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         igbvf_irq_disable(adapter);
1198
1199         if (!test_bit(__IGBVF_DOWN, &adapter->state))
1200                 igbvf_irq_enable(adapter);
1201
1202         if (hw->mac.ops.set_vfta(hw, vid, false)) {
1203                 dev_err(&adapter->pdev->dev,
1204                         "Failed to remove vlan id %d\n", vid);
1205                 return -EINVAL;
1206         }
1207         clear_bit(vid, adapter->active_vlans);
1208         return 0;
1209 }
1210
1211 static void igbvf_restore_vlan(struct igbvf_adapter *adapter)
1212 {
1213         u16 vid;
1214
1215         for_each_set_bit(vid, adapter->active_vlans, VLAN_N_VID)
1216                 igbvf_vlan_rx_add_vid(adapter->netdev, vid);
1217 }
1218
1219 /**
1220  * igbvf_configure_tx - Configure Transmit Unit after Reset
1221  * @adapter: board private structure
1222  *
1223  * Configure the Tx unit of the MAC after a reset.
1224  **/
1225 static void igbvf_configure_tx(struct igbvf_adapter *adapter)
1226 {
1227         struct e1000_hw *hw = &adapter->hw;
1228         struct igbvf_ring *tx_ring = adapter->tx_ring;
1229         u64 tdba;
1230         u32 txdctl, dca_txctrl;
1231
1232         /* disable transmits */
1233         txdctl = er32(TXDCTL(0));
1234         ew32(TXDCTL(0), txdctl & ~E1000_TXDCTL_QUEUE_ENABLE);
1235         e1e_flush();
1236         msleep(10);
1237
1238         /* Setup the HW Tx Head and Tail descriptor pointers */
1239         ew32(TDLEN(0), tx_ring->count * sizeof(union e1000_adv_tx_desc));
1240         tdba = tx_ring->dma;
1241         ew32(TDBAL(0), (tdba & DMA_BIT_MASK(32)));
1242         ew32(TDBAH(0), (tdba >> 32));
1243         ew32(TDH(0), 0);
1244         ew32(TDT(0), 0);
1245         tx_ring->head = E1000_TDH(0);
1246         tx_ring->tail = E1000_TDT(0);
1247
1248         /* Turn off Relaxed Ordering on head write-backs.  The writebacks
1249          * MUST be delivered in order or it will completely screw up
1250          * our bookeeping.
1251          */
1252         dca_txctrl = er32(DCA_TXCTRL(0));
1253         dca_txctrl &= ~E1000_DCA_TXCTRL_TX_WB_RO_EN;
1254         ew32(DCA_TXCTRL(0), dca_txctrl);
1255
1256         /* enable transmits */
1257         txdctl |= E1000_TXDCTL_QUEUE_ENABLE;
1258         ew32(TXDCTL(0), txdctl);
1259
1260         /* Setup Transmit Descriptor Settings for eop descriptor */
1261         adapter->txd_cmd = E1000_ADVTXD_DCMD_EOP | E1000_ADVTXD_DCMD_IFCS;
1262
1263         /* enable Report Status bit */
1264         adapter->txd_cmd |= E1000_ADVTXD_DCMD_RS;
1265 }
1266
1267 /**
1268  * igbvf_setup_srrctl - configure the receive control registers
1269  * @adapter: Board private structure
1270  **/
1271 static void igbvf_setup_srrctl(struct igbvf_adapter *adapter)
1272 {
1273         struct e1000_hw *hw = &adapter->hw;
1274         u32 srrctl = 0;
1275
1276         srrctl &= ~(E1000_SRRCTL_DESCTYPE_MASK |
1277                     E1000_SRRCTL_BSIZEHDR_MASK |
1278                     E1000_SRRCTL_BSIZEPKT_MASK);
1279
1280         /* Enable queue drop to avoid head of line blocking */
1281         srrctl |= E1000_SRRCTL_DROP_EN;
1282
1283         /* Setup buffer sizes */
1284         srrctl |= ALIGN(adapter->rx_buffer_len, 1024) >>
1285                   E1000_SRRCTL_BSIZEPKT_SHIFT;
1286
1287         if (adapter->rx_buffer_len < 2048) {
1288                 adapter->rx_ps_hdr_size = 0;
1289                 srrctl |= E1000_SRRCTL_DESCTYPE_ADV_ONEBUF;
1290         } else {
1291                 adapter->rx_ps_hdr_size = 128;
1292                 srrctl |= adapter->rx_ps_hdr_size <<
1293                           E1000_SRRCTL_BSIZEHDRSIZE_SHIFT;
1294                 srrctl |= E1000_SRRCTL_DESCTYPE_HDR_SPLIT_ALWAYS;
1295         }
1296
1297         ew32(SRRCTL(0), srrctl);
1298 }
1299
1300 /**
1301  * igbvf_configure_rx - Configure Receive Unit after Reset
1302  * @adapter: board private structure
1303  *
1304  * Configure the Rx unit of the MAC after a reset.
1305  **/
1306 static void igbvf_configure_rx(struct igbvf_adapter *adapter)
1307 {
1308         struct e1000_hw *hw = &adapter->hw;
1309         struct igbvf_ring *rx_ring = adapter->rx_ring;
1310         u64 rdba;
1311         u32 rdlen, rxdctl;
1312
1313         /* disable receives */
1314         rxdctl = er32(RXDCTL(0));
1315         ew32(RXDCTL(0), rxdctl & ~E1000_RXDCTL_QUEUE_ENABLE);
1316         e1e_flush();
1317         msleep(10);
1318
1319         rdlen = rx_ring->count * sizeof(union e1000_adv_rx_desc);
1320
1321         /*
1322          * Setup the HW Rx Head and Tail Descriptor Pointers and
1323          * the Base and Length of the Rx Descriptor Ring
1324          */
1325         rdba = rx_ring->dma;
1326         ew32(RDBAL(0), (rdba & DMA_BIT_MASK(32)));
1327         ew32(RDBAH(0), (rdba >> 32));
1328         ew32(RDLEN(0), rx_ring->count * sizeof(union e1000_adv_rx_desc));
1329         rx_ring->head = E1000_RDH(0);
1330         rx_ring->tail = E1000_RDT(0);
1331         ew32(RDH(0), 0);
1332         ew32(RDT(0), 0);
1333
1334         rxdctl |= E1000_RXDCTL_QUEUE_ENABLE;
1335         rxdctl &= 0xFFF00000;
1336         rxdctl |= IGBVF_RX_PTHRESH;
1337         rxdctl |= IGBVF_RX_HTHRESH << 8;
1338         rxdctl |= IGBVF_RX_WTHRESH << 16;
1339
1340         igbvf_set_rlpml(adapter);
1341
1342         /* enable receives */
1343         ew32(RXDCTL(0), rxdctl);
1344 }
1345
1346 /**
1347  * igbvf_set_multi - Multicast and Promiscuous mode set
1348  * @netdev: network interface device structure
1349  *
1350  * The set_multi entry point is called whenever the multicast address
1351  * list or the network interface flags are updated.  This routine is
1352  * responsible for configuring the hardware for proper multicast,
1353  * promiscuous mode, and all-multi behavior.
1354  **/
1355 static void igbvf_set_multi(struct net_device *netdev)
1356 {
1357         struct igbvf_adapter *adapter = netdev_priv(netdev);
1358         struct e1000_hw *hw = &adapter->hw;
1359         struct netdev_hw_addr *ha;
1360         u8  *mta_list = NULL;
1361         int i;
1362
1363         if (!netdev_mc_empty(netdev)) {
1364                 mta_list = kmalloc(netdev_mc_count(netdev) * 6, GFP_ATOMIC);
1365                 if (!mta_list) {
1366                         dev_err(&adapter->pdev->dev,
1367                                 "failed to allocate multicast filter list\n");
1368                         return;
1369                 }
1370         }
1371
1372         /* prepare a packed array of only addresses. */
1373         i = 0;
1374         netdev_for_each_mc_addr(ha, netdev)
1375                 memcpy(mta_list + (i++ * ETH_ALEN), ha->addr, ETH_ALEN);
1376
1377         hw->mac.ops.update_mc_addr_list(hw, mta_list, i, 0, 0);
1378         kfree(mta_list);
1379 }
1380
1381 /**
1382  * igbvf_configure - configure the hardware for Rx and Tx
1383  * @adapter: private board structure
1384  **/
1385 static void igbvf_configure(struct igbvf_adapter *adapter)
1386 {
1387         igbvf_set_multi(adapter->netdev);
1388
1389         igbvf_restore_vlan(adapter);
1390
1391         igbvf_configure_tx(adapter);
1392         igbvf_setup_srrctl(adapter);
1393         igbvf_configure_rx(adapter);
1394         igbvf_alloc_rx_buffers(adapter->rx_ring,
1395                                igbvf_desc_unused(adapter->rx_ring));
1396 }
1397
1398 /* igbvf_reset - bring the hardware into a known good state
1399  *
1400  * This function boots the hardware and enables some settings that
1401  * require a configuration cycle of the hardware - those cannot be
1402  * set/changed during runtime. After reset the device needs to be
1403  * properly configured for Rx, Tx etc.
