drivers/net/igbvf/netdev.c

   1 /*******************************************************************************
   2
   3   Intel(R) 82576 Virtual Function Linux driver
   4   Copyright(c) 2009 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 <net/checksum.h>
  39 #include <net/ip6_checksum.h>
  40 #include <linux/mii.h>
  41 #include <linux/ethtool.h>
  42 #include <linux/if_vlan.h>
  43 #include <linux/pm_qos_params.h>
  44
  45 #include "igbvf.h"
  46
  47 #define DRV_VERSION "1.0.0-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[] = "Copyright (c) 2009 Intel Corporation.";
  53
  54 static int igbvf_poll(struct napi_struct *napi, int budget);
  55 static void igbvf_reset(struct igbvf_adapter *);
  56 static void igbvf_set_interrupt_capability(struct igbvf_adapter *);
  57 static void igbvf_reset_interrupt_capability(struct igbvf_adapter *);
  58
  59 static struct igbvf_info igbvf_vf_info = {
  60         .mac                    = e1000_vfadapt,
  61         .flags                  = 0,
  62         .pba                    = 10,
  63         .init_ops               = e1000_init_function_pointers_vf,
  64 };
  65
  66 static const struct igbvf_info *igbvf_info_tbl[] = {
  67         [board_vf]              = &igbvf_vf_info,
  68 };
  69
  70 /**
  71  * igbvf_desc_unused - calculate if we have unused descriptors
  72  **/
  73 static int igbvf_desc_unused(struct igbvf_ring *ring)
  74 {
  75         if (ring->next_to_clean > ring->next_to_use)
  76                 return ring->next_to_clean - ring->next_to_use - 1;
  77
  78         return ring->count + ring->next_to_clean - ring->next_to_use - 1;
  79 }
  80
  81 /**
  82  * igbvf_receive_skb - helper function to handle Rx indications
  83  * @adapter: board private structure
  84  * @status: descriptor status field as written by hardware
  85  * @vlan: descriptor vlan field as written by hardware (no le/be conversion)
  86  * @skb: pointer to sk_buff to be indicated to stack
  87  **/
  88 static void igbvf_receive_skb(struct igbvf_adapter *adapter,
  89                               struct net_device *netdev,
  90                               struct sk_buff *skb,
  91                               u32 status, u16 vlan)
  92 {
  93         if (adapter->vlgrp && (status & E1000_RXD_STAT_VP))
  94                 vlan_hwaccel_receive_skb(skb, adapter->vlgrp,
  95                                          le16_to_cpu(vlan) &
  96                                          E1000_RXD_SPC_VLAN_MASK);
  97         else
  98                 netif_receive_skb(skb);
  99
 100         netdev->last_rx = jiffies;
 101 }
 102
 103 static inline void igbvf_rx_checksum_adv(struct igbvf_adapter *adapter,
 104                                          u32 status_err, struct sk_buff *skb)
 105 {
 106         skb->ip_summed = CHECKSUM_NONE;
 107
 108         /* Ignore Checksum bit is set or checksum is disabled through ethtool */
 109         if ((status_err & E1000_RXD_STAT_IXSM) ||
 110             (adapter->flags & IGBVF_FLAG_RX_CSUM_DISABLED))
 111                 return;
 112
 113         /* TCP/UDP checksum error bit is set */
 114         if (status_err &
 115             (E1000_RXDEXT_STATERR_TCPE | E1000_RXDEXT_STATERR_IPE)) {
 116                 /* let the stack verify checksum errors */
 117                 adapter->hw_csum_err++;
 118                 return;
 119         }
 120
 121         /* It must be a TCP or UDP packet with a valid checksum */
 122         if (status_err & (E1000_RXD_STAT_TCPCS | E1000_RXD_STAT_UDPCS))
 123                 skb->ip_summed = CHECKSUM_UNNECESSARY;
 124
 125         adapter->hw_csum_good++;
 126 }
 127
 128 /**
 129  * igbvf_alloc_rx_buffers - Replace used receive buffers; packet split
 130  * @rx_ring: address of ring structure to repopulate
 131  * @cleaned_count: number of buffers to repopulate
 132  **/
 133 static void igbvf_alloc_rx_buffers(struct igbvf_ring *rx_ring,
 134                                    int cleaned_count)
 135 {
 136         struct igbvf_adapter *adapter = rx_ring->adapter;
 137         struct net_device *netdev = adapter->netdev;
 138         struct pci_dev *pdev = adapter->pdev;
 139         union e1000_adv_rx_desc *rx_desc;
 140         struct igbvf_buffer *buffer_info;
 141         struct sk_buff *skb;
 142         unsigned int i;
 143         int bufsz;
 144
 145         i = rx_ring->next_to_use;
 146         buffer_info = &rx_ring->buffer_info[i];
 147
 148         if (adapter->rx_ps_hdr_size)
 149                 bufsz = adapter->rx_ps_hdr_size;
 150         else
 151                 bufsz = adapter->rx_buffer_len;
 152
 153         while (cleaned_count--) {
 154                 rx_desc = IGBVF_RX_DESC_ADV(*rx_ring, i);
 155
 156                 if (adapter->rx_ps_hdr_size && !buffer_info->page_dma) {
 157                         if (!buffer_info->page) {
 158                                 buffer_info->page = alloc_page(GFP_ATOMIC);
 159                                 if (!buffer_info->page) {
 160                                         adapter->alloc_rx_buff_failed++;
 161                                         goto no_buffers;
 162                                 }
 163                                 buffer_info->page_offset = 0;
 164                         } else {
 165                                 buffer_info->page_offset ^= PAGE_SIZE / 2;
 166                         }
 167                         buffer_info->page_dma =
 168                                 pci_map_page(pdev, buffer_info->page,
 169                                              buffer_info->page_offset,
 170                                              PAGE_SIZE / 2,
 171                                              PCI_DMA_FROMDEVICE);
 172                 }
 173
 174                 if (!buffer_info->skb) {
 175                         skb = netdev_alloc_skb(netdev, bufsz + NET_IP_ALIGN);
 176                         if (!skb) {
 177                                 adapter->alloc_rx_buff_failed++;
 178                                 goto no_buffers;
 179                         }
 180
 181                         /* Make buffer alignment 2 beyond a 16 byte boundary
 182                          * this will result in a 16 byte aligned IP header after
 183                          * the 14 byte MAC header is removed
 184                          */
 185                         skb_reserve(skb, NET_IP_ALIGN);
 186
 187                         buffer_info->skb = skb;
 188                         buffer_info->dma = pci_map_single(pdev, skb->data,
 189                                                           bufsz,
 190                                                           PCI_DMA_FROMDEVICE);
 191                 }
 192                 /* Refresh the desc even if buffer_addrs didn't change because
 193                  * each write-back erases this info. */
 194                 if (adapter->rx_ps_hdr_size) {
 195                         rx_desc->read.pkt_addr =
 196                              cpu_to_le64(buffer_info->page_dma);
 197                         rx_desc->read.hdr_addr = cpu_to_le64(buffer_info->dma);
 198                 } else {
 199                         rx_desc->read.pkt_addr =
 200                              cpu_to_le64(buffer_info->dma);
 201                         rx_desc->read.hdr_addr = 0;
 202                 }
 203
 204                 i++;
 205                 if (i == rx_ring->count)
 206                         i = 0;
 207                 buffer_info = &rx_ring->buffer_info[i];
 208         }
 209
 210 no_buffers:
 211         if (rx_ring->next_to_use != i) {
 212                 rx_ring->next_to_use = i;
 213                 if (i == 0)
 214                         i = (rx_ring->count - 1);
 215                 else
 216                         i--;
 217
 218                 /* Force memory writes to complete before letting h/w
 219                  * know there are new descriptors to fetch.  (Only
 220                  * applicable for weak-ordered memory model archs,
 221                  * such as IA-64). */
 222                 wmb();
 223                 writel(i, adapter->hw.hw_addr + rx_ring->tail);
 224         }
 225 }
 226
 227 /**
 228  * igbvf_clean_rx_irq - Send received data up the network stack; legacy
 229  * @adapter: board private structure
 230  *
 231  * the return value indicates whether actual cleaning was done, there
 232  * is no guarantee that everything was cleaned
 233  **/
 234 static bool igbvf_clean_rx_irq(struct igbvf_adapter *adapter,
 235                                int *work_done, int work_to_do)
 236 {
 237         struct igbvf_ring *rx_ring = adapter->rx_ring;
 238         struct net_device *netdev = adapter->netdev;
 239         struct pci_dev *pdev = adapter->pdev;
 240         union e1000_adv_rx_desc *rx_desc, *next_rxd;
 241         struct igbvf_buffer *buffer_info, *next_buffer;
 242         struct sk_buff *skb;
 243         bool cleaned = false;
 244         int cleaned_count = 0;
 245         unsigned int total_bytes = 0, total_packets = 0;
 246         unsigned int i;
 247         u32 length, hlen, staterr;
 248
 249         i = rx_ring->next_to_clean;
 250         rx_desc = IGBVF_RX_DESC_ADV(*rx_ring, i);
 251         staterr = le32_to_cpu(rx_desc->wb.upper.status_error);
 252
 253         while (staterr & E1000_RXD_STAT_DD) {
 254                 if (*work_done >= work_to_do)
 255                         break;
 256                 (*work_done)++;
 257
 258                 buffer_info = &rx_ring->buffer_info[i];
 259
 260                 /* HW will not DMA in data larger than the given buffer, even
 261                  * if it parses the (NFS, of course) header to be larger.  In
 262                  * that case, it fills the header buffer and spills the rest
 263                  * into the page.
 264                  */
 265                 hlen = (le16_to_cpu(rx_desc->wb.lower.lo_dword.hs_rss.hdr_info) &
 266                   E1000_RXDADV_HDRBUFLEN_MASK) >> E1000_RXDADV_HDRBUFLEN_SHIFT;
 267                 if (hlen > adapter->rx_ps_hdr_size)
 268                         hlen = adapter->rx_ps_hdr_size;
 269
 270                 length = le16_to_cpu(rx_desc->wb.upper.length);
 271                 cleaned = true;
 272                 cleaned_count++;
 273
 274                 skb = buffer_info->skb;
 275                 prefetch(skb->data - NET_IP_ALIGN);
 276                 buffer_info->skb = NULL;
 277                 if (!adapter->rx_ps_hdr_size) {
 278                         pci_unmap_single(pdev, buffer_info->dma,
 279                                          adapter->rx_buffer_len,
 280                                          PCI_DMA_FROMDEVICE);
 281                         buffer_info->dma = 0;
 282                         skb_put(skb, length);
 283                         goto send_up;
 284                 }
 285
 286                 if (!skb_shinfo(skb)->nr_frags) {
 287                         pci_unmap_single(pdev, buffer_info->dma,
 288                                          adapter->rx_ps_hdr_size,
 289                                          PCI_DMA_FROMDEVICE);
 290                         skb_put(skb, hlen);
 291                 }
 292
 293                 if (length) {
 294                         pci_unmap_page(pdev, buffer_info->page_dma,
 295                                        PAGE_SIZE / 2,
 296                                        PCI_DMA_FROMDEVICE);
 297                         buffer_info->page_dma = 0;
 298
 299                         skb_fill_page_desc(skb, skb_shinfo(skb)->nr_frags++,
 300                                            buffer_info->page,
 301                                            buffer_info->page_offset,
 302                                            length);
 303
 304                         if ((adapter->rx_buffer_len > (PAGE_SIZE / 2)) ||
 305                             (page_count(buffer_info->page) != 1))
 306                                 buffer_info->page = NULL;
 307                         else
 308                                 get_page(buffer_info->page);
 309
 310                         skb->len += length;
 311                         skb->data_len += length;
 312                         skb->truesize += length;
 313                 }
 314 send_up:
 315                 i++;
 316                 if (i == rx_ring->count)
 317                         i = 0;
 318                 next_rxd = IGBVF_RX_DESC_ADV(*rx_ring, i);
 319                 prefetch(next_rxd);
 320                 next_buffer = &rx_ring->buffer_info[i];
 321
 322                 if (!(staterr & E1000_RXD_STAT_EOP)) {
 323                         buffer_info->skb = next_buffer->skb;
 324                         buffer_info->dma = next_buffer->dma;
 325                         next_buffer->skb = skb;
 326                         next_buffer->dma = 0;
 327                         goto next_desc;
 328                 }
 329
 330                 if (staterr & E1000_RXDEXT_ERR_FRAME_ERR_MASK) {
 331                         dev_kfree_skb_irq(skb);
 332                         goto next_desc;
 333                 }
 334
 335                 total_bytes += skb->len;
 336                 total_packets++;
 337
 338                 igbvf_rx_checksum_adv(adapter, staterr, skb);
 339
 340                 skb->protocol = eth_type_trans(skb, netdev);
 341
 342                 igbvf_receive_skb(adapter, netdev, skb, staterr,
 343                                   rx_desc->wb.upper.vlan);
 344
 345                 netdev->last_rx = jiffies;
 346
 347 next_desc:
 348                 rx_desc->wb.upper.status_error = 0;
 349
 350                 /* return some buffers to hardware, one at a time is too slow */
 351                 if (cleaned_count >= IGBVF_RX_BUFFER_WRITE) {
 352                         igbvf_alloc_rx_buffers(rx_ring, cleaned_count);
 353                         cleaned_count = 0;
 354                 }
 355
 356                 /* use prefetched values */
 357                 rx_desc = next_rxd;
 358                 buffer_info = next_buffer;
 359
 360                 staterr = le32_to_cpu(rx_desc->wb.upper.status_error);
 361         }
 362
 363         rx_ring->next_to_clean = i;
 364         cleaned_count = igbvf_desc_unused(rx_ring);
 365
 366         if (cleaned_count)
 367                 igbvf_alloc_rx_buffers(rx_ring, cleaned_count);
 368
 369         adapter->total_rx_packets += total_packets;
 370         adapter->total_rx_bytes += total_bytes;
 371         adapter->net_stats.rx_bytes += total_bytes;
 372         adapter->net_stats.rx_packets += total_packets;
