git.oblomov.eu Git - linux-2.6/blob - drivers/net/igbvf/netdev.c

   1 /*******************************************************************************
   2
   3   Intel(R) 82576 Virtual Function Linux driver
   4   Copyright(c) 2009 Intel Corporation.
   5
   6   This program is free software; you can redistribute it and/or modify it
   7   under the terms and conditions of the GNU General Public License,
   8   version 2, as published by the Free Software Foundation.
   9
  10   This program is distributed in the hope it will be useful, but WITHOUT
  11   ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
  12   FITNESS FOR A PARTICULAR PURPOSE.  See the GNU General Public License for
  13   more details.
  14
  15   You should have received a copy of the GNU General Public License along with
  16   this program; if not, write to the Free Software Foundation, Inc.,
  17   51 Franklin St - Fifth Floor, Boston, MA 02110-1301 USA.
  18
  19   The full GNU General Public License is included in this distribution in
  20   the file called "COPYING".
  21
  22   Contact Information:
  23   e1000-devel Mailing List <e1000-devel@lists.sourceforge.net>
  24   Intel Corporation, 5200 N.E. Elam Young Parkway, Hillsboro, OR 97124-6497
  25
  26 *******************************************************************************/
  27
  28 #include <linux/module.h>
  29 #include <linux/types.h>
  30 #include <linux/init.h>
  31 #include <linux/pci.h>
  32 #include <linux/vmalloc.h>
  33 #include <linux/pagemap.h>
  34 #include <linux/delay.h>
  35 #include <linux/netdevice.h>
  36 #include <linux/tcp.h>
  37 #include <linux/ipv6.h>
  38 #include <net/checksum.h>
  39 #include <net/ip6_checksum.h>
  40 #include <linux/mii.h>
  41 #include <linux/ethtool.h>
  42 #include <linux/if_vlan.h>
  43 #include <linux/pm_qos_params.h>
  44
  45 #include "igbvf.h"
  46
  47 #define DRV_VERSION "1.0.0-k0"
  48 char igbvf_driver_name[] = "igbvf";
  49 const char igbvf_driver_version[] = DRV_VERSION;
  50 static const char igbvf_driver_string[] =
  51                                 "Intel(R) Virtual Function Network Driver";
  52 static const char igbvf_copyright[] = "Copyright (c) 2009 Intel Corporation.";
  53
  54 static int igbvf_poll(struct napi_struct *napi, int budget);
  55
  56 static struct igbvf_info igbvf_vf_info = {
  57         .mac                    = e1000_vfadapt,
  58         .flags                  = FLAG_HAS_JUMBO_FRAMES
  59                                   | FLAG_RX_CSUM_ENABLED,
  60         .pba                    = 10,
  61         .init_ops               = e1000_init_function_pointers_vf,
  62 };
  63
  64 static const struct igbvf_info *igbvf_info_tbl[] = {
  65         [board_vf]              = &igbvf_vf_info,
  66 };
  67
  68 /**
  69  * igbvf_desc_unused - calculate if we have unused descriptors
  70  **/
  71 static int igbvf_desc_unused(struct igbvf_ring *ring)
  72 {
  73         if (ring->next_to_clean > ring->next_to_use)
  74                 return ring->next_to_clean - ring->next_to_use - 1;
  75
  76         return ring->count + ring->next_to_clean - ring->next_to_use - 1;
  77 }
  78
  79 /**
  80  * igbvf_receive_skb - helper function to handle Rx indications
  81  * @adapter: board private structure
  82  * @status: descriptor status field as written by hardware
  83  * @vlan: descriptor vlan field as written by hardware (no le/be conversion)
  84  * @skb: pointer to sk_buff to be indicated to stack
  85  **/
  86 static void igbvf_receive_skb(struct igbvf_adapter *adapter,
  87                               struct net_device *netdev,
  88                               struct sk_buff *skb,
  89                               u32 status, u16 vlan)
  90 {
  91         if (adapter->vlgrp && (status & E1000_RXD_STAT_VP))
  92                 vlan_hwaccel_receive_skb(skb, adapter->vlgrp,
  93                                          le16_to_cpu(vlan) &
  94                                          E1000_RXD_SPC_VLAN_MASK);
  95         else
  96                 netif_receive_skb(skb);
  97
  98         netdev->last_rx = jiffies;
  99 }
 100
 101 static inline void igbvf_rx_checksum_adv(struct igbvf_adapter *adapter,
 102                                          u32 status_err, struct sk_buff *skb)
 103 {
 104         skb->ip_summed = CHECKSUM_NONE;
 105
 106         /* Ignore Checksum bit is set or checksum is disabled through ethtool */
 107         if ((status_err & E1000_RXD_STAT_IXSM))
 108                 return;
 109         /* TCP/UDP checksum error bit is set */
 110         if (status_err &
 111             (E1000_RXDEXT_STATERR_TCPE | E1000_RXDEXT_STATERR_IPE)) {
 112                 /* let the stack verify checksum errors */
 113                 adapter->hw_csum_err++;
 114                 return;
 115         }
 116         /* It must be a TCP or UDP packet with a valid checksum */
 117         if (status_err & (E1000_RXD_STAT_TCPCS | E1000_RXD_STAT_UDPCS))
 118                 skb->ip_summed = CHECKSUM_UNNECESSARY;
 119
 120         adapter->hw_csum_good++;
 121 }
 122
 123 /**
 124  * igbvf_alloc_rx_buffers - Replace used receive buffers; packet split
 125  * @rx_ring: address of ring structure to repopulate
 126  * @cleaned_count: number of buffers to repopulate
 127  **/
 128 static void igbvf_alloc_rx_buffers(struct igbvf_ring *rx_ring,
 129                                    int cleaned_count)
 130 {
 131         struct igbvf_adapter *adapter = rx_ring->adapter;
 132         struct net_device *netdev = adapter->netdev;
 133         struct pci_dev *pdev = adapter->pdev;
 134         union e1000_adv_rx_desc *rx_desc;
 135         struct igbvf_buffer *buffer_info;
 136         struct sk_buff *skb;
 137         unsigned int i;
 138         int bufsz;
 139
 140         i = rx_ring->next_to_use;
 141         buffer_info = &rx_ring->buffer_info[i];
 142
 143         if (adapter->rx_ps_hdr_size)
 144                 bufsz = adapter->rx_ps_hdr_size;
 145         else
 146                 bufsz = adapter->rx_buffer_len;
 147         bufsz += NET_IP_ALIGN;
 148
 149         while (cleaned_count--) {
 150                 rx_desc = IGBVF_RX_DESC_ADV(*rx_ring, i);
 151
 152                 if (adapter->rx_ps_hdr_size && !buffer_info->page_dma) {
 153                         if (!buffer_info->page) {
 154                                 buffer_info->page = alloc_page(GFP_ATOMIC);
 155                                 if (!buffer_info->page) {
 156                                         adapter->alloc_rx_buff_failed++;
 157                                         goto no_buffers;
 158                                 }
 159                                 buffer_info->page_offset = 0;
 160                         } else {
 161                                 buffer_info->page_offset ^= PAGE_SIZE / 2;
 162                         }
 163                         buffer_info->page_dma =
 164                                 pci_map_page(pdev, buffer_info->page,
 165                                              buffer_info->page_offset,
 166                                              PAGE_SIZE / 2,
 167                                              PCI_DMA_FROMDEVICE);
 168                 }
 169
 170                 if (!buffer_info->skb) {
 171                         skb = netdev_alloc_skb(netdev, bufsz);
 172                         if (!skb) {
 173                                 adapter->alloc_rx_buff_failed++;
 174                                 goto no_buffers;
 175                         }
 176
 177                         /* Make buffer alignment 2 beyond a 16 byte boundary
 178                          * this will result in a 16 byte aligned IP header after
 179                          * the 14 byte MAC header is removed
 180                          */
 181                         skb_reserve(skb, NET_IP_ALIGN);
 182
 183                         buffer_info->skb = skb;
 184                         buffer_info->dma = pci_map_single(pdev, skb->data,
 185                                                           bufsz,
 186                                                           PCI_DMA_FROMDEVICE);
 187                 }
 188                 /* Refresh the desc even if buffer_addrs didn't change because
 189                  * each write-back erases this info. */
 190                 if (adapter->rx_ps_hdr_size) {
 191                         rx_desc->read.pkt_addr =
 192                              cpu_to_le64(buffer_info->page_dma);
 193                         rx_desc->read.hdr_addr = cpu_to_le64(buffer_info->dma);
 194                 } else {
 195                         rx_desc->read.pkt_addr =
 196                              cpu_to_le64(buffer_info->dma);
 197                         rx_desc->read.hdr_addr = 0;
 198                 }
 199
 200                 i++;
 201                 if (i == rx_ring->count)
 202                         i = 0;
 203                 buffer_info = &rx_ring->buffer_info[i];
 204         }
 205
 206 no_buffers:
 207         if (rx_ring->next_to_use != i) {
 208                 rx_ring->next_to_use = i;
 209                 if (i == 0)
 210                         i = (rx_ring->count - 1);
 211                 else
 212                         i--;
 213
 214                 /* Force memory writes to complete before letting h/w
 215                  * know there are new descriptors to fetch.  (Only
 216                  * applicable for weak-ordered memory model archs,
 217                  * such as IA-64). */
 218                 wmb();
 219                 writel(i, adapter->hw.hw_addr + rx_ring->tail);
 220         }
 221 }
 222
 223 /**
 224  * igbvf_clean_rx_irq - Send received data up the network stack; legacy
 225  * @adapter: board private structure
 226  *
 227  * the return value indicates whether actual cleaning was done, there
 228  * is no guarantee that everything was cleaned
 229  **/
 230 static bool igbvf_clean_rx_irq(struct igbvf_adapter *adapter,
 231                                int *work_done, int work_to_do)
 232 {
 233         struct igbvf_ring *rx_ring = adapter->rx_ring;
 234         struct net_device *netdev = adapter->netdev;
 235         struct pci_dev *pdev = adapter->pdev;
 236         union e1000_adv_rx_desc *rx_desc, *next_rxd;
 237         struct igbvf_buffer *buffer_info, *next_buffer;
 238         struct sk_buff *skb;
 239         bool cleaned = false;
 240         int cleaned_count = 0;
 241         unsigned int total_bytes = 0, total_packets = 0;
 242         unsigned int i;
 243         u32 length, hlen, staterr;
 244
 245         i = rx_ring->next_to_clean;
 246         rx_desc = IGBVF_RX_DESC_ADV(*rx_ring, i);
 247         staterr = le32_to_cpu(rx_desc->wb.upper.status_error);
 248
 249         while (staterr & E1000_RXD_STAT_DD) {
 250                 if (*work_done >= work_to_do)
 251                         break;
 252                 (*work_done)++;
 253
 254                 buffer_info = &rx_ring->buffer_info[i];
 255
 256                 /* HW will not DMA in data larger than the given buffer, even
 257                  * if it parses the (NFS, of course) header to be larger.  In
 258                  * that case, it fills the header buffer and spills the rest
 259                  * into the page.
