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

   1 /*******************************************************************************
   2
   3   Intel(R) 82576 Virtual Function Linux driver
   4   Copyright(c) 2009 - 2010 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 #include <linux/module.h>
  29 #include <linux/types.h>
  30 #include <linux/init.h>
  31 #include <linux/pci.h>
  32 #include <linux/vmalloc.h>
  33 #include <linux/pagemap.h>
  34 #include <linux/delay.h>
  35 #include <linux/netdevice.h>
  36 #include <linux/tcp.h>
  37 #include <linux/ipv6.h>
  38 #include <linux/slab.h>
  39 #include <net/checksum.h>
  40 #include <net/ip6_checksum.h>
  41 #include <linux/mii.h>
  42 #include <linux/ethtool.h>
  43 #include <linux/if_vlan.h>
  44 #include <linux/prefetch.h>
  45
  46 #include "igbvf.h"
  47
  48 #define DRV_VERSION "2.0.0-k"
  49 char igbvf_driver_name[] = "igbvf";
  50 const char igbvf_driver_version[] = DRV_VERSION;
  51 static const char igbvf_driver_string[] =
  52                                 "Intel(R) Virtual Function Network Driver";
  53 static const char igbvf_copyright[] =
  54                                 "Copyright (c) 2009 - 2010 Intel Corporation.";
  55
  56 static int igbvf_poll(struct napi_struct *napi, int budget);
  57 static void igbvf_reset(struct igbvf_adapter *);
  58 static void igbvf_set_interrupt_capability(struct igbvf_adapter *);
  59 static void igbvf_reset_interrupt_capability(struct igbvf_adapter *);
  60
  61 static struct igbvf_info igbvf_vf_info = {
  62         .mac                    = e1000_vfadapt,
  63         .flags                  = 0,
  64         .pba                    = 10,
  65         .init_ops               = e1000_init_function_pointers_vf,
  66 };
  67
  68 static struct igbvf_info igbvf_i350_vf_info = {
  69         .mac                    = e1000_vfadapt_i350,
  70         .flags                  = 0,
  71         .pba                    = 10,
  72         .init_ops               = e1000_init_function_pointers_vf,
  73 };
  74
  75 static const struct igbvf_info *igbvf_info_tbl[] = {
  76         [board_vf]              = &igbvf_vf_info,
  77         [board_i350_vf]         = &igbvf_i350_vf_info,
  78 };
  79
  80 /**
  81  * igbvf_desc_unused - calculate if we have unused descriptors
  82  **/
  83 static int igbvf_desc_unused(struct igbvf_ring *ring)
  84 {
  85         if (ring->next_to_clean > ring->next_to_use)
  86                 return ring->next_to_clean - ring->next_to_use - 1;
  87
  88         return ring->count + ring->next_to_clean - ring->next_to_use - 1;
  89 }
  90
  91 /**
  92  * igbvf_receive_skb - helper function to handle Rx indications
  93  * @adapter: board private structure
  94  * @status: descriptor status field as written by hardware
  95  * @vlan: descriptor vlan field as written by hardware (no le/be conversion)
  96  * @skb: pointer to sk_buff to be indicated to stack
  97  **/
  98 static void igbvf_receive_skb(struct igbvf_adapter *adapter,
  99                               struct net_device *netdev,
 100                               struct sk_buff *skb,
 101                               u32 status, u16 vlan)
 102 {
 103         if (status & E1000_RXD_STAT_VP) {
 104                 u16 vid = le16_to_cpu(vlan) & E1000_RXD_SPC_VLAN_MASK;
 105
 106                 __vlan_hwaccel_put_tag(skb, vid);
 107         }
 108         netif_receive_skb(skb);
 109 }
 110
 111 static inline void igbvf_rx_checksum_adv(struct igbvf_adapter *adapter,
 112                                          u32 status_err, struct sk_buff *skb)
 113 {
 114         skb_checksum_none_assert(skb);
 115
 116         /* Ignore Checksum bit is set or checksum is disabled through ethtool */
 117         if ((status_err & E1000_RXD_STAT_IXSM) ||
 118             (adapter->flags & IGBVF_FLAG_RX_CSUM_DISABLED))
 119                 return;
 120
 121         /* TCP/UDP checksum error bit is set */
 122         if (status_err &
 123             (E1000_RXDEXT_STATERR_TCPE | E1000_RXDEXT_STATERR_IPE)) {
 124                 /* let the stack verify checksum errors */
 125                 adapter->hw_csum_err++;
 126                 return;
 127         }
 128
 129         /* It must be a TCP or UDP packet with a valid checksum */
 130         if (status_err & (E1000_RXD_STAT_TCPCS | E1000_RXD_STAT_UDPCS))
 131                 skb->ip_summed = CHECKSUM_UNNECESSARY;
 132
 133         adapter->hw_csum_good++;
 134 }
 135
 136 /**
 137  * igbvf_alloc_rx_buffers - Replace used receive buffers; packet split
 138  * @rx_ring: address of ring structure to repopulate
 139  * @cleaned_count: number of buffers to repopulate
 140  **/
 141 static void igbvf_alloc_rx_buffers(struct igbvf_ring *rx_ring,
 142                                    int cleaned_count)
 143 {
 144         struct igbvf_adapter *adapter = rx_ring->adapter;
 145         struct net_device *netdev = adapter->netdev;
 146         struct pci_dev *pdev = adapter->pdev;
 147         union e1000_adv_rx_desc *rx_desc;
 148         struct igbvf_buffer *buffer_info;
 149         struct sk_buff *skb;
 150         unsigned int i;
 151         int bufsz;
 152
 153         i = rx_ring->next_to_use;
 154         buffer_info = &rx_ring->buffer_info[i];
 155
 156         if (adapter->rx_ps_hdr_size)
 157                 bufsz = adapter->rx_ps_hdr_size;
 158         else
 159                 bufsz = adapter->rx_buffer_len;
 160
 161         while (cleaned_count--) {
 162                 rx_desc = IGBVF_RX_DESC_ADV(*rx_ring, i);
 163
 164                 if (adapter->rx_ps_hdr_size && !buffer_info->page_dma) {
 165                         if (!buffer_info->page) {
 166                                 buffer_info->page = alloc_page(GFP_ATOMIC);
 167                                 if (!buffer_info->page) {
 168                                         adapter->alloc_rx_buff_failed++;
 169                                         goto no_buffers;
 170                                 }
 171                                 buffer_info->page_offset = 0;
 172                         } else {
 173                                 buffer_info->page_offset ^= PAGE_SIZE / 2;
 174                         }
 175                         buffer_info->page_dma =
 176                                 dma_map_page(&pdev->dev, buffer_info->page,
 177                                              buffer_info->page_offset,
 178                                              PAGE_SIZE / 2,
 179                                              DMA_FROM_DEVICE);
 180                 }
 181
 182                 if (!buffer_info->skb) {
 183                         skb = netdev_alloc_skb_ip_align(netdev, bufsz);
 184                         if (!skb) {
 185                                 adapter->alloc_rx_buff_failed++;
 186                                 goto no_buffers;
 187                         }
 188
 189                         buffer_info->skb = skb;
 190                         buffer_info->dma = dma_map_single(&pdev->dev, skb->data,
 191                                                           bufsz,
 192                                                           DMA_FROM_DEVICE);
 193                 }
 194                 /* Refresh the desc even if buffer_addrs didn't change because
 195                  * each write-back erases this info. */
 196                 if (adapter->rx_ps_hdr_size) {
 197                         rx_desc->read.pkt_addr =
 198                              cpu_to_le64(buffer_info->page_dma);
 199                         rx_desc->read.hdr_addr = cpu_to_le64(buffer_info->dma);
 200                 } else {
 201                         rx_desc->read.pkt_addr =
 202                              cpu_to_le64(buffer_info->dma);
 203                         rx_desc->read.hdr_addr = 0;
 204                 }
 205
 206                 i++;
 207                 if (i == rx_ring->count)
 208                         i = 0;
 209                 buffer_info = &rx_ring->buffer_info[i];
 210         }
 211
 212 no_buffers:
 213         if (rx_ring->next_to_use != i) {
 214                 rx_ring->next_to_use = i;
 215                 if (i == 0)
 216                         i = (rx_ring->count - 1);
 217                 else
 218                         i--;
 219
 220                 /* Force memory writes to complete before letting h/w
 221                  * know there are new descriptors to fetch.  (Only
 222                  * applicable for weak-ordered memory model archs,
 223                  * such as IA-64). */
 224                 wmb();
 225                 writel(i, adapter->hw.hw_addr + rx_ring->tail);
 226         }
 227 }
 228
 229 /**
 230  * igbvf_clean_rx_irq - Send received data up the network stack; legacy
 231  * @adapter: board private structure
 232  *
 233  * the return value indicates whether actual cleaning was done, there
 234  * is no guarantee that everything was cleaned
 235  **/
 236 static bool igbvf_clean_rx_irq(struct igbvf_adapter *adapter,
 237                                int *work_done, int work_to_do)
 238 {
 239         struct igbvf_ring *rx_ring = adapter->rx_ring;
 240         struct net_device *netdev = adapter->netdev;
 241         struct pci_dev *pdev = adapter->pdev;
 242         union e1000_adv_rx_desc *rx_desc, *next_rxd;
 243         struct igbvf_buffer *buffer_info, *next_buffer;
 244         struct sk_buff *skb;
 245         bool cleaned = false;
 246         int cleaned_count = 0;
 247         unsigned int total_bytes = 0, total_packets = 0;
 248         unsigned int i;
 249         u32 length, hlen, staterr;
 250
 251         i = rx_ring->next_to_clean;
 252         rx_desc = IGBVF_RX_DESC_ADV(*rx_ring, i);
 253         staterr = le32_to_cpu(rx_desc->wb.upper.status_error);
 254
 255         while (staterr & E1000_RXD_STAT_DD) {
 256                 if (*work_done >= work_to_do)
 257                         break;
 258                 (*work_done)++;
 259                 rmb(); /* read descriptor and rx_buffer_info after status DD */
 260
 261                 buffer_info = &rx_ring->buffer_info[i];
 262
 263                 /* HW will not DMA in data larger than the given buffer, even
 264                  * if it parses the (NFS, of course) header to be larger.  In
 265                  * that case, it fills the header buffer and spills the rest
 266                  * into the page.
