drivers/net/e1000e/netdev.c

   1 /*******************************************************************************
   2
   3   Intel PRO/1000 Linux driver
   4   Copyright(c) 1999 - 2008 Intel Corporation.
   5
   6   This program is free software; you can redistribute it and/or modify it
   7   under the terms and conditions of the GNU General Public License,
   8   version 2, as published by the Free Software Foundation.
   9
  10   This program is distributed in the hope it will be useful, but WITHOUT
  11   ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
  12   FITNESS FOR A PARTICULAR PURPOSE.  See the GNU General Public License for
  13   more details.
  14
  15   You should have received a copy of the GNU General Public License along with
  16   this program; if not, write to the Free Software Foundation, Inc.,
  17   51 Franklin St - Fifth Floor, Boston, MA 02110-1301 USA.
  18
  19   The full GNU General Public License is included in this distribution in
  20   the file called "COPYING".
  21
  22   Contact Information:
  23   Linux NICS <linux.nics@intel.com>
  24   e1000-devel Mailing List <e1000-devel@lists.sourceforge.net>
  25   Intel Corporation, 5200 N.E. Elam Young Parkway, Hillsboro, OR 97124-6497
  26
  27 *******************************************************************************/
  28
  29 #include <linux/module.h>
  30 #include <linux/types.h>
  31 #include <linux/init.h>
  32 #include <linux/pci.h>
  33 #include <linux/vmalloc.h>
  34 #include <linux/pagemap.h>
  35 #include <linux/delay.h>
  36 #include <linux/netdevice.h>
  37 #include <linux/tcp.h>
  38 #include <linux/ipv6.h>
  39 #include <net/checksum.h>
  40 #include <net/ip6_checksum.h>
  41 #include <linux/mii.h>
  42 #include <linux/ethtool.h>
  43 #include <linux/if_vlan.h>
  44 #include <linux/cpu.h>
  45 #include <linux/smp.h>
  46 #include <linux/pm_qos_params.h>
  47
  48 #include "e1000.h"
  49
  50 #define DRV_VERSION "0.3.3.3-k6"
  51 char e1000e_driver_name[] = "e1000e";
  52 const char e1000e_driver_version[] = DRV_VERSION;
  53
  54 static const struct e1000_info *e1000_info_tbl[] = {
  55         [board_82571]           = &e1000_82571_info,
  56         [board_82572]           = &e1000_82572_info,
  57         [board_82573]           = &e1000_82573_info,
  58         [board_82574]           = &e1000_82574_info,
  59         [board_80003es2lan]     = &e1000_es2_info,
  60         [board_ich8lan]         = &e1000_ich8_info,
  61         [board_ich9lan]         = &e1000_ich9_info,
  62         [board_ich10lan]        = &e1000_ich10_info,
  63 };
  64
  65 #ifdef DEBUG
  66 /**
  67  * e1000_get_hw_dev_name - return device name string
  68  * used by hardware layer to print debugging information
  69  **/
  70 char *e1000e_get_hw_dev_name(struct e1000_hw *hw)
  71 {
  72         return hw->adapter->netdev->name;
  73 }
  74 #endif
  75
  76 /**
  77  * e1000_desc_unused - calculate if we have unused descriptors
  78  **/
  79 static int e1000_desc_unused(struct e1000_ring *ring)
  80 {
  81         if (ring->next_to_clean > ring->next_to_use)
  82                 return ring->next_to_clean - ring->next_to_use - 1;
  83
  84         return ring->count + ring->next_to_clean - ring->next_to_use - 1;
  85 }
  86
  87 /**
  88  * e1000_receive_skb - helper function to handle Rx indications
  89  * @adapter: board private structure
  90  * @status: descriptor status field as written by hardware
  91  * @vlan: descriptor vlan field as written by hardware (no le/be conversion)
  92  * @skb: pointer to sk_buff to be indicated to stack
  93  **/
  94 static void e1000_receive_skb(struct e1000_adapter *adapter,
  95                               struct net_device *netdev,
  96                               struct sk_buff *skb,
  97                               u8 status, __le16 vlan)
  98 {
  99         skb->protocol = eth_type_trans(skb, netdev);
 100
 101         if (adapter->vlgrp && (status & E1000_RXD_STAT_VP))
 102                 vlan_hwaccel_receive_skb(skb, adapter->vlgrp,
 103                                          le16_to_cpu(vlan));
 104         else
 105                 netif_receive_skb(skb);
 106
 107         netdev->last_rx = jiffies;
 108 }
 109
 110 /**
 111  * e1000_rx_checksum - Receive Checksum Offload for 82543
 112  * @adapter:     board private structure
 113  * @status_err:  receive descriptor status and error fields
 114  * @csum:       receive descriptor csum field
 115  * @sk_buff:     socket buffer with received data
 116  **/
 117 static void e1000_rx_checksum(struct e1000_adapter *adapter, u32 status_err,
 118                               u32 csum, struct sk_buff *skb)
 119 {
 120         u16 status = (u16)status_err;
 121         u8 errors = (u8)(status_err >> 24);
 122         skb->ip_summed = CHECKSUM_NONE;
 123
 124         /* Ignore Checksum bit is set */
 125         if (status & E1000_RXD_STAT_IXSM)
 126                 return;
 127         /* TCP/UDP checksum error bit is set */
 128         if (errors & E1000_RXD_ERR_TCPE) {
 129                 /* let the stack verify checksum errors */
 130                 adapter->hw_csum_err++;
 131                 return;
 132         }
 133
 134         /* TCP/UDP Checksum has not been calculated */
 135         if (!(status & (E1000_RXD_STAT_TCPCS | E1000_RXD_STAT_UDPCS)))
 136                 return;
 137
 138         /* It must be a TCP or UDP packet with a valid checksum */
 139         if (status & E1000_RXD_STAT_TCPCS) {
 140                 /* TCP checksum is good */
 141                 skb->ip_summed = CHECKSUM_UNNECESSARY;
 142         } else {
 143                 /*
 144                  * IP fragment with UDP payload
 145                  * Hardware complements the payload checksum, so we undo it
 146                  * and then put the value in host order for further stack use.
 147                  */
 148                 __sum16 sum = (__force __sum16)htons(csum);
 149                 skb->csum = csum_unfold(~sum);
 150                 skb->ip_summed = CHECKSUM_COMPLETE;
 151         }
 152         adapter->hw_csum_good++;
 153 }
 154
 155 /**
 156  * e1000_alloc_rx_buffers - Replace used receive buffers; legacy & extended
 157  * @adapter: address of board private structure
 158  **/
 159 static void e1000_alloc_rx_buffers(struct e1000_adapter *adapter,
 160                                    int cleaned_count)
 161 {
 162         struct net_device *netdev = adapter->netdev;
 163         struct pci_dev *pdev = adapter->pdev;
 164         struct e1000_ring *rx_ring = adapter->rx_ring;
 165         struct e1000_rx_desc *rx_desc;
 166         struct e1000_buffer *buffer_info;
 167         struct sk_buff *skb;
 168         unsigned int i;
 169         unsigned int bufsz = adapter->rx_buffer_len + NET_IP_ALIGN;
 170
 171         i = rx_ring->next_to_use;
 172         buffer_info = &rx_ring->buffer_info[i];
 173
 174         while (cleaned_count--) {
 175                 skb = buffer_info->skb;
 176                 if (skb) {
 177                         skb_trim(skb, 0);
 178                         goto map_skb;
 179                 }
 180
 181                 skb = netdev_alloc_skb(netdev, bufsz);
 182                 if (!skb) {
 183                         /* Better luck next round */
 184                         adapter->alloc_rx_buff_failed++;
 185                         break;
 186                 }
 187
 188                 /*
 189                  * Make buffer alignment 2 beyond a 16 byte boundary
 190                  * this will result in a 16 byte aligned IP header after
 191                  * the 14 byte MAC header is removed
 192                  */
 193                 skb_reserve(skb, NET_IP_ALIGN);
 194
 195                 buffer_info->skb = skb;
 196 map_skb:
 197                 buffer_info->dma = pci_map_single(pdev, skb->data,
 198                                                   adapter->rx_buffer_len,
 199                                                   PCI_DMA_FROMDEVICE);
 200                 if (pci_dma_mapping_error(pdev, buffer_info->dma)) {
 201                         dev_err(&pdev->dev, "RX DMA map failed\n");
 202                         adapter->rx_dma_failed++;
 203                         break;
 204                 }
 205
 206                 rx_desc = E1000_RX_DESC(*rx_ring, i);
 207                 rx_desc->buffer_addr = cpu_to_le64(buffer_info->dma);
 208
 209                 i++;
 210                 if (i == rx_ring->count)
 211                         i = 0;
 212                 buffer_info = &rx_ring->buffer_info[i];
 213         }
 214
 215         if (rx_ring->next_to_use != i) {
 216                 rx_ring->next_to_use = i;
 217                 if (i-- == 0)
 218                         i = (rx_ring->count - 1);
 219
 220                 /*
 221                  * Force memory writes to complete before letting h/w
 222                  * know there are new descriptors to fetch.  (Only
 223                  * applicable for weak-ordered memory model archs,
 224                  * such as IA-64).
 225                  */
 226                 wmb();
 227                 writel(i, adapter->hw.hw_addr + rx_ring->tail);
 228         }
 229 }
 230
 231 /**
 232  * e1000_alloc_rx_buffers_ps - Replace used receive buffers; packet split
 233  * @adapter: address of board private structure
 234  **/
 235 static void e1000_alloc_rx_buffers_ps(struct e1000_adapter *adapter,
 236                                       int cleaned_count)
 237 {
 238         struct net_device *netdev = adapter->netdev;
 239         struct pci_dev *pdev = adapter->pdev;
 240         union e1000_rx_desc_packet_split *rx_desc;
 241         struct e1000_ring *rx_ring = adapter->rx_ring;
 242         struct e1000_buffer *buffer_info;
 243         struct e1000_ps_page *ps_page;
 244         struct sk_buff *skb;
 245         unsigned int i, j;
 246
 247         i = rx_ring->next_to_use;
 248         buffer_info = &rx_ring->buffer_info[i];
 249
 250         while (cleaned_count--) {
 251                 rx_desc = E1000_RX_DESC_PS(*rx_ring, i);
 252
 253                 for (j = 0; j < PS_PAGE_BUFFERS; j++) {
 254                         ps_page = &buffer_info->ps_pages[j];
 255                         if (j >= adapter->rx_ps_pages) {
 256                                 /* all unused desc entries get hw null ptr */
 257                                 rx_desc->read.buffer_addr[j+1] = ~cpu_to_le64(0);
 258                                 continue;
 259                         }
 260                         if (!ps_page->page) {
 261                                 ps_page->page = alloc_page(GFP_ATOMIC);
 262                                 if (!ps_page->page) {
 263                                         adapter->alloc_rx_buff_failed++;
 264                                         goto no_buffers;
 265                                 }
 266                                 ps_page->dma = pci_map_page(pdev,
 267                                                    ps_page->page,
 268                                                    0, PAGE_SIZE,
 269                                                    PCI_DMA_FROMDEVICE);
 270                                 if (pci_dma_mapping_error(pdev, ps_page->dma)) {
 271                                         dev_err(&adapter->pdev->dev,
 272                                           "RX DMA page map failed\n");
 273                                         adapter->rx_dma_failed++;
 274                                         goto no_buffers;
 275                                 }
 276                         }
 277                         /*
 278                          * Refresh the desc even if buffer_addrs
 279                          * didn't change because each write-back
 280                          * erases this info.
 281                          */
 282                         rx_desc->read.buffer_addr[j+1] =
 283                              cpu_to_le64(ps_page->dma);
 284                 }
 285
 286                 skb = netdev_alloc_skb(netdev,
 287                                        adapter->rx_ps_bsize0 + NET_IP_ALIGN);
 288
 289                 if (!skb) {
 290                         adapter->alloc_rx_buff_failed++;
 291                         break;
 292                 }
 293
 294                 /*
 295                  * Make buffer alignment 2 beyond a 16 byte boundary
 296                  * this will result in a 16 byte aligned IP header after
 297                  * the 14 byte MAC header is removed
 298                  */
 299                 skb_reserve(skb, NET_IP_ALIGN);
 300
 301                 buffer_info->skb = skb;
 302                 buffer_info->dma = pci_map_single(pdev, skb->data,
 303                                                   adapter->rx_ps_bsize0,
 304                                                   PCI_DMA_FROMDEVICE);
 305                 if (pci_dma_mapping_error(pdev, buffer_info->dma)) {
 306                         dev_err(&pdev->dev, "RX DMA map failed\n");
 307                         adapter->rx_dma_failed++;
 308                         /* cleanup skb */
 309                         dev_kfree_skb_any(skb);
 310                         buffer_info->skb = NULL;
 311                         break;
 312                 }
 313
 314                 rx_desc->read.buffer_addr[0] = cpu_to_le64(buffer_info->dma);
 315
 316                 i++;
 317                 if (i == rx_ring->count)
 318                         i = 0;
 319                 buffer_info = &rx_ring->buffer_info[i];
 320         }
 321
 322 no_buffers:
 323         if (rx_ring->next_to_use != i) {
 324                 rx_ring->next_to_use = i;
 325
 326                 if (!(i--))
 327                         i = (rx_ring->count - 1);
 328
 329                 /*
 330                  * Force memory writes to complete before letting h/w
 331                  * know there are new descriptors to fetch.  (Only
 332                  * applicable for weak-ordered memory model archs,
 333                  * such as IA-64).
 334                  */
 335                 wmb();
 336                 /*
 337                  * Hardware increments by 16 bytes, but packet split
 338                  * descriptors are 32 bytes...so we increment tail
 339                  * twice as much.
 340                  */
 341                 writel(i<<1, adapter->hw.hw_addr + rx_ring->tail);
 342         }
 343 }
 344
 345 /**
 346  * e1000_alloc_jumbo_rx_buffers - Replace used jumbo receive buffers
 347  * @adapter: address of board private structure
 348  * @cleaned_count: number of buffers to allocate this pass
 349  **/
 350
 351 static void e1000_alloc_jumbo_rx_buffers(struct e1000_adapter *adapter,
 352                                          int cleaned_count)
 353 {
 354         struct net_device *netdev = adapter->netdev;
 355         struct pci_dev *pdev = adapter->pdev;
 356         struct e1000_rx_desc *rx_desc;
 357         struct e1000_ring *rx_ring = adapter->rx_ring;
 358         struct e1000_buffer *buffer_info;
 359         struct sk_buff *skb;
 360         unsigned int i;
 361         unsigned int bufsz = 256 -
 362                              16 /* for skb_reserve */ -
 363                              NET_IP_ALIGN;
 364
 365         i = rx_ring->next_to_use;
 366         buffer_info = &rx_ring->buffer_info[i];
 367
 368         while (cleaned_count--) {
 369                 skb = buffer_info->skb;
 370                 if (skb) {
 371                         skb_trim(skb, 0);
 372                         goto check_page;
 373                 }
 374
 375                 skb = netdev_alloc_skb(netdev, bufsz);
 376                 if (unlikely(!skb)) {
 377                         /* Better luck next round */
 378                         adapter->alloc_rx_buff_failed++;
 379                         break;
 380                 }
 381
 382                 /* Make buffer alignment 2 beyond a 16 byte boundary
 383                  * this will result in a 16 byte aligned IP header after
 384                  * the 14 byte MAC header is removed
 385                  */
 386                 skb_reserve(skb, NET_IP_ALIGN);
 387
 388                 buffer_info->skb = skb;
 389 check_page:
 390                 /* allocate a new page if necessary */
 391                 if (!buffer_info->page) {
 392                         buffer_info->page = alloc_page(GFP_ATOMIC);
 393                         if (unlikely(!buffer_info->page)) {
 394                                 adapter->alloc_rx_buff_failed++;
 395                                 break;
 396                         }
 397                 }
 398
 399                 if (!buffer_info->dma)
 400                         buffer_info->dma = pci_map_page(pdev,
 401                                                         buffer_info->page, 0,
 402                                                         PAGE_SIZE,
 403                                                         PCI_DMA_FROMDEVICE);
 404
 405                 rx_desc = E1000_RX_DESC(*rx_ring, i);
 406                 rx_desc->buffer_addr = cpu_to_le64(buffer_info->dma);
 407
 408                 if (unlikely(++i == rx_ring->count))
 409                         i = 0;
 410                 buffer_info = &rx_ring->buffer_info[i];
 411         }
 412
 413         if (likely(rx_ring->next_to_use != i)) {
 414                 rx_ring->next_to_use = i;
 415                 if (unlikely(i-- == 0))
 416                         i = (rx_ring->count - 1);
 417
 418                 /* Force memory writes to complete before letting h/w
 419                  * know there are new descriptors to fetch.  (Only
 420                  * applicable for weak-ordered memory model archs,
 421                  * such as IA-64). */
 422                 wmb();
 423                 writel(i, adapter->hw.hw_addr + rx_ring->tail);
 424         }
 425 }
 426
 427 /**
 428  * e1000_clean_rx_irq - Send received data up the network stack; legacy
 429  * @adapter: board private structure
 430  *
 431  * the return value indicates whether actual cleaning was done, there
 432  * is no guarantee that everything was cleaned
 433  **/
 434 static bool e1000_clean_rx_irq(struct e1000_adapter *adapter,
 435                                int *work_done, int work_to_do)
 436 {
 437         struct net_device *netdev = adapter->netdev;
 438         struct pci_dev *pdev = adapter->pdev;
 439         struct e1000_ring *rx_ring = adapter->rx_ring;
 440         struct e1000_rx_desc *rx_desc, *next_rxd;
 441         struct e1000_buffer *buffer_info, *next_buffer;
 442         u32 length;
 443         unsigned int i;
 444         int cleaned_count = 0;
 445         bool cleaned = 0;
 446         unsigned int total_rx_bytes = 0, total_rx_packets = 0;
 447
 448         i = rx_ring->next_to_clean;
 449         rx_desc = E1000_RX_DESC(*rx_ring, i);
 450         buffer_info = &rx_ring->buffer_info[i];
 451
 452         while (rx_desc->status & E1000_RXD_STAT_DD) {
 453                 struct sk_buff *skb;
 454                 u8 status;
 455
 456                 if (*work_done >= work_to_do)
 457                         break;
 458                 (*work_done)++;
 459
 460                 status = rx_desc->status;
 461                 skb = buffer_info->skb;
 462                 buffer_info->skb = NULL;
 463
 464                 prefetch(skb->data - NET_IP_ALIGN);
 465
 466                 i++;
 467                 if (i == rx_ring->count)
 468                         i = 0;
 469                 next_rxd = E1000_RX_DESC(*rx_ring, i);
 470                 prefetch(next_rxd);
 471
 472                 next_buffer = &rx_ring->buffer_info[i];
 473
 474                 cleaned = 1;
 475                 cleaned_count++;
 476                 pci_unmap_single(pdev,
 477                                  buffer_info->dma,
 478                                  adapter->rx_buffer_len,
 479                                  PCI_DMA_FROMDEVICE);
 480                 buffer_info->dma = 0;
 481
 482                 length = le16_to_cpu(rx_desc->length);
 483
 484                 /* !EOP means multiple descriptors were used to store a single
 485                  * packet, also make sure the frame isn't just CRC only */
 486                 if (!(status & E1000_RXD_STAT_EOP) || (length <= 4)) {
 487                         /* All receives must fit into a single buffer */
 488                         e_dbg("%s: Receive packet consumed multiple buffers\n",
 489                               netdev->name);
 490                         /* recycle */
 491                         buffer_info->skb = skb;
 492                         goto next_desc;
 493                 }
 494
 495                 if (rx_desc->errors & E1000_RXD_ERR_FRAME_ERR_MASK) {
 496                         /* recycle */
 497                         buffer_info->skb = skb;
 498                         goto next_desc;
 499                 }
 500
 501                 /* adjust length to remove Ethernet CRC */
 502                 if (!(adapter->flags2 & FLAG2_CRC_STRIPPING))
 503                         length -= 4;
 504
 505                 total_rx_bytes += length;
 506                 total_rx_packets++;
 507
 508                 /*
 509                  * code added for copybreak, this should improve
 510                  * performance for small packets with large amounts
 511                  * of reassembly being done in the stack
 512                  */
 513                 if (length < copybreak) {
 514                         struct sk_buff *new_skb =
 515                             netdev_alloc_skb(netdev, length + NET_IP_ALIGN);
 516                         if (new_skb) {
 517                                 skb_reserve(new_skb, NET_IP_ALIGN);
 518                                 skb_copy_to_linear_data_offset(new_skb,
 519                                                                -NET_IP_ALIGN,
 520                                                                (skb->data -
 521                                                                 NET_IP_ALIGN),
 522                                                                (length +
 523                                                                 NET_IP_ALIGN));
 524                                 /* save the skb in buffer_info as good */
 525                                 buffer_info->skb = skb;
 526                                 skb = new_skb;
 527                         }
 528                         /* else just continue with the old one */
 529                 }
 530                 /* end copybreak code */
 531                 skb_put(skb, length);
 532
 533                 /* Receive Checksum Offload */
 534                 e1000_rx_checksum(adapter,
 535                                   (u32)(status) |
 536                                   ((u32)(rx_desc->errors) << 24),
 537                                   le16_to_cpu(rx_desc->csum), skb);
 538
 539                 e1000_receive_skb(adapter, netdev, skb,status,rx_desc->special);
 540
 541 next_desc:
 542                 rx_desc->status = 0;
 543
 544                 /* return some buffers to hardware, one at a time is too slow */
 545                 if (cleaned_count >= E1000_RX_BUFFER_WRITE) {
 546                         adapter->alloc_rx_buf(adapter, cleaned_count);
 547                         cleaned_count = 0;
 548                 }
 549
 550                 /* use prefetched values */
 551                 rx_desc = next_rxd;
 552                 buffer_info = next_buffer;
 553         }
 554         rx_ring->next_to_clean = i;
 555
 556         cleaned_count = e1000_desc_unused(rx_ring);
 557         if (cleaned_count)
 558                 adapter->alloc_rx_buf(adapter, cleaned_count);
 559
 560         adapter->total_rx_bytes += total_rx_bytes;
 561         adapter->total_rx_packets += total_rx_packets;
 562         adapter->net_stats.rx_bytes += total_rx_bytes;
 563         adapter->net_stats.rx_packets += total_rx_packets;
 564         return cleaned;
 565 }
 566
 567 static void e1000_put_txbuf(struct e1000_adapter *adapter,
 568                              struct e1000_buffer *buffer_info)
 569 {
 570         if (buffer_info->dma) {
 571                 pci_unmap_page(adapter->pdev, buffer_info->dma,
 572                                buffer_info->length, PCI_DMA_TODEVICE);
 573                 buffer_info->dma = 0;
 574         }
 575         if (buffer_info->skb) {
 576                 dev_kfree_skb_any(buffer_info->skb);
 577                 buffer_info->skb = NULL;
 578         }
 579 }
 580
 581 static void e1000_print_tx_hang(struct e1000_adapter *adapter)
 582 {
 583         struct e1000_ring *tx_ring = adapter->tx_ring;
 584         unsigned int i = tx_ring->next_to_clean;
 585         unsigned int eop = tx_ring->buffer_info[i].next_to_watch;
 586         struct e1000_tx_desc *eop_desc = E1000_TX_DESC(*tx_ring, eop);
 587
 588         /* detected Tx unit hang */
 589         e_err("Detected Tx Unit Hang:\n"
 590               "  TDH                  <%x>\n"
 591               "  TDT                  <%x>\n"
 592               "  next_to_use          <%x>\n"
 593               "  next_to_clean        <%x>\n"
 594               "buffer_info[next_to_clean]:\n"
 595               "  time_stamp           <%lx>\n"
 596               "  next_to_watch        <%x>\n"
 597               "  jiffies              <%lx>\n"
 598               "  next_to_watch.status <%x>\n",
 599               readl(adapter->hw.hw_addr + tx_ring->head),
 600               readl(adapter->hw.hw_addr + tx_ring->tail),
 601               tx_ring->next_to_use,
 602               tx_ring->next_to_clean,
 603               tx_ring->buffer_info[eop].time_stamp,
 604               eop,
 605               jiffies,
 606               eop_desc->upper.fields.status);
 607 }
 608
 609 /**
 610  * e1000_clean_tx_irq - Reclaim resources after transmit completes
 611  * @adapter: board private structure
 612  *
 613  * the return value indicates whether actual cleaning was done, there
 614  * is no guarantee that everything was cleaned
 615  **/
 616 static bool e1000_clean_tx_irq(struct e1000_adapter *adapter)
 617 {
 618         struct net_device *netdev = adapter->netdev;
 619         struct e1000_hw *hw = &adapter->hw;
 620         struct e1000_ring *tx_ring = adapter->tx_ring;
 621         struct e1000_tx_desc *tx_desc, *eop_desc;
 622         struct e1000_buffer *buffer_info;
 623         unsigned int i, eop;
 624         unsigned int count = 0;
 625         bool cleaned = 0;
 626         unsigned int total_tx_bytes = 0, total_tx_packets = 0;
 627
 628         i = tx_ring->next_to_clean;
 629         eop = tx_ring->buffer_info[i].next_to_watch;
 630         eop_desc = E1000_TX_DESC(*tx_ring, eop);
 631
 632         while (eop_desc->upper.data & cpu_to_le32(E1000_TXD_STAT_DD)) {
 633                 for (cleaned = 0; !cleaned; ) {
 634                         tx_desc = E1000_TX_DESC(*tx_ring, i);
 635                         buffer_info = &tx_ring->buffer_info[i];
 636                         cleaned = (i == eop);
 637
 638                         if (cleaned) {
 639                                 struct sk_buff *skb = buffer_info->skb;
 640                                 unsigned int segs, bytecount;
 641                                 segs = skb_shinfo(skb)->gso_segs ?: 1;
 642                                 /* multiply data chunks by size of headers */
 643                                 bytecount = ((segs - 1) * skb_headlen(skb)) +
 644                                             skb->len;
 645                                 total_tx_packets += segs;
 646                                 total_tx_bytes += bytecount;
 647                         }
 648
 649                         e1000_put_txbuf(adapter, buffer_info);
 650                         tx_desc->upper.data = 0;
 651
 652                         i++;
 653                         if (i == tx_ring->count)
 654                                 i = 0;
 655                 }
 656
 657                 eop = tx_ring->buffer_info[i].next_to_watch;
 658                 eop_desc = E1000_TX_DESC(*tx_ring, eop);
 659 #define E1000_TX_WEIGHT 64
 660                 /* weight of a sort for tx, to avoid endless transmit cleanup */
 661                 if (count++ == E1000_TX_WEIGHT)
 662                         break;
 663         }
 664
 665         tx_ring->next_to_clean = i;
 666
 667 #define TX_WAKE_THRESHOLD 32
 668         if (cleaned && netif_carrier_ok(netdev) &&
 669                      e1000_desc_unused(tx_ring) >= TX_WAKE_THRESHOLD) {
 670                 /* Make sure that anybody stopping the queue after this
 671                  * sees the new next_to_clean.
