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_82574]           = &e1000_82574_info,
  59         [board_80003es2lan]     = &e1000_es2_info,
  60         [board_ich8lan]         = &e1000_ich8_info,
  61         [board_ich9lan]         = &e1000_ich9_info,
  62         [board_ich10lan]        = &e1000_ich10_info,
  63 };
  64
  65 #ifdef DEBUG
  66 /**
  67  * e1000_get_hw_dev_name - return device name string
  68  * used by hardware layer to print debugging information
  69  **/
  70 char *e1000e_get_hw_dev_name(struct e1000_hw *hw)
  71 {
  72         return hw->adapter->netdev->name;
  73 }
  74 #endif
  75
  76 /**
  77  * e1000_desc_unused - calculate if we have unused descriptors
  78  **/
  79 static int e1000_desc_unused(struct e1000_ring *ring)
  80 {
  81         if (ring->next_to_clean > ring->next_to_use)
  82                 return ring->next_to_clean - ring->next_to_use - 1;
  83
  84         return ring->count + ring->next_to_clean - ring->next_to_use - 1;
  85 }
  86
  87 /**
  88  * e1000_receive_skb - helper function to handle Rx indications
  89  * @adapter: board private structure
  90  * @status: descriptor status field as written by hardware
  91  * @vlan: descriptor vlan field as written by hardware (no le/be conversion)
  92  * @skb: pointer to sk_buff to be indicated to stack
  93  **/
  94 static void e1000_receive_skb(struct e1000_adapter *adapter,
  95                               struct net_device *netdev,
  96                               struct sk_buff *skb,
  97                               u8 status, __le16 vlan)
  98 {
  99         skb->protocol = eth_type_trans(skb, netdev);
 100
 101         if (adapter->vlgrp && (status & E1000_RXD_STAT_VP))
 102                 vlan_hwaccel_receive_skb(skb, adapter->vlgrp,
 103                                          le16_to_cpu(vlan));
 104         else
 105                 netif_receive_skb(skb);
 106
 107         netdev->last_rx = jiffies;
 108 }
 109
 110 /**
 111  * e1000_rx_checksum - Receive Checksum Offload for 82543
 112  * @adapter:     board private structure
 113  * @status_err:  receive descriptor status and error fields
 114  * @csum:       receive descriptor csum field
 115  * @sk_buff:     socket buffer with received data
 116  **/
 117 static void e1000_rx_checksum(struct e1000_adapter *adapter, u32 status_err,
 118                               u32 csum, struct sk_buff *skb)
 119 {
 120         u16 status = (u16)status_err;
 121         u8 errors = (u8)(status_err >> 24);
 122         skb->ip_summed = CHECKSUM_NONE;
 123
 124         /* Ignore Checksum bit is set */
 125         if (status & E1000_RXD_STAT_IXSM)
 126                 return;
 127         /* TCP/UDP checksum error bit is set */
 128         if (errors & E1000_RXD_ERR_TCPE) {
 129                 /* let the stack verify checksum errors */
 130                 adapter->hw_csum_err++;
 131                 return;
 132         }
 133
 134         /* TCP/UDP Checksum has not been calculated */
 135         if (!(status & (E1000_RXD_STAT_TCPCS | E1000_RXD_STAT_UDPCS)))
 136                 return;
 137
 138         /* It must be a TCP or UDP packet with a valid checksum */
 139         if (status & E1000_RXD_STAT_TCPCS) {
 140                 /* TCP checksum is good */
 141                 skb->ip_summed = CHECKSUM_UNNECESSARY;
 142         } else {
 143                 /*
 144                  * IP fragment with UDP payload
 145                  * Hardware complements the payload checksum, so we undo it
 146                  * and then put the value in host order for further stack use.
 147                  */
 148                 __sum16 sum = (__force __sum16)htons(csum);
 149                 skb->csum = csum_unfold(~sum);
 150                 skb->ip_summed = CHECKSUM_COMPLETE;
 151         }
 152         adapter->hw_csum_good++;
 153 }
 154
 155 /**
 156  * e1000_alloc_rx_buffers - Replace used receive buffers; legacy & extended
 157  * @adapter: address of board private structure
 158  **/
 159 static void e1000_alloc_rx_buffers(struct e1000_adapter *adapter,
 160                                    int cleaned_count)
 161 {
 162         struct net_device *netdev = adapter->netdev;
 163         struct pci_dev *pdev = adapter->pdev;
 164         struct e1000_ring *rx_ring = adapter->rx_ring;
 165         struct e1000_rx_desc *rx_desc;
 166         struct e1000_buffer *buffer_info;
 167         struct sk_buff *skb;
 168         unsigned int i;
 169         unsigned int bufsz = adapter->rx_buffer_len + NET_IP_ALIGN;
 170
 171         i = rx_ring->next_to_use;
 172         buffer_info = &rx_ring->buffer_info[i];
 173
 174         while (cleaned_count--) {
 175                 skb = buffer_info->skb;
 176                 if (skb) {
 177                         skb_trim(skb, 0);
 178                         goto map_skb;
 179                 }
 180
 181                 skb = netdev_alloc_skb(netdev, bufsz);
 182                 if (!skb) {
 183                         /* Better luck next round */
 184                         adapter->alloc_rx_buff_failed++;
 185                         break;
 186                 }
 187
 188                 /*
 189                  * Make buffer alignment 2 beyond a 16 byte boundary
 190                  * this will result in a 16 byte aligned IP header after
 191                  * the 14 byte MAC header is removed
 192                  */
 193                 skb_reserve(skb, NET_IP_ALIGN);
 194
 195                 buffer_info->skb = skb;
 196 map_skb:
 197                 buffer_info->dma = pci_map_single(pdev, skb->data,
 198                                                   adapter->rx_buffer_len,
 199                                                   PCI_DMA_FROMDEVICE);
 200                 if (pci_dma_mapping_error(pdev, buffer_info->dma)) {
 201                         dev_err(&pdev->dev, "RX DMA map failed\n");
 202                         adapter->rx_dma_failed++;
 203                         break;
 204                 }
 205
 206                 rx_desc = E1000_RX_DESC(*rx_ring, i);
 207                 rx_desc->buffer_addr = cpu_to_le64(buffer_info->dma);
 208
 209                 i++;
 210                 if (i == rx_ring->count)
 211                         i = 0;
 212                 buffer_info = &rx_ring->buffer_info[i];
 213         }
 214
 215         if (rx_ring->next_to_use != i) {
 216                 rx_ring->next_to_use = i;
 217                 if (i-- == 0)
 218                         i = (rx_ring->count - 1);
 219
 220                 /*
 221                  * Force memory writes to complete before letting h/w
 222                  * know there are new descriptors to fetch.  (Only
 223                  * applicable for weak-ordered memory model archs,
 224                  * such as IA-64).
 225                  */
 226                 wmb();
 227                 writel(i, adapter->hw.hw_addr + rx_ring->tail);
 228         }
 229 }
 230
 231 /**
 232  * e1000_alloc_rx_buffers_ps - Replace used receive buffers; packet split
 233  * @adapter: address of board private structure
 234  **/
 235 static void e1000_alloc_rx_buffers_ps(struct e1000_adapter *adapter,
 236                                       int cleaned_count)
 237 {
 238         struct net_device *netdev = adapter->netdev;
 239         struct pci_dev *pdev = adapter->pdev;
 240         union e1000_rx_desc_packet_split *rx_desc;
 241         struct e1000_ring *rx_ring = adapter->rx_ring;
 242         struct e1000_buffer *buffer_info;
 243         struct e1000_ps_page *ps_page;
 244         struct sk_buff *skb;
 245         unsigned int i, j;
 246
 247         i = rx_ring->next_to_use;
 248         buffer_info = &rx_ring->buffer_info[i];
 249
 250         while (cleaned_count--) {
 251                 rx_desc = E1000_RX_DESC_PS(*rx_ring, i);
 252
 253                 for (j = 0; j < PS_PAGE_BUFFERS; j++) {
 254                         ps_page = &buffer_info->ps_pages[j];
 255                         if (j >= adapter->rx_ps_pages) {
 256                                 /* all unused desc entries get hw null ptr */
 257                                 rx_desc->read.buffer_addr[j+1] = ~cpu_to_le64(0);
 258                                 continue;
 259                         }
 260                         if (!ps_page->page) {
 261                                 ps_page->page = alloc_page(GFP_ATOMIC);
 262                                 if (!ps_page->page) {
 263                                         adapter->alloc_rx_buff_failed++;
 264                                         goto no_buffers;
 265                                 }
 266                                 ps_page->dma = pci_map_page(pdev,
 267                                                    ps_page->page,
 268                                                    0, PAGE_SIZE,
 269                                                    PCI_DMA_FROMDEVICE);
 270                                 if (pci_dma_mapping_error(pdev, ps_page->dma)) {
 271                                         dev_err(&adapter->pdev->dev,
 272                                           "RX DMA page map failed\n");
 273                                         adapter->rx_dma_failed++;
 274                                         goto no_buffers;
 275                                 }
 276                         }
 277                         /*
 278                          * Refresh the desc even if buffer_addrs
 279                          * didn't change because each write-back
 280                          * erases this info.
 281                          */
 282                         rx_desc->read.buffer_addr[j+1] =
 283                              cpu_to_le64(ps_page->dma);
 284                 }
 285
 286                 skb = netdev_alloc_skb(netdev,
 287                                        adapter->rx_ps_bsize0 + NET_IP_ALIGN);
 288
 289                 if (!skb) {
 290                         adapter->alloc_rx_buff_failed++;
 291                         break;
 292                 }
 293
 294                 /*
 295                  * Make buffer alignment 2 beyond a 16 byte boundary
 296                  * this will result in a 16 byte aligned IP header after
 297                  * the 14 byte MAC header is removed
 298                  */
 299                 skb_reserve(skb, NET_IP_ALIGN);
 300
 301                 buffer_info->skb = skb;
 302                 buffer_info->dma = pci_map_single(pdev, skb->data,
 303                                                   adapter->rx_ps_bsize0,
 304                                                   PCI_DMA_FROMDEVICE);
 305                 if (pci_dma_mapping_error(pdev, buffer_info->dma)) {
 306                         dev_err(&pdev->dev, "RX DMA map failed\n");
 307                         adapter->rx_dma_failed++;
 308                         /* cleanup skb */
 309                         dev_kfree_skb_any(skb);
 310                         buffer_info->skb = NULL;
 311                         break;
 312                 }
 313
 314                 rx_desc->read.buffer_addr[0] = cpu_to_le64(buffer_info->dma);
 315
 316                 i++;
 317                 if (i == rx_ring->count)
 318                         i = 0;
 319                 buffer_info = &rx_ring->buffer_info[i];
 320         }
 321
 322 no_buffers:
 323         if (rx_ring->next_to_use != i) {
 324                 rx_ring->next_to_use = i;
 325
 326                 if (!(i--))
 327                         i = (rx_ring->count - 1);
 328
 329                 /*
 330                  * Force memory writes to complete before letting h/w
 331                  * know there are new descriptors to fetch.  (Only
 332                  * applicable for weak-ordered memory model archs,
 333                  * such as IA-64).
 334                  */
 335                 wmb();
 336                 /*
 337                  * Hardware increments by 16 bytes, but packet split
 338                  * descriptors are 32 bytes...so we increment tail
 339                  * twice as much.
 340                  */
 341                 writel(i<<1, adapter->hw.hw_addr + rx_ring->tail);
 342         }
 343 }
 344
 345 /**
 346  * e1000_alloc_jumbo_rx_buffers - Replace used jumbo receive buffers
 347  * @adapter: address of board private structure
 348  * @rx_ring: pointer to receive ring structure
 349  * @cleaned_count: number of buffers to allocate this pass
 350  **/
 351
 352 static void e1000_alloc_jumbo_rx_buffers(struct e1000_adapter *adapter,
 353                                          int cleaned_count)
 354 {
 355         struct net_device *netdev = adapter->netdev;
 356         struct pci_dev *pdev = adapter->pdev;
 357         struct e1000_rx_desc *rx_desc;
 358         struct e1000_ring *rx_ring = adapter->rx_ring;
 359         struct e1000_buffer *buffer_info;
 360         struct sk_buff *skb;
 361         unsigned int i;
 362         unsigned int bufsz = 256 -
 363                              16 /* for skb_reserve */ -
 364                              NET_IP_ALIGN;
 365
 366         i = rx_ring->next_to_use;
 367         buffer_info = &rx_ring->buffer_info[i];
 368
 369         while (cleaned_count--) {
 370                 skb = buffer_info->skb;
 371                 if (skb) {
 372                         skb_trim(skb, 0);
 373                         goto check_page;
 374                 }
 375
 376                 skb = netdev_alloc_skb(netdev, bufsz);
 377                 if (unlikely(!skb)) {
 378                         /* Better luck next round */
 379                         adapter->alloc_rx_buff_failed++;
 380                         break;
 381                 }
 382
 383                 /* Make buffer alignment 2 beyond a 16 byte boundary
 384                  * this will result in a 16 byte aligned IP header after
 385                  * the 14 byte MAC header is removed
 386                  */
 387                 skb_reserve(skb, NET_IP_ALIGN);
 388
 389                 buffer_info->skb = skb;
 390 check_page:
 391                 /* allocate a new page if necessary */
 392                 if (!buffer_info->page) {
 393                         buffer_info->page = alloc_page(GFP_ATOMIC);
 394                         if (unlikely(!buffer_info->page)) {
 395                                 adapter->alloc_rx_buff_failed++;
 396                                 break;
 397                         }
 398                 }
 399
 400                 if (!buffer_info->dma)
 401                         buffer_info->dma = pci_map_page(pdev,
 402                                                         buffer_info->page, 0,
 403                                                         PAGE_SIZE,
 404                                                         PCI_DMA_FROMDEVICE);
 405
 406                 rx_desc = E1000_RX_DESC(*rx_ring, i);
 407                 rx_desc->buffer_addr = cpu_to_le64(buffer_info->dma);
 408
 409                 if (unlikely(++i == rx_ring->count))
 410                         i = 0;
 411                 buffer_info = &rx_ring->buffer_info[i];
 412         }
 413
 414         if (likely(rx_ring->next_to_use != i)) {
 415                 rx_ring->next_to_use = i;
 416                 if (unlikely(i-- == 0))
 417                         i = (rx_ring->count - 1);
 418
 419                 /* Force memory writes to complete before letting h/w
 420                  * know there are new descriptors to fetch.  (Only
 421                  * applicable for weak-ordered memory model archs,
 422                  * such as IA-64). */
 423                 wmb();
 424                 writel(i, adapter->hw.hw_addr + rx_ring->tail);
 425         }
 426 }
 427
 428 /**
 429  * e1000_clean_rx_irq - Send received data up the network stack; legacy
 430  * @adapter: board private structure
 431  *
 432  * the return value indicates whether actual cleaning was done, there
 433  * is no guarantee that everything was cleaned
 434  **/
 435 static bool e1000_clean_rx_irq(struct e1000_adapter *adapter,
 436                                int *work_done, int work_to_do)
 437 {
 438         struct net_device *netdev = adapter->netdev;
 439         struct pci_dev *pdev = adapter->pdev;
 440         struct e1000_ring *rx_ring = adapter->rx_ring;
 441         struct e1000_rx_desc *rx_desc, *next_rxd;
 442         struct e1000_buffer *buffer_info, *next_buffer;
 443         u32 length;
 444         unsigned int i;
 445         int cleaned_count = 0;
 446         bool cleaned = 0;
 447         unsigned int total_rx_bytes = 0, total_rx_packets = 0;
 448
 449         i = rx_ring->next_to_clean;
 450         rx_desc = E1000_RX_DESC(*rx_ring, i);
 451         buffer_info = &rx_ring->buffer_info[i];
 452
 453         while (rx_desc->status & E1000_RXD_STAT_DD) {
 454                 struct sk_buff *skb;
 455                 u8 status;
 456
 457                 if (*work_done >= work_to_do)
 458                         break;
 459                 (*work_done)++;
 460
 461                 status = rx_desc->status;
 462                 skb = buffer_info->skb;
 463                 buffer_info->skb = NULL;
 464
 465                 prefetch(skb->data - NET_IP_ALIGN);
 466
 467                 i++;
 468                 if (i == rx_ring->count)
 469                         i = 0;
 470                 next_rxd = E1000_RX_DESC(*rx_ring, i);
 471                 prefetch(next_rxd);
 472
 473                 next_buffer = &rx_ring->buffer_info[i];
 474
 475                 cleaned = 1;
 476                 cleaned_count++;
 477                 pci_unmap_single(pdev,
 478                                  buffer_info->dma,
 479                                  adapter->rx_buffer_len,
 480                                  PCI_DMA_FROMDEVICE);
 481                 buffer_info->dma = 0;
 482
 483                 length = le16_to_cpu(rx_desc->length);
 484
 485                 /* !EOP means multiple descriptors were used to store a single
 486                  * packet, also make sure the frame isn't just CRC only */
 487                 if (!(status & E1000_RXD_STAT_EOP) || (length <= 4)) {
 488                         /* All receives must fit into a single buffer */
 489                         e_dbg("%s: Receive packet consumed multiple buffers\n",
 490                               netdev->name);
 491                         /* recycle */
 492                         buffer_info->skb = skb;
 493                         goto next_desc;
 494                 }
 495
 496                 if (rx_desc->errors & E1000_RXD_ERR_FRAME_ERR_MASK) {
 497                         /* recycle */
 498                         buffer_info->skb = skb;
 499                         goto next_desc;
 500                 }
 501
 502                 total_rx_bytes += length;
 503                 total_rx_packets++;
 504
 505                 /*
 506                  * code added for copybreak, this should improve
 507                  * performance for small packets with large amounts
 508                  * of reassembly being done in the stack
 509                  */
 510                 if (length < copybreak) {
 511                         struct sk_buff *new_skb =
 512                             netdev_alloc_skb(netdev, length + NET_IP_ALIGN);
 513                         if (new_skb) {
 514                                 skb_reserve(new_skb, NET_IP_ALIGN);
 515                                 skb_copy_to_linear_data_offset(new_skb,
 516                                                                -NET_IP_ALIGN,
 517                                                                (skb->data -
 518                                                                 NET_IP_ALIGN),
 519                                                                (length +
 520                                                                 NET_IP_ALIGN));
 521                                 /* save the skb in buffer_info as good */
 522                                 buffer_info->skb = skb;
 523                                 skb = new_skb;
 524                         }
 525                         /* else just continue with the old one */
 526                 }
 527                 /* end copybreak code */
 528                 skb_put(skb, length);
 529
 530                 /* Receive Checksum Offload */
 531                 e1000_rx_checksum(adapter,
 532                                   (u32)(status) |
 533                                   ((u32)(rx_desc->errors) << 24),
 534                                   le16_to_cpu(rx_desc->csum), skb);
 535
 536                 e1000_receive_skb(adapter, netdev, skb,status,rx_desc->special);
 537
 538 next_desc:
 539                 rx_desc->status = 0;
 540
 541                 /* return some buffers to hardware, one at a time is too slow */
 542                 if (cleaned_count >= E1000_RX_BUFFER_WRITE) {
 543                         adapter->alloc_rx_buf(adapter, cleaned_count);
 544                         cleaned_count = 0;
 545                 }
 546
 547                 /* use prefetched values */
 548                 rx_desc = next_rxd;
 549                 buffer_info = next_buffer;
 550         }
 551         rx_ring->next_to_clean = i;
 552
 553         cleaned_count = e1000_desc_unused(rx_ring);
 554         if (cleaned_count)
 555                 adapter->alloc_rx_buf(adapter, cleaned_count);
 556
 557         adapter->total_rx_bytes += total_rx_bytes;
 558         adapter->total_rx_packets += total_rx_packets;
 559         adapter->net_stats.rx_bytes += total_rx_bytes;
 560         adapter->net_stats.rx_packets += total_rx_packets;
 561         return cleaned;
 562 }
 563
 564 static void e1000_put_txbuf(struct e1000_adapter *adapter,
 565                              struct e1000_buffer *buffer_info)
 566 {
 567         if (buffer_info->dma) {
 568                 pci_unmap_page(adapter->pdev, buffer_info->dma,
 569                                buffer_info->length, PCI_DMA_TODEVICE);
 570                 buffer_info->dma = 0;
 571         }
 572         if (buffer_info->skb) {
 573                 dev_kfree_skb_any(buffer_info->skb);
 574                 buffer_info->skb = NULL;
 575         }
 576 }
 577
 578 static void e1000_print_tx_hang(struct e1000_adapter *adapter)
 579 {
 580         struct e1000_ring *tx_ring = adapter->tx_ring;
 581         unsigned int i = tx_ring->next_to_clean;
 582         unsigned int eop = tx_ring->buffer_info[i].next_to_watch;
 583         struct e1000_tx_desc *eop_desc = E1000_TX_DESC(*tx_ring, eop);
 584
 585         /* detected Tx unit hang */
 586         e_err("Detected Tx Unit Hang:\n"
 587               "  TDH                  <%x>\n"
 588               "  TDT                  <%x>\n"
 589               "  next_to_use          <%x>\n"
 590               "  next_to_clean        <%x>\n"
 591               "buffer_info[next_to_clean]:\n"
 592               "  time_stamp           <%lx>\n"
 593               "  next_to_watch        <%x>\n"
 594               "  jiffies              <%lx>\n"
 595               "  next_to_watch.status <%x>\n",
 596               readl(adapter->hw.hw_addr + tx_ring->head),
 597               readl(adapter->hw.hw_addr + tx_ring->tail),
 598               tx_ring->next_to_use,
 599               tx_ring->next_to_clean,
 600               tx_ring->buffer_info[eop].time_stamp,
 601               eop,
 602               jiffies,
 603               eop_desc->upper.fields.status);
 604 }
 605
 606 /**
 607  * e1000_clean_tx_irq - Reclaim resources after transmit completes
 608  * @adapter: board private structure
 609  *
 610  * the return value indicates whether actual cleaning was done, there
 611  * is no guarantee that everything was cleaned
 612  **/
 613 static bool e1000_clean_tx_irq(struct e1000_adapter *adapter)
 614 {
 615         struct net_device *netdev = adapter->netdev;
 616         struct e1000_hw *hw = &adapter->hw;
 617         struct e1000_ring *tx_ring = adapter->tx_ring;
 618         struct e1000_tx_desc *tx_desc, *eop_desc;
 619         struct e1000_buffer *buffer_info;
 620         unsigned int i, eop;
 621         unsigned int count = 0;
 622         bool cleaned = 0;
 623         unsigned int total_tx_bytes = 0, total_tx_packets = 0;
 624
 625         i = tx_ring->next_to_clean;
 626         eop = tx_ring->buffer_info[i].next_to_watch;
 627         eop_desc = E1000_TX_DESC(*tx_ring, eop);
 628
 629         while (eop_desc->upper.data & cpu_to_le32(E1000_TXD_STAT_DD)) {
 630                 for (cleaned = 0; !cleaned; ) {
 631                         tx_desc = E1000_TX_DESC(*tx_ring, i);
 632                         buffer_info = &tx_ring->buffer_info[i];
 633                         cleaned = (i == eop);
 634
 635                         if (cleaned) {
 636                                 struct sk_buff *skb = buffer_info->skb;
 637                                 unsigned int segs, bytecount;
 638                                 segs = skb_shinfo(skb)->gso_segs ?: 1;
 639                                 /* multiply data chunks by size of headers */
 640                                 bytecount = ((segs - 1) * skb_headlen(skb)) +
 641                                             skb->len;
 642                                 total_tx_packets += segs;
 643                                 total_tx_bytes += bytecount;
 644                         }
 645
 646                         e1000_put_txbuf(adapter, buffer_info);
 647                         tx_desc->upper.data = 0;
 648
 649                         i++;
 650                         if (i == tx_ring->count)
 651                                 i = 0;
 652                 }
 653
 654                 eop = tx_ring->buffer_info[i].next_to_watch;
 655                 eop_desc = E1000_TX_DESC(*tx_ring, eop);
 656 #define E1000_TX_WEIGHT 64
 657                 /* weight of a sort for tx, to avoid endless transmit cleanup */
 658                 if (count++ == E1000_TX_WEIGHT)
 659                         break;
 660         }
 661
 662         tx_ring->next_to_clean = i;
 663
 664 #define TX_WAKE_THRESHOLD 32
 665         if (cleaned && netif_carrier_ok(netdev) &&
 666                      e1000_desc_unused(tx_ring) >= TX_WAKE_THRESHOLD) {
 667                 /* Make sure that anybody stopping the queue after this
 668                  * sees the new next_to_clean.
