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 = vmalloc(size);
 434         if (!tx_ring->buffer_info)
 435                 goto err;
 436         memset(tx_ring->buffer_info, 0, size);
 437
 438         /* round up to nearest 4K */
 439         tx_ring->size = tx_ring->count * sizeof(union e1000_adv_tx_desc);
 440         tx_ring->size = ALIGN(tx_ring->size, 4096);
 441
 442         tx_ring->desc = dma_alloc_coherent(&pdev->dev, tx_ring->size,
 443                                            &tx_ring->dma, GFP_KERNEL);
 444
 445         if (!tx_ring->desc)
 446                 goto err;
 447
 448         tx_ring->adapter = adapter;
 449         tx_ring->next_to_use = 0;
 450         tx_ring->next_to_clean = 0;
 451
 452         return 0;
 453 err:
 454         vfree(tx_ring->buffer_info);
 455         dev_err(&adapter->pdev->dev,
 456                 "Unable to allocate memory for the transmit descriptor ring\n");
 457         return -ENOMEM;
 458 }
 459
 460 /**
 461  * igbvf_setup_rx_resources - allocate Rx resources (Descriptors)
 462  * @adapter: board private structure
 463  *
 464  * Returns 0 on success, negative on failure
 465  **/
 466 int igbvf_setup_rx_resources(struct igbvf_adapter *adapter,
 467                              struct igbvf_ring *rx_ring)
 468 {
 469         struct pci_dev *pdev = adapter->pdev;
 470         int size, desc_len;
 471
 472         size = sizeof(struct igbvf_buffer) * rx_ring->count;
 473         rx_ring->buffer_info = vmalloc(size);
 474         if (!rx_ring->buffer_info)
 475                 goto err;
 476         memset(rx_ring->buffer_info, 0, size);
 477
 478         desc_len = sizeof(union e1000_adv_rx_desc);
 479
 480         /* Round up to nearest 4K */
 481         rx_ring->size = rx_ring->count * desc_len;
 482         rx_ring->size = ALIGN(rx_ring->size, 4096);
 483
 484         rx_ring->desc = dma_alloc_coherent(&pdev->dev, rx_ring->size,
 485                                            &rx_ring->dma, GFP_KERNEL);
 486
 487         if (!rx_ring->desc)
 488                 goto err;
 489
 490         rx_ring->next_to_clean = 0;
 491         rx_ring->next_to_use = 0;
 492
 493         rx_ring->adapter = adapter;
 494
 495         return 0;
 496
 497 err:
 498         vfree(rx_ring->buffer_info);
 499         rx_ring->buffer_info = NULL;
 500         dev_err(&adapter->pdev->dev,
 501                 "Unable to allocate memory for the receive descriptor ring\n");
 502         return -ENOMEM;
 503 }
 504
 505 /**
 506  * igbvf_clean_tx_ring - Free Tx Buffers
 507  * @tx_ring: ring to be cleaned
 508  **/
 509 static void igbvf_clean_tx_ring(struct igbvf_ring *tx_ring)
 510 {
 511         struct igbvf_adapter *adapter = tx_ring->adapter;
 512         struct igbvf_buffer *buffer_info;
 513         unsigned long size;
 514         unsigned int i;
 515
 516         if (!tx_ring->buffer_info)
 517                 return;
 518
 519         /* Free all the Tx ring sk_buffs */
 520         for (i = 0; i < tx_ring->count; i++) {
 521                 buffer_info = &tx_ring->buffer_info[i];
 522                 igbvf_put_txbuf(adapter, buffer_info);
 523         }
 524
 525         size = sizeof(struct igbvf_buffer) * tx_ring->count;
 526         memset(tx_ring->buffer_info, 0, size);
 527
 528         /* Zero out the descriptor ring */
 529         memset(tx_ring->desc, 0, tx_ring->size);
 530
 531         tx_ring->next_to_use = 0;
 532         tx_ring->next_to_clean = 0;
 533
 534         writel(0, adapter->hw.hw_addr + tx_ring->head);
 535         writel(0, adapter->hw.hw_addr + tx_ring->tail);
 536 }
 537
 538 /**
 539  * igbvf_free_tx_resources - Free Tx Resources per Queue
 540  * @tx_ring: ring to free resources from
 541  *
 542  * Free all transmit software resources
 543  **/
 544 void igbvf_free_tx_resources(struct igbvf_ring *tx_ring)
 545 {
 546         struct pci_dev *pdev = tx_ring->adapter->pdev;
 547
 548         igbvf_clean_tx_ring(tx_ring);
 549
 550         vfree(tx_ring->buffer_info);
 551         tx_ring->buffer_info = NULL;
 552
 553         dma_free_coherent(&pdev->dev, tx_ring->size, tx_ring->desc,
 554                           tx_ring->dma);
 555
 556         tx_ring->desc = NULL;
 557 }
 558
 559 /**
 560  * igbvf_clean_rx_ring - Free Rx Buffers per Queue
 561  * @adapter: board private structure
 562  **/
 563 static void igbvf_clean_rx_ring(struct igbvf_ring *rx_ring)
 564 {
 565         struct igbvf_adapter *adapter = rx_ring->adapter;
 566         struct igbvf_buffer *buffer_info;
 567         struct pci_dev *pdev = adapter->pdev;
 568         unsigned long size;
 569         unsigned int i;
 570
 571         if (!rx_ring->buffer_info)
 572                 return;
 573
 574         /* Free all the Rx ring sk_buffs */
 575         for (i = 0; i < rx_ring->count; i++) {
 576                 buffer_info = &rx_ring->buffer_info[i];
 577                 if (buffer_info->dma) {
 578                         if (adapter->rx_ps_hdr_size){
 579                                 dma_unmap_single(&pdev->dev, buffer_info->dma,
 580                                                  adapter->rx_ps_hdr_size,
 581                                                  DMA_FROM_DEVICE);
 582                         } else {
 583                                 dma_unmap_single(&pdev->dev, buffer_info->dma,
 584                                                  adapter->rx_buffer_len,
 585                                                  DMA_FROM_DEVICE);
 586                         }
 587                         buffer_info->dma = 0;
 588                 }
 589
 590                 if (buffer_info->skb) {
 591                         dev_kfree_skb(buffer_info->skb);
 592                         buffer_info->skb = NULL;
 593                 }
 594
 595                 if (buffer_info->page) {
 596                         if (buffer_info->page_dma)
 597                                 dma_unmap_page(&pdev->dev,
 598                                                buffer_info->page_dma,
 599                                                PAGE_SIZE / 2,
 600                                                DMA_FROM_DEVICE);
 601                         put_page(buffer_info->page);
 602                         buffer_info->page = NULL;
 603                         buffer_info->page_dma = 0;
 604                         buffer_info->page_offset = 0;
 605                 }
 606         }
 607
 608         size = sizeof(struct igbvf_buffer) * rx_ring->count;
 609         memset(rx_ring->buffer_info, 0, size);
 610
 611         /* Zero out the descriptor ring */
 612         memset(rx_ring->desc, 0, rx_ring->size);
 613
 614         rx_ring->next_to_clean = 0;
 615         rx_ring->next_to_use = 0;
 616
 617         writel(0, adapter->hw.hw_addr + rx_ring->head);
 618         writel(0, adapter->hw.hw_addr + rx_ring->tail);
 619 }
 620
 621 /**
 622  * igbvf_free_rx_resources - Free Rx Resources
 623  * @rx_ring: ring to clean the resources from
 624  *
 625  * Free all receive software resources
 626  **/
 627
 628 void igbvf_free_rx_resources(struct igbvf_ring *rx_ring)
 629 {
 630         struct pci_dev *pdev = rx_ring->adapter->pdev;
 631
 632         igbvf_clean_rx_ring(rx_ring);
 633
 634         vfree(rx_ring->buffer_info);
 635         rx_ring->buffer_info = NULL;
 636
 637         dma_free_coherent(&pdev->dev, rx_ring->size, rx_ring->desc,
 638                           rx_ring->dma);
 639         rx_ring->desc = NULL;
 640 }
 641
 642 /**
 643  * igbvf_update_itr - update the dynamic ITR value based on statistics
 644  * @adapter: pointer to adapter
 645  * @itr_setting: current adapter->itr
 646  * @packets: the number of packets during this measurement interval
 647  * @bytes: the number of bytes during this measurement interval
 648  *
 649  *      Stores a new ITR value based on packets and byte
 650  *      counts during the last interrupt.  The advantage of per interrupt
 651  *      computation is faster updates and more accurate ITR for the current
 652  *      traffic pattern.  Constants in this function were computed
 653  *      based on theoretical maximum wire speed and thresholds were set based
 654  *      on testing data as well as attempting to minimize response time
 655  *      while increasing bulk throughput.  This functionality is controlled
 656  *      by the InterruptThrottleRate module parameter.
 657  **/
 658 static unsigned int igbvf_update_itr(struct igbvf_adapter *adapter,
 659                                      u16 itr_setting, int packets,
 660                                      int bytes)
 661 {
 662         unsigned int retval = itr_setting;
 663
 664         if (packets == 0)
 665                 goto update_itr_done;
 666
 667         switch (itr_setting) {
 668         case lowest_latency:
 669                 /* handle TSO and jumbo frames */
 670                 if (bytes/packets > 8000)
 671                         retval = bulk_latency;
 672                 else if ((packets < 5) && (bytes > 512))
 673                         retval = low_latency;
 674                 break;
 675         case low_latency:  /* 50 usec aka 20000 ints/s */
 676                 if (bytes > 10000) {
 677                         /* this if handles the TSO accounting */
 678                         if (bytes/packets > 8000)
 679                                 retval = bulk_latency;
 680                         else if ((packets < 10) || ((bytes/packets) > 1200))
 681                                 retval = bulk_latency;
 682                         else if ((packets > 35))
 683                                 retval = lowest_latency;
 684                 } else if (bytes/packets > 2000) {
 685                         retval = bulk_latency;
 686                 } else if (packets <= 2 && bytes < 512) {
 687                         retval = lowest_latency;
 688                 }
 689                 break;
 690         case bulk_latency: /* 250 usec aka 4000 ints/s */
 691                 if (bytes > 25000) {
 692                         if (packets > 35)
 693                                 retval = low_latency;
 694                 } else if (bytes < 6000) {
 695                         retval = low_latency;
 696                 }
 697                 break;
 698         }
 699
 700 update_itr_done:
 701         return retval;
 702 }
 703
 704 static void igbvf_set_itr(struct igbvf_adapter *adapter)
 705 {
 706         struct e1000_hw *hw = &adapter->hw;
 707         u16 current_itr;
 708         u32 new_itr = adapter->itr;
 709
 710         adapter->tx_itr = igbvf_update_itr(adapter, adapter->tx_itr,
 711                                            adapter->total_tx_packets,
 712                                            adapter->total_tx_bytes);
 713         /* conservative mode (itr 3) eliminates the lowest_latency setting */
 714         if (adapter->itr_setting == 3 && adapter->tx_itr == lowest_latency)
 715                 adapter->tx_itr = low_latency;
 716
 717         adapter->rx_itr = igbvf_update_itr(adapter, adapter->rx_itr,
 718                                            adapter->total_rx_packets,
 719                                            adapter->total_rx_bytes);
 720         /* conservative mode (itr 3) eliminates the lowest_latency setting */
 721         if (adapter->itr_setting == 3 && adapter->rx_itr == lowest_latency)
 722                 adapter->rx_itr = low_latency;
 723
 724         current_itr = max(adapter->rx_itr, adapter->tx_itr);
 725
 726         switch (current_itr) {
 727         /* counts and packets in update_itr are dependent on these numbers */
 728         case lowest_latency:
 729                 new_itr = 70000;
 730                 break;
 731         case low_latency:
 732                 new_itr = 20000; /* aka hwitr = ~200 */
 733                 break;
 734         case bulk_latency:
 735                 new_itr = 4000;
 736                 break;
 737         default:
 738                 break;
 739         }
 740
 741         if (new_itr != adapter->itr) {
 742                 /*
 743                  * this attempts to bias the interrupt rate towards Bulk
 744                  * by adding intermediate steps when interrupt rate is
 745                  * increasing
 746                  */
 747                 new_itr = new_itr > adapter->itr ?
