drivers/net/ethernet/intel/igbvf/netdev.c

   1 /*******************************************************************************
   2
   3   Intel(R) 82576 Virtual Function Linux driver
   4   Copyright(c) 2009 - 2012 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   e1000-devel Mailing List <e1000-devel@lists.sourceforge.net>
  24   Intel Corporation, 5200 N.E. Elam Young Parkway, Hillsboro, OR 97124-6497
  25
  26 *******************************************************************************/
  27
  28 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
  29
  30 #include <linux/module.h>
  31 #include <linux/types.h>
  32 #include <linux/init.h>
  33 #include <linux/pci.h>
  34 #include <linux/vmalloc.h>
  35 #include <linux/pagemap.h>
  36 #include <linux/delay.h>
  37 #include <linux/netdevice.h>
  38 #include <linux/tcp.h>
  39 #include <linux/ipv6.h>
  40 #include <linux/slab.h>
  41 #include <net/checksum.h>
  42 #include <net/ip6_checksum.h>
  43 #include <linux/mii.h>
  44 #include <linux/ethtool.h>
  45 #include <linux/if_vlan.h>
  46 #include <linux/prefetch.h>
  47
  48 #include "igbvf.h"
  49
  50 #define DRV_VERSION "2.0.1-k"
  51 char igbvf_driver_name[] = "igbvf";
  52 const char igbvf_driver_version[] = DRV_VERSION;
  53 static const char igbvf_driver_string[] =
  54                   "Intel(R) Gigabit Virtual Function Network Driver";
  55 static const char igbvf_copyright[] =
  56                   "Copyright (c) 2009 - 2012 Intel Corporation.";
  57
  58 static int igbvf_poll(struct napi_struct *napi, int budget);
  59 static void igbvf_reset(struct igbvf_adapter *);
  60 static void igbvf_set_interrupt_capability(struct igbvf_adapter *);
  61 static void igbvf_reset_interrupt_capability(struct igbvf_adapter *);
  62
  63 static struct igbvf_info igbvf_vf_info = {
  64         .mac                    = e1000_vfadapt,
  65         .flags                  = 0,
  66         .pba                    = 10,
  67         .init_ops               = e1000_init_function_pointers_vf,
  68 };
  69
  70 static struct igbvf_info igbvf_i350_vf_info = {
  71         .mac                    = e1000_vfadapt_i350,
  72         .flags                  = 0,
  73         .pba                    = 10,
  74         .init_ops               = e1000_init_function_pointers_vf,
  75 };
  76
  77 static const struct igbvf_info *igbvf_info_tbl[] = {
  78         [board_vf]              = &igbvf_vf_info,
  79         [board_i350_vf]         = &igbvf_i350_vf_info,
  80 };
  81
  82 /**
  83  * igbvf_desc_unused - calculate if we have unused descriptors
  84  **/
  85 static int igbvf_desc_unused(struct igbvf_ring *ring)
  86 {
  87         if (ring->next_to_clean > ring->next_to_use)
  88                 return ring->next_to_clean - ring->next_to_use - 1;
  89
  90         return ring->count + ring->next_to_clean - ring->next_to_use - 1;
  91 }
  92
  93 /**
  94  * igbvf_receive_skb - helper function to handle Rx indications
  95  * @adapter: board private structure
  96  * @status: descriptor status field as written by hardware
  97  * @vlan: descriptor vlan field as written by hardware (no le/be conversion)
  98  * @skb: pointer to sk_buff to be indicated to stack
  99  **/
 100 static void igbvf_receive_skb(struct igbvf_adapter *adapter,
 101                               struct net_device *netdev,
 102                               struct sk_buff *skb,
 103                               u32 status, u16 vlan)
 104 {
 105         if (status & E1000_RXD_STAT_VP) {
 106                 u16 vid = le16_to_cpu(vlan) & E1000_RXD_SPC_VLAN_MASK;
 107                 if (test_bit(vid, adapter->active_vlans))
 108                         __vlan_hwaccel_put_tag(skb, vid);
 109         }
 110         netif_receive_skb(skb);
 111 }
 112
 113 static inline void igbvf_rx_checksum_adv(struct igbvf_adapter *adapter,
 114                                          u32 status_err, struct sk_buff *skb)
 115 {
 116         skb_checksum_none_assert(skb);
 117
 118         /* Ignore Checksum bit is set or checksum is disabled through ethtool */
 119         if ((status_err & E1000_RXD_STAT_IXSM) ||
 120             (adapter->flags & IGBVF_FLAG_RX_CSUM_DISABLED))
 121                 return;
 122
 123         /* TCP/UDP checksum error bit is set */
 124         if (status_err &
 125             (E1000_RXDEXT_STATERR_TCPE | E1000_RXDEXT_STATERR_IPE)) {
 126                 /* let the stack verify checksum errors */
 127                 adapter->hw_csum_err++;
 128                 return;
 129         }
 130
 131         /* It must be a TCP or UDP packet with a valid checksum */
 132         if (status_err & (E1000_RXD_STAT_TCPCS | E1000_RXD_STAT_UDPCS))
 133                 skb->ip_summed = CHECKSUM_UNNECESSARY;
 134
 135         adapter->hw_csum_good++;
 136 }
 137
 138 /**
 139  * igbvf_alloc_rx_buffers - Replace used receive buffers; packet split
 140  * @rx_ring: address of ring structure to repopulate
 141  * @cleaned_count: number of buffers to repopulate
 142  **/
 143 static void igbvf_alloc_rx_buffers(struct igbvf_ring *rx_ring,
 144                                    int cleaned_count)
 145 {
 146         struct igbvf_adapter *adapter = rx_ring->adapter;
 147         struct net_device *netdev = adapter->netdev;
 148         struct pci_dev *pdev = adapter->pdev;
 149         union e1000_adv_rx_desc *rx_desc;
 150         struct igbvf_buffer *buffer_info;
 151         struct sk_buff *skb;
 152         unsigned int i;
 153         int bufsz;
 154
 155         i = rx_ring->next_to_use;
 156         buffer_info = &rx_ring->buffer_info[i];
 157
 158         if (adapter->rx_ps_hdr_size)
 159                 bufsz = adapter->rx_ps_hdr_size;
 160         else
 161                 bufsz = adapter->rx_buffer_len;
 162
 163         while (cleaned_count--) {
 164                 rx_desc = IGBVF_RX_DESC_ADV(*rx_ring, i);
 165
 166                 if (adapter->rx_ps_hdr_size && !buffer_info->page_dma) {
 167                         if (!buffer_info->page) {
 168                                 buffer_info->page = alloc_page(GFP_ATOMIC);
 169                                 if (!buffer_info->page) {
 170                                         adapter->alloc_rx_buff_failed++;
 171                                         goto no_buffers;
 172                                 }
 173                                 buffer_info->page_offset = 0;
 174                         } else {
 175                                 buffer_info->page_offset ^= PAGE_SIZE / 2;
 176                         }
 177                         buffer_info->page_dma =
 178                                 dma_map_page(&pdev->dev, buffer_info->page,
 179                                              buffer_info->page_offset,
 180                                              PAGE_SIZE / 2,
 181                                              DMA_FROM_DEVICE);
 182                 }
 183
 184                 if (!buffer_info->skb) {
 185                         skb = netdev_alloc_skb_ip_align(netdev, bufsz);
 186                         if (!skb) {
 187                                 adapter->alloc_rx_buff_failed++;
 188                                 goto no_buffers;
 189                         }
 190
 191                         buffer_info->skb = skb;
 192                         buffer_info->dma = dma_map_single(&pdev->dev, skb->data,
 193                                                           bufsz,
 194                                                           DMA_FROM_DEVICE);
 195                 }
 196                 /* Refresh the desc even if buffer_addrs didn't change because
 197                  * each write-back erases this info. */
 198                 if (adapter->rx_ps_hdr_size) {
 199                         rx_desc->read.pkt_addr =
 200                              cpu_to_le64(buffer_info->page_dma);
 201                         rx_desc->read.hdr_addr = cpu_to_le64(buffer_info->dma);
 202                 } else {
 203                         rx_desc->read.pkt_addr =
 204                              cpu_to_le64(buffer_info->dma);
 205                         rx_desc->read.hdr_addr = 0;
 206                 }
 207
 208                 i++;
 209                 if (i == rx_ring->count)
 210                         i = 0;
 211                 buffer_info = &rx_ring->buffer_info[i];
 212         }
 213
 214 no_buffers:
 215         if (rx_ring->next_to_use != i) {
 216                 rx_ring->next_to_use = i;
 217                 if (i == 0)
 218                         i = (rx_ring->count - 1);
 219                 else
 220                         i--;
 221
 222                 /* Force memory writes to complete before letting h/w
 223                  * know there are new descriptors to fetch.  (Only
 224                  * applicable for weak-ordered memory model archs,
 225                  * such as IA-64). */
 226                 wmb();
 227                 writel(i, adapter->hw.hw_addr + rx_ring->tail);
 228         }
 229 }
 230
 231 /**
 232  * igbvf_clean_rx_irq - Send received data up the network stack; legacy
 233  * @adapter: board private structure
 234  *
 235  * the return value indicates whether actual cleaning was done, there
 236  * is no guarantee that everything was cleaned
 237  **/
 238 static bool igbvf_clean_rx_irq(struct igbvf_adapter *adapter,
 239                                int *work_done, int work_to_do)
 240 {
 241         struct igbvf_ring *rx_ring = adapter->rx_ring;
 242         struct net_device *netdev = adapter->netdev;
 243         struct pci_dev *pdev = adapter->pdev;
 244         union e1000_adv_rx_desc *rx_desc, *next_rxd;
 245         struct igbvf_buffer *buffer_info, *next_buffer;
 246         struct sk_buff *skb;
 247         bool cleaned = false;
 248         int cleaned_count = 0;
 249         unsigned int total_bytes = 0, total_packets = 0;
 250         unsigned int i;
 251         u32 length, hlen, staterr;
 252
 253         i = rx_ring->next_to_clean;
 254         rx_desc = IGBVF_RX_DESC_ADV(*rx_ring, i);
 255         staterr = le32_to_cpu(rx_desc->wb.upper.status_error);
 256
 257         while (staterr & E1000_RXD_STAT_DD) {
 258                 if (*work_done >= work_to_do)
 259                         break;
 260                 (*work_done)++;
 261                 rmb(); /* read descriptor and rx_buffer_info after status DD */
 262
 263                 buffer_info = &rx_ring->buffer_info[i];
 264
 265                 /* HW will not DMA in data larger than the given buffer, even
 266                  * if it parses the (NFS, of course) header to be larger.  In
 267                  * that case, it fills the header buffer and spills the rest
 268                  * into the page.
