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 <linux/slab.h>
  39 #include <net/checksum.h>
  40 #include <net/ip6_checksum.h>
  41 #include <linux/mii.h>
  42 #include <linux/ethtool.h>
  43 #include <linux/if_vlan.h>
  44 #include <linux/pm_qos_params.h>
  45
  46 #include "igbvf.h"
  47
  48 #define DRV_VERSION "1.0.0-k0"
  49 char igbvf_driver_name[] = "igbvf";
  50 const char igbvf_driver_version[] = DRV_VERSION;
  51 static struct pm_qos_request_list igbvf_driver_pm_qos_req;
  52 static const char igbvf_driver_string[] =
  53                                 "Intel(R) Virtual Function Network Driver";
  54 static const char igbvf_copyright[] = "Copyright (c) 2009 Intel Corporation.";
  55
  56 static int igbvf_poll(struct napi_struct *napi, int budget);
  57 static void igbvf_reset(struct igbvf_adapter *);
  58 static void igbvf_set_interrupt_capability(struct igbvf_adapter *);
  59 static void igbvf_reset_interrupt_capability(struct igbvf_adapter *);
  60
  61 static struct igbvf_info igbvf_vf_info = {
  62         .mac                    = e1000_vfadapt,
  63         .flags                  = 0,
  64         .pba                    = 10,
  65         .init_ops               = e1000_init_function_pointers_vf,
  66 };
  67
  68 static const struct igbvf_info *igbvf_info_tbl[] = {
  69         [board_vf]              = &igbvf_vf_info,
  70 };
  71
  72 /**
  73  * igbvf_desc_unused - calculate if we have unused descriptors
  74  **/
  75 static int igbvf_desc_unused(struct igbvf_ring *ring)
  76 {
  77         if (ring->next_to_clean > ring->next_to_use)
  78                 return ring->next_to_clean - ring->next_to_use - 1;
  79
  80         return ring->count + ring->next_to_clean - ring->next_to_use - 1;
  81 }
  82
  83 /**
  84  * igbvf_receive_skb - helper function to handle Rx indications
  85  * @adapter: board private structure
  86  * @status: descriptor status field as written by hardware
  87  * @vlan: descriptor vlan field as written by hardware (no le/be conversion)
  88  * @skb: pointer to sk_buff to be indicated to stack
  89  **/
  90 static void igbvf_receive_skb(struct igbvf_adapter *adapter,
  91                               struct net_device *netdev,
  92                               struct sk_buff *skb,
  93                               u32 status, u16 vlan)
  94 {
  95         if (adapter->vlgrp && (status & E1000_RXD_STAT_VP))
  96                 vlan_hwaccel_receive_skb(skb, adapter->vlgrp,
  97                                          le16_to_cpu(vlan) &
  98                                          E1000_RXD_SPC_VLAN_MASK);
  99         else
 100                 netif_receive_skb(skb);
 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                                 dma_map_page(&pdev->dev, buffer_info->page,
 169                                              buffer_info->page_offset,
 170                                              PAGE_SIZE / 2,
 171                                              DMA_FROM_DEVICE);
 172                 }
 173
 174                 if (!buffer_info->skb) {
 175                         skb = netdev_alloc_skb_ip_align(netdev, bufsz);
 176                         if (!skb) {
 177                                 adapter->alloc_rx_buff_failed++;
 178                                 goto no_buffers;
 179                         }
 180
 181                         buffer_info->skb = skb;
 182                         buffer_info->dma = dma_map_single(&pdev->dev, skb->data,
 183                                                           bufsz,
 184                                                           DMA_FROM_DEVICE);
 185                 }
 186                 /* Refresh the desc even if buffer_addrs didn't change because
 187                  * each write-back erases this info. */
 188                 if (adapter->rx_ps_hdr_size) {
 189                         rx_desc->read.pkt_addr =
 190                              cpu_to_le64(buffer_info->page_dma);
 191                         rx_desc->read.hdr_addr = cpu_to_le64(buffer_info->dma);
 192                 } else {
 193                         rx_desc->read.pkt_addr =
 194                              cpu_to_le64(buffer_info->dma);
 195                         rx_desc->read.hdr_addr = 0;
 196                 }
 197
 198                 i++;
 199                 if (i == rx_ring->count)
 200                         i = 0;
 201                 buffer_info = &rx_ring->buffer_info[i];
 202         }
 203
 204 no_buffers:
 205         if (rx_ring->next_to_use != i) {
 206                 rx_ring->next_to_use = i;
 207                 if (i == 0)
 208                         i = (rx_ring->count - 1);
 209                 else
 210                         i--;
 211
 212                 /* Force memory writes to complete before letting h/w
 213                  * know there are new descriptors to fetch.  (Only
 214                  * applicable for weak-ordered memory model archs,
 215                  * such as IA-64). */
 216                 wmb();
 217                 writel(i, adapter->hw.hw_addr + rx_ring->tail);
 218         }
 219 }
 220
 221 /**
 222  * igbvf_clean_rx_irq - Send received data up the network stack; legacy
 223  * @adapter: board private structure
 224  *
 225  * the return value indicates whether actual cleaning was done, there
 226  * is no guarantee that everything was cleaned
 227  **/
 228 static bool igbvf_clean_rx_irq(struct igbvf_adapter *adapter,
 229                                int *work_done, int work_to_do)
 230 {
 231         struct igbvf_ring *rx_ring = adapter->rx_ring;
 232         struct net_device *netdev = adapter->netdev;
 233         struct pci_dev *pdev = adapter->pdev;
 234         union e1000_adv_rx_desc *rx_desc, *next_rxd;
 235         struct igbvf_buffer *buffer_info, *next_buffer;
 236         struct sk_buff *skb;
 237         bool cleaned = false;
 238         int cleaned_count = 0;
 239         unsigned int total_bytes = 0, total_packets = 0;
 240         unsigned int i;
 241         u32 length, hlen, staterr;
 242
 243         i = rx_ring->next_to_clean;
 244         rx_desc = IGBVF_RX_DESC_ADV(*rx_ring, i);
 245         staterr = le32_to_cpu(rx_desc->wb.upper.status_error);
 246
 247         while (staterr & E1000_RXD_STAT_DD) {
 248                 if (*work_done >= work_to_do)
 249                         break;
 250                 (*work_done)++;
 251                 rmb(); /* read descriptor and rx_buffer_info after status DD */
 252
 253                 buffer_info = &rx_ring->buffer_info[i];
 254
 255                 /* HW will not DMA in data larger than the given buffer, even
 256                  * if it parses the (NFS, of course) header to be larger.  In
 257                  * that case, it fills the header buffer and spills the rest
 258                  * into the page.
 259                  */
 260                 hlen = (le16_to_cpu(rx_desc->wb.lower.lo_dword.hs_rss.hdr_info) &
 261                   E1000_RXDADV_HDRBUFLEN_MASK) >> E1000_RXDADV_HDRBUFLEN_SHIFT;
 262                 if (hlen > adapter->rx_ps_hdr_size)
 263                         hlen = adapter->rx_ps_hdr_size;
 264
 265                 length = le16_to_cpu(rx_desc->wb.upper.length);
 266                 cleaned = true;
 267                 cleaned_count++;
 268
 269                 skb = buffer_info->skb;
 270                 prefetch(skb->data - NET_IP_ALIGN);
 271                 buffer_info->skb = NULL;
 272                 if (!adapter->rx_ps_hdr_size) {
 273                         dma_unmap_single(&pdev->dev, buffer_info->dma,
 274                                          adapter->rx_buffer_len,
 275                                          DMA_FROM_DEVICE);
 276                         buffer_info->dma = 0;
 277                         skb_put(skb, length);
 278                         goto send_up;
 279                 }
 280
 281                 if (!skb_shinfo(skb)->nr_frags) {
 282                         dma_unmap_single(&pdev->dev, buffer_info->dma,
 283                                          adapter->rx_ps_hdr_size,
 284                                          DMA_FROM_DEVICE);
 285                         skb_put(skb, hlen);
 286                 }
 287
 288                 if (length) {
 289                         dma_unmap_page(&pdev->dev, buffer_info->page_dma,
 290                                        PAGE_SIZE / 2,
 291                                        DMA_FROM_DEVICE);
 292                         buffer_info->page_dma = 0;
 293
 294                         skb_fill_page_desc(skb, skb_shinfo(skb)->nr_frags,
 295                                            buffer_info->page,
 296                                            buffer_info->page_offset,
 297                                            length);
 298
 299                         if ((adapter->rx_buffer_len > (PAGE_SIZE / 2)) ||
 300                             (page_count(buffer_info->page) != 1))
 301                                 buffer_info->page = NULL;
 302                         else
 303                                 get_page(buffer_info->page);
 304
 305                         skb->len += length;
 306                         skb->data_len += length;
 307                         skb->truesize += length;
 308                 }
 309 send_up:
 310                 i++;
 311                 if (i == rx_ring->count)
 312                         i = 0;
 313                 next_rxd = IGBVF_RX_DESC_ADV(*rx_ring, i);
 314                 prefetch(next_rxd);
 315                 next_buffer = &rx_ring->buffer_info[i];
 316
 317                 if (!(staterr & E1000_RXD_STAT_EOP)) {
 318                         buffer_info->skb = next_buffer->skb;
 319                         buffer_info->dma = next_buffer->dma;
 320                         next_buffer->skb = skb;
 321                         next_buffer->dma = 0;
 322                         goto next_desc;
 323                 }
 324
 325                 if (staterr & E1000_RXDEXT_ERR_FRAME_ERR_MASK) {
 326                         dev_kfree_skb_irq(skb);
 327                         goto next_desc;
 328                 }
 329
 330                 total_bytes += skb->len;
 331                 total_packets++;
 332
 333                 igbvf_rx_checksum_adv(adapter, staterr, skb);
 334
 335                 skb->protocol = eth_type_trans(skb, netdev);
 336
 337                 igbvf_receive_skb(adapter, netdev, skb, staterr,
 338                                   rx_desc->wb.upper.vlan);
 339
 340 next_desc:
 341                 rx_desc->wb.upper.status_error = 0;
 342
 343                 /* return some buffers to hardware, one at a time is too slow */
 344                 if (cleaned_count >= IGBVF_RX_BUFFER_WRITE) {
 345                         igbvf_alloc_rx_buffers(rx_ring, cleaned_count);
 346                         cleaned_count = 0;
 347                 }
 348
 349                 /* use prefetched values */
 350                 rx_desc = next_rxd;
 351                 buffer_info = next_buffer;
 352
 353                 staterr = le32_to_cpu(rx_desc->wb.upper.status_error);
 354         }
 355
 356         rx_ring->next_to_clean = i;
 357         cleaned_count = igbvf_desc_unused(rx_ring);
 358
 359         if (cleaned_count)
 360                 igbvf_alloc_rx_buffers(rx_ring, cleaned_count);
 361
 362         adapter->total_rx_packets += total_packets;
 363         adapter->total_rx_bytes += total_bytes;
 364         adapter->net_stats.rx_bytes += total_bytes;
 365         adapter->net_stats.rx_packets += total_packets;
 366         return cleaned;
 367 }
 368
 369 static void igbvf_put_txbuf(struct igbvf_adapter *adapter,
 370                             struct igbvf_buffer *buffer_info)
 371 {
 372         if (buffer_info->dma) {
 373                 if (buffer_info->mapped_as_page)
 374                         dma_unmap_page(&adapter->pdev->dev,
 375                                        buffer_info->dma,
 376                                        buffer_info->length,
 377                                        DMA_TO_DEVICE);
 378                 else
 379                         dma_unmap_single(&adapter->pdev->dev,
 380                                          buffer_info->dma,
 381                                          buffer_info->length,
 382                                          DMA_TO_DEVICE);
 383                 buffer_info->dma = 0;
 384         }
 385         if (buffer_info->skb) {
 386                 dev_kfree_skb_any(buffer_info->skb);
 387                 buffer_info->skb = NULL;
 388         }
 389         buffer_info->time_stamp = 0;
 390 }
 391
 392 static void igbvf_print_tx_hang(struct igbvf_adapter *adapter)
 393 {
 394         struct igbvf_ring *tx_ring = adapter->tx_ring;
 395         unsigned int i = tx_ring->next_to_clean;
 396         unsigned int eop = tx_ring->buffer_info[i].next_to_watch;
 397         union e1000_adv_tx_desc *eop_desc = IGBVF_TX_DESC_ADV(*tx_ring, eop);
 398
 399         /* detected Tx unit hang */
 400         dev_err(&adapter->pdev->dev,
 401                 "Detected Tx Unit Hang:\n"
 402                 "  TDH                  <%x>\n"
 403                 "  TDT                  <%x>\n"
 404                 "  next_to_use          <%x>\n"
 405                 "  next_to_clean        <%x>\n"
 406                 "buffer_info[next_to_clean]:\n"
 407                 "  time_stamp           <%lx>\n"
 408                 "  next_to_watch        <%x>\n"
 409                 "  jiffies              <%lx>\n"
 410                 "  next_to_watch.status <%x>\n",
 411                 readl(adapter->hw.hw_addr + tx_ring->head),
 412                 readl(adapter->hw.hw_addr + tx_ring->tail),
 413                 tx_ring->next_to_use,
 414                 tx_ring->next_to_clean,
 415                 tx_ring->buffer_info[eop].time_stamp,
 416                 eop,
 417                 jiffies,
 418                 eop_desc->wb.status);
 419 }
 420
 421 /**
 422  * igbvf_setup_tx_resources - allocate Tx resources (Descriptors)
 423  * @adapter: board private structure
 424  *
 425  * Return 0 on success, negative on failure
 426  **/
 427 int igbvf_setup_tx_resources(struct igbvf_adapter *adapter,
 428                              struct igbvf_ring *tx_ring)
 429 {
 430         struct pci_dev *pdev = adapter->pdev;
 431         int size;
 432
 433         size = sizeof(struct igbvf_buffer) * tx_ring->count;
 434         tx_ring->buffer_info = vmalloc(size);
 435         if (!tx_ring->buffer_info)
 436                 goto err;
 437         memset(tx_ring->buffer_info, 0, size);
 438
 439         /* round up to nearest 4K */
 440         tx_ring->size = tx_ring->count * sizeof(union e1000_adv_tx_desc);
 441         tx_ring->size = ALIGN(tx_ring->size, 4096);
 442
 443         tx_ring->desc = dma_alloc_coherent(&pdev->dev, tx_ring->size,
 444                                            &tx_ring->dma, GFP_KERNEL);
 445
 446         if (!tx_ring->desc)
