drivers/net/e1000e/netdev.c

   1 /*******************************************************************************
   2
   3   Intel PRO/1000 Linux driver
   4   Copyright(c) 1999 - 2007 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
  47 #include "e1000.h"
  48
  49 #define DRV_VERSION "0.2.0"
  50 char e1000e_driver_name[] = "e1000e";
  51 const char e1000e_driver_version[] = DRV_VERSION;
  52
  53 static const struct e1000_info *e1000_info_tbl[] = {
  54         [board_82571]           = &e1000_82571_info,
  55         [board_82572]           = &e1000_82572_info,
  56         [board_82573]           = &e1000_82573_info,
  57         [board_80003es2lan]     = &e1000_es2_info,
  58         [board_ich8lan]         = &e1000_ich8_info,
  59         [board_ich9lan]         = &e1000_ich9_info,
  60 };
  61
  62 #ifdef DEBUG
  63 /**
  64  * e1000_get_hw_dev_name - return device name string
  65  * used by hardware layer to print debugging information
  66  **/
  67 char *e1000e_get_hw_dev_name(struct e1000_hw *hw)
  68 {
  69         return hw->adapter->netdev->name;
  70 }
  71 #endif
  72
  73 /**
  74  * e1000_desc_unused - calculate if we have unused descriptors
  75  **/
  76 static int e1000_desc_unused(struct e1000_ring *ring)
  77 {
  78         if (ring->next_to_clean > ring->next_to_use)
  79                 return ring->next_to_clean - ring->next_to_use - 1;
  80
  81         return ring->count + ring->next_to_clean - ring->next_to_use - 1;
  82 }
  83
  84 /**
  85  * e1000_receive_skb - helper function to handle rx indications
  86  * @adapter: board private structure
  87  * @status: descriptor status field as written by hardware
  88  * @vlan: descriptor vlan field as written by hardware (no le/be conversion)
  89  * @skb: pointer to sk_buff to be indicated to stack
  90  **/
  91 static void e1000_receive_skb(struct e1000_adapter *adapter,
  92                               struct net_device *netdev,
  93                               struct sk_buff *skb,
  94                               u8 status, u16 vlan)
  95 {
  96         skb->protocol = eth_type_trans(skb, netdev);
  97
  98         if (adapter->vlgrp && (status & E1000_RXD_STAT_VP))
  99                 vlan_hwaccel_receive_skb(skb, adapter->vlgrp,
 100                                          le16_to_cpu(vlan) &
 101                                          E1000_RXD_SPC_VLAN_MASK);
 102         else
 103                 netif_receive_skb(skb);
 104
 105         netdev->last_rx = jiffies;
 106 }
 107
 108 /**
 109  * e1000_rx_checksum - Receive Checksum Offload for 82543
 110  * @adapter:     board private structure
 111  * @status_err:  receive descriptor status and error fields
 112  * @csum:       receive descriptor csum field
 113  * @sk_buff:     socket buffer with received data
 114  **/
 115 static void e1000_rx_checksum(struct e1000_adapter *adapter, u32 status_err,
 116                               u32 csum, struct sk_buff *skb)
 117 {
 118         u16 status = (u16)status_err;
 119         u8 errors = (u8)(status_err >> 24);
 120         skb->ip_summed = CHECKSUM_NONE;
 121
 122         /* Ignore Checksum bit is set */
 123         if (status & E1000_RXD_STAT_IXSM)
 124                 return;
 125         /* TCP/UDP checksum error bit is set */
 126         if (errors & E1000_RXD_ERR_TCPE) {
 127                 /* let the stack verify checksum errors */
 128                 adapter->hw_csum_err++;
 129                 return;
 130         }
 131
 132         /* TCP/UDP Checksum has not been calculated */
 133         if (!(status & (E1000_RXD_STAT_TCPCS | E1000_RXD_STAT_UDPCS)))
 134                 return;
 135
 136         /* It must be a TCP or UDP packet with a valid checksum */
 137         if (status & E1000_RXD_STAT_TCPCS) {
 138                 /* TCP checksum is good */
 139                 skb->ip_summed = CHECKSUM_UNNECESSARY;
 140         } else {
 141                 /* IP fragment with UDP payload */
 142                 /* Hardware complements the payload checksum, so we undo it
 143                  * and then put the value in host order for further stack use.
 144                  */
 145                 csum = ntohl(csum ^ 0xFFFF);
 146                 skb->csum = csum;
 147                 skb->ip_summed = CHECKSUM_COMPLETE;
 148         }
 149         adapter->hw_csum_good++;
 150 }
 151
 152 /**
 153  * e1000_alloc_rx_buffers - Replace used receive buffers; legacy & extended
 154  * @adapter: address of board private structure
 155  **/
 156 static void e1000_alloc_rx_buffers(struct e1000_adapter *adapter,
 157                                    int cleaned_count)
 158 {
 159         struct net_device *netdev = adapter->netdev;
 160         struct pci_dev *pdev = adapter->pdev;
 161         struct e1000_ring *rx_ring = adapter->rx_ring;
 162         struct e1000_rx_desc *rx_desc;
 163         struct e1000_buffer *buffer_info;
 164         struct sk_buff *skb;
 165         unsigned int i;
 166         unsigned int bufsz = adapter->rx_buffer_len + NET_IP_ALIGN;
 167
 168         i = rx_ring->next_to_use;
 169         buffer_info = &rx_ring->buffer_info[i];
 170
 171         while (cleaned_count--) {
 172                 skb = buffer_info->skb;
 173                 if (skb) {
 174                         skb_trim(skb, 0);
 175                         goto map_skb;
 176                 }
 177
 178                 skb = netdev_alloc_skb(netdev, bufsz);
 179                 if (!skb) {
 180                         /* Better luck next round */
 181                         adapter->alloc_rx_buff_failed++;
 182                         break;
 183                 }
 184
 185                 /* Make buffer alignment 2 beyond a 16 byte boundary
 186                  * this will result in a 16 byte aligned IP header after
 187                  * the 14 byte MAC header is removed
 188                  */
 189                 skb_reserve(skb, NET_IP_ALIGN);
 190
 191                 buffer_info->skb = skb;
 192 map_skb:
 193                 buffer_info->dma = pci_map_single(pdev, skb->data,
 194                                                   adapter->rx_buffer_len,
 195                                                   PCI_DMA_FROMDEVICE);
 196                 if (pci_dma_mapping_error(buffer_info->dma)) {
 197                         dev_err(&pdev->dev, "RX DMA map failed\n");
 198                         adapter->rx_dma_failed++;
 199                         break;
 200                 }
 201
 202                 rx_desc = E1000_RX_DESC(*rx_ring, i);
 203                 rx_desc->buffer_addr = cpu_to_le64(buffer_info->dma);
 204
 205                 i++;
 206                 if (i == rx_ring->count)
 207                         i = 0;
 208                 buffer_info = &rx_ring->buffer_info[i];
 209         }
 210
 211         if (rx_ring->next_to_use != i) {
 212                 rx_ring->next_to_use = i;
 213                 if (i-- == 0)
 214                         i = (rx_ring->count - 1);
 215
 216                 /* Force memory writes to complete before letting h/w
 217                  * know there are new descriptors to fetch.  (Only
 218                  * applicable for weak-ordered memory model archs,
 219                  * such as IA-64). */
 220                 wmb();
 221                 writel(i, adapter->hw.hw_addr + rx_ring->tail);
 222         }
 223 }
 224
 225 /**
 226  * e1000_alloc_rx_buffers_ps - Replace used receive buffers; packet split
 227  * @adapter: address of board private structure
 228  **/
 229 static void e1000_alloc_rx_buffers_ps(struct e1000_adapter *adapter,
 230                                       int cleaned_count)
 231 {
 232         struct net_device *netdev = adapter->netdev;
 233         struct pci_dev *pdev = adapter->pdev;
 234         union e1000_rx_desc_packet_split *rx_desc;
 235         struct e1000_ring *rx_ring = adapter->rx_ring;
 236         struct e1000_buffer *buffer_info;
 237         struct e1000_ps_page *ps_page;
 238         struct sk_buff *skb;
 239         unsigned int i, j;
 240
 241         i = rx_ring->next_to_use;
 242         buffer_info = &rx_ring->buffer_info[i];
 243
 244         while (cleaned_count--) {
 245                 rx_desc = E1000_RX_DESC_PS(*rx_ring, i);
 246
 247                 for (j = 0; j < PS_PAGE_BUFFERS; j++) {
 248                         ps_page = &buffer_info->ps_pages[j];
 249                         if (j >= adapter->rx_ps_pages) {
 250                                 /* all unused desc entries get hw null ptr */
 251                                 rx_desc->read.buffer_addr[j+1] = ~0;
 252                                 continue;
 253                         }
 254                         if (!ps_page->page) {
 255                                 ps_page->page = alloc_page(GFP_ATOMIC);
 256                                 if (!ps_page->page) {
 257                                         adapter->alloc_rx_buff_failed++;
 258                                         goto no_buffers;
 259                                 }
 260                                 ps_page->dma = pci_map_page(pdev,
 261                                                    ps_page->page,
 262                                                    0, PAGE_SIZE,
 263                                                    PCI_DMA_FROMDEVICE);
 264                                 if (pci_dma_mapping_error(ps_page->dma)) {
 265                                         dev_err(&adapter->pdev->dev,
 266                                           "RX DMA page map failed\n");
 267                                         adapter->rx_dma_failed++;
 268                                         goto no_buffers;
 269                                 }
 270                         }
 271                         /*
 272                          * Refresh the desc even if buffer_addrs
 273                          * didn't change because each write-back
 274                          * erases this info.
 275                          */
 276                         rx_desc->read.buffer_addr[j+1] =
 277                              cpu_to_le64(ps_page->dma);
 278                 }
 279
 280                 skb = netdev_alloc_skb(netdev,
 281                                        adapter->rx_ps_bsize0 + NET_IP_ALIGN);
 282
 283                 if (!skb) {
 284                         adapter->alloc_rx_buff_failed++;
 285                         break;
 286                 }
 287
 288                 /* Make buffer alignment 2 beyond a 16 byte boundary
 289                  * this will result in a 16 byte aligned IP header after
 290                  * the 14 byte MAC header is removed
 291                  */
 292                 skb_reserve(skb, NET_IP_ALIGN);
 293
 294                 buffer_info->skb = skb;
 295                 buffer_info->dma = pci_map_single(pdev, skb->data,
 296                                                   adapter->rx_ps_bsize0,
 297                                                   PCI_DMA_FROMDEVICE);
 298                 if (pci_dma_mapping_error(buffer_info->dma)) {
 299                         dev_err(&pdev->dev, "RX DMA map failed\n");
 300                         adapter->rx_dma_failed++;
 301                         /* cleanup skb */
 302                         dev_kfree_skb_any(skb);
 303                         buffer_info->skb = NULL;
 304                         break;
 305                 }
 306
 307                 rx_desc->read.buffer_addr[0] = cpu_to_le64(buffer_info->dma);
 308
 309                 i++;
 310                 if (i == rx_ring->count)
 311                         i = 0;
 312                 buffer_info = &rx_ring->buffer_info[i];
 313         }
 314
 315 no_buffers:
 316         if (rx_ring->next_to_use != i) {
 317                 rx_ring->next_to_use = i;
 318
 319                 if (!(i--))
 320                         i = (rx_ring->count - 1);
 321
 322                 /* Force memory writes to complete before letting h/w
 323                  * know there are new descriptors to fetch.  (Only
 324                  * applicable for weak-ordered memory model archs,
 325                  * such as IA-64). */
 326                 wmb();
 327                 /* Hardware increments by 16 bytes, but packet split
 328                  * descriptors are 32 bytes...so we increment tail
 329                  * twice as much.
 330                  */
 331                 writel(i<<1, adapter->hw.hw_addr + rx_ring->tail);
 332         }
 333 }
 334
 335 /**
 336  * e1000_alloc_rx_buffers_jumbo - Replace used jumbo receive buffers
 337  *
 338  * @adapter: address of board private structure
 339  * @cleaned_count: number of buffers to allocate this pass
 340  **/
 341 static void e1000_alloc_rx_buffers_jumbo(struct e1000_adapter *adapter,
 342                                          int cleaned_count)
 343 {
 344         struct net_device *netdev = adapter->netdev;
 345         struct pci_dev *pdev = adapter->pdev;
 346         struct e1000_ring *rx_ring = adapter->rx_ring;
 347         struct e1000_rx_desc *rx_desc;
 348         struct e1000_buffer *buffer_info;
 349         struct sk_buff *skb;
 350         unsigned int i;
 351         unsigned int bufsz = 256 -
 352                              16 /*for skb_reserve */ -
 353                              NET_IP_ALIGN;
 354
 355         i = rx_ring->next_to_use;
 356         buffer_info = &rx_ring->buffer_info[i];
 357
 358         while (cleaned_count--) {
 359                 skb = buffer_info->skb;
 360                 if (skb) {
 361                         skb_trim(skb, 0);
 362                         goto check_page;
 363                 }
 364
 365                 skb = netdev_alloc_skb(netdev, bufsz);
 366                 if (!skb) {
 367                         /* Better luck next round */
 368                         adapter->alloc_rx_buff_failed++;
 369                         break;
 370                 }
 371
 372                 /* Make buffer alignment 2 beyond a 16 byte boundary
 373                  * this will result in a 16 byte aligned IP header after
 374                  * the 14 byte MAC header is removed
 375                  */
 376                 skb_reserve(skb, NET_IP_ALIGN);
 377
 378                 buffer_info->skb = skb;
 379 check_page:
 380                 /* allocate a new page if necessary */
 381                 if (!buffer_info->page) {
 382                         buffer_info->page = alloc_page(GFP_ATOMIC);
 383                         if (!buffer_info->page) {
 384                                 adapter->alloc_rx_buff_failed++;
 385                                 break;
 386                         }
 387                 }
 388
 389                 if (!buffer_info->dma)
 390                         buffer_info->dma = pci_map_page(pdev,
 391                                                         buffer_info->page, 0,
 392                                                         PAGE_SIZE,
 393                                                         PCI_DMA_FROMDEVICE);
 394                 if (pci_dma_mapping_error(buffer_info->dma)) {
 395                         dev_err(&adapter->pdev->dev, "RX DMA page map failed\n");
 396                         adapter->rx_dma_failed++;
 397                         break;
 398                 }
 399
 400                 rx_desc = E1000_RX_DESC(*rx_ring, i);
 401                 rx_desc->buffer_addr = cpu_to_le64(buffer_info->dma);
 402
 403                 i++;
 404                 if (i == rx_ring->count)
 405                         i = 0;
 406                 buffer_info = &rx_ring->buffer_info[i];
 407         }
 408
 409         if (rx_ring->next_to_use != i) {
 410                 rx_ring->next_to_use = i;
 411                 if (i-- == 0)
 412                         i = (rx_ring->count - 1);
 413
 414                 /* Force memory writes to complete before letting h/w
 415                  * know there are new descriptors to fetch.  (Only
 416                  * applicable for weak-ordered memory model archs,
 417                  * such as IA-64). */
 418                 wmb();
 419                 writel(i, adapter->hw.hw_addr + rx_ring->tail);
 420         }
 421 }
 422
 423 /**
 424  * e1000_clean_rx_irq - Send received data up the network stack; legacy
 425  * @adapter: board private structure
 426  *
 427  * the return value indicates whether actual cleaning was done, there
 428  * is no guarantee that everything was cleaned
 429  **/
 430 static bool e1000_clean_rx_irq(struct e1000_adapter *adapter,
 431                                int *work_done, int work_to_do)
 432 {
 433         struct net_device *netdev = adapter->netdev;
 434         struct pci_dev *pdev = adapter->pdev;
 435         struct e1000_ring *rx_ring = adapter->rx_ring;
 436         struct e1000_rx_desc *rx_desc, *next_rxd;
 437         struct e1000_buffer *buffer_info, *next_buffer;
 438         u32 length;
 439         unsigned int i;
 440         int cleaned_count = 0;
 441         bool cleaned = 0;
 442         unsigned int total_rx_bytes = 0, total_rx_packets = 0;
 443
 444         i = rx_ring->next_to_clean;
 445         rx_desc = E1000_RX_DESC(*rx_ring, i);
 446         buffer_info = &rx_ring->buffer_info[i];
 447
 448         while (rx_desc->status & E1000_RXD_STAT_DD) {
 449                 struct sk_buff *skb;
 450                 u8 status;
 451
 452                 if (*work_done >= work_to_do)
 453                         break;
 454                 (*work_done)++;
 455
 456                 status = rx_desc->status;
 457                 skb = buffer_info->skb;
 458                 buffer_info->skb = NULL;
 459
 460                 prefetch(skb->data - NET_IP_ALIGN);
 461
 462                 i++;
 463                 if (i == rx_ring->count)
 464                         i = 0;
 465                 next_rxd = E1000_RX_DESC(*rx_ring, i);
 466                 prefetch(next_rxd);
 467
 468                 next_buffer = &rx_ring->buffer_info[i];
 469
 470                 cleaned = 1;
 471                 cleaned_count++;
 472                 pci_unmap_single(pdev,
 473                                  buffer_info->dma,
 474                                  adapter->rx_buffer_len,
 475                                  PCI_DMA_FROMDEVICE);
 476                 buffer_info->dma = 0;
 477
 478                 length = le16_to_cpu(rx_desc->length);
 479
 480                 /* !EOP means multiple descriptors were used to store a single
 481                  * packet, also make sure the frame isn't just CRC only */
 482                 if (!(status & E1000_RXD_STAT_EOP) || (length <= 4)) {
 483                         /* All receives must fit into a single buffer */
 484                         ndev_dbg(netdev, "%s: Receive packet consumed "
 485                                  "multiple buffers\n", netdev->name);
 486                         /* recycle */
 487                         buffer_info->skb = skb;
 488                         goto next_desc;
 489                 }
 490
 491                 if (rx_desc->errors & E1000_RXD_ERR_FRAME_ERR_MASK) {
 492                         /* recycle */
 493                         buffer_info->skb = skb;
 494                         goto next_desc;
 495                 }
 496
 497                 /* adjust length to remove Ethernet CRC */
 498                 length -= 4;
 499
 500                 /* probably a little skewed due to removing CRC */
 501                 total_rx_bytes += length;
 502                 total_rx_packets++;
 503
 504                 /* code added for copybreak, this should improve
 505                  * performance for small packets with large amounts
 506                  * of reassembly being done in the stack */
 507                 if (length < copybreak) {
 508                         struct sk_buff *new_skb =
 509                             netdev_alloc_skb(netdev, length + NET_IP_ALIGN);
 510                         if (new_skb) {
 511                                 skb_reserve(new_skb, NET_IP_ALIGN);
 512                                 memcpy(new_skb->data - NET_IP_ALIGN,
 513                                        skb->data - NET_IP_ALIGN,
 514                                        length + NET_IP_ALIGN);
 515                                 /* save the skb in buffer_info as good */
 516                                 buffer_info->skb = skb;
 517                                 skb = new_skb;
 518                         }
 519                         /* else just continue with the old one */
 520                 }
 521                 /* end copybreak code */
 522                 skb_put(skb, length);
 523
 524                 /* Receive Checksum Offload */
 525                 e1000_rx_checksum(adapter,
 526                                   (u32)(status) |
 527                                   ((u32)(rx_desc->errors) << 24),
 528                                   le16_to_cpu(rx_desc->csum), skb);
 529
 530                 e1000_receive_skb(adapter, netdev, skb,status,rx_desc->special);
 531
 532 next_desc:
 533                 rx_desc->status = 0;
 534
 535                 /* return some buffers to hardware, one at a time is too slow */
 536                 if (cleaned_count >= E1000_RX_BUFFER_WRITE) {
 537                         adapter->alloc_rx_buf(adapter, cleaned_count);
 538                         cleaned_count = 0;
 539                 }
 540
 541                 /* use prefetched values */
 542                 rx_desc = next_rxd;
 543                 buffer_info = next_buffer;
 544         }
 545         rx_ring->next_to_clean = i;
 546
 547         cleaned_count = e1000_desc_unused(rx_ring);
 548         if (cleaned_count)
 549                 adapter->alloc_rx_buf(adapter, cleaned_count);
 550
 551         adapter->total_rx_packets += total_rx_packets;
 552         adapter->total_rx_bytes += total_rx_bytes;
 553         return cleaned;
 554 }
 555
 556 static void e1000_consume_page(struct e1000_buffer *bi, struct sk_buff *skb,
 557                                u16 length)
 558 {
 559         bi->page = NULL;
 560         skb->len += length;
 561         skb->data_len += length;
 562         skb->truesize += length;
 563 }
 564
 565 static void e1000_put_txbuf(struct e1000_adapter *adapter,
 566                              struct e1000_buffer *buffer_info)
 567 {
 568         if (buffer_info->dma) {
 569                 pci_unmap_page(adapter->pdev, buffer_info->dma,
 570                                buffer_info->length, PCI_DMA_TODEVICE);
 571                 buffer_info->dma = 0;
 572         }
 573         if (buffer_info->skb) {
 574                 dev_kfree_skb_any(buffer_info->skb);
 575                 buffer_info->skb = NULL;
 576         }
 577 }
 578
 579 static void e1000_print_tx_hang(struct e1000_adapter *adapter)
 580 {
 581         struct e1000_ring *tx_ring = adapter->tx_ring;
 582         unsigned int i = tx_ring->next_to_clean;
 583         unsigned int eop = tx_ring->buffer_info[i].next_to_watch;
 584         struct e1000_tx_desc *eop_desc = E1000_TX_DESC(*tx_ring, eop);
 585         struct net_device *netdev = adapter->netdev;
 586
 587         /* detected Tx unit hang */
 588         ndev_err(netdev,
 589                  "Detected Tx Unit Hang:\n"
 590                  "  TDH                  <%x>\n"
 591                  "  TDT                  <%x>\n"
 592                  "  next_to_use          <%x>\n"
 593                  "  next_to_clean        <%x>\n"
 594                  "buffer_info[next_to_clean]:\n"
 595                  "  time_stamp           <%lx>\n"
 596                  "  next_to_watch        <%x>\n"
 597                  "  jiffies              <%lx>\n"
 598                  "  next_to_watch.status <%x>\n",
 599                  readl(adapter->hw.hw_addr + tx_ring->head),
 600                  readl(adapter->hw.hw_addr + tx_ring->tail),
 601                  tx_ring->next_to_use,
 602                  tx_ring->next_to_clean,
 603                  tx_ring->buffer_info[eop].time_stamp,
 604                  eop,
 605                  jiffies,
 606                  eop_desc->upper.fields.status);
 607 }
 608
 609 /**
 610  * e1000_clean_tx_irq - Reclaim resources after transmit completes
 611  * @adapter: board private structure
 612  *
 613  * the return value indicates whether actual cleaning was done, there
 614  * is no guarantee that everything was cleaned
 615  **/
 616 static bool e1000_clean_tx_irq(struct e1000_adapter *adapter)
 617 {
 618         struct net_device *netdev = adapter->netdev;
 619         struct e1000_hw *hw = &adapter->hw;
 620         struct e1000_ring *tx_ring = adapter->tx_ring;
 621         struct e1000_tx_desc *tx_desc, *eop_desc;
 622         struct e1000_buffer *buffer_info;
 623         unsigned int i, eop;
 624         unsigned int count = 0;
 625         bool cleaned = 0;
 626         unsigned int total_tx_bytes = 0, total_tx_packets = 0;
 627
 628         i = tx_ring->next_to_clean;
 629         eop = tx_ring->buffer_info[i].next_to_watch;
 630         eop_desc = E1000_TX_DESC(*tx_ring, eop);
 631
 632         while (eop_desc->upper.data & cpu_to_le32(E1000_TXD_STAT_DD)) {
 633                 for (cleaned = 0; !cleaned; ) {
 634                         tx_desc = E1000_TX_DESC(*tx_ring, i);
 635                         buffer_info = &tx_ring->buffer_info[i];
 636                         cleaned = (i == eop);
 637
 638                         if (cleaned) {
 639                                 struct sk_buff *skb = buffer_info->skb;
 640                                 unsigned int segs, bytecount;
 641                                 segs = skb_shinfo(skb)->gso_segs ?: 1;
 642                                 /* multiply data chunks by size of headers */
 643                                 bytecount = ((segs - 1) * skb_headlen(skb)) +
 644                                             skb->len;
 645                                 total_tx_packets += segs;
 646                                 total_tx_bytes += bytecount;
 647                         }
 648
 649                         e1000_put_txbuf(adapter, buffer_info);
 650                         tx_desc->upper.data = 0;
 651
 652                         i++;
 653                         if (i == tx_ring->count)
 654                                 i = 0;
 655                 }
 656
 657                 eop = tx_ring->buffer_info[i].next_to_watch;
 658                 eop_desc = E1000_TX_DESC(*tx_ring, eop);
 659 #define E1000_TX_WEIGHT 64
 660                 /* weight of a sort for tx, to avoid endless transmit cleanup */
 661                 if (count++ == E1000_TX_WEIGHT)
 662                         break;
 663         }
 664
 665         tx_ring->next_to_clean = i;
 666
 667 #define TX_WAKE_THRESHOLD 32
 668         if (cleaned && netif_carrier_ok(netdev) &&
 669                      e1000_desc_unused(tx_ring) >= TX_WAKE_THRESHOLD) {
 670                 /* Make sure that anybody stopping the queue after this
 671                  * sees the new next_to_clean.
