drivers/net/e1000e/netdev.c

   1 /*******************************************************************************
   2
   3   Intel PRO/1000 Linux driver
   4   Copyright(c) 1999 - 2008 Intel Corporation.
   5
   6   This program is free software; you can redistribute it and/or modify it
   7   under the terms and conditions of the GNU General Public License,
   8   version 2, as published by the Free Software Foundation.
   9
  10   This program is distributed in the hope it will be useful, but WITHOUT
  11   ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
  12   FITNESS FOR A PARTICULAR PURPOSE.  See the GNU General Public License for
  13   more details.
  14
  15   You should have received a copy of the GNU General Public License along with
  16   this program; if not, write to the Free Software Foundation, Inc.,
  17   51 Franklin St - Fifth Floor, Boston, MA 02110-1301 USA.
  18
  19   The full GNU General Public License is included in this distribution in
  20   the file called "COPYING".
  21
  22   Contact Information:
  23   Linux NICS <linux.nics@intel.com>
  24   e1000-devel Mailing List <e1000-devel@lists.sourceforge.net>
  25   Intel Corporation, 5200 N.E. Elam Young Parkway, Hillsboro, OR 97124-6497
  26
  27 *******************************************************************************/
  28
  29 #include <linux/module.h>
  30 #include <linux/types.h>
  31 #include <linux/init.h>
  32 #include <linux/pci.h>
  33 #include <linux/vmalloc.h>
  34 #include <linux/pagemap.h>
  35 #include <linux/delay.h>
  36 #include <linux/netdevice.h>
  37 #include <linux/tcp.h>
  38 #include <linux/ipv6.h>
  39 #include <net/checksum.h>
  40 #include <net/ip6_checksum.h>
  41 #include <linux/mii.h>
  42 #include <linux/ethtool.h>
  43 #include <linux/if_vlan.h>
  44 #include <linux/cpu.h>
  45 #include <linux/smp.h>
  46 #include <linux/pm_qos_params.h>
  47
  48 #include "e1000.h"
  49
  50 #define DRV_VERSION "0.3.3.3-k2"
  51 char e1000e_driver_name[] = "e1000e";
  52 const char e1000e_driver_version[] = DRV_VERSION;
  53
  54 static const struct e1000_info *e1000_info_tbl[] = {
  55         [board_82571]           = &e1000_82571_info,
  56         [board_82572]           = &e1000_82572_info,
  57         [board_82573]           = &e1000_82573_info,
  58         [board_80003es2lan]     = &e1000_es2_info,
  59         [board_ich8lan]         = &e1000_ich8_info,
  60         [board_ich9lan]         = &e1000_ich9_info,
  61 };
  62
  63 #ifdef DEBUG
  64 /**
  65  * e1000_get_hw_dev_name - return device name string
  66  * used by hardware layer to print debugging information
  67  **/
  68 char *e1000e_get_hw_dev_name(struct e1000_hw *hw)
  69 {
  70         return hw->adapter->netdev->name;
  71 }
  72 #endif
  73
  74 /**
  75  * e1000_desc_unused - calculate if we have unused descriptors
  76  **/
  77 static int e1000_desc_unused(struct e1000_ring *ring)
  78 {
  79         if (ring->next_to_clean > ring->next_to_use)
  80                 return ring->next_to_clean - ring->next_to_use - 1;
  81
  82         return ring->count + ring->next_to_clean - ring->next_to_use - 1;
  83 }
  84
  85 /**
  86  * e1000_receive_skb - helper function to handle Rx indications
  87  * @adapter: board private structure
  88  * @status: descriptor status field as written by hardware
  89  * @vlan: descriptor vlan field as written by hardware (no le/be conversion)
  90  * @skb: pointer to sk_buff to be indicated to stack
  91  **/
  92 static void e1000_receive_skb(struct e1000_adapter *adapter,
  93                               struct net_device *netdev,
  94                               struct sk_buff *skb,
  95                               u8 status, __le16 vlan)
  96 {
  97         skb->protocol = eth_type_trans(skb, netdev);
  98
  99         if (adapter->vlgrp && (status & E1000_RXD_STAT_VP))
 100                 vlan_hwaccel_receive_skb(skb, adapter->vlgrp,
 101                                          le16_to_cpu(vlan));
 102         else
 103                 netif_receive_skb(skb);
 104
 105         netdev->last_rx = jiffies;
 106 }
 107
 108 /**
 109  * e1000_rx_checksum - Receive Checksum Offload for 82543
 110  * @adapter:     board private structure
 111  * @status_err:  receive descriptor status and error fields
 112  * @csum:       receive descriptor csum field
 113  * @sk_buff:     socket buffer with received data
 114  **/
 115 static void e1000_rx_checksum(struct e1000_adapter *adapter, u32 status_err,
 116                               u32 csum, struct sk_buff *skb)
 117 {
 118         u16 status = (u16)status_err;
 119         u8 errors = (u8)(status_err >> 24);
 120         skb->ip_summed = CHECKSUM_NONE;
 121
 122         /* Ignore Checksum bit is set */
 123         if (status & E1000_RXD_STAT_IXSM)
 124                 return;
 125         /* TCP/UDP checksum error bit is set */
 126         if (errors & E1000_RXD_ERR_TCPE) {
 127                 /* let the stack verify checksum errors */
 128                 adapter->hw_csum_err++;
 129                 return;
 130         }
 131
 132         /* TCP/UDP Checksum has not been calculated */
 133         if (!(status & (E1000_RXD_STAT_TCPCS | E1000_RXD_STAT_UDPCS)))
 134                 return;
 135
 136         /* It must be a TCP or UDP packet with a valid checksum */
 137         if (status & E1000_RXD_STAT_TCPCS) {
 138                 /* TCP checksum is good */
 139                 skb->ip_summed = CHECKSUM_UNNECESSARY;
 140         } else {
 141                 /*
 142                  * IP fragment with UDP payload
 143                  * Hardware complements the payload checksum, so we undo it
 144                  * and then put the value in host order for further stack use.
 145                  */
 146                 __sum16 sum = (__force __sum16)htons(csum);
 147                 skb->csum = csum_unfold(~sum);
 148                 skb->ip_summed = CHECKSUM_COMPLETE;
 149         }
 150         adapter->hw_csum_good++;
 151 }
 152
 153 /**
 154  * e1000_alloc_rx_buffers - Replace used receive buffers; legacy & extended
 155  * @adapter: address of board private structure
 156  **/
 157 static void e1000_alloc_rx_buffers(struct e1000_adapter *adapter,
 158                                    int cleaned_count)
 159 {
 160         struct net_device *netdev = adapter->netdev;
 161         struct pci_dev *pdev = adapter->pdev;
 162         struct e1000_ring *rx_ring = adapter->rx_ring;
 163         struct e1000_rx_desc *rx_desc;
 164         struct e1000_buffer *buffer_info;
 165         struct sk_buff *skb;
 166         unsigned int i;
 167         unsigned int bufsz = adapter->rx_buffer_len + NET_IP_ALIGN;
 168
 169         i = rx_ring->next_to_use;
 170         buffer_info = &rx_ring->buffer_info[i];
 171
 172         while (cleaned_count--) {
 173                 skb = buffer_info->skb;
 174                 if (skb) {
 175                         skb_trim(skb, 0);
 176                         goto map_skb;
 177                 }
 178
 179                 skb = netdev_alloc_skb(netdev, bufsz);
 180                 if (!skb) {
 181                         /* Better luck next round */
 182                         adapter->alloc_rx_buff_failed++;
 183                         break;
 184                 }
 185
 186                 /*
 187                  * Make buffer alignment 2 beyond a 16 byte boundary
 188                  * this will result in a 16 byte aligned IP header after
 189                  * the 14 byte MAC header is removed
 190                  */
 191                 skb_reserve(skb, NET_IP_ALIGN);
 192
 193                 buffer_info->skb = skb;
 194 map_skb:
 195                 buffer_info->dma = pci_map_single(pdev, skb->data,
 196                                                   adapter->rx_buffer_len,
 197                                                   PCI_DMA_FROMDEVICE);
 198                 if (pci_dma_mapping_error(pdev, buffer_info->dma)) {
 199                         dev_err(&pdev->dev, "RX DMA map failed\n");
 200                         adapter->rx_dma_failed++;
 201                         break;
 202                 }
 203
 204                 rx_desc = E1000_RX_DESC(*rx_ring, i);
 205                 rx_desc->buffer_addr = cpu_to_le64(buffer_info->dma);
 206
 207                 i++;
 208                 if (i == rx_ring->count)
 209                         i = 0;
 210                 buffer_info = &rx_ring->buffer_info[i];
 211         }
 212
 213         if (rx_ring->next_to_use != i) {
 214                 rx_ring->next_to_use = i;
 215                 if (i-- == 0)
 216                         i = (rx_ring->count - 1);
 217
 218                 /*
 219                  * Force memory writes to complete before letting h/w
 220                  * know there are new descriptors to fetch.  (Only
 221                  * applicable for weak-ordered memory model archs,
 222                  * such as IA-64).
 223                  */
 224                 wmb();
 225                 writel(i, adapter->hw.hw_addr + rx_ring->tail);
 226         }
 227 }
 228
 229 /**
 230  * e1000_alloc_rx_buffers_ps - Replace used receive buffers; packet split
 231  * @adapter: address of board private structure
 232  **/
 233 static void e1000_alloc_rx_buffers_ps(struct e1000_adapter *adapter,
 234                                       int cleaned_count)
 235 {
 236         struct net_device *netdev = adapter->netdev;
 237         struct pci_dev *pdev = adapter->pdev;
 238         union e1000_rx_desc_packet_split *rx_desc;
 239         struct e1000_ring *rx_ring = adapter->rx_ring;
 240         struct e1000_buffer *buffer_info;
 241         struct e1000_ps_page *ps_page;
 242         struct sk_buff *skb;
 243         unsigned int i, j;
 244
 245         i = rx_ring->next_to_use;
 246         buffer_info = &rx_ring->buffer_info[i];
 247
 248         while (cleaned_count--) {
 249                 rx_desc = E1000_RX_DESC_PS(*rx_ring, i);
 250
 251                 for (j = 0; j < PS_PAGE_BUFFERS; j++) {
 252                         ps_page = &buffer_info->ps_pages[j];
 253                         if (j >= adapter->rx_ps_pages) {
 254                                 /* all unused desc entries get hw null ptr */
 255                                 rx_desc->read.buffer_addr[j+1] = ~cpu_to_le64(0);
 256                                 continue;
 257                         }
 258                         if (!ps_page->page) {
 259                                 ps_page->page = alloc_page(GFP_ATOMIC);
 260                                 if (!ps_page->page) {
 261                                         adapter->alloc_rx_buff_failed++;
 262                                         goto no_buffers;
 263                                 }
 264                                 ps_page->dma = pci_map_page(pdev,
 265                                                    ps_page->page,
 266                                                    0, PAGE_SIZE,
 267                                                    PCI_DMA_FROMDEVICE);
 268                                 if (pci_dma_mapping_error(pdev, ps_page->dma)) {
 269                                         dev_err(&adapter->pdev->dev,
 270                                           "RX DMA page map failed\n");
 271                                         adapter->rx_dma_failed++;
 272                                         goto no_buffers;
 273                                 }
 274                         }
 275                         /*
 276                          * Refresh the desc even if buffer_addrs
 277                          * didn't change because each write-back
 278                          * erases this info.
 279                          */
 280                         rx_desc->read.buffer_addr[j+1] =
 281                              cpu_to_le64(ps_page->dma);
 282                 }
 283
 284                 skb = netdev_alloc_skb(netdev,
 285                                        adapter->rx_ps_bsize0 + NET_IP_ALIGN);
 286
 287                 if (!skb) {
 288                         adapter->alloc_rx_buff_failed++;
 289                         break;
 290                 }
 291
 292                 /*
 293                  * Make buffer alignment 2 beyond a 16 byte boundary
 294                  * this will result in a 16 byte aligned IP header after
 295                  * the 14 byte MAC header is removed
 296                  */
 297                 skb_reserve(skb, NET_IP_ALIGN);
 298
 299                 buffer_info->skb = skb;
 300                 buffer_info->dma = pci_map_single(pdev, skb->data,
 301                                                   adapter->rx_ps_bsize0,
 302                                                   PCI_DMA_FROMDEVICE);
 303                 if (pci_dma_mapping_error(pdev, buffer_info->dma)) {
 304                         dev_err(&pdev->dev, "RX DMA map failed\n");
 305                         adapter->rx_dma_failed++;
 306                         /* cleanup skb */
 307                         dev_kfree_skb_any(skb);
 308                         buffer_info->skb = NULL;
 309                         break;
 310                 }
 311
 312                 rx_desc->read.buffer_addr[0] = cpu_to_le64(buffer_info->dma);
 313
 314                 i++;
 315                 if (i == rx_ring->count)
 316                         i = 0;
 317                 buffer_info = &rx_ring->buffer_info[i];
 318         }
 319
 320 no_buffers:
 321         if (rx_ring->next_to_use != i) {
 322                 rx_ring->next_to_use = i;
 323
 324                 if (!(i--))
 325                         i = (rx_ring->count - 1);
 326
 327                 /*
 328                  * Force memory writes to complete before letting h/w
 329                  * know there are new descriptors to fetch.  (Only
 330                  * applicable for weak-ordered memory model archs,
 331                  * such as IA-64).
 332                  */
 333                 wmb();
 334                 /*
 335                  * Hardware increments by 16 bytes, but packet split
 336                  * descriptors are 32 bytes...so we increment tail
 337                  * twice as much.
 338                  */
 339                 writel(i<<1, adapter->hw.hw_addr + rx_ring->tail);
 340         }
 341 }
 342
 343 /**
 344  * e1000_alloc_jumbo_rx_buffers - Replace used jumbo receive buffers
 345  * @adapter: address of board private structure
 346  * @rx_ring: pointer to receive ring structure
 347  * @cleaned_count: number of buffers to allocate this pass
 348  **/
 349
 350 static void e1000_alloc_jumbo_rx_buffers(struct e1000_adapter *adapter,
 351                                          int cleaned_count)
 352 {
 353         struct net_device *netdev = adapter->netdev;
 354         struct pci_dev *pdev = adapter->pdev;
 355         struct e1000_rx_desc *rx_desc;
 356         struct e1000_ring *rx_ring = adapter->rx_ring;
 357         struct e1000_buffer *buffer_info;
 358         struct sk_buff *skb;
 359         unsigned int i;
 360         unsigned int bufsz = 256 -
 361                              16 /* for skb_reserve */ -
 362                              NET_IP_ALIGN;
 363
 364         i = rx_ring->next_to_use;
 365         buffer_info = &rx_ring->buffer_info[i];
 366
 367         while (cleaned_count--) {
 368                 skb = buffer_info->skb;
 369                 if (skb) {
 370                         skb_trim(skb, 0);
 371                         goto check_page;
 372                 }
 373
 374                 skb = netdev_alloc_skb(netdev, bufsz);
 375                 if (unlikely(!skb)) {
 376                         /* Better luck next round */
 377                         adapter->alloc_rx_buff_failed++;
 378                         break;
 379                 }
 380
 381                 /* Make buffer alignment 2 beyond a 16 byte boundary
 382                  * this will result in a 16 byte aligned IP header after
 383                  * the 14 byte MAC header is removed
 384                  */
 385                 skb_reserve(skb, NET_IP_ALIGN);
 386
 387                 buffer_info->skb = skb;
 388 check_page:
 389                 /* allocate a new page if necessary */
 390                 if (!buffer_info->page) {
 391                         buffer_info->page = alloc_page(GFP_ATOMIC);
 392                         if (unlikely(!buffer_info->page)) {
 393                                 adapter->alloc_rx_buff_failed++;
 394                                 break;
 395                         }
 396                 }
 397
 398                 if (!buffer_info->dma)
 399                         buffer_info->dma = pci_map_page(pdev,
 400                                                         buffer_info->page, 0,
 401                                                         PAGE_SIZE,
 402                                                         PCI_DMA_FROMDEVICE);
 403
 404                 rx_desc = E1000_RX_DESC(*rx_ring, i);
 405                 rx_desc->buffer_addr = cpu_to_le64(buffer_info->dma);
 406
 407                 if (unlikely(++i == rx_ring->count))
 408                         i = 0;
 409                 buffer_info = &rx_ring->buffer_info[i];
 410         }
 411
 412         if (likely(rx_ring->next_to_use != i)) {
 413                 rx_ring->next_to_use = i;
 414                 if (unlikely(i-- == 0))
 415                         i = (rx_ring->count - 1);
 416
 417                 /* Force memory writes to complete before letting h/w
 418                  * know there are new descriptors to fetch.  (Only
 419                  * applicable for weak-ordered memory model archs,
 420                  * such as IA-64). */
 421                 wmb();
 422                 writel(i, adapter->hw.hw_addr + rx_ring->tail);
 423         }
 424 }
 425
 426 /**
 427  * e1000_clean_rx_irq - Send received data up the network stack; legacy
 428  * @adapter: board private structure
 429  *
 430  * the return value indicates whether actual cleaning was done, there
 431  * is no guarantee that everything was cleaned
 432  **/
 433 static bool e1000_clean_rx_irq(struct e1000_adapter *adapter,
 434                                int *work_done, int work_to_do)
 435 {
 436         struct net_device *netdev = adapter->netdev;
 437         struct pci_dev *pdev = adapter->pdev;
 438         struct e1000_ring *rx_ring = adapter->rx_ring;
 439         struct e1000_rx_desc *rx_desc, *next_rxd;
 440         struct e1000_buffer *buffer_info, *next_buffer;
 441         u32 length;
 442         unsigned int i;
 443         int cleaned_count = 0;
 444         bool cleaned = 0;
 445         unsigned int total_rx_bytes = 0, total_rx_packets = 0;
 446
 447         i = rx_ring->next_to_clean;
 448         rx_desc = E1000_RX_DESC(*rx_ring, i);
 449         buffer_info = &rx_ring->buffer_info[i];
 450
 451         while (rx_desc->status & E1000_RXD_STAT_DD) {
 452                 struct sk_buff *skb;
 453                 u8 status;
 454
 455                 if (*work_done >= work_to_do)
 456                         break;
 457                 (*work_done)++;
 458
 459                 status = rx_desc->status;
 460                 skb = buffer_info->skb;
 461                 buffer_info->skb = NULL;
 462
 463                 prefetch(skb->data - NET_IP_ALIGN);
 464
 465                 i++;
 466                 if (i == rx_ring->count)
 467                         i = 0;
 468                 next_rxd = E1000_RX_DESC(*rx_ring, i);
 469                 prefetch(next_rxd);
 470
 471                 next_buffer = &rx_ring->buffer_info[i];
 472
 473                 cleaned = 1;
 474                 cleaned_count++;
 475                 pci_unmap_single(pdev,
 476                                  buffer_info->dma,
 477                                  adapter->rx_buffer_len,
 478                                  PCI_DMA_FROMDEVICE);
 479                 buffer_info->dma = 0;
 480
 481                 length = le16_to_cpu(rx_desc->length);
 482
 483                 /* !EOP means multiple descriptors were used to store a single
 484                  * packet, also make sure the frame isn't just CRC only */
 485                 if (!(status & E1000_RXD_STAT_EOP) || (length <= 4)) {
 486                         /* All receives must fit into a single buffer */
 487                         e_dbg("%s: Receive packet consumed multiple buffers\n",
 488                               netdev->name);
 489                         /* recycle */
 490                         buffer_info->skb = skb;
 491                         goto next_desc;
 492                 }
 493
 494                 if (rx_desc->errors & E1000_RXD_ERR_FRAME_ERR_MASK) {
 495                         /* recycle */
 496                         buffer_info->skb = skb;
 497                         goto next_desc;
 498                 }
 499
 500                 total_rx_bytes += length;
 501                 total_rx_packets++;
 502
 503                 /*
 504                  * code added for copybreak, this should improve
 505                  * performance for small packets with large amounts
 506                  * of reassembly being done in the stack
 507                  */
 508                 if (length < copybreak) {
 509                         struct sk_buff *new_skb =
 510                             netdev_alloc_skb(netdev, length + NET_IP_ALIGN);
 511                         if (new_skb) {
 512                                 skb_reserve(new_skb, NET_IP_ALIGN);
 513                                 memcpy(new_skb->data - NET_IP_ALIGN,
 514                                        skb->data - NET_IP_ALIGN,
 515                                        length + NET_IP_ALIGN);
 516                                 /* save the skb in buffer_info as good */
 517                                 buffer_info->skb = skb;
 518                                 skb = new_skb;
 519                         }
 520                         /* else just continue with the old one */
 521                 }
 522                 /* end copybreak code */
 523                 skb_put(skb, length);
 524
 525                 /* Receive Checksum Offload */
 526                 e1000_rx_checksum(adapter,
 527                                   (u32)(status) |
 528                                   ((u32)(rx_desc->errors) << 24),
 529                                   le16_to_cpu(rx_desc->csum), skb);
 530
 531                 e1000_receive_skb(adapter, netdev, skb,status,rx_desc->special);
 532
 533 next_desc:
 534                 rx_desc->status = 0;
 535
 536                 /* return some buffers to hardware, one at a time is too slow */
 537                 if (cleaned_count >= E1000_RX_BUFFER_WRITE) {
 538                         adapter->alloc_rx_buf(adapter, cleaned_count);
 539                         cleaned_count = 0;
 540                 }
 541
 542                 /* use prefetched values */
 543                 rx_desc = next_rxd;
 544                 buffer_info = next_buffer;
 545         }
 546         rx_ring->next_to_clean = i;
 547
 548         cleaned_count = e1000_desc_unused(rx_ring);
 549         if (cleaned_count)
 550                 adapter->alloc_rx_buf(adapter, cleaned_count);
 551
 552         adapter->total_rx_bytes += total_rx_bytes;
 553         adapter->total_rx_packets += total_rx_packets;
 554         adapter->net_stats.rx_bytes += total_rx_bytes;
 555         adapter->net_stats.rx_packets += total_rx_packets;
 556         return cleaned;
 557 }
 558
 559 static void e1000_put_txbuf(struct e1000_adapter *adapter,
 560                              struct e1000_buffer *buffer_info)
 561 {
 562         if (buffer_info->dma) {
 563                 pci_unmap_page(adapter->pdev, buffer_info->dma,
 564                                buffer_info->length, PCI_DMA_TODEVICE);
 565                 buffer_info->dma = 0;
 566         }
 567         if (buffer_info->skb) {
 568                 dev_kfree_skb_any(buffer_info->skb);
 569                 buffer_info->skb = NULL;
 570         }
 571 }
 572
 573 static void e1000_print_tx_hang(struct e1000_adapter *adapter)
 574 {
 575         struct e1000_ring *tx_ring = adapter->tx_ring;
 576         unsigned int i = tx_ring->next_to_clean;
 577         unsigned int eop = tx_ring->buffer_info[i].next_to_watch;
 578         struct e1000_tx_desc *eop_desc = E1000_TX_DESC(*tx_ring, eop);
 579
 580         /* detected Tx unit hang */
 581         e_err("Detected Tx Unit Hang:\n"
 582               "  TDH                  <%x>\n"
 583               "  TDT                  <%x>\n"
 584               "  next_to_use          <%x>\n"
 585               "  next_to_clean        <%x>\n"
 586               "buffer_info[next_to_clean]:\n"
 587               "  time_stamp           <%lx>\n"
 588               "  next_to_watch        <%x>\n"
 589               "  jiffies              <%lx>\n"
 590               "  next_to_watch.status <%x>\n",
 591               readl(adapter->hw.hw_addr + tx_ring->head),
 592               readl(adapter->hw.hw_addr + tx_ring->tail),
 593               tx_ring->next_to_use,
 594               tx_ring->next_to_clean,
 595               tx_ring->buffer_info[eop].time_stamp,
 596               eop,
 597               jiffies,
 598               eop_desc->upper.fields.status);
 599 }
 600
 601 /**
 602  * e1000_clean_tx_irq - Reclaim resources after transmit completes
 603  * @adapter: board private structure
 604  *
 605  * the return value indicates whether actual cleaning was done, there
 606  * is no guarantee that everything was cleaned
 607  **/
 608 static bool e1000_clean_tx_irq(struct e1000_adapter *adapter)
 609 {
 610         struct net_device *netdev = adapter->netdev;
 611         struct e1000_hw *hw = &adapter->hw;
 612         struct e1000_ring *tx_ring = adapter->tx_ring;
 613         struct e1000_tx_desc *tx_desc, *eop_desc;
 614         struct e1000_buffer *buffer_info;
 615         unsigned int i, eop;
 616         unsigned int count = 0;
 617         bool cleaned = 0;
 618         unsigned int total_tx_bytes = 0, total_tx_packets = 0;
 619
 620         i = tx_ring->next_to_clean;
 621         eop = tx_ring->buffer_info[i].next_to_watch;
 622         eop_desc = E1000_TX_DESC(*tx_ring, eop);
 623
 624         while (eop_desc->upper.data & cpu_to_le32(E1000_TXD_STAT_DD)) {
 625                 for (cleaned = 0; !cleaned; ) {
 626                         tx_desc = E1000_TX_DESC(*tx_ring, i);
 627                         buffer_info = &tx_ring->buffer_info[i];
 628                         cleaned = (i == eop);
 629
 630                         if (cleaned) {
 631                                 struct sk_buff *skb = buffer_info->skb;
 632                                 unsigned int segs, bytecount;
 633                                 segs = skb_shinfo(skb)->gso_segs ?: 1;
 634                                 /* multiply data chunks by size of headers */
 635                                 bytecount = ((segs - 1) * skb_headlen(skb)) +
 636                                             skb->len;
 637                                 total_tx_packets += segs;
 638                                 total_tx_bytes += bytecount;
 639                         }
 640
 641                         e1000_put_txbuf(adapter, buffer_info);
 642                         tx_desc->upper.data = 0;
 643
 644                         i++;
 645                         if (i == tx_ring->count)
 646                                 i = 0;
 647                 }
 648
 649                 eop = tx_ring->buffer_info[i].next_to_watch;
 650                 eop_desc = E1000_TX_DESC(*tx_ring, eop);
 651 #define E1000_TX_WEIGHT 64
 652                 /* weight of a sort for tx, to avoid endless transmit cleanup */
 653                 if (count++ == E1000_TX_WEIGHT)
 654                         break;
 655         }
 656
 657         tx_ring->next_to_clean = i;
 658
 659 #define TX_WAKE_THRESHOLD 32
 660         if (cleaned && netif_carrier_ok(netdev) &&
 661                      e1000_desc_unused(tx_ring) >= TX_WAKE_THRESHOLD) {
 662                 /* Make sure that anybody stopping the queue after this
 663                  * sees the new next_to_clean.