 748                              min(adapter->itr + (new_itr >> 2), new_itr) :
 749                              new_itr;
 750                 adapter->itr = new_itr;
 751                 adapter->rx_ring->itr_val = 1952;
 752
 753                 if (adapter->msix_entries)
 754                         adapter->rx_ring->set_itr = 1;
 755                 else
 756                         ew32(ITR, 1952);
 757         }
 758 }
 759
 760 /**
 761  * igbvf_clean_tx_irq - Reclaim resources after transmit completes
 762  * @adapter: board private structure
 763  * returns true if ring is completely cleaned
 764  **/
 765 static bool igbvf_clean_tx_irq(struct igbvf_ring *tx_ring)
 766 {
 767         struct igbvf_adapter *adapter = tx_ring->adapter;
 768         struct e1000_hw *hw = &adapter->hw;
 769         struct net_device *netdev = adapter->netdev;
 770         struct igbvf_buffer *buffer_info;
 771         struct sk_buff *skb;
 772         union e1000_adv_tx_desc *tx_desc, *eop_desc;
 773         unsigned int total_bytes = 0, total_packets = 0;
 774         unsigned int i, eop, count = 0;
 775         bool cleaned = false;
 776
 777         i = tx_ring->next_to_clean;
 778         eop = tx_ring->buffer_info[i].next_to_watch;
 779         eop_desc = IGBVF_TX_DESC_ADV(*tx_ring, eop);
 780
 781         while ((eop_desc->wb.status & cpu_to_le32(E1000_TXD_STAT_DD)) &&
 782                (count < tx_ring->count)) {
 783                 rmb();  /* read buffer_info after eop_desc status */
 784                 for (cleaned = false; !cleaned; count++) {
 785                         tx_desc = IGBVF_TX_DESC_ADV(*tx_ring, i);
 786                         buffer_info = &tx_ring->buffer_info[i];
 787                         cleaned = (i == eop);
 788                         skb = buffer_info->skb;
 789
 790                         if (skb) {
 791                                 unsigned int segs, bytecount;
 792
 793                                 /* gso_segs is currently only valid for tcp */
 794                                 segs = skb_shinfo(skb)->gso_segs ?: 1;
 795                                 /* multiply data chunks by size of headers */
 796                                 bytecount = ((segs - 1) * skb_headlen(skb)) +
 797                                             skb->len;
 798                                 total_packets += segs;
 799                                 total_bytes += bytecount;
 800                         }
 801
 802                         igbvf_put_txbuf(adapter, buffer_info);
 803                         tx_desc->wb.status = 0;
 804
 805                         i++;
 806                         if (i == tx_ring->count)
 807                                 i = 0;
 808                 }
 809                 eop = tx_ring->buffer_info[i].next_to_watch;
 810                 eop_desc = IGBVF_TX_DESC_ADV(*tx_ring, eop);
 811         }
 812
 813         tx_ring->next_to_clean = i;
 814
 815         if (unlikely(count &&
 816                      netif_carrier_ok(netdev) &&
 817                      igbvf_desc_unused(tx_ring) >= IGBVF_TX_QUEUE_WAKE)) {
 818                 /* Make sure that anybody stopping the queue after this
 819                  * sees the new next_to_clean.
 820                  */
 821                 smp_mb();
 822                 if (netif_queue_stopped(netdev) &&
 823                     !(test_bit(__IGBVF_DOWN, &adapter->state))) {
 824                         netif_wake_queue(netdev);
 825                         ++adapter->restart_queue;
 826                 }
 827         }
 828
 829         if (adapter->detect_tx_hung) {
 830                 /* Detect a transmit hang in hardware, this serializes the
 831                  * check with the clearing of time_stamp and movement of i */
 832                 adapter->detect_tx_hung = false;
 833                 if (tx_ring->buffer_info[i].time_stamp &&
 834                     time_after(jiffies, tx_ring->buffer_info[i].time_stamp +
 835                                (adapter->tx_timeout_factor * HZ)) &&
 836                     !(er32(STATUS) & E1000_STATUS_TXOFF)) {
 837
 838                         tx_desc = IGBVF_TX_DESC_ADV(*tx_ring, i);
 839                         /* detected Tx unit hang */
 840                         igbvf_print_tx_hang(adapter);
 841
 842                         netif_stop_queue(netdev);
 843                 }
 844         }
 845         adapter->net_stats.tx_bytes += total_bytes;
 846         adapter->net_stats.tx_packets += total_packets;
 847         return count < tx_ring->count;
 848 }
 849
 850 static irqreturn_t igbvf_msix_other(int irq, void *data)
 851 {
 852         struct net_device *netdev = data;
 853         struct igbvf_adapter *adapter = netdev_priv(netdev);
 854         struct e1000_hw *hw = &adapter->hw;
 855
 856         adapter->int_counter1++;
 857
 858         netif_carrier_off(netdev);
 859         hw->mac.get_link_status = 1;
 860         if (!test_bit(__IGBVF_DOWN, &adapter->state))
 861                 mod_timer(&adapter->watchdog_timer, jiffies + 1);
 862
 863         ew32(EIMS, adapter->eims_other);
 864
 865         return IRQ_HANDLED;
 866 }
 867
 868 static irqreturn_t igbvf_intr_msix_tx(int irq, void *data)
 869 {
 870         struct net_device *netdev = data;
 871         struct igbvf_adapter *adapter = netdev_priv(netdev);
 872         struct e1000_hw *hw = &adapter->hw;
 873         struct igbvf_ring *tx_ring = adapter->tx_ring;
 874
 875
 876         adapter->total_tx_bytes = 0;
 877         adapter->total_tx_packets = 0;
 878
 879         /* auto mask will automatically reenable the interrupt when we write
 880          * EICS */
 881         if (!igbvf_clean_tx_irq(tx_ring))
 882                 /* Ring was not completely cleaned, so fire another interrupt */
 883                 ew32(EICS, tx_ring->eims_value);
 884         else
 885                 ew32(EIMS, tx_ring->eims_value);
 886
 887         return IRQ_HANDLED;
 888 }
 889
 890 static irqreturn_t igbvf_intr_msix_rx(int irq, void *data)
 891 {
 892         struct net_device *netdev = data;
 893         struct igbvf_adapter *adapter = netdev_priv(netdev);
 894
 895         adapter->int_counter0++;
 896
 897         /* Write the ITR value calculated at the end of the
 898          * previous interrupt.
 899          */
 900         if (adapter->rx_ring->set_itr) {
 901                 writel(adapter->rx_ring->itr_val,
 902                        adapter->hw.hw_addr + adapter->rx_ring->itr_register);
 903                 adapter->rx_ring->set_itr = 0;
 904         }
 905
 906         if (napi_schedule_prep(&adapter->rx_ring->napi)) {
 907                 adapter->total_rx_bytes = 0;
 908                 adapter->total_rx_packets = 0;
 909                 __napi_schedule(&adapter->rx_ring->napi);
 910         }
 911
 912         return IRQ_HANDLED;
 913 }
 914
 915 #define IGBVF_NO_QUEUE -1
 916
 917 static void igbvf_assign_vector(struct igbvf_adapter *adapter, int rx_queue,
 918                                 int tx_queue, int msix_vector)
 919 {
 920         struct e1000_hw *hw = &adapter->hw;
 921         u32 ivar, index;
 922
 923         /* 82576 uses a table-based method for assigning vectors.
 924            Each queue has a single entry in the table to which we write
 925            a vector number along with a "valid" bit.  Sadly, the layout
 926            of the table is somewhat counterintuitive. */
 927         if (rx_queue > IGBVF_NO_QUEUE) {
 928                 index = (rx_queue >> 1);
 929                 ivar = array_er32(IVAR0, index);
 930                 if (rx_queue & 0x1) {
 931                         /* vector goes into third byte of register */
 932                         ivar = ivar & 0xFF00FFFF;
 933                         ivar |= (msix_vector | E1000_IVAR_VALID) << 16;
 934                 } else {
 935                         /* vector goes into low byte of register */
 936                         ivar = ivar & 0xFFFFFF00;
 937                         ivar |= msix_vector | E1000_IVAR_VALID;
 938                 }
 939                 adapter->rx_ring[rx_queue].eims_value = 1 << msix_vector;
 940                 array_ew32(IVAR0, index, ivar);
 941         }
 942         if (tx_queue > IGBVF_NO_QUEUE) {
 943                 index = (tx_queue >> 1);
 944                 ivar = array_er32(IVAR0, index);
 945                 if (tx_queue & 0x1) {
 946                         /* vector goes into high byte of register */
 947                         ivar = ivar & 0x00FFFFFF;
 948                         ivar |= (msix_vector | E1000_IVAR_VALID) << 24;
 949                 } else {
 950                         /* vector goes into second byte of register */
 951                         ivar = ivar & 0xFFFF00FF;
 952                         ivar |= (msix_vector | E1000_IVAR_VALID) << 8;
 953                 }
 954                 adapter->tx_ring[tx_queue].eims_value = 1 << msix_vector;
 955                 array_ew32(IVAR0, index, ivar);
 956         }
 957 }
 958
 959 /**
 960  * igbvf_configure_msix - Configure MSI-X hardware
 961  *
 962  * igbvf_configure_msix sets up the hardware to properly
 963  * generate MSI-X interrupts.
 964  **/
 965 static void igbvf_configure_msix(struct igbvf_adapter *adapter)
 966 {
 967         u32 tmp;
 968         struct e1000_hw *hw = &adapter->hw;
 969         struct igbvf_ring *tx_ring = adapter->tx_ring;
 970         struct igbvf_ring *rx_ring = adapter->rx_ring;
 971         int vector = 0;
 972
 973         adapter->eims_enable_mask = 0;
 974
 975         igbvf_assign_vector(adapter, IGBVF_NO_QUEUE, 0, vector++);
 976         adapter->eims_enable_mask |= tx_ring->eims_value;
 977         if (tx_ring->itr_val)
 978                 writel(tx_ring->itr_val,
 979                        hw->hw_addr + tx_ring->itr_register);
 980         else
 981                 writel(1952, hw->hw_addr + tx_ring->itr_register);
 982
 983         igbvf_assign_vector(adapter, 0, IGBVF_NO_QUEUE, vector++);
 984         adapter->eims_enable_mask |= rx_ring->eims_value;
 985         if (rx_ring->itr_val)
 986                 writel(rx_ring->itr_val,
 987                        hw->hw_addr + rx_ring->itr_register);
 988         else
 989                 writel(1952, hw->hw_addr + rx_ring->itr_register);
 990
 991         /* set vector for other causes, i.e. link changes */
 992
 993         tmp = (vector++ | E1000_IVAR_VALID);
 994
 995         ew32(IVAR_MISC, tmp);
 996
 997         adapter->eims_enable_mask = (1 << (vector)) - 1;
 998         adapter->eims_other = 1 << (vector - 1);
 999         e1e_flush();
1000 }
1001
1002 static void igbvf_reset_interrupt_capability(struct igbvf_adapter *adapter)
1003 {
1004         if (adapter->msix_entries) {
1005                 pci_disable_msix(adapter->pdev);
1006                 kfree(adapter->msix_entries);
1007                 adapter->msix_entries = NULL;
1008         }
1009 }
1010
1011 /**
1012  * igbvf_set_interrupt_capability - set MSI or MSI-X if supported
1013  *
1014  * Attempt to configure interrupts using the best available
1015  * capabilities of the hardware and kernel.
1016  **/
1017 static void igbvf_set_interrupt_capability(struct igbvf_adapter *adapter)
1018 {
1019         int err = -ENOMEM;
1020         int i;
1021
1022         /* we allocate 3 vectors, 1 for tx, 1 for rx, one for pf messages */
1023         adapter->msix_entries = kcalloc(3, sizeof(struct msix_entry),
1024                                         GFP_KERNEL);
1025         if (adapter->msix_entries) {
1026                 for (i = 0; i < 3; i++)
1027                         adapter->msix_entries[i].entry = i;
1028
1029                 err = pci_enable_msix(adapter->pdev,
1030                                       adapter->msix_entries, 3);
1031         }
1032
1033         if (err) {
1034                 /* MSI-X failed */
1035                 dev_err(&adapter->pdev->dev,
1036                         "Failed to initialize MSI-X interrupts.\n");
1037                 igbvf_reset_interrupt_capability(adapter);
1038         }
1039 }
1040
1041 /**
1042  * igbvf_request_msix - Initialize MSI-X interrupts
1043  *
1044  * igbvf_request_msix allocates MSI-X vectors and requests interrupts from the
1045  * kernel.