 267                  */
 268                 hlen = (le16_to_cpu(rx_desc->wb.lower.lo_dword.hs_rss.hdr_info) &
 269                   E1000_RXDADV_HDRBUFLEN_MASK) >> E1000_RXDADV_HDRBUFLEN_SHIFT;
 270                 if (hlen > adapter->rx_ps_hdr_size)
 271                         hlen = adapter->rx_ps_hdr_size;
 272
 273                 length = le16_to_cpu(rx_desc->wb.upper.length);
 274                 cleaned = true;
 275                 cleaned_count++;
 276
 277                 skb = buffer_info->skb;
 278                 prefetch(skb->data - NET_IP_ALIGN);
 279                 buffer_info->skb = NULL;
 280                 if (!adapter->rx_ps_hdr_size) {
 281                         dma_unmap_single(&pdev->dev, buffer_info->dma,
 282                                          adapter->rx_buffer_len,
 283                                          DMA_FROM_DEVICE);
 284                         buffer_info->dma = 0;
 285                         skb_put(skb, length);
 286                         goto send_up;
 287                 }
 288
 289                 if (!skb_shinfo(skb)->nr_frags) {
 290                         dma_unmap_single(&pdev->dev, buffer_info->dma,
 291                                          adapter->rx_ps_hdr_size,
 292                                          DMA_FROM_DEVICE);
 293                         skb_put(skb, hlen);
 294                 }
 295
 296                 if (length) {
 297                         dma_unmap_page(&pdev->dev, buffer_info->page_dma,
 298                                        PAGE_SIZE / 2,
 299                                        DMA_FROM_DEVICE);
 300                         buffer_info->page_dma = 0;
 301
 302                         skb_fill_page_desc(skb, skb_shinfo(skb)->nr_frags,
 303                                            buffer_info->page,
 304                                            buffer_info->page_offset,
 305                                            length);
 306
 307                         if ((adapter->rx_buffer_len > (PAGE_SIZE / 2)) ||
 308                             (page_count(buffer_info->page) != 1))
 309                                 buffer_info->page = NULL;
 310                         else
 311                                 get_page(buffer_info->page);
 312
 313                         skb->len += length;
 314                         skb->data_len += length;
 315                         skb->truesize += length;
 316                 }
 317 send_up:
 318                 i++;
 319                 if (i == rx_ring->count)
 320                         i = 0;
 321                 next_rxd = IGBVF_RX_DESC_ADV(*rx_ring, i);
 322                 prefetch(next_rxd);
 323                 next_buffer = &rx_ring->buffer_info[i];
 324
 325                 if (!(staterr & E1000_RXD_STAT_EOP)) {
 326                         buffer_info->skb = next_buffer->skb;
 327                         buffer_info->dma = next_buffer->dma;
 328                         next_buffer->skb = skb;
 329                         next_buffer->dma = 0;
 330                         goto next_desc;
 331                 }
 332
 333                 if (staterr & E1000_RXDEXT_ERR_FRAME_ERR_MASK) {
 334                         dev_kfree_skb_irq(skb);
 335                         goto next_desc;
 336                 }
 337
 338                 total_bytes += skb->len;
 339                 total_packets++;
 340
 341                 igbvf_rx_checksum_adv(adapter, staterr, skb);
 342
 343                 skb->protocol = eth_type_trans(skb, netdev);
 344
 345                 igbvf_receive_skb(adapter, netdev, skb, staterr,
 346                                   rx_desc->wb.upper.vlan);
 347
 348 next_desc:
 349                 rx_desc->wb.upper.status_error = 0;
 350
 351                 /* return some buffers to hardware, one at a time is too slow */
 352                 if (cleaned_count >= IGBVF_RX_BUFFER_WRITE) {
 353                         igbvf_alloc_rx_buffers(rx_ring, cleaned_count);
 354                         cleaned_count = 0;
 355                 }
 356
 357                 /* use prefetched values */
 358                 rx_desc = next_rxd;
 359                 buffer_info = next_buffer;
 360
 361                 staterr = le32_to_cpu(rx_desc->wb.upper.status_error);
 362         }
 363
 364         rx_ring->next_to_clean = i;
 365         cleaned_count = igbvf_desc_unused(rx_ring);
 366
 367         if (cleaned_count)
 368                 igbvf_alloc_rx_buffers(rx_ring, cleaned_count);
 369
 370         adapter->total_rx_packets += total_packets;
 371         adapter->total_rx_bytes += total_bytes;
 372         adapter->net_stats.rx_bytes += total_bytes;
 373         adapter->net_stats.rx_packets += total_packets;
 374         return cleaned;
 375 }
 376
 377 static void igbvf_put_txbuf(struct igbvf_adapter *adapter,
 378                             struct igbvf_buffer *buffer_info)
 379 {
 380         if (buffer_info->dma) {
 381                 if (buffer_info->mapped_as_page)
 382                         dma_unmap_page(&adapter->pdev->dev,
 383                                        buffer_info->dma,
 384                                        buffer_info->length,
 385                                        DMA_TO_DEVICE);
 386                 else
 387                         dma_unmap_single(&adapter->pdev->dev,
 388                                          buffer_info->dma,
 389                                          buffer_info->length,
 390                                          DMA_TO_DEVICE);
 391                 buffer_info->dma = 0;
 392         }
 393         if (buffer_info->skb) {
 394                 dev_kfree_skb_any(buffer_info->skb);
 395                 buffer_info->skb = NULL;
 396         }
 397         buffer_info->time_stamp = 0;
 398 }
 399
 400 /**
 401  * igbvf_setup_tx_resources - allocate Tx resources (Descriptors)
 402  * @adapter: board private structure
 403  *
 404  * Return 0 on success, negative on failure
 405  **/
 406 int igbvf_setup_tx_resources(struct igbvf_adapter *adapter,
 407                              struct igbvf_ring *tx_ring)
 408 {
 409         struct pci_dev *pdev = adapter->pdev;
 410         int size;
 411
 412         size = sizeof(struct igbvf_buffer) * tx_ring->count;
 413         tx_ring->buffer_info = vzalloc(size);
 414         if (!tx_ring->buffer_info)
 415                 goto err;
 416
 417         /* round up to nearest 4K */
 418         tx_ring->size = tx_ring->count * sizeof(union e1000_adv_tx_desc);
 419         tx_ring->size = ALIGN(tx_ring->size, 4096);
 420
 421         tx_ring->desc = dma_alloc_coherent(&pdev->dev, tx_ring->size,
 422                                            &tx_ring->dma, GFP_KERNEL);
 423
 424         if (!tx_ring->desc)
 425                 goto err;
 426
 427         tx_ring->adapter = adapter;
 428         tx_ring->next_to_use = 0;
 429         tx_ring->next_to_clean = 0;
 430
 431         return 0;
 432 err:
 433         vfree(tx_ring->buffer_info);
 434         dev_err(&adapter->pdev->dev,
 435                 "Unable to allocate memory for the transmit descriptor ring\n");
 436         return -ENOMEM;
 437 }
 438
 439 /**
 440  * igbvf_setup_rx_resources - allocate Rx resources (Descriptors)
 441  * @adapter: board private structure
 442  *
 443  * Returns 0 on success, negative on failure
 444  **/
 445 int igbvf_setup_rx_resources(struct igbvf_adapter *adapter,
 446                              struct igbvf_ring *rx_ring)
 447 {
 448         struct pci_dev *pdev = adapter->pdev;
 449         int size, desc_len;
 450
 451         size = sizeof(struct igbvf_buffer) * rx_ring->count;
 452         rx_ring->buffer_info = vzalloc(size);
 453         if (!rx_ring->buffer_info)
 454                 goto err;
 455
 456         desc_len = sizeof(union e1000_adv_rx_desc);
 457
 458         /* Round up to nearest 4K */
 459         rx_ring->size = rx_ring->count * desc_len;
 460         rx_ring->size = ALIGN(rx_ring->size, 4096);
 461
 462         rx_ring->desc = dma_alloc_coherent(&pdev->dev, rx_ring->size,
 463                                            &rx_ring->dma, GFP_KERNEL);
 464
 465         if (!rx_ring->desc)
 466                 goto err;
 467
 468         rx_ring->next_to_clean = 0;
 469         rx_ring->next_to_use = 0;
 470
 471         rx_ring->adapter = adapter;
 472
 473         return 0;
 474
 475 err:
 476         vfree(rx_ring->buffer_info);
 477         rx_ring->buffer_info = NULL;
 478         dev_err(&adapter->pdev->dev,
 479                 "Unable to allocate memory for the receive descriptor ring\n");
 480         return -ENOMEM;
 481 }
 482
 483 /**
 484  * igbvf_clean_tx_ring - Free Tx Buffers
 485  * @tx_ring: ring to be cleaned
 486  **/
 487 static void igbvf_clean_tx_ring(struct igbvf_ring *tx_ring)
 488 {
 489         struct igbvf_adapter *adapter = tx_ring->adapter;
 490         struct igbvf_buffer *buffer_info;
 491         unsigned long size;
 492         unsigned int i;
 493
 494         if (!tx_ring->buffer_info)
 495                 return;
 496
 497         /* Free all the Tx ring sk_buffs */
 498         for (i = 0; i < tx_ring->count; i++) {
 499                 buffer_info = &tx_ring->buffer_info[i];
 500                 igbvf_put_txbuf(adapter, buffer_info);
 501         }
 502
 503         size = sizeof(struct igbvf_buffer) * tx_ring->count;
 504         memset(tx_ring->buffer_info, 0, size);
 505
 506         /* Zero out the descriptor ring */
 507         memset(tx_ring->desc, 0, tx_ring->size);
 508
 509         tx_ring->next_to_use = 0;
 510         tx_ring->next_to_clean = 0;
 511
 512         writel(0, adapter->hw.hw_addr + tx_ring->head);
 513         writel(0, adapter->hw.hw_addr + tx_ring->tail);
 514 }
 515
 516 /**
 517  * igbvf_free_tx_resources - Free Tx Resources per Queue
 518  * @tx_ring: ring to free resources from
 519  *
 520  * Free all transmit software resources
 521  **/
 522 void igbvf_free_tx_resources(struct igbvf_ring *tx_ring)
 523 {
 524         struct pci_dev *pdev = tx_ring->adapter->pdev;
 525
 526         igbvf_clean_tx_ring(tx_ring);
 527
 528         vfree(tx_ring->buffer_info);
 529         tx_ring->buffer_info = NULL;
 530
 531         dma_free_coherent(&pdev->dev, tx_ring->size, tx_ring->desc,
 532                           tx_ring->dma);
 533
 534         tx_ring->desc = NULL;
 535 }
 536
 537 /**
 538  * igbvf_clean_rx_ring - Free Rx Buffers per Queue
 539  * @adapter: board private structure
 540  **/
 541 static void igbvf_clean_rx_ring(struct igbvf_ring *rx_ring)
 542 {
 543         struct igbvf_adapter *adapter = rx_ring->adapter;
 544         struct igbvf_buffer *buffer_info;
 545         struct pci_dev *pdev = adapter->pdev;
 546         unsigned long size;
 547         unsigned int i;
 548
 549         if (!rx_ring->buffer_info)
 550                 return;
 551
 552         /* Free all the Rx ring sk_buffs */
 553         for (i = 0; i < rx_ring->count; i++) {
 554                 buffer_info = &rx_ring->buffer_info[i];
 555                 if (buffer_info->dma) {
 556                         if (adapter->rx_ps_hdr_size){
 557                                 dma_unmap_single(&pdev->dev, buffer_info->dma,
 558                                                  adapter->rx_ps_hdr_size,
 559                                                  DMA_FROM_DEVICE);
 560                         } else {
 561                                 dma_unmap_single(&pdev->dev, buffer_info->dma,
 562                                                  adapter->rx_buffer_len,
 563                                                  DMA_FROM_DEVICE);
 564                         }
 565                         buffer_info->dma = 0;
 566                 }
 567
 568                 if (buffer_info->skb) {
 569                         dev_kfree_skb(buffer_info->skb);
 570                         buffer_info->skb = NULL;
 571                 }
 572
 573                 if (buffer_info->page) {
 574                         if (buffer_info->page_dma)
 575                                 dma_unmap_page(&pdev->dev,
 576                                                buffer_info->page_dma,
 577                                                PAGE_SIZE / 2,
 578                                                DMA_FROM_DEVICE);
 579                         put_page(buffer_info->page);
 580                         buffer_info->page = NULL;
 581                         buffer_info->page_dma = 0;
 582                         buffer_info->page_offset = 0;
 583                 }
 584         }
 585
 586         size = sizeof(struct igbvf_buffer) * rx_ring->count;
 587         memset(rx_ring->buffer_info, 0, size);
 588
 589         /* Zero out the descriptor ring */
 590         memset(rx_ring->desc, 0, rx_ring->size);
 591
 592         rx_ring->next_to_clean = 0;
 593         rx_ring->next_to_use = 0;
 594
 595         writel(0, adapter->hw.hw_addr + rx_ring->head);
 596         writel(0, adapter->hw.hw_addr + rx_ring->tail);
 597 }
 598
 599 /**
 600  * igbvf_free_rx_resources - Free Rx Resources
 601  * @rx_ring: ring to clean the resources from
 602  *
 603  * Free all receive software resources
 604  **/
 605
 606 void igbvf_free_rx_resources(struct igbvf_ring *rx_ring)
 607 {
 608         struct pci_dev *pdev = rx_ring->adapter->pdev;
 609
 610         igbvf_clean_rx_ring(rx_ring);
 611
 612         vfree(rx_ring->buffer_info);
 613         rx_ring->buffer_info = NULL;
 614
 615         dma_free_coherent(&pdev->dev, rx_ring->size, rx_ring->desc,
 616                           rx_ring->dma);
 617         rx_ring->desc = NULL;
 618 }
 619
 620 /**
 621  * igbvf_update_itr - update the dynamic ITR value based on statistics
 622  * @adapter: pointer to adapter
 623  * @itr_setting: current adapter->itr
 624  * @packets: the number of packets during this measurement interval
 625  * @bytes: the number of bytes during this measurement interval
 626  *
 627  *      Stores a new ITR value based on packets and byte
 628  *      counts during the last interrupt.  The advantage of per interrupt
 629  *      computation is faster updates and more accurate ITR for the current
 630  *      traffic pattern.  Constants in this function were computed
 631  *      based on theoretical maximum wire speed and thresholds were set based
 632  *      on testing data as well as attempting to minimize response time
 633  *      while increasing bulk throughput.  This functionality is controlled
 634  *      by the InterruptThrottleRate module parameter.
 635  **/
 636 static unsigned int igbvf_update_itr(struct igbvf_adapter *adapter,
 637                                      u16 itr_setting, int packets,
 638                                      int bytes)
 639 {
 640         unsigned int retval = itr_setting;
 641
 642         if (packets == 0)
 643                 goto update_itr_done;
 644
 645         switch (itr_setting) {
 646         case lowest_latency:
 647                 /* handle TSO and jumbo frames */
 648                 if (bytes/packets > 8000)
 649                         retval = bulk_latency;
 650                 else if ((packets < 5) && (bytes > 512))
 651                         retval = low_latency;
 652                 break;
 653         case low_latency:  /* 50 usec aka 20000 ints/s */
 654                 if (bytes > 10000) {
 655                         /* this if handles the TSO accounting */
 656                         if (bytes/packets > 8000)
 657                                 retval = bulk_latency;
 658                         else if ((packets < 10) || ((bytes/packets) > 1200))
 659                                 retval = bulk_latency;
 660                         else if ((packets > 35))
 661                                 retval = lowest_latency;
 662                 } else if (bytes/packets > 2000) {
 663                         retval = bulk_latency;
 664                 } else if (packets <= 2 && bytes < 512) {
 665                         retval = lowest_latency;
 666                 }
 667                 break;
 668         case bulk_latency: /* 250 usec aka 4000 ints/s */
 669                 if (bytes > 25000) {
 670                         if (packets > 35)
 671                                 retval = low_latency;
 672                 } else if (bytes < 6000) {
 673                         retval = low_latency;
 674                 }
 675                 break;
 676         }
 677
 678 update_itr_done:
 679         return retval;
 680 }
 681
 682 static void igbvf_set_itr(struct igbvf_adapter *adapter)
 683 {
 684         struct e1000_hw *hw = &adapter->hw;
 685         u16 current_itr;
 686         u32 new_itr = adapter->itr;
 687
 688         adapter->tx_itr = igbvf_update_itr(adapter, adapter->tx_itr,
 689                                            adapter->total_tx_packets,
 690                                            adapter->total_tx_bytes);
 691         /* conservative mode (itr 3) eliminates the lowest_latency setting */
 692         if (adapter->itr_setting == 3 && adapter->tx_itr == lowest_latency)
 693                 adapter->tx_itr = low_latency;
 694
 695         adapter->rx_itr = igbvf_update_itr(adapter, adapter->rx_itr,
 696                                            adapter->total_rx_packets,
 697                                            adapter->total_rx_bytes);
 698         /* conservative mode (itr 3) eliminates the lowest_latency setting */
 699         if (adapter->itr_setting == 3 && adapter->rx_itr == lowest_latency)
 700                 adapter->rx_itr = low_latency;
 701
 702         current_itr = max(adapter->rx_itr, adapter->tx_itr);
 703
 704         switch (current_itr) {
 705         /* counts and packets in update_itr are dependent on these numbers */
 706         case lowest_latency:
 707                 new_itr = 70000;
 708                 break;
 709         case low_latency:
 710                 new_itr = 20000; /* aka hwitr = ~200 */
 711                 break;
 712         case bulk_latency:
 713                 new_itr = 4000;
 714                 break;
 715         default:
 716                 break;
 717         }
 718
 719         if (new_itr != adapter->itr) {
 720                 /*
 721                  * this attempts to bias the interrupt rate towards Bulk
 722                  * by adding intermediate steps when interrupt rate is
 723                  * increasing
 724                  */
 725                 new_itr = new_itr > adapter->itr ?