 664                  */
 665                 smp_mb();
 666
 667                 if (netif_queue_stopped(netdev) &&
 668                     !(test_bit(__E1000_DOWN, &adapter->state))) {
 669                         netif_wake_queue(netdev);
 670                         ++adapter->restart_queue;
 671                 }
 672         }
 673
 674         if (adapter->detect_tx_hung) {
 675                 /*
 676                  * Detect a transmit hang in hardware, this serializes the
 677                  * check with the clearing of time_stamp and movement of i
 678                  */
 679                 adapter->detect_tx_hung = 0;
 680                 if (tx_ring->buffer_info[eop].dma &&
 681                     time_after(jiffies, tx_ring->buffer_info[eop].time_stamp
 682                                + (adapter->tx_timeout_factor * HZ))
 683                     && !(er32(STATUS) & E1000_STATUS_TXOFF)) {
 684                         e1000_print_tx_hang(adapter);
 685                         netif_stop_queue(netdev);
 686                 }
 687         }
 688         adapter->total_tx_bytes += total_tx_bytes;
 689         adapter->total_tx_packets += total_tx_packets;
 690         adapter->net_stats.tx_bytes += total_tx_bytes;
 691         adapter->net_stats.tx_packets += total_tx_packets;
 692         return cleaned;
 693 }
 694
 695 /**
 696  * e1000_clean_rx_irq_ps - Send received data up the network stack; packet split
 697  * @adapter: board private structure
 698  *
 699  * the return value indicates whether actual cleaning was done, there
 700  * is no guarantee that everything was cleaned
 701  **/
 702 static bool e1000_clean_rx_irq_ps(struct e1000_adapter *adapter,
 703                                   int *work_done, int work_to_do)
 704 {
 705         union e1000_rx_desc_packet_split *rx_desc, *next_rxd;
 706         struct net_device *netdev = adapter->netdev;
 707         struct pci_dev *pdev = adapter->pdev;
 708         struct e1000_ring *rx_ring = adapter->rx_ring;
 709         struct e1000_buffer *buffer_info, *next_buffer;
 710         struct e1000_ps_page *ps_page;
 711         struct sk_buff *skb;
 712         unsigned int i, j;
 713         u32 length, staterr;
 714         int cleaned_count = 0;
 715         bool cleaned = 0;
 716         unsigned int total_rx_bytes = 0, total_rx_packets = 0;
 717
 718         i = rx_ring->next_to_clean;
 719         rx_desc = E1000_RX_DESC_PS(*rx_ring, i);
 720         staterr = le32_to_cpu(rx_desc->wb.middle.status_error);
 721         buffer_info = &rx_ring->buffer_info[i];
 722
 723         while (staterr & E1000_RXD_STAT_DD) {
 724                 if (*work_done >= work_to_do)
 725                         break;
 726                 (*work_done)++;
 727                 skb = buffer_info->skb;
 728
 729                 /* in the packet split case this is header only */
 730                 prefetch(skb->data - NET_IP_ALIGN);
 731
 732                 i++;
 733                 if (i == rx_ring->count)
 734                         i = 0;
 735                 next_rxd = E1000_RX_DESC_PS(*rx_ring, i);
 736                 prefetch(next_rxd);
 737
 738                 next_buffer = &rx_ring->buffer_info[i];
 739
 740                 cleaned = 1;
 741                 cleaned_count++;
 742                 pci_unmap_single(pdev, buffer_info->dma,
 743                                  adapter->rx_ps_bsize0,
 744                                  PCI_DMA_FROMDEVICE);
 745                 buffer_info->dma = 0;
 746
 747                 if (!(staterr & E1000_RXD_STAT_EOP)) {
 748                         e_dbg("%s: Packet Split buffers didn't pick up the "
 749                               "full packet\n", netdev->name);
 750                         dev_kfree_skb_irq(skb);
 751                         goto next_desc;
 752                 }
 753
 754                 if (staterr & E1000_RXDEXT_ERR_FRAME_ERR_MASK) {
 755                         dev_kfree_skb_irq(skb);
 756                         goto next_desc;
 757                 }
 758
 759                 length = le16_to_cpu(rx_desc->wb.middle.length0);
 760
 761                 if (!length) {
 762                         e_dbg("%s: Last part of the packet spanning multiple "
 763                               "descriptors\n", netdev->name);
 764                         dev_kfree_skb_irq(skb);
 765                         goto next_desc;
 766                 }
 767
 768                 /* Good Receive */
 769                 skb_put(skb, length);
 770
 771                 {
 772                 /*
 773                  * this looks ugly, but it seems compiler issues make it
 774                  * more efficient than reusing j
 775                  */
 776                 int l1 = le16_to_cpu(rx_desc->wb.upper.length[0]);
 777
 778                 /*
 779                  * page alloc/put takes too long and effects small packet
 780                  * throughput, so unsplit small packets and save the alloc/put
 781                  * only valid in softirq (napi) context to call kmap_*
 782                  */
 783                 if (l1 && (l1 <= copybreak) &&
 784                     ((length + l1) <= adapter->rx_ps_bsize0)) {
 785                         u8 *vaddr;
 786
 787                         ps_page = &buffer_info->ps_pages[0];
 788
 789                         /*
 790                          * there is no documentation about how to call
 791                          * kmap_atomic, so we can't hold the mapping
 792                          * very long
 793                          */
 794                         pci_dma_sync_single_for_cpu(pdev, ps_page->dma,
 795                                 PAGE_SIZE, PCI_DMA_FROMDEVICE);
 796                         vaddr = kmap_atomic(ps_page->page, KM_SKB_DATA_SOFTIRQ);
 797                         memcpy(skb_tail_pointer(skb), vaddr, l1);
 798                         kunmap_atomic(vaddr, KM_SKB_DATA_SOFTIRQ);
 799                         pci_dma_sync_single_for_device(pdev, ps_page->dma,
 800                                 PAGE_SIZE, PCI_DMA_FROMDEVICE);
 801
 802                         skb_put(skb, l1);
 803                         goto copydone;
 804                 } /* if */
 805                 }
 806
 807                 for (j = 0; j < PS_PAGE_BUFFERS; j++) {
 808                         length = le16_to_cpu(rx_desc->wb.upper.length[j]);
 809                         if (!length)
 810                                 break;
 811
 812                         ps_page = &buffer_info->ps_pages[j];
 813                         pci_unmap_page(pdev, ps_page->dma, PAGE_SIZE,
 814                                        PCI_DMA_FROMDEVICE);
 815                         ps_page->dma = 0;
 816                         skb_fill_page_desc(skb, j, ps_page->page, 0, length);
 817                         ps_page->page = NULL;
 818                         skb->len += length;
 819                         skb->data_len += length;
 820                         skb->truesize += length;
 821                 }
 822
 823 copydone:
 824                 total_rx_bytes += skb->len;
 825                 total_rx_packets++;
 826
 827                 e1000_rx_checksum(adapter, staterr, le16_to_cpu(
 828                         rx_desc->wb.lower.hi_dword.csum_ip.csum), skb);
 829
 830                 if (rx_desc->wb.upper.header_status &
 831                            cpu_to_le16(E1000_RXDPS_HDRSTAT_HDRSP))
 832                         adapter->rx_hdr_split++;
 833
 834                 e1000_receive_skb(adapter, netdev, skb,
 835                                   staterr, rx_desc->wb.middle.vlan);
 836
 837 next_desc:
 838                 rx_desc->wb.middle.status_error &= cpu_to_le32(~0xFF);
 839                 buffer_info->skb = NULL;
 840
 841                 /* return some buffers to hardware, one at a time is too slow */
 842                 if (cleaned_count >= E1000_RX_BUFFER_WRITE) {
 843                         adapter->alloc_rx_buf(adapter, cleaned_count);
 844                         cleaned_count = 0;
 845                 }
 846
 847                 /* use prefetched values */
 848                 rx_desc = next_rxd;
 849                 buffer_info = next_buffer;
 850
 851                 staterr = le32_to_cpu(rx_desc->wb.middle.status_error);
 852         }
 853         rx_ring->next_to_clean = i;
 854
 855         cleaned_count = e1000_desc_unused(rx_ring);
 856         if (cleaned_count)
 857                 adapter->alloc_rx_buf(adapter, cleaned_count);
 858
 859         adapter->total_rx_bytes += total_rx_bytes;
 860         adapter->total_rx_packets += total_rx_packets;
 861         adapter->net_stats.rx_bytes += total_rx_bytes;
 862         adapter->net_stats.rx_packets += total_rx_packets;
 863         return cleaned;
 864 }
 865
 866 /**
 867  * e1000_consume_page - helper function
 868  **/
 869 static void e1000_consume_page(struct e1000_buffer *bi, struct sk_buff *skb,
 870                                u16 length)
 871 {
 872         bi->page = NULL;
 873         skb->len += length;
 874         skb->data_len += length;
 875         skb->truesize += length;
 876 }
 877
 878 /**
 879  * e1000_clean_jumbo_rx_irq - Send received data up the network stack; legacy
 880  * @adapter: board private structure
 881  *
 882  * the return value indicates whether actual cleaning was done, there
 883  * is no guarantee that everything was cleaned
 884  **/
 885
 886 static bool e1000_clean_jumbo_rx_irq(struct e1000_adapter *adapter,
 887                                      int *work_done, int work_to_do)
 888 {
 889         struct net_device *netdev = adapter->netdev;
 890         struct pci_dev *pdev = adapter->pdev;
 891         struct e1000_ring *rx_ring = adapter->rx_ring;
 892         struct e1000_rx_desc *rx_desc, *next_rxd;
 893         struct e1000_buffer *buffer_info, *next_buffer;
 894         u32 length;
 895         unsigned int i;
 896         int cleaned_count = 0;
 897         bool cleaned = false;
 898         unsigned int total_rx_bytes=0, total_rx_packets=0;
 899
 900         i = rx_ring->next_to_clean;
 901         rx_desc = E1000_RX_DESC(*rx_ring, i);
 902         buffer_info = &rx_ring->buffer_info[i];
 903
 904         while (rx_desc->status & E1000_RXD_STAT_DD) {
 905                 struct sk_buff *skb;
 906                 u8 status;
 907
 908                 if (*work_done >= work_to_do)
 909                         break;
 910                 (*work_done)++;
 911
 912                 status = rx_desc->status;
 913                 skb = buffer_info->skb;
 914                 buffer_info->skb = NULL;
 915
 916                 ++i;
 917                 if (i == rx_ring->count)
 918                         i = 0;
 919                 next_rxd = E1000_RX_DESC(*rx_ring, i);
 920                 prefetch(next_rxd);
 921
 922                 next_buffer = &rx_ring->buffer_info[i];
 923
 924                 cleaned = true;
 925                 cleaned_count++;
 926                 pci_unmap_page(pdev, buffer_info->dma, PAGE_SIZE,
 927                                PCI_DMA_FROMDEVICE);
 928                 buffer_info->dma = 0;
 929
 930                 length = le16_to_cpu(rx_desc->length);
 931
 932                 /* errors is only valid for DD + EOP descriptors */
 933                 if (unlikely((status & E1000_RXD_STAT_EOP) &&
 934                     (rx_desc->errors & E1000_RXD_ERR_FRAME_ERR_MASK))) {
 935                                 /* recycle both page and skb */
 936                                 buffer_info->skb = skb;
 937                                 /* an error means any chain goes out the window
 938                                  * too */
 939                                 if (rx_ring->rx_skb_top)
 940                                         dev_kfree_skb(rx_ring->rx_skb_top);
 941                                 rx_ring->rx_skb_top = NULL;
 942                                 goto next_desc;
 943                 }
 944
 945 #define rxtop rx_ring->rx_skb_top
 946                 if (!(status & E1000_RXD_STAT_EOP)) {
 947                         /* this descriptor is only the beginning (or middle) */
 948                         if (!rxtop) {
 949                                 /* this is the beginning of a chain */
 950                                 rxtop = skb;
 951                                 skb_fill_page_desc(rxtop, 0, buffer_info->page,
 952                                                    0, length);
 953                         } else {
 954                                 /* this is the middle of a chain */
 955                                 skb_fill_page_desc(rxtop,
 956                                     skb_shinfo(rxtop)->nr_frags,
 957                                     buffer_info->page, 0, length);
 958                                 /* re-use the skb, only consumed the page */
 959                                 buffer_info->skb = skb;
 960                         }
 961                         e1000_consume_page(buffer_info, rxtop, length);
 962                         goto next_desc;
 963                 } else {
 964                         if (rxtop) {
 965                                 /* end of the chain */
 966                                 skb_fill_page_desc(rxtop,
 967                                     skb_shinfo(rxtop)->nr_frags,
 968                                     buffer_info->page, 0, length);
 969                                 /* re-use the current skb, we only consumed the
 970                                  * page */
 971                                 buffer_info->skb = skb;
 972                                 skb = rxtop;
 973                                 rxtop = NULL;
 974                                 e1000_consume_page(buffer_info, skb, length);
 975                         } else {
 976                                 /* no chain, got EOP, this buf is the packet
 977                                  * copybreak to save the put_page/alloc_page */
 978                                 if (length <= copybreak &&
 979                                     skb_tailroom(skb) >= length) {
 980                                         u8 *vaddr;
 981                                         vaddr = kmap_atomic(buffer_info->page,
 982                                                            KM_SKB_DATA_SOFTIRQ);
 983                                         memcpy(skb_tail_pointer(skb), vaddr,
 984                                                length);
 985                                         kunmap_atomic(vaddr,
 986                                                       KM_SKB_DATA_SOFTIRQ);
 987                                         /* re-use the page, so don't erase
 988                                          * buffer_info->page */
 989                                         skb_put(skb, length);
 990                                 } else {
 991                                         skb_fill_page_desc(skb, 0,
 992                                                            buffer_info->page, 0,
 993                                                            length);
 994                                         e1000_consume_page(buffer_info, skb,
 995                                                            length);
 996                                 }
 997                         }
 998                 }
 999
1000                 /* Receive Checksum Offload XXX recompute due to CRC strip? */
1001                 e1000_rx_checksum(adapter,
1002                                   (u32)(status) |
1003                                   ((u32)(rx_desc->errors) << 24),
1004                                   le16_to_cpu(rx_desc->csum), skb);
1005
1006                 /* probably a little skewed due to removing CRC */
1007                 total_rx_bytes += skb->len;
1008                 total_rx_packets++;
1009
1010                 /* eth type trans needs skb->data to point to something */
1011                 if (!pskb_may_pull(skb, ETH_HLEN)) {
1012                         e_err("pskb_may_pull failed.\n");
1013                         dev_kfree_skb(skb);
1014                         goto next_desc;
1015                 }
1016
1017                 e1000_receive_skb(adapter, netdev, skb, status,
1018                                   rx_desc->special);
1019
1020 next_desc:
1021                 rx_desc->status = 0;
1022
1023                 /* return some buffers to hardware, one at a time is too slow */
1024                 if (unlikely(cleaned_count >= E1000_RX_BUFFER_WRITE)) {
1025                         adapter->alloc_rx_buf(adapter, cleaned_count);
1026                         cleaned_count = 0;
1027                 }
1028
1029                 /* use prefetched values */
1030                 rx_desc = next_rxd;
1031                 buffer_info = next_buffer;
1032         }
1033         rx_ring->next_to_clean = i;
1034
1035         cleaned_count = e1000_desc_unused(rx_ring);
1036         if (cleaned_count)
1037                 adapter->alloc_rx_buf(adapter, cleaned_count);
1038
1039         adapter->total_rx_bytes += total_rx_bytes;
1040         adapter->total_rx_packets += total_rx_packets;
1041         adapter->net_stats.rx_bytes += total_rx_bytes;
1042         adapter->net_stats.rx_packets += total_rx_packets;
1043         return cleaned;
1044 }
1045
1046 /**
1047  * e1000_clean_rx_ring - Free Rx Buffers per Queue
1048  * @adapter: board private structure
1049  **/
1050 static void e1000_clean_rx_ring(struct e1000_adapter *adapter)
1051 {
1052         struct e1000_ring *rx_ring = adapter->rx_ring;
1053         struct e1000_buffer *buffer_info;
1054         struct e1000_ps_page *ps_page;
1055         struct pci_dev *pdev = adapter->pdev;
1056         unsigned int i, j;
1057
1058         /* Free all the Rx ring sk_buffs */
1059         for (i = 0; i < rx_ring->count; i++) {
1060                 buffer_info = &rx_ring->buffer_info[i];
1061                 if (buffer_info->dma) {
1062                         if (adapter->clean_rx == e1000_clean_rx_irq)
1063                                 pci_unmap_single(pdev, buffer_info->dma,
1064                                                  adapter->rx_buffer_len,
1065                                                  PCI_DMA_FROMDEVICE);
1066                         else if (adapter->clean_rx == e1000_clean_jumbo_rx_irq)
1067                                 pci_unmap_page(pdev, buffer_info->dma,
1068                                                PAGE_SIZE,
1069                                                PCI_DMA_FROMDEVICE);
1070                         else if (adapter->clean_rx == e1000_clean_rx_irq_ps)
1071                                 pci_unmap_single(pdev, buffer_info->dma,
1072                                                  adapter->rx_ps_bsize0,
1073                                                  PCI_DMA_FROMDEVICE);
1074                         buffer_info->dma = 0;
1075                 }
1076
1077                 if (buffer_info->page) {
1078                         put_page(buffer_info->page);
1079                         buffer_info->page = NULL;
1080                 }
1081
1082                 if (buffer_info->skb) {
1083                         dev_kfree_skb(buffer_info->skb);
1084                         buffer_info->skb = NULL;
1085                 }
1086
1087                 for (j = 0; j < PS_PAGE_BUFFERS; j++) {
1088                         ps_page = &buffer_info->ps_pages[j];
1089                         if (!ps_page->page)
1090                                 break;
1091                         pci_unmap_page(pdev, ps_page->dma, PAGE_SIZE,
1092                                        PCI_DMA_FROMDEVICE);
1093                         ps_page->dma = 0;
1094                         put_page(ps_page->page);
1095                         ps_page->page = NULL;
1096                 }
1097         }
1098
1099         /* there also may be some cached data from a chained receive */
1100         if (rx_ring->rx_skb_top) {
1101                 dev_kfree_skb(rx_ring->rx_skb_top);
1102                 rx_ring->rx_skb_top = NULL;
1103         }
1104
1105         /* Zero out the descriptor ring */
1106         memset(rx_ring->desc, 0, rx_ring->size);
1107
1108         rx_ring->next_to_clean = 0;
1109         rx_ring->next_to_use = 0;
1110
1111         writel(0, adapter->hw.hw_addr + rx_ring->head);
1112         writel(0, adapter->hw.hw_addr + rx_ring->tail);
1113 }
1114
1115 /**
1116  * e1000_intr_msi - Interrupt Handler
1117  * @irq: interrupt number
1118  * @data: pointer to a network interface device structure
1119  **/
1120 static irqreturn_t e1000_intr_msi(int irq, void *data)
1121 {
1122         struct net_device *netdev = data;
1123         struct e1000_adapter *adapter = netdev_priv(netdev);
1124         struct e1000_hw *hw = &adapter->hw;
1125         u32 icr = er32(ICR);
1126
1127         /*
1128          * read ICR disables interrupts using IAM
1129          */
1130
1131         if (icr & (E1000_ICR_RXSEQ | E1000_ICR_LSC)) {
1132                 hw->mac.get_link_status = 1;
1133                 /*
1134                  * ICH8 workaround-- Call gig speed drop workaround on cable
1135                  * disconnect (LSC) before accessing any PHY registers
1136                  */
1137                 if ((adapter->flags & FLAG_LSC_GIG_SPEED_DROP) &&
1138                     (!(er32(STATUS) & E1000_STATUS_LU)))
1139                         e1000e_gig_downshift_workaround_ich8lan(hw);
1140
1141                 /*
1142                  * 80003ES2LAN workaround-- For packet buffer work-around on
1143                  * link down event; disable receives here in the ISR and reset
1144                  * adapter in watchdog
1145                  */
1146                 if (netif_carrier_ok(netdev) &&
1147                     adapter->flags & FLAG_RX_NEEDS_RESTART) {
1148                         /* disable receives */
1149                         u32 rctl = er32(RCTL);
1150                         ew32(RCTL, rctl & ~E1000_RCTL_EN);
1151                         adapter->flags |= FLAG_RX_RESTART_NOW;
1152                 }
1153                 /* guard against interrupt when we're going down */
1154                 if (!test_bit(__E1000_DOWN, &adapter->state))
1155                         mod_timer(&adapter->watchdog_timer, jiffies + 1);
1156         }
1157
1158         if (netif_rx_schedule_prep(netdev, &adapter->napi)) {
1159                 adapter->total_tx_bytes = 0;
1160                 adapter->total_tx_packets = 0;
1161                 adapter->total_rx_bytes = 0;
1162                 adapter->total_rx_packets = 0;
1163                 __netif_rx_schedule(netdev, &adapter->napi);
1164         }
1165
1166         return IRQ_HANDLED;
1167 }
1168
1169 /**
1170  * e1000_intr - Interrupt Handler
1171  * @irq: interrupt number
1172  * @data: pointer to a network interface device structure
1173  **/
1174 static irqreturn_t e1000_intr(int irq, void *data)
1175 {
1176         struct net_device *netdev = data;
1177         struct e1000_adapter *adapter = netdev_priv(netdev);
1178         struct e1000_hw *hw = &adapter->hw;
1179
1180         u32 rctl, icr = er32(ICR);
1181         if (!icr)
1182                 return IRQ_NONE;  /* Not our interrupt */
1183
1184         /*
1185          * IMS will not auto-mask if INT_ASSERTED is not set, and if it is
1186          * not set, then the adapter didn't send an interrupt
1187          */
1188         if (!(icr & E1000_ICR_INT_ASSERTED))
1189                 return IRQ_NONE;
1190
1191         /*
1192          * Interrupt Auto-Mask...upon reading ICR,
1193          * interrupts are masked.  No need for the
1194          * IMC write
1195          */
1196
1197         if (icr & (E1000_ICR_RXSEQ | E1000_ICR_LSC)) {
1198                 hw->mac.get_link_status = 1;
1199                 /*
1200                  * ICH8 workaround-- Call gig speed drop workaround on cable
1201                  * disconnect (LSC) before accessing any PHY registers
1202                  */
1203                 if ((adapter->flags & FLAG_LSC_GIG_SPEED_DROP) &&
1204                     (!(er32(STATUS) & E1000_STATUS_LU)))
1205                         e1000e_gig_downshift_workaround_ich8lan(hw);
1206
1207                 /*
1208                  * 80003ES2LAN workaround--
1209                  * For packet buffer work-around on link down event;
1210                  * disable receives here in the ISR and
1211                  * reset adapter in watchdog
1212                  */
1213                 if (netif_carrier_ok(netdev) &&
1214                     (adapter->flags & FLAG_RX_NEEDS_RESTART)) {
1215                         /* disable receives */
1216                         rctl = er32(RCTL);
1217                         ew32(RCTL, rctl & ~E1000_RCTL_EN);
1218                         adapter->flags |= FLAG_RX_RESTART_NOW;
1219                 }
1220                 /* guard against interrupt when we're going down */
1221                 if (!test_bit(__E1000_DOWN, &adapter->state))
1222                         mod_timer(&adapter->watchdog_timer, jiffies + 1);
1223         }
1224
1225         if (netif_rx_schedule_prep(netdev, &adapter->napi)) {
1226                 adapter->total_tx_bytes = 0;
1227                 adapter->total_tx_packets = 0;
1228                 adapter->total_rx_bytes = 0;
1229                 adapter->total_rx_packets = 0;
1230                 __netif_rx_schedule(netdev, &adapter->napi);
1231         }
1232
1233         return IRQ_HANDLED;
1234 }
1235
1236 static int e1000_request_irq(struct e1000_adapter *adapter)
1237 {
1238         struct net_device *netdev = adapter->netdev;
1239         irq_handler_t handler = e1000_intr;
1240         int irq_flags = IRQF_SHARED;
1241         int err;
1242
1243         if (!pci_enable_msi(adapter->pdev)) {
1244                 adapter->flags |= FLAG_MSI_ENABLED;
1245                 handler = e1000_intr_msi;
1246                 irq_flags = 0;
1247         }
1248
1249         err = request_irq(adapter->pdev->irq, handler, irq_flags, netdev->name,
1250                           netdev);
1251         if (err) {
1252                 e_err("Unable to allocate %s interrupt (return: %d)\n",
1253                       adapter->flags & FLAG_MSI_ENABLED ? "MSI":"INTx", err);
1254                 if (adapter->flags & FLAG_MSI_ENABLED)
1255                         pci_disable_msi(adapter->pdev);
1256         }
1257
1258         return err;
1259 }
1260
1261 static void e1000_free_irq(struct e1000_adapter *adapter)
1262 {
1263         struct net_device *netdev = adapter->netdev;
1264
1265         free_irq(adapter->pdev->irq, netdev);
1266         if (adapter->flags & FLAG_MSI_ENABLED) {
1267                 pci_disable_msi(adapter->pdev);
1268                 adapter->flags &= ~FLAG_MSI_ENABLED;
1269         }
1270 }
1271
1272 /**
1273  * e1000_irq_disable - Mask off interrupt generation on the NIC
1274  **/
1275 static void e1000_irq_disable(struct e1000_adapter *adapter)
1276 {
1277         struct e1000_hw *hw = &adapter->hw;
1278
1279         ew32(IMC, ~0);
1280         e1e_flush();
1281         synchronize_irq(adapter->pdev->irq);
1282 }
1283
1284 /**
1285  * e1000_irq_enable - Enable default interrupt generation settings
1286  **/
1287 static void e1000_irq_enable(struct e1000_adapter *adapter)
1288 {
1289         struct e1000_hw *hw = &adapter->hw;
1290
1291         ew32(IMS, IMS_ENABLE_MASK);
1292         e1e_flush();
1293 }
1294
1295 /**
1296  * e1000_get_hw_control - get control of the h/w from f/w
1297  * @adapter: address of board private structure
1298  *
1299  * e1000_get_hw_control sets {CTRL_EXT|SWSM}:DRV_LOAD bit.
