drivers/net/e1000e/netdev.c

   1 /*******************************************************************************
   2
   3   Intel PRO/1000 Linux driver
   4   Copyright(c) 1999 - 2009 Intel Corporation.
   5
   6   This program is free software; you can redistribute it and/or modify it
   7   under the terms and conditions of the GNU General Public License,
   8   version 2, as published by the Free Software Foundation.
   9
  10   This program is distributed in the hope it will be useful, but WITHOUT
  11   ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
  12   FITNESS FOR A PARTICULAR PURPOSE.  See the GNU General Public License for
  13   more details.
  14
  15   You should have received a copy of the GNU General Public License along with
  16   this program; if not, write to the Free Software Foundation, Inc.,
  17   51 Franklin St - Fifth Floor, Boston, MA 02110-1301 USA.
  18
  19   The full GNU General Public License is included in this distribution in
  20   the file called "COPYING".
  21
  22   Contact Information:
  23   Linux NICS <linux.nics@intel.com>
  24   e1000-devel Mailing List <e1000-devel@lists.sourceforge.net>
  25   Intel Corporation, 5200 N.E. Elam Young Parkway, Hillsboro, OR 97124-6497
  26
  27 *******************************************************************************/
  28
  29 #include <linux/module.h>
  30 #include <linux/types.h>
  31 #include <linux/init.h>
  32 #include <linux/pci.h>
  33 #include <linux/vmalloc.h>
  34 #include <linux/pagemap.h>
  35 #include <linux/delay.h>
  36 #include <linux/netdevice.h>
  37 #include <linux/tcp.h>
  38 #include <linux/ipv6.h>
  39 #include <net/checksum.h>
  40 #include <net/ip6_checksum.h>
  41 #include <linux/mii.h>
  42 #include <linux/ethtool.h>
  43 #include <linux/if_vlan.h>
  44 #include <linux/cpu.h>
  45 #include <linux/smp.h>
  46 #include <linux/pm_qos_params.h>
  47 #include <linux/aer.h>
  48
  49 #include "e1000.h"
  50
  51 #define DRV_VERSION "1.0.2-k2"
  52 char e1000e_driver_name[] = "e1000e";
  53 const char e1000e_driver_version[] = DRV_VERSION;
  54
  55 static const struct e1000_info *e1000_info_tbl[] = {
  56         [board_82571]           = &e1000_82571_info,
  57         [board_82572]           = &e1000_82572_info,
  58         [board_82573]           = &e1000_82573_info,
  59         [board_82574]           = &e1000_82574_info,
  60         [board_82583]           = &e1000_82583_info,
  61         [board_80003es2lan]     = &e1000_es2_info,
  62         [board_ich8lan]         = &e1000_ich8_info,
  63         [board_ich9lan]         = &e1000_ich9_info,
  64         [board_ich10lan]        = &e1000_ich10_info,
  65         [board_pchlan]          = &e1000_pch_info,
  66 };
  67
  68 /**
  69  * e1000_desc_unused - calculate if we have unused descriptors
  70  **/
  71 static int e1000_desc_unused(struct e1000_ring *ring)
  72 {
  73         if (ring->next_to_clean > ring->next_to_use)
  74                 return ring->next_to_clean - ring->next_to_use - 1;
  75
  76         return ring->count + ring->next_to_clean - ring->next_to_use - 1;
  77 }
  78
  79 /**
  80  * e1000_receive_skb - helper function to handle Rx indications
  81  * @adapter: board private structure
  82  * @status: descriptor status field as written by hardware
  83  * @vlan: descriptor vlan field as written by hardware (no le/be conversion)
  84  * @skb: pointer to sk_buff to be indicated to stack
  85  **/
  86 static void e1000_receive_skb(struct e1000_adapter *adapter,
  87                               struct net_device *netdev,
  88                               struct sk_buff *skb,
  89                               u8 status, __le16 vlan)
  90 {
  91         skb->protocol = eth_type_trans(skb, netdev);
  92
  93         if (adapter->vlgrp && (status & E1000_RXD_STAT_VP))
  94                 vlan_gro_receive(&adapter->napi, adapter->vlgrp,
  95                                  le16_to_cpu(vlan), skb);
  96         else
  97                 napi_gro_receive(&adapter->napi, skb);
  98 }
  99
 100 /**
 101  * e1000_rx_checksum - Receive Checksum Offload for 82543
 102  * @adapter:     board private structure
 103  * @status_err:  receive descriptor status and error fields
 104  * @csum:       receive descriptor csum field
 105  * @sk_buff:     socket buffer with received data
 106  **/
 107 static void e1000_rx_checksum(struct e1000_adapter *adapter, u32 status_err,
 108                               u32 csum, struct sk_buff *skb)
 109 {
 110         u16 status = (u16)status_err;
 111         u8 errors = (u8)(status_err >> 24);
 112         skb->ip_summed = CHECKSUM_NONE;
 113
 114         /* Ignore Checksum bit is set */
 115         if (status & E1000_RXD_STAT_IXSM)
 116                 return;
 117         /* TCP/UDP checksum error bit is set */
 118         if (errors & E1000_RXD_ERR_TCPE) {
 119                 /* let the stack verify checksum errors */
 120                 adapter->hw_csum_err++;
 121                 return;
 122         }
 123
 124         /* TCP/UDP Checksum has not been calculated */
 125         if (!(status & (E1000_RXD_STAT_TCPCS | E1000_RXD_STAT_UDPCS)))
 126                 return;
 127
 128         /* It must be a TCP or UDP packet with a valid checksum */
 129         if (status & E1000_RXD_STAT_TCPCS) {
 130                 /* TCP checksum is good */
 131                 skb->ip_summed = CHECKSUM_UNNECESSARY;
 132         } else {
 133                 /*
 134                  * IP fragment with UDP payload
 135                  * Hardware complements the payload checksum, so we undo it
 136                  * and then put the value in host order for further stack use.
 137                  */
 138                 __sum16 sum = (__force __sum16)htons(csum);
 139                 skb->csum = csum_unfold(~sum);
 140                 skb->ip_summed = CHECKSUM_COMPLETE;
 141         }
 142         adapter->hw_csum_good++;
 143 }
 144
 145 /**
 146  * e1000_alloc_rx_buffers - Replace used receive buffers; legacy & extended
 147  * @adapter: address of board private structure
 148  **/
 149 static void e1000_alloc_rx_buffers(struct e1000_adapter *adapter,
 150                                    int cleaned_count)
 151 {
 152         struct net_device *netdev = adapter->netdev;
 153         struct pci_dev *pdev = adapter->pdev;
 154         struct e1000_ring *rx_ring = adapter->rx_ring;
 155         struct e1000_rx_desc *rx_desc;
 156         struct e1000_buffer *buffer_info;
 157         struct sk_buff *skb;
 158         unsigned int i;
 159         unsigned int bufsz = adapter->rx_buffer_len;
 160
 161         i = rx_ring->next_to_use;
 162         buffer_info = &rx_ring->buffer_info[i];
 163
 164         while (cleaned_count--) {
 165                 skb = buffer_info->skb;
 166                 if (skb) {
 167                         skb_trim(skb, 0);
 168                         goto map_skb;
 169                 }
 170
 171                 skb = netdev_alloc_skb_ip_align(netdev, bufsz);
 172                 if (!skb) {
 173                         /* Better luck next round */
 174                         adapter->alloc_rx_buff_failed++;
 175                         break;
 176                 }
 177
 178                 buffer_info->skb = skb;
 179 map_skb:
 180                 buffer_info->dma = pci_map_single(pdev, skb->data,
 181                                                   adapter->rx_buffer_len,
 182                                                   PCI_DMA_FROMDEVICE);
 183                 if (pci_dma_mapping_error(pdev, buffer_info->dma)) {
 184                         dev_err(&pdev->dev, "RX DMA map failed\n");
 185                         adapter->rx_dma_failed++;
 186                         break;
 187                 }
 188
 189                 rx_desc = E1000_RX_DESC(*rx_ring, i);
 190                 rx_desc->buffer_addr = cpu_to_le64(buffer_info->dma);
 191
 192                 i++;
 193                 if (i == rx_ring->count)
 194                         i = 0;
 195                 buffer_info = &rx_ring->buffer_info[i];
 196         }
 197
 198         if (rx_ring->next_to_use != i) {
 199                 rx_ring->next_to_use = i;
 200                 if (i-- == 0)
 201                         i = (rx_ring->count - 1);
 202
 203                 /*
 204                  * Force memory writes to complete before letting h/w
 205                  * know there are new descriptors to fetch.  (Only
 206                  * applicable for weak-ordered memory model archs,
 207                  * such as IA-64).
 208                  */
 209                 wmb();
 210                 writel(i, adapter->hw.hw_addr + rx_ring->tail);
 211         }
 212 }
 213
 214 /**
 215  * e1000_alloc_rx_buffers_ps - Replace used receive buffers; packet split
 216  * @adapter: address of board private structure
 217  **/
 218 static void e1000_alloc_rx_buffers_ps(struct e1000_adapter *adapter,
 219                                       int cleaned_count)
 220 {
 221         struct net_device *netdev = adapter->netdev;
 222         struct pci_dev *pdev = adapter->pdev;
 223         union e1000_rx_desc_packet_split *rx_desc;
 224         struct e1000_ring *rx_ring = adapter->rx_ring;
 225         struct e1000_buffer *buffer_info;
 226         struct e1000_ps_page *ps_page;
 227         struct sk_buff *skb;
 228         unsigned int i, j;
 229
 230         i = rx_ring->next_to_use;
 231         buffer_info = &rx_ring->buffer_info[i];
 232
 233         while (cleaned_count--) {
 234                 rx_desc = E1000_RX_DESC_PS(*rx_ring, i);
 235
 236                 for (j = 0; j < PS_PAGE_BUFFERS; j++) {
 237                         ps_page = &buffer_info->ps_pages[j];
 238                         if (j >= adapter->rx_ps_pages) {
 239                                 /* all unused desc entries get hw null ptr */
 240                                 rx_desc->read.buffer_addr[j+1] = ~cpu_to_le64(0);
 241                                 continue;
 242                         }
 243                         if (!ps_page->page) {
 244                                 ps_page->page = alloc_page(GFP_ATOMIC);
 245                                 if (!ps_page->page) {
 246                                         adapter->alloc_rx_buff_failed++;
 247                                         goto no_buffers;
 248                                 }
 249                                 ps_page->dma = pci_map_page(pdev,
 250                                                    ps_page->page,
 251                                                    0, PAGE_SIZE,
 252                                                    PCI_DMA_FROMDEVICE);
 253                                 if (pci_dma_mapping_error(pdev, ps_page->dma)) {
 254                                         dev_err(&adapter->pdev->dev,
 255                                           "RX DMA page map failed\n");
 256                                         adapter->rx_dma_failed++;
 257                                         goto no_buffers;
 258                                 }
 259                         }
 260                         /*
 261                          * Refresh the desc even if buffer_addrs
 262                          * didn't change because each write-back
 263                          * erases this info.
 264                          */
 265                         rx_desc->read.buffer_addr[j+1] =
 266                              cpu_to_le64(ps_page->dma);
 267                 }
 268
 269                 skb = netdev_alloc_skb_ip_align(netdev,
 270                                                 adapter->rx_ps_bsize0);
 271
 272                 if (!skb) {
 273                         adapter->alloc_rx_buff_failed++;
 274                         break;
 275                 }
 276
 277                 buffer_info->skb = skb;
 278                 buffer_info->dma = pci_map_single(pdev, skb->data,
 279                                                   adapter->rx_ps_bsize0,
 280                                                   PCI_DMA_FROMDEVICE);
 281                 if (pci_dma_mapping_error(pdev, buffer_info->dma)) {
 282                         dev_err(&pdev->dev, "RX DMA map failed\n");
 283                         adapter->rx_dma_failed++;
 284                         /* cleanup skb */
 285                         dev_kfree_skb_any(skb);
 286                         buffer_info->skb = NULL;
 287                         break;
 288                 }
 289
 290                 rx_desc->read.buffer_addr[0] = cpu_to_le64(buffer_info->dma);
 291
 292                 i++;
 293                 if (i == rx_ring->count)
 294                         i = 0;
 295                 buffer_info = &rx_ring->buffer_info[i];
 296         }
 297
 298 no_buffers:
 299         if (rx_ring->next_to_use != i) {
 300                 rx_ring->next_to_use = i;
 301
 302                 if (!(i--))
 303                         i = (rx_ring->count - 1);
 304
 305                 /*
 306                  * Force memory writes to complete before letting h/w
 307                  * know there are new descriptors to fetch.  (Only
 308                  * applicable for weak-ordered memory model archs,
 309                  * such as IA-64).
 310                  */
 311                 wmb();
 312                 /*
 313                  * Hardware increments by 16 bytes, but packet split
 314                  * descriptors are 32 bytes...so we increment tail
 315                  * twice as much.
 316                  */
 317                 writel(i<<1, adapter->hw.hw_addr + rx_ring->tail);
 318         }
 319 }
 320
 321 /**
 322  * e1000_alloc_jumbo_rx_buffers - Replace used jumbo receive buffers
 323  * @adapter: address of board private structure
 324  * @cleaned_count: number of buffers to allocate this pass
 325  **/
 326
 327 static void e1000_alloc_jumbo_rx_buffers(struct e1000_adapter *adapter,
 328                                          int cleaned_count)
 329 {
 330         struct net_device *netdev = adapter->netdev;
 331         struct pci_dev *pdev = adapter->pdev;
 332         struct e1000_rx_desc *rx_desc;
 333         struct e1000_ring *rx_ring = adapter->rx_ring;
 334         struct e1000_buffer *buffer_info;
 335         struct sk_buff *skb;
 336         unsigned int i;
 337         unsigned int bufsz = 256 - 16 /* for skb_reserve */;
 338
 339         i = rx_ring->next_to_use;
 340         buffer_info = &rx_ring->buffer_info[i];
 341
 342         while (cleaned_count--) {
 343                 skb = buffer_info->skb;
 344                 if (skb) {
 345                         skb_trim(skb, 0);
 346                         goto check_page;
 347                 }
 348
 349                 skb = netdev_alloc_skb_ip_align(netdev, bufsz);
 350                 if (unlikely(!skb)) {
 351                         /* Better luck next round */
 352                         adapter->alloc_rx_buff_failed++;
 353                         break;
 354                 }
 355
 356                 buffer_info->skb = skb;
 357 check_page:
 358                 /* allocate a new page if necessary */
 359                 if (!buffer_info->page) {
 360                         buffer_info->page = alloc_page(GFP_ATOMIC);
 361                         if (unlikely(!buffer_info->page)) {
 362                                 adapter->alloc_rx_buff_failed++;
 363                                 break;
 364                         }
 365                 }
 366
 367                 if (!buffer_info->dma)
 368                         buffer_info->dma = pci_map_page(pdev,
 369                                                         buffer_info->page, 0,
 370                                                         PAGE_SIZE,
 371                                                         PCI_DMA_FROMDEVICE);
 372
 373                 rx_desc = E1000_RX_DESC(*rx_ring, i);
 374                 rx_desc->buffer_addr = cpu_to_le64(buffer_info->dma);
 375
 376                 if (unlikely(++i == rx_ring->count))
 377                         i = 0;
 378                 buffer_info = &rx_ring->buffer_info[i];
 379         }
 380
 381         if (likely(rx_ring->next_to_use != i)) {
 382                 rx_ring->next_to_use = i;
 383                 if (unlikely(i-- == 0))
 384                         i = (rx_ring->count - 1);
 385
 386                 /* Force memory writes to complete before letting h/w
 387                  * know there are new descriptors to fetch.  (Only
 388                  * applicable for weak-ordered memory model archs,
 389                  * such as IA-64). */
 390                 wmb();
 391                 writel(i, adapter->hw.hw_addr + rx_ring->tail);
 392         }
 393 }
 394
 395 /**
 396  * e1000_clean_rx_irq - Send received data up the network stack; legacy
 397  * @adapter: board private structure
 398  *
 399  * the return value indicates whether actual cleaning was done, there
 400  * is no guarantee that everything was cleaned
 401  **/
 402 static bool e1000_clean_rx_irq(struct e1000_adapter *adapter,
 403                                int *work_done, int work_to_do)
 404 {
 405         struct net_device *netdev = adapter->netdev;
 406         struct pci_dev *pdev = adapter->pdev;
 407         struct e1000_hw *hw = &adapter->hw;
 408         struct e1000_ring *rx_ring = adapter->rx_ring;
 409         struct e1000_rx_desc *rx_desc, *next_rxd;
 410         struct e1000_buffer *buffer_info, *next_buffer;
 411         u32 length;
 412         unsigned int i;
 413         int cleaned_count = 0;
 414         bool cleaned = 0;
 415         unsigned int total_rx_bytes = 0, total_rx_packets = 0;
 416
 417         i = rx_ring->next_to_clean;
 418         rx_desc = E1000_RX_DESC(*rx_ring, i);
 419         buffer_info = &rx_ring->buffer_info[i];
 420
 421         while (rx_desc->status & E1000_RXD_STAT_DD) {
 422                 struct sk_buff *skb;
 423                 u8 status;
 424
 425                 if (*work_done >= work_to_do)
 426                         break;
 427                 (*work_done)++;
 428
 429                 status = rx_desc->status;
 430                 skb = buffer_info->skb;
 431                 buffer_info->skb = NULL;
 432
 433                 prefetch(skb->data - NET_IP_ALIGN);
 434
 435                 i++;
 436                 if (i == rx_ring->count)
 437                         i = 0;
 438                 next_rxd = E1000_RX_DESC(*rx_ring, i);
 439                 prefetch(next_rxd);
 440
 441                 next_buffer = &rx_ring->buffer_info[i];
 442
 443                 cleaned = 1;
 444                 cleaned_count++;
 445                 pci_unmap_single(pdev,
 446                                  buffer_info->dma,
 447                                  adapter->rx_buffer_len,
 448                                  PCI_DMA_FROMDEVICE);
 449                 buffer_info->dma = 0;
 450
 451                 length = le16_to_cpu(rx_desc->length);
 452
 453                 /*
 454                  * !EOP means multiple descriptors were used to store a single
 455                  * packet, if that's the case we need to toss it.  In fact, we
 456                  * need to toss every packet with the EOP bit clear and the
 457                  * next frame that _does_ have the EOP bit set, as it is by
 458                  * definition only a frame fragment
 459                  */
 460                 if (unlikely(!(status & E1000_RXD_STAT_EOP)))
 461                         adapter->flags2 |= FLAG2_IS_DISCARDING;
 462
 463                 if (adapter->flags2 & FLAG2_IS_DISCARDING) {
 464                         /* All receives must fit into a single buffer */
 465                         e_dbg("Receive packet consumed multiple buffers\n");
 466                         /* recycle */
 467                         buffer_info->skb = skb;
 468                         if (status & E1000_RXD_STAT_EOP)
 469                                 adapter->flags2 &= ~FLAG2_IS_DISCARDING;
 470                         goto next_desc;
 471                 }
 472
 473                 if (rx_desc->errors & E1000_RXD_ERR_FRAME_ERR_MASK) {
 474                         /* recycle */
 475                         buffer_info->skb = skb;
 476                         goto next_desc;
 477                 }
 478
 479                 /* adjust length to remove Ethernet CRC */
 480                 if (!(adapter->flags2 & FLAG2_CRC_STRIPPING))
 481                         length -= 4;
 482
 483                 total_rx_bytes += length;
 484                 total_rx_packets++;
 485
 486                 /*
 487                  * code added for copybreak, this should improve
 488                  * performance for small packets with large amounts
 489                  * of reassembly being done in the stack
 490                  */
 491                 if (length < copybreak) {
 492                         struct sk_buff *new_skb =
 493                             netdev_alloc_skb_ip_align(netdev, length);
 494                         if (new_skb) {
 495                                 skb_copy_to_linear_data_offset(new_skb,
 496                                                                -NET_IP_ALIGN,
 497                                                                (skb->data -
 498                                                                 NET_IP_ALIGN),
 499                                                                (length +
 500                                                                 NET_IP_ALIGN));
 501                                 /* save the skb in buffer_info as good */
 502                                 buffer_info->skb = skb;
 503                                 skb = new_skb;
 504                         }
 505                         /* else just continue with the old one */
 506                 }
 507                 /* end copybreak code */
 508                 skb_put(skb, length);
 509
 510                 /* Receive Checksum Offload */
 511                 e1000_rx_checksum(adapter,
 512                                   (u32)(status) |
 513                                   ((u32)(rx_desc->errors) << 24),
 514                                   le16_to_cpu(rx_desc->csum), skb);
 515
 516                 e1000_receive_skb(adapter, netdev, skb,status,rx_desc->special);
 517
 518 next_desc:
 519                 rx_desc->status = 0;
 520
 521                 /* return some buffers to hardware, one at a time is too slow */
 522                 if (cleaned_count >= E1000_RX_BUFFER_WRITE) {
 523                         adapter->alloc_rx_buf(adapter, cleaned_count);
 524                         cleaned_count = 0;
 525                 }
 526
 527                 /* use prefetched values */
 528                 rx_desc = next_rxd;
 529                 buffer_info = next_buffer;
 530         }
 531         rx_ring->next_to_clean = i;
 532
 533         cleaned_count = e1000_desc_unused(rx_ring);
 534         if (cleaned_count)
 535                 adapter->alloc_rx_buf(adapter, cleaned_count);
 536
 537         adapter->total_rx_bytes += total_rx_bytes;
 538         adapter->total_rx_packets += total_rx_packets;
 539         netdev->stats.rx_bytes += total_rx_bytes;
 540         netdev->stats.rx_packets += total_rx_packets;
 541         return cleaned;
 542 }
 543
 544 static void e1000_put_txbuf(struct e1000_adapter *adapter,
 545                              struct e1000_buffer *buffer_info)
 546 {
 547         if (buffer_info->dma) {
 548                 if (buffer_info->mapped_as_page)
 549                         pci_unmap_page(adapter->pdev, buffer_info->dma,
 550                                        buffer_info->length, PCI_DMA_TODEVICE);
 551                 else
 552                         pci_unmap_single(adapter->pdev, buffer_info->dma,
 553                                          buffer_info->length,
 554                                          PCI_DMA_TODEVICE);
 555                 buffer_info->dma = 0;
 556         }
 557         if (buffer_info->skb) {
 558                 dev_kfree_skb_any(buffer_info->skb);
 559                 buffer_info->skb = NULL;
 560         }
 561         buffer_info->time_stamp = 0;
 562 }
 563
 564 static void e1000_print_hw_hang(struct work_struct *work)
 565 {
 566         struct e1000_adapter *adapter = container_of(work,
 567                                                      struct e1000_adapter,
 568                                                      print_hang_task);
 569         struct e1000_ring *tx_ring = adapter->tx_ring;
 570         unsigned int i = tx_ring->next_to_clean;
 571         unsigned int eop = tx_ring->buffer_info[i].next_to_watch;
 572         struct e1000_tx_desc *eop_desc = E1000_TX_DESC(*tx_ring, eop);
 573         struct e1000_hw *hw = &adapter->hw;
 574         u16 phy_status, phy_1000t_status, phy_ext_status;
 575         u16 pci_status;
 576
 577         e1e_rphy(hw, PHY_STATUS, &phy_status);
 578         e1e_rphy(hw, PHY_1000T_STATUS, &phy_1000t_status);
 579         e1e_rphy(hw, PHY_EXT_STATUS, &phy_ext_status);
 580
 581         pci_read_config_word(adapter->pdev, PCI_STATUS, &pci_status);
 582
 583         /* detected Hardware unit hang */
 584         e_err("Detected Hardware Unit Hang:\n"
 585               "  TDH                  <%x>\n"
 586               "  TDT                  <%x>\n"
 587               "  next_to_use          <%x>\n"
 588               "  next_to_clean        <%x>\n"
 589               "buffer_info[next_to_clean]:\n"
 590               "  time_stamp           <%lx>\n"
 591               "  next_to_watch        <%x>\n"
 592               "  jiffies              <%lx>\n"
 593               "  next_to_watch.status <%x>\n"
 594               "MAC Status             <%x>\n"
 595               "PHY Status             <%x>\n"
 596               "PHY 1000BASE-T Status  <%x>\n"
 597               "PHY Extended Status    <%x>\n"
 598               "PCI 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               er32(STATUS),
 608               phy_status,
 609               phy_1000t_status,
 610               phy_ext_status,
 611               pci_status);
 612 }
 613
 614 /**
 615  * e1000_clean_tx_irq - Reclaim resources after transmit completes
 616  * @adapter: board private structure
 617  *
 618  * the return value indicates whether actual cleaning was done, there
 619  * is no guarantee that everything was cleaned
 620  **/
 621 static bool e1000_clean_tx_irq(struct e1000_adapter *adapter)
 622 {
 623         struct net_device *netdev = adapter->netdev;
 624         struct e1000_hw *hw = &adapter->hw;
 625         struct e1000_ring *tx_ring = adapter->tx_ring;
 626         struct e1000_tx_desc *tx_desc, *eop_desc;
 627         struct e1000_buffer *buffer_info;
 628         unsigned int i, eop;
 629         unsigned int count = 0;
 630         unsigned int total_tx_bytes = 0, total_tx_packets = 0;
 631
 632         i = tx_ring->next_to_clean;
 633         eop = tx_ring->buffer_info[i].next_to_watch;
 634         eop_desc = E1000_TX_DESC(*tx_ring, eop);
 635
 636         while ((eop_desc->upper.data & cpu_to_le32(E1000_TXD_STAT_DD)) &&
 637                (count < tx_ring->count)) {
 638                 bool cleaned = false;
 639                 for (; !cleaned; count++) {
 640                         tx_desc = E1000_TX_DESC(*tx_ring, i);
 641                         buffer_info = &tx_ring->buffer_info[i];
 642                         cleaned = (i == eop);
 643
 644                         if (cleaned) {
 645                                 struct sk_buff *skb = buffer_info->skb;
 646                                 unsigned int segs, bytecount;
 647                                 segs = skb_shinfo(skb)->gso_segs ?: 1;
 648                                 /* multiply data chunks by size of headers */
 649                                 bytecount = ((segs - 1) * skb_headlen(skb)) +
 650                                             skb->len;
 651                                 total_tx_packets += segs;
 652                                 total_tx_bytes += bytecount;
 653                         }
 654
 655                         e1000_put_txbuf(adapter, buffer_info);
 656                         tx_desc->upper.data = 0;
 657
 658                         i++;
 659                         if (i == tx_ring->count)
 660                                 i = 0;
 661                 }
 662
 663                 eop = tx_ring->buffer_info[i].next_to_watch;
 664                 eop_desc = E1000_TX_DESC(*tx_ring, eop);
 665         }
 666
 667         tx_ring->next_to_clean = i;
 668
 669 #define TX_WAKE_THRESHOLD 32
 670         if (count && netif_carrier_ok(netdev) &&
 671             e1000_desc_unused(tx_ring) >= TX_WAKE_THRESHOLD) {
 672                 /* Make sure that anybody stopping the queue after this
 673                  * sees the new next_to_clean.
