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(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(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(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                         ndev_dbg(netdev, "%s: Receive packet consumed "
 488                                  "multiple buffers\n", 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                                 memcpy(new_skb->data - NET_IP_ALIGN,
 514                                        skb->data - NET_IP_ALIGN,
 515                                        length + NET_IP_ALIGN);
 516                                 /* save the skb in buffer_info as good */
 517                                 buffer_info->skb = skb;
 518                                 skb = new_skb;
 519                         }
 520                         /* else just continue with the old one */
 521                 }
 522                 /* end copybreak code */
 523                 skb_put(skb, length);
 524
 525                 /* Receive Checksum Offload */
 526                 e1000_rx_checksum(adapter,
 527                                   (u32)(status) |
 528                                   ((u32)(rx_desc->errors) << 24),
 529                                   le16_to_cpu(rx_desc->csum), skb);
 530
 531                 e1000_receive_skb(adapter, netdev, skb,status,rx_desc->special);
 532
 533 next_desc:
 534                 rx_desc->status = 0;
 535
 536                 /* return some buffers to hardware, one at a time is too slow */
 537                 if (cleaned_count >= E1000_RX_BUFFER_WRITE) {
 538                         adapter->alloc_rx_buf(adapter, cleaned_count);
 539                         cleaned_count = 0;
 540                 }
 541
 542                 /* use prefetched values */
 543                 rx_desc = next_rxd;
 544                 buffer_info = next_buffer;
 545         }
 546         rx_ring->next_to_clean = i;
 547
 548         cleaned_count = e1000_desc_unused(rx_ring);
 549         if (cleaned_count)
 550                 adapter->alloc_rx_buf(adapter, cleaned_count);
 551
 552         adapter->total_rx_bytes += total_rx_bytes;
 553         adapter->total_rx_packets += total_rx_packets;
 554         adapter->net_stats.rx_bytes += total_rx_bytes;
 555         adapter->net_stats.rx_packets += total_rx_packets;
 556         return cleaned;
 557 }
 558
 559 static void e1000_put_txbuf(struct e1000_adapter *adapter,
 560                              struct e1000_buffer *buffer_info)
 561 {
 562         if (buffer_info->dma) {
 563                 pci_unmap_page(adapter->pdev, buffer_info->dma,
 564                                buffer_info->length, PCI_DMA_TODEVICE);
 565                 buffer_info->dma = 0;
 566         }
 567         if (buffer_info->skb) {
 568                 dev_kfree_skb_any(buffer_info->skb);
 569                 buffer_info->skb = NULL;
 570         }
 571 }
 572
 573 static void e1000_print_tx_hang(struct e1000_adapter *adapter)
 574 {
 575         struct e1000_ring *tx_ring = adapter->tx_ring;
 576         unsigned int i = tx_ring->next_to_clean;
 577         unsigned int eop = tx_ring->buffer_info[i].next_to_watch;
 578         struct e1000_tx_desc *eop_desc = E1000_TX_DESC(*tx_ring, eop);
 579         struct net_device *netdev = adapter->netdev;
 580
 581         /* detected Tx unit hang */
 582         ndev_err(netdev,
 583                  "Detected Tx Unit Hang:\n"
 584                  "  TDH                  <%x>\n"
 585                  "  TDT                  <%x>\n"
 586                  "  next_to_use          <%x>\n"
 587                  "  next_to_clean        <%x>\n"
 588                  "buffer_info[next_to_clean]:\n"
 589                  "  time_stamp           <%lx>\n"
 590                  "  next_to_watch        <%x>\n"
 591                  "  jiffies              <%lx>\n"
 592                  "  next_to_watch.status <%x>\n",
 593                  readl(adapter->hw.hw_addr + tx_ring->head),
 594                  readl(adapter->hw.hw_addr + tx_ring->tail),
 595                  tx_ring->next_to_use,
 596                  tx_ring->next_to_clean,
 597                  tx_ring->buffer_info[eop].time_stamp,
 598                  eop,
 599                  jiffies,
 600                  eop_desc->upper.fields.status);
 601 }
 602
 603 /**
 604  * e1000_clean_tx_irq - Reclaim resources after transmit completes
 605  * @adapter: board private structure
 606  *
 607  * the return value indicates whether actual cleaning was done, there
 608  * is no guarantee that everything was cleaned
 609  **/
 610 static bool e1000_clean_tx_irq(struct e1000_adapter *adapter)
 611 {
 612         struct net_device *netdev = adapter->netdev;
 613         struct e1000_hw *hw = &adapter->hw;
 614         struct e1000_ring *tx_ring = adapter->tx_ring;
 615         struct e1000_tx_desc *tx_desc, *eop_desc;
 616         struct e1000_buffer *buffer_info;
 617         unsigned int i, eop;
 618         unsigned int count = 0;
 619         bool cleaned = 0;
 620         unsigned int total_tx_bytes = 0, total_tx_packets = 0;
 621
 622         i = tx_ring->next_to_clean;
 623         eop = tx_ring->buffer_info[i].next_to_watch;
 624         eop_desc = E1000_TX_DESC(*tx_ring, eop);
 625
 626         while (eop_desc->upper.data & cpu_to_le32(E1000_TXD_STAT_DD)) {
 627                 for (cleaned = 0; !cleaned; ) {
 628                         tx_desc = E1000_TX_DESC(*tx_ring, i);
 629                         buffer_info = &tx_ring->buffer_info[i];
 630                         cleaned = (i == eop);
 631
 632                         if (cleaned) {
 633                                 struct sk_buff *skb = buffer_info->skb;
 634                                 unsigned int segs, bytecount;
 635                                 segs = skb_shinfo(skb)->gso_segs ?: 1;
 636                                 /* multiply data chunks by size of headers */
 637                                 bytecount = ((segs - 1) * skb_headlen(skb)) +
 638                                             skb->len;
 639                                 total_tx_packets += segs;
 640                                 total_tx_bytes += bytecount;
 641                         }
 642
 643                         e1000_put_txbuf(adapter, buffer_info);
 644                         tx_desc->upper.data = 0;
 645
 646                         i++;
 647                         if (i == tx_ring->count)
 648                                 i = 0;
 649                 }
 650
 651                 eop = tx_ring->buffer_info[i].next_to_watch;
 652                 eop_desc = E1000_TX_DESC(*tx_ring, eop);
 653 #define E1000_TX_WEIGHT 64
 654                 /* weight of a sort for tx, to avoid endless transmit cleanup */
 655                 if (count++ == E1000_TX_WEIGHT)
 656                         break;
 657         }
 658
 659         tx_ring->next_to_clean = i;
 660
 661 #define TX_WAKE_THRESHOLD 32
 662         if (cleaned && netif_carrier_ok(netdev) &&
 663                      e1000_desc_unused(tx_ring) >= TX_WAKE_THRESHOLD) {
 664                 /* Make sure that anybody stopping the queue after this
 665                  * sees the new next_to_clean.
 666                  */
 667                 smp_mb();
 668
 669                 if (netif_queue_stopped(netdev) &&
 670                     !(test_bit(__E1000_DOWN, &adapter->state))) {
 671                         netif_wake_queue(netdev);
 672                         ++adapter->restart_queue;
 673                 }
 674         }
 675
 676         if (adapter->detect_tx_hung) {
 677                 /*
 678                  * Detect a transmit hang in hardware, this serializes the
 679                  * check with the clearing of time_stamp and movement of i
 680                  */
 681                 adapter->detect_tx_hung = 0;
 682                 if (tx_ring->buffer_info[eop].dma &&
 683                     time_after(jiffies, tx_ring->buffer_info[eop].time_stamp
 684                                + (adapter->tx_timeout_factor * HZ))
 685                     && !(er32(STATUS) & E1000_STATUS_TXOFF)) {
 686                         e1000_print_tx_hang(adapter);
 687                         netif_stop_queue(netdev);
 688                 }
 689         }
 690         adapter->total_tx_bytes += total_tx_bytes;
 691         adapter->total_tx_packets += total_tx_packets;
 692         adapter->net_stats.tx_bytes += total_tx_bytes;
 693         adapter->net_stats.tx_packets += total_tx_packets;
 694         return cleaned;
 695 }
 696
 697 /**
 698  * e1000_clean_rx_irq_ps - Send received data up the network stack; packet split
 699  * @adapter: board private structure
 700  *
 701  * the return value indicates whether actual cleaning was done, there
 702  * is no guarantee that everything was cleaned
 703  **/
 704 static bool e1000_clean_rx_irq_ps(struct e1000_adapter *adapter,
 705                                   int *work_done, int work_to_do)
 706 {
 707         union e1000_rx_desc_packet_split *rx_desc, *next_rxd;
 708         struct net_device *netdev = adapter->netdev;
 709         struct pci_dev *pdev = adapter->pdev;
 710         struct e1000_ring *rx_ring = adapter->rx_ring;
 711         struct e1000_buffer *buffer_info, *next_buffer;
 712         struct e1000_ps_page *ps_page;
 713         struct sk_buff *skb;
 714         unsigned int i, j;
 715         u32 length, staterr;
 716         int cleaned_count = 0;
 717         bool cleaned = 0;
 718         unsigned int total_rx_bytes = 0, total_rx_packets = 0;
 719
 720         i = rx_ring->next_to_clean;
 721         rx_desc = E1000_RX_DESC_PS(*rx_ring, i);
 722         staterr = le32_to_cpu(rx_desc->wb.middle.status_error);
 723         buffer_info = &rx_ring->buffer_info[i];
 724
 725         while (staterr & E1000_RXD_STAT_DD) {
 726                 if (*work_done >= work_to_do)
 727                         break;
 728                 (*work_done)++;
 729                 skb = buffer_info->skb;
 730
 731                 /* in the packet split case this is header only */
 732                 prefetch(skb->data - NET_IP_ALIGN);
 733
 734                 i++;
 735                 if (i == rx_ring->count)
 736                         i = 0;
 737                 next_rxd = E1000_RX_DESC_PS(*rx_ring, i);
 738                 prefetch(next_rxd);
 739
 740                 next_buffer = &rx_ring->buffer_info[i];
 741
 742                 cleaned = 1;
 743                 cleaned_count++;
 744                 pci_unmap_single(pdev, buffer_info->dma,
 745                                  adapter->rx_ps_bsize0,
 746                                  PCI_DMA_FROMDEVICE);
 747                 buffer_info->dma = 0;
 748
 749                 if (!(staterr & E1000_RXD_STAT_EOP)) {
 750                         ndev_dbg(netdev, "%s: Packet Split buffers didn't pick "
 751                                  "up the full packet\n", netdev->name);
 752                         dev_kfree_skb_irq(skb);
 753                         goto next_desc;
 754                 }
 755
 756                 if (staterr & E1000_RXDEXT_ERR_FRAME_ERR_MASK) {
 757                         dev_kfree_skb_irq(skb);
 758                         goto next_desc;
 759                 }
 760
 761                 length = le16_to_cpu(rx_desc->wb.middle.length0);
 762
 763                 if (!length) {
 764                         ndev_dbg(netdev, "%s: Last part of the packet spanning"
 765                                  " multiple descriptors\n", netdev->name);
 766                         dev_kfree_skb_irq(skb);
 767                         goto next_desc;
 768                 }
 769
 770                 /* Good Receive */
 771                 skb_put(skb, length);
 772
 773                 {
 774                 /*
 775                  * this looks ugly, but it seems compiler issues make it
 776                  * more efficient than reusing j
 777                  */
 778                 int l1 = le16_to_cpu(rx_desc->wb.upper.length[0]);
 779
 780                 /*
 781                  * page alloc/put takes too long and effects small packet
 782                  * throughput, so unsplit small packets and save the alloc/put
 783                  * only valid in softirq (napi) context to call kmap_*
 784                  */
 785                 if (l1 && (l1 <= copybreak) &&
 786                     ((length + l1) <= adapter->rx_ps_bsize0)) {
 787                         u8 *vaddr;
 788
 789                         ps_page = &buffer_info->ps_pages[0];
 790
 791                         /*
 792                          * there is no documentation about how to call
 793                          * kmap_atomic, so we can't hold the mapping
 794                          * very long
 795                          */
 796                         pci_dma_sync_single_for_cpu(pdev, ps_page->dma,
 797                                 PAGE_SIZE, PCI_DMA_FROMDEVICE);
 798                         vaddr = kmap_atomic(ps_page->page, KM_SKB_DATA_SOFTIRQ);
 799                         memcpy(skb_tail_pointer(skb), vaddr, l1);
 800                         kunmap_atomic(vaddr, KM_SKB_DATA_SOFTIRQ);
 801                         pci_dma_sync_single_for_device(pdev, ps_page->dma,
 802                                 PAGE_SIZE, PCI_DMA_FROMDEVICE);
 803
 804                         skb_put(skb, l1);
 805                         goto copydone;
 806                 } /* if */
 807                 }
 808
 809                 for (j = 0; j < PS_PAGE_BUFFERS; j++) {
 810                         length = le16_to_cpu(rx_desc->wb.upper.length[j]);
 811                         if (!length)
 812                                 break;
 813
 814                         ps_page = &buffer_info->ps_pages[j];
 815                         pci_unmap_page(pdev, ps_page->dma, PAGE_SIZE,
 816                                        PCI_DMA_FROMDEVICE);
 817                         ps_page->dma = 0;
 818                         skb_fill_page_desc(skb, j, ps_page->page, 0, length);
 819                         ps_page->page = NULL;
 820                         skb->len += length;
 821                         skb->data_len += length;
 822                         skb->truesize += length;
 823                 }
 824
 825 copydone:
 826                 total_rx_bytes += skb->len;
 827                 total_rx_packets++;
 828
 829                 e1000_rx_checksum(adapter, staterr, le16_to_cpu(
 830                         rx_desc->wb.lower.hi_dword.csum_ip.csum), skb);
 831
 832                 if (rx_desc->wb.upper.header_status &
 833                            cpu_to_le16(E1000_RXDPS_HDRSTAT_HDRSP))
 834                         adapter->rx_hdr_split++;
 835
 836                 e1000_receive_skb(adapter, netdev, skb,
 837                                   staterr, rx_desc->wb.middle.vlan);
 838
 839 next_desc:
 840                 rx_desc->wb.middle.status_error &= cpu_to_le32(~0xFF);
 841                 buffer_info->skb = NULL;
 842
 843                 /* return some buffers to hardware, one at a time is too slow */
 844                 if (cleaned_count >= E1000_RX_BUFFER_WRITE) {
 845                         adapter->alloc_rx_buf(adapter, cleaned_count);
 846                         cleaned_count = 0;
 847                 }
 848
 849                 /* use prefetched values */
 850                 rx_desc = next_rxd;
 851                 buffer_info = next_buffer;
 852
 853                 staterr = le32_to_cpu(rx_desc->wb.middle.status_error);
 854         }
 855         rx_ring->next_to_clean = i;
 856
 857         cleaned_count = e1000_desc_unused(rx_ring);
 858         if (cleaned_count)
 859                 adapter->alloc_rx_buf(adapter, cleaned_count);
 860
 861         adapter->total_rx_bytes += total_rx_bytes;
 862         adapter->total_rx_packets += total_rx_packets;
 863         adapter->net_stats.rx_bytes += total_rx_bytes;
 864         adapter->net_stats.rx_packets += total_rx_packets;
 865         return cleaned;
 866 }
 867
 868 /**
 869  * e1000_consume_page - helper function
 870  **/
 871 static void e1000_consume_page(struct e1000_buffer *bi, struct sk_buff *skb,
 872                                u16 length)
 873 {
 874         bi->page = NULL;
 875         skb->len += length;
 876         skb->data_len += length;
 877         skb->truesize += length;
 878 }
 879
 880 /**
 881  * e1000_clean_jumbo_rx_irq - Send received data up the network stack; legacy
 882  * @adapter: board private structure
 883  *
 884  * the return value indicates whether actual cleaning was done, there
 885  * is no guarantee that everything was cleaned
 886  **/
 887
 888 static bool e1000_clean_jumbo_rx_irq(struct e1000_adapter *adapter,
 889                                      int *work_done, int work_to_do)
 890 {
 891         struct net_device *netdev = adapter->netdev;
 892         struct pci_dev *pdev = adapter->pdev;
 893         struct e1000_ring *rx_ring = adapter->rx_ring;
 894         struct e1000_rx_desc *rx_desc, *next_rxd;
 895         struct e1000_buffer *buffer_info, *next_buffer;
 896         u32 length;
 897         unsigned int i;
 898         int cleaned_count = 0;
 899         bool cleaned = false;
 900         unsigned int total_rx_bytes=0, total_rx_packets=0;
 901
 902         i = rx_ring->next_to_clean;
 903         rx_desc = E1000_RX_DESC(*rx_ring, i);
 904         buffer_info = &rx_ring->buffer_info[i];
 905
 906         while (rx_desc->status & E1000_RXD_STAT_DD) {
 907                 struct sk_buff *skb;
 908                 u8 status;
 909
 910                 if (*work_done >= work_to_do)
 911                         break;
 912                 (*work_done)++;
 913
 914                 status = rx_desc->status;
 915                 skb = buffer_info->skb;
 916                 buffer_info->skb = NULL;
 917
 918                 ++i;
 919                 if (i == rx_ring->count)
 920                         i = 0;
 921                 next_rxd = E1000_RX_DESC(*rx_ring, i);
 922                 prefetch(next_rxd);
 923
 924                 next_buffer = &rx_ring->buffer_info[i];
 925
 926                 cleaned = true;
 927                 cleaned_count++;
 928                 pci_unmap_page(pdev, buffer_info->dma, PAGE_SIZE,
 929                                PCI_DMA_FROMDEVICE);
 930                 buffer_info->dma = 0;
 931
 932                 length = le16_to_cpu(rx_desc->length);
 933
 934                 /* errors is only valid for DD + EOP descriptors */
 935                 if (unlikely((status & E1000_RXD_STAT_EOP) &&
 936                     (rx_desc->errors & E1000_RXD_ERR_FRAME_ERR_MASK))) {
 937                                 /* recycle both page and skb */
 938                                 buffer_info->skb = skb;
 939                                 /* an error means any chain goes out the window
 940                                  * too */
 941                                 if (rx_ring->rx_skb_top)
 942                                         dev_kfree_skb(rx_ring->rx_skb_top);
 943                                 rx_ring->rx_skb_top = NULL;
 944                                 goto next_desc;
 945                 }
 946
 947 #define rxtop rx_ring->rx_skb_top
 948                 if (!(status & E1000_RXD_STAT_EOP)) {
 949                         /* this descriptor is only the beginning (or middle) */
 950                         if (!rxtop) {
 951                                 /* this is the beginning of a chain */
 952                                 rxtop = skb;
 953                                 skb_fill_page_desc(rxtop, 0, buffer_info->page,
 954                                                    0, length);
 955                         } else {
 956                                 /* this is the middle of a chain */
 957                                 skb_fill_page_desc(rxtop,
 958                                     skb_shinfo(rxtop)->nr_frags,
 959                                     buffer_info->page, 0, length);
 960                                 /* re-use the skb, only consumed the page */
 961                                 buffer_info->skb = skb;
 962                         }
 963                         e1000_consume_page(buffer_info, rxtop, length);
 964                         goto next_desc;
 965                 } else {
 966                         if (rxtop) {
 967                                 /* end of the chain */
 968                                 skb_fill_page_desc(rxtop,
 969                                     skb_shinfo(rxtop)->nr_frags,
 970                                     buffer_info->page, 0, length);
 971                                 /* re-use the current skb, we only consumed the
 972                                  * page */
 973                                 buffer_info->skb = skb;
 974                                 skb = rxtop;
 975                                 rxtop = NULL;
 976                                 e1000_consume_page(buffer_info, skb, length);
 977                         } else {
 978                                 /* no chain, got EOP, this buf is the packet
 979                                  * copybreak to save the put_page/alloc_page */
 980                                 if (length <= copybreak &&
 981                                     skb_tailroom(skb) >= length) {
 982                                         u8 *vaddr;
 983                                         vaddr = kmap_atomic(buffer_info->page,
 984                                                            KM_SKB_DATA_SOFTIRQ);
 985                                         memcpy(skb_tail_pointer(skb), vaddr,
 986                                                length);
 987                                         kunmap_atomic(vaddr,
 988                                                       KM_SKB_DATA_SOFTIRQ);
 989                                         /* re-use the page, so don't erase