 672                  */
 673                 smp_mb();
 674
 675                 if (netif_queue_stopped(netdev) &&
 676                     !(test_bit(__E1000_DOWN, &adapter->state))) {
 677                         netif_wake_queue(netdev);
 678                         ++adapter->restart_queue;
 679                 }
 680         }
 681
 682         if (adapter->detect_tx_hung) {
 683                 /* Detect a transmit hang in hardware, this serializes the
 684                  * check with the clearing of time_stamp and movement of i */
 685                 adapter->detect_tx_hung = 0;
 686                 if (tx_ring->buffer_info[eop].dma &&
 687                     time_after(jiffies, tx_ring->buffer_info[eop].time_stamp
 688                                + (adapter->tx_timeout_factor * HZ))
 689                     && !(er32(STATUS) &
 690                          E1000_STATUS_TXOFF)) {
 691                         e1000_print_tx_hang(adapter);
 692                         netif_stop_queue(netdev);
 693                 }
 694         }
 695         adapter->total_tx_bytes += total_tx_bytes;
 696         adapter->total_tx_packets += total_tx_packets;
 697         return cleaned;
 698 }
 699
 700 /**
 701  * e1000_clean_rx_irq_jumbo - Send received data up the network stack; legacy
 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_jumbo(struct e1000_adapter *adapter,
 708                                      int *work_done, int work_to_do)
 709 {
 710         struct net_device *netdev = adapter->netdev;
 711         struct pci_dev *pdev = adapter->pdev;
 712         struct e1000_ring *rx_ring = adapter->rx_ring;
 713         struct e1000_rx_desc *rx_desc, *next_rxd;
 714         struct e1000_buffer *buffer_info, *next_buffer;
 715         u32 length;
 716         unsigned int i;
 717         int cleaned_count = 0;
 718         bool cleaned = 0;
 719         unsigned int total_rx_bytes = 0, total_rx_packets = 0;
 720
 721         i = rx_ring->next_to_clean;
 722         rx_desc = E1000_RX_DESC(*rx_ring, i);
 723         buffer_info = &rx_ring->buffer_info[i];
 724
 725         while (rx_desc->status & E1000_RXD_STAT_DD) {
 726                 struct sk_buff *skb;
 727                 u8 status;
 728
 729                 if (*work_done >= work_to_do)
 730                         break;
 731                 (*work_done)++;
 732
 733                 status = rx_desc->status;
 734                 skb = buffer_info->skb;
 735                 buffer_info->skb = NULL;
 736
 737                 i++;
 738                 if (i == rx_ring->count)
 739                         i = 0;
 740                 next_rxd = E1000_RX_DESC(*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_page(pdev,
 748                                buffer_info->dma,
 749                                PAGE_SIZE,
 750                                PCI_DMA_FROMDEVICE);
 751                 buffer_info->dma = 0;
 752
 753                 length = le16_to_cpu(rx_desc->length);
 754
 755                 /* errors is only valid for DD + EOP descriptors */
 756                 if ((status & E1000_RXD_STAT_EOP) &&
 757                     (rx_desc->errors & E1000_RXD_ERR_FRAME_ERR_MASK)) {
 758                         /* recycle both page and skb */
 759                         buffer_info->skb = skb;
 760                         /* an error means any chain goes out the window too */
 761                         if (rx_ring->rx_skb_top)
 762                                 dev_kfree_skb(rx_ring->rx_skb_top);
 763                         rx_ring->rx_skb_top = NULL;
 764                         goto next_desc;
 765                 }
 766
 767 #define rxtop rx_ring->rx_skb_top
 768                 if (!(status & E1000_RXD_STAT_EOP)) {
 769                         /* this descriptor is only the beginning (or middle) */
 770                         if (!rxtop) {
 771                                 /* this is the beginning of a chain */
 772                                 rxtop = skb;
 773                                 skb_fill_page_desc(rxtop, 0, buffer_info->page,
 774                                                    0, length);
 775                         } else {
 776                                 /* this is the middle of a chain */
 777                                 skb_fill_page_desc(rxtop,
 778                                                    skb_shinfo(rxtop)->nr_frags,
 779                                                    buffer_info->page, 0,
 780                                                    length);
 781                                 /* re-use the skb, only consumed the page */
 782                                 buffer_info->skb = skb;
 783                         }
 784                         e1000_consume_page(buffer_info, rxtop, length);
 785                         goto next_desc;
 786                 } else {
 787                         if (rxtop) {
 788                                 /* end of the chain */
 789                                 skb_fill_page_desc(rxtop,
 790                                     skb_shinfo(rxtop)->nr_frags,
 791                                     buffer_info->page, 0, length);
 792                                 /* re-use the current skb, we only consumed the
 793                                  * page */
 794                                 buffer_info->skb = skb;
 795                                 skb = rxtop;
 796                                 rxtop = NULL;
 797                                 e1000_consume_page(buffer_info, skb, length);
 798                         } else {
 799                                 /* no chain, got EOP, this buf is the packet
 800                                  * copybreak to save the put_page/alloc_page */
 801                                 if (length <= copybreak &&
 802                                     skb_tailroom(skb) >= length) {
 803                                         u8 *vaddr;
 804                                         vaddr = kmap_atomic(buffer_info->page,
 805                                                            KM_SKB_DATA_SOFTIRQ);
 806                                         memcpy(skb_tail_pointer(skb),
 807                                                vaddr, length);
 808                                         kunmap_atomic(vaddr,
 809                                                       KM_SKB_DATA_SOFTIRQ);
 810                                         /* re-use the page, so don't erase
 811                                          * buffer_info->page */
 812                                         skb_put(skb, length);
 813                                 } else {
 814                                         skb_fill_page_desc(skb, 0,
 815                                                            buffer_info->page, 0,
 816                                                            length);
 817                                         e1000_consume_page(buffer_info, skb,
 818                                                            length);
 819                                 }
 820                         }
 821                 }
 822
 823                 /* Receive Checksum Offload XXX recompute due to CRC strip? */
 824                 e1000_rx_checksum(adapter,
 825                                   (u32)(status) |
 826                                   ((u32)(rx_desc->errors) << 24),
 827                                   le16_to_cpu(rx_desc->csum), skb);
 828
 829                 pskb_trim(skb, skb->len - 4);
 830
 831                 /* probably a little skewed due to removing CRC */
 832                 total_rx_bytes += skb->len;
 833                 total_rx_packets++;
 834
 835                 /* eth type trans needs skb->data to point to something */
 836                 if (!pskb_may_pull(skb, ETH_HLEN)) {
 837                         ndev_err(netdev, "__pskb_pull_tail failed.\n");
 838                         dev_kfree_skb(skb);
 839                         goto next_desc;
 840                 }
 841
 842                 e1000_receive_skb(adapter, netdev, skb,status,rx_desc->special);
 843
 844 next_desc:
 845                 rx_desc->status = 0;
 846
 847                 /* return some buffers to hardware, one at a time is too slow */
 848                 if (cleaned_count >= E1000_RX_BUFFER_WRITE) {
 849                         adapter->alloc_rx_buf(adapter, cleaned_count);
 850                         cleaned_count = 0;
 851                 }
 852
 853                 /* use prefetched values */
 854                 rx_desc = next_rxd;
 855                 buffer_info = next_buffer;
 856         }
 857         rx_ring->next_to_clean = i;
 858
 859         cleaned_count = e1000_desc_unused(rx_ring);
 860         if (cleaned_count)
 861                 adapter->alloc_rx_buf(adapter, cleaned_count);
 862
 863         adapter->total_rx_packets += total_rx_packets;
 864         adapter->total_rx_bytes += total_rx_bytes;
 865         return cleaned;
 866 }
 867
 868 /**
 869  * e1000_clean_rx_irq_ps - Send received data up the network stack; packet split
 870  * @adapter: board private structure
 871  *
 872  * the return value indicates whether actual cleaning was done, there
 873  * is no guarantee that everything was cleaned
 874  **/
 875 static bool e1000_clean_rx_irq_ps(struct e1000_adapter *adapter,
 876                                   int *work_done, int work_to_do)
 877 {
 878         union e1000_rx_desc_packet_split *rx_desc, *next_rxd;
 879         struct net_device *netdev = adapter->netdev;
 880         struct pci_dev *pdev = adapter->pdev;
 881         struct e1000_ring *rx_ring = adapter->rx_ring;
 882         struct e1000_buffer *buffer_info, *next_buffer;
 883         struct e1000_ps_page *ps_page;
 884         struct sk_buff *skb;
 885         unsigned int i, j;
 886         u32 length, staterr;
 887         int cleaned_count = 0;
 888         bool cleaned = 0;
 889         unsigned int total_rx_bytes = 0, total_rx_packets = 0;
 890
 891         i = rx_ring->next_to_clean;
 892         rx_desc = E1000_RX_DESC_PS(*rx_ring, i);
 893         staterr = le32_to_cpu(rx_desc->wb.middle.status_error);
 894         buffer_info = &rx_ring->buffer_info[i];
 895
 896         while (staterr & E1000_RXD_STAT_DD) {
 897                 if (*work_done >= work_to_do)
 898                         break;
 899                 (*work_done)++;
 900                 skb = buffer_info->skb;
 901
 902                 /* in the packet split case this is header only */
 903                 prefetch(skb->data - NET_IP_ALIGN);
 904
 905                 i++;
 906                 if (i == rx_ring->count)
 907                         i = 0;
 908                 next_rxd = E1000_RX_DESC_PS(*rx_ring, i);
 909                 prefetch(next_rxd);
 910
 911                 next_buffer = &rx_ring->buffer_info[i];
 912
 913                 cleaned = 1;
 914                 cleaned_count++;
 915                 pci_unmap_single(pdev, buffer_info->dma,
 916                                  adapter->rx_ps_bsize0,
 917                                  PCI_DMA_FROMDEVICE);
 918                 buffer_info->dma = 0;
 919
 920                 if (!(staterr & E1000_RXD_STAT_EOP)) {
 921                         ndev_dbg(netdev, "%s: Packet Split buffers didn't pick "
 922                                  "up the full packet\n", netdev->name);
 923                         dev_kfree_skb_irq(skb);
 924                         goto next_desc;
 925                 }
 926
 927                 if (staterr & E1000_RXDEXT_ERR_FRAME_ERR_MASK) {
 928                         dev_kfree_skb_irq(skb);
 929                         goto next_desc;
 930                 }
 931
 932                 length = le16_to_cpu(rx_desc->wb.middle.length0);
 933
 934                 if (!length) {
 935                         ndev_dbg(netdev, "%s: Last part of the packet spanning"
 936                                  " multiple descriptors\n", netdev->name);
 937                         dev_kfree_skb_irq(skb);
 938                         goto next_desc;
 939                 }
 940
 941                 /* Good Receive */
 942                 skb_put(skb, length);
 943
 944                 {
 945                 /* this looks ugly, but it seems compiler issues make it
 946                    more efficient than reusing j */
 947                 int l1 = le16_to_cpu(rx_desc->wb.upper.length[0]);
 948
 949                 /* page alloc/put takes too long and effects small packet
 950                  * throughput, so unsplit small packets and save the alloc/put*/
 951                 if (l1 && (l1 <= copybreak) &&
 952                     ((length + l1) <= adapter->rx_ps_bsize0)) {
 953                         u8 *vaddr;
 954
 955                         ps_page = &buffer_info->ps_pages[0];
 956
 957                         /* there is no documentation about how to call
 958                          * kmap_atomic, so we can't hold the mapping
 959                          * very long */
 960                         pci_dma_sync_single_for_cpu(pdev, ps_page->dma,
 961                                 PAGE_SIZE, PCI_DMA_FROMDEVICE);
 962                         vaddr = kmap_atomic(ps_page->page, KM_SKB_DATA_SOFTIRQ);
 963                         memcpy(skb_tail_pointer(skb), vaddr, l1);
 964                         kunmap_atomic(vaddr, KM_SKB_DATA_SOFTIRQ);
 965                         pci_dma_sync_single_for_device(pdev, ps_page->dma,
 966                                 PAGE_SIZE, PCI_DMA_FROMDEVICE);
 967                         /* remove the CRC */
 968                         l1 -= 4;
 969                         skb_put(skb, l1);
 970                         goto copydone;
 971                 } /* if */
 972                 }
 973
 974                 for (j = 0; j < PS_PAGE_BUFFERS; j++) {
 975                         length = le16_to_cpu(rx_desc->wb.upper.length[j]);
 976                         if (!length)
 977                                 break;
 978
 979                         ps_page = &buffer_info->ps_pages[j];
 980                         pci_unmap_page(pdev, ps_page->dma, PAGE_SIZE,
 981                                        PCI_DMA_FROMDEVICE);
 982                         ps_page->dma = 0;
 983                         skb_fill_page_desc(skb, j, ps_page->page, 0, length);
 984                         ps_page->page = NULL;
 985                         skb->len += length;
 986                         skb->data_len += length;
 987                         skb->truesize += length;
 988                 }
 989
 990                 /* strip the ethernet crc, problem is we're using pages now so
 991                  * this whole operation can get a little cpu intensive */
 992                 pskb_trim(skb, skb->len - 4);
 993
 994 copydone:
 995                 total_rx_bytes += skb->len;
 996                 total_rx_packets++;
 997
 998                 e1000_rx_checksum(adapter, staterr, le16_to_cpu(
 999                         rx_desc->wb.lower.hi_dword.csum_ip.csum), skb);
1000
1001                 if (rx_desc->wb.upper.header_status &
1002                            cpu_to_le16(E1000_RXDPS_HDRSTAT_HDRSP))
1003                         adapter->rx_hdr_split++;
1004
1005                 e1000_receive_skb(adapter, netdev, skb,
1006                                   staterr, rx_desc->wb.middle.vlan);
1007
1008 next_desc:
1009                 rx_desc->wb.middle.status_error &= cpu_to_le32(~0xFF);
1010                 buffer_info->skb = NULL;
1011
1012                 /* return some buffers to hardware, one at a time is too slow */
1013                 if (cleaned_count >= E1000_RX_BUFFER_WRITE) {
1014                         adapter->alloc_rx_buf(adapter, cleaned_count);
1015                         cleaned_count = 0;
1016                 }
1017
1018                 /* use prefetched values */
1019                 rx_desc = next_rxd;
1020                 buffer_info = next_buffer;
1021
1022                 staterr = le32_to_cpu(rx_desc->wb.middle.status_error);
1023         }
1024         rx_ring->next_to_clean = i;
1025
1026         cleaned_count = e1000_desc_unused(rx_ring);
1027         if (cleaned_count)
1028                 adapter->alloc_rx_buf(adapter, cleaned_count);
1029
1030         adapter->total_rx_packets += total_rx_packets;
1031         adapter->total_rx_bytes += total_rx_bytes;
1032         return cleaned;
1033 }
1034
1035 /**
1036  * e1000_clean_rx_ring - Free Rx Buffers per Queue
1037  * @adapter: board private structure
1038  **/
1039 static void e1000_clean_rx_ring(struct e1000_adapter *adapter)
1040 {
1041         struct e1000_ring *rx_ring = adapter->rx_ring;
1042         struct e1000_buffer *buffer_info;
1043         struct e1000_ps_page *ps_page;
1044         struct pci_dev *pdev = adapter->pdev;
1045         unsigned int i, j;
1046
1047         /* Free all the Rx ring sk_buffs */
1048         for (i = 0; i < rx_ring->count; i++) {
1049                 buffer_info = &rx_ring->buffer_info[i];
1050                 if (buffer_info->dma) {
1051                         if (adapter->clean_rx == e1000_clean_rx_irq)
1052                                 pci_unmap_single(pdev, buffer_info->dma,
1053                                                  adapter->rx_buffer_len,
1054                                                  PCI_DMA_FROMDEVICE);
1055                         else if (adapter->clean_rx == e1000_clean_rx_irq_jumbo)
1056                                 pci_unmap_page(pdev, buffer_info->dma,
1057                                                PAGE_SIZE, PCI_DMA_FROMDEVICE);
1058                         else if (adapter->clean_rx == e1000_clean_rx_irq_ps)
1059                                 pci_unmap_single(pdev, buffer_info->dma,
1060                                                  adapter->rx_ps_bsize0,
1061                                                  PCI_DMA_FROMDEVICE);
1062                         buffer_info->dma = 0;
1063                 }
1064
1065                 if (buffer_info->page) {
1066                         put_page(buffer_info->page);
1067                         buffer_info->page = NULL;
1068                 }
1069
1070                 if (buffer_info->skb) {
1071                         dev_kfree_skb(buffer_info->skb);
1072                         buffer_info->skb = NULL;
1073                 }
1074
1075                 for (j = 0; j < PS_PAGE_BUFFERS; j++) {
1076                         ps_page = &buffer_info->ps_pages[j];
1077                         if (!ps_page->page)
1078                                 break;
1079                         pci_unmap_page(pdev, ps_page->dma, PAGE_SIZE,
1080                                        PCI_DMA_FROMDEVICE);
1081                         ps_page->dma = 0;
1082                         put_page(ps_page->page);
1083                         ps_page->page = NULL;
1084                 }
1085         }
1086
1087         /* there also may be some cached data from a chained receive */
1088         if (rx_ring->rx_skb_top) {
1089                 dev_kfree_skb(rx_ring->rx_skb_top);
1090                 rx_ring->rx_skb_top = NULL;
1091         }
1092
1093         /* Zero out the descriptor ring */
1094         memset(rx_ring->desc, 0, rx_ring->size);
1095
1096         rx_ring->next_to_clean = 0;
1097         rx_ring->next_to_use = 0;
1098
1099         writel(0, adapter->hw.hw_addr + rx_ring->head);
1100         writel(0, adapter->hw.hw_addr + rx_ring->tail);
1101 }
1102
1103 /**
1104  * e1000_intr_msi - Interrupt Handler
1105  * @irq: interrupt number
1106  * @data: pointer to a network interface device structure
1107  **/
1108 static irqreturn_t e1000_intr_msi(int irq, void *data)
1109 {
1110         struct net_device *netdev = data;
1111         struct e1000_adapter *adapter = netdev_priv(netdev);
1112         struct e1000_hw *hw = &adapter->hw;
1113         u32 icr = er32(ICR);
1114
1115         /* read ICR disables interrupts using IAM, so keep up with our
1116          * enable/disable accounting */
1117         atomic_inc(&adapter->irq_sem);
1118
1119         if (icr & (E1000_ICR_RXSEQ | E1000_ICR_LSC)) {
1120                 hw->mac.get_link_status = 1;
1121                 /* ICH8 workaround-- Call gig speed drop workaround on cable
1122                  * disconnect (LSC) before accessing any PHY registers */
1123                 if ((adapter->flags & FLAG_LSC_GIG_SPEED_DROP) &&
1124                     (!(er32(STATUS) & E1000_STATUS_LU)))
1125                         e1000e_gig_downshift_workaround_ich8lan(hw);
1126
1127                 /* 80003ES2LAN workaround-- For packet buffer work-around on
1128                  * link down event; disable receives here in the ISR and reset
1129                  * adapter in watchdog */
1130                 if (netif_carrier_ok(netdev) &&
1131                     adapter->flags & FLAG_RX_NEEDS_RESTART) {
1132                         /* disable receives */
1133                         u32 rctl = er32(RCTL);
1134                         ew32(RCTL, rctl & ~E1000_RCTL_EN);
1135                 }
1136                 /* guard against interrupt when we're going down */
1137                 if (!test_bit(__E1000_DOWN, &adapter->state))
1138                         mod_timer(&adapter->watchdog_timer, jiffies + 1);
1139         }
1140
1141         if (netif_rx_schedule_prep(netdev, &adapter->napi)) {
1142                 adapter->total_tx_bytes = 0;
1143                 adapter->total_tx_packets = 0;
1144                 adapter->total_rx_bytes = 0;
1145                 adapter->total_rx_packets = 0;
1146                 __netif_rx_schedule(netdev, &adapter->napi);
1147         } else {
1148                 atomic_dec(&adapter->irq_sem);
1149         }
1150
1151         return IRQ_HANDLED;
1152 }
1153
1154 /**
1155  * e1000_intr - Interrupt Handler
1156  * @irq: interrupt number
1157  * @data: pointer to a network interface device structure
1158  **/
1159 static irqreturn_t e1000_intr(int irq, void *data)
1160 {
1161         struct net_device *netdev = data;
1162         struct e1000_adapter *adapter = netdev_priv(netdev);
1163         struct e1000_hw *hw = &adapter->hw;
1164
1165         u32 rctl, icr = er32(ICR);
1166         if (!icr)
1167                 return IRQ_NONE;  /* Not our interrupt */
1168
1169         /* IMS will not auto-mask if INT_ASSERTED is not set, and if it is
1170          * not set, then the adapter didn't send an interrupt */
1171         if (!(icr & E1000_ICR_INT_ASSERTED))
1172                 return IRQ_NONE;
1173
1174         /* Interrupt Auto-Mask...upon reading ICR,
1175          * interrupts are masked.  No need for the
1176          * IMC write, but it does mean we should
1177          * account for it ASAP. */
1178         atomic_inc(&adapter->irq_sem);
1179
1180         if (icr & (E1000_ICR_RXSEQ | E1000_ICR_LSC)) {
1181                 hw->mac.get_link_status = 1;
1182                 /* ICH8 workaround-- Call gig speed drop workaround on cable
1183                  * disconnect (LSC) before accessing any PHY registers */
1184                 if ((adapter->flags & FLAG_LSC_GIG_SPEED_DROP) &&
1185                     (!(er32(STATUS) & E1000_STATUS_LU)))
1186                         e1000e_gig_downshift_workaround_ich8lan(hw);
1187
1188                 /* 80003ES2LAN workaround--
1189                  * For packet buffer work-around on link down event;
1190                  * disable receives here in the ISR and
1191                  * reset adapter in watchdog
1192                  */
1193                 if (netif_carrier_ok(netdev) &&
1194                     (adapter->flags & FLAG_RX_NEEDS_RESTART)) {
1195                         /* disable receives */
1196                         rctl = er32(RCTL);
1197                         ew32(RCTL, rctl & ~E1000_RCTL_EN);
1198                 }
1199                 /* guard against interrupt when we're going down */
1200                 if (!test_bit(__E1000_DOWN, &adapter->state))
1201                         mod_timer(&adapter->watchdog_timer, jiffies + 1);
1202         }
1203
1204         if (netif_rx_schedule_prep(netdev, &adapter->napi)) {
1205                 adapter->total_tx_bytes = 0;
1206                 adapter->total_tx_packets = 0;
1207                 adapter->total_rx_bytes = 0;
1208                 adapter->total_rx_packets = 0;
1209                 __netif_rx_schedule(netdev, &adapter->napi);
1210         } else {
1211                 atomic_dec(&adapter->irq_sem);
1212         }
1213
1214         return IRQ_HANDLED;
1215 }
1216
1217 static int e1000_request_irq(struct e1000_adapter *adapter)
1218 {
1219         struct net_device *netdev = adapter->netdev;
1220         void (*handler) = &e1000_intr;
1221         int irq_flags = IRQF_SHARED;
1222         int err;
1223
1224         err = pci_enable_msi(adapter->pdev);
1225         if (err) {
1226                 ndev_warn(netdev,
1227                  "Unable to allocate MSI interrupt Error: %d\n", err);
1228         } else {
1229                 adapter->flags |= FLAG_MSI_ENABLED;
1230                 handler = &e1000_intr_msi;
1231                 irq_flags = 0;
1232         }
1233
1234         err = request_irq(adapter->pdev->irq, handler, irq_flags, netdev->name,
1235                           netdev);
1236         if (err) {
1237                 if (adapter->flags & FLAG_MSI_ENABLED)
1238                         pci_disable_msi(adapter->pdev);
1239                 ndev_err(netdev,
1240                        "Unable to allocate interrupt Error: %d\n", err);
1241         }
1242
1243         return err;
1244 }
1245
1246 static void e1000_free_irq(struct e1000_adapter *adapter)
1247 {
1248         struct net_device *netdev = adapter->netdev;
1249
1250         free_irq(adapter->pdev->irq, netdev);
1251         if (adapter->flags & FLAG_MSI_ENABLED) {
1252                 pci_disable_msi(adapter->pdev);
1253                 adapter->flags &= ~FLAG_MSI_ENABLED;
1254         }
1255 }
1256
1257 /**
1258  * e1000_irq_disable - Mask off interrupt generation on the NIC
1259  **/
1260 static void e1000_irq_disable(struct e1000_adapter *adapter)
1261 {
1262         struct e1000_hw *hw = &adapter->hw;
1263
1264         atomic_inc(&adapter->irq_sem);
1265         ew32(IMC, ~0);
1266         e1e_flush();
1267         synchronize_irq(adapter->pdev->irq);
1268 }
1269
1270 /**
1271  * e1000_irq_enable - Enable default interrupt generation settings
1272  **/
1273 static void e1000_irq_enable(struct e1000_adapter *adapter)
1274 {
1275         struct e1000_hw *hw = &adapter->hw;
1276
1277         if (atomic_dec_and_test(&adapter->irq_sem)) {
1278                 ew32(IMS, IMS_ENABLE_MASK);
1279                 e1e_flush();
1280         }
1281 }
1282
1283 /**
1284  * e1000_get_hw_control - get control of the h/w from f/w
1285  * @adapter: address of board private structure
1286  *
1287  * e1000_get_hw_control sets {CTRL_EXT|FWSM}:DRV_LOAD bit.
