err.no 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-k2"
  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_80003es2lan]     = &e1000_es2_info,
  59         [board_ich8lan]         = &e1000_ich8_info,
  60         [board_ich9lan]         = &e1000_ich9_info,
  61 };
  62
  63 #ifdef DEBUG
  64 /**
  65  * e1000_get_hw_dev_name - return device name string
  66  * used by hardware layer to print debugging information
  67  **/
  68 char *e1000e_get_hw_dev_name(struct e1000_hw *hw)
  69 {
  70         return hw->adapter->netdev->name;
  71 }
  72 #endif
  73
  74 /**
  75  * e1000_desc_unused - calculate if we have unused descriptors
  76  **/
  77 static int e1000_desc_unused(struct e1000_ring *ring)
  78 {
  79         if (ring->next_to_clean > ring->next_to_use)
  80                 return ring->next_to_clean - ring->next_to_use - 1;
  81
  82         return ring->count + ring->next_to_clean - ring->next_to_use - 1;
  83 }
  84
  85 /**
  86  * e1000_receive_skb - helper function to handle Rx indications
  87  * @adapter: board private structure
  88  * @status: descriptor status field as written by hardware
  89  * @vlan: descriptor vlan field as written by hardware (no le/be conversion)
  90  * @skb: pointer to sk_buff to be indicated to stack
  91  **/
  92 static void e1000_receive_skb(struct e1000_adapter *adapter,
  93                               struct net_device *netdev,
  94                               struct sk_buff *skb,
  95                               u8 status, __le16 vlan)
  96 {
  97         skb->protocol = eth_type_trans(skb, netdev);
  98
  99         if (adapter->vlgrp && (status & E1000_RXD_STAT_VP))
 100                 vlan_hwaccel_receive_skb(skb, adapter->vlgrp,
 101                                          le16_to_cpu(vlan));
 102         else
 103                 netif_receive_skb(skb);
 104
 105         netdev->last_rx = jiffies;
 106 }
 107
 108 /**
 109  * e1000_rx_checksum - Receive Checksum Offload for 82543
 110  * @adapter:     board private structure
 111  * @status_err:  receive descriptor status and error fields
 112  * @csum:       receive descriptor csum field
 113  * @sk_buff:     socket buffer with received data
 114  **/
 115 static void e1000_rx_checksum(struct e1000_adapter *adapter, u32 status_err,
 116                               u32 csum, struct sk_buff *skb)
 117 {
 118         u16 status = (u16)status_err;
 119         u8 errors = (u8)(status_err >> 24);
 120         skb->ip_summed = CHECKSUM_NONE;
 121
 122         /* Ignore Checksum bit is set */
 123         if (status & E1000_RXD_STAT_IXSM)
 124                 return;
 125         /* TCP/UDP checksum error bit is set */
 126         if (errors & E1000_RXD_ERR_TCPE) {
 127                 /* let the stack verify checksum errors */
 128                 adapter->hw_csum_err++;
 129                 return;
 130         }
 131
 132         /* TCP/UDP Checksum has not been calculated */
 133         if (!(status & (E1000_RXD_STAT_TCPCS | E1000_RXD_STAT_UDPCS)))
 134                 return;
 135
 136         /* It must be a TCP or UDP packet with a valid checksum */
 137         if (status & E1000_RXD_STAT_TCPCS) {
 138                 /* TCP checksum is good */
 139                 skb->ip_summed = CHECKSUM_UNNECESSARY;
 140         } else {
 141                 /*
 142                  * IP fragment with UDP payload
 143                  * Hardware complements the payload checksum, so we undo it
 144                  * and then put the value in host order for further stack use.
 145                  */
 146                 __sum16 sum = (__force __sum16)htons(csum);
 147                 skb->csum = csum_unfold(~sum);
 148                 skb->ip_summed = CHECKSUM_COMPLETE;
 149         }
 150         adapter->hw_csum_good++;
 151 }
 152
 153 /**
 154  * e1000_alloc_rx_buffers - Replace used receive buffers; legacy & extended
 155  * @adapter: address of board private structure
 156  **/
 157 static void e1000_alloc_rx_buffers(struct e1000_adapter *adapter,
 158                                    int cleaned_count)
 159 {
 160         struct net_device *netdev = adapter->netdev;
 161         struct pci_dev *pdev = adapter->pdev;
 162         struct e1000_ring *rx_ring = adapter->rx_ring;
 163         struct e1000_rx_desc *rx_desc;
 164         struct e1000_buffer *buffer_info;
 165         struct sk_buff *skb;
 166         unsigned int i;
 167         unsigned int bufsz = adapter->rx_buffer_len + NET_IP_ALIGN;
 168
 169         i = rx_ring->next_to_use;
 170         buffer_info = &rx_ring->buffer_info[i];
 171
 172         while (cleaned_count--) {
 173                 skb = buffer_info->skb;
 174                 if (skb) {
 175                         skb_trim(skb, 0);
 176                         goto map_skb;
 177                 }
 178
 179                 skb = netdev_alloc_skb(netdev, bufsz);
 180                 if (!skb) {
 181                         /* Better luck next round */
 182                         adapter->alloc_rx_buff_failed++;
 183                         break;
 184                 }
 185
 186                 /*
 187                  * Make buffer alignment 2 beyond a 16 byte boundary
 188                  * this will result in a 16 byte aligned IP header after
 189                  * the 14 byte MAC header is removed
 190                  */
 191                 skb_reserve(skb, NET_IP_ALIGN);
 192
 193                 buffer_info->skb = skb;
 194 map_skb:
 195                 buffer_info->dma = pci_map_single(pdev, skb->data,
 196                                                   adapter->rx_buffer_len,
 197                                                   PCI_DMA_FROMDEVICE);
 198                 if (pci_dma_mapping_error(pdev, buffer_info->dma)) {
 199                         dev_err(&pdev->dev, "RX DMA map failed\n");
 200                         adapter->rx_dma_failed++;
 201                         break;
 202                 }
 203
 204                 rx_desc = E1000_RX_DESC(*rx_ring, i);
 205                 rx_desc->buffer_addr = cpu_to_le64(buffer_info->dma);
 206
 207                 i++;
 208                 if (i == rx_ring->count)
 209                         i = 0;
 210                 buffer_info = &rx_ring->buffer_info[i];
 211         }
 212
 213         if (rx_ring->next_to_use != i) {
 214                 rx_ring->next_to_use = i;
 215                 if (i-- == 0)
 216                         i = (rx_ring->count - 1);
 217
 218                 /*
 219                  * Force memory writes to complete before letting h/w
 220                  * know there are new descriptors to fetch.  (Only
 221                  * applicable for weak-ordered memory model archs,
 222                  * such as IA-64).
 223                  */
 224                 wmb();
 225                 writel(i, adapter->hw.hw_addr + rx_ring->tail);
 226         }
 227 }
 228
 229 /**
 230  * e1000_alloc_rx_buffers_ps - Replace used receive buffers; packet split
 231  * @adapter: address of board private structure
 232  **/
 233 static void e1000_alloc_rx_buffers_ps(struct e1000_adapter *adapter,
 234                                       int cleaned_count)
 235 {
 236         struct net_device *netdev = adapter->netdev;
 237         struct pci_dev *pdev = adapter->pdev;
 238         union e1000_rx_desc_packet_split *rx_desc;
 239         struct e1000_ring *rx_ring = adapter->rx_ring;
 240         struct e1000_buffer *buffer_info;
 241         struct e1000_ps_page *ps_page;
 242         struct sk_buff *skb;
 243         unsigned int i, j;
 244
 245         i = rx_ring->next_to_use;
 246         buffer_info = &rx_ring->buffer_info[i];
 247
 248         while (cleaned_count--) {
 249                 rx_desc = E1000_RX_DESC_PS(*rx_ring, i);
 250
 251                 for (j = 0; j < PS_PAGE_BUFFERS; j++) {
 252                         ps_page = &buffer_info->ps_pages[j];
 253                         if (j >= adapter->rx_ps_pages) {
 254                                 /* all unused desc entries get hw null ptr */
 255                                 rx_desc->read.buffer_addr[j+1] = ~cpu_to_le64(0);
 256                                 continue;
 257                         }
 258                         if (!ps_page->page) {
 259                                 ps_page->page = alloc_page(GFP_ATOMIC);
 260                                 if (!ps_page->page) {
 261                                         adapter->alloc_rx_buff_failed++;
 262                                         goto no_buffers;
 263                                 }
 264                                 ps_page->dma = pci_map_page(pdev,
 265                                                    ps_page->page,
 266                                                    0, PAGE_SIZE,
 267                                                    PCI_DMA_FROMDEVICE);
 268                                 if (pci_dma_mapping_error(pdev, ps_page->dma)) {
 269                                         dev_err(&adapter->pdev->dev,
 270                                           "RX DMA page map failed\n");
 271                                         adapter->rx_dma_failed++;
 272                                         goto no_buffers;
 273                                 }
 274                         }
 275                         /*
 276                          * Refresh the desc even if buffer_addrs
 277                          * didn't change because each write-back
 278                          * erases this info.
 279                          */
 280                         rx_desc->read.buffer_addr[j+1] =
 281                              cpu_to_le64(ps_page->dma);
 282                 }
 283
 284                 skb = netdev_alloc_skb(netdev,
 285                                        adapter->rx_ps_bsize0 + NET_IP_ALIGN);
 286
 287                 if (!skb) {
 288                         adapter->alloc_rx_buff_failed++;
 289                         break;
 290                 }
 291
 292                 /*
 293                  * Make buffer alignment 2 beyond a 16 byte boundary
 294                  * this will result in a 16 byte aligned IP header after
 295                  * the 14 byte MAC header is removed
 296                  */
 297                 skb_reserve(skb, NET_IP_ALIGN);
 298
 299                 buffer_info->skb = skb;
 300                 buffer_info->dma = pci_map_single(pdev, skb->data,
 301                                                   adapter->rx_ps_bsize0,
 302                                                   PCI_DMA_FROMDEVICE);
 303                 if (pci_dma_mapping_error(pdev, buffer_info->dma)) {
 304                         dev_err(&pdev->dev, "RX DMA map failed\n");
 305                         adapter->rx_dma_failed++;
 306                         /* cleanup skb */
 307                         dev_kfree_skb_any(skb);
 308                         buffer_info->skb = NULL;
 309                         break;
 310                 }
 311
 312                 rx_desc->read.buffer_addr[0] = cpu_to_le64(buffer_info->dma);
 313
 314                 i++;
 315                 if (i == rx_ring->count)
 316                         i = 0;
 317                 buffer_info = &rx_ring->buffer_info[i];
 318         }
 319
 320 no_buffers:
 321         if (rx_ring->next_to_use != i) {
 322                 rx_ring->next_to_use = i;
 323
 324                 if (!(i--))
 325                         i = (rx_ring->count - 1);
 326
 327                 /*
 328                  * Force memory writes to complete before letting h/w
 329                  * know there are new descriptors to fetch.  (Only
 330                  * applicable for weak-ordered memory model archs,
 331                  * such as IA-64).
 332                  */
 333                 wmb();
 334                 /*
 335                  * Hardware increments by 16 bytes, but packet split
 336                  * descriptors are 32 bytes...so we increment tail
 337                  * twice as much.
 338                  */
 339                 writel(i<<1, adapter->hw.hw_addr + rx_ring->tail);
 340         }
 341 }
 342
 343 /**
 344  * e1000_alloc_jumbo_rx_buffers - Replace used jumbo receive buffers
 345  * @adapter: address of board private structure
 346  * @rx_ring: pointer to receive ring structure
 347  * @cleaned_count: number of buffers to allocate this pass
 348  **/
 349
 350 static void e1000_alloc_jumbo_rx_buffers(struct e1000_adapter *adapter,
 351                                          int cleaned_count)
 352 {
 353         struct net_device *netdev = adapter->netdev;
 354         struct pci_dev *pdev = adapter->pdev;
 355         struct e1000_rx_desc *rx_desc;
 356         struct e1000_ring *rx_ring = adapter->rx_ring;
 357         struct e1000_buffer *buffer_info;
 358         struct sk_buff *skb;
 359         unsigned int i;
 360         unsigned int bufsz = 256 -
 361                              16 /* for skb_reserve */ -
 362                              NET_IP_ALIGN;
 363
 364         i = rx_ring->next_to_use;
 365         buffer_info = &rx_ring->buffer_info[i];
 366
 367         while (cleaned_count--) {
 368                 skb = buffer_info->skb;
 369                 if (skb) {
 370                         skb_trim(skb, 0);
 371                         goto check_page;
 372                 }
 373
 374                 skb = netdev_alloc_skb(netdev, bufsz);
 375                 if (unlikely(!skb)) {
 376                         /* Better luck next round */
 377                         adapter->alloc_rx_buff_failed++;
 378                         break;
 379                 }
 380
 381                 /* Make buffer alignment 2 beyond a 16 byte boundary
 382                  * this will result in a 16 byte aligned IP header after
 383                  * the 14 byte MAC header is removed
 384                  */
 385                 skb_reserve(skb, NET_IP_ALIGN);
 386
 387                 buffer_info->skb = skb;
 388 check_page:
 389                 /* allocate a new page if necessary */
 390                 if (!buffer_info->page) {
 391                         buffer_info->page = alloc_page(GFP_ATOMIC);
 392                         if (unlikely(!buffer_info->page)) {
 393                                 adapter->alloc_rx_buff_failed++;
 394                                 break;
 395                         }
 396                 }
 397
 398                 if (!buffer_info->dma)
 399                         buffer_info->dma = pci_map_page(pdev,
 400                                                         buffer_info->page, 0,
 401                                                         PAGE_SIZE,
 402                                                         PCI_DMA_FROMDEVICE);
 403
 404                 rx_desc = E1000_RX_DESC(*rx_ring, i);
 405                 rx_desc->buffer_addr = cpu_to_le64(buffer_info->dma);
 406
 407                 if (unlikely(++i == rx_ring->count))
 408                         i = 0;
 409                 buffer_info = &rx_ring->buffer_info[i];
 410         }
 411
 412         if (likely(rx_ring->next_to_use != i)) {
 413                 rx_ring->next_to_use = i;
 414                 if (unlikely(i-- == 0))
 415                         i = (rx_ring->count - 1);
 416
 417                 /* Force memory writes to complete before letting h/w
 418                  * know there are new descriptors to fetch.  (Only
 419                  * applicable for weak-ordered memory model archs,
 420                  * such as IA-64). */
 421                 wmb();
 422                 writel(i, adapter->hw.hw_addr + rx_ring->tail);
 423         }
 424 }
 425
 426 /**
 427  * e1000_clean_rx_irq - Send received data up the network stack; legacy
 428  * @adapter: board private structure
 429  *
 430  * the return value indicates whether actual cleaning was done, there
 431  * is no guarantee that everything was cleaned
 432  **/
 433 static bool e1000_clean_rx_irq(struct e1000_adapter *adapter,
 434                                int *work_done, int work_to_do)
 435 {
 436         struct net_device *netdev = adapter->netdev;
 437         struct pci_dev *pdev = adapter->pdev;
 438         struct e1000_ring *rx_ring = adapter->rx_ring;
 439         struct e1000_rx_desc *rx_desc, *next_rxd;
 440         struct e1000_buffer *buffer_info, *next_buffer;
 441         u32 length;
 442         unsigned int i;
 443         int cleaned_count = 0;
 444         bool cleaned = 0;
 445         unsigned int total_rx_bytes = 0, total_rx_packets = 0;
 446
 447         i = rx_ring->next_to_clean;
 448         rx_desc = E1000_RX_DESC(*rx_ring, i);
 449         buffer_info = &rx_ring->buffer_info[i];
 450
 451         while (rx_desc->status & E1000_RXD_STAT_DD) {
 452                 struct sk_buff *skb;
 453                 u8 status;
 454
 455                 if (*work_done >= work_to_do)
 456                         break;
 457                 (*work_done)++;
 458
 459                 status = rx_desc->status;
 460                 skb = buffer_info->skb;
 461                 buffer_info->skb = NULL;
 462
 463                 prefetch(skb->data - NET_IP_ALIGN);
 464
 465                 i++;
 466                 if (i == rx_ring->count)
 467                         i = 0;
 468                 next_rxd = E1000_RX_DESC(*rx_ring, i);
 469                 prefetch(next_rxd);
 470
 471                 next_buffer = &rx_ring->buffer_info[i];
 472
 473                 cleaned = 1;
 474                 cleaned_count++;
 475                 pci_unmap_single(pdev,
 476                                  buffer_info->dma,
 477                                  adapter->rx_buffer_len,
 478                                  PCI_DMA_FROMDEVICE);
 479                 buffer_info->dma = 0;
 480
 481                 length = le16_to_cpu(rx_desc->length);
 482
 483                 /* !EOP means multiple descriptors were used to store a single
 484                  * packet, also make sure the frame isn't just CRC only */
 485                 if (!(status & E1000_RXD_STAT_EOP) || (length <= 4)) {
 486                         /* All receives must fit into a single buffer */
 487                         e_dbg("%s: Receive packet consumed multiple buffers\n",
 488                               netdev->name);
 489                         /* recycle */
 490                         buffer_info->skb = skb;
 491                         goto next_desc;
 492                 }
 493
 494                 if (rx_desc->errors & E1000_RXD_ERR_FRAME_ERR_MASK) {
 495                         /* recycle */
 496                         buffer_info->skb = skb;
 497                         goto next_desc;
 498                 }
 499
 500                 total_rx_bytes += length;
 501                 total_rx_packets++;
 502
 503                 /*
 504                  * code added for copybreak, this should improve
 505                  * performance for small packets with large amounts
 506                  * of reassembly being done in the stack
 507                  */
 508                 if (length < copybreak) {
 509                         struct sk_buff *new_skb =
 510                             netdev_alloc_skb(netdev, length + NET_IP_ALIGN);
 511                         if (new_skb) {
 512                                 skb_reserve(new_skb, NET_IP_ALIGN);
 513                                 skb_copy_to_linear_data_offset(new_skb,
 514                                                                -NET_IP_ALIGN,
 515                                                                (skb->data -
 516                                                                 NET_IP_ALIGN),
 517                                                                (length +
 518                                                                 NET_IP_ALIGN));
 519                                 /* save the skb in buffer_info as good */
 520                                 buffer_info->skb = skb;
 521                                 skb = new_skb;
 522                         }
 523                         /* else just continue with the old one */
 524                 }
 525                 /* end copybreak code */
 526                 skb_put(skb, length);
 527
 528                 /* Receive Checksum Offload */
 529                 e1000_rx_checksum(adapter,
 530                                   (u32)(status) |
 531                                   ((u32)(rx_desc->errors) << 24),
 532                                   le16_to_cpu(rx_desc->csum), skb);
 533
 534                 e1000_receive_skb(adapter, netdev, skb,status,rx_desc->special);
 535
 536 next_desc:
 537                 rx_desc->status = 0;
 538
 539                 /* return some buffers to hardware, one at a time is too slow */
 540                 if (cleaned_count >= E1000_RX_BUFFER_WRITE) {
 541                         adapter->alloc_rx_buf(adapter, cleaned_count);
 542                         cleaned_count = 0;
 543                 }
 544
 545                 /* use prefetched values */
 546                 rx_desc = next_rxd;
 547                 buffer_info = next_buffer;
 548         }
 549         rx_ring->next_to_clean = i;
 550
 551         cleaned_count = e1000_desc_unused(rx_ring);
 552         if (cleaned_count)
 553                 adapter->alloc_rx_buf(adapter, cleaned_count);
 554
 555         adapter->total_rx_bytes += total_rx_bytes;
 556         adapter->total_rx_packets += total_rx_packets;
 557         adapter->net_stats.rx_bytes += total_rx_bytes;
 558         adapter->net_stats.rx_packets += total_rx_packets;
 559         return cleaned;
 560 }
 561
 562 static void e1000_put_txbuf(struct e1000_adapter *adapter,
 563                              struct e1000_buffer *buffer_info)
 564 {
 565         if (buffer_info->dma) {
 566                 pci_unmap_page(adapter->pdev, buffer_info->dma,
 567                                buffer_info->length, PCI_DMA_TODEVICE);
 568                 buffer_info->dma = 0;
 569         }
 570         if (buffer_info->skb) {
 571                 dev_kfree_skb_any(buffer_info->skb);
 572                 buffer_info->skb = NULL;
 573         }
 574 }
 575
 576 static void e1000_print_tx_hang(struct e1000_adapter *adapter)
 577 {
 578         struct e1000_ring *tx_ring = adapter->tx_ring;
 579         unsigned int i = tx_ring->next_to_clean;
 580         unsigned int eop = tx_ring->buffer_info[i].next_to_watch;
 581         struct e1000_tx_desc *eop_desc = E1000_TX_DESC(*tx_ring, eop);
 582
 583         /* detected Tx unit hang */
 584         e_err("Detected Tx Unit Hang:\n"
 585               "  TDH                  <%x>\n"
 586               "  TDT                  <%x>\n"
 587               "  next_to_use          <%x>\n"
 588               "  next_to_clean        <%x>\n"
 589               "buffer_info[next_to_clean]:\n"
 590               "  time_stamp           <%lx>\n"
 591               "  next_to_watch        <%x>\n"
 592               "  jiffies              <%lx>\n"
 593               "  next_to_watch.status <%x>\n",
 594               readl(adapter->hw.hw_addr + tx_ring->head),
 595               readl(adapter->hw.hw_addr + tx_ring->tail),
 596               tx_ring->next_to_use,
 597               tx_ring->next_to_clean,
 598               tx_ring->buffer_info[eop].time_stamp,
 599               eop,
 600               jiffies,
 601               eop_desc->upper.fields.status);
 602 }
 603
 604 /**
 605  * e1000_clean_tx_irq - Reclaim resources after transmit completes
 606  * @adapter: board private structure
 607  *
 608  * the return value indicates whether actual cleaning was done, there
 609  * is no guarantee that everything was cleaned
 610  **/
 611 static bool e1000_clean_tx_irq(struct e1000_adapter *adapter)
 612 {
 613         struct net_device *netdev = adapter->netdev;
 614         struct e1000_hw *hw = &adapter->hw;
 615         struct e1000_ring *tx_ring = adapter->tx_ring;
 616         struct e1000_tx_desc *tx_desc, *eop_desc;
 617         struct e1000_buffer *buffer_info;
 618         unsigned int i, eop;
 619         unsigned int count = 0;
 620         bool cleaned = 0;
 621         unsigned int total_tx_bytes = 0, total_tx_packets = 0;
 622
 623         i = tx_ring->next_to_clean;
 624         eop = tx_ring->buffer_info[i].next_to_watch;
 625         eop_desc = E1000_TX_DESC(*tx_ring, eop);
 626
 627         while (eop_desc->upper.data & cpu_to_le32(E1000_TXD_STAT_DD)) {
 628                 for (cleaned = 0; !cleaned; ) {
 629                         tx_desc = E1000_TX_DESC(*tx_ring, i);
 630                         buffer_info = &tx_ring->buffer_info[i];
 631                         cleaned = (i == eop);
 632
 633                         if (cleaned) {
 634                                 struct sk_buff *skb = buffer_info->skb;
 635                                 unsigned int segs, bytecount;
 636                                 segs = skb_shinfo(skb)->gso_segs ?: 1;
 637                                 /* multiply data chunks by size of headers */
 638                                 bytecount = ((segs - 1) * skb_headlen(skb)) +
 639                                             skb->len;
 640                                 total_tx_packets += segs;
 641                                 total_tx_bytes += bytecount;
 642                         }
 643
 644                         e1000_put_txbuf(adapter, buffer_info);
 645                         tx_desc->upper.data = 0;
 646
 647                         i++;
 648                         if (i == tx_ring->count)
 649                                 i = 0;
 650                 }
 651
 652                 eop = tx_ring->buffer_info[i].next_to_watch;
 653                 eop_desc = E1000_TX_DESC(*tx_ring, eop);
 654 #define E1000_TX_WEIGHT 64
 655                 /* weight of a sort for tx, to avoid endless transmit cleanup */
 656                 if (count++ == E1000_TX_WEIGHT)
 657                         break;
 658         }
 659
 660         tx_ring->next_to_clean = i;
 661
 662 #define TX_WAKE_THRESHOLD 32
 663         if (cleaned && netif_carrier_ok(netdev) &&
 664                      e1000_desc_unused(tx_ring) >= TX_WAKE_THRESHOLD) {
 665                 /* Make sure that anybody stopping the queue after this
 666                  * sees the new next_to_clean.
