drivers/net/e1000e/netdev.c

   1 /*******************************************************************************
   2
   3   Intel PRO/1000 Linux driver
   4   Copyright(c) 1999 - 2008 Intel Corporation.
   5
   6   This program is free software; you can redistribute it and/or modify it
   7   under the terms and conditions of the GNU General Public License,
   8   version 2, as published by the Free Software Foundation.
   9
  10   This program is distributed in the hope it will be useful, but WITHOUT
  11   ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
  12   FITNESS FOR A PARTICULAR PURPOSE.  See the GNU General Public License for
  13   more details.
  14
  15   You should have received a copy of the GNU General Public License along with
  16   this program; if not, write to the Free Software Foundation, Inc.,
  17   51 Franklin St - Fifth Floor, Boston, MA 02110-1301 USA.
  18
  19   The full GNU General Public License is included in this distribution in
  20   the file called "COPYING".
  21
  22   Contact Information:
  23   Linux NICS <linux.nics@intel.com>
  24   e1000-devel Mailing List <e1000-devel@lists.sourceforge.net>
  25   Intel Corporation, 5200 N.E. Elam Young Parkway, Hillsboro, OR 97124-6497
  26
  27 *******************************************************************************/
  28
  29 #include <linux/module.h>
  30 #include <linux/types.h>
  31 #include <linux/init.h>
  32 #include <linux/pci.h>
  33 #include <linux/vmalloc.h>
  34 #include <linux/pagemap.h>
  35 #include <linux/delay.h>
  36 #include <linux/netdevice.h>
  37 #include <linux/tcp.h>
  38 #include <linux/ipv6.h>
  39 #include <net/checksum.h>
  40 #include <net/ip6_checksum.h>
  41 #include <linux/mii.h>
  42 #include <linux/ethtool.h>
  43 #include <linux/if_vlan.h>
  44 #include <linux/cpu.h>
  45 #include <linux/smp.h>
  46 #include <linux/pm_qos_params.h>
  47 #include <linux/aer.h>
  48
  49 #include "e1000.h"
  50
  51 #define DRV_VERSION "0.3.3.4-k4"
  52 char e1000e_driver_name[] = "e1000e";
  53 const char e1000e_driver_version[] = DRV_VERSION;
  54
  55 static const struct e1000_info *e1000_info_tbl[] = {
  56         [board_82571]           = &e1000_82571_info,
  57         [board_82572]           = &e1000_82572_info,
  58         [board_82573]           = &e1000_82573_info,
  59         [board_82574]           = &e1000_82574_info,
  60         [board_82583]           = &e1000_82583_info,
  61         [board_80003es2lan]     = &e1000_es2_info,
  62         [board_ich8lan]         = &e1000_ich8_info,
  63         [board_ich9lan]         = &e1000_ich9_info,
  64         [board_ich10lan]        = &e1000_ich10_info,
  65 };
  66
  67 #ifdef DEBUG
  68 /**
  69  * e1000_get_hw_dev_name - return device name string
  70  * used by hardware layer to print debugging information
  71  **/
  72 char *e1000e_get_hw_dev_name(struct e1000_hw *hw)
  73 {
  74         return hw->adapter->netdev->name;
  75 }
  76 #endif
  77
  78 /**
  79  * e1000_desc_unused - calculate if we have unused descriptors
  80  **/
  81 static int e1000_desc_unused(struct e1000_ring *ring)
  82 {
  83         if (ring->next_to_clean > ring->next_to_use)
  84                 return ring->next_to_clean - ring->next_to_use - 1;
  85
  86         return ring->count + ring->next_to_clean - ring->next_to_use - 1;
  87 }
  88
  89 /**
  90  * e1000_receive_skb - helper function to handle Rx indications
  91  * @adapter: board private structure
  92  * @status: descriptor status field as written by hardware
  93  * @vlan: descriptor vlan field as written by hardware (no le/be conversion)
  94  * @skb: pointer to sk_buff to be indicated to stack
  95  **/
  96 static void e1000_receive_skb(struct e1000_adapter *adapter,
  97                               struct net_device *netdev,
  98                               struct sk_buff *skb,
  99                               u8 status, __le16 vlan)
 100 {
 101         skb->protocol = eth_type_trans(skb, netdev);
 102
 103         if (adapter->vlgrp && (status & E1000_RXD_STAT_VP))
 104                 vlan_gro_receive(&adapter->napi, adapter->vlgrp,
 105                                  le16_to_cpu(vlan), skb);
 106         else
 107                 napi_gro_receive(&adapter->napi, skb);
 108 }
 109
 110 /**
 111  * e1000_rx_checksum - Receive Checksum Offload for 82543
 112  * @adapter:     board private structure
 113  * @status_err:  receive descriptor status and error fields
 114  * @csum:       receive descriptor csum field
 115  * @sk_buff:     socket buffer with received data
 116  **/
 117 static void e1000_rx_checksum(struct e1000_adapter *adapter, u32 status_err,
 118                               u32 csum, struct sk_buff *skb)
 119 {
 120         u16 status = (u16)status_err;
 121         u8 errors = (u8)(status_err >> 24);
 122         skb->ip_summed = CHECKSUM_NONE;
 123
 124         /* Ignore Checksum bit is set */
 125         if (status & E1000_RXD_STAT_IXSM)
 126                 return;
 127         /* TCP/UDP checksum error bit is set */
 128         if (errors & E1000_RXD_ERR_TCPE) {
 129                 /* let the stack verify checksum errors */
 130                 adapter->hw_csum_err++;
 131                 return;
 132         }
 133
 134         /* TCP/UDP Checksum has not been calculated */
 135         if (!(status & (E1000_RXD_STAT_TCPCS | E1000_RXD_STAT_UDPCS)))
 136                 return;
 137
 138         /* It must be a TCP or UDP packet with a valid checksum */
 139         if (status & E1000_RXD_STAT_TCPCS) {
 140                 /* TCP checksum is good */
 141                 skb->ip_summed = CHECKSUM_UNNECESSARY;
 142         } else {
 143                 /*
 144                  * IP fragment with UDP payload
 145                  * Hardware complements the payload checksum, so we undo it
 146                  * and then put the value in host order for further stack use.
 147                  */
 148                 __sum16 sum = (__force __sum16)htons(csum);
 149                 skb->csum = csum_unfold(~sum);
 150                 skb->ip_summed = CHECKSUM_COMPLETE;
 151         }
 152         adapter->hw_csum_good++;
 153 }
 154
 155 /**
 156  * e1000_alloc_rx_buffers - Replace used receive buffers; legacy & extended
 157  * @adapter: address of board private structure
 158  **/
 159 static void e1000_alloc_rx_buffers(struct e1000_adapter *adapter,
 160                                    int cleaned_count)
 161 {
 162         struct net_device *netdev = adapter->netdev;
 163         struct pci_dev *pdev = adapter->pdev;
 164         struct e1000_ring *rx_ring = adapter->rx_ring;
 165         struct e1000_rx_desc *rx_desc;
 166         struct e1000_buffer *buffer_info;
 167         struct sk_buff *skb;
 168         unsigned int i;
 169         unsigned int bufsz = adapter->rx_buffer_len + NET_IP_ALIGN;
 170
 171         i = rx_ring->next_to_use;
 172         buffer_info = &rx_ring->buffer_info[i];
 173
 174         while (cleaned_count--) {
 175                 skb = buffer_info->skb;
 176                 if (skb) {
 177                         skb_trim(skb, 0);
 178                         goto map_skb;
 179                 }
 180
 181                 skb = netdev_alloc_skb(netdev, bufsz);
 182                 if (!skb) {
 183                         /* Better luck next round */
 184                         adapter->alloc_rx_buff_failed++;
 185                         break;
 186                 }
 187
 188                 /*
 189                  * Make buffer alignment 2 beyond a 16 byte boundary
 190                  * this will result in a 16 byte aligned IP header after
 191                  * the 14 byte MAC header is removed
 192                  */
 193                 skb_reserve(skb, NET_IP_ALIGN);
 194
 195                 buffer_info->skb = skb;
 196 map_skb:
 197                 buffer_info->dma = pci_map_single(pdev, skb->data,
 198                                                   adapter->rx_buffer_len,
 199                                                   PCI_DMA_FROMDEVICE);
 200                 if (pci_dma_mapping_error(pdev, buffer_info->dma)) {
 201                         dev_err(&pdev->dev, "RX DMA map failed\n");
 202                         adapter->rx_dma_failed++;
 203                         break;
 204                 }
 205
 206                 rx_desc = E1000_RX_DESC(*rx_ring, i);
 207                 rx_desc->buffer_addr = cpu_to_le64(buffer_info->dma);
 208
 209                 i++;
 210                 if (i == rx_ring->count)
 211                         i = 0;
 212                 buffer_info = &rx_ring->buffer_info[i];
 213         }
 214
 215         if (rx_ring->next_to_use != i) {
 216                 rx_ring->next_to_use = i;
 217                 if (i-- == 0)
 218                         i = (rx_ring->count - 1);
 219
 220                 /*
 221                  * Force memory writes to complete before letting h/w
 222                  * know there are new descriptors to fetch.  (Only
 223                  * applicable for weak-ordered memory model archs,
 224                  * such as IA-64).
 225                  */
 226                 wmb();
 227                 writel(i, adapter->hw.hw_addr + rx_ring->tail);
 228         }
 229 }
 230
 231 /**
 232  * e1000_alloc_rx_buffers_ps - Replace used receive buffers; packet split
 233  * @adapter: address of board private structure
 234  **/
 235 static void e1000_alloc_rx_buffers_ps(struct e1000_adapter *adapter,
 236                                       int cleaned_count)
 237 {
 238         struct net_device *netdev = adapter->netdev;
 239         struct pci_dev *pdev = adapter->pdev;
 240         union e1000_rx_desc_packet_split *rx_desc;
 241         struct e1000_ring *rx_ring = adapter->rx_ring;
 242         struct e1000_buffer *buffer_info;
 243         struct e1000_ps_page *ps_page;
 244         struct sk_buff *skb;
 245         unsigned int i, j;
 246
 247         i = rx_ring->next_to_use;
 248         buffer_info = &rx_ring->buffer_info[i];
 249
 250         while (cleaned_count--) {
 251                 rx_desc = E1000_RX_DESC_PS(*rx_ring, i);
 252
 253                 for (j = 0; j < PS_PAGE_BUFFERS; j++) {
 254                         ps_page = &buffer_info->ps_pages[j];
 255                         if (j >= adapter->rx_ps_pages) {
 256                                 /* all unused desc entries get hw null ptr */
 257                                 rx_desc->read.buffer_addr[j+1] = ~cpu_to_le64(0);
 258                                 continue;
 259                         }
 260                         if (!ps_page->page) {
 261                                 ps_page->page = alloc_page(GFP_ATOMIC);
 262                                 if (!ps_page->page) {
 263                                         adapter->alloc_rx_buff_failed++;
 264                                         goto no_buffers;
 265                                 }
 266                                 ps_page->dma = pci_map_page(pdev,
 267                                                    ps_page->page,
 268                                                    0, PAGE_SIZE,
 269                                                    PCI_DMA_FROMDEVICE);
 270                                 if (pci_dma_mapping_error(pdev, ps_page->dma)) {
 271                                         dev_err(&adapter->pdev->dev,
 272                                           "RX DMA page map failed\n");
 273                                         adapter->rx_dma_failed++;
 274                                         goto no_buffers;
 275                                 }
 276                         }
 277                         /*
 278                          * Refresh the desc even if buffer_addrs
 279                          * didn't change because each write-back
 280                          * erases this info.
 281                          */
 282                         rx_desc->read.buffer_addr[j+1] =
 283                              cpu_to_le64(ps_page->dma);
 284                 }
 285
 286                 skb = netdev_alloc_skb(netdev,
 287                                        adapter->rx_ps_bsize0 + NET_IP_ALIGN);
 288
 289                 if (!skb) {
 290                         adapter->alloc_rx_buff_failed++;
 291                         break;
 292                 }
 293
 294                 /*
 295                  * Make buffer alignment 2 beyond a 16 byte boundary
 296                  * this will result in a 16 byte aligned IP header after
 297                  * the 14 byte MAC header is removed
 298                  */
 299                 skb_reserve(skb, NET_IP_ALIGN);
 300
 301                 buffer_info->skb = skb;
 302                 buffer_info->dma = pci_map_single(pdev, skb->data,
 303                                                   adapter->rx_ps_bsize0,
 304                                                   PCI_DMA_FROMDEVICE);
 305                 if (pci_dma_mapping_error(pdev, buffer_info->dma)) {
 306                         dev_err(&pdev->dev, "RX DMA map failed\n");
 307                         adapter->rx_dma_failed++;
 308                         /* cleanup skb */
 309                         dev_kfree_skb_any(skb);
 310                         buffer_info->skb = NULL;
 311                         break;
 312                 }
 313
 314                 rx_desc->read.buffer_addr[0] = cpu_to_le64(buffer_info->dma);
 315
 316                 i++;
 317                 if (i == rx_ring->count)
 318                         i = 0;
 319                 buffer_info = &rx_ring->buffer_info[i];
 320         }
 321
 322 no_buffers:
 323         if (rx_ring->next_to_use != i) {
 324                 rx_ring->next_to_use = i;
 325
 326                 if (!(i--))
 327                         i = (rx_ring->count - 1);
 328
 329                 /*
 330                  * Force memory writes to complete before letting h/w
 331                  * know there are new descriptors to fetch.  (Only
 332                  * applicable for weak-ordered memory model archs,
 333                  * such as IA-64).
 334                  */
 335                 wmb();
 336                 /*
 337                  * Hardware increments by 16 bytes, but packet split
 338                  * descriptors are 32 bytes...so we increment tail
 339                  * twice as much.
 340                  */
 341                 writel(i<<1, adapter->hw.hw_addr + rx_ring->tail);
 342         }
 343 }
 344
 345 /**
 346  * e1000_alloc_jumbo_rx_buffers - Replace used jumbo receive buffers
 347  * @adapter: address of board private structure
 348  * @cleaned_count: number of buffers to allocate this pass
 349  **/
 350
 351 static void e1000_alloc_jumbo_rx_buffers(struct e1000_adapter *adapter,
 352                                          int cleaned_count)
 353 {
 354         struct net_device *netdev = adapter->netdev;
 355         struct pci_dev *pdev = adapter->pdev;
 356         struct e1000_rx_desc *rx_desc;
 357         struct e1000_ring *rx_ring = adapter->rx_ring;
 358         struct e1000_buffer *buffer_info;
 359         struct sk_buff *skb;
 360         unsigned int i;
 361         unsigned int bufsz = 256 -
 362                              16 /* for skb_reserve */ -
 363                              NET_IP_ALIGN;
 364
 365         i = rx_ring->next_to_use;
 366         buffer_info = &rx_ring->buffer_info[i];
 367
 368         while (cleaned_count--) {
 369                 skb = buffer_info->skb;
 370                 if (skb) {
 371                         skb_trim(skb, 0);
 372                         goto check_page;
 373                 }
 374
 375                 skb = netdev_alloc_skb(netdev, bufsz);
 376                 if (unlikely(!skb)) {
 377                         /* Better luck next round */
 378                         adapter->alloc_rx_buff_failed++;
 379                         break;
 380                 }
 381
 382                 /* Make buffer alignment 2 beyond a 16 byte boundary
 383                  * this will result in a 16 byte aligned IP header after
 384                  * the 14 byte MAC header is removed
 385                  */
 386                 skb_reserve(skb, NET_IP_ALIGN);
 387
 388                 buffer_info->skb = skb;
 389 check_page:
 390                 /* allocate a new page if necessary */
 391                 if (!buffer_info->page) {
 392                         buffer_info->page = alloc_page(GFP_ATOMIC);
 393                         if (unlikely(!buffer_info->page)) {
 394                                 adapter->alloc_rx_buff_failed++;
 395                                 break;
 396                         }
 397                 }
 398
 399                 if (!buffer_info->dma)
 400                         buffer_info->dma = pci_map_page(pdev,
 401                                                         buffer_info->page, 0,
 402                                                         PAGE_SIZE,
 403                                                         PCI_DMA_FROMDEVICE);
 404
 405                 rx_desc = E1000_RX_DESC(*rx_ring, i);
 406                 rx_desc->buffer_addr = cpu_to_le64(buffer_info->dma);
 407
 408                 if (unlikely(++i == rx_ring->count))
 409                         i = 0;
 410                 buffer_info = &rx_ring->buffer_info[i];
 411         }
 412
 413         if (likely(rx_ring->next_to_use != i)) {
 414                 rx_ring->next_to_use = i;
 415                 if (unlikely(i-- == 0))
 416                         i = (rx_ring->count - 1);
 417
 418                 /* Force memory writes to complete before letting h/w
 419                  * know there are new descriptors to fetch.  (Only
 420                  * applicable for weak-ordered memory model archs,
 421                  * such as IA-64). */
 422                 wmb();
 423                 writel(i, adapter->hw.hw_addr + rx_ring->tail);
 424         }
 425 }
 426
 427 /**
 428  * e1000_clean_rx_irq - Send received data up the network stack; legacy
 429  * @adapter: board private structure
 430  *
 431  * the return value indicates whether actual cleaning was done, there
 432  * is no guarantee that everything was cleaned
 433  **/
 434 static bool e1000_clean_rx_irq(struct e1000_adapter *adapter,
 435                                int *work_done, int work_to_do)
 436 {
 437         struct net_device *netdev = adapter->netdev;
 438         struct pci_dev *pdev = adapter->pdev;
 439         struct e1000_ring *rx_ring = adapter->rx_ring;
 440         struct e1000_rx_desc *rx_desc, *next_rxd;
 441         struct e1000_buffer *buffer_info, *next_buffer;
 442         u32 length;
 443         unsigned int i;
 444         int cleaned_count = 0;
 445         bool cleaned = 0;
 446         unsigned int total_rx_bytes = 0, total_rx_packets = 0;
 447
 448         i = rx_ring->next_to_clean;
 449         rx_desc = E1000_RX_DESC(*rx_ring, i);
 450         buffer_info = &rx_ring->buffer_info[i];
 451
 452         while (rx_desc->status & E1000_RXD_STAT_DD) {
 453                 struct sk_buff *skb;
 454                 u8 status;
 455
 456                 if (*work_done >= work_to_do)
 457                         break;
 458                 (*work_done)++;
 459
 460                 status = rx_desc->status;
 461                 skb = buffer_info->skb;
 462                 buffer_info->skb = NULL;
 463
 464                 prefetch(skb->data - NET_IP_ALIGN);
 465
 466                 i++;
 467                 if (i == rx_ring->count)
 468                         i = 0;
 469                 next_rxd = E1000_RX_DESC(*rx_ring, i);
 470                 prefetch(next_rxd);
 471
 472                 next_buffer = &rx_ring->buffer_info[i];
 473
 474                 cleaned = 1;
 475                 cleaned_count++;
 476                 pci_unmap_single(pdev,
 477                                  buffer_info->dma,
 478                                  adapter->rx_buffer_len,
 479                                  PCI_DMA_FROMDEVICE);
 480                 buffer_info->dma = 0;
 481
 482                 length = le16_to_cpu(rx_desc->length);
 483
 484                 /* !EOP means multiple descriptors were used to store a single
 485                  * packet, also make sure the frame isn't just CRC only */
 486                 if (!(status & E1000_RXD_STAT_EOP) || (length <= 4)) {
 487                         /* All receives must fit into a single buffer */
 488                         e_dbg("%s: Receive packet consumed multiple buffers\n",
 489                               netdev->name);
 490                         /* recycle */
 491                         buffer_info->skb = skb;
 492                         goto next_desc;
 493                 }
 494
 495                 if (rx_desc->errors & E1000_RXD_ERR_FRAME_ERR_MASK) {
 496                         /* recycle */
 497                         buffer_info->skb = skb;
 498                         goto next_desc;
 499                 }
 500
 501                 /* adjust length to remove Ethernet CRC */
 502                 if (!(adapter->flags2 & FLAG2_CRC_STRIPPING))
 503                         length -= 4;
 504
 505                 total_rx_bytes += length;
 506                 total_rx_packets++;
 507
 508                 /*
 509                  * code added for copybreak, this should improve
 510                  * performance for small packets with large amounts
 511                  * of reassembly being done in the stack
 512                  */
 513                 if (length < copybreak) {
 514                         struct sk_buff *new_skb =
 515                             netdev_alloc_skb(netdev, length + NET_IP_ALIGN);
 516                         if (new_skb) {
 517                                 skb_reserve(new_skb, NET_IP_ALIGN);
 518                                 skb_copy_to_linear_data_offset(new_skb,
 519                                                                -NET_IP_ALIGN,
 520                                                                (skb->data -
 521                                                                 NET_IP_ALIGN),
 522                                                                (length +
 523                                                                 NET_IP_ALIGN));
 524                                 /* save the skb in buffer_info as good */
 525                                 buffer_info->skb = skb;
 526                                 skb = new_skb;
 527                         }
 528                         /* else just continue with the old one */
 529                 }
 530                 /* end copybreak code */
 531                 skb_put(skb, length);
 532
 533                 /* Receive Checksum Offload */
 534                 e1000_rx_checksum(adapter,
 535                                   (u32)(status) |
 536                                   ((u32)(rx_desc->errors) << 24),
 537                                   le16_to_cpu(rx_desc->csum), skb);
 538
 539                 e1000_receive_skb(adapter, netdev, skb,status,rx_desc->special);
 540
 541 next_desc:
 542                 rx_desc->status = 0;
 543
 544                 /* return some buffers to hardware, one at a time is too slow */
 545                 if (cleaned_count >= E1000_RX_BUFFER_WRITE) {
 546                         adapter->alloc_rx_buf(adapter, cleaned_count);
 547                         cleaned_count = 0;
 548                 }
 549
 550                 /* use prefetched values */
 551                 rx_desc = next_rxd;
 552                 buffer_info = next_buffer;
 553         }
 554         rx_ring->next_to_clean = i;
 555
 556         cleaned_count = e1000_desc_unused(rx_ring);
 557         if (cleaned_count)
 558                 adapter->alloc_rx_buf(adapter, cleaned_count);
 559
 560         adapter->total_rx_bytes += total_rx_bytes;
 561         adapter->total_rx_packets += total_rx_packets;
 562         adapter->net_stats.rx_bytes += total_rx_bytes;
 563         adapter->net_stats.rx_packets += total_rx_packets;
 564         return cleaned;
 565 }
 566
 567 static void e1000_put_txbuf(struct e1000_adapter *adapter,
 568                              struct e1000_buffer *buffer_info)
 569 {
 570         buffer_info->dma = 0;
 571         if (buffer_info->skb) {
 572                 skb_dma_unmap(&adapter->pdev->dev, buffer_info->skb,
 573                               DMA_TO_DEVICE);
 574                 dev_kfree_skb_any(buffer_info->skb);
 575                 buffer_info->skb = NULL;
 576         }
 577         buffer_info->time_stamp = 0;
 578 }
 579
 580 static void e1000_print_tx_hang(struct e1000_adapter *adapter)
 581 {
 582         struct e1000_ring *tx_ring = adapter->tx_ring;
 583         unsigned int i = tx_ring->next_to_clean;
 584         unsigned int eop = tx_ring->buffer_info[i].next_to_watch;
 585         struct e1000_tx_desc *eop_desc = E1000_TX_DESC(*tx_ring, eop);
 586
 587         /* detected Tx unit hang */
 588         e_err("Detected Tx Unit Hang:\n"
 589               "  TDH                  <%x>\n"
 590               "  TDT                  <%x>\n"
 591               "  next_to_use          <%x>\n"
 592               "  next_to_clean        <%x>\n"
 593               "buffer_info[next_to_clean]:\n"
 594               "  time_stamp           <%lx>\n"
 595               "  next_to_watch        <%x>\n"
 596               "  jiffies              <%lx>\n"
 597               "  next_to_watch.status <%x>\n",
 598               readl(adapter->hw.hw_addr + tx_ring->head),
 599               readl(adapter->hw.hw_addr + tx_ring->tail),
 600               tx_ring->next_to_use,
 601               tx_ring->next_to_clean,
 602               tx_ring->buffer_info[eop].time_stamp,
 603               eop,
 604               jiffies,
 605               eop_desc->upper.fields.status);
 606 }
 607
 608 /**
 609  * e1000_clean_tx_irq - Reclaim resources after transmit completes
 610  * @adapter: board private structure
 611  *
 612  * the return value indicates whether actual cleaning was done, there
 613  * is no guarantee that everything was cleaned
 614  **/
 615 static bool e1000_clean_tx_irq(struct e1000_adapter *adapter)
 616 {
 617         struct net_device *netdev = adapter->netdev;
 618         struct e1000_hw *hw = &adapter->hw;
 619         struct e1000_ring *tx_ring = adapter->tx_ring;
 620         struct e1000_tx_desc *tx_desc, *eop_desc;
 621         struct e1000_buffer *buffer_info;
 622         unsigned int i, eop;
 623         unsigned int count = 0;
 624         unsigned int total_tx_bytes = 0, total_tx_packets = 0;
 625
 626         i = tx_ring->next_to_clean;
 627         eop = tx_ring->buffer_info[i].next_to_watch;
 628         eop_desc = E1000_TX_DESC(*tx_ring, eop);
 629
 630         while ((eop_desc->upper.data & cpu_to_le32(E1000_TXD_STAT_DD)) &&
 631                (count < tx_ring->count)) {
 632                 bool cleaned = false;
 633                 for (; !cleaned; count++) {
 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         }
 660
 661         tx_ring->next_to_clean = i;
 662
 663 #define TX_WAKE_THRESHOLD 32
 664         if (count && netif_carrier_ok(netdev) &&
 665             e1000_desc_unused(tx_ring) >= TX_WAKE_THRESHOLD) {
 666                 /* Make sure that anybody stopping the queue after this
 667                  * sees the new next_to_clean.