1046  **/
1047 static int igbvf_request_msix(struct igbvf_adapter *adapter)
1048 {
1049         struct net_device *netdev = adapter->netdev;
1050         int err = 0, vector = 0;
1051
1052         if (strlen(netdev->name) < (IFNAMSIZ - 5)) {
1053                 sprintf(adapter->tx_ring->name, "%s-tx-0", netdev->name);
1054                 sprintf(adapter->rx_ring->name, "%s-rx-0", netdev->name);
1055         } else {
1056                 memcpy(adapter->tx_ring->name, netdev->name, IFNAMSIZ);
1057                 memcpy(adapter->rx_ring->name, netdev->name, IFNAMSIZ);
1058         }
1059
1060         err = request_irq(adapter->msix_entries[vector].vector,
1061                           igbvf_intr_msix_tx, 0, adapter->tx_ring->name,
1062                           netdev);
1063         if (err)
1064                 goto out;
1065
1066         adapter->tx_ring->itr_register = E1000_EITR(vector);
1067         adapter->tx_ring->itr_val = 1952;
1068         vector++;
1069
1070         err = request_irq(adapter->msix_entries[vector].vector,
1071                           igbvf_intr_msix_rx, 0, adapter->rx_ring->name,
1072                           netdev);
1073         if (err)
1074                 goto out;
1075
1076         adapter->rx_ring->itr_register = E1000_EITR(vector);
1077         adapter->rx_ring->itr_val = 1952;
1078         vector++;
1079
1080         err = request_irq(adapter->msix_entries[vector].vector,
1081                           igbvf_msix_other, 0, netdev->name, netdev);
1082         if (err)
1083                 goto out;
1084
1085         igbvf_configure_msix(adapter);
1086         return 0;
1087 out:
1088         return err;
1089 }
1090
1091 /**
1092  * igbvf_alloc_queues - Allocate memory for all rings
1093  * @adapter: board private structure to initialize
1094  **/
1095 static int __devinit igbvf_alloc_queues(struct igbvf_adapter *adapter)
1096 {
1097         struct net_device *netdev = adapter->netdev;
1098
1099         adapter->tx_ring = kzalloc(sizeof(struct igbvf_ring), GFP_KERNEL);
1100         if (!adapter->tx_ring)
1101                 return -ENOMEM;
1102
1103         adapter->rx_ring = kzalloc(sizeof(struct igbvf_ring), GFP_KERNEL);
1104         if (!adapter->rx_ring) {
1105                 kfree(adapter->tx_ring);
1106                 return -ENOMEM;
1107         }
1108
1109         netif_napi_add(netdev, &adapter->rx_ring->napi, igbvf_poll, 64);
1110
1111         return 0;
1112 }
1113
1114 /**
1115  * igbvf_request_irq - initialize interrupts
1116  *
1117  * Attempts to configure interrupts using the best available
1118  * capabilities of the hardware and kernel.
1119  **/
1120 static int igbvf_request_irq(struct igbvf_adapter *adapter)
1121 {
1122         int err = -1;
1123
1124         /* igbvf supports msi-x only */
1125         if (adapter->msix_entries)
1126                 err = igbvf_request_msix(adapter);
1127
1128         if (!err)
1129                 return err;
1130
1131         dev_err(&adapter->pdev->dev,
1132                 "Unable to allocate interrupt, Error: %d\n", err);
1133
1134         return err;
1135 }
1136
1137 static void igbvf_free_irq(struct igbvf_adapter *adapter)
1138 {
1139         struct net_device *netdev = adapter->netdev;
1140         int vector;
1141
1142         if (adapter->msix_entries) {
1143                 for (vector = 0; vector < 3; vector++)
1144                         free_irq(adapter->msix_entries[vector].vector, netdev);
1145         }
1146 }
1147
1148 /**
1149  * igbvf_irq_disable - Mask off interrupt generation on the NIC
1150  **/
1151 static void igbvf_irq_disable(struct igbvf_adapter *adapter)
1152 {
1153         struct e1000_hw *hw = &adapter->hw;
1154
1155         ew32(EIMC, ~0);
1156
1157         if (adapter->msix_entries)
1158                 ew32(EIAC, 0);
1159 }
1160
1161 /**
1162  * igbvf_irq_enable - Enable default interrupt generation settings
1163  **/
1164 static void igbvf_irq_enable(struct igbvf_adapter *adapter)
1165 {
1166         struct e1000_hw *hw = &adapter->hw;
1167
1168         ew32(EIAC, adapter->eims_enable_mask);
1169         ew32(EIAM, adapter->eims_enable_mask);
1170         ew32(EIMS, adapter->eims_enable_mask);
1171 }
1172
1173 /**
1174  * igbvf_poll - NAPI Rx polling callback
1175  * @napi: struct associated with this polling callback
1176  * @budget: amount of packets driver is allowed to process this poll
1177  **/
1178 static int igbvf_poll(struct napi_struct *napi, int budget)
1179 {
1180         struct igbvf_ring *rx_ring = container_of(napi, struct igbvf_ring, napi);
1181         struct igbvf_adapter *adapter = rx_ring->adapter;
1182         struct e1000_hw *hw = &adapter->hw;
1183         int work_done = 0;
1184
1185         igbvf_clean_rx_irq(adapter, &work_done, budget);
1186
1187         /* If not enough Rx work done, exit the polling mode */
1188         if (work_done < budget) {
1189                 napi_complete(napi);
1190
1191                 if (adapter->itr_setting & 3)
1192                         igbvf_set_itr(adapter);
1193
1194                 if (!test_bit(__IGBVF_DOWN, &adapter->state))
1195                         ew32(EIMS, adapter->rx_ring->eims_value);
1196         }
1197
1198         return work_done;
1199 }
1200
1201 /**
1202  * igbvf_set_rlpml - set receive large packet maximum length
1203  * @adapter: board private structure
1204  *
1205  * Configure the maximum size of packets that will be received
1206  */
1207 static void igbvf_set_rlpml(struct igbvf_adapter *adapter)
1208 {
1209         int max_frame_size = adapter->max_frame_size;
1210         struct e1000_hw *hw = &adapter->hw;
1211
1212         if (adapter->vlgrp)
1213                 max_frame_size += VLAN_TAG_SIZE;
1214
1215         e1000_rlpml_set_vf(hw, max_frame_size);
1216 }
1217
1218 static void igbvf_vlan_rx_add_vid(struct net_device *netdev, u16 vid)
1219 {
1220         struct igbvf_adapter *adapter = netdev_priv(netdev);
1221         struct e1000_hw *hw = &adapter->hw;
1222
1223         if (hw->mac.ops.set_vfta(hw, vid, true))
1224                 dev_err(&adapter->pdev->dev, "Failed to add vlan id %d\n", vid);
1225 }
1226
1227 static void igbvf_vlan_rx_kill_vid(struct net_device *netdev, u16 vid)
1228 {
1229         struct igbvf_adapter *adapter = netdev_priv(netdev);
1230         struct e1000_hw *hw = &adapter->hw;
1231
1232         igbvf_irq_disable(adapter);
1233         vlan_group_set_device(adapter->vlgrp, vid, NULL);
1234
1235         if (!test_bit(__IGBVF_DOWN, &adapter->state))
1236                 igbvf_irq_enable(adapter);
1237
1238         if (hw->mac.ops.set_vfta(hw, vid, false))
1239                 dev_err(&adapter->pdev->dev,
1240                         "Failed to remove vlan id %d\n", vid);
1241 }
1242
1243 static void igbvf_vlan_rx_register(struct net_device *netdev,
1244                                    struct vlan_group *grp)
1245 {
1246         struct igbvf_adapter *adapter = netdev_priv(netdev);
1247
1248         adapter->vlgrp = grp;
1249 }
1250
1251 static void igbvf_restore_vlan(struct igbvf_adapter *adapter)
1252 {
1253         u16 vid;
1254
1255         if (!adapter->vlgrp)
1256                 return;
1257
1258         for (vid = 0; vid < VLAN_N_VID; vid++) {
1259                 if (!vlan_group_get_device(adapter->vlgrp, vid))
1260                         continue;
1261                 igbvf_vlan_rx_add_vid(adapter->netdev, vid);
1262         }
1263
1264         igbvf_set_rlpml(adapter);
1265 }
1266
1267 /**
1268  * igbvf_configure_tx - Configure Transmit Unit after Reset
1269  * @adapter: board private structure
1270  *
1271  * Configure the Tx unit of the MAC after a reset.
1272  **/
1273 static void igbvf_configure_tx(struct igbvf_adapter *adapter)
1274 {
1275         struct e1000_hw *hw = &adapter->hw;
1276         struct igbvf_ring *tx_ring = adapter->tx_ring;
1277         u64 tdba;
1278         u32 txdctl, dca_txctrl;
1279
1280         /* disable transmits */
1281         txdctl = er32(TXDCTL(0));
1282         ew32(TXDCTL(0), txdctl & ~E1000_TXDCTL_QUEUE_ENABLE);
1283         msleep(10);
1284
1285         /* Setup the HW Tx Head and Tail descriptor pointers */
1286         ew32(TDLEN(0), tx_ring->count * sizeof(union e1000_adv_tx_desc));
1287         tdba = tx_ring->dma;
1288         ew32(TDBAL(0), (tdba & DMA_BIT_MASK(32)));
1289         ew32(TDBAH(0), (tdba >> 32));
1290         ew32(TDH(0), 0);
1291         ew32(TDT(0), 0);
1292         tx_ring->head = E1000_TDH(0);
1293         tx_ring->tail = E1000_TDT(0);
1294
1295         /* Turn off Relaxed Ordering on head write-backs.  The writebacks
1296          * MUST be delivered in order or it will completely screw up
1297          * our bookeeping.
1298          */
1299         dca_txctrl = er32(DCA_TXCTRL(0));
1300         dca_txctrl &= ~E1000_DCA_TXCTRL_TX_WB_RO_EN;
1301         ew32(DCA_TXCTRL(0), dca_txctrl);
1302
1303         /* enable transmits */
1304         txdctl |= E1000_TXDCTL_QUEUE_ENABLE;
1305         ew32(TXDCTL(0), txdctl);
1306
1307         /* Setup Transmit Descriptor Settings for eop descriptor */
1308         adapter->txd_cmd = E1000_ADVTXD_DCMD_EOP | E1000_ADVTXD_DCMD_IFCS;
1309
1310         /* enable Report Status bit */
1311         adapter->txd_cmd |= E1000_ADVTXD_DCMD_RS;
1312 }
1313
1314 /**
1315  * igbvf_setup_srrctl - configure the receive control registers
1316  * @adapter: Board private structure
1317  **/
1318 static void igbvf_setup_srrctl(struct igbvf_adapter *adapter)
1319 {
1320         struct e1000_hw *hw = &adapter->hw;
1321         u32 srrctl = 0;
1322
1323         srrctl &= ~(E1000_SRRCTL_DESCTYPE_MASK |
1324                     E1000_SRRCTL_BSIZEHDR_MASK |
1325                     E1000_SRRCTL_BSIZEPKT_MASK);
1326
1327         /* Enable queue drop to avoid head of line blocking */
1328         srrctl |= E1000_SRRCTL_DROP_EN;
1329
1330         /* Setup buffer sizes */
1331         srrctl |= ALIGN(adapter->rx_buffer_len, 1024) >>
1332                   E1000_SRRCTL_BSIZEPKT_SHIFT;
1333
1334         if (adapter->rx_buffer_len < 2048) {
1335                 adapter->rx_ps_hdr_size = 0;
1336                 srrctl |= E1000_SRRCTL_DESCTYPE_ADV_ONEBUF;
1337         } else {
1338                 adapter->rx_ps_hdr_size = 128;
1339                 srrctl |= adapter->rx_ps_hdr_size <<
1340                           E1000_SRRCTL_BSIZEHDRSIZE_SHIFT;
1341                 srrctl |= E1000_SRRCTL_DESCTYPE_HDR_SPLIT_ALWAYS;
1342         }
1343
1344         ew32(SRRCTL(0), srrctl);
1345 }
1346
1347 /**
1348  * igbvf_configure_rx - Configure Receive Unit after Reset
1349  * @adapter: board private structure
1350  *
1351  * Configure the Rx unit of the MAC after a reset.
