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 const char igbvf_driver_string[] =
  52                                 "Intel(R) Virtual Function Network Driver";
  53 static const char igbvf_copyright[] = "Copyright (c) 2009 Intel Corporation.";
  54
  55 static int igbvf_poll(struct napi_struct *napi, int budget);
  56 static void igbvf_reset(struct igbvf_adapter *);
  57 static void igbvf_set_interrupt_capability(struct igbvf_adapter *);
  58 static void igbvf_reset_interrupt_capability(struct igbvf_adapter *);
  59
  60 static struct igbvf_info igbvf_vf_info = {
  61         .mac                    = e1000_vfadapt,
  62         .flags                  = 0,
  63         .pba                    = 10,
  64         .init_ops               = e1000_init_function_pointers_vf,
  65 };
  66
  67 static const struct igbvf_info *igbvf_info_tbl[] = {
  68         [board_vf]              = &igbvf_vf_info,
  69 };
  70
  71 /**
  72  * igbvf_desc_unused - calculate if we have unused descriptors
  73  **/
  74 static int igbvf_desc_unused(struct igbvf_ring *ring)
  75 {
  76         if (ring->next_to_clean > ring->next_to_use)
  77                 return ring->next_to_clean - ring->next_to_use - 1;
  78
  79         return ring->count + ring->next_to_clean - ring->next_to_use - 1;
  80 }
  81
  82 /**
  83  * igbvf_receive_skb - helper function to handle Rx indications
  84  * @adapter: board private structure
  85  * @status: descriptor status field as written by hardware
  86  * @vlan: descriptor vlan field as written by hardware (no le/be conversion)
  87  * @skb: pointer to sk_buff to be indicated to stack
  88  **/
  89 static void igbvf_receive_skb(struct igbvf_adapter *adapter,
  90                               struct net_device *netdev,
  91                               struct sk_buff *skb,
  92                               u32 status, u16 vlan)
  93 {
  94         if (adapter->vlgrp && (status & E1000_RXD_STAT_VP))
  95                 vlan_hwaccel_receive_skb(skb, adapter->vlgrp,
  96                                          le16_to_cpu(vlan) &
  97                                          E1000_RXD_SPC_VLAN_MASK);
  98         else
  99                 netif_receive_skb(skb);
 100 }
 101
 102 static inline void igbvf_rx_checksum_adv(struct igbvf_adapter *adapter,
 103                                          u32 status_err, struct sk_buff *skb)
 104 {
 105         skb->ip_summed = CHECKSUM_NONE;
 106
 107         /* Ignore Checksum bit is set or checksum is disabled through ethtool */
 108         if ((status_err & E1000_RXD_STAT_IXSM) ||
 109             (adapter->flags & IGBVF_FLAG_RX_CSUM_DISABLED))
 110                 return;
 111
 112         /* TCP/UDP checksum error bit is set */
 113         if (status_err &
 114             (E1000_RXDEXT_STATERR_TCPE | E1000_RXDEXT_STATERR_IPE)) {
 115                 /* let the stack verify checksum errors */
 116                 adapter->hw_csum_err++;
 117                 return;
 118         }
 119
 120         /* It must be a TCP or UDP packet with a valid checksum */
 121         if (status_err & (E1000_RXD_STAT_TCPCS | E1000_RXD_STAT_UDPCS))
 122                 skb->ip_summed = CHECKSUM_UNNECESSARY;
 123
 124         adapter->hw_csum_good++;
 125 }
 126
 127 /**
 128  * igbvf_alloc_rx_buffers - Replace used receive buffers; packet split
 129  * @rx_ring: address of ring structure to repopulate
 130  * @cleaned_count: number of buffers to repopulate
 131  **/
 132 static void igbvf_alloc_rx_buffers(struct igbvf_ring *rx_ring,
 133                                    int cleaned_count)
 134 {
 135         struct igbvf_adapter *adapter = rx_ring->adapter;
 136         struct net_device *netdev = adapter->netdev;
 137         struct pci_dev *pdev = adapter->pdev;
 138         union e1000_adv_rx_desc *rx_desc;
 139         struct igbvf_buffer *buffer_info;
 140         struct sk_buff *skb;
 141         unsigned int i;
 142         int bufsz;
 143
 144         i = rx_ring->next_to_use;
 145         buffer_info = &rx_ring->buffer_info[i];
 146
 147         if (adapter->rx_ps_hdr_size)
 148                 bufsz = adapter->rx_ps_hdr_size;
 149         else
 150                 bufsz = adapter->rx_buffer_len;
 151
 152         while (cleaned_count--) {
 153                 rx_desc = IGBVF_RX_DESC_ADV(*rx_ring, i);
 154
 155                 if (adapter->rx_ps_hdr_size && !buffer_info->page_dma) {
 156                         if (!buffer_info->page) {
 157                                 buffer_info->page = alloc_page(GFP_ATOMIC);
 158                                 if (!buffer_info->page) {
 159                                         adapter->alloc_rx_buff_failed++;
 160                                         goto no_buffers;
 161                                 }
 162                                 buffer_info->page_offset = 0;
 163                         } else {
 164                                 buffer_info->page_offset ^= PAGE_SIZE / 2;
 165                         }
 166                         buffer_info->page_dma =
 167                                 pci_map_page(pdev, buffer_info->page,
 168                                              buffer_info->page_offset,
 169                                              PAGE_SIZE / 2,
 170                                              PCI_DMA_FROMDEVICE);
 171                 }
 172
 173                 if (!buffer_info->skb) {
 174                         skb = netdev_alloc_skb_ip_align(netdev, bufsz);
 175                         if (!skb) {
 176                                 adapter->alloc_rx_buff_failed++;
 177                                 goto no_buffers;
 178                         }
 179
 180                         buffer_info->skb = skb;
 181                         buffer_info->dma = pci_map_single(pdev, skb->data,
 182                                                           bufsz,
 183                                                           PCI_DMA_FROMDEVICE);
 184                 }
 185                 /* Refresh the desc even if buffer_addrs didn't change because
 186                  * each write-back erases this info. */
 187                 if (adapter->rx_ps_hdr_size) {
 188                         rx_desc->read.pkt_addr =
 189                              cpu_to_le64(buffer_info->page_dma);
 190                         rx_desc->read.hdr_addr = cpu_to_le64(buffer_info->dma);
 191                 } else {
 192                         rx_desc->read.pkt_addr =
 193                              cpu_to_le64(buffer_info->dma);
 194                         rx_desc->read.hdr_addr = 0;
 195                 }
 196
 197                 i++;
 198                 if (i == rx_ring->count)
 199                         i = 0;
 200                 buffer_info = &rx_ring->buffer_info[i];
 201         }
 202
 203 no_buffers:
 204         if (rx_ring->next_to_use != i) {
 205                 rx_ring->next_to_use = i;
 206                 if (i == 0)
 207                         i = (rx_ring->count - 1);
 208                 else
 209                         i--;
 210
 211                 /* Force memory writes to complete before letting h/w
 212                  * know there are new descriptors to fetch.  (Only
 213                  * applicable for weak-ordered memory model archs,
 214                  * such as IA-64). */
 215                 wmb();
 216                 writel(i, adapter->hw.hw_addr + rx_ring->tail);
 217         }
 218 }
 219
 220 /**
 221  * igbvf_clean_rx_irq - Send received data up the network stack; legacy
 222  * @adapter: board private structure
 223  *
 224  * the return value indicates whether actual cleaning was done, there
 225  * is no guarantee that everything was cleaned
 226  **/
 227 static bool igbvf_clean_rx_irq(struct igbvf_adapter *adapter,
 228                                int *work_done, int work_to_do)
 229 {
 230         struct igbvf_ring *rx_ring = adapter->rx_ring;
 231         struct net_device *netdev = adapter->netdev;
 232         struct pci_dev *pdev = adapter->pdev;
 233         union e1000_adv_rx_desc *rx_desc, *next_rxd;
 234         struct igbvf_buffer *buffer_info, *next_buffer;
 235         struct sk_buff *skb;
 236         bool cleaned = false;
 237         int cleaned_count = 0;
 238         unsigned int total_bytes = 0, total_packets = 0;
 239         unsigned int i;
 240         u32 length, hlen, staterr;
 241
 242         i = rx_ring->next_to_clean;
 243         rx_desc = IGBVF_RX_DESC_ADV(*rx_ring, i);
 244         staterr = le32_to_cpu(rx_desc->wb.upper.status_error);
 245
 246         while (staterr & E1000_RXD_STAT_DD) {
 247                 if (*work_done >= work_to_do)
 248                         break;
 249                 (*work_done)++;
 250
 251                 buffer_info = &rx_ring->buffer_info[i];
 252
 253                 /* HW will not DMA in data larger than the given buffer, even
 254                  * if it parses the (NFS, of course) header to be larger.  In
 255                  * that case, it fills the header buffer and spills the rest
 256                  * into the page.
 257                  */
 258                 hlen = (le16_to_cpu(rx_desc->wb.lower.lo_dword.hs_rss.hdr_info) &
 259                   E1000_RXDADV_HDRBUFLEN_MASK) >> E1000_RXDADV_HDRBUFLEN_SHIFT;
 260                 if (hlen > adapter->rx_ps_hdr_size)
 261                         hlen = adapter->rx_ps_hdr_size;
 262
 263                 length = le16_to_cpu(rx_desc->wb.upper.length);
 264                 cleaned = true;
 265                 cleaned_count++;
 266
 267                 skb = buffer_info->skb;
 268                 prefetch(skb->data - NET_IP_ALIGN);
 269                 buffer_info->skb = NULL;
 270                 if (!adapter->rx_ps_hdr_size) {
 271                         pci_unmap_single(pdev, buffer_info->dma,
 272                                          adapter->rx_buffer_len,
 273                                          PCI_DMA_FROMDEVICE);
 274                         buffer_info->dma = 0;
 275                         skb_put(skb, length);
 276                         goto send_up;
 277                 }
 278
 279                 if (!skb_shinfo(skb)->nr_frags) {
 280                         pci_unmap_single(pdev, buffer_info->dma,
 281                                          adapter->rx_ps_hdr_size,
 282                                          PCI_DMA_FROMDEVICE);
 283                         skb_put(skb, hlen);
 284                 }
 285
 286                 if (length) {
 287                         pci_unmap_page(pdev, buffer_info->page_dma,
 288                                        PAGE_SIZE / 2,
 289                                        PCI_DMA_FROMDEVICE);
 290                         buffer_info->page_dma = 0;
 291
 292                         skb_fill_page_desc(skb, skb_shinfo(skb)->nr_frags++,
 293                                            buffer_info->page,
 294                                            buffer_info->page_offset,
 295                                            length);
 296
 297                         if ((adapter->rx_buffer_len > (PAGE_SIZE / 2)) ||
 298                             (page_count(buffer_info->page) != 1))
 299                                 buffer_info->page = NULL;
 300                         else
 301                                 get_page(buffer_info->page);
 302
 303                         skb->len += length;
 304                         skb->data_len += length;
 305                         skb->truesize += length;
 306                 }
 307 send_up:
 308                 i++;
 309                 if (i == rx_ring->count)
 310                         i = 0;
 311                 next_rxd = IGBVF_RX_DESC_ADV(*rx_ring, i);
 312                 prefetch(next_rxd);
 313                 next_buffer = &rx_ring->buffer_info[i];
 314
 315                 if (!(staterr & E1000_RXD_STAT_EOP)) {
 316                         buffer_info->skb = next_buffer->skb;
 317                         buffer_info->dma = next_buffer->dma;
 318                         next_buffer->skb = skb;
 319                         next_buffer->dma = 0;
 320                         goto next_desc;
 321                 }
 322
 323                 if (staterr & E1000_RXDEXT_ERR_FRAME_ERR_MASK) {
 324                         dev_kfree_skb_irq(skb);
 325                         goto next_desc;
 326                 }
 327
 328                 total_bytes += skb->len;
 329                 total_packets++;
 330
 331                 igbvf_rx_checksum_adv(adapter, staterr, skb);
 332
 333                 skb->protocol = eth_type_trans(skb, netdev);
 334
 335                 igbvf_receive_skb(adapter, netdev, skb, staterr,
 336                                   rx_desc->wb.upper.vlan);
 337
 338 next_desc:
 339                 rx_desc->wb.upper.status_error = 0;
 340
 341                 /* return some buffers to hardware, one at a time is too slow */
 342                 if (cleaned_count >= IGBVF_RX_BUFFER_WRITE) {
 343                         igbvf_alloc_rx_buffers(rx_ring, cleaned_count);
 344                         cleaned_count = 0;
 345                 }
 346
 347                 /* use prefetched values */
 348                 rx_desc = next_rxd;
 349                 buffer_info = next_buffer;
 350
 351                 staterr = le32_to_cpu(rx_desc->wb.upper.status_error);
 352         }
 353
 354         rx_ring->next_to_clean = i;
 355         cleaned_count = igbvf_desc_unused(rx_ring);
 356
 357         if (cleaned_count)
 358                 igbvf_alloc_rx_buffers(rx_ring, cleaned_count);
 359
 360         adapter->total_rx_packets += total_packets;
 361         adapter->total_rx_bytes += total_bytes;
 362         adapter->net_stats.rx_bytes += total_bytes;
 363         adapter->net_stats.rx_packets += total_packets;
 364         return cleaned;
 365 }
 366
 367 static void igbvf_put_txbuf(struct igbvf_adapter *adapter,
 368                             struct igbvf_buffer *buffer_info)
 369 {
 370         if (buffer_info->dma) {
 371                 if (buffer_info->mapped_as_page)
 372                         pci_unmap_page(adapter->pdev,
 373                                        buffer_info->dma,
 374                                        buffer_info->length,
 375                                        PCI_DMA_TODEVICE);
 376                 else
 377                         pci_unmap_single(adapter->pdev,