 668                  */
 669                 smp_mb();
 670
 671                 if (netif_queue_stopped(netdev) &&
 672                     !(test_bit(__E1000_DOWN, &adapter->state))) {
 673                         netif_wake_queue(netdev);
 674                         ++adapter->restart_queue;
 675                 }
 676         }
 677
 678         if (adapter->detect_tx_hung) {
 679                 /* Detect a transmit hang in hardware, this serializes the
 680                  * check with the clearing of time_stamp and movement of i */
 681                 adapter->detect_tx_hung = 0;
 682                 if (tx_ring->buffer_info[i].time_stamp &&
 683                     time_after(jiffies, tx_ring->buffer_info[i].time_stamp
 684                                + (adapter->tx_timeout_factor * HZ))
 685                     && !(er32(STATUS) & E1000_STATUS_TXOFF)) {
 686                         e1000_print_tx_hang(adapter);
 687                         netif_stop_queue(netdev);
 688                 }
 689         }
 690         adapter->total_tx_bytes += total_tx_bytes;
 691         adapter->total_tx_packets += total_tx_packets;
 692         adapter->net_stats.tx_bytes += total_tx_bytes;
 693         adapter->net_stats.tx_packets += total_tx_packets;
 694         return (count < tx_ring->count);
 695 }
 696
 697 /**
 698  * e1000_clean_rx_irq_ps - Send received data up the network stack; packet split
 699  * @adapter: board private structure
 700  *
 701  * the return value indicates whether actual cleaning was done, there
 702  * is no guarantee that everything was cleaned
 703  **/
 704 static bool e1000_clean_rx_irq_ps(struct e1000_adapter *adapter,
 705                                   int *work_done, int work_to_do)
 706 {
 707         union e1000_rx_desc_packet_split *rx_desc, *next_rxd;
 708         struct net_device *netdev = adapter->netdev;
 709         struct pci_dev *pdev = adapter->pdev;
 710         struct e1000_ring *rx_ring = adapter->rx_ring;
 711         struct e1000_buffer *buffer_info, *next_buffer;
 712         struct e1000_ps_page *ps_page;
 713         struct sk_buff *skb;
 714         unsigned int i, j;
 715         u32 length, staterr;
 716         int cleaned_count = 0;
 717         bool cleaned = 0;
 718         unsigned int total_rx_bytes = 0, total_rx_packets = 0;
 719
 720         i = rx_ring->next_to_clean;
 721         rx_desc = E1000_RX_DESC_PS(*rx_ring, i);
 722         staterr = le32_to_cpu(rx_desc->wb.middle.status_error);
 723         buffer_info = &rx_ring->buffer_info[i];
 724
 725         while (staterr & E1000_RXD_STAT_DD) {
 726                 if (*work_done >= work_to_do)
 727                         break;
 728                 (*work_done)++;
 729                 skb = buffer_info->skb;
 730
 731                 /* in the packet split case this is header only */
 732                 prefetch(skb->data - NET_IP_ALIGN);
 733
 734                 i++;
 735                 if (i == rx_ring->count)
 736                         i = 0;
 737                 next_rxd = E1000_RX_DESC_PS(*rx_ring, i);
 738                 prefetch(next_rxd);
 739
 740                 next_buffer = &rx_ring->buffer_info[i];
 741
 742                 cleaned = 1;
 743                 cleaned_count++;
 744                 pci_unmap_single(pdev, buffer_info->dma,
 745                                  adapter->rx_ps_bsize0,
 746                                  PCI_DMA_FROMDEVICE);
 747                 buffer_info->dma = 0;
 748
 749                 if (!(staterr & E1000_RXD_STAT_EOP)) {
 750                         e_dbg("%s: Packet Split buffers didn't pick up the "
 751                               "full packet\n", netdev->name);
 752                         dev_kfree_skb_irq(skb);
 753                         goto next_desc;
 754                 }
 755
 756                 if (staterr & E1000_RXDEXT_ERR_FRAME_ERR_MASK) {
 757                         dev_kfree_skb_irq(skb);
 758                         goto next_desc;
 759                 }
 760
 761                 length = le16_to_cpu(rx_desc->wb.middle.length0);
 762
 763                 if (!length) {
 764                         e_dbg("%s: Last part of the packet spanning multiple "
 765                               "descriptors\n", netdev->name);
 766                         dev_kfree_skb_irq(skb);
 767                         goto next_desc;
 768                 }
 769
 770                 /* Good Receive */
 771                 skb_put(skb, length);
 772
 773                 {
 774                 /*
 775                  * this looks ugly, but it seems compiler issues make it
 776                  * more efficient than reusing j
 777                  */
 778                 int l1 = le16_to_cpu(rx_desc->wb.upper.length[0]);
 779
 780                 /*
 781                  * page alloc/put takes too long and effects small packet
 782                  * throughput, so unsplit small packets and save the alloc/put
 783                  * only valid in softirq (napi) context to call kmap_*
 784                  */
 785                 if (l1 && (l1 <= copybreak) &&
 786                     ((length + l1) <= adapter->rx_ps_bsize0)) {
 787                         u8 *vaddr;
 788
 789                         ps_page = &buffer_info->ps_pages[0];
 790
 791                         /*
 792                          * there is no documentation about how to call
 793                          * kmap_atomic, so we can't hold the mapping
 794                          * very long
 795                          */
 796                         pci_dma_sync_single_for_cpu(pdev, ps_page->dma,
 797                                 PAGE_SIZE, PCI_DMA_FROMDEVICE);
 798                         vaddr = kmap_atomic(ps_page->page, KM_SKB_DATA_SOFTIRQ);
 799                         memcpy(skb_tail_pointer(skb), vaddr, l1);
 800                         kunmap_atomic(vaddr, KM_SKB_DATA_SOFTIRQ);
 801                         pci_dma_sync_single_for_device(pdev, ps_page->dma,
 802                                 PAGE_SIZE, PCI_DMA_FROMDEVICE);
 803
 804                         /* remove the CRC */
 805                         if (!(adapter->flags2 & FLAG2_CRC_STRIPPING))
 806                                 l1 -= 4;
 807
 808                         skb_put(skb, l1);
 809                         goto copydone;
 810                 } /* if */
 811                 }
 812
 813                 for (j = 0; j < PS_PAGE_BUFFERS; j++) {
 814                         length = le16_to_cpu(rx_desc->wb.upper.length[j]);
 815                         if (!length)
 816                                 break;
 817
 818                         ps_page = &buffer_info->ps_pages[j];
 819                         pci_unmap_page(pdev, ps_page->dma, PAGE_SIZE,
 820                                        PCI_DMA_FROMDEVICE);
 821                         ps_page->dma = 0;
 822                         skb_fill_page_desc(skb, j, ps_page->page, 0, length);
 823                         ps_page->page = NULL;
 824                         skb->len += length;
 825                         skb->data_len += length;
 826                         skb->truesize += length;
 827                 }
 828
 829                 /* strip the ethernet crc, problem is we're using pages now so
 830                  * this whole operation can get a little cpu intensive
 831                  */
 832                 if (!(adapter->flags2 & FLAG2_CRC_STRIPPING))
 833                         pskb_trim(skb, skb->len - 4);
 834
 835 copydone:
 836                 total_rx_bytes += skb->len;
 837                 total_rx_packets++;
 838
 839                 e1000_rx_checksum(adapter, staterr, le16_to_cpu(
 840                         rx_desc->wb.lower.hi_dword.csum_ip.csum), skb);
 841
 842                 if (rx_desc->wb.upper.header_status &
 843                            cpu_to_le16(E1000_RXDPS_HDRSTAT_HDRSP))
 844                         adapter->rx_hdr_split++;
 845
 846                 e1000_receive_skb(adapter, netdev, skb,
 847                                   staterr, rx_desc->wb.middle.vlan);
 848
 849 next_desc:
 850                 rx_desc->wb.middle.status_error &= cpu_to_le32(~0xFF);
 851                 buffer_info->skb = NULL;
 852
 853                 /* return some buffers to hardware, one at a time is too slow */
 854                 if (cleaned_count >= E1000_RX_BUFFER_WRITE) {
 855                         adapter->alloc_rx_buf(adapter, cleaned_count);
 856                         cleaned_count = 0;
 857                 }
 858
 859                 /* use prefetched values */
 860                 rx_desc = next_rxd;
 861                 buffer_info = next_buffer;
 862
 863                 staterr = le32_to_cpu(rx_desc->wb.middle.status_error);
 864         }
 865         rx_ring->next_to_clean = i;
 866
 867         cleaned_count = e1000_desc_unused(rx_ring);
 868         if (cleaned_count)
 869                 adapter->alloc_rx_buf(adapter, cleaned_count);
 870
 871         adapter->total_rx_bytes += total_rx_bytes;
 872         adapter->total_rx_packets += total_rx_packets;
 873         adapter->net_stats.rx_bytes += total_rx_bytes;
 874         adapter->net_stats.rx_packets += total_rx_packets;
 875         return cleaned;
 876 }
 877
 878 /**
 879  * e1000_consume_page - helper function
 880  **/
 881 static void e1000_consume_page(struct e1000_buffer *bi, struct sk_buff *skb,
 882                                u16 length)
 883 {
 884         bi->page = NULL;
 885         skb->len += length;
 886         skb->data_len += length;
 887         skb->truesize += length;
 888 }
 889
 890 /**
 891  * e1000_clean_jumbo_rx_irq - Send received data up the network stack; legacy
 892  * @adapter: board private structure
 893  *
 894  * the return value indicates whether actual cleaning was done, there
 895  * is no guarantee that everything was cleaned
 896  **/
 897
 898 static bool e1000_clean_jumbo_rx_irq(struct e1000_adapter *adapter,
 899                                      int *work_done, int work_to_do)
 900 {
 901         struct net_device *netdev = adapter->netdev;
 902         struct pci_dev *pdev = adapter->pdev;
 903         struct e1000_ring *rx_ring = adapter->rx_ring;
 904         struct e1000_rx_desc *rx_desc, *next_rxd;
 905         struct e1000_buffer *buffer_info, *next_buffer;
 906         u32 length;
 907         unsigned int i;
 908         int cleaned_count = 0;
 909         bool cleaned = false;
 910         unsigned int total_rx_bytes=0, total_rx_packets=0;
 911
 912         i = rx_ring->next_to_clean;
 913         rx_desc = E1000_RX_DESC(*rx_ring, i);
 914         buffer_info = &rx_ring->buffer_info[i];
 915
 916         while (rx_desc->status & E1000_RXD_STAT_DD) {
 917                 struct sk_buff *skb;
 918                 u8 status;
 919
 920                 if (*work_done >= work_to_do)
 921                         break;
 922                 (*work_done)++;
 923
 924                 status = rx_desc->status;
 925                 skb = buffer_info->skb;
 926                 buffer_info->skb = NULL;
 927
 928                 ++i;
 929                 if (i == rx_ring->count)
 930                         i = 0;
 931                 next_rxd = E1000_RX_DESC(*rx_ring, i);
 932                 prefetch(next_rxd);
 933
 934                 next_buffer = &rx_ring->buffer_info[i];
 935
 936                 cleaned = true;
 937                 cleaned_count++;
 938                 pci_unmap_page(pdev, buffer_info->dma, PAGE_SIZE,
 939                                PCI_DMA_FROMDEVICE);
 940                 buffer_info->dma = 0;
 941
 942                 length = le16_to_cpu(rx_desc->length);
 943
 944                 /* errors is only valid for DD + EOP descriptors */
 945                 if (unlikely((status & E1000_RXD_STAT_EOP) &&
 946                     (rx_desc->errors & E1000_RXD_ERR_FRAME_ERR_MASK))) {
 947                                 /* recycle both page and skb */
 948                                 buffer_info->skb = skb;
 949                                 /* an error means any chain goes out the window
 950                                  * too */
 951                                 if (rx_ring->rx_skb_top)
 952                                         dev_kfree_skb(rx_ring->rx_skb_top);
 953                                 rx_ring->rx_skb_top = NULL;
 954                                 goto next_desc;
 955                 }
 956
 957 #define rxtop rx_ring->rx_skb_top
 958                 if (!(status & E1000_RXD_STAT_EOP)) {
 959                         /* this descriptor is only the beginning (or middle) */
 960                         if (!rxtop) {
 961                                 /* this is the beginning of a chain */
 962                                 rxtop = skb;
 963                                 skb_fill_page_desc(rxtop, 0, buffer_info->page,
 964                                                    0, length);
 965                         } else {
 966                                 /* this is the middle of a chain */
 967                                 skb_fill_page_desc(rxtop,
 968                                     skb_shinfo(rxtop)->nr_frags,
 969                                     buffer_info->page, 0, length);
 970                                 /* re-use the skb, only consumed the page */
 971                                 buffer_info->skb = skb;
 972                         }
 973                         e1000_consume_page(buffer_info, rxtop, length);
 974                         goto next_desc;
 975                 } else {
 976                         if (rxtop) {
 977                                 /* end of the chain */
 978                                 skb_fill_page_desc(rxtop,
 979                                     skb_shinfo(rxtop)->nr_frags,
 980                                     buffer_info->page, 0, length);
 981                                 /* re-use the current skb, we only consumed the
 982                                  * page */
 983                                 buffer_info->skb = skb;
 984                                 skb = rxtop;
 985                                 rxtop = NULL;
 986                                 e1000_consume_page(buffer_info, skb, length);
 987                         } else {
 988                                 /* no chain, got EOP, this buf is the packet
 989                                  * copybreak to save the put_page/alloc_page */
 990                                 if (length <= copybreak &&
 991                                     skb_tailroom(skb) >= length) {
 992                                         u8 *vaddr;
 993                                         vaddr = kmap_atomic(buffer_info->page,
 994                                                            KM_SKB_DATA_SOFTIRQ);
 995                                         memcpy(skb_tail_pointer(skb), vaddr,
 996                                                length);
 997                                         kunmap_atomic(vaddr,
 998                                                       KM_SKB_DATA_SOFTIRQ);
 999                                         /* re-use the page, so don't erase
1000                                          * buffer_info->page */
1001                                         skb_put(skb, length);
1002                                 } else {
1003                                         skb_fill_page_desc(skb, 0,
1004                                                            buffer_info->page, 0,
1005                                                            length);
1006                                         e1000_consume_page(buffer_info, skb,
1007                                                            length);
1008                                 }
1009                         }
1010                 }
1011
1012                 /* Receive Checksum Offload XXX recompute due to CRC strip? */
1013                 e1000_rx_checksum(adapter,
1014                                   (u32)(status) |
1015                                   ((u32)(rx_desc->errors) << 24),
1016                                   le16_to_cpu(rx_desc->csum), skb);
1017
1018                 /* probably a little skewed due to removing CRC */
1019                 total_rx_bytes += skb->len;
1020                 total_rx_packets++;
1021
1022                 /* eth type trans needs skb->data to point to something */
1023                 if (!pskb_may_pull(skb, ETH_HLEN)) {
1024                         e_err("pskb_may_pull failed.\n");
1025                         dev_kfree_skb(skb);
1026                         goto next_desc;
1027                 }
1028
1029                 e1000_receive_skb(adapter, netdev, skb, status,
1030                                   rx_desc->special);
1031
1032 next_desc:
1033                 rx_desc->status = 0;
1034
1035                 /* return some buffers to hardware, one at a time is too slow */
1036                 if (unlikely(cleaned_count >= E1000_RX_BUFFER_WRITE)) {
1037                         adapter->alloc_rx_buf(adapter, cleaned_count);
1038                         cleaned_count = 0;
1039                 }
1040
1041                 /* use prefetched values */
1042                 rx_desc = next_rxd;
1043                 buffer_info = next_buffer;
1044         }
1045         rx_ring->next_to_clean = i;
1046
1047         cleaned_count = e1000_desc_unused(rx_ring);
1048         if (cleaned_count)
1049                 adapter->alloc_rx_buf(adapter, cleaned_count);
1050
1051         adapter->total_rx_bytes += total_rx_bytes;
1052         adapter->total_rx_packets += total_rx_packets;
1053         adapter->net_stats.rx_bytes += total_rx_bytes;
1054         adapter->net_stats.rx_packets += total_rx_packets;
1055         return cleaned;
1056 }
1057
1058 /**
1059  * e1000_clean_rx_ring - Free Rx Buffers per Queue
1060  * @adapter: board private structure
1061  **/
1062 static void e1000_clean_rx_ring(struct e1000_adapter *adapter)
1063 {
1064         struct e1000_ring *rx_ring = adapter->rx_ring;
1065         struct e1000_buffer *buffer_info;
1066         struct e1000_ps_page *ps_page;
1067         struct pci_dev *pdev = adapter->pdev;
1068         unsigned int i, j;
1069
1070         /* Free all the Rx ring sk_buffs */
1071         for (i = 0; i < rx_ring->count; i++) {
1072                 buffer_info = &rx_ring->buffer_info[i];
1073                 if (buffer_info->dma) {
1074                         if (adapter->clean_rx == e1000_clean_rx_irq)
1075                                 pci_unmap_single(pdev, buffer_info->dma,
1076                                                  adapter->rx_buffer_len,
1077                                                  PCI_DMA_FROMDEVICE);
1078                         else if (adapter->clean_rx == e1000_clean_jumbo_rx_irq)
1079                                 pci_unmap_page(pdev, buffer_info->dma,
1080                                                PAGE_SIZE,
1081                                                PCI_DMA_FROMDEVICE);
1082                         else if (adapter->clean_rx == e1000_clean_rx_irq_ps)
1083                                 pci_unmap_single(pdev, buffer_info->dma,
1084                                                  adapter->rx_ps_bsize0,
1085                                                  PCI_DMA_FROMDEVICE);
1086                         buffer_info->dma = 0;
1087                 }
1088
1089                 if (buffer_info->page) {
1090                         put_page(buffer_info->page);
1091                         buffer_info->page = NULL;
1092                 }
1093
1094                 if (buffer_info->skb) {
1095                         dev_kfree_skb(buffer_info->skb);
1096                         buffer_info->skb = NULL;
1097                 }
1098
1099                 for (j = 0; j < PS_PAGE_BUFFERS; j++) {
1100                         ps_page = &buffer_info->ps_pages[j];
1101                         if (!ps_page->page)
1102                                 break;
1103                         pci_unmap_page(pdev, ps_page->dma, PAGE_SIZE,
1104                                        PCI_DMA_FROMDEVICE);
1105                         ps_page->dma = 0;
1106                         put_page(ps_page->page);
1107                         ps_page->page = NULL;
1108                 }
1109         }
1110
1111         /* there also may be some cached data from a chained receive */
1112         if (rx_ring->rx_skb_top) {
1113                 dev_kfree_skb(rx_ring->rx_skb_top);
1114                 rx_ring->rx_skb_top = NULL;
1115         }
1116
1117         /* Zero out the descriptor ring */
1118         memset(rx_ring->desc, 0, rx_ring->size);
1119
1120         rx_ring->next_to_clean = 0;
1121         rx_ring->next_to_use = 0;
1122
1123         writel(0, adapter->hw.hw_addr + rx_ring->head);
1124         writel(0, adapter->hw.hw_addr + rx_ring->tail);
1125 }
1126
1127 static void e1000e_downshift_workaround(struct work_struct *work)
1128 {
1129         struct e1000_adapter *adapter = container_of(work,
1130                                         struct e1000_adapter, downshift_task);
1131
1132         e1000e_gig_downshift_workaround_ich8lan(&adapter->hw);
1133 }
1134
1135 /**
1136  * e1000_intr_msi - Interrupt Handler
1137  * @irq: interrupt number
1138  * @data: pointer to a network interface device structure
1139  **/
1140 static irqreturn_t e1000_intr_msi(int irq, void *data)
1141 {
1142         struct net_device *netdev = data;
1143         struct e1000_adapter *adapter = netdev_priv(netdev);
1144         struct e1000_hw *hw = &adapter->hw;
1145         u32 icr = er32(ICR);
1146
1147         /*
1148          * read ICR disables interrupts using IAM
1149          */
1150
1151         if (icr & E1000_ICR_LSC) {
1152                 hw->mac.get_link_status = 1;
1153                 /*
1154                  * ICH8 workaround-- Call gig speed drop workaround on cable
1155                  * disconnect (LSC) before accessing any PHY registers
1156                  */
1157                 if ((adapter->flags & FLAG_LSC_GIG_SPEED_DROP) &&
1158                     (!(er32(STATUS) & E1000_STATUS_LU)))
1159                         schedule_work(&adapter->downshift_task);
1160
1161                 /*
1162                  * 80003ES2LAN workaround-- For packet buffer work-around on
1163                  * link down event; disable receives here in the ISR and reset
1164                  * adapter in watchdog
1165                  */
1166                 if (netif_carrier_ok(netdev) &&
1167                     adapter->flags & FLAG_RX_NEEDS_RESTART) {
1168                         /* disable receives */
1169                         u32 rctl = er32(RCTL);
1170                         ew32(RCTL, rctl & ~E1000_RCTL_EN);
1171                         adapter->flags |= FLAG_RX_RESTART_NOW;
1172                 }
1173                 /* guard against interrupt when we're going down */
1174                 if (!test_bit(__E1000_DOWN, &adapter->state))
1175                         mod_timer(&adapter->watchdog_timer, jiffies + 1);
1176         }
1177
1178         if (napi_schedule_prep(&adapter->napi)) {
1179                 adapter->total_tx_bytes = 0;
1180                 adapter->total_tx_packets = 0;
1181                 adapter->total_rx_bytes = 0;
1182                 adapter->total_rx_packets = 0;
1183                 __napi_schedule(&adapter->napi);
1184         }
1185
1186         return IRQ_HANDLED;
1187 }
1188
1189 /**
1190  * e1000_intr - Interrupt Handler
1191  * @irq: interrupt number
1192  * @data: pointer to a network interface device structure
1193  **/
1194 static irqreturn_t e1000_intr(int irq, void *data)
1195 {
1196         struct net_device *netdev = data;
1197         struct e1000_adapter *adapter = netdev_priv(netdev);
1198         struct e1000_hw *hw = &adapter->hw;
1199         u32 rctl, icr = er32(ICR);
1200
1201         if (!icr)
1202                 return IRQ_NONE;  /* Not our interrupt */
1203
1204         /*
1205          * IMS will not auto-mask if INT_ASSERTED is not set, and if it is
1206          * not set, then the adapter didn't send an interrupt
1207          */
1208         if (!(icr & E1000_ICR_INT_ASSERTED))
1209                 return IRQ_NONE;
1210
1211         /*
1212          * Interrupt Auto-Mask...upon reading ICR,
1213          * interrupts are masked.  No need for the
1214          * IMC write
1215          */
1216
1217         if (icr & E1000_ICR_LSC) {
1218                 hw->mac.get_link_status = 1;
1219                 /*
1220                  * ICH8 workaround-- Call gig speed drop workaround on cable
1221                  * disconnect (LSC) before accessing any PHY registers
1222                  */
1223                 if ((adapter->flags & FLAG_LSC_GIG_SPEED_DROP) &&
1224                     (!(er32(STATUS) & E1000_STATUS_LU)))
1225                         schedule_work(&adapter->downshift_task);
1226
1227                 /*
1228                  * 80003ES2LAN workaround--
1229                  * For packet buffer work-around on link down event;
1230                  * disable receives here in the ISR and
1231                  * reset adapter in watchdog
1232                  */
1233                 if (netif_carrier_ok(netdev) &&
1234                     (adapter->flags & FLAG_RX_NEEDS_RESTART)) {
1235                         /* disable receives */
1236                         rctl = er32(RCTL);
1237                         ew32(RCTL, rctl & ~E1000_RCTL_EN);
1238                         adapter->flags |= FLAG_RX_RESTART_NOW;
1239                 }
1240                 /* guard against interrupt when we're going down */
1241                 if (!test_bit(__E1000_DOWN, &adapter->state))
1242                         mod_timer(&adapter->watchdog_timer, jiffies + 1);
1243         }
1244
1245         if (napi_schedule_prep(&adapter->napi)) {
1246                 adapter->total_tx_bytes = 0;
1247                 adapter->total_tx_packets = 0;
1248                 adapter->total_rx_bytes = 0;
1249                 adapter->total_rx_packets = 0;
1250                 __napi_schedule(&adapter->napi);
1251         }
1252
1253         return IRQ_HANDLED;
1254 }
1255
1256 static irqreturn_t e1000_msix_other(int irq, void *data)
1257 {
1258         struct net_device *netdev = data;
1259         struct e1000_adapter *adapter = netdev_priv(netdev);
1260         struct e1000_hw *hw = &adapter->hw;
1261         u32 icr = er32(ICR);
1262
1263         if (!(icr & E1000_ICR_INT_ASSERTED)) {
1264                 if (!test_bit(__E1000_DOWN, &adapter->state))
1265                         ew32(IMS, E1000_IMS_OTHER);
1266                 return IRQ_NONE;
1267         }
1268
1269         if (icr & adapter->eiac_mask)
1270                 ew32(ICS, (icr & adapter->eiac_mask));
1271
1272         if (icr & E1000_ICR_OTHER) {
1273                 if (!(icr & E1000_ICR_LSC))
1274                         goto no_link_interrupt;
1275                 hw->mac.get_link_status = 1;
1276                 /* guard against interrupt when we're going down */
1277                 if (!test_bit(__E1000_DOWN, &adapter->state))
1278                         mod_timer(&adapter->watchdog_timer, jiffies + 1);
1279         }
1280
1281 no_link_interrupt:
1282         if (!test_bit(__E1000_DOWN, &adapter->state))
1283                 ew32(IMS, E1000_IMS_LSC | E1000_IMS_OTHER);
1284
1285         return IRQ_HANDLED;
1286 }
1287
1288
1289 static irqreturn_t e1000_intr_msix_tx(int irq, void *data)
1290 {
1291         struct net_device *netdev = data;
1292         struct e1000_adapter *adapter = netdev_priv(netdev);
1293         struct e1000_hw *hw = &adapter->hw;
1294         struct e1000_ring *tx_ring = adapter->tx_ring;
1295
1296
1297         adapter->total_tx_bytes = 0;
1298         adapter->total_tx_packets = 0;
1299
1300         if (!e1000_clean_tx_irq(adapter))
1301                 /* Ring was not completely cleaned, so fire another interrupt */
1302                 ew32(ICS, tx_ring->ims_val);
1303
1304         return IRQ_HANDLED;
1305 }
1306
1307 static irqreturn_t e1000_intr_msix_rx(int irq, void *data)
1308 {
1309         struct net_device *netdev = data;
1310         struct e1000_adapter *adapter = netdev_priv(netdev);
1311
1312         /* Write the ITR value calculated at the end of the
1313          * previous interrupt.
1314          */
1315         if (adapter->rx_ring->set_itr) {
1316                 writel(1000000000 / (adapter->rx_ring->itr_val * 256),
1317                        adapter->hw.hw_addr + adapter->rx_ring->itr_register);
1318                 adapter->rx_ring->set_itr = 0;
1319         }
1320
1321         if (napi_schedule_prep(&adapter->napi)) {
1322                 adapter->total_rx_bytes = 0;
1323                 adapter->total_rx_packets = 0;
1324                 __napi_schedule(&adapter->napi);
1325         }
1326         return IRQ_HANDLED;
1327 }
1328
1329 /**
1330  * e1000_configure_msix - Configure MSI-X hardware
1331  *
1332  * e1000_configure_msix sets up the hardware to properly
1333  * generate MSI-X interrupts.
