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 + NET_IP_ALIGN;
 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(netdev, bufsz);
 183                 if (!skb) {
 184                         /* Better luck next round */
 185                         adapter->alloc_rx_buff_failed++;
 186                         break;
 187                 }
 188
 189                 /*
 190                  * Make buffer alignment 2 beyond a 16 byte boundary
 191                  * this will result in a 16 byte aligned IP header after
 192                  * the 14 byte MAC header is removed
 193                  */
 194                 skb_reserve(skb, NET_IP_ALIGN);
 195
 196                 buffer_info->skb = skb;
 197 map_skb:
 198                 buffer_info->dma = pci_map_single(pdev, skb->data,
 199                                                   adapter->rx_buffer_len,
 200                                                   PCI_DMA_FROMDEVICE);
 201                 if (pci_dma_mapping_error(pdev, buffer_info->dma)) {
 202                         dev_err(&pdev->dev, "RX DMA map failed\n");
 203                         adapter->rx_dma_failed++;
 204                         break;
 205                 }
 206
 207                 rx_desc = E1000_RX_DESC(*rx_ring, i);
 208                 rx_desc->buffer_addr = cpu_to_le64(buffer_info->dma);
 209
 210                 i++;
 211                 if (i == rx_ring->count)
 212                         i = 0;
 213                 buffer_info = &rx_ring->buffer_info[i];
 214         }
 215
 216         if (rx_ring->next_to_use != i) {
 217                 rx_ring->next_to_use = i;
 218                 if (i-- == 0)
 219                         i = (rx_ring->count - 1);
 220
 221                 /*
 222                  * Force memory writes to complete before letting h/w
 223                  * know there are new descriptors to fetch.  (Only
 224                  * applicable for weak-ordered memory model archs,
 225                  * such as IA-64).
 226                  */
 227                 wmb();
 228                 writel(i, adapter->hw.hw_addr + rx_ring->tail);
 229         }
 230 }
 231
 232 /**
 233  * e1000_alloc_rx_buffers_ps - Replace used receive buffers; packet split
 234  * @adapter: address of board private structure
 235  **/
 236 static void e1000_alloc_rx_buffers_ps(struct e1000_adapter *adapter,
 237                                       int cleaned_count)
 238 {
 239         struct net_device *netdev = adapter->netdev;
 240         struct pci_dev *pdev = adapter->pdev;
 241         union e1000_rx_desc_packet_split *rx_desc;
 242         struct e1000_ring *rx_ring = adapter->rx_ring;
 243         struct e1000_buffer *buffer_info;
 244         struct e1000_ps_page *ps_page;
 245         struct sk_buff *skb;
 246         unsigned int i, j;
 247
 248         i = rx_ring->next_to_use;
 249         buffer_info = &rx_ring->buffer_info[i];
 250
 251         while (cleaned_count--) {
 252                 rx_desc = E1000_RX_DESC_PS(*rx_ring, i);
 253
 254                 for (j = 0; j < PS_PAGE_BUFFERS; j++) {
 255                         ps_page = &buffer_info->ps_pages[j];
 256                         if (j >= adapter->rx_ps_pages) {
 257                                 /* all unused desc entries get hw null ptr */
 258                                 rx_desc->read.buffer_addr[j+1] = ~cpu_to_le64(0);
 259                                 continue;
 260                         }
 261                         if (!ps_page->page) {
 262                                 ps_page->page = alloc_page(GFP_ATOMIC);
 263                                 if (!ps_page->page) {
 264                                         adapter->alloc_rx_buff_failed++;
 265                                         goto no_buffers;
 266                                 }
 267                                 ps_page->dma = pci_map_page(pdev,
 268                                                    ps_page->page,
 269                                                    0, PAGE_SIZE,
 270                                                    PCI_DMA_FROMDEVICE);
 271                                 if (pci_dma_mapping_error(pdev, ps_page->dma)) {
 272                                         dev_err(&adapter->pdev->dev,
 273                                           "RX DMA page map failed\n");
 274                                         adapter->rx_dma_failed++;
 275                                         goto no_buffers;
 276                                 }
 277                         }
 278                         /*
 279                          * Refresh the desc even if buffer_addrs
 280                          * didn't change because each write-back
 281                          * erases this info.
 282                          */
 283                         rx_desc->read.buffer_addr[j+1] =
 284                              cpu_to_le64(ps_page->dma);
 285                 }
 286
 287                 skb = netdev_alloc_skb(netdev,
 288                                        adapter->rx_ps_bsize0 + NET_IP_ALIGN);
 289
 290                 if (!skb) {
 291                         adapter->alloc_rx_buff_failed++;
 292                         break;
 293                 }
 294
 295                 /*
 296                  * Make buffer alignment 2 beyond a 16 byte boundary
 297                  * this will result in a 16 byte aligned IP header after
 298                  * the 14 byte MAC header is removed
 299                  */
 300                 skb_reserve(skb, NET_IP_ALIGN);
 301
 302                 buffer_info->skb = skb;
 303                 buffer_info->dma = pci_map_single(pdev, skb->data,
 304                                                   adapter->rx_ps_bsize0,
 305                                                   PCI_DMA_FROMDEVICE);
 306                 if (pci_dma_mapping_error(pdev, buffer_info->dma)) {
 307                         dev_err(&pdev->dev, "RX DMA map failed\n");
 308                         adapter->rx_dma_failed++;
 309                         /* cleanup skb */
 310                         dev_kfree_skb_any(skb);
 311                         buffer_info->skb = NULL;
 312                         break;
 313                 }
 314
 315                 rx_desc->read.buffer_addr[0] = cpu_to_le64(buffer_info->dma);
 316
 317                 i++;
 318                 if (i == rx_ring->count)
 319                         i = 0;
 320                 buffer_info = &rx_ring->buffer_info[i];
 321         }
 322
 323 no_buffers:
 324         if (rx_ring->next_to_use != i) {
 325                 rx_ring->next_to_use = i;
 326
 327                 if (!(i--))
 328                         i = (rx_ring->count - 1);
 329
 330                 /*
 331                  * Force memory writes to complete before letting h/w
 332                  * know there are new descriptors to fetch.  (Only
 333                  * applicable for weak-ordered memory model archs,
 334                  * such as IA-64).
 335                  */
 336                 wmb();
 337                 /*
 338                  * Hardware increments by 16 bytes, but packet split
 339                  * descriptors are 32 bytes...so we increment tail
 340                  * twice as much.
 341                  */
 342                 writel(i<<1, adapter->hw.hw_addr + rx_ring->tail);
 343         }
 344 }
 345
 346 /**
 347  * e1000_alloc_jumbo_rx_buffers - Replace used jumbo receive buffers
 348  * @adapter: address of board private structure
 349  * @cleaned_count: number of buffers to allocate this pass
 350  **/
 351
 352 static void e1000_alloc_jumbo_rx_buffers(struct e1000_adapter *adapter,
 353                                          int cleaned_count)
 354 {
 355         struct net_device *netdev = adapter->netdev;
 356         struct pci_dev *pdev = adapter->pdev;
 357         struct e1000_rx_desc *rx_desc;
 358         struct e1000_ring *rx_ring = adapter->rx_ring;
 359         struct e1000_buffer *buffer_info;
 360         struct sk_buff *skb;
 361         unsigned int i;
 362         unsigned int bufsz = 256 -
 363                              16 /* for skb_reserve */ -
 364                              NET_IP_ALIGN;
 365
 366         i = rx_ring->next_to_use;
 367         buffer_info = &rx_ring->buffer_info[i];
 368
 369         while (cleaned_count--) {
 370                 skb = buffer_info->skb;
 371                 if (skb) {
 372                         skb_trim(skb, 0);
 373                         goto check_page;
 374                 }
 375
 376                 skb = netdev_alloc_skb(netdev, bufsz);
 377                 if (unlikely(!skb)) {
 378                         /* Better luck next round */
 379                         adapter->alloc_rx_buff_failed++;
 380                         break;
 381                 }
 382
 383                 /* Make buffer alignment 2 beyond a 16 byte boundary
 384                  * this will result in a 16 byte aligned IP header after
 385                  * the 14 byte MAC header is removed
 386                  */
 387                 skb_reserve(skb, NET_IP_ALIGN);
 388
 389                 buffer_info->skb = skb;
 390 check_page:
 391                 /* allocate a new page if necessary */
 392                 if (!buffer_info->page) {
 393                         buffer_info->page = alloc_page(GFP_ATOMIC);
 394                         if (unlikely(!buffer_info->page)) {
 395                                 adapter->alloc_rx_buff_failed++;
 396                                 break;
 397                         }
 398                 }
 399
 400                 if (!buffer_info->dma)
 401                         buffer_info->dma = pci_map_page(pdev,
 402                                                         buffer_info->page, 0,
 403                                                         PAGE_SIZE,
 404                                                         PCI_DMA_FROMDEVICE);
 405
 406                 rx_desc = E1000_RX_DESC(*rx_ring, i);
 407                 rx_desc->buffer_addr = cpu_to_le64(buffer_info->dma);
 408
 409                 if (unlikely(++i == rx_ring->count))
 410                         i = 0;
 411                 buffer_info = &rx_ring->buffer_info[i];
 412         }
 413
 414         if (likely(rx_ring->next_to_use != i)) {
 415                 rx_ring->next_to_use = i;
 416                 if (unlikely(i-- == 0))
 417                         i = (rx_ring->count - 1);
 418
 419                 /* Force memory writes to complete before letting h/w
 420                  * know there are new descriptors to fetch.  (Only
 421                  * applicable for weak-ordered memory model archs,
 422                  * such as IA-64). */
 423                 wmb();
 424                 writel(i, adapter->hw.hw_addr + rx_ring->tail);
 425         }
 426 }
 427
 428 /**
 429  * e1000_clean_rx_irq - Send received data up the network stack; legacy
 430  * @adapter: board private structure
 431  *
 432  * the return value indicates whether actual cleaning was done, there
 433  * is no guarantee that everything was cleaned
 434  **/
 435 static bool e1000_clean_rx_irq(struct e1000_adapter *adapter,
 436                                int *work_done, int work_to_do)
 437 {
 438         struct net_device *netdev = adapter->netdev;
 439         struct pci_dev *pdev = adapter->pdev;
 440         struct e1000_ring *rx_ring = adapter->rx_ring;
 441         struct e1000_rx_desc *rx_desc, *next_rxd;
 442         struct e1000_buffer *buffer_info, *next_buffer;
 443         u32 length;
 444         unsigned int i;
 445         int cleaned_count = 0;
 446         bool cleaned = 0;
 447         unsigned int total_rx_bytes = 0, total_rx_packets = 0;
 448
 449         i = rx_ring->next_to_clean;
 450         rx_desc = E1000_RX_DESC(*rx_ring, i);
 451         buffer_info = &rx_ring->buffer_info[i];
 452
 453         while (rx_desc->status & E1000_RXD_STAT_DD) {
 454                 struct sk_buff *skb;
 455                 u8 status;
 456
 457                 if (*work_done >= work_to_do)
 458                         break;
 459                 (*work_done)++;
 460
 461                 status = rx_desc->status;
 462                 skb = buffer_info->skb;
 463                 buffer_info->skb = NULL;
 464
 465                 prefetch(skb->data - NET_IP_ALIGN);
 466
 467                 i++;
 468                 if (i == rx_ring->count)
 469                         i = 0;
 470                 next_rxd = E1000_RX_DESC(*rx_ring, i);
 471                 prefetch(next_rxd);
 472
 473                 next_buffer = &rx_ring->buffer_info[i];
 474
 475                 cleaned = 1;
 476                 cleaned_count++;
 477                 pci_unmap_single(pdev,
 478                                  buffer_info->dma,
 479                                  adapter->rx_buffer_len,
 480                                  PCI_DMA_FROMDEVICE);
 481                 buffer_info->dma = 0;
 482
 483                 length = le16_to_cpu(rx_desc->length);
 484
 485                 /* !EOP means multiple descriptors were used to store a single
 486                  * packet, also make sure the frame isn't just CRC only */
 487                 if (!(status & E1000_RXD_STAT_EOP) || (length <= 4)) {
 488                         /* All receives must fit into a single buffer */
 489                         e_dbg("%s: Receive packet consumed multiple buffers\n",
 490                               netdev->name);
 491                         /* recycle */
 492                         buffer_info->skb = skb;
 493                         goto next_desc;
 494                 }
 495
 496                 if (rx_desc->errors & E1000_RXD_ERR_FRAME_ERR_MASK) {
 497                         /* recycle */
 498                         buffer_info->skb = skb;
 499                         goto next_desc;
 500                 }
 501
 502                 /* adjust length to remove Ethernet CRC */
 503                 if (!(adapter->flags2 & FLAG2_CRC_STRIPPING))
 504                         length -= 4;
 505
 506                 total_rx_bytes += length;
 507                 total_rx_packets++;
 508
 509                 /*
 510                  * code added for copybreak, this should improve
 511                  * performance for small packets with large amounts
 512                  * of reassembly being done in the stack
 513                  */
 514                 if (length < copybreak) {
 515                         struct sk_buff *new_skb =
 516                             netdev_alloc_skb(netdev, length + NET_IP_ALIGN);
 517                         if (new_skb) {
 518                                 skb_reserve(new_skb, NET_IP_ALIGN);
 519                                 skb_copy_to_linear_data_offset(new_skb,
 520                                                                -NET_IP_ALIGN,
 521                                                                (skb->data -
 522                                                                 NET_IP_ALIGN),
 523                                                                (length +
 524                                                                 NET_IP_ALIGN));
 525                                 /* save the skb in buffer_info as good */
 526                                 buffer_info->skb = skb;
 527                                 skb = new_skb;
 528                         }
 529                         /* else just continue with the old one */
 530                 }
 531                 /* end copybreak code */
 532                 skb_put(skb, length);
 533
 534                 /* Receive Checksum Offload */
 535                 e1000_rx_checksum(adapter,
 536                                   (u32)(status) |
 537                                   ((u32)(rx_desc->errors) << 24),
 538                                   le16_to_cpu(rx_desc->csum), skb);
 539
 540                 e1000_receive_skb(adapter, netdev, skb,status,rx_desc->special);
 541
 542 next_desc:
 543                 rx_desc->status = 0;
 544
 545                 /* return some buffers to hardware, one at a time is too slow */
 546                 if (cleaned_count >= E1000_RX_BUFFER_WRITE) {
 547                         adapter->alloc_rx_buf(adapter, cleaned_count);
 548                         cleaned_count = 0;
 549                 }
 550
 551                 /* use prefetched values */
 552                 rx_desc = next_rxd;
 553                 buffer_info = next_buffer;
 554         }
 555         rx_ring->next_to_clean = i;
 556
 557         cleaned_count = e1000_desc_unused(rx_ring);
 558         if (cleaned_count)
 559                 adapter->alloc_rx_buf(adapter, cleaned_count);
 560
 561         adapter->total_rx_bytes += total_rx_bytes;
 562         adapter->total_rx_packets += total_rx_packets;
 563         adapter->net_stats.rx_bytes += total_rx_bytes;
 564         adapter->net_stats.rx_packets += total_rx_packets;
 565         return cleaned;
 566 }
 567
 568 static void e1000_put_txbuf(struct e1000_adapter *adapter,
 569                              struct e1000_buffer *buffer_info)
 570 {
 571         buffer_info->dma = 0;
 572         if (buffer_info->skb) {
 573                 skb_dma_unmap(&adapter->pdev->dev, buffer_info->skb,
 574                               DMA_TO_DEVICE);
 575                 dev_kfree_skb_any(buffer_info->skb);
 576                 buffer_info->skb = NULL;
 577         }
 578         buffer_info->time_stamp = 0;
 579 }
 580
 581 static void e1000_print_tx_hang(struct e1000_adapter *adapter)
 582 {
 583         struct e1000_ring *tx_ring = adapter->tx_ring;
 584         unsigned int i = tx_ring->next_to_clean;
 585         unsigned int eop = tx_ring->buffer_info[i].next_to_watch;
 586         struct e1000_tx_desc *eop_desc = E1000_TX_DESC(*tx_ring, eop);
 587
 588         /* detected Tx unit hang */
 589         e_err("Detected Tx Unit Hang:\n"
 590               "  TDH                  <%x>\n"
 591               "  TDT                  <%x>\n"
 592               "  next_to_use          <%x>\n"
 593               "  next_to_clean        <%x>\n"
 594               "buffer_info[next_to_clean]:\n"
 595               "  time_stamp           <%lx>\n"
 596               "  next_to_watch        <%x>\n"
 597               "  jiffies              <%lx>\n"
 598               "  next_to_watch.status <%x>\n",
 599               readl(adapter->hw.hw_addr + tx_ring->head),
 600               readl(adapter->hw.hw_addr + tx_ring->tail),
 601               tx_ring->next_to_use,
 602               tx_ring->next_to_clean,
 603               tx_ring->buffer_info[eop].time_stamp,
 604               eop,
 605               jiffies,
 606               eop_desc->upper.fields.status);
 607 }
 608
 609 /**
 610  * e1000_clean_tx_irq - Reclaim resources after transmit completes
 611  * @adapter: board private structure
 612  *
 613  * the return value indicates whether actual cleaning was done, there
 614  * is no guarantee that everything was cleaned
 615  **/
 616 static bool e1000_clean_tx_irq(struct e1000_adapter *adapter)
 617 {
 618         struct net_device *netdev = adapter->netdev;
 619         struct e1000_hw *hw = &adapter->hw;
 620         struct e1000_ring *tx_ring = adapter->tx_ring;
 621         struct e1000_tx_desc *tx_desc, *eop_desc;
 622         struct e1000_buffer *buffer_info;
 623         unsigned int i, eop;
 624         unsigned int count = 0;
 625         unsigned int total_tx_bytes = 0, total_tx_packets = 0;
 626
 627         i = tx_ring->next_to_clean;
 628         eop = tx_ring->buffer_info[i].next_to_watch;
 629         eop_desc = E1000_TX_DESC(*tx_ring, eop);
 630
 631         while ((eop_desc->upper.data & cpu_to_le32(E1000_TXD_STAT_DD)) &&
 632                (count < tx_ring->count)) {
 633                 bool cleaned = false;
 634                 for (; !cleaned; count++) {
 635                         tx_desc = E1000_TX_DESC(*tx_ring, i);
 636                         buffer_info = &tx_ring->buffer_info[i];
 637                         cleaned = (i == eop);
 638
 639                         if (cleaned) {
 640                                 struct sk_buff *skb = buffer_info->skb;
 641                                 unsigned int segs, bytecount;
 642                                 segs = skb_shinfo(skb)->gso_segs ?: 1;
 643                                 /* multiply data chunks by size of headers */
 644                                 bytecount = ((segs - 1) * skb_headlen(skb)) +
 645                                             skb->len;
 646                                 total_tx_packets += segs;
 647                                 total_tx_bytes += bytecount;
 648                         }
 649
 650                         e1000_put_txbuf(adapter, buffer_info);
 651                         tx_desc->upper.data = 0;
 652
 653                         i++;
 654                         if (i == tx_ring->count)
 655                                 i = 0;
 656                 }
 657
 658                 eop = tx_ring->buffer_info[i].next_to_watch;
 659                 eop_desc = E1000_TX_DESC(*tx_ring, eop);
 660         }
 661
 662         tx_ring->next_to_clean = i;
 663
 664 #define TX_WAKE_THRESHOLD 32
 665         if (count && netif_carrier_ok(netdev) &&
 666             e1000_desc_unused(tx_ring) >= TX_WAKE_THRESHOLD) {
 667                 /* Make sure that anybody stopping the queue after this
 668                  * sees the new next_to_clean.