 674                  */
 675                 smp_mb();
 676
 677                 if (netif_queue_stopped(netdev) &&
 678                     !(test_bit(__E1000_DOWN, &adapter->state))) {
 679                         netif_wake_queue(netdev);
 680                         ++adapter->restart_queue;
 681                 }
 682         }
 683
 684         if (adapter->detect_tx_hung) {
 685                 /*
 686                  * Detect a transmit hang in hardware, this serializes the
 687                  * check with the clearing of time_stamp and movement of i
 688                  */
 689                 adapter->detect_tx_hung = 0;
 690                 if (tx_ring->buffer_info[i].time_stamp &&
 691                     time_after(jiffies, tx_ring->buffer_info[i].time_stamp
 692                                + (adapter->tx_timeout_factor * HZ)) &&
 693                     !(er32(STATUS) & E1000_STATUS_TXOFF)) {
 694                         schedule_work(&adapter->print_hang_task);
 695                         netif_stop_queue(netdev);
 696                 }
 697         }
 698         adapter->total_tx_bytes += total_tx_bytes;
 699         adapter->total_tx_packets += total_tx_packets;
 700         netdev->stats.tx_bytes += total_tx_bytes;
 701         netdev->stats.tx_packets += total_tx_packets;
 702         return (count < tx_ring->count);
 703 }
 704
 705 /**
 706  * e1000_clean_rx_irq_ps - Send received data up the network stack; packet split
 707  * @adapter: board private structure
 708  *
 709  * the return value indicates whether actual cleaning was done, there
 710  * is no guarantee that everything was cleaned
 711  **/
 712 static bool e1000_clean_rx_irq_ps(struct e1000_adapter *adapter,
 713                                   int *work_done, int work_to_do)
 714 {
 715         struct e1000_hw *hw = &adapter->hw;
 716         union e1000_rx_desc_packet_split *rx_desc, *next_rxd;
 717         struct net_device *netdev = adapter->netdev;
 718         struct pci_dev *pdev = adapter->pdev;
 719         struct e1000_ring *rx_ring = adapter->rx_ring;
 720         struct e1000_buffer *buffer_info, *next_buffer;
 721         struct e1000_ps_page *ps_page;
 722         struct sk_buff *skb;
 723         unsigned int i, j;
 724         u32 length, staterr;
 725         int cleaned_count = 0;
 726         bool cleaned = 0;
 727         unsigned int total_rx_bytes = 0, total_rx_packets = 0;
 728
 729         i = rx_ring->next_to_clean;
 730         rx_desc = E1000_RX_DESC_PS(*rx_ring, i);
 731         staterr = le32_to_cpu(rx_desc->wb.middle.status_error);
 732         buffer_info = &rx_ring->buffer_info[i];
 733
 734         while (staterr & E1000_RXD_STAT_DD) {
 735                 if (*work_done >= work_to_do)
 736                         break;
 737                 (*work_done)++;
 738                 skb = buffer_info->skb;
 739
 740                 /* in the packet split case this is header only */
 741                 prefetch(skb->data - NET_IP_ALIGN);
 742
 743                 i++;
 744                 if (i == rx_ring->count)
 745                         i = 0;
 746                 next_rxd = E1000_RX_DESC_PS(*rx_ring, i);
 747                 prefetch(next_rxd);
 748
 749                 next_buffer = &rx_ring->buffer_info[i];
 750
 751                 cleaned = 1;
 752                 cleaned_count++;
 753                 pci_unmap_single(pdev, buffer_info->dma,
 754                                  adapter->rx_ps_bsize0,
 755                                  PCI_DMA_FROMDEVICE);
 756                 buffer_info->dma = 0;
 757
 758                 /* see !EOP comment in other rx routine */
 759                 if (!(staterr & E1000_RXD_STAT_EOP))
 760                         adapter->flags2 |= FLAG2_IS_DISCARDING;
 761
 762                 if (adapter->flags2 & FLAG2_IS_DISCARDING) {
 763                         e_dbg("Packet Split buffers didn't pick up the full "
 764                               "packet\n");
 765                         dev_kfree_skb_irq(skb);
 766                         if (staterr & E1000_RXD_STAT_EOP)
 767                                 adapter->flags2 &= ~FLAG2_IS_DISCARDING;
 768                         goto next_desc;
 769                 }
 770
 771                 if (staterr & E1000_RXDEXT_ERR_FRAME_ERR_MASK) {
 772                         dev_kfree_skb_irq(skb);
 773                         goto next_desc;
 774                 }
 775
 776                 length = le16_to_cpu(rx_desc->wb.middle.length0);
 777
 778                 if (!length) {
 779                         e_dbg("Last part of the packet spanning multiple "
 780                               "descriptors\n");
 781                         dev_kfree_skb_irq(skb);
 782                         goto next_desc;
 783                 }
 784
 785                 /* Good Receive */
 786                 skb_put(skb, length);
 787
 788                 {
 789                 /*
 790                  * this looks ugly, but it seems compiler issues make it
 791                  * more efficient than reusing j
 792                  */
 793                 int l1 = le16_to_cpu(rx_desc->wb.upper.length[0]);
 794
 795                 /*
 796                  * page alloc/put takes too long and effects small packet
 797                  * throughput, so unsplit small packets and save the alloc/put
 798                  * only valid in softirq (napi) context to call kmap_*
 799                  */
 800                 if (l1 && (l1 <= copybreak) &&
 801                     ((length + l1) <= adapter->rx_ps_bsize0)) {
 802                         u8 *vaddr;
 803
 804                         ps_page = &buffer_info->ps_pages[0];
 805
 806                         /*
 807                          * there is no documentation about how to call
 808                          * kmap_atomic, so we can't hold the mapping
 809                          * very long
 810                          */
 811                         pci_dma_sync_single_for_cpu(pdev, ps_page->dma,
 812                                 PAGE_SIZE, PCI_DMA_FROMDEVICE);
 813                         vaddr = kmap_atomic(ps_page->page, KM_SKB_DATA_SOFTIRQ);
 814                         memcpy(skb_tail_pointer(skb), vaddr, l1);
 815                         kunmap_atomic(vaddr, KM_SKB_DATA_SOFTIRQ);
 816                         pci_dma_sync_single_for_device(pdev, ps_page->dma,
 817                                 PAGE_SIZE, PCI_DMA_FROMDEVICE);
 818
 819                         /* remove the CRC */
 820                         if (!(adapter->flags2 & FLAG2_CRC_STRIPPING))
 821                                 l1 -= 4;
 822
 823                         skb_put(skb, l1);
 824                         goto copydone;
 825                 } /* if */
 826                 }
 827
 828                 for (j = 0; j < PS_PAGE_BUFFERS; j++) {
 829                         length = le16_to_cpu(rx_desc->wb.upper.length[j]);
 830                         if (!length)
 831                                 break;
 832
 833                         ps_page = &buffer_info->ps_pages[j];
 834                         pci_unmap_page(pdev, ps_page->dma, PAGE_SIZE,
 835                                        PCI_DMA_FROMDEVICE);
 836                         ps_page->dma = 0;
 837                         skb_fill_page_desc(skb, j, ps_page->page, 0, length);
 838                         ps_page->page = NULL;
 839                         skb->len += length;
 840                         skb->data_len += length;
 841                         skb->truesize += length;
 842                 }
 843
 844                 /* strip the ethernet crc, problem is we're using pages now so
 845                  * this whole operation can get a little cpu intensive
 846                  */
 847                 if (!(adapter->flags2 & FLAG2_CRC_STRIPPING))
 848                         pskb_trim(skb, skb->len - 4);
 849
 850 copydone:
 851                 total_rx_bytes += skb->len;
 852                 total_rx_packets++;
 853
 854                 e1000_rx_checksum(adapter, staterr, le16_to_cpu(
 855                         rx_desc->wb.lower.hi_dword.csum_ip.csum), skb);
 856
 857                 if (rx_desc->wb.upper.header_status &
 858                            cpu_to_le16(E1000_RXDPS_HDRSTAT_HDRSP))
 859                         adapter->rx_hdr_split++;
 860
 861                 e1000_receive_skb(adapter, netdev, skb,
 862                                   staterr, rx_desc->wb.middle.vlan);
 863
 864 next_desc:
 865                 rx_desc->wb.middle.status_error &= cpu_to_le32(~0xFF);
 866                 buffer_info->skb = NULL;
 867
 868                 /* return some buffers to hardware, one at a time is too slow */
 869                 if (cleaned_count >= E1000_RX_BUFFER_WRITE) {
 870                         adapter->alloc_rx_buf(adapter, cleaned_count);
 871                         cleaned_count = 0;
 872                 }
 873
 874                 /* use prefetched values */
 875                 rx_desc = next_rxd;
 876                 buffer_info = next_buffer;
 877
 878                 staterr = le32_to_cpu(rx_desc->wb.middle.status_error);
 879         }
 880         rx_ring->next_to_clean = i;
 881
 882         cleaned_count = e1000_desc_unused(rx_ring);
 883         if (cleaned_count)
 884                 adapter->alloc_rx_buf(adapter, cleaned_count);
 885
 886         adapter->total_rx_bytes += total_rx_bytes;
 887         adapter->total_rx_packets += total_rx_packets;
 888         netdev->stats.rx_bytes += total_rx_bytes;
 889         netdev->stats.rx_packets += total_rx_packets;
 890         return cleaned;
 891 }
 892
 893 /**
 894  * e1000_consume_page - helper function
 895  **/
 896 static void e1000_consume_page(struct e1000_buffer *bi, struct sk_buff *skb,
 897                                u16 length)
 898 {
 899         bi->page = NULL;
 900         skb->len += length;
 901         skb->data_len += length;
 902         skb->truesize += length;
 903 }
 904
 905 /**
 906  * e1000_clean_jumbo_rx_irq - Send received data up the network stack; legacy
 907  * @adapter: board private structure
 908  *
 909  * the return value indicates whether actual cleaning was done, there
 910  * is no guarantee that everything was cleaned
 911  **/
 912
 913 static bool e1000_clean_jumbo_rx_irq(struct e1000_adapter *adapter,
 914                                      int *work_done, int work_to_do)
 915 {
 916         struct net_device *netdev = adapter->netdev;
 917         struct pci_dev *pdev = adapter->pdev;
 918         struct e1000_ring *rx_ring = adapter->rx_ring;
 919         struct e1000_rx_desc *rx_desc, *next_rxd;
 920         struct e1000_buffer *buffer_info, *next_buffer;
 921         u32 length;
 922         unsigned int i;
 923         int cleaned_count = 0;
 924         bool cleaned = false;
 925         unsigned int total_rx_bytes=0, total_rx_packets=0;
 926
 927         i = rx_ring->next_to_clean;
 928         rx_desc = E1000_RX_DESC(*rx_ring, i);
 929         buffer_info = &rx_ring->buffer_info[i];
 930
 931         while (rx_desc->status & E1000_RXD_STAT_DD) {
 932                 struct sk_buff *skb;
 933                 u8 status;
 934
 935                 if (*work_done >= work_to_do)
 936                         break;
 937                 (*work_done)++;
 938
 939                 status = rx_desc->status;
 940                 skb = buffer_info->skb;
 941                 buffer_info->skb = NULL;
 942
 943                 ++i;
 944                 if (i == rx_ring->count)
 945                         i = 0;
 946                 next_rxd = E1000_RX_DESC(*rx_ring, i);
 947                 prefetch(next_rxd);
 948
 949                 next_buffer = &rx_ring->buffer_info[i];
 950
 951                 cleaned = true;
 952                 cleaned_count++;
 953                 pci_unmap_page(pdev, buffer_info->dma, PAGE_SIZE,
 954                                PCI_DMA_FROMDEVICE);
 955                 buffer_info->dma = 0;
 956
 957                 length = le16_to_cpu(rx_desc->length);
 958
 959                 /* errors is only valid for DD + EOP descriptors */
 960                 if (unlikely((status & E1000_RXD_STAT_EOP) &&
 961                     (rx_desc->errors & E1000_RXD_ERR_FRAME_ERR_MASK))) {
 962                                 /* recycle both page and skb */
 963                                 buffer_info->skb = skb;
 964                                 /* an error means any chain goes out the window
 965                                  * too */
 966                                 if (rx_ring->rx_skb_top)
 967                                         dev_kfree_skb(rx_ring->rx_skb_top);
 968                                 rx_ring->rx_skb_top = NULL;
 969                                 goto next_desc;
 970                 }
 971
 972 #define rxtop rx_ring->rx_skb_top
 973                 if (!(status & E1000_RXD_STAT_EOP)) {
 974                         /* this descriptor is only the beginning (or middle) */
 975                         if (!rxtop) {
 976                                 /* this is the beginning of a chain */
 977                                 rxtop = skb;
 978                                 skb_fill_page_desc(rxtop, 0, buffer_info->page,
 979                                                    0, length);
 980                         } else {
 981                                 /* this is the middle of a chain */
 982                                 skb_fill_page_desc(rxtop,
 983                                     skb_shinfo(rxtop)->nr_frags,
 984                                     buffer_info->page, 0, length);
 985                                 /* re-use the skb, only consumed the page */
 986                                 buffer_info->skb = skb;
 987                         }
 988                         e1000_consume_page(buffer_info, rxtop, length);
 989                         goto next_desc;
 990                 } else {
 991                         if (rxtop) {
 992                                 /* end of the chain */
 993                                 skb_fill_page_desc(rxtop,
 994                                     skb_shinfo(rxtop)->nr_frags,
 995                                     buffer_info->page, 0, length);
 996                                 /* re-use the current skb, we only consumed the
 997                                  * page */
 998                                 buffer_info->skb = skb;
 999                                 skb = rxtop;
1000                                 rxtop = NULL;
1001                                 e1000_consume_page(buffer_info, skb, length);
1002                         } else {
1003                                 /* no chain, got EOP, this buf is the packet
1004                                  * copybreak to save the put_page/alloc_page */
1005                                 if (length <= copybreak &&
1006                                     skb_tailroom(skb) >= length) {
1007                                         u8 *vaddr;
1008                                         vaddr = kmap_atomic(buffer_info->page,
1009                                                            KM_SKB_DATA_SOFTIRQ);
1010                                         memcpy(skb_tail_pointer(skb), vaddr,
1011                                                length);
1012                                         kunmap_atomic(vaddr,
1013                                                       KM_SKB_DATA_SOFTIRQ);
1014                                         /* re-use the page, so don't erase
1015                                          * buffer_info->page */
1016                                         skb_put(skb, length);
1017                                 } else {
1018                                         skb_fill_page_desc(skb, 0,
1019                                                            buffer_info->page, 0,
1020                                                            length);
1021                                         e1000_consume_page(buffer_info, skb,
1022                                                            length);
1023                                 }
1024                         }
1025                 }
1026
1027                 /* Receive Checksum Offload XXX recompute due to CRC strip? */
1028                 e1000_rx_checksum(adapter,
1029                                   (u32)(status) |
1030                                   ((u32)(rx_desc->errors) << 24),
1031                                   le16_to_cpu(rx_desc->csum), skb);
1032
1033                 /* probably a little skewed due to removing CRC */
1034                 total_rx_bytes += skb->len;
1035                 total_rx_packets++;
1036
1037                 /* eth type trans needs skb->data to point to something */
1038                 if (!pskb_may_pull(skb, ETH_HLEN)) {
1039                         e_err("pskb_may_pull failed.\n");
1040                         dev_kfree_skb(skb);
1041                         goto next_desc;
1042                 }
1043
1044                 e1000_receive_skb(adapter, netdev, skb, status,
1045                                   rx_desc->special);
1046
1047 next_desc:
1048                 rx_desc->status = 0;
1049
1050                 /* return some buffers to hardware, one at a time is too slow */
1051                 if (unlikely(cleaned_count >= E1000_RX_BUFFER_WRITE)) {
1052                         adapter->alloc_rx_buf(adapter, cleaned_count);
1053                         cleaned_count = 0;
1054                 }
1055
1056                 /* use prefetched values */
1057                 rx_desc = next_rxd;
1058                 buffer_info = next_buffer;
1059         }
1060         rx_ring->next_to_clean = i;
1061
1062         cleaned_count = e1000_desc_unused(rx_ring);
1063         if (cleaned_count)
1064                 adapter->alloc_rx_buf(adapter, cleaned_count);
1065
1066         adapter->total_rx_bytes += total_rx_bytes;
1067         adapter->total_rx_packets += total_rx_packets;
1068         netdev->stats.rx_bytes += total_rx_bytes;
1069         netdev->stats.rx_packets += total_rx_packets;
1070         return cleaned;
1071 }
1072
1073 /**
1074  * e1000_clean_rx_ring - Free Rx Buffers per Queue
1075  * @adapter: board private structure
1076  **/
1077 static void e1000_clean_rx_ring(struct e1000_adapter *adapter)
1078 {
1079         struct e1000_ring *rx_ring = adapter->rx_ring;
1080         struct e1000_buffer *buffer_info;
1081         struct e1000_ps_page *ps_page;
1082         struct pci_dev *pdev = adapter->pdev;
1083         unsigned int i, j;
1084
1085         /* Free all the Rx ring sk_buffs */
1086         for (i = 0; i < rx_ring->count; i++) {
1087                 buffer_info = &rx_ring->buffer_info[i];
1088                 if (buffer_info->dma) {
1089                         if (adapter->clean_rx == e1000_clean_rx_irq)
1090                                 pci_unmap_single(pdev, buffer_info->dma,
1091                                                  adapter->rx_buffer_len,
1092                                                  PCI_DMA_FROMDEVICE);
1093                         else if (adapter->clean_rx == e1000_clean_jumbo_rx_irq)
1094                                 pci_unmap_page(pdev, buffer_info->dma,
1095                                                PAGE_SIZE,
1096                                                PCI_DMA_FROMDEVICE);
1097                         else if (adapter->clean_rx == e1000_clean_rx_irq_ps)
1098                                 pci_unmap_single(pdev, buffer_info->dma,
1099                                                  adapter->rx_ps_bsize0,
1100                                                  PCI_DMA_FROMDEVICE);
1101                         buffer_info->dma = 0;
1102                 }
1103
1104                 if (buffer_info->page) {
1105                         put_page(buffer_info->page);
1106                         buffer_info->page = NULL;
1107                 }
1108
1109                 if (buffer_info->skb) {
1110                         dev_kfree_skb(buffer_info->skb);
1111                         buffer_info->skb = NULL;
1112                 }
1113
1114                 for (j = 0; j < PS_PAGE_BUFFERS; j++) {
1115                         ps_page = &buffer_info->ps_pages[j];
1116                         if (!ps_page->page)
1117                                 break;
1118                         pci_unmap_page(pdev, ps_page->dma, PAGE_SIZE,
1119                                        PCI_DMA_FROMDEVICE);
1120                         ps_page->dma = 0;
1121                         put_page(ps_page->page);
1122                         ps_page->page = NULL;
1123                 }
1124         }
1125
1126         /* there also may be some cached data from a chained receive */
1127         if (rx_ring->rx_skb_top) {
1128                 dev_kfree_skb(rx_ring->rx_skb_top);
1129                 rx_ring->rx_skb_top = NULL;
1130         }
1131
1132         /* Zero out the descriptor ring */
1133         memset(rx_ring->desc, 0, rx_ring->size);
1134
1135         rx_ring->next_to_clean = 0;
1136         rx_ring->next_to_use = 0;
1137         adapter->flags2 &= ~FLAG2_IS_DISCARDING;
1138
1139         writel(0, adapter->hw.hw_addr + rx_ring->head);
1140         writel(0, adapter->hw.hw_addr + rx_ring->tail);
1141 }
1142
1143 static void e1000e_downshift_workaround(struct work_struct *work)
1144 {
1145         struct e1000_adapter *adapter = container_of(work,
1146                                         struct e1000_adapter, downshift_task);
1147
1148         e1000e_gig_downshift_workaround_ich8lan(&adapter->hw);
1149 }
1150
1151 /**
1152  * e1000_intr_msi - Interrupt Handler
1153  * @irq: interrupt number
1154  * @data: pointer to a network interface device structure
1155  **/
1156 static irqreturn_t e1000_intr_msi(int irq, void *data)
1157 {
1158         struct net_device *netdev = data;
1159         struct e1000_adapter *adapter = netdev_priv(netdev);
1160         struct e1000_hw *hw = &adapter->hw;
1161         u32 icr = er32(ICR);
1162
1163         /*
1164          * read ICR disables interrupts using IAM
1165          */
1166
1167         if (icr & E1000_ICR_LSC) {
1168                 hw->mac.get_link_status = 1;
1169                 /*
1170                  * ICH8 workaround-- Call gig speed drop workaround on cable
1171                  * disconnect (LSC) before accessing any PHY registers
1172                  */
1173                 if ((adapter->flags & FLAG_LSC_GIG_SPEED_DROP) &&
1174                     (!(er32(STATUS) & E1000_STATUS_LU)))
1175                         schedule_work(&adapter->downshift_task);
1176
1177                 /*
1178                  * 80003ES2LAN workaround-- For packet buffer work-around on
1179                  * link down event; disable receives here in the ISR and reset
1180                  * adapter in watchdog
1181                  */
1182                 if (netif_carrier_ok(netdev) &&
1183                     adapter->flags & FLAG_RX_NEEDS_RESTART) {
1184                         /* disable receives */
1185                         u32 rctl = er32(RCTL);
1186                         ew32(RCTL, rctl & ~E1000_RCTL_EN);
1187                         adapter->flags |= FLAG_RX_RESTART_NOW;
1188                 }
1189                 /* guard against interrupt when we're going down */
1190                 if (!test_bit(__E1000_DOWN, &adapter->state))
1191                         mod_timer(&adapter->watchdog_timer, jiffies + 1);
1192         }
1193
1194         if (napi_schedule_prep(&adapter->napi)) {
1195                 adapter->total_tx_bytes = 0;
1196                 adapter->total_tx_packets = 0;
1197                 adapter->total_rx_bytes = 0;
1198                 adapter->total_rx_packets = 0;
1199                 __napi_schedule(&adapter->napi);
1200         }
1201
1202         return IRQ_HANDLED;
1203 }
1204
1205 /**
1206  * e1000_intr - Interrupt Handler
1207  * @irq: interrupt number
1208  * @data: pointer to a network interface device structure
1209  **/
1210 static irqreturn_t e1000_intr(int irq, void *data)
1211 {
1212         struct net_device *netdev = data;
1213         struct e1000_adapter *adapter = netdev_priv(netdev);
1214         struct e1000_hw *hw = &adapter->hw;
1215         u32 rctl, icr = er32(ICR);
1216
1217         if (!icr || test_bit(__E1000_DOWN, &adapter->state))
1218                 return IRQ_NONE;  /* Not our interrupt */
1219
1220         /*
1221          * IMS will not auto-mask if INT_ASSERTED is not set, and if it is
1222          * not set, then the adapter didn't send an interrupt
1223          */
1224         if (!(icr & E1000_ICR_INT_ASSERTED))
1225                 return IRQ_NONE;
1226
1227         /*
1228          * Interrupt Auto-Mask...upon reading ICR,
1229          * interrupts are masked.  No need for the
1230          * IMC write
1231          */
1232
1233         if (icr & E1000_ICR_LSC) {
1234                 hw->mac.get_link_status = 1;
1235                 /*
1236                  * ICH8 workaround-- Call gig speed drop workaround on cable
1237                  * disconnect (LSC) before accessing any PHY registers
1238                  */
1239                 if ((adapter->flags & FLAG_LSC_GIG_SPEED_DROP) &&
1240                     (!(er32(STATUS) & E1000_STATUS_LU)))
1241                         schedule_work(&adapter->downshift_task);
1242
1243                 /*
1244                  * 80003ES2LAN workaround--
1245                  * For packet buffer work-around on link down event;
1246                  * disable receives here in the ISR and
1247                  * reset adapter in watchdog
1248                  */
1249                 if (netif_carrier_ok(netdev) &&
1250                     (adapter->flags & FLAG_RX_NEEDS_RESTART)) {
1251                         /* disable receives */
1252                         rctl = er32(RCTL);
1253                         ew32(RCTL, rctl & ~E1000_RCTL_EN);
1254                         adapter->flags |= FLAG_RX_RESTART_NOW;
1255                 }
1256                 /* guard against interrupt when we're going down */
1257                 if (!test_bit(__E1000_DOWN, &adapter->state))
1258                         mod_timer(&adapter->watchdog_timer, jiffies + 1);
1259         }
1260
1261         if (napi_schedule_prep(&adapter->napi)) {
1262                 adapter->total_tx_bytes = 0;
1263                 adapter->total_tx_packets = 0;
1264                 adapter->total_rx_bytes = 0;
1265                 adapter->total_rx_packets = 0;
1266                 __napi_schedule(&adapter->napi);
1267         }
1268
1269         return IRQ_HANDLED;
1270 }
1271
1272 static irqreturn_t e1000_msix_other(int irq, void *data)
1273 {
1274         struct net_device *netdev = data;
1275         struct e1000_adapter *adapter = netdev_priv(netdev);
1276         struct e1000_hw *hw = &adapter->hw;
1277         u32 icr = er32(ICR);
1278
1279         if (!(icr & E1000_ICR_INT_ASSERTED)) {
1280                 if (!test_bit(__E1000_DOWN, &adapter->state))
1281                         ew32(IMS, E1000_IMS_OTHER);
1282                 return IRQ_NONE;
1283         }
1284
1285         if (icr & adapter->eiac_mask)
1286                 ew32(ICS, (icr & adapter->eiac_mask));
1287
1288         if (icr & E1000_ICR_OTHER) {
1289                 if (!(icr & E1000_ICR_LSC))
1290                         goto no_link_interrupt;
1291                 hw->mac.get_link_status = 1;
1292                 /* guard against interrupt when we're going down */
1293                 if (!test_bit(__E1000_DOWN, &adapter->state))
1294                         mod_timer(&adapter->watchdog_timer, jiffies + 1);
1295         }
1296
1297 no_link_interrupt:
1298         if (!test_bit(__E1000_DOWN, &adapter->state))
1299                 ew32(IMS, E1000_IMS_LSC | E1000_IMS_OTHER);
1300
1301         return IRQ_HANDLED;
1302 }
1303
1304
1305 static irqreturn_t e1000_intr_msix_tx(int irq, void *data)
1306 {
1307         struct net_device *netdev = data;
1308         struct e1000_adapter *adapter = netdev_priv(netdev);
1309         struct e1000_hw *hw = &adapter->hw;
1310         struct e1000_ring *tx_ring = adapter->tx_ring;
1311
1312
1313         adapter->total_tx_bytes = 0;
1314         adapter->total_tx_packets = 0;
1315
1316         if (!e1000_clean_tx_irq(adapter))
1317                 /* Ring was not completely cleaned, so fire another interrupt */
1318                 ew32(ICS, tx_ring->ims_val);
1319
1320         return IRQ_HANDLED;
1321 }
1322
1323 static irqreturn_t e1000_intr_msix_rx(int irq, void *data)
1324 {
1325         struct net_device *netdev = data;
1326         struct e1000_adapter *adapter = netdev_priv(netdev);
1327
1328         /* Write the ITR value calculated at the end of the
1329          * previous interrupt.
1330          */
1331         if (adapter->rx_ring->set_itr) {
1332                 writel(1000000000 / (adapter->rx_ring->itr_val * 256),
1333                        adapter->hw.hw_addr + adapter->rx_ring->itr_register);
1334                 adapter->rx_ring->set_itr = 0;
1335         }
1336
1337         if (napi_schedule_prep(&adapter->napi)) {
1338                 adapter->total_rx_bytes = 0;
1339                 adapter->total_rx_packets = 0;
1340                 __napi_schedule(&adapter->napi);
1341         }
1342         return IRQ_HANDLED;
1343 }
1344
1345 /**
1346  * e1000_configure_msix - Configure MSI-X hardware
1347  *
1348  * e1000_configure_msix sets up the hardware to properly
1349  * generate MSI-X interrupts.
