git.oblomov.eu Git - linux-2.6/blob - 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  * @rx_ring: pointer to receive ring structure
 349  * @cleaned_count: number of buffers to allocate this pass
 350  **/
 351
 352 static void e1000_alloc_jumbo_rx_buffers(struct e1000_adapter *adapter,
 353                                          int cleaned_count)
 354 {
 355         struct net_device *netdev = adapter->netdev;
 356         struct pci_dev *pdev = adapter->pdev;
 357         struct e1000_rx_desc *rx_desc;
 358         struct e1000_ring *rx_ring = adapter->rx_ring;
 359         struct e1000_buffer *buffer_info;
 360         struct sk_buff *skb;
 361         unsigned int i;
 362         unsigned int bufsz = 256 -
 363                              16 /* for skb_reserve */ -
 364                              NET_IP_ALIGN;
 365
 366         i = rx_ring->next_to_use;
 367         buffer_info = &rx_ring->buffer_info[i];
 368
 369         while (cleaned_count--) {
 370                 skb = buffer_info->skb;
 371                 if (skb) {
 372                         skb_trim(skb, 0);
 373                         goto check_page;
 374                 }
 375
 376                 skb = netdev_alloc_skb(netdev, bufsz);
 377                 if (unlikely(!skb)) {
 378                         /* Better luck next round */
 379                         adapter->alloc_rx_buff_failed++;
 380                         break;
 381                 }
 382
 383                 /* Make buffer alignment 2 beyond a 16 byte boundary
 384                  * this will result in a 16 byte aligned IP header after
 385                  * the 14 byte MAC header is removed
 386                  */
 387                 skb_reserve(skb, NET_IP_ALIGN);
 388
 389                 buffer_info->skb = skb;
 390 check_page:
 391                 /* allocate a new page if necessary */
 392                 if (!buffer_info->page) {
 393                         buffer_info->page = alloc_page(GFP_ATOMIC);
 394                         if (unlikely(!buffer_info->page)) {
 395                                 adapter->alloc_rx_buff_failed++;
 396                                 break;
 397                         }
 398                 }
 399
 400                 if (!buffer_info->dma)
 401                         buffer_info->dma = pci_map_page(pdev,
 402                                                         buffer_info->page, 0,
 403                                                         PAGE_SIZE,
 404                                                         PCI_DMA_FROMDEVICE);
 405
 406                 rx_desc = E1000_RX_DESC(*rx_ring, i);
 407                 rx_desc->buffer_addr = cpu_to_le64(buffer_info->dma);
 408
 409                 if (unlikely(++i == rx_ring->count))
 410                         i = 0;
 411                 buffer_info = &rx_ring->buffer_info[i];
 412         }
 413
 414         if (likely(rx_ring->next_to_use != i)) {
 415                 rx_ring->next_to_use = i;
 416                 if (unlikely(i-- == 0))
 417                         i = (rx_ring->count - 1);
 418
 419                 /* Force memory writes to complete before letting h/w
 420                  * know there are new descriptors to fetch.  (Only
 421                  * applicable for weak-ordered memory model archs,
 422                  * such as IA-64). */
 423                 wmb();
 424                 writel(i, adapter->hw.hw_addr + rx_ring->tail);
 425         }
 426 }
 427
 428 /**
 429  * e1000_clean_rx_irq - Send received data up the network stack; legacy
 430  * @adapter: board private structure
 431  *
 432  * the return value indicates whether actual cleaning was done, there
 433  * is no guarantee that everything was cleaned
 434  **/
 435 static bool e1000_clean_rx_irq(struct e1000_adapter *adapter,
 436                                int *work_done, int work_to_do)
 437 {
 438         struct net_device *netdev = adapter->netdev;
 439         struct pci_dev *pdev = adapter->pdev;
 440         struct e1000_ring *rx_ring = adapter->rx_ring;
 441         struct e1000_rx_desc *rx_desc, *next_rxd;
 442         struct e1000_buffer *buffer_info, *next_buffer;
 443         u32 length;
 444         unsigned int i;
 445         int cleaned_count = 0;
 446         bool cleaned = 0;
 447         unsigned int total_rx_bytes = 0, total_rx_packets = 0;
 448
 449         i = rx_ring->next_to_clean;
 450         rx_desc = E1000_RX_DESC(*rx_ring, i);
 451         buffer_info = &rx_ring->buffer_info[i];
 452
 453         while (rx_desc->status & E1000_RXD_STAT_DD) {
 454                 struct sk_buff *skb;
 455                 u8 status;
 456
 457                 if (*work_done >= work_to_do)
 458                         break;
 459                 (*work_done)++;
 460
 461                 status = rx_desc->status;
 462                 skb = buffer_info->skb;
 463                 buffer_info->skb = NULL;
 464
 465                 prefetch(skb->data - NET_IP_ALIGN);
 466
 467                 i++;
 468                 if (i == rx_ring->count)
 469                         i = 0;
 470                 next_rxd = E1000_RX_DESC(*rx_ring, i);
 471                 prefetch(next_rxd);
 472
 473                 next_buffer = &rx_ring->buffer_info[i];
 474
 475                 cleaned = 1;
 476                 cleaned_count++;
 477                 pci_unmap_single(pdev,
 478                                  buffer_info->dma,
 479                                  adapter->rx_buffer_len,
 480                                  PCI_DMA_FROMDEVICE);
 481                 buffer_info->dma = 0;
 482
 483                 length = le16_to_cpu(rx_desc->length);
 484
 485                 /* !EOP means multiple descriptors were used to store a single
 486                  * packet, also make sure the frame isn't just CRC only */
 487                 if (!(status & E1000_RXD_STAT_EOP) || (length <= 4)) {
 488                         /* All receives must fit into a single buffer */
 489                         e_dbg("%s: Receive packet consumed multiple buffers\n",
 490                               netdev->name);
 491                         /* recycle */
 492                         buffer_info->skb = skb;
 493                         goto next_desc;
 494                 }
 495
 496                 if (rx_desc->errors & E1000_RXD_ERR_FRAME_ERR_MASK) {
 497                         /* recycle */
 498                         buffer_info->skb = skb;
 499                         goto next_desc;
 500                 }
 501
 502                 total_rx_bytes += length;
 503                 total_rx_packets++;
 504
 505                 /*
 506                  * code added for copybreak, this should improve
 507                  * performance for small packets with large amounts
 508                  * of reassembly being done in the stack
 509                  */
 510                 if (length < copybreak) {
 511                         struct sk_buff *new_skb =
 512                             netdev_alloc_skb(netdev, length + NET_IP_ALIGN);
 513                         if (new_skb) {
 514                                 skb_reserve(new_skb, NET_IP_ALIGN);
 515                                 skb_copy_to_linear_data_offset(new_skb,
 516                                                                -NET_IP_ALIGN,
 517                                                                (skb->data -
 518                                                                 NET_IP_ALIGN),
 519                                                                (length +
 520                                                                 NET_IP_ALIGN));
 521                                 /* save the skb in buffer_info as good */
 522                                 buffer_info->skb = skb;
 523                                 skb = new_skb;
 524                         }
 525                         /* else just continue with the old one */
 526                 }
 527                 /* end copybreak code */
 528                 skb_put(skb, length);
 529
 530                 /* Receive Checksum Offload */
 531                 e1000_rx_checksum(adapter,
 532                                   (u32)(status) |
 533                                   ((u32)(rx_desc->errors) << 24),
 534                                   le16_to_cpu(rx_desc->csum), skb);
 535
 536                 e1000_receive_skb(adapter, netdev, skb,status,rx_desc->special);
 537
 538 next_desc:
 539                 rx_desc->status = 0;
 540
 541                 /* return some buffers to hardware, one at a time is too slow */
 542                 if (cleaned_count >= E1000_RX_BUFFER_WRITE) {
 543                         adapter->alloc_rx_buf(adapter, cleaned_count);
 544                         cleaned_count = 0;
 545                 }
 546
 547                 /* use prefetched values */
 548                 rx_desc = next_rxd;
 549                 buffer_info = next_buffer;
 550         }
 551         rx_ring->next_to_clean = i;
 552
 553         cleaned_count = e1000_desc_unused(rx_ring);
 554         if (cleaned_count)
 555                 adapter->alloc_rx_buf(adapter, cleaned_count);
 556
 557         adapter->total_rx_bytes += total_rx_bytes;
 558         adapter->total_rx_packets += total_rx_packets;
 559         adapter->net_stats.rx_bytes += total_rx_bytes;
 560         adapter->net_stats.rx_packets += total_rx_packets;
 561         return cleaned;
 562 }
 563
 564 static void e1000_put_txbuf(struct e1000_adapter *adapter,
 565                              struct e1000_buffer *buffer_info)
 566 {
 567         if (buffer_info->dma) {
 568                 pci_unmap_page(adapter->pdev, buffer_info->dma,
 569                                buffer_info->length, PCI_DMA_TODEVICE);
 570                 buffer_info->dma = 0;
 571         }
 572         if (buffer_info->skb) {
 573                 dev_kfree_skb_any(buffer_info->skb);
 574                 buffer_info->skb = NULL;
 575         }
 576 }
 577
 578 static void e1000_print_tx_hang(struct e1000_adapter *adapter)
 579 {
 580         struct e1000_ring *tx_ring = adapter->tx_ring;
 581         unsigned int i = tx_ring->next_to_clean;
 582         unsigned int eop = tx_ring->buffer_info[i].next_to_watch;
 583         struct e1000_tx_desc *eop_desc = E1000_TX_DESC(*tx_ring, eop);
 584
 585         /* detected Tx unit hang */
 586         e_err("Detected Tx Unit Hang:\n"
 587               "  TDH                  <%x>\n"
 588               "  TDT                  <%x>\n"
 589               "  next_to_use          <%x>\n"
 590               "  next_to_clean        <%x>\n"
 591               "buffer_info[next_to_clean]:\n"
 592               "  time_stamp           <%lx>\n"
 593               "  next_to_watch        <%x>\n"
 594               "  jiffies              <%lx>\n"
 595               "  next_to_watch.status <%x>\n",
 596               readl(adapter->hw.hw_addr + tx_ring->head),
 597               readl(adapter->hw.hw_addr + tx_ring->tail),
 598               tx_ring->next_to_use,
 599               tx_ring->next_to_clean,
 600               tx_ring->buffer_info[eop].time_stamp,
 601               eop,
 602               jiffies,
 603               eop_desc->upper.fields.status);
 604 }
 605
 606 /**
 607  * e1000_clean_tx_irq - Reclaim resources after transmit completes
 608  * @adapter: board private structure
 609  *
 610  * the return value indicates whether actual cleaning was done, there
 611  * is no guarantee that everything was cleaned
 612  **/
 613 static bool e1000_clean_tx_irq(struct e1000_adapter *adapter)
 614 {
 615         struct net_device *netdev = adapter->netdev;
 616         struct e1000_hw *hw = &adapter->hw;
 617         struct e1000_ring *tx_ring = adapter->tx_ring;
 618         struct e1000_tx_desc *tx_desc, *eop_desc;
 619         struct e1000_buffer *buffer_info;
 620         unsigned int i, eop;
 621         unsigned int count = 0;
 622         bool cleaned = 0;
 623         unsigned int total_tx_bytes = 0, total_tx_packets = 0;
 624
 625         i = tx_ring->next_to_clean;
 626         eop = tx_ring->buffer_info[i].next_to_watch;
 627         eop_desc = E1000_TX_DESC(*tx_ring, eop);
 628
 629         while (eop_desc->upper.data & cpu_to_le32(E1000_TXD_STAT_DD)) {
 630                 for (cleaned = 0; !cleaned; ) {
 631                         tx_desc = E1000_TX_DESC(*tx_ring, i);
 632                         buffer_info = &tx_ring->buffer_info[i];
 633                         cleaned = (i == eop);
 634
 635                         if (cleaned) {
 636                                 struct sk_buff *skb = buffer_info->skb;
 637                                 unsigned int segs, bytecount;
 638                                 segs = skb_shinfo(skb)->gso_segs ?: 1;
 639                                 /* multiply data chunks by size of headers */
 640                                 bytecount = ((segs - 1) * skb_headlen(skb)) +
 641                                             skb->len;
 642                                 total_tx_packets += segs;
 643                                 total_tx_bytes += bytecount;
 644                         }
 645
 646                         e1000_put_txbuf(adapter, buffer_info);
 647                         tx_desc->upper.data = 0;
 648
 649                         i++;
 650                         if (i == tx_ring->count)
 651                                 i = 0;
 652                 }
 653
 654                 eop = tx_ring->buffer_info[i].next_to_watch;
 655                 eop_desc = E1000_TX_DESC(*tx_ring, eop);
 656 #define E1000_TX_WEIGHT 64
 657                 /* weight of a sort for tx, to avoid endless transmit cleanup */
 658                 if (count++ == E1000_TX_WEIGHT)
 659                         break;
 660         }
 661
 662         tx_ring->next_to_clean = i;
 663
 664 #define TX_WAKE_THRESHOLD 32
 665         if (cleaned && netif_carrier_ok(netdev) &&
 666                      e1000_desc_unused(tx_ring) >= TX_WAKE_THRESHOLD) {
 667                 /* Make sure that anybody stopping the queue after this
 668                  * sees the new next_to_clean.
 669                  */
 670                 smp_mb();
 671
 672                 if (netif_queue_stopped(netdev) &&
 673                     !(test_bit(__E1000_DOWN, &adapter->state))) {
 674                         netif_wake_queue(netdev);
 675                         ++adapter->restart_queue;
 676                 }
 677         }
 678
 679         if (adapter->detect_tx_hung) {
 680                 /*
 681                  * Detect a transmit hang in hardware, this serializes the
 682                  * check with the clearing of time_stamp and movement of i
 683                  */
 684                 adapter->detect_tx_hung = 0;
 685                 if (tx_ring->buffer_info[eop].dma &&
 686                     time_after(jiffies, tx_ring->buffer_info[eop].time_stamp
 687                                + (adapter->tx_timeout_factor * HZ))
 688                     && !(er32(STATUS) & E1000_STATUS_TXOFF)) {
 689                         e1000_print_tx_hang(adapter);
 690                         netif_stop_queue(netdev);
 691                 }
 692         }
 693         adapter->total_tx_bytes += total_tx_bytes;
 694         adapter->total_tx_packets += total_tx_packets;
 695         adapter->net_stats.tx_bytes += total_tx_bytes;
 696         adapter->net_stats.tx_packets += total_tx_packets;
 697         return cleaned;
 698 }
 699
 700 /**
 701  * e1000_clean_rx_irq_ps - Send received data up the network stack; packet split
 702  * @adapter: board private structure
 703  *
 704  * the return value indicates whether actual cleaning was done, there
 705  * is no guarantee that everything was cleaned
 706  **/
 707 static bool e1000_clean_rx_irq_ps(struct e1000_adapter *adapter,
 708                                   int *work_done, int work_to_do)
 709 {
 710         union e1000_rx_desc_packet_split *rx_desc, *next_rxd;
 711         struct net_device *netdev = adapter->netdev;
 712         struct pci_dev *pdev = adapter->pdev;
 713         struct e1000_ring *rx_ring = adapter->rx_ring;
 714         struct e1000_buffer *buffer_info, *next_buffer;
 715         struct e1000_ps_page *ps_page;
 716         struct sk_buff *skb;
 717         unsigned int i, j;
 718         u32 length, staterr;
 719         int cleaned_count = 0;
 720         bool cleaned = 0;
 721         unsigned int total_rx_bytes = 0, total_rx_packets = 0;
 722
 723         i = rx_ring->next_to_clean;
 724         rx_desc = E1000_RX_DESC_PS(*rx_ring, i);
 725         staterr = le32_to_cpu(rx_desc->wb.middle.status_error);
 726         buffer_info = &rx_ring->buffer_info[i];
 727
 728         while (staterr & E1000_RXD_STAT_DD) {
 729                 if (*work_done >= work_to_do)
 730                         break;
 731                 (*work_done)++;
 732                 skb = buffer_info->skb;
 733
 734                 /* in the packet split case this is header only */
 735                 prefetch(skb->data - NET_IP_ALIGN);
 736
 737                 i++;
 738                 if (i == rx_ring->count)
 739                         i = 0;
 740                 next_rxd = E1000_RX_DESC_PS(*rx_ring, i);
 741                 prefetch(next_rxd);
 742
 743                 next_buffer = &rx_ring->buffer_info[i];
 744
 745                 cleaned = 1;
 746                 cleaned_count++;
 747                 pci_unmap_single(pdev, buffer_info->dma,
 748                                  adapter->rx_ps_bsize0,
 749                                  PCI_DMA_FROMDEVICE);
 750                 buffer_info->dma = 0;
 751
 752                 if (!(staterr & E1000_RXD_STAT_EOP)) {
 753                         e_dbg("%s: Packet Split buffers didn't pick up the "
 754                               "full packet\n", netdev->name);
 755                         dev_kfree_skb_irq(skb);
 756                         goto next_desc;
 757                 }
 758
 759                 if (staterr & E1000_RXDEXT_ERR_FRAME_ERR_MASK) {
 760                         dev_kfree_skb_irq(skb);
 761                         goto next_desc;
 762                 }
 763
 764                 length = le16_to_cpu(rx_desc->wb.middle.length0);
 765
 766                 if (!length) {
 767                         e_dbg("%s: Last part of the packet spanning multiple "
 768                               "descriptors\n", netdev->name);
 769                         dev_kfree_skb_irq(skb);
 770                         goto next_desc;
 771                 }
 772
 773                 /* Good Receive */
 774                 skb_put(skb, length);
 775
 776                 {
 777                 /*
 778                  * this looks ugly, but it seems compiler issues make it
 779                  * more efficient than reusing j
 780                  */
 781                 int l1 = le16_to_cpu(rx_desc->wb.upper.length[0]);
 782
 783                 /*
 784                  * page alloc/put takes too long and effects small packet
 785                  * throughput, so unsplit small packets and save the alloc/put
 786                  * only valid in softirq (napi) context to call kmap_*
 787                  */
 788                 if (l1 && (l1 <= copybreak) &&
 789                     ((length + l1) <= adapter->rx_ps_bsize0)) {
 790                         u8 *vaddr;
 791
 792                         ps_page = &buffer_info->ps_pages[0];
 793
 794                         /*
 795                          * there is no documentation about how to call
 796                          * kmap_atomic, so we can't hold the mapping
 797                          * very long
 798                          */
 799                         pci_dma_sync_single_for_cpu(pdev, ps_page->dma,
 800                                 PAGE_SIZE, PCI_DMA_FROMDEVICE);
 801                         vaddr = kmap_atomic(ps_page->page, KM_SKB_DATA_SOFTIRQ);
 802                         memcpy(skb_tail_pointer(skb), vaddr, l1);
 803                         kunmap_atomic(vaddr, KM_SKB_DATA_SOFTIRQ);
 804                         pci_dma_sync_single_for_device(pdev, ps_page->dma,
 805                                 PAGE_SIZE, PCI_DMA_FROMDEVICE);
 806
 807                         skb_put(skb, l1);
 808                         goto copydone;
 809                 } /* if */
 810                 }
 811
 812                 for (j = 0; j < PS_PAGE_BUFFERS; j++) {
 813                         length = le16_to_cpu(rx_desc->wb.upper.length[j]);
 814                         if (!length)
 815                                 break;
 816
 817                         ps_page = &buffer_info->ps_pages[j];
 818                         pci_unmap_page(pdev, ps_page->dma, PAGE_SIZE,
 819                                        PCI_DMA_FROMDEVICE);
 820                         ps_page->dma = 0;
 821                         skb_fill_page_desc(skb, j, ps_page->page, 0, length);
 822                         ps_page->page = NULL;
 823                         skb->len += length;
 824                         skb->data_len += length;
 825                         skb->truesize += length;
 826                 }
 827
 828 copydone:
 829                 total_rx_bytes += skb->len;
 830                 total_rx_packets++;
 831
 832                 e1000_rx_checksum(adapter, staterr, le16_to_cpu(
 833                         rx_desc->wb.lower.hi_dword.csum_ip.csum), skb);
 834
 835                 if (rx_desc->wb.upper.header_status &
 836                            cpu_to_le16(E1000_RXDPS_HDRSTAT_HDRSP))
 837                         adapter->rx_hdr_split++;
 838
 839                 e1000_receive_skb(adapter, netdev, skb,
 840                                   staterr, rx_desc->wb.middle.vlan);
 841
 842 next_desc:
 843                 rx_desc->wb.middle.status_error &= cpu_to_le32(~0xFF);
 844                 buffer_info->skb = NULL;
 845
 846                 /* return some buffers to hardware, one at a time is too slow */
 847                 if (cleaned_count >= E1000_RX_BUFFER_WRITE) {
 848                         adapter->alloc_rx_buf(adapter, cleaned_count);
 849                         cleaned_count = 0;
 850                 }
 851
 852                 /* use prefetched values */
 853                 rx_desc = next_rxd;
 854                 buffer_info = next_buffer;
 855
 856                 staterr = le32_to_cpu(rx_desc->wb.middle.status_error);
 857         }
 858         rx_ring->next_to_clean = i;
 859
 860         cleaned_count = e1000_desc_unused(rx_ring);
 861         if (cleaned_count)
 862                 adapter->alloc_rx_buf(adapter, cleaned_count);
 863
 864         adapter->total_rx_bytes += total_rx_bytes;
 865         adapter->total_rx_packets += total_rx_packets;
 866         adapter->net_stats.rx_bytes += total_rx_bytes;
 867         adapter->net_stats.rx_packets += total_rx_packets;
 868         return cleaned;
 869 }
 870
 871 /**
 872  * e1000_consume_page - helper function
 873  **/
 874 static void e1000_consume_page(struct e1000_buffer *bi, struct sk_buff *skb,
 875                                u16 length)
 876 {
 877         bi->page = NULL;
 878         skb->len += length;
 879         skb->data_len += length;
 880         skb->truesize += length;
 881 }
 882
 883 /**
 884  * e1000_clean_jumbo_rx_irq - Send received data up the network stack; legacy
 885  * @adapter: board private structure
 886  *
 887  * the return value indicates whether actual cleaning was done, there
 888  * is no guarantee that everything was cleaned
 889  **/
 890
 891 static bool e1000_clean_jumbo_rx_irq(struct e1000_adapter *adapter,
 892                                      int *work_done, int work_to_do)
 893 {
 894         struct net_device *netdev = adapter->netdev;
 895         struct pci_dev *pdev = adapter->pdev;
 896         struct e1000_ring *rx_ring = adapter->rx_ring;
 897         struct e1000_rx_desc *rx_desc, *next_rxd;
 898         struct e1000_buffer *buffer_info, *next_buffer;
 899         u32 length;
 900         unsigned int i;
 901         int cleaned_count = 0;
 902         bool cleaned = false;
 903         unsigned int total_rx_bytes=0, total_rx_packets=0;
 904
 905         i = rx_ring->next_to_clean;
 906         rx_desc = E1000_RX_DESC(*rx_ring, i);
 907         buffer_info = &rx_ring->buffer_info[i];
 908
 909         while (rx_desc->status & E1000_RXD_STAT_DD) {
 910                 struct sk_buff *skb;
 911                 u8 status;
 912
 913                 if (*work_done >= work_to_do)
 914                         break;
 915                 (*work_done)++;
 916
 917                 status = rx_desc->status;
 918                 skb = buffer_info->skb;
 919                 buffer_info->skb = NULL;
 920
 921                 ++i;
 922                 if (i == rx_ring->count)
 923                         i = 0;
 924                 next_rxd = E1000_RX_DESC(*rx_ring, i);
 925                 prefetch(next_rxd);
 926
 927                 next_buffer = &rx_ring->buffer_info[i];
 928
 929                 cleaned = true;
 930                 cleaned_count++;
 931                 pci_unmap_page(pdev, buffer_info->dma, PAGE_SIZE,
 932                                PCI_DMA_FROMDEVICE);
 933                 buffer_info->dma = 0;
 934
 935                 length = le16_to_cpu(rx_desc->length);
 936
 937                 /* errors is only valid for DD + EOP descriptors */
 938                 if (unlikely((status & E1000_RXD_STAT_EOP) &&
 939                     (rx_desc->errors & E1000_RXD_ERR_FRAME_ERR_MASK))) {
 940                                 /* recycle both page and skb */
 941                                 buffer_info->skb = skb;
 942                                 /* an error means any chain goes out the window
 943                                  * too */
 944                                 if (rx_ring->rx_skb_top)
 945                                         dev_kfree_skb(rx_ring->rx_skb_top);
 946                                 rx_ring->rx_skb_top = NULL;
 947                                 goto next_desc;
 948                 }
 949
 950 #define rxtop rx_ring->rx_skb_top
 951                 if (!(status & E1000_RXD_STAT_EOP)) {
 952                         /* this descriptor is only the beginning (or middle) */
 953                         if (!rxtop) {
 954                                 /* this is the beginning of a chain */
 955                                 rxtop = skb;
 956                                 skb_fill_page_desc(rxtop, 0, buffer_info->page,
 957                                                    0, length);
 958                         } else {
 959                                 /* this is the middle of a chain */
 960                                 skb_fill_page_desc(rxtop,