 669                  */
 670                 smp_mb();
 671
 672                 if (netif_queue_stopped(netdev) &&
 673                     !(test_bit(__E1000_DOWN, &adapter->state))) {
 674                         netif_wake_queue(netdev);
 675                         ++adapter->restart_queue;
 676                 }
 677         }
 678
 679         if (adapter->detect_tx_hung) {
 680                 /* Detect a transmit hang in hardware, this serializes the
 681                  * check with the clearing of time_stamp and movement of i */
 682                 adapter->detect_tx_hung = 0;
 683                 if (tx_ring->buffer_info[i].time_stamp &&
 684                     time_after(jiffies, tx_ring->buffer_info[i].time_stamp
 685                                + (adapter->tx_timeout_factor * HZ))
 686                     && !(er32(STATUS) & E1000_STATUS_TXOFF)) {
 687                         e1000_print_tx_hang(adapter);
 688                         netif_stop_queue(netdev);
 689                 }
 690         }
 691         adapter->total_tx_bytes += total_tx_bytes;
 692         adapter->total_tx_packets += total_tx_packets;
 693         adapter->net_stats.tx_bytes += total_tx_bytes;
 694         adapter->net_stats.tx_packets += total_tx_packets;
 695         return (count < tx_ring->count);
 696 }
 697
 698 /**
 699  * e1000_clean_rx_irq_ps - Send received data up the network stack; packet split
 700  * @adapter: board private structure
 701  *
 702  * the return value indicates whether actual cleaning was done, there
 703  * is no guarantee that everything was cleaned
 704  **/
 705 static bool e1000_clean_rx_irq_ps(struct e1000_adapter *adapter,
 706                                   int *work_done, int work_to_do)
 707 {
 708         union e1000_rx_desc_packet_split *rx_desc, *next_rxd;
 709         struct net_device *netdev = adapter->netdev;
 710         struct pci_dev *pdev = adapter->pdev;
 711         struct e1000_ring *rx_ring = adapter->rx_ring;
 712         struct e1000_buffer *buffer_info, *next_buffer;
 713         struct e1000_ps_page *ps_page;
 714         struct sk_buff *skb;
 715         unsigned int i, j;
 716         u32 length, staterr;
 717         int cleaned_count = 0;
 718         bool cleaned = 0;
 719         unsigned int total_rx_bytes = 0, total_rx_packets = 0;
 720
 721         i = rx_ring->next_to_clean;
 722         rx_desc = E1000_RX_DESC_PS(*rx_ring, i);
 723         staterr = le32_to_cpu(rx_desc->wb.middle.status_error);
 724         buffer_info = &rx_ring->buffer_info[i];
 725
 726         while (staterr & E1000_RXD_STAT_DD) {
 727                 if (*work_done >= work_to_do)
 728                         break;
 729                 (*work_done)++;
 730                 skb = buffer_info->skb;
 731
 732                 /* in the packet split case this is header only */
 733                 prefetch(skb->data - NET_IP_ALIGN);
 734
 735                 i++;
 736                 if (i == rx_ring->count)
 737                         i = 0;
 738                 next_rxd = E1000_RX_DESC_PS(*rx_ring, i);
 739                 prefetch(next_rxd);
 740
 741                 next_buffer = &rx_ring->buffer_info[i];
 742
 743                 cleaned = 1;
 744                 cleaned_count++;
 745                 pci_unmap_single(pdev, buffer_info->dma,
 746                                  adapter->rx_ps_bsize0,
 747                                  PCI_DMA_FROMDEVICE);
 748                 buffer_info->dma = 0;
 749
 750                 if (!(staterr & E1000_RXD_STAT_EOP)) {
 751                         e_dbg("%s: Packet Split buffers didn't pick up the "
 752                               "full packet\n", netdev->name);
 753                         dev_kfree_skb_irq(skb);
 754                         goto next_desc;
 755                 }
 756
 757                 if (staterr & E1000_RXDEXT_ERR_FRAME_ERR_MASK) {
 758                         dev_kfree_skb_irq(skb);
 759                         goto next_desc;
 760                 }
 761
 762                 length = le16_to_cpu(rx_desc->wb.middle.length0);
 763
 764                 if (!length) {
 765                         e_dbg("%s: Last part of the packet spanning multiple "
 766                               "descriptors\n", netdev->name);
 767                         dev_kfree_skb_irq(skb);
 768                         goto next_desc;
 769                 }
 770
 771                 /* Good Receive */
 772                 skb_put(skb, length);
 773
 774                 {
 775                 /*
 776                  * this looks ugly, but it seems compiler issues make it
 777                  * more efficient than reusing j
 778                  */
 779                 int l1 = le16_to_cpu(rx_desc->wb.upper.length[0]);
 780
 781                 /*
 782                  * page alloc/put takes too long and effects small packet
 783                  * throughput, so unsplit small packets and save the alloc/put
 784                  * only valid in softirq (napi) context to call kmap_*
 785                  */
 786                 if (l1 && (l1 <= copybreak) &&
 787                     ((length + l1) <= adapter->rx_ps_bsize0)) {
 788                         u8 *vaddr;
 789
 790                         ps_page = &buffer_info->ps_pages[0];
 791
 792                         /*
 793                          * there is no documentation about how to call
 794                          * kmap_atomic, so we can't hold the mapping
 795                          * very long
 796                          */
 797                         pci_dma_sync_single_for_cpu(pdev, ps_page->dma,
 798                                 PAGE_SIZE, PCI_DMA_FROMDEVICE);
 799                         vaddr = kmap_atomic(ps_page->page, KM_SKB_DATA_SOFTIRQ);
 800                         memcpy(skb_tail_pointer(skb), vaddr, l1);
 801                         kunmap_atomic(vaddr, KM_SKB_DATA_SOFTIRQ);
 802                         pci_dma_sync_single_for_device(pdev, ps_page->dma,
 803                                 PAGE_SIZE, PCI_DMA_FROMDEVICE);
 804
 805                         /* remove the CRC */
 806                         if (!(adapter->flags2 & FLAG2_CRC_STRIPPING))
 807                                 l1 -= 4;
 808
 809                         skb_put(skb, l1);
 810                         goto copydone;
 811                 } /* if */
 812                 }
 813
 814                 for (j = 0; j < PS_PAGE_BUFFERS; j++) {
 815                         length = le16_to_cpu(rx_desc->wb.upper.length[j]);
 816                         if (!length)
 817                                 break;
 818
 819                         ps_page = &buffer_info->ps_pages[j];
 820                         pci_unmap_page(pdev, ps_page->dma, PAGE_SIZE,
 821                                        PCI_DMA_FROMDEVICE);
 822                         ps_page->dma = 0;
 823                         skb_fill_page_desc(skb, j, ps_page->page, 0, length);
 824                         ps_page->page = NULL;
 825                         skb->len += length;
 826                         skb->data_len += length;
 827                         skb->truesize += length;
 828                 }
 829
 830                 /* strip the ethernet crc, problem is we're using pages now so
 831                  * this whole operation can get a little cpu intensive
 832                  */
 833                 if (!(adapter->flags2 & FLAG2_CRC_STRIPPING))
 834                         pskb_trim(skb, skb->len - 4);
 835
 836 copydone:
 837                 total_rx_bytes += skb->len;
 838                 total_rx_packets++;
 839
 840                 e1000_rx_checksum(adapter, staterr, le16_to_cpu(
 841                         rx_desc->wb.lower.hi_dword.csum_ip.csum), skb);
 842
 843                 if (rx_desc->wb.upper.header_status &
 844                            cpu_to_le16(E1000_RXDPS_HDRSTAT_HDRSP))
 845                         adapter->rx_hdr_split++;
 846
 847                 e1000_receive_skb(adapter, netdev, skb,
 848                                   staterr, rx_desc->wb.middle.vlan);
 849
 850 next_desc:
 851                 rx_desc->wb.middle.status_error &= cpu_to_le32(~0xFF);
 852                 buffer_info->skb = NULL;
 853
 854                 /* return some buffers to hardware, one at a time is too slow */
 855                 if (cleaned_count >= E1000_RX_BUFFER_WRITE) {
 856                         adapter->alloc_rx_buf(adapter, cleaned_count);
 857                         cleaned_count = 0;
 858                 }
 859
 860                 /* use prefetched values */
 861                 rx_desc = next_rxd;
 862                 buffer_info = next_buffer;
 863
 864                 staterr = le32_to_cpu(rx_desc->wb.middle.status_error);
 865         }
 866         rx_ring->next_to_clean = i;
 867
 868         cleaned_count = e1000_desc_unused(rx_ring);
 869         if (cleaned_count)
 870                 adapter->alloc_rx_buf(adapter, cleaned_count);
 871
 872         adapter->total_rx_bytes += total_rx_bytes;
 873         adapter->total_rx_packets += total_rx_packets;
 874         adapter->net_stats.rx_bytes += total_rx_bytes;
 875         adapter->net_stats.rx_packets += total_rx_packets;
 876         return cleaned;
 877 }
 878
 879 /**
 880  * e1000_consume_page - helper function
 881  **/
 882 static void e1000_consume_page(struct e1000_buffer *bi, struct sk_buff *skb,
 883                                u16 length)
 884 {
 885         bi->page = NULL;
 886         skb->len += length;
 887         skb->data_len += length;
 888         skb->truesize += length;
 889 }
 890
 891 /**
 892  * e1000_clean_jumbo_rx_irq - Send received data up the network stack; legacy
 893  * @adapter: board private structure
 894  *
 895  * the return value indicates whether actual cleaning was done, there
 896  * is no guarantee that everything was cleaned
 897  **/
 898
 899 static bool e1000_clean_jumbo_rx_irq(struct e1000_adapter *adapter,
 900                                      int *work_done, int work_to_do)
 901 {
 902         struct net_device *netdev = adapter->netdev;
 903         struct pci_dev *pdev = adapter->pdev;
 904         struct e1000_ring *rx_ring = adapter->rx_ring;
 905         struct e1000_rx_desc *rx_desc, *next_rxd;
 906         struct e1000_buffer *buffer_info, *next_buffer;
 907         u32 length;
 908         unsigned int i;
 909         int cleaned_count = 0;
 910         bool cleaned = false;
 911         unsigned int total_rx_bytes=0, total_rx_packets=0;
 912
 913         i = rx_ring->next_to_clean;
 914         rx_desc = E1000_RX_DESC(*rx_ring, i);
 915         buffer_info = &rx_ring->buffer_info[i];
 916
 917         while (rx_desc->status & E1000_RXD_STAT_DD) {
 918                 struct sk_buff *skb;
 919                 u8 status;
 920
 921                 if (*work_done >= work_to_do)
 922                         break;
 923                 (*work_done)++;
 924
 925                 status = rx_desc->status;
 926                 skb = buffer_info->skb;
 927                 buffer_info->skb = NULL;
 928
 929                 ++i;
 930                 if (i == rx_ring->count)
 931                         i = 0;
 932                 next_rxd = E1000_RX_DESC(*rx_ring, i);
 933                 prefetch(next_rxd);
 934
 935                 next_buffer = &rx_ring->buffer_info[i];
 936
 937                 cleaned = true;
 938                 cleaned_count++;
 939                 pci_unmap_page(pdev, buffer_info->dma, PAGE_SIZE,
 940                                PCI_DMA_FROMDEVICE);
 941                 buffer_info->dma = 0;
 942
 943                 length = le16_to_cpu(rx_desc->length);
 944
 945                 /* errors is only valid for DD + EOP descriptors */
 946                 if (unlikely((status & E1000_RXD_STAT_EOP) &&
 947                     (rx_desc->errors & E1000_RXD_ERR_FRAME_ERR_MASK))) {
 948                                 /* recycle both page and skb */
 949                                 buffer_info->skb = skb;
 950                                 /* an error means any chain goes out the window
 951                                  * too */
 952                                 if (rx_ring->rx_skb_top)
 953                                         dev_kfree_skb(rx_ring->rx_skb_top);
 954                                 rx_ring->rx_skb_top = NULL;
 955                                 goto next_desc;
 956                 }
 957
 958 #define rxtop rx_ring->rx_skb_top
 959                 if (!(status & E1000_RXD_STAT_EOP)) {
 960                         /* this descriptor is only the beginning (or middle) */
 961                         if (!rxtop) {
 962                                 /* this is the beginning of a chain */
 963                                 rxtop = skb;
 964                                 skb_fill_page_desc(rxtop, 0, buffer_info->page,
 965                                                    0, length);
 966                         } else {
 967                                 /* this is the middle of a chain */
 968                                 skb_fill_page_desc(rxtop,
 969                                     skb_shinfo(rxtop)->nr_frags,
 970                                     buffer_info->page, 0, length);
 971                                 /* re-use the skb, only consumed the page */
 972                                 buffer_info->skb = skb;
 973                         }
 974                         e1000_consume_page(buffer_info, rxtop, length);
 975                         goto next_desc;
 976                 } else {
 977                         if (rxtop) {
 978                                 /* end of the chain */
 979                                 skb_fill_page_desc(rxtop,
 980                                     skb_shinfo(rxtop)->nr_frags,
 981                                     buffer_info->page, 0, length);
 982                                 /* re-use the current skb, we only consumed the
 983                                  * page */
 984                                 buffer_info->skb = skb;
 985                                 skb = rxtop;
 986                                 rxtop = NULL;
 987                                 e1000_consume_page(buffer_info, skb, length);
 988                         } else {
 989                                 /* no chain, got EOP, this buf is the packet
 990                                  * copybreak to save the put_page/alloc_page */
 991                                 if (length <= copybreak &&
 992                                     skb_tailroom(skb) >= length) {
 993                                         u8 *vaddr;
 994                                         vaddr = kmap_atomic(buffer_info->page,
 995                                                            KM_SKB_DATA_SOFTIRQ);
 996                                         memcpy(skb_tail_pointer(skb), vaddr,
 997                                                length);
 998                                         kunmap_atomic(vaddr,
 999                                                       KM_SKB_DATA_SOFTIRQ);
1000                                         /* re-use the page, so don't erase
1001                                          * buffer_info->page */
1002                                         skb_put(skb, length);
1003                                 } else {
1004                                         skb_fill_page_desc(skb, 0,
1005                                                            buffer_info->page, 0,
1006                                                            length);
1007                                         e1000_consume_page(buffer_info, skb,
1008                                                            length);
1009                                 }
1010                         }
1011                 }
1012
1013                 /* Receive Checksum Offload XXX recompute due to CRC strip? */
1014                 e1000_rx_checksum(adapter,
1015                                   (u32)(status) |
1016                                   ((u32)(rx_desc->errors) << 24),
1017                                   le16_to_cpu(rx_desc->csum), skb);
1018
1019                 /* probably a little skewed due to removing CRC */
1020                 total_rx_bytes += skb->len;
1021                 total_rx_packets++;
1022
1023                 /* eth type trans needs skb->data to point to something */
1024                 if (!pskb_may_pull(skb, ETH_HLEN)) {
1025                         e_err("pskb_may_pull failed.\n");
1026                         dev_kfree_skb(skb);
1027                         goto next_desc;
1028                 }
1029
1030                 e1000_receive_skb(adapter, netdev, skb, status,
1031                                   rx_desc->special);
1032
1033 next_desc:
1034                 rx_desc->status = 0;
1035
1036                 /* return some buffers to hardware, one at a time is too slow */
1037                 if (unlikely(cleaned_count >= E1000_RX_BUFFER_WRITE)) {
1038                         adapter->alloc_rx_buf(adapter, cleaned_count);
1039                         cleaned_count = 0;
1040                 }
1041
1042                 /* use prefetched values */
1043                 rx_desc = next_rxd;
1044                 buffer_info = next_buffer;
1045         }
1046         rx_ring->next_to_clean = i;
1047
1048         cleaned_count = e1000_desc_unused(rx_ring);
1049         if (cleaned_count)
1050                 adapter->alloc_rx_buf(adapter, cleaned_count);
1051
1052         adapter->total_rx_bytes += total_rx_bytes;
1053         adapter->total_rx_packets += total_rx_packets;
1054         adapter->net_stats.rx_bytes += total_rx_bytes;
1055         adapter->net_stats.rx_packets += total_rx_packets;
1056         return cleaned;
1057 }
1058
1059 /**
1060  * e1000_clean_rx_ring - Free Rx Buffers per Queue
1061  * @adapter: board private structure
1062  **/
1063 static void e1000_clean_rx_ring(struct e1000_adapter *adapter)
1064 {
1065         struct e1000_ring *rx_ring = adapter->rx_ring;
1066         struct e1000_buffer *buffer_info;
1067         struct e1000_ps_page *ps_page;
1068         struct pci_dev *pdev = adapter->pdev;
1069         unsigned int i, j;
1070
1071         /* Free all the Rx ring sk_buffs */
1072         for (i = 0; i < rx_ring->count; i++) {
1073                 buffer_info = &rx_ring->buffer_info[i];
1074                 if (buffer_info->dma) {
1075                         if (adapter->clean_rx == e1000_clean_rx_irq)
1076                                 pci_unmap_single(pdev, buffer_info->dma,
1077                                                  adapter->rx_buffer_len,
1078                                                  PCI_DMA_FROMDEVICE);
1079                         else if (adapter->clean_rx == e1000_clean_jumbo_rx_irq)
1080                                 pci_unmap_page(pdev, buffer_info->dma,
1081                                                PAGE_SIZE,
1082                                                PCI_DMA_FROMDEVICE);
1083                         else if (adapter->clean_rx == e1000_clean_rx_irq_ps)
1084                                 pci_unmap_single(pdev, buffer_info->dma,
1085                                                  adapter->rx_ps_bsize0,
1086                                                  PCI_DMA_FROMDEVICE);
1087                         buffer_info->dma = 0;
1088                 }
1089
1090                 if (buffer_info->page) {
1091                         put_page(buffer_info->page);
1092                         buffer_info->page = NULL;
1093                 }
1094
1095                 if (buffer_info->skb) {
1096                         dev_kfree_skb(buffer_info->skb);
1097                         buffer_info->skb = NULL;
1098                 }
1099
1100                 for (j = 0; j < PS_PAGE_BUFFERS; j++) {
1101                         ps_page = &buffer_info->ps_pages[j];
1102                         if (!ps_page->page)
1103                                 break;
1104                         pci_unmap_page(pdev, ps_page->dma, PAGE_SIZE,
1105                                        PCI_DMA_FROMDEVICE);
1106                         ps_page->dma = 0;
1107                         put_page(ps_page->page);
1108                         ps_page->page = NULL;
1109                 }
1110         }
1111
1112         /* there also may be some cached data from a chained receive */
1113         if (rx_ring->rx_skb_top) {
1114                 dev_kfree_skb(rx_ring->rx_skb_top);
1115                 rx_ring->rx_skb_top = NULL;
1116         }
1117
1118         /* Zero out the descriptor ring */
1119         memset(rx_ring->desc, 0, rx_ring->size);
1120
1121         rx_ring->next_to_clean = 0;
1122         rx_ring->next_to_use = 0;
1123
1124         writel(0, adapter->hw.hw_addr + rx_ring->head);
1125         writel(0, adapter->hw.hw_addr + rx_ring->tail);
1126 }
1127
1128 static void e1000e_downshift_workaround(struct work_struct *work)
1129 {
1130         struct e1000_adapter *adapter = container_of(work,
1131                                         struct e1000_adapter, downshift_task);
1132
1133         e1000e_gig_downshift_workaround_ich8lan(&adapter->hw);
1134 }
1135
1136 /**
1137  * e1000_intr_msi - Interrupt Handler
1138  * @irq: interrupt number
1139  * @data: pointer to a network interface device structure
1140  **/
1141 static irqreturn_t e1000_intr_msi(int irq, void *data)
1142 {
1143         struct net_device *netdev = data;
1144         struct e1000_adapter *adapter = netdev_priv(netdev);
1145         struct e1000_hw *hw = &adapter->hw;
1146         u32 icr = er32(ICR);
1147
1148         /*
1149          * read ICR disables interrupts using IAM
1150          */
1151
1152         if (icr & E1000_ICR_LSC) {
1153                 hw->mac.get_link_status = 1;
1154                 /*
1155                  * ICH8 workaround-- Call gig speed drop workaround on cable
1156                  * disconnect (LSC) before accessing any PHY registers
1157                  */
1158                 if ((adapter->flags & FLAG_LSC_GIG_SPEED_DROP) &&
1159                     (!(er32(STATUS) & E1000_STATUS_LU)))
1160                         schedule_work(&adapter->downshift_task);
1161
1162                 /*
1163                  * 80003ES2LAN workaround-- For packet buffer work-around on
1164                  * link down event; disable receives here in the ISR and reset
1165                  * adapter in watchdog
1166                  */
1167                 if (netif_carrier_ok(netdev) &&
1168                     adapter->flags & FLAG_RX_NEEDS_RESTART) {
1169                         /* disable receives */
1170                         u32 rctl = er32(RCTL);
1171                         ew32(RCTL, rctl & ~E1000_RCTL_EN);
1172                         adapter->flags |= FLAG_RX_RESTART_NOW;
1173                 }
1174                 /* guard against interrupt when we're going down */
1175                 if (!test_bit(__E1000_DOWN, &adapter->state))
1176                         mod_timer(&adapter->watchdog_timer, jiffies + 1);
1177         }
1178
1179         if (napi_schedule_prep(&adapter->napi)) {
1180                 adapter->total_tx_bytes = 0;
1181                 adapter->total_tx_packets = 0;
1182                 adapter->total_rx_bytes = 0;
1183                 adapter->total_rx_packets = 0;
1184                 __napi_schedule(&adapter->napi);
1185         }
1186
1187         return IRQ_HANDLED;
1188 }
1189
1190 /**
1191  * e1000_intr - Interrupt Handler
1192  * @irq: interrupt number
1193  * @data: pointer to a network interface device structure
1194  **/
1195 static irqreturn_t e1000_intr(int irq, void *data)
1196 {
1197         struct net_device *netdev = data;
1198         struct e1000_adapter *adapter = netdev_priv(netdev);
1199         struct e1000_hw *hw = &adapter->hw;
1200         u32 rctl, icr = er32(ICR);
1201
1202         if (!icr)
1203                 return IRQ_NONE;  /* Not our interrupt */
1204
1205         /*
1206          * IMS will not auto-mask if INT_ASSERTED is not set, and if it is
1207          * not set, then the adapter didn't send an interrupt
1208          */
1209         if (!(icr & E1000_ICR_INT_ASSERTED))
1210                 return IRQ_NONE;
1211
1212         /*
1213          * Interrupt Auto-Mask...upon reading ICR,
1214          * interrupts are masked.  No need for the
1215          * IMC write
1216          */
1217
1218         if (icr & E1000_ICR_LSC) {
1219                 hw->mac.get_link_status = 1;
1220                 /*
1221                  * ICH8 workaround-- Call gig speed drop workaround on cable
1222                  * disconnect (LSC) before accessing any PHY registers
1223                  */
1224                 if ((adapter->flags & FLAG_LSC_GIG_SPEED_DROP) &&
1225                     (!(er32(STATUS) & E1000_STATUS_LU)))
1226                         schedule_work(&adapter->downshift_task);
1227
1228                 /*
1229                  * 80003ES2LAN workaround--
1230                  * For packet buffer work-around on link down event;
1231                  * disable receives here in the ISR and
1232                  * reset adapter in watchdog
1233                  */
1234                 if (netif_carrier_ok(netdev) &&
1235                     (adapter->flags & FLAG_RX_NEEDS_RESTART)) {
1236                         /* disable receives */
1237                         rctl = er32(RCTL);
1238                         ew32(RCTL, rctl & ~E1000_RCTL_EN);
1239                         adapter->flags |= FLAG_RX_RESTART_NOW;
1240                 }
1241                 /* guard against interrupt when we're going down */
1242                 if (!test_bit(__E1000_DOWN, &adapter->state))
1243                         mod_timer(&adapter->watchdog_timer, jiffies + 1);
1244         }
1245
1246         if (napi_schedule_prep(&adapter->napi)) {
1247                 adapter->total_tx_bytes = 0;
1248                 adapter->total_tx_packets = 0;
1249                 adapter->total_rx_bytes = 0;
1250                 adapter->total_rx_packets = 0;
1251                 __napi_schedule(&adapter->napi);
1252         }
1253
1254         return IRQ_HANDLED;
1255 }
1256
1257 static irqreturn_t e1000_msix_other(int irq, void *data)
1258 {
1259         struct net_device *netdev = data;
1260         struct e1000_adapter *adapter = netdev_priv(netdev);
1261         struct e1000_hw *hw = &adapter->hw;
1262         u32 icr = er32(ICR);
1263
1264         if (!(icr & E1000_ICR_INT_ASSERTED)) {
1265                 if (!test_bit(__E1000_DOWN, &adapter->state))
1266                         ew32(IMS, E1000_IMS_OTHER);
1267                 return IRQ_NONE;
1268         }
1269
1270         if (icr & adapter->eiac_mask)
1271                 ew32(ICS, (icr & adapter->eiac_mask));
1272
1273         if (icr & E1000_ICR_OTHER) {
1274                 if (!(icr & E1000_ICR_LSC))
1275                         goto no_link_interrupt;
1276                 hw->mac.get_link_status = 1;
1277                 /* guard against interrupt when we're going down */
1278                 if (!test_bit(__E1000_DOWN, &adapter->state))
1279                         mod_timer(&adapter->watchdog_timer, jiffies + 1);
1280         }
1281
1282 no_link_interrupt:
1283         if (!test_bit(__E1000_DOWN, &adapter->state))
1284                 ew32(IMS, E1000_IMS_LSC | E1000_IMS_OTHER);
1285
1286         return IRQ_HANDLED;
1287 }
1288
1289
1290 static irqreturn_t e1000_intr_msix_tx(int irq, void *data)
1291 {
1292         struct net_device *netdev = data;
1293         struct e1000_adapter *adapter = netdev_priv(netdev);
1294         struct e1000_hw *hw = &adapter->hw;
1295         struct e1000_ring *tx_ring = adapter->tx_ring;
1296
1297
1298         adapter->total_tx_bytes = 0;
1299         adapter->total_tx_packets = 0;
1300
1301         if (!e1000_clean_tx_irq(adapter))
1302                 /* Ring was not completely cleaned, so fire another interrupt */
1303                 ew32(ICS, tx_ring->ims_val);
1304
1305         return IRQ_HANDLED;
1306 }
1307
1308 static irqreturn_t e1000_intr_msix_rx(int irq, void *data)
1309 {
1310         struct net_device *netdev = data;
1311         struct e1000_adapter *adapter = netdev_priv(netdev);
1312
1313         /* Write the ITR value calculated at the end of the
1314          * previous interrupt.
1315          */
1316         if (adapter->rx_ring->set_itr) {
1317                 writel(1000000000 / (adapter->rx_ring->itr_val * 256),
1318                        adapter->hw.hw_addr + adapter->rx_ring->itr_register);
1319                 adapter->rx_ring->set_itr = 0;
1320         }
1321
1322         if (napi_schedule_prep(&adapter->napi)) {
1323                 adapter->total_rx_bytes = 0;
1324                 adapter->total_rx_packets = 0;
1325                 __napi_schedule(&adapter->napi);
1326         }
1327         return IRQ_HANDLED;
1328 }
1329
1330 /**
1331  * e1000_configure_msix - Configure MSI-X hardware
1332  *
1333  * e1000_configure_msix sets up the hardware to properly
1334  * generate MSI-X interrupts.
1335  **/
1336 static void e1000_configure_msix(struct e1000_adapter *adapter)
1337 {
1338         struct e1000_hw *hw = &adapter->hw;
1339         struct e1000_ring *rx_ring = adapter->rx_ring;
1340         struct e1000_ring *tx_ring = adapter->tx_ring;
1341         int vector = 0;
1342         u32 ctrl_ext, ivar = 0;
1343
1344         adapter->eiac_mask = 0;
1345
1346         /* Workaround issue with spurious interrupts on 82574 in MSI-X mode */
1347         if (hw->mac.type == e1000_82574) {
1348                 u32 rfctl = er32(RFCTL);
1349                 rfctl |= E1000_RFCTL_ACK_DIS;
1350                 ew32(RFCTL, rfctl);
1351         }
1352
1353 #define E1000_IVAR_INT_ALLOC_VALID      0x8
1354         /* Configure Rx vector */
1355         rx_ring->ims_val = E1000_IMS_RXQ0;
1356         adapter->eiac_mask |= rx_ring->ims_val;
1357         if (rx_ring->itr_val)
1358                 writel(1000000000 / (rx_ring->itr_val * 256),
1359                        hw->hw_addr + rx_ring->itr_register);
1360         else
1361                 writel(1, hw->hw_addr + rx_ring->itr_register);
1362         ivar = E1000_IVAR_INT_ALLOC_VALID | vector;
1363
1364         /* Configure Tx vector */
1365         tx_ring->ims_val = E1000_IMS_TXQ0;
1366         vector++;
1367         if (tx_ring->itr_val)
1368                 writel(1000000000 / (tx_ring->itr_val * 256),
1369                        hw->hw_addr + tx_ring->itr_register);
1370         else
1371                 writel(1, hw->hw_addr + tx_ring->itr_register);
1372         adapter->eiac_mask |= tx_ring->ims_val;
1373         ivar |= ((E1000_IVAR_INT_ALLOC_VALID | vector) << 8);
1374
1375         /* set vector for Other Causes, e.g. link changes */
1376         vector++;
1377         ivar |= ((E1000_IVAR_INT_ALLOC_VALID | vector) << 16);
1378         if (rx_ring->itr_val)
1379                 writel(1000000000 / (rx_ring->itr_val * 256),
1380                        hw->hw_addr + E1000_EITR_82574(vector));
1381         else
1382                 writel(1, hw->hw_addr + E1000_EITR_82574(vector));
1383
1384         /* Cause Tx interrupts on every write back */
1385         ivar |= (1 << 31);
1386
1387         ew32(IVAR, ivar);
1388
1389         /* enable MSI-X PBA support */
1390         ctrl_ext = er32(CTRL_EXT);
1391         ctrl_ext |= E1000_CTRL_EXT_PBA_CLR;
1392
1393         /* Auto-Mask Other interrupts upon ICR read */
1394 #define E1000_EIAC_MASK_82574   0x01F00000
1395         ew32(IAM, ~E1000_EIAC_MASK_82574 | E1000_IMS_OTHER);
1396         ctrl_ext |= E1000_CTRL_EXT_EIAME;
1397         ew32(CTRL_EXT, ctrl_ext);
1398         e1e_flush();
1399 }
1400
1401 void e1000e_reset_interrupt_capability(struct e1000_adapter *adapter)
1402 {
1403         if (adapter->msix_entries) {
1404                 pci_disable_msix(adapter->pdev);
1405                 kfree(adapter->msix_entries);
1406                 adapter->msix_entries = NULL;
1407         } else if (adapter->flags & FLAG_MSI_ENABLED) {
1408                 pci_disable_msi(adapter->pdev);
1409                 adapter->flags &= ~FLAG_MSI_ENABLED;
1410         }
1411
1412         return;
1413 }
1414
1415 /**
1416  * e1000e_set_interrupt_capability - set MSI or MSI-X if supported
1417  *
1418  * Attempt to configure interrupts using the best available
1419  * capabilities of the hardware and kernel.
1420  **/
1421 void e1000e_set_interrupt_capability(struct e1000_adapter *adapter)
1422 {
1423         int err;
1424         int numvecs, i;
1425
1426
1427         switch (adapter->int_mode) {
1428         case E1000E_INT_MODE_MSIX:
1429                 if (adapter->flags & FLAG_HAS_MSIX) {
1430                         numvecs = 3; /* RxQ0, TxQ0 and other */
1431                         adapter->msix_entries = kcalloc(numvecs,
1432                                                       sizeof(struct msix_entry),
1433                                                       GFP_KERNEL);
1434                         if (adapter->msix_entries) {
1435                                 for (i = 0; i < numvecs; i++)
1436                                         adapter->msix_entries[i].entry = i;
1437
1438                                 err = pci_enable_msix(adapter->pdev,
1439                                                       adapter->msix_entries,
1440                                                       numvecs);
1441                                 if (err == 0)
1442                                         return;
1443                         }
1444                         /* MSI-X failed, so fall through and try MSI */
1445                         e_err("Failed to initialize MSI-X interrupts.  "
1446                               "Falling back to MSI interrupts.\n");
1447                         e1000e_reset_interrupt_capability(adapter);
1448                 }
1449                 adapter->int_mode = E1000E_INT_MODE_MSI;
1450                 /* Fall through */
1451         case E1000E_INT_MODE_MSI:
1452                 if (!pci_enable_msi(adapter->pdev)) {
1453                         adapter->flags |= FLAG_MSI_ENABLED;
1454                 } else {
1455                         adapter->int_mode = E1000E_INT_MODE_LEGACY;
1456                         e_err("Failed to initialize MSI interrupts.  Falling "
1457                               "back to legacy interrupts.\n");
1458                 }
1459                 /* Fall through */
1460         case E1000E_INT_MODE_LEGACY:
1461                 /* Don't do anything; this is the system default */
1462                 break;
1463         }
1464
1465         return;
1466 }
1467
1468 /**
1469  * e1000_request_msix - Initialize MSI-X interrupts
1470  *
1471  * e1000_request_msix allocates MSI-X vectors and requests interrupts from the
1472  * kernel.
1473  **/
1474 static int e1000_request_msix(struct e1000_adapter *adapter)
1475 {
1476         struct net_device *netdev = adapter->netdev;
1477         int err = 0, vector = 0;
1478
1479         if (strlen(netdev->name) < (IFNAMSIZ - 5))
1480                 sprintf(adapter->rx_ring->name, "%s-rx-0", netdev->name);
1481         else
1482                 memcpy(adapter->rx_ring->name, netdev->name, IFNAMSIZ);
1483         err = request_irq(adapter->msix_entries[vector].vector,
1484                           &e1000_intr_msix_rx, 0, adapter->rx_ring->name,
1485                           netdev);
1486         if (err)
1487                 goto out;
1488         adapter->rx_ring->itr_register = E1000_EITR_82574(vector);
1489         adapter->rx_ring->itr_val = adapter->itr;
1490         vector++;
1491
1492         if (strlen(netdev->name) < (IFNAMSIZ - 5))
1493                 sprintf(adapter->tx_ring->name, "%s-tx-0", netdev->name);
1494         else
1495                 memcpy(adapter->tx_ring->name, netdev->name, IFNAMSIZ);
1496         err = request_irq(adapter->msix_entries[vector].vector,
1497                           &e1000_intr_msix_tx, 0, adapter->tx_ring->name,
1498                           netdev);
1499         if (err)
1500                 goto out;
1501         adapter->tx_ring->itr_register = E1000_EITR_82574(vector);
1502         adapter->tx_ring->itr_val = adapter->itr;
1503         vector++;
1504
1505         err = request_irq(adapter->msix_entries[vector].vector,
1506                           &e1000_msix_other, 0, netdev->name, netdev);
1507         if (err)
1508                 goto out;
1509
1510         e1000_configure_msix(adapter);
1511         return 0;
1512 out:
1513         return err;
1514 }
1515
1516 /**
1517  * e1000_request_irq - initialize interrupts
1518  *
1519  * Attempts to configure interrupts using the best available
1520  * capabilities of the hardware and kernel.
