E1000_RCTL_VFE;
2557         } else {
2558                 if (netdev->flags & IFF_ALLMULTI) {
2559                         rctl |= E1000_RCTL_MPE;
2560                         rctl &= ~E1000_RCTL_UPE;
2561                 } else {
2562                         rctl &= ~(E1000_RCTL_UPE | E1000_RCTL_MPE);
2563                 }
2564                 if (adapter->flags & FLAG_HAS_HW_VLAN_FILTER)
2565                         rctl |= E1000_RCTL_VFE;
2566         }
2567
2568         ew32(RCTL, rctl);
2569
2570         if (netdev->mc_count) {
2571                 mta_list = kmalloc(netdev->mc_count * 6, GFP_ATOMIC);
2572                 if (!mta_list)
2573                         return;
2574
2575                 /* prepare a packed array of only addresses. */
2576                 mc_ptr = netdev->mc_list;
2577
2578                 for (i = 0; i < netdev->mc_count; i++) {
2579                         if (!mc_ptr)
2580                                 break;
2581                         memcpy(mta_list + (i*ETH_ALEN), mc_ptr->dmi_addr,
2582                                ETH_ALEN);
2583                         mc_ptr = mc_ptr->next;
2584                 }
2585
2586                 e1000_update_mc_addr_list(hw, mta_list, i, 1,
2587                                           mac->rar_entry_count);
2588                 kfree(mta_list);
2589         } else {
2590                 /*
2591                  * if we're called from probe, we might not have
2592                  * anything to do here, so clear out the list
2593                  */
2594                 e1000_update_mc_addr_list(hw, NULL, 0, 1, mac->rar_entry_count);
2595         }
2596 }
2597
2598 /**
2599  * e1000_configure - configure the hardware for Rx and Tx
2600  * @adapter: private board structure
2601  **/
2602 static void e1000_configure(struct e1000_adapter *adapter)
2603 {
2604         e1000_set_multi(adapter->netdev);
2605
2606         e1000_restore_vlan(adapter);
2607         e1000_init_manageability(adapter);
2608
2609         e1000_configure_tx(adapter);
2610         e1000_setup_rctl(adapter);
2611         e1000_configure_rx(adapter);
2612         adapter->alloc_rx_buf(adapter, e1000_desc_unused(adapter->rx_ring));
2613 }
2614
2615 /**
2616  * e1000e_power_up_phy - restore link in case the phy was powered down
2617  * @adapter: address of board private structure
2618  *
2619  * The phy may be powered down to save power and turn off link when the
2620  * driver is unloaded and wake on lan is not enabled (among others)
2621  * *** this routine MUST be followed by a call to e1000e_reset ***
2622  **/
2623 void e1000e_power_up_phy(struct e1000_adapter *adapter)
2624 {
2625         u16 mii_reg = 0;
2626
2627         /* Just clear the power down bit to wake the phy back up */
2628         if (adapter->hw.phy.media_type == e1000_media_type_copper) {
2629                 /*
2630                  * According to the manual, the phy will retain its
2631                  * settings across a power-down/up cycle
2632                  */
2633                 e1e_rphy(&adapter->hw, PHY_CONTROL, &mii_reg);
2634                 mii_reg &= ~MII_CR_POWER_DOWN;
2635                 e1e_wphy(&adapter->hw, PHY_CONTROL, mii_reg);
2636         }
2637
2638         adapter->hw.mac.ops.setup_link(&adapter->hw);
2639 }
2640
2641 /**
2642  * e1000_power_down_phy - Power down the PHY
2643  *
2644  * Power down the PHY so no link is implied when interface is down
2645  * The PHY cannot be powered down is management or WoL is active
2646  */
2647 static void e1000_power_down_phy(struct e1000_adapter *adapter)
2648 {
2649         struct e1000_hw *hw = &adapter->hw;
2650         u16 mii_reg;
2651
2652         /* WoL is enabled */
2653         if (adapter->wol)
2654                 return;
2655
2656         /* non-copper PHY? */
2657         if (adapter->hw.phy.media_type != e1000_media_type_copper)
2658                 return;
2659
2660         /* reset is blocked because of a SoL/IDER session */
2661         if (e1000e_check_mng_mode(hw) || e1000_check_reset_block(hw))
2662                 return;
2663
2664         /* manageability (AMT) is enabled */
2665         if (er32(MANC) & E1000_MANC_SMBUS_EN)
2666                 return;
2667
2668         /* power down the PHY */
2669         e1e_rphy(hw, PHY_CONTROL, &mii_reg);
2670         mii_reg |= MII_CR_POWER_DOWN;
2671         e1e_wphy(hw, PHY_CONTROL, mii_reg);
2672         mdelay(1);
2673 }
2674
2675 /**
2676  * e1000e_reset - bring the hardware into a known good state
2677  *
2678  * This function boots the hardware and enables some settings that
2679  * require a configuration cycle of the hardware - those cannot be
2680  * set/changed during runtime. After reset the device needs to be
2681  * properly configured for Rx, Tx etc.
2682  */
2683 void e1000e_reset(struct e1000_adapter *adapter)
2684 {
2685         struct e1000_mac_info *mac = &adapter->hw.mac;
2686         struct e1000_fc_info *fc = &adapter->hw.fc;
2687         struct e1000_hw *hw = &adapter->hw;
2688         u32 tx_space, min_tx_space, min_rx_space;
2689         u32 pba = adapter->pba;
2690         u16 hwm;
2691
2692         /* reset Packet Buffer Allocation to default */
2693         ew32(PBA, pba);
2694
2695         if (adapter->max_frame_size > ETH_FRAME_LEN + ETH_FCS_LEN) {
2696                 /*
2697                  * To maintain wire speed transmits, the Tx FIFO should be
2698                  * large enough to accommodate two full transmit packets,
2699                  * rounded up to the next 1KB and expressed in KB.  Likewise,
2700                  * the Rx FIFO should be large enough to accommodate at least
2701                  * one full receive packet and is similarly rounded up and
2702                  * expressed in KB.
2703                  */
2704                 pba = er32(PBA);
2705                 /* upper 16 bits has Tx packet buffer allocation size in KB */
2706                 tx_space = pba >> 16;
2707                 /* lower 16 bits has Rx packet buffer allocation size in KB */
2708                 pba &= 0xffff;
2709                 /*
2710                  * the Tx fifo also stores 16 bytes of information about the tx
2711                  * but don't include ethernet FCS because hardware appends it
2712                  */
2713                 min_tx_space = (adapter->max_frame_size +
2714                                 sizeof(struct e1000_tx_desc) -
2715                                 ETH_FCS_LEN) * 2;
2716                 min_tx_space = ALIGN(min_tx_space, 1024);
2717                 min_tx_space >>= 10;
2718                 /* software strips receive CRC, so leave room for it */
2719                 min_rx_space = adapter->max_frame_size;
2720                 min_rx_space = ALIGN(min_rx_space, 1024);
2721                 min_rx_space >>= 10;
2722
2723                 /*
2724                  * If current Tx allocation is less than the min Tx FIFO size,
2725                  * and the min Tx FIFO size is less than the current Rx FIFO
2726                  * allocation, take space away from current Rx allocation
2727                  */
2728                 if ((tx_space < min_tx_space) &&
2729                     ((min_tx_space - tx_space) < pba)) {
2730                         pba -= min_tx_space - tx_space;
2731
2732                         /*
2733                          * if short on Rx space, Rx wins and must trump tx
2734                          * adjustment or use Early Receive if available
2735                          */
2736                         if ((pba < min_rx_space) &&
2737                             (!(adapter->flags & FLAG_HAS_ERT)))
2738                                 /* ERT enabled in e1000_configure_rx */
2739                                 pba = min_rx_space;
2740                 }
2741
2742                 ew32(PBA, pba);
2743         }
2744
2745
2746         /*
2747          * flow control settings
2748          *
2749          * The high water mark must be low enough to fit two full frame
2750          * (or the size used for early receive) above it in the Rx FIFO.
2751          * Set it to the lower of:
2752          * - 90% of the Rx FIFO size, and
2753          * - the full Rx FIFO size minus the early receive size (for parts
2754          *   with ERT support assuming ERT set to E1000_ERT_2048), or
2755          * - the full Rx FIFO size minus two full frames
2756          */
2757         if ((adapter->flags & FLAG_HAS_ERT) &&
2758             (adapter->netdev->mtu > ETH_DATA_LEN))
2759                 hwm = min(((pba << 10) * 9 / 10),
2760                           ((pba << 10) - (E1000_ERT_2048 << 3)));
2761         else
2762                 hwm = min(((pba << 10) * 9 / 10),
2763                           ((pba << 10) - (2 * adapter->max_frame_size)));
2764
2765         fc->high_water = hwm & E1000_FCRTH_RTH; /* 8-byte granularity */
2766         fc->low_water = (fc->high_water - (2 * adapter->max_frame_size));
2767         fc->low_water &= E1000_FCRTL_RTL; /* 8-byte granularity */
2768
2769         if (adapter->flags & FLAG_DISABLE_FC_PAUSE_TIME)
2770                 fc->pause_time = 0xFFFF;
2771         else
2772                 fc->pause_time = E1000_FC_PAUSE_TIME;
2773         fc->send_xon = 1;
2774         fc->current_mode = fc->requested_mode;
2775
2776         /* Allow time for pending master requests to run */
2777         mac->ops.reset_hw(hw);
2778
2779         /*
2780          * For parts with AMT enabled, let the firmware know
2781          * that the network interface is in control
2782          */
2783         if (adapter->flags & FLAG_HAS_AMT)
2784                 e1000_get_hw_control(adapter);
2785
2786         ew32(WUC, 0);
2787         if (adapter->flags2 & FLAG2_HAS_PHY_WAKEUP)
2788                 e1e_wphy(&adapter->hw, BM_WUC, 0);
2789
2790         if (mac->ops.init_hw(hw))
2791                 e_err("Hardware Error\n");
2792
2793         e1000_update_mng_vlan(adapter);
2794
2795         /* Enable h/w to recognize an 802.1Q VLAN Ethernet packet */
2796         ew32(VET, ETH_P_8021Q);
2797
2798         e1000e_reset_adaptive(hw);
2799         e1000_get_phy_info(hw);
2800
2801         if ((adapter->flags & FLAG_HAS_SMART_POWER_DOWN) &&
2802             !(adapter->flags & FLAG_SMART_POWER_DOWN)) {
2803                 u16 phy_data = 0;
2804                 /*
2805                  * speed up time to link by disabling smart power down, ignore
2806                  * the return value of this function because there is nothing
2807                  * different we would do if it failed
2808                  */
2809                 e1e_rphy(hw, IGP02E1000_PHY_POWER_MGMT, &phy_data);
2810                 phy_data &= ~IGP02E1000_PM_SPD;
2811                 e1e_wphy(hw, IGP02E1000_PHY_POWER_MGMT, phy_data);
2812         }
2813 }
2814
2815 int e1000e_up(struct e1000_adapter *adapter)
2816 {
2817         struct e1000_hw *hw = &adapter->hw;
2818
2819         /* hardware has been reset, we need to reload some things */
2820         e1000_configure(adapter);
2821
2822         clear_bit(__E1000_DOWN, &adapter->state);
2823
2824         napi_enable(&adapter->napi);
2825         if (adapter->msix_entries)
2826                 e1000_configure_msix(adapter);
2827         e1000_irq_enable(adapter);
2828
2829         netif_wake_queue(adapter->netdev);
2830
2831         /* fire a link change interrupt to start the watchdog */
2832         ew32(ICS, E1000_ICS_LSC);
2833         return 0;
2834 }
2835
2836 void e1000e_down(struct e1000_adapter *adapter)
2837 {
2838         struct net_device *netdev = adapter->netdev;
2839         struct e1000_hw *hw = &adapter->hw;
2840         u32 tctl, rctl;
2841
2842         /*
2843          * signal that we're down so the interrupt handler does not
2844          * reschedule our watchdog timer
2845          */
2846         set_bit(__E1000_DOWN, &adapter->state);
2847
2848         /* disable receives in the hardware */
2849         rctl = er32(RCTL);
2850         ew32(RCTL, rctl & ~E1000_RCTL_EN);
2851         /* flush and sleep below */
2852
2853         netif_stop_queue(netdev);
2854
2855         /* disable transmits in the hardware */
2856         tctl = er32(TCTL);
2857         tctl &= ~E1000_TCTL_EN;
2858         ew32(TCTL, tctl);
2859         /* flush both disables and wait for them to finish */
2860         e1e_flush();
2861         msleep(10);
2862
2863         napi_disable(&adapter->napi);
2864         e1000_irq_disable(adapter);
2865
2866         del_timer_sync(&adapter->watchdog_timer);
2867         del_timer_sync(&adapter->phy_info_timer);
2868
2869         netdev->tx_queue_len = adapter->tx_queue_len;
2870         netif_carrier_off(netdev);
2871         adapter->link_speed = 0;
2872         adapter->link_duplex = 0;
2873
2874         if (!pci_channel_offline(adapter->pdev))
2875                 e1000e_reset(adapter);
2876         e1000_clean_tx_ring(adapter);
2877         e1000_clean_rx_ring(adapter);
2878
2879         /*
2880          * TODO: for power management, we could drop the link and
2881          * pci_disable_device here.
2882          */
2883 }
2884
2885 void e1000e_reinit_locked(struct e1000_adapter *adapter)
2886 {
2887         might_sleep();
2888         while (test_and_set_bit(__E1000_RESETTING, &adapter->state))
2889                 msleep(1);
2890         e1000e_down(adapter);
2891         e1000e_up(adapter);
2892         clear_bit(__E1000_RESETTING, &adapter->state);
2893 }
2894
2895 /**
2896  * e1000_sw_init - Initialize general software structures (struct e1000_adapter)
2897  * @adapter: board private structure to initialize
2898  *
2899  * e1000_sw_init initializes the Adapter private data structure.
2900  * Fields are initialized based on PCI device information and
2901  * OS network device settings (MTU size).