 990                                          * buffer_info->page */
 991                                         skb_put(skb, length);
 992                                 } else {
 993                                         skb_fill_page_desc(skb, 0,
 994                                                            buffer_info->page, 0,
 995                                                            length);
 996                                         e1000_consume_page(buffer_info, skb,
 997                                                            length);
 998                                 }
 999                         }
1000                 }
1001
1002                 /* Receive Checksum Offload XXX recompute due to CRC strip? */
1003                 e1000_rx_checksum(adapter,
1004                                   (u32)(status) |
1005                                   ((u32)(rx_desc->errors) << 24),
1006                                   le16_to_cpu(rx_desc->csum), skb);
1007
1008                 /* probably a little skewed due to removing CRC */
1009                 total_rx_bytes += skb->len;
1010                 total_rx_packets++;
1011
1012                 /* eth type trans needs skb->data to point to something */
1013                 if (!pskb_may_pull(skb, ETH_HLEN)) {
1014                         ndev_err(netdev, "pskb_may_pull failed.\n");
1015                         dev_kfree_skb(skb);
1016                         goto next_desc;
1017                 }
1018
1019                 e1000_receive_skb(adapter, netdev, skb, status,
1020                                   rx_desc->special);
1021
1022 next_desc:
1023                 rx_desc->status = 0;
1024
1025                 /* return some buffers to hardware, one at a time is too slow */
1026                 if (unlikely(cleaned_count >= E1000_RX_BUFFER_WRITE)) {
1027                         adapter->alloc_rx_buf(adapter, cleaned_count);
1028                         cleaned_count = 0;
1029                 }
1030
1031                 /* use prefetched values */
1032                 rx_desc = next_rxd;
1033                 buffer_info = next_buffer;
1034         }
1035         rx_ring->next_to_clean = i;
1036
1037         cleaned_count = e1000_desc_unused(rx_ring);
1038         if (cleaned_count)
1039                 adapter->alloc_rx_buf(adapter, cleaned_count);
1040
1041         adapter->total_rx_bytes += total_rx_bytes;
1042         adapter->total_rx_packets += total_rx_packets;
1043         adapter->net_stats.rx_bytes += total_rx_bytes;
1044         adapter->net_stats.rx_packets += total_rx_packets;
1045         return cleaned;
1046 }
1047
1048 /**
1049  * e1000_clean_rx_ring - Free Rx Buffers per Queue
1050  * @adapter: board private structure
1051  **/
1052 static void e1000_clean_rx_ring(struct e1000_adapter *adapter)
1053 {
1054         struct e1000_ring *rx_ring = adapter->rx_ring;
1055         struct e1000_buffer *buffer_info;
1056         struct e1000_ps_page *ps_page;
1057         struct pci_dev *pdev = adapter->pdev;
1058         unsigned int i, j;
1059
1060         /* Free all the Rx ring sk_buffs */
1061         for (i = 0; i < rx_ring->count; i++) {
1062                 buffer_info = &rx_ring->buffer_info[i];
1063                 if (buffer_info->dma) {
1064                         if (adapter->clean_rx == e1000_clean_rx_irq)
1065                                 pci_unmap_single(pdev, buffer_info->dma,
1066                                                  adapter->rx_buffer_len,
1067                                                  PCI_DMA_FROMDEVICE);
1068                         else if (adapter->clean_rx == e1000_clean_jumbo_rx_irq)
1069                                 pci_unmap_page(pdev, buffer_info->dma,
1070                                                PAGE_SIZE,
1071                                                PCI_DMA_FROMDEVICE);
1072                         else if (adapter->clean_rx == e1000_clean_rx_irq_ps)
1073                                 pci_unmap_single(pdev, buffer_info->dma,
1074                                                  adapter->rx_ps_bsize0,
1075                                                  PCI_DMA_FROMDEVICE);
1076                         buffer_info->dma = 0;
1077                 }
1078
1079                 if (buffer_info->page) {
1080                         put_page(buffer_info->page);
1081                         buffer_info->page = NULL;
1082                 }
1083
1084                 if (buffer_info->skb) {
1085                         dev_kfree_skb(buffer_info->skb);
1086                         buffer_info->skb = NULL;
1087                 }
1088
1089                 for (j = 0; j < PS_PAGE_BUFFERS; j++) {
1090                         ps_page = &buffer_info->ps_pages[j];
1091                         if (!ps_page->page)
1092                                 break;
1093                         pci_unmap_page(pdev, ps_page->dma, PAGE_SIZE,
1094                                        PCI_DMA_FROMDEVICE);
1095                         ps_page->dma = 0;
1096                         put_page(ps_page->page);
1097                         ps_page->page = NULL;
1098                 }
1099         }
1100
1101         /* there also may be some cached data from a chained receive */
1102         if (rx_ring->rx_skb_top) {
1103                 dev_kfree_skb(rx_ring->rx_skb_top);
1104                 rx_ring->rx_skb_top = NULL;
1105         }
1106
1107         /* Zero out the descriptor ring */
1108         memset(rx_ring->desc, 0, rx_ring->size);
1109
1110         rx_ring->next_to_clean = 0;
1111         rx_ring->next_to_use = 0;
1112
1113         writel(0, adapter->hw.hw_addr + rx_ring->head);
1114         writel(0, adapter->hw.hw_addr + rx_ring->tail);
1115 }
1116
1117 /**
1118  * e1000_intr_msi - Interrupt Handler
1119  * @irq: interrupt number
1120  * @data: pointer to a network interface device structure
1121  **/
1122 static irqreturn_t e1000_intr_msi(int irq, void *data)
1123 {
1124         struct net_device *netdev = data;
1125         struct e1000_adapter *adapter = netdev_priv(netdev);
1126         struct e1000_hw *hw = &adapter->hw;
1127         u32 icr = er32(ICR);
1128
1129         /*
1130          * read ICR disables interrupts using IAM
1131          */
1132
1133         if (icr & (E1000_ICR_RXSEQ | E1000_ICR_LSC)) {
1134                 hw->mac.get_link_status = 1;
1135                 /*
1136                  * ICH8 workaround-- Call gig speed drop workaround on cable
1137                  * disconnect (LSC) before accessing any PHY registers
1138                  */
1139                 if ((adapter->flags & FLAG_LSC_GIG_SPEED_DROP) &&
1140                     (!(er32(STATUS) & E1000_STATUS_LU)))
1141                         e1000e_gig_downshift_workaround_ich8lan(hw);
1142
1143                 /*
1144                  * 80003ES2LAN workaround-- For packet buffer work-around on
1145                  * link down event; disable receives here in the ISR and reset
1146                  * adapter in watchdog
1147                  */
1148                 if (netif_carrier_ok(netdev) &&
1149                     adapter->flags & FLAG_RX_NEEDS_RESTART) {
1150                         /* disable receives */
1151                         u32 rctl = er32(RCTL);
1152                         ew32(RCTL, rctl & ~E1000_RCTL_EN);
1153                         adapter->flags |= FLAG_RX_RESTART_NOW;
1154                 }
1155                 /* guard against interrupt when we're going down */
1156                 if (!test_bit(__E1000_DOWN, &adapter->state))
1157                         mod_timer(&adapter->watchdog_timer, jiffies + 1);
1158         }
1159
1160         if (netif_rx_schedule_prep(netdev, &adapter->napi)) {
1161                 adapter->total_tx_bytes = 0;
1162                 adapter->total_tx_packets = 0;
1163                 adapter->total_rx_bytes = 0;
1164                 adapter->total_rx_packets = 0;
1165                 __netif_rx_schedule(netdev, &adapter->napi);
1166         }
1167
1168         return IRQ_HANDLED;
1169 }
1170
1171 /**
1172  * e1000_intr - Interrupt Handler
1173  * @irq: interrupt number
1174  * @data: pointer to a network interface device structure
1175  **/
1176 static irqreturn_t e1000_intr(int irq, void *data)
1177 {
1178         struct net_device *netdev = data;
1179         struct e1000_adapter *adapter = netdev_priv(netdev);
1180         struct e1000_hw *hw = &adapter->hw;
1181
1182         u32 rctl, icr = er32(ICR);
1183         if (!icr)
1184                 return IRQ_NONE;  /* Not our interrupt */
1185
1186         /*
1187          * IMS will not auto-mask if INT_ASSERTED is not set, and if it is
1188          * not set, then the adapter didn't send an interrupt
1189          */
1190         if (!(icr & E1000_ICR_INT_ASSERTED))
1191                 return IRQ_NONE;
1192
1193         /*
1194          * Interrupt Auto-Mask...upon reading ICR,
1195          * interrupts are masked.  No need for the
1196          * IMC write
1197          */
1198
1199         if (icr & (E1000_ICR_RXSEQ | E1000_ICR_LSC)) {
1200                 hw->mac.get_link_status = 1;
1201                 /*
1202                  * ICH8 workaround-- Call gig speed drop workaround on cable
1203                  * disconnect (LSC) before accessing any PHY registers
1204                  */
1205                 if ((adapter->flags & FLAG_LSC_GIG_SPEED_DROP) &&
1206                     (!(er32(STATUS) & E1000_STATUS_LU)))
1207                         e1000e_gig_downshift_workaround_ich8lan(hw);
1208
1209                 /*
1210                  * 80003ES2LAN workaround--
1211                  * For packet buffer work-around on link down event;
1212                  * disable receives here in the ISR and
1213                  * reset adapter in watchdog
1214                  */
1215                 if (netif_carrier_ok(netdev) &&
1216                     (adapter->flags & FLAG_RX_NEEDS_RESTART)) {
1217                         /* disable receives */
1218                         rctl = er32(RCTL);
1219                         ew32(RCTL, rctl & ~E1000_RCTL_EN);
1220                         adapter->flags |= FLAG_RX_RESTART_NOW;
1221                 }
1222                 /* guard against interrupt when we're going down */
1223                 if (!test_bit(__E1000_DOWN, &adapter->state))
1224                         mod_timer(&adapter->watchdog_timer, jiffies + 1);
1225         }
1226
1227         if (netif_rx_schedule_prep(netdev, &adapter->napi)) {
1228                 adapter->total_tx_bytes = 0;
1229                 adapter->total_tx_packets = 0;
1230                 adapter->total_rx_bytes = 0;
1231                 adapter->total_rx_packets = 0;
1232                 __netif_rx_schedule(netdev, &adapter->napi);
1233         }
1234
1235         return IRQ_HANDLED;
1236 }
1237
1238 static int e1000_request_irq(struct e1000_adapter *adapter)
1239 {
1240         struct net_device *netdev = adapter->netdev;
1241         irq_handler_t handler = e1000_intr;
1242         int irq_flags = IRQF_SHARED;
1243         int err;
1244
1245         if (!pci_enable_msi(adapter->pdev)) {
1246                 adapter->flags |= FLAG_MSI_ENABLED;
1247                 handler = e1000_intr_msi;
1248                 irq_flags = 0;
1249         }
1250
1251         err = request_irq(adapter->pdev->irq, handler, irq_flags, netdev->name,
1252                           netdev);
1253         if (err) {
1254                 ndev_err(netdev,
1255                        "Unable to allocate %s interrupt (return: %d)\n",
1256                         adapter->flags & FLAG_MSI_ENABLED ? "MSI":"INTx",
1257                         err);
1258                 if (adapter->flags & FLAG_MSI_ENABLED)
1259                         pci_disable_msi(adapter->pdev);
1260         }
1261
1262         return err;
1263 }
1264
1265 static void e1000_free_irq(struct e1000_adapter *adapter)
1266 {
1267         struct net_device *netdev = adapter->netdev;
1268
1269         free_irq(adapter->pdev->irq, netdev);
1270         if (adapter->flags & FLAG_MSI_ENABLED) {
1271                 pci_disable_msi(adapter->pdev);
1272                 adapter->flags &= ~FLAG_MSI_ENABLED;
1273         }
1274 }
1275
1276 /**
1277  * e1000_irq_disable - Mask off interrupt generation on the NIC
1278  **/
1279 static void e1000_irq_disable(struct e1000_adapter *adapter)
1280 {
1281         struct e1000_hw *hw = &adapter->hw;
1282
1283         ew32(IMC, ~0);
1284         e1e_flush();
1285         synchronize_irq(adapter->pdev->irq);
1286 }
1287
1288 /**
1289  * e1000_irq_enable - Enable default interrupt generation settings
1290  **/
1291 static void e1000_irq_enable(struct e1000_adapter *adapter)
1292 {
1293         struct e1000_hw *hw = &adapter->hw;
1294
1295         ew32(IMS, IMS_ENABLE_MASK);
1296         e1e_flush();
1297 }
1298
1299 /**
1300  * e1000_get_hw_control - get control of the h/w from f/w
1301  * @adapter: address of board private structure
1302  *
1303  * e1000_get_hw_control sets {CTRL_EXT|SWSM}:DRV_LOAD bit.
1304  * For ASF and Pass Through versions of f/w this means that
1305  * the driver is loaded. For AMT version (only with 82573)
1306  * of the f/w this means that the network i/f is open.
1307  **/
1308 static void e1000_get_hw_control(struct e1000_adapter *adapter)
1309 {
1310         struct e1000_hw *hw = &adapter->hw;
1311         u32 ctrl_ext;
1312         u32 swsm;
1313
1314         /* Let firmware know the driver has taken over */
1315         if (adapter->flags & FLAG_HAS_SWSM_ON_LOAD) {
1316                 swsm = er32(SWSM);
1317                 ew32(SWSM, swsm | E1000_SWSM_DRV_LOAD);
1318         } else if (adapter->flags & FLAG_HAS_CTRLEXT_ON_LOAD) {
1319                 ctrl_ext = er32(CTRL_EXT);
1320                 ew32(CTRL_EXT, ctrl_ext | E1000_CTRL_EXT_DRV_LOAD);
1321         }
1322 }
1323
1324 /**
1325  * e1000_release_hw_control - release control of the h/w to f/w
1326  * @adapter: address of board private structure
1327  *
1328  * e1000_release_hw_control resets {CTRL_EXT|SWSM}:DRV_LOAD bit.
1329  * For ASF and Pass Through versions of f/w this means that the
1330  * driver is no longer loaded. For AMT version (only with 82573) i
1331  * of the f/w this means that the network i/f is closed.
1332  *
1333  **/
1334 static void e1000_release_hw_control(struct e1000_adapter *adapter)
1335 {
1336         struct e1000_hw *hw = &adapter->hw;
1337         u32 ctrl_ext;
1338         u32 swsm;
1339
1340         /* Let firmware taken over control of h/w */
1341         if (adapter->flags & FLAG_HAS_SWSM_ON_LOAD) {
1342                 swsm = er32(SWSM);
1343                 ew32(SWSM, swsm & ~E1000_SWSM_DRV_LOAD);
1344         } else if (adapter->flags & FLAG_HAS_CTRLEXT_ON_LOAD) {
1345                 ctrl_ext = er32(CTRL_EXT);
1346                 ew32(CTRL_EXT, ctrl_ext & ~E1000_CTRL_EXT_DRV_LOAD);
1347         }
1348 }
1349
1350 /**
1351  * @e1000_alloc_ring - allocate memory for a ring structure
1352  **/
1353 static int e1000_alloc_ring_dma(struct e1000_adapter *adapter,
1354                                 struct e1000_ring *ring)
1355 {
1356         struct pci_dev *pdev = adapter->pdev;
1357
1358         ring->desc = dma_alloc_coherent(&pdev->dev, ring->size, &ring->dma,
1359                                         GFP_KERNEL);
1360         if (!ring->desc)
1361                 return -ENOMEM;
1362
1363         return 0;
1364 }
1365
1366 /**
1367  * e1000e_setup_tx_resources - allocate Tx resources (Descriptors)
1368  * @adapter: board private structure
1369  *
1370  * Return 0 on success, negative on failure
1371  **/
1372 int e1000e_setup_tx_resources(struct e1000_adapter *adapter)
1373 {
1374         struct e1000_ring *tx_ring = adapter->tx_ring;
1375         int err = -ENOMEM, size;
1376
1377         size = sizeof(struct e1000_buffer) * tx_ring->count;
1378         tx_ring->buffer_info = vmalloc(size);
1379         if (!tx_ring->buffer_info)
1380                 goto err;
1381         memset(tx_ring->buffer_info, 0, size);
1382
1383         /* round up to nearest 4K */
1384         tx_ring->size = tx_ring->count * sizeof(struct e1000_tx_desc);
1385         tx_ring->size = ALIGN(tx_ring->size, 4096);
1386
1387         err = e1000_alloc_ring_dma(adapter, tx_ring);
1388         if (err)
1389                 goto err;
1390
1391         tx_ring->next_to_use = 0;
1392         tx_ring->next_to_clean = 0;
1393         spin_lock_init(&adapter->tx_queue_lock);
1394
1395         return 0;
1396 err:
1397         vfree(tx_ring->buffer_info);
1398         ndev_err(adapter->netdev,
1399         "Unable to allocate memory for the transmit descriptor ring\n");
1400         return err;
1401 }
1402
1403 /**
1404  * e1000e_setup_rx_resources - allocate Rx resources (Descriptors)
1405  * @adapter: board private structure
1406  *
1407  * Returns 0 on success, negative on failure
1408  **/
1409 int e1000e_setup_rx_resources(struct e1000_adapter *adapter)
1410 {
1411         struct e1000_ring *rx_ring = adapter->rx_ring;
1412         struct e1000_buffer *buffer_info;
1413         int i, size, desc_len, err = -ENOMEM;
1414
1415         size = sizeof(struct e1000_buffer) * rx_ring->count;
1416         rx_ring->buffer_info = vmalloc(size);
1417         if (!rx_ring->buffer_info)
1418                 goto err;
1419         memset(rx_ring->buffer_info, 0, size);
1420
1421         for (i = 0; i < rx_ring->count; i++) {
1422                 buffer_info = &rx_ring->buffer_info[i];
1423                 buffer_info->ps_pages = kcalloc(PS_PAGE_BUFFERS,
1424                                                 sizeof(struct e1000_ps_page),
1425                                                 GFP_KERNEL);
1426                 if (!buffer_info->ps_pages)
1427                         goto err_pages;
1428         }
1429
1430         desc_len = sizeof(union e1000_rx_desc_packet_split);
1431
1432         /* Round up to nearest 4K */
1433         rx_ring->size = rx_ring->count * desc_len;
1434         rx_ring->size = ALIGN(rx_ring->size, 4096);
1435
1436         err = e1000_alloc_ring_dma(adapter, rx_ring);
1437         if (err)
1438                 goto err_pages;
1439
1440         rx_ring->next_to_clean = 0;
1441         rx_ring->next_to_use = 0;
1442         rx_ring->rx_skb_top = NULL;
1443
1444         return 0;
1445
1446 err_pages:
1447         for (i = 0; i < rx_ring->count; i++) {
1448                 buffer_info = &rx_ring->buffer_info[i];
1449                 kfree(buffer_info->ps_pages);
1450         }
1451 err:
1452         vfree(rx_ring->buffer_info);
1453         ndev_err(adapter->netdev,
1454         "Unable to allocate memory for the transmit descriptor ring\n");
1455         return err;
1456 }
1457
1458 /**
1459  * e1000_clean_tx_ring - Free Tx Buffers
1460  * @adapter: board private structure
1461  **/
1462 static void e1000_clean_tx_ring(struct e1000_adapter *adapter)
1463 {
1464         struct e1000_ring *tx_ring = adapter->tx_ring;
1465         struct e1000_buffer *buffer_info;
1466         unsigned long size;
1467         unsigned int i;
1468
1469         for (i = 0; i < tx_ring->count; i++) {
1470                 buffer_info = &tx_ring->buffer_info[i];
1471                 e1000_put_txbuf(adapter, buffer_info);
1472         }
1473
1474         size = sizeof(struct e1000_buffer) * tx_ring->count;
1475         memset(tx_ring->buffer_info, 0, size);
1476
1477         memset(tx_ring->desc, 0, tx_ring->size);
1478
1479         tx_ring->next_to_use = 0;
1480         tx_ring->next_to_clean = 0;
1481
1482         writel(0, adapter->hw.hw_addr + tx_ring->head);
1483         writel(0, adapter->hw.hw_addr + tx_ring->tail);
1484 }
1485
1486 /**
1487  * e1000e_free_tx_resources - Free Tx Resources per Queue
1488  * @adapter: board private structure
1489  *
1490  * Free all transmit software resources
1491  **/
1492 void e1000e_free_tx_resources(struct e1000_adapter *adapter)
1493 {
1494         struct pci_dev *pdev = adapter->pdev;
1495         struct e1000_ring *tx_ring = adapter->tx_ring;
1496
1497         e1000_clean_tx_ring(adapter);
1498
1499         vfree(tx_ring->buffer_info);
1500         tx_ring->buffer_info = NULL;
1501
1502         dma_free_coherent(&pdev->dev, tx_ring->size, tx_ring->desc,
1503                           tx_ring->dma);
1504         tx_ring->desc = NULL;
1505 }
1506
1507 /**
1508  * e1000e_free_rx_resources - Free Rx Resources
1509  * @adapter: board private structure
1510  *
1511  * Free all receive software resources
1512  **/
1513
1514 void e1000e_free_rx_resources(struct e1000_adapter *adapter)
1515 {
1516         struct pci_dev *pdev = adapter->pdev;
1517         struct e1000_ring *rx_ring = adapter->rx_ring;
1518         int i;
1519
1520         e1000_clean_rx_ring(adapter);
1521
1522         for (i = 0; i < rx_ring->count; i++) {
1523                 kfree(rx_ring->buffer_info[i].ps_pages);
1524         }
1525
1526         vfree(rx_ring->buffer_info);
1527         rx_ring->buffer_info = NULL;
1528
1529         dma_free_coherent(&pdev->dev, rx_ring->size, rx_ring->desc,
1530                           rx_ring->dma);
1531         rx_ring->desc = NULL;
1532 }
1533
1534 /**
1535  * e1000_update_itr - update the dynamic ITR value based on statistics
1536  * @adapter: pointer to adapter
1537  * @itr_setting: current adapter->itr
1538  * @packets: the number of packets during this measurement interval
1539  * @bytes: the number of bytes during this measurement interval
1540  *
1541  *      Stores a new ITR value based on packets and byte
1542  *      counts during the last interrupt.  The advantage of per interrupt
1543  *      computation is faster updates and more accurate ITR for the current
1544  *      traffic pattern.  Constants in this function were computed
1545  *      based on theoretical maximum wire speed and thresholds were set based
1546  *      on testing data as well as attempting to minimize response time
1547  *      while increasing bulk throughput.