 260                  */
 261                 hlen = (le16_to_cpu(rx_desc->wb.lower.lo_dword.hs_rss.hdr_info) &
 262                   E1000_RXDADV_HDRBUFLEN_MASK) >> E1000_RXDADV_HDRBUFLEN_SHIFT;
 263                 if (hlen > adapter->rx_ps_hdr_size)
 264                         hlen = adapter->rx_ps_hdr_size;
 265
 266                 length = le16_to_cpu(rx_desc->wb.upper.length);
 267                 cleaned = true;
 268                 cleaned_count++;
 269
 270                 skb = buffer_info->skb;
 271                 prefetch(skb->data - NET_IP_ALIGN);
 272                 buffer_info->skb = NULL;
 273                 if (!adapter->rx_ps_hdr_size) {
 274                         pci_unmap_single(pdev, buffer_info->dma,
 275                                          adapter->rx_buffer_len,
 276                                          PCI_DMA_FROMDEVICE);
 277                         buffer_info->dma = 0;
 278                         skb_put(skb, length);
 279                         goto send_up;
 280                 }
 281
 282                 if (!skb_shinfo(skb)->nr_frags) {
 283                         pci_unmap_single(pdev, buffer_info->dma,
 284                                          adapter->rx_ps_hdr_size + NET_IP_ALIGN,
 285                                          PCI_DMA_FROMDEVICE);
 286                         skb_put(skb, hlen);
 287                 }
 288
 289                 if (length) {
 290                         pci_unmap_page(pdev, buffer_info->page_dma,
 291                                        PAGE_SIZE / 2,
 292                                        PCI_DMA_FROMDEVICE);
 293                         buffer_info->page_dma = 0;
 294
 295                         skb_fill_page_desc(skb, skb_shinfo(skb)->nr_frags++,
 296                                            buffer_info->page,
 297                                            buffer_info->page_offset,
 298                                            length);
 299
 300                         if ((adapter->rx_buffer_len > (PAGE_SIZE / 2)) ||
 301                             (page_count(buffer_info->page) != 1))
 302                                 buffer_info->page = NULL;
 303                         else
 304                                 get_page(buffer_info->page);
 305
 306                         skb->len += length;
 307                         skb->data_len += length;
 308                         skb->truesize += length;
 309                 }
 310 send_up:
 311                 i++;
 312                 if (i == rx_ring->count)
 313                         i = 0;
 314                 next_rxd = IGBVF_RX_DESC_ADV(*rx_ring, i);
 315                 prefetch(next_rxd);
 316                 next_buffer = &rx_ring->buffer_info[i];
 317
 318                 if (!(staterr & E1000_RXD_STAT_EOP)) {
 319                         buffer_info->skb = next_buffer->skb;
 320                         buffer_info->dma = next_buffer->dma;
 321                         next_buffer->skb = skb;
 322                         next_buffer->dma = 0;
 323                         goto next_desc;
 324                 }
 325
 326                 if (staterr & E1000_RXDEXT_ERR_FRAME_ERR_MASK) {
 327                         dev_kfree_skb_irq(skb);
 328                         goto next_desc;
 329                 }
 330
 331                 total_bytes += skb->len;
 332                 total_packets++;
 333
 334                 igbvf_rx_checksum_adv(adapter, staterr, skb);
 335
 336                 skb->protocol = eth_type_trans(skb, netdev);
 337
 338                 igbvf_receive_skb(adapter, netdev, skb, staterr,
 339                                   rx_desc->wb.upper.vlan);
 340
 341                 netdev->last_rx = jiffies;
 342
 343 next_desc:
 344                 rx_desc->wb.upper.status_error = 0;
 345
 346                 /* return some buffers to hardware, one at a time is too slow */
 347                 if (cleaned_count >= IGBVF_RX_BUFFER_WRITE) {
 348                         igbvf_alloc_rx_buffers(rx_ring, cleaned_count);
 349                         cleaned_count = 0;
 350                 }
 351
 352                 /* use prefetched values */
 353                 rx_desc = next_rxd;
 354                 buffer_info = next_buffer;
 355
 356                 staterr = le32_to_cpu(rx_desc->wb.upper.status_error);
 357         }
 358
 359         rx_ring->next_to_clean = i;
 360         cleaned_count = igbvf_desc_unused(rx_ring);
 361
 362         if (cleaned_count)
 363                 igbvf_alloc_rx_buffers(rx_ring, cleaned_count);
 364
 365         adapter->total_rx_packets += total_packets;
 366         adapter->total_rx_bytes += total_bytes;
 367         adapter->net_stats.rx_bytes += total_bytes;
 368         adapter->net_stats.rx_packets += total_packets;
 369         return cleaned;
 370 }
 371
 372 static void igbvf_put_txbuf(struct igbvf_adapter *adapter,
 373                             struct igbvf_buffer *buffer_info)
 374 {
 375         buffer_info->dma = 0;
 376         if (buffer_info->skb) {
 377                 skb_dma_unmap(&adapter->pdev->dev, buffer_info->skb,
 378                               DMA_TO_DEVICE);
 379                 dev_kfree_skb_any(buffer_info->skb);
 380                 buffer_info->skb = NULL;
 381         }
 382         buffer_info->time_stamp = 0;
 383 }
 384
 385 static void igbvf_print_tx_hang(struct igbvf_adapter *adapter)
 386 {
 387         struct igbvf_ring *tx_ring = adapter->tx_ring;
 388         unsigned int i = tx_ring->next_to_clean;
 389         unsigned int eop = tx_ring->buffer_info[i].next_to_watch;
 390         union e1000_adv_tx_desc *eop_desc = IGBVF_TX_DESC_ADV(*tx_ring, eop);
 391
 392         /* detected Tx unit hang */
 393         dev_err(&adapter->pdev->dev,
 394                 "Detected Tx Unit Hang:\n"
 395                 "  TDH                  <%x>\n"
 396                 "  TDT                  <%x>\n"
 397                 "  next_to_use          <%x>\n"
 398                 "  next_to_clean        <%x>\n"
 399                 "buffer_info[next_to_clean]:\n"
 400                 "  time_stamp           <%lx>\n"
 401                 "  next_to_watch        <%x>\n"
 402                 "  jiffies              <%lx>\n"
 403                 "  next_to_watch.status <%x>\n",
 404                 readl(adapter->hw.hw_addr + tx_ring->head),
 405                 readl(adapter->hw.hw_addr + tx_ring->tail),
 406                 tx_ring->next_to_use,
 407                 tx_ring->next_to_clean,
 408                 tx_ring->buffer_info[eop].time_stamp,
 409                 eop,
 410                 jiffies,
 411                 eop_desc->wb.status);
 412 }
 413
 414 /**
 415  * igbvf_setup_tx_resources - allocate Tx resources (Descriptors)
 416  * @adapter: board private structure
 417  *
 418  * Return 0 on success, negative on failure
 419  **/
 420 int igbvf_setup_tx_resources(struct igbvf_adapter *adapter,
 421                              struct igbvf_ring *tx_ring)
 422 {
 423         struct pci_dev *pdev = adapter->pdev;
 424         int size;
 425
 426         size = sizeof(struct igbvf_buffer) * tx_ring->count;
 427         tx_ring->buffer_info = vmalloc(size);
 428         if (!tx_ring->buffer_info)
 429                 goto err;
 430         memset(tx_ring->buffer_info, 0, size);
 431
 432         /* round up to nearest 4K */
 433         tx_ring->size = tx_ring->count * sizeof(union e1000_adv_tx_desc);
 434         tx_ring->size = ALIGN(tx_ring->size, 4096);
 435
 436         tx_ring->desc = pci_alloc_consistent(pdev, tx_ring->size,
 437                                              &tx_ring->dma);
 438
 439         if (!tx_ring->desc)
 440                 goto err;
 441
 442         tx_ring->adapter = adapter;
 443         tx_ring->next_to_use = 0;
 444         tx_ring->next_to_clean = 0;
 445
 446         return 0;
 447 err:
 448         vfree(tx_ring->buffer_info);
 449         dev_err(&adapter->pdev->dev,
 450                 "Unable to allocate memory for the transmit descriptor ring\n");
 451         return -ENOMEM;
 452 }
 453
 454 /**
 455  * igbvf_setup_rx_resources - allocate Rx resources (Descriptors)
 456  * @adapter: board private structure
 457  *
 458  * Returns 0 on success, negative on failure
 459  **/
 460 int igbvf_setup_rx_resources(struct igbvf_adapter *adapter,
 461                              struct igbvf_ring *rx_ring)
 462 {
 463         struct pci_dev *pdev = adapter->pdev;
 464         int size, desc_len;
 465
 466         size = sizeof(struct igbvf_buffer) * rx_ring->count;
 467         rx_ring->buffer_info = vmalloc(size);
 468         if (!rx_ring->buffer_info)
 469                 goto err;
 470         memset(rx_ring->buffer_info, 0, size);
 471
 472         desc_len = sizeof(union e1000_adv_rx_desc);
 473
 474         /* Round up to nearest 4K */
 475         rx_ring->size = rx_ring->count * desc_len;
 476         rx_ring->size = ALIGN(rx_ring->size, 4096);
 477
 478         rx_ring->desc = pci_alloc_consistent(pdev, rx_ring->size,
 479                                              &rx_ring->dma);
 480
 481         if (!rx_ring->desc)
 482                 goto err;
 483
 484         rx_ring->next_to_clean = 0;
 485         rx_ring->next_to_use = 0;
 486
 487         rx_ring->adapter = adapter;
 488
 489         return 0;
 490
 491 err:
 492         vfree(rx_ring->buffer_info);
 493         rx_ring->buffer_info = NULL;
 494         dev_err(&adapter->pdev->dev,
 495                 "Unable to allocate memory for the receive descriptor ring\n");
 496         return -ENOMEM;
 497 }
 498
 499 /**
 500  * igbvf_clean_tx_ring - Free Tx Buffers
 501  * @tx_ring: ring to be cleaned
 502  **/
 503 static void igbvf_clean_tx_ring(struct igbvf_ring *tx_ring)
 504 {
 505         struct igbvf_adapter *adapter = tx_ring->adapter;
 506         struct igbvf_buffer *buffer_info;
 507         unsigned long size;
 508         unsigned int i;
 509
 510         if (!tx_ring->buffer_info)
 511                 return;
 512
 513         /* Free all the Tx ring sk_buffs */
 514         for (i = 0; i < tx_ring->count; i++) {
 515                 buffer_info = &tx_ring->buffer_info[i];
 516                 igbvf_put_txbuf(adapter, buffer_info);
 517         }
 518
 519         size = sizeof(struct igbvf_buffer) * tx_ring->count;
 520         memset(tx_ring->buffer_info, 0, size);
 521
 522         /* Zero out the descriptor ring */
 523         memset(tx_ring->desc, 0, tx_ring->size);
 524
 525         tx_ring->next_to_use = 0;
 526         tx_ring->next_to_clean = 0;
 527
 528         writel(0, adapter->hw.hw_addr + tx_ring->head);
 529         writel(0, adapter->hw.hw_addr + tx_ring->tail);
 530 }
 531
 532 /**
 533  * igbvf_free_tx_resources - Free Tx Resources per Queue
 534  * @tx_ring: ring to free resources from
 535  *
 536  * Free all transmit software resources
 537  **/
 538 void igbvf_free_tx_resources(struct igbvf_ring *tx_ring)
 539 {
 540         struct pci_dev *pdev = tx_ring->adapter->pdev;
 541
 542         igbvf_clean_tx_ring(tx_ring);
 543
 544         vfree(tx_ring->buffer_info);
 545         tx_ring->buffer_info = NULL;
 546
 547         pci_free_consistent(pdev, tx_ring->size, tx_ring->desc, tx_ring->dma);
 548
 549         tx_ring->desc = NULL;
 550 }
 551
 552 /**
 553  * igbvf_clean_rx_ring - Free Rx Buffers per Queue
 554  * @adapter: board private structure
 555  **/
 556 static void igbvf_clean_rx_ring(struct igbvf_ring *rx_ring)
 557 {
 558         struct igbvf_adapter *adapter = rx_ring->adapter;
 559         struct igbvf_buffer *buffer_info;
 560         struct pci_dev *pdev = adapter->pdev;
 561         unsigned long size;
 562         unsigned int i;
 563
 564         if (!rx_ring->buffer_info)
 565                 return;
 566
 567         /* Free all the Rx ring sk_buffs */
 568         for (i = 0; i < rx_ring->count; i++) {
 569                 buffer_info = &rx_ring->buffer_info[i];
 570                 if (buffer_info->dma) {
 571                         if (adapter->rx_ps_hdr_size){
 572                                 pci_unmap_single(pdev, buffer_info->dma,
 573                                                  adapter->rx_ps_hdr_size,
 574                                                  PCI_DMA_FROMDEVICE);
 575                         } else {
 576                                 pci_unmap_single(pdev, buffer_info->dma,
 577                                                  adapter->rx_buffer_len,
 578                                                  PCI_DMA_FROMDEVICE);
 579                         }
 580                         buffer_info->dma = 0;
 581                 }
 582
 583                 if (buffer_info->skb) {
 584                         dev_kfree_skb(buffer_info->skb);
 585                         buffer_info->skb = NULL;
 586                 }
 587
 588                 if (buffer_info->page) {
 589                         if (buffer_info->page_dma)
 590                                 pci_unmap_page(pdev, buffer_info->page_dma,
 591                                                PAGE_SIZE / 2,
 592                                                PCI_DMA_FROMDEVICE);
 593                         put_page(buffer_info->page);
 594                         buffer_info->page = NULL;
 595                         buffer_info->page_dma = 0;
 596                         buffer_info->page_offset = 0;
 597                 }
 598         }
 599
 600         size = sizeof(struct igbvf_buffer) * rx_ring->count;
 601         memset(rx_ring->buffer_info, 0, size);
 602
 603         /* Zero out the descriptor ring */
 604         memset(rx_ring->desc, 0, rx_ring->size);
 605
 606         rx_ring->next_to_clean = 0;
 607         rx_ring->next_to_use = 0;
 608
 609         writel(0, adapter->hw.hw_addr + rx_ring->head);
 610         writel(0, adapter->hw.hw_addr + rx_ring->tail);
 611 }
 612
 613 /**
 614  * igbvf_free_rx_resources - Free Rx Resources
 615  * @rx_ring: ring to clean the resources from
 616  *
 617  * Free all receive software resources
 618  **/
 619
 620 void igbvf_free_rx_resources(struct igbvf_ring *rx_ring)
 621 {
 622         struct pci_dev *pdev = rx_ring->adapter->pdev;
 623
 624         igbvf_clean_rx_ring(rx_ring);
 625
 626         vfree(rx_ring->buffer_info);
 627         rx_ring->buffer_info = NULL;
 628
 629         dma_free_coherent(&pdev->dev, rx_ring->size, rx_ring->desc,
 630                           rx_ring->dma);
 631         rx_ring->desc = NULL;
 632 }
 633
 634 /**
 635  * igbvf_update_itr - update the dynamic ITR value based on statistics
 636  * @adapter: pointer to adapter
 637  * @itr_setting: current adapter->itr
 638  * @packets: the number of packets during this measurement interval
 639  * @bytes: the number of bytes during this measurement interval
 640  *
 641  *      Stores a new ITR value based on packets and byte
 642  *      counts during the last interrupt.  The advantage of per interrupt
 643  *      computation is faster updates and more accurate ITR for the current
 644  *      traffic pattern.  Constants in this function were computed
 645  *      based on theoretical maximum wire speed and thresholds were set based
 646  *      on testing data as well as attempting to minimize response time
 647  *      while increasing bulk throughput.  This functionality is controlled
 648  *      by the InterruptThrottleRate module parameter.
 649  **/
 650 static unsigned int igbvf_update_itr(struct igbvf_adapter *adapter,
 651                                      u16 itr_setting, int packets,
 652                                      int bytes)
 653 {
 654         unsigned int retval = itr_setting;
 655
 656         if (packets == 0)
 657                 goto update_itr_done;
 658
 659         switch (itr_setting) {
 660         case lowest_latency:
 661                 /* handle TSO and jumbo frames */
 662                 if (bytes/packets > 8000)
 663                         retval = bulk_latency;
 664                 else if ((packets < 5) && (bytes > 512))
 665                         retval = low_latency;
 666                 break;
 667         case low_latency:  /* 50 usec aka 20000 ints/s */
 668                 if (bytes > 10000) {
 669                         /* this if handles the TSO accounting */
 670                         if (bytes/packets > 8000)
 671                                 retval = bulk_latency;
 672                         else if ((packets < 10) || ((bytes/packets) > 1200))
 673                                 retval = bulk_latency;
 674                         else if ((packets > 35))
 675                                 retval = lowest_latency;
 676                 } else if (bytes/packets > 2000) {
 677                         retval = bulk_latency;
 678                 } else if (packets <= 2 && bytes < 512) {
 679                         retval = lowest_latency;
 680                 }
 681                 break;
 682         case bulk_latency: /* 250 usec aka 4000 ints/s */
 683                 if (bytes > 25000) {
 684                         if (packets > 35)
 685                                 retval = low_latency;
 686                 } else if (bytes < 6000) {
 687                         retval = low_latency;
 688                 }
 689                 break;
 690         }
 691
 692 update_itr_done:
 693         return retval;
 694 }
 695
 696 static void igbvf_set_itr(struct igbvf_adapter *adapter)
 697 {
 698         struct e1000_hw *hw = &adapter->hw;
 699         u16 current_itr;
 700         u32 new_itr = adapter->itr;
 701
 702         adapter->tx_itr = igbvf_update_itr(adapter, adapter->tx_itr,
 703                                            adapter->total_tx_packets,
 704                                            adapter->total_tx_bytes);
 705         /* conservative mode (itr 3) eliminates the lowest_latency setting */
 706         if (adapter->itr_setting == 3 && adapter->tx_itr == lowest_latency)
 707                 adapter->tx_itr = low_latency;
 708
 709         adapter->rx_itr = igbvf_update_itr(adapter, adapter->rx_itr,
 710                                            adapter->total_rx_packets,
 711                                            adapter->total_rx_bytes);
 712         /* conservative mode (itr 3) eliminates the lowest_latency setting */
 713         if (adapter->itr_setting == 3 && adapter->rx_itr == lowest_latency)
 714                 adapter->rx_itr = low_latency;
 715
 716         current_itr = max(adapter->rx_itr, adapter->tx_itr);
 717
 718         switch (current_itr) {
 719         /* counts and packets in update_itr are dependent on these numbers */
 720         case lowest_latency:
 721                 new_itr = 70000;
 722                 break;
 723         case low_latency:
 724                 new_itr = 20000; /* aka hwitr = ~200 */
 725                 break;
 726         case bulk_latency:
 727                 new_itr = 4000;
 728                 break;
 729         default:
 730                 break;
 731         }
 732
 733         if (new_itr != adapter->itr) {
 734                 /*
 735                  * this attempts to bias the interrupt rate towards Bulk
 736                  * by adding intermediate steps when interrupt rate is
 737                  * increasing
 738                  */
 739                 new_itr = new_itr > adapter->itr ?