1352  **/
1353 static void igbvf_configure_rx(struct igbvf_adapter *adapter)
1354 {
1355         struct e1000_hw *hw = &adapter->hw;
1356         struct igbvf_ring *rx_ring = adapter->rx_ring;
1357         u64 rdba;
1358         u32 rdlen, rxdctl;
1359
1360         /* disable receives */
1361         rxdctl = er32(RXDCTL(0));
1362         ew32(RXDCTL(0), rxdctl & ~E1000_RXDCTL_QUEUE_ENABLE);
1363         msleep(10);
1364
1365         rdlen = rx_ring->count * sizeof(union e1000_adv_rx_desc);
1366
1367         /*
1368          * Setup the HW Rx Head and Tail Descriptor Pointers and
1369          * the Base and Length of the Rx Descriptor Ring
1370          */
1371         rdba = rx_ring->dma;
1372         ew32(RDBAL(0), (rdba & DMA_BIT_MASK(32)));
1373         ew32(RDBAH(0), (rdba >> 32));
1374         ew32(RDLEN(0), rx_ring->count * sizeof(union e1000_adv_rx_desc));
1375         rx_ring->head = E1000_RDH(0);
1376         rx_ring->tail = E1000_RDT(0);
1377         ew32(RDH(0), 0);
1378         ew32(RDT(0), 0);
1379
1380         rxdctl |= E1000_RXDCTL_QUEUE_ENABLE;
1381         rxdctl &= 0xFFF00000;
1382         rxdctl |= IGBVF_RX_PTHRESH;
1383         rxdctl |= IGBVF_RX_HTHRESH << 8;
1384         rxdctl |= IGBVF_RX_WTHRESH << 16;
1385
1386         igbvf_set_rlpml(adapter);
1387
1388         /* enable receives */
1389         ew32(RXDCTL(0), rxdctl);
1390 }
1391
1392 /**
1393  * igbvf_set_multi - Multicast and Promiscuous mode set
1394  * @netdev: network interface device structure
1395  *
1396  * The set_multi entry point is called whenever the multicast address
1397  * list or the network interface flags are updated.  This routine is
1398  * responsible for configuring the hardware for proper multicast,
1399  * promiscuous mode, and all-multi behavior.
1400  **/
1401 static void igbvf_set_multi(struct net_device *netdev)
1402 {
1403         struct igbvf_adapter *adapter = netdev_priv(netdev);
1404         struct e1000_hw *hw = &adapter->hw;
1405         struct netdev_hw_addr *ha;
1406         u8  *mta_list = NULL;
1407         int i;
1408
1409         if (!netdev_mc_empty(netdev)) {
1410                 mta_list = kmalloc(netdev_mc_count(netdev) * 6, GFP_ATOMIC);
1411                 if (!mta_list) {
1412                         dev_err(&adapter->pdev->dev,
1413                                 "failed to allocate multicast filter list\n");
1414                         return;
1415                 }
1416         }
1417
1418         /* prepare a packed array of only addresses. */
1419         i = 0;
1420         netdev_for_each_mc_addr(ha, netdev)
1421                 memcpy(mta_list + (i++ * ETH_ALEN), ha->addr, ETH_ALEN);
1422
1423         hw->mac.ops.update_mc_addr_list(hw, mta_list, i, 0, 0);
1424         kfree(mta_list);
1425 }
1426
1427 /**
1428  * igbvf_configure - configure the hardware for Rx and Tx
1429  * @adapter: private board structure
1430  **/
1431 static void igbvf_configure(struct igbvf_adapter *adapter)
1432 {
1433         igbvf_set_multi(adapter->netdev);
1434
1435         igbvf_restore_vlan(adapter);
1436
1437         igbvf_configure_tx(adapter);
1438         igbvf_setup_srrctl(adapter);
1439         igbvf_configure_rx(adapter);
1440         igbvf_alloc_rx_buffers(adapter->rx_ring,
1441                                igbvf_desc_unused(adapter->rx_ring));
1442 }
1443
1444 /* igbvf_reset - bring the hardware into a known good state
1445  *
1446  * This function boots the hardware and enables some settings that
1447  * require a configuration cycle of the hardware - those cannot be
1448  * set/changed during runtime. After reset the device needs to be
1449  * properly configured for Rx, Tx etc.
1450  */
1451 static void igbvf_reset(struct igbvf_adapter *adapter)
1452 {
1453         struct e1000_mac_info *mac = &adapter->hw.mac;
1454         struct net_device *netdev = adapter->netdev;
1455         struct e1000_hw *hw = &adapter->hw;
1456
1457         /* Allow time for pending master requests to run */
1458         if (mac->ops.reset_hw(hw))
1459                 dev_err(&adapter->pdev->dev, "PF still resetting\n");
1460
1461         mac->ops.init_hw(hw);
1462
1463         if (is_valid_ether_addr(adapter->hw.mac.addr)) {
1464                 memcpy(netdev->dev_addr, adapter->hw.mac.addr,
1465                        netdev->addr_len);
1466                 memcpy(netdev->perm_addr, adapter->hw.mac.addr,
1467                        netdev->addr_len);
1468         }
1469
1470         adapter->last_reset = jiffies;
1471 }
1472
1473 int igbvf_up(struct igbvf_adapter *adapter)
1474 {
1475         struct e1000_hw *hw = &adapter->hw;
1476
1477         /* hardware has been reset, we need to reload some things */
1478         igbvf_configure(adapter);
1479
1480         clear_bit(__IGBVF_DOWN, &adapter->state);
1481
1482         napi_enable(&adapter->rx_ring->napi);
1483         if (adapter->msix_entries)
1484                 igbvf_configure_msix(adapter);
1485
1486         /* Clear any pending interrupts. */
1487         er32(EICR);
1488         igbvf_irq_enable(adapter);
1489
1490         /* start the watchdog */
1491         hw->mac.get_link_status = 1;
1492         mod_timer(&adapter->watchdog_timer, jiffies + 1);
1493
1494
1495         return 0;
1496 }
1497
1498 void igbvf_down(struct igbvf_adapter *adapter)
1499 {
1500         struct net_device *netdev = adapter->netdev;
1501         struct e1000_hw *hw = &adapter->hw;
1502         u32 rxdctl, txdctl;
1503
1504         /*
1505          * signal that we're down so the interrupt handler does not
1506          * reschedule our watchdog timer
1507          */
1508         set_bit(__IGBVF_DOWN, &adapter->state);
1509
1510         /* disable receives in the hardware */
1511         rxdctl = er32(RXDCTL(0));
1512         ew32(RXDCTL(0), rxdctl & ~E1000_RXDCTL_QUEUE_ENABLE);
1513
1514         netif_stop_queue(netdev);
1515
1516         /* disable transmits in the hardware */
1517         txdctl = er32(TXDCTL(0));
1518         ew32(TXDCTL(0), txdctl & ~E1000_TXDCTL_QUEUE_ENABLE);
1519
1520         /* flush both disables and wait for them to finish */
1521         e1e_flush();
1522         msleep(10);
1523
1524         napi_disable(&adapter->rx_ring->napi);
1525
1526         igbvf_irq_disable(adapter);
1527
1528         del_timer_sync(&adapter->watchdog_timer);
1529
1530         netif_carrier_off(netdev);
1531
1532         /* record the stats before reset*/
1533         igbvf_update_stats(adapter);
1534
1535         adapter->link_speed = 0;
1536         adapter->link_duplex = 0;
1537
1538         igbvf_reset(adapter);
1539         igbvf_clean_tx_ring(adapter->tx_ring);
1540         igbvf_clean_rx_ring(adapter->rx_ring);
1541 }
1542
1543 void igbvf_reinit_locked(struct igbvf_adapter *adapter)
1544 {
1545         might_sleep();
1546         while (test_and_set_bit(__IGBVF_RESETTING, &adapter->state))
1547                 msleep(1);
1548         igbvf_down(adapter);
1549         igbvf_up(adapter);
1550         clear_bit(__IGBVF_RESETTING, &adapter->state);
1551 }
1552
1553 /**
1554  * igbvf_sw_init - Initialize general software structures (struct igbvf_adapter)
1555  * @adapter: board private structure to initialize
1556  *
1557  * igbvf_sw_init initializes the Adapter private data structure.
1558  * Fields are initialized based on PCI device information and
1559  * OS network device settings (MTU size).
1560  **/
1561 static int __devinit igbvf_sw_init(struct igbvf_adapter *adapter)
1562 {
1563         struct net_device *netdev = adapter->netdev;
1564         s32 rc;
1565
1566         adapter->rx_buffer_len = ETH_FRAME_LEN + VLAN_HLEN + ETH_FCS_LEN;
1567         adapter->rx_ps_hdr_size = 0;
1568         adapter->max_frame_size = netdev->mtu + ETH_HLEN + ETH_FCS_LEN;
1569         adapter->min_frame_size = ETH_ZLEN + ETH_FCS_LEN;
1570
1571         adapter->tx_int_delay = 8;
1572         adapter->tx_abs_int_delay = 32;
1573         adapter->rx_int_delay = 0;
1574         adapter->rx_abs_int_delay = 8;
1575         adapter->itr_setting = 3;
1576         adapter->itr = 20000;
1577
1578         /* Set various function pointers */
1579         adapter->ei->init_ops(&adapter->hw);
1580
1581         rc = adapter->hw.mac.ops.init_params(&adapter->hw);
1582         if (rc)
1583                 return rc;
1584
1585         rc = adapter->hw.mbx.ops.init_params(&adapter->hw);
1586         if (rc)
1587                 return rc;
1588
1589         igbvf_set_interrupt_capability(adapter);
1590
1591         if (igbvf_alloc_queues(adapter))
1592                 return -ENOMEM;
1593
1594         spin_lock_init(&adapter->tx_queue_lock);
1595
1596         /* Explicitly disable IRQ since the NIC can be in any state. */
1597         igbvf_irq_disable(adapter);
1598
1599         spin_lock_init(&adapter->stats_lock);
1600
1601         set_bit(__IGBVF_DOWN, &adapter->state);
1602         return 0;
1603 }
1604
1605 static void igbvf_initialize_last_counter_stats(struct igbvf_adapter *adapter)
1606 {
1607         struct e1000_hw *hw = &adapter->hw;
1608
1609         adapter->stats.last_gprc = er32(VFGPRC);
1610         adapter->stats.last_gorc = er32(VFGORC);
1611         adapter->stats.last_gptc = er32(VFGPTC);
1612         adapter->stats.last_gotc = er32(VFGOTC);
1613         adapter->stats.last_mprc = er32(VFMPRC);
1614         adapter->stats.last_gotlbc = er32(VFGOTLBC);
1615         adapter->stats.last_gptlbc = er32(VFGPTLBC);
1616         adapter->stats.last_gorlbc = er32(VFGORLBC);
1617         adapter->stats.last_gprlbc = er32(VFGPRLBC);
1618
1619         adapter->stats.base_gprc = er32(VFGPRC);
1620         adapter->stats.base_gorc = er32(VFGORC);
1621         adapter->stats.base_gptc = er32(VFGPTC);
1622         adapter->stats.base_gotc = er32(VFGOTC);
1623         adapter->stats.base_mprc = er32(VFMPRC);
1624         adapter->stats.base_gotlbc = er32(VFGOTLBC);
1625         adapter->stats.base_gptlbc = er32(VFGPTLBC);
1626         adapter->stats.base_gorlbc = er32(VFGORLBC);
1627         adapter->stats.base_gprlbc = er32(VFGPRLBC);
1628 }
1629
1630 /**
1631  * igbvf_open - Called when a network interface is made active
1632  * @netdev: network interface device structure
1633  *
1634  * Returns 0 on success, negative value on failure
1635  *
1636  * The open entry point is called when a network interface is made
1637  * active by the system (IFF_UP).  At this point all resources needed
1638  * for transmit and receive operations are allocated, the interrupt
1639  * handler is registered with the OS, the watchdog timer is started,
1640  * and the stack is notified that the interface is ready.
1641  **/
1642 static int igbvf_open(struct net_device *netdev)
1643 {
1644         struct igbvf_adapter *adapter = netdev_priv(netdev);
1645         struct e1000_hw *hw = &adapter->hw;
1646         int err;
1647
1648         /* disallow open during test */
1649         if (test_bit(__IGBVF_TESTING, &adapter->state))
1650                 return -EBUSY;
1651
1652         /* allocate transmit descriptors */
1653         err = igbvf_setup_tx_resources(adapter, adapter->tx_ring);
1654         if (err)
1655                 goto err_setup_tx;
1656
1657         /* allocate receive descriptors */
1658         err = igbvf_setup_rx_resources(adapter, adapter->rx_ring);
1659         if (err)
1660                 goto err_setup_rx;
1661
1662         /*
1663          * before we allocate an interrupt, we must be ready to handle it.
1664          * Setting DEBUG_SHIRQ in the kernel makes it fire an interrupt
1665          * as soon as we call pci_request_irq, so we have to setup our
1666          * clean_rx handler before we do so.