 961                                     skb_shinfo(rxtop)->nr_frags,
 962                                     buffer_info->page, 0, length);
 963                                 /* re-use the skb, only consumed the page */
 964                                 buffer_info->skb = skb;
 965                         }
 966                         e1000_consume_page(buffer_info, rxtop, length);
 967                         goto next_desc;
 968                 } else {
 969                         if (rxtop) {
 970                                 /* end of the chain */
 971                                 skb_fill_page_desc(rxtop,
 972                                     skb_shinfo(rxtop)->nr_frags,
 973                                     buffer_info->page, 0, length);
 974                                 /* re-use the current skb, we only consumed the
 975                                  * page */
 976                                 buffer_info->skb = skb;
 977                                 skb = rxtop;
 978                                 rxtop = NULL;
 979                                 e1000_consume_page(buffer_info, skb, length);
 980                         } else {
 981                                 /* no chain, got EOP, this buf is the packet
 982                                  * copybreak to save the put_page/alloc_page */
 983                                 if (length <= copybreak &&
 984                                     skb_tailroom(skb) >= length) {
 985                                         u8 *vaddr;
 986                                         vaddr = kmap_atomic(buffer_info->page,
 987                                                            KM_SKB_DATA_SOFTIRQ);
 988                                         memcpy(skb_tail_pointer(skb), vaddr,
 989                                                length);
 990                                         kunmap_atomic(vaddr,
 991                                                       KM_SKB_DATA_SOFTIRQ);
 992                                         /* re-use the page, so don't erase
 993                                          * buffer_info->page */
 994                                         skb_put(skb, length);
 995                                 } else {
 996                                         skb_fill_page_desc(skb, 0,
 997                                                            buffer_info->page, 0,
 998                                                            length);
 999                                         e1000_consume_page(buffer_info, skb,
1000                                                            length);
1001                                 }
1002                         }
1003                 }
1004
1005                 /* Receive Checksum Offload XXX recompute due to CRC strip? */
1006                 e1000_rx_checksum(adapter,
1007                                   (u32)(status) |
1008                                   ((u32)(rx_desc->errors) << 24),
1009                                   le16_to_cpu(rx_desc->csum), skb);
1010
1011                 /* probably a little skewed due to removing CRC */
1012                 total_rx_bytes += skb->len;
1013                 total_rx_packets++;
1014
1015                 /* eth type trans needs skb->data to point to something */
1016                 if (!pskb_may_pull(skb, ETH_HLEN)) {
1017                         e_err("pskb_may_pull failed.\n");
1018                         dev_kfree_skb(skb);
1019                         goto next_desc;
1020                 }
1021
1022                 e1000_receive_skb(adapter, netdev, skb, status,
1023                                   rx_desc->special);
1024
1025 next_desc:
1026                 rx_desc->status = 0;
1027
1028                 /* return some buffers to hardware, one at a time is too slow */
1029                 if (unlikely(cleaned_count >= E1000_RX_BUFFER_WRITE)) {
1030                         adapter->alloc_rx_buf(adapter, cleaned_count);
1031                         cleaned_count = 0;
1032                 }
1033
1034                 /* use prefetched values */
1035                 rx_desc = next_rxd;
1036                 buffer_info = next_buffer;
1037         }
1038         rx_ring->next_to_clean = i;
1039
1040         cleaned_count = e1000_desc_unused(rx_ring);
1041         if (cleaned_count)
1042                 adapter->alloc_rx_buf(adapter, cleaned_count);
1043
1044         adapter->total_rx_bytes += total_rx_bytes;
1045         adapter->total_rx_packets += total_rx_packets;
1046         adapter->net_stats.rx_bytes += total_rx_bytes;
1047         adapter->net_stats.rx_packets += total_rx_packets;
1048         return cleaned;
1049 }
1050
1051 /**
1052  * e1000_clean_rx_ring - Free Rx Buffers per Queue
1053  * @adapter: board private structure
1054  **/
1055 static void e1000_clean_rx_ring(struct e1000_adapter *adapter)
1056 {
1057         struct e1000_ring *rx_ring = adapter->rx_ring;
1058         struct e1000_buffer *buffer_info;
1059         struct e1000_ps_page *ps_page;
1060         struct pci_dev *pdev = adapter->pdev;
1061         unsigned int i, j;
1062
1063         /* Free all the Rx ring sk_buffs */
1064         for (i = 0; i < rx_ring->count; i++) {
1065                 buffer_info = &rx_ring->buffer_info[i];
1066                 if (buffer_info->dma) {
1067                         if (adapter->clean_rx == e1000_clean_rx_irq)
1068                                 pci_unmap_single(pdev, buffer_info->dma,
1069                                                  adapter->rx_buffer_len,
1070                                                  PCI_DMA_FROMDEVICE);
1071                         else if (adapter->clean_rx == e1000_clean_jumbo_rx_irq)
1072                                 pci_unmap_page(pdev, buffer_info->dma,
1073                                                PAGE_SIZE,
1074                                                PCI_DMA_FROMDEVICE);
1075                         else if (adapter->clean_rx == e1000_clean_rx_irq_ps)
1076                                 pci_unmap_single(pdev, buffer_info->dma,
1077                                                  adapter->rx_ps_bsize0,
1078                                                  PCI_DMA_FROMDEVICE);
1079                         buffer_info->dma = 0;
1080                 }
1081
1082                 if (buffer_info->page) {
1083                         put_page(buffer_info->page);
1084                         buffer_info->page = NULL;
1085                 }
1086
1087                 if (buffer_info->skb) {
1088                         dev_kfree_skb(buffer_info->skb);
1089                         buffer_info->skb = NULL;
1090                 }
1091
1092                 for (j = 0; j < PS_PAGE_BUFFERS; j++) {
1093                         ps_page = &buffer_info->ps_pages[j];
1094                         if (!ps_page->page)
1095                                 break;
1096                         pci_unmap_page(pdev, ps_page->dma, PAGE_SIZE,
1097                                        PCI_DMA_FROMDEVICE);
1098                         ps_page->dma = 0;
1099                         put_page(ps_page->page);
1100                         ps_page->page = NULL;
1101                 }
1102         }
1103
1104         /* there also may be some cached data from a chained receive */
1105         if (rx_ring->rx_skb_top) {
1106                 dev_kfree_skb(rx_ring->rx_skb_top);
1107                 rx_ring->rx_skb_top = NULL;
1108         }
1109
1110         /* Zero out the descriptor ring */
1111         memset(rx_ring->desc, 0, rx_ring->size);
1112
1113         rx_ring->next_to_clean = 0;
1114         rx_ring->next_to_use = 0;
1115
1116         writel(0, adapter->hw.hw_addr + rx_ring->head);
1117         writel(0, adapter->hw.hw_addr + rx_ring->tail);
1118 }
1119
1120 static void e1000e_downshift_workaround(struct work_struct *work)
1121 {
1122         struct e1000_adapter *adapter = container_of(work,
1123                                         struct e1000_adapter, downshift_task);
1124
1125         e1000e_gig_downshift_workaround_ich8lan(&adapter->hw);
1126 }
1127
1128 /**
1129  * e1000_intr_msi - Interrupt Handler
1130  * @irq: interrupt number
1131  * @data: pointer to a network interface device structure
1132  **/
1133 static irqreturn_t e1000_intr_msi(int irq, void *data)
1134 {
1135         struct net_device *netdev = data;
1136         struct e1000_adapter *adapter = netdev_priv(netdev);
1137         struct e1000_hw *hw = &adapter->hw;
1138         u32 icr = er32(ICR);
1139
1140         /*
1141          * read ICR disables interrupts using IAM
1142          */
1143
1144         if (icr & (E1000_ICR_RXSEQ | E1000_ICR_LSC)) {
1145                 hw->mac.get_link_status = 1;
1146                 /*
1147                  * ICH8 workaround-- Call gig speed drop workaround on cable
1148                  * disconnect (LSC) before accessing any PHY registers
1149                  */
1150                 if ((adapter->flags & FLAG_LSC_GIG_SPEED_DROP) &&
1151                     (!(er32(STATUS) & E1000_STATUS_LU)))
1152                         schedule_work(&adapter->downshift_task);
1153
1154                 /*
1155                  * 80003ES2LAN workaround-- For packet buffer work-around on
1156                  * link down event; disable receives here in the ISR and reset
1157                  * adapter in watchdog
1158                  */
1159                 if (netif_carrier_ok(netdev) &&
1160                     adapter->flags & FLAG_RX_NEEDS_RESTART) {
1161                         /* disable receives */
1162                         u32 rctl = er32(RCTL);
1163                         ew32(RCTL, rctl & ~E1000_RCTL_EN);
1164                         adapter->flags |= FLAG_RX_RESTART_NOW;
1165                 }
1166                 /* guard against interrupt when we're going down */
1167                 if (!test_bit(__E1000_DOWN, &adapter->state))
1168                         mod_timer(&adapter->watchdog_timer, jiffies + 1);
1169         }
1170
1171         if (netif_rx_schedule_prep(netdev, &adapter->napi)) {
1172                 adapter->total_tx_bytes = 0;
1173                 adapter->total_tx_packets = 0;
1174                 adapter->total_rx_bytes = 0;
1175                 adapter->total_rx_packets = 0;
1176                 __netif_rx_schedule(netdev, &adapter->napi);
1177         }
1178
1179         return IRQ_HANDLED;
1180 }
1181
1182 /**
1183  * e1000_intr - Interrupt Handler
1184  * @irq: interrupt number
1185  * @data: pointer to a network interface device structure
1186  **/
1187 static irqreturn_t e1000_intr(int irq, void *data)
1188 {
1189         struct net_device *netdev = data;
1190         struct e1000_adapter *adapter = netdev_priv(netdev);
1191         struct e1000_hw *hw = &adapter->hw;
1192         u32 rctl, icr = er32(ICR);
1193
1194         if (!icr)
1195                 return IRQ_NONE;  /* Not our interrupt */
1196
1197         /*
1198          * IMS will not auto-mask if INT_ASSERTED is not set, and if it is
1199          * not set, then the adapter didn't send an interrupt
1200          */
1201         if (!(icr & E1000_ICR_INT_ASSERTED))
1202                 return IRQ_NONE;
1203
1204         /*
1205          * Interrupt Auto-Mask...upon reading ICR,
1206          * interrupts are masked.  No need for the
1207          * IMC write
1208          */
1209
1210         if (icr & (E1000_ICR_RXSEQ | E1000_ICR_LSC)) {
1211                 hw->mac.get_link_status = 1;
1212                 /*
1213                  * ICH8 workaround-- Call gig speed drop workaround on cable
1214                  * disconnect (LSC) before accessing any PHY registers
1215                  */
1216                 if ((adapter->flags & FLAG_LSC_GIG_SPEED_DROP) &&
1217                     (!(er32(STATUS) & E1000_STATUS_LU)))
1218                         schedule_work(&adapter->downshift_task);
1219
1220                 /*
1221                  * 80003ES2LAN workaround--
1222                  * For packet buffer work-around on link down event;
1223                  * disable receives here in the ISR and
1224                  * reset adapter in watchdog
1225                  */
1226                 if (netif_carrier_ok(netdev) &&
1227                     (adapter->flags & FLAG_RX_NEEDS_RESTART)) {
1228                         /* disable receives */
1229                         rctl = er32(RCTL);
1230                         ew32(RCTL, rctl & ~E1000_RCTL_EN);
1231                         adapter->flags |= FLAG_RX_RESTART_NOW;
1232                 }
1233                 /* guard against interrupt when we're going down */
1234                 if (!test_bit(__E1000_DOWN, &adapter->state))
1235                         mod_timer(&adapter->watchdog_timer, jiffies + 1);
1236         }
1237
1238         if (netif_rx_schedule_prep(netdev, &adapter->napi)) {
1239                 adapter->total_tx_bytes = 0;
1240                 adapter->total_tx_packets = 0;
1241                 adapter->total_rx_bytes = 0;
1242                 adapter->total_rx_packets = 0;
1243                 __netif_rx_schedule(netdev, &adapter->napi);
1244         }
1245
1246         return IRQ_HANDLED;
1247 }
1248
1249 static irqreturn_t e1000_msix_other(int irq, void *data)
1250 {
1251         struct net_device *netdev = data;
1252         struct e1000_adapter *adapter = netdev_priv(netdev);
1253         struct e1000_hw *hw = &adapter->hw;
1254         u32 icr = er32(ICR);
1255
1256         if (!(icr & E1000_ICR_INT_ASSERTED)) {
1257                 ew32(IMS, E1000_IMS_OTHER);
1258                 return IRQ_NONE;
1259         }
1260
1261         if (icr & adapter->eiac_mask)
1262                 ew32(ICS, (icr & adapter->eiac_mask));
1263
1264         if (icr & E1000_ICR_OTHER) {
1265                 if (!(icr & E1000_ICR_LSC))
1266                         goto no_link_interrupt;
1267                 hw->mac.get_link_status = 1;
1268                 /* guard against interrupt when we're going down */
1269                 if (!test_bit(__E1000_DOWN, &adapter->state))
1270                         mod_timer(&adapter->watchdog_timer, jiffies + 1);
1271         }
1272
1273 no_link_interrupt:
1274         ew32(IMS, E1000_IMS_LSC | E1000_IMS_OTHER);
1275
1276         return IRQ_HANDLED;
1277 }
1278
1279
1280 static irqreturn_t e1000_intr_msix_tx(int irq, void *data)
1281 {
1282         struct net_device *netdev = data;
1283         struct e1000_adapter *adapter = netdev_priv(netdev);
1284         struct e1000_hw *hw = &adapter->hw;
1285         struct e1000_ring *tx_ring = adapter->tx_ring;
1286
1287
1288         adapter->total_tx_bytes = 0;
1289         adapter->total_tx_packets = 0;
1290
1291         if (!e1000_clean_tx_irq(adapter))
1292                 /* Ring was not completely cleaned, so fire another interrupt */
1293                 ew32(ICS, tx_ring->ims_val);
1294
1295         return IRQ_HANDLED;
1296 }
1297
1298 static irqreturn_t e1000_intr_msix_rx(int irq, void *data)
1299 {
1300         struct net_device *netdev = data;
1301         struct e1000_adapter *adapter = netdev_priv(netdev);
1302
1303         /* Write the ITR value calculated at the end of the
1304          * previous interrupt.
1305          */
1306         if (adapter->rx_ring->set_itr) {
1307                 writel(1000000000 / (adapter->rx_ring->itr_val * 256),
1308                        adapter->hw.hw_addr + adapter->rx_ring->itr_register);
1309                 adapter->rx_ring->set_itr = 0;
1310         }
1311
1312         if (netif_rx_schedule_prep(netdev, &adapter->napi)) {
1313                 adapter->total_rx_bytes = 0;
1314                 adapter->total_rx_packets = 0;
1315                 __netif_rx_schedule(netdev, &adapter->napi);
1316         }
1317         return IRQ_HANDLED;
1318 }
1319
1320 /**
1321  * e1000_configure_msix - Configure MSI-X hardware
1322  *
1323  * e1000_configure_msix sets up the hardware to properly
1324  * generate MSI-X interrupts.
1325  **/
1326 static void e1000_configure_msix(struct e1000_adapter *adapter)
1327 {
1328         struct e1000_hw *hw = &adapter->hw;
1329         struct e1000_ring *rx_ring = adapter->rx_ring;
1330         struct e1000_ring *tx_ring = adapter->tx_ring;
1331         int vector = 0;
1332         u32 ctrl_ext, ivar = 0;
1333
1334         adapter->eiac_mask = 0;
1335
1336         /* Workaround issue with spurious interrupts on 82574 in MSI-X mode */
1337         if (hw->mac.type == e1000_82574) {
1338                 u32 rfctl = er32(RFCTL);
1339                 rfctl |= E1000_RFCTL_ACK_DIS;
1340                 ew32(RFCTL, rfctl);
1341         }
1342
1343 #define E1000_IVAR_INT_ALLOC_VALID      0x8
1344         /* Configure Rx vector */
1345         rx_ring->ims_val = E1000_IMS_RXQ0;
1346         adapter->eiac_mask |= rx_ring->ims_val;
1347         if (rx_ring->itr_val)
1348                 writel(1000000000 / (rx_ring->itr_val * 256),
1349                        hw->hw_addr + rx_ring->itr_register);
1350         else
1351                 writel(1, hw->hw_addr + rx_ring->itr_register);
1352         ivar = E1000_IVAR_INT_ALLOC_VALID | vector;
1353
1354         /* Configure Tx vector */
1355         tx_ring->ims_val = E1000_IMS_TXQ0;
1356         vector++;
1357         if (tx_ring->itr_val)
1358                 writel(1000000000 / (tx_ring->itr_val * 256),
1359                        hw->hw_addr + tx_ring->itr_register);
1360         else
1361                 writel(1, hw->hw_addr + tx_ring->itr_register);
1362         adapter->eiac_mask |= tx_ring->ims_val;
1363         ivar |= ((E1000_IVAR_INT_ALLOC_VALID | vector) << 8);
1364
1365         /* set vector for Other Causes, e.g. link changes */
1366         vector++;
1367         ivar |= ((E1000_IVAR_INT_ALLOC_VALID | vector) << 16);
1368         if (rx_ring->itr_val)
1369                 writel(1000000000 / (rx_ring->itr_val * 256),
1370                        hw->hw_addr + E1000_EITR_82574(vector));
1371         else
1372                 writel(1, hw->hw_addr + E1000_EITR_82574(vector));
1373
1374         /* Cause Tx interrupts on every write back */
1375         ivar |= (1 << 31);
1376
1377         ew32(IVAR, ivar);
1378
1379         /* enable MSI-X PBA support */
1380         ctrl_ext = er32(CTRL_EXT);
1381         ctrl_ext |= E1000_CTRL_EXT_PBA_CLR;
1382
1383         /* Auto-Mask Other interrupts upon ICR read */
1384 #define E1000_EIAC_MASK_82574   0x01F00000
1385         ew32(IAM, ~E1000_EIAC_MASK_82574 | E1000_IMS_OTHER);
1386         ctrl_ext |= E1000_CTRL_EXT_EIAME;
1387         ew32(CTRL_EXT, ctrl_ext);
1388         e1e_flush();
1389 }
1390
1391 void e1000e_reset_interrupt_capability(struct e1000_adapter *adapter)
1392 {
1393         if (adapter->msix_entries) {
1394                 pci_disable_msix(adapter->pdev);
1395                 kfree(adapter->msix_entries);
1396                 adapter->msix_entries = NULL;
1397         } else if (adapter->flags & FLAG_MSI_ENABLED) {
1398                 pci_disable_msi(adapter->pdev);
1399                 adapter->flags &= ~FLAG_MSI_ENABLED;
1400         }
1401
1402         return;
1403 }
1404
1405 /**
1406  * e1000e_set_interrupt_capability - set MSI or MSI-X if supported
1407  *
1408  * Attempt to configure interrupts using the best available
1409  * capabilities of the hardware and kernel.
1410  **/
1411 void e1000e_set_interrupt_capability(struct e1000_adapter *adapter)
1412 {
1413         int err;
1414         int numvecs, i;
1415
1416
1417         switch (adapter->int_mode) {
1418         case E1000E_INT_MODE_MSIX:
1419                 if (adapter->flags & FLAG_HAS_MSIX) {
1420                         numvecs = 3; /* RxQ0, TxQ0 and other */
1421                         adapter->msix_entries = kcalloc(numvecs,
1422                                                       sizeof(struct msix_entry),
1423                                                       GFP_KERNEL);
1424                         if (adapter->msix_entries) {
1425                                 for (i = 0; i < numvecs; i++)
1426                                         adapter->msix_entries[i].entry = i;
1427
1428                                 err = pci_enable_msix(adapter->pdev,
1429                                                       adapter->msix_entries,
1430                                                       numvecs);
1431                                 if (err == 0)
1432                                         return;
1433                         }
1434                         /* MSI-X failed, so fall through and try MSI */
1435                         e_err("Failed to initialize MSI-X interrupts.  "
1436                               "Falling back to MSI interrupts.\n");
1437                         e1000e_reset_interrupt_capability(adapter);
1438                 }
1439                 adapter->int_mode = E1000E_INT_MODE_MSI;
1440                 /* Fall through */
1441         case E1000E_INT_MODE_MSI:
1442                 if (!pci_enable_msi(adapter->pdev)) {
1443                         adapter->flags |= FLAG_MSI_ENABLED;
1444                 } else {
1445                         adapter->int_mode = E1000E_INT_MODE_LEGACY;
1446                         e_err("Failed to initialize MSI interrupts.  Falling "
1447                               "back to legacy interrupts.\n");
1448                 }
1449                 /* Fall through */
1450         case E1000E_INT_MODE_LEGACY:
1451                 /* Don't do anything; this is the system default */
1452                 break;
1453         }
1454
1455         return;
1456 }
1457
1458 /**
1459  * e1000_request_msix - Initialize MSI-X interrupts
1460  *
1461  * e1000_request_msix allocates MSI-X vectors and requests interrupts from the
1462  * kernel.
1463  **/
1464 static int e1000_request_msix(struct e1000_adapter *adapter)
1465 {
1466         struct net_device *netdev = adapter->netdev;
1467         int err = 0, vector = 0;
1468
1469         if (strlen(netdev->name) < (IFNAMSIZ - 5))
1470                 sprintf(adapter->rx_ring->name, "%s-rx0", netdev->name);
1471         else
1472                 memcpy(adapter->rx_ring->name, netdev->name, IFNAMSIZ);
1473         err = request_irq(adapter->msix_entries[vector].vector,
1474                           &e1000_intr_msix_rx, 0, adapter->rx_ring->name,
1475                           netdev);
1476         if (err)
1477                 goto out;
1478         adapter->rx_ring->itr_register = E1000_EITR_82574(vector);
1479         adapter->rx_ring->itr_val = adapter->itr;
1480         vector++;
1481
1482         if (strlen(netdev->name) < (IFNAMSIZ - 5))
1483                 sprintf(adapter->tx_ring->name, "%s-tx0", netdev->name);
1484         else
1485                 memcpy(adapter->tx_ring->name, netdev->name, IFNAMSIZ);
1486         err = request_irq(adapter->msix_entries[vector].vector,
1487                           &e1000_intr_msix_tx, 0, adapter->tx_ring->name,
1488                           netdev);
1489         if (err)
1490                 goto out;
1491         adapter->tx_ring->itr_register = E1000_EITR_82574(vector);
1492         adapter->tx_ring->itr_val = adapter->itr;
1493         vector++;
1494
1495         err = request_irq(adapter->msix_entries[vector].vector,
1496                           &e1000_msix_other, 0, netdev->name, netdev);
1497         if (err)
1498                 goto out;
1499
1500         e1000_configure_msix(adapter);
1501         return 0;
1502 out:
1503         return err;
1504 }
1505
1506 /**
1507  * e1000_request_irq - initialize interrupts
1508  *
1509  * Attempts to configure interrupts using the best available
1510  * capabilities of the hardware and kernel.
1511  **/
1512 static int e1000_request_irq(struct e1000_adapter *adapter)
1513 {
1514         struct net_device *netdev = adapter->netdev;
1515         int err;
1516
1517         if (adapter->msix_entries) {
1518                 err = e1000_request_msix(adapter);
1519                 if (!err)
1520                         return err;
1521                 /* fall back to MSI */
1522                 e1000e_reset_interrupt_capability(adapter);
1523                 adapter->int_mode = E1000E_INT_MODE_MSI;
1524                 e1000e_set_interrupt_capability(adapter);
1525         }
1526         if (adapter->flags & FLAG_MSI_ENABLED) {
1527                 err = request_irq(adapter->pdev->irq, &e1000_intr_msi, 0,
1528                                   netdev->name, netdev);
1529                 if (!err)
1530                         return err;
1531
1532                 /* fall back to legacy interrupt */
1533                 e1000e_reset_interrupt_capability(adapter);
1534                 adapter->int_mode = E1000E_INT_MODE_LEGACY;
1535         }
1536
1537         err = request_irq(adapter->pdev->irq, &e1000_intr, IRQF_SHARED,
1538                           netdev->name, netdev);
1539         if (err)
1540                 e_err("Unable to allocate interrupt, Error: %d\n", err);
1541
1542         return err;
1543 }
1544
1545 static void e1000_free_irq(struct e1000_adapter *adapter)
1546 {
1547         struct net_device *netdev = adapter->netdev;
1548
1549         if (adapter->msix_entries) {
1550                 int vector = 0;
1551
1552                 free_irq(adapter->msix_entries[vector].vector, netdev);
1553                 vector++;
1554
1555                 free_irq(adapter->msix_entries[vector].vector, netdev);
1556                 vector++;
1557
1558                 /* Other Causes interrupt vector */
1559                 free_irq(adapter->msix_entries[vector].vector, netdev);
1560                 return;
1561         }
1562
1563         free_irq(adapter->pdev->irq, netdev);
1564 }
1565
1566 /**
1567  * e1000_irq_disable - Mask off interrupt generation on the NIC
1568  **/
1569 static void e1000_irq_disable(struct e1000_adapter *adapter)
1570 {
1571         struct e1000_hw *hw = &adapter->hw;
1572
1573         ew32(IMC, ~0);
1574         if (adapter->msix_entries)
1575                 ew32(EIAC_82574, 0);
1576         e1e_flush();
1577         synchronize_irq(adapter->pdev->irq);
1578 }
1579
1580 /**
1581  * e1000_irq_enable - Enable default interrupt generation settings
1582  **/
1583 static void e1000_irq_enable(struct e1000_adapter *adapter)
1584 {
1585         struct e1000_hw *hw = &adapter->hw;
1586
1587         if (adapter->msix_entries) {
1588                 ew32(EIAC_82574, adapter->eiac_mask & E1000_EIAC_MASK_82574);
1589                 ew32(IMS, adapter->eiac_mask | E1000_IMS_OTHER | E1000_IMS_LSC);
1590         } else {
1591                 ew32(IMS, IMS_ENABLE_MASK);
1592         }
1593         e1e_flush();
1594 }
1595
1596 /**
1597  * e1000_get_hw_control - get control of the h/w from f/w
1598  * @adapter: address of board private structure
1599  *
1600  * e1000_get_hw_control sets {CTRL_EXT|SWSM}:DRV_LOAD bit.