 672                  */
 673                 smp_mb();
 674
 675                 if (netif_queue_stopped(netdev) &&
 676                     !(test_bit(__E1000_DOWN, &adapter->state))) {
 677                         netif_wake_queue(netdev);
 678                         ++adapter->restart_queue;
 679                 }
 680         }
 681
 682         if (adapter->detect_tx_hung) {
 683                 /*
 684                  * Detect a transmit hang in hardware, this serializes the
 685                  * check with the clearing of time_stamp and movement of i
 686                  */
 687                 adapter->detect_tx_hung = 0;
 688                 if (tx_ring->buffer_info[eop].dma &&
 689                     time_after(jiffies, tx_ring->buffer_info[eop].time_stamp
 690                                + (adapter->tx_timeout_factor * HZ))
 691                     && !(er32(STATUS) & E1000_STATUS_TXOFF)) {
 692                         e1000_print_tx_hang(adapter);
 693                         netif_stop_queue(netdev);
 694                 }
 695         }
 696         adapter->total_tx_bytes += total_tx_bytes;
 697         adapter->total_tx_packets += total_tx_packets;
 698         adapter->net_stats.tx_bytes += total_tx_bytes;
 699         adapter->net_stats.tx_packets += total_tx_packets;
 700         return cleaned;
 701 }
 702
 703 /**
 704  * e1000_clean_rx_irq_ps - Send received data up the network stack; packet split
 705  * @adapter: board private structure
 706  *
 707  * the return value indicates whether actual cleaning was done, there
 708  * is no guarantee that everything was cleaned
 709  **/
 710 static bool e1000_clean_rx_irq_ps(struct e1000_adapter *adapter,
 711                                   int *work_done, int work_to_do)
 712 {
 713         union e1000_rx_desc_packet_split *rx_desc, *next_rxd;
 714         struct net_device *netdev = adapter->netdev;
 715         struct pci_dev *pdev = adapter->pdev;
 716         struct e1000_ring *rx_ring = adapter->rx_ring;
 717         struct e1000_buffer *buffer_info, *next_buffer;
 718         struct e1000_ps_page *ps_page;
 719         struct sk_buff *skb;
 720         unsigned int i, j;
 721         u32 length, staterr;
 722         int cleaned_count = 0;
 723         bool cleaned = 0;
 724         unsigned int total_rx_bytes = 0, total_rx_packets = 0;
 725
 726         i = rx_ring->next_to_clean;
 727         rx_desc = E1000_RX_DESC_PS(*rx_ring, i);
 728         staterr = le32_to_cpu(rx_desc->wb.middle.status_error);
 729         buffer_info = &rx_ring->buffer_info[i];
 730
 731         while (staterr & E1000_RXD_STAT_DD) {
 732                 if (*work_done >= work_to_do)
 733                         break;
 734                 (*work_done)++;
 735                 skb = buffer_info->skb;
 736
 737                 /* in the packet split case this is header only */
 738                 prefetch(skb->data - NET_IP_ALIGN);
 739
 740                 i++;
 741                 if (i == rx_ring->count)
 742                         i = 0;
 743                 next_rxd = E1000_RX_DESC_PS(*rx_ring, i);
 744                 prefetch(next_rxd);
 745
 746                 next_buffer = &rx_ring->buffer_info[i];
 747
 748                 cleaned = 1;
 749                 cleaned_count++;
 750                 pci_unmap_single(pdev, buffer_info->dma,
 751                                  adapter->rx_ps_bsize0,
 752                                  PCI_DMA_FROMDEVICE);
 753                 buffer_info->dma = 0;
 754
 755                 if (!(staterr & E1000_RXD_STAT_EOP)) {
 756                         e_dbg("%s: Packet Split buffers didn't pick up the "
 757                               "full packet\n", netdev->name);
 758                         dev_kfree_skb_irq(skb);
 759                         goto next_desc;
 760                 }
 761
 762                 if (staterr & E1000_RXDEXT_ERR_FRAME_ERR_MASK) {
 763                         dev_kfree_skb_irq(skb);
 764                         goto next_desc;
 765                 }
 766
 767                 length = le16_to_cpu(rx_desc->wb.middle.length0);
 768
 769                 if (!length) {
 770                         e_dbg("%s: Last part of the packet spanning multiple "
 771                               "descriptors\n", netdev->name);
 772                         dev_kfree_skb_irq(skb);
 773                         goto next_desc;
 774                 }
 775
 776                 /* Good Receive */
 777                 skb_put(skb, length);
 778
 779                 {
 780                 /*
 781                  * this looks ugly, but it seems compiler issues make it
 782                  * more efficient than reusing j
 783                  */
 784                 int l1 = le16_to_cpu(rx_desc->wb.upper.length[0]);
 785
 786                 /*
 787                  * page alloc/put takes too long and effects small packet
 788                  * throughput, so unsplit small packets and save the alloc/put
 789                  * only valid in softirq (napi) context to call kmap_*
 790                  */
 791                 if (l1 && (l1 <= copybreak) &&
 792                     ((length + l1) <= adapter->rx_ps_bsize0)) {
 793                         u8 *vaddr;
 794
 795                         ps_page = &buffer_info->ps_pages[0];
 796
 797                         /*
 798                          * there is no documentation about how to call
 799                          * kmap_atomic, so we can't hold the mapping
 800                          * very long
 801                          */
 802                         pci_dma_sync_single_for_cpu(pdev, ps_page->dma,
 803                                 PAGE_SIZE, PCI_DMA_FROMDEVICE);
 804                         vaddr = kmap_atomic(ps_page->page, KM_SKB_DATA_SOFTIRQ);
 805                         memcpy(skb_tail_pointer(skb), vaddr, l1);
 806                         kunmap_atomic(vaddr, KM_SKB_DATA_SOFTIRQ);
 807                         pci_dma_sync_single_for_device(pdev, ps_page->dma,
 808                                 PAGE_SIZE, PCI_DMA_FROMDEVICE);
 809
 810                         /* remove the CRC */
 811                         if (!(adapter->flags2 & FLAG2_CRC_STRIPPING))
 812                                 l1 -= 4;
 813
 814                         skb_put(skb, l1);
 815                         goto copydone;
 816                 } /* if */
 817                 }
 818
 819                 for (j = 0; j < PS_PAGE_BUFFERS; j++) {
 820                         length = le16_to_cpu(rx_desc->wb.upper.length[j]);
 821                         if (!length)
 822                                 break;
 823
 824                         ps_page = &buffer_info->ps_pages[j];
 825                         pci_unmap_page(pdev, ps_page->dma, PAGE_SIZE,
 826                                        PCI_DMA_FROMDEVICE);
 827                         ps_page->dma = 0;
 828                         skb_fill_page_desc(skb, j, ps_page->page, 0, length);
 829                         ps_page->page = NULL;
 830                         skb->len += length;
 831                         skb->data_len += length;
 832                         skb->truesize += length;
 833                 }
 834
 835                 /* strip the ethernet crc, problem is we're using pages now so
 836                  * this whole operation can get a little cpu intensive
 837                  */
 838                 if (!(adapter->flags2 & FLAG2_CRC_STRIPPING))
 839                         pskb_trim(skb, skb->len - 4);
 840
 841 copydone:
 842                 total_rx_bytes += skb->len;
 843                 total_rx_packets++;
 844
 845                 e1000_rx_checksum(adapter, staterr, le16_to_cpu(
 846                         rx_desc->wb.lower.hi_dword.csum_ip.csum), skb);
 847
 848                 if (rx_desc->wb.upper.header_status &
 849                            cpu_to_le16(E1000_RXDPS_HDRSTAT_HDRSP))
 850                         adapter->rx_hdr_split++;
 851
 852                 e1000_receive_skb(adapter, netdev, skb,
 853                                   staterr, rx_desc->wb.middle.vlan);
 854
 855 next_desc:
 856                 rx_desc->wb.middle.status_error &= cpu_to_le32(~0xFF);
 857                 buffer_info->skb = NULL;
 858
 859                 /* return some buffers to hardware, one at a time is too slow */
 860                 if (cleaned_count >= E1000_RX_BUFFER_WRITE) {
 861                         adapter->alloc_rx_buf(adapter, cleaned_count);
 862                         cleaned_count = 0;
 863                 }
 864
 865                 /* use prefetched values */
 866                 rx_desc = next_rxd;
 867                 buffer_info = next_buffer;
 868
 869                 staterr = le32_to_cpu(rx_desc->wb.middle.status_error);
 870         }
 871         rx_ring->next_to_clean = i;
 872
 873         cleaned_count = e1000_desc_unused(rx_ring);
 874         if (cleaned_count)
 875                 adapter->alloc_rx_buf(adapter, cleaned_count);
 876
 877         adapter->total_rx_bytes += total_rx_bytes;
 878         adapter->total_rx_packets += total_rx_packets;
 879         adapter->net_stats.rx_bytes += total_rx_bytes;
 880         adapter->net_stats.rx_packets += total_rx_packets;
 881         return cleaned;
 882 }
 883
 884 /**
 885  * e1000_consume_page - helper function
 886  **/
 887 static void e1000_consume_page(struct e1000_buffer *bi, struct sk_buff *skb,
 888                                u16 length)
 889 {
 890         bi->page = NULL;
 891         skb->len += length;
 892         skb->data_len += length;
 893         skb->truesize += length;
 894 }
 895
 896 /**
 897  * e1000_clean_jumbo_rx_irq - Send received data up the network stack; legacy
 898  * @adapter: board private structure
 899  *
 900  * the return value indicates whether actual cleaning was done, there
 901  * is no guarantee that everything was cleaned
 902  **/
 903
 904 static bool e1000_clean_jumbo_rx_irq(struct e1000_adapter *adapter,
 905                                      int *work_done, int work_to_do)
 906 {
 907         struct net_device *netdev = adapter->netdev;
 908         struct pci_dev *pdev = adapter->pdev;
 909         struct e1000_ring *rx_ring = adapter->rx_ring;
 910         struct e1000_rx_desc *rx_desc, *next_rxd;
 911         struct e1000_buffer *buffer_info, *next_buffer;
 912         u32 length;
 913         unsigned int i;
 914         int cleaned_count = 0;
 915         bool cleaned = false;
 916         unsigned int total_rx_bytes=0, total_rx_packets=0;
 917
 918         i = rx_ring->next_to_clean;
 919         rx_desc = E1000_RX_DESC(*rx_ring, i);
 920         buffer_info = &rx_ring->buffer_info[i];
 921
 922         while (rx_desc->status & E1000_RXD_STAT_DD) {
 923                 struct sk_buff *skb;
 924                 u8 status;
 925
 926                 if (*work_done >= work_to_do)
 927                         break;
 928                 (*work_done)++;
 929
 930                 status = rx_desc->status;
 931                 skb = buffer_info->skb;
 932                 buffer_info->skb = NULL;
 933
 934                 ++i;
 935                 if (i == rx_ring->count)
 936                         i = 0;
 937                 next_rxd = E1000_RX_DESC(*rx_ring, i);
 938                 prefetch(next_rxd);
 939
 940                 next_buffer = &rx_ring->buffer_info[i];
 941
 942                 cleaned = true;
 943                 cleaned_count++;
 944                 pci_unmap_page(pdev, buffer_info->dma, PAGE_SIZE,
 945                                PCI_DMA_FROMDEVICE);
 946                 buffer_info->dma = 0;
 947
 948                 length = le16_to_cpu(rx_desc->length);
 949
 950                 /* errors is only valid for DD + EOP descriptors */
 951                 if (unlikely((status & E1000_RXD_STAT_EOP) &&
 952                     (rx_desc->errors & E1000_RXD_ERR_FRAME_ERR_MASK))) {
 953                                 /* recycle both page and skb */
 954                                 buffer_info->skb = skb;
 955                                 /* an error means any chain goes out the window
 956                                  * too */
 957                                 if (rx_ring->rx_skb_top)
 958                                         dev_kfree_skb(rx_ring->rx_skb_top);
 959                                 rx_ring->rx_skb_top = NULL;
 960                                 goto next_desc;
 961                 }
 962
 963 #define rxtop rx_ring->rx_skb_top
 964                 if (!(status & E1000_RXD_STAT_EOP)) {
 965                         /* this descriptor is only the beginning (or middle) */
 966                         if (!rxtop) {
 967                                 /* this is the beginning of a chain */
 968                                 rxtop = skb;
 969                                 skb_fill_page_desc(rxtop, 0, buffer_info->page,
 970                                                    0, length);
 971                         } else {
 972                                 /* this is the middle of a chain */
 973                                 skb_fill_page_desc(rxtop,
 974                                     skb_shinfo(rxtop)->nr_frags,
 975                                     buffer_info->page, 0, length);
 976                                 /* re-use the skb, only consumed the page */
 977                                 buffer_info->skb = skb;
 978                         }
 979                         e1000_consume_page(buffer_info, rxtop, length);
 980                         goto next_desc;
 981                 } else {
 982                         if (rxtop) {
 983                                 /* end of the chain */
 984                                 skb_fill_page_desc(rxtop,
 985                                     skb_shinfo(rxtop)->nr_frags,
 986                                     buffer_info->page, 0, length);
 987                                 /* re-use the current skb, we only consumed the
 988                                  * page */
 989                                 buffer_info->skb = skb;
 990                                 skb = rxtop;
 991                                 rxtop = NULL;
 992                                 e1000_consume_page(buffer_info, skb, length);
 993                         } else {
 994                                 /* no chain, got EOP, this buf is the packet
 995                                  * copybreak to save the put_page/alloc_page */
 996                                 if (length <= copybreak &&
 997                                     skb_tailroom(skb) >= length) {
 998                                         u8 *vaddr;
 999                                         vaddr = kmap_atomic(buffer_info->page,
1000                                                            KM_SKB_DATA_SOFTIRQ);
1001                                         memcpy(skb_tail_pointer(skb), vaddr,
1002                                                length);
1003                                         kunmap_atomic(vaddr,
1004                                                       KM_SKB_DATA_SOFTIRQ);
1005                                         /* re-use the page, so don't erase
1006                                          * buffer_info->page */
1007                                         skb_put(skb, length);
1008                                 } else {
1009                                         skb_fill_page_desc(skb, 0,
1010                                                            buffer_info->page, 0,
1011                                                            length);
1012                                         e1000_consume_page(buffer_info, skb,
1013                                                            length);
1014                                 }
1015                         }
1016                 }
1017
1018                 /* Receive Checksum Offload XXX recompute due to CRC strip? */
1019                 e1000_rx_checksum(adapter,
1020                                   (u32)(status) |
1021                                   ((u32)(rx_desc->errors) << 24),
1022                                   le16_to_cpu(rx_desc->csum), skb);
1023
1024                 /* probably a little skewed due to removing CRC */
1025                 total_rx_bytes += skb->len;
1026                 total_rx_packets++;
1027
1028                 /* eth type trans needs skb->data to point to something */
1029                 if (!pskb_may_pull(skb, ETH_HLEN)) {
1030                         e_err("pskb_may_pull failed.\n");
1031                         dev_kfree_skb(skb);
1032                         goto next_desc;
1033                 }
1034
1035                 e1000_receive_skb(adapter, netdev, skb, status,
1036                                   rx_desc->special);
1037
1038 next_desc:
1039                 rx_desc->status = 0;
1040
1041                 /* return some buffers to hardware, one at a time is too slow */
1042                 if (unlikely(cleaned_count >= E1000_RX_BUFFER_WRITE)) {
1043                         adapter->alloc_rx_buf(adapter, cleaned_count);
1044                         cleaned_count = 0;
1045                 }
1046
1047                 /* use prefetched values */
1048                 rx_desc = next_rxd;
1049                 buffer_info = next_buffer;
1050         }
1051         rx_ring->next_to_clean = i;
1052
1053         cleaned_count = e1000_desc_unused(rx_ring);
1054         if (cleaned_count)
1055                 adapter->alloc_rx_buf(adapter, cleaned_count);
1056
1057         adapter->total_rx_bytes += total_rx_bytes;
1058         adapter->total_rx_packets += total_rx_packets;
1059         adapter->net_stats.rx_bytes += total_rx_bytes;
1060         adapter->net_stats.rx_packets += total_rx_packets;
1061         return cleaned;
1062 }
1063
1064 /**
1065  * e1000_clean_rx_ring - Free Rx Buffers per Queue
1066  * @adapter: board private structure
1067  **/
1068 static void e1000_clean_rx_ring(struct e1000_adapter *adapter)
1069 {
1070         struct e1000_ring *rx_ring = adapter->rx_ring;
1071         struct e1000_buffer *buffer_info;
1072         struct e1000_ps_page *ps_page;
1073         struct pci_dev *pdev = adapter->pdev;
1074         unsigned int i, j;
1075
1076         /* Free all the Rx ring sk_buffs */
1077         for (i = 0; i < rx_ring->count; i++) {
1078                 buffer_info = &rx_ring->buffer_info[i];
1079                 if (buffer_info->dma) {
1080                         if (adapter->clean_rx == e1000_clean_rx_irq)
1081                                 pci_unmap_single(pdev, buffer_info->dma,
1082                                                  adapter->rx_buffer_len,
1083                                                  PCI_DMA_FROMDEVICE);
1084                         else if (adapter->clean_rx == e1000_clean_jumbo_rx_irq)
1085                                 pci_unmap_page(pdev, buffer_info->dma,
1086                                                PAGE_SIZE,
1087                                                PCI_DMA_FROMDEVICE);
1088                         else if (adapter->clean_rx == e1000_clean_rx_irq_ps)
1089                                 pci_unmap_single(pdev, buffer_info->dma,
1090                                                  adapter->rx_ps_bsize0,
1091                                                  PCI_DMA_FROMDEVICE);
1092                         buffer_info->dma = 0;
1093                 }
1094
1095                 if (buffer_info->page) {
1096                         put_page(buffer_info->page);
1097                         buffer_info->page = NULL;
1098                 }
1099
1100                 if (buffer_info->skb) {
1101                         dev_kfree_skb(buffer_info->skb);
1102                         buffer_info->skb = NULL;
1103                 }
1104
1105                 for (j = 0; j < PS_PAGE_BUFFERS; j++) {
1106                         ps_page = &buffer_info->ps_pages[j];
1107                         if (!ps_page->page)
1108                                 break;
1109                         pci_unmap_page(pdev, ps_page->dma, PAGE_SIZE,
1110                                        PCI_DMA_FROMDEVICE);
1111                         ps_page->dma = 0;
1112                         put_page(ps_page->page);
1113                         ps_page->page = NULL;
1114                 }
1115         }
1116
1117         /* there also may be some cached data from a chained receive */
1118         if (rx_ring->rx_skb_top) {
1119                 dev_kfree_skb(rx_ring->rx_skb_top);
1120                 rx_ring->rx_skb_top = NULL;
1121         }
1122
1123         /* Zero out the descriptor ring */
1124         memset(rx_ring->desc, 0, rx_ring->size);
1125
1126         rx_ring->next_to_clean = 0;
1127         rx_ring->next_to_use = 0;
1128
1129         writel(0, adapter->hw.hw_addr + rx_ring->head);
1130         writel(0, adapter->hw.hw_addr + rx_ring->tail);
1131 }
1132
1133 static void e1000e_downshift_workaround(struct work_struct *work)
1134 {
1135         struct e1000_adapter *adapter = container_of(work,
1136                                         struct e1000_adapter, downshift_task);
1137
1138         e1000e_gig_downshift_workaround_ich8lan(&adapter->hw);
1139 }
1140
1141 /**
1142  * e1000_intr_msi - Interrupt Handler
1143  * @irq: interrupt number
1144  * @data: pointer to a network interface device structure
1145  **/
1146 static irqreturn_t e1000_intr_msi(int irq, void *data)
1147 {
1148         struct net_device *netdev = data;
1149         struct e1000_adapter *adapter = netdev_priv(netdev);
1150         struct e1000_hw *hw = &adapter->hw;
1151         u32 icr = er32(ICR);
1152
1153         /*
1154          * read ICR disables interrupts using IAM
1155          */
1156
1157         if (icr & (E1000_ICR_RXSEQ | E1000_ICR_LSC)) {
1158                 hw->mac.get_link_status = 1;
1159                 /*
1160                  * ICH8 workaround-- Call gig speed drop workaround on cable
1161                  * disconnect (LSC) before accessing any PHY registers
1162                  */
1163                 if ((adapter->flags & FLAG_LSC_GIG_SPEED_DROP) &&
1164                     (!(er32(STATUS) & E1000_STATUS_LU)))
1165                         schedule_work(&adapter->downshift_task);
1166
1167                 /*
1168                  * 80003ES2LAN workaround-- For packet buffer work-around on
1169                  * link down event; disable receives here in the ISR and reset
1170                  * adapter in watchdog
1171                  */
1172                 if (netif_carrier_ok(netdev) &&
1173                     adapter->flags & FLAG_RX_NEEDS_RESTART) {
1174                         /* disable receives */
1175                         u32 rctl = er32(RCTL);
1176                         ew32(RCTL, rctl & ~E1000_RCTL_EN);
1177                         adapter->flags |= FLAG_RX_RESTART_NOW;
1178                 }
1179                 /* guard against interrupt when we're going down */
1180                 if (!test_bit(__E1000_DOWN, &adapter->state))
1181                         mod_timer(&adapter->watchdog_timer, jiffies + 1);
1182         }
1183
1184         if (netif_rx_schedule_prep(netdev, &adapter->napi)) {
1185                 adapter->total_tx_bytes = 0;
1186                 adapter->total_tx_packets = 0;
1187                 adapter->total_rx_bytes = 0;
1188                 adapter->total_rx_packets = 0;
1189                 __netif_rx_schedule(netdev, &adapter->napi);
1190         }
1191
1192         return IRQ_HANDLED;
1193 }
1194
1195 /**
1196  * e1000_intr - Interrupt Handler
1197  * @irq: interrupt number
1198  * @data: pointer to a network interface device structure
1199  **/
1200 static irqreturn_t e1000_intr(int irq, void *data)
1201 {
1202         struct net_device *netdev = data;
1203         struct e1000_adapter *adapter = netdev_priv(netdev);
1204         struct e1000_hw *hw = &adapter->hw;
1205         u32 rctl, icr = er32(ICR);
1206
1207         if (!icr)
1208                 return IRQ_NONE;  /* Not our interrupt */
1209
1210         /*
1211          * IMS will not auto-mask if INT_ASSERTED is not set, and if it is
1212          * not set, then the adapter didn't send an interrupt
1213          */
1214         if (!(icr & E1000_ICR_INT_ASSERTED))
1215                 return IRQ_NONE;
1216
1217         /*
1218          * Interrupt Auto-Mask...upon reading ICR,
1219          * interrupts are masked.  No need for the
1220          * IMC write
1221          */
1222
1223         if (icr & (E1000_ICR_RXSEQ | E1000_ICR_LSC)) {
1224                 hw->mac.get_link_status = 1;
1225                 /*
1226                  * ICH8 workaround-- Call gig speed drop workaround on cable
1227                  * disconnect (LSC) before accessing any PHY registers
1228                  */
1229                 if ((adapter->flags & FLAG_LSC_GIG_SPEED_DROP) &&
1230                     (!(er32(STATUS) & E1000_STATUS_LU)))
1231                         schedule_work(&adapter->downshift_task);
1232
1233                 /*
1234                  * 80003ES2LAN workaround--
1235                  * For packet buffer work-around on link down event;
1236                  * disable receives here in the ISR and
1237                  * reset adapter in watchdog
1238                  */
1239                 if (netif_carrier_ok(netdev) &&
1240                     (adapter->flags & FLAG_RX_NEEDS_RESTART)) {
1241                         /* disable receives */
1242                         rctl = er32(RCTL);
1243                         ew32(RCTL, rctl & ~E1000_RCTL_EN);
1244                         adapter->flags |= FLAG_RX_RESTART_NOW;
1245                 }
1246                 /* guard against interrupt when we're going down */
1247                 if (!test_bit(__E1000_DOWN, &adapter->state))
1248                         mod_timer(&adapter->watchdog_timer, jiffies + 1);
1249         }
1250
1251         if (netif_rx_schedule_prep(netdev, &adapter->napi)) {
1252                 adapter->total_tx_bytes = 0;
1253                 adapter->total_tx_packets = 0;
1254                 adapter->total_rx_bytes = 0;
1255                 adapter->total_rx_packets = 0;
1256                 __netif_rx_schedule(netdev, &adapter->napi);
1257         }
1258
1259         return IRQ_HANDLED;
1260 }
1261
1262 static irqreturn_t e1000_msix_other(int irq, void *data)
1263 {
1264         struct net_device *netdev = data;
1265         struct e1000_adapter *adapter = netdev_priv(netdev);
1266         struct e1000_hw *hw = &adapter->hw;
1267         u32 icr = er32(ICR);
1268
1269         if (!(icr & E1000_ICR_INT_ASSERTED)) {
1270                 ew32(IMS, E1000_IMS_OTHER);
1271                 return IRQ_NONE;
1272         }
1273
1274         if (icr & adapter->eiac_mask)
1275                 ew32(ICS, (icr & adapter->eiac_mask));
1276
1277         if (icr & E1000_ICR_OTHER) {
1278                 if (!(icr & E1000_ICR_LSC))
1279                         goto no_link_interrupt;
1280                 hw->mac.get_link_status = 1;
1281                 /* guard against interrupt when we're going down */
1282                 if (!test_bit(__E1000_DOWN, &adapter->state))
1283                         mod_timer(&adapter->watchdog_timer, jiffies + 1);
1284         }
1285
1286 no_link_interrupt:
1287         ew32(IMS, E1000_IMS_LSC | E1000_IMS_OTHER);
1288
1289         return IRQ_HANDLED;
1290 }
1291
1292
1293 static irqreturn_t e1000_intr_msix_tx(int irq, void *data)
1294 {
1295         struct net_device *netdev = data;
1296         struct e1000_adapter *adapter = netdev_priv(netdev);
1297         struct e1000_hw *hw = &adapter->hw;
1298         struct e1000_ring *tx_ring = adapter->tx_ring;
1299
1300
1301         adapter->total_tx_bytes = 0;
1302         adapter->total_tx_packets = 0;
1303
1304         if (!e1000_clean_tx_irq(adapter))
1305                 /* Ring was not completely cleaned, so fire another interrupt */
1306                 ew32(ICS, tx_ring->ims_val);
1307
1308         return IRQ_HANDLED;
1309 }
1310
1311 static irqreturn_t e1000_intr_msix_rx(int irq, void *data)
1312 {
1313         struct net_device *netdev = data;
1314         struct e1000_adapter *adapter = netdev_priv(netdev);
1315
1316         /* Write the ITR value calculated at the end of the
1317          * previous interrupt.