1300  * For ASF and Pass Through versions of f/w this means that
1301  * the driver is loaded. For AMT version (only with 82573)
1302  * of the f/w this means that the network i/f is open.
1303  **/
1304 static void e1000_get_hw_control(struct e1000_adapter *adapter)
1305 {
1306         struct e1000_hw *hw = &adapter->hw;
1307         u32 ctrl_ext;
1308         u32 swsm;
1309
1310         /* Let firmware know the driver has taken over */
1311         if (adapter->flags & FLAG_HAS_SWSM_ON_LOAD) {
1312                 swsm = er32(SWSM);
1313                 ew32(SWSM, swsm | E1000_SWSM_DRV_LOAD);
1314         } else if (adapter->flags & FLAG_HAS_CTRLEXT_ON_LOAD) {
1315                 ctrl_ext = er32(CTRL_EXT);
1316                 ew32(CTRL_EXT, ctrl_ext | E1000_CTRL_EXT_DRV_LOAD);
1317         }
1318 }
1319
1320 /**
1321  * e1000_release_hw_control - release control of the h/w to f/w
1322  * @adapter: address of board private structure
1323  *
1324  * e1000_release_hw_control resets {CTRL_EXT|SWSM}:DRV_LOAD bit.
1325  * For ASF and Pass Through versions of f/w this means that the
1326  * driver is no longer loaded. For AMT version (only with 82573) i
1327  * of the f/w this means that the network i/f is closed.
1328  *
1329  **/
1330 static void e1000_release_hw_control(struct e1000_adapter *adapter)
1331 {
1332         struct e1000_hw *hw = &adapter->hw;
1333         u32 ctrl_ext;
1334         u32 swsm;
1335
1336         /* Let firmware taken over control of h/w */
1337         if (adapter->flags & FLAG_HAS_SWSM_ON_LOAD) {
1338                 swsm = er32(SWSM);
1339                 ew32(SWSM, swsm & ~E1000_SWSM_DRV_LOAD);
1340         } else if (adapter->flags & FLAG_HAS_CTRLEXT_ON_LOAD) {
1341                 ctrl_ext = er32(CTRL_EXT);
1342                 ew32(CTRL_EXT, ctrl_ext & ~E1000_CTRL_EXT_DRV_LOAD);
1343         }
1344 }
1345
1346 /**
1347  * @e1000_alloc_ring - allocate memory for a ring structure
1348  **/
1349 static int e1000_alloc_ring_dma(struct e1000_adapter *adapter,
1350                                 struct e1000_ring *ring)
1351 {
1352         struct pci_dev *pdev = adapter->pdev;
1353
1354         ring->desc = dma_alloc_coherent(&pdev->dev, ring->size, &ring->dma,
1355                                         GFP_KERNEL);
1356         if (!ring->desc)
1357                 return -ENOMEM;
1358
1359         return 0;
1360 }
1361
1362 /**
1363  * e1000e_setup_tx_resources - allocate Tx resources (Descriptors)
1364  * @adapter: board private structure
1365  *
1366  * Return 0 on success, negative on failure
1367  **/
1368 int e1000e_setup_tx_resources(struct e1000_adapter *adapter)
1369 {
1370         struct e1000_ring *tx_ring = adapter->tx_ring;
1371         int err = -ENOMEM, size;
1372
1373         size = sizeof(struct e1000_buffer) * tx_ring->count;
1374         tx_ring->buffer_info = vmalloc(size);
1375         if (!tx_ring->buffer_info)
1376                 goto err;
1377         memset(tx_ring->buffer_info, 0, size);
1378
1379         /* round up to nearest 4K */
1380         tx_ring->size = tx_ring->count * sizeof(struct e1000_tx_desc);
1381         tx_ring->size = ALIGN(tx_ring->size, 4096);
1382
1383         err = e1000_alloc_ring_dma(adapter, tx_ring);
1384         if (err)
1385                 goto err;
1386
1387         tx_ring->next_to_use = 0;
1388         tx_ring->next_to_clean = 0;
1389         spin_lock_init(&adapter->tx_queue_lock);
1390
1391         return 0;
1392 err:
1393         vfree(tx_ring->buffer_info);
1394         e_err("Unable to allocate memory for the transmit descriptor ring\n");
1395         return err;
1396 }
1397
1398 /**
1399  * e1000e_setup_rx_resources - allocate Rx resources (Descriptors)
1400  * @adapter: board private structure
1401  *
1402  * Returns 0 on success, negative on failure
1403  **/
1404 int e1000e_setup_rx_resources(struct e1000_adapter *adapter)
1405 {
1406         struct e1000_ring *rx_ring = adapter->rx_ring;
1407         struct e1000_buffer *buffer_info;
1408         int i, size, desc_len, err = -ENOMEM;
1409
1410         size = sizeof(struct e1000_buffer) * rx_ring->count;
1411         rx_ring->buffer_info = vmalloc(size);
1412         if (!rx_ring->buffer_info)
1413                 goto err;
1414         memset(rx_ring->buffer_info, 0, size);
1415
1416         for (i = 0; i < rx_ring->count; i++) {
1417                 buffer_info = &rx_ring->buffer_info[i];
1418                 buffer_info->ps_pages = kcalloc(PS_PAGE_BUFFERS,
1419                                                 sizeof(struct e1000_ps_page),
1420                                                 GFP_KERNEL);
1421                 if (!buffer_info->ps_pages)
1422                         goto err_pages;
1423         }
1424
1425         desc_len = sizeof(union e1000_rx_desc_packet_split);
1426
1427         /* Round up to nearest 4K */
1428         rx_ring->size = rx_ring->count * desc_len;
1429         rx_ring->size = ALIGN(rx_ring->size, 4096);
1430
1431         err = e1000_alloc_ring_dma(adapter, rx_ring);
1432         if (err)
1433                 goto err_pages;
1434
1435         rx_ring->next_to_clean = 0;
1436         rx_ring->next_to_use = 0;
1437         rx_ring->rx_skb_top = NULL;
1438
1439         return 0;
1440
1441 err_pages:
1442         for (i = 0; i < rx_ring->count; i++) {
1443                 buffer_info = &rx_ring->buffer_info[i];
1444                 kfree(buffer_info->ps_pages);
1445         }
1446 err:
1447         vfree(rx_ring->buffer_info);
1448         e_err("Unable to allocate memory for the transmit descriptor ring\n");
1449         return err;
1450 }
1451
1452 /**
1453  * e1000_clean_tx_ring - Free Tx Buffers
1454  * @adapter: board private structure
1455  **/
1456 static void e1000_clean_tx_ring(struct e1000_adapter *adapter)
1457 {
1458         struct e1000_ring *tx_ring = adapter->tx_ring;
1459         struct e1000_buffer *buffer_info;
1460         unsigned long size;
1461         unsigned int i;
1462
1463         for (i = 0; i < tx_ring->count; i++) {
1464                 buffer_info = &tx_ring->buffer_info[i];
1465                 e1000_put_txbuf(adapter, buffer_info);
1466         }
1467
1468         size = sizeof(struct e1000_buffer) * tx_ring->count;
1469         memset(tx_ring->buffer_info, 0, size);
1470
1471         memset(tx_ring->desc, 0, tx_ring->size);
1472
1473         tx_ring->next_to_use = 0;
1474         tx_ring->next_to_clean = 0;
1475
1476         writel(0, adapter->hw.hw_addr + tx_ring->head);
1477         writel(0, adapter->hw.hw_addr + tx_ring->tail);
1478 }
1479
1480 /**
1481  * e1000e_free_tx_resources - Free Tx Resources per Queue
1482  * @adapter: board private structure
1483  *
1484  * Free all transmit software resources
1485  **/
1486 void e1000e_free_tx_resources(struct e1000_adapter *adapter)
1487 {
1488         struct pci_dev *pdev = adapter->pdev;
1489         struct e1000_ring *tx_ring = adapter->tx_ring;
1490
1491         e1000_clean_tx_ring(adapter);
1492
1493         vfree(tx_ring->buffer_info);
1494         tx_ring->buffer_info = NULL;
1495
1496         dma_free_coherent(&pdev->dev, tx_ring->size, tx_ring->desc,
1497                           tx_ring->dma);
1498         tx_ring->desc = NULL;
1499 }
1500
1501 /**
1502  * e1000e_free_rx_resources - Free Rx Resources
1503  * @adapter: board private structure
1504  *
1505  * Free all receive software resources
1506  **/
1507
1508 void e1000e_free_rx_resources(struct e1000_adapter *adapter)
1509 {
1510         struct pci_dev *pdev = adapter->pdev;
1511         struct e1000_ring *rx_ring = adapter->rx_ring;
1512         int i;
1513
1514         e1000_clean_rx_ring(adapter);
1515
1516         for (i = 0; i < rx_ring->count; i++) {
1517                 kfree(rx_ring->buffer_info[i].ps_pages);
1518         }
1519
1520         vfree(rx_ring->buffer_info);
1521         rx_ring->buffer_info = NULL;
1522
1523         dma_free_coherent(&pdev->dev, rx_ring->size, rx_ring->desc,
1524                           rx_ring->dma);
1525         rx_ring->desc = NULL;
1526 }
1527
1528 /**
1529  * e1000_update_itr - update the dynamic ITR value based on statistics
1530  * @adapter: pointer to adapter
1531  * @itr_setting: current adapter->itr
1532  * @packets: the number of packets during this measurement interval
1533  * @bytes: the number of bytes during this measurement interval
1534  *
1535  *      Stores a new ITR value based on packets and byte
1536  *      counts during the last interrupt.  The advantage of per interrupt
1537  *      computation is faster updates and more accurate ITR for the current
1538  *      traffic pattern.  Constants in this function were computed
1539  *      based on theoretical maximum wire speed and thresholds were set based
1540  *      on testing data as well as attempting to minimize response time
1541  *      while increasing bulk throughput.
1542  *      this functionality is controlled by the InterruptThrottleRate module
1543  *      parameter (see e1000_param.c)
1544  **/
1545 static unsigned int e1000_update_itr(struct e1000_adapter *adapter,
1546                                      u16 itr_setting, int packets,
1547                                      int bytes)
1548 {
1549         unsigned int retval = itr_setting;
1550
1551         if (packets == 0)
1552                 goto update_itr_done;
1553
1554         switch (itr_setting) {
1555         case lowest_latency:
1556                 /* handle TSO and jumbo frames */
1557                 if (bytes/packets > 8000)
1558                         retval = bulk_latency;
1559                 else if ((packets < 5) && (bytes > 512)) {
1560                         retval = low_latency;
1561                 }
1562                 break;
1563         case low_latency:  /* 50 usec aka 20000 ints/s */
1564                 if (bytes > 10000) {
1565                         /* this if handles the TSO accounting */
1566                         if (bytes/packets > 8000) {
1567                                 retval = bulk_latency;
1568                         } else if ((packets < 10) || ((bytes/packets) > 1200)) {
1569                                 retval = bulk_latency;
1570                         } else if ((packets > 35)) {
1571                                 retval = lowest_latency;
1572                         }
1573                 } else if (bytes/packets > 2000) {
1574                         retval = bulk_latency;
1575                 } else if (packets <= 2 && bytes < 512) {
1576                         retval = lowest_latency;
1577                 }
1578                 break;
1579         case bulk_latency: /* 250 usec aka 4000 ints/s */
1580                 if (bytes > 25000) {
1581                         if (packets > 35) {
1582                                 retval = low_latency;
1583                         }
1584                 } else if (bytes < 6000) {
1585                         retval = low_latency;
1586                 }
1587                 break;
1588         }
1589
1590 update_itr_done:
1591         return retval;
1592 }
1593
1594 static void e1000_set_itr(struct e1000_adapter *adapter)
1595 {
1596         struct e1000_hw *hw = &adapter->hw;
1597         u16 current_itr;
1598         u32 new_itr = adapter->itr;
1599
1600         /* for non-gigabit speeds, just fix the interrupt rate at 4000 */
1601         if (adapter->link_speed != SPEED_1000) {
1602                 current_itr = 0;
1603                 new_itr = 4000;
1604                 goto set_itr_now;
1605         }
1606
1607         adapter->tx_itr = e1000_update_itr(adapter,
1608                                     adapter->tx_itr,
1609                                     adapter->total_tx_packets,
1610                                     adapter->total_tx_bytes);
1611         /* conservative mode (itr 3) eliminates the lowest_latency setting */
1612         if (adapter->itr_setting == 3 && adapter->tx_itr == lowest_latency)
1613                 adapter->tx_itr = low_latency;
1614
1615         adapter->rx_itr = e1000_update_itr(adapter,
1616                                     adapter->rx_itr,
1617                                     adapter->total_rx_packets,
1618                                     adapter->total_rx_bytes);
1619         /* conservative mode (itr 3) eliminates the lowest_latency setting */
1620         if (adapter->itr_setting == 3 && adapter->rx_itr == lowest_latency)
1621                 adapter->rx_itr = low_latency;
1622
1623         current_itr = max(adapter->rx_itr, adapter->tx_itr);
1624
1625         switch (current_itr) {
1626         /* counts and packets in update_itr are dependent on these numbers */
1627         case lowest_latency:
1628                 new_itr = 70000;
1629                 break;
1630         case low_latency:
1631                 new_itr = 20000; /* aka hwitr = ~200 */
1632                 break;
1633         case bulk_latency:
1634                 new_itr = 4000;
1635                 break;
1636         default:
1637                 break;
1638         }
1639
1640 set_itr_now:
1641         if (new_itr != adapter->itr) {
1642                 /*
1643                  * this attempts to bias the interrupt rate towards Bulk
1644                  * by adding intermediate steps when interrupt rate is
1645                  * increasing
1646                  */
1647                 new_itr = new_itr > adapter->itr ?
1648                              min(adapter->itr + (new_itr >> 2), new_itr) :
1649                              new_itr;
1650                 adapter->itr = new_itr;
1651                 ew32(ITR, 1000000000 / (new_itr * 256));
1652         }
1653 }
1654
1655 /**
1656  * e1000_clean - NAPI Rx polling callback
1657  * @napi: struct associated with this polling callback
1658  * @budget: amount of packets driver is allowed to process this poll
1659  **/
1660 static int e1000_clean(struct napi_struct *napi, int budget)
1661 {
1662         struct e1000_adapter *adapter = container_of(napi, struct e1000_adapter, napi);
1663         struct net_device *poll_dev = adapter->netdev;
1664         int tx_cleaned = 0, work_done = 0;
1665
1666         /* Must NOT use netdev_priv macro here. */
1667         adapter = poll_dev->priv;
1668
1669         /*
1670          * e1000_clean is called per-cpu.  This lock protects
1671          * tx_ring from being cleaned by multiple cpus
1672          * simultaneously.  A failure obtaining the lock means
1673          * tx_ring is currently being cleaned anyway.
1674          */
1675         if (spin_trylock(&adapter->tx_queue_lock)) {
1676                 tx_cleaned = e1000_clean_tx_irq(adapter);
1677                 spin_unlock(&adapter->tx_queue_lock);
1678         }
1679
1680         adapter->clean_rx(adapter, &work_done, budget);
1681
1682         if (tx_cleaned)
1683                 work_done = budget;
1684
1685         /* If budget not fully consumed, exit the polling mode */
1686         if (work_done < budget) {
1687                 if (adapter->itr_setting & 3)
1688                         e1000_set_itr(adapter);
1689                 netif_rx_complete(poll_dev, napi);
1690                 e1000_irq_enable(adapter);
1691         }
1692
1693         return work_done;
1694 }
1695
1696 static void e1000_vlan_rx_add_vid(struct net_device *netdev, u16 vid)
1697 {
1698         struct e1000_adapter *adapter = netdev_priv(netdev);
1699         struct e1000_hw *hw = &adapter->hw;
1700         u32 vfta, index;
1701
1702         /* don't update vlan cookie if already programmed */
1703         if ((adapter->hw.mng_cookie.status &
1704              E1000_MNG_DHCP_COOKIE_STATUS_VLAN) &&
1705             (vid == adapter->mng_vlan_id))
1706                 return;
1707         /* add VID to filter table */
1708         index = (vid >> 5) & 0x7F;
1709         vfta = E1000_READ_REG_ARRAY(hw, E1000_VFTA, index);
1710         vfta |= (1 << (vid & 0x1F));
1711         e1000e_write_vfta(hw, index, vfta);
1712 }
1713
1714 static void e1000_vlan_rx_kill_vid(struct net_device *netdev, u16 vid)
1715 {
1716         struct e1000_adapter *adapter = netdev_priv(netdev);
1717         struct e1000_hw *hw = &adapter->hw;
1718         u32 vfta, index;
1719
1720         if (!test_bit(__E1000_DOWN, &adapter->state))
1721                 e1000_irq_disable(adapter);
1722         vlan_group_set_device(adapter->vlgrp, vid, NULL);
1723
1724         if (!test_bit(__E1000_DOWN, &adapter->state))
1725                 e1000_irq_enable(adapter);
1726
1727         if ((adapter->hw.mng_cookie.status &
1728              E1000_MNG_DHCP_COOKIE_STATUS_VLAN) &&
1729             (vid == adapter->mng_vlan_id)) {
1730                 /* release control to f/w */
1731                 e1000_release_hw_control(adapter);
1732                 return;
1733         }
1734
1735         /* remove VID from filter table */
1736         index = (vid >> 5) & 0x7F;
1737         vfta = E1000_READ_REG_ARRAY(hw, E1000_VFTA, index);
1738         vfta &= ~(1 << (vid & 0x1F));
1739         e1000e_write_vfta(hw, index, vfta);
1740 }
1741
1742 static void e1000_update_mng_vlan(struct e1000_adapter *adapter)
1743 {
1744         struct net_device *netdev = adapter->netdev;
1745         u16 vid = adapter->hw.mng_cookie.vlan_id;
1746         u16 old_vid = adapter->mng_vlan_id;
1747
1748         if (!adapter->vlgrp)
1749                 return;
1750
1751         if (!vlan_group_get_device(adapter->vlgrp, vid)) {
1752                 adapter->mng_vlan_id = E1000_MNG_VLAN_NONE;
1753                 if (adapter->hw.mng_cookie.status &
1754                         E1000_MNG_DHCP_COOKIE_STATUS_VLAN) {
1755                         e1000_vlan_rx_add_vid(netdev, vid);
1756                         adapter->mng_vlan_id = vid;
1757                 }
1758
1759                 if ((old_vid != (u16)E1000_MNG_VLAN_NONE) &&
1760                                 (vid != old_vid) &&
1761                     !vlan_group_get_device(adapter->vlgrp, old_vid))
1762                         e1000_vlan_rx_kill_vid(netdev, old_vid);
1763         } else {
1764                 adapter->mng_vlan_id = vid;
1765         }
1766 }
1767
1768
1769 static void e1000_vlan_rx_register(struct net_device *netdev,
1770                                    struct vlan_group *grp)
1771 {
1772         struct e1000_adapter *adapter = netdev_priv(netdev);
1773         struct e1000_hw *hw = &adapter->hw;
1774         u32 ctrl, rctl;
1775
1776         if (!test_bit(__E1000_DOWN, &adapter->state))
1777                 e1000_irq_disable(adapter);
1778         adapter->vlgrp = grp;
1779
1780         if (grp) {
1781                 /* enable VLAN tag insert/strip */
1782                 ctrl = er32(CTRL);
1783                 ctrl |= E1000_CTRL_VME;
1784                 ew32(CTRL, ctrl);
1785
1786                 if (adapter->flags & FLAG_HAS_HW_VLAN_FILTER) {
1787                         /* enable VLAN receive filtering */
1788                         rctl = er32(RCTL);
1789                         rctl &= ~E1000_RCTL_CFIEN;
1790                         ew32(RCTL, rctl);
1791                         e1000_update_mng_vlan(adapter);
1792                 }
1793         } else {
1794                 /* disable VLAN tag insert/strip */
1795                 ctrl = er32(CTRL);
1796                 ctrl &= ~E1000_CTRL_VME;
1797                 ew32(CTRL, ctrl);
1798
1799                 if (adapter->flags & FLAG_HAS_HW_VLAN_FILTER) {
1800                         if (adapter->mng_vlan_id !=
1801                             (u16)E1000_MNG_VLAN_NONE) {
1802                                 e1000_vlan_rx_kill_vid(netdev,
1803                                                        adapter->mng_vlan_id);
1804                                 adapter->mng_vlan_id = E1000_MNG_VLAN_NONE;
1805                         }
1806                 }
1807         }
1808
1809         if (!test_bit(__E1000_DOWN, &adapter->state))
1810                 e1000_irq_enable(adapter);
1811 }
1812
1813 static void e1000_restore_vlan(struct e1000_adapter *adapter)
1814 {
1815         u16 vid;
1816
1817         e1000_vlan_rx_register(adapter->netdev, adapter->vlgrp);
1818
1819         if (!adapter->vlgrp)
1820                 return;
1821
1822         for (vid = 0; vid < VLAN_GROUP_ARRAY_LEN; vid++) {
1823                 if (!vlan_group_get_device(adapter->vlgrp, vid))
1824                         continue;
1825                 e1000_vlan_rx_add_vid(adapter->netdev, vid);
1826         }
1827 }
1828
1829 static void e1000_init_manageability(struct e1000_adapter *adapter)
1830 {
1831         struct e1000_hw *hw = &adapter->hw;
1832         u32 manc, manc2h;
1833
1834         if (!(adapter->flags & FLAG_MNG_PT_ENABLED))
1835                 return;
1836
1837         manc = er32(MANC);
1838
1839         /*
1840          * enable receiving management packets to the host. this will probably
1841          * generate destination unreachable messages from the host OS, but
1842          * the packets will be handled on SMBUS
1843          */
1844         manc |= E1000_MANC_EN_MNG2HOST;
1845         manc2h = er32(MANC2H);
1846 #define E1000_MNG2HOST_PORT_623 (1 << 5)
1847 #define E1000_MNG2HOST_PORT_664 (1 << 6)
1848         manc2h |= E1000_MNG2HOST_PORT_623;
1849         manc2h |= E1000_MNG2HOST_PORT_664;
1850         ew32(MANC2H, manc2h);
1851         ew32(MANC, manc);
1852 }
1853
1854 /**
1855  * e1000_configure_tx - Configure 8254x Transmit Unit after Reset
1856  * @adapter: board private structure
1857  *
1858  * Configure the Tx unit of the MAC after a reset.