 373         return cleaned;
 374 }
 375
 376 static void igbvf_put_txbuf(struct igbvf_adapter *adapter,
 377                             struct igbvf_buffer *buffer_info)
 378 {
 379         buffer_info->dma = 0;
 380         if (buffer_info->skb) {
 381                 skb_dma_unmap(&adapter->pdev->dev, buffer_info->skb,
 382                               DMA_TO_DEVICE);
 383                 dev_kfree_skb_any(buffer_info->skb);
 384                 buffer_info->skb = NULL;
 385         }
 386         buffer_info->time_stamp = 0;
 387 }
 388
 389 static void igbvf_print_tx_hang(struct igbvf_adapter *adapter)
 390 {
 391         struct igbvf_ring *tx_ring = adapter->tx_ring;
 392         unsigned int i = tx_ring->next_to_clean;
 393         unsigned int eop = tx_ring->buffer_info[i].next_to_watch;
 394         union e1000_adv_tx_desc *eop_desc = IGBVF_TX_DESC_ADV(*tx_ring, eop);
 395
 396         /* detected Tx unit hang */
 397         dev_err(&adapter->pdev->dev,
 398                 "Detected Tx Unit Hang:\n"
 399                 "  TDH                  <%x>\n"
 400                 "  TDT                  <%x>\n"
 401                 "  next_to_use          <%x>\n"
 402                 "  next_to_clean        <%x>\n"
 403                 "buffer_info[next_to_clean]:\n"
 404                 "  time_stamp           <%lx>\n"
 405                 "  next_to_watch        <%x>\n"
 406                 "  jiffies              <%lx>\n"
 407                 "  next_to_watch.status <%x>\n",
 408                 readl(adapter->hw.hw_addr + tx_ring->head),
 409                 readl(adapter->hw.hw_addr + tx_ring->tail),
 410                 tx_ring->next_to_use,
 411                 tx_ring->next_to_clean,
 412                 tx_ring->buffer_info[eop].time_stamp,
 413                 eop,
 414                 jiffies,
 415                 eop_desc->wb.status);
 416 }
 417
 418 /**
 419  * igbvf_setup_tx_resources - allocate Tx resources (Descriptors)
 420  * @adapter: board private structure
 421  *
 422  * Return 0 on success, negative on failure
 423  **/
 424 int igbvf_setup_tx_resources(struct igbvf_adapter *adapter,
 425                              struct igbvf_ring *tx_ring)
 426 {
 427         struct pci_dev *pdev = adapter->pdev;
 428         int size;
 429
 430         size = sizeof(struct igbvf_buffer) * tx_ring->count;
 431         tx_ring->buffer_info = vmalloc(size);
 432         if (!tx_ring->buffer_info)
 433                 goto err;
 434         memset(tx_ring->buffer_info, 0, size);
 435
 436         /* round up to nearest 4K */
 437         tx_ring->size = tx_ring->count * sizeof(union e1000_adv_tx_desc);
 438         tx_ring->size = ALIGN(tx_ring->size, 4096);
 439
 440         tx_ring->desc = pci_alloc_consistent(pdev, tx_ring->size,
 441                                              &tx_ring->dma);
 442
 443         if (!tx_ring->desc)
 444                 goto err;
 445
 446         tx_ring->adapter = adapter;
 447         tx_ring->next_to_use = 0;
 448         tx_ring->next_to_clean = 0;
 449
 450         return 0;
 451 err:
 452         vfree(tx_ring->buffer_info);
 453         dev_err(&adapter->pdev->dev,
 454                 "Unable to allocate memory for the transmit descriptor ring\n");
 455         return -ENOMEM;
 456 }
 457
 458 /**
 459  * igbvf_setup_rx_resources - allocate Rx resources (Descriptors)
 460  * @adapter: board private structure
 461  *
 462  * Returns 0 on success, negative on failure
 463  **/
 464 int igbvf_setup_rx_resources(struct igbvf_adapter *adapter,
 465                              struct igbvf_ring *rx_ring)
 466 {
 467         struct pci_dev *pdev = adapter->pdev;
 468         int size, desc_len;
 469
 470         size = sizeof(struct igbvf_buffer) * rx_ring->count;
 471         rx_ring->buffer_info = vmalloc(size);
 472         if (!rx_ring->buffer_info)
 473                 goto err;
 474         memset(rx_ring->buffer_info, 0, size);
 475
 476         desc_len = sizeof(union e1000_adv_rx_desc);
 477
 478         /* Round up to nearest 4K */
 479         rx_ring->size = rx_ring->count * desc_len;
 480         rx_ring->size = ALIGN(rx_ring->size, 4096);
 481
 482         rx_ring->desc = pci_alloc_consistent(pdev, rx_ring->size,
 483                                              &rx_ring->dma);
 484
 485         if (!rx_ring->desc)
 486                 goto err;
 487
 488         rx_ring->next_to_clean = 0;
 489         rx_ring->next_to_use = 0;
 490
 491         rx_ring->adapter = adapter;
 492
 493         return 0;
 494
 495 err:
 496         vfree(rx_ring->buffer_info);
 497         rx_ring->buffer_info = NULL;
 498         dev_err(&adapter->pdev->dev,
 499                 "Unable to allocate memory for the receive descriptor ring\n");
 500         return -ENOMEM;
 501 }
 502
 503 /**
 504  * igbvf_clean_tx_ring - Free Tx Buffers
 505  * @tx_ring: ring to be cleaned
 506  **/
 507 static void igbvf_clean_tx_ring(struct igbvf_ring *tx_ring)
 508 {
 509         struct igbvf_adapter *adapter = tx_ring->adapter;
 510         struct igbvf_buffer *buffer_info;
 511         unsigned long size;
 512         unsigned int i;
 513
 514         if (!tx_ring->buffer_info)
 515                 return;
 516
 517         /* Free all the Tx ring sk_buffs */
 518         for (i = 0; i < tx_ring->count; i++) {
 519                 buffer_info = &tx_ring->buffer_info[i];
 520                 igbvf_put_txbuf(adapter, buffer_info);
 521         }
 522
 523         size = sizeof(struct igbvf_buffer) * tx_ring->count;
 524         memset(tx_ring->buffer_info, 0, size);
 525
 526         /* Zero out the descriptor ring */
 527         memset(tx_ring->desc, 0, tx_ring->size);
 528
 529         tx_ring->next_to_use = 0;
 530         tx_ring->next_to_clean = 0;
 531
 532         writel(0, adapter->hw.hw_addr + tx_ring->head);
 533         writel(0, adapter->hw.hw_addr + tx_ring->tail);
 534 }
 535
 536 /**
 537  * igbvf_free_tx_resources - Free Tx Resources per Queue
 538  * @tx_ring: ring to free resources from
 539  *
 540  * Free all transmit software resources
 541  **/
 542 void igbvf_free_tx_resources(struct igbvf_ring *tx_ring)
 543 {
 544         struct pci_dev *pdev = tx_ring->adapter->pdev;
 545
 546         igbvf_clean_tx_ring(tx_ring);
 547
 548         vfree(tx_ring->buffer_info);
 549         tx_ring->buffer_info = NULL;
 550
 551         pci_free_consistent(pdev, tx_ring->size, tx_ring->desc, tx_ring->dma);
 552
 553         tx_ring->desc = NULL;
 554 }
 555
 556 /**
 557  * igbvf_clean_rx_ring - Free Rx Buffers per Queue
 558  * @adapter: board private structure
 559  **/
 560 static void igbvf_clean_rx_ring(struct igbvf_ring *rx_ring)
 561 {
 562         struct igbvf_adapter *adapter = rx_ring->adapter;
 563         struct igbvf_buffer *buffer_info;
 564         struct pci_dev *pdev = adapter->pdev;
 565         unsigned long size;
 566         unsigned int i;
 567
 568         if (!rx_ring->buffer_info)
 569                 return;
 570
 571         /* Free all the Rx ring sk_buffs */
 572         for (i = 0; i < rx_ring->count; i++) {
 573                 buffer_info = &rx_ring->buffer_info[i];
 574                 if (buffer_info->dma) {
 575                         if (adapter->rx_ps_hdr_size){
 576                                 pci_unmap_single(pdev, buffer_info->dma,
 577                                                  adapter->rx_ps_hdr_size,
 578                                                  PCI_DMA_FROMDEVICE);
 579                         } else {
 580                                 pci_unmap_single(pdev, buffer_info->dma,
 581                                                  adapter->rx_buffer_len,
 582                                                  PCI_DMA_FROMDEVICE);
 583                         }
 584                         buffer_info->dma = 0;
 585                 }
 586
 587                 if (buffer_info->skb) {
 588                         dev_kfree_skb(buffer_info->skb);
 589                         buffer_info->skb = NULL;
 590                 }
 591
 592                 if (buffer_info->page) {
 593                         if (buffer_info->page_dma)
 594                                 pci_unmap_page(pdev, buffer_info->page_dma,
 595                                                PAGE_SIZE / 2,
 596                                                PCI_DMA_FROMDEVICE);
 597                         put_page(buffer_info->page);
 598                         buffer_info->page = NULL;
 599                         buffer_info->page_dma = 0;
 600                         buffer_info->page_offset = 0;
 601                 }
 602         }
 603
 604         size = sizeof(struct igbvf_buffer) * rx_ring->count;
 605         memset(rx_ring->buffer_info, 0, size);
 606
 607         /* Zero out the descriptor ring */
 608         memset(rx_ring->desc, 0, rx_ring->size);
 609
 610         rx_ring->next_to_clean = 0;
 611         rx_ring->next_to_use = 0;
 612
 613         writel(0, adapter->hw.hw_addr + rx_ring->head);
 614         writel(0, adapter->hw.hw_addr + rx_ring->tail);
 615 }
 616
 617 /**
 618  * igbvf_free_rx_resources - Free Rx Resources
 619  * @rx_ring: ring to clean the resources from
 620  *
 621  * Free all receive software resources
 622  **/
 623
 624 void igbvf_free_rx_resources(struct igbvf_ring *rx_ring)
 625 {
 626         struct pci_dev *pdev = rx_ring->adapter->pdev;
 627
 628         igbvf_clean_rx_ring(rx_ring);
 629
 630         vfree(rx_ring->buffer_info);
 631         rx_ring->buffer_info = NULL;
 632
 633         dma_free_coherent(&pdev->dev, rx_ring->size, rx_ring->desc,
 634                           rx_ring->dma);
 635         rx_ring->desc = NULL;
 636 }
 637
 638 /**
 639  * igbvf_update_itr - update the dynamic ITR value based on statistics
 640  * @adapter: pointer to adapter
 641  * @itr_setting: current adapter->itr
 642  * @packets: the number of packets during this measurement interval
 643  * @bytes: the number of bytes during this measurement interval
 644  *
 645  *      Stores a new ITR value based on packets and byte
 646  *      counts during the last interrupt.  The advantage of per interrupt
 647  *      computation is faster updates and more accurate ITR for the current
 648  *      traffic pattern.  Constants in this function were computed
 649  *      based on theoretical maximum wire speed and thresholds were set based
 650  *      on testing data as well as attempting to minimize response time
 651  *      while increasing bulk throughput.  This functionality is controlled
 652  *      by the InterruptThrottleRate module parameter.
 653  **/
 654 static unsigned int igbvf_update_itr(struct igbvf_adapter *adapter,
 655                                      u16 itr_setting, int packets,
 656                                      int bytes)
 657 {
 658         unsigned int retval = itr_setting;
 659
 660         if (packets == 0)
 661                 goto update_itr_done;
 662
 663         switch (itr_setting) {
 664         case lowest_latency:
 665                 /* handle TSO and jumbo frames */
 666                 if (bytes/packets > 8000)
 667                         retval = bulk_latency;
 668                 else if ((packets < 5) && (bytes > 512))
 669                         retval = low_latency;
 670                 break;
 671         case low_latency:  /* 50 usec aka 20000 ints/s */
 672                 if (bytes > 10000) {
 673                         /* this if handles the TSO accounting */
 674                         if (bytes/packets > 8000)
 675                                 retval = bulk_latency;
 676                         else if ((packets < 10) || ((bytes/packets) > 1200))
 677                                 retval = bulk_latency;
 678                         else if ((packets > 35))
 679                                 retval = lowest_latency;
 680                 } else if (bytes/packets > 2000) {
 681                         retval = bulk_latency;
 682                 } else if (packets <= 2 && bytes < 512) {
 683                         retval = lowest_latency;
 684                 }
 685                 break;
 686         case bulk_latency: /* 250 usec aka 4000 ints/s */
 687                 if (bytes > 25000) {
 688                         if (packets > 35)
 689                                 retval = low_latency;
 690                 } else if (bytes < 6000) {
 691                         retval = low_latency;
 692                 }
 693                 break;
 694         }
 695
 696 update_itr_done:
 697         return retval;
 698 }
 699
 700 static void igbvf_set_itr(struct igbvf_adapter *adapter)
 701 {
 702         struct e1000_hw *hw = &adapter->hw;
 703         u16 current_itr;
 704         u32 new_itr = adapter->itr;
 705
 706         adapter->tx_itr = igbvf_update_itr(adapter, adapter->tx_itr,
 707                                            adapter->total_tx_packets,
 708                                            adapter->total_tx_bytes);
 709         /* conservative mode (itr 3) eliminates the lowest_latency setting */
 710         if (adapter->itr_setting == 3 && adapter->tx_itr == lowest_latency)
 711                 adapter->tx_itr = low_latency;
 712
 713         adapter->rx_itr = igbvf_update_itr(adapter, adapter->rx_itr,
 714                                            adapter->total_rx_packets,
 715                                            adapter->total_rx_bytes);
 716         /* conservative mode (itr 3) eliminates the lowest_latency setting */
 717         if (adapter->itr_setting == 3 && adapter->rx_itr == lowest_latency)
 718                 adapter->rx_itr = low_latency;
 719
 720         current_itr = max(adapter->rx_itr, adapter->tx_itr);
 721
 722         switch (current_itr) {
 723         /* counts and packets in update_itr are dependent on these numbers */
 724         case lowest_latency:
 725                 new_itr = 70000;
 726                 break;
 727         case low_latency:
 728                 new_itr = 20000; /* aka hwitr = ~200 */
 729                 break;
 730         case bulk_latency:
 731                 new_itr = 4000;
 732                 break;
 733         default:
 734                 break;
 735         }
 736
 737         if (new_itr != adapter->itr) {
 738                 /*
 739                  * this attempts to bias the interrupt rate towards Bulk
 740                  * by adding intermediate steps when interrupt rate is
 741                  * increasing
 742                  */
 743                 new_itr = new_itr > adapter->itr ?