 669                  */
 670                 smp_mb();
 671
 672                 if (netif_queue_stopped(netdev) &&
 673                     !(test_bit(__E1000_DOWN, &adapter->state))) {
 674                         netif_wake_queue(netdev);
 675                         ++adapter->restart_queue;
 676                 }
 677         }
 678
 679         if (adapter->detect_tx_hung) {
 680                 /*
 681                  * Detect a transmit hang in hardware, this serializes the
 682                  * check with the clearing of time_stamp and movement of i
 683                  */
 684                 adapter->detect_tx_hung = 0;
 685                 if (tx_ring->buffer_info[eop].dma &&
 686                     time_after(jiffies, tx_ring->buffer_info[eop].time_stamp
 687                                + (adapter->tx_timeout_factor * HZ))
 688                     && !(er32(STATUS) & E1000_STATUS_TXOFF)) {
 689                         e1000_print_tx_hang(adapter);
 690                         netif_stop_queue(netdev);
 691                 }
 692         }
 693         adapter->total_tx_bytes += total_tx_bytes;
 694         adapter->total_tx_packets += total_tx_packets;
 695         adapter->net_stats.tx_bytes += total_tx_bytes;
 696         adapter->net_stats.tx_packets += total_tx_packets;
 697         return cleaned;
 698 }
 699
 700 /**
 701  * e1000_clean_rx_irq_ps - Send received data up the network stack; packet split
 702  * @adapter: board private structure
 703  *
 704  * the return value indicates whether actual cleaning was done, there
 705  * is no guarantee that everything was cleaned
 706  **/
 707 static bool e1000_clean_rx_irq_ps(struct e1000_adapter *adapter,
 708                                   int *work_done, int work_to_do)
 709 {
 710         union e1000_rx_desc_packet_split *rx_desc, *next_rxd;
 711         struct net_device *netdev = adapter->netdev;
 712         struct pci_dev *pdev = adapter->pdev;
 713         struct e1000_ring *rx_ring = adapter->rx_ring;
 714         struct e1000_buffer *buffer_info, *next_buffer;
 715         struct e1000_ps_page *ps_page;
 716         struct sk_buff *skb;
 717         unsigned int i, j;
 718         u32 length, staterr;
 719         int cleaned_count = 0;
 720         bool cleaned = 0;
 721         unsigned int total_rx_bytes = 0, total_rx_packets = 0;
 722
 723         i = rx_ring->next_to_clean;
 724         rx_desc = E1000_RX_DESC_PS(*rx_ring, i);
 725         staterr = le32_to_cpu(rx_desc->wb.middle.status_error);
 726         buffer_info = &rx_ring->buffer_info[i];
 727
 728         while (staterr & E1000_RXD_STAT_DD) {
 729                 if (*work_done >= work_to_do)
 730                         break;
 731                 (*work_done)++;
 732                 skb = buffer_info->skb;
 733
 734                 /* in the packet split case this is header only */
 735                 prefetch(skb->data - NET_IP_ALIGN);
 736
 737                 i++;
 738                 if (i == rx_ring->count)
 739                         i = 0;
 740                 next_rxd = E1000_RX_DESC_PS(*rx_ring, i);
 741                 prefetch(next_rxd);
 742
 743                 next_buffer = &rx_ring->buffer_info[i];
 744
 745                 cleaned = 1;
 746                 cleaned_count++;
 747                 pci_unmap_single(pdev, buffer_info->dma,
 748                                  adapter->rx_ps_bsize0,
 749                                  PCI_DMA_FROMDEVICE);
 750                 buffer_info->dma = 0;
 751
 752                 if (!(staterr & E1000_RXD_STAT_EOP)) {
 753                         e_dbg("%s: Packet Split buffers didn't pick up the "
 754                               "full packet\n", netdev->name);
 755                         dev_kfree_skb_irq(skb);
 756                         goto next_desc;
 757                 }
 758
 759                 if (staterr & E1000_RXDEXT_ERR_FRAME_ERR_MASK) {
 760                         dev_kfree_skb_irq(skb);
 761                         goto next_desc;
 762                 }
 763
 764                 length = le16_to_cpu(rx_desc->wb.middle.length0);
 765
 766                 if (!length) {
 767                         e_dbg("%s: Last part of the packet spanning multiple "
 768                               "descriptors\n", netdev->name);
 769                         dev_kfree_skb_irq(skb);
 770                         goto next_desc;
 771                 }
 772
 773                 /* Good Receive */
 774                 skb_put(skb, length);
 775
 776                 {
 777                 /*
 778                  * this looks ugly, but it seems compiler issues make it
 779                  * more efficient than reusing j
 780                  */
 781                 int l1 = le16_to_cpu(rx_desc->wb.upper.length[0]);
 782
 783                 /*
 784                  * page alloc/put takes too long and effects small packet
 785                  * throughput, so unsplit small packets and save the alloc/put
 786                  * only valid in softirq (napi) context to call kmap_*
 787                  */
 788                 if (l1 && (l1 <= copybreak) &&
 789                     ((length + l1) <= adapter->rx_ps_bsize0)) {
 790                         u8 *vaddr;
 791
 792                         ps_page = &buffer_info->ps_pages[0];
 793
 794                         /*
 795                          * there is no documentation about how to call
 796                          * kmap_atomic, so we can't hold the mapping
 797                          * very long
 798                          */
 799                         pci_dma_sync_single_for_cpu(pdev, ps_page->dma,
 800                                 PAGE_SIZE, PCI_DMA_FROMDEVICE);
 801                         vaddr = kmap_atomic(ps_page->page, KM_SKB_DATA_SOFTIRQ);
 802                         memcpy(skb_tail_pointer(skb), vaddr, l1);
 803                         kunmap_atomic(vaddr, KM_SKB_DATA_SOFTIRQ);
 804                         pci_dma_sync_single_for_device(pdev, ps_page->dma,
 805                                 PAGE_SIZE, PCI_DMA_FROMDEVICE);
 806
 807                         skb_put(skb, l1);
 808                         goto copydone;
 809                 } /* if */
 810                 }
 811
 812                 for (j = 0; j < PS_PAGE_BUFFERS; j++) {
 813                         length = le16_to_cpu(rx_desc->wb.upper.length[j]);
 814                         if (!length)
 815                                 break;
 816
 817                         ps_page = &buffer_info->ps_pages[j];
 818                         pci_unmap_page(pdev, ps_page->dma, PAGE_SIZE,
 819                                        PCI_DMA_FROMDEVICE);
 820                         ps_page->dma = 0;
 821                         skb_fill_page_desc(skb, j, ps_page->page, 0, length);
 822                         ps_page->page = NULL;
 823                         skb->len += length;
 824                         skb->data_len += length;
 825                         skb->truesize += length;
 826                 }
 827
 828 copydone:
 829                 total_rx_bytes += skb->len;
 830                 total_rx_packets++;
 831
 832                 e1000_rx_checksum(adapter, staterr, le16_to_cpu(
 833                         rx_desc->wb.lower.hi_dword.csum_ip.csum), skb);
 834
 835                 if (rx_desc->wb.upper.header_status &
 836                            cpu_to_le16(E1000_RXDPS_HDRSTAT_HDRSP))
 837                         adapter->rx_hdr_split++;
 838
 839                 e1000_receive_skb(adapter, netdev, skb,
 840                                   staterr, rx_desc->wb.middle.vlan);
 841
 842 next_desc:
 843                 rx_desc->wb.middle.status_error &= cpu_to_le32(~0xFF);
 844                 buffer_info->skb = NULL;
 845
 846                 /* return some buffers to hardware, one at a time is too slow */
 847                 if (cleaned_count >= E1000_RX_BUFFER_WRITE) {
 848                         adapter->alloc_rx_buf(adapter, cleaned_count);
 849                         cleaned_count = 0;
 850                 }
 851
 852                 /* use prefetched values */
 853                 rx_desc = next_rxd;
 854                 buffer_info = next_buffer;
 855
 856                 staterr = le32_to_cpu(rx_desc->wb.middle.status_error);
 857         }
 858         rx_ring->next_to_clean = i;
 859
 860         cleaned_count = e1000_desc_unused(rx_ring);
 861         if (cleaned_count)
 862                 adapter->alloc_rx_buf(adapter, cleaned_count);
 863
 864         adapter->total_rx_bytes += total_rx_bytes;
 865         adapter->total_rx_packets += total_rx_packets;
 866         adapter->net_stats.rx_bytes += total_rx_bytes;
 867         adapter->net_stats.rx_packets += total_rx_packets;
 868         return cleaned;
 869 }
 870
 871 /**
 872  * e1000_consume_page - helper function
 873  **/
 874 static void e1000_consume_page(struct e1000_buffer *bi, struct sk_buff *skb,
 875                                u16 length)
 876 {
 877         bi->page = NULL;
 878         skb->len += length;
 879         skb->data_len += length;
 880         skb->truesize += length;
 881 }
 882
 883 /**
 884  * e1000_clean_jumbo_rx_irq - Send received data up the network stack; legacy
 885  * @adapter: board private structure
 886  *
 887  * the return value indicates whether actual cleaning was done, there
 888  * is no guarantee that everything was cleaned
 889  **/
 890
 891 static bool e1000_clean_jumbo_rx_irq(struct e1000_adapter *adapter,
 892                                      int *work_done, int work_to_do)
 893 {
 894         struct net_device *netdev = adapter->netdev;
 895         struct pci_dev *pdev = adapter->pdev;
 896         struct e1000_ring *rx_ring = adapter->rx_ring;
 897         struct e1000_rx_desc *rx_desc, *next_rxd;
 898         struct e1000_buffer *buffer_info, *next_buffer;
 899         u32 length;
 900         unsigned int i;
 901         int cleaned_count = 0;
 902         bool cleaned = false;
 903         unsigned int total_rx_bytes=0, total_rx_packets=0;
 904
 905         i = rx_ring->next_to_clean;
 906         rx_desc = E1000_RX_DESC(*rx_ring, i);
 907         buffer_info = &rx_ring->buffer_info[i];
 908
 909         while (rx_desc->status & E1000_RXD_STAT_DD) {
 910                 struct sk_buff *skb;
 911                 u8 status;
 912
 913                 if (*work_done >= work_to_do)
 914                         break;
 915                 (*work_done)++;
 916
 917                 status = rx_desc->status;
 918                 skb = buffer_info->skb;
 919                 buffer_info->skb = NULL;
 920
 921                 ++i;
 922                 if (i == rx_ring->count)
 923                         i = 0;
 924                 next_rxd = E1000_RX_DESC(*rx_ring, i);
 925                 prefetch(next_rxd);
 926
 927                 next_buffer = &rx_ring->buffer_info[i];
 928
 929                 cleaned = true;
 930                 cleaned_count++;
 931                 pci_unmap_page(pdev, buffer_info->dma, PAGE_SIZE,
 932                                PCI_DMA_FROMDEVICE);
 933                 buffer_info->dma = 0;
 934
 935                 length = le16_to_cpu(rx_desc->length);
 936
 937                 /* errors is only valid for DD + EOP descriptors */
 938                 if (unlikely((status & E1000_RXD_STAT_EOP) &&
 939                     (rx_desc->errors & E1000_RXD_ERR_FRAME_ERR_MASK))) {
 940                                 /* recycle both page and skb */
 941                                 buffer_info->skb = skb;
 942                                 /* an error means any chain goes out the window
 943                                  * too */
 944                                 if (rx_ring->rx_skb_top)
 945                                         dev_kfree_skb(rx_ring->rx_skb_top);
 946                                 rx_ring->rx_skb_top = NULL;
 947                                 goto next_desc;
 948                 }
 949
 950 #define rxtop rx_ring->rx_skb_top
 951                 if (!(status & E1000_RXD_STAT_EOP)) {
 952                         /* this descriptor is only the beginning (or middle) */
 953                         if (!rxtop) {
 954                                 /* this is the beginning of a chain */
 955                                 rxtop = skb;
 956                                 skb_fill_page_desc(rxtop, 0, buffer_info->page,
 957                                                    0, length);
 958                         } else {
 959                                 /* this is the middle of a chain */
 960                                 skb_fill_page_desc(rxtop,
 961                                     skb_shinfo(rxtop)->nr_frags,
 962                                     buffer_info->page, 0, length);
 963                                 /* re-use the skb, only consumed the page */
 964                                 buffer_info->skb = skb;
 965                         }
 966                         e1000_consume_page(buffer_info, rxtop, length);
 967                         goto next_desc;
 968                 } else {
 969                         if (rxtop) {
 970                                 /* end of the chain */
 971                                 skb_fill_page_desc(rxtop,
 972                                     skb_shinfo(rxtop)->nr_frags,
 973                                     buffer_info->page, 0, length);
 974                                 /* re-use the current skb, we only consumed the
 975                                  * page */
 976                                 buffer_info->skb = skb;
 977                                 skb = rxtop;
 978                                 rxtop = NULL;
 979                                 e1000_consume_page(buffer_info, skb, length);
 980                         } else {
 981                                 /* no chain, got EOP, this buf is the packet
 982                                  * copybreak to save the put_page/alloc_page */
 983                                 if (length <= copybreak &&
 984                                     skb_tailroom(skb) >= length) {
 985                                         u8 *vaddr;
 986                                         vaddr = kmap_atomic(buffer_info->page,
 987                                                            KM_SKB_DATA_SOFTIRQ);
 988                                         memcpy(skb_tail_pointer(skb), vaddr,
 989                                                length);
 990                                         kunmap_atomic(vaddr,
 991                                                       KM_SKB_DATA_SOFTIRQ);
 992                                         /* re-use the page, so don't erase
 993                                          * buffer_info->page */
 994                                         skb_put(skb, length);
 995                                 } else {
 996                                         skb_fill_page_desc(skb, 0,
 997                                                            buffer_info->page, 0,
 998                                                            length);
 999                                         e1000_consume_page(buffer_info, skb,
1000                                                            length);
1001                                 }
1002                         }
1003                 }
1004
1005                 /* Receive Checksum Offload XXX recompute due to CRC strip? */
1006                 e1000_rx_checksum(adapter,
1007                                   (u32)(status) |
1008                                   ((u32)(rx_desc->errors) << 24),
1009                                   le16_to_cpu(rx_desc->csum), skb);
1010
1011                 /* probably a little skewed due to removing CRC */
1012                 total_rx_bytes += skb->len;
1013                 total_rx_packets++;
1014
1015                 /* eth type trans needs skb->data to point to something */
1016                 if (!pskb_may_pull(skb, ETH_HLEN)) {
1017                         e_err("pskb_may_pull failed.\n");
1018                         dev_kfree_skb(skb);
1019                         goto next_desc;
1020                 }
1021
1022                 e1000_receive_skb(adapter, netdev, skb, status,
1023                                   rx_desc->special);
1024
1025 next_desc:
1026                 rx_desc->status = 0;
1027
1028                 /* return some buffers to hardware, one at a time is too slow */
1029                 if (unlikely(cleaned_count >= E1000_RX_BUFFER_WRITE)) {
1030                         adapter->alloc_rx_buf(adapter, cleaned_count);
1031                         cleaned_count = 0;
1032                 }
1033
1034                 /* use prefetched values */
1035                 rx_desc = next_rxd;
1036                 buffer_info = next_buffer;
1037         }
1038         rx_ring->next_to_clean = i;
1039
1040         cleaned_count = e1000_desc_unused(rx_ring);
1041         if (cleaned_count)
1042                 adapter->alloc_rx_buf(adapter, cleaned_count);
1043
1044         adapter->total_rx_bytes += total_rx_bytes;
1045         adapter->total_rx_packets += total_rx_packets;
1046         adapter->net_stats.rx_bytes += total_rx_bytes;
1047         adapter->net_stats.rx_packets += total_rx_packets;
1048         return cleaned;
1049 }
1050
1051 /**
1052  * e1000_clean_rx_ring - Free Rx Buffers per Queue
1053  * @adapter: board private structure
1054  **/
1055 static void e1000_clean_rx_ring(struct e1000_adapter *adapter)
1056 {
1057         struct e1000_ring *rx_ring = adapter->rx_ring;
1058         struct e1000_buffer *buffer_info;
1059         struct e1000_ps_page *ps_page;
1060         struct pci_dev *pdev = adapter->pdev;
1061         unsigned int i, j;
1062
1063         /* Free all the Rx ring sk_buffs */
1064         for (i = 0; i < rx_ring->count; i++) {
1065                 buffer_info = &rx_ring->buffer_info[i];
1066                 if (buffer_info->dma) {
1067                         if (adapter->clean_rx == e1000_clean_rx_irq)
1068                                 pci_unmap_single(pdev, buffer_info->dma,
1069                                                  adapter->rx_buffer_len,
1070                                                  PCI_DMA_FROMDEVICE);
1071                         else if (adapter->clean_rx == e1000_clean_jumbo_rx_irq)
1072                                 pci_unmap_page(pdev, buffer_info->dma,
1073                                                PAGE_SIZE,
1074                                                PCI_DMA_FROMDEVICE);
1075                         else if (adapter->clean_rx == e1000_clean_rx_irq_ps)
1076                                 pci_unmap_single(pdev, buffer_info->dma,
1077                                                  adapter->rx_ps_bsize0,
1078                                                  PCI_DMA_FROMDEVICE);
1079                         buffer_info->dma = 0;
1080                 }
1081
1082                 if (buffer_info->page) {
1083                         put_page(buffer_info->page);
1084                         buffer_info->page = NULL;
1085                 }
1086
1087                 if (buffer_info->skb) {
1088                         dev_kfree_skb(buffer_info->skb);
1089                         buffer_info->skb = NULL;
1090                 }
1091
1092                 for (j = 0; j < PS_PAGE_BUFFERS; j++) {
1093                         ps_page = &buffer_info->ps_pages[j];
1094                         if (!ps_page->page)
1095                                 break;
1096                         pci_unmap_page(pdev, ps_page->dma, PAGE_SIZE,
1097                                        PCI_DMA_FROMDEVICE);
1098                         ps_page->dma = 0;
1099                         put_page(ps_page->page);
1100                         ps_page->page = NULL;
1101                 }
1102         }
1103
1104         /* there also may be some cached data from a chained receive */
1105         if (rx_ring->rx_skb_top) {
1106                 dev_kfree_skb(rx_ring->rx_skb_top);
1107                 rx_ring->rx_skb_top = NULL;
1108         }
1109
1110         /* Zero out the descriptor ring */
1111         memset(rx_ring->desc, 0, rx_ring->size);
1112
1113         rx_ring->next_to_clean = 0;
1114         rx_ring->next_to_use = 0;
1115
1116         writel(0, adapter->hw.hw_addr + rx_ring->head);
1117         writel(0, adapter->hw.hw_addr + rx_ring->tail);
1118 }
1119
1120 /**
1121  * e1000_intr_msi - Interrupt Handler
1122  * @irq: interrupt number
1123  * @data: pointer to a network interface device structure
1124  **/
1125 static irqreturn_t e1000_intr_msi(int irq, void *data)
1126 {
1127         struct net_device *netdev = data;
1128         struct e1000_adapter *adapter = netdev_priv(netdev);
1129         struct e1000_hw *hw = &adapter->hw;
1130         u32 icr = er32(ICR);
1131
1132         /*
1133          * read ICR disables interrupts using IAM
1134          */
1135
1136         if (icr & (E1000_ICR_RXSEQ | E1000_ICR_LSC)) {
1137                 hw->mac.get_link_status = 1;
1138                 /*
1139                  * ICH8 workaround-- Call gig speed drop workaround on cable
1140                  * disconnect (LSC) before accessing any PHY registers
1141                  */
1142                 if ((adapter->flags & FLAG_LSC_GIG_SPEED_DROP) &&
1143                     (!(er32(STATUS) & E1000_STATUS_LU)))
1144                         e1000e_gig_downshift_workaround_ich8lan(hw);
1145
1146                 /*
1147                  * 80003ES2LAN workaround-- For packet buffer work-around on
1148                  * link down event; disable receives here in the ISR and reset
1149                  * adapter in watchdog
1150                  */
1151                 if (netif_carrier_ok(netdev) &&
1152                     adapter->flags & FLAG_RX_NEEDS_RESTART) {
1153                         /* disable receives */
1154                         u32 rctl = er32(RCTL);
1155                         ew32(RCTL, rctl & ~E1000_RCTL_EN);
1156                         adapter->flags |= FLAG_RX_RESTART_NOW;
1157                 }
1158                 /* guard against interrupt when we're going down */
1159                 if (!test_bit(__E1000_DOWN, &adapter->state))
1160                         mod_timer(&adapter->watchdog_timer, jiffies + 1);
1161         }
1162
1163         if (netif_rx_schedule_prep(netdev, &adapter->napi)) {
1164                 adapter->total_tx_bytes = 0;
1165                 adapter->total_tx_packets = 0;
1166                 adapter->total_rx_bytes = 0;
1167                 adapter->total_rx_packets = 0;
1168                 __netif_rx_schedule(netdev, &adapter->napi);
1169         }
1170
1171         return IRQ_HANDLED;
1172 }
1173
1174 /**
1175  * e1000_intr - Interrupt Handler
1176  * @irq: interrupt number
1177  * @data: pointer to a network interface device structure
1178  **/
1179 static irqreturn_t e1000_intr(int irq, void *data)
1180 {
1181         struct net_device *netdev = data;
1182         struct e1000_adapter *adapter = netdev_priv(netdev);
1183         struct e1000_hw *hw = &adapter->hw;
1184         u32 rctl, icr = er32(ICR);
1185
1186         if (!icr)
1187                 return IRQ_NONE;  /* Not our interrupt */
1188
1189         /*
1190          * IMS will not auto-mask if INT_ASSERTED is not set, and if it is
1191          * not set, then the adapter didn't send an interrupt
1192          */
1193         if (!(icr & E1000_ICR_INT_ASSERTED))
1194                 return IRQ_NONE;
1195
1196         /*
1197          * Interrupt Auto-Mask...upon reading ICR,
1198          * interrupts are masked.  No need for the
1199          * IMC write
1200          */
1201
1202         if (icr & (E1000_ICR_RXSEQ | E1000_ICR_LSC)) {
1203                 hw->mac.get_link_status = 1;
1204                 /*
1205                  * ICH8 workaround-- Call gig speed drop workaround on cable
1206                  * disconnect (LSC) before accessing any PHY registers
1207                  */
1208                 if ((adapter->flags & FLAG_LSC_GIG_SPEED_DROP) &&
1209                     (!(er32(STATUS) & E1000_STATUS_LU)))
1210                         e1000e_gig_downshift_workaround_ich8lan(hw);
1211
1212                 /*
1213                  * 80003ES2LAN workaround--
1214                  * For packet buffer work-around on link down event;
1215                  * disable receives here in the ISR and
1216                  * reset adapter in watchdog
1217                  */
1218                 if (netif_carrier_ok(netdev) &&
1219                     (adapter->flags & FLAG_RX_NEEDS_RESTART)) {
1220                         /* disable receives */
1221                         rctl = er32(RCTL);
1222                         ew32(RCTL, rctl & ~E1000_RCTL_EN);
1223                         adapter->flags |= FLAG_RX_RESTART_NOW;
1224                 }
1225                 /* guard against interrupt when we're going down */
1226                 if (!test_bit(__E1000_DOWN, &adapter->state))
1227                         mod_timer(&adapter->watchdog_timer, jiffies + 1);
1228         }
1229
1230         if (netif_rx_schedule_prep(netdev, &adapter->napi)) {
1231                 adapter->total_tx_bytes = 0;
1232                 adapter->total_tx_packets = 0;
1233                 adapter->total_rx_bytes = 0;
1234                 adapter->total_rx_packets = 0;
1235                 __netif_rx_schedule(netdev, &adapter->napi);
1236         }
1237
1238         return IRQ_HANDLED;
1239 }
1240
1241 static irqreturn_t e1000_msix_other(int irq, void *data)
1242 {
1243         struct net_device *netdev = data;
1244         struct e1000_adapter *adapter = netdev_priv(netdev);
1245         struct e1000_hw *hw = &adapter->hw;
1246         u32 icr = er32(ICR);
1247
1248         if (!(icr & E1000_ICR_INT_ASSERTED)) {
1249                 ew32(IMS, E1000_IMS_OTHER);
1250                 return IRQ_NONE;
1251         }
1252
1253         if (icr & adapter->eiac_mask)
1254                 ew32(ICS, (icr & adapter->eiac_mask));
1255
1256         if (icr & E1000_ICR_OTHER) {
1257                 if (!(icr & E1000_ICR_LSC))
1258                         goto no_link_interrupt;
1259                 hw->mac.get_link_status = 1;
1260                 /* guard against interrupt when we're going down */
1261                 if (!test_bit(__E1000_DOWN, &adapter->state))
1262                         mod_timer(&adapter->watchdog_timer, jiffies + 1);
1263         }
1264
1265 no_link_interrupt:
1266         ew32(IMS, E1000_IMS_LSC | E1000_IMS_OTHER);
1267
1268         return IRQ_HANDLED;
1269 }
1270
1271
1272 static irqreturn_t e1000_intr_msix_tx(int irq, void *data)
1273 {
1274         struct net_device *netdev = data;
1275         struct e1000_adapter *adapter = netdev_priv(netdev);
1276         struct e1000_hw *hw = &adapter->hw;
1277         struct e1000_ring *tx_ring = adapter->tx_ring;
1278
1279
1280         adapter->total_tx_bytes = 0;
1281         adapter->total_tx_packets = 0;
1282
1283         if (!e1000_clean_tx_irq(adapter))
1284                 /* Ring was not completely cleaned, so fire another interrupt */
1285                 ew32(ICS, tx_ring->ims_val);
1286
1287         return IRQ_HANDLED;
1288 }
1289
1290 static irqreturn_t e1000_intr_msix_rx(int irq, void *data)
1291 {
1292         struct net_device *netdev = data;
1293         struct e1000_adapter *adapter = netdev_priv(netdev);
1294
1295         /* Write the ITR value calculated at the end of the
1296          * previous interrupt.
1297          */
1298         if (adapter->rx_ring->set_itr) {
1299                 writel(1000000000 / (adapter->rx_ring->itr_val * 256),
1300                        adapter->hw.hw_addr + adapter->rx_ring->itr_register);
1301                 adapter->rx_ring->set_itr = 0;
1302         }
1303
1304         if (netif_rx_schedule_prep(netdev, &adapter->napi)) {
1305                 adapter->total_rx_bytes = 0;
1306                 adapter->total_rx_packets = 0;
1307                 __netif_rx_schedule(netdev, &adapter->napi);
1308         }
1309         return IRQ_HANDLED;
1310 }
1311
1312 /**
1313  * e1000_configure_msix - Configure MSI-X hardware
1314  *
1315  * e1000_configure_msix sets up the hardware to properly
1316  * generate MSI-X interrupts.