 447                 goto err;
 448
 449         tx_ring->adapter = adapter;
 450         tx_ring->next_to_use = 0;
 451         tx_ring->next_to_clean = 0;
 452
 453         return 0;
 454 err:
 455         vfree(tx_ring->buffer_info);
 456         dev_err(&adapter->pdev->dev,
 457                 "Unable to allocate memory for the transmit descriptor ring\n");
 458         return -ENOMEM;
 459 }
 460
 461 /**
 462  * igbvf_setup_rx_resources - allocate Rx resources (Descriptors)
 463  * @adapter: board private structure
 464  *
 465  * Returns 0 on success, negative on failure
 466  **/
 467 int igbvf_setup_rx_resources(struct igbvf_adapter *adapter,
 468                              struct igbvf_ring *rx_ring)
 469 {
 470         struct pci_dev *pdev = adapter->pdev;
 471         int size, desc_len;
 472
 473         size = sizeof(struct igbvf_buffer) * rx_ring->count;
 474         rx_ring->buffer_info = vmalloc(size);
 475         if (!rx_ring->buffer_info)
 476                 goto err;
 477         memset(rx_ring->buffer_info, 0, size);
 478
 479         desc_len = sizeof(union e1000_adv_rx_desc);
 480
 481         /* Round up to nearest 4K */
 482         rx_ring->size = rx_ring->count * desc_len;
 483         rx_ring->size = ALIGN(rx_ring->size, 4096);
 484
 485         rx_ring->desc = dma_alloc_coherent(&pdev->dev, rx_ring->size,
 486                                            &rx_ring->dma, GFP_KERNEL);
 487
 488         if (!rx_ring->desc)
 489                 goto err;
 490
 491         rx_ring->next_to_clean = 0;
 492         rx_ring->next_to_use = 0;
 493
 494         rx_ring->adapter = adapter;
 495
 496         return 0;
 497
 498 err:
 499         vfree(rx_ring->buffer_info);
 500         rx_ring->buffer_info = NULL;
 501         dev_err(&adapter->pdev->dev,
 502                 "Unable to allocate memory for the receive descriptor ring\n");
 503         return -ENOMEM;
 504 }
 505
 506 /**
 507  * igbvf_clean_tx_ring - Free Tx Buffers
 508  * @tx_ring: ring to be cleaned
 509  **/
 510 static void igbvf_clean_tx_ring(struct igbvf_ring *tx_ring)
 511 {
 512         struct igbvf_adapter *adapter = tx_ring->adapter;
 513         struct igbvf_buffer *buffer_info;
 514         unsigned long size;
 515         unsigned int i;
 516
 517         if (!tx_ring->buffer_info)
 518                 return;
 519
 520         /* Free all the Tx ring sk_buffs */
 521         for (i = 0; i < tx_ring->count; i++) {
 522                 buffer_info = &tx_ring->buffer_info[i];
 523                 igbvf_put_txbuf(adapter, buffer_info);
 524         }
 525
 526         size = sizeof(struct igbvf_buffer) * tx_ring->count;
 527         memset(tx_ring->buffer_info, 0, size);
 528
 529         /* Zero out the descriptor ring */
 530         memset(tx_ring->desc, 0, tx_ring->size);
 531
 532         tx_ring->next_to_use = 0;
 533         tx_ring->next_to_clean = 0;
 534
 535         writel(0, adapter->hw.hw_addr + tx_ring->head);
 536         writel(0, adapter->hw.hw_addr + tx_ring->tail);
 537 }
 538
 539 /**
 540  * igbvf_free_tx_resources - Free Tx Resources per Queue
 541  * @tx_ring: ring to free resources from
 542  *
 543  * Free all transmit software resources
 544  **/
 545 void igbvf_free_tx_resources(struct igbvf_ring *tx_ring)
 546 {
 547         struct pci_dev *pdev = tx_ring->adapter->pdev;
 548
 549         igbvf_clean_tx_ring(tx_ring);
 550
 551         vfree(tx_ring->buffer_info);
 552         tx_ring->buffer_info = NULL;
 553
 554         dma_free_coherent(&pdev->dev, tx_ring->size, tx_ring->desc,
 555                           tx_ring->dma);
 556
 557         tx_ring->desc = NULL;
 558 }
 559
 560 /**
 561  * igbvf_clean_rx_ring - Free Rx Buffers per Queue
 562  * @adapter: board private structure
 563  **/
 564 static void igbvf_clean_rx_ring(struct igbvf_ring *rx_ring)
 565 {
 566         struct igbvf_adapter *adapter = rx_ring->adapter;
 567         struct igbvf_buffer *buffer_info;
 568         struct pci_dev *pdev = adapter->pdev;
 569         unsigned long size;
 570         unsigned int i;
 571
 572         if (!rx_ring->buffer_info)
 573                 return;
 574
 575         /* Free all the Rx ring sk_buffs */
 576         for (i = 0; i < rx_ring->count; i++) {
 577                 buffer_info = &rx_ring->buffer_info[i];
 578                 if (buffer_info->dma) {
 579                         if (adapter->rx_ps_hdr_size){
 580                                 dma_unmap_single(&pdev->dev, buffer_info->dma,
 581                                                  adapter->rx_ps_hdr_size,
 582                                                  DMA_FROM_DEVICE);
 583                         } else {
 584                                 dma_unmap_single(&pdev->dev, buffer_info->dma,
 585                                                  adapter->rx_buffer_len,
 586                                                  DMA_FROM_DEVICE);
 587                         }
 588                         buffer_info->dma = 0;
 589                 }
 590
 591                 if (buffer_info->skb) {
 592                         dev_kfree_skb(buffer_info->skb);
 593                         buffer_info->skb = NULL;
 594                 }
 595
 596                 if (buffer_info->page) {
 597                         if (buffer_info->page_dma)
 598                                 dma_unmap_page(&pdev->dev,
 599                                                buffer_info->page_dma,
 600                                                PAGE_SIZE / 2,
 601                                                DMA_FROM_DEVICE);
 602                         put_page(buffer_info->page);
 603                         buffer_info->page = NULL;
 604                         buffer_info->page_dma = 0;
 605                         buffer_info->page_offset = 0;
 606                 }
 607         }
 608
 609         size = sizeof(struct igbvf_buffer) * rx_ring->count;
 610         memset(rx_ring->buffer_info, 0, size);
 611
 612         /* Zero out the descriptor ring */
 613         memset(rx_ring->desc, 0, rx_ring->size);
 614
 615         rx_ring->next_to_clean = 0;
 616         rx_ring->next_to_use = 0;
 617
 618         writel(0, adapter->hw.hw_addr + rx_ring->head);
 619         writel(0, adapter->hw.hw_addr + rx_ring->tail);
 620 }
 621
 622 /**
 623  * igbvf_free_rx_resources - Free Rx Resources
 624  * @rx_ring: ring to clean the resources from
 625  *
 626  * Free all receive software resources
 627  **/
 628
 629 void igbvf_free_rx_resources(struct igbvf_ring *rx_ring)
 630 {
 631         struct pci_dev *pdev = rx_ring->adapter->pdev;
 632
 633         igbvf_clean_rx_ring(rx_ring);
 634
 635         vfree(rx_ring->buffer_info);
 636         rx_ring->buffer_info = NULL;
 637
 638         dma_free_coherent(&pdev->dev, rx_ring->size, rx_ring->desc,
 639                           rx_ring->dma);
 640         rx_ring->desc = NULL;
 641 }
 642
 643 /**
 644  * igbvf_update_itr - update the dynamic ITR value based on statistics
 645  * @adapter: pointer to adapter
 646  * @itr_setting: current adapter->itr
 647  * @packets: the number of packets during this measurement interval
 648  * @bytes: the number of bytes during this measurement interval
 649  *
 650  *      Stores a new ITR value based on packets and byte
 651  *      counts during the last interrupt.  The advantage of per interrupt
 652  *      computation is faster updates and more accurate ITR for the current
 653  *      traffic pattern.  Constants in this function were computed
 654  *      based on theoretical maximum wire speed and thresholds were set based
 655  *      on testing data as well as attempting to minimize response time
 656  *      while increasing bulk throughput.  This functionality is controlled
 657  *      by the InterruptThrottleRate module parameter.
 658  **/
 659 static unsigned int igbvf_update_itr(struct igbvf_adapter *adapter,
 660                                      u16 itr_setting, int packets,
 661                                      int bytes)
 662 {
 663         unsigned int retval = itr_setting;
 664
 665         if (packets == 0)
 666                 goto update_itr_done;
 667
 668         switch (itr_setting) {
 669         case lowest_latency:
 670                 /* handle TSO and jumbo frames */
 671                 if (bytes/packets > 8000)
 672                         retval = bulk_latency;
 673                 else if ((packets < 5) && (bytes > 512))
 674                         retval = low_latency;
 675                 break;
 676         case low_latency:  /* 50 usec aka 20000 ints/s */
 677                 if (bytes > 10000) {
 678                         /* this if handles the TSO accounting */
 679                         if (bytes/packets > 8000)
 680                                 retval = bulk_latency;
 681                         else if ((packets < 10) || ((bytes/packets) > 1200))
 682                                 retval = bulk_latency;
 683                         else if ((packets > 35))
 684                                 retval = lowest_latency;
 685                 } else if (bytes/packets > 2000) {
 686                         retval = bulk_latency;
 687                 } else if (packets <= 2 && bytes < 512) {
 688                         retval = lowest_latency;
 689                 }
 690                 break;
 691         case bulk_latency: /* 250 usec aka 4000 ints/s */
 692                 if (bytes > 25000) {
 693                         if (packets > 35)
 694                                 retval = low_latency;
 695                 } else if (bytes < 6000) {
 696                         retval = low_latency;
 697                 }
 698                 break;
 699         }
 700
 701 update_itr_done:
 702         return retval;
 703 }
 704
 705 static void igbvf_set_itr(struct igbvf_adapter *adapter)
 706 {
 707         struct e1000_hw *hw = &adapter->hw;
 708         u16 current_itr;
 709         u32 new_itr = adapter->itr;
 710
 711         adapter->tx_itr = igbvf_update_itr(adapter, adapter->tx_itr,
 712                                            adapter->total_tx_packets,
 713                                            adapter->total_tx_bytes);
 714         /* conservative mode (itr 3) eliminates the lowest_latency setting */
 715         if (adapter->itr_setting == 3 && adapter->tx_itr == lowest_latency)
 716                 adapter->tx_itr = low_latency;
 717
 718         adapter->rx_itr = igbvf_update_itr(adapter, adapter->rx_itr,
 719                                            adapter->total_rx_packets,
 720                                            adapter->total_rx_bytes);
 721         /* conservative mode (itr 3) eliminates the lowest_latency setting */
 722         if (adapter->itr_setting == 3 && adapter->rx_itr == lowest_latency)
 723                 adapter->rx_itr = low_latency;
 724
 725         current_itr = max(adapter->rx_itr, adapter->tx_itr);
 726
 727         switch (current_itr) {
 728         /* counts and packets in update_itr are dependent on these numbers */
 729         case lowest_latency:
 730                 new_itr = 70000;
 731                 break;
 732         case low_latency:
 733                 new_itr = 20000; /* aka hwitr = ~200 */
 734                 break;
 735         case bulk_latency:
 736                 new_itr = 4000;
 737                 break;
 738         default:
 739                 break;
 740         }
 741
 742         if (new_itr != adapter->itr) {
 743                 /*
 744                  * this attempts to bias the interrupt rate towards Bulk
 745                  * by adding intermediate steps when interrupt rate is
 746                  * increasing
 747                  */
 748                 new_itr = new_itr > adapter->itr ?
 749                              min(adapter->itr + (new_itr >> 2), new_itr) :
 750                              new_itr;
 751                 adapter->itr = new_itr;
 752                 adapter->rx_ring->itr_val = 1952;
 753
 754                 if (adapter->msix_entries)
 755                         adapter->rx_ring->set_itr = 1;
 756                 else
 757                         ew32(ITR, 1952);
 758         }
 759 }
 760
 761 /**
 762  * igbvf_clean_tx_irq - Reclaim resources after transmit completes
 763  * @adapter: board private structure
 764  * returns true if ring is completely cleaned
 765  **/
 766 static bool igbvf_clean_tx_irq(struct igbvf_ring *tx_ring)
 767 {
 768         struct igbvf_adapter *adapter = tx_ring->adapter;
 769         struct e1000_hw *hw = &adapter->hw;
 770         struct net_device *netdev = adapter->netdev;
 771         struct igbvf_buffer *buffer_info;
 772         struct sk_buff *skb;
 773         union e1000_adv_tx_desc *tx_desc, *eop_desc;
 774         unsigned int total_bytes = 0, total_packets = 0;
 775         unsigned int i, eop, count = 0;
 776         bool cleaned = false;
 777
 778         i = tx_ring->next_to_clean;
 779         eop = tx_ring->buffer_info[i].next_to_watch;
 780         eop_desc = IGBVF_TX_DESC_ADV(*tx_ring, eop);
 781
 782         while ((eop_desc->wb.status & cpu_to_le32(E1000_TXD_STAT_DD)) &&
 783                (count < tx_ring->count)) {
 784                 rmb();  /* read buffer_info after eop_desc status */
 785                 for (cleaned = false; !cleaned; count++) {
 786                         tx_desc = IGBVF_TX_DESC_ADV(*tx_ring, i);
 787                         buffer_info = &tx_ring->buffer_info[i];
 788                         cleaned = (i == eop);
 789                         skb = buffer_info->skb;
 790
 791                         if (skb) {
 792                                 unsigned int segs, bytecount;
 793
 794                                 /* gso_segs is currently only valid for tcp */
 795                                 segs = skb_shinfo(skb)->gso_segs ?: 1;
 796                                 /* multiply data chunks by size of headers */
 797                                 bytecount = ((segs - 1) * skb_headlen(skb)) +
 798                                             skb->len;
 799                                 total_packets += segs;
 800                                 total_bytes += bytecount;
 801                         }
 802
 803                         igbvf_put_txbuf(adapter, buffer_info);
 804                         tx_desc->wb.status = 0;
 805
 806                         i++;
 807                         if (i == tx_ring->count)
 808                                 i = 0;
 809                 }
 810                 eop = tx_ring->buffer_info[i].next_to_watch;
 811                 eop_desc = IGBVF_TX_DESC_ADV(*tx_ring, eop);
 812         }
 813
 814         tx_ring->next_to_clean = i;
 815
 816         if (unlikely(count &&
 817                      netif_carrier_ok(netdev) &&
 818                      igbvf_desc_unused(tx_ring) >= IGBVF_TX_QUEUE_WAKE)) {
 819                 /* Make sure that anybody stopping the queue after this
 820                  * sees the new next_to_clean.