2902  **/
2903 static int __devinit e1000_sw_init(struct e1000_adapter *adapter)
2904 {
2905         struct net_device *netdev = adapter->netdev;
2906
2907         adapter->rx_buffer_len = ETH_FRAME_LEN + VLAN_HLEN + ETH_FCS_LEN;
2908         adapter->rx_ps_bsize0 = 128;
2909         adapter->max_frame_size = netdev->mtu + ETH_HLEN + ETH_FCS_LEN;
2910         adapter->min_frame_size = ETH_ZLEN + ETH_FCS_LEN;
2911
2912         e1000e_set_interrupt_capability(adapter);
2913
2914         if (e1000_alloc_queues(adapter))
2915                 return -ENOMEM;
2916
2917         /* Explicitly disable IRQ since the NIC can be in any state. */
2918         e1000_irq_disable(adapter);
2919
2920         set_bit(__E1000_DOWN, &adapter->state);
2921         return 0;
2922 }
2923
2924 /**
2925  * e1000_intr_msi_test - Interrupt Handler
2926  * @irq: interrupt number
2927  * @data: pointer to a network interface device structure
2928  **/
2929 static irqreturn_t e1000_intr_msi_test(int irq, void *data)
2930 {
2931         struct net_device *netdev = data;
2932         struct e1000_adapter *adapter = netdev_priv(netdev);
2933         struct e1000_hw *hw = &adapter->hw;
2934         u32 icr = er32(ICR);
2935
2936         e_dbg("%s: icr is %08X\n", netdev->name, icr);
2937         if (icr & E1000_ICR_RXSEQ) {
2938                 adapter->flags &= ~FLAG_MSI_TEST_FAILED;
2939                 wmb();
2940         }
2941
2942         return IRQ_HANDLED;
2943 }
2944
2945 /**
2946  * e1000_test_msi_interrupt - Returns 0 for successful test
2947  * @adapter: board private struct
2948  *
2949  * code flow taken from tg3.c
2950  **/
2951 static int e1000_test_msi_interrupt(struct e1000_adapter *adapter)
2952 {
2953         struct net_device *netdev = adapter->netdev;
2954         struct e1000_hw *hw = &adapter->hw;
2955         int err;
2956
2957         /* poll_enable hasn't been called yet, so don't need disable */
2958         /* clear any pending events */
2959         er32(ICR);
2960
2961         /* free the real vector and request a test handler */
2962         e1000_free_irq(adapter);
2963         e1000e_reset_interrupt_capability(adapter);
2964
2965         /* Assume that the test fails, if it succeeds then the test
2966          * MSI irq handler will unset this flag */
2967         adapter->flags |= FLAG_MSI_TEST_FAILED;
2968
2969         err = pci_enable_msi(adapter->pdev);
2970         if (err)
2971                 goto msi_test_failed;
2972
2973         err = request_irq(adapter->pdev->irq, e1000_intr_msi_test, 0,
2974                           netdev->name, netdev);
2975         if (err) {
2976                 pci_disable_msi(adapter->pdev);
2977                 goto msi_test_failed;
2978         }
2979
2980         wmb();
2981
2982         e1000_irq_enable(adapter);
2983
2984         /* fire an unusual interrupt on the test handler */
2985         ew32(ICS, E1000_ICS_RXSEQ);
2986         e1e_flush();
2987         msleep(50);
2988
2989         e1000_irq_disable(adapter);
2990
2991         rmb();
2992
2993         if (adapter->flags & FLAG_MSI_TEST_FAILED) {
2994                 adapter->int_mode = E1000E_INT_MODE_LEGACY;
2995                 err = -EIO;
2996                 e_info("MSI interrupt test failed!\n");
2997         }
2998
2999         free_irq(adapter->pdev->irq, netdev);
3000         pci_disable_msi(adapter->pdev);
3001
3002         if (err == -EIO)
3003                 goto msi_test_failed;
3004
3005         /* okay so the test worked, restore settings */
3006         e_dbg("%s: MSI interrupt test succeeded!\n", netdev->name);
3007 msi_test_failed:
3008         e1000e_set_interrupt_capability(adapter);
3009         e1000_request_irq(adapter);
3010         return err;
3011 }
3012
3013 /**
3014  * e1000_test_msi - Returns 0 if MSI test succeeds or INTx mode is restored
3015  * @adapter: board private struct
3016  *
3017  * code flow taken from tg3.c, called with e1000 interrupts disabled.
3018  **/
3019 static int e1000_test_msi(struct e1000_adapter *adapter)
3020 {
3021         int err;
3022         u16 pci_cmd;
3023
3024         if (!(adapter->flags & FLAG_MSI_ENABLED))
3025                 return 0;
3026
3027         /* disable SERR in case the MSI write causes a master abort */
3028         pci_read_config_word(adapter->pdev, PCI_COMMAND, &pci_cmd);
3029         pci_write_config_word(adapter->pdev, PCI_COMMAND,
3030                               pci_cmd & ~PCI_COMMAND_SERR);
3031
3032         err = e1000_test_msi_interrupt(adapter);
3033
3034         /* restore previous setting of command word */
3035         pci_write_config_word(adapter->pdev, PCI_COMMAND, pci_cmd);
3036
3037         /* success ! */
3038         if (!err)
3039                 return 0;
3040
3041         /* EIO means MSI test failed */
3042         if (err != -EIO)
3043                 return err;
3044
3045         /* back to INTx mode */
3046         e_warn("MSI interrupt test failed, using legacy interrupt.\n");
3047
3048         e1000_free_irq(adapter);
3049
3050         err = e1000_request_irq(adapter);
3051
3052         return err;
3053 }
3054
3055 /**
3056  * e1000_open - Called when a network interface is made active
3057  * @netdev: network interface device structure
3058  *
3059  * Returns 0 on success, negative value on failure
3060  *
3061  * The open entry point is called when a network interface is made
3062  * active by the system (IFF_UP).  At this point all resources needed
3063  * for transmit and receive operations are allocated, the interrupt
3064  * handler is registered with the OS, the watchdog timer is started,
3065  * and the stack is notified that the interface is ready.
3066  **/
3067 static int e1000_open(struct net_device *netdev)
3068 {
3069         struct e1000_adapter *adapter = netdev_priv(netdev);
3070         struct e1000_hw *hw = &adapter->hw;
3071         int err;
3072
3073         /* disallow open during test */
3074         if (test_bit(__E1000_TESTING, &adapter->state))
3075                 return -EBUSY;
3076
3077         netif_carrier_off(netdev);
3078
3079         /* allocate transmit descriptors */
3080         err = e1000e_setup_tx_resources(adapter);
3081         if (err)
3082                 goto err_setup_tx;
3083
3084         /* allocate receive descriptors */
3085         err = e1000e_setup_rx_resources(adapter);
3086         if (err)
3087                 goto err_setup_rx;
3088
3089         e1000e_power_up_phy(adapter);
3090
3091         adapter->mng_vlan_id = E1000_MNG_VLAN_NONE;
3092         if ((adapter->hw.mng_cookie.status &
3093              E1000_MNG_DHCP_COOKIE_STATUS_VLAN))
3094                 e1000_update_mng_vlan(adapter);
3095
3096         /*
3097          * If AMT is enabled, let the firmware know that the network
3098          * interface is now open
3099          */
3100         if (adapter->flags & FLAG_HAS_AMT)
3101                 e1000_get_hw_control(adapter);
3102
3103         /*
3104          * before we allocate an interrupt, we must be ready to handle it.
3105          * Setting DEBUG_SHIRQ in the kernel makes it fire an interrupt
3106          * as soon as we call pci_request_irq, so we have to setup our
3107          * clean_rx handler before we do so.
3108          */
3109         e1000_configure(adapter);
3110
3111         err = e1000_request_irq(adapter);
3112         if (err)
3113                 goto err_req_irq;
3114
3115         /*
3116          * Work around PCIe errata with MSI interrupts causing some chipsets to
3117          * ignore e1000e MSI messages, which means we need to test our MSI
3118          * interrupt now
3119          */
3120         if (adapter->int_mode != E1000E_INT_MODE_LEGACY) {
3121                 err = e1000_test_msi(adapter);
3122                 if (err) {
3123                         e_err("Interrupt allocation failed\n");
3124                         goto err_req_irq;
3125                 }
3126         }
3127
3128         /* From here on the code is the same as e1000e_up() */
3129         clear_bit(__E1000_DOWN, &adapter->state);
3130
3131         napi_enable(&adapter->napi);
3132
3133         e1000_irq_enable(adapter);
3134
3135         netif_start_queue(netdev);
3136
3137         /* fire a link status change interrupt to start the watchdog */
3138         ew32(ICS, E1000_ICS_LSC);
3139
3140         return 0;
3141
3142 err_req_irq:
3143         e1000_release_hw_control(adapter);
3144         e1000_power_down_phy(adapter);
3145         e1000e_free_rx_resources(adapter);
3146 err_setup_rx:
3147         e1000e_free_tx_resources(adapter);
3148 err_setup_tx:
3149         e1000e_reset(adapter);
3150
3151         return err;
3152 }
3153
3154 /**
3155  * e1000_close - Disables a network interface
3156  * @netdev: network interface device structure
3157  *
3158  * Returns 0, this is not allowed to fail
3159  *
3160  * The close entry point is called when an interface is de-activated
3161  * by the OS.  The hardware is still under the drivers control, but
3162  * needs to be disabled.  A global MAC reset is issued to stop the
3163  * hardware, and all transmit and receive resources are freed.
3164  **/
3165 static int e1000_close(struct net_device *netdev)
3166 {
3167         struct e1000_adapter *adapter = netdev_priv(netdev);
3168
3169         WARN_ON(test_bit(__E1000_RESETTING, &adapter->state));
3170         e1000e_down(adapter);
3171         e1000_power_down_phy(adapter);
3172         e1000_free_irq(adapter);
3173
3174         e1000e_free_tx_resources(adapter);
3175         e1000e_free_rx_resources(adapter);
3176
3177         /*
3178          * kill manageability vlan ID if supported, but not if a vlan with
3179          * the same ID is registered on the host OS (let 8021q kill it)
3180          */
3181         if ((adapter->hw.mng_cookie.status &
3182                           E1000_MNG_DHCP_COOKIE_STATUS_VLAN) &&
3183              !(adapter->vlgrp &&
3184                vlan_group_get_device(adapter->vlgrp, adapter->mng_vlan_id)))
3185                 e1000_vlan_rx_kill_vid(netdev, adapter->mng_vlan_id);
3186
3187         /*
3188          * If AMT is enabled, let the firmware know that the network
3189          * interface is now closed
3190          */
3191         if (adapter->flags & FLAG_HAS_AMT)
3192                 e1000_release_hw_control(adapter);
3193
3194         return 0;
3195 }
3196 /**
3197  * e1000_set_mac - Change the Ethernet Address of the NIC
3198  * @netdev: network interface device structure
3199  * @p: pointer to an address structure
3200  *
3201  * Returns 0 on success, negative on failure
3202  **/
3203 static int e1000_set_mac(struct net_device *netdev, void *p)
3204 {
3205         struct e1000_adapter *adapter = netdev_priv(netdev);
3206         struct sockaddr *addr = p;
3207
3208         if (!is_valid_ether_addr(addr->sa_data))
3209                 return -EADDRNOTAVAIL;
3210
3211         memcpy(netdev->dev_addr, addr->sa_data, netdev->addr_len);
3212         memcpy(adapter->hw.mac.addr, addr->sa_data, netdev->addr_len);
3213
3214         e1000e_rar_set(&adapter->hw, adapter->hw.mac.addr, 0);
3215
3216         if (adapter->flags & FLAG_RESET_OVERWRITES_LAA) {
3217                 /* activate the work around */
3218                 e1000e_set_laa_state_82571(&adapter->hw, 1);
3219
3220                 /*
3221                  * Hold a copy of the LAA in RAR[14] This is done so that
3222                  * between the time RAR[0] gets clobbered  and the time it
3223                  * gets fixed (in e1000_watchdog), the actual LAA is in one
3224                  * of the RARs and no incoming packets directed to this port
3225                  * are dropped. Eventually the LAA will be in RAR[0] and
3226                  * RAR[14]
3227                  */
3228                 e1000e_rar_set(&adapter->hw,
3229                               adapter->hw.mac.addr,
3230                               adapter->hw.mac.rar_entry_count - 1);
3231         }
3232
3233         return 0;
3234 }
3235
3236 /**
3237  * e1000e_update_phy_task - work thread to update phy
3238  * @work: pointer to our work struct
3239  *
3240  * this worker thread exists because we must acquire a
3241  * semaphore to read the phy, which we could msleep while
3242  * waiting for it, and we can't msleep in a timer.
3243  **/
3244 static void e1000e_update_phy_task(struct work_struct *work)
3245 {
3246         struct e1000_adapter *adapter = container_of(work,
3247                                         struct e1000_adapter, update_phy_task);
3248         e1000_get_phy_info(&adapter->hw);
3249 }
3250
3251 /*
3252  * Need to wait a few seconds after link up to get diagnostic information from
3253  * the phy
3254  */
3255 static void e1000_update_phy_info(unsigned long data)
3256 {
3257         struct e1000_adapter *adapter = (struct e1000_adapter *) data;
3258         schedule_work(&adapter->update_phy_task);
3259 }
3260
3261 /**
3262  * e1000e_update_stats - Update the board statistics counters
3263  * @adapter: board private structure
3264  **/
3265 void e1000e_update_stats(struct e1000_adapter *adapter)
3266 {
3267         struct net_device *netdev = adapter->netdev;
3268         struct e1000_hw *hw = &adapter->hw;
3269         struct pci_dev *pdev = adapter->pdev;
3270         u16 phy_data;
3271
3272         /*
3273          * Prevent stats update while adapter is being reset, or if the pci
3274          * connection is down.