1667          */
1668         igbvf_configure(adapter);
1669
1670         err = igbvf_request_irq(adapter);
1671         if (err)
1672                 goto err_req_irq;
1673
1674         /* From here on the code is the same as igbvf_up() */
1675         clear_bit(__IGBVF_DOWN, &adapter->state);
1676
1677         napi_enable(&adapter->rx_ring->napi);
1678
1679         /* clear any pending interrupts */
1680         er32(EICR);
1681
1682         igbvf_irq_enable(adapter);
1683
1684         /* start the watchdog */
1685         hw->mac.get_link_status = 1;
1686         mod_timer(&adapter->watchdog_timer, jiffies + 1);
1687
1688         return 0;
1689
1690 err_req_irq:
1691         igbvf_free_rx_resources(adapter->rx_ring);
1692 err_setup_rx:
1693         igbvf_free_tx_resources(adapter->tx_ring);
1694 err_setup_tx:
1695         igbvf_reset(adapter);
1696
1697         return err;
1698 }
1699
1700 /**
1701  * igbvf_close - Disables a network interface
1702  * @netdev: network interface device structure
1703  *
1704  * Returns 0, this is not allowed to fail
1705  *
1706  * The close entry point is called when an interface is de-activated
1707  * by the OS.  The hardware is still under the drivers control, but
1708  * needs to be disabled.  A global MAC reset is issued to stop the
1709  * hardware, and all transmit and receive resources are freed.
1710  **/
1711 static int igbvf_close(struct net_device *netdev)
1712 {
1713         struct igbvf_adapter *adapter = netdev_priv(netdev);
1714
1715         WARN_ON(test_bit(__IGBVF_RESETTING, &adapter->state));
1716         igbvf_down(adapter);
1717
1718         igbvf_free_irq(adapter);
1719
1720         igbvf_free_tx_resources(adapter->tx_ring);
1721         igbvf_free_rx_resources(adapter->rx_ring);
1722
1723         return 0;
1724 }
1725 /**
1726  * igbvf_set_mac - Change the Ethernet Address of the NIC
1727  * @netdev: network interface device structure
1728  * @p: pointer to an address structure
1729  *
1730  * Returns 0 on success, negative on failure
1731  **/
1732 static int igbvf_set_mac(struct net_device *netdev, void *p)
1733 {
1734         struct igbvf_adapter *adapter = netdev_priv(netdev);
1735         struct e1000_hw *hw = &adapter->hw;
1736         struct sockaddr *addr = p;
1737
1738         if (!is_valid_ether_addr(addr->sa_data))
1739                 return -EADDRNOTAVAIL;
1740
1741         memcpy(hw->mac.addr, addr->sa_data, netdev->addr_len);
1742
1743         hw->mac.ops.rar_set(hw, hw->mac.addr, 0);
1744
1745         if (memcmp(addr->sa_data, hw->mac.addr, 6))
1746                 return -EADDRNOTAVAIL;
1747
1748         memcpy(netdev->dev_addr, addr->sa_data, netdev->addr_len);
1749
1750         return 0;
1751 }
1752
1753 #define UPDATE_VF_COUNTER(reg, name)                                    \
1754         {                                                               \
1755                 u32 current_counter = er32(reg);                        \
1756                 if (current_counter < adapter->stats.last_##name)       \
1757                         adapter->stats.name += 0x100000000LL;           \
1758                 adapter->stats.last_##name = current_counter;           \
1759                 adapter->stats.name &= 0xFFFFFFFF00000000LL;            \
1760                 adapter->stats.name |= current_counter;                 \
1761         }
1762
1763 /**
1764  * igbvf_update_stats - Update the board statistics counters
1765  * @adapter: board private structure
1766 **/
1767 void igbvf_update_stats(struct igbvf_adapter *adapter)
1768 {
1769         struct e1000_hw *hw = &adapter->hw;
1770         struct pci_dev *pdev = adapter->pdev;
1771
1772         /*
1773          * Prevent stats update while adapter is being reset, link is down
1774          * or if the pci connection is down.
1775          */
1776         if (adapter->link_speed == 0)
1777                 return;
1778
1779         if (test_bit(__IGBVF_RESETTING, &adapter->state))
1780                 return;
1781
1782         if (pci_channel_offline(pdev))
1783                 return;
1784
1785         UPDATE_VF_COUNTER(VFGPRC, gprc);
1786         UPDATE_VF_COUNTER(VFGORC, gorc);
1787         UPDATE_VF_COUNTER(VFGPTC, gptc);
1788         UPDATE_VF_COUNTER(VFGOTC, gotc);
1789         UPDATE_VF_COUNTER(VFMPRC, mprc);
1790         UPDATE_VF_COUNTER(VFGOTLBC, gotlbc);
1791         UPDATE_VF_COUNTER(VFGPTLBC, gptlbc);
1792         UPDATE_VF_COUNTER(VFGORLBC, gorlbc);
1793         UPDATE_VF_COUNTER(VFGPRLBC, gprlbc);
1794
1795         /* Fill out the OS statistics structure */
1796         adapter->net_stats.multicast = adapter->stats.mprc;
1797 }
1798
1799 static void igbvf_print_link_info(struct igbvf_adapter *adapter)
1800 {
1801         dev_info(&adapter->pdev->dev, "Link is Up %d Mbps %s\n",
1802                  adapter->link_speed,
1803                  ((adapter->link_duplex == FULL_DUPLEX) ?
1804                   "Full Duplex" : "Half Duplex"));
1805 }
1806
1807 static bool igbvf_has_link(struct igbvf_adapter *adapter)
1808 {
1809         struct e1000_hw *hw = &adapter->hw;
1810         s32 ret_val = E1000_SUCCESS;
1811         bool link_active;
1812
1813         /* If interface is down, stay link down */
1814         if (test_bit(__IGBVF_DOWN, &adapter->state))
1815                 return false;
1816
1817         ret_val = hw->mac.ops.check_for_link(hw);
1818         link_active = !hw->mac.get_link_status;
1819
1820         /* if check for link returns error we will need to reset */
1821         if (ret_val && time_after(jiffies, adapter->last_reset + (10 * HZ)))
1822                 schedule_work(&adapter->reset_task);
1823
1824         return link_active;
1825 }
1826
1827 /**
1828  * igbvf_watchdog - Timer Call-back
1829  * @data: pointer to adapter cast into an unsigned long
1830  **/
1831 static void igbvf_watchdog(unsigned long data)
1832 {
1833         struct igbvf_adapter *adapter = (struct igbvf_adapter *) data;
1834
1835         /* Do the rest outside of interrupt context */
1836         schedule_work(&adapter->watchdog_task);
1837 }
1838
1839 static void igbvf_watchdog_task(struct work_struct *work)
1840 {
1841         struct igbvf_adapter *adapter = container_of(work,
1842                                                      struct igbvf_adapter,
1843                                                      watchdog_task);
1844         struct net_device *netdev = adapter->netdev;
1845         struct e1000_mac_info *mac = &adapter->hw.mac;
1846         struct igbvf_ring *tx_ring = adapter->tx_ring;
1847         struct e1000_hw *hw = &adapter->hw;
1848         u32 link;
1849         int tx_pending = 0;
1850
1851         link = igbvf_has_link(adapter);
1852
1853         if (link) {
1854                 if (!netif_carrier_ok(netdev)) {
1855                         bool txb2b = 1;
1856
1857                         mac->ops.get_link_up_info(&adapter->hw,
1858                                                   &adapter->link_speed,
1859                                                   &adapter->link_duplex);
1860                         igbvf_print_link_info(adapter);
1861
1862                         /* adjust timeout factor according to speed/duplex */
1863                         adapter->tx_timeout_factor = 1;
1864                         switch (adapter->link_speed) {
1865                         case SPEED_10:
1866                                 txb2b = 0;
1867                                 adapter->tx_timeout_factor = 16;
1868                                 break;
1869                         case SPEED_100:
1870                                 txb2b = 0;
1871                                 /* maybe add some timeout factor ? */
1872                                 break;
1873                         }
1874
1875                         netif_carrier_on(netdev);
1876                         netif_wake_queue(netdev);
1877                 }
1878         } else {
1879                 if (netif_carrier_ok(netdev)) {
1880                         adapter->link_speed = 0;
1881                         adapter->link_duplex = 0;
1882                         dev_info(&adapter->pdev->dev, "Link is Down\n");
1883                         netif_carrier_off(netdev);
1884                         netif_stop_queue(netdev);
1885                 }
1886         }
1887
1888         if (netif_carrier_ok(netdev)) {
1889                 igbvf_update_stats(adapter);
1890         } else {
1891                 tx_pending = (igbvf_desc_unused(tx_ring) + 1 <
1892                               tx_ring->count);
1893                 if (tx_pending) {
1894                         /*
1895                          * We've lost link, so the controller stops DMA,
1896                          * but we've got queued Tx work that's never going
1897                          * to get done, so reset controller to flush Tx.
1898                          * (Do the reset outside of interrupt context).
1899                          */
1900                         adapter->tx_timeout_count++;
1901                         schedule_work(&adapter->reset_task);
1902                 }
1903         }
1904
1905         /* Cause software interrupt to ensure Rx ring is cleaned */
1906         ew32(EICS, adapter->rx_ring->eims_value);
1907
1908         /* Force detection of hung controller every watchdog period */
1909         adapter->detect_tx_hung = 1;
1910
1911         /* Reset the timer */
1912         if (!test_bit(__IGBVF_DOWN, &adapter->state))
1913                 mod_timer(&adapter->watchdog_timer,
1914                           round_jiffies(jiffies + (2 * HZ)));
1915 }
1916
1917 #define IGBVF_TX_FLAGS_CSUM             0x00000001
1918 #define IGBVF_TX_FLAGS_VLAN             0x00000002
1919 #define IGBVF_TX_FLAGS_TSO              0x00000004
1920 #define IGBVF_TX_FLAGS_IPV4             0x00000008
1921 #define IGBVF_TX_FLAGS_VLAN_MASK        0xffff0000
1922 #define IGBVF_TX_FLAGS_VLAN_SHIFT       16
1923
1924 static int igbvf_tso(struct igbvf_adapter *adapter,
1925                      struct igbvf_ring *tx_ring,
1926                      struct sk_buff *skb, u32 tx_flags, u8 *hdr_len)
1927 {
1928         struct e1000_adv_tx_context_desc *context_desc;
1929         unsigned int i;
1930         int err;
1931         struct igbvf_buffer *buffer_info;
1932         u32 info = 0, tu_cmd = 0;
1933         u32 mss_l4len_idx, l4len;
1934         *hdr_len = 0;
1935
1936         if (skb_header_cloned(skb)) {
1937                 err = pskb_expand_head(skb, 0, 0, GFP_ATOMIC);
1938                 if (err) {
1939                         dev_err(&adapter->pdev->dev,
1940                                 "igbvf_tso returning an error\n");
1941                         return err;
1942                 }
1943         }
1944
1945         l4len = tcp_hdrlen(skb);
1946         *hdr_len += l4len;
1947
1948         if (skb->protocol == htons(ETH_P_IP)) {
1949                 struct iphdr *iph = ip_hdr(skb);
1950                 iph->tot_len = 0;
1951                 iph->check = 0;
1952                 tcp_hdr(skb)->check = ~csum_tcpudp_magic(iph->saddr,
1953                                                          iph->daddr, 0,
1954                                                          IPPROTO_TCP,
1955                                                          0);
1956         } else if (skb_is_gso_v6(skb)) {
1957                 ipv6_hdr(skb)->payload_len = 0;
1958                 tcp_hdr(skb)->check = ~csum_ipv6_magic(&ipv6_hdr(skb)->saddr,
1959                                                        &ipv6_hdr(skb)->daddr,
1960                                                        0, IPPROTO_TCP, 0);
1961         }
1962
1963         i = tx_ring->next_to_use;
1964
1965         buffer_info = &tx_ring->buffer_info[i];
1966         context_desc = IGBVF_TX_CTXTDESC_ADV(*tx_ring, i);
1967         /* VLAN MACLEN IPLEN */
1968         if (tx_flags & IGBVF_TX_FLAGS_VLAN)
1969                 info |= (tx_flags & IGBVF_TX_FLAGS_VLAN_MASK);
1970         info |= (skb_network_offset(skb) << E1000_ADVTXD_MACLEN_SHIFT);
1971         *hdr_len += skb_network_offset(skb);
1972         info |= (skb_transport_header(skb) - skb_network_header(skb));
