drivers/net/igbvf/netdev.c

   1 /*******************************************************************************
   2
   3   Intel(R) 82576 Virtual Function Linux driver
   4   Copyright(c) 2009 - 2010 Intel Corporation.
   5
   6   This program is free software; you can redistribute it and/or modify it
   7   under the terms and conditions of the GNU General Public License,
   8   version 2, as published by the Free Software Foundation.
   9
  10   This program is distributed in the hope it will be useful, but WITHOUT
  11   ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
  12   FITNESS FOR A PARTICULAR PURPOSE.  See the GNU General Public License for
  13   more details.
  14
  15   You should have received a copy of the GNU General Public License along with
  16   this program; if not, write to the Free Software Foundation, Inc.,
  17   51 Franklin St - Fifth Floor, Boston, MA 02110-1301 USA.
  18
  19   The full GNU General Public License is included in this distribution in
  20   the file called "COPYING".
  21
  22   Contact Information:
  23   e1000-devel Mailing List <e1000-devel@lists.sourceforge.net>
  24   Intel Corporation, 5200 N.E. Elam Young Parkway, Hillsboro, OR 97124-6497
  25
  26 *******************************************************************************/
  27
  28 #include <linux/module.h>
  29 #include <linux/types.h>
  30 #include <linux/init.h>
  31 #include <linux/pci.h>
  32 #include <linux/vmalloc.h>
  33 #include <linux/pagemap.h>
  34 #include <linux/delay.h>
  35 #include <linux/netdevice.h>
  36 #include <linux/tcp.h>
  37 #include <linux/ipv6.h>
  38 #include <linux/slab.h>
  39 #include <net/checksum.h>
  40 #include <net/ip6_checksum.h>
  41 #include <linux/mii.h>
  42 #include <linux/ethtool.h>
  43 #include <linux/if_vlan.h>
  44
  45 #include "igbvf.h"
  46
  47 #define DRV_VERSION "1.0.8-k0"
  48 char igbvf_driver_name[] = "igbvf";
  49 const char igbvf_driver_version[] = DRV_VERSION;
  50 static const char igbvf_driver_string[] =
  51                                 "Intel(R) Virtual Function Network Driver";
  52 static const char igbvf_copyright[] =
  53                                 "Copyright (c) 2009 - 2010 Intel Corporation.";
  54
  55 static int igbvf_poll(struct napi_struct *napi, int budget);
  56 static void igbvf_reset(struct igbvf_adapter *);
  57 static void igbvf_set_interrupt_capability(struct igbvf_adapter *);
  58 static void igbvf_reset_interrupt_capability(struct igbvf_adapter *);
  59
  60 static struct igbvf_info igbvf_vf_info = {
  61         .mac                    = e1000_vfadapt,
  62         .flags                  = 0,
  63         .pba                    = 10,
  64         .init_ops               = e1000_init_function_pointers_vf,
  65 };
  66
  67 static const struct igbvf_info *igbvf_info_tbl[] = {
  68         [board_vf]              = &igbvf_vf_info,
  69 };
  70
  71 /**
  72  * igbvf_desc_unused - calculate if we have unused descriptors
  73  **/
  74 static int igbvf_desc_unused(struct igbvf_ring *ring)
  75 {
  76         if (ring->next_to_clean > ring->next_to_use)
  77                 return ring->next_to_clean - ring->next_to_use - 1;
  78
  79         return ring->count + ring->next_to_clean - ring->next_to_use - 1;
  80 }
  81
  82 /**
  83  * igbvf_receive_skb - helper function to handle Rx indications
  84  * @adapter: board private structure
  85  * @status: descriptor status field as written by hardware
  86  * @vlan: descriptor vlan field as written by hardware (no le/be conversion)
  87  * @skb: pointer to sk_buff to be indicated to stack
  88  **/
  89 static void igbvf_receive_skb(struct igbvf_adapter *adapter,
  90                               struct net_device *netdev,
  91                               struct sk_buff *skb,
  92                               u32 status, u16 vlan)
  93 {
  94         if (adapter->vlgrp && (status & E1000_RXD_STAT_VP))
  95                 vlan_hwaccel_receive_skb(skb, adapter->vlgrp,
  96                                          le16_to_cpu(vlan) &
  97                                          E1000_RXD_SPC_VLAN_MASK);
  98         else
  99                 netif_receive_skb(skb);
 100 }
 101
 102 static inline void igbvf_rx_checksum_adv(struct igbvf_adapter *adapter,
 103                                          u32 status_err, struct sk_buff *skb)
 104 {
 105         skb_checksum_none_assert(skb);
 106
 107         /* Ignore Checksum bit is set or checksum is disabled through ethtool */
 108         if ((status_err & E1000_RXD_STAT_IXSM) ||
 109             (adapter->flags & IGBVF_FLAG_RX_CSUM_DISABLED))
 110                 return;
 111
 112         /* TCP/UDP checksum error bit is set */
 113         if (status_err &
 114             (E1000_RXDEXT_STATERR_TCPE | E1000_RXDEXT_STATERR_IPE)) {
 115                 /* let the stack verify checksum errors */
 116                 adapter->hw_csum_err++;
 117                 return;
 118         }
 119
 120         /* It must be a TCP or UDP packet with a valid checksum */
 121         if (status_err & (E1000_RXD_STAT_TCPCS | E1000_RXD_STAT_UDPCS))
 122                 skb->ip_summed = CHECKSUM_UNNECESSARY;
 123
 124         adapter->hw_csum_good++;
 125 }
 126
 127 /**
 128  * igbvf_alloc_rx_buffers - Replace used receive buffers; packet split
 129  * @rx_ring: address of ring structure to repopulate
 130  * @cleaned_count: number of buffers to repopulate
 131  **/
 132 static void igbvf_alloc_rx_buffers(struct igbvf_ring *rx_ring,
 133                                    int cleaned_count)
 134 {
 135         struct igbvf_adapter *adapter = rx_ring->adapter;
 136         struct net_device *netdev = adapter->netdev;
 137         struct pci_dev *pdev = adapter->pdev;
 138         union e1000_adv_rx_desc *rx_desc;
 139         struct igbvf_buffer *buffer_info;
 140         struct sk_buff *skb;
 141         unsigned int i;
 142         int bufsz;
 143
 144         i = rx_ring->next_to_use;
 145         buffer_info = &rx_ring->buffer_info[i];
 146
 147         if (adapter->rx_ps_hdr_size)
 148                 bufsz = adapter->rx_ps_hdr_size;
 149         else
 150                 bufsz = adapter->rx_buffer_len;
 151
 152         while (cleaned_count--) {
 153                 rx_desc = IGBVF_RX_DESC_ADV(*rx_ring, i);
 154
 155                 if (adapter->rx_ps_hdr_size && !buffer_info->page_dma) {
 156                         if (!buffer_info->page) {
 157                                 buffer_info->page = alloc_page(GFP_ATOMIC);
 158                                 if (!buffer_info->page) {
 159                                         adapter->alloc_rx_buff_failed++;
 160                                         goto no_buffers;
 161                                 }
 162                                 buffer_info->page_offset = 0;
 163                         } else {
 164                                 buffer_info->page_offset ^= PAGE_SIZE / 2;
 165                         }
 166                         buffer_info->page_dma =
 167                                 dma_map_page(&pdev->dev, buffer_info->page,
 168                                              buffer_info->page_offset,
 169                                              PAGE_SIZE / 2,
 170                                              DMA_FROM_DEVICE);
 171                 }
 172
 173                 if (!buffer_info->skb) {
 174                         skb = netdev_alloc_skb_ip_align(netdev, bufsz);
 175                         if (!skb) {
 176                                 adapter->alloc_rx_buff_failed++;
 177                                 goto no_buffers;
 178                         }
 179
 180                         buffer_info->skb = skb;
 181                         buffer_info->dma = dma_map_single(&pdev->dev, skb->data,
 182                                                           bufsz,
 183                                                           DMA_FROM_DEVICE);
 184                 }
 185                 /* Refresh the desc even if buffer_addrs didn't change because
 186                  * each write-back erases this info. */
 187                 if (adapter->rx_ps_hdr_size) {
 188                         rx_desc->read.pkt_addr =
 189                              cpu_to_le64(buffer_info->page_dma);
 190                         rx_desc->read.hdr_addr = cpu_to_le64(buffer_info->dma);
 191                 } else {
 192                         rx_desc->read.pkt_addr =
 193                              cpu_to_le64(buffer_info->dma);
 194                         rx_desc->read.hdr_addr = 0;
 195                 }
 196
 197                 i++;
 198                 if (i == rx_ring->count)
 199                         i = 0;
 200                 buffer_info = &rx_ring->buffer_info[i];
 201         }
 202
 203 no_buffers:
 204         if (rx_ring->next_to_use != i) {
 205                 rx_ring->next_to_use = i;
 206                 if (i == 0)
 207                         i = (rx_ring->count - 1);
 208                 else
 209                         i--;
 210
 211                 /* Force memory writes to complete before letting h/w
 212                  * know there are new descriptors to fetch.  (Only
 213                  * applicable for weak-ordered memory model archs,
 214                  * such as IA-64). */
 215                 wmb();
 216                 writel(i, adapter->hw.hw_addr + rx_ring->tail);
 217         }
 218 }
 219
 220 /**
 221  * igbvf_clean_rx_irq - Send received data up the network stack; legacy
 222  * @adapter: board private structure
 223  *
 224  * the return value indicates whether actual cleaning was done, there
 225  * is no guarantee that everything was cleaned
 226  **/
 227 static bool igbvf_clean_rx_irq(struct igbvf_adapter *adapter,
 228                                int *work_done, int work_to_do)
 229 {
 230         struct igbvf_ring *rx_ring = adapter->rx_ring;
 231         struct net_device *netdev = adapter->netdev;
 232         struct pci_dev *pdev = adapter->pdev;
 233         union e1000_adv_rx_desc *rx_desc, *next_rxd;
 234         struct igbvf_buffer *buffer_info, *next_buffer;
 235         struct sk_buff *skb;
 236         bool cleaned = false;
 237         int cleaned_count = 0;
 238         unsigned int total_bytes = 0, total_packets = 0;
 239         unsigned int i;
 240         u32 length, hlen, staterr;
 241
 242         i = rx_ring->next_to_clean;
 243         rx_desc = IGBVF_RX_DESC_ADV(*rx_ring, i);
 244         staterr = le32_to_cpu(rx_desc->wb.upper.status_error);
 245
 246         while (staterr & E1000_RXD_STAT_DD) {
 247                 if (*work_done >= work_to_do)
 248                         break;
 249                 (*work_done)++;
 250                 rmb(); /* read descriptor and rx_buffer_info after status DD */
 251
 252                 buffer_info = &rx_ring->buffer_info[i];
 253
 254                 /* HW will not DMA in data larger than the given buffer, even
 255                  * if it parses the (NFS, of course) header to be larger.  In
 256                  * that case, it fills the header buffer and spills the rest
 257                  * into the page.