3275          */
3276         if (adapter->link_speed == 0)
3277                 return;
3278         if (pci_channel_offline(pdev))
3279                 return;
3280
3281         adapter->stats.crcerrs += er32(CRCERRS);
3282         adapter->stats.gprc += er32(GPRC);
3283         adapter->stats.gorc += er32(GORCL);
3284         er32(GORCH); /* Clear gorc */
3285         adapter->stats.bprc += er32(BPRC);
3286         adapter->stats.mprc += er32(MPRC);
3287         adapter->stats.roc += er32(ROC);
3288
3289         adapter->stats.mpc += er32(MPC);
3290         if ((hw->phy.type == e1000_phy_82578) ||
3291             (hw->phy.type == e1000_phy_82577)) {
3292                 e1e_rphy(hw, HV_SCC_UPPER, &phy_data);
3293                 e1e_rphy(hw, HV_SCC_LOWER, &phy_data);
3294                 adapter->stats.scc += phy_data;
3295
3296                 e1e_rphy(hw, HV_ECOL_UPPER, &phy_data);
3297                 e1e_rphy(hw, HV_ECOL_LOWER, &phy_data);
3298                 adapter->stats.ecol += phy_data;
3299
3300                 e1e_rphy(hw, HV_MCC_UPPER, &phy_data);
3301                 e1e_rphy(hw, HV_MCC_LOWER, &phy_data);
3302                 adapter->stats.mcc += phy_data;
3303
3304                 e1e_rphy(hw, HV_LATECOL_UPPER, &phy_data);
3305                 e1e_rphy(hw, HV_LATECOL_LOWER, &phy_data);
3306                 adapter->stats.latecol += phy_data;
3307
3308                 e1e_rphy(hw, HV_DC_UPPER, &phy_data);
3309                 e1e_rphy(hw, HV_DC_LOWER, &phy_data);
3310                 adapter->stats.dc += phy_data;
3311         } else {
3312                 adapter->stats.scc += er32(SCC);
3313                 adapter->stats.ecol += er32(ECOL);
3314                 adapter->stats.mcc += er32(MCC);
3315                 adapter->stats.latecol += er32(LATECOL);
3316                 adapter->stats.dc += er32(DC);
3317         }
3318         adapter->stats.xonrxc += er32(XONRXC);
3319         adapter->stats.xontxc += er32(XONTXC);
3320         adapter->stats.xoffrxc += er32(XOFFRXC);
3321         adapter->stats.xofftxc += er32(XOFFTXC);
3322         adapter->stats.gptc += er32(GPTC);
3323         adapter->stats.gotc += er32(GOTCL);
3324         er32(GOTCH); /* Clear gotc */
3325         adapter->stats.rnbc += er32(RNBC);
3326         adapter->stats.ruc += er32(RUC);
3327
3328         adapter->stats.mptc += er32(MPTC);
3329         adapter->stats.bptc += er32(BPTC);
3330
3331         /* used for adaptive IFS */
3332
3333         hw->mac.tx_packet_delta = er32(TPT);
3334         adapter->stats.tpt += hw->mac.tx_packet_delta;
3335         if ((hw->phy.type == e1000_phy_82578) ||
3336             (hw->phy.type == e1000_phy_82577)) {
3337                 e1e_rphy(hw, HV_COLC_UPPER, &phy_data);
3338                 e1e_rphy(hw, HV_COLC_LOWER, &phy_data);
3339                 hw->mac.collision_delta = phy_data;
3340         } else {
3341                 hw->mac.collision_delta = er32(COLC);
3342         }
3343         adapter->stats.colc += hw->mac.collision_delta;
3344
3345         adapter->stats.algnerrc += er32(ALGNERRC);
3346         adapter->stats.rxerrc += er32(RXERRC);
3347         if ((hw->phy.type == e1000_phy_82578) ||
3348             (hw->phy.type == e1000_phy_82577)) {
3349                 e1e_rphy(hw, HV_TNCRS_UPPER, &phy_data);
3350                 e1e_rphy(hw, HV_TNCRS_LOWER, &phy_data);
3351                 adapter->stats.tncrs += phy_data;
3352         } else {
3353                 if ((hw->mac.type != e1000_82574) &&
3354                     (hw->mac.type != e1000_82583))
3355                         adapter->stats.tncrs += er32(TNCRS);
3356         }
3357         adapter->stats.cexterr += er32(CEXTERR);
3358         adapter->stats.tsctc += er32(TSCTC);
3359         adapter->stats.tsctfc += er32(TSCTFC);
3360
3361         /* Fill out the OS statistics structure */
3362         netdev->stats.multicast = adapter->stats.mprc;
3363         netdev->stats.collisions = adapter->stats.colc;
3364
3365         /* Rx Errors */
3366
3367         /*
3368          * RLEC on some newer hardware can be incorrect so build
3369          * our own version based on RUC and ROC
3370          */
3371         netdev->stats.rx_errors = adapter->stats.rxerrc +
3372                 adapter->stats.crcerrs + adapter->stats.algnerrc +
3373                 adapter->stats.ruc + adapter->stats.roc +
3374                 adapter->stats.cexterr;
3375         netdev->stats.rx_length_errors = adapter->stats.ruc +
3376                                               adapter->stats.roc;
3377         netdev->stats.rx_crc_errors = adapter->stats.crcerrs;
3378         netdev->stats.rx_frame_errors = adapter->stats.algnerrc;
3379         netdev->stats.rx_missed_errors = adapter->stats.mpc;
3380
3381         /* Tx Errors */
3382         netdev->stats.tx_errors = adapter->stats.ecol +
3383                                        adapter->stats.latecol;
3384         netdev->stats.tx_aborted_errors = adapter->stats.ecol;
3385         netdev->stats.tx_window_errors = adapter->stats.latecol;
3386         netdev->stats.tx_carrier_errors = adapter->stats.tncrs;
3387
3388         /* Tx Dropped needs to be maintained elsewhere */
3389
3390         /* Management Stats */
3391         adapter->stats.mgptc += er32(MGTPTC);
3392         adapter->stats.mgprc += er32(MGTPRC);
3393         adapter->stats.mgpdc += er32(MGTPDC);
3394 }
3395
3396 /**
3397  * e1000_phy_read_status - Update the PHY register status snapshot
3398  * @adapter: board private structure
3399  **/
3400 static void e1000_phy_read_status(struct e1000_adapter *adapter)
3401 {
3402         struct e1000_hw *hw = &adapter->hw;
3403         struct e1000_phy_regs *phy = &adapter->phy_regs;
3404         int ret_val;
3405
3406         if ((er32(STATUS) & E1000_STATUS_LU) &&
3407             (adapter->hw.phy.media_type == e1000_media_type_copper)) {
3408                 ret_val  = e1e_rphy(hw, PHY_CONTROL, &phy->bmcr);
3409                 ret_val |= e1e_rphy(hw, PHY_STATUS, &phy->bmsr);
3410                 ret_val |= e1e_rphy(hw, PHY_AUTONEG_ADV, &phy->advertise);
3411                 ret_val |= e1e_rphy(hw, PHY_LP_ABILITY, &phy->lpa);
3412                 ret_val |= e1e_rphy(hw, PHY_AUTONEG_EXP, &phy->expansion);
3413                 ret_val |= e1e_rphy(hw, PHY_1000T_CTRL, &phy->ctrl1000);
3414                 ret_val |= e1e_rphy(hw, PHY_1000T_STATUS, &phy->stat1000);
3415                 ret_val |= e1e_rphy(hw, PHY_EXT_STATUS, &phy->estatus);
3416                 if (ret_val)
3417                         e_warn("Error reading PHY register\n");
3418         } else {
3419                 /*
3420                  * Do not read PHY registers if link is not up
3421                  * Set values to typical power-on defaults
3422                  */
3423                 phy->bmcr = (BMCR_SPEED1000 | BMCR_ANENABLE | BMCR_FULLDPLX);
3424                 phy->bmsr = (BMSR_100FULL | BMSR_100HALF | BMSR_10FULL |
3425                              BMSR_10HALF | BMSR_ESTATEN | BMSR_ANEGCAPABLE |
3426                              BMSR_ERCAP);
3427                 phy->advertise = (ADVERTISE_PAUSE_ASYM | ADVERTISE_PAUSE_CAP |
3428                                   ADVERTISE_ALL | ADVERTISE_CSMA);
3429                 phy->lpa = 0;
3430                 phy->expansion = EXPANSION_ENABLENPAGE;
3431                 phy->ctrl1000 = ADVERTISE_1000FULL;
3432                 phy->stat1000 = 0;
3433                 phy->estatus = (ESTATUS_1000_TFULL | ESTATUS_1000_THALF);
3434         }
3435 }
3436
3437 static void e1000_print_link_info(struct e1000_adapter *adapter)
3438 {
3439         struct e1000_hw *hw = &adapter->hw;
3440         u32 ctrl = er32(CTRL);
3441
3442         /* Link status message must follow this format for user tools */
3443         printk(KERN_INFO "e1000e: %s NIC Link is Up %d Mbps %s, "
3444                "Flow Control: %s\n",
3445                adapter->netdev->name,
3446                adapter->link_speed,
3447                (adapter->link_duplex == FULL_DUPLEX) ?
3448                                 "Full Duplex" : "Half Duplex",
3449                ((ctrl & E1000_CTRL_TFCE) && (ctrl & E1000_CTRL_RFCE)) ?
3450                                 "RX/TX" :
3451                ((ctrl & E1000_CTRL_RFCE) ? "RX" :
3452                ((ctrl & E1000_CTRL_TFCE) ? "TX" : "None" )));
3453 }
3454
3455 bool e1000_has_link(struct e1000_adapter *adapter)
3456 {
3457         struct e1000_hw *hw = &adapter->hw;
3458         bool link_active = 0;
3459         s32 ret_val = 0;
3460
3461         /*
3462          * get_link_status is set on LSC (link status) interrupt or
3463          * Rx sequence error interrupt.  get_link_status will stay
3464          * false until the check_for_link establishes link
3465          * for copper adapters ONLY
3466          */
3467         switch (hw->phy.media_type) {
3468         case e1000_media_type_copper:
3469                 if (hw->mac.get_link_status) {
3470                         ret_val = hw->mac.ops.check_for_link(hw);
3471                         link_active = !hw->mac.get_link_status;
3472                 } else {
3473                         link_active = 1;
3474                 }
3475                 break;
3476         case e1000_media_type_fiber:
3477                 ret_val = hw->mac.ops.check_for_link(hw);
3478                 link_active = !!(er32(STATUS) & E1000_STATUS_LU);
3479                 break;
3480         case e1000_media_type_internal_serdes:
3481                 ret_val = hw->mac.ops.check_for_link(hw);
3482                 link_active = adapter->hw.mac.serdes_has_link;
3483                 break;
3484         default:
3485         case e1000_media_type_unknown:
3486                 break;
3487         }
3488
3489         if ((ret_val == E1000_ERR_PHY) && (hw->phy.type == e1000_phy_igp_3) &&
3490             (er32(CTRL) & E1000_PHY_CTRL_GBE_DISABLE)) {
3491                 /* See e1000_kmrn_lock_loss_workaround_ich8lan() */
3492                 e_info("Gigabit has been disabled, downgrading speed\n");
3493         }
3494
3495         return link_active;
3496 }
3497
3498 static void e1000e_enable_receives(struct e1000_adapter *adapter)
3499 {
3500         /* make sure the receive unit is started */
3501         if ((adapter->flags & FLAG_RX_NEEDS_RESTART) &&
3502             (adapter->flags & FLAG_RX_RESTART_NOW)) {
3503                 struct e1000_hw *hw = &adapter->hw;
3504                 u32 rctl = er32(RCTL);
3505                 ew32(RCTL, rctl | E1000_RCTL_EN);
3506                 adapter->flags &= ~FLAG_RX_RESTART_NOW;
3507         }
3508 }
3509
3510 /**
3511  * e1000_watchdog - Timer Call-back
3512  * @data: pointer to adapter cast into an unsigned long
3513  **/
3514 static void e1000_watchdog(unsigned long data)
3515 {
3516         struct e1000_adapter *adapter = (struct e1000_adapter *) data;
3517
3518         /* Do the rest outside of interrupt context */
3519         schedule_work(&adapter->watchdog_task);
3520
3521         /* TODO: make this use queue_delayed_work() */
3522 }
3523
3524 static void e1000_watchdog_task(struct work_struct *work)
3525 {
3526         struct e1000_adapter *adapter = container_of(work,
3527                                         struct e1000_adapter, watchdog_task);
3528         struct net_device *netdev = adapter->netdev;
3529         struct e1000_mac_info *mac = &adapter->hw.mac;
3530         struct e1000_phy_info *phy = &adapter->hw.phy;
3531         struct e1000_ring *tx_ring = adapter->tx_ring;
3532         struct e1000_hw *hw = &adapter->hw;
3533         u32 link, tctl;
3534         int tx_pending = 0;
3535
3536         link = e1000_has_link(adapter);
3537         if ((netif_carrier_ok(netdev)) && link) {
3538                 e1000e_enable_receives(adapter);
3539                 goto link_up;
3540         }
3541
3542         if ((e1000e_enable_tx_pkt_filtering(hw)) &&
3543             (adapter->mng_vlan_id != adapter->hw.mng_cookie.vlan_id))
3544                 e1000_update_mng_vlan(adapter);
3545
3546         if (link) {
3547                 if (!netif_carrier_ok(netdev)) {
3548                         bool txb2b = 1;
3549                         /* update snapshot of PHY registers on LSC */
3550                         e1000_phy_read_status(adapter);
3551                         mac->ops.get_link_up_info(&adapter->hw,
3552                                                    &adapter->link_speed,
3553                                                    &adapter->link_duplex);
3554                         e1000_print_link_info(adapter);
3555                         /*
3556                          * On supported PHYs, check for duplex mismatch only
3557                          * if link has autonegotiated at 10/100 half
3558                          */
3559                         if ((hw->phy.type == e1000_phy_igp_3 ||
3560                              hw->phy.type == e1000_phy_bm) &&
3561                             (hw->mac.autoneg == true) &&
3562                             (adapter->link_speed == SPEED_10 ||
3563                              adapter->link_speed == SPEED_100) &&
3564                             (adapter->link_duplex == HALF_DUPLEX)) {
3565                                 u16 autoneg_exp;
3566
3567                                 e1e_rphy(hw, PHY_AUTONEG_EXP, &autoneg_exp);
3568
3569                                 if (!(autoneg_exp & NWAY_ER_LP_NWAY_CAPS))
3570                                         e_info("Autonegotiated half duplex but"
3571                                                " link partner cannot autoneg. "
3572                                                " Try forcing full duplex if "
3573                                                "link gets many collisions.\n");
3574                         }
3575
3576                         /*
3577                          * tweak tx_queue_len according to speed/duplex
3578                          * and adjust the timeout factor
3579                          */
3580                         netdev->tx_queue_len = adapter->tx_queue_len;
3581                         adapter->tx_timeout_factor = 1;
3582                         switch (adapter->link_speed) {
3583                         case SPEED_10:
3584                                 txb2b = 0;
3585                                 netdev->tx_queue_len = 10;
3586                                 adapter->tx_timeout_factor = 16;
3587                                 break;
3588                         case SPEED_100:
3589                                 txb2b = 0;
3590                                 netdev->tx_queue_len = 100;
3591                                 /* maybe add some timeout factor ? */
3592                                 break;
3593                         }
3594
3595                         /*
3596                          * workaround: re-program speed mode bit after
3597                          * link-up event
3598                          */
3599                         if ((adapter->flags & FLAG_TARC_SPEED_MODE_BIT) &&
3600                             !txb2b) {
3601                                 u32 tarc0;
3602                                 tarc0 = er32(TARC(0));
3603                                 tarc0 &= ~SPEED_MODE_BIT;
3604                                 ew32(TARC(0), tarc0);
3605                         }
3606
3607                         /*
3608                          * disable TSO for pcie and 10/100 speeds, to avoid
3609                          * some hardware issues
3610                          */
3611                         if (!(adapter->flags & FLAG_TSO_FORCE)) {
3612                                 switch (adapter->link_speed) {
3613                                 case SPEED_10:
3614                                 case SPEED_100:
3615                                         e_info("10/100 speed: disabling TSO\n");
3616                                         netdev->features &= ~NETIF_F_TSO;
3617                                         netdev->features &= ~NETIF_F_TSO6;
3618                                         break;
3619                                 case SPEED_1000:
3620                                         netdev->features |= NETIF_F_TSO;
3621                                         netdev->features |= NETIF_F_TSO6;
3622                                         break;
3623                                 default:
3624                                         /* oops */
3625                                         break;
3626                                 }
3627                         }
3628
3629                         /*
3630                          * enable transmits in the hardware, need to do this
3631                          * after setting TARC(0)
3632                          */
3633                         tctl = er32(TCTL);
3634                         tctl |= E1000_TCTL_EN;
3635                         ew32(TCTL, tctl);
3636
3637                         /*
3638                          * Perform any post-link-up configuration before
3639                          * reporting link up.