1350  **/
1351 static void e1000_configure_msix(struct e1000_adapter *adapter)
1352 {
1353         struct e1000_hw *hw = &adapter->hw;
1354         struct e1000_ring *rx_ring = adapter->rx_ring;
1355         struct e1000_ring *tx_ring = adapter->tx_ring;
1356         int vector = 0;
1357         u32 ctrl_ext, ivar = 0;
1358
1359         adapter->eiac_mask = 0;
1360
1361         /* Workaround issue with spurious interrupts on 82574 in MSI-X mode */
1362         if (hw->mac.type == e1000_82574) {
1363                 u32 rfctl = er32(RFCTL);
1364                 rfctl |= E1000_RFCTL_ACK_DIS;
1365                 ew32(RFCTL, rfctl);
1366         }
1367
1368 #define E1000_IVAR_INT_ALLOC_VALID      0x8
1369         /* Configure Rx vector */
1370         rx_ring->ims_val = E1000_IMS_RXQ0;
1371         adapter->eiac_mask |= rx_ring->ims_val;
1372         if (rx_ring->itr_val)
1373                 writel(1000000000 / (rx_ring->itr_val * 256),
1374                        hw->hw_addr + rx_ring->itr_register);
1375         else
1376                 writel(1, hw->hw_addr + rx_ring->itr_register);
1377         ivar = E1000_IVAR_INT_ALLOC_VALID | vector;
1378
1379         /* Configure Tx vector */
1380         tx_ring->ims_val = E1000_IMS_TXQ0;
1381         vector++;
1382         if (tx_ring->itr_val)
1383                 writel(1000000000 / (tx_ring->itr_val * 256),
1384                        hw->hw_addr + tx_ring->itr_register);
1385         else
1386                 writel(1, hw->hw_addr + tx_ring->itr_register);
1387         adapter->eiac_mask |= tx_ring->ims_val;
1388         ivar |= ((E1000_IVAR_INT_ALLOC_VALID | vector) << 8);
1389
1390         /* set vector for Other Causes, e.g. link changes */
1391         vector++;
1392         ivar |= ((E1000_IVAR_INT_ALLOC_VALID | vector) << 16);
1393         if (rx_ring->itr_val)
1394                 writel(1000000000 / (rx_ring->itr_val * 256),
1395                        hw->hw_addr + E1000_EITR_82574(vector));
1396         else
1397                 writel(1, hw->hw_addr + E1000_EITR_82574(vector));
1398
1399         /* Cause Tx interrupts on every write back */
1400         ivar |= (1 << 31);
1401
1402         ew32(IVAR, ivar);
1403
1404         /* enable MSI-X PBA support */
1405         ctrl_ext = er32(CTRL_EXT);
1406         ctrl_ext |= E1000_CTRL_EXT_PBA_CLR;
1407
1408         /* Auto-Mask Other interrupts upon ICR read */
1409 #define E1000_EIAC_MASK_82574   0x01F00000
1410         ew32(IAM, ~E1000_EIAC_MASK_82574 | E1000_IMS_OTHER);
1411         ctrl_ext |= E1000_CTRL_EXT_EIAME;
1412         ew32(CTRL_EXT, ctrl_ext);
1413         e1e_flush();
1414 }
1415
1416 void e1000e_reset_interrupt_capability(struct e1000_adapter *adapter)
1417 {
1418         if (adapter->msix_entries) {
1419                 pci_disable_msix(adapter->pdev);
1420                 kfree(adapter->msix_entries);
1421                 adapter->msix_entries = NULL;
1422         } else if (adapter->flags & FLAG_MSI_ENABLED) {
1423                 pci_disable_msi(adapter->pdev);
1424                 adapter->flags &= ~FLAG_MSI_ENABLED;
1425         }
1426
1427         return;
1428 }
1429
1430 /**
1431  * e1000e_set_interrupt_capability - set MSI or MSI-X if supported
1432  *
1433  * Attempt to configure interrupts using the best available
1434  * capabilities of the hardware and kernel.
1435  **/
1436 void e1000e_set_interrupt_capability(struct e1000_adapter *adapter)
1437 {
1438         int err;
1439         int numvecs, i;
1440
1441
1442         switch (adapter->int_mode) {
1443         case E1000E_INT_MODE_MSIX:
1444                 if (adapter->flags & FLAG_HAS_MSIX) {
1445                         numvecs = 3; /* RxQ0, TxQ0 and other */
1446                         adapter->msix_entries = kcalloc(numvecs,
1447                                                       sizeof(struct msix_entry),
1448                                                       GFP_KERNEL);
1449                         if (adapter->msix_entries) {
1450                                 for (i = 0; i < numvecs; i++)
1451                                         adapter->msix_entries[i].entry = i;
1452
1453                                 err = pci_enable_msix(adapter->pdev,
1454                                                       adapter->msix_entries,
1455                                                       numvecs);
1456                                 if (err == 0)
1457                                         return;
1458                         }
1459                         /* MSI-X failed, so fall through and try MSI */
1460                         e_err("Failed to initialize MSI-X interrupts.  "
1461                               "Falling back to MSI interrupts.\n");
1462                         e1000e_reset_interrupt_capability(adapter);
1463                 }
1464                 adapter->int_mode = E1000E_INT_MODE_MSI;
1465                 /* Fall through */
1466         case E1000E_INT_MODE_MSI:
1467                 if (!pci_enable_msi(adapter->pdev)) {
1468                         adapter->flags |= FLAG_MSI_ENABLED;
1469                 } else {
1470                         adapter->int_mode = E1000E_INT_MODE_LEGACY;
1471                         e_err("Failed to initialize MSI interrupts.  Falling "
1472                               "back to legacy interrupts.\n");
1473                 }
1474                 /* Fall through */
1475         case E1000E_INT_MODE_LEGACY:
1476                 /* Don't do anything; this is the system default */
1477                 break;
1478         }
1479
1480         return;
1481 }
1482
1483 /**
1484  * e1000_request_msix - Initialize MSI-X interrupts
1485  *
1486  * e1000_request_msix allocates MSI-X vectors and requests interrupts from the
1487  * kernel.
1488  **/
1489 static int e1000_request_msix(struct e1000_adapter *adapter)
1490 {
1491         struct net_device *netdev = adapter->netdev;
1492         int err = 0, vector = 0;
1493
1494         if (strlen(netdev->name) < (IFNAMSIZ - 5))
1495                 sprintf(adapter->rx_ring->name, "%s-rx-0", netdev->name);
1496         else
1497                 memcpy(adapter->rx_ring->name, netdev->name, IFNAMSIZ);
1498         err = request_irq(adapter->msix_entries[vector].vector,
1499                           e1000_intr_msix_rx, 0, adapter->rx_ring->name,
1500                           netdev);
1501         if (err)
1502                 goto out;
1503         adapter->rx_ring->itr_register = E1000_EITR_82574(vector);
1504         adapter->rx_ring->itr_val = adapter->itr;
1505         vector++;
1506
1507         if (strlen(netdev->name) < (IFNAMSIZ - 5))
1508                 sprintf(adapter->tx_ring->name, "%s-tx-0", netdev->name);
1509         else
1510                 memcpy(adapter->tx_ring->name, netdev->name, IFNAMSIZ);
1511         err = request_irq(adapter->msix_entries[vector].vector,
1512                           e1000_intr_msix_tx, 0, adapter->tx_ring->name,
1513                           netdev);
1514         if (err)
1515                 goto out;
1516         adapter->tx_ring->itr_register = E1000_EITR_82574(vector);
1517         adapter->tx_ring->itr_val = adapter->itr;
1518         vector++;
1519
1520         err = request_irq(adapter->msix_entries[vector].vector,
1521                           e1000_msix_other, 0, netdev->name, netdev);
1522         if (err)
1523                 goto out;
1524
1525         e1000_configure_msix(adapter);
1526         return 0;
1527 out:
1528         return err;
1529 }
1530
1531 /**
1532  * e1000_request_irq - initialize interrupts
1533  *
1534  * Attempts to configure interrupts using the best available
1535  * capabilities of the hardware and kernel.
1536  **/
1537 static int e1000_request_irq(struct e1000_adapter *adapter)
1538 {
1539         struct net_device *netdev = adapter->netdev;
1540         int err;
1541
1542         if (adapter->msix_entries) {
1543                 err = e1000_request_msix(adapter);
1544                 if (!err)
1545                         return err;
1546                 /* fall back to MSI */
1547                 e1000e_reset_interrupt_capability(adapter);
1548                 adapter->int_mode = E1000E_INT_MODE_MSI;
1549                 e1000e_set_interrupt_capability(adapter);
1550         }
1551         if (adapter->flags & FLAG_MSI_ENABLED) {
1552                 err = request_irq(adapter->pdev->irq, e1000_intr_msi, 0,
1553                                   netdev->name, netdev);
1554                 if (!err)
1555                         return err;
1556
1557                 /* fall back to legacy interrupt */
1558                 e1000e_reset_interrupt_capability(adapter);
1559                 adapter->int_mode = E1000E_INT_MODE_LEGACY;
1560         }
1561
1562         err = request_irq(adapter->pdev->irq, e1000_intr, IRQF_SHARED,
1563                           netdev->name, netdev);
1564         if (err)
1565                 e_err("Unable to allocate interrupt, Error: %d\n", err);
1566
1567         return err;
1568 }
1569
1570 static void e1000_free_irq(struct e1000_adapter *adapter)
1571 {
1572         struct net_device *netdev = adapter->netdev;
1573
1574         if (adapter->msix_entries) {
1575                 int vector = 0;
1576
1577                 free_irq(adapter->msix_entries[vector].vector, netdev);
1578                 vector++;
1579
1580                 free_irq(adapter->msix_entries[vector].vector, netdev);
1581                 vector++;
1582
1583                 /* Other Causes interrupt vector */
1584                 free_irq(adapter->msix_entries[vector].vector, netdev);
1585                 return;
1586         }
1587
1588         free_irq(adapter->pdev->irq, netdev);
1589 }
1590
1591 /**
1592  * e1000_irq_disable - Mask off interrupt generation on the NIC
1593  **/
1594 static void e1000_irq_disable(struct e1000_adapter *adapter)
1595 {
1596         struct e1000_hw *hw = &adapter->hw;
1597
1598         ew32(IMC, ~0);
1599         if (adapter->msix_entries)
1600                 ew32(EIAC_82574, 0);
1601         e1e_flush();
1602         synchronize_irq(adapter->pdev->irq);
1603 }
1604
1605 /**
1606  * e1000_irq_enable - Enable default interrupt generation settings
1607  **/
1608 static void e1000_irq_enable(struct e1000_adapter *adapter)
1609 {
1610         struct e1000_hw *hw = &adapter->hw;
1611
1612         if (adapter->msix_entries) {
1613                 ew32(EIAC_82574, adapter->eiac_mask & E1000_EIAC_MASK_82574);
1614                 ew32(IMS, adapter->eiac_mask | E1000_IMS_OTHER | E1000_IMS_LSC);
1615         } else {
1616                 ew32(IMS, IMS_ENABLE_MASK);
1617         }
1618         e1e_flush();
1619 }
1620
1621 /**
1622  * e1000_get_hw_control - get control of the h/w from f/w
1623  * @adapter: address of board private structure
1624  *
1625  * e1000_get_hw_control sets {CTRL_EXT|SWSM}:DRV_LOAD bit.
1626  * For ASF and Pass Through versions of f/w this means that
1627  * the driver is loaded. For AMT version (only with 82573)
1628  * of the f/w this means that the network i/f is open.
1629  **/
1630 static void e1000_get_hw_control(struct e1000_adapter *adapter)
1631 {
1632         struct e1000_hw *hw = &adapter->hw;
1633         u32 ctrl_ext;
1634         u32 swsm;
1635
1636         /* Let firmware know the driver has taken over */
1637         if (adapter->flags & FLAG_HAS_SWSM_ON_LOAD) {
1638                 swsm = er32(SWSM);
1639                 ew32(SWSM, swsm | E1000_SWSM_DRV_LOAD);
1640         } else if (adapter->flags & FLAG_HAS_CTRLEXT_ON_LOAD) {
1641                 ctrl_ext = er32(CTRL_EXT);
1642                 ew32(CTRL_EXT, ctrl_ext | E1000_CTRL_EXT_DRV_LOAD);
1643         }
1644 }
1645
1646 /**
1647  * e1000_release_hw_control - release control of the h/w to f/w
1648  * @adapter: address of board private structure
1649  *
1650  * e1000_release_hw_control resets {CTRL_EXT|SWSM}:DRV_LOAD bit.
1651  * For ASF and Pass Through versions of f/w this means that the
1652  * driver is no longer loaded. For AMT version (only with 82573) i
1653  * of the f/w this means that the network i/f is closed.
1654  *
1655  **/
1656 static void e1000_release_hw_control(struct e1000_adapter *adapter)
1657 {
1658         struct e1000_hw *hw = &adapter->hw;
1659         u32 ctrl_ext;
1660         u32 swsm;
1661
1662         /* Let firmware taken over control of h/w */
1663         if (adapter->flags & FLAG_HAS_SWSM_ON_LOAD) {
1664                 swsm = er32(SWSM);
1665                 ew32(SWSM, swsm & ~E1000_SWSM_DRV_LOAD);
1666         } else if (adapter->flags & FLAG_HAS_CTRLEXT_ON_LOAD) {
1667                 ctrl_ext = er32(CTRL_EXT);
1668                 ew32(CTRL_EXT, ctrl_ext & ~E1000_CTRL_EXT_DRV_LOAD);
1669         }
1670 }
1671
1672 /**
1673  * @e1000_alloc_ring - allocate memory for a ring structure
1674  **/
1675 static int e1000_alloc_ring_dma(struct e1000_adapter *adapter,
1676                                 struct e1000_ring *ring)
1677 {
1678         struct pci_dev *pdev = adapter->pdev;
1679
1680         ring->desc = dma_alloc_coherent(&pdev->dev, ring->size, &ring->dma,
1681                                         GFP_KERNEL);
1682         if (!ring->desc)
1683                 return -ENOMEM;
1684
1685         return 0;
1686 }
1687
1688 /**
1689  * e1000e_setup_tx_resources - allocate Tx resources (Descriptors)
1690  * @adapter: board private structure
1691  *
1692  * Return 0 on success, negative on failure
1693  **/
1694 int e1000e_setup_tx_resources(struct e1000_adapter *adapter)
1695 {
1696         struct e1000_ring *tx_ring = adapter->tx_ring;
1697         int err = -ENOMEM, size;
1698
1699         size = sizeof(struct e1000_buffer) * tx_ring->count;
1700         tx_ring->buffer_info = vmalloc(size);
1701         if (!tx_ring->buffer_info)
1702                 goto err;
1703         memset(tx_ring->buffer_info, 0, size);
1704
1705         /* round up to nearest 4K */
1706         tx_ring->size = tx_ring->count * sizeof(struct e1000_tx_desc);
1707         tx_ring->size = ALIGN(tx_ring->size, 4096);
1708
1709         err = e1000_alloc_ring_dma(adapter, tx_ring);
1710         if (err)
1711                 goto err;
1712
1713         tx_ring->next_to_use = 0;
1714         tx_ring->next_to_clean = 0;
1715
1716         return 0;
1717 err:
1718         vfree(tx_ring->buffer_info);
1719         e_err("Unable to allocate memory for the transmit descriptor ring\n");
1720         return err;
1721 }
1722
1723 /**
1724  * e1000e_setup_rx_resources - allocate Rx resources (Descriptors)
1725  * @adapter: board private structure
1726  *
1727  * Returns 0 on success, negative on failure
1728  **/
1729 int e1000e_setup_rx_resources(struct e1000_adapter *adapter)
1730 {
1731         struct e1000_ring *rx_ring = adapter->rx_ring;
1732         struct e1000_buffer *buffer_info;
1733         int i, size, desc_len, err = -ENOMEM;
1734
1735         size = sizeof(struct e1000_buffer) * rx_ring->count;
1736         rx_ring->buffer_info = vmalloc(size);
1737         if (!rx_ring->buffer_info)
1738                 goto err;
1739         memset(rx_ring->buffer_info, 0, size);
1740
1741         for (i = 0; i < rx_ring->count; i++) {
1742                 buffer_info = &rx_ring->buffer_info[i];
1743                 buffer_info->ps_pages = kcalloc(PS_PAGE_BUFFERS,
1744                                                 sizeof(struct e1000_ps_page),
1745                                                 GFP_KERNEL);
1746                 if (!buffer_info->ps_pages)
1747                         goto err_pages;
1748         }
1749
1750         desc_len = sizeof(union e1000_rx_desc_packet_split);
1751
1752         /* Round up to nearest 4K */
1753         rx_ring->size = rx_ring->count * desc_len;
1754         rx_ring->size = ALIGN(rx_ring->size, 4096);
1755
1756         err = e1000_alloc_ring_dma(adapter, rx_ring);
1757         if (err)
1758                 goto err_pages;
1759
1760         rx_ring->next_to_clean = 0;
1761         rx_ring->next_to_use = 0;
1762         rx_ring->rx_skb_top = NULL;
1763
1764         return 0;
1765
1766 err_pages:
1767         for (i = 0; i < rx_ring->count; i++) {
1768                 buffer_info = &rx_ring->buffer_info[i];
1769                 kfree(buffer_info->ps_pages);
1770         }
1771 err:
1772         vfree(rx_ring->buffer_info);
1773         e_err("Unable to allocate memory for the transmit descriptor ring\n");
1774         return err;
1775 }
1776
1777 /**
1778  * e1000_clean_tx_ring - Free Tx Buffers
1779  * @adapter: board private structure
1780  **/
1781 static void e1000_clean_tx_ring(struct e1000_adapter *adapter)
1782 {
1783         struct e1000_ring *tx_ring = adapter->tx_ring;
1784         struct e1000_buffer *buffer_info;
1785         unsigned long size;
1786         unsigned int i;
1787
1788         for (i = 0; i < tx_ring->count; i++) {
1789                 buffer_info = &tx_ring->buffer_info[i];
1790                 e1000_put_txbuf(adapter, buffer_info);
1791         }
1792
1793         size = sizeof(struct e1000_buffer) * tx_ring->count;
1794         memset(tx_ring->buffer_info, 0, size);
1795
1796         memset(tx_ring->desc, 0, tx_ring->size);
1797
1798         tx_ring->next_to_use = 0;
1799         tx_ring->next_to_clean = 0;
1800
1801         writel(0, adapter->hw.hw_addr + tx_ring->head);
1802         writel(0, adapter->hw.hw_addr + tx_ring->tail);
1803 }
1804
1805 /**
1806  * e1000e_free_tx_resources - Free Tx Resources per Queue
1807  * @adapter: board private structure
1808  *
1809  * Free all transmit software resources
1810  **/
1811 void e1000e_free_tx_resources(struct e1000_adapter *adapter)
1812 {
1813         struct pci_dev *pdev = adapter->pdev;
1814         struct e1000_ring *tx_ring = adapter->tx_ring;
1815
1816         e1000_clean_tx_ring(adapter);
1817
1818         vfree(tx_ring->buffer_info);
1819         tx_ring->buffer_info = NULL;
1820
1821         dma_free_coherent(&pdev->dev, tx_ring->size, tx_ring->desc,
1822                           tx_ring->dma);
1823         tx_ring->desc = NULL;
1824 }
1825
1826 /**
1827  * e1000e_free_rx_resources - Free Rx Resources
1828  * @adapter: board private structure
1829  *
1830  * Free all receive software resources
1831  **/
1832
1833 void e1000e_free_rx_resources(struct e1000_adapter *adapter)
1834 {
1835         struct pci_dev *pdev = adapter->pdev;
1836         struct e1000_ring *rx_ring = adapter->rx_ring;
1837         int i;
1838
1839         e1000_clean_rx_ring(adapter);
1840
1841         for (i = 0; i < rx_ring->count; i++) {
1842                 kfree(rx_ring->buffer_info[i].ps_pages);
1843         }
1844
1845         vfree(rx_ring->buffer_info);
1846         rx_ring->buffer_info = NULL;
1847
1848         dma_free_coherent(&pdev->dev, rx_ring->size, rx_ring->desc,
1849                           rx_ring->dma);
1850         rx_ring->desc = NULL;
1851 }
1852
1853 /**
1854  * e1000_update_itr - update the dynamic ITR value based on statistics
1855  * @adapter: pointer to adapter
1856  * @itr_setting: current adapter->itr
1857  * @packets: the number of packets during this measurement interval
1858  * @bytes: the number of bytes during this measurement interval
1859  *
1860  *      Stores a new ITR value based on packets and byte
1861  *      counts during the last interrupt.  The advantage of per interrupt
1862  *      computation is faster updates and more accurate ITR for the current
1863  *      traffic pattern.  Constants in this function were computed
1864  *      based on theoretical maximum wire speed and thresholds were set based
1865  *      on testing data as well as attempting to minimize response time
1866  *      while increasing bulk throughput.  This functionality is controlled
1867  *      by the InterruptThrottleRate module parameter.
1868  **/
1869 static unsigned int e1000_update_itr(struct e1000_adapter *adapter,
1870                                      u16 itr_setting, int packets,
1871                                      int bytes)
1872 {
1873         unsigned int retval = itr_setting;
1874
1875         if (packets == 0)
1876                 goto update_itr_done;
1877
1878         switch (itr_setting) {
1879         case lowest_latency:
1880                 /* handle TSO and jumbo frames */
1881                 if (bytes/packets > 8000)
1882                         retval = bulk_latency;
1883                 else if ((packets < 5) && (bytes > 512)) {
1884                         retval = low_latency;
1885                 }
1886                 break;
1887         case low_latency:  /* 50 usec aka 20000 ints/s */
1888                 if (bytes > 10000) {
1889                         /* this if handles the TSO accounting */
1890                         if (bytes/packets > 8000) {
1891                                 retval = bulk_latency;
1892                         } else if ((packets < 10) || ((bytes/packets) > 1200)) {
1893                                 retval = bulk_latency;
1894                         } else if ((packets > 35)) {
1895                                 retval = lowest_latency;
1896                         }
1897                 } else if (bytes/packets > 2000) {
1898                         retval = bulk_latency;
1899                 } else if (packets <= 2 && bytes < 512) {
1900                         retval = lowest_latency;
1901                 }
1902                 break;
1903         case bulk_latency: /* 250 usec aka 4000 ints/s */
1904                 if (bytes > 25000) {
1905                         if (packets > 35) {
1906                                 retval = low_latency;
1907                         }
1908                 } else if (bytes < 6000) {
1909                         retval = low_latency;
1910                 }
1911                 break;
1912         }
1913
1914 update_itr_done:
1915         return retval;
1916 }
1917
1918 static void e1000_set_itr(struct e1000_adapter *adapter)
1919 {
1920         struct e1000_hw *hw = &adapter->hw;
1921         u16 current_itr;
1922         u32 new_itr = adapter->itr;
1923
1924         /* for non-gigabit speeds, just fix the interrupt rate at 4000 */
1925         if (adapter->link_speed != SPEED_1000) {
1926                 current_itr = 0;
1927                 new_itr = 4000;
1928                 goto set_itr_now;
1929         }
1930
1931         adapter->tx_itr = e1000_update_itr(adapter,
1932                                     adapter->tx_itr,
1933                                     adapter->total_tx_packets,
1934                                     adapter->total_tx_bytes);
1935         /* conservative mode (itr 3) eliminates the lowest_latency setting */
1936         if (adapter->itr_setting == 3 && adapter->tx_itr == lowest_latency)
1937                 adapter->tx_itr = low_latency;
1938
1939         adapter->rx_itr = e1000_update_itr(adapter,
1940                                     adapter->rx_itr,
1941                                     adapter->total_rx_packets,
1942                                     adapter->total_rx_bytes);
1943         /* conservative mode (itr 3) eliminates the lowest_latency setting */
1944         if (adapter->itr_setting == 3 && adapter->rx_itr == lowest_latency)
1945                 adapter->rx_itr = low_latency;
1946
1947         current_itr = max(adapter->rx_itr, adapter->tx_itr);
1948
1949         switch (current_itr) {
1950         /* counts and packets in update_itr are dependent on these numbers */
1951         case lowest_latency:
1952                 new_itr = 70000;
1953                 break;
1954         case low_latency:
1955                 new_itr = 20000; /* aka hwitr = ~200 */
1956                 break;
1957         case bulk_latency:
1958                 new_itr = 4000;
1959                 break;
1960         default:
1961                 break;
1962         }
1963
1964 set_itr_now:
1965         if (new_itr != adapter->itr) {
1966                 /*
1967                  * this attempts to bias the interrupt rate towards Bulk
1968                  * by adding intermediate steps when interrupt rate is
1969                  * increasing
1970                  */
1971                 new_itr = new_itr > adapter->itr ?