 258                  */
 259                 hlen = (le16_to_cpu(rx_desc->wb.lower.lo_dword.hs_rss.hdr_info) &
 260                   E1000_RXDADV_HDRBUFLEN_MASK) >> E1000_RXDADV_HDRBUFLEN_SHIFT;
 261                 if (hlen > adapter->rx_ps_hdr_size)
 262                         hlen = adapter->rx_ps_hdr_size;
 263
 264                 length = le16_to_cpu(rx_desc->wb.upper.length);
 265                 cleaned = true;
 266                 cleaned_count++;
 267
 268                 skb = buffer_info->skb;
 269                 prefetch(skb->data - NET_IP_ALIGN);
 270                 buffer_info->skb = NULL;
 271                 if (!adapter->rx_ps_hdr_size) {
 272                         dma_unmap_single(&pdev->dev, buffer_info->dma,
 273                                          adapter->rx_buffer_len,
 274                                          DMA_FROM_DEVICE);
 275                         buffer_info->dma = 0;
 276                         skb_put(skb, length);
 277                         goto send_up;
 278                 }
 279
 280                 if (!skb_shinfo(skb)->nr_frags) {
 281                         dma_unmap_single(&pdev->dev, buffer_info->dma,
 282                                          adapter->rx_ps_hdr_size,
 283                                          DMA_FROM_DEVICE);
 284                         skb_put(skb, hlen);
 285                 }
 286
 287                 if (length) {
 288                         dma_unmap_page(&pdev->dev, buffer_info->page_dma,
 289                                        PAGE_SIZE / 2,
 290                                        DMA_FROM_DEVICE);
 291                         buffer_info->page_dma = 0;
 292
 293                         skb_fill_page_desc(skb, skb_shinfo(skb)->nr_frags,
 294                                            buffer_info->page,
 295                                            buffer_info->page_offset,
 296                                            length);
 297
 298                         if ((adapter->rx_buffer_len > (PAGE_SIZE / 2)) ||
 299                             (page_count(buffer_info->page) != 1))
 300                                 buffer_info->page = NULL;
 301                         else
 302                                 get_page(buffer_info->page);
 303
 304                         skb->len += length;
 305                         skb->data_len += length;
 306                         skb->truesize += length;
 307                 }
 308 send_up:
 309                 i++;
 310                 if (i == rx_ring->count)
 311                         i = 0;
 312                 next_rxd = IGBVF_RX_DESC_ADV(*rx_ring, i);
 313                 prefetch(next_rxd);
 314                 next_buffer = &rx_ring->buffer_info[i];
 315
 316                 if (!(staterr & E1000_RXD_STAT_EOP)) {
 317                         buffer_info->skb = next_buffer->skb;
 318                         buffer_info->dma = next_buffer->dma;
 319                         next_buffer->skb = skb;
 320                         next_buffer->dma = 0;
 321                         goto next_desc;
 322                 }
 323
 324                 if (staterr & E1000_RXDEXT_ERR_FRAME_ERR_MASK) {
 325                         dev_kfree_skb_irq(skb);
 326                         goto next_desc;
 327                 }
 328
 329                 total_bytes += skb->len;
 330                 total_packets++;
 331
 332                 igbvf_rx_checksum_adv(adapter, staterr, skb);
 333
 334                 skb->protocol = eth_type_trans(skb, netdev);
 335
 336                 igbvf_receive_skb(adapter, netdev, skb, staterr,
 337                                   rx_desc->wb.upper.vlan);
 338
 339 next_desc:
 340                 rx_desc->wb.upper.status_error = 0;
 341
 342                 /* return some buffers to hardware, one at a time is too slow */
 343                 if (cleaned_count >= IGBVF_RX_BUFFER_WRITE) {
 344                         igbvf_alloc_rx_buffers(rx_ring, cleaned_count);
 345                         cleaned_count = 0;
 346                 }
 347
 348                 /* use prefetched values */
 349                 rx_desc = next_rxd;
 350                 buffer_info = next_buffer;
 351
 352                 staterr = le32_to_cpu(rx_desc->wb.upper.status_error);
 353         }
 354
 355         rx_ring->next_to_clean = i;
 356         cleaned_count = igbvf_desc_unused(rx_ring);
 357
 358         if (cleaned_count)
 359                 igbvf_alloc_rx_buffers(rx_ring, cleaned_count);
 360
 361         adapter->total_rx_packets += total_packets;
 362         adapter->total_rx_bytes += total_bytes;
 363         adapter->net_stats.rx_bytes += total_bytes;
 364         adapter->net_stats.rx_packets += total_packets;
 365         return cleaned;
 366 }
 367
 368 static void igbvf_put_txbuf(struct igbvf_adapter *adapter,
 369                             struct igbvf_buffer *buffer_info)
 370 {
 371         if (buffer_info->dma) {
 372                 if (buffer_info->mapped_as_page)
 373                         dma_unmap_page(&adapter->pdev->dev,
 374                                        buffer_info->dma,
 375                                        buffer_info->length,
 376                                        DMA_TO_DEVICE);
 377                 else
 378                         dma_unmap_single(&adapter->pdev->dev,
 379                                          buffer_info->dma,
 380                                          buffer_info->length,
 381                                          DMA_TO_DEVICE);
 382                 buffer_info->dma = 0;
 383         }
 384         if (buffer_info->skb) {
 385                 dev_kfree_skb_any(buffer_info->skb);
 386                 buffer_info->skb = NULL;
 387         }
 388         buffer_info->time_stamp = 0;
 389 }
 390
 391 static void igbvf_print_tx_hang(struct igbvf_adapter *adapter)
 392 {
 393         struct igbvf_ring *tx_ring = adapter->tx_ring;
 394         unsigned int i = tx_ring->next_to_clean;
 395         unsigned int eop = tx_ring->buffer_info[i].next_to_watch;
 396         union e1000_adv_tx_desc *eop_desc = IGBVF_TX_DESC_ADV(*tx_ring, eop);
 397
 398         /* detected Tx unit hang */
 399         dev_err(&adapter->pdev->dev,
 400                 "Detected Tx Unit Hang:\n"
 401                 "  TDH                  <%x>\n"
 402                 "  TDT                  <%x>\n"
 403                 "  next_to_use          <%x>\n"
 404                 "  next_to_clean        <%x>\n"
 405                 "buffer_info[next_to_clean]:\n"
 406                 "  time_stamp           <%lx>\n"
 407                 "  next_to_watch        <%x>\n"
 408                 "  jiffies              <%lx>\n"
 409                 "  next_to_watch.status <%x>\n",
 410                 readl(adapter->hw.hw_addr + tx_ring->head),
 411                 readl(adapter->hw.hw_addr + tx_ring->tail),
 412                 tx_ring->next_to_use,
 413                 tx_ring->next_to_clean,
 414                 tx_ring->buffer_info[eop].time_stamp,
 415                 eop,
 416                 jiffies,
 417                 eop_desc->wb.status);
 418 }
 419
 420 /**
 421  * igbvf_setup_tx_resources - allocate Tx resources (Descriptors)
 422  * @adapter: board private structure
 423  *
 424  * Return 0 on success, negative on failure
 425  **/
 426 int igbvf_setup_tx_resources(struct igbvf_adapter *adapter,
 427                              struct igbvf_ring *tx_ring)
 428 {
 429         struct pci_dev *pdev = adapter->pdev;
 430         int size;
 431
 432         size = sizeof(struct igbvf_buffer) * tx_ring->count;
 433         tx_ring->buffer_info = vzalloc(size);
 434         if (!tx_ring->buffer_info)
 435                 goto err;
 436
 437         /* round up to nearest 4K */
 438         tx_ring->size = tx_ring->count * sizeof(union e1000_adv_tx_desc);
 439         tx_ring->size = ALIGN(tx_ring->size, 4096);
 440
 441         tx_ring->desc = dma_alloc_coherent(&pdev->dev, tx_ring->size,
 442                                            &tx_ring->dma, GFP_KERNEL);
 443
 444         if (!tx_ring->desc)
 445                 goto err;
 446
 447         tx_ring->adapter = adapter;
 448         tx_ring->next_to_use = 0;
 449         tx_ring->next_to_clean = 0;
 450
 451         return 0;
 452 err:
 453         vfree(tx_ring->buffer_info);
 454         dev_err(&adapter->pdev->dev,
 455                 "Unable to allocate memory for the transmit descriptor ring\n");
 456         return -ENOMEM;
 457 }
 458
 459 /**
 460  * igbvf_setup_rx_resources - allocate Rx resources (Descriptors)
 461  * @adapter: board private structure
 462  *
 463  * Returns 0 on success, negative on failure
 464  **/
 465 int igbvf_setup_rx_resources(struct igbvf_adapter *adapter,
 466                              struct igbvf_ring *rx_ring)
 467 {
 468         struct pci_dev *pdev = adapter->pdev;
 469         int size, desc_len;
 470
 471         size = sizeof(struct igbvf_buffer) * rx_ring->count;
 472         rx_ring->buffer_info = vzalloc(size);
 473         if (!rx_ring->buffer_info)
 474                 goto err;
 475
 476         desc_len = sizeof(union e1000_adv_rx_desc);
 477
 478         /* Round up to nearest 4K */
 479         rx_ring->size = rx_ring->count * desc_len;
 480         rx_ring->size = ALIGN(rx_ring->size, 4096);
 481
 482         rx_ring->desc = dma_alloc_coherent(&pdev->dev, rx_ring->size,
 483                                            &rx_ring->dma, GFP_KERNEL);
 484
 485         if (!rx_ring->desc)
 486                 goto err;
 487
 488         rx_ring->next_to_clean = 0;
 489         rx_ring->next_to_use = 0;
 490
 491         rx_ring->adapter = adapter;
 492
 493         return 0;
 494
 495 err:
 496         vfree(rx_ring->buffer_info);
 497         rx_ring->buffer_info = NULL;
 498         dev_err(&adapter->pdev->dev,
 499                 "Unable to allocate memory for the receive descriptor ring\n");
 500         return -ENOMEM;
 501 }
 502
 503 /**
 504  * igbvf_clean_tx_ring - Free Tx Buffers
 505  * @tx_ring: ring to be cleaned
 506  **/
 507 static void igbvf_clean_tx_ring(struct igbvf_ring *tx_ring)
 508 {
 509         struct igbvf_adapter *adapter = tx_ring->adapter;
 510         struct igbvf_buffer *buffer_info;
 511         unsigned long size;
 512         unsigned int i;
 513
 514         if (!tx_ring->buffer_info)
 515                 return;
 516
 517         /* Free all the Tx ring sk_buffs */
 518         for (i = 0; i < tx_ring->count; i++) {
 519                 buffer_info = &tx_ring->buffer_info[i];
 520                 igbvf_put_txbuf(adapter, buffer_info);
 521         }
 522
 523         size = sizeof(struct igbvf_buffer) * tx_ring->count;
 524         memset(tx_ring->buffer_info, 0, size);
 525
 526         /* Zero out the descriptor ring */
 527         memset(tx_ring->desc, 0, tx_ring->size);
 528
 529         tx_ring->next_to_use = 0;
 530         tx_ring->next_to_clean = 0;
 531
 532         writel(0, adapter->hw.hw_addr + tx_ring->head);
 533         writel(0, adapter->hw.hw_addr + tx_ring->tail);
 534 }
 535
 536 /**
 537  * igbvf_free_tx_resources - Free Tx Resources per Queue
 538  * @tx_ring: ring to free resources from
 539  *
 540  * Free all transmit software resources
 541  **/
 542 void igbvf_free_tx_resources(struct igbvf_ring *tx_ring)
 543 {
 544         struct pci_dev *pdev = tx_ring->adapter->pdev;
 545
 546         igbvf_clean_tx_ring(tx_ring);
 547
 548         vfree(tx_ring->buffer_info);
 549         tx_ring->buffer_info = NULL;
 550
 551         dma_free_coherent(&pdev->dev, tx_ring->size, tx_ring->desc,
 552                           tx_ring->dma);
 553
 554         tx_ring->desc = NULL;
 555 }
 556
 557 /**
 558  * igbvf_clean_rx_ring - Free Rx Buffers per Queue
 559  * @adapter: board private structure
 560  **/
 561 static void igbvf_clean_rx_ring(struct igbvf_ring *rx_ring)
 562 {
 563         struct igbvf_adapter *adapter = rx_ring->adapter;
 564         struct igbvf_buffer *buffer_info;
 565         struct pci_dev *pdev = adapter->pdev;
 566         unsigned long size;
 567         unsigned int i;
 568
 569         if (!rx_ring->buffer_info)
 570                 return;
 571
 572         /* Free all the Rx ring sk_buffs */
 573         for (i = 0; i < rx_ring->count; i++) {
 574                 buffer_info = &rx_ring->buffer_info[i];
 575                 if (buffer_info->dma) {
 576                         if (adapter->rx_ps_hdr_size){
 577                                 dma_unmap_single(&pdev->dev, buffer_info->dma,
 578                                                  adapter->rx_ps_hdr_size,
 579                                                  DMA_FROM_DEVICE);
 580                         } else {
 581                                 dma_unmap_single(&pdev->dev, buffer_info->dma,
 582                                                  adapter->rx_buffer_len,
 583                                                  DMA_FROM_DEVICE);
 584                         }
 585                         buffer_info->dma = 0;
 586                 }
 587
 588                 if (buffer_info->skb) {
 589                         dev_kfree_skb(buffer_info->skb);
 590                         buffer_info->skb = NULL;
 591                 }
 592
 593                 if (buffer_info->page) {
 594                         if (buffer_info->page_dma)
 595                                 dma_unmap_page(&pdev->dev,
 596                                                buffer_info->page_dma,
 597                                                PAGE_SIZE / 2,
 598                                                DMA_FROM_DEVICE);
 599                         put_page(buffer_info->page);
 600                         buffer_info->page = NULL;
 601                         buffer_info->page_dma = 0;
 602                         buffer_info->page_offset = 0;
 603                 }
 604         }
 605
 606         size = sizeof(struct igbvf_buffer) * rx_ring->count;
 607         memset(rx_ring->buffer_info, 0, size);
 608
 609         /* Zero out the descriptor ring */
 610         memset(rx_ring->desc, 0, rx_ring->size);
 611
 612         rx_ring->next_to_clean = 0;
 613         rx_ring->next_to_use = 0;
 614
 615         writel(0, adapter->hw.hw_addr + rx_ring->head);
 616         writel(0, adapter->hw.hw_addr + rx_ring->tail);
 617 }
 618
 619 /**
 620  * igbvf_free_rx_resources - Free Rx Resources
 621  * @rx_ring: ring to clean the resources from
 622  *
 623  * Free all receive software resources
 624  **/
 625
 626 void igbvf_free_rx_resources(struct igbvf_ring *rx_ring)
 627 {
 628         struct pci_dev *pdev = rx_ring->adapter->pdev;
 629
 630         igbvf_clean_rx_ring(rx_ring);
 631
 632         vfree(rx_ring->buffer_info);
 633         rx_ring->buffer_info = NULL;
 634
 635         dma_free_coherent(&pdev->dev, rx_ring->size, rx_ring->desc,
 636                           rx_ring->dma);
 637         rx_ring->desc = NULL;
 638 }
 639
 640 /**
 641  * igbvf_update_itr - update the dynamic ITR value based on statistics
 642  * @adapter: pointer to adapter
 643  * @itr_setting: current adapter->itr
 644  * @packets: the number of packets during this measurement interval
 645  * @bytes: the number of bytes during this measurement interval
 646  *
 647  *      Stores a new ITR value based on packets and byte
 648  *      counts during the last interrupt.  The advantage of per interrupt
 649  *      computation is faster updates and more accurate ITR for the current
 650  *      traffic pattern.  Constants in this function were computed
 651  *      based on theoretical maximum wire speed and thresholds were set based
 652  *      on testing data as well as attempting to minimize response time
 653  *      while increasing bulk throughput.  This functionality is controlled
 654  *      by the InterruptThrottleRate module parameter.
 655  **/
 656 static unsigned int igbvf_update_itr(struct igbvf_adapter *adapter,
 657                                      u16 itr_setting, int packets,
 658                                      int bytes)
 659 {
 660         unsigned int retval = itr_setting;
 661
 662         if (packets == 0)
 663                 goto update_itr_done;
 664
 665         switch (itr_setting) {
 666         case lowest_latency:
 667                 /* handle TSO and jumbo frames */
 668                 if (bytes/packets > 8000)
 669                         retval = bulk_latency;
 670                 else if ((packets < 5) && (bytes > 512))
 671                         retval = low_latency;
 672                 break;
 673         case low_latency:  /* 50 usec aka 20000 ints/s */
 674                 if (bytes > 10000) {
 675                         /* this if handles the TSO accounting */
 676                         if (bytes/packets > 8000)
 677                                 retval = bulk_latency;
 678                         else if ((packets < 10) || ((bytes/packets) > 1200))
 679                                 retval = bulk_latency;
 680                         else if ((packets > 35))
 681                                 retval = lowest_latency;
 682                 } else if (bytes/packets > 2000) {
 683                         retval = bulk_latency;
 684                 } else if (packets <= 2 && bytes < 512) {
 685                         retval = lowest_latency;
 686                 }
 687                 break;
 688         case bulk_latency: /* 250 usec aka 4000 ints/s */
 689                 if (bytes > 25000) {
 690                         if (packets > 35)
 691                                 retval = low_latency;
 692                 } else if (bytes < 6000) {
 693                         retval = low_latency;
 694                 }
 695                 break;
 696         }
 697
 698 update_itr_done:
 699         return retval;
 700 }
 701
 702 static void igbvf_set_itr(struct igbvf_adapter *adapter)
 703 {
 704         struct e1000_hw *hw = &adapter->hw;
 705         u16 current_itr;
 706         u32 new_itr = adapter->itr;
 707
 708         adapter->tx_itr = igbvf_update_itr(adapter, adapter->tx_itr,
 709                                            adapter->total_tx_packets,
 710                                            adapter->total_tx_bytes);
 711         /* conservative mode (itr 3) eliminates the lowest_latency setting */
 712         if (adapter->itr_setting == 3 && adapter->tx_itr == lowest_latency)
 713                 adapter->tx_itr = low_latency;
 714
 715         adapter->rx_itr = igbvf_update_itr(adapter, adapter->rx_itr,
 716                                            adapter->total_rx_packets,
 717                                            adapter->total_rx_bytes);
 718         /* conservative mode (itr 3) eliminates the lowest_latency setting */
 719         if (adapter->itr_setting == 3 && adapter->rx_itr == lowest_latency)
 720                 adapter->rx_itr = low_latency;
 721
 722         current_itr = max(adapter->rx_itr, adapter->tx_itr);
 723
 724         switch (current_itr) {
 725         /* counts and packets in update_itr are dependent on these numbers */
 726         case lowest_latency:
 727                 new_itr = 70000;
 728                 break;
 729         case low_latency:
 730                 new_itr = 20000; /* aka hwitr = ~200 */
 731                 break;
 732         case bulk_latency:
 733                 new_itr = 4000;
 734                 break;
 735         default:
 736                 break;
 737         }
 738
 739         if (new_itr != adapter->itr) {
 740                 /*
 741                  * this attempts to bias the interrupt rate towards Bulk
 742                  * by adding intermediate steps when interrupt rate is
 743                  * increasing
 744                  */
 745                 new_itr = new_itr > adapter->itr ?