 378                                          buffer_info->dma,
 379                                          buffer_info->length,
 380                                          PCI_DMA_TODEVICE);
 381                 buffer_info->dma = 0;
 382         }
 383         if (buffer_info->skb) {
 384                 dev_kfree_skb_any(buffer_info->skb);
 385                 buffer_info->skb = NULL;
 386         }
 387         buffer_info->time_stamp = 0;
 388 }
 389
 390 static void igbvf_print_tx_hang(struct igbvf_adapter *adapter)
 391 {
 392         struct igbvf_ring *tx_ring = adapter->tx_ring;
 393         unsigned int i = tx_ring->next_to_clean;
 394         unsigned int eop = tx_ring->buffer_info[i].next_to_watch;
 395         union e1000_adv_tx_desc *eop_desc = IGBVF_TX_DESC_ADV(*tx_ring, eop);
 396
 397         /* detected Tx unit hang */
 398         dev_err(&adapter->pdev->dev,
 399                 "Detected Tx Unit Hang:\n"
 400                 "  TDH                  <%x>\n"
 401                 "  TDT                  <%x>\n"
 402                 "  next_to_use          <%x>\n"
 403                 "  next_to_clean        <%x>\n"
 404                 "buffer_info[next_to_clean]:\n"
 405                 "  time_stamp           <%lx>\n"
 406                 "  next_to_watch        <%x>\n"
 407                 "  jiffies              <%lx>\n"
 408                 "  next_to_watch.status <%x>\n",
 409                 readl(adapter->hw.hw_addr + tx_ring->head),
 410                 readl(adapter->hw.hw_addr + tx_ring->tail),
 411                 tx_ring->next_to_use,
 412                 tx_ring->next_to_clean,
 413                 tx_ring->buffer_info[eop].time_stamp,
 414                 eop,
 415                 jiffies,
 416                 eop_desc->wb.status);
 417 }
 418
 419 /**
 420  * igbvf_setup_tx_resources - allocate Tx resources (Descriptors)
 421  * @adapter: board private structure
 422  *
 423  * Return 0 on success, negative on failure
 424  **/
 425 int igbvf_setup_tx_resources(struct igbvf_adapter *adapter,
 426                              struct igbvf_ring *tx_ring)
 427 {
 428         struct pci_dev *pdev = adapter->pdev;
 429         int size;
 430
 431         size = sizeof(struct igbvf_buffer) * tx_ring->count;
 432         tx_ring->buffer_info = vmalloc(size);
 433         if (!tx_ring->buffer_info)
 434                 goto err;
 435         memset(tx_ring->buffer_info, 0, size);
 436
 437         /* round up to nearest 4K */
 438         tx_ring->size = tx_ring->count * sizeof(union e1000_adv_tx_desc);
 439         tx_ring->size = ALIGN(tx_ring->size, 4096);
 440
 441         tx_ring->desc = pci_alloc_consistent(pdev, tx_ring->size,
 442                                              &tx_ring->dma);
 443
 444         if (!tx_ring->desc)
 445                 goto err;
 446
 447         tx_ring->adapter = adapter;
 448         tx_ring->next_to_use = 0;
 449         tx_ring->next_to_clean = 0;
 450
 451         return 0;
 452 err:
 453         vfree(tx_ring->buffer_info);
 454         dev_err(&adapter->pdev->dev,
 455                 "Unable to allocate memory for the transmit descriptor ring\n");
 456         return -ENOMEM;
 457 }
 458
 459 /**
 460  * igbvf_setup_rx_resources - allocate Rx resources (Descriptors)
 461  * @adapter: board private structure
 462  *
 463  * Returns 0 on success, negative on failure
 464  **/
 465 int igbvf_setup_rx_resources(struct igbvf_adapter *adapter,
 466                              struct igbvf_ring *rx_ring)
 467 {
 468         struct pci_dev *pdev = adapter->pdev;
 469         int size, desc_len;
 470
 471         size = sizeof(struct igbvf_buffer) * rx_ring->count;
 472         rx_ring->buffer_info = vmalloc(size);
 473         if (!rx_ring->buffer_info)
 474                 goto err;
 475         memset(rx_ring->buffer_info, 0, size);
 476
 477         desc_len = sizeof(union e1000_adv_rx_desc);
 478
 479         /* Round up to nearest 4K */
 480         rx_ring->size = rx_ring->count * desc_len;
 481         rx_ring->size = ALIGN(rx_ring->size, 4096);
 482
 483         rx_ring->desc = pci_alloc_consistent(pdev, rx_ring->size,
 484                                              &rx_ring->dma);
 485
 486         if (!rx_ring->desc)
 487                 goto err;
 488
 489         rx_ring->next_to_clean = 0;
 490         rx_ring->next_to_use = 0;
 491
 492         rx_ring->adapter = adapter;
 493
 494         return 0;
 495
 496 err:
 497         vfree(rx_ring->buffer_info);
 498         rx_ring->buffer_info = NULL;
 499         dev_err(&adapter->pdev->dev,
 500                 "Unable to allocate memory for the receive descriptor ring\n");
 501         return -ENOMEM;
 502 }
 503
 504 /**
 505  * igbvf_clean_tx_ring - Free Tx Buffers
 506  * @tx_ring: ring to be cleaned
 507  **/
 508 static void igbvf_clean_tx_ring(struct igbvf_ring *tx_ring)
 509 {
 510         struct igbvf_adapter *adapter = tx_ring->adapter;
 511         struct igbvf_buffer *buffer_info;
 512         unsigned long size;
 513         unsigned int i;
 514
 515         if (!tx_ring->buffer_info)
 516                 return;
 517
 518         /* Free all the Tx ring sk_buffs */
 519         for (i = 0; i < tx_ring->count; i++) {
 520                 buffer_info = &tx_ring->buffer_info[i];
 521                 igbvf_put_txbuf(adapter, buffer_info);
 522         }
 523
 524         size = sizeof(struct igbvf_buffer) * tx_ring->count;
 525         memset(tx_ring->buffer_info, 0, size);
 526
 527         /* Zero out the descriptor ring */
 528         memset(tx_ring->desc, 0, tx_ring->size);
 529
 530         tx_ring->next_to_use = 0;
 531         tx_ring->next_to_clean = 0;
 532
 533         writel(0, adapter->hw.hw_addr + tx_ring->head);
 534         writel(0, adapter->hw.hw_addr + tx_ring->tail);
 535 }
 536
 537 /**
 538  * igbvf_free_tx_resources - Free Tx Resources per Queue
 539  * @tx_ring: ring to free resources from
 540  *
 541  * Free all transmit software resources
 542  **/
 543 void igbvf_free_tx_resources(struct igbvf_ring *tx_ring)
 544 {
 545         struct pci_dev *pdev = tx_ring->adapter->pdev;
 546
 547         igbvf_clean_tx_ring(tx_ring);
 548
 549         vfree(tx_ring->buffer_info);
 550         tx_ring->buffer_info = NULL;
 551
 552         pci_free_consistent(pdev, tx_ring->size, tx_ring->desc, tx_ring->dma);
 553
 554         tx_ring->desc = NULL;
 555 }
 556
 557 /**
 558  * igbvf_clean_rx_ring - Free Rx Buffers per Queue
 559  * @adapter: board private structure
 560  **/
 561 static void igbvf_clean_rx_ring(struct igbvf_ring *rx_ring)
 562 {
 563         struct igbvf_adapter *adapter = rx_ring->adapter;
 564         struct igbvf_buffer *buffer_info;
 565         struct pci_dev *pdev = adapter->pdev;
 566         unsigned long size;
 567         unsigned int i;
 568
 569         if (!rx_ring->buffer_info)
 570                 return;
 571
 572         /* Free all the Rx ring sk_buffs */
 573         for (i = 0; i < rx_ring->count; i++) {
 574                 buffer_info = &rx_ring->buffer_info[i];
 575                 if (buffer_info->dma) {
 576                         if (adapter->rx_ps_hdr_size){
 577                                 pci_unmap_single(pdev, buffer_info->dma,
 578                                                  adapter->rx_ps_hdr_size,
 579                                                  PCI_DMA_FROMDEVICE);
 580                         } else {
 581                                 pci_unmap_single(pdev, buffer_info->dma,
 582                                                  adapter->rx_buffer_len,
 583                                                  PCI_DMA_FROMDEVICE);
 584                         }
 585                         buffer_info->dma = 0;
 586                 }
 587
 588                 if (buffer_info->skb) {
 589                         dev_kfree_skb(buffer_info->skb);
 590                         buffer_info->skb = NULL;
 591                 }
 592
 593                 if (buffer_info->page) {
 594                         if (buffer_info->page_dma)
 595                                 pci_unmap_page(pdev, buffer_info->page_dma,
 596                                                PAGE_SIZE / 2,
 597                                                PCI_DMA_FROMDEVICE);
 598                         put_page(buffer_info->page);
 599                         buffer_info->page = NULL;
 600                         buffer_info->page_dma = 0;
 601                         buffer_info->page_offset = 0;
 602                 }
 603         }
 604
 605         size = sizeof(struct igbvf_buffer) * rx_ring->count;
 606         memset(rx_ring->buffer_info, 0, size);
 607
 608         /* Zero out the descriptor ring */
 609         memset(rx_ring->desc, 0, rx_ring->size);
 610
 611         rx_ring->next_to_clean = 0;
 612         rx_ring->next_to_use = 0;
 613
 614         writel(0, adapter->hw.hw_addr + rx_ring->head);
 615         writel(0, adapter->hw.hw_addr + rx_ring->tail);
 616 }
 617
 618 /**
 619  * igbvf_free_rx_resources - Free Rx Resources
 620  * @rx_ring: ring to clean the resources from
 621  *
 622  * Free all receive software resources
 623  **/
 624
 625 void igbvf_free_rx_resources(struct igbvf_ring *rx_ring)
 626 {
 627         struct pci_dev *pdev = rx_ring->adapter->pdev;
 628
 629         igbvf_clean_rx_ring(rx_ring);
 630
 631         vfree(rx_ring->buffer_info);
 632         rx_ring->buffer_info = NULL;
 633
 634         dma_free_coherent(&pdev->dev, rx_ring->size, rx_ring->desc,
 635                           rx_ring->dma);
 636         rx_ring->desc = NULL;
 637 }
 638
 639 /**
 640  * igbvf_update_itr - update the dynamic ITR value based on statistics
 641  * @adapter: pointer to adapter
 642  * @itr_setting: current adapter->itr
 643  * @packets: the number of packets during this measurement interval
 644  * @bytes: the number of bytes during this measurement interval
 645  *
 646  *      Stores a new ITR value based on packets and byte
 647  *      counts during the last interrupt.  The advantage of per interrupt
 648  *      computation is faster updates and more accurate ITR for the current
 649  *      traffic pattern.  Constants in this function were computed
 650  *      based on theoretical maximum wire speed and thresholds were set based
 651  *      on testing data as well as attempting to minimize response time
 652  *      while increasing bulk throughput.  This functionality is controlled
 653  *      by the InterruptThrottleRate module parameter.
 654  **/
 655 static unsigned int igbvf_update_itr(struct igbvf_adapter *adapter,
 656                                      u16 itr_setting, int packets,
 657                                      int bytes)
 658 {
 659         unsigned int retval = itr_setting;
 660
 661         if (packets == 0)
 662                 goto update_itr_done;
 663
 664         switch (itr_setting) {
 665         case lowest_latency:
 666                 /* handle TSO and jumbo frames */
 667                 if (bytes/packets > 8000)
 668                         retval = bulk_latency;
 669                 else if ((packets < 5) && (bytes > 512))
 670                         retval = low_latency;
 671                 break;
 672         case low_latency:  /* 50 usec aka 20000 ints/s */
 673                 if (bytes > 10000) {
 674                         /* this if handles the TSO accounting */
 675                         if (bytes/packets > 8000)
 676                                 retval = bulk_latency;
 677                         else if ((packets < 10) || ((bytes/packets) > 1200))
 678                                 retval = bulk_latency;
 679                         else if ((packets > 35))
 680                                 retval = lowest_latency;
 681                 } else if (bytes/packets > 2000) {
 682                         retval = bulk_latency;
 683                 } else if (packets <= 2 && bytes < 512) {
 684                         retval = lowest_latency;
 685                 }
 686                 break;
 687         case bulk_latency: /* 250 usec aka 4000 ints/s */
 688                 if (bytes > 25000) {
 689                         if (packets > 35)
 690                                 retval = low_latency;
 691                 } else if (bytes < 6000) {
 692                         retval = low_latency;
 693                 }
 694                 break;
 695         }
 696
 697 update_itr_done:
 698         return retval;
 699 }
 700
 701 static void igbvf_set_itr(struct igbvf_adapter *adapter)
 702 {
 703         struct e1000_hw *hw = &adapter->hw;
 704         u16 current_itr;
 705         u32 new_itr = adapter->itr;
 706
 707         adapter->tx_itr = igbvf_update_itr(adapter, adapter->tx_itr,
 708                                            adapter->total_tx_packets,
 709                                            adapter->total_tx_bytes);
 710         /* conservative mode (itr 3) eliminates the lowest_latency setting */
 711         if (adapter->itr_setting == 3 && adapter->tx_itr == lowest_latency)
 712                 adapter->tx_itr = low_latency;
 713
 714         adapter->rx_itr = igbvf_update_itr(adapter, adapter->rx_itr,
 715                                            adapter->total_rx_packets,
 716                                            adapter->total_rx_bytes);
 717         /* conservative mode (itr 3) eliminates the lowest_latency setting */
 718         if (adapter->itr_setting == 3 && adapter->rx_itr == lowest_latency)
 719                 adapter->rx_itr = low_latency;
 720
 721         current_itr = max(adapter->rx_itr, adapter->tx_itr);
 722
 723         switch (current_itr) {
 724         /* counts and packets in update_itr are dependent on these numbers */
 725         case lowest_latency:
 726                 new_itr = 70000;
 727                 break;
 728         case low_latency:
 729                 new_itr = 20000; /* aka hwitr = ~200 */
 730                 break;
 731         case bulk_latency:
 732                 new_itr = 4000;
 733                 break;
 734         default:
 735                 break;
 736         }
 737
 738         if (new_itr != adapter->itr) {
 739                 /*
 740                  * this attempts to bias the interrupt rate towards Bulk
 741                  * by adding intermediate steps when interrupt rate is
 742                  * increasing
 743                  */
 744                 new_itr = new_itr > adapter->itr ?