 667                  */
 668                 smp_mb();
 669
 670                 if (netif_queue_stopped(netdev) &&
 671                     !(test_bit(__E1000_DOWN, &adapter->state))) {
 672                         netif_wake_queue(netdev);
 673                         ++adapter->restart_queue;
 674                 }
 675         }
 676
 677         if (adapter->detect_tx_hung) {
 678                 /*
 679                  * Detect a transmit hang in hardware, this serializes the
 680                  * check with the clearing of time_stamp and movement of i
 681                  */
 682                 adapter->detect_tx_hung = 0;
 683                 if (tx_ring->buffer_info[eop].dma &&
 684                     time_after(jiffies, tx_ring->buffer_info[eop].time_stamp
 685                                + (adapter->tx_timeout_factor * HZ))
 686                     && !(er32(STATUS) & E1000_STATUS_TXOFF)) {
 687                         e1000_print_tx_hang(adapter);
 688                         netif_stop_queue(netdev);
 689                 }
 690         }
 691         adapter->total_tx_bytes += total_tx_bytes;
 692         adapter->total_tx_packets += total_tx_packets;
 693         adapter->net_stats.tx_bytes += total_tx_bytes;
 694         adapter->net_stats.tx_packets += total_tx_packets;
 695         return cleaned;
 696 }
 697
 698 /**
 699  * e1000_clean_rx_irq_ps - Send received data up the network stack; packet split
 700  * @adapter: board private structure
 701  *
 702  * the return value indicates whether actual cleaning was done, there
 703  * is no guarantee that everything was cleaned
 704  **/
 705 static bool e1000_clean_rx_irq_ps(struct e1000_adapter *adapter,
 706                                   int *work_done, int work_to_do)
 707 {
 708         union e1000_rx_desc_packet_split *rx_desc, *next_rxd;
 709         struct net_device *netdev = adapter->netdev;
 710         struct pci_dev *pdev = adapter->pdev;
 711         struct e1000_ring *rx_ring = adapter->rx_ring;
 712         struct e1000_buffer *buffer_info, *next_buffer;
 713         struct e1000_ps_page *ps_page;
 714         struct sk_buff *skb;
 715         unsigned int i, j;
 716         u32 length, staterr;
 717         int cleaned_count = 0;
 718         bool cleaned = 0;
 719         unsigned int total_rx_bytes = 0, total_rx_packets = 0;
 720
 721         i = rx_ring->next_to_clean;
 722         rx_desc = E1000_RX_DESC_PS(*rx_ring, i);
 723         staterr = le32_to_cpu(rx_desc->wb.middle.status_error);
 724         buffer_info = &rx_ring->buffer_info[i];
 725
 726         while (staterr & E1000_RXD_STAT_DD) {
 727                 if (*work_done >= work_to_do)
 728                         break;
 729                 (*work_done)++;
 730                 skb = buffer_info->skb;
 731
 732                 /* in the packet split case this is header only */
 733                 prefetch(skb->data - NET_IP_ALIGN);
 734
 735                 i++;
 736                 if (i == rx_ring->count)
 737                         i = 0;
 738                 next_rxd = E1000_RX_DESC_PS(*rx_ring, i);
 739                 prefetch(next_rxd);
 740
 741                 next_buffer = &rx_ring->buffer_info[i];
 742
 743                 cleaned = 1;
 744                 cleaned_count++;
 745                 pci_unmap_single(pdev, buffer_info->dma,
 746                                  adapter->rx_ps_bsize0,
 747                                  PCI_DMA_FROMDEVICE);
 748                 buffer_info->dma = 0;
 749
 750                 if (!(staterr & E1000_RXD_STAT_EOP)) {
 751                         e_dbg("%s: Packet Split buffers didn't pick up the "
 752                               "full packet\n", netdev->name);
 753                         dev_kfree_skb_irq(skb);
 754                         goto next_desc;
 755                 }
 756
 757                 if (staterr & E1000_RXDEXT_ERR_FRAME_ERR_MASK) {
 758                         dev_kfree_skb_irq(skb);
 759                         goto next_desc;
 760                 }
 761
 762                 length = le16_to_cpu(rx_desc->wb.middle.length0);
 763
 764                 if (!length) {
 765                         e_dbg("%s: Last part of the packet spanning multiple "
 766                               "descriptors\n", netdev->name);
 767                         dev_kfree_skb_irq(skb);
 768                         goto next_desc;
 769                 }
 770
 771                 /* Good Receive */
 772                 skb_put(skb, length);
 773
 774                 {
 775                 /*
 776                  * this looks ugly, but it seems compiler issues make it
 777                  * more efficient than reusing j
 778                  */
 779                 int l1 = le16_to_cpu(rx_desc->wb.upper.length[0]);
 780
 781                 /*
 782                  * page alloc/put takes too long and effects small packet
 783                  * throughput, so unsplit small packets and save the alloc/put
 784                  * only valid in softirq (napi) context to call kmap_*
 785                  */
 786                 if (l1 && (l1 <= copybreak) &&
 787                     ((length + l1) <= adapter->rx_ps_bsize0)) {
 788                         u8 *vaddr;
 789
 790                         ps_page = &buffer_info->ps_pages[0];
 791
 792                         /*
 793                          * there is no documentation about how to call
 794                          * kmap_atomic, so we can't hold the mapping
 795                          * very long
 796                          */
 797                         pci_dma_sync_single_for_cpu(pdev, ps_page->dma,
 798                                 PAGE_SIZE, PCI_DMA_FROMDEVICE);
 799                         vaddr = kmap_atomic(ps_page->page, KM_SKB_DATA_SOFTIRQ);
 800                         memcpy(skb_tail_pointer(skb), vaddr, l1);
 801                         kunmap_atomic(vaddr, KM_SKB_DATA_SOFTIRQ);
 802                         pci_dma_sync_single_for_device(pdev, ps_page->dma,
 803                                 PAGE_SIZE, PCI_DMA_FROMDEVICE);
 804
 805                         skb_put(skb, l1);
 806                         goto copydone;
 807                 } /* if */
 808                 }
 809
 810                 for (j = 0; j < PS_PAGE_BUFFERS; j++) {
 811                         length = le16_to_cpu(rx_desc->wb.upper.length[j]);
 812                         if (!length)
 813                                 break;
 814
 815                         ps_page = &buffer_info->ps_pages[j];
 816                         pci_unmap_page(pdev, ps_page->dma, PAGE_SIZE,
 817                                        PCI_DMA_FROMDEVICE);
 818                         ps_page->dma = 0;
 819                         skb_fill_page_desc(skb, j, ps_page->page, 0, length);
 820                         ps_page->page = NULL;
 821                         skb->len += length;
 822                         skb->data_len += length;
 823                         skb->truesize += length;
 824                 }
 825
 826 copydone:
 827                 total_rx_bytes += skb->len;
 828                 total_rx_packets++;
 829
 830                 e1000_rx_checksum(adapter, staterr, le16_to_cpu(
 831                         rx_desc->wb.lower.hi_dword.csum_ip.csum), skb);
 832
 833                 if (rx_desc->wb.upper.header_status &
 834                            cpu_to_le16(E1000_RXDPS_HDRSTAT_HDRSP))
 835                         adapter->rx_hdr_split++;
 836
 837                 e1000_receive_skb(adapter, netdev, skb,
 838                                   staterr, rx_desc->wb.middle.vlan);
 839
 840 next_desc:
 841                 rx_desc->wb.middle.status_error &= cpu_to_le32(~0xFF);
 842                 buffer_info->skb = NULL;
 843
 844                 /* return some buffers to hardware, one at a time is too slow */
 845                 if (cleaned_count >= E1000_RX_BUFFER_WRITE) {
 846                         adapter->alloc_rx_buf(adapter, cleaned_count);
 847                         cleaned_count = 0;
 848                 }
 849
 850                 /* use prefetched values */
 851                 rx_desc = next_rxd;
 852                 buffer_info = next_buffer;
 853
 854                 staterr = le32_to_cpu(rx_desc->wb.middle.status_error);
 855         }
 856         rx_ring->next_to_clean = i;
 857
 858         cleaned_count = e1000_desc_unused(rx_ring);
 859         if (cleaned_count)
 860                 adapter->alloc_rx_buf(adapter, cleaned_count);
 861
 862         adapter->total_rx_bytes += total_rx_bytes;
 863         adapter->total_rx_packets += total_rx_packets;
 864         adapter->net_stats.rx_bytes += total_rx_bytes;
 865         adapter->net_stats.rx_packets += total_rx_packets;
 866         return cleaned;
 867 }
 868
 869 /**
 870  * e1000_consume_page - helper function
 871  **/
 872 static void e1000_consume_page(struct e1000_buffer *bi, struct sk_buff *skb,
 873                                u16 length)
 874 {
 875         bi->page = NULL;
 876         skb->len += length;
 877         skb->data_len += length;
 878         skb->truesize += length;
 879 }
 880
 881 /**
 882  * e1000_clean_jumbo_rx_irq - Send received data up the network stack; legacy
 883  * @adapter: board private structure
 884  *
 885  * the return value indicates whether actual cleaning was done, there
 886  * is no guarantee that everything was cleaned
 887  **/
 888
 889 static bool e1000_clean_jumbo_rx_irq(struct e1000_adapter *adapter,
 890                                      int *work_done, int work_to_do)
 891 {
 892         struct net_device *netdev = adapter->netdev;
 893         struct pci_dev *pdev = adapter->pdev;
 894         struct e1000_ring *rx_ring = adapter->rx_ring;
 895         struct e1000_rx_desc *rx_desc, *next_rxd;
 896         struct e1000_buffer *buffer_info, *next_buffer;
 897         u32 length;
 898         unsigned int i;
 899         int cleaned_count = 0;
 900         bool cleaned = false;
 901         unsigned int total_rx_bytes=0, total_rx_packets=0;
 902
 903         i = rx_ring->next_to_clean;
 904         rx_desc = E1000_RX_DESC(*rx_ring, i);
 905         buffer_info = &rx_ring->buffer_info[i];
 906
 907         while (rx_desc->status & E1000_RXD_STAT_DD) {
 908                 struct sk_buff *skb;
 909                 u8 status;
 910
 911                 if (*work_done >= work_to_do)
 912                         break;
 913                 (*work_done)++;
 914
 915                 status = rx_desc->status;
 916                 skb = buffer_info->skb;
 917                 buffer_info->skb = NULL;
 918
 919                 ++i;
 920                 if (i == rx_ring->count)
 921                         i = 0;
 922                 next_rxd = E1000_RX_DESC(*rx_ring, i);
 923                 prefetch(next_rxd);
 924
 925                 next_buffer = &rx_ring->buffer_info[i];
 926
 927                 cleaned = true;
 928                 cleaned_count++;
 929                 pci_unmap_page(pdev, buffer_info->dma, PAGE_SIZE,
 930                                PCI_DMA_FROMDEVICE);
 931                 buffer_info->dma = 0;
 932
 933                 length = le16_to_cpu(rx_desc->length);
 934
 935                 /* errors is only valid for DD + EOP descriptors */
 936                 if (unlikely((status & E1000_RXD_STAT_EOP) &&
 937                     (rx_desc->errors & E1000_RXD_ERR_FRAME_ERR_MASK))) {
 938                                 /* recycle both page and skb */
 939                                 buffer_info->skb = skb;
 940                                 /* an error means any chain goes out the window
 941                                  * too */
 942                                 if (rx_ring->rx_skb_top)
 943                                         dev_kfree_skb(rx_ring->rx_skb_top);
 944                                 rx_ring->rx_skb_top = NULL;
 945                                 goto next_desc;
 946                 }
 947
 948 #define rxtop rx_ring->rx_skb_top
 949                 if (!(status & E1000_RXD_STAT_EOP)) {
 950                         /* this descriptor is only the beginning (or middle) */
 951                         if (!rxtop) {
 952                                 /* this is the beginning of a chain */
 953                                 rxtop = skb;
 954                                 skb_fill_page_desc(rxtop, 0, buffer_info->page,
 955                                                    0, length);
 956                         } else {
 957                                 /* this is the middle of a chain */
 958                                 skb_fill_page_desc(rxtop,
 959                                     skb_shinfo(rxtop)->nr_frags,
 960                                     buffer_info->page, 0, length);
 961                                 /* re-use the skb, only consumed the page */
 962                                 buffer_info->skb = skb;
 963                         }
 964                         e1000_consume_page(buffer_info, rxtop, length);
 965                         goto next_desc;
 966                 } else {
 967                         if (rxtop) {
 968                                 /* end of the chain */
 969                                 skb_fill_page_desc(rxtop,
 970                                     skb_shinfo(rxtop)->nr_frags,
 971                                     buffer_info->page, 0, length);
 972                                 /* re-use the current skb, we only consumed the
 973                                  * page */
 974                                 buffer_info->skb = skb;
 975                                 skb = rxtop;
 976                                 rxtop = NULL;
 977                                 e1000_consume_page(buffer_info, skb, length);
 978                         } else {
 979                                 /* no chain, got EOP, this buf is the packet
 980                                  * copybreak to save the put_page/alloc_page */
 981                                 if (length <= copybreak &&
 982                                     skb_tailroom(skb) >= length) {
 983                                         u8 *vaddr;
 984                                         vaddr = kmap_atomic(buffer_info->page,
 985                                                            KM_SKB_DATA_SOFTIRQ);
 986                                         memcpy(skb_tail_pointer(skb), vaddr,
 987                                                length);
 988                                         kunmap_atomic(vaddr,
 989                                                       KM_SKB_DATA_SOFTIRQ);
 990                                         /* re-use the page, so don't erase
 991                                          * buffer_info->page */
 992                                         skb_put(skb, length);
 993                                 } else {
 994                                         skb_fill_page_desc(skb, 0,
 995                                                            buffer_info->page, 0,
 996                                                            length);
 997                                         e1000_consume_page(buffer_info, skb,
 998                                                            length);
 999                                 }
1000                         }
1001                 }
1002
1003                 /* Receive Checksum Offload XXX recompute due to CRC strip? */
1004                 e1000_rx_checksum(adapter,
1005                                   (u32)(status) |
1006                                   ((u32)(rx_desc->errors) << 24),
1007                                   le16_to_cpu(rx_desc->csum), skb);
1008
1009                 /* probably a little skewed due to removing CRC */
1010                 total_rx_bytes += skb->len;
1011                 total_rx_packets++;
1012
1013                 /* eth type trans needs skb->data to point to something */
1014                 if (!pskb_may_pull(skb, ETH_HLEN)) {
1015                         e_err("pskb_may_pull failed.\n");
1016                         dev_kfree_skb(skb);
1017                         goto next_desc;
1018                 }
1019
1020                 e1000_receive_skb(adapter, netdev, skb, status,
1021                                   rx_desc->special);
1022
1023 next_desc:
1024                 rx_desc->status = 0;
1025
1026                 /* return some buffers to hardware, one at a time is too slow */
1027                 if (unlikely(cleaned_count >= E1000_RX_BUFFER_WRITE)) {
1028                         adapter->alloc_rx_buf(adapter, cleaned_count);
1029                         cleaned_count = 0;
1030                 }
1031
1032                 /* use prefetched values */
1033                 rx_desc = next_rxd;
1034                 buffer_info = next_buffer;
1035         }
1036         rx_ring->next_to_clean = i;
1037
1038         cleaned_count = e1000_desc_unused(rx_ring);
1039         if (cleaned_count)
1040                 adapter->alloc_rx_buf(adapter, cleaned_count);
1041
1042         adapter->total_rx_bytes += total_rx_bytes;
1043         adapter->total_rx_packets += total_rx_packets;
1044         adapter->net_stats.rx_bytes += total_rx_bytes;
1045         adapter->net_stats.rx_packets += total_rx_packets;
1046         return cleaned;
1047 }
1048
1049 /**
1050  * e1000_clean_rx_ring - Free Rx Buffers per Queue
1051  * @adapter: board private structure
1052  **/
1053 static void e1000_clean_rx_ring(struct e1000_adapter *adapter)
1054 {
1055         struct e1000_ring *rx_ring = adapter->rx_ring;
1056         struct e1000_buffer *buffer_info;
1057         struct e1000_ps_page *ps_page;
1058         struct pci_dev *pdev = adapter->pdev;
1059         unsigned int i, j;
1060
1061         /* Free all the Rx ring sk_buffs */
1062         for (i = 0; i < rx_ring->count; i++) {
1063                 buffer_info = &rx_ring->buffer_info[i];
1064                 if (buffer_info->dma) {
1065                         if (adapter->clean_rx == e1000_clean_rx_irq)
1066                                 pci_unmap_single(pdev, buffer_info->dma,
1067                                                  adapter->rx_buffer_len,
1068                                                  PCI_DMA_FROMDEVICE);
1069                         else if (adapter->clean_rx == e1000_clean_jumbo_rx_irq)
1070                                 pci_unmap_page(pdev, buffer_info->dma,
1071                                                PAGE_SIZE,
1072                                                PCI_DMA_FROMDEVICE);
1073                         else if (adapter->clean_rx == e1000_clean_rx_irq_ps)
1074                                 pci_unmap_single(pdev, buffer_info->dma,
1075                                                  adapter->rx_ps_bsize0,
1076                                                  PCI_DMA_FROMDEVICE);
1077                         buffer_info->dma = 0;
1078                 }
1079
1080                 if (buffer_info->page) {
1081                         put_page(buffer_info->page);
1082                         buffer_info->page = NULL;
1083                 }
1084
1085                 if (buffer_info->skb) {
1086                         dev_kfree_skb(buffer_info->skb);
1087                         buffer_info->skb = NULL;
1088                 }
1089
1090                 for (j = 0; j < PS_PAGE_BUFFERS; j++) {
1091                         ps_page = &buffer_info->ps_pages[j];
1092                         if (!ps_page->page)
1093                                 break;
1094                         pci_unmap_page(pdev, ps_page->dma, PAGE_SIZE,
1095                                        PCI_DMA_FROMDEVICE);
1096                         ps_page->dma = 0;
1097                         put_page(ps_page->page);
1098                         ps_page->page = NULL;
1099                 }
1100         }
1101
1102         /* there also may be some cached data from a chained receive */
1103         if (rx_ring->rx_skb_top) {
1104                 dev_kfree_skb(rx_ring->rx_skb_top);
1105                 rx_ring->rx_skb_top = NULL;
1106         }
1107
1108         /* Zero out the descriptor ring */
1109         memset(rx_ring->desc, 0, rx_ring->size);
1110
1111         rx_ring->next_to_clean = 0;
1112         rx_ring->next_to_use = 0;
1113
1114         writel(0, adapter->hw.hw_addr + rx_ring->head);
1115         writel(0, adapter->hw.hw_addr + rx_ring->tail);
1116 }
1117
1118 /**
1119  * e1000_intr_msi - Interrupt Handler
1120  * @irq: interrupt number
1121  * @data: pointer to a network interface device structure
1122  **/
1123 static irqreturn_t e1000_intr_msi(int irq, void *data)
1124 {
1125         struct net_device *netdev = data;
1126         struct e1000_adapter *adapter = netdev_priv(netdev);
1127         struct e1000_hw *hw = &adapter->hw;
1128         u32 icr = er32(ICR);
1129
1130         /*
1131          * read ICR disables interrupts using IAM
1132          */
1133
1134         if (icr & (E1000_ICR_RXSEQ | E1000_ICR_LSC)) {
1135                 hw->mac.get_link_status = 1;
1136                 /*
1137                  * ICH8 workaround-- Call gig speed drop workaround on cable
1138                  * disconnect (LSC) before accessing any PHY registers
1139                  */
1140                 if ((adapter->flags & FLAG_LSC_GIG_SPEED_DROP) &&
1141                     (!(er32(STATUS) & E1000_STATUS_LU)))
1142                         e1000e_gig_downshift_workaround_ich8lan(hw);
1143
1144                 /*
1145                  * 80003ES2LAN workaround-- For packet buffer work-around on
1146                  * link down event; disable receives here in the ISR and reset
1147                  * adapter in watchdog
1148                  */
1149                 if (netif_carrier_ok(netdev) &&
1150                     adapter->flags & FLAG_RX_NEEDS_RESTART) {
1151                         /* disable receives */
1152                         u32 rctl = er32(RCTL);
1153                         ew32(RCTL, rctl & ~E1000_RCTL_EN);
1154                         adapter->flags |= FLAG_RX_RESTART_NOW;
1155                 }
1156                 /* guard against interrupt when we're going down */
1157                 if (!test_bit(__E1000_DOWN, &adapter->state))
1158                         mod_timer(&adapter->watchdog_timer, jiffies + 1);
1159         }
1160
1161         if (netif_rx_schedule_prep(netdev, &adapter->napi)) {
1162                 adapter->total_tx_bytes = 0;
1163                 adapter->total_tx_packets = 0;
1164                 adapter->total_rx_bytes = 0;
1165                 adapter->total_rx_packets = 0;
1166                 __netif_rx_schedule(netdev, &adapter->napi);
1167         }
1168
1169         return IRQ_HANDLED;
1170 }
1171
1172 /**
1173  * e1000_intr - Interrupt Handler
1174  * @irq: interrupt number
1175  * @data: pointer to a network interface device structure
1176  **/
1177 static irqreturn_t e1000_intr(int irq, void *data)
1178 {
1179         struct net_device *netdev = data;
1180         struct e1000_adapter *adapter = netdev_priv(netdev);
1181         struct e1000_hw *hw = &adapter->hw;
1182
1183         u32 rctl, icr = er32(ICR);
1184         if (!icr)
1185                 return IRQ_NONE;  /* Not our interrupt */
1186
1187         /*
1188          * IMS will not auto-mask if INT_ASSERTED is not set, and if it is
1189          * not set, then the adapter didn't send an interrupt
1190          */
1191         if (!(icr & E1000_ICR_INT_ASSERTED))
1192                 return IRQ_NONE;
1193
1194         /*
1195          * Interrupt Auto-Mask...upon reading ICR,
1196          * interrupts are masked.  No need for the
1197          * IMC write
1198          */
1199
1200         if (icr & (E1000_ICR_RXSEQ | E1000_ICR_LSC)) {
1201                 hw->mac.get_link_status = 1;
1202                 /*
1203                  * ICH8 workaround-- Call gig speed drop workaround on cable
1204                  * disconnect (LSC) before accessing any PHY registers
1205                  */
1206                 if ((adapter->flags & FLAG_LSC_GIG_SPEED_DROP) &&
1207                     (!(er32(STATUS) & E1000_STATUS_LU)))
1208                         e1000e_gig_downshift_workaround_ich8lan(hw);
1209
1210                 /*
1211                  * 80003ES2LAN workaround--
1212                  * For packet buffer work-around on link down event;
1213                  * disable receives here in the ISR and
1214                  * reset adapter in watchdog
1215                  */
1216                 if (netif_carrier_ok(netdev) &&
1217                     (adapter->flags & FLAG_RX_NEEDS_RESTART)) {
1218                         /* disable receives */
1219                         rctl = er32(RCTL);
1220                         ew32(RCTL, rctl & ~E1000_RCTL_EN);
1221                         adapter->flags |= FLAG_RX_RESTART_NOW;
1222                 }
1223                 /* guard against interrupt when we're going down */
1224                 if (!test_bit(__E1000_DOWN, &adapter->state))
1225                         mod_timer(&adapter->watchdog_timer, jiffies + 1);
1226         }
1227
1228         if (netif_rx_schedule_prep(netdev, &adapter->napi)) {
1229                 adapter->total_tx_bytes = 0;
1230                 adapter->total_tx_packets = 0;
1231                 adapter->total_rx_bytes = 0;
1232                 adapter->total_rx_packets = 0;
1233                 __netif_rx_schedule(netdev, &adapter->napi);
1234         }
1235
1236         return IRQ_HANDLED;
1237 }
1238
1239 static int e1000_request_irq(struct e1000_adapter *adapter)
1240 {
1241         struct net_device *netdev = adapter->netdev;
1242         irq_handler_t handler = e1000_intr;
1243         int irq_flags = IRQF_SHARED;
1244         int err;
1245
1246         if (!pci_enable_msi(adapter->pdev)) {
1247                 adapter->flags |= FLAG_MSI_ENABLED;
1248                 handler = e1000_intr_msi;
1249                 irq_flags = 0;
1250         }
1251
1252         err = request_irq(adapter->pdev->irq, handler, irq_flags, netdev->name,
1253                           netdev);
1254         if (err) {
1255                 e_err("Unable to allocate %s interrupt (return: %d)\n",
1256                       adapter->flags & FLAG_MSI_ENABLED ? "MSI":"INTx", err);
1257                 if (adapter->flags & FLAG_MSI_ENABLED)
1258                         pci_disable_msi(adapter->pdev);
1259         }
1260
1261         return err;
1262 }
1263
1264 static void e1000_free_irq(struct e1000_adapter *adapter)
1265 {
1266         struct net_device *netdev = adapter->netdev;
1267
1268         free_irq(adapter->pdev->irq, netdev);
1269         if (adapter->flags & FLAG_MSI_ENABLED) {
1270                 pci_disable_msi(adapter->pdev);
1271                 adapter->flags &= ~FLAG_MSI_ENABLED;
1272         }
1273 }
1274
1275 /**
1276  * e1000_irq_disable - Mask off interrupt generation on the NIC
1277  **/
1278 static void e1000_irq_disable(struct e1000_adapter *adapter)
1279 {
1280         struct e1000_hw *hw = &adapter->hw;
1281
1282         ew32(IMC, ~0);
1283         e1e_flush();
1284         synchronize_irq(adapter->pdev->irq);
1285 }
1286
1287 /**
1288  * e1000_irq_enable - Enable default interrupt generation settings
1289  **/
1290 static void e1000_irq_enable(struct e1000_adapter *adapter)
1291 {
1292         struct e1000_hw *hw = &adapter->hw;
1293
1294         ew32(IMS, IMS_ENABLE_MASK);
1295         e1e_flush();
1296 }
1297
1298 /**
1299  * e1000_get_hw_control - get control of the h/w from f/w
1300  * @adapter: address of board private structure
1301  *
1302  * e1000_get_hw_control sets {CTRL_EXT|SWSM}:DRV_LOAD bit.
