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 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
  29
  30 #include <linux/module.h>
  31 #include <linux/types.h>
  32 #include <linux/init.h>
  33 #include <linux/pci.h>
  34 #include <linux/vmalloc.h>
  35 #include <linux/pagemap.h>
  36 #include <linux/delay.h>
  37 #include <linux/netdevice.h>
  38 #include <linux/tcp.h>
  39 #include <linux/ipv6.h>
  40 #include <linux/slab.h>
  41 #include <net/checksum.h>
  42 #include <net/ip6_checksum.h>
  43 #include <linux/mii.h>
  44 #include <linux/ethtool.h>
  45 #include <linux/if_vlan.h>
  46 #include <linux/prefetch.h>
  47
  48 #include "igbvf.h"
  49
  50 #define DRV_VERSION "2.0.1-k"
  51 char igbvf_driver_name[] = "igbvf";
  52 const char igbvf_driver_version[] = DRV_VERSION;
  53 static const char igbvf_driver_string[] =
  54                   "Intel(R) Gigabit Virtual Function Network Driver";
  55 static const char igbvf_copyright[] =
  56                   "Copyright (c) 2009 - 2011 Intel Corporation.";
  57
  58 static int igbvf_poll(struct napi_struct *napi, int budget);
  59 static void igbvf_reset(struct igbvf_adapter *);
  60 static void igbvf_set_interrupt_capability(struct igbvf_adapter *);
  61 static void igbvf_reset_interrupt_capability(struct igbvf_adapter *);
  62
  63 static struct igbvf_info igbvf_vf_info = {
  64         .mac                    = e1000_vfadapt,
  65         .flags                  = 0,
  66         .pba                    = 10,
  67         .init_ops               = e1000_init_function_pointers_vf,
  68 };
  69
  70 static struct igbvf_info igbvf_i350_vf_info = {
  71         .mac                    = e1000_vfadapt_i350,
  72         .flags                  = 0,
  73         .pba                    = 10,
  74         .init_ops               = e1000_init_function_pointers_vf,
  75 };
  76
  77 static const struct igbvf_info *igbvf_info_tbl[] = {
  78         [board_vf]              = &igbvf_vf_info,
  79         [board_i350_vf]         = &igbvf_i350_vf_info,
  80 };
  81
  82 /**
  83  * igbvf_desc_unused - calculate if we have unused descriptors
  84  **/
  85 static int igbvf_desc_unused(struct igbvf_ring *ring)
  86 {
  87         if (ring->next_to_clean > ring->next_to_use)
  88                 return ring->next_to_clean - ring->next_to_use - 1;
  89
  90         return ring->count + ring->next_to_clean - ring->next_to_use - 1;
  91 }
  92
  93 /**
  94  * igbvf_receive_skb - helper function to handle Rx indications
  95  * @adapter: board private structure
  96  * @status: descriptor status field as written by hardware
  97  * @vlan: descriptor vlan field as written by hardware (no le/be conversion)
  98  * @skb: pointer to sk_buff to be indicated to stack
  99  **/
 100 static void igbvf_receive_skb(struct igbvf_adapter *adapter,
 101                               struct net_device *netdev,
 102                               struct sk_buff *skb,
 103                               u32 status, u16 vlan)
 104 {
 105         if (status & E1000_RXD_STAT_VP) {
 106                 u16 vid = le16_to_cpu(vlan) & E1000_RXD_SPC_VLAN_MASK;
 107                 if (test_bit(vid, adapter->active_vlans))
 108                         __vlan_hwaccel_put_tag(skb, vid);
 109         }
 110         netif_receive_skb(skb);
 111 }
 112
 113 static inline void igbvf_rx_checksum_adv(struct igbvf_adapter *adapter,
 114                                          u32 status_err, struct sk_buff *skb)
 115 {
 116         skb_checksum_none_assert(skb);
 117
 118         /* Ignore Checksum bit is set or checksum is disabled through ethtool */
 119         if ((status_err & E1000_RXD_STAT_IXSM) ||
 120             (adapter->flags & IGBVF_FLAG_RX_CSUM_DISABLED))
 121                 return;
 122
 123         /* TCP/UDP checksum error bit is set */
 124         if (status_err &
 125             (E1000_RXDEXT_STATERR_TCPE | E1000_RXDEXT_STATERR_IPE)) {
 126                 /* let the stack verify checksum errors */
 127                 adapter->hw_csum_err++;
 128                 return;
 129         }
 130
 131         /* It must be a TCP or UDP packet with a valid checksum */
 132         if (status_err & (E1000_RXD_STAT_TCPCS | E1000_RXD_STAT_UDPCS))
 133                 skb->ip_summed = CHECKSUM_UNNECESSARY;
 134
 135         adapter->hw_csum_good++;
 136 }
 137
 138 /**
 139  * igbvf_alloc_rx_buffers - Replace used receive buffers; packet split
 140  * @rx_ring: address of ring structure to repopulate
 141  * @cleaned_count: number of buffers to repopulate
 142  **/
 143 static void igbvf_alloc_rx_buffers(struct igbvf_ring *rx_ring,
 144                                    int cleaned_count)
 145 {
 146         struct igbvf_adapter *adapter = rx_ring->adapter;
 147         struct net_device *netdev = adapter->netdev;
 148         struct pci_dev *pdev = adapter->pdev;
 149         union e1000_adv_rx_desc *rx_desc;
 150         struct igbvf_buffer *buffer_info;
 151         struct sk_buff *skb;
 152         unsigned int i;
 153         int bufsz;
 154
 155         i = rx_ring->next_to_use;
 156         buffer_info = &rx_ring->buffer_info[i];
 157
 158         if (adapter->rx_ps_hdr_size)
 159                 bufsz = adapter->rx_ps_hdr_size;
 160         else
 161                 bufsz = adapter->rx_buffer_len;
 162
 163         while (cleaned_count--) {
 164                 rx_desc = IGBVF_RX_DESC_ADV(*rx_ring, i);
 165
 166                 if (adapter->rx_ps_hdr_size && !buffer_info->page_dma) {
 167                         if (!buffer_info->page) {
 168                                 buffer_info->page = alloc_page(GFP_ATOMIC);
 169                                 if (!buffer_info->page) {
 170                                         adapter->alloc_rx_buff_failed++;
 171                                         goto no_buffers;
 172                                 }
 173                                 buffer_info->page_offset = 0;
 174                         } else {
 175                                 buffer_info->page_offset ^= PAGE_SIZE / 2;
 176                         }
 177                         buffer_info->page_dma =
 178                                 dma_map_page(&pdev->dev, buffer_info->page,
 179                                              buffer_info->page_offset,
 180                                              PAGE_SIZE / 2,
 181                                              DMA_FROM_DEVICE);
 182                 }
 183
 184                 if (!buffer_info->skb) {
 185                         skb = netdev_alloc_skb_ip_align(netdev, bufsz);
 186                         if (!skb) {
 187                                 adapter->alloc_rx_buff_failed++;
 188                                 goto no_buffers;
 189                         }
 190
 191                         buffer_info->skb = skb;
 192                         buffer_info->dma = dma_map_single(&pdev->dev, skb->data,
 193                                                           bufsz,
 194                                                           DMA_FROM_DEVICE);
 195                 }
 196                 /* Refresh the desc even if buffer_addrs didn't change because
 197                  * each write-back erases this info. */
 198                 if (adapter->rx_ps_hdr_size) {
 199                         rx_desc->read.pkt_addr =
 200                              cpu_to_le64(buffer_info->page_dma);
 201                         rx_desc->read.hdr_addr = cpu_to_le64(buffer_info->dma);
 202                 } else {
 203                         rx_desc->read.pkt_addr =
 204                              cpu_to_le64(buffer_info->dma);
 205                         rx_desc->read.hdr_addr = 0;
 206                 }
 207
 208                 i++;
 209                 if (i == rx_ring->count)
 210                         i = 0;
 211                 buffer_info = &rx_ring->buffer_info[i];
 212         }
 213
 214 no_buffers:
 215         if (rx_ring->next_to_use != i) {
 216                 rx_ring->next_to_use = i;
 217                 if (i == 0)
 218                         i = (rx_ring->count - 1);
 219                 else
 220                         i--;
 221
 222                 /* Force memory writes to complete before letting h/w
 223                  * know there are new descriptors to fetch.  (Only
 224                  * applicable for weak-ordered memory model archs,
 225                  * such as IA-64). */
 226                 wmb();
 227                 writel(i, adapter->hw.hw_addr + rx_ring->tail);
 228         }
 229 }
 230
 231 /**
 232  * igbvf_clean_rx_irq - Send received data up the network stack; legacy
 233  * @adapter: board private structure
 234  *
 235  * the return value indicates whether actual cleaning was done, there
 236  * is no guarantee that everything was cleaned
 237  **/
 238 static bool igbvf_clean_rx_irq(struct igbvf_adapter *adapter,
 239                                int *work_done, int work_to_do)
 240 {
 241         struct igbvf_ring *rx_ring = adapter->rx_ring;
 242         struct net_device *netdev = adapter->netdev;
 243         struct pci_dev *pdev = adapter->pdev;
 244         union e1000_adv_rx_desc *rx_desc, *next_rxd;
 245         struct igbvf_buffer *buffer_info, *next_buffer;
 246         struct sk_buff *skb;
 247         bool cleaned = false;
 248         int cleaned_count = 0;
 249         unsigned int total_bytes = 0, total_packets = 0;
 250         unsigned int i;
 251         u32 length, hlen, staterr;
 252
 253         i = rx_ring->next_to_clean;
 254         rx_desc = IGBVF_RX_DESC_ADV(*rx_ring, i);
 255         staterr = le32_to_cpu(rx_desc->wb.upper.status_error);
 256
 257         while (staterr & E1000_RXD_STAT_DD) {
 258                 if (*work_done >= work_to_do)
 259                         break;
 260                 (*work_done)++;
 261                 rmb(); /* read descriptor and rx_buffer_info after status DD */
 262
 263                 buffer_info = &rx_ring->buffer_info[i];
 264
 265                 /* HW will not DMA in data larger than the given buffer, even
 266                  * if it parses the (NFS, of course) header to be larger.  In
 267                  * that case, it fills the header buffer and spills the rest
 268                  * into the page.