1288  * For ASF and Pass Through versions of f/w this means that
1289  * the driver is loaded. For AMT version (only with 82573)
1290  * of the f/w this means that the network i/f is open.
1291  **/
1292 static void e1000_get_hw_control(struct e1000_adapter *adapter)
1293 {
1294         struct e1000_hw *hw = &adapter->hw;
1295         u32 ctrl_ext;
1296         u32 swsm;
1297
1298         /* Let firmware know the driver has taken over */
1299         if (adapter->flags & FLAG_HAS_SWSM_ON_LOAD) {
1300                 swsm = er32(SWSM);
1301                 ew32(SWSM, swsm | E1000_SWSM_DRV_LOAD);
1302         } else if (adapter->flags & FLAG_HAS_CTRLEXT_ON_LOAD) {
1303                 ctrl_ext = er32(CTRL_EXT);
1304                 ew32(CTRL_EXT,
1305                                 ctrl_ext | E1000_CTRL_EXT_DRV_LOAD);
1306         }
1307 }
1308
1309 /**
1310  * e1000_release_hw_control - release control of the h/w to f/w
1311  * @adapter: address of board private structure
1312  *
1313  * e1000_release_hw_control resets {CTRL_EXT|FWSM}:DRV_LOAD bit.
1314  * For ASF and Pass Through versions of f/w this means that the
1315  * driver is no longer loaded. For AMT version (only with 82573) i
1316  * of the f/w this means that the network i/f is closed.
1317  *
1318  **/
1319 static void e1000_release_hw_control(struct e1000_adapter *adapter)
1320 {
1321         struct e1000_hw *hw = &adapter->hw;
1322         u32 ctrl_ext;
1323         u32 swsm;
1324
1325         /* Let firmware taken over control of h/w */
1326         if (adapter->flags & FLAG_HAS_SWSM_ON_LOAD) {
1327                 swsm = er32(SWSM);
1328                 ew32(SWSM, swsm & ~E1000_SWSM_DRV_LOAD);
1329         } else if (adapter->flags & FLAG_HAS_CTRLEXT_ON_LOAD) {
1330                 ctrl_ext = er32(CTRL_EXT);
1331                 ew32(CTRL_EXT,
1332                                 ctrl_ext & ~E1000_CTRL_EXT_DRV_LOAD);
1333         }
1334 }
1335
1336 static void e1000_release_manageability(struct e1000_adapter *adapter)
1337 {
1338         if (adapter->flags & FLAG_MNG_PT_ENABLED) {
1339                 struct e1000_hw *hw = &adapter->hw;
1340
1341                 u32 manc = er32(MANC);
1342
1343                 /* re-enable hardware interception of ARP */
1344                 manc |= E1000_MANC_ARP_EN;
1345                 manc &= ~E1000_MANC_EN_MNG2HOST;
1346
1347                 /* don't explicitly have to mess with MANC2H since
1348                  * MANC has an enable disable that gates MANC2H */
1349                 ew32(MANC, manc);
1350         }
1351 }
1352
1353 /**
1354  * @e1000_alloc_ring - allocate memory for a ring structure
1355  **/
1356 static int e1000_alloc_ring_dma(struct e1000_adapter *adapter,
1357                                 struct e1000_ring *ring)
1358 {
1359         struct pci_dev *pdev = adapter->pdev;
1360
1361         ring->desc = dma_alloc_coherent(&pdev->dev, ring->size, &ring->dma,
1362                                         GFP_KERNEL);
1363         if (!ring->desc)
1364                 return -ENOMEM;
1365
1366         return 0;
1367 }
1368
1369 /**
1370  * e1000e_setup_tx_resources - allocate Tx resources (Descriptors)
1371  * @adapter: board private structure
1372  *
1373  * Return 0 on success, negative on failure
1374  **/
1375 int e1000e_setup_tx_resources(struct e1000_adapter *adapter)
1376 {
1377         struct e1000_ring *tx_ring = adapter->tx_ring;
1378         int err = -ENOMEM, size;
1379
1380         size = sizeof(struct e1000_buffer) * tx_ring->count;
1381         tx_ring->buffer_info = vmalloc(size);
1382         if (!tx_ring->buffer_info)
1383                 goto err;
1384         memset(tx_ring->buffer_info, 0, size);
1385
1386         /* round up to nearest 4K */
1387         tx_ring->size = tx_ring->count * sizeof(struct e1000_tx_desc);
1388         tx_ring->size = ALIGN(tx_ring->size, 4096);
1389
1390         err = e1000_alloc_ring_dma(adapter, tx_ring);
1391         if (err)
1392                 goto err;
1393
1394         tx_ring->next_to_use = 0;
1395         tx_ring->next_to_clean = 0;
1396         spin_lock_init(&adapter->tx_queue_lock);
1397
1398         return 0;
1399 err:
1400         vfree(tx_ring->buffer_info);
1401         ndev_err(adapter->netdev,
1402         "Unable to allocate memory for the transmit descriptor ring\n");
1403         return err;
1404 }
1405
1406 /**
1407  * e1000e_setup_rx_resources - allocate Rx resources (Descriptors)
1408  * @adapter: board private structure
1409  *
1410  * Returns 0 on success, negative on failure
1411  **/
1412 int e1000e_setup_rx_resources(struct e1000_adapter *adapter)
1413 {
1414         struct e1000_ring *rx_ring = adapter->rx_ring;
1415         struct e1000_buffer *buffer_info;
1416         int i, size, desc_len, err = -ENOMEM;
1417
1418         size = sizeof(struct e1000_buffer) * rx_ring->count;
1419         rx_ring->buffer_info = vmalloc(size);
1420         if (!rx_ring->buffer_info)
1421                 goto err;
1422         memset(rx_ring->buffer_info, 0, size);
1423
1424         for (i = 0; i < rx_ring->count; i++) {
1425                 buffer_info = &rx_ring->buffer_info[i];
1426                 buffer_info->ps_pages = kcalloc(PS_PAGE_BUFFERS,
1427                                                 sizeof(struct e1000_ps_page),
1428                                                 GFP_KERNEL);
1429                 if (!buffer_info->ps_pages)
1430                         goto err_pages;
1431         }
1432
1433         desc_len = sizeof(union e1000_rx_desc_packet_split);
1434
1435         /* Round up to nearest 4K */
1436         rx_ring->size = rx_ring->count * desc_len;
1437         rx_ring->size = ALIGN(rx_ring->size, 4096);
1438
1439         err = e1000_alloc_ring_dma(adapter, rx_ring);
1440         if (err)
1441                 goto err_pages;
1442
1443         rx_ring->next_to_clean = 0;
1444         rx_ring->next_to_use = 0;
1445         rx_ring->rx_skb_top = NULL;
1446
1447         return 0;
1448
1449 err_pages:
1450         for (i = 0; i < rx_ring->count; i++) {
1451                 buffer_info = &rx_ring->buffer_info[i];
1452                 kfree(buffer_info->ps_pages);
1453         }
1454 err:
1455         vfree(rx_ring->buffer_info);
1456         ndev_err(adapter->netdev,
1457         "Unable to allocate memory for the transmit descriptor ring\n");
1458         return err;
1459 }
1460
1461 /**
1462  * e1000_clean_tx_ring - Free Tx Buffers
1463  * @adapter: board private structure
1464  **/
1465 static void e1000_clean_tx_ring(struct e1000_adapter *adapter)
1466 {
1467         struct e1000_ring *tx_ring = adapter->tx_ring;
1468         struct e1000_buffer *buffer_info;
1469         unsigned long size;
1470         unsigned int i;
1471
1472         for (i = 0; i < tx_ring->count; i++) {
1473                 buffer_info = &tx_ring->buffer_info[i];
1474                 e1000_put_txbuf(adapter, buffer_info);
1475         }
1476
1477         size = sizeof(struct e1000_buffer) * tx_ring->count;
1478         memset(tx_ring->buffer_info, 0, size);
1479
1480         memset(tx_ring->desc, 0, tx_ring->size);
1481
1482         tx_ring->next_to_use = 0;
1483         tx_ring->next_to_clean = 0;
1484
1485         writel(0, adapter->hw.hw_addr + tx_ring->head);
1486         writel(0, adapter->hw.hw_addr + tx_ring->tail);
1487 }
1488
1489 /**
1490  * e1000e_free_tx_resources - Free Tx Resources per Queue
1491  * @adapter: board private structure
1492  *
1493  * Free all transmit software resources
1494  **/
1495 void e1000e_free_tx_resources(struct e1000_adapter *adapter)
1496 {
1497         struct pci_dev *pdev = adapter->pdev;
1498         struct e1000_ring *tx_ring = adapter->tx_ring;
1499
1500         e1000_clean_tx_ring(adapter);
1501
1502         vfree(tx_ring->buffer_info);
1503         tx_ring->buffer_info = NULL;
1504
1505         dma_free_coherent(&pdev->dev, tx_ring->size, tx_ring->desc,
1506                           tx_ring->dma);
1507         tx_ring->desc = NULL;
1508 }
1509
1510 /**
1511  * e1000e_free_rx_resources - Free Rx Resources
1512  * @adapter: board private structure
1513  *
1514  * Free all receive software resources
1515  **/
1516
1517 void e1000e_free_rx_resources(struct e1000_adapter *adapter)
1518 {
1519         struct pci_dev *pdev = adapter->pdev;
1520         struct e1000_ring *rx_ring = adapter->rx_ring;
1521         int i;
1522
1523         e1000_clean_rx_ring(adapter);
1524
1525         for (i = 0; i < rx_ring->count; i++) {
1526                 kfree(rx_ring->buffer_info[i].ps_pages);
1527         }
1528
1529         vfree(rx_ring->buffer_info);
1530         rx_ring->buffer_info = NULL;
1531
1532         dma_free_coherent(&pdev->dev, rx_ring->size, rx_ring->desc,
1533                           rx_ring->dma);
1534         rx_ring->desc = NULL;
1535 }
1536
1537 /**
1538  * e1000_update_itr - update the dynamic ITR value based on statistics
1539  *      Stores a new ITR value based on packets and byte
1540  *      counts during the last interrupt.  The advantage of per interrupt
1541  *      computation is faster updates and more accurate ITR for the current
1542  *      traffic pattern.  Constants in this function were computed
1543  *      based on theoretical maximum wire speed and thresholds were set based
1544  *      on testing data as well as attempting to minimize response time
1545  *      while increasing bulk throughput.
1546  *      this functionality is controlled by the InterruptThrottleRate module
1547  *      parameter (see e1000_param.c)
1548  * @adapter: pointer to adapter
1549  * @itr_setting: current adapter->itr
1550  * @packets: the number of packets during this measurement interval
1551  * @bytes: the number of bytes during this measurement interval
1552  **/
1553 static unsigned int e1000_update_itr(struct e1000_adapter *adapter,
1554                                      u16 itr_setting, int packets,
1555                                      int bytes)
1556 {
1557         unsigned int retval = itr_setting;
1558
1559         if (packets == 0)
1560                 goto update_itr_done;
1561
1562         switch (itr_setting) {
1563         case lowest_latency:
1564                 /* handle TSO and jumbo frames */
1565                 if (bytes/packets > 8000)
1566                         retval = bulk_latency;
1567                 else if ((packets < 5) && (bytes > 512)) {
1568                         retval = low_latency;
1569                 }
1570                 break;
1571         case low_latency:  /* 50 usec aka 20000 ints/s */
1572                 if (bytes > 10000) {
1573                         /* this if handles the TSO accounting */
1574                         if (bytes/packets > 8000) {
1575                                 retval = bulk_latency;
1576                         } else if ((packets < 10) || ((bytes/packets) > 1200)) {
1577                                 retval = bulk_latency;
1578                         } else if ((packets > 35)) {
1579                                 retval = lowest_latency;
1580                         }
1581                 } else if (bytes/packets > 2000) {
1582                         retval = bulk_latency;
1583                 } else if (packets <= 2 && bytes < 512) {
1584                         retval = lowest_latency;
1585                 }
1586                 break;
1587         case bulk_latency: /* 250 usec aka 4000 ints/s */
1588                 if (bytes > 25000) {
1589                         if (packets > 35) {
1590                                 retval = low_latency;
1591                         }
1592                 } else if (bytes < 6000) {
1593                         retval = low_latency;
1594                 }
1595                 break;
1596         }
1597
1598 update_itr_done:
1599         return retval;
1600 }
1601
1602 static void e1000_set_itr(struct e1000_adapter *adapter)
1603 {
1604         struct e1000_hw *hw = &adapter->hw;
1605         u16 current_itr;
1606         u32 new_itr = adapter->itr;
1607
1608         /* for non-gigabit speeds, just fix the interrupt rate at 4000 */
1609         if (adapter->link_speed != SPEED_1000) {
1610                 current_itr = 0;
1611                 new_itr = 4000;
1612                 goto set_itr_now;
1613         }
1614
1615         adapter->tx_itr = e1000_update_itr(adapter,
1616                                     adapter->tx_itr,
1617                                     adapter->total_tx_packets,
1618                                     adapter->total_tx_bytes);
1619         /* conservative mode (itr 3) eliminates the lowest_latency setting */
1620         if (adapter->itr_setting == 3 && adapter->tx_itr == lowest_latency)
1621                 adapter->tx_itr = low_latency;
1622
1623         adapter->rx_itr = e1000_update_itr(adapter,
1624                                     adapter->rx_itr,
1625                                     adapter->total_rx_packets,
1626                                     adapter->total_rx_bytes);
1627         /* conservative mode (itr 3) eliminates the lowest_latency setting */
1628         if (adapter->itr_setting == 3 && adapter->rx_itr == lowest_latency)
1629                 adapter->rx_itr = low_latency;
1630
1631         current_itr = max(adapter->rx_itr, adapter->tx_itr);
1632
1633         switch (current_itr) {
1634         /* counts and packets in update_itr are dependent on these numbers */
1635         case lowest_latency:
1636                 new_itr = 70000;
1637                 break;
1638         case low_latency:
1639                 new_itr = 20000; /* aka hwitr = ~200 */
1640                 break;
1641         case bulk_latency:
1642                 new_itr = 4000;
1643                 break;
1644         default:
1645                 break;
1646         }
1647
1648 set_itr_now:
1649         if (new_itr != adapter->itr) {
1650                 /* this attempts to bias the interrupt rate towards Bulk
1651                  * by adding intermediate steps when interrupt rate is
1652                  * increasing */
1653                 new_itr = new_itr > adapter->itr ?