 821                  */
 822                 smp_mb();
 823                 if (netif_queue_stopped(netdev) &&
 824                     !(test_bit(__IGBVF_DOWN, &adapter->state))) {
 825                         netif_wake_queue(netdev);
 826                         ++adapter->restart_queue;
 827                 }
 828         }
 829
 830         if (adapter->detect_tx_hung) {
 831                 /* Detect a transmit hang in hardware, this serializes the
 832                  * check with the clearing of time_stamp and movement of i */
 833                 adapter->detect_tx_hung = false;
 834                 if (tx_ring->buffer_info[i].time_stamp &&
 835                     time_after(jiffies, tx_ring->buffer_info[i].time_stamp +
 836                                (adapter->tx_timeout_factor * HZ)) &&
 837                     !(er32(STATUS) & E1000_STATUS_TXOFF)) {
 838
 839                         tx_desc = IGBVF_TX_DESC_ADV(*tx_ring, i);
 840                         /* detected Tx unit hang */
 841                         igbvf_print_tx_hang(adapter);
 842
 843                         netif_stop_queue(netdev);
 844                 }
 845         }
 846         adapter->net_stats.tx_bytes += total_bytes;
 847         adapter->net_stats.tx_packets += total_packets;
 848         return (count < tx_ring->count);
 849 }
 850
 851 static irqreturn_t igbvf_msix_other(int irq, void *data)
 852 {
 853         struct net_device *netdev = data;
 854         struct igbvf_adapter *adapter = netdev_priv(netdev);
 855         struct e1000_hw *hw = &adapter->hw;
 856
 857         adapter->int_counter1++;
 858
 859         netif_carrier_off(netdev);
 860         hw->mac.get_link_status = 1;
 861         if (!test_bit(__IGBVF_DOWN, &adapter->state))
 862                 mod_timer(&adapter->watchdog_timer, jiffies + 1);
 863
 864         ew32(EIMS, adapter->eims_other);
 865
 866         return IRQ_HANDLED;
 867 }
 868
 869 static irqreturn_t igbvf_intr_msix_tx(int irq, void *data)
 870 {
 871         struct net_device *netdev = data;
 872         struct igbvf_adapter *adapter = netdev_priv(netdev);
 873         struct e1000_hw *hw = &adapter->hw;
 874         struct igbvf_ring *tx_ring = adapter->tx_ring;
 875
 876
 877         adapter->total_tx_bytes = 0;
 878         adapter->total_tx_packets = 0;
 879
 880         /* auto mask will automatically reenable the interrupt when we write
 881          * EICS */
 882         if (!igbvf_clean_tx_irq(tx_ring))
 883                 /* Ring was not completely cleaned, so fire another interrupt */
 884                 ew32(EICS, tx_ring->eims_value);
 885         else
 886                 ew32(EIMS, tx_ring->eims_value);
 887
 888         return IRQ_HANDLED;
 889 }
 890
 891 static irqreturn_t igbvf_intr_msix_rx(int irq, void *data)
 892 {
 893         struct net_device *netdev = data;
 894         struct igbvf_adapter *adapter = netdev_priv(netdev);
 895
 896         adapter->int_counter0++;
 897
 898         /* Write the ITR value calculated at the end of the
 899          * previous interrupt.
 900          */
 901         if (adapter->rx_ring->set_itr) {
 902                 writel(adapter->rx_ring->itr_val,
 903                        adapter->hw.hw_addr + adapter->rx_ring->itr_register);
 904                 adapter->rx_ring->set_itr = 0;
 905         }
 906
 907         if (napi_schedule_prep(&adapter->rx_ring->napi)) {
 908                 adapter->total_rx_bytes = 0;
 909                 adapter->total_rx_packets = 0;
 910                 __napi_schedule(&adapter->rx_ring->napi);
 911         }
 912
 913         return IRQ_HANDLED;
 914 }
 915
 916 #define IGBVF_NO_QUEUE -1
 917
 918 static void igbvf_assign_vector(struct igbvf_adapter *adapter, int rx_queue,
 919                                 int tx_queue, int msix_vector)
 920 {
 921         struct e1000_hw *hw = &adapter->hw;
 922         u32 ivar, index;
 923
 924         /* 82576 uses a table-based method for assigning vectors.
 925            Each queue has a single entry in the table to which we write
 926            a vector number along with a "valid" bit.  Sadly, the layout
 927            of the table is somewhat counterintuitive. */
 928         if (rx_queue > IGBVF_NO_QUEUE) {
 929                 index = (rx_queue >> 1);
 930                 ivar = array_er32(IVAR0, index);
 931                 if (rx_queue & 0x1) {
 932                         /* vector goes into third byte of register */
 933                         ivar = ivar & 0xFF00FFFF;
 934                         ivar |= (msix_vector | E1000_IVAR_VALID) << 16;
 935                 } else {
 936                         /* vector goes into low byte of register */
 937                         ivar = ivar & 0xFFFFFF00;
 938                         ivar |= msix_vector | E1000_IVAR_VALID;
 939                 }
 940                 adapter->rx_ring[rx_queue].eims_value = 1 << msix_vector;
 941                 array_ew32(IVAR0, index, ivar);
 942         }
 943         if (tx_queue > IGBVF_NO_QUEUE) {
 944                 index = (tx_queue >> 1);
 945                 ivar = array_er32(IVAR0, index);
 946                 if (tx_queue & 0x1) {
 947                         /* vector goes into high byte of register */
 948                         ivar = ivar & 0x00FFFFFF;
 949                         ivar |= (msix_vector | E1000_IVAR_VALID) << 24;
 950                 } else {
 951                         /* vector goes into second byte of register */
 952                         ivar = ivar & 0xFFFF00FF;
 953                         ivar |= (msix_vector | E1000_IVAR_VALID) << 8;
 954                 }
 955                 adapter->tx_ring[tx_queue].eims_value = 1 << msix_vector;
 956                 array_ew32(IVAR0, index, ivar);
 957         }
 958 }
 959
 960 /**
 961  * igbvf_configure_msix - Configure MSI-X hardware
 962  *
 963  * igbvf_configure_msix sets up the hardware to properly
 964  * generate MSI-X interrupts.
 965  **/
 966 static void igbvf_configure_msix(struct igbvf_adapter *adapter)
 967 {
 968         u32 tmp;
 969         struct e1000_hw *hw = &adapter->hw;
 970         struct igbvf_ring *tx_ring = adapter->tx_ring;
 971         struct igbvf_ring *rx_ring = adapter->rx_ring;
 972         int vector = 0;
 973
 974         adapter->eims_enable_mask = 0;
 975
 976         igbvf_assign_vector(adapter, IGBVF_NO_QUEUE, 0, vector++);
 977         adapter->eims_enable_mask |= tx_ring->eims_value;
 978         if (tx_ring->itr_val)
 979                 writel(tx_ring->itr_val,
 980                        hw->hw_addr + tx_ring->itr_register);
 981         else
 982                 writel(1952, hw->hw_addr + tx_ring->itr_register);
 983
 984         igbvf_assign_vector(adapter, 0, IGBVF_NO_QUEUE, vector++);
 985         adapter->eims_enable_mask |= rx_ring->eims_value;
 986         if (rx_ring->itr_val)
 987                 writel(rx_ring->itr_val,
 988                        hw->hw_addr + rx_ring->itr_register);
 989         else
 990                 writel(1952, hw->hw_addr + rx_ring->itr_register);
 991
 992         /* set vector for other causes, i.e. link changes */
 993
 994         tmp = (vector++ | E1000_IVAR_VALID);
 995
 996         ew32(IVAR_MISC, tmp);
 997
 998         adapter->eims_enable_mask = (1 << (vector)) - 1;
 999         adapter->eims_other = 1 << (vector - 1);
1000         e1e_flush();
1001 }
1002
1003 static void igbvf_reset_interrupt_capability(struct igbvf_adapter *adapter)
1004 {
1005         if (adapter->msix_entries) {
1006                 pci_disable_msix(adapter->pdev);
1007                 kfree(adapter->msix_entries);
1008                 adapter->msix_entries = NULL;
1009         }
1010 }
1011
1012 /**
1013  * igbvf_set_interrupt_capability - set MSI or MSI-X if supported
1014  *
1015  * Attempt to configure interrupts using the best available
1016  * capabilities of the hardware and kernel.
1017  **/
1018 static void igbvf_set_interrupt_capability(struct igbvf_adapter *adapter)
1019 {
1020         int err = -ENOMEM;
1021         int i;
1022
1023         /* we allocate 3 vectors, 1 for tx, 1 for rx, one for pf messages */
1024         adapter->msix_entries = kcalloc(3, sizeof(struct msix_entry),
1025                                         GFP_KERNEL);
1026         if (adapter->msix_entries) {
1027                 for (i = 0; i < 3; i++)
1028                         adapter->msix_entries[i].entry = i;
1029
1030                 err = pci_enable_msix(adapter->pdev,
1031                                       adapter->msix_entries, 3);
1032         }
1033
1034         if (err) {
1035                 /* MSI-X failed */
1036                 dev_err(&adapter->pdev->dev,
1037                         "Failed to initialize MSI-X interrupts.\n");
1038                 igbvf_reset_interrupt_capability(adapter);
1039         }
1040 }
1041
1042 /**
1043  * igbvf_request_msix - Initialize MSI-X interrupts
1044  *
1045  * igbvf_request_msix allocates MSI-X vectors and requests interrupts from the
1046  * kernel.
1047  **/
1048 static int igbvf_request_msix(struct igbvf_adapter *adapter)
1049 {
1050         struct net_device *netdev = adapter->netdev;
1051         int err = 0, vector = 0;
1052
1053         if (strlen(netdev->name) < (IFNAMSIZ - 5)) {
1054                 sprintf(adapter->tx_ring->name, "%s-tx-0", netdev->name);
1055                 sprintf(adapter->rx_ring->name, "%s-rx-0", netdev->name);
1056         } else {
1057                 memcpy(adapter->tx_ring->name, netdev->name, IFNAMSIZ);
1058                 memcpy(adapter->rx_ring->name, netdev->name, IFNAMSIZ);
1059         }
1060
1061         err = request_irq(adapter->msix_entries[vector].vector,
1062                           igbvf_intr_msix_tx, 0, adapter->tx_ring->name,
1063                           netdev);
1064         if (err)
1065                 goto out;
1066
1067         adapter->tx_ring->itr_register = E1000_EITR(vector);
1068         adapter->tx_ring->itr_val = 1952;
1069         vector++;
1070
1071         err = request_irq(adapter->msix_entries[vector].vector,
1072                           igbvf_intr_msix_rx, 0, adapter->rx_ring->name,
1073                           netdev);
1074         if (err)
1075                 goto out;
1076
1077         adapter->rx_ring->itr_register = E1000_EITR(vector);
1078         adapter->rx_ring->itr_val = 1952;
1079         vector++;
1080
1081         err = request_irq(adapter->msix_entries[vector].vector,
1082                           igbvf_msix_other, 0, netdev->name, netdev);
1083         if (err)
1084                 goto out;
1085
1086         igbvf_configure_msix(adapter);
1087         return 0;
1088 out:
1089         return err;
1090 }
1091
1092 /**
1093  * igbvf_alloc_queues - Allocate memory for all rings
1094  * @adapter: board private structure to initialize
1095  **/
1096 static int __devinit igbvf_alloc_queues(struct igbvf_adapter *adapter)
1097 {
1098         struct net_device *netdev = adapter->netdev;
1099
1100         adapter->tx_ring = kzalloc(sizeof(struct igbvf_ring), GFP_KERNEL);
1101         if (!adapter->tx_ring)
1102                 return -ENOMEM;
1103
1104         adapter->rx_ring = kzalloc(sizeof(struct igbvf_ring), GFP_KERNEL);
1105         if (!adapter->rx_ring) {
1106                 kfree(adapter->tx_ring);
1107                 return -ENOMEM;
1108         }
1109
1110         netif_napi_add(netdev, &adapter->rx_ring->napi, igbvf_poll, 64);
1111
1112         return 0;
1113 }
1114
1115 /**
1116  * igbvf_request_irq - initialize interrupts
1117  *
1118  * Attempts to configure interrupts using the best available
1119  * capabilities of the hardware and kernel.