 726                              min(adapter->itr + (new_itr >> 2), new_itr) :
 727                              new_itr;
 728                 adapter->itr = new_itr;
 729                 adapter->rx_ring->itr_val = 1952;
 730
 731                 if (adapter->msix_entries)
 732                         adapter->rx_ring->set_itr = 1;
 733                 else
 734                         ew32(ITR, 1952);
 735         }
 736 }
 737
 738 /**
 739  * igbvf_clean_tx_irq - Reclaim resources after transmit completes
 740  * @adapter: board private structure
 741  * returns true if ring is completely cleaned
 742  **/
 743 static bool igbvf_clean_tx_irq(struct igbvf_ring *tx_ring)
 744 {
 745         struct igbvf_adapter *adapter = tx_ring->adapter;
 746         struct net_device *netdev = adapter->netdev;
 747         struct igbvf_buffer *buffer_info;
 748         struct sk_buff *skb;
 749         union e1000_adv_tx_desc *tx_desc, *eop_desc;
 750         unsigned int total_bytes = 0, total_packets = 0;
 751         unsigned int i, eop, count = 0;
 752         bool cleaned = false;
 753
 754         i = tx_ring->next_to_clean;
 755         eop = tx_ring->buffer_info[i].next_to_watch;
 756         eop_desc = IGBVF_TX_DESC_ADV(*tx_ring, eop);
 757
 758         while ((eop_desc->wb.status & cpu_to_le32(E1000_TXD_STAT_DD)) &&
 759                (count < tx_ring->count)) {
 760                 rmb();  /* read buffer_info after eop_desc status */
 761                 for (cleaned = false; !cleaned; count++) {
 762                         tx_desc = IGBVF_TX_DESC_ADV(*tx_ring, i);
 763                         buffer_info = &tx_ring->buffer_info[i];
 764                         cleaned = (i == eop);
 765                         skb = buffer_info->skb;
 766
 767                         if (skb) {
 768                                 unsigned int segs, bytecount;
 769
 770                                 /* gso_segs is currently only valid for tcp */
 771                                 segs = skb_shinfo(skb)->gso_segs ?: 1;
 772                                 /* multiply data chunks by size of headers */
 773                                 bytecount = ((segs - 1) * skb_headlen(skb)) +
 774                                             skb->len;
 775                                 total_packets += segs;
 776                                 total_bytes += bytecount;
 777                         }
 778
 779                         igbvf_put_txbuf(adapter, buffer_info);
 780                         tx_desc->wb.status = 0;
 781
 782                         i++;
 783                         if (i == tx_ring->count)
 784                                 i = 0;
 785                 }
 786                 eop = tx_ring->buffer_info[i].next_to_watch;
 787                 eop_desc = IGBVF_TX_DESC_ADV(*tx_ring, eop);
 788         }
 789
 790         tx_ring->next_to_clean = i;
 791
 792         if (unlikely(count &&
 793                      netif_carrier_ok(netdev) &&
 794                      igbvf_desc_unused(tx_ring) >= IGBVF_TX_QUEUE_WAKE)) {
 795                 /* Make sure that anybody stopping the queue after this
 796                  * sees the new next_to_clean.
 797                  */
 798                 smp_mb();
 799                 if (netif_queue_stopped(netdev) &&
 800                     !(test_bit(__IGBVF_DOWN, &adapter->state))) {
 801                         netif_wake_queue(netdev);
 802                         ++adapter->restart_queue;
 803                 }
 804         }
 805
 806         adapter->net_stats.tx_bytes += total_bytes;
 807         adapter->net_stats.tx_packets += total_packets;
 808         return count < tx_ring->count;
 809 }
 810
 811 static irqreturn_t igbvf_msix_other(int irq, void *data)
 812 {
 813         struct net_device *netdev = data;
 814         struct igbvf_adapter *adapter = netdev_priv(netdev);
 815         struct e1000_hw *hw = &adapter->hw;
 816
 817         adapter->int_counter1++;
 818
 819         netif_carrier_off(netdev);
 820         hw->mac.get_link_status = 1;
 821         if (!test_bit(__IGBVF_DOWN, &adapter->state))
 822                 mod_timer(&adapter->watchdog_timer, jiffies + 1);
 823
 824         ew32(EIMS, adapter->eims_other);
 825
 826         return IRQ_HANDLED;
 827 }
 828
 829 static irqreturn_t igbvf_intr_msix_tx(int irq, void *data)
 830 {
 831         struct net_device *netdev = data;
 832         struct igbvf_adapter *adapter = netdev_priv(netdev);
 833         struct e1000_hw *hw = &adapter->hw;
 834         struct igbvf_ring *tx_ring = adapter->tx_ring;
 835
 836
 837         adapter->total_tx_bytes = 0;
 838         adapter->total_tx_packets = 0;
 839
 840         /* auto mask will automatically reenable the interrupt when we write
 841          * EICS */
 842         if (!igbvf_clean_tx_irq(tx_ring))
 843                 /* Ring was not completely cleaned, so fire another interrupt */
 844                 ew32(EICS, tx_ring->eims_value);
 845         else
 846                 ew32(EIMS, tx_ring->eims_value);
 847
 848         return IRQ_HANDLED;
 849 }
 850
 851 static irqreturn_t igbvf_intr_msix_rx(int irq, void *data)
 852 {
 853         struct net_device *netdev = data;
 854         struct igbvf_adapter *adapter = netdev_priv(netdev);
 855
 856         adapter->int_counter0++;
 857
 858         /* Write the ITR value calculated at the end of the
 859          * previous interrupt.
 860          */
 861         if (adapter->rx_ring->set_itr) {
 862                 writel(adapter->rx_ring->itr_val,
 863                        adapter->hw.hw_addr + adapter->rx_ring->itr_register);
 864                 adapter->rx_ring->set_itr = 0;
 865         }
 866
 867         if (napi_schedule_prep(&adapter->rx_ring->napi)) {
 868                 adapter->total_rx_bytes = 0;
 869                 adapter->total_rx_packets = 0;
 870                 __napi_schedule(&adapter->rx_ring->napi);
 871         }
 872
 873         return IRQ_HANDLED;
 874 }
 875
 876 #define IGBVF_NO_QUEUE -1
 877
 878 static void igbvf_assign_vector(struct igbvf_adapter *adapter, int rx_queue,
 879                                 int tx_queue, int msix_vector)
 880 {
 881         struct e1000_hw *hw = &adapter->hw;
 882         u32 ivar, index;
 883
 884         /* 82576 uses a table-based method for assigning vectors.
 885            Each queue has a single entry in the table to which we write
 886            a vector number along with a "valid" bit.  Sadly, the layout
 887            of the table is somewhat counterintuitive. */
 888         if (rx_queue > IGBVF_NO_QUEUE) {
 889                 index = (rx_queue >> 1);
 890                 ivar = array_er32(IVAR0, index);
 891                 if (rx_queue & 0x1) {
 892                         /* vector goes into third byte of register */
 893                         ivar = ivar & 0xFF00FFFF;
 894                         ivar |= (msix_vector | E1000_IVAR_VALID) << 16;
 895                 } else {
 896                         /* vector goes into low byte of register */
 897                         ivar = ivar & 0xFFFFFF00;
 898                         ivar |= msix_vector | E1000_IVAR_VALID;
 899                 }
 900                 adapter->rx_ring[rx_queue].eims_value = 1 << msix_vector;
 901                 array_ew32(IVAR0, index, ivar);
 902         }
 903         if (tx_queue > IGBVF_NO_QUEUE) {
 904                 index = (tx_queue >> 1);
 905                 ivar = array_er32(IVAR0, index);
 906                 if (tx_queue & 0x1) {
 907                         /* vector goes into high byte of register */
 908                         ivar = ivar & 0x00FFFFFF;
 909                         ivar |= (msix_vector | E1000_IVAR_VALID) << 24;
 910                 } else {
 911                         /* vector goes into second byte of register */
 912                         ivar = ivar & 0xFFFF00FF;
 913                         ivar |= (msix_vector | E1000_IVAR_VALID) << 8;
 914                 }
 915                 adapter->tx_ring[tx_queue].eims_value = 1 << msix_vector;
 916                 array_ew32(IVAR0, index, ivar);
 917         }
 918 }
 919
 920 /**
 921  * igbvf_configure_msix - Configure MSI-X hardware
 922  *
 923  * igbvf_configure_msix sets up the hardware to properly
 924  * generate MSI-X interrupts.
 925  **/
 926 static void igbvf_configure_msix(struct igbvf_adapter *adapter)
 927 {
 928         u32 tmp;
 929         struct e1000_hw *hw = &adapter->hw;
 930         struct igbvf_ring *tx_ring = adapter->tx_ring;
 931         struct igbvf_ring *rx_ring = adapter->rx_ring;
 932         int vector = 0;
 933
 934         adapter->eims_enable_mask = 0;
 935
 936         igbvf_assign_vector(adapter, IGBVF_NO_QUEUE, 0, vector++);
 937         adapter->eims_enable_mask |= tx_ring->eims_value;
 938         if (tx_ring->itr_val)
 939                 writel(tx_ring->itr_val,
 940                        hw->hw_addr + tx_ring->itr_register);
 941         else
 942                 writel(1952, hw->hw_addr + tx_ring->itr_register);
 943
 944         igbvf_assign_vector(adapter, 0, IGBVF_NO_QUEUE, vector++);
 945         adapter->eims_enable_mask |= rx_ring->eims_value;
 946         if (rx_ring->itr_val)
 947                 writel(rx_ring->itr_val,
 948                        hw->hw_addr + rx_ring->itr_register);
 949         else
 950                 writel(1952, hw->hw_addr + rx_ring->itr_register);
 951
 952         /* set vector for other causes, i.e. link changes */
 953
 954         tmp = (vector++ | E1000_IVAR_VALID);
 955
 956         ew32(IVAR_MISC, tmp);
 957
 958         adapter->eims_enable_mask = (1 << (vector)) - 1;
 959         adapter->eims_other = 1 << (vector - 1);
 960         e1e_flush();
 961 }
 962
 963 static void igbvf_reset_interrupt_capability(struct igbvf_adapter *adapter)
 964 {
 965         if (adapter->msix_entries) {
 966                 pci_disable_msix(adapter->pdev);
 967                 kfree(adapter->msix_entries);
 968                 adapter->msix_entries = NULL;
 969         }
 970 }
 971
 972 /**
 973  * igbvf_set_interrupt_capability - set MSI or MSI-X if supported
 974  *
 975  * Attempt to configure interrupts using the best available
 976  * capabilities of the hardware and kernel.
 977  **/
 978 static void igbvf_set_interrupt_capability(struct igbvf_adapter *adapter)
 979 {
 980         int err = -ENOMEM;
 981         int i;
 982
 983         /* we allocate 3 vectors, 1 for tx, 1 for rx, one for pf messages */
 984         adapter->msix_entries = kcalloc(3, sizeof(struct msix_entry),
 985                                         GFP_KERNEL);
 986         if (adapter->msix_entries) {
 987                 for (i = 0; i < 3; i++)
 988                         adapter->msix_entries[i].entry = i;
 989
 990                 err = pci_enable_msix(adapter->pdev,
 991                                       adapter->msix_entries, 3);
 992         }
 993
 994         if (err) {
 995                 /* MSI-X failed */
 996                 dev_err(&adapter->pdev->dev,
 997                         "Failed to initialize MSI-X interrupts.\n");
 998                 igbvf_reset_interrupt_capability(adapter);
 999         }
1000 }
1001
1002 /**
1003  * igbvf_request_msix - Initialize MSI-X interrupts
1004  *
1005  * igbvf_request_msix allocates MSI-X vectors and requests interrupts from the
1006  * kernel.
1007  **/
1008 static int igbvf_request_msix(struct igbvf_adapter *adapter)
1009 {
1010         struct net_device *netdev = adapter->netdev;
1011         int err = 0, vector = 0;
1012
1013         if (strlen(netdev->name) < (IFNAMSIZ - 5)) {
1014                 sprintf(adapter->tx_ring->name, "%s-tx-0", netdev->name);
1015                 sprintf(adapter->rx_ring->name, "%s-rx-0", netdev->name);
1016         } else {
1017                 memcpy(adapter->tx_ring->name, netdev->name, IFNAMSIZ);
1018                 memcpy(adapter->rx_ring->name, netdev->name, IFNAMSIZ);
1019         }
1020
1021         err = request_irq(adapter->msix_entries[vector].vector,
1022                           igbvf_intr_msix_tx, 0, adapter->tx_ring->name,
1023                           netdev);
1024         if (err)
1025                 goto out;
1026
1027         adapter->tx_ring->itr_register = E1000_EITR(vector);
1028         adapter->tx_ring->itr_val = 1952;
1029         vector++;
1030
1031         err = request_irq(adapter->msix_entries[vector].vector,
1032                           igbvf_intr_msix_rx, 0, adapter->rx_ring->name,
1033                           netdev);
1034         if (err)
1035                 goto out;
1036
1037         adapter->rx_ring->itr_register = E1000_EITR(vector);
1038         adapter->rx_ring->itr_val = 1952;
1039         vector++;
1040
1041         err = request_irq(adapter->msix_entries[vector].vector,
1042                           igbvf_msix_other, 0, netdev->name, netdev);
1043         if (err)
1044                 goto out;
1045
1046         igbvf_configure_msix(adapter);
1047         return 0;
1048 out:
1049         return err;
1050 }
1051
1052 /**
1053  * igbvf_alloc_queues - Allocate memory for all rings
1054  * @adapter: board private structure to initialize
1055  **/
1056 static int __devinit igbvf_alloc_queues(struct igbvf_adapter *adapter)
1057 {
1058         struct net_device *netdev = adapter->netdev;
1059
1060         adapter->tx_ring = kzalloc(sizeof(struct igbvf_ring), GFP_KERNEL);
1061         if (!adapter->tx_ring)
1062                 return -ENOMEM;
1063
1064         adapter->rx_ring = kzalloc(sizeof(struct igbvf_ring), GFP_KERNEL);
1065         if (!adapter->rx_ring) {
1066                 kfree(adapter->tx_ring);
1067                 return -ENOMEM;
1068         }
1069
1070         netif_napi_add(netdev, &adapter->rx_ring->napi, igbvf_poll, 64);
1071
1072         return 0;
1073 }
1074
1075 /**
1076  * igbvf_request_irq - initialize interrupts
1077  *
1078  * Attempts to configure interrupts using the best available
1079  * capabilities of the hardware and kernel.