1654                              min(adapter->itr + (new_itr >> 2), new_itr) :
1655                              new_itr;
1656                 adapter->itr = new_itr;
1657                 ew32(ITR, 1000000000 / (new_itr * 256));
1658         }
1659 }
1660
1661 /**
1662  * e1000_clean - NAPI Rx polling callback
1663  * @adapter: board private structure
1664  **/
1665 static int e1000_clean(struct napi_struct *napi, int budget)
1666 {
1667         struct e1000_adapter *adapter = container_of(napi, struct e1000_adapter, napi);
1668         struct net_device *poll_dev = adapter->netdev;
1669         int tx_cleaned = 0, work_done = 0;
1670
1671         /* Must NOT use netdev_priv macro here. */
1672         adapter = poll_dev->priv;
1673
1674         /* Keep link state information with original netdev */
1675         if (!netif_carrier_ok(poll_dev))
1676                 goto quit_polling;
1677
1678         /* e1000_clean is called per-cpu.  This lock protects
1679          * tx_ring from being cleaned by multiple cpus
1680          * simultaneously.  A failure obtaining the lock means
1681          * tx_ring is currently being cleaned anyway. */
1682         if (spin_trylock(&adapter->tx_queue_lock)) {
1683                 tx_cleaned = e1000_clean_tx_irq(adapter);
1684                 spin_unlock(&adapter->tx_queue_lock);
1685         }
1686
1687         adapter->clean_rx(adapter, &work_done, budget);
1688
1689         /* If no Tx and not enough Rx work done, exit the polling mode */
1690         if ((!tx_cleaned && (work_done < budget)) ||
1691            !netif_running(poll_dev)) {
1692 quit_polling:
1693                 if (adapter->itr_setting & 3)
1694                         e1000_set_itr(adapter);
1695                 netif_rx_complete(poll_dev, napi);
1696                 e1000_irq_enable(adapter);
1697         }
1698
1699         return work_done;
1700 }
1701
1702 static void e1000_vlan_rx_add_vid(struct net_device *netdev, u16 vid)
1703 {
1704         struct e1000_adapter *adapter = netdev_priv(netdev);
1705         struct e1000_hw *hw = &adapter->hw;
1706         u32 vfta, index;
1707
1708         /* don't update vlan cookie if already programmed */
1709         if ((adapter->hw.mng_cookie.status &
1710              E1000_MNG_DHCP_COOKIE_STATUS_VLAN) &&
1711             (vid == adapter->mng_vlan_id))
1712                 return;
1713         /* add VID to filter table */
1714         index = (vid >> 5) & 0x7F;
1715         vfta = E1000_READ_REG_ARRAY(hw, E1000_VFTA, index);
1716         vfta |= (1 << (vid & 0x1F));
1717         e1000e_write_vfta(hw, index, vfta);
1718 }
1719
1720 static void e1000_vlan_rx_kill_vid(struct net_device *netdev, u16 vid)
1721 {
1722         struct e1000_adapter *adapter = netdev_priv(netdev);
1723         struct e1000_hw *hw = &adapter->hw;
1724         u32 vfta, index;
1725
1726         e1000_irq_disable(adapter);
1727         vlan_group_set_device(adapter->vlgrp, vid, NULL);
1728         e1000_irq_enable(adapter);
1729
1730         if ((adapter->hw.mng_cookie.status &
1731              E1000_MNG_DHCP_COOKIE_STATUS_VLAN) &&
1732             (vid == adapter->mng_vlan_id)) {
1733                 /* release control to f/w */
1734                 e1000_release_hw_control(adapter);
1735                 return;
1736         }
1737
1738         /* remove VID from filter table */
1739         index = (vid >> 5) & 0x7F;
1740         vfta = E1000_READ_REG_ARRAY(hw, E1000_VFTA, index);
1741         vfta &= ~(1 << (vid & 0x1F));
1742         e1000e_write_vfta(hw, index, vfta);
1743 }
1744
1745 static void e1000_update_mng_vlan(struct e1000_adapter *adapter)
1746 {
1747         struct net_device *netdev = adapter->netdev;
1748         u16 vid = adapter->hw.mng_cookie.vlan_id;
1749         u16 old_vid = adapter->mng_vlan_id;
1750
1751         if (!adapter->vlgrp)
1752                 return;
1753
1754         if (!vlan_group_get_device(adapter->vlgrp, vid)) {
1755                 adapter->mng_vlan_id = E1000_MNG_VLAN_NONE;
1756                 if (adapter->hw.mng_cookie.status &
1757                         E1000_MNG_DHCP_COOKIE_STATUS_VLAN) {
1758                         e1000_vlan_rx_add_vid(netdev, vid);
1759                         adapter->mng_vlan_id = vid;
1760                 }
1761
1762                 if ((old_vid != (u16)E1000_MNG_VLAN_NONE) &&
1763                                 (vid != old_vid) &&
1764                     !vlan_group_get_device(adapter->vlgrp, old_vid))
1765                         e1000_vlan_rx_kill_vid(netdev, old_vid);
1766         } else {
1767                 adapter->mng_vlan_id = vid;
1768         }
1769 }
1770
1771
1772 static void e1000_vlan_rx_register(struct net_device *netdev,
1773                                    struct vlan_group *grp)
1774 {
1775         struct e1000_adapter *adapter = netdev_priv(netdev);
1776         struct e1000_hw *hw = &adapter->hw;
1777         u32 ctrl, rctl;
1778
1779         e1000_irq_disable(adapter);
1780         adapter->vlgrp = grp;
1781
1782         if (grp) {
1783                 /* enable VLAN tag insert/strip */
1784                 ctrl = er32(CTRL);
1785                 ctrl |= E1000_CTRL_VME;
1786                 ew32(CTRL, ctrl);
1787
1788                 if (adapter->flags & FLAG_HAS_HW_VLAN_FILTER) {
1789                         /* enable VLAN receive filtering */
1790                         rctl = er32(RCTL);
1791                         rctl |= E1000_RCTL_VFE;
1792                         rctl &= ~E1000_RCTL_CFIEN;
1793                         ew32(RCTL, rctl);
1794                         e1000_update_mng_vlan(adapter);
1795                 }
1796         } else {
1797                 /* disable VLAN tag insert/strip */
1798                 ctrl = er32(CTRL);
1799                 ctrl &= ~E1000_CTRL_VME;
1800                 ew32(CTRL, ctrl);
1801
1802                 if (adapter->flags & FLAG_HAS_HW_VLAN_FILTER) {
1803                         /* disable VLAN filtering */
1804                         rctl = er32(RCTL);
1805                         rctl &= ~E1000_RCTL_VFE;
1806                         ew32(RCTL, rctl);
1807                         if (adapter->mng_vlan_id !=
1808                             (u16)E1000_MNG_VLAN_NONE) {
1809                                 e1000_vlan_rx_kill_vid(netdev,
1810                                                        adapter->mng_vlan_id);
1811                                 adapter->mng_vlan_id = E1000_MNG_VLAN_NONE;
1812                         }
1813                 }
1814         }
1815
1816         e1000_irq_enable(adapter);
1817 }
1818
1819 static void e1000_restore_vlan(struct e1000_adapter *adapter)
1820 {
1821         u16 vid;
1822
1823         e1000_vlan_rx_register(adapter->netdev, adapter->vlgrp);
1824
1825         if (!adapter->vlgrp)
1826                 return;
1827
1828         for (vid = 0; vid < VLAN_GROUP_ARRAY_LEN; vid++) {
1829                 if (!vlan_group_get_device(adapter->vlgrp, vid))
1830                         continue;
1831                 e1000_vlan_rx_add_vid(adapter->netdev, vid);
1832         }
1833 }
1834
1835 static void e1000_init_manageability(struct e1000_adapter *adapter)
1836 {
1837         struct e1000_hw *hw = &adapter->hw;
1838         u32 manc, manc2h;
1839
1840         if (!(adapter->flags & FLAG_MNG_PT_ENABLED))
1841                 return;
1842
1843         manc = er32(MANC);
1844
1845         /* disable hardware interception of ARP */
1846         manc &= ~(E1000_MANC_ARP_EN);
1847
1848         /* enable receiving management packets to the host. this will probably
1849          * generate destination unreachable messages from the host OS, but
1850          * the packets will be handled on SMBUS */
1851         manc |= E1000_MANC_EN_MNG2HOST;
1852         manc2h = er32(MANC2H);
1853 #define E1000_MNG2HOST_PORT_623 (1 << 5)
1854 #define E1000_MNG2HOST_PORT_664 (1 << 6)
1855         manc2h |= E1000_MNG2HOST_PORT_623;
1856         manc2h |= E1000_MNG2HOST_PORT_664;
1857         ew32(MANC2H, manc2h);
1858         ew32(MANC, manc);
1859 }
1860
1861 /**
1862  * e1000_configure_tx - Configure 8254x Transmit Unit after Reset
1863  * @adapter: board private structure
1864  *
1865  * Configure the Tx unit of the MAC after a reset.
1866  **/
1867 static void e1000_configure_tx(struct e1000_adapter *adapter)
1868 {
1869         struct e1000_hw *hw = &adapter->hw;
1870         struct e1000_ring *tx_ring = adapter->tx_ring;
1871         u64 tdba;
1872         u32 tdlen, tctl, tipg, tarc;
1873         u32 ipgr1, ipgr2;
1874
1875         /* Setup the HW Tx Head and Tail descriptor pointers */
1876         tdba = tx_ring->dma;
1877         tdlen = tx_ring->count * sizeof(struct e1000_tx_desc);
1878         ew32(TDBAL, (tdba & DMA_32BIT_MASK));
1879         ew32(TDBAH, (tdba >> 32));
1880         ew32(TDLEN, tdlen);
1881         ew32(TDH, 0);
1882         ew32(TDT, 0);
1883         tx_ring->head = E1000_TDH;
1884         tx_ring->tail = E1000_TDT;
1885
1886         /* Set the default values for the Tx Inter Packet Gap timer */
1887         tipg = DEFAULT_82543_TIPG_IPGT_COPPER;          /*  8  */
1888         ipgr1 = DEFAULT_82543_TIPG_IPGR1;               /*  8  */
1889         ipgr2 = DEFAULT_82543_TIPG_IPGR2;               /*  6  */
1890
1891         if (adapter->flags & FLAG_TIPG_MEDIUM_FOR_80003ESLAN)
1892                 ipgr2 = DEFAULT_80003ES2LAN_TIPG_IPGR2; /*  7  */
1893
1894         tipg |= ipgr1 << E1000_TIPG_IPGR1_SHIFT;
1895         tipg |= ipgr2 << E1000_TIPG_IPGR2_SHIFT;
1896         ew32(TIPG, tipg);
1897
1898         /* Set the Tx Interrupt Delay register */
1899         ew32(TIDV, adapter->tx_int_delay);
1900         /* tx irq moderation */
1901         ew32(TADV, adapter->tx_abs_int_delay);
1902
1903         /* Program the Transmit Control Register */
1904         tctl = er32(TCTL);
1905         tctl &= ~E1000_TCTL_CT;
1906         tctl |= E1000_TCTL_PSP | E1000_TCTL_RTLC |
1907                 (E1000_COLLISION_THRESHOLD << E1000_CT_SHIFT);
1908
1909         if (adapter->flags & FLAG_TARC_SPEED_MODE_BIT) {
1910                 tarc = er32(TARC0);
1911                 /* set the speed mode bit, we'll clear it if we're not at
1912                  * gigabit link later */
1913 #define SPEED_MODE_BIT (1 << 21)
1914                 tarc |= SPEED_MODE_BIT;
1915                 ew32(TARC0, tarc);
1916         }
1917
1918         /* errata: program both queues to unweighted RR */
1919         if (adapter->flags & FLAG_TARC_SET_BIT_ZERO) {
1920                 tarc = er32(TARC0);
1921                 tarc |= 1;
1922                 ew32(TARC0, tarc);
1923                 tarc = er32(TARC1);
1924                 tarc |= 1;
1925                 ew32(TARC1, tarc);
1926         }
1927
1928         e1000e_config_collision_dist(hw);
1929
1930         /* Setup Transmit Descriptor Settings for eop descriptor */
1931         adapter->txd_cmd = E1000_TXD_CMD_EOP | E1000_TXD_CMD_IFCS;
1932
1933         /* only set IDE if we are delaying interrupts using the timers */
1934         if (adapter->tx_int_delay)
1935                 adapter->txd_cmd |= E1000_TXD_CMD_IDE;
1936
1937         /* enable Report Status bit */
1938         adapter->txd_cmd |= E1000_TXD_CMD_RS;
1939
1940         ew32(TCTL, tctl);
1941
1942         adapter->tx_queue_len = adapter->netdev->tx_queue_len;
1943 }
1944
1945 /**
1946  * e1000_setup_rctl - configure the receive control registers
1947  * @adapter: Board private structure
1948  **/
1949 #define PAGE_USE_COUNT(S) (((S) >> PAGE_SHIFT) + \
1950                            (((S) & (PAGE_SIZE - 1)) ? 1 : 0))
1951 static void e1000_setup_rctl(struct e1000_adapter *adapter)
1952 {
1953         struct e1000_hw *hw = &adapter->hw;
1954         u32 rctl, rfctl;
1955         u32 psrctl = 0;
1956         u32 pages = 0;
1957
1958         /* Program MC offset vector base */
1959         rctl = er32(RCTL);
1960         rctl &= ~(3 << E1000_RCTL_MO_SHIFT);
1961         rctl |= E1000_RCTL_EN | E1000_RCTL_BAM |
1962                 E1000_RCTL_LBM_NO | E1000_RCTL_RDMTS_HALF |
1963                 (adapter->hw.mac.mc_filter_type << E1000_RCTL_MO_SHIFT);
1964
1965         /* Do not Store bad packets */
1966         rctl &= ~E1000_RCTL_SBP;
1967
1968         /* Enable Long Packet receive */
1969         if (adapter->netdev->mtu <= ETH_DATA_LEN)
1970                 rctl &= ~E1000_RCTL_LPE;
1971         else
1972                 rctl |= E1000_RCTL_LPE;
1973
1974         /* Setup buffer sizes */
1975         rctl &= ~E1000_RCTL_SZ_4096;
1976         rctl |= E1000_RCTL_BSEX;
1977         switch (adapter->rx_buffer_len) {
1978         case 256:
1979                 rctl |= E1000_RCTL_SZ_256;
1980                 rctl &= ~E1000_RCTL_BSEX;
1981                 break;
1982         case 512:
1983                 rctl |= E1000_RCTL_SZ_512;
1984                 rctl &= ~E1000_RCTL_BSEX;
1985                 break;
1986         case 1024:
1987                 rctl |= E1000_RCTL_SZ_1024;
1988                 rctl &= ~E1000_RCTL_BSEX;
1989                 break;
1990         case 2048:
1991         default:
1992                 rctl |= E1000_RCTL_SZ_2048;
1993                 rctl &= ~E1000_RCTL_BSEX;
1994                 break;
1995         case 4096:
1996                 rctl |= E1000_RCTL_SZ_4096;
1997                 break;
1998         case 8192:
1999                 rctl |= E1000_RCTL_SZ_8192;
2000                 break;
2001         case 16384:
2002                 rctl |= E1000_RCTL_SZ_16384;
2003                 break;
2004         }
2005
2006         /*
2007          * 82571 and greater support packet-split where the protocol
2008          * header is placed in skb->data and the packet data is
2009          * placed in pages hanging off of skb_shinfo(skb)->nr_frags.
2010          * In the case of a non-split, skb->data is linearly filled,
2011          * followed by the page buffers.  Therefore, skb->data is
2012          * sized to hold the largest protocol header.
2013          *
2014          * allocations using alloc_page take too long for regular MTU
2015          * so only enable packet split for jumbo frames
2016          *
2017          * Using pages when the page size is greater than 16k wastes
2018          * a lot of memory, since we allocate 3 pages at all times
2019          * per packet.
2020          */
2021         adapter->rx_ps_pages = 0;
2022         pages = PAGE_USE_COUNT(adapter->netdev->mtu);
2023         if ((pages <= 3) && (PAGE_SIZE <= 16384) && (rctl & E1000_RCTL_LPE))
2024                 adapter->rx_ps_pages = pages;
2025
2026         if (adapter->rx_ps_pages) {
2027                 /* Configure extra packet-split registers */
2028                 rfctl = er32(RFCTL);
2029                 rfctl |= E1000_RFCTL_EXTEN;
2030                 /* disable packet split support for IPv6 extension headers,
2031                  * because some malformed IPv6 headers can hang the RX */
2032                 rfctl |= (E1000_RFCTL_IPV6_EX_DIS |
2033                           E1000_RFCTL_NEW_IPV6_EXT_DIS);
2034
2035                 ew32(RFCTL, rfctl);
2036
2037                 /* disable the stripping of CRC because it breaks
2038                  * BMC firmware connected over SMBUS */
2039                 rctl |= E1000_RCTL_DTYP_PS /* | E1000_RCTL_SECRC */;
2040
2041                 psrctl |= adapter->rx_ps_bsize0 >>
2042                         E1000_PSRCTL_BSIZE0_SHIFT;
2043
2044                 switch (adapter->rx_ps_pages) {
2045                 case 3:
2046                         psrctl |= PAGE_SIZE <<
2047                                 E1000_PSRCTL_BSIZE3_SHIFT;
2048                 case 2:
2049                         psrctl |= PAGE_SIZE <<
2050                                 E1000_PSRCTL_BSIZE2_SHIFT;
2051                 case 1:
2052                         psrctl |= PAGE_SIZE >>
2053                                 E1000_PSRCTL_BSIZE1_SHIFT;
2054                         break;
2055                 }
2056
2057                 ew32(PSRCTL, psrctl);
2058         }
2059
2060         ew32(RCTL, rctl);
2061 }
2062
2063 /**
2064  * e1000_configure_rx - Configure Receive Unit after Reset
2065  * @adapter: board private structure
2066  *
2067  * Configure the Rx unit of the MAC after a reset.
2068  **/
2069 static void e1000_configure_rx(struct e1000_adapter *adapter)
2070 {
2071         struct e1000_hw *hw = &adapter->hw;
2072         struct e1000_ring *rx_ring = adapter->rx_ring;
2073         u64 rdba;
2074         u32 rdlen, rctl, rxcsum, ctrl_ext;
2075
2076         if (adapter->rx_ps_pages) {
2077                 /* this is a 32 byte descriptor */
2078                 rdlen = rx_ring->count *
2079                         sizeof(union e1000_rx_desc_packet_split);
2080                 adapter->clean_rx = e1000_clean_rx_irq_ps;
2081                 adapter->alloc_rx_buf = e1000_alloc_rx_buffers_ps;
2082         } else if (adapter->netdev->mtu > ETH_FRAME_LEN + VLAN_HLEN + 4) {
2083                 rdlen = rx_ring->count *
2084                         sizeof(struct e1000_rx_desc);
2085                 adapter->clean_rx = e1000_clean_rx_irq_jumbo;
2086                 adapter->alloc_rx_buf = e1000_alloc_rx_buffers_jumbo;
2087         } else {
2088                 rdlen = rx_ring->count *
2089                         sizeof(struct e1000_rx_desc);
2090                 adapter->clean_rx = e1000_clean_rx_irq;
2091                 adapter->alloc_rx_buf = e1000_alloc_rx_buffers;
2092         }
2093
2094         /* disable receives while setting up the descriptors */
2095         rctl = er32(RCTL);
2096         ew32(RCTL, rctl & ~E1000_RCTL_EN);
2097         e1e_flush();
2098         msleep(10);
2099
2100         /* set the Receive Delay Timer Register */
2101         ew32(RDTR, adapter->rx_int_delay);
2102
2103         /* irq moderation */
2104         ew32(RADV, adapter->rx_abs_int_delay);
2105         if (adapter->itr_setting != 0)
2106                 ew32(ITR,
2107                         1000000000 / (adapter->itr * 256));
2108
2109         ctrl_ext = er32(CTRL_EXT);
2110         /* Reset delay timers after every interrupt */
2111         ctrl_ext |= E1000_CTRL_EXT_INT_TIMER_CLR;
2112         /* Auto-Mask interrupts upon ICR access */
2113         ctrl_ext |= E1000_CTRL_EXT_IAME;
2114         ew32(IAM, 0xffffffff);
2115         ew32(CTRL_EXT, ctrl_ext);
2116         e1e_flush();
2117
2118         /* Setup the HW Rx Head and Tail Descriptor Pointers and
2119          * the Base and Length of the Rx Descriptor Ring */
2120         rdba = rx_ring->dma;
2121         ew32(RDBAL, (rdba & DMA_32BIT_MASK));
2122         ew32(RDBAH, (rdba >> 32));
2123         ew32(RDLEN, rdlen);
2124         ew32(RDH, 0);
2125         ew32(RDT, 0);
2126         rx_ring->head = E1000_RDH;
2127         rx_ring->tail = E1000_RDT;
2128
2129         /* Enable Receive Checksum Offload for TCP and UDP */
2130         rxcsum = er32(RXCSUM);
2131         if (adapter->flags & FLAG_RX_CSUM_ENABLED) {
2132                 rxcsum |= E1000_RXCSUM_TUOFL;
2133
2134                 /* IPv4 payload checksum for UDP fragments must be
2135                  * used in conjunction with packet-split. */
2136                 if (adapter->rx_ps_pages)
2137                         rxcsum |= E1000_RXCSUM_IPPCSE;
2138         } else {
2139                 rxcsum &= ~E1000_RXCSUM_TUOFL;
2140                 /* no need to clear IPPCSE as it defaults to 0 */
2141         }
2142         ew32(RXCSUM, rxcsum);
2143
2144         /* Enable early receives on supported devices, only takes effect when
2145          * packet size is equal or larger than the specified value (in 8 byte
2146          * units), e.g. using jumbo frames when setting to E1000_ERT_2048 */
2147         if ((adapter->flags & FLAG_HAS_ERT) &&
2148             (adapter->netdev->mtu > ETH_DATA_LEN))
2149                 ew32(ERT, E1000_ERT_2048);
2150
2151         /* Enable Receives */
2152         ew32(RCTL, rctl);
2153 }
2154
2155 /**
2156  *  e1000_mc_addr_list_update - Update Multicast addresses
2157  *  @hw: pointer to the HW structure
2158  *  @mc_addr_list: array of multicast addresses to program
2159  *  @mc_addr_count: number of multicast addresses to program
2160  *  @rar_used_count: the first RAR register free to program
2161  *  @rar_count: total number of supported Receive Address Registers
2162  *
2163  *  Updates the Receive Address Registers and Multicast Table Array.
2164  *  The caller must have a packed mc_addr_list of multicast addresses.
2165  *  The parameter rar_count will usually be hw->mac.rar_entry_count
2166  *  unless there are workarounds that change this.  Currently no func pointer
2167  *  exists and all implementations are handled in the generic version of this
2168  *  function.
2169  **/
2170 static void e1000_mc_addr_list_update(struct e1000_hw *hw, u8 *mc_addr_list,
2171                                u32 mc_addr_count, u32 rar_used_count,
2172                                u32 rar_count)
2173 {
2174         hw->mac.ops.mc_addr_list_update(hw, mc_addr_list, mc_addr_count,
2175                                         rar_used_count, rar_count);
2176 }
2177
2178 /**
2179  * e1000_set_multi - Multicast and Promiscuous mode set
2180  * @netdev: network interface device structure
2181  *
2182  * The set_multi entry point is called whenever the multicast address
2183  * list or the network interface flags are updated.  This routine is
2184  * responsible for configuring the hardware for proper multicast,
2185  * promiscuous mode, and all-multi behavior.
2186  **/
2187 static void e1000_set_multi(struct net_device *netdev)
2188 {
2189         struct e1000_adapter *adapter = netdev_priv(netdev);
2190         struct e1000_hw *hw = &adapter->hw;
2191         struct e1000_mac_info *mac = &hw->mac;
2192         struct dev_mc_list *mc_ptr;
2193         u8  *mta_list;
2194         u32 rctl;
2195         int i;
2196
2197         /* Check for Promiscuous and All Multicast modes */
2198
2199         rctl = er32(RCTL);
2200
2201         if (netdev->flags & IFF_PROMISC) {
2202                 rctl |= (E1000_RCTL_UPE | E1000_RCTL_MPE);
2203         } else if (netdev->flags & IFF_ALLMULTI) {
2204                 rctl |= E1000_RCTL_MPE;
2205                 rctl &= ~E1000_RCTL_UPE;
2206         } else {
2207                 rctl &= ~(E1000_RCTL_UPE | E1000_RCTL_MPE);
2208         }
2209
2210         ew32(RCTL, rctl);
2211
2212         if (netdev->mc_count) {
2213                 mta_list = kmalloc(netdev->mc_count * 6, GFP_ATOMIC);
2214                 if (!mta_list)
2215                         return;
2216
2217                 /* prepare a packed array of only addresses. */
2218                 mc_ptr = netdev->mc_list;
2219
2220                 for (i = 0; i < netdev->mc_count; i++) {
2221                         if (!mc_ptr)
2222                                 break;
2223                         memcpy(mta_list + (i*ETH_ALEN), mc_ptr->dmi_addr,
2224                                ETH_ALEN);
2225                         mc_ptr = mc_ptr->next;
2226                 }
2227
2228                 e1000_mc_addr_list_update(hw, mta_list, i, 1,
2229                                           mac->rar_entry_count);
2230                 kfree(mta_list);
2231         } else {
2232                 /*
2233                  * if we're called from probe, we might not have
2234                  * anything to do here, so clear out the list
2235                  */
2236                 e1000_mc_addr_list_update(hw, NULL, 0, 1,
2237                                           mac->rar_entry_count);
2238         }
2239 }
2240
2241 /**
2242  * e1000_configure - configure the hardware for RX and TX
2243  * @adapter: private board structure
2244  **/
2245 static void e1000_configure(struct e1000_adapter *adapter)
2246 {
2247         e1000_set_multi(adapter->netdev);
2248
2249         e1000_restore_vlan(adapter);
2250         e1000_init_manageability(adapter);
2251
2252         e1000_configure_tx(adapter);
2253         e1000_setup_rctl(adapter);
2254         e1000_configure_rx(adapter);
2255         adapter->alloc_rx_buf(adapter,
2256                               e1000_desc_unused(adapter->rx_ring));
2257 }
2258
2259 /**
2260  * e1000e_power_up_phy - restore link in case the phy was powered down
2261  * @adapter: address of board private structure
2262  *
2263  * The phy may be powered down to save power and turn off link when the
2264  * driver is unloaded and wake on lan is not enabled (among others)
2265  * *** this routine MUST be followed by a call to e1000e_reset ***
2266  **/
2267 void e1000e_power_up_phy(struct e1000_adapter *adapter)
2268 {
2269         u16 mii_reg = 0;
2270
2271         /* Just clear the power down bit to wake the phy back up */
2272         if (adapter->hw.media_type == e1000_media_type_copper) {
2273                 /* according to the manual, the phy will retain its
2274                  * settings across a power-down/up cycle */
2275                 e1e_rphy(&adapter->hw, PHY_CONTROL, &mii_reg);
2276                 mii_reg &= ~MII_CR_POWER_DOWN;
2277                 e1e_wphy(&adapter->hw, PHY_CONTROL, mii_reg);
2278         }
2279
2280         adapter->hw.mac.ops.setup_link(&adapter->hw);
2281 }
2282
2283 /**
2284  * e1000_power_down_phy - Power down the PHY
2285  *
2286  * Power down the PHY so no link is implied when interface is down
2287  * The PHY cannot be powered down is management or WoL is active
2288  */
2289 static void e1000_power_down_phy(struct e1000_adapter *adapter)
2290 {
2291         struct e1000_hw *hw = &adapter->hw;
2292         u16 mii_reg;
2293
2294         /* WoL is enabled */
2295         if (!adapter->wol)
2296                 return;
2297
2298         /* non-copper PHY? */
2299         if (adapter->hw.media_type != e1000_media_type_copper)
2300                 return;
2301
2302         /* reset is blocked because of a SoL/IDER session */
2303         if (e1000e_check_mng_mode(hw) ||
2304             e1000_check_reset_block(hw))
2305                 return;
2306
2307         /* managebility (AMT) is enabled */
2308         if (er32(MANC) & E1000_MANC_SMBUS_EN)
2309                 return;
2310
2311         /* power down the PHY */
2312         e1e_rphy(hw, PHY_CONTROL, &mii_reg);
2313         mii_reg |= MII_CR_POWER_DOWN;
2314         e1e_wphy(hw, PHY_CONTROL, mii_reg);
2315         mdelay(1);
2316 }
2317
2318 /**
2319  * e1000e_reset - bring the hardware into a known good state
2320  *
2321  * This function boots the hardware and enables some settings that
2322  * require a configuration cycle of the hardware - those cannot be
2323  * set/changed during runtime. After reset the device needs to be
2324  * properly configured for rx, tx etc.