 745                              min(adapter->itr + (new_itr >> 2), new_itr) :
 746                              new_itr;
 747                 adapter->itr = new_itr;
 748                 adapter->rx_ring->itr_val = 1952;
 749
 750                 if (adapter->msix_entries)
 751                         adapter->rx_ring->set_itr = 1;
 752                 else
 753                         ew32(ITR, 1952);
 754         }
 755 }
 756
 757 /**
 758  * igbvf_clean_tx_irq - Reclaim resources after transmit completes
 759  * @adapter: board private structure
 760  * returns true if ring is completely cleaned
 761  **/
 762 static bool igbvf_clean_tx_irq(struct igbvf_ring *tx_ring)
 763 {
 764         struct igbvf_adapter *adapter = tx_ring->adapter;
 765         struct e1000_hw *hw = &adapter->hw;
 766         struct net_device *netdev = adapter->netdev;
 767         struct igbvf_buffer *buffer_info;
 768         struct sk_buff *skb;
 769         union e1000_adv_tx_desc *tx_desc, *eop_desc;
 770         unsigned int total_bytes = 0, total_packets = 0;
 771         unsigned int i, eop, count = 0;
 772         bool cleaned = false;
 773
 774         i = tx_ring->next_to_clean;
 775         eop = tx_ring->buffer_info[i].next_to_watch;
 776         eop_desc = IGBVF_TX_DESC_ADV(*tx_ring, eop);
 777
 778         while ((eop_desc->wb.status & cpu_to_le32(E1000_TXD_STAT_DD)) &&
 779                (count < tx_ring->count)) {
 780                 for (cleaned = false; !cleaned; count++) {
 781                         tx_desc = IGBVF_TX_DESC_ADV(*tx_ring, i);
 782                         buffer_info = &tx_ring->buffer_info[i];
 783                         cleaned = (i == eop);
 784                         skb = buffer_info->skb;
 785
 786                         if (skb) {
 787                                 unsigned int segs, bytecount;
 788
 789                                 /* gso_segs is currently only valid for tcp */
 790                                 segs = skb_shinfo(skb)->gso_segs ?: 1;
 791                                 /* multiply data chunks by size of headers */
 792                                 bytecount = ((segs - 1) * skb_headlen(skb)) +
 793                                             skb->len;
 794                                 total_packets += segs;
 795                                 total_bytes += bytecount;
 796                         }
 797
 798                         igbvf_put_txbuf(adapter, buffer_info);
 799                         tx_desc->wb.status = 0;
 800
 801                         i++;
 802                         if (i == tx_ring->count)
 803                                 i = 0;
 804                 }
 805                 eop = tx_ring->buffer_info[i].next_to_watch;
 806                 eop_desc = IGBVF_TX_DESC_ADV(*tx_ring, eop);
 807         }
 808
 809         tx_ring->next_to_clean = i;
 810
 811         if (unlikely(count &&
 812                      netif_carrier_ok(netdev) &&
 813                      igbvf_desc_unused(tx_ring) >= IGBVF_TX_QUEUE_WAKE)) {
 814                 /* Make sure that anybody stopping the queue after this
 815                  * sees the new next_to_clean.
 816                  */
 817                 smp_mb();
 818                 if (netif_queue_stopped(netdev) &&
 819                     !(test_bit(__IGBVF_DOWN, &adapter->state))) {
 820                         netif_wake_queue(netdev);
 821                         ++adapter->restart_queue;
 822                 }
 823         }
 824
 825         if (adapter->detect_tx_hung) {
 826                 /* Detect a transmit hang in hardware, this serializes the
 827                  * check with the clearing of time_stamp and movement of i */
 828                 adapter->detect_tx_hung = false;
 829                 if (tx_ring->buffer_info[i].time_stamp &&
 830                     time_after(jiffies, tx_ring->buffer_info[i].time_stamp +
 831                                (adapter->tx_timeout_factor * HZ)) &&
 832                     !(er32(STATUS) & E1000_STATUS_TXOFF)) {
 833
 834                         tx_desc = IGBVF_TX_DESC_ADV(*tx_ring, i);
 835                         /* detected Tx unit hang */
 836                         igbvf_print_tx_hang(adapter);
 837
 838                         netif_stop_queue(netdev);
 839                 }
 840         }
 841         adapter->net_stats.tx_bytes += total_bytes;
 842         adapter->net_stats.tx_packets += total_packets;
 843         return (count < tx_ring->count);
 844 }
 845
 846 static irqreturn_t igbvf_msix_other(int irq, void *data)
 847 {
 848         struct net_device *netdev = data;
 849         struct igbvf_adapter *adapter = netdev_priv(netdev);
 850         struct e1000_hw *hw = &adapter->hw;
 851
 852         adapter->int_counter1++;
 853
 854         netif_carrier_off(netdev);
 855         hw->mac.get_link_status = 1;
 856         if (!test_bit(__IGBVF_DOWN, &adapter->state))
 857                 mod_timer(&adapter->watchdog_timer, jiffies + 1);
 858
 859         ew32(EIMS, adapter->eims_other);
 860
 861         return IRQ_HANDLED;
 862 }
 863
 864 static irqreturn_t igbvf_intr_msix_tx(int irq, void *data)
 865 {
 866         struct net_device *netdev = data;
 867         struct igbvf_adapter *adapter = netdev_priv(netdev);
 868         struct e1000_hw *hw = &adapter->hw;
 869         struct igbvf_ring *tx_ring = adapter->tx_ring;
 870
 871
 872         adapter->total_tx_bytes = 0;
 873         adapter->total_tx_packets = 0;
 874
 875         /* auto mask will automatically reenable the interrupt when we write
 876          * EICS */
 877         if (!igbvf_clean_tx_irq(tx_ring))
 878                 /* Ring was not completely cleaned, so fire another interrupt */
 879                 ew32(EICS, tx_ring->eims_value);
 880         else
 881                 ew32(EIMS, tx_ring->eims_value);
 882
 883         return IRQ_HANDLED;
 884 }
 885
 886 static irqreturn_t igbvf_intr_msix_rx(int irq, void *data)
 887 {
 888         struct net_device *netdev = data;
 889         struct igbvf_adapter *adapter = netdev_priv(netdev);
 890
 891         adapter->int_counter0++;
 892
 893         /* Write the ITR value calculated at the end of the
 894          * previous interrupt.
 895          */
 896         if (adapter->rx_ring->set_itr) {
 897                 writel(adapter->rx_ring->itr_val,
 898                        adapter->hw.hw_addr + adapter->rx_ring->itr_register);
 899                 adapter->rx_ring->set_itr = 0;
 900         }
 901
 902         if (napi_schedule_prep(&adapter->rx_ring->napi)) {
 903                 adapter->total_rx_bytes = 0;
 904                 adapter->total_rx_packets = 0;
 905                 __napi_schedule(&adapter->rx_ring->napi);
 906         }
 907
 908         return IRQ_HANDLED;
 909 }
 910
 911 #define IGBVF_NO_QUEUE -1
 912
 913 static void igbvf_assign_vector(struct igbvf_adapter *adapter, int rx_queue,
 914                                 int tx_queue, int msix_vector)
 915 {
 916         struct e1000_hw *hw = &adapter->hw;
 917         u32 ivar, index;
 918
 919         /* 82576 uses a table-based method for assigning vectors.
 920            Each queue has a single entry in the table to which we write
 921            a vector number along with a "valid" bit.  Sadly, the layout
 922            of the table is somewhat counterintuitive. */
 923         if (rx_queue > IGBVF_NO_QUEUE) {
 924                 index = (rx_queue >> 1);
 925                 ivar = array_er32(IVAR0, index);
 926                 if (rx_queue & 0x1) {
 927                         /* vector goes into third byte of register */
 928                         ivar = ivar & 0xFF00FFFF;
 929                         ivar |= (msix_vector | E1000_IVAR_VALID) << 16;
 930                 } else {
 931                         /* vector goes into low byte of register */
 932                         ivar = ivar & 0xFFFFFF00;
 933                         ivar |= msix_vector | E1000_IVAR_VALID;
 934                 }
 935                 adapter->rx_ring[rx_queue].eims_value = 1 << msix_vector;
 936                 array_ew32(IVAR0, index, ivar);
 937         }
 938         if (tx_queue > IGBVF_NO_QUEUE) {
 939                 index = (tx_queue >> 1);
 940                 ivar = array_er32(IVAR0, index);
 941                 if (tx_queue & 0x1) {
 942                         /* vector goes into high byte of register */
 943                         ivar = ivar & 0x00FFFFFF;
 944                         ivar |= (msix_vector | E1000_IVAR_VALID) << 24;
 945                 } else {
 946                         /* vector goes into second byte of register */
 947                         ivar = ivar & 0xFFFF00FF;
 948                         ivar |= (msix_vector | E1000_IVAR_VALID) << 8;
 949                 }
 950                 adapter->tx_ring[tx_queue].eims_value = 1 << msix_vector;
 951                 array_ew32(IVAR0, index, ivar);
 952         }
 953 }
 954
 955 /**
 956  * igbvf_configure_msix - Configure MSI-X hardware
 957  *
 958  * igbvf_configure_msix sets up the hardware to properly
 959  * generate MSI-X interrupts.
 960  **/
 961 static void igbvf_configure_msix(struct igbvf_adapter *adapter)
 962 {
 963         u32 tmp;
 964         struct e1000_hw *hw = &adapter->hw;
 965         struct igbvf_ring *tx_ring = adapter->tx_ring;
 966         struct igbvf_ring *rx_ring = adapter->rx_ring;
 967         int vector = 0;
 968
 969         adapter->eims_enable_mask = 0;
 970
 971         igbvf_assign_vector(adapter, IGBVF_NO_QUEUE, 0, vector++);
 972         adapter->eims_enable_mask |= tx_ring->eims_value;
 973         if (tx_ring->itr_val)
 974                 writel(tx_ring->itr_val,
 975                        hw->hw_addr + tx_ring->itr_register);
 976         else
 977                 writel(1952, hw->hw_addr + tx_ring->itr_register);
 978
 979         igbvf_assign_vector(adapter, 0, IGBVF_NO_QUEUE, vector++);
 980         adapter->eims_enable_mask |= rx_ring->eims_value;
 981         if (rx_ring->itr_val)
 982                 writel(rx_ring->itr_val,
 983                        hw->hw_addr + rx_ring->itr_register);
 984         else
 985                 writel(1952, hw->hw_addr + rx_ring->itr_register);
 986
 987         /* set vector for other causes, i.e. link changes */
 988
 989         tmp = (vector++ | E1000_IVAR_VALID);
 990
 991         ew32(IVAR_MISC, tmp);
 992
 993         adapter->eims_enable_mask = (1 << (vector)) - 1;
 994         adapter->eims_other = 1 << (vector - 1);
 995         e1e_flush();
 996 }
 997
 998 static void igbvf_reset_interrupt_capability(struct igbvf_adapter *adapter)
 999 {
1000         if (adapter->msix_entries) {
1001                 pci_disable_msix(adapter->pdev);
1002                 kfree(adapter->msix_entries);
1003                 adapter->msix_entries = NULL;
1004         }
1005 }
1006
1007 /**
1008  * igbvf_set_interrupt_capability - set MSI or MSI-X if supported
1009  *
1010  * Attempt to configure interrupts using the best available
1011  * capabilities of the hardware and kernel.
1012  **/
1013 static void igbvf_set_interrupt_capability(struct igbvf_adapter *adapter)
1014 {
1015         int err = -ENOMEM;
1016         int i;
1017
1018         /* we allocate 3 vectors, 1 for tx, 1 for rx, one for pf messages */
1019         adapter->msix_entries = kcalloc(3, sizeof(struct msix_entry),
1020                                         GFP_KERNEL);
1021         if (adapter->msix_entries) {
1022                 for (i = 0; i < 3; i++)
1023                         adapter->msix_entries[i].entry = i;
1024
1025                 err = pci_enable_msix(adapter->pdev,
1026                                       adapter->msix_entries, 3);
1027         }
1028
1029         if (err) {
1030                 /* MSI-X failed */
1031                 dev_err(&adapter->pdev->dev,
1032                         "Failed to initialize MSI-X interrupts.\n");
1033                 igbvf_reset_interrupt_capability(adapter);
1034         }
1035 }
1036
1037 /**
1038  * igbvf_request_msix - Initialize MSI-X interrupts
1039  *
1040  * igbvf_request_msix allocates MSI-X vectors and requests interrupts from the
1041  * kernel.
1042  **/
1043 static int igbvf_request_msix(struct igbvf_adapter *adapter)
1044 {
1045         struct net_device *netdev = adapter->netdev;
1046         int err = 0, vector = 0;
1047
1048         if (strlen(netdev->name) < (IFNAMSIZ - 5)) {
1049                 sprintf(adapter->tx_ring->name, "%s-tx-0", netdev->name);
1050                 sprintf(adapter->rx_ring->name, "%s-rx-0", netdev->name);
1051         } else {
1052                 memcpy(adapter->tx_ring->name, netdev->name, IFNAMSIZ);
1053                 memcpy(adapter->rx_ring->name, netdev->name, IFNAMSIZ);
1054         }
1055
1056         err = request_irq(adapter->msix_entries[vector].vector,
1057                           igbvf_intr_msix_tx, 0, adapter->tx_ring->name,
1058                           netdev);
1059         if (err)
1060                 goto out;
1061
1062         adapter->tx_ring->itr_register = E1000_EITR(vector);
1063         adapter->tx_ring->itr_val = 1952;
1064         vector++;
1065
1066         err = request_irq(adapter->msix_entries[vector].vector,
1067                           igbvf_intr_msix_rx, 0, adapter->rx_ring->name,
1068                           netdev);
1069         if (err)
1070                 goto out;
1071
1072         adapter->rx_ring->itr_register = E1000_EITR(vector);
1073         adapter->rx_ring->itr_val = 1952;
1074         vector++;
1075
1076         err = request_irq(adapter->msix_entries[vector].vector,
1077                           igbvf_msix_other, 0, netdev->name, netdev);
1078         if (err)
1079                 goto out;
1080
1081         igbvf_configure_msix(adapter);
1082         return 0;
1083 out:
1084         return err;
1085 }
1086
1087 /**
1088  * igbvf_alloc_queues - Allocate memory for all rings
1089  * @adapter: board private structure to initialize
1090  **/
1091 static int __devinit igbvf_alloc_queues(struct igbvf_adapter *adapter)
1092 {
1093         struct net_device *netdev = adapter->netdev;
1094
1095         adapter->tx_ring = kzalloc(sizeof(struct igbvf_ring), GFP_KERNEL);
1096         if (!adapter->tx_ring)
1097                 return -ENOMEM;
1098
1099         adapter->rx_ring = kzalloc(sizeof(struct igbvf_ring), GFP_KERNEL);
1100         if (!adapter->rx_ring) {
1101                 kfree(adapter->tx_ring);
1102                 return -ENOMEM;
1103         }
1104
1105         netif_napi_add(netdev, &adapter->rx_ring->napi, igbvf_poll, 64);
1106
1107         return 0;
1108 }
1109
1110 /**
1111  * igbvf_request_irq - initialize interrupts
1112  *
1113  * Attempts to configure interrupts using the best available
1114  * capabilities of the hardware and kernel.