1120  **/
1121 static int igbvf_request_irq(struct igbvf_adapter *adapter)
1122 {
1123         int err = -1;
1124
1125         /* igbvf supports msi-x only */
1126         if (adapter->msix_entries)
1127                 err = igbvf_request_msix(adapter);
1128
1129         if (!err)
1130                 return err;
1131
1132         dev_err(&adapter->pdev->dev,
1133                 "Unable to allocate interrupt, Error: %d\n", err);
1134
1135         return err;
1136 }
1137
1138 static void igbvf_free_irq(struct igbvf_adapter *adapter)
1139 {
1140         struct net_device *netdev = adapter->netdev;
1141         int vector;
1142
1143         if (adapter->msix_entries) {
1144                 for (vector = 0; vector < 3; vector++)
1145                         free_irq(adapter->msix_entries[vector].vector, netdev);
1146         }
1147 }
1148
1149 /**
1150  * igbvf_irq_disable - Mask off interrupt generation on the NIC
1151  **/
1152 static void igbvf_irq_disable(struct igbvf_adapter *adapter)
1153 {
1154         struct e1000_hw *hw = &adapter->hw;
1155
1156         ew32(EIMC, ~0);
1157
1158         if (adapter->msix_entries)
1159                 ew32(EIAC, 0);
1160 }
1161
1162 /**
1163  * igbvf_irq_enable - Enable default interrupt generation settings
1164  **/
1165 static void igbvf_irq_enable(struct igbvf_adapter *adapter)
1166 {
1167         struct e1000_hw *hw = &adapter->hw;
1168
1169         ew32(EIAC, adapter->eims_enable_mask);
1170         ew32(EIAM, adapter->eims_enable_mask);
1171         ew32(EIMS, adapter->eims_enable_mask);
1172 }
1173
1174 /**
1175  * igbvf_poll - NAPI Rx polling callback
1176  * @napi: struct associated with this polling callback
1177  * @budget: amount of packets driver is allowed to process this poll
1178  **/
1179 static int igbvf_poll(struct napi_struct *napi, int budget)
1180 {
1181         struct igbvf_ring *rx_ring = container_of(napi, struct igbvf_ring, napi);
1182         struct igbvf_adapter *adapter = rx_ring->adapter;
1183         struct e1000_hw *hw = &adapter->hw;
1184         int work_done = 0;
1185
1186         igbvf_clean_rx_irq(adapter, &work_done, budget);
1187
1188         /* If not enough Rx work done, exit the polling mode */
1189         if (work_done < budget) {
1190                 napi_complete(napi);
1191
1192                 if (adapter->itr_setting & 3)
1193                         igbvf_set_itr(adapter);
1194
1195                 if (!test_bit(__IGBVF_DOWN, &adapter->state))
1196                         ew32(EIMS, adapter->rx_ring->eims_value);
1197         }
1198
1199         return work_done;
1200 }
1201
1202 /**
1203  * igbvf_set_rlpml - set receive large packet maximum length
1204  * @adapter: board private structure
1205  *
1206  * Configure the maximum size of packets that will be received
1207  */
1208 static void igbvf_set_rlpml(struct igbvf_adapter *adapter)
1209 {
1210         int max_frame_size = adapter->max_frame_size;
1211         struct e1000_hw *hw = &adapter->hw;
1212
1213         if (adapter->vlgrp)
1214                 max_frame_size += VLAN_TAG_SIZE;
1215
1216         e1000_rlpml_set_vf(hw, max_frame_size);
1217 }
1218
1219 static void igbvf_vlan_rx_add_vid(struct net_device *netdev, u16 vid)
1220 {
1221         struct igbvf_adapter *adapter = netdev_priv(netdev);
1222         struct e1000_hw *hw = &adapter->hw;
1223
1224         if (hw->mac.ops.set_vfta(hw, vid, true))
1225                 dev_err(&adapter->pdev->dev, "Failed to add vlan id %d\n", vid);
1226 }
1227
1228 static void igbvf_vlan_rx_kill_vid(struct net_device *netdev, u16 vid)
1229 {
1230         struct igbvf_adapter *adapter = netdev_priv(netdev);
1231         struct e1000_hw *hw = &adapter->hw;
1232
1233         igbvf_irq_disable(adapter);
1234         vlan_group_set_device(adapter->vlgrp, vid, NULL);
1235
1236         if (!test_bit(__IGBVF_DOWN, &adapter->state))
1237                 igbvf_irq_enable(adapter);
1238
1239         if (hw->mac.ops.set_vfta(hw, vid, false))
1240                 dev_err(&adapter->pdev->dev,
1241                         "Failed to remove vlan id %d\n", vid);
1242 }
1243
1244 static void igbvf_vlan_rx_register(struct net_device *netdev,
1245                                    struct vlan_group *grp)
1246 {
1247         struct igbvf_adapter *adapter = netdev_priv(netdev);
1248
1249         adapter->vlgrp = grp;
1250 }
1251
1252 static void igbvf_restore_vlan(struct igbvf_adapter *adapter)
1253 {
1254         u16 vid;
1255
1256         if (!adapter->vlgrp)
1257                 return;
1258
1259         for (vid = 0; vid < VLAN_GROUP_ARRAY_LEN; vid++) {
1260                 if (!vlan_group_get_device(adapter->vlgrp, vid))
1261                         continue;
1262                 igbvf_vlan_rx_add_vid(adapter->netdev, vid);
1263         }
1264
1265         igbvf_set_rlpml(adapter);
1266 }
1267
1268 /**
1269  * igbvf_configure_tx - Configure Transmit Unit after Reset
1270  * @adapter: board private structure
1271  *
1272  * Configure the Tx unit of the MAC after a reset.
1273  **/
1274 static void igbvf_configure_tx(struct igbvf_adapter *adapter)
1275 {
1276         struct e1000_hw *hw = &adapter->hw;
1277         struct igbvf_ring *tx_ring = adapter->tx_ring;
1278         u64 tdba;
1279         u32 txdctl, dca_txctrl;
1280
1281         /* disable transmits */
1282         txdctl = er32(TXDCTL(0));
1283         ew32(TXDCTL(0), txdctl & ~E1000_TXDCTL_QUEUE_ENABLE);
1284         msleep(10);
1285
1286         /* Setup the HW Tx Head and Tail descriptor pointers */
1287         ew32(TDLEN(0), tx_ring->count * sizeof(union e1000_adv_tx_desc));
1288         tdba = tx_ring->dma;
1289         ew32(TDBAL(0), (tdba & DMA_BIT_MASK(32)));
1290         ew32(TDBAH(0), (tdba >> 32));
1291         ew32(TDH(0), 0);
1292         ew32(TDT(0), 0);
1293         tx_ring->head = E1000_TDH(0);
1294         tx_ring->tail = E1000_TDT(0);
1295
1296         /* Turn off Relaxed Ordering on head write-backs.  The writebacks
1297          * MUST be delivered in order or it will completely screw up
1298          * our bookeeping.
1299          */
1300         dca_txctrl = er32(DCA_TXCTRL(0));
1301         dca_txctrl &= ~E1000_DCA_TXCTRL_TX_WB_RO_EN;
1302         ew32(DCA_TXCTRL(0), dca_txctrl);
1303
1304         /* enable transmits */
1305         txdctl |= E1000_TXDCTL_QUEUE_ENABLE;
1306         ew32(TXDCTL(0), txdctl);
1307
1308         /* Setup Transmit Descriptor Settings for eop descriptor */
1309         adapter->txd_cmd = E1000_ADVTXD_DCMD_EOP | E1000_ADVTXD_DCMD_IFCS;
1310
1311         /* enable Report Status bit */
1312         adapter->txd_cmd |= E1000_ADVTXD_DCMD_RS;
1313 }
1314
1315 /**
1316  * igbvf_setup_srrctl - configure the receive control registers
1317  * @adapter: Board private structure
1318  **/
1319 static void igbvf_setup_srrctl(struct igbvf_adapter *adapter)
1320 {
1321         struct e1000_hw *hw = &adapter->hw;
1322         u32 srrctl = 0;
1323
1324         srrctl &= ~(E1000_SRRCTL_DESCTYPE_MASK |
1325                     E1000_SRRCTL_BSIZEHDR_MASK |
1326                     E1000_SRRCTL_BSIZEPKT_MASK);
1327
1328         /* Enable queue drop to avoid head of line blocking */
1329         srrctl |= E1000_SRRCTL_DROP_EN;
1330
1331         /* Setup buffer sizes */
1332         srrctl |= ALIGN(adapter->rx_buffer_len, 1024) >>
1333                   E1000_SRRCTL_BSIZEPKT_SHIFT;
1334
1335         if (adapter->rx_buffer_len < 2048) {
1336                 adapter->rx_ps_hdr_size = 0;
1337                 srrctl |= E1000_SRRCTL_DESCTYPE_ADV_ONEBUF;
1338         } else {
1339                 adapter->rx_ps_hdr_size = 128;
1340                 srrctl |= adapter->rx_ps_hdr_size <<
1341                           E1000_SRRCTL_BSIZEHDRSIZE_SHIFT;
1342                 srrctl |= E1000_SRRCTL_DESCTYPE_HDR_SPLIT_ALWAYS;
1343         }
1344
1345         ew32(SRRCTL(0), srrctl);
1346 }
1347
1348 /**
1349  * igbvf_configure_rx - Configure Receive Unit after Reset
1350  * @adapter: board private structure
1351  *
1352  * Configure the Rx unit of the MAC after a reset.
1353  **/
1354 static void igbvf_configure_rx(struct igbvf_adapter *adapter)
1355 {
1356         struct e1000_hw *hw = &adapter->hw;
1357         struct igbvf_ring *rx_ring = adapter->rx_ring;
1358         u64 rdba;
1359         u32 rdlen, rxdctl;
1360
1361         /* disable receives */
1362         rxdctl = er32(RXDCTL(0));
1363         ew32(RXDCTL(0), rxdctl & ~E1000_RXDCTL_QUEUE_ENABLE);
1364         msleep(10);
1365
1366         rdlen = rx_ring->count * sizeof(union e1000_adv_rx_desc);
1367
1368         /*
1369          * Setup the HW Rx Head and Tail Descriptor Pointers and
1370          * the Base and Length of the Rx Descriptor Ring
1371          */
1372         rdba = rx_ring->dma;
1373         ew32(RDBAL(0), (rdba & DMA_BIT_MASK(32)));
1374         ew32(RDBAH(0), (rdba >> 32));
1375         ew32(RDLEN(0), rx_ring->count * sizeof(union e1000_adv_rx_desc));
1376         rx_ring->head = E1000_RDH(0);
1377         rx_ring->tail = E1000_RDT(0);
1378         ew32(RDH(0), 0);
1379         ew32(RDT(0), 0);
1380
1381         rxdctl |= E1000_RXDCTL_QUEUE_ENABLE;
1382         rxdctl &= 0xFFF00000;
1383         rxdctl |= IGBVF_RX_PTHRESH;
1384         rxdctl |= IGBVF_RX_HTHRESH << 8;
1385         rxdctl |= IGBVF_RX_WTHRESH << 16;
1386
1387         igbvf_set_rlpml(adapter);
1388
1389         /* enable receives */
1390         ew32(RXDCTL(0), rxdctl);
1391 }
1392
1393 /**
1394  * igbvf_set_multi - Multicast and Promiscuous mode set
1395  * @netdev: network interface device structure
1396  *
1397  * The set_multi entry point is called whenever the multicast address
1398  * list or the network interface flags are updated.  This routine is
1399  * responsible for configuring the hardware for proper multicast,
1400  * promiscuous mode, and all-multi behavior.
1401  **/
1402 static void igbvf_set_multi(struct net_device *netdev)
1403 {
1404         struct igbvf_adapter *adapter = netdev_priv(netdev);
1405         struct e1000_hw *hw = &adapter->hw;
1406         struct netdev_hw_addr *ha;
1407         u8  *mta_list = NULL;
1408         int i;
1409
1410         if (!netdev_mc_empty(netdev)) {
1411                 mta_list = kmalloc(netdev_mc_count(netdev) * 6, GFP_ATOMIC);
1412                 if (!mta_list) {
1413                         dev_err(&adapter->pdev->dev,
1414                                 "failed to allocate multicast filter list\n");
1415                         return;
1416                 }
1417         }
1418
1419         /* prepare a packed array of only addresses. */
1420         i = 0;
1421         netdev_for_each_mc_addr(ha, netdev)
1422                 memcpy(mta_list + (i++ * ETH_ALEN), ha->addr, ETH_ALEN);
1423
1424         hw->mac.ops.update_mc_addr_list(hw, mta_list, i, 0, 0);
1425         kfree(mta_list);
1426 }
1427
1428 /**
1429  * igbvf_configure - configure the hardware for Rx and Tx
1430  * @adapter: private board structure
1431  **/
1432 static void igbvf_configure(struct igbvf_adapter *adapter)
1433 {
1434         igbvf_set_multi(adapter->netdev);
1435
1436         igbvf_restore_vlan(adapter);
1437
1438         igbvf_configure_tx(adapter);
1439         igbvf_setup_srrctl(adapter);
1440         igbvf_configure_rx(adapter);
1441         igbvf_alloc_rx_buffers(adapter->rx_ring,
1442                                igbvf_desc_unused(adapter->rx_ring));
1443 }
1444
1445 /* igbvf_reset - bring the hardware into a known good state
1446  *
1447  * This function boots the hardware and enables some settings that
1448  * require a configuration cycle of the hardware - those cannot be
1449  * set/changed during runtime. After reset the device needs to be
1450  * properly configured for Rx, Tx etc.
1451  */
1452 static void igbvf_reset(struct igbvf_adapter *adapter)
1453 {
1454         struct e1000_mac_info *mac = &adapter->hw.mac;
1455         struct net_device *netdev = adapter->netdev;
1456         struct e1000_hw *hw = &adapter->hw;
1457
1458         /* Allow time for pending master requests to run */
1459         if (mac->ops.reset_hw(hw))
1460                 dev_err(&adapter->pdev->dev, "PF still resetting\n");
1461
1462         mac->ops.init_hw(hw);
1463
1464         if (is_valid_ether_addr(adapter->hw.mac.addr)) {
1465                 memcpy(netdev->dev_addr, adapter->hw.mac.addr,
1466                        netdev->addr_len);
1467                 memcpy(netdev->perm_addr, adapter->hw.mac.addr,
1468                        netdev->addr_len);
1469         }
1470
1471         adapter->last_reset = jiffies;
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         netif_carrier_off(netdev);
1532
1533         /* record the stats before reset*/
1534         igbvf_update_stats(adapter);
1535
1536         adapter->link_speed = 0;
1537         adapter->link_duplex = 0;
1538
1539         igbvf_reset(adapter);
1540         igbvf_clean_tx_ring(adapter->tx_ring);
1541         igbvf_clean_rx_ring(adapter->rx_ring);
1542 }
1543
1544 void igbvf_reinit_locked(struct igbvf_adapter *adapter)
1545 {
1546         might_sleep();
1547         while (test_and_set_bit(__IGBVF_RESETTING, &adapter->state))
1548                 msleep(1);
1549         igbvf_down(adapter);
1550         igbvf_up(adapter);
1551         clear_bit(__IGBVF_RESETTING, &adapter->state);
1552 }
1553
1554 /**
1555  * igbvf_sw_init - Initialize general software structures (struct igbvf_adapter)
1556  * @adapter: board private structure to initialize
1557  *
1558  * igbvf_sw_init initializes the Adapter private data structure.