 269                  */
 270                 hlen = (le16_to_cpu(rx_desc->wb.lower.lo_dword.hs_rss.hdr_info) &
 271                   E1000_RXDADV_HDRBUFLEN_MASK) >> E1000_RXDADV_HDRBUFLEN_SHIFT;
 272                 if (hlen > adapter->rx_ps_hdr_size)
 273                         hlen = adapter->rx_ps_hdr_size;
 274
 275                 length = le16_to_cpu(rx_desc->wb.upper.length);
 276                 cleaned = true;
 277                 cleaned_count++;
 278
 279                 skb = buffer_info->skb;
 280                 prefetch(skb->data - NET_IP_ALIGN);
 281                 buffer_info->skb = NULL;
 282                 if (!adapter->rx_ps_hdr_size) {
 283                         dma_unmap_single(&pdev->dev, buffer_info->dma,
 284                                          adapter->rx_buffer_len,
 285                                          DMA_FROM_DEVICE);
 286                         buffer_info->dma = 0;
 287                         skb_put(skb, length);
 288                         goto send_up;
 289                 }
 290
 291                 if (!skb_shinfo(skb)->nr_frags) {
 292                         dma_unmap_single(&pdev->dev, buffer_info->dma,
 293                                          adapter->rx_ps_hdr_size,
 294                                          DMA_FROM_DEVICE);
 295                         skb_put(skb, hlen);
 296                 }
 297
 298                 if (length) {
 299                         dma_unmap_page(&pdev->dev, buffer_info->page_dma,
 300                                        PAGE_SIZE / 2,
 301                                        DMA_FROM_DEVICE);
 302                         buffer_info->page_dma = 0;
 303
 304                         skb_fill_page_desc(skb, skb_shinfo(skb)->nr_frags,
 305                                            buffer_info->page,
 306                                            buffer_info->page_offset,
 307                                            length);
 308
 309                         if ((adapter->rx_buffer_len > (PAGE_SIZE / 2)) ||
 310                             (page_count(buffer_info->page) != 1))
 311                                 buffer_info->page = NULL;
 312                         else
 313                                 get_page(buffer_info->page);
 314
 315                         skb->len += length;
 316                         skb->data_len += length;
 317                         skb->truesize += PAGE_SIZE / 2;
 318                 }
 319 send_up:
 320                 i++;
 321                 if (i == rx_ring->count)
 322                         i = 0;
 323                 next_rxd = IGBVF_RX_DESC_ADV(*rx_ring, i);
 324                 prefetch(next_rxd);
 325                 next_buffer = &rx_ring->buffer_info[i];
 326
 327                 if (!(staterr & E1000_RXD_STAT_EOP)) {
 328                         buffer_info->skb = next_buffer->skb;
 329                         buffer_info->dma = next_buffer->dma;
 330                         next_buffer->skb = skb;
 331                         next_buffer->dma = 0;
 332                         goto next_desc;
 333                 }
 334
 335                 if (staterr & E1000_RXDEXT_ERR_FRAME_ERR_MASK) {
 336                         dev_kfree_skb_irq(skb);
 337                         goto next_desc;
 338                 }
 339
 340                 total_bytes += skb->len;
 341                 total_packets++;
 342
 343                 igbvf_rx_checksum_adv(adapter, staterr, skb);
 344
 345                 skb->protocol = eth_type_trans(skb, netdev);
 346
 347                 igbvf_receive_skb(adapter, netdev, skb, staterr,
 348                                   rx_desc->wb.upper.vlan);
 349
 350 next_desc:
 351                 rx_desc->wb.upper.status_error = 0;
 352
 353                 /* return some buffers to hardware, one at a time is too slow */
 354                 if (cleaned_count >= IGBVF_RX_BUFFER_WRITE) {
 355                         igbvf_alloc_rx_buffers(rx_ring, cleaned_count);
 356                         cleaned_count = 0;
 357                 }
 358
 359                 /* use prefetched values */
 360                 rx_desc = next_rxd;
 361                 buffer_info = next_buffer;
 362
 363                 staterr = le32_to_cpu(rx_desc->wb.upper.status_error);
 364         }
 365
 366         rx_ring->next_to_clean = i;
 367         cleaned_count = igbvf_desc_unused(rx_ring);
 368
 369         if (cleaned_count)
 370                 igbvf_alloc_rx_buffers(rx_ring, cleaned_count);
 371
 372         adapter->total_rx_packets += total_packets;
 373         adapter->total_rx_bytes += total_bytes;
 374         adapter->net_stats.rx_bytes += total_bytes;
 375         adapter->net_stats.rx_packets += total_packets;
 376         return cleaned;
 377 }
 378
 379 static void igbvf_put_txbuf(struct igbvf_adapter *adapter,
 380                             struct igbvf_buffer *buffer_info)
 381 {
 382         if (buffer_info->dma) {
 383                 if (buffer_info->mapped_as_page)
 384                         dma_unmap_page(&adapter->pdev->dev,
 385                                        buffer_info->dma,
 386                                        buffer_info->length,
 387                                        DMA_TO_DEVICE);
 388                 else
 389                         dma_unmap_single(&adapter->pdev->dev,
 390                                          buffer_info->dma,
 391                                          buffer_info->length,
 392                                          DMA_TO_DEVICE);
 393                 buffer_info->dma = 0;
 394         }
 395         if (buffer_info->skb) {
 396                 dev_kfree_skb_any(buffer_info->skb);
 397                 buffer_info->skb = NULL;
 398         }
 399         buffer_info->time_stamp = 0;
 400 }
 401
 402 /**
 403  * igbvf_setup_tx_resources - allocate Tx resources (Descriptors)
 404  * @adapter: board private structure
 405  *
 406  * Return 0 on success, negative on failure
 407  **/
 408 int igbvf_setup_tx_resources(struct igbvf_adapter *adapter,
 409                              struct igbvf_ring *tx_ring)
 410 {
 411         struct pci_dev *pdev = adapter->pdev;
 412         int size;
 413
 414         size = sizeof(struct igbvf_buffer) * tx_ring->count;
 415         tx_ring->buffer_info = vzalloc(size);
 416         if (!tx_ring->buffer_info)
 417                 goto err;
 418
 419         /* round up to nearest 4K */
 420         tx_ring->size = tx_ring->count * sizeof(union e1000_adv_tx_desc);
 421         tx_ring->size = ALIGN(tx_ring->size, 4096);
 422
 423         tx_ring->desc = dma_alloc_coherent(&pdev->dev, tx_ring->size,
 424                                            &tx_ring->dma, GFP_KERNEL);
 425
 426         if (!tx_ring->desc)
 427                 goto err;
 428
 429         tx_ring->adapter = adapter;
 430         tx_ring->next_to_use = 0;
 431         tx_ring->next_to_clean = 0;
 432
 433         return 0;
 434 err:
 435         vfree(tx_ring->buffer_info);
 436         dev_err(&adapter->pdev->dev,
 437                 "Unable to allocate memory for the transmit descriptor ring\n");
 438         return -ENOMEM;
 439 }
 440
 441 /**
 442  * igbvf_setup_rx_resources - allocate Rx resources (Descriptors)
 443  * @adapter: board private structure
 444  *
 445  * Returns 0 on success, negative on failure
 446  **/
 447 int igbvf_setup_rx_resources(struct igbvf_adapter *adapter,
 448                              struct igbvf_ring *rx_ring)
 449 {
 450         struct pci_dev *pdev = adapter->pdev;
 451         int size, desc_len;
 452
 453         size = sizeof(struct igbvf_buffer) * rx_ring->count;
 454         rx_ring->buffer_info = vzalloc(size);
 455         if (!rx_ring->buffer_info)
 456                 goto err;
 457
 458         desc_len = sizeof(union e1000_adv_rx_desc);
 459
 460         /* Round up to nearest 4K */
 461         rx_ring->size = rx_ring->count * desc_len;
 462         rx_ring->size = ALIGN(rx_ring->size, 4096);
 463
 464         rx_ring->desc = dma_alloc_coherent(&pdev->dev, rx_ring->size,
 465                                            &rx_ring->dma, GFP_KERNEL);
 466
 467         if (!rx_ring->desc)
 468                 goto err;
 469
 470         rx_ring->next_to_clean = 0;
 471         rx_ring->next_to_use = 0;
 472
 473         rx_ring->adapter = adapter;
 474
 475         return 0;
 476
 477 err:
 478         vfree(rx_ring->buffer_info);
 479         rx_ring->buffer_info = NULL;
 480         dev_err(&adapter->pdev->dev,
 481                 "Unable to allocate memory for the receive descriptor ring\n");
 482         return -ENOMEM;
 483 }
 484
 485 /**
 486  * igbvf_clean_tx_ring - Free Tx Buffers
 487  * @tx_ring: ring to be cleaned
 488  **/
 489 static void igbvf_clean_tx_ring(struct igbvf_ring *tx_ring)
 490 {
 491         struct igbvf_adapter *adapter = tx_ring->adapter;
 492         struct igbvf_buffer *buffer_info;
 493         unsigned long size;
 494         unsigned int i;
 495
 496         if (!tx_ring->buffer_info)
 497                 return;
 498
 499         /* Free all the Tx ring sk_buffs */
 500         for (i = 0; i < tx_ring->count; i++) {
 501                 buffer_info = &tx_ring->buffer_info[i];
 502                 igbvf_put_txbuf(adapter, buffer_info);
 503         }
 504
 505         size = sizeof(struct igbvf_buffer) * tx_ring->count;
 506         memset(tx_ring->buffer_info, 0, size);
 507
 508         /* Zero out the descriptor ring */
 509         memset(tx_ring->desc, 0, tx_ring->size);
 510
 511         tx_ring->next_to_use = 0;
 512         tx_ring->next_to_clean = 0;
 513
 514         writel(0, adapter->hw.hw_addr + tx_ring->head);
 515         writel(0, adapter->hw.hw_addr + tx_ring->tail);
 516 }
 517
 518 /**
 519  * igbvf_free_tx_resources - Free Tx Resources per Queue
 520  * @tx_ring: ring to free resources from
 521  *
 522  * Free all transmit software resources
 523  **/
 524 void igbvf_free_tx_resources(struct igbvf_ring *tx_ring)
 525 {
 526         struct pci_dev *pdev = tx_ring->adapter->pdev;
 527
 528         igbvf_clean_tx_ring(tx_ring);
 529
 530         vfree(tx_ring->buffer_info);
 531         tx_ring->buffer_info = NULL;
 532
 533         dma_free_coherent(&pdev->dev, tx_ring->size, tx_ring->desc,
 534                           tx_ring->dma);
 535
 536         tx_ring->desc = NULL;
 537 }
 538
 539 /**
 540  * igbvf_clean_rx_ring - Free Rx Buffers per Queue
 541  * @adapter: board private structure
 542  **/
 543 static void igbvf_clean_rx_ring(struct igbvf_ring *rx_ring)
 544 {
 545         struct igbvf_adapter *adapter = rx_ring->adapter;
 546         struct igbvf_buffer *buffer_info;
 547         struct pci_dev *pdev = adapter->pdev;
 548         unsigned long size;
 549         unsigned int i;
 550
 551         if (!rx_ring->buffer_info)
 552                 return;
 553
 554         /* Free all the Rx ring sk_buffs */
 555         for (i = 0; i < rx_ring->count; i++) {
 556                 buffer_info = &rx_ring->buffer_info[i];
 557                 if (buffer_info->dma) {
 558                         if (adapter->rx_ps_hdr_size){
 559                                 dma_unmap_single(&pdev->dev, buffer_info->dma,
 560                                                  adapter->rx_ps_hdr_size,
 561                                                  DMA_FROM_DEVICE);
 562                         } else {
 563                                 dma_unmap_single(&pdev->dev, buffer_info->dma,
 564                                                  adapter->rx_buffer_len,
 565                                                  DMA_FROM_DEVICE);
 566                         }
 567                         buffer_info->dma = 0;
 568                 }
 569
 570                 if (buffer_info->skb) {
 571                         dev_kfree_skb(buffer_info->skb);
 572                         buffer_info->skb = NULL;
 573                 }
 574
 575                 if (buffer_info->page) {
 576                         if (buffer_info->page_dma)
 577                                 dma_unmap_page(&pdev->dev,
 578                                                buffer_info->page_dma,
 579                                                PAGE_SIZE / 2,
 580                                                DMA_FROM_DEVICE);
 581                         put_page(buffer_info->page);
 582                         buffer_info->page = NULL;
 583                         buffer_info->page_dma = 0;
 584                         buffer_info->page_offset = 0;
 585                 }
 586         }
 587
 588         size = sizeof(struct igbvf_buffer) * rx_ring->count;
 589         memset(rx_ring->buffer_info, 0, size);
 590
 591         /* Zero out the descriptor ring */
 592         memset(rx_ring->desc, 0, rx_ring->size);
 593
 594         rx_ring->next_to_clean = 0;
 595         rx_ring->next_to_use = 0;
 596
 597         writel(0, adapter->hw.hw_addr + rx_ring->head);
 598         writel(0, adapter->hw.hw_addr + rx_ring->tail);
 599 }
 600
 601 /**
 602  * igbvf_free_rx_resources - Free Rx Resources
 603  * @rx_ring: ring to clean the resources from
 604  *
 605  * Free all receive software resources
 606  **/
 607
 608 void igbvf_free_rx_resources(struct igbvf_ring *rx_ring)
 609 {
 610         struct pci_dev *pdev = rx_ring->adapter->pdev;
 611
 612         igbvf_clean_rx_ring(rx_ring);
 613
 614         vfree(rx_ring->buffer_info);
 615         rx_ring->buffer_info = NULL;
 616
 617         dma_free_coherent(&pdev->dev, rx_ring->size, rx_ring->desc,
 618                           rx_ring->dma);
 619         rx_ring->desc = NULL;
 620 }
 621
 622 /**
 623  * igbvf_update_itr - update the dynamic ITR value based on statistics
 624  * @adapter: pointer to adapter
 625  * @itr_setting: current adapter->itr
 626  * @packets: the number of packets during this measurement interval
 627  * @bytes: the number of bytes during this measurement interval
 628  *
 629  *      Stores a new ITR value based on packets and byte
 630  *      counts during the last interrupt.  The advantage of per interrupt
 631  *      computation is faster updates and more accurate ITR for the current
 632  *      traffic pattern.  Constants in this function were computed
 633  *      based on theoretical maximum wire speed and thresholds were set based
 634  *      on testing data as well as attempting to minimize response time
 635  *      while increasing bulk throughput.
 636  **/
 637 static enum latency_range igbvf_update_itr(struct igbvf_adapter *adapter,
 638                                            enum latency_range itr_setting,
 639                                            int packets, int bytes)
 640 {
 641         enum latency_range retval = itr_setting;
 642
 643         if (packets == 0)
 644                 goto update_itr_done;
 645
 646         switch (itr_setting) {
 647         case lowest_latency:
 648                 /* handle TSO and jumbo frames */
 649                 if (bytes/packets > 8000)
 650                         retval = bulk_latency;
 651                 else if ((packets < 5) && (bytes > 512))
 652                         retval = low_latency;
 653                 break;
 654         case low_latency:  /* 50 usec aka 20000 ints/s */
 655                 if (bytes > 10000) {
 656                         /* this if handles the TSO accounting */
 657                         if (bytes/packets > 8000)
 658                                 retval = bulk_latency;
 659                         else if ((packets < 10) || ((bytes/packets) > 1200))
 660                                 retval = bulk_latency;
 661                         else if ((packets > 35))
 662                                 retval = lowest_latency;
 663                 } else if (bytes/packets > 2000) {
 664                         retval = bulk_latency;
 665                 } else if (packets <= 2 && bytes < 512) {
 666                         retval = lowest_latency;
 667                 }
 668                 break;
 669         case bulk_latency: /* 250 usec aka 4000 ints/s */
 670                 if (bytes > 25000) {
 671                         if (packets > 35)
 672                                 retval = low_latency;
 673                 } else if (bytes < 6000) {
 674                         retval = low_latency;
 675                 }
 676                 break;
 677         default:
 678                 break;
 679         }
 680
 681 update_itr_done:
 682         return retval;
 683 }
 684
 685 static int igbvf_range_to_itr(enum latency_range current_range)
 686 {
 687         int new_itr;
 688
 689         switch (current_range) {
 690         /* counts and packets in update_itr are dependent on these numbers */
 691         case lowest_latency:
 692                 new_itr = IGBVF_70K_ITR;
 693                 break;
 694         case low_latency:
 695                 new_itr = IGBVF_20K_ITR;
 696                 break;
 697         case bulk_latency:
 698                 new_itr = IGBVF_4K_ITR;
 699                 break;
 700         default:
 701                 new_itr = IGBVF_START_ITR;
 702                 break;
 703         }
 704         return new_itr;
 705 }
 706
 707 static void igbvf_set_itr(struct igbvf_adapter *adapter)
 708 {
 709         u32 new_itr;
 710
 711         adapter->tx_ring->itr_range =
 712                         igbvf_update_itr(adapter,
 713                                          adapter->tx_ring->itr_val,
 714                                          adapter->total_tx_packets,
 715                                          adapter->total_tx_bytes);
 716
 717         /* conservative mode (itr 3) eliminates the lowest_latency setting */
 718         if (adapter->requested_itr == 3 &&
 719             adapter->tx_ring->itr_range == lowest_latency)
 720                 adapter->tx_ring->itr_range = low_latency;
 721
 722         new_itr = igbvf_range_to_itr(adapter->tx_ring->itr_range);
 723
 724
 725         if (new_itr != adapter->tx_ring->itr_val) {
 726                 u32 current_itr = adapter->tx_ring->itr_val;
 727                 /*
 728                  * this attempts to bias the interrupt rate towards Bulk
 729                  * by adding intermediate steps when interrupt rate is
 730                  * increasing
 731                  */
 732                 new_itr = new_itr > current_itr ?
 733                              min(current_itr + (new_itr >> 2), new_itr) :
 734                              new_itr;
 735                 adapter->tx_ring->itr_val = new_itr;
 736
 737                 adapter->tx_ring->set_itr = 1;
 738         }
 739
 740         adapter->rx_ring->itr_range =
 741                         igbvf_update_itr(adapter, adapter->rx_ring->itr_val,
 742                                          adapter->total_rx_packets,
 743                                          adapter->total_rx_bytes);
 744         if (adapter->requested_itr == 3 &&
 745             adapter->rx_ring->itr_range == lowest_latency)
 746                 adapter->rx_ring->itr_range = low_latency;
 747
 748         new_itr = igbvf_range_to_itr(adapter->rx_ring->itr_range);
 749
 750         if (new_itr != adapter->rx_ring->itr_val) {
 751                 u32 current_itr = adapter->rx_ring->itr_val;
 752                 new_itr = new_itr > current_itr ?