1115  **/
1116 static int igbvf_request_irq(struct igbvf_adapter *adapter)
1117 {
1118         int err = -1;
1119
1120         /* igbvf supports msi-x only */
1121         if (adapter->msix_entries)
1122                 err = igbvf_request_msix(adapter);
1123
1124         if (!err)
1125                 return err;
1126
1127         dev_err(&adapter->pdev->dev,
1128                 "Unable to allocate interrupt, Error: %d\n", err);
1129
1130         return err;
1131 }
1132
1133 static void igbvf_free_irq(struct igbvf_adapter *adapter)
1134 {
1135         struct net_device *netdev = adapter->netdev;
1136         int vector;
1137
1138         if (adapter->msix_entries) {
1139                 for (vector = 0; vector < 3; vector++)
1140                         free_irq(adapter->msix_entries[vector].vector, netdev);
1141         }
1142 }
1143
1144 /**
1145  * igbvf_irq_disable - Mask off interrupt generation on the NIC
1146  **/
1147 static void igbvf_irq_disable(struct igbvf_adapter *adapter)
1148 {
1149         struct e1000_hw *hw = &adapter->hw;
1150
1151         ew32(EIMC, ~0);
1152
1153         if (adapter->msix_entries)
1154                 ew32(EIAC, 0);
1155 }
1156
1157 /**
1158  * igbvf_irq_enable - Enable default interrupt generation settings
1159  **/
1160 static void igbvf_irq_enable(struct igbvf_adapter *adapter)
1161 {
1162         struct e1000_hw *hw = &adapter->hw;
1163
1164         ew32(EIAC, adapter->eims_enable_mask);
1165         ew32(EIAM, adapter->eims_enable_mask);
1166         ew32(EIMS, adapter->eims_enable_mask);
1167 }
1168
1169 /**
1170  * igbvf_poll - NAPI Rx polling callback
1171  * @napi: struct associated with this polling callback
1172  * @budget: amount of packets driver is allowed to process this poll
1173  **/
1174 static int igbvf_poll(struct napi_struct *napi, int budget)
1175 {
1176         struct igbvf_ring *rx_ring = container_of(napi, struct igbvf_ring, napi);
1177         struct igbvf_adapter *adapter = rx_ring->adapter;
1178         struct e1000_hw *hw = &adapter->hw;
1179         int work_done = 0;
1180
1181         igbvf_clean_rx_irq(adapter, &work_done, budget);
1182
1183         /* If not enough Rx work done, exit the polling mode */
1184         if (work_done < budget) {
1185                 napi_complete(napi);
1186
1187                 if (adapter->itr_setting & 3)
1188                         igbvf_set_itr(adapter);
1189
1190                 if (!test_bit(__IGBVF_DOWN, &adapter->state))
1191                         ew32(EIMS, adapter->rx_ring->eims_value);
1192         }
1193
1194         return work_done;
1195 }
1196
1197 /**
1198  * igbvf_set_rlpml - set receive large packet maximum length
1199  * @adapter: board private structure
1200  *
1201  * Configure the maximum size of packets that will be received
1202  */
1203 static void igbvf_set_rlpml(struct igbvf_adapter *adapter)
1204 {
1205         int max_frame_size = adapter->max_frame_size;
1206         struct e1000_hw *hw = &adapter->hw;
1207
1208         if (adapter->vlgrp)
1209                 max_frame_size += VLAN_TAG_SIZE;
1210
1211         e1000_rlpml_set_vf(hw, max_frame_size);
1212 }
1213
1214 static void igbvf_vlan_rx_add_vid(struct net_device *netdev, u16 vid)
1215 {
1216         struct igbvf_adapter *adapter = netdev_priv(netdev);
1217         struct e1000_hw *hw = &adapter->hw;
1218
1219         if (hw->mac.ops.set_vfta(hw, vid, true))
1220                 dev_err(&adapter->pdev->dev, "Failed to add vlan id %d\n", vid);
1221 }
1222
1223 static void igbvf_vlan_rx_kill_vid(struct net_device *netdev, u16 vid)
1224 {
1225         struct igbvf_adapter *adapter = netdev_priv(netdev);
1226         struct e1000_hw *hw = &adapter->hw;
1227
1228         igbvf_irq_disable(adapter);
1229         vlan_group_set_device(adapter->vlgrp, vid, NULL);
1230
1231         if (!test_bit(__IGBVF_DOWN, &adapter->state))
1232                 igbvf_irq_enable(adapter);
1233
1234         if (hw->mac.ops.set_vfta(hw, vid, false))
1235                 dev_err(&adapter->pdev->dev,
1236                         "Failed to remove vlan id %d\n", vid);
1237 }
1238
1239 static void igbvf_vlan_rx_register(struct net_device *netdev,
1240                                    struct vlan_group *grp)
1241 {
1242         struct igbvf_adapter *adapter = netdev_priv(netdev);
1243
1244         adapter->vlgrp = grp;
1245 }
1246
1247 static void igbvf_restore_vlan(struct igbvf_adapter *adapter)
1248 {
1249         u16 vid;
1250
1251         if (!adapter->vlgrp)
1252                 return;
1253
1254         for (vid = 0; vid < VLAN_GROUP_ARRAY_LEN; vid++) {
1255                 if (!vlan_group_get_device(adapter->vlgrp, vid))
1256                         continue;
1257                 igbvf_vlan_rx_add_vid(adapter->netdev, vid);
1258         }
1259
1260         igbvf_set_rlpml(adapter);
1261 }
1262
1263 /**
1264  * igbvf_configure_tx - Configure Transmit Unit after Reset
1265  * @adapter: board private structure
1266  *
1267  * Configure the Tx unit of the MAC after a reset.
1268  **/
1269 static void igbvf_configure_tx(struct igbvf_adapter *adapter)
1270 {
1271         struct e1000_hw *hw = &adapter->hw;
1272         struct igbvf_ring *tx_ring = adapter->tx_ring;
1273         u64 tdba;
1274         u32 txdctl, dca_txctrl;
1275
1276         /* disable transmits */
1277         txdctl = er32(TXDCTL(0));
1278         ew32(TXDCTL(0), txdctl & ~E1000_TXDCTL_QUEUE_ENABLE);
1279         msleep(10);
1280
1281         /* Setup the HW Tx Head and Tail descriptor pointers */
1282         ew32(TDLEN(0), tx_ring->count * sizeof(union e1000_adv_tx_desc));
1283         tdba = tx_ring->dma;
1284         ew32(TDBAL(0), (tdba & DMA_BIT_MASK(32)));
1285         ew32(TDBAH(0), (tdba >> 32));
1286         ew32(TDH(0), 0);
1287         ew32(TDT(0), 0);
1288         tx_ring->head = E1000_TDH(0);
1289         tx_ring->tail = E1000_TDT(0);
1290
1291         /* Turn off Relaxed Ordering on head write-backs.  The writebacks
1292          * MUST be delivered in order or it will completely screw up
1293          * our bookeeping.
1294          */
1295         dca_txctrl = er32(DCA_TXCTRL(0));
1296         dca_txctrl &= ~E1000_DCA_TXCTRL_TX_WB_RO_EN;
1297         ew32(DCA_TXCTRL(0), dca_txctrl);
1298
1299         /* enable transmits */
1300         txdctl |= E1000_TXDCTL_QUEUE_ENABLE;
1301         ew32(TXDCTL(0), txdctl);
1302
1303         /* Setup Transmit Descriptor Settings for eop descriptor */
1304         adapter->txd_cmd = E1000_ADVTXD_DCMD_EOP | E1000_ADVTXD_DCMD_IFCS;
1305
1306         /* enable Report Status bit */
1307         adapter->txd_cmd |= E1000_ADVTXD_DCMD_RS;
1308 }
1309
1310 /**
1311  * igbvf_setup_srrctl - configure the receive control registers
1312  * @adapter: Board private structure
1313  **/
1314 static void igbvf_setup_srrctl(struct igbvf_adapter *adapter)
1315 {
1316         struct e1000_hw *hw = &adapter->hw;
1317         u32 srrctl = 0;
1318
1319         srrctl &= ~(E1000_SRRCTL_DESCTYPE_MASK |
1320                     E1000_SRRCTL_BSIZEHDR_MASK |
1321                     E1000_SRRCTL_BSIZEPKT_MASK);
1322
1323         /* Enable queue drop to avoid head of line blocking */
1324         srrctl |= E1000_SRRCTL_DROP_EN;
1325
1326         /* Setup buffer sizes */
1327         srrctl |= ALIGN(adapter->rx_buffer_len, 1024) >>
1328                   E1000_SRRCTL_BSIZEPKT_SHIFT;
1329
1330         if (adapter->rx_buffer_len < 2048) {
1331                 adapter->rx_ps_hdr_size = 0;
1332                 srrctl |= E1000_SRRCTL_DESCTYPE_ADV_ONEBUF;
1333         } else {
1334                 adapter->rx_ps_hdr_size = 128;
1335                 srrctl |= adapter->rx_ps_hdr_size <<
1336                           E1000_SRRCTL_BSIZEHDRSIZE_SHIFT;
1337                 srrctl |= E1000_SRRCTL_DESCTYPE_HDR_SPLIT_ALWAYS;
1338         }
1339
1340         ew32(SRRCTL(0), srrctl);
1341 }
1342
1343 /**
1344  * igbvf_configure_rx - Configure Receive Unit after Reset
1345  * @adapter: board private structure
1346  *
1347  * Configure the Rx unit of the MAC after a reset.
1348  **/
1349 static void igbvf_configure_rx(struct igbvf_adapter *adapter)
1350 {
1351         struct e1000_hw *hw = &adapter->hw;
1352         struct igbvf_ring *rx_ring = adapter->rx_ring;
1353         u64 rdba;
1354         u32 rdlen, rxdctl;
1355
1356         /* disable receives */
1357         rxdctl = er32(RXDCTL(0));
1358         ew32(RXDCTL(0), rxdctl & ~E1000_RXDCTL_QUEUE_ENABLE);
1359         msleep(10);
1360
1361         rdlen = rx_ring->count * sizeof(union e1000_adv_rx_desc);
1362
1363         /*
1364          * Setup the HW Rx Head and Tail Descriptor Pointers and
1365          * the Base and Length of the Rx Descriptor Ring
1366          */
1367         rdba = rx_ring->dma;
1368         ew32(RDBAL(0), (rdba & DMA_BIT_MASK(32)));
1369         ew32(RDBAH(0), (rdba >> 32));
1370         ew32(RDLEN(0), rx_ring->count * sizeof(union e1000_adv_rx_desc));
1371         rx_ring->head = E1000_RDH(0);
1372         rx_ring->tail = E1000_RDT(0);
1373         ew32(RDH(0), 0);
1374         ew32(RDT(0), 0);
1375
1376         rxdctl |= E1000_RXDCTL_QUEUE_ENABLE;
1377         rxdctl &= 0xFFF00000;
1378         rxdctl |= IGBVF_RX_PTHRESH;
1379         rxdctl |= IGBVF_RX_HTHRESH << 8;
1380         rxdctl |= IGBVF_RX_WTHRESH << 16;
1381
1382         igbvf_set_rlpml(adapter);
1383
1384         /* enable receives */
1385         ew32(RXDCTL(0), rxdctl);
1386 }
1387
1388 /**
1389  * igbvf_set_multi - Multicast and Promiscuous mode set
1390  * @netdev: network interface device structure
1391  *
1392  * The set_multi entry point is called whenever the multicast address
1393  * list or the network interface flags are updated.  This routine is
1394  * responsible for configuring the hardware for proper multicast,
1395  * promiscuous mode, and all-multi behavior.
1396  **/
1397 static void igbvf_set_multi(struct net_device *netdev)
1398 {
1399         struct igbvf_adapter *adapter = netdev_priv(netdev);
1400         struct e1000_hw *hw = &adapter->hw;
1401         struct dev_mc_list *mc_ptr;
1402         u8  *mta_list = NULL;
1403         int i;
1404
1405         if (!netdev_mc_empty(netdev)) {
1406                 mta_list = kmalloc(netdev_mc_count(netdev) * 6, GFP_ATOMIC);
1407                 if (!mta_list) {
1408                         dev_err(&adapter->pdev->dev,
1409                                 "failed to allocate multicast filter list\n");
1410                         return;
1411                 }
1412         }
1413
1414         /* prepare a packed array of only addresses. */
1415         i = 0;
1416         netdev_for_each_mc_addr(mc_ptr, netdev)
1417                 memcpy(mta_list + (i++ * ETH_ALEN), mc_ptr->dmi_addr, ETH_ALEN);
1418
1419         hw->mac.ops.update_mc_addr_list(hw, mta_list, i, 0, 0);
1420         kfree(mta_list);
1421 }
1422
1423 /**
1424  * igbvf_configure - configure the hardware for Rx and Tx
1425  * @adapter: private board structure
1426  **/
1427 static void igbvf_configure(struct igbvf_adapter *adapter)
1428 {
1429         igbvf_set_multi(adapter->netdev);
1430
1431         igbvf_restore_vlan(adapter);
1432
1433         igbvf_configure_tx(adapter);
1434         igbvf_setup_srrctl(adapter);
1435         igbvf_configure_rx(adapter);
1436         igbvf_alloc_rx_buffers(adapter->rx_ring,
1437                                igbvf_desc_unused(adapter->rx_ring));
1438 }
1439
1440 /* igbvf_reset - bring the hardware into a known good state
1441  *
1442  * This function boots the hardware and enables some settings that
1443  * require a configuration cycle of the hardware - those cannot be
1444  * set/changed during runtime. After reset the device needs to be
1445  * properly configured for Rx, Tx etc.