1318          */
1319         if (adapter->rx_ring->set_itr) {
1320                 writel(1000000000 / (adapter->rx_ring->itr_val * 256),
1321                        adapter->hw.hw_addr + adapter->rx_ring->itr_register);
1322                 adapter->rx_ring->set_itr = 0;
1323         }
1324
1325         if (netif_rx_schedule_prep(netdev, &adapter->napi)) {
1326                 adapter->total_rx_bytes = 0;
1327                 adapter->total_rx_packets = 0;
1328                 __netif_rx_schedule(netdev, &adapter->napi);
1329         }
1330         return IRQ_HANDLED;
1331 }
1332
1333 /**
1334  * e1000_configure_msix - Configure MSI-X hardware
1335  *
1336  * e1000_configure_msix sets up the hardware to properly
1337  * generate MSI-X interrupts.
1338  **/
1339 static void e1000_configure_msix(struct e1000_adapter *adapter)
1340 {
1341         struct e1000_hw *hw = &adapter->hw;
1342         struct e1000_ring *rx_ring = adapter->rx_ring;
1343         struct e1000_ring *tx_ring = adapter->tx_ring;
1344         int vector = 0;
1345         u32 ctrl_ext, ivar = 0;
1346
1347         adapter->eiac_mask = 0;
1348
1349         /* Workaround issue with spurious interrupts on 82574 in MSI-X mode */
1350         if (hw->mac.type == e1000_82574) {
1351                 u32 rfctl = er32(RFCTL);
1352                 rfctl |= E1000_RFCTL_ACK_DIS;
1353                 ew32(RFCTL, rfctl);
1354         }
1355
1356 #define E1000_IVAR_INT_ALLOC_VALID      0x8
1357         /* Configure Rx vector */
1358         rx_ring->ims_val = E1000_IMS_RXQ0;
1359         adapter->eiac_mask |= rx_ring->ims_val;
1360         if (rx_ring->itr_val)
1361                 writel(1000000000 / (rx_ring->itr_val * 256),
1362                        hw->hw_addr + rx_ring->itr_register);
1363         else
1364                 writel(1, hw->hw_addr + rx_ring->itr_register);
1365         ivar = E1000_IVAR_INT_ALLOC_VALID | vector;
1366
1367         /* Configure Tx vector */
1368         tx_ring->ims_val = E1000_IMS_TXQ0;
1369         vector++;
1370         if (tx_ring->itr_val)
1371                 writel(1000000000 / (tx_ring->itr_val * 256),
1372                        hw->hw_addr + tx_ring->itr_register);
1373         else
1374                 writel(1, hw->hw_addr + tx_ring->itr_register);
1375         adapter->eiac_mask |= tx_ring->ims_val;
1376         ivar |= ((E1000_IVAR_INT_ALLOC_VALID | vector) << 8);
1377
1378         /* set vector for Other Causes, e.g. link changes */
1379         vector++;
1380         ivar |= ((E1000_IVAR_INT_ALLOC_VALID | vector) << 16);
1381         if (rx_ring->itr_val)
1382                 writel(1000000000 / (rx_ring->itr_val * 256),
1383                        hw->hw_addr + E1000_EITR_82574(vector));
1384         else
1385                 writel(1, hw->hw_addr + E1000_EITR_82574(vector));
1386
1387         /* Cause Tx interrupts on every write back */
1388         ivar |= (1 << 31);
1389
1390         ew32(IVAR, ivar);
1391
1392         /* enable MSI-X PBA support */
1393         ctrl_ext = er32(CTRL_EXT);
1394         ctrl_ext |= E1000_CTRL_EXT_PBA_CLR;
1395
1396         /* Auto-Mask Other interrupts upon ICR read */
1397 #define E1000_EIAC_MASK_82574   0x01F00000
1398         ew32(IAM, ~E1000_EIAC_MASK_82574 | E1000_IMS_OTHER);
1399         ctrl_ext |= E1000_CTRL_EXT_EIAME;
1400         ew32(CTRL_EXT, ctrl_ext);
1401         e1e_flush();
1402 }
1403
1404 void e1000e_reset_interrupt_capability(struct e1000_adapter *adapter)
1405 {
1406         if (adapter->msix_entries) {
1407                 pci_disable_msix(adapter->pdev);
1408                 kfree(adapter->msix_entries);
1409                 adapter->msix_entries = NULL;
1410         } else if (adapter->flags & FLAG_MSI_ENABLED) {
1411                 pci_disable_msi(adapter->pdev);
1412                 adapter->flags &= ~FLAG_MSI_ENABLED;
1413         }
1414
1415         return;
1416 }
1417
1418 /**
1419  * e1000e_set_interrupt_capability - set MSI or MSI-X if supported
1420  *
1421  * Attempt to configure interrupts using the best available
1422  * capabilities of the hardware and kernel.
1423  **/
1424 void e1000e_set_interrupt_capability(struct e1000_adapter *adapter)
1425 {
1426         int err;
1427         int numvecs, i;
1428
1429
1430         switch (adapter->int_mode) {
1431         case E1000E_INT_MODE_MSIX:
1432                 if (adapter->flags & FLAG_HAS_MSIX) {
1433                         numvecs = 3; /* RxQ0, TxQ0 and other */
1434                         adapter->msix_entries = kcalloc(numvecs,
1435                                                       sizeof(struct msix_entry),
1436                                                       GFP_KERNEL);
1437                         if (adapter->msix_entries) {
1438                                 for (i = 0; i < numvecs; i++)
1439                                         adapter->msix_entries[i].entry = i;
1440
1441                                 err = pci_enable_msix(adapter->pdev,
1442                                                       adapter->msix_entries,
1443                                                       numvecs);
1444                                 if (err == 0)
1445                                         return;
1446                         }
1447                         /* MSI-X failed, so fall through and try MSI */
1448                         e_err("Failed to initialize MSI-X interrupts.  "
1449                               "Falling back to MSI interrupts.\n");
1450                         e1000e_reset_interrupt_capability(adapter);
1451                 }
1452                 adapter->int_mode = E1000E_INT_MODE_MSI;
1453                 /* Fall through */
1454         case E1000E_INT_MODE_MSI:
1455                 if (!pci_enable_msi(adapter->pdev)) {
1456                         adapter->flags |= FLAG_MSI_ENABLED;
1457                 } else {
1458                         adapter->int_mode = E1000E_INT_MODE_LEGACY;
1459                         e_err("Failed to initialize MSI interrupts.  Falling "
1460                               "back to legacy interrupts.\n");
1461                 }
1462                 /* Fall through */
1463         case E1000E_INT_MODE_LEGACY:
1464                 /* Don't do anything; this is the system default */
1465                 break;
1466         }
1467
1468         return;
1469 }
1470
1471 /**
1472  * e1000_request_msix - Initialize MSI-X interrupts
1473  *
1474  * e1000_request_msix allocates MSI-X vectors and requests interrupts from the
1475  * kernel.
1476  **/
1477 static int e1000_request_msix(struct e1000_adapter *adapter)
1478 {
1479         struct net_device *netdev = adapter->netdev;
1480         int err = 0, vector = 0;
1481
1482         if (strlen(netdev->name) < (IFNAMSIZ - 5))
1483                 sprintf(adapter->rx_ring->name, "%s-rx0", netdev->name);
1484         else
1485                 memcpy(adapter->rx_ring->name, netdev->name, IFNAMSIZ);
1486         err = request_irq(adapter->msix_entries[vector].vector,
1487                           &e1000_intr_msix_rx, 0, adapter->rx_ring->name,
1488                           netdev);
1489         if (err)
1490                 goto out;
1491         adapter->rx_ring->itr_register = E1000_EITR_82574(vector);
1492         adapter->rx_ring->itr_val = adapter->itr;
1493         vector++;
1494
1495         if (strlen(netdev->name) < (IFNAMSIZ - 5))
1496                 sprintf(adapter->tx_ring->name, "%s-tx0", netdev->name);
1497         else
1498                 memcpy(adapter->tx_ring->name, netdev->name, IFNAMSIZ);
1499         err = request_irq(adapter->msix_entries[vector].vector,
1500                           &e1000_intr_msix_tx, 0, adapter->tx_ring->name,
1501                           netdev);
1502         if (err)
1503                 goto out;
1504         adapter->tx_ring->itr_register = E1000_EITR_82574(vector);
1505         adapter->tx_ring->itr_val = adapter->itr;
1506         vector++;
1507
1508         err = request_irq(adapter->msix_entries[vector].vector,
1509                           &e1000_msix_other, 0, netdev->name, netdev);
1510         if (err)
1511                 goto out;
1512
1513         e1000_configure_msix(adapter);
1514         return 0;
1515 out:
1516         return err;
1517 }
1518
1519 /**
1520  * e1000_request_irq - initialize interrupts
1521  *
1522  * Attempts to configure interrupts using the best available
1523  * capabilities of the hardware and kernel.
1524  **/
1525 static int e1000_request_irq(struct e1000_adapter *adapter)
1526 {
1527         struct net_device *netdev = adapter->netdev;
1528         int err;
1529
1530         if (adapter->msix_entries) {
1531                 err = e1000_request_msix(adapter);
1532                 if (!err)
1533                         return err;
1534                 /* fall back to MSI */
1535                 e1000e_reset_interrupt_capability(adapter);
1536                 adapter->int_mode = E1000E_INT_MODE_MSI;
1537                 e1000e_set_interrupt_capability(adapter);
1538         }
1539         if (adapter->flags & FLAG_MSI_ENABLED) {
1540                 err = request_irq(adapter->pdev->irq, &e1000_intr_msi, 0,
1541                                   netdev->name, netdev);
1542                 if (!err)
1543                         return err;
1544
1545                 /* fall back to legacy interrupt */
1546                 e1000e_reset_interrupt_capability(adapter);
1547                 adapter->int_mode = E1000E_INT_MODE_LEGACY;
1548         }
1549
1550         err = request_irq(adapter->pdev->irq, &e1000_intr, IRQF_SHARED,
1551                           netdev->name, netdev);
1552         if (err)
1553                 e_err("Unable to allocate interrupt, Error: %d\n", err);
1554
1555         return err;
1556 }
1557
1558 static void e1000_free_irq(struct e1000_adapter *adapter)
1559 {
1560         struct net_device *netdev = adapter->netdev;
1561
1562         if (adapter->msix_entries) {
1563                 int vector = 0;
1564
1565                 free_irq(adapter->msix_entries[vector].vector, netdev);
1566                 vector++;
1567
1568                 free_irq(adapter->msix_entries[vector].vector, netdev);
1569                 vector++;
1570
1571                 /* Other Causes interrupt vector */
1572                 free_irq(adapter->msix_entries[vector].vector, netdev);
1573                 return;
1574         }
1575
1576         free_irq(adapter->pdev->irq, netdev);
1577 }
1578
1579 /**
1580  * e1000_irq_disable - Mask off interrupt generation on the NIC
1581  **/
1582 static void e1000_irq_disable(struct e1000_adapter *adapter)
1583 {
1584         struct e1000_hw *hw = &adapter->hw;
1585
1586         ew32(IMC, ~0);
1587         if (adapter->msix_entries)
1588                 ew32(EIAC_82574, 0);
1589         e1e_flush();
1590         synchronize_irq(adapter->pdev->irq);
1591 }
1592
1593 /**
1594  * e1000_irq_enable - Enable default interrupt generation settings
1595  **/
1596 static void e1000_irq_enable(struct e1000_adapter *adapter)
1597 {
1598         struct e1000_hw *hw = &adapter->hw;
1599
1600         if (adapter->msix_entries) {
1601                 ew32(EIAC_82574, adapter->eiac_mask & E1000_EIAC_MASK_82574);
1602                 ew32(IMS, adapter->eiac_mask | E1000_IMS_OTHER | E1000_IMS_LSC);
1603         } else {
1604                 ew32(IMS, IMS_ENABLE_MASK);
1605         }
1606         e1e_flush();
1607 }
1608
1609 /**
1610  * e1000_get_hw_control - get control of the h/w from f/w
1611  * @adapter: address of board private structure
1612  *
1613  * e1000_get_hw_control sets {CTRL_EXT|SWSM}:DRV_LOAD bit.
1614  * For ASF and Pass Through versions of f/w this means that
1615  * the driver is loaded. For AMT version (only with 82573)
1616  * of the f/w this means that the network i/f is open.
1617  **/
1618 static void e1000_get_hw_control(struct e1000_adapter *adapter)
1619 {
1620         struct e1000_hw *hw = &adapter->hw;
1621         u32 ctrl_ext;
1622         u32 swsm;
1623
1624         /* Let firmware know the driver has taken over */
1625         if (adapter->flags & FLAG_HAS_SWSM_ON_LOAD) {
1626                 swsm = er32(SWSM);
1627                 ew32(SWSM, swsm | E1000_SWSM_DRV_LOAD);
1628         } else if (adapter->flags & FLAG_HAS_CTRLEXT_ON_LOAD) {
1629                 ctrl_ext = er32(CTRL_EXT);
1630                 ew32(CTRL_EXT, ctrl_ext | E1000_CTRL_EXT_DRV_LOAD);
1631         }
1632 }
1633
1634 /**
1635  * e1000_release_hw_control - release control of the h/w to f/w
1636  * @adapter: address of board private structure
1637  *
1638  * e1000_release_hw_control resets {CTRL_EXT|SWSM}:DRV_LOAD bit.
1639  * For ASF and Pass Through versions of f/w this means that the
1640  * driver is no longer loaded. For AMT version (only with 82573) i
1641  * of the f/w this means that the network i/f is closed.
1642  *
1643  **/
1644 static void e1000_release_hw_control(struct e1000_adapter *adapter)
1645 {
1646         struct e1000_hw *hw = &adapter->hw;
1647         u32 ctrl_ext;
1648         u32 swsm;
1649
1650         /* Let firmware taken over control of h/w */
1651         if (adapter->flags & FLAG_HAS_SWSM_ON_LOAD) {
1652                 swsm = er32(SWSM);
1653                 ew32(SWSM, swsm & ~E1000_SWSM_DRV_LOAD);
1654         } else if (adapter->flags & FLAG_HAS_CTRLEXT_ON_LOAD) {
1655                 ctrl_ext = er32(CTRL_EXT);
1656                 ew32(CTRL_EXT, ctrl_ext & ~E1000_CTRL_EXT_DRV_LOAD);
1657         }
1658 }
1659
1660 /**
1661  * @e1000_alloc_ring - allocate memory for a ring structure
1662  **/
1663 static int e1000_alloc_ring_dma(struct e1000_adapter *adapter,
1664                                 struct e1000_ring *ring)
1665 {
1666         struct pci_dev *pdev = adapter->pdev;
1667
1668         ring->desc = dma_alloc_coherent(&pdev->dev, ring->size, &ring->dma,
1669                                         GFP_KERNEL);
1670         if (!ring->desc)
1671                 return -ENOMEM;
1672
1673         return 0;
1674 }
1675
1676 /**
1677  * e1000e_setup_tx_resources - allocate Tx resources (Descriptors)
1678  * @adapter: board private structure
1679  *
1680  * Return 0 on success, negative on failure
1681  **/
1682 int e1000e_setup_tx_resources(struct e1000_adapter *adapter)
1683 {
1684         struct e1000_ring *tx_ring = adapter->tx_ring;
1685         int err = -ENOMEM, size;
1686
1687         size = sizeof(struct e1000_buffer) * tx_ring->count;
1688         tx_ring->buffer_info = vmalloc(size);
1689         if (!tx_ring->buffer_info)
1690                 goto err;
1691         memset(tx_ring->buffer_info, 0, size);
1692
1693         /* round up to nearest 4K */
1694         tx_ring->size = tx_ring->count * sizeof(struct e1000_tx_desc);
1695         tx_ring->size = ALIGN(tx_ring->size, 4096);
1696
1697         err = e1000_alloc_ring_dma(adapter, tx_ring);
1698         if (err)
1699                 goto err;
1700
1701         tx_ring->next_to_use = 0;
1702         tx_ring->next_to_clean = 0;
1703         spin_lock_init(&adapter->tx_queue_lock);
1704
1705         return 0;
1706 err:
1707         vfree(tx_ring->buffer_info);
1708         e_err("Unable to allocate memory for the transmit descriptor ring\n");
1709         return err;
1710 }
1711
1712 /**
1713  * e1000e_setup_rx_resources - allocate Rx resources (Descriptors)
1714  * @adapter: board private structure
1715  *
1716  * Returns 0 on success, negative on failure
1717  **/
1718 int e1000e_setup_rx_resources(struct e1000_adapter *adapter)
1719 {
1720         struct e1000_ring *rx_ring = adapter->rx_ring;
1721         struct e1000_buffer *buffer_info;
1722         int i, size, desc_len, err = -ENOMEM;
1723
1724         size = sizeof(struct e1000_buffer) * rx_ring->count;
1725         rx_ring->buffer_info = vmalloc(size);
1726         if (!rx_ring->buffer_info)
1727                 goto err;
1728         memset(rx_ring->buffer_info, 0, size);
1729
1730         for (i = 0; i < rx_ring->count; i++) {
1731                 buffer_info = &rx_ring->buffer_info[i];
1732                 buffer_info->ps_pages = kcalloc(PS_PAGE_BUFFERS,
1733                                                 sizeof(struct e1000_ps_page),
1734                                                 GFP_KERNEL);
1735                 if (!buffer_info->ps_pages)
1736                         goto err_pages;
1737         }
1738
1739         desc_len = sizeof(union e1000_rx_desc_packet_split);
1740
1741         /* Round up to nearest 4K */
1742         rx_ring->size = rx_ring->count * desc_len;
1743         rx_ring->size = ALIGN(rx_ring->size, 4096);
1744
1745         err = e1000_alloc_ring_dma(adapter, rx_ring);
1746         if (err)
1747                 goto err_pages;
1748
1749         rx_ring->next_to_clean = 0;
1750         rx_ring->next_to_use = 0;
1751         rx_ring->rx_skb_top = NULL;
1752
1753         return 0;
1754
1755 err_pages:
1756         for (i = 0; i < rx_ring->count; i++) {
1757                 buffer_info = &rx_ring->buffer_info[i];
1758                 kfree(buffer_info->ps_pages);
1759         }
1760 err:
1761         vfree(rx_ring->buffer_info);
1762         e_err("Unable to allocate memory for the transmit descriptor ring\n");
1763         return err;
1764 }
1765
1766 /**
1767  * e1000_clean_tx_ring - Free Tx Buffers
1768  * @adapter: board private structure
1769  **/
1770 static void e1000_clean_tx_ring(struct e1000_adapter *adapter)
1771 {
1772         struct e1000_ring *tx_ring = adapter->tx_ring;
1773         struct e1000_buffer *buffer_info;
1774         unsigned long size;
1775         unsigned int i;
1776
1777         for (i = 0; i < tx_ring->count; i++) {
1778                 buffer_info = &tx_ring->buffer_info[i];
1779                 e1000_put_txbuf(adapter, buffer_info);
1780         }
1781
1782         size = sizeof(struct e1000_buffer) * tx_ring->count;
1783         memset(tx_ring->buffer_info, 0, size);
1784
1785         memset(tx_ring->desc, 0, tx_ring->size);
1786
1787         tx_ring->next_to_use = 0;
1788         tx_ring->next_to_clean = 0;
1789
1790         writel(0, adapter->hw.hw_addr + tx_ring->head);
1791         writel(0, adapter->hw.hw_addr + tx_ring->tail);
1792 }
1793
1794 /**
1795  * e1000e_free_tx_resources - Free Tx Resources per Queue
1796  * @adapter: board private structure
1797  *
1798  * Free all transmit software resources
1799  **/
1800 void e1000e_free_tx_resources(struct e1000_adapter *adapter)
1801 {
1802         struct pci_dev *pdev = adapter->pdev;
1803         struct e1000_ring *tx_ring = adapter->tx_ring;
1804
1805         e1000_clean_tx_ring(adapter);
1806
1807         vfree(tx_ring->buffer_info);
1808         tx_ring->buffer_info = NULL;
1809
1810         dma_free_coherent(&pdev->dev, tx_ring->size, tx_ring->desc,
1811                           tx_ring->dma);
1812         tx_ring->desc = NULL;
1813 }
1814
1815 /**
1816  * e1000e_free_rx_resources - Free Rx Resources
1817  * @adapter: board private structure
1818  *
1819  * Free all receive software resources
1820  **/
1821
1822 void e1000e_free_rx_resources(struct e1000_adapter *adapter)
1823 {
1824         struct pci_dev *pdev = adapter->pdev;
1825         struct e1000_ring *rx_ring = adapter->rx_ring;
1826         int i;
1827
1828         e1000_clean_rx_ring(adapter);
1829
1830         for (i = 0; i < rx_ring->count; i++) {
1831                 kfree(rx_ring->buffer_info[i].ps_pages);
1832         }
1833
1834         vfree(rx_ring->buffer_info);
1835         rx_ring->buffer_info = NULL;
1836
1837         dma_free_coherent(&pdev->dev, rx_ring->size, rx_ring->desc,
1838                           rx_ring->dma);
1839         rx_ring->desc = NULL;
1840 }
1841
1842 /**
1843  * e1000_update_itr - update the dynamic ITR value based on statistics
1844  * @adapter: pointer to adapter
1845  * @itr_setting: current adapter->itr
1846  * @packets: the number of packets during this measurement interval
1847  * @bytes: the number of bytes during this measurement interval
1848  *
1849  *      Stores a new ITR value based on packets and byte
1850  *      counts during the last interrupt.  The advantage of per interrupt
1851  *      computation is faster updates and more accurate ITR for the current
1852  *      traffic pattern.  Constants in this function were computed
1853  *      based on theoretical maximum wire speed and thresholds were set based
1854  *      on testing data as well as attempting to minimize response time
1855  *      while increasing bulk throughput.  This functionality is controlled
1856  *      by the InterruptThrottleRate module parameter.
1857  **/
1858 static unsigned int e1000_update_itr(struct e1000_adapter *adapter,
1859                                      u16 itr_setting, int packets,
1860                                      int bytes)
1861 {
1862         unsigned int retval = itr_setting;
1863
1864         if (packets == 0)
1865                 goto update_itr_done;
1866
1867         switch (itr_setting) {
1868         case lowest_latency:
1869                 /* handle TSO and jumbo frames */
1870                 if (bytes/packets > 8000)
1871                         retval = bulk_latency;
1872                 else if ((packets < 5) && (bytes > 512)) {
1873                         retval = low_latency;
1874                 }
1875                 break;
1876         case low_latency:  /* 50 usec aka 20000 ints/s */
1877                 if (bytes > 10000) {
1878                         /* this if handles the TSO accounting */
1879                         if (bytes/packets > 8000) {
1880                                 retval = bulk_latency;
1881                         } else if ((packets < 10) || ((bytes/packets) > 1200)) {
1882                                 retval = bulk_latency;
1883                         } else if ((packets > 35)) {
1884                                 retval = lowest_latency;
1885                         }
1886                 } else if (bytes/packets > 2000) {
1887                         retval = bulk_latency;
1888                 } else if (packets <= 2 && bytes < 512) {
1889                         retval = lowest_latency;
1890                 }
1891                 break;
1892         case bulk_latency: /* 250 usec aka 4000 ints/s */
1893                 if (bytes > 25000) {
1894                         if (packets > 35) {
1895                                 retval = low_latency;
1896                         }
1897                 } else if (bytes < 6000) {
1898                         retval = low_latency;
1899                 }
1900                 break;
1901         }
1902
1903 update_itr_done:
1904         return retval;
1905 }
1906
1907 static void e1000_set_itr(struct e1000_adapter *adapter)
1908 {
1909         struct e1000_hw *hw = &adapter->hw;
1910         u16 current_itr;
1911         u32 new_itr = adapter->itr;
1912
1913         /* for non-gigabit speeds, just fix the interrupt rate at 4000 */
1914         if (adapter->link_speed != SPEED_1000) {
1915                 current_itr = 0;
1916                 new_itr = 4000;
1917                 goto set_itr_now;
1918         }
1919
1920         adapter->tx_itr = e1000_update_itr(adapter,
1921                                     adapter->tx_itr,
1922                                     adapter->total_tx_packets,
1923                                     adapter->total_tx_bytes);
1924         /* conservative mode (itr 3) eliminates the lowest_latency setting */
1925         if (adapter->itr_setting == 3 && adapter->tx_itr == lowest_latency)
1926                 adapter->tx_itr = low_latency;
1927
1928         adapter->rx_itr = e1000_update_itr(adapter,
1929                                     adapter->rx_itr,
1930                                     adapter->total_rx_packets,
1931                                     adapter->total_rx_bytes);
1932         /* conservative mode (itr 3) eliminates the lowest_latency setting */
1933         if (adapter->itr_setting == 3 && adapter->rx_itr == lowest_latency)
1934                 adapter->rx_itr = low_latency;
1935
1936         current_itr = max(adapter->rx_itr, adapter->tx_itr);
1937
1938         switch (current_itr) {
1939         /* counts and packets in update_itr are dependent on these numbers */
1940         case lowest_latency:
1941                 new_itr = 70000;
1942                 break;
1943         case low_latency:
1944                 new_itr = 20000; /* aka hwitr = ~200 */
1945                 break;
1946         case bulk_latency:
1947                 new_itr = 4000;
1948                 break;
1949         default:
1950                 break;
1951         }
1952
1953 set_itr_now:
1954         if (new_itr != adapter->itr) {
1955                 /*
1956                  * this attempts to bias the interrupt rate towards Bulk
1957                  * by adding intermediate steps when interrupt rate is
1958                  * increasing
1959                  */
1960                 new_itr = new_itr > adapter->itr ?