1972                              min(adapter->itr + (new_itr >> 2), new_itr) :
1973                              new_itr;
1974                 adapter->itr = new_itr;
1975                 adapter->rx_ring->itr_val = new_itr;
1976                 if (adapter->msix_entries)
1977                         adapter->rx_ring->set_itr = 1;
1978                 else
1979                         ew32(ITR, 1000000000 / (new_itr * 256));
1980         }
1981 }
1982
1983 /**
1984  * e1000_alloc_queues - Allocate memory for all rings
1985  * @adapter: board private structure to initialize
1986  **/
1987 static int __devinit e1000_alloc_queues(struct e1000_adapter *adapter)
1988 {
1989         adapter->tx_ring = kzalloc(sizeof(struct e1000_ring), GFP_KERNEL);
1990         if (!adapter->tx_ring)
1991                 goto err;
1992
1993         adapter->rx_ring = kzalloc(sizeof(struct e1000_ring), GFP_KERNEL);
1994         if (!adapter->rx_ring)
1995                 goto err;
1996
1997         return 0;
1998 err:
1999         e_err("Unable to allocate memory for queues\n");
2000         kfree(adapter->rx_ring);
2001         kfree(adapter->tx_ring);
2002         return -ENOMEM;
2003 }
2004
2005 /**
2006  * e1000_clean - NAPI Rx polling callback
2007  * @napi: struct associated with this polling callback
2008  * @budget: amount of packets driver is allowed to process this poll
2009  **/
2010 static int e1000_clean(struct napi_struct *napi, int budget)
2011 {
2012         struct e1000_adapter *adapter = container_of(napi, struct e1000_adapter, napi);
2013         struct e1000_hw *hw = &adapter->hw;
2014         struct net_device *poll_dev = adapter->netdev;
2015         int tx_cleaned = 1, work_done = 0;
2016
2017         adapter = netdev_priv(poll_dev);
2018
2019         if (adapter->msix_entries &&
2020             !(adapter->rx_ring->ims_val & adapter->tx_ring->ims_val))
2021                 goto clean_rx;
2022
2023         tx_cleaned = e1000_clean_tx_irq(adapter);
2024
2025 clean_rx:
2026         adapter->clean_rx(adapter, &work_done, budget);
2027
2028         if (!tx_cleaned)
2029                 work_done = budget;
2030
2031         /* If budget not fully consumed, exit the polling mode */
2032         if (work_done < budget) {
2033                 if (adapter->itr_setting & 3)
2034                         e1000_set_itr(adapter);
2035                 napi_complete(napi);
2036                 if (!test_bit(__E1000_DOWN, &adapter->state)) {
2037                         if (adapter->msix_entries)
2038                                 ew32(IMS, adapter->rx_ring->ims_val);
2039                         else
2040                                 e1000_irq_enable(adapter);
2041                 }
2042         }
2043
2044         return work_done;
2045 }
2046
2047 static void e1000_vlan_rx_add_vid(struct net_device *netdev, u16 vid)
2048 {
2049         struct e1000_adapter *adapter = netdev_priv(netdev);
2050         struct e1000_hw *hw = &adapter->hw;
2051         u32 vfta, index;
2052
2053         /* don't update vlan cookie if already programmed */
2054         if ((adapter->hw.mng_cookie.status &
2055              E1000_MNG_DHCP_COOKIE_STATUS_VLAN) &&
2056             (vid == adapter->mng_vlan_id))
2057                 return;
2058
2059         /* add VID to filter table */
2060         if (adapter->flags & FLAG_HAS_HW_VLAN_FILTER) {
2061                 index = (vid >> 5) & 0x7F;
2062                 vfta = E1000_READ_REG_ARRAY(hw, E1000_VFTA, index);
2063                 vfta |= (1 << (vid & 0x1F));
2064                 hw->mac.ops.write_vfta(hw, index, vfta);
2065         }
2066 }
2067
2068 static void e1000_vlan_rx_kill_vid(struct net_device *netdev, u16 vid)
2069 {
2070         struct e1000_adapter *adapter = netdev_priv(netdev);
2071         struct e1000_hw *hw = &adapter->hw;
2072         u32 vfta, index;
2073
2074         if (!test_bit(__E1000_DOWN, &adapter->state))
2075                 e1000_irq_disable(adapter);
2076         vlan_group_set_device(adapter->vlgrp, vid, NULL);
2077
2078         if (!test_bit(__E1000_DOWN, &adapter->state))
2079                 e1000_irq_enable(adapter);
2080
2081         if ((adapter->hw.mng_cookie.status &
2082              E1000_MNG_DHCP_COOKIE_STATUS_VLAN) &&
2083             (vid == adapter->mng_vlan_id)) {
2084                 /* release control to f/w */
2085                 e1000_release_hw_control(adapter);
2086                 return;
2087         }
2088
2089         /* remove VID from filter table */
2090         if (adapter->flags & FLAG_HAS_HW_VLAN_FILTER) {
2091                 index = (vid >> 5) & 0x7F;
2092                 vfta = E1000_READ_REG_ARRAY(hw, E1000_VFTA, index);
2093                 vfta &= ~(1 << (vid & 0x1F));
2094                 hw->mac.ops.write_vfta(hw, index, vfta);
2095         }
2096 }
2097
2098 static void e1000_update_mng_vlan(struct e1000_adapter *adapter)
2099 {
2100         struct net_device *netdev = adapter->netdev;
2101         u16 vid = adapter->hw.mng_cookie.vlan_id;
2102         u16 old_vid = adapter->mng_vlan_id;
2103
2104         if (!adapter->vlgrp)
2105                 return;
2106
2107         if (!vlan_group_get_device(adapter->vlgrp, vid)) {
2108                 adapter->mng_vlan_id = E1000_MNG_VLAN_NONE;
2109                 if (adapter->hw.mng_cookie.status &
2110                         E1000_MNG_DHCP_COOKIE_STATUS_VLAN) {
2111                         e1000_vlan_rx_add_vid(netdev, vid);
2112                         adapter->mng_vlan_id = vid;
2113                 }
2114
2115                 if ((old_vid != (u16)E1000_MNG_VLAN_NONE) &&
2116                                 (vid != old_vid) &&
2117                     !vlan_group_get_device(adapter->vlgrp, old_vid))
2118                         e1000_vlan_rx_kill_vid(netdev, old_vid);
2119         } else {
2120                 adapter->mng_vlan_id = vid;
2121         }
2122 }
2123
2124
2125 static void e1000_vlan_rx_register(struct net_device *netdev,
2126                                    struct vlan_group *grp)
2127 {
2128         struct e1000_adapter *adapter = netdev_priv(netdev);
2129         struct e1000_hw *hw = &adapter->hw;
2130         u32 ctrl, rctl;
2131
2132         if (!test_bit(__E1000_DOWN, &adapter->state))
2133                 e1000_irq_disable(adapter);
2134         adapter->vlgrp = grp;
2135
2136         if (grp) {
2137                 /* enable VLAN tag insert/strip */
2138                 ctrl = er32(CTRL);
2139                 ctrl |= E1000_CTRL_VME;
2140                 ew32(CTRL, ctrl);
2141
2142                 if (adapter->flags & FLAG_HAS_HW_VLAN_FILTER) {
2143                         /* enable VLAN receive filtering */
2144                         rctl = er32(RCTL);
2145                         rctl &= ~E1000_RCTL_CFIEN;
2146                         ew32(RCTL, rctl);
2147                         e1000_update_mng_vlan(adapter);
2148                 }
2149         } else {
2150                 /* disable VLAN tag insert/strip */
2151                 ctrl = er32(CTRL);
2152                 ctrl &= ~E1000_CTRL_VME;
2153                 ew32(CTRL, ctrl);
2154
2155                 if (adapter->flags & FLAG_HAS_HW_VLAN_FILTER) {
2156                         if (adapter->mng_vlan_id !=
2157                             (u16)E1000_MNG_VLAN_NONE) {
2158                                 e1000_vlan_rx_kill_vid(netdev,
2159                                                        adapter->mng_vlan_id);
2160                                 adapter->mng_vlan_id = E1000_MNG_VLAN_NONE;
2161                         }
2162                 }
2163         }
2164
2165         if (!test_bit(__E1000_DOWN, &adapter->state))
2166                 e1000_irq_enable(adapter);
2167 }
2168
2169 static void e1000_restore_vlan(struct e1000_adapter *adapter)
2170 {
2171         u16 vid;
2172
2173         e1000_vlan_rx_register(adapter->netdev, adapter->vlgrp);
2174
2175         if (!adapter->vlgrp)
2176                 return;
2177
2178         for (vid = 0; vid < VLAN_GROUP_ARRAY_LEN; vid++) {
2179                 if (!vlan_group_get_device(adapter->vlgrp, vid))
2180                         continue;
2181                 e1000_vlan_rx_add_vid(adapter->netdev, vid);
2182         }
2183 }
2184
2185 static void e1000_init_manageability(struct e1000_adapter *adapter)
2186 {
2187         struct e1000_hw *hw = &adapter->hw;
2188         u32 manc, manc2h;
2189
2190         if (!(adapter->flags & FLAG_MNG_PT_ENABLED))
2191                 return;
2192
2193         manc = er32(MANC);
2194
2195         /*
2196          * enable receiving management packets to the host. this will probably
2197          * generate destination unreachable messages from the host OS, but
2198          * the packets will be handled on SMBUS
2199          */
2200         manc |= E1000_MANC_EN_MNG2HOST;
2201         manc2h = er32(MANC2H);
2202 #define E1000_MNG2HOST_PORT_623 (1 << 5)
2203 #define E1000_MNG2HOST_PORT_664 (1 << 6)
2204         manc2h |= E1000_MNG2HOST_PORT_623;
2205         manc2h |= E1000_MNG2HOST_PORT_664;
2206         ew32(MANC2H, manc2h);
2207         ew32(MANC, manc);
2208 }
2209
2210 /**
2211  * e1000_configure_tx - Configure 8254x Transmit Unit after Reset
2212  * @adapter: board private structure
2213  *
2214  * Configure the Tx unit of the MAC after a reset.
2215  **/
2216 static void e1000_configure_tx(struct e1000_adapter *adapter)
2217 {
2218         struct e1000_hw *hw = &adapter->hw;
2219         struct e1000_ring *tx_ring = adapter->tx_ring;
2220         u64 tdba;
2221         u32 tdlen, tctl, tipg, tarc;
2222         u32 ipgr1, ipgr2;
2223
2224         /* Setup the HW Tx Head and Tail descriptor pointers */
2225         tdba = tx_ring->dma;
2226         tdlen = tx_ring->count * sizeof(struct e1000_tx_desc);
2227         ew32(TDBAL, (tdba & DMA_BIT_MASK(32)));
2228         ew32(TDBAH, (tdba >> 32));
2229         ew32(TDLEN, tdlen);
2230         ew32(TDH, 0);
2231         ew32(TDT, 0);
2232         tx_ring->head = E1000_TDH;
2233         tx_ring->tail = E1000_TDT;
2234
2235         /* Set the default values for the Tx Inter Packet Gap timer */
2236         tipg = DEFAULT_82543_TIPG_IPGT_COPPER;          /*  8  */
2237         ipgr1 = DEFAULT_82543_TIPG_IPGR1;               /*  8  */
2238         ipgr2 = DEFAULT_82543_TIPG_IPGR2;               /*  6  */
2239
2240         if (adapter->flags & FLAG_TIPG_MEDIUM_FOR_80003ESLAN)
2241                 ipgr2 = DEFAULT_80003ES2LAN_TIPG_IPGR2; /*  7  */
2242
2243         tipg |= ipgr1 << E1000_TIPG_IPGR1_SHIFT;
2244         tipg |= ipgr2 << E1000_TIPG_IPGR2_SHIFT;
2245         ew32(TIPG, tipg);
2246
2247         /* Set the Tx Interrupt Delay register */
2248         ew32(TIDV, adapter->tx_int_delay);
2249         /* Tx irq moderation */
2250         ew32(TADV, adapter->tx_abs_int_delay);
2251
2252         /* Program the Transmit Control Register */
2253         tctl = er32(TCTL);
2254         tctl &= ~E1000_TCTL_CT;
2255         tctl |= E1000_TCTL_PSP | E1000_TCTL_RTLC |
2256                 (E1000_COLLISION_THRESHOLD << E1000_CT_SHIFT);
2257
2258         if (adapter->flags & FLAG_TARC_SPEED_MODE_BIT) {
2259                 tarc = er32(TARC(0));
2260                 /*
2261                  * set the speed mode bit, we'll clear it if we're not at
2262                  * gigabit link later
2263                  */
2264 #define SPEED_MODE_BIT (1 << 21)
2265                 tarc |= SPEED_MODE_BIT;
2266                 ew32(TARC(0), tarc);
2267         }
2268
2269         /* errata: program both queues to unweighted RR */
2270         if (adapter->flags & FLAG_TARC_SET_BIT_ZERO) {
2271                 tarc = er32(TARC(0));
2272                 tarc |= 1;
2273                 ew32(TARC(0), tarc);
2274                 tarc = er32(TARC(1));
2275                 tarc |= 1;
2276                 ew32(TARC(1), tarc);
2277         }
2278
2279         /* Setup Transmit Descriptor Settings for eop descriptor */
2280         adapter->txd_cmd = E1000_TXD_CMD_EOP | E1000_TXD_CMD_IFCS;
2281
2282         /* only set IDE if we are delaying interrupts using the timers */
2283         if (adapter->tx_int_delay)
2284                 adapter->txd_cmd |= E1000_TXD_CMD_IDE;
2285
2286         /* enable Report Status bit */
2287         adapter->txd_cmd |= E1000_TXD_CMD_RS;
2288
2289         ew32(TCTL, tctl);
2290
2291         e1000e_config_collision_dist(hw);
2292 }
2293
2294 /**
2295  * e1000_setup_rctl - configure the receive control registers
2296  * @adapter: Board private structure
2297  **/
2298 #define PAGE_USE_COUNT(S) (((S) >> PAGE_SHIFT) + \
2299                            (((S) & (PAGE_SIZE - 1)) ? 1 : 0))
2300 static void e1000_setup_rctl(struct e1000_adapter *adapter)
2301 {
2302         struct e1000_hw *hw = &adapter->hw;
2303         u32 rctl, rfctl;
2304         u32 psrctl = 0;
2305         u32 pages = 0;
2306
2307         /* Program MC offset vector base */
2308         rctl = er32(RCTL);
2309         rctl &= ~(3 << E1000_RCTL_MO_SHIFT);
2310         rctl |= E1000_RCTL_EN | E1000_RCTL_BAM |
2311                 E1000_RCTL_LBM_NO | E1000_RCTL_RDMTS_HALF |
2312                 (adapter->hw.mac.mc_filter_type << E1000_RCTL_MO_SHIFT);
2313
2314         /* Do not Store bad packets */
2315         rctl &= ~E1000_RCTL_SBP;
2316
2317         /* Enable Long Packet receive */
2318         if (adapter->netdev->mtu <= ETH_DATA_LEN)
2319                 rctl &= ~E1000_RCTL_LPE;
2320         else
2321                 rctl |= E1000_RCTL_LPE;
2322
2323         /* Some systems expect that the CRC is included in SMBUS traffic. The
2324          * hardware strips the CRC before sending to both SMBUS (BMC) and to
2325          * host memory when this is enabled
2326          */
2327         if (adapter->flags2 & FLAG2_CRC_STRIPPING)
2328                 rctl |= E1000_RCTL_SECRC;
2329
2330         /* Workaround Si errata on 82577 PHY - configure IPG for jumbos */
2331         if ((hw->phy.type == e1000_phy_82577) && (rctl & E1000_RCTL_LPE)) {
2332                 u16 phy_data;
2333
2334                 e1e_rphy(hw, PHY_REG(770, 26), &phy_data);
2335                 phy_data &= 0xfff8;
2336                 phy_data |= (1 << 2);
2337                 e1e_wphy(hw, PHY_REG(770, 26), phy_data);
2338
2339                 e1e_rphy(hw, 22, &phy_data);
2340                 phy_data &= 0x0fff;
2341                 phy_data |= (1 << 14);
2342                 e1e_wphy(hw, 0x10, 0x2823);
2343                 e1e_wphy(hw, 0x11, 0x0003);
2344                 e1e_wphy(hw, 22, phy_data);
2345         }
2346
2347         /* Setup buffer sizes */
2348         rctl &= ~E1000_RCTL_SZ_4096;
2349         rctl |= E1000_RCTL_BSEX;
2350         switch (adapter->rx_buffer_len) {
2351         case 2048:
2352         default:
2353                 rctl |= E1000_RCTL_SZ_2048;
2354                 rctl &= ~E1000_RCTL_BSEX;
2355                 break;
2356         case 4096:
2357                 rctl |= E1000_RCTL_SZ_4096;
2358                 break;
2359         case 8192:
2360                 rctl |= E1000_RCTL_SZ_8192;
2361                 break;
2362         case 16384:
2363                 rctl |= E1000_RCTL_SZ_16384;
2364                 break;
2365         }
2366
2367         /*
2368          * 82571 and greater support packet-split where the protocol
2369          * header is placed in skb->data and the packet data is
2370          * placed in pages hanging off of skb_shinfo(skb)->nr_frags.
2371          * In the case of a non-split, skb->data is linearly filled,
2372          * followed by the page buffers.  Therefore, skb->data is
2373          * sized to hold the largest protocol header.
2374          *
2375          * allocations using alloc_page take too long for regular MTU
2376          * so only enable packet split for jumbo frames
2377          *
2378          * Using pages when the page size is greater than 16k wastes
2379          * a lot of memory, since we allocate 3 pages at all times
2380          * per packet.
2381          */
2382         pages = PAGE_USE_COUNT(adapter->netdev->mtu);
2383         if (!(adapter->flags & FLAG_IS_ICH) && (pages <= 3) &&
2384             (PAGE_SIZE <= 16384) && (rctl & E1000_RCTL_LPE))
2385                 adapter->rx_ps_pages = pages;
2386         else
2387                 adapter->rx_ps_pages = 0;
2388
2389         if (adapter->rx_ps_pages) {
2390                 /* Configure extra packet-split registers */
2391                 rfctl = er32(RFCTL);
2392                 rfctl |= E1000_RFCTL_EXTEN;
2393                 /*
2394                  * disable packet split support for IPv6 extension headers,
2395                  * because some malformed IPv6 headers can hang the Rx
2396                  */
2397                 rfctl |= (E1000_RFCTL_IPV6_EX_DIS |
2398                           E1000_RFCTL_NEW_IPV6_EXT_DIS);
2399
2400                 ew32(RFCTL, rfctl);
2401
2402                 /* Enable Packet split descriptors */
2403                 rctl |= E1000_RCTL_DTYP_PS;
2404
2405                 psrctl |= adapter->rx_ps_bsize0 >>
2406                         E1000_PSRCTL_BSIZE0_SHIFT;
2407
2408                 switch (adapter->rx_ps_pages) {
2409                 case 3:
2410                         psrctl |= PAGE_SIZE <<
2411                                 E1000_PSRCTL_BSIZE3_SHIFT;
2412                 case 2:
2413                         psrctl |= PAGE_SIZE <<
2414                                 E1000_PSRCTL_BSIZE2_SHIFT;
2415                 case 1:
2416                         psrctl |= PAGE_SIZE >>
2417                                 E1000_PSRCTL_BSIZE1_SHIFT;
2418                         break;
2419                 }
2420
2421                 ew32(PSRCTL, psrctl);
2422         }
2423
2424         ew32(RCTL, rctl);
2425         /* just started the receive unit, no need to restart */
2426         adapter->flags &= ~FLAG_RX_RESTART_NOW;
2427 }
2428
2429 /**
2430  * e1000_configure_rx - Configure Receive Unit after Reset
2431  * @adapter: board private structure
2432  *
2433  * Configure the Rx unit of the MAC after a reset.
2434  **/
2435 static void e1000_configure_rx(struct e1000_adapter *adapter)
2436 {
2437         struct e1000_hw *hw = &adapter->hw;
2438         struct e1000_ring *rx_ring = adapter->rx_ring;
2439         u64 rdba;
2440         u32 rdlen, rctl, rxcsum, ctrl_ext;
2441
2442         if (adapter->rx_ps_pages) {
2443                 /* this is a 32 byte descriptor */
2444                 rdlen = rx_ring->count *
2445                         sizeof(union e1000_rx_desc_packet_split);
2446                 adapter->clean_rx = e1000_clean_rx_irq_ps;
2447                 adapter->alloc_rx_buf = e1000_alloc_rx_buffers_ps;
2448         } else if (adapter->netdev->mtu > ETH_FRAME_LEN + ETH_FCS_LEN) {
2449                 rdlen = rx_ring->count * sizeof(struct e1000_rx_desc);
2450                 adapter->clean_rx = e1000_clean_jumbo_rx_irq;
2451                 adapter->alloc_rx_buf = e1000_alloc_jumbo_rx_buffers;
2452         } else {
2453                 rdlen = rx_ring->count * sizeof(struct e1000_rx_desc);
2454                 adapter->clean_rx = e1000_clean_rx_irq;
2455                 adapter->alloc_rx_buf = e1000_alloc_rx_buffers;
2456         }
2457
2458         /* disable receives while setting up the descriptors */
2459         rctl = er32(RCTL);
2460         ew32(RCTL, rctl & ~E1000_RCTL_EN);
2461         e1e_flush();
2462         msleep(10);
2463
2464         /* set the Receive Delay Timer Register */
2465         ew32(RDTR, adapter->rx_int_delay);
2466
2467         /* irq moderation */
2468         ew32(RADV, adapter->rx_abs_int_delay);
2469         if (adapter->itr_setting != 0)
2470                 ew32(ITR, 1000000000 / (adapter->itr * 256));
2471
2472         ctrl_ext = er32(CTRL_EXT);
2473         /* Auto-Mask interrupts upon ICR access */
2474         ctrl_ext |= E1000_CTRL_EXT_IAME;
2475         ew32(IAM, 0xffffffff);
2476         ew32(CTRL_EXT, ctrl_ext);
2477         e1e_flush();
2478
2479         /*
2480          * Setup the HW Rx Head and Tail Descriptor Pointers and
2481          * the Base and Length of the Rx Descriptor Ring
2482          */
2483         rdba = rx_ring->dma;
2484         ew32(RDBAL, (rdba & DMA_BIT_MASK(32)));
2485         ew32(RDBAH, (rdba >> 32));
2486         ew32(RDLEN, rdlen);
2487         ew32(RDH, 0);
2488         ew32(RDT, 0);
2489         rx_ring->head = E1000_RDH;
2490         rx_ring->tail = E1000_RDT;
2491
2492         /* Enable Receive Checksum Offload for TCP and UDP */
2493         rxcsum = er32(RXCSUM);
2494         if (adapter->flags & FLAG_RX_CSUM_ENABLED) {
2495                 rxcsum |= E1000_RXCSUM_TUOFL;
2496
2497                 /*
2498                  * IPv4 payload checksum for UDP fragments must be
2499                  * used in conjunction with packet-split.
2500                  */
2501                 if (adapter->rx_ps_pages)
2502                         rxcsum |= E1000_RXCSUM_IPPCSE;
2503         } else {
2504                 rxcsum &= ~E1000_RXCSUM_TUOFL;
2505                 /* no need to clear IPPCSE as it defaults to 0 */
2506         }
2507         ew32(RXCSUM, rxcsum);
2508
2509         /*
2510          * Enable early receives on supported devices, only takes effect when
2511          * packet size is equal or larger than the specified value (in 8 byte
2512          * units), e.g. using jumbo frames when setting to E1000_ERT_2048
2513          */
2514         if (adapter->flags & FLAG_HAS_ERT) {
2515                 if (adapter->netdev->mtu > ETH_DATA_LEN) {
2516                         u32 rxdctl = er32(RXDCTL(0));
2517                         ew32(RXDCTL(0), rxdctl | 0x3);
2518                         ew32(ERT, E1000_ERT_2048 | (1 << 13));
2519                         /*
2520                          * With jumbo frames and early-receive enabled,
2521                          * excessive C-state transition latencies result in
2522                          * dropped transactions.
2523                          */
2524                         pm_qos_update_requirement(PM_QOS_CPU_DMA_LATENCY,
2525                                                   adapter->netdev->name, 55);
2526                 } else {
2527                         pm_qos_update_requirement(PM_QOS_CPU_DMA_LATENCY,
2528                                                   adapter->netdev->name,
2529                                                   PM_QOS_DEFAULT_VALUE);
2530                 }
2531         }
2532
2533         /* Enable Receives */
2534         ew32(RCTL, rctl);
2535 }
2536
2537 /**
2538  *  e1000_update_mc_addr_list - Update Multicast addresses
2539  *  @hw: pointer to the HW structure
2540  *  @mc_addr_list: array of multicast addresses to program
2541  *  @mc_addr_count: number of multicast addresses to program
2542  *
2543  *  Updates the Multicast Table Array.
2544  *  The caller must have a packed mc_addr_list of multicast addresses.
2545  **/
2546 static void e1000_update_mc_addr_list(struct e1000_hw *hw, u8 *mc_addr_list,
2547                                       u32 mc_addr_count)
2548 {
2549         hw->mac.ops.update_mc_addr_list(hw, mc_addr_list, mc_addr_count);
2550 }
2551
2552 /**
2553  * e1000_set_multi - Multicast and Promiscuous mode set
2554  * @netdev: network interface device structure
2555  *
2556  * The set_multi entry point is called whenever the multicast address
2557  * list or the network interface flags are updated.  This routine is
2558  * responsible for configuring the hardware for proper multicast,
2559  * promiscuous mode, and all-multi behavior.
2560  **/
2561 static void e1000_set_multi(struct net_device *netdev)
2562 {
2563         struct e1000_adapter *adapter = netdev_priv(netdev);
2564         struct e1000_hw *hw = &adapter->hw;
2565         struct dev_mc_list *mc_ptr;
2566         u8  *mta_list;
2567         u32 rctl;
2568         int i;
2569
2570         /* Check for Promiscuous and All Multicast modes */
2571
2572         rctl = er32(RCTL);
2573
2574         if (netdev->flags & IFF_PROMISC) {
2575                 rctl |= (E1000_RCTL_UPE | E1000_RCTL_MPE);
2576                 rctl &= ~E1000_RCTL_VFE;
2577         } else {
2578                 if (netdev->flags & IFF_ALLMULTI) {
2579                         rctl |= E1000_RCTL_MPE;
2580                         rctl &= ~E1000_RCTL_UPE;
2581                 } else {
2582                         rctl &= ~(E1000_RCTL_UPE | E1000_RCTL_MPE);
2583                 }
2584                 if (adapter->flags & FLAG_HAS_HW_VLAN_FILTER)
2585                         rctl |= E1000_RCTL_VFE;
2586         }
2587
2588         ew32(RCTL, rctl);
2589
2590         if (!netdev_mc_empty(netdev)) {
2591                 mta_list = kmalloc(netdev_mc_count(netdev) * 6, GFP_ATOMIC);
2592                 if (!mta_list)
2593                         return;
2594
2595                 /* prepare a packed array of only addresses. */
2596                 i = 0;
2597                 netdev_for_each_mc_addr(mc_ptr, netdev)
2598                         memcpy(mta_list + (i++ * ETH_ALEN),
2599                                mc_ptr->dmi_addr, ETH_ALEN);
2600
2601                 e1000_update_mc_addr_list(hw, mta_list, i);
2602                 kfree(mta_list);
2603         } else {
2604                 /*
2605                  * if we're called from probe, we might not have
2606                  * anything to do here, so clear out the list
2607                  */
2608                 e1000_update_mc_addr_list(hw, NULL, 0);
2609         }
2610 }
2611
2612 /**
2613  * e1000_configure - configure the hardware for Rx and Tx
2614  * @adapter: private board structure
2615  **/
2616 static void e1000_configure(struct e1000_adapter *adapter)
2617 {
2618         e1000_set_multi(adapter->netdev);
2619
2620         e1000_restore_vlan(adapter);
2621         e1000_init_manageability(adapter);
2622
2623         e1000_configure_tx(adapter);
2624         e1000_setup_rctl(adapter);
2625         e1000_configure_rx(adapter);
2626         adapter->alloc_rx_buf(adapter, e1000_desc_unused(adapter->rx_ring));
2627 }
2628
2629 /**
2630  * e1000e_power_up_phy - restore link in case the phy was powered down
2631  * @adapter: address of board private structure
2632  *
2633  * The phy may be powered down to save power and turn off link when the
2634  * driver is unloaded and wake on lan is not enabled (among others)
2635  * *** this routine MUST be followed by a call to e1000e_reset ***
2636  **/
2637 void e1000e_power_up_phy(struct e1000_adapter *adapter)
2638 {
2639         if (adapter->hw.phy.ops.power_up)
2640                 adapter->hw.phy.ops.power_up(&adapter->hw);
2641
2642         adapter->hw.mac.ops.setup_link(&adapter->hw);
2643 }
2644
2645 /**
2646  * e1000_power_down_phy - Power down the PHY
2647  *
2648  * Power down the PHY so no link is implied when interface is down.
2649  * The PHY cannot be powered down if management or WoL is active.
2650  */
2651 static void e1000_power_down_phy(struct e1000_adapter *adapter)
2652 {
2653         /* WoL is enabled */
2654         if (adapter->wol)
2655                 return;
2656
2657         if (adapter->hw.phy.ops.power_down)
2658                 adapter->hw.phy.ops.power_down(&adapter->hw);
2659 }
2660
2661 /**
2662  * e1000e_reset - bring the hardware into a known good state
2663  *
2664  * This function boots the hardware and enables some settings that
2665  * require a configuration cycle of the hardware - those cannot be
2666  * set/changed during runtime. After reset the device needs to be
2667  * properly configured for Rx, Tx etc.
2668  */
2669 void e1000e_reset(struct e1000_adapter *adapter)
2670 {
2671         struct e1000_mac_info *mac = &adapter->hw.mac;
2672         struct e1000_fc_info *fc = &adapter->hw.fc;
2673         struct e1000_hw *hw = &adapter->hw;
2674         u32 tx_space, min_tx_space, min_rx_space;
2675         u32 pba = adapter->pba;
2676         u16 hwm;
2677
2678         /* reset Packet Buffer Allocation to default */
2679         ew32(PBA, pba);
2680
2681         if (adapter->max_frame_size > ETH_FRAME_LEN + ETH_FCS_LEN) {
2682                 /*
2683                  * To maintain wire speed transmits, the Tx FIFO should be
2684                  * large enough to accommodate two full transmit packets,
2685                  * rounded up to the next 1KB and expressed in KB.  Likewise,
2686                  * the Rx FIFO should be large enough to accommodate at least
2687                  * one full receive packet and is similarly rounded up and
2688                  * expressed in KB.
2689                  */
2690                 pba = er32(PBA);
2691                 /* upper 16 bits has Tx packet buffer allocation size in KB */
2692                 tx_space = pba >> 16;
2693                 /* lower 16 bits has Rx packet buffer allocation size in KB */
2694                 pba &= 0xffff;
2695                 /*
2696                  * the Tx fifo also stores 16 bytes of information about the tx
2697                  * but don't include ethernet FCS because hardware appends it
2698                  */
2699                 min_tx_space = (adapter->max_frame_size +
2700                                 sizeof(struct e1000_tx_desc) -
2701                                 ETH_FCS_LEN) * 2;
2702                 min_tx_space = ALIGN(min_tx_space, 1024);
2703                 min_tx_space >>= 10;
2704                 /* software strips receive CRC, so leave room for it */
2705                 min_rx_space = adapter->max_frame_size;
2706                 min_rx_space = ALIGN(min_rx_space, 1024);
2707                 min_rx_space >>= 10;
2708
2709                 /*
2710                  * If current Tx allocation is less than the min Tx FIFO size,
2711                  * and the min Tx FIFO size is less than the current Rx FIFO
2712                  * allocation, take space away from current Rx allocation
2713                  */
2714                 if ((tx_space < min_tx_space) &&
2715                     ((min_tx_space - tx_space) < pba)) {
2716                         pba -= min_tx_space - tx_space;
2717
2718                         /*
2719                          * if short on Rx space, Rx wins and must trump tx
2720                          * adjustment or use Early Receive if available
2721                          */
2722                         if ((pba < min_rx_space) &&
2723                             (!(adapter->flags & FLAG_HAS_ERT)))
2724                                 /* ERT enabled in e1000_configure_rx */
2725                                 pba = min_rx_space;
2726                 }
2727
2728                 ew32(PBA, pba);
2729         }
2730
2731
2732         /*
2733          * flow control settings
2734          *
2735          * The high water mark must be low enough to fit one full frame
2736          * (or the size used for early receive) above it in the Rx FIFO.