1521  **/
1522 static int e1000_request_irq(struct e1000_adapter *adapter)
1523 {
1524         struct net_device *netdev = adapter->netdev;
1525         int err;
1526
1527         if (adapter->msix_entries) {
1528                 err = e1000_request_msix(adapter);
1529                 if (!err)
1530                         return err;
1531                 /* fall back to MSI */
1532                 e1000e_reset_interrupt_capability(adapter);
1533                 adapter->int_mode = E1000E_INT_MODE_MSI;
1534                 e1000e_set_interrupt_capability(adapter);
1535         }
1536         if (adapter->flags & FLAG_MSI_ENABLED) {
1537                 err = request_irq(adapter->pdev->irq, &e1000_intr_msi, 0,
1538                                   netdev->name, netdev);
1539                 if (!err)
1540                         return err;
1541
1542                 /* fall back to legacy interrupt */
1543                 e1000e_reset_interrupt_capability(adapter);
1544                 adapter->int_mode = E1000E_INT_MODE_LEGACY;
1545         }
1546
1547         err = request_irq(adapter->pdev->irq, &e1000_intr, IRQF_SHARED,
1548                           netdev->name, netdev);
1549         if (err)
1550                 e_err("Unable to allocate interrupt, Error: %d\n", err);
1551
1552         return err;
1553 }
1554
1555 static void e1000_free_irq(struct e1000_adapter *adapter)
1556 {
1557         struct net_device *netdev = adapter->netdev;
1558
1559         if (adapter->msix_entries) {
1560                 int vector = 0;
1561
1562                 free_irq(adapter->msix_entries[vector].vector, netdev);
1563                 vector++;
1564
1565                 free_irq(adapter->msix_entries[vector].vector, netdev);
1566                 vector++;
1567
1568                 /* Other Causes interrupt vector */
1569                 free_irq(adapter->msix_entries[vector].vector, netdev);
1570                 return;
1571         }
1572
1573         free_irq(adapter->pdev->irq, netdev);
1574 }
1575
1576 /**
1577  * e1000_irq_disable - Mask off interrupt generation on the NIC
1578  **/
1579 static void e1000_irq_disable(struct e1000_adapter *adapter)
1580 {
1581         struct e1000_hw *hw = &adapter->hw;
1582
1583         ew32(IMC, ~0);
1584         if (adapter->msix_entries)
1585                 ew32(EIAC_82574, 0);
1586         e1e_flush();
1587         synchronize_irq(adapter->pdev->irq);
1588 }
1589
1590 /**
1591  * e1000_irq_enable - Enable default interrupt generation settings
1592  **/
1593 static void e1000_irq_enable(struct e1000_adapter *adapter)
1594 {
1595         struct e1000_hw *hw = &adapter->hw;
1596
1597         if (adapter->msix_entries) {
1598                 ew32(EIAC_82574, adapter->eiac_mask & E1000_EIAC_MASK_82574);
1599                 ew32(IMS, adapter->eiac_mask | E1000_IMS_OTHER | E1000_IMS_LSC);
1600         } else {
1601                 ew32(IMS, IMS_ENABLE_MASK);
1602         }
1603         e1e_flush();
1604 }
1605
1606 /**
1607  * e1000_get_hw_control - get control of the h/w from f/w
1608  * @adapter: address of board private structure
1609  *
1610  * e1000_get_hw_control sets {CTRL_EXT|SWSM}:DRV_LOAD bit.
1611  * For ASF and Pass Through versions of f/w this means that
1612  * the driver is loaded. For AMT version (only with 82573)
1613  * of the f/w this means that the network i/f is open.
1614  **/
1615 static void e1000_get_hw_control(struct e1000_adapter *adapter)
1616 {
1617         struct e1000_hw *hw = &adapter->hw;
1618         u32 ctrl_ext;
1619         u32 swsm;
1620
1621         /* Let firmware know the driver has taken over */
1622         if (adapter->flags & FLAG_HAS_SWSM_ON_LOAD) {
1623                 swsm = er32(SWSM);
1624                 ew32(SWSM, swsm | E1000_SWSM_DRV_LOAD);
1625         } else if (adapter->flags & FLAG_HAS_CTRLEXT_ON_LOAD) {
1626                 ctrl_ext = er32(CTRL_EXT);
1627                 ew32(CTRL_EXT, ctrl_ext | E1000_CTRL_EXT_DRV_LOAD);
1628         }
1629 }
1630
1631 /**
1632  * e1000_release_hw_control - release control of the h/w to f/w
1633  * @adapter: address of board private structure
1634  *
1635  * e1000_release_hw_control resets {CTRL_EXT|SWSM}:DRV_LOAD bit.
1636  * For ASF and Pass Through versions of f/w this means that the
1637  * driver is no longer loaded. For AMT version (only with 82573) i
1638  * of the f/w this means that the network i/f is closed.
1639  *
1640  **/
1641 static void e1000_release_hw_control(struct e1000_adapter *adapter)
1642 {
1643         struct e1000_hw *hw = &adapter->hw;
1644         u32 ctrl_ext;
1645         u32 swsm;
1646
1647         /* Let firmware taken over control of h/w */
1648         if (adapter->flags & FLAG_HAS_SWSM_ON_LOAD) {
1649                 swsm = er32(SWSM);
1650                 ew32(SWSM, swsm & ~E1000_SWSM_DRV_LOAD);
1651         } else if (adapter->flags & FLAG_HAS_CTRLEXT_ON_LOAD) {
1652                 ctrl_ext = er32(CTRL_EXT);
1653                 ew32(CTRL_EXT, ctrl_ext & ~E1000_CTRL_EXT_DRV_LOAD);
1654         }
1655 }
1656
1657 /**
1658  * @e1000_alloc_ring - allocate memory for a ring structure
1659  **/
1660 static int e1000_alloc_ring_dma(struct e1000_adapter *adapter,
1661                                 struct e1000_ring *ring)
1662 {
1663         struct pci_dev *pdev = adapter->pdev;
1664
1665         ring->desc = dma_alloc_coherent(&pdev->dev, ring->size, &ring->dma,
1666                                         GFP_KERNEL);
1667         if (!ring->desc)
1668                 return -ENOMEM;
1669
1670         return 0;
1671 }
1672
1673 /**
1674  * e1000e_setup_tx_resources - allocate Tx resources (Descriptors)
1675  * @adapter: board private structure
1676  *
1677  * Return 0 on success, negative on failure
1678  **/
1679 int e1000e_setup_tx_resources(struct e1000_adapter *adapter)
1680 {
1681         struct e1000_ring *tx_ring = adapter->tx_ring;
1682         int err = -ENOMEM, size;
1683
1684         size = sizeof(struct e1000_buffer) * tx_ring->count;
1685         tx_ring->buffer_info = vmalloc(size);
1686         if (!tx_ring->buffer_info)
1687                 goto err;
1688         memset(tx_ring->buffer_info, 0, size);
1689
1690         /* round up to nearest 4K */
1691         tx_ring->size = tx_ring->count * sizeof(struct e1000_tx_desc);
1692         tx_ring->size = ALIGN(tx_ring->size, 4096);
1693
1694         err = e1000_alloc_ring_dma(adapter, tx_ring);
1695         if (err)
1696                 goto err;
1697
1698         tx_ring->next_to_use = 0;
1699         tx_ring->next_to_clean = 0;
1700
1701         return 0;
1702 err:
1703         vfree(tx_ring->buffer_info);
1704         e_err("Unable to allocate memory for the transmit descriptor ring\n");
1705         return err;
1706 }
1707
1708 /**
1709  * e1000e_setup_rx_resources - allocate Rx resources (Descriptors)
1710  * @adapter: board private structure
1711  *
1712  * Returns 0 on success, negative on failure
1713  **/
1714 int e1000e_setup_rx_resources(struct e1000_adapter *adapter)
1715 {
1716         struct e1000_ring *rx_ring = adapter->rx_ring;
1717         struct e1000_buffer *buffer_info;
1718         int i, size, desc_len, err = -ENOMEM;
1719
1720         size = sizeof(struct e1000_buffer) * rx_ring->count;
1721         rx_ring->buffer_info = vmalloc(size);
1722         if (!rx_ring->buffer_info)
1723                 goto err;
1724         memset(rx_ring->buffer_info, 0, size);
1725
1726         for (i = 0; i < rx_ring->count; i++) {
1727                 buffer_info = &rx_ring->buffer_info[i];
1728                 buffer_info->ps_pages = kcalloc(PS_PAGE_BUFFERS,
1729                                                 sizeof(struct e1000_ps_page),
1730                                                 GFP_KERNEL);
1731                 if (!buffer_info->ps_pages)
1732                         goto err_pages;
1733         }
1734
1735         desc_len = sizeof(union e1000_rx_desc_packet_split);
1736
1737         /* Round up to nearest 4K */
1738         rx_ring->size = rx_ring->count * desc_len;
1739         rx_ring->size = ALIGN(rx_ring->size, 4096);
1740
1741         err = e1000_alloc_ring_dma(adapter, rx_ring);
1742         if (err)
1743                 goto err_pages;
1744
1745         rx_ring->next_to_clean = 0;
1746         rx_ring->next_to_use = 0;
1747         rx_ring->rx_skb_top = NULL;
1748
1749         return 0;
1750
1751 err_pages:
1752         for (i = 0; i < rx_ring->count; i++) {
1753                 buffer_info = &rx_ring->buffer_info[i];
1754                 kfree(buffer_info->ps_pages);
1755         }
1756 err:
1757         vfree(rx_ring->buffer_info);
1758         e_err("Unable to allocate memory for the transmit descriptor ring\n");
1759         return err;
1760 }
1761
1762 /**
1763  * e1000_clean_tx_ring - Free Tx Buffers
1764  * @adapter: board private structure
1765  **/
1766 static void e1000_clean_tx_ring(struct e1000_adapter *adapter)
1767 {
1768         struct e1000_ring *tx_ring = adapter->tx_ring;
1769         struct e1000_buffer *buffer_info;
1770         unsigned long size;
1771         unsigned int i;
1772
1773         for (i = 0; i < tx_ring->count; i++) {
1774                 buffer_info = &tx_ring->buffer_info[i];
1775                 e1000_put_txbuf(adapter, buffer_info);
1776         }
1777
1778         size = sizeof(struct e1000_buffer) * tx_ring->count;
1779         memset(tx_ring->buffer_info, 0, size);
1780
1781         memset(tx_ring->desc, 0, tx_ring->size);
1782
1783         tx_ring->next_to_use = 0;
1784         tx_ring->next_to_clean = 0;
1785
1786         writel(0, adapter->hw.hw_addr + tx_ring->head);
1787         writel(0, adapter->hw.hw_addr + tx_ring->tail);
1788 }
1789
1790 /**
1791  * e1000e_free_tx_resources - Free Tx Resources per Queue
1792  * @adapter: board private structure
1793  *
1794  * Free all transmit software resources
1795  **/
1796 void e1000e_free_tx_resources(struct e1000_adapter *adapter)
1797 {
1798         struct pci_dev *pdev = adapter->pdev;
1799         struct e1000_ring *tx_ring = adapter->tx_ring;
1800
1801         e1000_clean_tx_ring(adapter);
1802
1803         vfree(tx_ring->buffer_info);
1804         tx_ring->buffer_info = NULL;
1805
1806         dma_free_coherent(&pdev->dev, tx_ring->size, tx_ring->desc,
1807                           tx_ring->dma);
1808         tx_ring->desc = NULL;
1809 }
1810
1811 /**
1812  * e1000e_free_rx_resources - Free Rx Resources
1813  * @adapter: board private structure
1814  *
1815  * Free all receive software resources
1816  **/
1817
1818 void e1000e_free_rx_resources(struct e1000_adapter *adapter)
1819 {
1820         struct pci_dev *pdev = adapter->pdev;
1821         struct e1000_ring *rx_ring = adapter->rx_ring;
1822         int i;
1823
1824         e1000_clean_rx_ring(adapter);
1825
1826         for (i = 0; i < rx_ring->count; i++) {
1827                 kfree(rx_ring->buffer_info[i].ps_pages);
1828         }
1829
1830         vfree(rx_ring->buffer_info);
1831         rx_ring->buffer_info = NULL;
1832
1833         dma_free_coherent(&pdev->dev, rx_ring->size, rx_ring->desc,
1834                           rx_ring->dma);
1835         rx_ring->desc = NULL;
1836 }
1837
1838 /**
1839  * e1000_update_itr - update the dynamic ITR value based on statistics
1840  * @adapter: pointer to adapter
1841  * @itr_setting: current adapter->itr
1842  * @packets: the number of packets during this measurement interval
1843  * @bytes: the number of bytes during this measurement interval
1844  *
1845  *      Stores a new ITR value based on packets and byte
1846  *      counts during the last interrupt.  The advantage of per interrupt
1847  *      computation is faster updates and more accurate ITR for the current
1848  *      traffic pattern.  Constants in this function were computed
1849  *      based on theoretical maximum wire speed and thresholds were set based
1850  *      on testing data as well as attempting to minimize response time
1851  *      while increasing bulk throughput.  This functionality is controlled
1852  *      by the InterruptThrottleRate module parameter.
1853  **/
1854 static unsigned int e1000_update_itr(struct e1000_adapter *adapter,
1855                                      u16 itr_setting, int packets,
1856                                      int bytes)
1857 {
1858         unsigned int retval = itr_setting;
1859
1860         if (packets == 0)
1861                 goto update_itr_done;
1862
1863         switch (itr_setting) {
1864         case lowest_latency:
1865                 /* handle TSO and jumbo frames */
1866                 if (bytes/packets > 8000)
1867                         retval = bulk_latency;
1868                 else if ((packets < 5) && (bytes > 512)) {
1869                         retval = low_latency;
1870                 }
1871                 break;
1872         case low_latency:  /* 50 usec aka 20000 ints/s */
1873                 if (bytes > 10000) {
1874                         /* this if handles the TSO accounting */
1875                         if (bytes/packets > 8000) {
1876                                 retval = bulk_latency;
1877                         } else if ((packets < 10) || ((bytes/packets) > 1200)) {
1878                                 retval = bulk_latency;
1879                         } else if ((packets > 35)) {
1880                                 retval = lowest_latency;
1881                         }
1882                 } else if (bytes/packets > 2000) {
1883                         retval = bulk_latency;
1884                 } else if (packets <= 2 && bytes < 512) {
1885                         retval = lowest_latency;
1886                 }
1887                 break;
1888         case bulk_latency: /* 250 usec aka 4000 ints/s */
1889                 if (bytes > 25000) {
1890                         if (packets > 35) {
1891                                 retval = low_latency;
1892                         }
1893                 } else if (bytes < 6000) {
1894                         retval = low_latency;
1895                 }
1896                 break;
1897         }
1898
1899 update_itr_done:
1900         return retval;
1901 }
1902
1903 static void e1000_set_itr(struct e1000_adapter *adapter)
1904 {
1905         struct e1000_hw *hw = &adapter->hw;
1906         u16 current_itr;
1907         u32 new_itr = adapter->itr;
1908
1909         /* for non-gigabit speeds, just fix the interrupt rate at 4000 */
1910         if (adapter->link_speed != SPEED_1000) {
1911                 current_itr = 0;
1912                 new_itr = 4000;
1913                 goto set_itr_now;
1914         }
1915
1916         adapter->tx_itr = e1000_update_itr(adapter,
1917                                     adapter->tx_itr,
1918                                     adapter->total_tx_packets,
1919                                     adapter->total_tx_bytes);
1920         /* conservative mode (itr 3) eliminates the lowest_latency setting */
1921         if (adapter->itr_setting == 3 && adapter->tx_itr == lowest_latency)
1922                 adapter->tx_itr = low_latency;
1923
1924         adapter->rx_itr = e1000_update_itr(adapter,
1925                                     adapter->rx_itr,
1926                                     adapter->total_rx_packets,
1927                                     adapter->total_rx_bytes);
1928         /* conservative mode (itr 3) eliminates the lowest_latency setting */
1929         if (adapter->itr_setting == 3 && adapter->rx_itr == lowest_latency)
1930                 adapter->rx_itr = low_latency;
1931
1932         current_itr = max(adapter->rx_itr, adapter->tx_itr);
1933
1934         switch (current_itr) {
1935         /* counts and packets in update_itr are dependent on these numbers */
1936         case lowest_latency:
1937                 new_itr = 70000;
1938                 break;
1939         case low_latency:
1940                 new_itr = 20000; /* aka hwitr = ~200 */
1941                 break;
1942         case bulk_latency:
1943                 new_itr = 4000;
1944                 break;
1945         default:
1946                 break;
1947         }
1948
1949 set_itr_now:
1950         if (new_itr != adapter->itr) {
1951                 /*
1952                  * this attempts to bias the interrupt rate towards Bulk
1953                  * by adding intermediate steps when interrupt rate is
1954                  * increasing
1955                  */
1956                 new_itr = new_itr > adapter->itr ?
1957                              min(adapter->itr + (new_itr >> 2), new_itr) :
1958                              new_itr;
1959                 adapter->itr = new_itr;
1960                 adapter->rx_ring->itr_val = new_itr;
1961                 if (adapter->msix_entries)
1962                         adapter->rx_ring->set_itr = 1;
1963                 else
1964                         ew32(ITR, 1000000000 / (new_itr * 256));
1965         }
1966 }
1967
1968 /**
1969  * e1000_alloc_queues - Allocate memory for all rings
1970  * @adapter: board private structure to initialize
1971  **/
1972 static int __devinit e1000_alloc_queues(struct e1000_adapter *adapter)
1973 {
1974         adapter->tx_ring = kzalloc(sizeof(struct e1000_ring), GFP_KERNEL);
1975         if (!adapter->tx_ring)
1976                 goto err;
1977
1978         adapter->rx_ring = kzalloc(sizeof(struct e1000_ring), GFP_KERNEL);
1979         if (!adapter->rx_ring)
1980                 goto err;
1981
1982         return 0;
1983 err:
1984         e_err("Unable to allocate memory for queues\n");
1985         kfree(adapter->rx_ring);
1986         kfree(adapter->tx_ring);
1987         return -ENOMEM;
1988 }
1989
1990 /**
1991  * e1000_clean - NAPI Rx polling callback
1992  * @napi: struct associated with this polling callback
1993  * @budget: amount of packets driver is allowed to process this poll
1994  **/
1995 static int e1000_clean(struct napi_struct *napi, int budget)
1996 {
1997         struct e1000_adapter *adapter = container_of(napi, struct e1000_adapter, napi);
1998         struct e1000_hw *hw = &adapter->hw;
1999         struct net_device *poll_dev = adapter->netdev;
2000         int tx_cleaned = 1, work_done = 0;
2001
2002         adapter = netdev_priv(poll_dev);
2003
2004         if (adapter->msix_entries &&
2005             !(adapter->rx_ring->ims_val & adapter->tx_ring->ims_val))
2006                 goto clean_rx;
2007
2008         tx_cleaned = e1000_clean_tx_irq(adapter);
2009
2010 clean_rx:
2011         adapter->clean_rx(adapter, &work_done, budget);
2012
2013         if (!tx_cleaned)
2014                 work_done = budget;
2015
2016         /* If budget not fully consumed, exit the polling mode */
2017         if (work_done < budget) {
2018                 if (adapter->itr_setting & 3)
2019                         e1000_set_itr(adapter);
2020                 napi_complete(napi);
2021                 if (!test_bit(__E1000_DOWN, &adapter->state)) {
2022                         if (adapter->msix_entries)
2023                                 ew32(IMS, adapter->rx_ring->ims_val);
2024                         else
2025                                 e1000_irq_enable(adapter);
2026                 }
2027         }
2028
2029         return work_done;
2030 }
2031
2032 static void e1000_vlan_rx_add_vid(struct net_device *netdev, u16 vid)
2033 {
2034         struct e1000_adapter *adapter = netdev_priv(netdev);
2035         struct e1000_hw *hw = &adapter->hw;
2036         u32 vfta, index;
2037
2038         /* don't update vlan cookie if already programmed */
2039         if ((adapter->hw.mng_cookie.status &
2040              E1000_MNG_DHCP_COOKIE_STATUS_VLAN) &&
2041             (vid == adapter->mng_vlan_id))
2042                 return;
2043         /* add VID to filter table */
2044         index = (vid >> 5) & 0x7F;
2045         vfta = E1000_READ_REG_ARRAY(hw, E1000_VFTA, index);
2046         vfta |= (1 << (vid & 0x1F));
2047         e1000e_write_vfta(hw, index, vfta);
2048 }
2049
2050 static void e1000_vlan_rx_kill_vid(struct net_device *netdev, u16 vid)
2051 {
2052         struct e1000_adapter *adapter = netdev_priv(netdev);
2053         struct e1000_hw *hw = &adapter->hw;
2054         u32 vfta, index;
2055
2056         if (!test_bit(__E1000_DOWN, &adapter->state))
2057                 e1000_irq_disable(adapter);
2058         vlan_group_set_device(adapter->vlgrp, vid, NULL);
2059
2060         if (!test_bit(__E1000_DOWN, &adapter->state))
2061                 e1000_irq_enable(adapter);
2062
2063         if ((adapter->hw.mng_cookie.status &
2064              E1000_MNG_DHCP_COOKIE_STATUS_VLAN) &&
2065             (vid == adapter->mng_vlan_id)) {
2066                 /* release control to f/w */
2067                 e1000_release_hw_control(adapter);
2068                 return;
2069         }
2070
2071         /* remove VID from filter table */
2072         index = (vid >> 5) & 0x7F;
2073         vfta = E1000_READ_REG_ARRAY(hw, E1000_VFTA, index);
2074         vfta &= ~(1 << (vid & 0x1F));
2075         e1000e_write_vfta(hw, index, vfta);
2076 }
2077
2078 static void e1000_update_mng_vlan(struct e1000_adapter *adapter)
2079 {
2080         struct net_device *netdev = adapter->netdev;
2081         u16 vid = adapter->hw.mng_cookie.vlan_id;
2082         u16 old_vid = adapter->mng_vlan_id;
2083
2084         if (!adapter->vlgrp)
2085                 return;
2086
2087         if (!vlan_group_get_device(adapter->vlgrp, vid)) {
2088                 adapter->mng_vlan_id = E1000_MNG_VLAN_NONE;
2089                 if (adapter->hw.mng_cookie.status &
2090                         E1000_MNG_DHCP_COOKIE_STATUS_VLAN) {
2091                         e1000_vlan_rx_add_vid(netdev, vid);
2092                         adapter->mng_vlan_id = vid;
2093                 }
2094
2095                 if ((old_vid != (u16)E1000_MNG_VLAN_NONE) &&
2096                                 (vid != old_vid) &&
2097                     !vlan_group_get_device(adapter->vlgrp, old_vid))
2098                         e1000_vlan_rx_kill_vid(netdev, old_vid);
2099         } else {
2100                 adapter->mng_vlan_id = vid;
2101         }
2102 }
2103
2104
2105 static void e1000_vlan_rx_register(struct net_device *netdev,
2106                                    struct vlan_group *grp)
2107 {
2108         struct e1000_adapter *adapter = netdev_priv(netdev);
2109         struct e1000_hw *hw = &adapter->hw;
2110         u32 ctrl, rctl;
2111
2112         if (!test_bit(__E1000_DOWN, &adapter->state))
2113                 e1000_irq_disable(adapter);
2114         adapter->vlgrp = grp;
2115
2116         if (grp) {
2117                 /* enable VLAN tag insert/strip */
2118                 ctrl = er32(CTRL);
2119                 ctrl |= E1000_CTRL_VME;
2120                 ew32(CTRL, ctrl);
2121
2122                 if (adapter->flags & FLAG_HAS_HW_VLAN_FILTER) {
2123                         /* enable VLAN receive filtering */
2124                         rctl = er32(RCTL);
2125                         rctl &= ~E1000_RCTL_CFIEN;
2126                         ew32(RCTL, rctl);
2127                         e1000_update_mng_vlan(adapter);
2128                 }
2129         } else {
2130                 /* disable VLAN tag insert/strip */
2131                 ctrl = er32(CTRL);
2132                 ctrl &= ~E1000_CTRL_VME;
2133                 ew32(CTRL, ctrl);
2134
2135                 if (adapter->flags & FLAG_HAS_HW_VLAN_FILTER) {
2136                         if (adapter->mng_vlan_id !=
2137                             (u16)E1000_MNG_VLAN_NONE) {
2138                                 e1000_vlan_rx_kill_vid(netdev,
2139                                                        adapter->mng_vlan_id);
2140                                 adapter->mng_vlan_id = E1000_MNG_VLAN_NONE;
2141                         }
2142                 }
2143         }
2144
2145         if (!test_bit(__E1000_DOWN, &adapter->state))
2146                 e1000_irq_enable(adapter);
2147 }
2148
2149 static void e1000_restore_vlan(struct e1000_adapter *adapter)
2150 {
2151         u16 vid;
2152
2153         e1000_vlan_rx_register(adapter->netdev, adapter->vlgrp);
2154
2155         if (!adapter->vlgrp)
2156                 return;
2157
2158         for (vid = 0; vid < VLAN_GROUP_ARRAY_LEN; vid++) {
2159                 if (!vlan_group_get_device(adapter->vlgrp, vid))
2160                         continue;
2161                 e1000_vlan_rx_add_vid(adapter->netdev, vid);
2162         }
2163 }
2164
2165 static void e1000_init_manageability(struct e1000_adapter *adapter)
2166 {
2167         struct e1000_hw *hw = &adapter->hw;
2168         u32 manc, manc2h;
2169
2170         if (!(adapter->flags & FLAG_MNG_PT_ENABLED))
2171                 return;
2172
2173         manc = er32(MANC);
2174
2175         /*
2176          * enable receiving management packets to the host. this will probably
2177          * generate destination unreachable messages from the host OS, but
2178          * the packets will be handled on SMBUS
2179          */
2180         manc |= E1000_MANC_EN_MNG2HOST;
2181         manc2h = er32(MANC2H);
2182 #define E1000_MNG2HOST_PORT_623 (1 << 5)
2183 #define E1000_MNG2HOST_PORT_664 (1 << 6)
2184         manc2h |= E1000_MNG2HOST_PORT_623;
2185         manc2h |= E1000_MNG2HOST_PORT_664;
2186         ew32(MANC2H, manc2h);
2187         ew32(MANC, manc);
2188 }
2189
2190 /**
2191  * e1000_configure_tx - Configure 8254x Transmit Unit after Reset
2192  * @adapter: board private structure
2193  *
2194  * Configure the Tx unit of the MAC after a reset.