1548  *      this functionality is controlled by the InterruptThrottleRate module
1549  *      parameter (see e1000_param.c)
1550  **/
1551 static unsigned int e1000_update_itr(struct e1000_adapter *adapter,
1552                                      u16 itr_setting, int packets,
1553                                      int bytes)
1554 {
1555         unsigned int retval = itr_setting;
1556
1557         if (packets == 0)
1558                 goto update_itr_done;
1559
1560         switch (itr_setting) {
1561         case lowest_latency:
1562                 /* handle TSO and jumbo frames */
1563                 if (bytes/packets > 8000)
1564                         retval = bulk_latency;
1565                 else if ((packets < 5) && (bytes > 512)) {
1566                         retval = low_latency;
1567                 }
1568                 break;
1569         case low_latency:  /* 50 usec aka 20000 ints/s */
1570                 if (bytes > 10000) {
1571                         /* this if handles the TSO accounting */
1572                         if (bytes/packets > 8000) {
1573                                 retval = bulk_latency;
1574                         } else if ((packets < 10) || ((bytes/packets) > 1200)) {
1575                                 retval = bulk_latency;
1576                         } else if ((packets > 35)) {
1577                                 retval = lowest_latency;
1578                         }
1579                 } else if (bytes/packets > 2000) {
1580                         retval = bulk_latency;
1581                 } else if (packets <= 2 && bytes < 512) {
1582                         retval = lowest_latency;
1583                 }
1584                 break;
1585         case bulk_latency: /* 250 usec aka 4000 ints/s */
1586                 if (bytes > 25000) {
1587                         if (packets > 35) {
1588                                 retval = low_latency;
1589                         }
1590                 } else if (bytes < 6000) {
1591                         retval = low_latency;
1592                 }
1593                 break;
1594         }
1595
1596 update_itr_done:
1597         return retval;
1598 }
1599
1600 static void e1000_set_itr(struct e1000_adapter *adapter)
1601 {
1602         struct e1000_hw *hw = &adapter->hw;
1603         u16 current_itr;
1604         u32 new_itr = adapter->itr;
1605
1606         /* for non-gigabit speeds, just fix the interrupt rate at 4000 */
1607         if (adapter->link_speed != SPEED_1000) {
1608                 current_itr = 0;
1609                 new_itr = 4000;
1610                 goto set_itr_now;
1611         }
1612
1613         adapter->tx_itr = e1000_update_itr(adapter,
1614                                     adapter->tx_itr,
1615                                     adapter->total_tx_packets,
1616                                     adapter->total_tx_bytes);
1617         /* conservative mode (itr 3) eliminates the lowest_latency setting */
1618         if (adapter->itr_setting == 3 && adapter->tx_itr == lowest_latency)
1619                 adapter->tx_itr = low_latency;
1620
1621         adapter->rx_itr = e1000_update_itr(adapter,
1622                                     adapter->rx_itr,
1623                                     adapter->total_rx_packets,
1624                                     adapter->total_rx_bytes);
1625         /* conservative mode (itr 3) eliminates the lowest_latency setting */
1626         if (adapter->itr_setting == 3 && adapter->rx_itr == lowest_latency)
1627                 adapter->rx_itr = low_latency;
1628
1629         current_itr = max(adapter->rx_itr, adapter->tx_itr);
1630
1631         switch (current_itr) {
1632         /* counts and packets in update_itr are dependent on these numbers */
1633         case lowest_latency:
1634                 new_itr = 70000;
1635                 break;
1636         case low_latency:
1637                 new_itr = 20000; /* aka hwitr = ~200 */
1638                 break;
1639         case bulk_latency:
1640                 new_itr = 4000;
1641                 break;
1642         default:
1643                 break;
1644         }
1645
1646 set_itr_now:
1647         if (new_itr != adapter->itr) {
1648                 /*
1649                  * this attempts to bias the interrupt rate towards Bulk
1650                  * by adding intermediate steps when interrupt rate is
1651                  * increasing
1652                  */
1653                 new_itr = new_itr > adapter->itr ?
1654                              min(adapter->itr + (new_itr >> 2), new_itr) :
1655                              new_itr;
1656                 adapter->itr = new_itr;
1657                 ew32(ITR, 1000000000 / (new_itr * 256));
1658         }
1659 }
1660
1661 /**
1662  * e1000_clean - NAPI Rx polling callback
1663  * @napi: struct associated with this polling callback
1664  * @budget: amount of packets driver is allowed to process this poll
1665  **/
1666 static int e1000_clean(struct napi_struct *napi, int budget)
1667 {
1668         struct e1000_adapter *adapter = container_of(napi, struct e1000_adapter, napi);
1669         struct net_device *poll_dev = adapter->netdev;
1670         int tx_cleaned = 0, work_done = 0;
1671
1672         /* Must NOT use netdev_priv macro here. */
1673         adapter = poll_dev->priv;
1674
1675         /*
1676          * e1000_clean is called per-cpu.  This lock protects
1677          * tx_ring from being cleaned by multiple cpus
1678          * simultaneously.  A failure obtaining the lock means
1679          * tx_ring is currently being cleaned anyway.
1680          */
1681         if (spin_trylock(&adapter->tx_queue_lock)) {
1682                 tx_cleaned = e1000_clean_tx_irq(adapter);
1683                 spin_unlock(&adapter->tx_queue_lock);
1684         }
1685
1686         adapter->clean_rx(adapter, &work_done, budget);
1687
1688         if (tx_cleaned)
1689                 work_done = budget;
1690
1691         /* If budget not fully consumed, exit the polling mode */
1692         if (work_done < budget) {
1693                 if (adapter->itr_setting & 3)
1694                         e1000_set_itr(adapter);
1695                 netif_rx_complete(poll_dev, napi);
1696                 e1000_irq_enable(adapter);
1697         }
1698
1699         return work_done;
1700 }
1701
1702 static void e1000_vlan_rx_add_vid(struct net_device *netdev, u16 vid)
1703 {
1704         struct e1000_adapter *adapter = netdev_priv(netdev);
1705         struct e1000_hw *hw = &adapter->hw;
1706         u32 vfta, index;
1707
1708         /* don't update vlan cookie if already programmed */
1709         if ((adapter->hw.mng_cookie.status &
1710              E1000_MNG_DHCP_COOKIE_STATUS_VLAN) &&
1711             (vid == adapter->mng_vlan_id))
1712                 return;
1713         /* add VID to filter table */
1714         index = (vid >> 5) & 0x7F;
1715         vfta = E1000_READ_REG_ARRAY(hw, E1000_VFTA, index);
1716         vfta |= (1 << (vid & 0x1F));
1717         e1000e_write_vfta(hw, index, vfta);
1718 }
1719
1720 static void e1000_vlan_rx_kill_vid(struct net_device *netdev, u16 vid)
1721 {
1722         struct e1000_adapter *adapter = netdev_priv(netdev);
1723         struct e1000_hw *hw = &adapter->hw;
1724         u32 vfta, index;
1725
1726         if (!test_bit(__E1000_DOWN, &adapter->state))
1727                 e1000_irq_disable(adapter);
1728         vlan_group_set_device(adapter->vlgrp, vid, NULL);
1729
1730         if (!test_bit(__E1000_DOWN, &adapter->state))
1731                 e1000_irq_enable(adapter);
1732
1733         if ((adapter->hw.mng_cookie.status &
1734              E1000_MNG_DHCP_COOKIE_STATUS_VLAN) &&
1735             (vid == adapter->mng_vlan_id)) {
1736                 /* release control to f/w */
1737                 e1000_release_hw_control(adapter);
1738                 return;
1739         }
1740
1741         /* remove VID from filter table */
1742         index = (vid >> 5) & 0x7F;
1743         vfta = E1000_READ_REG_ARRAY(hw, E1000_VFTA, index);
1744         vfta &= ~(1 << (vid & 0x1F));
1745         e1000e_write_vfta(hw, index, vfta);
1746 }
1747
1748 static void e1000_update_mng_vlan(struct e1000_adapter *adapter)
1749 {
1750         struct net_device *netdev = adapter->netdev;
1751         u16 vid = adapter->hw.mng_cookie.vlan_id;
1752         u16 old_vid = adapter->mng_vlan_id;
1753
1754         if (!adapter->vlgrp)
1755                 return;
1756
1757         if (!vlan_group_get_device(adapter->vlgrp, vid)) {
1758                 adapter->mng_vlan_id = E1000_MNG_VLAN_NONE;
1759                 if (adapter->hw.mng_cookie.status &
1760                         E1000_MNG_DHCP_COOKIE_STATUS_VLAN) {
1761                         e1000_vlan_rx_add_vid(netdev, vid);
1762                         adapter->mng_vlan_id = vid;
1763                 }
1764
1765                 if ((old_vid != (u16)E1000_MNG_VLAN_NONE) &&
1766                                 (vid != old_vid) &&
1767                     !vlan_group_get_device(adapter->vlgrp, old_vid))
1768                         e1000_vlan_rx_kill_vid(netdev, old_vid);
1769         } else {
1770                 adapter->mng_vlan_id = vid;
1771         }
1772 }
1773
1774
1775 static void e1000_vlan_rx_register(struct net_device *netdev,
1776                                    struct vlan_group *grp)
1777 {
1778         struct e1000_adapter *adapter = netdev_priv(netdev);
1779         struct e1000_hw *hw = &adapter->hw;
1780         u32 ctrl, rctl;
1781
1782         if (!test_bit(__E1000_DOWN, &adapter->state))
1783                 e1000_irq_disable(adapter);
1784         adapter->vlgrp = grp;
1785
1786         if (grp) {
1787                 /* enable VLAN tag insert/strip */
1788                 ctrl = er32(CTRL);
1789                 ctrl |= E1000_CTRL_VME;
1790                 ew32(CTRL, ctrl);
1791
1792                 if (adapter->flags & FLAG_HAS_HW_VLAN_FILTER) {
1793                         /* enable VLAN receive filtering */
1794                         rctl = er32(RCTL);
1795                         if (!(netdev->flags & IFF_PROMISC))
1796                                 rctl |= E1000_RCTL_VFE;
1797                         rctl &= ~E1000_RCTL_CFIEN;
1798                         ew32(RCTL, rctl);
1799                         e1000_update_mng_vlan(adapter);
1800                 }
1801         } else {
1802                 /* disable VLAN tag insert/strip */
1803                 ctrl = er32(CTRL);
1804                 ctrl &= ~E1000_CTRL_VME;
1805                 ew32(CTRL, ctrl);
1806
1807                 if (adapter->flags & FLAG_HAS_HW_VLAN_FILTER) {
1808                         /* disable VLAN filtering */
1809                         rctl = er32(RCTL);
1810                         rctl &= ~E1000_RCTL_VFE;
1811                         ew32(RCTL, rctl);
1812                         if (adapter->mng_vlan_id !=
1813                             (u16)E1000_MNG_VLAN_NONE) {
1814                                 e1000_vlan_rx_kill_vid(netdev,
1815                                                        adapter->mng_vlan_id);
1816                                 adapter->mng_vlan_id = E1000_MNG_VLAN_NONE;
1817                         }
1818                 }
1819         }
1820
1821         if (!test_bit(__E1000_DOWN, &adapter->state))
1822                 e1000_irq_enable(adapter);
1823 }
1824
1825 static void e1000_restore_vlan(struct e1000_adapter *adapter)
1826 {
1827         u16 vid;
1828
1829         e1000_vlan_rx_register(adapter->netdev, adapter->vlgrp);
1830
1831         if (!adapter->vlgrp)
1832                 return;
1833
1834         for (vid = 0; vid < VLAN_GROUP_ARRAY_LEN; vid++) {
1835                 if (!vlan_group_get_device(adapter->vlgrp, vid))
1836                         continue;
1837                 e1000_vlan_rx_add_vid(adapter->netdev, vid);
1838         }
1839 }
1840
1841 static void e1000_init_manageability(struct e1000_adapter *adapter)
1842 {
1843         struct e1000_hw *hw = &adapter->hw;
1844         u32 manc, manc2h;
1845
1846         if (!(adapter->flags & FLAG_MNG_PT_ENABLED))
1847                 return;
1848
1849         manc = er32(MANC);
1850
1851         /*
1852          * enable receiving management packets to the host. this will probably
1853          * generate destination unreachable messages from the host OS, but
1854          * the packets will be handled on SMBUS
1855          */
1856         manc |= E1000_MANC_EN_MNG2HOST;
1857         manc2h = er32(MANC2H);
1858 #define E1000_MNG2HOST_PORT_623 (1 << 5)
1859 #define E1000_MNG2HOST_PORT_664 (1 << 6)
1860         manc2h |= E1000_MNG2HOST_PORT_623;
1861         manc2h |= E1000_MNG2HOST_PORT_664;
1862         ew32(MANC2H, manc2h);
1863         ew32(MANC, manc);
1864 }
1865
1866 /**
1867  * e1000_configure_tx - Configure 8254x Transmit Unit after Reset
1868  * @adapter: board private structure
1869  *
1870  * Configure the Tx unit of the MAC after a reset.
1871  **/
1872 static void e1000_configure_tx(struct e1000_adapter *adapter)
1873 {
1874         struct e1000_hw *hw = &adapter->hw;
1875         struct e1000_ring *tx_ring = adapter->tx_ring;
1876         u64 tdba;
1877         u32 tdlen, tctl, tipg, tarc;
1878         u32 ipgr1, ipgr2;
1879
1880         /* Setup the HW Tx Head and Tail descriptor pointers */
1881         tdba = tx_ring->dma;
1882         tdlen = tx_ring->count * sizeof(struct e1000_tx_desc);
1883         ew32(TDBAL, (tdba & DMA_32BIT_MASK));
1884         ew32(TDBAH, (tdba >> 32));
1885         ew32(TDLEN, tdlen);
1886         ew32(TDH, 0);
1887         ew32(TDT, 0);
1888         tx_ring->head = E1000_TDH;
1889         tx_ring->tail = E1000_TDT;
1890
1891         /* Set the default values for the Tx Inter Packet Gap timer */
1892         tipg = DEFAULT_82543_TIPG_IPGT_COPPER;          /*  8  */
1893         ipgr1 = DEFAULT_82543_TIPG_IPGR1;               /*  8  */
1894         ipgr2 = DEFAULT_82543_TIPG_IPGR2;               /*  6  */
1895
1896         if (adapter->flags & FLAG_TIPG_MEDIUM_FOR_80003ESLAN)
1897                 ipgr2 = DEFAULT_80003ES2LAN_TIPG_IPGR2; /*  7  */
1898
1899         tipg |= ipgr1 << E1000_TIPG_IPGR1_SHIFT;
1900         tipg |= ipgr2 << E1000_TIPG_IPGR2_SHIFT;
1901         ew32(TIPG, tipg);
1902
1903         /* Set the Tx Interrupt Delay register */
1904         ew32(TIDV, adapter->tx_int_delay);
1905         /* Tx irq moderation */
1906         ew32(TADV, adapter->tx_abs_int_delay);
1907
1908         /* Program the Transmit Control Register */
1909         tctl = er32(TCTL);
1910         tctl &= ~E1000_TCTL_CT;
1911         tctl |= E1000_TCTL_PSP | E1000_TCTL_RTLC |
1912                 (E1000_COLLISION_THRESHOLD << E1000_CT_SHIFT);
1913
1914         if (adapter->flags & FLAG_TARC_SPEED_MODE_BIT) {
1915                 tarc = er32(TARC(0));
1916                 /*
1917                  * set the speed mode bit, we'll clear it if we're not at
1918                  * gigabit link later
1919                  */
1920 #define SPEED_MODE_BIT (1 << 21)
1921                 tarc |= SPEED_MODE_BIT;
1922                 ew32(TARC(0), tarc);
1923         }
1924
1925         /* errata: program both queues to unweighted RR */
1926         if (adapter->flags & FLAG_TARC_SET_BIT_ZERO) {
1927                 tarc = er32(TARC(0));
1928                 tarc |= 1;
1929                 ew32(TARC(0), tarc);
1930                 tarc = er32(TARC(1));
1931                 tarc |= 1;
1932                 ew32(TARC(1), tarc);
1933         }
1934
1935         e1000e_config_collision_dist(hw);
1936
1937         /* Setup Transmit Descriptor Settings for eop descriptor */
1938         adapter->txd_cmd = E1000_TXD_CMD_EOP | E1000_TXD_CMD_IFCS;
1939
1940         /* only set IDE if we are delaying interrupts using the timers */
1941         if (adapter->tx_int_delay)
1942                 adapter->txd_cmd |= E1000_TXD_CMD_IDE;
1943
1944         /* enable Report Status bit */
1945         adapter->txd_cmd |= E1000_TXD_CMD_RS;
1946
1947         ew32(TCTL, tctl);
1948
1949         adapter->tx_queue_len = adapter->netdev->tx_queue_len;
1950 }
1951
1952 /**
1953  * e1000_setup_rctl - configure the receive control registers
1954  * @adapter: Board private structure
1955  **/
1956 #define PAGE_USE_COUNT(S) (((S) >> PAGE_SHIFT) + \
1957                            (((S) & (PAGE_SIZE - 1)) ? 1 : 0))
1958 static void e1000_setup_rctl(struct e1000_adapter *adapter)
1959 {
1960         struct e1000_hw *hw = &adapter->hw;
1961         u32 rctl, rfctl;
1962         u32 psrctl = 0;
1963         u32 pages = 0;
1964
1965         /* Program MC offset vector base */
1966         rctl = er32(RCTL);
1967         rctl &= ~(3 << E1000_RCTL_MO_SHIFT);
1968         rctl |= E1000_RCTL_EN | E1000_RCTL_BAM |
1969                 E1000_RCTL_LBM_NO | E1000_RCTL_RDMTS_HALF |
1970                 (adapter->hw.mac.mc_filter_type << E1000_RCTL_MO_SHIFT);
1971
1972         /* Do not Store bad packets */
1973         rctl &= ~E1000_RCTL_SBP;
1974
1975         /* Enable Long Packet receive */
1976         if (adapter->netdev->mtu <= ETH_DATA_LEN)
1977                 rctl &= ~E1000_RCTL_LPE;
1978         else
1979                 rctl |= E1000_RCTL_LPE;
1980
1981         /* Enable hardware CRC frame stripping */
1982         rctl |= E1000_RCTL_SECRC;
1983
1984         /* Setup buffer sizes */
1985         rctl &= ~E1000_RCTL_SZ_4096;
1986         rctl |= E1000_RCTL_BSEX;
1987         switch (adapter->rx_buffer_len) {
1988         case 256:
1989                 rctl |= E1000_RCTL_SZ_256;
1990                 rctl &= ~E1000_RCTL_BSEX;
1991                 break;
1992         case 512:
1993                 rctl |= E1000_RCTL_SZ_512;
1994                 rctl &= ~E1000_RCTL_BSEX;
1995                 break;
1996         case 1024:
1997                 rctl |= E1000_RCTL_SZ_1024;
1998                 rctl &= ~E1000_RCTL_BSEX;
1999                 break;
2000         case 2048:
2001         default:
2002                 rctl |= E1000_RCTL_SZ_2048;
2003                 rctl &= ~E1000_RCTL_BSEX;
2004                 break;
2005         case 4096:
2006                 rctl |= E1000_RCTL_SZ_4096;
2007                 break;
2008         case 8192:
2009                 rctl |= E1000_RCTL_SZ_8192;
2010                 break;
2011         case 16384:
2012                 rctl |= E1000_RCTL_SZ_16384;
2013                 break;
2014         }
2015
2016         /*
2017          * 82571 and greater support packet-split where the protocol
2018          * header is placed in skb->data and the packet data is
2019          * placed in pages hanging off of skb_shinfo(skb)->nr_frags.
2020          * In the case of a non-split, skb->data is linearly filled,
2021          * followed by the page buffers.  Therefore, skb->data is
2022          * sized to hold the largest protocol header.
2023          *
2024          * allocations using alloc_page take too long for regular MTU
2025          * so only enable packet split for jumbo frames
2026          *
2027          * Using pages when the page size is greater than 16k wastes
2028          * a lot of memory, since we allocate 3 pages at all times
2029          * per packet.
2030          */
2031         pages = PAGE_USE_COUNT(adapter->netdev->mtu);
2032         if (!(adapter->flags & FLAG_IS_ICH) && (pages <= 3) &&
2033             (PAGE_SIZE <= 16384) && (rctl & E1000_RCTL_LPE))
2034                 adapter->rx_ps_pages = pages;
2035         else
2036                 adapter->rx_ps_pages = 0;
2037
2038         if (adapter->rx_ps_pages) {
2039                 /* Configure extra packet-split registers */
2040                 rfctl = er32(RFCTL);
2041                 rfctl |= E1000_RFCTL_EXTEN;
2042                 /*
2043                  * disable packet split support for IPv6 extension headers,
2044                  * because some malformed IPv6 headers can hang the Rx
2045                  */
2046                 rfctl |= (E1000_RFCTL_IPV6_EX_DIS |
2047                           E1000_RFCTL_NEW_IPV6_EXT_DIS);
2048
2049                 ew32(RFCTL, rfctl);
2050
2051                 /* Enable Packet split descriptors */
2052                 rctl |= E1000_RCTL_DTYP_PS;
2053
2054                 psrctl |= adapter->rx_ps_bsize0 >>
2055                         E1000_PSRCTL_BSIZE0_SHIFT;
2056
2057                 switch (adapter->rx_ps_pages) {
2058                 case 3:
2059                         psrctl |= PAGE_SIZE <<
2060                                 E1000_PSRCTL_BSIZE3_SHIFT;
2061                 case 2:
2062                         psrctl |= PAGE_SIZE <<
2063                                 E1000_PSRCTL_BSIZE2_SHIFT;
2064                 case 1:
2065                         psrctl |= PAGE_SIZE >>
2066                                 E1000_PSRCTL_BSIZE1_SHIFT;
2067                         break;
2068                 }
2069
2070                 ew32(PSRCTL, psrctl);
2071         }
2072
2073         ew32(RCTL, rctl);
2074         /* just started the receive unit, no need to restart */
2075         adapter->flags &= ~FLAG_RX_RESTART_NOW;
2076 }
2077
2078 /**
2079  * e1000_configure_rx - Configure Receive Unit after Reset
2080  * @adapter: board private structure
2081  *
2082  * Configure the Rx unit of the MAC after a reset.