 753                              min(current_itr + (new_itr >> 2), new_itr) :
 754                              new_itr;
 755                 adapter->rx_ring->itr_val = new_itr;
 756
 757                 adapter->rx_ring->set_itr = 1;
 758         }
 759 }
 760
 761 /**
 762  * igbvf_clean_tx_irq - Reclaim resources after transmit completes
 763  * @adapter: board private structure
 764  * returns true if ring is completely cleaned
 765  **/
 766 static bool igbvf_clean_tx_irq(struct igbvf_ring *tx_ring)
 767 {
 768         struct igbvf_adapter *adapter = tx_ring->adapter;
 769         struct net_device *netdev = adapter->netdev;
 770         struct igbvf_buffer *buffer_info;
 771         struct sk_buff *skb;
 772         union e1000_adv_tx_desc *tx_desc, *eop_desc;
 773         unsigned int total_bytes = 0, total_packets = 0;
 774         unsigned int i, eop, count = 0;
 775         bool cleaned = false;
 776
 777         i = tx_ring->next_to_clean;
 778         eop = tx_ring->buffer_info[i].next_to_watch;
 779         eop_desc = IGBVF_TX_DESC_ADV(*tx_ring, eop);
 780
 781         while ((eop_desc->wb.status & cpu_to_le32(E1000_TXD_STAT_DD)) &&
 782                (count < tx_ring->count)) {
 783                 rmb();  /* read buffer_info after eop_desc status */
 784                 for (cleaned = false; !cleaned; count++) {
 785                         tx_desc = IGBVF_TX_DESC_ADV(*tx_ring, i);
 786                         buffer_info = &tx_ring->buffer_info[i];
 787                         cleaned = (i == eop);
 788                         skb = buffer_info->skb;
 789
 790                         if (skb) {
 791                                 unsigned int segs, bytecount;
 792
 793                                 /* gso_segs is currently only valid for tcp */
 794                                 segs = skb_shinfo(skb)->gso_segs ?: 1;
 795                                 /* multiply data chunks by size of headers */
 796                                 bytecount = ((segs - 1) * skb_headlen(skb)) +
 797                                             skb->len;
 798                                 total_packets += segs;
 799                                 total_bytes += bytecount;
 800                         }
 801
 802                         igbvf_put_txbuf(adapter, buffer_info);
 803                         tx_desc->wb.status = 0;
 804
 805                         i++;
 806                         if (i == tx_ring->count)
 807                                 i = 0;
 808                 }
 809                 eop = tx_ring->buffer_info[i].next_to_watch;
 810                 eop_desc = IGBVF_TX_DESC_ADV(*tx_ring, eop);
 811         }
 812
 813         tx_ring->next_to_clean = i;
 814
 815         if (unlikely(count &&
 816                      netif_carrier_ok(netdev) &&
 817                      igbvf_desc_unused(tx_ring) >= IGBVF_TX_QUEUE_WAKE)) {
 818                 /* Make sure that anybody stopping the queue after this
 819                  * sees the new next_to_clean.
 820                  */
 821                 smp_mb();
 822                 if (netif_queue_stopped(netdev) &&
 823                     !(test_bit(__IGBVF_DOWN, &adapter->state))) {
 824                         netif_wake_queue(netdev);
 825                         ++adapter->restart_queue;
 826                 }
 827         }
 828
 829         adapter->net_stats.tx_bytes += total_bytes;
 830         adapter->net_stats.tx_packets += total_packets;
 831         return count < tx_ring->count;
 832 }
 833
 834 static irqreturn_t igbvf_msix_other(int irq, void *data)
 835 {
 836         struct net_device *netdev = data;
 837         struct igbvf_adapter *adapter = netdev_priv(netdev);
 838         struct e1000_hw *hw = &adapter->hw;
 839
 840         adapter->int_counter1++;
 841
 842         netif_carrier_off(netdev);
 843         hw->mac.get_link_status = 1;
 844         if (!test_bit(__IGBVF_DOWN, &adapter->state))
 845                 mod_timer(&adapter->watchdog_timer, jiffies + 1);
 846
 847         ew32(EIMS, adapter->eims_other);
 848
 849         return IRQ_HANDLED;
 850 }
 851
 852 static irqreturn_t igbvf_intr_msix_tx(int irq, void *data)
 853 {
 854         struct net_device *netdev = data;
 855         struct igbvf_adapter *adapter = netdev_priv(netdev);
 856         struct e1000_hw *hw = &adapter->hw;
 857         struct igbvf_ring *tx_ring = adapter->tx_ring;
 858
 859         if (tx_ring->set_itr) {
 860                 writel(tx_ring->itr_val,
 861                        adapter->hw.hw_addr + tx_ring->itr_register);
 862                 adapter->tx_ring->set_itr = 0;
 863         }
 864
 865         adapter->total_tx_bytes = 0;
 866         adapter->total_tx_packets = 0;
 867
 868         /* auto mask will automatically reenable the interrupt when we write
 869          * EICS */
 870         if (!igbvf_clean_tx_irq(tx_ring))
 871                 /* Ring was not completely cleaned, so fire another interrupt */
 872                 ew32(EICS, tx_ring->eims_value);
 873         else
 874                 ew32(EIMS, tx_ring->eims_value);
 875
 876         return IRQ_HANDLED;
 877 }
 878
 879 static irqreturn_t igbvf_intr_msix_rx(int irq, void *data)
 880 {
 881         struct net_device *netdev = data;
 882         struct igbvf_adapter *adapter = netdev_priv(netdev);
 883
 884         adapter->int_counter0++;
 885
 886         /* Write the ITR value calculated at the end of the
 887          * previous interrupt.
 888          */
 889         if (adapter->rx_ring->set_itr) {
 890                 writel(adapter->rx_ring->itr_val,
 891                        adapter->hw.hw_addr + adapter->rx_ring->itr_register);
 892                 adapter->rx_ring->set_itr = 0;
 893         }
 894
 895         if (napi_schedule_prep(&adapter->rx_ring->napi)) {
 896                 adapter->total_rx_bytes = 0;
 897                 adapter->total_rx_packets = 0;
 898                 __napi_schedule(&adapter->rx_ring->napi);
 899         }
 900
 901         return IRQ_HANDLED;
 902 }
 903
 904 #define IGBVF_NO_QUEUE -1
 905
 906 static void igbvf_assign_vector(struct igbvf_adapter *adapter, int rx_queue,
 907                                 int tx_queue, int msix_vector)
 908 {
 909         struct e1000_hw *hw = &adapter->hw;
 910         u32 ivar, index;
 911
 912         /* 82576 uses a table-based method for assigning vectors.
 913            Each queue has a single entry in the table to which we write
 914            a vector number along with a "valid" bit.  Sadly, the layout
 915            of the table is somewhat counterintuitive. */
 916         if (rx_queue > IGBVF_NO_QUEUE) {
 917                 index = (rx_queue >> 1);
 918                 ivar = array_er32(IVAR0, index);
 919                 if (rx_queue & 0x1) {
 920                         /* vector goes into third byte of register */
 921                         ivar = ivar & 0xFF00FFFF;
 922                         ivar |= (msix_vector | E1000_IVAR_VALID) << 16;
 923                 } else {
 924                         /* vector goes into low byte of register */
 925                         ivar = ivar & 0xFFFFFF00;
 926                         ivar |= msix_vector | E1000_IVAR_VALID;
 927                 }
 928                 adapter->rx_ring[rx_queue].eims_value = 1 << msix_vector;
 929                 array_ew32(IVAR0, index, ivar);
 930         }
 931         if (tx_queue > IGBVF_NO_QUEUE) {
 932                 index = (tx_queue >> 1);
 933                 ivar = array_er32(IVAR0, index);
 934                 if (tx_queue & 0x1) {
 935                         /* vector goes into high byte of register */
 936                         ivar = ivar & 0x00FFFFFF;
 937                         ivar |= (msix_vector | E1000_IVAR_VALID) << 24;
 938                 } else {
 939                         /* vector goes into second byte of register */
 940                         ivar = ivar & 0xFFFF00FF;
 941                         ivar |= (msix_vector | E1000_IVAR_VALID) << 8;
 942                 }
 943                 adapter->tx_ring[tx_queue].eims_value = 1 << msix_vector;
 944                 array_ew32(IVAR0, index, ivar);
 945         }
 946 }
 947
 948 /**
 949  * igbvf_configure_msix - Configure MSI-X hardware
 950  *
 951  * igbvf_configure_msix sets up the hardware to properly
 952  * generate MSI-X interrupts.
 953  **/
 954 static void igbvf_configure_msix(struct igbvf_adapter *adapter)
 955 {
 956         u32 tmp;
 957         struct e1000_hw *hw = &adapter->hw;
 958         struct igbvf_ring *tx_ring = adapter->tx_ring;
 959         struct igbvf_ring *rx_ring = adapter->rx_ring;
 960         int vector = 0;
 961
 962         adapter->eims_enable_mask = 0;
 963
 964         igbvf_assign_vector(adapter, IGBVF_NO_QUEUE, 0, vector++);
 965         adapter->eims_enable_mask |= tx_ring->eims_value;
 966         writel(tx_ring->itr_val, hw->hw_addr + tx_ring->itr_register);
 967         igbvf_assign_vector(adapter, 0, IGBVF_NO_QUEUE, vector++);
 968         adapter->eims_enable_mask |= rx_ring->eims_value;
 969         writel(rx_ring->itr_val, hw->hw_addr + rx_ring->itr_register);
 970
 971         /* set vector for other causes, i.e. link changes */
 972
 973         tmp = (vector++ | E1000_IVAR_VALID);
 974
 975         ew32(IVAR_MISC, tmp);
 976
 977         adapter->eims_enable_mask = (1 << (vector)) - 1;
 978         adapter->eims_other = 1 << (vector - 1);
 979         e1e_flush();
 980 }
 981
 982 static void igbvf_reset_interrupt_capability(struct igbvf_adapter *adapter)
 983 {
 984         if (adapter->msix_entries) {
 985                 pci_disable_msix(adapter->pdev);
 986                 kfree(adapter->msix_entries);
 987                 adapter->msix_entries = NULL;
 988         }
 989 }
 990
 991 /**
 992  * igbvf_set_interrupt_capability - set MSI or MSI-X if supported
 993  *
 994  * Attempt to configure interrupts using the best available
 995  * capabilities of the hardware and kernel.
 996  **/
 997 static void igbvf_set_interrupt_capability(struct igbvf_adapter *adapter)
 998 {
 999         int err = -ENOMEM;
1000         int i;
1001
1002         /* we allocate 3 vectors, 1 for tx, 1 for rx, one for pf messages */
1003         adapter->msix_entries = kcalloc(3, sizeof(struct msix_entry),
1004                                         GFP_KERNEL);
1005         if (adapter->msix_entries) {
1006                 for (i = 0; i < 3; i++)
1007                         adapter->msix_entries[i].entry = i;
1008
1009                 err = pci_enable_msix(adapter->pdev,
1010                                       adapter->msix_entries, 3);
1011         }
1012
1013         if (err) {
1014                 /* MSI-X failed */
1015                 dev_err(&adapter->pdev->dev,
1016                         "Failed to initialize MSI-X interrupts.\n");
1017                 igbvf_reset_interrupt_capability(adapter);
1018         }
1019 }
1020
1021 /**
1022  * igbvf_request_msix - Initialize MSI-X interrupts
1023  *
1024  * igbvf_request_msix allocates MSI-X vectors and requests interrupts from the
1025  * kernel.
1026  **/
1027 static int igbvf_request_msix(struct igbvf_adapter *adapter)
1028 {
1029         struct net_device *netdev = adapter->netdev;
1030         int err = 0, vector = 0;
1031
1032         if (strlen(netdev->name) < (IFNAMSIZ - 5)) {
1033                 sprintf(adapter->tx_ring->name, "%s-tx-0", netdev->name);
1034                 sprintf(adapter->rx_ring->name, "%s-rx-0", netdev->name);
1035         } else {
1036                 memcpy(adapter->tx_ring->name, netdev->name, IFNAMSIZ);
1037                 memcpy(adapter->rx_ring->name, netdev->name, IFNAMSIZ);
1038         }
1039
1040         err = request_irq(adapter->msix_entries[vector].vector,
1041                           igbvf_intr_msix_tx, 0, adapter->tx_ring->name,
1042                           netdev);
1043         if (err)
1044                 goto out;
1045
1046         adapter->tx_ring->itr_register = E1000_EITR(vector);
1047         adapter->tx_ring->itr_val = adapter->current_itr;
1048         vector++;
1049
1050         err = request_irq(adapter->msix_entries[vector].vector,
1051                           igbvf_intr_msix_rx, 0, adapter->rx_ring->name,
1052                           netdev);
1053         if (err)
1054                 goto out;
1055
1056         adapter->rx_ring->itr_register = E1000_EITR(vector);
1057         adapter->rx_ring->itr_val = adapter->current_itr;
1058         vector++;
1059
1060         err = request_irq(adapter->msix_entries[vector].vector,
1061                           igbvf_msix_other, 0, netdev->name, netdev);
1062         if (err)
1063                 goto out;
1064
1065         igbvf_configure_msix(adapter);
1066         return 0;
1067 out:
1068         return err;
1069 }
1070
1071 /**
1072  * igbvf_alloc_queues - Allocate memory for all rings
1073  * @adapter: board private structure to initialize
1074  **/
1075 static int __devinit igbvf_alloc_queues(struct igbvf_adapter *adapter)
1076 {
1077         struct net_device *netdev = adapter->netdev;
1078
1079         adapter->tx_ring = kzalloc(sizeof(struct igbvf_ring), GFP_KERNEL);
1080         if (!adapter->tx_ring)
1081                 return -ENOMEM;
1082
1083         adapter->rx_ring = kzalloc(sizeof(struct igbvf_ring), GFP_KERNEL);
1084         if (!adapter->rx_ring) {
1085                 kfree(adapter->tx_ring);
1086                 return -ENOMEM;
1087         }
1088
1089         netif_napi_add(netdev, &adapter->rx_ring->napi, igbvf_poll, 64);
1090
1091         return 0;
1092 }
1093
1094 /**
1095  * igbvf_request_irq - initialize interrupts
1096  *
1097  * Attempts to configure interrupts using the best available
1098  * capabilities of the hardware and kernel.