 746                              min(adapter->itr + (new_itr >> 2), new_itr) :
 747                              new_itr;
 748                 adapter->itr = new_itr;
 749                 adapter->rx_ring->itr_val = 1952;
 750
 751                 if (adapter->msix_entries)
 752                         adapter->rx_ring->set_itr = 1;
 753                 else
 754                         ew32(ITR, 1952);
 755         }
 756 }
 757
 758 /**
 759  * igbvf_clean_tx_irq - Reclaim resources after transmit completes
 760  * @adapter: board private structure
 761  * returns true if ring is completely cleaned
 762  **/
 763 static bool igbvf_clean_tx_irq(struct igbvf_ring *tx_ring)
 764 {
 765         struct igbvf_adapter *adapter = tx_ring->adapter;
 766         struct e1000_hw *hw = &adapter->hw;
 767         struct net_device *netdev = adapter->netdev;
 768         struct igbvf_buffer *buffer_info;
 769         struct sk_buff *skb;
 770         union e1000_adv_tx_desc *tx_desc, *eop_desc;
 771         unsigned int total_bytes = 0, total_packets = 0;
 772         unsigned int i, eop, count = 0;
 773         bool cleaned = false;
 774
 775         i = tx_ring->next_to_clean;
 776         eop = tx_ring->buffer_info[i].next_to_watch;
 777         eop_desc = IGBVF_TX_DESC_ADV(*tx_ring, eop);
 778
 779         while ((eop_desc->wb.status & cpu_to_le32(E1000_TXD_STAT_DD)) &&
 780                (count < tx_ring->count)) {
 781                 rmb();  /* read buffer_info after eop_desc status */
 782                 for (cleaned = false; !cleaned; count++) {
 783                         tx_desc = IGBVF_TX_DESC_ADV(*tx_ring, i);
 784                         buffer_info = &tx_ring->buffer_info[i];
 785                         cleaned = (i == eop);
 786                         skb = buffer_info->skb;
 787
 788                         if (skb) {
 789                                 unsigned int segs, bytecount;
 790
 791                                 /* gso_segs is currently only valid for tcp */
 792                                 segs = skb_shinfo(skb)->gso_segs ?: 1;
 793                                 /* multiply data chunks by size of headers */
 794                                 bytecount = ((segs - 1) * skb_headlen(skb)) +
 795                                             skb->len;
 796                                 total_packets += segs;
 797                                 total_bytes += bytecount;
 798                         }
 799
 800                         igbvf_put_txbuf(adapter, buffer_info);
 801                         tx_desc->wb.status = 0;
 802
 803                         i++;
 804                         if (i == tx_ring->count)
 805                                 i = 0;
 806                 }
 807                 eop = tx_ring->buffer_info[i].next_to_watch;
 808                 eop_desc = IGBVF_TX_DESC_ADV(*tx_ring, eop);
 809         }
 810
 811         tx_ring->next_to_clean = i;
 812
 813         if (unlikely(count &&
 814                      netif_carrier_ok(netdev) &&
 815                      igbvf_desc_unused(tx_ring) >= IGBVF_TX_QUEUE_WAKE)) {
 816                 /* Make sure that anybody stopping the queue after this
 817                  * sees the new next_to_clean.
 818                  */
 819                 smp_mb();
 820                 if (netif_queue_stopped(netdev) &&
 821                     !(test_bit(__IGBVF_DOWN, &adapter->state))) {
 822                         netif_wake_queue(netdev);
 823                         ++adapter->restart_queue;
 824                 }
 825         }
 826
 827         if (adapter->detect_tx_hung) {
 828                 /* Detect a transmit hang in hardware, this serializes the
 829                  * check with the clearing of time_stamp and movement of i */
 830                 adapter->detect_tx_hung = false;
 831                 if (tx_ring->buffer_info[i].time_stamp &&
 832                     time_after(jiffies, tx_ring->buffer_info[i].time_stamp +
 833                                (adapter->tx_timeout_factor * HZ)) &&
 834                     !(er32(STATUS) & E1000_STATUS_TXOFF)) {
 835
 836                         tx_desc = IGBVF_TX_DESC_ADV(*tx_ring, i);
 837                         /* detected Tx unit hang */
 838                         igbvf_print_tx_hang(adapter);
 839
 840                         netif_stop_queue(netdev);
 841                 }
 842         }
 843         adapter->net_stats.tx_bytes += total_bytes;
 844         adapter->net_stats.tx_packets += total_packets;
 845         return count < tx_ring->count;
 846 }
 847
 848 static irqreturn_t igbvf_msix_other(int irq, void *data)
 849 {
 850         struct net_device *netdev = data;
 851         struct igbvf_adapter *adapter = netdev_priv(netdev);
 852         struct e1000_hw *hw = &adapter->hw;
 853
 854         adapter->int_counter1++;
 855
 856         netif_carrier_off(netdev);
 857         hw->mac.get_link_status = 1;
 858         if (!test_bit(__IGBVF_DOWN, &adapter->state))
 859                 mod_timer(&adapter->watchdog_timer, jiffies + 1);
 860
 861         ew32(EIMS, adapter->eims_other);
 862
 863         return IRQ_HANDLED;
 864 }
 865
 866 static irqreturn_t igbvf_intr_msix_tx(int irq, void *data)
 867 {
 868         struct net_device *netdev = data;
 869         struct igbvf_adapter *adapter = netdev_priv(netdev);
 870         struct e1000_hw *hw = &adapter->hw;
 871         struct igbvf_ring *tx_ring = adapter->tx_ring;
 872
 873
 874         adapter->total_tx_bytes = 0;
 875         adapter->total_tx_packets = 0;
 876
 877         /* auto mask will automatically reenable the interrupt when we write
 878          * EICS */
 879         if (!igbvf_clean_tx_irq(tx_ring))
 880                 /* Ring was not completely cleaned, so fire another interrupt */
 881                 ew32(EICS, tx_ring->eims_value);
 882         else
 883                 ew32(EIMS, tx_ring->eims_value);
 884
 885         return IRQ_HANDLED;
 886 }
 887
 888 static irqreturn_t igbvf_intr_msix_rx(int irq, void *data)
 889 {
 890         struct net_device *netdev = data;
 891         struct igbvf_adapter *adapter = netdev_priv(netdev);
 892
 893         adapter->int_counter0++;
 894
 895         /* Write the ITR value calculated at the end of the
 896          * previous interrupt.
 897          */
 898         if (adapter->rx_ring->set_itr) {
 899                 writel(adapter->rx_ring->itr_val,
 900                        adapter->hw.hw_addr + adapter->rx_ring->itr_register);
 901                 adapter->rx_ring->set_itr = 0;
 902         }
 903
 904         if (napi_schedule_prep(&adapter->rx_ring->napi)) {
 905                 adapter->total_rx_bytes = 0;
 906                 adapter->total_rx_packets = 0;
 907                 __napi_schedule(&adapter->rx_ring->napi);
 908         }
 909
 910         return IRQ_HANDLED;
 911 }
 912
 913 #define IGBVF_NO_QUEUE -1
 914
 915 static void igbvf_assign_vector(struct igbvf_adapter *adapter, int rx_queue,
 916                                 int tx_queue, int msix_vector)
 917 {
 918         struct e1000_hw *hw = &adapter->hw;
 919         u32 ivar, index;
 920
 921         /* 82576 uses a table-based method for assigning vectors.
 922            Each queue has a single entry in the table to which we write
 923            a vector number along with a "valid" bit.  Sadly, the layout
 924            of the table is somewhat counterintuitive. */
 925         if (rx_queue > IGBVF_NO_QUEUE) {
 926                 index = (rx_queue >> 1);
 927                 ivar = array_er32(IVAR0, index);
 928                 if (rx_queue & 0x1) {
 929                         /* vector goes into third byte of register */
 930                         ivar = ivar & 0xFF00FFFF;
 931                         ivar |= (msix_vector | E1000_IVAR_VALID) << 16;
 932                 } else {
 933                         /* vector goes into low byte of register */
 934                         ivar = ivar & 0xFFFFFF00;
 935                         ivar |= msix_vector | E1000_IVAR_VALID;
 936                 }
 937                 adapter->rx_ring[rx_queue].eims_value = 1 << msix_vector;
 938                 array_ew32(IVAR0, index, ivar);
 939         }
 940         if (tx_queue > IGBVF_NO_QUEUE) {
 941                 index = (tx_queue >> 1);
 942                 ivar = array_er32(IVAR0, index);
 943                 if (tx_queue & 0x1) {
 944                         /* vector goes into high byte of register */
 945                         ivar = ivar & 0x00FFFFFF;
 946                         ivar |= (msix_vector | E1000_IVAR_VALID) << 24;
 947                 } else {
 948                         /* vector goes into second byte of register */
 949                         ivar = ivar & 0xFFFF00FF;
 950                         ivar |= (msix_vector | E1000_IVAR_VALID) << 8;
 951                 }
 952                 adapter->tx_ring[tx_queue].eims_value = 1 << msix_vector;
 953                 array_ew32(IVAR0, index, ivar);
 954         }
 955 }
 956
 957 /**
 958  * igbvf_configure_msix - Configure MSI-X hardware
 959  *
 960  * igbvf_configure_msix sets up the hardware to properly
 961  * generate MSI-X interrupts.
 962  **/
 963 static void igbvf_configure_msix(struct igbvf_adapter *adapter)
 964 {
 965         u32 tmp;
 966         struct e1000_hw *hw = &adapter->hw;
 967         struct igbvf_ring *tx_ring = adapter->tx_ring;
 968         struct igbvf_ring *rx_ring = adapter->rx_ring;
 969         int vector = 0;
 970
 971         adapter->eims_enable_mask = 0;
 972
 973         igbvf_assign_vector(adapter, IGBVF_NO_QUEUE, 0, vector++);
 974         adapter->eims_enable_mask |= tx_ring->eims_value;
 975         if (tx_ring->itr_val)
 976                 writel(tx_ring->itr_val,
 977                        hw->hw_addr + tx_ring->itr_register);
 978         else
 979                 writel(1952, hw->hw_addr + tx_ring->itr_register);
 980
 981         igbvf_assign_vector(adapter, 0, IGBVF_NO_QUEUE, vector++);
 982         adapter->eims_enable_mask |= rx_ring->eims_value;
 983         if (rx_ring->itr_val)
 984                 writel(rx_ring->itr_val,
 985                        hw->hw_addr + rx_ring->itr_register);
 986         else
 987                 writel(1952, hw->hw_addr + rx_ring->itr_register);
 988
 989         /* set vector for other causes, i.e. link changes */
 990
 991         tmp = (vector++ | E1000_IVAR_VALID);
 992
 993         ew32(IVAR_MISC, tmp);
 994
 995         adapter->eims_enable_mask = (1 << (vector)) - 1;
 996         adapter->eims_other = 1 << (vector - 1);
 997         e1e_flush();
 998 }
 999
1000 static void igbvf_reset_interrupt_capability(struct igbvf_adapter *adapter)
1001 {
1002         if (adapter->msix_entries) {
1003                 pci_disable_msix(adapter->pdev);
1004                 kfree(adapter->msix_entries);
1005                 adapter->msix_entries = NULL;
1006         }
1007 }
1008
1009 /**
1010  * igbvf_set_interrupt_capability - set MSI or MSI-X if supported
1011  *
1012  * Attempt to configure interrupts using the best available
1013  * capabilities of the hardware and kernel.
1014  **/
1015 static void igbvf_set_interrupt_capability(struct igbvf_adapter *adapter)
1016 {
1017         int err = -ENOMEM;
1018         int i;
1019
1020         /* we allocate 3 vectors, 1 for tx, 1 for rx, one for pf messages */
1021         adapter->msix_entries = kcalloc(3, sizeof(struct msix_entry),
1022                                         GFP_KERNEL);
1023         if (adapter->msix_entries) {
1024                 for (i = 0; i < 3; i++)
1025                         adapter->msix_entries[i].entry = i;
1026
1027                 err = pci_enable_msix(adapter->pdev,
1028                                       adapter->msix_entries, 3);
1029         }
1030
1031         if (err) {
1032                 /* MSI-X failed */
1033                 dev_err(&adapter->pdev->dev,
1034                         "Failed to initialize MSI-X interrupts.\n");
1035                 igbvf_reset_interrupt_capability(adapter);
1036         }
1037 }
1038
1039 /**
1040  * igbvf_request_msix - Initialize MSI-X interrupts
1041  *
1042  * igbvf_request_msix allocates MSI-X vectors and requests interrupts from the
1043  * kernel.
1044  **/
1045 static int igbvf_request_msix(struct igbvf_adapter *adapter)
1046 {
1047         struct net_device *netdev = adapter->netdev;
1048         int err = 0, vector = 0;
1049
1050         if (strlen(netdev->name) < (IFNAMSIZ - 5)) {
1051                 sprintf(adapter->tx_ring->name, "%s-tx-0", netdev->name);
1052                 sprintf(adapter->rx_ring->name, "%s-rx-0", netdev->name);
1053         } else {
1054                 memcpy(adapter->tx_ring->name, netdev->name, IFNAMSIZ);
1055                 memcpy(adapter->rx_ring->name, netdev->name, IFNAMSIZ);
1056         }
1057
1058         err = request_irq(adapter->msix_entries[vector].vector,
1059                           igbvf_intr_msix_tx, 0, adapter->tx_ring->name,
1060                           netdev);
1061         if (err)
1062                 goto out;
1063
1064         adapter->tx_ring->itr_register = E1000_EITR(vector);
1065         adapter->tx_ring->itr_val = 1952;
1066         vector++;
1067
1068         err = request_irq(adapter->msix_entries[vector].vector,
1069                           igbvf_intr_msix_rx, 0, adapter->rx_ring->name,
1070                           netdev);
1071         if (err)
1072                 goto out;
1073
1074         adapter->rx_ring->itr_register = E1000_EITR(vector);
1075         adapter->rx_ring->itr_val = 1952;
1076         vector++;
1077
1078         err = request_irq(adapter->msix_entries[vector].vector,
1079                           igbvf_msix_other, 0, netdev->name, netdev);
1080         if (err)
1081                 goto out;
1082
1083         igbvf_configure_msix(adapter);
1084         return 0;
1085 out:
1086         return err;
1087 }
1088
1089 /**
1090  * igbvf_alloc_queues - Allocate memory for all rings
1091  * @adapter: board private structure to initialize
1092  **/
1093 static int __devinit igbvf_alloc_queues(struct igbvf_adapter *adapter)
1094 {
1095         struct net_device *netdev = adapter->netdev;
1096
1097         adapter->tx_ring = kzalloc(sizeof(struct igbvf_ring), GFP_KERNEL);
1098         if (!adapter->tx_ring)
1099                 return -ENOMEM;
1100
1101         adapter->rx_ring = kzalloc(sizeof(struct igbvf_ring), GFP_KERNEL);
1102         if (!adapter->rx_ring) {
1103                 kfree(adapter->tx_ring);
1104                 return -ENOMEM;
1105         }
1106
1107         netif_napi_add(netdev, &adapter->rx_ring->napi, igbvf_poll, 64);
1108
1109         return 0;
1110 }
1111
1112 /**
1113  * igbvf_request_irq - initialize interrupts
1114  *
1115  * Attempts to configure interrupts using the best available
1116  * capabilities of the hardware and kernel.