 269                  */
 270                 hlen = (le16_to_cpu(rx_desc->wb.lower.lo_dword.hs_rss.hdr_info) &
 271                   E1000_RXDADV_HDRBUFLEN_MASK) >> E1000_RXDADV_HDRBUFLEN_SHIFT;
 272                 if (hlen > adapter->rx_ps_hdr_size)
 273                         hlen = adapter->rx_ps_hdr_size;
 274
 275                 length = le16_to_cpu(rx_desc->wb.upper.length);
 276                 cleaned = true;
 277                 cleaned_count++;
 278
 279                 skb = buffer_info->skb;
 280                 prefetch(skb->data - NET_IP_ALIGN);
 281                 buffer_info->skb = NULL;
 282                 if (!adapter->rx_ps_hdr_size) {
 283                         dma_unmap_single(&pdev->dev, buffer_info->dma,
 284                                          adapter->rx_buffer_len,
 285                                          DMA_FROM_DEVICE);
 286                         buffer_info->dma = 0;
 287                         skb_put(skb, length);
 288                         goto send_up;
 289                 }
 290
 291                 if (!skb_shinfo(skb)->nr_frags) {
 292                         dma_unmap_single(&pdev->dev, buffer_info->dma,
 293                                          adapter->rx_ps_hdr_size,
 294                                          DMA_FROM_DEVICE);
 295                         skb_put(skb, hlen);
 296                 }
 297
 298                 if (length) {
 299                         dma_unmap_page(&pdev->dev, buffer_info->page_dma,
 300                                        PAGE_SIZE / 2,
 301                                        DMA_FROM_DEVICE);
 302                         buffer_info->page_dma = 0;
 303
 304                         skb_fill_page_desc(skb, skb_shinfo(skb)->nr_frags,
 305                                            buffer_info->page,
 306                                            buffer_info->page_offset,
 307                                            length);
 308
 309                         if ((adapter->rx_buffer_len > (PAGE_SIZE / 2)) ||
 310                             (page_count(buffer_info->page) != 1))
 311                                 buffer_info->page = NULL;
 312                         else
 313                                 get_page(buffer_info->page);
 314
 315                         skb->len += length;
 316                         skb->data_len += length;
 317                         skb->truesize += PAGE_SIZE / 2;
 318                 }
 319 send_up:
 320                 i++;
 321                 if (i == rx_ring->count)
 322                         i = 0;
 323                 next_rxd = IGBVF_RX_DESC_ADV(*rx_ring, i);
 324                 prefetch(next_rxd);
 325                 next_buffer = &rx_ring->buffer_info[i];
 326
 327                 if (!(staterr & E1000_RXD_STAT_EOP)) {
 328                         buffer_info->skb = next_buffer->skb;
 329                         buffer_info->dma = next_buffer->dma;
 330                         next_buffer->skb = skb;
 331                         next_buffer->dma = 0;
 332                         goto next_desc;
 333                 }
 334
 335                 if (staterr & E1000_RXDEXT_ERR_FRAME_ERR_MASK) {
 336                         dev_kfree_skb_irq(skb);
 337                         goto next_desc;
 338                 }
 339
 340                 total_bytes += skb->len;
 341                 total_packets++;
 342
 343                 igbvf_rx_checksum_adv(adapter, staterr, skb);
 344
 345                 skb->protocol = eth_type_trans(skb, netdev);
 346
 347                 igbvf_receive_skb(adapter, netdev, skb, staterr,
 348                                   rx_desc->wb.upper.vlan);
 349
 350 next_desc:
 351                 rx_desc->wb.upper.status_error = 0;
 352
 353                 /* return some buffers to hardware, one at a time is too slow */
 354                 if (cleaned_count >= IGBVF_RX_BUFFER_WRITE) {
 355                         igbvf_alloc_rx_buffers(rx_ring, cleaned_count);
 356                         cleaned_count = 0;
 357                 }
 358
 359                 /* use prefetched values */
 360                 rx_desc = next_rxd;
 361                 buffer_info = next_buffer;
 362
 363                 staterr = le32_to_cpu(rx_desc->wb.upper.status_error);
 364         }
 365
 366         rx_ring->next_to_clean = i;
 367         cleaned_count = igbvf_desc_unused(rx_ring);
 368
 369         if (cleaned_count)
 370                 igbvf_alloc_rx_buffers(rx_ring, cleaned_count);
 371
 372         adapter->total_rx_packets += total_packets;
 373         adapter->total_rx_bytes += total_bytes;
 374         adapter->net_stats.rx_bytes += total_bytes;
 375         adapter->net_stats.rx_packets += total_packets;
 376         return cleaned;
 377 }
 378
 379 static void igbvf_put_txbuf(struct igbvf_adapter *adapter,
 380                             struct igbvf_buffer *buffer_info)
 381 {
 382         if (buffer_info->dma) {
 383                 if (buffer_info->mapped_as_page)
 384                         dma_unmap_page(&adapter->pdev->dev,
 385                                        buffer_info->dma,
 386                                        buffer_info->length,
 387                                        DMA_TO_DEVICE);
 388                 else
 389                         dma_unmap_single(&adapter->pdev->dev,
 390                                          buffer_info->dma,
 391                                          buffer_info->length,
 392                                          DMA_TO_DEVICE);
 393                 buffer_info->dma = 0;
 394         }
 395         if (buffer_info->skb) {
 396                 dev_kfree_skb_any(buffer_info->skb);
 397                 buffer_info->skb = NULL;
 398         }
 399         buffer_info->time_stamp = 0;
 400 }
 401
 402 /**
 403  * igbvf_setup_tx_resources - allocate Tx resources (Descriptors)
 404  * @adapter: board private structure
 405  *
 406  * Return 0 on success, negative on failure
 407  **/
 408 int igbvf_setup_tx_resources(struct igbvf_adapter *adapter,
 409                              struct igbvf_ring *tx_ring)
 410 {
 411         struct pci_dev *pdev = adapter->pdev;
 412         int size;
 413
 414         size = sizeof(struct igbvf_buffer) * tx_ring->count;
 415         tx_ring->buffer_info = vzalloc(size);
 416         if (!tx_ring->buffer_info)
 417                 goto err;
 418
 419         /* round up to nearest 4K */
 420         tx_ring->size = tx_ring->count * sizeof(union e1000_adv_tx_desc);
 421         tx_ring->size = ALIGN(tx_ring->size, 4096);
 422
 423         tx_ring->desc = dma_alloc_coherent(&pdev->dev, tx_ring->size,
 424                                            &tx_ring->dma, GFP_KERNEL);
 425
 426         if (!tx_ring->desc)
 427                 goto err;
 428
 429         tx_ring->adapter = adapter;
 430         tx_ring->next_to_use = 0;
 431         tx_ring->next_to_clean = 0;
 432
 433         return 0;
 434 err:
 435         vfree(tx_ring->buffer_info);
 436         dev_err(&adapter->pdev->dev,
 437                 "Unable to allocate memory for the transmit descriptor ring\n");
 438         return -ENOMEM;
 439 }
 440
 441 /**
 442  * igbvf_setup_rx_resources - allocate Rx resources (Descriptors)
 443  * @adapter: board private structure
 444  *
 445  * Returns 0 on success, negative on failure
 446  **/
 447 int igbvf_setup_rx_resources(struct igbvf_adapter *adapter,
 448                              struct igbvf_ring *rx_ring)
 449 {
 450         struct pci_dev *pdev = adapter->pdev;
 451         int size, desc_len;
 452
 453         size = sizeof(struct igbvf_buffer) * rx_ring->count;
 454         rx_ring->buffer_info = vzalloc(size);
 455         if (!rx_ring->buffer_info)
 456                 goto err;
 457
 458         desc_len = sizeof(union e1000_adv_rx_desc);
 459
 460         /* Round up to nearest 4K */
 461         rx_ring->size = rx_ring->count * desc_len;
 462         rx_ring->size = ALIGN(rx_ring->size, 4096);
 463
 464         rx_ring->desc = dma_alloc_coherent(&pdev->dev, rx_ring->size,
 465                                            &rx_ring->dma, GFP_KERNEL);
 466
 467         if (!rx_ring->desc)
 468                 goto err;
 469
 470         rx_ring->next_to_clean = 0;
 471         rx_ring->next_to_use = 0;
 472
 473         rx_ring->adapter = adapter;
 474
 475         return 0;
 476
 477 err:
 478         vfree(rx_ring->buffer_info);
 479         rx_ring->buffer_info = NULL;
 480         dev_err(&adapter->pdev->dev,
 481                 "Unable to allocate memory for the receive descriptor ring\n");
 482         return -ENOMEM;
 483 }
 484
 485 /**
 486  * igbvf_clean_tx_ring - Free Tx Buffers
 487  * @tx_ring: ring to be cleaned
 488  **/
 489 static void igbvf_clean_tx_ring(struct igbvf_ring *tx_ring)
 490 {
 491         struct igbvf_adapter *adapter = tx_ring->adapter;
 492         struct igbvf_buffer *buffer_info;
 493         unsigned long size;
 494         unsigned int i;
 495
 496         if (!tx_ring->buffer_info)
 497                 return;
 498
 499         /* Free all the Tx ring sk_buffs */
 500         for (i = 0; i < tx_ring->count; i++) {
 501                 buffer_info = &tx_ring->buffer_info[i];
 502                 igbvf_put_txbuf(adapter, buffer_info);
 503         }
 504
 505         size = sizeof(struct igbvf_buffer) * tx_ring->count;
 506         memset(tx_ring->buffer_info, 0, size);
 507
 508         /* Zero out the descriptor ring */
 509         memset(tx_ring->desc, 0, tx_ring->size);
 510
 511         tx_ring->next_to_use = 0;
 512         tx_ring->next_to_clean = 0;
 513
 514         writel(0, adapter->hw.hw_addr + tx_ring->head);
 515         writel(0, adapter->hw.hw_addr + tx_ring->tail);
 516 }
 517
 518 /**
 519  * igbvf_free_tx_resources - Free Tx Resources per Queue
 520  * @tx_ring: ring to free resources from
 521  *
 522  * Free all transmit software resources
 523  **/
 524 void igbvf_free_tx_resources(struct igbvf_ring *tx_ring)
 525 {
 526         struct pci_dev *pdev = tx_ring->adapter->pdev;
 527
 528         igbvf_clean_tx_ring(tx_ring);
 529
 530         vfree(tx_ring->buffer_info);
 531         tx_ring->buffer_info = NULL;
 532
 533         dma_free_coherent(&pdev->dev, tx_ring->size, tx_ring->desc,
 534                           tx_ring->dma);
 535
 536         tx_ring->desc = NULL;
 537 }
 538
 539 /**
 540  * igbvf_clean_rx_ring - Free Rx Buffers per Queue
 541  * @adapter: board private structure
 542  **/
 543 static void igbvf_clean_rx_ring(struct igbvf_ring *rx_ring)
 544 {
 545         struct igbvf_adapter *adapter = rx_ring->adapter;
 546         struct igbvf_buffer *buffer_info;
 547         struct pci_dev *pdev = adapter->pdev;
 548         unsigned long size;
 549         unsigned int i;
 550
 551         if (!rx_ring->buffer_info)