1099  **/
1100 static int igbvf_request_irq(struct igbvf_adapter *adapter)
1101 {
1102         int err = -1;
1103
1104         /* igbvf supports msi-x only */
1105         if (adapter->msix_entries)
1106                 err = igbvf_request_msix(adapter);
1107
1108         if (!err)
1109                 return err;
1110
1111         dev_err(&adapter->pdev->dev,
1112                 "Unable to allocate interrupt, Error: %d\n", err);
1113
1114         return err;
1115 }
1116
1117 static void igbvf_free_irq(struct igbvf_adapter *adapter)
1118 {
1119         struct net_device *netdev = adapter->netdev;
1120         int vector;
1121
1122         if (adapter->msix_entries) {
1123                 for (vector = 0; vector < 3; vector++)
1124                         free_irq(adapter->msix_entries[vector].vector, netdev);
1125         }
1126 }
1127
1128 /**
1129  * igbvf_irq_disable - Mask off interrupt generation on the NIC
1130  **/
1131 static void igbvf_irq_disable(struct igbvf_adapter *adapter)
1132 {
1133         struct e1000_hw *hw = &adapter->hw;
1134
1135         ew32(EIMC, ~0);
1136
1137         if (adapter->msix_entries)
1138                 ew32(EIAC, 0);
1139 }
1140
1141 /**
1142  * igbvf_irq_enable - Enable default interrupt generation settings
1143  **/
1144 static void igbvf_irq_enable(struct igbvf_adapter *adapter)
1145 {
1146         struct e1000_hw *hw = &adapter->hw;
1147
1148         ew32(EIAC, adapter->eims_enable_mask);
1149         ew32(EIAM, adapter->eims_enable_mask);
1150         ew32(EIMS, adapter->eims_enable_mask);
1151 }
1152
1153 /**
1154  * igbvf_poll - NAPI Rx polling callback
1155  * @napi: struct associated with this polling callback
1156  * @budget: amount of packets driver is allowed to process this poll
1157  **/
1158 static int igbvf_poll(struct napi_struct *napi, int budget)
1159 {
1160         struct igbvf_ring *rx_ring = container_of(napi, struct igbvf_ring, napi);
1161         struct igbvf_adapter *adapter = rx_ring->adapter;
1162         struct e1000_hw *hw = &adapter->hw;
1163         int work_done = 0;
1164
1165         igbvf_clean_rx_irq(adapter, &work_done, budget);
1166
1167         /* If not enough Rx work done, exit the polling mode */
1168         if (work_done < budget) {
1169                 napi_complete(napi);
1170
1171                 if (adapter->requested_itr & 3)
1172                         igbvf_set_itr(adapter);
1173
1174                 if (!test_bit(__IGBVF_DOWN, &adapter->state))
1175                         ew32(EIMS, adapter->rx_ring->eims_value);
1176         }
1177
1178         return work_done;
1179 }
1180
1181 /**
1182  * igbvf_set_rlpml - set receive large packet maximum length
1183  * @adapter: board private structure
1184  *
1185  * Configure the maximum size of packets that will be received
1186  */
1187 static void igbvf_set_rlpml(struct igbvf_adapter *adapter)
1188 {
1189         int max_frame_size;
1190         struct e1000_hw *hw = &adapter->hw;
1191
1192         max_frame_size = adapter->max_frame_size + VLAN_TAG_SIZE;
1193         e1000_rlpml_set_vf(hw, max_frame_size);
1194 }
1195
1196 static int igbvf_vlan_rx_add_vid(struct net_device *netdev, u16 vid)
1197 {
1198         struct igbvf_adapter *adapter = netdev_priv(netdev);
1199         struct e1000_hw *hw = &adapter->hw;
1200
1201         if (hw->mac.ops.set_vfta(hw, vid, true)) {
1202                 dev_err(&adapter->pdev->dev, "Failed to add vlan id %d\n", vid);
1203                 return -EINVAL;
1204         }
1205         set_bit(vid, adapter->active_vlans);
1206         return 0;
1207 }
1208
1209 static int igbvf_vlan_rx_kill_vid(struct net_device *netdev, u16 vid)
1210 {
1211         struct igbvf_adapter *adapter = netdev_priv(netdev);
1212         struct e1000_hw *hw = &adapter->hw;
1213
1214         if (hw->mac.ops.set_vfta(hw, vid, false)) {
1215                 dev_err(&adapter->pdev->dev,
1216                         "Failed to remove vlan id %d\n", vid);
1217                 return -EINVAL;
1218         }
1219         clear_bit(vid, adapter->active_vlans);
1220         return 0;
1221 }
1222
1223 static void igbvf_restore_vlan(struct igbvf_adapter *adapter)
1224 {
1225         u16 vid;
1226
1227         for_each_set_bit(vid, adapter->active_vlans, VLAN_N_VID)
1228                 igbvf_vlan_rx_add_vid(adapter->netdev, vid);
1229 }
1230
1231 /**
1232  * igbvf_configure_tx - Configure Transmit Unit after Reset
1233  * @adapter: board private structure
1234  *
1235  * Configure the Tx unit of the MAC after a reset.
1236  **/
1237 static void igbvf_configure_tx(struct igbvf_adapter *adapter)
1238 {
1239         struct e1000_hw *hw = &adapter->hw;
1240         struct igbvf_ring *tx_ring = adapter->tx_ring;
1241         u64 tdba;
1242         u32 txdctl, dca_txctrl;
1243
1244         /* disable transmits */
1245         txdctl = er32(TXDCTL(0));
1246         ew32(TXDCTL(0), txdctl & ~E1000_TXDCTL_QUEUE_ENABLE);
1247         e1e_flush();
1248         msleep(10);
1249
1250         /* Setup the HW Tx Head and Tail descriptor pointers */
1251         ew32(TDLEN(0), tx_ring->count * sizeof(union e1000_adv_tx_desc));
1252         tdba = tx_ring->dma;
1253         ew32(TDBAL(0), (tdba & DMA_BIT_MASK(32)));
1254         ew32(TDBAH(0), (tdba >> 32));
1255         ew32(TDH(0), 0);
1256         ew32(TDT(0), 0);
1257         tx_ring->head = E1000_TDH(0);
1258         tx_ring->tail = E1000_TDT(0);
1259
1260         /* Turn off Relaxed Ordering on head write-backs.  The writebacks
1261          * MUST be delivered in order or it will completely screw up
1262          * our bookeeping.
1263          */
1264         dca_txctrl = er32(DCA_TXCTRL(0));
1265         dca_txctrl &= ~E1000_DCA_TXCTRL_TX_WB_RO_EN;
1266         ew32(DCA_TXCTRL(0), dca_txctrl);
1267
1268         /* enable transmits */
1269         txdctl |= E1000_TXDCTL_QUEUE_ENABLE;
1270         ew32(TXDCTL(0), txdctl);
1271
1272         /* Setup Transmit Descriptor Settings for eop descriptor */
1273         adapter->txd_cmd = E1000_ADVTXD_DCMD_EOP | E1000_ADVTXD_DCMD_IFCS;
1274
1275         /* enable Report Status bit */
1276         adapter->txd_cmd |= E1000_ADVTXD_DCMD_RS;
1277 }
1278
1279 /**
1280  * igbvf_setup_srrctl - configure the receive control registers
1281  * @adapter: Board private structure
1282  **/
1283 static void igbvf_setup_srrctl(struct igbvf_adapter *adapter)
1284 {
1285         struct e1000_hw *hw = &adapter->hw;
1286         u32 srrctl = 0;
1287
1288         srrctl &= ~(E1000_SRRCTL_DESCTYPE_MASK |
1289                     E1000_SRRCTL_BSIZEHDR_MASK |
1290                     E1000_SRRCTL_BSIZEPKT_MASK);
1291
1292         /* Enable queue drop to avoid head of line blocking */
1293         srrctl |= E1000_SRRCTL_DROP_EN;
1294
1295         /* Setup buffer sizes */
1296         srrctl |= ALIGN(adapter->rx_buffer_len, 1024) >>
1297                   E1000_SRRCTL_BSIZEPKT_SHIFT;
1298
1299         if (adapter->rx_buffer_len < 2048) {
1300                 adapter->rx_ps_hdr_size = 0;
1301                 srrctl |= E1000_SRRCTL_DESCTYPE_ADV_ONEBUF;
1302         } else {
1303                 adapter->rx_ps_hdr_size = 128;
1304                 srrctl |= adapter->rx_ps_hdr_size <<
1305                           E1000_SRRCTL_BSIZEHDRSIZE_SHIFT;
1306                 srrctl |= E1000_SRRCTL_DESCTYPE_HDR_SPLIT_ALWAYS;
1307         }
1308
1309         ew32(SRRCTL(0), srrctl);
1310 }
1311
1312 /**
1313  * igbvf_configure_rx - Configure Receive Unit after Reset
1314  * @adapter: board private structure
1315  *
1316  * Configure the Rx unit of the MAC after a reset.
1317  **/
1318 static void igbvf_configure_rx(struct igbvf_adapter *adapter)
1319 {
1320         struct e1000_hw *hw = &adapter->hw;
1321         struct igbvf_ring *rx_ring = adapter->rx_ring;
1322         u64 rdba;
1323         u32 rdlen, rxdctl;
1324
1325         /* disable receives */
1326         rxdctl = er32(RXDCTL(0));
1327         ew32(RXDCTL(0), rxdctl & ~E1000_RXDCTL_QUEUE_ENABLE);
1328         e1e_flush();
1329         msleep(10);
1330
1331         rdlen = rx_ring->count * sizeof(union e1000_adv_rx_desc);
1332
1333         /*
1334          * Setup the HW Rx Head and Tail Descriptor Pointers and
1335          * the Base and Length of the Rx Descriptor Ring
1336          */
1337         rdba = rx_ring->dma;
1338         ew32(RDBAL(0), (rdba & DMA_BIT_MASK(32)));
1339         ew32(RDBAH(0), (rdba >> 32));
1340         ew32(RDLEN(0), rx_ring->count * sizeof(union e1000_adv_rx_desc));
1341         rx_ring->head = E1000_RDH(0);
1342         rx_ring->tail = E1000_RDT(0);
1343         ew32(RDH(0), 0);
1344         ew32(RDT(0), 0);
1345
1346         rxdctl |= E1000_RXDCTL_QUEUE_ENABLE;
1347         rxdctl &= 0xFFF00000;
1348         rxdctl |= IGBVF_RX_PTHRESH;
1349         rxdctl |= IGBVF_RX_HTHRESH << 8;
1350         rxdctl |= IGBVF_RX_WTHRESH << 16;
1351
1352         igbvf_set_rlpml(adapter);
1353
1354         /* enable receives */
1355         ew32(RXDCTL(0), rxdctl);
1356 }
1357
1358 /**
1359  * igbvf_set_multi - Multicast and Promiscuous mode set
1360  * @netdev: network interface device structure
1361  *
1362  * The set_multi entry point is called whenever the multicast address
1363  * list or the network interface flags are updated.  This routine is
1364  * responsible for configuring the hardware for proper multicast,
1365  * promiscuous mode, and all-multi behavior.
1366  **/
1367 static void igbvf_set_multi(struct net_device *netdev)
1368 {
1369         struct igbvf_adapter *adapter = netdev_priv(netdev);
1370         struct e1000_hw *hw = &adapter->hw;
1371         struct netdev_hw_addr *ha;
1372         u8  *mta_list = NULL;
1373         int i;
1374
1375         if (!netdev_mc_empty(netdev)) {
1376                 mta_list = kmalloc(netdev_mc_count(netdev) * 6, GFP_ATOMIC);
1377                 if (!mta_list) {
1378                         dev_err(&adapter->pdev->dev,
1379                                 "failed to allocate multicast filter list\n");
1380                         return;
1381                 }
1382         }
1383
1384         /* prepare a packed array of only addresses. */
1385         i = 0;
1386         netdev_for_each_mc_addr(ha, netdev)
1387                 memcpy(mta_list + (i++ * ETH_ALEN), ha->addr, ETH_ALEN);
1388
1389         hw->mac.ops.update_mc_addr_list(hw, mta_list, i, 0, 0);
1390         kfree(mta_list);
1391 }
1392
1393 /**
1394  * igbvf_configure - configure the hardware for Rx and Tx
1395  * @adapter: private board structure
1396  **/
1397 static void igbvf_configure(struct igbvf_adapter *adapter)
1398 {
1399         igbvf_set_multi(adapter->netdev);
1400
1401         igbvf_restore_vlan(adapter);
1402
1403         igbvf_configure_tx(adapter);
1404         igbvf_setup_srrctl(adapter);
1405         igbvf_configure_rx(adapter);
1406         igbvf_alloc_rx_buffers(adapter->rx_ring,
1407                                igbvf_desc_unused(adapter->rx_ring));
1408 }
1409
1410 /* igbvf_reset - bring the hardware into a known good state
1411  *
1412  * This function boots the hardware and enables some settings that
1413  * require a configuration cycle of the hardware - those cannot be
1414  * set/changed during runtime. After reset the device needs to be
1415  * properly configured for Rx, Tx etc.