1303  * For ASF and Pass Through versions of f/w this means that
1304  * the driver is loaded. For AMT version (only with 82573)
1305  * of the f/w this means that the network i/f is open.
1306  **/
1307 static void e1000_get_hw_control(struct e1000_adapter *adapter)
1308 {
1309         struct e1000_hw *hw = &adapter->hw;
1310         u32 ctrl_ext;
1311         u32 swsm;
1312
1313         /* Let firmware know the driver has taken over */
1314         if (adapter->flags & FLAG_HAS_SWSM_ON_LOAD) {
1315                 swsm = er32(SWSM);
1316                 ew32(SWSM, swsm | E1000_SWSM_DRV_LOAD);
1317         } else if (adapter->flags & FLAG_HAS_CTRLEXT_ON_LOAD) {
1318                 ctrl_ext = er32(CTRL_EXT);
1319                 ew32(CTRL_EXT, ctrl_ext | E1000_CTRL_EXT_DRV_LOAD);
1320         }
1321 }
1322
1323 /**
1324  * e1000_release_hw_control - release control of the h/w to f/w
1325  * @adapter: address of board private structure
1326  *
1327  * e1000_release_hw_control resets {CTRL_EXT|SWSM}:DRV_LOAD bit.
1328  * For ASF and Pass Through versions of f/w this means that the
1329  * driver is no longer loaded. For AMT version (only with 82573) i
1330  * of the f/w this means that the network i/f is closed.
1331  *
1332  **/
1333 static void e1000_release_hw_control(struct e1000_adapter *adapter)
1334 {
1335         struct e1000_hw *hw = &adapter->hw;
1336         u32 ctrl_ext;
1337         u32 swsm;
1338
1339         /* Let firmware taken over control of h/w */
1340         if (adapter->flags & FLAG_HAS_SWSM_ON_LOAD) {
1341                 swsm = er32(SWSM);
1342                 ew32(SWSM, swsm & ~E1000_SWSM_DRV_LOAD);
1343         } else if (adapter->flags & FLAG_HAS_CTRLEXT_ON_LOAD) {
1344                 ctrl_ext = er32(CTRL_EXT);
1345                 ew32(CTRL_EXT, ctrl_ext & ~E1000_CTRL_EXT_DRV_LOAD);
1346         }
1347 }
1348
1349 /**
1350  * @e1000_alloc_ring - allocate memory for a ring structure
1351  **/
1352 static int e1000_alloc_ring_dma(struct e1000_adapter *adapter,
1353                                 struct e1000_ring *ring)
1354 {
1355         struct pci_dev *pdev = adapter->pdev;
1356
1357         ring->desc = dma_alloc_coherent(&pdev->dev, ring->size, &ring->dma,
1358                                         GFP_KERNEL);
1359         if (!ring->desc)
1360                 return -ENOMEM;
1361
1362         return 0;
1363 }
1364
1365 /**
1366  * e1000e_setup_tx_resources - allocate Tx resources (Descriptors)
1367  * @adapter: board private structure
1368  *
1369  * Return 0 on success, negative on failure
1370  **/
1371 int e1000e_setup_tx_resources(struct e1000_adapter *adapter)
1372 {
1373         struct e1000_ring *tx_ring = adapter->tx_ring;
1374         int err = -ENOMEM, size;
1375
1376         size = sizeof(struct e1000_buffer) * tx_ring->count;
1377         tx_ring->buffer_info = vmalloc(size);
1378         if (!tx_ring->buffer_info)
1379                 goto err;
1380         memset(tx_ring->buffer_info, 0, size);
1381
1382         /* round up to nearest 4K */
1383         tx_ring->size = tx_ring->count * sizeof(struct e1000_tx_desc);
1384         tx_ring->size = ALIGN(tx_ring->size, 4096);
1385
1386         err = e1000_alloc_ring_dma(adapter, tx_ring);
1387         if (err)
1388                 goto err;
1389
1390         tx_ring->next_to_use = 0;
1391         tx_ring->next_to_clean = 0;
1392         spin_lock_init(&adapter->tx_queue_lock);
1393
1394         return 0;
1395 err:
1396         vfree(tx_ring->buffer_info);
1397         e_err("Unable to allocate memory for the transmit descriptor ring\n");
1398         return err;
1399 }
1400
1401 /**
1402  * e1000e_setup_rx_resources - allocate Rx resources (Descriptors)
1403  * @adapter: board private structure
1404  *
1405  * Returns 0 on success, negative on failure
1406  **/
1407 int e1000e_setup_rx_resources(struct e1000_adapter *adapter)
1408 {
1409         struct e1000_ring *rx_ring = adapter->rx_ring;
1410         struct e1000_buffer *buffer_info;
1411         int i, size, desc_len, err = -ENOMEM;
1412
1413         size = sizeof(struct e1000_buffer) * rx_ring->count;
1414         rx_ring->buffer_info = vmalloc(size);
1415         if (!rx_ring->buffer_info)
1416                 goto err;
1417         memset(rx_ring->buffer_info, 0, size);
1418
1419         for (i = 0; i < rx_ring->count; i++) {
1420                 buffer_info = &rx_ring->buffer_info[i];
1421                 buffer_info->ps_pages = kcalloc(PS_PAGE_BUFFERS,
1422                                                 sizeof(struct e1000_ps_page),
1423                                                 GFP_KERNEL);
1424                 if (!buffer_info->ps_pages)
1425                         goto err_pages;
1426         }
1427
1428         desc_len = sizeof(union e1000_rx_desc_packet_split);
1429
1430         /* Round up to nearest 4K */
1431         rx_ring->size = rx_ring->count * desc_len;
1432         rx_ring->size = ALIGN(rx_ring->size, 4096);
1433
1434         err = e1000_alloc_ring_dma(adapter, rx_ring);
1435         if (err)
1436                 goto err_pages;
1437
1438         rx_ring->next_to_clean = 0;
1439         rx_ring->next_to_use = 0;
1440         rx_ring->rx_skb_top = NULL;
1441
1442         return 0;
1443
1444 err_pages:
1445         for (i = 0; i < rx_ring->count; i++) {
1446                 buffer_info = &rx_ring->buffer_info[i];
1447                 kfree(buffer_info->ps_pages);
1448         }
1449 err:
1450         vfree(rx_ring->buffer_info);
1451         e_err("Unable to allocate memory for the transmit descriptor ring\n");
1452         return err;
1453 }
1454
1455 /**
1456  * e1000_clean_tx_ring - Free Tx Buffers
1457  * @adapter: board private structure
1458  **/
1459 static void e1000_clean_tx_ring(struct e1000_adapter *adapter)
1460 {
1461         struct e1000_ring *tx_ring = adapter->tx_ring;
1462         struct e1000_buffer *buffer_info;
1463         unsigned long size;
1464         unsigned int i;
1465
1466         for (i = 0; i < tx_ring->count; i++) {
1467                 buffer_info = &tx_ring->buffer_info[i];
1468                 e1000_put_txbuf(adapter, buffer_info);
1469         }
1470
1471         size = sizeof(struct e1000_buffer) * tx_ring->count;
1472         memset(tx_ring->buffer_info, 0, size);
1473
1474         memset(tx_ring->desc, 0, tx_ring->size);
1475
1476         tx_ring->next_to_use = 0;
1477         tx_ring->next_to_clean = 0;
1478
1479         writel(0, adapter->hw.hw_addr + tx_ring->head);
1480         writel(0, adapter->hw.hw_addr + tx_ring->tail);
1481 }
1482
1483 /**
1484  * e1000e_free_tx_resources - Free Tx Resources per Queue
1485  * @adapter: board private structure
1486  *
1487  * Free all transmit software resources
1488  **/
1489 void e1000e_free_tx_resources(struct e1000_adapter *adapter)
1490 {
1491         struct pci_dev *pdev = adapter->pdev;
1492         struct e1000_ring *tx_ring = adapter->tx_ring;
1493
1494         e1000_clean_tx_ring(adapter);
1495
1496         vfree(tx_ring->buffer_info);
1497         tx_ring->buffer_info = NULL;
1498
1499         dma_free_coherent(&pdev->dev, tx_ring->size, tx_ring->desc,
1500                           tx_ring->dma);
1501         tx_ring->desc = NULL;
1502 }
1503
1504 /**
1505  * e1000e_free_rx_resources - Free Rx Resources
1506  * @adapter: board private structure
1507  *
1508  * Free all receive software resources
1509  **/
1510
1511 void e1000e_free_rx_resources(struct e1000_adapter *adapter)
1512 {
1513         struct pci_dev *pdev = adapter->pdev;
1514         struct e1000_ring *rx_ring = adapter->rx_ring;
1515         int i;
1516
1517         e1000_clean_rx_ring(adapter);
1518
1519         for (i = 0; i < rx_ring->count; i++) {
1520                 kfree(rx_ring->buffer_info[i].ps_pages);
1521         }
1522
1523         vfree(rx_ring->buffer_info);
1524         rx_ring->buffer_info = NULL;
1525
1526         dma_free_coherent(&pdev->dev, rx_ring->size, rx_ring->desc,
1527                           rx_ring->dma);
1528         rx_ring->desc = NULL;
1529 }
1530
1531 /**
1532  * e1000_update_itr - update the dynamic ITR value based on statistics
1533  * @adapter: pointer to adapter
1534  * @itr_setting: current adapter->itr
1535  * @packets: the number of packets during this measurement interval
1536  * @bytes: the number of bytes during this measurement interval
1537  *
1538  *      Stores a new ITR value based on packets and byte
1539  *      counts during the last interrupt.  The advantage of per interrupt
1540  *      computation is faster updates and more accurate ITR for the current
1541  *      traffic pattern.  Constants in this function were computed
1542  *      based on theoretical maximum wire speed and thresholds were set based
1543  *      on testing data as well as attempting to minimize response time
1544  *      while increasing bulk throughput.
1545  *      this functionality is controlled by the InterruptThrottleRate module
1546  *      parameter (see e1000_param.c)
1547  **/
1548 static unsigned int e1000_update_itr(struct e1000_adapter *adapter,
1549                                      u16 itr_setting, int packets,
1550                                      int bytes)
1551 {
1552         unsigned int retval = itr_setting;
1553
1554         if (packets == 0)
1555                 goto update_itr_done;
1556
1557         switch (itr_setting) {
1558         case lowest_latency:
1559                 /* handle TSO and jumbo frames */
1560                 if (bytes/packets > 8000)
1561                         retval = bulk_latency;
1562                 else if ((packets < 5) && (bytes > 512)) {
1563                         retval = low_latency;
1564                 }
1565                 break;
1566         case low_latency:  /* 50 usec aka 20000 ints/s */
1567                 if (bytes > 10000) {
1568                         /* this if handles the TSO accounting */
1569                         if (bytes/packets > 8000) {
1570                                 retval = bulk_latency;
1571                         } else if ((packets < 10) || ((bytes/packets) > 1200)) {
1572                                 retval = bulk_latency;
1573                         } else if ((packets > 35)) {
1574                                 retval = lowest_latency;
1575                         }
1576                 } else if (bytes/packets > 2000) {
1577                         retval = bulk_latency;
1578                 } else if (packets <= 2 && bytes < 512) {
1579                         retval = lowest_latency;
1580                 }
1581                 break;
1582         case bulk_latency: /* 250 usec aka 4000 ints/s */
1583                 if (bytes > 25000) {
1584                         if (packets > 35) {
1585                                 retval = low_latency;
1586                         }
1587                 } else if (bytes < 6000) {
1588                         retval = low_latency;
1589                 }
1590                 break;
1591         }
1592
1593 update_itr_done:
1594         return retval;
1595 }
1596
1597 static void e1000_set_itr(struct e1000_adapter *adapter)
1598 {
1599         struct e1000_hw *hw = &adapter->hw;
1600         u16 current_itr;
1601         u32 new_itr = adapter->itr;
1602
1603         /* for non-gigabit speeds, just fix the interrupt rate at 4000 */
1604         if (adapter->link_speed != SPEED_1000) {
1605                 current_itr = 0;
1606                 new_itr = 4000;
1607                 goto set_itr_now;
1608         }
1609
1610         adapter->tx_itr = e1000_update_itr(adapter,
1611                                     adapter->tx_itr,
1612                                     adapter->total_tx_packets,
1613                                     adapter->total_tx_bytes);
1614         /* conservative mode (itr 3) eliminates the lowest_latency setting */
1615         if (adapter->itr_setting == 3 && adapter->tx_itr == lowest_latency)
1616                 adapter->tx_itr = low_latency;
1617
1618         adapter->rx_itr = e1000_update_itr(adapter,
1619                                     adapter->rx_itr,
1620                                     adapter->total_rx_packets,
1621                                     adapter->total_rx_bytes);
1622         /* conservative mode (itr 3) eliminates the lowest_latency setting */
1623         if (adapter->itr_setting == 3 && adapter->rx_itr == lowest_latency)
1624                 adapter->rx_itr = low_latency;
1625
1626         current_itr = max(adapter->rx_itr, adapter->tx_itr);
1627
1628         switch (current_itr) {
1629         /* counts and packets in update_itr are dependent on these numbers */
1630         case lowest_latency:
1631                 new_itr = 70000;
1632                 break;
1633         case low_latency:
1634                 new_itr = 20000; /* aka hwitr = ~200 */
1635                 break;
1636         case bulk_latency:
1637                 new_itr = 4000;
1638                 break;
1639         default:
1640                 break;
1641         }
1642
1643 set_itr_now:
1644         if (new_itr != adapter->itr) {
1645                 /*
1646                  * this attempts to bias the interrupt rate towards Bulk
1647                  * by adding intermediate steps when interrupt rate is
1648                  * increasing
1649                  */
1650                 new_itr = new_itr > adapter->itr ?
1651                              min(adapter->itr + (new_itr >> 2), new_itr) :
1652                              new_itr;
1653                 adapter->itr = new_itr;
1654                 ew32(ITR, 1000000000 / (new_itr * 256));
1655         }
1656 }
1657
1658 /**
1659  * e1000_clean - NAPI Rx polling callback
1660  * @napi: struct associated with this polling callback
1661  * @budget: amount of packets driver is allowed to process this poll
1662  **/
1663 static int e1000_clean(struct napi_struct *napi, int budget)
1664 {
1665         struct e1000_adapter *adapter = container_of(napi, struct e1000_adapter, napi);
1666         struct net_device *poll_dev = adapter->netdev;
1667         int tx_cleaned = 0, work_done = 0;
1668
1669         /* Must NOT use netdev_priv macro here. */
1670         adapter = poll_dev->priv;
1671
1672         /*
1673          * e1000_clean is called per-cpu.  This lock protects
1674          * tx_ring from being cleaned by multiple cpus
1675          * simultaneously.  A failure obtaining the lock means
1676          * tx_ring is currently being cleaned anyway.
1677          */
1678         if (spin_trylock(&adapter->tx_queue_lock)) {
1679                 tx_cleaned = e1000_clean_tx_irq(adapter);
1680                 spin_unlock(&adapter->tx_queue_lock);
1681         }
1682
1683         adapter->clean_rx(adapter, &work_done, budget);
1684
1685         if (tx_cleaned)
1686                 work_done = budget;
1687
1688         /* If budget not fully consumed, exit the polling mode */
1689         if (work_done < budget) {
1690                 if (adapter->itr_setting & 3)
1691                         e1000_set_itr(adapter);
1692                 netif_rx_complete(poll_dev, napi);
1693                 e1000_irq_enable(adapter);
1694         }
1695
1696         return work_done;
1697 }
1698
1699 static void e1000_vlan_rx_add_vid(struct net_device *netdev, u16 vid)
1700 {
1701         struct e1000_adapter *adapter = netdev_priv(netdev);
1702         struct e1000_hw *hw = &adapter->hw;
1703         u32 vfta, index;
1704
1705         /* don't update vlan cookie if already programmed */
1706         if ((adapter->hw.mng_cookie.status &
1707              E1000_MNG_DHCP_COOKIE_STATUS_VLAN) &&
1708             (vid == adapter->mng_vlan_id))
1709                 return;
1710         /* add VID to filter table */
1711         index = (vid >> 5) & 0x7F;
1712         vfta = E1000_READ_REG_ARRAY(hw, E1000_VFTA, index);
1713         vfta |= (1 << (vid & 0x1F));
1714         e1000e_write_vfta(hw, index, vfta);
1715 }
1716
1717 static void e1000_vlan_rx_kill_vid(struct net_device *netdev, u16 vid)
1718 {
1719         struct e1000_adapter *adapter = netdev_priv(netdev);
1720         struct e1000_hw *hw = &adapter->hw;
1721         u32 vfta, index;
1722
1723         if (!test_bit(__E1000_DOWN, &adapter->state))
1724                 e1000_irq_disable(adapter);
1725         vlan_group_set_device(adapter->vlgrp, vid, NULL);
1726
1727         if (!test_bit(__E1000_DOWN, &adapter->state))
1728                 e1000_irq_enable(adapter);
1729
1730         if ((adapter->hw.mng_cookie.status &
1731              E1000_MNG_DHCP_COOKIE_STATUS_VLAN) &&
1732             (vid == adapter->mng_vlan_id)) {
1733                 /* release control to f/w */
1734                 e1000_release_hw_control(adapter);
1735                 return;
1736         }
1737
1738         /* remove VID from filter table */
1739         index = (vid >> 5) & 0x7F;
1740         vfta = E1000_READ_REG_ARRAY(hw, E1000_VFTA, index);
1741         vfta &= ~(1 << (vid & 0x1F));
1742         e1000e_write_vfta(hw, index, vfta);
1743 }
1744
1745 static void e1000_update_mng_vlan(struct e1000_adapter *adapter)
1746 {
1747         struct net_device *netdev = adapter->netdev;
1748         u16 vid = adapter->hw.mng_cookie.vlan_id;
1749         u16 old_vid = adapter->mng_vlan_id;
1750
1751         if (!adapter->vlgrp)
1752                 return;
1753
1754         if (!vlan_group_get_device(adapter->vlgrp, vid)) {
1755                 adapter->mng_vlan_id = E1000_MNG_VLAN_NONE;
1756                 if (adapter->hw.mng_cookie.status &
1757                         E1000_MNG_DHCP_COOKIE_STATUS_VLAN) {
1758                         e1000_vlan_rx_add_vid(netdev, vid);
1759                         adapter->mng_vlan_id = vid;
1760                 }
1761
1762                 if ((old_vid != (u16)E1000_MNG_VLAN_NONE) &&
1763                                 (vid != old_vid) &&
1764                     !vlan_group_get_device(adapter->vlgrp, old_vid))
1765                         e1000_vlan_rx_kill_vid(netdev, old_vid);
1766         } else {
1767                 adapter->mng_vlan_id = vid;
1768         }
1769 }
1770
1771
1772 static void e1000_vlan_rx_register(struct net_device *netdev,
1773                                    struct vlan_group *grp)
1774 {
1775         struct e1000_adapter *adapter = netdev_priv(netdev);
1776         struct e1000_hw *hw = &adapter->hw;
1777         u32 ctrl, rctl;
1778
1779         if (!test_bit(__E1000_DOWN, &adapter->state))
1780                 e1000_irq_disable(adapter);
1781         adapter->vlgrp = grp;
1782
1783         if (grp) {
1784                 /* enable VLAN tag insert/strip */
1785                 ctrl = er32(CTRL);
1786                 ctrl |= E1000_CTRL_VME;
1787                 ew32(CTRL, ctrl);
1788
1789                 if (adapter->flags & FLAG_HAS_HW_VLAN_FILTER) {
1790                         /* enable VLAN receive filtering */
1791                         rctl = er32(RCTL);
1792                         rctl &= ~E1000_RCTL_CFIEN;
1793                         ew32(RCTL, rctl);
1794                         e1000_update_mng_vlan(adapter);
1795                 }
1796         } else {
1797                 /* disable VLAN tag insert/strip */
1798                 ctrl = er32(CTRL);
1799                 ctrl &= ~E1000_CTRL_VME;
1800                 ew32(CTRL, ctrl);
1801
1802                 if (adapter->flags & FLAG_HAS_HW_VLAN_FILTER) {
1803                         if (adapter->mng_vlan_id !=
1804                             (u16)E1000_MNG_VLAN_NONE) {
1805                                 e1000_vlan_rx_kill_vid(netdev,
1806                                                        adapter->mng_vlan_id);
1807                                 adapter->mng_vlan_id = E1000_MNG_VLAN_NONE;
1808                         }
1809                 }
1810         }
1811
1812         if (!test_bit(__E1000_DOWN, &adapter->state))
1813                 e1000_irq_enable(adapter);
1814 }
1815
1816 static void e1000_restore_vlan(struct e1000_adapter *adapter)
1817 {
1818         u16 vid;
1819
1820         e1000_vlan_rx_register(adapter->netdev, adapter->vlgrp);
1821
1822         if (!adapter->vlgrp)
1823                 return;
1824
1825         for (vid = 0; vid < VLAN_GROUP_ARRAY_LEN; vid++) {
1826                 if (!vlan_group_get_device(adapter->vlgrp, vid))
1827                         continue;
1828                 e1000_vlan_rx_add_vid(adapter->netdev, vid);
1829         }
1830 }
1831
1832 static void e1000_init_manageability(struct e1000_adapter *adapter)
1833 {
1834         struct e1000_hw *hw = &adapter->hw;
1835         u32 manc, manc2h;
1836
1837         if (!(adapter->flags & FLAG_MNG_PT_ENABLED))
1838                 return;
1839
1840         manc = er32(MANC);
1841
1842         /*
1843          * enable receiving management packets to the host. this will probably
1844          * generate destination unreachable messages from the host OS, but
1845          * the packets will be handled on SMBUS
1846          */
1847         manc |= E1000_MANC_EN_MNG2HOST;
1848         manc2h = er32(MANC2H);
1849 #define E1000_MNG2HOST_PORT_623 (1 << 5)
1850 #define E1000_MNG2HOST_PORT_664 (1 << 6)
1851         manc2h |= E1000_MNG2HOST_PORT_623;
1852         manc2h |= E1000_MNG2HOST_PORT_664;
1853         ew32(MANC2H, manc2h);
1854         ew32(MANC, manc);
1855 }
1856
1857 /**
1858  * e1000_configure_tx - Configure 8254x Transmit Unit after Reset
1859  * @adapter: board private structure
1860  *
1861  * Configure the Tx unit of the MAC after a reset.