1317  **/
1318 static void e1000_configure_msix(struct e1000_adapter *adapter)
1319 {
1320         struct e1000_hw *hw = &adapter->hw;
1321         struct e1000_ring *rx_ring = adapter->rx_ring;
1322         struct e1000_ring *tx_ring = adapter->tx_ring;
1323         int vector = 0;
1324         u32 ctrl_ext, ivar = 0;
1325
1326         adapter->eiac_mask = 0;
1327
1328         /* Workaround issue with spurious interrupts on 82574 in MSI-X mode */
1329         if (hw->mac.type == e1000_82574) {
1330                 u32 rfctl = er32(RFCTL);
1331                 rfctl |= E1000_RFCTL_ACK_DIS;
1332                 ew32(RFCTL, rfctl);
1333         }
1334
1335 #define E1000_IVAR_INT_ALLOC_VALID      0x8
1336         /* Configure Rx vector */
1337         rx_ring->ims_val = E1000_IMS_RXQ0;
1338         adapter->eiac_mask |= rx_ring->ims_val;
1339         if (rx_ring->itr_val)
1340                 writel(1000000000 / (rx_ring->itr_val * 256),
1341                        hw->hw_addr + rx_ring->itr_register);
1342         else
1343                 writel(1, hw->hw_addr + rx_ring->itr_register);
1344         ivar = E1000_IVAR_INT_ALLOC_VALID | vector;
1345
1346         /* Configure Tx vector */
1347         tx_ring->ims_val = E1000_IMS_TXQ0;
1348         vector++;
1349         if (tx_ring->itr_val)
1350                 writel(1000000000 / (tx_ring->itr_val * 256),
1351                        hw->hw_addr + tx_ring->itr_register);
1352         else
1353                 writel(1, hw->hw_addr + tx_ring->itr_register);
1354         adapter->eiac_mask |= tx_ring->ims_val;
1355         ivar |= ((E1000_IVAR_INT_ALLOC_VALID | vector) << 8);
1356
1357         /* set vector for Other Causes, e.g. link changes */
1358         vector++;
1359         ivar |= ((E1000_IVAR_INT_ALLOC_VALID | vector) << 16);
1360         if (rx_ring->itr_val)
1361                 writel(1000000000 / (rx_ring->itr_val * 256),
1362                        hw->hw_addr + E1000_EITR_82574(vector));
1363         else
1364                 writel(1, hw->hw_addr + E1000_EITR_82574(vector));
1365
1366         /* Cause Tx interrupts on every write back */
1367         ivar |= (1 << 31);
1368
1369         ew32(IVAR, ivar);
1370
1371         /* enable MSI-X PBA support */
1372         ctrl_ext = er32(CTRL_EXT);
1373         ctrl_ext |= E1000_CTRL_EXT_PBA_CLR;
1374
1375         /* Auto-Mask Other interrupts upon ICR read */
1376 #define E1000_EIAC_MASK_82574   0x01F00000
1377         ew32(IAM, ~E1000_EIAC_MASK_82574 | E1000_IMS_OTHER);
1378         ctrl_ext |= E1000_CTRL_EXT_EIAME;
1379         ew32(CTRL_EXT, ctrl_ext);
1380         e1e_flush();
1381 }
1382
1383 void e1000e_reset_interrupt_capability(struct e1000_adapter *adapter)
1384 {
1385         if (adapter->msix_entries) {
1386                 pci_disable_msix(adapter->pdev);
1387                 kfree(adapter->msix_entries);
1388                 adapter->msix_entries = NULL;
1389         } else if (adapter->flags & FLAG_MSI_ENABLED) {
1390                 pci_disable_msi(adapter->pdev);
1391                 adapter->flags &= ~FLAG_MSI_ENABLED;
1392         }
1393
1394         return;
1395 }
1396
1397 /**
1398  * e1000e_set_interrupt_capability - set MSI or MSI-X if supported
1399  *
1400  * Attempt to configure interrupts using the best available
1401  * capabilities of the hardware and kernel.
1402  **/
1403 void e1000e_set_interrupt_capability(struct e1000_adapter *adapter)
1404 {
1405         int err;
1406         int numvecs, i;
1407
1408
1409         switch (adapter->int_mode) {
1410         case E1000E_INT_MODE_MSIX:
1411                 if (adapter->flags & FLAG_HAS_MSIX) {
1412                         numvecs = 3; /* RxQ0, TxQ0 and other */
1413                         adapter->msix_entries = kcalloc(numvecs,
1414                                                       sizeof(struct msix_entry),
1415                                                       GFP_KERNEL);
1416                         if (adapter->msix_entries) {
1417                                 for (i = 0; i < numvecs; i++)
1418                                         adapter->msix_entries[i].entry = i;
1419
1420                                 err = pci_enable_msix(adapter->pdev,
1421                                                       adapter->msix_entries,
1422                                                       numvecs);
1423                                 if (err == 0)
1424                                         return;
1425                         }
1426                         /* MSI-X failed, so fall through and try MSI */
1427                         e_err("Failed to initialize MSI-X interrupts.  "
1428                               "Falling back to MSI interrupts.\n");
1429                         e1000e_reset_interrupt_capability(adapter);
1430                 }
1431                 adapter->int_mode = E1000E_INT_MODE_MSI;
1432                 /* Fall through */
1433         case E1000E_INT_MODE_MSI:
1434                 if (!pci_enable_msi(adapter->pdev)) {
1435                         adapter->flags |= FLAG_MSI_ENABLED;
1436                 } else {
1437                         adapter->int_mode = E1000E_INT_MODE_LEGACY;
1438                         e_err("Failed to initialize MSI interrupts.  Falling "
1439                               "back to legacy interrupts.\n");
1440                 }
1441                 /* Fall through */
1442         case E1000E_INT_MODE_LEGACY:
1443                 /* Don't do anything; this is the system default */
1444                 break;
1445         }
1446
1447         return;
1448 }
1449
1450 /**
1451  * e1000_request_msix - Initialize MSI-X interrupts
1452  *
1453  * e1000_request_msix allocates MSI-X vectors and requests interrupts from the
1454  * kernel.
1455  **/
1456 static int e1000_request_msix(struct e1000_adapter *adapter)
1457 {
1458         struct net_device *netdev = adapter->netdev;
1459         int err = 0, vector = 0;
1460
1461         if (strlen(netdev->name) < (IFNAMSIZ - 5))
1462                 sprintf(adapter->rx_ring->name, "%s-rx0", netdev->name);
1463         else
1464                 memcpy(adapter->rx_ring->name, netdev->name, IFNAMSIZ);
1465         err = request_irq(adapter->msix_entries[vector].vector,
1466                           &e1000_intr_msix_rx, 0, adapter->rx_ring->name,
1467                           netdev);
1468         if (err)
1469                 goto out;
1470         adapter->rx_ring->itr_register = E1000_EITR_82574(vector);
1471         adapter->rx_ring->itr_val = adapter->itr;
1472         vector++;
1473
1474         if (strlen(netdev->name) < (IFNAMSIZ - 5))
1475                 sprintf(adapter->tx_ring->name, "%s-tx0", netdev->name);
1476         else
1477                 memcpy(adapter->tx_ring->name, netdev->name, IFNAMSIZ);
1478         err = request_irq(adapter->msix_entries[vector].vector,
1479                           &e1000_intr_msix_tx, 0, adapter->tx_ring->name,
1480                           netdev);
1481         if (err)
1482                 goto out;
1483         adapter->tx_ring->itr_register = E1000_EITR_82574(vector);
1484         adapter->tx_ring->itr_val = adapter->itr;
1485         vector++;
1486
1487         err = request_irq(adapter->msix_entries[vector].vector,
1488                           &e1000_msix_other, 0, netdev->name, netdev);
1489         if (err)
1490                 goto out;
1491
1492         e1000_configure_msix(adapter);
1493         return 0;
1494 out:
1495         return err;
1496 }
1497
1498 /**
1499  * e1000_request_irq - initialize interrupts
1500  *
1501  * Attempts to configure interrupts using the best available
1502  * capabilities of the hardware and kernel.
1503  **/
1504 static int e1000_request_irq(struct e1000_adapter *adapter)
1505 {
1506         struct net_device *netdev = adapter->netdev;
1507         int err;
1508
1509         if (adapter->msix_entries) {
1510                 err = e1000_request_msix(adapter);
1511                 if (!err)
1512                         return err;
1513                 /* fall back to MSI */
1514                 e1000e_reset_interrupt_capability(adapter);
1515                 adapter->int_mode = E1000E_INT_MODE_MSI;
1516                 e1000e_set_interrupt_capability(adapter);
1517         }
1518         if (adapter->flags & FLAG_MSI_ENABLED) {
1519                 err = request_irq(adapter->pdev->irq, &e1000_intr_msi, 0,
1520                                   netdev->name, netdev);
1521                 if (!err)
1522                         return err;
1523
1524                 /* fall back to legacy interrupt */
1525                 e1000e_reset_interrupt_capability(adapter);
1526                 adapter->int_mode = E1000E_INT_MODE_LEGACY;
1527         }
1528
1529         err = request_irq(adapter->pdev->irq, &e1000_intr, IRQF_SHARED,
1530                           netdev->name, netdev);
1531         if (err)
1532                 e_err("Unable to allocate interrupt, Error: %d\n", err);
1533
1534         return err;
1535 }
1536
1537 static void e1000_free_irq(struct e1000_adapter *adapter)
1538 {
1539         struct net_device *netdev = adapter->netdev;
1540
1541         if (adapter->msix_entries) {
1542                 int vector = 0;
1543
1544                 free_irq(adapter->msix_entries[vector].vector, netdev);
1545                 vector++;
1546
1547                 free_irq(adapter->msix_entries[vector].vector, netdev);
1548                 vector++;
1549
1550                 /* Other Causes interrupt vector */
1551                 free_irq(adapter->msix_entries[vector].vector, netdev);
1552                 return;
1553         }
1554
1555         free_irq(adapter->pdev->irq, netdev);
1556 }
1557
1558 /**
1559  * e1000_irq_disable - Mask off interrupt generation on the NIC
1560  **/
1561 static void e1000_irq_disable(struct e1000_adapter *adapter)
1562 {
1563         struct e1000_hw *hw = &adapter->hw;
1564
1565         ew32(IMC, ~0);
1566         if (adapter->msix_entries)
1567                 ew32(EIAC_82574, 0);
1568         e1e_flush();
1569         synchronize_irq(adapter->pdev->irq);
1570 }
1571
1572 /**
1573  * e1000_irq_enable - Enable default interrupt generation settings
1574  **/
1575 static void e1000_irq_enable(struct e1000_adapter *adapter)
1576 {
1577         struct e1000_hw *hw = &adapter->hw;
1578
1579         if (adapter->msix_entries) {
1580                 ew32(EIAC_82574, adapter->eiac_mask & E1000_EIAC_MASK_82574);
1581                 ew32(IMS, adapter->eiac_mask | E1000_IMS_OTHER | E1000_IMS_LSC);
1582         } else {
1583                 ew32(IMS, IMS_ENABLE_MASK);
1584         }
1585         e1e_flush();
1586 }
1587
1588 /**
1589  * e1000_get_hw_control - get control of the h/w from f/w
1590  * @adapter: address of board private structure
1591  *
1592  * e1000_get_hw_control sets {CTRL_EXT|SWSM}:DRV_LOAD bit.
1593  * For ASF and Pass Through versions of f/w this means that
1594  * the driver is loaded. For AMT version (only with 82573)
1595  * of the f/w this means that the network i/f is open.
1596  **/
1597 static void e1000_get_hw_control(struct e1000_adapter *adapter)
1598 {
1599         struct e1000_hw *hw = &adapter->hw;
1600         u32 ctrl_ext;
1601         u32 swsm;
1602
1603         /* Let firmware know the driver has taken over */
1604         if (adapter->flags & FLAG_HAS_SWSM_ON_LOAD) {
1605                 swsm = er32(SWSM);
1606                 ew32(SWSM, swsm | E1000_SWSM_DRV_LOAD);
1607         } else if (adapter->flags & FLAG_HAS_CTRLEXT_ON_LOAD) {
1608                 ctrl_ext = er32(CTRL_EXT);
1609                 ew32(CTRL_EXT, ctrl_ext | E1000_CTRL_EXT_DRV_LOAD);
1610         }
1611 }
1612
1613 /**
1614  * e1000_release_hw_control - release control of the h/w to f/w
1615  * @adapter: address of board private structure
1616  *
1617  * e1000_release_hw_control resets {CTRL_EXT|SWSM}:DRV_LOAD bit.
1618  * For ASF and Pass Through versions of f/w this means that the
1619  * driver is no longer loaded. For AMT version (only with 82573) i
1620  * of the f/w this means that the network i/f is closed.
1621  *
1622  **/
1623 static void e1000_release_hw_control(struct e1000_adapter *adapter)
1624 {
1625         struct e1000_hw *hw = &adapter->hw;
1626         u32 ctrl_ext;
1627         u32 swsm;
1628
1629         /* Let firmware taken over control of h/w */
1630         if (adapter->flags & FLAG_HAS_SWSM_ON_LOAD) {
1631                 swsm = er32(SWSM);
1632                 ew32(SWSM, swsm & ~E1000_SWSM_DRV_LOAD);
1633         } else if (adapter->flags & FLAG_HAS_CTRLEXT_ON_LOAD) {
1634                 ctrl_ext = er32(CTRL_EXT);
1635                 ew32(CTRL_EXT, ctrl_ext & ~E1000_CTRL_EXT_DRV_LOAD);
1636         }
1637 }
1638
1639 /**
1640  * @e1000_alloc_ring - allocate memory for a ring structure
1641  **/
1642 static int e1000_alloc_ring_dma(struct e1000_adapter *adapter,
1643                                 struct e1000_ring *ring)
1644 {
1645         struct pci_dev *pdev = adapter->pdev;
1646
1647         ring->desc = dma_alloc_coherent(&pdev->dev, ring->size, &ring->dma,
1648                                         GFP_KERNEL);
1649         if (!ring->desc)
1650                 return -ENOMEM;
1651
1652         return 0;
1653 }
1654
1655 /**
1656  * e1000e_setup_tx_resources - allocate Tx resources (Descriptors)
1657  * @adapter: board private structure
1658  *
1659  * Return 0 on success, negative on failure
1660  **/
1661 int e1000e_setup_tx_resources(struct e1000_adapter *adapter)
1662 {
1663         struct e1000_ring *tx_ring = adapter->tx_ring;
1664         int err = -ENOMEM, size;
1665
1666         size = sizeof(struct e1000_buffer) * tx_ring->count;
1667         tx_ring->buffer_info = vmalloc(size);
1668         if (!tx_ring->buffer_info)
1669                 goto err;
1670         memset(tx_ring->buffer_info, 0, size);
1671
1672         /* round up to nearest 4K */
1673         tx_ring->size = tx_ring->count * sizeof(struct e1000_tx_desc);
1674         tx_ring->size = ALIGN(tx_ring->size, 4096);
1675
1676         err = e1000_alloc_ring_dma(adapter, tx_ring);
1677         if (err)
1678                 goto err;
1679
1680         tx_ring->next_to_use = 0;
1681         tx_ring->next_to_clean = 0;
1682         spin_lock_init(&adapter->tx_queue_lock);
1683
1684         return 0;
1685 err:
1686         vfree(tx_ring->buffer_info);
1687         e_err("Unable to allocate memory for the transmit descriptor ring\n");
1688         return err;
1689 }
1690
1691 /**
1692  * e1000e_setup_rx_resources - allocate Rx resources (Descriptors)
1693  * @adapter: board private structure
1694  *
1695  * Returns 0 on success, negative on failure
1696  **/
1697 int e1000e_setup_rx_resources(struct e1000_adapter *adapter)
1698 {
1699         struct e1000_ring *rx_ring = adapter->rx_ring;
1700         struct e1000_buffer *buffer_info;
1701         int i, size, desc_len, err = -ENOMEM;
1702
1703         size = sizeof(struct e1000_buffer) * rx_ring->count;
1704         rx_ring->buffer_info = vmalloc(size);
1705         if (!rx_ring->buffer_info)
1706                 goto err;
1707         memset(rx_ring->buffer_info, 0, size);
1708
1709         for (i = 0; i < rx_ring->count; i++) {
1710                 buffer_info = &rx_ring->buffer_info[i];
1711                 buffer_info->ps_pages = kcalloc(PS_PAGE_BUFFERS,
1712                                                 sizeof(struct e1000_ps_page),
1713                                                 GFP_KERNEL);
1714                 if (!buffer_info->ps_pages)
1715                         goto err_pages;
1716         }
1717
1718         desc_len = sizeof(union e1000_rx_desc_packet_split);
1719
1720         /* Round up to nearest 4K */
1721         rx_ring->size = rx_ring->count * desc_len;
1722         rx_ring->size = ALIGN(rx_ring->size, 4096);
1723
1724         err = e1000_alloc_ring_dma(adapter, rx_ring);
1725         if (err)
1726                 goto err_pages;
1727
1728         rx_ring->next_to_clean = 0;
1729         rx_ring->next_to_use = 0;
1730         rx_ring->rx_skb_top = NULL;
1731
1732         return 0;
1733
1734 err_pages:
1735         for (i = 0; i < rx_ring->count; i++) {
1736                 buffer_info = &rx_ring->buffer_info[i];
1737                 kfree(buffer_info->ps_pages);
1738         }
1739 err:
1740         vfree(rx_ring->buffer_info);
1741         e_err("Unable to allocate memory for the transmit descriptor ring\n");
1742         return err;
1743 }
1744
1745 /**
1746  * e1000_clean_tx_ring - Free Tx Buffers
1747  * @adapter: board private structure
1748  **/
1749 static void e1000_clean_tx_ring(struct e1000_adapter *adapter)
1750 {
1751         struct e1000_ring *tx_ring = adapter->tx_ring;
1752         struct e1000_buffer *buffer_info;
1753         unsigned long size;
1754         unsigned int i;
1755
1756         for (i = 0; i < tx_ring->count; i++) {
1757                 buffer_info = &tx_ring->buffer_info[i];
1758                 e1000_put_txbuf(adapter, buffer_info);
1759         }
1760
1761         size = sizeof(struct e1000_buffer) * tx_ring->count;
1762         memset(tx_ring->buffer_info, 0, size);
1763
1764         memset(tx_ring->desc, 0, tx_ring->size);
1765
1766         tx_ring->next_to_use = 0;
1767         tx_ring->next_to_clean = 0;
1768
1769         writel(0, adapter->hw.hw_addr + tx_ring->head);
1770         writel(0, adapter->hw.hw_addr + tx_ring->tail);
1771 }
1772
1773 /**
1774  * e1000e_free_tx_resources - Free Tx Resources per Queue
1775  * @adapter: board private structure
1776  *
1777  * Free all transmit software resources
1778  **/
1779 void e1000e_free_tx_resources(struct e1000_adapter *adapter)
1780 {
1781         struct pci_dev *pdev = adapter->pdev;
1782         struct e1000_ring *tx_ring = adapter->tx_ring;
1783
1784         e1000_clean_tx_ring(adapter);
1785
1786         vfree(tx_ring->buffer_info);
1787         tx_ring->buffer_info = NULL;
1788
1789         dma_free_coherent(&pdev->dev, tx_ring->size, tx_ring->desc,
1790                           tx_ring->dma);
1791         tx_ring->desc = NULL;
1792 }
1793
1794 /**
1795  * e1000e_free_rx_resources - Free Rx Resources
1796  * @adapter: board private structure
1797  *
1798  * Free all receive software resources
1799  **/
1800
1801 void e1000e_free_rx_resources(struct e1000_adapter *adapter)
1802 {
1803         struct pci_dev *pdev = adapter->pdev;
1804         struct e1000_ring *rx_ring = adapter->rx_ring;
1805         int i;
1806
1807         e1000_clean_rx_ring(adapter);
1808
1809         for (i = 0; i < rx_ring->count; i++) {
1810                 kfree(rx_ring->buffer_info[i].ps_pages);
1811         }
1812
1813         vfree(rx_ring->buffer_info);
1814         rx_ring->buffer_info = NULL;
1815
1816         dma_free_coherent(&pdev->dev, rx_ring->size, rx_ring->desc,
1817                           rx_ring->dma);
1818         rx_ring->desc = NULL;
1819 }
1820
1821 /**
1822  * e1000_update_itr - update the dynamic ITR value based on statistics
1823  * @adapter: pointer to adapter
1824  * @itr_setting: current adapter->itr
1825  * @packets: the number of packets during this measurement interval
1826  * @bytes: the number of bytes during this measurement interval
1827  *
1828  *      Stores a new ITR value based on packets and byte
1829  *      counts during the last interrupt.  The advantage of per interrupt
1830  *      computation is faster updates and more accurate ITR for the current
1831  *      traffic pattern.  Constants in this function were computed
1832  *      based on theoretical maximum wire speed and thresholds were set based
1833  *      on testing data as well as attempting to minimize response time
1834  *      while increasing bulk throughput.  This functionality is controlled
1835  *      by the InterruptThrottleRate module parameter.
1836  **/
1837 static unsigned int e1000_update_itr(struct e1000_adapter *adapter,
1838                                      u16 itr_setting, int packets,
1839                                      int bytes)
1840 {
1841         unsigned int retval = itr_setting;
1842
1843         if (packets == 0)
1844                 goto update_itr_done;
1845
1846         switch (itr_setting) {
1847         case lowest_latency:
1848                 /* handle TSO and jumbo frames */
1849                 if (bytes/packets > 8000)
1850                         retval = bulk_latency;
1851                 else if ((packets < 5) && (bytes > 512)) {
1852                         retval = low_latency;
1853                 }
1854                 break;
1855         case low_latency:  /* 50 usec aka 20000 ints/s */
1856                 if (bytes > 10000) {
1857                         /* this if handles the TSO accounting */
1858                         if (bytes/packets > 8000) {
1859                                 retval = bulk_latency;
1860                         } else if ((packets < 10) || ((bytes/packets) > 1200)) {
1861                                 retval = bulk_latency;
1862                         } else if ((packets > 35)) {
1863                                 retval = lowest_latency;
1864                         }
1865                 } else if (bytes/packets > 2000) {
1866                         retval = bulk_latency;
1867                 } else if (packets <= 2 && bytes < 512) {
1868                         retval = lowest_latency;
1869                 }
1870                 break;
1871         case bulk_latency: /* 250 usec aka 4000 ints/s */
1872                 if (bytes > 25000) {
1873                         if (packets > 35) {
1874                                 retval = low_latency;
1875                         }
1876                 } else if (bytes < 6000) {
1877                         retval = low_latency;
1878                 }
1879                 break;
1880         }
1881
1882 update_itr_done:
1883         return retval;
1884 }
1885
1886 static void e1000_set_itr(struct e1000_adapter *adapter)
1887 {
1888         struct e1000_hw *hw = &adapter->hw;
1889         u16 current_itr;
1890         u32 new_itr = adapter->itr;
1891
1892         /* for non-gigabit speeds, just fix the interrupt rate at 4000 */
1893         if (adapter->link_speed != SPEED_1000) {
1894                 current_itr = 0;
1895                 new_itr = 4000;
1896                 goto set_itr_now;
1897         }
1898
1899         adapter->tx_itr = e1000_update_itr(adapter,
1900                                     adapter->tx_itr,
1901                                     adapter->total_tx_packets,
1902                                     adapter->total_tx_bytes);
1903         /* conservative mode (itr 3) eliminates the lowest_latency setting */
1904         if (adapter->itr_setting == 3 && adapter->tx_itr == lowest_latency)
1905                 adapter->tx_itr = low_latency;
1906
1907         adapter->rx_itr = e1000_update_itr(adapter,
1908                                     adapter->rx_itr,
1909                                     adapter->total_rx_packets,
1910                                     adapter->total_rx_bytes);
1911         /* conservative mode (itr 3) eliminates the lowest_latency setting */
1912         if (adapter->itr_setting == 3 && adapter->rx_itr == lowest_latency)
1913                 adapter->rx_itr = low_latency;
1914
1915         current_itr = max(adapter->rx_itr, adapter->tx_itr);
1916
1917         switch (current_itr) {
1918         /* counts and packets in update_itr are dependent on these numbers */
1919         case lowest_latency:
1920                 new_itr = 70000;
1921                 break;
1922         case low_latency:
1923                 new_itr = 20000; /* aka hwitr = ~200 */
1924                 break;
1925         case bulk_latency:
1926                 new_itr = 4000;
1927                 break;
1928         default:
1929                 break;
1930         }
1931
1932 set_itr_now:
1933         if (new_itr != adapter->itr) {
1934                 /*
1935                  * this attempts to bias the interrupt rate towards Bulk
1936                  * by adding intermediate steps when interrupt rate is
1937                  * increasing
1938                  */
1939                 new_itr = new_itr > adapter->itr ?
1940                              min(adapter->itr + (new_itr >> 2), new_itr) :
1941                              new_itr;
1942                 adapter->itr = new_itr;
1943                 adapter->rx_ring->itr_val = new_itr;
1944                 if (adapter->msix_entries)
1945                         adapter->rx_ring->set_itr = 1;
1946                 else
1947                         ew32(ITR, 1000000000 / (new_itr * 256));
1948         }
1949 }
1950
1951 /**
1952  * e1000_alloc_queues - Allocate memory for all rings
1953  * @adapter: board private structure to initialize
1954  **/
1955 static int __devinit e1000_alloc_queues(struct e1000_adapter *adapter)
1956 {
1957         adapter->tx_ring = kzalloc(sizeof(struct e1000_ring), GFP_KERNEL);
1958         if (!adapter->tx_ring)
1959                 goto err;
1960
1961         adapter->rx_ring = kzalloc(sizeof(struct e1000_ring), GFP_KERNEL);
1962         if (!adapter->rx_ring)
1963                 goto err;
1964
1965         return 0;
1966 err:
1967         e_err("Unable to allocate memory for queues\n");
1968         kfree(adapter->rx_ring);
1969         kfree(adapter->tx_ring);
1970         return -ENOMEM;
1971 }
1972
1973 /**
1974  * e1000_clean - NAPI Rx polling callback
1975  * @napi: struct associated with this polling callback
1976  * @budget: amount of packets driver is allowed to process this poll
1977  **/
1978 static int e1000_clean(struct napi_struct *napi, int budget)
1979 {
1980         struct e1000_adapter *adapter = container_of(napi, struct e1000_adapter, napi);
1981         struct e1000_hw *hw = &adapter->hw;
1982         struct net_device *poll_dev = adapter->netdev;
1983         int tx_cleaned = 0, work_done = 0;
1984
1985         /* Must NOT use netdev_priv macro here. */
1986         adapter = poll_dev->priv;
1987
1988         if (adapter->msix_entries &&
1989             !(adapter->rx_ring->ims_val & adapter->tx_ring->ims_val))
1990                 goto clean_rx;
1991
1992         /*
1993          * e1000_clean is called per-cpu.  This lock protects
1994          * tx_ring from being cleaned by multiple cpus
1995          * simultaneously.  A failure obtaining the lock means
1996          * tx_ring is currently being cleaned anyway.