1334  **/
1335 static void e1000_configure_msix(struct e1000_adapter *adapter)
1336 {
1337         struct e1000_hw *hw = &adapter->hw;
1338         struct e1000_ring *rx_ring = adapter->rx_ring;
1339         struct e1000_ring *tx_ring = adapter->tx_ring;
1340         int vector = 0;
1341         u32 ctrl_ext, ivar = 0;
1342
1343         adapter->eiac_mask = 0;
1344
1345         /* Workaround issue with spurious interrupts on 82574 in MSI-X mode */
1346         if (hw->mac.type == e1000_82574) {
1347                 u32 rfctl = er32(RFCTL);
1348                 rfctl |= E1000_RFCTL_ACK_DIS;
1349                 ew32(RFCTL, rfctl);
1350         }
1351
1352 #define E1000_IVAR_INT_ALLOC_VALID      0x8
1353         /* Configure Rx vector */
1354         rx_ring->ims_val = E1000_IMS_RXQ0;
1355         adapter->eiac_mask |= rx_ring->ims_val;
1356         if (rx_ring->itr_val)
1357                 writel(1000000000 / (rx_ring->itr_val * 256),
1358                        hw->hw_addr + rx_ring->itr_register);
1359         else
1360                 writel(1, hw->hw_addr + rx_ring->itr_register);
1361         ivar = E1000_IVAR_INT_ALLOC_VALID | vector;
1362
1363         /* Configure Tx vector */
1364         tx_ring->ims_val = E1000_IMS_TXQ0;
1365         vector++;
1366         if (tx_ring->itr_val)
1367                 writel(1000000000 / (tx_ring->itr_val * 256),
1368                        hw->hw_addr + tx_ring->itr_register);
1369         else
1370                 writel(1, hw->hw_addr + tx_ring->itr_register);
1371         adapter->eiac_mask |= tx_ring->ims_val;
1372         ivar |= ((E1000_IVAR_INT_ALLOC_VALID | vector) << 8);
1373
1374         /* set vector for Other Causes, e.g. link changes */
1375         vector++;
1376         ivar |= ((E1000_IVAR_INT_ALLOC_VALID | vector) << 16);
1377         if (rx_ring->itr_val)
1378                 writel(1000000000 / (rx_ring->itr_val * 256),
1379                        hw->hw_addr + E1000_EITR_82574(vector));
1380         else
1381                 writel(1, hw->hw_addr + E1000_EITR_82574(vector));
1382
1383         /* Cause Tx interrupts on every write back */
1384         ivar |= (1 << 31);
1385
1386         ew32(IVAR, ivar);
1387
1388         /* enable MSI-X PBA support */
1389         ctrl_ext = er32(CTRL_EXT);
1390         ctrl_ext |= E1000_CTRL_EXT_PBA_CLR;
1391
1392         /* Auto-Mask Other interrupts upon ICR read */
1393 #define E1000_EIAC_MASK_82574   0x01F00000
1394         ew32(IAM, ~E1000_EIAC_MASK_82574 | E1000_IMS_OTHER);
1395         ctrl_ext |= E1000_CTRL_EXT_EIAME;
1396         ew32(CTRL_EXT, ctrl_ext);
1397         e1e_flush();
1398 }
1399
1400 void e1000e_reset_interrupt_capability(struct e1000_adapter *adapter)
1401 {
1402         if (adapter->msix_entries) {
1403                 pci_disable_msix(adapter->pdev);
1404                 kfree(adapter->msix_entries);
1405                 adapter->msix_entries = NULL;
1406         } else if (adapter->flags & FLAG_MSI_ENABLED) {
1407                 pci_disable_msi(adapter->pdev);
1408                 adapter->flags &= ~FLAG_MSI_ENABLED;
1409         }
1410
1411         return;
1412 }
1413
1414 /**
1415  * e1000e_set_interrupt_capability - set MSI or MSI-X if supported
1416  *
1417  * Attempt to configure interrupts using the best available
1418  * capabilities of the hardware and kernel.
1419  **/
1420 void e1000e_set_interrupt_capability(struct e1000_adapter *adapter)
1421 {
1422         int err;
1423         int numvecs, i;
1424
1425
1426         switch (adapter->int_mode) {
1427         case E1000E_INT_MODE_MSIX:
1428                 if (adapter->flags & FLAG_HAS_MSIX) {
1429                         numvecs = 3; /* RxQ0, TxQ0 and other */
1430                         adapter->msix_entries = kcalloc(numvecs,
1431                                                       sizeof(struct msix_entry),
1432                                                       GFP_KERNEL);
1433                         if (adapter->msix_entries) {
1434                                 for (i = 0; i < numvecs; i++)
1435                                         adapter->msix_entries[i].entry = i;
1436
1437                                 err = pci_enable_msix(adapter->pdev,
1438                                                       adapter->msix_entries,
1439                                                       numvecs);
1440                                 if (err == 0)
1441                                         return;
1442                         }
1443                         /* MSI-X failed, so fall through and try MSI */
1444                         e_err("Failed to initialize MSI-X interrupts.  "
1445                               "Falling back to MSI interrupts.\n");
1446                         e1000e_reset_interrupt_capability(adapter);
1447                 }
1448                 adapter->int_mode = E1000E_INT_MODE_MSI;
1449                 /* Fall through */
1450         case E1000E_INT_MODE_MSI:
1451                 if (!pci_enable_msi(adapter->pdev)) {
1452                         adapter->flags |= FLAG_MSI_ENABLED;
1453                 } else {
1454                         adapter->int_mode = E1000E_INT_MODE_LEGACY;
1455                         e_err("Failed to initialize MSI interrupts.  Falling "
1456                               "back to legacy interrupts.\n");
1457                 }
1458                 /* Fall through */
1459         case E1000E_INT_MODE_LEGACY:
1460                 /* Don't do anything; this is the system default */
1461                 break;
1462         }
1463
1464         return;
1465 }
1466
1467 /**
1468  * e1000_request_msix - Initialize MSI-X interrupts
1469  *
1470  * e1000_request_msix allocates MSI-X vectors and requests interrupts from the
1471  * kernel.
1472  **/
1473 static int e1000_request_msix(struct e1000_adapter *adapter)
1474 {
1475         struct net_device *netdev = adapter->netdev;
1476         int err = 0, vector = 0;
1477
1478         if (strlen(netdev->name) < (IFNAMSIZ - 5))
1479                 sprintf(adapter->rx_ring->name, "%s-rx-0", netdev->name);
1480         else
1481                 memcpy(adapter->rx_ring->name, netdev->name, IFNAMSIZ);
1482         err = request_irq(adapter->msix_entries[vector].vector,
1483                           &e1000_intr_msix_rx, 0, adapter->rx_ring->name,
1484                           netdev);
1485         if (err)
1486                 goto out;
1487         adapter->rx_ring->itr_register = E1000_EITR_82574(vector);
1488         adapter->rx_ring->itr_val = adapter->itr;
1489         vector++;
1490
1491         if (strlen(netdev->name) < (IFNAMSIZ - 5))
1492                 sprintf(adapter->tx_ring->name, "%s-tx-0", netdev->name);
1493         else
1494                 memcpy(adapter->tx_ring->name, netdev->name, IFNAMSIZ);
1495         err = request_irq(adapter->msix_entries[vector].vector,
1496                           &e1000_intr_msix_tx, 0, adapter->tx_ring->name,
1497                           netdev);
1498         if (err)
1499                 goto out;
1500         adapter->tx_ring->itr_register = E1000_EITR_82574(vector);
1501         adapter->tx_ring->itr_val = adapter->itr;
1502         vector++;
1503
1504         err = request_irq(adapter->msix_entries[vector].vector,
1505                           &e1000_msix_other, 0, netdev->name, netdev);
1506         if (err)
1507                 goto out;
1508
1509         e1000_configure_msix(adapter);
1510         return 0;
1511 out:
1512         return err;
1513 }
1514
1515 /**
1516  * e1000_request_irq - initialize interrupts
1517  *
1518  * Attempts to configure interrupts using the best available
1519  * capabilities of the hardware and kernel.
1520  **/
1521 static int e1000_request_irq(struct e1000_adapter *adapter)
1522 {
1523         struct net_device *netdev = adapter->netdev;
1524         int err;
1525
1526         if (adapter->msix_entries) {
1527                 err = e1000_request_msix(adapter);
1528                 if (!err)
1529                         return err;
1530                 /* fall back to MSI */
1531                 e1000e_reset_interrupt_capability(adapter);
1532                 adapter->int_mode = E1000E_INT_MODE_MSI;
1533                 e1000e_set_interrupt_capability(adapter);
1534         }
1535         if (adapter->flags & FLAG_MSI_ENABLED) {
1536                 err = request_irq(adapter->pdev->irq, &e1000_intr_msi, 0,
1537                                   netdev->name, netdev);
1538                 if (!err)
1539                         return err;
1540
1541                 /* fall back to legacy interrupt */
1542                 e1000e_reset_interrupt_capability(adapter);
1543                 adapter->int_mode = E1000E_INT_MODE_LEGACY;
1544         }
1545
1546         err = request_irq(adapter->pdev->irq, &e1000_intr, IRQF_SHARED,
1547                           netdev->name, netdev);
1548         if (err)
1549                 e_err("Unable to allocate interrupt, Error: %d\n", err);
1550
1551         return err;
1552 }
1553
1554 static void e1000_free_irq(struct e1000_adapter *adapter)
1555 {
1556         struct net_device *netdev = adapter->netdev;
1557
1558         if (adapter->msix_entries) {
1559                 int vector = 0;
1560
1561                 free_irq(adapter->msix_entries[vector].vector, netdev);
1562                 vector++;
1563
1564                 free_irq(adapter->msix_entries[vector].vector, netdev);
1565                 vector++;
1566
1567                 /* Other Causes interrupt vector */
1568                 free_irq(adapter->msix_entries[vector].vector, netdev);
1569                 return;
1570         }
1571
1572         free_irq(adapter->pdev->irq, netdev);
1573 }
1574
1575 /**
1576  * e1000_irq_disable - Mask off interrupt generation on the NIC
1577  **/
1578 static void e1000_irq_disable(struct e1000_adapter *adapter)
1579 {
1580         struct e1000_hw *hw = &adapter->hw;
1581
1582         ew32(IMC, ~0);
1583         if (adapter->msix_entries)
1584                 ew32(EIAC_82574, 0);
1585         e1e_flush();
1586         synchronize_irq(adapter->pdev->irq);
1587 }
1588
1589 /**
1590  * e1000_irq_enable - Enable default interrupt generation settings
1591  **/
1592 static void e1000_irq_enable(struct e1000_adapter *adapter)
1593 {
1594         struct e1000_hw *hw = &adapter->hw;
1595
1596         if (adapter->msix_entries) {
1597                 ew32(EIAC_82574, adapter->eiac_mask & E1000_EIAC_MASK_82574);
1598                 ew32(IMS, adapter->eiac_mask | E1000_IMS_OTHER | E1000_IMS_LSC);
1599         } else {
1600                 ew32(IMS, IMS_ENABLE_MASK);
1601         }
1602         e1e_flush();
1603 }
1604
1605 /**
1606  * e1000_get_hw_control - get control of the h/w from f/w
1607  * @adapter: address of board private structure
1608  *
1609  * e1000_get_hw_control sets {CTRL_EXT|SWSM}:DRV_LOAD bit.
1610  * For ASF and Pass Through versions of f/w this means that
1611  * the driver is loaded. For AMT version (only with 82573)
1612  * of the f/w this means that the network i/f is open.
1613  **/
1614 static void e1000_get_hw_control(struct e1000_adapter *adapter)
1615 {
1616         struct e1000_hw *hw = &adapter->hw;
1617         u32 ctrl_ext;
1618         u32 swsm;
1619
1620         /* Let firmware know the driver has taken over */
1621         if (adapter->flags & FLAG_HAS_SWSM_ON_LOAD) {
1622                 swsm = er32(SWSM);
1623                 ew32(SWSM, swsm | E1000_SWSM_DRV_LOAD);
1624         } else if (adapter->flags & FLAG_HAS_CTRLEXT_ON_LOAD) {
1625                 ctrl_ext = er32(CTRL_EXT);
1626                 ew32(CTRL_EXT, ctrl_ext | E1000_CTRL_EXT_DRV_LOAD);
1627         }
1628 }
1629
1630 /**
1631  * e1000_release_hw_control - release control of the h/w to f/w
1632  * @adapter: address of board private structure
1633  *
1634  * e1000_release_hw_control resets {CTRL_EXT|SWSM}:DRV_LOAD bit.
1635  * For ASF and Pass Through versions of f/w this means that the
1636  * driver is no longer loaded. For AMT version (only with 82573) i
1637  * of the f/w this means that the network i/f is closed.
1638  *
1639  **/
1640 static void e1000_release_hw_control(struct e1000_adapter *adapter)
1641 {
1642         struct e1000_hw *hw = &adapter->hw;
1643         u32 ctrl_ext;
1644         u32 swsm;
1645
1646         /* Let firmware taken over control of h/w */
1647         if (adapter->flags & FLAG_HAS_SWSM_ON_LOAD) {
1648                 swsm = er32(SWSM);
1649                 ew32(SWSM, swsm & ~E1000_SWSM_DRV_LOAD);
1650         } else if (adapter->flags & FLAG_HAS_CTRLEXT_ON_LOAD) {
1651                 ctrl_ext = er32(CTRL_EXT);
1652                 ew32(CTRL_EXT, ctrl_ext & ~E1000_CTRL_EXT_DRV_LOAD);
1653         }
1654 }
1655
1656 /**
1657  * @e1000_alloc_ring - allocate memory for a ring structure
1658  **/
1659 static int e1000_alloc_ring_dma(struct e1000_adapter *adapter,
1660                                 struct e1000_ring *ring)
1661 {
1662         struct pci_dev *pdev = adapter->pdev;
1663
1664         ring->desc = dma_alloc_coherent(&pdev->dev, ring->size, &ring->dma,
1665                                         GFP_KERNEL);
1666         if (!ring->desc)
1667                 return -ENOMEM;
1668
1669         return 0;
1670 }
1671
1672 /**
1673  * e1000e_setup_tx_resources - allocate Tx resources (Descriptors)
1674  * @adapter: board private structure
1675  *
1676  * Return 0 on success, negative on failure
1677  **/
1678 int e1000e_setup_tx_resources(struct e1000_adapter *adapter)
1679 {
1680         struct e1000_ring *tx_ring = adapter->tx_ring;
1681         int err = -ENOMEM, size;
1682
1683         size = sizeof(struct e1000_buffer) * tx_ring->count;
1684         tx_ring->buffer_info = vmalloc(size);
1685         if (!tx_ring->buffer_info)
1686                 goto err;
1687         memset(tx_ring->buffer_info, 0, size);
1688
1689         /* round up to nearest 4K */
1690         tx_ring->size = tx_ring->count * sizeof(struct e1000_tx_desc);
1691         tx_ring->size = ALIGN(tx_ring->size, 4096);
1692
1693         err = e1000_alloc_ring_dma(adapter, tx_ring);
1694         if (err)
1695                 goto err;
1696
1697         tx_ring->next_to_use = 0;
1698         tx_ring->next_to_clean = 0;
1699
1700         return 0;
1701 err:
1702         vfree(tx_ring->buffer_info);
1703         e_err("Unable to allocate memory for the transmit descriptor ring\n");
1704         return err;
1705 }
1706
1707 /**
1708  * e1000e_setup_rx_resources - allocate Rx resources (Descriptors)
1709  * @adapter: board private structure
1710  *
1711  * Returns 0 on success, negative on failure
1712  **/
1713 int e1000e_setup_rx_resources(struct e1000_adapter *adapter)
1714 {
1715         struct e1000_ring *rx_ring = adapter->rx_ring;
1716         struct e1000_buffer *buffer_info;
1717         int i, size, desc_len, err = -ENOMEM;
1718
1719         size = sizeof(struct e1000_buffer) * rx_ring->count;
1720         rx_ring->buffer_info = vmalloc(size);
1721         if (!rx_ring->buffer_info)
1722                 goto err;
1723         memset(rx_ring->buffer_info, 0, size);
1724
1725         for (i = 0; i < rx_ring->count; i++) {
1726                 buffer_info = &rx_ring->buffer_info[i];
1727                 buffer_info->ps_pages = kcalloc(PS_PAGE_BUFFERS,
1728                                                 sizeof(struct e1000_ps_page),
1729                                                 GFP_KERNEL);
1730                 if (!buffer_info->ps_pages)
1731                         goto err_pages;
1732         }
1733
1734         desc_len = sizeof(union e1000_rx_desc_packet_split);
1735
1736         /* Round up to nearest 4K */
1737         rx_ring->size = rx_ring->count * desc_len;
1738         rx_ring->size = ALIGN(rx_ring->size, 4096);
1739
1740         err = e1000_alloc_ring_dma(adapter, rx_ring);
1741         if (err)
1742                 goto err_pages;
1743
1744         rx_ring->next_to_clean = 0;
1745         rx_ring->next_to_use = 0;
1746         rx_ring->rx_skb_top = NULL;
1747
1748         return 0;
1749
1750 err_pages:
1751         for (i = 0; i < rx_ring->count; i++) {
1752                 buffer_info = &rx_ring->buffer_info[i];
1753                 kfree(buffer_info->ps_pages);
1754         }
1755 err:
1756         vfree(rx_ring->buffer_info);
1757         e_err("Unable to allocate memory for the transmit descriptor ring\n");
1758         return err;
1759 }
1760
1761 /**
1762  * e1000_clean_tx_ring - Free Tx Buffers
1763  * @adapter: board private structure
1764  **/
1765 static void e1000_clean_tx_ring(struct e1000_adapter *adapter)
1766 {
1767         struct e1000_ring *tx_ring = adapter->tx_ring;
1768         struct e1000_buffer *buffer_info;
1769         unsigned long size;
1770         unsigned int i;
1771
1772         for (i = 0; i < tx_ring->count; i++) {
1773                 buffer_info = &tx_ring->buffer_info[i];
1774                 e1000_put_txbuf(adapter, buffer_info);
1775         }
1776
1777         size = sizeof(struct e1000_buffer) * tx_ring->count;
1778         memset(tx_ring->buffer_info, 0, size);
1779
1780         memset(tx_ring->desc, 0, tx_ring->size);
1781
1782         tx_ring->next_to_use = 0;
1783         tx_ring->next_to_clean = 0;
1784
1785         writel(0, adapter->hw.hw_addr + tx_ring->head);
1786         writel(0, adapter->hw.hw_addr + tx_ring->tail);
1787 }
1788
1789 /**
1790  * e1000e_free_tx_resources - Free Tx Resources per Queue
1791  * @adapter: board private structure
1792  *
1793  * Free all transmit software resources
1794  **/
1795 void e1000e_free_tx_resources(struct e1000_adapter *adapter)
1796 {
1797         struct pci_dev *pdev = adapter->pdev;
1798         struct e1000_ring *tx_ring = adapter->tx_ring;
1799
1800         e1000_clean_tx_ring(adapter);
1801
1802         vfree(tx_ring->buffer_info);
1803         tx_ring->buffer_info = NULL;
1804
1805         dma_free_coherent(&pdev->dev, tx_ring->size, tx_ring->desc,
1806                           tx_ring->dma);
1807         tx_ring->desc = NULL;
1808 }
1809
1810 /**
1811  * e1000e_free_rx_resources - Free Rx Resources
1812  * @adapter: board private structure
1813  *
1814  * Free all receive software resources
1815  **/
1816
1817 void e1000e_free_rx_resources(struct e1000_adapter *adapter)
1818 {
1819         struct pci_dev *pdev = adapter->pdev;
1820         struct e1000_ring *rx_ring = adapter->rx_ring;
1821         int i;
1822
1823         e1000_clean_rx_ring(adapter);
1824
1825         for (i = 0; i < rx_ring->count; i++) {
1826                 kfree(rx_ring->buffer_info[i].ps_pages);
1827         }
1828
1829         vfree(rx_ring->buffer_info);
1830         rx_ring->buffer_info = NULL;
1831
1832         dma_free_coherent(&pdev->dev, rx_ring->size, rx_ring->desc,
1833                           rx_ring->dma);
1834         rx_ring->desc = NULL;
1835 }
1836
1837 /**
1838  * e1000_update_itr - update the dynamic ITR value based on statistics
1839  * @adapter: pointer to adapter
1840  * @itr_setting: current adapter->itr
1841  * @packets: the number of packets during this measurement interval
1842  * @bytes: the number of bytes during this measurement interval
1843  *
1844  *      Stores a new ITR value based on packets and byte
1845  *      counts during the last interrupt.  The advantage of per interrupt
1846  *      computation is faster updates and more accurate ITR for the current
1847  *      traffic pattern.  Constants in this function were computed
1848  *      based on theoretical maximum wire speed and thresholds were set based
1849  *      on testing data as well as attempting to minimize response time
1850  *      while increasing bulk throughput.  This functionality is controlled
1851  *      by the InterruptThrottleRate module parameter.
1852  **/
1853 static unsigned int e1000_update_itr(struct e1000_adapter *adapter,
1854                                      u16 itr_setting, int packets,
1855                                      int bytes)
1856 {
1857         unsigned int retval = itr_setting;
1858
1859         if (packets == 0)
1860                 goto update_itr_done;
1861
1862         switch (itr_setting) {
1863         case lowest_latency:
1864                 /* handle TSO and jumbo frames */
1865                 if (bytes/packets > 8000)
1866                         retval = bulk_latency;
1867                 else if ((packets < 5) && (bytes > 512)) {
1868                         retval = low_latency;
1869                 }
1870                 break;
1871         case low_latency:  /* 50 usec aka 20000 ints/s */
1872                 if (bytes > 10000) {
1873                         /* this if handles the TSO accounting */
1874                         if (bytes/packets > 8000) {
1875                                 retval = bulk_latency;
1876                         } else if ((packets < 10) || ((bytes/packets) > 1200)) {
1877                                 retval = bulk_latency;
1878                         } else if ((packets > 35)) {
1879                                 retval = lowest_latency;
1880                         }
1881                 } else if (bytes/packets > 2000) {
1882                         retval = bulk_latency;
1883                 } else if (packets <= 2 && bytes < 512) {
1884                         retval = lowest_latency;
1885                 }
1886                 break;
1887         case bulk_latency: /* 250 usec aka 4000 ints/s */
1888                 if (bytes > 25000) {
1889                         if (packets > 35) {
1890                                 retval = low_latency;
1891                         }
1892                 } else if (bytes < 6000) {
1893                         retval = low_latency;
1894                 }
1895                 break;
1896         }
1897
1898 update_itr_done:
1899         return retval;
1900 }
1901
1902 static void e1000_set_itr(struct e1000_adapter *adapter)
1903 {
1904         struct e1000_hw *hw = &adapter->hw;
1905         u16 current_itr;
1906         u32 new_itr = adapter->itr;
1907
1908         /* for non-gigabit speeds, just fix the interrupt rate at 4000 */
1909         if (adapter->link_speed != SPEED_1000) {
1910                 current_itr = 0;
1911                 new_itr = 4000;
1912                 goto set_itr_now;
1913         }
1914
1915         adapter->tx_itr = e1000_update_itr(adapter,
1916                                     adapter->tx_itr,
1917                                     adapter->total_tx_packets,
1918                                     adapter->total_tx_bytes);
1919         /* conservative mode (itr 3) eliminates the lowest_latency setting */
1920         if (adapter->itr_setting == 3 && adapter->tx_itr == lowest_latency)
1921                 adapter->tx_itr = low_latency;
1922
1923         adapter->rx_itr = e1000_update_itr(adapter,
1924                                     adapter->rx_itr,
1925                                     adapter->total_rx_packets,
1926                                     adapter->total_rx_bytes);
1927         /* conservative mode (itr 3) eliminates the lowest_latency setting */
1928         if (adapter->itr_setting == 3 && adapter->rx_itr == lowest_latency)
1929                 adapter->rx_itr = low_latency;
1930
1931         current_itr = max(adapter->rx_itr, adapter->tx_itr);
1932
1933         switch (current_itr) {
1934         /* counts and packets in update_itr are dependent on these numbers */
1935         case lowest_latency:
1936                 new_itr = 70000;
1937                 break;
1938         case low_latency:
1939                 new_itr = 20000; /* aka hwitr = ~200 */
1940                 break;
1941         case bulk_latency:
1942                 new_itr = 4000;
1943                 break;
1944         default:
1945                 break;
1946         }
1947
1948 set_itr_now:
1949         if (new_itr != adapter->itr) {
1950                 /*
1951                  * this attempts to bias the interrupt rate towards Bulk
1952                  * by adding intermediate steps when interrupt rate is
1953                  * increasing
1954                  */
1955                 new_itr = new_itr > adapter->itr ?