1862  **/
1863 static void e1000_configure_tx(struct e1000_adapter *adapter)
1864 {
1865         struct e1000_hw *hw = &adapter->hw;
1866         struct e1000_ring *tx_ring = adapter->tx_ring;
1867         u64 tdba;
1868         u32 tdlen, tctl, tipg, tarc;
1869         u32 ipgr1, ipgr2;
1870
1871         /* Setup the HW Tx Head and Tail descriptor pointers */
1872         tdba = tx_ring->dma;
1873         tdlen = tx_ring->count * sizeof(struct e1000_tx_desc);
1874         ew32(TDBAL, (tdba & DMA_32BIT_MASK));
1875         ew32(TDBAH, (tdba >> 32));
1876         ew32(TDLEN, tdlen);
1877         ew32(TDH, 0);
1878         ew32(TDT, 0);
1879         tx_ring->head = E1000_TDH;
1880         tx_ring->tail = E1000_TDT;
1881
1882         /* Set the default values for the Tx Inter Packet Gap timer */
1883         tipg = DEFAULT_82543_TIPG_IPGT_COPPER;          /*  8  */
1884         ipgr1 = DEFAULT_82543_TIPG_IPGR1;               /*  8  */
1885         ipgr2 = DEFAULT_82543_TIPG_IPGR2;               /*  6  */
1886
1887         if (adapter->flags & FLAG_TIPG_MEDIUM_FOR_80003ESLAN)
1888                 ipgr2 = DEFAULT_80003ES2LAN_TIPG_IPGR2; /*  7  */
1889
1890         tipg |= ipgr1 << E1000_TIPG_IPGR1_SHIFT;
1891         tipg |= ipgr2 << E1000_TIPG_IPGR2_SHIFT;
1892         ew32(TIPG, tipg);
1893
1894         /* Set the Tx Interrupt Delay register */
1895         ew32(TIDV, adapter->tx_int_delay);
1896         /* Tx irq moderation */
1897         ew32(TADV, adapter->tx_abs_int_delay);
1898
1899         /* Program the Transmit Control Register */
1900         tctl = er32(TCTL);
1901         tctl &= ~E1000_TCTL_CT;
1902         tctl |= E1000_TCTL_PSP | E1000_TCTL_RTLC |
1903                 (E1000_COLLISION_THRESHOLD << E1000_CT_SHIFT);
1904
1905         if (adapter->flags & FLAG_TARC_SPEED_MODE_BIT) {
1906                 tarc = er32(TARC(0));
1907                 /*
1908                  * set the speed mode bit, we'll clear it if we're not at
1909                  * gigabit link later
1910                  */
1911 #define SPEED_MODE_BIT (1 << 21)
1912                 tarc |= SPEED_MODE_BIT;
1913                 ew32(TARC(0), tarc);
1914         }
1915
1916         /* errata: program both queues to unweighted RR */
1917         if (adapter->flags & FLAG_TARC_SET_BIT_ZERO) {
1918                 tarc = er32(TARC(0));
1919                 tarc |= 1;
1920                 ew32(TARC(0), tarc);
1921                 tarc = er32(TARC(1));
1922                 tarc |= 1;
1923                 ew32(TARC(1), tarc);
1924         }
1925
1926         e1000e_config_collision_dist(hw);
1927
1928         /* Setup Transmit Descriptor Settings for eop descriptor */
1929         adapter->txd_cmd = E1000_TXD_CMD_EOP | E1000_TXD_CMD_IFCS;
1930
1931         /* only set IDE if we are delaying interrupts using the timers */
1932         if (adapter->tx_int_delay)
1933                 adapter->txd_cmd |= E1000_TXD_CMD_IDE;
1934
1935         /* enable Report Status bit */
1936         adapter->txd_cmd |= E1000_TXD_CMD_RS;
1937
1938         ew32(TCTL, tctl);
1939
1940         adapter->tx_queue_len = adapter->netdev->tx_queue_len;
1941 }
1942
1943 /**
1944  * e1000_setup_rctl - configure the receive control registers
1945  * @adapter: Board private structure
1946  **/
1947 #define PAGE_USE_COUNT(S) (((S) >> PAGE_SHIFT) + \
1948                            (((S) & (PAGE_SIZE - 1)) ? 1 : 0))
1949 static void e1000_setup_rctl(struct e1000_adapter *adapter)
1950 {
1951         struct e1000_hw *hw = &adapter->hw;
1952         u32 rctl, rfctl;
1953         u32 psrctl = 0;
1954         u32 pages = 0;
1955
1956         /* Program MC offset vector base */
1957         rctl = er32(RCTL);
1958         rctl &= ~(3 << E1000_RCTL_MO_SHIFT);
1959         rctl |= E1000_RCTL_EN | E1000_RCTL_BAM |
1960                 E1000_RCTL_LBM_NO | E1000_RCTL_RDMTS_HALF |
1961                 (adapter->hw.mac.mc_filter_type << E1000_RCTL_MO_SHIFT);
1962
1963         /* Do not Store bad packets */
1964         rctl &= ~E1000_RCTL_SBP;
1965
1966         /* Enable Long Packet receive */
1967         if (adapter->netdev->mtu <= ETH_DATA_LEN)
1968                 rctl &= ~E1000_RCTL_LPE;
1969         else
1970                 rctl |= E1000_RCTL_LPE;
1971
1972         /* Enable hardware CRC frame stripping */
1973         rctl |= E1000_RCTL_SECRC;
1974
1975         /* Setup buffer sizes */
1976         rctl &= ~E1000_RCTL_SZ_4096;
1977         rctl |= E1000_RCTL_BSEX;
1978         switch (adapter->rx_buffer_len) {
1979         case 256:
1980                 rctl |= E1000_RCTL_SZ_256;
1981                 rctl &= ~E1000_RCTL_BSEX;
1982                 break;
1983         case 512:
1984                 rctl |= E1000_RCTL_SZ_512;
1985                 rctl &= ~E1000_RCTL_BSEX;
1986                 break;
1987         case 1024:
1988                 rctl |= E1000_RCTL_SZ_1024;
1989                 rctl &= ~E1000_RCTL_BSEX;
1990                 break;
1991         case 2048:
1992         default:
1993                 rctl |= E1000_RCTL_SZ_2048;
1994                 rctl &= ~E1000_RCTL_BSEX;
1995                 break;
1996         case 4096:
1997                 rctl |= E1000_RCTL_SZ_4096;
1998                 break;
1999         case 8192:
2000                 rctl |= E1000_RCTL_SZ_8192;
2001                 break;
2002         case 16384:
2003                 rctl |= E1000_RCTL_SZ_16384;
2004                 break;
2005         }
2006
2007         /*
2008          * 82571 and greater support packet-split where the protocol
2009          * header is placed in skb->data and the packet data is
2010          * placed in pages hanging off of skb_shinfo(skb)->nr_frags.
2011          * In the case of a non-split, skb->data is linearly filled,
2012          * followed by the page buffers.  Therefore, skb->data is
2013          * sized to hold the largest protocol header.
2014          *
2015          * allocations using alloc_page take too long for regular MTU
2016          * so only enable packet split for jumbo frames
2017          *
2018          * Using pages when the page size is greater than 16k wastes
2019          * a lot of memory, since we allocate 3 pages at all times
2020          * per packet.
2021          */
2022         pages = PAGE_USE_COUNT(adapter->netdev->mtu);
2023         if (!(adapter->flags & FLAG_IS_ICH) && (pages <= 3) &&
2024             (PAGE_SIZE <= 16384) && (rctl & E1000_RCTL_LPE))
2025                 adapter->rx_ps_pages = pages;
2026         else
2027                 adapter->rx_ps_pages = 0;
2028
2029         if (adapter->rx_ps_pages) {
2030                 /* Configure extra packet-split registers */
2031                 rfctl = er32(RFCTL);
2032                 rfctl |= E1000_RFCTL_EXTEN;
2033                 /*
2034                  * disable packet split support for IPv6 extension headers,
2035                  * because some malformed IPv6 headers can hang the Rx
2036                  */
2037                 rfctl |= (E1000_RFCTL_IPV6_EX_DIS |
2038                           E1000_RFCTL_NEW_IPV6_EXT_DIS);
2039
2040                 ew32(RFCTL, rfctl);
2041
2042                 /* Enable Packet split descriptors */
2043                 rctl |= E1000_RCTL_DTYP_PS;
2044
2045                 psrctl |= adapter->rx_ps_bsize0 >>
2046                         E1000_PSRCTL_BSIZE0_SHIFT;
2047
2048                 switch (adapter->rx_ps_pages) {
2049                 case 3:
2050                         psrctl |= PAGE_SIZE <<
2051                                 E1000_PSRCTL_BSIZE3_SHIFT;
2052                 case 2:
2053                         psrctl |= PAGE_SIZE <<
2054                                 E1000_PSRCTL_BSIZE2_SHIFT;
2055                 case 1:
2056                         psrctl |= PAGE_SIZE >>
2057                                 E1000_PSRCTL_BSIZE1_SHIFT;
2058                         break;
2059                 }
2060
2061                 ew32(PSRCTL, psrctl);
2062         }
2063
2064         ew32(RCTL, rctl);
2065         /* just started the receive unit, no need to restart */
2066         adapter->flags &= ~FLAG_RX_RESTART_NOW;
2067 }
2068
2069 /**
2070  * e1000_configure_rx - Configure Receive Unit after Reset
2071  * @adapter: board private structure
2072  *
2073  * Configure the Rx unit of the MAC after a reset.
2074  **/
2075 static void e1000_configure_rx(struct e1000_adapter *adapter)
2076 {
2077         struct e1000_hw *hw = &adapter->hw;
2078         struct e1000_ring *rx_ring = adapter->rx_ring;
2079         u64 rdba;
2080         u32 rdlen, rctl, rxcsum, ctrl_ext;
2081
2082         if (adapter->rx_ps_pages) {
2083                 /* this is a 32 byte descriptor */
2084                 rdlen = rx_ring->count *
2085                         sizeof(union e1000_rx_desc_packet_split);
2086                 adapter->clean_rx = e1000_clean_rx_irq_ps;
2087                 adapter->alloc_rx_buf = e1000_alloc_rx_buffers_ps;
2088         } else if (adapter->netdev->mtu > ETH_FRAME_LEN + ETH_FCS_LEN) {
2089                 rdlen = rx_ring->count * sizeof(struct e1000_rx_desc);
2090                 adapter->clean_rx = e1000_clean_jumbo_rx_irq;
2091                 adapter->alloc_rx_buf = e1000_alloc_jumbo_rx_buffers;
2092         } else {
2093                 rdlen = rx_ring->count * sizeof(struct e1000_rx_desc);
2094                 adapter->clean_rx = e1000_clean_rx_irq;
2095                 adapter->alloc_rx_buf = e1000_alloc_rx_buffers;
2096         }
2097
2098         /* disable receives while setting up the descriptors */
2099         rctl = er32(RCTL);
2100         ew32(RCTL, rctl & ~E1000_RCTL_EN);
2101         e1e_flush();
2102         msleep(10);
2103
2104         /* set the Receive Delay Timer Register */
2105         ew32(RDTR, adapter->rx_int_delay);
2106
2107         /* irq moderation */
2108         ew32(RADV, adapter->rx_abs_int_delay);
2109         if (adapter->itr_setting != 0)
2110                 ew32(ITR, 1000000000 / (adapter->itr * 256));
2111
2112         ctrl_ext = er32(CTRL_EXT);
2113         /* Reset delay timers after every interrupt */
2114         ctrl_ext |= E1000_CTRL_EXT_INT_TIMER_CLR;
2115         /* Auto-Mask interrupts upon ICR access */
2116         ctrl_ext |= E1000_CTRL_EXT_IAME;
2117         ew32(IAM, 0xffffffff);
2118         ew32(CTRL_EXT, ctrl_ext);
2119         e1e_flush();
2120
2121         /*
2122          * Setup the HW Rx Head and Tail Descriptor Pointers and
2123          * the Base and Length of the Rx Descriptor Ring
2124          */
2125         rdba = rx_ring->dma;
2126         ew32(RDBAL, (rdba & DMA_32BIT_MASK));
2127         ew32(RDBAH, (rdba >> 32));
2128         ew32(RDLEN, rdlen);
2129         ew32(RDH, 0);
2130         ew32(RDT, 0);
2131         rx_ring->head = E1000_RDH;
2132         rx_ring->tail = E1000_RDT;
2133
2134         /* Enable Receive Checksum Offload for TCP and UDP */
2135         rxcsum = er32(RXCSUM);
2136         if (adapter->flags & FLAG_RX_CSUM_ENABLED) {
2137                 rxcsum |= E1000_RXCSUM_TUOFL;
2138
2139                 /*
2140                  * IPv4 payload checksum for UDP fragments must be
2141                  * used in conjunction with packet-split.
2142                  */
2143                 if (adapter->rx_ps_pages)
2144                         rxcsum |= E1000_RXCSUM_IPPCSE;
2145         } else {
2146                 rxcsum &= ~E1000_RXCSUM_TUOFL;
2147                 /* no need to clear IPPCSE as it defaults to 0 */
2148         }
2149         ew32(RXCSUM, rxcsum);
2150
2151         /*
2152          * Enable early receives on supported devices, only takes effect when
2153          * packet size is equal or larger than the specified value (in 8 byte
2154          * units), e.g. using jumbo frames when setting to E1000_ERT_2048
2155          */
2156         if ((adapter->flags & FLAG_HAS_ERT) &&
2157             (adapter->netdev->mtu > ETH_DATA_LEN)) {
2158                 u32 rxdctl = er32(RXDCTL(0));
2159                 ew32(RXDCTL(0), rxdctl | 0x3);
2160                 ew32(ERT, E1000_ERT_2048 | (1 << 13));
2161                 /*
2162                  * With jumbo frames and early-receive enabled, excessive
2163                  * C4->C2 latencies result in dropped transactions.
2164                  */
2165                 pm_qos_update_requirement(PM_QOS_CPU_DMA_LATENCY,
2166                                           e1000e_driver_name, 55);
2167         } else {
2168                 pm_qos_update_requirement(PM_QOS_CPU_DMA_LATENCY,
2169                                           e1000e_driver_name,
2170                                           PM_QOS_DEFAULT_VALUE);
2171         }
2172
2173         /* Enable Receives */
2174         ew32(RCTL, rctl);
2175 }
2176
2177 /**
2178  *  e1000_update_mc_addr_list - Update Multicast addresses
2179  *  @hw: pointer to the HW structure
2180  *  @mc_addr_list: array of multicast addresses to program
2181  *  @mc_addr_count: number of multicast addresses to program
2182  *  @rar_used_count: the first RAR register free to program
2183  *  @rar_count: total number of supported Receive Address Registers
2184  *
2185  *  Updates the Receive Address Registers and Multicast Table Array.
2186  *  The caller must have a packed mc_addr_list of multicast addresses.
2187  *  The parameter rar_count will usually be hw->mac.rar_entry_count
2188  *  unless there are workarounds that change this.  Currently no func pointer
2189  *  exists and all implementations are handled in the generic version of this
2190  *  function.
2191  **/
2192 static void e1000_update_mc_addr_list(struct e1000_hw *hw, u8 *mc_addr_list,
2193                                       u32 mc_addr_count, u32 rar_used_count,
2194                                       u32 rar_count)
2195 {
2196         hw->mac.ops.update_mc_addr_list(hw, mc_addr_list, mc_addr_count,
2197                                         rar_used_count, rar_count);
2198 }
2199
2200 /**
2201  * e1000_set_multi - Multicast and Promiscuous mode set
2202  * @netdev: network interface device structure
2203  *
2204  * The set_multi entry point is called whenever the multicast address
2205  * list or the network interface flags are updated.  This routine is
2206  * responsible for configuring the hardware for proper multicast,
2207  * promiscuous mode, and all-multi behavior.
2208  **/
2209 static void e1000_set_multi(struct net_device *netdev)
2210 {
2211         struct e1000_adapter *adapter = netdev_priv(netdev);
2212         struct e1000_hw *hw = &adapter->hw;
2213         struct e1000_mac_info *mac = &hw->mac;
2214         struct dev_mc_list *mc_ptr;
2215         u8  *mta_list;
2216         u32 rctl;
2217         int i;
2218
2219         /* Check for Promiscuous and All Multicast modes */
2220
2221         rctl = er32(RCTL);
2222
2223         if (netdev->flags & IFF_PROMISC) {
2224                 rctl |= (E1000_RCTL_UPE | E1000_RCTL_MPE);
2225                 rctl &= ~E1000_RCTL_VFE;
2226         } else {
2227                 if (netdev->flags & IFF_ALLMULTI) {
2228                         rctl |= E1000_RCTL_MPE;
2229                         rctl &= ~E1000_RCTL_UPE;
2230                 } else {
2231                         rctl &= ~(E1000_RCTL_UPE | E1000_RCTL_MPE);
2232                 }
2233                 if (adapter->flags & FLAG_HAS_HW_VLAN_FILTER)
2234                         rctl |= E1000_RCTL_VFE;
2235         }
2236
2237         ew32(RCTL, rctl);
2238
2239         if (netdev->mc_count) {
2240                 mta_list = kmalloc(netdev->mc_count * 6, GFP_ATOMIC);
2241                 if (!mta_list)
2242                         return;
2243
2244                 /* prepare a packed array of only addresses. */
2245                 mc_ptr = netdev->mc_list;
2246
2247                 for (i = 0; i < netdev->mc_count; i++) {
2248                         if (!mc_ptr)
2249                                 break;
2250                         memcpy(mta_list + (i*ETH_ALEN), mc_ptr->dmi_addr,
2251                                ETH_ALEN);
2252                         mc_ptr = mc_ptr->next;
2253                 }
2254
2255                 e1000_update_mc_addr_list(hw, mta_list, i, 1,
2256                                           mac->rar_entry_count);
2257                 kfree(mta_list);
2258         } else {
2259                 /*
2260                  * if we're called from probe, we might not have
2261                  * anything to do here, so clear out the list
2262                  */
2263                 e1000_update_mc_addr_list(hw, NULL, 0, 1, mac->rar_entry_count);
2264         }
2265 }
2266
2267 /**
2268  * e1000_configure - configure the hardware for Rx and Tx
2269  * @adapter: private board structure
2270  **/
2271 static void e1000_configure(struct e1000_adapter *adapter)
2272 {
2273         e1000_set_multi(adapter->netdev);
2274
2275         e1000_restore_vlan(adapter);
2276         e1000_init_manageability(adapter);
2277
2278         e1000_configure_tx(adapter);
2279         e1000_setup_rctl(adapter);
2280         e1000_configure_rx(adapter);
2281         adapter->alloc_rx_buf(adapter, e1000_desc_unused(adapter->rx_ring));
2282 }
2283
2284 /**
2285  * e1000e_power_up_phy - restore link in case the phy was powered down
2286  * @adapter: address of board private structure
2287  *
2288  * The phy may be powered down to save power and turn off link when the
2289  * driver is unloaded and wake on lan is not enabled (among others)
2290  * *** this routine MUST be followed by a call to e1000e_reset ***
2291  **/
2292 void e1000e_power_up_phy(struct e1000_adapter *adapter)
2293 {
2294         u16 mii_reg = 0;
2295
2296         /* Just clear the power down bit to wake the phy back up */
2297         if (adapter->hw.phy.media_type == e1000_media_type_copper) {
2298                 /*
2299                  * According to the manual, the phy will retain its
2300                  * settings across a power-down/up cycle
2301                  */
2302                 e1e_rphy(&adapter->hw, PHY_CONTROL, &mii_reg);
2303                 mii_reg &= ~MII_CR_POWER_DOWN;
2304                 e1e_wphy(&adapter->hw, PHY_CONTROL, mii_reg);
2305         }
2306
2307         adapter->hw.mac.ops.setup_link(&adapter->hw);
2308 }
2309
2310 /**
2311  * e1000_power_down_phy - Power down the PHY
2312  *
2313  * Power down the PHY so no link is implied when interface is down
2314  * The PHY cannot be powered down is management or WoL is active
2315  */
2316 static void e1000_power_down_phy(struct e1000_adapter *adapter)
2317 {
2318         struct e1000_hw *hw = &adapter->hw;
2319         u16 mii_reg;
2320
2321         /* WoL is enabled */
2322         if (adapter->wol)
2323                 return;
2324
2325         /* non-copper PHY? */
2326         if (adapter->hw.phy.media_type != e1000_media_type_copper)
2327                 return;
2328
2329         /* reset is blocked because of a SoL/IDER session */
2330         if (e1000e_check_mng_mode(hw) || e1000_check_reset_block(hw))
2331                 return;
2332
2333         /* manageability (AMT) is enabled */
2334         if (er32(MANC) & E1000_MANC_SMBUS_EN)
2335                 return;
2336
2337         /* power down the PHY */
2338         e1e_rphy(hw, PHY_CONTROL, &mii_reg);
2339         mii_reg |= MII_CR_POWER_DOWN;
2340         e1e_wphy(hw, PHY_CONTROL, mii_reg);
2341         mdelay(1);
2342 }
2343
2344 /**
2345  * e1000e_reset - bring the hardware into a known good state
2346  *
2347  * This function boots the hardware and enables some settings that
2348  * require a configuration cycle of the hardware - those cannot be
2349  * set/changed during runtime. After reset the device needs to be
2350  * properly configured for Rx, Tx etc.
2351  */
2352 void e1000e_reset(struct e1000_adapter *adapter)
2353 {
2354         struct e1000_mac_info *mac = &adapter->hw.mac;
2355         struct e1000_fc_info *fc = &adapter->hw.fc;
2356         struct e1000_hw *hw = &adapter->hw;
2357         u32 tx_space, min_tx_space, min_rx_space;
2358         u32 pba = adapter->pba;
2359         u16 hwm;
2360
2361         /* reset Packet Buffer Allocation to default */
2362         ew32(PBA, pba);
2363
2364         if (adapter->max_frame_size > ETH_FRAME_LEN + ETH_FCS_LEN) {
2365                 /*
2366                  * To maintain wire speed transmits, the Tx FIFO should be
2367                  * large enough to accommodate two full transmit packets,
2368                  * rounded up to the next 1KB and expressed in KB.  Likewise,
2369                  * the Rx FIFO should be large enough to accommodate at least
2370                  * one full receive packet and is similarly rounded up and
2371                  * expressed in KB.