 744                              min(adapter->itr + (new_itr >> 2), new_itr) :
 745                              new_itr;
 746                 adapter->itr = new_itr;
 747                 adapter->rx_ring->itr_val = 1952;
 748
 749                 if (adapter->msix_entries)
 750                         adapter->rx_ring->set_itr = 1;
 751                 else
 752                         ew32(ITR, 1952);
 753         }
 754 }
 755
 756 /**
 757  * igbvf_clean_tx_irq - Reclaim resources after transmit completes
 758  * @adapter: board private structure
 759  * returns true if ring is completely cleaned
 760  **/
 761 static bool igbvf_clean_tx_irq(struct igbvf_ring *tx_ring)
 762 {
 763         struct igbvf_adapter *adapter = tx_ring->adapter;
 764         struct e1000_hw *hw = &adapter->hw;
 765         struct net_device *netdev = adapter->netdev;
 766         struct igbvf_buffer *buffer_info;
 767         struct sk_buff *skb;
 768         union e1000_adv_tx_desc *tx_desc, *eop_desc;
 769         unsigned int total_bytes = 0, total_packets = 0;
 770         unsigned int i, eop, count = 0;
 771         bool cleaned = false;
 772
 773         i = tx_ring->next_to_clean;
 774         eop = tx_ring->buffer_info[i].next_to_watch;
 775         eop_desc = IGBVF_TX_DESC_ADV(*tx_ring, eop);
 776
 777         while ((eop_desc->wb.status & cpu_to_le32(E1000_TXD_STAT_DD)) &&
 778                (count < tx_ring->count)) {
 779                 for (cleaned = false; !cleaned; count++) {
 780                         tx_desc = IGBVF_TX_DESC_ADV(*tx_ring, i);
 781                         buffer_info = &tx_ring->buffer_info[i];
 782                         cleaned = (i == eop);
 783                         skb = buffer_info->skb;
 784
 785                         if (skb) {
 786                                 unsigned int segs, bytecount;
 787
 788                                 /* gso_segs is currently only valid for tcp */
 789                                 segs = skb_shinfo(skb)->gso_segs ?: 1;
 790                                 /* multiply data chunks by size of headers */
 791                                 bytecount = ((segs - 1) * skb_headlen(skb)) +
 792                                             skb->len;
 793                                 total_packets += segs;
 794                                 total_bytes += bytecount;
 795                         }
 796
 797                         igbvf_put_txbuf(adapter, buffer_info);
 798                         tx_desc->wb.status = 0;
 799
 800                         i++;
 801                         if (i == tx_ring->count)
 802                                 i = 0;
 803                 }
 804                 eop = tx_ring->buffer_info[i].next_to_watch;
 805                 eop_desc = IGBVF_TX_DESC_ADV(*tx_ring, eop);
 806         }
 807
 808         tx_ring->next_to_clean = i;
 809
 810         if (unlikely(count &&
 811                      netif_carrier_ok(netdev) &&
 812                      igbvf_desc_unused(tx_ring) >= IGBVF_TX_QUEUE_WAKE)) {
 813                 /* Make sure that anybody stopping the queue after this
 814                  * sees the new next_to_clean.
 815                  */
 816                 smp_mb();
 817                 if (netif_queue_stopped(netdev) &&
 818                     !(test_bit(__IGBVF_DOWN, &adapter->state))) {
 819                         netif_wake_queue(netdev);
 820                         ++adapter->restart_queue;
 821                 }
 822         }
 823
 824         if (adapter->detect_tx_hung) {
 825                 /* Detect a transmit hang in hardware, this serializes the
 826                  * check with the clearing of time_stamp and movement of i */
 827                 adapter->detect_tx_hung = false;
 828                 if (tx_ring->buffer_info[i].time_stamp &&
 829                     time_after(jiffies, tx_ring->buffer_info[i].time_stamp +
 830                                (adapter->tx_timeout_factor * HZ))
 831                     && !(er32(STATUS) & E1000_STATUS_TXOFF)) {
 832
 833                         tx_desc = IGBVF_TX_DESC_ADV(*tx_ring, i);
 834                         /* detected Tx unit hang */
 835                         igbvf_print_tx_hang(adapter);
 836
 837                         netif_stop_queue(netdev);
 838                 }
 839         }
 840         adapter->net_stats.tx_bytes += total_bytes;
 841         adapter->net_stats.tx_packets += total_packets;
 842         return (count < tx_ring->count);
 843 }
 844
 845 static irqreturn_t igbvf_msix_other(int irq, void *data)
 846 {
 847         struct net_device *netdev = data;
 848         struct igbvf_adapter *adapter = netdev_priv(netdev);
 849         struct e1000_hw *hw = &adapter->hw;
 850
 851         adapter->int_counter1++;
 852
 853         netif_carrier_off(netdev);
 854         hw->mac.get_link_status = 1;
 855         if (!test_bit(__IGBVF_DOWN, &adapter->state))
 856                 mod_timer(&adapter->watchdog_timer, jiffies + 1);
 857
 858         ew32(EIMS, adapter->eims_other);
 859
 860         return IRQ_HANDLED;
 861 }
 862
 863 static irqreturn_t igbvf_intr_msix_tx(int irq, void *data)
 864 {
 865         struct net_device *netdev = data;
 866         struct igbvf_adapter *adapter = netdev_priv(netdev);
 867         struct e1000_hw *hw = &adapter->hw;
 868         struct igbvf_ring *tx_ring = adapter->tx_ring;
 869
 870
 871         adapter->total_tx_bytes = 0;
 872         adapter->total_tx_packets = 0;
 873
 874         /* auto mask will automatically reenable the interrupt when we write
 875          * EICS */
 876         if (!igbvf_clean_tx_irq(tx_ring))
 877                 /* Ring was not completely cleaned, so fire another interrupt */
 878                 ew32(EICS, tx_ring->eims_value);
 879         else
 880                 ew32(EIMS, tx_ring->eims_value);
 881
 882         return IRQ_HANDLED;
 883 }
 884
 885 static irqreturn_t igbvf_intr_msix_rx(int irq, void *data)
 886 {
 887         struct net_device *netdev = data;
 888         struct igbvf_adapter *adapter = netdev_priv(netdev);
 889
 890         adapter->int_counter0++;
 891
 892         /* Write the ITR value calculated at the end of the
 893          * previous interrupt.
 894          */
 895         if (adapter->rx_ring->set_itr) {
 896                 writel(adapter->rx_ring->itr_val,
 897                        adapter->hw.hw_addr + adapter->rx_ring->itr_register);
 898                 adapter->rx_ring->set_itr = 0;
 899         }
 900
 901         if (napi_schedule_prep(&adapter->rx_ring->napi)) {
 902                 adapter->total_rx_bytes = 0;
 903                 adapter->total_rx_packets = 0;
 904                 __napi_schedule(&adapter->rx_ring->napi);
 905         }
 906
 907         return IRQ_HANDLED;
 908 }
 909
 910 #define IGBVF_NO_QUEUE -1
 911
 912 static void igbvf_assign_vector(struct igbvf_adapter *adapter, int rx_queue,
 913                                 int tx_queue, int msix_vector)
 914 {
 915         struct e1000_hw *hw = &adapter->hw;
 916         u32 ivar, index;
 917
 918         /* 82576 uses a table-based method for assigning vectors.
 919            Each queue has a single entry in the table to which we write
 920            a vector number along with a "valid" bit.  Sadly, the layout
 921            of the table is somewhat counterintuitive. */
 922         if (rx_queue > IGBVF_NO_QUEUE) {
 923                 index = (rx_queue >> 1);
 924                 ivar = array_er32(IVAR0, index);
 925                 if (rx_queue & 0x1) {
 926                         /* vector goes into third byte of register */
 927                         ivar = ivar & 0xFF00FFFF;
 928                         ivar |= (msix_vector | E1000_IVAR_VALID) << 16;
 929                 } else {
 930                         /* vector goes into low byte of register */
 931                         ivar = ivar & 0xFFFFFF00;
 932                         ivar |= msix_vector | E1000_IVAR_VALID;
 933                 }
 934                 adapter->rx_ring[rx_queue].eims_value = 1 << msix_vector;
 935                 array_ew32(IVAR0, index, ivar);
 936         }
 937         if (tx_queue > IGBVF_NO_QUEUE) {
 938                 index = (tx_queue >> 1);
 939                 ivar = array_er32(IVAR0, index);
 940                 if (tx_queue & 0x1) {
 941                         /* vector goes into high byte of register */
 942                         ivar = ivar & 0x00FFFFFF;
 943                         ivar |= (msix_vector | E1000_IVAR_VALID) << 24;
 944                 } else {
 945                         /* vector goes into second byte of register */
 946                         ivar = ivar & 0xFFFF00FF;
 947                         ivar |= (msix_vector | E1000_IVAR_VALID) << 8;
 948                 }
 949                 adapter->tx_ring[tx_queue].eims_value = 1 << msix_vector;
 950                 array_ew32(IVAR0, index, ivar);
 951         }
 952 }
 953
 954 /**
 955  * igbvf_configure_msix - Configure MSI-X hardware
 956  *
 957  * igbvf_configure_msix sets up the hardware to properly
 958  * generate MSI-X interrupts.
 959  **/
 960 static void igbvf_configure_msix(struct igbvf_adapter *adapter)
 961 {
 962         u32 tmp;
 963         struct e1000_hw *hw = &adapter->hw;
 964         struct igbvf_ring *tx_ring = adapter->tx_ring;
 965         struct igbvf_ring *rx_ring = adapter->rx_ring;
 966         int vector = 0;
 967
 968         adapter->eims_enable_mask = 0;
 969
 970         igbvf_assign_vector(adapter, IGBVF_NO_QUEUE, 0, vector++);
 971         adapter->eims_enable_mask |= tx_ring->eims_value;
 972         if (tx_ring->itr_val)
 973                 writel(tx_ring->itr_val,
 974                        hw->hw_addr + tx_ring->itr_register);
 975         else
 976                 writel(1952, hw->hw_addr + tx_ring->itr_register);
 977
 978         igbvf_assign_vector(adapter, 0, IGBVF_NO_QUEUE, vector++);
 979         adapter->eims_enable_mask |= rx_ring->eims_value;
 980         if (rx_ring->itr_val)
 981                 writel(rx_ring->itr_val,
 982                        hw->hw_addr + rx_ring->itr_register);
 983         else
 984                 writel(1952, hw->hw_addr + rx_ring->itr_register);
 985
 986         /* set vector for other causes, i.e. link changes */
 987
 988         tmp = (vector++ | E1000_IVAR_VALID);
 989
 990         ew32(IVAR_MISC, tmp);
 991
 992         adapter->eims_enable_mask = (1 << (vector)) - 1;
 993         adapter->eims_other = 1 << (vector - 1);
 994         e1e_flush();
 995 }
 996
 997 static void igbvf_reset_interrupt_capability(struct igbvf_adapter *adapter)
 998 {
 999         if (adapter->msix_entries) {
1000                 pci_disable_msix(adapter->pdev);
1001                 kfree(adapter->msix_entries);
1002                 adapter->msix_entries = NULL;
1003         }
1004 }
1005
1006 /**
1007  * igbvf_set_interrupt_capability - set MSI or MSI-X if supported
1008  *
1009  * Attempt to configure interrupts using the best available
1010  * capabilities of the hardware and kernel.
1011  **/
1012 static void igbvf_set_interrupt_capability(struct igbvf_adapter *adapter)
1013 {
1014         int err = -ENOMEM;
1015         int i;
1016
1017         /* we allocate 3 vectors, 1 for tx, 1 for rx, one for pf messages */
1018         adapter->msix_entries = kcalloc(3, sizeof(struct msix_entry),
1019                                         GFP_KERNEL);
1020         if (adapter->msix_entries) {
1021                 for (i = 0; i < 3; i++)
1022                         adapter->msix_entries[i].entry = i;
1023
1024                 err = pci_enable_msix(adapter->pdev,
1025                                       adapter->msix_entries, 3);
1026         }
1027
1028         if (err) {
1029                 /* MSI-X failed */
1030                 dev_err(&adapter->pdev->dev,
1031                         "Failed to initialize MSI-X interrupts.\n");
1032                 igbvf_reset_interrupt_capability(adapter);
1033         }
1034 }
1035
1036 /**
1037  * igbvf_request_msix - Initialize MSI-X interrupts
1038  *
1039  * igbvf_request_msix allocates MSI-X vectors and requests interrupts from the
1040  * kernel.
1041  **/
1042 static int igbvf_request_msix(struct igbvf_adapter *adapter)
1043 {
1044         struct net_device *netdev = adapter->netdev;
1045         int err = 0, vector = 0;
1046
1047         if (strlen(netdev->name) < (IFNAMSIZ - 5)) {
1048                 sprintf(adapter->tx_ring->name, "%s-tx-0", netdev->name);
1049                 sprintf(adapter->rx_ring->name, "%s-rx-0", netdev->name);
1050         } else {
1051                 memcpy(adapter->tx_ring->name, netdev->name, IFNAMSIZ);
1052                 memcpy(adapter->rx_ring->name, netdev->name, IFNAMSIZ);
1053         }
1054
1055         err = request_irq(adapter->msix_entries[vector].vector,
1056                           &igbvf_intr_msix_tx, 0, adapter->tx_ring->name,
1057                           netdev);
1058         if (err)
1059                 goto out;
1060
1061         adapter->tx_ring->itr_register = E1000_EITR(vector);
1062         adapter->tx_ring->itr_val = 1952;
1063         vector++;
1064
1065         err = request_irq(adapter->msix_entries[vector].vector,
1066                           &igbvf_intr_msix_rx, 0, adapter->rx_ring->name,
1067                           netdev);
1068         if (err)
1069                 goto out;
1070
1071         adapter->rx_ring->itr_register = E1000_EITR(vector);
1072         adapter->rx_ring->itr_val = 1952;
1073         vector++;
1074
1075         err = request_irq(adapter->msix_entries[vector].vector,
1076                           &igbvf_msix_other, 0, netdev->name, netdev);
1077         if (err)
1078                 goto out;
1079
1080         igbvf_configure_msix(adapter);
1081         return 0;
1082 out:
1083         return err;
1084 }
1085
1086 /**
1087  * igbvf_alloc_queues - Allocate memory for all rings
1088  * @adapter: board private structure to initialize
1089  **/
1090 static int __devinit igbvf_alloc_queues(struct igbvf_adapter *adapter)
1091 {
1092         struct net_device *netdev = adapter->netdev;
1093
1094         adapter->tx_ring = kzalloc(sizeof(struct igbvf_ring), GFP_KERNEL);
1095         if (!adapter->tx_ring)
1096                 return -ENOMEM;
1097
1098         adapter->rx_ring = kzalloc(sizeof(struct igbvf_ring), GFP_KERNEL);
1099         if (!adapter->rx_ring) {
1100                 kfree(adapter->tx_ring);
1101                 return -ENOMEM;
1102         }
1103
1104         netif_napi_add(netdev, &adapter->rx_ring->napi, igbvf_poll, 64);
1105
1106         return 0;
1107 }
1108
1109 /**
1110  * igbvf_request_irq - initialize interrupts
1111  *
1112  * Attempts to configure interrupts using the best available
1113  * capabilities of the hardware and kernel.