1117  **/
1118 static int igbvf_request_irq(struct igbvf_adapter *adapter)
1119 {
1120         int err = -1;
1121
1122         /* igbvf supports msi-x only */
1123         if (adapter->msix_entries)
1124                 err = igbvf_request_msix(adapter);
1125
1126         if (!err)
1127                 return err;
1128
1129         dev_err(&adapter->pdev->dev,
1130                 "Unable to allocate interrupt, Error: %d\n", err);
1131
1132         return err;
1133 }
1134
1135 static void igbvf_free_irq(struct igbvf_adapter *adapter)
1136 {
1137         struct net_device *netdev = adapter->netdev;
1138         int vector;
1139
1140         if (adapter->msix_entries) {
1141                 for (vector = 0; vector < 3; vector++)
1142                         free_irq(adapter->msix_entries[vector].vector, netdev);
1143         }
1144 }
1145
1146 /**
1147  * igbvf_irq_disable - Mask off interrupt generation on the NIC
1148  **/
1149 static void igbvf_irq_disable(struct igbvf_adapter *adapter)
1150 {
1151         struct e1000_hw *hw = &adapter->hw;
1152
1153         ew32(EIMC, ~0);
1154
1155         if (adapter->msix_entries)
1156                 ew32(EIAC, 0);
1157 }
1158
1159 /**
1160  * igbvf_irq_enable - Enable default interrupt generation settings
1161  **/
1162 static void igbvf_irq_enable(struct igbvf_adapter *adapter)
1163 {
1164         struct e1000_hw *hw = &adapter->hw;
1165
1166         ew32(EIAC, adapter->eims_enable_mask);
1167         ew32(EIAM, adapter->eims_enable_mask);
1168         ew32(EIMS, adapter->eims_enable_mask);
1169 }
1170
1171 /**
1172  * igbvf_poll - NAPI Rx polling callback
1173  * @napi: struct associated with this polling callback
1174  * @budget: amount of packets driver is allowed to process this poll
1175  **/
1176 static int igbvf_poll(struct napi_struct *napi, int budget)
1177 {
1178         struct igbvf_ring *rx_ring = container_of(napi, struct igbvf_ring, napi);
1179         struct igbvf_adapter *adapter = rx_ring->adapter;
1180         struct e1000_hw *hw = &adapter->hw;
1181         int work_done = 0;
1182
1183         igbvf_clean_rx_irq(adapter, &work_done, budget);
1184
1185         /* If not enough Rx work done, exit the polling mode */
1186         if (work_done < budget) {
1187                 napi_complete(napi);
1188
1189                 if (adapter->itr_setting & 3)
1190                         igbvf_set_itr(adapter);
1191
1192                 if (!test_bit(__IGBVF_DOWN, &adapter->state))
1193                         ew32(EIMS, adapter->rx_ring->eims_value);
1194         }
1195
1196         return work_done;
1197 }
1198
1199 /**
1200  * igbvf_set_rlpml - set receive large packet maximum length
1201  * @adapter: board private structure
1202  *
1203  * Configure the maximum size of packets that will be received
1204  */
1205 static void igbvf_set_rlpml(struct igbvf_adapter *adapter)
1206 {
1207         int max_frame_size = adapter->max_frame_size;
1208         struct e1000_hw *hw = &adapter->hw;
1209
1210         if (adapter->vlgrp)
1211                 max_frame_size += VLAN_TAG_SIZE;
1212
1213         e1000_rlpml_set_vf(hw, max_frame_size);
1214 }
1215
1216 static void igbvf_vlan_rx_add_vid(struct net_device *netdev, u16 vid)
1217 {
1218         struct igbvf_adapter *adapter = netdev_priv(netdev);
1219         struct e1000_hw *hw = &adapter->hw;
1220
1221         if (hw->mac.ops.set_vfta(hw, vid, true))
1222                 dev_err(&adapter->pdev->dev, "Failed to add vlan id %d\n", vid);
1223 }
1224
1225 static void igbvf_vlan_rx_kill_vid(struct net_device *netdev, u16 vid)
1226 {
1227         struct igbvf_adapter *adapter = netdev_priv(netdev);
1228         struct e1000_hw *hw = &adapter->hw;
1229
1230         igbvf_irq_disable(adapter);
1231         vlan_group_set_device(adapter->vlgrp, vid, NULL);
1232
1233         if (!test_bit(__IGBVF_DOWN, &adapter->state))
1234                 igbvf_irq_enable(adapter);
1235
1236         if (hw->mac.ops.set_vfta(hw, vid, false))
1237                 dev_err(&adapter->pdev->dev,
1238                         "Failed to remove vlan id %d\n", vid);
1239 }
1240
1241 static void igbvf_vlan_rx_register(struct net_device *netdev,
1242                                    struct vlan_group *grp)
1243 {
1244         struct igbvf_adapter *adapter = netdev_priv(netdev);
1245
1246         adapter->vlgrp = grp;
1247 }
1248
1249 static void igbvf_restore_vlan(struct igbvf_adapter *adapter)
1250 {
1251         u16 vid;
1252
1253         if (!adapter->vlgrp)
1254                 return;
1255
1256         for (vid = 0; vid < VLAN_N_VID; vid++) {
1257                 if (!vlan_group_get_device(adapter->vlgrp, vid))
1258                         continue;
1259                 igbvf_vlan_rx_add_vid(adapter->netdev, vid);
1260         }
1261
1262         igbvf_set_rlpml(adapter);
1263 }
1264
1265 /**
1266  * igbvf_configure_tx - Configure Transmit Unit after Reset
1267  * @adapter: board private structure
1268  *
1269  * Configure the Tx unit of the MAC after a reset.
1270  **/
1271 static void igbvf_configure_tx(struct igbvf_adapter *adapter)
1272 {
1273         struct e1000_hw *hw = &adapter->hw;
1274         struct igbvf_ring *tx_ring = adapter->tx_ring;
1275         u64 tdba;
1276         u32 txdctl, dca_txctrl;
1277
1278         /* disable transmits */
1279         txdctl = er32(TXDCTL(0));
1280         ew32(TXDCTL(0), txdctl & ~E1000_TXDCTL_QUEUE_ENABLE);
1281         msleep(10);
1282
1283         /* Setup the HW Tx Head and Tail descriptor pointers */
1284         ew32(TDLEN(0), tx_ring->count * sizeof(union e1000_adv_tx_desc));
1285         tdba = tx_ring->dma;
1286         ew32(TDBAL(0), (tdba & DMA_BIT_MASK(32)));
1287         ew32(TDBAH(0), (tdba >> 32));
1288         ew32(TDH(0), 0);
1289         ew32(TDT(0), 0);
1290         tx_ring->head = E1000_TDH(0);
1291         tx_ring->tail = E1000_TDT(0);
1292
1293         /* Turn off Relaxed Ordering on head write-backs.  The writebacks
1294          * MUST be delivered in order or it will completely screw up
1295          * our bookeeping.
1296          */
1297         dca_txctrl = er32(DCA_TXCTRL(0));
1298         dca_txctrl &= ~E1000_DCA_TXCTRL_TX_WB_RO_EN;
1299         ew32(DCA_TXCTRL(0), dca_txctrl);
1300
1301         /* enable transmits */
1302         txdctl |= E1000_TXDCTL_QUEUE_ENABLE;
1303         ew32(TXDCTL(0), txdctl);
1304
1305         /* Setup Transmit Descriptor Settings for eop descriptor */
1306         adapter->txd_cmd = E1000_ADVTXD_DCMD_EOP | E1000_ADVTXD_DCMD_IFCS;
1307
1308         /* enable Report Status bit */
1309         adapter->txd_cmd |= E1000_ADVTXD_DCMD_RS;
1310 }
1311
1312 /**
1313  * igbvf_setup_srrctl - configure the receive control registers
1314  * @adapter: Board private structure
1315  **/
1316 static void igbvf_setup_srrctl(struct igbvf_adapter *adapter)
1317 {
1318         struct e1000_hw *hw = &adapter->hw;
1319         u32 srrctl = 0;
1320
1321         srrctl &= ~(E1000_SRRCTL_DESCTYPE_MASK |
1322                     E1000_SRRCTL_BSIZEHDR_MASK |
1323                     E1000_SRRCTL_BSIZEPKT_MASK);
1324
1325         /* Enable queue drop to avoid head of line blocking */
1326         srrctl |= E1000_SRRCTL_DROP_EN;
1327
1328         /* Setup buffer sizes */
1329         srrctl |= ALIGN(adapter->rx_buffer_len, 1024) >>
1330                   E1000_SRRCTL_BSIZEPKT_SHIFT;
1331
1332         if (adapter->rx_buffer_len < 2048) {
1333                 adapter->rx_ps_hdr_size = 0;
1334                 srrctl |= E1000_SRRCTL_DESCTYPE_ADV_ONEBUF;
1335         } else {
1336                 adapter->rx_ps_hdr_size = 128;
1337                 srrctl |= adapter->rx_ps_hdr_size <<
1338                           E1000_SRRCTL_BSIZEHDRSIZE_SHIFT;
1339                 srrctl |= E1000_SRRCTL_DESCTYPE_HDR_SPLIT_ALWAYS;
1340         }
1341
1342         ew32(SRRCTL(0), srrctl);
1343 }
1344
1345 /**
1346  * igbvf_configure_rx - Configure Receive Unit after Reset
1347  * @adapter: board private structure
1348  *
1349  * Configure the Rx unit of the MAC after a reset.
1350  **/
1351 static void igbvf_configure_rx(struct igbvf_adapter *adapter)
1352 {
1353         struct e1000_hw *hw = &adapter->hw;
1354         struct igbvf_ring *rx_ring = adapter->rx_ring;
1355         u64 rdba;
1356         u32 rdlen, rxdctl;
1357
1358         /* disable receives */
1359         rxdctl = er32(RXDCTL(0));
1360         ew32(RXDCTL(0), rxdctl & ~E1000_RXDCTL_QUEUE_ENABLE);
1361         msleep(10);
1362
1363         rdlen = rx_ring->count * sizeof(union e1000_adv_rx_desc);
1364
1365         /*
1366          * Setup the HW Rx Head and Tail Descriptor Pointers and
1367          * the Base and Length of the Rx Descriptor Ring
1368          */
1369         rdba = rx_ring->dma;
1370         ew32(RDBAL(0), (rdba & DMA_BIT_MASK(32)));
1371         ew32(RDBAH(0), (rdba >> 32));
1372         ew32(RDLEN(0), rx_ring->count * sizeof(union e1000_adv_rx_desc));
1373         rx_ring->head = E1000_RDH(0);
1374         rx_ring->tail = E1000_RDT(0);
1375         ew32(RDH(0), 0);
1376         ew32(RDT(0), 0);
1377
1378         rxdctl |= E1000_RXDCTL_QUEUE_ENABLE;
1379         rxdctl &= 0xFFF00000;
1380         rxdctl |= IGBVF_RX_PTHRESH;
1381         rxdctl |= IGBVF_RX_HTHRESH << 8;
1382         rxdctl |= IGBVF_RX_WTHRESH << 16;
1383
1384         igbvf_set_rlpml(adapter);
1385
1386         /* enable receives */
1387         ew32(RXDCTL(0), rxdctl);
1388 }
1389
1390 /**
1391  * igbvf_set_multi - Multicast and Promiscuous mode set
1392  * @netdev: network interface device structure
1393  *
1394  * The set_multi entry point is called whenever the multicast address
1395  * list or the network interface flags are updated.  This routine is
1396  * responsible for configuring the hardware for proper multicast,
1397  * promiscuous mode, and all-multi behavior.
1398  **/
1399 static void igbvf_set_multi(struct net_device *netdev)
1400 {
1401         struct igbvf_adapter *adapter = netdev_priv(netdev);
1402         struct e1000_hw *hw = &adapter->hw;
1403         struct netdev_hw_addr *ha;
1404         u8  *mta_list = NULL;
1405         int i;
1406
1407         if (!netdev_mc_empty(netdev)) {
1408                 mta_list = kmalloc(netdev_mc_count(netdev) * 6, GFP_ATOMIC);
1409                 if (!mta_list) {
1410                         dev_err(&adapter->pdev->dev,
1411                                 "failed to allocate multicast filter list\n");
1412                         return;
1413                 }
1414         }
1415
1416         /* prepare a packed array of only addresses. */
1417         i = 0;
1418         netdev_for_each_mc_addr(ha, netdev)
1419                 memcpy(mta_list + (i++ * ETH_ALEN), ha->addr, ETH_ALEN);
1420
1421         hw->mac.ops.update_mc_addr_list(hw, mta_list, i, 0, 0);
1422         kfree(mta_list);
1423 }
1424
1425 /**
1426  * igbvf_configure - configure the hardware for Rx and Tx
1427  * @adapter: private board structure
1428  **/
1429 static void igbvf_configure(struct igbvf_adapter *adapter)
1430 {
1431         igbvf_set_multi(adapter->netdev);
1432
1433         igbvf_restore_vlan(adapter);
1434
1435         igbvf_configure_tx(adapter);
1436         igbvf_setup_srrctl(adapter);
1437         igbvf_configure_rx(adapter);
1438         igbvf_alloc_rx_buffers(adapter->rx_ring,
1439                                igbvf_desc_unused(adapter->rx_ring));
1440 }
1441
1442 /* igbvf_reset - bring the hardware into a known good state
1443  *
1444  * This function boots the hardware and enables some settings that
1445  * require a configuration cycle of the hardware - those cannot be
1446  * set/changed during runtime. After reset the device needs to be
1447  * properly configured for Rx, Tx etc.