1859  **/
1860 static void e1000_configure_tx(struct e1000_adapter *adapter)
1861 {
1862         struct e1000_hw *hw = &adapter->hw;
1863         struct e1000_ring *tx_ring = adapter->tx_ring;
1864         u64 tdba;
1865         u32 tdlen, tctl, tipg, tarc;
1866         u32 ipgr1, ipgr2;
1867
1868         /* Setup the HW Tx Head and Tail descriptor pointers */
1869         tdba = tx_ring->dma;
1870         tdlen = tx_ring->count * sizeof(struct e1000_tx_desc);
1871         ew32(TDBAL, (tdba & DMA_32BIT_MASK));
1872         ew32(TDBAH, (tdba >> 32));
1873         ew32(TDLEN, tdlen);
1874         ew32(TDH, 0);
1875         ew32(TDT, 0);
1876         tx_ring->head = E1000_TDH;
1877         tx_ring->tail = E1000_TDT;
1878
1879         /* Set the default values for the Tx Inter Packet Gap timer */
1880         tipg = DEFAULT_82543_TIPG_IPGT_COPPER;          /*  8  */
1881         ipgr1 = DEFAULT_82543_TIPG_IPGR1;               /*  8  */
1882         ipgr2 = DEFAULT_82543_TIPG_IPGR2;               /*  6  */
1883
1884         if (adapter->flags & FLAG_TIPG_MEDIUM_FOR_80003ESLAN)
1885                 ipgr2 = DEFAULT_80003ES2LAN_TIPG_IPGR2; /*  7  */
1886
1887         tipg |= ipgr1 << E1000_TIPG_IPGR1_SHIFT;
1888         tipg |= ipgr2 << E1000_TIPG_IPGR2_SHIFT;
1889         ew32(TIPG, tipg);
1890
1891         /* Set the Tx Interrupt Delay register */
1892         ew32(TIDV, adapter->tx_int_delay);
1893         /* Tx irq moderation */
1894         ew32(TADV, adapter->tx_abs_int_delay);
1895
1896         /* Program the Transmit Control Register */
1897         tctl = er32(TCTL);
1898         tctl &= ~E1000_TCTL_CT;
1899         tctl |= E1000_TCTL_PSP | E1000_TCTL_RTLC |
1900                 (E1000_COLLISION_THRESHOLD << E1000_CT_SHIFT);
1901
1902         if (adapter->flags & FLAG_TARC_SPEED_MODE_BIT) {
1903                 tarc = er32(TARC(0));
1904                 /*
1905                  * set the speed mode bit, we'll clear it if we're not at
1906                  * gigabit link later
1907                  */
1908 #define SPEED_MODE_BIT (1 << 21)
1909                 tarc |= SPEED_MODE_BIT;
1910                 ew32(TARC(0), tarc);
1911         }
1912
1913         /* errata: program both queues to unweighted RR */
1914         if (adapter->flags & FLAG_TARC_SET_BIT_ZERO) {
1915                 tarc = er32(TARC(0));
1916                 tarc |= 1;
1917                 ew32(TARC(0), tarc);
1918                 tarc = er32(TARC(1));
1919                 tarc |= 1;
1920                 ew32(TARC(1), tarc);
1921         }
1922
1923         e1000e_config_collision_dist(hw);
1924
1925         /* Setup Transmit Descriptor Settings for eop descriptor */
1926         adapter->txd_cmd = E1000_TXD_CMD_EOP | E1000_TXD_CMD_IFCS;
1927
1928         /* only set IDE if we are delaying interrupts using the timers */
1929         if (adapter->tx_int_delay)
1930                 adapter->txd_cmd |= E1000_TXD_CMD_IDE;
1931
1932         /* enable Report Status bit */
1933         adapter->txd_cmd |= E1000_TXD_CMD_RS;
1934
1935         ew32(TCTL, tctl);
1936
1937         adapter->tx_queue_len = adapter->netdev->tx_queue_len;
1938 }
1939
1940 /**
1941  * e1000_setup_rctl - configure the receive control registers
1942  * @adapter: Board private structure
1943  **/
1944 #define PAGE_USE_COUNT(S) (((S) >> PAGE_SHIFT) + \
1945                            (((S) & (PAGE_SIZE - 1)) ? 1 : 0))
1946 static void e1000_setup_rctl(struct e1000_adapter *adapter)
1947 {
1948         struct e1000_hw *hw = &adapter->hw;
1949         u32 rctl, rfctl;
1950         u32 psrctl = 0;
1951         u32 pages = 0;
1952
1953         /* Program MC offset vector base */
1954         rctl = er32(RCTL);
1955         rctl &= ~(3 << E1000_RCTL_MO_SHIFT);
1956         rctl |= E1000_RCTL_EN | E1000_RCTL_BAM |
1957                 E1000_RCTL_LBM_NO | E1000_RCTL_RDMTS_HALF |
1958                 (adapter->hw.mac.mc_filter_type << E1000_RCTL_MO_SHIFT);
1959
1960         /* Do not Store bad packets */
1961         rctl &= ~E1000_RCTL_SBP;
1962
1963         /* Enable Long Packet receive */
1964         if (adapter->netdev->mtu <= ETH_DATA_LEN)
1965                 rctl &= ~E1000_RCTL_LPE;
1966         else
1967                 rctl |= E1000_RCTL_LPE;
1968
1969         /* Enable hardware CRC frame stripping */
1970         rctl |= E1000_RCTL_SECRC;
1971
1972         /* Setup buffer sizes */
1973         rctl &= ~E1000_RCTL_SZ_4096;
1974         rctl |= E1000_RCTL_BSEX;
1975         switch (adapter->rx_buffer_len) {
1976         case 256:
1977                 rctl |= E1000_RCTL_SZ_256;
1978                 rctl &= ~E1000_RCTL_BSEX;
1979                 break;
1980         case 512:
1981                 rctl |= E1000_RCTL_SZ_512;
1982                 rctl &= ~E1000_RCTL_BSEX;
1983                 break;
1984         case 1024:
1985                 rctl |= E1000_RCTL_SZ_1024;
1986                 rctl &= ~E1000_RCTL_BSEX;
1987                 break;
1988         case 2048:
1989         default:
1990                 rctl |= E1000_RCTL_SZ_2048;
1991                 rctl &= ~E1000_RCTL_BSEX;
1992                 break;
1993         case 4096:
1994                 rctl |= E1000_RCTL_SZ_4096;
1995                 break;
1996         case 8192:
1997                 rctl |= E1000_RCTL_SZ_8192;
1998                 break;
1999         case 16384:
2000                 rctl |= E1000_RCTL_SZ_16384;
2001                 break;
2002         }
2003
2004         /*
2005          * 82571 and greater support packet-split where the protocol
2006          * header is placed in skb->data and the packet data is
2007          * placed in pages hanging off of skb_shinfo(skb)->nr_frags.
2008          * In the case of a non-split, skb->data is linearly filled,
2009          * followed by the page buffers.  Therefore, skb->data is
2010          * sized to hold the largest protocol header.
2011          *
2012          * allocations using alloc_page take too long for regular MTU
2013          * so only enable packet split for jumbo frames
2014          *
2015          * Using pages when the page size is greater than 16k wastes
2016          * a lot of memory, since we allocate 3 pages at all times
2017          * per packet.
2018          */
2019         pages = PAGE_USE_COUNT(adapter->netdev->mtu);
2020         if (!(adapter->flags & FLAG_IS_ICH) && (pages <= 3) &&
2021             (PAGE_SIZE <= 16384) && (rctl & E1000_RCTL_LPE))
2022                 adapter->rx_ps_pages = pages;
2023         else
2024                 adapter->rx_ps_pages = 0;
2025
2026         if (adapter->rx_ps_pages) {
2027                 /* Configure extra packet-split registers */
2028                 rfctl = er32(RFCTL);
2029                 rfctl |= E1000_RFCTL_EXTEN;
2030                 /*
2031                  * disable packet split support for IPv6 extension headers,
2032                  * because some malformed IPv6 headers can hang the Rx
2033                  */
2034                 rfctl |= (E1000_RFCTL_IPV6_EX_DIS |
2035                           E1000_RFCTL_NEW_IPV6_EXT_DIS);
2036
2037                 ew32(RFCTL, rfctl);
2038
2039                 /* Enable Packet split descriptors */
2040                 rctl |= E1000_RCTL_DTYP_PS;
2041
2042                 psrctl |= adapter->rx_ps_bsize0 >>
2043                         E1000_PSRCTL_BSIZE0_SHIFT;
2044
2045                 switch (adapter->rx_ps_pages) {
2046                 case 3:
2047                         psrctl |= PAGE_SIZE <<
2048                                 E1000_PSRCTL_BSIZE3_SHIFT;
2049                 case 2:
2050                         psrctl |= PAGE_SIZE <<
2051                                 E1000_PSRCTL_BSIZE2_SHIFT;
2052                 case 1:
2053                         psrctl |= PAGE_SIZE >>
2054                                 E1000_PSRCTL_BSIZE1_SHIFT;
2055                         break;
2056                 }
2057
2058                 ew32(PSRCTL, psrctl);
2059         }
2060
2061         ew32(RCTL, rctl);
2062         /* just started the receive unit, no need to restart */
2063         adapter->flags &= ~FLAG_RX_RESTART_NOW;
2064 }
2065
2066 /**
2067  * e1000_configure_rx - Configure Receive Unit after Reset
2068  * @adapter: board private structure
2069  *
2070  * Configure the Rx unit of the MAC after a reset.
2071  **/
2072 static void e1000_configure_rx(struct e1000_adapter *adapter)
2073 {
2074         struct e1000_hw *hw = &adapter->hw;
2075         struct e1000_ring *rx_ring = adapter->rx_ring;
2076         u64 rdba;
2077         u32 rdlen, rctl, rxcsum, ctrl_ext;
2078
2079         if (adapter->rx_ps_pages) {
2080                 /* this is a 32 byte descriptor */
2081                 rdlen = rx_ring->count *
2082                         sizeof(union e1000_rx_desc_packet_split);
2083                 adapter->clean_rx = e1000_clean_rx_irq_ps;
2084                 adapter->alloc_rx_buf = e1000_alloc_rx_buffers_ps;
2085         } else if (adapter->netdev->mtu > ETH_FRAME_LEN + ETH_FCS_LEN) {
2086                 rdlen = rx_ring->count * sizeof(struct e1000_rx_desc);
2087                 adapter->clean_rx = e1000_clean_jumbo_rx_irq;
2088                 adapter->alloc_rx_buf = e1000_alloc_jumbo_rx_buffers;
2089         } else {
2090                 rdlen = rx_ring->count * sizeof(struct e1000_rx_desc);
2091                 adapter->clean_rx = e1000_clean_rx_irq;
2092                 adapter->alloc_rx_buf = e1000_alloc_rx_buffers;
2093         }
2094
2095         /* disable receives while setting up the descriptors */
2096         rctl = er32(RCTL);
2097         ew32(RCTL, rctl & ~E1000_RCTL_EN);
2098         e1e_flush();
2099         msleep(10);
2100
2101         /* set the Receive Delay Timer Register */
2102         ew32(RDTR, adapter->rx_int_delay);
2103
2104         /* irq moderation */
2105         ew32(RADV, adapter->rx_abs_int_delay);
2106         if (adapter->itr_setting != 0)
2107                 ew32(ITR, 1000000000 / (adapter->itr * 256));
2108
2109         ctrl_ext = er32(CTRL_EXT);
2110         /* Reset delay timers after every interrupt */
2111         ctrl_ext |= E1000_CTRL_EXT_INT_TIMER_CLR;
2112         /* Auto-Mask interrupts upon ICR access */
2113         ctrl_ext |= E1000_CTRL_EXT_IAME;
2114         ew32(IAM, 0xffffffff);
2115         ew32(CTRL_EXT, ctrl_ext);
2116         e1e_flush();
2117
2118         /*
2119          * Setup the HW Rx Head and Tail Descriptor Pointers and
2120          * the Base and Length of the Rx Descriptor Ring
2121          */
2122         rdba = rx_ring->dma;
2123         ew32(RDBAL, (rdba & DMA_32BIT_MASK));
2124         ew32(RDBAH, (rdba >> 32));
2125         ew32(RDLEN, rdlen);
2126         ew32(RDH, 0);
2127         ew32(RDT, 0);
2128         rx_ring->head = E1000_RDH;
2129         rx_ring->tail = E1000_RDT;
2130
2131         /* Enable Receive Checksum Offload for TCP and UDP */
2132         rxcsum = er32(RXCSUM);
2133         if (adapter->flags & FLAG_RX_CSUM_ENABLED) {
2134                 rxcsum |= E1000_RXCSUM_TUOFL;
2135
2136                 /*
2137                  * IPv4 payload checksum for UDP fragments must be
2138                  * used in conjunction with packet-split.
2139                  */
2140                 if (adapter->rx_ps_pages)
2141                         rxcsum |= E1000_RXCSUM_IPPCSE;
2142         } else {
2143                 rxcsum &= ~E1000_RXCSUM_TUOFL;
2144                 /* no need to clear IPPCSE as it defaults to 0 */
2145         }
2146         ew32(RXCSUM, rxcsum);
2147
2148         /*
2149          * Enable early receives on supported devices, only takes effect when
2150          * packet size is equal or larger than the specified value (in 8 byte
2151          * units), e.g. using jumbo frames when setting to E1000_ERT_2048
2152          */
2153         if ((adapter->flags & FLAG_HAS_ERT) &&
2154             (adapter->netdev->mtu > ETH_DATA_LEN)) {
2155                 u32 rxdctl = er32(RXDCTL(0));
2156                 ew32(RXDCTL(0), rxdctl | 0x3);
2157                 ew32(ERT, E1000_ERT_2048 | (1 << 13));
2158                 /*
2159                  * With jumbo frames and early-receive enabled, excessive
2160                  * C4->C2 latencies result in dropped transactions.
2161                  */
2162                 pm_qos_update_requirement(PM_QOS_CPU_DMA_LATENCY,
2163                                           e1000e_driver_name, 55);
2164         } else {
2165                 pm_qos_update_requirement(PM_QOS_CPU_DMA_LATENCY,
2166                                           e1000e_driver_name,
2167                                           PM_QOS_DEFAULT_VALUE);
2168         }
2169
2170         /* Enable Receives */
2171         ew32(RCTL, rctl);
2172 }
2173
2174 /**
2175  *  e1000_update_mc_addr_list - Update Multicast addresses
2176  *  @hw: pointer to the HW structure
2177  *  @mc_addr_list: array of multicast addresses to program
2178  *  @mc_addr_count: number of multicast addresses to program
2179  *  @rar_used_count: the first RAR register free to program
2180  *  @rar_count: total number of supported Receive Address Registers
2181  *
2182  *  Updates the Receive Address Registers and Multicast Table Array.
2183  *  The caller must have a packed mc_addr_list of multicast addresses.
2184  *  The parameter rar_count will usually be hw->mac.rar_entry_count
2185  *  unless there are workarounds that change this.  Currently no func pointer
2186  *  exists and all implementations are handled in the generic version of this
2187  *  function.
2188  **/
2189 static void e1000_update_mc_addr_list(struct e1000_hw *hw, u8 *mc_addr_list,
2190                                       u32 mc_addr_count, u32 rar_used_count,
2191                                       u32 rar_count)
2192 {
2193         hw->mac.ops.update_mc_addr_list(hw, mc_addr_list, mc_addr_count,
2194                                         rar_used_count, rar_count);
2195 }
2196
2197 /**
2198  * e1000_set_multi - Multicast and Promiscuous mode set
2199  * @netdev: network interface device structure
2200  *
2201  * The set_multi entry point is called whenever the multicast address
2202  * list or the network interface flags are updated.  This routine is
2203  * responsible for configuring the hardware for proper multicast,
2204  * promiscuous mode, and all-multi behavior.
2205  **/
2206 static void e1000_set_multi(struct net_device *netdev)
2207 {
2208         struct e1000_adapter *adapter = netdev_priv(netdev);
2209         struct e1000_hw *hw = &adapter->hw;
2210         struct e1000_mac_info *mac = &hw->mac;
2211         struct dev_mc_list *mc_ptr;
2212         u8  *mta_list;
2213         u32 rctl;
2214         int i;
2215
2216         /* Check for Promiscuous and All Multicast modes */
2217
2218         rctl = er32(RCTL);
2219
2220         if (netdev->flags & IFF_PROMISC) {
2221                 rctl |= (E1000_RCTL_UPE | E1000_RCTL_MPE);
2222                 rctl &= ~E1000_RCTL_VFE;
2223         } else {
2224                 if (netdev->flags & IFF_ALLMULTI) {
2225                         rctl |= E1000_RCTL_MPE;
2226                         rctl &= ~E1000_RCTL_UPE;
2227                 } else {
2228                         rctl &= ~(E1000_RCTL_UPE | E1000_RCTL_MPE);
2229                 }
2230                 if (adapter->flags & FLAG_HAS_HW_VLAN_FILTER)
2231                         rctl |= E1000_RCTL_VFE;
2232         }
2233
2234         ew32(RCTL, rctl);
2235
2236         if (netdev->mc_count) {
2237                 mta_list = kmalloc(netdev->mc_count * 6, GFP_ATOMIC);
2238                 if (!mta_list)
2239                         return;
2240
2241                 /* prepare a packed array of only addresses. */
2242                 mc_ptr = netdev->mc_list;
2243
2244                 for (i = 0; i < netdev->mc_count; i++) {
2245                         if (!mc_ptr)
2246                                 break;
2247                         memcpy(mta_list + (i*ETH_ALEN), mc_ptr->dmi_addr,
2248                                ETH_ALEN);
2249                         mc_ptr = mc_ptr->next;
2250                 }
2251
2252                 e1000_update_mc_addr_list(hw, mta_list, i, 1,
2253                                           mac->rar_entry_count);
2254                 kfree(mta_list);
2255         } else {
2256                 /*
2257                  * if we're called from probe, we might not have
2258                  * anything to do here, so clear out the list
2259                  */
2260                 e1000_update_mc_addr_list(hw, NULL, 0, 1, mac->rar_entry_count);
2261         }
2262 }
2263
2264 /**
2265  * e1000_configure - configure the hardware for Rx and Tx
2266  * @adapter: private board structure
2267  **/
2268 static void e1000_configure(struct e1000_adapter *adapter)
2269 {
2270         e1000_set_multi(adapter->netdev);
2271
2272         e1000_restore_vlan(adapter);
2273         e1000_init_manageability(adapter);
2274
2275         e1000_configure_tx(adapter);
2276         e1000_setup_rctl(adapter);
2277         e1000_configure_rx(adapter);
2278         adapter->alloc_rx_buf(adapter, e1000_desc_unused(adapter->rx_ring));
2279 }
2280
2281 /**
2282  * e1000e_power_up_phy - restore link in case the phy was powered down
2283  * @adapter: address of board private structure
2284  *
2285  * The phy may be powered down to save power and turn off link when the
2286  * driver is unloaded and wake on lan is not enabled (among others)
2287  * *** this routine MUST be followed by a call to e1000e_reset ***
2288  **/
2289 void e1000e_power_up_phy(struct e1000_adapter *adapter)
2290 {
2291         u16 mii_reg = 0;
2292
2293         /* Just clear the power down bit to wake the phy back up */
2294         if (adapter->hw.phy.media_type == e1000_media_type_copper) {
2295                 /*
2296                  * According to the manual, the phy will retain its
2297                  * settings across a power-down/up cycle
2298                  */
2299                 e1e_rphy(&adapter->hw, PHY_CONTROL, &mii_reg);
2300                 mii_reg &= ~MII_CR_POWER_DOWN;
2301                 e1e_wphy(&adapter->hw, PHY_CONTROL, mii_reg);
2302         }
2303
2304         adapter->hw.mac.ops.setup_link(&adapter->hw);
2305 }
2306
2307 /**
2308  * e1000_power_down_phy - Power down the PHY
2309  *
2310  * Power down the PHY so no link is implied when interface is down
2311  * The PHY cannot be powered down is management or WoL is active
2312  */
2313 static void e1000_power_down_phy(struct e1000_adapter *adapter)
2314 {
2315         struct e1000_hw *hw = &adapter->hw;
2316         u16 mii_reg;
2317
2318         /* WoL is enabled */
2319         if (adapter->wol)
2320                 return;
2321
2322         /* non-copper PHY? */
2323         if (adapter->hw.phy.media_type != e1000_media_type_copper)
2324                 return;
2325
2326         /* reset is blocked because of a SoL/IDER session */
2327         if (e1000e_check_mng_mode(hw) || e1000_check_reset_block(hw))
2328                 return;
2329
2330         /* manageability (AMT) is enabled */
2331         if (er32(MANC) & E1000_MANC_SMBUS_EN)
2332                 return;
2333
2334         /* power down the PHY */
2335         e1e_rphy(hw, PHY_CONTROL, &mii_reg);
2336         mii_reg |= MII_CR_POWER_DOWN;
2337         e1e_wphy(hw, PHY_CONTROL, mii_reg);
2338         mdelay(1);
2339 }
2340
2341 /**
2342  * e1000e_reset - bring the hardware into a known good state
2343  *
2344  * This function boots the hardware and enables some settings that
2345  * require a configuration cycle of the hardware - those cannot be
2346  * set/changed during runtime. After reset the device needs to be
2347  * properly configured for Rx, Tx etc.