1973         *hdr_len += (skb_transport_header(skb) - skb_network_header(skb));
1974         context_desc->vlan_macip_lens = cpu_to_le32(info);
1975
1976         /* ADV DTYP TUCMD MKRLOC/ISCSIHEDLEN */
1977         tu_cmd |= (E1000_TXD_CMD_DEXT | E1000_ADVTXD_DTYP_CTXT);
1978
1979         if (skb->protocol == htons(ETH_P_IP))
1980                 tu_cmd |= E1000_ADVTXD_TUCMD_IPV4;
1981         tu_cmd |= E1000_ADVTXD_TUCMD_L4T_TCP;
1982
1983         context_desc->type_tucmd_mlhl = cpu_to_le32(tu_cmd);
1984
1985         /* MSS L4LEN IDX */
1986         mss_l4len_idx = (skb_shinfo(skb)->gso_size << E1000_ADVTXD_MSS_SHIFT);
1987         mss_l4len_idx |= (l4len << E1000_ADVTXD_L4LEN_SHIFT);
1988
1989         context_desc->mss_l4len_idx = cpu_to_le32(mss_l4len_idx);
1990         context_desc->seqnum_seed = 0;
1991
1992         buffer_info->time_stamp = jiffies;
1993         buffer_info->next_to_watch = i;
1994         buffer_info->dma = 0;
1995         i++;
1996         if (i == tx_ring->count)
1997                 i = 0;
1998
1999         tx_ring->next_to_use = i;
2000
2001         return true;
2002 }
2003
2004 static inline bool igbvf_tx_csum(struct igbvf_adapter *adapter,
2005                                  struct igbvf_ring *tx_ring,
2006                                  struct sk_buff *skb, u32 tx_flags)
2007 {
2008         struct e1000_adv_tx_context_desc *context_desc;
2009         unsigned int i;
2010         struct igbvf_buffer *buffer_info;
2011         u32 info = 0, tu_cmd = 0;
2012
2013         if ((skb->ip_summed == CHECKSUM_PARTIAL) ||
2014             (tx_flags & IGBVF_TX_FLAGS_VLAN)) {
2015                 i = tx_ring->next_to_use;
2016                 buffer_info = &tx_ring->buffer_info[i];
2017                 context_desc = IGBVF_TX_CTXTDESC_ADV(*tx_ring, i);
2018
2019                 if (tx_flags & IGBVF_TX_FLAGS_VLAN)
2020                         info |= (tx_flags & IGBVF_TX_FLAGS_VLAN_MASK);
2021
2022                 info |= (skb_network_offset(skb) << E1000_ADVTXD_MACLEN_SHIFT);
2023                 if (skb->ip_summed == CHECKSUM_PARTIAL)
2024                         info |= (skb_transport_header(skb) -
2025                                  skb_network_header(skb));
2026
2027
2028                 context_desc->vlan_macip_lens = cpu_to_le32(info);
2029
2030                 tu_cmd |= (E1000_TXD_CMD_DEXT | E1000_ADVTXD_DTYP_CTXT);
2031
2032                 if (skb->ip_summed == CHECKSUM_PARTIAL) {
2033                         switch (skb->protocol) {
2034                         case __constant_htons(ETH_P_IP):
2035                                 tu_cmd |= E1000_ADVTXD_TUCMD_IPV4;
2036                                 if (ip_hdr(skb)->protocol == IPPROTO_TCP)
2037                                         tu_cmd |= E1000_ADVTXD_TUCMD_L4T_TCP;
2038                                 break;
2039                         case __constant_htons(ETH_P_IPV6):
2040                                 if (ipv6_hdr(skb)->nexthdr == IPPROTO_TCP)
2041                                         tu_cmd |= E1000_ADVTXD_TUCMD_L4T_TCP;
2042                                 break;
2043                         default:
2044                                 break;
2045                         }
2046                 }
2047
2048                 context_desc->type_tucmd_mlhl = cpu_to_le32(tu_cmd);
2049                 context_desc->seqnum_seed = 0;
2050                 context_desc->mss_l4len_idx = 0;
2051
2052                 buffer_info->time_stamp = jiffies;
2053                 buffer_info->next_to_watch = i;
2054                 buffer_info->dma = 0;
2055                 i++;
2056                 if (i == tx_ring->count)
2057                         i = 0;
2058                 tx_ring->next_to_use = i;
2059
2060                 return true;
2061         }
2062
2063         return false;
2064 }
2065
2066 static int igbvf_maybe_stop_tx(struct net_device *netdev, int size)
2067 {
2068         struct igbvf_adapter *adapter = netdev_priv(netdev);
2069
2070         /* there is enough descriptors then we don't need to worry  */
2071         if (igbvf_desc_unused(adapter->tx_ring) >= size)
2072                 return 0;
2073
2074         netif_stop_queue(netdev);
2075
2076         smp_mb();
2077
2078         /* We need to check again just in case room has been made available */
2079         if (igbvf_desc_unused(adapter->tx_ring) < size)
2080                 return -EBUSY;
2081
2082         netif_wake_queue(netdev);
2083
2084         ++adapter->restart_queue;
2085         return 0;
2086 }
2087
2088 #define IGBVF_MAX_TXD_PWR       16
2089 #define IGBVF_MAX_DATA_PER_TXD  (1 << IGBVF_MAX_TXD_PWR)
2090
2091 static inline int igbvf_tx_map_adv(struct igbvf_adapter *adapter,
2092                                    struct igbvf_ring *tx_ring,
2093                                    struct sk_buff *skb,
2094                                    unsigned int first)
2095 {
2096         struct igbvf_buffer *buffer_info;
2097         struct pci_dev *pdev = adapter->pdev;
2098         unsigned int len = skb_headlen(skb);
2099         unsigned int count = 0, i;
2100         unsigned int f;
2101
2102         i = tx_ring->next_to_use;
2103
2104         buffer_info = &tx_ring->buffer_info[i];
2105         BUG_ON(len >= IGBVF_MAX_DATA_PER_TXD);
2106         buffer_info->length = len;
2107         /* set time_stamp *before* dma to help avoid a possible race */
2108         buffer_info->time_stamp = jiffies;
2109         buffer_info->next_to_watch = i;
2110         buffer_info->mapped_as_page = false;
2111         buffer_info->dma = dma_map_single(&pdev->dev, skb->data, len,
2112                                           DMA_TO_DEVICE);
2113         if (dma_mapping_error(&pdev->dev, buffer_info->dma))
2114                 goto dma_error;
2115
2116
2117         for (f = 0; f < skb_shinfo(skb)->nr_frags; f++) {
2118                 struct skb_frag_struct *frag;
2119
2120                 count++;
2121                 i++;
2122                 if (i == tx_ring->count)
2123                         i = 0;
2124
2125                 frag = &skb_shinfo(skb)->frags[f];
2126                 len = frag->size;
2127
2128                 buffer_info = &tx_ring->buffer_info[i];
2129                 BUG_ON(len >= IGBVF_MAX_DATA_PER_TXD);
2130                 buffer_info->length = len;
2131                 buffer_info->time_stamp = jiffies;
2132                 buffer_info->next_to_watch = i;
2133                 buffer_info->mapped_as_page = true;
2134                 buffer_info->dma = dma_map_page(&pdev->dev,
2135                                                 frag->page,
2136                                                 frag->page_offset,
2137                                                 len,
2138                                                 DMA_TO_DEVICE);
2139                 if (dma_mapping_error(&pdev->dev, buffer_info->dma))
2140                         goto dma_error;
2141         }
2142
2143         tx_ring->buffer_info[i].skb = skb;
2144         tx_ring->buffer_info[first].next_to_watch = i;
2145
2146         return ++count;
2147
2148 dma_error:
2149         dev_err(&pdev->dev, "TX DMA map failed\n");
2150
2151         /* clear timestamp and dma mappings for failed buffer_info mapping */
2152         buffer_info->dma = 0;
2153         buffer_info->time_stamp = 0;
2154         buffer_info->length = 0;
2155         buffer_info->next_to_watch = 0;
2156         buffer_info->mapped_as_page = false;
2157         if (count)
2158                 count--;
2159
2160         /* clear timestamp and dma mappings for remaining portion of packet */
2161         while (count--) {
2162                 if (i==0)
2163                         i += tx_ring->count;
2164                 i--;
2165                 buffer_info = &tx_ring->buffer_info[i];
2166                 igbvf_put_txbuf(adapter, buffer_info);
2167         }
2168
2169         return 0;
2170 }
2171
2172 static inline void igbvf_tx_queue_adv(struct igbvf_adapter *adapter,
2173                                       struct igbvf_ring *tx_ring,
2174                                       int tx_flags, int count, u32 paylen,
2175                                       u8 hdr_len)
2176 {
2177         union e1000_adv_tx_desc *tx_desc = NULL;
2178         struct igbvf_buffer *buffer_info;
2179         u32 olinfo_status = 0, cmd_type_len;
2180         unsigned int i;
2181
2182         cmd_type_len = (E1000_ADVTXD_DTYP_DATA | E1000_ADVTXD_DCMD_IFCS |
2183                         E1000_ADVTXD_DCMD_DEXT);
2184
2185         if (tx_flags & IGBVF_TX_FLAGS_VLAN)
2186                 cmd_type_len |= E1000_ADVTXD_DCMD_VLE;
2187
2188         if (tx_flags & IGBVF_TX_FLAGS_TSO) {
2189                 cmd_type_len |= E1000_ADVTXD_DCMD_TSE;
2190
2191                 /* insert tcp checksum */
2192                 olinfo_status |= E1000_TXD_POPTS_TXSM << 8;
2193
2194                 /* insert ip checksum */
2195                 if (tx_flags & IGBVF_TX_FLAGS_IPV4)
2196                         olinfo_status |= E1000_TXD_POPTS_IXSM << 8;
2197
2198         } else if (tx_flags & IGBVF_TX_FLAGS_CSUM) {
2199                 olinfo_status |= E1000_TXD_POPTS_TXSM << 8;
2200         }
2201
2202         olinfo_status |= ((paylen - hdr_len) << E1000_ADVTXD_PAYLEN_SHIFT);
2203
2204         i = tx_ring->next_to_use;
2205         while (count--) {
2206                 buffer_info = &tx_ring->buffer_info[i];
2207                 tx_desc = IGBVF_TX_DESC_ADV(*tx_ring, i);
2208                 tx_desc->read.buffer_addr = cpu_to_le64(buffer_info->dma);
2209                 tx_desc->read.cmd_type_len =
2210                          cpu_to_le32(cmd_type_len | buffer_info->length);
2211                 tx_desc->read.olinfo_status = cpu_to_le32(olinfo_status);
2212                 i++;
2213                 if (i == tx_ring->count)
2214                         i = 0;
2215         }
2216
2217         tx_desc->read.cmd_type_len |= cpu_to_le32(adapter->txd_cmd);
2218         /* Force memory writes to complete before letting h/w
2219          * know there are new descriptors to fetch.  (Only
2220          * applicable for weak-ordered memory model archs,
2221          * such as IA-64). */
2222         wmb();
2223
2224         tx_ring->next_to_use = i;
2225         writel(i, adapter->hw.hw_addr + tx_ring->tail);
2226         /* we need this if more than one processor can write to our tail
2227          * at a time, it syncronizes IO on IA64/Altix systems */
2228         mmiowb();
2229 }
2230
2231 static netdev_tx_t igbvf_xmit_frame_ring_adv(struct sk_buff *skb,
2232                                              struct net_device *netdev,
2233                                              struct igbvf_ring *tx_ring)
2234 {
2235         struct igbvf_adapter *adapter = netdev_priv(netdev);
2236         unsigned int first, tx_flags = 0;
2237         u8 hdr_len = 0;
2238         int count = 0;
2239         int tso = 0;
2240
2241         if (test_bit(__IGBVF_DOWN, &adapter->state)) {
2242                 dev_kfree_skb_any(skb);
2243                 return NETDEV_TX_OK;
2244         }
2245
2246         if (skb->len <= 0) {
2247                 dev_kfree_skb_any(skb);
2248                 return NETDEV_TX_OK;
2249         }
2250
2251         /*
2252          * need: count + 4 desc gap to keep tail from touching
2253          *       + 2 desc gap to keep tail from touching head,
2254          *       + 1 desc for skb->data,
2255          *       + 1 desc for context descriptor,
2256          * head, otherwise try next time
2257          */
2258         if (igbvf_maybe_stop_tx(netdev, skb_shinfo(skb)->nr_frags + 4)) {
2259                 /* this is a hard error */
2260                 return NETDEV_TX_BUSY;
2261         }
2262
2263         if (adapter->vlgrp && vlan_tx_tag_present(skb)) {
2264                 tx_flags |= IGBVF_TX_FLAGS_VLAN;
2265                 tx_flags |= (vlan_tx_tag_get(skb) << IGBVF_TX_FLAGS_VLAN_SHIFT);
2266         }
2267
2268         if (skb->protocol == htons(ETH_P_IP))
2269                 tx_flags |= IGBVF_TX_FLAGS_IPV4;
2270
2271         first = tx_ring->next_to_use;
2272
2273         tso = skb_is_gso(skb) ?