 552                 return;
 553
 554         /* Free all the Rx ring sk_buffs */
 555         for (i = 0; i < rx_ring->count; i++) {
 556                 buffer_info = &rx_ring->buffer_info[i];
 557                 if (buffer_info->dma) {
 558                         if (adapter->rx_ps_hdr_size){
 559                                 dma_unmap_single(&pdev->dev, buffer_info->dma,
 560                                                  adapter->rx_ps_hdr_size,
 561                                                  DMA_FROM_DEVICE);
 562                         } else {
 563                                 dma_unmap_single(&pdev->dev, buffer_info->dma,
 564                                                  adapter->rx_buffer_len,
 565                                                  DMA_FROM_DEVICE);
 566                         }
 567                         buffer_info->dma = 0;
 568                 }
 569
 570                 if (buffer_info->skb) {
 571                         dev_kfree_skb(buffer_info->skb);
 572                         buffer_info->skb = NULL;
 573                 }
 574
 575                 if (buffer_info->page) {
 576                         if (buffer_info->page_dma)
 577                                 dma_unmap_page(&pdev->dev,
 578                                                buffer_info->page_dma,
 579                                                PAGE_SIZE / 2,
 580                                                DMA_FROM_DEVICE);
 581                         put_page(buffer_info->page);
 582                         buffer_info->page = NULL;
 583                         buffer_info->page_dma = 0;
 584                         buffer_info->page_offset = 0;
 585                 }
 586         }
 587
 588         size = sizeof(struct igbvf_buffer) * rx_ring->count;
 589         memset(rx_ring->buffer_info, 0, size);
 590
 591         /* Zero out the descriptor ring */
 592         memset(rx_ring->desc, 0, rx_ring->size);
 593
 594         rx_ring->next_to_clean = 0;
 595         rx_ring->next_to_use = 0;
 596
 597         writel(0, adapter->hw.hw_addr + rx_ring->head);
 598         writel(0, adapter->hw.hw_addr + rx_ring->tail);
 599 }
 600
 601 /**
 602  * igbvf_free_rx_resources - Free Rx Resources
 603  * @rx_ring: ring to clean the resources from
 604  *
 605  * Free all receive software resources
 606  **/
 607
 608 void igbvf_free_rx_resources(struct igbvf_ring *rx_ring)
 609 {
 610         struct pci_dev *pdev = rx_ring->adapter->pdev;
 611
 612         igbvf_clean_rx_ring(rx_ring);
 613
 614         vfree(rx_ring->buffer_info);
 615         rx_ring->buffer_info = NULL;
 616
 617         dma_free_coherent(&pdev->dev, rx_ring->size, rx_ring->desc,
 618                           rx_ring->dma);
 619         rx_ring->desc = NULL;
 620 }
 621
 622 /**
 623  * igbvf_update_itr - update the dynamic ITR value based on statistics
 624  * @adapter: pointer to adapter
 625  * @itr_setting: current adapter->itr
 626  * @packets: the number of packets during this measurement interval
 627  * @bytes: the number of bytes during this measurement interval
 628  *
 629  *      Stores a new ITR value based on packets and byte
 630  *      counts during the last interrupt.  The advantage of per interrupt
 631  *      computation is faster updates and more accurate ITR for the current
 632  *      traffic pattern.  Constants in this function were computed
 633  *      based on theoretical maximum wire speed and thresholds were set based
 634  *      on testing data as well as attempting to minimize response time
 635  *      while increasing bulk throughput.  This functionality is controlled
 636  *      by the InterruptThrottleRate module parameter.
 637  **/
 638 static unsigned int igbvf_update_itr(struct igbvf_adapter *adapter,
 639                                      u16 itr_setting, int packets,
 640                                      int bytes)
 641 {
 642         unsigned int retval = itr_setting;
 643
 644         if (packets == 0)
 645                 goto update_itr_done;
 646
 647         switch (itr_setting) {
 648         case lowest_latency:
 649                 /* handle TSO and jumbo frames */
 650                 if (bytes/packets > 8000)
 651                         retval = bulk_latency;
 652                 else if ((packets < 5) && (bytes > 512))
 653                         retval = low_latency;
 654                 break;
 655         case low_latency:  /* 50 usec aka 20000 ints/s */
 656                 if (bytes > 10000) {
 657                         /* this if handles the TSO accounting */
 658                         if (bytes/packets > 8000)
 659                                 retval = bulk_latency;
 660                         else if ((packets < 10) || ((bytes/packets) > 1200))
 661                                 retval = bulk_latency;
 662                         else if ((packets > 35))
 663                                 retval = lowest_latency;
 664                 } else if (bytes/packets > 2000) {
 665                         retval = bulk_latency;
 666                 } else if (packets <= 2 && bytes < 512) {
 667                         retval = lowest_latency;
 668                 }
 669                 break;
 670         case bulk_latency: /* 250 usec aka 4000 ints/s */
 671                 if (bytes > 25000) {
 672                         if (packets > 35)
 673                                 retval = low_latency;
 674                 } else if (bytes < 6000) {
 675                         retval = low_latency;
 676                 }
 677                 break;
 678         }
 679
 680 update_itr_done:
 681         return retval;
 682 }
 683
 684 static void igbvf_set_itr(struct igbvf_adapter *adapter)
 685 {
 686         struct e1000_hw *hw = &adapter->hw;
 687         u16 current_itr;
 688         u32 new_itr = adapter->itr;
 689
 690         adapter->tx_itr = igbvf_update_itr(adapter, adapter->tx_itr,
 691                                            adapter->total_tx_packets,
 692                                            adapter->total_tx_bytes);
 693         /* conservative mode (itr 3) eliminates the lowest_latency setting */
 694         if (adapter->itr_setting == 3 && adapter->tx_itr == lowest_latency)
 695                 adapter->tx_itr = low_latency;
 696
 697         adapter->rx_itr = igbvf_update_itr(adapter, adapter->rx_itr,
 698                                            adapter->total_rx_packets,
 699                                            adapter->total_rx_bytes);
 700         /* conservative mode (itr 3) eliminates the lowest_latency setting */
 701         if (adapter->itr_setting == 3 && adapter->rx_itr == lowest_latency)
 702                 adapter->rx_itr = low_latency;
 703
 704         current_itr = max(adapter->rx_itr, adapter->tx_itr);
 705
 706         switch (current_itr) {
 707         /* counts and packets in update_itr are dependent on these numbers */
 708         case lowest_latency:
 709                 new_itr = 70000;
 710                 break;
 711         case low_latency:
 712                 new_itr = 20000; /* aka hwitr = ~200 */
 713                 break;
 714         case bulk_latency:
 715                 new_itr = 4000;
 716                 break;
 717         default:
 718                 break;
 719         }
 720
 721         if (new_itr != adapter->itr) {
 722                 /*
 723                  * this attempts to bias the interrupt rate towards Bulk
 724                  * by adding intermediate steps when interrupt rate is
 725                  * increasing
 726                  */
 727                 new_itr = new_itr > adapter->itr ?
 728                              min(adapter->itr + (new_itr >> 2), new_itr) :
 729                              new_itr;
 730                 adapter->itr = new_itr;
 731                 adapter->rx_ring->itr_val = 1952;
 732
 733                 if (adapter->msix_entries)
 734                         adapter->rx_ring->set_itr = 1;
 735                 else
 736                         ew32(ITR, 1952);
 737         }
 738 }
 739
 740 /**
 741  * igbvf_clean_tx_irq - Reclaim resources after transmit completes
 742  * @adapter: board private structure
 743  * returns true if ring is completely cleaned
 744  **/
 745 static bool igbvf_clean_tx_irq(struct igbvf_ring *tx_ring)
 746 {
 747         struct igbvf_adapter *adapter = tx_ring->adapter;
 748         struct net_device *netdev = adapter->netdev;
 749         struct igbvf_buffer *buffer_info;
 750         struct sk_buff *skb;
 751         union e1000_adv_tx_desc *tx_desc, *eop_desc;
 752         unsigned int total_bytes = 0, total_packets = 0;
 753         unsigned int i, eop, count = 0;
 754         bool cleaned = false;
 755
 756         i = tx_ring->next_to_clean;
 757         eop = tx_ring->buffer_info[i].next_to_watch;
 758         eop_desc = IGBVF_TX_DESC_ADV(*tx_ring, eop);
 759
 760         while ((eop_desc->wb.status & cpu_to_le32(E1000_TXD_STAT_DD)) &&
 761                (count < tx_ring->count)) {
 762                 rmb();  /* read buffer_info after eop_desc status */
 763                 for (cleaned = false; !cleaned; count++) {
 764                         tx_desc = IGBVF_TX_DESC_ADV(*tx_ring, i);
 765                         buffer_info = &tx_ring->buffer_info[i];
 766                         cleaned = (i == eop);
 767                         skb = buffer_info->skb;
 768
 769                         if (skb) {
 770                                 unsigned int segs, bytecount;
 771
 772                                 /* gso_segs is currently only valid for tcp */
 773                                 segs = skb_shinfo(skb)->gso_segs ?: 1;
 774                                 /* multiply data chunks by size of headers */
 775                                 bytecount = ((segs - 1) * skb_headlen(skb)) +
 776                                             skb->len;
 777                                 total_packets += segs;
 778                                 total_bytes += bytecount;
 779                         }
 780
 781                         igbvf_put_txbuf(adapter, buffer_info);
 782                         tx_desc->wb.status = 0;
 783
 784                         i++;
 785                         if (i == tx_ring->count)
 786                                 i = 0;
 787                 }
 788                 eop = tx_ring->buffer_info[i].next_to_watch;
 789                 eop_desc = IGBVF_TX_DESC_ADV(*tx_ring, eop);
 790         }
 791
 792         tx_ring->next_to_clean = i;
 793
 794         if (unlikely(count &&
 795                      netif_carrier_ok(netdev) &&
 796                      igbvf_desc_unused(tx_ring) >= IGBVF_TX_QUEUE_WAKE)) {
 797                 /* Make sure that anybody stopping the queue after this
 798                  * sees the new next_to_clean.