 740                              min(adapter->itr + (new_itr >> 2), new_itr) :
 741                              new_itr;
 742                 adapter->itr = new_itr;
 743                 adapter->rx_ring->itr_val = 1952;
 744
 745                 if (adapter->msix_entries)
 746                         adapter->rx_ring->set_itr = 1;
 747                 else
 748                         ew32(ITR, 1952);
 749         }
 750 }
 751
 752 /**
 753  * igbvf_clean_tx_irq - Reclaim resources after transmit completes
 754  * @adapter: board private structure
 755  * returns true if ring is completely cleaned
 756  **/
 757 static bool igbvf_clean_tx_irq(struct igbvf_ring *tx_ring)
 758 {
 759         struct igbvf_adapter *adapter = tx_ring->adapter;
 760         struct e1000_hw *hw = &adapter->hw;
 761         struct net_device *netdev = adapter->netdev;
 762         struct igbvf_buffer *buffer_info;
 763         struct sk_buff *skb;
 764         union e1000_adv_tx_desc *tx_desc, *eop_desc;
 765         unsigned int total_bytes = 0, total_packets = 0;
 766         unsigned int i, eop, count = 0;
 767         bool cleaned = false;
 768
 769         i = tx_ring->next_to_clean;
 770         eop = tx_ring->buffer_info[i].next_to_watch;
 771         eop_desc = IGBVF_TX_DESC_ADV(*tx_ring, eop);
 772
 773         while ((eop_desc->wb.status & cpu_to_le32(E1000_TXD_STAT_DD)) &&
 774                (count < tx_ring->count)) {
 775                 for (cleaned = false; !cleaned; count++) {
 776                         tx_desc = IGBVF_TX_DESC_ADV(*tx_ring, i);
 777                         buffer_info = &tx_ring->buffer_info[i];
 778                         cleaned = (i == eop);
 779                         skb = buffer_info->skb;
 780
 781                         if (skb) {
 782                                 unsigned int segs, bytecount;
 783
 784                                 /* gso_segs is currently only valid for tcp */
 785                                 segs = skb_shinfo(skb)->gso_segs ?: 1;
 786                                 /* multiply data chunks by size of headers */
 787                                 bytecount = ((segs - 1) * skb_headlen(skb)) +
 788                                             skb->len;
 789                                 total_packets += segs;
 790                                 total_bytes += bytecount;
 791                         }
 792
 793                         igbvf_put_txbuf(adapter, buffer_info);
 794                         tx_desc->wb.status = 0;
 795
 796                         i++;
 797                         if (i == tx_ring->count)
 798                                 i = 0;
 799                 }
 800                 eop = tx_ring->buffer_info[i].next_to_watch;
 801                 eop_desc = IGBVF_TX_DESC_ADV(*tx_ring, eop);
 802         }
 803
 804         tx_ring->next_to_clean = i;
 805
 806         if (unlikely(count &&
 807                      netif_carrier_ok(netdev) &&
 808                      igbvf_desc_unused(tx_ring) >= IGBVF_TX_QUEUE_WAKE)) {
 809                 /* Make sure that anybody stopping the queue after this
 810                  * sees the new next_to_clean.
 811                  */
 812                 smp_mb();
 813                 if (netif_queue_stopped(netdev) &&
 814                     !(test_bit(__IGBVF_DOWN, &adapter->state))) {
 815                         netif_wake_queue(netdev);
 816                         ++adapter->restart_queue;
 817                 }
 818         }
 819
 820         if (adapter->detect_tx_hung) {
 821                 /* Detect a transmit hang in hardware, this serializes the
 822                  * check with the clearing of time_stamp and movement of i */
 823                 adapter->detect_tx_hung = false;
 824                 if (tx_ring->buffer_info[i].time_stamp &&
 825                     time_after(jiffies, tx_ring->buffer_info[i].time_stamp +
 826                                (adapter->tx_timeout_factor * HZ))
 827                     && !(er32(STATUS) & E1000_STATUS_TXOFF)) {
 828
 829                         tx_desc = IGBVF_TX_DESC_ADV(*tx_ring, i);
 830                         /* detected Tx unit hang */
 831                         igbvf_print_tx_hang(adapter);
 832
 833                         netif_stop_queue(netdev);
 834                 }
 835         }
 836         adapter->net_stats.tx_bytes += total_bytes;
 837         adapter->net_stats.tx_packets += total_packets;
 838         return (count < tx_ring->count);
 839 }
 840
 841 static irqreturn_t igbvf_msix_other(int irq, void *data)
 842 {
 843         struct net_device *netdev = data;
 844         struct igbvf_adapter *adapter = netdev_priv(netdev);
 845         struct e1000_hw *hw = &adapter->hw;
 846
 847         adapter->int_counter1++;
 848
 849         netif_carrier_off(netdev);
 850         hw->mac.get_link_status = 1;
 851         if (!test_bit(__IGBVF_DOWN, &adapter->state))
 852                 mod_timer(&adapter->watchdog_timer, jiffies + 1);
 853
 854         ew32(EIMS, adapter->eims_other);
 855
 856         return IRQ_HANDLED;
 857 }
 858
 859 static irqreturn_t igbvf_intr_msix_tx(int irq, void *data)
 860 {
 861         struct net_device *netdev = data;
 862         struct igbvf_adapter *adapter = netdev_priv(netdev);
 863         struct e1000_hw *hw = &adapter->hw;
 864         struct igbvf_ring *tx_ring = adapter->tx_ring;
 865
 866
 867         adapter->total_tx_bytes = 0;
 868         adapter->total_tx_packets = 0;
 869
 870         /* auto mask will automatically reenable the interrupt when we write
 871          * EICS */
 872         if (!igbvf_clean_tx_irq(tx_ring))
 873                 /* Ring was not completely cleaned, so fire another interrupt */
 874                 ew32(EICS, tx_ring->eims_value);
 875         else
 876                 ew32(EIMS, tx_ring->eims_value);
 877
 878         return IRQ_HANDLED;
 879 }
 880
 881 static irqreturn_t igbvf_intr_msix_rx(int irq, void *data)
 882 {
 883         struct net_device *netdev = data;
 884         struct igbvf_adapter *adapter = netdev_priv(netdev);
 885
 886         adapter->int_counter0++;
 887
 888         /* Write the ITR value calculated at the end of the
 889          * previous interrupt.
 890          */
 891         if (adapter->rx_ring->set_itr) {
 892                 writel(adapter->rx_ring->itr_val,
 893                        adapter->hw.hw_addr + adapter->rx_ring->itr_register);
 894                 adapter->rx_ring->set_itr = 0;
 895         }
 896
 897         if (napi_schedule_prep(&adapter->rx_ring->napi)) {
 898                 adapter->total_rx_bytes = 0;
 899                 adapter->total_rx_packets = 0;
 900                 __napi_schedule(&adapter->rx_ring->napi);
 901         }
 902
 903         return IRQ_HANDLED;
 904 }
 905
 906 #define IGBVF_NO_QUEUE -1
 907
 908 static void igbvf_assign_vector(struct igbvf_adapter *adapter, int rx_queue,
 909                                 int tx_queue, int msix_vector)
 910 {
 911         struct e1000_hw *hw = &adapter->hw;
 912         u32 ivar, index;
 913
 914         /* 82576 uses a table-based method for assigning vectors.
 915            Each queue has a single entry in the table to which we write
 916            a vector number along with a "valid" bit.  Sadly, the layout
 917            of the table is somewhat counterintuitive. */
 918         if (rx_queue > IGBVF_NO_QUEUE) {
 919                 index = (rx_queue >> 1);
 920                 ivar = array_er32(IVAR0, index);
 921                 if (rx_queue & 0x1) {
 922                         /* vector goes into third byte of register */
 923                         ivar = ivar & 0xFF00FFFF;
 924                         ivar |= (msix_vector | E1000_IVAR_VALID) << 16;
 925                 } else {
 926                         /* vector goes into low byte of register */
 927                         ivar = ivar & 0xFFFFFF00;
 928                         ivar |= msix_vector | E1000_IVAR_VALID;
 929                 }
 930                 adapter->rx_ring[rx_queue].eims_value = 1 << msix_vector;
 931                 array_ew32(IVAR0, index, ivar);
 932         }
 933         if (tx_queue > IGBVF_NO_QUEUE) {
 934                 index = (tx_queue >> 1);
 935                 ivar = array_er32(IVAR0, index);
 936                 if (tx_queue & 0x1) {
 937                         /* vector goes into high byte of register */
 938                         ivar = ivar & 0x00FFFFFF;
 939                         ivar |= (msix_vector | E1000_IVAR_VALID) << 24;
 940                 } else {
 941                         /* vector goes into second byte of register */
 942                         ivar = ivar & 0xFFFF00FF;
 943                         ivar |= (msix_vector | E1000_IVAR_VALID) << 8;
 944                 }
 945                 adapter->tx_ring[tx_queue].eims_value = 1 << msix_vector;
 946                 array_ew32(IVAR0, index, ivar);
 947         }
 948 }
 949
 950 /**
 951  * igbvf_configure_msix - Configure MSI-X hardware
 952  *
 953  * igbvf_configure_msix sets up the hardware to properly
 954  * generate MSI-X interrupts.
 955  **/
 956 static void igbvf_configure_msix(struct igbvf_adapter *adapter)
 957 {
 958         u32 tmp;
 959         struct e1000_hw *hw = &adapter->hw;
 960         struct igbvf_ring *tx_ring = adapter->tx_ring;
 961         struct igbvf_ring *rx_ring = adapter->rx_ring;
 962         int vector = 0;
 963
 964         adapter->eims_enable_mask = 0;
 965
 966         igbvf_assign_vector(adapter, IGBVF_NO_QUEUE, 0, vector++);
 967         adapter->eims_enable_mask |= tx_ring->eims_value;
 968         if (tx_ring->itr_val)
 969                 writel(tx_ring->itr_val,
 970                        hw->hw_addr + tx_ring->itr_register);
 971         else
 972                 writel(1952, hw->hw_addr + tx_ring->itr_register);
 973
 974         igbvf_assign_vector(adapter, 0, IGBVF_NO_QUEUE, vector++);
 975         adapter->eims_enable_mask |= rx_ring->eims_value;
 976         if (rx_ring->itr_val)
 977                 writel(rx_ring->itr_val,
 978                        hw->hw_addr + rx_ring->itr_register);
 979         else
 980                 writel(1952, hw->hw_addr + rx_ring->itr_register);
 981
 982         /* set vector for other causes, i.e. link changes */
 983
 984         tmp = (vector++ | E1000_IVAR_VALID);
 985
 986         ew32(IVAR_MISC, tmp);
 987
 988         adapter->eims_enable_mask = (1 << (vector)) - 1;
 989         adapter->eims_other = 1 << (vector - 1);
 990         e1e_flush();
 991 }
 992
 993 void igbvf_reset_interrupt_capability(struct igbvf_adapter *adapter)
 994 {
 995         if (adapter->msix_entries) {
 996                 pci_disable_msix(adapter->pdev);
 997                 kfree(adapter->msix_entries);
 998                 adapter->msix_entries = NULL;
 999         }
1000 }
1001
1002 /**
1003  * igbvf_set_interrupt_capability - set MSI or MSI-X if supported
1004  *
1005  * Attempt to configure interrupts using the best available
1006  * capabilities of the hardware and kernel.
1007  **/
1008 void igbvf_set_interrupt_capability(struct igbvf_adapter *adapter)
1009 {
1010         int err = -ENOMEM;
1011         int i;
1012
1013         /* we allocate 3 vectors, 1 for tx, 1 for rx, one for pf messages */
1014         adapter->msix_entries = kcalloc(3, sizeof(struct msix_entry),
1015                                         GFP_KERNEL);
1016         if (adapter->msix_entries) {
1017                 for (i = 0; i < 3; i++)
1018                         adapter->msix_entries[i].entry = i;
1019
1020                 err = pci_enable_msix(adapter->pdev,
1021                                       adapter->msix_entries, 3);
1022         }
1023
1024         if (err) {
1025                 /* MSI-X failed */
1026                 dev_err(&adapter->pdev->dev,
1027                         "Failed to initialize MSI-X interrupts.\n");
1028                 igbvf_reset_interrupt_capability(adapter);
1029         }
1030 }
1031
1032 /**
1033  * igbvf_request_msix - Initialize MSI-X interrupts
1034  *
1035  * igbvf_request_msix allocates MSI-X vectors and requests interrupts from the
1036  * kernel.
1037  **/
1038 static int igbvf_request_msix(struct igbvf_adapter *adapter)
1039 {
1040         struct net_device *netdev = adapter->netdev;
1041         int err = 0, vector = 0;
1042
1043         if (strlen(netdev->name) < (IFNAMSIZ - 5)) {
1044                 sprintf(adapter->tx_ring->name, "%s-tx-0", netdev->name);
1045                 sprintf(adapter->rx_ring->name, "%s-rx-0", netdev->name);
1046         } else {
1047                 memcpy(adapter->tx_ring->name, netdev->name, IFNAMSIZ);
1048                 memcpy(adapter->rx_ring->name, netdev->name, IFNAMSIZ);
1049         }
1050
1051         err = request_irq(adapter->msix_entries[vector].vector,
1052                           &igbvf_intr_msix_tx, 0, adapter->tx_ring->name,
1053                           netdev);
1054         if (err)
1055                 goto out;
1056
1057         adapter->tx_ring->itr_register = E1000_EITR(vector);
1058         adapter->tx_ring->itr_val = 1952;
1059         vector++;
1060
1061         err = request_irq(adapter->msix_entries[vector].vector,
1062                           &igbvf_intr_msix_rx, 0, adapter->rx_ring->name,
1063                           netdev);
1064         if (err)
1065                 goto out;
1066
1067         adapter->rx_ring->itr_register = E1000_EITR(vector);
1068         adapter->rx_ring->itr_val = 1952;
1069         vector++;
1070
1071         err = request_irq(adapter->msix_entries[vector].vector,
1072                           &igbvf_msix_other, 0, netdev->name, netdev);
1073         if (err)
1074                 goto out;
1075
1076         igbvf_configure_msix(adapter);
1077         return 0;
1078 out:
1079         return err;
1080 }
1081
1082 /**
1083  * igbvf_alloc_queues - Allocate memory for all rings
1084  * @adapter: board private structure to initialize
1085  **/
1086 static int __devinit igbvf_alloc_queues(struct igbvf_adapter *adapter)
1087 {
1088         struct net_device *netdev = adapter->netdev;
1089
1090         adapter->tx_ring = kzalloc(sizeof(struct igbvf_ring), GFP_KERNEL);
1091         if (!adapter->tx_ring)
1092                 return -ENOMEM;
1093
1094         adapter->rx_ring = kzalloc(sizeof(struct igbvf_ring), GFP_KERNEL);
1095         if (!adapter->rx_ring) {
1096                 kfree(adapter->tx_ring);
1097                 return -ENOMEM;
1098         }
1099
1100         netif_napi_add(netdev, &adapter->rx_ring->napi, igbvf_poll, 64);
1101
1102         return 0;
1103 }
1104
1105 /**
1106  * igbvf_request_irq - initialize interrupts
1107  *
1108  * Attempts to configure interrupts using the best available
1109  * capabilities of the hardware and kernel.