2274                 igbvf_tso(adapter, tx_ring, skb, tx_flags, &hdr_len) : 0;
2275         if (unlikely(tso < 0)) {
2276                 dev_kfree_skb_any(skb);
2277                 return NETDEV_TX_OK;
2278         }
2279
2280         if (tso)
2281                 tx_flags |= IGBVF_TX_FLAGS_TSO;
2282         else if (igbvf_tx_csum(adapter, tx_ring, skb, tx_flags) &&
2283                  (skb->ip_summed == CHECKSUM_PARTIAL))
2284                 tx_flags |= IGBVF_TX_FLAGS_CSUM;
2285
2286         /*
2287          * count reflects descriptors mapped, if 0 then mapping error
2288          * has occured and we need to rewind the descriptor queue
2289          */
2290         count = igbvf_tx_map_adv(adapter, tx_ring, skb, first);
2291
2292         if (count) {
2293                 igbvf_tx_queue_adv(adapter, tx_ring, tx_flags, count,
2294                                    skb->len, hdr_len);
2295                 /* Make sure there is space in the ring for the next send. */
2296                 igbvf_maybe_stop_tx(netdev, MAX_SKB_FRAGS + 4);
2297         } else {
2298                 dev_kfree_skb_any(skb);
2299                 tx_ring->buffer_info[first].time_stamp = 0;
2300                 tx_ring->next_to_use = first;
2301         }
2302
2303         return NETDEV_TX_OK;
2304 }
2305
2306 static netdev_tx_t igbvf_xmit_frame(struct sk_buff *skb,
2307                                     struct net_device *netdev)
2308 {
2309         struct igbvf_adapter *adapter = netdev_priv(netdev);
2310         struct igbvf_ring *tx_ring;
2311
2312         if (test_bit(__IGBVF_DOWN, &adapter->state)) {
2313                 dev_kfree_skb_any(skb);
2314                 return NETDEV_TX_OK;
2315         }
2316
2317         tx_ring = &adapter->tx_ring[0];
2318
2319         return igbvf_xmit_frame_ring_adv(skb, netdev, tx_ring);
2320 }
2321
2322 /**
2323  * igbvf_tx_timeout - Respond to a Tx Hang
2324  * @netdev: network interface device structure
2325  **/
2326 static void igbvf_tx_timeout(struct net_device *netdev)
2327 {
2328         struct igbvf_adapter *adapter = netdev_priv(netdev);
2329
2330         /* Do the reset outside of interrupt context */
2331         adapter->tx_timeout_count++;
2332         schedule_work(&adapter->reset_task);
2333 }
2334
2335 static void igbvf_reset_task(struct work_struct *work)
2336 {
2337         struct igbvf_adapter *adapter;
2338         adapter = container_of(work, struct igbvf_adapter, reset_task);
2339
2340         igbvf_reinit_locked(adapter);
2341 }
2342
2343 /**
2344  * igbvf_get_stats - Get System Network Statistics
2345  * @netdev: network interface device structure
2346  *
2347  * Returns the address of the device statistics structure.
2348  * The statistics are actually updated from the timer callback.
2349  **/
2350 static struct net_device_stats *igbvf_get_stats(struct net_device *netdev)
2351 {
2352         struct igbvf_adapter *adapter = netdev_priv(netdev);
2353
2354         /* only return the current stats */
2355         return &adapter->net_stats;
2356 }
2357
2358 /**
2359  * igbvf_change_mtu - Change the Maximum Transfer Unit
2360  * @netdev: network interface device structure
2361  * @new_mtu: new value for maximum frame size
2362  *
2363  * Returns 0 on success, negative on failure
2364  **/
2365 static int igbvf_change_mtu(struct net_device *netdev, int new_mtu)
2366 {
2367         struct igbvf_adapter *adapter = netdev_priv(netdev);
2368         int max_frame = new_mtu + ETH_HLEN + ETH_FCS_LEN;
2369
2370         if ((new_mtu < 68) || (max_frame > MAX_JUMBO_FRAME_SIZE)) {
2371                 dev_err(&adapter->pdev->dev, "Invalid MTU setting\n");
2372                 return -EINVAL;
2373         }
2374
2375 #define MAX_STD_JUMBO_FRAME_SIZE 9234
2376         if (max_frame > MAX_STD_JUMBO_FRAME_SIZE) {
2377                 dev_err(&adapter->pdev->dev, "MTU > 9216 not supported.\n");
2378                 return -EINVAL;
2379         }
2380
2381         while (test_and_set_bit(__IGBVF_RESETTING, &adapter->state))
2382                 msleep(1);
2383         /* igbvf_down has a dependency on max_frame_size */
2384         adapter->max_frame_size = max_frame;
2385         if (netif_running(netdev))
2386                 igbvf_down(adapter);
2387
2388         /*
2389          * NOTE: netdev_alloc_skb reserves 16 bytes, and typically NET_IP_ALIGN
2390          * means we reserve 2 more, this pushes us to allocate from the next
2391          * larger slab size.
2392          * i.e. RXBUFFER_2048 --> size-4096 slab
2393          * However with the new *_jumbo_rx* routines, jumbo receives will use
2394          * fragmented skbs
2395          */
2396
2397         if (max_frame <= 1024)
2398                 adapter->rx_buffer_len = 1024;
2399         else if (max_frame <= 2048)
2400                 adapter->rx_buffer_len = 2048;
2401         else
2402 #if (PAGE_SIZE / 2) > 16384
2403                 adapter->rx_buffer_len = 16384;
2404 #else
2405                 adapter->rx_buffer_len = PAGE_SIZE / 2;
2406 #endif
2407
2408
2409         /* adjust allocation if LPE protects us, and we aren't using SBP */
2410         if ((max_frame == ETH_FRAME_LEN + ETH_FCS_LEN) ||
2411              (max_frame == ETH_FRAME_LEN + VLAN_HLEN + ETH_FCS_LEN))
2412                 adapter->rx_buffer_len = ETH_FRAME_LEN + VLAN_HLEN +
2413                                          ETH_FCS_LEN;
2414
2415         dev_info(&adapter->pdev->dev, "changing MTU from %d to %d\n",
2416                  netdev->mtu, new_mtu);
2417         netdev->mtu = new_mtu;
2418
2419         if (netif_running(netdev))
2420                 igbvf_up(adapter);
2421         else
2422                 igbvf_reset(adapter);
2423
2424         clear_bit(__IGBVF_RESETTING, &adapter->state);
2425
2426         return 0;
2427 }
2428
2429 static int igbvf_ioctl(struct net_device *netdev, struct ifreq *ifr, int cmd)
2430 {
2431         switch (cmd) {
2432         default:
2433                 return -EOPNOTSUPP;
2434         }
2435 }
2436
2437 static int igbvf_suspend(struct pci_dev *pdev, pm_message_t state)
2438 {
2439         struct net_device *netdev = pci_get_drvdata(pdev);
2440         struct igbvf_adapter *adapter = netdev_priv(netdev);
2441 #ifdef CONFIG_PM
2442         int retval = 0;
2443 #endif
2444
2445         netif_device_detach(netdev);
2446
2447         if (netif_running(netdev)) {
2448                 WARN_ON(test_bit(__IGBVF_RESETTING, &adapter->state));
2449                 igbvf_down(adapter);
2450                 igbvf_free_irq(adapter);
2451         }
2452
2453 #ifdef CONFIG_PM
2454         retval = pci_save_state(pdev);
2455         if (retval)
2456                 return retval;
2457 #endif
2458
2459         pci_disable_device(pdev);
2460
2461         return 0;
2462 }
2463
2464 #ifdef CONFIG_PM
2465 static int igbvf_resume(struct pci_dev *pdev)
2466 {
2467         struct net_device *netdev = pci_get_drvdata(pdev);
2468         struct igbvf_adapter *adapter = netdev_priv(netdev);
2469         u32 err;
2470
2471         pci_restore_state(pdev);
2472         err = pci_enable_device_mem(pdev);
2473         if (err) {
2474                 dev_err(&pdev->dev, "Cannot enable PCI device from suspend\n");
2475                 return err;
2476         }
2477
2478         pci_set_master(pdev);
2479
2480         if (netif_running(netdev)) {
2481                 err = igbvf_request_irq(adapter);
2482                 if (err)
2483                         return err;
2484         }
2485
2486         igbvf_reset(adapter);
2487
2488         if (netif_running(netdev))
2489                 igbvf_up(adapter);
2490
2491         netif_device_attach(netdev);
2492
2493         return 0;
2494 }
2495 #endif
2496
2497 static void igbvf_shutdown(struct pci_dev *pdev)
2498 {
2499         igbvf_suspend(pdev, PMSG_SUSPEND);
2500 }
2501
2502 #ifdef CONFIG_NET_POLL_CONTROLLER
2503 /*
2504  * Polling 'interrupt' - used by things like netconsole to send skbs
2505  * without having to re-enable interrupts. It's not called while
2506  * the interrupt routine is executing.
2507  */
2508 static void igbvf_netpoll(struct net_device *netdev)
2509 {
2510         struct igbvf_adapter *adapter = netdev_priv(netdev);
2511
2512         disable_irq(adapter->pdev->irq);
2513
2514         igbvf_clean_tx_irq(adapter->tx_ring);
2515
2516         enable_irq(adapter->pdev->irq);
2517 }
2518 #endif
2519
2520 /**
2521  * igbvf_io_error_detected - called when PCI error is detected
2522  * @pdev: Pointer to PCI device
2523  * @state: The current pci connection state
2524  *
2525  * This function is called after a PCI bus error affecting
2526  * this device has been detected.
2527  */
2528 static pci_ers_result_t igbvf_io_error_detected(struct pci_dev *pdev,
2529                                                 pci_channel_state_t state)
2530 {
2531         struct net_device *netdev = pci_get_drvdata(pdev);
2532         struct igbvf_adapter *adapter = netdev_priv(netdev);
2533
2534         netif_device_detach(netdev);
2535
2536         if (state == pci_channel_io_perm_failure)
2537                 return PCI_ERS_RESULT_DISCONNECT;
2538
2539         if (netif_running(netdev))
2540                 igbvf_down(adapter);
2541         pci_disable_device(pdev);
2542
2543         /* Request a slot slot reset. */
2544         return PCI_ERS_RESULT_NEED_RESET;
2545 }
2546
2547 /**
2548  * igbvf_io_slot_reset - called after the pci bus has been reset.
2549  * @pdev: Pointer to PCI device
2550  *
2551  * Restart the card from scratch, as if from a cold-boot. Implementation
2552  * resembles the first-half of the igbvf_resume routine.
2553  */
2554 static pci_ers_result_t igbvf_io_slot_reset(struct pci_dev *pdev)
2555 {
2556         struct net_device *netdev = pci_get_drvdata(pdev);
2557         struct igbvf_adapter *adapter = netdev_priv(netdev);
2558
2559         if (pci_enable_device_mem(pdev)) {
2560                 dev_err(&pdev->dev,
2561                         "Cannot re-enable PCI device after reset.\n");
2562                 return PCI_ERS_RESULT_DISCONNECT;
2563         }
2564         pci_set_master(pdev);
2565
2566         igbvf_reset(adapter);
2567
2568         return PCI_ERS_RESULT_RECOVERED;
2569 }
2570
2571 /**
2572  * igbvf_io_resume - called when traffic can start flowing again.
2573  * @pdev: Pointer to PCI device
2574  *
2575  * This callback is called when the error recovery driver tells us that
2576  * its OK to resume normal operation. Implementation resembles the
2577  * second-half of the igbvf_resume routine.
2578  */
2579 static void igbvf_io_resume(struct pci_dev *pdev)
2580 {
2581         struct net_device *netdev = pci_get_drvdata(pdev);
2582         struct igbvf_adapter *adapter = netdev_priv(netdev);
2583
2584         if (netif_running(netdev)) {
2585                 if (igbvf_up(adapter)) {
2586                         dev_err(&pdev->dev,
2587                                 "can't bring device back up after reset\n");
2588                         return;
2589                 }
2590         }
2591
2592         netif_device_attach(netdev);
2593 }
2594
2595 static void igbvf_print_device_info(struct igbvf_adapter *adapter)
2596 {
2597         struct e1000_hw *hw = &adapter->hw;
2598         struct net_device *netdev = adapter->netdev;
2599         struct pci_dev *pdev = adapter->pdev;
2600
2601         dev_info(&pdev->dev, "Intel(R) 82576 Virtual Function\n");
2602         dev_info(&pdev->dev, "Address: %pM\n", netdev->dev_addr);
2603         dev_info(&pdev->dev, "MAC: %d\n", hw->mac.type);
2604 }
2605
2606 static const struct net_device_ops igbvf_netdev_ops = {
2607         .ndo_open                       = igbvf_open,
2608         .ndo_stop                       = igbvf_close,
2609         .ndo_start_xmit                 = igbvf_xmit_frame,
2610         .ndo_get_stats                  = igbvf_get_stats,
2611         .ndo_set_multicast_list         = igbvf_set_multi,
2612         .ndo_set_mac_address            = igbvf_set_mac,
2613         .ndo_change_mtu                 = igbvf_change_mtu,
2614         .ndo_do_ioctl                   = igbvf_ioctl,
2615         .ndo_tx_timeout                 = igbvf_tx_timeout,
2616         .ndo_vlan_rx_register           = igbvf_vlan_rx_register,
2617         .ndo_vlan_rx_add_vid            = igbvf_vlan_rx_add_vid,
2618         .ndo_vlan_rx_kill_vid           = igbvf_vlan_rx_kill_vid,
2619 #ifdef CONFIG_NET_POLL_CONTROLLER
2620         .ndo_poll_controller            = igbvf_netpoll,
2621 #endif
2622 };
2623
2624 /**
2625  * igbvf_probe - Device Initialization Routine
2626  * @pdev: PCI device information struct
2627  * @ent: entry in igbvf_pci_tbl
2628  *
2629  * Returns 0 on success, negative on failure
2630  *
2631  * igbvf_probe initializes an adapter identified by a pci_dev structure.