3640                          */
3641                         if (phy->ops.cfg_on_link_up)
3642                                 phy->ops.cfg_on_link_up(hw);
3643
3644                         netif_carrier_on(netdev);
3645
3646                         if (!test_bit(__E1000_DOWN, &adapter->state))
3647                                 mod_timer(&adapter->phy_info_timer,
3648                                           round_jiffies(jiffies + 2 * HZ));
3649                 }
3650         } else {
3651                 if (netif_carrier_ok(netdev)) {
3652                         adapter->link_speed = 0;
3653                         adapter->link_duplex = 0;
3654                         /* Link status message must follow this format */
3655                         printk(KERN_INFO "e1000e: %s NIC Link is Down\n",
3656                                adapter->netdev->name);
3657                         netif_carrier_off(netdev);
3658                         if (!test_bit(__E1000_DOWN, &adapter->state))
3659                                 mod_timer(&adapter->phy_info_timer,
3660                                           round_jiffies(jiffies + 2 * HZ));
3661
3662                         if (adapter->flags & FLAG_RX_NEEDS_RESTART)
3663                                 schedule_work(&adapter->reset_task);
3664                 }
3665         }
3666
3667 link_up:
3668         e1000e_update_stats(adapter);
3669
3670         mac->tx_packet_delta = adapter->stats.tpt - adapter->tpt_old;
3671         adapter->tpt_old = adapter->stats.tpt;
3672         mac->collision_delta = adapter->stats.colc - adapter->colc_old;
3673         adapter->colc_old = adapter->stats.colc;
3674
3675         adapter->gorc = adapter->stats.gorc - adapter->gorc_old;
3676         adapter->gorc_old = adapter->stats.gorc;
3677         adapter->gotc = adapter->stats.gotc - adapter->gotc_old;
3678         adapter->gotc_old = adapter->stats.gotc;
3679
3680         e1000e_update_adaptive(&adapter->hw);
3681
3682         if (!netif_carrier_ok(netdev)) {
3683                 tx_pending = (e1000_desc_unused(tx_ring) + 1 <
3684                                tx_ring->count);
3685                 if (tx_pending) {
3686                         /*
3687                          * We've lost link, so the controller stops DMA,
3688                          * but we've got queued Tx work that's never going
3689                          * to get done, so reset controller to flush Tx.
3690                          * (Do the reset outside of interrupt context).
3691                          */
3692                         adapter->tx_timeout_count++;
3693                         schedule_work(&adapter->reset_task);
3694                         /* return immediately since reset is imminent */
3695                         return;
3696                 }
3697         }
3698
3699         /* Cause software interrupt to ensure Rx ring is cleaned */
3700         if (adapter->msix_entries)
3701                 ew32(ICS, adapter->rx_ring->ims_val);
3702         else
3703                 ew32(ICS, E1000_ICS_RXDMT0);
3704
3705         /* Force detection of hung controller every watchdog period */
3706         adapter->detect_tx_hung = 1;
3707
3708         /*
3709          * With 82571 controllers, LAA may be overwritten due to controller
3710          * reset from the other port. Set the appropriate LAA in RAR[0]
3711          */
3712         if (e1000e_get_laa_state_82571(hw))
3713                 e1000e_rar_set(hw, adapter->hw.mac.addr, 0);
3714
3715         /* Reset the timer */
3716         if (!test_bit(__E1000_DOWN, &adapter->state))
3717                 mod_timer(&adapter->watchdog_timer,
3718                           round_jiffies(jiffies + 2 * HZ));
3719 }
3720
3721 #define E1000_TX_FLAGS_CSUM             0x00000001
3722 #define E1000_TX_FLAGS_VLAN             0x00000002
3723 #define E1000_TX_FLAGS_TSO              0x00000004
3724 #define E1000_TX_FLAGS_IPV4             0x00000008
3725 #define E1000_TX_FLAGS_VLAN_MASK        0xffff0000
3726 #define E1000_TX_FLAGS_VLAN_SHIFT       16
3727
3728 static int e1000_tso(struct e1000_adapter *adapter,
3729                      struct sk_buff *skb)
3730 {
3731         struct e1000_ring *tx_ring = adapter->tx_ring;
3732         struct e1000_context_desc *context_desc;
3733         struct e1000_buffer *buffer_info;
3734         unsigned int i;
3735         u32 cmd_length = 0;
3736         u16 ipcse = 0, tucse, mss;
3737         u8 ipcss, ipcso, tucss, tucso, hdr_len;
3738         int err;
3739
3740         if (skb_is_gso(skb)) {
3741                 if (skb_header_cloned(skb)) {
3742                         err = pskb_expand_head(skb, 0, 0, GFP_ATOMIC);
3743                         if (err)
3744                                 return err;
3745                 }
3746
3747                 hdr_len = skb_transport_offset(skb) + tcp_hdrlen(skb);
3748                 mss = skb_shinfo(skb)->gso_size;
3749                 if (skb->protocol == htons(ETH_P_IP)) {
3750                         struct iphdr *iph = ip_hdr(skb);
3751                         iph->tot_len = 0;
3752                         iph->check = 0;
3753                         tcp_hdr(skb)->check = ~csum_tcpudp_magic(iph->saddr,
3754                                                                  iph->daddr, 0,
3755                                                                  IPPROTO_TCP,
3756                                                                  0);
3757                         cmd_length = E1000_TXD_CMD_IP;
3758                         ipcse = skb_transport_offset(skb) - 1;
3759                 } else if (skb_shinfo(skb)->gso_type == SKB_GSO_TCPV6) {
3760                         ipv6_hdr(skb)->payload_len = 0;
3761                         tcp_hdr(skb)->check =
3762                                 ~csum_ipv6_magic(&ipv6_hdr(skb)->saddr,
3763                                                  &ipv6_hdr(skb)->daddr,
3764                                                  0, IPPROTO_TCP, 0);
3765                         ipcse = 0;
3766                 }
3767                 ipcss = skb_network_offset(skb);
3768                 ipcso = (void *)&(ip_hdr(skb)->check) - (void *)skb->data;
3769                 tucss = skb_transport_offset(skb);
3770                 tucso = (void *)&(tcp_hdr(skb)->check) - (void *)skb->data;
3771                 tucse = 0;
3772
3773                 cmd_length |= (E1000_TXD_CMD_DEXT | E1000_TXD_CMD_TSE |
3774                                E1000_TXD_CMD_TCP | (skb->len - (hdr_len)));
3775
3776                 i = tx_ring->next_to_use;
3777                 context_desc = E1000_CONTEXT_DESC(*tx_ring, i);
3778                 buffer_info = &tx_ring->buffer_info[i];
3779
3780                 context_desc->lower_setup.ip_fields.ipcss  = ipcss;
3781                 context_desc->lower_setup.ip_fields.ipcso  = ipcso;
3782                 context_desc->lower_setup.ip_fields.ipcse  = cpu_to_le16(ipcse);
3783                 context_desc->upper_setup.tcp_fields.tucss = tucss;
3784                 context_desc->upper_setup.tcp_fields.tucso = tucso;
3785                 context_desc->upper_setup.tcp_fields.tucse = cpu_to_le16(tucse);
3786                 context_desc->tcp_seg_setup.fields.mss     = cpu_to_le16(mss);
3787                 context_desc->tcp_seg_setup.fields.hdr_len = hdr_len;
3788                 context_desc->cmd_and_length = cpu_to_le32(cmd_length);
3789
3790                 buffer_info->time_stamp = jiffies;
3791                 buffer_info->next_to_watch = i;
3792
3793                 i++;
3794                 if (i == tx_ring->count)
3795                         i = 0;
3796                 tx_ring->next_to_use = i;
3797
3798                 return 1;
3799         }
3800
3801         return 0;
3802 }
3803
3804 static bool e1000_tx_csum(struct e1000_adapter *adapter, struct sk_buff *skb)
3805 {
3806         struct e1000_ring *tx_ring = adapter->tx_ring;
3807         struct e1000_context_desc *context_desc;
3808         struct e1000_buffer *buffer_info;
3809         unsigned int i;
3810         u8 css;
3811         u32 cmd_len = E1000_TXD_CMD_DEXT;
3812         __be16 protocol;
3813
3814         if (skb->ip_summed != CHECKSUM_PARTIAL)
3815                 return 0;
3816
3817         if (skb->protocol == cpu_to_be16(ETH_P_8021Q))
3818                 protocol = vlan_eth_hdr(skb)->h_vlan_encapsulated_proto;
3819         else
3820                 protocol = skb->protocol;
3821
3822         switch (protocol) {
3823         case cpu_to_be16(ETH_P_IP):
3824                 if (ip_hdr(skb)->protocol == IPPROTO_TCP)
3825                         cmd_len |= E1000_TXD_CMD_TCP;
3826                 break;
3827         case cpu_to_be16(ETH_P_IPV6):
3828                 /* XXX not handling all IPV6 headers */
3829                 if (ipv6_hdr(skb)->nexthdr == IPPROTO_TCP)
3830                         cmd_len |= E1000_TXD_CMD_TCP;
3831                 break;
3832         default:
3833                 if (unlikely(net_ratelimit()))
3834                         e_warn("checksum_partial proto=%x!\n",
3835                                be16_to_cpu(protocol));
3836                 break;
3837         }
3838
3839         css = skb_transport_offset(skb);
3840
3841         i = tx_ring->next_to_use;
3842         buffer_info = &tx_ring->buffer_info[i];
3843         context_desc = E1000_CONTEXT_DESC(*tx_ring, i);
3844
3845         context_desc->lower_setup.ip_config = 0;
3846         context_desc->upper_setup.tcp_fields.tucss = css;
3847         context_desc->upper_setup.tcp_fields.tucso =
3848                                 css + skb->csum_offset;
3849         context_desc->upper_setup.tcp_fields.tucse = 0;
3850         context_desc->tcp_seg_setup.data = 0;
3851         context_desc->cmd_and_length = cpu_to_le32(cmd_len);
3852
3853         buffer_info->time_stamp = jiffies;
3854         buffer_info->next_to_watch = i;
3855
3856         i++;
3857         if (i == tx_ring->count)
3858                 i = 0;
3859         tx_ring->next_to_use = i;
3860
3861         return 1;
3862 }
3863
3864 #define E1000_MAX_PER_TXD       8192
3865 #define E1000_MAX_TXD_PWR       12
3866
3867 static int e1000_tx_map(struct e1000_adapter *adapter,
3868                         struct sk_buff *skb, unsigned int first,
3869                         unsigned int max_per_txd, unsigned int nr_frags,
3870                         unsigned int mss)
3871 {
3872         struct e1000_ring *tx_ring = adapter->tx_ring;
3873         struct e1000_buffer *buffer_info;
3874         unsigned int len = skb_headlen(skb);
3875         unsigned int offset, size, count = 0, i;
3876         unsigned int f;
3877         dma_addr_t *map;
3878
3879         i = tx_ring->next_to_use;
3880
3881         if (skb_dma_map(&adapter->pdev->dev, skb, DMA_TO_DEVICE)) {
3882                 dev_err(&adapter->pdev->dev, "TX DMA map failed\n");
3883                 adapter->tx_dma_failed++;
3884                 return 0;
3885         }
3886
3887         map = skb_shinfo(skb)->dma_maps;
3888         offset = 0;
3889
3890         while (len) {
3891                 buffer_info = &tx_ring->buffer_info[i];
3892                 size = min(len, max_per_txd);
3893
3894                 buffer_info->length = size;
3895                 buffer_info->time_stamp = jiffies;
3896                 buffer_info->next_to_watch = i;
3897                 buffer_info->dma = skb_shinfo(skb)->dma_head + offset;
3898                 count++;
3899
3900                 len -= size;
3901                 offset += size;
3902
3903                 if (len) {
3904                         i++;
3905                         if (i == tx_ring->count)
3906                                 i = 0;
3907                 }
3908         }
3909
3910         for (f = 0; f < nr_frags; f++) {
3911                 struct skb_frag_struct *frag;
3912
3913                 frag = &skb_shinfo(skb)->frags[f];
3914                 len = frag->size;
3915                 offset = 0;
3916
3917                 while (len) {
3918                         i++;
3919                         if (i == tx_ring->count)
3920                                 i = 0;
3921
3922                         buffer_info = &tx_ring->buffer_info[i];
3923                         size = min(len, max_per_txd);
3924
3925                         buffer_info->length = size;
3926                         buffer_info->time_stamp = jiffies;
3927                         buffer_info->next_to_watch = i;
3928                         buffer_info->dma = map[f] + offset;
3929
3930                         len -= size;
3931                         offset += size;
3932                         count++;
3933                 }
3934         }
3935
3936         tx_ring->buffer_info[i].skb = skb;
3937         tx_ring->buffer_info[first].next_to_watch = i;
3938
3939         return count;
3940 }
3941
3942 static void e1000_tx_queue(struct e1000_adapter *adapter,
3943                            int tx_flags, int count)
3944 {
3945         struct e1000_ring *tx_ring = adapter->tx_ring;
3946         struct e1000_tx_desc *tx_desc = NULL;
3947         struct e1000_buffer *buffer_info;
3948         u32 txd_upper = 0, txd_lower = E1000_TXD_CMD_IFCS;
3949         unsigned int i;
3950
3951         if (tx_flags & E1000_TX_FLAGS_TSO) {
3952                 txd_lower |= E1000_TXD_CMD_DEXT | E1000_TXD_DTYP_D |
3953                              E1000_TXD_CMD_TSE;
3954                 txd_upper |= E1000_TXD_POPTS_TXSM << 8;
3955
3956                 if (tx_flags & E1000_TX_FLAGS_IPV4)
3957                         txd_upper |= E1000_TXD_POPTS_IXSM << 8;
3958         }
3959
3960         if (tx_flags & E1000_TX_FLAGS_CSUM) {
3961                 txd_lower |= E1000_TXD_CMD_DEXT | E1000_TXD_DTYP_D;
3962                 txd_upper |= E1000_TXD_POPTS_TXSM << 8;
3963         }
3964
3965         if (tx_flags & E1000_TX_FLAGS_VLAN) {
3966                 txd_lower |= E1000_TXD_CMD_VLE;
3967                 txd_upper |= (tx_flags & E1000_TX_FLAGS_VLAN_MASK);
3968         }
3969
3970         i = tx_ring->next_to_use;
3971
3972         while (count--) {
3973                 buffer_info = &tx_ring->buffer_info[i];
3974                 tx_desc = E1000_TX_DESC(*tx_ring, i);
3975                 tx_desc->buffer_addr = cpu_to_le64(buffer_info->dma);
3976                 tx_desc->lower.data =
3977                         cpu_to_le32(txd_lower | buffer_info->length);
3978                 tx_desc->upper.data = cpu_to_le32(txd_upper);
3979
3980                 i++;
3981                 if (i == tx_ring->count)
3982                         i = 0;
3983         }
3984
3985         tx_desc->lower.data |= cpu_to_le32(adapter->txd_cmd);
3986
3987         /*
3988          * Force memory writes to complete before letting h/w
3989          * know there are new descriptors to fetch.  (Only
3990          * applicable for weak-ordered memory model archs,
3991          * such as IA-64).