 799                  */
 800                 smp_mb();
 801                 if (netif_queue_stopped(netdev) &&
 802                     !(test_bit(__IGBVF_DOWN, &adapter->state))) {
 803                         netif_wake_queue(netdev);
 804                         ++adapter->restart_queue;
 805                 }
 806         }
 807
 808         adapter->net_stats.tx_bytes += total_bytes;
 809         adapter->net_stats.tx_packets += total_packets;
 810         return count < tx_ring->count;
 811 }
 812
 813 static irqreturn_t igbvf_msix_other(int irq, void *data)
 814 {
 815         struct net_device *netdev = data;
 816         struct igbvf_adapter *adapter = netdev_priv(netdev);
 817         struct e1000_hw *hw = &adapter->hw;
 818
 819         adapter->int_counter1++;
 820
 821         netif_carrier_off(netdev);
 822         hw->mac.get_link_status = 1;
 823         if (!test_bit(__IGBVF_DOWN, &adapter->state))
 824                 mod_timer(&adapter->watchdog_timer, jiffies + 1);
 825
 826         ew32(EIMS, adapter->eims_other);
 827
 828         return IRQ_HANDLED;
 829 }
 830
 831 static irqreturn_t igbvf_intr_msix_tx(int irq, void *data)
 832 {
 833         struct net_device *netdev = data;
 834         struct igbvf_adapter *adapter = netdev_priv(netdev);
 835         struct e1000_hw *hw = &adapter->hw;
 836         struct igbvf_ring *tx_ring = adapter->tx_ring;
 837
 838
 839         adapter->total_tx_bytes = 0;
 840         adapter->total_tx_packets = 0;
 841
 842         /* auto mask will automatically reenable the interrupt when we write
 843          * EICS */
 844         if (!igbvf_clean_tx_irq(tx_ring))
 845                 /* Ring was not completely cleaned, so fire another interrupt */
 846                 ew32(EICS, tx_ring->eims_value);
 847         else
 848                 ew32(EIMS, tx_ring->eims_value);
 849
 850         return IRQ_HANDLED;
 851 }
 852
 853 static irqreturn_t igbvf_intr_msix_rx(int irq, void *data)
 854 {
 855         struct net_device *netdev = data;
 856         struct igbvf_adapter *adapter = netdev_priv(netdev);
 857
 858         adapter->int_counter0++;
 859
 860         /* Write the ITR value calculated at the end of the
 861          * previous interrupt.
 862          */
 863         if (adapter->rx_ring->set_itr) {
 864                 writel(adapter->rx_ring->itr_val,
 865                        adapter->hw.hw_addr + adapter->rx_ring->itr_register);
 866                 adapter->rx_ring->set_itr = 0;
 867         }
 868
 869         if (napi_schedule_prep(&adapter->rx_ring->napi)) {
 870                 adapter->total_rx_bytes = 0;
 871                 adapter->total_rx_packets = 0;
 872                 __napi_schedule(&adapter->rx_ring->napi);
 873         }
 874
 875         return IRQ_HANDLED;
 876 }
 877
 878 #define IGBVF_NO_QUEUE -1
 879
 880 static void igbvf_assign_vector(struct igbvf_adapter *adapter, int rx_queue,
 881                                 int tx_queue, int msix_vector)
 882 {
 883         struct e1000_hw *hw = &adapter->hw;
 884         u32 ivar, index;
 885
 886         /* 82576 uses a table-based method for assigning vectors.
 887            Each queue has a single entry in the table to which we write
 888            a vector number along with a "valid" bit.  Sadly, the layout
 889            of the table is somewhat counterintuitive. */
 890         if (rx_queue > IGBVF_NO_QUEUE) {
 891                 index = (rx_queue >> 1);
 892                 ivar = array_er32(IVAR0, index);
 893                 if (rx_queue & 0x1) {
 894                         /* vector goes into third byte of register */
 895                         ivar = ivar & 0xFF00FFFF;
 896                         ivar |= (msix_vector | E1000_IVAR_VALID) << 16;
 897                 } else {
 898                         /* vector goes into low byte of register */
 899                         ivar = ivar & 0xFFFFFF00;
 900                         ivar |= msix_vector | E1000_IVAR_VALID;
 901                 }
 902                 adapter->rx_ring[rx_queue].eims_value = 1 << msix_vector;
 903                 array_ew32(IVAR0, index, ivar);
 904         }
 905         if (tx_queue > IGBVF_NO_QUEUE) {
 906                 index = (tx_queue >> 1);
 907                 ivar = array_er32(IVAR0, index);
 908                 if (tx_queue & 0x1) {
 909                         /* vector goes into high byte of register */
 910                         ivar = ivar & 0x00FFFFFF;
 911                         ivar |= (msix_vector | E1000_IVAR_VALID) << 24;
 912                 } else {
 913                         /* vector goes into second byte of register */
 914                         ivar = ivar & 0xFFFF00FF;
 915                         ivar |= (msix_vector | E1000_IVAR_VALID) << 8;
 916                 }
 917                 adapter->tx_ring[tx_queue].eims_value = 1 << msix_vector;
 918                 array_ew32(IVAR0, index, ivar);
 919         }
 920 }
 921
 922 /**
 923  * igbvf_configure_msix - Configure MSI-X hardware
 924  *
 925  * igbvf_configure_msix sets up the hardware to properly
 926  * generate MSI-X interrupts.
 927  **/
 928 static void igbvf_configure_msix(struct igbvf_adapter *adapter)
 929 {
 930         u32 tmp;
 931         struct e1000_hw *hw = &adapter->hw;
 932         struct igbvf_ring *tx_ring = adapter->tx_ring;
 933         struct igbvf_ring *rx_ring = adapter->rx_ring;
 934         int vector = 0;
 935
 936         adapter->eims_enable_mask = 0;
 937
 938         igbvf_assign_vector(adapter, IGBVF_NO_QUEUE, 0, vector++);
 939         adapter->eims_enable_mask |= tx_ring->eims_value;
 940         if (tx_ring->itr_val)
 941                 writel(tx_ring->itr_val,
 942                        hw->hw_addr + tx_ring->itr_register);
 943         else
 944                 writel(1952, hw->hw_addr + tx_ring->itr_register);
 945
 946         igbvf_assign_vector(adapter, 0, IGBVF_NO_QUEUE, vector++);
 947         adapter->eims_enable_mask |= rx_ring->eims_value;
 948         if (rx_ring->itr_val)
 949                 writel(rx_ring->itr_val,
 950                        hw->hw_addr + rx_ring->itr_register);
 951         else
 952                 writel(1952, hw->hw_addr + rx_ring->itr_register);
 953
 954         /* set vector for other causes, i.e. link changes */
 955
 956         tmp = (vector++ | E1000_IVAR_VALID);
 957
 958         ew32(IVAR_MISC, tmp);
 959
 960         adapter->eims_enable_mask = (1 << (vector)) - 1;
 961         adapter->eims_other = 1 << (vector - 1);
 962         e1e_flush();
 963 }
 964
 965 static void igbvf_reset_interrupt_capability(struct igbvf_adapter *adapter)
 966 {
 967         if (adapter->msix_entries) {
 968                 pci_disable_msix(adapter->pdev);
 969                 kfree(adapter->msix_entries);
 970                 adapter->msix_entries = NULL;
 971         }
 972 }
 973
 974 /**
 975  * igbvf_set_interrupt_capability - set MSI or MSI-X if supported
 976  *
 977  * Attempt to configure interrupts using the best available
 978  * capabilities of the hardware and kernel.
 979  **/
 980 static void igbvf_set_interrupt_capability(struct igbvf_adapter *adapter)
 981 {
 982         int err = -ENOMEM;
 983         int i;
 984
 985         /* we allocate 3 vectors, 1 for tx, 1 for rx, one for pf messages */
 986         adapter->msix_entries = kcalloc(3, sizeof(struct msix_entry),
 987                                         GFP_KERNEL);
 988         if (adapter->msix_entries) {
 989                 for (i = 0; i < 3; i++)
 990                         adapter->msix_entries[i].entry = i;
 991
 992                 err = pci_enable_msix(adapter->pdev,
 993                                       adapter->msix_entries, 3);
 994         }
 995
 996         if (err) {
 997                 /* MSI-X failed */
 998                 dev_err(&adapter->pdev->dev,
 999                         "Failed to initialize MSI-X interrupts.\n");
1000                 igbvf_reset_interrupt_capability(adapter);
1001         }
1002 }
1003
1004 /**
1005  * igbvf_request_msix - Initialize MSI-X interrupts
1006  *
1007  * igbvf_request_msix allocates MSI-X vectors and requests interrupts from the
1008  * kernel.
1009  **/
1010 static int igbvf_request_msix(struct igbvf_adapter *adapter)
1011 {
1012         struct net_device *netdev = adapter->netdev;
1013         int err = 0, vector = 0;
1014
1015         if (strlen(netdev->name) < (IFNAMSIZ - 5)) {
1016                 sprintf(adapter->tx_ring->name, "%s-tx-0", netdev->name);
1017                 sprintf(adapter->rx_ring->name, "%s-rx-0", netdev->name);
1018         } else {
1019                 memcpy(adapter->tx_ring->name, netdev->name, IFNAMSIZ);
1020                 memcpy(adapter->rx_ring->name, netdev->name, IFNAMSIZ);
1021         }
1022
1023         err = request_irq(adapter->msix_entries[vector].vector,
1024                           igbvf_intr_msix_tx, 0, adapter->tx_ring->name,
1025                           netdev);
1026         if (err)
1027                 goto out;
1028
1029         adapter->tx_ring->itr_register = E1000_EITR(vector);
1030         adapter->tx_ring->itr_val = 1952;
1031         vector++;
1032
1033         err = request_irq(adapter->msix_entries[vector].vector,
1034                           igbvf_intr_msix_rx, 0, adapter->rx_ring->name,
1035                           netdev);
1036         if (err)
1037                 goto out;
1038
1039         adapter->rx_ring->itr_register = E1000_EITR(vector);
1040         adapter->rx_ring->itr_val = 1952;
1041         vector++;
1042
1043         err = request_irq(adapter->msix_entries[vector].vector,
1044                           igbvf_msix_other, 0, netdev->name, netdev);
1045         if (err)
1046                 goto out;
1047
1048         igbvf_configure_msix(adapter);
1049         return 0;
1050 out:
1051         return err;
1052 }
1053
1054 /**
1055  * igbvf_alloc_queues - Allocate memory for all rings
1056  * @adapter: board private structure to initialize
1057  **/
1058 static int __devinit igbvf_alloc_queues(struct igbvf_adapter *adapter)
1059 {
1060         struct net_device *netdev = adapter->netdev;
1061
1062         adapter->tx_ring = kzalloc(sizeof(struct igbvf_ring), GFP_KERNEL);
1063         if (!adapter->tx_ring)
1064                 return -ENOMEM;
1065
1066         adapter->rx_ring = kzalloc(sizeof(struct igbvf_ring), GFP_KERNEL);
1067         if (!adapter->rx_ring) {
1068                 kfree(adapter->tx_ring);
1069                 return -ENOMEM;
1070         }
1071
1072         netif_napi_add(netdev, &adapter->rx_ring->napi, igbvf_poll, 64);
1073
1074         return 0;
1075 }
1076
1077 /**
1078  * igbvf_request_irq - initialize interrupts
1079  *
1080  * Attempts to configure interrupts using the best available
1081  * capabilities of the hardware and kernel.