2632  * The OS initialization, configuring of the adapter private structure,
2633  * and a hardware reset occur.
2634  **/
2635 static int __devinit igbvf_probe(struct pci_dev *pdev,
2636                                  const struct pci_device_id *ent)
2637 {
2638         struct net_device *netdev;
2639         struct igbvf_adapter *adapter;
2640         struct e1000_hw *hw;
2641         const struct igbvf_info *ei = igbvf_info_tbl[ent->driver_data];
2642
2643         static int cards_found;
2644         int err, pci_using_dac;
2645
2646         err = pci_enable_device_mem(pdev);
2647         if (err)
2648                 return err;
2649
2650         pci_using_dac = 0;
2651         err = dma_set_mask(&pdev->dev, DMA_BIT_MASK(64));
2652         if (!err) {
2653                 err = dma_set_coherent_mask(&pdev->dev, DMA_BIT_MASK(64));
2654                 if (!err)
2655                         pci_using_dac = 1;
2656         } else {
2657                 err = dma_set_mask(&pdev->dev, DMA_BIT_MASK(32));
2658                 if (err) {
2659                         err = dma_set_coherent_mask(&pdev->dev,
2660                                                     DMA_BIT_MASK(32));
2661                         if (err) {
2662                                 dev_err(&pdev->dev, "No usable DMA "
2663                                         "configuration, aborting\n");
2664                                 goto err_dma;
2665                         }
2666                 }
2667         }
2668
2669         err = pci_request_regions(pdev, igbvf_driver_name);
2670         if (err)
2671                 goto err_pci_reg;
2672
2673         pci_set_master(pdev);
2674
2675         err = -ENOMEM;
2676         netdev = alloc_etherdev(sizeof(struct igbvf_adapter));
2677         if (!netdev)
2678                 goto err_alloc_etherdev;
2679
2680         SET_NETDEV_DEV(netdev, &pdev->dev);
2681
2682         pci_set_drvdata(pdev, netdev);
2683         adapter = netdev_priv(netdev);
2684         hw = &adapter->hw;
2685         adapter->netdev = netdev;
2686         adapter->pdev = pdev;
2687         adapter->ei = ei;
2688         adapter->pba = ei->pba;
2689         adapter->flags = ei->flags;
2690         adapter->hw.back = adapter;
2691         adapter->hw.mac.type = ei->mac;
2692         adapter->msg_enable = (1 << NETIF_MSG_DRV | NETIF_MSG_PROBE) - 1;
2693
2694         /* PCI config space info */
2695
2696         hw->vendor_id = pdev->vendor;
2697         hw->device_id = pdev->device;
2698         hw->subsystem_vendor_id = pdev->subsystem_vendor;
2699         hw->subsystem_device_id = pdev->subsystem_device;
2700
2701         pci_read_config_byte(pdev, PCI_REVISION_ID, &hw->revision_id);
2702
2703         err = -EIO;
2704         adapter->hw.hw_addr = ioremap(pci_resource_start(pdev, 0),
2705                                       pci_resource_len(pdev, 0));
2706
2707         if (!adapter->hw.hw_addr)
2708                 goto err_ioremap;
2709
2710         if (ei->get_variants) {
2711                 err = ei->get_variants(adapter);
2712                 if (err)
2713                         goto err_ioremap;
2714         }
2715
2716         /* setup adapter struct */
2717         err = igbvf_sw_init(adapter);
2718         if (err)
2719                 goto err_sw_init;
2720
2721         /* construct the net_device struct */
2722         netdev->netdev_ops = &igbvf_netdev_ops;
2723
2724         igbvf_set_ethtool_ops(netdev);
2725         netdev->watchdog_timeo = 5 * HZ;
2726         strncpy(netdev->name, pci_name(pdev), sizeof(netdev->name) - 1);
2727
2728         adapter->bd_number = cards_found++;
2729
2730         netdev->features = NETIF_F_SG |
2731                            NETIF_F_IP_CSUM |
2732                            NETIF_F_HW_VLAN_TX |
2733                            NETIF_F_HW_VLAN_RX |
2734                            NETIF_F_HW_VLAN_FILTER;
2735
2736         netdev->features |= NETIF_F_IPV6_CSUM;
2737         netdev->features |= NETIF_F_TSO;
2738         netdev->features |= NETIF_F_TSO6;
2739
2740         if (pci_using_dac)
2741                 netdev->features |= NETIF_F_HIGHDMA;
2742
2743         netdev->vlan_features |= NETIF_F_TSO;
2744         netdev->vlan_features |= NETIF_F_TSO6;
2745         netdev->vlan_features |= NETIF_F_IP_CSUM;
2746         netdev->vlan_features |= NETIF_F_IPV6_CSUM;
2747         netdev->vlan_features |= NETIF_F_SG;
2748
2749         /*reset the controller to put the device in a known good state */
2750         err = hw->mac.ops.reset_hw(hw);
2751         if (err) {
2752                 dev_info(&pdev->dev,
2753                          "PF still in reset state, assigning new address."
2754                          " Is the PF interface up?\n");
2755                 dev_hw_addr_random(adapter->netdev, hw->mac.addr);
2756         } else {
2757                 err = hw->mac.ops.read_mac_addr(hw);
2758                 if (err) {
2759                         dev_err(&pdev->dev, "Error reading MAC address\n");
2760                         goto err_hw_init;
2761                 }
2762         }
2763
2764         memcpy(netdev->dev_addr, adapter->hw.mac.addr, netdev->addr_len);
2765         memcpy(netdev->perm_addr, adapter->hw.mac.addr, netdev->addr_len);
2766
2767         if (!is_valid_ether_addr(netdev->perm_addr)) {
2768                 dev_err(&pdev->dev, "Invalid MAC Address: %pM\n",
2769                         netdev->dev_addr);
2770                 err = -EIO;
2771                 goto err_hw_init;
2772         }
2773
2774         setup_timer(&adapter->watchdog_timer, &igbvf_watchdog,
2775                     (unsigned long) adapter);
2776
2777         INIT_WORK(&adapter->reset_task, igbvf_reset_task);
2778         INIT_WORK(&adapter->watchdog_task, igbvf_watchdog_task);
2779
2780         /* ring size defaults */
2781         adapter->rx_ring->count = 1024;
2782         adapter->tx_ring->count = 1024;
2783
2784         /* reset the hardware with the new settings */
2785         igbvf_reset(adapter);
2786
2787         strcpy(netdev->name, "eth%d");
2788         err = register_netdev(netdev);
2789         if (err)
2790                 goto err_hw_init;
2791
2792         /* tell the stack to leave us alone until igbvf_open() is called */
2793         netif_carrier_off(netdev);
2794         netif_stop_queue(netdev);
2795
2796         igbvf_print_device_info(adapter);
2797
2798         igbvf_initialize_last_counter_stats(adapter);
2799
2800         return 0;
2801
2802 err_hw_init:
2803         kfree(adapter->tx_ring);
2804         kfree(adapter->rx_ring);
2805 err_sw_init:
2806         igbvf_reset_interrupt_capability(adapter);
2807         iounmap(adapter->hw.hw_addr);
2808 err_ioremap:
2809         free_netdev(netdev);
2810 err_alloc_etherdev:
2811         pci_release_regions(pdev);
2812 err_pci_reg:
2813 err_dma:
2814         pci_disable_device(pdev);
2815         return err;
2816 }
2817
2818 /**
2819  * igbvf_remove - Device Removal Routine
2820  * @pdev: PCI device information struct
2821  *
2822  * igbvf_remove is called by the PCI subsystem to alert the driver
2823  * that it should release a PCI device.  The could be caused by a
2824  * Hot-Plug event, or because the driver is going to be removed from
2825  * memory.
2826  **/
2827 static void __devexit igbvf_remove(struct pci_dev *pdev)
2828 {
2829         struct net_device *netdev = pci_get_drvdata(pdev);
2830         struct igbvf_adapter *adapter = netdev_priv(netdev);
2831         struct e1000_hw *hw = &adapter->hw;
2832
2833         /*
2834          * flush_scheduled work may reschedule our watchdog task, so
2835          * explicitly disable watchdog tasks from being rescheduled
2836          */
2837         set_bit(__IGBVF_DOWN, &adapter->state);
2838         del_timer_sync(&adapter->watchdog_timer);
2839
2840         flush_scheduled_work();
2841
2842         unregister_netdev(netdev);
2843
2844         igbvf_reset_interrupt_capability(adapter);
2845
2846         /*
2847          * it is important to delete the napi struct prior to freeing the
2848          * rx ring so that you do not end up with null pointer refs
2849          */
2850         netif_napi_del(&adapter->rx_ring->napi);
2851         kfree(adapter->tx_ring);
2852         kfree(adapter->rx_ring);
2853
2854         iounmap(hw->hw_addr);
2855         if (hw->flash_address)
2856                 iounmap(hw->flash_address);
2857         pci_release_regions(pdev);
2858
2859         free_netdev(netdev);
2860
2861         pci_disable_device(pdev);
2862 }
2863
2864 /* PCI Error Recovery (ERS) */
2865 static struct pci_error_handlers igbvf_err_handler = {
2866         .error_detected = igbvf_io_error_detected,
2867         .slot_reset = igbvf_io_slot_reset,
2868         .resume = igbvf_io_resume,
2869 };
2870
2871 static DEFINE_PCI_DEVICE_TABLE(igbvf_pci_tbl) = {
2872         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82576_VF), board_vf },
2873         { } /* terminate list */
2874 };
2875 MODULE_DEVICE_TABLE(pci, igbvf_pci_tbl);
2876
2877 /* PCI Device API Driver */
2878 static struct pci_driver igbvf_driver = {
2879         .name     = igbvf_driver_name,
2880         .id_table = igbvf_pci_tbl,
2881         .probe    = igbvf_probe,
2882         .remove   = __devexit_p(igbvf_remove),
2883 #ifdef CONFIG_PM
2884         /* Power Management Hooks */
2885         .suspend  = igbvf_suspend,
2886         .resume   = igbvf_resume,
2887 #endif
2888         .shutdown = igbvf_shutdown,
2889         .err_handler = &igbvf_err_handler
2890 };
2891
2892 /**
2893  * igbvf_init_module - Driver Registration Routine
2894  *
2895  * igbvf_init_module is the first routine called when the driver is
2896  * loaded. All it does is register with the PCI subsystem.
2897  **/
2898 static int __init igbvf_init_module(void)
2899 {
2900         int ret;
2901         printk(KERN_INFO "%s - version %s\n",
2902                igbvf_driver_string, igbvf_driver_version);
2903         printk(KERN_INFO "%s\n", igbvf_copyright);
2904
2905         ret = pci_register_driver(&igbvf_driver);
2906
2907         return ret;
2908 }
2909 module_init(igbvf_init_module);
2910
2911 /**
2912  * igbvf_exit_module - Driver Exit Cleanup Routine
2913  *
2914  * igbvf_exit_module is called just before the driver is removed
2915  * from memory.
2916  **/
2917 static void __exit igbvf_exit_module(void)
2918 {
2919         pci_unregister_driver(&igbvf_driver);
2920 }
2921 module_exit(igbvf_exit_module);
2922
2923
2924 MODULE_AUTHOR("Intel Corporation, <e1000-devel@lists.sourceforge.net>");
2925 MODULE_DESCRIPTION("Intel(R) 82576 Virtual Function Network Driver");
2926 MODULE_LICENSE("GPL");
2927 MODULE_VERSION(DRV_VERSION);
2928
2929 /* netdev.c */