1559  * Fields are initialized based on PCI device information and
1560  * OS network device settings (MTU size).
1561  **/
1562 static int __devinit igbvf_sw_init(struct igbvf_adapter *adapter)
1563 {
1564         struct net_device *netdev = adapter->netdev;
1565         s32 rc;
1566
1567         adapter->rx_buffer_len = ETH_FRAME_LEN + VLAN_HLEN + ETH_FCS_LEN;
1568         adapter->rx_ps_hdr_size = 0;
1569         adapter->max_frame_size = netdev->mtu + ETH_HLEN + ETH_FCS_LEN;
1570         adapter->min_frame_size = ETH_ZLEN + ETH_FCS_LEN;
1571
1572         adapter->tx_int_delay = 8;
1573         adapter->tx_abs_int_delay = 32;
1574         adapter->rx_int_delay = 0;
1575         adapter->rx_abs_int_delay = 8;
1576         adapter->itr_setting = 3;
1577         adapter->itr = 20000;
1578
1579         /* Set various function pointers */
1580         adapter->ei->init_ops(&adapter->hw);
1581
1582         rc = adapter->hw.mac.ops.init_params(&adapter->hw);
1583         if (rc)
1584                 return rc;
1585
1586         rc = adapter->hw.mbx.ops.init_params(&adapter->hw);
1587         if (rc)
1588                 return rc;
1589
1590         igbvf_set_interrupt_capability(adapter);
1591
1592         if (igbvf_alloc_queues(adapter))
1593                 return -ENOMEM;
1594
1595         spin_lock_init(&adapter->tx_queue_lock);
1596
1597         /* Explicitly disable IRQ since the NIC can be in any state. */
1598         igbvf_irq_disable(adapter);
1599
1600         spin_lock_init(&adapter->stats_lock);
1601
1602         set_bit(__IGBVF_DOWN, &adapter->state);
1603         return 0;
1604 }
1605
1606 static void igbvf_initialize_last_counter_stats(struct igbvf_adapter *adapter)
1607 {
1608         struct e1000_hw *hw = &adapter->hw;
1609
1610         adapter->stats.last_gprc = er32(VFGPRC);
1611         adapter->stats.last_gorc = er32(VFGORC);
1612         adapter->stats.last_gptc = er32(VFGPTC);
1613         adapter->stats.last_gotc = er32(VFGOTC);
1614         adapter->stats.last_mprc = er32(VFMPRC);
1615         adapter->stats.last_gotlbc = er32(VFGOTLBC);
1616         adapter->stats.last_gptlbc = er32(VFGPTLBC);
1617         adapter->stats.last_gorlbc = er32(VFGORLBC);
1618         adapter->stats.last_gprlbc = er32(VFGPRLBC);
1619
1620         adapter->stats.base_gprc = er32(VFGPRC);
1621         adapter->stats.base_gorc = er32(VFGORC);
1622         adapter->stats.base_gptc = er32(VFGPTC);
1623         adapter->stats.base_gotc = er32(VFGOTC);
1624         adapter->stats.base_mprc = er32(VFMPRC);
1625         adapter->stats.base_gotlbc = er32(VFGOTLBC);
1626         adapter->stats.base_gptlbc = er32(VFGPTLBC);
1627         adapter->stats.base_gorlbc = er32(VFGORLBC);
1628         adapter->stats.base_gprlbc = er32(VFGPRLBC);
1629 }
1630
1631 /**
1632  * igbvf_open - Called when a network interface is made active
1633  * @netdev: network interface device structure
1634  *
1635  * Returns 0 on success, negative value on failure
1636  *
1637  * The open entry point is called when a network interface is made
1638  * active by the system (IFF_UP).  At this point all resources needed
1639  * for transmit and receive operations are allocated, the interrupt
1640  * handler is registered with the OS, the watchdog timer is started,
1641  * and the stack is notified that the interface is ready.
1642  **/
1643 static int igbvf_open(struct net_device *netdev)
1644 {
1645         struct igbvf_adapter *adapter = netdev_priv(netdev);
1646         struct e1000_hw *hw = &adapter->hw;
1647         int err;
1648
1649         /* disallow open during test */
1650         if (test_bit(__IGBVF_TESTING, &adapter->state))
1651                 return -EBUSY;
1652
1653         /* allocate transmit descriptors */
1654         err = igbvf_setup_tx_resources(adapter, adapter->tx_ring);
1655         if (err)
1656                 goto err_setup_tx;
1657
1658         /* allocate receive descriptors */
1659         err = igbvf_setup_rx_resources(adapter, adapter->rx_ring);
1660         if (err)
1661                 goto err_setup_rx;
1662
1663         /*
1664          * before we allocate an interrupt, we must be ready to handle it.
1665          * Setting DEBUG_SHIRQ in the kernel makes it fire an interrupt
1666          * as soon as we call pci_request_irq, so we have to setup our
1667          * clean_rx handler before we do so.
1668          */
1669         igbvf_configure(adapter);
1670
1671         err = igbvf_request_irq(adapter);
1672         if (err)
1673                 goto err_req_irq;
1674
1675         /* From here on the code is the same as igbvf_up() */
1676         clear_bit(__IGBVF_DOWN, &adapter->state);
1677
1678         napi_enable(&adapter->rx_ring->napi);
1679
1680         /* clear any pending interrupts */
1681         er32(EICR);
1682
1683         igbvf_irq_enable(adapter);
1684
1685         /* start the watchdog */
1686         hw->mac.get_link_status = 1;
1687         mod_timer(&adapter->watchdog_timer, jiffies + 1);
1688
1689         return 0;
1690
1691 err_req_irq:
1692         igbvf_free_rx_resources(adapter->rx_ring);
1693 err_setup_rx:
1694         igbvf_free_tx_resources(adapter->tx_ring);
1695 err_setup_tx:
1696         igbvf_reset(adapter);
1697
1698         return err;
1699 }
1700
1701 /**
1702  * igbvf_close - Disables a network interface
1703  * @netdev: network interface device structure
1704  *
1705  * Returns 0, this is not allowed to fail
1706  *
1707  * The close entry point is called when an interface is de-activated
1708  * by the OS.  The hardware is still under the drivers control, but
1709  * needs to be disabled.  A global MAC reset is issued to stop the
1710  * hardware, and all transmit and receive resources are freed.
1711  **/
1712 static int igbvf_close(struct net_device *netdev)
1713 {
1714         struct igbvf_adapter *adapter = netdev_priv(netdev);
1715
1716         WARN_ON(test_bit(__IGBVF_RESETTING, &adapter->state));
1717         igbvf_down(adapter);
1718
1719         igbvf_free_irq(adapter);
1720
1721         igbvf_free_tx_resources(adapter->tx_ring);
1722         igbvf_free_rx_resources(adapter->rx_ring);
1723
1724         return 0;
1725 }
1726 /**
1727  * igbvf_set_mac - Change the Ethernet Address of the NIC
1728  * @netdev: network interface device structure
1729  * @p: pointer to an address structure
1730  *
1731  * Returns 0 on success, negative on failure
1732  **/
1733 static int igbvf_set_mac(struct net_device *netdev, void *p)
1734 {
1735         struct igbvf_adapter *adapter = netdev_priv(netdev);
1736         struct e1000_hw *hw = &adapter->hw;
1737         struct sockaddr *addr = p;
1738
1739         if (!is_valid_ether_addr(addr->sa_data))
1740                 return -EADDRNOTAVAIL;
1741
1742         memcpy(hw->mac.addr, addr->sa_data, netdev->addr_len);
1743
1744         hw->mac.ops.rar_set(hw, hw->mac.addr, 0);
1745
1746         if (memcmp(addr->sa_data, hw->mac.addr, 6))
1747                 return -EADDRNOTAVAIL;
1748
1749         memcpy(netdev->dev_addr, addr->sa_data, netdev->addr_len);
1750
1751         return 0;
1752 }
1753
1754 #define UPDATE_VF_COUNTER(reg, name)                                    \
1755         {                                                               \
1756                 u32 current_counter = er32(reg);                        \
1757                 if (current_counter < adapter->stats.last_##name)       \
1758                         adapter->stats.name += 0x100000000LL;           \
1759                 adapter->stats.last_##name = current_counter;           \
1760                 adapter->stats.name &= 0xFFFFFFFF00000000LL;            \
1761                 adapter->stats.name |= current_counter;                 \
1762         }
1763
1764 /**
1765  * igbvf_update_stats - Update the board statistics counters
1766  * @adapter: board private structure
1767 **/
1768 void igbvf_update_stats(struct igbvf_adapter *adapter)
1769 {
1770         struct e1000_hw *hw = &adapter->hw;
1771         struct pci_dev *pdev = adapter->pdev;
1772
1773         /*
1774          * Prevent stats update while adapter is being reset, link is down
1775          * or if the pci connection is down.
1776          */
1777         if (adapter->link_speed == 0)
1778                 return;
1779
1780         if (test_bit(__IGBVF_RESETTING, &adapter->state))
1781                 return;
1782
1783         if (pci_channel_offline(pdev))
1784                 return;
1785
1786         UPDATE_VF_COUNTER(VFGPRC, gprc);
1787         UPDATE_VF_COUNTER(VFGORC, gorc);
1788         UPDATE_VF_COUNTER(VFGPTC, gptc);
1789         UPDATE_VF_COUNTER(VFGOTC, gotc);
1790         UPDATE_VF_COUNTER(VFMPRC, mprc);
1791         UPDATE_VF_COUNTER(VFGOTLBC, gotlbc);
1792         UPDATE_VF_COUNTER(VFGPTLBC, gptlbc);
1793         UPDATE_VF_COUNTER(VFGORLBC, gorlbc);
1794         UPDATE_VF_COUNTER(VFGPRLBC, gprlbc);
1795
1796         /* Fill out the OS statistics structure */
1797         adapter->net_stats.multicast = adapter->stats.mprc;
1798 }
1799
1800 static void igbvf_print_link_info(struct igbvf_adapter *adapter)
1801 {
1802         dev_info(&adapter->pdev->dev, "Link is Up %d Mbps %s\n",
1803                  adapter->link_speed,
1804                  ((adapter->link_duplex == FULL_DUPLEX) ?
1805                   "Full Duplex" : "Half Duplex"));
1806 }
1807
1808 static bool igbvf_has_link(struct igbvf_adapter *adapter)
1809 {
1810         struct e1000_hw *hw = &adapter->hw;
1811         s32 ret_val = E1000_SUCCESS;
1812         bool link_active;
1813
1814         /* If interface is down, stay link down */
1815         if (test_bit(__IGBVF_DOWN, &adapter->state))
1816                 return false;
1817
1818         ret_val = hw->mac.ops.check_for_link(hw);
1819         link_active = !hw->mac.get_link_status;
1820
1821         /* if check for link returns error we will need to reset */
1822         if (ret_val && time_after(jiffies, adapter->last_reset + (10 * HZ)))
1823                 schedule_work(&adapter->reset_task);
1824
1825         return link_active;
1826 }
1827
1828 /**
1829  * igbvf_watchdog - Timer Call-back
1830  * @data: pointer to adapter cast into an unsigned long
1831  **/
1832 static void igbvf_watchdog(unsigned long data)
1833 {
1834         struct igbvf_adapter *adapter = (struct igbvf_adapter *) data;
1835
1836         /* Do the rest outside of interrupt context */
1837         schedule_work(&adapter->watchdog_task);
1838 }
1839
1840 static void igbvf_watchdog_task(struct work_struct *work)
1841 {
1842         struct igbvf_adapter *adapter = container_of(work,
1843                                                      struct igbvf_adapter,
1844                                                      watchdog_task);
1845         struct net_device *netdev = adapter->netdev;
1846         struct e1000_mac_info *mac = &adapter->hw.mac;
1847         struct igbvf_ring *tx_ring = adapter->tx_ring;
1848         struct e1000_hw *hw = &adapter->hw;
1849         u32 link;
1850         int tx_pending = 0;
1851
1852         link = igbvf_has_link(adapter);
1853
1854         if (link) {
1855                 if (!netif_carrier_ok(netdev)) {
1856                         bool txb2b = 1;
1857
1858                         mac->ops.get_link_up_info(&adapter->hw,
1859                                                   &adapter->link_speed,
1860                                                   &adapter->link_duplex);
1861                         igbvf_print_link_info(adapter);
1862
1863                         /* adjust timeout factor according to speed/duplex */
1864                         adapter->tx_timeout_factor = 1;
1865                         switch (adapter->link_speed) {
1866                         case SPEED_10:
1867                                 txb2b = 0;
1868                                 adapter->tx_timeout_factor = 16;
1869                                 break;
1870                         case SPEED_100:
1871                                 txb2b = 0;
1872                                 /* maybe add some timeout factor ? */
1873                                 break;
1874                         }
1875
1876                         netif_carrier_on(netdev);
1877                         netif_wake_queue(netdev);
1878                 }
1879         } else {
1880                 if (netif_carrier_ok(netdev)) {
1881                         adapter->link_speed = 0;
1882                         adapter->link_duplex = 0;
1883                         dev_info(&adapter->pdev->dev, "Link is Down\n");
1884                         netif_carrier_off(netdev);
1885                         netif_stop_queue(netdev);
1886                 }
1887         }
1888
1889         if (netif_carrier_ok(netdev)) {
1890                 igbvf_update_stats(adapter);
1891         } else {
1892                 tx_pending = (igbvf_desc_unused(tx_ring) + 1 <
1893                               tx_ring->count);
1894                 if (tx_pending) {
1895                         /*
1896                          * We've lost link, so the controller stops DMA,
1897                          * but we've got queued Tx work that's never going
1898                          * to get done, so reset controller to flush Tx.