1997          */
1998         if (spin_trylock(&adapter->tx_queue_lock)) {
1999                 tx_cleaned = e1000_clean_tx_irq(adapter);
2000                 spin_unlock(&adapter->tx_queue_lock);
2001         }
2002
2003 clean_rx:
2004         adapter->clean_rx(adapter, &work_done, budget);
2005
2006         if (tx_cleaned)
2007                 work_done = budget;
2008
2009         /* If budget not fully consumed, exit the polling mode */
2010         if (work_done < budget) {
2011                 if (adapter->itr_setting & 3)
2012                         e1000_set_itr(adapter);
2013                 netif_rx_complete(poll_dev, napi);
2014                 if (adapter->msix_entries)
2015                         ew32(IMS, adapter->rx_ring->ims_val);
2016                 else
2017                         e1000_irq_enable(adapter);
2018         }
2019
2020         return work_done;
2021 }
2022
2023 static void e1000_vlan_rx_add_vid(struct net_device *netdev, u16 vid)
2024 {
2025         struct e1000_adapter *adapter = netdev_priv(netdev);
2026         struct e1000_hw *hw = &adapter->hw;
2027         u32 vfta, index;
2028
2029         /* don't update vlan cookie if already programmed */
2030         if ((adapter->hw.mng_cookie.status &
2031              E1000_MNG_DHCP_COOKIE_STATUS_VLAN) &&
2032             (vid == adapter->mng_vlan_id))
2033                 return;
2034         /* add VID to filter table */
2035         index = (vid >> 5) & 0x7F;
2036         vfta = E1000_READ_REG_ARRAY(hw, E1000_VFTA, index);
2037         vfta |= (1 << (vid & 0x1F));
2038         e1000e_write_vfta(hw, index, vfta);
2039 }
2040
2041 static void e1000_vlan_rx_kill_vid(struct net_device *netdev, u16 vid)
2042 {
2043         struct e1000_adapter *adapter = netdev_priv(netdev);
2044         struct e1000_hw *hw = &adapter->hw;
2045         u32 vfta, index;
2046
2047         if (!test_bit(__E1000_DOWN, &adapter->state))
2048                 e1000_irq_disable(adapter);
2049         vlan_group_set_device(adapter->vlgrp, vid, NULL);
2050
2051         if (!test_bit(__E1000_DOWN, &adapter->state))
2052                 e1000_irq_enable(adapter);
2053
2054         if ((adapter->hw.mng_cookie.status &
2055              E1000_MNG_DHCP_COOKIE_STATUS_VLAN) &&
2056             (vid == adapter->mng_vlan_id)) {
2057                 /* release control to f/w */
2058                 e1000_release_hw_control(adapter);
2059                 return;
2060         }
2061
2062         /* remove VID from filter table */
2063         index = (vid >> 5) & 0x7F;
2064         vfta = E1000_READ_REG_ARRAY(hw, E1000_VFTA, index);
2065         vfta &= ~(1 << (vid & 0x1F));
2066         e1000e_write_vfta(hw, index, vfta);
2067 }
2068
2069 static void e1000_update_mng_vlan(struct e1000_adapter *adapter)
2070 {
2071         struct net_device *netdev = adapter->netdev;
2072         u16 vid = adapter->hw.mng_cookie.vlan_id;
2073         u16 old_vid = adapter->mng_vlan_id;
2074
2075         if (!adapter->vlgrp)
2076                 return;
2077
2078         if (!vlan_group_get_device(adapter->vlgrp, vid)) {
2079                 adapter->mng_vlan_id = E1000_MNG_VLAN_NONE;
2080                 if (adapter->hw.mng_cookie.status &
2081                         E1000_MNG_DHCP_COOKIE_STATUS_VLAN) {
2082                         e1000_vlan_rx_add_vid(netdev, vid);
2083                         adapter->mng_vlan_id = vid;
2084                 }
2085
2086                 if ((old_vid != (u16)E1000_MNG_VLAN_NONE) &&
2087                                 (vid != old_vid) &&
2088                     !vlan_group_get_device(adapter->vlgrp, old_vid))
2089                         e1000_vlan_rx_kill_vid(netdev, old_vid);
2090         } else {
2091                 adapter->mng_vlan_id = vid;
2092         }
2093 }
2094
2095
2096 static void e1000_vlan_rx_register(struct net_device *netdev,
2097                                    struct vlan_group *grp)
2098 {
2099         struct e1000_adapter *adapter = netdev_priv(netdev);
2100         struct e1000_hw *hw = &adapter->hw;
2101         u32 ctrl, rctl;
2102
2103         if (!test_bit(__E1000_DOWN, &adapter->state))
2104                 e1000_irq_disable(adapter);
2105         adapter->vlgrp = grp;
2106
2107         if (grp) {
2108                 /* enable VLAN tag insert/strip */
2109                 ctrl = er32(CTRL);
2110                 ctrl |= E1000_CTRL_VME;
2111                 ew32(CTRL, ctrl);
2112
2113                 if (adapter->flags & FLAG_HAS_HW_VLAN_FILTER) {
2114                         /* enable VLAN receive filtering */
2115                         rctl = er32(RCTL);
2116                         rctl &= ~E1000_RCTL_CFIEN;
2117                         ew32(RCTL, rctl);
2118                         e1000_update_mng_vlan(adapter);
2119                 }
2120         } else {
2121                 /* disable VLAN tag insert/strip */
2122                 ctrl = er32(CTRL);
2123                 ctrl &= ~E1000_CTRL_VME;
2124                 ew32(CTRL, ctrl);
2125
2126                 if (adapter->flags & FLAG_HAS_HW_VLAN_FILTER) {
2127                         if (adapter->mng_vlan_id !=
2128                             (u16)E1000_MNG_VLAN_NONE) {
2129                                 e1000_vlan_rx_kill_vid(netdev,
2130                                                        adapter->mng_vlan_id);
2131                                 adapter->mng_vlan_id = E1000_MNG_VLAN_NONE;
2132                         }
2133                 }
2134         }
2135
2136         if (!test_bit(__E1000_DOWN, &adapter->state))
2137                 e1000_irq_enable(adapter);
2138 }
2139
2140 static void e1000_restore_vlan(struct e1000_adapter *adapter)
2141 {
2142         u16 vid;
2143
2144         e1000_vlan_rx_register(adapter->netdev, adapter->vlgrp);
2145
2146         if (!adapter->vlgrp)
2147                 return;
2148
2149         for (vid = 0; vid < VLAN_GROUP_ARRAY_LEN; vid++) {
2150                 if (!vlan_group_get_device(adapter->vlgrp, vid))
2151                         continue;
2152                 e1000_vlan_rx_add_vid(adapter->netdev, vid);
2153         }
2154 }
2155
2156 static void e1000_init_manageability(struct e1000_adapter *adapter)
2157 {
2158         struct e1000_hw *hw = &adapter->hw;
2159         u32 manc, manc2h;
2160
2161         if (!(adapter->flags & FLAG_MNG_PT_ENABLED))
2162                 return;
2163
2164         manc = er32(MANC);
2165
2166         /*
2167          * enable receiving management packets to the host. this will probably
2168          * generate destination unreachable messages from the host OS, but
2169          * the packets will be handled on SMBUS
2170          */
2171         manc |= E1000_MANC_EN_MNG2HOST;
2172         manc2h = er32(MANC2H);
2173 #define E1000_MNG2HOST_PORT_623 (1 << 5)
2174 #define E1000_MNG2HOST_PORT_664 (1 << 6)
2175         manc2h |= E1000_MNG2HOST_PORT_623;
2176         manc2h |= E1000_MNG2HOST_PORT_664;
2177         ew32(MANC2H, manc2h);
2178         ew32(MANC, manc);
2179 }
2180
2181 /**
2182  * e1000_configure_tx - Configure 8254x Transmit Unit after Reset
2183  * @adapter: board private structure
2184  *
2185  * Configure the Tx unit of the MAC after a reset.
2186  **/
2187 static void e1000_configure_tx(struct e1000_adapter *adapter)
2188 {
2189         struct e1000_hw *hw = &adapter->hw;
2190         struct e1000_ring *tx_ring = adapter->tx_ring;
2191         u64 tdba;
2192         u32 tdlen, tctl, tipg, tarc;
2193         u32 ipgr1, ipgr2;
2194
2195         /* Setup the HW Tx Head and Tail descriptor pointers */
2196         tdba = tx_ring->dma;
2197         tdlen = tx_ring->count * sizeof(struct e1000_tx_desc);
2198         ew32(TDBAL, (tdba & DMA_32BIT_MASK));
2199         ew32(TDBAH, (tdba >> 32));
2200         ew32(TDLEN, tdlen);
2201         ew32(TDH, 0);
2202         ew32(TDT, 0);
2203         tx_ring->head = E1000_TDH;
2204         tx_ring->tail = E1000_TDT;
2205
2206         /* Set the default values for the Tx Inter Packet Gap timer */
2207         tipg = DEFAULT_82543_TIPG_IPGT_COPPER;          /*  8  */
2208         ipgr1 = DEFAULT_82543_TIPG_IPGR1;               /*  8  */
2209         ipgr2 = DEFAULT_82543_TIPG_IPGR2;               /*  6  */
2210
2211         if (adapter->flags & FLAG_TIPG_MEDIUM_FOR_80003ESLAN)
2212                 ipgr2 = DEFAULT_80003ES2LAN_TIPG_IPGR2; /*  7  */
2213
2214         tipg |= ipgr1 << E1000_TIPG_IPGR1_SHIFT;
2215         tipg |= ipgr2 << E1000_TIPG_IPGR2_SHIFT;
2216         ew32(TIPG, tipg);
2217
2218         /* Set the Tx Interrupt Delay register */
2219         ew32(TIDV, adapter->tx_int_delay);
2220         /* Tx irq moderation */
2221         ew32(TADV, adapter->tx_abs_int_delay);
2222
2223         /* Program the Transmit Control Register */
2224         tctl = er32(TCTL);
2225         tctl &= ~E1000_TCTL_CT;
2226         tctl |= E1000_TCTL_PSP | E1000_TCTL_RTLC |
2227                 (E1000_COLLISION_THRESHOLD << E1000_CT_SHIFT);
2228
2229         if (adapter->flags & FLAG_TARC_SPEED_MODE_BIT) {
2230                 tarc = er32(TARC(0));
2231                 /*
2232                  * set the speed mode bit, we'll clear it if we're not at
2233                  * gigabit link later
2234                  */
2235 #define SPEED_MODE_BIT (1 << 21)
2236                 tarc |= SPEED_MODE_BIT;
2237                 ew32(TARC(0), tarc);
2238         }
2239
2240         /* errata: program both queues to unweighted RR */
2241         if (adapter->flags & FLAG_TARC_SET_BIT_ZERO) {
2242                 tarc = er32(TARC(0));
2243                 tarc |= 1;
2244                 ew32(TARC(0), tarc);
2245                 tarc = er32(TARC(1));
2246                 tarc |= 1;
2247                 ew32(TARC(1), tarc);
2248         }
2249
2250         e1000e_config_collision_dist(hw);
2251
2252         /* Setup Transmit Descriptor Settings for eop descriptor */
2253         adapter->txd_cmd = E1000_TXD_CMD_EOP | E1000_TXD_CMD_IFCS;
2254
2255         /* only set IDE if we are delaying interrupts using the timers */
2256         if (adapter->tx_int_delay)
2257                 adapter->txd_cmd |= E1000_TXD_CMD_IDE;
2258
2259         /* enable Report Status bit */
2260         adapter->txd_cmd |= E1000_TXD_CMD_RS;
2261
2262         ew32(TCTL, tctl);
2263
2264         adapter->tx_queue_len = adapter->netdev->tx_queue_len;
2265 }
2266
2267 /**
2268  * e1000_setup_rctl - configure the receive control registers
2269  * @adapter: Board private structure
2270  **/
2271 #define PAGE_USE_COUNT(S) (((S) >> PAGE_SHIFT) + \
2272                            (((S) & (PAGE_SIZE - 1)) ? 1 : 0))
2273 static void e1000_setup_rctl(struct e1000_adapter *adapter)
2274 {
2275         struct e1000_hw *hw = &adapter->hw;
2276         u32 rctl, rfctl;
2277         u32 psrctl = 0;
2278         u32 pages = 0;
2279
2280         /* Program MC offset vector base */
2281         rctl = er32(RCTL);
2282         rctl &= ~(3 << E1000_RCTL_MO_SHIFT);
2283         rctl |= E1000_RCTL_EN | E1000_RCTL_BAM |
2284                 E1000_RCTL_LBM_NO | E1000_RCTL_RDMTS_HALF |
2285                 (adapter->hw.mac.mc_filter_type << E1000_RCTL_MO_SHIFT);
2286
2287         /* Do not Store bad packets */
2288         rctl &= ~E1000_RCTL_SBP;
2289
2290         /* Enable Long Packet receive */
2291         if (adapter->netdev->mtu <= ETH_DATA_LEN)
2292                 rctl &= ~E1000_RCTL_LPE;
2293         else
2294                 rctl |= E1000_RCTL_LPE;
2295
2296         /* Enable hardware CRC frame stripping */
2297         rctl |= E1000_RCTL_SECRC;
2298
2299         /* Setup buffer sizes */
2300         rctl &= ~E1000_RCTL_SZ_4096;
2301         rctl |= E1000_RCTL_BSEX;
2302         switch (adapter->rx_buffer_len) {
2303         case 256:
2304                 rctl |= E1000_RCTL_SZ_256;
2305                 rctl &= ~E1000_RCTL_BSEX;
2306                 break;
2307         case 512:
2308                 rctl |= E1000_RCTL_SZ_512;
2309                 rctl &= ~E1000_RCTL_BSEX;
2310                 break;
2311         case 1024:
2312                 rctl |= E1000_RCTL_SZ_1024;
2313                 rctl &= ~E1000_RCTL_BSEX;
2314                 break;
2315         case 2048:
2316         default:
2317                 rctl |= E1000_RCTL_SZ_2048;
2318                 rctl &= ~E1000_RCTL_BSEX;
2319                 break;
2320         case 4096:
2321                 rctl |= E1000_RCTL_SZ_4096;
2322                 break;
2323         case 8192:
2324                 rctl |= E1000_RCTL_SZ_8192;
2325                 break;
2326         case 16384:
2327                 rctl |= E1000_RCTL_SZ_16384;
2328                 break;
2329         }
2330
2331         /*
2332          * 82571 and greater support packet-split where the protocol
2333          * header is placed in skb->data and the packet data is
2334          * placed in pages hanging off of skb_shinfo(skb)->nr_frags.
2335          * In the case of a non-split, skb->data is linearly filled,
2336          * followed by the page buffers.  Therefore, skb->data is
2337          * sized to hold the largest protocol header.
2338          *
2339          * allocations using alloc_page take too long for regular MTU
2340          * so only enable packet split for jumbo frames
2341          *
2342          * Using pages when the page size is greater than 16k wastes
2343          * a lot of memory, since we allocate 3 pages at all times
2344          * per packet.
2345          */
2346         pages = PAGE_USE_COUNT(adapter->netdev->mtu);
2347         if (!(adapter->flags & FLAG_IS_ICH) && (pages <= 3) &&
2348             (PAGE_SIZE <= 16384) && (rctl & E1000_RCTL_LPE))
2349                 adapter->rx_ps_pages = pages;
2350         else
2351                 adapter->rx_ps_pages = 0;
2352
2353         if (adapter->rx_ps_pages) {
2354                 /* Configure extra packet-split registers */
2355                 rfctl = er32(RFCTL);
2356                 rfctl |= E1000_RFCTL_EXTEN;
2357                 /*
2358                  * disable packet split support for IPv6 extension headers,
2359                  * because some malformed IPv6 headers can hang the Rx
2360                  */
2361                 rfctl |= (E1000_RFCTL_IPV6_EX_DIS |
2362                           E1000_RFCTL_NEW_IPV6_EXT_DIS);
2363
2364                 ew32(RFCTL, rfctl);
2365
2366                 /* Enable Packet split descriptors */
2367                 rctl |= E1000_RCTL_DTYP_PS;
2368
2369                 psrctl |= adapter->rx_ps_bsize0 >>
2370                         E1000_PSRCTL_BSIZE0_SHIFT;
2371
2372                 switch (adapter->rx_ps_pages) {
2373                 case 3:
2374                         psrctl |= PAGE_SIZE <<
2375                                 E1000_PSRCTL_BSIZE3_SHIFT;
2376                 case 2:
2377                         psrctl |= PAGE_SIZE <<
2378                                 E1000_PSRCTL_BSIZE2_SHIFT;
2379                 case 1:
2380                         psrctl |= PAGE_SIZE >>
2381                                 E1000_PSRCTL_BSIZE1_SHIFT;
2382                         break;
2383                 }
2384
2385                 ew32(PSRCTL, psrctl);
2386         }
2387
2388         ew32(RCTL, rctl);
2389         /* just started the receive unit, no need to restart */
2390         adapter->flags &= ~FLAG_RX_RESTART_NOW;
2391 }
2392
2393 /**
2394  * e1000_configure_rx - Configure Receive Unit after Reset
2395  * @adapter: board private structure
2396  *
2397  * Configure the Rx unit of the MAC after a reset.
2398  **/
2399 static void e1000_configure_rx(struct e1000_adapter *adapter)
2400 {
2401         struct e1000_hw *hw = &adapter->hw;
2402         struct e1000_ring *rx_ring = adapter->rx_ring;
2403         u64 rdba;
2404         u32 rdlen, rctl, rxcsum, ctrl_ext;
2405
2406         if (adapter->rx_ps_pages) {
2407                 /* this is a 32 byte descriptor */
2408                 rdlen = rx_ring->count *
2409                         sizeof(union e1000_rx_desc_packet_split);
2410                 adapter->clean_rx = e1000_clean_rx_irq_ps;
2411                 adapter->alloc_rx_buf = e1000_alloc_rx_buffers_ps;
2412         } else if (adapter->netdev->mtu > ETH_FRAME_LEN + ETH_FCS_LEN) {
2413                 rdlen = rx_ring->count * sizeof(struct e1000_rx_desc);
2414                 adapter->clean_rx = e1000_clean_jumbo_rx_irq;
2415                 adapter->alloc_rx_buf = e1000_alloc_jumbo_rx_buffers;
2416         } else {
2417                 rdlen = rx_ring->count * sizeof(struct e1000_rx_desc);
2418                 adapter->clean_rx = e1000_clean_rx_irq;
2419                 adapter->alloc_rx_buf = e1000_alloc_rx_buffers;
2420         }
2421
2422         /* disable receives while setting up the descriptors */
2423         rctl = er32(RCTL);
2424         ew32(RCTL, rctl & ~E1000_RCTL_EN);
2425         e1e_flush();
2426         msleep(10);
2427
2428         /* set the Receive Delay Timer Register */
2429         ew32(RDTR, adapter->rx_int_delay);
2430
2431         /* irq moderation */
2432         ew32(RADV, adapter->rx_abs_int_delay);
2433         if (adapter->itr_setting != 0)
2434                 ew32(ITR, 1000000000 / (adapter->itr * 256));
2435
2436         ctrl_ext = er32(CTRL_EXT);
2437         /* Reset delay timers after every interrupt */
2438         ctrl_ext |= E1000_CTRL_EXT_INT_TIMER_CLR;
2439         /* Auto-Mask interrupts upon ICR access */
2440         ctrl_ext |= E1000_CTRL_EXT_IAME;
2441         ew32(IAM, 0xffffffff);
2442         ew32(CTRL_EXT, ctrl_ext);
2443         e1e_flush();
2444
2445         /*
2446          * Setup the HW Rx Head and Tail Descriptor Pointers and
2447          * the Base and Length of the Rx Descriptor Ring
2448          */
2449         rdba = rx_ring->dma;
2450         ew32(RDBAL, (rdba & DMA_32BIT_MASK));
2451         ew32(RDBAH, (rdba >> 32));
2452         ew32(RDLEN, rdlen);
2453         ew32(RDH, 0);
2454         ew32(RDT, 0);
2455         rx_ring->head = E1000_RDH;
2456         rx_ring->tail = E1000_RDT;
2457
2458         /* Enable Receive Checksum Offload for TCP and UDP */
2459         rxcsum = er32(RXCSUM);
2460         if (adapter->flags & FLAG_RX_CSUM_ENABLED) {
2461                 rxcsum |= E1000_RXCSUM_TUOFL;
2462
2463                 /*
2464                  * IPv4 payload checksum for UDP fragments must be
2465                  * used in conjunction with packet-split.
2466                  */
2467                 if (adapter->rx_ps_pages)
2468                         rxcsum |= E1000_RXCSUM_IPPCSE;
2469         } else {
2470                 rxcsum &= ~E1000_RXCSUM_TUOFL;
2471                 /* no need to clear IPPCSE as it defaults to 0 */
2472         }
2473         ew32(RXCSUM, rxcsum);
2474
2475         /*
2476          * Enable early receives on supported devices, only takes effect when
2477          * packet size is equal or larger than the specified value (in 8 byte
2478          * units), e.g. using jumbo frames when setting to E1000_ERT_2048
2479          */
2480         if ((adapter->flags & FLAG_HAS_ERT) &&
2481             (adapter->netdev->mtu > ETH_DATA_LEN)) {
2482                 u32 rxdctl = er32(RXDCTL(0));
2483                 ew32(RXDCTL(0), rxdctl | 0x3);
2484                 ew32(ERT, E1000_ERT_2048 | (1 << 13));
2485                 /*
2486                  * With jumbo frames and early-receive enabled, excessive
2487                  * C4->C2 latencies result in dropped transactions.
2488                  */
2489                 pm_qos_update_requirement(PM_QOS_CPU_DMA_LATENCY,
2490                                           e1000e_driver_name, 55);
2491         } else {
2492                 pm_qos_update_requirement(PM_QOS_CPU_DMA_LATENCY,
2493                                           e1000e_driver_name,
2494                                           PM_QOS_DEFAULT_VALUE);
2495         }
2496
2497         /* Enable Receives */
2498         ew32(RCTL, rctl);
2499 }
2500
2501 /**
2502  *  e1000_update_mc_addr_list - Update Multicast addresses
2503  *  @hw: pointer to the HW structure
2504  *  @mc_addr_list: array of multicast addresses to program
2505  *  @mc_addr_count: number of multicast addresses to program
2506  *  @rar_used_count: the first RAR register free to program
2507  *  @rar_count: total number of supported Receive Address Registers
2508  *
2509  *  Updates the Receive Address Registers and Multicast Table Array.
2510  *  The caller must have a packed mc_addr_list of multicast addresses.
2511  *  The parameter rar_count will usually be hw->mac.rar_entry_count
2512  *  unless there are workarounds that change this.  Currently no func pointer
2513  *  exists and all implementations are handled in the generic version of this
2514  *  function.
2515  **/
2516 static void e1000_update_mc_addr_list(struct e1000_hw *hw, u8 *mc_addr_list,
2517                                       u32 mc_addr_count, u32 rar_used_count,
2518                                       u32 rar_count)
2519 {
2520         hw->mac.ops.update_mc_addr_list(hw, mc_addr_list, mc_addr_count,
2521                                         rar_used_count, rar_count);
2522 }
2523
2524 /**
2525  * e1000_set_multi - Multicast and Promiscuous mode set
2526  * @netdev: network interface device structure
2527  *
2528  * The set_multi entry point is called whenever the multicast address
2529  * list or the network interface flags are updated.  This routine is
2530  * responsible for configuring the hardware for proper multicast,
2531  * promiscuous mode, and all-multi behavior.
2532  **/
2533 static void e1000_set_multi(struct net_device *netdev)
2534 {
2535         struct e1000_adapter *adapter = netdev_priv(netdev);
2536         struct e1000_hw *hw = &adapter->hw;
2537         struct e1000_mac_info *mac = &hw->mac;
2538         struct dev_mc_list *mc_ptr;
2539         u8  *mta_list;
2540         u32 rctl;
2541         int i;
2542
2543         /* Check for Promiscuous and All Multicast modes */
2544
2545         rctl = er32(RCTL);
2546
2547         if (netdev->flags & IFF_PROMISC) {
2548                 rctl |= (E1000_RCTL_UPE | E1000_RCTL_MPE);
2549                 rctl &= ~E1000_RCTL_VFE;
2550         } else {
2551                 if (netdev->flags & IFF_ALLMULTI) {
2552                         rctl |= E1000_RCTL_MPE;
2553                         rctl &= ~E1000_RCTL_UPE;
2554                 } else {
2555                         rctl &= ~(E1000_RCTL_UPE | E1000_RCTL_MPE);
2556                 }
2557                 if (adapter->flags & FLAG_HAS_HW_VLAN_FILTER)
2558                         rctl |= E1000_RCTL_VFE;
2559         }
2560
2561         ew32(RCTL, rctl);
2562
2563         if (netdev->mc_count) {
2564                 mta_list = kmalloc(netdev->mc_count * 6, GFP_ATOMIC);
2565                 if (!mta_list)
2566                         return;
2567
2568                 /* prepare a packed array of only addresses. */
2569                 mc_ptr = netdev->mc_list;
2570
2571                 for (i = 0; i < netdev->mc_count; i++) {
2572                         if (!mc_ptr)
2573                                 break;
2574                         memcpy(mta_list + (i*ETH_ALEN), mc_ptr->dmi_addr,
2575                                ETH_ALEN);
2576                         mc_ptr = mc_ptr->next;
2577                 }
2578
2579                 e1000_update_mc_addr_list(hw, mta_list, i, 1,
2580                                           mac->rar_entry_count);
2581                 kfree(mta_list);
2582         } else {
2583                 /*
2584                  * if we're called from probe, we might not have
2585                  * anything to do here, so clear out the list
2586                  */
2587                 e1000_update_mc_addr_list(hw, NULL, 0, 1, mac->rar_entry_count);
2588         }
2589 }
2590
2591 /**
2592  * e1000_configure - configure the hardware for Rx and Tx
2593  * @adapter: private board structure
2594  **/
2595 static void e1000_configure(struct e1000_adapter *adapter)
2596 {
2597         e1000_set_multi(adapter->netdev);
2598
2599         e1000_restore_vlan(adapter);
2600         e1000_init_manageability(adapter);
2601
2602         e1000_configure_tx(adapter);
2603         e1000_setup_rctl(adapter);
2604         e1000_configure_rx(adapter);
2605         adapter->alloc_rx_buf(adapter, e1000_desc_unused(adapter->rx_ring));
2606 }
2607
2608 /**
2609  * e1000e_power_up_phy - restore link in case the phy was powered down
2610  * @adapter: address of board private structure
2611  *
2612  * The phy may be powered down to save power and turn off link when the
2613  * driver is unloaded and wake on lan is not enabled (among others)
2614  * *** this routine MUST be followed by a call to e1000e_reset ***
2615  **/
2616 void e1000e_power_up_phy(struct e1000_adapter *adapter)
2617 {
2618         u16 mii_reg = 0;
2619
2620         /* Just clear the power down bit to wake the phy back up */
2621         if (adapter->hw.phy.media_type == e1000_media_type_copper) {
2622                 /*
2623                  * According to the manual, the phy will retain its
2624                  * settings across a power-down/up cycle
2625                  */
2626                 e1e_rphy(&adapter->hw, PHY_CONTROL, &mii_reg);
2627                 mii_reg &= ~MII_CR_POWER_DOWN;
2628                 e1e_wphy(&adapter->hw, PHY_CONTROL, mii_reg);
2629         }
2630
2631         adapter->hw.mac.ops.setup_link(&adapter->hw);
2632 }
2633
2634 /**
2635  * e1000_power_down_phy - Power down the PHY
2636  *
2637  * Power down the PHY so no link is implied when interface is down
2638  * The PHY cannot be powered down is management or WoL is active
2639  */
2640 static void e1000_power_down_phy(struct e1000_adapter *adapter)
2641 {
2642         struct e1000_hw *hw = &adapter->hw;
2643         u16 mii_reg;
2644
2645         /* WoL is enabled */
2646         if (adapter->wol)
2647                 return;
2648
2649         /* non-copper PHY? */
2650         if (adapter->hw.phy.media_type != e1000_media_type_copper)
2651                 return;
2652
2653         /* reset is blocked because of a SoL/IDER session */
2654         if (e1000e_check_mng_mode(hw) || e1000_check_reset_block(hw))
2655                 return;
2656
2657         /* manageability (AMT) is enabled */
2658         if (er32(MANC) & E1000_MANC_SMBUS_EN)
2659                 return;
2660
2661         /* power down the PHY */
2662         e1e_rphy(hw, PHY_CONTROL, &mii_reg);
2663         mii_reg |= MII_CR_POWER_DOWN;
2664         e1e_wphy(hw, PHY_CONTROL, mii_reg);
2665         mdelay(1);
2666 }
2667
2668 /**
2669  * e1000e_reset - bring the hardware into a known good state
2670  *
2671  * This function boots the hardware and enables some settings that
2672  * require a configuration cycle of the hardware - those cannot be
2673  * set/changed during runtime. After reset the device needs to be
2674  * properly configured for Rx, Tx etc.