2372                  */
2373                 pba = er32(PBA);
2374                 /* upper 16 bits has Tx packet buffer allocation size in KB */
2375                 tx_space = pba >> 16;
2376                 /* lower 16 bits has Rx packet buffer allocation size in KB */
2377                 pba &= 0xffff;
2378                 /*
2379                  * the Tx fifo also stores 16 bytes of information about the tx
2380                  * but don't include ethernet FCS because hardware appends it
2381                  */
2382                 min_tx_space = (adapter->max_frame_size +
2383                                 sizeof(struct e1000_tx_desc) -
2384                                 ETH_FCS_LEN) * 2;
2385                 min_tx_space = ALIGN(min_tx_space, 1024);
2386                 min_tx_space >>= 10;
2387                 /* software strips receive CRC, so leave room for it */
2388                 min_rx_space = adapter->max_frame_size;
2389                 min_rx_space = ALIGN(min_rx_space, 1024);
2390                 min_rx_space >>= 10;
2391
2392                 /*
2393                  * If current Tx allocation is less than the min Tx FIFO size,
2394                  * and the min Tx FIFO size is less than the current Rx FIFO
2395                  * allocation, take space away from current Rx allocation
2396                  */
2397                 if ((tx_space < min_tx_space) &&
2398                     ((min_tx_space - tx_space) < pba)) {
2399                         pba -= min_tx_space - tx_space;
2400
2401                         /*
2402                          * if short on Rx space, Rx wins and must trump tx
2403                          * adjustment or use Early Receive if available
2404                          */
2405                         if ((pba < min_rx_space) &&
2406                             (!(adapter->flags & FLAG_HAS_ERT)))
2407                                 /* ERT enabled in e1000_configure_rx */
2408                                 pba = min_rx_space;
2409                 }
2410
2411                 ew32(PBA, pba);
2412         }
2413
2414
2415         /*
2416          * flow control settings
2417          *
2418          * The high water mark must be low enough to fit one full frame
2419          * (or the size used for early receive) above it in the Rx FIFO.
2420          * Set it to the lower of:
2421          * - 90% of the Rx FIFO size, and
2422          * - the full Rx FIFO size minus the early receive size (for parts
2423          *   with ERT support assuming ERT set to E1000_ERT_2048), or
2424          * - the full Rx FIFO size minus one full frame
2425          */
2426         if (adapter->flags & FLAG_HAS_ERT)
2427                 hwm = min(((pba << 10) * 9 / 10),
2428                           ((pba << 10) - (E1000_ERT_2048 << 3)));
2429         else
2430                 hwm = min(((pba << 10) * 9 / 10),
2431                           ((pba << 10) - adapter->max_frame_size));
2432
2433         fc->high_water = hwm & 0xFFF8; /* 8-byte granularity */
2434         fc->low_water = fc->high_water - 8;
2435
2436         if (adapter->flags & FLAG_DISABLE_FC_PAUSE_TIME)
2437                 fc->pause_time = 0xFFFF;
2438         else
2439                 fc->pause_time = E1000_FC_PAUSE_TIME;
2440         fc->send_xon = 1;
2441         fc->type = fc->original_type;
2442
2443         /* Allow time for pending master requests to run */
2444         mac->ops.reset_hw(hw);
2445
2446         /*
2447          * For parts with AMT enabled, let the firmware know
2448          * that the network interface is in control
2449          */
2450         if (adapter->flags & FLAG_HAS_AMT)
2451                 e1000_get_hw_control(adapter);
2452
2453         ew32(WUC, 0);
2454
2455         if (mac->ops.init_hw(hw))
2456                 e_err("Hardware Error\n");
2457
2458         e1000_update_mng_vlan(adapter);
2459
2460         /* Enable h/w to recognize an 802.1Q VLAN Ethernet packet */
2461         ew32(VET, ETH_P_8021Q);
2462
2463         e1000e_reset_adaptive(hw);
2464         e1000_get_phy_info(hw);
2465
2466         if (!(adapter->flags & FLAG_SMART_POWER_DOWN)) {
2467                 u16 phy_data = 0;
2468                 /*
2469                  * speed up time to link by disabling smart power down, ignore
2470                  * the return value of this function because there is nothing
2471                  * different we would do if it failed
2472                  */
2473                 e1e_rphy(hw, IGP02E1000_PHY_POWER_MGMT, &phy_data);
2474                 phy_data &= ~IGP02E1000_PM_SPD;
2475                 e1e_wphy(hw, IGP02E1000_PHY_POWER_MGMT, phy_data);
2476         }
2477 }
2478
2479 int e1000e_up(struct e1000_adapter *adapter)
2480 {
2481         struct e1000_hw *hw = &adapter->hw;
2482
2483         /* hardware has been reset, we need to reload some things */
2484         e1000_configure(adapter);
2485
2486         clear_bit(__E1000_DOWN, &adapter->state);
2487
2488         napi_enable(&adapter->napi);
2489         e1000_irq_enable(adapter);
2490
2491         /* fire a link change interrupt to start the watchdog */
2492         ew32(ICS, E1000_ICS_LSC);
2493         return 0;
2494 }
2495
2496 void e1000e_down(struct e1000_adapter *adapter)
2497 {
2498         struct net_device *netdev = adapter->netdev;
2499         struct e1000_hw *hw = &adapter->hw;
2500         u32 tctl, rctl;
2501
2502         /*
2503          * signal that we're down so the interrupt handler does not
2504          * reschedule our watchdog timer
2505          */
2506         set_bit(__E1000_DOWN, &adapter->state);
2507
2508         /* disable receives in the hardware */
2509         rctl = er32(RCTL);
2510         ew32(RCTL, rctl & ~E1000_RCTL_EN);
2511         /* flush and sleep below */
2512
2513         netif_tx_stop_all_queues(netdev);
2514
2515         /* disable transmits in the hardware */
2516         tctl = er32(TCTL);
2517         tctl &= ~E1000_TCTL_EN;
2518         ew32(TCTL, tctl);
2519         /* flush both disables and wait for them to finish */
2520         e1e_flush();
2521         msleep(10);
2522
2523         napi_disable(&adapter->napi);
2524         e1000_irq_disable(adapter);
2525
2526         del_timer_sync(&adapter->watchdog_timer);
2527         del_timer_sync(&adapter->phy_info_timer);
2528
2529         netdev->tx_queue_len = adapter->tx_queue_len;
2530         netif_carrier_off(netdev);
2531         adapter->link_speed = 0;
2532         adapter->link_duplex = 0;
2533
2534         if (!pci_channel_offline(adapter->pdev))
2535                 e1000e_reset(adapter);
2536         e1000_clean_tx_ring(adapter);
2537         e1000_clean_rx_ring(adapter);
2538
2539         /*
2540          * TODO: for power management, we could drop the link and
2541          * pci_disable_device here.
2542          */
2543 }
2544
2545 void e1000e_reinit_locked(struct e1000_adapter *adapter)
2546 {
2547         might_sleep();
2548         while (test_and_set_bit(__E1000_RESETTING, &adapter->state))
2549                 msleep(1);
2550         e1000e_down(adapter);
2551         e1000e_up(adapter);
2552         clear_bit(__E1000_RESETTING, &adapter->state);
2553 }
2554
2555 /**
2556  * e1000_sw_init - Initialize general software structures (struct e1000_adapter)
2557  * @adapter: board private structure to initialize
2558  *
2559  * e1000_sw_init initializes the Adapter private data structure.
2560  * Fields are initialized based on PCI device information and
2561  * OS network device settings (MTU size).
2562  **/
2563 static int __devinit e1000_sw_init(struct e1000_adapter *adapter)
2564 {
2565         struct net_device *netdev = adapter->netdev;
2566
2567         adapter->rx_buffer_len = ETH_FRAME_LEN + VLAN_HLEN + ETH_FCS_LEN;
2568         adapter->rx_ps_bsize0 = 128;
2569         adapter->max_frame_size = netdev->mtu + ETH_HLEN + ETH_FCS_LEN;
2570         adapter->min_frame_size = ETH_ZLEN + ETH_FCS_LEN;
2571
2572         adapter->tx_ring = kzalloc(sizeof(struct e1000_ring), GFP_KERNEL);
2573         if (!adapter->tx_ring)
2574                 goto err;
2575
2576         adapter->rx_ring = kzalloc(sizeof(struct e1000_ring), GFP_KERNEL);
2577         if (!adapter->rx_ring)
2578                 goto err;
2579
2580         spin_lock_init(&adapter->tx_queue_lock);
2581
2582         /* Explicitly disable IRQ since the NIC can be in any state. */
2583         e1000_irq_disable(adapter);
2584
2585         spin_lock_init(&adapter->stats_lock);
2586
2587         set_bit(__E1000_DOWN, &adapter->state);
2588         return 0;
2589
2590 err:
2591         e_err("Unable to allocate memory for queues\n");
2592         kfree(adapter->rx_ring);
2593         kfree(adapter->tx_ring);
2594         return -ENOMEM;
2595 }
2596
2597 /**
2598  * e1000_open - Called when a network interface is made active
2599  * @netdev: network interface device structure
2600  *
2601  * Returns 0 on success, negative value on failure
2602  *
2603  * The open entry point is called when a network interface is made
2604  * active by the system (IFF_UP).  At this point all resources needed
2605  * for transmit and receive operations are allocated, the interrupt
2606  * handler is registered with the OS, the watchdog timer is started,
2607  * and the stack is notified that the interface is ready.
2608  **/
2609 static int e1000_open(struct net_device *netdev)
2610 {
2611         struct e1000_adapter *adapter = netdev_priv(netdev);
2612         struct e1000_hw *hw = &adapter->hw;
2613         int err;
2614
2615         /* disallow open during test */
2616         if (test_bit(__E1000_TESTING, &adapter->state))
2617                 return -EBUSY;
2618
2619         /* allocate transmit descriptors */
2620         err = e1000e_setup_tx_resources(adapter);
2621         if (err)
2622                 goto err_setup_tx;
2623
2624         /* allocate receive descriptors */
2625         err = e1000e_setup_rx_resources(adapter);
2626         if (err)
2627                 goto err_setup_rx;
2628
2629         e1000e_power_up_phy(adapter);
2630
2631         adapter->mng_vlan_id = E1000_MNG_VLAN_NONE;
2632         if ((adapter->hw.mng_cookie.status &
2633              E1000_MNG_DHCP_COOKIE_STATUS_VLAN))
2634                 e1000_update_mng_vlan(adapter);
2635
2636         /*
2637          * If AMT is enabled, let the firmware know that the network
2638          * interface is now open
2639          */
2640         if (adapter->flags & FLAG_HAS_AMT)
2641                 e1000_get_hw_control(adapter);
2642
2643         /*
2644          * before we allocate an interrupt, we must be ready to handle it.
2645          * Setting DEBUG_SHIRQ in the kernel makes it fire an interrupt
2646          * as soon as we call pci_request_irq, so we have to setup our
2647          * clean_rx handler before we do so.
2648          */
2649         e1000_configure(adapter);
2650
2651         err = e1000_request_irq(adapter);
2652         if (err)
2653                 goto err_req_irq;
2654
2655         /* From here on the code is the same as e1000e_up() */
2656         clear_bit(__E1000_DOWN, &adapter->state);
2657
2658         napi_enable(&adapter->napi);
2659
2660         e1000_irq_enable(adapter);
2661
2662         netif_tx_start_all_queues(netdev);
2663
2664         /* fire a link status change interrupt to start the watchdog */
2665         ew32(ICS, E1000_ICS_LSC);
2666
2667         return 0;
2668
2669 err_req_irq:
2670         e1000_release_hw_control(adapter);
2671         e1000_power_down_phy(adapter);
2672         e1000e_free_rx_resources(adapter);
2673 err_setup_rx:
2674         e1000e_free_tx_resources(adapter);
2675 err_setup_tx:
2676         e1000e_reset(adapter);
2677
2678         return err;
2679 }
2680
2681 /**
2682  * e1000_close - Disables a network interface
2683  * @netdev: network interface device structure
2684  *
2685  * Returns 0, this is not allowed to fail
2686  *
2687  * The close entry point is called when an interface is de-activated
2688  * by the OS.  The hardware is still under the drivers control, but
2689  * needs to be disabled.  A global MAC reset is issued to stop the
2690  * hardware, and all transmit and receive resources are freed.
2691  **/
2692 static int e1000_close(struct net_device *netdev)
2693 {
2694         struct e1000_adapter *adapter = netdev_priv(netdev);
2695
2696         WARN_ON(test_bit(__E1000_RESETTING, &adapter->state));
2697         e1000e_down(adapter);
2698         e1000_power_down_phy(adapter);
2699         e1000_free_irq(adapter);
2700
2701         e1000e_free_tx_resources(adapter);
2702         e1000e_free_rx_resources(adapter);
2703
2704         /*
2705          * kill manageability vlan ID if supported, but not if a vlan with
2706          * the same ID is registered on the host OS (let 8021q kill it)
2707          */
2708         if ((adapter->hw.mng_cookie.status &
2709                           E1000_MNG_DHCP_COOKIE_STATUS_VLAN) &&
2710              !(adapter->vlgrp &&
2711                vlan_group_get_device(adapter->vlgrp, adapter->mng_vlan_id)))
2712                 e1000_vlan_rx_kill_vid(netdev, adapter->mng_vlan_id);
2713
2714         /*
2715          * If AMT is enabled, let the firmware know that the network
2716          * interface is now closed
2717          */
2718         if (adapter->flags & FLAG_HAS_AMT)
2719                 e1000_release_hw_control(adapter);
2720
2721         return 0;
2722 }
2723 /**
2724  * e1000_set_mac - Change the Ethernet Address of the NIC
2725  * @netdev: network interface device structure
2726  * @p: pointer to an address structure
2727  *
2728  * Returns 0 on success, negative on failure
2729  **/
2730 static int e1000_set_mac(struct net_device *netdev, void *p)
2731 {
2732         struct e1000_adapter *adapter = netdev_priv(netdev);
2733         struct sockaddr *addr = p;
2734
2735         if (!is_valid_ether_addr(addr->sa_data))
2736                 return -EADDRNOTAVAIL;
2737
2738         memcpy(netdev->dev_addr, addr->sa_data, netdev->addr_len);
2739         memcpy(adapter->hw.mac.addr, addr->sa_data, netdev->addr_len);
2740
2741         e1000e_rar_set(&adapter->hw, adapter->hw.mac.addr, 0);
2742
2743         if (adapter->flags & FLAG_RESET_OVERWRITES_LAA) {
2744                 /* activate the work around */
2745                 e1000e_set_laa_state_82571(&adapter->hw, 1);
2746
2747                 /*
2748                  * Hold a copy of the LAA in RAR[14] This is done so that
2749                  * between the time RAR[0] gets clobbered  and the time it
2750                  * gets fixed (in e1000_watchdog), the actual LAA is in one
2751                  * of the RARs and no incoming packets directed to this port
2752                  * are dropped. Eventually the LAA will be in RAR[0] and
2753                  * RAR[14]
2754                  */
2755                 e1000e_rar_set(&adapter->hw,
2756                               adapter->hw.mac.addr,
2757                               adapter->hw.mac.rar_entry_count - 1);
2758         }
2759
2760         return 0;
2761 }
2762
2763 /*
2764  * Need to wait a few seconds after link up to get diagnostic information from
2765  * the phy
2766  */
2767 static void e1000_update_phy_info(unsigned long data)
2768 {
2769         struct e1000_adapter *adapter = (struct e1000_adapter *) data;
2770         e1000_get_phy_info(&adapter->hw);
2771 }
2772
2773 /**
2774  * e1000e_update_stats - Update the board statistics counters
2775  * @adapter: board private structure
2776  **/
2777 void e1000e_update_stats(struct e1000_adapter *adapter)
2778 {
2779         struct e1000_hw *hw = &adapter->hw;
2780         struct pci_dev *pdev = adapter->pdev;
2781         unsigned long irq_flags;
2782         u16 phy_tmp;
2783
2784 #define PHY_IDLE_ERROR_COUNT_MASK 0x00FF
2785
2786         /*
2787          * Prevent stats update while adapter is being reset, or if the pci
2788          * connection is down.
2789          */
2790         if (adapter->link_speed == 0)
2791                 return;
2792         if (pci_channel_offline(pdev))
2793                 return;
2794
2795         spin_lock_irqsave(&adapter->stats_lock, irq_flags);
2796
2797         /*
2798          * these counters are modified from e1000_adjust_tbi_stats,
2799          * called from the interrupt context, so they must only
2800          * be written while holding adapter->stats_lock
2801          */
2802
2803         adapter->stats.crcerrs += er32(CRCERRS);
2804         adapter->stats.gprc += er32(GPRC);
2805         adapter->stats.gorc += er32(GORCL);
2806         er32(GORCH); /* Clear gorc */
2807         adapter->stats.bprc += er32(BPRC);
2808         adapter->stats.mprc += er32(MPRC);
2809         adapter->stats.roc += er32(ROC);
2810
2811         adapter->stats.mpc += er32(MPC);
2812         adapter->stats.scc += er32(SCC);
2813         adapter->stats.ecol += er32(ECOL);
2814         adapter->stats.mcc += er32(MCC);
2815         adapter->stats.latecol += er32(LATECOL);
2816         adapter->stats.dc += er32(DC);
2817         adapter->stats.xonrxc += er32(XONRXC);
2818         adapter->stats.xontxc += er32(XONTXC);
2819         adapter->stats.xoffrxc += er32(XOFFRXC);
2820         adapter->stats.xofftxc += er32(XOFFTXC);
2821         adapter->stats.gptc += er32(GPTC);
2822         adapter->stats.gotc += er32(GOTCL);
2823         er32(GOTCH); /* Clear gotc */
2824         adapter->stats.rnbc += er32(RNBC);
2825         adapter->stats.ruc += er32(RUC);
2826
2827         adapter->stats.mptc += er32(MPTC);
2828         adapter->stats.bptc += er32(BPTC);
2829
2830         /* used for adaptive IFS */
2831
2832         hw->mac.tx_packet_delta = er32(TPT);
2833         adapter->stats.tpt += hw->mac.tx_packet_delta;
2834         hw->mac.collision_delta = er32(COLC);
2835         adapter->stats.colc += hw->mac.collision_delta;
2836
2837         adapter->stats.algnerrc += er32(ALGNERRC);
2838         adapter->stats.rxerrc += er32(RXERRC);
2839         adapter->stats.tncrs += er32(TNCRS);
2840         adapter->stats.cexterr += er32(CEXTERR);
2841         adapter->stats.tsctc += er32(TSCTC);
2842         adapter->stats.tsctfc += er32(TSCTFC);
2843
2844         /* Fill out the OS statistics structure */
2845         adapter->net_stats.multicast = adapter->stats.mprc;
2846         adapter->net_stats.collisions = adapter->stats.colc;
2847
2848         /* Rx Errors */
2849
2850         /*
2851          * RLEC on some newer hardware can be incorrect so build
2852          * our own version based on RUC and ROC
2853          */
2854         adapter->net_stats.rx_errors = adapter->stats.rxerrc +
2855                 adapter->stats.crcerrs + adapter->stats.algnerrc +
2856                 adapter->stats.ruc + adapter->stats.roc +
2857                 adapter->stats.cexterr;
2858         adapter->net_stats.rx_length_errors = adapter->stats.ruc +
2859                                               adapter->stats.roc;
2860         adapter->net_stats.rx_crc_errors = adapter->stats.crcerrs;
2861         adapter->net_stats.rx_frame_errors = adapter->stats.algnerrc;
2862         adapter->net_stats.rx_missed_errors = adapter->stats.mpc;
2863
2864         /* Tx Errors */
2865         adapter->net_stats.tx_errors = adapter->stats.ecol +
2866                                        adapter->stats.latecol;
2867         adapter->net_stats.tx_aborted_errors = adapter->stats.ecol;
2868         adapter->net_stats.tx_window_errors = adapter->stats.latecol;
2869         adapter->net_stats.tx_carrier_errors = adapter->stats.tncrs;
2870
2871         /* Tx Dropped needs to be maintained elsewhere */
2872
2873         /* Phy Stats */
2874         if (hw->phy.media_type == e1000_media_type_copper) {
2875                 if ((adapter->link_speed == SPEED_1000) &&
2876                    (!e1e_rphy(hw, PHY_1000T_STATUS, &phy_tmp))) {
2877                         phy_tmp &= PHY_IDLE_ERROR_COUNT_MASK;
2878                         adapter->phy_stats.idle_errors += phy_tmp;
2879                 }
2880         }
2881
2882         /* Management Stats */
2883         adapter->stats.mgptc += er32(MGTPTC);
2884         adapter->stats.mgprc += er32(MGTPRC);
2885         adapter->stats.mgpdc += er32(MGTPDC);
2886
2887         spin_unlock_irqrestore(&adapter->stats_lock, irq_flags);
2888 }
2889
2890 /**
2891  * e1000_phy_read_status - Update the PHY register status snapshot
2892  * @adapter: board private structure
2893  **/
2894 static void e1000_phy_read_status(struct e1000_adapter *adapter)
2895 {
2896         struct e1000_hw *hw = &adapter->hw;
2897         struct e1000_phy_regs *phy = &adapter->phy_regs;
2898         int ret_val;
2899         unsigned long irq_flags;
2900
2901
2902         spin_lock_irqsave(&adapter->stats_lock, irq_flags);
2903
2904         if ((er32(STATUS) & E1000_STATUS_LU) &&
2905             (adapter->hw.phy.media_type == e1000_media_type_copper)) {
2906                 ret_val  = e1e_rphy(hw, PHY_CONTROL, &phy->bmcr);
2907                 ret_val |= e1e_rphy(hw, PHY_STATUS, &phy->bmsr);
2908                 ret_val |= e1e_rphy(hw, PHY_AUTONEG_ADV, &phy->advertise);
2909                 ret_val |= e1e_rphy(hw, PHY_LP_ABILITY, &phy->lpa);
2910                 ret_val |= e1e_rphy(hw, PHY_AUTONEG_EXP, &phy->expansion);
2911                 ret_val |= e1e_rphy(hw, PHY_1000T_CTRL, &phy->ctrl1000);
2912                 ret_val |= e1e_rphy(hw, PHY_1000T_STATUS, &phy->stat1000);
2913                 ret_val |= e1e_rphy(hw, PHY_EXT_STATUS, &phy->estatus);
2914                 if (ret_val)
2915                         e_warn("Error reading PHY register\n");
2916         } else {
2917                 /*
2918                  * Do not read PHY registers if link is not up
2919                  * Set values to typical power-on defaults
2920                  */
2921                 phy->bmcr = (BMCR_SPEED1000 | BMCR_ANENABLE | BMCR_FULLDPLX);
2922                 phy->bmsr = (BMSR_100FULL | BMSR_100HALF | BMSR_10FULL |
2923                              BMSR_10HALF | BMSR_ESTATEN | BMSR_ANEGCAPABLE |
2924                              BMSR_ERCAP);
2925                 phy->advertise = (ADVERTISE_PAUSE_ASYM | ADVERTISE_PAUSE_CAP |
2926                                   ADVERTISE_ALL | ADVERTISE_CSMA);
2927                 phy->lpa = 0;
2928                 phy->expansion = EXPANSION_ENABLENPAGE;
2929                 phy->ctrl1000 = ADVERTISE_1000FULL;
2930                 phy->stat1000 = 0;
2931                 phy->estatus = (ESTATUS_1000_TFULL | ESTATUS_1000_THALF);
2932         }
2933
2934         spin_unlock_irqrestore(&adapter->stats_lock, irq_flags);
2935 }
2936
2937 static void e1000_print_link_info(struct e1000_adapter *adapter)
2938 {
2939         struct e1000_hw *hw = &adapter->hw;
2940         u32 ctrl = er32(CTRL);
2941
2942         e_info("Link is Up %d Mbps %s, Flow Control: %s\n",
2943                adapter->link_speed,
2944                (adapter->link_duplex == FULL_DUPLEX) ?