1416  */
1417 static void igbvf_reset(struct igbvf_adapter *adapter)
1418 {
1419         struct e1000_mac_info *mac = &adapter->hw.mac;
1420         struct net_device *netdev = adapter->netdev;
1421         struct e1000_hw *hw = &adapter->hw;
1422
1423         /* Allow time for pending master requests to run */
1424         if (mac->ops.reset_hw(hw))
1425                 dev_err(&adapter->pdev->dev, "PF still resetting\n");
1426
1427         mac->ops.init_hw(hw);
1428
1429         if (is_valid_ether_addr(adapter->hw.mac.addr)) {
1430                 memcpy(netdev->dev_addr, adapter->hw.mac.addr,
1431                        netdev->addr_len);
1432                 memcpy(netdev->perm_addr, adapter->hw.mac.addr,
1433                        netdev->addr_len);
1434         }
1435
1436         adapter->last_reset = jiffies;
1437 }
1438
1439 int igbvf_up(struct igbvf_adapter *adapter)
1440 {
1441         struct e1000_hw *hw = &adapter->hw;
1442
1443         /* hardware has been reset, we need to reload some things */
1444         igbvf_configure(adapter);
1445
1446         clear_bit(__IGBVF_DOWN, &adapter->state);
1447
1448         napi_enable(&adapter->rx_ring->napi);
1449         if (adapter->msix_entries)
1450                 igbvf_configure_msix(adapter);
1451
1452         /* Clear any pending interrupts. */
1453         er32(EICR);
1454         igbvf_irq_enable(adapter);
1455
1456         /* start the watchdog */
1457         hw->mac.get_link_status = 1;
1458         mod_timer(&adapter->watchdog_timer, jiffies + 1);
1459
1460
1461         return 0;
1462 }
1463
1464 void igbvf_down(struct igbvf_adapter *adapter)
1465 {
1466         struct net_device *netdev = adapter->netdev;
1467         struct e1000_hw *hw = &adapter->hw;
1468         u32 rxdctl, txdctl;
1469
1470         /*
1471          * signal that we're down so the interrupt handler does not
1472          * reschedule our watchdog timer
1473          */
1474         set_bit(__IGBVF_DOWN, &adapter->state);
1475
1476         /* disable receives in the hardware */
1477         rxdctl = er32(RXDCTL(0));
1478         ew32(RXDCTL(0), rxdctl & ~E1000_RXDCTL_QUEUE_ENABLE);
1479
1480         netif_stop_queue(netdev);
1481
1482         /* disable transmits in the hardware */
1483         txdctl = er32(TXDCTL(0));
1484         ew32(TXDCTL(0), txdctl & ~E1000_TXDCTL_QUEUE_ENABLE);
1485
1486         /* flush both disables and wait for them to finish */
1487         e1e_flush();
1488         msleep(10);
1489
1490         napi_disable(&adapter->rx_ring->napi);
1491
1492         igbvf_irq_disable(adapter);
1493
1494         del_timer_sync(&adapter->watchdog_timer);
1495
1496         netif_carrier_off(netdev);
1497
1498         /* record the stats before reset*/
1499         igbvf_update_stats(adapter);
1500
1501         adapter->link_speed = 0;
1502         adapter->link_duplex = 0;
1503
1504         igbvf_reset(adapter);
1505         igbvf_clean_tx_ring(adapter->tx_ring);
1506         igbvf_clean_rx_ring(adapter->rx_ring);
1507 }
1508
1509 void igbvf_reinit_locked(struct igbvf_adapter *adapter)
1510 {
1511         might_sleep();
1512         while (test_and_set_bit(__IGBVF_RESETTING, &adapter->state))
1513                 msleep(1);
1514         igbvf_down(adapter);
1515         igbvf_up(adapter);
1516         clear_bit(__IGBVF_RESETTING, &adapter->state);
1517 }
1518
1519 /**
1520  * igbvf_sw_init - Initialize general software structures (struct igbvf_adapter)
1521  * @adapter: board private structure to initialize
1522  *
1523  * igbvf_sw_init initializes the Adapter private data structure.
1524  * Fields are initialized based on PCI device information and
1525  * OS network device settings (MTU size).
1526  **/
1527 static int __devinit igbvf_sw_init(struct igbvf_adapter *adapter)
1528 {
1529         struct net_device *netdev = adapter->netdev;
1530         s32 rc;
1531
1532         adapter->rx_buffer_len = ETH_FRAME_LEN + VLAN_HLEN + ETH_FCS_LEN;
1533         adapter->rx_ps_hdr_size = 0;
1534         adapter->max_frame_size = netdev->mtu + ETH_HLEN + ETH_FCS_LEN;
1535         adapter->min_frame_size = ETH_ZLEN + ETH_FCS_LEN;
1536
1537         adapter->tx_int_delay = 8;
1538         adapter->tx_abs_int_delay = 32;
1539         adapter->rx_int_delay = 0;
1540         adapter->rx_abs_int_delay = 8;
1541         adapter->requested_itr = 3;
1542         adapter->current_itr = IGBVF_START_ITR;
1543
1544         /* Set various function pointers */
1545         adapter->ei->init_ops(&adapter->hw);
1546
1547         rc = adapter->hw.mac.ops.init_params(&adapter->hw);
1548         if (rc)
1549                 return rc;
1550
1551         rc = adapter->hw.mbx.ops.init_params(&adapter->hw);
1552         if (rc)
1553                 return rc;
1554
1555         igbvf_set_interrupt_capability(adapter);
1556
1557         if (igbvf_alloc_queues(adapter))
1558                 return -ENOMEM;
1559
1560         spin_lock_init(&adapter->tx_queue_lock);
1561
1562         /* Explicitly disable IRQ since the NIC can be in any state. */
1563         igbvf_irq_disable(adapter);
1564
1565         spin_lock_init(&adapter->stats_lock);
1566
1567         set_bit(__IGBVF_DOWN, &adapter->state);
1568         return 0;
1569 }
1570
1571 static void igbvf_initialize_last_counter_stats(struct igbvf_adapter *adapter)
1572 {
1573         struct e1000_hw *hw = &adapter->hw;
1574
1575         adapter->stats.last_gprc = er32(VFGPRC);
1576         adapter->stats.last_gorc = er32(VFGORC);
1577         adapter->stats.last_gptc = er32(VFGPTC);
1578         adapter->stats.last_gotc = er32(VFGOTC);
1579         adapter->stats.last_mprc = er32(VFMPRC);
1580         adapter->stats.last_gotlbc = er32(VFGOTLBC);
1581         adapter->stats.last_gptlbc = er32(VFGPTLBC);
1582         adapter->stats.last_gorlbc = er32(VFGORLBC);
1583         adapter->stats.last_gprlbc = er32(VFGPRLBC);
1584
1585         adapter->stats.base_gprc = er32(VFGPRC);
1586         adapter->stats.base_gorc = er32(VFGORC);
1587         adapter->stats.base_gptc = er32(VFGPTC);
1588         adapter->stats.base_gotc = er32(VFGOTC);
1589         adapter->stats.base_mprc = er32(VFMPRC);
1590         adapter->stats.base_gotlbc = er32(VFGOTLBC);
1591         adapter->stats.base_gptlbc = er32(VFGPTLBC);
1592         adapter->stats.base_gorlbc = er32(VFGORLBC);
1593         adapter->stats.base_gprlbc = er32(VFGPRLBC);
1594 }
1595
1596 /**
1597  * igbvf_open - Called when a network interface is made active
1598  * @netdev: network interface device structure
1599  *
1600  * Returns 0 on success, negative value on failure
1601  *
1602  * The open entry point is called when a network interface is made
1603  * active by the system (IFF_UP).  At this point all resources needed
1604  * for transmit and receive operations are allocated, the interrupt
1605  * handler is registered with the OS, the watchdog timer is started,
1606  * and the stack is notified that the interface is ready.
1607  **/
1608 static int igbvf_open(struct net_device *netdev)
1609 {
1610         struct igbvf_adapter *adapter = netdev_priv(netdev);
1611         struct e1000_hw *hw = &adapter->hw;
1612         int err;
1613
1614         /* disallow open during test */
1615         if (test_bit(__IGBVF_TESTING, &adapter->state))
1616                 return -EBUSY;
1617
1618         /* allocate transmit descriptors */
1619         err = igbvf_setup_tx_resources(adapter, adapter->tx_ring);
1620         if (err)
1621                 goto err_setup_tx;
1622
1623         /* allocate receive descriptors */
1624         err = igbvf_setup_rx_resources(adapter, adapter->rx_ring);
1625         if (err)
1626                 goto err_setup_rx;
1627
1628         /*
1629          * before we allocate an interrupt, we must be ready to handle it.
1630          * Setting DEBUG_SHIRQ in the kernel makes it fire an interrupt
1631          * as soon as we call pci_request_irq, so we have to setup our
1632          * clean_rx handler before we do so.
1633          */
1634         igbvf_configure(adapter);
1635
1636         err = igbvf_request_irq(adapter);
1637         if (err)
1638                 goto err_req_irq;
1639
1640         /* From here on the code is the same as igbvf_up() */
1641         clear_bit(__IGBVF_DOWN, &adapter->state);
1642
1643         napi_enable(&adapter->rx_ring->napi);
1644
1645         /* clear any pending interrupts */
1646         er32(EICR);
1647
1648         igbvf_irq_enable(adapter);
1649
1650         /* start the watchdog */
1651         hw->mac.get_link_status = 1;
1652         mod_timer(&adapter->watchdog_timer, jiffies + 1);
1653
1654         return 0;
1655
1656 err_req_irq:
1657         igbvf_free_rx_resources(adapter->rx_ring);
1658 err_setup_rx:
1659         igbvf_free_tx_resources(adapter->tx_ring);
1660 err_setup_tx:
1661         igbvf_reset(adapter);
1662
1663         return err;
1664 }
1665
1666 /**
1667  * igbvf_close - Disables a network interface
1668  * @netdev: network interface device structure
1669  *
1670  * Returns 0, this is not allowed to fail
1671  *
1672  * The close entry point is called when an interface is de-activated
1673  * by the OS.  The hardware is still under the drivers control, but
1674  * needs to be disabled.  A global MAC reset is issued to stop the
1675  * hardware, and all transmit and receive resources are freed.
1676  **/
1677 static int igbvf_close(struct net_device *netdev)
1678 {
1679         struct igbvf_adapter *adapter = netdev_priv(netdev);
1680
1681         WARN_ON(test_bit(__IGBVF_RESETTING, &adapter->state));
1682         igbvf_down(adapter);
1683
1684         igbvf_free_irq(adapter);
1685
1686         igbvf_free_tx_resources(adapter->tx_ring);
1687         igbvf_free_rx_resources(adapter->rx_ring);
1688
1689         return 0;
1690 }
1691 /**
1692  * igbvf_set_mac - Change the Ethernet Address of the NIC
1693  * @netdev: network interface device structure
1694  * @p: pointer to an address structure
1695  *
1696  * Returns 0 on success, negative on failure
1697  **/
1698 static int igbvf_set_mac(struct net_device *netdev, void *p)
1699 {
1700         struct igbvf_adapter *adapter = netdev_priv(netdev);
1701         struct e1000_hw *hw = &adapter->hw;
1702         struct sockaddr *addr = p;
1703
1704         if (!is_valid_ether_addr(addr->sa_data))
1705                 return -EADDRNOTAVAIL;
1706
1707         memcpy(hw->mac.addr, addr->sa_data, netdev->addr_len);
1708
1709         hw->mac.ops.rar_set(hw, hw->mac.addr, 0);
1710
1711         if (memcmp(addr->sa_data, hw->mac.addr, 6))
1712                 return -EADDRNOTAVAIL;
1713
1714         memcpy(netdev->dev_addr, addr->sa_data, netdev->addr_len);
1715
1716         return 0;
1717 }
1718
1719 #define UPDATE_VF_COUNTER(reg, name)                                    \
1720         {                                                               \
1721                 u32 current_counter = er32(reg);                        \
1722                 if (current_counter < adapter->stats.last_##name)       \
1723                         adapter->stats.name += 0x100000000LL;           \
1724                 adapter->stats.last_##name = current_counter;           \
1725                 adapter->stats.name &= 0xFFFFFFFF00000000LL;            \
1726                 adapter->stats.name |= current_counter;                 \
1727         }
1728
1729 /**
1730  * igbvf_update_stats - Update the board statistics counters
1731  * @adapter: board private structure
1732 **/
1733 void igbvf_update_stats(struct igbvf_adapter *adapter)
1734 {
1735         struct e1000_hw *hw = &adapter->hw;
1736         struct pci_dev *pdev = adapter->pdev;
1737
1738         /*
1739          * Prevent stats update while adapter is being reset, link is down
1740          * or if the pci connection is down.
1741          */
1742         if (adapter->link_speed == 0)
1743                 return;
1744
1745         if (test_bit(__IGBVF_RESETTING, &adapter->state))
1746                 return;
1747
1748         if (pci_channel_offline(pdev))
1749                 return;
1750
1751         UPDATE_VF_COUNTER(VFGPRC, gprc);
1752         UPDATE_VF_COUNTER(VFGORC, gorc);
1753         UPDATE_VF_COUNTER(VFGPTC, gptc);
1754         UPDATE_VF_COUNTER(VFGOTC, gotc);
1755         UPDATE_VF_COUNTER(VFMPRC, mprc);
1756         UPDATE_VF_COUNTER(VFGOTLBC, gotlbc);
1757         UPDATE_VF_COUNTER(VFGPTLBC, gptlbc);
1758         UPDATE_VF_COUNTER(VFGORLBC, gorlbc);
1759         UPDATE_VF_COUNTER(VFGPRLBC, gprlbc);
1760
1761         /* Fill out the OS statistics structure */
1762         adapter->net_stats.multicast = adapter->stats.mprc;
1763 }
1764
1765 static void igbvf_print_link_info(struct igbvf_adapter *adapter)
1766 {
1767         dev_info(&adapter->pdev->dev, "Link is Up %d Mbps %s Duplex\n",
1768                  adapter->link_speed,
1769                  adapter->link_duplex == FULL_DUPLEX ? "Full" : "Half");
1770 }
1771
1772 static bool igbvf_has_link(struct igbvf_adapter *adapter)
1773 {
1774         struct e1000_hw *hw = &adapter->hw;
1775         s32 ret_val = E1000_SUCCESS;
1776         bool link_active;
1777
1778         /* If interface is down, stay link down */
1779         if (test_bit(__IGBVF_DOWN, &adapter->state))
1780                 return false;
1781
1782         ret_val = hw->mac.ops.check_for_link(hw);
1783         link_active = !hw->mac.get_link_status;
1784
1785         /* if check for link returns error we will need to reset */
1786         if (ret_val && time_after(jiffies, adapter->last_reset + (10 * HZ)))
1787                 schedule_work(&adapter->reset_task);
1788
1789         return link_active;
1790 }
1791
1792 /**
1793  * igbvf_watchdog - Timer Call-back
1794  * @data: pointer to adapter cast into an unsigned long
1795  **/
1796 static void igbvf_watchdog(unsigned long data)
1797 {
1798         struct igbvf_adapter *adapter = (struct igbvf_adapter *) data;
1799
1800         /* Do the rest outside of interrupt context */
1801         schedule_work(&adapter->watchdog_task);
1802 }
1803
1804 static void igbvf_watchdog_task(struct work_struct *work)
1805 {
1806         struct igbvf_adapter *adapter = container_of(work,
1807                                                      struct igbvf_adapter,
1808                                                      watchdog_task);
1809         struct net_device *netdev = adapter->netdev;
1810         struct e1000_mac_info *mac = &adapter->hw.mac;
1811         struct igbvf_ring *tx_ring = adapter->tx_ring;
1812         struct e1000_hw *hw = &adapter->hw;
1813         u32 link;
1814         int tx_pending = 0;
1815
1816         link = igbvf_has_link(adapter);
1817
1818         if (link) {
1819                 if (!netif_carrier_ok(netdev)) {
1820                         mac->ops.get_link_up_info(&adapter->hw,
1821                                                   &adapter->link_speed,
1822                                                   &adapter->link_duplex);
1823                         igbvf_print_link_info(adapter);
1824
1825                         netif_carrier_on(netdev);
1826                         netif_wake_queue(netdev);
1827                 }
1828         } else {
1829                 if (netif_carrier_ok(netdev)) {
1830                         adapter->link_speed = 0;
1831                         adapter->link_duplex = 0;
1832                         dev_info(&adapter->pdev->dev, "Link is Down\n");
1833                         netif_carrier_off(netdev);
1834                         netif_stop_queue(netdev);
1835                 }
1836         }
1837
1838         if (netif_carrier_ok(netdev)) {
1839                 igbvf_update_stats(adapter);
1840         } else {
1841                 tx_pending = (igbvf_desc_unused(tx_ring) + 1 <
1842                               tx_ring->count);
1843                 if (tx_pending) {
1844                         /*
1845                          * We've lost link, so the controller stops DMA,
1846                          * but we've got queued Tx work that's never going
1847                          * to get done, so reset controller to flush Tx.