2737          * Set it to the lower of:
2738          * - 90% of the Rx FIFO size, and
2739          * - the full Rx FIFO size minus the early receive size (for parts
2740          *   with ERT support assuming ERT set to E1000_ERT_2048), or
2741          * - the full Rx FIFO size minus one full frame
2742          */
2743         if (hw->mac.type == e1000_pchlan) {
2744                 /*
2745                  * Workaround PCH LOM adapter hangs with certain network
2746                  * loads.  If hangs persist, try disabling Tx flow control.
2747                  */
2748                 if (adapter->netdev->mtu > ETH_DATA_LEN) {
2749                         fc->high_water = 0x3500;
2750                         fc->low_water  = 0x1500;
2751                 } else {
2752                         fc->high_water = 0x5000;
2753                         fc->low_water  = 0x3000;
2754                 }
2755         } else {
2756                 if ((adapter->flags & FLAG_HAS_ERT) &&
2757                     (adapter->netdev->mtu > ETH_DATA_LEN))
2758                         hwm = min(((pba << 10) * 9 / 10),
2759                                   ((pba << 10) - (E1000_ERT_2048 << 3)));
2760                 else
2761                         hwm = min(((pba << 10) * 9 / 10),
2762                                   ((pba << 10) - adapter->max_frame_size));
2763
2764                 fc->high_water = hwm & E1000_FCRTH_RTH; /* 8-byte granularity */
2765                 fc->low_water = fc->high_water - 8;
2766         }
2767
2768         if (adapter->flags & FLAG_DISABLE_FC_PAUSE_TIME)
2769                 fc->pause_time = 0xFFFF;
2770         else
2771                 fc->pause_time = E1000_FC_PAUSE_TIME;
2772         fc->send_xon = 1;
2773         fc->current_mode = fc->requested_mode;
2774
2775         /* Allow time for pending master requests to run */
2776         mac->ops.reset_hw(hw);
2777
2778         /*
2779          * For parts with AMT enabled, let the firmware know
2780          * that the network interface is in control
2781          */
2782         if (adapter->flags & FLAG_HAS_AMT)
2783                 e1000_get_hw_control(adapter);
2784
2785         ew32(WUC, 0);
2786         if (adapter->flags2 & FLAG2_HAS_PHY_WAKEUP)
2787                 e1e_wphy(&adapter->hw, BM_WUC, 0);
2788
2789         if (mac->ops.init_hw(hw))
2790                 e_err("Hardware Error\n");
2791
2792         /* additional part of the flow-control workaround above */
2793         if (hw->mac.type == e1000_pchlan)
2794                 ew32(FCRTV_PCH, 0x1000);
2795
2796         e1000_update_mng_vlan(adapter);
2797
2798         /* Enable h/w to recognize an 802.1Q VLAN Ethernet packet */
2799         ew32(VET, ETH_P_8021Q);
2800
2801         e1000e_reset_adaptive(hw);
2802         e1000_get_phy_info(hw);
2803
2804         if ((adapter->flags & FLAG_HAS_SMART_POWER_DOWN) &&
2805             !(adapter->flags & FLAG_SMART_POWER_DOWN)) {
2806                 u16 phy_data = 0;
2807                 /*
2808                  * speed up time to link by disabling smart power down, ignore
2809                  * the return value of this function because there is nothing
2810                  * different we would do if it failed
2811                  */
2812                 e1e_rphy(hw, IGP02E1000_PHY_POWER_MGMT, &phy_data);
2813                 phy_data &= ~IGP02E1000_PM_SPD;
2814                 e1e_wphy(hw, IGP02E1000_PHY_POWER_MGMT, phy_data);
2815         }
2816 }
2817
2818 int e1000e_up(struct e1000_adapter *adapter)
2819 {
2820         struct e1000_hw *hw = &adapter->hw;
2821
2822         /* DMA latency requirement to workaround early-receive/jumbo issue */
2823         if (adapter->flags & FLAG_HAS_ERT)
2824                 pm_qos_add_requirement(PM_QOS_CPU_DMA_LATENCY,
2825                                        adapter->netdev->name,
2826                                        PM_QOS_DEFAULT_VALUE);
2827
2828         /* hardware has been reset, we need to reload some things */
2829         e1000_configure(adapter);
2830
2831         clear_bit(__E1000_DOWN, &adapter->state);
2832
2833         napi_enable(&adapter->napi);
2834         if (adapter->msix_entries)
2835                 e1000_configure_msix(adapter);
2836         e1000_irq_enable(adapter);
2837
2838         netif_wake_queue(adapter->netdev);
2839
2840         /* fire a link change interrupt to start the watchdog */
2841         ew32(ICS, E1000_ICS_LSC);
2842         return 0;
2843 }
2844
2845 void e1000e_down(struct e1000_adapter *adapter)
2846 {
2847         struct net_device *netdev = adapter->netdev;
2848         struct e1000_hw *hw = &adapter->hw;
2849         u32 tctl, rctl;
2850
2851         /*
2852          * signal that we're down so the interrupt handler does not
2853          * reschedule our watchdog timer
2854          */
2855         set_bit(__E1000_DOWN, &adapter->state);
2856
2857         /* disable receives in the hardware */
2858         rctl = er32(RCTL);
2859         ew32(RCTL, rctl & ~E1000_RCTL_EN);
2860         /* flush and sleep below */
2861
2862         netif_stop_queue(netdev);
2863
2864         /* disable transmits in the hardware */
2865         tctl = er32(TCTL);
2866         tctl &= ~E1000_TCTL_EN;
2867         ew32(TCTL, tctl);
2868         /* flush both disables and wait for them to finish */
2869         e1e_flush();
2870         msleep(10);
2871
2872         napi_disable(&adapter->napi);
2873         e1000_irq_disable(adapter);
2874
2875         del_timer_sync(&adapter->watchdog_timer);
2876         del_timer_sync(&adapter->phy_info_timer);
2877
2878         netif_carrier_off(netdev);
2879         adapter->link_speed = 0;
2880         adapter->link_duplex = 0;
2881
2882         if (!pci_channel_offline(adapter->pdev))
2883                 e1000e_reset(adapter);
2884         e1000_clean_tx_ring(adapter);
2885         e1000_clean_rx_ring(adapter);
2886
2887         if (adapter->flags & FLAG_HAS_ERT)
2888                 pm_qos_remove_requirement(PM_QOS_CPU_DMA_LATENCY,
2889                                           adapter->netdev->name);
2890
2891         /*
2892          * TODO: for power management, we could drop the link and
2893          * pci_disable_device here.
2894          */
2895 }
2896
2897 void e1000e_reinit_locked(struct e1000_adapter *adapter)
2898 {
2899         might_sleep();
2900         while (test_and_set_bit(__E1000_RESETTING, &adapter->state))
2901                 msleep(1);
2902         e1000e_down(adapter);
2903         e1000e_up(adapter);
2904         clear_bit(__E1000_RESETTING, &adapter->state);
2905 }
2906
2907 /**
2908  * e1000_sw_init - Initialize general software structures (struct e1000_adapter)
2909  * @adapter: board private structure to initialize
2910  *
2911  * e1000_sw_init initializes the Adapter private data structure.
2912  * Fields are initialized based on PCI device information and
2913  * OS network device settings (MTU size).
2914  **/
2915 static int __devinit e1000_sw_init(struct e1000_adapter *adapter)
2916 {
2917         struct net_device *netdev = adapter->netdev;
2918
2919         adapter->rx_buffer_len = ETH_FRAME_LEN + VLAN_HLEN + ETH_FCS_LEN;
2920         adapter->rx_ps_bsize0 = 128;
2921         adapter->max_frame_size = netdev->mtu + ETH_HLEN + ETH_FCS_LEN;
2922         adapter->min_frame_size = ETH_ZLEN + ETH_FCS_LEN;
2923
2924         e1000e_set_interrupt_capability(adapter);
2925
2926         if (e1000_alloc_queues(adapter))
2927                 return -ENOMEM;
2928
2929         /* Explicitly disable IRQ since the NIC can be in any state. */
2930         e1000_irq_disable(adapter);
2931
2932         set_bit(__E1000_DOWN, &adapter->state);
2933         return 0;
2934 }
2935
2936 /**
2937  * e1000_intr_msi_test - Interrupt Handler
2938  * @irq: interrupt number
2939  * @data: pointer to a network interface device structure
2940  **/
2941 static irqreturn_t e1000_intr_msi_test(int irq, void *data)
2942 {
2943         struct net_device *netdev = data;
2944         struct e1000_adapter *adapter = netdev_priv(netdev);
2945         struct e1000_hw *hw = &adapter->hw;
2946         u32 icr = er32(ICR);
2947
2948         e_dbg("icr is %08X\n", icr);
2949         if (icr & E1000_ICR_RXSEQ) {
2950                 adapter->flags &= ~FLAG_MSI_TEST_FAILED;
2951                 wmb();
2952         }
2953
2954         return IRQ_HANDLED;
2955 }
2956
2957 /**
2958  * e1000_test_msi_interrupt - Returns 0 for successful test
2959  * @adapter: board private struct
2960  *
2961  * code flow taken from tg3.c
2962  **/
2963 static int e1000_test_msi_interrupt(struct e1000_adapter *adapter)
2964 {
2965         struct net_device *netdev = adapter->netdev;
2966         struct e1000_hw *hw = &adapter->hw;
2967         int err;
2968
2969         /* poll_enable hasn't been called yet, so don't need disable */
2970         /* clear any pending events */
2971         er32(ICR);
2972
2973         /* free the real vector and request a test handler */
2974         e1000_free_irq(adapter);
2975         e1000e_reset_interrupt_capability(adapter);
2976
2977         /* Assume that the test fails, if it succeeds then the test
2978          * MSI irq handler will unset this flag */
2979         adapter->flags |= FLAG_MSI_TEST_FAILED;
2980
2981         err = pci_enable_msi(adapter->pdev);
2982         if (err)
2983                 goto msi_test_failed;
2984
2985         err = request_irq(adapter->pdev->irq, e1000_intr_msi_test, 0,
2986                           netdev->name, netdev);
2987         if (err) {
2988                 pci_disable_msi(adapter->pdev);
2989                 goto msi_test_failed;
2990         }
2991
2992         wmb();
2993
2994         e1000_irq_enable(adapter);
2995
2996         /* fire an unusual interrupt on the test handler */
2997         ew32(ICS, E1000_ICS_RXSEQ);
2998         e1e_flush();
2999         msleep(50);
3000
3001         e1000_irq_disable(adapter);
3002
3003         rmb();
3004
3005         if (adapter->flags & FLAG_MSI_TEST_FAILED) {
3006                 adapter->int_mode = E1000E_INT_MODE_LEGACY;
3007                 err = -EIO;
3008                 e_info("MSI interrupt test failed!\n");
3009         }
3010
3011         free_irq(adapter->pdev->irq, netdev);
3012         pci_disable_msi(adapter->pdev);
3013
3014         if (err == -EIO)
3015                 goto msi_test_failed;
3016
3017         /* okay so the test worked, restore settings */
3018         e_dbg("MSI interrupt test succeeded!\n");
3019 msi_test_failed:
3020         e1000e_set_interrupt_capability(adapter);
3021         e1000_request_irq(adapter);
3022         return err;
3023 }
3024
3025 /**
3026  * e1000_test_msi - Returns 0 if MSI test succeeds or INTx mode is restored
3027  * @adapter: board private struct
3028  *
3029  * code flow taken from tg3.c, called with e1000 interrupts disabled.
3030  **/
3031 static int e1000_test_msi(struct e1000_adapter *adapter)
3032 {
3033         int err;
3034         u16 pci_cmd;
3035
3036         if (!(adapter->flags & FLAG_MSI_ENABLED))
3037                 return 0;
3038
3039         /* disable SERR in case the MSI write causes a master abort */
3040         pci_read_config_word(adapter->pdev, PCI_COMMAND, &pci_cmd);
3041         pci_write_config_word(adapter->pdev, PCI_COMMAND,
3042                               pci_cmd & ~PCI_COMMAND_SERR);
3043
3044         err = e1000_test_msi_interrupt(adapter);
3045
3046         /* restore previous setting of command word */
3047         pci_write_config_word(adapter->pdev, PCI_COMMAND, pci_cmd);
3048
3049         /* success ! */
3050         if (!err)
3051                 return 0;
3052
3053         /* EIO means MSI test failed */
3054         if (err != -EIO)
3055                 return err;
3056
3057         /* back to INTx mode */
3058         e_warn("MSI interrupt test failed, using legacy interrupt.\n");
3059
3060         e1000_free_irq(adapter);
3061
3062         err = e1000_request_irq(adapter);
3063
3064         return err;
3065 }
3066
3067 /**
3068  * e1000_open - Called when a network interface is made active
3069  * @netdev: network interface device structure
3070  *
3071  * Returns 0 on success, negative value on failure
3072  *
3073  * The open entry point is called when a network interface is made
3074  * active by the system (IFF_UP).  At this point all resources needed
3075  * for transmit and receive operations are allocated, the interrupt
3076  * handler is registered with the OS, the watchdog timer is started,
3077  * and the stack is notified that the interface is ready.
3078  **/
3079 static int e1000_open(struct net_device *netdev)
3080 {
3081         struct e1000_adapter *adapter = netdev_priv(netdev);
3082         struct e1000_hw *hw = &adapter->hw;
3083         int err;
3084
3085         /* disallow open during test */
3086         if (test_bit(__E1000_TESTING, &adapter->state))
3087                 return -EBUSY;
3088
3089         netif_carrier_off(netdev);
3090
3091         /* allocate transmit descriptors */
3092         err = e1000e_setup_tx_resources(adapter);
3093         if (err)
3094                 goto err_setup_tx;
3095
3096         /* allocate receive descriptors */
3097         err = e1000e_setup_rx_resources(adapter);
3098         if (err)
3099                 goto err_setup_rx;
3100
3101         e1000e_power_up_phy(adapter);
3102
3103         adapter->mng_vlan_id = E1000_MNG_VLAN_NONE;
3104         if ((adapter->hw.mng_cookie.status &
3105              E1000_MNG_DHCP_COOKIE_STATUS_VLAN))
3106                 e1000_update_mng_vlan(adapter);
3107
3108         /*
3109          * If AMT is enabled, let the firmware know that the network
3110          * interface is now open
3111          */
3112         if (adapter->flags & FLAG_HAS_AMT)
3113                 e1000_get_hw_control(adapter);
3114
3115         /*
3116          * before we allocate an interrupt, we must be ready to handle it.
3117          * Setting DEBUG_SHIRQ in the kernel makes it fire an interrupt
3118          * as soon as we call pci_request_irq, so we have to setup our
3119          * clean_rx handler before we do so.
3120          */
3121         e1000_configure(adapter);
3122
3123         err = e1000_request_irq(adapter);
3124         if (err)
3125                 goto err_req_irq;
3126
3127         /*
3128          * Work around PCIe errata with MSI interrupts causing some chipsets to
3129          * ignore e1000e MSI messages, which means we need to test our MSI
3130          * interrupt now
3131          */
3132         if (adapter->int_mode != E1000E_INT_MODE_LEGACY) {
3133                 err = e1000_test_msi(adapter);
3134                 if (err) {
3135                         e_err("Interrupt allocation failed\n");
3136                         goto err_req_irq;
3137                 }
3138         }
3139
3140         /* From here on the code is the same as e1000e_up() */
3141         clear_bit(__E1000_DOWN, &adapter->state);
3142
3143         napi_enable(&adapter->napi);
3144
3145         e1000_irq_enable(adapter);
3146
3147         netif_start_queue(netdev);
3148
3149         /* fire a link status change interrupt to start the watchdog */
3150         ew32(ICS, E1000_ICS_LSC);
3151
3152         return 0;
3153
3154 err_req_irq:
3155         e1000_release_hw_control(adapter);
3156         e1000_power_down_phy(adapter);
3157         e1000e_free_rx_resources(adapter);
3158 err_setup_rx:
3159         e1000e_free_tx_resources(adapter);
3160 err_setup_tx:
3161         e1000e_reset(adapter);
3162
3163         return err;
3164 }
3165
3166 /**
3167  * e1000_close - Disables a network interface
3168  * @netdev: network interface device structure
3169  *
3170  * Returns 0, this is not allowed to fail
3171  *
3172  * The close entry point is called when an interface is de-activated
3173  * by the OS.  The hardware is still under the drivers control, but
3174  * needs to be disabled.  A global MAC reset is issued to stop the
3175  * hardware, and all transmit and receive resources are freed.
3176  **/
3177 static int e1000_close(struct net_device *netdev)
3178 {
3179         struct e1000_adapter *adapter = netdev_priv(netdev);
3180
3181         WARN_ON(test_bit(__E1000_RESETTING, &adapter->state));
3182         e1000e_down(adapter);
3183         e1000_power_down_phy(adapter);
3184         e1000_free_irq(adapter);
3185
3186         e1000e_free_tx_resources(adapter);
3187         e1000e_free_rx_resources(adapter);
3188
3189         /*
3190          * kill manageability vlan ID if supported, but not if a vlan with
3191          * the same ID is registered on the host OS (let 8021q kill it)
3192          */
3193         if ((adapter->hw.mng_cookie.status &
3194                           E1000_MNG_DHCP_COOKIE_STATUS_VLAN) &&
3195              !(adapter->vlgrp &&
3196                vlan_group_get_device(adapter->vlgrp, adapter->mng_vlan_id)))
3197                 e1000_vlan_rx_kill_vid(netdev, adapter->mng_vlan_id);
3198
3199         /*
3200          * If AMT is enabled, let the firmware know that the network
3201          * interface is now closed
3202          */
3203         if (adapter->flags & FLAG_HAS_AMT)
3204                 e1000_release_hw_control(adapter);
3205
3206         return 0;
3207 }
3208 /**
3209  * e1000_set_mac - Change the Ethernet Address of the NIC
3210  * @netdev: network interface device structure
3211  * @p: pointer to an address structure
3212  *
3213  * Returns 0 on success, negative on failure
3214  **/
3215 static int e1000_set_mac(struct net_device *netdev, void *p)
3216 {
3217         struct e1000_adapter *adapter = netdev_priv(netdev);
3218         struct sockaddr *addr = p;
3219
3220         if (!is_valid_ether_addr(addr->sa_data))
3221                 return -EADDRNOTAVAIL;
3222
3223         memcpy(netdev->dev_addr, addr->sa_data, netdev->addr_len);
3224         memcpy(adapter->hw.mac.addr, addr->sa_data, netdev->addr_len);
3225
3226         e1000e_rar_set(&adapter->hw, adapter->hw.mac.addr, 0);
3227
3228         if (adapter->flags & FLAG_RESET_OVERWRITES_LAA) {
3229                 /* activate the work around */
3230                 e1000e_set_laa_state_82571(&adapter->hw, 1);
3231
3232                 /*
3233                  * Hold a copy of the LAA in RAR[14] This is done so that
3234                  * between the time RAR[0] gets clobbered  and the time it
3235                  * gets fixed (in e1000_watchdog), the actual LAA is in one
3236                  * of the RARs and no incoming packets directed to this port
3237                  * are dropped. Eventually the LAA will be in RAR[0] and
3238                  * RAR[14]
3239                  */
3240                 e1000e_rar_set(&adapter->hw,
3241                               adapter->hw.mac.addr,
3242                               adapter->hw.mac.rar_entry_count - 1);
3243         }
3244
3245         return 0;
3246 }
3247
3248 /**
3249  * e1000e_update_phy_task - work thread to update phy
3250  * @work: pointer to our work struct
3251  *
3252  * this worker thread exists because we must acquire a
3253  * semaphore to read the phy, which we could msleep while
3254  * waiting for it, and we can't msleep in a timer.
3255  **/
3256 static void e1000e_update_phy_task(struct work_struct *work)
3257 {
3258         struct e1000_adapter *adapter = container_of(work,
3259                                         struct e1000_adapter, update_phy_task);
3260         e1000_get_phy_info(&adapter->hw);
3261 }
3262
3263 /*
3264  * Need to wait a few seconds after link up to get diagnostic information from
3265  * the phy
3266  */
3267 static void e1000_update_phy_info(unsigned long data)
3268 {
3269         struct e1000_adapter *adapter = (struct e1000_adapter *) data;
3270         schedule_work(&adapter->update_phy_task);
3271 }
3272
3273 /**
3274  * e1000e_update_stats - Update the board statistics counters
3275  * @adapter: board private structure
3276  **/
3277 void e1000e_update_stats(struct e1000_adapter *adapter)
3278 {
3279         struct net_device *netdev = adapter->netdev;
3280         struct e1000_hw *hw = &adapter->hw;
3281         struct pci_dev *pdev = adapter->pdev;
3282         u16 phy_data;
3283
3284         /*
3285          * Prevent stats update while adapter is being reset, or if the pci
3286          * connection is down.
3287          */
3288         if (adapter->link_speed == 0)
3289                 return;
3290         if (pci_channel_offline(pdev))
3291                 return;
3292
3293         adapter->stats.crcerrs += er32(CRCERRS);
3294         adapter->stats.gprc += er32(GPRC);
3295         adapter->stats.gorc += er32(GORCL);
3296         er32(GORCH); /* Clear gorc */
3297         adapter->stats.bprc += er32(BPRC);
3298         adapter->stats.mprc += er32(MPRC);
3299         adapter->stats.roc += er32(ROC);
3300
3301         adapter->stats.mpc += er32(MPC);
3302         if ((hw->phy.type == e1000_phy_82578) ||
3303             (hw->phy.type == e1000_phy_82577)) {
3304                 e1e_rphy(hw, HV_SCC_UPPER, &phy_data);
3305                 if (!e1e_rphy(hw, HV_SCC_LOWER, &phy_data))
3306                         adapter->stats.scc += phy_data;
3307
3308                 e1e_rphy(hw, HV_ECOL_UPPER, &phy_data);
3309                 if (!e1e_rphy(hw, HV_ECOL_LOWER, &phy_data))
3310                         adapter->stats.ecol += phy_data;
3311
3312                 e1e_rphy(hw, HV_MCC_UPPER, &phy_data);
3313                 if (!e1e_rphy(hw, HV_MCC_LOWER, &phy_data))
3314                         adapter->stats.mcc += phy_data;
3315
3316                 e1e_rphy(hw, HV_LATECOL_UPPER, &phy_data);
3317                 if (!e1e_rphy(hw, HV_LATECOL_LOWER, &phy_data))
3318                         adapter->stats.latecol += phy_data;
3319
3320                 e1e_rphy(hw, HV_DC_UPPER, &phy_data);
3321                 if (!e1e_rphy(hw, HV_DC_LOWER, &phy_data))
3322                         adapter->stats.dc += phy_data;
3323         } else {
3324                 adapter->stats.scc += er32(SCC);
3325                 adapter->stats.ecol += er32(ECOL);
3326                 adapter->stats.mcc += er32(MCC);
3327                 adapter->stats.latecol += er32(LATECOL);
3328                 adapter->stats.dc += er32(DC);
3329         }
3330         adapter->stats.xonrxc += er32(XONRXC);
3331         adapter->stats.xontxc += er32(XONTXC);
3332         adapter->stats.xoffrxc += er32(XOFFRXC);
3333         adapter->stats.xofftxc += er32(XOFFTXC);
3334         adapter->stats.gptc += er32(GPTC);
3335         adapter->stats.gotc += er32(GOTCL);
3336         er32(GOTCH); /* Clear gotc */
3337         adapter->stats.rnbc += er32(RNBC);
3338         adapter->stats.ruc += er32(RUC);
3339
3340         adapter->stats.mptc += er32(MPTC);
3341         adapter->stats.bptc += er32(BPTC);
3342
3343         /* used for adaptive IFS */
3344
3345         hw->mac.tx_packet_delta = er32(TPT);
3346         adapter->stats.tpt += hw->mac.tx_packet_delta;
3347         if ((hw->phy.type == e1000_phy_82578) ||
3348             (hw->phy.type == e1000_phy_82577)) {
3349                 e1e_rphy(hw, HV_COLC_UPPER, &phy_data);
3350                 if (!e1e_rphy(hw, HV_COLC_LOWER, &phy_data))
3351                         hw->mac.collision_delta = phy_data;
3352         } else {
3353                 hw->mac.collision_delta = er32(COLC);
3354         }
3355         adapter->stats.colc += hw->mac.collision_delta;
3356
3357         adapter->stats.algnerrc += er32(ALGNERRC);
3358         adapter->stats.rxerrc += er32(RXERRC);
3359         if ((hw->phy.type == e1000_phy_82578) ||
3360             (hw->phy.type == e1000_phy_82577)) {
3361                 e1e_rphy(hw, HV_TNCRS_UPPER, &phy_data);
3362                 if (!e1e_rphy(hw, HV_TNCRS_LOWER, &phy_data))
3363                         adapter->stats.tncrs += phy_data;
3364         } else {
3365                 if ((hw->mac.type != e1000_82574) &&
3366                     (hw->mac.type != e1000_82583))
3367                         adapter->stats.tncrs += er32(TNCRS);
3368         }
3369         adapter->stats.cexterr += er32(CEXTERR);
3370         adapter->stats.tsctc += er32(TSCTC);
3371         adapter->stats.tsctfc += er32(TSCTFC);
3372
3373         /* Fill out the OS statistics structure */
3374         netdev->stats.multicast = adapter->stats.mprc;
3375         netdev->stats.collisions = adapter->stats.colc;
3376
3377         /* Rx Errors */
3378
3379         /*
3380          * RLEC on some newer hardware can be incorrect so build
3381          * our own version based on RUC and ROC
3382          */
3383         netdev->stats.rx_errors = adapter->stats.rxerrc +
3384                 adapter->stats.crcerrs + adapter->stats.algnerrc +
3385                 adapter->stats.ruc + adapter->stats.roc +
3386                 adapter->stats.cexterr;
3387         netdev->stats.rx_length_errors = adapter->stats.ruc +
3388                                               adapter->stats.roc;
3389         netdev->stats.rx_crc_errors = adapter->stats.crcerrs;
3390         netdev->stats.rx_frame_errors = adapter->stats.algnerrc;
3391         netdev->stats.rx_missed_errors = adapter->stats.mpc;
3392
3393         /* Tx Errors */
3394         netdev->stats.tx_errors = adapter->stats.ecol +
3395                                        adapter->stats.latecol;
3396         netdev->stats.tx_aborted_errors = adapter->stats.ecol;
3397         netdev->stats.tx_window_errors = adapter->stats.latecol;
3398         netdev->stats.tx_carrier_errors = adapter->stats.tncrs;
3399
3400         /* Tx Dropped needs to be maintained elsewhere */
3401
3402         /* Management Stats */
3403         adapter->stats.mgptc += er32(MGTPTC);
3404         adapter->stats.mgprc += er32(MGTPRC);
3405         adapter->stats.mgpdc += er32(MGTPDC);
3406 }
3407
3408 /**
3409  * e1000_phy_read_status - Update the PHY register status snapshot
3410  * @adapter: board private structure
3411  **/
3412 static void e1000_phy_read_status(struct e1000_adapter *adapter)
3413 {
3414         struct e1000_hw *hw = &adapter->hw;
3415         struct e1000_phy_regs *phy = &adapter->phy_regs;
3416         int ret_val;
3417
3418         if ((er32(STATUS) & E1000_STATUS_LU) &&
3419             (adapter->hw.phy.media_type == e1000_media_type_copper)) {
3420                 ret_val  = e1e_rphy(hw, PHY_CONTROL, &phy->bmcr);
3421                 ret_val |= e1e_rphy(hw, PHY_STATUS, &phy->bmsr);
3422                 ret_val |= e1e_rphy(hw, PHY_AUTONEG_ADV, &phy->advertise);
3423                 ret_val |= e1e_rphy(hw, PHY_LP_ABILITY, &phy->lpa);
3424                 ret_val |= e1e_rphy(hw, PHY_AUTONEG_EXP, &phy->expansion);
3425                 ret_val |= e1e_rphy(hw, PHY_1000T_CTRL, &phy->ctrl1000);
3426                 ret_val |= e1e_rphy(hw, PHY_1000T_STATUS, &phy->stat1000);
3427                 ret_val |= e1e_rphy(hw, PHY_EXT_STATUS, &phy->estatus);
3428                 if (ret_val)
3429                         e_warn("Error reading PHY register\n");
3430         } else {
3431                 /*
3432                  * Do not read PHY registers if link is not up
3433                  * Set values to typical power-on defaults
3434                  */
3435                 phy->bmcr = (BMCR_SPEED1000 | BMCR_ANENABLE | BMCR_FULLDPLX);
3436                 phy->bmsr = (BMSR_100FULL | BMSR_100HALF | BMSR_10FULL |
3437                              BMSR_10HALF | BMSR_ESTATEN | BMSR_ANEGCAPABLE |
3438                              BMSR_ERCAP);
3439                 phy->advertise = (ADVERTISE_PAUSE_ASYM | ADVERTISE_PAUSE_CAP |
3440                                   ADVERTISE_ALL | ADVERTISE_CSMA);
3441                 phy->lpa = 0;
3442                 phy->expansion = EXPANSION_ENABLENPAGE;
3443                 phy->ctrl1000 = ADVERTISE_1000FULL;
3444                 phy->stat1000 = 0;
3445                 phy->estatus = (ESTATUS_1000_TFULL | ESTATUS_1000_THALF);
3446         }
3447 }
3448
3449 static void e1000_print_link_info(struct e1000_adapter *adapter)
3450 {
3451         struct e1000_hw *hw = &adapter->hw;
3452         u32 ctrl = er32(CTRL);
3453
3454         /* Link status message must follow this format for user tools */
3455         printk(KERN_INFO "e1000e: %s NIC Link is Up %d Mbps %s, "
3456                "Flow Control: %s\n",
3457                adapter->netdev->name,
3458                adapter->link_speed,
3459                (adapter->link_duplex == FULL_DUPLEX) ?