2945                                 "Full Duplex" : "Half Duplex",
2946                ((ctrl & E1000_CTRL_TFCE) && (ctrl & E1000_CTRL_RFCE)) ?
2947                                 "RX/TX" :
2948                ((ctrl & E1000_CTRL_RFCE) ? "RX" :
2949                ((ctrl & E1000_CTRL_TFCE) ? "TX" : "None" )));
2950 }
2951
2952 static bool e1000_has_link(struct e1000_adapter *adapter)
2953 {
2954         struct e1000_hw *hw = &adapter->hw;
2955         bool link_active = 0;
2956         s32 ret_val = 0;
2957
2958         /*
2959          * get_link_status is set on LSC (link status) interrupt or
2960          * Rx sequence error interrupt.  get_link_status will stay
2961          * false until the check_for_link establishes link
2962          * for copper adapters ONLY
2963          */
2964         switch (hw->phy.media_type) {
2965         case e1000_media_type_copper:
2966                 if (hw->mac.get_link_status) {
2967                         ret_val = hw->mac.ops.check_for_link(hw);
2968                         link_active = !hw->mac.get_link_status;
2969                 } else {
2970                         link_active = 1;
2971                 }
2972                 break;
2973         case e1000_media_type_fiber:
2974                 ret_val = hw->mac.ops.check_for_link(hw);
2975                 link_active = !!(er32(STATUS) & E1000_STATUS_LU);
2976                 break;
2977         case e1000_media_type_internal_serdes:
2978                 ret_val = hw->mac.ops.check_for_link(hw);
2979                 link_active = adapter->hw.mac.serdes_has_link;
2980                 break;
2981         default:
2982         case e1000_media_type_unknown:
2983                 break;
2984         }
2985
2986         if ((ret_val == E1000_ERR_PHY) && (hw->phy.type == e1000_phy_igp_3) &&
2987             (er32(CTRL) & E1000_PHY_CTRL_GBE_DISABLE)) {
2988                 /* See e1000_kmrn_lock_loss_workaround_ich8lan() */
2989                 e_info("Gigabit has been disabled, downgrading speed\n");
2990         }
2991
2992         return link_active;
2993 }
2994
2995 static void e1000e_enable_receives(struct e1000_adapter *adapter)
2996 {
2997         /* make sure the receive unit is started */
2998         if ((adapter->flags & FLAG_RX_NEEDS_RESTART) &&
2999             (adapter->flags & FLAG_RX_RESTART_NOW)) {
3000                 struct e1000_hw *hw = &adapter->hw;
3001                 u32 rctl = er32(RCTL);
3002                 ew32(RCTL, rctl | E1000_RCTL_EN);
3003                 adapter->flags &= ~FLAG_RX_RESTART_NOW;
3004         }
3005 }
3006
3007 /**
3008  * e1000_watchdog - Timer Call-back
3009  * @data: pointer to adapter cast into an unsigned long
3010  **/
3011 static void e1000_watchdog(unsigned long data)
3012 {
3013         struct e1000_adapter *adapter = (struct e1000_adapter *) data;
3014
3015         /* Do the rest outside of interrupt context */
3016         schedule_work(&adapter->watchdog_task);
3017
3018         /* TODO: make this use queue_delayed_work() */
3019 }
3020
3021 static void e1000_watchdog_task(struct work_struct *work)
3022 {
3023         struct e1000_adapter *adapter = container_of(work,
3024                                         struct e1000_adapter, watchdog_task);
3025         struct net_device *netdev = adapter->netdev;
3026         struct e1000_mac_info *mac = &adapter->hw.mac;
3027         struct e1000_ring *tx_ring = adapter->tx_ring;
3028         struct e1000_hw *hw = &adapter->hw;
3029         u32 link, tctl;
3030         int tx_pending = 0;
3031
3032         link = e1000_has_link(adapter);
3033         if ((netif_carrier_ok(netdev)) && link) {
3034                 e1000e_enable_receives(adapter);
3035                 goto link_up;
3036         }
3037
3038         if ((e1000e_enable_tx_pkt_filtering(hw)) &&
3039             (adapter->mng_vlan_id != adapter->hw.mng_cookie.vlan_id))
3040                 e1000_update_mng_vlan(adapter);
3041
3042         if (link) {
3043                 if (!netif_carrier_ok(netdev)) {
3044                         bool txb2b = 1;
3045                         /* update snapshot of PHY registers on LSC */
3046                         e1000_phy_read_status(adapter);
3047                         mac->ops.get_link_up_info(&adapter->hw,
3048                                                    &adapter->link_speed,
3049                                                    &adapter->link_duplex);
3050                         e1000_print_link_info(adapter);
3051                         /*
3052                          * tweak tx_queue_len according to speed/duplex
3053                          * and adjust the timeout factor
3054                          */
3055                         netdev->tx_queue_len = adapter->tx_queue_len;
3056                         adapter->tx_timeout_factor = 1;
3057                         switch (adapter->link_speed) {
3058                         case SPEED_10:
3059                                 txb2b = 0;
3060                                 netdev->tx_queue_len = 10;
3061                                 adapter->tx_timeout_factor = 16;
3062                                 break;
3063                         case SPEED_100:
3064                                 txb2b = 0;
3065                                 netdev->tx_queue_len = 100;
3066                                 /* maybe add some timeout factor ? */
3067                                 break;
3068                         }
3069
3070                         /*
3071                          * workaround: re-program speed mode bit after
3072                          * link-up event
3073                          */
3074                         if ((adapter->flags & FLAG_TARC_SPEED_MODE_BIT) &&
3075                             !txb2b) {
3076                                 u32 tarc0;
3077                                 tarc0 = er32(TARC(0));
3078                                 tarc0 &= ~SPEED_MODE_BIT;
3079                                 ew32(TARC(0), tarc0);
3080                         }
3081
3082                         /*
3083                          * disable TSO for pcie and 10/100 speeds, to avoid
3084                          * some hardware issues
3085                          */
3086                         if (!(adapter->flags & FLAG_TSO_FORCE)) {
3087                                 switch (adapter->link_speed) {
3088                                 case SPEED_10:
3089                                 case SPEED_100:
3090                                         e_info("10/100 speed: disabling TSO\n");
3091                                         netdev->features &= ~NETIF_F_TSO;
3092                                         netdev->features &= ~NETIF_F_TSO6;
3093                                         break;
3094                                 case SPEED_1000:
3095                                         netdev->features |= NETIF_F_TSO;
3096                                         netdev->features |= NETIF_F_TSO6;
3097                                         break;
3098                                 default:
3099                                         /* oops */
3100                                         break;
3101                                 }
3102                         }
3103
3104                         /*
3105                          * enable transmits in the hardware, need to do this
3106                          * after setting TARC(0)
3107                          */
3108                         tctl = er32(TCTL);
3109                         tctl |= E1000_TCTL_EN;
3110                         ew32(TCTL, tctl);
3111
3112                         netif_carrier_on(netdev);
3113                         netif_tx_wake_all_queues(netdev);
3114
3115                         if (!test_bit(__E1000_DOWN, &adapter->state))
3116                                 mod_timer(&adapter->phy_info_timer,
3117                                           round_jiffies(jiffies + 2 * HZ));
3118                 }
3119         } else {
3120                 if (netif_carrier_ok(netdev)) {
3121                         adapter->link_speed = 0;
3122                         adapter->link_duplex = 0;
3123                         e_info("Link is Down\n");
3124                         netif_carrier_off(netdev);
3125                         netif_tx_stop_all_queues(netdev);
3126                         if (!test_bit(__E1000_DOWN, &adapter->state))
3127                                 mod_timer(&adapter->phy_info_timer,
3128                                           round_jiffies(jiffies + 2 * HZ));
3129
3130                         if (adapter->flags & FLAG_RX_NEEDS_RESTART)
3131                                 schedule_work(&adapter->reset_task);
3132                 }
3133         }
3134
3135 link_up:
3136         e1000e_update_stats(adapter);
3137
3138         mac->tx_packet_delta = adapter->stats.tpt - adapter->tpt_old;
3139         adapter->tpt_old = adapter->stats.tpt;
3140         mac->collision_delta = adapter->stats.colc - adapter->colc_old;
3141         adapter->colc_old = adapter->stats.colc;
3142
3143         adapter->gorc = adapter->stats.gorc - adapter->gorc_old;
3144         adapter->gorc_old = adapter->stats.gorc;
3145         adapter->gotc = adapter->stats.gotc - adapter->gotc_old;
3146         adapter->gotc_old = adapter->stats.gotc;
3147
3148         e1000e_update_adaptive(&adapter->hw);
3149
3150         if (!netif_carrier_ok(netdev)) {
3151                 tx_pending = (e1000_desc_unused(tx_ring) + 1 <
3152                                tx_ring->count);
3153                 if (tx_pending) {
3154                         /*
3155                          * We've lost link, so the controller stops DMA,
3156                          * but we've got queued Tx work that's never going
3157                          * to get done, so reset controller to flush Tx.
3158                          * (Do the reset outside of interrupt context).
3159                          */
3160                         adapter->tx_timeout_count++;
3161                         schedule_work(&adapter->reset_task);
3162                 }
3163         }
3164
3165         /* Cause software interrupt to ensure Rx ring is cleaned */
3166         ew32(ICS, E1000_ICS_RXDMT0);
3167
3168         /* Force detection of hung controller every watchdog period */
3169         adapter->detect_tx_hung = 1;
3170
3171         /*
3172          * With 82571 controllers, LAA may be overwritten due to controller
3173          * reset from the other port. Set the appropriate LAA in RAR[0]
3174          */
3175         if (e1000e_get_laa_state_82571(hw))
3176                 e1000e_rar_set(hw, adapter->hw.mac.addr, 0);
3177
3178         /* Reset the timer */
3179         if (!test_bit(__E1000_DOWN, &adapter->state))
3180                 mod_timer(&adapter->watchdog_timer,
3181                           round_jiffies(jiffies + 2 * HZ));
3182 }
3183
3184 #define E1000_TX_FLAGS_CSUM             0x00000001
3185 #define E1000_TX_FLAGS_VLAN             0x00000002
3186 #define E1000_TX_FLAGS_TSO              0x00000004
3187 #define E1000_TX_FLAGS_IPV4             0x00000008
3188 #define E1000_TX_FLAGS_VLAN_MASK        0xffff0000
3189 #define E1000_TX_FLAGS_VLAN_SHIFT       16
3190
3191 static int e1000_tso(struct e1000_adapter *adapter,
3192                      struct sk_buff *skb)
3193 {
3194         struct e1000_ring *tx_ring = adapter->tx_ring;
3195         struct e1000_context_desc *context_desc;
3196         struct e1000_buffer *buffer_info;
3197         unsigned int i;
3198         u32 cmd_length = 0;
3199         u16 ipcse = 0, tucse, mss;
3200         u8 ipcss, ipcso, tucss, tucso, hdr_len;
3201         int err;
3202
3203         if (skb_is_gso(skb)) {
3204                 if (skb_header_cloned(skb)) {
3205                         err = pskb_expand_head(skb, 0, 0, GFP_ATOMIC);
3206                         if (err)
3207                                 return err;
3208                 }
3209
3210                 hdr_len = skb_transport_offset(skb) + tcp_hdrlen(skb);
3211                 mss = skb_shinfo(skb)->gso_size;
3212                 if (skb->protocol == htons(ETH_P_IP)) {
3213                         struct iphdr *iph = ip_hdr(skb);
3214                         iph->tot_len = 0;
3215                         iph->check = 0;
3216                         tcp_hdr(skb)->check = ~csum_tcpudp_magic(iph->saddr,
3217                                                                  iph->daddr, 0,
3218                                                                  IPPROTO_TCP,
3219                                                                  0);
3220                         cmd_length = E1000_TXD_CMD_IP;
3221                         ipcse = skb_transport_offset(skb) - 1;
3222                 } else if (skb_shinfo(skb)->gso_type == SKB_GSO_TCPV6) {
3223                         ipv6_hdr(skb)->payload_len = 0;
3224                         tcp_hdr(skb)->check =
3225                                 ~csum_ipv6_magic(&ipv6_hdr(skb)->saddr,
3226                                                  &ipv6_hdr(skb)->daddr,
3227                                                  0, IPPROTO_TCP, 0);
3228                         ipcse = 0;
3229                 }
3230                 ipcss = skb_network_offset(skb);
3231                 ipcso = (void *)&(ip_hdr(skb)->check) - (void *)skb->data;
3232                 tucss = skb_transport_offset(skb);
3233                 tucso = (void *)&(tcp_hdr(skb)->check) - (void *)skb->data;
3234                 tucse = 0;
3235
3236                 cmd_length |= (E1000_TXD_CMD_DEXT | E1000_TXD_CMD_TSE |
3237                                E1000_TXD_CMD_TCP | (skb->len - (hdr_len)));
3238
3239                 i = tx_ring->next_to_use;
3240                 context_desc = E1000_CONTEXT_DESC(*tx_ring, i);
3241                 buffer_info = &tx_ring->buffer_info[i];
3242
3243                 context_desc->lower_setup.ip_fields.ipcss  = ipcss;
3244                 context_desc->lower_setup.ip_fields.ipcso  = ipcso;
3245                 context_desc->lower_setup.ip_fields.ipcse  = cpu_to_le16(ipcse);
3246                 context_desc->upper_setup.tcp_fields.tucss = tucss;
3247                 context_desc->upper_setup.tcp_fields.tucso = tucso;
3248                 context_desc->upper_setup.tcp_fields.tucse = cpu_to_le16(tucse);
3249                 context_desc->tcp_seg_setup.fields.mss     = cpu_to_le16(mss);
3250                 context_desc->tcp_seg_setup.fields.hdr_len = hdr_len;
3251                 context_desc->cmd_and_length = cpu_to_le32(cmd_length);
3252
3253                 buffer_info->time_stamp = jiffies;
3254                 buffer_info->next_to_watch = i;
3255
3256                 i++;
3257                 if (i == tx_ring->count)
3258                         i = 0;
3259                 tx_ring->next_to_use = i;
3260
3261                 return 1;
3262         }
3263
3264         return 0;
3265 }
3266
3267 static bool e1000_tx_csum(struct e1000_adapter *adapter, struct sk_buff *skb)
3268 {
3269         struct e1000_ring *tx_ring = adapter->tx_ring;
3270         struct e1000_context_desc *context_desc;
3271         struct e1000_buffer *buffer_info;
3272         unsigned int i;
3273         u8 css;
3274
3275         if (skb->ip_summed == CHECKSUM_PARTIAL) {
3276                 css = skb_transport_offset(skb);
3277
3278                 i = tx_ring->next_to_use;
3279                 buffer_info = &tx_ring->buffer_info[i];
3280                 context_desc = E1000_CONTEXT_DESC(*tx_ring, i);
3281
3282                 context_desc->lower_setup.ip_config = 0;
3283                 context_desc->upper_setup.tcp_fields.tucss = css;
3284                 context_desc->upper_setup.tcp_fields.tucso =
3285                                         css + skb->csum_offset;
3286                 context_desc->upper_setup.tcp_fields.tucse = 0;
3287                 context_desc->tcp_seg_setup.data = 0;
3288                 context_desc->cmd_and_length = cpu_to_le32(E1000_TXD_CMD_DEXT);
3289
3290                 buffer_info->time_stamp = jiffies;
3291                 buffer_info->next_to_watch = i;
3292
3293                 i++;
3294                 if (i == tx_ring->count)
3295                         i = 0;
3296                 tx_ring->next_to_use = i;
3297
3298                 return 1;
3299         }
3300
3301         return 0;
3302 }
3303
3304 #define E1000_MAX_PER_TXD       8192
3305 #define E1000_MAX_TXD_PWR       12
3306
3307 static int e1000_tx_map(struct e1000_adapter *adapter,
3308                         struct sk_buff *skb, unsigned int first,
3309                         unsigned int max_per_txd, unsigned int nr_frags,
3310                         unsigned int mss)
3311 {
3312         struct e1000_ring *tx_ring = adapter->tx_ring;
3313         struct e1000_buffer *buffer_info;
3314         unsigned int len = skb->len - skb->data_len;
3315         unsigned int offset = 0, size, count = 0, i;
3316         unsigned int f;
3317
3318         i = tx_ring->next_to_use;
3319
3320         while (len) {
3321                 buffer_info = &tx_ring->buffer_info[i];
3322                 size = min(len, max_per_txd);
3323
3324                 /* Workaround for premature desc write-backs
3325                  * in TSO mode.  Append 4-byte sentinel desc */
3326                 if (mss && !nr_frags && size == len && size > 8)
3327                         size -= 4;
3328
3329                 buffer_info->length = size;
3330                 /* set time_stamp *before* dma to help avoid a possible race */
3331                 buffer_info->time_stamp = jiffies;
3332                 buffer_info->dma =
3333                         pci_map_single(adapter->pdev,
3334                                 skb->data + offset,
3335                                 size,
3336                                 PCI_DMA_TODEVICE);
3337                 if (pci_dma_mapping_error(adapter->pdev, buffer_info->dma)) {
3338                         dev_err(&adapter->pdev->dev, "TX DMA map failed\n");
3339                         adapter->tx_dma_failed++;
3340                         return -1;
3341                 }
3342                 buffer_info->next_to_watch = i;
3343
3344                 len -= size;
3345                 offset += size;
3346                 count++;
3347                 i++;
3348                 if (i == tx_ring->count)
3349                         i = 0;
3350         }
3351
3352         for (f = 0; f < nr_frags; f++) {
3353                 struct skb_frag_struct *frag;
3354
3355                 frag = &skb_shinfo(skb)->frags[f];
3356                 len = frag->size;
3357                 offset = frag->page_offset;
3358
3359                 while (len) {
3360                         buffer_info = &tx_ring->buffer_info[i];
3361                         size = min(len, max_per_txd);
3362                         /* Workaround for premature desc write-backs
3363                          * in TSO mode.  Append 4-byte sentinel desc */
3364                         if (mss && f == (nr_frags-1) && size == len && size > 8)
3365                                 size -= 4;
3366
3367                         buffer_info->length = size;
3368                         buffer_info->time_stamp = jiffies;
3369                         buffer_info->dma =
3370                                 pci_map_page(adapter->pdev,
3371                                         frag->page,
3372                                         offset,
3373                                         size,
3374                                         PCI_DMA_TODEVICE);
3375                         if (pci_dma_mapping_error(adapter->pdev,
3376                                                   buffer_info->dma)) {
3377                                 dev_err(&adapter->pdev->dev,
3378                                         "TX DMA page map failed\n");
3379                                 adapter->tx_dma_failed++;
3380                                 return -1;
3381                         }
3382
3383                         buffer_info->next_to_watch = i;
3384
3385                         len -= size;
3386                         offset += size;
3387                         count++;
3388
3389                         i++;
3390                         if (i == tx_ring->count)
3391                                 i = 0;
3392                 }
3393         }
3394
3395         if (i == 0)
3396                 i = tx_ring->count - 1;
3397         else
3398                 i--;
3399
3400         tx_ring->buffer_info[i].skb = skb;
3401         tx_ring->buffer_info[first].next_to_watch = i;
3402
3403         return count;
3404 }
3405
3406 static void e1000_tx_queue(struct e1000_adapter *adapter,
3407                            int tx_flags, int count)
3408 {
3409         struct e1000_ring *tx_ring = adapter->tx_ring;
3410         struct e1000_tx_desc *tx_desc = NULL;
3411         struct e1000_buffer *buffer_info;
3412         u32 txd_upper = 0, txd_lower = E1000_TXD_CMD_IFCS;
3413         unsigned int i;
3414
3415         if (tx_flags & E1000_TX_FLAGS_TSO) {
3416                 txd_lower |= E1000_TXD_CMD_DEXT | E1000_TXD_DTYP_D |
3417                              E1000_TXD_CMD_TSE;
3418                 txd_upper |= E1000_TXD_POPTS_TXSM << 8;
3419
3420                 if (tx_flags & E1000_TX_FLAGS_IPV4)
3421                         txd_upper |= E1000_TXD_POPTS_IXSM << 8;
3422         }
3423
3424         if (tx_flags & E1000_TX_FLAGS_CSUM) {
3425                 txd_lower |= E1000_TXD_CMD_DEXT | E1000_TXD_DTYP_D;
3426                 txd_upper |= E1000_TXD_POPTS_TXSM << 8;
3427         }
3428
3429         if (tx_flags & E1000_TX_FLAGS_VLAN) {
3430                 txd_lower |= E1000_TXD_CMD_VLE;
3431                 txd_upper |= (tx_flags & E1000_TX_FLAGS_VLAN_MASK);
3432         }
3433
3434         i = tx_ring->next_to_use;
3435
3436         while (count--) {
3437                 buffer_info = &tx_ring->buffer_info[i];
3438                 tx_desc = E1000_TX_DESC(*tx_ring, i);
3439                 tx_desc->buffer_addr = cpu_to_le64(buffer_info->dma);
3440                 tx_desc->lower.data =
3441                         cpu_to_le32(txd_lower | buffer_info->length);
3442                 tx_desc->upper.data = cpu_to_le32(txd_upper);
3443
3444                 i++;
3445                 if (i == tx_ring->count)
3446                         i = 0;
3447         }
3448
3449         tx_desc->lower.data |= cpu_to_le32(adapter->txd_cmd);
3450
3451         /*
3452          * Force memory writes to complete before letting h/w
3453          * know there are new descriptors to fetch.  (Only
3454          * applicable for weak-ordered memory model archs,
3455          * such as IA-64).
3456          */
3457         wmb();
3458
3459         tx_ring->next_to_use = i;
3460         writel(i, adapter->hw.hw_addr + tx_ring->tail);
3461         /*
3462          * we need this if more than one processor can write to our tail
3463          * at a time, it synchronizes IO on IA64/Altix systems
3464          */
3465         mmiowb();
3466 }
3467
3468 #define MINIMUM_DHCP_PACKET_SIZE 282
3469 static int e1000_transfer_dhcp_info(struct e1000_adapter *adapter,
3470                                     struct sk_buff *skb)
3471 {
3472         struct e1000_hw *hw =  &adapter->hw;
3473         u16 length, offset;
3474
3475         if (vlan_tx_tag_present(skb)) {
3476                 if (!((vlan_tx_tag_get(skb) == adapter->hw.mng_cookie.vlan_id)
3477                     && (adapter->hw.mng_cookie.status &
3478                         E1000_MNG_DHCP_COOKIE_STATUS_VLAN)))
3479                         return 0;
3480         }
3481
3482         if (skb->len <= MINIMUM_DHCP_PACKET_SIZE)
3483                 return 0;
3484
3485         if (((struct ethhdr *) skb->data)->h_proto != htons(ETH_P_IP))
3486                 return 0;
3487
3488         {
3489                 const struct iphdr *ip = (struct iphdr *)((u8 *)skb->data+14);
3490                 struct udphdr *udp;
3491
3492                 if (ip->protocol != IPPROTO_UDP)
3493                         return 0;
3494
3495                 udp = (struct udphdr *)((u8 *)ip + (ip->ihl << 2));
3496                 if (ntohs(udp->dest) != 67)
3497                         return 0;
3498
3499                 offset = (u8 *)udp + 8 - skb->data;
3500                 length = skb->len - offset;
3501                 return e1000e_mng_write_dhcp_info(hw, (u8 *)udp + 8, length);
3502         }
3503
3504         return 0;
3505 }
3506
3507 static int __e1000_maybe_stop_tx(struct net_device *netdev, int size)
3508 {
3509         struct e1000_adapter *adapter = netdev_priv(netdev);
3510
3511         netif_stop_queue(netdev);
3512         /*
3513          * Herbert's original patch had:
3514          *  smp_mb__after_netif_stop_queue();
3515          * but since that doesn't exist yet, just open code it.
3516          */
3517         smp_mb();
3518
3519         /*
3520          * We need to check again in a case another CPU has just
3521          * made room available.