1114  **/
1115 static int igbvf_request_irq(struct igbvf_adapter *adapter)
1116 {
1117         int err = -1;
1118
1119         /* igbvf supports msi-x only */
1120         if (adapter->msix_entries)
1121                 err = igbvf_request_msix(adapter);
1122
1123         if (!err)
1124                 return err;
1125
1126         dev_err(&adapter->pdev->dev,
1127                 "Unable to allocate interrupt, Error: %d\n", err);
1128
1129         return err;
1130 }
1131
1132 static void igbvf_free_irq(struct igbvf_adapter *adapter)
1133 {
1134         struct net_device *netdev = adapter->netdev;
1135         int vector;
1136
1137         if (adapter->msix_entries) {
1138                 for (vector = 0; vector < 3; vector++)
1139                         free_irq(adapter->msix_entries[vector].vector, netdev);
1140         }
1141 }
1142
1143 /**
1144  * igbvf_irq_disable - Mask off interrupt generation on the NIC
1145  **/
1146 static void igbvf_irq_disable(struct igbvf_adapter *adapter)
1147 {
1148         struct e1000_hw *hw = &adapter->hw;
1149
1150         ew32(EIMC, ~0);
1151
1152         if (adapter->msix_entries)
1153                 ew32(EIAC, 0);
1154 }
1155
1156 /**
1157  * igbvf_irq_enable - Enable default interrupt generation settings
1158  **/
1159 static void igbvf_irq_enable(struct igbvf_adapter *adapter)
1160 {
1161         struct e1000_hw *hw = &adapter->hw;
1162
1163         ew32(EIAC, adapter->eims_enable_mask);
1164         ew32(EIAM, adapter->eims_enable_mask);
1165         ew32(EIMS, adapter->eims_enable_mask);
1166 }
1167
1168 /**
1169  * igbvf_poll - NAPI Rx polling callback
1170  * @napi: struct associated with this polling callback
1171  * @budget: amount of packets driver is allowed to process this poll
1172  **/
1173 static int igbvf_poll(struct napi_struct *napi, int budget)
1174 {
1175         struct igbvf_ring *rx_ring = container_of(napi, struct igbvf_ring, napi);
1176         struct igbvf_adapter *adapter = rx_ring->adapter;
1177         struct e1000_hw *hw = &adapter->hw;
1178         int work_done = 0;
1179
1180         igbvf_clean_rx_irq(adapter, &work_done, budget);
1181
1182         /* If not enough Rx work done, exit the polling mode */
1183         if (work_done < budget) {
1184                 napi_complete(napi);
1185
1186                 if (adapter->itr_setting & 3)
1187                         igbvf_set_itr(adapter);
1188
1189                 if (!test_bit(__IGBVF_DOWN, &adapter->state))
1190                         ew32(EIMS, adapter->rx_ring->eims_value);
1191         }
1192
1193         return work_done;
1194 }
1195
1196 /**
1197  * igbvf_set_rlpml - set receive large packet maximum length
1198  * @adapter: board private structure
1199  *
1200  * Configure the maximum size of packets that will be received
1201  */
1202 static void igbvf_set_rlpml(struct igbvf_adapter *adapter)
1203 {
1204         int max_frame_size = adapter->max_frame_size;
1205         struct e1000_hw *hw = &adapter->hw;
1206
1207         if (adapter->vlgrp)
1208                 max_frame_size += VLAN_TAG_SIZE;
1209
1210         e1000_rlpml_set_vf(hw, max_frame_size);
1211 }
1212
1213 static void igbvf_vlan_rx_add_vid(struct net_device *netdev, u16 vid)
1214 {
1215         struct igbvf_adapter *adapter = netdev_priv(netdev);
1216         struct e1000_hw *hw = &adapter->hw;
1217
1218         if (hw->mac.ops.set_vfta(hw, vid, true))
1219                 dev_err(&adapter->pdev->dev, "Failed to add vlan id %d\n", vid);
1220 }
1221
1222 static void igbvf_vlan_rx_kill_vid(struct net_device *netdev, u16 vid)
1223 {
1224         struct igbvf_adapter *adapter = netdev_priv(netdev);
1225         struct e1000_hw *hw = &adapter->hw;
1226
1227         igbvf_irq_disable(adapter);
1228         vlan_group_set_device(adapter->vlgrp, vid, NULL);
1229
1230         if (!test_bit(__IGBVF_DOWN, &adapter->state))
1231                 igbvf_irq_enable(adapter);
1232
1233         if (hw->mac.ops.set_vfta(hw, vid, false))
1234                 dev_err(&adapter->pdev->dev,
1235                         "Failed to remove vlan id %d\n", vid);
1236 }
1237
1238 static void igbvf_vlan_rx_register(struct net_device *netdev,
1239                                    struct vlan_group *grp)
1240 {
1241         struct igbvf_adapter *adapter = netdev_priv(netdev);
1242
1243         adapter->vlgrp = grp;
1244 }
1245
1246 static void igbvf_restore_vlan(struct igbvf_adapter *adapter)
1247 {
1248         u16 vid;
1249
1250         if (!adapter->vlgrp)
1251                 return;
1252
1253         for (vid = 0; vid < VLAN_GROUP_ARRAY_LEN; vid++) {
1254                 if (!vlan_group_get_device(adapter->vlgrp, vid))
1255                         continue;
1256                 igbvf_vlan_rx_add_vid(adapter->netdev, vid);
1257         }
1258
1259         igbvf_set_rlpml(adapter);
1260 }
1261
1262 /**
1263  * igbvf_configure_tx - Configure Transmit Unit after Reset
1264  * @adapter: board private structure
1265  *
1266  * Configure the Tx unit of the MAC after a reset.
1267  **/
1268 static void igbvf_configure_tx(struct igbvf_adapter *adapter)
1269 {
1270         struct e1000_hw *hw = &adapter->hw;
1271         struct igbvf_ring *tx_ring = adapter->tx_ring;
1272         u64 tdba;
1273         u32 txdctl, dca_txctrl;
1274
1275         /* disable transmits */
1276         txdctl = er32(TXDCTL(0));
1277         ew32(TXDCTL(0), txdctl & ~E1000_TXDCTL_QUEUE_ENABLE);
1278         msleep(10);
1279
1280         /* Setup the HW Tx Head and Tail descriptor pointers */
1281         ew32(TDLEN(0), tx_ring->count * sizeof(union e1000_adv_tx_desc));
1282         tdba = tx_ring->dma;
1283         ew32(TDBAL(0), (tdba & DMA_BIT_MASK(32)));
1284         ew32(TDBAH(0), (tdba >> 32));
1285         ew32(TDH(0), 0);
1286         ew32(TDT(0), 0);
1287         tx_ring->head = E1000_TDH(0);
1288         tx_ring->tail = E1000_TDT(0);
1289
1290         /* Turn off Relaxed Ordering on head write-backs.  The writebacks
1291          * MUST be delivered in order or it will completely screw up
1292          * our bookeeping.
1293          */
1294         dca_txctrl = er32(DCA_TXCTRL(0));
1295         dca_txctrl &= ~E1000_DCA_TXCTRL_TX_WB_RO_EN;
1296         ew32(DCA_TXCTRL(0), dca_txctrl);
1297
1298         /* enable transmits */
1299         txdctl |= E1000_TXDCTL_QUEUE_ENABLE;
1300         ew32(TXDCTL(0), txdctl);
1301
1302         /* Setup Transmit Descriptor Settings for eop descriptor */
1303         adapter->txd_cmd = E1000_ADVTXD_DCMD_EOP | E1000_ADVTXD_DCMD_IFCS;
1304
1305         /* enable Report Status bit */
1306         adapter->txd_cmd |= E1000_ADVTXD_DCMD_RS;
1307
1308         adapter->tx_queue_len = adapter->netdev->tx_queue_len;
1309 }
1310
1311 /**
1312  * igbvf_setup_srrctl - configure the receive control registers
1313  * @adapter: Board private structure
1314  **/
1315 static void igbvf_setup_srrctl(struct igbvf_adapter *adapter)
1316 {
1317         struct e1000_hw *hw = &adapter->hw;
1318         u32 srrctl = 0;
1319
1320         srrctl &= ~(E1000_SRRCTL_DESCTYPE_MASK |
1321                     E1000_SRRCTL_BSIZEHDR_MASK |
1322                     E1000_SRRCTL_BSIZEPKT_MASK);
1323
1324         /* Enable queue drop to avoid head of line blocking */
1325         srrctl |= E1000_SRRCTL_DROP_EN;
1326
1327         /* Setup buffer sizes */
1328         srrctl |= ALIGN(adapter->rx_buffer_len, 1024) >>
1329                   E1000_SRRCTL_BSIZEPKT_SHIFT;
1330
1331         if (adapter->rx_buffer_len < 2048) {
1332                 adapter->rx_ps_hdr_size = 0;
1333                 srrctl |= E1000_SRRCTL_DESCTYPE_ADV_ONEBUF;
1334         } else {
1335                 adapter->rx_ps_hdr_size = 128;
1336                 srrctl |= adapter->rx_ps_hdr_size <<
1337                           E1000_SRRCTL_BSIZEHDRSIZE_SHIFT;
1338                 srrctl |= E1000_SRRCTL_DESCTYPE_HDR_SPLIT_ALWAYS;
1339         }
1340
1341         ew32(SRRCTL(0), srrctl);
1342 }
1343
1344 /**
1345  * igbvf_configure_rx - Configure Receive Unit after Reset
1346  * @adapter: board private structure
1347  *
1348  * Configure the Rx unit of the MAC after a reset.
1349  **/
1350 static void igbvf_configure_rx(struct igbvf_adapter *adapter)
1351 {
1352         struct e1000_hw *hw = &adapter->hw;
1353         struct igbvf_ring *rx_ring = adapter->rx_ring;
1354         u64 rdba;
1355         u32 rdlen, rxdctl;
1356
1357         /* disable receives */
1358         rxdctl = er32(RXDCTL(0));
1359         ew32(RXDCTL(0), rxdctl & ~E1000_RXDCTL_QUEUE_ENABLE);
1360         msleep(10);
1361
1362         rdlen = rx_ring->count * sizeof(union e1000_adv_rx_desc);
1363
1364         /*
1365          * Setup the HW Rx Head and Tail Descriptor Pointers and
1366          * the Base and Length of the Rx Descriptor Ring
1367          */
1368         rdba = rx_ring->dma;
1369         ew32(RDBAL(0), (rdba & DMA_BIT_MASK(32)));
1370         ew32(RDBAH(0), (rdba >> 32));
1371         ew32(RDLEN(0), rx_ring->count * sizeof(union e1000_adv_rx_desc));
1372         rx_ring->head = E1000_RDH(0);
1373         rx_ring->tail = E1000_RDT(0);
1374         ew32(RDH(0), 0);
1375         ew32(RDT(0), 0);
1376
1377         rxdctl |= E1000_RXDCTL_QUEUE_ENABLE;
1378         rxdctl &= 0xFFF00000;
1379         rxdctl |= IGBVF_RX_PTHRESH;
1380         rxdctl |= IGBVF_RX_HTHRESH << 8;
1381         rxdctl |= IGBVF_RX_WTHRESH << 16;
1382
1383         igbvf_set_rlpml(adapter);
1384
1385         /* enable receives */
1386         ew32(RXDCTL(0), rxdctl);
1387 }
1388
1389 /**
1390  * igbvf_set_multi - Multicast and Promiscuous mode set
1391  * @netdev: network interface device structure
1392  *
1393  * The set_multi entry point is called whenever the multicast address
1394  * list or the network interface flags are updated.  This routine is
1395  * responsible for configuring the hardware for proper multicast,
1396  * promiscuous mode, and all-multi behavior.
1397  **/
1398 static void igbvf_set_multi(struct net_device *netdev)
1399 {
1400         struct igbvf_adapter *adapter = netdev_priv(netdev);
1401         struct e1000_hw *hw = &adapter->hw;
1402         struct dev_mc_list *mc_ptr;
1403         u8  *mta_list = NULL;
1404         int i;
1405
1406         if (netdev->mc_count) {
1407                 mta_list = kmalloc(netdev->mc_count * 6, GFP_ATOMIC);
1408                 if (!mta_list) {
1409                         dev_err(&adapter->pdev->dev,
1410                                 "failed to allocate multicast filter list\n");
1411                         return;
1412                 }
1413         }
1414
1415         /* prepare a packed array of only addresses. */
1416         mc_ptr = netdev->mc_list;
1417
1418         for (i = 0; i < netdev->mc_count; i++) {
1419                 if (!mc_ptr)
1420                         break;
1421                 memcpy(mta_list + (i*ETH_ALEN), mc_ptr->dmi_addr,
1422                        ETH_ALEN);
1423                 mc_ptr = mc_ptr->next;
1424         }
1425
1426         hw->mac.ops.update_mc_addr_list(hw, mta_list, i, 0, 0);
1427         kfree(mta_list);
1428 }
1429
1430 /**
1431  * igbvf_configure - configure the hardware for Rx and Tx
1432  * @adapter: private board structure
1433  **/
1434 static void igbvf_configure(struct igbvf_adapter *adapter)
1435 {
1436         igbvf_set_multi(adapter->netdev);
1437
1438         igbvf_restore_vlan(adapter);
1439
1440         igbvf_configure_tx(adapter);
1441         igbvf_setup_srrctl(adapter);
1442         igbvf_configure_rx(adapter);
1443         igbvf_alloc_rx_buffers(adapter->rx_ring,
1444                                igbvf_desc_unused(adapter->rx_ring));
1445 }
1446
1447 /* igbvf_reset - bring the hardware into a known good state
1448  *
1449  * This function boots the hardware and enables some settings that
1450  * require a configuration cycle of the hardware - those cannot be
1451  * set/changed during runtime. After reset the device needs to be
1452  * properly configured for Rx, Tx etc.