1956                              min(adapter->itr + (new_itr >> 2), new_itr) :
1957                              new_itr;
1958                 adapter->itr = new_itr;
1959                 adapter->rx_ring->itr_val = new_itr;
1960                 if (adapter->msix_entries)
1961                         adapter->rx_ring->set_itr = 1;
1962                 else
1963                         ew32(ITR, 1000000000 / (new_itr * 256));
1964         }
1965 }
1966
1967 /**
1968  * e1000_alloc_queues - Allocate memory for all rings
1969  * @adapter: board private structure to initialize
1970  **/
1971 static int __devinit e1000_alloc_queues(struct e1000_adapter *adapter)
1972 {
1973         adapter->tx_ring = kzalloc(sizeof(struct e1000_ring), GFP_KERNEL);
1974         if (!adapter->tx_ring)
1975                 goto err;
1976
1977         adapter->rx_ring = kzalloc(sizeof(struct e1000_ring), GFP_KERNEL);
1978         if (!adapter->rx_ring)
1979                 goto err;
1980
1981         return 0;
1982 err:
1983         e_err("Unable to allocate memory for queues\n");
1984         kfree(adapter->rx_ring);
1985         kfree(adapter->tx_ring);
1986         return -ENOMEM;
1987 }
1988
1989 /**
1990  * e1000_clean - NAPI Rx polling callback
1991  * @napi: struct associated with this polling callback
1992  * @budget: amount of packets driver is allowed to process this poll
1993  **/
1994 static int e1000_clean(struct napi_struct *napi, int budget)
1995 {
1996         struct e1000_adapter *adapter = container_of(napi, struct e1000_adapter, napi);
1997         struct e1000_hw *hw = &adapter->hw;
1998         struct net_device *poll_dev = adapter->netdev;
1999         int tx_cleaned = 0, work_done = 0;
2000
2001         adapter = netdev_priv(poll_dev);
2002
2003         if (adapter->msix_entries &&
2004             !(adapter->rx_ring->ims_val & adapter->tx_ring->ims_val))
2005                 goto clean_rx;
2006
2007         tx_cleaned = e1000_clean_tx_irq(adapter);
2008
2009 clean_rx:
2010         adapter->clean_rx(adapter, &work_done, budget);
2011
2012         if (!tx_cleaned)
2013                 work_done = budget;
2014
2015         /* If budget not fully consumed, exit the polling mode */
2016         if (work_done < budget) {
2017                 if (adapter->itr_setting & 3)
2018                         e1000_set_itr(adapter);
2019                 napi_complete(napi);
2020                 if (!test_bit(__E1000_DOWN, &adapter->state)) {
2021                         if (adapter->msix_entries)
2022                                 ew32(IMS, adapter->rx_ring->ims_val);
2023                         else
2024                                 e1000_irq_enable(adapter);
2025                 }
2026         }
2027
2028         return work_done;
2029 }
2030
2031 static void e1000_vlan_rx_add_vid(struct net_device *netdev, u16 vid)
2032 {
2033         struct e1000_adapter *adapter = netdev_priv(netdev);
2034         struct e1000_hw *hw = &adapter->hw;
2035         u32 vfta, index;
2036
2037         /* don't update vlan cookie if already programmed */
2038         if ((adapter->hw.mng_cookie.status &
2039              E1000_MNG_DHCP_COOKIE_STATUS_VLAN) &&
2040             (vid == adapter->mng_vlan_id))
2041                 return;
2042         /* add VID to filter table */
2043         index = (vid >> 5) & 0x7F;
2044         vfta = E1000_READ_REG_ARRAY(hw, E1000_VFTA, index);
2045         vfta |= (1 << (vid & 0x1F));
2046         e1000e_write_vfta(hw, index, vfta);
2047 }
2048
2049 static void e1000_vlan_rx_kill_vid(struct net_device *netdev, u16 vid)
2050 {
2051         struct e1000_adapter *adapter = netdev_priv(netdev);
2052         struct e1000_hw *hw = &adapter->hw;
2053         u32 vfta, index;
2054
2055         if (!test_bit(__E1000_DOWN, &adapter->state))
2056                 e1000_irq_disable(adapter);
2057         vlan_group_set_device(adapter->vlgrp, vid, NULL);
2058
2059         if (!test_bit(__E1000_DOWN, &adapter->state))
2060                 e1000_irq_enable(adapter);
2061
2062         if ((adapter->hw.mng_cookie.status &
2063              E1000_MNG_DHCP_COOKIE_STATUS_VLAN) &&
2064             (vid == adapter->mng_vlan_id)) {
2065                 /* release control to f/w */
2066                 e1000_release_hw_control(adapter);
2067                 return;
2068         }
2069
2070         /* remove VID from filter table */
2071         index = (vid >> 5) & 0x7F;
2072         vfta = E1000_READ_REG_ARRAY(hw, E1000_VFTA, index);
2073         vfta &= ~(1 << (vid & 0x1F));
2074         e1000e_write_vfta(hw, index, vfta);
2075 }
2076
2077 static void e1000_update_mng_vlan(struct e1000_adapter *adapter)
2078 {
2079         struct net_device *netdev = adapter->netdev;
2080         u16 vid = adapter->hw.mng_cookie.vlan_id;
2081         u16 old_vid = adapter->mng_vlan_id;
2082
2083         if (!adapter->vlgrp)
2084                 return;
2085
2086         if (!vlan_group_get_device(adapter->vlgrp, vid)) {
2087                 adapter->mng_vlan_id = E1000_MNG_VLAN_NONE;
2088                 if (adapter->hw.mng_cookie.status &
2089                         E1000_MNG_DHCP_COOKIE_STATUS_VLAN) {
2090                         e1000_vlan_rx_add_vid(netdev, vid);
2091                         adapter->mng_vlan_id = vid;
2092                 }
2093
2094                 if ((old_vid != (u16)E1000_MNG_VLAN_NONE) &&
2095                                 (vid != old_vid) &&
2096                     !vlan_group_get_device(adapter->vlgrp, old_vid))
2097                         e1000_vlan_rx_kill_vid(netdev, old_vid);
2098         } else {
2099                 adapter->mng_vlan_id = vid;
2100         }
2101 }
2102
2103
2104 static void e1000_vlan_rx_register(struct net_device *netdev,
2105                                    struct vlan_group *grp)
2106 {
2107         struct e1000_adapter *adapter = netdev_priv(netdev);
2108         struct e1000_hw *hw = &adapter->hw;
2109         u32 ctrl, rctl;
2110
2111         if (!test_bit(__E1000_DOWN, &adapter->state))
2112                 e1000_irq_disable(adapter);
2113         adapter->vlgrp = grp;
2114
2115         if (grp) {
2116                 /* enable VLAN tag insert/strip */
2117                 ctrl = er32(CTRL);
2118                 ctrl |= E1000_CTRL_VME;
2119                 ew32(CTRL, ctrl);
2120
2121                 if (adapter->flags & FLAG_HAS_HW_VLAN_FILTER) {
2122                         /* enable VLAN receive filtering */
2123                         rctl = er32(RCTL);
2124                         rctl &= ~E1000_RCTL_CFIEN;
2125                         ew32(RCTL, rctl);
2126                         e1000_update_mng_vlan(adapter);
2127                 }
2128         } else {
2129                 /* disable VLAN tag insert/strip */
2130                 ctrl = er32(CTRL);
2131                 ctrl &= ~E1000_CTRL_VME;
2132                 ew32(CTRL, ctrl);
2133
2134                 if (adapter->flags & FLAG_HAS_HW_VLAN_FILTER) {
2135                         if (adapter->mng_vlan_id !=
2136                             (u16)E1000_MNG_VLAN_NONE) {
2137                                 e1000_vlan_rx_kill_vid(netdev,
2138                                                        adapter->mng_vlan_id);
2139                                 adapter->mng_vlan_id = E1000_MNG_VLAN_NONE;
2140                         }
2141                 }
2142         }
2143
2144         if (!test_bit(__E1000_DOWN, &adapter->state))
2145                 e1000_irq_enable(adapter);
2146 }
2147
2148 static void e1000_restore_vlan(struct e1000_adapter *adapter)
2149 {
2150         u16 vid;
2151
2152         e1000_vlan_rx_register(adapter->netdev, adapter->vlgrp);
2153
2154         if (!adapter->vlgrp)
2155                 return;
2156
2157         for (vid = 0; vid < VLAN_GROUP_ARRAY_LEN; vid++) {
2158                 if (!vlan_group_get_device(adapter->vlgrp, vid))
2159                         continue;
2160                 e1000_vlan_rx_add_vid(adapter->netdev, vid);
2161         }
2162 }
2163
2164 static void e1000_init_manageability(struct e1000_adapter *adapter)
2165 {
2166         struct e1000_hw *hw = &adapter->hw;
2167         u32 manc, manc2h;
2168
2169         if (!(adapter->flags & FLAG_MNG_PT_ENABLED))
2170                 return;
2171
2172         manc = er32(MANC);
2173
2174         /*
2175          * enable receiving management packets to the host. this will probably
2176          * generate destination unreachable messages from the host OS, but
2177          * the packets will be handled on SMBUS
2178          */
2179         manc |= E1000_MANC_EN_MNG2HOST;
2180         manc2h = er32(MANC2H);
2181 #define E1000_MNG2HOST_PORT_623 (1 << 5)
2182 #define E1000_MNG2HOST_PORT_664 (1 << 6)
2183         manc2h |= E1000_MNG2HOST_PORT_623;
2184         manc2h |= E1000_MNG2HOST_PORT_664;
2185         ew32(MANC2H, manc2h);
2186         ew32(MANC, manc);
2187 }
2188
2189 /**
2190  * e1000_configure_tx - Configure 8254x Transmit Unit after Reset
2191  * @adapter: board private structure
2192  *
2193  * Configure the Tx unit of the MAC after a reset.
2194  **/
2195 static void e1000_configure_tx(struct e1000_adapter *adapter)
2196 {
2197         struct e1000_hw *hw = &adapter->hw;
2198         struct e1000_ring *tx_ring = adapter->tx_ring;
2199         u64 tdba;
2200         u32 tdlen, tctl, tipg, tarc;
2201         u32 ipgr1, ipgr2;
2202
2203         /* Setup the HW Tx Head and Tail descriptor pointers */
2204         tdba = tx_ring->dma;
2205         tdlen = tx_ring->count * sizeof(struct e1000_tx_desc);
2206         ew32(TDBAL, (tdba & DMA_BIT_MASK(32)));
2207         ew32(TDBAH, (tdba >> 32));
2208         ew32(TDLEN, tdlen);
2209         ew32(TDH, 0);
2210         ew32(TDT, 0);
2211         tx_ring->head = E1000_TDH;
2212         tx_ring->tail = E1000_TDT;
2213
2214         /* Set the default values for the Tx Inter Packet Gap timer */
2215         tipg = DEFAULT_82543_TIPG_IPGT_COPPER;          /*  8  */
2216         ipgr1 = DEFAULT_82543_TIPG_IPGR1;               /*  8  */
2217         ipgr2 = DEFAULT_82543_TIPG_IPGR2;               /*  6  */
2218
2219         if (adapter->flags & FLAG_TIPG_MEDIUM_FOR_80003ESLAN)
2220                 ipgr2 = DEFAULT_80003ES2LAN_TIPG_IPGR2; /*  7  */
2221
2222         tipg |= ipgr1 << E1000_TIPG_IPGR1_SHIFT;
2223         tipg |= ipgr2 << E1000_TIPG_IPGR2_SHIFT;
2224         ew32(TIPG, tipg);
2225
2226         /* Set the Tx Interrupt Delay register */
2227         ew32(TIDV, adapter->tx_int_delay);
2228         /* Tx irq moderation */
2229         ew32(TADV, adapter->tx_abs_int_delay);
2230
2231         /* Program the Transmit Control Register */
2232         tctl = er32(TCTL);
2233         tctl &= ~E1000_TCTL_CT;
2234         tctl |= E1000_TCTL_PSP | E1000_TCTL_RTLC |
2235                 (E1000_COLLISION_THRESHOLD << E1000_CT_SHIFT);
2236
2237         if (adapter->flags & FLAG_TARC_SPEED_MODE_BIT) {
2238                 tarc = er32(TARC(0));
2239                 /*
2240                  * set the speed mode bit, we'll clear it if we're not at
2241                  * gigabit link later
2242                  */
2243 #define SPEED_MODE_BIT (1 << 21)
2244                 tarc |= SPEED_MODE_BIT;
2245                 ew32(TARC(0), tarc);
2246         }
2247
2248         /* errata: program both queues to unweighted RR */
2249         if (adapter->flags & FLAG_TARC_SET_BIT_ZERO) {
2250                 tarc = er32(TARC(0));
2251                 tarc |= 1;
2252                 ew32(TARC(0), tarc);
2253                 tarc = er32(TARC(1));
2254                 tarc |= 1;
2255                 ew32(TARC(1), tarc);
2256         }
2257
2258         e1000e_config_collision_dist(hw);
2259
2260         /* Setup Transmit Descriptor Settings for eop descriptor */
2261         adapter->txd_cmd = E1000_TXD_CMD_EOP | E1000_TXD_CMD_IFCS;
2262
2263         /* only set IDE if we are delaying interrupts using the timers */
2264         if (adapter->tx_int_delay)
2265                 adapter->txd_cmd |= E1000_TXD_CMD_IDE;
2266
2267         /* enable Report Status bit */
2268         adapter->txd_cmd |= E1000_TXD_CMD_RS;
2269
2270         ew32(TCTL, tctl);
2271
2272         adapter->tx_queue_len = adapter->netdev->tx_queue_len;
2273 }
2274
2275 /**
2276  * e1000_setup_rctl - configure the receive control registers
2277  * @adapter: Board private structure
2278  **/
2279 #define PAGE_USE_COUNT(S) (((S) >> PAGE_SHIFT) + \
2280                            (((S) & (PAGE_SIZE - 1)) ? 1 : 0))
2281 static void e1000_setup_rctl(struct e1000_adapter *adapter)
2282 {
2283         struct e1000_hw *hw = &adapter->hw;
2284         u32 rctl, rfctl;
2285         u32 psrctl = 0;
2286         u32 pages = 0;
2287
2288         /* Program MC offset vector base */
2289         rctl = er32(RCTL);
2290         rctl &= ~(3 << E1000_RCTL_MO_SHIFT);
2291         rctl |= E1000_RCTL_EN | E1000_RCTL_BAM |
2292                 E1000_RCTL_LBM_NO | E1000_RCTL_RDMTS_HALF |
2293                 (adapter->hw.mac.mc_filter_type << E1000_RCTL_MO_SHIFT);
2294
2295         /* Do not Store bad packets */
2296         rctl &= ~E1000_RCTL_SBP;
2297
2298         /* Enable Long Packet receive */
2299         if (adapter->netdev->mtu <= ETH_DATA_LEN)
2300                 rctl &= ~E1000_RCTL_LPE;
2301         else
2302                 rctl |= E1000_RCTL_LPE;
2303
2304         /* Some systems expect that the CRC is included in SMBUS traffic. The
2305          * hardware strips the CRC before sending to both SMBUS (BMC) and to
2306          * host memory when this is enabled
2307          */
2308         if (adapter->flags2 & FLAG2_CRC_STRIPPING)
2309                 rctl |= E1000_RCTL_SECRC;
2310
2311         /* Setup buffer sizes */
2312         rctl &= ~E1000_RCTL_SZ_4096;
2313         rctl |= E1000_RCTL_BSEX;
2314         switch (adapter->rx_buffer_len) {
2315         case 256:
2316                 rctl |= E1000_RCTL_SZ_256;
2317                 rctl &= ~E1000_RCTL_BSEX;
2318                 break;
2319         case 512:
2320                 rctl |= E1000_RCTL_SZ_512;
2321                 rctl &= ~E1000_RCTL_BSEX;
2322                 break;
2323         case 1024:
2324                 rctl |= E1000_RCTL_SZ_1024;
2325                 rctl &= ~E1000_RCTL_BSEX;
2326                 break;
2327         case 2048:
2328         default:
2329                 rctl |= E1000_RCTL_SZ_2048;
2330                 rctl &= ~E1000_RCTL_BSEX;
2331                 break;
2332         case 4096:
2333                 rctl |= E1000_RCTL_SZ_4096;
2334                 break;
2335         case 8192:
2336                 rctl |= E1000_RCTL_SZ_8192;
2337                 break;
2338         case 16384:
2339                 rctl |= E1000_RCTL_SZ_16384;
2340                 break;
2341         }
2342
2343         /*
2344          * 82571 and greater support packet-split where the protocol
2345          * header is placed in skb->data and the packet data is
2346          * placed in pages hanging off of skb_shinfo(skb)->nr_frags.
2347          * In the case of a non-split, skb->data is linearly filled,
2348          * followed by the page buffers.  Therefore, skb->data is
2349          * sized to hold the largest protocol header.
2350          *
2351          * allocations using alloc_page take too long for regular MTU
2352          * so only enable packet split for jumbo frames
2353          *
2354          * Using pages when the page size is greater than 16k wastes
2355          * a lot of memory, since we allocate 3 pages at all times
2356          * per packet.
2357          */
2358         pages = PAGE_USE_COUNT(adapter->netdev->mtu);
2359         if (!(adapter->flags & FLAG_IS_ICH) && (pages <= 3) &&
2360             (PAGE_SIZE <= 16384) && (rctl & E1000_RCTL_LPE))
2361                 adapter->rx_ps_pages = pages;
2362         else
2363                 adapter->rx_ps_pages = 0;
2364
2365         if (adapter->rx_ps_pages) {
2366                 /* Configure extra packet-split registers */
2367                 rfctl = er32(RFCTL);
2368                 rfctl |= E1000_RFCTL_EXTEN;
2369                 /*
2370                  * disable packet split support for IPv6 extension headers,
2371                  * because some malformed IPv6 headers can hang the Rx
2372                  */
2373                 rfctl |= (E1000_RFCTL_IPV6_EX_DIS |
2374                           E1000_RFCTL_NEW_IPV6_EXT_DIS);
2375
2376                 ew32(RFCTL, rfctl);
2377
2378                 /* Enable Packet split descriptors */
2379                 rctl |= E1000_RCTL_DTYP_PS;
2380
2381                 psrctl |= adapter->rx_ps_bsize0 >>
2382                         E1000_PSRCTL_BSIZE0_SHIFT;
2383
2384                 switch (adapter->rx_ps_pages) {
2385                 case 3:
2386                         psrctl |= PAGE_SIZE <<
2387                                 E1000_PSRCTL_BSIZE3_SHIFT;
2388                 case 2:
2389                         psrctl |= PAGE_SIZE <<
2390                                 E1000_PSRCTL_BSIZE2_SHIFT;
2391                 case 1:
2392                         psrctl |= PAGE_SIZE >>
2393                                 E1000_PSRCTL_BSIZE1_SHIFT;
2394                         break;
2395                 }
2396
2397                 ew32(PSRCTL, psrctl);
2398         }
2399
2400         ew32(RCTL, rctl);
2401         /* just started the receive unit, no need to restart */
2402         adapter->flags &= ~FLAG_RX_RESTART_NOW;
2403 }
2404
2405 /**
2406  * e1000_configure_rx - Configure Receive Unit after Reset
2407  * @adapter: board private structure
2408  *
2409  * Configure the Rx unit of the MAC after a reset.
2410  **/
2411 static void e1000_configure_rx(struct e1000_adapter *adapter)
2412 {
2413         struct e1000_hw *hw = &adapter->hw;
2414         struct e1000_ring *rx_ring = adapter->rx_ring;
2415         u64 rdba;
2416         u32 rdlen, rctl, rxcsum, ctrl_ext;
2417
2418         if (adapter->rx_ps_pages) {
2419                 /* this is a 32 byte descriptor */
2420                 rdlen = rx_ring->count *
2421                         sizeof(union e1000_rx_desc_packet_split);
2422                 adapter->clean_rx = e1000_clean_rx_irq_ps;
2423                 adapter->alloc_rx_buf = e1000_alloc_rx_buffers_ps;
2424         } else if (adapter->netdev->mtu > ETH_FRAME_LEN + ETH_FCS_LEN) {
2425                 rdlen = rx_ring->count * sizeof(struct e1000_rx_desc);
2426                 adapter->clean_rx = e1000_clean_jumbo_rx_irq;
2427                 adapter->alloc_rx_buf = e1000_alloc_jumbo_rx_buffers;
2428         } else {
2429                 rdlen = rx_ring->count * sizeof(struct e1000_rx_desc);
2430                 adapter->clean_rx = e1000_clean_rx_irq;
2431                 adapter->alloc_rx_buf = e1000_alloc_rx_buffers;
2432         }
2433
2434         /* disable receives while setting up the descriptors */
2435         rctl = er32(RCTL);
2436         ew32(RCTL, rctl & ~E1000_RCTL_EN);
2437         e1e_flush();
2438         msleep(10);
2439
2440         /* set the Receive Delay Timer Register */
2441         ew32(RDTR, adapter->rx_int_delay);
2442
2443         /* irq moderation */
2444         ew32(RADV, adapter->rx_abs_int_delay);
2445         if (adapter->itr_setting != 0)
2446                 ew32(ITR, 1000000000 / (adapter->itr * 256));
2447
2448         ctrl_ext = er32(CTRL_EXT);
2449         /* Reset delay timers after every interrupt */
2450         ctrl_ext |= E1000_CTRL_EXT_INT_TIMER_CLR;
2451         /* Auto-Mask interrupts upon ICR access */
2452         ctrl_ext |= E1000_CTRL_EXT_IAME;
2453         ew32(IAM, 0xffffffff);
2454         ew32(CTRL_EXT, ctrl_ext);
2455         e1e_flush();
2456
2457         /*
2458          * Setup the HW Rx Head and Tail Descriptor Pointers and
2459          * the Base and Length of the Rx Descriptor Ring
2460          */
2461         rdba = rx_ring->dma;
2462         ew32(RDBAL, (rdba & DMA_BIT_MASK(32)));
2463         ew32(RDBAH, (rdba >> 32));
2464         ew32(RDLEN, rdlen);
2465         ew32(RDH, 0);
2466         ew32(RDT, 0);
2467         rx_ring->head = E1000_RDH;
2468         rx_ring->tail = E1000_RDT;
2469
2470         /* Enable Receive Checksum Offload for TCP and UDP */
2471         rxcsum = er32(RXCSUM);
2472         if (adapter->flags & FLAG_RX_CSUM_ENABLED) {
2473                 rxcsum |= E1000_RXCSUM_TUOFL;
2474
2475                 /*
2476                  * IPv4 payload checksum for UDP fragments must be
2477                  * used in conjunction with packet-split.
2478                  */
2479                 if (adapter->rx_ps_pages)
2480                         rxcsum |= E1000_RXCSUM_IPPCSE;
2481         } else {
2482                 rxcsum &= ~E1000_RXCSUM_TUOFL;
2483                 /* no need to clear IPPCSE as it defaults to 0 */
2484         }
2485         ew32(RXCSUM, rxcsum);
2486
2487         /*
2488          * Enable early receives on supported devices, only takes effect when
2489          * packet size is equal or larger than the specified value (in 8 byte
2490          * units), e.g. using jumbo frames when setting to E1000_ERT_2048
2491          */
2492         if ((adapter->flags & FLAG_HAS_ERT) &&
2493             (adapter->netdev->mtu > ETH_DATA_LEN)) {
2494                 u32 rxdctl = er32(RXDCTL(0));
2495                 ew32(RXDCTL(0), rxdctl | 0x3);
2496                 ew32(ERT, E1000_ERT_2048 | (1 << 13));
2497                 /*
2498                  * With jumbo frames and early-receive enabled, excessive
2499                  * C4->C2 latencies result in dropped transactions.
2500                  */
2501                 pm_qos_update_requirement(PM_QOS_CPU_DMA_LATENCY,
2502                                           e1000e_driver_name, 55);
2503         } else {
2504                 pm_qos_update_requirement(PM_QOS_CPU_DMA_LATENCY,
2505                                           e1000e_driver_name,
2506                                           PM_QOS_DEFAULT_VALUE);
2507         }
2508
2509         /* Enable Receives */
2510         ew32(RCTL, rctl);
2511 }
2512
2513 /**
2514  *  e1000_update_mc_addr_list - Update Multicast addresses
2515  *  @hw: pointer to the HW structure
2516  *  @mc_addr_list: array of multicast addresses to program
2517  *  @mc_addr_count: number of multicast addresses to program
2518  *  @rar_used_count: the first RAR register free to program
2519  *  @rar_count: total number of supported Receive Address Registers
2520  *
2521  *  Updates the Receive Address Registers and Multicast Table Array.
2522  *  The caller must have a packed mc_addr_list of multicast addresses.
2523  *  The parameter rar_count will usually be hw->mac.rar_entry_count
2524  *  unless there are workarounds that change this.  Currently no func pointer
2525  *  exists and all implementations are handled in the generic version of this
2526  *  function.
2527  **/
2528 static void e1000_update_mc_addr_list(struct e1000_hw *hw, u8 *mc_addr_list,
2529                                       u32 mc_addr_count, u32 rar_used_count,
2530                                       u32 rar_count)
2531 {
2532         hw->mac.ops.update_mc_addr_list(hw, mc_addr_list, mc_addr_count,
2533                                         rar_used_count, rar_count);
2534 }
2535
2536 /**
2537  * e1000_set_multi - Multicast and Promiscuous mode set
2538  * @netdev: network interface device structure
2539  *
2540  * The set_multi entry point is called whenever the multicast address
2541  * list or the network interface flags are updated.  This routine is
2542  * responsible for configuring the hardware for proper multicast,
2543  * promiscuous mode, and all-multi behavior.
2544  **/
2545 static void e1000_set_multi(struct net_device *netdev)
2546 {
2547         struct e1000_adapter *adapter = netdev_priv(netdev);
2548         struct e1000_hw *hw = &adapter->hw;
2549         struct e1000_mac_info *mac = &hw->mac;
2550         struct dev_mc_list *mc_ptr;
2551         u8  *mta_list;
2552         u32 rctl;
2553         int i;
2554
2555         /* Check for Promiscuous and All Multicast modes */
2556
2557         rctl = er32(RCTL);
2558
2559         if (netdev->flags & IFF_PROMISC) {
2560                 rctl |= (E1000_RCTL_UPE | E1000_RCTL_MPE);
2561                 rctl &= ~E1000_RCTL_VFE;
2562         } else {
2563                 if (netdev->flags & IFF_ALLMULTI) {
2564                         rctl |= E1000_RCTL_MPE;
2565                         rctl &= ~E1000_RCTL_UPE;
2566                 } else {
2567                         rctl &= ~(E1000_RCTL_UPE | E1000_RCTL_MPE);
2568                 }
2569                 if (adapter->flags & FLAG_HAS_HW_VLAN_FILTER)
2570                         rctl |= E1000_RCTL_VFE;
2571         }
2572
2573         ew32(RCTL, rctl);
2574
2575         if (netdev->mc_count) {
2576                 mta_list = kmalloc(netdev->mc_count * 6, GFP_ATOMIC);
2577                 if (!mta_list)
2578                         return;
2579
2580                 /* prepare a packed array of only addresses. */
2581                 mc_ptr = netdev->mc_list;
2582
2583                 for (i = 0; i < netdev->mc_count; i++) {
2584                         if (!mc_ptr)
2585                                 break;
2586                         memcpy(mta_list + (i*ETH_ALEN), mc_ptr->dmi_addr,
2587                                ETH_ALEN);
2588                         mc_ptr = mc_ptr->next;
2589                 }
2590
2591                 e1000_update_mc_addr_list(hw, mta_list, i, 1,
2592                                           mac->rar_entry_count);
2593                 kfree(mta_list);
2594         } else {
2595                 /*
2596                  * if we're called from probe, we might not have
2597                  * anything to do here, so clear out the list
2598                  */
2599                 e1000_update_mc_addr_list(hw, NULL, 0, 1, mac->rar_entry_count);
2600         }
2601 }
2602
2603 /**
2604  * e1000_configure - configure the hardware for Rx and Tx
2605  * @adapter: private board structure
2606  **/
2607 static void e1000_configure(struct e1000_adapter *adapter)
2608 {
2609         e1000_set_multi(adapter->netdev);
2610
2611         e1000_restore_vlan(adapter);
2612         e1000_init_manageability(adapter);
2613
2614         e1000_configure_tx(adapter);
2615         e1000_setup_rctl(adapter);
2616         e1000_configure_rx(adapter);
2617         adapter->alloc_rx_buf(adapter, e1000_desc_unused(adapter->rx_ring));
2618 }
2619
2620 /**
2621  * e1000e_power_up_phy - restore link in case the phy was powered down
2622  * @adapter: address of board private structure
2623  *
2624  * The phy may be powered down to save power and turn off link when the
2625  * driver is unloaded and wake on lan is not enabled (among others)
2626  * *** this routine MUST be followed by a call to e1000e_reset ***
2627  **/
2628 void e1000e_power_up_phy(struct e1000_adapter *adapter)
2629 {
2630         u16 mii_reg = 0;
2631
2632         /* Just clear the power down bit to wake the phy back up */
2633         if (adapter->hw.phy.media_type == e1000_media_type_copper) {
2634                 /*
2635                  * According to the manual, the phy will retain its
2636                  * settings across a power-down/up cycle
2637                  */
2638                 e1e_rphy(&adapter->hw, PHY_CONTROL, &mii_reg);
2639                 mii_reg &= ~MII_CR_POWER_DOWN;
2640                 e1e_wphy(&adapter->hw, PHY_CONTROL, mii_reg);
2641         }
2642
2643         adapter->hw.mac.ops.setup_link(&adapter->hw);
2644 }
2645
2646 /**
2647  * e1000_power_down_phy - Power down the PHY
2648  *
2649  * Power down the PHY so no link is implied when interface is down
2650  * The PHY cannot be powered down is management or WoL is active
2651  */
2652 static void e1000_power_down_phy(struct e1000_adapter *adapter)
2653 {
2654         struct e1000_hw *hw = &adapter->hw;
2655         u16 mii_reg;
2656
2657         /* WoL is enabled */
2658         if (adapter->wol)
2659                 return;
2660
2661         /* non-copper PHY? */
2662         if (adapter->hw.phy.media_type != e1000_media_type_copper)
2663                 return;
2664
2665         /* reset is blocked because of a SoL/IDER session */
2666         if (e1000e_check_mng_mode(hw) || e1000_check_reset_block(hw))
2667                 return;
2668
2669         /* manageability (AMT) is enabled */
2670         if (er32(MANC) & E1000_MANC_SMBUS_EN)
2671                 return;
2672
2673         /* power down the PHY */
2674         e1e_rphy(hw, PHY_CONTROL, &mii_reg);
2675         mii_reg |= MII_CR_POWER_DOWN;
2676         e1e_wphy(hw, PHY_CONTROL, mii_reg);
2677         mdelay(1);
2678 }
2679
2680 /**
2681  * e1000e_reset - bring the hardware into a known good state
2682  *
2683  * This function boots the hardware and enables some settings that
2684  * require a configuration cycle of the hardware - those cannot be
2685  * set/changed during runtime. After reset the device needs to be
2686  * properly configured for Rx, Tx etc.