1080  **/
1081 static int igbvf_request_irq(struct igbvf_adapter *adapter)
1082 {
1083         int err = -1;
1084
1085         /* igbvf supports msi-x only */
1086         if (adapter->msix_entries)
1087                 err = igbvf_request_msix(adapter);
1088
1089         if (!err)
1090                 return err;
1091
1092         dev_err(&adapter->pdev->dev,
1093                 "Unable to allocate interrupt, Error: %d\n", err);
1094
1095         return err;
1096 }
1097
1098 static void igbvf_free_irq(struct igbvf_adapter *adapter)
1099 {
1100         struct net_device *netdev = adapter->netdev;
1101         int vector;
1102
1103         if (adapter->msix_entries) {
1104                 for (vector = 0; vector < 3; vector++)
1105                         free_irq(adapter->msix_entries[vector].vector, netdev);
1106         }
1107 }
1108
1109 /**
1110  * igbvf_irq_disable - Mask off interrupt generation on the NIC
1111  **/
1112 static void igbvf_irq_disable(struct igbvf_adapter *adapter)
1113 {
1114         struct e1000_hw *hw = &adapter->hw;
1115
1116         ew32(EIMC, ~0);
1117
1118         if (adapter->msix_entries)
1119                 ew32(EIAC, 0);
1120 }
1121
1122 /**
1123  * igbvf_irq_enable - Enable default interrupt generation settings
1124  **/
1125 static void igbvf_irq_enable(struct igbvf_adapter *adapter)
1126 {
1127         struct e1000_hw *hw = &adapter->hw;
1128
1129         ew32(EIAC, adapter->eims_enable_mask);
1130         ew32(EIAM, adapter->eims_enable_mask);
1131         ew32(EIMS, adapter->eims_enable_mask);
1132 }
1133
1134 /**
1135  * igbvf_poll - NAPI Rx polling callback
1136  * @napi: struct associated with this polling callback
1137  * @budget: amount of packets driver is allowed to process this poll
1138  **/
1139 static int igbvf_poll(struct napi_struct *napi, int budget)
1140 {
1141         struct igbvf_ring *rx_ring = container_of(napi, struct igbvf_ring, napi);
1142         struct igbvf_adapter *adapter = rx_ring->adapter;
1143         struct e1000_hw *hw = &adapter->hw;
1144         int work_done = 0;
1145
1146         igbvf_clean_rx_irq(adapter, &work_done, budget);
1147
1148         /* If not enough Rx work done, exit the polling mode */
1149         if (work_done < budget) {
1150                 napi_complete(napi);
1151
1152                 if (adapter->itr_setting & 3)
1153                         igbvf_set_itr(adapter);
1154
1155                 if (!test_bit(__IGBVF_DOWN, &adapter->state))
1156                         ew32(EIMS, adapter->rx_ring->eims_value);
1157         }
1158
1159         return work_done;
1160 }
1161
1162 /**
1163  * igbvf_set_rlpml - set receive large packet maximum length
1164  * @adapter: board private structure
1165  *
1166  * Configure the maximum size of packets that will be received
1167  */
1168 static void igbvf_set_rlpml(struct igbvf_adapter *adapter)
1169 {
1170         int max_frame_size;
1171         struct e1000_hw *hw = &adapter->hw;
1172
1173         max_frame_size = adapter->max_frame_size + VLAN_TAG_SIZE;
1174         e1000_rlpml_set_vf(hw, max_frame_size);
1175 }
1176
1177 static void igbvf_vlan_rx_add_vid(struct net_device *netdev, u16 vid)
1178 {
1179         struct igbvf_adapter *adapter = netdev_priv(netdev);
1180         struct e1000_hw *hw = &adapter->hw;
1181
1182         if (hw->mac.ops.set_vfta(hw, vid, true))
1183                 dev_err(&adapter->pdev->dev, "Failed to add vlan id %d\n", vid);
1184         else
1185                 set_bit(vid, adapter->active_vlans);
1186 }
1187
1188 static void igbvf_vlan_rx_kill_vid(struct net_device *netdev, u16 vid)
1189 {
1190         struct igbvf_adapter *adapter = netdev_priv(netdev);
1191         struct e1000_hw *hw = &adapter->hw;
1192
1193         igbvf_irq_disable(adapter);
1194
1195         if (!test_bit(__IGBVF_DOWN, &adapter->state))
1196                 igbvf_irq_enable(adapter);
1197
1198         if (hw->mac.ops.set_vfta(hw, vid, false))
1199                 dev_err(&adapter->pdev->dev,
1200                         "Failed to remove vlan id %d\n", vid);
1201         else
1202                 clear_bit(vid, adapter->active_vlans);
1203 }
1204
1205 static void igbvf_restore_vlan(struct igbvf_adapter *adapter)
1206 {
1207         u16 vid;
1208
1209         for_each_set_bit(vid, adapter->active_vlans, VLAN_N_VID)
1210                 igbvf_vlan_rx_add_vid(adapter->netdev, vid);
1211 }
1212
1213 /**
1214  * igbvf_configure_tx - Configure Transmit Unit after Reset
1215  * @adapter: board private structure
1216  *
1217  * Configure the Tx unit of the MAC after a reset.
1218  **/
1219 static void igbvf_configure_tx(struct igbvf_adapter *adapter)
1220 {
1221         struct e1000_hw *hw = &adapter->hw;
1222         struct igbvf_ring *tx_ring = adapter->tx_ring;
1223         u64 tdba;
1224         u32 txdctl, dca_txctrl;
1225
1226         /* disable transmits */
1227         txdctl = er32(TXDCTL(0));
1228         ew32(TXDCTL(0), txdctl & ~E1000_TXDCTL_QUEUE_ENABLE);
1229         e1e_flush();
1230         msleep(10);
1231
1232         /* Setup the HW Tx Head and Tail descriptor pointers */
1233         ew32(TDLEN(0), tx_ring->count * sizeof(union e1000_adv_tx_desc));
1234         tdba = tx_ring->dma;
1235         ew32(TDBAL(0), (tdba & DMA_BIT_MASK(32)));
1236         ew32(TDBAH(0), (tdba >> 32));
1237         ew32(TDH(0), 0);
1238         ew32(TDT(0), 0);
1239         tx_ring->head = E1000_TDH(0);
1240         tx_ring->tail = E1000_TDT(0);
1241
1242         /* Turn off Relaxed Ordering on head write-backs.  The writebacks
1243          * MUST be delivered in order or it will completely screw up
1244          * our bookeeping.
1245          */
1246         dca_txctrl = er32(DCA_TXCTRL(0));
1247         dca_txctrl &= ~E1000_DCA_TXCTRL_TX_WB_RO_EN;
1248         ew32(DCA_TXCTRL(0), dca_txctrl);
1249
1250         /* enable transmits */
1251         txdctl |= E1000_TXDCTL_QUEUE_ENABLE;
1252         ew32(TXDCTL(0), txdctl);
1253
1254         /* Setup Transmit Descriptor Settings for eop descriptor */
1255         adapter->txd_cmd = E1000_ADVTXD_DCMD_EOP | E1000_ADVTXD_DCMD_IFCS;
1256
1257         /* enable Report Status bit */
1258         adapter->txd_cmd |= E1000_ADVTXD_DCMD_RS;
1259 }
1260
1261 /**
1262  * igbvf_setup_srrctl - configure the receive control registers
1263  * @adapter: Board private structure
1264  **/
1265 static void igbvf_setup_srrctl(struct igbvf_adapter *adapter)
1266 {
1267         struct e1000_hw *hw = &adapter->hw;
1268         u32 srrctl = 0;
1269
1270         srrctl &= ~(E1000_SRRCTL_DESCTYPE_MASK |
1271                     E1000_SRRCTL_BSIZEHDR_MASK |
1272                     E1000_SRRCTL_BSIZEPKT_MASK);
1273
1274         /* Enable queue drop to avoid head of line blocking */
1275         srrctl |= E1000_SRRCTL_DROP_EN;
1276
1277         /* Setup buffer sizes */
1278         srrctl |= ALIGN(adapter->rx_buffer_len, 1024) >>
1279                   E1000_SRRCTL_BSIZEPKT_SHIFT;
1280
1281         if (adapter->rx_buffer_len < 2048) {
1282                 adapter->rx_ps_hdr_size = 0;
1283                 srrctl |= E1000_SRRCTL_DESCTYPE_ADV_ONEBUF;
1284         } else {
1285                 adapter->rx_ps_hdr_size = 128;
1286                 srrctl |= adapter->rx_ps_hdr_size <<
1287                           E1000_SRRCTL_BSIZEHDRSIZE_SHIFT;
1288                 srrctl |= E1000_SRRCTL_DESCTYPE_HDR_SPLIT_ALWAYS;
1289         }
1290
1291         ew32(SRRCTL(0), srrctl);
1292 }
1293
1294 /**
1295  * igbvf_configure_rx - Configure Receive Unit after Reset
1296  * @adapter: board private structure
1297  *
1298  * Configure the Rx unit of the MAC after a reset.
1299  **/
1300 static void igbvf_configure_rx(struct igbvf_adapter *adapter)
1301 {
1302         struct e1000_hw *hw = &adapter->hw;
1303         struct igbvf_ring *rx_ring = adapter->rx_ring;
1304         u64 rdba;
1305         u32 rdlen, rxdctl;
1306
1307         /* disable receives */
1308         rxdctl = er32(RXDCTL(0));
1309         ew32(RXDCTL(0), rxdctl & ~E1000_RXDCTL_QUEUE_ENABLE);
1310         e1e_flush();
1311         msleep(10);
1312
1313         rdlen = rx_ring->count * sizeof(union e1000_adv_rx_desc);
1314
1315         /*
1316          * Setup the HW Rx Head and Tail Descriptor Pointers and
1317          * the Base and Length of the Rx Descriptor Ring
1318          */
1319         rdba = rx_ring->dma;
1320         ew32(RDBAL(0), (rdba & DMA_BIT_MASK(32)));
1321         ew32(RDBAH(0), (rdba >> 32));
1322         ew32(RDLEN(0), rx_ring->count * sizeof(union e1000_adv_rx_desc));
1323         rx_ring->head = E1000_RDH(0);
1324         rx_ring->tail = E1000_RDT(0);
1325         ew32(RDH(0), 0);
1326         ew32(RDT(0), 0);
1327
1328         rxdctl |= E1000_RXDCTL_QUEUE_ENABLE;
1329         rxdctl &= 0xFFF00000;
1330         rxdctl |= IGBVF_RX_PTHRESH;
1331         rxdctl |= IGBVF_RX_HTHRESH << 8;
1332         rxdctl |= IGBVF_RX_WTHRESH << 16;
1333
1334         igbvf_set_rlpml(adapter);
1335
1336         /* enable receives */
1337         ew32(RXDCTL(0), rxdctl);
1338 }
1339
1340 /**
1341  * igbvf_set_multi - Multicast and Promiscuous mode set
1342  * @netdev: network interface device structure
1343  *
1344  * The set_multi entry point is called whenever the multicast address
1345  * list or the network interface flags are updated.  This routine is
1346  * responsible for configuring the hardware for proper multicast,
1347  * promiscuous mode, and all-multi behavior.
1348  **/
1349 static void igbvf_set_multi(struct net_device *netdev)
1350 {
1351         struct igbvf_adapter *adapter = netdev_priv(netdev);
1352         struct e1000_hw *hw = &adapter->hw;
1353         struct netdev_hw_addr *ha;
1354         u8  *mta_list = NULL;
1355         int i;
1356
1357         if (!netdev_mc_empty(netdev)) {
1358                 mta_list = kmalloc(netdev_mc_count(netdev) * 6, GFP_ATOMIC);
1359                 if (!mta_list) {
1360                         dev_err(&adapter->pdev->dev,
1361                                 "failed to allocate multicast filter list\n");
1362                         return;
1363                 }
1364         }
1365
1366         /* prepare a packed array of only addresses. */
1367         i = 0;
1368         netdev_for_each_mc_addr(ha, netdev)
1369                 memcpy(mta_list + (i++ * ETH_ALEN), ha->addr, ETH_ALEN);
1370
1371         hw->mac.ops.update_mc_addr_list(hw, mta_list, i, 0, 0);
1372         kfree(mta_list);
1373 }
1374
1375 /**
1376  * igbvf_configure - configure the hardware for Rx and Tx
1377  * @adapter: private board structure
1378  **/
1379 static void igbvf_configure(struct igbvf_adapter *adapter)
1380 {
1381         igbvf_set_multi(adapter->netdev);
1382
1383         igbvf_restore_vlan(adapter);
1384
1385         igbvf_configure_tx(adapter);
1386         igbvf_setup_srrctl(adapter);
1387         igbvf_configure_rx(adapter);
1388         igbvf_alloc_rx_buffers(adapter->rx_ring,
1389                                igbvf_desc_unused(adapter->rx_ring));
1390 }
1391
1392 /* igbvf_reset - bring the hardware into a known good state
1393  *
1394  * This function boots the hardware and enables some settings that
1395  * require a configuration cycle of the hardware - those cannot be
1396  * set/changed during runtime. After reset the device needs to be
1397  * properly configured for Rx, Tx etc.