2675  */
2676 void e1000e_reset(struct e1000_adapter *adapter)
2677 {
2678         struct e1000_mac_info *mac = &adapter->hw.mac;
2679         struct e1000_fc_info *fc = &adapter->hw.fc;
2680         struct e1000_hw *hw = &adapter->hw;
2681         u32 tx_space, min_tx_space, min_rx_space;
2682         u32 pba = adapter->pba;
2683         u16 hwm;
2684
2685         /* reset Packet Buffer Allocation to default */
2686         ew32(PBA, pba);
2687
2688         if (adapter->max_frame_size > ETH_FRAME_LEN + ETH_FCS_LEN) {
2689                 /*
2690                  * To maintain wire speed transmits, the Tx FIFO should be
2691                  * large enough to accommodate two full transmit packets,
2692                  * rounded up to the next 1KB and expressed in KB.  Likewise,
2693                  * the Rx FIFO should be large enough to accommodate at least
2694                  * one full receive packet and is similarly rounded up and
2695                  * expressed in KB.
2696                  */
2697                 pba = er32(PBA);
2698                 /* upper 16 bits has Tx packet buffer allocation size in KB */
2699                 tx_space = pba >> 16;
2700                 /* lower 16 bits has Rx packet buffer allocation size in KB */
2701                 pba &= 0xffff;
2702                 /*
2703                  * the Tx fifo also stores 16 bytes of information about the tx
2704                  * but don't include ethernet FCS because hardware appends it
2705                  */
2706                 min_tx_space = (adapter->max_frame_size +
2707                                 sizeof(struct e1000_tx_desc) -
2708                                 ETH_FCS_LEN) * 2;
2709                 min_tx_space = ALIGN(min_tx_space, 1024);
2710                 min_tx_space >>= 10;
2711                 /* software strips receive CRC, so leave room for it */
2712                 min_rx_space = adapter->max_frame_size;
2713                 min_rx_space = ALIGN(min_rx_space, 1024);
2714                 min_rx_space >>= 10;
2715
2716                 /*
2717                  * If current Tx allocation is less than the min Tx FIFO size,
2718                  * and the min Tx FIFO size is less than the current Rx FIFO
2719                  * allocation, take space away from current Rx allocation
2720                  */
2721                 if ((tx_space < min_tx_space) &&
2722                     ((min_tx_space - tx_space) < pba)) {
2723                         pba -= min_tx_space - tx_space;
2724
2725                         /*
2726                          * if short on Rx space, Rx wins and must trump tx
2727                          * adjustment or use Early Receive if available
2728                          */
2729                         if ((pba < min_rx_space) &&
2730                             (!(adapter->flags & FLAG_HAS_ERT)))
2731                                 /* ERT enabled in e1000_configure_rx */
2732                                 pba = min_rx_space;
2733                 }
2734
2735                 ew32(PBA, pba);
2736         }
2737
2738
2739         /*
2740          * flow control settings
2741          *
2742          * The high water mark must be low enough to fit one full frame
2743          * (or the size used for early receive) above it in the Rx FIFO.
2744          * Set it to the lower of:
2745          * - 90% of the Rx FIFO size, and
2746          * - the full Rx FIFO size minus the early receive size (for parts
2747          *   with ERT support assuming ERT set to E1000_ERT_2048), or
2748          * - the full Rx FIFO size minus one full frame
2749          */
2750         if (adapter->flags & FLAG_HAS_ERT)
2751                 hwm = min(((pba << 10) * 9 / 10),
2752                           ((pba << 10) - (E1000_ERT_2048 << 3)));
2753         else
2754                 hwm = min(((pba << 10) * 9 / 10),
2755                           ((pba << 10) - adapter->max_frame_size));
2756
2757         fc->high_water = hwm & 0xFFF8; /* 8-byte granularity */
2758         fc->low_water = fc->high_water - 8;
2759
2760         if (adapter->flags & FLAG_DISABLE_FC_PAUSE_TIME)
2761                 fc->pause_time = 0xFFFF;
2762         else
2763                 fc->pause_time = E1000_FC_PAUSE_TIME;
2764         fc->send_xon = 1;
2765         fc->type = fc->original_type;
2766
2767         /* Allow time for pending master requests to run */
2768         mac->ops.reset_hw(hw);
2769
2770         /*
2771          * For parts with AMT enabled, let the firmware know
2772          * that the network interface is in control
2773          */
2774         if (adapter->flags & FLAG_HAS_AMT)
2775                 e1000_get_hw_control(adapter);
2776
2777         ew32(WUC, 0);
2778
2779         if (mac->ops.init_hw(hw))
2780                 e_err("Hardware Error\n");
2781
2782         e1000_update_mng_vlan(adapter);
2783
2784         /* Enable h/w to recognize an 802.1Q VLAN Ethernet packet */
2785         ew32(VET, ETH_P_8021Q);
2786
2787         e1000e_reset_adaptive(hw);
2788         e1000_get_phy_info(hw);
2789
2790         if (!(adapter->flags & FLAG_SMART_POWER_DOWN)) {
2791                 u16 phy_data = 0;
2792                 /*
2793                  * speed up time to link by disabling smart power down, ignore
2794                  * the return value of this function because there is nothing
2795                  * different we would do if it failed
2796                  */
2797                 e1e_rphy(hw, IGP02E1000_PHY_POWER_MGMT, &phy_data);
2798                 phy_data &= ~IGP02E1000_PM_SPD;
2799                 e1e_wphy(hw, IGP02E1000_PHY_POWER_MGMT, phy_data);
2800         }
2801 }
2802
2803 int e1000e_up(struct e1000_adapter *adapter)
2804 {
2805         struct e1000_hw *hw = &adapter->hw;
2806
2807         /* hardware has been reset, we need to reload some things */
2808         e1000_configure(adapter);
2809
2810         clear_bit(__E1000_DOWN, &adapter->state);
2811
2812         napi_enable(&adapter->napi);
2813         if (adapter->msix_entries)
2814                 e1000_configure_msix(adapter);
2815         e1000_irq_enable(adapter);
2816
2817         /* fire a link change interrupt to start the watchdog */
2818         ew32(ICS, E1000_ICS_LSC);
2819         return 0;
2820 }
2821
2822 void e1000e_down(struct e1000_adapter *adapter)
2823 {
2824         struct net_device *netdev = adapter->netdev;
2825         struct e1000_hw *hw = &adapter->hw;
2826         u32 tctl, rctl;
2827
2828         /*
2829          * signal that we're down so the interrupt handler does not
2830          * reschedule our watchdog timer
2831          */
2832         set_bit(__E1000_DOWN, &adapter->state);
2833
2834         /* disable receives in the hardware */
2835         rctl = er32(RCTL);
2836         ew32(RCTL, rctl & ~E1000_RCTL_EN);
2837         /* flush and sleep below */
2838
2839         netif_tx_stop_all_queues(netdev);
2840
2841         /* disable transmits in the hardware */
2842         tctl = er32(TCTL);
2843         tctl &= ~E1000_TCTL_EN;
2844         ew32(TCTL, tctl);
2845         /* flush both disables and wait for them to finish */
2846         e1e_flush();
2847         msleep(10);
2848
2849         napi_disable(&adapter->napi);
2850         e1000_irq_disable(adapter);
2851
2852         del_timer_sync(&adapter->watchdog_timer);
2853         del_timer_sync(&adapter->phy_info_timer);
2854
2855         netdev->tx_queue_len = adapter->tx_queue_len;
2856         netif_carrier_off(netdev);
2857         adapter->link_speed = 0;
2858         adapter->link_duplex = 0;
2859
2860         if (!pci_channel_offline(adapter->pdev))
2861                 e1000e_reset(adapter);
2862         e1000_clean_tx_ring(adapter);
2863         e1000_clean_rx_ring(adapter);
2864
2865         /*
2866          * TODO: for power management, we could drop the link and
2867          * pci_disable_device here.
2868          */
2869 }
2870
2871 void e1000e_reinit_locked(struct e1000_adapter *adapter)
2872 {
2873         might_sleep();
2874         while (test_and_set_bit(__E1000_RESETTING, &adapter->state))
2875                 msleep(1);
2876         e1000e_down(adapter);
2877         e1000e_up(adapter);
2878         clear_bit(__E1000_RESETTING, &adapter->state);
2879 }
2880
2881 /**
2882  * e1000_sw_init - Initialize general software structures (struct e1000_adapter)
2883  * @adapter: board private structure to initialize
2884  *
2885  * e1000_sw_init initializes the Adapter private data structure.
2886  * Fields are initialized based on PCI device information and
2887  * OS network device settings (MTU size).
2888  **/
2889 static int __devinit e1000_sw_init(struct e1000_adapter *adapter)
2890 {
2891         struct net_device *netdev = adapter->netdev;
2892
2893         adapter->rx_buffer_len = ETH_FRAME_LEN + VLAN_HLEN + ETH_FCS_LEN;
2894         adapter->rx_ps_bsize0 = 128;
2895         adapter->max_frame_size = netdev->mtu + ETH_HLEN + ETH_FCS_LEN;
2896         adapter->min_frame_size = ETH_ZLEN + ETH_FCS_LEN;
2897
2898         e1000e_set_interrupt_capability(adapter);
2899
2900         if (e1000_alloc_queues(adapter))
2901                 return -ENOMEM;
2902
2903         spin_lock_init(&adapter->tx_queue_lock);
2904
2905         /* Explicitly disable IRQ since the NIC can be in any state. */
2906         e1000_irq_disable(adapter);
2907
2908         spin_lock_init(&adapter->stats_lock);
2909
2910         set_bit(__E1000_DOWN, &adapter->state);
2911         return 0;
2912 }
2913
2914 /**
2915  * e1000_intr_msi_test - Interrupt Handler
2916  * @irq: interrupt number
2917  * @data: pointer to a network interface device structure
2918  **/
2919 static irqreturn_t e1000_intr_msi_test(int irq, void *data)
2920 {
2921         struct net_device *netdev = data;
2922         struct e1000_adapter *adapter = netdev_priv(netdev);
2923         struct e1000_hw *hw = &adapter->hw;
2924         u32 icr = er32(ICR);
2925
2926         e_dbg("%s: icr is %08X\n", netdev->name, icr);
2927         if (icr & E1000_ICR_RXSEQ) {
2928                 adapter->flags &= ~FLAG_MSI_TEST_FAILED;
2929                 wmb();
2930         }
2931
2932         return IRQ_HANDLED;
2933 }
2934
2935 /**
2936  * e1000_test_msi_interrupt - Returns 0 for successful test
2937  * @adapter: board private struct
2938  *
2939  * code flow taken from tg3.c
2940  **/
2941 static int e1000_test_msi_interrupt(struct e1000_adapter *adapter)
2942 {
2943         struct net_device *netdev = adapter->netdev;
2944         struct e1000_hw *hw = &adapter->hw;
2945         int err;
2946
2947         /* poll_enable hasn't been called yet, so don't need disable */
2948         /* clear any pending events */
2949         er32(ICR);
2950
2951         /* free the real vector and request a test handler */
2952         e1000_free_irq(adapter);
2953         e1000e_reset_interrupt_capability(adapter);
2954
2955         /* Assume that the test fails, if it succeeds then the test
2956          * MSI irq handler will unset this flag */
2957         adapter->flags |= FLAG_MSI_TEST_FAILED;
2958
2959         err = pci_enable_msi(adapter->pdev);
2960         if (err)
2961                 goto msi_test_failed;
2962
2963         err = request_irq(adapter->pdev->irq, &e1000_intr_msi_test, 0,
2964                           netdev->name, netdev);
2965         if (err) {
2966                 pci_disable_msi(adapter->pdev);
2967                 goto msi_test_failed;
2968         }
2969
2970         wmb();
2971
2972         e1000_irq_enable(adapter);
2973
2974         /* fire an unusual interrupt on the test handler */
2975         ew32(ICS, E1000_ICS_RXSEQ);
2976         e1e_flush();
2977         msleep(50);
2978
2979         e1000_irq_disable(adapter);
2980
2981         rmb();
2982
2983         if (adapter->flags & FLAG_MSI_TEST_FAILED) {
2984                 adapter->int_mode = E1000E_INT_MODE_LEGACY;
2985                 err = -EIO;
2986                 e_info("MSI interrupt test failed!\n");
2987         }
2988
2989         free_irq(adapter->pdev->irq, netdev);
2990         pci_disable_msi(adapter->pdev);
2991
2992         if (err == -EIO)
2993                 goto msi_test_failed;
2994
2995         /* okay so the test worked, restore settings */
2996         e_dbg("%s: MSI interrupt test succeeded!\n", netdev->name);
2997 msi_test_failed:
2998         e1000e_set_interrupt_capability(adapter);
2999         e1000_request_irq(adapter);
3000         return err;
3001 }
3002
3003 /**
3004  * e1000_test_msi - Returns 0 if MSI test succeeds or INTx mode is restored
3005  * @adapter: board private struct
3006  *
3007  * code flow taken from tg3.c, called with e1000 interrupts disabled.
3008  **/
3009 static int e1000_test_msi(struct e1000_adapter *adapter)
3010 {
3011         int err;
3012         u16 pci_cmd;
3013
3014         if (!(adapter->flags & FLAG_MSI_ENABLED))
3015                 return 0;
3016
3017         /* disable SERR in case the MSI write causes a master abort */
3018         pci_read_config_word(adapter->pdev, PCI_COMMAND, &pci_cmd);
3019         pci_write_config_word(adapter->pdev, PCI_COMMAND,
3020                               pci_cmd & ~PCI_COMMAND_SERR);
3021
3022         err = e1000_test_msi_interrupt(adapter);
3023
3024         /* restore previous setting of command word */
3025         pci_write_config_word(adapter->pdev, PCI_COMMAND, pci_cmd);
3026
3027         /* success ! */
3028         if (!err)
3029                 return 0;
3030
3031         /* EIO means MSI test failed */
3032         if (err != -EIO)
3033                 return err;
3034
3035         /* back to INTx mode */
3036         e_warn("MSI interrupt test failed, using legacy interrupt.\n");
3037
3038         e1000_free_irq(adapter);
3039
3040         err = e1000_request_irq(adapter);
3041
3042         return err;
3043 }
3044
3045 /**
3046  * e1000_open - Called when a network interface is made active
3047  * @netdev: network interface device structure
3048  *
3049  * Returns 0 on success, negative value on failure
3050  *
3051  * The open entry point is called when a network interface is made
3052  * active by the system (IFF_UP).  At this point all resources needed
3053  * for transmit and receive operations are allocated, the interrupt
3054  * handler is registered with the OS, the watchdog timer is started,
3055  * and the stack is notified that the interface is ready.
3056  **/
3057 static int e1000_open(struct net_device *netdev)
3058 {
3059         struct e1000_adapter *adapter = netdev_priv(netdev);
3060         struct e1000_hw *hw = &adapter->hw;
3061         int err;
3062
3063         /* disallow open during test */
3064         if (test_bit(__E1000_TESTING, &adapter->state))
3065                 return -EBUSY;
3066
3067         /* allocate transmit descriptors */
3068         err = e1000e_setup_tx_resources(adapter);
3069         if (err)
3070                 goto err_setup_tx;
3071
3072         /* allocate receive descriptors */
3073         err = e1000e_setup_rx_resources(adapter);
3074         if (err)
3075                 goto err_setup_rx;
3076
3077         e1000e_power_up_phy(adapter);
3078
3079         adapter->mng_vlan_id = E1000_MNG_VLAN_NONE;
3080         if ((adapter->hw.mng_cookie.status &
3081              E1000_MNG_DHCP_COOKIE_STATUS_VLAN))
3082                 e1000_update_mng_vlan(adapter);
3083
3084         /*
3085          * If AMT is enabled, let the firmware know that the network
3086          * interface is now open
3087          */
3088         if (adapter->flags & FLAG_HAS_AMT)
3089                 e1000_get_hw_control(adapter);
3090
3091         /*
3092          * before we allocate an interrupt, we must be ready to handle it.
3093          * Setting DEBUG_SHIRQ in the kernel makes it fire an interrupt
3094          * as soon as we call pci_request_irq, so we have to setup our
3095          * clean_rx handler before we do so.
3096          */
3097         e1000_configure(adapter);
3098
3099         err = e1000_request_irq(adapter);
3100         if (err)
3101                 goto err_req_irq;
3102
3103         /*
3104          * Work around PCIe errata with MSI interrupts causing some chipsets to
3105          * ignore e1000e MSI messages, which means we need to test our MSI
3106          * interrupt now
3107          */
3108         if (adapter->int_mode != E1000E_INT_MODE_LEGACY) {
3109                 err = e1000_test_msi(adapter);
3110                 if (err) {
3111                         e_err("Interrupt allocation failed\n");
3112                         goto err_req_irq;
3113                 }
3114         }
3115
3116         /* From here on the code is the same as e1000e_up() */
3117         clear_bit(__E1000_DOWN, &adapter->state);
3118
3119         napi_enable(&adapter->napi);
3120
3121         e1000_irq_enable(adapter);
3122
3123         netif_tx_start_all_queues(netdev);
3124
3125         /* fire a link status change interrupt to start the watchdog */
3126         ew32(ICS, E1000_ICS_LSC);
3127
3128         return 0;
3129
3130 err_req_irq:
3131         e1000_release_hw_control(adapter);
3132         e1000_power_down_phy(adapter);
3133         e1000e_free_rx_resources(adapter);
3134 err_setup_rx:
3135         e1000e_free_tx_resources(adapter);
3136 err_setup_tx:
3137         e1000e_reset(adapter);
3138
3139         return err;
3140 }
3141
3142 /**
3143  * e1000_close - Disables a network interface
3144  * @netdev: network interface device structure
3145  *
3146  * Returns 0, this is not allowed to fail
3147  *
3148  * The close entry point is called when an interface is de-activated
3149  * by the OS.  The hardware is still under the drivers control, but
3150  * needs to be disabled.  A global MAC reset is issued to stop the
3151  * hardware, and all transmit and receive resources are freed.
3152  **/
3153 static int e1000_close(struct net_device *netdev)
3154 {
3155         struct e1000_adapter *adapter = netdev_priv(netdev);
3156
3157         WARN_ON(test_bit(__E1000_RESETTING, &adapter->state));
3158         e1000e_down(adapter);
3159         e1000_power_down_phy(adapter);
3160         e1000_free_irq(adapter);
3161
3162         e1000e_free_tx_resources(adapter);
3163         e1000e_free_rx_resources(adapter);
3164
3165         /*
3166          * kill manageability vlan ID if supported, but not if a vlan with
3167          * the same ID is registered on the host OS (let 8021q kill it)
3168          */
3169         if ((adapter->hw.mng_cookie.status &
3170                           E1000_MNG_DHCP_COOKIE_STATUS_VLAN) &&
3171              !(adapter->vlgrp &&
3172                vlan_group_get_device(adapter->vlgrp, adapter->mng_vlan_id)))
3173                 e1000_vlan_rx_kill_vid(netdev, adapter->mng_vlan_id);
3174
3175         /*
3176          * If AMT is enabled, let the firmware know that the network
3177          * interface is now closed
3178          */
3179         if (adapter->flags & FLAG_HAS_AMT)
3180                 e1000_release_hw_control(adapter);
3181
3182         return 0;
3183 }
3184 /**
3185  * e1000_set_mac - Change the Ethernet Address of the NIC
3186  * @netdev: network interface device structure
3187  * @p: pointer to an address structure
3188  *
3189  * Returns 0 on success, negative on failure
3190  **/
3191 static int e1000_set_mac(struct net_device *netdev, void *p)
3192 {
3193         struct e1000_adapter *adapter = netdev_priv(netdev);
3194         struct sockaddr *addr = p;
3195
3196         if (!is_valid_ether_addr(addr->sa_data))
3197                 return -EADDRNOTAVAIL;
3198
3199         memcpy(netdev->dev_addr, addr->sa_data, netdev->addr_len);
3200         memcpy(adapter->hw.mac.addr, addr->sa_data, netdev->addr_len);
3201
3202         e1000e_rar_set(&adapter->hw, adapter->hw.mac.addr, 0);
3203
3204         if (adapter->flags & FLAG_RESET_OVERWRITES_LAA) {
3205                 /* activate the work around */
3206                 e1000e_set_laa_state_82571(&adapter->hw, 1);
3207
3208                 /*
3209                  * Hold a copy of the LAA in RAR[14] This is done so that
3210                  * between the time RAR[0] gets clobbered  and the time it
3211                  * gets fixed (in e1000_watchdog), the actual LAA is in one
3212                  * of the RARs and no incoming packets directed to this port
3213                  * are dropped. Eventually the LAA will be in RAR[0] and
3214                  * RAR[14]
3215                  */
3216                 e1000e_rar_set(&adapter->hw,
3217                               adapter->hw.mac.addr,
3218                               adapter->hw.mac.rar_entry_count - 1);
3219         }
3220
3221         return 0;
3222 }
3223
3224 /*
3225  * Need to wait a few seconds after link up to get diagnostic information from
3226  * the phy
3227  */
3228 static void e1000_update_phy_info(unsigned long data)
3229 {
3230         struct e1000_adapter *adapter = (struct e1000_adapter *) data;
3231         e1000_get_phy_info(&adapter->hw);
3232 }
3233
3234 /**
3235  * e1000e_update_stats - Update the board statistics counters
3236  * @adapter: board private structure
3237  **/
3238 void e1000e_update_stats(struct e1000_adapter *adapter)
3239 {
3240         struct e1000_hw *hw = &adapter->hw;
3241         struct pci_dev *pdev = adapter->pdev;
3242         unsigned long irq_flags;
3243         u16 phy_tmp;
3244
3245 #define PHY_IDLE_ERROR_COUNT_MASK 0x00FF
3246
3247         /*
3248          * Prevent stats update while adapter is being reset, or if the pci
3249          * connection is down.