2195  **/
2196 static void e1000_configure_tx(struct e1000_adapter *adapter)
2197 {
2198         struct e1000_hw *hw = &adapter->hw;
2199         struct e1000_ring *tx_ring = adapter->tx_ring;
2200         u64 tdba;
2201         u32 tdlen, tctl, tipg, tarc;
2202         u32 ipgr1, ipgr2;
2203
2204         /* Setup the HW Tx Head and Tail descriptor pointers */
2205         tdba = tx_ring->dma;
2206         tdlen = tx_ring->count * sizeof(struct e1000_tx_desc);
2207         ew32(TDBAL, (tdba & DMA_BIT_MASK(32)));
2208         ew32(TDBAH, (tdba >> 32));
2209         ew32(TDLEN, tdlen);
2210         ew32(TDH, 0);
2211         ew32(TDT, 0);
2212         tx_ring->head = E1000_TDH;
2213         tx_ring->tail = E1000_TDT;
2214
2215         /* Set the default values for the Tx Inter Packet Gap timer */
2216         tipg = DEFAULT_82543_TIPG_IPGT_COPPER;          /*  8  */
2217         ipgr1 = DEFAULT_82543_TIPG_IPGR1;               /*  8  */
2218         ipgr2 = DEFAULT_82543_TIPG_IPGR2;               /*  6  */
2219
2220         if (adapter->flags & FLAG_TIPG_MEDIUM_FOR_80003ESLAN)
2221                 ipgr2 = DEFAULT_80003ES2LAN_TIPG_IPGR2; /*  7  */
2222
2223         tipg |= ipgr1 << E1000_TIPG_IPGR1_SHIFT;
2224         tipg |= ipgr2 << E1000_TIPG_IPGR2_SHIFT;
2225         ew32(TIPG, tipg);
2226
2227         /* Set the Tx Interrupt Delay register */
2228         ew32(TIDV, adapter->tx_int_delay);
2229         /* Tx irq moderation */
2230         ew32(TADV, adapter->tx_abs_int_delay);
2231
2232         /* Program the Transmit Control Register */
2233         tctl = er32(TCTL);
2234         tctl &= ~E1000_TCTL_CT;
2235         tctl |= E1000_TCTL_PSP | E1000_TCTL_RTLC |
2236                 (E1000_COLLISION_THRESHOLD << E1000_CT_SHIFT);
2237
2238         if (adapter->flags & FLAG_TARC_SPEED_MODE_BIT) {
2239                 tarc = er32(TARC(0));
2240                 /*
2241                  * set the speed mode bit, we'll clear it if we're not at
2242                  * gigabit link later
2243                  */
2244 #define SPEED_MODE_BIT (1 << 21)
2245                 tarc |= SPEED_MODE_BIT;
2246                 ew32(TARC(0), tarc);
2247         }
2248
2249         /* errata: program both queues to unweighted RR */
2250         if (adapter->flags & FLAG_TARC_SET_BIT_ZERO) {
2251                 tarc = er32(TARC(0));
2252                 tarc |= 1;
2253                 ew32(TARC(0), tarc);
2254                 tarc = er32(TARC(1));
2255                 tarc |= 1;
2256                 ew32(TARC(1), tarc);
2257         }
2258
2259         /* Setup Transmit Descriptor Settings for eop descriptor */
2260         adapter->txd_cmd = E1000_TXD_CMD_EOP | E1000_TXD_CMD_IFCS;
2261
2262         /* only set IDE if we are delaying interrupts using the timers */
2263         if (adapter->tx_int_delay)
2264                 adapter->txd_cmd |= E1000_TXD_CMD_IDE;
2265
2266         /* enable Report Status bit */
2267         adapter->txd_cmd |= E1000_TXD_CMD_RS;
2268
2269         ew32(TCTL, tctl);
2270
2271         e1000e_config_collision_dist(hw);
2272
2273         adapter->tx_queue_len = adapter->netdev->tx_queue_len;
2274 }
2275
2276 /**
2277  * e1000_setup_rctl - configure the receive control registers
2278  * @adapter: Board private structure
2279  **/
2280 #define PAGE_USE_COUNT(S) (((S) >> PAGE_SHIFT) + \
2281                            (((S) & (PAGE_SIZE - 1)) ? 1 : 0))
2282 static void e1000_setup_rctl(struct e1000_adapter *adapter)
2283 {
2284         struct e1000_hw *hw = &adapter->hw;
2285         u32 rctl, rfctl;
2286         u32 psrctl = 0;
2287         u32 pages = 0;
2288
2289         /* Program MC offset vector base */
2290         rctl = er32(RCTL);
2291         rctl &= ~(3 << E1000_RCTL_MO_SHIFT);
2292         rctl |= E1000_RCTL_EN | E1000_RCTL_BAM |
2293                 E1000_RCTL_LBM_NO | E1000_RCTL_RDMTS_HALF |
2294                 (adapter->hw.mac.mc_filter_type << E1000_RCTL_MO_SHIFT);
2295
2296         /* Do not Store bad packets */
2297         rctl &= ~E1000_RCTL_SBP;
2298
2299         /* Enable Long Packet receive */
2300         if (adapter->netdev->mtu <= ETH_DATA_LEN)
2301                 rctl &= ~E1000_RCTL_LPE;
2302         else
2303                 rctl |= E1000_RCTL_LPE;
2304
2305         /* Some systems expect that the CRC is included in SMBUS traffic. The
2306          * hardware strips the CRC before sending to both SMBUS (BMC) and to
2307          * host memory when this is enabled
2308          */
2309         if (adapter->flags2 & FLAG2_CRC_STRIPPING)
2310                 rctl |= E1000_RCTL_SECRC;
2311
2312         /* Workaround Si errata on 82577 PHY - configure IPG for jumbos */
2313         if ((hw->phy.type == e1000_phy_82577) && (rctl & E1000_RCTL_LPE)) {
2314                 u16 phy_data;
2315
2316                 e1e_rphy(hw, PHY_REG(770, 26), &phy_data);
2317                 phy_data &= 0xfff8;
2318                 phy_data |= (1 << 2);
2319                 e1e_wphy(hw, PHY_REG(770, 26), phy_data);
2320
2321                 e1e_rphy(hw, 22, &phy_data);
2322                 phy_data &= 0x0fff;
2323                 phy_data |= (1 << 14);
2324                 e1e_wphy(hw, 0x10, 0x2823);
2325                 e1e_wphy(hw, 0x11, 0x0003);
2326                 e1e_wphy(hw, 22, phy_data);
2327         }
2328
2329         /* Setup buffer sizes */
2330         rctl &= ~E1000_RCTL_SZ_4096;
2331         rctl |= E1000_RCTL_BSEX;
2332         switch (adapter->rx_buffer_len) {
2333         case 256:
2334                 rctl |= E1000_RCTL_SZ_256;
2335                 rctl &= ~E1000_RCTL_BSEX;
2336                 break;
2337         case 512:
2338                 rctl |= E1000_RCTL_SZ_512;
2339                 rctl &= ~E1000_RCTL_BSEX;
2340                 break;
2341         case 1024:
2342                 rctl |= E1000_RCTL_SZ_1024;
2343                 rctl &= ~E1000_RCTL_BSEX;
2344                 break;
2345         case 2048:
2346         default:
2347                 rctl |= E1000_RCTL_SZ_2048;
2348                 rctl &= ~E1000_RCTL_BSEX;
2349                 break;
2350         case 4096:
2351                 rctl |= E1000_RCTL_SZ_4096;
2352                 break;
2353         case 8192:
2354                 rctl |= E1000_RCTL_SZ_8192;
2355                 break;
2356         case 16384:
2357                 rctl |= E1000_RCTL_SZ_16384;
2358                 break;
2359         }
2360
2361         /*
2362          * 82571 and greater support packet-split where the protocol
2363          * header is placed in skb->data and the packet data is
2364          * placed in pages hanging off of skb_shinfo(skb)->nr_frags.
2365          * In the case of a non-split, skb->data is linearly filled,
2366          * followed by the page buffers.  Therefore, skb->data is
2367          * sized to hold the largest protocol header.
2368          *
2369          * allocations using alloc_page take too long for regular MTU
2370          * so only enable packet split for jumbo frames
2371          *
2372          * Using pages when the page size is greater than 16k wastes
2373          * a lot of memory, since we allocate 3 pages at all times
2374          * per packet.
2375          */
2376         pages = PAGE_USE_COUNT(adapter->netdev->mtu);
2377         if (!(adapter->flags & FLAG_IS_ICH) && (pages <= 3) &&
2378             (PAGE_SIZE <= 16384) && (rctl & E1000_RCTL_LPE))
2379                 adapter->rx_ps_pages = pages;
2380         else
2381                 adapter->rx_ps_pages = 0;
2382
2383         if (adapter->rx_ps_pages) {
2384                 /* Configure extra packet-split registers */
2385                 rfctl = er32(RFCTL);
2386                 rfctl |= E1000_RFCTL_EXTEN;
2387                 /*
2388                  * disable packet split support for IPv6 extension headers,
2389                  * because some malformed IPv6 headers can hang the Rx
2390                  */
2391                 rfctl |= (E1000_RFCTL_IPV6_EX_DIS |
2392                           E1000_RFCTL_NEW_IPV6_EXT_DIS);
2393
2394                 ew32(RFCTL, rfctl);
2395
2396                 /* Enable Packet split descriptors */
2397                 rctl |= E1000_RCTL_DTYP_PS;
2398
2399                 psrctl |= adapter->rx_ps_bsize0 >>
2400                         E1000_PSRCTL_BSIZE0_SHIFT;
2401
2402                 switch (adapter->rx_ps_pages) {
2403                 case 3:
2404                         psrctl |= PAGE_SIZE <<
2405                                 E1000_PSRCTL_BSIZE3_SHIFT;
2406                 case 2:
2407                         psrctl |= PAGE_SIZE <<
2408                                 E1000_PSRCTL_BSIZE2_SHIFT;
2409                 case 1:
2410                         psrctl |= PAGE_SIZE >>
2411                                 E1000_PSRCTL_BSIZE1_SHIFT;
2412                         break;
2413                 }
2414
2415                 ew32(PSRCTL, psrctl);
2416         }
2417
2418         ew32(RCTL, rctl);
2419         /* just started the receive unit, no need to restart */
2420         adapter->flags &= ~FLAG_RX_RESTART_NOW;
2421 }
2422
2423 /**
2424  * e1000_configure_rx - Configure Receive Unit after Reset
2425  * @adapter: board private structure
2426  *
2427  * Configure the Rx unit of the MAC after a reset.
2428  **/
2429 static void e1000_configure_rx(struct e1000_adapter *adapter)
2430 {
2431         struct e1000_hw *hw = &adapter->hw;
2432         struct e1000_ring *rx_ring = adapter->rx_ring;
2433         u64 rdba;
2434         u32 rdlen, rctl, rxcsum, ctrl_ext;
2435
2436         if (adapter->rx_ps_pages) {
2437                 /* this is a 32 byte descriptor */
2438                 rdlen = rx_ring->count *
2439                         sizeof(union e1000_rx_desc_packet_split);
2440                 adapter->clean_rx = e1000_clean_rx_irq_ps;
2441                 adapter->alloc_rx_buf = e1000_alloc_rx_buffers_ps;
2442         } else if (adapter->netdev->mtu > ETH_FRAME_LEN + ETH_FCS_LEN) {
2443                 rdlen = rx_ring->count * sizeof(struct e1000_rx_desc);
2444                 adapter->clean_rx = e1000_clean_jumbo_rx_irq;
2445                 adapter->alloc_rx_buf = e1000_alloc_jumbo_rx_buffers;
2446         } else {
2447                 rdlen = rx_ring->count * sizeof(struct e1000_rx_desc);
2448                 adapter->clean_rx = e1000_clean_rx_irq;
2449                 adapter->alloc_rx_buf = e1000_alloc_rx_buffers;
2450         }
2451
2452         /* disable receives while setting up the descriptors */
2453         rctl = er32(RCTL);
2454         ew32(RCTL, rctl & ~E1000_RCTL_EN);
2455         e1e_flush();
2456         msleep(10);
2457
2458         /* set the Receive Delay Timer Register */
2459         ew32(RDTR, adapter->rx_int_delay);
2460
2461         /* irq moderation */
2462         ew32(RADV, adapter->rx_abs_int_delay);
2463         if (adapter->itr_setting != 0)
2464                 ew32(ITR, 1000000000 / (adapter->itr * 256));
2465
2466         ctrl_ext = er32(CTRL_EXT);
2467         /* Reset delay timers after every interrupt */
2468         ctrl_ext |= E1000_CTRL_EXT_INT_TIMER_CLR;
2469         /* Auto-Mask interrupts upon ICR access */
2470         ctrl_ext |= E1000_CTRL_EXT_IAME;
2471         ew32(IAM, 0xffffffff);
2472         ew32(CTRL_EXT, ctrl_ext);
2473         e1e_flush();
2474
2475         /*
2476          * Setup the HW Rx Head and Tail Descriptor Pointers and
2477          * the Base and Length of the Rx Descriptor Ring
2478          */
2479         rdba = rx_ring->dma;
2480         ew32(RDBAL, (rdba & DMA_BIT_MASK(32)));
2481         ew32(RDBAH, (rdba >> 32));
2482         ew32(RDLEN, rdlen);
2483         ew32(RDH, 0);
2484         ew32(RDT, 0);
2485         rx_ring->head = E1000_RDH;
2486         rx_ring->tail = E1000_RDT;
2487
2488         /* Enable Receive Checksum Offload for TCP and UDP */
2489         rxcsum = er32(RXCSUM);
2490         if (adapter->flags & FLAG_RX_CSUM_ENABLED) {
2491                 rxcsum |= E1000_RXCSUM_TUOFL;
2492
2493                 /*
2494                  * IPv4 payload checksum for UDP fragments must be
2495                  * used in conjunction with packet-split.
2496                  */
2497                 if (adapter->rx_ps_pages)
2498                         rxcsum |= E1000_RXCSUM_IPPCSE;
2499         } else {
2500                 rxcsum &= ~E1000_RXCSUM_TUOFL;
2501                 /* no need to clear IPPCSE as it defaults to 0 */
2502         }
2503         ew32(RXCSUM, rxcsum);
2504
2505         /*
2506          * Enable early receives on supported devices, only takes effect when
2507          * packet size is equal or larger than the specified value (in 8 byte
2508          * units), e.g. using jumbo frames when setting to E1000_ERT_2048
2509          */
2510         if ((adapter->flags & FLAG_HAS_ERT) &&
2511             (adapter->netdev->mtu > ETH_DATA_LEN)) {
2512                 u32 rxdctl = er32(RXDCTL(0));
2513                 ew32(RXDCTL(0), rxdctl | 0x3);
2514                 ew32(ERT, E1000_ERT_2048 | (1 << 13));
2515                 /*
2516                  * With jumbo frames and early-receive enabled, excessive
2517                  * C4->C2 latencies result in dropped transactions.
2518                  */
2519                 pm_qos_update_requirement(PM_QOS_CPU_DMA_LATENCY,
2520                                           e1000e_driver_name, 55);
2521         } else {
2522                 pm_qos_update_requirement(PM_QOS_CPU_DMA_LATENCY,
2523                                           e1000e_driver_name,
2524                                           PM_QOS_DEFAULT_VALUE);
2525         }
2526
2527         /* Enable Receives */
2528         ew32(RCTL, rctl);
2529 }
2530
2531 /**
2532  *  e1000_update_mc_addr_list - Update Multicast addresses
2533  *  @hw: pointer to the HW structure
2534  *  @mc_addr_list: array of multicast addresses to program
2535  *  @mc_addr_count: number of multicast addresses to program
2536  *  @rar_used_count: the first RAR register free to program
2537  *  @rar_count: total number of supported Receive Address Registers
2538  *
2539  *  Updates the Receive Address Registers and Multicast Table Array.
2540  *  The caller must have a packed mc_addr_list of multicast addresses.
2541  *  The parameter rar_count will usually be hw->mac.rar_entry_count
2542  *  unless there are workarounds that change this.  Currently no func pointer
2543  *  exists and all implementations are handled in the generic version of this
2544  *  function.
2545  **/
2546 static void e1000_update_mc_addr_list(struct e1000_hw *hw, u8 *mc_addr_list,
2547                                       u32 mc_addr_count, u32 rar_used_count,
2548                                       u32 rar_count)
2549 {
2550         hw->mac.ops.update_mc_addr_list(hw, mc_addr_list, mc_addr_count,
2551                                         rar_used_count, rar_count);
2552 }
2553
2554 /**
2555  * e1000_set_multi - Multicast and Promiscuous mode set
2556  * @netdev: network interface device structure
2557  *
2558  * The set_multi entry point is called whenever the multicast address
2559  * list or the network interface flags are updated.  This routine is
2560  * responsible for configuring the hardware for proper multicast,
2561  * promiscuous mode, and all-multi behavior.
2562  **/
2563 static void e1000_set_multi(struct net_device *netdev)
2564 {
2565         struct e1000_adapter *adapter = netdev_priv(netdev);
2566         struct e1000_hw *hw = &adapter->hw;
2567         struct e1000_mac_info *mac = &hw->mac;
2568         struct dev_mc_list *mc_ptr;
2569         u8  *mta_list;
2570         u32 rctl;
2571         int i;
2572
2573         /* Check for Promiscuous and All Multicast modes */
2574
2575         rctl = er32(RCTL);
2576
2577         if (netdev->flags & IFF_PROMISC) {
2578                 rctl |= (E1000_RCTL_UPE | E1000_RCTL_MPE);
2579                 rctl &= ~E1000_RCTL_VFE;
2580         } else {
2581                 if (netdev->flags & IFF_ALLMULTI) {
2582                         rctl |= E1000_RCTL_MPE;
2583                         rctl &= ~E1000_RCTL_UPE;
2584                 } else {
2585                         rctl &= ~(E1000_RCTL_UPE | E1000_RCTL_MPE);
2586                 }
2587                 if (adapter->flags & FLAG_HAS_HW_VLAN_FILTER)
2588                         rctl |= E1000_RCTL_VFE;
2589         }
2590
2591         ew32(RCTL, rctl);
2592
2593         if (netdev->mc_count) {
2594                 mta_list = kmalloc(netdev->mc_count * 6, GFP_ATOMIC);
2595                 if (!mta_list)
2596                         return;
2597
2598                 /* prepare a packed array of only addresses. */
2599                 mc_ptr = netdev->mc_list;
2600
2601                 for (i = 0; i < netdev->mc_count; i++) {
2602                         if (!mc_ptr)
2603                                 break;
2604                         memcpy(mta_list + (i*ETH_ALEN), mc_ptr->dmi_addr,
2605                                ETH_ALEN);
2606                         mc_ptr = mc_ptr->next;
2607                 }
2608
2609                 e1000_update_mc_addr_list(hw, mta_list, i, 1,
2610                                           mac->rar_entry_count);
2611                 kfree(mta_list);
2612         } else {
2613                 /*
2614                  * if we're called from probe, we might not have
2615                  * anything to do here, so clear out the list
2616                  */
2617                 e1000_update_mc_addr_list(hw, NULL, 0, 1, mac->rar_entry_count);
2618         }
2619 }
2620
2621 /**
2622  * e1000_configure - configure the hardware for Rx and Tx
2623  * @adapter: private board structure
2624  **/
2625 static void e1000_configure(struct e1000_adapter *adapter)
2626 {
2627         e1000_set_multi(adapter->netdev);
2628
2629         e1000_restore_vlan(adapter);
2630         e1000_init_manageability(adapter);
2631
2632         e1000_configure_tx(adapter);
2633         e1000_setup_rctl(adapter);
2634         e1000_configure_rx(adapter);
2635         adapter->alloc_rx_buf(adapter, e1000_desc_unused(adapter->rx_ring));
2636 }
2637
2638 /**
2639  * e1000e_power_up_phy - restore link in case the phy was powered down
2640  * @adapter: address of board private structure
2641  *
2642  * The phy may be powered down to save power and turn off link when the
2643  * driver is unloaded and wake on lan is not enabled (among others)
2644  * *** this routine MUST be followed by a call to e1000e_reset ***
2645  **/
2646 void e1000e_power_up_phy(struct e1000_adapter *adapter)
2647 {
2648         u16 mii_reg = 0;
2649
2650         /* Just clear the power down bit to wake the phy back up */
2651         if (adapter->hw.phy.media_type == e1000_media_type_copper) {
2652                 /*
2653                  * According to the manual, the phy will retain its
2654                  * settings across a power-down/up cycle
2655                  */
2656                 e1e_rphy(&adapter->hw, PHY_CONTROL, &mii_reg);
2657                 mii_reg &= ~MII_CR_POWER_DOWN;
2658                 e1e_wphy(&adapter->hw, PHY_CONTROL, mii_reg);
2659         }
2660
2661         adapter->hw.mac.ops.setup_link(&adapter->hw);
2662 }
2663
2664 /**
2665  * e1000_power_down_phy - Power down the PHY
2666  *
2667  * Power down the PHY so no link is implied when interface is down
2668  * The PHY cannot be powered down is management or WoL is active
2669  */
2670 static void e1000_power_down_phy(struct e1000_adapter *adapter)
2671 {
2672         struct e1000_hw *hw = &adapter->hw;
2673         u16 mii_reg;
2674
2675         /* WoL is enabled */
2676         if (adapter->wol)
2677                 return;
2678
2679         /* non-copper PHY? */
2680         if (adapter->hw.phy.media_type != e1000_media_type_copper)
2681                 return;
2682
2683         /* reset is blocked because of a SoL/IDER session */
2684         if (e1000e_check_mng_mode(hw) || e1000_check_reset_block(hw))
2685                 return;
2686
2687         /* manageability (AMT) is enabled */
2688         if (er32(MANC) & E1000_MANC_SMBUS_EN)
2689                 return;
2690
2691         /* power down the PHY */
2692         e1e_rphy(hw, PHY_CONTROL, &mii_reg);
2693         mii_reg |= MII_CR_POWER_DOWN;
2694         e1e_wphy(hw, PHY_CONTROL, mii_reg);
2695         mdelay(1);
2696 }
2697
2698 /**
2699  * e1000e_reset - bring the hardware into a known good state
2700  *
2701  * This function boots the hardware and enables some settings that
2702  * require a configuration cycle of the hardware - those cannot be
2703  * set/changed during runtime. After reset the device needs to be
2704  * properly configured for Rx, Tx etc.
2705  */
2706 void e1000e_reset(struct e1000_adapter *adapter)
2707 {
2708         struct e1000_mac_info *mac = &adapter->hw.mac;
2709         struct e1000_fc_info *fc = &adapter->hw.fc;
2710         struct e1000_hw *hw = &adapter->hw;
2711         u32 tx_space, min_tx_space, min_rx_space;
2712         u32 pba = adapter->pba;
2713         u16 hwm;
2714
2715         /* reset Packet Buffer Allocation to default */
2716         ew32(PBA, pba);
2717
2718         if (adapter->max_frame_size > ETH_FRAME_LEN + ETH_FCS_LEN) {
2719                 /*
2720                  * To maintain wire speed transmits, the Tx FIFO should be
2721                  * large enough to accommodate two full transmit packets,
2722                  * rounded up to the next 1KB and expressed in KB.  Likewise,
2723                  * the Rx FIFO should be large enough to accommodate at least
2724                  * one full receive packet and is similarly rounded up and
2725                  * expressed in KB.
2726                  */
2727                 pba = er32(PBA);
2728                 /* upper 16 bits has Tx packet buffer allocation size in KB */
2729                 tx_space = pba >> 16;
2730                 /* lower 16 bits has Rx packet buffer allocation size in KB */
2731                 pba &= 0xffff;
2732                 /*
2733                  * the Tx fifo also stores 16 bytes of information about the tx
2734                  * but don't include ethernet FCS because hardware appends it
2735                  */
2736                 min_tx_space = (adapter->max_frame_size +
2737                                 sizeof(struct e1000_tx_desc) -
2738                                 ETH_FCS_LEN) * 2;
2739                 min_tx_space = ALIGN(min_tx_space, 1024);
2740                 min_tx_space >>= 10;
2741                 /* software strips receive CRC, so leave room for it */
2742                 min_rx_space = adapter->max_frame_size;
2743                 min_rx_space = ALIGN(min_rx_space, 1024);
2744                 min_rx_space >>= 10;
2745
2746                 /*
2747                  * If current Tx allocation is less than the min Tx FIFO size,
2748                  * and the min Tx FIFO size is less than the current Rx FIFO
2749                  * allocation, take space away from current Rx allocation
2750                  */
2751                 if ((tx_space < min_tx_space) &&
2752                     ((min_tx_space - tx_space) < pba)) {
2753                         pba -= min_tx_space - tx_space;
2754
2755                         /*
2756                          * if short on Rx space, Rx wins and must trump tx
2757                          * adjustment or use Early Receive if available
2758                          */
2759                         if ((pba < min_rx_space) &&
2760                             (!(adapter->flags & FLAG_HAS_ERT)))
2761                                 /* ERT enabled in e1000_configure_rx */
2762                                 pba = min_rx_space;
2763                 }
2764
2765                 ew32(PBA, pba);
2766         }
2767
2768
2769         /*
2770          * flow control settings
2771          *
2772          * The high water mark must be low enough to fit two full frame
2773          * (or the size used for early receive) above it in the Rx FIFO.