1961                              min(adapter->itr + (new_itr >> 2), new_itr) :
1962                              new_itr;
1963                 adapter->itr = new_itr;
1964                 adapter->rx_ring->itr_val = new_itr;
1965                 if (adapter->msix_entries)
1966                         adapter->rx_ring->set_itr = 1;
1967                 else
1968                         ew32(ITR, 1000000000 / (new_itr * 256));
1969         }
1970 }
1971
1972 /**
1973  * e1000_alloc_queues - Allocate memory for all rings
1974  * @adapter: board private structure to initialize
1975  **/
1976 static int __devinit e1000_alloc_queues(struct e1000_adapter *adapter)
1977 {
1978         adapter->tx_ring = kzalloc(sizeof(struct e1000_ring), GFP_KERNEL);
1979         if (!adapter->tx_ring)
1980                 goto err;
1981
1982         adapter->rx_ring = kzalloc(sizeof(struct e1000_ring), GFP_KERNEL);
1983         if (!adapter->rx_ring)
1984                 goto err;
1985
1986         return 0;
1987 err:
1988         e_err("Unable to allocate memory for queues\n");
1989         kfree(adapter->rx_ring);
1990         kfree(adapter->tx_ring);
1991         return -ENOMEM;
1992 }
1993
1994 /**
1995  * e1000_clean - NAPI Rx polling callback
1996  * @napi: struct associated with this polling callback
1997  * @budget: amount of packets driver is allowed to process this poll
1998  **/
1999 static int e1000_clean(struct napi_struct *napi, int budget)
2000 {
2001         struct e1000_adapter *adapter = container_of(napi, struct e1000_adapter, napi);
2002         struct e1000_hw *hw = &adapter->hw;
2003         struct net_device *poll_dev = adapter->netdev;
2004         int tx_cleaned = 0, work_done = 0;
2005
2006         /* Must NOT use netdev_priv macro here. */
2007         adapter = poll_dev->priv;
2008
2009         if (adapter->msix_entries &&
2010             !(adapter->rx_ring->ims_val & adapter->tx_ring->ims_val))
2011                 goto clean_rx;
2012
2013         /*
2014          * e1000_clean is called per-cpu.  This lock protects
2015          * tx_ring from being cleaned by multiple cpus
2016          * simultaneously.  A failure obtaining the lock means
2017          * tx_ring is currently being cleaned anyway.
2018          */
2019         if (spin_trylock(&adapter->tx_queue_lock)) {
2020                 tx_cleaned = e1000_clean_tx_irq(adapter);
2021                 spin_unlock(&adapter->tx_queue_lock);
2022         }
2023
2024 clean_rx:
2025         adapter->clean_rx(adapter, &work_done, budget);
2026
2027         if (tx_cleaned)
2028                 work_done = budget;
2029
2030         /* If budget not fully consumed, exit the polling mode */
2031         if (work_done < budget) {
2032                 if (adapter->itr_setting & 3)
2033                         e1000_set_itr(adapter);
2034                 netif_rx_complete(poll_dev, napi);
2035                 if (adapter->msix_entries)
2036                         ew32(IMS, adapter->rx_ring->ims_val);
2037                 else
2038                         e1000_irq_enable(adapter);
2039         }
2040
2041         return work_done;
2042 }
2043
2044 static void e1000_vlan_rx_add_vid(struct net_device *netdev, u16 vid)
2045 {
2046         struct e1000_adapter *adapter = netdev_priv(netdev);
2047         struct e1000_hw *hw = &adapter->hw;
2048         u32 vfta, index;
2049
2050         /* don't update vlan cookie if already programmed */
2051         if ((adapter->hw.mng_cookie.status &
2052              E1000_MNG_DHCP_COOKIE_STATUS_VLAN) &&
2053             (vid == adapter->mng_vlan_id))
2054                 return;
2055         /* add VID to filter table */
2056         index = (vid >> 5) & 0x7F;
2057         vfta = E1000_READ_REG_ARRAY(hw, E1000_VFTA, index);
2058         vfta |= (1 << (vid & 0x1F));
2059         e1000e_write_vfta(hw, index, vfta);
2060 }
2061
2062 static void e1000_vlan_rx_kill_vid(struct net_device *netdev, u16 vid)
2063 {
2064         struct e1000_adapter *adapter = netdev_priv(netdev);
2065         struct e1000_hw *hw = &adapter->hw;
2066         u32 vfta, index;
2067
2068         if (!test_bit(__E1000_DOWN, &adapter->state))
2069                 e1000_irq_disable(adapter);
2070         vlan_group_set_device(adapter->vlgrp, vid, NULL);
2071
2072         if (!test_bit(__E1000_DOWN, &adapter->state))
2073                 e1000_irq_enable(adapter);
2074
2075         if ((adapter->hw.mng_cookie.status &
2076              E1000_MNG_DHCP_COOKIE_STATUS_VLAN) &&
2077             (vid == adapter->mng_vlan_id)) {
2078                 /* release control to f/w */
2079                 e1000_release_hw_control(adapter);
2080                 return;
2081         }
2082
2083         /* remove VID from filter table */
2084         index = (vid >> 5) & 0x7F;
2085         vfta = E1000_READ_REG_ARRAY(hw, E1000_VFTA, index);
2086         vfta &= ~(1 << (vid & 0x1F));
2087         e1000e_write_vfta(hw, index, vfta);
2088 }
2089
2090 static void e1000_update_mng_vlan(struct e1000_adapter *adapter)
2091 {
2092         struct net_device *netdev = adapter->netdev;
2093         u16 vid = adapter->hw.mng_cookie.vlan_id;
2094         u16 old_vid = adapter->mng_vlan_id;
2095
2096         if (!adapter->vlgrp)
2097                 return;
2098
2099         if (!vlan_group_get_device(adapter->vlgrp, vid)) {
2100                 adapter->mng_vlan_id = E1000_MNG_VLAN_NONE;
2101                 if (adapter->hw.mng_cookie.status &
2102                         E1000_MNG_DHCP_COOKIE_STATUS_VLAN) {
2103                         e1000_vlan_rx_add_vid(netdev, vid);
2104                         adapter->mng_vlan_id = vid;
2105                 }
2106
2107                 if ((old_vid != (u16)E1000_MNG_VLAN_NONE) &&
2108                                 (vid != old_vid) &&
2109                     !vlan_group_get_device(adapter->vlgrp, old_vid))
2110                         e1000_vlan_rx_kill_vid(netdev, old_vid);
2111         } else {
2112                 adapter->mng_vlan_id = vid;
2113         }
2114 }
2115
2116
2117 static void e1000_vlan_rx_register(struct net_device *netdev,
2118                                    struct vlan_group *grp)
2119 {
2120         struct e1000_adapter *adapter = netdev_priv(netdev);
2121         struct e1000_hw *hw = &adapter->hw;
2122         u32 ctrl, rctl;
2123
2124         if (!test_bit(__E1000_DOWN, &adapter->state))
2125                 e1000_irq_disable(adapter);
2126         adapter->vlgrp = grp;
2127
2128         if (grp) {
2129                 /* enable VLAN tag insert/strip */
2130                 ctrl = er32(CTRL);
2131                 ctrl |= E1000_CTRL_VME;
2132                 ew32(CTRL, ctrl);
2133
2134                 if (adapter->flags & FLAG_HAS_HW_VLAN_FILTER) {
2135                         /* enable VLAN receive filtering */
2136                         rctl = er32(RCTL);
2137                         rctl &= ~E1000_RCTL_CFIEN;
2138                         ew32(RCTL, rctl);
2139                         e1000_update_mng_vlan(adapter);
2140                 }
2141         } else {
2142                 /* disable VLAN tag insert/strip */
2143                 ctrl = er32(CTRL);
2144                 ctrl &= ~E1000_CTRL_VME;
2145                 ew32(CTRL, ctrl);
2146
2147                 if (adapter->flags & FLAG_HAS_HW_VLAN_FILTER) {
2148                         if (adapter->mng_vlan_id !=
2149                             (u16)E1000_MNG_VLAN_NONE) {
2150                                 e1000_vlan_rx_kill_vid(netdev,
2151                                                        adapter->mng_vlan_id);
2152                                 adapter->mng_vlan_id = E1000_MNG_VLAN_NONE;
2153                         }
2154                 }
2155         }
2156
2157         if (!test_bit(__E1000_DOWN, &adapter->state))
2158                 e1000_irq_enable(adapter);
2159 }
2160
2161 static void e1000_restore_vlan(struct e1000_adapter *adapter)
2162 {
2163         u16 vid;
2164
2165         e1000_vlan_rx_register(adapter->netdev, adapter->vlgrp);
2166
2167         if (!adapter->vlgrp)
2168                 return;
2169
2170         for (vid = 0; vid < VLAN_GROUP_ARRAY_LEN; vid++) {
2171                 if (!vlan_group_get_device(adapter->vlgrp, vid))
2172                         continue;
2173                 e1000_vlan_rx_add_vid(adapter->netdev, vid);
2174         }
2175 }
2176
2177 static void e1000_init_manageability(struct e1000_adapter *adapter)
2178 {
2179         struct e1000_hw *hw = &adapter->hw;
2180         u32 manc, manc2h;
2181
2182         if (!(adapter->flags & FLAG_MNG_PT_ENABLED))
2183                 return;
2184
2185         manc = er32(MANC);
2186
2187         /*
2188          * enable receiving management packets to the host. this will probably
2189          * generate destination unreachable messages from the host OS, but
2190          * the packets will be handled on SMBUS
2191          */
2192         manc |= E1000_MANC_EN_MNG2HOST;
2193         manc2h = er32(MANC2H);
2194 #define E1000_MNG2HOST_PORT_623 (1 << 5)
2195 #define E1000_MNG2HOST_PORT_664 (1 << 6)
2196         manc2h |= E1000_MNG2HOST_PORT_623;
2197         manc2h |= E1000_MNG2HOST_PORT_664;
2198         ew32(MANC2H, manc2h);
2199         ew32(MANC, manc);
2200 }
2201
2202 /**
2203  * e1000_configure_tx - Configure 8254x Transmit Unit after Reset
2204  * @adapter: board private structure
2205  *
2206  * Configure the Tx unit of the MAC after a reset.
2207  **/
2208 static void e1000_configure_tx(struct e1000_adapter *adapter)
2209 {
2210         struct e1000_hw *hw = &adapter->hw;
2211         struct e1000_ring *tx_ring = adapter->tx_ring;
2212         u64 tdba;
2213         u32 tdlen, tctl, tipg, tarc;
2214         u32 ipgr1, ipgr2;
2215
2216         /* Setup the HW Tx Head and Tail descriptor pointers */
2217         tdba = tx_ring->dma;
2218         tdlen = tx_ring->count * sizeof(struct e1000_tx_desc);
2219         ew32(TDBAL, (tdba & DMA_32BIT_MASK));
2220         ew32(TDBAH, (tdba >> 32));
2221         ew32(TDLEN, tdlen);
2222         ew32(TDH, 0);
2223         ew32(TDT, 0);
2224         tx_ring->head = E1000_TDH;
2225         tx_ring->tail = E1000_TDT;
2226
2227         /* Set the default values for the Tx Inter Packet Gap timer */
2228         tipg = DEFAULT_82543_TIPG_IPGT_COPPER;          /*  8  */
2229         ipgr1 = DEFAULT_82543_TIPG_IPGR1;               /*  8  */
2230         ipgr2 = DEFAULT_82543_TIPG_IPGR2;               /*  6  */
2231
2232         if (adapter->flags & FLAG_TIPG_MEDIUM_FOR_80003ESLAN)
2233                 ipgr2 = DEFAULT_80003ES2LAN_TIPG_IPGR2; /*  7  */
2234
2235         tipg |= ipgr1 << E1000_TIPG_IPGR1_SHIFT;
2236         tipg |= ipgr2 << E1000_TIPG_IPGR2_SHIFT;
2237         ew32(TIPG, tipg);
2238
2239         /* Set the Tx Interrupt Delay register */
2240         ew32(TIDV, adapter->tx_int_delay);
2241         /* Tx irq moderation */
2242         ew32(TADV, adapter->tx_abs_int_delay);
2243
2244         /* Program the Transmit Control Register */
2245         tctl = er32(TCTL);
2246         tctl &= ~E1000_TCTL_CT;
2247         tctl |= E1000_TCTL_PSP | E1000_TCTL_RTLC |
2248                 (E1000_COLLISION_THRESHOLD << E1000_CT_SHIFT);
2249
2250         if (adapter->flags & FLAG_TARC_SPEED_MODE_BIT) {
2251                 tarc = er32(TARC(0));
2252                 /*
2253                  * set the speed mode bit, we'll clear it if we're not at
2254                  * gigabit link later
2255                  */
2256 #define SPEED_MODE_BIT (1 << 21)
2257                 tarc |= SPEED_MODE_BIT;
2258                 ew32(TARC(0), tarc);
2259         }
2260
2261         /* errata: program both queues to unweighted RR */
2262         if (adapter->flags & FLAG_TARC_SET_BIT_ZERO) {
2263                 tarc = er32(TARC(0));
2264                 tarc |= 1;
2265                 ew32(TARC(0), tarc);
2266                 tarc = er32(TARC(1));
2267                 tarc |= 1;
2268                 ew32(TARC(1), tarc);
2269         }
2270
2271         e1000e_config_collision_dist(hw);
2272
2273         /* Setup Transmit Descriptor Settings for eop descriptor */
2274         adapter->txd_cmd = E1000_TXD_CMD_EOP | E1000_TXD_CMD_IFCS;
2275
2276         /* only set IDE if we are delaying interrupts using the timers */
2277         if (adapter->tx_int_delay)
2278                 adapter->txd_cmd |= E1000_TXD_CMD_IDE;
2279
2280         /* enable Report Status bit */
2281         adapter->txd_cmd |= E1000_TXD_CMD_RS;
2282
2283         ew32(TCTL, tctl);
2284
2285         adapter->tx_queue_len = adapter->netdev->tx_queue_len;
2286 }
2287
2288 /**
2289  * e1000_setup_rctl - configure the receive control registers
2290  * @adapter: Board private structure
2291  **/
2292 #define PAGE_USE_COUNT(S) (((S) >> PAGE_SHIFT) + \
2293                            (((S) & (PAGE_SIZE - 1)) ? 1 : 0))
2294 static void e1000_setup_rctl(struct e1000_adapter *adapter)
2295 {
2296         struct e1000_hw *hw = &adapter->hw;
2297         u32 rctl, rfctl;
2298         u32 psrctl = 0;
2299         u32 pages = 0;
2300
2301         /* Program MC offset vector base */
2302         rctl = er32(RCTL);
2303         rctl &= ~(3 << E1000_RCTL_MO_SHIFT);
2304         rctl |= E1000_RCTL_EN | E1000_RCTL_BAM |
2305                 E1000_RCTL_LBM_NO | E1000_RCTL_RDMTS_HALF |
2306                 (adapter->hw.mac.mc_filter_type << E1000_RCTL_MO_SHIFT);
2307
2308         /* Do not Store bad packets */
2309         rctl &= ~E1000_RCTL_SBP;
2310
2311         /* Enable Long Packet receive */
2312         if (adapter->netdev->mtu <= ETH_DATA_LEN)
2313                 rctl &= ~E1000_RCTL_LPE;
2314         else
2315                 rctl |= E1000_RCTL_LPE;
2316
2317         /* Some systems expect that the CRC is included in SMBUS traffic. The
2318          * hardware strips the CRC before sending to both SMBUS (BMC) and to
2319          * host memory when this is enabled
2320          */
2321         if (adapter->flags2 & FLAG2_CRC_STRIPPING)
2322                 rctl |= E1000_RCTL_SECRC;
2323
2324         /* Setup buffer sizes */
2325         rctl &= ~E1000_RCTL_SZ_4096;
2326         rctl |= E1000_RCTL_BSEX;
2327         switch (adapter->rx_buffer_len) {
2328         case 256:
2329                 rctl |= E1000_RCTL_SZ_256;
2330                 rctl &= ~E1000_RCTL_BSEX;
2331                 break;
2332         case 512:
2333                 rctl |= E1000_RCTL_SZ_512;
2334                 rctl &= ~E1000_RCTL_BSEX;
2335                 break;
2336         case 1024:
2337                 rctl |= E1000_RCTL_SZ_1024;
2338                 rctl &= ~E1000_RCTL_BSEX;
2339                 break;
2340         case 2048:
2341         default:
2342                 rctl |= E1000_RCTL_SZ_2048;
2343                 rctl &= ~E1000_RCTL_BSEX;
2344                 break;
2345         case 4096:
2346                 rctl |= E1000_RCTL_SZ_4096;
2347                 break;
2348         case 8192:
2349                 rctl |= E1000_RCTL_SZ_8192;
2350                 break;
2351         case 16384:
2352                 rctl |= E1000_RCTL_SZ_16384;
2353                 break;
2354         }
2355
2356         /*
2357          * 82571 and greater support packet-split where the protocol
2358          * header is placed in skb->data and the packet data is
2359          * placed in pages hanging off of skb_shinfo(skb)->nr_frags.
2360          * In the case of a non-split, skb->data is linearly filled,
2361          * followed by the page buffers.  Therefore, skb->data is
2362          * sized to hold the largest protocol header.
2363          *
2364          * allocations using alloc_page take too long for regular MTU
2365          * so only enable packet split for jumbo frames
2366          *
2367          * Using pages when the page size is greater than 16k wastes
2368          * a lot of memory, since we allocate 3 pages at all times
2369          * per packet.
2370          */
2371         pages = PAGE_USE_COUNT(adapter->netdev->mtu);
2372         if (!(adapter->flags & FLAG_IS_ICH) && (pages <= 3) &&
2373             (PAGE_SIZE <= 16384) && (rctl & E1000_RCTL_LPE))
2374                 adapter->rx_ps_pages = pages;
2375         else
2376                 adapter->rx_ps_pages = 0;
2377
2378         if (adapter->rx_ps_pages) {
2379                 /* Configure extra packet-split registers */
2380                 rfctl = er32(RFCTL);
2381                 rfctl |= E1000_RFCTL_EXTEN;
2382                 /*
2383                  * disable packet split support for IPv6 extension headers,
2384                  * because some malformed IPv6 headers can hang the Rx
2385                  */
2386                 rfctl |= (E1000_RFCTL_IPV6_EX_DIS |
2387                           E1000_RFCTL_NEW_IPV6_EXT_DIS);
2388
2389                 ew32(RFCTL, rfctl);
2390
2391                 /* Enable Packet split descriptors */
2392                 rctl |= E1000_RCTL_DTYP_PS;
2393
2394                 psrctl |= adapter->rx_ps_bsize0 >>
2395                         E1000_PSRCTL_BSIZE0_SHIFT;
2396
2397                 switch (adapter->rx_ps_pages) {
2398                 case 3:
2399                         psrctl |= PAGE_SIZE <<
2400                                 E1000_PSRCTL_BSIZE3_SHIFT;
2401                 case 2:
2402                         psrctl |= PAGE_SIZE <<
2403                                 E1000_PSRCTL_BSIZE2_SHIFT;
2404                 case 1:
2405                         psrctl |= PAGE_SIZE >>
2406                                 E1000_PSRCTL_BSIZE1_SHIFT;
2407                         break;
2408                 }
2409
2410                 ew32(PSRCTL, psrctl);
2411         }
2412
2413         ew32(RCTL, rctl);
2414         /* just started the receive unit, no need to restart */
2415         adapter->flags &= ~FLAG_RX_RESTART_NOW;
2416 }
2417
2418 /**
2419  * e1000_configure_rx - Configure Receive Unit after Reset
2420  * @adapter: board private structure
2421  *
2422  * Configure the Rx unit of the MAC after a reset.
2423  **/
2424 static void e1000_configure_rx(struct e1000_adapter *adapter)
2425 {
2426         struct e1000_hw *hw = &adapter->hw;
2427         struct e1000_ring *rx_ring = adapter->rx_ring;
2428         u64 rdba;
2429         u32 rdlen, rctl, rxcsum, ctrl_ext;
2430
2431         if (adapter->rx_ps_pages) {
2432                 /* this is a 32 byte descriptor */
2433                 rdlen = rx_ring->count *
2434                         sizeof(union e1000_rx_desc_packet_split);
2435                 adapter->clean_rx = e1000_clean_rx_irq_ps;
2436                 adapter->alloc_rx_buf = e1000_alloc_rx_buffers_ps;
2437         } else if (adapter->netdev->mtu > ETH_FRAME_LEN + ETH_FCS_LEN) {
2438                 rdlen = rx_ring->count * sizeof(struct e1000_rx_desc);
2439                 adapter->clean_rx = e1000_clean_jumbo_rx_irq;
2440                 adapter->alloc_rx_buf = e1000_alloc_jumbo_rx_buffers;
2441         } else {
2442                 rdlen = rx_ring->count * sizeof(struct e1000_rx_desc);
2443                 adapter->clean_rx = e1000_clean_rx_irq;
2444                 adapter->alloc_rx_buf = e1000_alloc_rx_buffers;
2445         }
2446
2447         /* disable receives while setting up the descriptors */
2448         rctl = er32(RCTL);
2449         ew32(RCTL, rctl & ~E1000_RCTL_EN);
2450         e1e_flush();
2451         msleep(10);
2452
2453         /* set the Receive Delay Timer Register */
2454         ew32(RDTR, adapter->rx_int_delay);
2455
2456         /* irq moderation */
2457         ew32(RADV, adapter->rx_abs_int_delay);
2458         if (adapter->itr_setting != 0)
2459                 ew32(ITR, 1000000000 / (adapter->itr * 256));
2460
2461         ctrl_ext = er32(CTRL_EXT);
2462         /* Reset delay timers after every interrupt */
2463         ctrl_ext |= E1000_CTRL_EXT_INT_TIMER_CLR;
2464         /* Auto-Mask interrupts upon ICR access */
2465         ctrl_ext |= E1000_CTRL_EXT_IAME;
2466         ew32(IAM, 0xffffffff);
2467         ew32(CTRL_EXT, ctrl_ext);
2468         e1e_flush();
2469
2470         /*
2471          * Setup the HW Rx Head and Tail Descriptor Pointers and
2472          * the Base and Length of the Rx Descriptor Ring
2473          */
2474         rdba = rx_ring->dma;
2475         ew32(RDBAL, (rdba & DMA_32BIT_MASK));
2476         ew32(RDBAH, (rdba >> 32));
2477         ew32(RDLEN, rdlen);
2478         ew32(RDH, 0);
2479         ew32(RDT, 0);
2480         rx_ring->head = E1000_RDH;
2481         rx_ring->tail = E1000_RDT;
2482
2483         /* Enable Receive Checksum Offload for TCP and UDP */
2484         rxcsum = er32(RXCSUM);
2485         if (adapter->flags & FLAG_RX_CSUM_ENABLED) {
2486                 rxcsum |= E1000_RXCSUM_TUOFL;
2487
2488                 /*
2489                  * IPv4 payload checksum for UDP fragments must be
2490                  * used in conjunction with packet-split.
2491                  */
2492                 if (adapter->rx_ps_pages)
2493                         rxcsum |= E1000_RXCSUM_IPPCSE;
2494         } else {
2495                 rxcsum &= ~E1000_RXCSUM_TUOFL;
2496                 /* no need to clear IPPCSE as it defaults to 0 */
2497         }
2498         ew32(RXCSUM, rxcsum);
2499
2500         /*
2501          * Enable early receives on supported devices, only takes effect when
2502          * packet size is equal or larger than the specified value (in 8 byte
2503          * units), e.g. using jumbo frames when setting to E1000_ERT_2048
2504          */
2505         if ((adapter->flags & FLAG_HAS_ERT) &&
2506             (adapter->netdev->mtu > ETH_DATA_LEN)) {
2507                 u32 rxdctl = er32(RXDCTL(0));
2508                 ew32(RXDCTL(0), rxdctl | 0x3);
2509                 ew32(ERT, E1000_ERT_2048 | (1 << 13));
2510                 /*
2511                  * With jumbo frames and early-receive enabled, excessive
2512                  * C4->C2 latencies result in dropped transactions.
2513                  */
2514                 pm_qos_update_requirement(PM_QOS_CPU_DMA_LATENCY,
2515                                           e1000e_driver_name, 55);
2516         } else {
2517                 pm_qos_update_requirement(PM_QOS_CPU_DMA_LATENCY,
2518                                           e1000e_driver_name,
2519                                           PM_QOS_DEFAULT_VALUE);
2520         }
2521
2522         /* Enable Receives */
2523         ew32(RCTL, rctl);
2524 }
2525
2526 /**
2527  *  e1000_update_mc_addr_list - Update Multicast addresses
2528  *  @hw: pointer to the HW structure
2529  *  @mc_addr_list: array of multicast addresses to program
2530  *  @mc_addr_count: number of multicast addresses to program
2531  *  @rar_used_count: the first RAR register free to program
2532  *  @rar_count: total number of supported Receive Address Registers
2533  *
2534  *  Updates the Receive Address Registers and Multicast Table Array.
2535  *  The caller must have a packed mc_addr_list of multicast addresses.
2536  *  The parameter rar_count will usually be hw->mac.rar_entry_count
2537  *  unless there are workarounds that change this.  Currently no func pointer
2538  *  exists and all implementations are handled in the generic version of this
2539  *  function.
2540  **/
2541 static void e1000_update_mc_addr_list(struct e1000_hw *hw, u8 *mc_addr_list,
2542                                       u32 mc_addr_count, u32 rar_used_count,
2543                                       u32 rar_count)
2544 {
2545         hw->mac.ops.update_mc_addr_list(hw, mc_addr_list, mc_addr_count,
2546                                         rar_used_count, rar_count);
2547 }
2548
2549 /**
2550  * e1000_set_multi - Multicast and Promiscuous mode set
2551  * @netdev: network interface device structure
2552  *
2553  * The set_multi entry point is called whenever the multicast address
2554  * list or the network interface flags are updated.  This routine is
2555  * responsible for configuring the hardware for proper multicast,
2556  * promiscuous mode, and all-multi behavior.