3250          */
3251         if (adapter->link_speed == 0)
3252                 return;
3253         if (pci_channel_offline(pdev))
3254                 return;
3255
3256         spin_lock_irqsave(&adapter->stats_lock, irq_flags);
3257
3258         /*
3259          * these counters are modified from e1000_adjust_tbi_stats,
3260          * called from the interrupt context, so they must only
3261          * be written while holding adapter->stats_lock
3262          */
3263
3264         adapter->stats.crcerrs += er32(CRCERRS);
3265         adapter->stats.gprc += er32(GPRC);
3266         adapter->stats.gorc += er32(GORCL);
3267         er32(GORCH); /* Clear gorc */
3268         adapter->stats.bprc += er32(BPRC);
3269         adapter->stats.mprc += er32(MPRC);
3270         adapter->stats.roc += er32(ROC);
3271
3272         adapter->stats.mpc += er32(MPC);
3273         adapter->stats.scc += er32(SCC);
3274         adapter->stats.ecol += er32(ECOL);
3275         adapter->stats.mcc += er32(MCC);
3276         adapter->stats.latecol += er32(LATECOL);
3277         adapter->stats.dc += er32(DC);
3278         adapter->stats.xonrxc += er32(XONRXC);
3279         adapter->stats.xontxc += er32(XONTXC);
3280         adapter->stats.xoffrxc += er32(XOFFRXC);
3281         adapter->stats.xofftxc += er32(XOFFTXC);
3282         adapter->stats.gptc += er32(GPTC);
3283         adapter->stats.gotc += er32(GOTCL);
3284         er32(GOTCH); /* Clear gotc */
3285         adapter->stats.rnbc += er32(RNBC);
3286         adapter->stats.ruc += er32(RUC);
3287
3288         adapter->stats.mptc += er32(MPTC);
3289         adapter->stats.bptc += er32(BPTC);
3290
3291         /* used for adaptive IFS */
3292
3293         hw->mac.tx_packet_delta = er32(TPT);
3294         adapter->stats.tpt += hw->mac.tx_packet_delta;
3295         hw->mac.collision_delta = er32(COLC);
3296         adapter->stats.colc += hw->mac.collision_delta;
3297
3298         adapter->stats.algnerrc += er32(ALGNERRC);
3299         adapter->stats.rxerrc += er32(RXERRC);
3300         if (hw->mac.type != e1000_82574)
3301                 adapter->stats.tncrs += er32(TNCRS);
3302         adapter->stats.cexterr += er32(CEXTERR);
3303         adapter->stats.tsctc += er32(TSCTC);
3304         adapter->stats.tsctfc += er32(TSCTFC);
3305
3306         /* Fill out the OS statistics structure */
3307         adapter->net_stats.multicast = adapter->stats.mprc;
3308         adapter->net_stats.collisions = adapter->stats.colc;
3309
3310         /* Rx Errors */
3311
3312         /*
3313          * RLEC on some newer hardware can be incorrect so build
3314          * our own version based on RUC and ROC
3315          */
3316         adapter->net_stats.rx_errors = adapter->stats.rxerrc +
3317                 adapter->stats.crcerrs + adapter->stats.algnerrc +
3318                 adapter->stats.ruc + adapter->stats.roc +
3319                 adapter->stats.cexterr;
3320         adapter->net_stats.rx_length_errors = adapter->stats.ruc +
3321                                               adapter->stats.roc;
3322         adapter->net_stats.rx_crc_errors = adapter->stats.crcerrs;
3323         adapter->net_stats.rx_frame_errors = adapter->stats.algnerrc;
3324         adapter->net_stats.rx_missed_errors = adapter->stats.mpc;
3325
3326         /* Tx Errors */
3327         adapter->net_stats.tx_errors = adapter->stats.ecol +
3328                                        adapter->stats.latecol;
3329         adapter->net_stats.tx_aborted_errors = adapter->stats.ecol;
3330         adapter->net_stats.tx_window_errors = adapter->stats.latecol;
3331         adapter->net_stats.tx_carrier_errors = adapter->stats.tncrs;
3332
3333         /* Tx Dropped needs to be maintained elsewhere */
3334
3335         /* Phy Stats */
3336         if (hw->phy.media_type == e1000_media_type_copper) {
3337                 if ((adapter->link_speed == SPEED_1000) &&
3338                    (!e1e_rphy(hw, PHY_1000T_STATUS, &phy_tmp))) {
3339                         phy_tmp &= PHY_IDLE_ERROR_COUNT_MASK;
3340                         adapter->phy_stats.idle_errors += phy_tmp;
3341                 }
3342         }
3343
3344         /* Management Stats */
3345         adapter->stats.mgptc += er32(MGTPTC);
3346         adapter->stats.mgprc += er32(MGTPRC);
3347         adapter->stats.mgpdc += er32(MGTPDC);
3348
3349         spin_unlock_irqrestore(&adapter->stats_lock, irq_flags);
3350 }
3351
3352 /**
3353  * e1000_phy_read_status - Update the PHY register status snapshot
3354  * @adapter: board private structure
3355  **/
3356 static void e1000_phy_read_status(struct e1000_adapter *adapter)
3357 {
3358         struct e1000_hw *hw = &adapter->hw;
3359         struct e1000_phy_regs *phy = &adapter->phy_regs;
3360         int ret_val;
3361         unsigned long irq_flags;
3362
3363
3364         spin_lock_irqsave(&adapter->stats_lock, irq_flags);
3365
3366         if ((er32(STATUS) & E1000_STATUS_LU) &&
3367             (adapter->hw.phy.media_type == e1000_media_type_copper)) {
3368                 ret_val  = e1e_rphy(hw, PHY_CONTROL, &phy->bmcr);
3369                 ret_val |= e1e_rphy(hw, PHY_STATUS, &phy->bmsr);
3370                 ret_val |= e1e_rphy(hw, PHY_AUTONEG_ADV, &phy->advertise);
3371                 ret_val |= e1e_rphy(hw, PHY_LP_ABILITY, &phy->lpa);
3372                 ret_val |= e1e_rphy(hw, PHY_AUTONEG_EXP, &phy->expansion);
3373                 ret_val |= e1e_rphy(hw, PHY_1000T_CTRL, &phy->ctrl1000);
3374                 ret_val |= e1e_rphy(hw, PHY_1000T_STATUS, &phy->stat1000);
3375                 ret_val |= e1e_rphy(hw, PHY_EXT_STATUS, &phy->estatus);
3376                 if (ret_val)
3377                         e_warn("Error reading PHY register\n");
3378         } else {
3379                 /*
3380                  * Do not read PHY registers if link is not up
3381                  * Set values to typical power-on defaults
3382                  */
3383                 phy->bmcr = (BMCR_SPEED1000 | BMCR_ANENABLE | BMCR_FULLDPLX);
3384                 phy->bmsr = (BMSR_100FULL | BMSR_100HALF | BMSR_10FULL |
3385                              BMSR_10HALF | BMSR_ESTATEN | BMSR_ANEGCAPABLE |
3386                              BMSR_ERCAP);
3387                 phy->advertise = (ADVERTISE_PAUSE_ASYM | ADVERTISE_PAUSE_CAP |
3388                                   ADVERTISE_ALL | ADVERTISE_CSMA);
3389                 phy->lpa = 0;
3390                 phy->expansion = EXPANSION_ENABLENPAGE;
3391                 phy->ctrl1000 = ADVERTISE_1000FULL;
3392                 phy->stat1000 = 0;
3393                 phy->estatus = (ESTATUS_1000_TFULL | ESTATUS_1000_THALF);
3394         }
3395
3396         spin_unlock_irqrestore(&adapter->stats_lock, irq_flags);
3397 }
3398
3399 static void e1000_print_link_info(struct e1000_adapter *adapter)
3400 {
3401         struct e1000_hw *hw = &adapter->hw;
3402         u32 ctrl = er32(CTRL);
3403
3404         e_info("Link is Up %d Mbps %s, Flow Control: %s\n",
3405                adapter->link_speed,
3406                (adapter->link_duplex == FULL_DUPLEX) ?
3407                                 "Full Duplex" : "Half Duplex",
3408                ((ctrl & E1000_CTRL_TFCE) && (ctrl & E1000_CTRL_RFCE)) ?
3409                                 "RX/TX" :
3410                ((ctrl & E1000_CTRL_RFCE) ? "RX" :
3411                ((ctrl & E1000_CTRL_TFCE) ? "TX" : "None" )));
3412 }
3413
3414 static bool e1000_has_link(struct e1000_adapter *adapter)
3415 {
3416         struct e1000_hw *hw = &adapter->hw;
3417         bool link_active = 0;
3418         s32 ret_val = 0;
3419
3420         /*
3421          * get_link_status is set on LSC (link status) interrupt or
3422          * Rx sequence error interrupt.  get_link_status will stay
3423          * false until the check_for_link establishes link
3424          * for copper adapters ONLY
3425          */
3426         switch (hw->phy.media_type) {
3427         case e1000_media_type_copper:
3428                 if (hw->mac.get_link_status) {
3429                         ret_val = hw->mac.ops.check_for_link(hw);
3430                         link_active = !hw->mac.get_link_status;
3431                 } else {
3432                         link_active = 1;
3433                 }
3434                 break;
3435         case e1000_media_type_fiber:
3436                 ret_val = hw->mac.ops.check_for_link(hw);
3437                 link_active = !!(er32(STATUS) & E1000_STATUS_LU);
3438                 break;
3439         case e1000_media_type_internal_serdes:
3440                 ret_val = hw->mac.ops.check_for_link(hw);
3441                 link_active = adapter->hw.mac.serdes_has_link;
3442                 break;
3443         default:
3444         case e1000_media_type_unknown:
3445                 break;
3446         }
3447
3448         if ((ret_val == E1000_ERR_PHY) && (hw->phy.type == e1000_phy_igp_3) &&
3449             (er32(CTRL) & E1000_PHY_CTRL_GBE_DISABLE)) {
3450                 /* See e1000_kmrn_lock_loss_workaround_ich8lan() */
3451                 e_info("Gigabit has been disabled, downgrading speed\n");
3452         }
3453
3454         return link_active;
3455 }
3456
3457 static void e1000e_enable_receives(struct e1000_adapter *adapter)
3458 {
3459         /* make sure the receive unit is started */
3460         if ((adapter->flags & FLAG_RX_NEEDS_RESTART) &&
3461             (adapter->flags & FLAG_RX_RESTART_NOW)) {
3462                 struct e1000_hw *hw = &adapter->hw;
3463                 u32 rctl = er32(RCTL);
3464                 ew32(RCTL, rctl | E1000_RCTL_EN);
3465                 adapter->flags &= ~FLAG_RX_RESTART_NOW;
3466         }
3467 }
3468
3469 /**
3470  * e1000_watchdog - Timer Call-back
3471  * @data: pointer to adapter cast into an unsigned long
3472  **/
3473 static void e1000_watchdog(unsigned long data)
3474 {
3475         struct e1000_adapter *adapter = (struct e1000_adapter *) data;
3476
3477         /* Do the rest outside of interrupt context */
3478         schedule_work(&adapter->watchdog_task);
3479
3480         /* TODO: make this use queue_delayed_work() */
3481 }
3482
3483 static void e1000_watchdog_task(struct work_struct *work)
3484 {
3485         struct e1000_adapter *adapter = container_of(work,
3486                                         struct e1000_adapter, watchdog_task);
3487         struct net_device *netdev = adapter->netdev;
3488         struct e1000_mac_info *mac = &adapter->hw.mac;
3489         struct e1000_ring *tx_ring = adapter->tx_ring;
3490         struct e1000_hw *hw = &adapter->hw;
3491         u32 link, tctl;
3492         int tx_pending = 0;
3493
3494         link = e1000_has_link(adapter);
3495         if ((netif_carrier_ok(netdev)) && link) {
3496                 e1000e_enable_receives(adapter);
3497                 goto link_up;
3498         }
3499
3500         if ((e1000e_enable_tx_pkt_filtering(hw)) &&
3501             (adapter->mng_vlan_id != adapter->hw.mng_cookie.vlan_id))
3502                 e1000_update_mng_vlan(adapter);
3503
3504         if (link) {
3505                 if (!netif_carrier_ok(netdev)) {
3506                         bool txb2b = 1;
3507                         /* update snapshot of PHY registers on LSC */
3508                         e1000_phy_read_status(adapter);
3509                         mac->ops.get_link_up_info(&adapter->hw,
3510                                                    &adapter->link_speed,
3511                                                    &adapter->link_duplex);
3512                         e1000_print_link_info(adapter);
3513                         /*
3514                          * On supported PHYs, check for duplex mismatch only
3515                          * if link has autonegotiated at 10/100 half
3516                          */
3517                         if ((hw->phy.type == e1000_phy_igp_3 ||
3518                              hw->phy.type == e1000_phy_bm) &&
3519                             (hw->mac.autoneg == true) &&
3520                             (adapter->link_speed == SPEED_10 ||
3521                              adapter->link_speed == SPEED_100) &&
3522                             (adapter->link_duplex == HALF_DUPLEX)) {
3523                                 u16 autoneg_exp;
3524
3525                                 e1e_rphy(hw, PHY_AUTONEG_EXP, &autoneg_exp);
3526
3527                                 if (!(autoneg_exp & NWAY_ER_LP_NWAY_CAPS))
3528                                         e_info("Autonegotiated half duplex but"
3529                                                " link partner cannot autoneg. "
3530                                                " Try forcing full duplex if "
3531                                                "link gets many collisions.\n");
3532                         }
3533
3534                         /*
3535                          * tweak tx_queue_len according to speed/duplex
3536                          * and adjust the timeout factor
3537                          */
3538                         netdev->tx_queue_len = adapter->tx_queue_len;
3539                         adapter->tx_timeout_factor = 1;
3540                         switch (adapter->link_speed) {
3541                         case SPEED_10:
3542                                 txb2b = 0;
3543                                 netdev->tx_queue_len = 10;
3544                                 adapter->tx_timeout_factor = 16;
3545                                 break;
3546                         case SPEED_100:
3547                                 txb2b = 0;
3548                                 netdev->tx_queue_len = 100;
3549                                 /* maybe add some timeout factor ? */
3550                                 break;
3551                         }
3552
3553                         /*
3554                          * workaround: re-program speed mode bit after
3555                          * link-up event
3556                          */
3557                         if ((adapter->flags & FLAG_TARC_SPEED_MODE_BIT) &&
3558                             !txb2b) {
3559                                 u32 tarc0;
3560                                 tarc0 = er32(TARC(0));
3561                                 tarc0 &= ~SPEED_MODE_BIT;
3562                                 ew32(TARC(0), tarc0);
3563                         }
3564
3565                         /*
3566                          * disable TSO for pcie and 10/100 speeds, to avoid
3567                          * some hardware issues
3568                          */
3569                         if (!(adapter->flags & FLAG_TSO_FORCE)) {
3570                                 switch (adapter->link_speed) {
3571                                 case SPEED_10:
3572                                 case SPEED_100:
3573                                         e_info("10/100 speed: disabling TSO\n");
3574                                         netdev->features &= ~NETIF_F_TSO;
3575                                         netdev->features &= ~NETIF_F_TSO6;
3576                                         break;
3577                                 case SPEED_1000:
3578                                         netdev->features |= NETIF_F_TSO;
3579                                         netdev->features |= NETIF_F_TSO6;
3580                                         break;
3581                                 default:
3582                                         /* oops */
3583                                         break;
3584                                 }
3585                         }
3586
3587                         /*
3588                          * enable transmits in the hardware, need to do this
3589                          * after setting TARC(0)
3590                          */
3591                         tctl = er32(TCTL);
3592                         tctl |= E1000_TCTL_EN;
3593                         ew32(TCTL, tctl);
3594
3595                         netif_carrier_on(netdev);
3596                         netif_tx_wake_all_queues(netdev);
3597
3598                         if (!test_bit(__E1000_DOWN, &adapter->state))
3599                                 mod_timer(&adapter->phy_info_timer,
3600                                           round_jiffies(jiffies + 2 * HZ));
3601                 }
3602         } else {
3603                 if (netif_carrier_ok(netdev)) {
3604                         adapter->link_speed = 0;
3605                         adapter->link_duplex = 0;
3606                         e_info("Link is Down\n");
3607                         netif_carrier_off(netdev);
3608                         netif_tx_stop_all_queues(netdev);
3609                         if (!test_bit(__E1000_DOWN, &adapter->state))
3610                                 mod_timer(&adapter->phy_info_timer,
3611                                           round_jiffies(jiffies + 2 * HZ));
3612
3613                         if (adapter->flags & FLAG_RX_NEEDS_RESTART)
3614                                 schedule_work(&adapter->reset_task);
3615                 }
3616         }
3617
3618 link_up:
3619         e1000e_update_stats(adapter);
3620
3621         mac->tx_packet_delta = adapter->stats.tpt - adapter->tpt_old;
3622         adapter->tpt_old = adapter->stats.tpt;
3623         mac->collision_delta = adapter->stats.colc - adapter->colc_old;
3624         adapter->colc_old = adapter->stats.colc;
3625
3626         adapter->gorc = adapter->stats.gorc - adapter->gorc_old;
3627         adapter->gorc_old = adapter->stats.gorc;
3628         adapter->gotc = adapter->stats.gotc - adapter->gotc_old;
3629         adapter->gotc_old = adapter->stats.gotc;
3630
3631         e1000e_update_adaptive(&adapter->hw);
3632
3633         if (!netif_carrier_ok(netdev)) {
3634                 tx_pending = (e1000_desc_unused(tx_ring) + 1 <
3635                                tx_ring->count);
3636                 if (tx_pending) {
3637                         /*
3638                          * We've lost link, so the controller stops DMA,
3639                          * but we've got queued Tx work that's never going
3640                          * to get done, so reset controller to flush Tx.
3641                          * (Do the reset outside of interrupt context).
3642                          */
3643                         adapter->tx_timeout_count++;
3644                         schedule_work(&adapter->reset_task);
3645                 }
3646         }
3647
3648         /* Cause software interrupt to ensure Rx ring is cleaned */
3649         if (adapter->msix_entries)
3650                 ew32(ICS, adapter->rx_ring->ims_val);
3651         else
3652                 ew32(ICS, E1000_ICS_RXDMT0);
3653
3654         /* Force detection of hung controller every watchdog period */
3655         adapter->detect_tx_hung = 1;
3656
3657         /*
3658          * With 82571 controllers, LAA may be overwritten due to controller
3659          * reset from the other port. Set the appropriate LAA in RAR[0]
3660          */
3661         if (e1000e_get_laa_state_82571(hw))
3662                 e1000e_rar_set(hw, adapter->hw.mac.addr, 0);
3663
3664         /* Reset the timer */
3665         if (!test_bit(__E1000_DOWN, &adapter->state))
3666                 mod_timer(&adapter->watchdog_timer,
3667                           round_jiffies(jiffies + 2 * HZ));
3668 }
3669
3670 #define E1000_TX_FLAGS_CSUM             0x00000001
3671 #define E1000_TX_FLAGS_VLAN             0x00000002
3672 #define E1000_TX_FLAGS_TSO              0x00000004
3673 #define E1000_TX_FLAGS_IPV4             0x00000008
3674 #define E1000_TX_FLAGS_VLAN_MASK        0xffff0000
3675 #define E1000_TX_FLAGS_VLAN_SHIFT       16
3676
3677 static int e1000_tso(struct e1000_adapter *adapter,
3678                      struct sk_buff *skb)
3679 {
3680         struct e1000_ring *tx_ring = adapter->tx_ring;
3681         struct e1000_context_desc *context_desc;
3682         struct e1000_buffer *buffer_info;
3683         unsigned int i;
3684         u32 cmd_length = 0;
3685         u16 ipcse = 0, tucse, mss;
3686         u8 ipcss, ipcso, tucss, tucso, hdr_len;
3687         int err;
3688
3689         if (skb_is_gso(skb)) {
3690                 if (skb_header_cloned(skb)) {
3691                         err = pskb_expand_head(skb, 0, 0, GFP_ATOMIC);
3692                         if (err)
3693                                 return err;
3694                 }
3695
3696                 hdr_len = skb_transport_offset(skb) + tcp_hdrlen(skb);
3697                 mss = skb_shinfo(skb)->gso_size;
3698                 if (skb->protocol == htons(ETH_P_IP)) {
3699                         struct iphdr *iph = ip_hdr(skb);
3700                         iph->tot_len = 0;
3701                         iph->check = 0;
3702                         tcp_hdr(skb)->check = ~csum_tcpudp_magic(iph->saddr,
3703                                                                  iph->daddr, 0,
3704                                                                  IPPROTO_TCP,
3705                                                                  0);
3706                         cmd_length = E1000_TXD_CMD_IP;
3707                         ipcse = skb_transport_offset(skb) - 1;
3708                 } else if (skb_shinfo(skb)->gso_type == SKB_GSO_TCPV6) {
3709                         ipv6_hdr(skb)->payload_len = 0;
3710                         tcp_hdr(skb)->check =
3711                                 ~csum_ipv6_magic(&ipv6_hdr(skb)->saddr,
3712                                                  &ipv6_hdr(skb)->daddr,
3713                                                  0, IPPROTO_TCP, 0);
3714                         ipcse = 0;
3715                 }
3716                 ipcss = skb_network_offset(skb);
3717                 ipcso = (void *)&(ip_hdr(skb)->check) - (void *)skb->data;
3718                 tucss = skb_transport_offset(skb);
3719                 tucso = (void *)&(tcp_hdr(skb)->check) - (void *)skb->data;
3720                 tucse = 0;
3721
3722                 cmd_length |= (E1000_TXD_CMD_DEXT | E1000_TXD_CMD_TSE |
3723                                E1000_TXD_CMD_TCP | (skb->len - (hdr_len)));
3724
3725                 i = tx_ring->next_to_use;
3726                 context_desc = E1000_CONTEXT_DESC(*tx_ring, i);
3727                 buffer_info = &tx_ring->buffer_info[i];
3728
3729                 context_desc->lower_setup.ip_fields.ipcss  = ipcss;
3730                 context_desc->lower_setup.ip_fields.ipcso  = ipcso;
3731                 context_desc->lower_setup.ip_fields.ipcse  = cpu_to_le16(ipcse);
3732                 context_desc->upper_setup.tcp_fields.tucss = tucss;
3733                 context_desc->upper_setup.tcp_fields.tucso = tucso;
3734                 context_desc->upper_setup.tcp_fields.tucse = cpu_to_le16(tucse);
3735                 context_desc->tcp_seg_setup.fields.mss     = cpu_to_le16(mss);
3736                 context_desc->tcp_seg_setup.fields.hdr_len = hdr_len;
3737                 context_desc->cmd_and_length = cpu_to_le32(cmd_length);
3738
3739                 buffer_info->time_stamp = jiffies;
3740                 buffer_info->next_to_watch = i;
3741
3742                 i++;
3743                 if (i == tx_ring->count)
3744                         i = 0;
3745                 tx_ring->next_to_use = i;
3746
3747                 return 1;
3748         }
3749
3750         return 0;
3751 }
3752
3753 static bool e1000_tx_csum(struct e1000_adapter *adapter, struct sk_buff *skb)
3754 {
3755         struct e1000_ring *tx_ring = adapter->tx_ring;
3756         struct e1000_context_desc *context_desc;
3757         struct e1000_buffer *buffer_info;
3758         unsigned int i;
3759         u8 css;
3760
3761         if (skb->ip_summed == CHECKSUM_PARTIAL) {
3762                 css = skb_transport_offset(skb);
3763
3764                 i = tx_ring->next_to_use;
3765                 buffer_info = &tx_ring->buffer_info[i];
3766                 context_desc = E1000_CONTEXT_DESC(*tx_ring, i);
3767
3768                 context_desc->lower_setup.ip_config = 0;
3769                 context_desc->upper_setup.tcp_fields.tucss = css;
3770                 context_desc->upper_setup.tcp_fields.tucso =
3771                                         css + skb->csum_offset;
3772                 context_desc->upper_setup.tcp_fields.tucse = 0;
3773                 context_desc->tcp_seg_setup.data = 0;
3774                 context_desc->cmd_and_length = cpu_to_le32(E1000_TXD_CMD_DEXT);
3775
3776                 buffer_info->time_stamp = jiffies;
3777                 buffer_info->next_to_watch = i;
3778
3779                 i++;
3780                 if (i == tx_ring->count)
3781                         i = 0;
3782                 tx_ring->next_to_use = i;
3783
3784                 return 1;
3785         }
3786
3787         return 0;
3788 }
3789
3790 #define E1000_MAX_PER_TXD       8192
3791 #define E1000_MAX_TXD_PWR       12
3792
3793 static int e1000_tx_map(struct e1000_adapter *adapter,
3794                         struct sk_buff *skb, unsigned int first,
3795                         unsigned int max_per_txd, unsigned int nr_frags,
3796                         unsigned int mss)
3797 {
3798         struct e1000_ring *tx_ring = adapter->tx_ring;
3799         struct e1000_buffer *buffer_info;
3800         unsigned int len = skb->len - skb->data_len;
3801         unsigned int offset = 0, size, count = 0, i;
3802         unsigned int f;
3803
3804         i = tx_ring->next_to_use;
3805
3806         while (len) {
3807                 buffer_info = &tx_ring->buffer_info[i];
3808                 size = min(len, max_per_txd);
3809
3810                 /* Workaround for premature desc write-backs
3811                  * in TSO mode.  Append 4-byte sentinel desc */
3812                 if (mss && !nr_frags && size == len && size > 8)
3813                         size -= 4;
3814
3815                 buffer_info->length = size;
3816                 /* set time_stamp *before* dma to help avoid a possible race */
3817                 buffer_info->time_stamp = jiffies;
3818                 buffer_info->dma =
3819                         pci_map_single(adapter->pdev,
3820                                 skb->data + offset,
3821                                 size,
3822                                 PCI_DMA_TODEVICE);
3823                 if (pci_dma_mapping_error(adapter->pdev, buffer_info->dma)) {
3824                         dev_err(&adapter->pdev->dev, "TX DMA map failed\n");
3825                         adapter->tx_dma_failed++;
3826                         return -1;
3827                 }
3828                 buffer_info->next_to_watch = i;
3829
3830                 len -= size;
3831                 offset += size;
3832                 count++;
3833                 i++;
3834                 if (i == tx_ring->count)
3835                         i = 0;
3836         }
3837
3838         for (f = 0; f < nr_frags; f++) {
3839                 struct skb_frag_struct *frag;
3840
3841                 frag = &skb_shinfo(skb)->frags[f];
3842                 len = frag->size;
3843                 offset = frag->page_offset;
3844
3845                 while (len) {
3846                         buffer_info = &tx_ring->buffer_info[i];
3847                         size = min(len, max_per_txd);
3848                         /* Workaround for premature desc write-backs
3849                          * in TSO mode.  Append 4-byte sentinel desc */
3850                         if (mss && f == (nr_frags-1) && size == len && size > 8)
3851                                 size -= 4;
3852
3853                         buffer_info->length = size;
3854                         buffer_info->time_stamp = jiffies;
3855                         buffer_info->dma =
3856                                 pci_map_page(adapter->pdev,
3857                                         frag->page,
3858                                         offset,
3859                                         size,
3860                                         PCI_DMA_TODEVICE);
3861                         if (pci_dma_mapping_error(adapter->pdev,
3862                                                   buffer_info->dma)) {
3863                                 dev_err(&adapter->pdev->dev,
3864                                         "TX DMA page map failed\n");
3865                                 adapter->tx_dma_failed++;
3866                                 return -1;
3867                         }
3868
3869                         buffer_info->next_to_watch = i;
3870
3871                         len -= size;
3872                         offset += size;
3873                         count++;
3874
3875                         i++;
3876                         if (i == tx_ring->count)
3877                                 i = 0;
3878                 }
3879         }
3880
3881         if (i == 0)
3882                 i = tx_ring->count - 1;
3883         else
3884                 i--;
3885
3886         tx_ring->buffer_info[i].skb = skb;
3887         tx_ring->buffer_info[first].next_to_watch = i;
3888
3889         return count;
3890 }
3891
3892 static void e1000_tx_queue(struct e1000_adapter *adapter,
3893                            int tx_flags, int count)
3894 {
3895         struct e1000_ring *tx_ring = adapter->tx_ring;
3896         struct e1000_tx_desc *tx_desc = NULL;
3897         struct e1000_buffer *buffer_info;
3898         u32 txd_upper = 0, txd_lower = E1000_TXD_CMD_IFCS;
3899         unsigned int i;
3900
3901         if (tx_flags & E1000_TX_FLAGS_TSO) {
3902                 txd_lower |= E1000_TXD_CMD_DEXT | E1000_TXD_DTYP_D |
3903                              E1000_TXD_CMD_TSE;
3904                 txd_upper |= E1000_TXD_POPTS_TXSM << 8;
3905
3906                 if (tx_flags & E1000_TX_FLAGS_IPV4)
3907                         txd_upper |= E1000_TXD_POPTS_IXSM << 8;
3908         }
3909
3910         if (tx_flags & E1000_TX_FLAGS_CSUM) {
3911                 txd_lower |= E1000_TXD_CMD_DEXT | E1000_TXD_DTYP_D;
3912                 txd_upper |= E1000_TXD_POPTS_TXSM << 8;
3913         }
3914
3915         if (tx_flags & E1000_TX_FLAGS_VLAN) {
3916                 txd_lower |= E1000_TXD_CMD_VLE;
3917                 txd_upper |= (tx_flags & E1000_TX_FLAGS_VLAN_MASK);
3918         }
3919
3920         i = tx_ring->next_to_use;
3921
3922         while (count--) {
3923                 buffer_info = &tx_ring->buffer_info[i];
3924                 tx_desc = E1000_TX_DESC(*tx_ring, i);
3925                 tx_desc->buffer_addr = cpu_to_le64(buffer_info->dma);
3926                 tx_desc->lower.data =
3927                         cpu_to_le32(txd_lower | buffer_info->length);
3928                 tx_desc->upper.data = cpu_to_le32(txd_upper);
3929
3930                 i++;
3931                 if (i == tx_ring->count)
3932                         i = 0;
3933         }
3934
3935         tx_desc->lower.data |= cpu_to_le32(adapter->txd_cmd);
3936
3937         /*
3938          * Force memory writes to complete before letting h/w
3939          * know there are new descriptors to fetch.  (Only
3940          * applicable for weak-ordered memory model archs,
3941          * such as IA-64).