1448  */
1449 static void igbvf_reset(struct igbvf_adapter *adapter)
1450 {
1451         struct e1000_mac_info *mac = &adapter->hw.mac;
1452         struct net_device *netdev = adapter->netdev;
1453         struct e1000_hw *hw = &adapter->hw;
1454
1455         /* Allow time for pending master requests to run */
1456         if (mac->ops.reset_hw(hw))
1457                 dev_err(&adapter->pdev->dev, "PF still resetting\n");
1458
1459         mac->ops.init_hw(hw);
1460
1461         if (is_valid_ether_addr(adapter->hw.mac.addr)) {
1462                 memcpy(netdev->dev_addr, adapter->hw.mac.addr,
1463                        netdev->addr_len);
1464                 memcpy(netdev->perm_addr, adapter->hw.mac.addr,
1465                        netdev->addr_len);
1466         }
1467
1468         adapter->last_reset = jiffies;
1469 }
1470
1471 int igbvf_up(struct igbvf_adapter *adapter)
1472 {
1473         struct e1000_hw *hw = &adapter->hw;
1474
1475         /* hardware has been reset, we need to reload some things */
1476         igbvf_configure(adapter);
1477
1478         clear_bit(__IGBVF_DOWN, &adapter->state);
1479
1480         napi_enable(&adapter->rx_ring->napi);
1481         if (adapter->msix_entries)
1482                 igbvf_configure_msix(adapter);
1483
1484         /* Clear any pending interrupts. */
1485         er32(EICR);
1486         igbvf_irq_enable(adapter);
1487
1488         /* start the watchdog */
1489         hw->mac.get_link_status = 1;
1490         mod_timer(&adapter->watchdog_timer, jiffies + 1);
1491
1492
1493         return 0;
1494 }
1495
1496 void igbvf_down(struct igbvf_adapter *adapter)
1497 {
1498         struct net_device *netdev = adapter->netdev;
1499         struct e1000_hw *hw = &adapter->hw;
1500         u32 rxdctl, txdctl;
1501
1502         /*
1503          * signal that we're down so the interrupt handler does not
1504          * reschedule our watchdog timer
1505          */
1506         set_bit(__IGBVF_DOWN, &adapter->state);
1507
1508         /* disable receives in the hardware */
1509         rxdctl = er32(RXDCTL(0));
1510         ew32(RXDCTL(0), rxdctl & ~E1000_RXDCTL_QUEUE_ENABLE);
1511
1512         netif_stop_queue(netdev);
1513
1514         /* disable transmits in the hardware */
1515         txdctl = er32(TXDCTL(0));
1516         ew32(TXDCTL(0), txdctl & ~E1000_TXDCTL_QUEUE_ENABLE);
1517
1518         /* flush both disables and wait for them to finish */
1519         e1e_flush();
1520         msleep(10);
1521
1522         napi_disable(&adapter->rx_ring->napi);
1523
1524         igbvf_irq_disable(adapter);
1525
1526         del_timer_sync(&adapter->watchdog_timer);
1527
1528         netif_carrier_off(netdev);
1529
1530         /* record the stats before reset*/
1531         igbvf_update_stats(adapter);
1532
1533         adapter->link_speed = 0;
1534         adapter->link_duplex = 0;
1535
1536         igbvf_reset(adapter);
1537         igbvf_clean_tx_ring(adapter->tx_ring);
1538         igbvf_clean_rx_ring(adapter->rx_ring);
1539 }
1540
1541 void igbvf_reinit_locked(struct igbvf_adapter *adapter)
1542 {
1543         might_sleep();
1544         while (test_and_set_bit(__IGBVF_RESETTING, &adapter->state))
1545                 msleep(1);
1546         igbvf_down(adapter);
1547         igbvf_up(adapter);
1548         clear_bit(__IGBVF_RESETTING, &adapter->state);
1549 }
1550
1551 /**
1552  * igbvf_sw_init - Initialize general software structures (struct igbvf_adapter)
1553  * @adapter: board private structure to initialize
1554  *
1555  * igbvf_sw_init initializes the Adapter private data structure.
1556  * Fields are initialized based on PCI device information and
1557  * OS network device settings (MTU size).
1558  **/
1559 static int __devinit igbvf_sw_init(struct igbvf_adapter *adapter)
1560 {
1561         struct net_device *netdev = adapter->netdev;
1562         s32 rc;
1563
1564         adapter->rx_buffer_len = ETH_FRAME_LEN + VLAN_HLEN + ETH_FCS_LEN;
1565         adapter->rx_ps_hdr_size = 0;
1566         adapter->max_frame_size = netdev->mtu + ETH_HLEN + ETH_FCS_LEN;
1567         adapter->min_frame_size = ETH_ZLEN + ETH_FCS_LEN;
1568
1569         adapter->tx_int_delay = 8;
1570         adapter->tx_abs_int_delay = 32;
1571         adapter->rx_int_delay = 0;
1572         adapter->rx_abs_int_delay = 8;
1573         adapter->itr_setting = 3;
1574         adapter->itr = 20000;
1575
1576         /* Set various function pointers */
1577         adapter->ei->init_ops(&adapter->hw);
1578
1579         rc = adapter->hw.mac.ops.init_params(&adapter->hw);
1580         if (rc)
1581                 return rc;
1582
1583         rc = adapter->hw.mbx.ops.init_params(&adapter->hw);
1584         if (rc)
1585                 return rc;
1586
1587         igbvf_set_interrupt_capability(adapter);
1588
1589         if (igbvf_alloc_queues(adapter))
1590                 return -ENOMEM;
1591
1592         spin_lock_init(&adapter->tx_queue_lock);
1593
1594         /* Explicitly disable IRQ since the NIC can be in any state. */
1595         igbvf_irq_disable(adapter);
1596
1597         spin_lock_init(&adapter->stats_lock);
1598
1599         set_bit(__IGBVF_DOWN, &adapter->state);
1600         return 0;
1601 }
1602
1603 static void igbvf_initialize_last_counter_stats(struct igbvf_adapter *adapter)
1604 {
1605         struct e1000_hw *hw = &adapter->hw;
1606
1607         adapter->stats.last_gprc = er32(VFGPRC);
1608         adapter->stats.last_gorc = er32(VFGORC);
1609         adapter->stats.last_gptc = er32(VFGPTC);
1610         adapter->stats.last_gotc = er32(VFGOTC);
1611         adapter->stats.last_mprc = er32(VFMPRC);
1612         adapter->stats.last_gotlbc = er32(VFGOTLBC);
1613         adapter->stats.last_gptlbc = er32(VFGPTLBC);
1614         adapter->stats.last_gorlbc = er32(VFGORLBC);
1615         adapter->stats.last_gprlbc = er32(VFGPRLBC);
1616
1617         adapter->stats.base_gprc = er32(VFGPRC);
1618         adapter->stats.base_gorc = er32(VFGORC);
1619         adapter->stats.base_gptc = er32(VFGPTC);
1620         adapter->stats.base_gotc = er32(VFGOTC);
1621         adapter->stats.base_mprc = er32(VFMPRC);
1622         adapter->stats.base_gotlbc = er32(VFGOTLBC);
1623         adapter->stats.base_gptlbc = er32(VFGPTLBC);
1624         adapter->stats.base_gorlbc = er32(VFGORLBC);
1625         adapter->stats.base_gprlbc = er32(VFGPRLBC);
1626 }
1627
1628 /**
1629  * igbvf_open - Called when a network interface is made active
1630  * @netdev: network interface device structure
1631  *
1632  * Returns 0 on success, negative value on failure
1633  *
1634  * The open entry point is called when a network interface is made
1635  * active by the system (IFF_UP).  At this point all resources needed
1636  * for transmit and receive operations are allocated, the interrupt
1637  * handler is registered with the OS, the watchdog timer is started,
1638  * and the stack is notified that the interface is ready.
1639  **/
1640 static int igbvf_open(struct net_device *netdev)
1641 {
1642         struct igbvf_adapter *adapter = netdev_priv(netdev);
1643         struct e1000_hw *hw = &adapter->hw;
1644         int err;
1645
1646         /* disallow open during test */
1647         if (test_bit(__IGBVF_TESTING, &adapter->state))
1648                 return -EBUSY;
1649
1650         /* allocate transmit descriptors */
1651         err = igbvf_setup_tx_resources(adapter, adapter->tx_ring);
1652         if (err)
1653                 goto err_setup_tx;
1654
1655         /* allocate receive descriptors */
1656         err = igbvf_setup_rx_resources(adapter, adapter->rx_ring);
1657         if (err)
1658                 goto err_setup_rx;
1659
1660         /*
1661          * before we allocate an interrupt, we must be ready to handle it.
1662          * Setting DEBUG_SHIRQ in the kernel makes it fire an interrupt
1663          * as soon as we call pci_request_irq, so we have to setup our
1664          * clean_rx handler before we do so.
1665          */
1666         igbvf_configure(adapter);
1667
1668         err = igbvf_request_irq(adapter);
1669         if (err)
1670                 goto err_req_irq;
1671
1672         /* From here on the code is the same as igbvf_up() */
1673         clear_bit(__IGBVF_DOWN, &adapter->state);
1674
1675         napi_enable(&adapter->rx_ring->napi);
1676
1677         /* clear any pending interrupts */
1678         er32(EICR);
1679
1680         igbvf_irq_enable(adapter);
1681
1682         /* start the watchdog */
1683         hw->mac.get_link_status = 1;
1684         mod_timer(&adapter->watchdog_timer, jiffies + 1);
1685
1686         return 0;
1687
1688 err_req_irq:
1689         igbvf_free_rx_resources(adapter->rx_ring);
1690 err_setup_rx:
1691         igbvf_free_tx_resources(adapter->tx_ring);
1692 err_setup_tx:
1693         igbvf_reset(adapter);
1694
1695         return err;
1696 }
1697
1698 /**
1699  * igbvf_close - Disables a network interface
1700  * @netdev: network interface device structure
1701  *
1702  * Returns 0, this is not allowed to fail
1703  *
1704  * The close entry point is called when an interface is de-activated
1705  * by the OS.  The hardware is still under the drivers control, but
1706  * needs to be disabled.  A global MAC reset is issued to stop the
1707  * hardware, and all transmit and receive resources are freed.
1708  **/
1709 static int igbvf_close(struct net_device *netdev)
1710 {
1711         struct igbvf_adapter *adapter = netdev_priv(netdev);
1712
1713         WARN_ON(test_bit(__IGBVF_RESETTING, &adapter->state));
1714         igbvf_down(adapter);
1715
1716         igbvf_free_irq(adapter);
1717
1718         igbvf_free_tx_resources(adapter->tx_ring);
1719         igbvf_free_rx_resources(adapter->rx_ring);
1720
1721         return 0;
1722 }
1723 /**
1724  * igbvf_set_mac - Change the Ethernet Address of the NIC
1725  * @netdev: network interface device structure
1726  * @p: pointer to an address structure
1727  *
1728  * Returns 0 on success, negative on failure
1729  **/
1730 static int igbvf_set_mac(struct net_device *netdev, void *p)
1731 {
1732         struct igbvf_adapter *adapter = netdev_priv(netdev);
1733         struct e1000_hw *hw = &adapter->hw;
1734         struct sockaddr *addr = p;
1735
1736         if (!is_valid_ether_addr(addr->sa_data))
1737                 return -EADDRNOTAVAIL;
1738
1739         memcpy(hw->mac.addr, addr->sa_data, netdev->addr_len);
1740
1741         hw->mac.ops.rar_set(hw, hw->mac.addr, 0);
1742
1743         if (memcmp(addr->sa_data, hw->mac.addr, 6))
1744                 return -EADDRNOTAVAIL;
1745
1746         memcpy(netdev->dev_addr, addr->sa_data, netdev->addr_len);
1747
1748         return 0;
1749 }
1750
1751 #define UPDATE_VF_COUNTER(reg, name)                                    \
1752         {                                                               \
1753                 u32 current_counter = er32(reg);                        \
1754                 if (current_counter < adapter->stats.last_##name)       \
1755                         adapter->stats.name += 0x100000000LL;           \
1756                 adapter->stats.last_##name = current_counter;           \
1757                 adapter->stats.name &= 0xFFFFFFFF00000000LL;            \
1758                 adapter->stats.name |= current_counter;                 \
1759         }
1760
1761 /**
1762  * igbvf_update_stats - Update the board statistics counters
1763  * @adapter: board private structure
1764 **/
1765 void igbvf_update_stats(struct igbvf_adapter *adapter)
1766 {
1767         struct e1000_hw *hw = &adapter->hw;
1768         struct pci_dev *pdev = adapter->pdev;
1769
1770         /*
1771          * Prevent stats update while adapter is being reset, link is down
1772          * or if the pci connection is down.
1773          */
1774         if (adapter->link_speed == 0)
1775                 return;
1776
1777         if (test_bit(__IGBVF_RESETTING, &adapter->state))
1778                 return;
1779
1780         if (pci_channel_offline(pdev))
1781                 return;
1782
1783         UPDATE_VF_COUNTER(VFGPRC, gprc);
1784         UPDATE_VF_COUNTER(VFGORC, gorc);
1785         UPDATE_VF_COUNTER(VFGPTC, gptc);
1786         UPDATE_VF_COUNTER(VFGOTC, gotc);
1787         UPDATE_VF_COUNTER(VFMPRC, mprc);
1788         UPDATE_VF_COUNTER(VFGOTLBC, gotlbc);
1789         UPDATE_VF_COUNTER(VFGPTLBC, gptlbc);
1790         UPDATE_VF_COUNTER(VFGORLBC, gorlbc);
1791         UPDATE_VF_COUNTER(VFGPRLBC, gprlbc);
1792
1793         /* Fill out the OS statistics structure */
1794         adapter->net_stats.multicast = adapter->stats.mprc;
1795 }
1796
1797 static void igbvf_print_link_info(struct igbvf_adapter *adapter)
1798 {
1799         dev_info(&adapter->pdev->dev, "Link is Up %d Mbps %s\n",
1800                  adapter->link_speed,
1801                  ((adapter->link_duplex == FULL_DUPLEX) ?