1404  */
1405 static void igbvf_reset(struct igbvf_adapter *adapter)
1406 {
1407         struct e1000_mac_info *mac = &adapter->hw.mac;
1408         struct net_device *netdev = adapter->netdev;
1409         struct e1000_hw *hw = &adapter->hw;
1410
1411         /* Allow time for pending master requests to run */
1412         if (mac->ops.reset_hw(hw))
1413                 dev_err(&adapter->pdev->dev, "PF still resetting\n");
1414
1415         mac->ops.init_hw(hw);
1416
1417         if (is_valid_ether_addr(adapter->hw.mac.addr)) {
1418                 memcpy(netdev->dev_addr, adapter->hw.mac.addr,
1419                        netdev->addr_len);
1420                 memcpy(netdev->perm_addr, adapter->hw.mac.addr,
1421                        netdev->addr_len);
1422         }
1423
1424         adapter->last_reset = jiffies;
1425 }
1426
1427 int igbvf_up(struct igbvf_adapter *adapter)
1428 {
1429         struct e1000_hw *hw = &adapter->hw;
1430
1431         /* hardware has been reset, we need to reload some things */
1432         igbvf_configure(adapter);
1433
1434         clear_bit(__IGBVF_DOWN, &adapter->state);
1435
1436         napi_enable(&adapter->rx_ring->napi);
1437         if (adapter->msix_entries)
1438                 igbvf_configure_msix(adapter);
1439
1440         /* Clear any pending interrupts. */
1441         er32(EICR);
1442         igbvf_irq_enable(adapter);
1443
1444         /* start the watchdog */
1445         hw->mac.get_link_status = 1;
1446         mod_timer(&adapter->watchdog_timer, jiffies + 1);
1447
1448
1449         return 0;
1450 }
1451
1452 void igbvf_down(struct igbvf_adapter *adapter)
1453 {
1454         struct net_device *netdev = adapter->netdev;
1455         struct e1000_hw *hw = &adapter->hw;
1456         u32 rxdctl, txdctl;
1457
1458         /*
1459          * signal that we're down so the interrupt handler does not
1460          * reschedule our watchdog timer
1461          */
1462         set_bit(__IGBVF_DOWN, &adapter->state);
1463
1464         /* disable receives in the hardware */
1465         rxdctl = er32(RXDCTL(0));
1466         ew32(RXDCTL(0), rxdctl & ~E1000_RXDCTL_QUEUE_ENABLE);
1467
1468         netif_stop_queue(netdev);
1469
1470         /* disable transmits in the hardware */
1471         txdctl = er32(TXDCTL(0));
1472         ew32(TXDCTL(0), txdctl & ~E1000_TXDCTL_QUEUE_ENABLE);
1473
1474         /* flush both disables and wait for them to finish */
1475         e1e_flush();
1476         msleep(10);
1477
1478         napi_disable(&adapter->rx_ring->napi);
1479
1480         igbvf_irq_disable(adapter);
1481
1482         del_timer_sync(&adapter->watchdog_timer);
1483
1484         netif_carrier_off(netdev);
1485
1486         /* record the stats before reset*/
1487         igbvf_update_stats(adapter);
1488
1489         adapter->link_speed = 0;
1490         adapter->link_duplex = 0;
1491
1492         igbvf_reset(adapter);
1493         igbvf_clean_tx_ring(adapter->tx_ring);
1494         igbvf_clean_rx_ring(adapter->rx_ring);
1495 }
1496
1497 void igbvf_reinit_locked(struct igbvf_adapter *adapter)
1498 {
1499         might_sleep();
1500         while (test_and_set_bit(__IGBVF_RESETTING, &adapter->state))
1501                 msleep(1);
1502         igbvf_down(adapter);
1503         igbvf_up(adapter);
1504         clear_bit(__IGBVF_RESETTING, &adapter->state);
1505 }
1506
1507 /**
1508  * igbvf_sw_init - Initialize general software structures (struct igbvf_adapter)
1509  * @adapter: board private structure to initialize
1510  *
1511  * igbvf_sw_init initializes the Adapter private data structure.
1512  * Fields are initialized based on PCI device information and
1513  * OS network device settings (MTU size).
1514  **/
1515 static int __devinit igbvf_sw_init(struct igbvf_adapter *adapter)
1516 {
1517         struct net_device *netdev = adapter->netdev;
1518         s32 rc;
1519
1520         adapter->rx_buffer_len = ETH_FRAME_LEN + VLAN_HLEN + ETH_FCS_LEN;
1521         adapter->rx_ps_hdr_size = 0;
1522         adapter->max_frame_size = netdev->mtu + ETH_HLEN + ETH_FCS_LEN;
1523         adapter->min_frame_size = ETH_ZLEN + ETH_FCS_LEN;
1524
1525         adapter->tx_int_delay = 8;
1526         adapter->tx_abs_int_delay = 32;
1527         adapter->rx_int_delay = 0;
1528         adapter->rx_abs_int_delay = 8;
1529         adapter->itr_setting = 3;
1530         adapter->itr = 20000;
1531
1532         /* Set various function pointers */
1533         adapter->ei->init_ops(&adapter->hw);
1534
1535         rc = adapter->hw.mac.ops.init_params(&adapter->hw);
1536         if (rc)
1537                 return rc;
1538
1539         rc = adapter->hw.mbx.ops.init_params(&adapter->hw);
1540         if (rc)
1541                 return rc;
1542
1543         igbvf_set_interrupt_capability(adapter);
1544
1545         if (igbvf_alloc_queues(adapter))
1546                 return -ENOMEM;
1547
1548         spin_lock_init(&adapter->tx_queue_lock);
1549
1550         /* Explicitly disable IRQ since the NIC can be in any state. */
1551         igbvf_irq_disable(adapter);
1552
1553         spin_lock_init(&adapter->stats_lock);
1554
1555         set_bit(__IGBVF_DOWN, &adapter->state);
1556         return 0;
1557 }
1558
1559 static void igbvf_initialize_last_counter_stats(struct igbvf_adapter *adapter)
1560 {
1561         struct e1000_hw *hw = &adapter->hw;
1562
1563         adapter->stats.last_gprc = er32(VFGPRC);
1564         adapter->stats.last_gorc = er32(VFGORC);
1565         adapter->stats.last_gptc = er32(VFGPTC);
1566         adapter->stats.last_gotc = er32(VFGOTC);
1567         adapter->stats.last_mprc = er32(VFMPRC);
1568         adapter->stats.last_gotlbc = er32(VFGOTLBC);
1569         adapter->stats.last_gptlbc = er32(VFGPTLBC);
1570         adapter->stats.last_gorlbc = er32(VFGORLBC);
1571         adapter->stats.last_gprlbc = er32(VFGPRLBC);
1572
1573         adapter->stats.base_gprc = er32(VFGPRC);
1574         adapter->stats.base_gorc = er32(VFGORC);
1575         adapter->stats.base_gptc = er32(VFGPTC);
1576         adapter->stats.base_gotc = er32(VFGOTC);
1577         adapter->stats.base_mprc = er32(VFMPRC);
1578         adapter->stats.base_gotlbc = er32(VFGOTLBC);
1579         adapter->stats.base_gptlbc = er32(VFGPTLBC);
1580         adapter->stats.base_gorlbc = er32(VFGORLBC);
1581         adapter->stats.base_gprlbc = er32(VFGPRLBC);
1582 }
1583
1584 /**
1585  * igbvf_open - Called when a network interface is made active
1586  * @netdev: network interface device structure
1587  *
1588  * Returns 0 on success, negative value on failure
1589  *
1590  * The open entry point is called when a network interface is made
1591  * active by the system (IFF_UP).  At this point all resources needed
1592  * for transmit and receive operations are allocated, the interrupt
1593  * handler is registered with the OS, the watchdog timer is started,
1594  * and the stack is notified that the interface is ready.