3522          */
3523         if (e1000_desc_unused(adapter->tx_ring) < size)
3524                 return -EBUSY;
3525
3526         /* A reprieve! */
3527         netif_start_queue(netdev);
3528         ++adapter->restart_queue;
3529         return 0;
3530 }
3531
3532 static int e1000_maybe_stop_tx(struct net_device *netdev, int size)
3533 {
3534         struct e1000_adapter *adapter = netdev_priv(netdev);
3535
3536         if (e1000_desc_unused(adapter->tx_ring) >= size)
3537                 return 0;
3538         return __e1000_maybe_stop_tx(netdev, size);
3539 }
3540
3541 #define TXD_USE_COUNT(S, X) (((S) >> (X)) + 1 )
3542 static int e1000_xmit_frame(struct sk_buff *skb, struct net_device *netdev)
3543 {
3544         struct e1000_adapter *adapter = netdev_priv(netdev);
3545         struct e1000_ring *tx_ring = adapter->tx_ring;
3546         unsigned int first;
3547         unsigned int max_per_txd = E1000_MAX_PER_TXD;
3548         unsigned int max_txd_pwr = E1000_MAX_TXD_PWR;
3549         unsigned int tx_flags = 0;
3550         unsigned int len = skb->len - skb->data_len;
3551         unsigned long irq_flags;
3552         unsigned int nr_frags;
3553         unsigned int mss;
3554         int count = 0;
3555         int tso;
3556         unsigned int f;
3557
3558         if (test_bit(__E1000_DOWN, &adapter->state)) {
3559                 dev_kfree_skb_any(skb);
3560                 return NETDEV_TX_OK;
3561         }
3562
3563         if (skb->len <= 0) {
3564                 dev_kfree_skb_any(skb);
3565                 return NETDEV_TX_OK;
3566         }
3567
3568         mss = skb_shinfo(skb)->gso_size;
3569         /*
3570          * The controller does a simple calculation to
3571          * make sure there is enough room in the FIFO before
3572          * initiating the DMA for each buffer.  The calc is:
3573          * 4 = ceil(buffer len/mss).  To make sure we don't
3574          * overrun the FIFO, adjust the max buffer len if mss
3575          * drops.
3576          */
3577         if (mss) {
3578                 u8 hdr_len;
3579                 max_per_txd = min(mss << 2, max_per_txd);
3580                 max_txd_pwr = fls(max_per_txd) - 1;
3581
3582                 /*
3583                  * TSO Workaround for 82571/2/3 Controllers -- if skb->data
3584                  * points to just header, pull a few bytes of payload from
3585                  * frags into skb->data
3586                  */
3587                 hdr_len = skb_transport_offset(skb) + tcp_hdrlen(skb);
3588                 /*
3589                  * we do this workaround for ES2LAN, but it is un-necessary,
3590                  * avoiding it could save a lot of cycles
3591                  */
3592                 if (skb->data_len && (hdr_len == len)) {
3593                         unsigned int pull_size;
3594
3595                         pull_size = min((unsigned int)4, skb->data_len);
3596                         if (!__pskb_pull_tail(skb, pull_size)) {
3597                                 e_err("__pskb_pull_tail failed.\n");
3598                                 dev_kfree_skb_any(skb);
3599                                 return NETDEV_TX_OK;
3600                         }
3601                         len = skb->len - skb->data_len;
3602                 }
3603         }
3604
3605         /* reserve a descriptor for the offload context */
3606         if ((mss) || (skb->ip_summed == CHECKSUM_PARTIAL))
3607                 count++;
3608         count++;
3609
3610         count += TXD_USE_COUNT(len, max_txd_pwr);
3611
3612         nr_frags = skb_shinfo(skb)->nr_frags;
3613         for (f = 0; f < nr_frags; f++)
3614                 count += TXD_USE_COUNT(skb_shinfo(skb)->frags[f].size,
3615                                        max_txd_pwr);
3616
3617         if (adapter->hw.mac.tx_pkt_filtering)
3618                 e1000_transfer_dhcp_info(adapter, skb);
3619
3620         if (!spin_trylock_irqsave(&adapter->tx_queue_lock, irq_flags))
3621                 /* Collision - tell upper layer to requeue */
3622                 return NETDEV_TX_LOCKED;
3623
3624         /*
3625          * need: count + 2 desc gap to keep tail from touching
3626          * head, otherwise try next time
3627          */
3628         if (e1000_maybe_stop_tx(netdev, count + 2)) {
3629                 spin_unlock_irqrestore(&adapter->tx_queue_lock, irq_flags);
3630                 return NETDEV_TX_BUSY;
3631         }
3632
3633         if (adapter->vlgrp && vlan_tx_tag_present(skb)) {
3634                 tx_flags |= E1000_TX_FLAGS_VLAN;
3635                 tx_flags |= (vlan_tx_tag_get(skb) << E1000_TX_FLAGS_VLAN_SHIFT);
3636         }
3637
3638         first = tx_ring->next_to_use;
3639
3640         tso = e1000_tso(adapter, skb);
3641         if (tso < 0) {
3642                 dev_kfree_skb_any(skb);
3643                 spin_unlock_irqrestore(&adapter->tx_queue_lock, irq_flags);
3644                 return NETDEV_TX_OK;
3645         }
3646
3647         if (tso)
3648                 tx_flags |= E1000_TX_FLAGS_TSO;
3649         else if (e1000_tx_csum(adapter, skb))
3650                 tx_flags |= E1000_TX_FLAGS_CSUM;
3651
3652         /*
3653          * Old method was to assume IPv4 packet by default if TSO was enabled.
3654          * 82571 hardware supports TSO capabilities for IPv6 as well...
3655          * no longer assume, we must.
3656          */
3657         if (skb->protocol == htons(ETH_P_IP))
3658                 tx_flags |= E1000_TX_FLAGS_IPV4;
3659
3660         count = e1000_tx_map(adapter, skb, first, max_per_txd, nr_frags, mss);
3661         if (count < 0) {
3662                 /* handle pci_map_single() error in e1000_tx_map */
3663                 dev_kfree_skb_any(skb);
3664                 spin_unlock_irqrestore(&adapter->tx_queue_lock, irq_flags);
3665                 return NETDEV_TX_OK;
3666         }
3667
3668         e1000_tx_queue(adapter, tx_flags, count);
3669
3670         netdev->trans_start = jiffies;
3671
3672         /* Make sure there is space in the ring for the next send. */
3673         e1000_maybe_stop_tx(netdev, MAX_SKB_FRAGS + 2);
3674
3675         spin_unlock_irqrestore(&adapter->tx_queue_lock, irq_flags);
3676         return NETDEV_TX_OK;
3677 }
3678
3679 /**
3680  * e1000_tx_timeout - Respond to a Tx Hang
3681  * @netdev: network interface device structure
3682  **/
3683 static void e1000_tx_timeout(struct net_device *netdev)
3684 {
3685         struct e1000_adapter *adapter = netdev_priv(netdev);
3686
3687         /* Do the reset outside of interrupt context */
3688         adapter->tx_timeout_count++;
3689         schedule_work(&adapter->reset_task);
3690 }
3691
3692 static void e1000_reset_task(struct work_struct *work)
3693 {
3694         struct e1000_adapter *adapter;
3695         adapter = container_of(work, struct e1000_adapter, reset_task);
3696
3697         e1000e_reinit_locked(adapter);
3698 }
3699
3700 /**
3701  * e1000_get_stats - Get System Network Statistics
3702  * @netdev: network interface device structure
3703  *
3704  * Returns the address of the device statistics structure.
3705  * The statistics are actually updated from the timer callback.
3706  **/
3707 static struct net_device_stats *e1000_get_stats(struct net_device *netdev)
3708 {
3709         struct e1000_adapter *adapter = netdev_priv(netdev);
3710
3711         /* only return the current stats */
3712         return &adapter->net_stats;
3713 }
3714
3715 /**
3716  * e1000_change_mtu - Change the Maximum Transfer Unit
3717  * @netdev: network interface device structure
3718  * @new_mtu: new value for maximum frame size
3719  *
3720  * Returns 0 on success, negative on failure
3721  **/
3722 static int e1000_change_mtu(struct net_device *netdev, int new_mtu)
3723 {
3724         struct e1000_adapter *adapter = netdev_priv(netdev);
3725         int max_frame = new_mtu + ETH_HLEN + ETH_FCS_LEN;
3726
3727         if ((max_frame < ETH_ZLEN + ETH_FCS_LEN) ||
3728             (max_frame > MAX_JUMBO_FRAME_SIZE)) {
3729                 e_err("Invalid MTU setting\n");
3730                 return -EINVAL;
3731         }
3732
3733         /* Jumbo frame size limits */
3734         if (max_frame > ETH_FRAME_LEN + ETH_FCS_LEN) {
3735                 if (!(adapter->flags & FLAG_HAS_JUMBO_FRAMES)) {
3736                         e_err("Jumbo Frames not supported.\n");
3737                         return -EINVAL;
3738                 }
3739                 if (adapter->hw.phy.type == e1000_phy_ife) {
3740                         e_err("Jumbo Frames not supported.\n");
3741                         return -EINVAL;
3742                 }
3743         }
3744
3745 #define MAX_STD_JUMBO_FRAME_SIZE 9234
3746         if (max_frame > MAX_STD_JUMBO_FRAME_SIZE) {
3747                 e_err("MTU > 9216 not supported.\n");
3748                 return -EINVAL;
3749         }
3750
3751         while (test_and_set_bit(__E1000_RESETTING, &adapter->state))
3752                 msleep(1);
3753         /* e1000e_down has a dependency on max_frame_size */
3754         adapter->max_frame_size = max_frame;
3755         if (netif_running(netdev))
3756                 e1000e_down(adapter);
3757
3758         /*
3759          * NOTE: netdev_alloc_skb reserves 16 bytes, and typically NET_IP_ALIGN
3760          * means we reserve 2 more, this pushes us to allocate from the next
3761          * larger slab size.
3762          * i.e. RXBUFFER_2048 --> size-4096 slab
3763          * However with the new *_jumbo_rx* routines, jumbo receives will use
3764          * fragmented skbs
3765          */
3766
3767         if (max_frame <= 256)
3768                 adapter->rx_buffer_len = 256;
3769         else if (max_frame <= 512)
3770                 adapter->rx_buffer_len = 512;
3771         else if (max_frame <= 1024)
3772                 adapter->rx_buffer_len = 1024;
3773         else if (max_frame <= 2048)
3774                 adapter->rx_buffer_len = 2048;
3775         else
3776                 adapter->rx_buffer_len = 4096;
3777
3778         /* adjust allocation if LPE protects us, and we aren't using SBP */
3779         if ((max_frame == ETH_FRAME_LEN + ETH_FCS_LEN) ||
3780              (max_frame == ETH_FRAME_LEN + VLAN_HLEN + ETH_FCS_LEN))
3781                 adapter->rx_buffer_len = ETH_FRAME_LEN + VLAN_HLEN
3782                                          + ETH_FCS_LEN;
3783
3784         e_info("changing MTU from %d to %d\n", netdev->mtu, new_mtu);
3785         netdev->mtu = new_mtu;
3786
3787         if (netif_running(netdev))
3788                 e1000e_up(adapter);
3789         else
3790                 e1000e_reset(adapter);
3791
3792         clear_bit(__E1000_RESETTING, &adapter->state);
3793
3794         return 0;
3795 }
3796
3797 static int e1000_mii_ioctl(struct net_device *netdev, struct ifreq *ifr,
3798                            int cmd)
3799 {
3800         struct e1000_adapter *adapter = netdev_priv(netdev);
3801         struct mii_ioctl_data *data = if_mii(ifr);
3802
3803         if (adapter->hw.phy.media_type != e1000_media_type_copper)
3804                 return -EOPNOTSUPP;
3805
3806         switch (cmd) {
3807         case SIOCGMIIPHY:
3808                 data->phy_id = adapter->hw.phy.addr;
3809                 break;
3810         case SIOCGMIIREG:
3811                 if (!capable(CAP_NET_ADMIN))
3812                         return -EPERM;
3813                 switch (data->reg_num & 0x1F) {
3814                 case MII_BMCR:
3815                         data->val_out = adapter->phy_regs.bmcr;
3816                         break;
3817                 case MII_BMSR:
3818                         data->val_out = adapter->phy_regs.bmsr;
3819                         break;
3820                 case MII_PHYSID1:
3821                         data->val_out = (adapter->hw.phy.id >> 16);
3822                         break;
3823                 case MII_PHYSID2:
3824                         data->val_out = (adapter->hw.phy.id & 0xFFFF);
3825                         break;
3826                 case MII_ADVERTISE:
3827                         data->val_out = adapter->phy_regs.advertise;
3828                         break;
3829                 case MII_LPA:
3830                         data->val_out = adapter->phy_regs.lpa;
3831                         break;
3832                 case MII_EXPANSION:
3833                         data->val_out = adapter->phy_regs.expansion;
3834                         break;
3835                 case MII_CTRL1000:
3836                         data->val_out = adapter->phy_regs.ctrl1000;
3837                         break;
3838                 case MII_STAT1000:
3839                         data->val_out = adapter->phy_regs.stat1000;
3840                         break;
3841                 case MII_ESTATUS:
3842                         data->val_out = adapter->phy_regs.estatus;
3843                         break;
3844                 default:
3845                         return -EIO;
3846                 }
3847                 break;
3848         case SIOCSMIIREG:
3849         default:
3850                 return -EOPNOTSUPP;
3851         }
3852         return 0;
3853 }
3854
3855 static int e1000_ioctl(struct net_device *netdev, struct ifreq *ifr, int cmd)
3856 {
3857         switch (cmd) {
3858         case SIOCGMIIPHY:
3859         case SIOCGMIIREG:
3860         case SIOCSMIIREG:
3861                 return e1000_mii_ioctl(netdev, ifr, cmd);
3862         default:
3863                 return -EOPNOTSUPP;
3864         }
3865 }
3866
3867 static int e1000_suspend(struct pci_dev *pdev, pm_message_t state)
3868 {
3869         struct net_device *netdev = pci_get_drvdata(pdev);
3870         struct e1000_adapter *adapter = netdev_priv(netdev);
3871         struct e1000_hw *hw = &adapter->hw;
3872         u32 ctrl, ctrl_ext, rctl, status;
3873         u32 wufc = adapter->wol;
3874         int retval = 0;
3875
3876         netif_device_detach(netdev);
3877
3878         if (netif_running(netdev)) {
3879                 WARN_ON(test_bit(__E1000_RESETTING, &adapter->state));
3880                 e1000e_down(adapter);
3881                 e1000_free_irq(adapter);
3882         }
3883
3884         retval = pci_save_state(pdev);
3885         if (retval)
3886                 return retval;
3887
3888         status = er32(STATUS);
3889         if (status & E1000_STATUS_LU)
3890                 wufc &= ~E1000_WUFC_LNKC;
3891
3892         if (wufc) {
3893                 e1000_setup_rctl(adapter);
3894                 e1000_set_multi(netdev);
3895
3896                 /* turn on all-multi mode if wake on multicast is enabled */
3897                 if (wufc & E1000_WUFC_MC) {
3898                         rctl = er32(RCTL);
3899                         rctl |= E1000_RCTL_MPE;
3900                         ew32(RCTL, rctl);
3901                 }
3902
3903                 ctrl = er32(CTRL);
3904                 /* advertise wake from D3Cold */
3905                 #define E1000_CTRL_ADVD3WUC 0x00100000
3906                 /* phy power management enable */
3907                 #define E1000_CTRL_EN_PHY_PWR_MGMT 0x00200000
3908                 ctrl |= E1000_CTRL_ADVD3WUC |
3909                         E1000_CTRL_EN_PHY_PWR_MGMT;
3910                 ew32(CTRL, ctrl);
3911
3912                 if (adapter->hw.phy.media_type == e1000_media_type_fiber ||
3913                     adapter->hw.phy.media_type ==
3914                     e1000_media_type_internal_serdes) {
3915                         /* keep the laser running in D3 */
3916                         ctrl_ext = er32(CTRL_EXT);
3917                         ctrl_ext |= E1000_CTRL_EXT_SDP7_DATA;
3918                         ew32(CTRL_EXT, ctrl_ext);
3919                 }
3920
3921                 if (adapter->flags & FLAG_IS_ICH)
3922                         e1000e_disable_gig_wol_ich8lan(&adapter->hw);
3923
3924                 /* Allow time for pending master requests to run */
3925                 e1000e_disable_pcie_master(&adapter->hw);
3926
3927                 ew32(WUC, E1000_WUC_PME_EN);
3928                 ew32(WUFC, wufc);
3929                 pci_enable_wake(pdev, PCI_D3hot, 1);
3930                 pci_enable_wake(pdev, PCI_D3cold, 1);
3931         } else {
3932                 ew32(WUC, 0);
3933                 ew32(WUFC, 0);
3934                 pci_enable_wake(pdev, PCI_D3hot, 0);
3935                 pci_enable_wake(pdev, PCI_D3cold, 0);
3936         }
3937
3938         /* make sure adapter isn't asleep if manageability is enabled */
3939         if (adapter->flags & FLAG_MNG_PT_ENABLED) {
3940                 pci_enable_wake(pdev, PCI_D3hot, 1);
3941                 pci_enable_wake(pdev, PCI_D3cold, 1);
3942         }
3943
3944         if (adapter->hw.phy.type == e1000_phy_igp_3)
3945                 e1000e_igp3_phy_powerdown_workaround_ich8lan(&adapter->hw);
3946
3947         /*
3948          * Release control of h/w to f/w.  If f/w is AMT enabled, this
3949          * would have already happened in close and is redundant.
3950          */
3951         e1000_release_hw_control(adapter);
3952
3953         pci_disable_device(pdev);
3954
3955         pci_set_power_state(pdev, pci_choose_state(pdev, state));
3956
3957         return 0;
3958 }
3959
3960 static void e1000e_disable_l1aspm(struct pci_dev *pdev)
3961 {
3962         int pos;
3963         u16 val;
3964
3965         /*
3966          * 82573 workaround - disable L1 ASPM on mobile chipsets
3967          *
3968          * L1 ASPM on various mobile (ich7) chipsets do not behave properly
3969          * resulting in lost data or garbage information on the pci-e link
3970          * level. This could result in (false) bad EEPROM checksum errors,
3971          * long ping times (up to 2s) or even a system freeze/hang.
3972          *
3973          * Unfortunately this feature saves about 1W power consumption when
3974          * active.
3975          */
3976         pos = pci_find_capability(pdev, PCI_CAP_ID_EXP);
3977         pci_read_config_word(pdev, pos + PCI_EXP_LNKCTL, &val);
3978         if (val & 0x2) {
3979                 dev_warn(&pdev->dev, "Disabling L1 ASPM\n");
3980                 val &= ~0x2;
3981                 pci_write_config_word(pdev, pos + PCI_EXP_LNKCTL, val);
3982         }
3983 }
3984
3985 #ifdef CONFIG_PM
3986 static int e1000_resume(struct pci_dev *pdev)
3987 {
3988         struct net_device *netdev = pci_get_drvdata(pdev);
3989         struct e1000_adapter *adapter = netdev_priv(netdev);
3990         struct e1000_hw *hw = &adapter->hw;
3991         u32 err;
3992
3993         pci_set_power_state(pdev, PCI_D0);
3994         pci_restore_state(pdev);
3995         e1000e_disable_l1aspm(pdev);
3996
3997         err = pci_enable_device_mem(pdev);
3998         if (err) {
3999                 dev_err(&pdev->dev,
4000                         "Cannot enable PCI device from suspend\n");
4001                 return err;
4002         }
4003
4004         pci_set_master(pdev);
4005
4006         pci_enable_wake(pdev, PCI_D3hot, 0);
4007         pci_enable_wake(pdev, PCI_D3cold, 0);
4008
4009         if (netif_running(netdev)) {
4010                 err = e1000_request_irq(adapter);
4011                 if (err)
4012                         return err;
4013         }
4014
4015         e1000e_power_up_phy(adapter);
4016         e1000e_reset(adapter);
4017         ew32(WUS, ~0);
4018
4019         e1000_init_manageability(adapter);
4020
4021         if (netif_running(netdev))
4022                 e1000e_up(adapter);
4023
4024         netif_device_attach(netdev);
4025
4026         /*
4027          * If the controller has AMT, do not set DRV_LOAD until the interface
4028          * is up.  For all other cases, let the f/w know that the h/w is now
4029          * under the control of the driver.
4030          */
4031         if (!(adapter->flags & FLAG_HAS_AMT))
4032                 e1000_get_hw_control(adapter);
4033
4034         return 0;
4035 }
4036 #endif
4037
4038 static void e1000_shutdown(struct pci_dev *pdev)
4039 {
4040         e1000_suspend(pdev, PMSG_SUSPEND);
4041 }
4042
4043 #ifdef CONFIG_NET_POLL_CONTROLLER
4044 /*
4045  * Polling 'interrupt' - used by things like netconsole to send skbs
4046  * without having to re-enable interrupts. It's not called while
4047  * the interrupt routine is executing.
4048  */
4049 static void e1000_netpoll(struct net_device *netdev)
4050 {
4051         struct e1000_adapter *adapter = netdev_priv(netdev);
4052
4053         disable_irq(adapter->pdev->irq);
4054         e1000_intr(adapter->pdev->irq, netdev);
4055
4056         enable_irq(adapter->pdev->irq);
4057 }
4058 #endif
4059
4060 /**
4061  * e1000_io_error_detected - called when PCI error is detected
4062  * @pdev: Pointer to PCI device
4063  * @state: The current pci connection state
4064  *
4065  * This function is called after a PCI bus error affecting
4066  * this device has been detected.
4067  */
4068 static pci_ers_result_t e1000_io_error_detected(struct pci_dev *pdev,
4069                                                 pci_channel_state_t state)
4070 {
4071         struct net_device *netdev = pci_get_drvdata(pdev);
4072         struct e1000_adapter *adapter = netdev_priv(netdev);
4073
4074         netif_device_detach(netdev);
4075
4076         if (netif_running(netdev))
4077                 e1000e_down(adapter);
4078         pci_disable_device(pdev);
4079
4080         /* Request a slot slot reset. */
4081         return PCI_ERS_RESULT_NEED_RESET;
4082 }
4083
4084 /**
4085  * e1000_io_slot_reset - called after the pci bus has been reset.
4086  * @pdev: Pointer to PCI device
4087  *
4088  * Restart the card from scratch, as if from a cold-boot. Implementation
4089  * resembles the first-half of the e1000_resume routine.
4090  */
4091 static pci_ers_result_t e1000_io_slot_reset(struct pci_dev *pdev)
4092 {
4093         struct net_device *netdev = pci_get_drvdata(pdev);
4094         struct e1000_adapter *adapter = netdev_priv(netdev);
4095         struct e1000_hw *hw = &adapter->hw;
4096         int err;
4097
4098         e1000e_disable_l1aspm(pdev);
4099         err = pci_enable_device_mem(pdev);
4100         if (err) {
4101                 dev_err(&pdev->dev,
4102                         "Cannot re-enable PCI device after reset.\n");
4103                 return PCI_ERS_RESULT_DISCONNECT;
4104         }
4105         pci_set_master(pdev);
4106         pci_restore_state(pdev);
4107
4108         pci_enable_wake(pdev, PCI_D3hot, 0);
4109         pci_enable_wake(pdev, PCI_D3cold, 0);
4110
4111         e1000e_reset(adapter);
4112         ew32(WUS, ~0);
4113
4114         return PCI_ERS_RESULT_RECOVERED;
4115 }
4116
4117 /**
4118  * e1000_io_resume - called when traffic can start flowing again.
4119  * @pdev: Pointer to PCI device
4120  *
4121  * This callback is called when the error recovery driver tells us that
4122  * its OK to resume normal operation. Implementation resembles the
4123  * second-half of the e1000_resume routine.