2557  **/
2558 static void e1000_set_multi(struct net_device *netdev)
2559 {
2560         struct e1000_adapter *adapter = netdev_priv(netdev);
2561         struct e1000_hw *hw = &adapter->hw;
2562         struct e1000_mac_info *mac = &hw->mac;
2563         struct dev_mc_list *mc_ptr;
2564         u8  *mta_list;
2565         u32 rctl;
2566         int i;
2567
2568         /* Check for Promiscuous and All Multicast modes */
2569
2570         rctl = er32(RCTL);
2571
2572         if (netdev->flags & IFF_PROMISC) {
2573                 rctl |= (E1000_RCTL_UPE | E1000_RCTL_MPE);
2574                 rctl &= ~E1000_RCTL_VFE;
2575         } else {
2576                 if (netdev->flags & IFF_ALLMULTI) {
2577                         rctl |= E1000_RCTL_MPE;
2578                         rctl &= ~E1000_RCTL_UPE;
2579                 } else {
2580                         rctl &= ~(E1000_RCTL_UPE | E1000_RCTL_MPE);
2581                 }
2582                 if (adapter->flags & FLAG_HAS_HW_VLAN_FILTER)
2583                         rctl |= E1000_RCTL_VFE;
2584         }
2585
2586         ew32(RCTL, rctl);
2587
2588         if (netdev->mc_count) {
2589                 mta_list = kmalloc(netdev->mc_count * 6, GFP_ATOMIC);
2590                 if (!mta_list)
2591                         return;
2592
2593                 /* prepare a packed array of only addresses. */
2594                 mc_ptr = netdev->mc_list;
2595
2596                 for (i = 0; i < netdev->mc_count; i++) {
2597                         if (!mc_ptr)
2598                                 break;
2599                         memcpy(mta_list + (i*ETH_ALEN), mc_ptr->dmi_addr,
2600                                ETH_ALEN);
2601                         mc_ptr = mc_ptr->next;
2602                 }
2603
2604                 e1000_update_mc_addr_list(hw, mta_list, i, 1,
2605                                           mac->rar_entry_count);
2606                 kfree(mta_list);
2607         } else {
2608                 /*
2609                  * if we're called from probe, we might not have
2610                  * anything to do here, so clear out the list
2611                  */
2612                 e1000_update_mc_addr_list(hw, NULL, 0, 1, mac->rar_entry_count);
2613         }
2614 }
2615
2616 /**
2617  * e1000_configure - configure the hardware for Rx and Tx
2618  * @adapter: private board structure
2619  **/
2620 static void e1000_configure(struct e1000_adapter *adapter)
2621 {
2622         e1000_set_multi(adapter->netdev);
2623
2624         e1000_restore_vlan(adapter);
2625         e1000_init_manageability(adapter);
2626
2627         e1000_configure_tx(adapter);
2628         e1000_setup_rctl(adapter);
2629         e1000_configure_rx(adapter);
2630         adapter->alloc_rx_buf(adapter, e1000_desc_unused(adapter->rx_ring));
2631 }
2632
2633 /**
2634  * e1000e_power_up_phy - restore link in case the phy was powered down
2635  * @adapter: address of board private structure
2636  *
2637  * The phy may be powered down to save power and turn off link when the
2638  * driver is unloaded and wake on lan is not enabled (among others)
2639  * *** this routine MUST be followed by a call to e1000e_reset ***
2640  **/
2641 void e1000e_power_up_phy(struct e1000_adapter *adapter)
2642 {
2643         u16 mii_reg = 0;
2644
2645         /* Just clear the power down bit to wake the phy back up */
2646         if (adapter->hw.phy.media_type == e1000_media_type_copper) {
2647                 /*
2648                  * According to the manual, the phy will retain its
2649                  * settings across a power-down/up cycle
2650                  */
2651                 e1e_rphy(&adapter->hw, PHY_CONTROL, &mii_reg);
2652                 mii_reg &= ~MII_CR_POWER_DOWN;
2653                 e1e_wphy(&adapter->hw, PHY_CONTROL, mii_reg);
2654         }
2655
2656         adapter->hw.mac.ops.setup_link(&adapter->hw);
2657 }
2658
2659 /**
2660  * e1000_power_down_phy - Power down the PHY
2661  *
2662  * Power down the PHY so no link is implied when interface is down
2663  * The PHY cannot be powered down is management or WoL is active
2664  */
2665 static void e1000_power_down_phy(struct e1000_adapter *adapter)
2666 {
2667         struct e1000_hw *hw = &adapter->hw;
2668         u16 mii_reg;
2669
2670         /* WoL is enabled */
2671         if (adapter->wol)
2672                 return;
2673
2674         /* non-copper PHY? */
2675         if (adapter->hw.phy.media_type != e1000_media_type_copper)
2676                 return;
2677
2678         /* reset is blocked because of a SoL/IDER session */
2679         if (e1000e_check_mng_mode(hw) || e1000_check_reset_block(hw))
2680                 return;
2681
2682         /* manageability (AMT) is enabled */
2683         if (er32(MANC) & E1000_MANC_SMBUS_EN)
2684                 return;
2685
2686         /* power down the PHY */
2687         e1e_rphy(hw, PHY_CONTROL, &mii_reg);
2688         mii_reg |= MII_CR_POWER_DOWN;
2689         e1e_wphy(hw, PHY_CONTROL, mii_reg);
2690         mdelay(1);
2691 }
2692
2693 /**
2694  * e1000e_reset - bring the hardware into a known good state
2695  *
2696  * This function boots the hardware and enables some settings that
2697  * require a configuration cycle of the hardware - those cannot be
2698  * set/changed during runtime. After reset the device needs to be
2699  * properly configured for Rx, Tx etc.
2700  */
2701 void e1000e_reset(struct e1000_adapter *adapter)
2702 {
2703         struct e1000_mac_info *mac = &adapter->hw.mac;
2704         struct e1000_fc_info *fc = &adapter->hw.fc;
2705         struct e1000_hw *hw = &adapter->hw;
2706         u32 tx_space, min_tx_space, min_rx_space;
2707         u32 pba = adapter->pba;
2708         u16 hwm;
2709
2710         /* reset Packet Buffer Allocation to default */
2711         ew32(PBA, pba);
2712
2713         if (adapter->max_frame_size > ETH_FRAME_LEN + ETH_FCS_LEN) {
2714                 /*
2715                  * To maintain wire speed transmits, the Tx FIFO should be
2716                  * large enough to accommodate two full transmit packets,
2717                  * rounded up to the next 1KB and expressed in KB.  Likewise,
2718                  * the Rx FIFO should be large enough to accommodate at least
2719                  * one full receive packet and is similarly rounded up and
2720                  * expressed in KB.
2721                  */
2722                 pba = er32(PBA);
2723                 /* upper 16 bits has Tx packet buffer allocation size in KB */
2724                 tx_space = pba >> 16;
2725                 /* lower 16 bits has Rx packet buffer allocation size in KB */
2726                 pba &= 0xffff;
2727                 /*
2728                  * the Tx fifo also stores 16 bytes of information about the tx
2729                  * but don't include ethernet FCS because hardware appends it
2730                  */
2731                 min_tx_space = (adapter->max_frame_size +
2732                                 sizeof(struct e1000_tx_desc) -
2733                                 ETH_FCS_LEN) * 2;
2734                 min_tx_space = ALIGN(min_tx_space, 1024);
2735                 min_tx_space >>= 10;
2736                 /* software strips receive CRC, so leave room for it */
2737                 min_rx_space = adapter->max_frame_size;
2738                 min_rx_space = ALIGN(min_rx_space, 1024);
2739                 min_rx_space >>= 10;
2740
2741                 /*
2742                  * If current Tx allocation is less than the min Tx FIFO size,
2743                  * and the min Tx FIFO size is less than the current Rx FIFO
2744                  * allocation, take space away from current Rx allocation
2745                  */
2746                 if ((tx_space < min_tx_space) &&
2747                     ((min_tx_space - tx_space) < pba)) {
2748                         pba -= min_tx_space - tx_space;
2749
2750                         /*
2751                          * if short on Rx space, Rx wins and must trump tx
2752                          * adjustment or use Early Receive if available
2753                          */
2754                         if ((pba < min_rx_space) &&
2755                             (!(adapter->flags & FLAG_HAS_ERT)))
2756                                 /* ERT enabled in e1000_configure_rx */
2757                                 pba = min_rx_space;
2758                 }
2759
2760                 ew32(PBA, pba);
2761         }
2762
2763
2764         /*
2765          * flow control settings
2766          *
2767          * The high water mark must be low enough to fit one full frame
2768          * (or the size used for early receive) above it in the Rx FIFO.
2769          * Set it to the lower of:
2770          * - 90% of the Rx FIFO size, and
2771          * - the full Rx FIFO size minus the early receive size (for parts
2772          *   with ERT support assuming ERT set to E1000_ERT_2048), or
2773          * - the full Rx FIFO size minus one full frame
2774          */
2775         if (adapter->flags & FLAG_HAS_ERT)
2776                 hwm = min(((pba << 10) * 9 / 10),
2777                           ((pba << 10) - (E1000_ERT_2048 << 3)));
2778         else
2779                 hwm = min(((pba << 10) * 9 / 10),
2780                           ((pba << 10) - adapter->max_frame_size));
2781
2782         fc->high_water = hwm & 0xFFF8; /* 8-byte granularity */
2783         fc->low_water = fc->high_water - 8;
2784
2785         if (adapter->flags & FLAG_DISABLE_FC_PAUSE_TIME)
2786                 fc->pause_time = 0xFFFF;
2787         else
2788                 fc->pause_time = E1000_FC_PAUSE_TIME;
2789         fc->send_xon = 1;
2790         fc->type = fc->original_type;
2791
2792         /* Allow time for pending master requests to run */
2793         mac->ops.reset_hw(hw);
2794
2795         /*
2796          * For parts with AMT enabled, let the firmware know
2797          * that the network interface is in control
2798          */
2799         if (adapter->flags & FLAG_HAS_AMT)
2800                 e1000_get_hw_control(adapter);
2801
2802         ew32(WUC, 0);
2803
2804         if (mac->ops.init_hw(hw))
2805                 e_err("Hardware Error\n");
2806
2807         e1000_update_mng_vlan(adapter);
2808
2809         /* Enable h/w to recognize an 802.1Q VLAN Ethernet packet */
2810         ew32(VET, ETH_P_8021Q);
2811
2812         e1000e_reset_adaptive(hw);
2813         e1000_get_phy_info(hw);
2814
2815         if (!(adapter->flags & FLAG_SMART_POWER_DOWN)) {
2816                 u16 phy_data = 0;
2817                 /*
2818                  * speed up time to link by disabling smart power down, ignore
2819                  * the return value of this function because there is nothing
2820                  * different we would do if it failed
2821                  */
2822                 e1e_rphy(hw, IGP02E1000_PHY_POWER_MGMT, &phy_data);
2823                 phy_data &= ~IGP02E1000_PM_SPD;
2824                 e1e_wphy(hw, IGP02E1000_PHY_POWER_MGMT, phy_data);
2825         }
2826 }
2827
2828 int e1000e_up(struct e1000_adapter *adapter)
2829 {
2830         struct e1000_hw *hw = &adapter->hw;
2831
2832         /* hardware has been reset, we need to reload some things */
2833         e1000_configure(adapter);
2834
2835         clear_bit(__E1000_DOWN, &adapter->state);
2836
2837         napi_enable(&adapter->napi);
2838         if (adapter->msix_entries)
2839                 e1000_configure_msix(adapter);
2840         e1000_irq_enable(adapter);
2841
2842         /* fire a link change interrupt to start the watchdog */
2843         ew32(ICS, E1000_ICS_LSC);
2844         return 0;
2845 }
2846
2847 void e1000e_down(struct e1000_adapter *adapter)
2848 {
2849         struct net_device *netdev = adapter->netdev;
2850         struct e1000_hw *hw = &adapter->hw;
2851         u32 tctl, rctl;
2852
2853         /*
2854          * signal that we're down so the interrupt handler does not
2855          * reschedule our watchdog timer
2856          */
2857         set_bit(__E1000_DOWN, &adapter->state);
2858
2859         /* disable receives in the hardware */
2860         rctl = er32(RCTL);
2861         ew32(RCTL, rctl & ~E1000_RCTL_EN);
2862         /* flush and sleep below */
2863
2864         netif_tx_stop_all_queues(netdev);
2865
2866         /* disable transmits in the hardware */
2867         tctl = er32(TCTL);
2868         tctl &= ~E1000_TCTL_EN;
2869         ew32(TCTL, tctl);
2870         /* flush both disables and wait for them to finish */
2871         e1e_flush();
2872         msleep(10);
2873
2874         napi_disable(&adapter->napi);
2875         e1000_irq_disable(adapter);
2876
2877         del_timer_sync(&adapter->watchdog_timer);
2878         del_timer_sync(&adapter->phy_info_timer);
2879
2880         netdev->tx_queue_len = adapter->tx_queue_len;
2881         netif_carrier_off(netdev);
2882         adapter->link_speed = 0;
2883         adapter->link_duplex = 0;
2884
2885         if (!pci_channel_offline(adapter->pdev))
2886                 e1000e_reset(adapter);
2887         e1000_clean_tx_ring(adapter);
2888         e1000_clean_rx_ring(adapter);
2889
2890         /*
2891          * TODO: for power management, we could drop the link and
2892          * pci_disable_device here.
2893          */
2894 }
2895
2896 void e1000e_reinit_locked(struct e1000_adapter *adapter)
2897 {
2898         might_sleep();
2899         while (test_and_set_bit(__E1000_RESETTING, &adapter->state))
2900                 msleep(1);
2901         e1000e_down(adapter);
2902         e1000e_up(adapter);
2903         clear_bit(__E1000_RESETTING, &adapter->state);
2904 }
2905
2906 /**
2907  * e1000_sw_init - Initialize general software structures (struct e1000_adapter)
2908  * @adapter: board private structure to initialize
2909  *
2910  * e1000_sw_init initializes the Adapter private data structure.
2911  * Fields are initialized based on PCI device information and
2912  * OS network device settings (MTU size).
2913  **/
2914 static int __devinit e1000_sw_init(struct e1000_adapter *adapter)
2915 {
2916         struct net_device *netdev = adapter->netdev;
2917
2918         adapter->rx_buffer_len = ETH_FRAME_LEN + VLAN_HLEN + ETH_FCS_LEN;
2919         adapter->rx_ps_bsize0 = 128;
2920         adapter->max_frame_size = netdev->mtu + ETH_HLEN + ETH_FCS_LEN;
2921         adapter->min_frame_size = ETH_ZLEN + ETH_FCS_LEN;
2922
2923         e1000e_set_interrupt_capability(adapter);
2924
2925         if (e1000_alloc_queues(adapter))
2926                 return -ENOMEM;
2927
2928         spin_lock_init(&adapter->tx_queue_lock);
2929
2930         /* Explicitly disable IRQ since the NIC can be in any state. */
2931         e1000_irq_disable(adapter);
2932
2933         set_bit(__E1000_DOWN, &adapter->state);
2934         return 0;
2935 }
2936
2937 /**
2938  * e1000_intr_msi_test - Interrupt Handler
2939  * @irq: interrupt number
2940  * @data: pointer to a network interface device structure
2941  **/
2942 static irqreturn_t e1000_intr_msi_test(int irq, void *data)
2943 {
2944         struct net_device *netdev = data;
2945         struct e1000_adapter *adapter = netdev_priv(netdev);
2946         struct e1000_hw *hw = &adapter->hw;
2947         u32 icr = er32(ICR);
2948
2949         e_dbg("%s: icr is %08X\n", netdev->name, icr);
2950         if (icr & E1000_ICR_RXSEQ) {
2951                 adapter->flags &= ~FLAG_MSI_TEST_FAILED;
2952                 wmb();
2953         }
2954
2955         return IRQ_HANDLED;
2956 }
2957
2958 /**
2959  * e1000_test_msi_interrupt - Returns 0 for successful test
2960  * @adapter: board private struct
2961  *
2962  * code flow taken from tg3.c
2963  **/
2964 static int e1000_test_msi_interrupt(struct e1000_adapter *adapter)
2965 {
2966         struct net_device *netdev = adapter->netdev;
2967         struct e1000_hw *hw = &adapter->hw;
2968         int err;
2969
2970         /* poll_enable hasn't been called yet, so don't need disable */
2971         /* clear any pending events */
2972         er32(ICR);
2973
2974         /* free the real vector and request a test handler */
2975         e1000_free_irq(adapter);
2976         e1000e_reset_interrupt_capability(adapter);
2977
2978         /* Assume that the test fails, if it succeeds then the test
2979          * MSI irq handler will unset this flag */
2980         adapter->flags |= FLAG_MSI_TEST_FAILED;
2981
2982         err = pci_enable_msi(adapter->pdev);
2983         if (err)
2984                 goto msi_test_failed;
2985
2986         err = request_irq(adapter->pdev->irq, &e1000_intr_msi_test, 0,
2987                           netdev->name, netdev);
2988         if (err) {
2989                 pci_disable_msi(adapter->pdev);
2990                 goto msi_test_failed;
2991         }
2992
2993         wmb();
2994
2995         e1000_irq_enable(adapter);
2996
2997         /* fire an unusual interrupt on the test handler */
2998         ew32(ICS, E1000_ICS_RXSEQ);
2999         e1e_flush();
3000         msleep(50);
3001
3002         e1000_irq_disable(adapter);
3003
3004         rmb();
3005
3006         if (adapter->flags & FLAG_MSI_TEST_FAILED) {
3007                 adapter->int_mode = E1000E_INT_MODE_LEGACY;
3008                 err = -EIO;
3009                 e_info("MSI interrupt test failed!\n");
3010         }
3011
3012         free_irq(adapter->pdev->irq, netdev);
3013         pci_disable_msi(adapter->pdev);
3014
3015         if (err == -EIO)
3016                 goto msi_test_failed;
3017
3018         /* okay so the test worked, restore settings */
3019         e_dbg("%s: MSI interrupt test succeeded!\n", netdev->name);
3020 msi_test_failed:
3021         e1000e_set_interrupt_capability(adapter);
3022         e1000_request_irq(adapter);
3023         return err;
3024 }
3025
3026 /**
3027  * e1000_test_msi - Returns 0 if MSI test succeeds or INTx mode is restored
3028  * @adapter: board private struct
3029  *
3030  * code flow taken from tg3.c, called with e1000 interrupts disabled.
3031  **/
3032 static int e1000_test_msi(struct e1000_adapter *adapter)
3033 {
3034         int err;
3035         u16 pci_cmd;
3036
3037         if (!(adapter->flags & FLAG_MSI_ENABLED))
3038                 return 0;
3039
3040         /* disable SERR in case the MSI write causes a master abort */
3041         pci_read_config_word(adapter->pdev, PCI_COMMAND, &pci_cmd);
3042         pci_write_config_word(adapter->pdev, PCI_COMMAND,
3043                               pci_cmd & ~PCI_COMMAND_SERR);
3044
3045         err = e1000_test_msi_interrupt(adapter);
3046
3047         /* restore previous setting of command word */
3048         pci_write_config_word(adapter->pdev, PCI_COMMAND, pci_cmd);
3049
3050         /* success ! */
3051         if (!err)
3052                 return 0;
3053
3054         /* EIO means MSI test failed */
3055         if (err != -EIO)
3056                 return err;
3057
3058         /* back to INTx mode */
3059         e_warn("MSI interrupt test failed, using legacy interrupt.\n");
3060
3061         e1000_free_irq(adapter);
3062
3063         err = e1000_request_irq(adapter);
3064
3065         return err;
3066 }
3067
3068 /**
3069  * e1000_open - Called when a network interface is made active
3070  * @netdev: network interface device structure
3071  *
3072  * Returns 0 on success, negative value on failure
3073  *
3074  * The open entry point is called when a network interface is made
3075  * active by the system (IFF_UP).  At this point all resources needed
3076  * for transmit and receive operations are allocated, the interrupt
3077  * handler is registered with the OS, the watchdog timer is started,
3078  * and the stack is notified that the interface is ready.
3079  **/
3080 static int e1000_open(struct net_device *netdev)
3081 {
3082         struct e1000_adapter *adapter = netdev_priv(netdev);
3083         struct e1000_hw *hw = &adapter->hw;
3084         int err;
3085
3086         /* disallow open during test */
3087         if (test_bit(__E1000_TESTING, &adapter->state))
3088                 return -EBUSY;
3089
3090         /* allocate transmit descriptors */
3091         err = e1000e_setup_tx_resources(adapter);
3092         if (err)
3093                 goto err_setup_tx;
3094
3095         /* allocate receive descriptors */
3096         err = e1000e_setup_rx_resources(adapter);
3097         if (err)
3098                 goto err_setup_rx;
3099
3100         e1000e_power_up_phy(adapter);
3101
3102         adapter->mng_vlan_id = E1000_MNG_VLAN_NONE;
3103         if ((adapter->hw.mng_cookie.status &
3104              E1000_MNG_DHCP_COOKIE_STATUS_VLAN))
3105                 e1000_update_mng_vlan(adapter);
3106
3107         /*
3108          * If AMT is enabled, let the firmware know that the network
3109          * interface is now open
3110          */
3111         if (adapter->flags & FLAG_HAS_AMT)
3112                 e1000_get_hw_control(adapter);
3113
3114         /*
3115          * before we allocate an interrupt, we must be ready to handle it.
3116          * Setting DEBUG_SHIRQ in the kernel makes it fire an interrupt
3117          * as soon as we call pci_request_irq, so we have to setup our
3118          * clean_rx handler before we do so.
3119          */
3120         e1000_configure(adapter);
3121
3122         err = e1000_request_irq(adapter);
3123         if (err)
3124                 goto err_req_irq;
3125
3126         /*
3127          * Work around PCIe errata with MSI interrupts causing some chipsets to
3128          * ignore e1000e MSI messages, which means we need to test our MSI
3129          * interrupt now
3130          */
3131         if (adapter->int_mode != E1000E_INT_MODE_LEGACY) {
3132                 err = e1000_test_msi(adapter);
3133                 if (err) {
3134                         e_err("Interrupt allocation failed\n");
3135                         goto err_req_irq;
3136                 }
3137         }
3138
3139         /* From here on the code is the same as e1000e_up() */
3140         clear_bit(__E1000_DOWN, &adapter->state);
3141
3142         napi_enable(&adapter->napi);
3143
3144         e1000_irq_enable(adapter);
3145
3146         netif_tx_start_all_queues(netdev);
3147
3148         /* fire a link status change interrupt to start the watchdog */
3149         ew32(ICS, E1000_ICS_LSC);
3150
3151         return 0;
3152
3153 err_req_irq:
3154         e1000_release_hw_control(adapter);
3155         e1000_power_down_phy(adapter);
3156         e1000e_free_rx_resources(adapter);
3157 err_setup_rx:
3158         e1000e_free_tx_resources(adapter);
3159 err_setup_tx:
3160         e1000e_reset(adapter);
3161
3162         return err;
3163 }
3164
3165 /**
3166  * e1000_close - Disables a network interface
3167  * @netdev: network interface device structure
3168  *
3169  * Returns 0, this is not allowed to fail
3170  *
3171  * The close entry point is called when an interface is de-activated
3172  * by the OS.  The hardware is still under the drivers control, but
3173  * needs to be disabled.  A global MAC reset is issued to stop the
3174  * hardware, and all transmit and receive resources are freed.
3175  **/
3176 static int e1000_close(struct net_device *netdev)
3177 {
3178         struct e1000_adapter *adapter = netdev_priv(netdev);
3179
3180         WARN_ON(test_bit(__E1000_RESETTING, &adapter->state));
3181         e1000e_down(adapter);
3182         e1000_power_down_phy(adapter);
3183         e1000_free_irq(adapter);
3184
3185         e1000e_free_tx_resources(adapter);
3186         e1000e_free_rx_resources(adapter);
3187
3188         /*
3189          * kill manageability vlan ID if supported, but not if a vlan with
3190          * the same ID is registered on the host OS (let 8021q kill it)
3191          */
3192         if ((adapter->hw.mng_cookie.status &
3193                           E1000_MNG_DHCP_COOKIE_STATUS_VLAN) &&
3194              !(adapter->vlgrp &&
3195                vlan_group_get_device(adapter->vlgrp, adapter->mng_vlan_id)))
3196                 e1000_vlan_rx_kill_vid(netdev, adapter->mng_vlan_id);
3197
3198         /*
3199          * If AMT is enabled, let the firmware know that the network
3200          * interface is now closed
3201          */
3202         if (adapter->flags & FLAG_HAS_AMT)
3203                 e1000_release_hw_control(adapter);
3204
3205         return 0;
3206 }
3207 /**
3208  * e1000_set_mac - Change the Ethernet Address of the NIC
3209  * @netdev: network interface device structure
3210  * @p: pointer to an address structure
3211  *
3212  * Returns 0 on success, negative on failure
3213  **/
3214 static int e1000_set_mac(struct net_device *netdev, void *p)
3215 {
3216         struct e1000_adapter *adapter = netdev_priv(netdev);
3217         struct sockaddr *addr = p;
3218
3219         if (!is_valid_ether_addr(addr->sa_data))
3220                 return -EADDRNOTAVAIL;
3221
3222         memcpy(netdev->dev_addr, addr->sa_data, netdev->addr_len);
3223         memcpy(adapter->hw.mac.addr, addr->sa_data, netdev->addr_len);
3224
3225         e1000e_rar_set(&adapter->hw, adapter->hw.mac.addr, 0);
3226
3227         if (adapter->flags & FLAG_RESET_OVERWRITES_LAA) {
3228                 /* activate the work around */
3229                 e1000e_set_laa_state_82571(&adapter->hw, 1);
3230
3231                 /*
3232                  * Hold a copy of the LAA in RAR[14] This is done so that
3233                  * between the time RAR[0] gets clobbered  and the time it
3234                  * gets fixed (in e1000_watchdog), the actual LAA is in one
3235                  * of the RARs and no incoming packets directed to this port
3236                  * are dropped. Eventually the LAA will be in RAR[0] and
3237                  * RAR[14]
3238                  */
3239                 e1000e_rar_set(&adapter->hw,
3240                               adapter->hw.mac.addr,
3241                               adapter->hw.mac.rar_entry_count - 1);
3242         }
3243
3244         return 0;
3245 }
3246
3247 /**
3248  * e1000e_update_phy_task - work thread to update phy
3249  * @work: pointer to our work struct
3250  *
3251  * this worker thread exists because we must acquire a
3252  * semaphore to read the phy, which we could msleep while
3253  * waiting for it, and we can't msleep in a timer.
3254  **/
3255 static void e1000e_update_phy_task(struct work_struct *work)
3256 {
3257         struct e1000_adapter *adapter = container_of(work,
3258                                         struct e1000_adapter, update_phy_task);
3259         e1000_get_phy_info(&adapter->hw);
3260 }
3261
3262 /*
3263  * Need to wait a few seconds after link up to get diagnostic information from
3264  * the phy
3265  */
3266 static void e1000_update_phy_info(unsigned long data)
3267 {
3268         struct e1000_adapter *adapter = (struct e1000_adapter *) data;
3269         schedule_work(&adapter->update_phy_task);
3270 }
3271
3272 /**
3273  * e1000e_update_stats - Update the board statistics counters
3274  * @adapter: board private structure
3275  **/
3276 void e1000e_update_stats(struct e1000_adapter *adapter)
3277 {
3278         struct e1000_hw *hw = &adapter->hw;
3279         struct pci_dev *pdev = adapter->pdev;
3280
3281         /*
3282          * Prevent stats update while adapter is being reset, or if the pci
3283          * connection is down.