1601  * For ASF and Pass Through versions of f/w this means that
1602  * the driver is loaded. For AMT version (only with 82573)
1603  * of the f/w this means that the network i/f is open.
1604  **/
1605 static void e1000_get_hw_control(struct e1000_adapter *adapter)
1606 {
1607         struct e1000_hw *hw = &adapter->hw;
1608         u32 ctrl_ext;
1609         u32 swsm;
1610
1611         /* Let firmware know the driver has taken over */
1612         if (adapter->flags & FLAG_HAS_SWSM_ON_LOAD) {
1613                 swsm = er32(SWSM);
1614                 ew32(SWSM, swsm | E1000_SWSM_DRV_LOAD);
1615         } else if (adapter->flags & FLAG_HAS_CTRLEXT_ON_LOAD) {
1616                 ctrl_ext = er32(CTRL_EXT);
1617                 ew32(CTRL_EXT, ctrl_ext | E1000_CTRL_EXT_DRV_LOAD);
1618         }
1619 }
1620
1621 /**
1622  * e1000_release_hw_control - release control of the h/w to f/w
1623  * @adapter: address of board private structure
1624  *
1625  * e1000_release_hw_control resets {CTRL_EXT|SWSM}:DRV_LOAD bit.
1626  * For ASF and Pass Through versions of f/w this means that the
1627  * driver is no longer loaded. For AMT version (only with 82573) i
1628  * of the f/w this means that the network i/f is closed.
1629  *
1630  **/
1631 static void e1000_release_hw_control(struct e1000_adapter *adapter)
1632 {
1633         struct e1000_hw *hw = &adapter->hw;
1634         u32 ctrl_ext;
1635         u32 swsm;
1636
1637         /* Let firmware taken over control of h/w */
1638         if (adapter->flags & FLAG_HAS_SWSM_ON_LOAD) {
1639                 swsm = er32(SWSM);
1640                 ew32(SWSM, swsm & ~E1000_SWSM_DRV_LOAD);
1641         } else if (adapter->flags & FLAG_HAS_CTRLEXT_ON_LOAD) {
1642                 ctrl_ext = er32(CTRL_EXT);
1643                 ew32(CTRL_EXT, ctrl_ext & ~E1000_CTRL_EXT_DRV_LOAD);
1644         }
1645 }
1646
1647 /**
1648  * @e1000_alloc_ring - allocate memory for a ring structure
1649  **/
1650 static int e1000_alloc_ring_dma(struct e1000_adapter *adapter,
1651                                 struct e1000_ring *ring)
1652 {
1653         struct pci_dev *pdev = adapter->pdev;
1654
1655         ring->desc = dma_alloc_coherent(&pdev->dev, ring->size, &ring->dma,
1656                                         GFP_KERNEL);
1657         if (!ring->desc)
1658                 return -ENOMEM;
1659
1660         return 0;
1661 }
1662
1663 /**
1664  * e1000e_setup_tx_resources - allocate Tx resources (Descriptors)
1665  * @adapter: board private structure
1666  *
1667  * Return 0 on success, negative on failure
1668  **/
1669 int e1000e_setup_tx_resources(struct e1000_adapter *adapter)
1670 {
1671         struct e1000_ring *tx_ring = adapter->tx_ring;
1672         int err = -ENOMEM, size;
1673
1674         size = sizeof(struct e1000_buffer) * tx_ring->count;
1675         tx_ring->buffer_info = vmalloc(size);
1676         if (!tx_ring->buffer_info)
1677                 goto err;
1678         memset(tx_ring->buffer_info, 0, size);
1679
1680         /* round up to nearest 4K */
1681         tx_ring->size = tx_ring->count * sizeof(struct e1000_tx_desc);
1682         tx_ring->size = ALIGN(tx_ring->size, 4096);
1683
1684         err = e1000_alloc_ring_dma(adapter, tx_ring);
1685         if (err)
1686                 goto err;
1687
1688         tx_ring->next_to_use = 0;
1689         tx_ring->next_to_clean = 0;
1690         spin_lock_init(&adapter->tx_queue_lock);
1691
1692         return 0;
1693 err:
1694         vfree(tx_ring->buffer_info);
1695         e_err("Unable to allocate memory for the transmit descriptor ring\n");
1696         return err;
1697 }
1698
1699 /**
1700  * e1000e_setup_rx_resources - allocate Rx resources (Descriptors)
1701  * @adapter: board private structure
1702  *
1703  * Returns 0 on success, negative on failure
1704  **/
1705 int e1000e_setup_rx_resources(struct e1000_adapter *adapter)
1706 {
1707         struct e1000_ring *rx_ring = adapter->rx_ring;
1708         struct e1000_buffer *buffer_info;
1709         int i, size, desc_len, err = -ENOMEM;
1710
1711         size = sizeof(struct e1000_buffer) * rx_ring->count;
1712         rx_ring->buffer_info = vmalloc(size);
1713         if (!rx_ring->buffer_info)
1714                 goto err;
1715         memset(rx_ring->buffer_info, 0, size);
1716
1717         for (i = 0; i < rx_ring->count; i++) {
1718                 buffer_info = &rx_ring->buffer_info[i];
1719                 buffer_info->ps_pages = kcalloc(PS_PAGE_BUFFERS,
1720                                                 sizeof(struct e1000_ps_page),
1721                                                 GFP_KERNEL);
1722                 if (!buffer_info->ps_pages)
1723                         goto err_pages;
1724         }
1725
1726         desc_len = sizeof(union e1000_rx_desc_packet_split);
1727
1728         /* Round up to nearest 4K */
1729         rx_ring->size = rx_ring->count * desc_len;
1730         rx_ring->size = ALIGN(rx_ring->size, 4096);
1731
1732         err = e1000_alloc_ring_dma(adapter, rx_ring);
1733         if (err)
1734                 goto err_pages;
1735
1736         rx_ring->next_to_clean = 0;
1737         rx_ring->next_to_use = 0;
1738         rx_ring->rx_skb_top = NULL;
1739
1740         return 0;
1741
1742 err_pages:
1743         for (i = 0; i < rx_ring->count; i++) {
1744                 buffer_info = &rx_ring->buffer_info[i];
1745                 kfree(buffer_info->ps_pages);
1746         }
1747 err:
1748         vfree(rx_ring->buffer_info);
1749         e_err("Unable to allocate memory for the transmit descriptor ring\n");
1750         return err;
1751 }
1752
1753 /**
1754  * e1000_clean_tx_ring - Free Tx Buffers
1755  * @adapter: board private structure
1756  **/
1757 static void e1000_clean_tx_ring(struct e1000_adapter *adapter)
1758 {
1759         struct e1000_ring *tx_ring = adapter->tx_ring;
1760         struct e1000_buffer *buffer_info;
1761         unsigned long size;
1762         unsigned int i;
1763
1764         for (i = 0; i < tx_ring->count; i++) {
1765                 buffer_info = &tx_ring->buffer_info[i];
1766                 e1000_put_txbuf(adapter, buffer_info);
1767         }
1768
1769         size = sizeof(struct e1000_buffer) * tx_ring->count;
1770         memset(tx_ring->buffer_info, 0, size);
1771
1772         memset(tx_ring->desc, 0, tx_ring->size);
1773
1774         tx_ring->next_to_use = 0;
1775         tx_ring->next_to_clean = 0;
1776
1777         writel(0, adapter->hw.hw_addr + tx_ring->head);
1778         writel(0, adapter->hw.hw_addr + tx_ring->tail);
1779 }
1780
1781 /**
1782  * e1000e_free_tx_resources - Free Tx Resources per Queue
1783  * @adapter: board private structure
1784  *
1785  * Free all transmit software resources
1786  **/
1787 void e1000e_free_tx_resources(struct e1000_adapter *adapter)
1788 {
1789         struct pci_dev *pdev = adapter->pdev;
1790         struct e1000_ring *tx_ring = adapter->tx_ring;
1791
1792         e1000_clean_tx_ring(adapter);
1793
1794         vfree(tx_ring->buffer_info);
1795         tx_ring->buffer_info = NULL;
1796
1797         dma_free_coherent(&pdev->dev, tx_ring->size, tx_ring->desc,
1798                           tx_ring->dma);
1799         tx_ring->desc = NULL;
1800 }
1801
1802 /**
1803  * e1000e_free_rx_resources - Free Rx Resources
1804  * @adapter: board private structure
1805  *
1806  * Free all receive software resources
1807  **/
1808
1809 void e1000e_free_rx_resources(struct e1000_adapter *adapter)
1810 {
1811         struct pci_dev *pdev = adapter->pdev;
1812         struct e1000_ring *rx_ring = adapter->rx_ring;
1813         int i;
1814
1815         e1000_clean_rx_ring(adapter);
1816
1817         for (i = 0; i < rx_ring->count; i++) {
1818                 kfree(rx_ring->buffer_info[i].ps_pages);
1819         }
1820
1821         vfree(rx_ring->buffer_info);
1822         rx_ring->buffer_info = NULL;
1823
1824         dma_free_coherent(&pdev->dev, rx_ring->size, rx_ring->desc,
1825                           rx_ring->dma);
1826         rx_ring->desc = NULL;
1827 }
1828
1829 /**
1830  * e1000_update_itr - update the dynamic ITR value based on statistics
1831  * @adapter: pointer to adapter
1832  * @itr_setting: current adapter->itr
1833  * @packets: the number of packets during this measurement interval
1834  * @bytes: the number of bytes during this measurement interval
1835  *
1836  *      Stores a new ITR value based on packets and byte
1837  *      counts during the last interrupt.  The advantage of per interrupt
1838  *      computation is faster updates and more accurate ITR for the current
1839  *      traffic pattern.  Constants in this function were computed
1840  *      based on theoretical maximum wire speed and thresholds were set based
1841  *      on testing data as well as attempting to minimize response time
1842  *      while increasing bulk throughput.  This functionality is controlled
1843  *      by the InterruptThrottleRate module parameter.
1844  **/
1845 static unsigned int e1000_update_itr(struct e1000_adapter *adapter,
1846                                      u16 itr_setting, int packets,
1847                                      int bytes)
1848 {
1849         unsigned int retval = itr_setting;
1850
1851         if (packets == 0)
1852                 goto update_itr_done;
1853
1854         switch (itr_setting) {
1855         case lowest_latency:
1856                 /* handle TSO and jumbo frames */
1857                 if (bytes/packets > 8000)
1858                         retval = bulk_latency;
1859                 else if ((packets < 5) && (bytes > 512)) {
1860                         retval = low_latency;
1861                 }
1862                 break;
1863         case low_latency:  /* 50 usec aka 20000 ints/s */
1864                 if (bytes > 10000) {
1865                         /* this if handles the TSO accounting */
1866                         if (bytes/packets > 8000) {
1867                                 retval = bulk_latency;
1868                         } else if ((packets < 10) || ((bytes/packets) > 1200)) {
1869                                 retval = bulk_latency;
1870                         } else if ((packets > 35)) {
1871                                 retval = lowest_latency;
1872                         }
1873                 } else if (bytes/packets > 2000) {
1874                         retval = bulk_latency;
1875                 } else if (packets <= 2 && bytes < 512) {
1876                         retval = lowest_latency;
1877                 }
1878                 break;
1879         case bulk_latency: /* 250 usec aka 4000 ints/s */
1880                 if (bytes > 25000) {
1881                         if (packets > 35) {
1882                                 retval = low_latency;
1883                         }
1884                 } else if (bytes < 6000) {
1885                         retval = low_latency;
1886                 }
1887                 break;
1888         }
1889
1890 update_itr_done:
1891         return retval;
1892 }
1893
1894 static void e1000_set_itr(struct e1000_adapter *adapter)
1895 {
1896         struct e1000_hw *hw = &adapter->hw;
1897         u16 current_itr;
1898         u32 new_itr = adapter->itr;
1899
1900         /* for non-gigabit speeds, just fix the interrupt rate at 4000 */
1901         if (adapter->link_speed != SPEED_1000) {
1902                 current_itr = 0;
1903                 new_itr = 4000;
1904                 goto set_itr_now;
1905         }
1906
1907         adapter->tx_itr = e1000_update_itr(adapter,
1908                                     adapter->tx_itr,
1909                                     adapter->total_tx_packets,
1910                                     adapter->total_tx_bytes);
1911         /* conservative mode (itr 3) eliminates the lowest_latency setting */
1912         if (adapter->itr_setting == 3 && adapter->tx_itr == lowest_latency)
1913                 adapter->tx_itr = low_latency;
1914
1915         adapter->rx_itr = e1000_update_itr(adapter,
1916                                     adapter->rx_itr,
1917                                     adapter->total_rx_packets,
1918                                     adapter->total_rx_bytes);
1919         /* conservative mode (itr 3) eliminates the lowest_latency setting */
1920         if (adapter->itr_setting == 3 && adapter->rx_itr == lowest_latency)
1921                 adapter->rx_itr = low_latency;
1922
1923         current_itr = max(adapter->rx_itr, adapter->tx_itr);
1924
1925         switch (current_itr) {
1926         /* counts and packets in update_itr are dependent on these numbers */
1927         case lowest_latency:
1928                 new_itr = 70000;
1929                 break;
1930         case low_latency:
1931                 new_itr = 20000; /* aka hwitr = ~200 */
1932                 break;
1933         case bulk_latency:
1934                 new_itr = 4000;
1935                 break;
1936         default:
1937                 break;
1938         }
1939
1940 set_itr_now:
1941         if (new_itr != adapter->itr) {
1942                 /*
1943                  * this attempts to bias the interrupt rate towards Bulk
1944                  * by adding intermediate steps when interrupt rate is
1945                  * increasing
1946                  */
1947                 new_itr = new_itr > adapter->itr ?
1948                              min(adapter->itr + (new_itr >> 2), new_itr) :
1949                              new_itr;
1950                 adapter->itr = new_itr;
1951                 adapter->rx_ring->itr_val = new_itr;
1952                 if (adapter->msix_entries)
1953                         adapter->rx_ring->set_itr = 1;
1954                 else
1955                         ew32(ITR, 1000000000 / (new_itr * 256));
1956         }
1957 }
1958
1959 /**
1960  * e1000_alloc_queues - Allocate memory for all rings
1961  * @adapter: board private structure to initialize
1962  **/
1963 static int __devinit e1000_alloc_queues(struct e1000_adapter *adapter)
1964 {
1965         adapter->tx_ring = kzalloc(sizeof(struct e1000_ring), GFP_KERNEL);
1966         if (!adapter->tx_ring)
1967                 goto err;
1968
1969         adapter->rx_ring = kzalloc(sizeof(struct e1000_ring), GFP_KERNEL);
1970         if (!adapter->rx_ring)
1971                 goto err;
1972
1973         return 0;
1974 err:
1975         e_err("Unable to allocate memory for queues\n");
1976         kfree(adapter->rx_ring);
1977         kfree(adapter->tx_ring);
1978         return -ENOMEM;
1979 }
1980
1981 /**
1982  * e1000_clean - NAPI Rx polling callback
1983  * @napi: struct associated with this polling callback
1984  * @budget: amount of packets driver is allowed to process this poll
1985  **/
1986 static int e1000_clean(struct napi_struct *napi, int budget)
1987 {
1988         struct e1000_adapter *adapter = container_of(napi, struct e1000_adapter, napi);
1989         struct e1000_hw *hw = &adapter->hw;
1990         struct net_device *poll_dev = adapter->netdev;
1991         int tx_cleaned = 0, work_done = 0;
1992
1993         /* Must NOT use netdev_priv macro here. */
1994         adapter = poll_dev->priv;
1995
1996         if (adapter->msix_entries &&
1997             !(adapter->rx_ring->ims_val & adapter->tx_ring->ims_val))
1998                 goto clean_rx;
1999
2000         /*
2001          * e1000_clean is called per-cpu.  This lock protects
2002          * tx_ring from being cleaned by multiple cpus
2003          * simultaneously.  A failure obtaining the lock means
2004          * tx_ring is currently being cleaned anyway.
2005          */
2006         if (spin_trylock(&adapter->tx_queue_lock)) {
2007                 tx_cleaned = e1000_clean_tx_irq(adapter);
2008                 spin_unlock(&adapter->tx_queue_lock);
2009         }
2010
2011 clean_rx:
2012         adapter->clean_rx(adapter, &work_done, budget);
2013
2014         if (tx_cleaned)
2015                 work_done = budget;
2016
2017         /* If budget not fully consumed, exit the polling mode */
2018         if (work_done < budget) {
2019                 if (adapter->itr_setting & 3)
2020                         e1000_set_itr(adapter);
2021                 netif_rx_complete(poll_dev, napi);
2022                 if (adapter->msix_entries)
2023                         ew32(IMS, adapter->rx_ring->ims_val);
2024                 else
2025                         e1000_irq_enable(adapter);
2026         }
2027
2028         return work_done;
2029 }
2030
2031 static void e1000_vlan_rx_add_vid(struct net_device *netdev, u16 vid)
2032 {
2033         struct e1000_adapter *adapter = netdev_priv(netdev);
2034         struct e1000_hw *hw = &adapter->hw;
2035         u32 vfta, index;
2036
2037         /* don't update vlan cookie if already programmed */
2038         if ((adapter->hw.mng_cookie.status &
2039              E1000_MNG_DHCP_COOKIE_STATUS_VLAN) &&
2040             (vid == adapter->mng_vlan_id))
2041                 return;
2042         /* add VID to filter table */
2043         index = (vid >> 5) & 0x7F;
2044         vfta = E1000_READ_REG_ARRAY(hw, E1000_VFTA, index);
2045         vfta |= (1 << (vid & 0x1F));
2046         e1000e_write_vfta(hw, index, vfta);
2047 }
2048
2049 static void e1000_vlan_rx_kill_vid(struct net_device *netdev, u16 vid)
2050 {
2051         struct e1000_adapter *adapter = netdev_priv(netdev);
2052         struct e1000_hw *hw = &adapter->hw;
2053         u32 vfta, index;
2054
2055         if (!test_bit(__E1000_DOWN, &adapter->state))
2056                 e1000_irq_disable(adapter);
2057         vlan_group_set_device(adapter->vlgrp, vid, NULL);
2058
2059         if (!test_bit(__E1000_DOWN, &adapter->state))
2060                 e1000_irq_enable(adapter);
2061
2062         if ((adapter->hw.mng_cookie.status &
2063              E1000_MNG_DHCP_COOKIE_STATUS_VLAN) &&
2064             (vid == adapter->mng_vlan_id)) {
2065                 /* release control to f/w */
2066                 e1000_release_hw_control(adapter);
2067                 return;
2068         }
2069
2070         /* remove VID from filter table */
2071         index = (vid >> 5) & 0x7F;
2072         vfta = E1000_READ_REG_ARRAY(hw, E1000_VFTA, index);
2073         vfta &= ~(1 << (vid & 0x1F));
2074         e1000e_write_vfta(hw, index, vfta);
2075 }
2076
2077 static void e1000_update_mng_vlan(struct e1000_adapter *adapter)
2078 {
2079         struct net_device *netdev = adapter->netdev;
2080         u16 vid = adapter->hw.mng_cookie.vlan_id;
2081         u16 old_vid = adapter->mng_vlan_id;
2082
2083         if (!adapter->vlgrp)
2084                 return;
2085
2086         if (!vlan_group_get_device(adapter->vlgrp, vid)) {
2087                 adapter->mng_vlan_id = E1000_MNG_VLAN_NONE;
2088                 if (adapter->hw.mng_cookie.status &
2089                         E1000_MNG_DHCP_COOKIE_STATUS_VLAN) {
2090                         e1000_vlan_rx_add_vid(netdev, vid);
2091                         adapter->mng_vlan_id = vid;
2092                 }
2093
2094                 if ((old_vid != (u16)E1000_MNG_VLAN_NONE) &&
2095                                 (vid != old_vid) &&
2096                     !vlan_group_get_device(adapter->vlgrp, old_vid))
2097                         e1000_vlan_rx_kill_vid(netdev, old_vid);
2098         } else {
2099                 adapter->mng_vlan_id = vid;
2100         }
2101 }
2102
2103
2104 static void e1000_vlan_rx_register(struct net_device *netdev,
2105                                    struct vlan_group *grp)
2106 {
2107         struct e1000_adapter *adapter = netdev_priv(netdev);
2108         struct e1000_hw *hw = &adapter->hw;
2109         u32 ctrl, rctl;
2110
2111         if (!test_bit(__E1000_DOWN, &adapter->state))
2112                 e1000_irq_disable(adapter);
2113         adapter->vlgrp = grp;
2114
2115         if (grp) {
2116                 /* enable VLAN tag insert/strip */
2117                 ctrl = er32(CTRL);
2118                 ctrl |= E1000_CTRL_VME;
2119                 ew32(CTRL, ctrl);
2120
2121                 if (adapter->flags & FLAG_HAS_HW_VLAN_FILTER) {
2122                         /* enable VLAN receive filtering */
2123                         rctl = er32(RCTL);
2124                         rctl &= ~E1000_RCTL_CFIEN;
2125                         ew32(RCTL, rctl);
2126                         e1000_update_mng_vlan(adapter);
2127                 }
2128         } else {
2129                 /* disable 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                         if (adapter->mng_vlan_id !=
2136                             (u16)E1000_MNG_VLAN_NONE) {
2137                                 e1000_vlan_rx_kill_vid(netdev,
2138                                                        adapter->mng_vlan_id);
2139                                 adapter->mng_vlan_id = E1000_MNG_VLAN_NONE;
2140                         }
2141                 }
2142         }
2143
2144         if (!test_bit(__E1000_DOWN, &adapter->state))
2145                 e1000_irq_enable(adapter);
2146 }
2147
2148 static void e1000_restore_vlan(struct e1000_adapter *adapter)
2149 {
2150         u16 vid;
2151
2152         e1000_vlan_rx_register(adapter->netdev, adapter->vlgrp);
2153
2154         if (!adapter->vlgrp)
2155                 return;
2156
2157         for (vid = 0; vid < VLAN_GROUP_ARRAY_LEN; vid++) {
2158                 if (!vlan_group_get_device(adapter->vlgrp, vid))
2159                         continue;
2160                 e1000_vlan_rx_add_vid(adapter->netdev, vid);
2161         }
2162 }
2163
2164 static void e1000_init_manageability(struct e1000_adapter *adapter)
2165 {
2166         struct e1000_hw *hw = &adapter->hw;
2167         u32 manc, manc2h;
2168
2169         if (!(adapter->flags & FLAG_MNG_PT_ENABLED))
2170                 return;
2171
2172         manc = er32(MANC);
2173
2174         /*
2175          * enable receiving management packets to the host. this will probably
2176          * generate destination unreachable messages from the host OS, but
2177          * the packets will be handled on SMBUS
2178          */
2179         manc |= E1000_MANC_EN_MNG2HOST;
2180         manc2h = er32(MANC2H);
2181 #define E1000_MNG2HOST_PORT_623 (1 << 5)
2182 #define E1000_MNG2HOST_PORT_664 (1 << 6)
2183         manc2h |= E1000_MNG2HOST_PORT_623;
2184         manc2h |= E1000_MNG2HOST_PORT_664;
2185         ew32(MANC2H, manc2h);
2186         ew32(MANC, manc);
2187 }
2188
2189 /**
2190  * e1000_configure_tx - Configure 8254x Transmit Unit after Reset
2191  * @adapter: board private structure
2192  *
2193  * Configure the Tx unit of the MAC after a reset.