3992          */
3993         wmb();
3994
3995         tx_ring->next_to_use = i;
3996         writel(i, adapter->hw.hw_addr + tx_ring->tail);
3997         /*
3998          * we need this if more than one processor can write to our tail
3999          * at a time, it synchronizes IO on IA64/Altix systems
4000          */
4001         mmiowb();
4002 }
4003
4004 #define MINIMUM_DHCP_PACKET_SIZE 282
4005 static int e1000_transfer_dhcp_info(struct e1000_adapter *adapter,
4006                                     struct sk_buff *skb)
4007 {
4008         struct e1000_hw *hw =  &adapter->hw;
4009         u16 length, offset;
4010
4011         if (vlan_tx_tag_present(skb)) {
4012                 if (!((vlan_tx_tag_get(skb) == adapter->hw.mng_cookie.vlan_id)
4013                     && (adapter->hw.mng_cookie.status &
4014                         E1000_MNG_DHCP_COOKIE_STATUS_VLAN)))
4015                         return 0;
4016         }
4017
4018         if (skb->len <= MINIMUM_DHCP_PACKET_SIZE)
4019                 return 0;
4020
4021         if (((struct ethhdr *) skb->data)->h_proto != htons(ETH_P_IP))
4022                 return 0;
4023
4024         {
4025                 const struct iphdr *ip = (struct iphdr *)((u8 *)skb->data+14);
4026                 struct udphdr *udp;
4027
4028                 if (ip->protocol != IPPROTO_UDP)
4029                         return 0;
4030
4031                 udp = (struct udphdr *)((u8 *)ip + (ip->ihl << 2));
4032                 if (ntohs(udp->dest) != 67)
4033                         return 0;
4034
4035                 offset = (u8 *)udp + 8 - skb->data;
4036                 length = skb->len - offset;
4037                 return e1000e_mng_write_dhcp_info(hw, (u8 *)udp + 8, length);
4038         }
4039
4040         return 0;
4041 }
4042
4043 static int __e1000_maybe_stop_tx(struct net_device *netdev, int size)
4044 {
4045         struct e1000_adapter *adapter = netdev_priv(netdev);
4046
4047         netif_stop_queue(netdev);
4048         /*
4049          * Herbert's original patch had:
4050          *  smp_mb__after_netif_stop_queue();
4051          * but since that doesn't exist yet, just open code it.
4052          */
4053         smp_mb();
4054
4055         /*
4056          * We need to check again in a case another CPU has just
4057          * made room available.
4058          */
4059         if (e1000_desc_unused(adapter->tx_ring) < size)
4060                 return -EBUSY;
4061
4062         /* A reprieve! */
4063         netif_start_queue(netdev);
4064         ++adapter->restart_queue;
4065         return 0;
4066 }
4067
4068 static int e1000_maybe_stop_tx(struct net_device *netdev, int size)
4069 {
4070         struct e1000_adapter *adapter = netdev_priv(netdev);
4071
4072         if (e1000_desc_unused(adapter->tx_ring) >= size)
4073                 return 0;
4074         return __e1000_maybe_stop_tx(netdev, size);
4075 }
4076
4077 #define TXD_USE_COUNT(S, X) (((S) >> (X)) + 1 )
4078 static netdev_tx_t e1000_xmit_frame(struct sk_buff *skb,
4079                                     struct net_device *netdev)
4080 {
4081         struct e1000_adapter *adapter = netdev_priv(netdev);
4082         struct e1000_ring *tx_ring = adapter->tx_ring;
4083         unsigned int first;
4084         unsigned int max_per_txd = E1000_MAX_PER_TXD;
4085         unsigned int max_txd_pwr = E1000_MAX_TXD_PWR;
4086         unsigned int tx_flags = 0;
4087         unsigned int len = skb->len - skb->data_len;
4088         unsigned int nr_frags;
4089         unsigned int mss;
4090         int count = 0;
4091         int tso;
4092         unsigned int f;
4093
4094         if (test_bit(__E1000_DOWN, &adapter->state)) {
4095                 dev_kfree_skb_any(skb);
4096                 return NETDEV_TX_OK;
4097         }
4098
4099         if (skb->len <= 0) {
4100                 dev_kfree_skb_any(skb);
4101                 return NETDEV_TX_OK;
4102         }
4103
4104         mss = skb_shinfo(skb)->gso_size;
4105         /*
4106          * The controller does a simple calculation to
4107          * make sure there is enough room in the FIFO before
4108          * initiating the DMA for each buffer.  The calc is:
4109          * 4 = ceil(buffer len/mss).  To make sure we don't
4110          * overrun the FIFO, adjust the max buffer len if mss
4111          * drops.
4112          */
4113         if (mss) {
4114                 u8 hdr_len;
4115                 max_per_txd = min(mss << 2, max_per_txd);
4116                 max_txd_pwr = fls(max_per_txd) - 1;
4117
4118                 /*
4119                  * TSO Workaround for 82571/2/3 Controllers -- if skb->data
4120                  * points to just header, pull a few bytes of payload from
4121                  * frags into skb->data
4122                  */
4123                 hdr_len = skb_transport_offset(skb) + tcp_hdrlen(skb);
4124                 /*
4125                  * we do this workaround for ES2LAN, but it is un-necessary,
4126                  * avoiding it could save a lot of cycles
4127                  */
4128                 if (skb->data_len && (hdr_len == len)) {
4129                         unsigned int pull_size;
4130
4131                         pull_size = min((unsigned int)4, skb->data_len);
4132                         if (!__pskb_pull_tail(skb, pull_size)) {
4133                                 e_err("__pskb_pull_tail failed.\n");
4134                                 dev_kfree_skb_any(skb);
4135                                 return NETDEV_TX_OK;
4136                         }
4137                         len = skb->len - skb->data_len;
4138                 }
4139         }
4140
4141         /* reserve a descriptor for the offload context */
4142         if ((mss) || (skb->ip_summed == CHECKSUM_PARTIAL))
4143                 count++;
4144         count++;
4145
4146         count += TXD_USE_COUNT(len, max_txd_pwr);
4147
4148         nr_frags = skb_shinfo(skb)->nr_frags;
4149         for (f = 0; f < nr_frags; f++)
4150                 count += TXD_USE_COUNT(skb_shinfo(skb)->frags[f].size,
4151                                        max_txd_pwr);
4152
4153         if (adapter->hw.mac.tx_pkt_filtering)
4154                 e1000_transfer_dhcp_info(adapter, skb);
4155
4156         /*
4157          * need: count + 2 desc gap to keep tail from touching
4158          * head, otherwise try next time
4159          */
4160         if (e1000_maybe_stop_tx(netdev, count + 2))
4161                 return NETDEV_TX_BUSY;
4162
4163         if (adapter->vlgrp && vlan_tx_tag_present(skb)) {
4164                 tx_flags |= E1000_TX_FLAGS_VLAN;
4165                 tx_flags |= (vlan_tx_tag_get(skb) << E1000_TX_FLAGS_VLAN_SHIFT);
4166         }
4167
4168         first = tx_ring->next_to_use;
4169
4170         tso = e1000_tso(adapter, skb);
4171         if (tso < 0) {
4172                 dev_kfree_skb_any(skb);
4173                 return NETDEV_TX_OK;
4174         }
4175
4176         if (tso)
4177                 tx_flags |= E1000_TX_FLAGS_TSO;
4178         else if (e1000_tx_csum(adapter, skb))
4179                 tx_flags |= E1000_TX_FLAGS_CSUM;
4180
4181         /*
4182          * Old method was to assume IPv4 packet by default if TSO was enabled.
4183          * 82571 hardware supports TSO capabilities for IPv6 as well...
4184          * no longer assume, we must.
4185          */
4186         if (skb->protocol == htons(ETH_P_IP))
4187                 tx_flags |= E1000_TX_FLAGS_IPV4;
4188
4189         /* if count is 0 then mapping error has occured */
4190         count = e1000_tx_map(adapter, skb, first, max_per_txd, nr_frags, mss);
4191         if (count) {
4192                 e1000_tx_queue(adapter, tx_flags, count);
4193                 /* Make sure there is space in the ring for the next send. */
4194                 e1000_maybe_stop_tx(netdev, MAX_SKB_FRAGS + 2);
4195
4196         } else {
4197                 dev_kfree_skb_any(skb);
4198                 tx_ring->buffer_info[first].time_stamp = 0;
4199                 tx_ring->next_to_use = first;
4200         }
4201
4202         return NETDEV_TX_OK;
4203 }
4204
4205 /**
4206  * e1000_tx_timeout - Respond to a Tx Hang
4207  * @netdev: network interface device structure
4208  **/
4209 static void e1000_tx_timeout(struct net_device *netdev)
4210 {
4211         struct e1000_adapter *adapter = netdev_priv(netdev);
4212
4213         /* Do the reset outside of interrupt context */
4214         adapter->tx_timeout_count++;
4215         schedule_work(&adapter->reset_task);
4216 }
4217
4218 static void e1000_reset_task(struct work_struct *work)
4219 {
4220         struct e1000_adapter *adapter;
4221         adapter = container_of(work, struct e1000_adapter, reset_task);
4222
4223         e1000e_reinit_locked(adapter);
4224 }
4225
4226 /**
4227  * e1000_get_stats - Get System Network Statistics
4228  * @netdev: network interface device structure
4229  *
4230  * Returns the address of the device statistics structure.
4231  * The statistics are actually updated from the timer callback.
4232  **/
4233 static struct net_device_stats *e1000_get_stats(struct net_device *netdev)
4234 {
4235         /* only return the current stats */
4236         return &netdev->stats;
4237 }
4238
4239 /**
4240  * e1000_change_mtu - Change the Maximum Transfer Unit
4241  * @netdev: network interface device structure
4242  * @new_mtu: new value for maximum frame size
4243  *
4244  * Returns 0 on success, negative on failure
4245  **/
4246 static int e1000_change_mtu(struct net_device *netdev, int new_mtu)
4247 {
4248         struct e1000_adapter *adapter = netdev_priv(netdev);
4249         int max_frame = new_mtu + ETH_HLEN + ETH_FCS_LEN;
4250
4251         /* Jumbo frame support */
4252         if ((max_frame > ETH_FRAME_LEN + ETH_FCS_LEN) &&
4253             !(adapter->flags & FLAG_HAS_JUMBO_FRAMES)) {
4254                 e_err("Jumbo Frames not supported.\n");
4255                 return -EINVAL;
4256         }
4257
4258         /* Supported frame sizes */
4259         if ((new_mtu < ETH_ZLEN + ETH_FCS_LEN + VLAN_HLEN) ||
4260             (max_frame > adapter->max_hw_frame_size)) {
4261                 e_err("Unsupported MTU setting\n");
4262                 return -EINVAL;
4263         }
4264
4265         while (test_and_set_bit(__E1000_RESETTING, &adapter->state))
4266                 msleep(1);
4267         /* e1000e_down has a dependency on max_frame_size */
4268         adapter->max_frame_size = max_frame;
4269         if (netif_running(netdev))
4270                 e1000e_down(adapter);
4271
4272         /*
4273          * NOTE: netdev_alloc_skb reserves 16 bytes, and typically NET_IP_ALIGN
4274          * means we reserve 2 more, this pushes us to allocate from the next
4275          * larger slab size.