3284          */
3285         if (adapter->link_speed == 0)
3286                 return;
3287         if (pci_channel_offline(pdev))
3288                 return;
3289
3290         adapter->stats.crcerrs += er32(CRCERRS);
3291         adapter->stats.gprc += er32(GPRC);
3292         adapter->stats.gorc += er32(GORCL);
3293         er32(GORCH); /* Clear gorc */
3294         adapter->stats.bprc += er32(BPRC);
3295         adapter->stats.mprc += er32(MPRC);
3296         adapter->stats.roc += er32(ROC);
3297
3298         adapter->stats.mpc += er32(MPC);
3299         adapter->stats.scc += er32(SCC);
3300         adapter->stats.ecol += er32(ECOL);
3301         adapter->stats.mcc += er32(MCC);
3302         adapter->stats.latecol += er32(LATECOL);
3303         adapter->stats.dc += er32(DC);
3304         adapter->stats.xonrxc += er32(XONRXC);
3305         adapter->stats.xontxc += er32(XONTXC);
3306         adapter->stats.xoffrxc += er32(XOFFRXC);
3307         adapter->stats.xofftxc += er32(XOFFTXC);
3308         adapter->stats.gptc += er32(GPTC);
3309         adapter->stats.gotc += er32(GOTCL);
3310         er32(GOTCH); /* Clear gotc */
3311         adapter->stats.rnbc += er32(RNBC);
3312         adapter->stats.ruc += er32(RUC);
3313
3314         adapter->stats.mptc += er32(MPTC);
3315         adapter->stats.bptc += er32(BPTC);
3316
3317         /* used for adaptive IFS */
3318
3319         hw->mac.tx_packet_delta = er32(TPT);
3320         adapter->stats.tpt += hw->mac.tx_packet_delta;
3321         hw->mac.collision_delta = er32(COLC);
3322         adapter->stats.colc += hw->mac.collision_delta;
3323
3324         adapter->stats.algnerrc += er32(ALGNERRC);
3325         adapter->stats.rxerrc += er32(RXERRC);
3326         if (hw->mac.type != e1000_82574)
3327                 adapter->stats.tncrs += er32(TNCRS);
3328         adapter->stats.cexterr += er32(CEXTERR);
3329         adapter->stats.tsctc += er32(TSCTC);
3330         adapter->stats.tsctfc += er32(TSCTFC);
3331
3332         /* Fill out the OS statistics structure */
3333         adapter->net_stats.multicast = adapter->stats.mprc;
3334         adapter->net_stats.collisions = adapter->stats.colc;
3335
3336         /* Rx Errors */
3337
3338         /*
3339          * RLEC on some newer hardware can be incorrect so build
3340          * our own version based on RUC and ROC
3341          */
3342         adapter->net_stats.rx_errors = adapter->stats.rxerrc +
3343                 adapter->stats.crcerrs + adapter->stats.algnerrc +
3344                 adapter->stats.ruc + adapter->stats.roc +
3345                 adapter->stats.cexterr;
3346         adapter->net_stats.rx_length_errors = adapter->stats.ruc +
3347                                               adapter->stats.roc;
3348         adapter->net_stats.rx_crc_errors = adapter->stats.crcerrs;
3349         adapter->net_stats.rx_frame_errors = adapter->stats.algnerrc;
3350         adapter->net_stats.rx_missed_errors = adapter->stats.mpc;
3351
3352         /* Tx Errors */
3353         adapter->net_stats.tx_errors = adapter->stats.ecol +
3354                                        adapter->stats.latecol;
3355         adapter->net_stats.tx_aborted_errors = adapter->stats.ecol;
3356         adapter->net_stats.tx_window_errors = adapter->stats.latecol;
3357         adapter->net_stats.tx_carrier_errors = adapter->stats.tncrs;
3358
3359         /* Tx Dropped needs to be maintained elsewhere */
3360
3361         /* Management Stats */
3362         adapter->stats.mgptc += er32(MGTPTC);
3363         adapter->stats.mgprc += er32(MGTPRC);
3364         adapter->stats.mgpdc += er32(MGTPDC);
3365 }
3366
3367 /**
3368  * e1000_phy_read_status - Update the PHY register status snapshot
3369  * @adapter: board private structure
3370  **/
3371 static void e1000_phy_read_status(struct e1000_adapter *adapter)
3372 {
3373         struct e1000_hw *hw = &adapter->hw;
3374         struct e1000_phy_regs *phy = &adapter->phy_regs;
3375         int ret_val;
3376
3377         if ((er32(STATUS) & E1000_STATUS_LU) &&
3378             (adapter->hw.phy.media_type == e1000_media_type_copper)) {
3379                 ret_val  = e1e_rphy(hw, PHY_CONTROL, &phy->bmcr);
3380                 ret_val |= e1e_rphy(hw, PHY_STATUS, &phy->bmsr);
3381                 ret_val |= e1e_rphy(hw, PHY_AUTONEG_ADV, &phy->advertise);
3382                 ret_val |= e1e_rphy(hw, PHY_LP_ABILITY, &phy->lpa);
3383                 ret_val |= e1e_rphy(hw, PHY_AUTONEG_EXP, &phy->expansion);
3384                 ret_val |= e1e_rphy(hw, PHY_1000T_CTRL, &phy->ctrl1000);
3385                 ret_val |= e1e_rphy(hw, PHY_1000T_STATUS, &phy->stat1000);
3386                 ret_val |= e1e_rphy(hw, PHY_EXT_STATUS, &phy->estatus);
3387                 if (ret_val)
3388                         e_warn("Error reading PHY register\n");
3389         } else {
3390                 /*
3391                  * Do not read PHY registers if link is not up
3392                  * Set values to typical power-on defaults
3393                  */
3394                 phy->bmcr = (BMCR_SPEED1000 | BMCR_ANENABLE | BMCR_FULLDPLX);
3395                 phy->bmsr = (BMSR_100FULL | BMSR_100HALF | BMSR_10FULL |
3396                              BMSR_10HALF | BMSR_ESTATEN | BMSR_ANEGCAPABLE |
3397                              BMSR_ERCAP);
3398                 phy->advertise = (ADVERTISE_PAUSE_ASYM | ADVERTISE_PAUSE_CAP |
3399                                   ADVERTISE_ALL | ADVERTISE_CSMA);
3400                 phy->lpa = 0;
3401                 phy->expansion = EXPANSION_ENABLENPAGE;
3402                 phy->ctrl1000 = ADVERTISE_1000FULL;
3403                 phy->stat1000 = 0;
3404                 phy->estatus = (ESTATUS_1000_TFULL | ESTATUS_1000_THALF);
3405         }
3406 }
3407
3408 static void e1000_print_link_info(struct e1000_adapter *adapter)
3409 {
3410         struct e1000_hw *hw = &adapter->hw;
3411         u32 ctrl = er32(CTRL);
3412
3413         e_info("Link is Up %d Mbps %s, Flow Control: %s\n",
3414                adapter->link_speed,
3415                (adapter->link_duplex == FULL_DUPLEX) ?
3416                                 "Full Duplex" : "Half Duplex",
3417                ((ctrl & E1000_CTRL_TFCE) && (ctrl & E1000_CTRL_RFCE)) ?
3418                                 "RX/TX" :
3419                ((ctrl & E1000_CTRL_RFCE) ? "RX" :
3420                ((ctrl & E1000_CTRL_TFCE) ? "TX" : "None" )));
3421 }
3422
3423 static bool e1000_has_link(struct e1000_adapter *adapter)
3424 {
3425         struct e1000_hw *hw = &adapter->hw;
3426         bool link_active = 0;
3427         s32 ret_val = 0;
3428
3429         /*
3430          * get_link_status is set on LSC (link status) interrupt or
3431          * Rx sequence error interrupt.  get_link_status will stay
3432          * false until the check_for_link establishes link
3433          * for copper adapters ONLY
3434          */
3435         switch (hw->phy.media_type) {
3436         case e1000_media_type_copper:
3437                 if (hw->mac.get_link_status) {
3438                         ret_val = hw->mac.ops.check_for_link(hw);
3439                         link_active = !hw->mac.get_link_status;
3440                 } else {
3441                         link_active = 1;
3442                 }
3443                 break;
3444         case e1000_media_type_fiber:
3445                 ret_val = hw->mac.ops.check_for_link(hw);
3446                 link_active = !!(er32(STATUS) & E1000_STATUS_LU);
3447                 break;
3448         case e1000_media_type_internal_serdes:
3449                 ret_val = hw->mac.ops.check_for_link(hw);
3450                 link_active = adapter->hw.mac.serdes_has_link;
3451                 break;
3452         default:
3453         case e1000_media_type_unknown:
3454                 break;
3455         }
3456
3457         if ((ret_val == E1000_ERR_PHY) && (hw->phy.type == e1000_phy_igp_3) &&
3458             (er32(CTRL) & E1000_PHY_CTRL_GBE_DISABLE)) {
3459                 /* See e1000_kmrn_lock_loss_workaround_ich8lan() */
3460                 e_info("Gigabit has been disabled, downgrading speed\n");
3461         }
3462
3463         return link_active;
3464 }
3465
3466 static void e1000e_enable_receives(struct e1000_adapter *adapter)
3467 {
3468         /* make sure the receive unit is started */
3469         if ((adapter->flags & FLAG_RX_NEEDS_RESTART) &&
3470             (adapter->flags & FLAG_RX_RESTART_NOW)) {
3471                 struct e1000_hw *hw = &adapter->hw;
3472                 u32 rctl = er32(RCTL);
3473                 ew32(RCTL, rctl | E1000_RCTL_EN);
3474                 adapter->flags &= ~FLAG_RX_RESTART_NOW;
3475         }
3476 }
3477
3478 /**
3479  * e1000_watchdog - Timer Call-back
3480  * @data: pointer to adapter cast into an unsigned long
3481  **/
3482 static void e1000_watchdog(unsigned long data)
3483 {
3484         struct e1000_adapter *adapter = (struct e1000_adapter *) data;
3485
3486         /* Do the rest outside of interrupt context */
3487         schedule_work(&adapter->watchdog_task);
3488
3489         /* TODO: make this use queue_delayed_work() */
3490 }
3491
3492 static void e1000_watchdog_task(struct work_struct *work)
3493 {
3494         struct e1000_adapter *adapter = container_of(work,
3495                                         struct e1000_adapter, watchdog_task);
3496         struct net_device *netdev = adapter->netdev;
3497         struct e1000_mac_info *mac = &adapter->hw.mac;
3498         struct e1000_ring *tx_ring = adapter->tx_ring;
3499         struct e1000_hw *hw = &adapter->hw;
3500         u32 link, tctl;
3501         int tx_pending = 0;
3502
3503         link = e1000_has_link(adapter);
3504         if ((netif_carrier_ok(netdev)) && link) {
3505                 e1000e_enable_receives(adapter);
3506                 goto link_up;
3507         }
3508
3509         if ((e1000e_enable_tx_pkt_filtering(hw)) &&
3510             (adapter->mng_vlan_id != adapter->hw.mng_cookie.vlan_id))
3511                 e1000_update_mng_vlan(adapter);
3512
3513         if (link) {
3514                 if (!netif_carrier_ok(netdev)) {
3515                         bool txb2b = 1;
3516                         /* update snapshot of PHY registers on LSC */
3517                         e1000_phy_read_status(adapter);
3518                         mac->ops.get_link_up_info(&adapter->hw,
3519                                                    &adapter->link_speed,
3520                                                    &adapter->link_duplex);
3521                         e1000_print_link_info(adapter);
3522                         /*
3523                          * On supported PHYs, check for duplex mismatch only
3524                          * if link has autonegotiated at 10/100 half
3525                          */
3526                         if ((hw->phy.type == e1000_phy_igp_3 ||
3527                              hw->phy.type == e1000_phy_bm) &&
3528                             (hw->mac.autoneg == true) &&
3529                             (adapter->link_speed == SPEED_10 ||
3530                              adapter->link_speed == SPEED_100) &&
3531                             (adapter->link_duplex == HALF_DUPLEX)) {
3532                                 u16 autoneg_exp;
3533
3534                                 e1e_rphy(hw, PHY_AUTONEG_EXP, &autoneg_exp);
3535
3536                                 if (!(autoneg_exp & NWAY_ER_LP_NWAY_CAPS))
3537                                         e_info("Autonegotiated half duplex but"
3538                                                " link partner cannot autoneg. "
3539                                                " Try forcing full duplex if "
3540                                                "link gets many collisions.\n");
3541                         }
3542
3543                         /*
3544                          * tweak tx_queue_len according to speed/duplex
3545                          * and adjust the timeout factor
3546                          */
3547                         netdev->tx_queue_len = adapter->tx_queue_len;
3548                         adapter->tx_timeout_factor = 1;
3549                         switch (adapter->link_speed) {
3550                         case SPEED_10:
3551                                 txb2b = 0;
3552                                 netdev->tx_queue_len = 10;
3553                                 adapter->tx_timeout_factor = 16;
3554                                 break;
3555                         case SPEED_100:
3556                                 txb2b = 0;
3557                                 netdev->tx_queue_len = 100;
3558                                 /* maybe add some timeout factor ? */
3559                                 break;
3560                         }
3561
3562                         /*
3563                          * workaround: re-program speed mode bit after
3564                          * link-up event
3565                          */
3566                         if ((adapter->flags & FLAG_TARC_SPEED_MODE_BIT) &&
3567                             !txb2b) {
3568                                 u32 tarc0;
3569                                 tarc0 = er32(TARC(0));
3570                                 tarc0 &= ~SPEED_MODE_BIT;
3571                                 ew32(TARC(0), tarc0);
3572                         }
3573
3574                         /*
3575                          * disable TSO for pcie and 10/100 speeds, to avoid
3576                          * some hardware issues
3577                          */
3578                         if (!(adapter->flags & FLAG_TSO_FORCE)) {
3579                                 switch (adapter->link_speed) {
3580                                 case SPEED_10:
3581                                 case SPEED_100:
3582                                         e_info("10/100 speed: disabling TSO\n");
3583                                         netdev->features &= ~NETIF_F_TSO;
3584                                         netdev->features &= ~NETIF_F_TSO6;
3585                                         break;
3586                                 case SPEED_1000:
3587                                         netdev->features |= NETIF_F_TSO;
3588                                         netdev->features |= NETIF_F_TSO6;
3589                                         break;
3590                                 default:
3591                                         /* oops */
3592                                         break;
3593                                 }
3594                         }
3595
3596                         /*
3597                          * enable transmits in the hardware, need to do this
3598                          * after setting TARC(0)
3599                          */
3600                         tctl = er32(TCTL);
3601                         tctl |= E1000_TCTL_EN;
3602                         ew32(TCTL, tctl);
3603
3604                         netif_carrier_on(netdev);
3605                         netif_tx_wake_all_queues(netdev);
3606
3607                         if (!test_bit(__E1000_DOWN, &adapter->state))
3608                                 mod_timer(&adapter->phy_info_timer,
3609                                           round_jiffies(jiffies + 2 * HZ));
3610                 }
3611         } else {
3612                 if (netif_carrier_ok(netdev)) {
3613                         adapter->link_speed = 0;
3614                         adapter->link_duplex = 0;
3615                         e_info("Link is Down\n");
3616                         netif_carrier_off(netdev);
3617                         netif_tx_stop_all_queues(netdev);
3618                         if (!test_bit(__E1000_DOWN, &adapter->state))
3619                                 mod_timer(&adapter->phy_info_timer,
3620                                           round_jiffies(jiffies + 2 * HZ));
3621
3622                         if (adapter->flags & FLAG_RX_NEEDS_RESTART)
3623                                 schedule_work(&adapter->reset_task);
3624                 }
3625         }
3626
3627 link_up:
3628         e1000e_update_stats(adapter);
3629
3630         mac->tx_packet_delta = adapter->stats.tpt - adapter->tpt_old;
3631         adapter->tpt_old = adapter->stats.tpt;
3632         mac->collision_delta = adapter->stats.colc - adapter->colc_old;
3633         adapter->colc_old = adapter->stats.colc;
3634
3635         adapter->gorc = adapter->stats.gorc - adapter->gorc_old;
3636         adapter->gorc_old = adapter->stats.gorc;
3637         adapter->gotc = adapter->stats.gotc - adapter->gotc_old;
3638         adapter->gotc_old = adapter->stats.gotc;
3639
3640         e1000e_update_adaptive(&adapter->hw);
3641
3642         if (!netif_carrier_ok(netdev)) {
3643                 tx_pending = (e1000_desc_unused(tx_ring) + 1 <
3644                                tx_ring->count);
3645                 if (tx_pending) {
3646                         /*
3647                          * We've lost link, so the controller stops DMA,
3648                          * but we've got queued Tx work that's never going
3649                          * to get done, so reset controller to flush Tx.
3650                          * (Do the reset outside of interrupt context).
3651                          */
3652                         adapter->tx_timeout_count++;
3653                         schedule_work(&adapter->reset_task);
3654                 }
3655         }
3656
3657         /* Cause software interrupt to ensure Rx ring is cleaned */
3658         if (adapter->msix_entries)
3659                 ew32(ICS, adapter->rx_ring->ims_val);
3660         else
3661                 ew32(ICS, E1000_ICS_RXDMT0);
3662
3663         /* Force detection of hung controller every watchdog period */
3664         adapter->detect_tx_hung = 1;
3665
3666         /*
3667          * With 82571 controllers, LAA may be overwritten due to controller
3668          * reset from the other port. Set the appropriate LAA in RAR[0]
3669          */
3670         if (e1000e_get_laa_state_82571(hw))
3671                 e1000e_rar_set(hw, adapter->hw.mac.addr, 0);
3672
3673         /* Reset the timer */
3674         if (!test_bit(__E1000_DOWN, &adapter->state))
3675                 mod_timer(&adapter->watchdog_timer,
3676                           round_jiffies(jiffies + 2 * HZ));
3677 }
3678
3679 #define E1000_TX_FLAGS_CSUM             0x00000001
3680 #define E1000_TX_FLAGS_VLAN             0x00000002
3681 #define E1000_TX_FLAGS_TSO              0x00000004
3682 #define E1000_TX_FLAGS_IPV4             0x00000008
3683 #define E1000_TX_FLAGS_VLAN_MASK        0xffff0000
3684 #define E1000_TX_FLAGS_VLAN_SHIFT       16
3685
3686 static int e1000_tso(struct e1000_adapter *adapter,
3687                      struct sk_buff *skb)
3688 {
3689         struct e1000_ring *tx_ring = adapter->tx_ring;
3690         struct e1000_context_desc *context_desc;
3691         struct e1000_buffer *buffer_info;
3692         unsigned int i;
3693         u32 cmd_length = 0;
3694         u16 ipcse = 0, tucse, mss;
3695         u8 ipcss, ipcso, tucss, tucso, hdr_len;
3696         int err;
3697
3698         if (skb_is_gso(skb)) {
3699                 if (skb_header_cloned(skb)) {
3700                         err = pskb_expand_head(skb, 0, 0, GFP_ATOMIC);
3701                         if (err)
3702                                 return err;
3703                 }
3704
3705                 hdr_len = skb_transport_offset(skb) + tcp_hdrlen(skb);
3706                 mss = skb_shinfo(skb)->gso_size;
3707                 if (skb->protocol == htons(ETH_P_IP)) {
3708                         struct iphdr *iph = ip_hdr(skb);
3709                         iph->tot_len = 0;
3710                         iph->check = 0;
3711                         tcp_hdr(skb)->check = ~csum_tcpudp_magic(iph->saddr,
3712                                                                  iph->daddr, 0,
3713                                                                  IPPROTO_TCP,
3714                                                                  0);
3715                         cmd_length = E1000_TXD_CMD_IP;
3716                         ipcse = skb_transport_offset(skb) - 1;
3717                 } else if (skb_shinfo(skb)->gso_type == SKB_GSO_TCPV6) {
3718                         ipv6_hdr(skb)->payload_len = 0;
3719                         tcp_hdr(skb)->check =
3720                                 ~csum_ipv6_magic(&ipv6_hdr(skb)->saddr,
3721                                                  &ipv6_hdr(skb)->daddr,
3722                                                  0, IPPROTO_TCP, 0);
3723                         ipcse = 0;
3724                 }
3725                 ipcss = skb_network_offset(skb);
3726                 ipcso = (void *)&(ip_hdr(skb)->check) - (void *)skb->data;
3727                 tucss = skb_transport_offset(skb);
3728                 tucso = (void *)&(tcp_hdr(skb)->check) - (void *)skb->data;
3729                 tucse = 0;
3730
3731                 cmd_length |= (E1000_TXD_CMD_DEXT | E1000_TXD_CMD_TSE |
3732                                E1000_TXD_CMD_TCP | (skb->len - (hdr_len)));
3733
3734                 i = tx_ring->next_to_use;
3735                 context_desc = E1000_CONTEXT_DESC(*tx_ring, i);
3736                 buffer_info = &tx_ring->buffer_info[i];
3737
3738                 context_desc->lower_setup.ip_fields.ipcss  = ipcss;
3739                 context_desc->lower_setup.ip_fields.ipcso  = ipcso;
3740                 context_desc->lower_setup.ip_fields.ipcse  = cpu_to_le16(ipcse);
3741                 context_desc->upper_setup.tcp_fields.tucss = tucss;
3742                 context_desc->upper_setup.tcp_fields.tucso = tucso;
3743                 context_desc->upper_setup.tcp_fields.tucse = cpu_to_le16(tucse);
3744                 context_desc->tcp_seg_setup.fields.mss     = cpu_to_le16(mss);
3745                 context_desc->tcp_seg_setup.fields.hdr_len = hdr_len;
3746                 context_desc->cmd_and_length = cpu_to_le32(cmd_length);
3747
3748                 buffer_info->time_stamp = jiffies;
3749                 buffer_info->next_to_watch = i;
3750
3751                 i++;
3752                 if (i == tx_ring->count)
3753                         i = 0;
3754                 tx_ring->next_to_use = i;
3755
3756                 return 1;
3757         }
3758
3759         return 0;
3760 }
3761
3762 static bool e1000_tx_csum(struct e1000_adapter *adapter, struct sk_buff *skb)
3763 {
3764         struct e1000_ring *tx_ring = adapter->tx_ring;
3765         struct e1000_context_desc *context_desc;
3766         struct e1000_buffer *buffer_info;
3767         unsigned int i;
3768         u8 css;
3769         u32 cmd_len = E1000_TXD_CMD_DEXT;
3770
3771         if (skb->ip_summed != CHECKSUM_PARTIAL)
3772                 return 0;
3773
3774         switch (skb->protocol) {
3775         case __constant_htons(ETH_P_IP):
3776                 if (ip_hdr(skb)->protocol == IPPROTO_TCP)
3777                         cmd_len |= E1000_TXD_CMD_TCP;
3778                 break;
3779         case __constant_htons(ETH_P_IPV6):
3780                 /* XXX not handling all IPV6 headers */
3781                 if (ipv6_hdr(skb)->nexthdr == IPPROTO_TCP)
3782                         cmd_len |= E1000_TXD_CMD_TCP;
3783                 break;
3784         default:
3785                 if (unlikely(net_ratelimit()))
3786                         e_warn("checksum_partial proto=%x!\n", skb->protocol);
3787                 break;
3788         }
3789
3790         css = skb_transport_offset(skb);
3791
3792         i = tx_ring->next_to_use;
3793         buffer_info = &tx_ring->buffer_info[i];
3794         context_desc = E1000_CONTEXT_DESC(*tx_ring, i);
3795
3796         context_desc->lower_setup.ip_config = 0;
3797         context_desc->upper_setup.tcp_fields.tucss = css;
3798         context_desc->upper_setup.tcp_fields.tucso =
3799                                 css + skb->csum_offset;
3800         context_desc->upper_setup.tcp_fields.tucse = 0;
3801         context_desc->tcp_seg_setup.data = 0;
3802         context_desc->cmd_and_length = cpu_to_le32(cmd_len);
3803
3804         buffer_info->time_stamp = jiffies;
3805         buffer_info->next_to_watch = i;
3806
3807         i++;
3808         if (i == tx_ring->count)
3809                 i = 0;
3810         tx_ring->next_to_use = i;
3811
3812         return 1;
3813 }
3814
3815 #define E1000_MAX_PER_TXD       8192
3816 #define E1000_MAX_TXD_PWR       12
3817
3818 static int e1000_tx_map(struct e1000_adapter *adapter,
3819                         struct sk_buff *skb, unsigned int first,
3820                         unsigned int max_per_txd, unsigned int nr_frags,
3821                         unsigned int mss)
3822 {
3823         struct e1000_ring *tx_ring = adapter->tx_ring;
3824         struct e1000_buffer *buffer_info;
3825         unsigned int len = skb->len - skb->data_len;
3826         unsigned int offset = 0, size, count = 0, i;
3827         unsigned int f;
3828
3829         i = tx_ring->next_to_use;
3830
3831         while (len) {
3832                 buffer_info = &tx_ring->buffer_info[i];
3833                 size = min(len, max_per_txd);
3834
3835                 /* Workaround for premature desc write-backs
3836                  * in TSO mode.  Append 4-byte sentinel desc */
3837                 if (mss && !nr_frags && size == len && size > 8)
3838                         size -= 4;
3839
3840                 buffer_info->length = size;
3841                 /* set time_stamp *before* dma to help avoid a possible race */
3842                 buffer_info->time_stamp = jiffies;
3843                 buffer_info->dma =
3844                         pci_map_single(adapter->pdev,
3845                                 skb->data + offset,
3846                                 size,
3847                                 PCI_DMA_TODEVICE);
3848                 if (pci_dma_mapping_error(adapter->pdev, buffer_info->dma)) {
3849                         dev_err(&adapter->pdev->dev, "TX DMA map failed\n");
3850                         adapter->tx_dma_failed++;
3851                         return -1;
3852                 }
3853                 buffer_info->next_to_watch = i;
3854
3855                 len -= size;
3856                 offset += size;
3857                 count++;
3858                 i++;
3859                 if (i == tx_ring->count)
3860                         i = 0;
3861         }
3862
3863         for (f = 0; f < nr_frags; f++) {
3864                 struct skb_frag_struct *frag;
3865
3866                 frag = &skb_shinfo(skb)->frags[f];
3867                 len = frag->size;
3868                 offset = frag->page_offset;
3869
3870                 while (len) {
3871                         buffer_info = &tx_ring->buffer_info[i];
3872                         size = min(len, max_per_txd);
3873                         /* Workaround for premature desc write-backs
3874                          * in TSO mode.  Append 4-byte sentinel desc */
3875                         if (mss && f == (nr_frags-1) && size == len && size > 8)
3876                                 size -= 4;
3877
3878                         buffer_info->length = size;
3879                         buffer_info->time_stamp = jiffies;
3880                         buffer_info->dma =
3881                                 pci_map_page(adapter->pdev,
3882                                         frag->page,
3883                                         offset,
3884                                         size,
3885                                         PCI_DMA_TODEVICE);
3886                         if (pci_dma_mapping_error(adapter->pdev,
3887                                                   buffer_info->dma)) {
3888                                 dev_err(&adapter->pdev->dev,
3889                                         "TX DMA page map failed\n");
3890                                 adapter->tx_dma_failed++;
3891                                 return -1;
3892                         }
3893
3894                         buffer_info->next_to_watch = i;
3895
3896                         len -= size;
3897                         offset += size;
3898                         count++;
3899
3900                         i++;
3901                         if (i == tx_ring->count)
3902                                 i = 0;
3903                 }
3904         }
3905
3906         if (i == 0)
3907                 i = tx_ring->count - 1;
3908         else
3909                 i--;
3910
3911         tx_ring->buffer_info[i].skb = skb;
3912         tx_ring->buffer_info[first].next_to_watch = i;
3913
3914         return count;
3915 }
3916
3917 static void e1000_tx_queue(struct e1000_adapter *adapter,
3918                            int tx_flags, int count)
3919 {
3920         struct e1000_ring *tx_ring = adapter->tx_ring;
3921         struct e1000_tx_desc *tx_desc = NULL;
3922         struct e1000_buffer *buffer_info;
3923         u32 txd_upper = 0, txd_lower = E1000_TXD_CMD_IFCS;
3924         unsigned int i;
3925
3926         if (tx_flags & E1000_TX_FLAGS_TSO) {
3927                 txd_lower |= E1000_TXD_CMD_DEXT | E1000_TXD_DTYP_D |
3928                              E1000_TXD_CMD_TSE;
3929                 txd_upper |= E1000_TXD_POPTS_TXSM << 8;
3930
3931                 if (tx_flags & E1000_TX_FLAGS_IPV4)
3932                         txd_upper |= E1000_TXD_POPTS_IXSM << 8;
3933         }
3934
3935         if (tx_flags & E1000_TX_FLAGS_CSUM) {
3936                 txd_lower |= E1000_TXD_CMD_DEXT | E1000_TXD_DTYP_D;
3937                 txd_upper |= E1000_TXD_POPTS_TXSM << 8;
3938         }
3939
3940         if (tx_flags & E1000_TX_FLAGS_VLAN) {
3941                 txd_lower |= E1000_TXD_CMD_VLE;
3942                 txd_upper |= (tx_flags & E1000_TX_FLAGS_VLAN_MASK);
3943         }
3944
3945         i = tx_ring->next_to_use;
3946
3947         while (count--) {
3948                 buffer_info = &tx_ring->buffer_info[i];
3949                 tx_desc = E1000_TX_DESC(*tx_ring, i);
3950                 tx_desc->buffer_addr = cpu_to_le64(buffer_info->dma);
3951                 tx_desc->lower.data =
3952                         cpu_to_le32(txd_lower | buffer_info->length);
3953                 tx_desc->upper.data = cpu_to_le32(txd_upper);
3954
3955                 i++;
3956                 if (i == tx_ring->count)
3957                         i = 0;
3958         }
3959
3960         tx_desc->lower.data |= cpu_to_le32(adapter->txd_cmd);
3961
3962         /*
3963          * Force memory writes to complete before letting h/w
3964          * know there are new descriptors to fetch.  (Only
3965          * applicable for weak-ordered memory model archs,
3966          * such as IA-64).