3942          */
3943         wmb();
3944
3945         tx_ring->next_to_use = i;
3946         writel(i, adapter->hw.hw_addr + tx_ring->tail);
3947         /*
3948          * we need this if more than one processor can write to our tail
3949          * at a time, it synchronizes IO on IA64/Altix systems
3950          */
3951         mmiowb();
3952 }
3953
3954 #define MINIMUM_DHCP_PACKET_SIZE 282
3955 static int e1000_transfer_dhcp_info(struct e1000_adapter *adapter,
3956                                     struct sk_buff *skb)
3957 {
3958         struct e1000_hw *hw =  &adapter->hw;
3959         u16 length, offset;
3960
3961         if (vlan_tx_tag_present(skb)) {
3962                 if (!((vlan_tx_tag_get(skb) == adapter->hw.mng_cookie.vlan_id)
3963                     && (adapter->hw.mng_cookie.status &
3964                         E1000_MNG_DHCP_COOKIE_STATUS_VLAN)))
3965                         return 0;
3966         }
3967
3968         if (skb->len <= MINIMUM_DHCP_PACKET_SIZE)
3969                 return 0;
3970
3971         if (((struct ethhdr *) skb->data)->h_proto != htons(ETH_P_IP))
3972                 return 0;
3973
3974         {
3975                 const struct iphdr *ip = (struct iphdr *)((u8 *)skb->data+14);
3976                 struct udphdr *udp;
3977
3978                 if (ip->protocol != IPPROTO_UDP)
3979                         return 0;
3980
3981                 udp = (struct udphdr *)((u8 *)ip + (ip->ihl << 2));
3982                 if (ntohs(udp->dest) != 67)
3983                         return 0;
3984
3985                 offset = (u8 *)udp + 8 - skb->data;
3986                 length = skb->len - offset;
3987                 return e1000e_mng_write_dhcp_info(hw, (u8 *)udp + 8, length);
3988         }
3989
3990         return 0;
3991 }
3992
3993 static int __e1000_maybe_stop_tx(struct net_device *netdev, int size)
3994 {
3995         struct e1000_adapter *adapter = netdev_priv(netdev);
3996
3997         netif_stop_queue(netdev);
3998         /*
3999          * Herbert's original patch had:
4000          *  smp_mb__after_netif_stop_queue();
4001          * but since that doesn't exist yet, just open code it.
4002          */
4003         smp_mb();
4004
4005         /*
4006          * We need to check again in a case another CPU has just
4007          * made room available.
4008          */
4009         if (e1000_desc_unused(adapter->tx_ring) < size)
4010                 return -EBUSY;
4011
4012         /* A reprieve! */
4013         netif_start_queue(netdev);
4014         ++adapter->restart_queue;
4015         return 0;
4016 }
4017
4018 static int e1000_maybe_stop_tx(struct net_device *netdev, int size)
4019 {
4020         struct e1000_adapter *adapter = netdev_priv(netdev);
4021
4022         if (e1000_desc_unused(adapter->tx_ring) >= size)
4023                 return 0;
4024         return __e1000_maybe_stop_tx(netdev, size);
4025 }
4026
4027 #define TXD_USE_COUNT(S, X) (((S) >> (X)) + 1 )
4028 static int e1000_xmit_frame(struct sk_buff *skb, struct net_device *netdev)
4029 {
4030         struct e1000_adapter *adapter = netdev_priv(netdev);
4031         struct e1000_ring *tx_ring = adapter->tx_ring;
4032         unsigned int first;
4033         unsigned int max_per_txd = E1000_MAX_PER_TXD;
4034         unsigned int max_txd_pwr = E1000_MAX_TXD_PWR;
4035         unsigned int tx_flags = 0;
4036         unsigned int len = skb->len - skb->data_len;
4037         unsigned long irq_flags;
4038         unsigned int nr_frags;
4039         unsigned int mss;
4040         int count = 0;
4041         int tso;
4042         unsigned int f;
4043
4044         if (test_bit(__E1000_DOWN, &adapter->state)) {
4045                 dev_kfree_skb_any(skb);
4046                 return NETDEV_TX_OK;
4047         }
4048
4049         if (skb->len <= 0) {
4050                 dev_kfree_skb_any(skb);
4051                 return NETDEV_TX_OK;
4052         }
4053
4054         mss = skb_shinfo(skb)->gso_size;
4055         /*
4056          * The controller does a simple calculation to
4057          * make sure there is enough room in the FIFO before
4058          * initiating the DMA for each buffer.  The calc is:
4059          * 4 = ceil(buffer len/mss).  To make sure we don't
4060          * overrun the FIFO, adjust the max buffer len if mss
4061          * drops.
4062          */
4063         if (mss) {
4064                 u8 hdr_len;
4065                 max_per_txd = min(mss << 2, max_per_txd);
4066                 max_txd_pwr = fls(max_per_txd) - 1;
4067
4068                 /*
4069                  * TSO Workaround for 82571/2/3 Controllers -- if skb->data
4070                  * points to just header, pull a few bytes of payload from
4071                  * frags into skb->data
4072                  */
4073                 hdr_len = skb_transport_offset(skb) + tcp_hdrlen(skb);
4074                 /*
4075                  * we do this workaround for ES2LAN, but it is un-necessary,
4076                  * avoiding it could save a lot of cycles
4077                  */
4078                 if (skb->data_len && (hdr_len == len)) {
4079                         unsigned int pull_size;
4080
4081                         pull_size = min((unsigned int)4, skb->data_len);
4082                         if (!__pskb_pull_tail(skb, pull_size)) {
4083                                 e_err("__pskb_pull_tail failed.\n");
4084                                 dev_kfree_skb_any(skb);
4085                                 return NETDEV_TX_OK;
4086                         }
4087                         len = skb->len - skb->data_len;
4088                 }
4089         }
4090
4091         /* reserve a descriptor for the offload context */
4092         if ((mss) || (skb->ip_summed == CHECKSUM_PARTIAL))
4093                 count++;
4094         count++;
4095
4096         count += TXD_USE_COUNT(len, max_txd_pwr);
4097
4098         nr_frags = skb_shinfo(skb)->nr_frags;
4099         for (f = 0; f < nr_frags; f++)
4100                 count += TXD_USE_COUNT(skb_shinfo(skb)->frags[f].size,
4101                                        max_txd_pwr);
4102
4103         if (adapter->hw.mac.tx_pkt_filtering)
4104                 e1000_transfer_dhcp_info(adapter, skb);
4105
4106         if (!spin_trylock_irqsave(&adapter->tx_queue_lock, irq_flags))
4107                 /* Collision - tell upper layer to requeue */
4108                 return NETDEV_TX_LOCKED;
4109
4110         /*
4111          * need: count + 2 desc gap to keep tail from touching
4112          * head, otherwise try next time
4113          */
4114         if (e1000_maybe_stop_tx(netdev, count + 2)) {
4115                 spin_unlock_irqrestore(&adapter->tx_queue_lock, irq_flags);
4116                 return NETDEV_TX_BUSY;
4117         }
4118
4119         if (adapter->vlgrp && vlan_tx_tag_present(skb)) {
4120                 tx_flags |= E1000_TX_FLAGS_VLAN;
4121                 tx_flags |= (vlan_tx_tag_get(skb) << E1000_TX_FLAGS_VLAN_SHIFT);
4122         }
4123
4124         first = tx_ring->next_to_use;
4125
4126         tso = e1000_tso(adapter, skb);
4127         if (tso < 0) {
4128                 dev_kfree_skb_any(skb);
4129                 spin_unlock_irqrestore(&adapter->tx_queue_lock, irq_flags);
4130                 return NETDEV_TX_OK;
4131         }
4132
4133         if (tso)
4134                 tx_flags |= E1000_TX_FLAGS_TSO;
4135         else if (e1000_tx_csum(adapter, skb))
4136                 tx_flags |= E1000_TX_FLAGS_CSUM;
4137
4138         /*
4139          * Old method was to assume IPv4 packet by default if TSO was enabled.
4140          * 82571 hardware supports TSO capabilities for IPv6 as well...
4141          * no longer assume, we must.
4142          */
4143         if (skb->protocol == htons(ETH_P_IP))
4144                 tx_flags |= E1000_TX_FLAGS_IPV4;
4145
4146         count = e1000_tx_map(adapter, skb, first, max_per_txd, nr_frags, mss);
4147         if (count < 0) {
4148                 /* handle pci_map_single() error in e1000_tx_map */
4149                 dev_kfree_skb_any(skb);
4150                 spin_unlock_irqrestore(&adapter->tx_queue_lock, irq_flags);
4151                 return NETDEV_TX_OK;
4152         }
4153
4154         e1000_tx_queue(adapter, tx_flags, count);
4155
4156         netdev->trans_start = jiffies;
4157
4158         /* Make sure there is space in the ring for the next send. */
4159         e1000_maybe_stop_tx(netdev, MAX_SKB_FRAGS + 2);
4160
4161         spin_unlock_irqrestore(&adapter->tx_queue_lock, irq_flags);
4162         return NETDEV_TX_OK;
4163 }
4164
4165 /**
4166  * e1000_tx_timeout - Respond to a Tx Hang
4167  * @netdev: network interface device structure
4168  **/
4169 static void e1000_tx_timeout(struct net_device *netdev)
4170 {
4171         struct e1000_adapter *adapter = netdev_priv(netdev);
4172
4173         /* Do the reset outside of interrupt context */
4174         adapter->tx_timeout_count++;
4175         schedule_work(&adapter->reset_task);
4176 }
4177
4178 static void e1000_reset_task(struct work_struct *work)
4179 {
4180         struct e1000_adapter *adapter;
4181         adapter = container_of(work, struct e1000_adapter, reset_task);
4182
4183         e1000e_reinit_locked(adapter);
4184 }
4185
4186 /**
4187  * e1000_get_stats - Get System Network Statistics
4188  * @netdev: network interface device structure
4189  *
4190  * Returns the address of the device statistics structure.
4191  * The statistics are actually updated from the timer callback.
4192  **/
4193 static struct net_device_stats *e1000_get_stats(struct net_device *netdev)
4194 {
4195         struct e1000_adapter *adapter = netdev_priv(netdev);
4196
4197         /* only return the current stats */
4198         return &adapter->net_stats;
4199 }
4200
4201 /**
4202  * e1000_change_mtu - Change the Maximum Transfer Unit
4203  * @netdev: network interface device structure
4204  * @new_mtu: new value for maximum frame size
4205  *
4206  * Returns 0 on success, negative on failure
4207  **/
4208 static int e1000_change_mtu(struct net_device *netdev, int new_mtu)
4209 {
4210         struct e1000_adapter *adapter = netdev_priv(netdev);
4211         int max_frame = new_mtu + ETH_HLEN + ETH_FCS_LEN;
4212
4213         if ((new_mtu < ETH_ZLEN + ETH_FCS_LEN + VLAN_HLEN) ||
4214             (max_frame > MAX_JUMBO_FRAME_SIZE)) {
4215                 e_err("Invalid MTU setting\n");
4216                 return -EINVAL;
4217         }
4218
4219         /* Jumbo frame size limits */
4220         if (max_frame > ETH_FRAME_LEN + ETH_FCS_LEN) {
4221                 if (!(adapter->flags & FLAG_HAS_JUMBO_FRAMES)) {
4222                         e_err("Jumbo Frames not supported.\n");
4223                         return -EINVAL;
4224                 }
4225                 if (adapter->hw.phy.type == e1000_phy_ife) {
4226                         e_err("Jumbo Frames not supported.\n");
4227                         return -EINVAL;
4228                 }
4229         }
4230
4231 #define MAX_STD_JUMBO_FRAME_SIZE 9234
4232         if (max_frame > MAX_STD_JUMBO_FRAME_SIZE) {
4233                 e_err("MTU > 9216 not supported.\n");
4234                 return -EINVAL;
4235         }
4236
4237         while (test_and_set_bit(__E1000_RESETTING, &adapter->state))
4238                 msleep(1);
4239         /* e1000e_down has a dependency on max_frame_size */
4240         adapter->max_frame_size = max_frame;
4241         if (netif_running(netdev))
4242                 e1000e_down(adapter);
4243
4244         /*
4245          * NOTE: netdev_alloc_skb reserves 16 bytes, and typically NET_IP_ALIGN
4246          * means we reserve 2 more, this pushes us to allocate from the next
4247          * larger slab size.
4248          * i.e. RXBUFFER_2048 --> size-4096 slab
4249          * However with the new *_jumbo_rx* routines, jumbo receives will use
4250          * fragmented skbs
4251          */
4252
4253         if (max_frame <= 256)
4254                 adapter->rx_buffer_len = 256;
4255         else if (max_frame <= 512)
4256                 adapter->rx_buffer_len = 512;
4257         else if (max_frame <= 1024)
4258                 adapter->rx_buffer_len = 1024;
4259         else if (max_frame <= 2048)
4260                 adapter->rx_buffer_len = 2048;
4261         else
4262                 adapter->rx_buffer_len = 4096;
4263
4264         /* adjust allocation if LPE protects us, and we aren't using SBP */
4265         if ((max_frame == ETH_FRAME_LEN + ETH_FCS_LEN) ||
4266              (max_frame == ETH_FRAME_LEN + VLAN_HLEN + ETH_FCS_LEN))
4267                 adapter->rx_buffer_len = ETH_FRAME_LEN + VLAN_HLEN
4268                                          + ETH_FCS_LEN;
4269
4270         e_info("changing MTU from %d to %d\n", netdev->mtu, new_mtu);
4271         netdev->mtu = new_mtu;
4272
4273         if (netif_running(netdev))
4274                 e1000e_up(adapter);
4275         else
4276                 e1000e_reset(adapter);
4277
4278         clear_bit(__E1000_RESETTING, &adapter->state);
4279
4280         return 0;
4281 }
4282
4283 static int e1000_mii_ioctl(struct net_device *netdev, struct ifreq *ifr,
4284                            int cmd)
4285 {
4286         struct e1000_adapter *adapter = netdev_priv(netdev);
4287         struct mii_ioctl_data *data = if_mii(ifr);
4288
4289         if (adapter->hw.phy.media_type != e1000_media_type_copper)
4290                 return -EOPNOTSUPP;
4291
4292         switch (cmd) {
4293         case SIOCGMIIPHY:
4294                 data->phy_id = adapter->hw.phy.addr;
4295                 break;
4296         case SIOCGMIIREG:
4297                 if (!capable(CAP_NET_ADMIN))
4298                         return -EPERM;
4299                 switch (data->reg_num & 0x1F) {
4300                 case MII_BMCR:
4301                         data->val_out = adapter->phy_regs.bmcr;
4302                         break;
4303                 case MII_BMSR:
4304                         data->val_out = adapter->phy_regs.bmsr;
4305                         break;
4306                 case MII_PHYSID1:
4307                         data->val_out = (adapter->hw.phy.id >> 16);
4308                         break;
4309                 case MII_PHYSID2:
4310                         data->val_out = (adapter->hw.phy.id & 0xFFFF);
4311                         break;
4312                 case MII_ADVERTISE:
4313                         data->val_out = adapter->phy_regs.advertise;
4314                         break;
4315                 case MII_LPA:
4316                         data->val_out = adapter->phy_regs.lpa;
4317                         break;
4318                 case MII_EXPANSION:
4319                         data->val_out = adapter->phy_regs.expansion;
4320                         break;
4321                 case MII_CTRL1000:
4322                         data->val_out = adapter->phy_regs.ctrl1000;
4323                         break;
4324                 case MII_STAT1000:
4325                         data->val_out = adapter->phy_regs.stat1000;
4326                         break;
4327                 case MII_ESTATUS:
4328                         data->val_out = adapter->phy_regs.estatus;
4329                         break;
4330                 default:
4331                         return -EIO;
4332                 }
4333                 break;
4334         case SIOCSMIIREG:
4335         default:
4336                 return -EOPNOTSUPP;
4337         }
4338         return 0;
4339 }
4340
4341 static int e1000_ioctl(struct net_device *netdev, struct ifreq *ifr, int cmd)
4342 {
4343         switch (cmd) {
4344         case SIOCGMIIPHY:
4345         case SIOCGMIIREG:
4346         case SIOCSMIIREG:
4347                 return e1000_mii_ioctl(netdev, ifr, cmd);
4348         default:
4349                 return -EOPNOTSUPP;
4350         }
4351 }
4352
4353 static int e1000_suspend(struct pci_dev *pdev, pm_message_t state)
4354 {
4355         struct net_device *netdev = pci_get_drvdata(pdev);
4356         struct e1000_adapter *adapter = netdev_priv(netdev);
4357         struct e1000_hw *hw = &adapter->hw;
4358         u32 ctrl, ctrl_ext, rctl, status;
4359         u32 wufc = adapter->wol;
4360         int retval = 0;
4361
4362         netif_device_detach(netdev);
4363
4364         if (netif_running(netdev)) {
4365                 WARN_ON(test_bit(__E1000_RESETTING, &adapter->state));
4366                 e1000e_down(adapter);
4367                 e1000_free_irq(adapter);
4368         }
4369         e1000e_reset_interrupt_capability(adapter);
4370
4371         retval = pci_save_state(pdev);
4372         if (retval)
4373                 return retval;
4374
4375         status = er32(STATUS);
4376         if (status & E1000_STATUS_LU)
4377                 wufc &= ~E1000_WUFC_LNKC;
4378
4379         if (wufc) {
4380                 e1000_setup_rctl(adapter);
4381                 e1000_set_multi(netdev);
4382
4383                 /* turn on all-multi mode if wake on multicast is enabled */
4384                 if (wufc & E1000_WUFC_MC) {
4385                         rctl = er32(RCTL);
4386                         rctl |= E1000_RCTL_MPE;
4387                         ew32(RCTL, rctl);
4388                 }
4389
4390                 ctrl = er32(CTRL);
4391                 /* advertise wake from D3Cold */
4392                 #define E1000_CTRL_ADVD3WUC 0x00100000
4393                 /* phy power management enable */
4394                 #define E1000_CTRL_EN_PHY_PWR_MGMT 0x00200000
4395                 ctrl |= E1000_CTRL_ADVD3WUC |
4396                         E1000_CTRL_EN_PHY_PWR_MGMT;
4397                 ew32(CTRL, ctrl);
4398
4399                 if (adapter->hw.phy.media_type == e1000_media_type_fiber ||
4400                     adapter->hw.phy.media_type ==
4401                     e1000_media_type_internal_serdes) {
4402                         /* keep the laser running in D3 */
4403                         ctrl_ext = er32(CTRL_EXT);
4404                         ctrl_ext |= E1000_CTRL_EXT_SDP7_DATA;
4405                         ew32(CTRL_EXT, ctrl_ext);
4406                 }
4407
4408                 if (adapter->flags & FLAG_IS_ICH)
4409                         e1000e_disable_gig_wol_ich8lan(&adapter->hw);
4410
4411                 /* Allow time for pending master requests to run */
4412                 e1000e_disable_pcie_master(&adapter->hw);
4413
4414                 ew32(WUC, E1000_WUC_PME_EN);
4415                 ew32(WUFC, wufc);
4416                 pci_enable_wake(pdev, PCI_D3hot, 1);
4417                 pci_enable_wake(pdev, PCI_D3cold, 1);
4418         } else {
4419                 ew32(WUC, 0);
4420                 ew32(WUFC, 0);
4421                 pci_enable_wake(pdev, PCI_D3hot, 0);
4422                 pci_enable_wake(pdev, PCI_D3cold, 0);
4423         }
4424
4425         /* make sure adapter isn't asleep if manageability is enabled */
4426         if (adapter->flags & FLAG_MNG_PT_ENABLED) {
4427                 pci_enable_wake(pdev, PCI_D3hot, 1);
4428                 pci_enable_wake(pdev, PCI_D3cold, 1);
4429         }
4430
4431         if (adapter->hw.phy.type == e1000_phy_igp_3)
4432                 e1000e_igp3_phy_powerdown_workaround_ich8lan(&adapter->hw);
4433
4434         /*
4435          * Release control of h/w to f/w.  If f/w is AMT enabled, this
4436          * would have already happened in close and is redundant.
4437          */
4438         e1000_release_hw_control(adapter);
4439
4440         pci_disable_device(pdev);
4441
4442         pci_set_power_state(pdev, pci_choose_state(pdev, state));
4443
4444         return 0;
4445 }
4446
4447 static void e1000e_disable_l1aspm(struct pci_dev *pdev)
4448 {
4449         int pos;
4450         u16 val;
4451
4452         /*
4453          * 82573 workaround - disable L1 ASPM on mobile chipsets
4454          *
4455          * L1 ASPM on various mobile (ich7) chipsets do not behave properly
4456          * resulting in lost data or garbage information on the pci-e link
4457          * level. This could result in (false) bad EEPROM checksum errors,
4458          * long ping times (up to 2s) or even a system freeze/hang.
4459          *
4460          * Unfortunately this feature saves about 1W power consumption when
4461          * active.
4462          */
4463         pos = pci_find_capability(pdev, PCI_CAP_ID_EXP);
4464         pci_read_config_word(pdev, pos + PCI_EXP_LNKCTL, &val);
4465         if (val & 0x2) {
4466                 dev_warn(&pdev->dev, "Disabling L1 ASPM\n");
4467                 val &= ~0x2;
4468                 pci_write_config_word(pdev, pos + PCI_EXP_LNKCTL, val);
4469         }
4470 }
4471
4472 #ifdef CONFIG_PM
4473 static int e1000_resume(struct pci_dev *pdev)
4474 {
4475         struct net_device *netdev = pci_get_drvdata(pdev);
4476         struct e1000_adapter *adapter = netdev_priv(netdev);
4477         struct e1000_hw *hw = &adapter->hw;
4478         u32 err;
4479
4480         pci_set_power_state(pdev, PCI_D0);
4481         pci_restore_state(pdev);
4482         e1000e_disable_l1aspm(pdev);
4483
4484         err = pci_enable_device_mem(pdev);
4485         if (err) {
4486                 dev_err(&pdev->dev,
4487                         "Cannot enable PCI device from suspend\n");
4488                 return err;
4489         }
4490
4491         pci_set_master(pdev);
4492
4493         pci_enable_wake(pdev, PCI_D3hot, 0);
4494         pci_enable_wake(pdev, PCI_D3cold, 0);
4495
4496         e1000e_set_interrupt_capability(adapter);
4497         if (netif_running(netdev)) {
4498                 err = e1000_request_irq(adapter);
4499                 if (err)
4500                         return err;
4501         }
4502
4503         e1000e_power_up_phy(adapter);
4504         e1000e_reset(adapter);
4505         ew32(WUS, ~0);
4506
4507         e1000_init_manageability(adapter);
4508
4509         if (netif_running(netdev))
4510                 e1000e_up(adapter);
4511
4512         netif_device_attach(netdev);
4513
4514         /*
4515          * If the controller has AMT, do not set DRV_LOAD until the interface
4516          * is up.  For all other cases, let the f/w know that the h/w is now
4517          * under the control of the driver.
4518          */
4519         if (!(adapter->flags & FLAG_HAS_AMT))
4520                 e1000_get_hw_control(adapter);
4521
4522         return 0;
4523 }
4524 #endif
4525
4526 static void e1000_shutdown(struct pci_dev *pdev)
4527 {
4528         e1000_suspend(pdev, PMSG_SUSPEND);
4529 }
4530
4531 #ifdef CONFIG_NET_POLL_CONTROLLER
4532 /*
4533  * Polling 'interrupt' - used by things like netconsole to send skbs
4534  * without having to re-enable interrupts. It's not called while
4535  * the interrupt routine is executing.
4536  */
4537 static void e1000_netpoll(struct net_device *netdev)
4538 {
4539         struct e1000_adapter *adapter = netdev_priv(netdev);
4540
4541         disable_irq(adapter->pdev->irq);
4542         e1000_intr(adapter->pdev->irq, netdev);
4543
4544         enable_irq(adapter->pdev->irq);
4545 }
4546 #endif
4547
4548 /**
4549  * e1000_io_error_detected - called when PCI error is detected
4550  * @pdev: Pointer to PCI device
4551  * @state: The current pci connection state
4552  *
4553  * This function is called after a PCI bus error affecting
4554  * this device has been detected.