1446  */
1447 static void igbvf_reset(struct igbvf_adapter *adapter)
1448 {
1449         struct e1000_mac_info *mac = &adapter->hw.mac;
1450         struct net_device *netdev = adapter->netdev;
1451         struct e1000_hw *hw = &adapter->hw;
1452
1453         /* Allow time for pending master requests to run */
1454         if (mac->ops.reset_hw(hw))
1455                 dev_err(&adapter->pdev->dev, "PF still resetting\n");
1456
1457         mac->ops.init_hw(hw);
1458
1459         if (is_valid_ether_addr(adapter->hw.mac.addr)) {
1460                 memcpy(netdev->dev_addr, adapter->hw.mac.addr,
1461                        netdev->addr_len);
1462                 memcpy(netdev->perm_addr, adapter->hw.mac.addr,
1463                        netdev->addr_len);
1464         }
1465
1466         adapter->last_reset = jiffies;
1467 }
1468
1469 int igbvf_up(struct igbvf_adapter *adapter)
1470 {
1471         struct e1000_hw *hw = &adapter->hw;
1472
1473         /* hardware has been reset, we need to reload some things */
1474         igbvf_configure(adapter);
1475
1476         clear_bit(__IGBVF_DOWN, &adapter->state);
1477
1478         napi_enable(&adapter->rx_ring->napi);
1479         if (adapter->msix_entries)
1480                 igbvf_configure_msix(adapter);
1481
1482         /* Clear any pending interrupts. */
1483         er32(EICR);
1484         igbvf_irq_enable(adapter);
1485
1486         /* start the watchdog */
1487         hw->mac.get_link_status = 1;
1488         mod_timer(&adapter->watchdog_timer, jiffies + 1);
1489
1490
1491         return 0;
1492 }
1493
1494 void igbvf_down(struct igbvf_adapter *adapter)
1495 {
1496         struct net_device *netdev = adapter->netdev;
1497         struct e1000_hw *hw = &adapter->hw;
1498         u32 rxdctl, txdctl;
1499
1500         /*
1501          * signal that we're down so the interrupt handler does not
1502          * reschedule our watchdog timer
1503          */
1504         set_bit(__IGBVF_DOWN, &adapter->state);
1505
1506         /* disable receives in the hardware */
1507         rxdctl = er32(RXDCTL(0));
1508         ew32(RXDCTL(0), rxdctl & ~E1000_RXDCTL_QUEUE_ENABLE);
1509
1510         netif_stop_queue(netdev);
1511
1512         /* disable transmits in the hardware */
1513         txdctl = er32(TXDCTL(0));
1514         ew32(TXDCTL(0), txdctl & ~E1000_TXDCTL_QUEUE_ENABLE);
1515
1516         /* flush both disables and wait for them to finish */
1517         e1e_flush();
1518         msleep(10);
1519
1520         napi_disable(&adapter->rx_ring->napi);
1521
1522         igbvf_irq_disable(adapter);
1523
1524         del_timer_sync(&adapter->watchdog_timer);
1525
1526         netif_carrier_off(netdev);
1527
1528         /* record the stats before reset*/
1529         igbvf_update_stats(adapter);
1530
1531         adapter->link_speed = 0;
1532         adapter->link_duplex = 0;
1533
1534         igbvf_reset(adapter);
1535         igbvf_clean_tx_ring(adapter->tx_ring);
1536         igbvf_clean_rx_ring(adapter->rx_ring);
1537 }
1538
1539 void igbvf_reinit_locked(struct igbvf_adapter *adapter)
1540 {
1541         might_sleep();
1542         while (test_and_set_bit(__IGBVF_RESETTING, &adapter->state))
1543                 msleep(1);
1544         igbvf_down(adapter);
1545         igbvf_up(adapter);
1546         clear_bit(__IGBVF_RESETTING, &adapter->state);
1547 }
1548
1549 /**
1550  * igbvf_sw_init - Initialize general software structures (struct igbvf_adapter)
1551  * @adapter: board private structure to initialize
1552  *
1553  * igbvf_sw_init initializes the Adapter private data structure.
1554  * Fields are initialized based on PCI device information and
1555  * OS network device settings (MTU size).
1556  **/
1557 static int __devinit igbvf_sw_init(struct igbvf_adapter *adapter)
1558 {
1559         struct net_device *netdev = adapter->netdev;
1560         s32 rc;
1561
1562         adapter->rx_buffer_len = ETH_FRAME_LEN + VLAN_HLEN + ETH_FCS_LEN;
1563         adapter->rx_ps_hdr_size = 0;
1564         adapter->max_frame_size = netdev->mtu + ETH_HLEN + ETH_FCS_LEN;
1565         adapter->min_frame_size = ETH_ZLEN + ETH_FCS_LEN;
1566
1567         adapter->tx_int_delay = 8;
1568         adapter->tx_abs_int_delay = 32;
1569         adapter->rx_int_delay = 0;
1570         adapter->rx_abs_int_delay = 8;
1571         adapter->itr_setting = 3;
1572         adapter->itr = 20000;
1573
1574         /* Set various function pointers */
1575         adapter->ei->init_ops(&adapter->hw);
1576
1577         rc = adapter->hw.mac.ops.init_params(&adapter->hw);
1578         if (rc)
1579                 return rc;
1580
1581         rc = adapter->hw.mbx.ops.init_params(&adapter->hw);
1582         if (rc)
1583                 return rc;
1584
1585         igbvf_set_interrupt_capability(adapter);
1586
1587         if (igbvf_alloc_queues(adapter))
1588                 return -ENOMEM;
1589
1590         spin_lock_init(&adapter->tx_queue_lock);
1591
1592         /* Explicitly disable IRQ since the NIC can be in any state. */
1593         igbvf_irq_disable(adapter);
1594
1595         spin_lock_init(&adapter->stats_lock);
1596
1597         set_bit(__IGBVF_DOWN, &adapter->state);
1598         return 0;
1599 }
1600
1601 static void igbvf_initialize_last_counter_stats(struct igbvf_adapter *adapter)
1602 {
1603         struct e1000_hw *hw = &adapter->hw;
1604
1605         adapter->stats.last_gprc = er32(VFGPRC);
1606         adapter->stats.last_gorc = er32(VFGORC);
1607         adapter->stats.last_gptc = er32(VFGPTC);
1608         adapter->stats.last_gotc = er32(VFGOTC);
1609         adapter->stats.last_mprc = er32(VFMPRC);
1610         adapter->stats.last_gotlbc = er32(VFGOTLBC);
1611         adapter->stats.last_gptlbc = er32(VFGPTLBC);
1612         adapter->stats.last_gorlbc = er32(VFGORLBC);
1613         adapter->stats.last_gprlbc = er32(VFGPRLBC);
1614
1615         adapter->stats.base_gprc = er32(VFGPRC);
1616         adapter->stats.base_gorc = er32(VFGORC);
1617         adapter->stats.base_gptc = er32(VFGPTC);
1618         adapter->stats.base_gotc = er32(VFGOTC);
1619         adapter->stats.base_mprc = er32(VFMPRC);
1620         adapter->stats.base_gotlbc = er32(VFGOTLBC);
1621         adapter->stats.base_gptlbc = er32(VFGPTLBC);
1622         adapter->stats.base_gorlbc = er32(VFGORLBC);
1623         adapter->stats.base_gprlbc = er32(VFGPRLBC);
1624 }
1625
1626 /**
1627  * igbvf_open - Called when a network interface is made active
1628  * @netdev: network interface device structure
1629  *
1630  * Returns 0 on success, negative value on failure
1631  *
1632  * The open entry point is called when a network interface is made
1633  * active by the system (IFF_UP).  At this point all resources needed
1634  * for transmit and receive operations are allocated, the interrupt
1635  * handler is registered with the OS, the watchdog timer is started,
1636  * and the stack is notified that the interface is ready.
1637  **/
1638 static int igbvf_open(struct net_device *netdev)
1639 {
1640         struct igbvf_adapter *adapter = netdev_priv(netdev);
1641         struct e1000_hw *hw = &adapter->hw;
1642         int err;
1643
1644         /* disallow open during test */
1645         if (test_bit(__IGBVF_TESTING, &adapter->state))
1646                 return -EBUSY;
1647
1648         /* allocate transmit descriptors */
1649         err = igbvf_setup_tx_resources(adapter, adapter->tx_ring);
1650         if (err)
1651                 goto err_setup_tx;
1652
1653         /* allocate receive descriptors */
1654         err = igbvf_setup_rx_resources(adapter, adapter->rx_ring);
1655         if (err)
1656                 goto err_setup_rx;
1657
1658         /*
1659          * before we allocate an interrupt, we must be ready to handle it.
1660          * Setting DEBUG_SHIRQ in the kernel makes it fire an interrupt
1661          * as soon as we call pci_request_irq, so we have to setup our
1662          * clean_rx handler before we do so.
1663          */
1664         igbvf_configure(adapter);
1665
1666         err = igbvf_request_irq(adapter);
1667         if (err)
1668                 goto err_req_irq;
1669
1670         /* From here on the code is the same as igbvf_up() */
1671         clear_bit(__IGBVF_DOWN, &adapter->state);
1672
1673         napi_enable(&adapter->rx_ring->napi);
1674
1675         /* clear any pending interrupts */
1676         er32(EICR);
1677
1678         igbvf_irq_enable(adapter);
1679
1680         /* start the watchdog */
1681         hw->mac.get_link_status = 1;
1682         mod_timer(&adapter->watchdog_timer, jiffies + 1);
1683
1684         return 0;
1685
1686 err_req_irq:
1687         igbvf_free_rx_resources(adapter->rx_ring);
1688 err_setup_rx:
1689         igbvf_free_tx_resources(adapter->tx_ring);
1690 err_setup_tx:
1691         igbvf_reset(adapter);
1692
1693         return err;
1694 }
1695
1696 /**
1697  * igbvf_close - Disables a network interface
1698  * @netdev: network interface device structure
1699  *
1700  * Returns 0, this is not allowed to fail
1701  *
1702  * The close entry point is called when an interface is de-activated
1703  * by the OS.  The hardware is still under the drivers control, but
1704  * needs to be disabled.  A global MAC reset is issued to stop the
1705  * hardware, and all transmit and receive resources are freed.
1706  **/
1707 static int igbvf_close(struct net_device *netdev)
1708 {
1709         struct igbvf_adapter *adapter = netdev_priv(netdev);
1710
1711         WARN_ON(test_bit(__IGBVF_RESETTING, &adapter->state));
1712         igbvf_down(adapter);
1713
1714         igbvf_free_irq(adapter);
1715
1716         igbvf_free_tx_resources(adapter->tx_ring);
1717         igbvf_free_rx_resources(adapter->rx_ring);
1718
1719         return 0;
1720 }
1721 /**
1722  * igbvf_set_mac - Change the Ethernet Address of the NIC
1723  * @netdev: network interface device structure
1724  * @p: pointer to an address structure
1725  *
1726  * Returns 0 on success, negative on failure
1727  **/
1728 static int igbvf_set_mac(struct net_device *netdev, void *p)
1729 {
1730         struct igbvf_adapter *adapter = netdev_priv(netdev);
1731         struct e1000_hw *hw = &adapter->hw;
1732         struct sockaddr *addr = p;
1733
1734         if (!is_valid_ether_addr(addr->sa_data))
1735                 return -EADDRNOTAVAIL;
1736
1737         memcpy(hw->mac.addr, addr->sa_data, netdev->addr_len);
1738
1739         hw->mac.ops.rar_set(hw, hw->mac.addr, 0);
1740
1741         if (memcmp(addr->sa_data, hw->mac.addr, 6))
1742                 return -EADDRNOTAVAIL;
1743
1744         memcpy(netdev->dev_addr, addr->sa_data, netdev->addr_len);
1745
1746         return 0;
1747 }
1748
1749 #define UPDATE_VF_COUNTER(reg, name)                                    \
1750         {                                                               \
1751                 u32 current_counter = er32(reg);                        \
1752                 if (current_counter < adapter->stats.last_##name)       \
1753                         adapter->stats.name += 0x100000000LL;           \
1754                 adapter->stats.last_##name = current_counter;           \
1755                 adapter->stats.name &= 0xFFFFFFFF00000000LL;            \
1756                 adapter->stats.name |= current_counter;                 \
1757         }
1758
1759 /**
1760  * igbvf_update_stats - Update the board statistics counters
1761  * @adapter: board private structure
1762 **/
1763 void igbvf_update_stats(struct igbvf_adapter *adapter)
1764 {
1765         struct e1000_hw *hw = &adapter->hw;
1766         struct pci_dev *pdev = adapter->pdev;
1767
1768         /*
1769          * Prevent stats update while adapter is being reset, link is down
1770          * or if the pci connection is down.
1771          */
1772         if (adapter->link_speed == 0)
1773                 return;
1774
1775         if (test_bit(__IGBVF_RESETTING, &adapter->state))
1776                 return;
1777
1778         if (pci_channel_offline(pdev))
1779                 return;
1780
1781         UPDATE_VF_COUNTER(VFGPRC, gprc);
1782         UPDATE_VF_COUNTER(VFGORC, gorc);
1783         UPDATE_VF_COUNTER(VFGPTC, gptc);
1784         UPDATE_VF_COUNTER(VFGOTC, gotc);
1785         UPDATE_VF_COUNTER(VFMPRC, mprc);
1786         UPDATE_VF_COUNTER(VFGOTLBC, gotlbc);
1787         UPDATE_VF_COUNTER(VFGPTLBC, gptlbc);
1788         UPDATE_VF_COUNTER(VFGORLBC, gorlbc);
1789         UPDATE_VF_COUNTER(VFGPRLBC, gprlbc);
1790
1791         /* Fill out the OS statistics structure */
1792         adapter->net_stats.multicast = adapter->stats.mprc;
1793 }
1794
1795 static void igbvf_print_link_info(struct igbvf_adapter *adapter)
1796 {
1797         dev_info(&adapter->pdev->dev, "Link is Up %d Mbps %s\n",
1798                  adapter->link_speed,
1799                  ((adapter->link_duplex == FULL_DUPLEX) ?
1800                   "Full Duplex" : "Half Duplex"));
1801 }
1802
1803 static bool igbvf_has_link(struct igbvf_adapter *adapter)
1804 {
1805         struct e1000_hw *hw = &adapter->hw;
1806         s32 ret_val = E1000_SUCCESS;
1807         bool link_active;
1808
1809         /* If interface is down, stay link down */
1810         if (test_bit(__IGBVF_DOWN, &adapter->state))
1811                 return false;
1812
1813         ret_val = hw->mac.ops.check_for_link(hw);
1814         link_active = !hw->mac.get_link_status;
1815
1816         /* if check for link returns error we will need to reset */
1817         if (ret_val && time_after(jiffies, adapter->last_reset + (10 * HZ)))
1818                 schedule_work(&adapter->reset_task);
1819
1820         return link_active;
1821 }
1822
1823 /**
1824  * igbvf_watchdog - Timer Call-back
1825  * @data: pointer to adapter cast into an unsigned long
1826  **/
1827 static void igbvf_watchdog(unsigned long data)
1828 {
1829         struct igbvf_adapter *adapter = (struct igbvf_adapter *) data;
1830
1831         /* Do the rest outside of interrupt context */
1832         schedule_work(&adapter->watchdog_task);
1833 }
1834
1835 static void igbvf_watchdog_task(struct work_struct *work)
1836 {
1837         struct igbvf_adapter *adapter = container_of(work,
1838                                                      struct igbvf_adapter,
1839                                                      watchdog_task);
1840         struct net_device *netdev = adapter->netdev;
1841         struct e1000_mac_info *mac = &adapter->hw.mac;
1842         struct igbvf_ring *tx_ring = adapter->tx_ring;
1843         struct e1000_hw *hw = &adapter->hw;
1844         u32 link;
1845         int tx_pending = 0;
1846
1847         link = igbvf_has_link(adapter);
1848
1849         if (link) {
1850                 if (!netif_carrier_ok(netdev)) {
1851                         bool txb2b = 1;
1852
1853                         mac->ops.get_link_up_info(&adapter->hw,
1854                                                   &adapter->link_speed,
1855                                                   &adapter->link_duplex);
1856                         igbvf_print_link_info(adapter);
1857
1858                         /* adjust timeout factor according to speed/duplex */
1859                         adapter->tx_timeout_factor = 1;
1860                         switch (adapter->link_speed) {
1861                         case SPEED_10:
1862                                 txb2b = 0;
1863                                 adapter->tx_timeout_factor = 16;
1864                                 break;
1865                         case SPEED_100:
1866                                 txb2b = 0;
1867                                 /* maybe add some timeout factor ? */
1868                                 break;
1869                         }
1870
1871                         netif_carrier_on(netdev);
1872                         netif_wake_queue(netdev);
1873                 }
1874         } else {
1875                 if (netif_carrier_ok(netdev)) {
1876                         adapter->link_speed = 0;
1877                         adapter->link_duplex = 0;
1878                         dev_info(&adapter->pdev->dev, "Link is Down\n");
1879                         netif_carrier_off(netdev);
1880                         netif_stop_queue(netdev);
1881                 }
1882         }
1883
1884         if (netif_carrier_ok(netdev)) {
1885                 igbvf_update_stats(adapter);
1886         } else {
1887                 tx_pending = (igbvf_desc_unused(tx_ring) + 1 <
1888                               tx_ring->count);
1889                 if (tx_pending) {
1890                         /*
1891                          * We've lost link, so the controller stops DMA,
1892                          * but we've got queued Tx work that's never going
1893                          * to get done, so reset controller to flush Tx.