1110  **/
1111 static int igbvf_request_irq(struct igbvf_adapter *adapter)
1112 {
1113         int err = -1;
1114
1115         /* igbvf supports msi-x only */
1116         if (adapter->msix_entries)
1117                 err = igbvf_request_msix(adapter);
1118
1119         if (!err)
1120                 return err;
1121
1122         dev_err(&adapter->pdev->dev,
1123                 "Unable to allocate interrupt, Error: %d\n", err);
1124
1125         return err;
1126 }
1127
1128 static void igbvf_free_irq(struct igbvf_adapter *adapter)
1129 {
1130         struct net_device *netdev = adapter->netdev;
1131         int vector;
1132
1133         if (adapter->msix_entries) {
1134                 for (vector = 0; vector < 3; vector++)
1135                         free_irq(adapter->msix_entries[vector].vector, netdev);
1136         }
1137 }
1138
1139 /**
1140  * igbvf_irq_disable - Mask off interrupt generation on the NIC
1141  **/
1142 static void igbvf_irq_disable(struct igbvf_adapter *adapter)
1143 {
1144         struct e1000_hw *hw = &adapter->hw;
1145
1146         ew32(EIMC, ~0);
1147
1148         if (adapter->msix_entries)
1149                 ew32(EIAC, 0);
1150 }
1151
1152 /**
1153  * igbvf_irq_enable - Enable default interrupt generation settings
1154  **/
1155 static void igbvf_irq_enable(struct igbvf_adapter *adapter)
1156 {
1157         struct e1000_hw *hw = &adapter->hw;
1158
1159         ew32(EIAC, adapter->eims_enable_mask);
1160         ew32(EIAM, adapter->eims_enable_mask);
1161         ew32(EIMS, adapter->eims_enable_mask);
1162 }
1163
1164 /**
1165  * igbvf_poll - NAPI Rx polling callback
1166  * @napi: struct associated with this polling callback
1167  * @budget: amount of packets driver is allowed to process this poll
1168  **/
1169 static int igbvf_poll(struct napi_struct *napi, int budget)
1170 {
1171         struct igbvf_ring *rx_ring = container_of(napi, struct igbvf_ring, napi);
1172         struct igbvf_adapter *adapter = rx_ring->adapter;
1173         struct e1000_hw *hw = &adapter->hw;
1174         int work_done = 0;
1175
1176         igbvf_clean_rx_irq(adapter, &work_done, budget);
1177
1178         /* If not enough Rx work done, exit the polling mode */
1179         if (work_done < budget) {
1180                 napi_complete(napi);
1181
1182                 if (adapter->itr_setting & 3)
1183                         igbvf_set_itr(adapter);
1184
1185                 if (!test_bit(__IGBVF_DOWN, &adapter->state))
1186                         ew32(EIMS, adapter->rx_ring->eims_value);
1187         }
1188
1189         return work_done;
1190 }
1191
1192 /**
1193  * igbvf_set_rlpml - set receive large packet maximum length
1194  * @adapter: board private structure
1195  *
1196  * Configure the maximum size of packets that will be received
1197  */
1198 static void igbvf_set_rlpml(struct igbvf_adapter *adapter)
1199 {
1200         int max_frame_size = adapter->max_frame_size;
1201         struct e1000_hw *hw = &adapter->hw;
1202
1203         if (adapter->vlgrp)
1204                 max_frame_size += VLAN_TAG_SIZE;
1205
1206         e1000_rlpml_set_vf(hw, max_frame_size);
1207 }
1208
1209 static void igbvf_vlan_rx_add_vid(struct net_device *netdev, u16 vid)
1210 {
1211         struct igbvf_adapter *adapter = netdev_priv(netdev);
1212         struct e1000_hw *hw = &adapter->hw;
1213
1214         if (hw->mac.ops.set_vfta(hw, vid, true))
1215                 dev_err(&adapter->pdev->dev, "Failed to add vlan id %d\n", vid);
1216 }
1217
1218 static void igbvf_vlan_rx_kill_vid(struct net_device *netdev, u16 vid)
1219 {
1220         struct igbvf_adapter *adapter = netdev_priv(netdev);
1221         struct e1000_hw *hw = &adapter->hw;
1222
1223         igbvf_irq_disable(adapter);
1224         vlan_group_set_device(adapter->vlgrp, vid, NULL);
1225
1226         if (!test_bit(__IGBVF_DOWN, &adapter->state))
1227                 igbvf_irq_enable(adapter);
1228
1229         if (hw->mac.ops.set_vfta(hw, vid, false))
1230                 dev_err(&adapter->pdev->dev,
1231                         "Failed to remove vlan id %d\n", vid);
1232 }
1233
1234 static void igbvf_vlan_rx_register(struct net_device *netdev,
1235                                    struct vlan_group *grp)
1236 {
1237         struct igbvf_adapter *adapter = netdev_priv(netdev);
1238
1239         adapter->vlgrp = grp;
1240 }
1241
1242 static void igbvf_restore_vlan(struct igbvf_adapter *adapter)
1243 {
1244         u16 vid;
1245
1246         if (!adapter->vlgrp)
1247                 return;
1248
1249         for (vid = 0; vid < VLAN_GROUP_ARRAY_LEN; vid++) {
1250                 if (!vlan_group_get_device(adapter->vlgrp, vid))
1251                         continue;
1252                 igbvf_vlan_rx_add_vid(adapter->netdev, vid);
1253         }
1254
1255         igbvf_set_rlpml(adapter);
1256 }
1257
1258 /**
1259  * igbvf_configure_tx - Configure Transmit Unit after Reset
1260  * @adapter: board private structure
1261  *
1262  * Configure the Tx unit of the MAC after a reset.
1263  **/
1264 static void igbvf_configure_tx(struct igbvf_adapter *adapter)
1265 {
1266         struct e1000_hw *hw = &adapter->hw;
1267         struct igbvf_ring *tx_ring = adapter->tx_ring;
1268         u64 tdba;
1269         u32 txdctl, dca_txctrl;
1270
1271         /* disable transmits */
1272         txdctl = er32(TXDCTL(0));
1273         ew32(TXDCTL(0), txdctl & ~E1000_TXDCTL_QUEUE_ENABLE);
1274         msleep(10);
1275
1276         /* Setup the HW Tx Head and Tail descriptor pointers */
1277         ew32(TDLEN(0), tx_ring->count * sizeof(union e1000_adv_tx_desc));
1278         tdba = tx_ring->dma;
1279         ew32(TDBAL(0), (tdba & DMA_32BIT_MASK));
1280         ew32(TDBAH(0), (tdba >> 32));
1281         ew32(TDH(0), 0);
1282         ew32(TDT(0), 0);
1283         tx_ring->head = E1000_TDH(0);
1284         tx_ring->tail = E1000_TDT(0);
1285
1286         /* Turn off Relaxed Ordering on head write-backs.  The writebacks
1287          * MUST be delivered in order or it will completely screw up
1288          * our bookeeping.
1289          */
1290         dca_txctrl = er32(DCA_TXCTRL(0));
1291         dca_txctrl &= ~E1000_DCA_TXCTRL_TX_WB_RO_EN;
1292         ew32(DCA_TXCTRL(0), dca_txctrl);
1293
1294         /* enable transmits */
1295         txdctl |= E1000_TXDCTL_QUEUE_ENABLE;
1296         ew32(TXDCTL(0), txdctl);
1297
1298         /* Setup Transmit Descriptor Settings for eop descriptor */
1299         adapter->txd_cmd = E1000_ADVTXD_DCMD_EOP | E1000_ADVTXD_DCMD_IFCS;
1300
1301         /* enable Report Status bit */
1302         adapter->txd_cmd |= E1000_ADVTXD_DCMD_RS;
1303
1304         adapter->tx_queue_len = adapter->netdev->tx_queue_len;
1305 }
1306
1307 /**
1308  * igbvf_setup_srrctl - configure the receive control registers
1309  * @adapter: Board private structure
1310  **/
1311 static void igbvf_setup_srrctl(struct igbvf_adapter *adapter)
1312 {
1313         struct e1000_hw *hw = &adapter->hw;
1314         u32 srrctl = 0;
1315
1316         srrctl &= ~(E1000_SRRCTL_DESCTYPE_MASK |
1317                     E1000_SRRCTL_BSIZEHDR_MASK |
1318                     E1000_SRRCTL_BSIZEPKT_MASK);
1319
1320         /* Enable queue drop to avoid head of line blocking */
1321         srrctl |= E1000_SRRCTL_DROP_EN;
1322
1323         /* Setup buffer sizes */
1324         srrctl |= ALIGN(adapter->rx_buffer_len, 1024) >>
1325                   E1000_SRRCTL_BSIZEPKT_SHIFT;
1326
1327         if (adapter->rx_buffer_len < 2048) {
1328                 adapter->rx_ps_hdr_size = 0;
1329                 srrctl |= E1000_SRRCTL_DESCTYPE_ADV_ONEBUF;
1330         } else {
1331                 adapter->rx_ps_hdr_size = 128;
1332                 srrctl |= adapter->rx_ps_hdr_size <<
1333                           E1000_SRRCTL_BSIZEHDRSIZE_SHIFT;
1334                 srrctl |= E1000_SRRCTL_DESCTYPE_HDR_SPLIT_ALWAYS;
1335         }
1336
1337         ew32(SRRCTL(0), srrctl);
1338 }
1339
1340 /**
1341  * igbvf_configure_rx - Configure Receive Unit after Reset
1342  * @adapter: board private structure
1343  *
1344  * Configure the Rx unit of the MAC after a reset.
1345  **/
1346 static void igbvf_configure_rx(struct igbvf_adapter *adapter)
1347 {
1348         struct e1000_hw *hw = &adapter->hw;
1349         struct igbvf_ring *rx_ring = adapter->rx_ring;
1350         u64 rdba;
1351         u32 rdlen, rxdctl;
1352
1353         /* disable receives */
1354         rxdctl = er32(RXDCTL(0));
1355         ew32(RXDCTL(0), rxdctl & ~E1000_RXDCTL_QUEUE_ENABLE);
1356         msleep(10);
1357
1358         rdlen = rx_ring->count * sizeof(union e1000_adv_rx_desc);
1359
1360         /*
1361          * Setup the HW Rx Head and Tail Descriptor Pointers and
1362          * the Base and Length of the Rx Descriptor Ring
1363          */
1364         rdba = rx_ring->dma;
1365         ew32(RDBAL(0), (rdba & DMA_32BIT_MASK));
1366         ew32(RDBAH(0), (rdba >> 32));
1367         ew32(RDLEN(0), rx_ring->count * sizeof(union e1000_adv_rx_desc));
1368         rx_ring->head = E1000_RDH(0);
1369         rx_ring->tail = E1000_RDT(0);
1370         ew32(RDH(0), 0);
1371         ew32(RDT(0), 0);
1372
1373         rxdctl |= E1000_RXDCTL_QUEUE_ENABLE;
1374         rxdctl &= 0xFFF00000;
1375         rxdctl |= IGBVF_RX_PTHRESH;
1376         rxdctl |= IGBVF_RX_HTHRESH << 8;
1377         rxdctl |= IGBVF_RX_WTHRESH << 16;
1378
1379         igbvf_set_rlpml(adapter);
1380
1381         /* enable receives */
1382         ew32(RXDCTL(0), rxdctl);
1383 }
1384
1385 /**
1386  * igbvf_set_multi - Multicast and Promiscuous mode set
1387  * @netdev: network interface device structure
1388  *
1389  * The set_multi entry point is called whenever the multicast address
1390  * list or the network interface flags are updated.  This routine is
1391  * responsible for configuring the hardware for proper multicast,
1392  * promiscuous mode, and all-multi behavior.
1393  **/
1394 static void igbvf_set_multi(struct net_device *netdev)
1395 {
1396         struct igbvf_adapter *adapter = netdev_priv(netdev);
1397         struct e1000_hw *hw = &adapter->hw;
1398         struct dev_mc_list *mc_ptr;
1399         u8  *mta_list = NULL;
1400         int i;
1401
1402         if (netdev->mc_count) {
1403                 mta_list = kmalloc(netdev->mc_count * 6, GFP_ATOMIC);
1404                 if (!mta_list) {
1405                         dev_err(&adapter->pdev->dev,
1406                                 "failed to allocate multicast filter list\n");
1407                         return;
1408                 }
1409         }
1410
1411         /* prepare a packed array of only addresses. */
1412         mc_ptr = netdev->mc_list;
1413
1414         for (i = 0; i < netdev->mc_count; i++) {
1415                 if (!mc_ptr)
1416                         break;
1417                 memcpy(mta_list + (i*ETH_ALEN), mc_ptr->dmi_addr,
1418                        ETH_ALEN);
1419                 mc_ptr = mc_ptr->next;
1420         }
1421
1422         hw->mac.ops.update_mc_addr_list(hw, mta_list, i, 0, 0);
1423         kfree(mta_list);
1424 }
1425
1426 /**
1427  * igbvf_configure - configure the hardware for Rx and Tx
1428  * @adapter: private board structure
1429  **/
1430 static void igbvf_configure(struct igbvf_adapter *adapter)
1431 {
1432         igbvf_set_multi(adapter->netdev);
1433
1434         igbvf_restore_vlan(adapter);
1435
1436         igbvf_configure_tx(adapter);
1437         igbvf_setup_srrctl(adapter);
1438         igbvf_configure_rx(adapter);
1439         igbvf_alloc_rx_buffers(adapter->rx_ring,
1440                                igbvf_desc_unused(adapter->rx_ring));
1441 }
1442
1443 /* igbvf_reset - bring the hardware into a known good state
1444  *
1445  * This function boots the hardware and enables some settings that
1446  * require a configuration cycle of the hardware - those cannot be
1447  * set/changed during runtime. After reset the device needs to be
1448  * properly configured for Rx, Tx etc.