1453  */
1454 static void igbvf_reset(struct igbvf_adapter *adapter)
1455 {
1456         struct e1000_mac_info *mac = &adapter->hw.mac;
1457         struct net_device *netdev = adapter->netdev;
1458         struct e1000_hw *hw = &adapter->hw;
1459
1460         /* Allow time for pending master requests to run */
1461         if (mac->ops.reset_hw(hw))
1462                 dev_err(&adapter->pdev->dev, "PF still resetting\n");
1463
1464         mac->ops.init_hw(hw);
1465
1466         if (is_valid_ether_addr(adapter->hw.mac.addr)) {
1467                 memcpy(netdev->dev_addr, adapter->hw.mac.addr,
1468                        netdev->addr_len);
1469                 memcpy(netdev->perm_addr, adapter->hw.mac.addr,
1470                        netdev->addr_len);
1471         }
1472 }
1473
1474 int igbvf_up(struct igbvf_adapter *adapter)
1475 {
1476         struct e1000_hw *hw = &adapter->hw;
1477
1478         /* hardware has been reset, we need to reload some things */
1479         igbvf_configure(adapter);
1480
1481         clear_bit(__IGBVF_DOWN, &adapter->state);
1482
1483         napi_enable(&adapter->rx_ring->napi);
1484         if (adapter->msix_entries)
1485                 igbvf_configure_msix(adapter);
1486
1487         /* Clear any pending interrupts. */
1488         er32(EICR);
1489         igbvf_irq_enable(adapter);
1490
1491         /* start the watchdog */
1492         hw->mac.get_link_status = 1;
1493         mod_timer(&adapter->watchdog_timer, jiffies + 1);
1494
1495
1496         return 0;
1497 }
1498
1499 void igbvf_down(struct igbvf_adapter *adapter)
1500 {
1501         struct net_device *netdev = adapter->netdev;
1502         struct e1000_hw *hw = &adapter->hw;
1503         u32 rxdctl, txdctl;
1504
1505         /*
1506          * signal that we're down so the interrupt handler does not
1507          * reschedule our watchdog timer
1508          */
1509         set_bit(__IGBVF_DOWN, &adapter->state);
1510
1511         /* disable receives in the hardware */
1512         rxdctl = er32(RXDCTL(0));
1513         ew32(RXDCTL(0), rxdctl & ~E1000_RXDCTL_QUEUE_ENABLE);
1514
1515         netif_stop_queue(netdev);
1516
1517         /* disable transmits in the hardware */
1518         txdctl = er32(TXDCTL(0));
1519         ew32(TXDCTL(0), txdctl & ~E1000_TXDCTL_QUEUE_ENABLE);
1520
1521         /* flush both disables and wait for them to finish */
1522         e1e_flush();
1523         msleep(10);
1524
1525         napi_disable(&adapter->rx_ring->napi);
1526
1527         igbvf_irq_disable(adapter);
1528
1529         del_timer_sync(&adapter->watchdog_timer);
1530
1531         netdev->tx_queue_len = adapter->tx_queue_len;
1532         netif_carrier_off(netdev);
1533
1534         /* record the stats before reset*/
1535         igbvf_update_stats(adapter);
1536
1537         adapter->link_speed = 0;
1538         adapter->link_duplex = 0;
1539
1540         igbvf_reset(adapter);
1541         igbvf_clean_tx_ring(adapter->tx_ring);
1542         igbvf_clean_rx_ring(adapter->rx_ring);
1543 }
1544
1545 void igbvf_reinit_locked(struct igbvf_adapter *adapter)
1546 {
1547         might_sleep();
1548         while (test_and_set_bit(__IGBVF_RESETTING, &adapter->state))
1549                 msleep(1);
1550         igbvf_down(adapter);
1551         igbvf_up(adapter);
1552         clear_bit(__IGBVF_RESETTING, &adapter->state);
1553 }
1554
1555 /**
1556  * igbvf_sw_init - Initialize general software structures (struct igbvf_adapter)
1557  * @adapter: board private structure to initialize
1558  *
1559  * igbvf_sw_init initializes the Adapter private data structure.
1560  * Fields are initialized based on PCI device information and
1561  * OS network device settings (MTU size).
1562  **/
1563 static int __devinit igbvf_sw_init(struct igbvf_adapter *adapter)
1564 {
1565         struct net_device *netdev = adapter->netdev;
1566         s32 rc;
1567
1568         adapter->rx_buffer_len = ETH_FRAME_LEN + VLAN_HLEN + ETH_FCS_LEN;
1569         adapter->rx_ps_hdr_size = 0;
1570         adapter->max_frame_size = netdev->mtu + ETH_HLEN + ETH_FCS_LEN;
1571         adapter->min_frame_size = ETH_ZLEN + ETH_FCS_LEN;
1572
1573         adapter->tx_int_delay = 8;
1574         adapter->tx_abs_int_delay = 32;
1575         adapter->rx_int_delay = 0;
1576         adapter->rx_abs_int_delay = 8;
1577         adapter->itr_setting = 3;
1578         adapter->itr = 20000;
1579
1580         /* Set various function pointers */
1581         adapter->ei->init_ops(&adapter->hw);
1582
1583         rc = adapter->hw.mac.ops.init_params(&adapter->hw);
1584         if (rc)
1585                 return rc;
1586
1587         rc = adapter->hw.mbx.ops.init_params(&adapter->hw);
1588         if (rc)
1589                 return rc;
1590
1591         igbvf_set_interrupt_capability(adapter);
1592
1593         if (igbvf_alloc_queues(adapter))
1594                 return -ENOMEM;
1595
1596         spin_lock_init(&adapter->tx_queue_lock);
1597
1598         /* Explicitly disable IRQ since the NIC can be in any state. */
1599         igbvf_irq_disable(adapter);
1600
1601         spin_lock_init(&adapter->stats_lock);
1602
1603         set_bit(__IGBVF_DOWN, &adapter->state);
1604         return 0;
1605 }
1606
1607 static void igbvf_initialize_last_counter_stats(struct igbvf_adapter *adapter)
1608 {
1609         struct e1000_hw *hw = &adapter->hw;
1610
1611         adapter->stats.last_gprc = er32(VFGPRC);
1612         adapter->stats.last_gorc = er32(VFGORC);
1613         adapter->stats.last_gptc = er32(VFGPTC);
1614         adapter->stats.last_gotc = er32(VFGOTC);
1615         adapter->stats.last_mprc = er32(VFMPRC);
1616         adapter->stats.last_gotlbc = er32(VFGOTLBC);
1617         adapter->stats.last_gptlbc = er32(VFGPTLBC);
1618         adapter->stats.last_gorlbc = er32(VFGORLBC);
1619         adapter->stats.last_gprlbc = er32(VFGPRLBC);
1620
1621         adapter->stats.base_gprc = er32(VFGPRC);
1622         adapter->stats.base_gorc = er32(VFGORC);
1623         adapter->stats.base_gptc = er32(VFGPTC);
1624         adapter->stats.base_gotc = er32(VFGOTC);
1625         adapter->stats.base_mprc = er32(VFMPRC);
1626         adapter->stats.base_gotlbc = er32(VFGOTLBC);
1627         adapter->stats.base_gptlbc = er32(VFGPTLBC);
1628         adapter->stats.base_gorlbc = er32(VFGORLBC);
1629         adapter->stats.base_gprlbc = er32(VFGPRLBC);
1630 }
1631
1632 /**
1633  * igbvf_open - Called when a network interface is made active
1634  * @netdev: network interface device structure
1635  *
1636  * Returns 0 on success, negative value on failure
1637  *
1638  * The open entry point is called when a network interface is made
1639  * active by the system (IFF_UP).  At this point all resources needed
1640  * for transmit and receive operations are allocated, the interrupt
1641  * handler is registered with the OS, the watchdog timer is started,
1642  * and the stack is notified that the interface is ready.
1643  **/
1644 static int igbvf_open(struct net_device *netdev)
1645 {
1646         struct igbvf_adapter *adapter = netdev_priv(netdev);
1647         struct e1000_hw *hw = &adapter->hw;
1648         int err;
1649
1650         /* disallow open during test */
1651         if (test_bit(__IGBVF_TESTING, &adapter->state))
1652                 return -EBUSY;
1653
1654         /* allocate transmit descriptors */
1655         err = igbvf_setup_tx_resources(adapter, adapter->tx_ring);
1656         if (err)
1657                 goto err_setup_tx;
1658
1659         /* allocate receive descriptors */
1660         err = igbvf_setup_rx_resources(adapter, adapter->rx_ring);
1661         if (err)
1662                 goto err_setup_rx;
1663
1664         /*
1665          * before we allocate an interrupt, we must be ready to handle it.
1666          * Setting DEBUG_SHIRQ in the kernel makes it fire an interrupt
1667          * as soon as we call pci_request_irq, so we have to setup our
1668          * clean_rx handler before we do so.
1669          */
1670         igbvf_configure(adapter);
1671
1672         err = igbvf_request_irq(adapter);
1673         if (err)
1674                 goto err_req_irq;
1675
1676         /* From here on the code is the same as igbvf_up() */
1677         clear_bit(__IGBVF_DOWN, &adapter->state);
1678
1679         napi_enable(&adapter->rx_ring->napi);
1680
1681         /* clear any pending interrupts */
1682         er32(EICR);
1683
1684         igbvf_irq_enable(adapter);
1685
1686         /* start the watchdog */
1687         hw->mac.get_link_status = 1;
1688         mod_timer(&adapter->watchdog_timer, jiffies + 1);
1689
1690         return 0;
1691
1692 err_req_irq:
1693         igbvf_free_rx_resources(adapter->rx_ring);
1694 err_setup_rx:
1695         igbvf_free_tx_resources(adapter->tx_ring);
1696 err_setup_tx:
1697         igbvf_reset(adapter);
1698
1699         return err;
1700 }
1701
1702 /**
1703  * igbvf_close - Disables a network interface
1704  * @netdev: network interface device structure
1705  *
1706  * Returns 0, this is not allowed to fail
1707  *
1708  * The close entry point is called when an interface is de-activated
1709  * by the OS.  The hardware is still under the drivers control, but
1710  * needs to be disabled.  A global MAC reset is issued to stop the
1711  * hardware, and all transmit and receive resources are freed.
1712  **/
1713 static int igbvf_close(struct net_device *netdev)
1714 {
1715         struct igbvf_adapter *adapter = netdev_priv(netdev);
1716
1717         WARN_ON(test_bit(__IGBVF_RESETTING, &adapter->state));
1718         igbvf_down(adapter);
1719
1720         igbvf_free_irq(adapter);
1721
1722         igbvf_free_tx_resources(adapter->tx_ring);
1723         igbvf_free_rx_resources(adapter->rx_ring);
1724
1725         return 0;
1726 }
1727 /**
1728  * igbvf_set_mac - Change the Ethernet Address of the NIC
1729  * @netdev: network interface device structure
1730  * @p: pointer to an address structure
1731  *
1732  * Returns 0 on success, negative on failure
1733  **/
1734 static int igbvf_set_mac(struct net_device *netdev, void *p)
1735 {
1736         struct igbvf_adapter *adapter = netdev_priv(netdev);
1737         struct e1000_hw *hw = &adapter->hw;
1738         struct sockaddr *addr = p;
1739
1740         if (!is_valid_ether_addr(addr->sa_data))
1741                 return -EADDRNOTAVAIL;
1742
1743         memcpy(hw->mac.addr, addr->sa_data, netdev->addr_len);
1744
1745         hw->mac.ops.rar_set(hw, hw->mac.addr, 0);
1746
1747         if (memcmp(addr->sa_data, hw->mac.addr, 6))
1748                 return -EADDRNOTAVAIL;
1749
1750         memcpy(netdev->dev_addr, addr->sa_data, netdev->addr_len);
1751
1752         return 0;
1753 }
1754
1755 #define UPDATE_VF_COUNTER(reg, name)                                    \
1756         {                                                               \
1757                 u32 current_counter = er32(reg);                        \
1758                 if (current_counter < adapter->stats.last_##name)       \
1759                         adapter->stats.name += 0x100000000LL;           \
1760                 adapter->stats.last_##name = current_counter;           \
1761                 adapter->stats.name &= 0xFFFFFFFF00000000LL;            \
1762                 adapter->stats.name |= current_counter;                 \
1763         }
1764
1765 /**
1766  * igbvf_update_stats - Update the board statistics counters
1767  * @adapter: board private structure
1768 **/
1769 void igbvf_update_stats(struct igbvf_adapter *adapter)
1770 {
1771         struct e1000_hw *hw = &adapter->hw;
1772         struct pci_dev *pdev = adapter->pdev;
1773
1774         /*
1775          * Prevent stats update while adapter is being reset, link is down
1776          * or if the pci connection is down.
1777          */
1778         if (adapter->link_speed == 0)
1779                 return;
1780
1781         if (test_bit(__IGBVF_RESETTING, &adapter->state))
1782                 return;
1783
1784         if (pci_channel_offline(pdev))
1785                 return;
1786
1787         UPDATE_VF_COUNTER(VFGPRC, gprc);
1788         UPDATE_VF_COUNTER(VFGORC, gorc);
1789         UPDATE_VF_COUNTER(VFGPTC, gptc);
1790         UPDATE_VF_COUNTER(VFGOTC, gotc);
1791         UPDATE_VF_COUNTER(VFMPRC, mprc);
1792         UPDATE_VF_COUNTER(VFGOTLBC, gotlbc);
1793         UPDATE_VF_COUNTER(VFGPTLBC, gptlbc);
1794         UPDATE_VF_COUNTER(VFGORLBC, gorlbc);
1795         UPDATE_VF_COUNTER(VFGPRLBC, gprlbc);
1796
1797         /* Fill out the OS statistics structure */
1798         adapter->net_stats.multicast = adapter->stats.mprc;
1799 }
1800
1801 static void igbvf_print_link_info(struct igbvf_adapter *adapter)
1802 {
1803         dev_info(&adapter->pdev->dev, "Link is Up %d Mbps %s\n",
1804                  adapter->link_speed,
1805                  ((adapter->link_duplex == FULL_DUPLEX) ?