3967          */
3968         wmb();
3969
3970         tx_ring->next_to_use = i;
3971         writel(i, adapter->hw.hw_addr + tx_ring->tail);
3972         /*
3973          * we need this if more than one processor can write to our tail
3974          * at a time, it synchronizes IO on IA64/Altix systems
3975          */
3976         mmiowb();
3977 }
3978
3979 #define MINIMUM_DHCP_PACKET_SIZE 282
3980 static int e1000_transfer_dhcp_info(struct e1000_adapter *adapter,
3981                                     struct sk_buff *skb)
3982 {
3983         struct e1000_hw *hw =  &adapter->hw;
3984         u16 length, offset;
3985
3986         if (vlan_tx_tag_present(skb)) {
3987                 if (!((vlan_tx_tag_get(skb) == adapter->hw.mng_cookie.vlan_id)
3988                     && (adapter->hw.mng_cookie.status &
3989                         E1000_MNG_DHCP_COOKIE_STATUS_VLAN)))
3990                         return 0;
3991         }
3992
3993         if (skb->len <= MINIMUM_DHCP_PACKET_SIZE)
3994                 return 0;
3995
3996         if (((struct ethhdr *) skb->data)->h_proto != htons(ETH_P_IP))
3997                 return 0;
3998
3999         {
4000                 const struct iphdr *ip = (struct iphdr *)((u8 *)skb->data+14);
4001                 struct udphdr *udp;
4002
4003                 if (ip->protocol != IPPROTO_UDP)
4004                         return 0;
4005
4006                 udp = (struct udphdr *)((u8 *)ip + (ip->ihl << 2));
4007                 if (ntohs(udp->dest) != 67)
4008                         return 0;
4009
4010                 offset = (u8 *)udp + 8 - skb->data;
4011                 length = skb->len - offset;
4012                 return e1000e_mng_write_dhcp_info(hw, (u8 *)udp + 8, length);
4013         }
4014
4015         return 0;
4016 }
4017
4018 static int __e1000_maybe_stop_tx(struct net_device *netdev, int size)
4019 {
4020         struct e1000_adapter *adapter = netdev_priv(netdev);
4021
4022         netif_stop_queue(netdev);
4023         /*
4024          * Herbert's original patch had:
4025          *  smp_mb__after_netif_stop_queue();
4026          * but since that doesn't exist yet, just open code it.
4027          */
4028         smp_mb();
4029
4030         /*
4031          * We need to check again in a case another CPU has just
4032          * made room available.
4033          */
4034         if (e1000_desc_unused(adapter->tx_ring) < size)
4035                 return -EBUSY;
4036
4037         /* A reprieve! */
4038         netif_start_queue(netdev);
4039         ++adapter->restart_queue;
4040         return 0;
4041 }
4042
4043 static int e1000_maybe_stop_tx(struct net_device *netdev, int size)
4044 {
4045         struct e1000_adapter *adapter = netdev_priv(netdev);
4046
4047         if (e1000_desc_unused(adapter->tx_ring) >= size)
4048                 return 0;
4049         return __e1000_maybe_stop_tx(netdev, size);
4050 }
4051
4052 #define TXD_USE_COUNT(S, X) (((S) >> (X)) + 1 )
4053 static int e1000_xmit_frame(struct sk_buff *skb, struct net_device *netdev)
4054 {
4055         struct e1000_adapter *adapter = netdev_priv(netdev);
4056         struct e1000_ring *tx_ring = adapter->tx_ring;
4057         unsigned int first;
4058         unsigned int max_per_txd = E1000_MAX_PER_TXD;
4059         unsigned int max_txd_pwr = E1000_MAX_TXD_PWR;
4060         unsigned int tx_flags = 0;
4061         unsigned int len = skb->len - skb->data_len;
4062         unsigned long irq_flags;
4063         unsigned int nr_frags;
4064         unsigned int mss;
4065         int count = 0;
4066         int tso;
4067         unsigned int f;
4068
4069         if (test_bit(__E1000_DOWN, &adapter->state)) {
4070                 dev_kfree_skb_any(skb);
4071                 return NETDEV_TX_OK;
4072         }
4073
4074         if (skb->len <= 0) {
4075                 dev_kfree_skb_any(skb);
4076                 return NETDEV_TX_OK;
4077         }
4078
4079         mss = skb_shinfo(skb)->gso_size;
4080         /*
4081          * The controller does a simple calculation to
4082          * make sure there is enough room in the FIFO before
4083          * initiating the DMA for each buffer.  The calc is:
4084          * 4 = ceil(buffer len/mss).  To make sure we don't
4085          * overrun the FIFO, adjust the max buffer len if mss
4086          * drops.
4087          */
4088         if (mss) {
4089                 u8 hdr_len;
4090                 max_per_txd = min(mss << 2, max_per_txd);
4091                 max_txd_pwr = fls(max_per_txd) - 1;
4092
4093                 /*
4094                  * TSO Workaround for 82571/2/3 Controllers -- if skb->data
4095                  * points to just header, pull a few bytes of payload from
4096                  * frags into skb->data
4097                  */
4098                 hdr_len = skb_transport_offset(skb) + tcp_hdrlen(skb);
4099                 /*
4100                  * we do this workaround for ES2LAN, but it is un-necessary,
4101                  * avoiding it could save a lot of cycles
4102                  */
4103                 if (skb->data_len && (hdr_len == len)) {
4104                         unsigned int pull_size;
4105
4106                         pull_size = min((unsigned int)4, skb->data_len);
4107                         if (!__pskb_pull_tail(skb, pull_size)) {
4108                                 e_err("__pskb_pull_tail failed.\n");
4109                                 dev_kfree_skb_any(skb);
4110                                 return NETDEV_TX_OK;
4111                         }
4112                         len = skb->len - skb->data_len;
4113                 }
4114         }
4115
4116         /* reserve a descriptor for the offload context */
4117         if ((mss) || (skb->ip_summed == CHECKSUM_PARTIAL))
4118                 count++;
4119         count++;
4120
4121         count += TXD_USE_COUNT(len, max_txd_pwr);
4122
4123         nr_frags = skb_shinfo(skb)->nr_frags;
4124         for (f = 0; f < nr_frags; f++)
4125                 count += TXD_USE_COUNT(skb_shinfo(skb)->frags[f].size,
4126                                        max_txd_pwr);
4127
4128         if (adapter->hw.mac.tx_pkt_filtering)
4129                 e1000_transfer_dhcp_info(adapter, skb);
4130
4131         if (!spin_trylock_irqsave(&adapter->tx_queue_lock, irq_flags))
4132                 /* Collision - tell upper layer to requeue */
4133                 return NETDEV_TX_LOCKED;
4134
4135         /*
4136          * need: count + 2 desc gap to keep tail from touching
4137          * head, otherwise try next time
4138          */
4139         if (e1000_maybe_stop_tx(netdev, count + 2)) {
4140                 spin_unlock_irqrestore(&adapter->tx_queue_lock, irq_flags);
4141                 return NETDEV_TX_BUSY;
4142         }
4143
4144         if (adapter->vlgrp && vlan_tx_tag_present(skb)) {
4145                 tx_flags |= E1000_TX_FLAGS_VLAN;
4146                 tx_flags |= (vlan_tx_tag_get(skb) << E1000_TX_FLAGS_VLAN_SHIFT);
4147         }
4148
4149         first = tx_ring->next_to_use;
4150
4151         tso = e1000_tso(adapter, skb);
4152         if (tso < 0) {
4153                 dev_kfree_skb_any(skb);
4154                 spin_unlock_irqrestore(&adapter->tx_queue_lock, irq_flags);
4155                 return NETDEV_TX_OK;
4156         }
4157
4158         if (tso)
4159                 tx_flags |= E1000_TX_FLAGS_TSO;
4160         else if (e1000_tx_csum(adapter, skb))
4161                 tx_flags |= E1000_TX_FLAGS_CSUM;
4162
4163         /*
4164          * Old method was to assume IPv4 packet by default if TSO was enabled.
4165          * 82571 hardware supports TSO capabilities for IPv6 as well...
4166          * no longer assume, we must.
4167          */
4168         if (skb->protocol == htons(ETH_P_IP))
4169                 tx_flags |= E1000_TX_FLAGS_IPV4;
4170
4171         count = e1000_tx_map(adapter, skb, first, max_per_txd, nr_frags, mss);
4172         if (count < 0) {
4173                 /* handle pci_map_single() error in e1000_tx_map */
4174                 dev_kfree_skb_any(skb);
4175                 spin_unlock_irqrestore(&adapter->tx_queue_lock, irq_flags);
4176                 return NETDEV_TX_OK;
4177         }
4178
4179         e1000_tx_queue(adapter, tx_flags, count);
4180
4181         netdev->trans_start = jiffies;
4182
4183         /* Make sure there is space in the ring for the next send. */
4184         e1000_maybe_stop_tx(netdev, MAX_SKB_FRAGS + 2);
4185
4186         spin_unlock_irqrestore(&adapter->tx_queue_lock, irq_flags);
4187         return NETDEV_TX_OK;
4188 }
4189
4190 /**
4191  * e1000_tx_timeout - Respond to a Tx Hang
4192  * @netdev: network interface device structure
4193  **/
4194 static void e1000_tx_timeout(struct net_device *netdev)
4195 {
4196         struct e1000_adapter *adapter = netdev_priv(netdev);
4197
4198         /* Do the reset outside of interrupt context */
4199         adapter->tx_timeout_count++;
4200         schedule_work(&adapter->reset_task);
4201 }
4202
4203 static void e1000_reset_task(struct work_struct *work)
4204 {
4205         struct e1000_adapter *adapter;
4206         adapter = container_of(work, struct e1000_adapter, reset_task);
4207
4208         e1000e_reinit_locked(adapter);
4209 }
4210
4211 /**
4212  * e1000_get_stats - Get System Network Statistics
4213  * @netdev: network interface device structure
4214  *
4215  * Returns the address of the device statistics structure.
4216  * The statistics are actually updated from the timer callback.
4217  **/
4218 static struct net_device_stats *e1000_get_stats(struct net_device *netdev)
4219 {
4220         struct e1000_adapter *adapter = netdev_priv(netdev);
4221
4222         /* only return the current stats */
4223         return &adapter->net_stats;
4224 }
4225
4226 /**
4227  * e1000_change_mtu - Change the Maximum Transfer Unit
4228  * @netdev: network interface device structure
4229  * @new_mtu: new value for maximum frame size
4230  *
4231  * Returns 0 on success, negative on failure
4232  **/
4233 static int e1000_change_mtu(struct net_device *netdev, int new_mtu)
4234 {
4235         struct e1000_adapter *adapter = netdev_priv(netdev);
4236         int max_frame = new_mtu + ETH_HLEN + ETH_FCS_LEN;
4237
4238         if ((new_mtu < ETH_ZLEN + ETH_FCS_LEN + VLAN_HLEN) ||
4239             (max_frame > MAX_JUMBO_FRAME_SIZE)) {
4240                 e_err("Invalid MTU setting\n");
4241                 return -EINVAL;
4242         }
4243
4244         /* Jumbo frame size limits */
4245         if (max_frame > ETH_FRAME_LEN + ETH_FCS_LEN) {
4246                 if (!(adapter->flags & FLAG_HAS_JUMBO_FRAMES)) {
4247                         e_err("Jumbo Frames not supported.\n");
4248                         return -EINVAL;
4249                 }
4250                 if (adapter->hw.phy.type == e1000_phy_ife) {
4251                         e_err("Jumbo Frames not supported.\n");
4252                         return -EINVAL;
4253                 }
4254         }
4255
4256 #define MAX_STD_JUMBO_FRAME_SIZE 9234
4257         if (max_frame > MAX_STD_JUMBO_FRAME_SIZE) {
4258                 e_err("MTU > 9216 not supported.\n");
4259                 return -EINVAL;
4260         }
4261
4262         while (test_and_set_bit(__E1000_RESETTING, &adapter->state))
4263                 msleep(1);
4264         /* e1000e_down has a dependency on max_frame_size */
4265         adapter->max_frame_size = max_frame;
4266         if (netif_running(netdev))
4267                 e1000e_down(adapter);
4268
4269         /*
4270          * NOTE: netdev_alloc_skb reserves 16 bytes, and typically NET_IP_ALIGN
4271          * means we reserve 2 more, this pushes us to allocate from the next
4272          * larger slab size.
4273          * i.e. RXBUFFER_2048 --> size-4096 slab
4274          * However with the new *_jumbo_rx* routines, jumbo receives will use
4275          * fragmented skbs
4276          */
4277
4278         if (max_frame <= 256)
4279                 adapter->rx_buffer_len = 256;
4280         else if (max_frame <= 512)
4281                 adapter->rx_buffer_len = 512;
4282         else if (max_frame <= 1024)
4283                 adapter->rx_buffer_len = 1024;
4284         else if (max_frame <= 2048)
4285                 adapter->rx_buffer_len = 2048;
4286         else
4287                 adapter->rx_buffer_len = 4096;
4288
4289         /* adjust allocation if LPE protects us, and we aren't using SBP */
4290         if ((max_frame == ETH_FRAME_LEN + ETH_FCS_LEN) ||
4291              (max_frame == ETH_FRAME_LEN + VLAN_HLEN + ETH_FCS_LEN))
4292                 adapter->rx_buffer_len = ETH_FRAME_LEN + VLAN_HLEN
4293                                          + ETH_FCS_LEN;
4294
4295         e_info("changing MTU from %d to %d\n", netdev->mtu, new_mtu);
4296         netdev->mtu = new_mtu;
4297
4298         if (netif_running(netdev))
4299                 e1000e_up(adapter);
4300         else
4301                 e1000e_reset(adapter);
4302
4303         clear_bit(__E1000_RESETTING, &adapter->state);
4304
4305         return 0;
4306 }
4307
4308 static int e1000_mii_ioctl(struct net_device *netdev, struct ifreq *ifr,
4309                            int cmd)
4310 {
4311         struct e1000_adapter *adapter = netdev_priv(netdev);
4312         struct mii_ioctl_data *data = if_mii(ifr);
4313
4314         if (adapter->hw.phy.media_type != e1000_media_type_copper)
4315                 return -EOPNOTSUPP;
4316
4317         switch (cmd) {
4318         case SIOCGMIIPHY:
4319                 data->phy_id = adapter->hw.phy.addr;
4320                 break;
4321         case SIOCGMIIREG:
4322                 if (!capable(CAP_NET_ADMIN))
4323                         return -EPERM;
4324                 switch (data->reg_num & 0x1F) {
4325                 case MII_BMCR:
4326                         data->val_out = adapter->phy_regs.bmcr;
4327                         break;
4328                 case MII_BMSR:
4329                         data->val_out = adapter->phy_regs.bmsr;
4330                         break;
4331                 case MII_PHYSID1:
4332                         data->val_out = (adapter->hw.phy.id >> 16);
4333                         break;
4334                 case MII_PHYSID2:
4335                         data->val_out = (adapter->hw.phy.id & 0xFFFF);
4336                         break;
4337                 case MII_ADVERTISE:
4338                         data->val_out = adapter->phy_regs.advertise;
4339                         break;
4340                 case MII_LPA:
4341                         data->val_out = adapter->phy_regs.lpa;
4342                         break;
4343                 case MII_EXPANSION:
4344                         data->val_out = adapter->phy_regs.expansion;
4345                         break;
4346                 case MII_CTRL1000:
4347                         data->val_out = adapter->phy_regs.ctrl1000;
4348                         break;
4349                 case MII_STAT1000:
4350                         data->val_out = adapter->phy_regs.stat1000;
4351                         break;
4352                 case MII_ESTATUS:
4353                         data->val_out = adapter->phy_regs.estatus;
4354                         break;
4355                 default:
4356                         return -EIO;
4357                 }
4358                 break;
4359         case SIOCSMIIREG:
4360         default:
4361                 return -EOPNOTSUPP;
4362         }
4363         return 0;
4364 }
4365
4366 static int e1000_ioctl(struct net_device *netdev, struct ifreq *ifr, int cmd)
4367 {
4368         switch (cmd) {
4369         case SIOCGMIIPHY:
4370         case SIOCGMIIREG:
4371         case SIOCSMIIREG:
4372                 return e1000_mii_ioctl(netdev, ifr, cmd);
4373         default:
4374                 return -EOPNOTSUPP;
4375         }
4376 }
4377
4378 static int e1000_suspend(struct pci_dev *pdev, pm_message_t state)
4379 {
4380         struct net_device *netdev = pci_get_drvdata(pdev);
4381         struct e1000_adapter *adapter = netdev_priv(netdev);
4382         struct e1000_hw *hw = &adapter->hw;
4383         u32 ctrl, ctrl_ext, rctl, status;
4384         u32 wufc = adapter->wol;
4385         int retval = 0;
4386
4387         netif_device_detach(netdev);
4388
4389         if (netif_running(netdev)) {
4390                 WARN_ON(test_bit(__E1000_RESETTING, &adapter->state));
4391                 e1000e_down(adapter);
4392                 e1000_free_irq(adapter);
4393         }
4394         e1000e_reset_interrupt_capability(adapter);
4395
4396         retval = pci_save_state(pdev);
4397         if (retval)
4398                 return retval;
4399
4400         status = er32(STATUS);
4401         if (status & E1000_STATUS_LU)
4402                 wufc &= ~E1000_WUFC_LNKC;
4403
4404         if (wufc) {
4405                 e1000_setup_rctl(adapter);
4406                 e1000_set_multi(netdev);
4407
4408                 /* turn on all-multi mode if wake on multicast is enabled */
4409                 if (wufc & E1000_WUFC_MC) {
4410                         rctl = er32(RCTL);
4411                         rctl |= E1000_RCTL_MPE;
4412                         ew32(RCTL, rctl);
4413                 }
4414
4415                 ctrl = er32(CTRL);
4416                 /* advertise wake from D3Cold */
4417                 #define E1000_CTRL_ADVD3WUC 0x00100000
4418                 /* phy power management enable */
4419                 #define E1000_CTRL_EN_PHY_PWR_MGMT 0x00200000
4420                 ctrl |= E1000_CTRL_ADVD3WUC |
4421                         E1000_CTRL_EN_PHY_PWR_MGMT;
4422                 ew32(CTRL, ctrl);
4423
4424                 if (adapter->hw.phy.media_type == e1000_media_type_fiber ||
4425                     adapter->hw.phy.media_type ==
4426                     e1000_media_type_internal_serdes) {
4427                         /* keep the laser running in D3 */
4428                         ctrl_ext = er32(CTRL_EXT);
4429                         ctrl_ext |= E1000_CTRL_EXT_SDP7_DATA;
4430                         ew32(CTRL_EXT, ctrl_ext);
4431                 }
4432
4433                 if (adapter->flags & FLAG_IS_ICH)
4434                         e1000e_disable_gig_wol_ich8lan(&adapter->hw);
4435
4436                 /* Allow time for pending master requests to run */
4437                 e1000e_disable_pcie_master(&adapter->hw);
4438
4439                 ew32(WUC, E1000_WUC_PME_EN);
4440                 ew32(WUFC, wufc);
4441                 pci_enable_wake(pdev, PCI_D3hot, 1);
4442                 pci_enable_wake(pdev, PCI_D3cold, 1);
4443         } else {
4444                 ew32(WUC, 0);
4445                 ew32(WUFC, 0);
4446                 pci_enable_wake(pdev, PCI_D3hot, 0);
4447                 pci_enable_wake(pdev, PCI_D3cold, 0);
4448         }
4449
4450         /* make sure adapter isn't asleep if manageability is enabled */
4451         if (adapter->flags & FLAG_MNG_PT_ENABLED) {
4452                 pci_enable_wake(pdev, PCI_D3hot, 1);
4453                 pci_enable_wake(pdev, PCI_D3cold, 1);
4454         }
4455
4456         if (adapter->hw.phy.type == e1000_phy_igp_3)
4457                 e1000e_igp3_phy_powerdown_workaround_ich8lan(&adapter->hw);
4458
4459         /*
4460          * Release control of h/w to f/w.  If f/w is AMT enabled, this
4461          * would have already happened in close and is redundant.
4462          */
4463         e1000_release_hw_control(adapter);
4464
4465         pci_disable_device(pdev);
4466
4467         pci_set_power_state(pdev, pci_choose_state(pdev, state));
4468
4469         return 0;
4470 }
4471
4472 static void e1000e_disable_l1aspm(struct pci_dev *pdev)
4473 {
4474         int pos;
4475         u16 val;
4476
4477         /*
4478          * 82573 workaround - disable L1 ASPM on mobile chipsets
4479          *
4480          * L1 ASPM on various mobile (ich7) chipsets do not behave properly
4481          * resulting in lost data or garbage information on the pci-e link
4482          * level. This could result in (false) bad EEPROM checksum errors,
4483          * long ping times (up to 2s) or even a system freeze/hang.
4484          *
4485          * Unfortunately this feature saves about 1W power consumption when
4486          * active.
4487          */
4488         pos = pci_find_capability(pdev, PCI_CAP_ID_EXP);
4489         pci_read_config_word(pdev, pos + PCI_EXP_LNKCTL, &val);
4490         if (val & 0x2) {
4491                 dev_warn(&pdev->dev, "Disabling L1 ASPM\n");
4492                 val &= ~0x2;
4493                 pci_write_config_word(pdev, pos + PCI_EXP_LNKCTL, val);
4494         }
4495 }
4496
4497 #ifdef CONFIG_PM
4498 static int e1000_resume(struct pci_dev *pdev)
4499 {
4500         struct net_device *netdev = pci_get_drvdata(pdev);
4501         struct e1000_adapter *adapter = netdev_priv(netdev);
4502         struct e1000_hw *hw = &adapter->hw;
4503         u32 err;
4504
4505         pci_set_power_state(pdev, PCI_D0);
4506         pci_restore_state(pdev);
4507         e1000e_disable_l1aspm(pdev);
4508
4509         err = pci_enable_device_mem(pdev);
4510         if (err) {
4511                 dev_err(&pdev->dev,
4512                         "Cannot enable PCI device from suspend\n");
4513                 return err;
4514         }
4515
4516         pci_set_master(pdev);
4517
4518         pci_enable_wake(pdev, PCI_D3hot, 0);
4519         pci_enable_wake(pdev, PCI_D3cold, 0);
4520
4521         e1000e_set_interrupt_capability(adapter);
4522         if (netif_running(netdev)) {
4523                 err = e1000_request_irq(adapter);
4524                 if (err)
4525                         return err;
4526         }
4527
4528         e1000e_power_up_phy(adapter);
4529         e1000e_reset(adapter);
4530         ew32(WUS, ~0);
4531
4532         e1000_init_manageability(adapter);
4533
4534         if (netif_running(netdev))
4535                 e1000e_up(adapter);
4536
4537         netif_device_attach(netdev);
4538
4539         /*
4540          * If the controller has AMT, do not set DRV_LOAD until the interface
4541          * is up.  For all other cases, let the f/w know that the h/w is now
4542          * under the control of the driver.
4543          */
4544         if (!(adapter->flags & FLAG_HAS_AMT))
4545                 e1000_get_hw_control(adapter);
4546
4547         return 0;
4548 }
4549 #endif
4550
4551 static void e1000_shutdown(struct pci_dev *pdev)
4552 {
4553         e1000_suspend(pdev, PMSG_SUSPEND);
4554 }
4555
4556 #ifdef CONFIG_NET_POLL_CONTROLLER
4557 /*
4558  * Polling 'interrupt' - used by things like netconsole to send skbs
4559  * without having to re-enable interrupts. It's not called while
4560  * the interrupt routine is executing.
4561  */
4562 static void e1000_netpoll(struct net_device *netdev)
4563 {
4564         struct e1000_adapter *adapter = netdev_priv(netdev);
4565
4566         disable_irq(adapter->pdev->irq);
4567         e1000_intr(adapter->pdev->irq, netdev);
4568
4569         enable_irq(adapter->pdev->irq);
4570 }
4571 #endif
4572
4573 /**
4574  * e1000_io_error_detected - called when PCI error is detected
4575  * @pdev: Pointer to PCI device
4576  * @state: The current pci connection state
4577  *
4578  * This function is called after a PCI bus error affecting
4579  * this device has been detected.