1449  */
1450 void igbvf_reset(struct igbvf_adapter *adapter)
1451 {
1452         struct e1000_mac_info *mac = &adapter->hw.mac;
1453         struct net_device *netdev = adapter->netdev;
1454         struct e1000_hw *hw = &adapter->hw;
1455
1456         /* Allow time for pending master requests to run */
1457         if (mac->ops.reset_hw(hw))
1458                 dev_err(&adapter->pdev->dev, "PF still resetting\n");
1459
1460         mac->ops.init_hw(hw);
1461
1462         if (is_valid_ether_addr(adapter->hw.mac.addr)) {
1463                 memcpy(netdev->dev_addr, adapter->hw.mac.addr,
1464                        netdev->addr_len);
1465                 memcpy(netdev->perm_addr, adapter->hw.mac.addr,
1466                        netdev->addr_len);
1467         }
1468 }
1469
1470 int igbvf_up(struct igbvf_adapter *adapter)
1471 {
1472         struct e1000_hw *hw = &adapter->hw;
1473
1474         /* hardware has been reset, we need to reload some things */
1475         igbvf_configure(adapter);
1476
1477         clear_bit(__IGBVF_DOWN, &adapter->state);
1478
1479         napi_enable(&adapter->rx_ring->napi);
1480         if (adapter->msix_entries)
1481                 igbvf_configure_msix(adapter);
1482
1483         /* Clear any pending interrupts. */
1484         er32(EICR);
1485         igbvf_irq_enable(adapter);
1486
1487         /* start the watchdog */
1488         hw->mac.get_link_status = 1;
1489         mod_timer(&adapter->watchdog_timer, jiffies + 1);
1490
1491
1492         return 0;
1493 }
1494
1495 void igbvf_down(struct igbvf_adapter *adapter)
1496 {
1497         struct net_device *netdev = adapter->netdev;
1498         struct e1000_hw *hw = &adapter->hw;
1499         u32 rxdctl, txdctl;
1500
1501         /*
1502          * signal that we're down so the interrupt handler does not
1503          * reschedule our watchdog timer
1504          */
1505         set_bit(__IGBVF_DOWN, &adapter->state);
1506
1507         /* disable receives in the hardware */
1508         rxdctl = er32(RXDCTL(0));
1509         ew32(RXDCTL(0), rxdctl & ~E1000_RXDCTL_QUEUE_ENABLE);
1510
1511         netif_stop_queue(netdev);
1512
1513         /* disable transmits in the hardware */
1514         txdctl = er32(TXDCTL(0));
1515         ew32(TXDCTL(0), txdctl & ~E1000_TXDCTL_QUEUE_ENABLE);
1516
1517         /* flush both disables and wait for them to finish */
1518         e1e_flush();
1519         msleep(10);
1520
1521         napi_disable(&adapter->rx_ring->napi);
1522
1523         igbvf_irq_disable(adapter);
1524
1525         del_timer_sync(&adapter->watchdog_timer);
1526
1527         netdev->tx_queue_len = adapter->tx_queue_len;
1528         netif_carrier_off(netdev);
1529
1530         /* record the stats before reset*/
1531         igbvf_update_stats(adapter);
1532
1533         adapter->link_speed = 0;
1534         adapter->link_duplex = 0;
1535
1536         igbvf_reset(adapter);
1537         igbvf_clean_tx_ring(adapter->tx_ring);
1538         igbvf_clean_rx_ring(adapter->rx_ring);
1539 }
1540
1541 void igbvf_reinit_locked(struct igbvf_adapter *adapter)
1542 {
1543         might_sleep();
1544         while (test_and_set_bit(__IGBVF_RESETTING, &adapter->state))
1545                 msleep(1);
1546         igbvf_down(adapter);
1547         igbvf_up(adapter);
1548         clear_bit(__IGBVF_RESETTING, &adapter->state);
1549 }
1550
1551 /**
1552  * igbvf_sw_init - Initialize general software structures (struct igbvf_adapter)
1553  * @adapter: board private structure to initialize
1554  *
1555  * igbvf_sw_init initializes the Adapter private data structure.
1556  * Fields are initialized based on PCI device information and
1557  * OS network device settings (MTU size).
1558  **/
1559 static int __devinit igbvf_sw_init(struct igbvf_adapter *adapter)
1560 {
1561         struct net_device *netdev = adapter->netdev;
1562         s32 rc;
1563
1564         adapter->rx_buffer_len = ETH_FRAME_LEN + VLAN_HLEN + ETH_FCS_LEN;
1565         adapter->rx_ps_hdr_size = 0;
1566         adapter->max_frame_size = netdev->mtu + ETH_HLEN + ETH_FCS_LEN;
1567         adapter->min_frame_size = ETH_ZLEN + ETH_FCS_LEN;
1568
1569         adapter->tx_int_delay = 8;
1570         adapter->tx_abs_int_delay = 32;
1571         adapter->rx_int_delay = 0;
1572         adapter->rx_abs_int_delay = 8;
1573         adapter->itr_setting = 3;
1574         adapter->itr = 20000;
1575
1576         /* Set various function pointers */
1577         adapter->ei->init_ops(&adapter->hw);
1578
1579         rc = adapter->hw.mac.ops.init_params(&adapter->hw);
1580         if (rc)
1581                 return rc;
1582
1583         rc = adapter->hw.mbx.ops.init_params(&adapter->hw);
1584         if (rc)
1585                 return rc;
1586
1587         igbvf_set_interrupt_capability(adapter);
1588
1589         if (igbvf_alloc_queues(adapter))
1590                 return -ENOMEM;
1591
1592         spin_lock_init(&adapter->tx_queue_lock);
1593
1594         /* Explicitly disable IRQ since the NIC can be in any state. */
1595         igbvf_irq_disable(adapter);
1596
1597         spin_lock_init(&adapter->stats_lock);
1598
1599         set_bit(__IGBVF_DOWN, &adapter->state);
1600         return 0;
1601 }
1602
1603 static void igbvf_initialize_last_counter_stats(struct igbvf_adapter *adapter)
1604 {
1605         struct e1000_hw *hw = &adapter->hw;
1606
1607         adapter->stats.last_gprc = er32(VFGPRC);
1608         adapter->stats.last_gorc = er32(VFGORC);
1609         adapter->stats.last_gptc = er32(VFGPTC);
1610         adapter->stats.last_gotc = er32(VFGOTC);
1611         adapter->stats.last_mprc = er32(VFMPRC);
1612         adapter->stats.last_gotlbc = er32(VFGOTLBC);
1613         adapter->stats.last_gptlbc = er32(VFGPTLBC);
1614         adapter->stats.last_gorlbc = er32(VFGORLBC);
1615         adapter->stats.last_gprlbc = er32(VFGPRLBC);
1616
1617         adapter->stats.base_gprc = er32(VFGPRC);
1618         adapter->stats.base_gorc = er32(VFGORC);
1619         adapter->stats.base_gptc = er32(VFGPTC);
1620         adapter->stats.base_gotc = er32(VFGOTC);
1621         adapter->stats.base_mprc = er32(VFMPRC);
1622         adapter->stats.base_gotlbc = er32(VFGOTLBC);
1623         adapter->stats.base_gptlbc = er32(VFGPTLBC);
1624         adapter->stats.base_gorlbc = er32(VFGORLBC);
1625         adapter->stats.base_gprlbc = er32(VFGPRLBC);
1626 }
1627
1628 /**
1629  * igbvf_open - Called when a network interface is made active
1630  * @netdev: network interface device structure
1631  *
1632  * Returns 0 on success, negative value on failure
1633  *
1634  * The open entry point is called when a network interface is made
1635  * active by the system (IFF_UP).  At this point all resources needed
1636  * for transmit and receive operations are allocated, the interrupt
1637  * handler is registered with the OS, the watchdog timer is started,
1638  * and the stack is notified that the interface is ready.
1639  **/
1640 static int igbvf_open(struct net_device *netdev)
1641 {
1642         struct igbvf_adapter *adapter = netdev_priv(netdev);
1643         struct e1000_hw *hw = &adapter->hw;
1644         int err;
1645
1646         /* disallow open during test */
1647         if (test_bit(__IGBVF_TESTING, &adapter->state))
1648                 return -EBUSY;
1649
1650         /* allocate transmit descriptors */
1651         err = igbvf_setup_tx_resources(adapter, adapter->tx_ring);
1652         if (err)
1653                 goto err_setup_tx;
1654
1655         /* allocate receive descriptors */
1656         err = igbvf_setup_rx_resources(adapter, adapter->rx_ring);
1657         if (err)
1658                 goto err_setup_rx;
1659
1660         /*
1661          * before we allocate an interrupt, we must be ready to handle it.
1662          * Setting DEBUG_SHIRQ in the kernel makes it fire an interrupt
1663          * as soon as we call pci_request_irq, so we have to setup our
1664          * clean_rx handler before we do so.
1665          */
1666         igbvf_configure(adapter);
1667
1668         err = igbvf_request_irq(adapter);
1669         if (err)
1670                 goto err_req_irq;
1671
1672         /* From here on the code is the same as igbvf_up() */
1673         clear_bit(__IGBVF_DOWN, &adapter->state);
1674
1675         napi_enable(&adapter->rx_ring->napi);
1676
1677         /* clear any pending interrupts */
1678         er32(EICR);
1679
1680         igbvf_irq_enable(adapter);
1681
1682         /* start the watchdog */
1683         hw->mac.get_link_status = 1;
1684         mod_timer(&adapter->watchdog_timer, jiffies + 1);
1685
1686         return 0;
1687
1688 err_req_irq:
1689         igbvf_free_rx_resources(adapter->rx_ring);
1690 err_setup_rx:
1691         igbvf_free_tx_resources(adapter->tx_ring);
1692 err_setup_tx:
1693         igbvf_reset(adapter);
1694
1695         return err;
1696 }
1697
1698 /**
1699  * igbvf_close - Disables a network interface
1700  * @netdev: network interface device structure
1701  *
1702  * Returns 0, this is not allowed to fail
1703  *
1704  * The close entry point is called when an interface is de-activated
1705  * by the OS.  The hardware is still under the drivers control, but
1706  * needs to be disabled.  A global MAC reset is issued to stop the
1707  * hardware, and all transmit and receive resources are freed.
1708  **/
1709 static int igbvf_close(struct net_device *netdev)
1710 {
1711         struct igbvf_adapter *adapter = netdev_priv(netdev);
1712
1713         WARN_ON(test_bit(__IGBVF_RESETTING, &adapter->state));
1714         igbvf_down(adapter);
1715
1716         igbvf_free_irq(adapter);
1717
1718         igbvf_free_tx_resources(adapter->tx_ring);
1719         igbvf_free_rx_resources(adapter->rx_ring);
1720
1721         return 0;
1722 }
1723 /**
1724  * igbvf_set_mac - Change the Ethernet Address of the NIC
1725  * @netdev: network interface device structure
1726  * @p: pointer to an address structure
1727  *
1728  * Returns 0 on success, negative on failure
1729  **/
1730 static int igbvf_set_mac(struct net_device *netdev, void *p)
1731 {
1732         struct igbvf_adapter *adapter = netdev_priv(netdev);
1733         struct e1000_hw *hw = &adapter->hw;
1734         struct sockaddr *addr = p;
1735
1736         if (!is_valid_ether_addr(addr->sa_data))
1737                 return -EADDRNOTAVAIL;
1738
1739         memcpy(hw->mac.addr, addr->sa_data, netdev->addr_len);
1740
1741         hw->mac.ops.rar_set(hw, hw->mac.addr, 0);
1742
1743         if (memcmp(addr->sa_data, hw->mac.addr, 6))
1744                 return -EADDRNOTAVAIL;
1745
1746         memcpy(netdev->dev_addr, addr->sa_data, netdev->addr_len);
1747
1748         return 0;
1749 }
1750
1751 #define UPDATE_VF_COUNTER(reg, name)                                    \
1752         {                                                               \
1753                 u32 current_counter = er32(reg);                        \
1754                 if (current_counter < adapter->stats.last_##name)       \
1755                         adapter->stats.name += 0x100000000LL;           \
1756                 adapter->stats.last_##name = current_counter;           \
1757                 adapter->stats.name &= 0xFFFFFFFF00000000LL;            \
1758                 adapter->stats.name |= current_counter;                 \
1759         }
1760
1761 /**
1762  * igbvf_update_stats - Update the board statistics counters
1763  * @adapter: board private structure
1764 **/
1765 void igbvf_update_stats(struct igbvf_adapter *adapter)
1766 {
1767         struct e1000_hw *hw = &adapter->hw;
1768         struct pci_dev *pdev = adapter->pdev;
1769
1770         /*
1771          * Prevent stats update while adapter is being reset, link is down
1772          * or if the pci connection is down.
1773          */
1774         if (adapter->link_speed == 0)
1775                 return;
1776
1777         if (test_bit(__IGBVF_RESETTING, &adapter->state))
1778                 return;
1779
1780         if (pci_channel_offline(pdev))
1781                 return;
1782
1783         UPDATE_VF_COUNTER(VFGPRC, gprc);
1784         UPDATE_VF_COUNTER(VFGORC, gorc);
1785         UPDATE_VF_COUNTER(VFGPTC, gptc);
1786         UPDATE_VF_COUNTER(VFGOTC, gotc);
1787         UPDATE_VF_COUNTER(VFMPRC, mprc);
1788         UPDATE_VF_COUNTER(VFGOTLBC, gotlbc);
1789         UPDATE_VF_COUNTER(VFGPTLBC, gptlbc);
1790         UPDATE_VF_COUNTER(VFGORLBC, gorlbc);
1791         UPDATE_VF_COUNTER(VFGPRLBC, gprlbc);
1792
1793         /* Fill out the OS statistics structure */
1794         adapter->net_stats.multicast = adapter->stats.mprc;
1795 }
1796
1797 static void igbvf_print_link_info(struct igbvf_adapter *adapter)
1798 {
1799         dev_info(&adapter->pdev->dev, "Link is Up %d Mbps %s\n",
1800                  adapter->link_speed,
1801                  ((adapter->link_duplex == FULL_DUPLEX) ?