2774          * Set it to the lower of:
2775          * - 90% of the Rx FIFO size, and
2776          * - the full Rx FIFO size minus the early receive size (for parts
2777          *   with ERT support assuming ERT set to E1000_ERT_2048), or
2778          * - the full Rx FIFO size minus two full frames
2779          */
2780         if ((adapter->flags & FLAG_HAS_ERT) &&
2781             (adapter->netdev->mtu > ETH_DATA_LEN))
2782                 hwm = min(((pba << 10) * 9 / 10),
2783                           ((pba << 10) - (E1000_ERT_2048 << 3)));
2784         else
2785                 hwm = min(((pba << 10) * 9 / 10),
2786                           ((pba << 10) - (2 * adapter->max_frame_size)));
2787
2788         fc->high_water = hwm & E1000_FCRTH_RTH; /* 8-byte granularity */
2789         fc->low_water = (fc->high_water - (2 * adapter->max_frame_size));
2790         fc->low_water &= E1000_FCRTL_RTL; /* 8-byte granularity */
2791
2792         if (adapter->flags & FLAG_DISABLE_FC_PAUSE_TIME)
2793                 fc->pause_time = 0xFFFF;
2794         else
2795                 fc->pause_time = E1000_FC_PAUSE_TIME;
2796         fc->send_xon = 1;
2797         fc->current_mode = fc->requested_mode;
2798
2799         /* Allow time for pending master requests to run */
2800         mac->ops.reset_hw(hw);
2801
2802         /*
2803          * For parts with AMT enabled, let the firmware know
2804          * that the network interface is in control
2805          */
2806         if (adapter->flags & FLAG_HAS_AMT)
2807                 e1000_get_hw_control(adapter);
2808
2809         ew32(WUC, 0);
2810         if (adapter->flags2 & FLAG2_HAS_PHY_WAKEUP)
2811                 e1e_wphy(&adapter->hw, BM_WUC, 0);
2812
2813         if (mac->ops.init_hw(hw))
2814                 e_err("Hardware Error\n");
2815
2816         e1000_update_mng_vlan(adapter);
2817
2818         /* Enable h/w to recognize an 802.1Q VLAN Ethernet packet */
2819         ew32(VET, ETH_P_8021Q);
2820
2821         e1000e_reset_adaptive(hw);
2822         e1000_get_phy_info(hw);
2823
2824         if ((adapter->flags & FLAG_HAS_SMART_POWER_DOWN) &&
2825             !(adapter->flags & FLAG_SMART_POWER_DOWN)) {
2826                 u16 phy_data = 0;
2827                 /*
2828                  * speed up time to link by disabling smart power down, ignore
2829                  * the return value of this function because there is nothing
2830                  * different we would do if it failed
2831                  */
2832                 e1e_rphy(hw, IGP02E1000_PHY_POWER_MGMT, &phy_data);
2833                 phy_data &= ~IGP02E1000_PM_SPD;
2834                 e1e_wphy(hw, IGP02E1000_PHY_POWER_MGMT, phy_data);
2835         }
2836 }
2837
2838 int e1000e_up(struct e1000_adapter *adapter)
2839 {
2840         struct e1000_hw *hw = &adapter->hw;
2841
2842         /* hardware has been reset, we need to reload some things */
2843         e1000_configure(adapter);
2844
2845         clear_bit(__E1000_DOWN, &adapter->state);
2846
2847         napi_enable(&adapter->napi);
2848         if (adapter->msix_entries)
2849                 e1000_configure_msix(adapter);
2850         e1000_irq_enable(adapter);
2851
2852         netif_wake_queue(adapter->netdev);
2853
2854         /* fire a link change interrupt to start the watchdog */
2855         ew32(ICS, E1000_ICS_LSC);
2856         return 0;
2857 }
2858
2859 void e1000e_down(struct e1000_adapter *adapter)
2860 {
2861         struct net_device *netdev = adapter->netdev;
2862         struct e1000_hw *hw = &adapter->hw;
2863         u32 tctl, rctl;
2864
2865         /*
2866          * signal that we're down so the interrupt handler does not
2867          * reschedule our watchdog timer
2868          */
2869         set_bit(__E1000_DOWN, &adapter->state);
2870
2871         /* disable receives in the hardware */
2872         rctl = er32(RCTL);
2873         ew32(RCTL, rctl & ~E1000_RCTL_EN);
2874         /* flush and sleep below */
2875
2876         netif_stop_queue(netdev);
2877
2878         /* disable transmits in the hardware */
2879         tctl = er32(TCTL);
2880         tctl &= ~E1000_TCTL_EN;
2881         ew32(TCTL, tctl);
2882         /* flush both disables and wait for them to finish */
2883         e1e_flush();
2884         msleep(10);
2885
2886         napi_disable(&adapter->napi);
2887         e1000_irq_disable(adapter);
2888
2889         del_timer_sync(&adapter->watchdog_timer);
2890         del_timer_sync(&adapter->phy_info_timer);
2891
2892         netdev->tx_queue_len = adapter->tx_queue_len;
2893         netif_carrier_off(netdev);
2894         adapter->link_speed = 0;
2895         adapter->link_duplex = 0;
2896
2897         if (!pci_channel_offline(adapter->pdev))
2898                 e1000e_reset(adapter);
2899         e1000_clean_tx_ring(adapter);
2900         e1000_clean_rx_ring(adapter);
2901
2902         /*
2903          * TODO: for power management, we could drop the link and
2904          * pci_disable_device here.
2905          */
2906 }
2907
2908 void e1000e_reinit_locked(struct e1000_adapter *adapter)
2909 {
2910         might_sleep();
2911         while (test_and_set_bit(__E1000_RESETTING, &adapter->state))
2912                 msleep(1);
2913         e1000e_down(adapter);
2914         e1000e_up(adapter);
2915         clear_bit(__E1000_RESETTING, &adapter->state);
2916 }
2917
2918 /**
2919  * e1000_sw_init - Initialize general software structures (struct e1000_adapter)
2920  * @adapter: board private structure to initialize
2921  *
2922  * e1000_sw_init initializes the Adapter private data structure.
2923  * Fields are initialized based on PCI device information and
2924  * OS network device settings (MTU size).
2925  **/
2926 static int __devinit e1000_sw_init(struct e1000_adapter *adapter)
2927 {
2928         struct net_device *netdev = adapter->netdev;
2929
2930         adapter->rx_buffer_len = ETH_FRAME_LEN + VLAN_HLEN + ETH_FCS_LEN;
2931         adapter->rx_ps_bsize0 = 128;
2932         adapter->max_frame_size = netdev->mtu + ETH_HLEN + ETH_FCS_LEN;
2933         adapter->min_frame_size = ETH_ZLEN + ETH_FCS_LEN;
2934
2935         e1000e_set_interrupt_capability(adapter);
2936
2937         if (e1000_alloc_queues(adapter))
2938                 return -ENOMEM;
2939
2940         /* Explicitly disable IRQ since the NIC can be in any state. */
2941         e1000_irq_disable(adapter);
2942
2943         set_bit(__E1000_DOWN, &adapter->state);
2944         return 0;
2945 }
2946
2947 /**
2948  * e1000_intr_msi_test - Interrupt Handler
2949  * @irq: interrupt number
2950  * @data: pointer to a network interface device structure
2951  **/
2952 static irqreturn_t e1000_intr_msi_test(int irq, void *data)
2953 {
2954         struct net_device *netdev = data;
2955         struct e1000_adapter *adapter = netdev_priv(netdev);
2956         struct e1000_hw *hw = &adapter->hw;
2957         u32 icr = er32(ICR);
2958
2959         e_dbg("%s: icr is %08X\n", netdev->name, icr);
2960         if (icr & E1000_ICR_RXSEQ) {
2961                 adapter->flags &= ~FLAG_MSI_TEST_FAILED;
2962                 wmb();
2963         }
2964
2965         return IRQ_HANDLED;
2966 }
2967
2968 /**
2969  * e1000_test_msi_interrupt - Returns 0 for successful test
2970  * @adapter: board private struct
2971  *
2972  * code flow taken from tg3.c
2973  **/
2974 static int e1000_test_msi_interrupt(struct e1000_adapter *adapter)
2975 {
2976         struct net_device *netdev = adapter->netdev;
2977         struct e1000_hw *hw = &adapter->hw;
2978         int err;
2979
2980         /* poll_enable hasn't been called yet, so don't need disable */
2981         /* clear any pending events */
2982         er32(ICR);
2983
2984         /* free the real vector and request a test handler */
2985         e1000_free_irq(adapter);
2986         e1000e_reset_interrupt_capability(adapter);
2987
2988         /* Assume that the test fails, if it succeeds then the test
2989          * MSI irq handler will unset this flag */
2990         adapter->flags |= FLAG_MSI_TEST_FAILED;
2991
2992         err = pci_enable_msi(adapter->pdev);
2993         if (err)
2994                 goto msi_test_failed;
2995
2996         err = request_irq(adapter->pdev->irq, &e1000_intr_msi_test, 0,
2997                           netdev->name, netdev);
2998         if (err) {
2999                 pci_disable_msi(adapter->pdev);
3000                 goto msi_test_failed;
3001         }
3002
3003         wmb();
3004
3005         e1000_irq_enable(adapter);
3006
3007         /* fire an unusual interrupt on the test handler */
3008         ew32(ICS, E1000_ICS_RXSEQ);
3009         e1e_flush();
3010         msleep(50);
3011
3012         e1000_irq_disable(adapter);
3013
3014         rmb();
3015
3016         if (adapter->flags & FLAG_MSI_TEST_FAILED) {
3017                 adapter->int_mode = E1000E_INT_MODE_LEGACY;
3018                 err = -EIO;
3019                 e_info("MSI interrupt test failed!\n");
3020         }
3021
3022         free_irq(adapter->pdev->irq, netdev);
3023         pci_disable_msi(adapter->pdev);
3024
3025         if (err == -EIO)
3026                 goto msi_test_failed;
3027
3028         /* okay so the test worked, restore settings */
3029         e_dbg("%s: MSI interrupt test succeeded!\n", netdev->name);
3030 msi_test_failed:
3031         e1000e_set_interrupt_capability(adapter);
3032         e1000_request_irq(adapter);
3033         return err;
3034 }
3035
3036 /**
3037  * e1000_test_msi - Returns 0 if MSI test succeeds or INTx mode is restored
3038  * @adapter: board private struct
3039  *
3040  * code flow taken from tg3.c, called with e1000 interrupts disabled.
3041  **/
3042 static int e1000_test_msi(struct e1000_adapter *adapter)
3043 {
3044         int err;
3045         u16 pci_cmd;
3046
3047         if (!(adapter->flags & FLAG_MSI_ENABLED))
3048                 return 0;
3049
3050         /* disable SERR in case the MSI write causes a master abort */
3051         pci_read_config_word(adapter->pdev, PCI_COMMAND, &pci_cmd);
3052         pci_write_config_word(adapter->pdev, PCI_COMMAND,
3053                               pci_cmd & ~PCI_COMMAND_SERR);
3054
3055         err = e1000_test_msi_interrupt(adapter);
3056
3057         /* restore previous setting of command word */
3058         pci_write_config_word(adapter->pdev, PCI_COMMAND, pci_cmd);
3059
3060         /* success ! */
3061         if (!err)
3062                 return 0;
3063
3064         /* EIO means MSI test failed */
3065         if (err != -EIO)
3066                 return err;
3067
3068         /* back to INTx mode */
3069         e_warn("MSI interrupt test failed, using legacy interrupt.\n");
3070
3071         e1000_free_irq(adapter);
3072
3073         err = e1000_request_irq(adapter);
3074
3075         return err;
3076 }
3077
3078 /**
3079  * e1000_open - Called when a network interface is made active
3080  * @netdev: network interface device structure
3081  *
3082  * Returns 0 on success, negative value on failure
3083  *
3084  * The open entry point is called when a network interface is made
3085  * active by the system (IFF_UP).  At this point all resources needed
3086  * for transmit and receive operations are allocated, the interrupt
3087  * handler is registered with the OS, the watchdog timer is started,
3088  * and the stack is notified that the interface is ready.
3089  **/
3090 static int e1000_open(struct net_device *netdev)
3091 {
3092         struct e1000_adapter *adapter = netdev_priv(netdev);
3093         struct e1000_hw *hw = &adapter->hw;
3094         int err;
3095
3096         /* disallow open during test */
3097         if (test_bit(__E1000_TESTING, &adapter->state))
3098                 return -EBUSY;
3099
3100         netif_carrier_off(netdev);
3101
3102         /* allocate transmit descriptors */
3103         err = e1000e_setup_tx_resources(adapter);
3104         if (err)
3105                 goto err_setup_tx;
3106
3107         /* allocate receive descriptors */
3108         err = e1000e_setup_rx_resources(adapter);
3109         if (err)
3110                 goto err_setup_rx;
3111
3112         e1000e_power_up_phy(adapter);
3113
3114         adapter->mng_vlan_id = E1000_MNG_VLAN_NONE;
3115         if ((adapter->hw.mng_cookie.status &
3116              E1000_MNG_DHCP_COOKIE_STATUS_VLAN))
3117                 e1000_update_mng_vlan(adapter);
3118
3119         /*
3120          * If AMT is enabled, let the firmware know that the network
3121          * interface is now open
3122          */
3123         if (adapter->flags & FLAG_HAS_AMT)
3124                 e1000_get_hw_control(adapter);
3125
3126         /*
3127          * before we allocate an interrupt, we must be ready to handle it.
3128          * Setting DEBUG_SHIRQ in the kernel makes it fire an interrupt
3129          * as soon as we call pci_request_irq, so we have to setup our
3130          * clean_rx handler before we do so.
3131          */
3132         e1000_configure(adapter);
3133
3134         err = e1000_request_irq(adapter);
3135         if (err)
3136                 goto err_req_irq;
3137
3138         /*
3139          * Work around PCIe errata with MSI interrupts causing some chipsets to
3140          * ignore e1000e MSI messages, which means we need to test our MSI
3141          * interrupt now
3142          */
3143         if (adapter->int_mode != E1000E_INT_MODE_LEGACY) {
3144                 err = e1000_test_msi(adapter);
3145                 if (err) {
3146                         e_err("Interrupt allocation failed\n");
3147                         goto err_req_irq;
3148                 }
3149         }
3150
3151         /* From here on the code is the same as e1000e_up() */
3152         clear_bit(__E1000_DOWN, &adapter->state);
3153
3154         napi_enable(&adapter->napi);
3155
3156         e1000_irq_enable(adapter);
3157
3158         netif_start_queue(netdev);
3159
3160         /* fire a link status change interrupt to start the watchdog */
3161         ew32(ICS, E1000_ICS_LSC);
3162
3163         return 0;
3164
3165 err_req_irq:
3166         e1000_release_hw_control(adapter);
3167         e1000_power_down_phy(adapter);
3168         e1000e_free_rx_resources(adapter);
3169 err_setup_rx:
3170         e1000e_free_tx_resources(adapter);
3171 err_setup_tx:
3172         e1000e_reset(adapter);
3173
3174         return err;
3175 }
3176
3177 /**
3178  * e1000_close - Disables a network interface
3179  * @netdev: network interface device structure
3180  *
3181  * Returns 0, this is not allowed to fail
3182  *
3183  * The close entry point is called when an interface is de-activated
3184  * by the OS.  The hardware is still under the drivers control, but
3185  * needs to be disabled.  A global MAC reset is issued to stop the
3186  * hardware, and all transmit and receive resources are freed.
3187  **/
3188 static int e1000_close(struct net_device *netdev)
3189 {
3190         struct e1000_adapter *adapter = netdev_priv(netdev);
3191
3192         WARN_ON(test_bit(__E1000_RESETTING, &adapter->state));
3193         e1000e_down(adapter);
3194         e1000_power_down_phy(adapter);
3195         e1000_free_irq(adapter);
3196
3197         e1000e_free_tx_resources(adapter);
3198         e1000e_free_rx_resources(adapter);
3199
3200         /*
3201          * kill manageability vlan ID if supported, but not if a vlan with
3202          * the same ID is registered on the host OS (let 8021q kill it)
3203          */
3204         if ((adapter->hw.mng_cookie.status &
3205                           E1000_MNG_DHCP_COOKIE_STATUS_VLAN) &&
3206              !(adapter->vlgrp &&
3207                vlan_group_get_device(adapter->vlgrp, adapter->mng_vlan_id)))
3208                 e1000_vlan_rx_kill_vid(netdev, adapter->mng_vlan_id);
3209
3210         /*
3211          * If AMT is enabled, let the firmware know that the network
3212          * interface is now closed
3213          */
3214         if (adapter->flags & FLAG_HAS_AMT)
3215                 e1000_release_hw_control(adapter);
3216
3217         return 0;
3218 }
3219 /**
3220  * e1000_set_mac - Change the Ethernet Address of the NIC
3221  * @netdev: network interface device structure
3222  * @p: pointer to an address structure
3223  *
3224  * Returns 0 on success, negative on failure
3225  **/
3226 static int e1000_set_mac(struct net_device *netdev, void *p)
3227 {
3228         struct e1000_adapter *adapter = netdev_priv(netdev);
3229         struct sockaddr *addr = p;
3230
3231         if (!is_valid_ether_addr(addr->sa_data))
3232                 return -EADDRNOTAVAIL;
3233
3234         memcpy(netdev->dev_addr, addr->sa_data, netdev->addr_len);
3235         memcpy(adapter->hw.mac.addr, addr->sa_data, netdev->addr_len);
3236
3237         e1000e_rar_set(&adapter->hw, adapter->hw.mac.addr, 0);
3238
3239         if (adapter->flags & FLAG_RESET_OVERWRITES_LAA) {
3240                 /* activate the work around */
3241                 e1000e_set_laa_state_82571(&adapter->hw, 1);
3242
3243                 /*
3244                  * Hold a copy of the LAA in RAR[14] This is done so that
3245                  * between the time RAR[0] gets clobbered  and the time it
3246                  * gets fixed (in e1000_watchdog), the actual LAA is in one
3247                  * of the RARs and no incoming packets directed to this port
3248                  * are dropped. Eventually the LAA will be in RAR[0] and
3249                  * RAR[14]
3250                  */
3251                 e1000e_rar_set(&adapter->hw,
3252                               adapter->hw.mac.addr,
3253                               adapter->hw.mac.rar_entry_count - 1);
3254         }
3255
3256         return 0;
3257 }
3258
3259 /**
3260  * e1000e_update_phy_task - work thread to update phy
3261  * @work: pointer to our work struct
3262  *
3263  * this worker thread exists because we must acquire a
3264  * semaphore to read the phy, which we could msleep while
3265  * waiting for it, and we can't msleep in a timer.
3266  **/
3267 static void e1000e_update_phy_task(struct work_struct *work)
3268 {
3269         struct e1000_adapter *adapter = container_of(work,
3270                                         struct e1000_adapter, update_phy_task);
3271         e1000_get_phy_info(&adapter->hw);
3272 }
3273
3274 /*
3275  * Need to wait a few seconds after link up to get diagnostic information from
3276  * the phy
3277  */
3278 static void e1000_update_phy_info(unsigned long data)
3279 {
3280         struct e1000_adapter *adapter = (struct e1000_adapter *) data;
3281         schedule_work(&adapter->update_phy_task);
3282 }
3283
3284 /**
3285  * e1000e_update_stats - Update the board statistics counters
3286  * @adapter: board private structure
3287  **/
3288 void e1000e_update_stats(struct e1000_adapter *adapter)
3289 {
3290         struct e1000_hw *hw = &adapter->hw;
3291         struct pci_dev *pdev = adapter->pdev;
3292         u16 phy_data;
3293
3294         /*
3295          * Prevent stats update while adapter is being reset, or if the pci
3296          * connection is down.
3297          */
3298         if (adapter->link_speed == 0)
3299                 return;
3300         if (pci_channel_offline(pdev))
3301                 return;
3302
3303         adapter->stats.crcerrs += er32(CRCERRS);
3304         adapter->stats.gprc += er32(GPRC);
3305         adapter->stats.gorc += er32(GORCL);
3306         er32(GORCH); /* Clear gorc */
3307         adapter->stats.bprc += er32(BPRC);
3308         adapter->stats.mprc += er32(MPRC);
3309         adapter->stats.roc += er32(ROC);
3310
3311         adapter->stats.mpc += er32(MPC);
3312         if ((hw->phy.type == e1000_phy_82578) ||
3313             (hw->phy.type == e1000_phy_82577)) {
3314                 e1e_rphy(hw, HV_SCC_UPPER, &phy_data);
3315                 e1e_rphy(hw, HV_SCC_LOWER, &phy_data);
3316                 adapter->stats.scc += phy_data;
3317
3318                 e1e_rphy(hw, HV_ECOL_UPPER, &phy_data);
3319                 e1e_rphy(hw, HV_ECOL_LOWER, &phy_data);
3320                 adapter->stats.ecol += phy_data;
3321
3322                 e1e_rphy(hw, HV_MCC_UPPER, &phy_data);
3323                 e1e_rphy(hw, HV_MCC_LOWER, &phy_data);
3324                 adapter->stats.mcc += phy_data;
3325
3326                 e1e_rphy(hw, HV_LATECOL_UPPER, &phy_data);
3327                 e1e_rphy(hw, HV_LATECOL_LOWER, &phy_data);
3328                 adapter->stats.latecol += phy_data;
3329
3330                 e1e_rphy(hw, HV_DC_UPPER, &phy_data);
3331                 e1e_rphy(hw, HV_DC_LOWER, &phy_data);
3332                 adapter->stats.dc += phy_data;
3333         } else {
3334                 adapter->stats.scc += er32(SCC);
3335                 adapter->stats.ecol += er32(ECOL);
3336                 adapter->stats.mcc += er32(MCC);
3337                 adapter->stats.latecol += er32(LATECOL);
3338                 adapter->stats.dc += er32(DC);
3339         }
3340         adapter->stats.xonrxc += er32(XONRXC);
3341         adapter->stats.xontxc += er32(XONTXC);
3342         adapter->stats.xoffrxc += er32(XOFFRXC);
3343         adapter->stats.xofftxc += er32(XOFFTXC);
3344         adapter->stats.gptc += er32(GPTC);
3345         adapter->stats.gotc += er32(GOTCL);
3346         er32(GOTCH); /* Clear gotc */
3347         adapter->stats.rnbc += er32(RNBC);
3348         adapter->stats.ruc += er32(RUC);
3349
3350         adapter->stats.mptc += er32(MPTC);
3351         adapter->stats.bptc += er32(BPTC);
3352
3353         /* used for adaptive IFS */
3354
3355         hw->mac.tx_packet_delta = er32(TPT);
3356         adapter->stats.tpt += hw->mac.tx_packet_delta;
3357         if ((hw->phy.type == e1000_phy_82578) ||
3358             (hw->phy.type == e1000_phy_82577)) {
3359                 e1e_rphy(hw, HV_COLC_UPPER, &phy_data);
3360                 e1e_rphy(hw, HV_COLC_LOWER, &phy_data);
3361                 hw->mac.collision_delta = phy_data;
3362         } else {
3363                 hw->mac.collision_delta = er32(COLC);
3364         }
3365         adapter->stats.colc += hw->mac.collision_delta;
3366
3367         adapter->stats.algnerrc += er32(ALGNERRC);
3368         adapter->stats.rxerrc += er32(RXERRC);
3369         if ((hw->phy.type == e1000_phy_82578) ||
3370             (hw->phy.type == e1000_phy_82577)) {
3371                 e1e_rphy(hw, HV_TNCRS_UPPER, &phy_data);
3372                 e1e_rphy(hw, HV_TNCRS_LOWER, &phy_data);
3373                 adapter->stats.tncrs += phy_data;
3374         } else {
3375                 if ((hw->mac.type != e1000_82574) &&
3376                     (hw->mac.type != e1000_82583))
3377                         adapter->stats.tncrs += er32(TNCRS);
3378         }
3379         adapter->stats.cexterr += er32(CEXTERR);
3380         adapter->stats.tsctc += er32(TSCTC);
3381         adapter->stats.tsctfc += er32(TSCTFC);
3382
3383         /* Fill out the OS statistics structure */
3384         adapter->net_stats.multicast = adapter->stats.mprc;
3385         adapter->net_stats.collisions = adapter->stats.colc;
3386
3387         /* Rx Errors */
3388
3389         /*
3390          * RLEC on some newer hardware can be incorrect so build
3391          * our own version based on RUC and ROC
3392          */
3393         adapter->net_stats.rx_errors = adapter->stats.rxerrc +
3394                 adapter->stats.crcerrs + adapter->stats.algnerrc +
3395                 adapter->stats.ruc + adapter->stats.roc +
3396                 adapter->stats.cexterr;
3397         adapter->net_stats.rx_length_errors = adapter->stats.ruc +
3398                                               adapter->stats.roc;
3399         adapter->net_stats.rx_crc_errors = adapter->stats.crcerrs;
3400         adapter->net_stats.rx_frame_errors = adapter->stats.algnerrc;
3401         adapter->net_stats.rx_missed_errors = adapter->stats.mpc;
3402
3403         /* Tx Errors */
3404         adapter->net_stats.tx_errors = adapter->stats.ecol +
3405                                        adapter->stats.latecol;
3406         adapter->net_stats.tx_aborted_errors = adapter->stats.ecol;
3407         adapter->net_stats.tx_window_errors = adapter->stats.latecol;
3408         adapter->net_stats.tx_carrier_errors = adapter->stats.tncrs;
3409
3410         /* Tx Dropped needs to be maintained elsewhere */
3411
3412         /* Management Stats */
3413         adapter->stats.mgptc += er32(MGTPTC);
3414         adapter->stats.mgprc += er32(MGTPRC);
3415         adapter->stats.mgpdc += er32(MGTPDC);
3416 }
3417
3418 /**
3419  * e1000_phy_read_status - Update the PHY register status snapshot
3420  * @adapter: board private structure
3421  **/
3422 static void e1000_phy_read_status(struct e1000_adapter *adapter)
3423 {
3424         struct e1000_hw *hw = &adapter->hw;
3425         struct e1000_phy_regs *phy = &adapter->phy_regs;
3426         int ret_val;
3427
3428         if ((er32(STATUS) & E1000_STATUS_LU) &&
3429             (adapter->hw.phy.media_type == e1000_media_type_copper)) {
3430                 ret_val  = e1e_rphy(hw, PHY_CONTROL, &phy->bmcr);
3431                 ret_val |= e1e_rphy(hw, PHY_STATUS, &phy->bmsr);
3432                 ret_val |= e1e_rphy(hw, PHY_AUTONEG_ADV, &phy->advertise);
3433                 ret_val |= e1e_rphy(hw, PHY_LP_ABILITY, &phy->lpa);
3434                 ret_val |= e1e_rphy(hw, PHY_AUTONEG_EXP, &phy->expansion);
3435                 ret_val |= e1e_rphy(hw, PHY_1000T_CTRL, &phy->ctrl1000);
3436                 ret_val |= e1e_rphy(hw, PHY_1000T_STATUS, &phy->stat1000);
3437                 ret_val |= e1e_rphy(hw, PHY_EXT_STATUS, &phy->estatus);
3438                 if (ret_val)
3439                         e_warn("Error reading PHY register\n");
3440         } else {
3441                 /*
3442                  * Do not read PHY registers if link is not up
3443                  * Set values to typical power-on defaults
3444                  */
3445                 phy->bmcr = (BMCR_SPEED1000 | BMCR_ANENABLE | BMCR_FULLDPLX);
3446                 phy->bmsr = (BMSR_100FULL | BMSR_100HALF | BMSR_10FULL |
3447                              BMSR_10HALF | BMSR_ESTATEN | BMSR_ANEGCAPABLE |
3448                              BMSR_ERCAP);
3449                 phy->advertise = (ADVERTISE_PAUSE_ASYM | ADVERTISE_PAUSE_CAP |
3450                                   ADVERTISE_ALL | ADVERTISE_CSMA);
3451                 phy->lpa = 0;
3452                 phy->expansion = EXPANSION_ENABLENPAGE;
3453                 phy->ctrl1000 = ADVERTISE_1000FULL;
3454                 phy->stat1000 = 0;
3455                 phy->estatus = (ESTATUS_1000_TFULL | ESTATUS_1000_THALF);
3456         }
3457 }
3458
3459 static void e1000_print_link_info(struct e1000_adapter *adapter)
3460 {
3461         struct e1000_hw *hw = &adapter->hw;
3462         u32 ctrl = er32(CTRL);
3463
3464         /* Link status message must follow this format for user tools */
3465         printk(KERN_INFO "e1000e: %s NIC Link is Up %d Mbps %s, "
3466                "Flow Control: %s\n",
3467                adapter->netdev->name,
3468                adapter->link_speed,
3469                (adapter->link_duplex == FULL_DUPLEX) ?
3470                                 "Full Duplex" : "Half Duplex",
3471                ((ctrl & E1000_CTRL_TFCE) && (ctrl & E1000_CTRL_RFCE)) ?