1595  **/
1596 static int igbvf_open(struct net_device *netdev)
1597 {
1598         struct igbvf_adapter *adapter = netdev_priv(netdev);
1599         struct e1000_hw *hw = &adapter->hw;
1600         int err;
1601
1602         /* disallow open during test */
1603         if (test_bit(__IGBVF_TESTING, &adapter->state))
1604                 return -EBUSY;
1605
1606         /* allocate transmit descriptors */
1607         err = igbvf_setup_tx_resources(adapter, adapter->tx_ring);
1608         if (err)
1609                 goto err_setup_tx;
1610
1611         /* allocate receive descriptors */
1612         err = igbvf_setup_rx_resources(adapter, adapter->rx_ring);
1613         if (err)
1614                 goto err_setup_rx;
1615
1616         /*
1617          * before we allocate an interrupt, we must be ready to handle it.
1618          * Setting DEBUG_SHIRQ in the kernel makes it fire an interrupt
1619          * as soon as we call pci_request_irq, so we have to setup our
1620          * clean_rx handler before we do so.
1621          */
1622         igbvf_configure(adapter);
1623
1624         err = igbvf_request_irq(adapter);
1625         if (err)
1626                 goto err_req_irq;
1627
1628         /* From here on the code is the same as igbvf_up() */
1629         clear_bit(__IGBVF_DOWN, &adapter->state);
1630
1631         napi_enable(&adapter->rx_ring->napi);
1632
1633         /* clear any pending interrupts */
1634         er32(EICR);
1635
1636         igbvf_irq_enable(adapter);
1637
1638         /* start the watchdog */
1639         hw->mac.get_link_status = 1;
1640         mod_timer(&adapter->watchdog_timer, jiffies + 1);
1641
1642         return 0;
1643
1644 err_req_irq:
1645         igbvf_free_rx_resources(adapter->rx_ring);
1646 err_setup_rx:
1647         igbvf_free_tx_resources(adapter->tx_ring);
1648 err_setup_tx:
1649         igbvf_reset(adapter);
1650
1651         return err;
1652 }
1653
1654 /**
1655  * igbvf_close - Disables a network interface
1656  * @netdev: network interface device structure
1657  *
1658  * Returns 0, this is not allowed to fail
1659  *
1660  * The close entry point is called when an interface is de-activated
1661  * by the OS.  The hardware is still under the drivers control, but
1662  * needs to be disabled.  A global MAC reset is issued to stop the
1663  * hardware, and all transmit and receive resources are freed.
1664  **/
1665 static int igbvf_close(struct net_device *netdev)
1666 {
1667         struct igbvf_adapter *adapter = netdev_priv(netdev);
1668
1669         WARN_ON(test_bit(__IGBVF_RESETTING, &adapter->state));
1670         igbvf_down(adapter);
1671
1672         igbvf_free_irq(adapter);
1673
1674         igbvf_free_tx_resources(adapter->tx_ring);
1675         igbvf_free_rx_resources(adapter->rx_ring);
1676
1677         return 0;
1678 }
1679 /**
1680  * igbvf_set_mac - Change the Ethernet Address of the NIC
1681  * @netdev: network interface device structure
1682  * @p: pointer to an address structure
1683  *
1684  * Returns 0 on success, negative on failure
1685  **/
1686 static int igbvf_set_mac(struct net_device *netdev, void *p)
1687 {
1688         struct igbvf_adapter *adapter = netdev_priv(netdev);
1689         struct e1000_hw *hw = &adapter->hw;
1690         struct sockaddr *addr = p;
1691
1692         if (!is_valid_ether_addr(addr->sa_data))
1693                 return -EADDRNOTAVAIL;
1694
1695         memcpy(hw->mac.addr, addr->sa_data, netdev->addr_len);
1696
1697         hw->mac.ops.rar_set(hw, hw->mac.addr, 0);
1698
1699         if (memcmp(addr->sa_data, hw->mac.addr, 6))
1700                 return -EADDRNOTAVAIL;
1701
1702         memcpy(netdev->dev_addr, addr->sa_data, netdev->addr_len);
1703
1704         return 0;
1705 }
1706
1707 #define UPDATE_VF_COUNTER(reg, name)                                    \
1708         {                                                               \
1709                 u32 current_counter = er32(reg);                        \
1710                 if (current_counter < adapter->stats.last_##name)       \
1711                         adapter->stats.name += 0x100000000LL;           \
1712                 adapter->stats.last_##name = current_counter;           \
1713                 adapter->stats.name &= 0xFFFFFFFF00000000LL;            \
1714                 adapter->stats.name |= current_counter;                 \
1715         }
1716
1717 /**
1718  * igbvf_update_stats - Update the board statistics counters
1719  * @adapter: board private structure
1720 **/
1721 void igbvf_update_stats(struct igbvf_adapter *adapter)
1722 {
1723         struct e1000_hw *hw = &adapter->hw;
1724         struct pci_dev *pdev = adapter->pdev;
1725
1726         /*
1727          * Prevent stats update while adapter is being reset, link is down
1728          * or if the pci connection is down.
1729          */
1730         if (adapter->link_speed == 0)
1731                 return;
1732
1733         if (test_bit(__IGBVF_RESETTING, &adapter->state))
1734                 return;
1735
1736         if (pci_channel_offline(pdev))
1737                 return;
1738
1739         UPDATE_VF_COUNTER(VFGPRC, gprc);
1740         UPDATE_VF_COUNTER(VFGORC, gorc);
1741         UPDATE_VF_COUNTER(VFGPTC, gptc);
1742         UPDATE_VF_COUNTER(VFGOTC, gotc);
1743         UPDATE_VF_COUNTER(VFMPRC, mprc);
1744         UPDATE_VF_COUNTER(VFGOTLBC, gotlbc);
1745         UPDATE_VF_COUNTER(VFGPTLBC, gptlbc);
1746         UPDATE_VF_COUNTER(VFGORLBC, gorlbc);
1747         UPDATE_VF_COUNTER(VFGPRLBC, gprlbc);
1748
1749         /* Fill out the OS statistics structure */
1750         adapter->net_stats.multicast = adapter->stats.mprc;
1751 }
1752
1753 static void igbvf_print_link_info(struct igbvf_adapter *adapter)
1754 {
1755         dev_info(&adapter->pdev->dev, "Link is Up %d Mbps %s Duplex\n",
1756                  adapter->link_speed,
1757                  adapter->link_duplex == FULL_DUPLEX ? "Full" : "Half");
1758 }
1759
1760 static bool igbvf_has_link(struct igbvf_adapter *adapter)
1761 {
1762         struct e1000_hw *hw = &adapter->hw;
1763         s32 ret_val = E1000_SUCCESS;
1764         bool link_active;
1765
1766         /* If interface is down, stay link down */
1767         if (test_bit(__IGBVF_DOWN, &adapter->state))
1768                 return false;
1769
1770         ret_val = hw->mac.ops.check_for_link(hw);
1771         link_active = !hw->mac.get_link_status;
1772
1773         /* if check for link returns error we will need to reset */
1774         if (ret_val && time_after(jiffies, adapter->last_reset + (10 * HZ)))
1775                 schedule_work(&adapter->reset_task);
1776
1777         return link_active;
1778 }
1779
1780 /**
1781  * igbvf_watchdog - Timer Call-back
1782  * @data: pointer to adapter cast into an unsigned long
1783  **/
1784 static void igbvf_watchdog(unsigned long data)
1785 {
1786         struct igbvf_adapter *adapter = (struct igbvf_adapter *) data;
1787
1788         /* Do the rest outside of interrupt context */
1789         schedule_work(&adapter->watchdog_task);
1790 }
1791
1792 static void igbvf_watchdog_task(struct work_struct *work)
1793 {
1794         struct igbvf_adapter *adapter = container_of(work,
1795                                                      struct igbvf_adapter,
1796                                                      watchdog_task);
1797         struct net_device *netdev = adapter->netdev;
1798         struct e1000_mac_info *mac = &adapter->hw.mac;
1799         struct igbvf_ring *tx_ring = adapter->tx_ring;
1800         struct e1000_hw *hw = &adapter->hw;
1801         u32 link;
1802         int tx_pending = 0;
1803
1804         link = igbvf_has_link(adapter);
1805
1806         if (link) {
1807                 if (!netif_carrier_ok(netdev)) {
1808                         mac->ops.get_link_up_info(&adapter->hw,
1809                                                   &adapter->link_speed,
1810                                                   &adapter->link_duplex);
1811                         igbvf_print_link_info(adapter);
1812
1813                         netif_carrier_on(netdev);
1814                         netif_wake_queue(netdev);
1815                 }
1816         } else {
1817                 if (netif_carrier_ok(netdev)) {
1818                         adapter->link_speed = 0;
1819                         adapter->link_duplex = 0;
1820                         dev_info(&adapter->pdev->dev, "Link is Down\n");
1821                         netif_carrier_off(netdev);
1822                         netif_stop_queue(netdev);
1823                 }
1824         }
1825
1826         if (netif_carrier_ok(netdev)) {
1827                 igbvf_update_stats(adapter);
1828         } else {
1829                 tx_pending = (igbvf_desc_unused(tx_ring) + 1 <
1830                               tx_ring->count);
1831                 if (tx_pending) {
1832                         /*
1833                          * We've lost link, so the controller stops DMA,
1834                          * but we've got queued Tx work that's never going
1835                          * to get done, so reset controller to flush Tx.