2194  **/
2195 static void e1000_configure_tx(struct e1000_adapter *adapter)
2196 {
2197         struct e1000_hw *hw = &adapter->hw;
2198         struct e1000_ring *tx_ring = adapter->tx_ring;
2199         u64 tdba;
2200         u32 tdlen, tctl, tipg, tarc;
2201         u32 ipgr1, ipgr2;
2202
2203         /* Setup the HW Tx Head and Tail descriptor pointers */
2204         tdba = tx_ring->dma;
2205         tdlen = tx_ring->count * sizeof(struct e1000_tx_desc);
2206         ew32(TDBAL, (tdba & DMA_32BIT_MASK));
2207         ew32(TDBAH, (tdba >> 32));
2208         ew32(TDLEN, tdlen);
2209         ew32(TDH, 0);
2210         ew32(TDT, 0);
2211         tx_ring->head = E1000_TDH;
2212         tx_ring->tail = E1000_TDT;
2213
2214         /* Set the default values for the Tx Inter Packet Gap timer */
2215         tipg = DEFAULT_82543_TIPG_IPGT_COPPER;          /*  8  */
2216         ipgr1 = DEFAULT_82543_TIPG_IPGR1;               /*  8  */
2217         ipgr2 = DEFAULT_82543_TIPG_IPGR2;               /*  6  */
2218
2219         if (adapter->flags & FLAG_TIPG_MEDIUM_FOR_80003ESLAN)
2220                 ipgr2 = DEFAULT_80003ES2LAN_TIPG_IPGR2; /*  7  */
2221
2222         tipg |= ipgr1 << E1000_TIPG_IPGR1_SHIFT;
2223         tipg |= ipgr2 << E1000_TIPG_IPGR2_SHIFT;
2224         ew32(TIPG, tipg);
2225
2226         /* Set the Tx Interrupt Delay register */
2227         ew32(TIDV, adapter->tx_int_delay);
2228         /* Tx irq moderation */
2229         ew32(TADV, adapter->tx_abs_int_delay);
2230
2231         /* Program the Transmit Control Register */
2232         tctl = er32(TCTL);
2233         tctl &= ~E1000_TCTL_CT;
2234         tctl |= E1000_TCTL_PSP | E1000_TCTL_RTLC |
2235                 (E1000_COLLISION_THRESHOLD << E1000_CT_SHIFT);
2236
2237         if (adapter->flags & FLAG_TARC_SPEED_MODE_BIT) {
2238                 tarc = er32(TARC(0));
2239                 /*
2240                  * set the speed mode bit, we'll clear it if we're not at
2241                  * gigabit link later
2242                  */
2243 #define SPEED_MODE_BIT (1 << 21)
2244                 tarc |= SPEED_MODE_BIT;
2245                 ew32(TARC(0), tarc);
2246         }
2247
2248         /* errata: program both queues to unweighted RR */
2249         if (adapter->flags & FLAG_TARC_SET_BIT_ZERO) {
2250                 tarc = er32(TARC(0));
2251                 tarc |= 1;
2252                 ew32(TARC(0), tarc);
2253                 tarc = er32(TARC(1));
2254                 tarc |= 1;
2255                 ew32(TARC(1), tarc);
2256         }
2257
2258         e1000e_config_collision_dist(hw);
2259
2260         /* Setup Transmit Descriptor Settings for eop descriptor */
2261         adapter->txd_cmd = E1000_TXD_CMD_EOP | E1000_TXD_CMD_IFCS;
2262
2263         /* only set IDE if we are delaying interrupts using the timers */
2264         if (adapter->tx_int_delay)
2265                 adapter->txd_cmd |= E1000_TXD_CMD_IDE;
2266
2267         /* enable Report Status bit */
2268         adapter->txd_cmd |= E1000_TXD_CMD_RS;
2269
2270         ew32(TCTL, tctl);
2271
2272         adapter->tx_queue_len = adapter->netdev->tx_queue_len;
2273 }
2274
2275 /**
2276  * e1000_setup_rctl - configure the receive control registers
2277  * @adapter: Board private structure
2278  **/
2279 #define PAGE_USE_COUNT(S) (((S) >> PAGE_SHIFT) + \
2280                            (((S) & (PAGE_SIZE - 1)) ? 1 : 0))
2281 static void e1000_setup_rctl(struct e1000_adapter *adapter)
2282 {
2283         struct e1000_hw *hw = &adapter->hw;
2284         u32 rctl, rfctl;
2285         u32 psrctl = 0;
2286         u32 pages = 0;
2287
2288         /* Program MC offset vector base */
2289         rctl = er32(RCTL);
2290         rctl &= ~(3 << E1000_RCTL_MO_SHIFT);
2291         rctl |= E1000_RCTL_EN | E1000_RCTL_BAM |
2292                 E1000_RCTL_LBM_NO | E1000_RCTL_RDMTS_HALF |
2293                 (adapter->hw.mac.mc_filter_type << E1000_RCTL_MO_SHIFT);
2294
2295         /* Do not Store bad packets */
2296         rctl &= ~E1000_RCTL_SBP;
2297
2298         /* Enable Long Packet receive */
2299         if (adapter->netdev->mtu <= ETH_DATA_LEN)
2300                 rctl &= ~E1000_RCTL_LPE;
2301         else
2302                 rctl |= E1000_RCTL_LPE;
2303
2304         /* Enable hardware CRC frame stripping */
2305         rctl |= E1000_RCTL_SECRC;
2306
2307         /* Setup buffer sizes */
2308         rctl &= ~E1000_RCTL_SZ_4096;
2309         rctl |= E1000_RCTL_BSEX;
2310         switch (adapter->rx_buffer_len) {
2311         case 256:
2312                 rctl |= E1000_RCTL_SZ_256;
2313                 rctl &= ~E1000_RCTL_BSEX;
2314                 break;
2315         case 512:
2316                 rctl |= E1000_RCTL_SZ_512;
2317                 rctl &= ~E1000_RCTL_BSEX;
2318                 break;
2319         case 1024:
2320                 rctl |= E1000_RCTL_SZ_1024;
2321                 rctl &= ~E1000_RCTL_BSEX;
2322                 break;
2323         case 2048:
2324         default:
2325                 rctl |= E1000_RCTL_SZ_2048;
2326                 rctl &= ~E1000_RCTL_BSEX;
2327                 break;
2328         case 4096:
2329                 rctl |= E1000_RCTL_SZ_4096;
2330                 break;
2331         case 8192:
2332                 rctl |= E1000_RCTL_SZ_8192;
2333                 break;
2334         case 16384:
2335                 rctl |= E1000_RCTL_SZ_16384;
2336                 break;
2337         }
2338
2339         /*
2340          * 82571 and greater support packet-split where the protocol
2341          * header is placed in skb->data and the packet data is
2342          * placed in pages hanging off of skb_shinfo(skb)->nr_frags.
2343          * In the case of a non-split, skb->data is linearly filled,
2344          * followed by the page buffers.  Therefore, skb->data is
2345          * sized to hold the largest protocol header.
2346          *
2347          * allocations using alloc_page take too long for regular MTU
2348          * so only enable packet split for jumbo frames
2349          *
2350          * Using pages when the page size is greater than 16k wastes
2351          * a lot of memory, since we allocate 3 pages at all times
2352          * per packet.
2353          */
2354         pages = PAGE_USE_COUNT(adapter->netdev->mtu);
2355         if (!(adapter->flags & FLAG_IS_ICH) && (pages <= 3) &&
2356             (PAGE_SIZE <= 16384) && (rctl & E1000_RCTL_LPE))
2357                 adapter->rx_ps_pages = pages;
2358         else
2359                 adapter->rx_ps_pages = 0;
2360
2361         if (adapter->rx_ps_pages) {
2362                 /* Configure extra packet-split registers */
2363                 rfctl = er32(RFCTL);
2364                 rfctl |= E1000_RFCTL_EXTEN;
2365                 /*
2366                  * disable packet split support for IPv6 extension headers,
2367                  * because some malformed IPv6 headers can hang the Rx
2368                  */
2369                 rfctl |= (E1000_RFCTL_IPV6_EX_DIS |
2370                           E1000_RFCTL_NEW_IPV6_EXT_DIS);
2371
2372                 ew32(RFCTL, rfctl);
2373
2374                 /* Enable Packet split descriptors */
2375                 rctl |= E1000_RCTL_DTYP_PS;
2376
2377                 psrctl |= adapter->rx_ps_bsize0 >>
2378                         E1000_PSRCTL_BSIZE0_SHIFT;
2379
2380                 switch (adapter->rx_ps_pages) {
2381                 case 3:
2382                         psrctl |= PAGE_SIZE <<
2383                                 E1000_PSRCTL_BSIZE3_SHIFT;
2384                 case 2:
2385                         psrctl |= PAGE_SIZE <<
2386                                 E1000_PSRCTL_BSIZE2_SHIFT;
2387                 case 1:
2388                         psrctl |= PAGE_SIZE >>
2389                                 E1000_PSRCTL_BSIZE1_SHIFT;
2390                         break;
2391                 }
2392
2393                 ew32(PSRCTL, psrctl);
2394         }
2395
2396         ew32(RCTL, rctl);
2397         /* just started the receive unit, no need to restart */
2398         adapter->flags &= ~FLAG_RX_RESTART_NOW;
2399 }
2400
2401 /**
2402  * e1000_configure_rx - Configure Receive Unit after Reset
2403  * @adapter: board private structure
2404  *
2405  * Configure the Rx unit of the MAC after a reset.
2406  **/
2407 static void e1000_configure_rx(struct e1000_adapter *adapter)
2408 {
2409         struct e1000_hw *hw = &adapter->hw;
2410         struct e1000_ring *rx_ring = adapter->rx_ring;
2411         u64 rdba;
2412         u32 rdlen, rctl, rxcsum, ctrl_ext;
2413
2414         if (adapter->rx_ps_pages) {
2415                 /* this is a 32 byte descriptor */
2416                 rdlen = rx_ring->count *
2417                         sizeof(union e1000_rx_desc_packet_split);
2418                 adapter->clean_rx = e1000_clean_rx_irq_ps;
2419                 adapter->alloc_rx_buf = e1000_alloc_rx_buffers_ps;
2420         } else if (adapter->netdev->mtu > ETH_FRAME_LEN + ETH_FCS_LEN) {
2421                 rdlen = rx_ring->count * sizeof(struct e1000_rx_desc);
2422                 adapter->clean_rx = e1000_clean_jumbo_rx_irq;
2423                 adapter->alloc_rx_buf = e1000_alloc_jumbo_rx_buffers;
2424         } else {
2425                 rdlen = rx_ring->count * sizeof(struct e1000_rx_desc);
2426                 adapter->clean_rx = e1000_clean_rx_irq;
2427                 adapter->alloc_rx_buf = e1000_alloc_rx_buffers;
2428         }
2429
2430         /* disable receives while setting up the descriptors */
2431         rctl = er32(RCTL);
2432         ew32(RCTL, rctl & ~E1000_RCTL_EN);
2433         e1e_flush();
2434         msleep(10);
2435
2436         /* set the Receive Delay Timer Register */
2437         ew32(RDTR, adapter->rx_int_delay);
2438
2439         /* irq moderation */
2440         ew32(RADV, adapter->rx_abs_int_delay);
2441         if (adapter->itr_setting != 0)
2442                 ew32(ITR, 1000000000 / (adapter->itr * 256));
2443
2444         ctrl_ext = er32(CTRL_EXT);
2445         /* Reset delay timers after every interrupt */
2446         ctrl_ext |= E1000_CTRL_EXT_INT_TIMER_CLR;
2447         /* Auto-Mask interrupts upon ICR access */
2448         ctrl_ext |= E1000_CTRL_EXT_IAME;
2449         ew32(IAM, 0xffffffff);
2450         ew32(CTRL_EXT, ctrl_ext);
2451         e1e_flush();
2452
2453         /*
2454          * Setup the HW Rx Head and Tail Descriptor Pointers and
2455          * the Base and Length of the Rx Descriptor Ring
2456          */
2457         rdba = rx_ring->dma;
2458         ew32(RDBAL, (rdba & DMA_32BIT_MASK));
2459         ew32(RDBAH, (rdba >> 32));
2460         ew32(RDLEN, rdlen);
2461         ew32(RDH, 0);
2462         ew32(RDT, 0);
2463         rx_ring->head = E1000_RDH;
2464         rx_ring->tail = E1000_RDT;
2465
2466         /* Enable Receive Checksum Offload for TCP and UDP */
2467         rxcsum = er32(RXCSUM);
2468         if (adapter->flags & FLAG_RX_CSUM_ENABLED) {
2469                 rxcsum |= E1000_RXCSUM_TUOFL;
2470
2471                 /*
2472                  * IPv4 payload checksum for UDP fragments must be
2473                  * used in conjunction with packet-split.
2474                  */
2475                 if (adapter->rx_ps_pages)
2476                         rxcsum |= E1000_RXCSUM_IPPCSE;
2477         } else {
2478                 rxcsum &= ~E1000_RXCSUM_TUOFL;
2479                 /* no need to clear IPPCSE as it defaults to 0 */
2480         }
2481         ew32(RXCSUM, rxcsum);
2482
2483         /*
2484          * Enable early receives on supported devices, only takes effect when
2485          * packet size is equal or larger than the specified value (in 8 byte
2486          * units), e.g. using jumbo frames when setting to E1000_ERT_2048
2487          */
2488         if ((adapter->flags & FLAG_HAS_ERT) &&
2489             (adapter->netdev->mtu > ETH_DATA_LEN)) {
2490                 u32 rxdctl = er32(RXDCTL(0));
2491                 ew32(RXDCTL(0), rxdctl | 0x3);
2492                 ew32(ERT, E1000_ERT_2048 | (1 << 13));
2493                 /*
2494                  * With jumbo frames and early-receive enabled, excessive
2495                  * C4->C2 latencies result in dropped transactions.
2496                  */
2497                 pm_qos_update_requirement(PM_QOS_CPU_DMA_LATENCY,
2498                                           e1000e_driver_name, 55);
2499         } else {
2500                 pm_qos_update_requirement(PM_QOS_CPU_DMA_LATENCY,
2501                                           e1000e_driver_name,
2502                                           PM_QOS_DEFAULT_VALUE);
2503         }
2504
2505         /* Enable Receives */
2506         ew32(RCTL, rctl);
2507 }
2508
2509 /**
2510  *  e1000_update_mc_addr_list - Update Multicast addresses
2511  *  @hw: pointer to the HW structure
2512  *  @mc_addr_list: array of multicast addresses to program
2513  *  @mc_addr_count: number of multicast addresses to program
2514  *  @rar_used_count: the first RAR register free to program
2515  *  @rar_count: total number of supported Receive Address Registers
2516  *
2517  *  Updates the Receive Address Registers and Multicast Table Array.
2518  *  The caller must have a packed mc_addr_list of multicast addresses.
2519  *  The parameter rar_count will usually be hw->mac.rar_entry_count
2520  *  unless there are workarounds that change this.  Currently no func pointer
2521  *  exists and all implementations are handled in the generic version of this
2522  *  function.
2523  **/
2524 static void e1000_update_mc_addr_list(struct e1000_hw *hw, u8 *mc_addr_list,
2525                                       u32 mc_addr_count, u32 rar_used_count,
2526                                       u32 rar_count)
2527 {
2528         hw->mac.ops.update_mc_addr_list(hw, mc_addr_list, mc_addr_count,
2529                                         rar_used_count, rar_count);
2530 }
2531
2532 /**
2533  * e1000_set_multi - Multicast and Promiscuous mode set
2534  * @netdev: network interface device structure
2535  *
2536  * The set_multi entry point is called whenever the multicast address
2537  * list or the network interface flags are updated.  This routine is
2538  * responsible for configuring the hardware for proper multicast,
2539  * promiscuous mode, and all-multi behavior.
2540  **/
2541 static void e1000_set_multi(struct net_device *netdev)
2542 {
2543         struct e1000_adapter *adapter = netdev_priv(netdev);
2544         struct e1000_hw *hw = &adapter->hw;
2545         struct e1000_mac_info *mac = &hw->mac;
2546         struct dev_mc_list *mc_ptr;
2547         u8  *mta_list;
2548         u32 rctl;
2549         int i;
2550
2551         /* Check for Promiscuous and All Multicast modes */
2552
2553         rctl = er32(RCTL);
2554
2555         if (netdev->flags & IFF_PROMISC) {
2556                 rctl |= (E1000_RCTL_UPE | E1000_RCTL_MPE);
2557                 rctl &= ~E1000_RCTL_VFE;
2558         } else {
2559                 if (netdev->flags & IFF_ALLMULTI) {
2560                         rctl |= E1000_RCTL_MPE;
2561                         rctl &= ~E1000_RCTL_UPE;
2562                 } else {
2563                         rctl &= ~(E1000_RCTL_UPE | E1000_RCTL_MPE);
2564                 }
2565                 if (adapter->flags & FLAG_HAS_HW_VLAN_FILTER)
2566                         rctl |= E1000_RCTL_VFE;
2567         }
2568
2569         ew32(RCTL, rctl);
2570
2571         if (netdev->mc_count) {
2572                 mta_list = kmalloc(netdev->mc_count * 6, GFP_ATOMIC);
2573                 if (!mta_list)
2574                         return;
2575
2576                 /* prepare a packed array of only addresses. */
2577                 mc_ptr = netdev->mc_list;
2578
2579                 for (i = 0; i < netdev->mc_count; i++) {
2580                         if (!mc_ptr)
2581                                 break;
2582                         memcpy(mta_list + (i*ETH_ALEN), mc_ptr->dmi_addr,
2583                                ETH_ALEN);
2584                         mc_ptr = mc_ptr->next;
2585                 }
2586
2587                 e1000_update_mc_addr_list(hw, mta_list, i, 1,
2588                                           mac->rar_entry_count);
2589                 kfree(mta_list);
2590         } else {
2591                 /*
2592                  * if we're called from probe, we might not have
2593                  * anything to do here, so clear out the list
2594                  */
2595                 e1000_update_mc_addr_list(hw, NULL, 0, 1, mac->rar_entry_count);
2596         }
2597 }
2598
2599 /**
2600  * e1000_configure - configure the hardware for Rx and Tx
2601  * @adapter: private board structure
2602  **/
2603 static void e1000_configure(struct e1000_adapter *adapter)
2604 {
2605         e1000_set_multi(adapter->netdev);
2606
2607         e1000_restore_vlan(adapter);
2608         e1000_init_manageability(adapter);
2609
2610         e1000_configure_tx(adapter);
2611         e1000_setup_rctl(adapter);
2612         e1000_configure_rx(adapter);
2613         adapter->alloc_rx_buf(adapter, e1000_desc_unused(adapter->rx_ring));
2614 }
2615
2616 /**
2617  * e1000e_power_up_phy - restore link in case the phy was powered down
2618  * @adapter: address of board private structure
2619  *
2620  * The phy may be powered down to save power and turn off link when the
2621  * driver is unloaded and wake on lan is not enabled (among others)
2622  * *** this routine MUST be followed by a call to e1000e_reset ***
2623  **/
2624 void e1000e_power_up_phy(struct e1000_adapter *adapter)
2625 {
2626         u16 mii_reg = 0;
2627
2628         /* Just clear the power down bit to wake the phy back up */
2629         if (adapter->hw.phy.media_type == e1000_media_type_copper) {
2630                 /*
2631                  * According to the manual, the phy will retain its
2632                  * settings across a power-down/up cycle
2633                  */
2634                 e1e_rphy(&adapter->hw, PHY_CONTROL, &mii_reg);
2635                 mii_reg &= ~MII_CR_POWER_DOWN;
2636                 e1e_wphy(&adapter->hw, PHY_CONTROL, mii_reg);
2637         }
2638
2639         adapter->hw.mac.ops.setup_link(&adapter->hw);
2640 }
2641
2642 /**
2643  * e1000_power_down_phy - Power down the PHY
2644  *
2645  * Power down the PHY so no link is implied when interface is down
2646  * The PHY cannot be powered down is management or WoL is active
2647  */
2648 static void e1000_power_down_phy(struct e1000_adapter *adapter)
2649 {
2650         struct e1000_hw *hw = &adapter->hw;
2651         u16 mii_reg;
2652
2653         /* WoL is enabled */
2654         if (adapter->wol)
2655                 return;
2656
2657         /* non-copper PHY? */
2658         if (adapter->hw.phy.media_type != e1000_media_type_copper)
2659                 return;
2660
2661         /* reset is blocked because of a SoL/IDER session */
2662         if (e1000e_check_mng_mode(hw) || e1000_check_reset_block(hw))
2663                 return;
2664
2665         /* manageability (AMT) is enabled */
2666         if (er32(MANC) & E1000_MANC_SMBUS_EN)
2667                 return;
2668
2669         /* power down the PHY */
2670         e1e_rphy(hw, PHY_CONTROL, &mii_reg);
2671         mii_reg |= MII_CR_POWER_DOWN;
2672         e1e_wphy(hw, PHY_CONTROL, mii_reg);
2673         mdelay(1);
2674 }
2675
2676 /**
2677  * e1000e_reset - bring the hardware into a known good state
2678  *
2679  * This function boots the hardware and enables some settings that
2680  * require a configuration cycle of the hardware - those cannot be
2681  * set/changed during runtime. After reset the device needs to be
2682  * properly configured for Rx, Tx etc.
2683  */
2684 void e1000e_reset(struct e1000_adapter *adapter)
2685 {
2686         struct e1000_mac_info *mac = &adapter->hw.mac;
2687         struct e1000_fc_info *fc = &adapter->hw.fc;
2688         struct e1000_hw *hw = &adapter->hw;
2689         u32 tx_space, min_tx_space, min_rx_space;
2690         u32 pba = adapter->pba;
2691         u16 hwm;
2692
2693         /* reset Packet Buffer Allocation to default */
2694         ew32(PBA, pba);
2695
2696         if (adapter->max_frame_size > ETH_FRAME_LEN + ETH_FCS_LEN) {
2697                 /*
2698                  * To maintain wire speed transmits, the Tx FIFO should be
2699                  * large enough to accommodate two full transmit packets,
2700                  * rounded up to the next 1KB and expressed in KB.  Likewise,
2701                  * the Rx FIFO should be large enough to accommodate at least
2702                  * one full receive packet and is similarly rounded up and
2703                  * expressed in KB.
2704                  */
2705                 pba = er32(PBA);
2706                 /* upper 16 bits has Tx packet buffer allocation size in KB */
2707                 tx_space = pba >> 16;
2708                 /* lower 16 bits has Rx packet buffer allocation size in KB */
2709                 pba &= 0xffff;
2710                 /*
2711                  * the Tx fifo also stores 16 bytes of information about the tx
2712                  * but don't include ethernet FCS because hardware appends it
2713                  */
2714                 min_tx_space = (adapter->max_frame_size +
2715                                 sizeof(struct e1000_tx_desc) -
2716                                 ETH_FCS_LEN) * 2;
2717                 min_tx_space = ALIGN(min_tx_space, 1024);
2718                 min_tx_space >>= 10;
2719                 /* software strips receive CRC, so leave room for it */
2720                 min_rx_space = adapter->max_frame_size;
2721                 min_rx_space = ALIGN(min_rx_space, 1024);
2722                 min_rx_space >>= 10;
2723
2724                 /*
2725                  * If current Tx allocation is less than the min Tx FIFO size,
2726                  * and the min Tx FIFO size is less than the current Rx FIFO
2727                  * allocation, take space away from current Rx allocation
2728                  */
2729                 if ((tx_space < min_tx_space) &&
2730                     ((min_tx_space - tx_space) < pba)) {
2731                         pba -= min_tx_space - tx_space;
2732
2733                         /*
2734                          * if short on Rx space, Rx wins and must trump tx
2735                          * adjustment or use Early Receive if available
2736                          */
2737                         if ((pba < min_rx_space) &&
2738                             (!(adapter->flags & FLAG_HAS_ERT)))
2739                                 /* ERT enabled in e1000_configure_rx */
2740                                 pba = min_rx_space;
2741                 }
2742
2743                 ew32(PBA, pba);
2744         }
2745
2746
2747         /*
2748          * flow control settings
2749          *
2750          * The high water mark must be low enough to fit one full frame
2751          * (or the size used for early receive) above it in the Rx FIFO.