1802                   "Full Duplex" : "Half Duplex"));
1803 }
1804
1805 static bool igbvf_has_link(struct igbvf_adapter *adapter)
1806 {
1807         struct e1000_hw *hw = &adapter->hw;
1808         s32 ret_val = E1000_SUCCESS;
1809         bool link_active;
1810
1811         ret_val = hw->mac.ops.check_for_link(hw);
1812         link_active = !hw->mac.get_link_status;
1813
1814         /* if check for link returns error we will need to reset */
1815         if (ret_val)
1816                 schedule_work(&adapter->reset_task);
1817
1818         return link_active;
1819 }
1820
1821 /**
1822  * igbvf_watchdog - Timer Call-back
1823  * @data: pointer to adapter cast into an unsigned long
1824  **/
1825 static void igbvf_watchdog(unsigned long data)
1826 {
1827         struct igbvf_adapter *adapter = (struct igbvf_adapter *) data;
1828
1829         /* Do the rest outside of interrupt context */
1830         schedule_work(&adapter->watchdog_task);
1831 }
1832
1833 static void igbvf_watchdog_task(struct work_struct *work)
1834 {
1835         struct igbvf_adapter *adapter = container_of(work,
1836                                                      struct igbvf_adapter,
1837                                                      watchdog_task);
1838         struct net_device *netdev = adapter->netdev;
1839         struct e1000_mac_info *mac = &adapter->hw.mac;
1840         struct igbvf_ring *tx_ring = adapter->tx_ring;
1841         struct e1000_hw *hw = &adapter->hw;
1842         u32 link;
1843         int tx_pending = 0;
1844
1845         link = igbvf_has_link(adapter);
1846
1847         if (link) {
1848                 if (!netif_carrier_ok(netdev)) {
1849                         bool txb2b = 1;
1850
1851                         mac->ops.get_link_up_info(&adapter->hw,
1852                                                   &adapter->link_speed,
1853                                                   &adapter->link_duplex);
1854                         igbvf_print_link_info(adapter);
1855
1856                         /*
1857                          * tweak tx_queue_len according to speed/duplex
1858                          * and adjust the timeout factor
1859                          */
1860                         netdev->tx_queue_len = adapter->tx_queue_len;
1861                         adapter->tx_timeout_factor = 1;
1862                         switch (adapter->link_speed) {
1863                         case SPEED_10:
1864                                 txb2b = 0;
1865                                 netdev->tx_queue_len = 10;
1866                                 adapter->tx_timeout_factor = 16;
1867                                 break;
1868                         case SPEED_100:
1869                                 txb2b = 0;
1870                                 netdev->tx_queue_len = 100;
1871                                 /* maybe add some timeout factor ? */
1872                                 break;
1873                         }
1874
1875                         netif_carrier_on(netdev);
1876                         netif_wake_queue(netdev);
1877                 }
1878         } else {
1879                 if (netif_carrier_ok(netdev)) {
1880                         adapter->link_speed = 0;
1881                         adapter->link_duplex = 0;
1882                         dev_info(&adapter->pdev->dev, "Link is Down\n");
1883                         netif_carrier_off(netdev);
1884                         netif_stop_queue(netdev);
1885                 }
1886         }
1887
1888         if (netif_carrier_ok(netdev)) {
1889                 igbvf_update_stats(adapter);
1890         } else {
1891                 tx_pending = (igbvf_desc_unused(tx_ring) + 1 <
1892                               tx_ring->count);
1893                 if (tx_pending) {
1894                         /*
1895                          * We've lost link, so the controller stops DMA,
1896                          * but we've got queued Tx work that's never going
1897                          * to get done, so reset controller to flush Tx.
1898                          * (Do the reset outside of interrupt context).
1899                          */
1900                         adapter->tx_timeout_count++;
1901                         schedule_work(&adapter->reset_task);
1902                 }
1903         }
1904
1905         /* Cause software interrupt to ensure Rx ring is cleaned */
1906         ew32(EICS, adapter->rx_ring->eims_value);
1907
1908         /* Force detection of hung controller every watchdog period */
1909         adapter->detect_tx_hung = 1;
1910
1911         /* Reset the timer */
1912         if (!test_bit(__IGBVF_DOWN, &adapter->state))
1913                 mod_timer(&adapter->watchdog_timer,
1914                           round_jiffies(jiffies + (2 * HZ)));
1915 }
1916
1917 #define IGBVF_TX_FLAGS_CSUM             0x00000001
1918 #define IGBVF_TX_FLAGS_VLAN             0x00000002
1919 #define IGBVF_TX_FLAGS_TSO              0x00000004
1920 #define IGBVF_TX_FLAGS_IPV4             0x00000008
1921 #define IGBVF_TX_FLAGS_VLAN_MASK        0xffff0000
1922 #define IGBVF_TX_FLAGS_VLAN_SHIFT       16
1923
1924 static int igbvf_tso(struct igbvf_adapter *adapter,
1925                      struct igbvf_ring *tx_ring,
1926                      struct sk_buff *skb, u32 tx_flags, u8 *hdr_len)
1927 {
1928         struct e1000_adv_tx_context_desc *context_desc;
1929         unsigned int i;
1930         int err;
1931         struct igbvf_buffer *buffer_info;
1932         u32 info = 0, tu_cmd = 0;
1933         u32 mss_l4len_idx, l4len;
1934         *hdr_len = 0;
1935
1936         if (skb_header_cloned(skb)) {
1937                 err = pskb_expand_head(skb, 0, 0, GFP_ATOMIC);
1938                 if (err) {
1939                         dev_err(&adapter->pdev->dev,
1940                                 "igbvf_tso returning an error\n");
1941                         return err;
1942                 }
1943         }
1944
1945         l4len = tcp_hdrlen(skb);
1946         *hdr_len += l4len;
1947
1948         if (skb->protocol == htons(ETH_P_IP)) {
1949                 struct iphdr *iph = ip_hdr(skb);
1950                 iph->tot_len = 0;
1951                 iph->check = 0;
1952                 tcp_hdr(skb)->check = ~csum_tcpudp_magic(iph->saddr,
1953                                                          iph->daddr, 0,
1954                                                          IPPROTO_TCP,
1955                                                          0);
1956         } else if (skb_shinfo(skb)->gso_type == SKB_GSO_TCPV6) {
1957                 ipv6_hdr(skb)->payload_len = 0;
1958                 tcp_hdr(skb)->check = ~csum_ipv6_magic(&ipv6_hdr(skb)->saddr,
1959                                                        &ipv6_hdr(skb)->daddr,
1960                                                        0, IPPROTO_TCP, 0);
1961         }
1962
1963         i = tx_ring->next_to_use;
1964
1965         buffer_info = &tx_ring->buffer_info[i];
1966         context_desc = IGBVF_TX_CTXTDESC_ADV(*tx_ring, i);
1967         /* VLAN MACLEN IPLEN */
1968         if (tx_flags & IGBVF_TX_FLAGS_VLAN)
1969                 info |= (tx_flags & IGBVF_TX_FLAGS_VLAN_MASK);
1970         info |= (skb_network_offset(skb) << E1000_ADVTXD_MACLEN_SHIFT);
1971         *hdr_len += skb_network_offset(skb);
1972         info |= (skb_transport_header(skb) - skb_network_header(skb));
1973         *hdr_len += (skb_transport_header(skb) - skb_network_header(skb));
1974         context_desc->vlan_macip_lens = cpu_to_le32(info);
1975
1976         /* ADV DTYP TUCMD MKRLOC/ISCSIHEDLEN */
1977         tu_cmd |= (E1000_TXD_CMD_DEXT | E1000_ADVTXD_DTYP_CTXT);
1978
1979         if (skb->protocol == htons(ETH_P_IP))
1980                 tu_cmd |= E1000_ADVTXD_TUCMD_IPV4;
1981         tu_cmd |= E1000_ADVTXD_TUCMD_L4T_TCP;
1982
1983         context_desc->type_tucmd_mlhl = cpu_to_le32(tu_cmd);
1984
1985         /* MSS L4LEN IDX */
1986         mss_l4len_idx = (skb_shinfo(skb)->gso_size << E1000_ADVTXD_MSS_SHIFT);
1987         mss_l4len_idx |= (l4len << E1000_ADVTXD_L4LEN_SHIFT);
1988
1989         context_desc->mss_l4len_idx = cpu_to_le32(mss_l4len_idx);
1990         context_desc->seqnum_seed = 0;
1991
1992         buffer_info->time_stamp = jiffies;
1993         buffer_info->next_to_watch = i;
1994         buffer_info->dma = 0;
1995         i++;
1996         if (i == tx_ring->count)
1997                 i = 0;
1998
1999         tx_ring->next_to_use = i;
2000
2001         return true;
2002 }
2003
2004 static inline bool igbvf_tx_csum(struct igbvf_adapter *adapter,
2005                                  struct igbvf_ring *tx_ring,
2006                                  struct sk_buff *skb, u32 tx_flags)
2007 {
2008         struct e1000_adv_tx_context_desc *context_desc;
2009         unsigned int i;
2010         struct igbvf_buffer *buffer_info;
2011         u32 info = 0, tu_cmd = 0;
2012
2013         if ((skb->ip_summed == CHECKSUM_PARTIAL) ||
2014             (tx_flags & IGBVF_TX_FLAGS_VLAN)) {
2015                 i = tx_ring->next_to_use;
2016                 buffer_info = &tx_ring->buffer_info[i];
2017                 context_desc = IGBVF_TX_CTXTDESC_ADV(*tx_ring, i);
2018
2019                 if (tx_flags & IGBVF_TX_FLAGS_VLAN)
2020                         info |= (tx_flags & IGBVF_TX_FLAGS_VLAN_MASK);
2021
2022                 info |= (skb_network_offset(skb) << E1000_ADVTXD_MACLEN_SHIFT);
2023                 if (skb->ip_summed == CHECKSUM_PARTIAL)
2024                         info |= (skb_transport_header(skb) -
2025                                  skb_network_header(skb));
2026
2027
2028                 context_desc->vlan_macip_lens = cpu_to_le32(info);
2029
2030                 tu_cmd |= (E1000_TXD_CMD_DEXT | E1000_ADVTXD_DTYP_CTXT);
2031
2032                 if (skb->ip_summed == CHECKSUM_PARTIAL) {
2033                         switch (skb->protocol) {
2034                         case __constant_htons(ETH_P_IP):
2035                                 tu_cmd |= E1000_ADVTXD_TUCMD_IPV4;
2036                                 if (ip_hdr(skb)->protocol == IPPROTO_TCP)
2037                                         tu_cmd |= E1000_ADVTXD_TUCMD_L4T_TCP;
2038                                 break;
2039                         case __constant_htons(ETH_P_IPV6):
2040                                 if (ipv6_hdr(skb)->nexthdr == IPPROTO_TCP)
2041                                         tu_cmd |= E1000_ADVTXD_TUCMD_L4T_TCP;
2042                                 break;
2043                         default:
2044                                 break;
2045                         }
2046                 }
2047
2048                 context_desc->type_tucmd_mlhl = cpu_to_le32(tu_cmd);
2049                 context_desc->seqnum_seed = 0;
2050                 context_desc->mss_l4len_idx = 0;
2051
2052                 buffer_info->time_stamp = jiffies;
2053                 buffer_info->next_to_watch = i;
2054                 buffer_info->dma = 0;
2055                 i++;
2056                 if (i == tx_ring->count)
2057                         i = 0;
2058                 tx_ring->next_to_use = i;
2059
2060                 return true;
2061         }
2062
2063         return false;
2064 }
2065
2066 static int igbvf_maybe_stop_tx(struct net_device *netdev, int size)
2067 {
2068         struct igbvf_adapter *adapter = netdev_priv(netdev);
2069
2070         /* there is enough descriptors then we don't need to worry  */
2071         if (igbvf_desc_unused(adapter->tx_ring) >= size)
2072                 return 0;
2073
2074         netif_stop_queue(netdev);
2075
2076         smp_mb();
2077
2078         /* We need to check again just in case room has been made available */
2079         if (igbvf_desc_unused(adapter->tx_ring) < size)
2080                 return -EBUSY;
2081
2082         netif_wake_queue(netdev);
2083
2084         ++adapter->restart_queue;
2085         return 0;
2086 }
2087
2088 #define IGBVF_MAX_TXD_PWR       16
2089 #define IGBVF_MAX_DATA_PER_TXD  (1 << IGBVF_MAX_TXD_PWR)
2090
2091 static inline int igbvf_tx_map_adv(struct igbvf_adapter *adapter,
2092                                    struct igbvf_ring *tx_ring,
2093                                    struct sk_buff *skb,
2094                                    unsigned int first)
2095 {
2096         struct igbvf_buffer *buffer_info;
2097         unsigned int len = skb_headlen(skb);
2098         unsigned int count = 0, i;
2099         unsigned int f;
2100         dma_addr_t *map;
2101
2102         i = tx_ring->next_to_use;
2103
2104         if (skb_dma_map(&adapter->pdev->dev, skb, DMA_TO_DEVICE)) {
2105                 dev_err(&adapter->pdev->dev, "TX DMA map failed\n");
2106                 return 0;
2107         }
2108
2109         map = skb_shinfo(skb)->dma_maps;
2110
2111         buffer_info = &tx_ring->buffer_info[i];
2112         BUG_ON(len >= IGBVF_MAX_DATA_PER_TXD);
2113         buffer_info->length = len;
2114         /* set time_stamp *before* dma to help avoid a possible race */
2115         buffer_info->time_stamp = jiffies;
2116         buffer_info->next_to_watch = i;
2117         buffer_info->dma = map[count];
2118         count++;
2119
2120         for (f = 0; f < skb_shinfo(skb)->nr_frags; f++) {
2121                 struct skb_frag_struct *frag;
2122
2123                 i++;
2124                 if (i == tx_ring->count)
2125                         i = 0;
2126
2127                 frag = &skb_shinfo(skb)->frags[f];
2128                 len = frag->size;
2129
2130                 buffer_info = &tx_ring->buffer_info[i];
2131                 BUG_ON(len >= IGBVF_MAX_DATA_PER_TXD);
2132                 buffer_info->length = len;
2133                 buffer_info->time_stamp = jiffies;
2134                 buffer_info->next_to_watch = i;
2135                 buffer_info->dma = map[count];
2136                 count++;
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
2145 static inline void igbvf_tx_queue_adv(struct igbvf_adapter *adapter,
2146                                       struct igbvf_ring *tx_ring,
2147                                       int tx_flags, int count, u32 paylen,
2148                                       u8 hdr_len)
2149 {
2150         union e1000_adv_tx_desc *tx_desc = NULL;
2151         struct igbvf_buffer *buffer_info;
2152         u32 olinfo_status = 0, cmd_type_len;
2153         unsigned int i;
2154
2155         cmd_type_len = (E1000_ADVTXD_DTYP_DATA | E1000_ADVTXD_DCMD_IFCS |
2156                         E1000_ADVTXD_DCMD_DEXT);
2157
2158         if (tx_flags & IGBVF_TX_FLAGS_VLAN)
2159                 cmd_type_len |= E1000_ADVTXD_DCMD_VLE;
2160
2161         if (tx_flags & IGBVF_TX_FLAGS_TSO) {
2162                 cmd_type_len |= E1000_ADVTXD_DCMD_TSE;
2163
2164                 /* insert tcp checksum */
2165                 olinfo_status |= E1000_TXD_POPTS_TXSM << 8;
2166
2167                 /* insert ip checksum */
2168                 if (tx_flags & IGBVF_TX_FLAGS_IPV4)
2169                         olinfo_status |= E1000_TXD_POPTS_IXSM << 8;
2170
2171         } else if (tx_flags & IGBVF_TX_FLAGS_CSUM) {
2172                 olinfo_status |= E1000_TXD_POPTS_TXSM << 8;
2173         }
2174
2175         olinfo_status |= ((paylen - hdr_len) << E1000_ADVTXD_PAYLEN_SHIFT);
2176
2177         i = tx_ring->next_to_use;
2178         while (count--) {
2179                 buffer_info = &tx_ring->buffer_info[i];
2180                 tx_desc = IGBVF_TX_DESC_ADV(*tx_ring, i);
2181                 tx_desc->read.buffer_addr = cpu_to_le64(buffer_info->dma);
2182                 tx_desc->read.cmd_type_len =
2183                          cpu_to_le32(cmd_type_len | buffer_info->length);
2184                 tx_desc->read.olinfo_status = cpu_to_le32(olinfo_status);
2185                 i++;
2186                 if (i == tx_ring->count)
2187                         i = 0;
2188         }
2189
2190         tx_desc->read.cmd_type_len |= cpu_to_le32(adapter->txd_cmd);
2191         /* Force memory writes to complete before letting h/w
2192          * know there are new descriptors to fetch.  (Only
2193          * applicable for weak-ordered memory model archs,
2194          * such as IA-64). */
2195         wmb();
2196
2197         tx_ring->next_to_use = i;
2198         writel(i, adapter->hw.hw_addr + tx_ring->tail);
2199         /* we need this if more than one processor can write to our tail
2200          * at a time, it syncronizes IO on IA64/Altix systems */
2201         mmiowb();
2202 }
2203
2204 static int igbvf_xmit_frame_ring_adv(struct sk_buff *skb,
2205                                    struct net_device *netdev,
2206                                    struct igbvf_ring *tx_ring)
2207 {
2208         struct igbvf_adapter *adapter = netdev_priv(netdev);
2209         unsigned int first, tx_flags = 0;
2210         u8 hdr_len = 0;
2211         int count = 0;
2212         int tso = 0;
2213
2214         if (test_bit(__IGBVF_DOWN, &adapter->state)) {
2215                 dev_kfree_skb_any(skb);
2216                 return NETDEV_TX_OK;
2217         }
2218
2219         if (skb->len <= 0) {
2220                 dev_kfree_skb_any(skb);
2221                 return NETDEV_TX_OK;
2222         }
2223
2224         /*
2225          * need: count + 4 desc gap to keep tail from touching
2226          *       + 2 desc gap to keep tail from touching head,
2227          *       + 1 desc for skb->data,
2228          *       + 1 desc for context descriptor,
2229          * head, otherwise try next time
2230          */
2231         if (igbvf_maybe_stop_tx(netdev, skb_shinfo(skb)->nr_frags + 4)) {
2232                 /* this is a hard error */
2233                 return NETDEV_TX_BUSY;
2234         }
2235
2236         if (adapter->vlgrp && vlan_tx_tag_present(skb)) {
2237                 tx_flags |= IGBVF_TX_FLAGS_VLAN;
2238                 tx_flags |= (vlan_tx_tag_get(skb) << IGBVF_TX_FLAGS_VLAN_SHIFT);
2239         }
2240
2241         if (skb->protocol == htons(ETH_P_IP))
2242                 tx_flags |= IGBVF_TX_FLAGS_IPV4;
2243
2244         first = tx_ring->next_to_use;
2245
2246         tso = skb_is_gso(skb) ?