2687  */
2688 void e1000e_reset(struct e1000_adapter *adapter)
2689 {
2690         struct e1000_mac_info *mac = &adapter->hw.mac;
2691         struct e1000_fc_info *fc = &adapter->hw.fc;
2692         struct e1000_hw *hw = &adapter->hw;
2693         u32 tx_space, min_tx_space, min_rx_space;
2694         u32 pba = adapter->pba;
2695         u16 hwm;
2696
2697         /* reset Packet Buffer Allocation to default */
2698         ew32(PBA, pba);
2699
2700         if (adapter->max_frame_size > ETH_FRAME_LEN + ETH_FCS_LEN) {
2701                 /*
2702                  * To maintain wire speed transmits, the Tx FIFO should be
2703                  * large enough to accommodate two full transmit packets,
2704                  * rounded up to the next 1KB and expressed in KB.  Likewise,
2705                  * the Rx FIFO should be large enough to accommodate at least
2706                  * one full receive packet and is similarly rounded up and
2707                  * expressed in KB.
2708                  */
2709                 pba = er32(PBA);
2710                 /* upper 16 bits has Tx packet buffer allocation size in KB */
2711                 tx_space = pba >> 16;
2712                 /* lower 16 bits has Rx packet buffer allocation size in KB */
2713                 pba &= 0xffff;
2714                 /*
2715                  * the Tx fifo also stores 16 bytes of information about the tx
2716                  * but don't include ethernet FCS because hardware appends it
2717                  */
2718                 min_tx_space = (adapter->max_frame_size +
2719                                 sizeof(struct e1000_tx_desc) -
2720                                 ETH_FCS_LEN) * 2;
2721                 min_tx_space = ALIGN(min_tx_space, 1024);
2722                 min_tx_space >>= 10;
2723                 /* software strips receive CRC, so leave room for it */
2724                 min_rx_space = adapter->max_frame_size;
2725                 min_rx_space = ALIGN(min_rx_space, 1024);
2726                 min_rx_space >>= 10;
2727
2728                 /*
2729                  * If current Tx allocation is less than the min Tx FIFO size,
2730                  * and the min Tx FIFO size is less than the current Rx FIFO
2731                  * allocation, take space away from current Rx allocation
2732                  */
2733                 if ((tx_space < min_tx_space) &&
2734                     ((min_tx_space - tx_space) < pba)) {
2735                         pba -= min_tx_space - tx_space;
2736
2737                         /*
2738                          * if short on Rx space, Rx wins and must trump tx
2739                          * adjustment or use Early Receive if available
2740                          */
2741                         if ((pba < min_rx_space) &&
2742                             (!(adapter->flags & FLAG_HAS_ERT)))
2743                                 /* ERT enabled in e1000_configure_rx */
2744                                 pba = min_rx_space;
2745                 }
2746
2747                 ew32(PBA, pba);
2748         }
2749
2750
2751         /*
2752          * flow control settings
2753          *
2754          * The high water mark must be low enough to fit one full frame
2755          * (or the size used for early receive) above it in the Rx FIFO.
2756          * Set it to the lower of:
2757          * - 90% of the Rx FIFO size, and
2758          * - the full Rx FIFO size minus the early receive size (for parts
2759          *   with ERT support assuming ERT set to E1000_ERT_2048), or
2760          * - the full Rx FIFO size minus one full frame
2761          */
2762         if (adapter->flags & FLAG_HAS_ERT)
2763                 hwm = min(((pba << 10) * 9 / 10),
2764                           ((pba << 10) - (E1000_ERT_2048 << 3)));
2765         else
2766                 hwm = min(((pba << 10) * 9 / 10),
2767                           ((pba << 10) - adapter->max_frame_size));
2768
2769         fc->high_water = hwm & 0xFFF8; /* 8-byte granularity */
2770         fc->low_water = fc->high_water - 8;
2771
2772         if (adapter->flags & FLAG_DISABLE_FC_PAUSE_TIME)
2773                 fc->pause_time = 0xFFFF;
2774         else
2775                 fc->pause_time = E1000_FC_PAUSE_TIME;
2776         fc->send_xon = 1;
2777         fc->current_mode = fc->requested_mode;
2778
2779         /* Allow time for pending master requests to run */
2780         mac->ops.reset_hw(hw);
2781
2782         /*
2783          * For parts with AMT enabled, let the firmware know
2784          * that the network interface is in control
2785          */
2786         if (adapter->flags & FLAG_HAS_AMT)
2787                 e1000_get_hw_control(adapter);
2788
2789         ew32(WUC, 0);
2790
2791         if (mac->ops.init_hw(hw))
2792                 e_err("Hardware Error\n");
2793
2794         e1000_update_mng_vlan(adapter);
2795
2796         /* Enable h/w to recognize an 802.1Q VLAN Ethernet packet */
2797         ew32(VET, ETH_P_8021Q);
2798
2799         e1000e_reset_adaptive(hw);
2800         e1000_get_phy_info(hw);
2801
2802         if (!(adapter->flags & FLAG_SMART_POWER_DOWN)) {
2803                 u16 phy_data = 0;
2804                 /*
2805                  * speed up time to link by disabling smart power down, ignore
2806                  * the return value of this function because there is nothing
2807                  * different we would do if it failed
2808                  */
2809                 e1e_rphy(hw, IGP02E1000_PHY_POWER_MGMT, &phy_data);
2810                 phy_data &= ~IGP02E1000_PM_SPD;
2811                 e1e_wphy(hw, IGP02E1000_PHY_POWER_MGMT, phy_data);
2812         }
2813 }
2814
2815 int e1000e_up(struct e1000_adapter *adapter)
2816 {
2817         struct e1000_hw *hw = &adapter->hw;
2818
2819         /* hardware has been reset, we need to reload some things */
2820         e1000_configure(adapter);
2821
2822         clear_bit(__E1000_DOWN, &adapter->state);
2823
2824         napi_enable(&adapter->napi);
2825         if (adapter->msix_entries)
2826                 e1000_configure_msix(adapter);
2827         e1000_irq_enable(adapter);
2828
2829         netif_wake_queue(adapter->netdev);
2830
2831         /* fire a link change interrupt to start the watchdog */
2832         ew32(ICS, E1000_ICS_LSC);
2833         return 0;
2834 }
2835
2836 void e1000e_down(struct e1000_adapter *adapter)
2837 {
2838         struct net_device *netdev = adapter->netdev;
2839         struct e1000_hw *hw = &adapter->hw;
2840         u32 tctl, rctl;
2841
2842         /*
2843          * signal that we're down so the interrupt handler does not
2844          * reschedule our watchdog timer
2845          */
2846         set_bit(__E1000_DOWN, &adapter->state);
2847
2848         /* disable receives in the hardware */
2849         rctl = er32(RCTL);
2850         ew32(RCTL, rctl & ~E1000_RCTL_EN);
2851         /* flush and sleep below */
2852
2853         netif_stop_queue(netdev);
2854
2855         /* disable transmits in the hardware */
2856         tctl = er32(TCTL);
2857         tctl &= ~E1000_TCTL_EN;
2858         ew32(TCTL, tctl);
2859         /* flush both disables and wait for them to finish */
2860         e1e_flush();
2861         msleep(10);
2862
2863         napi_disable(&adapter->napi);
2864         e1000_irq_disable(adapter);
2865
2866         del_timer_sync(&adapter->watchdog_timer);
2867         del_timer_sync(&adapter->phy_info_timer);
2868
2869         netdev->tx_queue_len = adapter->tx_queue_len;
2870         netif_carrier_off(netdev);
2871         adapter->link_speed = 0;
2872         adapter->link_duplex = 0;
2873
2874         if (!pci_channel_offline(adapter->pdev))
2875                 e1000e_reset(adapter);
2876         e1000_clean_tx_ring(adapter);
2877         e1000_clean_rx_ring(adapter);
2878
2879         /*
2880          * TODO: for power management, we could drop the link and
2881          * pci_disable_device here.
2882          */
2883 }
2884
2885 void e1000e_reinit_locked(struct e1000_adapter *adapter)
2886 {
2887         might_sleep();
2888         while (test_and_set_bit(__E1000_RESETTING, &adapter->state))
2889                 msleep(1);
2890         e1000e_down(adapter);
2891         e1000e_up(adapter);
2892         clear_bit(__E1000_RESETTING, &adapter->state);
2893 }
2894
2895 /**
2896  * e1000_sw_init - Initialize general software structures (struct e1000_adapter)
2897  * @adapter: board private structure to initialize
2898  *
2899  * e1000_sw_init initializes the Adapter private data structure.
2900  * Fields are initialized based on PCI device information and
2901  * OS network device settings (MTU size).
2902  **/
2903 static int __devinit e1000_sw_init(struct e1000_adapter *adapter)
2904 {
2905         struct net_device *netdev = adapter->netdev;
2906
2907         adapter->rx_buffer_len = ETH_FRAME_LEN + VLAN_HLEN + ETH_FCS_LEN;
2908         adapter->rx_ps_bsize0 = 128;
2909         adapter->max_frame_size = netdev->mtu + ETH_HLEN + ETH_FCS_LEN;
2910         adapter->min_frame_size = ETH_ZLEN + ETH_FCS_LEN;
2911
2912         e1000e_set_interrupt_capability(adapter);
2913
2914         if (e1000_alloc_queues(adapter))
2915                 return -ENOMEM;
2916
2917         /* Explicitly disable IRQ since the NIC can be in any state. */
2918         e1000_irq_disable(adapter);
2919
2920         set_bit(__E1000_DOWN, &adapter->state);
2921         return 0;
2922 }
2923
2924 /**
2925  * e1000_intr_msi_test - Interrupt Handler
2926  * @irq: interrupt number
2927  * @data: pointer to a network interface device structure
2928  **/
2929 static irqreturn_t e1000_intr_msi_test(int irq, void *data)
2930 {
2931         struct net_device *netdev = data;
2932         struct e1000_adapter *adapter = netdev_priv(netdev);
2933         struct e1000_hw *hw = &adapter->hw;
2934         u32 icr = er32(ICR);
2935
2936         e_dbg("%s: icr is %08X\n", netdev->name, icr);
2937         if (icr & E1000_ICR_RXSEQ) {
2938                 adapter->flags &= ~FLAG_MSI_TEST_FAILED;
2939                 wmb();
2940         }
2941
2942         return IRQ_HANDLED;
2943 }
2944
2945 /**
2946  * e1000_test_msi_interrupt - Returns 0 for successful test
2947  * @adapter: board private struct
2948  *
2949  * code flow taken from tg3.c
2950  **/
2951 static int e1000_test_msi_interrupt(struct e1000_adapter *adapter)
2952 {
2953         struct net_device *netdev = adapter->netdev;
2954         struct e1000_hw *hw = &adapter->hw;
2955         int err;
2956
2957         /* poll_enable hasn't been called yet, so don't need disable */
2958         /* clear any pending events */
2959         er32(ICR);
2960
2961         /* free the real vector and request a test handler */
2962         e1000_free_irq(adapter);
2963         e1000e_reset_interrupt_capability(adapter);
2964
2965         /* Assume that the test fails, if it succeeds then the test
2966          * MSI irq handler will unset this flag */
2967         adapter->flags |= FLAG_MSI_TEST_FAILED;
2968
2969         err = pci_enable_msi(adapter->pdev);
2970         if (err)
2971                 goto msi_test_failed;
2972
2973         err = request_irq(adapter->pdev->irq, &e1000_intr_msi_test, 0,
2974                           netdev->name, netdev);
2975         if (err) {
2976                 pci_disable_msi(adapter->pdev);
2977                 goto msi_test_failed;
2978         }
2979
2980         wmb();
2981
2982         e1000_irq_enable(adapter);
2983
2984         /* fire an unusual interrupt on the test handler */
2985         ew32(ICS, E1000_ICS_RXSEQ);
2986         e1e_flush();
2987         msleep(50);
2988
2989         e1000_irq_disable(adapter);
2990
2991         rmb();
2992
2993         if (adapter->flags & FLAG_MSI_TEST_FAILED) {
2994                 adapter->int_mode = E1000E_INT_MODE_LEGACY;
2995                 err = -EIO;
2996                 e_info("MSI interrupt test failed!\n");
2997         }
2998
2999         free_irq(adapter->pdev->irq, netdev);
3000         pci_disable_msi(adapter->pdev);
3001
3002         if (err == -EIO)
3003                 goto msi_test_failed;
3004
3005         /* okay so the test worked, restore settings */
3006         e_dbg("%s: MSI interrupt test succeeded!\n", netdev->name);
3007 msi_test_failed:
3008         e1000e_set_interrupt_capability(adapter);
3009         e1000_request_irq(adapter);
3010         return err;
3011 }
3012
3013 /**
3014  * e1000_test_msi - Returns 0 if MSI test succeeds or INTx mode is restored
3015  * @adapter: board private struct
3016  *
3017  * code flow taken from tg3.c, called with e1000 interrupts disabled.
3018  **/
3019 static int e1000_test_msi(struct e1000_adapter *adapter)
3020 {
3021         int err;
3022         u16 pci_cmd;
3023
3024         if (!(adapter->flags & FLAG_MSI_ENABLED))
3025                 return 0;
3026
3027         /* disable SERR in case the MSI write causes a master abort */
3028         pci_read_config_word(adapter->pdev, PCI_COMMAND, &pci_cmd);
3029         pci_write_config_word(adapter->pdev, PCI_COMMAND,
3030                               pci_cmd & ~PCI_COMMAND_SERR);
3031
3032         err = e1000_test_msi_interrupt(adapter);
3033
3034         /* restore previous setting of command word */
3035         pci_write_config_word(adapter->pdev, PCI_COMMAND, pci_cmd);
3036
3037         /* success ! */
3038         if (!err)
3039                 return 0;
3040
3041         /* EIO means MSI test failed */
3042         if (err != -EIO)
3043                 return err;
3044
3045         /* back to INTx mode */
3046         e_warn("MSI interrupt test failed, using legacy interrupt.\n");
3047
3048         e1000_free_irq(adapter);
3049
3050         err = e1000_request_irq(adapter);
3051
3052         return err;
3053 }
3054
3055 /**
3056  * e1000_open - Called when a network interface is made active
3057  * @netdev: network interface device structure
3058  *
3059  * Returns 0 on success, negative value on failure
3060  *
3061  * The open entry point is called when a network interface is made
3062  * active by the system (IFF_UP).  At this point all resources needed
3063  * for transmit and receive operations are allocated, the interrupt
3064  * handler is registered with the OS, the watchdog timer is started,
3065  * and the stack is notified that the interface is ready.
3066  **/
3067 static int e1000_open(struct net_device *netdev)
3068 {
3069         struct e1000_adapter *adapter = netdev_priv(netdev);
3070         struct e1000_hw *hw = &adapter->hw;
3071         int err;
3072
3073         /* disallow open during test */
3074         if (test_bit(__E1000_TESTING, &adapter->state))
3075                 return -EBUSY;
3076
3077         netif_carrier_off(netdev);
3078
3079         /* allocate transmit descriptors */
3080         err = e1000e_setup_tx_resources(adapter);
3081         if (err)
3082                 goto err_setup_tx;
3083
3084         /* allocate receive descriptors */
3085         err = e1000e_setup_rx_resources(adapter);
3086         if (err)
3087                 goto err_setup_rx;
3088
3089         e1000e_power_up_phy(adapter);
3090
3091         adapter->mng_vlan_id = E1000_MNG_VLAN_NONE;
3092         if ((adapter->hw.mng_cookie.status &
3093              E1000_MNG_DHCP_COOKIE_STATUS_VLAN))
3094                 e1000_update_mng_vlan(adapter);
3095
3096         /*
3097          * If AMT is enabled, let the firmware know that the network
3098          * interface is now open
3099          */
3100         if (adapter->flags & FLAG_HAS_AMT)
3101                 e1000_get_hw_control(adapter);
3102
3103         /*
3104          * before we allocate an interrupt, we must be ready to handle it.
3105          * Setting DEBUG_SHIRQ in the kernel makes it fire an interrupt
3106          * as soon as we call pci_request_irq, so we have to setup our
3107          * clean_rx handler before we do so.
3108          */
3109         e1000_configure(adapter);
3110
3111         err = e1000_request_irq(adapter);
3112         if (err)
3113                 goto err_req_irq;
3114
3115         /*
3116          * Work around PCIe errata with MSI interrupts causing some chipsets to
3117          * ignore e1000e MSI messages, which means we need to test our MSI
3118          * interrupt now
3119          */
3120         if (adapter->int_mode != E1000E_INT_MODE_LEGACY) {
3121                 err = e1000_test_msi(adapter);
3122                 if (err) {
3123                         e_err("Interrupt allocation failed\n");
3124                         goto err_req_irq;
3125                 }
3126         }
3127
3128         /* From here on the code is the same as e1000e_up() */
3129         clear_bit(__E1000_DOWN, &adapter->state);
3130
3131         napi_enable(&adapter->napi);
3132
3133         e1000_irq_enable(adapter);
3134
3135         netif_start_queue(netdev);
3136
3137         /* fire a link status change interrupt to start the watchdog */
3138         ew32(ICS, E1000_ICS_LSC);
3139
3140         return 0;
3141
3142 err_req_irq:
3143         e1000_release_hw_control(adapter);
3144         e1000_power_down_phy(adapter);
3145         e1000e_free_rx_resources(adapter);
3146 err_setup_rx:
3147         e1000e_free_tx_resources(adapter);
3148 err_setup_tx:
3149         e1000e_reset(adapter);
3150
3151         return err;
3152 }
3153
3154 /**
3155  * e1000_close - Disables a network interface
3156  * @netdev: network interface device structure
3157  *
3158  * Returns 0, this is not allowed to fail
3159  *
3160  * The close entry point is called when an interface is de-activated
3161  * by the OS.  The hardware is still under the drivers control, but
3162  * needs to be disabled.  A global MAC reset is issued to stop the
3163  * hardware, and all transmit and receive resources are freed.
3164  **/
3165 static int e1000_close(struct net_device *netdev)
3166 {
3167         struct e1000_adapter *adapter = netdev_priv(netdev);
3168
3169         WARN_ON(test_bit(__E1000_RESETTING, &adapter->state));
3170         e1000e_down(adapter);
3171         e1000_power_down_phy(adapter);
3172         e1000_free_irq(adapter);
3173
3174         e1000e_free_tx_resources(adapter);
3175         e1000e_free_rx_resources(adapter);
3176
3177         /*
3178          * kill manageability vlan ID if supported, but not if a vlan with
3179          * the same ID is registered on the host OS (let 8021q kill it)
3180          */
3181         if ((adapter->hw.mng_cookie.status &
3182                           E1000_MNG_DHCP_COOKIE_STATUS_VLAN) &&
3183              !(adapter->vlgrp &&
3184                vlan_group_get_device(adapter->vlgrp, adapter->mng_vlan_id)))
3185                 e1000_vlan_rx_kill_vid(netdev, adapter->mng_vlan_id);
3186
3187         /*
3188          * If AMT is enabled, let the firmware know that the network
3189          * interface is now closed
3190          */
3191         if (adapter->flags & FLAG_HAS_AMT)
3192                 e1000_release_hw_control(adapter);
3193
3194         return 0;
3195 }
3196 /**
3197  * e1000_set_mac - Change the Ethernet Address of the NIC
3198  * @netdev: network interface device structure
3199  * @p: pointer to an address structure
3200  *
3201  * Returns 0 on success, negative on failure
3202  **/
3203 static int e1000_set_mac(struct net_device *netdev, void *p)
3204 {
3205         struct e1000_adapter *adapter = netdev_priv(netdev);
3206         struct sockaddr *addr = p;
3207
3208         if (!is_valid_ether_addr(addr->sa_data))
3209                 return -EADDRNOTAVAIL;
3210
3211         memcpy(netdev->dev_addr, addr->sa_data, netdev->addr_len);
3212         memcpy(adapter->hw.mac.addr, addr->sa_data, netdev->addr_len);
3213
3214         e1000e_rar_set(&adapter->hw, adapter->hw.mac.addr, 0);
3215
3216         if (adapter->flags & FLAG_RESET_OVERWRITES_LAA) {
3217                 /* activate the work around */
3218                 e1000e_set_laa_state_82571(&adapter->hw, 1);
3219
3220                 /*
3221                  * Hold a copy of the LAA in RAR[14] This is done so that
3222                  * between the time RAR[0] gets clobbered  and the time it
3223                  * gets fixed (in e1000_watchdog), the actual LAA is in one
3224                  * of the RARs and no incoming packets directed to this port
3225                  * are dropped. Eventually the LAA will be in RAR[0] and
3226                  * RAR[14]
3227                  */
3228                 e1000e_rar_set(&adapter->hw,
3229                               adapter->hw.mac.addr,
3230                               adapter->hw.mac.rar_entry_count - 1);
3231         }
3232
3233         return 0;
3234 }
3235
3236 /**
3237  * e1000e_update_phy_task - work thread to update phy
3238  * @work: pointer to our work struct
3239  *
3240  * this worker thread exists because we must acquire a
3241  * semaphore to read the phy, which we could msleep while
3242  * waiting for it, and we can't msleep in a timer.
3243  **/
3244 static void e1000e_update_phy_task(struct work_struct *work)
3245 {
3246         struct e1000_adapter *adapter = container_of(work,
3247                                         struct e1000_adapter, update_phy_task);
3248         e1000_get_phy_info(&adapter->hw);
3249 }
3250
3251 /*
3252  * Need to wait a few seconds after link up to get diagnostic information from
3253  * the phy
3254  */
3255 static void e1000_update_phy_info(unsigned long data)
3256 {
3257         struct e1000_adapter *adapter = (struct e1000_adapter *) data;
3258         schedule_work(&adapter->update_phy_task);
3259 }
3260
3261 /**
3262  * e1000e_update_stats - Update the board statistics counters
3263  * @adapter: board private structure
3264  **/
3265 void e1000e_update_stats(struct e1000_adapter *adapter)
3266 {
3267         struct e1000_hw *hw = &adapter->hw;
3268         struct pci_dev *pdev = adapter->pdev;
3269
3270         /*
3271          * Prevent stats update while adapter is being reset, or if the pci
3272          * connection is down.