2348  */
2349 void e1000e_reset(struct e1000_adapter *adapter)
2350 {
2351         struct e1000_mac_info *mac = &adapter->hw.mac;
2352         struct e1000_fc_info *fc = &adapter->hw.fc;
2353         struct e1000_hw *hw = &adapter->hw;
2354         u32 tx_space, min_tx_space, min_rx_space;
2355         u32 pba = adapter->pba;
2356         u16 hwm;
2357
2358         /* reset Packet Buffer Allocation to default */
2359         ew32(PBA, pba);
2360
2361         if (adapter->max_frame_size > ETH_FRAME_LEN + ETH_FCS_LEN) {
2362                 /*
2363                  * To maintain wire speed transmits, the Tx FIFO should be
2364                  * large enough to accommodate two full transmit packets,
2365                  * rounded up to the next 1KB and expressed in KB.  Likewise,
2366                  * the Rx FIFO should be large enough to accommodate at least
2367                  * one full receive packet and is similarly rounded up and
2368                  * expressed in KB.
2369                  */
2370                 pba = er32(PBA);
2371                 /* upper 16 bits has Tx packet buffer allocation size in KB */
2372                 tx_space = pba >> 16;
2373                 /* lower 16 bits has Rx packet buffer allocation size in KB */
2374                 pba &= 0xffff;
2375                 /*
2376                  * the Tx fifo also stores 16 bytes of information about the tx
2377                  * but don't include ethernet FCS because hardware appends it
2378                  */
2379                 min_tx_space = (adapter->max_frame_size +
2380                                 sizeof(struct e1000_tx_desc) -
2381                                 ETH_FCS_LEN) * 2;
2382                 min_tx_space = ALIGN(min_tx_space, 1024);
2383                 min_tx_space >>= 10;
2384                 /* software strips receive CRC, so leave room for it */
2385                 min_rx_space = adapter->max_frame_size;
2386                 min_rx_space = ALIGN(min_rx_space, 1024);
2387                 min_rx_space >>= 10;
2388
2389                 /*
2390                  * If current Tx allocation is less than the min Tx FIFO size,
2391                  * and the min Tx FIFO size is less than the current Rx FIFO
2392                  * allocation, take space away from current Rx allocation
2393                  */
2394                 if ((tx_space < min_tx_space) &&
2395                     ((min_tx_space - tx_space) < pba)) {
2396                         pba -= min_tx_space - tx_space;
2397
2398                         /*
2399                          * if short on Rx space, Rx wins and must trump tx
2400                          * adjustment or use Early Receive if available
2401                          */
2402                         if ((pba < min_rx_space) &&
2403                             (!(adapter->flags & FLAG_HAS_ERT)))
2404                                 /* ERT enabled in e1000_configure_rx */
2405                                 pba = min_rx_space;
2406                 }
2407
2408                 ew32(PBA, pba);
2409         }
2410
2411
2412         /*
2413          * flow control settings
2414          *
2415          * The high water mark must be low enough to fit one full frame
2416          * (or the size used for early receive) above it in the Rx FIFO.
2417          * Set it to the lower of:
2418          * - 90% of the Rx FIFO size, and
2419          * - the full Rx FIFO size minus the early receive size (for parts
2420          *   with ERT support assuming ERT set to E1000_ERT_2048), or
2421          * - the full Rx FIFO size minus one full frame
2422          */
2423         if (adapter->flags & FLAG_HAS_ERT)
2424                 hwm = min(((pba << 10) * 9 / 10),
2425                           ((pba << 10) - (E1000_ERT_2048 << 3)));
2426         else
2427                 hwm = min(((pba << 10) * 9 / 10),
2428                           ((pba << 10) - adapter->max_frame_size));
2429
2430         fc->high_water = hwm & 0xFFF8; /* 8-byte granularity */
2431         fc->low_water = fc->high_water - 8;
2432
2433         if (adapter->flags & FLAG_DISABLE_FC_PAUSE_TIME)
2434                 fc->pause_time = 0xFFFF;
2435         else
2436                 fc->pause_time = E1000_FC_PAUSE_TIME;
2437         fc->send_xon = 1;
2438         fc->type = fc->original_type;
2439
2440         /* Allow time for pending master requests to run */
2441         mac->ops.reset_hw(hw);
2442
2443         /*
2444          * For parts with AMT enabled, let the firmware know
2445          * that the network interface is in control
2446          */
2447         if (adapter->flags & FLAG_HAS_AMT)
2448                 e1000_get_hw_control(adapter);
2449
2450         ew32(WUC, 0);
2451
2452         if (mac->ops.init_hw(hw))
2453                 e_err("Hardware Error\n");
2454
2455         e1000_update_mng_vlan(adapter);
2456
2457         /* Enable h/w to recognize an 802.1Q VLAN Ethernet packet */
2458         ew32(VET, ETH_P_8021Q);
2459
2460         e1000e_reset_adaptive(hw);
2461         e1000_get_phy_info(hw);
2462
2463         if (!(adapter->flags & FLAG_SMART_POWER_DOWN)) {
2464                 u16 phy_data = 0;
2465                 /*
2466                  * speed up time to link by disabling smart power down, ignore
2467                  * the return value of this function because there is nothing
2468                  * different we would do if it failed
2469                  */
2470                 e1e_rphy(hw, IGP02E1000_PHY_POWER_MGMT, &phy_data);
2471                 phy_data &= ~IGP02E1000_PM_SPD;
2472                 e1e_wphy(hw, IGP02E1000_PHY_POWER_MGMT, phy_data);
2473         }
2474 }
2475
2476 int e1000e_up(struct e1000_adapter *adapter)
2477 {
2478         struct e1000_hw *hw = &adapter->hw;
2479
2480         /* hardware has been reset, we need to reload some things */
2481         e1000_configure(adapter);
2482
2483         clear_bit(__E1000_DOWN, &adapter->state);
2484
2485         napi_enable(&adapter->napi);
2486         e1000_irq_enable(adapter);
2487
2488         /* fire a link change interrupt to start the watchdog */
2489         ew32(ICS, E1000_ICS_LSC);
2490         return 0;
2491 }
2492
2493 void e1000e_down(struct e1000_adapter *adapter)
2494 {
2495         struct net_device *netdev = adapter->netdev;
2496         struct e1000_hw *hw = &adapter->hw;
2497         u32 tctl, rctl;
2498
2499         /*
2500          * signal that we're down so the interrupt handler does not
2501          * reschedule our watchdog timer
2502          */
2503         set_bit(__E1000_DOWN, &adapter->state);
2504
2505         /* disable receives in the hardware */
2506         rctl = er32(RCTL);
2507         ew32(RCTL, rctl & ~E1000_RCTL_EN);
2508         /* flush and sleep below */
2509
2510         netif_tx_stop_all_queues(netdev);
2511
2512         /* disable transmits in the hardware */
2513         tctl = er32(TCTL);
2514         tctl &= ~E1000_TCTL_EN;
2515         ew32(TCTL, tctl);
2516         /* flush both disables and wait for them to finish */
2517         e1e_flush();
2518         msleep(10);
2519
2520         napi_disable(&adapter->napi);
2521         e1000_irq_disable(adapter);
2522
2523         del_timer_sync(&adapter->watchdog_timer);
2524         del_timer_sync(&adapter->phy_info_timer);
2525
2526         netdev->tx_queue_len = adapter->tx_queue_len;
2527         netif_carrier_off(netdev);
2528         adapter->link_speed = 0;
2529         adapter->link_duplex = 0;
2530
2531         if (!pci_channel_offline(adapter->pdev))
2532                 e1000e_reset(adapter);
2533         e1000_clean_tx_ring(adapter);
2534         e1000_clean_rx_ring(adapter);
2535
2536         /*
2537          * TODO: for power management, we could drop the link and
2538          * pci_disable_device here.
2539          */
2540 }
2541
2542 void e1000e_reinit_locked(struct e1000_adapter *adapter)
2543 {
2544         might_sleep();
2545         while (test_and_set_bit(__E1000_RESETTING, &adapter->state))
2546                 msleep(1);
2547         e1000e_down(adapter);
2548         e1000e_up(adapter);
2549         clear_bit(__E1000_RESETTING, &adapter->state);
2550 }
2551
2552 /**
2553  * e1000_sw_init - Initialize general software structures (struct e1000_adapter)
2554  * @adapter: board private structure to initialize
2555  *
2556  * e1000_sw_init initializes the Adapter private data structure.
2557  * Fields are initialized based on PCI device information and
2558  * OS network device settings (MTU size).
2559  **/
2560 static int __devinit e1000_sw_init(struct e1000_adapter *adapter)
2561 {
2562         struct net_device *netdev = adapter->netdev;
2563
2564         adapter->rx_buffer_len = ETH_FRAME_LEN + VLAN_HLEN + ETH_FCS_LEN;
2565         adapter->rx_ps_bsize0 = 128;
2566         adapter->max_frame_size = netdev->mtu + ETH_HLEN + ETH_FCS_LEN;
2567         adapter->min_frame_size = ETH_ZLEN + ETH_FCS_LEN;
2568
2569         adapter->tx_ring = kzalloc(sizeof(struct e1000_ring), GFP_KERNEL);
2570         if (!adapter->tx_ring)
2571                 goto err;
2572
2573         adapter->rx_ring = kzalloc(sizeof(struct e1000_ring), GFP_KERNEL);
2574         if (!adapter->rx_ring)
2575                 goto err;
2576
2577         spin_lock_init(&adapter->tx_queue_lock);
2578
2579         /* Explicitly disable IRQ since the NIC can be in any state. */
2580         e1000_irq_disable(adapter);
2581
2582         spin_lock_init(&adapter->stats_lock);
2583
2584         set_bit(__E1000_DOWN, &adapter->state);
2585         return 0;
2586
2587 err:
2588         e_err("Unable to allocate memory for queues\n");
2589         kfree(adapter->rx_ring);
2590         kfree(adapter->tx_ring);
2591         return -ENOMEM;
2592 }
2593
2594 /**
2595  * e1000_open - Called when a network interface is made active
2596  * @netdev: network interface device structure
2597  *
2598  * Returns 0 on success, negative value on failure
2599  *
2600  * The open entry point is called when a network interface is made
2601  * active by the system (IFF_UP).  At this point all resources needed
2602  * for transmit and receive operations are allocated, the interrupt
2603  * handler is registered with the OS, the watchdog timer is started,
2604  * and the stack is notified that the interface is ready.
2605  **/
2606 static int e1000_open(struct net_device *netdev)
2607 {
2608         struct e1000_adapter *adapter = netdev_priv(netdev);
2609         struct e1000_hw *hw = &adapter->hw;
2610         int err;
2611
2612         /* disallow open during test */
2613         if (test_bit(__E1000_TESTING, &adapter->state))
2614                 return -EBUSY;
2615
2616         /* allocate transmit descriptors */
2617         err = e1000e_setup_tx_resources(adapter);
2618         if (err)
2619                 goto err_setup_tx;
2620
2621         /* allocate receive descriptors */
2622         err = e1000e_setup_rx_resources(adapter);
2623         if (err)
2624                 goto err_setup_rx;
2625
2626         e1000e_power_up_phy(adapter);
2627
2628         adapter->mng_vlan_id = E1000_MNG_VLAN_NONE;
2629         if ((adapter->hw.mng_cookie.status &
2630              E1000_MNG_DHCP_COOKIE_STATUS_VLAN))
2631                 e1000_update_mng_vlan(adapter);
2632
2633         /*
2634          * If AMT is enabled, let the firmware know that the network
2635          * interface is now open
2636          */
2637         if (adapter->flags & FLAG_HAS_AMT)
2638                 e1000_get_hw_control(adapter);
2639
2640         /*
2641          * before we allocate an interrupt, we must be ready to handle it.
2642          * Setting DEBUG_SHIRQ in the kernel makes it fire an interrupt
2643          * as soon as we call pci_request_irq, so we have to setup our
2644          * clean_rx handler before we do so.
2645          */
2646         e1000_configure(adapter);
2647
2648         err = e1000_request_irq(adapter);
2649         if (err)
2650                 goto err_req_irq;
2651
2652         /* From here on the code is the same as e1000e_up() */
2653         clear_bit(__E1000_DOWN, &adapter->state);
2654
2655         napi_enable(&adapter->napi);
2656
2657         e1000_irq_enable(adapter);
2658
2659         netif_tx_start_all_queues(netdev);
2660
2661         /* fire a link status change interrupt to start the watchdog */
2662         ew32(ICS, E1000_ICS_LSC);
2663
2664         return 0;
2665
2666 err_req_irq:
2667         e1000_release_hw_control(adapter);
2668         e1000_power_down_phy(adapter);
2669         e1000e_free_rx_resources(adapter);
2670 err_setup_rx:
2671         e1000e_free_tx_resources(adapter);
2672 err_setup_tx:
2673         e1000e_reset(adapter);
2674
2675         return err;
2676 }
2677
2678 /**
2679  * e1000_close - Disables a network interface
2680  * @netdev: network interface device structure
2681  *
2682  * Returns 0, this is not allowed to fail
2683  *
2684  * The close entry point is called when an interface is de-activated
2685  * by the OS.  The hardware is still under the drivers control, but
2686  * needs to be disabled.  A global MAC reset is issued to stop the
2687  * hardware, and all transmit and receive resources are freed.
2688  **/
2689 static int e1000_close(struct net_device *netdev)
2690 {
2691         struct e1000_adapter *adapter = netdev_priv(netdev);
2692
2693         WARN_ON(test_bit(__E1000_RESETTING, &adapter->state));
2694         e1000e_down(adapter);
2695         e1000_power_down_phy(adapter);
2696         e1000_free_irq(adapter);
2697
2698         e1000e_free_tx_resources(adapter);
2699         e1000e_free_rx_resources(adapter);
2700
2701         /*
2702          * kill manageability vlan ID if supported, but not if a vlan with
2703          * the same ID is registered on the host OS (let 8021q kill it)
2704          */
2705         if ((adapter->hw.mng_cookie.status &
2706                           E1000_MNG_DHCP_COOKIE_STATUS_VLAN) &&
2707              !(adapter->vlgrp &&
2708                vlan_group_get_device(adapter->vlgrp, adapter->mng_vlan_id)))
2709                 e1000_vlan_rx_kill_vid(netdev, adapter->mng_vlan_id);
2710
2711         /*
2712          * If AMT is enabled, let the firmware know that the network
2713          * interface is now closed
2714          */
2715         if (adapter->flags & FLAG_HAS_AMT)
2716                 e1000_release_hw_control(adapter);
2717
2718         return 0;
2719 }
2720 /**
2721  * e1000_set_mac - Change the Ethernet Address of the NIC
2722  * @netdev: network interface device structure
2723  * @p: pointer to an address structure
2724  *
2725  * Returns 0 on success, negative on failure
2726  **/
2727 static int e1000_set_mac(struct net_device *netdev, void *p)
2728 {
2729         struct e1000_adapter *adapter = netdev_priv(netdev);
2730         struct sockaddr *addr = p;
2731
2732         if (!is_valid_ether_addr(addr->sa_data))
2733                 return -EADDRNOTAVAIL;
2734
2735         memcpy(netdev->dev_addr, addr->sa_data, netdev->addr_len);
2736         memcpy(adapter->hw.mac.addr, addr->sa_data, netdev->addr_len);
2737
2738         e1000e_rar_set(&adapter->hw, adapter->hw.mac.addr, 0);
2739
2740         if (adapter->flags & FLAG_RESET_OVERWRITES_LAA) {
2741                 /* activate the work around */
2742                 e1000e_set_laa_state_82571(&adapter->hw, 1);
2743
2744                 /*
2745                  * Hold a copy of the LAA in RAR[14] This is done so that
2746                  * between the time RAR[0] gets clobbered  and the time it
2747                  * gets fixed (in e1000_watchdog), the actual LAA is in one
2748                  * of the RARs and no incoming packets directed to this port
2749                  * are dropped. Eventually the LAA will be in RAR[0] and
2750                  * RAR[14]
2751                  */
2752                 e1000e_rar_set(&adapter->hw,
2753                               adapter->hw.mac.addr,
2754                               adapter->hw.mac.rar_entry_count - 1);
2755         }
2756
2757         return 0;
2758 }
2759
2760 /*
2761  * Need to wait a few seconds after link up to get diagnostic information from
2762  * the phy
2763  */
2764 static void e1000_update_phy_info(unsigned long data)
2765 {
2766         struct e1000_adapter *adapter = (struct e1000_adapter *) data;
2767         e1000_get_phy_info(&adapter->hw);
2768 }
2769
2770 /**
2771  * e1000e_update_stats - Update the board statistics counters
2772  * @adapter: board private structure
2773  **/
2774 void e1000e_update_stats(struct e1000_adapter *adapter)
2775 {
2776         struct e1000_hw *hw = &adapter->hw;
2777         struct pci_dev *pdev = adapter->pdev;
2778         unsigned long irq_flags;
2779         u16 phy_tmp;
2780
2781 #define PHY_IDLE_ERROR_COUNT_MASK 0x00FF
2782
2783         /*
2784          * Prevent stats update while adapter is being reset, or if the pci
2785          * connection is down.
2786          */
2787         if (adapter->link_speed == 0)
2788                 return;
2789         if (pci_channel_offline(pdev))
2790                 return;
2791
2792         spin_lock_irqsave(&adapter->stats_lock, irq_flags);
2793
2794         /*
2795          * these counters are modified from e1000_adjust_tbi_stats,
2796          * called from the interrupt context, so they must only
2797          * be written while holding adapter->stats_lock
2798          */
2799
2800         adapter->stats.crcerrs += er32(CRCERRS);
2801         adapter->stats.gprc += er32(GPRC);
2802         adapter->stats.gorc += er32(GORCL);
2803         er32(GORCH); /* Clear gorc */
2804         adapter->stats.bprc += er32(BPRC);
2805         adapter->stats.mprc += er32(MPRC);
2806         adapter->stats.roc += er32(ROC);
2807
2808         adapter->stats.mpc += er32(MPC);
2809         adapter->stats.scc += er32(SCC);
2810         adapter->stats.ecol += er32(ECOL);
2811         adapter->stats.mcc += er32(MCC);
2812         adapter->stats.latecol += er32(LATECOL);
2813         adapter->stats.dc += er32(DC);
2814         adapter->stats.xonrxc += er32(XONRXC);
2815         adapter->stats.xontxc += er32(XONTXC);
2816         adapter->stats.xoffrxc += er32(XOFFRXC);
2817         adapter->stats.xofftxc += er32(XOFFTXC);
2818         adapter->stats.gptc += er32(GPTC);
2819         adapter->stats.gotc += er32(GOTCL);
2820         er32(GOTCH); /* Clear gotc */
2821         adapter->stats.rnbc += er32(RNBC);
2822         adapter->stats.ruc += er32(RUC);
2823
2824         adapter->stats.mptc += er32(MPTC);
2825         adapter->stats.bptc += er32(BPTC);
2826
2827         /* used for adaptive IFS */
2828
2829         hw->mac.tx_packet_delta = er32(TPT);
2830         adapter->stats.tpt += hw->mac.tx_packet_delta;
2831         hw->mac.collision_delta = er32(COLC);
2832         adapter->stats.colc += hw->mac.collision_delta;
2833
2834         adapter->stats.algnerrc += er32(ALGNERRC);
2835         adapter->stats.rxerrc += er32(RXERRC);
2836         adapter->stats.tncrs += er32(TNCRS);
2837         adapter->stats.cexterr += er32(CEXTERR);
2838         adapter->stats.tsctc += er32(TSCTC);
2839         adapter->stats.tsctfc += er32(TSCTFC);
2840
2841         /* Fill out the OS statistics structure */
2842         adapter->net_stats.multicast = adapter->stats.mprc;
2843         adapter->net_stats.collisions = adapter->stats.colc;
2844
2845         /* Rx Errors */
2846
2847         /*
2848          * RLEC on some newer hardware can be incorrect so build
2849          * our own version based on RUC and ROC
2850          */
2851         adapter->net_stats.rx_errors = adapter->stats.rxerrc +
2852                 adapter->stats.crcerrs + adapter->stats.algnerrc +
2853                 adapter->stats.ruc + adapter->stats.roc +
2854                 adapter->stats.cexterr;
2855         adapter->net_stats.rx_length_errors = adapter->stats.ruc +
2856                                               adapter->stats.roc;
2857         adapter->net_stats.rx_crc_errors = adapter->stats.crcerrs;
2858         adapter->net_stats.rx_frame_errors = adapter->stats.algnerrc;
2859         adapter->net_stats.rx_missed_errors = adapter->stats.mpc;
2860
2861         /* Tx Errors */
2862         adapter->net_stats.tx_errors = adapter->stats.ecol +
2863                                        adapter->stats.latecol;
2864         adapter->net_stats.tx_aborted_errors = adapter->stats.ecol;
2865         adapter->net_stats.tx_window_errors = adapter->stats.latecol;
2866         adapter->net_stats.tx_carrier_errors = adapter->stats.tncrs;
2867
2868         /* Tx Dropped needs to be maintained elsewhere */
2869
2870         /* Phy Stats */
2871         if (hw->phy.media_type == e1000_media_type_copper) {
2872                 if ((adapter->link_speed == SPEED_1000) &&
2873                    (!e1e_rphy(hw, PHY_1000T_STATUS, &phy_tmp))) {
2874                         phy_tmp &= PHY_IDLE_ERROR_COUNT_MASK;
2875                         adapter->phy_stats.idle_errors += phy_tmp;
2876                 }
2877         }
2878
2879         /* Management Stats */
2880         adapter->stats.mgptc += er32(MGTPTC);
2881         adapter->stats.mgprc += er32(MGTPRC);
2882         adapter->stats.mgpdc += er32(MGTPDC);
2883
2884         spin_unlock_irqrestore(&adapter->stats_lock, irq_flags);
2885 }
2886
2887 /**
2888  * e1000_phy_read_status - Update the PHY register status snapshot
2889  * @adapter: board private structure
2890  **/
2891 static void e1000_phy_read_status(struct e1000_adapter *adapter)
2892 {
2893         struct e1000_hw *hw = &adapter->hw;
2894         struct e1000_phy_regs *phy = &adapter->phy_regs;
2895         int ret_val;
2896         unsigned long irq_flags;
2897
2898
2899         spin_lock_irqsave(&adapter->stats_lock, irq_flags);
2900
2901         if ((er32(STATUS) & E1000_STATUS_LU) &&
2902             (adapter->hw.phy.media_type == e1000_media_type_copper)) {
2903                 ret_val  = e1e_rphy(hw, PHY_CONTROL, &phy->bmcr);
2904                 ret_val |= e1e_rphy(hw, PHY_STATUS, &phy->bmsr);
2905                 ret_val |= e1e_rphy(hw, PHY_AUTONEG_ADV, &phy->advertise);
2906                 ret_val |= e1e_rphy(hw, PHY_LP_ABILITY, &phy->lpa);
2907                 ret_val |= e1e_rphy(hw, PHY_AUTONEG_EXP, &phy->expansion);
2908                 ret_val |= e1e_rphy(hw, PHY_1000T_CTRL, &phy->ctrl1000);
2909                 ret_val |= e1e_rphy(hw, PHY_1000T_STATUS, &phy->stat1000);
2910                 ret_val |= e1e_rphy(hw, PHY_EXT_STATUS, &phy->estatus);
2911                 if (ret_val)
2912                         e_warn("Error reading PHY register\n");
2913         } else {
2914                 /*
2915                  * Do not read PHY registers if link is not up
2916                  * Set values to typical power-on defaults
2917                  */
2918                 phy->bmcr = (BMCR_SPEED1000 | BMCR_ANENABLE | BMCR_FULLDPLX);
2919                 phy->bmsr = (BMSR_100FULL | BMSR_100HALF | BMSR_10FULL |
2920                              BMSR_10HALF | BMSR_ESTATEN | BMSR_ANEGCAPABLE |
2921                              BMSR_ERCAP);
2922                 phy->advertise = (ADVERTISE_PAUSE_ASYM | ADVERTISE_PAUSE_CAP |
2923                                   ADVERTISE_ALL | ADVERTISE_CSMA);
2924                 phy->lpa = 0;
2925                 phy->expansion = EXPANSION_ENABLENPAGE;
2926                 phy->ctrl1000 = ADVERTISE_1000FULL;
2927                 phy->stat1000 = 0;
2928                 phy->estatus = (ESTATUS_1000_TFULL | ESTATUS_1000_THALF);
2929         }
2930
2931         spin_unlock_irqrestore(&adapter->stats_lock, irq_flags);
2932 }
2933
2934 static void e1000_print_link_info(struct e1000_adapter *adapter)
2935 {
2936         struct e1000_hw *hw = &adapter->hw;
2937         u32 ctrl = er32(CTRL);
2938
2939         e_info("Link is Up %d Mbps %s, Flow Control: %s\n",
2940                adapter->link_speed,
2941                (adapter->link_duplex == FULL_DUPLEX) ?