1806                   "Full Duplex" : "Half Duplex"));
1807 }
1808
1809 static bool igbvf_has_link(struct igbvf_adapter *adapter)
1810 {
1811         struct e1000_hw *hw = &adapter->hw;
1812         s32 ret_val = E1000_SUCCESS;
1813         bool link_active;
1814
1815         ret_val = hw->mac.ops.check_for_link(hw);
1816         link_active = !hw->mac.get_link_status;
1817
1818         /* if check for link returns error we will need to reset */
1819         if (ret_val)
1820                 schedule_work(&adapter->reset_task);
1821
1822         return link_active;
1823 }
1824
1825 /**
1826  * igbvf_watchdog - Timer Call-back
1827  * @data: pointer to adapter cast into an unsigned long
1828  **/
1829 static void igbvf_watchdog(unsigned long data)
1830 {
1831         struct igbvf_adapter *adapter = (struct igbvf_adapter *) data;
1832
1833         /* Do the rest outside of interrupt context */
1834         schedule_work(&adapter->watchdog_task);
1835 }
1836
1837 static void igbvf_watchdog_task(struct work_struct *work)
1838 {
1839         struct igbvf_adapter *adapter = container_of(work,
1840                                                      struct igbvf_adapter,
1841                                                      watchdog_task);
1842         struct net_device *netdev = adapter->netdev;
1843         struct e1000_mac_info *mac = &adapter->hw.mac;
1844         struct igbvf_ring *tx_ring = adapter->tx_ring;
1845         struct e1000_hw *hw = &adapter->hw;
1846         u32 link;
1847         int tx_pending = 0;
1848
1849         link = igbvf_has_link(adapter);
1850
1851         if (link) {
1852                 if (!netif_carrier_ok(netdev)) {
1853                         bool txb2b = 1;
1854
1855                         mac->ops.get_link_up_info(&adapter->hw,
1856                                                   &adapter->link_speed,
1857                                                   &adapter->link_duplex);
1858                         igbvf_print_link_info(adapter);
1859
1860                         /*
1861                          * tweak tx_queue_len according to speed/duplex
1862                          * and adjust the timeout factor
1863                          */
1864                         netdev->tx_queue_len = adapter->tx_queue_len;
1865                         adapter->tx_timeout_factor = 1;
1866                         switch (adapter->link_speed) {
1867                         case SPEED_10:
1868                                 txb2b = 0;
1869                                 netdev->tx_queue_len = 10;
1870                                 adapter->tx_timeout_factor = 16;
1871                                 break;
1872                         case SPEED_100:
1873                                 txb2b = 0;
1874                                 netdev->tx_queue_len = 100;
1875                                 /* maybe add some timeout factor ? */
1876                                 break;
1877                         }
1878
1879                         netif_carrier_on(netdev);
1880                         netif_wake_queue(netdev);
1881                 }
1882         } else {
1883                 if (netif_carrier_ok(netdev)) {
1884                         adapter->link_speed = 0;
1885                         adapter->link_duplex = 0;
1886                         dev_info(&adapter->pdev->dev, "Link is Down\n");
1887                         netif_carrier_off(netdev);
1888                         netif_stop_queue(netdev);
1889                 }
1890         }
1891
1892         if (netif_carrier_ok(netdev)) {
1893                 igbvf_update_stats(adapter);
1894         } else {
1895                 tx_pending = (igbvf_desc_unused(tx_ring) + 1 <
1896                               tx_ring->count);
1897                 if (tx_pending) {
1898                         /*
1899                          * We've lost link, so the controller stops DMA,
1900                          * but we've got queued Tx work that's never going
1901                          * to get done, so reset controller to flush Tx.
1902                          * (Do the reset outside of interrupt context).
1903                          */
1904                         adapter->tx_timeout_count++;
1905                         schedule_work(&adapter->reset_task);
1906                 }
1907         }
1908
1909         /* Cause software interrupt to ensure Rx ring is cleaned */
1910         ew32(EICS, adapter->rx_ring->eims_value);
1911
1912         /* Force detection of hung controller every watchdog period */
1913         adapter->detect_tx_hung = 1;
1914
1915         /* Reset the timer */
1916         if (!test_bit(__IGBVF_DOWN, &adapter->state))
1917                 mod_timer(&adapter->watchdog_timer,
1918                           round_jiffies(jiffies + (2 * HZ)));
1919 }
1920
1921 #define IGBVF_TX_FLAGS_CSUM             0x00000001
1922 #define IGBVF_TX_FLAGS_VLAN             0x00000002
1923 #define IGBVF_TX_FLAGS_TSO              0x00000004
1924 #define IGBVF_TX_FLAGS_IPV4             0x00000008
1925 #define IGBVF_TX_FLAGS_VLAN_MASK        0xffff0000
1926 #define IGBVF_TX_FLAGS_VLAN_SHIFT       16
1927
1928 static int igbvf_tso(struct igbvf_adapter *adapter,
1929                      struct igbvf_ring *tx_ring,
1930                      struct sk_buff *skb, u32 tx_flags, u8 *hdr_len)
1931 {
1932         struct e1000_adv_tx_context_desc *context_desc;
1933         unsigned int i;
1934         int err;
1935         struct igbvf_buffer *buffer_info;
1936         u32 info = 0, tu_cmd = 0;
1937         u32 mss_l4len_idx, l4len;
1938         *hdr_len = 0;
1939
1940         if (skb_header_cloned(skb)) {
1941                 err = pskb_expand_head(skb, 0, 0, GFP_ATOMIC);
1942                 if (err) {
1943                         dev_err(&adapter->pdev->dev,
1944                                 "igbvf_tso returning an error\n");
1945                         return err;
1946                 }
1947         }
1948
1949         l4len = tcp_hdrlen(skb);
1950         *hdr_len += l4len;
1951
1952         if (skb->protocol == htons(ETH_P_IP)) {
1953                 struct iphdr *iph = ip_hdr(skb);
1954                 iph->tot_len = 0;
1955                 iph->check = 0;
1956                 tcp_hdr(skb)->check = ~csum_tcpudp_magic(iph->saddr,
1957                                                          iph->daddr, 0,
1958                                                          IPPROTO_TCP,
1959                                                          0);
1960         } else if (skb_shinfo(skb)->gso_type == SKB_GSO_TCPV6) {
1961                 ipv6_hdr(skb)->payload_len = 0;
1962                 tcp_hdr(skb)->check = ~csum_ipv6_magic(&ipv6_hdr(skb)->saddr,
1963                                                        &ipv6_hdr(skb)->daddr,
1964                                                        0, IPPROTO_TCP, 0);
1965         }
1966
1967         i = tx_ring->next_to_use;
1968
1969         buffer_info = &tx_ring->buffer_info[i];
1970         context_desc = IGBVF_TX_CTXTDESC_ADV(*tx_ring, i);
1971         /* VLAN MACLEN IPLEN */
1972         if (tx_flags & IGBVF_TX_FLAGS_VLAN)
1973                 info |= (tx_flags & IGBVF_TX_FLAGS_VLAN_MASK);
1974         info |= (skb_network_offset(skb) << E1000_ADVTXD_MACLEN_SHIFT);
1975         *hdr_len += skb_network_offset(skb);
1976         info |= (skb_transport_header(skb) - skb_network_header(skb));
1977         *hdr_len += (skb_transport_header(skb) - skb_network_header(skb));
1978         context_desc->vlan_macip_lens = cpu_to_le32(info);
1979
1980         /* ADV DTYP TUCMD MKRLOC/ISCSIHEDLEN */
1981         tu_cmd |= (E1000_TXD_CMD_DEXT | E1000_ADVTXD_DTYP_CTXT);
1982
1983         if (skb->protocol == htons(ETH_P_IP))
1984                 tu_cmd |= E1000_ADVTXD_TUCMD_IPV4;
1985         tu_cmd |= E1000_ADVTXD_TUCMD_L4T_TCP;
1986
1987         context_desc->type_tucmd_mlhl = cpu_to_le32(tu_cmd);
1988
1989         /* MSS L4LEN IDX */
1990         mss_l4len_idx = (skb_shinfo(skb)->gso_size << E1000_ADVTXD_MSS_SHIFT);
1991         mss_l4len_idx |= (l4len << E1000_ADVTXD_L4LEN_SHIFT);
1992
1993         context_desc->mss_l4len_idx = cpu_to_le32(mss_l4len_idx);
1994         context_desc->seqnum_seed = 0;
1995
1996         buffer_info->time_stamp = jiffies;
1997         buffer_info->next_to_watch = i;
1998         buffer_info->dma = 0;
1999         i++;
2000         if (i == tx_ring->count)
2001                 i = 0;
2002
2003         tx_ring->next_to_use = i;
2004
2005         return true;
2006 }
2007
2008 static inline bool igbvf_tx_csum(struct igbvf_adapter *adapter,
2009                                  struct igbvf_ring *tx_ring,
2010                                  struct sk_buff *skb, u32 tx_flags)
2011 {
2012         struct e1000_adv_tx_context_desc *context_desc;
2013         unsigned int i;
2014         struct igbvf_buffer *buffer_info;
2015         u32 info = 0, tu_cmd = 0;
2016
2017         if ((skb->ip_summed == CHECKSUM_PARTIAL) ||
2018             (tx_flags & IGBVF_TX_FLAGS_VLAN)) {
2019                 i = tx_ring->next_to_use;
2020                 buffer_info = &tx_ring->buffer_info[i];
2021                 context_desc = IGBVF_TX_CTXTDESC_ADV(*tx_ring, i);
2022
2023                 if (tx_flags & IGBVF_TX_FLAGS_VLAN)
2024                         info |= (tx_flags & IGBVF_TX_FLAGS_VLAN_MASK);
2025
2026                 info |= (skb_network_offset(skb) << E1000_ADVTXD_MACLEN_SHIFT);
2027                 if (skb->ip_summed == CHECKSUM_PARTIAL)
2028                         info |= (skb_transport_header(skb) -
2029                                  skb_network_header(skb));
2030
2031
2032                 context_desc->vlan_macip_lens = cpu_to_le32(info);
2033
2034                 tu_cmd |= (E1000_TXD_CMD_DEXT | E1000_ADVTXD_DTYP_CTXT);
2035
2036                 if (skb->ip_summed == CHECKSUM_PARTIAL) {
2037                         switch (skb->protocol) {
2038                         case __constant_htons(ETH_P_IP):
2039                                 tu_cmd |= E1000_ADVTXD_TUCMD_IPV4;
2040                                 if (ip_hdr(skb)->protocol == IPPROTO_TCP)
2041                                         tu_cmd |= E1000_ADVTXD_TUCMD_L4T_TCP;
2042                                 break;
2043                         case __constant_htons(ETH_P_IPV6):
2044                                 if (ipv6_hdr(skb)->nexthdr == IPPROTO_TCP)
2045                                         tu_cmd |= E1000_ADVTXD_TUCMD_L4T_TCP;
2046                                 break;
2047                         default:
2048                                 break;
2049                         }
2050                 }
2051
2052                 context_desc->type_tucmd_mlhl = cpu_to_le32(tu_cmd);
2053                 context_desc->seqnum_seed = 0;
2054                 context_desc->mss_l4len_idx = 0;
2055
2056                 buffer_info->time_stamp = jiffies;
2057                 buffer_info->next_to_watch = i;
2058                 buffer_info->dma = 0;
2059                 i++;
2060                 if (i == tx_ring->count)
2061                         i = 0;
2062                 tx_ring->next_to_use = i;
2063
2064                 return true;
2065         }
2066
2067         return false;
2068 }
2069
2070 static int igbvf_maybe_stop_tx(struct net_device *netdev, int size)
2071 {
2072         struct igbvf_adapter *adapter = netdev_priv(netdev);
2073
2074         /* there is enough descriptors then we don't need to worry  */
2075         if (igbvf_desc_unused(adapter->tx_ring) >= size)
2076                 return 0;
2077
2078         netif_stop_queue(netdev);
2079
2080         smp_mb();
2081
2082         /* We need to check again just in case room has been made available */
2083         if (igbvf_desc_unused(adapter->tx_ring) < size)
2084                 return -EBUSY;
2085
2086         netif_wake_queue(netdev);
2087
2088         ++adapter->restart_queue;
2089         return 0;
2090 }
2091
2092 #define IGBVF_MAX_TXD_PWR       16
2093 #define IGBVF_MAX_DATA_PER_TXD  (1 << IGBVF_MAX_TXD_PWR)
2094
2095 static inline int igbvf_tx_map_adv(struct igbvf_adapter *adapter,
2096                                    struct igbvf_ring *tx_ring,
2097                                    struct sk_buff *skb,
2098                                    unsigned int first)
2099 {
2100         struct igbvf_buffer *buffer_info;
2101         unsigned int len = skb_headlen(skb);
2102         unsigned int count = 0, i;
2103         unsigned int f;
2104         dma_addr_t *map;
2105
2106         i = tx_ring->next_to_use;
2107
2108         if (skb_dma_map(&adapter->pdev->dev, skb, DMA_TO_DEVICE)) {
2109                 dev_err(&adapter->pdev->dev, "TX DMA map failed\n");
2110                 return 0;
2111         }
2112
2113         map = skb_shinfo(skb)->dma_maps;
2114
2115         buffer_info = &tx_ring->buffer_info[i];
2116         BUG_ON(len >= IGBVF_MAX_DATA_PER_TXD);
2117         buffer_info->length = len;
2118         /* set time_stamp *before* dma to help avoid a possible race */
2119         buffer_info->time_stamp = jiffies;
2120         buffer_info->next_to_watch = i;
2121         buffer_info->dma = skb_shinfo(skb)->dma_head;
2122
2123         for (f = 0; f < skb_shinfo(skb)->nr_frags; f++) {
2124                 struct skb_frag_struct *frag;
2125
2126                 i++;
2127                 if (i == tx_ring->count)
2128                         i = 0;
2129
2130                 frag = &skb_shinfo(skb)->frags[f];
2131                 len = frag->size;
2132
2133                 buffer_info = &tx_ring->buffer_info[i];
2134                 BUG_ON(len >= IGBVF_MAX_DATA_PER_TXD);
2135                 buffer_info->length = len;
2136                 buffer_info->time_stamp = jiffies;
2137                 buffer_info->next_to_watch = i;
2138                 buffer_info->dma = map[count];
2139                 count++;
2140         }
2141
2142         tx_ring->buffer_info[i].skb = skb;
2143         tx_ring->buffer_info[first].next_to_watch = i;
2144
2145         return count + 1;
2146 }
2147
2148 static inline void igbvf_tx_queue_adv(struct igbvf_adapter *adapter,
2149                                       struct igbvf_ring *tx_ring,
2150                                       int tx_flags, int count, u32 paylen,
2151                                       u8 hdr_len)
2152 {
2153         union e1000_adv_tx_desc *tx_desc = NULL;
2154         struct igbvf_buffer *buffer_info;
2155         u32 olinfo_status = 0, cmd_type_len;
2156         unsigned int i;
2157
2158         cmd_type_len = (E1000_ADVTXD_DTYP_DATA | E1000_ADVTXD_DCMD_IFCS |
2159                         E1000_ADVTXD_DCMD_DEXT);
2160
2161         if (tx_flags & IGBVF_TX_FLAGS_VLAN)
2162                 cmd_type_len |= E1000_ADVTXD_DCMD_VLE;
2163
2164         if (tx_flags & IGBVF_TX_FLAGS_TSO) {
2165                 cmd_type_len |= E1000_ADVTXD_DCMD_TSE;
2166
2167                 /* insert tcp checksum */
2168                 olinfo_status |= E1000_TXD_POPTS_TXSM << 8;
2169
2170                 /* insert ip checksum */
2171                 if (tx_flags & IGBVF_TX_FLAGS_IPV4)
2172                         olinfo_status |= E1000_TXD_POPTS_IXSM << 8;
2173
2174         } else if (tx_flags & IGBVF_TX_FLAGS_CSUM) {
2175                 olinfo_status |= E1000_TXD_POPTS_TXSM << 8;
2176         }
2177
2178         olinfo_status |= ((paylen - hdr_len) << E1000_ADVTXD_PAYLEN_SHIFT);
2179
2180         i = tx_ring->next_to_use;
2181         while (count--) {
2182                 buffer_info = &tx_ring->buffer_info[i];
2183                 tx_desc = IGBVF_TX_DESC_ADV(*tx_ring, i);
2184                 tx_desc->read.buffer_addr = cpu_to_le64(buffer_info->dma);
2185                 tx_desc->read.cmd_type_len =
2186                          cpu_to_le32(cmd_type_len | buffer_info->length);
2187                 tx_desc->read.olinfo_status = cpu_to_le32(olinfo_status);
2188                 i++;
2189                 if (i == tx_ring->count)
2190                         i = 0;
2191         }
2192
2193         tx_desc->read.cmd_type_len |= cpu_to_le32(adapter->txd_cmd);
2194         /* Force memory writes to complete before letting h/w
2195          * know there are new descriptors to fetch.  (Only
2196          * applicable for weak-ordered memory model archs,
2197          * such as IA-64). */
2198         wmb();
2199
2200         tx_ring->next_to_use = i;
2201         writel(i, adapter->hw.hw_addr + tx_ring->tail);
2202         /* we need this if more than one processor can write to our tail
2203          * at a time, it syncronizes IO on IA64/Altix systems */
2204         mmiowb();
2205 }
2206
2207 static int igbvf_xmit_frame_ring_adv(struct sk_buff *skb,
2208                                    struct net_device *netdev,
2209                                    struct igbvf_ring *tx_ring)
2210 {
2211         struct igbvf_adapter *adapter = netdev_priv(netdev);
2212         unsigned int first, tx_flags = 0;
2213         u8 hdr_len = 0;
2214         int count = 0;
2215         int tso = 0;
2216
2217         if (test_bit(__IGBVF_DOWN, &adapter->state)) {
2218                 dev_kfree_skb_any(skb);
2219                 return NETDEV_TX_OK;
2220         }
2221
2222         if (skb->len <= 0) {
2223                 dev_kfree_skb_any(skb);
2224                 return NETDEV_TX_OK;
2225         }
2226
2227         /*
2228          * need: count + 4 desc gap to keep tail from touching
2229          *       + 2 desc gap to keep tail from touching head,
2230          *       + 1 desc for skb->data,
2231          *       + 1 desc for context descriptor,
2232          * head, otherwise try next time
2233          */
2234         if (igbvf_maybe_stop_tx(netdev, skb_shinfo(skb)->nr_frags + 4)) {
2235                 /* this is a hard error */
2236                 return NETDEV_TX_BUSY;
2237         }
2238
2239         if (adapter->vlgrp && vlan_tx_tag_present(skb)) {
2240                 tx_flags |= IGBVF_TX_FLAGS_VLAN;
2241                 tx_flags |= (vlan_tx_tag_get(skb) << IGBVF_TX_FLAGS_VLAN_SHIFT);
2242         }
2243
2244         if (skb->protocol == htons(ETH_P_IP))
2245                 tx_flags |= IGBVF_TX_FLAGS_IPV4;
2246
2247         first = tx_ring->next_to_use;
2248
2249         tso = skb_is_gso(skb) ?