2752          * Set it to the lower of:
2753          * - 90% of the Rx FIFO size, and
2754          * - the full Rx FIFO size minus the early receive size (for parts
2755          *   with ERT support assuming ERT set to E1000_ERT_2048), or
2756          * - the full Rx FIFO size minus one full frame
2757          */
2758         if (adapter->flags & FLAG_HAS_ERT)
2759                 hwm = min(((pba << 10) * 9 / 10),
2760                           ((pba << 10) - (E1000_ERT_2048 << 3)));
2761         else
2762                 hwm = min(((pba << 10) * 9 / 10),
2763                           ((pba << 10) - adapter->max_frame_size));
2764
2765         fc->high_water = hwm & 0xFFF8; /* 8-byte granularity */
2766         fc->low_water = fc->high_water - 8;
2767
2768         if (adapter->flags & FLAG_DISABLE_FC_PAUSE_TIME)
2769                 fc->pause_time = 0xFFFF;
2770         else
2771                 fc->pause_time = E1000_FC_PAUSE_TIME;
2772         fc->send_xon = 1;
2773         fc->type = fc->original_type;
2774
2775         /* Allow time for pending master requests to run */
2776         mac->ops.reset_hw(hw);
2777
2778         /*
2779          * For parts with AMT enabled, let the firmware know
2780          * that the network interface is in control
2781          */
2782         if (adapter->flags & FLAG_HAS_AMT)
2783                 e1000_get_hw_control(adapter);
2784
2785         ew32(WUC, 0);
2786
2787         if (mac->ops.init_hw(hw))
2788                 e_err("Hardware Error\n");
2789
2790         e1000_update_mng_vlan(adapter);
2791
2792         /* Enable h/w to recognize an 802.1Q VLAN Ethernet packet */
2793         ew32(VET, ETH_P_8021Q);
2794
2795         e1000e_reset_adaptive(hw);
2796         e1000_get_phy_info(hw);
2797
2798         if (!(adapter->flags & FLAG_SMART_POWER_DOWN)) {
2799                 u16 phy_data = 0;
2800                 /*
2801                  * speed up time to link by disabling smart power down, ignore
2802                  * the return value of this function because there is nothing
2803                  * different we would do if it failed
2804                  */
2805                 e1e_rphy(hw, IGP02E1000_PHY_POWER_MGMT, &phy_data);
2806                 phy_data &= ~IGP02E1000_PM_SPD;
2807                 e1e_wphy(hw, IGP02E1000_PHY_POWER_MGMT, phy_data);
2808         }
2809 }
2810
2811 int e1000e_up(struct e1000_adapter *adapter)
2812 {
2813         struct e1000_hw *hw = &adapter->hw;
2814
2815         /* hardware has been reset, we need to reload some things */
2816         e1000_configure(adapter);
2817
2818         clear_bit(__E1000_DOWN, &adapter->state);
2819
2820         napi_enable(&adapter->napi);
2821         if (adapter->msix_entries)
2822                 e1000_configure_msix(adapter);
2823         e1000_irq_enable(adapter);
2824
2825         /* fire a link change interrupt to start the watchdog */
2826         ew32(ICS, E1000_ICS_LSC);
2827         return 0;
2828 }
2829
2830 void e1000e_down(struct e1000_adapter *adapter)
2831 {
2832         struct net_device *netdev = adapter->netdev;
2833         struct e1000_hw *hw = &adapter->hw;
2834         u32 tctl, rctl;
2835
2836         /*
2837          * signal that we're down so the interrupt handler does not
2838          * reschedule our watchdog timer
2839          */
2840         set_bit(__E1000_DOWN, &adapter->state);
2841
2842         /* disable receives in the hardware */
2843         rctl = er32(RCTL);
2844         ew32(RCTL, rctl & ~E1000_RCTL_EN);
2845         /* flush and sleep below */
2846
2847         netif_tx_stop_all_queues(netdev);
2848
2849         /* disable transmits in the hardware */
2850         tctl = er32(TCTL);
2851         tctl &= ~E1000_TCTL_EN;
2852         ew32(TCTL, tctl);
2853         /* flush both disables and wait for them to finish */
2854         e1e_flush();
2855         msleep(10);
2856
2857         napi_disable(&adapter->napi);
2858         e1000_irq_disable(adapter);
2859
2860         del_timer_sync(&adapter->watchdog_timer);
2861         del_timer_sync(&adapter->phy_info_timer);
2862
2863         netdev->tx_queue_len = adapter->tx_queue_len;
2864         netif_carrier_off(netdev);
2865         adapter->link_speed = 0;
2866         adapter->link_duplex = 0;
2867
2868         if (!pci_channel_offline(adapter->pdev))
2869                 e1000e_reset(adapter);
2870         e1000_clean_tx_ring(adapter);
2871         e1000_clean_rx_ring(adapter);
2872
2873         /*
2874          * TODO: for power management, we could drop the link and
2875          * pci_disable_device here.
2876          */
2877 }
2878
2879 void e1000e_reinit_locked(struct e1000_adapter *adapter)
2880 {
2881         might_sleep();
2882         while (test_and_set_bit(__E1000_RESETTING, &adapter->state))
2883                 msleep(1);
2884         e1000e_down(adapter);
2885         e1000e_up(adapter);
2886         clear_bit(__E1000_RESETTING, &adapter->state);
2887 }
2888
2889 /**
2890  * e1000_sw_init - Initialize general software structures (struct e1000_adapter)
2891  * @adapter: board private structure to initialize
2892  *
2893  * e1000_sw_init initializes the Adapter private data structure.
2894  * Fields are initialized based on PCI device information and
2895  * OS network device settings (MTU size).
2896  **/
2897 static int __devinit e1000_sw_init(struct e1000_adapter *adapter)
2898 {
2899         struct net_device *netdev = adapter->netdev;
2900
2901         adapter->rx_buffer_len = ETH_FRAME_LEN + VLAN_HLEN + ETH_FCS_LEN;
2902         adapter->rx_ps_bsize0 = 128;
2903         adapter->max_frame_size = netdev->mtu + ETH_HLEN + ETH_FCS_LEN;
2904         adapter->min_frame_size = ETH_ZLEN + ETH_FCS_LEN;
2905
2906         e1000e_set_interrupt_capability(adapter);
2907
2908         if (e1000_alloc_queues(adapter))
2909                 return -ENOMEM;
2910
2911         spin_lock_init(&adapter->tx_queue_lock);
2912
2913         /* Explicitly disable IRQ since the NIC can be in any state. */
2914         e1000_irq_disable(adapter);
2915
2916         set_bit(__E1000_DOWN, &adapter->state);
2917         return 0;
2918 }
2919
2920 /**
2921  * e1000_intr_msi_test - Interrupt Handler
2922  * @irq: interrupt number
2923  * @data: pointer to a network interface device structure
2924  **/
2925 static irqreturn_t e1000_intr_msi_test(int irq, void *data)
2926 {
2927         struct net_device *netdev = data;
2928         struct e1000_adapter *adapter = netdev_priv(netdev);
2929         struct e1000_hw *hw = &adapter->hw;
2930         u32 icr = er32(ICR);
2931
2932         e_dbg("%s: icr is %08X\n", netdev->name, icr);
2933         if (icr & E1000_ICR_RXSEQ) {
2934                 adapter->flags &= ~FLAG_MSI_TEST_FAILED;
2935                 wmb();
2936         }
2937
2938         return IRQ_HANDLED;
2939 }
2940
2941 /**
2942  * e1000_test_msi_interrupt - Returns 0 for successful test
2943  * @adapter: board private struct
2944  *
2945  * code flow taken from tg3.c
2946  **/
2947 static int e1000_test_msi_interrupt(struct e1000_adapter *adapter)
2948 {
2949         struct net_device *netdev = adapter->netdev;
2950         struct e1000_hw *hw = &adapter->hw;
2951         int err;
2952
2953         /* poll_enable hasn't been called yet, so don't need disable */
2954         /* clear any pending events */
2955         er32(ICR);
2956
2957         /* free the real vector and request a test handler */
2958         e1000_free_irq(adapter);
2959         e1000e_reset_interrupt_capability(adapter);
2960
2961         /* Assume that the test fails, if it succeeds then the test
2962          * MSI irq handler will unset this flag */
2963         adapter->flags |= FLAG_MSI_TEST_FAILED;
2964
2965         err = pci_enable_msi(adapter->pdev);
2966         if (err)
2967                 goto msi_test_failed;
2968
2969         err = request_irq(adapter->pdev->irq, &e1000_intr_msi_test, 0,
2970                           netdev->name, netdev);
2971         if (err) {
2972                 pci_disable_msi(adapter->pdev);
2973                 goto msi_test_failed;
2974         }
2975
2976         wmb();
2977
2978         e1000_irq_enable(adapter);
2979
2980         /* fire an unusual interrupt on the test handler */
2981         ew32(ICS, E1000_ICS_RXSEQ);
2982         e1e_flush();
2983         msleep(50);
2984
2985         e1000_irq_disable(adapter);
2986
2987         rmb();
2988
2989         if (adapter->flags & FLAG_MSI_TEST_FAILED) {
2990                 adapter->int_mode = E1000E_INT_MODE_LEGACY;
2991                 err = -EIO;
2992                 e_info("MSI interrupt test failed!\n");
2993         }
2994
2995         free_irq(adapter->pdev->irq, netdev);
2996         pci_disable_msi(adapter->pdev);
2997
2998         if (err == -EIO)
2999                 goto msi_test_failed;
3000
3001         /* okay so the test worked, restore settings */
3002         e_dbg("%s: MSI interrupt test succeeded!\n", netdev->name);
3003 msi_test_failed:
3004         e1000e_set_interrupt_capability(adapter);
3005         e1000_request_irq(adapter);
3006         return err;
3007 }
3008
3009 /**
3010  * e1000_test_msi - Returns 0 if MSI test succeeds or INTx mode is restored
3011  * @adapter: board private struct
3012  *
3013  * code flow taken from tg3.c, called with e1000 interrupts disabled.
3014  **/
3015 static int e1000_test_msi(struct e1000_adapter *adapter)
3016 {
3017         int err;
3018         u16 pci_cmd;
3019
3020         if (!(adapter->flags & FLAG_MSI_ENABLED))
3021                 return 0;
3022
3023         /* disable SERR in case the MSI write causes a master abort */
3024         pci_read_config_word(adapter->pdev, PCI_COMMAND, &pci_cmd);
3025         pci_write_config_word(adapter->pdev, PCI_COMMAND,
3026                               pci_cmd & ~PCI_COMMAND_SERR);
3027
3028         err = e1000_test_msi_interrupt(adapter);
3029
3030         /* restore previous setting of command word */
3031         pci_write_config_word(adapter->pdev, PCI_COMMAND, pci_cmd);
3032
3033         /* success ! */
3034         if (!err)
3035                 return 0;
3036
3037         /* EIO means MSI test failed */
3038         if (err != -EIO)
3039                 return err;
3040
3041         /* back to INTx mode */
3042         e_warn("MSI interrupt test failed, using legacy interrupt.\n");
3043
3044         e1000_free_irq(adapter);
3045
3046         err = e1000_request_irq(adapter);
3047
3048         return err;
3049 }
3050
3051 /**
3052  * e1000_open - Called when a network interface is made active
3053  * @netdev: network interface device structure
3054  *
3055  * Returns 0 on success, negative value on failure
3056  *
3057  * The open entry point is called when a network interface is made
3058  * active by the system (IFF_UP).  At this point all resources needed
3059  * for transmit and receive operations are allocated, the interrupt
3060  * handler is registered with the OS, the watchdog timer is started,
3061  * and the stack is notified that the interface is ready.
3062  **/
3063 static int e1000_open(struct net_device *netdev)
3064 {
3065         struct e1000_adapter *adapter = netdev_priv(netdev);
3066         struct e1000_hw *hw = &adapter->hw;
3067         int err;
3068
3069         /* disallow open during test */
3070         if (test_bit(__E1000_TESTING, &adapter->state))
3071                 return -EBUSY;
3072
3073         /* allocate transmit descriptors */
3074         err = e1000e_setup_tx_resources(adapter);
3075         if (err)
3076                 goto err_setup_tx;
3077
3078         /* allocate receive descriptors */
3079         err = e1000e_setup_rx_resources(adapter);
3080         if (err)
3081                 goto err_setup_rx;
3082
3083         e1000e_power_up_phy(adapter);
3084
3085         adapter->mng_vlan_id = E1000_MNG_VLAN_NONE;
3086         if ((adapter->hw.mng_cookie.status &
3087              E1000_MNG_DHCP_COOKIE_STATUS_VLAN))
3088                 e1000_update_mng_vlan(adapter);
3089
3090         /*
3091          * If AMT is enabled, let the firmware know that the network
3092          * interface is now open
3093          */
3094         if (adapter->flags & FLAG_HAS_AMT)
3095                 e1000_get_hw_control(adapter);
3096
3097         /*
3098          * before we allocate an interrupt, we must be ready to handle it.
3099          * Setting DEBUG_SHIRQ in the kernel makes it fire an interrupt
3100          * as soon as we call pci_request_irq, so we have to setup our
3101          * clean_rx handler before we do so.
3102          */
3103         e1000_configure(adapter);
3104
3105         err = e1000_request_irq(adapter);
3106         if (err)
3107                 goto err_req_irq;
3108
3109         /*
3110          * Work around PCIe errata with MSI interrupts causing some chipsets to
3111          * ignore e1000e MSI messages, which means we need to test our MSI
3112          * interrupt now
3113          */
3114         if (adapter->int_mode != E1000E_INT_MODE_LEGACY) {
3115                 err = e1000_test_msi(adapter);
3116                 if (err) {
3117                         e_err("Interrupt allocation failed\n");
3118                         goto err_req_irq;
3119                 }
3120         }
3121
3122         /* From here on the code is the same as e1000e_up() */
3123         clear_bit(__E1000_DOWN, &adapter->state);
3124
3125         napi_enable(&adapter->napi);
3126
3127         e1000_irq_enable(adapter);
3128
3129         netif_tx_start_all_queues(netdev);
3130
3131         /* fire a link status change interrupt to start the watchdog */
3132         ew32(ICS, E1000_ICS_LSC);
3133
3134         return 0;
3135
3136 err_req_irq:
3137         e1000_release_hw_control(adapter);
3138         e1000_power_down_phy(adapter);
3139         e1000e_free_rx_resources(adapter);
3140 err_setup_rx:
3141         e1000e_free_tx_resources(adapter);
3142 err_setup_tx:
3143         e1000e_reset(adapter);
3144
3145         return err;
3146 }
3147
3148 /**
3149  * e1000_close - Disables a network interface
3150  * @netdev: network interface device structure
3151  *
3152  * Returns 0, this is not allowed to fail
3153  *
3154  * The close entry point is called when an interface is de-activated
3155  * by the OS.  The hardware is still under the drivers control, but
3156  * needs to be disabled.  A global MAC reset is issued to stop the
3157  * hardware, and all transmit and receive resources are freed.
3158  **/
3159 static int e1000_close(struct net_device *netdev)
3160 {
3161         struct e1000_adapter *adapter = netdev_priv(netdev);
3162
3163         WARN_ON(test_bit(__E1000_RESETTING, &adapter->state));
3164         e1000e_down(adapter);
3165         e1000_power_down_phy(adapter);
3166         e1000_free_irq(adapter);
3167
3168         e1000e_free_tx_resources(adapter);
3169         e1000e_free_rx_resources(adapter);
3170
3171         /*
3172          * kill manageability vlan ID if supported, but not if a vlan with
3173          * the same ID is registered on the host OS (let 8021q kill it)
3174          */
3175         if ((adapter->hw.mng_cookie.status &
3176                           E1000_MNG_DHCP_COOKIE_STATUS_VLAN) &&
3177              !(adapter->vlgrp &&
3178                vlan_group_get_device(adapter->vlgrp, adapter->mng_vlan_id)))
3179                 e1000_vlan_rx_kill_vid(netdev, adapter->mng_vlan_id);
3180
3181         /*
3182          * If AMT is enabled, let the firmware know that the network
3183          * interface is now closed
3184          */
3185         if (adapter->flags & FLAG_HAS_AMT)
3186                 e1000_release_hw_control(adapter);
3187
3188         return 0;
3189 }
3190 /**
3191  * e1000_set_mac - Change the Ethernet Address of the NIC
3192  * @netdev: network interface device structure
3193  * @p: pointer to an address structure
3194  *
3195  * Returns 0 on success, negative on failure
3196  **/
3197 static int e1000_set_mac(struct net_device *netdev, void *p)
3198 {
3199         struct e1000_adapter *adapter = netdev_priv(netdev);
3200         struct sockaddr *addr = p;
3201
3202         if (!is_valid_ether_addr(addr->sa_data))
3203                 return -EADDRNOTAVAIL;
3204
3205         memcpy(netdev->dev_addr, addr->sa_data, netdev->addr_len);
3206         memcpy(adapter->hw.mac.addr, addr->sa_data, netdev->addr_len);
3207
3208         e1000e_rar_set(&adapter->hw, adapter->hw.mac.addr, 0);
3209
3210         if (adapter->flags & FLAG_RESET_OVERWRITES_LAA) {
3211                 /* activate the work around */
3212                 e1000e_set_laa_state_82571(&adapter->hw, 1);
3213
3214                 /*
3215                  * Hold a copy of the LAA in RAR[14] This is done so that
3216                  * between the time RAR[0] gets clobbered  and the time it
3217                  * gets fixed (in e1000_watchdog), the actual LAA is in one
3218                  * of the RARs and no incoming packets directed to this port
3219                  * are dropped. Eventually the LAA will be in RAR[0] and
3220                  * RAR[14]
3221                  */
3222                 e1000e_rar_set(&adapter->hw,
3223                               adapter->hw.mac.addr,
3224                               adapter->hw.mac.rar_entry_count - 1);
3225         }
3226
3227         return 0;
3228 }
3229
3230 /**
3231  * e1000e_update_phy_task - work thread to update phy
3232  * @work: pointer to our work struct
3233  *
3234  * this worker thread exists because we must acquire a
3235  * semaphore to read the phy, which we could msleep while
3236  * waiting for it, and we can't msleep in a timer.
3237  **/
3238 static void e1000e_update_phy_task(struct work_struct *work)
3239 {
3240         struct e1000_adapter *adapter = container_of(work,
3241                                         struct e1000_adapter, update_phy_task);
3242         e1000_get_phy_info(&adapter->hw);
3243 }
3244
3245 /*
3246  * Need to wait a few seconds after link up to get diagnostic information from
3247  * the phy
3248  */
3249 static void e1000_update_phy_info(unsigned long data)
3250 {
3251         struct e1000_adapter *adapter = (struct e1000_adapter *) data;
3252         schedule_work(&adapter->update_phy_task);
3253 }
3254
3255 /**
3256  * e1000e_update_stats - Update the board statistics counters
3257  * @adapter: board private structure
3258  **/
3259 void e1000e_update_stats(struct e1000_adapter *adapter)
3260 {
3261         struct e1000_hw *hw = &adapter->hw;
3262         struct pci_dev *pdev = adapter->pdev;
3263
3264         /*
3265          * Prevent stats update while adapter is being reset, or if the pci
3266          * connection is down.
3267          */
3268         if (adapter->link_speed == 0)
3269                 return;
3270         if (pci_channel_offline(pdev))
3271                 return;
3272
3273         adapter->stats.crcerrs += er32(CRCERRS);
3274         adapter->stats.gprc += er32(GPRC);
3275         adapter->stats.gorc += er32(GORCL);
3276         er32(GORCH); /* Clear gorc */
3277         adapter->stats.bprc += er32(BPRC);
3278         adapter->stats.mprc += er32(MPRC);
3279         adapter->stats.roc += er32(ROC);
3280
3281         adapter->stats.mpc += er32(MPC);
3282         adapter->stats.scc += er32(SCC);
3283         adapter->stats.ecol += er32(ECOL);
3284         adapter->stats.mcc += er32(MCC);
3285         adapter->stats.latecol += er32(LATECOL);
3286         adapter->stats.dc += er32(DC);
3287         adapter->stats.xonrxc += er32(XONRXC);
3288         adapter->stats.xontxc += er32(XONTXC);
3289         adapter->stats.xoffrxc += er32(XOFFRXC);
3290         adapter->stats.xofftxc += er32(XOFFTXC);
3291         adapter->stats.gptc += er32(GPTC);
3292         adapter->stats.gotc += er32(GOTCL);
3293         er32(GOTCH); /* Clear gotc */
3294         adapter->stats.rnbc += er32(RNBC);
3295         adapter->stats.ruc += er32(RUC);
3296
3297         adapter->stats.mptc += er32(MPTC);
3298         adapter->stats.bptc += er32(BPTC);
3299
3300         /* used for adaptive IFS */
3301
3302         hw->mac.tx_packet_delta = er32(TPT);
3303         adapter->stats.tpt += hw->mac.tx_packet_delta;
3304         hw->mac.collision_delta = er32(COLC);
3305         adapter->stats.colc += hw->mac.collision_delta;
3306
3307         adapter->stats.algnerrc += er32(ALGNERRC);
3308         adapter->stats.rxerrc += er32(RXERRC);
3309         if (hw->mac.type != e1000_82574)
3310                 adapter->stats.tncrs += er32(TNCRS);
3311         adapter->stats.cexterr += er32(CEXTERR);
3312         adapter->stats.tsctc += er32(TSCTC);
3313         adapter->stats.tsctfc += er32(TSCTFC);
3314
3315         /* Fill out the OS statistics structure */
3316         adapter->net_stats.multicast = adapter->stats.mprc;
3317         adapter->net_stats.collisions = adapter->stats.colc;
3318
3319         /* Rx Errors */
3320
3321         /*
3322          * RLEC on some newer hardware can be incorrect so build
3323          * our own version based on RUC and ROC
3324          */
3325         adapter->net_stats.rx_errors = adapter->stats.rxerrc +
3326                 adapter->stats.crcerrs + adapter->stats.algnerrc +
3327                 adapter->stats.ruc + adapter->stats.roc +
3328                 adapter->stats.cexterr;
3329         adapter->net_stats.rx_length_errors = adapter->stats.ruc +
3330                                               adapter->stats.roc;
3331         adapter->net_stats.rx_crc_errors = adapter->stats.crcerrs;
3332         adapter->net_stats.rx_frame_errors = adapter->stats.algnerrc;
3333         adapter->net_stats.rx_missed_errors = adapter->stats.mpc;
3334
3335         /* Tx Errors */
3336         adapter->net_stats.tx_errors = adapter->stats.ecol +
3337                                        adapter->stats.latecol;
3338         adapter->net_stats.tx_aborted_errors = adapter->stats.ecol;
3339         adapter->net_stats.tx_window_errors = adapter->stats.latecol;
3340         adapter->net_stats.tx_carrier_errors = adapter->stats.tncrs;
3341
3342         /* Tx Dropped needs to be maintained elsewhere */
3343
3344         /* Management Stats */
3345         adapter->stats.mgptc += er32(MGTPTC);
3346         adapter->stats.mgprc += er32(MGTPRC);
3347         adapter->stats.mgpdc += er32(MGTPDC);
3348 }
3349
3350 /**
3351  * e1000_phy_read_status - Update the PHY register status snapshot
3352  * @adapter: board private structure
3353  **/
3354 static void e1000_phy_read_status(struct e1000_adapter *adapter)
3355 {
3356         struct e1000_hw *hw = &adapter->hw;
3357         struct e1000_phy_regs *phy = &adapter->phy_regs;
3358         int ret_val;
3359
3360         if ((er32(STATUS) & E1000_STATUS_LU) &&
3361             (adapter->hw.phy.media_type == e1000_media_type_copper)) {
3362                 ret_val  = e1e_rphy(hw, PHY_CONTROL, &phy->bmcr);
3363                 ret_val |= e1e_rphy(hw, PHY_STATUS, &phy->bmsr);
3364                 ret_val |= e1e_rphy(hw, PHY_AUTONEG_ADV, &phy->advertise);
3365                 ret_val |= e1e_rphy(hw, PHY_LP_ABILITY, &phy->lpa);
3366                 ret_val |= e1e_rphy(hw, PHY_AUTONEG_EXP, &phy->expansion);
3367                 ret_val |= e1e_rphy(hw, PHY_1000T_CTRL, &phy->ctrl1000);
3368                 ret_val |= e1e_rphy(hw, PHY_1000T_STATUS, &phy->stat1000);
3369                 ret_val |= e1e_rphy(hw, PHY_EXT_STATUS, &phy->estatus);
3370                 if (ret_val)
3371                         e_warn("Error reading PHY register\n");
3372         } else {
3373                 /*
3374                  * Do not read PHY registers if link is not up
3375                  * Set values to typical power-on defaults
3376                  */
3377                 phy->bmcr = (BMCR_SPEED1000 | BMCR_ANENABLE | BMCR_FULLDPLX);
3378                 phy->bmsr = (BMSR_100FULL | BMSR_100HALF | BMSR_10FULL |
3379                              BMSR_10HALF | BMSR_ESTATEN | BMSR_ANEGCAPABLE |
3380                              BMSR_ERCAP);
3381                 phy->advertise = (ADVERTISE_PAUSE_ASYM | ADVERTISE_PAUSE_CAP |
3382                                   ADVERTISE_ALL | ADVERTISE_CSMA);
3383                 phy->lpa = 0;
3384                 phy->expansion = EXPANSION_ENABLENPAGE;
3385                 phy->ctrl1000 = ADVERTISE_1000FULL;
3386                 phy->stat1000 = 0;
3387                 phy->estatus = (ESTATUS_1000_TFULL | ESTATUS_1000_THALF);
3388         }
3389 }
3390
3391 static void e1000_print_link_info(struct e1000_adapter *adapter)
3392 {
3393         struct e1000_hw *hw = &adapter->hw;
3394         u32 ctrl = er32(CTRL);
3395
3396         e_info("Link is Up %d Mbps %s, Flow Control: %s\n",
3397                adapter->link_speed,
3398                (adapter->link_duplex == FULL_DUPLEX) ?