1802                   "Full Duplex" : "Half Duplex"));
1803 }
1804
1805 static bool igbvf_has_link(struct igbvf_adapter *adapter)
1806 {
1807         struct e1000_hw *hw = &adapter->hw;
1808         s32 ret_val = E1000_SUCCESS;
1809         bool link_active;
1810
1811         /* If interface is down, stay link down */
1812         if (test_bit(__IGBVF_DOWN, &adapter->state))
1813                 return false;
1814
1815         ret_val = hw->mac.ops.check_for_link(hw);
1816         link_active = !hw->mac.get_link_status;
1817
1818         /* if check for link returns error we will need to reset */
1819         if (ret_val && time_after(jiffies, adapter->last_reset + (10 * HZ)))
1820                 schedule_work(&adapter->reset_task);
1821
1822         return link_active;
1823 }
1824
1825 /**
1826  * igbvf_watchdog - Timer Call-back
1827  * @data: pointer to adapter cast into an unsigned long
1828  **/
1829 static void igbvf_watchdog(unsigned long data)
1830 {
1831         struct igbvf_adapter *adapter = (struct igbvf_adapter *) data;
1832
1833         /* Do the rest outside of interrupt context */
1834         schedule_work(&adapter->watchdog_task);
1835 }
1836
1837 static void igbvf_watchdog_task(struct work_struct *work)
1838 {
1839         struct igbvf_adapter *adapter = container_of(work,
1840                                                      struct igbvf_adapter,
1841                                                      watchdog_task);
1842         struct net_device *netdev = adapter->netdev;
1843         struct e1000_mac_info *mac = &adapter->hw.mac;
1844         struct igbvf_ring *tx_ring = adapter->tx_ring;
1845         struct e1000_hw *hw = &adapter->hw;
1846         u32 link;
1847         int tx_pending = 0;
1848
1849         link = igbvf_has_link(adapter);
1850
1851         if (link) {
1852                 if (!netif_carrier_ok(netdev)) {
1853                         mac->ops.get_link_up_info(&adapter->hw,
1854                                                   &adapter->link_speed,
1855                                                   &adapter->link_duplex);
1856                         igbvf_print_link_info(adapter);
1857
1858                         /* adjust timeout factor according to speed/duplex */
1859                         adapter->tx_timeout_factor = 1;
1860                         switch (adapter->link_speed) {
1861                         case SPEED_10:
1862                                 adapter->tx_timeout_factor = 16;
1863                                 break;
1864                         case SPEED_100:
1865                                 /* maybe add some timeout factor ? */
1866                                 break;
1867                         }
1868
1869                         netif_carrier_on(netdev);
1870                         netif_wake_queue(netdev);
1871                 }
1872         } else {
1873                 if (netif_carrier_ok(netdev)) {
1874                         adapter->link_speed = 0;
1875                         adapter->link_duplex = 0;
1876                         dev_info(&adapter->pdev->dev, "Link is Down\n");
1877                         netif_carrier_off(netdev);
1878                         netif_stop_queue(netdev);
1879                 }
1880         }
1881
1882         if (netif_carrier_ok(netdev)) {
1883                 igbvf_update_stats(adapter);
1884         } else {
1885                 tx_pending = (igbvf_desc_unused(tx_ring) + 1 <
1886                               tx_ring->count);
1887                 if (tx_pending) {
1888                         /*
1889                          * We've lost link, so the controller stops DMA,
1890                          * but we've got queued Tx work that's never going
1891                          * to get done, so reset controller to flush Tx.
1892                          * (Do the reset outside of interrupt context).
1893                          */
1894                         adapter->tx_timeout_count++;
1895                         schedule_work(&adapter->reset_task);
1896                 }
1897         }
1898
1899         /* Cause software interrupt to ensure Rx ring is cleaned */
1900         ew32(EICS, adapter->rx_ring->eims_value);
1901
1902         /* Force detection of hung controller every watchdog period */
1903         adapter->detect_tx_hung = 1;
1904
1905         /* Reset the timer */
1906         if (!test_bit(__IGBVF_DOWN, &adapter->state))
1907                 mod_timer(&adapter->watchdog_timer,
1908                           round_jiffies(jiffies + (2 * HZ)));
1909 }
1910
1911 #define IGBVF_TX_FLAGS_CSUM             0x00000001
1912 #define IGBVF_TX_FLAGS_VLAN             0x00000002
1913 #define IGBVF_TX_FLAGS_TSO              0x00000004
1914 #define IGBVF_TX_FLAGS_IPV4             0x00000008
1915 #define IGBVF_TX_FLAGS_VLAN_MASK        0xffff0000
1916 #define IGBVF_TX_FLAGS_VLAN_SHIFT       16
1917
1918 static int igbvf_tso(struct igbvf_adapter *adapter,
1919                      struct igbvf_ring *tx_ring,
1920                      struct sk_buff *skb, u32 tx_flags, u8 *hdr_len)
1921 {
1922         struct e1000_adv_tx_context_desc *context_desc;
1923         unsigned int i;
1924         int err;
1925         struct igbvf_buffer *buffer_info;
1926         u32 info = 0, tu_cmd = 0;
1927         u32 mss_l4len_idx, l4len;
1928         *hdr_len = 0;
1929
1930         if (skb_header_cloned(skb)) {
1931                 err = pskb_expand_head(skb, 0, 0, GFP_ATOMIC);
1932                 if (err) {
1933                         dev_err(&adapter->pdev->dev,
1934                                 "igbvf_tso returning an error\n");
1935                         return err;
1936                 }
1937         }
1938
1939         l4len = tcp_hdrlen(skb);
1940         *hdr_len += l4len;
1941
1942         if (skb->protocol == htons(ETH_P_IP)) {
1943                 struct iphdr *iph = ip_hdr(skb);
1944                 iph->tot_len = 0;
1945                 iph->check = 0;
1946                 tcp_hdr(skb)->check = ~csum_tcpudp_magic(iph->saddr,
1947                                                          iph->daddr, 0,
1948                                                          IPPROTO_TCP,
1949                                                          0);
1950         } else if (skb_is_gso_v6(skb)) {
1951                 ipv6_hdr(skb)->payload_len = 0;
1952                 tcp_hdr(skb)->check = ~csum_ipv6_magic(&ipv6_hdr(skb)->saddr,
1953                                                        &ipv6_hdr(skb)->daddr,
1954                                                        0, IPPROTO_TCP, 0);
1955         }
1956
1957         i = tx_ring->next_to_use;
1958
1959         buffer_info = &tx_ring->buffer_info[i];
1960         context_desc = IGBVF_TX_CTXTDESC_ADV(*tx_ring, i);
1961         /* VLAN MACLEN IPLEN */
1962         if (tx_flags & IGBVF_TX_FLAGS_VLAN)
1963                 info |= (tx_flags & IGBVF_TX_FLAGS_VLAN_MASK);
1964         info |= (skb_network_offset(skb) << E1000_ADVTXD_MACLEN_SHIFT);
1965         *hdr_len += skb_network_offset(skb);
1966         info |= (skb_transport_header(skb) - skb_network_header(skb));
1967         *hdr_len += (skb_transport_header(skb) - skb_network_header(skb));
1968         context_desc->vlan_macip_lens = cpu_to_le32(info);
1969
1970         /* ADV DTYP TUCMD MKRLOC/ISCSIHEDLEN */
1971         tu_cmd |= (E1000_TXD_CMD_DEXT | E1000_ADVTXD_DTYP_CTXT);
1972
1973         if (skb->protocol == htons(ETH_P_IP))
1974                 tu_cmd |= E1000_ADVTXD_TUCMD_IPV4;
1975         tu_cmd |= E1000_ADVTXD_TUCMD_L4T_TCP;
1976
1977         context_desc->type_tucmd_mlhl = cpu_to_le32(tu_cmd);
1978
1979         /* MSS L4LEN IDX */
1980         mss_l4len_idx = (skb_shinfo(skb)->gso_size << E1000_ADVTXD_MSS_SHIFT);
1981         mss_l4len_idx |= (l4len << E1000_ADVTXD_L4LEN_SHIFT);
1982
1983         context_desc->mss_l4len_idx = cpu_to_le32(mss_l4len_idx);
1984         context_desc->seqnum_seed = 0;
1985
1986         buffer_info->time_stamp = jiffies;
1987         buffer_info->next_to_watch = i;
1988         buffer_info->dma = 0;
1989         i++;
1990         if (i == tx_ring->count)
1991                 i = 0;
1992
1993         tx_ring->next_to_use = i;
1994
1995         return true;
1996 }
1997
1998 static inline bool igbvf_tx_csum(struct igbvf_adapter *adapter,
1999                                  struct igbvf_ring *tx_ring,
2000                                  struct sk_buff *skb, u32 tx_flags)
2001 {
2002         struct e1000_adv_tx_context_desc *context_desc;
2003         unsigned int i;
2004         struct igbvf_buffer *buffer_info;
2005         u32 info = 0, tu_cmd = 0;
2006
2007         if ((skb->ip_summed == CHECKSUM_PARTIAL) ||
2008             (tx_flags & IGBVF_TX_FLAGS_VLAN)) {
2009                 i = tx_ring->next_to_use;
2010                 buffer_info = &tx_ring->buffer_info[i];
2011                 context_desc = IGBVF_TX_CTXTDESC_ADV(*tx_ring, i);
2012
2013                 if (tx_flags & IGBVF_TX_FLAGS_VLAN)
2014                         info |= (tx_flags & IGBVF_TX_FLAGS_VLAN_MASK);
2015
2016                 info |= (skb_network_offset(skb) << E1000_ADVTXD_MACLEN_SHIFT);
2017                 if (skb->ip_summed == CHECKSUM_PARTIAL)
2018                         info |= (skb_transport_header(skb) -
2019                                  skb_network_header(skb));
2020
2021
2022                 context_desc->vlan_macip_lens = cpu_to_le32(info);
2023
2024                 tu_cmd |= (E1000_TXD_CMD_DEXT | E1000_ADVTXD_DTYP_CTXT);
2025
2026                 if (skb->ip_summed == CHECKSUM_PARTIAL) {
2027                         switch (skb->protocol) {
2028                         case __constant_htons(ETH_P_IP):
2029                                 tu_cmd |= E1000_ADVTXD_TUCMD_IPV4;
2030                                 if (ip_hdr(skb)->protocol == IPPROTO_TCP)
2031                                         tu_cmd |= E1000_ADVTXD_TUCMD_L4T_TCP;
2032                                 break;
2033                         case __constant_htons(ETH_P_IPV6):
2034                                 if (ipv6_hdr(skb)->nexthdr == IPPROTO_TCP)
2035                                         tu_cmd |= E1000_ADVTXD_TUCMD_L4T_TCP;
2036                                 break;
2037                         default:
2038                                 break;
2039                         }
2040                 }
2041
2042                 context_desc->type_tucmd_mlhl = cpu_to_le32(tu_cmd);
2043                 context_desc->seqnum_seed = 0;
2044                 context_desc->mss_l4len_idx = 0;
2045
2046                 buffer_info->time_stamp = jiffies;
2047                 buffer_info->next_to_watch = i;
2048                 buffer_info->dma = 0;
2049                 i++;
2050                 if (i == tx_ring->count)
2051                         i = 0;
2052                 tx_ring->next_to_use = i;
2053
2054                 return true;
2055         }
2056
2057         return false;
2058 }
2059
2060 static int igbvf_maybe_stop_tx(struct net_device *netdev, int size)
2061 {
2062         struct igbvf_adapter *adapter = netdev_priv(netdev);
2063
2064         /* there is enough descriptors then we don't need to worry  */
2065         if (igbvf_desc_unused(adapter->tx_ring) >= size)
2066                 return 0;
2067
2068         netif_stop_queue(netdev);
2069
2070         smp_mb();
2071
2072         /* We need to check again just in case room has been made available */
2073         if (igbvf_desc_unused(adapter->tx_ring) < size)
2074                 return -EBUSY;
2075
2076         netif_wake_queue(netdev);
2077
2078         ++adapter->restart_queue;
2079         return 0;
2080 }
2081
2082 #define IGBVF_MAX_TXD_PWR       16
2083 #define IGBVF_MAX_DATA_PER_TXD  (1 << IGBVF_MAX_TXD_PWR)
2084
2085 static inline int igbvf_tx_map_adv(struct igbvf_adapter *adapter,
2086                                    struct igbvf_ring *tx_ring,
2087                                    struct sk_buff *skb,
2088                                    unsigned int first)
2089 {
2090         struct igbvf_buffer *buffer_info;
2091         struct pci_dev *pdev = adapter->pdev;
2092         unsigned int len = skb_headlen(skb);
2093         unsigned int count = 0, i;
2094         unsigned int f;
2095
2096         i = tx_ring->next_to_use;
2097
2098         buffer_info = &tx_ring->buffer_info[i];
2099         BUG_ON(len >= IGBVF_MAX_DATA_PER_TXD);
2100         buffer_info->length = len;
2101         /* set time_stamp *before* dma to help avoid a possible race */
2102         buffer_info->time_stamp = jiffies;
2103         buffer_info->next_to_watch = i;
2104         buffer_info->mapped_as_page = false;
2105         buffer_info->dma = dma_map_single(&pdev->dev, skb->data, len,
2106                                           DMA_TO_DEVICE);
2107         if (dma_mapping_error(&pdev->dev, buffer_info->dma))
2108                 goto dma_error;
2109
2110
2111         for (f = 0; f < skb_shinfo(skb)->nr_frags; f++) {
2112                 struct skb_frag_struct *frag;
2113
2114                 count++;
2115                 i++;
2116                 if (i == tx_ring->count)
2117                         i = 0;
2118
2119                 frag = &skb_shinfo(skb)->frags[f];
2120                 len = frag->size;
2121
2122                 buffer_info = &tx_ring->buffer_info[i];
2123                 BUG_ON(len >= IGBVF_MAX_DATA_PER_TXD);
2124                 buffer_info->length = len;
2125                 buffer_info->time_stamp = jiffies;
2126                 buffer_info->next_to_watch = i;
2127                 buffer_info->mapped_as_page = true;
2128                 buffer_info->dma = dma_map_page(&pdev->dev,
2129                                                 frag->page,
2130                                                 frag->page_offset,
2131                                                 len,
2132                                                 DMA_TO_DEVICE);
2133                 if (dma_mapping_error(&pdev->dev, buffer_info->dma))
2134                         goto dma_error;
2135         }
2136
2137         tx_ring->buffer_info[i].skb = skb;
2138         tx_ring->buffer_info[first].next_to_watch = i;
2139
2140         return ++count;
2141
2142 dma_error:
2143         dev_err(&pdev->dev, "TX DMA map failed\n");
2144
2145         /* clear timestamp and dma mappings for failed buffer_info mapping */
2146         buffer_info->dma = 0;
2147         buffer_info->time_stamp = 0;
2148         buffer_info->length = 0;
2149         buffer_info->next_to_watch = 0;
2150         buffer_info->mapped_as_page = false;
2151         if (count)
2152                 count--;
2153
2154         /* clear timestamp and dma mappings for remaining portion of packet */
2155         while (count--) {
2156                 if (i==0)
2157                         i += tx_ring->count;
2158                 i--;
2159                 buffer_info = &tx_ring->buffer_info[i];
2160                 igbvf_put_txbuf(adapter, buffer_info);
2161         }
2162
2163         return 0;
2164 }
2165
2166 static inline void igbvf_tx_queue_adv(struct igbvf_adapter *adapter,
2167                                       struct igbvf_ring *tx_ring,
2168                                       int tx_flags, int count, u32 paylen,
2169                                       u8 hdr_len)
2170 {
2171         union e1000_adv_tx_desc *tx_desc = NULL;
2172         struct igbvf_buffer *buffer_info;
2173         u32 olinfo_status = 0, cmd_type_len;
2174         unsigned int i;
2175
2176         cmd_type_len = (E1000_ADVTXD_DTYP_DATA | E1000_ADVTXD_DCMD_IFCS |
2177                         E1000_ADVTXD_DCMD_DEXT);
2178
2179         if (tx_flags & IGBVF_TX_FLAGS_VLAN)
2180                 cmd_type_len |= E1000_ADVTXD_DCMD_VLE;
2181
2182         if (tx_flags & IGBVF_TX_FLAGS_TSO) {
2183                 cmd_type_len |= E1000_ADVTXD_DCMD_TSE;
2184
2185                 /* insert tcp checksum */
2186                 olinfo_status |= E1000_TXD_POPTS_TXSM << 8;
2187
2188                 /* insert ip checksum */
2189                 if (tx_flags & IGBVF_TX_FLAGS_IPV4)
2190                         olinfo_status |= E1000_TXD_POPTS_IXSM << 8;
2191
2192         } else if (tx_flags & IGBVF_TX_FLAGS_CSUM) {
2193                 olinfo_status |= E1000_TXD_POPTS_TXSM << 8;
2194         }
2195
2196         olinfo_status |= ((paylen - hdr_len) << E1000_ADVTXD_PAYLEN_SHIFT);
2197
2198         i = tx_ring->next_to_use;
2199         while (count--) {
2200                 buffer_info = &tx_ring->buffer_info[i];
2201                 tx_desc = IGBVF_TX_DESC_ADV(*tx_ring, i);
2202                 tx_desc->read.buffer_addr = cpu_to_le64(buffer_info->dma);
2203                 tx_desc->read.cmd_type_len =
2204                          cpu_to_le32(cmd_type_len | buffer_info->length);
2205                 tx_desc->read.olinfo_status = cpu_to_le32(olinfo_status);
2206                 i++;
2207                 if (i == tx_ring->count)
2208                         i = 0;
2209         }
2210
2211         tx_desc->read.cmd_type_len |= cpu_to_le32(adapter->txd_cmd);
2212         /* Force memory writes to complete before letting h/w
2213          * know there are new descriptors to fetch.  (Only
2214          * applicable for weak-ordered memory model archs,
2215          * such as IA-64). */
2216         wmb();
2217
2218         tx_ring->next_to_use = i;
2219         writel(i, adapter->hw.hw_addr + tx_ring->tail);
2220         /* we need this if more than one processor can write to our tail
2221          * at a time, it syncronizes IO on IA64/Altix systems */
2222         mmiowb();
2223 }
2224
2225 static netdev_tx_t igbvf_xmit_frame_ring_adv(struct sk_buff *skb,
2226                                              struct net_device *netdev,
2227                                              struct igbvf_ring *tx_ring)
2228 {
2229         struct igbvf_adapter *adapter = netdev_priv(netdev);
2230         unsigned int first, tx_flags = 0;
2231         u8 hdr_len = 0;
2232         int count = 0;
2233         int tso = 0;
2234
2235         if (test_bit(__IGBVF_DOWN, &adapter->state)) {
2236                 dev_kfree_skb_any(skb);
2237                 return NETDEV_TX_OK;
2238         }
2239
2240         if (skb->len <= 0) {
2241                 dev_kfree_skb_any(skb);
2242                 return NETDEV_TX_OK;
2243         }
2244
2245         /*
2246          * need: count + 4 desc gap to keep tail from touching
2247          *       + 2 desc gap to keep tail from touching head,
2248          *       + 1 desc for skb->data,
2249          *       + 1 desc for context descriptor,
2250          * head, otherwise try next time
2251          */
2252         if (igbvf_maybe_stop_tx(netdev, skb_shinfo(skb)->nr_frags + 4)) {
2253                 /* this is a hard error */
2254                 return NETDEV_TX_BUSY;
2255         }
2256
2257         if (adapter->vlgrp && vlan_tx_tag_present(skb)) {
2258                 tx_flags |= IGBVF_TX_FLAGS_VLAN;
2259                 tx_flags |= (vlan_tx_tag_get(skb) << IGBVF_TX_FLAGS_VLAN_SHIFT);
2260         }
2261
2262         if (skb->protocol == htons(ETH_P_IP))
2263                 tx_flags |= IGBVF_TX_FLAGS_IPV4;
2264
2265         first = tx_ring->next_to_use;
2266
2267         tso = skb_is_gso(skb) ?