2250                 igbvf_tso(adapter, tx_ring, skb, tx_flags, &hdr_len) : 0;
2251         if (unlikely(tso < 0)) {
2252                 dev_kfree_skb_any(skb);
2253                 return NETDEV_TX_OK;
2254         }
2255
2256         if (tso)
2257                 tx_flags |= IGBVF_TX_FLAGS_TSO;
2258         else if (igbvf_tx_csum(adapter, tx_ring, skb, tx_flags) &&
2259                  (skb->ip_summed == CHECKSUM_PARTIAL))
2260                 tx_flags |= IGBVF_TX_FLAGS_CSUM;
2261
2262         /*
2263          * count reflects descriptors mapped, if 0 then mapping error
2264          * has occured and we need to rewind the descriptor queue
2265          */
2266         count = igbvf_tx_map_adv(adapter, tx_ring, skb, first);
2267
2268         if (count) {
2269                 igbvf_tx_queue_adv(adapter, tx_ring, tx_flags, count,
2270                                    skb->len, hdr_len);
2271                 /* Make sure there is space in the ring for the next send. */
2272                 igbvf_maybe_stop_tx(netdev, MAX_SKB_FRAGS + 4);
2273         } else {
2274                 dev_kfree_skb_any(skb);
2275                 tx_ring->buffer_info[first].time_stamp = 0;
2276                 tx_ring->next_to_use = first;
2277         }
2278
2279         return NETDEV_TX_OK;
2280 }
2281
2282 static int igbvf_xmit_frame(struct sk_buff *skb, struct net_device *netdev)
2283 {
2284         struct igbvf_adapter *adapter = netdev_priv(netdev);
2285         struct igbvf_ring *tx_ring;
2286         int retval;
2287
2288         if (test_bit(__IGBVF_DOWN, &adapter->state)) {
2289                 dev_kfree_skb_any(skb);
2290                 return NETDEV_TX_OK;
2291         }
2292
2293         tx_ring = &adapter->tx_ring[0];
2294
2295         retval = igbvf_xmit_frame_ring_adv(skb, netdev, tx_ring);
2296
2297         return retval;
2298 }
2299
2300 /**
2301  * igbvf_tx_timeout - Respond to a Tx Hang
2302  * @netdev: network interface device structure
2303  **/
2304 static void igbvf_tx_timeout(struct net_device *netdev)
2305 {
2306         struct igbvf_adapter *adapter = netdev_priv(netdev);
2307
2308         /* Do the reset outside of interrupt context */
2309         adapter->tx_timeout_count++;
2310         schedule_work(&adapter->reset_task);
2311 }
2312
2313 static void igbvf_reset_task(struct work_struct *work)
2314 {
2315         struct igbvf_adapter *adapter;
2316         adapter = container_of(work, struct igbvf_adapter, reset_task);
2317
2318         igbvf_reinit_locked(adapter);
2319 }
2320
2321 /**
2322  * igbvf_get_stats - Get System Network Statistics
2323  * @netdev: network interface device structure
2324  *
2325  * Returns the address of the device statistics structure.
2326  * The statistics are actually updated from the timer callback.
2327  **/
2328 static struct net_device_stats *igbvf_get_stats(struct net_device *netdev)
2329 {
2330         struct igbvf_adapter *adapter = netdev_priv(netdev);
2331
2332         /* only return the current stats */
2333         return &adapter->net_stats;
2334 }
2335
2336 /**
2337  * igbvf_change_mtu - Change the Maximum Transfer Unit
2338  * @netdev: network interface device structure
2339  * @new_mtu: new value for maximum frame size
2340  *
2341  * Returns 0 on success, negative on failure
2342  **/
2343 static int igbvf_change_mtu(struct net_device *netdev, int new_mtu)
2344 {
2345         struct igbvf_adapter *adapter = netdev_priv(netdev);
2346         int max_frame = new_mtu + ETH_HLEN + ETH_FCS_LEN;
2347
2348         if ((new_mtu < 68) || (max_frame > MAX_JUMBO_FRAME_SIZE)) {
2349                 dev_err(&adapter->pdev->dev, "Invalid MTU setting\n");
2350                 return -EINVAL;
2351         }
2352
2353 #define MAX_STD_JUMBO_FRAME_SIZE 9234
2354         if (max_frame > MAX_STD_JUMBO_FRAME_SIZE) {
2355                 dev_err(&adapter->pdev->dev, "MTU > 9216 not supported.\n");
2356                 return -EINVAL;
2357         }
2358
2359         while (test_and_set_bit(__IGBVF_RESETTING, &adapter->state))
2360                 msleep(1);
2361         /* igbvf_down has a dependency on max_frame_size */
2362         adapter->max_frame_size = max_frame;
2363         if (netif_running(netdev))
2364                 igbvf_down(adapter);
2365
2366         /*
2367          * NOTE: netdev_alloc_skb reserves 16 bytes, and typically NET_IP_ALIGN
2368          * means we reserve 2 more, this pushes us to allocate from the next
2369          * larger slab size.
2370          * i.e. RXBUFFER_2048 --> size-4096 slab
2371          * However with the new *_jumbo_rx* routines, jumbo receives will use
2372          * fragmented skbs
2373          */
2374
2375         if (max_frame <= 1024)
2376                 adapter->rx_buffer_len = 1024;
2377         else if (max_frame <= 2048)
2378                 adapter->rx_buffer_len = 2048;
2379         else
2380 #if (PAGE_SIZE / 2) > 16384
2381                 adapter->rx_buffer_len = 16384;
2382 #else
2383                 adapter->rx_buffer_len = PAGE_SIZE / 2;
2384 #endif
2385
2386
2387         /* adjust allocation if LPE protects us, and we aren't using SBP */
2388         if ((max_frame == ETH_FRAME_LEN + ETH_FCS_LEN) ||
2389              (max_frame == ETH_FRAME_LEN + VLAN_HLEN + ETH_FCS_LEN))
2390                 adapter->rx_buffer_len = ETH_FRAME_LEN + VLAN_HLEN +
2391                                          ETH_FCS_LEN;
2392
2393         dev_info(&adapter->pdev->dev, "changing MTU from %d to %d\n",
2394                  netdev->mtu, new_mtu);
2395         netdev->mtu = new_mtu;
2396
2397         if (netif_running(netdev))
2398                 igbvf_up(adapter);
2399         else
2400                 igbvf_reset(adapter);
2401
2402         clear_bit(__IGBVF_RESETTING, &adapter->state);
2403
2404         return 0;
2405 }
2406
2407 static int igbvf_ioctl(struct net_device *netdev, struct ifreq *ifr, int cmd)
2408 {
2409         switch (cmd) {
2410         default:
2411                 return -EOPNOTSUPP;
2412         }
2413 }
2414
2415 static int igbvf_suspend(struct pci_dev *pdev, pm_message_t state)
2416 {
2417         struct net_device *netdev = pci_get_drvdata(pdev);
2418         struct igbvf_adapter *adapter = netdev_priv(netdev);
2419 #ifdef CONFIG_PM
2420         int retval = 0;
2421 #endif
2422
2423         netif_device_detach(netdev);
2424
2425         if (netif_running(netdev)) {
2426                 WARN_ON(test_bit(__IGBVF_RESETTING, &adapter->state));
2427                 igbvf_down(adapter);
2428                 igbvf_free_irq(adapter);
2429         }
2430
2431 #ifdef CONFIG_PM
2432         retval = pci_save_state(pdev);
2433         if (retval)
2434                 return retval;
2435 #endif
2436
2437         pci_disable_device(pdev);
2438
2439         return 0;
2440 }
2441
2442 #ifdef CONFIG_PM
2443 static int igbvf_resume(struct pci_dev *pdev)
2444 {
2445         struct net_device *netdev = pci_get_drvdata(pdev);
2446         struct igbvf_adapter *adapter = netdev_priv(netdev);
2447         u32 err;
2448
2449         pci_restore_state(pdev);
2450         err = pci_enable_device_mem(pdev);
2451         if (err) {
2452                 dev_err(&pdev->dev, "Cannot enable PCI device from suspend\n");
2453                 return err;
2454         }
2455
2456         pci_set_master(pdev);
2457
2458         if (netif_running(netdev)) {
2459                 err = igbvf_request_irq(adapter);
2460                 if (err)
2461                         return err;
2462         }
2463
2464         igbvf_reset(adapter);
2465
2466         if (netif_running(netdev))
2467                 igbvf_up(adapter);
2468
2469         netif_device_attach(netdev);
2470
2471         return 0;
2472 }
2473 #endif
2474
2475 static void igbvf_shutdown(struct pci_dev *pdev)
2476 {
2477         igbvf_suspend(pdev, PMSG_SUSPEND);
2478 }
2479
2480 #ifdef CONFIG_NET_POLL_CONTROLLER
2481 /*
2482  * Polling 'interrupt' - used by things like netconsole to send skbs
2483  * without having to re-enable interrupts. It's not called while
2484  * the interrupt routine is executing.
2485  */
2486 static void igbvf_netpoll(struct net_device *netdev)
2487 {
2488         struct igbvf_adapter *adapter = netdev_priv(netdev);
2489
2490         disable_irq(adapter->pdev->irq);
2491
2492         igbvf_clean_tx_irq(adapter->tx_ring);
2493
2494         enable_irq(adapter->pdev->irq);
2495 }
2496 #endif
2497
2498 /**
2499  * igbvf_io_error_detected - called when PCI error is detected
2500  * @pdev: Pointer to PCI device
2501  * @state: The current pci connection state
2502  *
2503  * This function is called after a PCI bus error affecting
2504  * this device has been detected.
2505  */
2506 static pci_ers_result_t igbvf_io_error_detected(struct pci_dev *pdev,
2507                                                 pci_channel_state_t state)
2508 {
2509         struct net_device *netdev = pci_get_drvdata(pdev);
2510         struct igbvf_adapter *adapter = netdev_priv(netdev);
2511
2512         netif_device_detach(netdev);
2513
2514         if (state == pci_channel_io_perm_failure)
2515                 return PCI_ERS_RESULT_DISCONNECT;
2516
2517         if (netif_running(netdev))
2518                 igbvf_down(adapter);
2519         pci_disable_device(pdev);
2520
2521         /* Request a slot slot reset. */
2522         return PCI_ERS_RESULT_NEED_RESET;
2523 }
2524
2525 /**
2526  * igbvf_io_slot_reset - called after the pci bus has been reset.
2527  * @pdev: Pointer to PCI device
2528  *
2529  * Restart the card from scratch, as if from a cold-boot. Implementation
2530  * resembles the first-half of the igbvf_resume routine.
2531  */
2532 static pci_ers_result_t igbvf_io_slot_reset(struct pci_dev *pdev)
2533 {
2534         struct net_device *netdev = pci_get_drvdata(pdev);
2535         struct igbvf_adapter *adapter = netdev_priv(netdev);
2536
2537         if (pci_enable_device_mem(pdev)) {
2538                 dev_err(&pdev->dev,
2539                         "Cannot re-enable PCI device after reset.\n");
2540                 return PCI_ERS_RESULT_DISCONNECT;
2541         }
2542         pci_set_master(pdev);
2543
2544         igbvf_reset(adapter);
2545
2546         return PCI_ERS_RESULT_RECOVERED;
2547 }
2548
2549 /**
2550  * igbvf_io_resume - called when traffic can start flowing again.
2551  * @pdev: Pointer to PCI device
2552  *
2553  * This callback is called when the error recovery driver tells us that
2554  * its OK to resume normal operation. Implementation resembles the
2555  * second-half of the igbvf_resume routine.