1836                          * (Do the reset outside of interrupt context).
1837                          */
1838                         adapter->tx_timeout_count++;
1839                         schedule_work(&adapter->reset_task);
1840                 }
1841         }
1842
1843         /* Cause software interrupt to ensure Rx ring is cleaned */
1844         ew32(EICS, adapter->rx_ring->eims_value);
1845
1846         /* Reset the timer */
1847         if (!test_bit(__IGBVF_DOWN, &adapter->state))
1848                 mod_timer(&adapter->watchdog_timer,
1849                           round_jiffies(jiffies + (2 * HZ)));
1850 }
1851
1852 #define IGBVF_TX_FLAGS_CSUM             0x00000001
1853 #define IGBVF_TX_FLAGS_VLAN             0x00000002
1854 #define IGBVF_TX_FLAGS_TSO              0x00000004
1855 #define IGBVF_TX_FLAGS_IPV4             0x00000008
1856 #define IGBVF_TX_FLAGS_VLAN_MASK        0xffff0000
1857 #define IGBVF_TX_FLAGS_VLAN_SHIFT       16
1858
1859 static int igbvf_tso(struct igbvf_adapter *adapter,
1860                      struct igbvf_ring *tx_ring,
1861                      struct sk_buff *skb, u32 tx_flags, u8 *hdr_len)
1862 {
1863         struct e1000_adv_tx_context_desc *context_desc;
1864         unsigned int i;
1865         int err;
1866         struct igbvf_buffer *buffer_info;
1867         u32 info = 0, tu_cmd = 0;
1868         u32 mss_l4len_idx, l4len;
1869         *hdr_len = 0;
1870
1871         if (skb_header_cloned(skb)) {
1872                 err = pskb_expand_head(skb, 0, 0, GFP_ATOMIC);
1873                 if (err) {
1874                         dev_err(&adapter->pdev->dev,
1875                                 "igbvf_tso returning an error\n");
1876                         return err;
1877                 }
1878         }
1879
1880         l4len = tcp_hdrlen(skb);
1881         *hdr_len += l4len;
1882
1883         if (skb->protocol == htons(ETH_P_IP)) {
1884                 struct iphdr *iph = ip_hdr(skb);
1885                 iph->tot_len = 0;
1886                 iph->check = 0;
1887                 tcp_hdr(skb)->check = ~csum_tcpudp_magic(iph->saddr,
1888                                                          iph->daddr, 0,
1889                                                          IPPROTO_TCP,
1890                                                          0);
1891         } else if (skb_is_gso_v6(skb)) {
1892                 ipv6_hdr(skb)->payload_len = 0;
1893                 tcp_hdr(skb)->check = ~csum_ipv6_magic(&ipv6_hdr(skb)->saddr,
1894                                                        &ipv6_hdr(skb)->daddr,
1895                                                        0, IPPROTO_TCP, 0);
1896         }
1897
1898         i = tx_ring->next_to_use;
1899
1900         buffer_info = &tx_ring->buffer_info[i];
1901         context_desc = IGBVF_TX_CTXTDESC_ADV(*tx_ring, i);
1902         /* VLAN MACLEN IPLEN */
1903         if (tx_flags & IGBVF_TX_FLAGS_VLAN)
1904                 info |= (tx_flags & IGBVF_TX_FLAGS_VLAN_MASK);
1905         info |= (skb_network_offset(skb) << E1000_ADVTXD_MACLEN_SHIFT);
1906         *hdr_len += skb_network_offset(skb);
1907         info |= (skb_transport_header(skb) - skb_network_header(skb));
1908         *hdr_len += (skb_transport_header(skb) - skb_network_header(skb));
1909         context_desc->vlan_macip_lens = cpu_to_le32(info);
1910
1911         /* ADV DTYP TUCMD MKRLOC/ISCSIHEDLEN */
1912         tu_cmd |= (E1000_TXD_CMD_DEXT | E1000_ADVTXD_DTYP_CTXT);
1913
1914         if (skb->protocol == htons(ETH_P_IP))
1915                 tu_cmd |= E1000_ADVTXD_TUCMD_IPV4;
1916         tu_cmd |= E1000_ADVTXD_TUCMD_L4T_TCP;
1917
1918         context_desc->type_tucmd_mlhl = cpu_to_le32(tu_cmd);
1919
1920         /* MSS L4LEN IDX */
1921         mss_l4len_idx = (skb_shinfo(skb)->gso_size << E1000_ADVTXD_MSS_SHIFT);
1922         mss_l4len_idx |= (l4len << E1000_ADVTXD_L4LEN_SHIFT);
1923
1924         context_desc->mss_l4len_idx = cpu_to_le32(mss_l4len_idx);
1925         context_desc->seqnum_seed = 0;
1926
1927         buffer_info->time_stamp = jiffies;
1928         buffer_info->next_to_watch = i;
1929         buffer_info->dma = 0;
1930         i++;
1931         if (i == tx_ring->count)
1932                 i = 0;
1933
1934         tx_ring->next_to_use = i;
1935
1936         return true;
1937 }
1938
1939 static inline bool igbvf_tx_csum(struct igbvf_adapter *adapter,
1940                                  struct igbvf_ring *tx_ring,
1941                                  struct sk_buff *skb, u32 tx_flags)
1942 {
1943         struct e1000_adv_tx_context_desc *context_desc;
1944         unsigned int i;
1945         struct igbvf_buffer *buffer_info;
1946         u32 info = 0, tu_cmd = 0;
1947
1948         if ((skb->ip_summed == CHECKSUM_PARTIAL) ||
1949             (tx_flags & IGBVF_TX_FLAGS_VLAN)) {
1950                 i = tx_ring->next_to_use;
1951                 buffer_info = &tx_ring->buffer_info[i];
1952                 context_desc = IGBVF_TX_CTXTDESC_ADV(*tx_ring, i);
1953
1954                 if (tx_flags & IGBVF_TX_FLAGS_VLAN)
1955                         info |= (tx_flags & IGBVF_TX_FLAGS_VLAN_MASK);
1956
1957                 info |= (skb_network_offset(skb) << E1000_ADVTXD_MACLEN_SHIFT);
1958                 if (skb->ip_summed == CHECKSUM_PARTIAL)
1959                         info |= (skb_transport_header(skb) -
1960                                  skb_network_header(skb));
1961
1962
1963                 context_desc->vlan_macip_lens = cpu_to_le32(info);
1964
1965                 tu_cmd |= (E1000_TXD_CMD_DEXT | E1000_ADVTXD_DTYP_CTXT);
1966
1967                 if (skb->ip_summed == CHECKSUM_PARTIAL) {
1968                         switch (skb->protocol) {
1969                         case __constant_htons(ETH_P_IP):
1970                                 tu_cmd |= E1000_ADVTXD_TUCMD_IPV4;
1971                                 if (ip_hdr(skb)->protocol == IPPROTO_TCP)
1972                                         tu_cmd |= E1000_ADVTXD_TUCMD_L4T_TCP;
1973                                 break;
1974                         case __constant_htons(ETH_P_IPV6):
1975                                 if (ipv6_hdr(skb)->nexthdr == IPPROTO_TCP)
1976                                         tu_cmd |= E1000_ADVTXD_TUCMD_L4T_TCP;
1977                                 break;
1978                         default:
1979                                 break;
1980                         }
1981                 }
1982
1983                 context_desc->type_tucmd_mlhl = cpu_to_le32(tu_cmd);
1984                 context_desc->seqnum_seed = 0;
1985                 context_desc->mss_l4len_idx = 0;
1986
1987                 buffer_info->time_stamp = jiffies;
1988                 buffer_info->next_to_watch = i;
1989                 buffer_info->dma = 0;
1990                 i++;
1991                 if (i == tx_ring->count)
1992                         i = 0;
1993                 tx_ring->next_to_use = i;
1994
1995                 return true;
1996         }
1997
1998         return false;
1999 }
2000
2001 static int igbvf_maybe_stop_tx(struct net_device *netdev, int size)
2002 {
2003         struct igbvf_adapter *adapter = netdev_priv(netdev);
2004
2005         /* there is enough descriptors then we don't need to worry  */
2006         if (igbvf_desc_unused(adapter->tx_ring) >= size)
2007                 return 0;
2008
2009         netif_stop_queue(netdev);
2010
2011         smp_mb();
2012
2013         /* We need to check again just in case room has been made available */
2014         if (igbvf_desc_unused(adapter->tx_ring) < size)