1398  */
1399 static void igbvf_reset(struct igbvf_adapter *adapter)
1400 {
1401         struct e1000_mac_info *mac = &adapter->hw.mac;
1402         struct net_device *netdev = adapter->netdev;
1403         struct e1000_hw *hw = &adapter->hw;
1404
1405         /* Allow time for pending master requests to run */
1406         if (mac->ops.reset_hw(hw))
1407                 dev_err(&adapter->pdev->dev, "PF still resetting\n");
1408
1409         mac->ops.init_hw(hw);
1410
1411         if (is_valid_ether_addr(adapter->hw.mac.addr)) {
1412                 memcpy(netdev->dev_addr, adapter->hw.mac.addr,
1413                        netdev->addr_len);
1414                 memcpy(netdev->perm_addr, adapter->hw.mac.addr,
1415                        netdev->addr_len);
1416         }
1417
1418         adapter->last_reset = jiffies;
1419 }
1420
1421 int igbvf_up(struct igbvf_adapter *adapter)
1422 {
1423         struct e1000_hw *hw = &adapter->hw;
1424
1425         /* hardware has been reset, we need to reload some things */
1426         igbvf_configure(adapter);
1427
1428         clear_bit(__IGBVF_DOWN, &adapter->state);
1429
1430         napi_enable(&adapter->rx_ring->napi);
1431         if (adapter->msix_entries)
1432                 igbvf_configure_msix(adapter);
1433
1434         /* Clear any pending interrupts. */
1435         er32(EICR);
1436         igbvf_irq_enable(adapter);
1437
1438         /* start the watchdog */
1439         hw->mac.get_link_status = 1;
1440         mod_timer(&adapter->watchdog_timer, jiffies + 1);
1441
1442
1443         return 0;
1444 }
1445
1446 void igbvf_down(struct igbvf_adapter *adapter)
1447 {
1448         struct net_device *netdev = adapter->netdev;
1449         struct e1000_hw *hw = &adapter->hw;
1450         u32 rxdctl, txdctl;
1451
1452         /*
1453          * signal that we're down so the interrupt handler does not
1454          * reschedule our watchdog timer
1455          */
1456         set_bit(__IGBVF_DOWN, &adapter->state);
1457
1458         /* disable receives in the hardware */
1459         rxdctl = er32(RXDCTL(0));
1460         ew32(RXDCTL(0), rxdctl & ~E1000_RXDCTL_QUEUE_ENABLE);
1461
1462         netif_stop_queue(netdev);
1463
1464         /* disable transmits in the hardware */
1465         txdctl = er32(TXDCTL(0));
1466         ew32(TXDCTL(0), txdctl & ~E1000_TXDCTL_QUEUE_ENABLE);
1467
1468         /* flush both disables and wait for them to finish */
1469         e1e_flush();
1470         msleep(10);
1471
1472         napi_disable(&adapter->rx_ring->napi);
1473
1474         igbvf_irq_disable(adapter);
1475
1476         del_timer_sync(&adapter->watchdog_timer);
1477
1478         netif_carrier_off(netdev);
1479
1480         /* record the stats before reset*/
1481         igbvf_update_stats(adapter);
1482
1483         adapter->link_speed = 0;
1484         adapter->link_duplex = 0;
1485
1486         igbvf_reset(adapter);
1487         igbvf_clean_tx_ring(adapter->tx_ring);
1488         igbvf_clean_rx_ring(adapter->rx_ring);
1489 }
1490
1491 void igbvf_reinit_locked(struct igbvf_adapter *adapter)
1492 {
1493         might_sleep();
1494         while (test_and_set_bit(__IGBVF_RESETTING, &adapter->state))
1495                 msleep(1);
1496         igbvf_down(adapter);
1497         igbvf_up(adapter);
1498         clear_bit(__IGBVF_RESETTING, &adapter->state);
1499 }
1500
1501 /**
1502  * igbvf_sw_init - Initialize general software structures (struct igbvf_adapter)
1503  * @adapter: board private structure to initialize
1504  *
1505  * igbvf_sw_init initializes the Adapter private data structure.
1506  * Fields are initialized based on PCI device information and
1507  * OS network device settings (MTU size).
1508  **/
1509 static int __devinit igbvf_sw_init(struct igbvf_adapter *adapter)
1510 {
1511         struct net_device *netdev = adapter->netdev;
1512         s32 rc;
1513
1514         adapter->rx_buffer_len = ETH_FRAME_LEN + VLAN_HLEN + ETH_FCS_LEN;
1515         adapter->rx_ps_hdr_size = 0;
1516         adapter->max_frame_size = netdev->mtu + ETH_HLEN + ETH_FCS_LEN;
1517         adapter->min_frame_size = ETH_ZLEN + ETH_FCS_LEN;
1518
1519         adapter->tx_int_delay = 8;
1520         adapter->tx_abs_int_delay = 32;
1521         adapter->rx_int_delay = 0;
1522         adapter->rx_abs_int_delay = 8;
1523         adapter->itr_setting = 3;
1524         adapter->itr = 20000;
1525
1526         /* Set various function pointers */
1527         adapter->ei->init_ops(&adapter->hw);
1528
1529         rc = adapter->hw.mac.ops.init_params(&adapter->hw);
1530         if (rc)
1531                 return rc;
1532
1533         rc = adapter->hw.mbx.ops.init_params(&adapter->hw);
1534         if (rc)
1535                 return rc;
1536
1537         igbvf_set_interrupt_capability(adapter);
1538
1539         if (igbvf_alloc_queues(adapter))
1540                 return -ENOMEM;
1541
1542         spin_lock_init(&adapter->tx_queue_lock);
1543
1544         /* Explicitly disable IRQ since the NIC can be in any state. */
1545         igbvf_irq_disable(adapter);
1546
1547         spin_lock_init(&adapter->stats_lock);
1548
1549         set_bit(__IGBVF_DOWN, &adapter->state);
1550         return 0;
1551 }
1552
1553 static void igbvf_initialize_last_counter_stats(struct igbvf_adapter *adapter)
1554 {
1555         struct e1000_hw *hw = &adapter->hw;
1556
1557         adapter->stats.last_gprc = er32(VFGPRC);
1558         adapter->stats.last_gorc = er32(VFGORC);
1559         adapter->stats.last_gptc = er32(VFGPTC);
1560         adapter->stats.last_gotc = er32(VFGOTC);
1561         adapter->stats.last_mprc = er32(VFMPRC);
1562         adapter->stats.last_gotlbc = er32(VFGOTLBC);
1563         adapter->stats.last_gptlbc = er32(VFGPTLBC);
1564         adapter->stats.last_gorlbc = er32(VFGORLBC);
1565         adapter->stats.last_gprlbc = er32(VFGPRLBC);
1566
1567         adapter->stats.base_gprc = er32(VFGPRC);
1568         adapter->stats.base_gorc = er32(VFGORC);
1569         adapter->stats.base_gptc = er32(VFGPTC);
1570         adapter->stats.base_gotc = er32(VFGOTC);
1571         adapter->stats.base_mprc = er32(VFMPRC);
1572         adapter->stats.base_gotlbc = er32(VFGOTLBC);
1573         adapter->stats.base_gptlbc = er32(VFGPTLBC);
1574         adapter->stats.base_gorlbc = er32(VFGORLBC);
1575         adapter->stats.base_gprlbc = er32(VFGPRLBC);
1576 }
1577
1578 /**
1579  * igbvf_open - Called when a network interface is made active
1580  * @netdev: network interface device structure
1581  *
1582  * Returns 0 on success, negative value on failure
1583  *
1584  * The open entry point is called when a network interface is made
1585  * active by the system (IFF_UP).  At this point all resources needed
1586  * for transmit and receive operations are allocated, the interrupt
1587  * handler is registered with the OS, the watchdog timer is started,
1588  * and the stack is notified that the interface is ready.
1589  **/
1590 static int igbvf_open(struct net_device *netdev)
1591 {
1592         struct igbvf_adapter *adapter = netdev_priv(netdev);
1593         struct e1000_hw *hw = &adapter->hw;
1594         int err;
1595
1596         /* disallow open during test */
1597         if (test_bit(__IGBVF_TESTING, &adapter->state))
1598                 return -EBUSY;
1599
1600         /* allocate transmit descriptors */
1601         err = igbvf_setup_tx_resources(adapter, adapter->tx_ring);
1602         if (err)
1603                 goto err_setup_tx;
1604
1605         /* allocate receive descriptors */
1606         err = igbvf_setup_rx_resources(adapter, adapter->rx_ring);
1607         if (err)
1608                 goto err_setup_rx;
1609
1610         /*
1611          * before we allocate an interrupt, we must be ready to handle it.
1612          * Setting DEBUG_SHIRQ in the kernel makes it fire an interrupt
1613          * as soon as we call pci_request_irq, so we have to setup our
1614          * clean_rx handler before we do so.
1615          */
1616         igbvf_configure(adapter);
1617
1618         err = igbvf_request_irq(adapter);
1619         if (err)
1620                 goto err_req_irq;
1621
1622         /* From here on the code is the same as igbvf_up() */
1623         clear_bit(__IGBVF_DOWN, &adapter->state);
1624
1625         napi_enable(&adapter->rx_ring->napi);
1626
1627         /* clear any pending interrupts */
1628         er32(EICR);
1629
1630         igbvf_irq_enable(adapter);
1631
1632         /* start the watchdog */
1633         hw->mac.get_link_status = 1;
1634         mod_timer(&adapter->watchdog_timer, jiffies + 1);
1635
1636         return 0;
1637
1638 err_req_irq:
1639         igbvf_free_rx_resources(adapter->rx_ring);
1640 err_setup_rx:
1641         igbvf_free_tx_resources(adapter->tx_ring);
1642 err_setup_tx:
1643         igbvf_reset(adapter);
1644
1645         return err;
1646 }
1647
1648 /**
1649  * igbvf_close - Disables a network interface
1650  * @netdev: network interface device structure
1651  *
1652  * Returns 0, this is not allowed to fail
1653  *
1654  * The close entry point is called when an interface is de-activated
1655  * by the OS.  The hardware is still under the drivers control, but
1656  * needs to be disabled.  A global MAC reset is issued to stop the
1657  * hardware, and all transmit and receive resources are freed.
1658  **/
1659 static int igbvf_close(struct net_device *netdev)
1660 {
1661         struct igbvf_adapter *adapter = netdev_priv(netdev);
1662
1663         WARN_ON(test_bit(__IGBVF_RESETTING, &adapter->state));
1664         igbvf_down(adapter);
1665
1666         igbvf_free_irq(adapter);
1667
1668         igbvf_free_tx_resources(adapter->tx_ring);
1669         igbvf_free_rx_resources(adapter->rx_ring);
1670
1671         return 0;
1672 }
1673 /**
1674  * igbvf_set_mac - Change the Ethernet Address of the NIC
1675  * @netdev: network interface device structure
1676  * @p: pointer to an address structure
1677  *
1678  * Returns 0 on success, negative on failure
1679  **/
1680 static int igbvf_set_mac(struct net_device *netdev, void *p)
1681 {
1682         struct igbvf_adapter *adapter = netdev_priv(netdev);
1683         struct e1000_hw *hw = &adapter->hw;
1684         struct sockaddr *addr = p;
1685
1686         if (!is_valid_ether_addr(addr->sa_data))
1687                 return -EADDRNOTAVAIL;
1688
1689         memcpy(hw->mac.addr, addr->sa_data, netdev->addr_len);
1690
1691         hw->mac.ops.rar_set(hw, hw->mac.addr, 0);
1692
1693         if (memcmp(addr->sa_data, hw->mac.addr, 6))
1694                 return -EADDRNOTAVAIL;
1695
1696         memcpy(netdev->dev_addr, addr->sa_data, netdev->addr_len);
1697
1698         return 0;
1699 }
1700
1701 #define UPDATE_VF_COUNTER(reg, name)                                    \
1702         {                                                               \
1703                 u32 current_counter = er32(reg);                        \
1704                 if (current_counter < adapter->stats.last_##name)       \
1705                         adapter->stats.name += 0x100000000LL;           \
1706                 adapter->stats.last_##name = current_counter;           \
1707                 adapter->stats.name &= 0xFFFFFFFF00000000LL;            \
1708                 adapter->stats.name |= current_counter;                 \
1709         }
1710
1711 /**
1712  * igbvf_update_stats - Update the board statistics counters
1713  * @adapter: board private structure
1714 **/
1715 void igbvf_update_stats(struct igbvf_adapter *adapter)
1716 {
1717         struct e1000_hw *hw = &adapter->hw;
1718         struct pci_dev *pdev = adapter->pdev;
1719
1720         /*
1721          * Prevent stats update while adapter is being reset, link is down
1722          * or if the pci connection is down.
1723          */
1724         if (adapter->link_speed == 0)
1725                 return;
1726
1727         if (test_bit(__IGBVF_RESETTING, &adapter->state))
1728                 return;
1729
1730         if (pci_channel_offline(pdev))
1731                 return;
1732
1733         UPDATE_VF_COUNTER(VFGPRC, gprc);
1734         UPDATE_VF_COUNTER(VFGORC, gorc);
1735         UPDATE_VF_COUNTER(VFGPTC, gptc);
1736         UPDATE_VF_COUNTER(VFGOTC, gotc);
1737         UPDATE_VF_COUNTER(VFMPRC, mprc);
1738         UPDATE_VF_COUNTER(VFGOTLBC, gotlbc);
1739         UPDATE_VF_COUNTER(VFGPTLBC, gptlbc);
1740         UPDATE_VF_COUNTER(VFGORLBC, gorlbc);
1741         UPDATE_VF_COUNTER(VFGPRLBC, gprlbc);
1742
1743         /* Fill out the OS statistics structure */
1744         adapter->net_stats.multicast = adapter->stats.mprc;
1745 }
1746
1747 static void igbvf_print_link_info(struct igbvf_adapter *adapter)
1748 {
1749         dev_info(&adapter->pdev->dev, "Link is Up %d Mbps %s\n",
1750                  adapter->link_speed,
1751                  ((adapter->link_duplex == FULL_DUPLEX) ?
1752                   "Full Duplex" : "Half Duplex"));
1753 }
1754
1755 static bool igbvf_has_link(struct igbvf_adapter *adapter)
1756 {
1757         struct e1000_hw *hw = &adapter->hw;
1758         s32 ret_val = E1000_SUCCESS;
1759         bool link_active;
1760
1761         /* If interface is down, stay link down */
1762         if (test_bit(__IGBVF_DOWN, &adapter->state))
1763                 return false;
1764
1765         ret_val = hw->mac.ops.check_for_link(hw);
1766         link_active = !hw->mac.get_link_status;
1767
1768         /* if check for link returns error we will need to reset */
1769         if (ret_val && time_after(jiffies, adapter->last_reset + (10 * HZ)))
1770                 schedule_work(&adapter->reset_task);
1771
1772         return link_active;
1773 }
1774
1775 /**
1776  * igbvf_watchdog - Timer Call-back
1777  * @data: pointer to adapter cast into an unsigned long
1778  **/
1779 static void igbvf_watchdog(unsigned long data)
1780 {
1781         struct igbvf_adapter *adapter = (struct igbvf_adapter *) data;
1782
1783         /* Do the rest outside of interrupt context */
1784         schedule_work(&adapter->watchdog_task);
1785 }
1786
1787 static void igbvf_watchdog_task(struct work_struct *work)
1788 {
1789         struct igbvf_adapter *adapter = container_of(work,
1790                                                      struct igbvf_adapter,
1791                                                      watchdog_task);
1792         struct net_device *netdev = adapter->netdev;
1793         struct e1000_mac_info *mac = &adapter->hw.mac;
1794         struct igbvf_ring *tx_ring = adapter->tx_ring;
1795         struct e1000_hw *hw = &adapter->hw;
1796         u32 link;
1797         int tx_pending = 0;
1798
1799         link = igbvf_has_link(adapter);
1800
1801         if (link) {
1802                 if (!netif_carrier_ok(netdev)) {
1803                         mac->ops.get_link_up_info(&adapter->hw,
1804                                                   &adapter->link_speed,
1805                                                   &adapter->link_duplex);
1806                         igbvf_print_link_info(adapter);
1807
1808                         netif_carrier_on(netdev);
1809                         netif_wake_queue(netdev);
1810                 }
1811         } else {
1812                 if (netif_carrier_ok(netdev)) {
1813                         adapter->link_speed = 0;
1814                         adapter->link_duplex = 0;
1815                         dev_info(&adapter->pdev->dev, "Link is Down\n");
1816                         netif_carrier_off(netdev);
1817                         netif_stop_queue(netdev);
1818                 }
1819         }
1820
1821         if (netif_carrier_ok(netdev)) {
1822                 igbvf_update_stats(adapter);
1823         } else {
1824                 tx_pending = (igbvf_desc_unused(tx_ring) + 1 <
1825                               tx_ring->count);
1826                 if (tx_pending) {
1827                         /*
1828                          * We've lost link, so the controller stops DMA,
1829                          * but we've got queued Tx work that's never going
1830                          * to get done, so reset controller to flush Tx.