3460                                 "Full Duplex" : "Half Duplex",
3461                ((ctrl & E1000_CTRL_TFCE) && (ctrl & E1000_CTRL_RFCE)) ?
3462                                 "RX/TX" :
3463                ((ctrl & E1000_CTRL_RFCE) ? "RX" :
3464                ((ctrl & E1000_CTRL_TFCE) ? "TX" : "None" )));
3465 }
3466
3467 bool e1000e_has_link(struct e1000_adapter *adapter)
3468 {
3469         struct e1000_hw *hw = &adapter->hw;
3470         bool link_active = 0;
3471         s32 ret_val = 0;
3472
3473         /*
3474          * get_link_status is set on LSC (link status) interrupt or
3475          * Rx sequence error interrupt.  get_link_status will stay
3476          * false until the check_for_link establishes link
3477          * for copper adapters ONLY
3478          */
3479         switch (hw->phy.media_type) {
3480         case e1000_media_type_copper:
3481                 if (hw->mac.get_link_status) {
3482                         ret_val = hw->mac.ops.check_for_link(hw);
3483                         link_active = !hw->mac.get_link_status;
3484                 } else {
3485                         link_active = 1;
3486                 }
3487                 break;
3488         case e1000_media_type_fiber:
3489                 ret_val = hw->mac.ops.check_for_link(hw);
3490                 link_active = !!(er32(STATUS) & E1000_STATUS_LU);
3491                 break;
3492         case e1000_media_type_internal_serdes:
3493                 ret_val = hw->mac.ops.check_for_link(hw);
3494                 link_active = adapter->hw.mac.serdes_has_link;
3495                 break;
3496         default:
3497         case e1000_media_type_unknown:
3498                 break;
3499         }
3500
3501         if ((ret_val == E1000_ERR_PHY) && (hw->phy.type == e1000_phy_igp_3) &&
3502             (er32(CTRL) & E1000_PHY_CTRL_GBE_DISABLE)) {
3503                 /* See e1000_kmrn_lock_loss_workaround_ich8lan() */
3504                 e_info("Gigabit has been disabled, downgrading speed\n");
3505         }
3506
3507         return link_active;
3508 }
3509
3510 static void e1000e_enable_receives(struct e1000_adapter *adapter)
3511 {
3512         /* make sure the receive unit is started */
3513         if ((adapter->flags & FLAG_RX_NEEDS_RESTART) &&
3514             (adapter->flags & FLAG_RX_RESTART_NOW)) {
3515                 struct e1000_hw *hw = &adapter->hw;
3516                 u32 rctl = er32(RCTL);
3517                 ew32(RCTL, rctl | E1000_RCTL_EN);
3518                 adapter->flags &= ~FLAG_RX_RESTART_NOW;
3519         }
3520 }
3521
3522 /**
3523  * e1000_watchdog - Timer Call-back
3524  * @data: pointer to adapter cast into an unsigned long
3525  **/
3526 static void e1000_watchdog(unsigned long data)
3527 {
3528         struct e1000_adapter *adapter = (struct e1000_adapter *) data;
3529
3530         /* Do the rest outside of interrupt context */
3531         schedule_work(&adapter->watchdog_task);
3532
3533         /* TODO: make this use queue_delayed_work() */
3534 }
3535
3536 static void e1000_watchdog_task(struct work_struct *work)
3537 {
3538         struct e1000_adapter *adapter = container_of(work,
3539                                         struct e1000_adapter, watchdog_task);
3540         struct net_device *netdev = adapter->netdev;
3541         struct e1000_mac_info *mac = &adapter->hw.mac;
3542         struct e1000_phy_info *phy = &adapter->hw.phy;
3543         struct e1000_ring *tx_ring = adapter->tx_ring;
3544         struct e1000_hw *hw = &adapter->hw;
3545         u32 link, tctl;
3546         int tx_pending = 0;
3547
3548         link = e1000e_has_link(adapter);
3549         if ((netif_carrier_ok(netdev)) && link) {
3550                 e1000e_enable_receives(adapter);
3551                 goto link_up;
3552         }
3553
3554         if ((e1000e_enable_tx_pkt_filtering(hw)) &&
3555             (adapter->mng_vlan_id != adapter->hw.mng_cookie.vlan_id))
3556                 e1000_update_mng_vlan(adapter);
3557
3558         if (link) {
3559                 if (!netif_carrier_ok(netdev)) {
3560                         bool txb2b = 1;
3561                         /* update snapshot of PHY registers on LSC */
3562                         e1000_phy_read_status(adapter);
3563                         mac->ops.get_link_up_info(&adapter->hw,
3564                                                    &adapter->link_speed,
3565                                                    &adapter->link_duplex);
3566                         e1000_print_link_info(adapter);
3567                         /*
3568                          * On supported PHYs, check for duplex mismatch only
3569                          * if link has autonegotiated at 10/100 half
3570                          */
3571                         if ((hw->phy.type == e1000_phy_igp_3 ||
3572                              hw->phy.type == e1000_phy_bm) &&
3573                             (hw->mac.autoneg == true) &&
3574                             (adapter->link_speed == SPEED_10 ||
3575                              adapter->link_speed == SPEED_100) &&
3576                             (adapter->link_duplex == HALF_DUPLEX)) {
3577                                 u16 autoneg_exp;
3578
3579                                 e1e_rphy(hw, PHY_AUTONEG_EXP, &autoneg_exp);
3580
3581                                 if (!(autoneg_exp & NWAY_ER_LP_NWAY_CAPS))
3582                                         e_info("Autonegotiated half duplex but"
3583                                                " link partner cannot autoneg. "
3584                                                " Try forcing full duplex if "
3585                                                "link gets many collisions.\n");
3586                         }
3587
3588                         /* adjust timeout factor according to speed/duplex */
3589                         adapter->tx_timeout_factor = 1;
3590                         switch (adapter->link_speed) {
3591                         case SPEED_10:
3592                                 txb2b = 0;
3593                                 adapter->tx_timeout_factor = 16;
3594                                 break;
3595                         case SPEED_100:
3596                                 txb2b = 0;
3597                                 adapter->tx_timeout_factor = 10;
3598                                 break;
3599                         }
3600
3601                         /*
3602                          * workaround: re-program speed mode bit after
3603                          * link-up event
3604                          */
3605                         if ((adapter->flags & FLAG_TARC_SPEED_MODE_BIT) &&
3606                             !txb2b) {
3607                                 u32 tarc0;
3608                                 tarc0 = er32(TARC(0));
3609                                 tarc0 &= ~SPEED_MODE_BIT;
3610                                 ew32(TARC(0), tarc0);
3611                         }
3612
3613                         /*
3614                          * disable TSO for pcie and 10/100 speeds, to avoid
3615                          * some hardware issues
3616                          */
3617                         if (!(adapter->flags & FLAG_TSO_FORCE)) {
3618                                 switch (adapter->link_speed) {
3619                                 case SPEED_10:
3620                                 case SPEED_100:
3621                                         e_info("10/100 speed: disabling TSO\n");
3622                                         netdev->features &= ~NETIF_F_TSO;
3623                                         netdev->features &= ~NETIF_F_TSO6;
3624                                         break;
3625                                 case SPEED_1000:
3626                                         netdev->features |= NETIF_F_TSO;
3627                                         netdev->features |= NETIF_F_TSO6;
3628                                         break;
3629                                 default:
3630                                         /* oops */
3631                                         break;
3632                                 }
3633                         }
3634
3635                         /*
3636                          * enable transmits in the hardware, need to do this
3637                          * after setting TARC(0)
3638                          */
3639                         tctl = er32(TCTL);
3640                         tctl |= E1000_TCTL_EN;
3641                         ew32(TCTL, tctl);
3642
3643                         /*
3644                          * Perform any post-link-up configuration before
3645                          * reporting link up.
3646                          */
3647                         if (phy->ops.cfg_on_link_up)
3648                                 phy->ops.cfg_on_link_up(hw);
3649
3650                         netif_carrier_on(netdev);
3651
3652                         if (!test_bit(__E1000_DOWN, &adapter->state))
3653                                 mod_timer(&adapter->phy_info_timer,
3654                                           round_jiffies(jiffies + 2 * HZ));
3655                 }
3656         } else {
3657                 if (netif_carrier_ok(netdev)) {
3658                         adapter->link_speed = 0;
3659                         adapter->link_duplex = 0;
3660                         /* Link status message must follow this format */
3661                         printk(KERN_INFO "e1000e: %s NIC Link is Down\n",
3662                                adapter->netdev->name);
3663                         netif_carrier_off(netdev);
3664                         if (!test_bit(__E1000_DOWN, &adapter->state))
3665                                 mod_timer(&adapter->phy_info_timer,
3666                                           round_jiffies(jiffies + 2 * HZ));
3667
3668                         if (adapter->flags & FLAG_RX_NEEDS_RESTART)
3669                                 schedule_work(&adapter->reset_task);
3670                 }
3671         }
3672
3673 link_up:
3674         e1000e_update_stats(adapter);
3675
3676         mac->tx_packet_delta = adapter->stats.tpt - adapter->tpt_old;
3677         adapter->tpt_old = adapter->stats.tpt;
3678         mac->collision_delta = adapter->stats.colc - adapter->colc_old;
3679         adapter->colc_old = adapter->stats.colc;
3680
3681         adapter->gorc = adapter->stats.gorc - adapter->gorc_old;
3682         adapter->gorc_old = adapter->stats.gorc;
3683         adapter->gotc = adapter->stats.gotc - adapter->gotc_old;
3684         adapter->gotc_old = adapter->stats.gotc;
3685
3686         e1000e_update_adaptive(&adapter->hw);
3687
3688         if (!netif_carrier_ok(netdev)) {
3689                 tx_pending = (e1000_desc_unused(tx_ring) + 1 <
3690                                tx_ring->count);
3691                 if (tx_pending) {
3692                         /*
3693                          * We've lost link, so the controller stops DMA,
3694                          * but we've got queued Tx work that's never going
3695                          * to get done, so reset controller to flush Tx.
3696                          * (Do the reset outside of interrupt context).
3697                          */
3698                         adapter->tx_timeout_count++;
3699                         schedule_work(&adapter->reset_task);
3700                         /* return immediately since reset is imminent */
3701                         return;
3702                 }
3703         }
3704
3705         /* Cause software interrupt to ensure Rx ring is cleaned */
3706         if (adapter->msix_entries)
3707                 ew32(ICS, adapter->rx_ring->ims_val);
3708         else
3709                 ew32(ICS, E1000_ICS_RXDMT0);
3710
3711         /* Force detection of hung controller every watchdog period */
3712         adapter->detect_tx_hung = 1;
3713
3714         /*
3715          * With 82571 controllers, LAA may be overwritten due to controller
3716          * reset from the other port. Set the appropriate LAA in RAR[0]
3717          */
3718         if (e1000e_get_laa_state_82571(hw))
3719                 e1000e_rar_set(hw, adapter->hw.mac.addr, 0);
3720
3721         /* Reset the timer */
3722         if (!test_bit(__E1000_DOWN, &adapter->state))
3723                 mod_timer(&adapter->watchdog_timer,
3724                           round_jiffies(jiffies + 2 * HZ));
3725 }
3726
3727 #define E1000_TX_FLAGS_CSUM             0x00000001
3728 #define E1000_TX_FLAGS_VLAN             0x00000002
3729 #define E1000_TX_FLAGS_TSO              0x00000004
3730 #define E1000_TX_FLAGS_IPV4             0x00000008
3731 #define E1000_TX_FLAGS_VLAN_MASK        0xffff0000
3732 #define E1000_TX_FLAGS_VLAN_SHIFT       16
3733
3734 static int e1000_tso(struct e1000_adapter *adapter,
3735                      struct sk_buff *skb)
3736 {
3737         struct e1000_ring *tx_ring = adapter->tx_ring;
3738         struct e1000_context_desc *context_desc;
3739         struct e1000_buffer *buffer_info;
3740         unsigned int i;
3741         u32 cmd_length = 0;
3742         u16 ipcse = 0, tucse, mss;
3743         u8 ipcss, ipcso, tucss, tucso, hdr_len;
3744         int err;
3745
3746         if (!skb_is_gso(skb))
3747                 return 0;
3748
3749         if (skb_header_cloned(skb)) {
3750                 err = pskb_expand_head(skb, 0, 0, GFP_ATOMIC);
3751                 if (err)
3752                         return err;
3753         }
3754
3755         hdr_len = skb_transport_offset(skb) + tcp_hdrlen(skb);
3756         mss = skb_shinfo(skb)->gso_size;
3757         if (skb->protocol == htons(ETH_P_IP)) {
3758                 struct iphdr *iph = ip_hdr(skb);
3759                 iph->tot_len = 0;
3760                 iph->check = 0;
3761                 tcp_hdr(skb)->check = ~csum_tcpudp_magic(iph->saddr, iph->daddr,
3762                                                          0, IPPROTO_TCP, 0);
3763                 cmd_length = E1000_TXD_CMD_IP;
3764                 ipcse = skb_transport_offset(skb) - 1;
3765         } else if (skb_is_gso_v6(skb)) {
3766                 ipv6_hdr(skb)->payload_len = 0;
3767                 tcp_hdr(skb)->check = ~csum_ipv6_magic(&ipv6_hdr(skb)->saddr,
3768                                                        &ipv6_hdr(skb)->daddr,
3769                                                        0, IPPROTO_TCP, 0);
3770                 ipcse = 0;
3771         }
3772         ipcss = skb_network_offset(skb);
3773         ipcso = (void *)&(ip_hdr(skb)->check) - (void *)skb->data;
3774         tucss = skb_transport_offset(skb);
3775         tucso = (void *)&(tcp_hdr(skb)->check) - (void *)skb->data;
3776         tucse = 0;
3777
3778         cmd_length |= (E1000_TXD_CMD_DEXT | E1000_TXD_CMD_TSE |
3779                        E1000_TXD_CMD_TCP | (skb->len - (hdr_len)));
3780
3781         i = tx_ring->next_to_use;
3782         context_desc = E1000_CONTEXT_DESC(*tx_ring, i);
3783         buffer_info = &tx_ring->buffer_info[i];
3784
3785         context_desc->lower_setup.ip_fields.ipcss  = ipcss;
3786         context_desc->lower_setup.ip_fields.ipcso  = ipcso;
3787         context_desc->lower_setup.ip_fields.ipcse  = cpu_to_le16(ipcse);
3788         context_desc->upper_setup.tcp_fields.tucss = tucss;
3789         context_desc->upper_setup.tcp_fields.tucso = tucso;
3790         context_desc->upper_setup.tcp_fields.tucse = cpu_to_le16(tucse);
3791         context_desc->tcp_seg_setup.fields.mss     = cpu_to_le16(mss);
3792         context_desc->tcp_seg_setup.fields.hdr_len = hdr_len;
3793         context_desc->cmd_and_length = cpu_to_le32(cmd_length);
3794
3795         buffer_info->time_stamp = jiffies;
3796         buffer_info->next_to_watch = i;
3797
3798         i++;
3799         if (i == tx_ring->count)
3800                 i = 0;
3801         tx_ring->next_to_use = i;
3802
3803         return 1;
3804 }
3805
3806 static bool e1000_tx_csum(struct e1000_adapter *adapter, struct sk_buff *skb)
3807 {
3808         struct e1000_ring *tx_ring = adapter->tx_ring;
3809         struct e1000_context_desc *context_desc;
3810         struct e1000_buffer *buffer_info;
3811         unsigned int i;
3812         u8 css;
3813         u32 cmd_len = E1000_TXD_CMD_DEXT;
3814         __be16 protocol;
3815
3816         if (skb->ip_summed != CHECKSUM_PARTIAL)
3817                 return 0;
3818
3819         if (skb->protocol == cpu_to_be16(ETH_P_8021Q))
3820                 protocol = vlan_eth_hdr(skb)->h_vlan_encapsulated_proto;
3821         else
3822                 protocol = skb->protocol;
3823
3824         switch (protocol) {
3825         case cpu_to_be16(ETH_P_IP):
3826                 if (ip_hdr(skb)->protocol == IPPROTO_TCP)
3827                         cmd_len |= E1000_TXD_CMD_TCP;
3828                 break;
3829         case cpu_to_be16(ETH_P_IPV6):
3830                 /* XXX not handling all IPV6 headers */
3831                 if (ipv6_hdr(skb)->nexthdr == IPPROTO_TCP)
3832                         cmd_len |= E1000_TXD_CMD_TCP;
3833                 break;
3834         default:
3835                 if (unlikely(net_ratelimit()))
3836                         e_warn("checksum_partial proto=%x!\n",
3837                                be16_to_cpu(protocol));
3838                 break;
3839         }
3840
3841         css = skb_transport_offset(skb);
3842
3843         i = tx_ring->next_to_use;
3844         buffer_info = &tx_ring->buffer_info[i];
3845         context_desc = E1000_CONTEXT_DESC(*tx_ring, i);
3846
3847         context_desc->lower_setup.ip_config = 0;
3848         context_desc->upper_setup.tcp_fields.tucss = css;
3849         context_desc->upper_setup.tcp_fields.tucso =
3850                                 css + skb->csum_offset;
3851         context_desc->upper_setup.tcp_fields.tucse = 0;
3852         context_desc->tcp_seg_setup.data = 0;
3853         context_desc->cmd_and_length = cpu_to_le32(cmd_len);
3854
3855         buffer_info->time_stamp = jiffies;
3856         buffer_info->next_to_watch = i;
3857
3858         i++;
3859         if (i == tx_ring->count)
3860                 i = 0;
3861         tx_ring->next_to_use = i;
3862
3863         return 1;
3864 }
3865
3866 #define E1000_MAX_PER_TXD       8192
3867 #define E1000_MAX_TXD_PWR       12
3868
3869 static int e1000_tx_map(struct e1000_adapter *adapter,
3870                         struct sk_buff *skb, unsigned int first,
3871                         unsigned int max_per_txd, unsigned int nr_frags,
3872                         unsigned int mss)
3873 {
3874         struct e1000_ring *tx_ring = adapter->tx_ring;
3875         struct pci_dev *pdev = adapter->pdev;
3876         struct e1000_buffer *buffer_info;
3877         unsigned int len = skb_headlen(skb);
3878         unsigned int offset = 0, size, count = 0, i;
3879         unsigned int f;
3880
3881         i = tx_ring->next_to_use;
3882
3883         while (len) {
3884                 buffer_info = &tx_ring->buffer_info[i];
3885                 size = min(len, max_per_txd);
3886
3887                 buffer_info->length = size;
3888                 buffer_info->time_stamp = jiffies;
3889                 buffer_info->next_to_watch = i;
3890                 buffer_info->dma = pci_map_single(pdev, skb->data + offset,
3891                                                   size, PCI_DMA_TODEVICE);
3892                 buffer_info->mapped_as_page = false;
3893                 if (pci_dma_mapping_error(pdev, buffer_info->dma))
3894                         goto dma_error;
3895
3896                 len -= size;
3897                 offset += size;
3898                 count++;
3899
3900                 if (len) {
3901                         i++;
3902                         if (i == tx_ring->count)
3903                                 i = 0;
3904                 }
3905         }
3906
3907         for (f = 0; f < nr_frags; f++) {
3908                 struct skb_frag_struct *frag;
3909
3910                 frag = &skb_shinfo(skb)->frags[f];
3911                 len = frag->size;
3912                 offset = frag->page_offset;
3913
3914                 while (len) {
3915                         i++;
3916                         if (i == tx_ring->count)
3917                                 i = 0;
3918
3919                         buffer_info = &tx_ring->buffer_info[i];
3920                         size = min(len, max_per_txd);
3921
3922                         buffer_info->length = size;
3923                         buffer_info->time_stamp = jiffies;
3924                         buffer_info->next_to_watch = i;
3925                         buffer_info->dma = pci_map_page(pdev, frag->page,
3926                                                         offset, size,
3927                                                         PCI_DMA_TODEVICE);
3928                         buffer_info->mapped_as_page = true;
3929                         if (pci_dma_mapping_error(pdev, buffer_info->dma))
3930                                 goto dma_error;
3931
3932                         len -= size;
3933                         offset += size;
3934                         count++;
3935                 }
3936         }
3937
3938         tx_ring->buffer_info[i].skb = skb;
3939         tx_ring->buffer_info[first].next_to_watch = i;
3940
3941         return count;
3942
3943 dma_error:
3944         dev_err(&pdev->dev, "TX DMA map failed\n");
3945         buffer_info->dma = 0;
3946         if (count)
3947                 count--;
3948
3949         while (count--) {
3950                 if (i==0)
3951                         i += tx_ring->count;
3952                 i--;
3953                 buffer_info = &tx_ring->buffer_info[i];
3954                 e1000_put_txbuf(adapter, buffer_info);;
3955         }
3956
3957         return 0;
3958 }
3959
3960 static void e1000_tx_queue(struct e1000_adapter *adapter,
3961                            int tx_flags, int count)
3962 {
3963         struct e1000_ring *tx_ring = adapter->tx_ring;
3964         struct e1000_tx_desc *tx_desc = NULL;
3965         struct e1000_buffer *buffer_info;
3966         u32 txd_upper = 0, txd_lower = E1000_TXD_CMD_IFCS;
3967         unsigned int i;
3968
3969         if (tx_flags & E1000_TX_FLAGS_TSO) {
3970                 txd_lower |= E1000_TXD_CMD_DEXT | E1000_TXD_DTYP_D |
3971                              E1000_TXD_CMD_TSE;
3972                 txd_upper |= E1000_TXD_POPTS_TXSM << 8;
3973
3974                 if (tx_flags & E1000_TX_FLAGS_IPV4)
3975                         txd_upper |= E1000_TXD_POPTS_IXSM << 8;
3976         }
3977
3978         if (tx_flags & E1000_TX_FLAGS_CSUM) {
3979                 txd_lower |= E1000_TXD_CMD_DEXT | E1000_TXD_DTYP_D;
3980                 txd_upper |= E1000_TXD_POPTS_TXSM << 8;
3981         }
3982
3983         if (tx_flags & E1000_TX_FLAGS_VLAN) {
3984                 txd_lower |= E1000_TXD_CMD_VLE;
3985                 txd_upper |= (tx_flags & E1000_TX_FLAGS_VLAN_MASK);
3986         }
3987
3988         i = tx_ring->next_to_use;
3989
3990         while (count--) {
3991                 buffer_info = &tx_ring->buffer_info[i];
3992                 tx_desc = E1000_TX_DESC(*tx_ring, i);
3993                 tx_desc->buffer_addr = cpu_to_le64(buffer_info->dma);
3994                 tx_desc->lower.data =
3995                         cpu_to_le32(txd_lower | buffer_info->length);
3996                 tx_desc->upper.data = cpu_to_le32(txd_upper);
3997
3998                 i++;
3999                 if (i == tx_ring->count)
4000                         i = 0;
4001         }
4002
4003         tx_desc->lower.data |= cpu_to_le32(adapter->txd_cmd);
4004
4005         /*
4006          * Force memory writes to complete before letting h/w
4007          * know there are new descriptors to fetch.  (Only
4008          * applicable for weak-ordered memory model archs,
4009          * such as IA-64).
4010          */
4011         wmb();
4012
4013         tx_ring->next_to_use = i;
4014         writel(i, adapter->hw.hw_addr + tx_ring->tail);
4015         /*
4016          * we need this if more than one processor can write to our tail
4017          * at a time, it synchronizes IO on IA64/Altix systems
4018          */
4019         mmiowb();
4020 }
4021
4022 #define MINIMUM_DHCP_PACKET_SIZE 282
4023 static int e1000_transfer_dhcp_info(struct e1000_adapter *adapter,
4024                                     struct sk_buff *skb)
4025 {
4026         struct e1000_hw *hw =  &adapter->hw;
4027         u16 length, offset;
4028
4029         if (vlan_tx_tag_present(skb)) {
4030                 if (!((vlan_tx_tag_get(skb) == adapter->hw.mng_cookie.vlan_id) &&
4031                     (adapter->hw.mng_cookie.status &
4032                         E1000_MNG_DHCP_COOKIE_STATUS_VLAN)))
4033                         return 0;
4034         }
4035
4036         if (skb->len <= MINIMUM_DHCP_PACKET_SIZE)
4037                 return 0;
4038
4039         if (((struct ethhdr *) skb->data)->h_proto != htons(ETH_P_IP))
4040                 return 0;
4041
4042         {
4043                 const struct iphdr *ip = (struct iphdr *)((u8 *)skb->data+14);
4044                 struct udphdr *udp;
4045
4046                 if (ip->protocol != IPPROTO_UDP)
4047                         return 0;
4048
4049                 udp = (struct udphdr *)((u8 *)ip + (ip->ihl << 2));
4050                 if (ntohs(udp->dest) != 67)
4051                         return 0;
4052
4053                 offset = (u8 *)udp + 8 - skb->data;
4054                 length = skb->len - offset;
4055                 return e1000e_mng_write_dhcp_info(hw, (u8 *)udp + 8, length);
4056         }
4057
4058         return 0;
4059 }
4060
4061 static int __e1000_maybe_stop_tx(struct net_device *netdev, int size)
4062 {
4063         struct e1000_adapter *adapter = netdev_priv(netdev);
4064
4065         netif_stop_queue(netdev);
4066         /*
4067          * Herbert's original patch had:
4068          *  smp_mb__after_netif_stop_queue();
4069          * but since that doesn't exist yet, just open code it.
4070          */
4071         smp_mb();
4072
4073         /*
4074          * We need to check again in a case another CPU has just
4075          * made room available.