4276          * i.e. RXBUFFER_2048 --> size-4096 slab
4277          * However with the new *_jumbo_rx* routines, jumbo receives will use
4278          * fragmented skbs
4279          */
4280
4281         if (max_frame <= 256)
4282                 adapter->rx_buffer_len = 256;
4283         else if (max_frame <= 512)
4284                 adapter->rx_buffer_len = 512;
4285         else if (max_frame <= 1024)
4286                 adapter->rx_buffer_len = 1024;
4287         else if (max_frame <= 2048)
4288                 adapter->rx_buffer_len = 2048;
4289         else
4290                 adapter->rx_buffer_len = 4096;
4291
4292         /* adjust allocation if LPE protects us, and we aren't using SBP */
4293         if ((max_frame == ETH_FRAME_LEN + ETH_FCS_LEN) ||
4294              (max_frame == ETH_FRAME_LEN + VLAN_HLEN + ETH_FCS_LEN))
4295                 adapter->rx_buffer_len = ETH_FRAME_LEN + VLAN_HLEN
4296                                          + ETH_FCS_LEN;
4297
4298         e_info("changing MTU from %d to %d\n", netdev->mtu, new_mtu);
4299         netdev->mtu = new_mtu;
4300
4301         if (netif_running(netdev))
4302                 e1000e_up(adapter);
4303         else
4304                 e1000e_reset(adapter);
4305
4306         clear_bit(__E1000_RESETTING, &adapter->state);
4307
4308         return 0;
4309 }
4310
4311 static int e1000_mii_ioctl(struct net_device *netdev, struct ifreq *ifr,
4312                            int cmd)
4313 {
4314         struct e1000_adapter *adapter = netdev_priv(netdev);
4315         struct mii_ioctl_data *data = if_mii(ifr);
4316
4317         if (adapter->hw.phy.media_type != e1000_media_type_copper)
4318                 return -EOPNOTSUPP;
4319
4320         switch (cmd) {
4321         case SIOCGMIIPHY:
4322                 data->phy_id = adapter->hw.phy.addr;
4323                 break;
4324         case SIOCGMIIREG:
4325                 switch (data->reg_num & 0x1F) {
4326                 case MII_BMCR:
4327                         data->val_out = adapter->phy_regs.bmcr;
4328                         break;
4329                 case MII_BMSR:
4330                         data->val_out = adapter->phy_regs.bmsr;
4331                         break;
4332                 case MII_PHYSID1:
4333                         data->val_out = (adapter->hw.phy.id >> 16);
4334                         break;
4335                 case MII_PHYSID2:
4336                         data->val_out = (adapter->hw.phy.id & 0xFFFF);
4337                         break;
4338                 case MII_ADVERTISE:
4339                         data->val_out = adapter->phy_regs.advertise;
4340                         break;
4341                 case MII_LPA:
4342                         data->val_out = adapter->phy_regs.lpa;
4343                         break;
4344                 case MII_EXPANSION:
4345                         data->val_out = adapter->phy_regs.expansion;
4346                         break;
4347                 case MII_CTRL1000:
4348                         data->val_out = adapter->phy_regs.ctrl1000;
4349                         break;
4350                 case MII_STAT1000:
4351                         data->val_out = adapter->phy_regs.stat1000;
4352                         break;
4353                 case MII_ESTATUS:
4354                         data->val_out = adapter->phy_regs.estatus;
4355                         break;
4356                 default:
4357                         return -EIO;
4358                 }
4359                 break;
4360         case SIOCSMIIREG:
4361         default:
4362                 return -EOPNOTSUPP;
4363         }
4364         return 0;
4365 }
4366
4367 static int e1000_ioctl(struct net_device *netdev, struct ifreq *ifr, int cmd)
4368 {
4369         switch (cmd) {
4370         case SIOCGMIIPHY:
4371         case SIOCGMIIREG:
4372         case SIOCSMIIREG:
4373                 return e1000_mii_ioctl(netdev, ifr, cmd);
4374         default:
4375                 return -EOPNOTSUPP;
4376         }
4377 }
4378
4379 static int e1000_init_phy_wakeup(struct e1000_adapter *adapter, u32 wufc)
4380 {
4381         struct e1000_hw *hw = &adapter->hw;
4382         u32 i, mac_reg;
4383         u16 phy_reg;
4384         int retval = 0;
4385
4386         /* copy MAC RARs to PHY RARs */
4387         for (i = 0; i < adapter->hw.mac.rar_entry_count; i++) {
4388                 mac_reg = er32(RAL(i));
4389                 e1e_wphy(hw, BM_RAR_L(i), (u16)(mac_reg & 0xFFFF));
4390                 e1e_wphy(hw, BM_RAR_M(i), (u16)((mac_reg >> 16) & 0xFFFF));
4391                 mac_reg = er32(RAH(i));
4392                 e1e_wphy(hw, BM_RAR_H(i), (u16)(mac_reg & 0xFFFF));
4393                 e1e_wphy(hw, BM_RAR_CTRL(i), (u16)((mac_reg >> 16) & 0xFFFF));
4394         }
4395
4396         /* copy MAC MTA to PHY MTA */
4397         for (i = 0; i < adapter->hw.mac.mta_reg_count; i++) {
4398                 mac_reg = E1000_READ_REG_ARRAY(hw, E1000_MTA, i);
4399                 e1e_wphy(hw, BM_MTA(i), (u16)(mac_reg & 0xFFFF));
4400                 e1e_wphy(hw, BM_MTA(i) + 1, (u16)((mac_reg >> 16) & 0xFFFF));
4401         }
4402
4403         /* configure PHY Rx Control register */
4404         e1e_rphy(&adapter->hw, BM_RCTL, &phy_reg);
4405         mac_reg = er32(RCTL);
4406         if (mac_reg & E1000_RCTL_UPE)
4407                 phy_reg |= BM_RCTL_UPE;
4408         if (mac_reg & E1000_RCTL_MPE)
4409                 phy_reg |= BM_RCTL_MPE;
4410         phy_reg &= ~(BM_RCTL_MO_MASK);
4411         if (mac_reg & E1000_RCTL_MO_3)
4412                 phy_reg |= (((mac_reg & E1000_RCTL_MO_3) >> E1000_RCTL_MO_SHIFT)
4413                                 << BM_RCTL_MO_SHIFT);
4414         if (mac_reg & E1000_RCTL_BAM)
4415                 phy_reg |= BM_RCTL_BAM;
4416         if (mac_reg & E1000_RCTL_PMCF)
4417                 phy_reg |= BM_RCTL_PMCF;
4418         mac_reg = er32(CTRL);
4419         if (mac_reg & E1000_CTRL_RFCE)
4420                 phy_reg |= BM_RCTL_RFCE;
4421         e1e_wphy(&adapter->hw, BM_RCTL, phy_reg);
4422
4423         /* enable PHY wakeup in MAC register */
4424         ew32(WUFC, wufc);
4425         ew32(WUC, E1000_WUC_PHY_WAKE | E1000_WUC_PME_EN);
4426
4427         /* configure and enable PHY wakeup in PHY registers */
4428         e1e_wphy(&adapter->hw, BM_WUFC, wufc);
4429         e1e_wphy(&adapter->hw, BM_WUC, E1000_WUC_PME_EN);
4430
4431         /* activate PHY wakeup */
4432         retval = hw->phy.ops.acquire_phy(hw);
4433         if (retval) {
4434                 e_err("Could not acquire PHY\n");
4435                 return retval;
4436         }
4437         e1000e_write_phy_reg_mdic(hw, IGP01E1000_PHY_PAGE_SELECT,
4438                                  (BM_WUC_ENABLE_PAGE << IGP_PAGE_SHIFT));
4439         retval = e1000e_read_phy_reg_mdic(hw, BM_WUC_ENABLE_REG, &phy_reg);
4440         if (retval) {
4441                 e_err("Could not read PHY page 769\n");
4442                 goto out;
4443         }
4444         phy_reg |= BM_WUC_ENABLE_BIT | BM_WUC_HOST_WU_BIT;
4445         retval = e1000e_write_phy_reg_mdic(hw, BM_WUC_ENABLE_REG, phy_reg);
4446         if (retval)
4447                 e_err("Could not set PHY Host Wakeup bit\n");
4448 out:
4449         hw->phy.ops.release_phy(hw);
4450
4451         return retval;
4452 }
4453
4454 static int __e1000_shutdown(struct pci_dev *pdev, bool *enable_wake)
4455 {
4456         struct net_device *netdev = pci_get_drvdata(pdev);
4457         struct e1000_adapter *adapter = netdev_priv(netdev);
4458         struct e1000_hw *hw = &adapter->hw;
4459         u32 ctrl, ctrl_ext, rctl, status;
4460         u32 wufc = adapter->wol;
4461         int retval = 0;
4462
4463         netif_device_detach(netdev);
4464
4465         if (netif_running(netdev)) {
4466                 WARN_ON(test_bit(__E1000_RESETTING, &adapter->state));
4467                 e1000e_down(adapter);
4468                 e1000_free_irq(adapter);
4469         }
4470         e1000e_reset_interrupt_capability(adapter);
4471
4472         retval = pci_save_state(pdev);
4473         if (retval)
4474                 return retval;
4475
4476         status = er32(STATUS);
4477         if (status & E1000_STATUS_LU)
4478                 wufc &= ~E1000_WUFC_LNKC;
4479
4480         if (wufc) {
4481                 e1000_setup_rctl(adapter);
4482                 e1000_set_multi(netdev);
4483
4484                 /* turn on all-multi mode if wake on multicast is enabled */
4485                 if (wufc & E1000_WUFC_MC) {
4486                         rctl = er32(RCTL);
4487                         rctl |= E1000_RCTL_MPE;
4488                         ew32(RCTL, rctl);
4489                 }
4490
4491                 ctrl = er32(CTRL);
4492                 /* advertise wake from D3Cold */
4493                 #define E1000_CTRL_ADVD3WUC 0x00100000
4494                 /* phy power management enable */
4495                 #define E1000_CTRL_EN_PHY_PWR_MGMT 0x00200000
4496                 ctrl |= E1000_CTRL_ADVD3WUC;
4497                 if (!(adapter->flags2 & FLAG2_HAS_PHY_WAKEUP))
4498                         ctrl |= E1000_CTRL_EN_PHY_PWR_MGMT;
4499                 ew32(CTRL, ctrl);
4500
4501                 if (adapter->hw.phy.media_type == e1000_media_type_fiber ||
4502                     adapter->hw.phy.media_type ==
4503                     e1000_media_type_internal_serdes) {
4504                         /* keep the laser running in D3 */
4505                         ctrl_ext = er32(CTRL_EXT);
4506                         ctrl_ext |= E1000_CTRL_EXT_SDP7_DATA;
4507                         ew32(CTRL_EXT, ctrl_ext);
4508                 }
4509
4510                 if (adapter->flags & FLAG_IS_ICH)
4511                         e1000e_disable_gig_wol_ich8lan(&adapter->hw);
4512
4513                 /* Allow time for pending master requests to run */
4514                 e1000e_disable_pcie_master(&adapter->hw);
4515
4516                 if (adapter->flags2 & FLAG2_HAS_PHY_WAKEUP) {
4517                         /* enable wakeup by the PHY */
4518                         retval = e1000_init_phy_wakeup(adapter, wufc);
4519                         if (retval)
4520                                 return retval;
4521                 } else {
4522                         /* enable wakeup by the MAC */
4523                         ew32(WUFC, wufc);
4524                         ew32(WUC, E1000_WUC_PME_EN);
4525                 }
4526         } else {
4527                 ew32(WUC, 0);
4528                 ew32(WUFC, 0);
4529         }
4530
4531         *enable_wake = !!wufc;
4532
4533         /* make sure adapter isn't asleep if manageability is enabled */
4534         if ((adapter->flags & FLAG_MNG_PT_ENABLED) ||
4535             (hw->mac.ops.check_mng_mode(hw)))
4536                 *enable_wake = true;
4537
4538         if (adapter->hw.phy.type == e1000_phy_igp_3)
4539                 e1000e_igp3_phy_powerdown_workaround_ich8lan(&adapter->hw);
4540
4541         /*
4542          * Release control of h/w to f/w.  If f/w is AMT enabled, this
4543          * would have already happened in close and is redundant.
4544          */
4545         e1000_release_hw_control(adapter);
4546
4547         pci_disable_device(pdev);
4548
4549         return 0;
4550 }
4551
4552 static void e1000_power_off(struct pci_dev *pdev, bool sleep, bool wake)
4553 {
4554         if (sleep && wake) {
4555                 pci_prepare_to_sleep(pdev);
4556                 return;
4557         }
4558
4559         pci_wake_from_d3(pdev, wake);
4560         pci_set_power_state(pdev, PCI_D3hot);
4561 }
4562
4563 static void e1000_complete_shutdown(struct pci_dev *pdev, bool sleep,
4564                                     bool wake)
4565 {
4566         struct net_device *netdev = pci_get_drvdata(pdev);
4567         struct e1000_adapter *adapter = netdev_priv(netdev);
4568
4569         /*
4570          * The pci-e switch on some quad port adapters will report a
4571          * correctable error when the MAC transitions from D0 to D3.  To
4572          * prevent this we need to mask off the correctable errors on the
4573          * downstream port of the pci-e switch.
4574          */
4575         if (adapter->flags & FLAG_IS_QUAD_PORT) {
4576                 struct pci_dev *us_dev = pdev->bus->self;
4577                 int pos = pci_find_capability(us_dev, PCI_CAP_ID_EXP);
4578                 u16 devctl;
4579
4580                 pci_read_config_word(us_dev, pos + PCI_EXP_DEVCTL, &devctl);
4581                 pci_write_config_word(us_dev, pos + PCI_EXP_DEVCTL,
4582                                       (devctl & ~PCI_EXP_DEVCTL_CERE));
4583
4584                 e1000_power_off(pdev, sleep, wake);
4585
4586                 pci_write_config_word(us_dev, pos + PCI_EXP_DEVCTL, devctl);
4587         } else {
4588                 e1000_power_off(pdev, sleep, wake);
4589         }
4590 }
4591
4592 static void e1000e_disable_l1aspm(struct pci_dev *pdev)
4593 {
4594         int pos;
4595         u16 val;
4596
4597         /*
4598          * 82573 workaround - disable L1 ASPM on mobile chipsets
4599          *
4600          * L1 ASPM on various mobile (ich7) chipsets do not behave properly
4601          * resulting in lost data or garbage information on the pci-e link
4602          * level. This could result in (false) bad EEPROM checksum errors,
4603          * long ping times (up to 2s) or even a system freeze/hang.
4604          *
4605          * Unfortunately this feature saves about 1W power consumption when
4606          * active.
4607          */
4608         pos = pci_find_capability(pdev, PCI_CAP_ID_EXP);
4609         pci_read_config_word(pdev, pos + PCI_EXP_LNKCTL, &val);
4610         if (val & 0x2) {
4611                 dev_warn(&pdev->dev, "Disabling L1 ASPM\n");
4612                 val &= ~0x2;
4613                 pci_write_config_word(pdev, pos + PCI_EXP_LNKCTL, val);
4614         }
4615 }
4616
4617 #ifdef CONFIG_PM
4618 static int e1000_suspend(struct pci_dev *pdev, pm_message_t state)
4619 {
4620         int retval;
4621         bool wake;
4622
4623         retval = __e1000_shutdown(pdev, &wake);
4624         if (!retval)
4625                 e1000_complete_shutdown(pdev, true, wake);
4626
4627         return retval;
4628 }
4629
4630 static int e1000_resume(struct pci_dev *pdev)
4631 {
4632         struct net_device *netdev = pci_get_drvdata(pdev);
4633         struct e1000_adapter *adapter = netdev_priv(netdev);
4634         struct e1000_hw *hw = &adapter->hw;
4635         u32 err;
4636
4637         pci_set_power_state(pdev, PCI_D0);
4638         pci_restore_state(pdev);
4639         e1000e_disable_l1aspm(pdev);
4640
4641         err = pci_enable_device_mem(pdev);
4642         if (err) {
4643                 dev_err(&pdev->dev,
4644                         "Cannot enable PCI device from suspend\n");
4645                 return err;
4646         }
4647
4648         pci_set_master(pdev);
4649
4650         pci_enable_wake(pdev, PCI_D3hot, 0);
4651         pci_enable_wake(pdev, PCI_D3cold, 0);
4652
4653         e1000e_set_interrupt_capability(adapter);
4654         if (netif_running(netdev)) {
4655                 err = e1000_request_irq(adapter);
4656                 if (err)
4657                         return err;
4658         }
4659
4660         e1000e_power_up_phy(adapter);
4661
4662         /* report the system wakeup cause from S3/S4 */
4663         if (adapter->flags2 & FLAG2_HAS_PHY_WAKEUP) {
4664                 u16 phy_data;
4665
4666                 e1e_rphy(&adapter->hw, BM_WUS, &phy_data);
4667                 if (phy_data) {
4668                         e_info("PHY Wakeup cause - %s\n",
4669                                 phy_data & E1000_WUS_EX ? "Unicast Packet" :
4670                                 phy_data & E1000_WUS_MC ? "Multicast Packet" :
4671                                 phy_data & E1000_WUS_BC ? "Broadcast Packet" :
4672                                 phy_data & E1000_WUS_MAG ? "Magic Packet" :
4673                                 phy_data & E1000_WUS_LNKC ? "Link Status "
4674                                 " Change" : "other");
4675                 }
4676                 e1e_wphy(&adapter->hw, BM_WUS, ~0);
4677         } else {
4678                 u32 wus = er32(WUS);
4679                 if (wus) {
4680                         e_info("MAC Wakeup cause - %s\n",
4681                                 wus & E1000_WUS_EX ? "Unicast Packet" :
4682                                 wus & E1000_WUS_MC ? "Multicast Packet" :
4683                                 wus & E1000_WUS_BC ? "Broadcast Packet" :
4684                                 wus & E1000_WUS_MAG ? "Magic Packet" :
4685                                 wus & E1000_WUS_LNKC ? "Link Status Change" :
4686                                 "other");
4687                 }
4688                 ew32(WUS, ~0);
4689         }
4690
4691         e1000e_reset(adapter);
4692
4693         e1000_init_manageability(adapter);
4694
4695         if (netif_running(netdev))
4696                 e1000e_up(adapter);
4697
4698         netif_device_attach(netdev);
4699
4700         /*
4701          * If the controller has AMT, do not set DRV_LOAD until the interface
4702          * is up.  For all other cases, let the f/w know that the h/w is now
4703          * under the control of the driver.