4580  */
4581 static pci_ers_result_t e1000_io_error_detected(struct pci_dev *pdev,
4582                                                 pci_channel_state_t state)
4583 {
4584         struct net_device *netdev = pci_get_drvdata(pdev);
4585         struct e1000_adapter *adapter = netdev_priv(netdev);
4586
4587         netif_device_detach(netdev);
4588
4589         if (netif_running(netdev))
4590                 e1000e_down(adapter);
4591         pci_disable_device(pdev);
4592
4593         /* Request a slot slot reset. */
4594         return PCI_ERS_RESULT_NEED_RESET;
4595 }
4596
4597 /**
4598  * e1000_io_slot_reset - called after the pci bus has been reset.
4599  * @pdev: Pointer to PCI device
4600  *
4601  * Restart the card from scratch, as if from a cold-boot. Implementation
4602  * resembles the first-half of the e1000_resume routine.
4603  */
4604 static pci_ers_result_t e1000_io_slot_reset(struct pci_dev *pdev)
4605 {
4606         struct net_device *netdev = pci_get_drvdata(pdev);
4607         struct e1000_adapter *adapter = netdev_priv(netdev);
4608         struct e1000_hw *hw = &adapter->hw;
4609         int err;
4610
4611         e1000e_disable_l1aspm(pdev);
4612         err = pci_enable_device_mem(pdev);
4613         if (err) {
4614                 dev_err(&pdev->dev,
4615                         "Cannot re-enable PCI device after reset.\n");
4616                 return PCI_ERS_RESULT_DISCONNECT;
4617         }
4618         pci_set_master(pdev);
4619         pci_restore_state(pdev);
4620
4621         pci_enable_wake(pdev, PCI_D3hot, 0);
4622         pci_enable_wake(pdev, PCI_D3cold, 0);
4623
4624         e1000e_reset(adapter);
4625         ew32(WUS, ~0);
4626
4627         return PCI_ERS_RESULT_RECOVERED;
4628 }
4629
4630 /**
4631  * e1000_io_resume - called when traffic can start flowing again.
4632  * @pdev: Pointer to PCI device
4633  *
4634  * This callback is called when the error recovery driver tells us that
4635  * its OK to resume normal operation. Implementation resembles the
4636  * second-half of the e1000_resume routine.
4637  */
4638 static void e1000_io_resume(struct pci_dev *pdev)
4639 {
4640         struct net_device *netdev = pci_get_drvdata(pdev);
4641         struct e1000_adapter *adapter = netdev_priv(netdev);
4642
4643         e1000_init_manageability(adapter);
4644
4645         if (netif_running(netdev)) {
4646                 if (e1000e_up(adapter)) {
4647                         dev_err(&pdev->dev,
4648                                 "can't bring device back up after reset\n");
4649                         return;
4650                 }
4651         }
4652
4653         netif_device_attach(netdev);
4654
4655         /*
4656          * If the controller has AMT, do not set DRV_LOAD until the interface
4657          * is up.  For all other cases, let the f/w know that the h/w is now
4658          * under the control of the driver.
4659          */
4660         if (!(adapter->flags & FLAG_HAS_AMT))
4661                 e1000_get_hw_control(adapter);
4662
4663 }
4664
4665 static void e1000_print_device_info(struct e1000_adapter *adapter)
4666 {
4667         struct e1000_hw *hw = &adapter->hw;
4668         struct net_device *netdev = adapter->netdev;
4669         u32 pba_num;
4670
4671         /* print bus type/speed/width info */
4672         e_info("(PCI Express:2.5GB/s:%s) %02x:%02x:%02x:%02x:%02x:%02x\n",
4673                /* bus width */
4674                ((hw->bus.width == e1000_bus_width_pcie_x4) ? "Width x4" :
4675                 "Width x1"),
4676                /* MAC address */
4677                netdev->dev_addr[0], netdev->dev_addr[1],
4678                netdev->dev_addr[2], netdev->dev_addr[3],
4679                netdev->dev_addr[4], netdev->dev_addr[5]);
4680         e_info("Intel(R) PRO/%s Network Connection\n",
4681                (hw->phy.type == e1000_phy_ife) ? "10/100" : "1000");
4682         e1000e_read_pba_num(hw, &pba_num);
4683         e_info("MAC: %d, PHY: %d, PBA No: %06x-%03x\n",
4684                hw->mac.type, hw->phy.type, (pba_num >> 8), (pba_num & 0xff));
4685 }
4686
4687 static void e1000_eeprom_checks(struct e1000_adapter *adapter)
4688 {
4689         struct e1000_hw *hw = &adapter->hw;
4690         int ret_val;
4691         u16 buf = 0;
4692
4693         if (hw->mac.type != e1000_82573)
4694                 return;
4695
4696         ret_val = e1000_read_nvm(hw, NVM_INIT_CONTROL2_REG, 1, &buf);
4697         if (!(le16_to_cpu(buf) & (1 << 0))) {
4698                 /* Deep Smart Power Down (DSPD) */
4699                 dev_warn(&adapter->pdev->dev,
4700                          "Warning: detected DSPD enabled in EEPROM\n");
4701         }
4702
4703         ret_val = e1000_read_nvm(hw, NVM_INIT_3GIO_3, 1, &buf);
4704         if (le16_to_cpu(buf) & (3 << 2)) {
4705                 /* ASPM enable */
4706                 dev_warn(&adapter->pdev->dev,
4707                          "Warning: detected ASPM enabled in EEPROM\n");
4708         }
4709 }
4710
4711 /**
4712  * e1000_probe - Device Initialization Routine
4713  * @pdev: PCI device information struct
4714  * @ent: entry in e1000_pci_tbl
4715  *
4716  * Returns 0 on success, negative on failure
4717  *
4718  * e1000_probe initializes an adapter identified by a pci_dev structure.
4719  * The OS initialization, configuring of the adapter private structure,
4720  * and a hardware reset occur.
4721  **/
4722 static int __devinit e1000_probe(struct pci_dev *pdev,
4723                                  const struct pci_device_id *ent)
4724 {
4725         struct net_device *netdev;
4726         struct e1000_adapter *adapter;
4727         struct e1000_hw *hw;
4728         const struct e1000_info *ei = e1000_info_tbl[ent->driver_data];
4729         resource_size_t mmio_start, mmio_len;
4730         resource_size_t flash_start, flash_len;
4731
4732         static int cards_found;
4733         int i, err, pci_using_dac;
4734         u16 eeprom_data = 0;
4735         u16 eeprom_apme_mask = E1000_EEPROM_APME;
4736
4737         e1000e_disable_l1aspm(pdev);
4738
4739         err = pci_enable_device_mem(pdev);
4740         if (err)
4741                 return err;
4742
4743         pci_using_dac = 0;
4744         err = pci_set_dma_mask(pdev, DMA_64BIT_MASK);
4745         if (!err) {
4746                 err = pci_set_consistent_dma_mask(pdev, DMA_64BIT_MASK);
4747                 if (!err)
4748                         pci_using_dac = 1;
4749         } else {
4750                 err = pci_set_dma_mask(pdev, DMA_32BIT_MASK);
4751                 if (err) {
4752                         err = pci_set_consistent_dma_mask(pdev,
4753                                                           DMA_32BIT_MASK);
4754                         if (err) {
4755                                 dev_err(&pdev->dev, "No usable DMA "
4756                                         "configuration, aborting\n");
4757                                 goto err_dma;
4758                         }
4759                 }
4760         }
4761
4762         err = pci_request_selected_regions(pdev,
4763                                           pci_select_bars(pdev, IORESOURCE_MEM),
4764                                           e1000e_driver_name);
4765         if (err)
4766                 goto err_pci_reg;
4767
4768         pci_set_master(pdev);
4769         pci_save_state(pdev);
4770
4771         err = -ENOMEM;
4772         netdev = alloc_etherdev(sizeof(struct e1000_adapter));
4773         if (!netdev)
4774                 goto err_alloc_etherdev;
4775
4776         SET_NETDEV_DEV(netdev, &pdev->dev);
4777
4778         pci_set_drvdata(pdev, netdev);
4779         adapter = netdev_priv(netdev);
4780         hw = &adapter->hw;
4781         adapter->netdev = netdev;
4782         adapter->pdev = pdev;
4783         adapter->ei = ei;
4784         adapter->pba = ei->pba;
4785         adapter->flags = ei->flags;
4786         adapter->flags2 = ei->flags2;
4787         adapter->hw.adapter = adapter;
4788         adapter->hw.mac.type = ei->mac;
4789         adapter->msg_enable = (1 << NETIF_MSG_DRV | NETIF_MSG_PROBE) - 1;
4790
4791         mmio_start = pci_resource_start(pdev, 0);
4792         mmio_len = pci_resource_len(pdev, 0);
4793
4794         err = -EIO;
4795         adapter->hw.hw_addr = ioremap(mmio_start, mmio_len);
4796         if (!adapter->hw.hw_addr)
4797                 goto err_ioremap;
4798
4799         if ((adapter->flags & FLAG_HAS_FLASH) &&
4800             (pci_resource_flags(pdev, 1) & IORESOURCE_MEM)) {
4801                 flash_start = pci_resource_start(pdev, 1);
4802                 flash_len = pci_resource_len(pdev, 1);
4803                 adapter->hw.flash_address = ioremap(flash_start, flash_len);
4804                 if (!adapter->hw.flash_address)
4805                         goto err_flashmap;
4806         }
4807
4808         /* construct the net_device struct */
4809         netdev->open                    = &e1000_open;
4810         netdev->stop                    = &e1000_close;
4811         netdev->hard_start_xmit         = &e1000_xmit_frame;
4812         netdev->get_stats               = &e1000_get_stats;
4813         netdev->set_multicast_list      = &e1000_set_multi;
4814         netdev->set_mac_address         = &e1000_set_mac;
4815         netdev->change_mtu              = &e1000_change_mtu;
4816         netdev->do_ioctl                = &e1000_ioctl;
4817         e1000e_set_ethtool_ops(netdev);
4818         netdev->tx_timeout              = &e1000_tx_timeout;
4819         netdev->watchdog_timeo          = 5 * HZ;
4820         netif_napi_add(netdev, &adapter->napi, e1000_clean, 64);
4821         netdev->vlan_rx_register        = e1000_vlan_rx_register;
4822         netdev->vlan_rx_add_vid         = e1000_vlan_rx_add_vid;
4823         netdev->vlan_rx_kill_vid        = e1000_vlan_rx_kill_vid;
4824 #ifdef CONFIG_NET_POLL_CONTROLLER
4825         netdev->poll_controller         = e1000_netpoll;
4826 #endif
4827         strncpy(netdev->name, pci_name(pdev), sizeof(netdev->name) - 1);
4828
4829         netdev->mem_start = mmio_start;
4830         netdev->mem_end = mmio_start + mmio_len;
4831
4832         adapter->bd_number = cards_found++;
4833
4834         e1000e_check_options(adapter);
4835
4836         /* setup adapter struct */
4837         err = e1000_sw_init(adapter);
4838         if (err)
4839                 goto err_sw_init;
4840
4841         err = -EIO;
4842
4843         memcpy(&hw->mac.ops, ei->mac_ops, sizeof(hw->mac.ops));
4844         memcpy(&hw->nvm.ops, ei->nvm_ops, sizeof(hw->nvm.ops));
4845         memcpy(&hw->phy.ops, ei->phy_ops, sizeof(hw->phy.ops));
4846
4847         err = ei->get_variants(adapter);
4848         if (err)
4849                 goto err_hw_init;
4850
4851         if ((adapter->flags & FLAG_IS_ICH) &&
4852             (adapter->flags & FLAG_READ_ONLY_NVM))
4853                 e1000e_write_protect_nvm_ich8lan(&adapter->hw);
4854
4855         hw->mac.ops.get_bus_info(&adapter->hw);
4856
4857         adapter->hw.phy.autoneg_wait_to_complete = 0;
4858
4859         /* Copper options */
4860         if (adapter->hw.phy.media_type == e1000_media_type_copper) {
4861                 adapter->hw.phy.mdix = AUTO_ALL_MODES;
4862                 adapter->hw.phy.disable_polarity_correction = 0;
4863                 adapter->hw.phy.ms_type = e1000_ms_hw_default;
4864         }
4865
4866         if (e1000_check_reset_block(&adapter->hw))
4867                 e_info("PHY reset is blocked due to SOL/IDER session.\n");
4868
4869         netdev->features = NETIF_F_SG |
4870                            NETIF_F_HW_CSUM |
4871                            NETIF_F_HW_VLAN_TX |
4872                            NETIF_F_HW_VLAN_RX;
4873
4874         if (adapter->flags & FLAG_HAS_HW_VLAN_FILTER)
4875                 netdev->features |= NETIF_F_HW_VLAN_FILTER;
4876
4877         netdev->features |= NETIF_F_TSO;
4878         netdev->features |= NETIF_F_TSO6;
4879
4880         netdev->vlan_features |= NETIF_F_TSO;
4881         netdev->vlan_features |= NETIF_F_TSO6;
4882         netdev->vlan_features |= NETIF_F_HW_CSUM;
4883         netdev->vlan_features |= NETIF_F_SG;
4884
4885         if (pci_using_dac)
4886                 netdev->features |= NETIF_F_HIGHDMA;
4887
4888         /*
4889          * We should not be using LLTX anymore, but we are still Tx faster with
4890          * it.
4891          */
4892         netdev->features |= NETIF_F_LLTX;
4893
4894         if (e1000e_enable_mng_pass_thru(&adapter->hw))
4895                 adapter->flags |= FLAG_MNG_PT_ENABLED;
4896
4897         /*
4898          * before reading the NVM, reset the controller to
4899          * put the device in a known good starting state
4900          */
4901         adapter->hw.mac.ops.reset_hw(&adapter->hw);
4902
4903         /*
4904          * systems with ASPM and others may see the checksum fail on the first
4905          * attempt. Let's give it a few tries
4906          */
4907         for (i = 0;; i++) {
4908                 if (e1000_validate_nvm_checksum(&adapter->hw) >= 0)
4909                         break;
4910                 if (i == 2) {
4911                         e_err("The NVM Checksum Is Not Valid\n");
4912                         err = -EIO;
4913                         goto err_eeprom;
4914                 }
4915         }
4916
4917         e1000_eeprom_checks(adapter);
4918
4919         /* copy the MAC address out of the NVM */
4920         if (e1000e_read_mac_addr(&adapter->hw))
4921                 e_err("NVM Read Error while reading MAC address\n");
4922
4923         memcpy(netdev->dev_addr, adapter->hw.mac.addr, netdev->addr_len);
4924         memcpy(netdev->perm_addr, adapter->hw.mac.addr, netdev->addr_len);
4925
4926         if (!is_valid_ether_addr(netdev->perm_addr)) {
4927                 e_err("Invalid MAC Address: %02x:%02x:%02x:%02x:%02x:%02x\n",
4928                       netdev->perm_addr[0], netdev->perm_addr[1],
4929                       netdev->perm_addr[2], netdev->perm_addr[3],
4930                       netdev->perm_addr[4], netdev->perm_addr[5]);
4931                 err = -EIO;
4932                 goto err_eeprom;
4933         }
4934
4935         init_timer(&adapter->watchdog_timer);
4936         adapter->watchdog_timer.function = &e1000_watchdog;
4937         adapter->watchdog_timer.data = (unsigned long) adapter;
4938
4939         init_timer(&adapter->phy_info_timer);
4940         adapter->phy_info_timer.function = &e1000_update_phy_info;
4941         adapter->phy_info_timer.data = (unsigned long) adapter;
4942
4943         INIT_WORK(&adapter->reset_task, e1000_reset_task);
4944         INIT_WORK(&adapter->watchdog_task, e1000_watchdog_task);
4945         INIT_WORK(&adapter->downshift_task, e1000e_downshift_workaround);
4946         INIT_WORK(&adapter->update_phy_task, e1000e_update_phy_task);
4947
4948         /* Initialize link parameters. User can change them with ethtool */
4949         adapter->hw.mac.autoneg = 1;
4950         adapter->fc_autoneg = 1;
4951         adapter->hw.fc.original_type = e1000_fc_default;
4952         adapter->hw.fc.type = e1000_fc_default;
4953         adapter->hw.phy.autoneg_advertised = 0x2f;
4954
4955         /* ring size defaults */
4956         adapter->rx_ring->count = 256;
4957         adapter->tx_ring->count = 256;
4958
4959         /*
4960          * Initial Wake on LAN setting - If APM wake is enabled in
4961          * the EEPROM, enable the ACPI Magic Packet filter
4962          */
4963         if (adapter->flags & FLAG_APME_IN_WUC) {
4964                 /* APME bit in EEPROM is mapped to WUC.APME */
4965                 eeprom_data = er32(WUC);
4966                 eeprom_apme_mask = E1000_WUC_APME;
4967         } else if (adapter->flags & FLAG_APME_IN_CTRL3) {
4968                 if (adapter->flags & FLAG_APME_CHECK_PORT_B &&
4969                     (adapter->hw.bus.func == 1))
4970                         e1000_read_nvm(&adapter->hw,
4971                                 NVM_INIT_CONTROL3_PORT_B, 1, &eeprom_data);
4972                 else
4973                         e1000_read_nvm(&adapter->hw,
4974                                 NVM_INIT_CONTROL3_PORT_A, 1, &eeprom_data);
4975         }
4976
4977         /* fetch WoL from EEPROM */
4978         if (eeprom_data & eeprom_apme_mask)
4979                 adapter->eeprom_wol |= E1000_WUFC_MAG;
4980
4981         /*
4982          * now that we have the eeprom settings, apply the special cases
4983          * where the eeprom may be wrong or the board simply won't support
4984          * wake on lan on a particular port
4985          */
4986         if (!(adapter->flags & FLAG_HAS_WOL))
4987                 adapter->eeprom_wol = 0;
4988
4989         /* initialize the wol settings based on the eeprom settings */
4990         adapter->wol = adapter->eeprom_wol;
4991         device_set_wakeup_enable(&adapter->pdev->dev, adapter->wol);
4992
4993         /* reset the hardware with the new settings */
4994         e1000e_reset(adapter);
4995
4996         /*
4997          * If the controller has AMT, do not set DRV_LOAD until the interface
4998          * is up.  For all other cases, let the f/w know that the h/w is now
4999          * under the control of the driver.
5000          */
5001         if (!(adapter->flags & FLAG_HAS_AMT))
5002                 e1000_get_hw_control(adapter);
5003
5004         /* tell the stack to leave us alone until e1000_open() is called */
5005         netif_carrier_off(netdev);
5006         netif_tx_stop_all_queues(netdev);
5007
5008         strcpy(netdev->name, "eth%d");
5009         err = register_netdev(netdev);
5010         if (err)
5011                 goto err_register;
5012
5013         e1000_print_device_info(adapter);
5014
5015         return 0;
5016
5017 err_register:
5018         if (!(adapter->flags & FLAG_HAS_AMT))
5019                 e1000_release_hw_control(adapter);
5020 err_eeprom:
5021         if (!e1000_check_reset_block(&adapter->hw))
5022                 e1000_phy_hw_reset(&adapter->hw);
5023 err_hw_init:
5024
5025         kfree(adapter->tx_ring);
5026         kfree(adapter->rx_ring);
5027 err_sw_init:
5028         if (adapter->hw.flash_address)
5029                 iounmap(adapter->hw.flash_address);
5030         e1000e_reset_interrupt_capability(adapter);
5031 err_flashmap:
5032         iounmap(adapter->hw.hw_addr);
5033 err_ioremap:
5034         free_netdev(netdev);
5035 err_alloc_etherdev:
5036         pci_release_selected_regions(pdev,
5037                                      pci_select_bars(pdev, IORESOURCE_MEM));
5038 err_pci_reg:
5039 err_dma:
5040         pci_disable_device(pdev);
5041         return err;
5042 }
5043
5044 /**
5045  * e1000_remove - Device Removal Routine
5046  * @pdev: PCI device information struct
5047  *
5048  * e1000_remove is called by the PCI subsystem to alert the driver
5049  * that it should release a PCI device.  The could be caused by a
5050  * Hot-Plug event, or because the driver is going to be removed from
5051  * memory.
5052  **/
5053 static void __devexit e1000_remove(struct pci_dev *pdev)
5054 {
5055         struct net_device *netdev = pci_get_drvdata(pdev);
5056         struct e1000_adapter *adapter = netdev_priv(netdev);
5057
5058         /*
5059          * flush_scheduled work may reschedule our watchdog task, so
5060          * explicitly disable watchdog tasks from being rescheduled
5061          */
5062         set_bit(__E1000_DOWN, &adapter->state);
5063         del_timer_sync(&adapter->watchdog_timer);
5064         del_timer_sync(&adapter->phy_info_timer);
5065
5066         flush_scheduled_work();
5067
5068         /*
5069          * Release control of h/w to f/w.  If f/w is AMT enabled, this
5070          * would have already happened in close and is redundant.
5071          */
5072         e1000_release_hw_control(adapter);
5073
5074         unregister_netdev(netdev);
5075
5076         if (!e1000_check_reset_block(&adapter->hw))
5077                 e1000_phy_hw_reset(&adapter->hw);
5078
5079         e1000e_reset_interrupt_capability(adapter);
5080         kfree(adapter->tx_ring);
5081         kfree(adapter->rx_ring);
5082
5083         iounmap(adapter->hw.hw_addr);
5084         if (adapter->hw.flash_address)
5085                 iounmap(adapter->hw.flash_address);
5086         pci_release_selected_regions(pdev,
5087                                      pci_select_bars(pdev, IORESOURCE_MEM));
5088
5089         free_netdev(netdev);
5090
5091         pci_disable_device(pdev);
5092 }
5093
5094 /* PCI Error Recovery (ERS) */
5095 static struct pci_error_handlers e1000_err_handler = {
5096         .error_detected = e1000_io_error_detected,
5097         .slot_reset = e1000_io_slot_reset,
5098         .resume = e1000_io_resume,
5099 };
5100
5101 static struct pci_device_id e1000_pci_tbl[] = {
5102         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82571EB_COPPER), board_82571 },
5103         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82571EB_FIBER), board_82571 },
5104         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82571EB_QUAD_COPPER), board_82571 },
5105         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82571EB_QUAD_COPPER_LP), board_82571 },
5106         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82571EB_QUAD_FIBER), board_82571 },
5107         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82571EB_SERDES), board_82571 },
5108         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82571EB_SERDES_DUAL), board_82571 },
5109         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82571EB_SERDES_QUAD), board_82571 },
5110         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82571PT_QUAD_COPPER), board_82571 },
5111
5112         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82572EI), board_82572 },
5113         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82572EI_COPPER), board_82572 },
5114         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82572EI_FIBER), board_82572 },
5115         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82572EI_SERDES), board_82572 },
5116
5117         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82573E), board_82573 },
5118         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82573E_IAMT), board_82573 },
5119         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82573L), board_82573 },
5120
5121         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82574L), board_82574 },
5122
5123         { PCI_VDEVICE(INTEL, E1000_DEV_ID_80003ES2LAN_COPPER_DPT),
5124           board_80003es2lan },
5125         { PCI_VDEVICE(INTEL, E1000_DEV_ID_80003ES2LAN_COPPER_SPT),
5126           board_80003es2lan },
5127         { PCI_VDEVICE(INTEL, E1000_DEV_ID_80003ES2LAN_SERDES_DPT),
5128           board_80003es2lan },
5129         { PCI_VDEVICE(INTEL, E1000_DEV_ID_80003ES2LAN_SERDES_SPT),
5130           board_80003es2lan },
5131
5132         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH8_IFE), board_ich8lan },
5133         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH8_IFE_G), board_ich8lan },
5134         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH8_IFE_GT), board_ich8lan },
5135         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH8_IGP_AMT), board_ich8lan },
5136         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH8_IGP_C), board_ich8lan },
5137         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH8_IGP_M), board_ich8lan },
5138         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH8_IGP_M_AMT), board_ich8lan },
5139
5140         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH9_IFE), board_ich9lan },
5141         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH9_IFE_G), board_ich9lan },
5142         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH9_IFE_GT), board_ich9lan },
5143         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH9_IGP_AMT), board_ich9lan },
5144         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH9_IGP_C), board_ich9lan },
5145         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH9_BM), board_ich9lan },
5146         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH9_IGP_M), board_ich9lan },
5147         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH9_IGP_M_AMT), board_ich9lan },
5148         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH9_IGP_M_V), board_ich9lan },
5149
5150         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH10_R_BM_LM), board_ich9lan },
5151         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH10_R_BM_LF), board_ich9lan },
5152         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH10_R_BM_V), board_ich9lan },
5153
5154         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH10_D_BM_LM), board_ich10lan },
5155         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH10_D_BM_LF), board_ich10lan },
5156
5157         { }     /* terminate list */
5158 };
5159 MODULE_DEVICE_TABLE(pci, e1000_pci_tbl);
5160
5161 /* PCI Device API Driver */
5162 static struct pci_driver e1000_driver = {
5163         .name     = e1000e_driver_name,
5164         .id_table = e1000_pci_tbl,
5165         .probe    = e1000_probe,
5166         .remove   = __devexit_p(e1000_remove),
5167 #ifdef CONFIG_PM
5168         /* Power Management Hooks */
5169         .suspend  = e1000_suspend,
5170         .resume   = e1000_resume,
5171 #endif
5172         .shutdown = e1000_shutdown,
5173         .err_handler = &e1000_err_handler
5174 };
5175
5176 /**
5177  * e1000_init_module - Driver Registration Routine
5178  *
5179  * e1000_init_module is the first routine called when the driver is
5180  * loaded. All it does is register with the PCI subsystem.
5181  **/
5182 static int __init e1000_init_module(void)
5183 {
5184         int ret;
5185         printk(KERN_INFO "%s: Intel(R) PRO/1000 Network Driver - %s\n",
5186                e1000e_driver_name, e1000e_driver_version);
5187         printk(KERN_INFO "%s: Copyright (c) 1999-2008 Intel Corporation.\n",
5188                e1000e_driver_name);
5189         ret = pci_register_driver(&e1000_driver);
5190         pm_qos_add_requirement(PM_QOS_CPU_DMA_LATENCY, e1000e_driver_name,
5191                                PM_QOS_DEFAULT_VALUE);
5192
5193         return ret;
5194 }
5195 module_init(e1000_init_module);
5196
5197 /**
5198  * e1000_exit_module - Driver Exit Cleanup Routine
5199  *
5200  * e1000_exit_module is called just before the driver is removed
5201  * from memory.
5202  **/
5203 static void __exit e1000_exit_module(void)
5204 {
5205         pci_unregister_driver(&e1000_driver);
5206         pm_qos_remove_requirement(PM_QOS_CPU_DMA_LATENCY, e1000e_driver_name);
5207 }
5208 module_exit(e1000_exit_module);
5209
5210
5211 MODULE_AUTHOR("Intel Corporation, <linux.nics@intel.com>");
5212 MODULE_DESCRIPTION("Intel(R) PRO/1000 Network Driver");
5213 MODULE_LICENSE("GPL");
5214 MODULE_VERSION(DRV_VERSION);
5215
5216 /* e1000_main.c */