4076          */
4077         if (e1000_desc_unused(adapter->tx_ring) < size)
4078                 return -EBUSY;
4079
4080         /* A reprieve! */
4081         netif_start_queue(netdev);
4082         ++adapter->restart_queue;
4083         return 0;
4084 }
4085
4086 static int e1000_maybe_stop_tx(struct net_device *netdev, int size)
4087 {
4088         struct e1000_adapter *adapter = netdev_priv(netdev);
4089
4090         if (e1000_desc_unused(adapter->tx_ring) >= size)
4091                 return 0;
4092         return __e1000_maybe_stop_tx(netdev, size);
4093 }
4094
4095 #define TXD_USE_COUNT(S, X) (((S) >> (X)) + 1 )
4096 static netdev_tx_t e1000_xmit_frame(struct sk_buff *skb,
4097                                     struct net_device *netdev)
4098 {
4099         struct e1000_adapter *adapter = netdev_priv(netdev);
4100         struct e1000_ring *tx_ring = adapter->tx_ring;
4101         unsigned int first;
4102         unsigned int max_per_txd = E1000_MAX_PER_TXD;
4103         unsigned int max_txd_pwr = E1000_MAX_TXD_PWR;
4104         unsigned int tx_flags = 0;
4105         unsigned int len = skb->len - skb->data_len;
4106         unsigned int nr_frags;
4107         unsigned int mss;
4108         int count = 0;
4109         int tso;
4110         unsigned int f;
4111
4112         if (test_bit(__E1000_DOWN, &adapter->state)) {
4113                 dev_kfree_skb_any(skb);
4114                 return NETDEV_TX_OK;
4115         }
4116
4117         if (skb->len <= 0) {
4118                 dev_kfree_skb_any(skb);
4119                 return NETDEV_TX_OK;
4120         }
4121
4122         mss = skb_shinfo(skb)->gso_size;
4123         /*
4124          * The controller does a simple calculation to
4125          * make sure there is enough room in the FIFO before
4126          * initiating the DMA for each buffer.  The calc is:
4127          * 4 = ceil(buffer len/mss).  To make sure we don't
4128          * overrun the FIFO, adjust the max buffer len if mss
4129          * drops.
4130          */
4131         if (mss) {
4132                 u8 hdr_len;
4133                 max_per_txd = min(mss << 2, max_per_txd);
4134                 max_txd_pwr = fls(max_per_txd) - 1;
4135
4136                 /*
4137                  * TSO Workaround for 82571/2/3 Controllers -- if skb->data
4138                  * points to just header, pull a few bytes of payload from
4139                  * frags into skb->data
4140                  */
4141                 hdr_len = skb_transport_offset(skb) + tcp_hdrlen(skb);
4142                 /*
4143                  * we do this workaround for ES2LAN, but it is un-necessary,
4144                  * avoiding it could save a lot of cycles
4145                  */
4146                 if (skb->data_len && (hdr_len == len)) {
4147                         unsigned int pull_size;
4148
4149                         pull_size = min((unsigned int)4, skb->data_len);
4150                         if (!__pskb_pull_tail(skb, pull_size)) {
4151                                 e_err("__pskb_pull_tail failed.\n");
4152                                 dev_kfree_skb_any(skb);
4153                                 return NETDEV_TX_OK;
4154                         }
4155                         len = skb->len - skb->data_len;
4156                 }
4157         }
4158
4159         /* reserve a descriptor for the offload context */
4160         if ((mss) || (skb->ip_summed == CHECKSUM_PARTIAL))
4161                 count++;
4162         count++;
4163
4164         count += TXD_USE_COUNT(len, max_txd_pwr);
4165
4166         nr_frags = skb_shinfo(skb)->nr_frags;
4167         for (f = 0; f < nr_frags; f++)
4168                 count += TXD_USE_COUNT(skb_shinfo(skb)->frags[f].size,
4169                                        max_txd_pwr);
4170
4171         if (adapter->hw.mac.tx_pkt_filtering)
4172                 e1000_transfer_dhcp_info(adapter, skb);
4173
4174         /*
4175          * need: count + 2 desc gap to keep tail from touching
4176          * head, otherwise try next time
4177          */
4178         if (e1000_maybe_stop_tx(netdev, count + 2))
4179                 return NETDEV_TX_BUSY;
4180
4181         if (adapter->vlgrp && vlan_tx_tag_present(skb)) {
4182                 tx_flags |= E1000_TX_FLAGS_VLAN;
4183                 tx_flags |= (vlan_tx_tag_get(skb) << E1000_TX_FLAGS_VLAN_SHIFT);
4184         }
4185
4186         first = tx_ring->next_to_use;
4187
4188         tso = e1000_tso(adapter, skb);
4189         if (tso < 0) {
4190                 dev_kfree_skb_any(skb);
4191                 return NETDEV_TX_OK;
4192         }
4193
4194         if (tso)
4195                 tx_flags |= E1000_TX_FLAGS_TSO;
4196         else if (e1000_tx_csum(adapter, skb))
4197                 tx_flags |= E1000_TX_FLAGS_CSUM;
4198
4199         /*
4200          * Old method was to assume IPv4 packet by default if TSO was enabled.
4201          * 82571 hardware supports TSO capabilities for IPv6 as well...
4202          * no longer assume, we must.
4203          */
4204         if (skb->protocol == htons(ETH_P_IP))
4205                 tx_flags |= E1000_TX_FLAGS_IPV4;
4206
4207         /* if count is 0 then mapping error has occured */
4208         count = e1000_tx_map(adapter, skb, first, max_per_txd, nr_frags, mss);
4209         if (count) {
4210                 e1000_tx_queue(adapter, tx_flags, count);
4211                 /* Make sure there is space in the ring for the next send. */
4212                 e1000_maybe_stop_tx(netdev, MAX_SKB_FRAGS + 2);
4213
4214         } else {
4215                 dev_kfree_skb_any(skb);
4216                 tx_ring->buffer_info[first].time_stamp = 0;
4217                 tx_ring->next_to_use = first;
4218         }
4219
4220         return NETDEV_TX_OK;
4221 }
4222
4223 /**
4224  * e1000_tx_timeout - Respond to a Tx Hang
4225  * @netdev: network interface device structure
4226  **/
4227 static void e1000_tx_timeout(struct net_device *netdev)
4228 {
4229         struct e1000_adapter *adapter = netdev_priv(netdev);
4230
4231         /* Do the reset outside of interrupt context */
4232         adapter->tx_timeout_count++;
4233         schedule_work(&adapter->reset_task);
4234 }
4235
4236 static void e1000_reset_task(struct work_struct *work)
4237 {
4238         struct e1000_adapter *adapter;
4239         adapter = container_of(work, struct e1000_adapter, reset_task);
4240
4241         e1000e_reinit_locked(adapter);
4242 }
4243
4244 /**
4245  * e1000_get_stats - Get System Network Statistics
4246  * @netdev: network interface device structure
4247  *
4248  * Returns the address of the device statistics structure.
4249  * The statistics are actually updated from the timer callback.
4250  **/
4251 static struct net_device_stats *e1000_get_stats(struct net_device *netdev)
4252 {
4253         /* only return the current stats */
4254         return &netdev->stats;
4255 }
4256
4257 /**
4258  * e1000_change_mtu - Change the Maximum Transfer Unit
4259  * @netdev: network interface device structure
4260  * @new_mtu: new value for maximum frame size
4261  *
4262  * Returns 0 on success, negative on failure
4263  **/
4264 static int e1000_change_mtu(struct net_device *netdev, int new_mtu)
4265 {
4266         struct e1000_adapter *adapter = netdev_priv(netdev);
4267         int max_frame = new_mtu + ETH_HLEN + ETH_FCS_LEN;
4268
4269         /* Jumbo frame support */
4270         if ((max_frame > ETH_FRAME_LEN + ETH_FCS_LEN) &&
4271             !(adapter->flags & FLAG_HAS_JUMBO_FRAMES)) {
4272                 e_err("Jumbo Frames not supported.\n");
4273                 return -EINVAL;
4274         }
4275
4276         /* Supported frame sizes */
4277         if ((new_mtu < ETH_ZLEN + ETH_FCS_LEN + VLAN_HLEN) ||
4278             (max_frame > adapter->max_hw_frame_size)) {
4279                 e_err("Unsupported MTU setting\n");
4280                 return -EINVAL;
4281         }
4282
4283         while (test_and_set_bit(__E1000_RESETTING, &adapter->state))
4284                 msleep(1);
4285         /* e1000e_down -> e1000e_reset dependent on max_frame_size & mtu */
4286         adapter->max_frame_size = max_frame;
4287         e_info("changing MTU from %d to %d\n", netdev->mtu, new_mtu);
4288         netdev->mtu = new_mtu;
4289         if (netif_running(netdev))
4290                 e1000e_down(adapter);
4291
4292         /*
4293          * NOTE: netdev_alloc_skb reserves 16 bytes, and typically NET_IP_ALIGN
4294          * means we reserve 2 more, this pushes us to allocate from the next
4295          * larger slab size.
4296          * i.e. RXBUFFER_2048 --> size-4096 slab
4297          * However with the new *_jumbo_rx* routines, jumbo receives will use
4298          * fragmented skbs
4299          */
4300
4301         if (max_frame <= 2048)
4302                 adapter->rx_buffer_len = 2048;
4303         else
4304                 adapter->rx_buffer_len = 4096;
4305
4306         /* adjust allocation if LPE protects us, and we aren't using SBP */
4307         if ((max_frame == ETH_FRAME_LEN + ETH_FCS_LEN) ||
4308              (max_frame == ETH_FRAME_LEN + VLAN_HLEN + ETH_FCS_LEN))
4309                 adapter->rx_buffer_len = ETH_FRAME_LEN + VLAN_HLEN
4310                                          + ETH_FCS_LEN;
4311
4312         if (netif_running(netdev))
4313                 e1000e_up(adapter);
4314         else
4315                 e1000e_reset(adapter);
4316
4317         clear_bit(__E1000_RESETTING, &adapter->state);
4318
4319         return 0;
4320 }
4321
4322 static int e1000_mii_ioctl(struct net_device *netdev, struct ifreq *ifr,
4323                            int cmd)
4324 {
4325         struct e1000_adapter *adapter = netdev_priv(netdev);
4326         struct mii_ioctl_data *data = if_mii(ifr);
4327
4328         if (adapter->hw.phy.media_type != e1000_media_type_copper)
4329                 return -EOPNOTSUPP;
4330
4331         switch (cmd) {
4332         case SIOCGMIIPHY:
4333                 data->phy_id = adapter->hw.phy.addr;
4334                 break;
4335         case SIOCGMIIREG:
4336                 e1000_phy_read_status(adapter);
4337
4338                 switch (data->reg_num & 0x1F) {
4339                 case MII_BMCR:
4340                         data->val_out = adapter->phy_regs.bmcr;
4341                         break;
4342                 case MII_BMSR:
4343                         data->val_out = adapter->phy_regs.bmsr;
4344                         break;
4345                 case MII_PHYSID1:
4346                         data->val_out = (adapter->hw.phy.id >> 16);
4347                         break;
4348                 case MII_PHYSID2:
4349                         data->val_out = (adapter->hw.phy.id & 0xFFFF);
4350                         break;
4351                 case MII_ADVERTISE:
4352                         data->val_out = adapter->phy_regs.advertise;
4353                         break;
4354                 case MII_LPA:
4355                         data->val_out = adapter->phy_regs.lpa;
4356                         break;
4357                 case MII_EXPANSION:
4358                         data->val_out = adapter->phy_regs.expansion;
4359                         break;
4360                 case MII_CTRL1000:
4361                         data->val_out = adapter->phy_regs.ctrl1000;
4362                         break;
4363                 case MII_STAT1000:
4364                         data->val_out = adapter->phy_regs.stat1000;
4365                         break;
4366                 case MII_ESTATUS:
4367                         data->val_out = adapter->phy_regs.estatus;
4368                         break;
4369                 default:
4370                         return -EIO;
4371                 }
4372                 break;
4373         case SIOCSMIIREG:
4374         default:
4375                 return -EOPNOTSUPP;
4376         }
4377         return 0;
4378 }
4379
4380 static int e1000_ioctl(struct net_device *netdev, struct ifreq *ifr, int cmd)
4381 {
4382         switch (cmd) {
4383         case SIOCGMIIPHY:
4384         case SIOCGMIIREG:
4385         case SIOCSMIIREG:
4386                 return e1000_mii_ioctl(netdev, ifr, cmd);
4387         default:
4388                 return -EOPNOTSUPP;
4389         }
4390 }
4391
4392 static int e1000_init_phy_wakeup(struct e1000_adapter *adapter, u32 wufc)
4393 {
4394         struct e1000_hw *hw = &adapter->hw;
4395         u32 i, mac_reg;
4396         u16 phy_reg;
4397         int retval = 0;
4398
4399         /* copy MAC RARs to PHY RARs */
4400         for (i = 0; i < adapter->hw.mac.rar_entry_count; i++) {
4401                 mac_reg = er32(RAL(i));
4402                 e1e_wphy(hw, BM_RAR_L(i), (u16)(mac_reg & 0xFFFF));
4403                 e1e_wphy(hw, BM_RAR_M(i), (u16)((mac_reg >> 16) & 0xFFFF));
4404                 mac_reg = er32(RAH(i));
4405                 e1e_wphy(hw, BM_RAR_H(i), (u16)(mac_reg & 0xFFFF));
4406                 e1e_wphy(hw, BM_RAR_CTRL(i), (u16)((mac_reg >> 16) & 0xFFFF));
4407         }
4408
4409         /* copy MAC MTA to PHY MTA */
4410         for (i = 0; i < adapter->hw.mac.mta_reg_count; i++) {
4411                 mac_reg = E1000_READ_REG_ARRAY(hw, E1000_MTA, i);
4412                 e1e_wphy(hw, BM_MTA(i), (u16)(mac_reg & 0xFFFF));
4413                 e1e_wphy(hw, BM_MTA(i) + 1, (u16)((mac_reg >> 16) & 0xFFFF));
4414         }
4415
4416         /* configure PHY Rx Control register */
4417         e1e_rphy(&adapter->hw, BM_RCTL, &phy_reg);
4418         mac_reg = er32(RCTL);
4419         if (mac_reg & E1000_RCTL_UPE)
4420                 phy_reg |= BM_RCTL_UPE;
4421         if (mac_reg & E1000_RCTL_MPE)
4422                 phy_reg |= BM_RCTL_MPE;
4423         phy_reg &= ~(BM_RCTL_MO_MASK);
4424         if (mac_reg & E1000_RCTL_MO_3)
4425                 phy_reg |= (((mac_reg & E1000_RCTL_MO_3) >> E1000_RCTL_MO_SHIFT)
4426                                 << BM_RCTL_MO_SHIFT);
4427         if (mac_reg & E1000_RCTL_BAM)
4428                 phy_reg |= BM_RCTL_BAM;
4429         if (mac_reg & E1000_RCTL_PMCF)
4430                 phy_reg |= BM_RCTL_PMCF;
4431         mac_reg = er32(CTRL);
4432         if (mac_reg & E1000_CTRL_RFCE)
4433                 phy_reg |= BM_RCTL_RFCE;
4434         e1e_wphy(&adapter->hw, BM_RCTL, phy_reg);
4435
4436         /* enable PHY wakeup in MAC register */
4437         ew32(WUFC, wufc);
4438         ew32(WUC, E1000_WUC_PHY_WAKE | E1000_WUC_PME_EN);
4439
4440         /* configure and enable PHY wakeup in PHY registers */
4441         e1e_wphy(&adapter->hw, BM_WUFC, wufc);
4442         e1e_wphy(&adapter->hw, BM_WUC, E1000_WUC_PME_EN);
4443
4444         /* activate PHY wakeup */
4445         retval = hw->phy.ops.acquire(hw);
4446         if (retval) {
4447                 e_err("Could not acquire PHY\n");
4448                 return retval;
4449         }
4450         e1000e_write_phy_reg_mdic(hw, IGP01E1000_PHY_PAGE_SELECT,
4451                                  (BM_WUC_ENABLE_PAGE << IGP_PAGE_SHIFT));
4452         retval = e1000e_read_phy_reg_mdic(hw, BM_WUC_ENABLE_REG, &phy_reg);
4453         if (retval) {
4454                 e_err("Could not read PHY page 769\n");
4455                 goto out;
4456         }
4457         phy_reg |= BM_WUC_ENABLE_BIT | BM_WUC_HOST_WU_BIT;
4458         retval = e1000e_write_phy_reg_mdic(hw, BM_WUC_ENABLE_REG, phy_reg);
4459         if (retval)
4460                 e_err("Could not set PHY Host Wakeup bit\n");
4461 out:
4462         hw->phy.ops.release(hw);
4463
4464         return retval;
4465 }
4466
4467 static int __e1000_shutdown(struct pci_dev *pdev, bool *enable_wake)
4468 {
4469         struct net_device *netdev = pci_get_drvdata(pdev);
4470         struct e1000_adapter *adapter = netdev_priv(netdev);
4471         struct e1000_hw *hw = &adapter->hw;
4472         u32 ctrl, ctrl_ext, rctl, status;
4473         u32 wufc = adapter->wol;
4474         int retval = 0;
4475
4476         netif_device_detach(netdev);
4477
4478         if (netif_running(netdev)) {
4479                 WARN_ON(test_bit(__E1000_RESETTING, &adapter->state));
4480                 e1000e_down(adapter);
4481                 e1000_free_irq(adapter);
4482         }
4483         e1000e_reset_interrupt_capability(adapter);
4484
4485         retval = pci_save_state(pdev);
4486         if (retval)
4487                 return retval;
4488
4489         status = er32(STATUS);
4490         if (status & E1000_STATUS_LU)
4491                 wufc &= ~E1000_WUFC_LNKC;
4492
4493         if (wufc) {
4494                 e1000_setup_rctl(adapter);
4495                 e1000_set_multi(netdev);
4496
4497                 /* turn on all-multi mode if wake on multicast is enabled */
4498                 if (wufc & E1000_WUFC_MC) {
4499                         rctl = er32(RCTL);
4500                         rctl |= E1000_RCTL_MPE;
4501                         ew32(RCTL, rctl);
4502                 }
4503
4504                 ctrl = er32(CTRL);
4505                 /* advertise wake from D3Cold */
4506                 #define E1000_CTRL_ADVD3WUC 0x00100000
4507                 /* phy power management enable */
4508                 #define E1000_CTRL_EN_PHY_PWR_MGMT 0x00200000
4509                 ctrl |= E1000_CTRL_ADVD3WUC;
4510                 if (!(adapter->flags2 & FLAG2_HAS_PHY_WAKEUP))
4511                         ctrl |= E1000_CTRL_EN_PHY_PWR_MGMT;
4512                 ew32(CTRL, ctrl);
4513
4514                 if (adapter->hw.phy.media_type == e1000_media_type_fiber ||
4515                     adapter->hw.phy.media_type ==
4516                     e1000_media_type_internal_serdes) {
4517                         /* keep the laser running in D3 */
4518                         ctrl_ext = er32(CTRL_EXT);
4519                         ctrl_ext |= E1000_CTRL_EXT_SDP3_DATA;
4520                         ew32(CTRL_EXT, ctrl_ext);
4521                 }
4522
4523                 if (adapter->flags & FLAG_IS_ICH)
4524                         e1000e_disable_gig_wol_ich8lan(&adapter->hw);
4525
4526                 /* Allow time for pending master requests to run */
4527                 e1000e_disable_pcie_master(&adapter->hw);
4528
4529                 if (adapter->flags2 & FLAG2_HAS_PHY_WAKEUP) {
4530                         /* enable wakeup by the PHY */
4531                         retval = e1000_init_phy_wakeup(adapter, wufc);
4532                         if (retval)
4533                                 return retval;
4534                 } else {
4535                         /* enable wakeup by the MAC */
4536                         ew32(WUFC, wufc);
4537                         ew32(WUC, E1000_WUC_PME_EN);
4538                 }
4539         } else {
4540                 ew32(WUC, 0);
4541                 ew32(WUFC, 0);
4542         }
4543
4544         *enable_wake = !!wufc;
4545
4546         /* make sure adapter isn't asleep if manageability is enabled */
4547         if ((adapter->flags & FLAG_MNG_PT_ENABLED) ||
4548             (hw->mac.ops.check_mng_mode(hw)))
4549                 *enable_wake = true;
4550
4551         if (adapter->hw.phy.type == e1000_phy_igp_3)
4552                 e1000e_igp3_phy_powerdown_workaround_ich8lan(&adapter->hw);
4553
4554         /*
4555          * Release control of h/w to f/w.  If f/w is AMT enabled, this
4556          * would have already happened in close and is redundant.
4557          */
4558         e1000_release_hw_control(adapter);
4559
4560         pci_disable_device(pdev);
4561
4562         return 0;
4563 }
4564
4565 static void e1000_power_off(struct pci_dev *pdev, bool sleep, bool wake)
4566 {
4567         if (sleep && wake) {
4568                 pci_prepare_to_sleep(pdev);
4569                 return;
4570         }
4571
4572         pci_wake_from_d3(pdev, wake);
4573         pci_set_power_state(pdev, PCI_D3hot);
4574 }
4575
4576 static void e1000_complete_shutdown(struct pci_dev *pdev, bool sleep,
4577                                     bool wake)
4578 {
4579         struct net_device *netdev = pci_get_drvdata(pdev);
4580         struct e1000_adapter *adapter = netdev_priv(netdev);
4581
4582         /*
4583          * The pci-e switch on some quad port adapters will report a
4584          * correctable error when the MAC transitions from D0 to D3.  To
4585          * prevent this we need to mask off the correctable errors on the
4586          * downstream port of the pci-e switch.
4587          */
4588         if (adapter->flags & FLAG_IS_QUAD_PORT) {
4589                 struct pci_dev *us_dev = pdev->bus->self;
4590                 int pos = pci_find_capability(us_dev, PCI_CAP_ID_EXP);
4591                 u16 devctl;
4592
4593                 pci_read_config_word(us_dev, pos + PCI_EXP_DEVCTL, &devctl);
4594                 pci_write_config_word(us_dev, pos + PCI_EXP_DEVCTL,
4595                                       (devctl & ~PCI_EXP_DEVCTL_CERE));
4596
4597                 e1000_power_off(pdev, sleep, wake);
4598
4599                 pci_write_config_word(us_dev, pos + PCI_EXP_DEVCTL, devctl);
4600         } else {
4601                 e1000_power_off(pdev, sleep, wake);
4602         }
4603 }
4604
4605 static void e1000e_disable_l1aspm(struct pci_dev *pdev)
4606 {
4607         int pos;
4608         u16 val;
4609
4610         /*
4611          * 82573 workaround - disable L1 ASPM on mobile chipsets
4612          *
4613          * L1 ASPM on various mobile (ich7) chipsets do not behave properly
4614          * resulting in lost data or garbage information on the pci-e link
4615          * level. This could result in (false) bad EEPROM checksum errors,
4616          * long ping times (up to 2s) or even a system freeze/hang.
4617          *
4618          * Unfortunately this feature saves about 1W power consumption when
4619          * active.
4620          */
4621         pos = pci_find_capability(pdev, PCI_CAP_ID_EXP);
4622         pci_read_config_word(pdev, pos + PCI_EXP_LNKCTL, &val);
4623         if (val & 0x2) {
4624                 dev_warn(&pdev->dev, "Disabling L1 ASPM\n");
4625                 val &= ~0x2;
4626                 pci_write_config_word(pdev, pos + PCI_EXP_LNKCTL, val);
4627         }
4628 }
4629
4630 #ifdef CONFIG_PM
4631 static int e1000_suspend(struct pci_dev *pdev, pm_message_t state)
4632 {
4633         int retval;
4634         bool wake;
4635
4636         retval = __e1000_shutdown(pdev, &wake);
4637         if (!retval)
4638                 e1000_complete_shutdown(pdev, true, wake);
4639
4640         return retval;
4641 }
4642
4643 static int e1000_resume(struct pci_dev *pdev)
4644 {
4645         struct net_device *netdev = pci_get_drvdata(pdev);
4646         struct e1000_adapter *adapter = netdev_priv(netdev);
4647         struct e1000_hw *hw = &adapter->hw;
4648         u32 err;
4649
4650         pci_set_power_state(pdev, PCI_D0);
4651         pci_restore_state(pdev);
4652         pci_save_state(pdev);
4653         e1000e_disable_l1aspm(pdev);
4654
4655         err = pci_enable_device_mem(pdev);
4656         if (err) {
4657                 dev_err(&pdev->dev,
4658                         "Cannot enable PCI device from suspend\n");
4659                 return err;
4660         }
4661
4662         pci_set_master(pdev);
4663
4664         pci_enable_wake(pdev, PCI_D3hot, 0);
4665         pci_enable_wake(pdev, PCI_D3cold, 0);
4666
4667         e1000e_set_interrupt_capability(adapter);
4668         if (netif_running(netdev)) {
4669                 err = e1000_request_irq(adapter);
4670                 if (err)
4671                         return err;
4672         }
4673
4674         e1000e_power_up_phy(adapter);
4675
4676         /* report the system wakeup cause from S3/S4 */
4677         if (adapter->flags2 & FLAG2_HAS_PHY_WAKEUP) {
4678                 u16 phy_data;
4679
4680                 e1e_rphy(&adapter->hw, BM_WUS, &phy_data);
4681                 if (phy_data) {
4682                         e_info("PHY Wakeup cause - %s\n",
4683                                 phy_data & E1000_WUS_EX ? "Unicast Packet" :
4684                                 phy_data & E1000_WUS_MC ? "Multicast Packet" :
4685                                 phy_data & E1000_WUS_BC ? "Broadcast Packet" :
4686                                 phy_data & E1000_WUS_MAG ? "Magic Packet" :
4687                                 phy_data & E1000_WUS_LNKC ? "Link Status "
4688                                 " Change" : "other");
4689                 }
4690                 e1e_wphy(&adapter->hw, BM_WUS, ~0);
4691         } else {
4692                 u32 wus = er32(WUS);
4693                 if (wus) {
4694                         e_info("MAC Wakeup cause - %s\n",
4695                                 wus & E1000_WUS_EX ? "Unicast Packet" :
4696                                 wus & E1000_WUS_MC ? "Multicast Packet" :
4697                                 wus & E1000_WUS_BC ? "Broadcast Packet" :
4698                                 wus & E1000_WUS_MAG ? "Magic Packet" :
4699                                 wus & E1000_WUS_LNKC ? "Link Status Change" :
4700                                 "other");
4701                 }
4702                 ew32(WUS, ~0);
4703         }
4704
4705         e1000e_reset(adapter);
4706
4707         e1000_init_manageability(adapter);
4708
4709         if (netif_running(netdev))
4710                 e1000e_up(adapter);
4711
4712         netif_device_attach(netdev);
4713
4714         /*
4715          * If the controller has AMT, do not set DRV_LOAD until the interface
4716          * is up.  For all other cases, let the f/w know that the h/w is now
4717          * under the control of the driver.
4718          */
4719         if (!(adapter->flags & FLAG_HAS_AMT))
4720                 e1000_get_hw_control(adapter);
4721
4722         return 0;
4723 }
4724 #endif
4725
4726 static void e1000_shutdown(struct pci_dev *pdev)
4727 {
4728         bool wake = false;
4729
4730         __e1000_shutdown(pdev, &wake);
4731
4732         if (system_state == SYSTEM_POWER_OFF)
4733                 e1000_complete_shutdown(pdev, false, wake);
4734 }
4735
4736 #ifdef CONFIG_NET_POLL_CONTROLLER
4737 /*
4738  * Polling 'interrupt' - used by things like netconsole to send skbs
4739  * without having to re-enable interrupts. It's not called while
4740  * the interrupt routine is executing.
4741  */
4742 static void e1000_netpoll(struct net_device *netdev)
4743 {
4744         struct e1000_adapter *adapter = netdev_priv(netdev);
4745
4746         disable_irq(adapter->pdev->irq);
4747         e1000_intr(adapter->pdev->irq, netdev);
4748
4749         enable_irq(adapter->pdev->irq);
4750 }
4751 #endif
4752
4753 /**
4754  * e1000_io_error_detected - called when PCI error is detected
4755  * @pdev: Pointer to PCI device
4756  * @state: The current pci connection state
4757  *
4758  * This function is called after a PCI bus error affecting
4759  * this device has been detected.