1082  **/
1083 static int igbvf_request_irq(struct igbvf_adapter *adapter)
1084 {
1085         int err = -1;
1086
1087         /* igbvf supports msi-x only */
1088         if (adapter->msix_entries)
1089                 err = igbvf_request_msix(adapter);
1090
1091         if (!err)
1092                 return err;
1093
1094         dev_err(&adapter->pdev->dev,
1095                 "Unable to allocate interrupt, Error: %d\n", err);
1096
1097         return err;
1098 }
1099
1100 static void igbvf_free_irq(struct igbvf_adapter *adapter)
1101 {
1102         struct net_device *netdev = adapter->netdev;
1103         int vector;
1104
1105         if (adapter->msix_entries) {
1106                 for (vector = 0; vector < 3; vector++)
1107                         free_irq(adapter->msix_entries[vector].vector, netdev);
1108         }
1109 }
1110
1111 /**
1112  * igbvf_irq_disable - Mask off interrupt generation on the NIC
1113  **/
1114 static void igbvf_irq_disable(struct igbvf_adapter *adapter)
1115 {
1116         struct e1000_hw *hw = &adapter->hw;
1117
1118         ew32(EIMC, ~0);
1119
1120         if (adapter->msix_entries)
1121                 ew32(EIAC, 0);
1122 }
1123
1124 /**
1125  * igbvf_irq_enable - Enable default interrupt generation settings
1126  **/
1127 static void igbvf_irq_enable(struct igbvf_adapter *adapter)
1128 {
1129         struct e1000_hw *hw = &adapter->hw;
1130
1131         ew32(EIAC, adapter->eims_enable_mask);
1132         ew32(EIAM, adapter->eims_enable_mask);
1133         ew32(EIMS, adapter->eims_enable_mask);
1134 }
1135
1136 /**
1137  * igbvf_poll - NAPI Rx polling callback
1138  * @napi: struct associated with this polling callback
1139  * @budget: amount of packets driver is allowed to process this poll
1140  **/
1141 static int igbvf_poll(struct napi_struct *napi, int budget)
1142 {
1143         struct igbvf_ring *rx_ring = container_of(napi, struct igbvf_ring, napi);
1144         struct igbvf_adapter *adapter = rx_ring->adapter;
1145         struct e1000_hw *hw = &adapter->hw;
1146         int work_done = 0;
1147
1148         igbvf_clean_rx_irq(adapter, &work_done, budget);
1149
1150         /* If not enough Rx work done, exit the polling mode */
1151         if (work_done < budget) {
1152                 napi_complete(napi);
1153
1154                 if (adapter->itr_setting & 3)
1155                         igbvf_set_itr(adapter);
1156
1157                 if (!test_bit(__IGBVF_DOWN, &adapter->state))
1158                         ew32(EIMS, adapter->rx_ring->eims_value);
1159         }
1160
1161         return work_done;
1162 }
1163
1164 /**
1165  * igbvf_set_rlpml - set receive large packet maximum length
1166  * @adapter: board private structure
1167  *
1168  * Configure the maximum size of packets that will be received
1169  */
1170 static void igbvf_set_rlpml(struct igbvf_adapter *adapter)
1171 {
1172         int max_frame_size;
1173         struct e1000_hw *hw = &adapter->hw;
1174
1175         max_frame_size = adapter->max_frame_size + VLAN_TAG_SIZE;
1176         e1000_rlpml_set_vf(hw, max_frame_size);
1177 }
1178
1179 static int igbvf_vlan_rx_add_vid(struct net_device *netdev, u16 vid)
1180 {
1181         struct igbvf_adapter *adapter = netdev_priv(netdev);
1182         struct e1000_hw *hw = &adapter->hw;
1183
1184         if (hw->mac.ops.set_vfta(hw, vid, true)) {
1185                 dev_err(&adapter->pdev->dev, "Failed to add vlan id %d\n", vid);
1186                 return -EINVAL;
1187         }
1188         set_bit(vid, adapter->active_vlans);
1189         return 0;
1190 }
1191
1192 static int igbvf_vlan_rx_kill_vid(struct net_device *netdev, u16 vid)
1193 {
1194         struct igbvf_adapter *adapter = netdev_priv(netdev);
1195         struct e1000_hw *hw = &adapter->hw;
1196
1197         if (hw->mac.ops.set_vfta(hw, vid, false)) {
1198                 dev_err(&adapter->pdev->dev,
1199                         "Failed to remove vlan id %d\n", vid);
1200                 return -EINVAL;
1201         }
1202         clear_bit(vid, adapter->active_vlans);
1203         return 0;
1204 }
1205
1206 static void igbvf_restore_vlan(struct igbvf_adapter *adapter)
1207 {
1208         u16 vid;
1209
1210         for_each_set_bit(vid, adapter->active_vlans, VLAN_N_VID)
1211                 igbvf_vlan_rx_add_vid(adapter->netdev, vid);
1212 }
1213
1214 /**
1215  * igbvf_configure_tx - Configure Transmit Unit after Reset
1216  * @adapter: board private structure
1217  *
1218  * Configure the Tx unit of the MAC after a reset.
1219  **/
1220 static void igbvf_configure_tx(struct igbvf_adapter *adapter)
1221 {
1222         struct e1000_hw *hw = &adapter->hw;
1223         struct igbvf_ring *tx_ring = adapter->tx_ring;
1224         u64 tdba;
1225         u32 txdctl, dca_txctrl;
1226
1227         /* disable transmits */
1228         txdctl = er32(TXDCTL(0));
1229         ew32(TXDCTL(0), txdctl & ~E1000_TXDCTL_QUEUE_ENABLE);
1230         e1e_flush();
1231         msleep(10);
1232
1233         /* Setup the HW Tx Head and Tail descriptor pointers */
1234         ew32(TDLEN(0), tx_ring->count * sizeof(union e1000_adv_tx_desc));
1235         tdba = tx_ring->dma;
1236         ew32(TDBAL(0), (tdba & DMA_BIT_MASK(32)));
1237         ew32(TDBAH(0), (tdba >> 32));
1238         ew32(TDH(0), 0);
1239         ew32(TDT(0), 0);
1240         tx_ring->head = E1000_TDH(0);
1241         tx_ring->tail = E1000_TDT(0);
1242
1243         /* Turn off Relaxed Ordering on head write-backs.  The writebacks
1244          * MUST be delivered in order or it will completely screw up
1245          * our bookeeping.
1246          */
1247         dca_txctrl = er32(DCA_TXCTRL(0));
1248         dca_txctrl &= ~E1000_DCA_TXCTRL_TX_WB_RO_EN;
1249         ew32(DCA_TXCTRL(0), dca_txctrl);
1250
1251         /* enable transmits */
1252         txdctl |= E1000_TXDCTL_QUEUE_ENABLE;
1253         ew32(TXDCTL(0), txdctl);
1254
1255         /* Setup Transmit Descriptor Settings for eop descriptor */
1256         adapter->txd_cmd = E1000_ADVTXD_DCMD_EOP | E1000_ADVTXD_DCMD_IFCS;
1257
1258         /* enable Report Status bit */
1259         adapter->txd_cmd |= E1000_ADVTXD_DCMD_RS;
1260 }
1261
1262 /**
1263  * igbvf_setup_srrctl - configure the receive control registers
1264  * @adapter: Board private structure
1265  **/
1266 static void igbvf_setup_srrctl(struct igbvf_adapter *adapter)
1267 {
1268         struct e1000_hw *hw = &adapter->hw;
1269         u32 srrctl = 0;
1270
1271         srrctl &= ~(E1000_SRRCTL_DESCTYPE_MASK |
1272                     E1000_SRRCTL_BSIZEHDR_MASK |
1273                     E1000_SRRCTL_BSIZEPKT_MASK);
1274
1275         /* Enable queue drop to avoid head of line blocking */
1276         srrctl |= E1000_SRRCTL_DROP_EN;
1277
1278         /* Setup buffer sizes */
1279         srrctl |= ALIGN(adapter->rx_buffer_len, 1024) >>
1280                   E1000_SRRCTL_BSIZEPKT_SHIFT;
1281
1282         if (adapter->rx_buffer_len < 2048) {
1283                 adapter->rx_ps_hdr_size = 0;
1284                 srrctl |= E1000_SRRCTL_DESCTYPE_ADV_ONEBUF;
1285         } else {
1286                 adapter->rx_ps_hdr_size = 128;
1287                 srrctl |= adapter->rx_ps_hdr_size <<
1288                           E1000_SRRCTL_BSIZEHDRSIZE_SHIFT;
1289                 srrctl |= E1000_SRRCTL_DESCTYPE_HDR_SPLIT_ALWAYS;
1290         }
1291
1292         ew32(SRRCTL(0), srrctl);
1293 }
1294
1295 /**
1296  * igbvf_configure_rx - Configure Receive Unit after Reset
1297  * @adapter: board private structure
1298  *
1299  * Configure the Rx unit of the MAC after a reset.