1899                          * (Do the reset outside of interrupt context).
1900                          */
1901                         adapter->tx_timeout_count++;
1902                         schedule_work(&adapter->reset_task);
1903                 }
1904         }
1905
1906         /* Cause software interrupt to ensure Rx ring is cleaned */
1907         ew32(EICS, adapter->rx_ring->eims_value);
1908
1909         /* Force detection of hung controller every watchdog period */
1910         adapter->detect_tx_hung = 1;
1911
1912         /* Reset the timer */
1913         if (!test_bit(__IGBVF_DOWN, &adapter->state))
1914                 mod_timer(&adapter->watchdog_timer,
1915                           round_jiffies(jiffies + (2 * HZ)));
1916 }
1917
1918 #define IGBVF_TX_FLAGS_CSUM             0x00000001
1919 #define IGBVF_TX_FLAGS_VLAN             0x00000002
1920 #define IGBVF_TX_FLAGS_TSO              0x00000004
1921 #define IGBVF_TX_FLAGS_IPV4             0x00000008
1922 #define IGBVF_TX_FLAGS_VLAN_MASK        0xffff0000
1923 #define IGBVF_TX_FLAGS_VLAN_SHIFT       16
1924
1925 static int igbvf_tso(struct igbvf_adapter *adapter,
1926                      struct igbvf_ring *tx_ring,
1927                      struct sk_buff *skb, u32 tx_flags, u8 *hdr_len)
1928 {
1929         struct e1000_adv_tx_context_desc *context_desc;
1930         unsigned int i;
1931         int err;
1932         struct igbvf_buffer *buffer_info;
1933         u32 info = 0, tu_cmd = 0;
1934         u32 mss_l4len_idx, l4len;
1935         *hdr_len = 0;
1936
1937         if (skb_header_cloned(skb)) {
1938                 err = pskb_expand_head(skb, 0, 0, GFP_ATOMIC);
1939                 if (err) {
1940                         dev_err(&adapter->pdev->dev,
1941                                 "igbvf_tso returning an error\n");
1942                         return err;
1943                 }
1944         }
1945
1946         l4len = tcp_hdrlen(skb);
1947         *hdr_len += l4len;
1948
1949         if (skb->protocol == htons(ETH_P_IP)) {
1950                 struct iphdr *iph = ip_hdr(skb);
1951                 iph->tot_len = 0;
1952                 iph->check = 0;
1953                 tcp_hdr(skb)->check = ~csum_tcpudp_magic(iph->saddr,
1954                                                          iph->daddr, 0,
1955                                                          IPPROTO_TCP,
1956                                                          0);
1957         } else if (skb_is_gso_v6(skb)) {
1958                 ipv6_hdr(skb)->payload_len = 0;
1959                 tcp_hdr(skb)->check = ~csum_ipv6_magic(&ipv6_hdr(skb)->saddr,
1960                                                        &ipv6_hdr(skb)->daddr,
1961                                                        0, IPPROTO_TCP, 0);
1962         }
1963
1964         i = tx_ring->next_to_use;
1965
1966         buffer_info = &tx_ring->buffer_info[i];
1967         context_desc = IGBVF_TX_CTXTDESC_ADV(*tx_ring, i);
1968         /* VLAN MACLEN IPLEN */
1969         if (tx_flags & IGBVF_TX_FLAGS_VLAN)
1970                 info |= (tx_flags & IGBVF_TX_FLAGS_VLAN_MASK);
1971         info |= (skb_network_offset(skb) << E1000_ADVTXD_MACLEN_SHIFT);
1972         *hdr_len += skb_network_offset(skb);
1973         info |= (skb_transport_header(skb) - skb_network_header(skb));
1974         *hdr_len += (skb_transport_header(skb) - skb_network_header(skb));
1975         context_desc->vlan_macip_lens = cpu_to_le32(info);
1976
1977         /* ADV DTYP TUCMD MKRLOC/ISCSIHEDLEN */
1978         tu_cmd |= (E1000_TXD_CMD_DEXT | E1000_ADVTXD_DTYP_CTXT);
1979
1980         if (skb->protocol == htons(ETH_P_IP))
1981                 tu_cmd |= E1000_ADVTXD_TUCMD_IPV4;
1982         tu_cmd |= E1000_ADVTXD_TUCMD_L4T_TCP;
1983
1984         context_desc->type_tucmd_mlhl = cpu_to_le32(tu_cmd);
1985
1986         /* MSS L4LEN IDX */
1987         mss_l4len_idx = (skb_shinfo(skb)->gso_size << E1000_ADVTXD_MSS_SHIFT);
1988         mss_l4len_idx |= (l4len << E1000_ADVTXD_L4LEN_SHIFT);
1989
1990         context_desc->mss_l4len_idx = cpu_to_le32(mss_l4len_idx);
1991         context_desc->seqnum_seed = 0;
1992
1993         buffer_info->time_stamp = jiffies;
1994         buffer_info->next_to_watch = i;
1995         buffer_info->dma = 0;
1996         i++;
1997         if (i == tx_ring->count)
1998                 i = 0;
1999
2000         tx_ring->next_to_use = i;
2001
2002         return true;
2003 }
2004
2005 static inline bool igbvf_tx_csum(struct igbvf_adapter *adapter,
2006                                  struct igbvf_ring *tx_ring,
2007                                  struct sk_buff *skb, u32 tx_flags)
2008 {
2009         struct e1000_adv_tx_context_desc *context_desc;
2010         unsigned int i;
2011         struct igbvf_buffer *buffer_info;
2012         u32 info = 0, tu_cmd = 0;
2013
2014         if ((skb->ip_summed == CHECKSUM_PARTIAL) ||
2015             (tx_flags & IGBVF_TX_FLAGS_VLAN)) {
2016                 i = tx_ring->next_to_use;
2017                 buffer_info = &tx_ring->buffer_info[i];
2018                 context_desc = IGBVF_TX_CTXTDESC_ADV(*tx_ring, i);
2019
2020                 if (tx_flags & IGBVF_TX_FLAGS_VLAN)
2021                         info |= (tx_flags & IGBVF_TX_FLAGS_VLAN_MASK);
2022
2023                 info |= (skb_network_offset(skb) << E1000_ADVTXD_MACLEN_SHIFT);
2024                 if (skb->ip_summed == CHECKSUM_PARTIAL)
2025                         info |= (skb_transport_header(skb) -
2026                                  skb_network_header(skb));
2027
2028
2029                 context_desc->vlan_macip_lens = cpu_to_le32(info);
2030
2031                 tu_cmd |= (E1000_TXD_CMD_DEXT | E1000_ADVTXD_DTYP_CTXT);
2032
2033                 if (skb->ip_summed == CHECKSUM_PARTIAL) {
2034                         switch (skb->protocol) {
2035                         case __constant_htons(ETH_P_IP):
2036                                 tu_cmd |= E1000_ADVTXD_TUCMD_IPV4;
2037                                 if (ip_hdr(skb)->protocol == IPPROTO_TCP)
2038                                         tu_cmd |= E1000_ADVTXD_TUCMD_L4T_TCP;
2039                                 break;
2040                         case __constant_htons(ETH_P_IPV6):
2041                                 if (ipv6_hdr(skb)->nexthdr == IPPROTO_TCP)
2042                                         tu_cmd |= E1000_ADVTXD_TUCMD_L4T_TCP;
2043                                 break;
2044                         default:
2045                                 break;
2046                         }
2047                 }
2048
2049                 context_desc->type_tucmd_mlhl = cpu_to_le32(tu_cmd);
2050                 context_desc->seqnum_seed = 0;
2051                 context_desc->mss_l4len_idx = 0;
2052
2053                 buffer_info->time_stamp = jiffies;
2054                 buffer_info->next_to_watch = i;
2055                 buffer_info->dma = 0;
2056                 i++;
2057                 if (i == tx_ring->count)
2058                         i = 0;
2059                 tx_ring->next_to_use = i;
2060
2061                 return true;
2062         }
2063
2064         return false;
2065 }
2066
2067 static int igbvf_maybe_stop_tx(struct net_device *netdev, int size)
2068 {
2069         struct igbvf_adapter *adapter = netdev_priv(netdev);
2070
2071         /* there is enough descriptors then we don't need to worry  */
2072         if (igbvf_desc_unused(adapter->tx_ring) >= size)
2073                 return 0;
2074
2075         netif_stop_queue(netdev);
2076
2077         smp_mb();
2078
2079         /* We need to check again just in case room has been made available */
2080         if (igbvf_desc_unused(adapter->tx_ring) < size)
2081                 return -EBUSY;
2082
2083         netif_wake_queue(netdev);
2084
2085         ++adapter->restart_queue;
2086         return 0;
2087 }
2088
2089 #define IGBVF_MAX_TXD_PWR       16
2090 #define IGBVF_MAX_DATA_PER_TXD  (1 << IGBVF_MAX_TXD_PWR)
2091
2092 static inline int igbvf_tx_map_adv(struct igbvf_adapter *adapter,
2093                                    struct igbvf_ring *tx_ring,
2094                                    struct sk_buff *skb,
2095                                    unsigned int first)
2096 {
2097         struct igbvf_buffer *buffer_info;
2098         struct pci_dev *pdev = adapter->pdev;
2099         unsigned int len = skb_headlen(skb);
2100         unsigned int count = 0, i;
2101         unsigned int f;
2102
2103         i = tx_ring->next_to_use;
2104
2105         buffer_info = &tx_ring->buffer_info[i];
2106         BUG_ON(len >= IGBVF_MAX_DATA_PER_TXD);
2107         buffer_info->length = len;
2108         /* set time_stamp *before* dma to help avoid a possible race */
2109         buffer_info->time_stamp = jiffies;
2110         buffer_info->next_to_watch = i;
2111         buffer_info->mapped_as_page = false;
2112         buffer_info->dma = dma_map_single(&pdev->dev, skb->data, len,
2113                                           DMA_TO_DEVICE);
2114         if (dma_mapping_error(&pdev->dev, buffer_info->dma))
2115                 goto dma_error;
2116
2117
2118         for (f = 0; f < skb_shinfo(skb)->nr_frags; f++) {
2119                 struct skb_frag_struct *frag;
2120
2121                 count++;
2122                 i++;
2123                 if (i == tx_ring->count)
2124                         i = 0;
2125
2126                 frag = &skb_shinfo(skb)->frags[f];
2127                 len = frag->size;
2128
2129                 buffer_info = &tx_ring->buffer_info[i];
2130                 BUG_ON(len >= IGBVF_MAX_DATA_PER_TXD);
2131                 buffer_info->length = len;
2132                 buffer_info->time_stamp = jiffies;
2133                 buffer_info->next_to_watch = i;
2134                 buffer_info->mapped_as_page = true;
2135                 buffer_info->dma = dma_map_page(&pdev->dev,
2136                                                 frag->page,
2137                                                 frag->page_offset,
2138                                                 len,
2139                                                 DMA_TO_DEVICE);
2140                 if (dma_mapping_error(&pdev->dev, buffer_info->dma))
2141                         goto dma_error;
2142         }
2143
2144         tx_ring->buffer_info[i].skb = skb;
2145         tx_ring->buffer_info[first].next_to_watch = i;
2146
2147         return ++count;
2148
2149 dma_error:
2150         dev_err(&pdev->dev, "TX DMA map failed\n");
2151
2152         /* clear timestamp and dma mappings for failed buffer_info mapping */
2153         buffer_info->dma = 0;
2154         buffer_info->time_stamp = 0;
2155         buffer_info->length = 0;
2156         buffer_info->next_to_watch = 0;
2157         buffer_info->mapped_as_page = false;
2158         if (count)
2159                 count--;
2160
2161         /* clear timestamp and dma mappings for remaining portion of packet */
2162         while (count--) {
2163                 if (i==0)
2164                         i += tx_ring->count;
2165                 i--;
2166                 buffer_info = &tx_ring->buffer_info[i];
2167                 igbvf_put_txbuf(adapter, buffer_info);
2168         }
2169
2170         return 0;
2171 }
2172
2173 static inline void igbvf_tx_queue_adv(struct igbvf_adapter *adapter,
2174                                       struct igbvf_ring *tx_ring,
2175                                       int tx_flags, int count, u32 paylen,
2176                                       u8 hdr_len)
2177 {
2178         union e1000_adv_tx_desc *tx_desc = NULL;
2179         struct igbvf_buffer *buffer_info;
2180         u32 olinfo_status = 0, cmd_type_len;
2181         unsigned int i;
2182
2183         cmd_type_len = (E1000_ADVTXD_DTYP_DATA | E1000_ADVTXD_DCMD_IFCS |
2184                         E1000_ADVTXD_DCMD_DEXT);
2185
2186         if (tx_flags & IGBVF_TX_FLAGS_VLAN)
2187                 cmd_type_len |= E1000_ADVTXD_DCMD_VLE;
2188
2189         if (tx_flags & IGBVF_TX_FLAGS_TSO) {
2190                 cmd_type_len |= E1000_ADVTXD_DCMD_TSE;
2191
2192                 /* insert tcp checksum */
2193                 olinfo_status |= E1000_TXD_POPTS_TXSM << 8;
2194
2195                 /* insert ip checksum */
2196                 if (tx_flags & IGBVF_TX_FLAGS_IPV4)
2197                         olinfo_status |= E1000_TXD_POPTS_IXSM << 8;
2198
2199         } else if (tx_flags & IGBVF_TX_FLAGS_CSUM) {
2200                 olinfo_status |= E1000_TXD_POPTS_TXSM << 8;
2201         }
2202
2203         olinfo_status |= ((paylen - hdr_len) << E1000_ADVTXD_PAYLEN_SHIFT);
2204
2205         i = tx_ring->next_to_use;
2206         while (count--) {
2207                 buffer_info = &tx_ring->buffer_info[i];
2208                 tx_desc = IGBVF_TX_DESC_ADV(*tx_ring, i);
2209                 tx_desc->read.buffer_addr = cpu_to_le64(buffer_info->dma);
2210                 tx_desc->read.cmd_type_len =
2211                          cpu_to_le32(cmd_type_len | buffer_info->length);
2212                 tx_desc->read.olinfo_status = cpu_to_le32(olinfo_status);
2213                 i++;
2214                 if (i == tx_ring->count)
2215                         i = 0;
2216         }
2217
2218         tx_desc->read.cmd_type_len |= cpu_to_le32(adapter->txd_cmd);
2219         /* Force memory writes to complete before letting h/w
2220          * know there are new descriptors to fetch.  (Only
2221          * applicable for weak-ordered memory model archs,
2222          * such as IA-64). */
2223         wmb();
2224
2225         tx_ring->next_to_use = i;
2226         writel(i, adapter->hw.hw_addr + tx_ring->tail);
2227         /* we need this if more than one processor can write to our tail
2228          * at a time, it syncronizes IO on IA64/Altix systems */
2229         mmiowb();
2230 }
2231
2232 static netdev_tx_t igbvf_xmit_frame_ring_adv(struct sk_buff *skb,
2233                                              struct net_device *netdev,
2234                                              struct igbvf_ring *tx_ring)
2235 {
2236         struct igbvf_adapter *adapter = netdev_priv(netdev);
2237         unsigned int first, tx_flags = 0;
2238         u8 hdr_len = 0;
2239         int count = 0;
2240         int tso = 0;
2241
2242         if (test_bit(__IGBVF_DOWN, &adapter->state)) {
2243                 dev_kfree_skb_any(skb);
2244                 return NETDEV_TX_OK;
2245         }
2246
2247         if (skb->len <= 0) {
2248                 dev_kfree_skb_any(skb);
2249                 return NETDEV_TX_OK;
2250         }
2251
2252         /*
2253          * need: count + 4 desc gap to keep tail from touching
2254          *       + 2 desc gap to keep tail from touching head,
2255          *       + 1 desc for skb->data,
2256          *       + 1 desc for context descriptor,
2257          * head, otherwise try next time
2258          */
2259         if (igbvf_maybe_stop_tx(netdev, skb_shinfo(skb)->nr_frags + 4)) {
2260                 /* this is a hard error */
2261                 return NETDEV_TX_BUSY;
2262         }
2263
2264         if (adapter->vlgrp && vlan_tx_tag_present(skb)) {
2265                 tx_flags |= IGBVF_TX_FLAGS_VLAN;
2266                 tx_flags |= (vlan_tx_tag_get(skb) << IGBVF_TX_FLAGS_VLAN_SHIFT);
2267         }
2268
2269         if (skb->protocol == htons(ETH_P_IP))
2270                 tx_flags |= IGBVF_TX_FLAGS_IPV4;
2271
2272         first = tx_ring->next_to_use;
2273
2274         tso = skb_is_gso(skb) ?