2247                 igbvf_tso(adapter, tx_ring, skb, tx_flags, &hdr_len) : 0;
2248         if (unlikely(tso < 0)) {
2249                 dev_kfree_skb_any(skb);
2250                 return NETDEV_TX_OK;
2251         }
2252
2253         if (tso)
2254                 tx_flags |= IGBVF_TX_FLAGS_TSO;
2255         else if (igbvf_tx_csum(adapter, tx_ring, skb, tx_flags) &&
2256                  (skb->ip_summed == CHECKSUM_PARTIAL))
2257                 tx_flags |= IGBVF_TX_FLAGS_CSUM;
2258
2259         /*
2260          * count reflects descriptors mapped, if 0 then mapping error
2261          * has occured and we need to rewind the descriptor queue
2262          */
2263         count = igbvf_tx_map_adv(adapter, tx_ring, skb, first);
2264
2265         if (count) {
2266                 igbvf_tx_queue_adv(adapter, tx_ring, tx_flags, count,
2267                                    skb->len, hdr_len);
2268                 netdev->trans_start = jiffies;
2269                 /* Make sure there is space in the ring for the next send. */
2270                 igbvf_maybe_stop_tx(netdev, MAX_SKB_FRAGS + 4);
2271         } else {
2272                 dev_kfree_skb_any(skb);
2273                 tx_ring->buffer_info[first].time_stamp = 0;
2274                 tx_ring->next_to_use = first;
2275         }
2276
2277         return NETDEV_TX_OK;
2278 }
2279
2280 static int igbvf_xmit_frame(struct sk_buff *skb, struct net_device *netdev)
2281 {
2282         struct igbvf_adapter *adapter = netdev_priv(netdev);
2283         struct igbvf_ring *tx_ring;
2284         int retval;
2285
2286         if (test_bit(__IGBVF_DOWN, &adapter->state)) {
2287                 dev_kfree_skb_any(skb);
2288                 return NETDEV_TX_OK;
2289         }
2290
2291         tx_ring = &adapter->tx_ring[0];
2292
2293         retval = igbvf_xmit_frame_ring_adv(skb, netdev, tx_ring);
2294
2295         return retval;
2296 }
2297
2298 /**
2299  * igbvf_tx_timeout - Respond to a Tx Hang
2300  * @netdev: network interface device structure
2301  **/
2302 static void igbvf_tx_timeout(struct net_device *netdev)
2303 {
2304         struct igbvf_adapter *adapter = netdev_priv(netdev);
2305
2306         /* Do the reset outside of interrupt context */
2307         adapter->tx_timeout_count++;
2308         schedule_work(&adapter->reset_task);
2309 }
2310
2311 static void igbvf_reset_task(struct work_struct *work)
2312 {
2313         struct igbvf_adapter *adapter;
2314         adapter = container_of(work, struct igbvf_adapter, reset_task);
2315
2316         igbvf_reinit_locked(adapter);
2317 }
2318
2319 /**
2320  * igbvf_get_stats - Get System Network Statistics
2321  * @netdev: network interface device structure
2322  *
2323  * Returns the address of the device statistics structure.
2324  * The statistics are actually updated from the timer callback.
2325  **/
2326 static struct net_device_stats *igbvf_get_stats(struct net_device *netdev)
2327 {
2328         struct igbvf_adapter *adapter = netdev_priv(netdev);
2329
2330         /* only return the current stats */
2331         return &adapter->net_stats;
2332 }
2333
2334 /**
2335  * igbvf_change_mtu - Change the Maximum Transfer Unit
2336  * @netdev: network interface device structure
2337  * @new_mtu: new value for maximum frame size
2338  *
2339  * Returns 0 on success, negative on failure
2340  **/
2341 static int igbvf_change_mtu(struct net_device *netdev, int new_mtu)
2342 {
2343         struct igbvf_adapter *adapter = netdev_priv(netdev);
2344         int max_frame = new_mtu + ETH_HLEN + ETH_FCS_LEN;
2345
2346         if ((new_mtu < 68) || (max_frame > MAX_JUMBO_FRAME_SIZE)) {
2347                 dev_err(&adapter->pdev->dev, "Invalid MTU setting\n");
2348                 return -EINVAL;
2349         }
2350
2351         /* Jumbo frame size limits */
2352         if (max_frame > ETH_FRAME_LEN + ETH_FCS_LEN) {
2353                 if (!(adapter->flags & FLAG_HAS_JUMBO_FRAMES)) {
2354                         dev_err(&adapter->pdev->dev,
2355                                 "Jumbo Frames not supported.\n");
2356                         return -EINVAL;
2357                 }
2358         }
2359
2360 #define MAX_STD_JUMBO_FRAME_SIZE 9234
2361         if (max_frame > MAX_STD_JUMBO_FRAME_SIZE) {
2362                 dev_err(&adapter->pdev->dev, "MTU > 9216 not supported.\n");
2363                 return -EINVAL;
2364         }
2365
2366         while (test_and_set_bit(__IGBVF_RESETTING, &adapter->state))
2367                 msleep(1);
2368         /* igbvf_down has a dependency on max_frame_size */
2369         adapter->max_frame_size = max_frame;
2370         if (netif_running(netdev))
2371                 igbvf_down(adapter);
2372
2373         /*
2374          * NOTE: netdev_alloc_skb reserves 16 bytes, and typically NET_IP_ALIGN
2375          * means we reserve 2 more, this pushes us to allocate from the next
2376          * larger slab size.
2377          * i.e. RXBUFFER_2048 --> size-4096 slab
2378          * However with the new *_jumbo_rx* routines, jumbo receives will use
2379          * fragmented skbs
2380          */
2381
2382         if (max_frame <= 1024)
2383                 adapter->rx_buffer_len = 1024;
2384         else if (max_frame <= 2048)
2385                 adapter->rx_buffer_len = 2048;
2386         else
2387 #if (PAGE_SIZE / 2) > 16384
2388                 adapter->rx_buffer_len = 16384;
2389 #else
2390                 adapter->rx_buffer_len = PAGE_SIZE / 2;
2391 #endif
2392
2393
2394         /* adjust allocation if LPE protects us, and we aren't using SBP */
2395         if ((max_frame == ETH_FRAME_LEN + ETH_FCS_LEN) ||
2396              (max_frame == ETH_FRAME_LEN + VLAN_HLEN + ETH_FCS_LEN))
2397                 adapter->rx_buffer_len = ETH_FRAME_LEN + VLAN_HLEN +
2398                                          ETH_FCS_LEN;
2399
2400         dev_info(&adapter->pdev->dev, "changing MTU from %d to %d\n",
2401                  netdev->mtu, new_mtu);
2402         netdev->mtu = new_mtu;
2403
2404         if (netif_running(netdev))
2405                 igbvf_up(adapter);
2406         else
2407                 igbvf_reset(adapter);
2408
2409         clear_bit(__IGBVF_RESETTING, &adapter->state);
2410
2411         return 0;
2412 }
2413
2414 static int igbvf_ioctl(struct net_device *netdev, struct ifreq *ifr, int cmd)
2415 {
2416         switch (cmd) {
2417         default:
2418                 return -EOPNOTSUPP;
2419         }
2420 }
2421
2422 static int igbvf_suspend(struct pci_dev *pdev, pm_message_t state)
2423 {
2424         struct net_device *netdev = pci_get_drvdata(pdev);
2425         struct igbvf_adapter *adapter = netdev_priv(netdev);
2426 #ifdef CONFIG_PM
2427         int retval = 0;
2428 #endif
2429
2430         netif_device_detach(netdev);
2431
2432         if (netif_running(netdev)) {
2433                 WARN_ON(test_bit(__IGBVF_RESETTING, &adapter->state));
2434                 igbvf_down(adapter);
2435                 igbvf_free_irq(adapter);
2436         }
2437
2438 #ifdef CONFIG_PM
2439         retval = pci_save_state(pdev);
2440         if (retval)
2441                 return retval;
2442 #endif
2443
2444         pci_disable_device(pdev);
2445
2446         return 0;
2447 }
2448
2449 #ifdef CONFIG_PM
2450 static int igbvf_resume(struct pci_dev *pdev)
2451 {
2452         struct net_device *netdev = pci_get_drvdata(pdev);
2453         struct igbvf_adapter *adapter = netdev_priv(netdev);
2454         u32 err;
2455
2456         pci_restore_state(pdev);
2457         err = pci_enable_device_mem(pdev);
2458         if (err) {
2459                 dev_err(&pdev->dev, "Cannot enable PCI device from suspend\n");
2460                 return err;
2461         }
2462
2463         pci_set_master(pdev);
2464
2465         if (netif_running(netdev)) {
2466                 err = igbvf_request_irq(adapter);
2467                 if (err)
2468                         return err;
2469         }
2470
2471         igbvf_reset(adapter);
2472
2473         if (netif_running(netdev))
2474                 igbvf_up(adapter);
2475
2476         netif_device_attach(netdev);
2477
2478         return 0;
2479 }
2480 #endif
2481
2482 static void igbvf_shutdown(struct pci_dev *pdev)
2483 {
2484         igbvf_suspend(pdev, PMSG_SUSPEND);
2485 }
2486
2487 #ifdef CONFIG_NET_POLL_CONTROLLER
2488 /*
2489  * Polling 'interrupt' - used by things like netconsole to send skbs
2490  * without having to re-enable interrupts. It's not called while
2491  * the interrupt routine is executing.
2492  */
2493 static void igbvf_netpoll(struct net_device *netdev)
2494 {
2495         struct igbvf_adapter *adapter = netdev_priv(netdev);
2496
2497         disable_irq(adapter->pdev->irq);
2498
2499         igbvf_clean_tx_irq(adapter->tx_ring);
2500
2501         enable_irq(adapter->pdev->irq);
2502 }
2503 #endif
2504
2505 /**
2506  * igbvf_io_error_detected - called when PCI error is detected
2507  * @pdev: Pointer to PCI device
2508  * @state: The current pci connection state
2509  *
2510  * This function is called after a PCI bus error affecting
2511  * this device has been detected.
2512  */
2513 static pci_ers_result_t igbvf_io_error_detected(struct pci_dev *pdev,
2514                                                 pci_channel_state_t state)
2515 {
2516         struct net_device *netdev = pci_get_drvdata(pdev);
2517         struct igbvf_adapter *adapter = netdev_priv(netdev);
2518
2519         netif_device_detach(netdev);
2520
2521         if (netif_running(netdev))
2522                 igbvf_down(adapter);
2523         pci_disable_device(pdev);
2524
2525         /* Request a slot slot reset. */
2526         return PCI_ERS_RESULT_NEED_RESET;
2527 }
2528
2529 /**
2530  * igbvf_io_slot_reset - called after the pci bus has been reset.
2531  * @pdev: Pointer to PCI device
2532  *
2533  * Restart the card from scratch, as if from a cold-boot. Implementation
2534  * resembles the first-half of the igbvf_resume routine.
2535  */
2536 static pci_ers_result_t igbvf_io_slot_reset(struct pci_dev *pdev)
2537 {
2538         struct net_device *netdev = pci_get_drvdata(pdev);
2539         struct igbvf_adapter *adapter = netdev_priv(netdev);
2540
2541         if (pci_enable_device_mem(pdev)) {
2542                 dev_err(&pdev->dev,
2543                         "Cannot re-enable PCI device after reset.\n");
2544                 return PCI_ERS_RESULT_DISCONNECT;
2545         }
2546         pci_set_master(pdev);
2547
2548         igbvf_reset(adapter);
2549
2550         return PCI_ERS_RESULT_RECOVERED;
2551 }
2552
2553 /**
2554  * igbvf_io_resume - called when traffic can start flowing again.