3273          */
3274         if (adapter->link_speed == 0)
3275                 return;
3276         if (pci_channel_offline(pdev))
3277                 return;
3278
3279         adapter->stats.crcerrs += er32(CRCERRS);
3280         adapter->stats.gprc += er32(GPRC);
3281         adapter->stats.gorc += er32(GORCL);
3282         er32(GORCH); /* Clear gorc */
3283         adapter->stats.bprc += er32(BPRC);
3284         adapter->stats.mprc += er32(MPRC);
3285         adapter->stats.roc += er32(ROC);
3286
3287         adapter->stats.mpc += er32(MPC);
3288         adapter->stats.scc += er32(SCC);
3289         adapter->stats.ecol += er32(ECOL);
3290         adapter->stats.mcc += er32(MCC);
3291         adapter->stats.latecol += er32(LATECOL);
3292         adapter->stats.dc += er32(DC);
3293         adapter->stats.xonrxc += er32(XONRXC);
3294         adapter->stats.xontxc += er32(XONTXC);
3295         adapter->stats.xoffrxc += er32(XOFFRXC);
3296         adapter->stats.xofftxc += er32(XOFFTXC);
3297         adapter->stats.gptc += er32(GPTC);
3298         adapter->stats.gotc += er32(GOTCL);
3299         er32(GOTCH); /* Clear gotc */
3300         adapter->stats.rnbc += er32(RNBC);
3301         adapter->stats.ruc += er32(RUC);
3302
3303         adapter->stats.mptc += er32(MPTC);
3304         adapter->stats.bptc += er32(BPTC);
3305
3306         /* used for adaptive IFS */
3307
3308         hw->mac.tx_packet_delta = er32(TPT);
3309         adapter->stats.tpt += hw->mac.tx_packet_delta;
3310         hw->mac.collision_delta = er32(COLC);
3311         adapter->stats.colc += hw->mac.collision_delta;
3312
3313         adapter->stats.algnerrc += er32(ALGNERRC);
3314         adapter->stats.rxerrc += er32(RXERRC);
3315         if ((hw->mac.type != e1000_82574) && (hw->mac.type != e1000_82583))
3316                 adapter->stats.tncrs += er32(TNCRS);
3317         adapter->stats.cexterr += er32(CEXTERR);
3318         adapter->stats.tsctc += er32(TSCTC);
3319         adapter->stats.tsctfc += er32(TSCTFC);
3320
3321         /* Fill out the OS statistics structure */
3322         adapter->net_stats.multicast = adapter->stats.mprc;
3323         adapter->net_stats.collisions = adapter->stats.colc;
3324
3325         /* Rx Errors */
3326
3327         /*
3328          * RLEC on some newer hardware can be incorrect so build
3329          * our own version based on RUC and ROC
3330          */
3331         adapter->net_stats.rx_errors = adapter->stats.rxerrc +
3332                 adapter->stats.crcerrs + adapter->stats.algnerrc +
3333                 adapter->stats.ruc + adapter->stats.roc +
3334                 adapter->stats.cexterr;
3335         adapter->net_stats.rx_length_errors = adapter->stats.ruc +
3336                                               adapter->stats.roc;
3337         adapter->net_stats.rx_crc_errors = adapter->stats.crcerrs;
3338         adapter->net_stats.rx_frame_errors = adapter->stats.algnerrc;
3339         adapter->net_stats.rx_missed_errors = adapter->stats.mpc;
3340
3341         /* Tx Errors */
3342         adapter->net_stats.tx_errors = adapter->stats.ecol +
3343                                        adapter->stats.latecol;
3344         adapter->net_stats.tx_aborted_errors = adapter->stats.ecol;
3345         adapter->net_stats.tx_window_errors = adapter->stats.latecol;
3346         adapter->net_stats.tx_carrier_errors = adapter->stats.tncrs;
3347
3348         /* Tx Dropped needs to be maintained elsewhere */
3349
3350         /* Management Stats */
3351         adapter->stats.mgptc += er32(MGTPTC);
3352         adapter->stats.mgprc += er32(MGTPRC);
3353         adapter->stats.mgpdc += er32(MGTPDC);
3354 }
3355
3356 /**
3357  * e1000_phy_read_status - Update the PHY register status snapshot
3358  * @adapter: board private structure
3359  **/
3360 static void e1000_phy_read_status(struct e1000_adapter *adapter)
3361 {
3362         struct e1000_hw *hw = &adapter->hw;
3363         struct e1000_phy_regs *phy = &adapter->phy_regs;
3364         int ret_val;
3365
3366         if ((er32(STATUS) & E1000_STATUS_LU) &&
3367             (adapter->hw.phy.media_type == e1000_media_type_copper)) {
3368                 ret_val  = e1e_rphy(hw, PHY_CONTROL, &phy->bmcr);
3369                 ret_val |= e1e_rphy(hw, PHY_STATUS, &phy->bmsr);
3370                 ret_val |= e1e_rphy(hw, PHY_AUTONEG_ADV, &phy->advertise);
3371                 ret_val |= e1e_rphy(hw, PHY_LP_ABILITY, &phy->lpa);
3372                 ret_val |= e1e_rphy(hw, PHY_AUTONEG_EXP, &phy->expansion);
3373                 ret_val |= e1e_rphy(hw, PHY_1000T_CTRL, &phy->ctrl1000);
3374                 ret_val |= e1e_rphy(hw, PHY_1000T_STATUS, &phy->stat1000);
3375                 ret_val |= e1e_rphy(hw, PHY_EXT_STATUS, &phy->estatus);
3376                 if (ret_val)
3377                         e_warn("Error reading PHY register\n");
3378         } else {
3379                 /*
3380                  * Do not read PHY registers if link is not up
3381                  * Set values to typical power-on defaults
3382                  */
3383                 phy->bmcr = (BMCR_SPEED1000 | BMCR_ANENABLE | BMCR_FULLDPLX);
3384                 phy->bmsr = (BMSR_100FULL | BMSR_100HALF | BMSR_10FULL |
3385                              BMSR_10HALF | BMSR_ESTATEN | BMSR_ANEGCAPABLE |
3386                              BMSR_ERCAP);
3387                 phy->advertise = (ADVERTISE_PAUSE_ASYM | ADVERTISE_PAUSE_CAP |
3388                                   ADVERTISE_ALL | ADVERTISE_CSMA);
3389                 phy->lpa = 0;
3390                 phy->expansion = EXPANSION_ENABLENPAGE;
3391                 phy->ctrl1000 = ADVERTISE_1000FULL;
3392                 phy->stat1000 = 0;
3393                 phy->estatus = (ESTATUS_1000_TFULL | ESTATUS_1000_THALF);
3394         }
3395 }
3396
3397 static void e1000_print_link_info(struct e1000_adapter *adapter)
3398 {
3399         struct e1000_hw *hw = &adapter->hw;
3400         u32 ctrl = er32(CTRL);
3401
3402         /* Link status message must follow this format for user tools */
3403         printk(KERN_INFO "e1000e: %s NIC Link is Up %d Mbps %s, "
3404                "Flow Control: %s\n",
3405                adapter->netdev->name,
3406                adapter->link_speed,
3407                (adapter->link_duplex == FULL_DUPLEX) ?
3408                                 "Full Duplex" : "Half Duplex",
3409                ((ctrl & E1000_CTRL_TFCE) && (ctrl & E1000_CTRL_RFCE)) ?
3410                                 "RX/TX" :
3411                ((ctrl & E1000_CTRL_RFCE) ? "RX" :
3412                ((ctrl & E1000_CTRL_TFCE) ? "TX" : "None" )));
3413 }
3414
3415 bool e1000_has_link(struct e1000_adapter *adapter)
3416 {
3417         struct e1000_hw *hw = &adapter->hw;
3418         bool link_active = 0;
3419         s32 ret_val = 0;
3420
3421         /*
3422          * get_link_status is set on LSC (link status) interrupt or
3423          * Rx sequence error interrupt.  get_link_status will stay
3424          * false until the check_for_link establishes link
3425          * for copper adapters ONLY
3426          */
3427         switch (hw->phy.media_type) {
3428         case e1000_media_type_copper:
3429                 if (hw->mac.get_link_status) {
3430                         ret_val = hw->mac.ops.check_for_link(hw);
3431                         link_active = !hw->mac.get_link_status;
3432                 } else {
3433                         link_active = 1;
3434                 }
3435                 break;
3436         case e1000_media_type_fiber:
3437                 ret_val = hw->mac.ops.check_for_link(hw);
3438                 link_active = !!(er32(STATUS) & E1000_STATUS_LU);
3439                 break;
3440         case e1000_media_type_internal_serdes:
3441                 ret_val = hw->mac.ops.check_for_link(hw);
3442                 link_active = adapter->hw.mac.serdes_has_link;
3443                 break;
3444         default:
3445         case e1000_media_type_unknown:
3446                 break;
3447         }
3448
3449         if ((ret_val == E1000_ERR_PHY) && (hw->phy.type == e1000_phy_igp_3) &&
3450             (er32(CTRL) & E1000_PHY_CTRL_GBE_DISABLE)) {
3451                 /* See e1000_kmrn_lock_loss_workaround_ich8lan() */
3452                 e_info("Gigabit has been disabled, downgrading speed\n");
3453         }
3454
3455         return link_active;
3456 }
3457
3458 static void e1000e_enable_receives(struct e1000_adapter *adapter)
3459 {
3460         /* make sure the receive unit is started */
3461         if ((adapter->flags & FLAG_RX_NEEDS_RESTART) &&
3462             (adapter->flags & FLAG_RX_RESTART_NOW)) {
3463                 struct e1000_hw *hw = &adapter->hw;
3464                 u32 rctl = er32(RCTL);
3465                 ew32(RCTL, rctl | E1000_RCTL_EN);
3466                 adapter->flags &= ~FLAG_RX_RESTART_NOW;
3467         }
3468 }
3469
3470 /**
3471  * e1000_watchdog - Timer Call-back
3472  * @data: pointer to adapter cast into an unsigned long
3473  **/
3474 static void e1000_watchdog(unsigned long data)
3475 {
3476         struct e1000_adapter *adapter = (struct e1000_adapter *) data;
3477
3478         /* Do the rest outside of interrupt context */
3479         schedule_work(&adapter->watchdog_task);
3480
3481         /* TODO: make this use queue_delayed_work() */
3482 }
3483
3484 static void e1000_watchdog_task(struct work_struct *work)
3485 {
3486         struct e1000_adapter *adapter = container_of(work,
3487                                         struct e1000_adapter, watchdog_task);
3488         struct net_device *netdev = adapter->netdev;
3489         struct e1000_mac_info *mac = &adapter->hw.mac;
3490         struct e1000_phy_info *phy = &adapter->hw.phy;
3491         struct e1000_ring *tx_ring = adapter->tx_ring;
3492         struct e1000_hw *hw = &adapter->hw;
3493         u32 link, tctl;
3494         int tx_pending = 0;
3495
3496         link = e1000_has_link(adapter);
3497         if ((netif_carrier_ok(netdev)) && link) {
3498                 e1000e_enable_receives(adapter);
3499                 goto link_up;
3500         }
3501
3502         if ((e1000e_enable_tx_pkt_filtering(hw)) &&
3503             (adapter->mng_vlan_id != adapter->hw.mng_cookie.vlan_id))
3504                 e1000_update_mng_vlan(adapter);
3505
3506         if (link) {
3507                 if (!netif_carrier_ok(netdev)) {
3508                         bool txb2b = 1;
3509                         /* update snapshot of PHY registers on LSC */
3510                         e1000_phy_read_status(adapter);
3511                         mac->ops.get_link_up_info(&adapter->hw,
3512                                                    &adapter->link_speed,
3513                                                    &adapter->link_duplex);
3514                         e1000_print_link_info(adapter);
3515                         /*
3516                          * On supported PHYs, check for duplex mismatch only
3517                          * if link has autonegotiated at 10/100 half
3518                          */
3519                         if ((hw->phy.type == e1000_phy_igp_3 ||
3520                              hw->phy.type == e1000_phy_bm) &&
3521                             (hw->mac.autoneg == true) &&
3522                             (adapter->link_speed == SPEED_10 ||
3523                              adapter->link_speed == SPEED_100) &&
3524                             (adapter->link_duplex == HALF_DUPLEX)) {
3525                                 u16 autoneg_exp;
3526
3527                                 e1e_rphy(hw, PHY_AUTONEG_EXP, &autoneg_exp);
3528
3529                                 if (!(autoneg_exp & NWAY_ER_LP_NWAY_CAPS))
3530                                         e_info("Autonegotiated half duplex but"
3531                                                " link partner cannot autoneg. "
3532                                                " Try forcing full duplex if "
3533                                                "link gets many collisions.\n");
3534                         }
3535
3536                         /*
3537                          * tweak tx_queue_len according to speed/duplex
3538                          * and adjust the timeout factor
3539                          */
3540                         netdev->tx_queue_len = adapter->tx_queue_len;
3541                         adapter->tx_timeout_factor = 1;
3542                         switch (adapter->link_speed) {
3543                         case SPEED_10:
3544                                 txb2b = 0;
3545                                 netdev->tx_queue_len = 10;
3546                                 adapter->tx_timeout_factor = 16;
3547                                 break;
3548                         case SPEED_100:
3549                                 txb2b = 0;
3550                                 netdev->tx_queue_len = 100;
3551                                 /* maybe add some timeout factor ? */
3552                                 break;
3553                         }
3554
3555                         /*
3556                          * workaround: re-program speed mode bit after
3557                          * link-up event
3558                          */
3559                         if ((adapter->flags & FLAG_TARC_SPEED_MODE_BIT) &&
3560                             !txb2b) {
3561                                 u32 tarc0;
3562                                 tarc0 = er32(TARC(0));
3563                                 tarc0 &= ~SPEED_MODE_BIT;
3564                                 ew32(TARC(0), tarc0);
3565                         }
3566
3567                         /*
3568                          * disable TSO for pcie and 10/100 speeds, to avoid
3569                          * some hardware issues
3570                          */
3571                         if (!(adapter->flags & FLAG_TSO_FORCE)) {
3572                                 switch (adapter->link_speed) {
3573                                 case SPEED_10:
3574                                 case SPEED_100:
3575                                         e_info("10/100 speed: disabling TSO\n");
3576                                         netdev->features &= ~NETIF_F_TSO;
3577                                         netdev->features &= ~NETIF_F_TSO6;
3578                                         break;
3579                                 case SPEED_1000:
3580                                         netdev->features |= NETIF_F_TSO;
3581                                         netdev->features |= NETIF_F_TSO6;
3582                                         break;
3583                                 default:
3584                                         /* oops */
3585                                         break;
3586                                 }
3587                         }
3588
3589                         /*
3590                          * enable transmits in the hardware, need to do this
3591                          * after setting TARC(0)
3592                          */
3593                         tctl = er32(TCTL);
3594                         tctl |= E1000_TCTL_EN;
3595                         ew32(TCTL, tctl);
3596
3597                         /*
3598                          * Perform any post-link-up configuration before
3599                          * reporting link up.
3600                          */
3601                         if (phy->ops.cfg_on_link_up)
3602                                 phy->ops.cfg_on_link_up(hw);
3603
3604                         netif_carrier_on(netdev);
3605
3606                         if (!test_bit(__E1000_DOWN, &adapter->state))
3607                                 mod_timer(&adapter->phy_info_timer,
3608                                           round_jiffies(jiffies + 2 * HZ));
3609                 }
3610         } else {
3611                 if (netif_carrier_ok(netdev)) {
3612                         adapter->link_speed = 0;
3613                         adapter->link_duplex = 0;
3614                         /* Link status message must follow this format */
3615                         printk(KERN_INFO "e1000e: %s NIC Link is Down\n",
3616                                adapter->netdev->name);
3617                         netif_carrier_off(netdev);
3618                         if (!test_bit(__E1000_DOWN, &adapter->state))
3619                                 mod_timer(&adapter->phy_info_timer,
3620                                           round_jiffies(jiffies + 2 * HZ));
3621
3622                         if (adapter->flags & FLAG_RX_NEEDS_RESTART)
3623                                 schedule_work(&adapter->reset_task);
3624                 }
3625         }
3626
3627 link_up:
3628         e1000e_update_stats(adapter);
3629
3630         mac->tx_packet_delta = adapter->stats.tpt - adapter->tpt_old;
3631         adapter->tpt_old = adapter->stats.tpt;
3632         mac->collision_delta = adapter->stats.colc - adapter->colc_old;
3633         adapter->colc_old = adapter->stats.colc;
3634
3635         adapter->gorc = adapter->stats.gorc - adapter->gorc_old;
3636         adapter->gorc_old = adapter->stats.gorc;
3637         adapter->gotc = adapter->stats.gotc - adapter->gotc_old;
3638         adapter->gotc_old = adapter->stats.gotc;
3639
3640         e1000e_update_adaptive(&adapter->hw);
3641
3642         if (!netif_carrier_ok(netdev)) {
3643                 tx_pending = (e1000_desc_unused(tx_ring) + 1 <
3644                                tx_ring->count);
3645                 if (tx_pending) {
3646                         /*
3647                          * We've lost link, so the controller stops DMA,
3648                          * but we've got queued Tx work that's never going
3649                          * to get done, so reset controller to flush Tx.
3650                          * (Do the reset outside of interrupt context).
3651                          */
3652                         adapter->tx_timeout_count++;
3653                         schedule_work(&adapter->reset_task);
3654                         /* return immediately since reset is imminent */
3655                         return;
3656                 }
3657         }
3658
3659         /* Cause software interrupt to ensure Rx ring is cleaned */
3660         if (adapter->msix_entries)
3661                 ew32(ICS, adapter->rx_ring->ims_val);
3662         else
3663                 ew32(ICS, E1000_ICS_RXDMT0);
3664
3665         /* Force detection of hung controller every watchdog period */
3666         adapter->detect_tx_hung = 1;
3667
3668         /*
3669          * With 82571 controllers, LAA may be overwritten due to controller
3670          * reset from the other port. Set the appropriate LAA in RAR[0]
3671          */
3672         if (e1000e_get_laa_state_82571(hw))
3673                 e1000e_rar_set(hw, adapter->hw.mac.addr, 0);
3674
3675         /* Reset the timer */
3676         if (!test_bit(__E1000_DOWN, &adapter->state))
3677                 mod_timer(&adapter->watchdog_timer,
3678                           round_jiffies(jiffies + 2 * HZ));
3679 }
3680
3681 #define E1000_TX_FLAGS_CSUM             0x00000001
3682 #define E1000_TX_FLAGS_VLAN             0x00000002
3683 #define E1000_TX_FLAGS_TSO              0x00000004
3684 #define E1000_TX_FLAGS_IPV4             0x00000008
3685 #define E1000_TX_FLAGS_VLAN_MASK        0xffff0000
3686 #define E1000_TX_FLAGS_VLAN_SHIFT       16
3687
3688 static int e1000_tso(struct e1000_adapter *adapter,
3689                      struct sk_buff *skb)
3690 {
3691         struct e1000_ring *tx_ring = adapter->tx_ring;
3692         struct e1000_context_desc *context_desc;
3693         struct e1000_buffer *buffer_info;
3694         unsigned int i;
3695         u32 cmd_length = 0;
3696         u16 ipcse = 0, tucse, mss;
3697         u8 ipcss, ipcso, tucss, tucso, hdr_len;
3698         int err;
3699
3700         if (skb_is_gso(skb)) {
3701                 if (skb_header_cloned(skb)) {
3702                         err = pskb_expand_head(skb, 0, 0, GFP_ATOMIC);
3703                         if (err)
3704                                 return err;
3705                 }
3706
3707                 hdr_len = skb_transport_offset(skb) + tcp_hdrlen(skb);
3708                 mss = skb_shinfo(skb)->gso_size;
3709                 if (skb->protocol == htons(ETH_P_IP)) {
3710                         struct iphdr *iph = ip_hdr(skb);
3711                         iph->tot_len = 0;
3712                         iph->check = 0;
3713                         tcp_hdr(skb)->check = ~csum_tcpudp_magic(iph->saddr,
3714                                                                  iph->daddr, 0,
3715                                                                  IPPROTO_TCP,
3716                                                                  0);
3717                         cmd_length = E1000_TXD_CMD_IP;
3718                         ipcse = skb_transport_offset(skb) - 1;
3719                 } else if (skb_shinfo(skb)->gso_type == SKB_GSO_TCPV6) {
3720                         ipv6_hdr(skb)->payload_len = 0;
3721                         tcp_hdr(skb)->check =
3722                                 ~csum_ipv6_magic(&ipv6_hdr(skb)->saddr,
3723                                                  &ipv6_hdr(skb)->daddr,
3724                                                  0, IPPROTO_TCP, 0);
3725                         ipcse = 0;
3726                 }
3727                 ipcss = skb_network_offset(skb);
3728                 ipcso = (void *)&(ip_hdr(skb)->check) - (void *)skb->data;
3729                 tucss = skb_transport_offset(skb);
3730                 tucso = (void *)&(tcp_hdr(skb)->check) - (void *)skb->data;
3731                 tucse = 0;
3732
3733                 cmd_length |= (E1000_TXD_CMD_DEXT | E1000_TXD_CMD_TSE |
3734                                E1000_TXD_CMD_TCP | (skb->len - (hdr_len)));
3735
3736                 i = tx_ring->next_to_use;
3737                 context_desc = E1000_CONTEXT_DESC(*tx_ring, i);
3738                 buffer_info = &tx_ring->buffer_info[i];
3739
3740                 context_desc->lower_setup.ip_fields.ipcss  = ipcss;
3741                 context_desc->lower_setup.ip_fields.ipcso  = ipcso;
3742                 context_desc->lower_setup.ip_fields.ipcse  = cpu_to_le16(ipcse);
3743                 context_desc->upper_setup.tcp_fields.tucss = tucss;
3744                 context_desc->upper_setup.tcp_fields.tucso = tucso;
3745                 context_desc->upper_setup.tcp_fields.tucse = cpu_to_le16(tucse);
3746                 context_desc->tcp_seg_setup.fields.mss     = cpu_to_le16(mss);
3747                 context_desc->tcp_seg_setup.fields.hdr_len = hdr_len;
3748                 context_desc->cmd_and_length = cpu_to_le32(cmd_length);
3749
3750                 buffer_info->time_stamp = jiffies;
3751                 buffer_info->next_to_watch = i;
3752
3753                 i++;
3754                 if (i == tx_ring->count)
3755                         i = 0;
3756                 tx_ring->next_to_use = i;
3757
3758                 return 1;
3759         }
3760
3761         return 0;
3762 }
3763
3764 static bool e1000_tx_csum(struct e1000_adapter *adapter, struct sk_buff *skb)
3765 {
3766         struct e1000_ring *tx_ring = adapter->tx_ring;
3767         struct e1000_context_desc *context_desc;
3768         struct e1000_buffer *buffer_info;
3769         unsigned int i;
3770         u8 css;
3771         u32 cmd_len = E1000_TXD_CMD_DEXT;
3772         __be16 protocol;
3773
3774         if (skb->ip_summed != CHECKSUM_PARTIAL)
3775                 return 0;
3776
3777         if (skb->protocol == cpu_to_be16(ETH_P_8021Q))
3778                 protocol = vlan_eth_hdr(skb)->h_vlan_encapsulated_proto;
3779         else
3780                 protocol = skb->protocol;
3781
3782         switch (protocol) {
3783         case cpu_to_be16(ETH_P_IP):
3784                 if (ip_hdr(skb)->protocol == IPPROTO_TCP)
3785                         cmd_len |= E1000_TXD_CMD_TCP;
3786                 break;
3787         case cpu_to_be16(ETH_P_IPV6):
3788                 /* XXX not handling all IPV6 headers */
3789                 if (ipv6_hdr(skb)->nexthdr == IPPROTO_TCP)
3790                         cmd_len |= E1000_TXD_CMD_TCP;
3791                 break;
3792         default:
3793                 if (unlikely(net_ratelimit()))
3794                         e_warn("checksum_partial proto=%x!\n",
3795                                be16_to_cpu(protocol));
3796                 break;
3797         }
3798
3799         css = skb_transport_offset(skb);
3800
3801         i = tx_ring->next_to_use;
3802         buffer_info = &tx_ring->buffer_info[i];
3803         context_desc = E1000_CONTEXT_DESC(*tx_ring, i);
3804
3805         context_desc->lower_setup.ip_config = 0;
3806         context_desc->upper_setup.tcp_fields.tucss = css;
3807         context_desc->upper_setup.tcp_fields.tucso =
3808                                 css + skb->csum_offset;
3809         context_desc->upper_setup.tcp_fields.tucse = 0;
3810         context_desc->tcp_seg_setup.data = 0;
3811         context_desc->cmd_and_length = cpu_to_le32(cmd_len);
3812
3813         buffer_info->time_stamp = jiffies;
3814         buffer_info->next_to_watch = i;
3815
3816         i++;
3817         if (i == tx_ring->count)
3818                 i = 0;
3819         tx_ring->next_to_use = i;
3820
3821         return 1;
3822 }
3823
3824 #define E1000_MAX_PER_TXD       8192
3825 #define E1000_MAX_TXD_PWR       12
3826
3827 static int e1000_tx_map(struct e1000_adapter *adapter,
3828                         struct sk_buff *skb, unsigned int first,
3829                         unsigned int max_per_txd, unsigned int nr_frags,
3830                         unsigned int mss)
3831 {
3832         struct e1000_ring *tx_ring = adapter->tx_ring;
3833         struct e1000_buffer *buffer_info;
3834         unsigned int len = skb_headlen(skb);
3835         unsigned int offset, size, count = 0, i;
3836         unsigned int f;
3837         dma_addr_t *map;
3838
3839         i = tx_ring->next_to_use;
3840
3841         if (skb_dma_map(&adapter->pdev->dev, skb, DMA_TO_DEVICE)) {
3842                 dev_err(&adapter->pdev->dev, "TX DMA map failed\n");
3843                 adapter->tx_dma_failed++;
3844                 return 0;
3845         }
3846
3847         map = skb_shinfo(skb)->dma_maps;
3848         offset = 0;
3849
3850         while (len) {
3851                 buffer_info = &tx_ring->buffer_info[i];
3852                 size = min(len, max_per_txd);
3853
3854                 buffer_info->length = size;
3855                 buffer_info->time_stamp = jiffies;
3856                 buffer_info->next_to_watch = i;
3857                 buffer_info->dma = map[0] + offset;
3858                 count++;
3859
3860                 len -= size;
3861                 offset += size;
3862
3863                 if (len) {
3864                         i++;
3865                         if (i == tx_ring->count)
3866                                 i = 0;
3867                 }
3868         }
3869
3870         for (f = 0; f < nr_frags; f++) {
3871                 struct skb_frag_struct *frag;
3872
3873                 frag = &skb_shinfo(skb)->frags[f];
3874                 len = frag->size;
3875                 offset = 0;
3876
3877                 while (len) {
3878                         i++;
3879                         if (i == tx_ring->count)
3880                                 i = 0;
3881
3882                         buffer_info = &tx_ring->buffer_info[i];
3883                         size = min(len, max_per_txd);
3884
3885                         buffer_info->length = size;
3886                         buffer_info->time_stamp = jiffies;
3887                         buffer_info->next_to_watch = i;
3888                         buffer_info->dma = map[f + 1] + offset;
3889
3890                         len -= size;
3891                         offset += size;
3892                         count++;
3893                 }
3894         }
3895
3896         tx_ring->buffer_info[i].skb = skb;
3897         tx_ring->buffer_info[first].next_to_watch = i;
3898
3899         return count;
3900 }
3901
3902 static void e1000_tx_queue(struct e1000_adapter *adapter,
3903                            int tx_flags, int count)
3904 {
3905         struct e1000_ring *tx_ring = adapter->tx_ring;
3906         struct e1000_tx_desc *tx_desc = NULL;
3907         struct e1000_buffer *buffer_info;
3908         u32 txd_upper = 0, txd_lower = E1000_TXD_CMD_IFCS;
3909         unsigned int i;
3910
3911         if (tx_flags & E1000_TX_FLAGS_TSO) {
3912                 txd_lower |= E1000_TXD_CMD_DEXT | E1000_TXD_DTYP_D |
3913                              E1000_TXD_CMD_TSE;
3914                 txd_upper |= E1000_TXD_POPTS_TXSM << 8;
3915
3916                 if (tx_flags & E1000_TX_FLAGS_IPV4)
3917                         txd_upper |= E1000_TXD_POPTS_IXSM << 8;
3918         }
3919
3920         if (tx_flags & E1000_TX_FLAGS_CSUM) {
3921                 txd_lower |= E1000_TXD_CMD_DEXT | E1000_TXD_DTYP_D;
3922                 txd_upper |= E1000_TXD_POPTS_TXSM << 8;
3923         }
3924
3925         if (tx_flags & E1000_TX_FLAGS_VLAN) {
3926                 txd_lower |= E1000_TXD_CMD_VLE;
3927                 txd_upper |= (tx_flags & E1000_TX_FLAGS_VLAN_MASK);
3928         }
3929
3930         i = tx_ring->next_to_use;
3931
3932         while (count--) {
3933                 buffer_info = &tx_ring->buffer_info[i];
3934                 tx_desc = E1000_TX_DESC(*tx_ring, i);
3935                 tx_desc->buffer_addr = cpu_to_le64(buffer_info->dma);
3936                 tx_desc->lower.data =
3937                         cpu_to_le32(txd_lower | buffer_info->length);
3938                 tx_desc->upper.data = cpu_to_le32(txd_upper);
3939
3940                 i++;
3941                 if (i == tx_ring->count)
3942                         i = 0;
3943         }
3944
3945         tx_desc->lower.data |= cpu_to_le32(adapter->txd_cmd);
3946
3947         /*
3948          * Force memory writes to complete before letting h/w
3949          * know there are new descriptors to fetch.  (Only
3950          * applicable for weak-ordered memory model archs,
3951          * such as IA-64).