2015                 return -EBUSY;
2016
2017         netif_wake_queue(netdev);
2018
2019         ++adapter->restart_queue;
2020         return 0;
2021 }
2022
2023 #define IGBVF_MAX_TXD_PWR       16
2024 #define IGBVF_MAX_DATA_PER_TXD  (1 << IGBVF_MAX_TXD_PWR)
2025
2026 static inline int igbvf_tx_map_adv(struct igbvf_adapter *adapter,
2027                                    struct igbvf_ring *tx_ring,
2028                                    struct sk_buff *skb,
2029                                    unsigned int first)
2030 {
2031         struct igbvf_buffer *buffer_info;
2032         struct pci_dev *pdev = adapter->pdev;
2033         unsigned int len = skb_headlen(skb);
2034         unsigned int count = 0, i;
2035         unsigned int f;
2036
2037         i = tx_ring->next_to_use;
2038
2039         buffer_info = &tx_ring->buffer_info[i];
2040         BUG_ON(len >= IGBVF_MAX_DATA_PER_TXD);
2041         buffer_info->length = len;
2042         /* set time_stamp *before* dma to help avoid a possible race */
2043         buffer_info->time_stamp = jiffies;
2044         buffer_info->next_to_watch = i;
2045         buffer_info->mapped_as_page = false;
2046         buffer_info->dma = dma_map_single(&pdev->dev, skb->data, len,
2047                                           DMA_TO_DEVICE);
2048         if (dma_mapping_error(&pdev->dev, buffer_info->dma))
2049                 goto dma_error;
2050
2051
2052         for (f = 0; f < skb_shinfo(skb)->nr_frags; f++) {
2053                 const struct skb_frag_struct *frag;
2054
2055                 count++;
2056                 i++;
2057                 if (i == tx_ring->count)
2058                         i = 0;
2059
2060                 frag = &skb_shinfo(skb)->frags[f];
2061                 len = skb_frag_size(frag);
2062
2063                 buffer_info = &tx_ring->buffer_info[i];
2064                 BUG_ON(len >= IGBVF_MAX_DATA_PER_TXD);
2065                 buffer_info->length = len;
2066                 buffer_info->time_stamp = jiffies;
2067                 buffer_info->next_to_watch = i;
2068                 buffer_info->mapped_as_page = true;
2069                 buffer_info->dma = skb_frag_dma_map(&pdev->dev, frag, 0, len,
2070                                                 DMA_TO_DEVICE);
2071                 if (dma_mapping_error(&pdev->dev, buffer_info->dma))
2072                         goto dma_error;
2073         }
2074
2075         tx_ring->buffer_info[i].skb = skb;
2076         tx_ring->buffer_info[first].next_to_watch = i;
2077
2078         return ++count;
2079
2080 dma_error:
2081         dev_err(&pdev->dev, "TX DMA map failed\n");
2082
2083         /* clear timestamp and dma mappings for failed buffer_info mapping */
2084         buffer_info->dma = 0;
2085         buffer_info->time_stamp = 0;
2086         buffer_info->length = 0;
2087         buffer_info->next_to_watch = 0;
2088         buffer_info->mapped_as_page = false;
2089         if (count)
2090                 count--;
2091
2092         /* clear timestamp and dma mappings for remaining portion of packet */
2093         while (count--) {
2094                 if (i==0)
2095                         i += tx_ring->count;
2096                 i--;
2097                 buffer_info = &tx_ring->buffer_info[i];
2098                 igbvf_put_txbuf(adapter, buffer_info);
2099         }
2100
2101         return 0;
2102 }
2103
2104 static inline void igbvf_tx_queue_adv(struct igbvf_adapter *adapter,
2105                                       struct igbvf_ring *tx_ring,
2106                                       int tx_flags, int count, u32 paylen,
2107                                       u8 hdr_len)
2108 {
2109         union e1000_adv_tx_desc *tx_desc = NULL;
2110         struct igbvf_buffer *buffer_info;
2111         u32 olinfo_status = 0, cmd_type_len;
2112         unsigned int i;
2113
2114         cmd_type_len = (E1000_ADVTXD_DTYP_DATA | E1000_ADVTXD_DCMD_IFCS |
2115                         E1000_ADVTXD_DCMD_DEXT);
2116
2117         if (tx_flags & IGBVF_TX_FLAGS_VLAN)
2118                 cmd_type_len |= E1000_ADVTXD_DCMD_VLE;
2119
2120         if (tx_flags & IGBVF_TX_FLAGS_TSO) {
2121                 cmd_type_len |= E1000_ADVTXD_DCMD_TSE;
2122
2123                 /* insert tcp checksum */
2124                 olinfo_status |= E1000_TXD_POPTS_TXSM << 8;
2125
2126                 /* insert ip checksum */
2127                 if (tx_flags & IGBVF_TX_FLAGS_IPV4)
2128                         olinfo_status |= E1000_TXD_POPTS_IXSM << 8;
2129
2130         } else if (tx_flags & IGBVF_TX_FLAGS_CSUM) {
2131                 olinfo_status |= E1000_TXD_POPTS_TXSM << 8;
2132         }
2133
2134         olinfo_status |= ((paylen - hdr_len) << E1000_ADVTXD_PAYLEN_SHIFT);
2135
2136         i = tx_ring->next_to_use;
2137         while (count--) {
2138                 buffer_info = &tx_ring->buffer_info[i];
2139                 tx_desc = IGBVF_TX_DESC_ADV(*tx_ring, i);
2140                 tx_desc->read.buffer_addr = cpu_to_le64(buffer_info->dma);
2141                 tx_desc->read.cmd_type_len =
2142                          cpu_to_le32(cmd_type_len | buffer_info->length);
2143                 tx_desc->read.olinfo_status = cpu_to_le32(olinfo_status);
2144                 i++;
2145                 if (i == tx_ring->count)
2146                         i = 0;
2147         }
2148
2149         tx_desc->read.cmd_type_len |= cpu_to_le32(adapter->txd_cmd);
2150         /* Force memory writes to complete before letting h/w
2151          * know there are new descriptors to fetch.  (Only
2152          * applicable for weak-ordered memory model archs,
2153          * such as IA-64). */
2154         wmb();
2155
2156         tx_ring->next_to_use = i;
2157         writel(i, adapter->hw.hw_addr + tx_ring->tail);
2158         /* we need this if more than one processor can write to our tail
2159          * at a time, it syncronizes IO on IA64/Altix systems */
2160         mmiowb();
2161 }
2162
2163 static netdev_tx_t igbvf_xmit_frame_ring_adv(struct sk_buff *skb,
2164                                              struct net_device *netdev,
2165                                              struct igbvf_ring *tx_ring)
2166 {
2167         struct igbvf_adapter *adapter = netdev_priv(netdev);
2168         unsigned int first, tx_flags = 0;
2169         u8 hdr_len = 0;
2170         int count = 0;
2171         int tso = 0;
2172
2173         if (test_bit(__IGBVF_DOWN, &adapter->state)) {
2174                 dev_kfree_skb_any(skb);
2175                 return NETDEV_TX_OK;
2176         }
2177
2178         if (skb->len <= 0) {
2179                 dev_kfree_skb_any(skb);
2180                 return NETDEV_TX_OK;
2181         }
2182
2183         /*
2184          * need: count + 4 desc gap to keep tail from touching
2185          *       + 2 desc gap to keep tail from touching head,
2186          *       + 1 desc for skb->data,
2187          *       + 1 desc for context descriptor,
2188          * head, otherwise try next time
2189          */
2190         if (igbvf_maybe_stop_tx(netdev, skb_shinfo(skb)->nr_frags + 4)) {
2191                 /* this is a hard error */
2192                 return NETDEV_TX_BUSY;
2193         }
2194
2195         if (vlan_tx_tag_present(skb)) {
2196                 tx_flags |= IGBVF_TX_FLAGS_VLAN;
2197                 tx_flags |= (vlan_tx_tag_get(skb) << IGBVF_TX_FLAGS_VLAN_SHIFT);
2198         }
2199
2200         if (skb->protocol == htons(ETH_P_IP))
2201                 tx_flags |= IGBVF_TX_FLAGS_IPV4;
2202
2203         first = tx_ring->next_to_use;
2204
2205         tso = skb_is_gso(skb) ?