2942                                 "Full Duplex" : "Half Duplex",
2943                ((ctrl & E1000_CTRL_TFCE) && (ctrl & E1000_CTRL_RFCE)) ?
2944                                 "RX/TX" :
2945                ((ctrl & E1000_CTRL_RFCE) ? "RX" :
2946                ((ctrl & E1000_CTRL_TFCE) ? "TX" : "None" )));
2947 }
2948
2949 static bool e1000_has_link(struct e1000_adapter *adapter)
2950 {
2951         struct e1000_hw *hw = &adapter->hw;
2952         bool link_active = 0;
2953         s32 ret_val = 0;
2954
2955         /*
2956          * get_link_status is set on LSC (link status) interrupt or
2957          * Rx sequence error interrupt.  get_link_status will stay
2958          * false until the check_for_link establishes link
2959          * for copper adapters ONLY
2960          */
2961         switch (hw->phy.media_type) {
2962         case e1000_media_type_copper:
2963                 if (hw->mac.get_link_status) {
2964                         ret_val = hw->mac.ops.check_for_link(hw);
2965                         link_active = !hw->mac.get_link_status;
2966                 } else {
2967                         link_active = 1;
2968                 }
2969                 break;
2970         case e1000_media_type_fiber:
2971                 ret_val = hw->mac.ops.check_for_link(hw);
2972                 link_active = !!(er32(STATUS) & E1000_STATUS_LU);
2973                 break;
2974         case e1000_media_type_internal_serdes:
2975                 ret_val = hw->mac.ops.check_for_link(hw);
2976                 link_active = adapter->hw.mac.serdes_has_link;
2977                 break;
2978         default:
2979         case e1000_media_type_unknown:
2980                 break;
2981         }
2982
2983         if ((ret_val == E1000_ERR_PHY) && (hw->phy.type == e1000_phy_igp_3) &&
2984             (er32(CTRL) & E1000_PHY_CTRL_GBE_DISABLE)) {
2985                 /* See e1000_kmrn_lock_loss_workaround_ich8lan() */
2986                 e_info("Gigabit has been disabled, downgrading speed\n");
2987         }
2988
2989         return link_active;
2990 }
2991
2992 static void e1000e_enable_receives(struct e1000_adapter *adapter)
2993 {
2994         /* make sure the receive unit is started */
2995         if ((adapter->flags & FLAG_RX_NEEDS_RESTART) &&
2996             (adapter->flags & FLAG_RX_RESTART_NOW)) {
2997                 struct e1000_hw *hw = &adapter->hw;
2998                 u32 rctl = er32(RCTL);
2999                 ew32(RCTL, rctl | E1000_RCTL_EN);
3000                 adapter->flags &= ~FLAG_RX_RESTART_NOW;
3001         }
3002 }
3003
3004 /**
3005  * e1000_watchdog - Timer Call-back
3006  * @data: pointer to adapter cast into an unsigned long
3007  **/
3008 static void e1000_watchdog(unsigned long data)
3009 {
3010         struct e1000_adapter *adapter = (struct e1000_adapter *) data;
3011
3012         /* Do the rest outside of interrupt context */
3013         schedule_work(&adapter->watchdog_task);
3014
3015         /* TODO: make this use queue_delayed_work() */
3016 }
3017
3018 static void e1000_watchdog_task(struct work_struct *work)
3019 {
3020         struct e1000_adapter *adapter = container_of(work,
3021                                         struct e1000_adapter, watchdog_task);
3022         struct net_device *netdev = adapter->netdev;
3023         struct e1000_mac_info *mac = &adapter->hw.mac;
3024         struct e1000_ring *tx_ring = adapter->tx_ring;
3025         struct e1000_hw *hw = &adapter->hw;
3026         u32 link, tctl;
3027         int tx_pending = 0;
3028
3029         link = e1000_has_link(adapter);
3030         if ((netif_carrier_ok(netdev)) && link) {
3031                 e1000e_enable_receives(adapter);
3032                 goto link_up;
3033         }
3034
3035         if ((e1000e_enable_tx_pkt_filtering(hw)) &&
3036             (adapter->mng_vlan_id != adapter->hw.mng_cookie.vlan_id))
3037                 e1000_update_mng_vlan(adapter);
3038
3039         if (link) {
3040                 if (!netif_carrier_ok(netdev)) {
3041                         bool txb2b = 1;
3042                         /* update snapshot of PHY registers on LSC */
3043                         e1000_phy_read_status(adapter);
3044                         mac->ops.get_link_up_info(&adapter->hw,
3045                                                    &adapter->link_speed,
3046                                                    &adapter->link_duplex);
3047                         e1000_print_link_info(adapter);
3048                         /*
3049                          * tweak tx_queue_len according to speed/duplex
3050                          * and adjust the timeout factor
3051                          */
3052                         netdev->tx_queue_len = adapter->tx_queue_len;
3053                         adapter->tx_timeout_factor = 1;
3054                         switch (adapter->link_speed) {
3055                         case SPEED_10:
3056                                 txb2b = 0;
3057                                 netdev->tx_queue_len = 10;
3058                                 adapter->tx_timeout_factor = 14;
3059                                 break;
3060                         case SPEED_100:
3061                                 txb2b = 0;
3062                                 netdev->tx_queue_len = 100;
3063                                 /* maybe add some timeout factor ? */
3064                                 break;
3065                         }
3066
3067                         /*
3068                          * workaround: re-program speed mode bit after
3069                          * link-up event
3070                          */
3071                         if ((adapter->flags & FLAG_TARC_SPEED_MODE_BIT) &&
3072                             !txb2b) {
3073                                 u32 tarc0;
3074                                 tarc0 = er32(TARC(0));
3075                                 tarc0 &= ~SPEED_MODE_BIT;
3076                                 ew32(TARC(0), tarc0);
3077                         }
3078
3079                         /*
3080                          * disable TSO for pcie and 10/100 speeds, to avoid
3081                          * some hardware issues
3082                          */
3083                         if (!(adapter->flags & FLAG_TSO_FORCE)) {
3084                                 switch (adapter->link_speed) {
3085                                 case SPEED_10:
3086                                 case SPEED_100:
3087                                         e_info("10/100 speed: disabling TSO\n");
3088                                         netdev->features &= ~NETIF_F_TSO;
3089                                         netdev->features &= ~NETIF_F_TSO6;
3090                                         break;
3091                                 case SPEED_1000:
3092                                         netdev->features |= NETIF_F_TSO;
3093                                         netdev->features |= NETIF_F_TSO6;
3094                                         break;
3095                                 default:
3096                                         /* oops */
3097                                         break;
3098                                 }
3099                         }
3100
3101                         /*
3102                          * enable transmits in the hardware, need to do this
3103                          * after setting TARC(0)
3104                          */
3105                         tctl = er32(TCTL);
3106                         tctl |= E1000_TCTL_EN;
3107                         ew32(TCTL, tctl);
3108
3109                         netif_carrier_on(netdev);
3110                         netif_tx_wake_all_queues(netdev);
3111
3112                         if (!test_bit(__E1000_DOWN, &adapter->state))
3113                                 mod_timer(&adapter->phy_info_timer,
3114                                           round_jiffies(jiffies + 2 * HZ));
3115                 }
3116         } else {
3117                 if (netif_carrier_ok(netdev)) {
3118                         adapter->link_speed = 0;
3119                         adapter->link_duplex = 0;
3120                         e_info("Link is Down\n");
3121                         netif_carrier_off(netdev);
3122                         netif_tx_stop_all_queues(netdev);
3123                         if (!test_bit(__E1000_DOWN, &adapter->state))
3124                                 mod_timer(&adapter->phy_info_timer,
3125                                           round_jiffies(jiffies + 2 * HZ));
3126
3127                         if (adapter->flags & FLAG_RX_NEEDS_RESTART)
3128                                 schedule_work(&adapter->reset_task);
3129                 }
3130         }
3131
3132 link_up:
3133         e1000e_update_stats(adapter);
3134
3135         mac->tx_packet_delta = adapter->stats.tpt - adapter->tpt_old;
3136         adapter->tpt_old = adapter->stats.tpt;
3137         mac->collision_delta = adapter->stats.colc - adapter->colc_old;
3138         adapter->colc_old = adapter->stats.colc;
3139
3140         adapter->gorc = adapter->stats.gorc - adapter->gorc_old;
3141         adapter->gorc_old = adapter->stats.gorc;
3142         adapter->gotc = adapter->stats.gotc - adapter->gotc_old;
3143         adapter->gotc_old = adapter->stats.gotc;
3144
3145         e1000e_update_adaptive(&adapter->hw);
3146
3147         if (!netif_carrier_ok(netdev)) {
3148                 tx_pending = (e1000_desc_unused(tx_ring) + 1 <
3149                                tx_ring->count);
3150                 if (tx_pending) {
3151                         /*
3152                          * We've lost link, so the controller stops DMA,
3153                          * but we've got queued Tx work that's never going
3154                          * to get done, so reset controller to flush Tx.
3155                          * (Do the reset outside of interrupt context).
3156                          */
3157                         adapter->tx_timeout_count++;
3158                         schedule_work(&adapter->reset_task);
3159                 }
3160         }
3161
3162         /* Cause software interrupt to ensure Rx ring is cleaned */
3163         ew32(ICS, E1000_ICS_RXDMT0);
3164
3165         /* Force detection of hung controller every watchdog period */
3166         adapter->detect_tx_hung = 1;
3167
3168         /*
3169          * With 82571 controllers, LAA may be overwritten due to controller
3170          * reset from the other port. Set the appropriate LAA in RAR[0]
3171          */
3172         if (e1000e_get_laa_state_82571(hw))
3173                 e1000e_rar_set(hw, adapter->hw.mac.addr, 0);
3174
3175         /* Reset the timer */
3176         if (!test_bit(__E1000_DOWN, &adapter->state))
3177                 mod_timer(&adapter->watchdog_timer,
3178                           round_jiffies(jiffies + 2 * HZ));
3179 }
3180
3181 #define E1000_TX_FLAGS_CSUM             0x00000001
3182 #define E1000_TX_FLAGS_VLAN             0x00000002
3183 #define E1000_TX_FLAGS_TSO              0x00000004
3184 #define E1000_TX_FLAGS_IPV4             0x00000008
3185 #define E1000_TX_FLAGS_VLAN_MASK        0xffff0000
3186 #define E1000_TX_FLAGS_VLAN_SHIFT       16
3187
3188 static int e1000_tso(struct e1000_adapter *adapter,
3189                      struct sk_buff *skb)
3190 {
3191         struct e1000_ring *tx_ring = adapter->tx_ring;
3192         struct e1000_context_desc *context_desc;
3193         struct e1000_buffer *buffer_info;
3194         unsigned int i;
3195         u32 cmd_length = 0;
3196         u16 ipcse = 0, tucse, mss;
3197         u8 ipcss, ipcso, tucss, tucso, hdr_len;
3198         int err;
3199
3200         if (skb_is_gso(skb)) {
3201                 if (skb_header_cloned(skb)) {
3202                         err = pskb_expand_head(skb, 0, 0, GFP_ATOMIC);
3203                         if (err)
3204                                 return err;
3205                 }
3206
3207                 hdr_len = skb_transport_offset(skb) + tcp_hdrlen(skb);
3208                 mss = skb_shinfo(skb)->gso_size;
3209                 if (skb->protocol == htons(ETH_P_IP)) {
3210                         struct iphdr *iph = ip_hdr(skb);
3211                         iph->tot_len = 0;
3212                         iph->check = 0;
3213                         tcp_hdr(skb)->check = ~csum_tcpudp_magic(iph->saddr,
3214                                                                  iph->daddr, 0,
3215                                                                  IPPROTO_TCP,
3216                                                                  0);
3217                         cmd_length = E1000_TXD_CMD_IP;
3218                         ipcse = skb_transport_offset(skb) - 1;
3219                 } else if (skb_shinfo(skb)->gso_type == SKB_GSO_TCPV6) {
3220                         ipv6_hdr(skb)->payload_len = 0;
3221                         tcp_hdr(skb)->check =
3222                                 ~csum_ipv6_magic(&ipv6_hdr(skb)->saddr,
3223                                                  &ipv6_hdr(skb)->daddr,
3224                                                  0, IPPROTO_TCP, 0);
3225                         ipcse = 0;
3226                 }
3227                 ipcss = skb_network_offset(skb);
3228                 ipcso = (void *)&(ip_hdr(skb)->check) - (void *)skb->data;
3229                 tucss = skb_transport_offset(skb);
3230                 tucso = (void *)&(tcp_hdr(skb)->check) - (void *)skb->data;
3231                 tucse = 0;
3232
3233                 cmd_length |= (E1000_TXD_CMD_DEXT | E1000_TXD_CMD_TSE |
3234                                E1000_TXD_CMD_TCP | (skb->len - (hdr_len)));
3235
3236                 i = tx_ring->next_to_use;
3237                 context_desc = E1000_CONTEXT_DESC(*tx_ring, i);
3238                 buffer_info = &tx_ring->buffer_info[i];
3239
3240                 context_desc->lower_setup.ip_fields.ipcss  = ipcss;
3241                 context_desc->lower_setup.ip_fields.ipcso  = ipcso;
3242                 context_desc->lower_setup.ip_fields.ipcse  = cpu_to_le16(ipcse);
3243                 context_desc->upper_setup.tcp_fields.tucss = tucss;
3244                 context_desc->upper_setup.tcp_fields.tucso = tucso;
3245                 context_desc->upper_setup.tcp_fields.tucse = cpu_to_le16(tucse);
3246                 context_desc->tcp_seg_setup.fields.mss     = cpu_to_le16(mss);
3247                 context_desc->tcp_seg_setup.fields.hdr_len = hdr_len;
3248                 context_desc->cmd_and_length = cpu_to_le32(cmd_length);
3249
3250                 buffer_info->time_stamp = jiffies;
3251                 buffer_info->next_to_watch = i;
3252
3253                 i++;
3254                 if (i == tx_ring->count)
3255                         i = 0;
3256                 tx_ring->next_to_use = i;
3257
3258                 return 1;
3259         }
3260
3261         return 0;
3262 }
3263
3264 static bool e1000_tx_csum(struct e1000_adapter *adapter, struct sk_buff *skb)
3265 {
3266         struct e1000_ring *tx_ring = adapter->tx_ring;
3267         struct e1000_context_desc *context_desc;
3268         struct e1000_buffer *buffer_info;
3269         unsigned int i;
3270         u8 css;
3271
3272         if (skb->ip_summed == CHECKSUM_PARTIAL) {
3273                 css = skb_transport_offset(skb);
3274
3275                 i = tx_ring->next_to_use;
3276                 buffer_info = &tx_ring->buffer_info[i];
3277                 context_desc = E1000_CONTEXT_DESC(*tx_ring, i);
3278
3279                 context_desc->lower_setup.ip_config = 0;
3280                 context_desc->upper_setup.tcp_fields.tucss = css;
3281                 context_desc->upper_setup.tcp_fields.tucso =
3282                                         css + skb->csum_offset;
3283                 context_desc->upper_setup.tcp_fields.tucse = 0;
3284                 context_desc->tcp_seg_setup.data = 0;
3285                 context_desc->cmd_and_length = cpu_to_le32(E1000_TXD_CMD_DEXT);
3286
3287                 buffer_info->time_stamp = jiffies;
3288                 buffer_info->next_to_watch = i;
3289
3290                 i++;
3291                 if (i == tx_ring->count)
3292                         i = 0;
3293                 tx_ring->next_to_use = i;
3294
3295                 return 1;
3296         }
3297
3298         return 0;
3299 }
3300
3301 #define E1000_MAX_PER_TXD       8192
3302 #define E1000_MAX_TXD_PWR       12
3303
3304 static int e1000_tx_map(struct e1000_adapter *adapter,
3305                         struct sk_buff *skb, unsigned int first,
3306                         unsigned int max_per_txd, unsigned int nr_frags,
3307                         unsigned int mss)
3308 {
3309         struct e1000_ring *tx_ring = adapter->tx_ring;
3310         struct e1000_buffer *buffer_info;
3311         unsigned int len = skb->len - skb->data_len;
3312         unsigned int offset = 0, size, count = 0, i;
3313         unsigned int f;
3314
3315         i = tx_ring->next_to_use;
3316
3317         while (len) {
3318                 buffer_info = &tx_ring->buffer_info[i];
3319                 size = min(len, max_per_txd);
3320
3321                 /* Workaround for premature desc write-backs
3322                  * in TSO mode.  Append 4-byte sentinel desc */
3323                 if (mss && !nr_frags && size == len && size > 8)
3324                         size -= 4;
3325
3326                 buffer_info->length = size;
3327                 /* set time_stamp *before* dma to help avoid a possible race */
3328                 buffer_info->time_stamp = jiffies;
3329                 buffer_info->dma =
3330                         pci_map_single(adapter->pdev,
3331                                 skb->data + offset,
3332                                 size,
3333                                 PCI_DMA_TODEVICE);
3334                 if (pci_dma_mapping_error(adapter->pdev, buffer_info->dma)) {
3335                         dev_err(&adapter->pdev->dev, "TX DMA map failed\n");
3336                         adapter->tx_dma_failed++;
3337                         return -1;
3338                 }
3339                 buffer_info->next_to_watch = i;
3340
3341                 len -= size;
3342                 offset += size;
3343                 count++;
3344                 i++;
3345                 if (i == tx_ring->count)
3346                         i = 0;
3347         }
3348
3349         for (f = 0; f < nr_frags; f++) {
3350                 struct skb_frag_struct *frag;
3351
3352                 frag = &skb_shinfo(skb)->frags[f];
3353                 len = frag->size;
3354                 offset = frag->page_offset;
3355
3356                 while (len) {
3357                         buffer_info = &tx_ring->buffer_info[i];
3358                         size = min(len, max_per_txd);
3359                         /* Workaround for premature desc write-backs
3360                          * in TSO mode.  Append 4-byte sentinel desc */
3361                         if (mss && f == (nr_frags-1) && size == len && size > 8)
3362                                 size -= 4;
3363
3364                         buffer_info->length = size;
3365                         buffer_info->time_stamp = jiffies;
3366                         buffer_info->dma =
3367                                 pci_map_page(adapter->pdev,
3368                                         frag->page,
3369                                         offset,
3370                                         size,
3371                                         PCI_DMA_TODEVICE);
3372                         if (pci_dma_mapping_error(adapter->pdev,
3373                                                   buffer_info->dma)) {
3374                                 dev_err(&adapter->pdev->dev,
3375                                         "TX DMA page map failed\n");
3376                                 adapter->tx_dma_failed++;
3377                                 return -1;
3378                         }
3379
3380                         buffer_info->next_to_watch = i;
3381
3382                         len -= size;
3383                         offset += size;
3384                         count++;
3385
3386                         i++;
3387                         if (i == tx_ring->count)
3388                                 i = 0;
3389                 }
3390         }
3391
3392         if (i == 0)
3393                 i = tx_ring->count - 1;
3394         else
3395                 i--;
3396
3397         tx_ring->buffer_info[i].skb = skb;
3398         tx_ring->buffer_info[first].next_to_watch = i;
3399
3400         return count;
3401 }
3402
3403 static void e1000_tx_queue(struct e1000_adapter *adapter,
3404                            int tx_flags, int count)
3405 {
3406         struct e1000_ring *tx_ring = adapter->tx_ring;
3407         struct e1000_tx_desc *tx_desc = NULL;
3408         struct e1000_buffer *buffer_info;
3409         u32 txd_upper = 0, txd_lower = E1000_TXD_CMD_IFCS;
3410         unsigned int i;
3411
3412         if (tx_flags & E1000_TX_FLAGS_TSO) {
3413                 txd_lower |= E1000_TXD_CMD_DEXT | E1000_TXD_DTYP_D |
3414                              E1000_TXD_CMD_TSE;
3415                 txd_upper |= E1000_TXD_POPTS_TXSM << 8;
3416
3417                 if (tx_flags & E1000_TX_FLAGS_IPV4)
3418                         txd_upper |= E1000_TXD_POPTS_IXSM << 8;
3419         }
3420
3421         if (tx_flags & E1000_TX_FLAGS_CSUM) {
3422                 txd_lower |= E1000_TXD_CMD_DEXT | E1000_TXD_DTYP_D;
3423                 txd_upper |= E1000_TXD_POPTS_TXSM << 8;
3424         }
3425
3426         if (tx_flags & E1000_TX_FLAGS_VLAN) {
3427                 txd_lower |= E1000_TXD_CMD_VLE;
3428                 txd_upper |= (tx_flags & E1000_TX_FLAGS_VLAN_MASK);
3429         }
3430
3431         i = tx_ring->next_to_use;
3432
3433         while (count--) {
3434                 buffer_info = &tx_ring->buffer_info[i];
3435                 tx_desc = E1000_TX_DESC(*tx_ring, i);
3436                 tx_desc->buffer_addr = cpu_to_le64(buffer_info->dma);
3437                 tx_desc->lower.data =
3438                         cpu_to_le32(txd_lower | buffer_info->length);
3439                 tx_desc->upper.data = cpu_to_le32(txd_upper);
3440
3441                 i++;
3442                 if (i == tx_ring->count)
3443                         i = 0;
3444         }
3445
3446         tx_desc->lower.data |= cpu_to_le32(adapter->txd_cmd);
3447
3448         /*
3449          * Force memory writes to complete before letting h/w
3450          * know there are new descriptors to fetch.  (Only
3451          * applicable for weak-ordered memory model archs,
3452          * such as IA-64).
3453          */
3454         wmb();
3455
3456         tx_ring->next_to_use = i;
3457         writel(i, adapter->hw.hw_addr + tx_ring->tail);
3458         /*
3459          * we need this if more than one processor can write to our tail
3460          * at a time, it synchronizes IO on IA64/Altix systems
3461          */
3462         mmiowb();
3463 }
3464
3465 #define MINIMUM_DHCP_PACKET_SIZE 282
3466 static int e1000_transfer_dhcp_info(struct e1000_adapter *adapter,
3467                                     struct sk_buff *skb)
3468 {
3469         struct e1000_hw *hw =  &adapter->hw;
3470         u16 length, offset;
3471
3472         if (vlan_tx_tag_present(skb)) {
3473                 if (!((vlan_tx_tag_get(skb) == adapter->hw.mng_cookie.vlan_id)
3474                     && (adapter->hw.mng_cookie.status &
3475                         E1000_MNG_DHCP_COOKIE_STATUS_VLAN)))
3476                         return 0;
3477         }
3478
3479         if (skb->len <= MINIMUM_DHCP_PACKET_SIZE)
3480                 return 0;
3481
3482         if (((struct ethhdr *) skb->data)->h_proto != htons(ETH_P_IP))
3483                 return 0;
3484
3485         {
3486                 const struct iphdr *ip = (struct iphdr *)((u8 *)skb->data+14);
3487                 struct udphdr *udp;
3488
3489                 if (ip->protocol != IPPROTO_UDP)
3490                         return 0;
3491
3492                 udp = (struct udphdr *)((u8 *)ip + (ip->ihl << 2));
3493                 if (ntohs(udp->dest) != 67)
3494                         return 0;
3495
3496                 offset = (u8 *)udp + 8 - skb->data;
3497                 length = skb->len - offset;
3498                 return e1000e_mng_write_dhcp_info(hw, (u8 *)udp + 8, length);
3499         }
3500
3501         return 0;
3502 }
3503
3504 static int __e1000_maybe_stop_tx(struct net_device *netdev, int size)
3505 {
3506         struct e1000_adapter *adapter = netdev_priv(netdev);
3507
3508         netif_stop_queue(netdev);
3509         /*
3510          * Herbert's original patch had:
3511          *  smp_mb__after_netif_stop_queue();
3512          * but since that doesn't exist yet, just open code it.