2275                 igbvf_tso(adapter, tx_ring, skb, tx_flags, &hdr_len) : 0;
2276         if (unlikely(tso < 0)) {
2277                 dev_kfree_skb_any(skb);
2278                 return NETDEV_TX_OK;
2279         }
2280
2281         if (tso)
2282                 tx_flags |= IGBVF_TX_FLAGS_TSO;
2283         else if (igbvf_tx_csum(adapter, tx_ring, skb, tx_flags) &&
2284                  (skb->ip_summed == CHECKSUM_PARTIAL))
2285                 tx_flags |= IGBVF_TX_FLAGS_CSUM;
2286
2287         /*
2288          * count reflects descriptors mapped, if 0 then mapping error
2289          * has occured and we need to rewind the descriptor queue
2290          */
2291         count = igbvf_tx_map_adv(adapter, tx_ring, skb, first);
2292
2293         if (count) {
2294                 igbvf_tx_queue_adv(adapter, tx_ring, tx_flags, count,
2295                                    skb->len, hdr_len);
2296                 /* Make sure there is space in the ring for the next send. */
2297                 igbvf_maybe_stop_tx(netdev, MAX_SKB_FRAGS + 4);
2298         } else {
2299                 dev_kfree_skb_any(skb);
2300                 tx_ring->buffer_info[first].time_stamp = 0;
2301                 tx_ring->next_to_use = first;
2302         }
2303
2304         return NETDEV_TX_OK;
2305 }
2306
2307 static netdev_tx_t igbvf_xmit_frame(struct sk_buff *skb,
2308                                     struct net_device *netdev)
2309 {
2310         struct igbvf_adapter *adapter = netdev_priv(netdev);
2311         struct igbvf_ring *tx_ring;
2312
2313         if (test_bit(__IGBVF_DOWN, &adapter->state)) {
2314                 dev_kfree_skb_any(skb);
2315                 return NETDEV_TX_OK;
2316         }
2317
2318         tx_ring = &adapter->tx_ring[0];
2319
2320         return igbvf_xmit_frame_ring_adv(skb, netdev, tx_ring);
2321 }
2322
2323 /**
2324  * igbvf_tx_timeout - Respond to a Tx Hang
2325  * @netdev: network interface device structure
2326  **/
2327 static void igbvf_tx_timeout(struct net_device *netdev)
2328 {
2329         struct igbvf_adapter *adapter = netdev_priv(netdev);
2330
2331         /* Do the reset outside of interrupt context */
2332         adapter->tx_timeout_count++;
2333         schedule_work(&adapter->reset_task);
2334 }
2335
2336 static void igbvf_reset_task(struct work_struct *work)
2337 {
2338         struct igbvf_adapter *adapter;
2339         adapter = container_of(work, struct igbvf_adapter, reset_task);
2340
2341         igbvf_reinit_locked(adapter);
2342 }
2343
2344 /**
2345  * igbvf_get_stats - Get System Network Statistics
2346  * @netdev: network interface device structure
2347  *
2348  * Returns the address of the device statistics structure.
2349  * The statistics are actually updated from the timer callback.
2350  **/
2351 static struct net_device_stats *igbvf_get_stats(struct net_device *netdev)
2352 {
2353         struct igbvf_adapter *adapter = netdev_priv(netdev);
2354
2355         /* only return the current stats */
2356         return &adapter->net_stats;
2357 }
2358
2359 /**
2360  * igbvf_change_mtu - Change the Maximum Transfer Unit
2361  * @netdev: network interface device structure
2362  * @new_mtu: new value for maximum frame size
2363  *
2364  * Returns 0 on success, negative on failure
2365  **/
2366 static int igbvf_change_mtu(struct net_device *netdev, int new_mtu)
2367 {
2368         struct igbvf_adapter *adapter = netdev_priv(netdev);
2369         int max_frame = new_mtu + ETH_HLEN + ETH_FCS_LEN;
2370
2371         if ((new_mtu < 68) || (max_frame > MAX_JUMBO_FRAME_SIZE)) {
2372                 dev_err(&adapter->pdev->dev, "Invalid MTU setting\n");
2373                 return -EINVAL;
2374         }
2375
2376 #define MAX_STD_JUMBO_FRAME_SIZE 9234
2377         if (max_frame > MAX_STD_JUMBO_FRAME_SIZE) {
2378                 dev_err(&adapter->pdev->dev, "MTU > 9216 not supported.\n");
2379                 return -EINVAL;
2380         }
2381
2382         while (test_and_set_bit(__IGBVF_RESETTING, &adapter->state))
2383                 msleep(1);
2384         /* igbvf_down has a dependency on max_frame_size */
2385         adapter->max_frame_size = max_frame;
2386         if (netif_running(netdev))
2387                 igbvf_down(adapter);
2388
2389         /*
2390          * NOTE: netdev_alloc_skb reserves 16 bytes, and typically NET_IP_ALIGN
2391          * means we reserve 2 more, this pushes us to allocate from the next
2392          * larger slab size.
2393          * i.e. RXBUFFER_2048 --> size-4096 slab
2394          * However with the new *_jumbo_rx* routines, jumbo receives will use
2395          * fragmented skbs
2396          */
2397
2398         if (max_frame <= 1024)
2399                 adapter->rx_buffer_len = 1024;
2400         else if (max_frame <= 2048)
2401                 adapter->rx_buffer_len = 2048;
2402         else
2403 #if (PAGE_SIZE / 2) > 16384
2404                 adapter->rx_buffer_len = 16384;
2405 #else
2406                 adapter->rx_buffer_len = PAGE_SIZE / 2;
2407 #endif
2408
2409
2410         /* adjust allocation if LPE protects us, and we aren't using SBP */
2411         if ((max_frame == ETH_FRAME_LEN + ETH_FCS_LEN) ||
2412              (max_frame == ETH_FRAME_LEN + VLAN_HLEN + ETH_FCS_LEN))
2413                 adapter->rx_buffer_len = ETH_FRAME_LEN + VLAN_HLEN +
2414                                          ETH_FCS_LEN;
2415
2416         dev_info(&adapter->pdev->dev, "changing MTU from %d to %d\n",
2417                  netdev->mtu, new_mtu);
2418         netdev->mtu = new_mtu;
2419
2420         if (netif_running(netdev))
2421                 igbvf_up(adapter);
2422         else
2423                 igbvf_reset(adapter);
2424
2425         clear_bit(__IGBVF_RESETTING, &adapter->state);
2426
2427         return 0;
2428 }
2429
2430 static int igbvf_ioctl(struct net_device *netdev, struct ifreq *ifr, int cmd)
2431 {
2432         switch (cmd) {
2433         default:
2434                 return -EOPNOTSUPP;
2435         }
2436 }
2437
2438 static int igbvf_suspend(struct pci_dev *pdev, pm_message_t state)
2439 {
2440         struct net_device *netdev = pci_get_drvdata(pdev);
2441         struct igbvf_adapter *adapter = netdev_priv(netdev);
2442 #ifdef CONFIG_PM
2443         int retval = 0;
2444 #endif
2445
2446         netif_device_detach(netdev);
2447
2448         if (netif_running(netdev)) {
2449                 WARN_ON(test_bit(__IGBVF_RESETTING, &adapter->state));
2450                 igbvf_down(adapter);
2451                 igbvf_free_irq(adapter);
2452         }
2453
2454 #ifdef CONFIG_PM
2455         retval = pci_save_state(pdev);
2456         if (retval)
2457                 return retval;
2458 #endif
2459
2460         pci_disable_device(pdev);
2461
2462         return 0;
2463 }
2464
2465 #ifdef CONFIG_PM
2466 static int igbvf_resume(struct pci_dev *pdev)
2467 {
2468         struct net_device *netdev = pci_get_drvdata(pdev);
2469         struct igbvf_adapter *adapter = netdev_priv(netdev);
2470         u32 err;
2471
2472         pci_restore_state(pdev);
2473         err = pci_enable_device_mem(pdev);
2474         if (err) {
2475                 dev_err(&pdev->dev, "Cannot enable PCI device from suspend\n");
2476                 return err;
2477         }
2478
2479         pci_set_master(pdev);
2480
2481         if (netif_running(netdev)) {
2482                 err = igbvf_request_irq(adapter);
2483                 if (err)
2484                         return err;
2485         }
2486
2487         igbvf_reset(adapter);
2488
2489         if (netif_running(netdev))
2490                 igbvf_up(adapter);
2491
2492         netif_device_attach(netdev);
2493
2494         return 0;
2495 }
2496 #endif
2497
2498 static void igbvf_shutdown(struct pci_dev *pdev)
2499 {
2500         igbvf_suspend(pdev, PMSG_SUSPEND);
2501 }
2502
2503 #ifdef CONFIG_NET_POLL_CONTROLLER
2504 /*
2505  * Polling 'interrupt' - used by things like netconsole to send skbs
2506  * without having to re-enable interrupts. It's not called while
2507  * the interrupt routine is executing.
2508  */
2509 static void igbvf_netpoll(struct net_device *netdev)
2510 {
2511         struct igbvf_adapter *adapter = netdev_priv(netdev);
2512
2513         disable_irq(adapter->pdev->irq);
2514
2515         igbvf_clean_tx_irq(adapter->tx_ring);
2516
2517         enable_irq(adapter->pdev->irq);
2518 }
2519 #endif
2520
2521 /**
2522  * igbvf_io_error_detected - called when PCI error is detected
2523  * @pdev: Pointer to PCI device
2524  * @state: The current pci connection state
2525  *
2526  * This function is called after a PCI bus error affecting
2527  * this device has been detected.
2528  */
2529 static pci_ers_result_t igbvf_io_error_detected(struct pci_dev *pdev,
2530                                                 pci_channel_state_t state)
2531 {
2532         struct net_device *netdev = pci_get_drvdata(pdev);
2533         struct igbvf_adapter *adapter = netdev_priv(netdev);
2534
2535         netif_device_detach(netdev);
2536
2537         if (state == pci_channel_io_perm_failure)
2538                 return PCI_ERS_RESULT_DISCONNECT;
2539
2540         if (netif_running(netdev))
2541                 igbvf_down(adapter);
2542         pci_disable_device(pdev);
2543
2544         /* Request a slot slot reset. */
2545         return PCI_ERS_RESULT_NEED_RESET;
2546 }
2547
2548 /**
2549  * igbvf_io_slot_reset - called after the pci bus has been reset.
2550  * @pdev: Pointer to PCI device
2551  *
2552  * Restart the card from scratch, as if from a cold-boot. Implementation
2553  * resembles the first-half of the igbvf_resume routine.
2554  */
2555 static pci_ers_result_t igbvf_io_slot_reset(struct pci_dev *pdev)
2556 {
2557         struct net_device *netdev = pci_get_drvdata(pdev);
2558         struct igbvf_adapter *adapter = netdev_priv(netdev);
2559
2560         if (pci_enable_device_mem(pdev)) {
2561                 dev_err(&pdev->dev,
2562                         "Cannot re-enable PCI device after reset.\n");
2563                 return PCI_ERS_RESULT_DISCONNECT;
2564         }
2565         pci_set_master(pdev);
2566
2567         igbvf_reset(adapter);
2568
2569         return PCI_ERS_RESULT_RECOVERED;
2570 }
2571
2572 /**
2573  * igbvf_io_resume - called when traffic can start flowing again.
2574  * @pdev: Pointer to PCI device
2575  *
2576  * This callback is called when the error recovery driver tells us that
2577  * its OK to resume normal operation. Implementation resembles the
2578  * second-half of the igbvf_resume routine.