1848                          * (Do the reset outside of interrupt context).
1849                          */
1850                         adapter->tx_timeout_count++;
1851                         schedule_work(&adapter->reset_task);
1852                 }
1853         }
1854
1855         /* Cause software interrupt to ensure Rx ring is cleaned */
1856         ew32(EICS, adapter->rx_ring->eims_value);
1857
1858         /* Reset the timer */
1859         if (!test_bit(__IGBVF_DOWN, &adapter->state))
1860                 mod_timer(&adapter->watchdog_timer,
1861                           round_jiffies(jiffies + (2 * HZ)));
1862 }
1863
1864 #define IGBVF_TX_FLAGS_CSUM             0x00000001
1865 #define IGBVF_TX_FLAGS_VLAN             0x00000002
1866 #define IGBVF_TX_FLAGS_TSO              0x00000004
1867 #define IGBVF_TX_FLAGS_IPV4             0x00000008
1868 #define IGBVF_TX_FLAGS_VLAN_MASK        0xffff0000
1869 #define IGBVF_TX_FLAGS_VLAN_SHIFT       16
1870
1871 static int igbvf_tso(struct igbvf_adapter *adapter,
1872                      struct igbvf_ring *tx_ring,
1873                      struct sk_buff *skb, u32 tx_flags, u8 *hdr_len)
1874 {
1875         struct e1000_adv_tx_context_desc *context_desc;
1876         unsigned int i;
1877         int err;
1878         struct igbvf_buffer *buffer_info;
1879         u32 info = 0, tu_cmd = 0;
1880         u32 mss_l4len_idx, l4len;
1881         *hdr_len = 0;
1882
1883         if (skb_header_cloned(skb)) {
1884                 err = pskb_expand_head(skb, 0, 0, GFP_ATOMIC);
1885                 if (err) {
1886                         dev_err(&adapter->pdev->dev,
1887                                 "igbvf_tso returning an error\n");
1888                         return err;
1889                 }
1890         }
1891
1892         l4len = tcp_hdrlen(skb);
1893         *hdr_len += l4len;
1894
1895         if (skb->protocol == htons(ETH_P_IP)) {
1896                 struct iphdr *iph = ip_hdr(skb);
1897                 iph->tot_len = 0;
1898                 iph->check = 0;
1899                 tcp_hdr(skb)->check = ~csum_tcpudp_magic(iph->saddr,
1900                                                          iph->daddr, 0,
1901                                                          IPPROTO_TCP,
1902                                                          0);
1903         } else if (skb_is_gso_v6(skb)) {
1904                 ipv6_hdr(skb)->payload_len = 0;
1905                 tcp_hdr(skb)->check = ~csum_ipv6_magic(&ipv6_hdr(skb)->saddr,
1906                                                        &ipv6_hdr(skb)->daddr,
1907                                                        0, IPPROTO_TCP, 0);
1908         }
1909
1910         i = tx_ring->next_to_use;
1911
1912         buffer_info = &tx_ring->buffer_info[i];
1913         context_desc = IGBVF_TX_CTXTDESC_ADV(*tx_ring, i);
1914         /* VLAN MACLEN IPLEN */
1915         if (tx_flags & IGBVF_TX_FLAGS_VLAN)
1916                 info |= (tx_flags & IGBVF_TX_FLAGS_VLAN_MASK);
1917         info |= (skb_network_offset(skb) << E1000_ADVTXD_MACLEN_SHIFT);
1918         *hdr_len += skb_network_offset(skb);
1919         info |= (skb_transport_header(skb) - skb_network_header(skb));
1920         *hdr_len += (skb_transport_header(skb) - skb_network_header(skb));
1921         context_desc->vlan_macip_lens = cpu_to_le32(info);
1922
1923         /* ADV DTYP TUCMD MKRLOC/ISCSIHEDLEN */
1924         tu_cmd |= (E1000_TXD_CMD_DEXT | E1000_ADVTXD_DTYP_CTXT);
1925
1926         if (skb->protocol == htons(ETH_P_IP))
1927                 tu_cmd |= E1000_ADVTXD_TUCMD_IPV4;
1928         tu_cmd |= E1000_ADVTXD_TUCMD_L4T_TCP;
1929
1930         context_desc->type_tucmd_mlhl = cpu_to_le32(tu_cmd);
1931
1932         /* MSS L4LEN IDX */
1933         mss_l4len_idx = (skb_shinfo(skb)->gso_size << E1000_ADVTXD_MSS_SHIFT);
1934         mss_l4len_idx |= (l4len << E1000_ADVTXD_L4LEN_SHIFT);
1935
1936         context_desc->mss_l4len_idx = cpu_to_le32(mss_l4len_idx);
1937         context_desc->seqnum_seed = 0;
1938
1939         buffer_info->time_stamp = jiffies;
1940         buffer_info->next_to_watch = i;
1941         buffer_info->dma = 0;
1942         i++;
1943         if (i == tx_ring->count)
1944                 i = 0;
1945
1946         tx_ring->next_to_use = i;
1947
1948         return true;
1949 }
1950
1951 static inline bool igbvf_tx_csum(struct igbvf_adapter *adapter,
1952                                  struct igbvf_ring *tx_ring,
1953                                  struct sk_buff *skb, u32 tx_flags)
1954 {
1955         struct e1000_adv_tx_context_desc *context_desc;
1956         unsigned int i;
1957         struct igbvf_buffer *buffer_info;
1958         u32 info = 0, tu_cmd = 0;
1959
1960         if ((skb->ip_summed == CHECKSUM_PARTIAL) ||
1961             (tx_flags & IGBVF_TX_FLAGS_VLAN)) {
1962                 i = tx_ring->next_to_use;
1963                 buffer_info = &tx_ring->buffer_info[i];
1964                 context_desc = IGBVF_TX_CTXTDESC_ADV(*tx_ring, i);
1965
1966                 if (tx_flags & IGBVF_TX_FLAGS_VLAN)
1967                         info |= (tx_flags & IGBVF_TX_FLAGS_VLAN_MASK);
1968
1969                 info |= (skb_network_offset(skb) << E1000_ADVTXD_MACLEN_SHIFT);
1970                 if (skb->ip_summed == CHECKSUM_PARTIAL)
1971                         info |= (skb_transport_header(skb) -
1972                                  skb_network_header(skb));
1973
1974
1975                 context_desc->vlan_macip_lens = cpu_to_le32(info);
1976
1977                 tu_cmd |= (E1000_TXD_CMD_DEXT | E1000_ADVTXD_DTYP_CTXT);
1978
1979                 if (skb->ip_summed == CHECKSUM_PARTIAL) {
1980                         switch (skb->protocol) {
1981                         case __constant_htons(ETH_P_IP):
1982                                 tu_cmd |= E1000_ADVTXD_TUCMD_IPV4;
1983                                 if (ip_hdr(skb)->protocol == IPPROTO_TCP)
1984                                         tu_cmd |= E1000_ADVTXD_TUCMD_L4T_TCP;
1985                                 break;
1986                         case __constant_htons(ETH_P_IPV6):
1987                                 if (ipv6_hdr(skb)->nexthdr == IPPROTO_TCP)
1988                                         tu_cmd |= E1000_ADVTXD_TUCMD_L4T_TCP;
1989                                 break;
1990                         default:
1991                                 break;
1992                         }
1993                 }
1994
1995                 context_desc->type_tucmd_mlhl = cpu_to_le32(tu_cmd);
1996                 context_desc->seqnum_seed = 0;
1997                 context_desc->mss_l4len_idx = 0;
1998
1999                 buffer_info->time_stamp = jiffies;
2000                 buffer_info->next_to_watch = i;
2001                 buffer_info->dma = 0;
2002                 i++;
2003                 if (i == tx_ring->count)
2004                         i = 0;
2005                 tx_ring->next_to_use = i;
2006
2007                 return true;
2008         }
2009
2010         return false;
2011 }
2012
2013 static int igbvf_maybe_stop_tx(struct net_device *netdev, int size)
2014 {
2015         struct igbvf_adapter *adapter = netdev_priv(netdev);
2016
2017         /* there is enough descriptors then we don't need to worry  */
2018         if (igbvf_desc_unused(adapter->tx_ring) >= size)
2019                 return 0;
2020
2021         netif_stop_queue(netdev);
2022
2023         smp_mb();
2024
2025         /* We need to check again just in case room has been made available */
2026         if (igbvf_desc_unused(adapter->tx_ring) < size)
2027                 return -EBUSY;
2028
2029         netif_wake_queue(netdev);
2030
2031         ++adapter->restart_queue;
2032         return 0;
2033 }
2034
2035 #define IGBVF_MAX_TXD_PWR       16
2036 #define IGBVF_MAX_DATA_PER_TXD  (1 << IGBVF_MAX_TXD_PWR)
2037
2038 static inline int igbvf_tx_map_adv(struct igbvf_adapter *adapter,
2039                                    struct igbvf_ring *tx_ring,
2040                                    struct sk_buff *skb,
2041                                    unsigned int first)
2042 {
2043         struct igbvf_buffer *buffer_info;
2044         struct pci_dev *pdev = adapter->pdev;
2045         unsigned int len = skb_headlen(skb);
2046         unsigned int count = 0, i;
2047         unsigned int f;
2048
2049         i = tx_ring->next_to_use;
2050
2051         buffer_info = &tx_ring->buffer_info[i];
2052         BUG_ON(len >= IGBVF_MAX_DATA_PER_TXD);
2053         buffer_info->length = len;
2054         /* set time_stamp *before* dma to help avoid a possible race */
2055         buffer_info->time_stamp = jiffies;
2056         buffer_info->next_to_watch = i;
2057         buffer_info->mapped_as_page = false;
2058         buffer_info->dma = dma_map_single(&pdev->dev, skb->data, len,
2059                                           DMA_TO_DEVICE);
2060         if (dma_mapping_error(&pdev->dev, buffer_info->dma))
2061                 goto dma_error;
2062
2063
2064         for (f = 0; f < skb_shinfo(skb)->nr_frags; f++) {
2065                 const struct skb_frag_struct *frag;
2066
2067                 count++;
2068                 i++;
2069                 if (i == tx_ring->count)
2070                         i = 0;
2071
2072                 frag = &skb_shinfo(skb)->frags[f];
2073                 len = skb_frag_size(frag);
2074
2075                 buffer_info = &tx_ring->buffer_info[i];
2076                 BUG_ON(len >= IGBVF_MAX_DATA_PER_TXD);
2077                 buffer_info->length = len;
2078                 buffer_info->time_stamp = jiffies;
2079                 buffer_info->next_to_watch = i;
2080                 buffer_info->mapped_as_page = true;
2081                 buffer_info->dma = skb_frag_dma_map(&pdev->dev, frag, 0, len,
2082                                                 DMA_TO_DEVICE);
2083                 if (dma_mapping_error(&pdev->dev, buffer_info->dma))
2084                         goto dma_error;
2085         }
2086
2087         tx_ring->buffer_info[i].skb = skb;
2088         tx_ring->buffer_info[first].next_to_watch = i;
2089
2090         return ++count;
2091
2092 dma_error:
2093         dev_err(&pdev->dev, "TX DMA map failed\n");
2094
2095         /* clear timestamp and dma mappings for failed buffer_info mapping */
2096         buffer_info->dma = 0;
2097         buffer_info->time_stamp = 0;
2098         buffer_info->length = 0;
2099         buffer_info->next_to_watch = 0;
2100         buffer_info->mapped_as_page = false;
2101         if (count)
2102                 count--;
2103
2104         /* clear timestamp and dma mappings for remaining portion of packet */
2105         while (count--) {
2106                 if (i==0)
2107                         i += tx_ring->count;
2108                 i--;
2109                 buffer_info = &tx_ring->buffer_info[i];
2110                 igbvf_put_txbuf(adapter, buffer_info);
2111         }
2112
2113         return 0;
2114 }
2115
2116 static inline void igbvf_tx_queue_adv(struct igbvf_adapter *adapter,
2117                                       struct igbvf_ring *tx_ring,
2118                                       int tx_flags, int count, u32 paylen,
2119                                       u8 hdr_len)
2120 {
2121         union e1000_adv_tx_desc *tx_desc = NULL;
2122         struct igbvf_buffer *buffer_info;
2123         u32 olinfo_status = 0, cmd_type_len;
2124         unsigned int i;
2125
2126         cmd_type_len = (E1000_ADVTXD_DTYP_DATA | E1000_ADVTXD_DCMD_IFCS |
2127                         E1000_ADVTXD_DCMD_DEXT);
2128
2129         if (tx_flags & IGBVF_TX_FLAGS_VLAN)
2130                 cmd_type_len |= E1000_ADVTXD_DCMD_VLE;
2131
2132         if (tx_flags & IGBVF_TX_FLAGS_TSO) {
2133                 cmd_type_len |= E1000_ADVTXD_DCMD_TSE;
2134
2135                 /* insert tcp checksum */
2136                 olinfo_status |= E1000_TXD_POPTS_TXSM << 8;
2137
2138                 /* insert ip checksum */
2139                 if (tx_flags & IGBVF_TX_FLAGS_IPV4)
2140                         olinfo_status |= E1000_TXD_POPTS_IXSM << 8;
2141
2142         } else if (tx_flags & IGBVF_TX_FLAGS_CSUM) {
2143                 olinfo_status |= E1000_TXD_POPTS_TXSM << 8;
2144         }
2145
2146         olinfo_status |= ((paylen - hdr_len) << E1000_ADVTXD_PAYLEN_SHIFT);
2147
2148         i = tx_ring->next_to_use;
2149         while (count--) {
2150                 buffer_info = &tx_ring->buffer_info[i];
2151                 tx_desc = IGBVF_TX_DESC_ADV(*tx_ring, i);
2152                 tx_desc->read.buffer_addr = cpu_to_le64(buffer_info->dma);
2153                 tx_desc->read.cmd_type_len =
2154                          cpu_to_le32(cmd_type_len | buffer_info->length);
2155                 tx_desc->read.olinfo_status = cpu_to_le32(olinfo_status);
2156                 i++;
2157                 if (i == tx_ring->count)
2158                         i = 0;
2159         }
2160
2161         tx_desc->read.cmd_type_len |= cpu_to_le32(adapter->txd_cmd);
2162         /* Force memory writes to complete before letting h/w
2163          * know there are new descriptors to fetch.  (Only
2164          * applicable for weak-ordered memory model archs,
2165          * such as IA-64). */
2166         wmb();
2167
2168         tx_ring->next_to_use = i;
2169         writel(i, adapter->hw.hw_addr + tx_ring->tail);
2170         /* we need this if more than one processor can write to our tail
2171          * at a time, it syncronizes IO on IA64/Altix systems */
2172         mmiowb();
2173 }
2174
2175 static netdev_tx_t igbvf_xmit_frame_ring_adv(struct sk_buff *skb,
2176                                              struct net_device *netdev,
2177                                              struct igbvf_ring *tx_ring)
2178 {
2179         struct igbvf_adapter *adapter = netdev_priv(netdev);
2180         unsigned int first, tx_flags = 0;
2181         u8 hdr_len = 0;
2182         int count = 0;
2183         int tso = 0;
2184
2185         if (test_bit(__IGBVF_DOWN, &adapter->state)) {
2186                 dev_kfree_skb_any(skb);
2187                 return NETDEV_TX_OK;
2188         }
2189
2190         if (skb->len <= 0) {
2191                 dev_kfree_skb_any(skb);
2192                 return NETDEV_TX_OK;
2193         }
2194
2195         /*
2196          * need: count + 4 desc gap to keep tail from touching
2197          *       + 2 desc gap to keep tail from touching head,
2198          *       + 1 desc for skb->data,
2199          *       + 1 desc for context descriptor,
2200          * head, otherwise try next time
2201          */
2202         if (igbvf_maybe_stop_tx(netdev, skb_shinfo(skb)->nr_frags + 4)) {
2203                 /* this is a hard error */
2204                 return NETDEV_TX_BUSY;
2205         }
2206
2207         if (vlan_tx_tag_present(skb)) {
2208                 tx_flags |= IGBVF_TX_FLAGS_VLAN;
2209                 tx_flags |= (vlan_tx_tag_get(skb) << IGBVF_TX_FLAGS_VLAN_SHIFT);
2210         }
2211
2212         if (skb->protocol == htons(ETH_P_IP))
2213                 tx_flags |= IGBVF_TX_FLAGS_IPV4;
2214
2215         first = tx_ring->next_to_use;
2216
2217         tso = skb_is_gso(skb) ?