2555  * @pdev: Pointer to PCI device
2556  *
2557  * This callback is called when the error recovery driver tells us that
2558  * its OK to resume normal operation. Implementation resembles the
2559  * second-half of the igbvf_resume routine.
2560  */
2561 static void igbvf_io_resume(struct pci_dev *pdev)
2562 {
2563         struct net_device *netdev = pci_get_drvdata(pdev);
2564         struct igbvf_adapter *adapter = netdev_priv(netdev);
2565
2566         if (netif_running(netdev)) {
2567                 if (igbvf_up(adapter)) {
2568                         dev_err(&pdev->dev,
2569                                 "can't bring device back up after reset\n");
2570                         return;
2571                 }
2572         }
2573
2574         netif_device_attach(netdev);
2575 }
2576
2577 static void igbvf_print_device_info(struct igbvf_adapter *adapter)
2578 {
2579         struct e1000_hw *hw = &adapter->hw;
2580         struct net_device *netdev = adapter->netdev;
2581         struct pci_dev *pdev = adapter->pdev;
2582
2583         dev_info(&pdev->dev, "Intel(R) 82576 Virtual Function\n");
2584         dev_info(&pdev->dev, "Address: %02x:%02x:%02x:%02x:%02x:%02x\n",
2585                  /* MAC address */
2586                  netdev->dev_addr[0], netdev->dev_addr[1],
2587                  netdev->dev_addr[2], netdev->dev_addr[3],
2588                  netdev->dev_addr[4], netdev->dev_addr[5]);
2589         dev_info(&pdev->dev, "MAC: %d\n", hw->mac.type);
2590 }
2591
2592 static const struct net_device_ops igbvf_netdev_ops = {
2593         .ndo_open                       = igbvf_open,
2594         .ndo_stop                       = igbvf_close,
2595         .ndo_start_xmit                 = igbvf_xmit_frame,
2596         .ndo_get_stats                  = igbvf_get_stats,
2597         .ndo_set_multicast_list         = igbvf_set_multi,
2598         .ndo_set_mac_address            = igbvf_set_mac,
2599         .ndo_change_mtu                 = igbvf_change_mtu,
2600         .ndo_do_ioctl                   = igbvf_ioctl,
2601         .ndo_tx_timeout                 = igbvf_tx_timeout,
2602         .ndo_vlan_rx_register           = igbvf_vlan_rx_register,
2603         .ndo_vlan_rx_add_vid            = igbvf_vlan_rx_add_vid,
2604         .ndo_vlan_rx_kill_vid           = igbvf_vlan_rx_kill_vid,
2605 #ifdef CONFIG_NET_POLL_CONTROLLER
2606         .ndo_poll_controller            = igbvf_netpoll,
2607 #endif
2608 };
2609
2610 /**
2611  * igbvf_probe - Device Initialization Routine
2612  * @pdev: PCI device information struct
2613  * @ent: entry in igbvf_pci_tbl
2614  *
2615  * Returns 0 on success, negative on failure
2616  *
2617  * igbvf_probe initializes an adapter identified by a pci_dev structure.
2618  * The OS initialization, configuring of the adapter private structure,
2619  * and a hardware reset occur.
2620  **/
2621 static int __devinit igbvf_probe(struct pci_dev *pdev,
2622                                  const struct pci_device_id *ent)
2623 {
2624         struct net_device *netdev;
2625         struct igbvf_adapter *adapter;
2626         struct e1000_hw *hw;
2627         const struct igbvf_info *ei = igbvf_info_tbl[ent->driver_data];
2628
2629         static int cards_found;
2630         int err, pci_using_dac;
2631
2632         err = pci_enable_device_mem(pdev);
2633         if (err)
2634                 return err;
2635
2636         pci_using_dac = 0;
2637         err = pci_set_dma_mask(pdev, DMA_64BIT_MASK);
2638         if (!err) {
2639                 err = pci_set_consistent_dma_mask(pdev, DMA_64BIT_MASK);
2640                 if (!err)
2641                         pci_using_dac = 1;
2642         } else {
2643                 err = pci_set_dma_mask(pdev, DMA_32BIT_MASK);
2644                 if (err) {
2645                         err = pci_set_consistent_dma_mask(pdev, DMA_32BIT_MASK);
2646                         if (err) {
2647                                 dev_err(&pdev->dev, "No usable DMA "
2648                                         "configuration, aborting\n");
2649                                 goto err_dma;
2650                         }
2651                 }
2652         }
2653
2654         err = pci_request_regions(pdev, igbvf_driver_name);
2655         if (err)
2656                 goto err_pci_reg;
2657
2658         pci_set_master(pdev);
2659
2660         err = -ENOMEM;
2661         netdev = alloc_etherdev(sizeof(struct igbvf_adapter));
2662         if (!netdev)
2663                 goto err_alloc_etherdev;
2664
2665         SET_NETDEV_DEV(netdev, &pdev->dev);
2666
2667         pci_set_drvdata(pdev, netdev);
2668         adapter = netdev_priv(netdev);
2669         hw = &adapter->hw;
2670         adapter->netdev = netdev;
2671         adapter->pdev = pdev;
2672         adapter->ei = ei;
2673         adapter->pba = ei->pba;
2674         adapter->flags = ei->flags;
2675         adapter->hw.back = adapter;
2676         adapter->hw.mac.type = ei->mac;
2677         adapter->msg_enable = (1 << NETIF_MSG_DRV | NETIF_MSG_PROBE) - 1;
2678
2679         /* PCI config space info */
2680
2681         hw->vendor_id = pdev->vendor;
2682         hw->device_id = pdev->device;
2683         hw->subsystem_vendor_id = pdev->subsystem_vendor;
2684         hw->subsystem_device_id = pdev->subsystem_device;
2685
2686         pci_read_config_byte(pdev, PCI_REVISION_ID, &hw->revision_id);
2687
2688         err = -EIO;
2689         adapter->hw.hw_addr = ioremap(pci_resource_start(pdev, 0),
2690                                       pci_resource_len(pdev, 0));
2691
2692         if (!adapter->hw.hw_addr)
2693                 goto err_ioremap;
2694
2695         if (ei->get_variants) {
2696                 err = ei->get_variants(adapter);
2697                 if (err)
2698                         goto err_ioremap;
2699         }
2700
2701         /* setup adapter struct */
2702         err = igbvf_sw_init(adapter);
2703         if (err)
2704                 goto err_sw_init;
2705
2706         /* construct the net_device struct */
2707         netdev->netdev_ops = &igbvf_netdev_ops;
2708
2709         igbvf_set_ethtool_ops(netdev);
2710         netdev->watchdog_timeo = 5 * HZ;
2711         strncpy(netdev->name, pci_name(pdev), sizeof(netdev->name) - 1);
2712
2713         adapter->bd_number = cards_found++;
2714
2715         netdev->features = NETIF_F_SG |
2716                            NETIF_F_IP_CSUM |
2717                            NETIF_F_HW_VLAN_TX |
2718                            NETIF_F_HW_VLAN_RX |
2719                            NETIF_F_HW_VLAN_FILTER;
2720
2721         netdev->features |= NETIF_F_IPV6_CSUM;
2722         netdev->features |= NETIF_F_TSO;
2723         netdev->features |= NETIF_F_TSO6;
2724
2725         if (pci_using_dac)
2726                 netdev->features |= NETIF_F_HIGHDMA;
2727
2728         netdev->vlan_features |= NETIF_F_TSO;
2729         netdev->vlan_features |= NETIF_F_TSO6;
2730         netdev->vlan_features |= NETIF_F_IP_CSUM;
2731         netdev->vlan_features |= NETIF_F_IPV6_CSUM;
2732         netdev->vlan_features |= NETIF_F_SG;
2733
2734         /*reset the controller to put the device in a known good state */
2735         err = hw->mac.ops.reset_hw(hw);
2736         if (err) {
2737                 dev_info(&pdev->dev,
2738                          "PF still in reset state, assigning new address\n");
2739                 random_ether_addr(hw->mac.addr);
2740         } else {
2741                 err = hw->mac.ops.read_mac_addr(hw);
2742                 if (err) {
2743                         dev_err(&pdev->dev, "Error reading MAC address\n");
2744                         goto err_hw_init;
2745                 }
2746         }
2747
2748         memcpy(netdev->dev_addr, adapter->hw.mac.addr, netdev->addr_len);
2749         memcpy(netdev->perm_addr, adapter->hw.mac.addr, netdev->addr_len);
2750
2751         if (!is_valid_ether_addr(netdev->perm_addr)) {
2752                 dev_err(&pdev->dev, "Invalid MAC Address: "
2753                         "%02x:%02x:%02x:%02x:%02x:%02x\n",
2754                         netdev->dev_addr[0], netdev->dev_addr[1],
2755                         netdev->dev_addr[2], netdev->dev_addr[3],
2756                         netdev->dev_addr[4], netdev->dev_addr[5]);
2757                 err = -EIO;
2758                 goto err_hw_init;
2759         }
2760
2761         setup_timer(&adapter->watchdog_timer, &igbvf_watchdog,
2762                     (unsigned long) adapter);
2763
2764         INIT_WORK(&adapter->reset_task, igbvf_reset_task);
2765         INIT_WORK(&adapter->watchdog_task, igbvf_watchdog_task);
2766
2767         /* ring size defaults */
2768         adapter->rx_ring->count = 1024;
2769         adapter->tx_ring->count = 1024;
2770
2771         /* reset the hardware with the new settings */
2772         igbvf_reset(adapter);
2773
2774         /* tell the stack to leave us alone until igbvf_open() is called */
2775         netif_carrier_off(netdev);
2776         netif_stop_queue(netdev);
2777
2778         strcpy(netdev->name, "eth%d");
2779         err = register_netdev(netdev);
2780         if (err)
2781                 goto err_hw_init;
2782
2783         igbvf_print_device_info(adapter);
2784
2785         igbvf_initialize_last_counter_stats(adapter);
2786
2787         return 0;
2788
2789 err_hw_init:
2790         kfree(adapter->tx_ring);
2791         kfree(adapter->rx_ring);
2792 err_sw_init:
2793         igbvf_reset_interrupt_capability(adapter);
2794         iounmap(adapter->hw.hw_addr);
2795 err_ioremap:
2796         free_netdev(netdev);
2797 err_alloc_etherdev:
2798         pci_release_regions(pdev);
2799 err_pci_reg:
2800 err_dma:
2801         pci_disable_device(pdev);
2802         return err;
2803 }
2804
2805 /**
2806  * igbvf_remove - Device Removal Routine
2807  * @pdev: PCI device information struct
2808  *
2809  * igbvf_remove is called by the PCI subsystem to alert the driver
2810  * that it should release a PCI device.  The could be caused by a
2811  * Hot-Plug event, or because the driver is going to be removed from
2812  * memory.
2813  **/
2814 static void __devexit igbvf_remove(struct pci_dev *pdev)
2815 {
2816         struct net_device *netdev = pci_get_drvdata(pdev);
2817         struct igbvf_adapter *adapter = netdev_priv(netdev);
2818         struct e1000_hw *hw = &adapter->hw;
2819
2820         /*
2821          * flush_scheduled work may reschedule our watchdog task, so
2822          * explicitly disable watchdog tasks from being rescheduled
2823          */
2824         set_bit(__IGBVF_DOWN, &adapter->state);
2825         del_timer_sync(&adapter->watchdog_timer);
2826
2827         flush_scheduled_work();
2828
2829         unregister_netdev(netdev);
2830
2831         igbvf_reset_interrupt_capability(adapter);
2832
2833         /*
2834          * it is important to delete the napi struct prior to freeing the
2835          * rx ring so that you do not end up with null pointer refs
2836          */
2837         netif_napi_del(&adapter->rx_ring->napi);
2838         kfree(adapter->tx_ring);
2839         kfree(adapter->rx_ring);
2840
2841         iounmap(hw->hw_addr);
2842         if (hw->flash_address)
2843                 iounmap(hw->flash_address);
2844         pci_release_regions(pdev);
2845
2846         free_netdev(netdev);
2847
2848         pci_disable_device(pdev);
2849 }
2850
2851 /* PCI Error Recovery (ERS) */
2852 static struct pci_error_handlers igbvf_err_handler = {
2853         .error_detected = igbvf_io_error_detected,
2854         .slot_reset = igbvf_io_slot_reset,
2855         .resume = igbvf_io_resume,
2856 };
2857
2858 static struct pci_device_id igbvf_pci_tbl[] = {
2859         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82576_VF), board_vf },
2860         { } /* terminate list */
2861 };
2862 MODULE_DEVICE_TABLE(pci, igbvf_pci_tbl);
2863
2864 /* PCI Device API Driver */
2865 static struct pci_driver igbvf_driver = {
2866         .name     = igbvf_driver_name,
2867         .id_table = igbvf_pci_tbl,
2868         .probe    = igbvf_probe,
2869         .remove   = __devexit_p(igbvf_remove),
2870 #ifdef CONFIG_PM
2871         /* Power Management Hooks */
2872         .suspend  = igbvf_suspend,
2873         .resume   = igbvf_resume,
2874 #endif
2875         .shutdown = igbvf_shutdown,
2876         .err_handler = &igbvf_err_handler
2877 };
2878
2879 /**
2880  * igbvf_init_module - Driver Registration Routine
2881  *
2882  * igbvf_init_module is the first routine called when the driver is
2883  * loaded. All it does is register with the PCI subsystem.
2884  **/
2885 static int __init igbvf_init_module(void)
2886 {
2887         int ret;
2888         printk(KERN_INFO "%s - version %s\n",
2889                igbvf_driver_string, igbvf_driver_version);
2890         printk(KERN_INFO "%s\n", igbvf_copyright);
2891
2892         ret = pci_register_driver(&igbvf_driver);
2893         pm_qos_add_requirement(PM_QOS_CPU_DMA_LATENCY, igbvf_driver_name,
2894                                PM_QOS_DEFAULT_VALUE);
2895
2896         return ret;
2897 }
2898 module_init(igbvf_init_module);
2899
2900 /**
2901  * igbvf_exit_module - Driver Exit Cleanup Routine
2902  *
2903  * igbvf_exit_module is called just before the driver is removed
2904  * from memory.
2905  **/
2906 static void __exit igbvf_exit_module(void)
2907 {
2908         pci_unregister_driver(&igbvf_driver);
2909         pm_qos_remove_requirement(PM_QOS_CPU_DMA_LATENCY, igbvf_driver_name);
2910 }
2911 module_exit(igbvf_exit_module);
2912
2913
2914 MODULE_AUTHOR("Intel Corporation, <e1000-devel@lists.sourceforge.net>");
2915 MODULE_DESCRIPTION("Intel(R) 82576 Virtual Function Network Driver");
2916 MODULE_LICENSE("GPL");
2917 MODULE_VERSION(DRV_VERSION);
2918
2919 /* netdev.c */