3472                                 "RX/TX" :
3473                ((ctrl & E1000_CTRL_RFCE) ? "RX" :
3474                ((ctrl & E1000_CTRL_TFCE) ? "TX" : "None" )));
3475 }
3476
3477 bool e1000_has_link(struct e1000_adapter *adapter)
3478 {
3479         struct e1000_hw *hw = &adapter->hw;
3480         bool link_active = 0;
3481         s32 ret_val = 0;
3482
3483         /*
3484          * get_link_status is set on LSC (link status) interrupt or
3485          * Rx sequence error interrupt.  get_link_status will stay
3486          * false until the check_for_link establishes link
3487          * for copper adapters ONLY
3488          */
3489         switch (hw->phy.media_type) {
3490         case e1000_media_type_copper:
3491                 if (hw->mac.get_link_status) {
3492                         ret_val = hw->mac.ops.check_for_link(hw);
3493                         link_active = !hw->mac.get_link_status;
3494                 } else {
3495                         link_active = 1;
3496                 }
3497                 break;
3498         case e1000_media_type_fiber:
3499                 ret_val = hw->mac.ops.check_for_link(hw);
3500                 link_active = !!(er32(STATUS) & E1000_STATUS_LU);
3501                 break;
3502         case e1000_media_type_internal_serdes:
3503                 ret_val = hw->mac.ops.check_for_link(hw);
3504                 link_active = adapter->hw.mac.serdes_has_link;
3505                 break;
3506         default:
3507         case e1000_media_type_unknown:
3508                 break;
3509         }
3510
3511         if ((ret_val == E1000_ERR_PHY) && (hw->phy.type == e1000_phy_igp_3) &&
3512             (er32(CTRL) & E1000_PHY_CTRL_GBE_DISABLE)) {
3513                 /* See e1000_kmrn_lock_loss_workaround_ich8lan() */
3514                 e_info("Gigabit has been disabled, downgrading speed\n");
3515         }
3516
3517         return link_active;
3518 }
3519
3520 static void e1000e_enable_receives(struct e1000_adapter *adapter)
3521 {
3522         /* make sure the receive unit is started */
3523         if ((adapter->flags & FLAG_RX_NEEDS_RESTART) &&
3524             (adapter->flags & FLAG_RX_RESTART_NOW)) {
3525                 struct e1000_hw *hw = &adapter->hw;
3526                 u32 rctl = er32(RCTL);
3527                 ew32(RCTL, rctl | E1000_RCTL_EN);
3528                 adapter->flags &= ~FLAG_RX_RESTART_NOW;
3529         }
3530 }
3531
3532 /**
3533  * e1000_watchdog - Timer Call-back
3534  * @data: pointer to adapter cast into an unsigned long
3535  **/
3536 static void e1000_watchdog(unsigned long data)
3537 {
3538         struct e1000_adapter *adapter = (struct e1000_adapter *) data;
3539
3540         /* Do the rest outside of interrupt context */
3541         schedule_work(&adapter->watchdog_task);
3542
3543         /* TODO: make this use queue_delayed_work() */
3544 }
3545
3546 static void e1000_watchdog_task(struct work_struct *work)
3547 {
3548         struct e1000_adapter *adapter = container_of(work,
3549                                         struct e1000_adapter, watchdog_task);
3550         struct net_device *netdev = adapter->netdev;
3551         struct e1000_mac_info *mac = &adapter->hw.mac;
3552         struct e1000_phy_info *phy = &adapter->hw.phy;
3553         struct e1000_ring *tx_ring = adapter->tx_ring;
3554         struct e1000_hw *hw = &adapter->hw;
3555         u32 link, tctl;
3556         int tx_pending = 0;
3557
3558         link = e1000_has_link(adapter);
3559         if ((netif_carrier_ok(netdev)) && link) {
3560                 e1000e_enable_receives(adapter);
3561                 goto link_up;
3562         }
3563
3564         if ((e1000e_enable_tx_pkt_filtering(hw)) &&
3565             (adapter->mng_vlan_id != adapter->hw.mng_cookie.vlan_id))
3566                 e1000_update_mng_vlan(adapter);
3567
3568         if (link) {
3569                 if (!netif_carrier_ok(netdev)) {
3570                         bool txb2b = 1;
3571                         /* update snapshot of PHY registers on LSC */
3572                         e1000_phy_read_status(adapter);
3573                         mac->ops.get_link_up_info(&adapter->hw,
3574                                                    &adapter->link_speed,
3575                                                    &adapter->link_duplex);
3576                         e1000_print_link_info(adapter);
3577                         /*
3578                          * On supported PHYs, check for duplex mismatch only
3579                          * if link has autonegotiated at 10/100 half
3580                          */
3581                         if ((hw->phy.type == e1000_phy_igp_3 ||
3582                              hw->phy.type == e1000_phy_bm) &&
3583                             (hw->mac.autoneg == true) &&
3584                             (adapter->link_speed == SPEED_10 ||
3585                              adapter->link_speed == SPEED_100) &&
3586                             (adapter->link_duplex == HALF_DUPLEX)) {
3587                                 u16 autoneg_exp;
3588
3589                                 e1e_rphy(hw, PHY_AUTONEG_EXP, &autoneg_exp);
3590
3591                                 if (!(autoneg_exp & NWAY_ER_LP_NWAY_CAPS))
3592                                         e_info("Autonegotiated half duplex but"
3593                                                " link partner cannot autoneg. "
3594                                                " Try forcing full duplex if "
3595                                                "link gets many collisions.\n");
3596                         }
3597
3598                         /*
3599                          * tweak tx_queue_len according to speed/duplex
3600                          * and adjust the timeout factor
3601                          */
3602                         netdev->tx_queue_len = adapter->tx_queue_len;
3603                         adapter->tx_timeout_factor = 1;
3604                         switch (adapter->link_speed) {
3605                         case SPEED_10:
3606                                 txb2b = 0;
3607                                 netdev->tx_queue_len = 10;
3608                                 adapter->tx_timeout_factor = 16;
3609                                 break;
3610                         case SPEED_100:
3611                                 txb2b = 0;
3612                                 netdev->tx_queue_len = 100;
3613                                 /* maybe add some timeout factor ? */
3614                                 break;
3615                         }
3616
3617                         /*
3618                          * workaround: re-program speed mode bit after
3619                          * link-up event
3620                          */
3621                         if ((adapter->flags & FLAG_TARC_SPEED_MODE_BIT) &&
3622                             !txb2b) {
3623                                 u32 tarc0;
3624                                 tarc0 = er32(TARC(0));
3625                                 tarc0 &= ~SPEED_MODE_BIT;
3626                                 ew32(TARC(0), tarc0);
3627                         }
3628
3629                         /*
3630                          * disable TSO for pcie and 10/100 speeds, to avoid
3631                          * some hardware issues
3632                          */
3633                         if (!(adapter->flags & FLAG_TSO_FORCE)) {
3634                                 switch (adapter->link_speed) {
3635                                 case SPEED_10:
3636                                 case SPEED_100:
3637                                         e_info("10/100 speed: disabling TSO\n");
3638                                         netdev->features &= ~NETIF_F_TSO;
3639                                         netdev->features &= ~NETIF_F_TSO6;
3640                                         break;
3641                                 case SPEED_1000:
3642                                         netdev->features |= NETIF_F_TSO;
3643                                         netdev->features |= NETIF_F_TSO6;
3644                                         break;
3645                                 default:
3646                                         /* oops */
3647                                         break;
3648                                 }
3649                         }
3650
3651                         /*
3652                          * enable transmits in the hardware, need to do this
3653                          * after setting TARC(0)
3654                          */
3655                         tctl = er32(TCTL);
3656                         tctl |= E1000_TCTL_EN;
3657                         ew32(TCTL, tctl);
3658
3659                         /*
3660                          * Perform any post-link-up configuration before
3661                          * reporting link up.
3662                          */
3663                         if (phy->ops.cfg_on_link_up)
3664                                 phy->ops.cfg_on_link_up(hw);
3665
3666                         netif_carrier_on(netdev);
3667
3668                         if (!test_bit(__E1000_DOWN, &adapter->state))
3669                                 mod_timer(&adapter->phy_info_timer,
3670                                           round_jiffies(jiffies + 2 * HZ));
3671                 }
3672         } else {
3673                 if (netif_carrier_ok(netdev)) {
3674                         adapter->link_speed = 0;
3675                         adapter->link_duplex = 0;
3676                         /* Link status message must follow this format */
3677                         printk(KERN_INFO "e1000e: %s NIC Link is Down\n",
3678                                adapter->netdev->name);
3679                         netif_carrier_off(netdev);
3680                         if (!test_bit(__E1000_DOWN, &adapter->state))
3681                                 mod_timer(&adapter->phy_info_timer,
3682                                           round_jiffies(jiffies + 2 * HZ));
3683
3684                         if (adapter->flags & FLAG_RX_NEEDS_RESTART)
3685                                 schedule_work(&adapter->reset_task);
3686                 }
3687         }
3688
3689 link_up:
3690         e1000e_update_stats(adapter);
3691
3692         mac->tx_packet_delta = adapter->stats.tpt - adapter->tpt_old;
3693         adapter->tpt_old = adapter->stats.tpt;
3694         mac->collision_delta = adapter->stats.colc - adapter->colc_old;
3695         adapter->colc_old = adapter->stats.colc;
3696
3697         adapter->gorc = adapter->stats.gorc - adapter->gorc_old;
3698         adapter->gorc_old = adapter->stats.gorc;
3699         adapter->gotc = adapter->stats.gotc - adapter->gotc_old;
3700         adapter->gotc_old = adapter->stats.gotc;
3701
3702         e1000e_update_adaptive(&adapter->hw);
3703
3704         if (!netif_carrier_ok(netdev)) {
3705                 tx_pending = (e1000_desc_unused(tx_ring) + 1 <
3706                                tx_ring->count);
3707                 if (tx_pending) {
3708                         /*
3709                          * We've lost link, so the controller stops DMA,
3710                          * but we've got queued Tx work that's never going
3711                          * to get done, so reset controller to flush Tx.
3712                          * (Do the reset outside of interrupt context).
3713                          */
3714                         adapter->tx_timeout_count++;
3715                         schedule_work(&adapter->reset_task);
3716                         /* return immediately since reset is imminent */
3717                         return;
3718                 }
3719         }
3720
3721         /* Cause software interrupt to ensure Rx ring is cleaned */
3722         if (adapter->msix_entries)
3723                 ew32(ICS, adapter->rx_ring->ims_val);
3724         else
3725                 ew32(ICS, E1000_ICS_RXDMT0);
3726
3727         /* Force detection of hung controller every watchdog period */
3728         adapter->detect_tx_hung = 1;
3729
3730         /*
3731          * With 82571 controllers, LAA may be overwritten due to controller
3732          * reset from the other port. Set the appropriate LAA in RAR[0]
3733          */
3734         if (e1000e_get_laa_state_82571(hw))
3735                 e1000e_rar_set(hw, adapter->hw.mac.addr, 0);
3736
3737         /* Reset the timer */
3738         if (!test_bit(__E1000_DOWN, &adapter->state))
3739                 mod_timer(&adapter->watchdog_timer,
3740                           round_jiffies(jiffies + 2 * HZ));
3741 }
3742
3743 #define E1000_TX_FLAGS_CSUM             0x00000001
3744 #define E1000_TX_FLAGS_VLAN             0x00000002
3745 #define E1000_TX_FLAGS_TSO              0x00000004
3746 #define E1000_TX_FLAGS_IPV4             0x00000008
3747 #define E1000_TX_FLAGS_VLAN_MASK        0xffff0000
3748 #define E1000_TX_FLAGS_VLAN_SHIFT       16
3749
3750 static int e1000_tso(struct e1000_adapter *adapter,
3751                      struct sk_buff *skb)
3752 {
3753         struct e1000_ring *tx_ring = adapter->tx_ring;
3754         struct e1000_context_desc *context_desc;
3755         struct e1000_buffer *buffer_info;
3756         unsigned int i;
3757         u32 cmd_length = 0;
3758         u16 ipcse = 0, tucse, mss;
3759         u8 ipcss, ipcso, tucss, tucso, hdr_len;
3760         int err;
3761
3762         if (skb_is_gso(skb)) {
3763                 if (skb_header_cloned(skb)) {
3764                         err = pskb_expand_head(skb, 0, 0, GFP_ATOMIC);
3765                         if (err)
3766                                 return err;
3767                 }
3768
3769                 hdr_len = skb_transport_offset(skb) + tcp_hdrlen(skb);
3770                 mss = skb_shinfo(skb)->gso_size;
3771                 if (skb->protocol == htons(ETH_P_IP)) {
3772                         struct iphdr *iph = ip_hdr(skb);
3773                         iph->tot_len = 0;
3774                         iph->check = 0;
3775                         tcp_hdr(skb)->check = ~csum_tcpudp_magic(iph->saddr,
3776                                                                  iph->daddr, 0,
3777                                                                  IPPROTO_TCP,
3778                                                                  0);
3779                         cmd_length = E1000_TXD_CMD_IP;
3780                         ipcse = skb_transport_offset(skb) - 1;
3781                 } else if (skb_shinfo(skb)->gso_type == SKB_GSO_TCPV6) {
3782                         ipv6_hdr(skb)->payload_len = 0;
3783                         tcp_hdr(skb)->check =
3784                                 ~csum_ipv6_magic(&ipv6_hdr(skb)->saddr,
3785                                                  &ipv6_hdr(skb)->daddr,
3786                                                  0, IPPROTO_TCP, 0);
3787                         ipcse = 0;
3788                 }
3789                 ipcss = skb_network_offset(skb);
3790                 ipcso = (void *)&(ip_hdr(skb)->check) - (void *)skb->data;
3791                 tucss = skb_transport_offset(skb);
3792                 tucso = (void *)&(tcp_hdr(skb)->check) - (void *)skb->data;
3793                 tucse = 0;
3794
3795                 cmd_length |= (E1000_TXD_CMD_DEXT | E1000_TXD_CMD_TSE |
3796                                E1000_TXD_CMD_TCP | (skb->len - (hdr_len)));
3797
3798                 i = tx_ring->next_to_use;
3799                 context_desc = E1000_CONTEXT_DESC(*tx_ring, i);
3800                 buffer_info = &tx_ring->buffer_info[i];
3801
3802                 context_desc->lower_setup.ip_fields.ipcss  = ipcss;
3803                 context_desc->lower_setup.ip_fields.ipcso  = ipcso;
3804                 context_desc->lower_setup.ip_fields.ipcse  = cpu_to_le16(ipcse);
3805                 context_desc->upper_setup.tcp_fields.tucss = tucss;
3806                 context_desc->upper_setup.tcp_fields.tucso = tucso;
3807                 context_desc->upper_setup.tcp_fields.tucse = cpu_to_le16(tucse);
3808                 context_desc->tcp_seg_setup.fields.mss     = cpu_to_le16(mss);
3809                 context_desc->tcp_seg_setup.fields.hdr_len = hdr_len;
3810                 context_desc->cmd_and_length = cpu_to_le32(cmd_length);
3811
3812                 buffer_info->time_stamp = jiffies;
3813                 buffer_info->next_to_watch = i;
3814
3815                 i++;
3816                 if (i == tx_ring->count)
3817                         i = 0;
3818                 tx_ring->next_to_use = i;
3819
3820                 return 1;
3821         }
3822
3823         return 0;
3824 }
3825
3826 static bool e1000_tx_csum(struct e1000_adapter *adapter, struct sk_buff *skb)
3827 {
3828         struct e1000_ring *tx_ring = adapter->tx_ring;
3829         struct e1000_context_desc *context_desc;
3830         struct e1000_buffer *buffer_info;
3831         unsigned int i;
3832         u8 css;
3833         u32 cmd_len = E1000_TXD_CMD_DEXT;
3834         __be16 protocol;
3835
3836         if (skb->ip_summed != CHECKSUM_PARTIAL)
3837                 return 0;
3838
3839         if (skb->protocol == cpu_to_be16(ETH_P_8021Q))
3840                 protocol = vlan_eth_hdr(skb)->h_vlan_encapsulated_proto;
3841         else
3842                 protocol = skb->protocol;
3843
3844         switch (protocol) {
3845         case cpu_to_be16(ETH_P_IP):
3846                 if (ip_hdr(skb)->protocol == IPPROTO_TCP)
3847                         cmd_len |= E1000_TXD_CMD_TCP;
3848                 break;
3849         case cpu_to_be16(ETH_P_IPV6):
3850                 /* XXX not handling all IPV6 headers */
3851                 if (ipv6_hdr(skb)->nexthdr == IPPROTO_TCP)
3852                         cmd_len |= E1000_TXD_CMD_TCP;
3853                 break;
3854         default:
3855                 if (unlikely(net_ratelimit()))
3856                         e_warn("checksum_partial proto=%x!\n",
3857                                be16_to_cpu(protocol));
3858                 break;
3859         }
3860
3861         css = skb_transport_offset(skb);
3862
3863         i = tx_ring->next_to_use;
3864         buffer_info = &tx_ring->buffer_info[i];
3865         context_desc = E1000_CONTEXT_DESC(*tx_ring, i);
3866
3867         context_desc->lower_setup.ip_config = 0;
3868         context_desc->upper_setup.tcp_fields.tucss = css;
3869         context_desc->upper_setup.tcp_fields.tucso =
3870                                 css + skb->csum_offset;
3871         context_desc->upper_setup.tcp_fields.tucse = 0;
3872         context_desc->tcp_seg_setup.data = 0;
3873         context_desc->cmd_and_length = cpu_to_le32(cmd_len);
3874
3875         buffer_info->time_stamp = jiffies;
3876         buffer_info->next_to_watch = i;
3877
3878         i++;
3879         if (i == tx_ring->count)
3880                 i = 0;
3881         tx_ring->next_to_use = i;
3882
3883         return 1;
3884 }
3885
3886 #define E1000_MAX_PER_TXD       8192
3887 #define E1000_MAX_TXD_PWR       12
3888
3889 static int e1000_tx_map(struct e1000_adapter *adapter,
3890                         struct sk_buff *skb, unsigned int first,
3891                         unsigned int max_per_txd, unsigned int nr_frags,
3892                         unsigned int mss)
3893 {
3894         struct e1000_ring *tx_ring = adapter->tx_ring;
3895         struct e1000_buffer *buffer_info;
3896         unsigned int len = skb_headlen(skb);
3897         unsigned int offset, size, count = 0, i;
3898         unsigned int f;
3899         dma_addr_t *map;
3900
3901         i = tx_ring->next_to_use;
3902
3903         if (skb_dma_map(&adapter->pdev->dev, skb, DMA_TO_DEVICE)) {
3904                 dev_err(&adapter->pdev->dev, "TX DMA map failed\n");
3905                 adapter->tx_dma_failed++;
3906                 return 0;
3907         }
3908
3909         map = skb_shinfo(skb)->dma_maps;
3910         offset = 0;
3911
3912         while (len) {
3913                 buffer_info = &tx_ring->buffer_info[i];
3914                 size = min(len, max_per_txd);
3915
3916                 buffer_info->length = size;
3917                 buffer_info->time_stamp = jiffies;
3918                 buffer_info->next_to_watch = i;
3919                 buffer_info->dma = skb_shinfo(skb)->dma_head + offset;
3920                 count++;
3921
3922                 len -= size;
3923                 offset += size;
3924
3925                 if (len) {
3926                         i++;
3927                         if (i == tx_ring->count)
3928                                 i = 0;
3929                 }
3930         }
3931
3932         for (f = 0; f < nr_frags; f++) {
3933                 struct skb_frag_struct *frag;
3934
3935                 frag = &skb_shinfo(skb)->frags[f];
3936                 len = frag->size;
3937                 offset = 0;
3938
3939                 while (len) {
3940                         i++;
3941                         if (i == tx_ring->count)
3942                                 i = 0;
3943
3944                         buffer_info = &tx_ring->buffer_info[i];
3945                         size = min(len, max_per_txd);
3946
3947                         buffer_info->length = size;
3948                         buffer_info->time_stamp = jiffies;
3949                         buffer_info->next_to_watch = i;
3950                         buffer_info->dma = map[f] + offset;
3951
3952                         len -= size;
3953                         offset += size;
3954                         count++;
3955                 }
3956         }
3957
3958         tx_ring->buffer_info[i].skb = skb;
3959         tx_ring->buffer_info[first].next_to_watch = i;
3960
3961         return count;
3962 }
3963
3964 static void e1000_tx_queue(struct e1000_adapter *adapter,
3965                            int tx_flags, int count)
3966 {
3967         struct e1000_ring *tx_ring = adapter->tx_ring;
3968         struct e1000_tx_desc *tx_desc = NULL;
3969         struct e1000_buffer *buffer_info;
3970         u32 txd_upper = 0, txd_lower = E1000_TXD_CMD_IFCS;
3971         unsigned int i;
3972
3973         if (tx_flags & E1000_TX_FLAGS_TSO) {
3974                 txd_lower |= E1000_TXD_CMD_DEXT | E1000_TXD_DTYP_D |
3975                              E1000_TXD_CMD_TSE;
3976                 txd_upper |= E1000_TXD_POPTS_TXSM << 8;
3977
3978                 if (tx_flags & E1000_TX_FLAGS_IPV4)
3979                         txd_upper |= E1000_TXD_POPTS_IXSM << 8;
3980         }
3981
3982         if (tx_flags & E1000_TX_FLAGS_CSUM) {
3983                 txd_lower |= E1000_TXD_CMD_DEXT | E1000_TXD_DTYP_D;
3984                 txd_upper |= E1000_TXD_POPTS_TXSM << 8;
3985         }
3986
3987         if (tx_flags & E1000_TX_FLAGS_VLAN) {
3988                 txd_lower |= E1000_TXD_CMD_VLE;
3989                 txd_upper |= (tx_flags & E1000_TX_FLAGS_VLAN_MASK);
3990         }
3991
3992         i = tx_ring->next_to_use;
3993
3994         while (count--) {
3995                 buffer_info = &tx_ring->buffer_info[i];
3996                 tx_desc = E1000_TX_DESC(*tx_ring, i);
3997                 tx_desc->buffer_addr = cpu_to_le64(buffer_info->dma);
3998                 tx_desc->lower.data =
3999                         cpu_to_le32(txd_lower | buffer_info->length);
4000                 tx_desc->upper.data = cpu_to_le32(txd_upper);
4001
4002                 i++;
4003                 if (i == tx_ring->count)
4004                         i = 0;
4005         }
4006
4007         tx_desc->lower.data |= cpu_to_le32(adapter->txd_cmd);
4008
4009         /*
4010          * Force memory writes to complete before letting h/w
4011          * know there are new descriptors to fetch.  (Only
4012          * applicable for weak-ordered memory model archs,
4013          * such as IA-64).
4014          */
4015         wmb();
4016
4017         tx_ring->next_to_use = i;
4018         writel(i, adapter->hw.hw_addr + tx_ring->tail);
4019         /*
4020          * we need this if more than one processor can write to our tail
4021          * at a time, it synchronizes IO on IA64/Altix systems
4022          */
4023         mmiowb();
4024 }
4025
4026 #define MINIMUM_DHCP_PACKET_SIZE 282
4027 static int e1000_transfer_dhcp_info(struct e1000_adapter *adapter,
4028                                     struct sk_buff *skb)
4029 {
4030         struct e1000_hw *hw =  &adapter->hw;
4031         u16 length, offset;
4032
4033         if (vlan_tx_tag_present(skb)) {
4034                 if (!((vlan_tx_tag_get(skb) == adapter->hw.mng_cookie.vlan_id)
4035                     && (adapter->hw.mng_cookie.status &
4036                         E1000_MNG_DHCP_COOKIE_STATUS_VLAN)))
4037                         return 0;
4038         }
4039
4040         if (skb->len <= MINIMUM_DHCP_PACKET_SIZE)
4041                 return 0;
4042
4043         if (((struct ethhdr *) skb->data)->h_proto != htons(ETH_P_IP))
4044                 return 0;
4045
4046         {
4047                 const struct iphdr *ip = (struct iphdr *)((u8 *)skb->data+14);
4048                 struct udphdr *udp;
4049
4050                 if (ip->protocol != IPPROTO_UDP)
4051                         return 0;
4052
4053                 udp = (struct udphdr *)((u8 *)ip + (ip->ihl << 2));
4054                 if (ntohs(udp->dest) != 67)
4055                         return 0;
4056
4057                 offset = (u8 *)udp + 8 - skb->data;
4058                 length = skb->len - offset;
4059                 return e1000e_mng_write_dhcp_info(hw, (u8 *)udp + 8, length);
4060         }
4061
4062         return 0;
4063 }
4064
4065 static int __e1000_maybe_stop_tx(struct net_device *netdev, int size)
4066 {
4067         struct e1000_adapter *adapter = netdev_priv(netdev);
4068
4069         netif_stop_queue(netdev);
4070         /*
4071          * Herbert's original patch had:
4072          *  smp_mb__after_netif_stop_queue();
4073          * but since that doesn't exist yet, just open code it.
4074          */
4075         smp_mb();
4076
4077         /*
4078          * We need to check again in a case another CPU has just
4079          * made room available.
4080          */
4081         if (e1000_desc_unused(adapter->tx_ring) < size)
4082                 return -EBUSY;
4083
4084         /* A reprieve! */
4085         netif_start_queue(netdev);
4086         ++adapter->restart_queue;
4087         return 0;
4088 }
4089
4090 static int e1000_maybe_stop_tx(struct net_device *netdev, int size)
4091 {
4092         struct e1000_adapter *adapter = netdev_priv(netdev);
4093
4094         if (e1000_desc_unused(adapter->tx_ring) >= size)
4095                 return 0;
4096         return __e1000_maybe_stop_tx(netdev, size);
4097 }
4098
4099 #define TXD_USE_COUNT(S, X) (((S) >> (X)) + 1 )
4100 static int e1000_xmit_frame(struct sk_buff *skb, struct net_device *netdev)
4101 {
4102         struct e1000_adapter *adapter = netdev_priv(netdev);
4103         struct e1000_ring *tx_ring = adapter->tx_ring;
4104         unsigned int first;
4105         unsigned int max_per_txd = E1000_MAX_PER_TXD;
4106         unsigned int max_txd_pwr = E1000_MAX_TXD_PWR;
4107         unsigned int tx_flags = 0;
4108         unsigned int len = skb->len - skb->data_len;
4109         unsigned int nr_frags;
4110         unsigned int mss;
4111         int count = 0;
4112         int tso;
4113         unsigned int f;
4114
4115         if (test_bit(__E1000_DOWN, &adapter->state)) {
4116                 dev_kfree_skb_any(skb);
4117                 return NETDEV_TX_OK;
4118         }
4119
4120         if (skb->len <= 0) {
4121                 dev_kfree_skb_any(skb);
4122                 return NETDEV_TX_OK;
4123         }
4124
4125         mss = skb_shinfo(skb)->gso_size;
4126         /*
4127          * The controller does a simple calculation to
4128          * make sure there is enough room in the FIFO before
4129          * initiating the DMA for each buffer.  The calc is:
4130          * 4 = ceil(buffer len/mss).  To make sure we don't
4131          * overrun the FIFO, adjust the max buffer len if mss
4132          * drops.