2325  */
2326 void e1000e_reset(struct e1000_adapter *adapter)
2327 {
2328         struct e1000_mac_info *mac = &adapter->hw.mac;
2329         struct e1000_hw *hw = &adapter->hw;
2330         u32 tx_space, min_tx_space, min_rx_space;
2331         u32 pba;
2332         u16 hwm;
2333
2334         ew32(PBA, adapter->pba);
2335
2336         if (mac->max_frame_size > ETH_FRAME_LEN + ETH_FCS_LEN ) {
2337                 /* To maintain wire speed transmits, the Tx FIFO should be
2338                  * large enough to accommodate two full transmit packets,
2339                  * rounded up to the next 1KB and expressed in KB.  Likewise,
2340                  * the Rx FIFO should be large enough to accommodate at least
2341                  * one full receive packet and is similarly rounded up and
2342                  * expressed in KB. */
2343                 pba = er32(PBA);
2344                 /* upper 16 bits has Tx packet buffer allocation size in KB */
2345                 tx_space = pba >> 16;
2346                 /* lower 16 bits has Rx packet buffer allocation size in KB */
2347                 pba &= 0xffff;
2348                 /* the tx fifo also stores 16 bytes of information about the tx
2349                  * but don't include ethernet FCS because hardware appends it */
2350                 min_tx_space = (mac->max_frame_size +
2351                                 sizeof(struct e1000_tx_desc) -
2352                                 ETH_FCS_LEN) * 2;
2353                 min_tx_space = ALIGN(min_tx_space, 1024);
2354                 min_tx_space >>= 10;
2355                 /* software strips receive CRC, so leave room for it */
2356                 min_rx_space = mac->max_frame_size;
2357                 min_rx_space = ALIGN(min_rx_space, 1024);
2358                 min_rx_space >>= 10;
2359
2360                 /* If current Tx allocation is less than the min Tx FIFO size,
2361                  * and the min Tx FIFO size is less than the current Rx FIFO
2362                  * allocation, take space away from current Rx allocation */
2363                 if ((tx_space < min_tx_space) &&
2364                     ((min_tx_space - tx_space) < pba)) {
2365                         pba -= min_tx_space - tx_space;
2366
2367                         /* if short on rx space, rx wins and must trump tx
2368                          * adjustment or use Early Receive if available */
2369                         if ((pba < min_rx_space) &&
2370                             (!(adapter->flags & FLAG_HAS_ERT)))
2371                                 /* ERT enabled in e1000_configure_rx */
2372                                 pba = min_rx_space;
2373                 }
2374
2375                 ew32(PBA, pba);
2376         }
2377
2378
2379         /* flow control settings */
2380         /* The high water mark must be low enough to fit one full frame
2381          * (or the size used for early receive) above it in the Rx FIFO.
2382          * Set it to the lower of:
2383          * - 90% of the Rx FIFO size, and
2384          * - the full Rx FIFO size minus the early receive size (for parts
2385          *   with ERT support assuming ERT set to E1000_ERT_2048), or
2386          * - the full Rx FIFO size minus one full frame */
2387         if (adapter->flags & FLAG_HAS_ERT)
2388                 hwm = min(((adapter->pba << 10) * 9 / 10),
2389                           ((adapter->pba << 10) - (E1000_ERT_2048 << 3)));
2390         else
2391                 hwm = min(((adapter->pba << 10) * 9 / 10),
2392                           ((adapter->pba << 10) - mac->max_frame_size));
2393
2394         mac->fc_high_water = hwm & 0xFFF8; /* 8-byte granularity */
2395         mac->fc_low_water = mac->fc_high_water - 8;
2396
2397         if (adapter->flags & FLAG_DISABLE_FC_PAUSE_TIME)
2398                 mac->fc_pause_time = 0xFFFF;
2399         else
2400                 mac->fc_pause_time = E1000_FC_PAUSE_TIME;
2401         mac->fc = mac->original_fc;
2402
2403         /* Allow time for pending master requests to run */
2404         mac->ops.reset_hw(hw);
2405         ew32(WUC, 0);
2406
2407         if (mac->ops.init_hw(hw))
2408                 ndev_err(adapter->netdev, "Hardware Error\n");
2409
2410         e1000_update_mng_vlan(adapter);
2411
2412         /* Enable h/w to recognize an 802.1Q VLAN Ethernet packet */
2413         ew32(VET, ETH_P_8021Q);
2414
2415         e1000e_reset_adaptive(hw);
2416         e1000_get_phy_info(hw);
2417
2418         if (!(adapter->flags & FLAG_SMART_POWER_DOWN)) {
2419                 u16 phy_data = 0;
2420                 /* speed up time to link by disabling smart power down, ignore
2421                  * the return value of this function because there is nothing
2422                  * different we would do if it failed */
2423                 e1e_rphy(hw, IGP02E1000_PHY_POWER_MGMT, &phy_data);
2424                 phy_data &= ~IGP02E1000_PM_SPD;
2425                 e1e_wphy(hw, IGP02E1000_PHY_POWER_MGMT, phy_data);
2426         }
2427
2428         e1000_release_manageability(adapter);
2429 }
2430
2431 int e1000e_up(struct e1000_adapter *adapter)
2432 {
2433         struct e1000_hw *hw = &adapter->hw;
2434
2435         /* hardware has been reset, we need to reload some things */
2436         e1000_configure(adapter);
2437
2438         clear_bit(__E1000_DOWN, &adapter->state);
2439
2440         napi_enable(&adapter->napi);
2441         e1000_irq_enable(adapter);
2442
2443         /* fire a link change interrupt to start the watchdog */
2444         ew32(ICS, E1000_ICS_LSC);
2445         return 0;
2446 }
2447
2448 void e1000e_down(struct e1000_adapter *adapter)
2449 {
2450         struct net_device *netdev = adapter->netdev;
2451         struct e1000_hw *hw = &adapter->hw;
2452         u32 tctl, rctl;
2453
2454         /* signal that we're down so the interrupt handler does not
2455          * reschedule our watchdog timer */
2456         set_bit(__E1000_DOWN, &adapter->state);
2457
2458         /* disable receives in the hardware */
2459         rctl = er32(RCTL);
2460         ew32(RCTL, rctl & ~E1000_RCTL_EN);
2461         /* flush and sleep below */
2462
2463         netif_stop_queue(netdev);
2464
2465         /* disable transmits in the hardware */
2466         tctl = er32(TCTL);
2467         tctl &= ~E1000_TCTL_EN;
2468         ew32(TCTL, tctl);
2469         /* flush both disables and wait for them to finish */
2470         e1e_flush();
2471         msleep(10);
2472
2473         napi_disable(&adapter->napi);
2474         e1000_irq_disable(adapter);
2475
2476         del_timer_sync(&adapter->watchdog_timer);
2477         del_timer_sync(&adapter->phy_info_timer);
2478
2479         netdev->tx_queue_len = adapter->tx_queue_len;
2480         netif_carrier_off(netdev);
2481         adapter->link_speed = 0;
2482         adapter->link_duplex = 0;
2483
2484         e1000e_reset(adapter);
2485         e1000_clean_tx_ring(adapter);
2486         e1000_clean_rx_ring(adapter);
2487
2488         /*
2489          * TODO: for power management, we could drop the link and
2490          * pci_disable_device here.
2491          */
2492 }
2493
2494 void e1000e_reinit_locked(struct e1000_adapter *adapter)
2495 {
2496         might_sleep();
2497         while (test_and_set_bit(__E1000_RESETTING, &adapter->state))
2498                 msleep(1);
2499         e1000e_down(adapter);
2500         e1000e_up(adapter);
2501         clear_bit(__E1000_RESETTING, &adapter->state);
2502 }
2503
2504 /**
2505  * e1000_sw_init - Initialize general software structures (struct e1000_adapter)
2506  * @adapter: board private structure to initialize
2507  *
2508  * e1000_sw_init initializes the Adapter private data structure.
2509  * Fields are initialized based on PCI device information and
2510  * OS network device settings (MTU size).
2511  **/
2512 static int __devinit e1000_sw_init(struct e1000_adapter *adapter)
2513 {
2514         struct e1000_hw *hw = &adapter->hw;
2515         struct net_device *netdev = adapter->netdev;
2516
2517         adapter->rx_buffer_len = ETH_FRAME_LEN + VLAN_HLEN + ETH_FCS_LEN;
2518         adapter->rx_ps_bsize0 = 128;
2519         hw->mac.max_frame_size = netdev->mtu + ETH_HLEN + ETH_FCS_LEN;
2520         hw->mac.min_frame_size = ETH_ZLEN + ETH_FCS_LEN;
2521
2522         adapter->tx_ring = kzalloc(sizeof(struct e1000_ring), GFP_KERNEL);
2523         if (!adapter->tx_ring)
2524                 goto err;
2525
2526         adapter->rx_ring = kzalloc(sizeof(struct e1000_ring), GFP_KERNEL);
2527         if (!adapter->rx_ring)
2528                 goto err;
2529
2530         spin_lock_init(&adapter->tx_queue_lock);
2531
2532         /* Explicitly disable IRQ since the NIC can be in any state. */
2533         atomic_set(&adapter->irq_sem, 0);
2534         e1000_irq_disable(adapter);
2535
2536         spin_lock_init(&adapter->stats_lock);
2537
2538         set_bit(__E1000_DOWN, &adapter->state);
2539         return 0;
2540
2541 err:
2542         ndev_err(netdev, "Unable to allocate memory for queues\n");
2543         kfree(adapter->rx_ring);
2544         kfree(adapter->tx_ring);
2545         return -ENOMEM;
2546 }
2547
2548 /**
2549  * e1000_open - Called when a network interface is made active
2550  * @netdev: network interface device structure
2551  *
2552  * Returns 0 on success, negative value on failure
2553  *
2554  * The open entry point is called when a network interface is made
2555  * active by the system (IFF_UP).  At this point all resources needed
2556  * for transmit and receive operations are allocated, the interrupt
2557  * handler is registered with the OS, the watchdog timer is started,
2558  * and the stack is notified that the interface is ready.
2559  **/
2560 static int e1000_open(struct net_device *netdev)
2561 {
2562         struct e1000_adapter *adapter = netdev_priv(netdev);
2563         struct e1000_hw *hw = &adapter->hw;
2564         int err;
2565
2566         /* disallow open during test */
2567         if (test_bit(__E1000_TESTING, &adapter->state))
2568                 return -EBUSY;
2569
2570         /* allocate transmit descriptors */
2571         err = e1000e_setup_tx_resources(adapter);
2572         if (err)
2573                 goto err_setup_tx;
2574
2575         /* allocate receive descriptors */
2576         err = e1000e_setup_rx_resources(adapter);
2577         if (err)
2578                 goto err_setup_rx;
2579
2580         e1000e_power_up_phy(adapter);
2581
2582         adapter->mng_vlan_id = E1000_MNG_VLAN_NONE;
2583         if ((adapter->hw.mng_cookie.status &
2584              E1000_MNG_DHCP_COOKIE_STATUS_VLAN))
2585                 e1000_update_mng_vlan(adapter);
2586
2587         /* If AMT is enabled, let the firmware know that the network
2588          * interface is now open */
2589         if ((adapter->flags & FLAG_HAS_AMT) &&
2590             e1000e_check_mng_mode(&adapter->hw))
2591                 e1000_get_hw_control(adapter);
2592
2593         /* before we allocate an interrupt, we must be ready to handle it.
2594          * Setting DEBUG_SHIRQ in the kernel makes it fire an interrupt
2595          * as soon as we call pci_request_irq, so we have to setup our
2596          * clean_rx handler before we do so.  */
2597         e1000_configure(adapter);
2598
2599         err = e1000_request_irq(adapter);
2600         if (err)
2601                 goto err_req_irq;
2602
2603         /* From here on the code is the same as e1000e_up() */
2604         clear_bit(__E1000_DOWN, &adapter->state);
2605
2606         napi_enable(&adapter->napi);
2607
2608         e1000_irq_enable(adapter);
2609
2610         /* fire a link status change interrupt to start the watchdog */
2611         ew32(ICS, E1000_ICS_LSC);
2612
2613         return 0;
2614
2615 err_req_irq:
2616         e1000_release_hw_control(adapter);
2617         e1000_power_down_phy(adapter);
2618         e1000e_free_rx_resources(adapter);
2619 err_setup_rx:
2620         e1000e_free_tx_resources(adapter);
2621 err_setup_tx:
2622         e1000e_reset(adapter);
2623
2624         return err;
2625 }
2626
2627 /**
2628  * e1000_close - Disables a network interface
2629  * @netdev: network interface device structure
2630  *
2631  * Returns 0, this is not allowed to fail
2632  *
2633  * The close entry point is called when an interface is de-activated
2634  * by the OS.  The hardware is still under the drivers control, but
2635  * needs to be disabled.  A global MAC reset is issued to stop the
2636  * hardware, and all transmit and receive resources are freed.
2637  **/
2638 static int e1000_close(struct net_device *netdev)
2639 {
2640         struct e1000_adapter *adapter = netdev_priv(netdev);
2641
2642         WARN_ON(test_bit(__E1000_RESETTING, &adapter->state));
2643         e1000e_down(adapter);
2644         e1000_power_down_phy(adapter);
2645         e1000_free_irq(adapter);
2646
2647         e1000e_free_tx_resources(adapter);
2648         e1000e_free_rx_resources(adapter);
2649
2650         /* kill manageability vlan ID if supported, but not if a vlan with
2651          * the same ID is registered on the host OS (let 8021q kill it) */
2652         if ((adapter->hw.mng_cookie.status &
2653                           E1000_MNG_DHCP_COOKIE_STATUS_VLAN) &&
2654              !(adapter->vlgrp &&
2655                vlan_group_get_device(adapter->vlgrp, adapter->mng_vlan_id)))
2656                 e1000_vlan_rx_kill_vid(netdev, adapter->mng_vlan_id);
2657
2658         /* If AMT is enabled, let the firmware know that the network
2659          * interface is now closed */
2660         if ((adapter->flags & FLAG_HAS_AMT) &&
2661             e1000e_check_mng_mode(&adapter->hw))
2662                 e1000_release_hw_control(adapter);
2663
2664         return 0;
2665 }
2666 /**
2667  * e1000_set_mac - Change the Ethernet Address of the NIC
2668  * @netdev: network interface device structure
2669  * @p: pointer to an address structure
2670  *
2671  * Returns 0 on success, negative on failure
2672  **/
2673 static int e1000_set_mac(struct net_device *netdev, void *p)
2674 {
2675         struct e1000_adapter *adapter = netdev_priv(netdev);
2676         struct sockaddr *addr = p;
2677
2678         if (!is_valid_ether_addr(addr->sa_data))
2679                 return -EADDRNOTAVAIL;
2680
2681         memcpy(netdev->dev_addr, addr->sa_data, netdev->addr_len);
2682         memcpy(adapter->hw.mac.addr, addr->sa_data, netdev->addr_len);
2683
2684         e1000e_rar_set(&adapter->hw, adapter->hw.mac.addr, 0);
2685
2686         if (adapter->flags & FLAG_RESET_OVERWRITES_LAA) {
2687                 /* activate the work around */
2688                 e1000e_set_laa_state_82571(&adapter->hw, 1);
2689
2690                 /* Hold a copy of the LAA in RAR[14] This is done so that
2691                  * between the time RAR[0] gets clobbered  and the time it
2692                  * gets fixed (in e1000_watchdog), the actual LAA is in one
2693                  * of the RARs and no incoming packets directed to this port
2694                  * are dropped. Eventually the LAA will be in RAR[0] and
2695                  * RAR[14] */
2696                 e1000e_rar_set(&adapter->hw,
2697                               adapter->hw.mac.addr,
2698                               adapter->hw.mac.rar_entry_count - 1);
2699         }
2700
2701         return 0;
2702 }
2703
2704 /* Need to wait a few seconds after link up to get diagnostic information from
2705  * the phy */
2706 static void e1000_update_phy_info(unsigned long data)
2707 {
2708         struct e1000_adapter *adapter = (struct e1000_adapter *) data;
2709         e1000_get_phy_info(&adapter->hw);
2710 }
2711
2712 /**
2713  * e1000e_update_stats - Update the board statistics counters
2714  * @adapter: board private structure
2715  **/
2716 void e1000e_update_stats(struct e1000_adapter *adapter)
2717 {
2718         struct e1000_hw *hw = &adapter->hw;
2719         struct pci_dev *pdev = adapter->pdev;
2720         unsigned long irq_flags;
2721         u16 phy_tmp;
2722
2723 #define PHY_IDLE_ERROR_COUNT_MASK 0x00FF
2724
2725         /*
2726          * Prevent stats update while adapter is being reset, or if the pci
2727          * connection is down.
2728          */
2729         if (adapter->link_speed == 0)
2730                 return;
2731         if (pci_channel_offline(pdev))
2732                 return;
2733
2734         spin_lock_irqsave(&adapter->stats_lock, irq_flags);
2735
2736         /* these counters are modified from e1000_adjust_tbi_stats,
2737          * called from the interrupt context, so they must only
2738          * be written while holding adapter->stats_lock
2739          */
2740
2741         adapter->stats.crcerrs += er32(CRCERRS);
2742         adapter->stats.gprc += er32(GPRC);
2743         adapter->stats.gorcl += er32(GORCL);
2744         adapter->stats.gorch += er32(GORCH);
2745         adapter->stats.bprc += er32(BPRC);
2746         adapter->stats.mprc += er32(MPRC);
2747         adapter->stats.roc += er32(ROC);
2748
2749         if (adapter->flags & FLAG_HAS_STATS_PTC_PRC) {
2750                 adapter->stats.prc64 += er32(PRC64);
2751                 adapter->stats.prc127 += er32(PRC127);
2752                 adapter->stats.prc255 += er32(PRC255);
2753                 adapter->stats.prc511 += er32(PRC511);
2754                 adapter->stats.prc1023 += er32(PRC1023);
2755                 adapter->stats.prc1522 += er32(PRC1522);
2756                 adapter->stats.symerrs += er32(SYMERRS);
2757                 adapter->stats.sec += er32(SEC);
2758         }
2759
2760         adapter->stats.mpc += er32(MPC);
2761         adapter->stats.scc += er32(SCC);
2762         adapter->stats.ecol += er32(ECOL);
2763         adapter->stats.mcc += er32(MCC);
2764         adapter->stats.latecol += er32(LATECOL);
2765         adapter->stats.dc += er32(DC);
2766         adapter->stats.rlec += er32(RLEC);
2767         adapter->stats.xonrxc += er32(XONRXC);
2768         adapter->stats.xontxc += er32(XONTXC);
2769         adapter->stats.xoffrxc += er32(XOFFRXC);
2770         adapter->stats.xofftxc += er32(XOFFTXC);
2771         adapter->stats.fcruc += er32(FCRUC);
2772         adapter->stats.gptc += er32(GPTC);
2773         adapter->stats.gotcl += er32(GOTCL);
2774         adapter->stats.gotch += er32(GOTCH);
2775         adapter->stats.rnbc += er32(RNBC);
2776         adapter->stats.ruc += er32(RUC);
2777         adapter->stats.rfc += er32(RFC);
2778         adapter->stats.rjc += er32(RJC);
2779         adapter->stats.torl += er32(TORL);
2780         adapter->stats.torh += er32(TORH);
2781         adapter->stats.totl += er32(TOTL);
2782         adapter->stats.toth += er32(TOTH);
2783         adapter->stats.tpr += er32(TPR);
2784
2785         if (adapter->flags & FLAG_HAS_STATS_PTC_PRC) {
2786                 adapter->stats.ptc64 += er32(PTC64);
2787                 adapter->stats.ptc127 += er32(PTC127);
2788                 adapter->stats.ptc255 += er32(PTC255);
2789                 adapter->stats.ptc511 += er32(PTC511);
2790                 adapter->stats.ptc1023 += er32(PTC1023);
2791                 adapter->stats.ptc1522 += er32(PTC1522);
2792         }
2793
2794         adapter->stats.mptc += er32(MPTC);
2795         adapter->stats.bptc += er32(BPTC);
2796
2797         /* used for adaptive IFS */
2798
2799         hw->mac.tx_packet_delta = er32(TPT);
2800         adapter->stats.tpt += hw->mac.tx_packet_delta;
2801         hw->mac.collision_delta = er32(COLC);
2802         adapter->stats.colc += hw->mac.collision_delta;
2803
2804         adapter->stats.algnerrc += er32(ALGNERRC);
2805         adapter->stats.rxerrc += er32(RXERRC);
2806         adapter->stats.tncrs += er32(TNCRS);
2807         adapter->stats.cexterr += er32(CEXTERR);
2808         adapter->stats.tsctc += er32(TSCTC);
2809         adapter->stats.tsctfc += er32(TSCTFC);
2810
2811         adapter->stats.iac += er32(IAC);
2812
2813         if (adapter->flags & FLAG_HAS_STATS_ICR_ICT) {
2814                 adapter->stats.icrxoc += er32(ICRXOC);
2815                 adapter->stats.icrxptc += er32(ICRXPTC);
2816                 adapter->stats.icrxatc += er32(ICRXATC);
2817                 adapter->stats.ictxptc += er32(ICTXPTC);
2818                 adapter->stats.ictxatc += er32(ICTXATC);
2819                 adapter->stats.ictxqec += er32(ICTXQEC);
2820                 adapter->stats.ictxqmtc += er32(ICTXQMTC);
2821                 adapter->stats.icrxdmtc += er32(ICRXDMTC);
2822         }
2823
2824         /* Fill out the OS statistics structure */
2825         adapter->net_stats.rx_packets = adapter->stats.gprc;
2826         adapter->net_stats.tx_packets = adapter->stats.gptc;
2827         adapter->net_stats.rx_bytes = adapter->stats.gorcl;
2828         adapter->net_stats.tx_bytes = adapter->stats.gotcl;
2829         adapter->net_stats.multicast = adapter->stats.mprc;
2830         adapter->net_stats.collisions = adapter->stats.colc;
2831
2832         /* Rx Errors */
2833
2834         /* RLEC on some newer hardware can be incorrect so build
2835         * our own version based on RUC and ROC */
2836         adapter->net_stats.rx_errors = adapter->stats.rxerrc +
2837                 adapter->stats.crcerrs + adapter->stats.algnerrc +
2838                 adapter->stats.ruc + adapter->stats.roc +
2839                 adapter->stats.cexterr;
2840         adapter->net_stats.rx_length_errors = adapter->stats.ruc +
2841                                               adapter->stats.roc;
2842         adapter->net_stats.rx_crc_errors = adapter->stats.crcerrs;
2843         adapter->net_stats.rx_frame_errors = adapter->stats.algnerrc;
2844         adapter->net_stats.rx_missed_errors = adapter->stats.mpc;
2845
2846         /* Tx Errors */
2847         adapter->net_stats.tx_errors = adapter->stats.ecol +
2848                                        adapter->stats.latecol;
2849         adapter->net_stats.tx_aborted_errors = adapter->stats.ecol;
2850         adapter->net_stats.tx_window_errors = adapter->stats.latecol;
2851         adapter->net_stats.tx_carrier_errors = adapter->stats.tncrs;
2852
2853         /* Tx Dropped needs to be maintained elsewhere */
2854
2855         /* Phy Stats */
2856         if (hw->media_type == e1000_media_type_copper) {
2857                 if ((adapter->link_speed == SPEED_1000) &&
2858                    (!e1e_rphy(hw, PHY_1000T_STATUS, &phy_tmp))) {
2859                         phy_tmp &= PHY_IDLE_ERROR_COUNT_MASK;
2860                         adapter->phy_stats.idle_errors += phy_tmp;
2861                 }
2862         }
2863
2864         /* Management Stats */
2865         adapter->stats.mgptc += er32(MGTPTC);
2866         adapter->stats.mgprc += er32(MGTPRC);
2867         adapter->stats.mgpdc += er32(MGTPDC);
2868
2869         spin_unlock_irqrestore(&adapter->stats_lock, irq_flags);
2870 }
2871
2872 static void e1000_print_link_info(struct e1000_adapter *adapter)
2873 {
2874         struct net_device *netdev = adapter->netdev;
2875         struct e1000_hw *hw = &adapter->hw;
2876         u32 ctrl = er32(CTRL);
2877
2878         ndev_info(netdev,
2879                 "Link is Up %d Mbps %s, Flow Control: %s\n",
2880                 adapter->link_speed,
2881                 (adapter->link_duplex == FULL_DUPLEX) ?