2206                 igbvf_tso(adapter, tx_ring, skb, tx_flags, &hdr_len) : 0;
2207         if (unlikely(tso < 0)) {
2208                 dev_kfree_skb_any(skb);
2209                 return NETDEV_TX_OK;
2210         }
2211
2212         if (tso)
2213                 tx_flags |= IGBVF_TX_FLAGS_TSO;
2214         else if (igbvf_tx_csum(adapter, tx_ring, skb, tx_flags) &&
2215                  (skb->ip_summed == CHECKSUM_PARTIAL))
2216                 tx_flags |= IGBVF_TX_FLAGS_CSUM;
2217
2218         /*
2219          * count reflects descriptors mapped, if 0 then mapping error
2220          * has occurred and we need to rewind the descriptor queue
2221          */
2222         count = igbvf_tx_map_adv(adapter, tx_ring, skb, first);
2223
2224         if (count) {
2225                 igbvf_tx_queue_adv(adapter, tx_ring, tx_flags, count,
2226                                    skb->len, hdr_len);
2227                 /* Make sure there is space in the ring for the next send. */
2228                 igbvf_maybe_stop_tx(netdev, MAX_SKB_FRAGS + 4);
2229         } else {
2230                 dev_kfree_skb_any(skb);
2231                 tx_ring->buffer_info[first].time_stamp = 0;
2232                 tx_ring->next_to_use = first;
2233         }
2234
2235         return NETDEV_TX_OK;
2236 }
2237
2238 static netdev_tx_t igbvf_xmit_frame(struct sk_buff *skb,
2239                                     struct net_device *netdev)
2240 {
2241         struct igbvf_adapter *adapter = netdev_priv(netdev);
2242         struct igbvf_ring *tx_ring;
2243
2244         if (test_bit(__IGBVF_DOWN, &adapter->state)) {
2245                 dev_kfree_skb_any(skb);
2246                 return NETDEV_TX_OK;
2247         }
2248
2249         tx_ring = &adapter->tx_ring[0];
2250
2251         return igbvf_xmit_frame_ring_adv(skb, netdev, tx_ring);
2252 }
2253
2254 /**
2255  * igbvf_tx_timeout - Respond to a Tx Hang
2256  * @netdev: network interface device structure
2257  **/
2258 static void igbvf_tx_timeout(struct net_device *netdev)
2259 {
2260         struct igbvf_adapter *adapter = netdev_priv(netdev);
2261
2262         /* Do the reset outside of interrupt context */
2263         adapter->tx_timeout_count++;
2264         schedule_work(&adapter->reset_task);
2265 }
2266
2267 static void igbvf_reset_task(struct work_struct *work)
2268 {
2269         struct igbvf_adapter *adapter;
2270         adapter = container_of(work, struct igbvf_adapter, reset_task);
2271
2272         igbvf_reinit_locked(adapter);
2273 }
2274
2275 /**
2276  * igbvf_get_stats - Get System Network Statistics
2277  * @netdev: network interface device structure
2278  *
2279  * Returns the address of the device statistics structure.
2280  * The statistics are actually updated from the timer callback.
2281  **/
2282 static struct net_device_stats *igbvf_get_stats(struct net_device *netdev)
2283 {
2284         struct igbvf_adapter *adapter = netdev_priv(netdev);
2285
2286         /* only return the current stats */
2287         return &adapter->net_stats;
2288 }
2289
2290 /**
2291  * igbvf_change_mtu - Change the Maximum Transfer Unit
2292  * @netdev: network interface device structure
2293  * @new_mtu: new value for maximum frame size
2294  *
2295  * Returns 0 on success, negative on failure
2296  **/
2297 static int igbvf_change_mtu(struct net_device *netdev, int new_mtu)
2298 {
2299         struct igbvf_adapter *adapter = netdev_priv(netdev);
2300         int max_frame = new_mtu + ETH_HLEN + ETH_FCS_LEN;
2301
2302         if ((new_mtu < 68) || (max_frame > MAX_JUMBO_FRAME_SIZE)) {
2303                 dev_err(&adapter->pdev->dev, "Invalid MTU setting\n");
2304                 return -EINVAL;
2305         }
2306
2307 #define MAX_STD_JUMBO_FRAME_SIZE 9234
2308         if (max_frame > MAX_STD_JUMBO_FRAME_SIZE) {
2309                 dev_err(&adapter->pdev->dev, "MTU > 9216 not supported.\n");
2310                 return -EINVAL;
2311         }
2312
2313         while (test_and_set_bit(__IGBVF_RESETTING, &adapter->state))
2314                 msleep(1);
2315         /* igbvf_down has a dependency on max_frame_size */
2316         adapter->max_frame_size = max_frame;
2317         if (netif_running(netdev))
2318                 igbvf_down(adapter);
2319
2320         /*
2321          * NOTE: netdev_alloc_skb reserves 16 bytes, and typically NET_IP_ALIGN
2322          * means we reserve 2 more, this pushes us to allocate from the next
2323          * larger slab size.
2324          * i.e. RXBUFFER_2048 --> size-4096 slab
2325          * However with the new *_jumbo_rx* routines, jumbo receives will use
2326          * fragmented skbs
2327          */
2328
2329         if (max_frame <= 1024)
2330                 adapter->rx_buffer_len = 1024;
2331         else if (max_frame <= 2048)
2332                 adapter->rx_buffer_len = 2048;
2333         else
2334 #if (PAGE_SIZE / 2) > 16384
2335                 adapter->rx_buffer_len = 16384;
2336 #else
2337                 adapter->rx_buffer_len = PAGE_SIZE / 2;
2338 #endif
2339
2340
2341         /* adjust allocation if LPE protects us, and we aren't using SBP */
2342         if ((max_frame == ETH_FRAME_LEN + ETH_FCS_LEN) ||
2343              (max_frame == ETH_FRAME_LEN + VLAN_HLEN + ETH_FCS_LEN))
2344                 adapter->rx_buffer_len = ETH_FRAME_LEN + VLAN_HLEN +
2345                                          ETH_FCS_LEN;
2346
2347         dev_info(&adapter->pdev->dev, "changing MTU from %d to %d\n",
2348                  netdev->mtu, new_mtu);
2349         netdev->mtu = new_mtu;
2350
2351         if (netif_running(netdev))
2352                 igbvf_up(adapter);
2353         else
2354                 igbvf_reset(adapter);
2355
2356         clear_bit(__IGBVF_RESETTING, &adapter->state);
2357
2358         return 0;
2359 }
2360
2361 static int igbvf_ioctl(struct net_device *netdev, struct ifreq *ifr, int cmd)
2362 {
2363         switch (cmd) {
2364         default:
2365                 return -EOPNOTSUPP;
2366         }
2367 }
2368
2369 static int igbvf_suspend(struct pci_dev *pdev, pm_message_t state)
2370 {
2371         struct net_device *netdev = pci_get_drvdata(pdev);
2372         struct igbvf_adapter *adapter = netdev_priv(netdev);
2373 #ifdef CONFIG_PM
2374         int retval = 0;
2375 #endif
2376
2377         netif_device_detach(netdev);
2378
2379         if (netif_running(netdev)) {
2380                 WARN_ON(test_bit(__IGBVF_RESETTING, &adapter->state));
2381                 igbvf_down(adapter);
2382                 igbvf_free_irq(adapter);
2383         }
2384
2385 #ifdef CONFIG_PM
2386         retval = pci_save_state(pdev);
2387         if (retval)
2388                 return retval;
2389 #endif
2390
2391         pci_disable_device(pdev);
2392
2393         return 0;
2394 }
2395
2396 #ifdef CONFIG_PM
2397 static int igbvf_resume(struct pci_dev *pdev)
2398 {
2399         struct net_device *netdev = pci_get_drvdata(pdev);
2400         struct igbvf_adapter *adapter = netdev_priv(netdev);
2401         u32 err;
2402
2403         pci_restore_state(pdev);
2404         err = pci_enable_device_mem(pdev);
2405         if (err) {
2406                 dev_err(&pdev->dev, "Cannot enable PCI device from suspend\n");
2407                 return err;
2408         }
2409
2410         pci_set_master(pdev);
2411
2412         if (netif_running(netdev)) {
2413                 err = igbvf_request_irq(adapter);
2414                 if (err)
2415                         return err;
2416         }
2417
2418         igbvf_reset(adapter);
2419
2420         if (netif_running(netdev))
2421                 igbvf_up(adapter);
2422
2423         netif_device_attach(netdev);
2424
2425         return 0;
2426 }
2427 #endif
2428
2429 static void igbvf_shutdown(struct pci_dev *pdev)
2430 {
2431         igbvf_suspend(pdev, PMSG_SUSPEND);
2432 }
2433
2434 #ifdef CONFIG_NET_POLL_CONTROLLER
2435 /*
2436  * Polling 'interrupt' - used by things like netconsole to send skbs
2437  * without having to re-enable interrupts. It's not called while
2438  * the interrupt routine is executing.