4555  */
4556 static pci_ers_result_t e1000_io_error_detected(struct pci_dev *pdev,
4557                                                 pci_channel_state_t state)
4558 {
4559         struct net_device *netdev = pci_get_drvdata(pdev);
4560         struct e1000_adapter *adapter = netdev_priv(netdev);
4561
4562         netif_device_detach(netdev);
4563
4564         if (netif_running(netdev))
4565                 e1000e_down(adapter);
4566         pci_disable_device(pdev);
4567
4568         /* Request a slot slot reset. */
4569         return PCI_ERS_RESULT_NEED_RESET;
4570 }
4571
4572 /**
4573  * e1000_io_slot_reset - called after the pci bus has been reset.
4574  * @pdev: Pointer to PCI device
4575  *
4576  * Restart the card from scratch, as if from a cold-boot. Implementation
4577  * resembles the first-half of the e1000_resume routine.
4578  */
4579 static pci_ers_result_t e1000_io_slot_reset(struct pci_dev *pdev)
4580 {
4581         struct net_device *netdev = pci_get_drvdata(pdev);
4582         struct e1000_adapter *adapter = netdev_priv(netdev);
4583         struct e1000_hw *hw = &adapter->hw;
4584         int err;
4585
4586         e1000e_disable_l1aspm(pdev);
4587         err = pci_enable_device_mem(pdev);
4588         if (err) {
4589                 dev_err(&pdev->dev,
4590                         "Cannot re-enable PCI device after reset.\n");
4591                 return PCI_ERS_RESULT_DISCONNECT;
4592         }
4593         pci_set_master(pdev);
4594         pci_restore_state(pdev);
4595
4596         pci_enable_wake(pdev, PCI_D3hot, 0);
4597         pci_enable_wake(pdev, PCI_D3cold, 0);
4598
4599         e1000e_reset(adapter);
4600         ew32(WUS, ~0);
4601
4602         return PCI_ERS_RESULT_RECOVERED;
4603 }
4604
4605 /**
4606  * e1000_io_resume - called when traffic can start flowing again.
4607  * @pdev: Pointer to PCI device
4608  *
4609  * This callback is called when the error recovery driver tells us that
4610  * its OK to resume normal operation. Implementation resembles the
4611  * second-half of the e1000_resume routine.
4612  */
4613 static void e1000_io_resume(struct pci_dev *pdev)
4614 {
4615         struct net_device *netdev = pci_get_drvdata(pdev);
4616         struct e1000_adapter *adapter = netdev_priv(netdev);
4617
4618         e1000_init_manageability(adapter);
4619
4620         if (netif_running(netdev)) {
4621                 if (e1000e_up(adapter)) {
4622                         dev_err(&pdev->dev,
4623                                 "can't bring device back up after reset\n");
4624                         return;
4625                 }
4626         }
4627
4628         netif_device_attach(netdev);
4629
4630         /*
4631          * If the controller has AMT, do not set DRV_LOAD until the interface
4632          * is up.  For all other cases, let the f/w know that the h/w is now
4633          * under the control of the driver.
4634          */
4635         if (!(adapter->flags & FLAG_HAS_AMT))
4636                 e1000_get_hw_control(adapter);
4637
4638 }
4639
4640 static void e1000_print_device_info(struct e1000_adapter *adapter)
4641 {
4642         struct e1000_hw *hw = &adapter->hw;
4643         struct net_device *netdev = adapter->netdev;
4644         u32 pba_num;
4645
4646         /* print bus type/speed/width info */
4647         e_info("(PCI Express:2.5GB/s:%s) %02x:%02x:%02x:%02x:%02x:%02x\n",
4648                /* bus width */
4649                ((hw->bus.width == e1000_bus_width_pcie_x4) ? "Width x4" :
4650                 "Width x1"),
4651                /* MAC address */
4652                netdev->dev_addr[0], netdev->dev_addr[1],
4653                netdev->dev_addr[2], netdev->dev_addr[3],
4654                netdev->dev_addr[4], netdev->dev_addr[5]);
4655         e_info("Intel(R) PRO/%s Network Connection\n",
4656                (hw->phy.type == e1000_phy_ife) ? "10/100" : "1000");
4657         e1000e_read_pba_num(hw, &pba_num);
4658         e_info("MAC: %d, PHY: %d, PBA No: %06x-%03x\n",
4659                hw->mac.type, hw->phy.type, (pba_num >> 8), (pba_num & 0xff));
4660 }
4661
4662 static void e1000_eeprom_checks(struct e1000_adapter *adapter)
4663 {
4664         struct e1000_hw *hw = &adapter->hw;
4665         int ret_val;
4666         u16 buf = 0;
4667
4668         if (hw->mac.type != e1000_82573)
4669                 return;
4670
4671         ret_val = e1000_read_nvm(hw, NVM_INIT_CONTROL2_REG, 1, &buf);
4672         if (!(le16_to_cpu(buf) & (1 << 0))) {
4673                 /* Deep Smart Power Down (DSPD) */
4674                 e_warn("Warning: detected DSPD enabled in EEPROM\n");
4675         }
4676
4677         ret_val = e1000_read_nvm(hw, NVM_INIT_3GIO_3, 1, &buf);
4678         if (le16_to_cpu(buf) & (3 << 2)) {
4679                 /* ASPM enable */
4680                 e_warn("Warning: detected ASPM enabled in EEPROM\n");
4681         }
4682 }
4683
4684 /**
4685  * e1000_probe - Device Initialization Routine
4686  * @pdev: PCI device information struct
4687  * @ent: entry in e1000_pci_tbl
4688  *
4689  * Returns 0 on success, negative on failure
4690  *
4691  * e1000_probe initializes an adapter identified by a pci_dev structure.
4692  * The OS initialization, configuring of the adapter private structure,
4693  * and a hardware reset occur.
4694  **/
4695 static int __devinit e1000_probe(struct pci_dev *pdev,
4696                                  const struct pci_device_id *ent)
4697 {
4698         struct net_device *netdev;
4699         struct e1000_adapter *adapter;
4700         struct e1000_hw *hw;
4701         const struct e1000_info *ei = e1000_info_tbl[ent->driver_data];
4702         resource_size_t mmio_start, mmio_len;
4703         resource_size_t flash_start, flash_len;
4704
4705         static int cards_found;
4706         int i, err, pci_using_dac;
4707         u16 eeprom_data = 0;
4708         u16 eeprom_apme_mask = E1000_EEPROM_APME;
4709
4710         e1000e_disable_l1aspm(pdev);
4711
4712         err = pci_enable_device_mem(pdev);
4713         if (err)
4714                 return err;
4715
4716         pci_using_dac = 0;
4717         err = pci_set_dma_mask(pdev, DMA_64BIT_MASK);
4718         if (!err) {
4719                 err = pci_set_consistent_dma_mask(pdev, DMA_64BIT_MASK);
4720                 if (!err)
4721                         pci_using_dac = 1;
4722         } else {
4723                 err = pci_set_dma_mask(pdev, DMA_32BIT_MASK);
4724                 if (err) {
4725                         err = pci_set_consistent_dma_mask(pdev,
4726                                                           DMA_32BIT_MASK);
4727                         if (err) {
4728                                 dev_err(&pdev->dev, "No usable DMA "
4729                                         "configuration, aborting\n");
4730                                 goto err_dma;
4731                         }
4732                 }
4733         }
4734
4735         err = pci_request_selected_regions(pdev,
4736                                           pci_select_bars(pdev, IORESOURCE_MEM),
4737                                           e1000e_driver_name);
4738         if (err)
4739                 goto err_pci_reg;
4740
4741         pci_set_master(pdev);
4742         pci_save_state(pdev);
4743
4744         err = -ENOMEM;
4745         netdev = alloc_etherdev(sizeof(struct e1000_adapter));
4746         if (!netdev)
4747                 goto err_alloc_etherdev;
4748
4749         SET_NETDEV_DEV(netdev, &pdev->dev);
4750
4751         pci_set_drvdata(pdev, netdev);
4752         adapter = netdev_priv(netdev);
4753         hw = &adapter->hw;
4754         adapter->netdev = netdev;
4755         adapter->pdev = pdev;
4756         adapter->ei = ei;
4757         adapter->pba = ei->pba;
4758         adapter->flags = ei->flags;
4759         adapter->hw.adapter = adapter;
4760         adapter->hw.mac.type = ei->mac;
4761         adapter->msg_enable = (1 << NETIF_MSG_DRV | NETIF_MSG_PROBE) - 1;
4762
4763         mmio_start = pci_resource_start(pdev, 0);
4764         mmio_len = pci_resource_len(pdev, 0);
4765
4766         err = -EIO;
4767         adapter->hw.hw_addr = ioremap(mmio_start, mmio_len);
4768         if (!adapter->hw.hw_addr)
4769                 goto err_ioremap;
4770
4771         if ((adapter->flags & FLAG_HAS_FLASH) &&
4772             (pci_resource_flags(pdev, 1) & IORESOURCE_MEM)) {
4773                 flash_start = pci_resource_start(pdev, 1);
4774                 flash_len = pci_resource_len(pdev, 1);
4775                 adapter->hw.flash_address = ioremap(flash_start, flash_len);
4776                 if (!adapter->hw.flash_address)
4777                         goto err_flashmap;
4778         }
4779
4780         /* construct the net_device struct */
4781         netdev->open                    = &e1000_open;
4782         netdev->stop                    = &e1000_close;
4783         netdev->hard_start_xmit         = &e1000_xmit_frame;
4784         netdev->get_stats               = &e1000_get_stats;
4785         netdev->set_multicast_list      = &e1000_set_multi;
4786         netdev->set_mac_address         = &e1000_set_mac;
4787         netdev->change_mtu              = &e1000_change_mtu;
4788         netdev->do_ioctl                = &e1000_ioctl;
4789         e1000e_set_ethtool_ops(netdev);
4790         netdev->tx_timeout              = &e1000_tx_timeout;
4791         netdev->watchdog_timeo          = 5 * HZ;
4792         netif_napi_add(netdev, &adapter->napi, e1000_clean, 64);
4793         netdev->vlan_rx_register        = e1000_vlan_rx_register;
4794         netdev->vlan_rx_add_vid         = e1000_vlan_rx_add_vid;
4795         netdev->vlan_rx_kill_vid        = e1000_vlan_rx_kill_vid;
4796 #ifdef CONFIG_NET_POLL_CONTROLLER
4797         netdev->poll_controller         = e1000_netpoll;
4798 #endif
4799         strncpy(netdev->name, pci_name(pdev), sizeof(netdev->name) - 1);
4800
4801         netdev->mem_start = mmio_start;
4802         netdev->mem_end = mmio_start + mmio_len;
4803
4804         adapter->bd_number = cards_found++;
4805
4806         e1000e_check_options(adapter);
4807
4808         /* setup adapter struct */
4809         err = e1000_sw_init(adapter);
4810         if (err)
4811                 goto err_sw_init;
4812
4813         err = -EIO;
4814
4815         memcpy(&hw->mac.ops, ei->mac_ops, sizeof(hw->mac.ops));
4816         memcpy(&hw->nvm.ops, ei->nvm_ops, sizeof(hw->nvm.ops));
4817         memcpy(&hw->phy.ops, ei->phy_ops, sizeof(hw->phy.ops));
4818
4819         err = ei->get_variants(adapter);
4820         if (err)
4821                 goto err_hw_init;
4822
4823         hw->mac.ops.get_bus_info(&adapter->hw);
4824
4825         adapter->hw.phy.autoneg_wait_to_complete = 0;
4826
4827         /* Copper options */
4828         if (adapter->hw.phy.media_type == e1000_media_type_copper) {
4829                 adapter->hw.phy.mdix = AUTO_ALL_MODES;
4830                 adapter->hw.phy.disable_polarity_correction = 0;
4831                 adapter->hw.phy.ms_type = e1000_ms_hw_default;
4832         }
4833
4834         if (e1000_check_reset_block(&adapter->hw))
4835                 e_info("PHY reset is blocked due to SOL/IDER session.\n");
4836
4837         netdev->features = NETIF_F_SG |
4838                            NETIF_F_HW_CSUM |
4839                            NETIF_F_HW_VLAN_TX |
4840                            NETIF_F_HW_VLAN_RX;
4841
4842         if (adapter->flags & FLAG_HAS_HW_VLAN_FILTER)
4843                 netdev->features |= NETIF_F_HW_VLAN_FILTER;
4844
4845         netdev->features |= NETIF_F_TSO;
4846         netdev->features |= NETIF_F_TSO6;
4847
4848         netdev->vlan_features |= NETIF_F_TSO;
4849         netdev->vlan_features |= NETIF_F_TSO6;
4850         netdev->vlan_features |= NETIF_F_HW_CSUM;
4851         netdev->vlan_features |= NETIF_F_SG;
4852
4853         if (pci_using_dac)
4854                 netdev->features |= NETIF_F_HIGHDMA;
4855
4856         /*
4857          * We should not be using LLTX anymore, but we are still Tx faster with
4858          * it.
4859          */
4860         netdev->features |= NETIF_F_LLTX;
4861
4862         if (e1000e_enable_mng_pass_thru(&adapter->hw))
4863                 adapter->flags |= FLAG_MNG_PT_ENABLED;
4864
4865         /*
4866          * before reading the NVM, reset the controller to
4867          * put the device in a known good starting state
4868          */
4869         adapter->hw.mac.ops.reset_hw(&adapter->hw);
4870
4871         /*
4872          * systems with ASPM and others may see the checksum fail on the first
4873          * attempt. Let's give it a few tries
4874          */
4875         for (i = 0;; i++) {
4876                 if (e1000_validate_nvm_checksum(&adapter->hw) >= 0)
4877                         break;
4878                 if (i == 2) {
4879                         e_err("The NVM Checksum Is Not Valid\n");
4880                         err = -EIO;
4881                         goto err_eeprom;
4882                 }
4883         }
4884
4885         e1000_eeprom_checks(adapter);
4886
4887         /* copy the MAC address out of the NVM */
4888         if (e1000e_read_mac_addr(&adapter->hw))
4889                 e_err("NVM Read Error while reading MAC address\n");
4890
4891         memcpy(netdev->dev_addr, adapter->hw.mac.addr, netdev->addr_len);
4892         memcpy(netdev->perm_addr, adapter->hw.mac.addr, netdev->addr_len);
4893
4894         if (!is_valid_ether_addr(netdev->perm_addr)) {
4895                 e_err("Invalid MAC Address: %02x:%02x:%02x:%02x:%02x:%02x\n",
4896                       netdev->perm_addr[0], netdev->perm_addr[1],
4897                       netdev->perm_addr[2], netdev->perm_addr[3],
4898                       netdev->perm_addr[4], netdev->perm_addr[5]);
4899                 err = -EIO;
4900                 goto err_eeprom;
4901         }
4902
4903         init_timer(&adapter->watchdog_timer);
4904         adapter->watchdog_timer.function = &e1000_watchdog;
4905         adapter->watchdog_timer.data = (unsigned long) adapter;
4906
4907         init_timer(&adapter->phy_info_timer);
4908         adapter->phy_info_timer.function = &e1000_update_phy_info;
4909         adapter->phy_info_timer.data = (unsigned long) adapter;
4910
4911         INIT_WORK(&adapter->reset_task, e1000_reset_task);
4912         INIT_WORK(&adapter->watchdog_task, e1000_watchdog_task);
4913
4914         /* Initialize link parameters. User can change them with ethtool */
4915         adapter->hw.mac.autoneg = 1;
4916         adapter->fc_autoneg = 1;
4917         adapter->hw.fc.original_type = e1000_fc_default;
4918         adapter->hw.fc.type = e1000_fc_default;
4919         adapter->hw.phy.autoneg_advertised = 0x2f;
4920
4921         /* ring size defaults */
4922         adapter->rx_ring->count = 256;
4923         adapter->tx_ring->count = 256;
4924
4925         /*
4926          * Initial Wake on LAN setting - If APM wake is enabled in
4927          * the EEPROM, enable the ACPI Magic Packet filter
4928          */
4929         if (adapter->flags & FLAG_APME_IN_WUC) {
4930                 /* APME bit in EEPROM is mapped to WUC.APME */
4931                 eeprom_data = er32(WUC);
4932                 eeprom_apme_mask = E1000_WUC_APME;
4933         } else if (adapter->flags & FLAG_APME_IN_CTRL3) {
4934                 if (adapter->flags & FLAG_APME_CHECK_PORT_B &&
4935                     (adapter->hw.bus.func == 1))
4936                         e1000_read_nvm(&adapter->hw,
4937                                 NVM_INIT_CONTROL3_PORT_B, 1, &eeprom_data);
4938                 else
4939                         e1000_read_nvm(&adapter->hw,
4940                                 NVM_INIT_CONTROL3_PORT_A, 1, &eeprom_data);
4941         }
4942
4943         /* fetch WoL from EEPROM */
4944         if (eeprom_data & eeprom_apme_mask)
4945                 adapter->eeprom_wol |= E1000_WUFC_MAG;
4946
4947         /*
4948          * now that we have the eeprom settings, apply the special cases
4949          * where the eeprom may be wrong or the board simply won't support
4950          * wake on lan on a particular port
4951          */
4952         if (!(adapter->flags & FLAG_HAS_WOL))
4953                 adapter->eeprom_wol = 0;
4954
4955         /* initialize the wol settings based on the eeprom settings */
4956         adapter->wol = adapter->eeprom_wol;
4957
4958         /* reset the hardware with the new settings */
4959         e1000e_reset(adapter);
4960
4961         /*
4962          * If the controller has AMT, do not set DRV_LOAD until the interface
4963          * is up.  For all other cases, let the f/w know that the h/w is now
4964          * under the control of the driver.
4965          */
4966         if (!(adapter->flags & FLAG_HAS_AMT))
4967                 e1000_get_hw_control(adapter);
4968
4969         /* tell the stack to leave us alone until e1000_open() is called */
4970         netif_carrier_off(netdev);
4971         netif_tx_stop_all_queues(netdev);
4972
4973         strcpy(netdev->name, "eth%d");
4974         err = register_netdev(netdev);
4975         if (err)
4976                 goto err_register;
4977
4978         e1000_print_device_info(adapter);
4979
4980         return 0;
4981
4982 err_register:
4983         if (!(adapter->flags & FLAG_HAS_AMT))
4984                 e1000_release_hw_control(adapter);
4985 err_eeprom:
4986         if (!e1000_check_reset_block(&adapter->hw))
4987                 e1000_phy_hw_reset(&adapter->hw);
4988 err_hw_init:
4989
4990         kfree(adapter->tx_ring);
4991         kfree(adapter->rx_ring);
4992 err_sw_init:
4993         if (adapter->hw.flash_address)
4994                 iounmap(adapter->hw.flash_address);
4995 err_flashmap:
4996         iounmap(adapter->hw.hw_addr);
4997 err_ioremap:
4998         free_netdev(netdev);
4999 err_alloc_etherdev:
5000         pci_release_selected_regions(pdev,
5001                                      pci_select_bars(pdev, IORESOURCE_MEM));
5002 err_pci_reg:
5003 err_dma:
5004         pci_disable_device(pdev);
5005         return err;
5006 }
5007
5008 /**
5009  * e1000_remove - Device Removal Routine
5010  * @pdev: PCI device information struct
5011  *
5012  * e1000_remove is called by the PCI subsystem to alert the driver
5013  * that it should release a PCI device.  The could be caused by a
5014  * Hot-Plug event, or because the driver is going to be removed from
5015  * memory.
5016  **/
5017 static void __devexit e1000_remove(struct pci_dev *pdev)
5018 {
5019         struct net_device *netdev = pci_get_drvdata(pdev);
5020         struct e1000_adapter *adapter = netdev_priv(netdev);
5021
5022         /*
5023          * flush_scheduled work may reschedule our watchdog task, so
5024          * explicitly disable watchdog tasks from being rescheduled
5025          */
5026         set_bit(__E1000_DOWN, &adapter->state);
5027         del_timer_sync(&adapter->watchdog_timer);
5028         del_timer_sync(&adapter->phy_info_timer);
5029
5030         flush_scheduled_work();
5031
5032         /*
5033          * Release control of h/w to f/w.  If f/w is AMT enabled, this
5034          * would have already happened in close and is redundant.
5035          */
5036         e1000_release_hw_control(adapter);
5037
5038         unregister_netdev(netdev);
5039
5040         if (!e1000_check_reset_block(&adapter->hw))
5041                 e1000_phy_hw_reset(&adapter->hw);
5042
5043         e1000e_reset_interrupt_capability(adapter);
5044         kfree(adapter->tx_ring);
5045         kfree(adapter->rx_ring);
5046
5047         iounmap(adapter->hw.hw_addr);
5048         if (adapter->hw.flash_address)
5049                 iounmap(adapter->hw.flash_address);
5050         pci_release_selected_regions(pdev,
5051                                      pci_select_bars(pdev, IORESOURCE_MEM));
5052
5053         free_netdev(netdev);
5054
5055         pci_disable_device(pdev);
5056 }
5057
5058 /* PCI Error Recovery (ERS) */
5059 static struct pci_error_handlers e1000_err_handler = {
5060         .error_detected = e1000_io_error_detected,
5061         .slot_reset = e1000_io_slot_reset,
5062         .resume = e1000_io_resume,
5063 };
5064
5065 static struct pci_device_id e1000_pci_tbl[] = {
5066         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82571EB_COPPER), board_82571 },
5067         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82571EB_FIBER), board_82571 },
5068         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82571EB_QUAD_COPPER), board_82571 },
5069         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82571EB_QUAD_COPPER_LP), board_82571 },
5070         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82571EB_QUAD_FIBER), board_82571 },
5071         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82571EB_SERDES), board_82571 },
5072         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82571EB_SERDES_DUAL), board_82571 },
5073         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82571EB_SERDES_QUAD), board_82571 },
5074         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82571PT_QUAD_COPPER), board_82571 },
5075
5076         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82572EI), board_82572 },
5077         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82572EI_COPPER), board_82572 },
5078         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82572EI_FIBER), board_82572 },
5079         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82572EI_SERDES), board_82572 },
5080
5081         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82573E), board_82573 },
5082         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82573E_IAMT), board_82573 },
5083         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82573L), board_82573 },
5084
5085         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82574L), board_82574 },
5086
5087         { PCI_VDEVICE(INTEL, E1000_DEV_ID_80003ES2LAN_COPPER_DPT),
5088           board_80003es2lan },
5089         { PCI_VDEVICE(INTEL, E1000_DEV_ID_80003ES2LAN_COPPER_SPT),
5090           board_80003es2lan },
5091         { PCI_VDEVICE(INTEL, E1000_DEV_ID_80003ES2LAN_SERDES_DPT),
5092           board_80003es2lan },
5093         { PCI_VDEVICE(INTEL, E1000_DEV_ID_80003ES2LAN_SERDES_SPT),
5094           board_80003es2lan },
5095
5096         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH8_IFE), board_ich8lan },
5097         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH8_IFE_G), board_ich8lan },
5098         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH8_IFE_GT), board_ich8lan },
5099         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH8_IGP_AMT), board_ich8lan },
5100         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH8_IGP_C), board_ich8lan },
5101         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH8_IGP_M), board_ich8lan },
5102         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH8_IGP_M_AMT), board_ich8lan },
5103
5104         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH9_IFE), board_ich9lan },
5105         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH9_IFE_G), board_ich9lan },
5106         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH9_IFE_GT), board_ich9lan },
5107         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH9_IGP_AMT), board_ich9lan },
5108         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH9_IGP_C), board_ich9lan },
5109         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH9_BM), board_ich9lan },
5110         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH9_IGP_M), board_ich9lan },
5111         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH9_IGP_M_AMT), board_ich9lan },
5112         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH9_IGP_M_V), board_ich9lan },
5113
5114         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH10_R_BM_LM), board_ich9lan },
5115         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH10_R_BM_LF), board_ich9lan },
5116         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH10_R_BM_V), board_ich9lan },
5117
5118         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH10_D_BM_LM), board_ich10lan },
5119         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH10_D_BM_LF), board_ich10lan },
5120
5121         { }     /* terminate list */
5122 };
5123 MODULE_DEVICE_TABLE(pci, e1000_pci_tbl);
5124
5125 /* PCI Device API Driver */
5126 static struct pci_driver e1000_driver = {
5127         .name     = e1000e_driver_name,
5128         .id_table = e1000_pci_tbl,
5129         .probe    = e1000_probe,
5130         .remove   = __devexit_p(e1000_remove),
5131 #ifdef CONFIG_PM
5132         /* Power Management Hooks */
5133         .suspend  = e1000_suspend,
5134         .resume   = e1000_resume,
5135 #endif
5136         .shutdown = e1000_shutdown,
5137         .err_handler = &e1000_err_handler
5138 };
5139
5140 /**
5141  * e1000_init_module - Driver Registration Routine
5142  *
5143  * e1000_init_module is the first routine called when the driver is
5144  * loaded. All it does is register with the PCI subsystem.
5145  **/
5146 static int __init e1000_init_module(void)
5147 {
5148         int ret;
5149         printk(KERN_INFO "%s: Intel(R) PRO/1000 Network Driver - %s\n",
5150                e1000e_driver_name, e1000e_driver_version);
5151         printk(KERN_INFO "%s: Copyright (c) 1999-2008 Intel Corporation.\n",
5152                e1000e_driver_name);
5153         ret = pci_register_driver(&e1000_driver);
5154         pm_qos_add_requirement(PM_QOS_CPU_DMA_LATENCY, e1000e_driver_name,
5155                                PM_QOS_DEFAULT_VALUE);
5156
5157         return ret;
5158 }
5159 module_init(e1000_init_module);
5160
5161 /**
5162  * e1000_exit_module - Driver Exit Cleanup Routine
5163  *
5164  * e1000_exit_module is called just before the driver is removed
5165  * from memory.
5166  **/
5167 static void __exit e1000_exit_module(void)
5168 {
5169         pci_unregister_driver(&e1000_driver);
5170         pm_qos_remove_requirement(PM_QOS_CPU_DMA_LATENCY, e1000e_driver_name);
5171 }
5172 module_exit(e1000_exit_module);
5173
5174
5175 MODULE_AUTHOR("Intel Corporation, <linux.nics@intel.com>");
5176 MODULE_DESCRIPTION("Intel(R) PRO/1000 Network Driver");
5177 MODULE_LICENSE("GPL");
5178 MODULE_VERSION(DRV_VERSION);
5179
5180 /* e1000_main.c */