4760  */
4761 static pci_ers_result_t e1000_io_error_detected(struct pci_dev *pdev,
4762                                                 pci_channel_state_t state)
4763 {
4764         struct net_device *netdev = pci_get_drvdata(pdev);
4765         struct e1000_adapter *adapter = netdev_priv(netdev);
4766
4767         netif_device_detach(netdev);
4768
4769         if (state == pci_channel_io_perm_failure)
4770                 return PCI_ERS_RESULT_DISCONNECT;
4771
4772         if (netif_running(netdev))
4773                 e1000e_down(adapter);
4774         pci_disable_device(pdev);
4775
4776         /* Request a slot slot reset. */
4777         return PCI_ERS_RESULT_NEED_RESET;
4778 }
4779
4780 /**
4781  * e1000_io_slot_reset - called after the pci bus has been reset.
4782  * @pdev: Pointer to PCI device
4783  *
4784  * Restart the card from scratch, as if from a cold-boot. Implementation
4785  * resembles the first-half of the e1000_resume routine.
4786  */
4787 static pci_ers_result_t e1000_io_slot_reset(struct pci_dev *pdev)
4788 {
4789         struct net_device *netdev = pci_get_drvdata(pdev);
4790         struct e1000_adapter *adapter = netdev_priv(netdev);
4791         struct e1000_hw *hw = &adapter->hw;
4792         int err;
4793         pci_ers_result_t result;
4794
4795         e1000e_disable_l1aspm(pdev);
4796         err = pci_enable_device_mem(pdev);
4797         if (err) {
4798                 dev_err(&pdev->dev,
4799                         "Cannot re-enable PCI device after reset.\n");
4800                 result = PCI_ERS_RESULT_DISCONNECT;
4801         } else {
4802                 pci_set_master(pdev);
4803                 pci_restore_state(pdev);
4804                 pci_save_state(pdev);
4805
4806                 pci_enable_wake(pdev, PCI_D3hot, 0);
4807                 pci_enable_wake(pdev, PCI_D3cold, 0);
4808
4809                 e1000e_reset(adapter);
4810                 ew32(WUS, ~0);
4811                 result = PCI_ERS_RESULT_RECOVERED;
4812         }
4813
4814         pci_cleanup_aer_uncorrect_error_status(pdev);
4815
4816         return result;
4817 }
4818
4819 /**
4820  * e1000_io_resume - called when traffic can start flowing again.
4821  * @pdev: Pointer to PCI device
4822  *
4823  * This callback is called when the error recovery driver tells us that
4824  * its OK to resume normal operation. Implementation resembles the
4825  * second-half of the e1000_resume routine.
4826  */
4827 static void e1000_io_resume(struct pci_dev *pdev)
4828 {
4829         struct net_device *netdev = pci_get_drvdata(pdev);
4830         struct e1000_adapter *adapter = netdev_priv(netdev);
4831
4832         e1000_init_manageability(adapter);
4833
4834         if (netif_running(netdev)) {
4835                 if (e1000e_up(adapter)) {
4836                         dev_err(&pdev->dev,
4837                                 "can't bring device back up after reset\n");
4838                         return;
4839                 }
4840         }
4841
4842         netif_device_attach(netdev);
4843
4844         /*
4845          * If the controller has AMT, do not set DRV_LOAD until the interface
4846          * is up.  For all other cases, let the f/w know that the h/w is now
4847          * under the control of the driver.
4848          */
4849         if (!(adapter->flags & FLAG_HAS_AMT))
4850                 e1000_get_hw_control(adapter);
4851
4852 }
4853
4854 static void e1000_print_device_info(struct e1000_adapter *adapter)
4855 {
4856         struct e1000_hw *hw = &adapter->hw;
4857         struct net_device *netdev = adapter->netdev;
4858         u32 pba_num;
4859
4860         /* print bus type/speed/width info */
4861         e_info("(PCI Express:2.5GB/s:%s) %pM\n",
4862                /* bus width */
4863                ((hw->bus.width == e1000_bus_width_pcie_x4) ? "Width x4" :
4864                 "Width x1"),
4865                /* MAC address */
4866                netdev->dev_addr);
4867         e_info("Intel(R) PRO/%s Network Connection\n",
4868                (hw->phy.type == e1000_phy_ife) ? "10/100" : "1000");
4869         e1000e_read_pba_num(hw, &pba_num);
4870         e_info("MAC: %d, PHY: %d, PBA No: %06x-%03x\n",
4871                hw->mac.type, hw->phy.type, (pba_num >> 8), (pba_num & 0xff));
4872 }
4873
4874 static void e1000_eeprom_checks(struct e1000_adapter *adapter)
4875 {
4876         struct e1000_hw *hw = &adapter->hw;
4877         int ret_val;
4878         u16 buf = 0;
4879
4880         if (hw->mac.type != e1000_82573)
4881                 return;
4882
4883         ret_val = e1000_read_nvm(hw, NVM_INIT_CONTROL2_REG, 1, &buf);
4884         if (!ret_val && (!(le16_to_cpu(buf) & (1 << 0)))) {
4885                 /* Deep Smart Power Down (DSPD) */
4886                 dev_warn(&adapter->pdev->dev,
4887                          "Warning: detected DSPD enabled in EEPROM\n");
4888         }
4889
4890         ret_val = e1000_read_nvm(hw, NVM_INIT_3GIO_3, 1, &buf);
4891         if (!ret_val && (le16_to_cpu(buf) & (3 << 2))) {
4892                 /* ASPM enable */
4893                 dev_warn(&adapter->pdev->dev,
4894                          "Warning: detected ASPM enabled in EEPROM\n");
4895         }
4896 }
4897
4898 static const struct net_device_ops e1000e_netdev_ops = {
4899         .ndo_open               = e1000_open,
4900         .ndo_stop               = e1000_close,
4901         .ndo_start_xmit         = e1000_xmit_frame,
4902         .ndo_get_stats          = e1000_get_stats,
4903         .ndo_set_multicast_list = e1000_set_multi,
4904         .ndo_set_mac_address    = e1000_set_mac,
4905         .ndo_change_mtu         = e1000_change_mtu,
4906         .ndo_do_ioctl           = e1000_ioctl,
4907         .ndo_tx_timeout         = e1000_tx_timeout,
4908         .ndo_validate_addr      = eth_validate_addr,
4909
4910         .ndo_vlan_rx_register   = e1000_vlan_rx_register,
4911         .ndo_vlan_rx_add_vid    = e1000_vlan_rx_add_vid,
4912         .ndo_vlan_rx_kill_vid   = e1000_vlan_rx_kill_vid,
4913 #ifdef CONFIG_NET_POLL_CONTROLLER
4914         .ndo_poll_controller    = e1000_netpoll,
4915 #endif
4916 };
4917
4918 /**
4919  * e1000_probe - Device Initialization Routine
4920  * @pdev: PCI device information struct
4921  * @ent: entry in e1000_pci_tbl
4922  *
4923  * Returns 0 on success, negative on failure
4924  *
4925  * e1000_probe initializes an adapter identified by a pci_dev structure.
4926  * The OS initialization, configuring of the adapter private structure,
4927  * and a hardware reset occur.
4928  **/
4929 static int __devinit e1000_probe(struct pci_dev *pdev,
4930                                  const struct pci_device_id *ent)
4931 {
4932         struct net_device *netdev;
4933         struct e1000_adapter *adapter;
4934         struct e1000_hw *hw;
4935         const struct e1000_info *ei = e1000_info_tbl[ent->driver_data];
4936         resource_size_t mmio_start, mmio_len;
4937         resource_size_t flash_start, flash_len;
4938
4939         static int cards_found;
4940         int i, err, pci_using_dac;
4941         u16 eeprom_data = 0;
4942         u16 eeprom_apme_mask = E1000_EEPROM_APME;
4943
4944         e1000e_disable_l1aspm(pdev);
4945
4946         err = pci_enable_device_mem(pdev);
4947         if (err)
4948                 return err;
4949
4950         pci_using_dac = 0;
4951         err = pci_set_dma_mask(pdev, DMA_BIT_MASK(64));
4952         if (!err) {
4953                 err = pci_set_consistent_dma_mask(pdev, DMA_BIT_MASK(64));
4954                 if (!err)
4955                         pci_using_dac = 1;
4956         } else {
4957                 err = pci_set_dma_mask(pdev, DMA_BIT_MASK(32));
4958                 if (err) {
4959                         err = pci_set_consistent_dma_mask(pdev,
4960                                                           DMA_BIT_MASK(32));
4961                         if (err) {
4962                                 dev_err(&pdev->dev, "No usable DMA "
4963                                         "configuration, aborting\n");
4964                                 goto err_dma;
4965                         }
4966                 }
4967         }
4968
4969         err = pci_request_selected_regions_exclusive(pdev,
4970                                           pci_select_bars(pdev, IORESOURCE_MEM),
4971                                           e1000e_driver_name);
4972         if (err)
4973                 goto err_pci_reg;
4974
4975         /* AER (Advanced Error Reporting) hooks */
4976         pci_enable_pcie_error_reporting(pdev);
4977
4978         pci_set_master(pdev);
4979         /* PCI config space info */
4980         err = pci_save_state(pdev);
4981         if (err)
4982                 goto err_alloc_etherdev;
4983
4984         err = -ENOMEM;
4985         netdev = alloc_etherdev(sizeof(struct e1000_adapter));
4986         if (!netdev)
4987                 goto err_alloc_etherdev;
4988
4989         SET_NETDEV_DEV(netdev, &pdev->dev);
4990
4991         pci_set_drvdata(pdev, netdev);
4992         adapter = netdev_priv(netdev);
4993         hw = &adapter->hw;
4994         adapter->netdev = netdev;
4995         adapter->pdev = pdev;
4996         adapter->ei = ei;
4997         adapter->pba = ei->pba;
4998         adapter->flags = ei->flags;
4999         adapter->flags2 = ei->flags2;
5000         adapter->hw.adapter = adapter;
5001         adapter->hw.mac.type = ei->mac;
5002         adapter->max_hw_frame_size = ei->max_hw_frame_size;
5003         adapter->msg_enable = (1 << NETIF_MSG_DRV | NETIF_MSG_PROBE) - 1;
5004
5005         mmio_start = pci_resource_start(pdev, 0);
5006         mmio_len = pci_resource_len(pdev, 0);
5007
5008         err = -EIO;
5009         adapter->hw.hw_addr = ioremap(mmio_start, mmio_len);
5010         if (!adapter->hw.hw_addr)
5011                 goto err_ioremap;
5012
5013         if ((adapter->flags & FLAG_HAS_FLASH) &&
5014             (pci_resource_flags(pdev, 1) & IORESOURCE_MEM)) {
5015                 flash_start = pci_resource_start(pdev, 1);
5016                 flash_len = pci_resource_len(pdev, 1);
5017                 adapter->hw.flash_address = ioremap(flash_start, flash_len);
5018                 if (!adapter->hw.flash_address)
5019                         goto err_flashmap;
5020         }
5021
5022         /* construct the net_device struct */
5023         netdev->netdev_ops              = &e1000e_netdev_ops;
5024         e1000e_set_ethtool_ops(netdev);
5025         netdev->watchdog_timeo          = 5 * HZ;
5026         netif_napi_add(netdev, &adapter->napi, e1000_clean, 64);
5027         strncpy(netdev->name, pci_name(pdev), sizeof(netdev->name) - 1);
5028
5029         netdev->mem_start = mmio_start;
5030         netdev->mem_end = mmio_start + mmio_len;
5031
5032         adapter->bd_number = cards_found++;
5033
5034         e1000e_check_options(adapter);
5035
5036         /* setup adapter struct */
5037         err = e1000_sw_init(adapter);
5038         if (err)
5039                 goto err_sw_init;
5040
5041         err = -EIO;
5042
5043         memcpy(&hw->mac.ops, ei->mac_ops, sizeof(hw->mac.ops));
5044         memcpy(&hw->nvm.ops, ei->nvm_ops, sizeof(hw->nvm.ops));
5045         memcpy(&hw->phy.ops, ei->phy_ops, sizeof(hw->phy.ops));
5046
5047         err = ei->get_variants(adapter);
5048         if (err)
5049                 goto err_hw_init;
5050
5051         if ((adapter->flags & FLAG_IS_ICH) &&
5052             (adapter->flags & FLAG_READ_ONLY_NVM))
5053                 e1000e_write_protect_nvm_ich8lan(&adapter->hw);
5054
5055         hw->mac.ops.get_bus_info(&adapter->hw);
5056
5057         adapter->hw.phy.autoneg_wait_to_complete = 0;
5058
5059         /* Copper options */
5060         if (adapter->hw.phy.media_type == e1000_media_type_copper) {
5061                 adapter->hw.phy.mdix = AUTO_ALL_MODES;
5062                 adapter->hw.phy.disable_polarity_correction = 0;
5063                 adapter->hw.phy.ms_type = e1000_ms_hw_default;
5064         }
5065
5066         if (e1000_check_reset_block(&adapter->hw))
5067                 e_info("PHY reset is blocked due to SOL/IDER session.\n");
5068
5069         netdev->features = NETIF_F_SG |
5070                            NETIF_F_HW_CSUM |
5071                            NETIF_F_HW_VLAN_TX |
5072                            NETIF_F_HW_VLAN_RX;
5073
5074         if (adapter->flags & FLAG_HAS_HW_VLAN_FILTER)
5075                 netdev->features |= NETIF_F_HW_VLAN_FILTER;
5076
5077         netdev->features |= NETIF_F_TSO;
5078         netdev->features |= NETIF_F_TSO6;
5079
5080         netdev->vlan_features |= NETIF_F_TSO;
5081         netdev->vlan_features |= NETIF_F_TSO6;
5082         netdev->vlan_features |= NETIF_F_HW_CSUM;
5083         netdev->vlan_features |= NETIF_F_SG;
5084
5085         if (pci_using_dac)
5086                 netdev->features |= NETIF_F_HIGHDMA;
5087
5088         if (e1000e_enable_mng_pass_thru(&adapter->hw))
5089                 adapter->flags |= FLAG_MNG_PT_ENABLED;
5090
5091         /*
5092          * before reading the NVM, reset the controller to
5093          * put the device in a known good starting state
5094          */
5095         adapter->hw.mac.ops.reset_hw(&adapter->hw);
5096
5097         /*
5098          * systems with ASPM and others may see the checksum fail on the first
5099          * attempt. Let's give it a few tries
5100          */
5101         for (i = 0;; i++) {
5102                 if (e1000_validate_nvm_checksum(&adapter->hw) >= 0)
5103                         break;
5104                 if (i == 2) {
5105                         e_err("The NVM Checksum Is Not Valid\n");
5106                         err = -EIO;
5107                         goto err_eeprom;
5108                 }
5109         }
5110
5111         e1000_eeprom_checks(adapter);
5112
5113         /* copy the MAC address */
5114         if (e1000e_read_mac_addr(&adapter->hw))
5115                 e_err("NVM Read Error while reading MAC address\n");
5116
5117         memcpy(netdev->dev_addr, adapter->hw.mac.addr, netdev->addr_len);
5118         memcpy(netdev->perm_addr, adapter->hw.mac.addr, netdev->addr_len);
5119
5120         if (!is_valid_ether_addr(netdev->perm_addr)) {
5121                 e_err("Invalid MAC Address: %pM\n", netdev->perm_addr);
5122                 err = -EIO;
5123                 goto err_eeprom;
5124         }
5125
5126         init_timer(&adapter->watchdog_timer);
5127         adapter->watchdog_timer.function = &e1000_watchdog;
5128         adapter->watchdog_timer.data = (unsigned long) adapter;
5129
5130         init_timer(&adapter->phy_info_timer);
5131         adapter->phy_info_timer.function = &e1000_update_phy_info;
5132         adapter->phy_info_timer.data = (unsigned long) adapter;
5133
5134         INIT_WORK(&adapter->reset_task, e1000_reset_task);
5135         INIT_WORK(&adapter->watchdog_task, e1000_watchdog_task);
5136         INIT_WORK(&adapter->downshift_task, e1000e_downshift_workaround);
5137         INIT_WORK(&adapter->update_phy_task, e1000e_update_phy_task);
5138         INIT_WORK(&adapter->print_hang_task, e1000_print_hw_hang);
5139
5140         /* Initialize link parameters. User can change them with ethtool */
5141         adapter->hw.mac.autoneg = 1;
5142         adapter->fc_autoneg = 1;
5143         adapter->hw.fc.requested_mode = e1000_fc_default;
5144         adapter->hw.fc.current_mode = e1000_fc_default;
5145         adapter->hw.phy.autoneg_advertised = 0x2f;
5146
5147         /* ring size defaults */
5148         adapter->rx_ring->count = 256;
5149         adapter->tx_ring->count = 256;
5150
5151         /*
5152          * Initial Wake on LAN setting - If APM wake is enabled in
5153          * the EEPROM, enable the ACPI Magic Packet filter
5154          */
5155         if (adapter->flags & FLAG_APME_IN_WUC) {
5156                 /* APME bit in EEPROM is mapped to WUC.APME */
5157                 eeprom_data = er32(WUC);
5158                 eeprom_apme_mask = E1000_WUC_APME;
5159                 if (eeprom_data & E1000_WUC_PHY_WAKE)
5160                         adapter->flags2 |= FLAG2_HAS_PHY_WAKEUP;
5161         } else if (adapter->flags & FLAG_APME_IN_CTRL3) {
5162                 if (adapter->flags & FLAG_APME_CHECK_PORT_B &&
5163                     (adapter->hw.bus.func == 1))
5164                         e1000_read_nvm(&adapter->hw,
5165                                 NVM_INIT_CONTROL3_PORT_B, 1, &eeprom_data);
5166                 else
5167                         e1000_read_nvm(&adapter->hw,
5168                                 NVM_INIT_CONTROL3_PORT_A, 1, &eeprom_data);
5169         }
5170
5171         /* fetch WoL from EEPROM */
5172         if (eeprom_data & eeprom_apme_mask)
5173                 adapter->eeprom_wol |= E1000_WUFC_MAG;
5174
5175         /*
5176          * now that we have the eeprom settings, apply the special cases
5177          * where the eeprom may be wrong or the board simply won't support
5178          * wake on lan on a particular port
5179          */
5180         if (!(adapter->flags & FLAG_HAS_WOL))
5181                 adapter->eeprom_wol = 0;
5182
5183         /* initialize the wol settings based on the eeprom settings */
5184         adapter->wol = adapter->eeprom_wol;
5185         device_set_wakeup_enable(&adapter->pdev->dev, adapter->wol);
5186
5187         /* save off EEPROM version number */
5188         e1000_read_nvm(&adapter->hw, 5, 1, &adapter->eeprom_vers);
5189
5190         /* reset the hardware with the new settings */
5191         e1000e_reset(adapter);
5192
5193         /*
5194          * If the controller has AMT, do not set DRV_LOAD until the interface
5195          * is up.  For all other cases, let the f/w know that the h/w is now
5196          * under the control of the driver.
5197          */
5198         if (!(adapter->flags & FLAG_HAS_AMT))
5199                 e1000_get_hw_control(adapter);
5200
5201         strcpy(netdev->name, "eth%d");
5202         err = register_netdev(netdev);
5203         if (err)
5204                 goto err_register;
5205
5206         /* carrier off reporting is important to ethtool even BEFORE open */
5207         netif_carrier_off(netdev);
5208
5209         e1000_print_device_info(adapter);
5210
5211         return 0;
5212
5213 err_register:
5214         if (!(adapter->flags & FLAG_HAS_AMT))
5215                 e1000_release_hw_control(adapter);
5216 err_eeprom:
5217         if (!e1000_check_reset_block(&adapter->hw))
5218                 e1000_phy_hw_reset(&adapter->hw);
5219 err_hw_init:
5220
5221         kfree(adapter->tx_ring);
5222         kfree(adapter->rx_ring);
5223 err_sw_init:
5224         if (adapter->hw.flash_address)
5225                 iounmap(adapter->hw.flash_address);
5226         e1000e_reset_interrupt_capability(adapter);
5227 err_flashmap:
5228         iounmap(adapter->hw.hw_addr);
5229 err_ioremap:
5230         free_netdev(netdev);
5231 err_alloc_etherdev:
5232         pci_release_selected_regions(pdev,
5233                                      pci_select_bars(pdev, IORESOURCE_MEM));
5234 err_pci_reg:
5235 err_dma:
5236         pci_disable_device(pdev);
5237         return err;
5238 }
5239
5240 /**
5241  * e1000_remove - Device Removal Routine
5242  * @pdev: PCI device information struct
5243  *
5244  * e1000_remove is called by the PCI subsystem to alert the driver
5245  * that it should release a PCI device.  The could be caused by a
5246  * Hot-Plug event, or because the driver is going to be removed from
5247  * memory.
5248  **/
5249 static void __devexit e1000_remove(struct pci_dev *pdev)
5250 {
5251         struct net_device *netdev = pci_get_drvdata(pdev);
5252         struct e1000_adapter *adapter = netdev_priv(netdev);
5253
5254         /*
5255          * flush_scheduled work may reschedule our watchdog task, so
5256          * explicitly disable watchdog tasks from being rescheduled
5257          */
5258         set_bit(__E1000_DOWN, &adapter->state);
5259         del_timer_sync(&adapter->watchdog_timer);
5260         del_timer_sync(&adapter->phy_info_timer);
5261
5262         cancel_work_sync(&adapter->reset_task);
5263         cancel_work_sync(&adapter->watchdog_task);
5264         cancel_work_sync(&adapter->downshift_task);
5265         cancel_work_sync(&adapter->update_phy_task);
5266         cancel_work_sync(&adapter->print_hang_task);
5267         flush_scheduled_work();
5268
5269         if (!(netdev->flags & IFF_UP))
5270                 e1000_power_down_phy(adapter);
5271
5272         unregister_netdev(netdev);
5273
5274         /*
5275          * Release control of h/w to f/w.  If f/w is AMT enabled, this
5276          * would have already happened in close and is redundant.
5277          */
5278         e1000_release_hw_control(adapter);
5279
5280         e1000e_reset_interrupt_capability(adapter);
5281         kfree(adapter->tx_ring);
5282         kfree(adapter->rx_ring);
5283
5284         iounmap(adapter->hw.hw_addr);
5285         if (adapter->hw.flash_address)
5286                 iounmap(adapter->hw.flash_address);
5287         pci_release_selected_regions(pdev,
5288                                      pci_select_bars(pdev, IORESOURCE_MEM));
5289
5290         free_netdev(netdev);
5291
5292         /* AER disable */
5293         pci_disable_pcie_error_reporting(pdev);
5294
5295         pci_disable_device(pdev);
5296 }
5297
5298 /* PCI Error Recovery (ERS) */
5299 static struct pci_error_handlers e1000_err_handler = {
5300         .error_detected = e1000_io_error_detected,
5301         .slot_reset = e1000_io_slot_reset,
5302         .resume = e1000_io_resume,
5303 };
5304
5305 static DEFINE_PCI_DEVICE_TABLE(e1000_pci_tbl) = {
5306         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82571EB_COPPER), board_82571 },
5307         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82571EB_FIBER), board_82571 },
5308         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82571EB_QUAD_COPPER), board_82571 },
5309         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82571EB_QUAD_COPPER_LP), board_82571 },
5310         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82571EB_QUAD_FIBER), board_82571 },
5311         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82571EB_SERDES), board_82571 },
5312         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82571EB_SERDES_DUAL), board_82571 },
5313         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82571EB_SERDES_QUAD), board_82571 },
5314         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82571PT_QUAD_COPPER), board_82571 },
5315
5316         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82572EI), board_82572 },
5317         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82572EI_COPPER), board_82572 },
5318         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82572EI_FIBER), board_82572 },
5319         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82572EI_SERDES), board_82572 },
5320
5321         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82573E), board_82573 },
5322         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82573E_IAMT), board_82573 },
5323         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82573L), board_82573 },
5324
5325         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82574L), board_82574 },
5326         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82574LA), board_82574 },
5327         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82583V), board_82583 },
5328
5329         { PCI_VDEVICE(INTEL, E1000_DEV_ID_80003ES2LAN_COPPER_DPT),
5330           board_80003es2lan },
5331         { PCI_VDEVICE(INTEL, E1000_DEV_ID_80003ES2LAN_COPPER_SPT),
5332           board_80003es2lan },
5333         { PCI_VDEVICE(INTEL, E1000_DEV_ID_80003ES2LAN_SERDES_DPT),
5334           board_80003es2lan },
5335         { PCI_VDEVICE(INTEL, E1000_DEV_ID_80003ES2LAN_SERDES_SPT),
5336           board_80003es2lan },
5337
5338         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH8_IFE), board_ich8lan },
5339         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH8_IFE_G), board_ich8lan },
5340         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH8_IFE_GT), board_ich8lan },
5341         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH8_IGP_AMT), board_ich8lan },
5342         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH8_IGP_C), board_ich8lan },
5343         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH8_IGP_M), board_ich8lan },
5344         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH8_IGP_M_AMT), board_ich8lan },
5345         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH8_82567V_3), board_ich8lan },
5346
5347         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH9_IFE), board_ich9lan },
5348         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH9_IFE_G), board_ich9lan },
5349         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH9_IFE_GT), board_ich9lan },
5350         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH9_IGP_AMT), board_ich9lan },
5351         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH9_IGP_C), board_ich9lan },
5352         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH9_BM), board_ich9lan },
5353         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH9_IGP_M), board_ich9lan },
5354         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH9_IGP_M_AMT), board_ich9lan },
5355         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH9_IGP_M_V), board_ich9lan },
5356
5357         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH10_R_BM_LM), board_ich9lan },
5358         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH10_R_BM_LF), board_ich9lan },
5359         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH10_R_BM_V), board_ich9lan },
5360
5361         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH10_D_BM_LM), board_ich10lan },
5362         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH10_D_BM_LF), board_ich10lan },
5363
5364         { PCI_VDEVICE(INTEL, E1000_DEV_ID_PCH_M_HV_LM), board_pchlan },
5365         { PCI_VDEVICE(INTEL, E1000_DEV_ID_PCH_M_HV_LC), board_pchlan },
5366         { PCI_VDEVICE(INTEL, E1000_DEV_ID_PCH_D_HV_DM), board_pchlan },
5367         { PCI_VDEVICE(INTEL, E1000_DEV_ID_PCH_D_HV_DC), board_pchlan },
5368
5369         { }     /* terminate list */
5370 };
5371 MODULE_DEVICE_TABLE(pci, e1000_pci_tbl);
5372
5373 /* PCI Device API Driver */
5374 static struct pci_driver e1000_driver = {
5375         .name     = e1000e_driver_name,
5376         .id_table = e1000_pci_tbl,
5377         .probe    = e1000_probe,
5378         .remove   = __devexit_p(e1000_remove),
5379 #ifdef CONFIG_PM
5380         /* Power Management Hooks */
5381         .suspend  = e1000_suspend,
5382         .resume   = e1000_resume,
5383 #endif
5384         .shutdown = e1000_shutdown,
5385         .err_handler = &e1000_err_handler
5386 };
5387
5388 /**
5389  * e1000_init_module - Driver Registration Routine
5390  *
5391  * e1000_init_module is the first routine called when the driver is
5392  * loaded. All it does is register with the PCI subsystem.
5393  **/
5394 static int __init e1000_init_module(void)
5395 {
5396         int ret;
5397         printk(KERN_INFO "%s: Intel(R) PRO/1000 Network Driver - %s\n",
5398                e1000e_driver_name, e1000e_driver_version);
5399         printk(KERN_INFO "%s: Copyright (c) 1999 - 2009 Intel Corporation.\n",
5400                e1000e_driver_name);
5401         ret = pci_register_driver(&e1000_driver);
5402
5403         return ret;
5404 }
5405 module_init(e1000_init_module);
5406
5407 /**
5408  * e1000_exit_module - Driver Exit Cleanup Routine
5409  *
5410  * e1000_exit_module is called just before the driver is removed
5411  * from memory.
5412  **/
5413 static void __exit e1000_exit_module(void)
5414 {
5415         pci_unregister_driver(&e1000_driver);
5416 }
5417 module_exit(e1000_exit_module);
5418
5419
5420 MODULE_AUTHOR("Intel Corporation, <linux.nics@intel.com>");
5421 MODULE_DESCRIPTION("Intel(R) PRO/1000 Network Driver");
5422 MODULE_LICENSE("GPL");
5423 MODULE_VERSION(DRV_VERSION);
5424
5425 /* e1000_main.c */