1831                          * (Do the reset outside of interrupt context).
1832                          */
1833                         adapter->tx_timeout_count++;
1834                         schedule_work(&adapter->reset_task);
1835                 }
1836         }
1837
1838         /* Cause software interrupt to ensure Rx ring is cleaned */
1839         ew32(EICS, adapter->rx_ring->eims_value);
1840
1841         /* Reset the timer */
1842         if (!test_bit(__IGBVF_DOWN, &adapter->state))
1843                 mod_timer(&adapter->watchdog_timer,
1844                           round_jiffies(jiffies + (2 * HZ)));
1845 }
1846
1847 #define IGBVF_TX_FLAGS_CSUM             0x00000001
1848 #define IGBVF_TX_FLAGS_VLAN             0x00000002
1849 #define IGBVF_TX_FLAGS_TSO              0x00000004
1850 #define IGBVF_TX_FLAGS_IPV4             0x00000008
1851 #define IGBVF_TX_FLAGS_VLAN_MASK        0xffff0000
1852 #define IGBVF_TX_FLAGS_VLAN_SHIFT       16
1853
1854 static int igbvf_tso(struct igbvf_adapter *adapter,
1855                      struct igbvf_ring *tx_ring,
1856                      struct sk_buff *skb, u32 tx_flags, u8 *hdr_len)
1857 {
1858         struct e1000_adv_tx_context_desc *context_desc;
1859         unsigned int i;
1860         int err;
1861         struct igbvf_buffer *buffer_info;
1862         u32 info = 0, tu_cmd = 0;
1863         u32 mss_l4len_idx, l4len;
1864         *hdr_len = 0;
1865
1866         if (skb_header_cloned(skb)) {
1867                 err = pskb_expand_head(skb, 0, 0, GFP_ATOMIC);
1868                 if (err) {
1869                         dev_err(&adapter->pdev->dev,
1870                                 "igbvf_tso returning an error\n");
1871                         return err;
1872                 }
1873         }
1874
1875         l4len = tcp_hdrlen(skb);
1876         *hdr_len += l4len;
1877
1878         if (skb->protocol == htons(ETH_P_IP)) {
1879                 struct iphdr *iph = ip_hdr(skb);
1880                 iph->tot_len = 0;
1881                 iph->check = 0;
1882                 tcp_hdr(skb)->check = ~csum_tcpudp_magic(iph->saddr,
1883                                                          iph->daddr, 0,
1884                                                          IPPROTO_TCP,
1885                                                          0);
1886         } else if (skb_is_gso_v6(skb)) {
1887                 ipv6_hdr(skb)->payload_len = 0;
1888                 tcp_hdr(skb)->check = ~csum_ipv6_magic(&ipv6_hdr(skb)->saddr,
1889                                                        &ipv6_hdr(skb)->daddr,
1890                                                        0, IPPROTO_TCP, 0);
1891         }
1892
1893         i = tx_ring->next_to_use;
1894
1895         buffer_info = &tx_ring->buffer_info[i];
1896         context_desc = IGBVF_TX_CTXTDESC_ADV(*tx_ring, i);
1897         /* VLAN MACLEN IPLEN */
1898         if (tx_flags & IGBVF_TX_FLAGS_VLAN)
1899                 info |= (tx_flags & IGBVF_TX_FLAGS_VLAN_MASK);
1900         info |= (skb_network_offset(skb) << E1000_ADVTXD_MACLEN_SHIFT);
1901         *hdr_len += skb_network_offset(skb);
1902         info |= (skb_transport_header(skb) - skb_network_header(skb));
1903         *hdr_len += (skb_transport_header(skb) - skb_network_header(skb));
1904         context_desc->vlan_macip_lens = cpu_to_le32(info);
1905
1906         /* ADV DTYP TUCMD MKRLOC/ISCSIHEDLEN */
1907         tu_cmd |= (E1000_TXD_CMD_DEXT | E1000_ADVTXD_DTYP_CTXT);
1908
1909         if (skb->protocol == htons(ETH_P_IP))
1910                 tu_cmd |= E1000_ADVTXD_TUCMD_IPV4;
1911         tu_cmd |= E1000_ADVTXD_TUCMD_L4T_TCP;
1912
1913         context_desc->type_tucmd_mlhl = cpu_to_le32(tu_cmd);
1914
1915         /* MSS L4LEN IDX */
1916         mss_l4len_idx = (skb_shinfo(skb)->gso_size << E1000_ADVTXD_MSS_SHIFT);
1917         mss_l4len_idx |= (l4len << E1000_ADVTXD_L4LEN_SHIFT);
1918
1919         context_desc->mss_l4len_idx = cpu_to_le32(mss_l4len_idx);
1920         context_desc->seqnum_seed = 0;
1921
1922         buffer_info->time_stamp = jiffies;
1923         buffer_info->next_to_watch = i;
1924         buffer_info->dma = 0;
1925         i++;
1926         if (i == tx_ring->count)
1927                 i = 0;
1928
1929         tx_ring->next_to_use = i;
1930
1931         return true;
1932 }
1933
1934 static inline bool igbvf_tx_csum(struct igbvf_adapter *adapter,
1935                                  struct igbvf_ring *tx_ring,
1936                                  struct sk_buff *skb, u32 tx_flags)
1937 {
1938         struct e1000_adv_tx_context_desc *context_desc;
1939         unsigned int i;
1940         struct igbvf_buffer *buffer_info;
1941         u32 info = 0, tu_cmd = 0;
1942
1943         if ((skb->ip_summed == CHECKSUM_PARTIAL) ||
1944             (tx_flags & IGBVF_TX_FLAGS_VLAN)) {
1945                 i = tx_ring->next_to_use;
1946                 buffer_info = &tx_ring->buffer_info[i];
1947                 context_desc = IGBVF_TX_CTXTDESC_ADV(*tx_ring, i);
1948
1949                 if (tx_flags & IGBVF_TX_FLAGS_VLAN)
1950                         info |= (tx_flags & IGBVF_TX_FLAGS_VLAN_MASK);
1951
1952                 info |= (skb_network_offset(skb) << E1000_ADVTXD_MACLEN_SHIFT);
1953                 if (skb->ip_summed == CHECKSUM_PARTIAL)
1954                         info |= (skb_transport_header(skb) -
1955                                  skb_network_header(skb));
1956
1957
1958                 context_desc->vlan_macip_lens = cpu_to_le32(info);
1959
1960                 tu_cmd |= (E1000_TXD_CMD_DEXT | E1000_ADVTXD_DTYP_CTXT);
1961
1962                 if (skb->ip_summed == CHECKSUM_PARTIAL) {
1963                         switch (skb->protocol) {
1964                         case __constant_htons(ETH_P_IP):
1965                                 tu_cmd |= E1000_ADVTXD_TUCMD_IPV4;
1966                                 if (ip_hdr(skb)->protocol == IPPROTO_TCP)
1967                                         tu_cmd |= E1000_ADVTXD_TUCMD_L4T_TCP;
1968                                 break;
1969                         case __constant_htons(ETH_P_IPV6):
1970                                 if (ipv6_hdr(skb)->nexthdr == IPPROTO_TCP)
1971                                         tu_cmd |= E1000_ADVTXD_TUCMD_L4T_TCP;
1972                                 break;
1973                         default:
1974                                 break;
1975                         }
1976                 }
1977
1978                 context_desc->type_tucmd_mlhl = cpu_to_le32(tu_cmd);
1979                 context_desc->seqnum_seed = 0;
1980                 context_desc->mss_l4len_idx = 0;
1981
1982                 buffer_info->time_stamp = jiffies;
1983                 buffer_info->next_to_watch = i;
1984                 buffer_info->dma = 0;
1985                 i++;
1986                 if (i == tx_ring->count)
1987                         i = 0;
1988                 tx_ring->next_to_use = i;
1989
1990                 return true;
1991         }
1992
1993         return false;
1994 }
1995
1996 static int igbvf_maybe_stop_tx(struct net_device *netdev, int size)
1997 {
1998         struct igbvf_adapter *adapter = netdev_priv(netdev);
1999
2000         /* there is enough descriptors then we don't need to worry  */
2001         if (igbvf_desc_unused(adapter->tx_ring) >= size)
2002                 return 0;
2003
2004         netif_stop_queue(netdev);
2005
2006         smp_mb();
2007
2008         /* We need to check again just in case room has been made available */
2009         if (igbvf_desc_unused(adapter->tx_ring) < size)
2010                 return -EBUSY;
2011
2012         netif_wake_queue(netdev);
2013
2014         ++adapter->restart_queue;
2015         return 0;
2016 }
2017
2018 #define IGBVF_MAX_TXD_PWR       16
2019 #define IGBVF_MAX_DATA_PER_TXD  (1 << IGBVF_MAX_TXD_PWR)
2020
2021 static inline int igbvf_tx_map_adv(struct igbvf_adapter *adapter,
2022                                    struct igbvf_ring *tx_ring,
2023                                    struct sk_buff *skb,
2024                                    unsigned int first)
2025 {
2026         struct igbvf_buffer *buffer_info;
2027         struct pci_dev *pdev = adapter->pdev;
2028         unsigned int len = skb_headlen(skb);
2029         unsigned int count = 0, i;
2030         unsigned int f;
2031
2032         i = tx_ring->next_to_use;
2033
2034         buffer_info = &tx_ring->buffer_info[i];
2035         BUG_ON(len >= IGBVF_MAX_DATA_PER_TXD);
2036         buffer_info->length = len;
2037         /* set time_stamp *before* dma to help avoid a possible race */
2038         buffer_info->time_stamp = jiffies;
2039         buffer_info->next_to_watch = i;
2040         buffer_info->mapped_as_page = false;
2041         buffer_info->dma = dma_map_single(&pdev->dev, skb->data, len,
2042                                           DMA_TO_DEVICE);
2043         if (dma_mapping_error(&pdev->dev, buffer_info->dma))
2044                 goto dma_error;
2045
2046
2047         for (f = 0; f < skb_shinfo(skb)->nr_frags; f++) {
2048                 struct skb_frag_struct *frag;
2049
2050                 count++;
2051                 i++;
2052                 if (i == tx_ring->count)
2053                         i = 0;
2054
2055                 frag = &skb_shinfo(skb)->frags[f];
2056                 len = frag->size;
2057
2058                 buffer_info = &tx_ring->buffer_info[i];
2059                 BUG_ON(len >= IGBVF_MAX_DATA_PER_TXD);
2060                 buffer_info->length = len;
2061                 buffer_info->time_stamp = jiffies;
2062                 buffer_info->next_to_watch = i;
2063                 buffer_info->mapped_as_page = true;
2064                 buffer_info->dma = skb_frag_dma_map(&pdev->dev, frag, 0, len,
2065                                                 DMA_TO_DEVICE);
2066                 if (dma_mapping_error(&pdev->dev, buffer_info->dma))
2067                         goto dma_error;
2068         }
2069
2070         tx_ring->buffer_info[i].skb = skb;
2071         tx_ring->buffer_info[first].next_to_watch = i;
2072
2073         return ++count;
2074
2075 dma_error:
2076         dev_err(&pdev->dev, "TX DMA map failed\n");
2077
2078         /* clear timestamp and dma mappings for failed buffer_info mapping */
2079         buffer_info->dma = 0;
2080         buffer_info->time_stamp = 0;
2081         buffer_info->length = 0;
2082         buffer_info->next_to_watch = 0;
2083         buffer_info->mapped_as_page = false;
2084         if (count)
2085                 count--;
2086
2087         /* clear timestamp and dma mappings for remaining portion of packet */
2088         while (count--) {
2089                 if (i==0)
2090                         i += tx_ring->count;
2091                 i--;
2092                 buffer_info = &tx_ring->buffer_info[i];
2093                 igbvf_put_txbuf(adapter, buffer_info);
2094         }
2095
2096         return 0;
2097 }
2098
2099 static inline void igbvf_tx_queue_adv(struct igbvf_adapter *adapter,
2100                                       struct igbvf_ring *tx_ring,
2101                                       int tx_flags, int count, u32 paylen,
2102                                       u8 hdr_len)
2103 {
2104         union e1000_adv_tx_desc *tx_desc = NULL;
2105         struct igbvf_buffer *buffer_info;
2106         u32 olinfo_status = 0, cmd_type_len;
2107         unsigned int i;
2108
2109         cmd_type_len = (E1000_ADVTXD_DTYP_DATA | E1000_ADVTXD_DCMD_IFCS |
2110                         E1000_ADVTXD_DCMD_DEXT);
2111
2112         if (tx_flags & IGBVF_TX_FLAGS_VLAN)
2113                 cmd_type_len |= E1000_ADVTXD_DCMD_VLE;
2114
2115         if (tx_flags & IGBVF_TX_FLAGS_TSO) {
2116                 cmd_type_len |= E1000_ADVTXD_DCMD_TSE;
2117
2118                 /* insert tcp checksum */
2119                 olinfo_status |= E1000_TXD_POPTS_TXSM << 8;
2120
2121                 /* insert ip checksum */
2122                 if (tx_flags & IGBVF_TX_FLAGS_IPV4)
2123                         olinfo_status |= E1000_TXD_POPTS_IXSM << 8;
2124
2125         } else if (tx_flags & IGBVF_TX_FLAGS_CSUM) {
2126                 olinfo_status |= E1000_TXD_POPTS_TXSM << 8;
2127         }
2128
2129         olinfo_status |= ((paylen - hdr_len) << E1000_ADVTXD_PAYLEN_SHIFT);
2130
2131         i = tx_ring->next_to_use;
2132         while (count--) {
2133                 buffer_info = &tx_ring->buffer_info[i];
2134                 tx_desc = IGBVF_TX_DESC_ADV(*tx_ring, i);
2135                 tx_desc->read.buffer_addr = cpu_to_le64(buffer_info->dma);
2136                 tx_desc->read.cmd_type_len =
2137                          cpu_to_le32(cmd_type_len | buffer_info->length);
2138                 tx_desc->read.olinfo_status = cpu_to_le32(olinfo_status);
2139                 i++;
2140                 if (i == tx_ring->count)
2141                         i = 0;
2142         }
2143
2144         tx_desc->read.cmd_type_len |= cpu_to_le32(adapter->txd_cmd);
2145         /* Force memory writes to complete before letting h/w
2146          * know there are new descriptors to fetch.  (Only
2147          * applicable for weak-ordered memory model archs,
2148          * such as IA-64). */
2149         wmb();
2150
2151         tx_ring->next_to_use = i;
2152         writel(i, adapter->hw.hw_addr + tx_ring->tail);
2153         /* we need this if more than one processor can write to our tail
2154          * at a time, it syncronizes IO on IA64/Altix systems */
2155         mmiowb();
2156 }
2157
2158 static netdev_tx_t igbvf_xmit_frame_ring_adv(struct sk_buff *skb,
2159                                              struct net_device *netdev,
2160                                              struct igbvf_ring *tx_ring)
2161 {
2162         struct igbvf_adapter *adapter = netdev_priv(netdev);
2163         unsigned int first, tx_flags = 0;
2164         u8 hdr_len = 0;
2165         int count = 0;
2166         int tso = 0;
2167
2168         if (test_bit(__IGBVF_DOWN, &adapter->state)) {
2169                 dev_kfree_skb_any(skb);
2170                 return NETDEV_TX_OK;
2171         }
2172
2173         if (skb->len <= 0) {
2174                 dev_kfree_skb_any(skb);
2175                 return NETDEV_TX_OK;
2176         }
2177
2178         /*
2179          * need: count + 4 desc gap to keep tail from touching
2180          *       + 2 desc gap to keep tail from touching head,
2181          *       + 1 desc for skb->data,
2182          *       + 1 desc for context descriptor,
2183          * head, otherwise try next time
2184          */
2185         if (igbvf_maybe_stop_tx(netdev, skb_shinfo(skb)->nr_frags + 4)) {
2186                 /* this is a hard error */
2187                 return NETDEV_TX_BUSY;
2188         }
2189
2190         if (vlan_tx_tag_present(skb)) {
2191                 tx_flags |= IGBVF_TX_FLAGS_VLAN;
2192                 tx_flags |= (vlan_tx_tag_get(skb) << IGBVF_TX_FLAGS_VLAN_SHIFT);
2193         }
2194
2195         if (skb->protocol == htons(ETH_P_IP))
2196                 tx_flags |= IGBVF_TX_FLAGS_IPV4;
2197
2198         first = tx_ring->next_to_use;
2199
2200         tso = skb_is_gso(skb) ?