1300  **/
1301 static void igbvf_configure_rx(struct igbvf_adapter *adapter)
1302 {
1303         struct e1000_hw *hw = &adapter->hw;
1304         struct igbvf_ring *rx_ring = adapter->rx_ring;
1305         u64 rdba;
1306         u32 rdlen, rxdctl;
1307
1308         /* disable receives */
1309         rxdctl = er32(RXDCTL(0));
1310         ew32(RXDCTL(0), rxdctl & ~E1000_RXDCTL_QUEUE_ENABLE);
1311         e1e_flush();
1312         msleep(10);
1313
1314         rdlen = rx_ring->count * sizeof(union e1000_adv_rx_desc);
1315
1316         /*
1317          * Setup the HW Rx Head and Tail Descriptor Pointers and
1318          * the Base and Length of the Rx Descriptor Ring
1319          */
1320         rdba = rx_ring->dma;
1321         ew32(RDBAL(0), (rdba & DMA_BIT_MASK(32)));
1322         ew32(RDBAH(0), (rdba >> 32));
1323         ew32(RDLEN(0), rx_ring->count * sizeof(union e1000_adv_rx_desc));
1324         rx_ring->head = E1000_RDH(0);
1325         rx_ring->tail = E1000_RDT(0);
1326         ew32(RDH(0), 0);
1327         ew32(RDT(0), 0);
1328
1329         rxdctl |= E1000_RXDCTL_QUEUE_ENABLE;
1330         rxdctl &= 0xFFF00000;
1331         rxdctl |= IGBVF_RX_PTHRESH;
1332         rxdctl |= IGBVF_RX_HTHRESH << 8;
1333         rxdctl |= IGBVF_RX_WTHRESH << 16;
1334
1335         igbvf_set_rlpml(adapter);
1336
1337         /* enable receives */
1338         ew32(RXDCTL(0), rxdctl);
1339 }
1340
1341 /**
1342  * igbvf_set_multi - Multicast and Promiscuous mode set
1343  * @netdev: network interface device structure
1344  *
1345  * The set_multi entry point is called whenever the multicast address
1346  * list or the network interface flags are updated.  This routine is
1347  * responsible for configuring the hardware for proper multicast,
1348  * promiscuous mode, and all-multi behavior.
1349  **/
1350 static void igbvf_set_multi(struct net_device *netdev)
1351 {
1352         struct igbvf_adapter *adapter = netdev_priv(netdev);
1353         struct e1000_hw *hw = &adapter->hw;
1354         struct netdev_hw_addr *ha;
1355         u8  *mta_list = NULL;
1356         int i;
1357
1358         if (!netdev_mc_empty(netdev)) {
1359                 mta_list = kmalloc(netdev_mc_count(netdev) * 6, GFP_ATOMIC);
1360                 if (!mta_list) {
1361                         dev_err(&adapter->pdev->dev,
1362                                 "failed to allocate multicast filter list\n");
1363                         return;
1364                 }
1365         }
1366
1367         /* prepare a packed array of only addresses. */
1368         i = 0;
1369         netdev_for_each_mc_addr(ha, netdev)
1370                 memcpy(mta_list + (i++ * ETH_ALEN), ha->addr, ETH_ALEN);
1371
1372         hw->mac.ops.update_mc_addr_list(hw, mta_list, i, 0, 0);
1373         kfree(mta_list);
1374 }
1375
1376 /**
1377  * igbvf_configure - configure the hardware for Rx and Tx
1378  * @adapter: private board structure
1379  **/
1380 static void igbvf_configure(struct igbvf_adapter *adapter)
1381 {
1382         igbvf_set_multi(adapter->netdev);
1383
1384         igbvf_restore_vlan(adapter);
1385
1386         igbvf_configure_tx(adapter);
1387         igbvf_setup_srrctl(adapter);
1388         igbvf_configure_rx(adapter);
1389         igbvf_alloc_rx_buffers(adapter->rx_ring,
1390                                igbvf_desc_unused(adapter->rx_ring));
1391 }
1392
1393 /* igbvf_reset - bring the hardware into a known good state
1394  *
1395  * This function boots the hardware and enables some settings that
1396  * require a configuration cycle of the hardware - those cannot be
1397  * set/changed during runtime. After reset the device needs to be
1398  * properly configured for Rx, Tx etc.
1399  */
1400 static void igbvf_reset(struct igbvf_adapter *adapter)
1401 {
1402         struct e1000_mac_info *mac = &adapter->hw.mac;
1403         struct net_device *netdev = adapter->netdev;
1404         struct e1000_hw *hw = &adapter->hw;
1405
1406         /* Allow time for pending master requests to run */
1407         if (mac->ops.reset_hw(hw))
1408                 dev_err(&adapter->pdev->dev, "PF still resetting\n");
1409
1410         mac->ops.init_hw(hw);
1411
1412         if (is_valid_ether_addr(adapter->hw.mac.addr)) {
1413                 memcpy(netdev->dev_addr, adapter->hw.mac.addr,
1414                        netdev->addr_len);
1415                 memcpy(netdev->perm_addr, adapter->hw.mac.addr,
1416                        netdev->addr_len);
1417         }
1418
1419         adapter->last_reset = jiffies;
1420 }
1421
1422 int igbvf_up(struct igbvf_adapter *adapter)
1423 {
1424         struct e1000_hw *hw = &adapter->hw;
1425
1426         /* hardware has been reset, we need to reload some things */
1427         igbvf_configure(adapter);
1428
1429         clear_bit(__IGBVF_DOWN, &adapter->state);
1430
1431         napi_enable(&adapter->rx_ring->napi);
1432         if (adapter->msix_entries)
1433                 igbvf_configure_msix(adapter);
1434
1435         /* Clear any pending interrupts. */
1436         er32(EICR);
1437         igbvf_irq_enable(adapter);
1438
1439         /* start the watchdog */
1440         hw->mac.get_link_status = 1;
1441         mod_timer(&adapter->watchdog_timer, jiffies + 1);
1442
1443
1444         return 0;
1445 }
1446
1447 void igbvf_down(struct igbvf_adapter *adapter)
1448 {
1449         struct net_device *netdev = adapter->netdev;
1450         struct e1000_hw *hw = &adapter->hw;
1451         u32 rxdctl, txdctl;
1452
1453         /*
1454          * signal that we're down so the interrupt handler does not
1455          * reschedule our watchdog timer
1456          */
1457         set_bit(__IGBVF_DOWN, &adapter->state);
1458
1459         /* disable receives in the hardware */
1460         rxdctl = er32(RXDCTL(0));
1461         ew32(RXDCTL(0), rxdctl & ~E1000_RXDCTL_QUEUE_ENABLE);
1462
1463         netif_stop_queue(netdev);
1464
1465         /* disable transmits in the hardware */
1466         txdctl = er32(TXDCTL(0));
1467         ew32(TXDCTL(0), txdctl & ~E1000_TXDCTL_QUEUE_ENABLE);
1468
1469         /* flush both disables and wait for them to finish */
1470         e1e_flush();
1471         msleep(10);
1472
1473         napi_disable(&adapter->rx_ring->napi);
1474
1475         igbvf_irq_disable(adapter);
1476
1477         del_timer_sync(&adapter->watchdog_timer);
1478
1479         netif_carrier_off(netdev);
1480
1481         /* record the stats before reset*/
1482         igbvf_update_stats(adapter);
1483
1484         adapter->link_speed = 0;
1485         adapter->link_duplex = 0;
1486
1487         igbvf_reset(adapter);
1488         igbvf_clean_tx_ring(adapter->tx_ring);
1489         igbvf_clean_rx_ring(adapter->rx_ring);
1490 }
1491
1492 void igbvf_reinit_locked(struct igbvf_adapter *adapter)
1493 {
1494         might_sleep();
1495         while (test_and_set_bit(__IGBVF_RESETTING, &adapter->state))
1496                 msleep(1);
1497         igbvf_down(adapter);
1498         igbvf_up(adapter);
1499         clear_bit(__IGBVF_RESETTING, &adapter->state);
1500 }
1501
1502 /**
1503  * igbvf_sw_init - Initialize general software structures (struct igbvf_adapter)
1504  * @adapter: board private structure to initialize
1505  *
1506  * igbvf_sw_init initializes the Adapter private data structure.
1507  * Fields are initialized based on PCI device information and
1508  * OS network device settings (MTU size).
1509  **/
1510 static int __devinit igbvf_sw_init(struct igbvf_adapter *adapter)
1511 {
1512         struct net_device *netdev = adapter->netdev;
1513         s32 rc;
1514
1515         adapter->rx_buffer_len = ETH_FRAME_LEN + VLAN_HLEN + ETH_FCS_LEN;
1516         adapter->rx_ps_hdr_size = 0;
1517         adapter->max_frame_size = netdev->mtu + ETH_HLEN + ETH_FCS_LEN;
1518         adapter->min_frame_size = ETH_ZLEN + ETH_FCS_LEN;
1519
1520         adapter->tx_int_delay = 8;
1521         adapter->tx_abs_int_delay = 32;
1522         adapter->rx_int_delay = 0;
1523         adapter->rx_abs_int_delay = 8;
1524         adapter->itr_setting = 3;
1525         adapter->itr = 20000;
1526
1527         /* Set various function pointers */
1528         adapter->ei->init_ops(&adapter->hw);
1529
1530         rc = adapter->hw.mac.ops.init_params(&adapter->hw);
1531         if (rc)
1532                 return rc;
1533
1534         rc = adapter->hw.mbx.ops.init_params(&adapter->hw);
1535         if (rc)
1536                 return rc;
1537
1538         igbvf_set_interrupt_capability(adapter);
1539
1540         if (igbvf_alloc_queues(adapter))
1541                 return -ENOMEM;
1542
1543         spin_lock_init(&adapter->tx_queue_lock);
1544
1545         /* Explicitly disable IRQ since the NIC can be in any state. */
1546         igbvf_irq_disable(adapter);
1547
1548         spin_lock_init(&adapter->stats_lock);
1549
1550         set_bit(__IGBVF_DOWN, &adapter->state);
1551         return 0;
1552 }
1553
1554 static void igbvf_initialize_last_counter_stats(struct igbvf_adapter *adapter)
1555 {
1556         struct e1000_hw *hw = &adapter->hw;
1557
1558         adapter->stats.last_gprc = er32(VFGPRC);
1559         adapter->stats.last_gorc = er32(VFGORC);
1560         adapter->stats.last_gptc = er32(VFGPTC);
1561         adapter->stats.last_gotc = er32(VFGOTC);
1562         adapter->stats.last_mprc = er32(VFMPRC);
1563         adapter->stats.last_gotlbc = er32(VFGOTLBC);
1564         adapter->stats.last_gptlbc = er32(VFGPTLBC);
1565         adapter->stats.last_gorlbc = er32(VFGORLBC);
1566         adapter->stats.last_gprlbc = er32(VFGPRLBC);
1567
1568         adapter->stats.base_gprc = er32(VFGPRC);
1569         adapter->stats.base_gorc = er32(VFGORC);
1570         adapter->stats.base_gptc = er32(VFGPTC);
1571         adapter->stats.base_gotc = er32(VFGOTC);
1572         adapter->stats.base_mprc = er32(VFMPRC);
1573         adapter->stats.base_gotlbc = er32(VFGOTLBC);
1574         adapter->stats.base_gptlbc = er32(VFGPTLBC);
1575         adapter->stats.base_gorlbc = er32(VFGORLBC);
1576         adapter->stats.base_gprlbc = er32(VFGPRLBC);
1577 }
1578
1579 /**
1580  * igbvf_open - Called when a network interface is made active
1581  * @netdev: network interface device structure
1582  *
1583  * Returns 0 on success, negative value on failure
1584  *
1585  * The open entry point is called when a network interface is made
1586  * active by the system (IFF_UP).  At this point all resources needed
1587  * for transmit and receive operations are allocated, the interrupt
1588  * handler is registered with the OS, the watchdog timer is started,
1589  * and the stack is notified that the interface is ready.