3513          */
3514         smp_mb();
3515
3516         /*
3517          * We need to check again in a case another CPU has just
3518          * made room available.
3519          */
3520         if (e1000_desc_unused(adapter->tx_ring) < size)
3521                 return -EBUSY;
3522
3523         /* A reprieve! */
3524         netif_start_queue(netdev);
3525         ++adapter->restart_queue;
3526         return 0;
3527 }
3528
3529 static int e1000_maybe_stop_tx(struct net_device *netdev, int size)
3530 {
3531         struct e1000_adapter *adapter = netdev_priv(netdev);
3532
3533         if (e1000_desc_unused(adapter->tx_ring) >= size)
3534                 return 0;
3535         return __e1000_maybe_stop_tx(netdev, size);
3536 }
3537
3538 #define TXD_USE_COUNT(S, X) (((S) >> (X)) + 1 )
3539 static int e1000_xmit_frame(struct sk_buff *skb, struct net_device *netdev)
3540 {
3541         struct e1000_adapter *adapter = netdev_priv(netdev);
3542         struct e1000_ring *tx_ring = adapter->tx_ring;
3543         unsigned int first;
3544         unsigned int max_per_txd = E1000_MAX_PER_TXD;
3545         unsigned int max_txd_pwr = E1000_MAX_TXD_PWR;
3546         unsigned int tx_flags = 0;
3547         unsigned int len = skb->len - skb->data_len;
3548         unsigned long irq_flags;
3549         unsigned int nr_frags;
3550         unsigned int mss;
3551         int count = 0;
3552         int tso;
3553         unsigned int f;
3554
3555         if (test_bit(__E1000_DOWN, &adapter->state)) {
3556                 dev_kfree_skb_any(skb);
3557                 return NETDEV_TX_OK;
3558         }
3559
3560         if (skb->len <= 0) {
3561                 dev_kfree_skb_any(skb);
3562                 return NETDEV_TX_OK;
3563         }
3564
3565         mss = skb_shinfo(skb)->gso_size;
3566         /*
3567          * The controller does a simple calculation to
3568          * make sure there is enough room in the FIFO before
3569          * initiating the DMA for each buffer.  The calc is:
3570          * 4 = ceil(buffer len/mss).  To make sure we don't
3571          * overrun the FIFO, adjust the max buffer len if mss
3572          * drops.
3573          */
3574         if (mss) {
3575                 u8 hdr_len;
3576                 max_per_txd = min(mss << 2, max_per_txd);
3577                 max_txd_pwr = fls(max_per_txd) - 1;
3578
3579                 /*
3580                  * TSO Workaround for 82571/2/3 Controllers -- if skb->data
3581                  * points to just header, pull a few bytes of payload from
3582                  * frags into skb->data
3583                  */
3584                 hdr_len = skb_transport_offset(skb) + tcp_hdrlen(skb);
3585                 /*
3586                  * we do this workaround for ES2LAN, but it is un-necessary,
3587                  * avoiding it could save a lot of cycles
3588                  */
3589                 if (skb->data_len && (hdr_len == len)) {
3590                         unsigned int pull_size;
3591
3592                         pull_size = min((unsigned int)4, skb->data_len);
3593                         if (!__pskb_pull_tail(skb, pull_size)) {
3594                                 e_err("__pskb_pull_tail failed.\n");
3595                                 dev_kfree_skb_any(skb);
3596                                 return NETDEV_TX_OK;
3597                         }
3598                         len = skb->len - skb->data_len;
3599                 }
3600         }
3601
3602         /* reserve a descriptor for the offload context */
3603         if ((mss) || (skb->ip_summed == CHECKSUM_PARTIAL))
3604                 count++;
3605         count++;
3606
3607         count += TXD_USE_COUNT(len, max_txd_pwr);
3608
3609         nr_frags = skb_shinfo(skb)->nr_frags;
3610         for (f = 0; f < nr_frags; f++)
3611                 count += TXD_USE_COUNT(skb_shinfo(skb)->frags[f].size,
3612                                        max_txd_pwr);
3613
3614         if (adapter->hw.mac.tx_pkt_filtering)
3615                 e1000_transfer_dhcp_info(adapter, skb);
3616
3617         if (!spin_trylock_irqsave(&adapter->tx_queue_lock, irq_flags))
3618                 /* Collision - tell upper layer to requeue */
3619                 return NETDEV_TX_LOCKED;
3620
3621         /*
3622          * need: count + 2 desc gap to keep tail from touching
3623          * head, otherwise try next time
3624          */
3625         if (e1000_maybe_stop_tx(netdev, count + 2)) {
3626                 spin_unlock_irqrestore(&adapter->tx_queue_lock, irq_flags);
3627                 return NETDEV_TX_BUSY;
3628         }
3629
3630         if (adapter->vlgrp && vlan_tx_tag_present(skb)) {
3631                 tx_flags |= E1000_TX_FLAGS_VLAN;
3632                 tx_flags |= (vlan_tx_tag_get(skb) << E1000_TX_FLAGS_VLAN_SHIFT);
3633         }
3634
3635         first = tx_ring->next_to_use;
3636
3637         tso = e1000_tso(adapter, skb);
3638         if (tso < 0) {
3639                 dev_kfree_skb_any(skb);
3640                 spin_unlock_irqrestore(&adapter->tx_queue_lock, irq_flags);
3641                 return NETDEV_TX_OK;
3642         }
3643
3644         if (tso)
3645                 tx_flags |= E1000_TX_FLAGS_TSO;
3646         else if (e1000_tx_csum(adapter, skb))
3647                 tx_flags |= E1000_TX_FLAGS_CSUM;
3648
3649         /*
3650          * Old method was to assume IPv4 packet by default if TSO was enabled.
3651          * 82571 hardware supports TSO capabilities for IPv6 as well...
3652          * no longer assume, we must.
3653          */
3654         if (skb->protocol == htons(ETH_P_IP))
3655                 tx_flags |= E1000_TX_FLAGS_IPV4;
3656
3657         count = e1000_tx_map(adapter, skb, first, max_per_txd, nr_frags, mss);
3658         if (count < 0) {
3659                 /* handle pci_map_single() error in e1000_tx_map */
3660                 dev_kfree_skb_any(skb);
3661                 spin_unlock_irqrestore(&adapter->tx_queue_lock, irq_flags);
3662                 return NETDEV_TX_OK;
3663         }
3664
3665         e1000_tx_queue(adapter, tx_flags, count);
3666
3667         netdev->trans_start = jiffies;
3668
3669         /* Make sure there is space in the ring for the next send. */
3670         e1000_maybe_stop_tx(netdev, MAX_SKB_FRAGS + 2);
3671
3672         spin_unlock_irqrestore(&adapter->tx_queue_lock, irq_flags);
3673         return NETDEV_TX_OK;
3674 }
3675
3676 /**
3677  * e1000_tx_timeout - Respond to a Tx Hang
3678  * @netdev: network interface device structure
3679  **/
3680 static void e1000_tx_timeout(struct net_device *netdev)
3681 {
3682         struct e1000_adapter *adapter = netdev_priv(netdev);
3683
3684         /* Do the reset outside of interrupt context */
3685         adapter->tx_timeout_count++;
3686         schedule_work(&adapter->reset_task);
3687 }
3688
3689 static void e1000_reset_task(struct work_struct *work)
3690 {
3691         struct e1000_adapter *adapter;
3692         adapter = container_of(work, struct e1000_adapter, reset_task);
3693
3694         e1000e_reinit_locked(adapter);
3695 }
3696
3697 /**
3698  * e1000_get_stats - Get System Network Statistics
3699  * @netdev: network interface device structure
3700  *
3701  * Returns the address of the device statistics structure.
3702  * The statistics are actually updated from the timer callback.
3703  **/
3704 static struct net_device_stats *e1000_get_stats(struct net_device *netdev)
3705 {
3706         struct e1000_adapter *adapter = netdev_priv(netdev);
3707
3708         /* only return the current stats */
3709         return &adapter->net_stats;
3710 }
3711
3712 /**
3713  * e1000_change_mtu - Change the Maximum Transfer Unit
3714  * @netdev: network interface device structure
3715  * @new_mtu: new value for maximum frame size
3716  *
3717  * Returns 0 on success, negative on failure
3718  **/
3719 static int e1000_change_mtu(struct net_device *netdev, int new_mtu)
3720 {
3721         struct e1000_adapter *adapter = netdev_priv(netdev);
3722         int max_frame = new_mtu + ETH_HLEN + ETH_FCS_LEN;
3723
3724         if ((max_frame < ETH_ZLEN + ETH_FCS_LEN) ||
3725             (max_frame > MAX_JUMBO_FRAME_SIZE)) {
3726                 e_err("Invalid MTU setting\n");
3727                 return -EINVAL;
3728         }
3729
3730         /* Jumbo frame size limits */
3731         if (max_frame > ETH_FRAME_LEN + ETH_FCS_LEN) {
3732                 if (!(adapter->flags & FLAG_HAS_JUMBO_FRAMES)) {
3733                         e_err("Jumbo Frames not supported.\n");
3734                         return -EINVAL;
3735                 }
3736                 if (adapter->hw.phy.type == e1000_phy_ife) {
3737                         e_err("Jumbo Frames not supported.\n");
3738                         return -EINVAL;
3739                 }
3740         }
3741
3742 #define MAX_STD_JUMBO_FRAME_SIZE 9234
3743         if (max_frame > MAX_STD_JUMBO_FRAME_SIZE) {
3744                 e_err("MTU > 9216 not supported.\n");
3745                 return -EINVAL;
3746         }
3747
3748         while (test_and_set_bit(__E1000_RESETTING, &adapter->state))
3749                 msleep(1);
3750         /* e1000e_down has a dependency on max_frame_size */
3751         adapter->max_frame_size = max_frame;
3752         if (netif_running(netdev))
3753                 e1000e_down(adapter);
3754
3755         /*
3756          * NOTE: netdev_alloc_skb reserves 16 bytes, and typically NET_IP_ALIGN
3757          * means we reserve 2 more, this pushes us to allocate from the next
3758          * larger slab size.
3759          * i.e. RXBUFFER_2048 --> size-4096 slab
3760          * However with the new *_jumbo_rx* routines, jumbo receives will use
3761          * fragmented skbs
3762          */
3763
3764         if (max_frame <= 256)
3765                 adapter->rx_buffer_len = 256;
3766         else if (max_frame <= 512)
3767                 adapter->rx_buffer_len = 512;
3768         else if (max_frame <= 1024)
3769                 adapter->rx_buffer_len = 1024;
3770         else if (max_frame <= 2048)
3771                 adapter->rx_buffer_len = 2048;
3772         else
3773                 adapter->rx_buffer_len = 4096;
3774
3775         /* adjust allocation if LPE protects us, and we aren't using SBP */
3776         if ((max_frame == ETH_FRAME_LEN + ETH_FCS_LEN) ||
3777              (max_frame == ETH_FRAME_LEN + VLAN_HLEN + ETH_FCS_LEN))
3778                 adapter->rx_buffer_len = ETH_FRAME_LEN + VLAN_HLEN
3779                                          + ETH_FCS_LEN;
3780
3781         e_info("changing MTU from %d to %d\n", netdev->mtu, new_mtu);
3782         netdev->mtu = new_mtu;
3783
3784         if (netif_running(netdev))
3785                 e1000e_up(adapter);
3786         else
3787                 e1000e_reset(adapter);
3788
3789         clear_bit(__E1000_RESETTING, &adapter->state);
3790
3791         return 0;
3792 }
3793
3794 static int e1000_mii_ioctl(struct net_device *netdev, struct ifreq *ifr,
3795                            int cmd)
3796 {
3797         struct e1000_adapter *adapter = netdev_priv(netdev);
3798         struct mii_ioctl_data *data = if_mii(ifr);
3799
3800         if (adapter->hw.phy.media_type != e1000_media_type_copper)
3801                 return -EOPNOTSUPP;
3802
3803         switch (cmd) {
3804         case SIOCGMIIPHY:
3805                 data->phy_id = adapter->hw.phy.addr;
3806                 break;
3807         case SIOCGMIIREG:
3808                 if (!capable(CAP_NET_ADMIN))
3809                         return -EPERM;
3810                 switch (data->reg_num & 0x1F) {
3811                 case MII_BMCR:
3812                         data->val_out = adapter->phy_regs.bmcr;
3813                         break;
3814                 case MII_BMSR:
3815                         data->val_out = adapter->phy_regs.bmsr;
3816                         break;
3817                 case MII_PHYSID1:
3818                         data->val_out = (adapter->hw.phy.id >> 16);
3819                         break;
3820                 case MII_PHYSID2:
3821                         data->val_out = (adapter->hw.phy.id & 0xFFFF);
3822                         break;
3823                 case MII_ADVERTISE:
3824                         data->val_out = adapter->phy_regs.advertise;
3825                         break;
3826                 case MII_LPA:
3827                         data->val_out = adapter->phy_regs.lpa;
3828                         break;
3829                 case MII_EXPANSION:
3830                         data->val_out = adapter->phy_regs.expansion;
3831                         break;
3832                 case MII_CTRL1000:
3833                         data->val_out = adapter->phy_regs.ctrl1000;
3834                         break;
3835                 case MII_STAT1000:
3836                         data->val_out = adapter->phy_regs.stat1000;
3837                         break;
3838                 case MII_ESTATUS:
3839                         data->val_out = adapter->phy_regs.estatus;
3840                         break;
3841                 default:
3842                         return -EIO;
3843                 }
3844                 break;
3845         case SIOCSMIIREG:
3846         default:
3847                 return -EOPNOTSUPP;
3848         }
3849         return 0;
3850 }
3851
3852 static int e1000_ioctl(struct net_device *netdev, struct ifreq *ifr, int cmd)
3853 {
3854         switch (cmd) {
3855         case SIOCGMIIPHY:
3856         case SIOCGMIIREG:
3857         case SIOCSMIIREG:
3858                 return e1000_mii_ioctl(netdev, ifr, cmd);
3859         default:
3860                 return -EOPNOTSUPP;
3861         }
3862 }
3863
3864 static int e1000_suspend(struct pci_dev *pdev, pm_message_t state)
3865 {
3866         struct net_device *netdev = pci_get_drvdata(pdev);
3867         struct e1000_adapter *adapter = netdev_priv(netdev);
3868         struct e1000_hw *hw = &adapter->hw;
3869         u32 ctrl, ctrl_ext, rctl, status;
3870         u32 wufc = adapter->wol;
3871         int retval = 0;
3872
3873         netif_device_detach(netdev);
3874
3875         if (netif_running(netdev)) {
3876                 WARN_ON(test_bit(__E1000_RESETTING, &adapter->state));
3877                 e1000e_down(adapter);
3878                 e1000_free_irq(adapter);
3879         }
3880
3881         retval = pci_save_state(pdev);
3882         if (retval)
3883                 return retval;
3884
3885         status = er32(STATUS);
3886         if (status & E1000_STATUS_LU)
3887                 wufc &= ~E1000_WUFC_LNKC;
3888
3889         if (wufc) {
3890                 e1000_setup_rctl(adapter);
3891                 e1000_set_multi(netdev);
3892
3893                 /* turn on all-multi mode if wake on multicast is enabled */
3894                 if (wufc & E1000_WUFC_MC) {
3895                         rctl = er32(RCTL);
3896                         rctl |= E1000_RCTL_MPE;
3897                         ew32(RCTL, rctl);
3898                 }
3899
3900                 ctrl = er32(CTRL);
3901                 /* advertise wake from D3Cold */
3902                 #define E1000_CTRL_ADVD3WUC 0x00100000
3903                 /* phy power management enable */
3904                 #define E1000_CTRL_EN_PHY_PWR_MGMT 0x00200000
3905                 ctrl |= E1000_CTRL_ADVD3WUC |
3906                         E1000_CTRL_EN_PHY_PWR_MGMT;
3907                 ew32(CTRL, ctrl);
3908
3909                 if (adapter->hw.phy.media_type == e1000_media_type_fiber ||
3910                     adapter->hw.phy.media_type ==
3911                     e1000_media_type_internal_serdes) {
3912                         /* keep the laser running in D3 */
3913                         ctrl_ext = er32(CTRL_EXT);
3914                         ctrl_ext |= E1000_CTRL_EXT_SDP7_DATA;
3915                         ew32(CTRL_EXT, ctrl_ext);
3916                 }
3917
3918                 if (adapter->flags & FLAG_IS_ICH)
3919                         e1000e_disable_gig_wol_ich8lan(&adapter->hw);
3920
3921                 /* Allow time for pending master requests to run */
3922                 e1000e_disable_pcie_master(&adapter->hw);
3923
3924                 ew32(WUC, E1000_WUC_PME_EN);
3925                 ew32(WUFC, wufc);
3926                 pci_enable_wake(pdev, PCI_D3hot, 1);
3927                 pci_enable_wake(pdev, PCI_D3cold, 1);
3928         } else {
3929                 ew32(WUC, 0);
3930                 ew32(WUFC, 0);
3931                 pci_enable_wake(pdev, PCI_D3hot, 0);
3932                 pci_enable_wake(pdev, PCI_D3cold, 0);
3933         }
3934
3935         /* make sure adapter isn't asleep if manageability is enabled */
3936         if (adapter->flags & FLAG_MNG_PT_ENABLED) {
3937                 pci_enable_wake(pdev, PCI_D3hot, 1);
3938                 pci_enable_wake(pdev, PCI_D3cold, 1);
3939         }
3940
3941         if (adapter->hw.phy.type == e1000_phy_igp_3)
3942                 e1000e_igp3_phy_powerdown_workaround_ich8lan(&adapter->hw);
3943
3944         /*
3945          * Release control of h/w to f/w.  If f/w is AMT enabled, this
3946          * would have already happened in close and is redundant.
3947          */
3948         e1000_release_hw_control(adapter);
3949
3950         pci_disable_device(pdev);
3951
3952         pci_set_power_state(pdev, pci_choose_state(pdev, state));
3953
3954         return 0;
3955 }
3956
3957 static void e1000e_disable_l1aspm(struct pci_dev *pdev)
3958 {
3959         int pos;
3960         u16 val;
3961
3962         /*
3963          * 82573 workaround - disable L1 ASPM on mobile chipsets
3964          *
3965          * L1 ASPM on various mobile (ich7) chipsets do not behave properly
3966          * resulting in lost data or garbage information on the pci-e link
3967          * level. This could result in (false) bad EEPROM checksum errors,
3968          * long ping times (up to 2s) or even a system freeze/hang.
3969          *
3970          * Unfortunately this feature saves about 1W power consumption when
3971          * active.
3972          */
3973         pos = pci_find_capability(pdev, PCI_CAP_ID_EXP);
3974         pci_read_config_word(pdev, pos + PCI_EXP_LNKCTL, &val);
3975         if (val & 0x2) {
3976                 dev_warn(&pdev->dev, "Disabling L1 ASPM\n");
3977                 val &= ~0x2;
3978                 pci_write_config_word(pdev, pos + PCI_EXP_LNKCTL, val);
3979         }
3980 }
3981
3982 #ifdef CONFIG_PM
3983 static int e1000_resume(struct pci_dev *pdev)
3984 {
3985         struct net_device *netdev = pci_get_drvdata(pdev);
3986         struct e1000_adapter *adapter = netdev_priv(netdev);
3987         struct e1000_hw *hw = &adapter->hw;
3988         u32 err;
3989
3990         pci_set_power_state(pdev, PCI_D0);
3991         pci_restore_state(pdev);
3992         e1000e_disable_l1aspm(pdev);
3993
3994         err = pci_enable_device_mem(pdev);
3995         if (err) {
3996                 dev_err(&pdev->dev,
3997                         "Cannot enable PCI device from suspend\n");
3998                 return err;
3999         }
4000
4001         pci_set_master(pdev);
4002
4003         pci_enable_wake(pdev, PCI_D3hot, 0);
4004         pci_enable_wake(pdev, PCI_D3cold, 0);
4005
4006         if (netif_running(netdev)) {
4007                 err = e1000_request_irq(adapter);
4008                 if (err)
4009                         return err;
4010         }
4011
4012         e1000e_power_up_phy(adapter);
4013         e1000e_reset(adapter);
4014         ew32(WUS, ~0);
4015
4016         e1000_init_manageability(adapter);
4017
4018         if (netif_running(netdev))
4019                 e1000e_up(adapter);
4020
4021         netif_device_attach(netdev);
4022
4023         /*
4024          * If the controller has AMT, do not set DRV_LOAD until the interface
4025          * is up.  For all other cases, let the f/w know that the h/w is now
4026          * under the control of the driver.
4027          */
4028         if (!(adapter->flags & FLAG_HAS_AMT))
4029                 e1000_get_hw_control(adapter);
4030
4031         return 0;
4032 }
4033 #endif
4034
4035 static void e1000_shutdown(struct pci_dev *pdev)
4036 {
4037         e1000_suspend(pdev, PMSG_SUSPEND);
4038 }
4039
4040 #ifdef CONFIG_NET_POLL_CONTROLLER
4041 /*
4042  * Polling 'interrupt' - used by things like netconsole to send skbs
4043  * without having to re-enable interrupts. It's not called while
4044  * the interrupt routine is executing.
4045  */
4046 static void e1000_netpoll(struct net_device *netdev)
4047 {
4048         struct e1000_adapter *adapter = netdev_priv(netdev);
4049
4050         disable_irq(adapter->pdev->irq);
4051         e1000_intr(adapter->pdev->irq, netdev);
4052
4053         enable_irq(adapter->pdev->irq);
4054 }
4055 #endif
4056
4057 /**
4058  * e1000_io_error_detected - called when PCI error is detected
4059  * @pdev: Pointer to PCI device
4060  * @state: The current pci connection state
4061  *
4062  * This function is called after a PCI bus error affecting
4063  * this device has been detected.
4064  */
4065 static pci_ers_result_t e1000_io_error_detected(struct pci_dev *pdev,
4066                                                 pci_channel_state_t state)
4067 {
4068         struct net_device *netdev = pci_get_drvdata(pdev);
4069         struct e1000_adapter *adapter = netdev_priv(netdev);
4070
4071         netif_device_detach(netdev);
4072
4073         if (netif_running(netdev))
4074                 e1000e_down(adapter);
4075         pci_disable_device(pdev);
4076
4077         /* Request a slot slot reset. */
4078         return PCI_ERS_RESULT_NEED_RESET;
4079 }
4080
4081 /**
4082  * e1000_io_slot_reset - called after the pci bus has been reset.
4083  * @pdev: Pointer to PCI device
4084  *
4085  * Restart the card from scratch, as if from a cold-boot. Implementation
4086  * resembles the first-half of the e1000_resume routine.
4087  */
4088 static pci_ers_result_t e1000_io_slot_reset(struct pci_dev *pdev)
4089 {
4090         struct net_device *netdev = pci_get_drvdata(pdev);
4091         struct e1000_adapter *adapter = netdev_priv(netdev);
4092         struct e1000_hw *hw = &adapter->hw;
4093         int err;
4094
4095         e1000e_disable_l1aspm(pdev);
4096         err = pci_enable_device_mem(pdev);
4097         if (err) {
4098                 dev_err(&pdev->dev,
4099                         "Cannot re-enable PCI device after reset.\n");
4100                 return PCI_ERS_RESULT_DISCONNECT;
4101         }
4102         pci_set_master(pdev);
4103         pci_restore_state(pdev);
4104
4105         pci_enable_wake(pdev, PCI_D3hot, 0);
4106         pci_enable_wake(pdev, PCI_D3cold, 0);
4107
4108         e1000e_reset(adapter);
4109         ew32(WUS, ~0);
4110
4111         return PCI_ERS_RESULT_RECOVERED;
4112 }
4113
4114 /**
4115  * e1000_io_resume - called when traffic can start flowing again.