2083  **/
2084 static void e1000_configure_rx(struct e1000_adapter *adapter)
2085 {
2086         struct e1000_hw *hw = &adapter->hw;
2087         struct e1000_ring *rx_ring = adapter->rx_ring;
2088         u64 rdba;
2089         u32 rdlen, rctl, rxcsum, ctrl_ext;
2090
2091         if (adapter->rx_ps_pages) {
2092                 /* this is a 32 byte descriptor */
2093                 rdlen = rx_ring->count *
2094                         sizeof(union e1000_rx_desc_packet_split);
2095                 adapter->clean_rx = e1000_clean_rx_irq_ps;
2096                 adapter->alloc_rx_buf = e1000_alloc_rx_buffers_ps;
2097         } else if (adapter->netdev->mtu > ETH_FRAME_LEN + ETH_FCS_LEN) {
2098                 rdlen = rx_ring->count * sizeof(struct e1000_rx_desc);
2099                 adapter->clean_rx = e1000_clean_jumbo_rx_irq;
2100                 adapter->alloc_rx_buf = e1000_alloc_jumbo_rx_buffers;
2101         } else {
2102                 rdlen = rx_ring->count * sizeof(struct e1000_rx_desc);
2103                 adapter->clean_rx = e1000_clean_rx_irq;
2104                 adapter->alloc_rx_buf = e1000_alloc_rx_buffers;
2105         }
2106
2107         /* disable receives while setting up the descriptors */
2108         rctl = er32(RCTL);
2109         ew32(RCTL, rctl & ~E1000_RCTL_EN);
2110         e1e_flush();
2111         msleep(10);
2112
2113         /* set the Receive Delay Timer Register */
2114         ew32(RDTR, adapter->rx_int_delay);
2115
2116         /* irq moderation */
2117         ew32(RADV, adapter->rx_abs_int_delay);
2118         if (adapter->itr_setting != 0)
2119                 ew32(ITR, 1000000000 / (adapter->itr * 256));
2120
2121         ctrl_ext = er32(CTRL_EXT);
2122         /* Reset delay timers after every interrupt */
2123         ctrl_ext |= E1000_CTRL_EXT_INT_TIMER_CLR;
2124         /* Auto-Mask interrupts upon ICR access */
2125         ctrl_ext |= E1000_CTRL_EXT_IAME;
2126         ew32(IAM, 0xffffffff);
2127         ew32(CTRL_EXT, ctrl_ext);
2128         e1e_flush();
2129
2130         /*
2131          * Setup the HW Rx Head and Tail Descriptor Pointers and
2132          * the Base and Length of the Rx Descriptor Ring
2133          */
2134         rdba = rx_ring->dma;
2135         ew32(RDBAL, (rdba & DMA_32BIT_MASK));
2136         ew32(RDBAH, (rdba >> 32));
2137         ew32(RDLEN, rdlen);
2138         ew32(RDH, 0);
2139         ew32(RDT, 0);
2140         rx_ring->head = E1000_RDH;
2141         rx_ring->tail = E1000_RDT;
2142
2143         /* Enable Receive Checksum Offload for TCP and UDP */
2144         rxcsum = er32(RXCSUM);
2145         if (adapter->flags & FLAG_RX_CSUM_ENABLED) {
2146                 rxcsum |= E1000_RXCSUM_TUOFL;
2147
2148                 /*
2149                  * IPv4 payload checksum for UDP fragments must be
2150                  * used in conjunction with packet-split.
2151                  */
2152                 if (adapter->rx_ps_pages)
2153                         rxcsum |= E1000_RXCSUM_IPPCSE;
2154         } else {
2155                 rxcsum &= ~E1000_RXCSUM_TUOFL;
2156                 /* no need to clear IPPCSE as it defaults to 0 */
2157         }
2158         ew32(RXCSUM, rxcsum);
2159
2160         /*
2161          * Enable early receives on supported devices, only takes effect when
2162          * packet size is equal or larger than the specified value (in 8 byte
2163          * units), e.g. using jumbo frames when setting to E1000_ERT_2048
2164          */
2165         if ((adapter->flags & FLAG_HAS_ERT) &&
2166             (adapter->netdev->mtu > ETH_DATA_LEN)) {
2167                 u32 rxdctl = er32(RXDCTL(0));
2168                 ew32(RXDCTL(0), rxdctl | 0x3);
2169                 ew32(ERT, E1000_ERT_2048 | (1 << 13));
2170                 /*
2171                  * With jumbo frames and early-receive enabled, excessive
2172                  * C4->C2 latencies result in dropped transactions.
2173                  */
2174                 pm_qos_update_requirement(PM_QOS_CPU_DMA_LATENCY,
2175                                           e1000e_driver_name, 55);
2176         } else {
2177                 pm_qos_update_requirement(PM_QOS_CPU_DMA_LATENCY,
2178                                           e1000e_driver_name,
2179                                           PM_QOS_DEFAULT_VALUE);
2180         }
2181
2182         /* Enable Receives */
2183         ew32(RCTL, rctl);
2184 }
2185
2186 /**
2187  *  e1000_update_mc_addr_list - Update Multicast addresses
2188  *  @hw: pointer to the HW structure
2189  *  @mc_addr_list: array of multicast addresses to program
2190  *  @mc_addr_count: number of multicast addresses to program
2191  *  @rar_used_count: the first RAR register free to program
2192  *  @rar_count: total number of supported Receive Address Registers
2193  *
2194  *  Updates the Receive Address Registers and Multicast Table Array.
2195  *  The caller must have a packed mc_addr_list of multicast addresses.
2196  *  The parameter rar_count will usually be hw->mac.rar_entry_count
2197  *  unless there are workarounds that change this.  Currently no func pointer
2198  *  exists and all implementations are handled in the generic version of this
2199  *  function.
2200  **/
2201 static void e1000_update_mc_addr_list(struct e1000_hw *hw, u8 *mc_addr_list,
2202                                       u32 mc_addr_count, u32 rar_used_count,
2203                                       u32 rar_count)
2204 {
2205         hw->mac.ops.update_mc_addr_list(hw, mc_addr_list, mc_addr_count,
2206                                         rar_used_count, rar_count);
2207 }
2208
2209 /**
2210  * e1000_set_multi - Multicast and Promiscuous mode set
2211  * @netdev: network interface device structure
2212  *
2213  * The set_multi entry point is called whenever the multicast address
2214  * list or the network interface flags are updated.  This routine is
2215  * responsible for configuring the hardware for proper multicast,
2216  * promiscuous mode, and all-multi behavior.
2217  **/
2218 static void e1000_set_multi(struct net_device *netdev)
2219 {
2220         struct e1000_adapter *adapter = netdev_priv(netdev);
2221         struct e1000_hw *hw = &adapter->hw;
2222         struct e1000_mac_info *mac = &hw->mac;
2223         struct dev_mc_list *mc_ptr;
2224         u8  *mta_list;
2225         u32 rctl;
2226         int i;
2227
2228         /* Check for Promiscuous and All Multicast modes */
2229
2230         rctl = er32(RCTL);
2231
2232         if (netdev->flags & IFF_PROMISC) {
2233                 rctl |= (E1000_RCTL_UPE | E1000_RCTL_MPE);
2234                 rctl &= ~E1000_RCTL_VFE;
2235         } else {
2236                 if (netdev->flags & IFF_ALLMULTI) {
2237                         rctl |= E1000_RCTL_MPE;
2238                         rctl &= ~E1000_RCTL_UPE;
2239                 } else {
2240                         rctl &= ~(E1000_RCTL_UPE | E1000_RCTL_MPE);
2241                 }
2242                 if (adapter->vlgrp && adapter->flags & FLAG_HAS_HW_VLAN_FILTER)
2243                         rctl |= E1000_RCTL_VFE;
2244         }
2245
2246         ew32(RCTL, rctl);
2247
2248         if (netdev->mc_count) {
2249                 mta_list = kmalloc(netdev->mc_count * 6, GFP_ATOMIC);
2250                 if (!mta_list)
2251                         return;
2252
2253                 /* prepare a packed array of only addresses. */
2254                 mc_ptr = netdev->mc_list;
2255
2256                 for (i = 0; i < netdev->mc_count; i++) {
2257                         if (!mc_ptr)
2258                                 break;
2259                         memcpy(mta_list + (i*ETH_ALEN), mc_ptr->dmi_addr,
2260                                ETH_ALEN);
2261                         mc_ptr = mc_ptr->next;
2262                 }
2263
2264                 e1000_update_mc_addr_list(hw, mta_list, i, 1,
2265                                           mac->rar_entry_count);
2266                 kfree(mta_list);
2267         } else {
2268                 /*
2269                  * if we're called from probe, we might not have
2270                  * anything to do here, so clear out the list
2271                  */
2272                 e1000_update_mc_addr_list(hw, NULL, 0, 1, mac->rar_entry_count);
2273         }
2274 }
2275
2276 /**
2277  * e1000_configure - configure the hardware for Rx and Tx
2278  * @adapter: private board structure
2279  **/
2280 static void e1000_configure(struct e1000_adapter *adapter)
2281 {
2282         e1000_set_multi(adapter->netdev);
2283
2284         e1000_restore_vlan(adapter);
2285         e1000_init_manageability(adapter);
2286
2287         e1000_configure_tx(adapter);
2288         e1000_setup_rctl(adapter);
2289         e1000_configure_rx(adapter);
2290         adapter->alloc_rx_buf(adapter, e1000_desc_unused(adapter->rx_ring));
2291 }
2292
2293 /**
2294  * e1000e_power_up_phy - restore link in case the phy was powered down
2295  * @adapter: address of board private structure
2296  *
2297  * The phy may be powered down to save power and turn off link when the
2298  * driver is unloaded and wake on lan is not enabled (among others)
2299  * *** this routine MUST be followed by a call to e1000e_reset ***
2300  **/
2301 void e1000e_power_up_phy(struct e1000_adapter *adapter)
2302 {
2303         u16 mii_reg = 0;
2304
2305         /* Just clear the power down bit to wake the phy back up */
2306         if (adapter->hw.phy.media_type == e1000_media_type_copper) {
2307                 /*
2308                  * According to the manual, the phy will retain its
2309                  * settings across a power-down/up cycle
2310                  */
2311                 e1e_rphy(&adapter->hw, PHY_CONTROL, &mii_reg);
2312                 mii_reg &= ~MII_CR_POWER_DOWN;
2313                 e1e_wphy(&adapter->hw, PHY_CONTROL, mii_reg);
2314         }
2315
2316         adapter->hw.mac.ops.setup_link(&adapter->hw);
2317 }
2318
2319 /**
2320  * e1000_power_down_phy - Power down the PHY
2321  *
2322  * Power down the PHY so no link is implied when interface is down
2323  * The PHY cannot be powered down is management or WoL is active
2324  */
2325 static void e1000_power_down_phy(struct e1000_adapter *adapter)
2326 {
2327         struct e1000_hw *hw = &adapter->hw;
2328         u16 mii_reg;
2329
2330         /* WoL is enabled */
2331         if (adapter->wol)
2332                 return;
2333
2334         /* non-copper PHY? */
2335         if (adapter->hw.phy.media_type != e1000_media_type_copper)
2336                 return;
2337
2338         /* reset is blocked because of a SoL/IDER session */
2339         if (e1000e_check_mng_mode(hw) || e1000_check_reset_block(hw))
2340                 return;
2341
2342         /* manageability (AMT) is enabled */
2343         if (er32(MANC) & E1000_MANC_SMBUS_EN)
2344                 return;
2345
2346         /* power down the PHY */
2347         e1e_rphy(hw, PHY_CONTROL, &mii_reg);
2348         mii_reg |= MII_CR_POWER_DOWN;
2349         e1e_wphy(hw, PHY_CONTROL, mii_reg);
2350         mdelay(1);
2351 }
2352
2353 /**
2354  * e1000e_reset - bring the hardware into a known good state
2355  *
2356  * This function boots the hardware and enables some settings that
2357  * require a configuration cycle of the hardware - those cannot be
2358  * set/changed during runtime. After reset the device needs to be
2359  * properly configured for Rx, Tx etc.
2360  */
2361 void e1000e_reset(struct e1000_adapter *adapter)
2362 {
2363         struct e1000_mac_info *mac = &adapter->hw.mac;
2364         struct e1000_fc_info *fc = &adapter->hw.fc;
2365         struct e1000_hw *hw = &adapter->hw;
2366         u32 tx_space, min_tx_space, min_rx_space;
2367         u32 pba = adapter->pba;
2368         u16 hwm;
2369
2370         /* reset Packet Buffer Allocation to default */
2371         ew32(PBA, pba);
2372
2373         if (adapter->max_frame_size > ETH_FRAME_LEN + ETH_FCS_LEN) {
2374                 /*
2375                  * To maintain wire speed transmits, the Tx FIFO should be
2376                  * large enough to accommodate two full transmit packets,
2377                  * rounded up to the next 1KB and expressed in KB.  Likewise,
2378                  * the Rx FIFO should be large enough to accommodate at least
2379                  * one full receive packet and is similarly rounded up and
2380                  * expressed in KB.
2381                  */
2382                 pba = er32(PBA);
2383                 /* upper 16 bits has Tx packet buffer allocation size in KB */
2384                 tx_space = pba >> 16;
2385                 /* lower 16 bits has Rx packet buffer allocation size in KB */
2386                 pba &= 0xffff;
2387                 /*
2388                  * the Tx fifo also stores 16 bytes of information about the tx
2389                  * but don't include ethernet FCS because hardware appends it
2390                  */
2391                 min_tx_space = (adapter->max_frame_size +
2392                                 sizeof(struct e1000_tx_desc) -
2393                                 ETH_FCS_LEN) * 2;
2394                 min_tx_space = ALIGN(min_tx_space, 1024);
2395                 min_tx_space >>= 10;
2396                 /* software strips receive CRC, so leave room for it */
2397                 min_rx_space = adapter->max_frame_size;
2398                 min_rx_space = ALIGN(min_rx_space, 1024);
2399                 min_rx_space >>= 10;
2400
2401                 /*
2402                  * If current Tx allocation is less than the min Tx FIFO size,
2403                  * and the min Tx FIFO size is less than the current Rx FIFO
2404                  * allocation, take space away from current Rx allocation
2405                  */
2406                 if ((tx_space < min_tx_space) &&
2407                     ((min_tx_space - tx_space) < pba)) {
2408                         pba -= min_tx_space - tx_space;
2409
2410                         /*
2411                          * if short on Rx space, Rx wins and must trump tx
2412                          * adjustment or use Early Receive if available
2413                          */
2414                         if ((pba < min_rx_space) &&
2415                             (!(adapter->flags & FLAG_HAS_ERT)))
2416                                 /* ERT enabled in e1000_configure_rx */
2417                                 pba = min_rx_space;
2418                 }
2419
2420                 ew32(PBA, pba);
2421         }
2422
2423
2424         /*
2425          * flow control settings
2426          *
2427          * The high water mark must be low enough to fit one full frame
2428          * (or the size used for early receive) above it in the Rx FIFO.
2429          * Set it to the lower of:
2430          * - 90% of the Rx FIFO size, and
2431          * - the full Rx FIFO size minus the early receive size (for parts
2432          *   with ERT support assuming ERT set to E1000_ERT_2048), or
2433          * - the full Rx FIFO size minus one full frame
2434          */
2435         if (adapter->flags & FLAG_HAS_ERT)
2436                 hwm = min(((pba << 10) * 9 / 10),
2437                           ((pba << 10) - (E1000_ERT_2048 << 3)));
2438         else
2439                 hwm = min(((pba << 10) * 9 / 10),
2440                           ((pba << 10) - adapter->max_frame_size));
2441
2442         fc->high_water = hwm & 0xFFF8; /* 8-byte granularity */
2443         fc->low_water = fc->high_water - 8;
2444
2445         if (adapter->flags & FLAG_DISABLE_FC_PAUSE_TIME)
2446                 fc->pause_time = 0xFFFF;
2447         else
2448                 fc->pause_time = E1000_FC_PAUSE_TIME;
2449         fc->send_xon = 1;
2450         fc->type = fc->original_type;
2451
2452         /* Allow time for pending master requests to run */
2453         mac->ops.reset_hw(hw);
2454
2455         /*
2456          * For parts with AMT enabled, let the firmware know
2457          * that the network interface is in control
2458          */
2459         if ((adapter->flags & FLAG_HAS_AMT) && e1000e_check_mng_mode(hw))
2460                 e1000_get_hw_control(adapter);
2461
2462         ew32(WUC, 0);
2463
2464         if (mac->ops.init_hw(hw))
2465                 ndev_err(adapter->netdev, "Hardware Error\n");
2466
2467         e1000_update_mng_vlan(adapter);
2468
2469         /* Enable h/w to recognize an 802.1Q VLAN Ethernet packet */
2470         ew32(VET, ETH_P_8021Q);
2471
2472         e1000e_reset_adaptive(hw);
2473         e1000_get_phy_info(hw);
2474
2475         if (!(adapter->flags & FLAG_SMART_POWER_DOWN)) {
2476                 u16 phy_data = 0;
2477                 /*
2478                  * speed up time to link by disabling smart power down, ignore
2479                  * the return value of this function because there is nothing
2480                  * different we would do if it failed
2481                  */
2482                 e1e_rphy(hw, IGP02E1000_PHY_POWER_MGMT, &phy_data);
2483                 phy_data &= ~IGP02E1000_PM_SPD;
2484                 e1e_wphy(hw, IGP02E1000_PHY_POWER_MGMT, phy_data);
2485         }
2486 }
2487
2488 int e1000e_up(struct e1000_adapter *adapter)
2489 {
2490         struct e1000_hw *hw = &adapter->hw;
2491
2492         /* hardware has been reset, we need to reload some things */
2493         e1000_configure(adapter);
2494
2495         clear_bit(__E1000_DOWN, &adapter->state);
2496
2497         napi_enable(&adapter->napi);
2498         e1000_irq_enable(adapter);
2499
2500         /* fire a link change interrupt to start the watchdog */
2501         ew32(ICS, E1000_ICS_LSC);
2502         return 0;
2503 }
2504
2505 void e1000e_down(struct e1000_adapter *adapter)
2506 {
2507         struct net_device *netdev = adapter->netdev;
2508         struct e1000_hw *hw = &adapter->hw;
2509         u32 tctl, rctl;
2510
2511         /*
2512          * signal that we're down so the interrupt handler does not
2513          * reschedule our watchdog timer
2514          */
2515         set_bit(__E1000_DOWN, &adapter->state);
2516
2517         /* disable receives in the hardware */
2518         rctl = er32(RCTL);
2519         ew32(RCTL, rctl & ~E1000_RCTL_EN);
2520         /* flush and sleep below */
2521
2522         netif_stop_queue(netdev);
2523
2524         /* disable transmits in the hardware */
2525         tctl = er32(TCTL);
2526         tctl &= ~E1000_TCTL_EN;
2527         ew32(TCTL, tctl);
2528         /* flush both disables and wait for them to finish */
2529         e1e_flush();
2530         msleep(10);
2531
2532         napi_disable(&adapter->napi);
2533         e1000_irq_disable(adapter);
2534
2535         del_timer_sync(&adapter->watchdog_timer);
2536         del_timer_sync(&adapter->phy_info_timer);
2537
2538         netdev->tx_queue_len = adapter->tx_queue_len;
2539         netif_carrier_off(netdev);
2540         adapter->link_speed = 0;
2541         adapter->link_duplex = 0;
2542
2543         if (!pci_channel_offline(adapter->pdev))
2544                 e1000e_reset(adapter);
2545         e1000_clean_tx_ring(adapter);
2546         e1000_clean_rx_ring(adapter);
2547
2548         /*
2549          * TODO: for power management, we could drop the link and
2550          * pci_disable_device here.
2551          */
2552 }
2553
2554 void e1000e_reinit_locked(struct e1000_adapter *adapter)
2555 {
2556         might_sleep();
2557         while (test_and_set_bit(__E1000_RESETTING, &adapter->state))
2558                 msleep(1);
2559         e1000e_down(adapter);
2560         e1000e_up(adapter);
2561         clear_bit(__E1000_RESETTING, &adapter->state);
2562 }
2563
2564 /**
2565  * e1000_sw_init - Initialize general software structures (struct e1000_adapter)
2566  * @adapter: board private structure to initialize
2567  *
2568  * e1000_sw_init initializes the Adapter private data structure.
2569  * Fields are initialized based on PCI device information and
2570  * OS network device settings (MTU size).
2571  **/
2572 static int __devinit e1000_sw_init(struct e1000_adapter *adapter)
2573 {
2574         struct net_device *netdev = adapter->netdev;
2575
2576         adapter->rx_buffer_len = ETH_FRAME_LEN + VLAN_HLEN + ETH_FCS_LEN;
2577         adapter->rx_ps_bsize0 = 128;
2578         adapter->max_frame_size = netdev->mtu + ETH_HLEN + ETH_FCS_LEN;
2579         adapter->min_frame_size = ETH_ZLEN + ETH_FCS_LEN;
2580
2581         adapter->tx_ring = kzalloc(sizeof(struct e1000_ring), GFP_KERNEL);
2582         if (!adapter->tx_ring)
2583                 goto err;
2584
2585         adapter->rx_ring = kzalloc(sizeof(struct e1000_ring), GFP_KERNEL);
2586         if (!adapter->rx_ring)
2587                 goto err;
2588
2589         spin_lock_init(&adapter->tx_queue_lock);
2590
2591         /* Explicitly disable IRQ since the NIC can be in any state. */
2592         e1000_irq_disable(adapter);
2593
2594         spin_lock_init(&adapter->stats_lock);
2595
2596         set_bit(__E1000_DOWN, &adapter->state);
2597         return 0;
2598
2599 err:
2600         ndev_err(netdev, "Unable to allocate memory for queues\n");
2601         kfree(adapter->rx_ring);
2602         kfree(adapter->tx_ring);
2603         return -ENOMEM;
2604 }
2605
2606 /**
2607  * e1000_open - Called when a network interface is made active
2608  * @netdev: network interface device structure
2609  *
2610  * Returns 0 on success, negative value on failure
2611  *
2612  * The open entry point is called when a network interface is made
2613  * active by the system (IFF_UP).  At this point all resources needed
2614  * for transmit and receive operations are allocated, the interrupt
2615  * handler is registered with the OS, the watchdog timer is started,
2616  * and the stack is notified that the interface is ready.