3399                                 "Full Duplex" : "Half Duplex",
3400                ((ctrl & E1000_CTRL_TFCE) && (ctrl & E1000_CTRL_RFCE)) ?
3401                                 "RX/TX" :
3402                ((ctrl & E1000_CTRL_RFCE) ? "RX" :
3403                ((ctrl & E1000_CTRL_TFCE) ? "TX" : "None" )));
3404 }
3405
3406 static bool e1000_has_link(struct e1000_adapter *adapter)
3407 {
3408         struct e1000_hw *hw = &adapter->hw;
3409         bool link_active = 0;
3410         s32 ret_val = 0;
3411
3412         /*
3413          * get_link_status is set on LSC (link status) interrupt or
3414          * Rx sequence error interrupt.  get_link_status will stay
3415          * false until the check_for_link establishes link
3416          * for copper adapters ONLY
3417          */
3418         switch (hw->phy.media_type) {
3419         case e1000_media_type_copper:
3420                 if (hw->mac.get_link_status) {
3421                         ret_val = hw->mac.ops.check_for_link(hw);
3422                         link_active = !hw->mac.get_link_status;
3423                 } else {
3424                         link_active = 1;
3425                 }
3426                 break;
3427         case e1000_media_type_fiber:
3428                 ret_val = hw->mac.ops.check_for_link(hw);
3429                 link_active = !!(er32(STATUS) & E1000_STATUS_LU);
3430                 break;
3431         case e1000_media_type_internal_serdes:
3432                 ret_val = hw->mac.ops.check_for_link(hw);
3433                 link_active = adapter->hw.mac.serdes_has_link;
3434                 break;
3435         default:
3436         case e1000_media_type_unknown:
3437                 break;
3438         }
3439
3440         if ((ret_val == E1000_ERR_PHY) && (hw->phy.type == e1000_phy_igp_3) &&
3441             (er32(CTRL) & E1000_PHY_CTRL_GBE_DISABLE)) {
3442                 /* See e1000_kmrn_lock_loss_workaround_ich8lan() */
3443                 e_info("Gigabit has been disabled, downgrading speed\n");
3444         }
3445
3446         return link_active;
3447 }
3448
3449 static void e1000e_enable_receives(struct e1000_adapter *adapter)
3450 {
3451         /* make sure the receive unit is started */
3452         if ((adapter->flags & FLAG_RX_NEEDS_RESTART) &&
3453             (adapter->flags & FLAG_RX_RESTART_NOW)) {
3454                 struct e1000_hw *hw = &adapter->hw;
3455                 u32 rctl = er32(RCTL);
3456                 ew32(RCTL, rctl | E1000_RCTL_EN);
3457                 adapter->flags &= ~FLAG_RX_RESTART_NOW;
3458         }
3459 }
3460
3461 /**
3462  * e1000_watchdog - Timer Call-back
3463  * @data: pointer to adapter cast into an unsigned long
3464  **/
3465 static void e1000_watchdog(unsigned long data)
3466 {
3467         struct e1000_adapter *adapter = (struct e1000_adapter *) data;
3468
3469         /* Do the rest outside of interrupt context */
3470         schedule_work(&adapter->watchdog_task);
3471
3472         /* TODO: make this use queue_delayed_work() */
3473 }
3474
3475 static void e1000_watchdog_task(struct work_struct *work)
3476 {
3477         struct e1000_adapter *adapter = container_of(work,
3478                                         struct e1000_adapter, watchdog_task);
3479         struct net_device *netdev = adapter->netdev;
3480         struct e1000_mac_info *mac = &adapter->hw.mac;
3481         struct e1000_ring *tx_ring = adapter->tx_ring;
3482         struct e1000_hw *hw = &adapter->hw;
3483         u32 link, tctl;
3484         int tx_pending = 0;
3485
3486         link = e1000_has_link(adapter);
3487         if ((netif_carrier_ok(netdev)) && link) {
3488                 e1000e_enable_receives(adapter);
3489                 goto link_up;
3490         }
3491
3492         if ((e1000e_enable_tx_pkt_filtering(hw)) &&
3493             (adapter->mng_vlan_id != adapter->hw.mng_cookie.vlan_id))
3494                 e1000_update_mng_vlan(adapter);
3495
3496         if (link) {
3497                 if (!netif_carrier_ok(netdev)) {
3498                         bool txb2b = 1;
3499                         /* update snapshot of PHY registers on LSC */
3500                         e1000_phy_read_status(adapter);
3501                         mac->ops.get_link_up_info(&adapter->hw,
3502                                                    &adapter->link_speed,
3503                                                    &adapter->link_duplex);
3504                         e1000_print_link_info(adapter);
3505                         /*
3506                          * On supported PHYs, check for duplex mismatch only
3507                          * if link has autonegotiated at 10/100 half
3508                          */
3509                         if ((hw->phy.type == e1000_phy_igp_3 ||
3510                              hw->phy.type == e1000_phy_bm) &&
3511                             (hw->mac.autoneg == true) &&
3512                             (adapter->link_speed == SPEED_10 ||
3513                              adapter->link_speed == SPEED_100) &&
3514                             (adapter->link_duplex == HALF_DUPLEX)) {
3515                                 u16 autoneg_exp;
3516
3517                                 e1e_rphy(hw, PHY_AUTONEG_EXP, &autoneg_exp);
3518
3519                                 if (!(autoneg_exp & NWAY_ER_LP_NWAY_CAPS))
3520                                         e_info("Autonegotiated half duplex but"
3521                                                " link partner cannot autoneg. "
3522                                                " Try forcing full duplex if "
3523                                                "link gets many collisions.\n");
3524                         }
3525
3526                         /*
3527                          * tweak tx_queue_len according to speed/duplex
3528                          * and adjust the timeout factor
3529                          */
3530                         netdev->tx_queue_len = adapter->tx_queue_len;
3531                         adapter->tx_timeout_factor = 1;
3532                         switch (adapter->link_speed) {
3533                         case SPEED_10:
3534                                 txb2b = 0;
3535                                 netdev->tx_queue_len = 10;
3536                                 adapter->tx_timeout_factor = 16;
3537                                 break;
3538                         case SPEED_100:
3539                                 txb2b = 0;
3540                                 netdev->tx_queue_len = 100;
3541                                 /* maybe add some timeout factor ? */
3542                                 break;
3543                         }
3544
3545                         /*
3546                          * workaround: re-program speed mode bit after
3547                          * link-up event
3548                          */
3549                         if ((adapter->flags & FLAG_TARC_SPEED_MODE_BIT) &&
3550                             !txb2b) {
3551                                 u32 tarc0;
3552                                 tarc0 = er32(TARC(0));
3553                                 tarc0 &= ~SPEED_MODE_BIT;
3554                                 ew32(TARC(0), tarc0);
3555                         }
3556
3557                         /*
3558                          * disable TSO for pcie and 10/100 speeds, to avoid
3559                          * some hardware issues
3560                          */
3561                         if (!(adapter->flags & FLAG_TSO_FORCE)) {
3562                                 switch (adapter->link_speed) {
3563                                 case SPEED_10:
3564                                 case SPEED_100:
3565                                         e_info("10/100 speed: disabling TSO\n");
3566                                         netdev->features &= ~NETIF_F_TSO;
3567                                         netdev->features &= ~NETIF_F_TSO6;
3568                                         break;
3569                                 case SPEED_1000:
3570                                         netdev->features |= NETIF_F_TSO;
3571                                         netdev->features |= NETIF_F_TSO6;
3572                                         break;
3573                                 default:
3574                                         /* oops */
3575                                         break;
3576                                 }
3577                         }
3578
3579                         /*
3580                          * enable transmits in the hardware, need to do this
3581                          * after setting TARC(0)
3582                          */
3583                         tctl = er32(TCTL);
3584                         tctl |= E1000_TCTL_EN;
3585                         ew32(TCTL, tctl);
3586
3587                         netif_carrier_on(netdev);
3588                         netif_tx_wake_all_queues(netdev);
3589
3590                         if (!test_bit(__E1000_DOWN, &adapter->state))
3591                                 mod_timer(&adapter->phy_info_timer,
3592                                           round_jiffies(jiffies + 2 * HZ));
3593                 }
3594         } else {
3595                 if (netif_carrier_ok(netdev)) {
3596                         adapter->link_speed = 0;
3597                         adapter->link_duplex = 0;
3598                         e_info("Link is Down\n");
3599                         netif_carrier_off(netdev);
3600                         netif_tx_stop_all_queues(netdev);
3601                         if (!test_bit(__E1000_DOWN, &adapter->state))
3602                                 mod_timer(&adapter->phy_info_timer,
3603                                           round_jiffies(jiffies + 2 * HZ));
3604
3605                         if (adapter->flags & FLAG_RX_NEEDS_RESTART)
3606                                 schedule_work(&adapter->reset_task);
3607                 }
3608         }
3609
3610 link_up:
3611         e1000e_update_stats(adapter);
3612
3613         mac->tx_packet_delta = adapter->stats.tpt - adapter->tpt_old;
3614         adapter->tpt_old = adapter->stats.tpt;
3615         mac->collision_delta = adapter->stats.colc - adapter->colc_old;
3616         adapter->colc_old = adapter->stats.colc;
3617
3618         adapter->gorc = adapter->stats.gorc - adapter->gorc_old;
3619         adapter->gorc_old = adapter->stats.gorc;
3620         adapter->gotc = adapter->stats.gotc - adapter->gotc_old;
3621         adapter->gotc_old = adapter->stats.gotc;
3622
3623         e1000e_update_adaptive(&adapter->hw);
3624
3625         if (!netif_carrier_ok(netdev)) {
3626                 tx_pending = (e1000_desc_unused(tx_ring) + 1 <
3627                                tx_ring->count);
3628                 if (tx_pending) {
3629                         /*
3630                          * We've lost link, so the controller stops DMA,
3631                          * but we've got queued Tx work that's never going
3632                          * to get done, so reset controller to flush Tx.
3633                          * (Do the reset outside of interrupt context).
3634                          */
3635                         adapter->tx_timeout_count++;
3636                         schedule_work(&adapter->reset_task);
3637                 }
3638         }
3639
3640         /* Cause software interrupt to ensure Rx ring is cleaned */
3641         if (adapter->msix_entries)
3642                 ew32(ICS, adapter->rx_ring->ims_val);
3643         else
3644                 ew32(ICS, E1000_ICS_RXDMT0);
3645
3646         /* Force detection of hung controller every watchdog period */
3647         adapter->detect_tx_hung = 1;
3648
3649         /*
3650          * With 82571 controllers, LAA may be overwritten due to controller
3651          * reset from the other port. Set the appropriate LAA in RAR[0]
3652          */
3653         if (e1000e_get_laa_state_82571(hw))
3654                 e1000e_rar_set(hw, adapter->hw.mac.addr, 0);
3655
3656         /* Reset the timer */
3657         if (!test_bit(__E1000_DOWN, &adapter->state))
3658                 mod_timer(&adapter->watchdog_timer,
3659                           round_jiffies(jiffies + 2 * HZ));
3660 }
3661
3662 #define E1000_TX_FLAGS_CSUM             0x00000001
3663 #define E1000_TX_FLAGS_VLAN             0x00000002
3664 #define E1000_TX_FLAGS_TSO              0x00000004
3665 #define E1000_TX_FLAGS_IPV4             0x00000008
3666 #define E1000_TX_FLAGS_VLAN_MASK        0xffff0000
3667 #define E1000_TX_FLAGS_VLAN_SHIFT       16
3668
3669 static int e1000_tso(struct e1000_adapter *adapter,
3670                      struct sk_buff *skb)
3671 {
3672         struct e1000_ring *tx_ring = adapter->tx_ring;
3673         struct e1000_context_desc *context_desc;
3674         struct e1000_buffer *buffer_info;
3675         unsigned int i;
3676         u32 cmd_length = 0;
3677         u16 ipcse = 0, tucse, mss;
3678         u8 ipcss, ipcso, tucss, tucso, hdr_len;
3679         int err;
3680
3681         if (skb_is_gso(skb)) {
3682                 if (skb_header_cloned(skb)) {
3683                         err = pskb_expand_head(skb, 0, 0, GFP_ATOMIC);
3684                         if (err)
3685                                 return err;
3686                 }
3687
3688                 hdr_len = skb_transport_offset(skb) + tcp_hdrlen(skb);
3689                 mss = skb_shinfo(skb)->gso_size;
3690                 if (skb->protocol == htons(ETH_P_IP)) {
3691                         struct iphdr *iph = ip_hdr(skb);
3692                         iph->tot_len = 0;
3693                         iph->check = 0;
3694                         tcp_hdr(skb)->check = ~csum_tcpudp_magic(iph->saddr,
3695                                                                  iph->daddr, 0,
3696                                                                  IPPROTO_TCP,
3697                                                                  0);
3698                         cmd_length = E1000_TXD_CMD_IP;
3699                         ipcse = skb_transport_offset(skb) - 1;
3700                 } else if (skb_shinfo(skb)->gso_type == SKB_GSO_TCPV6) {
3701                         ipv6_hdr(skb)->payload_len = 0;
3702                         tcp_hdr(skb)->check =
3703                                 ~csum_ipv6_magic(&ipv6_hdr(skb)->saddr,
3704                                                  &ipv6_hdr(skb)->daddr,
3705                                                  0, IPPROTO_TCP, 0);
3706                         ipcse = 0;
3707                 }
3708                 ipcss = skb_network_offset(skb);
3709                 ipcso = (void *)&(ip_hdr(skb)->check) - (void *)skb->data;
3710                 tucss = skb_transport_offset(skb);
3711                 tucso = (void *)&(tcp_hdr(skb)->check) - (void *)skb->data;
3712                 tucse = 0;
3713
3714                 cmd_length |= (E1000_TXD_CMD_DEXT | E1000_TXD_CMD_TSE |
3715                                E1000_TXD_CMD_TCP | (skb->len - (hdr_len)));
3716
3717                 i = tx_ring->next_to_use;
3718                 context_desc = E1000_CONTEXT_DESC(*tx_ring, i);
3719                 buffer_info = &tx_ring->buffer_info[i];
3720
3721                 context_desc->lower_setup.ip_fields.ipcss  = ipcss;
3722                 context_desc->lower_setup.ip_fields.ipcso  = ipcso;
3723                 context_desc->lower_setup.ip_fields.ipcse  = cpu_to_le16(ipcse);
3724                 context_desc->upper_setup.tcp_fields.tucss = tucss;
3725                 context_desc->upper_setup.tcp_fields.tucso = tucso;
3726                 context_desc->upper_setup.tcp_fields.tucse = cpu_to_le16(tucse);
3727                 context_desc->tcp_seg_setup.fields.mss     = cpu_to_le16(mss);
3728                 context_desc->tcp_seg_setup.fields.hdr_len = hdr_len;
3729                 context_desc->cmd_and_length = cpu_to_le32(cmd_length);
3730
3731                 buffer_info->time_stamp = jiffies;
3732                 buffer_info->next_to_watch = i;
3733
3734                 i++;
3735                 if (i == tx_ring->count)
3736                         i = 0;
3737                 tx_ring->next_to_use = i;
3738
3739                 return 1;
3740         }
3741
3742         return 0;
3743 }
3744
3745 static bool e1000_tx_csum(struct e1000_adapter *adapter, struct sk_buff *skb)
3746 {
3747         struct e1000_ring *tx_ring = adapter->tx_ring;
3748         struct e1000_context_desc *context_desc;
3749         struct e1000_buffer *buffer_info;
3750         unsigned int i;
3751         u8 css;
3752         u32 cmd_len = E1000_TXD_CMD_DEXT;
3753
3754         if (skb->ip_summed != CHECKSUM_PARTIAL)
3755                 return 0;
3756
3757         switch (skb->protocol) {
3758         case __constant_htons(ETH_P_IP):
3759                 if (ip_hdr(skb)->protocol == IPPROTO_TCP)
3760                         cmd_len |= E1000_TXD_CMD_TCP;
3761                 break;
3762         case __constant_htons(ETH_P_IPV6):
3763                 /* XXX not handling all IPV6 headers */
3764                 if (ipv6_hdr(skb)->nexthdr == IPPROTO_TCP)
3765                         cmd_len |= E1000_TXD_CMD_TCP;
3766                 break;
3767         default:
3768                 if (unlikely(net_ratelimit()))
3769                         e_warn("checksum_partial proto=%x!\n", skb->protocol);
3770                 break;
3771         }
3772
3773         css = skb_transport_offset(skb);
3774
3775         i = tx_ring->next_to_use;
3776         buffer_info = &tx_ring->buffer_info[i];
3777         context_desc = E1000_CONTEXT_DESC(*tx_ring, i);
3778
3779         context_desc->lower_setup.ip_config = 0;
3780         context_desc->upper_setup.tcp_fields.tucss = css;
3781         context_desc->upper_setup.tcp_fields.tucso =
3782                                 css + skb->csum_offset;
3783         context_desc->upper_setup.tcp_fields.tucse = 0;
3784         context_desc->tcp_seg_setup.data = 0;
3785         context_desc->cmd_and_length = cpu_to_le32(cmd_len);
3786
3787         buffer_info->time_stamp = jiffies;
3788         buffer_info->next_to_watch = i;
3789
3790         i++;
3791         if (i == tx_ring->count)
3792                 i = 0;
3793         tx_ring->next_to_use = i;
3794
3795         return 1;
3796 }
3797
3798 #define E1000_MAX_PER_TXD       8192
3799 #define E1000_MAX_TXD_PWR       12
3800
3801 static int e1000_tx_map(struct e1000_adapter *adapter,
3802                         struct sk_buff *skb, unsigned int first,
3803                         unsigned int max_per_txd, unsigned int nr_frags,
3804                         unsigned int mss)
3805 {
3806         struct e1000_ring *tx_ring = adapter->tx_ring;
3807         struct e1000_buffer *buffer_info;
3808         unsigned int len = skb->len - skb->data_len;
3809         unsigned int offset = 0, size, count = 0, i;
3810         unsigned int f;
3811
3812         i = tx_ring->next_to_use;
3813
3814         while (len) {
3815                 buffer_info = &tx_ring->buffer_info[i];
3816                 size = min(len, max_per_txd);
3817
3818                 /* Workaround for premature desc write-backs
3819                  * in TSO mode.  Append 4-byte sentinel desc */
3820                 if (mss && !nr_frags && size == len && size > 8)
3821                         size -= 4;
3822
3823                 buffer_info->length = size;
3824                 /* set time_stamp *before* dma to help avoid a possible race */
3825                 buffer_info->time_stamp = jiffies;
3826                 buffer_info->dma =
3827                         pci_map_single(adapter->pdev,
3828                                 skb->data + offset,
3829                                 size,
3830                                 PCI_DMA_TODEVICE);
3831                 if (pci_dma_mapping_error(adapter->pdev, buffer_info->dma)) {
3832                         dev_err(&adapter->pdev->dev, "TX DMA map failed\n");
3833                         adapter->tx_dma_failed++;
3834                         return -1;
3835                 }
3836                 buffer_info->next_to_watch = i;
3837
3838                 len -= size;
3839                 offset += size;
3840                 count++;
3841                 i++;
3842                 if (i == tx_ring->count)
3843                         i = 0;
3844         }
3845
3846         for (f = 0; f < nr_frags; f++) {
3847                 struct skb_frag_struct *frag;
3848
3849                 frag = &skb_shinfo(skb)->frags[f];
3850                 len = frag->size;
3851                 offset = frag->page_offset;
3852
3853                 while (len) {
3854                         buffer_info = &tx_ring->buffer_info[i];
3855                         size = min(len, max_per_txd);
3856                         /* Workaround for premature desc write-backs
3857                          * in TSO mode.  Append 4-byte sentinel desc */
3858                         if (mss && f == (nr_frags-1) && size == len && size > 8)
3859                                 size -= 4;
3860
3861                         buffer_info->length = size;
3862                         buffer_info->time_stamp = jiffies;
3863                         buffer_info->dma =
3864                                 pci_map_page(adapter->pdev,
3865                                         frag->page,
3866                                         offset,
3867                                         size,
3868                                         PCI_DMA_TODEVICE);
3869                         if (pci_dma_mapping_error(adapter->pdev,
3870                                                   buffer_info->dma)) {
3871                                 dev_err(&adapter->pdev->dev,
3872                                         "TX DMA page map failed\n");
3873                                 adapter->tx_dma_failed++;
3874                                 return -1;
3875                         }
3876
3877                         buffer_info->next_to_watch = i;
3878
3879                         len -= size;
3880                         offset += size;
3881                         count++;
3882
3883                         i++;
3884                         if (i == tx_ring->count)
3885                                 i = 0;
3886                 }
3887         }
3888
3889         if (i == 0)
3890                 i = tx_ring->count - 1;
3891         else
3892                 i--;
3893
3894         tx_ring->buffer_info[i].skb = skb;
3895         tx_ring->buffer_info[first].next_to_watch = i;
3896
3897         return count;
3898 }
3899
3900 static void e1000_tx_queue(struct e1000_adapter *adapter,
3901                            int tx_flags, int count)
3902 {
3903         struct e1000_ring *tx_ring = adapter->tx_ring;
3904         struct e1000_tx_desc *tx_desc = NULL;
3905         struct e1000_buffer *buffer_info;
3906         u32 txd_upper = 0, txd_lower = E1000_TXD_CMD_IFCS;
3907         unsigned int i;
3908
3909         if (tx_flags & E1000_TX_FLAGS_TSO) {
3910                 txd_lower |= E1000_TXD_CMD_DEXT | E1000_TXD_DTYP_D |
3911                              E1000_TXD_CMD_TSE;
3912                 txd_upper |= E1000_TXD_POPTS_TXSM << 8;
3913
3914                 if (tx_flags & E1000_TX_FLAGS_IPV4)
3915                         txd_upper |= E1000_TXD_POPTS_IXSM << 8;
3916         }
3917
3918         if (tx_flags & E1000_TX_FLAGS_CSUM) {
3919                 txd_lower |= E1000_TXD_CMD_DEXT | E1000_TXD_DTYP_D;
3920                 txd_upper |= E1000_TXD_POPTS_TXSM << 8;
3921         }
3922
3923         if (tx_flags & E1000_TX_FLAGS_VLAN) {
3924                 txd_lower |= E1000_TXD_CMD_VLE;
3925                 txd_upper |= (tx_flags & E1000_TX_FLAGS_VLAN_MASK);
3926         }
3927
3928         i = tx_ring->next_to_use;
3929
3930         while (count--) {
3931                 buffer_info = &tx_ring->buffer_info[i];
3932                 tx_desc = E1000_TX_DESC(*tx_ring, i);
3933                 tx_desc->buffer_addr = cpu_to_le64(buffer_info->dma);
3934                 tx_desc->lower.data =
3935                         cpu_to_le32(txd_lower | buffer_info->length);
3936                 tx_desc->upper.data = cpu_to_le32(txd_upper);
3937
3938                 i++;
3939                 if (i == tx_ring->count)
3940                         i = 0;
3941         }
3942
3943         tx_desc->lower.data |= cpu_to_le32(adapter->txd_cmd);
3944
3945         /*
3946          * Force memory writes to complete before letting h/w
3947          * know there are new descriptors to fetch.  (Only
3948          * applicable for weak-ordered memory model archs,
3949          * such as IA-64).
3950          */
3951         wmb();
3952
3953         tx_ring->next_to_use = i;
3954         writel(i, adapter->hw.hw_addr + tx_ring->tail);
3955         /*
3956          * we need this if more than one processor can write to our tail
3957          * at a time, it synchronizes IO on IA64/Altix systems
3958          */
3959         mmiowb();
3960 }
3961
3962 #define MINIMUM_DHCP_PACKET_SIZE 282
3963 static int e1000_transfer_dhcp_info(struct e1000_adapter *adapter,
3964                                     struct sk_buff *skb)
3965 {
3966         struct e1000_hw *hw =  &adapter->hw;
3967         u16 length, offset;
3968
3969         if (vlan_tx_tag_present(skb)) {
3970                 if (!((vlan_tx_tag_get(skb) == adapter->hw.mng_cookie.vlan_id)
3971                     && (adapter->hw.mng_cookie.status &
3972                         E1000_MNG_DHCP_COOKIE_STATUS_VLAN)))
3973                         return 0;
3974         }
3975
3976         if (skb->len <= MINIMUM_DHCP_PACKET_SIZE)
3977                 return 0;
3978
3979         if (((struct ethhdr *) skb->data)->h_proto != htons(ETH_P_IP))
3980                 return 0;
3981
3982         {
3983                 const struct iphdr *ip = (struct iphdr *)((u8 *)skb->data+14);
3984                 struct udphdr *udp;
3985
3986                 if (ip->protocol != IPPROTO_UDP)
3987                         return 0;
3988
3989                 udp = (struct udphdr *)((u8 *)ip + (ip->ihl << 2));
3990                 if (ntohs(udp->dest) != 67)
3991                         return 0;
3992
3993                 offset = (u8 *)udp + 8 - skb->data;
3994                 length = skb->len - offset;
3995                 return e1000e_mng_write_dhcp_info(hw, (u8 *)udp + 8, length);
3996         }
3997
3998         return 0;
3999 }
4000
4001 static int __e1000_maybe_stop_tx(struct net_device *netdev, int size)
4002 {
4003         struct e1000_adapter *adapter = netdev_priv(netdev);
4004
4005         netif_stop_queue(netdev);
4006         /*
4007          * Herbert's original patch had:
4008          *  smp_mb__after_netif_stop_queue();
4009          * but since that doesn't exist yet, just open code it.