2439  */
2440 static void igbvf_netpoll(struct net_device *netdev)
2441 {
2442         struct igbvf_adapter *adapter = netdev_priv(netdev);
2443
2444         disable_irq(adapter->pdev->irq);
2445
2446         igbvf_clean_tx_irq(adapter->tx_ring);
2447
2448         enable_irq(adapter->pdev->irq);
2449 }
2450 #endif
2451
2452 /**
2453  * igbvf_io_error_detected - called when PCI error is detected
2454  * @pdev: Pointer to PCI device
2455  * @state: The current pci connection state
2456  *
2457  * This function is called after a PCI bus error affecting
2458  * this device has been detected.
2459  */
2460 static pci_ers_result_t igbvf_io_error_detected(struct pci_dev *pdev,
2461                                                 pci_channel_state_t state)
2462 {
2463         struct net_device *netdev = pci_get_drvdata(pdev);
2464         struct igbvf_adapter *adapter = netdev_priv(netdev);
2465
2466         netif_device_detach(netdev);
2467
2468         if (state == pci_channel_io_perm_failure)
2469                 return PCI_ERS_RESULT_DISCONNECT;
2470
2471         if (netif_running(netdev))
2472                 igbvf_down(adapter);
2473         pci_disable_device(pdev);
2474
2475         /* Request a slot slot reset. */
2476         return PCI_ERS_RESULT_NEED_RESET;
2477 }
2478
2479 /**
2480  * igbvf_io_slot_reset - called after the pci bus has been reset.
2481  * @pdev: Pointer to PCI device
2482  *
2483  * Restart the card from scratch, as if from a cold-boot. Implementation
2484  * resembles the first-half of the igbvf_resume routine.
2485  */
2486 static pci_ers_result_t igbvf_io_slot_reset(struct pci_dev *pdev)
2487 {
2488         struct net_device *netdev = pci_get_drvdata(pdev);
2489         struct igbvf_adapter *adapter = netdev_priv(netdev);
2490
2491         if (pci_enable_device_mem(pdev)) {
2492                 dev_err(&pdev->dev,
2493                         "Cannot re-enable PCI device after reset.\n");
2494                 return PCI_ERS_RESULT_DISCONNECT;
2495         }
2496         pci_set_master(pdev);
2497
2498         igbvf_reset(adapter);
2499
2500         return PCI_ERS_RESULT_RECOVERED;
2501 }
2502
2503 /**
2504  * igbvf_io_resume - called when traffic can start flowing again.
2505  * @pdev: Pointer to PCI device
2506  *
2507  * This callback is called when the error recovery driver tells us that
2508  * its OK to resume normal operation. Implementation resembles the
2509  * second-half of the igbvf_resume routine.
2510  */
2511 static void igbvf_io_resume(struct pci_dev *pdev)
2512 {
2513         struct net_device *netdev = pci_get_drvdata(pdev);
2514         struct igbvf_adapter *adapter = netdev_priv(netdev);
2515
2516         if (netif_running(netdev)) {
2517                 if (igbvf_up(adapter)) {
2518                         dev_err(&pdev->dev,
2519                                 "can't bring device back up after reset\n");
2520                         return;
2521                 }
2522         }
2523
2524         netif_device_attach(netdev);
2525 }
2526
2527 static void igbvf_print_device_info(struct igbvf_adapter *adapter)
2528 {
2529         struct e1000_hw *hw = &adapter->hw;
2530         struct net_device *netdev = adapter->netdev;
2531         struct pci_dev *pdev = adapter->pdev;
2532
2533         if (hw->mac.type == e1000_vfadapt_i350)
2534                 dev_info(&pdev->dev, "Intel(R) I350 Virtual Function\n");
2535         else
2536                 dev_info(&pdev->dev, "Intel(R) 82576 Virtual Function\n");
2537         dev_info(&pdev->dev, "Address: %pM\n", netdev->dev_addr);
2538 }
2539
2540 static int igbvf_set_features(struct net_device *netdev,
2541         netdev_features_t features)
2542 {
2543         struct igbvf_adapter *adapter = netdev_priv(netdev);
2544
2545         if (features & NETIF_F_RXCSUM)
2546                 adapter->flags &= ~IGBVF_FLAG_RX_CSUM_DISABLED;
2547         else
2548                 adapter->flags |= IGBVF_FLAG_RX_CSUM_DISABLED;
2549
2550         return 0;
2551 }
2552
2553 static const struct net_device_ops igbvf_netdev_ops = {
2554         .ndo_open                       = igbvf_open,
2555         .ndo_stop                       = igbvf_close,
2556         .ndo_start_xmit                 = igbvf_xmit_frame,
2557         .ndo_get_stats                  = igbvf_get_stats,
2558         .ndo_set_rx_mode                = igbvf_set_multi,
2559         .ndo_set_mac_address            = igbvf_set_mac,
2560         .ndo_change_mtu                 = igbvf_change_mtu,
2561         .ndo_do_ioctl                   = igbvf_ioctl,
2562         .ndo_tx_timeout                 = igbvf_tx_timeout,
2563         .ndo_vlan_rx_add_vid            = igbvf_vlan_rx_add_vid,
2564         .ndo_vlan_rx_kill_vid           = igbvf_vlan_rx_kill_vid,
2565 #ifdef CONFIG_NET_POLL_CONTROLLER
2566         .ndo_poll_controller            = igbvf_netpoll,
2567 #endif
2568         .ndo_set_features               = igbvf_set_features,
2569 };
2570
2571 /**
2572  * igbvf_probe - Device Initialization Routine
2573  * @pdev: PCI device information struct
2574  * @ent: entry in igbvf_pci_tbl
2575  *
2576  * Returns 0 on success, negative on failure
2577  *
2578  * igbvf_probe initializes an adapter identified by a pci_dev structure.
2579  * The OS initialization, configuring of the adapter private structure,
2580  * and a hardware reset occur.
2581  **/
2582 static int __devinit igbvf_probe(struct pci_dev *pdev,
2583                                  const struct pci_device_id *ent)
2584 {
2585         struct net_device *netdev;
2586         struct igbvf_adapter *adapter;
2587         struct e1000_hw *hw;
2588         const struct igbvf_info *ei = igbvf_info_tbl[ent->driver_data];
2589
2590         static int cards_found;
2591         int err, pci_using_dac;
2592
2593         err = pci_enable_device_mem(pdev);
2594         if (err)
2595                 return err;
2596
2597         pci_using_dac = 0;
2598         err = dma_set_mask(&pdev->dev, DMA_BIT_MASK(64));
2599         if (!err) {
2600                 err = dma_set_coherent_mask(&pdev->dev, DMA_BIT_MASK(64));
2601                 if (!err)
2602                         pci_using_dac = 1;
2603         } else {
2604                 err = dma_set_mask(&pdev->dev, DMA_BIT_MASK(32));
2605                 if (err) {
2606                         err = dma_set_coherent_mask(&pdev->dev,
2607                                                     DMA_BIT_MASK(32));
2608                         if (err) {
2609                                 dev_err(&pdev->dev, "No usable DMA "
2610                                         "configuration, aborting\n");
2611                                 goto err_dma;
2612                         }
2613                 }
2614         }
2615
2616         err = pci_request_regions(pdev, igbvf_driver_name);
2617         if (err)
2618                 goto err_pci_reg;
2619
2620         pci_set_master(pdev);
2621
2622         err = -ENOMEM;
2623         netdev = alloc_etherdev(sizeof(struct igbvf_adapter));
2624         if (!netdev)
2625                 goto err_alloc_etherdev;
2626
2627         SET_NETDEV_DEV(netdev, &pdev->dev);
2628
2629         pci_set_drvdata(pdev, netdev);
2630         adapter = netdev_priv(netdev);
2631         hw = &adapter->hw;
2632         adapter->netdev = netdev;
2633         adapter->pdev = pdev;
2634         adapter->ei = ei;
2635         adapter->pba = ei->pba;
2636         adapter->flags = ei->flags;
2637         adapter->hw.back = adapter;
2638         adapter->hw.mac.type = ei->mac;
2639         adapter->msg_enable = (1 << NETIF_MSG_DRV | NETIF_MSG_PROBE) - 1;
2640
2641         /* PCI config space info */
2642
2643         hw->vendor_id = pdev->vendor;
2644         hw->device_id = pdev->device;
2645         hw->subsystem_vendor_id = pdev->subsystem_vendor;
2646         hw->subsystem_device_id = pdev->subsystem_device;
2647         hw->revision_id = pdev->revision;
2648
2649         err = -EIO;
2650         adapter->hw.hw_addr = ioremap(pci_resource_start(pdev, 0),
2651                                       pci_resource_len(pdev, 0));
2652
2653         if (!adapter->hw.hw_addr)
2654                 goto err_ioremap;
2655
2656         if (ei->get_variants) {
2657                 err = ei->get_variants(adapter);
2658                 if (err)
2659                         goto err_ioremap;
2660         }
2661
2662         /* setup adapter struct */
2663         err = igbvf_sw_init(adapter);
2664         if (err)
2665                 goto err_sw_init;
2666
2667         /* construct the net_device struct */
2668         netdev->netdev_ops = &igbvf_netdev_ops;
2669
2670         igbvf_set_ethtool_ops(netdev);
2671         netdev->watchdog_timeo = 5 * HZ;
2672         strncpy(netdev->name, pci_name(pdev), sizeof(netdev->name) - 1);
2673
2674         adapter->bd_number = cards_found++;
2675
2676         netdev->hw_features = NETIF_F_SG |
2677                            NETIF_F_IP_CSUM |
2678                            NETIF_F_IPV6_CSUM |
2679                            NETIF_F_TSO |
2680                            NETIF_F_TSO6 |
2681                            NETIF_F_RXCSUM;
2682
2683         netdev->features = netdev->hw_features |
2684                            NETIF_F_HW_VLAN_TX |
2685                            NETIF_F_HW_VLAN_RX |
2686                            NETIF_F_HW_VLAN_FILTER;
2687
2688         if (pci_using_dac)
2689                 netdev->features |= NETIF_F_HIGHDMA;
2690
2691         netdev->vlan_features |= NETIF_F_TSO;
2692         netdev->vlan_features |= NETIF_F_TSO6;
2693         netdev->vlan_features |= NETIF_F_IP_CSUM;
2694         netdev->vlan_features |= NETIF_F_IPV6_CSUM;
2695         netdev->vlan_features |= NETIF_F_SG;
2696
2697         /*reset the controller to put the device in a known good state */
2698         err = hw->mac.ops.reset_hw(hw);
2699         if (err) {
2700                 dev_info(&pdev->dev,
2701                          "PF still in reset state, assigning new address."