4704          */
4705         if (!(adapter->flags & FLAG_HAS_AMT))
4706                 e1000_get_hw_control(adapter);
4707
4708         return 0;
4709 }
4710 #endif
4711
4712 static void e1000_shutdown(struct pci_dev *pdev)
4713 {
4714         bool wake = false;
4715
4716         __e1000_shutdown(pdev, &wake);
4717
4718         if (system_state == SYSTEM_POWER_OFF)
4719                 e1000_complete_shutdown(pdev, false, wake);
4720 }
4721
4722 #ifdef CONFIG_NET_POLL_CONTROLLER
4723 /*
4724  * Polling 'interrupt' - used by things like netconsole to send skbs
4725  * without having to re-enable interrupts. It's not called while
4726  * the interrupt routine is executing.
4727  */
4728 static void e1000_netpoll(struct net_device *netdev)
4729 {
4730         struct e1000_adapter *adapter = netdev_priv(netdev);
4731
4732         disable_irq(adapter->pdev->irq);
4733         e1000_intr(adapter->pdev->irq, netdev);
4734
4735         enable_irq(adapter->pdev->irq);
4736 }
4737 #endif
4738
4739 /**
4740  * e1000_io_error_detected - called when PCI error is detected
4741  * @pdev: Pointer to PCI device
4742  * @state: The current pci connection state
4743  *
4744  * This function is called after a PCI bus error affecting
4745  * this device has been detected.
4746  */
4747 static pci_ers_result_t e1000_io_error_detected(struct pci_dev *pdev,
4748                                                 pci_channel_state_t state)
4749 {
4750         struct net_device *netdev = pci_get_drvdata(pdev);
4751         struct e1000_adapter *adapter = netdev_priv(netdev);
4752
4753         netif_device_detach(netdev);
4754
4755         if (state == pci_channel_io_perm_failure)
4756                 return PCI_ERS_RESULT_DISCONNECT;
4757
4758         if (netif_running(netdev))
4759                 e1000e_down(adapter);
4760         pci_disable_device(pdev);
4761
4762         /* Request a slot slot reset. */
4763         return PCI_ERS_RESULT_NEED_RESET;
4764 }
4765
4766 /**
4767  * e1000_io_slot_reset - called after the pci bus has been reset.
4768  * @pdev: Pointer to PCI device
4769  *
4770  * Restart the card from scratch, as if from a cold-boot. Implementation
4771  * resembles the first-half of the e1000_resume routine.
4772  */
4773 static pci_ers_result_t e1000_io_slot_reset(struct pci_dev *pdev)
4774 {
4775         struct net_device *netdev = pci_get_drvdata(pdev);
4776         struct e1000_adapter *adapter = netdev_priv(netdev);
4777         struct e1000_hw *hw = &adapter->hw;
4778         int err;
4779         pci_ers_result_t result;
4780
4781         e1000e_disable_l1aspm(pdev);
4782         err = pci_enable_device_mem(pdev);
4783         if (err) {
4784                 dev_err(&pdev->dev,
4785                         "Cannot re-enable PCI device after reset.\n");
4786                 result = PCI_ERS_RESULT_DISCONNECT;
4787         } else {
4788                 pci_set_master(pdev);
4789                 pci_restore_state(pdev);
4790
4791                 pci_enable_wake(pdev, PCI_D3hot, 0);
4792                 pci_enable_wake(pdev, PCI_D3cold, 0);
4793
4794                 e1000e_reset(adapter);
4795                 ew32(WUS, ~0);
4796                 result = PCI_ERS_RESULT_RECOVERED;
4797         }
4798
4799         pci_cleanup_aer_uncorrect_error_status(pdev);
4800
4801         return result;
4802 }
4803
4804 /**
4805  * e1000_io_resume - called when traffic can start flowing again.
4806  * @pdev: Pointer to PCI device
4807  *
4808  * This callback is called when the error recovery driver tells us that
4809  * its OK to resume normal operation. Implementation resembles the
4810  * second-half of the e1000_resume routine.
4811  */
4812 static void e1000_io_resume(struct pci_dev *pdev)
4813 {
4814         struct net_device *netdev = pci_get_drvdata(pdev);
4815         struct e1000_adapter *adapter = netdev_priv(netdev);
4816
4817         e1000_init_manageability(adapter);
4818
4819         if (netif_running(netdev)) {
4820                 if (e1000e_up(adapter)) {
4821                         dev_err(&pdev->dev,
4822                                 "can't bring device back up after reset\n");
4823                         return;
4824                 }
4825         }
4826
4827         netif_device_attach(netdev);
4828
4829         /*
4830          * If the controller has AMT, do not set DRV_LOAD until the interface
4831          * is up.  For all other cases, let the f/w know that the h/w is now
4832          * under the control of the driver.
4833          */
4834         if (!(adapter->flags & FLAG_HAS_AMT))
4835                 e1000_get_hw_control(adapter);
4836
4837 }
4838
4839 static void e1000_print_device_info(struct e1000_adapter *adapter)
4840 {
4841         struct e1000_hw *hw = &adapter->hw;
4842         struct net_device *netdev = adapter->netdev;
4843         u32 pba_num;
4844
4845         /* print bus type/speed/width info */
4846         e_info("(PCI Express:2.5GB/s:%s) %pM\n",
4847                /* bus width */
4848                ((hw->bus.width == e1000_bus_width_pcie_x4) ? "Width x4" :
4849                 "Width x1"),
4850                /* MAC address */
4851                netdev->dev_addr);
4852         e_info("Intel(R) PRO/%s Network Connection\n",
4853                (hw->phy.type == e1000_phy_ife) ? "10/100" : "1000");
4854         e1000e_read_pba_num(hw, &pba_num);
4855         e_info("MAC: %d, PHY: %d, PBA No: %06x-%03x\n",
4856                hw->mac.type, hw->phy.type, (pba_num >> 8), (pba_num & 0xff));
4857 }
4858
4859 static void e1000_eeprom_checks(struct e1000_adapter *adapter)
4860 {
4861         struct e1000_hw *hw = &adapter->hw;
4862         int ret_val;
4863         u16 buf = 0;
4864
4865         if (hw->mac.type != e1000_82573)
4866                 return;
4867
4868         ret_val = e1000_read_nvm(hw, NVM_INIT_CONTROL2_REG, 1, &buf);
4869         if (!ret_val && (!(le16_to_cpu(buf) & (1 << 0)))) {
4870                 /* Deep Smart Power Down (DSPD) */
4871                 dev_warn(&adapter->pdev->dev,
4872                          "Warning: detected DSPD enabled in EEPROM\n");
4873         }
4874
4875         ret_val = e1000_read_nvm(hw, NVM_INIT_3GIO_3, 1, &buf);
4876         if (!ret_val && (le16_to_cpu(buf) & (3 << 2))) {
4877                 /* ASPM enable */
4878                 dev_warn(&adapter->pdev->dev,
4879                          "Warning: detected ASPM enabled in EEPROM\n");
4880         }
4881 }
4882
4883 static const struct net_device_ops e1000e_netdev_ops = {
4884         .ndo_open               = e1000_open,
4885         .ndo_stop               = e1000_close,
4886         .ndo_start_xmit         = e1000_xmit_frame,
4887         .ndo_get_stats          = e1000_get_stats,
4888         .ndo_set_multicast_list = e1000_set_multi,
4889         .ndo_set_mac_address    = e1000_set_mac,
4890         .ndo_change_mtu         = e1000_change_mtu,
4891         .ndo_do_ioctl           = e1000_ioctl,
4892         .ndo_tx_timeout         = e1000_tx_timeout,
4893         .ndo_validate_addr      = eth_validate_addr,
4894
4895         .ndo_vlan_rx_register   = e1000_vlan_rx_register,
4896         .ndo_vlan_rx_add_vid    = e1000_vlan_rx_add_vid,
4897         .ndo_vlan_rx_kill_vid   = e1000_vlan_rx_kill_vid,
4898 #ifdef CONFIG_NET_POLL_CONTROLLER
4899         .ndo_poll_controller    = e1000_netpoll,
4900 #endif
4901 };
4902
4903 /**
4904  * e1000_probe - Device Initialization Routine
4905  * @pdev: PCI device information struct
4906  * @ent: entry in e1000_pci_tbl
4907  *
4908  * Returns 0 on success, negative on failure
4909  *
4910  * e1000_probe initializes an adapter identified by a pci_dev structure.
4911  * The OS initialization, configuring of the adapter private structure,
4912  * and a hardware reset occur.
4913  **/
4914 static int __devinit e1000_probe(struct pci_dev *pdev,
4915                                  const struct pci_device_id *ent)
4916 {
4917         struct net_device *netdev;
4918         struct e1000_adapter *adapter;
4919         struct e1000_hw *hw;
4920         const struct e1000_info *ei = e1000_info_tbl[ent->driver_data];
4921         resource_size_t mmio_start, mmio_len;
4922         resource_size_t flash_start, flash_len;
4923
4924         static int cards_found;
4925         int i, err, pci_using_dac;
4926         u16 eeprom_data = 0;
4927         u16 eeprom_apme_mask = E1000_EEPROM_APME;
4928
4929         e1000e_disable_l1aspm(pdev);
4930
4931         err = pci_enable_device_mem(pdev);
4932         if (err)
4933                 return err;
4934
4935         pci_using_dac = 0;
4936         err = pci_set_dma_mask(pdev, DMA_BIT_MASK(64));
4937         if (!err) {
4938                 err = pci_set_consistent_dma_mask(pdev, DMA_BIT_MASK(64));
4939                 if (!err)
4940                         pci_using_dac = 1;
4941         } else {
4942                 err = pci_set_dma_mask(pdev, DMA_BIT_MASK(32));
4943                 if (err) {
4944                         err = pci_set_consistent_dma_mask(pdev,
4945                                                           DMA_BIT_MASK(32));
4946                         if (err) {
4947                                 dev_err(&pdev->dev, "No usable DMA "
4948                                         "configuration, aborting\n");
4949                                 goto err_dma;
4950                         }
4951                 }
4952         }
4953
4954         err = pci_request_selected_regions_exclusive(pdev,
4955                                           pci_select_bars(pdev, IORESOURCE_MEM),
4956                                           e1000e_driver_name);
4957         if (err)
4958                 goto err_pci_reg;
4959
4960         /* AER (Advanced Error Reporting) hooks */
4961         pci_enable_pcie_error_reporting(pdev);
4962
4963         pci_set_master(pdev);
4964         /* PCI config space info */
4965         err = pci_save_state(pdev);
4966         if (err)
4967                 goto err_alloc_etherdev;
4968
4969         err = -ENOMEM;
4970         netdev = alloc_etherdev(sizeof(struct e1000_adapter));
4971         if (!netdev)
4972                 goto err_alloc_etherdev;
4973
4974         SET_NETDEV_DEV(netdev, &pdev->dev);
4975
4976         pci_set_drvdata(pdev, netdev);
4977         adapter = netdev_priv(netdev);
4978         hw = &adapter->hw;
4979         adapter->netdev = netdev;
4980         adapter->pdev = pdev;
4981         adapter->ei = ei;
4982         adapter->pba = ei->pba;
4983         adapter->flags = ei->flags;
4984         adapter->flags2 = ei->flags2;
4985         adapter->hw.adapter = adapter;
4986         adapter->hw.mac.type = ei->mac;
4987         adapter->max_hw_frame_size = ei->max_hw_frame_size;
4988         adapter->msg_enable = (1 << NETIF_MSG_DRV | NETIF_MSG_PROBE) - 1;
4989
4990         mmio_start = pci_resource_start(pdev, 0);
4991         mmio_len = pci_resource_len(pdev, 0);
4992
4993         err = -EIO;
4994         adapter->hw.hw_addr = ioremap(mmio_start, mmio_len);
4995         if (!adapter->hw.hw_addr)
4996                 goto err_ioremap;
4997
4998         if ((adapter->flags & FLAG_HAS_FLASH) &&
4999             (pci_resource_flags(pdev, 1) & IORESOURCE_MEM)) {
5000                 flash_start = pci_resource_start(pdev, 1);
5001                 flash_len = pci_resource_len(pdev, 1);
5002                 adapter->hw.flash_address = ioremap(flash_start, flash_len);
5003                 if (!adapter->hw.flash_address)
5004                         goto err_flashmap;
5005         }
5006
5007         /* construct the net_device struct */
5008         netdev->netdev_ops              = &e1000e_netdev_ops;
5009         e1000e_set_ethtool_ops(netdev);
5010         netdev->watchdog_timeo          = 5 * HZ;
5011         netif_napi_add(netdev, &adapter->napi, e1000_clean, 64);
5012         strncpy(netdev->name, pci_name(pdev), sizeof(netdev->name) - 1);
5013
5014         netdev->mem_start = mmio_start;
5015         netdev->mem_end = mmio_start + mmio_len;
5016
5017         adapter->bd_number = cards_found++;
5018
5019         e1000e_check_options(adapter);
5020
5021         /* setup adapter struct */
5022         err = e1000_sw_init(adapter);
5023         if (err)
5024                 goto err_sw_init;
5025
5026         err = -EIO;
5027
5028         memcpy(&hw->mac.ops, ei->mac_ops, sizeof(hw->mac.ops));
5029         memcpy(&hw->nvm.ops, ei->nvm_ops, sizeof(hw->nvm.ops));
5030         memcpy(&hw->phy.ops, ei->phy_ops, sizeof(hw->phy.ops));
5031
5032         err = ei->get_variants(adapter);
5033         if (err)
5034                 goto err_hw_init;
5035
5036         if ((adapter->flags & FLAG_IS_ICH) &&
5037             (adapter->flags & FLAG_READ_ONLY_NVM))
5038                 e1000e_write_protect_nvm_ich8lan(&adapter->hw);
5039
5040         hw->mac.ops.get_bus_info(&adapter->hw);
5041
5042         adapter->hw.phy.autoneg_wait_to_complete = 0;
5043
5044         /* Copper options */
5045         if (adapter->hw.phy.media_type == e1000_media_type_copper) {
5046                 adapter->hw.phy.mdix = AUTO_ALL_MODES;
5047                 adapter->hw.phy.disable_polarity_correction = 0;
5048                 adapter->hw.phy.ms_type = e1000_ms_hw_default;
5049         }
5050
5051         if (e1000_check_reset_block(&adapter->hw))
5052                 e_info("PHY reset is blocked due to SOL/IDER session.\n");
5053
5054         netdev->features = NETIF_F_SG |
5055                            NETIF_F_HW_CSUM |
5056                            NETIF_F_HW_VLAN_TX |
5057                            NETIF_F_HW_VLAN_RX;
5058
5059         if (adapter->flags & FLAG_HAS_HW_VLAN_FILTER)
5060                 netdev->features |= NETIF_F_HW_VLAN_FILTER;
5061
5062         netdev->features |= NETIF_F_TSO;
5063         netdev->features |= NETIF_F_TSO6;
5064
5065         netdev->vlan_features |= NETIF_F_TSO;
5066         netdev->vlan_features |= NETIF_F_TSO6;
5067         netdev->vlan_features |= NETIF_F_HW_CSUM;
5068         netdev->vlan_features |= NETIF_F_SG;
5069
5070         if (pci_using_dac)
5071                 netdev->features |= NETIF_F_HIGHDMA;
5072
5073         if (e1000e_enable_mng_pass_thru(&adapter->hw))
5074                 adapter->flags |= FLAG_MNG_PT_ENABLED;
5075
5076         /*
5077          * before reading the NVM, reset the controller to
5078          * put the device in a known good starting state
5079          */
5080         adapter->hw.mac.ops.reset_hw(&adapter->hw);
5081
5082         /*
5083          * systems with ASPM and others may see the checksum fail on the first
5084          * attempt. Let's give it a few tries
5085          */