4010          */
4011         smp_mb();
4012
4013         /*
4014          * We need to check again in a case another CPU has just
4015          * made room available.
4016          */
4017         if (e1000_desc_unused(adapter->tx_ring) < size)
4018                 return -EBUSY;
4019
4020         /* A reprieve! */
4021         netif_start_queue(netdev);
4022         ++adapter->restart_queue;
4023         return 0;
4024 }
4025
4026 static int e1000_maybe_stop_tx(struct net_device *netdev, int size)
4027 {
4028         struct e1000_adapter *adapter = netdev_priv(netdev);
4029
4030         if (e1000_desc_unused(adapter->tx_ring) >= size)
4031                 return 0;
4032         return __e1000_maybe_stop_tx(netdev, size);
4033 }
4034
4035 #define TXD_USE_COUNT(S, X) (((S) >> (X)) + 1 )
4036 static int e1000_xmit_frame(struct sk_buff *skb, struct net_device *netdev)
4037 {
4038         struct e1000_adapter *adapter = netdev_priv(netdev);
4039         struct e1000_ring *tx_ring = adapter->tx_ring;
4040         unsigned int first;
4041         unsigned int max_per_txd = E1000_MAX_PER_TXD;
4042         unsigned int max_txd_pwr = E1000_MAX_TXD_PWR;
4043         unsigned int tx_flags = 0;
4044         unsigned int len = skb->len - skb->data_len;
4045         unsigned long irq_flags;
4046         unsigned int nr_frags;
4047         unsigned int mss;
4048         int count = 0;
4049         int tso;
4050         unsigned int f;
4051
4052         if (test_bit(__E1000_DOWN, &adapter->state)) {
4053                 dev_kfree_skb_any(skb);
4054                 return NETDEV_TX_OK;
4055         }
4056
4057         if (skb->len <= 0) {
4058                 dev_kfree_skb_any(skb);
4059                 return NETDEV_TX_OK;
4060         }
4061
4062         mss = skb_shinfo(skb)->gso_size;
4063         /*
4064          * The controller does a simple calculation to
4065          * make sure there is enough room in the FIFO before
4066          * initiating the DMA for each buffer.  The calc is:
4067          * 4 = ceil(buffer len/mss).  To make sure we don't
4068          * overrun the FIFO, adjust the max buffer len if mss
4069          * drops.
4070          */
4071         if (mss) {
4072                 u8 hdr_len;
4073                 max_per_txd = min(mss << 2, max_per_txd);
4074                 max_txd_pwr = fls(max_per_txd) - 1;
4075
4076                 /*
4077                  * TSO Workaround for 82571/2/3 Controllers -- if skb->data
4078                  * points to just header, pull a few bytes of payload from
4079                  * frags into skb->data
4080                  */
4081                 hdr_len = skb_transport_offset(skb) + tcp_hdrlen(skb);
4082                 /*
4083                  * we do this workaround for ES2LAN, but it is un-necessary,
4084                  * avoiding it could save a lot of cycles
4085                  */
4086                 if (skb->data_len && (hdr_len == len)) {
4087                         unsigned int pull_size;
4088
4089                         pull_size = min((unsigned int)4, skb->data_len);
4090                         if (!__pskb_pull_tail(skb, pull_size)) {
4091                                 e_err("__pskb_pull_tail failed.\n");
4092                                 dev_kfree_skb_any(skb);
4093                                 return NETDEV_TX_OK;
4094                         }
4095                         len = skb->len - skb->data_len;
4096                 }
4097         }
4098
4099         /* reserve a descriptor for the offload context */
4100         if ((mss) || (skb->ip_summed == CHECKSUM_PARTIAL))
4101                 count++;
4102         count++;
4103
4104         count += TXD_USE_COUNT(len, max_txd_pwr);
4105
4106         nr_frags = skb_shinfo(skb)->nr_frags;
4107         for (f = 0; f < nr_frags; f++)
4108                 count += TXD_USE_COUNT(skb_shinfo(skb)->frags[f].size,
4109                                        max_txd_pwr);
4110
4111         if (adapter->hw.mac.tx_pkt_filtering)
4112                 e1000_transfer_dhcp_info(adapter, skb);
4113
4114         if (!spin_trylock_irqsave(&adapter->tx_queue_lock, irq_flags))
4115                 /* Collision - tell upper layer to requeue */
4116                 return NETDEV_TX_LOCKED;
4117
4118         /*
4119          * need: count + 2 desc gap to keep tail from touching
4120          * head, otherwise try next time
4121          */
4122         if (e1000_maybe_stop_tx(netdev, count + 2)) {
4123                 spin_unlock_irqrestore(&adapter->tx_queue_lock, irq_flags);
4124                 return NETDEV_TX_BUSY;
4125         }
4126
4127         if (adapter->vlgrp && vlan_tx_tag_present(skb)) {
4128                 tx_flags |= E1000_TX_FLAGS_VLAN;
4129                 tx_flags |= (vlan_tx_tag_get(skb) << E1000_TX_FLAGS_VLAN_SHIFT);
4130         }
4131
4132         first = tx_ring->next_to_use;
4133
4134         tso = e1000_tso(adapter, skb);
4135         if (tso < 0) {
4136                 dev_kfree_skb_any(skb);
4137                 spin_unlock_irqrestore(&adapter->tx_queue_lock, irq_flags);
4138                 return NETDEV_TX_OK;
4139         }
4140
4141         if (tso)
4142                 tx_flags |= E1000_TX_FLAGS_TSO;
4143         else if (e1000_tx_csum(adapter, skb))
4144                 tx_flags |= E1000_TX_FLAGS_CSUM;
4145
4146         /*
4147          * Old method was to assume IPv4 packet by default if TSO was enabled.
4148          * 82571 hardware supports TSO capabilities for IPv6 as well...
4149          * no longer assume, we must.
4150          */
4151         if (skb->protocol == htons(ETH_P_IP))
4152                 tx_flags |= E1000_TX_FLAGS_IPV4;
4153
4154         count = e1000_tx_map(adapter, skb, first, max_per_txd, nr_frags, mss);
4155         if (count < 0) {
4156                 /* handle pci_map_single() error in e1000_tx_map */
4157                 dev_kfree_skb_any(skb);
4158                 spin_unlock_irqrestore(&adapter->tx_queue_lock, irq_flags);
4159                 return NETDEV_TX_OK;
4160         }
4161
4162         e1000_tx_queue(adapter, tx_flags, count);
4163
4164         netdev->trans_start = jiffies;
4165
4166         /* Make sure there is space in the ring for the next send. */
4167         e1000_maybe_stop_tx(netdev, MAX_SKB_FRAGS + 2);
4168
4169         spin_unlock_irqrestore(&adapter->tx_queue_lock, irq_flags);
4170         return NETDEV_TX_OK;
4171 }
4172
4173 /**
4174  * e1000_tx_timeout - Respond to a Tx Hang
4175  * @netdev: network interface device structure
4176  **/
4177 static void e1000_tx_timeout(struct net_device *netdev)
4178 {
4179         struct e1000_adapter *adapter = netdev_priv(netdev);
4180
4181         /* Do the reset outside of interrupt context */
4182         adapter->tx_timeout_count++;
4183         schedule_work(&adapter->reset_task);
4184 }
4185
4186 static void e1000_reset_task(struct work_struct *work)
4187 {
4188         struct e1000_adapter *adapter;
4189         adapter = container_of(work, struct e1000_adapter, reset_task);
4190
4191         e1000e_reinit_locked(adapter);
4192 }
4193
4194 /**
4195  * e1000_get_stats - Get System Network Statistics
4196  * @netdev: network interface device structure
4197  *
4198  * Returns the address of the device statistics structure.
4199  * The statistics are actually updated from the timer callback.
4200  **/
4201 static struct net_device_stats *e1000_get_stats(struct net_device *netdev)
4202 {
4203         struct e1000_adapter *adapter = netdev_priv(netdev);
4204
4205         /* only return the current stats */
4206         return &adapter->net_stats;
4207 }
4208
4209 /**
4210  * e1000_change_mtu - Change the Maximum Transfer Unit
4211  * @netdev: network interface device structure
4212  * @new_mtu: new value for maximum frame size
4213  *
4214  * Returns 0 on success, negative on failure
4215  **/
4216 static int e1000_change_mtu(struct net_device *netdev, int new_mtu)
4217 {
4218         struct e1000_adapter *adapter = netdev_priv(netdev);
4219         int max_frame = new_mtu + ETH_HLEN + ETH_FCS_LEN;
4220
4221         if ((new_mtu < ETH_ZLEN + ETH_FCS_LEN + VLAN_HLEN) ||
4222             (max_frame > MAX_JUMBO_FRAME_SIZE)) {
4223                 e_err("Invalid MTU setting\n");
4224                 return -EINVAL;
4225         }
4226
4227         /* Jumbo frame size limits */
4228         if (max_frame > ETH_FRAME_LEN + ETH_FCS_LEN) {
4229                 if (!(adapter->flags & FLAG_HAS_JUMBO_FRAMES)) {
4230                         e_err("Jumbo Frames not supported.\n");
4231                         return -EINVAL;
4232                 }
4233                 if (adapter->hw.phy.type == e1000_phy_ife) {
4234                         e_err("Jumbo Frames not supported.\n");
4235                         return -EINVAL;
4236                 }
4237         }
4238
4239 #define MAX_STD_JUMBO_FRAME_SIZE 9234
4240         if (max_frame > MAX_STD_JUMBO_FRAME_SIZE) {
4241                 e_err("MTU > 9216 not supported.\n");
4242                 return -EINVAL;
4243         }
4244
4245         while (test_and_set_bit(__E1000_RESETTING, &adapter->state))
4246                 msleep(1);
4247         /* e1000e_down has a dependency on max_frame_size */
4248         adapter->max_frame_size = max_frame;
4249         if (netif_running(netdev))
4250                 e1000e_down(adapter);
4251
4252         /*
4253          * NOTE: netdev_alloc_skb reserves 16 bytes, and typically NET_IP_ALIGN
4254          * means we reserve 2 more, this pushes us to allocate from the next
4255          * larger slab size.
4256          * i.e. RXBUFFER_2048 --> size-4096 slab
4257          * However with the new *_jumbo_rx* routines, jumbo receives will use
4258          * fragmented skbs
4259          */
4260
4261         if (max_frame <= 256)
4262                 adapter->rx_buffer_len = 256;
4263         else if (max_frame <= 512)
4264                 adapter->rx_buffer_len = 512;
4265         else if (max_frame <= 1024)
4266                 adapter->rx_buffer_len = 1024;
4267         else if (max_frame <= 2048)
4268                 adapter->rx_buffer_len = 2048;
4269         else
4270                 adapter->rx_buffer_len = 4096;
4271
4272         /* adjust allocation if LPE protects us, and we aren't using SBP */
4273         if ((max_frame == ETH_FRAME_LEN + ETH_FCS_LEN) ||
4274              (max_frame == ETH_FRAME_LEN + VLAN_HLEN + ETH_FCS_LEN))
4275                 adapter->rx_buffer_len = ETH_FRAME_LEN + VLAN_HLEN
4276                                          + ETH_FCS_LEN;
4277
4278         e_info("changing MTU from %d to %d\n", netdev->mtu, new_mtu);
4279         netdev->mtu = new_mtu;
4280
4281         if (netif_running(netdev))
4282                 e1000e_up(adapter);
4283         else
4284                 e1000e_reset(adapter);
4285
4286         clear_bit(__E1000_RESETTING, &adapter->state);
4287
4288         return 0;
4289 }
4290
4291 static int e1000_mii_ioctl(struct net_device *netdev, struct ifreq *ifr,
4292                            int cmd)
4293 {
4294         struct e1000_adapter *adapter = netdev_priv(netdev);
4295         struct mii_ioctl_data *data = if_mii(ifr);
4296
4297         if (adapter->hw.phy.media_type != e1000_media_type_copper)
4298                 return -EOPNOTSUPP;
4299
4300         switch (cmd) {
4301         case SIOCGMIIPHY:
4302                 data->phy_id = adapter->hw.phy.addr;
4303                 break;
4304         case SIOCGMIIREG:
4305                 if (!capable(CAP_NET_ADMIN))
4306                         return -EPERM;
4307                 switch (data->reg_num & 0x1F) {
4308                 case MII_BMCR:
4309                         data->val_out = adapter->phy_regs.bmcr;
4310                         break;
4311                 case MII_BMSR:
4312                         data->val_out = adapter->phy_regs.bmsr;
4313                         break;
4314                 case MII_PHYSID1:
4315                         data->val_out = (adapter->hw.phy.id >> 16);
4316                         break;
4317                 case MII_PHYSID2:
4318                         data->val_out = (adapter->hw.phy.id & 0xFFFF);
4319                         break;
4320                 case MII_ADVERTISE:
4321                         data->val_out = adapter->phy_regs.advertise;
4322                         break;
4323                 case MII_LPA:
4324                         data->val_out = adapter->phy_regs.lpa;
4325                         break;
4326                 case MII_EXPANSION:
4327                         data->val_out = adapter->phy_regs.expansion;
4328                         break;
4329                 case MII_CTRL1000:
4330                         data->val_out = adapter->phy_regs.ctrl1000;
4331                         break;
4332                 case MII_STAT1000:
4333                         data->val_out = adapter->phy_regs.stat1000;
4334                         break;
4335                 case MII_ESTATUS:
4336                         data->val_out = adapter->phy_regs.estatus;
4337                         break;
4338                 default:
4339                         return -EIO;
4340                 }
4341                 break;
4342         case SIOCSMIIREG:
4343         default:
4344                 return -EOPNOTSUPP;
4345         }
4346         return 0;
4347 }
4348
4349 static int e1000_ioctl(struct net_device *netdev, struct ifreq *ifr, int cmd)
4350 {
4351         switch (cmd) {
4352         case SIOCGMIIPHY:
4353         case SIOCGMIIREG:
4354         case SIOCSMIIREG:
4355                 return e1000_mii_ioctl(netdev, ifr, cmd);
4356         default:
4357                 return -EOPNOTSUPP;
4358         }
4359 }
4360
4361 static int e1000_suspend(struct pci_dev *pdev, pm_message_t state)
4362 {
4363         struct net_device *netdev = pci_get_drvdata(pdev);
4364         struct e1000_adapter *adapter = netdev_priv(netdev);
4365         struct e1000_hw *hw = &adapter->hw;
4366         u32 ctrl, ctrl_ext, rctl, status;
4367         u32 wufc = adapter->wol;
4368         int retval = 0;
4369
4370         netif_device_detach(netdev);
4371
4372         if (netif_running(netdev)) {
4373                 WARN_ON(test_bit(__E1000_RESETTING, &adapter->state));
4374                 e1000e_down(adapter);
4375                 e1000_free_irq(adapter);
4376         }
4377         e1000e_reset_interrupt_capability(adapter);
4378
4379         retval = pci_save_state(pdev);
4380         if (retval)
4381                 return retval;
4382
4383         status = er32(STATUS);
4384         if (status & E1000_STATUS_LU)
4385                 wufc &= ~E1000_WUFC_LNKC;
4386
4387         if (wufc) {
4388                 e1000_setup_rctl(adapter);
4389                 e1000_set_multi(netdev);
4390
4391                 /* turn on all-multi mode if wake on multicast is enabled */
4392                 if (wufc & E1000_WUFC_MC) {
4393                         rctl = er32(RCTL);
4394                         rctl |= E1000_RCTL_MPE;
4395                         ew32(RCTL, rctl);
4396                 }
4397
4398                 ctrl = er32(CTRL);
4399                 /* advertise wake from D3Cold */
4400                 #define E1000_CTRL_ADVD3WUC 0x00100000
4401                 /* phy power management enable */
4402                 #define E1000_CTRL_EN_PHY_PWR_MGMT 0x00200000
4403                 ctrl |= E1000_CTRL_ADVD3WUC |
4404                         E1000_CTRL_EN_PHY_PWR_MGMT;
4405                 ew32(CTRL, ctrl);
4406
4407                 if (adapter->hw.phy.media_type == e1000_media_type_fiber ||
4408                     adapter->hw.phy.media_type ==
4409                     e1000_media_type_internal_serdes) {
4410                         /* keep the laser running in D3 */
4411                         ctrl_ext = er32(CTRL_EXT);
4412                         ctrl_ext |= E1000_CTRL_EXT_SDP7_DATA;
4413                         ew32(CTRL_EXT, ctrl_ext);
4414                 }
4415
4416                 if (adapter->flags & FLAG_IS_ICH)
4417                         e1000e_disable_gig_wol_ich8lan(&adapter->hw);
4418
4419                 /* Allow time for pending master requests to run */
4420                 e1000e_disable_pcie_master(&adapter->hw);
4421
4422                 ew32(WUC, E1000_WUC_PME_EN);
4423                 ew32(WUFC, wufc);
4424                 pci_enable_wake(pdev, PCI_D3hot, 1);
4425                 pci_enable_wake(pdev, PCI_D3cold, 1);
4426         } else {
4427                 ew32(WUC, 0);
4428                 ew32(WUFC, 0);
4429                 pci_enable_wake(pdev, PCI_D3hot, 0);
4430                 pci_enable_wake(pdev, PCI_D3cold, 0);
4431         }
4432
4433         /* make sure adapter isn't asleep if manageability is enabled */
4434         if (adapter->flags & FLAG_MNG_PT_ENABLED) {
4435                 pci_enable_wake(pdev, PCI_D3hot, 1);
4436                 pci_enable_wake(pdev, PCI_D3cold, 1);
4437         }
4438
4439         if (adapter->hw.phy.type == e1000_phy_igp_3)
4440                 e1000e_igp3_phy_powerdown_workaround_ich8lan(&adapter->hw);
4441
4442         /*
4443          * Release control of h/w to f/w.  If f/w is AMT enabled, this
4444          * would have already happened in close and is redundant.
4445          */
4446         e1000_release_hw_control(adapter);
4447
4448         pci_disable_device(pdev);
4449
4450         pci_set_power_state(pdev, pci_choose_state(pdev, state));
4451
4452         return 0;
4453 }
4454
4455 static void e1000e_disable_l1aspm(struct pci_dev *pdev)
4456 {
4457         int pos;
4458         u16 val;
4459
4460         /*
4461          * 82573 workaround - disable L1 ASPM on mobile chipsets
4462          *
4463          * L1 ASPM on various mobile (ich7) chipsets do not behave properly
4464          * resulting in lost data or garbage information on the pci-e link
4465          * level. This could result in (false) bad EEPROM checksum errors,
4466          * long ping times (up to 2s) or even a system freeze/hang.
4467          *
4468          * Unfortunately this feature saves about 1W power consumption when
4469          * active.
4470          */
4471         pos = pci_find_capability(pdev, PCI_CAP_ID_EXP);
4472         pci_read_config_word(pdev, pos + PCI_EXP_LNKCTL, &val);
4473         if (val & 0x2) {
4474                 dev_warn(&pdev->dev, "Disabling L1 ASPM\n");
4475                 val &= ~0x2;
4476                 pci_write_config_word(pdev, pos + PCI_EXP_LNKCTL, val);
4477         }
4478 }
4479
4480 #ifdef CONFIG_PM
4481 static int e1000_resume(struct pci_dev *pdev)
4482 {
4483         struct net_device *netdev = pci_get_drvdata(pdev);
4484         struct e1000_adapter *adapter = netdev_priv(netdev);
4485         struct e1000_hw *hw = &adapter->hw;
4486         u32 err;
4487
4488         pci_set_power_state(pdev, PCI_D0);
4489         pci_restore_state(pdev);
4490         e1000e_disable_l1aspm(pdev);
4491
4492         err = pci_enable_device_mem(pdev);
4493         if (err) {
4494                 dev_err(&pdev->dev,
4495                         "Cannot enable PCI device from suspend\n");
4496                 return err;
4497         }
4498
4499         pci_set_master(pdev);
4500
4501         pci_enable_wake(pdev, PCI_D3hot, 0);
4502         pci_enable_wake(pdev, PCI_D3cold, 0);
4503
4504         e1000e_set_interrupt_capability(adapter);
4505         if (netif_running(netdev)) {
4506                 err = e1000_request_irq(adapter);
4507                 if (err)
4508                         return err;
4509         }
4510
4511         e1000e_power_up_phy(adapter);
4512         e1000e_reset(adapter);
4513         ew32(WUS, ~0);
4514
4515         e1000_init_manageability(adapter);
4516
4517         if (netif_running(netdev))
4518                 e1000e_up(adapter);
4519
4520         netif_device_attach(netdev);
4521
4522         /*
4523          * If the controller has AMT, do not set DRV_LOAD until the interface
4524          * is up.  For all other cases, let the f/w know that the h/w is now
4525          * under the control of the driver.
4526          */
4527         if (!(adapter->flags & FLAG_HAS_AMT))
4528                 e1000_get_hw_control(adapter);
4529
4530         return 0;
4531 }
4532 #endif
4533
4534 static void e1000_shutdown(struct pci_dev *pdev)
4535 {
4536         e1000_suspend(pdev, PMSG_SUSPEND);
4537 }
4538
4539 #ifdef CONFIG_NET_POLL_CONTROLLER
4540 /*
4541  * Polling 'interrupt' - used by things like netconsole to send skbs
4542  * without having to re-enable interrupts. It's not called while
4543  * the interrupt routine is executing.
4544  */
4545 static void e1000_netpoll(struct net_device *netdev)
4546 {
4547         struct e1000_adapter *adapter = netdev_priv(netdev);
4548
4549         disable_irq(adapter->pdev->irq);
4550         e1000_intr(adapter->pdev->irq, netdev);
4551
4552         enable_irq(adapter->pdev->irq);
4553 }
4554 #endif
4555
4556 /**
4557  * e1000_io_error_detected - called when PCI error is detected
4558  * @pdev: Pointer to PCI device
4559  * @state: The current pci connection state
4560  *
4561  * This function is called after a PCI bus error affecting
4562  * this device has been detected.
4563  */
4564 static pci_ers_result_t e1000_io_error_detected(struct pci_dev *pdev,
4565                                                 pci_channel_state_t state)
4566 {
4567         struct net_device *netdev = pci_get_drvdata(pdev);
4568         struct e1000_adapter *adapter = netdev_priv(netdev);
4569
4570         netif_device_detach(netdev);
4571
4572         if (netif_running(netdev))
4573                 e1000e_down(adapter);
4574         pci_disable_device(pdev);
4575
4576         /* Request a slot slot reset. */
4577         return PCI_ERS_RESULT_NEED_RESET;
4578 }
4579
4580 /**
4581  * e1000_io_slot_reset - called after the pci bus has been reset.
4582  * @pdev: Pointer to PCI device
4583  *
4584  * Restart the card from scratch, as if from a cold-boot. Implementation
4585  * resembles the first-half of the e1000_resume routine.
4586  */
4587 static pci_ers_result_t e1000_io_slot_reset(struct pci_dev *pdev)
4588 {
4589         struct net_device *netdev = pci_get_drvdata(pdev);
4590         struct e1000_adapter *adapter = netdev_priv(netdev);
4591         struct e1000_hw *hw = &adapter->hw;
4592         int err;
4593
4594         e1000e_disable_l1aspm(pdev);
4595         err = pci_enable_device_mem(pdev);
4596         if (err) {
4597                 dev_err(&pdev->dev,
4598                         "Cannot re-enable PCI device after reset.\n");
4599                 return PCI_ERS_RESULT_DISCONNECT;
4600         }
4601         pci_set_master(pdev);
4602         pci_restore_state(pdev);
4603
4604         pci_enable_wake(pdev, PCI_D3hot, 0);
4605         pci_enable_wake(pdev, PCI_D3cold, 0);
4606
4607         e1000e_reset(adapter);
4608         ew32(WUS, ~0);
4609
4610         return PCI_ERS_RESULT_RECOVERED;
4611 }
4612
4613 /**
4614  * e1000_io_resume - called when traffic can start flowing again.