2579  */
2580 static void igbvf_io_resume(struct pci_dev *pdev)
2581 {
2582         struct net_device *netdev = pci_get_drvdata(pdev);
2583         struct igbvf_adapter *adapter = netdev_priv(netdev);
2584
2585         if (netif_running(netdev)) {
2586                 if (igbvf_up(adapter)) {
2587                         dev_err(&pdev->dev,
2588                                 "can't bring device back up after reset\n");
2589                         return;
2590                 }
2591         }
2592
2593         netif_device_attach(netdev);
2594 }
2595
2596 static void igbvf_print_device_info(struct igbvf_adapter *adapter)
2597 {
2598         struct e1000_hw *hw = &adapter->hw;
2599         struct net_device *netdev = adapter->netdev;
2600         struct pci_dev *pdev = adapter->pdev;
2601
2602         dev_info(&pdev->dev, "Intel(R) 82576 Virtual Function\n");
2603         dev_info(&pdev->dev, "Address: %pM\n", netdev->dev_addr);
2604         dev_info(&pdev->dev, "MAC: %d\n", hw->mac.type);
2605 }
2606
2607 static const struct net_device_ops igbvf_netdev_ops = {
2608         .ndo_open                       = igbvf_open,
2609         .ndo_stop                       = igbvf_close,
2610         .ndo_start_xmit                 = igbvf_xmit_frame,
2611         .ndo_get_stats                  = igbvf_get_stats,
2612         .ndo_set_multicast_list         = igbvf_set_multi,
2613         .ndo_set_mac_address            = igbvf_set_mac,
2614         .ndo_change_mtu                 = igbvf_change_mtu,
2615         .ndo_do_ioctl                   = igbvf_ioctl,
2616         .ndo_tx_timeout                 = igbvf_tx_timeout,
2617         .ndo_vlan_rx_register           = igbvf_vlan_rx_register,
2618         .ndo_vlan_rx_add_vid            = igbvf_vlan_rx_add_vid,
2619         .ndo_vlan_rx_kill_vid           = igbvf_vlan_rx_kill_vid,
2620 #ifdef CONFIG_NET_POLL_CONTROLLER
2621         .ndo_poll_controller            = igbvf_netpoll,
2622 #endif
2623 };
2624
2625 /**
2626  * igbvf_probe - Device Initialization Routine
2627  * @pdev: PCI device information struct
2628  * @ent: entry in igbvf_pci_tbl
2629  *
2630  * Returns 0 on success, negative on failure
2631  *
2632  * igbvf_probe initializes an adapter identified by a pci_dev structure.
2633  * The OS initialization, configuring of the adapter private structure,
2634  * and a hardware reset occur.
2635  **/
2636 static int __devinit igbvf_probe(struct pci_dev *pdev,
2637                                  const struct pci_device_id *ent)
2638 {
2639         struct net_device *netdev;
2640         struct igbvf_adapter *adapter;
2641         struct e1000_hw *hw;
2642         const struct igbvf_info *ei = igbvf_info_tbl[ent->driver_data];
2643
2644         static int cards_found;
2645         int err, pci_using_dac;
2646
2647         err = pci_enable_device_mem(pdev);
2648         if (err)
2649                 return err;
2650
2651         pci_using_dac = 0;
2652         err = dma_set_mask(&pdev->dev, DMA_BIT_MASK(64));
2653         if (!err) {
2654                 err = dma_set_coherent_mask(&pdev->dev, DMA_BIT_MASK(64));
2655                 if (!err)
2656                         pci_using_dac = 1;
2657         } else {
2658                 err = dma_set_mask(&pdev->dev, DMA_BIT_MASK(32));
2659                 if (err) {
2660                         err = dma_set_coherent_mask(&pdev->dev,
2661                                                     DMA_BIT_MASK(32));
2662                         if (err) {
2663                                 dev_err(&pdev->dev, "No usable DMA "
2664                                         "configuration, aborting\n");
2665                                 goto err_dma;
2666                         }
2667                 }
2668         }
2669
2670         err = pci_request_regions(pdev, igbvf_driver_name);
2671         if (err)
2672                 goto err_pci_reg;
2673
2674         pci_set_master(pdev);
2675
2676         err = -ENOMEM;
2677         netdev = alloc_etherdev(sizeof(struct igbvf_adapter));
2678         if (!netdev)
2679                 goto err_alloc_etherdev;
2680
2681         SET_NETDEV_DEV(netdev, &pdev->dev);
2682
2683         pci_set_drvdata(pdev, netdev);
2684         adapter = netdev_priv(netdev);
2685         hw = &adapter->hw;
2686         adapter->netdev = netdev;
2687         adapter->pdev = pdev;
2688         adapter->ei = ei;
2689         adapter->pba = ei->pba;
2690         adapter->flags = ei->flags;
2691         adapter->hw.back = adapter;
2692         adapter->hw.mac.type = ei->mac;
2693         adapter->msg_enable = (1 << NETIF_MSG_DRV | NETIF_MSG_PROBE) - 1;
2694
2695         /* PCI config space info */
2696
2697         hw->vendor_id = pdev->vendor;
2698         hw->device_id = pdev->device;
2699         hw->subsystem_vendor_id = pdev->subsystem_vendor;
2700         hw->subsystem_device_id = pdev->subsystem_device;
2701
2702         pci_read_config_byte(pdev, PCI_REVISION_ID, &hw->revision_id);
2703
2704         err = -EIO;
2705         adapter->hw.hw_addr = ioremap(pci_resource_start(pdev, 0),
2706                                       pci_resource_len(pdev, 0));
2707
2708         if (!adapter->hw.hw_addr)
2709                 goto err_ioremap;
2710
2711         if (ei->get_variants) {
2712                 err = ei->get_variants(adapter);
2713                 if (err)
2714                         goto err_ioremap;
2715         }
2716
2717         /* setup adapter struct */
2718         err = igbvf_sw_init(adapter);
2719         if (err)
2720                 goto err_sw_init;
2721
2722         /* construct the net_device struct */
2723         netdev->netdev_ops = &igbvf_netdev_ops;
2724
2725         igbvf_set_ethtool_ops(netdev);
2726         netdev->watchdog_timeo = 5 * HZ;
2727         strncpy(netdev->name, pci_name(pdev), sizeof(netdev->name) - 1);
2728
2729         adapter->bd_number = cards_found++;
2730
2731         netdev->features = NETIF_F_SG |
2732                            NETIF_F_IP_CSUM |
2733                            NETIF_F_HW_VLAN_TX |
2734                            NETIF_F_HW_VLAN_RX |
2735                            NETIF_F_HW_VLAN_FILTER;
2736
2737         netdev->features |= NETIF_F_IPV6_CSUM;
2738         netdev->features |= NETIF_F_TSO;
2739         netdev->features |= NETIF_F_TSO6;
2740
2741         if (pci_using_dac)
2742                 netdev->features |= NETIF_F_HIGHDMA;
2743
2744         netdev->vlan_features |= NETIF_F_TSO;
2745         netdev->vlan_features |= NETIF_F_TSO6;
2746         netdev->vlan_features |= NETIF_F_IP_CSUM;
2747         netdev->vlan_features |= NETIF_F_IPV6_CSUM;
2748         netdev->vlan_features |= NETIF_F_SG;
2749
2750         /*reset the controller to put the device in a known good state */
2751         err = hw->mac.ops.reset_hw(hw);
2752         if (err) {
2753                 dev_info(&pdev->dev,
2754                          "PF still in reset state, assigning new address."
2755                          " Is the PF interface up?\n");
2756                 dev_hw_addr_random(adapter->netdev, hw->mac.addr);
2757         } else {
2758                 err = hw->mac.ops.read_mac_addr(hw);
2759                 if (err) {
2760                         dev_err(&pdev->dev, "Error reading MAC address\n");
2761                         goto err_hw_init;
2762                 }
2763         }
2764
2765         memcpy(netdev->dev_addr, adapter->hw.mac.addr, netdev->addr_len);
2766         memcpy(netdev->perm_addr, adapter->hw.mac.addr, netdev->addr_len);
2767
2768         if (!is_valid_ether_addr(netdev->perm_addr)) {
2769                 dev_err(&pdev->dev, "Invalid MAC Address: %pM\n",
2770                         netdev->dev_addr);
2771                 err = -EIO;
2772                 goto err_hw_init;
2773         }
2774
2775         setup_timer(&adapter->watchdog_timer, &igbvf_watchdog,
2776                     (unsigned long) adapter);
2777
2778         INIT_WORK(&adapter->reset_task, igbvf_reset_task);
2779         INIT_WORK(&adapter->watchdog_task, igbvf_watchdog_task);
2780
2781         /* ring size defaults */
2782         adapter->rx_ring->count = 1024;
2783         adapter->tx_ring->count = 1024;
2784
2785         /* reset the hardware with the new settings */
2786         igbvf_reset(adapter);
2787
2788         /* tell the stack to leave us alone until igbvf_open() is called */
2789         netif_carrier_off(netdev);
2790         netif_stop_queue(netdev);
2791
2792         strcpy(netdev->name, "eth%d");
2793         err = register_netdev(netdev);
2794         if (err)
2795                 goto err_hw_init;
2796
2797         igbvf_print_device_info(adapter);
2798
2799         igbvf_initialize_last_counter_stats(adapter);
2800
2801         return 0;
2802
2803 err_hw_init:
2804         kfree(adapter->tx_ring);
2805         kfree(adapter->rx_ring);
2806 err_sw_init:
2807         igbvf_reset_interrupt_capability(adapter);
2808         iounmap(adapter->hw.hw_addr);
2809 err_ioremap:
2810         free_netdev(netdev);
2811 err_alloc_etherdev:
2812         pci_release_regions(pdev);
2813 err_pci_reg:
2814 err_dma:
2815         pci_disable_device(pdev);
2816         return err;
2817 }
2818
2819 /**
2820  * igbvf_remove - Device Removal Routine
2821  * @pdev: PCI device information struct
2822  *
2823  * igbvf_remove is called by the PCI subsystem to alert the driver
2824  * that it should release a PCI device.  The could be caused by a
2825  * Hot-Plug event, or because the driver is going to be removed from
2826  * memory.
2827  **/
2828 static void __devexit igbvf_remove(struct pci_dev *pdev)
2829 {
2830         struct net_device *netdev = pci_get_drvdata(pdev);
2831         struct igbvf_adapter *adapter = netdev_priv(netdev);
2832         struct e1000_hw *hw = &adapter->hw;
2833
2834         /*
2835          * flush_scheduled work may reschedule our watchdog task, so
2836          * explicitly disable watchdog tasks from being rescheduled
2837          */
2838         set_bit(__IGBVF_DOWN, &adapter->state);
2839         del_timer_sync(&adapter->watchdog_timer);
2840
2841         flush_scheduled_work();
2842
2843         unregister_netdev(netdev);
2844
2845         igbvf_reset_interrupt_capability(adapter);
2846
2847         /*
2848          * it is important to delete the napi struct prior to freeing the
2849          * rx ring so that you do not end up with null pointer refs
2850          */
2851         netif_napi_del(&adapter->rx_ring->napi);
2852         kfree(adapter->tx_ring);
2853         kfree(adapter->rx_ring);
2854
2855         iounmap(hw->hw_addr);
2856         if (hw->flash_address)
2857                 iounmap(hw->flash_address);
2858         pci_release_regions(pdev);
2859
2860         free_netdev(netdev);
2861
2862         pci_disable_device(pdev);
2863 }
2864
2865 /* PCI Error Recovery (ERS) */
2866 static struct pci_error_handlers igbvf_err_handler = {
2867         .error_detected = igbvf_io_error_detected,
2868         .slot_reset = igbvf_io_slot_reset,
2869         .resume = igbvf_io_resume,
2870 };
2871
2872 static DEFINE_PCI_DEVICE_TABLE(igbvf_pci_tbl) = {
2873         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82576_VF), board_vf },
2874         { } /* terminate list */
2875 };
2876 MODULE_DEVICE_TABLE(pci, igbvf_pci_tbl);
2877
2878 /* PCI Device API Driver */
2879 static struct pci_driver igbvf_driver = {
2880         .name     = igbvf_driver_name,
2881         .id_table = igbvf_pci_tbl,
2882         .probe    = igbvf_probe,
2883         .remove   = __devexit_p(igbvf_remove),
2884 #ifdef CONFIG_PM
2885         /* Power Management Hooks */
2886         .suspend  = igbvf_suspend,
2887         .resume   = igbvf_resume,
2888 #endif
2889         .shutdown = igbvf_shutdown,
2890         .err_handler = &igbvf_err_handler
2891 };
2892
2893 /**
2894  * igbvf_init_module - Driver Registration Routine
2895  *
2896  * igbvf_init_module is the first routine called when the driver is
2897  * loaded. All it does is register with the PCI subsystem.
2898  **/
2899 static int __init igbvf_init_module(void)
2900 {
2901         int ret;
2902         printk(KERN_INFO "%s - version %s\n",
2903                igbvf_driver_string, igbvf_driver_version);
2904         printk(KERN_INFO "%s\n", igbvf_copyright);
2905
2906         ret = pci_register_driver(&igbvf_driver);
2907         pm_qos_add_request(&igbvf_driver_pm_qos_req, PM_QOS_CPU_DMA_LATENCY,
2908                            PM_QOS_DEFAULT_VALUE);
2909
2910         return ret;
2911 }
2912 module_init(igbvf_init_module);
2913
2914 /**
2915  * igbvf_exit_module - Driver Exit Cleanup Routine
2916  *
2917  * igbvf_exit_module is called just before the driver is removed
2918  * from memory.
2919  **/
2920 static void __exit igbvf_exit_module(void)
2921 {
2922         pci_unregister_driver(&igbvf_driver);
2923         pm_qos_remove_request(&igbvf_driver_pm_qos_req);
2924 }
2925 module_exit(igbvf_exit_module);
2926
2927
2928 MODULE_AUTHOR("Intel Corporation, <e1000-devel@lists.sourceforge.net>");
2929 MODULE_DESCRIPTION("Intel(R) 82576 Virtual Function Network Driver");
2930 MODULE_LICENSE("GPL");
2931 MODULE_VERSION(DRV_VERSION);
2932
2933 /* netdev.c */