4133          */
4134         if (mss) {
4135                 u8 hdr_len;
4136                 max_per_txd = min(mss << 2, max_per_txd);
4137                 max_txd_pwr = fls(max_per_txd) - 1;
4138
4139                 /*
4140                  * TSO Workaround for 82571/2/3 Controllers -- if skb->data
4141                  * points to just header, pull a few bytes of payload from
4142                  * frags into skb->data
4143                  */
4144                 hdr_len = skb_transport_offset(skb) + tcp_hdrlen(skb);
4145                 /*
4146                  * we do this workaround for ES2LAN, but it is un-necessary,
4147                  * avoiding it could save a lot of cycles
4148                  */
4149                 if (skb->data_len && (hdr_len == len)) {
4150                         unsigned int pull_size;
4151
4152                         pull_size = min((unsigned int)4, skb->data_len);
4153                         if (!__pskb_pull_tail(skb, pull_size)) {
4154                                 e_err("__pskb_pull_tail failed.\n");
4155                                 dev_kfree_skb_any(skb);
4156                                 return NETDEV_TX_OK;
4157                         }
4158                         len = skb->len - skb->data_len;
4159                 }
4160         }
4161
4162         /* reserve a descriptor for the offload context */
4163         if ((mss) || (skb->ip_summed == CHECKSUM_PARTIAL))
4164                 count++;
4165         count++;
4166
4167         count += TXD_USE_COUNT(len, max_txd_pwr);
4168
4169         nr_frags = skb_shinfo(skb)->nr_frags;
4170         for (f = 0; f < nr_frags; f++)
4171                 count += TXD_USE_COUNT(skb_shinfo(skb)->frags[f].size,
4172                                        max_txd_pwr);
4173
4174         if (adapter->hw.mac.tx_pkt_filtering)
4175                 e1000_transfer_dhcp_info(adapter, skb);
4176
4177         /*
4178          * need: count + 2 desc gap to keep tail from touching
4179          * head, otherwise try next time
4180          */
4181         if (e1000_maybe_stop_tx(netdev, count + 2))
4182                 return NETDEV_TX_BUSY;
4183
4184         if (adapter->vlgrp && vlan_tx_tag_present(skb)) {
4185                 tx_flags |= E1000_TX_FLAGS_VLAN;
4186                 tx_flags |= (vlan_tx_tag_get(skb) << E1000_TX_FLAGS_VLAN_SHIFT);
4187         }
4188
4189         first = tx_ring->next_to_use;
4190
4191         tso = e1000_tso(adapter, skb);
4192         if (tso < 0) {
4193                 dev_kfree_skb_any(skb);
4194                 return NETDEV_TX_OK;
4195         }
4196
4197         if (tso)
4198                 tx_flags |= E1000_TX_FLAGS_TSO;
4199         else if (e1000_tx_csum(adapter, skb))
4200                 tx_flags |= E1000_TX_FLAGS_CSUM;
4201
4202         /*
4203          * Old method was to assume IPv4 packet by default if TSO was enabled.
4204          * 82571 hardware supports TSO capabilities for IPv6 as well...
4205          * no longer assume, we must.
4206          */
4207         if (skb->protocol == htons(ETH_P_IP))
4208                 tx_flags |= E1000_TX_FLAGS_IPV4;
4209
4210         /* if count is 0 then mapping error has occured */
4211         count = e1000_tx_map(adapter, skb, first, max_per_txd, nr_frags, mss);
4212         if (count) {
4213                 e1000_tx_queue(adapter, tx_flags, count);
4214                 /* Make sure there is space in the ring for the next send. */
4215                 e1000_maybe_stop_tx(netdev, MAX_SKB_FRAGS + 2);
4216
4217         } else {
4218                 dev_kfree_skb_any(skb);
4219                 tx_ring->buffer_info[first].time_stamp = 0;
4220                 tx_ring->next_to_use = first;
4221         }
4222
4223         return NETDEV_TX_OK;
4224 }
4225
4226 /**
4227  * e1000_tx_timeout - Respond to a Tx Hang
4228  * @netdev: network interface device structure
4229  **/
4230 static void e1000_tx_timeout(struct net_device *netdev)
4231 {
4232         struct e1000_adapter *adapter = netdev_priv(netdev);
4233
4234         /* Do the reset outside of interrupt context */
4235         adapter->tx_timeout_count++;
4236         schedule_work(&adapter->reset_task);
4237 }
4238
4239 static void e1000_reset_task(struct work_struct *work)
4240 {
4241         struct e1000_adapter *adapter;
4242         adapter = container_of(work, struct e1000_adapter, reset_task);
4243
4244         e1000e_reinit_locked(adapter);
4245 }
4246
4247 /**
4248  * e1000_get_stats - Get System Network Statistics
4249  * @netdev: network interface device structure
4250  *
4251  * Returns the address of the device statistics structure.
4252  * The statistics are actually updated from the timer callback.
4253  **/
4254 static struct net_device_stats *e1000_get_stats(struct net_device *netdev)
4255 {
4256         struct e1000_adapter *adapter = netdev_priv(netdev);
4257
4258         /* only return the current stats */
4259         return &adapter->net_stats;
4260 }
4261
4262 /**
4263  * e1000_change_mtu - Change the Maximum Transfer Unit
4264  * @netdev: network interface device structure
4265  * @new_mtu: new value for maximum frame size
4266  *
4267  * Returns 0 on success, negative on failure
4268  **/
4269 static int e1000_change_mtu(struct net_device *netdev, int new_mtu)
4270 {
4271         struct e1000_adapter *adapter = netdev_priv(netdev);
4272         int max_frame = new_mtu + ETH_HLEN + ETH_FCS_LEN;
4273
4274         /* Jumbo frame support */
4275         if ((max_frame > ETH_FRAME_LEN + ETH_FCS_LEN) &&
4276             !(adapter->flags & FLAG_HAS_JUMBO_FRAMES)) {
4277                 e_err("Jumbo Frames not supported.\n");
4278                 return -EINVAL;
4279         }
4280
4281         /* Supported frame sizes */
4282         if ((new_mtu < ETH_ZLEN + ETH_FCS_LEN + VLAN_HLEN) ||
4283             (max_frame > adapter->max_hw_frame_size)) {
4284                 e_err("Unsupported MTU setting\n");
4285                 return -EINVAL;
4286         }
4287
4288         while (test_and_set_bit(__E1000_RESETTING, &adapter->state))
4289                 msleep(1);
4290         /* e1000e_down has a dependency on max_frame_size */
4291         adapter->max_frame_size = max_frame;
4292         if (netif_running(netdev))
4293                 e1000e_down(adapter);
4294
4295         /*
4296          * NOTE: netdev_alloc_skb reserves 16 bytes, and typically NET_IP_ALIGN
4297          * means we reserve 2 more, this pushes us to allocate from the next
4298          * larger slab size.
4299          * i.e. RXBUFFER_2048 --> size-4096 slab
4300          * However with the new *_jumbo_rx* routines, jumbo receives will use
4301          * fragmented skbs
4302          */
4303
4304         if (max_frame <= 256)
4305                 adapter->rx_buffer_len = 256;
4306         else if (max_frame <= 512)
4307                 adapter->rx_buffer_len = 512;
4308         else if (max_frame <= 1024)
4309                 adapter->rx_buffer_len = 1024;
4310         else if (max_frame <= 2048)
4311                 adapter->rx_buffer_len = 2048;
4312         else
4313                 adapter->rx_buffer_len = 4096;
4314
4315         /* adjust allocation if LPE protects us, and we aren't using SBP */
4316         if ((max_frame == ETH_FRAME_LEN + ETH_FCS_LEN) ||
4317              (max_frame == ETH_FRAME_LEN + VLAN_HLEN + ETH_FCS_LEN))
4318                 adapter->rx_buffer_len = ETH_FRAME_LEN + VLAN_HLEN
4319                                          + ETH_FCS_LEN;
4320
4321         e_info("changing MTU from %d to %d\n", netdev->mtu, new_mtu);
4322         netdev->mtu = new_mtu;
4323
4324         if (netif_running(netdev))
4325                 e1000e_up(adapter);
4326         else
4327                 e1000e_reset(adapter);
4328
4329         clear_bit(__E1000_RESETTING, &adapter->state);
4330
4331         return 0;
4332 }
4333
4334 static int e1000_mii_ioctl(struct net_device *netdev, struct ifreq *ifr,
4335                            int cmd)
4336 {
4337         struct e1000_adapter *adapter = netdev_priv(netdev);
4338         struct mii_ioctl_data *data = if_mii(ifr);
4339
4340         if (adapter->hw.phy.media_type != e1000_media_type_copper)
4341                 return -EOPNOTSUPP;
4342
4343         switch (cmd) {
4344         case SIOCGMIIPHY:
4345                 data->phy_id = adapter->hw.phy.addr;
4346                 break;
4347         case SIOCGMIIREG:
4348                 if (!capable(CAP_NET_ADMIN))
4349                         return -EPERM;
4350                 switch (data->reg_num & 0x1F) {
4351                 case MII_BMCR:
4352                         data->val_out = adapter->phy_regs.bmcr;
4353                         break;
4354                 case MII_BMSR:
4355                         data->val_out = adapter->phy_regs.bmsr;
4356                         break;
4357                 case MII_PHYSID1:
4358                         data->val_out = (adapter->hw.phy.id >> 16);
4359                         break;
4360                 case MII_PHYSID2:
4361                         data->val_out = (adapter->hw.phy.id & 0xFFFF);
4362                         break;
4363                 case MII_ADVERTISE:
4364                         data->val_out = adapter->phy_regs.advertise;
4365                         break;
4366                 case MII_LPA:
4367                         data->val_out = adapter->phy_regs.lpa;
4368                         break;
4369                 case MII_EXPANSION:
4370                         data->val_out = adapter->phy_regs.expansion;
4371                         break;
4372                 case MII_CTRL1000:
4373                         data->val_out = adapter->phy_regs.ctrl1000;
4374                         break;
4375                 case MII_STAT1000:
4376                         data->val_out = adapter->phy_regs.stat1000;
4377                         break;
4378                 case MII_ESTATUS:
4379                         data->val_out = adapter->phy_regs.estatus;
4380                         break;
4381                 default:
4382                         return -EIO;
4383                 }
4384                 break;
4385         case SIOCSMIIREG:
4386         default:
4387                 return -EOPNOTSUPP;
4388         }
4389         return 0;
4390 }
4391
4392 static int e1000_ioctl(struct net_device *netdev, struct ifreq *ifr, int cmd)
4393 {
4394         switch (cmd) {
4395         case SIOCGMIIPHY:
4396         case SIOCGMIIREG:
4397         case SIOCSMIIREG:
4398                 return e1000_mii_ioctl(netdev, ifr, cmd);
4399         default:
4400                 return -EOPNOTSUPP;
4401         }
4402 }
4403
4404 static int e1000_init_phy_wakeup(struct e1000_adapter *adapter, u32 wufc)
4405 {
4406         struct e1000_hw *hw = &adapter->hw;
4407         u32 i, mac_reg;
4408         u16 phy_reg;
4409         int retval = 0;
4410
4411         /* copy MAC RARs to PHY RARs */
4412         for (i = 0; i < adapter->hw.mac.rar_entry_count; i++) {
4413                 mac_reg = er32(RAL(i));
4414                 e1e_wphy(hw, BM_RAR_L(i), (u16)(mac_reg & 0xFFFF));
4415                 e1e_wphy(hw, BM_RAR_M(i), (u16)((mac_reg >> 16) & 0xFFFF));
4416                 mac_reg = er32(RAH(i));
4417                 e1e_wphy(hw, BM_RAR_H(i), (u16)(mac_reg & 0xFFFF));
4418                 e1e_wphy(hw, BM_RAR_CTRL(i), (u16)((mac_reg >> 16) & 0xFFFF));
4419         }
4420
4421         /* copy MAC MTA to PHY MTA */
4422         for (i = 0; i < adapter->hw.mac.mta_reg_count; i++) {
4423                 mac_reg = E1000_READ_REG_ARRAY(hw, E1000_MTA, i);
4424                 e1e_wphy(hw, BM_MTA(i), (u16)(mac_reg & 0xFFFF));
4425                 e1e_wphy(hw, BM_MTA(i) + 1, (u16)((mac_reg >> 16) & 0xFFFF));
4426         }
4427
4428         /* configure PHY Rx Control register */
4429         e1e_rphy(&adapter->hw, BM_RCTL, &phy_reg);
4430         mac_reg = er32(RCTL);
4431         if (mac_reg & E1000_RCTL_UPE)
4432                 phy_reg |= BM_RCTL_UPE;
4433         if (mac_reg & E1000_RCTL_MPE)
4434                 phy_reg |= BM_RCTL_MPE;
4435         phy_reg &= ~(BM_RCTL_MO_MASK);
4436         if (mac_reg & E1000_RCTL_MO_3)
4437                 phy_reg |= (((mac_reg & E1000_RCTL_MO_3) >> E1000_RCTL_MO_SHIFT)
4438                                 << BM_RCTL_MO_SHIFT);
4439         if (mac_reg & E1000_RCTL_BAM)
4440                 phy_reg |= BM_RCTL_BAM;
4441         if (mac_reg & E1000_RCTL_PMCF)
4442                 phy_reg |= BM_RCTL_PMCF;
4443         mac_reg = er32(CTRL);
4444         if (mac_reg & E1000_CTRL_RFCE)
4445                 phy_reg |= BM_RCTL_RFCE;
4446         e1e_wphy(&adapter->hw, BM_RCTL, phy_reg);
4447
4448         /* enable PHY wakeup in MAC register */
4449         ew32(WUFC, wufc);
4450         ew32(WUC, E1000_WUC_PHY_WAKE | E1000_WUC_PME_EN);
4451
4452         /* configure and enable PHY wakeup in PHY registers */
4453         e1e_wphy(&adapter->hw, BM_WUFC, wufc);
4454         e1e_wphy(&adapter->hw, BM_WUC, E1000_WUC_PME_EN);
4455
4456         /* activate PHY wakeup */
4457         retval = hw->phy.ops.acquire_phy(hw);
4458         if (retval) {
4459                 e_err("Could not acquire PHY\n");
4460                 return retval;
4461         }
4462         e1000e_write_phy_reg_mdic(hw, IGP01E1000_PHY_PAGE_SELECT,
4463                                  (BM_WUC_ENABLE_PAGE << IGP_PAGE_SHIFT));
4464         retval = e1000e_read_phy_reg_mdic(hw, BM_WUC_ENABLE_REG, &phy_reg);
4465         if (retval) {
4466                 e_err("Could not read PHY page 769\n");
4467                 goto out;
4468         }
4469         phy_reg |= BM_WUC_ENABLE_BIT | BM_WUC_HOST_WU_BIT;
4470         retval = e1000e_write_phy_reg_mdic(hw, BM_WUC_ENABLE_REG, phy_reg);
4471         if (retval)
4472                 e_err("Could not set PHY Host Wakeup bit\n");
4473 out:
4474         hw->phy.ops.release_phy(hw);
4475
4476         return retval;
4477 }
4478
4479 static int __e1000_shutdown(struct pci_dev *pdev, bool *enable_wake)
4480 {
4481         struct net_device *netdev = pci_get_drvdata(pdev);
4482         struct e1000_adapter *adapter = netdev_priv(netdev);
4483         struct e1000_hw *hw = &adapter->hw;
4484         u32 ctrl, ctrl_ext, rctl, status;
4485         u32 wufc = adapter->wol;
4486         int retval = 0;
4487
4488         netif_device_detach(netdev);
4489
4490         if (netif_running(netdev)) {
4491                 WARN_ON(test_bit(__E1000_RESETTING, &adapter->state));
4492                 e1000e_down(adapter);
4493                 e1000_free_irq(adapter);
4494         }
4495         e1000e_reset_interrupt_capability(adapter);
4496
4497         retval = pci_save_state(pdev);
4498         if (retval)
4499                 return retval;
4500
4501         status = er32(STATUS);
4502         if (status & E1000_STATUS_LU)
4503                 wufc &= ~E1000_WUFC_LNKC;
4504
4505         if (wufc) {
4506                 e1000_setup_rctl(adapter);
4507                 e1000_set_multi(netdev);
4508
4509                 /* turn on all-multi mode if wake on multicast is enabled */
4510                 if (wufc & E1000_WUFC_MC) {
4511                         rctl = er32(RCTL);
4512                         rctl |= E1000_RCTL_MPE;
4513                         ew32(RCTL, rctl);
4514                 }
4515
4516                 ctrl = er32(CTRL);
4517                 /* advertise wake from D3Cold */
4518                 #define E1000_CTRL_ADVD3WUC 0x00100000
4519                 /* phy power management enable */
4520                 #define E1000_CTRL_EN_PHY_PWR_MGMT 0x00200000
4521                 ctrl |= E1000_CTRL_ADVD3WUC;
4522                 if (!(adapter->flags2 & FLAG2_HAS_PHY_WAKEUP))
4523                         ctrl |= E1000_CTRL_EN_PHY_PWR_MGMT;
4524                 ew32(CTRL, ctrl);
4525
4526                 if (adapter->hw.phy.media_type == e1000_media_type_fiber ||
4527                     adapter->hw.phy.media_type ==
4528                     e1000_media_type_internal_serdes) {
4529                         /* keep the laser running in D3 */
4530                         ctrl_ext = er32(CTRL_EXT);
4531                         ctrl_ext |= E1000_CTRL_EXT_SDP7_DATA;
4532                         ew32(CTRL_EXT, ctrl_ext);
4533                 }
4534
4535                 if (adapter->flags & FLAG_IS_ICH)
4536                         e1000e_disable_gig_wol_ich8lan(&adapter->hw);
4537
4538                 /* Allow time for pending master requests to run */
4539                 e1000e_disable_pcie_master(&adapter->hw);
4540
4541                 if (adapter->flags2 & FLAG2_HAS_PHY_WAKEUP) {
4542                         /* enable wakeup by the PHY */
4543                         retval = e1000_init_phy_wakeup(adapter, wufc);
4544                         if (retval)
4545                                 return retval;
4546                 } else {
4547                         /* enable wakeup by the MAC */
4548                         ew32(WUFC, wufc);
4549                         ew32(WUC, E1000_WUC_PME_EN);
4550                 }
4551         } else {
4552                 ew32(WUC, 0);
4553                 ew32(WUFC, 0);
4554         }
4555
4556         *enable_wake = !!wufc;
4557
4558         /* make sure adapter isn't asleep if manageability is enabled */
4559         if ((adapter->flags & FLAG_MNG_PT_ENABLED) ||
4560             (hw->mac.ops.check_mng_mode(hw)))
4561                 *enable_wake = true;
4562
4563         if (adapter->hw.phy.type == e1000_phy_igp_3)
4564                 e1000e_igp3_phy_powerdown_workaround_ich8lan(&adapter->hw);
4565
4566         /*
4567          * Release control of h/w to f/w.  If f/w is AMT enabled, this
4568          * would have already happened in close and is redundant.
4569          */
4570         e1000_release_hw_control(adapter);
4571
4572         pci_disable_device(pdev);
4573
4574         return 0;
4575 }
4576
4577 static void e1000_power_off(struct pci_dev *pdev, bool sleep, bool wake)
4578 {
4579         if (sleep && wake) {
4580                 pci_prepare_to_sleep(pdev);
4581                 return;
4582         }
4583
4584         pci_wake_from_d3(pdev, wake);
4585         pci_set_power_state(pdev, PCI_D3hot);
4586 }
4587
4588 static void e1000_complete_shutdown(struct pci_dev *pdev, bool sleep,
4589                                     bool wake)
4590 {
4591         struct net_device *netdev = pci_get_drvdata(pdev);
4592         struct e1000_adapter *adapter = netdev_priv(netdev);
4593
4594         /*
4595          * The pci-e switch on some quad port adapters will report a
4596          * correctable error when the MAC transitions from D0 to D3.  To
4597          * prevent this we need to mask off the correctable errors on the
4598          * downstream port of the pci-e switch.
4599          */
4600         if (adapter->flags & FLAG_IS_QUAD_PORT) {
4601                 struct pci_dev *us_dev = pdev->bus->self;
4602                 int pos = pci_find_capability(us_dev, PCI_CAP_ID_EXP);
4603                 u16 devctl;
4604
4605                 pci_read_config_word(us_dev, pos + PCI_EXP_DEVCTL, &devctl);
4606                 pci_write_config_word(us_dev, pos + PCI_EXP_DEVCTL,
4607                                       (devctl & ~PCI_EXP_DEVCTL_CERE));
4608
4609                 e1000_power_off(pdev, sleep, wake);
4610
4611                 pci_write_config_word(us_dev, pos + PCI_EXP_DEVCTL, devctl);
4612         } else {
4613                 e1000_power_off(pdev, sleep, wake);
4614         }
4615 }
4616
4617 static void e1000e_disable_l1aspm(struct pci_dev *pdev)
4618 {
4619         int pos;
4620         u16 val;
4621
4622         /*
4623          * 82573 workaround - disable L1 ASPM on mobile chipsets
4624          *
4625          * L1 ASPM on various mobile (ich7) chipsets do not behave properly
4626          * resulting in lost data or garbage information on the pci-e link
4627          * level. This could result in (false) bad EEPROM checksum errors,
4628          * long ping times (up to 2s) or even a system freeze/hang.
4629          *
4630          * Unfortunately this feature saves about 1W power consumption when
4631          * active.
4632          */
4633         pos = pci_find_capability(pdev, PCI_CAP_ID_EXP);
4634         pci_read_config_word(pdev, pos + PCI_EXP_LNKCTL, &val);
4635         if (val & 0x2) {
4636                 dev_warn(&pdev->dev, "Disabling L1 ASPM\n");
4637                 val &= ~0x2;
4638                 pci_write_config_word(pdev, pos + PCI_EXP_LNKCTL, val);
4639         }
4640 }
4641
4642 #ifdef CONFIG_PM
4643 static int e1000_suspend(struct pci_dev *pdev, pm_message_t state)
4644 {
4645         int retval;
4646         bool wake;
4647
4648         retval = __e1000_shutdown(pdev, &wake);
4649         if (!retval)
4650                 e1000_complete_shutdown(pdev, true, wake);
4651
4652         return retval;
4653 }
4654
4655 static int e1000_resume(struct pci_dev *pdev)
4656 {
4657         struct net_device *netdev = pci_get_drvdata(pdev);
4658         struct e1000_adapter *adapter = netdev_priv(netdev);
4659         struct e1000_hw *hw = &adapter->hw;
4660         u32 err;
4661
4662         pci_set_power_state(pdev, PCI_D0);
4663         pci_restore_state(pdev);
4664         e1000e_disable_l1aspm(pdev);
4665
4666         err = pci_enable_device_mem(pdev);
4667         if (err) {
4668                 dev_err(&pdev->dev,
4669                         "Cannot enable PCI device from suspend\n");
4670                 return err;
4671         }
4672
4673         pci_set_master(pdev);
4674
4675         pci_enable_wake(pdev, PCI_D3hot, 0);
4676         pci_enable_wake(pdev, PCI_D3cold, 0);
4677
4678         e1000e_set_interrupt_capability(adapter);
4679         if (netif_running(netdev)) {
4680                 err = e1000_request_irq(adapter);
4681                 if (err)
4682                         return err;
4683         }
4684
4685         e1000e_power_up_phy(adapter);
4686
4687         /* report the system wakeup cause from S3/S4 */
4688         if (adapter->flags2 & FLAG2_HAS_PHY_WAKEUP) {
4689                 u16 phy_data;
4690
4691                 e1e_rphy(&adapter->hw, BM_WUS, &phy_data);
4692                 if (phy_data) {
4693                         e_info("PHY Wakeup cause - %s\n",
4694                                 phy_data & E1000_WUS_EX ? "Unicast Packet" :
4695                                 phy_data & E1000_WUS_MC ? "Multicast Packet" :
4696                                 phy_data & E1000_WUS_BC ? "Broadcast Packet" :
4697                                 phy_data & E1000_WUS_MAG ? "Magic Packet" :
4698                                 phy_data & E1000_WUS_LNKC ? "Link Status "
4699                                 " Change" : "other");
4700                 }
4701                 e1e_wphy(&adapter->hw, BM_WUS, ~0);
4702         } else {
4703                 u32 wus = er32(WUS);
4704                 if (wus) {
4705                         e_info("MAC Wakeup cause - %s\n",
4706                                 wus & E1000_WUS_EX ? "Unicast Packet" :
4707                                 wus & E1000_WUS_MC ? "Multicast Packet" :
4708                                 wus & E1000_WUS_BC ? "Broadcast Packet" :
4709                                 wus & E1000_WUS_MAG ? "Magic Packet" :
4710                                 wus & E1000_WUS_LNKC ? "Link Status Change" :
4711                                 "other");
4712                 }
4713                 ew32(WUS, ~0);
4714         }
4715
4716         e1000e_reset(adapter);
4717
4718         e1000_init_manageability(adapter);
4719
4720         if (netif_running(netdev))
4721                 e1000e_up(adapter);
4722
4723         netif_device_attach(netdev);
4724
4725         /*
4726          * If the controller has AMT, do not set DRV_LOAD until the interface
4727          * is up.  For all other cases, let the f/w know that the h/w is now
4728          * under the control of the driver.
4729          */
4730         if (!(adapter->flags & FLAG_HAS_AMT))
4731                 e1000_get_hw_control(adapter);
4732
4733         return 0;
4734 }
4735 #endif
4736
4737 static void e1000_shutdown(struct pci_dev *pdev)
4738 {
4739         bool wake = false;
4740
4741         __e1000_shutdown(pdev, &wake);
4742
4743         if (system_state == SYSTEM_POWER_OFF)
4744                 e1000_complete_shutdown(pdev, false, wake);
4745 }
4746
4747 #ifdef CONFIG_NET_POLL_CONTROLLER
4748 /*
4749  * Polling 'interrupt' - used by things like netconsole to send skbs
4750  * without having to re-enable interrupts. It's not called while
4751  * the interrupt routine is executing.
4752  */
4753 static void e1000_netpoll(struct net_device *netdev)
4754 {
4755         struct e1000_adapter *adapter = netdev_priv(netdev);
4756
4757         disable_irq(adapter->pdev->irq);
4758         e1000_intr(adapter->pdev->irq, netdev);
4759
4760         enable_irq(adapter->pdev->irq);
4761 }
4762 #endif
4763
4764 /**
4765  * e1000_io_error_detected - called when PCI error is detected
4766  * @pdev: Pointer to PCI device
4767  * @state: The current pci connection state
4768  *
4769  * This function is called after a PCI bus error affecting
4770  * this device has been detected.
4771  */
4772 static pci_ers_result_t e1000_io_error_detected(struct pci_dev *pdev,
4773                                                 pci_channel_state_t state)
4774 {
4775         struct net_device *netdev = pci_get_drvdata(pdev);
4776         struct e1000_adapter *adapter = netdev_priv(netdev);
4777
4778         netif_device_detach(netdev);
4779
4780         if (state == pci_channel_io_perm_failure)
4781                 return PCI_ERS_RESULT_DISCONNECT;
4782
4783         if (netif_running(netdev))
4784                 e1000e_down(adapter);
4785         pci_disable_device(pdev);
4786
4787         /* Request a slot slot reset. */
4788         return PCI_ERS_RESULT_NEED_RESET;
4789 }
4790
4791 /**
4792  * e1000_io_slot_reset - called after the pci bus has been reset.
4793  * @pdev: Pointer to PCI device
4794  *
4795  * Restart the card from scratch, as if from a cold-boot. Implementation
4796  * resembles the first-half of the e1000_resume routine.