2218                 igbvf_tso(adapter, tx_ring, skb, tx_flags, &hdr_len) : 0;
2219         if (unlikely(tso < 0)) {
2220                 dev_kfree_skb_any(skb);
2221                 return NETDEV_TX_OK;
2222         }
2223
2224         if (tso)
2225                 tx_flags |= IGBVF_TX_FLAGS_TSO;
2226         else if (igbvf_tx_csum(adapter, tx_ring, skb, tx_flags) &&
2227                  (skb->ip_summed == CHECKSUM_PARTIAL))
2228                 tx_flags |= IGBVF_TX_FLAGS_CSUM;
2229
2230         /*
2231          * count reflects descriptors mapped, if 0 then mapping error
2232          * has occurred and we need to rewind the descriptor queue
2233          */
2234         count = igbvf_tx_map_adv(adapter, tx_ring, skb, first);
2235
2236         if (count) {
2237                 igbvf_tx_queue_adv(adapter, tx_ring, tx_flags, count,
2238                                    skb->len, hdr_len);
2239                 /* Make sure there is space in the ring for the next send. */
2240                 igbvf_maybe_stop_tx(netdev, MAX_SKB_FRAGS + 4);
2241         } else {
2242                 dev_kfree_skb_any(skb);
2243                 tx_ring->buffer_info[first].time_stamp = 0;
2244                 tx_ring->next_to_use = first;
2245         }
2246
2247         return NETDEV_TX_OK;
2248 }
2249
2250 static netdev_tx_t igbvf_xmit_frame(struct sk_buff *skb,
2251                                     struct net_device *netdev)
2252 {
2253         struct igbvf_adapter *adapter = netdev_priv(netdev);
2254         struct igbvf_ring *tx_ring;
2255
2256         if (test_bit(__IGBVF_DOWN, &adapter->state)) {
2257                 dev_kfree_skb_any(skb);
2258                 return NETDEV_TX_OK;
2259         }
2260
2261         tx_ring = &adapter->tx_ring[0];
2262
2263         return igbvf_xmit_frame_ring_adv(skb, netdev, tx_ring);
2264 }
2265
2266 /**
2267  * igbvf_tx_timeout - Respond to a Tx Hang
2268  * @netdev: network interface device structure
2269  **/
2270 static void igbvf_tx_timeout(struct net_device *netdev)
2271 {
2272         struct igbvf_adapter *adapter = netdev_priv(netdev);
2273
2274         /* Do the reset outside of interrupt context */
2275         adapter->tx_timeout_count++;
2276         schedule_work(&adapter->reset_task);
2277 }
2278
2279 static void igbvf_reset_task(struct work_struct *work)
2280 {
2281         struct igbvf_adapter *adapter;
2282         adapter = container_of(work, struct igbvf_adapter, reset_task);
2283
2284         igbvf_reinit_locked(adapter);
2285 }
2286
2287 /**
2288  * igbvf_get_stats - Get System Network Statistics
2289  * @netdev: network interface device structure
2290  *
2291  * Returns the address of the device statistics structure.
2292  * The statistics are actually updated from the timer callback.
2293  **/
2294 static struct net_device_stats *igbvf_get_stats(struct net_device *netdev)
2295 {
2296         struct igbvf_adapter *adapter = netdev_priv(netdev);
2297
2298         /* only return the current stats */
2299         return &adapter->net_stats;
2300 }
2301
2302 /**
2303  * igbvf_change_mtu - Change the Maximum Transfer Unit
2304  * @netdev: network interface device structure
2305  * @new_mtu: new value for maximum frame size
2306  *
2307  * Returns 0 on success, negative on failure
2308  **/
2309 static int igbvf_change_mtu(struct net_device *netdev, int new_mtu)
2310 {
2311         struct igbvf_adapter *adapter = netdev_priv(netdev);
2312         int max_frame = new_mtu + ETH_HLEN + ETH_FCS_LEN;
2313
2314         if ((new_mtu < 68) || (max_frame > MAX_JUMBO_FRAME_SIZE)) {
2315                 dev_err(&adapter->pdev->dev, "Invalid MTU setting\n");
2316                 return -EINVAL;
2317         }
2318
2319 #define MAX_STD_JUMBO_FRAME_SIZE 9234
2320         if (max_frame > MAX_STD_JUMBO_FRAME_SIZE) {
2321                 dev_err(&adapter->pdev->dev, "MTU > 9216 not supported.\n");
2322                 return -EINVAL;
2323         }
2324
2325         while (test_and_set_bit(__IGBVF_RESETTING, &adapter->state))
2326                 msleep(1);
2327         /* igbvf_down has a dependency on max_frame_size */
2328         adapter->max_frame_size = max_frame;
2329         if (netif_running(netdev))
2330                 igbvf_down(adapter);
2331
2332         /*
2333          * NOTE: netdev_alloc_skb reserves 16 bytes, and typically NET_IP_ALIGN
2334          * means we reserve 2 more, this pushes us to allocate from the next
2335          * larger slab size.
2336          * i.e. RXBUFFER_2048 --> size-4096 slab
2337          * However with the new *_jumbo_rx* routines, jumbo receives will use
2338          * fragmented skbs
2339          */
2340
2341         if (max_frame <= 1024)
2342                 adapter->rx_buffer_len = 1024;
2343         else if (max_frame <= 2048)
2344                 adapter->rx_buffer_len = 2048;
2345         else
2346 #if (PAGE_SIZE / 2) > 16384
2347                 adapter->rx_buffer_len = 16384;
2348 #else
2349                 adapter->rx_buffer_len = PAGE_SIZE / 2;
2350 #endif
2351
2352
2353         /* adjust allocation if LPE protects us, and we aren't using SBP */
2354         if ((max_frame == ETH_FRAME_LEN + ETH_FCS_LEN) ||
2355              (max_frame == ETH_FRAME_LEN + VLAN_HLEN + ETH_FCS_LEN))
2356                 adapter->rx_buffer_len = ETH_FRAME_LEN + VLAN_HLEN +
2357                                          ETH_FCS_LEN;
2358
2359         dev_info(&adapter->pdev->dev, "changing MTU from %d to %d\n",
2360                  netdev->mtu, new_mtu);
2361         netdev->mtu = new_mtu;
2362
2363         if (netif_running(netdev))
2364                 igbvf_up(adapter);
2365         else
2366                 igbvf_reset(adapter);
2367
2368         clear_bit(__IGBVF_RESETTING, &adapter->state);
2369
2370         return 0;
2371 }
2372
2373 static int igbvf_ioctl(struct net_device *netdev, struct ifreq *ifr, int cmd)
2374 {
2375         switch (cmd) {
2376         default:
2377                 return -EOPNOTSUPP;
2378         }
2379 }
2380
2381 static int igbvf_suspend(struct pci_dev *pdev, pm_message_t state)
2382 {
2383         struct net_device *netdev = pci_get_drvdata(pdev);
2384         struct igbvf_adapter *adapter = netdev_priv(netdev);
2385 #ifdef CONFIG_PM
2386         int retval = 0;
2387 #endif
2388
2389         netif_device_detach(netdev);
2390
2391         if (netif_running(netdev)) {
2392                 WARN_ON(test_bit(__IGBVF_RESETTING, &adapter->state));
2393                 igbvf_down(adapter);
2394                 igbvf_free_irq(adapter);
2395         }
2396
2397 #ifdef CONFIG_PM
2398         retval = pci_save_state(pdev);
2399         if (retval)
2400                 return retval;
2401 #endif
2402
2403         pci_disable_device(pdev);
2404
2405         return 0;
2406 }
2407
2408 #ifdef CONFIG_PM
2409 static int igbvf_resume(struct pci_dev *pdev)
2410 {
2411         struct net_device *netdev = pci_get_drvdata(pdev);
2412         struct igbvf_adapter *adapter = netdev_priv(netdev);
2413         u32 err;
2414
2415         pci_restore_state(pdev);
2416         err = pci_enable_device_mem(pdev);
2417         if (err) {
2418                 dev_err(&pdev->dev, "Cannot enable PCI device from suspend\n");
2419                 return err;
2420         }
2421
2422         pci_set_master(pdev);
2423
2424         if (netif_running(netdev)) {
2425                 err = igbvf_request_irq(adapter);
2426                 if (err)
2427                         return err;
2428         }
2429
2430         igbvf_reset(adapter);
2431
2432         if (netif_running(netdev))
2433                 igbvf_up(adapter);
2434
2435         netif_device_attach(netdev);
2436
2437         return 0;
2438 }
2439 #endif
2440
2441 static void igbvf_shutdown(struct pci_dev *pdev)
2442 {
2443         igbvf_suspend(pdev, PMSG_SUSPEND);
2444 }
2445
2446 #ifdef CONFIG_NET_POLL_CONTROLLER
2447 /*
2448  * Polling 'interrupt' - used by things like netconsole to send skbs
2449  * without having to re-enable interrupts. It's not called while
2450  * the interrupt routine is executing.
2451  */
2452 static void igbvf_netpoll(struct net_device *netdev)
2453 {
2454         struct igbvf_adapter *adapter = netdev_priv(netdev);
2455
2456         disable_irq(adapter->pdev->irq);
2457
2458         igbvf_clean_tx_irq(adapter->tx_ring);
2459
2460         enable_irq(adapter->pdev->irq);
2461 }
2462 #endif
2463
2464 /**
2465  * igbvf_io_error_detected - called when PCI error is detected
2466  * @pdev: Pointer to PCI device
2467  * @state: The current pci connection state
2468  *
2469  * This function is called after a PCI bus error affecting
2470  * this device has been detected.
2471  */
2472 static pci_ers_result_t igbvf_io_error_detected(struct pci_dev *pdev,
2473                                                 pci_channel_state_t state)
2474 {
2475         struct net_device *netdev = pci_get_drvdata(pdev);
2476         struct igbvf_adapter *adapter = netdev_priv(netdev);
2477
2478         netif_device_detach(netdev);
2479
2480         if (state == pci_channel_io_perm_failure)
2481                 return PCI_ERS_RESULT_DISCONNECT;
2482
2483         if (netif_running(netdev))
2484                 igbvf_down(adapter);
2485         pci_disable_device(pdev);
2486
2487         /* Request a slot slot reset. */
2488         return PCI_ERS_RESULT_NEED_RESET;
2489 }
2490
2491 /**
2492  * igbvf_io_slot_reset - called after the pci bus has been reset.
2493  * @pdev: Pointer to PCI device
2494  *
2495  * Restart the card from scratch, as if from a cold-boot. Implementation
2496  * resembles the first-half of the igbvf_resume routine.
2497  */
2498 static pci_ers_result_t igbvf_io_slot_reset(struct pci_dev *pdev)
2499 {
2500         struct net_device *netdev = pci_get_drvdata(pdev);
2501         struct igbvf_adapter *adapter = netdev_priv(netdev);
2502
2503         if (pci_enable_device_mem(pdev)) {
2504                 dev_err(&pdev->dev,
2505                         "Cannot re-enable PCI device after reset.\n");
2506                 return PCI_ERS_RESULT_DISCONNECT;
2507         }
2508         pci_set_master(pdev);
2509
2510         igbvf_reset(adapter);
2511
2512         return PCI_ERS_RESULT_RECOVERED;
2513 }
2514
2515 /**
2516  * igbvf_io_resume - called when traffic can start flowing again.
2517  * @pdev: Pointer to PCI device
2518  *
2519  * This callback is called when the error recovery driver tells us that
2520  * its OK to resume normal operation. Implementation resembles the
2521  * second-half of the igbvf_resume routine.