1894                          * (Do the reset outside of interrupt context).
1895                          */
1896                         adapter->tx_timeout_count++;
1897                         schedule_work(&adapter->reset_task);
1898                 }
1899         }
1900
1901         /* Cause software interrupt to ensure Rx ring is cleaned */
1902         ew32(EICS, adapter->rx_ring->eims_value);
1903
1904         /* Force detection of hung controller every watchdog period */
1905         adapter->detect_tx_hung = 1;
1906
1907         /* Reset the timer */
1908         if (!test_bit(__IGBVF_DOWN, &adapter->state))
1909                 mod_timer(&adapter->watchdog_timer,
1910                           round_jiffies(jiffies + (2 * HZ)));
1911 }
1912
1913 #define IGBVF_TX_FLAGS_CSUM             0x00000001
1914 #define IGBVF_TX_FLAGS_VLAN             0x00000002
1915 #define IGBVF_TX_FLAGS_TSO              0x00000004
1916 #define IGBVF_TX_FLAGS_IPV4             0x00000008
1917 #define IGBVF_TX_FLAGS_VLAN_MASK        0xffff0000
1918 #define IGBVF_TX_FLAGS_VLAN_SHIFT       16
1919
1920 static int igbvf_tso(struct igbvf_adapter *adapter,
1921                      struct igbvf_ring *tx_ring,
1922                      struct sk_buff *skb, u32 tx_flags, u8 *hdr_len)
1923 {
1924         struct e1000_adv_tx_context_desc *context_desc;
1925         unsigned int i;
1926         int err;
1927         struct igbvf_buffer *buffer_info;
1928         u32 info = 0, tu_cmd = 0;
1929         u32 mss_l4len_idx, l4len;
1930         *hdr_len = 0;
1931
1932         if (skb_header_cloned(skb)) {
1933                 err = pskb_expand_head(skb, 0, 0, GFP_ATOMIC);
1934                 if (err) {
1935                         dev_err(&adapter->pdev->dev,
1936                                 "igbvf_tso returning an error\n");
1937                         return err;
1938                 }
1939         }
1940
1941         l4len = tcp_hdrlen(skb);
1942         *hdr_len += l4len;
1943
1944         if (skb->protocol == htons(ETH_P_IP)) {
1945                 struct iphdr *iph = ip_hdr(skb);
1946                 iph->tot_len = 0;
1947                 iph->check = 0;
1948                 tcp_hdr(skb)->check = ~csum_tcpudp_magic(iph->saddr,
1949                                                          iph->daddr, 0,
1950                                                          IPPROTO_TCP,
1951                                                          0);
1952         } else if (skb_is_gso_v6(skb)) {
1953                 ipv6_hdr(skb)->payload_len = 0;
1954                 tcp_hdr(skb)->check = ~csum_ipv6_magic(&ipv6_hdr(skb)->saddr,
1955                                                        &ipv6_hdr(skb)->daddr,
1956                                                        0, IPPROTO_TCP, 0);
1957         }
1958
1959         i = tx_ring->next_to_use;
1960
1961         buffer_info = &tx_ring->buffer_info[i];
1962         context_desc = IGBVF_TX_CTXTDESC_ADV(*tx_ring, i);
1963         /* VLAN MACLEN IPLEN */
1964         if (tx_flags & IGBVF_TX_FLAGS_VLAN)
1965                 info |= (tx_flags & IGBVF_TX_FLAGS_VLAN_MASK);
1966         info |= (skb_network_offset(skb) << E1000_ADVTXD_MACLEN_SHIFT);
1967         *hdr_len += skb_network_offset(skb);
1968         info |= (skb_transport_header(skb) - skb_network_header(skb));
1969         *hdr_len += (skb_transport_header(skb) - skb_network_header(skb));
1970         context_desc->vlan_macip_lens = cpu_to_le32(info);
1971
1972         /* ADV DTYP TUCMD MKRLOC/ISCSIHEDLEN */
1973         tu_cmd |= (E1000_TXD_CMD_DEXT | E1000_ADVTXD_DTYP_CTXT);
1974
1975         if (skb->protocol == htons(ETH_P_IP))
1976                 tu_cmd |= E1000_ADVTXD_TUCMD_IPV4;
1977         tu_cmd |= E1000_ADVTXD_TUCMD_L4T_TCP;
1978
1979         context_desc->type_tucmd_mlhl = cpu_to_le32(tu_cmd);
1980
1981         /* MSS L4LEN IDX */
1982         mss_l4len_idx = (skb_shinfo(skb)->gso_size << E1000_ADVTXD_MSS_SHIFT);
1983         mss_l4len_idx |= (l4len << E1000_ADVTXD_L4LEN_SHIFT);
1984
1985         context_desc->mss_l4len_idx = cpu_to_le32(mss_l4len_idx);
1986         context_desc->seqnum_seed = 0;
1987
1988         buffer_info->time_stamp = jiffies;
1989         buffer_info->next_to_watch = i;
1990         buffer_info->dma = 0;
1991         i++;
1992         if (i == tx_ring->count)
1993                 i = 0;
1994
1995         tx_ring->next_to_use = i;
1996
1997         return true;
1998 }
1999
2000 static inline bool igbvf_tx_csum(struct igbvf_adapter *adapter,
2001                                  struct igbvf_ring *tx_ring,
2002                                  struct sk_buff *skb, u32 tx_flags)
2003 {
2004         struct e1000_adv_tx_context_desc *context_desc;
2005         unsigned int i;
2006         struct igbvf_buffer *buffer_info;
2007         u32 info = 0, tu_cmd = 0;
2008
2009         if ((skb->ip_summed == CHECKSUM_PARTIAL) ||
2010             (tx_flags & IGBVF_TX_FLAGS_VLAN)) {
2011                 i = tx_ring->next_to_use;
2012                 buffer_info = &tx_ring->buffer_info[i];
2013                 context_desc = IGBVF_TX_CTXTDESC_ADV(*tx_ring, i);
2014
2015                 if (tx_flags & IGBVF_TX_FLAGS_VLAN)
2016                         info |= (tx_flags & IGBVF_TX_FLAGS_VLAN_MASK);
2017
2018                 info |= (skb_network_offset(skb) << E1000_ADVTXD_MACLEN_SHIFT);
2019                 if (skb->ip_summed == CHECKSUM_PARTIAL)
2020                         info |= (skb_transport_header(skb) -
2021                                  skb_network_header(skb));
2022
2023
2024                 context_desc->vlan_macip_lens = cpu_to_le32(info);
2025
2026                 tu_cmd |= (E1000_TXD_CMD_DEXT | E1000_ADVTXD_DTYP_CTXT);
2027
2028                 if (skb->ip_summed == CHECKSUM_PARTIAL) {
2029                         switch (skb->protocol) {
2030                         case __constant_htons(ETH_P_IP):
2031                                 tu_cmd |= E1000_ADVTXD_TUCMD_IPV4;
2032                                 if (ip_hdr(skb)->protocol == IPPROTO_TCP)
2033                                         tu_cmd |= E1000_ADVTXD_TUCMD_L4T_TCP;
2034                                 break;
2035                         case __constant_htons(ETH_P_IPV6):
2036                                 if (ipv6_hdr(skb)->nexthdr == IPPROTO_TCP)
2037                                         tu_cmd |= E1000_ADVTXD_TUCMD_L4T_TCP;
2038                                 break;
2039                         default:
2040                                 break;
2041                         }
2042                 }
2043
2044                 context_desc->type_tucmd_mlhl = cpu_to_le32(tu_cmd);
2045                 context_desc->seqnum_seed = 0;
2046                 context_desc->mss_l4len_idx = 0;
2047
2048                 buffer_info->time_stamp = jiffies;
2049                 buffer_info->next_to_watch = i;
2050                 buffer_info->dma = 0;
2051                 i++;
2052                 if (i == tx_ring->count)
2053                         i = 0;
2054                 tx_ring->next_to_use = i;
2055
2056                 return true;
2057         }
2058
2059         return false;
2060 }
2061
2062 static int igbvf_maybe_stop_tx(struct net_device *netdev, int size)
2063 {
2064         struct igbvf_adapter *adapter = netdev_priv(netdev);
2065
2066         /* there is enough descriptors then we don't need to worry  */
2067         if (igbvf_desc_unused(adapter->tx_ring) >= size)
2068                 return 0;
2069
2070         netif_stop_queue(netdev);
2071
2072         smp_mb();
2073
2074         /* We need to check again just in case room has been made available */
2075         if (igbvf_desc_unused(adapter->tx_ring) < size)
2076                 return -EBUSY;
2077
2078         netif_wake_queue(netdev);
2079
2080         ++adapter->restart_queue;
2081         return 0;
2082 }
2083
2084 #define IGBVF_MAX_TXD_PWR       16
2085 #define IGBVF_MAX_DATA_PER_TXD  (1 << IGBVF_MAX_TXD_PWR)
2086
2087 static inline int igbvf_tx_map_adv(struct igbvf_adapter *adapter,
2088                                    struct igbvf_ring *tx_ring,
2089                                    struct sk_buff *skb,
2090                                    unsigned int first)
2091 {
2092         struct igbvf_buffer *buffer_info;
2093         struct pci_dev *pdev = adapter->pdev;
2094         unsigned int len = skb_headlen(skb);
2095         unsigned int count = 0, i;
2096         unsigned int f;
2097
2098         i = tx_ring->next_to_use;
2099
2100         buffer_info = &tx_ring->buffer_info[i];
2101         BUG_ON(len >= IGBVF_MAX_DATA_PER_TXD);
2102         buffer_info->length = len;
2103         /* set time_stamp *before* dma to help avoid a possible race */
2104         buffer_info->time_stamp = jiffies;
2105         buffer_info->next_to_watch = i;
2106         buffer_info->mapped_as_page = false;
2107         buffer_info->dma = pci_map_single(pdev, skb->data, len,
2108                                           PCI_DMA_TODEVICE);
2109         if (pci_dma_mapping_error(pdev, buffer_info->dma))
2110                 goto dma_error;
2111
2112
2113         for (f = 0; f < skb_shinfo(skb)->nr_frags; f++) {
2114                 struct skb_frag_struct *frag;
2115
2116                 count++;
2117                 i++;
2118                 if (i == tx_ring->count)
2119                         i = 0;
2120
2121                 frag = &skb_shinfo(skb)->frags[f];
2122                 len = frag->size;
2123
2124                 buffer_info = &tx_ring->buffer_info[i];
2125                 BUG_ON(len >= IGBVF_MAX_DATA_PER_TXD);
2126                 buffer_info->length = len;
2127                 buffer_info->time_stamp = jiffies;
2128                 buffer_info->next_to_watch = i;
2129                 buffer_info->mapped_as_page = true;
2130                 buffer_info->dma = pci_map_page(pdev,
2131                                                 frag->page,
2132                                                 frag->page_offset,
2133                                                 len,
2134                                                 PCI_DMA_TODEVICE);
2135                 if (pci_dma_mapping_error(pdev, buffer_info->dma))
2136                         goto dma_error;
2137         }
2138
2139         tx_ring->buffer_info[i].skb = skb;
2140         tx_ring->buffer_info[first].next_to_watch = i;
2141
2142         return ++count;
2143
2144 dma_error:
2145         dev_err(&pdev->dev, "TX DMA map failed\n");
2146
2147         /* clear timestamp and dma mappings for failed buffer_info mapping */
2148         buffer_info->dma = 0;
2149         buffer_info->time_stamp = 0;
2150         buffer_info->length = 0;
2151         buffer_info->next_to_watch = 0;
2152         buffer_info->mapped_as_page = false;
2153         if (count)
2154                 count--;
2155
2156         /* clear timestamp and dma mappings for remaining portion of packet */
2157         while (count--) {
2158                 if (i==0)
2159                         i += tx_ring->count;
2160                 i--;
2161                 buffer_info = &tx_ring->buffer_info[i];
2162                 igbvf_put_txbuf(adapter, buffer_info);
2163         }
2164
2165         return 0;
2166 }
2167
2168 static inline void igbvf_tx_queue_adv(struct igbvf_adapter *adapter,
2169                                       struct igbvf_ring *tx_ring,
2170                                       int tx_flags, int count, u32 paylen,
2171                                       u8 hdr_len)
2172 {
2173         union e1000_adv_tx_desc *tx_desc = NULL;
2174         struct igbvf_buffer *buffer_info;
2175         u32 olinfo_status = 0, cmd_type_len;
2176         unsigned int i;
2177
2178         cmd_type_len = (E1000_ADVTXD_DTYP_DATA | E1000_ADVTXD_DCMD_IFCS |
2179                         E1000_ADVTXD_DCMD_DEXT);
2180
2181         if (tx_flags & IGBVF_TX_FLAGS_VLAN)
2182                 cmd_type_len |= E1000_ADVTXD_DCMD_VLE;
2183
2184         if (tx_flags & IGBVF_TX_FLAGS_TSO) {
2185                 cmd_type_len |= E1000_ADVTXD_DCMD_TSE;
2186
2187                 /* insert tcp checksum */
2188                 olinfo_status |= E1000_TXD_POPTS_TXSM << 8;
2189
2190                 /* insert ip checksum */
2191                 if (tx_flags & IGBVF_TX_FLAGS_IPV4)
2192                         olinfo_status |= E1000_TXD_POPTS_IXSM << 8;
2193
2194         } else if (tx_flags & IGBVF_TX_FLAGS_CSUM) {
2195                 olinfo_status |= E1000_TXD_POPTS_TXSM << 8;
2196         }
2197
2198         olinfo_status |= ((paylen - hdr_len) << E1000_ADVTXD_PAYLEN_SHIFT);
2199
2200         i = tx_ring->next_to_use;
2201         while (count--) {
2202                 buffer_info = &tx_ring->buffer_info[i];
2203                 tx_desc = IGBVF_TX_DESC_ADV(*tx_ring, i);
2204                 tx_desc->read.buffer_addr = cpu_to_le64(buffer_info->dma);
2205                 tx_desc->read.cmd_type_len =
2206                          cpu_to_le32(cmd_type_len | buffer_info->length);
2207                 tx_desc->read.olinfo_status = cpu_to_le32(olinfo_status);
2208                 i++;
2209                 if (i == tx_ring->count)
2210                         i = 0;
2211         }
2212
2213         tx_desc->read.cmd_type_len |= cpu_to_le32(adapter->txd_cmd);
2214         /* Force memory writes to complete before letting h/w
2215          * know there are new descriptors to fetch.  (Only
2216          * applicable for weak-ordered memory model archs,
2217          * such as IA-64). */
2218         wmb();
2219
2220         tx_ring->next_to_use = i;
2221         writel(i, adapter->hw.hw_addr + tx_ring->tail);
2222         /* we need this if more than one processor can write to our tail
2223          * at a time, it syncronizes IO on IA64/Altix systems */
2224         mmiowb();
2225 }
2226
2227 static netdev_tx_t igbvf_xmit_frame_ring_adv(struct sk_buff *skb,
2228                                              struct net_device *netdev,
2229                                              struct igbvf_ring *tx_ring)
2230 {
2231         struct igbvf_adapter *adapter = netdev_priv(netdev);
2232         unsigned int first, tx_flags = 0;
2233         u8 hdr_len = 0;
2234         int count = 0;
2235         int tso = 0;
2236
2237         if (test_bit(__IGBVF_DOWN, &adapter->state)) {
2238                 dev_kfree_skb_any(skb);
2239                 return NETDEV_TX_OK;
2240         }
2241
2242         if (skb->len <= 0) {
2243                 dev_kfree_skb_any(skb);
2244                 return NETDEV_TX_OK;
2245         }
2246
2247         /*
2248          * need: count + 4 desc gap to keep tail from touching
2249          *       + 2 desc gap to keep tail from touching head,
2250          *       + 1 desc for skb->data,
2251          *       + 1 desc for context descriptor,
2252          * head, otherwise try next time
2253          */
2254         if (igbvf_maybe_stop_tx(netdev, skb_shinfo(skb)->nr_frags + 4)) {
2255                 /* this is a hard error */
2256                 return NETDEV_TX_BUSY;
2257         }
2258
2259         if (adapter->vlgrp && vlan_tx_tag_present(skb)) {
2260                 tx_flags |= IGBVF_TX_FLAGS_VLAN;
2261                 tx_flags |= (vlan_tx_tag_get(skb) << IGBVF_TX_FLAGS_VLAN_SHIFT);
2262         }
2263
2264         if (skb->protocol == htons(ETH_P_IP))
2265                 tx_flags |= IGBVF_TX_FLAGS_IPV4;
2266
2267         first = tx_ring->next_to_use;
2268
2269         tso = skb_is_gso(skb) ?