4124  */
4125 static void e1000_io_resume(struct pci_dev *pdev)
4126 {
4127         struct net_device *netdev = pci_get_drvdata(pdev);
4128         struct e1000_adapter *adapter = netdev_priv(netdev);
4129
4130         e1000_init_manageability(adapter);
4131
4132         if (netif_running(netdev)) {
4133                 if (e1000e_up(adapter)) {
4134                         dev_err(&pdev->dev,
4135                                 "can't bring device back up after reset\n");
4136                         return;
4137                 }
4138         }
4139
4140         netif_device_attach(netdev);
4141
4142         /*
4143          * If the controller has AMT, do not set DRV_LOAD until the interface
4144          * is up.  For all other cases, let the f/w know that the h/w is now
4145          * under the control of the driver.
4146          */
4147         if (!(adapter->flags & FLAG_HAS_AMT))
4148                 e1000_get_hw_control(adapter);
4149
4150 }
4151
4152 static void e1000_print_device_info(struct e1000_adapter *adapter)
4153 {
4154         struct e1000_hw *hw = &adapter->hw;
4155         struct net_device *netdev = adapter->netdev;
4156         u32 pba_num;
4157
4158         /* print bus type/speed/width info */
4159         e_info("(PCI Express:2.5GB/s:%s) %02x:%02x:%02x:%02x:%02x:%02x\n",
4160                /* bus width */
4161                ((hw->bus.width == e1000_bus_width_pcie_x4) ? "Width x4" :
4162                 "Width x1"),
4163                /* MAC address */
4164                netdev->dev_addr[0], netdev->dev_addr[1],
4165                netdev->dev_addr[2], netdev->dev_addr[3],
4166                netdev->dev_addr[4], netdev->dev_addr[5]);
4167         e_info("Intel(R) PRO/%s Network Connection\n",
4168                (hw->phy.type == e1000_phy_ife) ? "10/100" : "1000");
4169         e1000e_read_pba_num(hw, &pba_num);
4170         e_info("MAC: %d, PHY: %d, PBA No: %06x-%03x\n",
4171                hw->mac.type, hw->phy.type, (pba_num >> 8), (pba_num & 0xff));
4172 }
4173
4174 static void e1000_eeprom_checks(struct e1000_adapter *adapter)
4175 {
4176         struct e1000_hw *hw = &adapter->hw;
4177         int ret_val;
4178         u16 buf = 0;
4179
4180         if (hw->mac.type != e1000_82573)
4181                 return;
4182
4183         ret_val = e1000_read_nvm(hw, NVM_INIT_CONTROL2_REG, 1, &buf);
4184         if (!(le16_to_cpu(buf) & (1 << 0))) {
4185                 /* Deep Smart Power Down (DSPD) */
4186                 e_warn("Warning: detected DSPD enabled in EEPROM\n");
4187         }
4188
4189         ret_val = e1000_read_nvm(hw, NVM_INIT_3GIO_3, 1, &buf);
4190         if (le16_to_cpu(buf) & (3 << 2)) {
4191                 /* ASPM enable */
4192                 e_warn("Warning: detected ASPM enabled in EEPROM\n");
4193         }
4194 }
4195
4196 /**
4197  * e1000_probe - Device Initialization Routine
4198  * @pdev: PCI device information struct
4199  * @ent: entry in e1000_pci_tbl
4200  *
4201  * Returns 0 on success, negative on failure
4202  *
4203  * e1000_probe initializes an adapter identified by a pci_dev structure.
4204  * The OS initialization, configuring of the adapter private structure,
4205  * and a hardware reset occur.
4206  **/
4207 static int __devinit e1000_probe(struct pci_dev *pdev,
4208                                  const struct pci_device_id *ent)
4209 {
4210         struct net_device *netdev;
4211         struct e1000_adapter *adapter;
4212         struct e1000_hw *hw;
4213         const struct e1000_info *ei = e1000_info_tbl[ent->driver_data];
4214         resource_size_t mmio_start, mmio_len;
4215         resource_size_t flash_start, flash_len;
4216
4217         static int cards_found;
4218         int i, err, pci_using_dac;
4219         u16 eeprom_data = 0;
4220         u16 eeprom_apme_mask = E1000_EEPROM_APME;
4221
4222         e1000e_disable_l1aspm(pdev);
4223
4224         err = pci_enable_device_mem(pdev);
4225         if (err)
4226                 return err;
4227
4228         pci_using_dac = 0;
4229         err = pci_set_dma_mask(pdev, DMA_64BIT_MASK);
4230         if (!err) {
4231                 err = pci_set_consistent_dma_mask(pdev, DMA_64BIT_MASK);
4232                 if (!err)
4233                         pci_using_dac = 1;
4234         } else {
4235                 err = pci_set_dma_mask(pdev, DMA_32BIT_MASK);
4236                 if (err) {
4237                         err = pci_set_consistent_dma_mask(pdev,
4238                                                           DMA_32BIT_MASK);
4239                         if (err) {
4240                                 dev_err(&pdev->dev, "No usable DMA "
4241                                         "configuration, aborting\n");
4242                                 goto err_dma;
4243                         }
4244                 }
4245         }
4246
4247         err = pci_request_selected_regions(pdev,
4248                                           pci_select_bars(pdev, IORESOURCE_MEM),
4249                                           e1000e_driver_name);
4250         if (err)
4251                 goto err_pci_reg;
4252
4253         pci_set_master(pdev);
4254         pci_save_state(pdev);
4255
4256         err = -ENOMEM;
4257         netdev = alloc_etherdev(sizeof(struct e1000_adapter));
4258         if (!netdev)
4259                 goto err_alloc_etherdev;
4260
4261         SET_NETDEV_DEV(netdev, &pdev->dev);
4262
4263         pci_set_drvdata(pdev, netdev);
4264         adapter = netdev_priv(netdev);
4265         hw = &adapter->hw;
4266         adapter->netdev = netdev;
4267         adapter->pdev = pdev;
4268         adapter->ei = ei;
4269         adapter->pba = ei->pba;
4270         adapter->flags = ei->flags;
4271         adapter->hw.adapter = adapter;
4272         adapter->hw.mac.type = ei->mac;
4273         adapter->msg_enable = (1 << NETIF_MSG_DRV | NETIF_MSG_PROBE) - 1;
4274
4275         mmio_start = pci_resource_start(pdev, 0);
4276         mmio_len = pci_resource_len(pdev, 0);
4277
4278         err = -EIO;
4279         adapter->hw.hw_addr = ioremap(mmio_start, mmio_len);
4280         if (!adapter->hw.hw_addr)
4281                 goto err_ioremap;
4282
4283         if ((adapter->flags & FLAG_HAS_FLASH) &&
4284             (pci_resource_flags(pdev, 1) & IORESOURCE_MEM)) {
4285                 flash_start = pci_resource_start(pdev, 1);
4286                 flash_len = pci_resource_len(pdev, 1);
4287                 adapter->hw.flash_address = ioremap(flash_start, flash_len);
4288                 if (!adapter->hw.flash_address)
4289                         goto err_flashmap;
4290         }
4291
4292         /* construct the net_device struct */
4293         netdev->open                    = &e1000_open;
4294         netdev->stop                    = &e1000_close;
4295         netdev->hard_start_xmit         = &e1000_xmit_frame;
4296         netdev->get_stats               = &e1000_get_stats;
4297         netdev->set_multicast_list      = &e1000_set_multi;
4298         netdev->set_mac_address         = &e1000_set_mac;
4299         netdev->change_mtu              = &e1000_change_mtu;
4300         netdev->do_ioctl                = &e1000_ioctl;
4301         e1000e_set_ethtool_ops(netdev);
4302         netdev->tx_timeout              = &e1000_tx_timeout;
4303         netdev->watchdog_timeo          = 5 * HZ;
4304         netif_napi_add(netdev, &adapter->napi, e1000_clean, 64);
4305         netdev->vlan_rx_register        = e1000_vlan_rx_register;
4306         netdev->vlan_rx_add_vid         = e1000_vlan_rx_add_vid;
4307         netdev->vlan_rx_kill_vid        = e1000_vlan_rx_kill_vid;
4308 #ifdef CONFIG_NET_POLL_CONTROLLER
4309         netdev->poll_controller         = e1000_netpoll;
4310 #endif
4311         strncpy(netdev->name, pci_name(pdev), sizeof(netdev->name) - 1);
4312
4313         netdev->mem_start = mmio_start;
4314         netdev->mem_end = mmio_start + mmio_len;
4315
4316         adapter->bd_number = cards_found++;
4317
4318         /* setup adapter struct */
4319         err = e1000_sw_init(adapter);
4320         if (err)
4321                 goto err_sw_init;
4322
4323         err = -EIO;
4324
4325         memcpy(&hw->mac.ops, ei->mac_ops, sizeof(hw->mac.ops));
4326         memcpy(&hw->nvm.ops, ei->nvm_ops, sizeof(hw->nvm.ops));
4327         memcpy(&hw->phy.ops, ei->phy_ops, sizeof(hw->phy.ops));
4328
4329         err = ei->get_variants(adapter);
4330         if (err)
4331                 goto err_hw_init;
4332
4333         hw->mac.ops.get_bus_info(&adapter->hw);
4334
4335         adapter->hw.phy.autoneg_wait_to_complete = 0;
4336
4337         /* Copper options */
4338         if (adapter->hw.phy.media_type == e1000_media_type_copper) {
4339                 adapter->hw.phy.mdix = AUTO_ALL_MODES;
4340                 adapter->hw.phy.disable_polarity_correction = 0;
4341                 adapter->hw.phy.ms_type = e1000_ms_hw_default;
4342         }
4343
4344         if (e1000_check_reset_block(&adapter->hw))
4345                 e_info("PHY reset is blocked due to SOL/IDER session.\n");
4346
4347         netdev->features = NETIF_F_SG |
4348                            NETIF_F_HW_CSUM |
4349                            NETIF_F_HW_VLAN_TX |
4350                            NETIF_F_HW_VLAN_RX;
4351
4352         if (adapter->flags & FLAG_HAS_HW_VLAN_FILTER)
4353                 netdev->features |= NETIF_F_HW_VLAN_FILTER;
4354
4355         netdev->features |= NETIF_F_TSO;
4356         netdev->features |= NETIF_F_TSO6;
4357
4358         netdev->vlan_features |= NETIF_F_TSO;
4359         netdev->vlan_features |= NETIF_F_TSO6;
4360         netdev->vlan_features |= NETIF_F_HW_CSUM;
4361         netdev->vlan_features |= NETIF_F_SG;
4362
4363         if (pci_using_dac)
4364                 netdev->features |= NETIF_F_HIGHDMA;
4365
4366         /*
4367          * We should not be using LLTX anymore, but we are still Tx faster with
4368          * it.
4369          */
4370         netdev->features |= NETIF_F_LLTX;
4371
4372         if (e1000e_enable_mng_pass_thru(&adapter->hw))
4373                 adapter->flags |= FLAG_MNG_PT_ENABLED;
4374
4375         /*
4376          * before reading the NVM, reset the controller to
4377          * put the device in a known good starting state
4378          */
4379         adapter->hw.mac.ops.reset_hw(&adapter->hw);
4380
4381         /*
4382          * systems with ASPM and others may see the checksum fail on the first
4383          * attempt. Let's give it a few tries
4384          */
4385         for (i = 0;; i++) {
4386                 if (e1000_validate_nvm_checksum(&adapter->hw) >= 0)
4387                         break;
4388                 if (i == 2) {
4389                         e_err("The NVM Checksum Is Not Valid\n");
4390                         err = -EIO;
4391                         goto err_eeprom;
4392                 }
4393         }
4394
4395         e1000_eeprom_checks(adapter);
4396
4397         /* copy the MAC address out of the NVM */
4398         if (e1000e_read_mac_addr(&adapter->hw))
4399                 e_err("NVM Read Error while reading MAC address\n");
4400
4401         memcpy(netdev->dev_addr, adapter->hw.mac.addr, netdev->addr_len);
4402         memcpy(netdev->perm_addr, adapter->hw.mac.addr, netdev->addr_len);
4403
4404         if (!is_valid_ether_addr(netdev->perm_addr)) {
4405                 e_err("Invalid MAC Address: %02x:%02x:%02x:%02x:%02x:%02x\n",
4406                       netdev->perm_addr[0], netdev->perm_addr[1],
4407                       netdev->perm_addr[2], netdev->perm_addr[3],
4408                       netdev->perm_addr[4], netdev->perm_addr[5]);
4409                 err = -EIO;
4410                 goto err_eeprom;
4411         }
4412
4413         init_timer(&adapter->watchdog_timer);
4414         adapter->watchdog_timer.function = &e1000_watchdog;
4415         adapter->watchdog_timer.data = (unsigned long) adapter;
4416
4417         init_timer(&adapter->phy_info_timer);
4418         adapter->phy_info_timer.function = &e1000_update_phy_info;
4419         adapter->phy_info_timer.data = (unsigned long) adapter;
4420
4421         INIT_WORK(&adapter->reset_task, e1000_reset_task);
4422         INIT_WORK(&adapter->watchdog_task, e1000_watchdog_task);
4423
4424         e1000e_check_options(adapter);
4425
4426         /* Initialize link parameters. User can change them with ethtool */
4427         adapter->hw.mac.autoneg = 1;
4428         adapter->fc_autoneg = 1;
4429         adapter->hw.fc.original_type = e1000_fc_default;
4430         adapter->hw.fc.type = e1000_fc_default;
4431         adapter->hw.phy.autoneg_advertised = 0x2f;
4432
4433         /* ring size defaults */
4434         adapter->rx_ring->count = 256;
4435         adapter->tx_ring->count = 256;
4436
4437         /*
4438          * Initial Wake on LAN setting - If APM wake is enabled in
4439          * the EEPROM, enable the ACPI Magic Packet filter
4440          */
4441         if (adapter->flags & FLAG_APME_IN_WUC) {
4442                 /* APME bit in EEPROM is mapped to WUC.APME */
4443                 eeprom_data = er32(WUC);
4444                 eeprom_apme_mask = E1000_WUC_APME;
4445         } else if (adapter->flags & FLAG_APME_IN_CTRL3) {
4446                 if (adapter->flags & FLAG_APME_CHECK_PORT_B &&
4447                     (adapter->hw.bus.func == 1))
4448                         e1000_read_nvm(&adapter->hw,
4449                                 NVM_INIT_CONTROL3_PORT_B, 1, &eeprom_data);
4450                 else
4451                         e1000_read_nvm(&adapter->hw,
4452                                 NVM_INIT_CONTROL3_PORT_A, 1, &eeprom_data);
4453         }
4454
4455         /* fetch WoL from EEPROM */
4456         if (eeprom_data & eeprom_apme_mask)
4457                 adapter->eeprom_wol |= E1000_WUFC_MAG;
4458
4459         /*
4460          * now that we have the eeprom settings, apply the special cases
4461          * where the eeprom may be wrong or the board simply won't support
4462          * wake on lan on a particular port
4463          */
4464         if (!(adapter->flags & FLAG_HAS_WOL))
4465                 adapter->eeprom_wol = 0;
4466
4467         /* initialize the wol settings based on the eeprom settings */
4468         adapter->wol = adapter->eeprom_wol;
4469
4470         /* reset the hardware with the new settings */
4471         e1000e_reset(adapter);
4472
4473         /*
4474          * If the controller has AMT, do not set DRV_LOAD until the interface
4475          * is up.  For all other cases, let the f/w know that the h/w is now
4476          * under the control of the driver.
4477          */
4478         if (!(adapter->flags & FLAG_HAS_AMT))
4479                 e1000_get_hw_control(adapter);
4480
4481         /* tell the stack to leave us alone until e1000_open() is called */
4482         netif_carrier_off(netdev);
4483         netif_tx_stop_all_queues(netdev);
4484
4485         strcpy(netdev->name, "eth%d");
4486         err = register_netdev(netdev);
4487         if (err)
4488                 goto err_register;
4489
4490         e1000_print_device_info(adapter);
4491
4492         return 0;
4493
4494 err_register:
4495         if (!(adapter->flags & FLAG_HAS_AMT))
4496                 e1000_release_hw_control(adapter);
4497 err_eeprom:
4498         if (!e1000_check_reset_block(&adapter->hw))
4499                 e1000_phy_hw_reset(&adapter->hw);
4500 err_hw_init:
4501
4502         kfree(adapter->tx_ring);
4503         kfree(adapter->rx_ring);
4504 err_sw_init:
4505         if (adapter->hw.flash_address)
4506                 iounmap(adapter->hw.flash_address);
4507 err_flashmap:
4508         iounmap(adapter->hw.hw_addr);
4509 err_ioremap:
4510         free_netdev(netdev);
4511 err_alloc_etherdev:
4512         pci_release_selected_regions(pdev,
4513                                      pci_select_bars(pdev, IORESOURCE_MEM));
4514 err_pci_reg:
4515 err_dma:
4516         pci_disable_device(pdev);
4517         return err;
4518 }
4519
4520 /**
4521  * e1000_remove - Device Removal Routine
4522  * @pdev: PCI device information struct
4523  *
4524  * e1000_remove is called by the PCI subsystem to alert the driver
4525  * that it should release a PCI device.  The could be caused by a
4526  * Hot-Plug event, or because the driver is going to be removed from
4527  * memory.
4528  **/
4529 static void __devexit e1000_remove(struct pci_dev *pdev)
4530 {
4531         struct net_device *netdev = pci_get_drvdata(pdev);
4532         struct e1000_adapter *adapter = netdev_priv(netdev);
4533
4534         /*
4535          * flush_scheduled work may reschedule our watchdog task, so
4536          * explicitly disable watchdog tasks from being rescheduled
4537          */
4538         set_bit(__E1000_DOWN, &adapter->state);
4539         del_timer_sync(&adapter->watchdog_timer);
4540         del_timer_sync(&adapter->phy_info_timer);
4541
4542         flush_scheduled_work();
4543
4544         /*
4545          * Release control of h/w to f/w.  If f/w is AMT enabled, this
4546          * would have already happened in close and is redundant.
4547          */
4548         e1000_release_hw_control(adapter);
4549
4550         unregister_netdev(netdev);
4551
4552         if (!e1000_check_reset_block(&adapter->hw))
4553                 e1000_phy_hw_reset(&adapter->hw);
4554
4555         kfree(adapter->tx_ring);
4556         kfree(adapter->rx_ring);
4557
4558         iounmap(adapter->hw.hw_addr);
4559         if (adapter->hw.flash_address)
4560                 iounmap(adapter->hw.flash_address);
4561         pci_release_selected_regions(pdev,
4562                                      pci_select_bars(pdev, IORESOURCE_MEM));
4563
4564         free_netdev(netdev);
4565
4566         pci_disable_device(pdev);
4567 }
4568
4569 /* PCI Error Recovery (ERS) */
4570 static struct pci_error_handlers e1000_err_handler = {
4571         .error_detected = e1000_io_error_detected,
4572         .slot_reset = e1000_io_slot_reset,
4573         .resume = e1000_io_resume,
4574 };
4575
4576 static struct pci_device_id e1000_pci_tbl[] = {
4577         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82571EB_COPPER), board_82571 },
4578         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82571EB_FIBER), board_82571 },
4579         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82571EB_QUAD_COPPER), board_82571 },
4580         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82571EB_QUAD_COPPER_LP), board_82571 },
4581         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82571EB_QUAD_FIBER), board_82571 },
4582         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82571EB_SERDES), board_82571 },
4583         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82571EB_SERDES_DUAL), board_82571 },
4584         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82571EB_SERDES_QUAD), board_82571 },
4585         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82571PT_QUAD_COPPER), board_82571 },
4586
4587         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82572EI), board_82572 },
4588         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82572EI_COPPER), board_82572 },
4589         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82572EI_FIBER), board_82572 },
4590         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82572EI_SERDES), board_82572 },
4591
4592         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82573E), board_82573 },
4593         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82573E_IAMT), board_82573 },
4594         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82573L), board_82573 },
4595
4596         { PCI_VDEVICE(INTEL, E1000_DEV_ID_80003ES2LAN_COPPER_DPT),
4597           board_80003es2lan },
4598         { PCI_VDEVICE(INTEL, E1000_DEV_ID_80003ES2LAN_COPPER_SPT),
4599           board_80003es2lan },
4600         { PCI_VDEVICE(INTEL, E1000_DEV_ID_80003ES2LAN_SERDES_DPT),
4601           board_80003es2lan },
4602         { PCI_VDEVICE(INTEL, E1000_DEV_ID_80003ES2LAN_SERDES_SPT),
4603           board_80003es2lan },
4604
4605         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH8_IFE), board_ich8lan },
4606         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH8_IFE_G), board_ich8lan },
4607         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH8_IFE_GT), board_ich8lan },
4608         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH8_IGP_AMT), board_ich8lan },
4609         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH8_IGP_C), board_ich8lan },
4610         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH8_IGP_M), board_ich8lan },
4611         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH8_IGP_M_AMT), board_ich8lan },
4612
4613         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH9_IFE), board_ich9lan },
4614         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH9_IFE_G), board_ich9lan },
4615         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH9_IFE_GT), board_ich9lan },
4616         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH9_IGP_AMT), board_ich9lan },
4617         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH9_IGP_C), board_ich9lan },
4618         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH9_IGP_M), board_ich9lan },
4619         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH9_IGP_M_AMT), board_ich9lan },
4620         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH9_IGP_M_V), board_ich9lan },
4621
4622         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH10_R_BM_LM), board_ich9lan },
4623         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH10_R_BM_LF), board_ich9lan },
4624         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH10_R_BM_V), board_ich9lan },
4625
4626         { }     /* terminate list */
4627 };
4628 MODULE_DEVICE_TABLE(pci, e1000_pci_tbl);
4629
4630 /* PCI Device API Driver */
4631 static struct pci_driver e1000_driver = {
4632         .name     = e1000e_driver_name,
4633         .id_table = e1000_pci_tbl,
4634         .probe    = e1000_probe,
4635         .remove   = __devexit_p(e1000_remove),
4636 #ifdef CONFIG_PM
4637         /* Power Management Hooks */
4638         .suspend  = e1000_suspend,
4639         .resume   = e1000_resume,
4640 #endif
4641         .shutdown = e1000_shutdown,
4642         .err_handler = &e1000_err_handler
4643 };
4644
4645 /**
4646  * e1000_init_module - Driver Registration Routine
4647  *
4648  * e1000_init_module is the first routine called when the driver is
4649  * loaded. All it does is register with the PCI subsystem.
4650  **/
4651 static int __init e1000_init_module(void)
4652 {
4653         int ret;
4654         printk(KERN_INFO "%s: Intel(R) PRO/1000 Network Driver - %s\n",
4655                e1000e_driver_name, e1000e_driver_version);
4656         printk(KERN_INFO "%s: Copyright (c) 1999-2008 Intel Corporation.\n",
4657                e1000e_driver_name);
4658         ret = pci_register_driver(&e1000_driver);
4659         pm_qos_add_requirement(PM_QOS_CPU_DMA_LATENCY, e1000e_driver_name,
4660                                PM_QOS_DEFAULT_VALUE);
4661
4662         return ret;
4663 }
4664 module_init(e1000_init_module);
4665
4666 /**
4667  * e1000_exit_module - Driver Exit Cleanup Routine
4668  *
4669  * e1000_exit_module is called just before the driver is removed
4670  * from memory.
4671  **/
4672 static void __exit e1000_exit_module(void)
4673 {
4674         pci_unregister_driver(&e1000_driver);
4675         pm_qos_remove_requirement(PM_QOS_CPU_DMA_LATENCY, e1000e_driver_name);
4676 }
4677 module_exit(e1000_exit_module);
4678
4679
4680 MODULE_AUTHOR("Intel Corporation, <linux.nics@intel.com>");
4681 MODULE_DESCRIPTION("Intel(R) PRO/1000 Network Driver");
4682 MODULE_LICENSE("GPL");
4683 MODULE_VERSION(DRV_VERSION);
4684
4685 /* e1000_main.c */