2201                 igbvf_tso(adapter, tx_ring, skb, tx_flags, &hdr_len) : 0;
2202         if (unlikely(tso < 0)) {
2203                 dev_kfree_skb_any(skb);
2204                 return NETDEV_TX_OK;
2205         }
2206
2207         if (tso)
2208                 tx_flags |= IGBVF_TX_FLAGS_TSO;
2209         else if (igbvf_tx_csum(adapter, tx_ring, skb, tx_flags) &&
2210                  (skb->ip_summed == CHECKSUM_PARTIAL))
2211                 tx_flags |= IGBVF_TX_FLAGS_CSUM;
2212
2213         /*
2214          * count reflects descriptors mapped, if 0 then mapping error
2215          * has occurred and we need to rewind the descriptor queue
2216          */
2217         count = igbvf_tx_map_adv(adapter, tx_ring, skb, first);
2218
2219         if (count) {
2220                 igbvf_tx_queue_adv(adapter, tx_ring, tx_flags, count,
2221                                    skb->len, hdr_len);
2222                 /* Make sure there is space in the ring for the next send. */
2223                 igbvf_maybe_stop_tx(netdev, MAX_SKB_FRAGS + 4);
2224         } else {
2225                 dev_kfree_skb_any(skb);
2226                 tx_ring->buffer_info[first].time_stamp = 0;
2227                 tx_ring->next_to_use = first;
2228         }
2229
2230         return NETDEV_TX_OK;
2231 }
2232
2233 static netdev_tx_t igbvf_xmit_frame(struct sk_buff *skb,
2234                                     struct net_device *netdev)
2235 {
2236         struct igbvf_adapter *adapter = netdev_priv(netdev);
2237         struct igbvf_ring *tx_ring;
2238
2239         if (test_bit(__IGBVF_DOWN, &adapter->state)) {
2240                 dev_kfree_skb_any(skb);
2241                 return NETDEV_TX_OK;
2242         }
2243
2244         tx_ring = &adapter->tx_ring[0];
2245
2246         return igbvf_xmit_frame_ring_adv(skb, netdev, tx_ring);
2247 }
2248
2249 /**
2250  * igbvf_tx_timeout - Respond to a Tx Hang
2251  * @netdev: network interface device structure
2252  **/
2253 static void igbvf_tx_timeout(struct net_device *netdev)
2254 {
2255         struct igbvf_adapter *adapter = netdev_priv(netdev);
2256
2257         /* Do the reset outside of interrupt context */
2258         adapter->tx_timeout_count++;
2259         schedule_work(&adapter->reset_task);
2260 }
2261
2262 static void igbvf_reset_task(struct work_struct *work)
2263 {
2264         struct igbvf_adapter *adapter;
2265         adapter = container_of(work, struct igbvf_adapter, reset_task);
2266
2267         igbvf_reinit_locked(adapter);
2268 }
2269
2270 /**
2271  * igbvf_get_stats - Get System Network Statistics
2272  * @netdev: network interface device structure
2273  *
2274  * Returns the address of the device statistics structure.
2275  * The statistics are actually updated from the timer callback.
2276  **/
2277 static struct net_device_stats *igbvf_get_stats(struct net_device *netdev)
2278 {
2279         struct igbvf_adapter *adapter = netdev_priv(netdev);
2280
2281         /* only return the current stats */
2282         return &adapter->net_stats;
2283 }
2284
2285 /**
2286  * igbvf_change_mtu - Change the Maximum Transfer Unit
2287  * @netdev: network interface device structure
2288  * @new_mtu: new value for maximum frame size
2289  *
2290  * Returns 0 on success, negative on failure
2291  **/
2292 static int igbvf_change_mtu(struct net_device *netdev, int new_mtu)
2293 {
2294         struct igbvf_adapter *adapter = netdev_priv(netdev);
2295         int max_frame = new_mtu + ETH_HLEN + ETH_FCS_LEN;
2296
2297         if ((new_mtu < 68) || (max_frame > MAX_JUMBO_FRAME_SIZE)) {
2298                 dev_err(&adapter->pdev->dev, "Invalid MTU setting\n");
2299                 return -EINVAL;
2300         }
2301
2302 #define MAX_STD_JUMBO_FRAME_SIZE 9234
2303         if (max_frame > MAX_STD_JUMBO_FRAME_SIZE) {
2304                 dev_err(&adapter->pdev->dev, "MTU > 9216 not supported.\n");
2305                 return -EINVAL;
2306         }
2307
2308         while (test_and_set_bit(__IGBVF_RESETTING, &adapter->state))
2309                 msleep(1);
2310         /* igbvf_down has a dependency on max_frame_size */
2311         adapter->max_frame_size = max_frame;
2312         if (netif_running(netdev))
2313                 igbvf_down(adapter);
2314
2315         /*
2316          * NOTE: netdev_alloc_skb reserves 16 bytes, and typically NET_IP_ALIGN
2317          * means we reserve 2 more, this pushes us to allocate from the next
2318          * larger slab size.
2319          * i.e. RXBUFFER_2048 --> size-4096 slab
2320          * However with the new *_jumbo_rx* routines, jumbo receives will use
2321          * fragmented skbs
2322          */
2323
2324         if (max_frame <= 1024)
2325                 adapter->rx_buffer_len = 1024;
2326         else if (max_frame <= 2048)
2327                 adapter->rx_buffer_len = 2048;
2328         else
2329 #if (PAGE_SIZE / 2) > 16384
2330                 adapter->rx_buffer_len = 16384;
2331 #else
2332                 adapter->rx_buffer_len = PAGE_SIZE / 2;
2333 #endif
2334
2335
2336         /* adjust allocation if LPE protects us, and we aren't using SBP */
2337         if ((max_frame == ETH_FRAME_LEN + ETH_FCS_LEN) ||
2338              (max_frame == ETH_FRAME_LEN + VLAN_HLEN + ETH_FCS_LEN))
2339                 adapter->rx_buffer_len = ETH_FRAME_LEN + VLAN_HLEN +
2340                                          ETH_FCS_LEN;
2341
2342         dev_info(&adapter->pdev->dev, "changing MTU from %d to %d\n",
2343                  netdev->mtu, new_mtu);
2344         netdev->mtu = new_mtu;
2345
2346         if (netif_running(netdev))
2347                 igbvf_up(adapter);
2348         else
2349                 igbvf_reset(adapter);
2350
2351         clear_bit(__IGBVF_RESETTING, &adapter->state);
2352
2353         return 0;
2354 }
2355
2356 static int igbvf_ioctl(struct net_device *netdev, struct ifreq *ifr, int cmd)
2357 {
2358         switch (cmd) {
2359         default:
2360                 return -EOPNOTSUPP;
2361         }
2362 }
2363
2364 static int igbvf_suspend(struct pci_dev *pdev, pm_message_t state)
2365 {
2366         struct net_device *netdev = pci_get_drvdata(pdev);
2367         struct igbvf_adapter *adapter = netdev_priv(netdev);
2368 #ifdef CONFIG_PM
2369         int retval = 0;
2370 #endif
2371
2372         netif_device_detach(netdev);
2373
2374         if (netif_running(netdev)) {
2375                 WARN_ON(test_bit(__IGBVF_RESETTING, &adapter->state));
2376                 igbvf_down(adapter);
2377                 igbvf_free_irq(adapter);
2378         }
2379
2380 #ifdef CONFIG_PM
2381         retval = pci_save_state(pdev);
2382         if (retval)
2383                 return retval;
2384 #endif
2385
2386         pci_disable_device(pdev);
2387
2388         return 0;
2389 }
2390
2391 #ifdef CONFIG_PM
2392 static int igbvf_resume(struct pci_dev *pdev)
2393 {
2394         struct net_device *netdev = pci_get_drvdata(pdev);
2395         struct igbvf_adapter *adapter = netdev_priv(netdev);
2396         u32 err;
2397
2398         pci_restore_state(pdev);
2399         err = pci_enable_device_mem(pdev);
2400         if (err) {
2401                 dev_err(&pdev->dev, "Cannot enable PCI device from suspend\n");
2402                 return err;
2403         }
2404
2405         pci_set_master(pdev);
2406
2407         if (netif_running(netdev)) {
2408                 err = igbvf_request_irq(adapter);
2409                 if (err)
2410                         return err;
2411         }
2412
2413         igbvf_reset(adapter);
2414
2415         if (netif_running(netdev))
2416                 igbvf_up(adapter);
2417
2418         netif_device_attach(netdev);
2419
2420         return 0;
2421 }
2422 #endif
2423
2424 static void igbvf_shutdown(struct pci_dev *pdev)
2425 {
2426         igbvf_suspend(pdev, PMSG_SUSPEND);
2427 }
2428
2429 #ifdef CONFIG_NET_POLL_CONTROLLER
2430 /*
2431  * Polling 'interrupt' - used by things like netconsole to send skbs
2432  * without having to re-enable interrupts. It's not called while
2433  * the interrupt routine is executing.
2434  */
2435 static void igbvf_netpoll(struct net_device *netdev)
2436 {
2437         struct igbvf_adapter *adapter = netdev_priv(netdev);
2438
2439         disable_irq(adapter->pdev->irq);
2440
2441         igbvf_clean_tx_irq(adapter->tx_ring);
2442
2443         enable_irq(adapter->pdev->irq);
2444 }
2445 #endif
2446
2447 /**
2448  * igbvf_io_error_detected - called when PCI error is detected
2449  * @pdev: Pointer to PCI device
2450  * @state: The current pci connection state
2451  *
2452  * This function is called after a PCI bus error affecting
2453  * this device has been detected.
2454  */
2455 static pci_ers_result_t igbvf_io_error_detected(struct pci_dev *pdev,
2456                                                 pci_channel_state_t state)
2457 {
2458         struct net_device *netdev = pci_get_drvdata(pdev);
2459         struct igbvf_adapter *adapter = netdev_priv(netdev);
2460
2461         netif_device_detach(netdev);
2462
2463         if (state == pci_channel_io_perm_failure)
2464                 return PCI_ERS_RESULT_DISCONNECT;
2465
2466         if (netif_running(netdev))
2467                 igbvf_down(adapter);
2468         pci_disable_device(pdev);
2469
2470         /* Request a slot slot reset. */
2471         return PCI_ERS_RESULT_NEED_RESET;
2472 }
2473
2474 /**
2475  * igbvf_io_slot_reset - called after the pci bus has been reset.
2476  * @pdev: Pointer to PCI device
2477  *
2478  * Restart the card from scratch, as if from a cold-boot. Implementation
2479  * resembles the first-half of the igbvf_resume routine.
2480  */
2481 static pci_ers_result_t igbvf_io_slot_reset(struct pci_dev *pdev)
2482 {
2483         struct net_device *netdev = pci_get_drvdata(pdev);
2484         struct igbvf_adapter *adapter = netdev_priv(netdev);
2485
2486         if (pci_enable_device_mem(pdev)) {
2487                 dev_err(&pdev->dev,
2488                         "Cannot re-enable PCI device after reset.\n");
2489                 return PCI_ERS_RESULT_DISCONNECT;
2490         }
2491         pci_set_master(pdev);
2492
2493         igbvf_reset(adapter);
2494
2495         return PCI_ERS_RESULT_RECOVERED;
2496 }
2497
2498 /**
2499  * igbvf_io_resume - called when traffic can start flowing again.