2270                 igbvf_tso(adapter, tx_ring, skb, tx_flags, &hdr_len) : 0;
2271         if (unlikely(tso < 0)) {
2272                 dev_kfree_skb_any(skb);
2273                 return NETDEV_TX_OK;
2274         }
2275
2276         if (tso)
2277                 tx_flags |= IGBVF_TX_FLAGS_TSO;
2278         else if (igbvf_tx_csum(adapter, tx_ring, skb, tx_flags) &&
2279                  (skb->ip_summed == CHECKSUM_PARTIAL))
2280                 tx_flags |= IGBVF_TX_FLAGS_CSUM;
2281
2282         /*
2283          * count reflects descriptors mapped, if 0 then mapping error
2284          * has occured and we need to rewind the descriptor queue
2285          */
2286         count = igbvf_tx_map_adv(adapter, tx_ring, skb, first);
2287
2288         if (count) {
2289                 igbvf_tx_queue_adv(adapter, tx_ring, tx_flags, count,
2290                                    skb->len, hdr_len);
2291                 /* Make sure there is space in the ring for the next send. */
2292                 igbvf_maybe_stop_tx(netdev, MAX_SKB_FRAGS + 4);
2293         } else {
2294                 dev_kfree_skb_any(skb);
2295                 tx_ring->buffer_info[first].time_stamp = 0;
2296                 tx_ring->next_to_use = first;
2297         }
2298
2299         return NETDEV_TX_OK;
2300 }
2301
2302 static netdev_tx_t igbvf_xmit_frame(struct sk_buff *skb,
2303                                     struct net_device *netdev)
2304 {
2305         struct igbvf_adapter *adapter = netdev_priv(netdev);
2306         struct igbvf_ring *tx_ring;
2307
2308         if (test_bit(__IGBVF_DOWN, &adapter->state)) {
2309                 dev_kfree_skb_any(skb);
2310                 return NETDEV_TX_OK;
2311         }
2312
2313         tx_ring = &adapter->tx_ring[0];
2314
2315         return igbvf_xmit_frame_ring_adv(skb, netdev, tx_ring);
2316 }
2317
2318 /**
2319  * igbvf_tx_timeout - Respond to a Tx Hang
2320  * @netdev: network interface device structure
2321  **/
2322 static void igbvf_tx_timeout(struct net_device *netdev)
2323 {
2324         struct igbvf_adapter *adapter = netdev_priv(netdev);
2325
2326         /* Do the reset outside of interrupt context */
2327         adapter->tx_timeout_count++;
2328         schedule_work(&adapter->reset_task);
2329 }
2330
2331 static void igbvf_reset_task(struct work_struct *work)
2332 {
2333         struct igbvf_adapter *adapter;
2334         adapter = container_of(work, struct igbvf_adapter, reset_task);
2335
2336         igbvf_reinit_locked(adapter);
2337 }
2338
2339 /**
2340  * igbvf_get_stats - Get System Network Statistics
2341  * @netdev: network interface device structure
2342  *
2343  * Returns the address of the device statistics structure.
2344  * The statistics are actually updated from the timer callback.
2345  **/
2346 static struct net_device_stats *igbvf_get_stats(struct net_device *netdev)
2347 {
2348         struct igbvf_adapter *adapter = netdev_priv(netdev);
2349
2350         /* only return the current stats */
2351         return &adapter->net_stats;
2352 }
2353
2354 /**
2355  * igbvf_change_mtu - Change the Maximum Transfer Unit
2356  * @netdev: network interface device structure
2357  * @new_mtu: new value for maximum frame size
2358  *
2359  * Returns 0 on success, negative on failure
2360  **/
2361 static int igbvf_change_mtu(struct net_device *netdev, int new_mtu)
2362 {
2363         struct igbvf_adapter *adapter = netdev_priv(netdev);
2364         int max_frame = new_mtu + ETH_HLEN + ETH_FCS_LEN;
2365
2366         if ((new_mtu < 68) || (max_frame > MAX_JUMBO_FRAME_SIZE)) {
2367                 dev_err(&adapter->pdev->dev, "Invalid MTU setting\n");
2368                 return -EINVAL;
2369         }
2370
2371 #define MAX_STD_JUMBO_FRAME_SIZE 9234
2372         if (max_frame > MAX_STD_JUMBO_FRAME_SIZE) {
2373                 dev_err(&adapter->pdev->dev, "MTU > 9216 not supported.\n");
2374                 return -EINVAL;
2375         }
2376
2377         while (test_and_set_bit(__IGBVF_RESETTING, &adapter->state))
2378                 msleep(1);
2379         /* igbvf_down has a dependency on max_frame_size */
2380         adapter->max_frame_size = max_frame;
2381         if (netif_running(netdev))
2382                 igbvf_down(adapter);
2383
2384         /*
2385          * NOTE: netdev_alloc_skb reserves 16 bytes, and typically NET_IP_ALIGN
2386          * means we reserve 2 more, this pushes us to allocate from the next
2387          * larger slab size.
2388          * i.e. RXBUFFER_2048 --> size-4096 slab
2389          * However with the new *_jumbo_rx* routines, jumbo receives will use
2390          * fragmented skbs
2391          */
2392
2393         if (max_frame <= 1024)
2394                 adapter->rx_buffer_len = 1024;
2395         else if (max_frame <= 2048)
2396                 adapter->rx_buffer_len = 2048;
2397         else
2398 #if (PAGE_SIZE / 2) > 16384
2399                 adapter->rx_buffer_len = 16384;
2400 #else
2401                 adapter->rx_buffer_len = PAGE_SIZE / 2;
2402 #endif
2403
2404
2405         /* adjust allocation if LPE protects us, and we aren't using SBP */
2406         if ((max_frame == ETH_FRAME_LEN + ETH_FCS_LEN) ||
2407              (max_frame == ETH_FRAME_LEN + VLAN_HLEN + ETH_FCS_LEN))
2408                 adapter->rx_buffer_len = ETH_FRAME_LEN + VLAN_HLEN +
2409                                          ETH_FCS_LEN;
2410
2411         dev_info(&adapter->pdev->dev, "changing MTU from %d to %d\n",
2412                  netdev->mtu, new_mtu);
2413         netdev->mtu = new_mtu;
2414
2415         if (netif_running(netdev))
2416                 igbvf_up(adapter);
2417         else
2418                 igbvf_reset(adapter);
2419
2420         clear_bit(__IGBVF_RESETTING, &adapter->state);
2421
2422         return 0;
2423 }
2424
2425 static int igbvf_ioctl(struct net_device *netdev, struct ifreq *ifr, int cmd)
2426 {
2427         switch (cmd) {
2428         default:
2429                 return -EOPNOTSUPP;
2430         }
2431 }
2432
2433 static int igbvf_suspend(struct pci_dev *pdev, pm_message_t state)
2434 {
2435         struct net_device *netdev = pci_get_drvdata(pdev);
2436         struct igbvf_adapter *adapter = netdev_priv(netdev);
2437 #ifdef CONFIG_PM
2438         int retval = 0;
2439 #endif
2440
2441         netif_device_detach(netdev);
2442
2443         if (netif_running(netdev)) {
2444                 WARN_ON(test_bit(__IGBVF_RESETTING, &adapter->state));
2445                 igbvf_down(adapter);
2446                 igbvf_free_irq(adapter);
2447         }
2448
2449 #ifdef CONFIG_PM
2450         retval = pci_save_state(pdev);
2451         if (retval)
2452                 return retval;
2453 #endif
2454
2455         pci_disable_device(pdev);
2456
2457         return 0;
2458 }
2459
2460 #ifdef CONFIG_PM
2461 static int igbvf_resume(struct pci_dev *pdev)
2462 {
2463         struct net_device *netdev = pci_get_drvdata(pdev);
2464         struct igbvf_adapter *adapter = netdev_priv(netdev);
2465         u32 err;
2466
2467         pci_restore_state(pdev);
2468         err = pci_enable_device_mem(pdev);
2469         if (err) {
2470                 dev_err(&pdev->dev, "Cannot enable PCI device from suspend\n");
2471                 return err;
2472         }
2473
2474         pci_set_master(pdev);
2475
2476         if (netif_running(netdev)) {
2477                 err = igbvf_request_irq(adapter);
2478                 if (err)
2479                         return err;
2480         }
2481
2482         igbvf_reset(adapter);
2483
2484         if (netif_running(netdev))
2485                 igbvf_up(adapter);
2486
2487         netif_device_attach(netdev);
2488
2489         return 0;
2490 }
2491 #endif
2492
2493 static void igbvf_shutdown(struct pci_dev *pdev)
2494 {
2495         igbvf_suspend(pdev, PMSG_SUSPEND);
2496 }
2497
2498 #ifdef CONFIG_NET_POLL_CONTROLLER
2499 /*
2500  * Polling 'interrupt' - used by things like netconsole to send skbs
2501  * without having to re-enable interrupts. It's not called while
2502  * the interrupt routine is executing.
2503  */
2504 static void igbvf_netpoll(struct net_device *netdev)
2505 {
2506         struct igbvf_adapter *adapter = netdev_priv(netdev);
2507
2508         disable_irq(adapter->pdev->irq);
2509
2510         igbvf_clean_tx_irq(adapter->tx_ring);
2511
2512         enable_irq(adapter->pdev->irq);
2513 }
2514 #endif
2515
2516 /**
2517  * igbvf_io_error_detected - called when PCI error is detected
2518  * @pdev: Pointer to PCI device
2519  * @state: The current pci connection state
2520  *
2521  * This function is called after a PCI bus error affecting
2522  * this device has been detected.
2523  */
2524 static pci_ers_result_t igbvf_io_error_detected(struct pci_dev *pdev,
2525                                                 pci_channel_state_t state)
2526 {
2527         struct net_device *netdev = pci_get_drvdata(pdev);
2528         struct igbvf_adapter *adapter = netdev_priv(netdev);
2529
2530         netif_device_detach(netdev);
2531
2532         if (state == pci_channel_io_perm_failure)
2533                 return PCI_ERS_RESULT_DISCONNECT;
2534
2535         if (netif_running(netdev))
2536                 igbvf_down(adapter);
2537         pci_disable_device(pdev);
2538
2539         /* Request a slot slot reset. */
2540         return PCI_ERS_RESULT_NEED_RESET;
2541 }
2542
2543 /**
2544  * igbvf_io_slot_reset - called after the pci bus has been reset.
2545  * @pdev: Pointer to PCI device
2546  *
2547  * Restart the card from scratch, as if from a cold-boot. Implementation
2548  * resembles the first-half of the igbvf_resume routine.
2549  */
2550 static pci_ers_result_t igbvf_io_slot_reset(struct pci_dev *pdev)
2551 {
2552         struct net_device *netdev = pci_get_drvdata(pdev);
2553         struct igbvf_adapter *adapter = netdev_priv(netdev);
2554
2555         if (pci_enable_device_mem(pdev)) {
2556                 dev_err(&pdev->dev,
2557                         "Cannot re-enable PCI device after reset.\n");
2558                 return PCI_ERS_RESULT_DISCONNECT;
2559         }
2560         pci_set_master(pdev);
2561
2562         igbvf_reset(adapter);
2563
2564         return PCI_ERS_RESULT_RECOVERED;
2565 }
2566
2567 /**
2568  * igbvf_io_resume - called when traffic can start flowing again.