2882                                 "Full Duplex" : "Half Duplex",
2883                 ((ctrl & E1000_CTRL_TFCE) && (ctrl & E1000_CTRL_RFCE)) ?
2884                                 "RX/TX" :
2885                 ((ctrl & E1000_CTRL_RFCE) ? "RX" :
2886                 ((ctrl & E1000_CTRL_TFCE) ? "TX" : "None" )));
2887 }
2888
2889 /**
2890  * e1000_watchdog - Timer Call-back
2891  * @data: pointer to adapter cast into an unsigned long
2892  **/
2893 static void e1000_watchdog(unsigned long data)
2894 {
2895         struct e1000_adapter *adapter = (struct e1000_adapter *) data;
2896
2897         /* Do the rest outside of interrupt context */
2898         schedule_work(&adapter->watchdog_task);
2899
2900         /* TODO: make this use queue_delayed_work() */
2901 }
2902
2903 static void e1000_watchdog_task(struct work_struct *work)
2904 {
2905         struct e1000_adapter *adapter = container_of(work,
2906                                         struct e1000_adapter, watchdog_task);
2907
2908         struct net_device *netdev = adapter->netdev;
2909         struct e1000_mac_info *mac = &adapter->hw.mac;
2910         struct e1000_ring *tx_ring = adapter->tx_ring;
2911         struct e1000_hw *hw = &adapter->hw;
2912         u32 link, tctl;
2913         s32 ret_val;
2914         int tx_pending = 0;
2915
2916         if ((netif_carrier_ok(netdev)) &&
2917             (er32(STATUS) & E1000_STATUS_LU))
2918                 goto link_up;
2919
2920         ret_val = mac->ops.check_for_link(hw);
2921         if ((ret_val == E1000_ERR_PHY) &&
2922             (adapter->hw.phy.type == e1000_phy_igp_3) &&
2923             (er32(CTRL) &
2924              E1000_PHY_CTRL_GBE_DISABLE)) {
2925                 /* See e1000_kmrn_lock_loss_workaround_ich8lan() */
2926                 ndev_info(netdev,
2927                         "Gigabit has been disabled, downgrading speed\n");
2928         }
2929
2930         if ((e1000e_enable_tx_pkt_filtering(hw)) &&
2931             (adapter->mng_vlan_id != adapter->hw.mng_cookie.vlan_id))
2932                 e1000_update_mng_vlan(adapter);
2933
2934         if ((adapter->hw.media_type == e1000_media_type_internal_serdes) &&
2935            !(er32(TXCW) & E1000_TXCW_ANE))
2936                 link = adapter->hw.mac.serdes_has_link;
2937         else
2938                 link = er32(STATUS) & E1000_STATUS_LU;
2939
2940         if (link) {
2941                 if (!netif_carrier_ok(netdev)) {
2942                         bool txb2b = 1;
2943                         mac->ops.get_link_up_info(&adapter->hw,
2944                                                    &adapter->link_speed,
2945                                                    &adapter->link_duplex);
2946                         e1000_print_link_info(adapter);
2947                         /* tweak tx_queue_len according to speed/duplex
2948                          * and adjust the timeout factor */
2949                         netdev->tx_queue_len = adapter->tx_queue_len;
2950                         adapter->tx_timeout_factor = 1;
2951                         switch (adapter->link_speed) {
2952                         case SPEED_10:
2953                                 txb2b = 0;
2954                                 netdev->tx_queue_len = 10;
2955                                 adapter->tx_timeout_factor = 14;
2956                                 break;
2957                         case SPEED_100:
2958                                 txb2b = 0;
2959                                 netdev->tx_queue_len = 100;
2960                                 /* maybe add some timeout factor ? */
2961                                 break;
2962                         }
2963
2964                         /* workaround: re-program speed mode bit after
2965                          * link-up event */
2966                         if ((adapter->flags & FLAG_TARC_SPEED_MODE_BIT) &&
2967                             !txb2b) {
2968                                 u32 tarc0;
2969                                 tarc0 = er32(TARC0);
2970                                 tarc0 &= ~SPEED_MODE_BIT;
2971                                 ew32(TARC0, tarc0);
2972                         }
2973
2974                         /* disable TSO for pcie and 10/100 speeds, to avoid
2975                          * some hardware issues */
2976                         if (!(adapter->flags & FLAG_TSO_FORCE)) {
2977                                 switch (adapter->link_speed) {
2978                                 case SPEED_10:
2979                                 case SPEED_100:
2980                                         ndev_info(netdev,
2981                                         "10/100 speed: disabling TSO\n");
2982                                         netdev->features &= ~NETIF_F_TSO;
2983                                         netdev->features &= ~NETIF_F_TSO6;
2984                                         break;
2985                                 case SPEED_1000:
2986                                         netdev->features |= NETIF_F_TSO;
2987                                         netdev->features |= NETIF_F_TSO6;
2988                                         break;
2989                                 default:
2990                                         /* oops */
2991                                         break;
2992                                 }
2993                         }
2994
2995                         /* enable transmits in the hardware, need to do this
2996                          * after setting TARC0 */
2997                         tctl = er32(TCTL);
2998                         tctl |= E1000_TCTL_EN;
2999                         ew32(TCTL, tctl);
3000
3001                         netif_carrier_on(netdev);
3002                         netif_wake_queue(netdev);
3003
3004                         if (!test_bit(__E1000_DOWN, &adapter->state))
3005                                 mod_timer(&adapter->phy_info_timer,
3006                                           round_jiffies(jiffies + 2 * HZ));
3007                 } else {
3008                         /* make sure the receive unit is started */
3009                         if (adapter->flags & FLAG_RX_NEEDS_RESTART) {
3010                                 u32 rctl = er32(RCTL);
3011                                 ew32(RCTL, rctl |
3012                                                 E1000_RCTL_EN);
3013                         }
3014                 }
3015         } else {
3016                 if (netif_carrier_ok(netdev)) {
3017                         adapter->link_speed = 0;
3018                         adapter->link_duplex = 0;
3019                         ndev_info(netdev, "Link is Down\n");
3020                         netif_carrier_off(netdev);
3021                         netif_stop_queue(netdev);
3022                         if (!test_bit(__E1000_DOWN, &adapter->state))
3023                                 mod_timer(&adapter->phy_info_timer,
3024                                           round_jiffies(jiffies + 2 * HZ));
3025
3026                         if (adapter->flags & FLAG_RX_NEEDS_RESTART)
3027                                 schedule_work(&adapter->reset_task);
3028                 }
3029         }
3030
3031 link_up:
3032         e1000e_update_stats(adapter);
3033
3034         mac->tx_packet_delta = adapter->stats.tpt - adapter->tpt_old;
3035         adapter->tpt_old = adapter->stats.tpt;
3036         mac->collision_delta = adapter->stats.colc - adapter->colc_old;
3037         adapter->colc_old = adapter->stats.colc;
3038
3039         adapter->gorcl = adapter->stats.gorcl - adapter->gorcl_old;
3040         adapter->gorcl_old = adapter->stats.gorcl;
3041         adapter->gotcl = adapter->stats.gotcl - adapter->gotcl_old;
3042         adapter->gotcl_old = adapter->stats.gotcl;
3043
3044         e1000e_update_adaptive(&adapter->hw);
3045
3046         if (!netif_carrier_ok(netdev)) {
3047                 tx_pending = (e1000_desc_unused(tx_ring) + 1 <
3048                                tx_ring->count);
3049                 if (tx_pending) {
3050                         /* We've lost link, so the controller stops DMA,
3051                          * but we've got queued Tx work that's never going
3052                          * to get done, so reset controller to flush Tx.
3053                          * (Do the reset outside of interrupt context). */
3054                         adapter->tx_timeout_count++;
3055                         schedule_work(&adapter->reset_task);
3056                 }
3057         }
3058
3059         /* Cause software interrupt to ensure rx ring is cleaned */
3060         ew32(ICS, E1000_ICS_RXDMT0);
3061
3062         /* Force detection of hung controller every watchdog period */
3063         adapter->detect_tx_hung = 1;
3064
3065         /* With 82571 controllers, LAA may be overwritten due to controller
3066          * reset from the other port. Set the appropriate LAA in RAR[0] */
3067         if (e1000e_get_laa_state_82571(hw))
3068                 e1000e_rar_set(hw, adapter->hw.mac.addr, 0);
3069
3070         /* Reset the timer */
3071         if (!test_bit(__E1000_DOWN, &adapter->state))
3072                 mod_timer(&adapter->watchdog_timer,
3073                           round_jiffies(jiffies + 2 * HZ));
3074 }
3075
3076 #define E1000_TX_FLAGS_CSUM             0x00000001
3077 #define E1000_TX_FLAGS_VLAN             0x00000002
3078 #define E1000_TX_FLAGS_TSO              0x00000004
3079 #define E1000_TX_FLAGS_IPV4             0x00000008
3080 #define E1000_TX_FLAGS_VLAN_MASK        0xffff0000
3081 #define E1000_TX_FLAGS_VLAN_SHIFT       16
3082
3083 static int e1000_tso(struct e1000_adapter *adapter,
3084                      struct sk_buff *skb)
3085 {
3086         struct e1000_ring *tx_ring = adapter->tx_ring;
3087         struct e1000_context_desc *context_desc;
3088         struct e1000_buffer *buffer_info;
3089         unsigned int i;
3090         u32 cmd_length = 0;
3091         u16 ipcse = 0, tucse, mss;
3092         u8 ipcss, ipcso, tucss, tucso, hdr_len;
3093         int err;
3094
3095         if (skb_is_gso(skb)) {
3096                 if (skb_header_cloned(skb)) {
3097                         err = pskb_expand_head(skb, 0, 0, GFP_ATOMIC);
3098                         if (err)
3099                                 return err;
3100                 }
3101
3102                 hdr_len = skb_transport_offset(skb) + tcp_hdrlen(skb);
3103                 mss = skb_shinfo(skb)->gso_size;
3104                 if (skb->protocol == htons(ETH_P_IP)) {
3105                         struct iphdr *iph = ip_hdr(skb);
3106                         iph->tot_len = 0;
3107                         iph->check = 0;
3108                         tcp_hdr(skb)->check = ~csum_tcpudp_magic(iph->saddr,
3109                                                                  iph->daddr, 0,
3110                                                                  IPPROTO_TCP,
3111                                                                  0);
3112                         cmd_length = E1000_TXD_CMD_IP;
3113                         ipcse = skb_transport_offset(skb) - 1;
3114                 } else if (skb_shinfo(skb)->gso_type == SKB_GSO_TCPV6) {
3115                         ipv6_hdr(skb)->payload_len = 0;
3116                         tcp_hdr(skb)->check =
3117                                 ~csum_ipv6_magic(&ipv6_hdr(skb)->saddr,
3118                                                  &ipv6_hdr(skb)->daddr,
3119                                                  0, IPPROTO_TCP, 0);
3120                         ipcse = 0;
3121                 }
3122                 ipcss = skb_network_offset(skb);
3123                 ipcso = (void *)&(ip_hdr(skb)->check) - (void *)skb->data;
3124                 tucss = skb_transport_offset(skb);
3125                 tucso = (void *)&(tcp_hdr(skb)->check) - (void *)skb->data;
3126                 tucse = 0;
3127
3128                 cmd_length |= (E1000_TXD_CMD_DEXT | E1000_TXD_CMD_TSE |
3129                                E1000_TXD_CMD_TCP | (skb->len - (hdr_len)));
3130
3131                 i = tx_ring->next_to_use;
3132                 context_desc = E1000_CONTEXT_DESC(*tx_ring, i);
3133                 buffer_info = &tx_ring->buffer_info[i];
3134
3135                 context_desc->lower_setup.ip_fields.ipcss  = ipcss;
3136                 context_desc->lower_setup.ip_fields.ipcso  = ipcso;
3137                 context_desc->lower_setup.ip_fields.ipcse  = cpu_to_le16(ipcse);
3138                 context_desc->upper_setup.tcp_fields.tucss = tucss;
3139                 context_desc->upper_setup.tcp_fields.tucso = tucso;
3140                 context_desc->upper_setup.tcp_fields.tucse = cpu_to_le16(tucse);
3141                 context_desc->tcp_seg_setup.fields.mss     = cpu_to_le16(mss);
3142                 context_desc->tcp_seg_setup.fields.hdr_len = hdr_len;
3143                 context_desc->cmd_and_length = cpu_to_le32(cmd_length);
3144
3145                 buffer_info->time_stamp = jiffies;
3146                 buffer_info->next_to_watch = i;
3147
3148                 i++;
3149                 if (i == tx_ring->count)
3150                         i = 0;
3151                 tx_ring->next_to_use = i;
3152
3153                 return 1;
3154         }
3155
3156         return 0;
3157 }
3158
3159 static bool e1000_tx_csum(struct e1000_adapter *adapter, struct sk_buff *skb)
3160 {
3161         struct e1000_ring *tx_ring = adapter->tx_ring;
3162         struct e1000_context_desc *context_desc;
3163         struct e1000_buffer *buffer_info;
3164         unsigned int i;
3165         u8 css;
3166
3167         if (skb->ip_summed == CHECKSUM_PARTIAL) {
3168                 css = skb_transport_offset(skb);
3169
3170                 i = tx_ring->next_to_use;
3171                 buffer_info = &tx_ring->buffer_info[i];
3172                 context_desc = E1000_CONTEXT_DESC(*tx_ring, i);
3173
3174                 context_desc->lower_setup.ip_config = 0;
3175                 context_desc->upper_setup.tcp_fields.tucss = css;
3176                 context_desc->upper_setup.tcp_fields.tucso =
3177                                         css + skb->csum_offset;
3178                 context_desc->upper_setup.tcp_fields.tucse = 0;
3179                 context_desc->tcp_seg_setup.data = 0;
3180                 context_desc->cmd_and_length = cpu_to_le32(E1000_TXD_CMD_DEXT);
3181
3182                 buffer_info->time_stamp = jiffies;
3183                 buffer_info->next_to_watch = i;
3184
3185                 i++;
3186                 if (i == tx_ring->count)
3187                         i = 0;
3188                 tx_ring->next_to_use = i;
3189
3190                 return 1;
3191         }
3192
3193         return 0;
3194 }
3195
3196 #define E1000_MAX_PER_TXD       8192
3197 #define E1000_MAX_TXD_PWR       12
3198
3199 static int e1000_tx_map(struct e1000_adapter *adapter,
3200                         struct sk_buff *skb, unsigned int first,
3201                         unsigned int max_per_txd, unsigned int nr_frags,
3202                         unsigned int mss)
3203 {
3204         struct e1000_ring *tx_ring = adapter->tx_ring;
3205         struct e1000_buffer *buffer_info;
3206         unsigned int len = skb->len - skb->data_len;
3207         unsigned int offset = 0, size, count = 0, i;
3208         unsigned int f;
3209
3210         i = tx_ring->next_to_use;
3211
3212         while (len) {
3213                 buffer_info = &tx_ring->buffer_info[i];
3214                 size = min(len, max_per_txd);
3215
3216                 /* Workaround for premature desc write-backs
3217                  * in TSO mode.  Append 4-byte sentinel desc */
3218                 if (mss && !nr_frags && size == len && size > 8)
3219                         size -= 4;
3220
3221                 buffer_info->length = size;
3222                 /* set time_stamp *before* dma to help avoid a possible race */
3223                 buffer_info->time_stamp = jiffies;
3224                 buffer_info->dma =
3225                         pci_map_single(adapter->pdev,
3226                                 skb->data + offset,
3227                                 size,
3228                                 PCI_DMA_TODEVICE);
3229                 if (pci_dma_mapping_error(buffer_info->dma)) {
3230                         dev_err(&adapter->pdev->dev, "TX DMA map failed\n");
3231                         adapter->tx_dma_failed++;
3232                         return -1;
3233                 }
3234                 buffer_info->next_to_watch = i;
3235
3236                 len -= size;
3237                 offset += size;
3238                 count++;
3239                 i++;
3240                 if (i == tx_ring->count)
3241                         i = 0;
3242         }
3243
3244         for (f = 0; f < nr_frags; f++) {
3245                 struct skb_frag_struct *frag;
3246
3247                 frag = &skb_shinfo(skb)->frags[f];
3248                 len = frag->size;
3249                 offset = frag->page_offset;
3250
3251                 while (len) {
3252                         buffer_info = &tx_ring->buffer_info[i];
3253                         size = min(len, max_per_txd);
3254                         /* Workaround for premature desc write-backs
3255                          * in TSO mode.  Append 4-byte sentinel desc */
3256                         if (mss && f == (nr_frags-1) && size == len && size > 8)
3257                                 size -= 4;
3258
3259                         buffer_info->length = size;
3260                         buffer_info->time_stamp = jiffies;
3261                         buffer_info->dma =
3262                                 pci_map_page(adapter->pdev,
3263                                         frag->page,
3264                                         offset,
3265                                         size,
3266                                         PCI_DMA_TODEVICE);
3267                         if (pci_dma_mapping_error(buffer_info->dma)) {
3268                                 dev_err(&adapter->pdev->dev,
3269                                         "TX DMA page map failed\n");
3270                                 adapter->tx_dma_failed++;
3271                                 return -1;
3272                         }
3273
3274                         buffer_info->next_to_watch = i;
3275
3276                         len -= size;
3277                         offset += size;
3278                         count++;
3279
3280                         i++;
3281                         if (i == tx_ring->count)
3282                                 i = 0;
3283                 }
3284         }
3285
3286         if (i == 0)
3287                 i = tx_ring->count - 1;
3288         else
3289                 i--;
3290
3291         tx_ring->buffer_info[i].skb = skb;
3292         tx_ring->buffer_info[first].next_to_watch = i;
3293
3294         return count;
3295 }
3296
3297 static void e1000_tx_queue(struct e1000_adapter *adapter,
3298                            int tx_flags, int count)
3299 {
3300         struct e1000_ring *tx_ring = adapter->tx_ring;
3301         struct e1000_tx_desc *tx_desc = NULL;
3302         struct e1000_buffer *buffer_info;
3303         u32 txd_upper = 0, txd_lower = E1000_TXD_CMD_IFCS;
3304         unsigned int i;
3305
3306         if (tx_flags & E1000_TX_FLAGS_TSO) {
3307                 txd_lower |= E1000_TXD_CMD_DEXT | E1000_TXD_DTYP_D |
3308                              E1000_TXD_CMD_TSE;
3309                 txd_upper |= E1000_TXD_POPTS_TXSM << 8;
3310
3311                 if (tx_flags & E1000_TX_FLAGS_IPV4)
3312                         txd_upper |= E1000_TXD_POPTS_IXSM << 8;
3313         }
3314
3315         if (tx_flags & E1000_TX_FLAGS_CSUM) {
3316                 txd_lower |= E1000_TXD_CMD_DEXT | E1000_TXD_DTYP_D;
3317                 txd_upper |= E1000_TXD_POPTS_TXSM << 8;
3318         }
3319
3320         if (tx_flags & E1000_TX_FLAGS_VLAN) {
3321                 txd_lower |= E1000_TXD_CMD_VLE;
3322                 txd_upper |= (tx_flags & E1000_TX_FLAGS_VLAN_MASK);
3323         }
3324
3325         i = tx_ring->next_to_use;
3326
3327         while (count--) {
3328                 buffer_info = &tx_ring->buffer_info[i];
3329                 tx_desc = E1000_TX_DESC(*tx_ring, i);
3330                 tx_desc->buffer_addr = cpu_to_le64(buffer_info->dma);
3331                 tx_desc->lower.data =
3332                         cpu_to_le32(txd_lower | buffer_info->length);
3333                 tx_desc->upper.data = cpu_to_le32(txd_upper);
3334
3335                 i++;
3336                 if (i == tx_ring->count)
3337                         i = 0;
3338         }
3339
3340         tx_desc->lower.data |= cpu_to_le32(adapter->txd_cmd);
3341
3342         /* Force memory writes to complete before letting h/w
3343          * know there are new descriptors to fetch.  (Only
3344          * applicable for weak-ordered memory model archs,
3345          * such as IA-64). */
3346         wmb();
3347
3348         tx_ring->next_to_use = i;
3349         writel(i, adapter->hw.hw_addr + tx_ring->tail);
3350         /* we need this if more than one processor can write to our tail