1590  **/
1591 static int igbvf_open(struct net_device *netdev)
1592 {
1593         struct igbvf_adapter *adapter = netdev_priv(netdev);
1594         struct e1000_hw *hw = &adapter->hw;
1595         int err;
1596
1597         /* disallow open during test */
1598         if (test_bit(__IGBVF_TESTING, &adapter->state))
1599                 return -EBUSY;
1600
1601         /* allocate transmit descriptors */
1602         err = igbvf_setup_tx_resources(adapter, adapter->tx_ring);
1603         if (err)
1604                 goto err_setup_tx;
1605
1606         /* allocate receive descriptors */
1607         err = igbvf_setup_rx_resources(adapter, adapter->rx_ring);
1608         if (err)
1609                 goto err_setup_rx;
1610
1611         /*
1612          * before we allocate an interrupt, we must be ready to handle it.
1613          * Setting DEBUG_SHIRQ in the kernel makes it fire an interrupt
1614          * as soon as we call pci_request_irq, so we have to setup our
1615          * clean_rx handler before we do so.
1616          */
1617         igbvf_configure(adapter);
1618
1619         err = igbvf_request_irq(adapter);
1620         if (err)
1621                 goto err_req_irq;
1622
1623         /* From here on the code is the same as igbvf_up() */
1624         clear_bit(__IGBVF_DOWN, &adapter->state);
1625
1626         napi_enable(&adapter->rx_ring->napi);
1627
1628         /* clear any pending interrupts */
1629         er32(EICR);
1630
1631         igbvf_irq_enable(adapter);
1632
1633         /* start the watchdog */
1634         hw->mac.get_link_status = 1;
1635         mod_timer(&adapter->watchdog_timer, jiffies + 1);
1636
1637         return 0;
1638
1639 err_req_irq:
1640         igbvf_free_rx_resources(adapter->rx_ring);
1641 err_setup_rx:
1642         igbvf_free_tx_resources(adapter->tx_ring);
1643 err_setup_tx:
1644         igbvf_reset(adapter);
1645
1646         return err;
1647 }
1648
1649 /**
1650  * igbvf_close - Disables a network interface
1651  * @netdev: network interface device structure
1652  *
1653  * Returns 0, this is not allowed to fail
1654  *
1655  * The close entry point is called when an interface is de-activated
1656  * by the OS.  The hardware is still under the drivers control, but
1657  * needs to be disabled.  A global MAC reset is issued to stop the
1658  * hardware, and all transmit and receive resources are freed.
1659  **/
1660 static int igbvf_close(struct net_device *netdev)
1661 {
1662         struct igbvf_adapter *adapter = netdev_priv(netdev);
1663
1664         WARN_ON(test_bit(__IGBVF_RESETTING, &adapter->state));
1665         igbvf_down(adapter);
1666
1667         igbvf_free_irq(adapter);
1668
1669         igbvf_free_tx_resources(adapter->tx_ring);
1670         igbvf_free_rx_resources(adapter->rx_ring);
1671
1672         return 0;
1673 }
1674 /**
1675  * igbvf_set_mac - Change the Ethernet Address of the NIC
1676  * @netdev: network interface device structure
1677  * @p: pointer to an address structure
1678  *
1679  * Returns 0 on success, negative on failure
1680  **/
1681 static int igbvf_set_mac(struct net_device *netdev, void *p)
1682 {
1683         struct igbvf_adapter *adapter = netdev_priv(netdev);
1684         struct e1000_hw *hw = &adapter->hw;
1685         struct sockaddr *addr = p;
1686
1687         if (!is_valid_ether_addr(addr->sa_data))
1688                 return -EADDRNOTAVAIL;
1689
1690         memcpy(hw->mac.addr, addr->sa_data, netdev->addr_len);
1691
1692         hw->mac.ops.rar_set(hw, hw->mac.addr, 0);
1693
1694         if (memcmp(addr->sa_data, hw->mac.addr, 6))
1695                 return -EADDRNOTAVAIL;
1696
1697         memcpy(netdev->dev_addr, addr->sa_data, netdev->addr_len);
1698
1699         return 0;
1700 }
1701
1702 #define UPDATE_VF_COUNTER(reg, name)                                    \
1703         {                                                               \
1704                 u32 current_counter = er32(reg);                        \
1705                 if (current_counter < adapter->stats.last_##name)       \
1706                         adapter->stats.name += 0x100000000LL;           \
1707                 adapter->stats.last_##name = current_counter;           \
1708                 adapter->stats.name &= 0xFFFFFFFF00000000LL;            \
1709                 adapter->stats.name |= current_counter;                 \
1710         }
1711
1712 /**
1713  * igbvf_update_stats - Update the board statistics counters
1714  * @adapter: board private structure
1715 **/
1716 void igbvf_update_stats(struct igbvf_adapter *adapter)
1717 {
1718         struct e1000_hw *hw = &adapter->hw;
1719         struct pci_dev *pdev = adapter->pdev;
1720
1721         /*
1722          * Prevent stats update while adapter is being reset, link is down
1723          * or if the pci connection is down.
1724          */
1725         if (adapter->link_speed == 0)
1726                 return;
1727
1728         if (test_bit(__IGBVF_RESETTING, &adapter->state))
1729                 return;
1730
1731         if (pci_channel_offline(pdev))
1732                 return;
1733
1734         UPDATE_VF_COUNTER(VFGPRC, gprc);
1735         UPDATE_VF_COUNTER(VFGORC, gorc);
1736         UPDATE_VF_COUNTER(VFGPTC, gptc);
1737         UPDATE_VF_COUNTER(VFGOTC, gotc);
1738         UPDATE_VF_COUNTER(VFMPRC, mprc);
1739         UPDATE_VF_COUNTER(VFGOTLBC, gotlbc);
1740         UPDATE_VF_COUNTER(VFGPTLBC, gptlbc);
1741         UPDATE_VF_COUNTER(VFGORLBC, gorlbc);
1742         UPDATE_VF_COUNTER(VFGPRLBC, gprlbc);
1743
1744         /* Fill out the OS statistics structure */
1745         adapter->net_stats.multicast = adapter->stats.mprc;
1746 }
1747
1748 static void igbvf_print_link_info(struct igbvf_adapter *adapter)
1749 {
1750         dev_info(&adapter->pdev->dev, "Link is Up %d Mbps %s Duplex\n",
1751                  adapter->link_speed,
1752                  adapter->link_duplex == FULL_DUPLEX ? "Full" : "Half");
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                 const 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 = skb_frag_size(frag);
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         if (hw->mac.type == e1000_vfadapt_i350)
2529                 dev_info(&pdev->dev, "Intel(R) I350 Virtual Function\n");
2530         else
2531                 dev_info(&pdev->dev, "Intel(R) 82576 Virtual Function\n");
2532         dev_info(&pdev->dev, "Address: %pM\n", netdev->dev_addr);
2533 }
2534
2535 static int igbvf_set_features(struct net_device *netdev,
2536         netdev_features_t features)
2537 {
2538         struct igbvf_adapter *adapter = netdev_priv(netdev);
2539
2540         if (features & NETIF_F_RXCSUM)
2541                 adapter->flags &= ~IGBVF_FLAG_RX_CSUM_DISABLED;
2542         else
2543                 adapter->flags |= IGBVF_FLAG_RX_CSUM_DISABLED;
2544
2545         return 0;
2546 }
2547
2548 static const struct net_device_ops igbvf_netdev_ops = {
2549         .ndo_open                       = igbvf_open,
2550         .ndo_stop                       = igbvf_close,
2551         .ndo_start_xmit                 = igbvf_xmit_frame,
2552         .ndo_get_stats                  = igbvf_get_stats,
2553         .ndo_set_rx_mode                = igbvf_set_multi,
2554         .ndo_set_mac_address            = igbvf_set_mac,
2555         .ndo_change_mtu                 = igbvf_change_mtu,
2556         .ndo_do_ioctl                   = igbvf_ioctl,
2557         .ndo_tx_timeout                 = igbvf_tx_timeout,
2558         .ndo_vlan_rx_add_vid            = igbvf_vlan_rx_add_vid,
2559         .ndo_vlan_rx_kill_vid           = igbvf_vlan_rx_kill_vid,
2560 #ifdef CONFIG_NET_POLL_CONTROLLER
2561         .ndo_poll_controller            = igbvf_netpoll,
2562 #endif
2563         .ndo_set_features               = igbvf_set_features,
2564 };
2565
2566 /**
2567  * igbvf_probe - Device Initialization Routine
2568  * @pdev: PCI device information struct
2569  * @ent: entry in igbvf_pci_tbl
2570  *
2571  * Returns 0 on success, negative on failure
2572  *
2573  * igbvf_probe initializes an adapter identified by a pci_dev structure.
2574  * The OS initialization, configuring of the adapter private structure,
2575  * and a hardware reset occur.