2617  **/
2618 static int e1000_open(struct net_device *netdev)
2619 {
2620         struct e1000_adapter *adapter = netdev_priv(netdev);
2621         struct e1000_hw *hw = &adapter->hw;
2622         int err;
2623
2624         /* disallow open during test */
2625         if (test_bit(__E1000_TESTING, &adapter->state))
2626                 return -EBUSY;
2627
2628         /* allocate transmit descriptors */
2629         err = e1000e_setup_tx_resources(adapter);
2630         if (err)
2631                 goto err_setup_tx;
2632
2633         /* allocate receive descriptors */
2634         err = e1000e_setup_rx_resources(adapter);
2635         if (err)
2636                 goto err_setup_rx;
2637
2638         e1000e_power_up_phy(adapter);
2639
2640         adapter->mng_vlan_id = E1000_MNG_VLAN_NONE;
2641         if ((adapter->hw.mng_cookie.status &
2642              E1000_MNG_DHCP_COOKIE_STATUS_VLAN))
2643                 e1000_update_mng_vlan(adapter);
2644
2645         /*
2646          * If AMT is enabled, let the firmware know that the network
2647          * interface is now open
2648          */
2649         if ((adapter->flags & FLAG_HAS_AMT) &&
2650             e1000e_check_mng_mode(&adapter->hw))
2651                 e1000_get_hw_control(adapter);
2652
2653         /*
2654          * before we allocate an interrupt, we must be ready to handle it.
2655          * Setting DEBUG_SHIRQ in the kernel makes it fire an interrupt
2656          * as soon as we call pci_request_irq, so we have to setup our
2657          * clean_rx handler before we do so.
2658          */
2659         e1000_configure(adapter);
2660
2661         err = e1000_request_irq(adapter);
2662         if (err)
2663                 goto err_req_irq;
2664
2665         /* From here on the code is the same as e1000e_up() */
2666         clear_bit(__E1000_DOWN, &adapter->state);
2667
2668         napi_enable(&adapter->napi);
2669
2670         e1000_irq_enable(adapter);
2671
2672         /* fire a link status change interrupt to start the watchdog */
2673         ew32(ICS, E1000_ICS_LSC);
2674
2675         return 0;
2676
2677 err_req_irq:
2678         e1000_release_hw_control(adapter);
2679         e1000_power_down_phy(adapter);
2680         e1000e_free_rx_resources(adapter);
2681 err_setup_rx:
2682         e1000e_free_tx_resources(adapter);
2683 err_setup_tx:
2684         e1000e_reset(adapter);
2685
2686         return err;
2687 }
2688
2689 /**
2690  * e1000_close - Disables a network interface
2691  * @netdev: network interface device structure
2692  *
2693  * Returns 0, this is not allowed to fail
2694  *
2695  * The close entry point is called when an interface is de-activated
2696  * by the OS.  The hardware is still under the drivers control, but
2697  * needs to be disabled.  A global MAC reset is issued to stop the
2698  * hardware, and all transmit and receive resources are freed.
2699  **/
2700 static int e1000_close(struct net_device *netdev)
2701 {
2702         struct e1000_adapter *adapter = netdev_priv(netdev);
2703
2704         WARN_ON(test_bit(__E1000_RESETTING, &adapter->state));
2705         e1000e_down(adapter);
2706         e1000_power_down_phy(adapter);
2707         e1000_free_irq(adapter);
2708
2709         e1000e_free_tx_resources(adapter);
2710         e1000e_free_rx_resources(adapter);
2711
2712         /*
2713          * kill manageability vlan ID if supported, but not if a vlan with
2714          * the same ID is registered on the host OS (let 8021q kill it)
2715          */
2716         if ((adapter->hw.mng_cookie.status &
2717                           E1000_MNG_DHCP_COOKIE_STATUS_VLAN) &&
2718              !(adapter->vlgrp &&
2719                vlan_group_get_device(adapter->vlgrp, adapter->mng_vlan_id)))
2720                 e1000_vlan_rx_kill_vid(netdev, adapter->mng_vlan_id);
2721
2722         /*
2723          * If AMT is enabled, let the firmware know that the network
2724          * interface is now closed
2725          */
2726         if ((adapter->flags & FLAG_HAS_AMT) &&
2727             e1000e_check_mng_mode(&adapter->hw))
2728                 e1000_release_hw_control(adapter);
2729
2730         return 0;
2731 }
2732 /**
2733  * e1000_set_mac - Change the Ethernet Address of the NIC
2734  * @netdev: network interface device structure
2735  * @p: pointer to an address structure
2736  *
2737  * Returns 0 on success, negative on failure
2738  **/
2739 static int e1000_set_mac(struct net_device *netdev, void *p)
2740 {
2741         struct e1000_adapter *adapter = netdev_priv(netdev);
2742         struct sockaddr *addr = p;
2743
2744         if (!is_valid_ether_addr(addr->sa_data))
2745                 return -EADDRNOTAVAIL;
2746
2747         memcpy(netdev->dev_addr, addr->sa_data, netdev->addr_len);
2748         memcpy(adapter->hw.mac.addr, addr->sa_data, netdev->addr_len);
2749
2750         e1000e_rar_set(&adapter->hw, adapter->hw.mac.addr, 0);
2751
2752         if (adapter->flags & FLAG_RESET_OVERWRITES_LAA) {
2753                 /* activate the work around */
2754                 e1000e_set_laa_state_82571(&adapter->hw, 1);
2755
2756                 /*
2757                  * Hold a copy of the LAA in RAR[14] This is done so that
2758                  * between the time RAR[0] gets clobbered  and the time it
2759                  * gets fixed (in e1000_watchdog), the actual LAA is in one
2760                  * of the RARs and no incoming packets directed to this port
2761                  * are dropped. Eventually the LAA will be in RAR[0] and
2762                  * RAR[14]
2763                  */
2764                 e1000e_rar_set(&adapter->hw,
2765                               adapter->hw.mac.addr,
2766                               adapter->hw.mac.rar_entry_count - 1);
2767         }
2768
2769         return 0;
2770 }
2771
2772 /*
2773  * Need to wait a few seconds after link up to get diagnostic information from
2774  * the phy
2775  */
2776 static void e1000_update_phy_info(unsigned long data)
2777 {
2778         struct e1000_adapter *adapter = (struct e1000_adapter *) data;
2779         e1000_get_phy_info(&adapter->hw);
2780 }
2781
2782 /**
2783  * e1000e_update_stats - Update the board statistics counters
2784  * @adapter: board private structure
2785  **/
2786 void e1000e_update_stats(struct e1000_adapter *adapter)
2787 {
2788         struct e1000_hw *hw = &adapter->hw;
2789         struct pci_dev *pdev = adapter->pdev;
2790         unsigned long irq_flags;
2791         u16 phy_tmp;
2792
2793 #define PHY_IDLE_ERROR_COUNT_MASK 0x00FF
2794
2795         /*
2796          * Prevent stats update while adapter is being reset, or if the pci
2797          * connection is down.
2798          */
2799         if (adapter->link_speed == 0)
2800                 return;
2801         if (pci_channel_offline(pdev))
2802                 return;
2803
2804         spin_lock_irqsave(&adapter->stats_lock, irq_flags);
2805
2806         /*
2807          * these counters are modified from e1000_adjust_tbi_stats,
2808          * called from the interrupt context, so they must only
2809          * be written while holding adapter->stats_lock
2810          */
2811
2812         adapter->stats.crcerrs += er32(CRCERRS);
2813         adapter->stats.gprc += er32(GPRC);
2814         adapter->stats.gorc += er32(GORCL);
2815         er32(GORCH); /* Clear gorc */
2816         adapter->stats.bprc += er32(BPRC);
2817         adapter->stats.mprc += er32(MPRC);
2818         adapter->stats.roc += er32(ROC);
2819
2820         adapter->stats.mpc += er32(MPC);
2821         adapter->stats.scc += er32(SCC);
2822         adapter->stats.ecol += er32(ECOL);
2823         adapter->stats.mcc += er32(MCC);
2824         adapter->stats.latecol += er32(LATECOL);
2825         adapter->stats.dc += er32(DC);
2826         adapter->stats.xonrxc += er32(XONRXC);
2827         adapter->stats.xontxc += er32(XONTXC);
2828         adapter->stats.xoffrxc += er32(XOFFRXC);
2829         adapter->stats.xofftxc += er32(XOFFTXC);
2830         adapter->stats.gptc += er32(GPTC);
2831         adapter->stats.gotc += er32(GOTCL);
2832         er32(GOTCH); /* Clear gotc */
2833         adapter->stats.rnbc += er32(RNBC);
2834         adapter->stats.ruc += er32(RUC);
2835
2836         adapter->stats.mptc += er32(MPTC);
2837         adapter->stats.bptc += er32(BPTC);
2838
2839         /* used for adaptive IFS */
2840
2841         hw->mac.tx_packet_delta = er32(TPT);
2842         adapter->stats.tpt += hw->mac.tx_packet_delta;
2843         hw->mac.collision_delta = er32(COLC);
2844         adapter->stats.colc += hw->mac.collision_delta;
2845
2846         adapter->stats.algnerrc += er32(ALGNERRC);
2847         adapter->stats.rxerrc += er32(RXERRC);
2848         adapter->stats.tncrs += er32(TNCRS);
2849         adapter->stats.cexterr += er32(CEXTERR);
2850         adapter->stats.tsctc += er32(TSCTC);
2851         adapter->stats.tsctfc += er32(TSCTFC);
2852
2853         /* Fill out the OS statistics structure */
2854         adapter->net_stats.multicast = adapter->stats.mprc;
2855         adapter->net_stats.collisions = adapter->stats.colc;
2856
2857         /* Rx Errors */
2858
2859         /*
2860          * RLEC on some newer hardware can be incorrect so build
2861          * our own version based on RUC and ROC
2862          */
2863         adapter->net_stats.rx_errors = adapter->stats.rxerrc +
2864                 adapter->stats.crcerrs + adapter->stats.algnerrc +
2865                 adapter->stats.ruc + adapter->stats.roc +
2866                 adapter->stats.cexterr;
2867         adapter->net_stats.rx_length_errors = adapter->stats.ruc +
2868                                               adapter->stats.roc;
2869         adapter->net_stats.rx_crc_errors = adapter->stats.crcerrs;
2870         adapter->net_stats.rx_frame_errors = adapter->stats.algnerrc;
2871         adapter->net_stats.rx_missed_errors = adapter->stats.mpc;
2872
2873         /* Tx Errors */
2874         adapter->net_stats.tx_errors = adapter->stats.ecol +
2875                                        adapter->stats.latecol;
2876         adapter->net_stats.tx_aborted_errors = adapter->stats.ecol;
2877         adapter->net_stats.tx_window_errors = adapter->stats.latecol;
2878         adapter->net_stats.tx_carrier_errors = adapter->stats.tncrs;
2879
2880         /* Tx Dropped needs to be maintained elsewhere */
2881
2882         /* Phy Stats */
2883         if (hw->phy.media_type == e1000_media_type_copper) {
2884                 if ((adapter->link_speed == SPEED_1000) &&
2885                    (!e1e_rphy(hw, PHY_1000T_STATUS, &phy_tmp))) {
2886                         phy_tmp &= PHY_IDLE_ERROR_COUNT_MASK;
2887                         adapter->phy_stats.idle_errors += phy_tmp;
2888                 }
2889         }
2890
2891         /* Management Stats */
2892         adapter->stats.mgptc += er32(MGTPTC);
2893         adapter->stats.mgprc += er32(MGTPRC);
2894         adapter->stats.mgpdc += er32(MGTPDC);
2895
2896         spin_unlock_irqrestore(&adapter->stats_lock, irq_flags);
2897 }
2898
2899 /**
2900  * e1000_phy_read_status - Update the PHY register status snapshot
2901  * @adapter: board private structure
2902  **/
2903 static void e1000_phy_read_status(struct e1000_adapter *adapter)
2904 {
2905         struct e1000_hw *hw = &adapter->hw;
2906         struct e1000_phy_regs *phy = &adapter->phy_regs;
2907         int ret_val;
2908         unsigned long irq_flags;
2909
2910
2911         spin_lock_irqsave(&adapter->stats_lock, irq_flags);
2912
2913         if ((er32(STATUS) & E1000_STATUS_LU) &&
2914             (adapter->hw.phy.media_type == e1000_media_type_copper)) {
2915                 ret_val  = e1e_rphy(hw, PHY_CONTROL, &phy->bmcr);
2916                 ret_val |= e1e_rphy(hw, PHY_STATUS, &phy->bmsr);
2917                 ret_val |= e1e_rphy(hw, PHY_AUTONEG_ADV, &phy->advertise);
2918                 ret_val |= e1e_rphy(hw, PHY_LP_ABILITY, &phy->lpa);
2919                 ret_val |= e1e_rphy(hw, PHY_AUTONEG_EXP, &phy->expansion);
2920                 ret_val |= e1e_rphy(hw, PHY_1000T_CTRL, &phy->ctrl1000);
2921                 ret_val |= e1e_rphy(hw, PHY_1000T_STATUS, &phy->stat1000);
2922                 ret_val |= e1e_rphy(hw, PHY_EXT_STATUS, &phy->estatus);
2923                 if (ret_val)
2924                         ndev_warn(adapter->netdev,
2925                                   "Error reading PHY register\n");
2926         } else {
2927                 /*
2928                  * Do not read PHY registers if link is not up
2929                  * Set values to typical power-on defaults
2930                  */
2931                 phy->bmcr = (BMCR_SPEED1000 | BMCR_ANENABLE | BMCR_FULLDPLX);
2932                 phy->bmsr = (BMSR_100FULL | BMSR_100HALF | BMSR_10FULL |
2933                              BMSR_10HALF | BMSR_ESTATEN | BMSR_ANEGCAPABLE |
2934                              BMSR_ERCAP);
2935                 phy->advertise = (ADVERTISE_PAUSE_ASYM | ADVERTISE_PAUSE_CAP |
2936                                   ADVERTISE_ALL | ADVERTISE_CSMA);
2937                 phy->lpa = 0;
2938                 phy->expansion = EXPANSION_ENABLENPAGE;
2939                 phy->ctrl1000 = ADVERTISE_1000FULL;
2940                 phy->stat1000 = 0;
2941                 phy->estatus = (ESTATUS_1000_TFULL | ESTATUS_1000_THALF);
2942         }
2943
2944         spin_unlock_irqrestore(&adapter->stats_lock, irq_flags);
2945 }
2946
2947 static void e1000_print_link_info(struct e1000_adapter *adapter)
2948 {
2949         struct e1000_hw *hw = &adapter->hw;
2950         struct net_device *netdev = adapter->netdev;
2951         u32 ctrl = er32(CTRL);
2952
2953         ndev_info(netdev,
2954                 "Link is Up %d Mbps %s, Flow Control: %s\n",
2955                 adapter->link_speed,
2956                 (adapter->link_duplex == FULL_DUPLEX) ?
2957                                 "Full Duplex" : "Half Duplex",
2958                 ((ctrl & E1000_CTRL_TFCE) && (ctrl & E1000_CTRL_RFCE)) ?
2959                                 "RX/TX" :
2960                 ((ctrl & E1000_CTRL_RFCE) ? "RX" :
2961                 ((ctrl & E1000_CTRL_TFCE) ? "TX" : "None" )));
2962 }
2963
2964 static bool e1000_has_link(struct e1000_adapter *adapter)
2965 {
2966         struct e1000_hw *hw = &adapter->hw;
2967         bool link_active = 0;
2968         s32 ret_val = 0;
2969
2970         /*
2971          * get_link_status is set on LSC (link status) interrupt or
2972          * Rx sequence error interrupt.  get_link_status will stay
2973          * false until the check_for_link establishes link
2974          * for copper adapters ONLY
2975          */
2976         switch (hw->phy.media_type) {
2977         case e1000_media_type_copper:
2978                 if (hw->mac.get_link_status) {
2979                         ret_val = hw->mac.ops.check_for_link(hw);
2980                         link_active = !hw->mac.get_link_status;
2981                 } else {
2982                         link_active = 1;
2983                 }
2984                 break;
2985         case e1000_media_type_fiber:
2986                 ret_val = hw->mac.ops.check_for_link(hw);
2987                 link_active = !!(er32(STATUS) & E1000_STATUS_LU);
2988                 break;
2989         case e1000_media_type_internal_serdes:
2990                 ret_val = hw->mac.ops.check_for_link(hw);
2991                 link_active = adapter->hw.mac.serdes_has_link;
2992                 break;
2993         default:
2994         case e1000_media_type_unknown:
2995                 break;
2996         }
2997
2998         if ((ret_val == E1000_ERR_PHY) && (hw->phy.type == e1000_phy_igp_3) &&
2999             (er32(CTRL) & E1000_PHY_CTRL_GBE_DISABLE)) {
3000                 /* See e1000_kmrn_lock_loss_workaround_ich8lan() */
3001                 ndev_info(adapter->netdev,
3002                           "Gigabit has been disabled, downgrading speed\n");
3003         }
3004
3005         return link_active;
3006 }
3007
3008 static void e1000e_enable_receives(struct e1000_adapter *adapter)
3009 {
3010         /* make sure the receive unit is started */
3011         if ((adapter->flags & FLAG_RX_NEEDS_RESTART) &&
3012             (adapter->flags & FLAG_RX_RESTART_NOW)) {
3013                 struct e1000_hw *hw = &adapter->hw;
3014                 u32 rctl = er32(RCTL);
3015                 ew32(RCTL, rctl | E1000_RCTL_EN);
3016                 adapter->flags &= ~FLAG_RX_RESTART_NOW;
3017         }
3018 }
3019
3020 /**
3021  * e1000_watchdog - Timer Call-back
3022  * @data: pointer to adapter cast into an unsigned long
3023  **/
3024 static void e1000_watchdog(unsigned long data)
3025 {
3026         struct e1000_adapter *adapter = (struct e1000_adapter *) data;
3027
3028         /* Do the rest outside of interrupt context */
3029         schedule_work(&adapter->watchdog_task);
3030
3031         /* TODO: make this use queue_delayed_work() */
3032 }
3033
3034 static void e1000_watchdog_task(struct work_struct *work)
3035 {
3036         struct e1000_adapter *adapter = container_of(work,
3037                                         struct e1000_adapter, watchdog_task);
3038         struct net_device *netdev = adapter->netdev;
3039         struct e1000_mac_info *mac = &adapter->hw.mac;
3040         struct e1000_ring *tx_ring = adapter->tx_ring;
3041         struct e1000_hw *hw = &adapter->hw;
3042         u32 link, tctl;
3043         int tx_pending = 0;
3044
3045         link = e1000_has_link(adapter);
3046         if ((netif_carrier_ok(netdev)) && link) {
3047                 e1000e_enable_receives(adapter);
3048                 goto link_up;
3049         }
3050
3051         if ((e1000e_enable_tx_pkt_filtering(hw)) &&
3052             (adapter->mng_vlan_id != adapter->hw.mng_cookie.vlan_id))
3053                 e1000_update_mng_vlan(adapter);
3054
3055         if (link) {
3056                 if (!netif_carrier_ok(netdev)) {
3057                         bool txb2b = 1;
3058                         /* update snapshot of PHY registers on LSC */
3059                         e1000_phy_read_status(adapter);
3060                         mac->ops.get_link_up_info(&adapter->hw,
3061                                                    &adapter->link_speed,
3062                                                    &adapter->link_duplex);
3063                         e1000_print_link_info(adapter);
3064                         /*
3065                          * tweak tx_queue_len according to speed/duplex
3066                          * and adjust the timeout factor
3067                          */
3068                         netdev->tx_queue_len = adapter->tx_queue_len;
3069                         adapter->tx_timeout_factor = 1;
3070                         switch (adapter->link_speed) {
3071                         case SPEED_10:
3072                                 txb2b = 0;
3073                                 netdev->tx_queue_len = 10;
3074                                 adapter->tx_timeout_factor = 14;
3075                                 break;
3076                         case SPEED_100:
3077                                 txb2b = 0;
3078                                 netdev->tx_queue_len = 100;
3079                                 /* maybe add some timeout factor ? */
3080                                 break;
3081                         }
3082
3083                         /*
3084                          * workaround: re-program speed mode bit after
3085                          * link-up event
3086                          */
3087                         if ((adapter->flags & FLAG_TARC_SPEED_MODE_BIT) &&
3088                             !txb2b) {
3089                                 u32 tarc0;
3090                                 tarc0 = er32(TARC(0));
3091                                 tarc0 &= ~SPEED_MODE_BIT;
3092                                 ew32(TARC(0), tarc0);
3093                         }
3094
3095                         /*
3096                          * disable TSO for pcie and 10/100 speeds, to avoid
3097                          * some hardware issues
3098                          */
3099                         if (!(adapter->flags & FLAG_TSO_FORCE)) {
3100                                 switch (adapter->link_speed) {
3101                                 case SPEED_10:
3102                                 case SPEED_100:
3103                                         ndev_info(netdev,
3104                                         "10/100 speed: disabling TSO\n");
3105                                         netdev->features &= ~NETIF_F_TSO;
3106                                         netdev->features &= ~NETIF_F_TSO6;
3107                                         break;
3108                                 case SPEED_1000:
3109                                         netdev->features |= NETIF_F_TSO;
3110                                         netdev->features |= NETIF_F_TSO6;
3111                                         break;
3112                                 default:
3113                                         /* oops */
3114                                         break;
3115                                 }
3116                         }
3117
3118                         /*
3119                          * enable transmits in the hardware, need to do this
3120                          * after setting TARC(0)
3121                          */
3122                         tctl = er32(TCTL);
3123                         tctl |= E1000_TCTL_EN;
3124                         ew32(TCTL, tctl);
3125
3126                         netif_carrier_on(netdev);
3127                         netif_wake_queue(netdev);
3128
3129                         if (!test_bit(__E1000_DOWN, &adapter->state))
3130                                 mod_timer(&adapter->phy_info_timer,
3131                                           round_jiffies(jiffies + 2 * HZ));
3132                 }
3133         } else {
3134                 if (netif_carrier_ok(netdev)) {
3135                         adapter->link_speed = 0;
3136                         adapter->link_duplex = 0;
3137                         ndev_info(netdev, "Link is Down\n");
3138                         netif_carrier_off(netdev);
3139                         netif_stop_queue(netdev);
3140                         if (!test_bit(__E1000_DOWN, &adapter->state))
3141                                 mod_timer(&adapter->phy_info_timer,
3142                                           round_jiffies(jiffies + 2 * HZ));
3143
3144                         if (adapter->flags & FLAG_RX_NEEDS_RESTART)
3145                                 schedule_work(&adapter->reset_task);
3146                 }
3147         }
3148
3149 link_up:
3150         e1000e_update_stats(adapter);
3151
3152         mac->tx_packet_delta = adapter->stats.tpt - adapter->tpt_old;
3153         adapter->tpt_old = adapter->stats.tpt;
3154         mac->collision_delta = adapter->stats.colc - adapter->colc_old;
3155         adapter->colc_old = adapter->stats.colc;
3156
3157         adapter->gorc = adapter->stats.gorc - adapter->gorc_old;
3158         adapter->gorc_old = adapter->stats.gorc;
3159         adapter->gotc = adapter->stats.gotc - adapter->gotc_old;
3160         adapter->gotc_old = adapter->stats.gotc;
3161
3162         e1000e_update_adaptive(&adapter->hw);
3163
3164         if (!netif_carrier_ok(netdev)) {
3165                 tx_pending = (e1000_desc_unused(tx_ring) + 1 <
3166                                tx_ring->count);
3167                 if (tx_pending) {
3168                         /*
3169                          * We've lost link, so the controller stops DMA,
3170                          * but we've got queued Tx work that's never going
3171                          * to get done, so reset controller to flush Tx.