3351          * at a time, it synchronizes IO on IA64/Altix systems */
3352         mmiowb();
3353 }
3354
3355 #define MINIMUM_DHCP_PACKET_SIZE 282
3356 static int e1000_transfer_dhcp_info(struct e1000_adapter *adapter,
3357                                     struct sk_buff *skb)
3358 {
3359         struct e1000_hw *hw =  &adapter->hw;
3360         u16 length, offset;
3361
3362         if (vlan_tx_tag_present(skb)) {
3363                 if (!((vlan_tx_tag_get(skb) == adapter->hw.mng_cookie.vlan_id)
3364                     && (adapter->hw.mng_cookie.status &
3365                         E1000_MNG_DHCP_COOKIE_STATUS_VLAN)))
3366                         return 0;
3367         }
3368
3369         if (skb->len <= MINIMUM_DHCP_PACKET_SIZE)
3370                 return 0;
3371
3372         if (((struct ethhdr *) skb->data)->h_proto != htons(ETH_P_IP))
3373                 return 0;
3374
3375         {
3376                 const struct iphdr *ip = (struct iphdr *)((u8 *)skb->data+14);
3377                 struct udphdr *udp;
3378
3379                 if (ip->protocol != IPPROTO_UDP)
3380                         return 0;
3381
3382                 udp = (struct udphdr *)((u8 *)ip + (ip->ihl << 2));
3383                 if (ntohs(udp->dest) != 67)
3384                         return 0;
3385
3386                 offset = (u8 *)udp + 8 - skb->data;
3387                 length = skb->len - offset;
3388                 return e1000e_mng_write_dhcp_info(hw, (u8 *)udp + 8, length);
3389         }
3390
3391         return 0;
3392 }
3393
3394 static int __e1000_maybe_stop_tx(struct net_device *netdev, int size)
3395 {
3396         struct e1000_adapter *adapter = netdev_priv(netdev);
3397
3398         netif_stop_queue(netdev);
3399         /* Herbert's original patch had:
3400          *  smp_mb__after_netif_stop_queue();
3401          * but since that doesn't exist yet, just open code it. */
3402         smp_mb();
3403
3404         /* We need to check again in a case another CPU has just
3405          * made room available. */
3406         if (e1000_desc_unused(adapter->tx_ring) < size)
3407                 return -EBUSY;
3408
3409         /* A reprieve! */
3410         netif_start_queue(netdev);
3411         ++adapter->restart_queue;
3412         return 0;
3413 }
3414
3415 static int e1000_maybe_stop_tx(struct net_device *netdev, int size)
3416 {
3417         struct e1000_adapter *adapter = netdev_priv(netdev);
3418
3419         if (e1000_desc_unused(adapter->tx_ring) >= size)
3420                 return 0;
3421         return __e1000_maybe_stop_tx(netdev, size);
3422 }
3423
3424 #define TXD_USE_COUNT(S, X) (((S) >> (X)) + 1 )
3425 static int e1000_xmit_frame(struct sk_buff *skb, struct net_device *netdev)
3426 {
3427         struct e1000_adapter *adapter = netdev_priv(netdev);
3428         struct e1000_ring *tx_ring = adapter->tx_ring;
3429         unsigned int first;
3430         unsigned int max_per_txd = E1000_MAX_PER_TXD;
3431         unsigned int max_txd_pwr = E1000_MAX_TXD_PWR;
3432         unsigned int tx_flags = 0;
3433         unsigned int len = skb->len - skb->data_len;
3434         unsigned long irq_flags;
3435         unsigned int nr_frags;
3436         unsigned int mss;
3437         int count = 0;
3438         int tso;
3439         unsigned int f;
3440
3441         if (test_bit(__E1000_DOWN, &adapter->state)) {
3442                 dev_kfree_skb_any(skb);
3443                 return NETDEV_TX_OK;
3444         }
3445
3446         if (skb->len <= 0) {
3447                 dev_kfree_skb_any(skb);
3448                 return NETDEV_TX_OK;
3449         }
3450
3451         mss = skb_shinfo(skb)->gso_size;
3452         /* The controller does a simple calculation to
3453          * make sure there is enough room in the FIFO before
3454          * initiating the DMA for each buffer.  The calc is:
3455          * 4 = ceil(buffer len/mss).  To make sure we don't
3456          * overrun the FIFO, adjust the max buffer len if mss
3457          * drops. */
3458         if (mss) {
3459                 u8 hdr_len;
3460                 max_per_txd = min(mss << 2, max_per_txd);
3461                 max_txd_pwr = fls(max_per_txd) - 1;
3462
3463                 /* TSO Workaround for 82571/2/3 Controllers -- if skb->data
3464                 * points to just header, pull a few bytes of payload from
3465                 * frags into skb->data */
3466                 hdr_len = skb_transport_offset(skb) + tcp_hdrlen(skb);
3467                 if (skb->data_len && (hdr_len == len)) {
3468                         unsigned int pull_size;
3469
3470                         pull_size = min((unsigned int)4, skb->data_len);
3471                         if (!__pskb_pull_tail(skb, pull_size)) {
3472                                 ndev_err(netdev,
3473                                          "__pskb_pull_tail failed.\n");
3474                                 dev_kfree_skb_any(skb);
3475                                 return NETDEV_TX_OK;
3476                         }
3477                         len = skb->len - skb->data_len;
3478                 }
3479         }
3480
3481         /* reserve a descriptor for the offload context */
3482         if ((mss) || (skb->ip_summed == CHECKSUM_PARTIAL))
3483                 count++;
3484         count++;
3485
3486         count += TXD_USE_COUNT(len, max_txd_pwr);
3487
3488         nr_frags = skb_shinfo(skb)->nr_frags;
3489         for (f = 0; f < nr_frags; f++)
3490                 count += TXD_USE_COUNT(skb_shinfo(skb)->frags[f].size,
3491                                        max_txd_pwr);
3492
3493         if (adapter->hw.mac.tx_pkt_filtering)
3494                 e1000_transfer_dhcp_info(adapter, skb);
3495
3496         if (!spin_trylock_irqsave(&adapter->tx_queue_lock, irq_flags))
3497                 /* Collision - tell upper layer to requeue */
3498                 return NETDEV_TX_LOCKED;
3499
3500         /* need: count + 2 desc gap to keep tail from touching
3501          * head, otherwise try next time */
3502         if (e1000_maybe_stop_tx(netdev, count + 2)) {
3503                 spin_unlock_irqrestore(&adapter->tx_queue_lock, irq_flags);
3504                 return NETDEV_TX_BUSY;
3505         }
3506
3507         if (adapter->vlgrp && vlan_tx_tag_present(skb)) {
3508                 tx_flags |= E1000_TX_FLAGS_VLAN;
3509                 tx_flags |= (vlan_tx_tag_get(skb) << E1000_TX_FLAGS_VLAN_SHIFT);
3510         }
3511
3512         first = tx_ring->next_to_use;
3513
3514         tso = e1000_tso(adapter, skb);
3515         if (tso < 0) {
3516                 dev_kfree_skb_any(skb);
3517                 spin_unlock_irqrestore(&adapter->tx_queue_lock, irq_flags);
3518                 return NETDEV_TX_OK;
3519         }
3520
3521         if (tso)
3522                 tx_flags |= E1000_TX_FLAGS_TSO;
3523         else if (e1000_tx_csum(adapter, skb))
3524                 tx_flags |= E1000_TX_FLAGS_CSUM;
3525
3526         /* Old method was to assume IPv4 packet by default if TSO was enabled.
3527          * 82571 hardware supports TSO capabilities for IPv6 as well...
3528          * no longer assume, we must. */
3529         if (skb->protocol == htons(ETH_P_IP))
3530                 tx_flags |= E1000_TX_FLAGS_IPV4;
3531
3532         count = e1000_tx_map(adapter, skb, first, max_per_txd, nr_frags, mss);
3533         if (count < 0) {
3534                 /* handle pci_map_single() error in e1000_tx_map */
3535                 dev_kfree_skb_any(skb);
3536                 spin_unlock_irqrestore(&adapter->tx_queue_lock, irq_flags);
3537                 return NETDEV_TX_OK;
3538         }
3539
3540         e1000_tx_queue(adapter, tx_flags, count);
3541
3542         netdev->trans_start = jiffies;
3543
3544         /* Make sure there is space in the ring for the next send. */
3545         e1000_maybe_stop_tx(netdev, MAX_SKB_FRAGS + 2);
3546
3547         spin_unlock_irqrestore(&adapter->tx_queue_lock, irq_flags);
3548         return NETDEV_TX_OK;
3549 }
3550
3551 /**
3552  * e1000_tx_timeout - Respond to a Tx Hang
3553  * @netdev: network interface device structure
3554  **/
3555 static void e1000_tx_timeout(struct net_device *netdev)
3556 {
3557         struct e1000_adapter *adapter = netdev_priv(netdev);
3558
3559         /* Do the reset outside of interrupt context */
3560         adapter->tx_timeout_count++;
3561         schedule_work(&adapter->reset_task);
3562 }
3563
3564 static void e1000_reset_task(struct work_struct *work)
3565 {
3566         struct e1000_adapter *adapter;
3567         adapter = container_of(work, struct e1000_adapter, reset_task);
3568
3569         e1000e_reinit_locked(adapter);
3570 }
3571
3572 /**
3573  * e1000_get_stats - Get System Network Statistics
3574  * @netdev: network interface device structure
3575  *
3576  * Returns the address of the device statistics structure.
3577  * The statistics are actually updated from the timer callback.
3578  **/
3579 static struct net_device_stats *e1000_get_stats(struct net_device *netdev)
3580 {
3581         struct e1000_adapter *adapter = netdev_priv(netdev);
3582
3583         /* only return the current stats */
3584         return &adapter->net_stats;
3585 }
3586
3587 /**
3588  * e1000_change_mtu - Change the Maximum Transfer Unit
3589  * @netdev: network interface device structure
3590  * @new_mtu: new value for maximum frame size
3591  *
3592  * Returns 0 on success, negative on failure
3593  **/
3594 static int e1000_change_mtu(struct net_device *netdev, int new_mtu)
3595 {
3596         struct e1000_adapter *adapter = netdev_priv(netdev);
3597         int max_frame = new_mtu + ETH_HLEN + ETH_FCS_LEN;
3598
3599         if ((max_frame < ETH_ZLEN + ETH_FCS_LEN) ||
3600             (max_frame > MAX_JUMBO_FRAME_SIZE)) {
3601                 ndev_err(netdev, "Invalid MTU setting\n");
3602                 return -EINVAL;
3603         }
3604
3605         /* Jumbo frame size limits */
3606         if (max_frame > ETH_FRAME_LEN + ETH_FCS_LEN) {
3607                 if (!(adapter->flags & FLAG_HAS_JUMBO_FRAMES)) {
3608                         ndev_err(netdev, "Jumbo Frames not supported.\n");
3609                         return -EINVAL;
3610                 }
3611                 if (adapter->hw.phy.type == e1000_phy_ife) {
3612                         ndev_err(netdev, "Jumbo Frames not supported.\n");
3613                         return -EINVAL;
3614                 }
3615         }
3616
3617 #define MAX_STD_JUMBO_FRAME_SIZE 9234
3618         if (max_frame > MAX_STD_JUMBO_FRAME_SIZE) {
3619                 ndev_err(netdev, "MTU > 9216 not supported.\n");
3620                 return -EINVAL;
3621         }
3622
3623         while (test_and_set_bit(__E1000_RESETTING, &adapter->state))
3624                 msleep(1);
3625         /* e1000e_down has a dependency on max_frame_size */
3626         adapter->hw.mac.max_frame_size = max_frame;
3627         if (netif_running(netdev))
3628                 e1000e_down(adapter);
3629
3630         /* NOTE: netdev_alloc_skb reserves 16 bytes, and typically NET_IP_ALIGN
3631          * means we reserve 2 more, this pushes us to allocate from the next
3632          * larger slab size.
3633          * i.e. RXBUFFER_2048 --> size-4096 slab
3634          *  however with the new *_jumbo* routines, jumbo receives will use
3635          *  fragmented skbs */
3636
3637         if (max_frame <= 256)
3638                 adapter->rx_buffer_len = 256;
3639         else if (max_frame <= 512)
3640                 adapter->rx_buffer_len = 512;
3641         else if (max_frame <= 1024)
3642                 adapter->rx_buffer_len = 1024;
3643         else if (max_frame <= 2048)
3644                 adapter->rx_buffer_len = 2048;
3645         else
3646                 adapter->rx_buffer_len = 4096;
3647
3648         /* adjust allocation if LPE protects us, and we aren't using SBP */
3649         if ((max_frame == ETH_FRAME_LEN + ETH_FCS_LEN) ||
3650              (max_frame == ETH_FRAME_LEN + VLAN_HLEN + ETH_FCS_LEN))
3651                 adapter->rx_buffer_len = ETH_FRAME_LEN + VLAN_HLEN
3652                                          + ETH_FCS_LEN ;
3653
3654         ndev_info(netdev, "changing MTU from %d to %d\n",
3655                 netdev->mtu, new_mtu);
3656         netdev->mtu = new_mtu;
3657
3658         if (netif_running(netdev))
3659                 e1000e_up(adapter);
3660         else
3661                 e1000e_reset(adapter);
3662
3663         clear_bit(__E1000_RESETTING, &adapter->state);
3664
3665         return 0;
3666 }
3667
3668 static int e1000_mii_ioctl(struct net_device *netdev, struct ifreq *ifr,
3669                            int cmd)
3670 {
3671         struct e1000_adapter *adapter = netdev_priv(netdev);
3672         struct mii_ioctl_data *data = if_mii(ifr);
3673         unsigned long irq_flags;
3674
3675         if (adapter->hw.media_type != e1000_media_type_copper)
3676                 return -EOPNOTSUPP;
3677
3678         switch (cmd) {
3679         case SIOCGMIIPHY:
3680                 data->phy_id = adapter->hw.phy.addr;
3681                 break;
3682         case SIOCGMIIREG:
3683                 if (!capable(CAP_NET_ADMIN))
3684                         return -EPERM;
3685                 spin_lock_irqsave(&adapter->stats_lock, irq_flags);
3686                 if (e1e_rphy(&adapter->hw, data->reg_num & 0x1F,
3687                                    &data->val_out)) {
3688                         spin_unlock_irqrestore(&adapter->stats_lock, irq_flags);
3689                         return -EIO;
3690                 }
3691                 spin_unlock_irqrestore(&adapter->stats_lock, irq_flags);
3692                 break;
3693         case SIOCSMIIREG:
3694         default:
3695                 return -EOPNOTSUPP;
3696         }
3697         return 0;
3698 }
3699
3700 static int e1000_ioctl(struct net_device *netdev, struct ifreq *ifr, int cmd)
3701 {
3702         switch (cmd) {
3703         case SIOCGMIIPHY:
3704         case SIOCGMIIREG:
3705         case SIOCSMIIREG:
3706                 return e1000_mii_ioctl(netdev, ifr, cmd);
3707         default:
3708                 return -EOPNOTSUPP;
3709         }
3710 }
3711
3712 static int e1000_suspend(struct pci_dev *pdev, pm_message_t state)
3713 {
3714         struct net_device *netdev = pci_get_drvdata(pdev);
3715         struct e1000_adapter *adapter = netdev_priv(netdev);
3716         struct e1000_hw *hw = &adapter->hw;
3717         u32 ctrl, ctrl_ext, rctl, status;
3718         u32 wufc = adapter->wol;
3719         int retval = 0;
3720
3721         netif_device_detach(netdev);
3722
3723         if (netif_running(netdev)) {
3724                 WARN_ON(test_bit(__E1000_RESETTING, &adapter->state));
3725                 e1000e_down(adapter);
3726                 e1000_free_irq(adapter);
3727         }
3728
3729         retval = pci_save_state(pdev);
3730         if (retval)
3731                 return retval;
3732
3733         status = er32(STATUS);
3734         if (status & E1000_STATUS_LU)
3735                 wufc &= ~E1000_WUFC_LNKC;
3736
3737         if (wufc) {
3738                 e1000_setup_rctl(adapter);
3739                 e1000_set_multi(netdev);
3740
3741                 /* turn on all-multi mode if wake on multicast is enabled */
3742                 if (wufc & E1000_WUFC_MC) {
3743                         rctl = er32(RCTL);
3744                         rctl |= E1000_RCTL_MPE;
3745                         ew32(RCTL, rctl);
3746                 }
3747
3748                 ctrl = er32(CTRL);
3749                 /* advertise wake from D3Cold */
3750                 #define E1000_CTRL_ADVD3WUC 0x00100000
3751                 /* phy power management enable */
3752                 #define E1000_CTRL_EN_PHY_PWR_MGMT 0x00200000
3753                 ctrl |= E1000_CTRL_ADVD3WUC |
3754                         E1000_CTRL_EN_PHY_PWR_MGMT;
3755                 ew32(CTRL, ctrl);
3756
3757                 if (adapter->hw.media_type == e1000_media_type_fiber ||
3758                    adapter->hw.media_type == e1000_media_type_internal_serdes) {
3759                         /* keep the laser running in D3 */
3760                         ctrl_ext = er32(CTRL_EXT);
3761                         ctrl_ext |= E1000_CTRL_EXT_SDP7_DATA;
3762                         ew32(CTRL_EXT, ctrl_ext);
3763                 }
3764
3765                 /* Allow time for pending master requests to run */
3766                 e1000e_disable_pcie_master(&adapter->hw);
3767
3768                 ew32(WUC, E1000_WUC_PME_EN);
3769                 ew32(WUFC, wufc);
3770                 pci_enable_wake(pdev, PCI_D3hot, 1);
3771                 pci_enable_wake(pdev, PCI_D3cold, 1);
3772         } else {
3773                 ew32(WUC, 0);
3774                 ew32(WUFC, 0);
3775                 pci_enable_wake(pdev, PCI_D3hot, 0);
3776                 pci_enable_wake(pdev, PCI_D3cold, 0);
3777         }
3778
3779         e1000_release_manageability(adapter);
3780
3781         /* make sure adapter isn't asleep if manageability is enabled */
3782         if (adapter->flags & FLAG_MNG_PT_ENABLED) {
3783                 pci_enable_wake(pdev, PCI_D3hot, 1);
3784                 pci_enable_wake(pdev, PCI_D3cold, 1);
3785         }
3786
3787         if (adapter->hw.phy.type == e1000_phy_igp_3)
3788                 e1000e_igp3_phy_powerdown_workaround_ich8lan(&adapter->hw);
3789
3790         /* Release control of h/w to f/w.  If f/w is AMT enabled, this
3791          * would have already happened in close and is redundant. */
3792         e1000_release_hw_control(adapter);
3793
3794         pci_disable_device(pdev);
3795
3796         pci_set_power_state(pdev, pci_choose_state(pdev, state));
3797
3798         return 0;
3799 }
3800
3801 #ifdef CONFIG_PM
3802 static int e1000_resume(struct pci_dev *pdev)
3803 {
3804         struct net_device *netdev = pci_get_drvdata(pdev);
3805         struct e1000_adapter *adapter = netdev_priv(netdev);
3806         struct e1000_hw *hw = &adapter->hw;
3807         u32 err;
3808
3809         pci_set_power_state(pdev, PCI_D0);
3810         pci_restore_state(pdev);
3811         err = pci_enable_device(pdev);
3812         if (err) {
3813                 dev_err(&pdev->dev,
3814                         "Cannot enable PCI device from suspend\n");
3815                 return err;
3816         }
3817
3818         pci_set_master(pdev);
3819
3820         pci_enable_wake(pdev, PCI_D3hot, 0);
3821         pci_enable_wake(pdev, PCI_D3cold, 0);
3822
3823         if (netif_running(netdev)) {
3824                 err = e1000_request_irq(adapter);
3825                 if (err)
3826                         return err;
3827         }
3828
3829         e1000e_power_up_phy(adapter);
3830         e1000e_reset(adapter);
3831         ew32(WUS, ~0);
3832
3833         e1000_init_manageability(adapter);
3834
3835         if (netif_running(netdev))
3836                 e1000e_up(adapter);
3837
3838         netif_device_attach(netdev);
3839
3840         /* If the controller has AMT, do not set DRV_LOAD until the interface
3841          * is up.  For all other cases, let the f/w know that the h/w is now
3842          * under the control of the driver. */
3843         if (!(adapter->flags & FLAG_HAS_AMT) || !e1000e_check_mng_mode(&adapter->hw))
3844                 e1000_get_hw_control(adapter);
3845
3846         return 0;
3847 }
3848 #endif
3849
3850 static void e1000_shutdown(struct pci_dev *pdev)
3851 {
3852         e1000_suspend(pdev, PMSG_SUSPEND);
3853 }
3854
3855 #ifdef CONFIG_NET_POLL_CONTROLLER
3856 /*
3857  * Polling 'interrupt' - used by things like netconsole to send skbs
3858  * without having to re-enable interrupts. It's not called while
3859  * the interrupt routine is executing.
3860  */
3861 static void e1000_netpoll(struct net_device *netdev)
3862 {
3863         struct e1000_adapter *adapter = netdev_priv(netdev);
3864
3865         disable_irq(adapter->pdev->irq);
3866         e1000_intr(adapter->pdev->irq, netdev);
3867
3868         e1000_clean_tx_irq(adapter);
3869
3870         enable_irq(adapter->pdev->irq);
3871 }
3872 #endif
3873
3874 /**
3875  * e1000_io_error_detected - called when PCI error is detected
3876  * @pdev: Pointer to PCI device
3877  * @state: The current pci connection state
3878  *
3879  * This function is called after a PCI bus error affecting
3880  * this device has been detected.
3881  */
3882 static pci_ers_result_t e1000_io_error_detected(struct pci_dev *pdev,
3883                                                 pci_channel_state_t state)
3884 {
3885         struct net_device *netdev = pci_get_drvdata(pdev);
3886         struct e1000_adapter *adapter = netdev_priv(netdev);
3887
3888         netif_device_detach(netdev);
3889
3890         if (netif_running(netdev))
3891                 e1000e_down(adapter);
3892         pci_disable_device(pdev);
3893
3894         /* Request a slot slot reset. */
3895         return PCI_ERS_RESULT_NEED_RESET;
3896 }
3897
3898 /**
3899  * e1000_io_slot_reset - called after the pci bus has been reset.
3900  * @pdev: Pointer to PCI device
3901  *
3902  * Restart the card from scratch, as if from a cold-boot. Implementation
3903  * resembles the first-half of the e1000_resume routine.