2500  * @pdev: Pointer to PCI device
2501  *
2502  * This callback is called when the error recovery driver tells us that
2503  * its OK to resume normal operation. Implementation resembles the
2504  * second-half of the igbvf_resume routine.
2505  */
2506 static void igbvf_io_resume(struct pci_dev *pdev)
2507 {
2508         struct net_device *netdev = pci_get_drvdata(pdev);
2509         struct igbvf_adapter *adapter = netdev_priv(netdev);
2510
2511         if (netif_running(netdev)) {
2512                 if (igbvf_up(adapter)) {
2513                         dev_err(&pdev->dev,
2514                                 "can't bring device back up after reset\n");
2515                         return;
2516                 }
2517         }
2518
2519         netif_device_attach(netdev);
2520 }
2521
2522 static void igbvf_print_device_info(struct igbvf_adapter *adapter)
2523 {
2524         struct e1000_hw *hw = &adapter->hw;
2525         struct net_device *netdev = adapter->netdev;
2526         struct pci_dev *pdev = adapter->pdev;
2527
2528         dev_info(&pdev->dev, "Intel(R) 82576 Virtual Function\n");
2529         dev_info(&pdev->dev, "Address: %pM\n", netdev->dev_addr);
2530         dev_info(&pdev->dev, "MAC: %d\n", hw->mac.type);
2531 }
2532
2533 static const struct net_device_ops igbvf_netdev_ops = {
2534         .ndo_open                       = igbvf_open,
2535         .ndo_stop                       = igbvf_close,
2536         .ndo_start_xmit                 = igbvf_xmit_frame,
2537         .ndo_get_stats                  = igbvf_get_stats,
2538         .ndo_set_rx_mode                = igbvf_set_multi,
2539         .ndo_set_mac_address            = igbvf_set_mac,
2540         .ndo_change_mtu                 = igbvf_change_mtu,
2541         .ndo_do_ioctl                   = igbvf_ioctl,
2542         .ndo_tx_timeout                 = igbvf_tx_timeout,
2543         .ndo_vlan_rx_add_vid            = igbvf_vlan_rx_add_vid,
2544         .ndo_vlan_rx_kill_vid           = igbvf_vlan_rx_kill_vid,
2545 #ifdef CONFIG_NET_POLL_CONTROLLER
2546         .ndo_poll_controller            = igbvf_netpoll,
2547 #endif
2548 };
2549
2550 /**
2551  * igbvf_probe - Device Initialization Routine
2552  * @pdev: PCI device information struct
2553  * @ent: entry in igbvf_pci_tbl
2554  *
2555  * Returns 0 on success, negative on failure
2556  *
2557  * igbvf_probe initializes an adapter identified by a pci_dev structure.
2558  * The OS initialization, configuring of the adapter private structure,
2559  * and a hardware reset occur.
2560  **/
2561 static int __devinit igbvf_probe(struct pci_dev *pdev,
2562                                  const struct pci_device_id *ent)
2563 {
2564         struct net_device *netdev;
2565         struct igbvf_adapter *adapter;
2566         struct e1000_hw *hw;
2567         const struct igbvf_info *ei = igbvf_info_tbl[ent->driver_data];
2568
2569         static int cards_found;
2570         int err, pci_using_dac;
2571
2572         err = pci_enable_device_mem(pdev);
2573         if (err)
2574                 return err;
2575
2576         pci_using_dac = 0;
2577         err = dma_set_mask(&pdev->dev, DMA_BIT_MASK(64));
2578         if (!err) {
2579                 err = dma_set_coherent_mask(&pdev->dev, DMA_BIT_MASK(64));
2580                 if (!err)
2581                         pci_using_dac = 1;
2582         } else {
2583                 err = dma_set_mask(&pdev->dev, DMA_BIT_MASK(32));
2584                 if (err) {
2585                         err = dma_set_coherent_mask(&pdev->dev,
2586                                                     DMA_BIT_MASK(32));
2587                         if (err) {
2588                                 dev_err(&pdev->dev, "No usable DMA "
2589                                         "configuration, aborting\n");
2590                                 goto err_dma;
2591                         }
2592                 }
2593         }
2594
2595         err = pci_request_regions(pdev, igbvf_driver_name);
2596         if (err)
2597                 goto err_pci_reg;
2598
2599         pci_set_master(pdev);
2600
2601         err = -ENOMEM;
2602         netdev = alloc_etherdev(sizeof(struct igbvf_adapter));
2603         if (!netdev)
2604                 goto err_alloc_etherdev;
2605
2606         SET_NETDEV_DEV(netdev, &pdev->dev);
2607
2608         pci_set_drvdata(pdev, netdev);
2609         adapter = netdev_priv(netdev);
2610         hw = &adapter->hw;
2611         adapter->netdev = netdev;
2612         adapter->pdev = pdev;
2613         adapter->ei = ei;
2614         adapter->pba = ei->pba;
2615         adapter->flags = ei->flags;
2616         adapter->hw.back = adapter;
2617         adapter->hw.mac.type = ei->mac;
2618         adapter->msg_enable = (1 << NETIF_MSG_DRV | NETIF_MSG_PROBE) - 1;
2619
2620         /* PCI config space info */
2621
2622         hw->vendor_id = pdev->vendor;
2623         hw->device_id = pdev->device;
2624         hw->subsystem_vendor_id = pdev->subsystem_vendor;
2625         hw->subsystem_device_id = pdev->subsystem_device;
2626         hw->revision_id = pdev->revision;
2627
2628         err = -EIO;
2629         adapter->hw.hw_addr = ioremap(pci_resource_start(pdev, 0),
2630                                       pci_resource_len(pdev, 0));
2631
2632         if (!adapter->hw.hw_addr)
2633                 goto err_ioremap;
2634
2635         if (ei->get_variants) {
2636                 err = ei->get_variants(adapter);
2637                 if (err)
2638                         goto err_ioremap;
2639         }
2640
2641         /* setup adapter struct */
2642         err = igbvf_sw_init(adapter);
2643         if (err)
2644                 goto err_sw_init;
2645
2646         /* construct the net_device struct */
2647         netdev->netdev_ops = &igbvf_netdev_ops;
2648
2649         igbvf_set_ethtool_ops(netdev);
2650         netdev->watchdog_timeo = 5 * HZ;
2651         strncpy(netdev->name, pci_name(pdev), sizeof(netdev->name) - 1);
2652
2653         adapter->bd_number = cards_found++;
2654
2655         netdev->features = NETIF_F_SG |
2656                            NETIF_F_IP_CSUM |
2657                            NETIF_F_HW_VLAN_TX |
2658                            NETIF_F_HW_VLAN_RX |
2659                            NETIF_F_HW_VLAN_FILTER;
2660
2661         netdev->features |= NETIF_F_IPV6_CSUM;
2662         netdev->features |= NETIF_F_TSO;
2663         netdev->features |= NETIF_F_TSO6;
2664
2665         if (pci_using_dac)
2666                 netdev->features |= NETIF_F_HIGHDMA;
2667
2668         netdev->vlan_features |= NETIF_F_TSO;
2669         netdev->vlan_features |= NETIF_F_TSO6;
2670         netdev->vlan_features |= NETIF_F_IP_CSUM;
2671         netdev->vlan_features |= NETIF_F_IPV6_CSUM;
2672         netdev->vlan_features |= NETIF_F_SG;
2673
2674         /*reset the controller to put the device in a known good state */
2675         err = hw->mac.ops.reset_hw(hw);
2676         if (err) {
2677                 dev_info(&pdev->dev,
2678                          "PF still in reset state, assigning new address."
2679                          " Is the PF interface up?\n");
2680                 dev_hw_addr_random(adapter->netdev, hw->mac.addr);
2681         } else {
2682                 err = hw->mac.ops.read_mac_addr(hw);
2683                 if (err) {
2684                         dev_err(&pdev->dev, "Error reading MAC address\n");
2685                         goto err_hw_init;
2686                 }
2687         }
2688
2689         memcpy(netdev->dev_addr, adapter->hw.mac.addr, netdev->addr_len);
2690         memcpy(netdev->perm_addr, adapter->hw.mac.addr, netdev->addr_len);
2691
2692         if (!is_valid_ether_addr(netdev->perm_addr)) {
2693                 dev_err(&pdev->dev, "Invalid MAC Address: %pM\n",
2694                         netdev->dev_addr);
2695                 err = -EIO;
2696                 goto err_hw_init;
2697         }
2698
2699         setup_timer(&adapter->watchdog_timer, &igbvf_watchdog,
2700                     (unsigned long) adapter);
2701
2702         INIT_WORK(&adapter->reset_task, igbvf_reset_task);
2703         INIT_WORK(&adapter->watchdog_task, igbvf_watchdog_task);
2704
2705         /* ring size defaults */
2706         adapter->rx_ring->count = 1024;
2707         adapter->tx_ring->count = 1024;
2708
2709         /* reset the hardware with the new settings */
2710         igbvf_reset(adapter);
2711
2712         strcpy(netdev->name, "eth%d");
2713         err = register_netdev(netdev);
2714         if (err)
2715                 goto err_hw_init;
2716
2717         /* tell the stack to leave us alone until igbvf_open() is called */
2718         netif_carrier_off(netdev);
2719         netif_stop_queue(netdev);
2720
2721         igbvf_print_device_info(adapter);
2722
2723         igbvf_initialize_last_counter_stats(adapter);
2724
2725         return 0;
2726
2727 err_hw_init:
2728         kfree(adapter->tx_ring);
2729         kfree(adapter->rx_ring);
2730 err_sw_init:
2731         igbvf_reset_interrupt_capability(adapter);
2732         iounmap(adapter->hw.hw_addr);
2733 err_ioremap:
2734         free_netdev(netdev);
2735 err_alloc_etherdev:
2736         pci_release_regions(pdev);
2737 err_pci_reg:
2738 err_dma:
2739         pci_disable_device(pdev);
2740         return err;
2741 }
2742
2743 /**
2744  * igbvf_remove - Device Removal Routine
2745  * @pdev: PCI device information struct
2746  *
2747  * igbvf_remove is called by the PCI subsystem to alert the driver
2748  * that it should release a PCI device.  The could be caused by a
2749  * Hot-Plug event, or because the driver is going to be removed from
2750  * memory.
2751  **/
2752 static void __devexit igbvf_remove(struct pci_dev *pdev)
2753 {
2754         struct net_device *netdev = pci_get_drvdata(pdev);
2755         struct igbvf_adapter *adapter = netdev_priv(netdev);
2756         struct e1000_hw *hw = &adapter->hw;
2757
2758         /*
2759          * The watchdog timer may be rescheduled, so explicitly
2760          * disable it from being rescheduled.
2761          */
2762         set_bit(__IGBVF_DOWN, &adapter->state);
2763         del_timer_sync(&adapter->watchdog_timer);
2764
2765         cancel_work_sync(&adapter->reset_task);
2766         cancel_work_sync(&adapter->watchdog_task);
2767
2768         unregister_netdev(netdev);
2769
2770         igbvf_reset_interrupt_capability(adapter);
2771
2772         /*
2773          * it is important to delete the napi struct prior to freeing the
2774          * rx ring so that you do not end up with null pointer refs
2775          */
2776         netif_napi_del(&adapter->rx_ring->napi);
2777         kfree(adapter->tx_ring);
2778         kfree(adapter->rx_ring);
2779
2780         iounmap(hw->hw_addr);
2781         if (hw->flash_address)
2782                 iounmap(hw->flash_address);
2783         pci_release_regions(pdev);
2784
2785         free_netdev(netdev);
2786
2787         pci_disable_device(pdev);
2788 }
2789
2790 /* PCI Error Recovery (ERS) */
2791 static struct pci_error_handlers igbvf_err_handler = {
2792         .error_detected = igbvf_io_error_detected,
2793         .slot_reset = igbvf_io_slot_reset,
2794         .resume = igbvf_io_resume,
2795 };
2796
2797 static DEFINE_PCI_DEVICE_TABLE(igbvf_pci_tbl) = {
2798         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82576_VF), board_vf },
2799         { PCI_VDEVICE(INTEL, E1000_DEV_ID_I350_VF), board_i350_vf },
2800         { } /* terminate list */
2801 };
2802 MODULE_DEVICE_TABLE(pci, igbvf_pci_tbl);
2803
2804 /* PCI Device API Driver */
2805 static struct pci_driver igbvf_driver = {
2806         .name     = igbvf_driver_name,
2807         .id_table = igbvf_pci_tbl,
2808         .probe    = igbvf_probe,
2809         .remove   = __devexit_p(igbvf_remove),
2810 #ifdef CONFIG_PM
2811         /* Power Management Hooks */
2812         .suspend  = igbvf_suspend,
2813         .resume   = igbvf_resume,
2814 #endif
2815         .shutdown = igbvf_shutdown,
2816         .err_handler = &igbvf_err_handler
2817 };
2818
2819 /**
2820  * igbvf_init_module - Driver Registration Routine
2821  *
2822  * igbvf_init_module is the first routine called when the driver is
2823  * loaded. All it does is register with the PCI subsystem.
2824  **/
2825 static int __init igbvf_init_module(void)
2826 {
2827         int ret;
2828         printk(KERN_INFO "%s - version %s\n",
2829                igbvf_driver_string, igbvf_driver_version);
2830         printk(KERN_INFO "%s\n", igbvf_copyright);
2831
2832         ret = pci_register_driver(&igbvf_driver);
2833
2834         return ret;
2835 }
2836 module_init(igbvf_init_module);
2837
2838 /**
2839  * igbvf_exit_module - Driver Exit Cleanup Routine
2840  *
2841  * igbvf_exit_module is called just before the driver is removed
2842  * from memory.
2843  **/
2844 static void __exit igbvf_exit_module(void)
2845 {
2846         pci_unregister_driver(&igbvf_driver);
2847 }
2848 module_exit(igbvf_exit_module);
2849
2850
2851 MODULE_AUTHOR("Intel Corporation, <e1000-devel@lists.sourceforge.net>");
2852 MODULE_DESCRIPTION("Intel(R) 82576 Virtual Function Network Driver");
2853 MODULE_LICENSE("GPL");
2854 MODULE_VERSION(DRV_VERSION);
2855
2856 /* netdev.c */