3172                          * (Do the reset outside of interrupt context).
3173                          */
3174                         adapter->tx_timeout_count++;
3175                         schedule_work(&adapter->reset_task);
3176                 }
3177         }
3178
3179         /* Cause software interrupt to ensure Rx ring is cleaned */
3180         ew32(ICS, E1000_ICS_RXDMT0);
3181
3182         /* Force detection of hung controller every watchdog period */
3183         adapter->detect_tx_hung = 1;
3184
3185         /*
3186          * With 82571 controllers, LAA may be overwritten due to controller
3187          * reset from the other port. Set the appropriate LAA in RAR[0]
3188          */
3189         if (e1000e_get_laa_state_82571(hw))
3190                 e1000e_rar_set(hw, adapter->hw.mac.addr, 0);
3191
3192         /* Reset the timer */
3193         if (!test_bit(__E1000_DOWN, &adapter->state))
3194                 mod_timer(&adapter->watchdog_timer,
3195                           round_jiffies(jiffies + 2 * HZ));
3196 }
3197
3198 #define E1000_TX_FLAGS_CSUM             0x00000001
3199 #define E1000_TX_FLAGS_VLAN             0x00000002
3200 #define E1000_TX_FLAGS_TSO              0x00000004
3201 #define E1000_TX_FLAGS_IPV4             0x00000008
3202 #define E1000_TX_FLAGS_VLAN_MASK        0xffff0000
3203 #define E1000_TX_FLAGS_VLAN_SHIFT       16
3204
3205 static int e1000_tso(struct e1000_adapter *adapter,
3206                      struct sk_buff *skb)
3207 {
3208         struct e1000_ring *tx_ring = adapter->tx_ring;
3209         struct e1000_context_desc *context_desc;
3210         struct e1000_buffer *buffer_info;
3211         unsigned int i;
3212         u32 cmd_length = 0;
3213         u16 ipcse = 0, tucse, mss;
3214         u8 ipcss, ipcso, tucss, tucso, hdr_len;
3215         int err;
3216
3217         if (skb_is_gso(skb)) {
3218                 if (skb_header_cloned(skb)) {
3219                         err = pskb_expand_head(skb, 0, 0, GFP_ATOMIC);
3220                         if (err)
3221                                 return err;
3222                 }
3223
3224                 hdr_len = skb_transport_offset(skb) + tcp_hdrlen(skb);
3225                 mss = skb_shinfo(skb)->gso_size;
3226                 if (skb->protocol == htons(ETH_P_IP)) {
3227                         struct iphdr *iph = ip_hdr(skb);
3228                         iph->tot_len = 0;
3229                         iph->check = 0;
3230                         tcp_hdr(skb)->check = ~csum_tcpudp_magic(iph->saddr,
3231                                                                  iph->daddr, 0,
3232                                                                  IPPROTO_TCP,
3233                                                                  0);
3234                         cmd_length = E1000_TXD_CMD_IP;
3235                         ipcse = skb_transport_offset(skb) - 1;
3236                 } else if (skb_shinfo(skb)->gso_type == SKB_GSO_TCPV6) {
3237                         ipv6_hdr(skb)->payload_len = 0;
3238                         tcp_hdr(skb)->check =
3239                                 ~csum_ipv6_magic(&ipv6_hdr(skb)->saddr,
3240                                                  &ipv6_hdr(skb)->daddr,
3241                                                  0, IPPROTO_TCP, 0);
3242                         ipcse = 0;
3243                 }
3244                 ipcss = skb_network_offset(skb);
3245                 ipcso = (void *)&(ip_hdr(skb)->check) - (void *)skb->data;
3246                 tucss = skb_transport_offset(skb);
3247                 tucso = (void *)&(tcp_hdr(skb)->check) - (void *)skb->data;
3248                 tucse = 0;
3249
3250                 cmd_length |= (E1000_TXD_CMD_DEXT | E1000_TXD_CMD_TSE |
3251                                E1000_TXD_CMD_TCP | (skb->len - (hdr_len)));
3252
3253                 i = tx_ring->next_to_use;
3254                 context_desc = E1000_CONTEXT_DESC(*tx_ring, i);
3255                 buffer_info = &tx_ring->buffer_info[i];
3256
3257                 context_desc->lower_setup.ip_fields.ipcss  = ipcss;
3258                 context_desc->lower_setup.ip_fields.ipcso  = ipcso;
3259                 context_desc->lower_setup.ip_fields.ipcse  = cpu_to_le16(ipcse);
3260                 context_desc->upper_setup.tcp_fields.tucss = tucss;
3261                 context_desc->upper_setup.tcp_fields.tucso = tucso;
3262                 context_desc->upper_setup.tcp_fields.tucse = cpu_to_le16(tucse);
3263                 context_desc->tcp_seg_setup.fields.mss     = cpu_to_le16(mss);
3264                 context_desc->tcp_seg_setup.fields.hdr_len = hdr_len;
3265                 context_desc->cmd_and_length = cpu_to_le32(cmd_length);
3266
3267                 buffer_info->time_stamp = jiffies;
3268                 buffer_info->next_to_watch = i;
3269
3270                 i++;
3271                 if (i == tx_ring->count)
3272                         i = 0;
3273                 tx_ring->next_to_use = i;
3274
3275                 return 1;
3276         }
3277
3278         return 0;
3279 }
3280
3281 static bool e1000_tx_csum(struct e1000_adapter *adapter, struct sk_buff *skb)
3282 {
3283         struct e1000_ring *tx_ring = adapter->tx_ring;
3284         struct e1000_context_desc *context_desc;
3285         struct e1000_buffer *buffer_info;
3286         unsigned int i;
3287         u8 css;
3288
3289         if (skb->ip_summed == CHECKSUM_PARTIAL) {
3290                 css = skb_transport_offset(skb);
3291
3292                 i = tx_ring->next_to_use;
3293                 buffer_info = &tx_ring->buffer_info[i];
3294                 context_desc = E1000_CONTEXT_DESC(*tx_ring, i);
3295
3296                 context_desc->lower_setup.ip_config = 0;
3297                 context_desc->upper_setup.tcp_fields.tucss = css;
3298                 context_desc->upper_setup.tcp_fields.tucso =
3299                                         css + skb->csum_offset;
3300                 context_desc->upper_setup.tcp_fields.tucse = 0;
3301                 context_desc->tcp_seg_setup.data = 0;
3302                 context_desc->cmd_and_length = cpu_to_le32(E1000_TXD_CMD_DEXT);
3303
3304                 buffer_info->time_stamp = jiffies;
3305                 buffer_info->next_to_watch = i;
3306
3307                 i++;
3308                 if (i == tx_ring->count)
3309                         i = 0;
3310                 tx_ring->next_to_use = i;
3311
3312                 return 1;
3313         }
3314
3315         return 0;
3316 }
3317
3318 #define E1000_MAX_PER_TXD       8192
3319 #define E1000_MAX_TXD_PWR       12
3320
3321 static int e1000_tx_map(struct e1000_adapter *adapter,
3322                         struct sk_buff *skb, unsigned int first,
3323                         unsigned int max_per_txd, unsigned int nr_frags,
3324                         unsigned int mss)
3325 {
3326         struct e1000_ring *tx_ring = adapter->tx_ring;
3327         struct e1000_buffer *buffer_info;
3328         unsigned int len = skb->len - skb->data_len;
3329         unsigned int offset = 0, size, count = 0, i;
3330         unsigned int f;
3331
3332         i = tx_ring->next_to_use;
3333
3334         while (len) {
3335                 buffer_info = &tx_ring->buffer_info[i];
3336                 size = min(len, max_per_txd);
3337
3338                 /* Workaround for premature desc write-backs
3339                  * in TSO mode.  Append 4-byte sentinel desc */
3340                 if (mss && !nr_frags && size == len && size > 8)
3341                         size -= 4;
3342
3343                 buffer_info->length = size;
3344                 /* set time_stamp *before* dma to help avoid a possible race */
3345                 buffer_info->time_stamp = jiffies;
3346                 buffer_info->dma =
3347                         pci_map_single(adapter->pdev,
3348                                 skb->data + offset,
3349                                 size,
3350                                 PCI_DMA_TODEVICE);
3351                 if (pci_dma_mapping_error(buffer_info->dma)) {
3352                         dev_err(&adapter->pdev->dev, "TX DMA map failed\n");
3353                         adapter->tx_dma_failed++;
3354                         return -1;
3355                 }
3356                 buffer_info->next_to_watch = i;
3357
3358                 len -= size;
3359                 offset += size;
3360                 count++;
3361                 i++;
3362                 if (i == tx_ring->count)
3363                         i = 0;
3364         }
3365
3366         for (f = 0; f < nr_frags; f++) {
3367                 struct skb_frag_struct *frag;
3368
3369                 frag = &skb_shinfo(skb)->frags[f];
3370                 len = frag->size;
3371                 offset = frag->page_offset;
3372
3373                 while (len) {
3374                         buffer_info = &tx_ring->buffer_info[i];
3375                         size = min(len, max_per_txd);
3376                         /* Workaround for premature desc write-backs
3377                          * in TSO mode.  Append 4-byte sentinel desc */
3378                         if (mss && f == (nr_frags-1) && size == len && size > 8)
3379                                 size -= 4;
3380
3381                         buffer_info->length = size;
3382                         buffer_info->time_stamp = jiffies;
3383                         buffer_info->dma =
3384                                 pci_map_page(adapter->pdev,
3385                                         frag->page,
3386                                         offset,
3387                                         size,
3388                                         PCI_DMA_TODEVICE);
3389                         if (pci_dma_mapping_error(buffer_info->dma)) {
3390                                 dev_err(&adapter->pdev->dev,
3391                                         "TX DMA page map failed\n");
3392                                 adapter->tx_dma_failed++;
3393                                 return -1;
3394                         }
3395
3396                         buffer_info->next_to_watch = i;
3397
3398                         len -= size;
3399                         offset += size;
3400                         count++;
3401
3402                         i++;
3403                         if (i == tx_ring->count)
3404                                 i = 0;
3405                 }
3406         }
3407
3408         if (i == 0)
3409                 i = tx_ring->count - 1;
3410         else
3411                 i--;
3412
3413         tx_ring->buffer_info[i].skb = skb;
3414         tx_ring->buffer_info[first].next_to_watch = i;
3415
3416         return count;
3417 }
3418
3419 static void e1000_tx_queue(struct e1000_adapter *adapter,
3420                            int tx_flags, int count)
3421 {
3422         struct e1000_ring *tx_ring = adapter->tx_ring;
3423         struct e1000_tx_desc *tx_desc = NULL;
3424         struct e1000_buffer *buffer_info;
3425         u32 txd_upper = 0, txd_lower = E1000_TXD_CMD_IFCS;
3426         unsigned int i;
3427
3428         if (tx_flags & E1000_TX_FLAGS_TSO) {
3429                 txd_lower |= E1000_TXD_CMD_DEXT | E1000_TXD_DTYP_D |
3430                              E1000_TXD_CMD_TSE;
3431                 txd_upper |= E1000_TXD_POPTS_TXSM << 8;
3432
3433                 if (tx_flags & E1000_TX_FLAGS_IPV4)
3434                         txd_upper |= E1000_TXD_POPTS_IXSM << 8;
3435         }
3436
3437         if (tx_flags & E1000_TX_FLAGS_CSUM) {
3438                 txd_lower |= E1000_TXD_CMD_DEXT | E1000_TXD_DTYP_D;
3439                 txd_upper |= E1000_TXD_POPTS_TXSM << 8;
3440         }
3441
3442         if (tx_flags & E1000_TX_FLAGS_VLAN) {
3443                 txd_lower |= E1000_TXD_CMD_VLE;
3444                 txd_upper |= (tx_flags & E1000_TX_FLAGS_VLAN_MASK);
3445         }
3446
3447         i = tx_ring->next_to_use;
3448
3449         while (count--) {
3450                 buffer_info = &tx_ring->buffer_info[i];
3451                 tx_desc = E1000_TX_DESC(*tx_ring, i);
3452                 tx_desc->buffer_addr = cpu_to_le64(buffer_info->dma);
3453                 tx_desc->lower.data =
3454                         cpu_to_le32(txd_lower | buffer_info->length);
3455                 tx_desc->upper.data = cpu_to_le32(txd_upper);
3456
3457                 i++;
3458                 if (i == tx_ring->count)
3459                         i = 0;
3460         }
3461
3462         tx_desc->lower.data |= cpu_to_le32(adapter->txd_cmd);
3463
3464         /*
3465          * Force memory writes to complete before letting h/w
3466          * know there are new descriptors to fetch.  (Only
3467          * applicable for weak-ordered memory model archs,
3468          * such as IA-64).
3469          */
3470         wmb();
3471
3472         tx_ring->next_to_use = i;
3473         writel(i, adapter->hw.hw_addr + tx_ring->tail);
3474         /*
3475          * we need this if more than one processor can write to our tail
3476          * at a time, it synchronizes IO on IA64/Altix systems
3477          */
3478         mmiowb();
3479 }
3480
3481 #define MINIMUM_DHCP_PACKET_SIZE 282
3482 static int e1000_transfer_dhcp_info(struct e1000_adapter *adapter,
3483                                     struct sk_buff *skb)
3484 {
3485         struct e1000_hw *hw =  &adapter->hw;
3486         u16 length, offset;
3487
3488         if (vlan_tx_tag_present(skb)) {
3489                 if (!((vlan_tx_tag_get(skb) == adapter->hw.mng_cookie.vlan_id)
3490                     && (adapter->hw.mng_cookie.status &
3491                         E1000_MNG_DHCP_COOKIE_STATUS_VLAN)))
3492                         return 0;
3493         }
3494
3495         if (skb->len <= MINIMUM_DHCP_PACKET_SIZE)
3496                 return 0;
3497
3498         if (((struct ethhdr *) skb->data)->h_proto != htons(ETH_P_IP))
3499                 return 0;
3500
3501         {
3502                 const struct iphdr *ip = (struct iphdr *)((u8 *)skb->data+14);
3503                 struct udphdr *udp;
3504
3505                 if (ip->protocol != IPPROTO_UDP)
3506                         return 0;
3507
3508                 udp = (struct udphdr *)((u8 *)ip + (ip->ihl << 2));
3509                 if (ntohs(udp->dest) != 67)
3510                         return 0;
3511
3512                 offset = (u8 *)udp + 8 - skb->data;
3513                 length = skb->len - offset;
3514                 return e1000e_mng_write_dhcp_info(hw, (u8 *)udp + 8, length);
3515         }
3516
3517         return 0;
3518 }
3519
3520 static int __e1000_maybe_stop_tx(struct net_device *netdev, int size)
3521 {
3522         struct e1000_adapter *adapter = netdev_priv(netdev);
3523
3524         netif_stop_queue(netdev);
3525         /*
3526          * Herbert's original patch had:
3527          *  smp_mb__after_netif_stop_queue();
3528          * but since that doesn't exist yet, just open code it.
3529          */
3530         smp_mb();
3531
3532         /*
3533          * We need to check again in a case another CPU has just
3534          * made room available.
3535          */
3536         if (e1000_desc_unused(adapter->tx_ring) < size)
3537                 return -EBUSY;
3538
3539         /* A reprieve! */
3540         netif_start_queue(netdev);
3541         ++adapter->restart_queue;
3542         return 0;
3543 }
3544
3545 static int e1000_maybe_stop_tx(struct net_device *netdev, int size)
3546 {
3547         struct e1000_adapter *adapter = netdev_priv(netdev);
3548
3549         if (e1000_desc_unused(adapter->tx_ring) >= size)
3550                 return 0;
3551         return __e1000_maybe_stop_tx(netdev, size);
3552 }
3553
3554 #define TXD_USE_COUNT(S, X) (((S) >> (X)) + 1 )
3555 static int e1000_xmit_frame(struct sk_buff *skb, struct net_device *netdev)
3556 {
3557         struct e1000_adapter *adapter = netdev_priv(netdev);
3558         struct e1000_ring *tx_ring = adapter->tx_ring;
3559         unsigned int first;
3560         unsigned int max_per_txd = E1000_MAX_PER_TXD;
3561         unsigned int max_txd_pwr = E1000_MAX_TXD_PWR;
3562         unsigned int tx_flags = 0;
3563         unsigned int len = skb->len - skb->data_len;
3564         unsigned long irq_flags;
3565         unsigned int nr_frags;
3566         unsigned int mss;
3567         int count = 0;
3568         int tso;
3569         unsigned int f;
3570
3571         if (test_bit(__E1000_DOWN, &adapter->state)) {
3572                 dev_kfree_skb_any(skb);
3573                 return NETDEV_TX_OK;
3574         }
3575
3576         if (skb->len <= 0) {
3577                 dev_kfree_skb_any(skb);
3578                 return NETDEV_TX_OK;
3579         }
3580
3581         mss = skb_shinfo(skb)->gso_size;
3582         /*
3583          * The controller does a simple calculation to
3584          * make sure there is enough room in the FIFO before
3585          * initiating the DMA for each buffer.  The calc is:
3586          * 4 = ceil(buffer len/mss).  To make sure we don't
3587          * overrun the FIFO, adjust the max buffer len if mss
3588          * drops.
3589          */
3590         if (mss) {
3591                 u8 hdr_len;
3592                 max_per_txd = min(mss << 2, max_per_txd);
3593                 max_txd_pwr = fls(max_per_txd) - 1;
3594
3595                 /*
3596                  * TSO Workaround for 82571/2/3 Controllers -- if skb->data
3597                  * points to just header, pull a few bytes of payload from
3598                  * frags into skb->data
3599                  */
3600                 hdr_len = skb_transport_offset(skb) + tcp_hdrlen(skb);
3601                 /*
3602                  * we do this workaround for ES2LAN, but it is un-necessary,
3603                  * avoiding it could save a lot of cycles
3604                  */
3605                 if (skb->data_len && (hdr_len == len)) {
3606                         unsigned int pull_size;
3607
3608                         pull_size = min((unsigned int)4, skb->data_len);
3609                         if (!__pskb_pull_tail(skb, pull_size)) {
3610                                 ndev_err(netdev,
3611                                          "__pskb_pull_tail failed.\n");
3612                                 dev_kfree_skb_any(skb);
3613                                 return NETDEV_TX_OK;
3614                         }
3615                         len = skb->len - skb->data_len;
3616                 }
3617         }
3618
3619         /* reserve a descriptor for the offload context */
3620         if ((mss) || (skb->ip_summed == CHECKSUM_PARTIAL))
3621                 count++;
3622         count++;
3623
3624         count += TXD_USE_COUNT(len, max_txd_pwr);
3625
3626         nr_frags = skb_shinfo(skb)->nr_frags;
3627         for (f = 0; f < nr_frags; f++)
3628                 count += TXD_USE_COUNT(skb_shinfo(skb)->frags[f].size,
3629                                        max_txd_pwr);
3630
3631         if (adapter->hw.mac.tx_pkt_filtering)
3632                 e1000_transfer_dhcp_info(adapter, skb);
3633
3634         if (!spin_trylock_irqsave(&adapter->tx_queue_lock, irq_flags))
3635                 /* Collision - tell upper layer to requeue */
3636                 return NETDEV_TX_LOCKED;
3637
3638         /*
3639          * need: count + 2 desc gap to keep tail from touching
3640          * head, otherwise try next time
3641          */
3642         if (e1000_maybe_stop_tx(netdev, count + 2)) {
3643                 spin_unlock_irqrestore(&adapter->tx_queue_lock, irq_flags);
3644                 return NETDEV_TX_BUSY;
3645         }
3646
3647         if (adapter->vlgrp && vlan_tx_tag_present(skb)) {
3648                 tx_flags |= E1000_TX_FLAGS_VLAN;
3649                 tx_flags |= (vlan_tx_tag_get(skb) << E1000_TX_FLAGS_VLAN_SHIFT);
3650         }
3651
3652         first = tx_ring->next_to_use;
3653
3654         tso = e1000_tso(adapter, skb);
3655         if (tso < 0) {
3656                 dev_kfree_skb_any(skb);
3657                 spin_unlock_irqrestore(&adapter->tx_queue_lock, irq_flags);
3658                 return NETDEV_TX_OK;
3659         }
3660
3661         if (tso)
3662                 tx_flags |= E1000_TX_FLAGS_TSO;
3663         else if (e1000_tx_csum(adapter, skb))
3664                 tx_flags |= E1000_TX_FLAGS_CSUM;
3665
3666         /*
3667          * Old method was to assume IPv4 packet by default if TSO was enabled.