2702                          " Is the PF interface up?\n");
2703                 dev_hw_addr_random(adapter->netdev, hw->mac.addr);
2704         } else {
2705                 err = hw->mac.ops.read_mac_addr(hw);
2706                 if (err) {
2707                         dev_err(&pdev->dev, "Error reading MAC address\n");
2708                         goto err_hw_init;
2709                 }
2710         }
2711
2712         memcpy(netdev->dev_addr, adapter->hw.mac.addr, netdev->addr_len);
2713         memcpy(netdev->perm_addr, adapter->hw.mac.addr, netdev->addr_len);
2714
2715         if (!is_valid_ether_addr(netdev->perm_addr)) {
2716                 dev_err(&pdev->dev, "Invalid MAC Address: %pM\n",
2717                         netdev->dev_addr);
2718                 err = -EIO;
2719                 goto err_hw_init;
2720         }
2721
2722         setup_timer(&adapter->watchdog_timer, &igbvf_watchdog,
2723                     (unsigned long) adapter);
2724
2725         INIT_WORK(&adapter->reset_task, igbvf_reset_task);
2726         INIT_WORK(&adapter->watchdog_task, igbvf_watchdog_task);
2727
2728         /* ring size defaults */
2729         adapter->rx_ring->count = 1024;
2730         adapter->tx_ring->count = 1024;
2731
2732         /* reset the hardware with the new settings */
2733         igbvf_reset(adapter);
2734
2735         strcpy(netdev->name, "eth%d");
2736         err = register_netdev(netdev);
2737         if (err)
2738                 goto err_hw_init;
2739
2740         /* tell the stack to leave us alone until igbvf_open() is called */
2741         netif_carrier_off(netdev);
2742         netif_stop_queue(netdev);
2743
2744         igbvf_print_device_info(adapter);
2745
2746         igbvf_initialize_last_counter_stats(adapter);
2747
2748         return 0;
2749
2750 err_hw_init:
2751         kfree(adapter->tx_ring);
2752         kfree(adapter->rx_ring);
2753 err_sw_init:
2754         igbvf_reset_interrupt_capability(adapter);
2755         iounmap(adapter->hw.hw_addr);
2756 err_ioremap:
2757         free_netdev(netdev);
2758 err_alloc_etherdev:
2759         pci_release_regions(pdev);
2760 err_pci_reg:
2761 err_dma:
2762         pci_disable_device(pdev);
2763         return err;
2764 }
2765
2766 /**
2767  * igbvf_remove - Device Removal Routine
2768  * @pdev: PCI device information struct
2769  *
2770  * igbvf_remove is called by the PCI subsystem to alert the driver
2771  * that it should release a PCI device.  The could be caused by a
2772  * Hot-Plug event, or because the driver is going to be removed from
2773  * memory.
2774  **/
2775 static void __devexit igbvf_remove(struct pci_dev *pdev)
2776 {
2777         struct net_device *netdev = pci_get_drvdata(pdev);
2778         struct igbvf_adapter *adapter = netdev_priv(netdev);
2779         struct e1000_hw *hw = &adapter->hw;
2780
2781         /*
2782          * The watchdog timer may be rescheduled, so explicitly
2783          * disable it from being rescheduled.
2784          */
2785         set_bit(__IGBVF_DOWN, &adapter->state);
2786         del_timer_sync(&adapter->watchdog_timer);
2787
2788         cancel_work_sync(&adapter->reset_task);
2789         cancel_work_sync(&adapter->watchdog_task);
2790
2791         unregister_netdev(netdev);
2792
2793         igbvf_reset_interrupt_capability(adapter);
2794
2795         /*
2796          * it is important to delete the napi struct prior to freeing the
2797          * rx ring so that you do not end up with null pointer refs
2798          */
2799         netif_napi_del(&adapter->rx_ring->napi);
2800         kfree(adapter->tx_ring);
2801         kfree(adapter->rx_ring);
2802
2803         iounmap(hw->hw_addr);
2804         if (hw->flash_address)
2805                 iounmap(hw->flash_address);
2806         pci_release_regions(pdev);
2807
2808         free_netdev(netdev);
2809
2810         pci_disable_device(pdev);
2811 }
2812
2813 /* PCI Error Recovery (ERS) */
2814 static struct pci_error_handlers igbvf_err_handler = {
2815         .error_detected = igbvf_io_error_detected,
2816         .slot_reset = igbvf_io_slot_reset,
2817         .resume = igbvf_io_resume,
2818 };
2819
2820 static DEFINE_PCI_DEVICE_TABLE(igbvf_pci_tbl) = {
2821         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82576_VF), board_vf },
2822         { PCI_VDEVICE(INTEL, E1000_DEV_ID_I350_VF), board_i350_vf },
2823         { } /* terminate list */
2824 };
2825 MODULE_DEVICE_TABLE(pci, igbvf_pci_tbl);
2826
2827 /* PCI Device API Driver */
2828 static struct pci_driver igbvf_driver = {
2829         .name     = igbvf_driver_name,
2830         .id_table = igbvf_pci_tbl,
2831         .probe    = igbvf_probe,
2832         .remove   = __devexit_p(igbvf_remove),
2833 #ifdef CONFIG_PM
2834         /* Power Management Hooks */
2835         .suspend  = igbvf_suspend,
2836         .resume   = igbvf_resume,
2837 #endif
2838         .shutdown = igbvf_shutdown,
2839         .err_handler = &igbvf_err_handler
2840 };
2841
2842 /**
2843  * igbvf_init_module - Driver Registration Routine
2844  *
2845  * igbvf_init_module is the first routine called when the driver is
2846  * loaded. All it does is register with the PCI subsystem.
2847  **/
2848 static int __init igbvf_init_module(void)
2849 {
2850         int ret;
2851         pr_info("%s - version %s\n", igbvf_driver_string, igbvf_driver_version);
2852         pr_info("%s\n", igbvf_copyright);
2853
2854         ret = pci_register_driver(&igbvf_driver);
2855
2856         return ret;
2857 }
2858 module_init(igbvf_init_module);
2859
2860 /**
2861  * igbvf_exit_module - Driver Exit Cleanup Routine
2862  *
2863  * igbvf_exit_module is called just before the driver is removed
2864  * from memory.
2865  **/
2866 static void __exit igbvf_exit_module(void)
2867 {
2868         pci_unregister_driver(&igbvf_driver);
2869 }
2870 module_exit(igbvf_exit_module);
2871
2872
2873 MODULE_AUTHOR("Intel Corporation, <e1000-devel@lists.sourceforge.net>");
2874 MODULE_DESCRIPTION("Intel(R) Gigabit Virtual Function Network Driver");
2875 MODULE_LICENSE("GPL");
2876 MODULE_VERSION(DRV_VERSION);
2877
2878 /* netdev.c */