2576  **/
2577 static int __devinit igbvf_probe(struct pci_dev *pdev,
2578                                  const struct pci_device_id *ent)
2579 {
2580         struct net_device *netdev;
2581         struct igbvf_adapter *adapter;
2582         struct e1000_hw *hw;
2583         const struct igbvf_info *ei = igbvf_info_tbl[ent->driver_data];
2584
2585         static int cards_found;
2586         int err, pci_using_dac;
2587
2588         err = pci_enable_device_mem(pdev);
2589         if (err)
2590                 return err;
2591
2592         pci_using_dac = 0;
2593         err = dma_set_mask(&pdev->dev, DMA_BIT_MASK(64));
2594         if (!err) {
2595                 err = dma_set_coherent_mask(&pdev->dev, DMA_BIT_MASK(64));
2596                 if (!err)
2597                         pci_using_dac = 1;
2598         } else {
2599                 err = dma_set_mask(&pdev->dev, DMA_BIT_MASK(32));
2600                 if (err) {
2601                         err = dma_set_coherent_mask(&pdev->dev,
2602                                                     DMA_BIT_MASK(32));
2603                         if (err) {
2604                                 dev_err(&pdev->dev, "No usable DMA "
2605                                         "configuration, aborting\n");
2606                                 goto err_dma;
2607                         }
2608                 }
2609         }
2610
2611         err = pci_request_regions(pdev, igbvf_driver_name);
2612         if (err)
2613                 goto err_pci_reg;
2614
2615         pci_set_master(pdev);
2616
2617         err = -ENOMEM;
2618         netdev = alloc_etherdev(sizeof(struct igbvf_adapter));
2619         if (!netdev)
2620                 goto err_alloc_etherdev;
2621
2622         SET_NETDEV_DEV(netdev, &pdev->dev);
2623
2624         pci_set_drvdata(pdev, netdev);
2625         adapter = netdev_priv(netdev);
2626         hw = &adapter->hw;
2627         adapter->netdev = netdev;
2628         adapter->pdev = pdev;
2629         adapter->ei = ei;
2630         adapter->pba = ei->pba;
2631         adapter->flags = ei->flags;
2632         adapter->hw.back = adapter;
2633         adapter->hw.mac.type = ei->mac;
2634         adapter->msg_enable = (1 << NETIF_MSG_DRV | NETIF_MSG_PROBE) - 1;
2635
2636         /* PCI config space info */
2637
2638         hw->vendor_id = pdev->vendor;
2639         hw->device_id = pdev->device;
2640         hw->subsystem_vendor_id = pdev->subsystem_vendor;
2641         hw->subsystem_device_id = pdev->subsystem_device;
2642         hw->revision_id = pdev->revision;
2643
2644         err = -EIO;
2645         adapter->hw.hw_addr = ioremap(pci_resource_start(pdev, 0),
2646                                       pci_resource_len(pdev, 0));
2647
2648         if (!adapter->hw.hw_addr)
2649                 goto err_ioremap;
2650
2651         if (ei->get_variants) {
2652                 err = ei->get_variants(adapter);
2653                 if (err)
2654                         goto err_ioremap;
2655         }
2656
2657         /* setup adapter struct */
2658         err = igbvf_sw_init(adapter);
2659         if (err)
2660                 goto err_sw_init;
2661
2662         /* construct the net_device struct */
2663         netdev->netdev_ops = &igbvf_netdev_ops;
2664
2665         igbvf_set_ethtool_ops(netdev);
2666         netdev->watchdog_timeo = 5 * HZ;
2667         strncpy(netdev->name, pci_name(pdev), sizeof(netdev->name) - 1);
2668
2669         adapter->bd_number = cards_found++;
2670
2671         netdev->hw_features = NETIF_F_SG |
2672                            NETIF_F_IP_CSUM |
2673                            NETIF_F_IPV6_CSUM |
2674                            NETIF_F_TSO |
2675                            NETIF_F_TSO6 |
2676                            NETIF_F_RXCSUM;
2677
2678         netdev->features = netdev->hw_features |
2679                            NETIF_F_HW_VLAN_TX |
2680                            NETIF_F_HW_VLAN_RX |
2681                            NETIF_F_HW_VLAN_FILTER;
2682
2683         if (pci_using_dac)
2684                 netdev->features |= NETIF_F_HIGHDMA;
2685
2686         netdev->vlan_features |= NETIF_F_TSO;
2687         netdev->vlan_features |= NETIF_F_TSO6;
2688         netdev->vlan_features |= NETIF_F_IP_CSUM;
2689         netdev->vlan_features |= NETIF_F_IPV6_CSUM;
2690         netdev->vlan_features |= NETIF_F_SG;
2691
2692         /*reset the controller to put the device in a known good state */
2693         err = hw->mac.ops.reset_hw(hw);
2694         if (err) {
2695                 dev_info(&pdev->dev,
2696                          "PF still in reset state, assigning new address."
2697                          " Is the PF interface up?\n");
2698                 dev_hw_addr_random(adapter->netdev, hw->mac.addr);
2699         } else {
2700                 err = hw->mac.ops.read_mac_addr(hw);
2701                 if (err) {
2702                         dev_err(&pdev->dev, "Error reading MAC address\n");
2703                         goto err_hw_init;
2704                 }
2705         }
2706
2707         memcpy(netdev->dev_addr, adapter->hw.mac.addr, netdev->addr_len);
2708         memcpy(netdev->perm_addr, adapter->hw.mac.addr, netdev->addr_len);
2709
2710         if (!is_valid_ether_addr(netdev->perm_addr)) {
2711                 dev_err(&pdev->dev, "Invalid MAC Address: %pM\n",
2712                         netdev->dev_addr);
2713                 err = -EIO;
2714                 goto err_hw_init;
2715         }
2716
2717         setup_timer(&adapter->watchdog_timer, &igbvf_watchdog,
2718                     (unsigned long) adapter);
2719
2720         INIT_WORK(&adapter->reset_task, igbvf_reset_task);
2721         INIT_WORK(&adapter->watchdog_task, igbvf_watchdog_task);
2722
2723         /* ring size defaults */
2724         adapter->rx_ring->count = 1024;
2725         adapter->tx_ring->count = 1024;
2726
2727         /* reset the hardware with the new settings */
2728         igbvf_reset(adapter);
2729
2730         strcpy(netdev->name, "eth%d");
2731         err = register_netdev(netdev);
2732         if (err)
2733                 goto err_hw_init;
2734
2735         /* tell the stack to leave us alone until igbvf_open() is called */
2736         netif_carrier_off(netdev);
2737         netif_stop_queue(netdev);
2738
2739         igbvf_print_device_info(adapter);
2740
2741         igbvf_initialize_last_counter_stats(adapter);
2742
2743         return 0;
2744
2745 err_hw_init:
2746         kfree(adapter->tx_ring);
2747         kfree(adapter->rx_ring);
2748 err_sw_init:
2749         igbvf_reset_interrupt_capability(adapter);
2750         iounmap(adapter->hw.hw_addr);
2751 err_ioremap:
2752         free_netdev(netdev);
2753 err_alloc_etherdev:
2754         pci_release_regions(pdev);
2755 err_pci_reg:
2756 err_dma:
2757         pci_disable_device(pdev);
2758         return err;
2759 }
2760
2761 /**
2762  * igbvf_remove - Device Removal Routine
2763  * @pdev: PCI device information struct
2764  *
2765  * igbvf_remove is called by the PCI subsystem to alert the driver
2766  * that it should release a PCI device.  The could be caused by a
2767  * Hot-Plug event, or because the driver is going to be removed from
2768  * memory.
2769  **/
2770 static void __devexit igbvf_remove(struct pci_dev *pdev)
2771 {
2772         struct net_device *netdev = pci_get_drvdata(pdev);
2773         struct igbvf_adapter *adapter = netdev_priv(netdev);
2774         struct e1000_hw *hw = &adapter->hw;
2775
2776         /*
2777          * The watchdog timer may be rescheduled, so explicitly
2778          * disable it from being rescheduled.
2779          */
2780         set_bit(__IGBVF_DOWN, &adapter->state);
2781         del_timer_sync(&adapter->watchdog_timer);
2782
2783         cancel_work_sync(&adapter->reset_task);
2784         cancel_work_sync(&adapter->watchdog_task);
2785
2786         unregister_netdev(netdev);
2787
2788         igbvf_reset_interrupt_capability(adapter);
2789
2790         /*
2791          * it is important to delete the napi struct prior to freeing the
2792          * rx ring so that you do not end up with null pointer refs
2793          */
2794         netif_napi_del(&adapter->rx_ring->napi);
2795         kfree(adapter->tx_ring);
2796         kfree(adapter->rx_ring);
2797
2798         iounmap(hw->hw_addr);
2799         if (hw->flash_address)
2800                 iounmap(hw->flash_address);
2801         pci_release_regions(pdev);
2802
2803         free_netdev(netdev);
2804
2805         pci_disable_device(pdev);
2806 }
2807
2808 /* PCI Error Recovery (ERS) */
2809 static struct pci_error_handlers igbvf_err_handler = {
2810         .error_detected = igbvf_io_error_detected,
2811         .slot_reset = igbvf_io_slot_reset,
2812         .resume = igbvf_io_resume,
2813 };
2814
2815 static DEFINE_PCI_DEVICE_TABLE(igbvf_pci_tbl) = {
2816         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82576_VF), board_vf },
2817         { PCI_VDEVICE(INTEL, E1000_DEV_ID_I350_VF), board_i350_vf },
2818         { } /* terminate list */
2819 };
2820 MODULE_DEVICE_TABLE(pci, igbvf_pci_tbl);
2821
2822 /* PCI Device API Driver */
2823 static struct pci_driver igbvf_driver = {
2824         .name     = igbvf_driver_name,
2825         .id_table = igbvf_pci_tbl,
2826         .probe    = igbvf_probe,
2827         .remove   = __devexit_p(igbvf_remove),
2828 #ifdef CONFIG_PM
2829         /* Power Management Hooks */
2830         .suspend  = igbvf_suspend,
2831         .resume   = igbvf_resume,
2832 #endif
2833         .shutdown = igbvf_shutdown,
2834         .err_handler = &igbvf_err_handler
2835 };
2836
2837 /**
2838  * igbvf_init_module - Driver Registration Routine
2839  *
2840  * igbvf_init_module is the first routine called when the driver is
2841  * loaded. All it does is register with the PCI subsystem.
2842  **/
2843 static int __init igbvf_init_module(void)
2844 {
2845         int ret;
2846         pr_info("%s - version %s\n", igbvf_driver_string, igbvf_driver_version);
2847         pr_info("%s\n", igbvf_copyright);
2848
2849         ret = pci_register_driver(&igbvf_driver);
2850
2851         return ret;
2852 }
2853 module_init(igbvf_init_module);
2854
2855 /**
2856  * igbvf_exit_module - Driver Exit Cleanup Routine
2857  *
2858  * igbvf_exit_module is called just before the driver is removed
2859  * from memory.
2860  **/
2861 static void __exit igbvf_exit_module(void)
2862 {
2863         pci_unregister_driver(&igbvf_driver);
2864 }
2865 module_exit(igbvf_exit_module);
2866
2867
2868 MODULE_AUTHOR("Intel Corporation, <e1000-devel@lists.sourceforge.net>");
2869 MODULE_DESCRIPTION("Intel(R) Gigabit Virtual Function Network Driver");
2870 MODULE_LICENSE("GPL");
2871 MODULE_VERSION(DRV_VERSION);
2872
2873 /* netdev.c */