2556  */
2557 static void igbvf_io_resume(struct pci_dev *pdev)
2558 {
2559         struct net_device *netdev = pci_get_drvdata(pdev);
2560         struct igbvf_adapter *adapter = netdev_priv(netdev);
2561
2562         if (netif_running(netdev)) {
2563                 if (igbvf_up(adapter)) {
2564                         dev_err(&pdev->dev,
2565                                 "can't bring device back up after reset\n");
2566                         return;
2567                 }
2568         }
2569
2570         netif_device_attach(netdev);
2571 }
2572
2573 static void igbvf_print_device_info(struct igbvf_adapter *adapter)
2574 {
2575         struct e1000_hw *hw = &adapter->hw;
2576         struct net_device *netdev = adapter->netdev;
2577         struct pci_dev *pdev = adapter->pdev;
2578
2579         dev_info(&pdev->dev, "Intel(R) 82576 Virtual Function\n");
2580         dev_info(&pdev->dev, "Address: %02x:%02x:%02x:%02x:%02x:%02x\n",
2581                  /* MAC address */
2582                  netdev->dev_addr[0], netdev->dev_addr[1],
2583                  netdev->dev_addr[2], netdev->dev_addr[3],
2584                  netdev->dev_addr[4], netdev->dev_addr[5]);
2585         dev_info(&pdev->dev, "MAC: %d\n", hw->mac.type);
2586 }
2587
2588 static const struct net_device_ops igbvf_netdev_ops = {
2589         .ndo_open                       = igbvf_open,
2590         .ndo_stop                       = igbvf_close,
2591         .ndo_start_xmit                 = igbvf_xmit_frame,
2592         .ndo_get_stats                  = igbvf_get_stats,
2593         .ndo_set_multicast_list         = igbvf_set_multi,
2594         .ndo_set_mac_address            = igbvf_set_mac,
2595         .ndo_change_mtu                 = igbvf_change_mtu,
2596         .ndo_do_ioctl                   = igbvf_ioctl,
2597         .ndo_tx_timeout                 = igbvf_tx_timeout,
2598         .ndo_vlan_rx_register           = igbvf_vlan_rx_register,
2599         .ndo_vlan_rx_add_vid            = igbvf_vlan_rx_add_vid,
2600         .ndo_vlan_rx_kill_vid           = igbvf_vlan_rx_kill_vid,
2601 #ifdef CONFIG_NET_POLL_CONTROLLER
2602         .ndo_poll_controller            = igbvf_netpoll,
2603 #endif
2604 };
2605
2606 /**
2607  * igbvf_probe - Device Initialization Routine
2608  * @pdev: PCI device information struct
2609  * @ent: entry in igbvf_pci_tbl
2610  *
2611  * Returns 0 on success, negative on failure
2612  *
2613  * igbvf_probe initializes an adapter identified by a pci_dev structure.
2614  * The OS initialization, configuring of the adapter private structure,
2615  * and a hardware reset occur.
2616  **/
2617 static int __devinit igbvf_probe(struct pci_dev *pdev,
2618                                  const struct pci_device_id *ent)
2619 {
2620         struct net_device *netdev;
2621         struct igbvf_adapter *adapter;
2622         struct e1000_hw *hw;
2623         const struct igbvf_info *ei = igbvf_info_tbl[ent->driver_data];
2624
2625         static int cards_found;
2626         int err, pci_using_dac;
2627
2628         err = pci_enable_device_mem(pdev);
2629         if (err)
2630                 return err;
2631
2632         pci_using_dac = 0;
2633         err = pci_set_dma_mask(pdev, DMA_BIT_MASK(64));
2634         if (!err) {
2635                 err = pci_set_consistent_dma_mask(pdev, DMA_BIT_MASK(64));
2636                 if (!err)
2637                         pci_using_dac = 1;
2638         } else {
2639                 err = pci_set_dma_mask(pdev, DMA_BIT_MASK(32));
2640                 if (err) {
2641                         err = pci_set_consistent_dma_mask(pdev,
2642                                                           DMA_BIT_MASK(32));
2643                         if (err) {
2644                                 dev_err(&pdev->dev, "No usable DMA "
2645                                         "configuration, aborting\n");
2646                                 goto err_dma;
2647                         }
2648                 }
2649         }
2650
2651         err = pci_request_regions(pdev, igbvf_driver_name);
2652         if (err)
2653                 goto err_pci_reg;
2654
2655         pci_set_master(pdev);
2656
2657         err = -ENOMEM;
2658         netdev = alloc_etherdev(sizeof(struct igbvf_adapter));
2659         if (!netdev)
2660                 goto err_alloc_etherdev;
2661
2662         SET_NETDEV_DEV(netdev, &pdev->dev);
2663
2664         pci_set_drvdata(pdev, netdev);
2665         adapter = netdev_priv(netdev);
2666         hw = &adapter->hw;
2667         adapter->netdev = netdev;
2668         adapter->pdev = pdev;
2669         adapter->ei = ei;
2670         adapter->pba = ei->pba;
2671         adapter->flags = ei->flags;
2672         adapter->hw.back = adapter;
2673         adapter->hw.mac.type = ei->mac;
2674         adapter->msg_enable = (1 << NETIF_MSG_DRV | NETIF_MSG_PROBE) - 1;
2675
2676         /* PCI config space info */
2677
2678         hw->vendor_id = pdev->vendor;
2679         hw->device_id = pdev->device;
2680         hw->subsystem_vendor_id = pdev->subsystem_vendor;
2681         hw->subsystem_device_id = pdev->subsystem_device;
2682
2683         pci_read_config_byte(pdev, PCI_REVISION_ID, &hw->revision_id);
2684
2685         err = -EIO;
2686         adapter->hw.hw_addr = ioremap(pci_resource_start(pdev, 0),
2687                                       pci_resource_len(pdev, 0));
2688
2689         if (!adapter->hw.hw_addr)
2690                 goto err_ioremap;
2691
2692         if (ei->get_variants) {
2693                 err = ei->get_variants(adapter);
2694                 if (err)
2695                         goto err_ioremap;
2696         }
2697
2698         /* setup adapter struct */
2699         err = igbvf_sw_init(adapter);
2700         if (err)
2701                 goto err_sw_init;
2702
2703         /* construct the net_device struct */
2704         netdev->netdev_ops = &igbvf_netdev_ops;
2705
2706         igbvf_set_ethtool_ops(netdev);
2707         netdev->watchdog_timeo = 5 * HZ;
2708         strncpy(netdev->name, pci_name(pdev), sizeof(netdev->name) - 1);
2709
2710         adapter->bd_number = cards_found++;
2711
2712         netdev->features = NETIF_F_SG |
2713                            NETIF_F_IP_CSUM |
2714                            NETIF_F_HW_VLAN_TX |
2715                            NETIF_F_HW_VLAN_RX |
2716                            NETIF_F_HW_VLAN_FILTER;
2717
2718         netdev->features |= NETIF_F_IPV6_CSUM;
2719         netdev->features |= NETIF_F_TSO;
2720         netdev->features |= NETIF_F_TSO6;
2721
2722         if (pci_using_dac)
2723                 netdev->features |= NETIF_F_HIGHDMA;
2724
2725         netdev->vlan_features |= NETIF_F_TSO;
2726         netdev->vlan_features |= NETIF_F_TSO6;
2727         netdev->vlan_features |= NETIF_F_IP_CSUM;
2728         netdev->vlan_features |= NETIF_F_IPV6_CSUM;
2729         netdev->vlan_features |= NETIF_F_SG;
2730
2731         /*reset the controller to put the device in a known good state */
2732         err = hw->mac.ops.reset_hw(hw);
2733         if (err) {
2734                 dev_info(&pdev->dev,
2735                          "PF still in reset state, assigning new address\n");
2736                 random_ether_addr(hw->mac.addr);
2737         } else {
2738                 err = hw->mac.ops.read_mac_addr(hw);
2739                 if (err) {
2740                         dev_err(&pdev->dev, "Error reading MAC address\n");
2741                         goto err_hw_init;
2742                 }
2743         }
2744
2745         memcpy(netdev->dev_addr, adapter->hw.mac.addr, netdev->addr_len);
2746         memcpy(netdev->perm_addr, adapter->hw.mac.addr, netdev->addr_len);
2747
2748         if (!is_valid_ether_addr(netdev->perm_addr)) {
2749                 dev_err(&pdev->dev, "Invalid MAC Address: "
2750                         "%02x:%02x:%02x:%02x:%02x:%02x\n",
2751                         netdev->dev_addr[0], netdev->dev_addr[1],
2752                         netdev->dev_addr[2], netdev->dev_addr[3],
2753                         netdev->dev_addr[4], netdev->dev_addr[5]);
2754                 err = -EIO;
2755                 goto err_hw_init;
2756         }
2757
2758         setup_timer(&adapter->watchdog_timer, &igbvf_watchdog,
2759                     (unsigned long) adapter);
2760
2761         INIT_WORK(&adapter->reset_task, igbvf_reset_task);
2762         INIT_WORK(&adapter->watchdog_task, igbvf_watchdog_task);
2763
2764         /* ring size defaults */
2765         adapter->rx_ring->count = 1024;
2766         adapter->tx_ring->count = 1024;
2767
2768         /* reset the hardware with the new settings */
2769         igbvf_reset(adapter);
2770
2771         /* tell the stack to leave us alone until igbvf_open() is called */
2772         netif_carrier_off(netdev);
2773         netif_stop_queue(netdev);
2774
2775         strcpy(netdev->name, "eth%d");
2776         err = register_netdev(netdev);
2777         if (err)
2778                 goto err_hw_init;
2779
2780         igbvf_print_device_info(adapter);
2781
2782         igbvf_initialize_last_counter_stats(adapter);
2783
2784         return 0;
2785
2786 err_hw_init:
2787         kfree(adapter->tx_ring);
2788         kfree(adapter->rx_ring);
2789 err_sw_init:
2790         igbvf_reset_interrupt_capability(adapter);
2791         iounmap(adapter->hw.hw_addr);
2792 err_ioremap:
2793         free_netdev(netdev);
2794 err_alloc_etherdev:
2795         pci_release_regions(pdev);
2796 err_pci_reg:
2797 err_dma:
2798         pci_disable_device(pdev);
2799         return err;
2800 }
2801
2802 /**
2803  * igbvf_remove - Device Removal Routine
2804  * @pdev: PCI device information struct
2805  *
2806  * igbvf_remove is called by the PCI subsystem to alert the driver
2807  * that it should release a PCI device.  The could be caused by a
2808  * Hot-Plug event, or because the driver is going to be removed from
2809  * memory.
2810  **/
2811 static void __devexit igbvf_remove(struct pci_dev *pdev)
2812 {
2813         struct net_device *netdev = pci_get_drvdata(pdev);
2814         struct igbvf_adapter *adapter = netdev_priv(netdev);
2815         struct e1000_hw *hw = &adapter->hw;
2816
2817         /*
2818          * flush_scheduled work may reschedule our watchdog task, so
2819          * explicitly disable watchdog tasks from being rescheduled
2820          */
2821         set_bit(__IGBVF_DOWN, &adapter->state);
2822         del_timer_sync(&adapter->watchdog_timer);
2823
2824         flush_scheduled_work();
2825
2826         unregister_netdev(netdev);
2827
2828         igbvf_reset_interrupt_capability(adapter);
2829
2830         /*
2831          * it is important to delete the napi struct prior to freeing the
2832          * rx ring so that you do not end up with null pointer refs
2833          */
2834         netif_napi_del(&adapter->rx_ring->napi);
2835         kfree(adapter->tx_ring);
2836         kfree(adapter->rx_ring);
2837
2838         iounmap(hw->hw_addr);
2839         if (hw->flash_address)
2840                 iounmap(hw->flash_address);
2841         pci_release_regions(pdev);
2842
2843         free_netdev(netdev);
2844
2845         pci_disable_device(pdev);
2846 }
2847
2848 /* PCI Error Recovery (ERS) */
2849 static struct pci_error_handlers igbvf_err_handler = {
2850         .error_detected = igbvf_io_error_detected,
2851         .slot_reset = igbvf_io_slot_reset,
2852         .resume = igbvf_io_resume,
2853 };
2854
2855 static struct pci_device_id igbvf_pci_tbl[] = {
2856         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82576_VF), board_vf },
2857         { } /* terminate list */
2858 };
2859 MODULE_DEVICE_TABLE(pci, igbvf_pci_tbl);
2860
2861 /* PCI Device API Driver */
2862 static struct pci_driver igbvf_driver = {
2863         .name     = igbvf_driver_name,
2864         .id_table = igbvf_pci_tbl,
2865         .probe    = igbvf_probe,
2866         .remove   = __devexit_p(igbvf_remove),
2867 #ifdef CONFIG_PM
2868         /* Power Management Hooks */
2869         .suspend  = igbvf_suspend,
2870         .resume   = igbvf_resume,
2871 #endif
2872         .shutdown = igbvf_shutdown,
2873         .err_handler = &igbvf_err_handler
2874 };
2875
2876 /**
2877  * igbvf_init_module - Driver Registration Routine
2878  *
2879  * igbvf_init_module is the first routine called when the driver is
2880  * loaded. All it does is register with the PCI subsystem.
2881  **/
2882 static int __init igbvf_init_module(void)
2883 {
2884         int ret;
2885         printk(KERN_INFO "%s - version %s\n",
2886                igbvf_driver_string, igbvf_driver_version);
2887         printk(KERN_INFO "%s\n", igbvf_copyright);
2888
2889         ret = pci_register_driver(&igbvf_driver);
2890         pm_qos_add_requirement(PM_QOS_CPU_DMA_LATENCY, igbvf_driver_name,
2891                                PM_QOS_DEFAULT_VALUE);
2892
2893         return ret;
2894 }
2895 module_init(igbvf_init_module);
2896
2897 /**
2898  * igbvf_exit_module - Driver Exit Cleanup Routine
2899  *
2900  * igbvf_exit_module is called just before the driver is removed
2901  * from memory.
2902  **/
2903 static void __exit igbvf_exit_module(void)
2904 {
2905         pci_unregister_driver(&igbvf_driver);
2906         pm_qos_remove_requirement(PM_QOS_CPU_DMA_LATENCY, igbvf_driver_name);
2907 }
2908 module_exit(igbvf_exit_module);
2909
2910
2911 MODULE_AUTHOR("Intel Corporation, <e1000-devel@lists.sourceforge.net>");
2912 MODULE_DESCRIPTION("Intel(R) 82576 Virtual Function Network Driver");
2913 MODULE_LICENSE("GPL");
2914 MODULE_VERSION(DRV_VERSION);
2915
2916 /* netdev.c */