3668          * 82571 hardware supports TSO capabilities for IPv6 as well...
3669          * no longer assume, we must.
3670          */
3671         if (skb->protocol == htons(ETH_P_IP))
3672                 tx_flags |= E1000_TX_FLAGS_IPV4;
3673
3674         count = e1000_tx_map(adapter, skb, first, max_per_txd, nr_frags, mss);
3675         if (count < 0) {
3676                 /* handle pci_map_single() error in e1000_tx_map */
3677                 dev_kfree_skb_any(skb);
3678                 spin_unlock_irqrestore(&adapter->tx_queue_lock, irq_flags);
3679                 return NETDEV_TX_OK;
3680         }
3681
3682         e1000_tx_queue(adapter, tx_flags, count);
3683
3684         netdev->trans_start = jiffies;
3685
3686         /* Make sure there is space in the ring for the next send. */
3687         e1000_maybe_stop_tx(netdev, MAX_SKB_FRAGS + 2);
3688
3689         spin_unlock_irqrestore(&adapter->tx_queue_lock, irq_flags);
3690         return NETDEV_TX_OK;
3691 }
3692
3693 /**
3694  * e1000_tx_timeout - Respond to a Tx Hang
3695  * @netdev: network interface device structure
3696  **/
3697 static void e1000_tx_timeout(struct net_device *netdev)
3698 {
3699         struct e1000_adapter *adapter = netdev_priv(netdev);
3700
3701         /* Do the reset outside of interrupt context */
3702         adapter->tx_timeout_count++;
3703         schedule_work(&adapter->reset_task);
3704 }
3705
3706 static void e1000_reset_task(struct work_struct *work)
3707 {
3708         struct e1000_adapter *adapter;
3709         adapter = container_of(work, struct e1000_adapter, reset_task);
3710
3711         e1000e_reinit_locked(adapter);
3712 }
3713
3714 /**
3715  * e1000_get_stats - Get System Network Statistics
3716  * @netdev: network interface device structure
3717  *
3718  * Returns the address of the device statistics structure.
3719  * The statistics are actually updated from the timer callback.
3720  **/
3721 static struct net_device_stats *e1000_get_stats(struct net_device *netdev)
3722 {
3723         struct e1000_adapter *adapter = netdev_priv(netdev);
3724
3725         /* only return the current stats */
3726         return &adapter->net_stats;
3727 }
3728
3729 /**
3730  * e1000_change_mtu - Change the Maximum Transfer Unit
3731  * @netdev: network interface device structure
3732  * @new_mtu: new value for maximum frame size
3733  *
3734  * Returns 0 on success, negative on failure
3735  **/
3736 static int e1000_change_mtu(struct net_device *netdev, int new_mtu)
3737 {
3738         struct e1000_adapter *adapter = netdev_priv(netdev);
3739         int max_frame = new_mtu + ETH_HLEN + ETH_FCS_LEN;
3740
3741         if ((max_frame < ETH_ZLEN + ETH_FCS_LEN) ||
3742             (max_frame > MAX_JUMBO_FRAME_SIZE)) {
3743                 ndev_err(netdev, "Invalid MTU setting\n");
3744                 return -EINVAL;
3745         }
3746
3747         /* Jumbo frame size limits */
3748         if (max_frame > ETH_FRAME_LEN + ETH_FCS_LEN) {
3749                 if (!(adapter->flags & FLAG_HAS_JUMBO_FRAMES)) {
3750                         ndev_err(netdev, "Jumbo Frames not supported.\n");
3751                         return -EINVAL;
3752                 }
3753                 if (adapter->hw.phy.type == e1000_phy_ife) {
3754                         ndev_err(netdev, "Jumbo Frames not supported.\n");
3755                         return -EINVAL;
3756                 }
3757         }
3758
3759 #define MAX_STD_JUMBO_FRAME_SIZE 9234
3760         if (max_frame > MAX_STD_JUMBO_FRAME_SIZE) {
3761                 ndev_err(netdev, "MTU > 9216 not supported.\n");
3762                 return -EINVAL;
3763         }
3764
3765         while (test_and_set_bit(__E1000_RESETTING, &adapter->state))
3766                 msleep(1);
3767         /* e1000e_down has a dependency on max_frame_size */
3768         adapter->max_frame_size = max_frame;
3769         if (netif_running(netdev))
3770                 e1000e_down(adapter);
3771
3772         /*
3773          * NOTE: netdev_alloc_skb reserves 16 bytes, and typically NET_IP_ALIGN
3774          * means we reserve 2 more, this pushes us to allocate from the next
3775          * larger slab size.
3776          * i.e. RXBUFFER_2048 --> size-4096 slab
3777          * However with the new *_jumbo_rx* routines, jumbo receives will use
3778          * fragmented skbs
3779          */
3780
3781         if (max_frame <= 256)
3782                 adapter->rx_buffer_len = 256;
3783         else if (max_frame <= 512)
3784                 adapter->rx_buffer_len = 512;
3785         else if (max_frame <= 1024)
3786                 adapter->rx_buffer_len = 1024;
3787         else if (max_frame <= 2048)
3788                 adapter->rx_buffer_len = 2048;
3789         else
3790                 adapter->rx_buffer_len = 4096;
3791
3792         /* adjust allocation if LPE protects us, and we aren't using SBP */
3793         if ((max_frame == ETH_FRAME_LEN + ETH_FCS_LEN) ||
3794              (max_frame == ETH_FRAME_LEN + VLAN_HLEN + ETH_FCS_LEN))
3795                 adapter->rx_buffer_len = ETH_FRAME_LEN + VLAN_HLEN
3796                                          + ETH_FCS_LEN;
3797
3798         ndev_info(netdev, "changing MTU from %d to %d\n",
3799                 netdev->mtu, new_mtu);
3800         netdev->mtu = new_mtu;
3801
3802         if (netif_running(netdev))
3803                 e1000e_up(adapter);
3804         else
3805                 e1000e_reset(adapter);
3806
3807         clear_bit(__E1000_RESETTING, &adapter->state);
3808
3809         return 0;
3810 }
3811
3812 static int e1000_mii_ioctl(struct net_device *netdev, struct ifreq *ifr,
3813                            int cmd)
3814 {
3815         struct e1000_adapter *adapter = netdev_priv(netdev);
3816         struct mii_ioctl_data *data = if_mii(ifr);
3817
3818         if (adapter->hw.phy.media_type != e1000_media_type_copper)
3819                 return -EOPNOTSUPP;
3820
3821         switch (cmd) {
3822         case SIOCGMIIPHY:
3823                 data->phy_id = adapter->hw.phy.addr;
3824                 break;
3825         case SIOCGMIIREG:
3826                 if (!capable(CAP_NET_ADMIN))
3827                         return -EPERM;
3828                 switch (data->reg_num & 0x1F) {
3829                 case MII_BMCR:
3830                         data->val_out = adapter->phy_regs.bmcr;
3831                         break;
3832                 case MII_BMSR:
3833                         data->val_out = adapter->phy_regs.bmsr;
3834                         break;
3835                 case MII_PHYSID1:
3836                         data->val_out = (adapter->hw.phy.id >> 16);
3837                         break;
3838                 case MII_PHYSID2:
3839                         data->val_out = (adapter->hw.phy.id & 0xFFFF);
3840                         break;
3841                 case MII_ADVERTISE:
3842                         data->val_out = adapter->phy_regs.advertise;
3843                         break;
3844                 case MII_LPA:
3845                         data->val_out = adapter->phy_regs.lpa;
3846                         break;
3847                 case MII_EXPANSION:
3848                         data->val_out = adapter->phy_regs.expansion;
3849                         break;
3850                 case MII_CTRL1000:
3851                         data->val_out = adapter->phy_regs.ctrl1000;
3852                         break;
3853                 case MII_STAT1000:
3854                         data->val_out = adapter->phy_regs.stat1000;
3855                         break;
3856                 case MII_ESTATUS:
3857                         data->val_out = adapter->phy_regs.estatus;
3858                         break;
3859                 default:
3860                         return -EIO;
3861                 }
3862                 break;
3863         case SIOCSMIIREG:
3864         default:
3865                 return -EOPNOTSUPP;
3866         }
3867         return 0;
3868 }
3869
3870 static int e1000_ioctl(struct net_device *netdev, struct ifreq *ifr, int cmd)
3871 {
3872         switch (cmd) {
3873         case SIOCGMIIPHY:
3874         case SIOCGMIIREG:
3875         case SIOCSMIIREG:
3876                 return e1000_mii_ioctl(netdev, ifr, cmd);
3877         default:
3878                 return -EOPNOTSUPP;
3879         }
3880 }
3881
3882 static int e1000_suspend(struct pci_dev *pdev, pm_message_t state)
3883 {
3884         struct net_device *netdev = pci_get_drvdata(pdev);
3885         struct e1000_adapter *adapter = netdev_priv(netdev);
3886         struct e1000_hw *hw = &adapter->hw;
3887         u32 ctrl, ctrl_ext, rctl, status;
3888         u32 wufc = adapter->wol;
3889         int retval = 0;
3890
3891         netif_device_detach(netdev);
3892
3893         if (netif_running(netdev)) {
3894                 WARN_ON(test_bit(__E1000_RESETTING, &adapter->state));
3895                 e1000e_down(adapter);
3896                 e1000_free_irq(adapter);
3897         }
3898
3899         retval = pci_save_state(pdev);
3900         if (retval)
3901                 return retval;
3902
3903         status = er32(STATUS);
3904         if (status & E1000_STATUS_LU)
3905                 wufc &= ~E1000_WUFC_LNKC;
3906
3907         if (wufc) {
3908                 e1000_setup_rctl(adapter);
3909                 e1000_set_multi(netdev);
3910
3911                 /* turn on all-multi mode if wake on multicast is enabled */
3912                 if (wufc & E1000_WUFC_MC) {
3913                         rctl = er32(RCTL);
3914                         rctl |= E1000_RCTL_MPE;
3915                         ew32(RCTL, rctl);
3916                 }
3917
3918                 ctrl = er32(CTRL);
3919                 /* advertise wake from D3Cold */
3920                 #define E1000_CTRL_ADVD3WUC 0x00100000
3921                 /* phy power management enable */
3922                 #define E1000_CTRL_EN_PHY_PWR_MGMT 0x00200000
3923                 ctrl |= E1000_CTRL_ADVD3WUC |
3924                         E1000_CTRL_EN_PHY_PWR_MGMT;
3925                 ew32(CTRL, ctrl);
3926
3927                 if (adapter->hw.phy.media_type == e1000_media_type_fiber ||
3928                     adapter->hw.phy.media_type ==
3929                     e1000_media_type_internal_serdes) {
3930                         /* keep the laser running in D3 */
3931                         ctrl_ext = er32(CTRL_EXT);
3932                         ctrl_ext |= E1000_CTRL_EXT_SDP7_DATA;
3933                         ew32(CTRL_EXT, ctrl_ext);
3934                 }
3935
3936                 if (adapter->flags & FLAG_IS_ICH)
3937                         e1000e_disable_gig_wol_ich8lan(&adapter->hw);
3938
3939                 /* Allow time for pending master requests to run */
3940                 e1000e_disable_pcie_master(&adapter->hw);
3941
3942                 ew32(WUC, E1000_WUC_PME_EN);
3943                 ew32(WUFC, wufc);
3944                 pci_enable_wake(pdev, PCI_D3hot, 1);
3945                 pci_enable_wake(pdev, PCI_D3cold, 1);
3946         } else {
3947                 ew32(WUC, 0);
3948                 ew32(WUFC, 0);
3949                 pci_enable_wake(pdev, PCI_D3hot, 0);
3950                 pci_enable_wake(pdev, PCI_D3cold, 0);
3951         }
3952
3953         /* make sure adapter isn't asleep if manageability is enabled */
3954         if (adapter->flags & FLAG_MNG_PT_ENABLED) {
3955                 pci_enable_wake(pdev, PCI_D3hot, 1);
3956                 pci_enable_wake(pdev, PCI_D3cold, 1);
3957         }
3958
3959         if (adapter->hw.phy.type == e1000_phy_igp_3)
3960                 e1000e_igp3_phy_powerdown_workaround_ich8lan(&adapter->hw);
3961
3962         /*
3963          * Release control of h/w to f/w.  If f/w is AMT enabled, this
3964          * would have already happened in close and is redundant.
3965          */
3966         e1000_release_hw_control(adapter);
3967
3968         pci_disable_device(pdev);
3969
3970         pci_set_power_state(pdev, pci_choose_state(pdev, state));
3971
3972         return 0;
3973 }
3974
3975 static void e1000e_disable_l1aspm(struct pci_dev *pdev)
3976 {
3977         int pos;
3978         u16 val;
3979
3980         /*
3981          * 82573 workaround - disable L1 ASPM on mobile chipsets
3982          *
3983          * L1 ASPM on various mobile (ich7) chipsets do not behave properly
3984          * resulting in lost data or garbage information on the pci-e link
3985          * level. This could result in (false) bad EEPROM checksum errors,
3986          * long ping times (up to 2s) or even a system freeze/hang.
3987          *
3988          * Unfortunately this feature saves about 1W power consumption when
3989          * active.
3990          */
3991         pos = pci_find_capability(pdev, PCI_CAP_ID_EXP);
3992         pci_read_config_word(pdev, pos + PCI_EXP_LNKCTL, &val);
3993         if (val & 0x2) {
3994                 dev_warn(&pdev->dev, "Disabling L1 ASPM\n");
3995                 val &= ~0x2;
3996                 pci_write_config_word(pdev, pos + PCI_EXP_LNKCTL, val);
3997         }
3998 }
3999
4000 #ifdef CONFIG_PM
4001 static int e1000_resume(struct pci_dev *pdev)
4002 {
4003         struct net_device *netdev = pci_get_drvdata(pdev);
4004         struct e1000_adapter *adapter = netdev_priv(netdev);
4005         struct e1000_hw *hw = &adapter->hw;
4006         u32 err;
4007
4008         pci_set_power_state(pdev, PCI_D0);
4009         pci_restore_state(pdev);
4010         e1000e_disable_l1aspm(pdev);
4011
4012         if (adapter->need_ioport)
4013                 err = pci_enable_device(pdev);
4014         else
4015                 err = pci_enable_device_mem(pdev);
4016         if (err) {
4017                 dev_err(&pdev->dev,
4018                         "Cannot enable PCI device from suspend\n");
4019                 return err;
4020         }
4021
4022         pci_set_master(pdev);
4023
4024         pci_enable_wake(pdev, PCI_D3hot, 0);
4025         pci_enable_wake(pdev, PCI_D3cold, 0);
4026
4027         if (netif_running(netdev)) {
4028                 err = e1000_request_irq(adapter);
4029                 if (err)
4030                         return err;
4031         }
4032
4033         e1000e_power_up_phy(adapter);
4034         e1000e_reset(adapter);
4035         ew32(WUS, ~0);
4036
4037         e1000_init_manageability(adapter);
4038
4039         if (netif_running(netdev))
4040                 e1000e_up(adapter);
4041
4042         netif_device_attach(netdev);
4043
4044         /*
4045          * If the controller has AMT, do not set DRV_LOAD until the interface
4046          * is up.  For all other cases, let the f/w know that the h/w is now
4047          * under the control of the driver.
4048          */
4049         if (!(adapter->flags & FLAG_HAS_AMT) || !e1000e_check_mng_mode(&adapter->hw))
4050                 e1000_get_hw_control(adapter);
4051
4052         return 0;
4053 }
4054 #endif
4055
4056 static void e1000_shutdown(struct pci_dev *pdev)
4057 {
4058         e1000_suspend(pdev, PMSG_SUSPEND);
4059 }
4060
4061 #ifdef CONFIG_NET_POLL_CONTROLLER
4062 /*
4063  * Polling 'interrupt' - used by things like netconsole to send skbs
4064  * without having to re-enable interrupts. It's not called while
4065  * the interrupt routine is executing.
4066  */
4067 static void e1000_netpoll(struct net_device *netdev)
4068 {
4069         struct e1000_adapter *adapter = netdev_priv(netdev);
4070
4071         disable_irq(adapter->pdev->irq);
4072         e1000_intr(adapter->pdev->irq, netdev);
4073
4074         e1000_clean_tx_irq(adapter);
4075
4076         enable_irq(adapter->pdev->irq);
4077 }
4078 #endif
4079
4080 /**
4081  * e1000_io_error_detected - called when PCI error is detected
4082  * @pdev: Pointer to PCI device
4083  * @state: The current pci connection state
4084  *
4085  * This function is called after a PCI bus error affecting
4086  * this device has been detected.
4087  */
4088 static pci_ers_result_t e1000_io_error_detected(struct pci_dev *pdev,
4089                                                 pci_channel_state_t state)
4090 {
4091         struct net_device *netdev = pci_get_drvdata(pdev);
4092         struct e1000_adapter *adapter = netdev_priv(netdev);
4093
4094         netif_device_detach(netdev);
4095
4096         if (netif_running(netdev))
4097                 e1000e_down(adapter);
4098         pci_disable_device(pdev);
4099
4100         /* Request a slot slot reset. */
4101         return PCI_ERS_RESULT_NEED_RESET;
4102 }
4103
4104 /**
4105  * e1000_io_slot_reset - called after the pci bus has been reset.
4106  * @pdev: Pointer to PCI device
4107  *
4108  * Restart the card from scratch, as if from a cold-boot. Implementation
4109  * resembles the first-half of the e1000_resume routine.
4110  */
4111 static pci_ers_result_t e1000_io_slot_reset(struct pci_dev *pdev)
4112 {
4113         struct net_device *netdev = pci_get_drvdata(pdev);
4114         struct e1000_adapter *adapter = netdev_priv(netdev);
4115         struct e1000_hw *hw = &adapter->hw;
4116         int err;
4117
4118         e1000e_disable_l1aspm(pdev);
4119         if (adapter->need_ioport)
4120                 err = pci_enable_device(pdev);
4121         else
4122                 err = pci_enable_device_mem(pdev);
4123         if (err) {
4124                 dev_err(&pdev->dev,
4125                         "Cannot re-enable PCI device after reset.\n");
4126                 return PCI_ERS_RESULT_DISCONNECT;
4127         }
4128         pci_set_master(pdev);
4129         pci_restore_state(pdev);
4130
4131         pci_enable_wake(pdev, PCI_D3hot, 0);
4132         pci_enable_wake(pdev, PCI_D3cold, 0);
4133
4134         e1000e_reset(adapter);
4135         ew32(WUS, ~0);
4136
4137         return PCI_ERS_RESULT_RECOVERED;
4138 }
4139
4140 /**
4141  * e1000_io_resume - called when traffic can start flowing again.
4142  * @pdev: Pointer to PCI device
4143  *
4144  * This callback is called when the error recovery driver tells us that
4145  * its OK to resume normal operation. Implementation resembles the
4146  * second-half of the e1000_resume routine.
4147  */
4148 static void e1000_io_resume(struct pci_dev *pdev)
4149 {
4150         struct net_device *netdev = pci_get_drvdata(pdev);
4151         struct e1000_adapter *adapter = netdev_priv(netdev);
4152
4153         e1000_init_manageability(adapter);
4154
4155         if (netif_running(netdev)) {
4156                 if (e1000e_up(adapter)) {
4157                         dev_err(&pdev->dev,
4158                                 "can't bring device back up after reset\n");
4159                         return;
4160                 }
4161         }
4162
4163         netif_device_attach(netdev);
4164
4165         /*
4166          * If the controller has AMT, do not set DRV_LOAD until the interface
4167          * is up.  For all other cases, let the f/w know that the h/w is now
4168          * under the control of the driver.
4169          */
4170         if (!(adapter->flags & FLAG_HAS_AMT) ||
4171             !e1000e_check_mng_mode(&adapter->hw))
4172                 e1000_get_hw_control(adapter);
4173
4174 }
4175
4176 static void e1000_print_device_info(struct e1000_adapter *adapter)
4177 {
4178         struct e1000_hw *hw = &adapter->hw;
4179         struct net_device *netdev = adapter->netdev;
4180         u32 pba_num;
4181
4182         /* print bus type/speed/width info */
4183         ndev_info(netdev, "(PCI Express:2.5GB/s:%s) "
4184                   "%02x:%02x:%02x:%02x:%02x:%02x\n",
4185                   /* bus width */
4186                  ((hw->bus.width == e1000_bus_width_pcie_x4) ? "Width x4" :
4187                   "Width x1"),
4188                   /* MAC address */
4189                   netdev->dev_addr[0], netdev->dev_addr[1],
4190                   netdev->dev_addr[2], netdev->dev_addr[3],
4191                   netdev->dev_addr[4], netdev->dev_addr[5]);
4192         ndev_info(netdev, "Intel(R) PRO/%s Network Connection\n",
4193                   (hw->phy.type == e1000_phy_ife)
4194                    ? "10/100" : "1000");
4195         e1000e_read_pba_num(hw, &pba_num);
4196         ndev_info(netdev, "MAC: %d, PHY: %d, PBA No: %06x-%03x\n",
4197                   hw->mac.type, hw->phy.type,
4198                   (pba_num >> 8), (pba_num & 0xff));
4199 }
4200
4201 /**
4202  * e1000e_is_need_ioport - determine if an adapter needs ioport resources or not
4203  * @pdev: PCI device information struct
4204  *
4205  * Returns true if an adapters needs ioport resources
4206  **/
4207 static int e1000e_is_need_ioport(struct pci_dev *pdev)
4208 {
4209         switch (pdev->device) {
4210         /* Currently there are no adapters that need ioport resources */
4211         default:
4212                 return false;
4213         }
4214 }
4215
4216 /**
4217  * e1000_probe - Device Initialization Routine
4218  * @pdev: PCI device information struct
4219  * @ent: entry in e1000_pci_tbl
4220  *
4221  * Returns 0 on success, negative on failure
4222  *
4223  * e1000_probe initializes an adapter identified by a pci_dev structure.