5086         for (i = 0;; i++) {
5087                 if (e1000_validate_nvm_checksum(&adapter->hw) >= 0)
5088                         break;
5089                 if (i == 2) {
5090                         e_err("The NVM Checksum Is Not Valid\n");
5091                         err = -EIO;
5092                         goto err_eeprom;
5093                 }
5094         }
5095
5096         e1000_eeprom_checks(adapter);
5097
5098         /* copy the MAC address out of the NVM */
5099         if (e1000e_read_mac_addr(&adapter->hw))
5100                 e_err("NVM Read Error while reading MAC address\n");
5101
5102         memcpy(netdev->dev_addr, adapter->hw.mac.addr, netdev->addr_len);
5103         memcpy(netdev->perm_addr, adapter->hw.mac.addr, netdev->addr_len);
5104
5105         if (!is_valid_ether_addr(netdev->perm_addr)) {
5106                 e_err("Invalid MAC Address: %pM\n", netdev->perm_addr);
5107                 err = -EIO;
5108                 goto err_eeprom;
5109         }
5110
5111         init_timer(&adapter->watchdog_timer);
5112         adapter->watchdog_timer.function = &e1000_watchdog;
5113         adapter->watchdog_timer.data = (unsigned long) adapter;
5114
5115         init_timer(&adapter->phy_info_timer);
5116         adapter->phy_info_timer.function = &e1000_update_phy_info;
5117         adapter->phy_info_timer.data = (unsigned long) adapter;
5118
5119         INIT_WORK(&adapter->reset_task, e1000_reset_task);
5120         INIT_WORK(&adapter->watchdog_task, e1000_watchdog_task);
5121         INIT_WORK(&adapter->downshift_task, e1000e_downshift_workaround);
5122         INIT_WORK(&adapter->update_phy_task, e1000e_update_phy_task);
5123
5124         /* Initialize link parameters. User can change them with ethtool */
5125         adapter->hw.mac.autoneg = 1;
5126         adapter->fc_autoneg = 1;
5127         adapter->hw.fc.requested_mode = e1000_fc_default;
5128         adapter->hw.fc.current_mode = e1000_fc_default;
5129         adapter->hw.phy.autoneg_advertised = 0x2f;
5130
5131         /* ring size defaults */
5132         adapter->rx_ring->count = 256;
5133         adapter->tx_ring->count = 256;
5134
5135         /*
5136          * Initial Wake on LAN setting - If APM wake is enabled in
5137          * the EEPROM, enable the ACPI Magic Packet filter
5138          */
5139         if (adapter->flags & FLAG_APME_IN_WUC) {
5140                 /* APME bit in EEPROM is mapped to WUC.APME */
5141                 eeprom_data = er32(WUC);
5142                 eeprom_apme_mask = E1000_WUC_APME;
5143                 if (eeprom_data & E1000_WUC_PHY_WAKE)
5144                         adapter->flags2 |= FLAG2_HAS_PHY_WAKEUP;
5145         } else if (adapter->flags & FLAG_APME_IN_CTRL3) {
5146                 if (adapter->flags & FLAG_APME_CHECK_PORT_B &&
5147                     (adapter->hw.bus.func == 1))
5148                         e1000_read_nvm(&adapter->hw,
5149                                 NVM_INIT_CONTROL3_PORT_B, 1, &eeprom_data);
5150                 else
5151                         e1000_read_nvm(&adapter->hw,
5152                                 NVM_INIT_CONTROL3_PORT_A, 1, &eeprom_data);
5153         }
5154
5155         /* fetch WoL from EEPROM */
5156         if (eeprom_data & eeprom_apme_mask)
5157                 adapter->eeprom_wol |= E1000_WUFC_MAG;
5158
5159         /*
5160          * now that we have the eeprom settings, apply the special cases
5161          * where the eeprom may be wrong or the board simply won't support
5162          * wake on lan on a particular port
5163          */
5164         if (!(adapter->flags & FLAG_HAS_WOL))
5165                 adapter->eeprom_wol = 0;
5166
5167         /* initialize the wol settings based on the eeprom settings */
5168         adapter->wol = adapter->eeprom_wol;
5169         device_set_wakeup_enable(&adapter->pdev->dev, adapter->wol);
5170
5171         /* save off EEPROM version number */
5172         e1000_read_nvm(&adapter->hw, 5, 1, &adapter->eeprom_vers);
5173
5174         /* reset the hardware with the new settings */
5175         e1000e_reset(adapter);
5176
5177         /*
5178          * If the controller has AMT, do not set DRV_LOAD until the interface
5179          * is up.  For all other cases, let the f/w know that the h/w is now
5180          * under the control of the driver.
5181          */
5182         if (!(adapter->flags & FLAG_HAS_AMT))
5183                 e1000_get_hw_control(adapter);
5184
5185         strcpy(netdev->name, "eth%d");
5186         err = register_netdev(netdev);
5187         if (err)
5188                 goto err_register;
5189
5190         /* carrier off reporting is important to ethtool even BEFORE open */
5191         netif_carrier_off(netdev);
5192
5193         e1000_print_device_info(adapter);
5194
5195         return 0;
5196
5197 err_register:
5198         if (!(adapter->flags & FLAG_HAS_AMT))
5199                 e1000_release_hw_control(adapter);
5200 err_eeprom:
5201         if (!e1000_check_reset_block(&adapter->hw))
5202                 e1000_phy_hw_reset(&adapter->hw);
5203 err_hw_init:
5204
5205         kfree(adapter->tx_ring);
5206         kfree(adapter->rx_ring);
5207 err_sw_init:
5208         if (adapter->hw.flash_address)
5209                 iounmap(adapter->hw.flash_address);
5210         e1000e_reset_interrupt_capability(adapter);
5211 err_flashmap:
5212         iounmap(adapter->hw.hw_addr);
5213 err_ioremap:
5214         free_netdev(netdev);
5215 err_alloc_etherdev:
5216         pci_release_selected_regions(pdev,
5217                                      pci_select_bars(pdev, IORESOURCE_MEM));
5218 err_pci_reg:
5219 err_dma:
5220         pci_disable_device(pdev);
5221         return err;
5222 }
5223
5224 /**
5225  * e1000_remove - Device Removal Routine
5226  * @pdev: PCI device information struct
5227  *
5228  * e1000_remove is called by the PCI subsystem to alert the driver
5229  * that it should release a PCI device.  The could be caused by a
5230  * Hot-Plug event, or because the driver is going to be removed from
5231  * memory.
5232  **/
5233 static void __devexit e1000_remove(struct pci_dev *pdev)
5234 {
5235         struct net_device *netdev = pci_get_drvdata(pdev);
5236         struct e1000_adapter *adapter = netdev_priv(netdev);
5237
5238         /*
5239          * flush_scheduled work may reschedule our watchdog task, so
5240          * explicitly disable watchdog tasks from being rescheduled
5241          */
5242         set_bit(__E1000_DOWN, &adapter->state);
5243         del_timer_sync(&adapter->watchdog_timer);
5244         del_timer_sync(&adapter->phy_info_timer);
5245
5246         flush_scheduled_work();
5247
5248         /*
5249          * Release control of h/w to f/w.  If f/w is AMT enabled, this
5250          * would have already happened in close and is redundant.
5251          */
5252         e1000_release_hw_control(adapter);
5253
5254         unregister_netdev(netdev);
5255
5256         if (!e1000_check_reset_block(&adapter->hw))
5257                 e1000_phy_hw_reset(&adapter->hw);
5258
5259         e1000e_reset_interrupt_capability(adapter);
5260         kfree(adapter->tx_ring);
5261         kfree(adapter->rx_ring);
5262
5263         iounmap(adapter->hw.hw_addr);
5264         if (adapter->hw.flash_address)
5265                 iounmap(adapter->hw.flash_address);
5266         pci_release_selected_regions(pdev,
5267                                      pci_select_bars(pdev, IORESOURCE_MEM));
5268
5269         free_netdev(netdev);
5270
5271         /* AER disable */
5272         pci_disable_pcie_error_reporting(pdev);
5273
5274         pci_disable_device(pdev);
5275 }
5276
5277 /* PCI Error Recovery (ERS) */
5278 static struct pci_error_handlers e1000_err_handler = {
5279         .error_detected = e1000_io_error_detected,
5280         .slot_reset = e1000_io_slot_reset,
5281         .resume = e1000_io_resume,
5282 };
5283
5284 static struct pci_device_id e1000_pci_tbl[] = {
5285         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82571EB_COPPER), board_82571 },
5286         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82571EB_FIBER), board_82571 },
5287         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82571EB_QUAD_COPPER), board_82571 },
5288         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82571EB_QUAD_COPPER_LP), board_82571 },
5289         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82571EB_QUAD_FIBER), board_82571 },
5290         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82571EB_SERDES), board_82571 },
5291         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82571EB_SERDES_DUAL), board_82571 },
5292         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82571EB_SERDES_QUAD), board_82571 },
5293         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82571PT_QUAD_COPPER), board_82571 },
5294
5295         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82572EI), board_82572 },
5296         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82572EI_COPPER), board_82572 },
5297         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82572EI_FIBER), board_82572 },
5298         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82572EI_SERDES), board_82572 },
5299
5300         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82573E), board_82573 },
5301         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82573E_IAMT), board_82573 },
5302         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82573L), board_82573 },
5303
5304         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82574L), board_82574 },
5305         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82574LA), board_82574 },
5306         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82583V), board_82583 },
5307
5308         { PCI_VDEVICE(INTEL, E1000_DEV_ID_80003ES2LAN_COPPER_DPT),
5309           board_80003es2lan },
5310         { PCI_VDEVICE(INTEL, E1000_DEV_ID_80003ES2LAN_COPPER_SPT),
5311           board_80003es2lan },
5312         { PCI_VDEVICE(INTEL, E1000_DEV_ID_80003ES2LAN_SERDES_DPT),
5313           board_80003es2lan },
5314         { PCI_VDEVICE(INTEL, E1000_DEV_ID_80003ES2LAN_SERDES_SPT),
5315           board_80003es2lan },
5316
5317         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH8_IFE), board_ich8lan },
5318         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH8_IFE_G), board_ich8lan },
5319         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH8_IFE_GT), board_ich8lan },
5320         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH8_IGP_AMT), board_ich8lan },
5321         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH8_IGP_C), board_ich8lan },
5322         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH8_IGP_M), board_ich8lan },
5323         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH8_IGP_M_AMT), board_ich8lan },
5324
5325         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH9_IFE), board_ich9lan },
5326         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH9_IFE_G), board_ich9lan },
5327         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH9_IFE_GT), board_ich9lan },
5328         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH9_IGP_AMT), board_ich9lan },
5329         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH9_IGP_C), board_ich9lan },
5330         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH9_BM), board_ich9lan },
5331         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH9_IGP_M), board_ich9lan },
5332         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH9_IGP_M_AMT), board_ich9lan },
5333         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH9_IGP_M_V), board_ich9lan },
5334
5335         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH10_R_BM_LM), board_ich9lan },
5336         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH10_R_BM_LF), board_ich9lan },
5337         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH10_R_BM_V), board_ich9lan },
5338
5339         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH10_D_BM_LM), board_ich10lan },
5340         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH10_D_BM_LF), board_ich10lan },
5341
5342         { PCI_VDEVICE(INTEL, E1000_DEV_ID_PCH_M_HV_LM), board_pchlan },
5343         { PCI_VDEVICE(INTEL, E1000_DEV_ID_PCH_M_HV_LC), board_pchlan },
5344         { PCI_VDEVICE(INTEL, E1000_DEV_ID_PCH_D_HV_DM), board_pchlan },
5345         { PCI_VDEVICE(INTEL, E1000_DEV_ID_PCH_D_HV_DC), board_pchlan },
5346
5347         { }     /* terminate list */
5348 };
5349 MODULE_DEVICE_TABLE(pci, e1000_pci_tbl);
5350
5351 /* PCI Device API Driver */
5352 static struct pci_driver e1000_driver = {
5353         .name     = e1000e_driver_name,
5354         .id_table = e1000_pci_tbl,
5355         .probe    = e1000_probe,
5356         .remove   = __devexit_p(e1000_remove),
5357 #ifdef CONFIG_PM
5358         /* Power Management Hooks */
5359         .suspend  = e1000_suspend,
5360         .resume   = e1000_resume,
5361 #endif
5362         .shutdown = e1000_shutdown,
5363         .err_handler = &e1000_err_handler
5364 };
5365
5366 /**
5367  * e1000_init_module - Driver Registration Routine
5368  *
5369  * e1000_init_module is the first routine called when the driver is
5370  * loaded. All it does is register with the PCI subsystem.
5371  **/
5372 static int __init e1000_init_module(void)
5373 {
5374         int ret;
5375         printk(KERN_INFO "%s: Intel(R) PRO/1000 Network Driver - %s\n",
5376                e1000e_driver_name, e1000e_driver_version);
5377         printk(KERN_INFO "%s: Copyright (c) 1999-2008 Intel Corporation.\n",
5378                e1000e_driver_name);
5379         ret = pci_register_driver(&e1000_driver);
5380         pm_qos_add_requirement(PM_QOS_CPU_DMA_LATENCY, e1000e_driver_name,
5381                                PM_QOS_DEFAULT_VALUE);
5382
5383         return ret;
5384 }
5385 module_init(e1000_init_module);
5386
5387 /**
5388  * e1000_exit_module - Driver Exit Cleanup Routine
5389  *
5390  * e1000_exit_module is called just before the driver is removed
5391  * from memory.
5392  **/
5393 static void __exit e1000_exit_module(void)
5394 {
5395         pci_unregister_driver(&e1000_driver);
5396         pm_qos_remove_requirement(PM_QOS_CPU_DMA_LATENCY, e1000e_driver_name);
5397 }
5398 module_exit(e1000_exit_module);
5399
5400
5401 MODULE_AUTHOR("Intel Corporation, <linux.nics@intel.com>");
5402 MODULE_DESCRIPTION("Intel(R) PRO/1000 Network Driver");
5403 MODULE_LICENSE("GPL");
5404 MODULE_VERSION(DRV_VERSION);
5405
5406 /* e1000_main.c */