2268                 igbvf_tso(adapter, tx_ring, skb, tx_flags, &hdr_len) : 0;
2269         if (unlikely(tso < 0)) {
2270                 dev_kfree_skb_any(skb);
2271                 return NETDEV_TX_OK;
2272         }
2273
2274         if (tso)
2275                 tx_flags |= IGBVF_TX_FLAGS_TSO;
2276         else if (igbvf_tx_csum(adapter, tx_ring, skb, tx_flags) &&
2277                  (skb->ip_summed == CHECKSUM_PARTIAL))
2278                 tx_flags |= IGBVF_TX_FLAGS_CSUM;
2279
2280         /*
2281          * count reflects descriptors mapped, if 0 then mapping error
2282          * has occured and we need to rewind the descriptor queue
2283          */
2284         count = igbvf_tx_map_adv(adapter, tx_ring, skb, first);
2285
2286         if (count) {
2287                 igbvf_tx_queue_adv(adapter, tx_ring, tx_flags, count,
2288                                    skb->len, hdr_len);
2289                 /* Make sure there is space in the ring for the next send. */
2290                 igbvf_maybe_stop_tx(netdev, MAX_SKB_FRAGS + 4);
2291         } else {
2292                 dev_kfree_skb_any(skb);
2293                 tx_ring->buffer_info[first].time_stamp = 0;
2294                 tx_ring->next_to_use = first;
2295         }
2296
2297         return NETDEV_TX_OK;
2298 }
2299
2300 static netdev_tx_t igbvf_xmit_frame(struct sk_buff *skb,
2301                                     struct net_device *netdev)
2302 {
2303         struct igbvf_adapter *adapter = netdev_priv(netdev);
2304         struct igbvf_ring *tx_ring;
2305
2306         if (test_bit(__IGBVF_DOWN, &adapter->state)) {
2307                 dev_kfree_skb_any(skb);
2308                 return NETDEV_TX_OK;
2309         }
2310
2311         tx_ring = &adapter->tx_ring[0];
2312
2313         return igbvf_xmit_frame_ring_adv(skb, netdev, tx_ring);
2314 }
2315
2316 /**
2317  * igbvf_tx_timeout - Respond to a Tx Hang
2318  * @netdev: network interface device structure
2319  **/
2320 static void igbvf_tx_timeout(struct net_device *netdev)
2321 {
2322         struct igbvf_adapter *adapter = netdev_priv(netdev);
2323
2324         /* Do the reset outside of interrupt context */
2325         adapter->tx_timeout_count++;
2326         schedule_work(&adapter->reset_task);
2327 }
2328
2329 static void igbvf_reset_task(struct work_struct *work)
2330 {
2331         struct igbvf_adapter *adapter;
2332         adapter = container_of(work, struct igbvf_adapter, reset_task);
2333
2334         igbvf_reinit_locked(adapter);
2335 }
2336
2337 /**
2338  * igbvf_get_stats - Get System Network Statistics
2339  * @netdev: network interface device structure
2340  *
2341  * Returns the address of the device statistics structure.
2342  * The statistics are actually updated from the timer callback.
2343  **/
2344 static struct net_device_stats *igbvf_get_stats(struct net_device *netdev)
2345 {
2346         struct igbvf_adapter *adapter = netdev_priv(netdev);
2347
2348         /* only return the current stats */
2349         return &adapter->net_stats;
2350 }
2351
2352 /**
2353  * igbvf_change_mtu - Change the Maximum Transfer Unit
2354  * @netdev: network interface device structure
2355  * @new_mtu: new value for maximum frame size
2356  *
2357  * Returns 0 on success, negative on failure
2358  **/
2359 static int igbvf_change_mtu(struct net_device *netdev, int new_mtu)
2360 {
2361         struct igbvf_adapter *adapter = netdev_priv(netdev);
2362         int max_frame = new_mtu + ETH_HLEN + ETH_FCS_LEN;
2363
2364         if ((new_mtu < 68) || (max_frame > MAX_JUMBO_FRAME_SIZE)) {
2365                 dev_err(&adapter->pdev->dev, "Invalid MTU setting\n");
2366                 return -EINVAL;
2367         }
2368
2369 #define MAX_STD_JUMBO_FRAME_SIZE 9234
2370         if (max_frame > MAX_STD_JUMBO_FRAME_SIZE) {
2371                 dev_err(&adapter->pdev->dev, "MTU > 9216 not supported.\n");
2372                 return -EINVAL;
2373         }
2374
2375         while (test_and_set_bit(__IGBVF_RESETTING, &adapter->state))
2376                 msleep(1);
2377         /* igbvf_down has a dependency on max_frame_size */
2378         adapter->max_frame_size = max_frame;
2379         if (netif_running(netdev))
2380                 igbvf_down(adapter);
2381
2382         /*
2383          * NOTE: netdev_alloc_skb reserves 16 bytes, and typically NET_IP_ALIGN
2384          * means we reserve 2 more, this pushes us to allocate from the next
2385          * larger slab size.
2386          * i.e. RXBUFFER_2048 --> size-4096 slab
2387          * However with the new *_jumbo_rx* routines, jumbo receives will use
2388          * fragmented skbs
2389          */
2390
2391         if (max_frame <= 1024)
2392                 adapter->rx_buffer_len = 1024;
2393         else if (max_frame <= 2048)
2394                 adapter->rx_buffer_len = 2048;
2395         else
2396 #if (PAGE_SIZE / 2) > 16384
2397                 adapter->rx_buffer_len = 16384;
2398 #else
2399                 adapter->rx_buffer_len = PAGE_SIZE / 2;
2400 #endif
2401
2402
2403         /* adjust allocation if LPE protects us, and we aren't using SBP */
2404         if ((max_frame == ETH_FRAME_LEN + ETH_FCS_LEN) ||
2405              (max_frame == ETH_FRAME_LEN + VLAN_HLEN + ETH_FCS_LEN))
2406                 adapter->rx_buffer_len = ETH_FRAME_LEN + VLAN_HLEN +
2407                                          ETH_FCS_LEN;
2408
2409         dev_info(&adapter->pdev->dev, "changing MTU from %d to %d\n",
2410                  netdev->mtu, new_mtu);
2411         netdev->mtu = new_mtu;
2412
2413         if (netif_running(netdev))
2414                 igbvf_up(adapter);
2415         else
2416                 igbvf_reset(adapter);
2417
2418         clear_bit(__IGBVF_RESETTING, &adapter->state);
2419
2420         return 0;
2421 }
2422
2423 static int igbvf_ioctl(struct net_device *netdev, struct ifreq *ifr, int cmd)
2424 {
2425         switch (cmd) {
2426         default:
2427                 return -EOPNOTSUPP;
2428         }
2429 }
2430
2431 static int igbvf_suspend(struct pci_dev *pdev, pm_message_t state)
2432 {
2433         struct net_device *netdev = pci_get_drvdata(pdev);
2434         struct igbvf_adapter *adapter = netdev_priv(netdev);
2435 #ifdef CONFIG_PM
2436         int retval = 0;
2437 #endif
2438
2439         netif_device_detach(netdev);
2440
2441         if (netif_running(netdev)) {
2442                 WARN_ON(test_bit(__IGBVF_RESETTING, &adapter->state));
2443                 igbvf_down(adapter);
2444                 igbvf_free_irq(adapter);
2445         }
2446
2447 #ifdef CONFIG_PM
2448         retval = pci_save_state(pdev);
2449         if (retval)
2450                 return retval;
2451 #endif
2452
2453         pci_disable_device(pdev);
2454
2455         return 0;
2456 }
2457
2458 #ifdef CONFIG_PM
2459 static int igbvf_resume(struct pci_dev *pdev)
2460 {
2461         struct net_device *netdev = pci_get_drvdata(pdev);
2462         struct igbvf_adapter *adapter = netdev_priv(netdev);
2463         u32 err;
2464
2465         pci_restore_state(pdev);
2466         err = pci_enable_device_mem(pdev);
2467         if (err) {
2468                 dev_err(&pdev->dev, "Cannot enable PCI device from suspend\n");
2469                 return err;
2470         }
2471
2472         pci_set_master(pdev);
2473
2474         if (netif_running(netdev)) {
2475                 err = igbvf_request_irq(adapter);
2476                 if (err)
2477                         return err;
2478         }
2479
2480         igbvf_reset(adapter);
2481
2482         if (netif_running(netdev))
2483                 igbvf_up(adapter);
2484
2485         netif_device_attach(netdev);
2486
2487         return 0;
2488 }
2489 #endif
2490
2491 static void igbvf_shutdown(struct pci_dev *pdev)
2492 {
2493         igbvf_suspend(pdev, PMSG_SUSPEND);
2494 }
2495
2496 #ifdef CONFIG_NET_POLL_CONTROLLER
2497 /*
2498  * Polling 'interrupt' - used by things like netconsole to send skbs
2499  * without having to re-enable interrupts. It's not called while
2500  * the interrupt routine is executing.
2501  */
2502 static void igbvf_netpoll(struct net_device *netdev)
2503 {
2504         struct igbvf_adapter *adapter = netdev_priv(netdev);
2505
2506         disable_irq(adapter->pdev->irq);
2507
2508         igbvf_clean_tx_irq(adapter->tx_ring);
2509
2510         enable_irq(adapter->pdev->irq);
2511 }
2512 #endif
2513
2514 /**
2515  * igbvf_io_error_detected - called when PCI error is detected
2516  * @pdev: Pointer to PCI device
2517  * @state: The current pci connection state
2518  *
2519  * This function is called after a PCI bus error affecting
2520  * this device has been detected.
2521  */
2522 static pci_ers_result_t igbvf_io_error_detected(struct pci_dev *pdev,
2523                                                 pci_channel_state_t state)
2524 {
2525         struct net_device *netdev = pci_get_drvdata(pdev);
2526         struct igbvf_adapter *adapter = netdev_priv(netdev);
2527
2528         netif_device_detach(netdev);
2529
2530         if (state == pci_channel_io_perm_failure)
2531                 return PCI_ERS_RESULT_DISCONNECT;
2532
2533         if (netif_running(netdev))
2534                 igbvf_down(adapter);
2535         pci_disable_device(pdev);
2536
2537         /* Request a slot slot reset. */
2538         return PCI_ERS_RESULT_NEED_RESET;
2539 }
2540
2541 /**
2542  * igbvf_io_slot_reset - called after the pci bus has been reset.
2543  * @pdev: Pointer to PCI device
2544  *
2545  * Restart the card from scratch, as if from a cold-boot. Implementation
2546  * resembles the first-half of the igbvf_resume routine.