4224  * The OS initialization, configuring of the adapter private structure,
4225  * and a hardware reset occur.
4226  **/
4227 static int __devinit e1000_probe(struct pci_dev *pdev,
4228                                  const struct pci_device_id *ent)
4229 {
4230         struct net_device *netdev;
4231         struct e1000_adapter *adapter;
4232         struct e1000_hw *hw;
4233         const struct e1000_info *ei = e1000_info_tbl[ent->driver_data];
4234         resource_size_t mmio_start, mmio_len;
4235         resource_size_t flash_start, flash_len;
4236
4237         static int cards_found;
4238         int i, err, pci_using_dac;
4239         u16 eeprom_data = 0;
4240         u16 eeprom_apme_mask = E1000_EEPROM_APME;
4241         int bars, need_ioport;
4242
4243         e1000e_disable_l1aspm(pdev);
4244
4245         /* do not allocate ioport bars when not needed */
4246         need_ioport = e1000e_is_need_ioport(pdev);
4247         if (need_ioport) {
4248                 bars = pci_select_bars(pdev, IORESOURCE_MEM | IORESOURCE_IO);
4249                 err = pci_enable_device(pdev);
4250         } else {
4251                 bars = pci_select_bars(pdev, IORESOURCE_MEM);
4252                 err = pci_enable_device_mem(pdev);
4253         }
4254         if (err)
4255                 return err;
4256
4257         pci_using_dac = 0;
4258         err = pci_set_dma_mask(pdev, DMA_64BIT_MASK);
4259         if (!err) {
4260                 err = pci_set_consistent_dma_mask(pdev, DMA_64BIT_MASK);
4261                 if (!err)
4262                         pci_using_dac = 1;
4263         } else {
4264                 err = pci_set_dma_mask(pdev, DMA_32BIT_MASK);
4265                 if (err) {
4266                         err = pci_set_consistent_dma_mask(pdev,
4267                                                           DMA_32BIT_MASK);
4268                         if (err) {
4269                                 dev_err(&pdev->dev, "No usable DMA "
4270                                         "configuration, aborting\n");
4271                                 goto err_dma;
4272                         }
4273                 }
4274         }
4275
4276         err = pci_request_selected_regions(pdev, bars, e1000e_driver_name);
4277         if (err)
4278                 goto err_pci_reg;
4279
4280         pci_set_master(pdev);
4281         pci_save_state(pdev);
4282
4283         err = -ENOMEM;
4284         netdev = alloc_etherdev(sizeof(struct e1000_adapter));
4285         if (!netdev)
4286                 goto err_alloc_etherdev;
4287
4288         SET_NETDEV_DEV(netdev, &pdev->dev);
4289
4290         pci_set_drvdata(pdev, netdev);
4291         adapter = netdev_priv(netdev);
4292         hw = &adapter->hw;
4293         adapter->netdev = netdev;
4294         adapter->pdev = pdev;
4295         adapter->ei = ei;
4296         adapter->pba = ei->pba;
4297         adapter->flags = ei->flags;
4298         adapter->hw.adapter = adapter;
4299         adapter->hw.mac.type = ei->mac;
4300         adapter->msg_enable = (1 << NETIF_MSG_DRV | NETIF_MSG_PROBE) - 1;
4301         adapter->bars = bars;
4302         adapter->need_ioport = need_ioport;
4303
4304         mmio_start = pci_resource_start(pdev, 0);
4305         mmio_len = pci_resource_len(pdev, 0);
4306
4307         err = -EIO;
4308         adapter->hw.hw_addr = ioremap(mmio_start, mmio_len);
4309         if (!adapter->hw.hw_addr)
4310                 goto err_ioremap;
4311
4312         if ((adapter->flags & FLAG_HAS_FLASH) &&
4313             (pci_resource_flags(pdev, 1) & IORESOURCE_MEM)) {
4314                 flash_start = pci_resource_start(pdev, 1);
4315                 flash_len = pci_resource_len(pdev, 1);
4316                 adapter->hw.flash_address = ioremap(flash_start, flash_len);
4317                 if (!adapter->hw.flash_address)
4318                         goto err_flashmap;
4319         }
4320
4321         /* construct the net_device struct */
4322         netdev->open                    = &e1000_open;
4323         netdev->stop                    = &e1000_close;
4324         netdev->hard_start_xmit         = &e1000_xmit_frame;
4325         netdev->get_stats               = &e1000_get_stats;
4326         netdev->set_multicast_list      = &e1000_set_multi;
4327         netdev->set_mac_address         = &e1000_set_mac;
4328         netdev->change_mtu              = &e1000_change_mtu;
4329         netdev->do_ioctl                = &e1000_ioctl;
4330         e1000e_set_ethtool_ops(netdev);
4331         netdev->tx_timeout              = &e1000_tx_timeout;
4332         netdev->watchdog_timeo          = 5 * HZ;
4333         netif_napi_add(netdev, &adapter->napi, e1000_clean, 64);
4334         netdev->vlan_rx_register        = e1000_vlan_rx_register;
4335         netdev->vlan_rx_add_vid         = e1000_vlan_rx_add_vid;
4336         netdev->vlan_rx_kill_vid        = e1000_vlan_rx_kill_vid;
4337 #ifdef CONFIG_NET_POLL_CONTROLLER
4338         netdev->poll_controller         = e1000_netpoll;
4339 #endif
4340         strncpy(netdev->name, pci_name(pdev), sizeof(netdev->name) - 1);
4341
4342         netdev->mem_start = mmio_start;
4343         netdev->mem_end = mmio_start + mmio_len;
4344
4345         adapter->bd_number = cards_found++;
4346
4347         /* setup adapter struct */
4348         err = e1000_sw_init(adapter);
4349         if (err)
4350                 goto err_sw_init;
4351
4352         err = -EIO;
4353
4354         memcpy(&hw->mac.ops, ei->mac_ops, sizeof(hw->mac.ops));
4355         memcpy(&hw->nvm.ops, ei->nvm_ops, sizeof(hw->nvm.ops));
4356         memcpy(&hw->phy.ops, ei->phy_ops, sizeof(hw->phy.ops));
4357
4358         err = ei->get_variants(adapter);
4359         if (err)
4360                 goto err_hw_init;
4361
4362         hw->mac.ops.get_bus_info(&adapter->hw);
4363
4364         adapter->hw.phy.autoneg_wait_to_complete = 0;
4365
4366         /* Copper options */
4367         if (adapter->hw.phy.media_type == e1000_media_type_copper) {
4368                 adapter->hw.phy.mdix = AUTO_ALL_MODES;
4369                 adapter->hw.phy.disable_polarity_correction = 0;
4370                 adapter->hw.phy.ms_type = e1000_ms_hw_default;
4371         }
4372
4373         if (e1000_check_reset_block(&adapter->hw))
4374                 ndev_info(netdev,
4375                           "PHY reset is blocked due to SOL/IDER session.\n");
4376
4377         netdev->features = NETIF_F_SG |
4378                            NETIF_F_HW_CSUM |
4379                            NETIF_F_HW_VLAN_TX |
4380                            NETIF_F_HW_VLAN_RX;
4381
4382         if (adapter->flags & FLAG_HAS_HW_VLAN_FILTER)
4383                 netdev->features |= NETIF_F_HW_VLAN_FILTER;
4384
4385         netdev->features |= NETIF_F_TSO;
4386         netdev->features |= NETIF_F_TSO6;
4387
4388         netdev->vlan_features |= NETIF_F_TSO;
4389         netdev->vlan_features |= NETIF_F_TSO6;
4390         netdev->vlan_features |= NETIF_F_HW_CSUM;
4391         netdev->vlan_features |= NETIF_F_SG;
4392
4393         if (pci_using_dac)
4394                 netdev->features |= NETIF_F_HIGHDMA;
4395
4396         /*
4397          * We should not be using LLTX anymore, but we are still Tx faster with
4398          * it.
4399          */
4400         netdev->features |= NETIF_F_LLTX;
4401
4402         if (e1000e_enable_mng_pass_thru(&adapter->hw))
4403                 adapter->flags |= FLAG_MNG_PT_ENABLED;
4404
4405         /*
4406          * before reading the NVM, reset the controller to
4407          * put the device in a known good starting state
4408          */
4409         adapter->hw.mac.ops.reset_hw(&adapter->hw);
4410
4411         /*
4412          * systems with ASPM and others may see the checksum fail on the first
4413          * attempt. Let's give it a few tries
4414          */
4415         for (i = 0;; i++) {
4416                 if (e1000_validate_nvm_checksum(&adapter->hw) >= 0)
4417                         break;
4418                 if (i == 2) {
4419                         ndev_err(netdev, "The NVM Checksum Is Not Valid\n");
4420                         err = -EIO;
4421                         goto err_eeprom;
4422                 }
4423         }
4424
4425         /* copy the MAC address out of the NVM */
4426         if (e1000e_read_mac_addr(&adapter->hw))
4427                 ndev_err(netdev, "NVM Read Error while reading MAC address\n");
4428
4429         memcpy(netdev->dev_addr, adapter->hw.mac.addr, netdev->addr_len);
4430         memcpy(netdev->perm_addr, adapter->hw.mac.addr, netdev->addr_len);
4431
4432         if (!is_valid_ether_addr(netdev->perm_addr)) {
4433                 ndev_err(netdev, "Invalid MAC Address: "
4434                          "%02x:%02x:%02x:%02x:%02x:%02x\n",
4435                          netdev->perm_addr[0], netdev->perm_addr[1],
4436                          netdev->perm_addr[2], netdev->perm_addr[3],
4437                          netdev->perm_addr[4], netdev->perm_addr[5]);
4438                 err = -EIO;
4439                 goto err_eeprom;
4440         }
4441
4442         init_timer(&adapter->watchdog_timer);
4443         adapter->watchdog_timer.function = &e1000_watchdog;
4444         adapter->watchdog_timer.data = (unsigned long) adapter;
4445
4446         init_timer(&adapter->phy_info_timer);
4447         adapter->phy_info_timer.function = &e1000_update_phy_info;
4448         adapter->phy_info_timer.data = (unsigned long) adapter;
4449
4450         INIT_WORK(&adapter->reset_task, e1000_reset_task);
4451         INIT_WORK(&adapter->watchdog_task, e1000_watchdog_task);
4452
4453         e1000e_check_options(adapter);
4454
4455         /* Initialize link parameters. User can change them with ethtool */
4456         adapter->hw.mac.autoneg = 1;
4457         adapter->fc_autoneg = 1;
4458         adapter->hw.fc.original_type = e1000_fc_default;
4459         adapter->hw.fc.type = e1000_fc_default;
4460         adapter->hw.phy.autoneg_advertised = 0x2f;
4461
4462         /* ring size defaults */
4463         adapter->rx_ring->count = 256;
4464         adapter->tx_ring->count = 256;
4465
4466         /*
4467          * Initial Wake on LAN setting - If APM wake is enabled in
4468          * the EEPROM, enable the ACPI Magic Packet filter
4469          */
4470         if (adapter->flags & FLAG_APME_IN_WUC) {
4471                 /* APME bit in EEPROM is mapped to WUC.APME */
4472                 eeprom_data = er32(WUC);
4473                 eeprom_apme_mask = E1000_WUC_APME;
4474         } else if (adapter->flags & FLAG_APME_IN_CTRL3) {
4475                 if (adapter->flags & FLAG_APME_CHECK_PORT_B &&
4476                     (adapter->hw.bus.func == 1))
4477                         e1000_read_nvm(&adapter->hw,
4478                                 NVM_INIT_CONTROL3_PORT_B, 1, &eeprom_data);
4479                 else
4480                         e1000_read_nvm(&adapter->hw,
4481                                 NVM_INIT_CONTROL3_PORT_A, 1, &eeprom_data);
4482         }
4483
4484         /* fetch WoL from EEPROM */
4485         if (eeprom_data & eeprom_apme_mask)
4486                 adapter->eeprom_wol |= E1000_WUFC_MAG;
4487
4488         /*
4489          * now that we have the eeprom settings, apply the special cases
4490          * where the eeprom may be wrong or the board simply won't support
4491          * wake on lan on a particular port
4492          */
4493         if (!(adapter->flags & FLAG_HAS_WOL))
4494                 adapter->eeprom_wol = 0;
4495
4496         /* initialize the wol settings based on the eeprom settings */
4497         adapter->wol = adapter->eeprom_wol;
4498
4499         /* reset the hardware with the new settings */
4500         e1000e_reset(adapter);
4501
4502         /*
4503          * If the controller has AMT, do not set DRV_LOAD until the interface
4504          * is up.  For all other cases, let the f/w know that the h/w is now
4505          * under the control of the driver.
4506          */
4507         if (!(adapter->flags & FLAG_HAS_AMT) ||
4508             !e1000e_check_mng_mode(&adapter->hw))
4509                 e1000_get_hw_control(adapter);
4510
4511         /* tell the stack to leave us alone until e1000_open() is called */
4512         netif_carrier_off(netdev);
4513         netif_stop_queue(netdev);
4514
4515         strcpy(netdev->name, "eth%d");
4516         err = register_netdev(netdev);
4517         if (err)
4518                 goto err_register;
4519
4520         e1000_print_device_info(adapter);
4521
4522         return 0;
4523
4524 err_register:
4525 err_hw_init:
4526         e1000_release_hw_control(adapter);
4527 err_eeprom:
4528         if (!e1000_check_reset_block(&adapter->hw))
4529                 e1000_phy_hw_reset(&adapter->hw);
4530
4531         if (adapter->hw.flash_address)
4532                 iounmap(adapter->hw.flash_address);
4533
4534 err_flashmap:
4535         kfree(adapter->tx_ring);
4536         kfree(adapter->rx_ring);
4537 err_sw_init:
4538         iounmap(adapter->hw.hw_addr);
4539 err_ioremap:
4540         free_netdev(netdev);
4541 err_alloc_etherdev:
4542         pci_release_selected_regions(pdev, bars);
4543 err_pci_reg:
4544 err_dma:
4545         pci_disable_device(pdev);
4546         return err;
4547 }
4548
4549 /**
4550  * e1000_remove - Device Removal Routine
4551  * @pdev: PCI device information struct
4552  *
4553  * e1000_remove is called by the PCI subsystem to alert the driver
4554  * that it should release a PCI device.  The could be caused by a
4555  * Hot-Plug event, or because the driver is going to be removed from
4556  * memory.
4557  **/
4558 static void __devexit e1000_remove(struct pci_dev *pdev)
4559 {
4560         struct net_device *netdev = pci_get_drvdata(pdev);
4561         struct e1000_adapter *adapter = netdev_priv(netdev);
4562
4563         /*
4564          * flush_scheduled work may reschedule our watchdog task, so
4565          * explicitly disable watchdog tasks from being rescheduled
4566          */
4567         set_bit(__E1000_DOWN, &adapter->state);
4568         del_timer_sync(&adapter->watchdog_timer);
4569         del_timer_sync(&adapter->phy_info_timer);
4570
4571         flush_scheduled_work();
4572
4573         /*
4574          * Release control of h/w to f/w.  If f/w is AMT enabled, this
4575          * would have already happened in close and is redundant.
4576          */
4577         e1000_release_hw_control(adapter);
4578
4579         unregister_netdev(netdev);
4580
4581         if (!e1000_check_reset_block(&adapter->hw))
4582                 e1000_phy_hw_reset(&adapter->hw);
4583
4584         kfree(adapter->tx_ring);
4585         kfree(adapter->rx_ring);
4586
4587         iounmap(adapter->hw.hw_addr);
4588         if (adapter->hw.flash_address)
4589                 iounmap(adapter->hw.flash_address);
4590         pci_release_selected_regions(pdev, adapter->bars);
4591
4592         free_netdev(netdev);
4593
4594         pci_disable_device(pdev);
4595 }
4596
4597 /* PCI Error Recovery (ERS) */
4598 static struct pci_error_handlers e1000_err_handler = {
4599         .error_detected = e1000_io_error_detected,
4600         .slot_reset = e1000_io_slot_reset,
4601         .resume = e1000_io_resume,
4602 };
4603
4604 static struct pci_device_id e1000_pci_tbl[] = {
4605         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82571EB_COPPER), board_82571 },
4606         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82571EB_FIBER), board_82571 },
4607         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82571EB_QUAD_COPPER), board_82571 },
4608         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82571EB_QUAD_COPPER_LP), board_82571 },
4609         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82571EB_QUAD_FIBER), board_82571 },
4610         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82571EB_SERDES), board_82571 },
4611         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82571EB_SERDES_DUAL), board_82571 },
4612         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82571EB_SERDES_QUAD), board_82571 },
4613         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82571PT_QUAD_COPPER), board_82571 },
4614
4615         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82572EI), board_82572 },
4616         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82572EI_COPPER), board_82572 },
4617         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82572EI_FIBER), board_82572 },
4618         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82572EI_SERDES), board_82572 },
4619
4620         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82573E), board_82573 },
4621         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82573E_IAMT), board_82573 },
4622         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82573L), board_82573 },
4623
4624         { PCI_VDEVICE(INTEL, E1000_DEV_ID_80003ES2LAN_COPPER_DPT),
4625           board_80003es2lan },
4626         { PCI_VDEVICE(INTEL, E1000_DEV_ID_80003ES2LAN_COPPER_SPT),
4627           board_80003es2lan },
4628         { PCI_VDEVICE(INTEL, E1000_DEV_ID_80003ES2LAN_SERDES_DPT),
4629           board_80003es2lan },
4630         { PCI_VDEVICE(INTEL, E1000_DEV_ID_80003ES2LAN_SERDES_SPT),
4631           board_80003es2lan },
4632
4633         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH8_IFE), board_ich8lan },
4634         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH8_IFE_G), board_ich8lan },
4635         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH8_IFE_GT), board_ich8lan },
4636         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH8_IGP_AMT), board_ich8lan },
4637         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH8_IGP_C), board_ich8lan },
4638         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH8_IGP_M), board_ich8lan },
4639         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH8_IGP_M_AMT), board_ich8lan },
4640
4641         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH9_IFE), board_ich9lan },
4642         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH9_IFE_G), board_ich9lan },
4643         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH9_IFE_GT), board_ich9lan },
4644         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH9_IGP_AMT), board_ich9lan },
4645         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH9_IGP_C), board_ich9lan },
4646         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH9_IGP_M), board_ich9lan },
4647         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH9_IGP_M_AMT), board_ich9lan },
4648         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH9_IGP_M_V), board_ich9lan },
4649
4650         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH10_R_BM_LM), board_ich9lan },
4651         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH10_R_BM_LF), board_ich9lan },
4652         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH10_R_BM_V), board_ich9lan },
4653
4654         { }     /* terminate list */
4655 };
4656 MODULE_DEVICE_TABLE(pci, e1000_pci_tbl);
4657
4658 /* PCI Device API Driver */
4659 static struct pci_driver e1000_driver = {
4660         .name     = e1000e_driver_name,
4661         .id_table = e1000_pci_tbl,
4662         .probe    = e1000_probe,
4663         .remove   = __devexit_p(e1000_remove),
4664 #ifdef CONFIG_PM
4665         /* Power Management Hooks */
4666         .suspend  = e1000_suspend,
4667         .resume   = e1000_resume,
4668 #endif
4669         .shutdown = e1000_shutdown,
4670         .err_handler = &e1000_err_handler
4671 };
4672
4673 /**
4674  * e1000_init_module - Driver Registration Routine
4675  *
4676  * e1000_init_module is the first routine called when the driver is
4677  * loaded. All it does is register with the PCI subsystem.
4678  **/
4679 static int __init e1000_init_module(void)
4680 {
4681         int ret;
4682         printk(KERN_INFO "%s: Intel(R) PRO/1000 Network Driver - %s\n",
4683                e1000e_driver_name, e1000e_driver_version);
4684         printk(KERN_INFO "%s: Copyright (c) 1999-2008 Intel Corporation.\n",
4685                e1000e_driver_name);
4686         ret = pci_register_driver(&e1000_driver);
4687         pm_qos_add_requirement(PM_QOS_CPU_DMA_LATENCY, e1000e_driver_name,
4688                                PM_QOS_DEFAULT_VALUE);
4689
4690         return ret;
4691 }
4692 module_init(e1000_init_module);
4693
4694 /**
4695  * e1000_exit_module - Driver Exit Cleanup Routine
4696  *
4697  * e1000_exit_module is called just before the driver is removed
4698  * from memory.
4699  **/
4700 static void __exit e1000_exit_module(void)
4701 {
4702         pci_unregister_driver(&e1000_driver);
4703         pm_qos_remove_requirement(PM_QOS_CPU_DMA_LATENCY, e1000e_driver_name);
4704 }
4705 module_exit(e1000_exit_module);
4706
4707
4708 MODULE_AUTHOR("Intel Corporation, <linux.nics@intel.com>");
4709 MODULE_DESCRIPTION("Intel(R) PRO/1000 Network Driver");
4710 MODULE_LICENSE("GPL");
4711 MODULE_VERSION(DRV_VERSION);
4712
4713 /* e1000_main.c */