3952          */
3953         wmb();
3954
3955         tx_ring->next_to_use = i;
3956         writel(i, adapter->hw.hw_addr + tx_ring->tail);
3957         /*
3958          * we need this if more than one processor can write to our tail
3959          * at a time, it synchronizes IO on IA64/Altix systems
3960          */
3961         mmiowb();
3962 }
3963
3964 #define MINIMUM_DHCP_PACKET_SIZE 282
3965 static int e1000_transfer_dhcp_info(struct e1000_adapter *adapter,
3966                                     struct sk_buff *skb)
3967 {
3968         struct e1000_hw *hw =  &adapter->hw;
3969         u16 length, offset;
3970
3971         if (vlan_tx_tag_present(skb)) {
3972                 if (!((vlan_tx_tag_get(skb) == adapter->hw.mng_cookie.vlan_id)
3973                     && (adapter->hw.mng_cookie.status &
3974                         E1000_MNG_DHCP_COOKIE_STATUS_VLAN)))
3975                         return 0;
3976         }
3977
3978         if (skb->len <= MINIMUM_DHCP_PACKET_SIZE)
3979                 return 0;
3980
3981         if (((struct ethhdr *) skb->data)->h_proto != htons(ETH_P_IP))
3982                 return 0;
3983
3984         {
3985                 const struct iphdr *ip = (struct iphdr *)((u8 *)skb->data+14);
3986                 struct udphdr *udp;
3987
3988                 if (ip->protocol != IPPROTO_UDP)
3989                         return 0;
3990
3991                 udp = (struct udphdr *)((u8 *)ip + (ip->ihl << 2));
3992                 if (ntohs(udp->dest) != 67)
3993                         return 0;
3994
3995                 offset = (u8 *)udp + 8 - skb->data;
3996                 length = skb->len - offset;
3997                 return e1000e_mng_write_dhcp_info(hw, (u8 *)udp + 8, length);
3998         }
3999
4000         return 0;
4001 }
4002
4003 static int __e1000_maybe_stop_tx(struct net_device *netdev, int size)
4004 {
4005         struct e1000_adapter *adapter = netdev_priv(netdev);
4006
4007         netif_stop_queue(netdev);
4008         /*
4009          * Herbert's original patch had:
4010          *  smp_mb__after_netif_stop_queue();
4011          * but since that doesn't exist yet, just open code it.
4012          */
4013         smp_mb();
4014
4015         /*
4016          * We need to check again in a case another CPU has just
4017          * made room available.
4018          */
4019         if (e1000_desc_unused(adapter->tx_ring) < size)
4020                 return -EBUSY;
4021
4022         /* A reprieve! */
4023         netif_start_queue(netdev);
4024         ++adapter->restart_queue;
4025         return 0;
4026 }
4027
4028 static int e1000_maybe_stop_tx(struct net_device *netdev, int size)
4029 {
4030         struct e1000_adapter *adapter = netdev_priv(netdev);
4031
4032         if (e1000_desc_unused(adapter->tx_ring) >= size)
4033                 return 0;
4034         return __e1000_maybe_stop_tx(netdev, size);
4035 }
4036
4037 #define TXD_USE_COUNT(S, X) (((S) >> (X)) + 1 )
4038 static int e1000_xmit_frame(struct sk_buff *skb, struct net_device *netdev)
4039 {
4040         struct e1000_adapter *adapter = netdev_priv(netdev);
4041         struct e1000_ring *tx_ring = adapter->tx_ring;
4042         unsigned int first;
4043         unsigned int max_per_txd = E1000_MAX_PER_TXD;
4044         unsigned int max_txd_pwr = E1000_MAX_TXD_PWR;
4045         unsigned int tx_flags = 0;
4046         unsigned int len = skb->len - skb->data_len;
4047         unsigned int nr_frags;
4048         unsigned int mss;
4049         int count = 0;
4050         int tso;
4051         unsigned int f;
4052
4053         if (test_bit(__E1000_DOWN, &adapter->state)) {
4054                 dev_kfree_skb_any(skb);
4055                 return NETDEV_TX_OK;
4056         }
4057
4058         if (skb->len <= 0) {
4059                 dev_kfree_skb_any(skb);
4060                 return NETDEV_TX_OK;
4061         }
4062
4063         mss = skb_shinfo(skb)->gso_size;
4064         /*
4065          * The controller does a simple calculation to
4066          * make sure there is enough room in the FIFO before
4067          * initiating the DMA for each buffer.  The calc is:
4068          * 4 = ceil(buffer len/mss).  To make sure we don't
4069          * overrun the FIFO, adjust the max buffer len if mss
4070          * drops.
4071          */
4072         if (mss) {
4073                 u8 hdr_len;
4074                 max_per_txd = min(mss << 2, max_per_txd);
4075                 max_txd_pwr = fls(max_per_txd) - 1;
4076
4077                 /*
4078                  * TSO Workaround for 82571/2/3 Controllers -- if skb->data
4079                  * points to just header, pull a few bytes of payload from
4080                  * frags into skb->data
4081                  */
4082                 hdr_len = skb_transport_offset(skb) + tcp_hdrlen(skb);
4083                 /*
4084                  * we do this workaround for ES2LAN, but it is un-necessary,
4085                  * avoiding it could save a lot of cycles
4086                  */
4087                 if (skb->data_len && (hdr_len == len)) {
4088                         unsigned int pull_size;
4089
4090                         pull_size = min((unsigned int)4, skb->data_len);
4091                         if (!__pskb_pull_tail(skb, pull_size)) {
4092                                 e_err("__pskb_pull_tail failed.\n");
4093                                 dev_kfree_skb_any(skb);
4094                                 return NETDEV_TX_OK;
4095                         }
4096                         len = skb->len - skb->data_len;
4097                 }
4098         }
4099
4100         /* reserve a descriptor for the offload context */
4101         if ((mss) || (skb->ip_summed == CHECKSUM_PARTIAL))
4102                 count++;
4103         count++;
4104
4105         count += TXD_USE_COUNT(len, max_txd_pwr);
4106
4107         nr_frags = skb_shinfo(skb)->nr_frags;
4108         for (f = 0; f < nr_frags; f++)
4109                 count += TXD_USE_COUNT(skb_shinfo(skb)->frags[f].size,
4110                                        max_txd_pwr);
4111
4112         if (adapter->hw.mac.tx_pkt_filtering)
4113                 e1000_transfer_dhcp_info(adapter, skb);
4114
4115         /*
4116          * need: count + 2 desc gap to keep tail from touching
4117          * head, otherwise try next time
4118          */
4119         if (e1000_maybe_stop_tx(netdev, count + 2))
4120                 return NETDEV_TX_BUSY;
4121
4122         if (adapter->vlgrp && vlan_tx_tag_present(skb)) {
4123                 tx_flags |= E1000_TX_FLAGS_VLAN;
4124                 tx_flags |= (vlan_tx_tag_get(skb) << E1000_TX_FLAGS_VLAN_SHIFT);
4125         }
4126
4127         first = tx_ring->next_to_use;
4128
4129         tso = e1000_tso(adapter, skb);
4130         if (tso < 0) {
4131                 dev_kfree_skb_any(skb);
4132                 return NETDEV_TX_OK;
4133         }
4134
4135         if (tso)
4136                 tx_flags |= E1000_TX_FLAGS_TSO;
4137         else if (e1000_tx_csum(adapter, skb))
4138                 tx_flags |= E1000_TX_FLAGS_CSUM;
4139
4140         /*
4141          * Old method was to assume IPv4 packet by default if TSO was enabled.
4142          * 82571 hardware supports TSO capabilities for IPv6 as well...
4143          * no longer assume, we must.
4144          */
4145         if (skb->protocol == htons(ETH_P_IP))
4146                 tx_flags |= E1000_TX_FLAGS_IPV4;
4147
4148         /* if count is 0 then mapping error has occured */
4149         count = e1000_tx_map(adapter, skb, first, max_per_txd, nr_frags, mss);
4150         if (count) {
4151                 e1000_tx_queue(adapter, tx_flags, count);
4152                 netdev->trans_start = jiffies;
4153                 /* Make sure there is space in the ring for the next send. */
4154                 e1000_maybe_stop_tx(netdev, MAX_SKB_FRAGS + 2);
4155
4156         } else {
4157                 dev_kfree_skb_any(skb);
4158                 tx_ring->buffer_info[first].time_stamp = 0;
4159                 tx_ring->next_to_use = first;
4160         }
4161
4162         return NETDEV_TX_OK;
4163 }
4164
4165 /**
4166  * e1000_tx_timeout - Respond to a Tx Hang
4167  * @netdev: network interface device structure
4168  **/
4169 static void e1000_tx_timeout(struct net_device *netdev)
4170 {
4171         struct e1000_adapter *adapter = netdev_priv(netdev);
4172
4173         /* Do the reset outside of interrupt context */
4174         adapter->tx_timeout_count++;
4175         schedule_work(&adapter->reset_task);
4176 }
4177
4178 static void e1000_reset_task(struct work_struct *work)
4179 {
4180         struct e1000_adapter *adapter;
4181         adapter = container_of(work, struct e1000_adapter, reset_task);
4182
4183         e1000e_reinit_locked(adapter);
4184 }
4185
4186 /**
4187  * e1000_get_stats - Get System Network Statistics
4188  * @netdev: network interface device structure
4189  *
4190  * Returns the address of the device statistics structure.
4191  * The statistics are actually updated from the timer callback.
4192  **/
4193 static struct net_device_stats *e1000_get_stats(struct net_device *netdev)
4194 {
4195         struct e1000_adapter *adapter = netdev_priv(netdev);
4196
4197         /* only return the current stats */
4198         return &adapter->net_stats;
4199 }
4200
4201 /**
4202  * e1000_change_mtu - Change the Maximum Transfer Unit
4203  * @netdev: network interface device structure
4204  * @new_mtu: new value for maximum frame size
4205  *
4206  * Returns 0 on success, negative on failure
4207  **/
4208 static int e1000_change_mtu(struct net_device *netdev, int new_mtu)
4209 {
4210         struct e1000_adapter *adapter = netdev_priv(netdev);
4211         int max_frame = new_mtu + ETH_HLEN + ETH_FCS_LEN;
4212
4213         if ((new_mtu < ETH_ZLEN + ETH_FCS_LEN + VLAN_HLEN) ||
4214             (max_frame > MAX_JUMBO_FRAME_SIZE)) {
4215                 e_err("Invalid MTU setting\n");
4216                 return -EINVAL;
4217         }
4218
4219         /* Jumbo frame size limits */
4220         if (max_frame > ETH_FRAME_LEN + ETH_FCS_LEN) {
4221                 if (!(adapter->flags & FLAG_HAS_JUMBO_FRAMES)) {
4222                         e_err("Jumbo Frames not supported.\n");
4223                         return -EINVAL;
4224                 }
4225                 if (adapter->hw.phy.type == e1000_phy_ife) {
4226                         e_err("Jumbo Frames not supported.\n");
4227                         return -EINVAL;
4228                 }
4229         }
4230
4231 #define MAX_STD_JUMBO_FRAME_SIZE 9234
4232         if (max_frame > MAX_STD_JUMBO_FRAME_SIZE) {
4233                 e_err("MTU > 9216 not supported.\n");
4234                 return -EINVAL;
4235         }
4236
4237         while (test_and_set_bit(__E1000_RESETTING, &adapter->state))
4238                 msleep(1);
4239         /* e1000e_down has a dependency on max_frame_size */
4240         adapter->max_frame_size = max_frame;
4241         if (netif_running(netdev))
4242                 e1000e_down(adapter);
4243
4244         /*
4245          * NOTE: netdev_alloc_skb reserves 16 bytes, and typically NET_IP_ALIGN
4246          * means we reserve 2 more, this pushes us to allocate from the next
4247          * larger slab size.
4248          * i.e. RXBUFFER_2048 --> size-4096 slab
4249          * However with the new *_jumbo_rx* routines, jumbo receives will use
4250          * fragmented skbs
4251          */
4252
4253         if (max_frame <= 256)
4254                 adapter->rx_buffer_len = 256;
4255         else if (max_frame <= 512)
4256                 adapter->rx_buffer_len = 512;
4257         else if (max_frame <= 1024)
4258                 adapter->rx_buffer_len = 1024;
4259         else if (max_frame <= 2048)
4260                 adapter->rx_buffer_len = 2048;
4261         else
4262                 adapter->rx_buffer_len = 4096;
4263
4264         /* adjust allocation if LPE protects us, and we aren't using SBP */
4265         if ((max_frame == ETH_FRAME_LEN + ETH_FCS_LEN) ||
4266              (max_frame == ETH_FRAME_LEN + VLAN_HLEN + ETH_FCS_LEN))
4267                 adapter->rx_buffer_len = ETH_FRAME_LEN + VLAN_HLEN
4268                                          + ETH_FCS_LEN;
4269
4270         e_info("changing MTU from %d to %d\n", netdev->mtu, new_mtu);
4271         netdev->mtu = new_mtu;
4272
4273         if (netif_running(netdev))
4274                 e1000e_up(adapter);
4275         else
4276                 e1000e_reset(adapter);
4277
4278         clear_bit(__E1000_RESETTING, &adapter->state);
4279
4280         return 0;
4281 }
4282
4283 static int e1000_mii_ioctl(struct net_device *netdev, struct ifreq *ifr,
4284                            int cmd)
4285 {
4286         struct e1000_adapter *adapter = netdev_priv(netdev);
4287         struct mii_ioctl_data *data = if_mii(ifr);
4288
4289         if (adapter->hw.phy.media_type != e1000_media_type_copper)
4290                 return -EOPNOTSUPP;
4291
4292         switch (cmd) {
4293         case SIOCGMIIPHY:
4294                 data->phy_id = adapter->hw.phy.addr;
4295                 break;
4296         case SIOCGMIIREG:
4297                 if (!capable(CAP_NET_ADMIN))
4298                         return -EPERM;
4299                 switch (data->reg_num & 0x1F) {
4300                 case MII_BMCR:
4301                         data->val_out = adapter->phy_regs.bmcr;
4302                         break;
4303                 case MII_BMSR:
4304                         data->val_out = adapter->phy_regs.bmsr;
4305                         break;
4306                 case MII_PHYSID1:
4307                         data->val_out = (adapter->hw.phy.id >> 16);
4308                         break;
4309                 case MII_PHYSID2:
4310                         data->val_out = (adapter->hw.phy.id & 0xFFFF);
4311                         break;
4312                 case MII_ADVERTISE:
4313                         data->val_out = adapter->phy_regs.advertise;
4314                         break;
4315                 case MII_LPA:
4316                         data->val_out = adapter->phy_regs.lpa;
4317                         break;
4318                 case MII_EXPANSION:
4319                         data->val_out = adapter->phy_regs.expansion;
4320                         break;
4321                 case MII_CTRL1000:
4322                         data->val_out = adapter->phy_regs.ctrl1000;
4323                         break;
4324                 case MII_STAT1000:
4325                         data->val_out = adapter->phy_regs.stat1000;
4326                         break;
4327                 case MII_ESTATUS:
4328                         data->val_out = adapter->phy_regs.estatus;
4329                         break;
4330                 default:
4331                         return -EIO;
4332                 }
4333                 break;
4334         case SIOCSMIIREG:
4335         default:
4336                 return -EOPNOTSUPP;
4337         }
4338         return 0;
4339 }
4340
4341 static int e1000_ioctl(struct net_device *netdev, struct ifreq *ifr, int cmd)
4342 {
4343         switch (cmd) {
4344         case SIOCGMIIPHY:
4345         case SIOCGMIIREG:
4346         case SIOCSMIIREG:
4347                 return e1000_mii_ioctl(netdev, ifr, cmd);
4348         default:
4349                 return -EOPNOTSUPP;
4350         }
4351 }
4352
4353 static int __e1000_shutdown(struct pci_dev *pdev, bool *enable_wake)
4354 {
4355         struct net_device *netdev = pci_get_drvdata(pdev);
4356         struct e1000_adapter *adapter = netdev_priv(netdev);
4357         struct e1000_hw *hw = &adapter->hw;
4358         u32 ctrl, ctrl_ext, rctl, status;
4359         u32 wufc = adapter->wol;
4360         int retval = 0;
4361
4362         netif_device_detach(netdev);
4363
4364         if (netif_running(netdev)) {
4365                 WARN_ON(test_bit(__E1000_RESETTING, &adapter->state));
4366                 e1000e_down(adapter);
4367                 e1000_free_irq(adapter);
4368         }
4369         e1000e_reset_interrupt_capability(adapter);
4370
4371         retval = pci_save_state(pdev);
4372         if (retval)
4373                 return retval;
4374
4375         status = er32(STATUS);
4376         if (status & E1000_STATUS_LU)
4377                 wufc &= ~E1000_WUFC_LNKC;
4378
4379         if (wufc) {
4380                 e1000_setup_rctl(adapter);
4381                 e1000_set_multi(netdev);
4382
4383                 /* turn on all-multi mode if wake on multicast is enabled */
4384                 if (wufc & E1000_WUFC_MC) {
4385                         rctl = er32(RCTL);
4386                         rctl |= E1000_RCTL_MPE;
4387                         ew32(RCTL, rctl);
4388                 }
4389
4390                 ctrl = er32(CTRL);
4391                 /* advertise wake from D3Cold */
4392                 #define E1000_CTRL_ADVD3WUC 0x00100000
4393                 /* phy power management enable */
4394                 #define E1000_CTRL_EN_PHY_PWR_MGMT 0x00200000
4395                 ctrl |= E1000_CTRL_ADVD3WUC |
4396                         E1000_CTRL_EN_PHY_PWR_MGMT;
4397                 ew32(CTRL, ctrl);
4398
4399                 if (adapter->hw.phy.media_type == e1000_media_type_fiber ||
4400                     adapter->hw.phy.media_type ==
4401                     e1000_media_type_internal_serdes) {
4402                         /* keep the laser running in D3 */
4403                         ctrl_ext = er32(CTRL_EXT);
4404                         ctrl_ext |= E1000_CTRL_EXT_SDP7_DATA;
4405                         ew32(CTRL_EXT, ctrl_ext);
4406                 }
4407
4408                 if (adapter->flags & FLAG_IS_ICH)
4409                         e1000e_disable_gig_wol_ich8lan(&adapter->hw);
4410
4411                 /* Allow time for pending master requests to run */
4412                 e1000e_disable_pcie_master(&adapter->hw);
4413
4414                 ew32(WUC, E1000_WUC_PME_EN);
4415                 ew32(WUFC, wufc);
4416         } else {
4417                 ew32(WUC, 0);
4418                 ew32(WUFC, 0);
4419         }
4420
4421         *enable_wake = !!wufc;
4422
4423         /* make sure adapter isn't asleep if manageability is enabled */
4424         if (adapter->flags & FLAG_MNG_PT_ENABLED)
4425                 *enable_wake = true;
4426
4427         if (adapter->hw.phy.type == e1000_phy_igp_3)
4428                 e1000e_igp3_phy_powerdown_workaround_ich8lan(&adapter->hw);
4429
4430         /*
4431          * Release control of h/w to f/w.  If f/w is AMT enabled, this
4432          * would have already happened in close and is redundant.
4433          */
4434         e1000_release_hw_control(adapter);
4435
4436         pci_disable_device(pdev);
4437
4438         return 0;
4439 }
4440
4441 static void e1000_power_off(struct pci_dev *pdev, bool sleep, bool wake)
4442 {
4443         if (sleep && wake) {
4444                 pci_prepare_to_sleep(pdev);
4445                 return;
4446         }
4447
4448         pci_wake_from_d3(pdev, wake);
4449         pci_set_power_state(pdev, PCI_D3hot);
4450 }
4451
4452 static void e1000_complete_shutdown(struct pci_dev *pdev, bool sleep,
4453                                     bool wake)
4454 {
4455         struct net_device *netdev = pci_get_drvdata(pdev);
4456         struct e1000_adapter *adapter = netdev_priv(netdev);
4457
4458         /*
4459          * The pci-e switch on some quad port adapters will report a
4460          * correctable error when the MAC transitions from D0 to D3.  To
4461          * prevent this we need to mask off the correctable errors on the
4462          * downstream port of the pci-e switch.
4463          */
4464         if (adapter->flags & FLAG_IS_QUAD_PORT) {
4465                 struct pci_dev *us_dev = pdev->bus->self;
4466                 int pos = pci_find_capability(us_dev, PCI_CAP_ID_EXP);
4467                 u16 devctl;
4468
4469                 pci_read_config_word(us_dev, pos + PCI_EXP_DEVCTL, &devctl);
4470                 pci_write_config_word(us_dev, pos + PCI_EXP_DEVCTL,
4471                                       (devctl & ~PCI_EXP_DEVCTL_CERE));
4472
4473                 e1000_power_off(pdev, sleep, wake);
4474
4475                 pci_write_config_word(us_dev, pos + PCI_EXP_DEVCTL, devctl);
4476         } else {
4477                 e1000_power_off(pdev, sleep, wake);
4478         }
4479 }
4480
4481 static void e1000e_disable_l1aspm(struct pci_dev *pdev)
4482 {
4483         int pos;
4484         u16 val;
4485
4486         /*
4487          * 82573 workaround - disable L1 ASPM on mobile chipsets
4488          *
4489          * L1 ASPM on various mobile (ich7) chipsets do not behave properly
4490          * resulting in lost data or garbage information on the pci-e link
4491          * level. This could result in (false) bad EEPROM checksum errors,
4492          * long ping times (up to 2s) or even a system freeze/hang.
4493          *
4494          * Unfortunately this feature saves about 1W power consumption when
4495          * active.
4496          */
4497         pos = pci_find_capability(pdev, PCI_CAP_ID_EXP);
4498         pci_read_config_word(pdev, pos + PCI_EXP_LNKCTL, &val);
4499         if (val & 0x2) {
4500                 dev_warn(&pdev->dev, "Disabling L1 ASPM\n");
4501                 val &= ~0x2;
4502                 pci_write_config_word(pdev, pos + PCI_EXP_LNKCTL, val);
4503         }
4504 }
4505
4506 #ifdef CONFIG_PM
4507 static int e1000_suspend(struct pci_dev *pdev, pm_message_t state)
4508 {
4509         int retval;
4510         bool wake;
4511
4512         retval = __e1000_shutdown(pdev, &wake);
4513         if (!retval)
4514                 e1000_complete_shutdown(pdev, true, wake);
4515
4516         return retval;
4517 }
4518
4519 static int e1000_resume(struct pci_dev *pdev)
4520 {
4521         struct net_device *netdev = pci_get_drvdata(pdev);
4522         struct e1000_adapter *adapter = netdev_priv(netdev);
4523         struct e1000_hw *hw = &adapter->hw;
4524         u32 err;
4525
4526         pci_set_power_state(pdev, PCI_D0);
4527         pci_restore_state(pdev);
4528         e1000e_disable_l1aspm(pdev);
4529
4530         err = pci_enable_device_mem(pdev);
4531         if (err) {
4532                 dev_err(&pdev->dev,
4533                         "Cannot enable PCI device from suspend\n");
4534                 return err;
4535         }
4536
4537         /* AER (Advanced Error Reporting) hooks */
4538         err = pci_enable_pcie_error_reporting(pdev);
4539         if (err) {
4540                 dev_err(&pdev->dev, "pci_enable_pcie_error_reporting failed "
4541                                     "0x%x\n", err);
4542                 /* non-fatal, continue */
4543         }
4544
4545         pci_set_master(pdev);
4546
4547         pci_enable_wake(pdev, PCI_D3hot, 0);
4548         pci_enable_wake(pdev, PCI_D3cold, 0);
4549
4550         e1000e_set_interrupt_capability(adapter);
4551         if (netif_running(netdev)) {
4552                 err = e1000_request_irq(adapter);
4553                 if (err)
4554                         return err;
4555         }
4556
4557         e1000e_power_up_phy(adapter);
4558         e1000e_reset(adapter);
4559         ew32(WUS, ~0);
4560
4561         e1000_init_manageability(adapter);
4562
4563         if (netif_running(netdev))
4564                 e1000e_up(adapter);
4565
4566         netif_device_attach(netdev);
4567
4568         /*
4569          * If the controller has AMT, do not set DRV_LOAD until the interface
4570          * is up.  For all other cases, let the f/w know that the h/w is now
4571          * under the control of the driver.
4572          */
4573         if (!(adapter->flags & FLAG_HAS_AMT))
4574                 e1000_get_hw_control(adapter);
4575
4576         return 0;
4577 }
4578 #endif
4579
4580 static void e1000_shutdown(struct pci_dev *pdev)
4581 {
4582         bool wake = false;
4583
4584         __e1000_shutdown(pdev, &wake);
4585
4586         if (system_state == SYSTEM_POWER_OFF)
4587                 e1000_complete_shutdown(pdev, false, wake);
4588 }
4589
4590 #ifdef CONFIG_NET_POLL_CONTROLLER
4591 /*
4592  * Polling 'interrupt' - used by things like netconsole to send skbs
4593  * without having to re-enable interrupts. It's not called while
4594  * the interrupt routine is executing.
4595  */
4596 static void e1000_netpoll(struct net_device *netdev)
4597 {
4598         struct e1000_adapter *adapter = netdev_priv(netdev);
4599
4600         disable_irq(adapter->pdev->irq);
4601         e1000_intr(adapter->pdev->irq, netdev);
4602
4603         enable_irq(adapter->pdev->irq);
4604 }
4605 #endif
4606
4607 /**
4608  * e1000_io_error_detected - called when PCI error is detected
4609  * @pdev: Pointer to PCI device
4610  * @state: The current pci connection state
4611  *
4612  * This function is called after a PCI bus error affecting
4613  * this device has been detected.
4614  */
4615 static pci_ers_result_t e1000_io_error_detected(struct pci_dev *pdev,
4616                                                 pci_channel_state_t state)
4617 {
4618         struct net_device *netdev = pci_get_drvdata(pdev);
4619         struct e1000_adapter *adapter = netdev_priv(netdev);
4620
4621         netif_device_detach(netdev);
4622
4623         if (netif_running(netdev))
4624                 e1000e_down(adapter);
4625         pci_disable_device(pdev);
4626
4627         /* Request a slot slot reset. */
4628         return PCI_ERS_RESULT_NEED_RESET;
4629 }
4630
4631 /**
4632  * e1000_io_slot_reset - called after the pci bus has been reset.