4615  * @pdev: Pointer to PCI device
4616  *
4617  * This callback is called when the error recovery driver tells us that
4618  * its OK to resume normal operation. Implementation resembles the
4619  * second-half of the e1000_resume routine.
4620  */
4621 static void e1000_io_resume(struct pci_dev *pdev)
4622 {
4623         struct net_device *netdev = pci_get_drvdata(pdev);
4624         struct e1000_adapter *adapter = netdev_priv(netdev);
4625
4626         e1000_init_manageability(adapter);
4627
4628         if (netif_running(netdev)) {
4629                 if (e1000e_up(adapter)) {
4630                         dev_err(&pdev->dev,
4631                                 "can't bring device back up after reset\n");
4632                         return;
4633                 }
4634         }
4635
4636         netif_device_attach(netdev);
4637
4638         /*
4639          * If the controller has AMT, do not set DRV_LOAD until the interface
4640          * is up.  For all other cases, let the f/w know that the h/w is now
4641          * under the control of the driver.
4642          */
4643         if (!(adapter->flags & FLAG_HAS_AMT))
4644                 e1000_get_hw_control(adapter);
4645
4646 }
4647
4648 static void e1000_print_device_info(struct e1000_adapter *adapter)
4649 {
4650         struct e1000_hw *hw = &adapter->hw;
4651         struct net_device *netdev = adapter->netdev;
4652         u32 pba_num;
4653
4654         /* print bus type/speed/width info */
4655         e_info("(PCI Express:2.5GB/s:%s) %02x:%02x:%02x:%02x:%02x:%02x\n",
4656                /* bus width */
4657                ((hw->bus.width == e1000_bus_width_pcie_x4) ? "Width x4" :
4658                 "Width x1"),
4659                /* MAC address */
4660                netdev->dev_addr[0], netdev->dev_addr[1],
4661                netdev->dev_addr[2], netdev->dev_addr[3],
4662                netdev->dev_addr[4], netdev->dev_addr[5]);
4663         e_info("Intel(R) PRO/%s Network Connection\n",
4664                (hw->phy.type == e1000_phy_ife) ? "10/100" : "1000");
4665         e1000e_read_pba_num(hw, &pba_num);
4666         e_info("MAC: %d, PHY: %d, PBA No: %06x-%03x\n",
4667                hw->mac.type, hw->phy.type, (pba_num >> 8), (pba_num & 0xff));
4668 }
4669
4670 static void e1000_eeprom_checks(struct e1000_adapter *adapter)
4671 {
4672         struct e1000_hw *hw = &adapter->hw;
4673         int ret_val;
4674         u16 buf = 0;
4675
4676         if (hw->mac.type != e1000_82573)
4677                 return;
4678
4679         ret_val = e1000_read_nvm(hw, NVM_INIT_CONTROL2_REG, 1, &buf);
4680         if (!(le16_to_cpu(buf) & (1 << 0))) {
4681                 /* Deep Smart Power Down (DSPD) */
4682                 dev_warn(&adapter->pdev->dev,
4683                          "Warning: detected DSPD enabled in EEPROM\n");
4684         }
4685
4686         ret_val = e1000_read_nvm(hw, NVM_INIT_3GIO_3, 1, &buf);
4687         if (le16_to_cpu(buf) & (3 << 2)) {
4688                 /* ASPM enable */
4689                 dev_warn(&adapter->pdev->dev,
4690                          "Warning: detected ASPM enabled in EEPROM\n");
4691         }
4692 }
4693
4694 /**
4695  * e1000_probe - Device Initialization Routine
4696  * @pdev: PCI device information struct
4697  * @ent: entry in e1000_pci_tbl
4698  *
4699  * Returns 0 on success, negative on failure
4700  *
4701  * e1000_probe initializes an adapter identified by a pci_dev structure.
4702  * The OS initialization, configuring of the adapter private structure,
4703  * and a hardware reset occur.
4704  **/
4705 static int __devinit e1000_probe(struct pci_dev *pdev,
4706                                  const struct pci_device_id *ent)
4707 {
4708         struct net_device *netdev;
4709         struct e1000_adapter *adapter;
4710         struct e1000_hw *hw;
4711         const struct e1000_info *ei = e1000_info_tbl[ent->driver_data];
4712         resource_size_t mmio_start, mmio_len;
4713         resource_size_t flash_start, flash_len;
4714
4715         static int cards_found;
4716         int i, err, pci_using_dac;
4717         u16 eeprom_data = 0;
4718         u16 eeprom_apme_mask = E1000_EEPROM_APME;
4719
4720         e1000e_disable_l1aspm(pdev);
4721
4722         err = pci_enable_device_mem(pdev);
4723         if (err)
4724                 return err;
4725
4726         pci_using_dac = 0;
4727         err = pci_set_dma_mask(pdev, DMA_64BIT_MASK);
4728         if (!err) {
4729                 err = pci_set_consistent_dma_mask(pdev, DMA_64BIT_MASK);
4730                 if (!err)
4731                         pci_using_dac = 1;
4732         } else {
4733                 err = pci_set_dma_mask(pdev, DMA_32BIT_MASK);
4734                 if (err) {
4735                         err = pci_set_consistent_dma_mask(pdev,
4736                                                           DMA_32BIT_MASK);
4737                         if (err) {
4738                                 dev_err(&pdev->dev, "No usable DMA "
4739                                         "configuration, aborting\n");
4740                                 goto err_dma;
4741                         }
4742                 }
4743         }
4744
4745         err = pci_request_selected_regions(pdev,
4746                                           pci_select_bars(pdev, IORESOURCE_MEM),
4747                                           e1000e_driver_name);
4748         if (err)
4749                 goto err_pci_reg;
4750
4751         pci_set_master(pdev);
4752         pci_save_state(pdev);
4753
4754         err = -ENOMEM;
4755         netdev = alloc_etherdev(sizeof(struct e1000_adapter));
4756         if (!netdev)
4757                 goto err_alloc_etherdev;
4758
4759         SET_NETDEV_DEV(netdev, &pdev->dev);
4760
4761         pci_set_drvdata(pdev, netdev);
4762         adapter = netdev_priv(netdev);
4763         hw = &adapter->hw;
4764         adapter->netdev = netdev;
4765         adapter->pdev = pdev;
4766         adapter->ei = ei;
4767         adapter->pba = ei->pba;
4768         adapter->flags = ei->flags;
4769         adapter->hw.adapter = adapter;
4770         adapter->hw.mac.type = ei->mac;
4771         adapter->msg_enable = (1 << NETIF_MSG_DRV | NETIF_MSG_PROBE) - 1;
4772
4773         mmio_start = pci_resource_start(pdev, 0);
4774         mmio_len = pci_resource_len(pdev, 0);
4775
4776         err = -EIO;
4777         adapter->hw.hw_addr = ioremap(mmio_start, mmio_len);
4778         if (!adapter->hw.hw_addr)
4779                 goto err_ioremap;
4780
4781         if ((adapter->flags & FLAG_HAS_FLASH) &&
4782             (pci_resource_flags(pdev, 1) & IORESOURCE_MEM)) {
4783                 flash_start = pci_resource_start(pdev, 1);
4784                 flash_len = pci_resource_len(pdev, 1);
4785                 adapter->hw.flash_address = ioremap(flash_start, flash_len);
4786                 if (!adapter->hw.flash_address)
4787                         goto err_flashmap;
4788         }
4789
4790         /* construct the net_device struct */
4791         netdev->open                    = &e1000_open;
4792         netdev->stop                    = &e1000_close;
4793         netdev->hard_start_xmit         = &e1000_xmit_frame;
4794         netdev->get_stats               = &e1000_get_stats;
4795         netdev->set_multicast_list      = &e1000_set_multi;
4796         netdev->set_mac_address         = &e1000_set_mac;
4797         netdev->change_mtu              = &e1000_change_mtu;
4798         netdev->do_ioctl                = &e1000_ioctl;
4799         e1000e_set_ethtool_ops(netdev);
4800         netdev->tx_timeout              = &e1000_tx_timeout;
4801         netdev->watchdog_timeo          = 5 * HZ;
4802         netif_napi_add(netdev, &adapter->napi, e1000_clean, 64);
4803         netdev->vlan_rx_register        = e1000_vlan_rx_register;
4804         netdev->vlan_rx_add_vid         = e1000_vlan_rx_add_vid;
4805         netdev->vlan_rx_kill_vid        = e1000_vlan_rx_kill_vid;
4806 #ifdef CONFIG_NET_POLL_CONTROLLER
4807         netdev->poll_controller         = e1000_netpoll;
4808 #endif
4809         strncpy(netdev->name, pci_name(pdev), sizeof(netdev->name) - 1);
4810
4811         netdev->mem_start = mmio_start;
4812         netdev->mem_end = mmio_start + mmio_len;
4813
4814         adapter->bd_number = cards_found++;
4815
4816         e1000e_check_options(adapter);
4817
4818         /* setup adapter struct */
4819         err = e1000_sw_init(adapter);
4820         if (err)
4821                 goto err_sw_init;
4822
4823         err = -EIO;
4824
4825         memcpy(&hw->mac.ops, ei->mac_ops, sizeof(hw->mac.ops));
4826         memcpy(&hw->nvm.ops, ei->nvm_ops, sizeof(hw->nvm.ops));
4827         memcpy(&hw->phy.ops, ei->phy_ops, sizeof(hw->phy.ops));
4828
4829         err = ei->get_variants(adapter);
4830         if (err)
4831                 goto err_hw_init;
4832
4833         if ((adapter->flags & FLAG_IS_ICH) &&
4834             (adapter->flags & FLAG_READ_ONLY_NVM))
4835                 e1000e_write_protect_nvm_ich8lan(&adapter->hw);
4836
4837         hw->mac.ops.get_bus_info(&adapter->hw);
4838
4839         adapter->hw.phy.autoneg_wait_to_complete = 0;
4840
4841         /* Copper options */
4842         if (adapter->hw.phy.media_type == e1000_media_type_copper) {
4843                 adapter->hw.phy.mdix = AUTO_ALL_MODES;
4844                 adapter->hw.phy.disable_polarity_correction = 0;
4845                 adapter->hw.phy.ms_type = e1000_ms_hw_default;
4846         }
4847
4848         if (e1000_check_reset_block(&adapter->hw))
4849                 e_info("PHY reset is blocked due to SOL/IDER session.\n");
4850
4851         netdev->features = NETIF_F_SG |
4852                            NETIF_F_HW_CSUM |
4853                            NETIF_F_HW_VLAN_TX |
4854                            NETIF_F_HW_VLAN_RX;
4855
4856         if (adapter->flags & FLAG_HAS_HW_VLAN_FILTER)
4857                 netdev->features |= NETIF_F_HW_VLAN_FILTER;
4858
4859         netdev->features |= NETIF_F_TSO;
4860         netdev->features |= NETIF_F_TSO6;
4861
4862         netdev->vlan_features |= NETIF_F_TSO;
4863         netdev->vlan_features |= NETIF_F_TSO6;
4864         netdev->vlan_features |= NETIF_F_HW_CSUM;
4865         netdev->vlan_features |= NETIF_F_SG;
4866
4867         if (pci_using_dac)
4868                 netdev->features |= NETIF_F_HIGHDMA;
4869
4870         /*
4871          * We should not be using LLTX anymore, but we are still Tx faster with
4872          * it.
4873          */
4874         netdev->features |= NETIF_F_LLTX;
4875
4876         if (e1000e_enable_mng_pass_thru(&adapter->hw))
4877                 adapter->flags |= FLAG_MNG_PT_ENABLED;
4878
4879         /*
4880          * before reading the NVM, reset the controller to
4881          * put the device in a known good starting state
4882          */
4883         adapter->hw.mac.ops.reset_hw(&adapter->hw);
4884
4885         /*
4886          * systems with ASPM and others may see the checksum fail on the first
4887          * attempt. Let's give it a few tries
4888          */
4889         for (i = 0;; i++) {
4890                 if (e1000_validate_nvm_checksum(&adapter->hw) >= 0)
4891                         break;
4892                 if (i == 2) {
4893                         e_err("The NVM Checksum Is Not Valid\n");
4894                         err = -EIO;
4895                         goto err_eeprom;
4896                 }
4897         }
4898
4899         e1000_eeprom_checks(adapter);
4900
4901         /* copy the MAC address out of the NVM */
4902         if (e1000e_read_mac_addr(&adapter->hw))
4903                 e_err("NVM Read Error while reading MAC address\n");
4904
4905         memcpy(netdev->dev_addr, adapter->hw.mac.addr, netdev->addr_len);
4906         memcpy(netdev->perm_addr, adapter->hw.mac.addr, netdev->addr_len);
4907
4908         if (!is_valid_ether_addr(netdev->perm_addr)) {
4909                 e_err("Invalid MAC Address: %02x:%02x:%02x:%02x:%02x:%02x\n",
4910                       netdev->perm_addr[0], netdev->perm_addr[1],
4911                       netdev->perm_addr[2], netdev->perm_addr[3],
4912                       netdev->perm_addr[4], netdev->perm_addr[5]);
4913                 err = -EIO;
4914                 goto err_eeprom;
4915         }
4916
4917         init_timer(&adapter->watchdog_timer);
4918         adapter->watchdog_timer.function = &e1000_watchdog;
4919         adapter->watchdog_timer.data = (unsigned long) adapter;
4920
4921         init_timer(&adapter->phy_info_timer);
4922         adapter->phy_info_timer.function = &e1000_update_phy_info;
4923         adapter->phy_info_timer.data = (unsigned long) adapter;
4924
4925         INIT_WORK(&adapter->reset_task, e1000_reset_task);
4926         INIT_WORK(&adapter->watchdog_task, e1000_watchdog_task);
4927         INIT_WORK(&adapter->downshift_task, e1000e_downshift_workaround);
4928         INIT_WORK(&adapter->update_phy_task, e1000e_update_phy_task);
4929
4930         /* Initialize link parameters. User can change them with ethtool */
4931         adapter->hw.mac.autoneg = 1;
4932         adapter->fc_autoneg = 1;
4933         adapter->hw.fc.original_type = e1000_fc_default;
4934         adapter->hw.fc.type = e1000_fc_default;
4935         adapter->hw.phy.autoneg_advertised = 0x2f;
4936
4937         /* ring size defaults */
4938         adapter->rx_ring->count = 256;
4939         adapter->tx_ring->count = 256;
4940
4941         /*
4942          * Initial Wake on LAN setting - If APM wake is enabled in
4943          * the EEPROM, enable the ACPI Magic Packet filter
4944          */
4945         if (adapter->flags & FLAG_APME_IN_WUC) {
4946                 /* APME bit in EEPROM is mapped to WUC.APME */
4947                 eeprom_data = er32(WUC);
4948                 eeprom_apme_mask = E1000_WUC_APME;
4949         } else if (adapter->flags & FLAG_APME_IN_CTRL3) {
4950                 if (adapter->flags & FLAG_APME_CHECK_PORT_B &&
4951                     (adapter->hw.bus.func == 1))
4952                         e1000_read_nvm(&adapter->hw,
4953                                 NVM_INIT_CONTROL3_PORT_B, 1, &eeprom_data);
4954                 else
4955                         e1000_read_nvm(&adapter->hw,
4956                                 NVM_INIT_CONTROL3_PORT_A, 1, &eeprom_data);
4957         }
4958
4959         /* fetch WoL from EEPROM */
4960         if (eeprom_data & eeprom_apme_mask)
4961                 adapter->eeprom_wol |= E1000_WUFC_MAG;
4962
4963         /*
4964          * now that we have the eeprom settings, apply the special cases
4965          * where the eeprom may be wrong or the board simply won't support
4966          * wake on lan on a particular port
4967          */
4968         if (!(adapter->flags & FLAG_HAS_WOL))
4969                 adapter->eeprom_wol = 0;
4970
4971         /* initialize the wol settings based on the eeprom settings */
4972         adapter->wol = adapter->eeprom_wol;
4973
4974         /* reset the hardware with the new settings */
4975         e1000e_reset(adapter);
4976
4977         /*
4978          * If the controller has AMT, do not set DRV_LOAD until the interface
4979          * is up.  For all other cases, let the f/w know that the h/w is now
4980          * under the control of the driver.
4981          */
4982         if (!(adapter->flags & FLAG_HAS_AMT))
4983                 e1000_get_hw_control(adapter);
4984
4985         /* tell the stack to leave us alone until e1000_open() is called */
4986         netif_carrier_off(netdev);
4987         netif_tx_stop_all_queues(netdev);
4988
4989         strcpy(netdev->name, "eth%d");
4990         err = register_netdev(netdev);
4991         if (err)
4992                 goto err_register;
4993
4994         e1000_print_device_info(adapter);
4995
4996         return 0;
4997
4998 err_register:
4999         if (!(adapter->flags & FLAG_HAS_AMT))
5000                 e1000_release_hw_control(adapter);
5001 err_eeprom:
5002         if (!e1000_check_reset_block(&adapter->hw))
5003                 e1000_phy_hw_reset(&adapter->hw);
5004 err_hw_init:
5005
5006         kfree(adapter->tx_ring);
5007         kfree(adapter->rx_ring);
5008 err_sw_init:
5009         if (adapter->hw.flash_address)
5010                 iounmap(adapter->hw.flash_address);
5011 err_flashmap:
5012         iounmap(adapter->hw.hw_addr);
5013 err_ioremap:
5014         free_netdev(netdev);
5015 err_alloc_etherdev:
5016         pci_release_selected_regions(pdev,
5017                                      pci_select_bars(pdev, IORESOURCE_MEM));
5018 err_pci_reg:
5019 err_dma:
5020         pci_disable_device(pdev);
5021         return err;
5022 }
5023
5024 /**
5025  * e1000_remove - Device Removal Routine
5026  * @pdev: PCI device information struct
5027  *
5028  * e1000_remove is called by the PCI subsystem to alert the driver
5029  * that it should release a PCI device.  The could be caused by a
5030  * Hot-Plug event, or because the driver is going to be removed from
5031  * memory.
5032  **/
5033 static void __devexit e1000_remove(struct pci_dev *pdev)
5034 {
5035         struct net_device *netdev = pci_get_drvdata(pdev);
5036         struct e1000_adapter *adapter = netdev_priv(netdev);
5037
5038         /*
5039          * flush_scheduled work may reschedule our watchdog task, so
5040          * explicitly disable watchdog tasks from being rescheduled
5041          */
5042         set_bit(__E1000_DOWN, &adapter->state);
5043         del_timer_sync(&adapter->watchdog_timer);
5044         del_timer_sync(&adapter->phy_info_timer);
5045
5046         flush_scheduled_work();
5047
5048         /*
5049          * Release control of h/w to f/w.  If f/w is AMT enabled, this
5050          * would have already happened in close and is redundant.
5051          */
5052         e1000_release_hw_control(adapter);
5053
5054         unregister_netdev(netdev);
5055
5056         if (!e1000_check_reset_block(&adapter->hw))
5057                 e1000_phy_hw_reset(&adapter->hw);
5058
5059         e1000e_reset_interrupt_capability(adapter);
5060         kfree(adapter->tx_ring);
5061         kfree(adapter->rx_ring);
5062
5063         iounmap(adapter->hw.hw_addr);
5064         if (adapter->hw.flash_address)
5065                 iounmap(adapter->hw.flash_address);
5066         pci_release_selected_regions(pdev,
5067                                      pci_select_bars(pdev, IORESOURCE_MEM));
5068
5069         free_netdev(netdev);
5070
5071         pci_disable_device(pdev);
5072 }
5073
5074 /* PCI Error Recovery (ERS) */
5075 static struct pci_error_handlers e1000_err_handler = {
5076         .error_detected = e1000_io_error_detected,
5077         .slot_reset = e1000_io_slot_reset,
5078         .resume = e1000_io_resume,
5079 };
5080
5081 static struct pci_device_id e1000_pci_tbl[] = {
5082         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82571EB_COPPER), board_82571 },
5083         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82571EB_FIBER), board_82571 },
5084         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82571EB_QUAD_COPPER), board_82571 },
5085         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82571EB_QUAD_COPPER_LP), board_82571 },
5086         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82571EB_QUAD_FIBER), board_82571 },
5087         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82571EB_SERDES), board_82571 },
5088         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82571EB_SERDES_DUAL), board_82571 },
5089         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82571EB_SERDES_QUAD), board_82571 },
5090         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82571PT_QUAD_COPPER), board_82571 },
5091
5092         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82572EI), board_82572 },
5093         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82572EI_COPPER), board_82572 },
5094         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82572EI_FIBER), board_82572 },
5095         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82572EI_SERDES), board_82572 },
5096
5097         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82573E), board_82573 },
5098         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82573E_IAMT), board_82573 },
5099         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82573L), board_82573 },
5100
5101         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82574L), board_82574 },
5102
5103         { PCI_VDEVICE(INTEL, E1000_DEV_ID_80003ES2LAN_COPPER_DPT),
5104           board_80003es2lan },
5105         { PCI_VDEVICE(INTEL, E1000_DEV_ID_80003ES2LAN_COPPER_SPT),
5106           board_80003es2lan },
5107         { PCI_VDEVICE(INTEL, E1000_DEV_ID_80003ES2LAN_SERDES_DPT),
5108           board_80003es2lan },
5109         { PCI_VDEVICE(INTEL, E1000_DEV_ID_80003ES2LAN_SERDES_SPT),
5110           board_80003es2lan },
5111
5112         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH8_IFE), board_ich8lan },
5113         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH8_IFE_G), board_ich8lan },
5114         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH8_IFE_GT), board_ich8lan },
5115         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH8_IGP_AMT), board_ich8lan },
5116         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH8_IGP_C), board_ich8lan },
5117         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH8_IGP_M), board_ich8lan },
5118         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH8_IGP_M_AMT), board_ich8lan },
5119
5120         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH9_IFE), board_ich9lan },
5121         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH9_IFE_G), board_ich9lan },
5122         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH9_IFE_GT), board_ich9lan },
5123         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH9_IGP_AMT), board_ich9lan },
5124         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH9_IGP_C), board_ich9lan },
5125         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH9_BM), board_ich9lan },
5126         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH9_IGP_M), board_ich9lan },
5127         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH9_IGP_M_AMT), board_ich9lan },
5128         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH9_IGP_M_V), board_ich9lan },
5129
5130         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH10_R_BM_LM), board_ich9lan },
5131         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH10_R_BM_LF), board_ich9lan },
5132         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH10_R_BM_V), board_ich9lan },
5133
5134         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH10_D_BM_LM), board_ich10lan },
5135         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH10_D_BM_LF), board_ich10lan },
5136
5137         { }     /* terminate list */
5138 };
5139 MODULE_DEVICE_TABLE(pci, e1000_pci_tbl);
5140
5141 /* PCI Device API Driver */
5142 static struct pci_driver e1000_driver = {
5143         .name     = e1000e_driver_name,
5144         .id_table = e1000_pci_tbl,
5145         .probe    = e1000_probe,
5146         .remove   = __devexit_p(e1000_remove),
5147 #ifdef CONFIG_PM
5148         /* Power Management Hooks */
5149         .suspend  = e1000_suspend,
5150         .resume   = e1000_resume,
5151 #endif
5152         .shutdown = e1000_shutdown,
5153         .err_handler = &e1000_err_handler
5154 };
5155
5156 /**
5157  * e1000_init_module - Driver Registration Routine
5158  *
5159  * e1000_init_module is the first routine called when the driver is
5160  * loaded. All it does is register with the PCI subsystem.
5161  **/
5162 static int __init e1000_init_module(void)
5163 {
5164         int ret;
5165         printk(KERN_INFO "%s: Intel(R) PRO/1000 Network Driver - %s\n",
5166                e1000e_driver_name, e1000e_driver_version);
5167         printk(KERN_INFO "%s: Copyright (c) 1999-2008 Intel Corporation.\n",
5168                e1000e_driver_name);
5169         ret = pci_register_driver(&e1000_driver);
5170         pm_qos_add_requirement(PM_QOS_CPU_DMA_LATENCY, e1000e_driver_name,
5171                                PM_QOS_DEFAULT_VALUE);
5172
5173         return ret;
5174 }
5175 module_init(e1000_init_module);
5176
5177 /**
5178  * e1000_exit_module - Driver Exit Cleanup Routine
5179  *
5180  * e1000_exit_module is called just before the driver is removed
5181  * from memory.
5182  **/
5183 static void __exit e1000_exit_module(void)
5184 {
5185         pci_unregister_driver(&e1000_driver);
5186         pm_qos_remove_requirement(PM_QOS_CPU_DMA_LATENCY, e1000e_driver_name);
5187 }
5188 module_exit(e1000_exit_module);
5189
5190
5191 MODULE_AUTHOR("Intel Corporation, <linux.nics@intel.com>");
5192 MODULE_DESCRIPTION("Intel(R) PRO/1000 Network Driver");
5193 MODULE_LICENSE("GPL");
5194 MODULE_VERSION(DRV_VERSION);
5195
5196 /* e1000_main.c */