2547  */
2548 static pci_ers_result_t igbvf_io_slot_reset(struct pci_dev *pdev)
2549 {
2550         struct net_device *netdev = pci_get_drvdata(pdev);
2551         struct igbvf_adapter *adapter = netdev_priv(netdev);
2552
2553         if (pci_enable_device_mem(pdev)) {
2554                 dev_err(&pdev->dev,
2555                         "Cannot re-enable PCI device after reset.\n");
2556                 return PCI_ERS_RESULT_DISCONNECT;
2557         }
2558         pci_set_master(pdev);
2559
2560         igbvf_reset(adapter);
2561
2562         return PCI_ERS_RESULT_RECOVERED;
2563 }
2564
2565 /**
2566  * igbvf_io_resume - called when traffic can start flowing again.
2567  * @pdev: Pointer to PCI device
2568  *
2569  * This callback is called when the error recovery driver tells us that
2570  * its OK to resume normal operation. Implementation resembles the
2571  * second-half of the igbvf_resume routine.
2572  */
2573 static void igbvf_io_resume(struct pci_dev *pdev)
2574 {
2575         struct net_device *netdev = pci_get_drvdata(pdev);
2576         struct igbvf_adapter *adapter = netdev_priv(netdev);
2577
2578         if (netif_running(netdev)) {
2579                 if (igbvf_up(adapter)) {
2580                         dev_err(&pdev->dev,
2581                                 "can't bring device back up after reset\n");
2582                         return;
2583                 }
2584         }
2585
2586         netif_device_attach(netdev);
2587 }
2588
2589 static void igbvf_print_device_info(struct igbvf_adapter *adapter)
2590 {
2591         struct e1000_hw *hw = &adapter->hw;
2592         struct net_device *netdev = adapter->netdev;
2593         struct pci_dev *pdev = adapter->pdev;
2594
2595         dev_info(&pdev->dev, "Intel(R) 82576 Virtual Function\n");
2596         dev_info(&pdev->dev, "Address: %pM\n", netdev->dev_addr);
2597         dev_info(&pdev->dev, "MAC: %d\n", hw->mac.type);
2598 }
2599
2600 static const struct net_device_ops igbvf_netdev_ops = {
2601         .ndo_open                       = igbvf_open,
2602         .ndo_stop                       = igbvf_close,
2603         .ndo_start_xmit                 = igbvf_xmit_frame,
2604         .ndo_get_stats                  = igbvf_get_stats,
2605         .ndo_set_multicast_list         = igbvf_set_multi,
2606         .ndo_set_mac_address            = igbvf_set_mac,
2607         .ndo_change_mtu                 = igbvf_change_mtu,
2608         .ndo_do_ioctl                   = igbvf_ioctl,
2609         .ndo_tx_timeout                 = igbvf_tx_timeout,
2610         .ndo_vlan_rx_register           = igbvf_vlan_rx_register,
2611         .ndo_vlan_rx_add_vid            = igbvf_vlan_rx_add_vid,
2612         .ndo_vlan_rx_kill_vid           = igbvf_vlan_rx_kill_vid,
2613 #ifdef CONFIG_NET_POLL_CONTROLLER
2614         .ndo_poll_controller            = igbvf_netpoll,
2615 #endif
2616 };
2617
2618 /**
2619  * igbvf_probe - Device Initialization Routine
2620  * @pdev: PCI device information struct
2621  * @ent: entry in igbvf_pci_tbl
2622  *
2623  * Returns 0 on success, negative on failure
2624  *
2625  * igbvf_probe initializes an adapter identified by a pci_dev structure.
2626  * The OS initialization, configuring of the adapter private structure,
2627  * and a hardware reset occur.
2628  **/
2629 static int __devinit igbvf_probe(struct pci_dev *pdev,
2630                                  const struct pci_device_id *ent)
2631 {
2632         struct net_device *netdev;
2633         struct igbvf_adapter *adapter;
2634         struct e1000_hw *hw;
2635         const struct igbvf_info *ei = igbvf_info_tbl[ent->driver_data];
2636
2637         static int cards_found;
2638         int err, pci_using_dac;
2639
2640         err = pci_enable_device_mem(pdev);
2641         if (err)
2642                 return err;
2643
2644         pci_using_dac = 0;
2645         err = dma_set_mask(&pdev->dev, DMA_BIT_MASK(64));
2646         if (!err) {
2647                 err = dma_set_coherent_mask(&pdev->dev, DMA_BIT_MASK(64));
2648                 if (!err)
2649                         pci_using_dac = 1;
2650         } else {
2651                 err = dma_set_mask(&pdev->dev, DMA_BIT_MASK(32));
2652                 if (err) {
2653                         err = dma_set_coherent_mask(&pdev->dev,
2654                                                     DMA_BIT_MASK(32));
2655                         if (err) {
2656                                 dev_err(&pdev->dev, "No usable DMA "
2657                                         "configuration, aborting\n");
2658                                 goto err_dma;
2659                         }
2660                 }
2661         }
2662
2663         err = pci_request_regions(pdev, igbvf_driver_name);
2664         if (err)
2665                 goto err_pci_reg;
2666
2667         pci_set_master(pdev);
2668
2669         err = -ENOMEM;
2670         netdev = alloc_etherdev(sizeof(struct igbvf_adapter));
2671         if (!netdev)
2672                 goto err_alloc_etherdev;
2673
2674         SET_NETDEV_DEV(netdev, &pdev->dev);
2675
2676         pci_set_drvdata(pdev, netdev);
2677         adapter = netdev_priv(netdev);
2678         hw = &adapter->hw;
2679         adapter->netdev = netdev;
2680         adapter->pdev = pdev;
2681         adapter->ei = ei;
2682         adapter->pba = ei->pba;
2683         adapter->flags = ei->flags;
2684         adapter->hw.back = adapter;
2685         adapter->hw.mac.type = ei->mac;
2686         adapter->msg_enable = (1 << NETIF_MSG_DRV | NETIF_MSG_PROBE) - 1;
2687
2688         /* PCI config space info */
2689
2690         hw->vendor_id = pdev->vendor;
2691         hw->device_id = pdev->device;
2692         hw->subsystem_vendor_id = pdev->subsystem_vendor;
2693         hw->subsystem_device_id = pdev->subsystem_device;
2694
2695         pci_read_config_byte(pdev, PCI_REVISION_ID, &hw->revision_id);
2696
2697         err = -EIO;
2698         adapter->hw.hw_addr = ioremap(pci_resource_start(pdev, 0),
2699                                       pci_resource_len(pdev, 0));
2700
2701         if (!adapter->hw.hw_addr)
2702                 goto err_ioremap;
2703
2704         if (ei->get_variants) {
2705                 err = ei->get_variants(adapter);
2706                 if (err)
2707                         goto err_ioremap;
2708         }
2709
2710         /* setup adapter struct */
2711         err = igbvf_sw_init(adapter);
2712         if (err)
2713                 goto err_sw_init;
2714
2715         /* construct the net_device struct */
2716         netdev->netdev_ops = &igbvf_netdev_ops;
2717
2718         igbvf_set_ethtool_ops(netdev);
2719         netdev->watchdog_timeo = 5 * HZ;
2720         strncpy(netdev->name, pci_name(pdev), sizeof(netdev->name) - 1);
2721
2722         adapter->bd_number = cards_found++;
2723
2724         netdev->features = NETIF_F_SG |
2725                            NETIF_F_IP_CSUM |
2726                            NETIF_F_HW_VLAN_TX |
2727                            NETIF_F_HW_VLAN_RX |
2728                            NETIF_F_HW_VLAN_FILTER;
2729
2730         netdev->features |= NETIF_F_IPV6_CSUM;
2731         netdev->features |= NETIF_F_TSO;
2732         netdev->features |= NETIF_F_TSO6;
2733
2734         if (pci_using_dac)
2735                 netdev->features |= NETIF_F_HIGHDMA;
2736
2737         netdev->vlan_features |= NETIF_F_TSO;
2738         netdev->vlan_features |= NETIF_F_TSO6;
2739         netdev->vlan_features |= NETIF_F_IP_CSUM;
2740         netdev->vlan_features |= NETIF_F_IPV6_CSUM;
2741         netdev->vlan_features |= NETIF_F_SG;
2742
2743         /*reset the controller to put the device in a known good state */
2744         err = hw->mac.ops.reset_hw(hw);
2745         if (err) {
2746                 dev_info(&pdev->dev,
2747                          "PF still in reset state, assigning new address."
2748                          " Is the PF interface up?\n");
2749                 dev_hw_addr_random(adapter->netdev, hw->mac.addr);
2750         } else {
2751                 err = hw->mac.ops.read_mac_addr(hw);
2752                 if (err) {
2753                         dev_err(&pdev->dev, "Error reading MAC address\n");
2754                         goto err_hw_init;
2755                 }
2756         }
2757
2758         memcpy(netdev->dev_addr, adapter->hw.mac.addr, netdev->addr_len);
2759         memcpy(netdev->perm_addr, adapter->hw.mac.addr, netdev->addr_len);
2760
2761         if (!is_valid_ether_addr(netdev->perm_addr)) {
2762                 dev_err(&pdev->dev, "Invalid MAC Address: %pM\n",
2763                         netdev->dev_addr);
2764                 err = -EIO;
2765                 goto err_hw_init;
2766         }
2767
2768         setup_timer(&adapter->watchdog_timer, &igbvf_watchdog,
2769                     (unsigned long) adapter);
2770
2771         INIT_WORK(&adapter->reset_task, igbvf_reset_task);
2772         INIT_WORK(&adapter->watchdog_task, igbvf_watchdog_task);
2773
2774         /* ring size defaults */
2775         adapter->rx_ring->count = 1024;
2776         adapter->tx_ring->count = 1024;
2777
2778         /* reset the hardware with the new settings */
2779         igbvf_reset(adapter);
2780
2781         strcpy(netdev->name, "eth%d");
2782         err = register_netdev(netdev);
2783         if (err)
2784                 goto err_hw_init;
2785
2786         /* tell the stack to leave us alone until igbvf_open() is called */
2787         netif_carrier_off(netdev);
2788         netif_stop_queue(netdev);
2789
2790         igbvf_print_device_info(adapter);
2791
2792         igbvf_initialize_last_counter_stats(adapter);
2793
2794         return 0;
2795
2796 err_hw_init:
2797         kfree(adapter->tx_ring);
2798         kfree(adapter->rx_ring);
2799 err_sw_init:
2800         igbvf_reset_interrupt_capability(adapter);
2801         iounmap(adapter->hw.hw_addr);
2802 err_ioremap:
2803         free_netdev(netdev);
2804 err_alloc_etherdev:
2805         pci_release_regions(pdev);
2806 err_pci_reg:
2807 err_dma:
2808         pci_disable_device(pdev);
2809         return err;
2810 }
2811
2812 /**
2813  * igbvf_remove - Device Removal Routine
2814  * @pdev: PCI device information struct
2815  *
2816  * igbvf_remove is called by the PCI subsystem to alert the driver
2817  * that it should release a PCI device.  The could be caused by a
2818  * Hot-Plug event, or because the driver is going to be removed from
2819  * memory.
2820  **/
2821 static void __devexit igbvf_remove(struct pci_dev *pdev)
2822 {
2823         struct net_device *netdev = pci_get_drvdata(pdev);
2824         struct igbvf_adapter *adapter = netdev_priv(netdev);
2825         struct e1000_hw *hw = &adapter->hw;
2826
2827         /*
2828          * The watchdog timer may be rescheduled, so explicitly
2829          * disable it from being rescheduled.
2830          */
2831         set_bit(__IGBVF_DOWN, &adapter->state);
2832         del_timer_sync(&adapter->watchdog_timer);
2833
2834         cancel_work_sync(&adapter->reset_task);
2835         cancel_work_sync(&adapter->watchdog_task);
2836
2837         unregister_netdev(netdev);
2838
2839         igbvf_reset_interrupt_capability(adapter);
2840
2841         /*
2842          * it is important to delete the napi struct prior to freeing the
2843          * rx ring so that you do not end up with null pointer refs
2844          */
2845         netif_napi_del(&adapter->rx_ring->napi);
2846         kfree(adapter->tx_ring);
2847         kfree(adapter->rx_ring);
2848
2849         iounmap(hw->hw_addr);
2850         if (hw->flash_address)
2851                 iounmap(hw->flash_address);
2852         pci_release_regions(pdev);
2853
2854         free_netdev(netdev);
2855
2856         pci_disable_device(pdev);
2857 }
2858
2859 /* PCI Error Recovery (ERS) */
2860 static struct pci_error_handlers igbvf_err_handler = {
2861         .error_detected = igbvf_io_error_detected,
2862         .slot_reset = igbvf_io_slot_reset,
2863         .resume = igbvf_io_resume,
2864 };
2865
2866 static DEFINE_PCI_DEVICE_TABLE(igbvf_pci_tbl) = {
2867         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82576_VF), board_vf },
2868         { } /* terminate list */
2869 };
2870 MODULE_DEVICE_TABLE(pci, igbvf_pci_tbl);
2871
2872 /* PCI Device API Driver */
2873 static struct pci_driver igbvf_driver = {
2874         .name     = igbvf_driver_name,
2875         .id_table = igbvf_pci_tbl,
2876         .probe    = igbvf_probe,
2877         .remove   = __devexit_p(igbvf_remove),
2878 #ifdef CONFIG_PM
2879         /* Power Management Hooks */
2880         .suspend  = igbvf_suspend,
2881         .resume   = igbvf_resume,
2882 #endif
2883         .shutdown = igbvf_shutdown,
2884         .err_handler = &igbvf_err_handler
2885 };
2886
2887 /**
2888  * igbvf_init_module - Driver Registration Routine
2889  *
2890  * igbvf_init_module is the first routine called when the driver is
2891  * loaded. All it does is register with the PCI subsystem.
2892  **/
2893 static int __init igbvf_init_module(void)
2894 {
2895         int ret;
2896         printk(KERN_INFO "%s - version %s\n",
2897                igbvf_driver_string, igbvf_driver_version);
2898         printk(KERN_INFO "%s\n", igbvf_copyright);
2899
2900         ret = pci_register_driver(&igbvf_driver);
2901
2902         return ret;
2903 }
2904 module_init(igbvf_init_module);
2905
2906 /**
2907  * igbvf_exit_module - Driver Exit Cleanup Routine
2908  *
2909  * igbvf_exit_module is called just before the driver is removed
2910  * from memory.
2911  **/
2912 static void __exit igbvf_exit_module(void)
2913 {
2914         pci_unregister_driver(&igbvf_driver);
2915 }
2916 module_exit(igbvf_exit_module);
2917
2918
2919 MODULE_AUTHOR("Intel Corporation, <e1000-devel@lists.sourceforge.net>");
2920 MODULE_DESCRIPTION("Intel(R) 82576 Virtual Function Network Driver");
2921 MODULE_LICENSE("GPL");
2922 MODULE_VERSION(DRV_VERSION);
2923
2924 /* netdev.c */