4116  * @pdev: Pointer to PCI device
4117  *
4118  * This callback is called when the error recovery driver tells us that
4119  * its OK to resume normal operation. Implementation resembles the
4120  * second-half of the e1000_resume routine.
4121  */
4122 static void e1000_io_resume(struct pci_dev *pdev)
4123 {
4124         struct net_device *netdev = pci_get_drvdata(pdev);
4125         struct e1000_adapter *adapter = netdev_priv(netdev);
4126
4127         e1000_init_manageability(adapter);
4128
4129         if (netif_running(netdev)) {
4130                 if (e1000e_up(adapter)) {
4131                         dev_err(&pdev->dev,
4132                                 "can't bring device back up after reset\n");
4133                         return;
4134                 }
4135         }
4136
4137         netif_device_attach(netdev);
4138
4139         /*
4140          * If the controller has AMT, do not set DRV_LOAD until the interface
4141          * is up.  For all other cases, let the f/w know that the h/w is now
4142          * under the control of the driver.
4143          */
4144         if (!(adapter->flags & FLAG_HAS_AMT))
4145                 e1000_get_hw_control(adapter);
4146
4147 }
4148
4149 static void e1000_print_device_info(struct e1000_adapter *adapter)
4150 {
4151         struct e1000_hw *hw = &adapter->hw;
4152         struct net_device *netdev = adapter->netdev;
4153         u32 pba_num;
4154
4155         /* print bus type/speed/width info */
4156         e_info("(PCI Express:2.5GB/s:%s) %02x:%02x:%02x:%02x:%02x:%02x\n",
4157                /* bus width */
4158                ((hw->bus.width == e1000_bus_width_pcie_x4) ? "Width x4" :
4159                 "Width x1"),
4160                /* MAC address */
4161                netdev->dev_addr[0], netdev->dev_addr[1],
4162                netdev->dev_addr[2], netdev->dev_addr[3],
4163                netdev->dev_addr[4], netdev->dev_addr[5]);
4164         e_info("Intel(R) PRO/%s Network Connection\n",
4165                (hw->phy.type == e1000_phy_ife) ? "10/100" : "1000");
4166         e1000e_read_pba_num(hw, &pba_num);
4167         e_info("MAC: %d, PHY: %d, PBA No: %06x-%03x\n",
4168                hw->mac.type, hw->phy.type, (pba_num >> 8), (pba_num & 0xff));
4169 }
4170
4171 static void e1000_eeprom_checks(struct e1000_adapter *adapter)
4172 {
4173         struct e1000_hw *hw = &adapter->hw;
4174         int ret_val;
4175         u16 buf = 0;
4176
4177         if (hw->mac.type != e1000_82573)
4178                 return;
4179
4180         ret_val = e1000_read_nvm(hw, NVM_INIT_CONTROL2_REG, 1, &buf);
4181         if (!(le16_to_cpu(buf) & (1 << 0))) {
4182                 /* Deep Smart Power Down (DSPD) */
4183                 e_warn("Warning: detected DSPD enabled in EEPROM\n");
4184         }
4185
4186         ret_val = e1000_read_nvm(hw, NVM_INIT_3GIO_3, 1, &buf);
4187         if (le16_to_cpu(buf) & (3 << 2)) {
4188                 /* ASPM enable */
4189                 e_warn("Warning: detected ASPM enabled in EEPROM\n");
4190         }
4191 }
4192
4193 /**
4194  * e1000_probe - Device Initialization Routine
4195  * @pdev: PCI device information struct
4196  * @ent: entry in e1000_pci_tbl
4197  *
4198  * Returns 0 on success, negative on failure
4199  *
4200  * e1000_probe initializes an adapter identified by a pci_dev structure.
4201  * The OS initialization, configuring of the adapter private structure,
4202  * and a hardware reset occur.
4203  **/
4204 static int __devinit e1000_probe(struct pci_dev *pdev,
4205                                  const struct pci_device_id *ent)
4206 {
4207         struct net_device *netdev;
4208         struct e1000_adapter *adapter;
4209         struct e1000_hw *hw;
4210         const struct e1000_info *ei = e1000_info_tbl[ent->driver_data];
4211         resource_size_t mmio_start, mmio_len;
4212         resource_size_t flash_start, flash_len;
4213
4214         static int cards_found;
4215         int i, err, pci_using_dac;
4216         u16 eeprom_data = 0;
4217         u16 eeprom_apme_mask = E1000_EEPROM_APME;
4218
4219         e1000e_disable_l1aspm(pdev);
4220
4221         err = pci_enable_device_mem(pdev);
4222         if (err)
4223                 return err;
4224
4225         pci_using_dac = 0;
4226         err = pci_set_dma_mask(pdev, DMA_64BIT_MASK);
4227         if (!err) {
4228                 err = pci_set_consistent_dma_mask(pdev, DMA_64BIT_MASK);
4229                 if (!err)
4230                         pci_using_dac = 1;
4231         } else {
4232                 err = pci_set_dma_mask(pdev, DMA_32BIT_MASK);
4233                 if (err) {
4234                         err = pci_set_consistent_dma_mask(pdev,
4235                                                           DMA_32BIT_MASK);
4236                         if (err) {
4237                                 dev_err(&pdev->dev, "No usable DMA "
4238                                         "configuration, aborting\n");
4239                                 goto err_dma;
4240                         }
4241                 }
4242         }
4243
4244         err = pci_request_selected_regions(pdev,
4245                                           pci_select_bars(pdev, IORESOURCE_MEM),
4246                                           e1000e_driver_name);
4247         if (err)
4248                 goto err_pci_reg;
4249
4250         pci_set_master(pdev);
4251         pci_save_state(pdev);
4252
4253         err = -ENOMEM;
4254         netdev = alloc_etherdev(sizeof(struct e1000_adapter));
4255         if (!netdev)
4256                 goto err_alloc_etherdev;
4257
4258         SET_NETDEV_DEV(netdev, &pdev->dev);
4259
4260         pci_set_drvdata(pdev, netdev);
4261         adapter = netdev_priv(netdev);
4262         hw = &adapter->hw;
4263         adapter->netdev = netdev;
4264         adapter->pdev = pdev;
4265         adapter->ei = ei;
4266         adapter->pba = ei->pba;
4267         adapter->flags = ei->flags;
4268         adapter->hw.adapter = adapter;
4269         adapter->hw.mac.type = ei->mac;
4270         adapter->msg_enable = (1 << NETIF_MSG_DRV | NETIF_MSG_PROBE) - 1;
4271
4272         mmio_start = pci_resource_start(pdev, 0);
4273         mmio_len = pci_resource_len(pdev, 0);
4274
4275         err = -EIO;
4276         adapter->hw.hw_addr = ioremap(mmio_start, mmio_len);
4277         if (!adapter->hw.hw_addr)
4278                 goto err_ioremap;
4279
4280         if ((adapter->flags & FLAG_HAS_FLASH) &&
4281             (pci_resource_flags(pdev, 1) & IORESOURCE_MEM)) {
4282                 flash_start = pci_resource_start(pdev, 1);
4283                 flash_len = pci_resource_len(pdev, 1);
4284                 adapter->hw.flash_address = ioremap(flash_start, flash_len);
4285                 if (!adapter->hw.flash_address)
4286                         goto err_flashmap;
4287         }
4288
4289         /* construct the net_device struct */
4290         netdev->open                    = &e1000_open;
4291         netdev->stop                    = &e1000_close;
4292         netdev->hard_start_xmit         = &e1000_xmit_frame;
4293         netdev->get_stats               = &e1000_get_stats;
4294         netdev->set_multicast_list      = &e1000_set_multi;
4295         netdev->set_mac_address         = &e1000_set_mac;
4296         netdev->change_mtu              = &e1000_change_mtu;
4297         netdev->do_ioctl                = &e1000_ioctl;
4298         e1000e_set_ethtool_ops(netdev);
4299         netdev->tx_timeout              = &e1000_tx_timeout;
4300         netdev->watchdog_timeo          = 5 * HZ;
4301         netif_napi_add(netdev, &adapter->napi, e1000_clean, 64);
4302         netdev->vlan_rx_register        = e1000_vlan_rx_register;
4303         netdev->vlan_rx_add_vid         = e1000_vlan_rx_add_vid;
4304         netdev->vlan_rx_kill_vid        = e1000_vlan_rx_kill_vid;
4305 #ifdef CONFIG_NET_POLL_CONTROLLER
4306         netdev->poll_controller         = e1000_netpoll;
4307 #endif
4308         strncpy(netdev->name, pci_name(pdev), sizeof(netdev->name) - 1);
4309
4310         netdev->mem_start = mmio_start;
4311         netdev->mem_end = mmio_start + mmio_len;
4312
4313         adapter->bd_number = cards_found++;
4314
4315         /* setup adapter struct */
4316         err = e1000_sw_init(adapter);
4317         if (err)
4318                 goto err_sw_init;
4319
4320         err = -EIO;
4321
4322         memcpy(&hw->mac.ops, ei->mac_ops, sizeof(hw->mac.ops));
4323         memcpy(&hw->nvm.ops, ei->nvm_ops, sizeof(hw->nvm.ops));
4324         memcpy(&hw->phy.ops, ei->phy_ops, sizeof(hw->phy.ops));
4325
4326         err = ei->get_variants(adapter);
4327         if (err)
4328                 goto err_hw_init;
4329
4330         hw->mac.ops.get_bus_info(&adapter->hw);
4331
4332         adapter->hw.phy.autoneg_wait_to_complete = 0;
4333
4334         /* Copper options */
4335         if (adapter->hw.phy.media_type == e1000_media_type_copper) {
4336                 adapter->hw.phy.mdix = AUTO_ALL_MODES;
4337                 adapter->hw.phy.disable_polarity_correction = 0;
4338                 adapter->hw.phy.ms_type = e1000_ms_hw_default;
4339         }
4340
4341         if (e1000_check_reset_block(&adapter->hw))
4342                 e_info("PHY reset is blocked due to SOL/IDER session.\n");
4343
4344         netdev->features = NETIF_F_SG |
4345                            NETIF_F_HW_CSUM |
4346                            NETIF_F_HW_VLAN_TX |
4347                            NETIF_F_HW_VLAN_RX;
4348
4349         if (adapter->flags & FLAG_HAS_HW_VLAN_FILTER)
4350                 netdev->features |= NETIF_F_HW_VLAN_FILTER;
4351
4352         netdev->features |= NETIF_F_TSO;
4353         netdev->features |= NETIF_F_TSO6;
4354
4355         netdev->vlan_features |= NETIF_F_TSO;
4356         netdev->vlan_features |= NETIF_F_TSO6;
4357         netdev->vlan_features |= NETIF_F_HW_CSUM;
4358         netdev->vlan_features |= NETIF_F_SG;
4359
4360         if (pci_using_dac)
4361                 netdev->features |= NETIF_F_HIGHDMA;
4362
4363         /*
4364          * We should not be using LLTX anymore, but we are still Tx faster with
4365          * it.
4366          */
4367         netdev->features |= NETIF_F_LLTX;
4368
4369         if (e1000e_enable_mng_pass_thru(&adapter->hw))
4370                 adapter->flags |= FLAG_MNG_PT_ENABLED;
4371
4372         /*
4373          * before reading the NVM, reset the controller to
4374          * put the device in a known good starting state
4375          */
4376         adapter->hw.mac.ops.reset_hw(&adapter->hw);
4377
4378         /*
4379          * systems with ASPM and others may see the checksum fail on the first
4380          * attempt. Let's give it a few tries
4381          */
4382         for (i = 0;; i++) {
4383                 if (e1000_validate_nvm_checksum(&adapter->hw) >= 0)
4384                         break;
4385                 if (i == 2) {
4386                         e_err("The NVM Checksum Is Not Valid\n");
4387                         err = -EIO;
4388                         goto err_eeprom;
4389                 }
4390         }
4391
4392         e1000_eeprom_checks(adapter);
4393
4394         /* copy the MAC address out of the NVM */
4395         if (e1000e_read_mac_addr(&adapter->hw))
4396                 e_err("NVM Read Error while reading MAC address\n");
4397
4398         memcpy(netdev->dev_addr, adapter->hw.mac.addr, netdev->addr_len);
4399         memcpy(netdev->perm_addr, adapter->hw.mac.addr, netdev->addr_len);
4400
4401         if (!is_valid_ether_addr(netdev->perm_addr)) {
4402                 e_err("Invalid MAC Address: %02x:%02x:%02x:%02x:%02x:%02x\n",
4403                       netdev->perm_addr[0], netdev->perm_addr[1],
4404                       netdev->perm_addr[2], netdev->perm_addr[3],
4405                       netdev->perm_addr[4], netdev->perm_addr[5]);
4406                 err = -EIO;
4407                 goto err_eeprom;
4408         }
4409
4410         init_timer(&adapter->watchdog_timer);
4411         adapter->watchdog_timer.function = &e1000_watchdog;
4412         adapter->watchdog_timer.data = (unsigned long) adapter;
4413
4414         init_timer(&adapter->phy_info_timer);
4415         adapter->phy_info_timer.function = &e1000_update_phy_info;
4416         adapter->phy_info_timer.data = (unsigned long) adapter;
4417
4418         INIT_WORK(&adapter->reset_task, e1000_reset_task);
4419         INIT_WORK(&adapter->watchdog_task, e1000_watchdog_task);
4420
4421         e1000e_check_options(adapter);
4422
4423         /* Initialize link parameters. User can change them with ethtool */
4424         adapter->hw.mac.autoneg = 1;
4425         adapter->fc_autoneg = 1;
4426         adapter->hw.fc.original_type = e1000_fc_default;
4427         adapter->hw.fc.type = e1000_fc_default;
4428         adapter->hw.phy.autoneg_advertised = 0x2f;
4429
4430         /* ring size defaults */
4431         adapter->rx_ring->count = 256;
4432         adapter->tx_ring->count = 256;
4433
4434         /*
4435          * Initial Wake on LAN setting - If APM wake is enabled in
4436          * the EEPROM, enable the ACPI Magic Packet filter
4437          */
4438         if (adapter->flags & FLAG_APME_IN_WUC) {
4439                 /* APME bit in EEPROM is mapped to WUC.APME */
4440                 eeprom_data = er32(WUC);
4441                 eeprom_apme_mask = E1000_WUC_APME;
4442         } else if (adapter->flags & FLAG_APME_IN_CTRL3) {
4443                 if (adapter->flags & FLAG_APME_CHECK_PORT_B &&
4444                     (adapter->hw.bus.func == 1))
4445                         e1000_read_nvm(&adapter->hw,
4446                                 NVM_INIT_CONTROL3_PORT_B, 1, &eeprom_data);
4447                 else
4448                         e1000_read_nvm(&adapter->hw,
4449                                 NVM_INIT_CONTROL3_PORT_A, 1, &eeprom_data);
4450         }
4451
4452         /* fetch WoL from EEPROM */
4453         if (eeprom_data & eeprom_apme_mask)
4454                 adapter->eeprom_wol |= E1000_WUFC_MAG;
4455
4456         /*
4457          * now that we have the eeprom settings, apply the special cases
4458          * where the eeprom may be wrong or the board simply won't support
4459          * wake on lan on a particular port
4460          */
4461         if (!(adapter->flags & FLAG_HAS_WOL))
4462                 adapter->eeprom_wol = 0;
4463
4464         /* initialize the wol settings based on the eeprom settings */
4465         adapter->wol = adapter->eeprom_wol;
4466
4467         /* reset the hardware with the new settings */
4468         e1000e_reset(adapter);
4469
4470         /*
4471          * If the controller has AMT, do not set DRV_LOAD until the interface
4472          * is up.  For all other cases, let the f/w know that the h/w is now
4473          * under the control of the driver.
4474          */
4475         if (!(adapter->flags & FLAG_HAS_AMT))
4476                 e1000_get_hw_control(adapter);
4477
4478         /* tell the stack to leave us alone until e1000_open() is called */
4479         netif_carrier_off(netdev);
4480         netif_tx_stop_all_queues(netdev);
4481
4482         strcpy(netdev->name, "eth%d");
4483         err = register_netdev(netdev);
4484         if (err)
4485                 goto err_register;
4486
4487         e1000_print_device_info(adapter);
4488
4489         return 0;
4490
4491 err_register:
4492         if (!(adapter->flags & FLAG_HAS_AMT))
4493                 e1000_release_hw_control(adapter);
4494 err_eeprom:
4495         if (!e1000_check_reset_block(&adapter->hw))
4496                 e1000_phy_hw_reset(&adapter->hw);
4497 err_hw_init:
4498
4499         kfree(adapter->tx_ring);
4500         kfree(adapter->rx_ring);
4501 err_sw_init:
4502         if (adapter->hw.flash_address)
4503                 iounmap(adapter->hw.flash_address);
4504 err_flashmap:
4505         iounmap(adapter->hw.hw_addr);
4506 err_ioremap:
4507         free_netdev(netdev);
4508 err_alloc_etherdev:
4509         pci_release_selected_regions(pdev,
4510                                      pci_select_bars(pdev, IORESOURCE_MEM));
4511 err_pci_reg:
4512 err_dma:
4513         pci_disable_device(pdev);
4514         return err;
4515 }
4516
4517 /**
4518  * e1000_remove - Device Removal Routine
4519  * @pdev: PCI device information struct
4520  *
4521  * e1000_remove is called by the PCI subsystem to alert the driver
4522  * that it should release a PCI device.  The could be caused by a
4523  * Hot-Plug event, or because the driver is going to be removed from
4524  * memory.
4525  **/
4526 static void __devexit e1000_remove(struct pci_dev *pdev)
4527 {
4528         struct net_device *netdev = pci_get_drvdata(pdev);
4529         struct e1000_adapter *adapter = netdev_priv(netdev);
4530
4531         /*
4532          * flush_scheduled work may reschedule our watchdog task, so
4533          * explicitly disable watchdog tasks from being rescheduled
4534          */
4535         set_bit(__E1000_DOWN, &adapter->state);
4536         del_timer_sync(&adapter->watchdog_timer);
4537         del_timer_sync(&adapter->phy_info_timer);
4538
4539         flush_scheduled_work();
4540
4541         /*
4542          * Release control of h/w to f/w.  If f/w is AMT enabled, this
4543          * would have already happened in close and is redundant.
4544          */
4545         e1000_release_hw_control(adapter);
4546
4547         unregister_netdev(netdev);
4548
4549         if (!e1000_check_reset_block(&adapter->hw))
4550                 e1000_phy_hw_reset(&adapter->hw);
4551
4552         kfree(adapter->tx_ring);
4553         kfree(adapter->rx_ring);
4554
4555         iounmap(adapter->hw.hw_addr);
4556         if (adapter->hw.flash_address)
4557                 iounmap(adapter->hw.flash_address);
4558         pci_release_selected_regions(pdev,
4559                                      pci_select_bars(pdev, IORESOURCE_MEM));
4560
4561         free_netdev(netdev);
4562
4563         pci_disable_device(pdev);
4564 }
4565
4566 /* PCI Error Recovery (ERS) */
4567 static struct pci_error_handlers e1000_err_handler = {
4568         .error_detected = e1000_io_error_detected,
4569         .slot_reset = e1000_io_slot_reset,
4570         .resume = e1000_io_resume,
4571 };
4572
4573 static struct pci_device_id e1000_pci_tbl[] = {
4574         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82571EB_COPPER), board_82571 },
4575         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82571EB_FIBER), board_82571 },
4576         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82571EB_QUAD_COPPER), board_82571 },
4577         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82571EB_QUAD_COPPER_LP), board_82571 },
4578         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82571EB_QUAD_FIBER), board_82571 },
4579         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82571EB_SERDES), board_82571 },
4580         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82571EB_SERDES_DUAL), board_82571 },
4581         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82571EB_SERDES_QUAD), board_82571 },
4582         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82571PT_QUAD_COPPER), board_82571 },
4583
4584         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82572EI), board_82572 },
4585         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82572EI_COPPER), board_82572 },
4586         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82572EI_FIBER), board_82572 },
4587         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82572EI_SERDES), board_82572 },
4588
4589         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82573E), board_82573 },
4590         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82573E_IAMT), board_82573 },
4591         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82573L), board_82573 },
4592
4593         { PCI_VDEVICE(INTEL, E1000_DEV_ID_80003ES2LAN_COPPER_DPT),
4594           board_80003es2lan },
4595         { PCI_VDEVICE(INTEL, E1000_DEV_ID_80003ES2LAN_COPPER_SPT),
4596           board_80003es2lan },
4597         { PCI_VDEVICE(INTEL, E1000_DEV_ID_80003ES2LAN_SERDES_DPT),
4598           board_80003es2lan },
4599         { PCI_VDEVICE(INTEL, E1000_DEV_ID_80003ES2LAN_SERDES_SPT),
4600           board_80003es2lan },
4601
4602         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH8_IFE), board_ich8lan },
4603         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH8_IFE_G), board_ich8lan },
4604         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH8_IFE_GT), board_ich8lan },
4605         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH8_IGP_AMT), board_ich8lan },
4606         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH8_IGP_C), board_ich8lan },
4607         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH8_IGP_M), board_ich8lan },
4608         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH8_IGP_M_AMT), board_ich8lan },
4609
4610         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH9_IFE), board_ich9lan },
4611         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH9_IFE_G), board_ich9lan },
4612         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH9_IFE_GT), board_ich9lan },
4613         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH9_IGP_AMT), board_ich9lan },
4614         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH9_IGP_C), board_ich9lan },
4615         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH9_IGP_M), board_ich9lan },
4616         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH9_IGP_M_AMT), board_ich9lan },
4617         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH9_IGP_M_V), board_ich9lan },
4618
4619         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH10_R_BM_LM), board_ich9lan },
4620         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH10_R_BM_LF), board_ich9lan },
4621         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH10_R_BM_V), board_ich9lan },
4622
4623         { }     /* terminate list */
4624 };
4625 MODULE_DEVICE_TABLE(pci, e1000_pci_tbl);
4626
4627 /* PCI Device API Driver */
4628 static struct pci_driver e1000_driver = {
4629         .name     = e1000e_driver_name,
4630         .id_table = e1000_pci_tbl,
4631         .probe    = e1000_probe,
4632         .remove   = __devexit_p(e1000_remove),
4633 #ifdef CONFIG_PM
4634         /* Power Management Hooks */
4635         .suspend  = e1000_suspend,
4636         .resume   = e1000_resume,
4637 #endif
4638         .shutdown = e1000_shutdown,
4639         .err_handler = &e1000_err_handler
4640 };
4641
4642 /**
4643  * e1000_init_module - Driver Registration Routine
4644  *
4645  * e1000_init_module is the first routine called when the driver is
4646  * loaded. All it does is register with the PCI subsystem.
4647  **/
4648 static int __init e1000_init_module(void)
4649 {
4650         int ret;
4651         printk(KERN_INFO "%s: Intel(R) PRO/1000 Network Driver - %s\n",
4652                e1000e_driver_name, e1000e_driver_version);
4653         printk(KERN_INFO "%s: Copyright (c) 1999-2008 Intel Corporation.\n",
4654                e1000e_driver_name);
4655         ret = pci_register_driver(&e1000_driver);
4656         pm_qos_add_requirement(PM_QOS_CPU_DMA_LATENCY, e1000e_driver_name,
4657                                PM_QOS_DEFAULT_VALUE);
4658
4659         return ret;
4660 }
4661 module_init(e1000_init_module);
4662
4663 /**
4664  * e1000_exit_module - Driver Exit Cleanup Routine
4665  *
4666  * e1000_exit_module is called just before the driver is removed
4667  * from memory.
4668  **/
4669 static void __exit e1000_exit_module(void)
4670 {
4671         pci_unregister_driver(&e1000_driver);
4672         pm_qos_remove_requirement(PM_QOS_CPU_DMA_LATENCY, e1000e_driver_name);
4673 }
4674 module_exit(e1000_exit_module);
4675
4676
4677 MODULE_AUTHOR("Intel Corporation, <linux.nics@intel.com>");
4678 MODULE_DESCRIPTION("Intel(R) PRO/1000 Network Driver");
4679 MODULE_LICENSE("GPL");
4680 MODULE_VERSION(DRV_VERSION);
4681
4682 /* e1000_main.c */