4633  * @pdev: Pointer to PCI device
4634  *
4635  * Restart the card from scratch, as if from a cold-boot. Implementation
4636  * resembles the first-half of the e1000_resume routine.
4637  */
4638 static pci_ers_result_t e1000_io_slot_reset(struct pci_dev *pdev)
4639 {
4640         struct net_device *netdev = pci_get_drvdata(pdev);
4641         struct e1000_adapter *adapter = netdev_priv(netdev);
4642         struct e1000_hw *hw = &adapter->hw;
4643         int err;
4644         pci_ers_result_t result;
4645
4646         e1000e_disable_l1aspm(pdev);
4647         err = pci_enable_device_mem(pdev);
4648         if (err) {
4649                 dev_err(&pdev->dev,
4650                         "Cannot re-enable PCI device after reset.\n");
4651                 result = PCI_ERS_RESULT_DISCONNECT;
4652         } else {
4653                 pci_set_master(pdev);
4654                 pci_restore_state(pdev);
4655
4656                 pci_enable_wake(pdev, PCI_D3hot, 0);
4657                 pci_enable_wake(pdev, PCI_D3cold, 0);
4658
4659                 e1000e_reset(adapter);
4660                 ew32(WUS, ~0);
4661                 result = PCI_ERS_RESULT_RECOVERED;
4662         }
4663
4664         pci_cleanup_aer_uncorrect_error_status(pdev);
4665
4666         return result;
4667 }
4668
4669 /**
4670  * e1000_io_resume - called when traffic can start flowing again.
4671  * @pdev: Pointer to PCI device
4672  *
4673  * This callback is called when the error recovery driver tells us that
4674  * its OK to resume normal operation. Implementation resembles the
4675  * second-half of the e1000_resume routine.
4676  */
4677 static void e1000_io_resume(struct pci_dev *pdev)
4678 {
4679         struct net_device *netdev = pci_get_drvdata(pdev);
4680         struct e1000_adapter *adapter = netdev_priv(netdev);
4681
4682         e1000_init_manageability(adapter);
4683
4684         if (netif_running(netdev)) {
4685                 if (e1000e_up(adapter)) {
4686                         dev_err(&pdev->dev,
4687                                 "can't bring device back up after reset\n");
4688                         return;
4689                 }
4690         }
4691
4692         netif_device_attach(netdev);
4693
4694         /*
4695          * If the controller has AMT, do not set DRV_LOAD until the interface
4696          * is up.  For all other cases, let the f/w know that the h/w is now
4697          * under the control of the driver.
4698          */
4699         if (!(adapter->flags & FLAG_HAS_AMT))
4700                 e1000_get_hw_control(adapter);
4701
4702 }
4703
4704 static void e1000_print_device_info(struct e1000_adapter *adapter)
4705 {
4706         struct e1000_hw *hw = &adapter->hw;
4707         struct net_device *netdev = adapter->netdev;
4708         u32 pba_num;
4709
4710         /* print bus type/speed/width info */
4711         e_info("(PCI Express:2.5GB/s:%s) %pM\n",
4712                /* bus width */
4713                ((hw->bus.width == e1000_bus_width_pcie_x4) ? "Width x4" :
4714                 "Width x1"),
4715                /* MAC address */
4716                netdev->dev_addr);
4717         e_info("Intel(R) PRO/%s Network Connection\n",
4718                (hw->phy.type == e1000_phy_ife) ? "10/100" : "1000");
4719         e1000e_read_pba_num(hw, &pba_num);
4720         e_info("MAC: %d, PHY: %d, PBA No: %06x-%03x\n",
4721                hw->mac.type, hw->phy.type, (pba_num >> 8), (pba_num & 0xff));
4722 }
4723
4724 static void e1000_eeprom_checks(struct e1000_adapter *adapter)
4725 {
4726         struct e1000_hw *hw = &adapter->hw;
4727         int ret_val;
4728         u16 buf = 0;
4729
4730         if (hw->mac.type != e1000_82573)
4731                 return;
4732
4733         ret_val = e1000_read_nvm(hw, NVM_INIT_CONTROL2_REG, 1, &buf);
4734         if (!ret_val && (!(le16_to_cpu(buf) & (1 << 0)))) {
4735                 /* Deep Smart Power Down (DSPD) */
4736                 dev_warn(&adapter->pdev->dev,
4737                          "Warning: detected DSPD enabled in EEPROM\n");
4738         }
4739
4740         ret_val = e1000_read_nvm(hw, NVM_INIT_3GIO_3, 1, &buf);
4741         if (!ret_val && (le16_to_cpu(buf) & (3 << 2))) {
4742                 /* ASPM enable */
4743                 dev_warn(&adapter->pdev->dev,
4744                          "Warning: detected ASPM enabled in EEPROM\n");
4745         }
4746 }
4747
4748 static const struct net_device_ops e1000e_netdev_ops = {
4749         .ndo_open               = e1000_open,
4750         .ndo_stop               = e1000_close,
4751         .ndo_start_xmit         = e1000_xmit_frame,
4752         .ndo_get_stats          = e1000_get_stats,
4753         .ndo_set_multicast_list = e1000_set_multi,
4754         .ndo_set_mac_address    = e1000_set_mac,
4755         .ndo_change_mtu         = e1000_change_mtu,
4756         .ndo_do_ioctl           = e1000_ioctl,
4757         .ndo_tx_timeout         = e1000_tx_timeout,
4758         .ndo_validate_addr      = eth_validate_addr,
4759
4760         .ndo_vlan_rx_register   = e1000_vlan_rx_register,
4761         .ndo_vlan_rx_add_vid    = e1000_vlan_rx_add_vid,
4762         .ndo_vlan_rx_kill_vid   = e1000_vlan_rx_kill_vid,
4763 #ifdef CONFIG_NET_POLL_CONTROLLER
4764         .ndo_poll_controller    = e1000_netpoll,
4765 #endif
4766 };
4767
4768 /**
4769  * e1000_probe - Device Initialization Routine
4770  * @pdev: PCI device information struct
4771  * @ent: entry in e1000_pci_tbl
4772  *
4773  * Returns 0 on success, negative on failure
4774  *
4775  * e1000_probe initializes an adapter identified by a pci_dev structure.
4776  * The OS initialization, configuring of the adapter private structure,
4777  * and a hardware reset occur.
4778  **/
4779 static int __devinit e1000_probe(struct pci_dev *pdev,
4780                                  const struct pci_device_id *ent)
4781 {
4782         struct net_device *netdev;
4783         struct e1000_adapter *adapter;
4784         struct e1000_hw *hw;
4785         const struct e1000_info *ei = e1000_info_tbl[ent->driver_data];
4786         resource_size_t mmio_start, mmio_len;
4787         resource_size_t flash_start, flash_len;
4788
4789         static int cards_found;
4790         int i, err, pci_using_dac;
4791         u16 eeprom_data = 0;
4792         u16 eeprom_apme_mask = E1000_EEPROM_APME;
4793
4794         e1000e_disable_l1aspm(pdev);
4795
4796         err = pci_enable_device_mem(pdev);
4797         if (err)
4798                 return err;
4799
4800         pci_using_dac = 0;
4801         err = pci_set_dma_mask(pdev, DMA_BIT_MASK(64));
4802         if (!err) {
4803                 err = pci_set_consistent_dma_mask(pdev, DMA_BIT_MASK(64));
4804                 if (!err)
4805                         pci_using_dac = 1;
4806         } else {
4807                 err = pci_set_dma_mask(pdev, DMA_BIT_MASK(32));
4808                 if (err) {
4809                         err = pci_set_consistent_dma_mask(pdev,
4810                                                           DMA_BIT_MASK(32));
4811                         if (err) {
4812                                 dev_err(&pdev->dev, "No usable DMA "
4813                                         "configuration, aborting\n");
4814                                 goto err_dma;
4815                         }
4816                 }
4817         }
4818
4819         err = pci_request_selected_regions_exclusive(pdev,
4820                                           pci_select_bars(pdev, IORESOURCE_MEM),
4821                                           e1000e_driver_name);
4822         if (err)
4823                 goto err_pci_reg;
4824
4825         pci_set_master(pdev);
4826         /* PCI config space info */
4827         err = pci_save_state(pdev);
4828         if (err)
4829                 goto err_alloc_etherdev;
4830
4831         err = -ENOMEM;
4832         netdev = alloc_etherdev(sizeof(struct e1000_adapter));
4833         if (!netdev)
4834                 goto err_alloc_etherdev;
4835
4836         SET_NETDEV_DEV(netdev, &pdev->dev);
4837
4838         pci_set_drvdata(pdev, netdev);
4839         adapter = netdev_priv(netdev);
4840         hw = &adapter->hw;
4841         adapter->netdev = netdev;
4842         adapter->pdev = pdev;
4843         adapter->ei = ei;
4844         adapter->pba = ei->pba;
4845         adapter->flags = ei->flags;
4846         adapter->flags2 = ei->flags2;
4847         adapter->hw.adapter = adapter;
4848         adapter->hw.mac.type = ei->mac;
4849         adapter->msg_enable = (1 << NETIF_MSG_DRV | NETIF_MSG_PROBE) - 1;
4850
4851         mmio_start = pci_resource_start(pdev, 0);
4852         mmio_len = pci_resource_len(pdev, 0);
4853
4854         err = -EIO;
4855         adapter->hw.hw_addr = ioremap(mmio_start, mmio_len);
4856         if (!adapter->hw.hw_addr)
4857                 goto err_ioremap;
4858
4859         if ((adapter->flags & FLAG_HAS_FLASH) &&
4860             (pci_resource_flags(pdev, 1) & IORESOURCE_MEM)) {
4861                 flash_start = pci_resource_start(pdev, 1);
4862                 flash_len = pci_resource_len(pdev, 1);
4863                 adapter->hw.flash_address = ioremap(flash_start, flash_len);
4864                 if (!adapter->hw.flash_address)
4865                         goto err_flashmap;
4866         }
4867
4868         /* construct the net_device struct */
4869         netdev->netdev_ops              = &e1000e_netdev_ops;
4870         e1000e_set_ethtool_ops(netdev);
4871         netdev->watchdog_timeo          = 5 * HZ;
4872         netif_napi_add(netdev, &adapter->napi, e1000_clean, 64);
4873         strncpy(netdev->name, pci_name(pdev), sizeof(netdev->name) - 1);
4874
4875         netdev->mem_start = mmio_start;
4876         netdev->mem_end = mmio_start + mmio_len;
4877
4878         adapter->bd_number = cards_found++;
4879
4880         e1000e_check_options(adapter);
4881
4882         /* setup adapter struct */
4883         err = e1000_sw_init(adapter);
4884         if (err)
4885                 goto err_sw_init;
4886
4887         err = -EIO;
4888
4889         memcpy(&hw->mac.ops, ei->mac_ops, sizeof(hw->mac.ops));
4890         memcpy(&hw->nvm.ops, ei->nvm_ops, sizeof(hw->nvm.ops));
4891         memcpy(&hw->phy.ops, ei->phy_ops, sizeof(hw->phy.ops));
4892
4893         err = ei->get_variants(adapter);
4894         if (err)
4895                 goto err_hw_init;
4896
4897         if ((adapter->flags & FLAG_IS_ICH) &&
4898             (adapter->flags & FLAG_READ_ONLY_NVM))
4899                 e1000e_write_protect_nvm_ich8lan(&adapter->hw);
4900
4901         hw->mac.ops.get_bus_info(&adapter->hw);
4902
4903         adapter->hw.phy.autoneg_wait_to_complete = 0;
4904
4905         /* Copper options */
4906         if (adapter->hw.phy.media_type == e1000_media_type_copper) {
4907                 adapter->hw.phy.mdix = AUTO_ALL_MODES;
4908                 adapter->hw.phy.disable_polarity_correction = 0;
4909                 adapter->hw.phy.ms_type = e1000_ms_hw_default;
4910         }
4911
4912         if (e1000_check_reset_block(&adapter->hw))
4913                 e_info("PHY reset is blocked due to SOL/IDER session.\n");
4914
4915         netdev->features = NETIF_F_SG |
4916                            NETIF_F_HW_CSUM |
4917                            NETIF_F_HW_VLAN_TX |
4918                            NETIF_F_HW_VLAN_RX;
4919
4920         if (adapter->flags & FLAG_HAS_HW_VLAN_FILTER)
4921                 netdev->features |= NETIF_F_HW_VLAN_FILTER;
4922
4923         netdev->features |= NETIF_F_TSO;
4924         netdev->features |= NETIF_F_TSO6;
4925
4926         netdev->vlan_features |= NETIF_F_TSO;
4927         netdev->vlan_features |= NETIF_F_TSO6;
4928         netdev->vlan_features |= NETIF_F_HW_CSUM;
4929         netdev->vlan_features |= NETIF_F_SG;
4930
4931         if (pci_using_dac)
4932                 netdev->features |= NETIF_F_HIGHDMA;
4933
4934         if (e1000e_enable_mng_pass_thru(&adapter->hw))
4935                 adapter->flags |= FLAG_MNG_PT_ENABLED;
4936
4937         /*
4938          * before reading the NVM, reset the controller to
4939          * put the device in a known good starting state
4940          */
4941         adapter->hw.mac.ops.reset_hw(&adapter->hw);
4942
4943         /*
4944          * systems with ASPM and others may see the checksum fail on the first
4945          * attempt. Let's give it a few tries
4946          */
4947         for (i = 0;; i++) {
4948                 if (e1000_validate_nvm_checksum(&adapter->hw) >= 0)
4949                         break;
4950                 if (i == 2) {
4951                         e_err("The NVM Checksum Is Not Valid\n");
4952                         err = -EIO;
4953                         goto err_eeprom;
4954                 }
4955         }
4956
4957         e1000_eeprom_checks(adapter);
4958
4959         /* copy the MAC address out of the NVM */
4960         if (e1000e_read_mac_addr(&adapter->hw))
4961                 e_err("NVM Read Error while reading MAC address\n");
4962
4963         memcpy(netdev->dev_addr, adapter->hw.mac.addr, netdev->addr_len);
4964         memcpy(netdev->perm_addr, adapter->hw.mac.addr, netdev->addr_len);
4965
4966         if (!is_valid_ether_addr(netdev->perm_addr)) {
4967                 e_err("Invalid MAC Address: %pM\n", netdev->perm_addr);
4968                 err = -EIO;
4969                 goto err_eeprom;
4970         }
4971
4972         init_timer(&adapter->watchdog_timer);
4973         adapter->watchdog_timer.function = &e1000_watchdog;
4974         adapter->watchdog_timer.data = (unsigned long) adapter;
4975
4976         init_timer(&adapter->phy_info_timer);
4977         adapter->phy_info_timer.function = &e1000_update_phy_info;
4978         adapter->phy_info_timer.data = (unsigned long) adapter;
4979
4980         INIT_WORK(&adapter->reset_task, e1000_reset_task);
4981         INIT_WORK(&adapter->watchdog_task, e1000_watchdog_task);
4982         INIT_WORK(&adapter->downshift_task, e1000e_downshift_workaround);
4983         INIT_WORK(&adapter->update_phy_task, e1000e_update_phy_task);
4984
4985         /* Initialize link parameters. User can change them with ethtool */
4986         adapter->hw.mac.autoneg = 1;
4987         adapter->fc_autoneg = 1;
4988         adapter->hw.fc.requested_mode = e1000_fc_default;
4989         adapter->hw.fc.current_mode = e1000_fc_default;
4990         adapter->hw.phy.autoneg_advertised = 0x2f;
4991
4992         /* ring size defaults */
4993         adapter->rx_ring->count = 256;
4994         adapter->tx_ring->count = 256;
4995
4996         /*
4997          * Initial Wake on LAN setting - If APM wake is enabled in
4998          * the EEPROM, enable the ACPI Magic Packet filter
4999          */
5000         if (adapter->flags & FLAG_APME_IN_WUC) {
5001                 /* APME bit in EEPROM is mapped to WUC.APME */
5002                 eeprom_data = er32(WUC);
5003                 eeprom_apme_mask = E1000_WUC_APME;
5004         } else if (adapter->flags & FLAG_APME_IN_CTRL3) {
5005                 if (adapter->flags & FLAG_APME_CHECK_PORT_B &&
5006                     (adapter->hw.bus.func == 1))
5007                         e1000_read_nvm(&adapter->hw,
5008                                 NVM_INIT_CONTROL3_PORT_B, 1, &eeprom_data);
5009                 else
5010                         e1000_read_nvm(&adapter->hw,
5011                                 NVM_INIT_CONTROL3_PORT_A, 1, &eeprom_data);
5012         }
5013
5014         /* fetch WoL from EEPROM */
5015         if (eeprom_data & eeprom_apme_mask)
5016                 adapter->eeprom_wol |= E1000_WUFC_MAG;
5017
5018         /*
5019          * now that we have the eeprom settings, apply the special cases
5020          * where the eeprom may be wrong or the board simply won't support
5021          * wake on lan on a particular port
5022          */
5023         if (!(adapter->flags & FLAG_HAS_WOL))
5024                 adapter->eeprom_wol = 0;
5025
5026         /* initialize the wol settings based on the eeprom settings */
5027         adapter->wol = adapter->eeprom_wol;
5028         device_set_wakeup_enable(&adapter->pdev->dev, adapter->wol);
5029
5030         /* save off EEPROM version number */
5031         e1000_read_nvm(&adapter->hw, 5, 1, &adapter->eeprom_vers);
5032
5033         /* reset the hardware with the new settings */
5034         e1000e_reset(adapter);
5035
5036         /*
5037          * If the controller has AMT, do not set DRV_LOAD until the interface
5038          * is up.  For all other cases, let the f/w know that the h/w is now
5039          * under the control of the driver.
5040          */
5041         if (!(adapter->flags & FLAG_HAS_AMT))
5042                 e1000_get_hw_control(adapter);
5043
5044         strcpy(netdev->name, "eth%d");
5045         err = register_netdev(netdev);
5046         if (err)
5047                 goto err_register;
5048
5049         /* carrier off reporting is important to ethtool even BEFORE open */
5050         netif_carrier_off(netdev);
5051
5052         e1000_print_device_info(adapter);
5053
5054         return 0;
5055
5056 err_register:
5057         if (!(adapter->flags & FLAG_HAS_AMT))
5058                 e1000_release_hw_control(adapter);
5059 err_eeprom:
5060         if (!e1000_check_reset_block(&adapter->hw))
5061                 e1000_phy_hw_reset(&adapter->hw);
5062 err_hw_init:
5063
5064         kfree(adapter->tx_ring);
5065         kfree(adapter->rx_ring);
5066 err_sw_init:
5067         if (adapter->hw.flash_address)
5068                 iounmap(adapter->hw.flash_address);
5069         e1000e_reset_interrupt_capability(adapter);
5070 err_flashmap:
5071         iounmap(adapter->hw.hw_addr);
5072 err_ioremap:
5073         free_netdev(netdev);
5074 err_alloc_etherdev:
5075         pci_release_selected_regions(pdev,
5076                                      pci_select_bars(pdev, IORESOURCE_MEM));
5077 err_pci_reg:
5078 err_dma:
5079         pci_disable_device(pdev);
5080         return err;
5081 }
5082
5083 /**
5084  * e1000_remove - Device Removal Routine
5085  * @pdev: PCI device information struct
5086  *
5087  * e1000_remove is called by the PCI subsystem to alert the driver
5088  * that it should release a PCI device.  The could be caused by a
5089  * Hot-Plug event, or because the driver is going to be removed from
5090  * memory.
5091  **/
5092 static void __devexit e1000_remove(struct pci_dev *pdev)
5093 {
5094         struct net_device *netdev = pci_get_drvdata(pdev);
5095         struct e1000_adapter *adapter = netdev_priv(netdev);
5096         int err;
5097
5098         /*
5099          * flush_scheduled work may reschedule our watchdog task, so
5100          * explicitly disable watchdog tasks from being rescheduled
5101          */
5102         set_bit(__E1000_DOWN, &adapter->state);
5103         del_timer_sync(&adapter->watchdog_timer);
5104         del_timer_sync(&adapter->phy_info_timer);
5105
5106         flush_scheduled_work();
5107
5108         /*
5109          * Release control of h/w to f/w.  If f/w is AMT enabled, this
5110          * would have already happened in close and is redundant.
5111          */
5112         e1000_release_hw_control(adapter);
5113
5114         unregister_netdev(netdev);
5115
5116         if (!e1000_check_reset_block(&adapter->hw))
5117                 e1000_phy_hw_reset(&adapter->hw);
5118
5119         e1000e_reset_interrupt_capability(adapter);
5120         kfree(adapter->tx_ring);
5121         kfree(adapter->rx_ring);
5122
5123         iounmap(adapter->hw.hw_addr);
5124         if (adapter->hw.flash_address)
5125                 iounmap(adapter->hw.flash_address);
5126         pci_release_selected_regions(pdev,
5127                                      pci_select_bars(pdev, IORESOURCE_MEM));
5128
5129         free_netdev(netdev);
5130
5131         /* AER disable */
5132         err = pci_disable_pcie_error_reporting(pdev);
5133         if (err)
5134                 dev_err(&pdev->dev,
5135                         "pci_disable_pcie_error_reporting failed 0x%x\n", err);
5136
5137         pci_disable_device(pdev);
5138 }
5139
5140 /* PCI Error Recovery (ERS) */
5141 static struct pci_error_handlers e1000_err_handler = {
5142         .error_detected = e1000_io_error_detected,
5143         .slot_reset = e1000_io_slot_reset,
5144         .resume = e1000_io_resume,
5145 };
5146
5147 static struct pci_device_id e1000_pci_tbl[] = {
5148         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82571EB_COPPER), board_82571 },
5149         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82571EB_FIBER), board_82571 },
5150         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82571EB_QUAD_COPPER), board_82571 },
5151         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82571EB_QUAD_COPPER_LP), board_82571 },
5152         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82571EB_QUAD_FIBER), board_82571 },
5153         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82571EB_SERDES), board_82571 },
5154         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82571EB_SERDES_DUAL), board_82571 },
5155         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82571EB_SERDES_QUAD), board_82571 },
5156         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82571PT_QUAD_COPPER), board_82571 },
5157
5158         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82572EI), board_82572 },
5159         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82572EI_COPPER), board_82572 },
5160         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82572EI_FIBER), board_82572 },
5161         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82572EI_SERDES), board_82572 },
5162
5163         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82573E), board_82573 },
5164         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82573E_IAMT), board_82573 },
5165         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82573L), board_82573 },
5166
5167         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82574L), board_82574 },
5168         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82574LA), board_82574 },
5169         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82583V), board_82583 },
5170
5171         { PCI_VDEVICE(INTEL, E1000_DEV_ID_80003ES2LAN_COPPER_DPT),
5172           board_80003es2lan },
5173         { PCI_VDEVICE(INTEL, E1000_DEV_ID_80003ES2LAN_COPPER_SPT),
5174           board_80003es2lan },
5175         { PCI_VDEVICE(INTEL, E1000_DEV_ID_80003ES2LAN_SERDES_DPT),
5176           board_80003es2lan },
5177         { PCI_VDEVICE(INTEL, E1000_DEV_ID_80003ES2LAN_SERDES_SPT),
5178           board_80003es2lan },
5179
5180         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH8_IFE), board_ich8lan },
5181         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH8_IFE_G), board_ich8lan },
5182         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH8_IFE_GT), board_ich8lan },
5183         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH8_IGP_AMT), board_ich8lan },
5184         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH8_IGP_C), board_ich8lan },
5185         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH8_IGP_M), board_ich8lan },
5186         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH8_IGP_M_AMT), board_ich8lan },
5187
5188         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH9_IFE), board_ich9lan },
5189         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH9_IFE_G), board_ich9lan },
5190         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH9_IFE_GT), board_ich9lan },
5191         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH9_IGP_AMT), board_ich9lan },
5192         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH9_IGP_C), board_ich9lan },
5193         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH9_BM), board_ich9lan },
5194         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH9_IGP_M), board_ich9lan },
5195         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH9_IGP_M_AMT), board_ich9lan },
5196         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH9_IGP_M_V), board_ich9lan },
5197
5198         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH10_R_BM_LM), board_ich9lan },
5199         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH10_R_BM_LF), board_ich9lan },
5200         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH10_R_BM_V), board_ich9lan },
5201
5202         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH10_D_BM_LM), board_ich10lan },
5203         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH10_D_BM_LF), board_ich10lan },
5204
5205         { }     /* terminate list */
5206 };
5207 MODULE_DEVICE_TABLE(pci, e1000_pci_tbl);
5208
5209 /* PCI Device API Driver */
5210 static struct pci_driver e1000_driver = {
5211         .name     = e1000e_driver_name,
5212         .id_table = e1000_pci_tbl,
5213         .probe    = e1000_probe,
5214         .remove   = __devexit_p(e1000_remove),
5215 #ifdef CONFIG_PM
5216         /* Power Management Hooks */
5217         .suspend  = e1000_suspend,
5218         .resume   = e1000_resume,
5219 #endif
5220         .shutdown = e1000_shutdown,
5221         .err_handler = &e1000_err_handler
5222 };
5223
5224 /**
5225  * e1000_init_module - Driver Registration Routine
5226  *
5227  * e1000_init_module is the first routine called when the driver is
5228  * loaded. All it does is register with the PCI subsystem.
5229  **/
5230 static int __init e1000_init_module(void)
5231 {
5232         int ret;
5233         printk(KERN_INFO "%s: Intel(R) PRO/1000 Network Driver - %s\n",
5234                e1000e_driver_name, e1000e_driver_version);
5235         printk(KERN_INFO "%s: Copyright (c) 1999-2008 Intel Corporation.\n",
5236                e1000e_driver_name);
5237         ret = pci_register_driver(&e1000_driver);
5238         pm_qos_add_requirement(PM_QOS_CPU_DMA_LATENCY, e1000e_driver_name,
5239                                PM_QOS_DEFAULT_VALUE);
5240
5241         return ret;
5242 }
5243 module_init(e1000_init_module);
5244
5245 /**
5246  * e1000_exit_module - Driver Exit Cleanup Routine
5247  *
5248  * e1000_exit_module is called just before the driver is removed
5249  * from memory.
5250  **/
5251 static void __exit e1000_exit_module(void)
5252 {
5253         pci_unregister_driver(&e1000_driver);
5254         pm_qos_remove_requirement(PM_QOS_CPU_DMA_LATENCY, e1000e_driver_name);
5255 }
5256 module_exit(e1000_exit_module);
5257
5258
5259 MODULE_AUTHOR("Intel Corporation, <linux.nics@intel.com>");
5260 MODULE_DESCRIPTION("Intel(R) PRO/1000 Network Driver");
5261 MODULE_LICENSE("GPL");
5262 MODULE_VERSION(DRV_VERSION);
5263
5264 /* e1000_main.c */