2522  */
2523 static void igbvf_io_resume(struct pci_dev *pdev)
2524 {
2525         struct net_device *netdev = pci_get_drvdata(pdev);
2526         struct igbvf_adapter *adapter = netdev_priv(netdev);
2527
2528         if (netif_running(netdev)) {
2529                 if (igbvf_up(adapter)) {
2530                         dev_err(&pdev->dev,
2531                                 "can't bring device back up after reset\n");
2532                         return;
2533                 }
2534         }
2535
2536         netif_device_attach(netdev);
2537 }
2538
2539 static void igbvf_print_device_info(struct igbvf_adapter *adapter)
2540 {
2541         struct e1000_hw *hw = &adapter->hw;
2542         struct net_device *netdev = adapter->netdev;
2543         struct pci_dev *pdev = adapter->pdev;
2544
2545         if (hw->mac.type == e1000_vfadapt_i350)
2546                 dev_info(&pdev->dev, "Intel(R) I350 Virtual Function\n");
2547         else
2548                 dev_info(&pdev->dev, "Intel(R) 82576 Virtual Function\n");
2549         dev_info(&pdev->dev, "Address: %pM\n", netdev->dev_addr);
2550 }
2551
2552 static int igbvf_set_features(struct net_device *netdev,
2553         netdev_features_t features)
2554 {
2555         struct igbvf_adapter *adapter = netdev_priv(netdev);
2556
2557         if (features & NETIF_F_RXCSUM)
2558                 adapter->flags &= ~IGBVF_FLAG_RX_CSUM_DISABLED;
2559         else
2560                 adapter->flags |= IGBVF_FLAG_RX_CSUM_DISABLED;
2561
2562         return 0;
2563 }
2564
2565 static const struct net_device_ops igbvf_netdev_ops = {
2566         .ndo_open                       = igbvf_open,
2567         .ndo_stop                       = igbvf_close,
2568         .ndo_start_xmit                 = igbvf_xmit_frame,
2569         .ndo_get_stats                  = igbvf_get_stats,
2570         .ndo_set_rx_mode                = igbvf_set_multi,
2571         .ndo_set_mac_address            = igbvf_set_mac,
2572         .ndo_change_mtu                 = igbvf_change_mtu,
2573         .ndo_do_ioctl                   = igbvf_ioctl,
2574         .ndo_tx_timeout                 = igbvf_tx_timeout,
2575         .ndo_vlan_rx_add_vid            = igbvf_vlan_rx_add_vid,
2576         .ndo_vlan_rx_kill_vid           = igbvf_vlan_rx_kill_vid,
2577 #ifdef CONFIG_NET_POLL_CONTROLLER
2578         .ndo_poll_controller            = igbvf_netpoll,
2579 #endif
2580         .ndo_set_features               = igbvf_set_features,
2581 };
2582
2583 /**
2584  * igbvf_probe - Device Initialization Routine
2585  * @pdev: PCI device information struct
2586  * @ent: entry in igbvf_pci_tbl
2587  *
2588  * Returns 0 on success, negative on failure
2589  *
2590  * igbvf_probe initializes an adapter identified by a pci_dev structure.
2591  * The OS initialization, configuring of the adapter private structure,
2592  * and a hardware reset occur.
2593  **/
2594 static int __devinit igbvf_probe(struct pci_dev *pdev,
2595                                  const struct pci_device_id *ent)
2596 {
2597         struct net_device *netdev;
2598         struct igbvf_adapter *adapter;
2599         struct e1000_hw *hw;
2600         const struct igbvf_info *ei = igbvf_info_tbl[ent->driver_data];
2601
2602         static int cards_found;
2603         int err, pci_using_dac;
2604
2605         err = pci_enable_device_mem(pdev);
2606         if (err)
2607                 return err;
2608
2609         pci_using_dac = 0;
2610         err = dma_set_mask(&pdev->dev, DMA_BIT_MASK(64));
2611         if (!err) {
2612                 err = dma_set_coherent_mask(&pdev->dev, DMA_BIT_MASK(64));
2613                 if (!err)
2614                         pci_using_dac = 1;
2615         } else {
2616                 err = dma_set_mask(&pdev->dev, DMA_BIT_MASK(32));
2617                 if (err) {
2618                         err = dma_set_coherent_mask(&pdev->dev,
2619                                                     DMA_BIT_MASK(32));
2620                         if (err) {
2621                                 dev_err(&pdev->dev, "No usable DMA "
2622                                         "configuration, aborting\n");
2623                                 goto err_dma;
2624                         }
2625                 }
2626         }
2627
2628         err = pci_request_regions(pdev, igbvf_driver_name);
2629         if (err)
2630                 goto err_pci_reg;
2631
2632         pci_set_master(pdev);
2633
2634         err = -ENOMEM;
2635         netdev = alloc_etherdev(sizeof(struct igbvf_adapter));
2636         if (!netdev)
2637                 goto err_alloc_etherdev;
2638
2639         SET_NETDEV_DEV(netdev, &pdev->dev);
2640
2641         pci_set_drvdata(pdev, netdev);
2642         adapter = netdev_priv(netdev);
2643         hw = &adapter->hw;
2644         adapter->netdev = netdev;
2645         adapter->pdev = pdev;
2646         adapter->ei = ei;
2647         adapter->pba = ei->pba;
2648         adapter->flags = ei->flags;
2649         adapter->hw.back = adapter;
2650         adapter->hw.mac.type = ei->mac;
2651         adapter->msg_enable = (1 << NETIF_MSG_DRV | NETIF_MSG_PROBE) - 1;
2652
2653         /* PCI config space info */
2654
2655         hw->vendor_id = pdev->vendor;
2656         hw->device_id = pdev->device;
2657         hw->subsystem_vendor_id = pdev->subsystem_vendor;
2658         hw->subsystem_device_id = pdev->subsystem_device;
2659         hw->revision_id = pdev->revision;
2660
2661         err = -EIO;
2662         adapter->hw.hw_addr = ioremap(pci_resource_start(pdev, 0),
2663                                       pci_resource_len(pdev, 0));
2664
2665         if (!adapter->hw.hw_addr)
2666                 goto err_ioremap;
2667
2668         if (ei->get_variants) {
2669                 err = ei->get_variants(adapter);
2670                 if (err)
2671                         goto err_ioremap;
2672         }
2673
2674         /* setup adapter struct */
2675         err = igbvf_sw_init(adapter);
2676         if (err)
2677                 goto err_sw_init;
2678
2679         /* construct the net_device struct */
2680         netdev->netdev_ops = &igbvf_netdev_ops;
2681
2682         igbvf_set_ethtool_ops(netdev);
2683         netdev->watchdog_timeo = 5 * HZ;
2684         strncpy(netdev->name, pci_name(pdev), sizeof(netdev->name) - 1);
2685
2686         adapter->bd_number = cards_found++;
2687
2688         netdev->hw_features = NETIF_F_SG |
2689                            NETIF_F_IP_CSUM |
2690                            NETIF_F_IPV6_CSUM |
2691                            NETIF_F_TSO |
2692                            NETIF_F_TSO6 |
2693                            NETIF_F_RXCSUM;
2694
2695         netdev->features = netdev->hw_features |
2696                            NETIF_F_HW_VLAN_TX |
2697                            NETIF_F_HW_VLAN_RX |
2698                            NETIF_F_HW_VLAN_FILTER;
2699
2700         if (pci_using_dac)
2701                 netdev->features |= NETIF_F_HIGHDMA;
2702
2703         netdev->vlan_features |= NETIF_F_TSO;
2704         netdev->vlan_features |= NETIF_F_TSO6;
2705         netdev->vlan_features |= NETIF_F_IP_CSUM;
2706         netdev->vlan_features |= NETIF_F_IPV6_CSUM;
2707         netdev->vlan_features |= NETIF_F_SG;
2708
2709         /*reset the controller to put the device in a known good state */
2710         err = hw->mac.ops.reset_hw(hw);
2711         if (err) {
2712                 dev_info(&pdev->dev,
2713                          "PF still in reset state, assigning new address."
2714                          " Is the PF interface up?\n");
2715                 eth_hw_addr_random(netdev);
2716                 memcpy(adapter->hw.mac.addr, netdev->dev_addr,
2717                         netdev->addr_len);
2718         } else {
2719                 err = hw->mac.ops.read_mac_addr(hw);
2720                 if (err) {
2721                         dev_err(&pdev->dev, "Error reading MAC address\n");
2722                         goto err_hw_init;
2723                 }
2724                 memcpy(netdev->dev_addr, adapter->hw.mac.addr,
2725                         netdev->addr_len);
2726         }
2727
2728         if (!is_valid_ether_addr(netdev->perm_addr)) {
2729                 dev_err(&pdev->dev, "Invalid MAC Address: %pM\n",
2730                         netdev->dev_addr);
2731                 err = -EIO;
2732                 goto err_hw_init;
2733         }
2734
2735         memcpy(netdev->perm_addr, adapter->hw.mac.addr, netdev->addr_len);
2736
2737         setup_timer(&adapter->watchdog_timer, &igbvf_watchdog,
2738                     (unsigned long) adapter);
2739
2740         INIT_WORK(&adapter->reset_task, igbvf_reset_task);
2741         INIT_WORK(&adapter->watchdog_task, igbvf_watchdog_task);
2742
2743         /* ring size defaults */
2744         adapter->rx_ring->count = 1024;
2745         adapter->tx_ring->count = 1024;
2746
2747         /* reset the hardware with the new settings */
2748         igbvf_reset(adapter);
2749
2750         strcpy(netdev->name, "eth%d");
2751         err = register_netdev(netdev);
2752         if (err)
2753                 goto err_hw_init;
2754
2755         /* tell the stack to leave us alone until igbvf_open() is called */
2756         netif_carrier_off(netdev);
2757         netif_stop_queue(netdev);
2758
2759         igbvf_print_device_info(adapter);
2760
2761         igbvf_initialize_last_counter_stats(adapter);
2762
2763         return 0;
2764
2765 err_hw_init:
2766         kfree(adapter->tx_ring);
2767         kfree(adapter->rx_ring);
2768 err_sw_init:
2769         igbvf_reset_interrupt_capability(adapter);
2770         iounmap(adapter->hw.hw_addr);
2771 err_ioremap:
2772         free_netdev(netdev);
2773 err_alloc_etherdev:
2774         pci_release_regions(pdev);
2775 err_pci_reg:
2776 err_dma:
2777         pci_disable_device(pdev);
2778         return err;
2779 }
2780
2781 /**
2782  * igbvf_remove - Device Removal Routine
2783  * @pdev: PCI device information struct
2784  *
2785  * igbvf_remove is called by the PCI subsystem to alert the driver
2786  * that it should release a PCI device.  The could be caused by a
2787  * Hot-Plug event, or because the driver is going to be removed from
2788  * memory.
2789  **/
2790 static void __devexit igbvf_remove(struct pci_dev *pdev)
2791 {
2792         struct net_device *netdev = pci_get_drvdata(pdev);
2793         struct igbvf_adapter *adapter = netdev_priv(netdev);
2794         struct e1000_hw *hw = &adapter->hw;
2795
2796         /*
2797          * The watchdog timer may be rescheduled, so explicitly
2798          * disable it from being rescheduled.
2799          */
2800         set_bit(__IGBVF_DOWN, &adapter->state);
2801         del_timer_sync(&adapter->watchdog_timer);
2802
2803         cancel_work_sync(&adapter->reset_task);
2804         cancel_work_sync(&adapter->watchdog_task);
2805
2806         unregister_netdev(netdev);
2807
2808         igbvf_reset_interrupt_capability(adapter);
2809
2810         /*
2811          * it is important to delete the napi struct prior to freeing the
2812          * rx ring so that you do not end up with null pointer refs
2813          */
2814         netif_napi_del(&adapter->rx_ring->napi);
2815         kfree(adapter->tx_ring);
2816         kfree(adapter->rx_ring);
2817
2818         iounmap(hw->hw_addr);
2819         if (hw->flash_address)
2820                 iounmap(hw->flash_address);
2821         pci_release_regions(pdev);
2822
2823         free_netdev(netdev);
2824
2825         pci_disable_device(pdev);
2826 }
2827
2828 /* PCI Error Recovery (ERS) */
2829 static struct pci_error_handlers igbvf_err_handler = {
2830         .error_detected = igbvf_io_error_detected,
2831         .slot_reset = igbvf_io_slot_reset,
2832         .resume = igbvf_io_resume,
2833 };
2834
2835 static DEFINE_PCI_DEVICE_TABLE(igbvf_pci_tbl) = {
2836         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82576_VF), board_vf },
2837         { PCI_VDEVICE(INTEL, E1000_DEV_ID_I350_VF), board_i350_vf },
2838         { } /* terminate list */
2839 };
2840 MODULE_DEVICE_TABLE(pci, igbvf_pci_tbl);
2841
2842 /* PCI Device API Driver */
2843 static struct pci_driver igbvf_driver = {
2844         .name     = igbvf_driver_name,
2845         .id_table = igbvf_pci_tbl,
2846         .probe    = igbvf_probe,
2847         .remove   = __devexit_p(igbvf_remove),
2848 #ifdef CONFIG_PM
2849         /* Power Management Hooks */
2850         .suspend  = igbvf_suspend,
2851         .resume   = igbvf_resume,
2852 #endif
2853         .shutdown = igbvf_shutdown,
2854         .err_handler = &igbvf_err_handler
2855 };
2856
2857 /**
2858  * igbvf_init_module - Driver Registration Routine
2859  *
2860  * igbvf_init_module is the first routine called when the driver is
2861  * loaded. All it does is register with the PCI subsystem.
2862  **/
2863 static int __init igbvf_init_module(void)
2864 {
2865         int ret;
2866         pr_info("%s - version %s\n", igbvf_driver_string, igbvf_driver_version);
2867         pr_info("%s\n", igbvf_copyright);
2868
2869         ret = pci_register_driver(&igbvf_driver);
2870
2871         return ret;
2872 }
2873 module_init(igbvf_init_module);
2874
2875 /**
2876  * igbvf_exit_module - Driver Exit Cleanup Routine
2877  *
2878  * igbvf_exit_module is called just before the driver is removed
2879  * from memory.
2880  **/
2881 static void __exit igbvf_exit_module(void)
2882 {
2883         pci_unregister_driver(&igbvf_driver);
2884 }
2885 module_exit(igbvf_exit_module);
2886
2887
2888 MODULE_AUTHOR("Intel Corporation, <e1000-devel@lists.sourceforge.net>");
2889 MODULE_DESCRIPTION("Intel(R) Gigabit Virtual Function Network Driver");
2890 MODULE_LICENSE("GPL");
2891 MODULE_VERSION(DRV_VERSION);
2892
2893 /* netdev.c */