4797  */
4798 static pci_ers_result_t e1000_io_slot_reset(struct pci_dev *pdev)
4799 {
4800         struct net_device *netdev = pci_get_drvdata(pdev);
4801         struct e1000_adapter *adapter = netdev_priv(netdev);
4802         struct e1000_hw *hw = &adapter->hw;
4803         int err;
4804         pci_ers_result_t result;
4805
4806         e1000e_disable_l1aspm(pdev);
4807         err = pci_enable_device_mem(pdev);
4808         if (err) {
4809                 dev_err(&pdev->dev,
4810                         "Cannot re-enable PCI device after reset.\n");
4811                 result = PCI_ERS_RESULT_DISCONNECT;
4812         } else {
4813                 pci_set_master(pdev);
4814                 pci_restore_state(pdev);
4815
4816                 pci_enable_wake(pdev, PCI_D3hot, 0);
4817                 pci_enable_wake(pdev, PCI_D3cold, 0);
4818
4819                 e1000e_reset(adapter);
4820                 ew32(WUS, ~0);
4821                 result = PCI_ERS_RESULT_RECOVERED;
4822         }
4823
4824         pci_cleanup_aer_uncorrect_error_status(pdev);
4825
4826         return result;
4827 }
4828
4829 /**
4830  * e1000_io_resume - called when traffic can start flowing again.
4831  * @pdev: Pointer to PCI device
4832  *
4833  * This callback is called when the error recovery driver tells us that
4834  * its OK to resume normal operation. Implementation resembles the
4835  * second-half of the e1000_resume routine.
4836  */
4837 static void e1000_io_resume(struct pci_dev *pdev)
4838 {
4839         struct net_device *netdev = pci_get_drvdata(pdev);
4840         struct e1000_adapter *adapter = netdev_priv(netdev);
4841
4842         e1000_init_manageability(adapter);
4843
4844         if (netif_running(netdev)) {
4845                 if (e1000e_up(adapter)) {
4846                         dev_err(&pdev->dev,
4847                                 "can't bring device back up after reset\n");
4848                         return;
4849                 }
4850         }
4851
4852         netif_device_attach(netdev);
4853
4854         /*
4855          * If the controller has AMT, do not set DRV_LOAD until the interface
4856          * is up.  For all other cases, let the f/w know that the h/w is now
4857          * under the control of the driver.
4858          */
4859         if (!(adapter->flags & FLAG_HAS_AMT))
4860                 e1000_get_hw_control(adapter);
4861
4862 }
4863
4864 static void e1000_print_device_info(struct e1000_adapter *adapter)
4865 {
4866         struct e1000_hw *hw = &adapter->hw;
4867         struct net_device *netdev = adapter->netdev;
4868         u32 pba_num;
4869
4870         /* print bus type/speed/width info */
4871         e_info("(PCI Express:2.5GB/s:%s) %pM\n",
4872                /* bus width */
4873                ((hw->bus.width == e1000_bus_width_pcie_x4) ? "Width x4" :
4874                 "Width x1"),
4875                /* MAC address */
4876                netdev->dev_addr);
4877         e_info("Intel(R) PRO/%s Network Connection\n",
4878                (hw->phy.type == e1000_phy_ife) ? "10/100" : "1000");
4879         e1000e_read_pba_num(hw, &pba_num);
4880         e_info("MAC: %d, PHY: %d, PBA No: %06x-%03x\n",
4881                hw->mac.type, hw->phy.type, (pba_num >> 8), (pba_num & 0xff));
4882 }
4883
4884 static void e1000_eeprom_checks(struct e1000_adapter *adapter)
4885 {
4886         struct e1000_hw *hw = &adapter->hw;
4887         int ret_val;
4888         u16 buf = 0;
4889
4890         if (hw->mac.type != e1000_82573)
4891                 return;
4892
4893         ret_val = e1000_read_nvm(hw, NVM_INIT_CONTROL2_REG, 1, &buf);
4894         if (!ret_val && (!(le16_to_cpu(buf) & (1 << 0)))) {
4895                 /* Deep Smart Power Down (DSPD) */
4896                 dev_warn(&adapter->pdev->dev,
4897                          "Warning: detected DSPD enabled in EEPROM\n");
4898         }
4899
4900         ret_val = e1000_read_nvm(hw, NVM_INIT_3GIO_3, 1, &buf);
4901         if (!ret_val && (le16_to_cpu(buf) & (3 << 2))) {
4902                 /* ASPM enable */
4903                 dev_warn(&adapter->pdev->dev,
4904                          "Warning: detected ASPM enabled in EEPROM\n");
4905         }
4906 }
4907
4908 static const struct net_device_ops e1000e_netdev_ops = {
4909         .ndo_open               = e1000_open,
4910         .ndo_stop               = e1000_close,
4911         .ndo_start_xmit         = e1000_xmit_frame,
4912         .ndo_get_stats          = e1000_get_stats,
4913         .ndo_set_multicast_list = e1000_set_multi,
4914         .ndo_set_mac_address    = e1000_set_mac,
4915         .ndo_change_mtu         = e1000_change_mtu,
4916         .ndo_do_ioctl           = e1000_ioctl,
4917         .ndo_tx_timeout         = e1000_tx_timeout,
4918         .ndo_validate_addr      = eth_validate_addr,
4919
4920         .ndo_vlan_rx_register   = e1000_vlan_rx_register,
4921         .ndo_vlan_rx_add_vid    = e1000_vlan_rx_add_vid,
4922         .ndo_vlan_rx_kill_vid   = e1000_vlan_rx_kill_vid,
4923 #ifdef CONFIG_NET_POLL_CONTROLLER
4924         .ndo_poll_controller    = e1000_netpoll,
4925 #endif
4926 };
4927
4928 /**
4929  * e1000_probe - Device Initialization Routine
4930  * @pdev: PCI device information struct
4931  * @ent: entry in e1000_pci_tbl
4932  *
4933  * Returns 0 on success, negative on failure
4934  *
4935  * e1000_probe initializes an adapter identified by a pci_dev structure.
4936  * The OS initialization, configuring of the adapter private structure,
4937  * and a hardware reset occur.
4938  **/
4939 static int __devinit e1000_probe(struct pci_dev *pdev,
4940                                  const struct pci_device_id *ent)
4941 {
4942         struct net_device *netdev;
4943         struct e1000_adapter *adapter;
4944         struct e1000_hw *hw;
4945         const struct e1000_info *ei = e1000_info_tbl[ent->driver_data];
4946         resource_size_t mmio_start, mmio_len;
4947         resource_size_t flash_start, flash_len;
4948
4949         static int cards_found;
4950         int i, err, pci_using_dac;
4951         u16 eeprom_data = 0;
4952         u16 eeprom_apme_mask = E1000_EEPROM_APME;
4953
4954         e1000e_disable_l1aspm(pdev);
4955
4956         err = pci_enable_device_mem(pdev);
4957         if (err)
4958                 return err;
4959
4960         pci_using_dac = 0;
4961         err = pci_set_dma_mask(pdev, DMA_BIT_MASK(64));
4962         if (!err) {
4963                 err = pci_set_consistent_dma_mask(pdev, DMA_BIT_MASK(64));
4964                 if (!err)
4965                         pci_using_dac = 1;
4966         } else {
4967                 err = pci_set_dma_mask(pdev, DMA_BIT_MASK(32));
4968                 if (err) {
4969                         err = pci_set_consistent_dma_mask(pdev,
4970                                                           DMA_BIT_MASK(32));
4971                         if (err) {
4972                                 dev_err(&pdev->dev, "No usable DMA "
4973                                         "configuration, aborting\n");
4974                                 goto err_dma;
4975                         }
4976                 }
4977         }
4978
4979         err = pci_request_selected_regions_exclusive(pdev,
4980                                           pci_select_bars(pdev, IORESOURCE_MEM),
4981                                           e1000e_driver_name);
4982         if (err)
4983                 goto err_pci_reg;
4984
4985         /* AER (Advanced Error Reporting) hooks */
4986         err = pci_enable_pcie_error_reporting(pdev);
4987         if (err) {
4988                 dev_err(&pdev->dev, "pci_enable_pcie_error_reporting failed "
4989                         "0x%x\n", err);
4990                 /* non-fatal, continue */
4991         }
4992
4993         pci_set_master(pdev);
4994         /* PCI config space info */
4995         err = pci_save_state(pdev);
4996         if (err)
4997                 goto err_alloc_etherdev;
4998
4999         err = -ENOMEM;
5000         netdev = alloc_etherdev(sizeof(struct e1000_adapter));
5001         if (!netdev)
5002                 goto err_alloc_etherdev;
5003
5004         SET_NETDEV_DEV(netdev, &pdev->dev);
5005
5006         pci_set_drvdata(pdev, netdev);
5007         adapter = netdev_priv(netdev);
5008         hw = &adapter->hw;
5009         adapter->netdev = netdev;
5010         adapter->pdev = pdev;
5011         adapter->ei = ei;
5012         adapter->pba = ei->pba;
5013         adapter->flags = ei->flags;
5014         adapter->flags2 = ei->flags2;
5015         adapter->hw.adapter = adapter;
5016         adapter->hw.mac.type = ei->mac;
5017         adapter->max_hw_frame_size = ei->max_hw_frame_size;
5018         adapter->msg_enable = (1 << NETIF_MSG_DRV | NETIF_MSG_PROBE) - 1;
5019
5020         mmio_start = pci_resource_start(pdev, 0);
5021         mmio_len = pci_resource_len(pdev, 0);
5022
5023         err = -EIO;
5024         adapter->hw.hw_addr = ioremap(mmio_start, mmio_len);
5025         if (!adapter->hw.hw_addr)
5026                 goto err_ioremap;
5027
5028         if ((adapter->flags & FLAG_HAS_FLASH) &&
5029             (pci_resource_flags(pdev, 1) & IORESOURCE_MEM)) {
5030                 flash_start = pci_resource_start(pdev, 1);
5031                 flash_len = pci_resource_len(pdev, 1);
5032                 adapter->hw.flash_address = ioremap(flash_start, flash_len);
5033                 if (!adapter->hw.flash_address)
5034                         goto err_flashmap;
5035         }
5036
5037         /* construct the net_device struct */
5038         netdev->netdev_ops              = &e1000e_netdev_ops;
5039         e1000e_set_ethtool_ops(netdev);
5040         netdev->watchdog_timeo          = 5 * HZ;
5041         netif_napi_add(netdev, &adapter->napi, e1000_clean, 64);
5042         strncpy(netdev->name, pci_name(pdev), sizeof(netdev->name) - 1);
5043
5044         netdev->mem_start = mmio_start;
5045         netdev->mem_end = mmio_start + mmio_len;
5046
5047         adapter->bd_number = cards_found++;
5048
5049         e1000e_check_options(adapter);
5050
5051         /* setup adapter struct */
5052         err = e1000_sw_init(adapter);
5053         if (err)
5054                 goto err_sw_init;
5055
5056         err = -EIO;
5057
5058         memcpy(&hw->mac.ops, ei->mac_ops, sizeof(hw->mac.ops));
5059         memcpy(&hw->nvm.ops, ei->nvm_ops, sizeof(hw->nvm.ops));
5060         memcpy(&hw->phy.ops, ei->phy_ops, sizeof(hw->phy.ops));
5061
5062         err = ei->get_variants(adapter);
5063         if (err)
5064                 goto err_hw_init;
5065
5066         if ((adapter->flags & FLAG_IS_ICH) &&
5067             (adapter->flags & FLAG_READ_ONLY_NVM))
5068                 e1000e_write_protect_nvm_ich8lan(&adapter->hw);
5069
5070         hw->mac.ops.get_bus_info(&adapter->hw);
5071
5072         adapter->hw.phy.autoneg_wait_to_complete = 0;
5073
5074         /* Copper options */
5075         if (adapter->hw.phy.media_type == e1000_media_type_copper) {
5076                 adapter->hw.phy.mdix = AUTO_ALL_MODES;
5077                 adapter->hw.phy.disable_polarity_correction = 0;
5078                 adapter->hw.phy.ms_type = e1000_ms_hw_default;
5079         }
5080
5081         if (e1000_check_reset_block(&adapter->hw))
5082                 e_info("PHY reset is blocked due to SOL/IDER session.\n");
5083
5084         netdev->features = NETIF_F_SG |
5085                            NETIF_F_HW_CSUM |
5086                            NETIF_F_HW_VLAN_TX |
5087                            NETIF_F_HW_VLAN_RX;
5088
5089         if (adapter->flags & FLAG_HAS_HW_VLAN_FILTER)
5090                 netdev->features |= NETIF_F_HW_VLAN_FILTER;
5091
5092         netdev->features |= NETIF_F_TSO;
5093         netdev->features |= NETIF_F_TSO6;
5094
5095         netdev->vlan_features |= NETIF_F_TSO;
5096         netdev->vlan_features |= NETIF_F_TSO6;
5097         netdev->vlan_features |= NETIF_F_HW_CSUM;
5098         netdev->vlan_features |= NETIF_F_SG;
5099
5100         if (pci_using_dac)
5101                 netdev->features |= NETIF_F_HIGHDMA;
5102
5103         if (e1000e_enable_mng_pass_thru(&adapter->hw))
5104                 adapter->flags |= FLAG_MNG_PT_ENABLED;
5105
5106         /*
5107          * before reading the NVM, reset the controller to
5108          * put the device in a known good starting state
5109          */
5110         adapter->hw.mac.ops.reset_hw(&adapter->hw);
5111
5112         /*
5113          * systems with ASPM and others may see the checksum fail on the first
5114          * attempt. Let's give it a few tries
5115          */
5116         for (i = 0;; i++) {
5117                 if (e1000_validate_nvm_checksum(&adapter->hw) >= 0)
5118                         break;
5119                 if (i == 2) {
5120                         e_err("The NVM Checksum Is Not Valid\n");
5121                         err = -EIO;
5122                         goto err_eeprom;
5123                 }
5124         }
5125
5126         e1000_eeprom_checks(adapter);
5127
5128         /* copy the MAC address out of the NVM */
5129         if (e1000e_read_mac_addr(&adapter->hw))
5130                 e_err("NVM Read Error while reading MAC address\n");
5131
5132         memcpy(netdev->dev_addr, adapter->hw.mac.addr, netdev->addr_len);
5133         memcpy(netdev->perm_addr, adapter->hw.mac.addr, netdev->addr_len);
5134
5135         if (!is_valid_ether_addr(netdev->perm_addr)) {
5136                 e_err("Invalid MAC Address: %pM\n", netdev->perm_addr);
5137                 err = -EIO;
5138                 goto err_eeprom;
5139         }
5140
5141         init_timer(&adapter->watchdog_timer);
5142         adapter->watchdog_timer.function = &e1000_watchdog;
5143         adapter->watchdog_timer.data = (unsigned long) adapter;
5144
5145         init_timer(&adapter->phy_info_timer);
5146         adapter->phy_info_timer.function = &e1000_update_phy_info;
5147         adapter->phy_info_timer.data = (unsigned long) adapter;
5148
5149         INIT_WORK(&adapter->reset_task, e1000_reset_task);
5150         INIT_WORK(&adapter->watchdog_task, e1000_watchdog_task);
5151         INIT_WORK(&adapter->downshift_task, e1000e_downshift_workaround);
5152         INIT_WORK(&adapter->update_phy_task, e1000e_update_phy_task);
5153
5154         /* Initialize link parameters. User can change them with ethtool */
5155         adapter->hw.mac.autoneg = 1;
5156         adapter->fc_autoneg = 1;
5157         adapter->hw.fc.requested_mode = e1000_fc_default;
5158         adapter->hw.fc.current_mode = e1000_fc_default;
5159         adapter->hw.phy.autoneg_advertised = 0x2f;
5160
5161         /* ring size defaults */
5162         adapter->rx_ring->count = 256;
5163         adapter->tx_ring->count = 256;
5164
5165         /*
5166          * Initial Wake on LAN setting - If APM wake is enabled in
5167          * the EEPROM, enable the ACPI Magic Packet filter
5168          */
5169         if (adapter->flags & FLAG_APME_IN_WUC) {
5170                 /* APME bit in EEPROM is mapped to WUC.APME */
5171                 eeprom_data = er32(WUC);
5172                 eeprom_apme_mask = E1000_WUC_APME;
5173                 if (eeprom_data & E1000_WUC_PHY_WAKE)
5174                         adapter->flags2 |= FLAG2_HAS_PHY_WAKEUP;
5175         } else if (adapter->flags & FLAG_APME_IN_CTRL3) {
5176                 if (adapter->flags & FLAG_APME_CHECK_PORT_B &&
5177                     (adapter->hw.bus.func == 1))
5178                         e1000_read_nvm(&adapter->hw,
5179                                 NVM_INIT_CONTROL3_PORT_B, 1, &eeprom_data);
5180                 else
5181                         e1000_read_nvm(&adapter->hw,
5182                                 NVM_INIT_CONTROL3_PORT_A, 1, &eeprom_data);
5183         }
5184
5185         /* fetch WoL from EEPROM */
5186         if (eeprom_data & eeprom_apme_mask)
5187                 adapter->eeprom_wol |= E1000_WUFC_MAG;
5188
5189         /*
5190          * now that we have the eeprom settings, apply the special cases
5191          * where the eeprom may be wrong or the board simply won't support
5192          * wake on lan on a particular port
5193          */
5194         if (!(adapter->flags & FLAG_HAS_WOL))
5195                 adapter->eeprom_wol = 0;
5196
5197         /* initialize the wol settings based on the eeprom settings */
5198         adapter->wol = adapter->eeprom_wol;
5199         device_set_wakeup_enable(&adapter->pdev->dev, adapter->wol);
5200
5201         /* save off EEPROM version number */
5202         e1000_read_nvm(&adapter->hw, 5, 1, &adapter->eeprom_vers);
5203
5204         /* reset the hardware with the new settings */
5205         e1000e_reset(adapter);
5206
5207         /*
5208          * If the controller has AMT, do not set DRV_LOAD until the interface
5209          * is up.  For all other cases, let the f/w know that the h/w is now
5210          * under the control of the driver.
5211          */
5212         if (!(adapter->flags & FLAG_HAS_AMT))
5213                 e1000_get_hw_control(adapter);
5214
5215         strcpy(netdev->name, "eth%d");
5216         err = register_netdev(netdev);
5217         if (err)
5218                 goto err_register;
5219
5220         /* carrier off reporting is important to ethtool even BEFORE open */
5221         netif_carrier_off(netdev);
5222
5223         e1000_print_device_info(adapter);
5224
5225         return 0;
5226
5227 err_register:
5228         if (!(adapter->flags & FLAG_HAS_AMT))
5229                 e1000_release_hw_control(adapter);
5230 err_eeprom:
5231         if (!e1000_check_reset_block(&adapter->hw))
5232                 e1000_phy_hw_reset(&adapter->hw);
5233 err_hw_init:
5234
5235         kfree(adapter->tx_ring);
5236         kfree(adapter->rx_ring);
5237 err_sw_init:
5238         if (adapter->hw.flash_address)
5239                 iounmap(adapter->hw.flash_address);
5240         e1000e_reset_interrupt_capability(adapter);
5241 err_flashmap:
5242         iounmap(adapter->hw.hw_addr);
5243 err_ioremap:
5244         free_netdev(netdev);
5245 err_alloc_etherdev:
5246         pci_release_selected_regions(pdev,
5247                                      pci_select_bars(pdev, IORESOURCE_MEM));
5248 err_pci_reg:
5249 err_dma:
5250         pci_disable_device(pdev);
5251         return err;
5252 }
5253
5254 /**
5255  * e1000_remove - Device Removal Routine
5256  * @pdev: PCI device information struct
5257  *
5258  * e1000_remove is called by the PCI subsystem to alert the driver
5259  * that it should release a PCI device.  The could be caused by a
5260  * Hot-Plug event, or because the driver is going to be removed from
5261  * memory.
5262  **/
5263 static void __devexit e1000_remove(struct pci_dev *pdev)
5264 {
5265         struct net_device *netdev = pci_get_drvdata(pdev);
5266         struct e1000_adapter *adapter = netdev_priv(netdev);
5267         int err;
5268
5269         /*
5270          * flush_scheduled work may reschedule our watchdog task, so
5271          * explicitly disable watchdog tasks from being rescheduled
5272          */
5273         set_bit(__E1000_DOWN, &adapter->state);
5274         del_timer_sync(&adapter->watchdog_timer);
5275         del_timer_sync(&adapter->phy_info_timer);
5276
5277         flush_scheduled_work();
5278
5279         /*
5280          * Release control of h/w to f/w.  If f/w is AMT enabled, this
5281          * would have already happened in close and is redundant.
5282          */
5283         e1000_release_hw_control(adapter);
5284
5285         unregister_netdev(netdev);
5286
5287         if (!e1000_check_reset_block(&adapter->hw))
5288                 e1000_phy_hw_reset(&adapter->hw);
5289
5290         e1000e_reset_interrupt_capability(adapter);
5291         kfree(adapter->tx_ring);
5292         kfree(adapter->rx_ring);
5293
5294         iounmap(adapter->hw.hw_addr);
5295         if (adapter->hw.flash_address)
5296                 iounmap(adapter->hw.flash_address);
5297         pci_release_selected_regions(pdev,
5298                                      pci_select_bars(pdev, IORESOURCE_MEM));
5299
5300         free_netdev(netdev);
5301
5302         /* AER disable */
5303         err = pci_disable_pcie_error_reporting(pdev);
5304         if (err)
5305                 dev_err(&pdev->dev,
5306                         "pci_disable_pcie_error_reporting failed 0x%x\n", err);
5307
5308         pci_disable_device(pdev);
5309 }
5310
5311 /* PCI Error Recovery (ERS) */
5312 static struct pci_error_handlers e1000_err_handler = {
5313         .error_detected = e1000_io_error_detected,
5314         .slot_reset = e1000_io_slot_reset,
5315         .resume = e1000_io_resume,
5316 };
5317
5318 static struct pci_device_id e1000_pci_tbl[] = {
5319         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82571EB_COPPER), board_82571 },
5320         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82571EB_FIBER), board_82571 },
5321         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82571EB_QUAD_COPPER), board_82571 },
5322         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82571EB_QUAD_COPPER_LP), board_82571 },
5323         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82571EB_QUAD_FIBER), board_82571 },
5324         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82571EB_SERDES), board_82571 },
5325         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82571EB_SERDES_DUAL), board_82571 },
5326         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82571EB_SERDES_QUAD), board_82571 },
5327         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82571PT_QUAD_COPPER), board_82571 },
5328
5329         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82572EI), board_82572 },
5330         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82572EI_COPPER), board_82572 },
5331         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82572EI_FIBER), board_82572 },
5332         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82572EI_SERDES), board_82572 },
5333
5334         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82573E), board_82573 },
5335         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82573E_IAMT), board_82573 },
5336         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82573L), board_82573 },
5337
5338         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82574L), board_82574 },
5339         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82574LA), board_82574 },
5340         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82583V), board_82583 },
5341
5342         { PCI_VDEVICE(INTEL, E1000_DEV_ID_80003ES2LAN_COPPER_DPT),
5343           board_80003es2lan },
5344         { PCI_VDEVICE(INTEL, E1000_DEV_ID_80003ES2LAN_COPPER_SPT),
5345           board_80003es2lan },
5346         { PCI_VDEVICE(INTEL, E1000_DEV_ID_80003ES2LAN_SERDES_DPT),
5347           board_80003es2lan },
5348         { PCI_VDEVICE(INTEL, E1000_DEV_ID_80003ES2LAN_SERDES_SPT),
5349           board_80003es2lan },
5350
5351         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH8_IFE), board_ich8lan },
5352         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH8_IFE_G), board_ich8lan },
5353         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH8_IFE_GT), board_ich8lan },
5354         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH8_IGP_AMT), board_ich8lan },
5355         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH8_IGP_C), board_ich8lan },
5356         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH8_IGP_M), board_ich8lan },
5357         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH8_IGP_M_AMT), board_ich8lan },
5358
5359         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH9_IFE), board_ich9lan },
5360         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH9_IFE_G), board_ich9lan },
5361         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH9_IFE_GT), board_ich9lan },
5362         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH9_IGP_AMT), board_ich9lan },
5363         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH9_IGP_C), board_ich9lan },
5364         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH9_BM), board_ich9lan },
5365         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH9_IGP_M), board_ich9lan },
5366         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH9_IGP_M_AMT), board_ich9lan },
5367         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH9_IGP_M_V), board_ich9lan },
5368
5369         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH10_R_BM_LM), board_ich9lan },
5370         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH10_R_BM_LF), board_ich9lan },
5371         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH10_R_BM_V), board_ich9lan },
5372
5373         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH10_D_BM_LM), board_ich10lan },
5374         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH10_D_BM_LF), board_ich10lan },
5375
5376         { PCI_VDEVICE(INTEL, E1000_DEV_ID_PCH_M_HV_LM), board_pchlan },
5377         { PCI_VDEVICE(INTEL, E1000_DEV_ID_PCH_M_HV_LC), board_pchlan },
5378         { PCI_VDEVICE(INTEL, E1000_DEV_ID_PCH_D_HV_DM), board_pchlan },
5379         { PCI_VDEVICE(INTEL, E1000_DEV_ID_PCH_D_HV_DC), board_pchlan },
5380
5381         { }     /* terminate list */
5382 };
5383 MODULE_DEVICE_TABLE(pci, e1000_pci_tbl);
5384
5385 /* PCI Device API Driver */
5386 static struct pci_driver e1000_driver = {
5387         .name     = e1000e_driver_name,
5388         .id_table = e1000_pci_tbl,
5389         .probe    = e1000_probe,
5390         .remove   = __devexit_p(e1000_remove),
5391 #ifdef CONFIG_PM
5392         /* Power Management Hooks */
5393         .suspend  = e1000_suspend,
5394         .resume   = e1000_resume,
5395 #endif
5396         .shutdown = e1000_shutdown,
5397         .err_handler = &e1000_err_handler
5398 };
5399
5400 /**
5401  * e1000_init_module - Driver Registration Routine
5402  *
5403  * e1000_init_module is the first routine called when the driver is
5404  * loaded. All it does is register with the PCI subsystem.
5405  **/
5406 static int __init e1000_init_module(void)
5407 {
5408         int ret;
5409         printk(KERN_INFO "%s: Intel(R) PRO/1000 Network Driver - %s\n",
5410                e1000e_driver_name, e1000e_driver_version);
5411         printk(KERN_INFO "%s: Copyright (c) 1999-2008 Intel Corporation.\n",
5412                e1000e_driver_name);
5413         ret = pci_register_driver(&e1000_driver);
5414         pm_qos_add_requirement(PM_QOS_CPU_DMA_LATENCY, e1000e_driver_name,
5415                                PM_QOS_DEFAULT_VALUE);
5416
5417         return ret;
5418 }
5419 module_init(e1000_init_module);
5420
5421 /**
5422  * e1000_exit_module - Driver Exit Cleanup Routine
5423  *
5424  * e1000_exit_module is called just before the driver is removed
5425  * from memory.
5426  **/
5427 static void __exit e1000_exit_module(void)
5428 {
5429         pci_unregister_driver(&e1000_driver);
5430         pm_qos_remove_requirement(PM_QOS_CPU_DMA_LATENCY, e1000e_driver_name);
5431 }
5432 module_exit(e1000_exit_module);
5433
5434
5435 MODULE_AUTHOR("Intel Corporation, <linux.nics@intel.com>");
5436 MODULE_DESCRIPTION("Intel(R) PRO/1000 Network Driver");
5437 MODULE_LICENSE("GPL");
5438 MODULE_VERSION(DRV_VERSION);
5439
5440 /* e1000_main.c */