2569  * @pdev: Pointer to PCI device
2570  *
2571  * This callback is called when the error recovery driver tells us that
2572  * its OK to resume normal operation. Implementation resembles the
2573  * second-half of the igbvf_resume routine.
2574  */
2575 static void igbvf_io_resume(struct pci_dev *pdev)
2576 {
2577         struct net_device *netdev = pci_get_drvdata(pdev);
2578         struct igbvf_adapter *adapter = netdev_priv(netdev);
2579
2580         if (netif_running(netdev)) {
2581                 if (igbvf_up(adapter)) {
2582                         dev_err(&pdev->dev,
2583                                 "can't bring device back up after reset\n");
2584                         return;
2585                 }
2586         }
2587
2588         netif_device_attach(netdev);
2589 }
2590
2591 static void igbvf_print_device_info(struct igbvf_adapter *adapter)
2592 {
2593         struct e1000_hw *hw = &adapter->hw;
2594         struct net_device *netdev = adapter->netdev;
2595         struct pci_dev *pdev = adapter->pdev;
2596
2597         dev_info(&pdev->dev, "Intel(R) 82576 Virtual Function\n");
2598         dev_info(&pdev->dev, "Address: %pM\n", netdev->dev_addr);
2599         dev_info(&pdev->dev, "MAC: %d\n", hw->mac.type);
2600 }
2601
2602 static const struct net_device_ops igbvf_netdev_ops = {
2603         .ndo_open                       = igbvf_open,
2604         .ndo_stop                       = igbvf_close,
2605         .ndo_start_xmit                 = igbvf_xmit_frame,
2606         .ndo_get_stats                  = igbvf_get_stats,
2607         .ndo_set_multicast_list         = igbvf_set_multi,
2608         .ndo_set_mac_address            = igbvf_set_mac,
2609         .ndo_change_mtu                 = igbvf_change_mtu,
2610         .ndo_do_ioctl                   = igbvf_ioctl,
2611         .ndo_tx_timeout                 = igbvf_tx_timeout,
2612         .ndo_vlan_rx_register           = igbvf_vlan_rx_register,
2613         .ndo_vlan_rx_add_vid            = igbvf_vlan_rx_add_vid,
2614         .ndo_vlan_rx_kill_vid           = igbvf_vlan_rx_kill_vid,
2615 #ifdef CONFIG_NET_POLL_CONTROLLER
2616         .ndo_poll_controller            = igbvf_netpoll,
2617 #endif
2618 };
2619
2620 /**
2621  * igbvf_probe - Device Initialization Routine
2622  * @pdev: PCI device information struct
2623  * @ent: entry in igbvf_pci_tbl
2624  *
2625  * Returns 0 on success, negative on failure
2626  *
2627  * igbvf_probe initializes an adapter identified by a pci_dev structure.
2628  * The OS initialization, configuring of the adapter private structure,
2629  * and a hardware reset occur.
2630  **/
2631 static int __devinit igbvf_probe(struct pci_dev *pdev,
2632                                  const struct pci_device_id *ent)
2633 {
2634         struct net_device *netdev;
2635         struct igbvf_adapter *adapter;
2636         struct e1000_hw *hw;
2637         const struct igbvf_info *ei = igbvf_info_tbl[ent->driver_data];
2638
2639         static int cards_found;
2640         int err, pci_using_dac;
2641
2642         err = pci_enable_device_mem(pdev);
2643         if (err)
2644                 return err;
2645
2646         pci_using_dac = 0;
2647         err = pci_set_dma_mask(pdev, DMA_BIT_MASK(64));
2648         if (!err) {
2649                 err = pci_set_consistent_dma_mask(pdev, DMA_BIT_MASK(64));
2650                 if (!err)
2651                         pci_using_dac = 1;
2652         } else {
2653                 err = pci_set_dma_mask(pdev, DMA_BIT_MASK(32));
2654                 if (err) {
2655                         err = pci_set_consistent_dma_mask(pdev,
2656                                                           DMA_BIT_MASK(32));
2657                         if (err) {
2658                                 dev_err(&pdev->dev, "No usable DMA "
2659                                         "configuration, aborting\n");
2660                                 goto err_dma;
2661                         }
2662                 }
2663         }
2664
2665         err = pci_request_regions(pdev, igbvf_driver_name);
2666         if (err)
2667                 goto err_pci_reg;
2668
2669         pci_set_master(pdev);
2670
2671         err = -ENOMEM;
2672         netdev = alloc_etherdev(sizeof(struct igbvf_adapter));
2673         if (!netdev)
2674                 goto err_alloc_etherdev;
2675
2676         SET_NETDEV_DEV(netdev, &pdev->dev);
2677
2678         pci_set_drvdata(pdev, netdev);
2679         adapter = netdev_priv(netdev);
2680         hw = &adapter->hw;
2681         adapter->netdev = netdev;
2682         adapter->pdev = pdev;
2683         adapter->ei = ei;
2684         adapter->pba = ei->pba;
2685         adapter->flags = ei->flags;
2686         adapter->hw.back = adapter;
2687         adapter->hw.mac.type = ei->mac;
2688         adapter->msg_enable = (1 << NETIF_MSG_DRV | NETIF_MSG_PROBE) - 1;
2689
2690         /* PCI config space info */
2691
2692         hw->vendor_id = pdev->vendor;
2693         hw->device_id = pdev->device;
2694         hw->subsystem_vendor_id = pdev->subsystem_vendor;
2695         hw->subsystem_device_id = pdev->subsystem_device;
2696
2697         pci_read_config_byte(pdev, PCI_REVISION_ID, &hw->revision_id);
2698
2699         err = -EIO;
2700         adapter->hw.hw_addr = ioremap(pci_resource_start(pdev, 0),
2701                                       pci_resource_len(pdev, 0));
2702
2703         if (!adapter->hw.hw_addr)
2704                 goto err_ioremap;
2705
2706         if (ei->get_variants) {
2707                 err = ei->get_variants(adapter);
2708                 if (err)
2709                         goto err_ioremap;
2710         }
2711
2712         /* setup adapter struct */
2713         err = igbvf_sw_init(adapter);
2714         if (err)
2715                 goto err_sw_init;
2716
2717         /* construct the net_device struct */
2718         netdev->netdev_ops = &igbvf_netdev_ops;
2719
2720         igbvf_set_ethtool_ops(netdev);
2721         netdev->watchdog_timeo = 5 * HZ;
2722         strncpy(netdev->name, pci_name(pdev), sizeof(netdev->name) - 1);
2723
2724         adapter->bd_number = cards_found++;
2725
2726         netdev->features = NETIF_F_SG |
2727                            NETIF_F_IP_CSUM |
2728                            NETIF_F_HW_VLAN_TX |
2729                            NETIF_F_HW_VLAN_RX |
2730                            NETIF_F_HW_VLAN_FILTER;
2731
2732         netdev->features |= NETIF_F_IPV6_CSUM;
2733         netdev->features |= NETIF_F_TSO;
2734         netdev->features |= NETIF_F_TSO6;
2735
2736         if (pci_using_dac)
2737                 netdev->features |= NETIF_F_HIGHDMA;
2738
2739         netdev->vlan_features |= NETIF_F_TSO;
2740         netdev->vlan_features |= NETIF_F_TSO6;
2741         netdev->vlan_features |= NETIF_F_IP_CSUM;
2742         netdev->vlan_features |= NETIF_F_IPV6_CSUM;
2743         netdev->vlan_features |= NETIF_F_SG;
2744
2745         /*reset the controller to put the device in a known good state */
2746         err = hw->mac.ops.reset_hw(hw);
2747         if (err) {
2748                 dev_info(&pdev->dev,
2749                          "PF still in reset state, assigning new address."
2750                          " Is the PF interface up?\n");
2751                 random_ether_addr(hw->mac.addr);
2752         } else {
2753                 err = hw->mac.ops.read_mac_addr(hw);
2754                 if (err) {
2755                         dev_err(&pdev->dev, "Error reading MAC address\n");
2756                         goto err_hw_init;
2757                 }
2758         }
2759
2760         memcpy(netdev->dev_addr, adapter->hw.mac.addr, netdev->addr_len);
2761         memcpy(netdev->perm_addr, adapter->hw.mac.addr, netdev->addr_len);
2762
2763         if (!is_valid_ether_addr(netdev->perm_addr)) {
2764                 dev_err(&pdev->dev, "Invalid MAC Address: %pM\n",
2765                         netdev->dev_addr);
2766                 err = -EIO;
2767                 goto err_hw_init;
2768         }
2769
2770         setup_timer(&adapter->watchdog_timer, &igbvf_watchdog,
2771                     (unsigned long) adapter);
2772
2773         INIT_WORK(&adapter->reset_task, igbvf_reset_task);
2774         INIT_WORK(&adapter->watchdog_task, igbvf_watchdog_task);
2775
2776         /* ring size defaults */
2777         adapter->rx_ring->count = 1024;
2778         adapter->tx_ring->count = 1024;
2779
2780         /* reset the hardware with the new settings */
2781         igbvf_reset(adapter);
2782
2783         /* tell the stack to leave us alone until igbvf_open() is called */
2784         netif_carrier_off(netdev);
2785         netif_stop_queue(netdev);
2786
2787         strcpy(netdev->name, "eth%d");
2788         err = register_netdev(netdev);
2789         if (err)
2790                 goto err_hw_init;
2791
2792         igbvf_print_device_info(adapter);
2793
2794         igbvf_initialize_last_counter_stats(adapter);
2795
2796         return 0;
2797
2798 err_hw_init:
2799         kfree(adapter->tx_ring);
2800         kfree(adapter->rx_ring);
2801 err_sw_init:
2802         igbvf_reset_interrupt_capability(adapter);
2803         iounmap(adapter->hw.hw_addr);
2804 err_ioremap:
2805         free_netdev(netdev);
2806 err_alloc_etherdev:
2807         pci_release_regions(pdev);
2808 err_pci_reg:
2809 err_dma:
2810         pci_disable_device(pdev);
2811         return err;
2812 }
2813
2814 /**
2815  * igbvf_remove - Device Removal Routine
2816  * @pdev: PCI device information struct
2817  *
2818  * igbvf_remove is called by the PCI subsystem to alert the driver
2819  * that it should release a PCI device.  The could be caused by a
2820  * Hot-Plug event, or because the driver is going to be removed from
2821  * memory.
2822  **/
2823 static void __devexit igbvf_remove(struct pci_dev *pdev)
2824 {
2825         struct net_device *netdev = pci_get_drvdata(pdev);
2826         struct igbvf_adapter *adapter = netdev_priv(netdev);
2827         struct e1000_hw *hw = &adapter->hw;
2828
2829         /*
2830          * flush_scheduled work may reschedule our watchdog task, so
2831          * explicitly disable watchdog tasks from being rescheduled
2832          */
2833         set_bit(__IGBVF_DOWN, &adapter->state);
2834         del_timer_sync(&adapter->watchdog_timer);
2835
2836         flush_scheduled_work();
2837
2838         unregister_netdev(netdev);
2839
2840         igbvf_reset_interrupt_capability(adapter);
2841
2842         /*
2843          * it is important to delete the napi struct prior to freeing the
2844          * rx ring so that you do not end up with null pointer refs
2845          */
2846         netif_napi_del(&adapter->rx_ring->napi);
2847         kfree(adapter->tx_ring);
2848         kfree(adapter->rx_ring);
2849
2850         iounmap(hw->hw_addr);
2851         if (hw->flash_address)
2852                 iounmap(hw->flash_address);
2853         pci_release_regions(pdev);
2854
2855         free_netdev(netdev);
2856
2857         pci_disable_device(pdev);
2858 }
2859
2860 /* PCI Error Recovery (ERS) */
2861 static struct pci_error_handlers igbvf_err_handler = {
2862         .error_detected = igbvf_io_error_detected,
2863         .slot_reset = igbvf_io_slot_reset,
2864         .resume = igbvf_io_resume,
2865 };
2866
2867 static DEFINE_PCI_DEVICE_TABLE(igbvf_pci_tbl) = {
2868         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82576_VF), board_vf },
2869         { } /* terminate list */
2870 };
2871 MODULE_DEVICE_TABLE(pci, igbvf_pci_tbl);
2872
2873 /* PCI Device API Driver */
2874 static struct pci_driver igbvf_driver = {
2875         .name     = igbvf_driver_name,
2876         .id_table = igbvf_pci_tbl,
2877         .probe    = igbvf_probe,
2878         .remove   = __devexit_p(igbvf_remove),
2879 #ifdef CONFIG_PM
2880         /* Power Management Hooks */
2881         .suspend  = igbvf_suspend,
2882         .resume   = igbvf_resume,
2883 #endif
2884         .shutdown = igbvf_shutdown,
2885         .err_handler = &igbvf_err_handler
2886 };
2887
2888 /**
2889  * igbvf_init_module - Driver Registration Routine
2890  *
2891  * igbvf_init_module is the first routine called when the driver is
2892  * loaded. All it does is register with the PCI subsystem.
2893  **/
2894 static int __init igbvf_init_module(void)
2895 {
2896         int ret;
2897         printk(KERN_INFO "%s - version %s\n",
2898                igbvf_driver_string, igbvf_driver_version);
2899         printk(KERN_INFO "%s\n", igbvf_copyright);
2900
2901         ret = pci_register_driver(&igbvf_driver);
2902         pm_qos_add_requirement(PM_QOS_CPU_DMA_LATENCY, igbvf_driver_name,
2903                                PM_QOS_DEFAULT_VALUE);
2904
2905         return ret;
2906 }
2907 module_init(igbvf_init_module);
2908
2909 /**
2910  * igbvf_exit_module - Driver Exit Cleanup Routine
2911  *
2912  * igbvf_exit_module is called just before the driver is removed
2913  * from memory.
2914  **/
2915 static void __exit igbvf_exit_module(void)
2916 {
2917         pci_unregister_driver(&igbvf_driver);
2918         pm_qos_remove_requirement(PM_QOS_CPU_DMA_LATENCY, igbvf_driver_name);
2919 }
2920 module_exit(igbvf_exit_module);
2921
2922
2923 MODULE_AUTHOR("Intel Corporation, <e1000-devel@lists.sourceforge.net>");
2924 MODULE_DESCRIPTION("Intel(R) 82576 Virtual Function Network Driver");
2925 MODULE_LICENSE("GPL");
2926 MODULE_VERSION(DRV_VERSION);
2927
2928 /* netdev.c */