3904  */
3905 static pci_ers_result_t e1000_io_slot_reset(struct pci_dev *pdev)
3906 {
3907         struct net_device *netdev = pci_get_drvdata(pdev);
3908         struct e1000_adapter *adapter = netdev_priv(netdev);
3909         struct e1000_hw *hw = &adapter->hw;
3910
3911         if (pci_enable_device(pdev)) {
3912                 dev_err(&pdev->dev,
3913                         "Cannot re-enable PCI device after reset.\n");
3914                 return PCI_ERS_RESULT_DISCONNECT;
3915         }
3916         pci_set_master(pdev);
3917
3918         pci_enable_wake(pdev, PCI_D3hot, 0);
3919         pci_enable_wake(pdev, PCI_D3cold, 0);
3920
3921         e1000e_reset(adapter);
3922         ew32(WUS, ~0);
3923
3924         return PCI_ERS_RESULT_RECOVERED;
3925 }
3926
3927 /**
3928  * e1000_io_resume - called when traffic can start flowing again.
3929  * @pdev: Pointer to PCI device
3930  *
3931  * This callback is called when the error recovery driver tells us that
3932  * its OK to resume normal operation. Implementation resembles the
3933  * second-half of the e1000_resume routine.
3934  */
3935 static void e1000_io_resume(struct pci_dev *pdev)
3936 {
3937         struct net_device *netdev = pci_get_drvdata(pdev);
3938         struct e1000_adapter *adapter = netdev_priv(netdev);
3939
3940         e1000_init_manageability(adapter);
3941
3942         if (netif_running(netdev)) {
3943                 if (e1000e_up(adapter)) {
3944                         dev_err(&pdev->dev,
3945                                 "can't bring device back up after reset\n");
3946                         return;
3947                 }
3948         }
3949
3950         netif_device_attach(netdev);
3951
3952         /* If the controller has AMT, do not set DRV_LOAD until the interface
3953          * is up.  For all other cases, let the f/w know that the h/w is now
3954          * under the control of the driver. */
3955         if (!(adapter->flags & FLAG_HAS_AMT) ||
3956             !e1000e_check_mng_mode(&adapter->hw))
3957                 e1000_get_hw_control(adapter);
3958
3959 }
3960
3961 static void e1000_print_device_info(struct e1000_adapter *adapter)
3962 {
3963         struct e1000_hw *hw = &adapter->hw;
3964         struct net_device *netdev = adapter->netdev;
3965         u32 part_num;
3966
3967         /* print bus type/speed/width info */
3968         ndev_info(netdev, "(PCI Express:2.5GB/s:%s) "
3969                   "%02x:%02x:%02x:%02x:%02x:%02x\n",
3970                   /* bus width */
3971                  ((hw->bus.width == e1000_bus_width_pcie_x4) ? "Width x4" :
3972                   "Width x1"),
3973                   /* MAC address */
3974                   netdev->dev_addr[0], netdev->dev_addr[1],
3975                   netdev->dev_addr[2], netdev->dev_addr[3],
3976                   netdev->dev_addr[4], netdev->dev_addr[5]);
3977         ndev_info(netdev, "Intel(R) PRO/%s Network Connection\n",
3978                   (hw->phy.type == e1000_phy_ife)
3979                    ? "10/100" : "1000");
3980         e1000e_read_part_num(hw, &part_num);
3981         ndev_info(netdev, "MAC: %d, PHY: %d, PBA No: %06x-%03x\n",
3982                   hw->mac.type, hw->phy.type,
3983                   (part_num >> 8), (part_num & 0xff));
3984 }
3985
3986 /**
3987  * e1000_probe - Device Initialization Routine
3988  * @pdev: PCI device information struct
3989  * @ent: entry in e1000_pci_tbl
3990  *
3991  * Returns 0 on success, negative on failure
3992  *
3993  * e1000_probe initializes an adapter identified by a pci_dev structure.
3994  * The OS initialization, configuring of the adapter private structure,
3995  * and a hardware reset occur.
3996  **/
3997 static int __devinit e1000_probe(struct pci_dev *pdev,
3998                                  const struct pci_device_id *ent)
3999 {
4000         struct net_device *netdev;
4001         struct e1000_adapter *adapter;
4002         struct e1000_hw *hw;
4003         const struct e1000_info *ei = e1000_info_tbl[ent->driver_data];
4004         unsigned long mmio_start, mmio_len;
4005         unsigned long flash_start, flash_len;
4006
4007         static int cards_found;
4008         int i, err, pci_using_dac;
4009         u16 eeprom_data = 0;
4010         u16 eeprom_apme_mask = E1000_EEPROM_APME;
4011
4012         err = pci_enable_device(pdev);
4013         if (err)
4014                 return err;
4015
4016         pci_using_dac = 0;
4017         err = pci_set_dma_mask(pdev, DMA_64BIT_MASK);
4018         if (!err) {
4019                 err = pci_set_consistent_dma_mask(pdev, DMA_64BIT_MASK);
4020                 if (!err)
4021                         pci_using_dac = 1;
4022         } else {
4023                 err = pci_set_dma_mask(pdev, DMA_32BIT_MASK);
4024                 if (err) {
4025                         err = pci_set_consistent_dma_mask(pdev,
4026                                                           DMA_32BIT_MASK);
4027                         if (err) {
4028                                 dev_err(&pdev->dev, "No usable DMA "
4029                                         "configuration, aborting\n");
4030                                 goto err_dma;
4031                         }
4032                 }
4033         }
4034
4035         err = pci_request_regions(pdev, e1000e_driver_name);
4036         if (err)
4037                 goto err_pci_reg;
4038
4039         pci_set_master(pdev);
4040
4041         err = -ENOMEM;
4042         netdev = alloc_etherdev(sizeof(struct e1000_adapter));
4043         if (!netdev)
4044                 goto err_alloc_etherdev;
4045
4046         SET_NETDEV_DEV(netdev, &pdev->dev);
4047
4048         pci_set_drvdata(pdev, netdev);
4049         adapter = netdev_priv(netdev);
4050         hw = &adapter->hw;
4051         adapter->netdev = netdev;
4052         adapter->pdev = pdev;
4053         adapter->ei = ei;
4054         adapter->pba = ei->pba;
4055         adapter->flags = ei->flags;
4056         adapter->hw.adapter = adapter;
4057         adapter->hw.mac.type = ei->mac;
4058         adapter->msg_enable = (1 << NETIF_MSG_DRV | NETIF_MSG_PROBE) - 1;
4059
4060         mmio_start = pci_resource_start(pdev, 0);
4061         mmio_len = pci_resource_len(pdev, 0);
4062
4063         err = -EIO;
4064         adapter->hw.hw_addr = ioremap(mmio_start, mmio_len);
4065         if (!adapter->hw.hw_addr)
4066                 goto err_ioremap;
4067
4068         if ((adapter->flags & FLAG_HAS_FLASH) &&
4069             (pci_resource_flags(pdev, 1) & IORESOURCE_MEM)) {
4070                 flash_start = pci_resource_start(pdev, 1);
4071                 flash_len = pci_resource_len(pdev, 1);
4072                 adapter->hw.flash_address = ioremap(flash_start, flash_len);
4073                 if (!adapter->hw.flash_address)
4074                         goto err_flashmap;
4075         }
4076
4077         /* construct the net_device struct */
4078         netdev->open                    = &e1000_open;
4079         netdev->stop                    = &e1000_close;
4080         netdev->hard_start_xmit         = &e1000_xmit_frame;
4081         netdev->get_stats               = &e1000_get_stats;
4082         netdev->set_multicast_list      = &e1000_set_multi;
4083         netdev->set_mac_address         = &e1000_set_mac;
4084         netdev->change_mtu              = &e1000_change_mtu;
4085         netdev->do_ioctl                = &e1000_ioctl;
4086         e1000e_set_ethtool_ops(netdev);
4087         netdev->tx_timeout              = &e1000_tx_timeout;
4088         netdev->watchdog_timeo          = 5 * HZ;
4089         netif_napi_add(netdev, &adapter->napi, e1000_clean, 64);
4090         netdev->vlan_rx_register        = e1000_vlan_rx_register;
4091         netdev->vlan_rx_add_vid         = e1000_vlan_rx_add_vid;
4092         netdev->vlan_rx_kill_vid        = e1000_vlan_rx_kill_vid;
4093 #ifdef CONFIG_NET_POLL_CONTROLLER
4094         netdev->poll_controller         = e1000_netpoll;
4095 #endif
4096         strncpy(netdev->name, pci_name(pdev), sizeof(netdev->name) - 1);
4097
4098         netdev->mem_start = mmio_start;
4099         netdev->mem_end = mmio_start + mmio_len;
4100
4101         adapter->bd_number = cards_found++;
4102
4103         /* setup adapter struct */
4104         err = e1000_sw_init(adapter);
4105         if (err)
4106                 goto err_sw_init;
4107
4108         err = -EIO;
4109
4110         memcpy(&hw->mac.ops, ei->mac_ops, sizeof(hw->mac.ops));
4111         memcpy(&hw->nvm.ops, ei->nvm_ops, sizeof(hw->nvm.ops));
4112         memcpy(&hw->phy.ops, ei->phy_ops, sizeof(hw->phy.ops));
4113
4114         err = ei->get_invariants(adapter);
4115         if (err)
4116                 goto err_hw_init;
4117
4118         hw->mac.ops.get_bus_info(&adapter->hw);
4119
4120         adapter->hw.phy.wait_for_link = 0;
4121
4122         /* Copper options */
4123         if (adapter->hw.media_type == e1000_media_type_copper) {
4124                 adapter->hw.phy.mdix = AUTO_ALL_MODES;
4125                 adapter->hw.phy.disable_polarity_correction = 0;
4126                 adapter->hw.phy.ms_type = e1000_ms_hw_default;
4127         }
4128
4129         if (e1000_check_reset_block(&adapter->hw))
4130                 ndev_info(netdev,
4131                           "PHY reset is blocked due to SOL/IDER session.\n");
4132
4133         netdev->features = NETIF_F_SG |
4134                            NETIF_F_HW_CSUM |
4135                            NETIF_F_HW_VLAN_TX |
4136                            NETIF_F_HW_VLAN_RX;
4137
4138         if (adapter->flags & FLAG_HAS_HW_VLAN_FILTER)
4139                 netdev->features |= NETIF_F_HW_VLAN_FILTER;
4140
4141         netdev->features |= NETIF_F_TSO;
4142         netdev->features |= NETIF_F_TSO6;
4143
4144         if (pci_using_dac)
4145                 netdev->features |= NETIF_F_HIGHDMA;
4146
4147         /* We should not be using LLTX anymore, but we are still TX faster with
4148          * it. */
4149         netdev->features |= NETIF_F_LLTX;
4150
4151         if (e1000e_enable_mng_pass_thru(&adapter->hw))
4152                 adapter->flags |= FLAG_MNG_PT_ENABLED;
4153
4154         /* before reading the NVM, reset the controller to
4155          * put the device in a known good starting state */
4156         adapter->hw.mac.ops.reset_hw(&adapter->hw);
4157
4158         /*
4159          * systems with ASPM and others may see the checksum fail on the first
4160          * attempt. Let's give it a few tries
4161          */
4162         for (i = 0;; i++) {
4163                 if (e1000_validate_nvm_checksum(&adapter->hw) >= 0)
4164                         break;
4165                 if (i == 2) {
4166                         ndev_err(netdev, "The NVM Checksum Is Not Valid\n");
4167                         err = -EIO;
4168                         goto err_eeprom;
4169                 }
4170         }
4171
4172         /* copy the MAC address out of the NVM */
4173         if (e1000e_read_mac_addr(&adapter->hw))
4174                 ndev_err(netdev, "NVM Read Error while reading MAC address\n");
4175
4176         memcpy(netdev->dev_addr, adapter->hw.mac.addr, netdev->addr_len);
4177         memcpy(netdev->perm_addr, adapter->hw.mac.addr, netdev->addr_len);
4178
4179         if (!is_valid_ether_addr(netdev->perm_addr)) {
4180                 ndev_err(netdev, "Invalid MAC Address: "
4181                          "%02x:%02x:%02x:%02x:%02x:%02x\n",
4182                          netdev->perm_addr[0], netdev->perm_addr[1],
4183                          netdev->perm_addr[2], netdev->perm_addr[3],
4184                          netdev->perm_addr[4], netdev->perm_addr[5]);
4185                 err = -EIO;
4186                 goto err_eeprom;
4187         }
4188
4189         init_timer(&adapter->watchdog_timer);
4190         adapter->watchdog_timer.function = &e1000_watchdog;
4191         adapter->watchdog_timer.data = (unsigned long) adapter;
4192
4193         init_timer(&adapter->phy_info_timer);
4194         adapter->phy_info_timer.function = &e1000_update_phy_info;
4195         adapter->phy_info_timer.data = (unsigned long) adapter;
4196
4197         INIT_WORK(&adapter->reset_task, e1000_reset_task);
4198         INIT_WORK(&adapter->watchdog_task, e1000_watchdog_task);
4199
4200         e1000e_check_options(adapter);
4201
4202         /* Initialize link parameters. User can change them with ethtool */
4203         adapter->hw.mac.autoneg = 1;
4204         adapter->fc_autoneg = 1;
4205         adapter->hw.mac.original_fc = e1000_fc_default;
4206         adapter->hw.mac.fc = e1000_fc_default;
4207         adapter->hw.phy.autoneg_advertised = 0x2f;
4208
4209         /* ring size defaults */
4210         adapter->rx_ring->count = 256;
4211         adapter->tx_ring->count = 256;
4212
4213         /*
4214          * Initial Wake on LAN setting - If APM wake is enabled in
4215          * the EEPROM, enable the ACPI Magic Packet filter
4216          */
4217         if (adapter->flags & FLAG_APME_IN_WUC) {
4218                 /* APME bit in EEPROM is mapped to WUC.APME */
4219                 eeprom_data = er32(WUC);
4220                 eeprom_apme_mask = E1000_WUC_APME;
4221         } else if (adapter->flags & FLAG_APME_IN_CTRL3) {
4222                 if (adapter->flags & FLAG_APME_CHECK_PORT_B &&
4223                     (adapter->hw.bus.func == 1))
4224                         e1000_read_nvm(&adapter->hw,
4225                                 NVM_INIT_CONTROL3_PORT_B, 1, &eeprom_data);
4226                 else
4227                         e1000_read_nvm(&adapter->hw,
4228                                 NVM_INIT_CONTROL3_PORT_A, 1, &eeprom_data);
4229         }
4230
4231         /* fetch WoL from EEPROM */
4232         if (eeprom_data & eeprom_apme_mask)
4233                 adapter->eeprom_wol |= E1000_WUFC_MAG;
4234
4235         /*
4236          * now that we have the eeprom settings, apply the special cases
4237          * where the eeprom may be wrong or the board simply won't support
4238          * wake on lan on a particular port
4239          */
4240         if (!(adapter->flags & FLAG_HAS_WOL))
4241                 adapter->eeprom_wol = 0;
4242
4243         /* initialize the wol settings based on the eeprom settings */
4244         adapter->wol = adapter->eeprom_wol;
4245
4246         /* reset the hardware with the new settings */
4247         e1000e_reset(adapter);
4248
4249         /* If the controller has AMT, do not set DRV_LOAD until the interface
4250          * is up.  For all other cases, let the f/w know that the h/w is now
4251          * under the control of the driver. */
4252         if (!(adapter->flags & FLAG_HAS_AMT) ||
4253             !e1000e_check_mng_mode(&adapter->hw))
4254                 e1000_get_hw_control(adapter);
4255
4256         /* tell the stack to leave us alone until e1000_open() is called */
4257         netif_carrier_off(netdev);
4258         netif_stop_queue(netdev);
4259
4260         strcpy(netdev->name, "eth%d");
4261         err = register_netdev(netdev);
4262         if (err)
4263                 goto err_register;
4264
4265         e1000_print_device_info(adapter);
4266
4267         return 0;
4268
4269 err_register:
4270 err_hw_init:
4271         e1000_release_hw_control(adapter);
4272 err_eeprom:
4273         if (!e1000_check_reset_block(&adapter->hw))
4274                 e1000_phy_hw_reset(&adapter->hw);
4275
4276         if (adapter->hw.flash_address)
4277                 iounmap(adapter->hw.flash_address);
4278
4279 err_flashmap:
4280         kfree(adapter->tx_ring);
4281         kfree(adapter->rx_ring);
4282 err_sw_init:
4283         iounmap(adapter->hw.hw_addr);
4284 err_ioremap:
4285         free_netdev(netdev);
4286 err_alloc_etherdev:
4287         pci_release_regions(pdev);
4288 err_pci_reg:
4289 err_dma:
4290         pci_disable_device(pdev);
4291         return err;
4292 }
4293
4294 /**
4295  * e1000_remove - Device Removal Routine
4296  * @pdev: PCI device information struct
4297  *
4298  * e1000_remove is called by the PCI subsystem to alert the driver
4299  * that it should release a PCI device.  The could be caused by a
4300  * Hot-Plug event, or because the driver is going to be removed from
4301  * memory.
4302  **/
4303 static void __devexit e1000_remove(struct pci_dev *pdev)
4304 {
4305         struct net_device *netdev = pci_get_drvdata(pdev);
4306         struct e1000_adapter *adapter = netdev_priv(netdev);
4307
4308         /* flush_scheduled work may reschedule our watchdog task, so
4309          * explicitly disable watchdog tasks from being rescheduled  */
4310         set_bit(__E1000_DOWN, &adapter->state);
4311         del_timer_sync(&adapter->watchdog_timer);
4312         del_timer_sync(&adapter->phy_info_timer);
4313
4314         flush_scheduled_work();
4315
4316         e1000_release_manageability(adapter);
4317
4318         /* Release control of h/w to f/w.  If f/w is AMT enabled, this
4319          * would have already happened in close and is redundant. */
4320         e1000_release_hw_control(adapter);
4321
4322         unregister_netdev(netdev);
4323
4324         if (!e1000_check_reset_block(&adapter->hw))
4325                 e1000_phy_hw_reset(&adapter->hw);
4326
4327         kfree(adapter->tx_ring);
4328         kfree(adapter->rx_ring);
4329
4330         iounmap(adapter->hw.hw_addr);
4331         if (adapter->hw.flash_address)
4332                 iounmap(adapter->hw.flash_address);
4333         pci_release_regions(pdev);
4334
4335         free_netdev(netdev);
4336
4337         pci_disable_device(pdev);
4338 }
4339
4340 /* PCI Error Recovery (ERS) */
4341 static struct pci_error_handlers e1000_err_handler = {
4342         .error_detected = e1000_io_error_detected,
4343         .slot_reset = e1000_io_slot_reset,
4344         .resume = e1000_io_resume,
4345 };
4346
4347 static struct pci_device_id e1000_pci_tbl[] = {
4348         /*
4349          * Support for 82571/2/3, es2lan and ich8 will be phased in
4350          * stepwise.
4351
4352         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82571EB_COPPER), board_82571 },
4353         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82571EB_FIBER), board_82571 },
4354         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82571EB_QUAD_COPPER), board_82571 },
4355         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82571EB_QUAD_COPPER_LP), board_82571 },
4356         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82571EB_QUAD_FIBER), board_82571 },
4357         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82571EB_SERDES), board_82571 },
4358         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82572EI), board_82572 },
4359         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82572EI_COPPER), board_82572 },
4360         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82572EI_FIBER), board_82572 },
4361         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82572EI_SERDES), board_82572 },
4362         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82573E), board_82573 },
4363         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82573E_IAMT), board_82573 },
4364         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82573L), board_82573 },
4365         { PCI_VDEVICE(INTEL, E1000_DEV_ID_80003ES2LAN_COPPER_DPT),
4366           board_80003es2lan },
4367         { PCI_VDEVICE(INTEL, E1000_DEV_ID_80003ES2LAN_COPPER_SPT),
4368           board_80003es2lan },
4369         { PCI_VDEVICE(INTEL, E1000_DEV_ID_80003ES2LAN_SERDES_DPT),
4370           board_80003es2lan },
4371         { PCI_VDEVICE(INTEL, E1000_DEV_ID_80003ES2LAN_SERDES_SPT),
4372           board_80003es2lan },
4373         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH8_IFE), board_ich8lan },
4374         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH8_IFE_G), board_ich8lan },
4375         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH8_IFE_GT), board_ich8lan },
4376         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH8_IGP_AMT), board_ich8lan },
4377         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH8_IGP_C), board_ich8lan },
4378         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH8_IGP_M), board_ich8lan },
4379         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH8_IGP_M_AMT), board_ich8lan },
4380         */
4381
4382         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH9_IFE), board_ich9lan },
4383         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH9_IFE_G), board_ich9lan },
4384         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH9_IFE_GT), board_ich9lan },
4385         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH9_IGP_AMT), board_ich9lan },
4386         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH9_IGP_C), board_ich9lan },
4387
4388         { }     /* terminate list */
4389 };
4390 MODULE_DEVICE_TABLE(pci, e1000_pci_tbl);
4391
4392 /* PCI Device API Driver */
4393 static struct pci_driver e1000_driver = {
4394         .name     = e1000e_driver_name,
4395         .id_table = e1000_pci_tbl,
4396         .probe    = e1000_probe,
4397         .remove   = __devexit_p(e1000_remove),
4398 #ifdef CONFIG_PM
4399         /* Power Managment Hooks */
4400         .suspend  = e1000_suspend,
4401         .resume   = e1000_resume,
4402 #endif
4403         .shutdown = e1000_shutdown,
4404         .err_handler = &e1000_err_handler
4405 };
4406
4407 /**
4408  * e1000_init_module - Driver Registration Routine
4409  *
4410  * e1000_init_module is the first routine called when the driver is
4411  * loaded. All it does is register with the PCI subsystem.
4412  **/
4413 static int __init e1000_init_module(void)
4414 {
4415         int ret;
4416         printk(KERN_INFO "%s: Intel(R) PRO/1000 Network Driver - %s\n",
4417                e1000e_driver_name, e1000e_driver_version);
4418         printk(KERN_INFO "%s: Copyright (c) 1999-2007 Intel Corporation.\n",
4419                e1000e_driver_name);
4420         ret = pci_register_driver(&e1000_driver);
4421
4422         return ret;
4423 }
4424 module_init(e1000_init_module);
4425
4426 /**
4427  * e1000_exit_module - Driver Exit Cleanup Routine
4428  *
4429  * e1000_exit_module is called just before the driver is removed
4430  * from memory.
4431  **/
4432 static void __exit e1000_exit_module(void)
4433 {
4434         pci_unregister_driver(&e1000_driver);
4435 }
4436 module_exit(e1000_exit_module);
4437
4438
4439 MODULE_AUTHOR("Intel Corporation, <linux.nics@intel.com>");
4440 MODULE_DESCRIPTION("Intel(R) PRO/1000 Network Driver");
4441 MODULE_LICENSE("GPL");
4442 MODULE_VERSION(DRV_VERSION);
4443
4444 /* e1000_main.c */