drivers/net/igbvf/netdev.c

   1 /*******************************************************************************
   2
   3   Intel(R) 82576 Virtual Function Linux driver
   4   Copyright(c) 2009 - 2010 Intel Corporation.
   5
   6   This program is free software; you can redistribute it and/or modify it
   7   under the terms and conditions of the GNU General Public License,
   8   version 2, as published by the Free Software Foundation.
   9
  10   This program is distributed in the hope it will be useful, but WITHOUT
  11   ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
  12   FITNESS FOR A PARTICULAR PURPOSE.  See the GNU General Public License for
  13   more details.
  14
  15   You should have received a copy of the GNU General Public License along with
  16   this program; if not, write to the Free Software Foundation, Inc.,
  17   51 Franklin St - Fifth Floor, Boston, MA 02110-1301 USA.
  18
  19   The full GNU General Public License is included in this distribution in
  20   the file called "COPYING".
  21
  22   Contact Information:
  23   e1000-devel Mailing List <e1000-devel@lists.sourceforge.net>
  24   Intel Corporation, 5200 N.E. Elam Young Parkway, Hillsboro, OR 97124-6497
  25
  26 *******************************************************************************/
  27
  28 #include <linux/module.h>
  29 #include <linux/types.h>
  30 #include <linux/init.h>
  31 #include <linux/pci.h>
  32 #include <linux/vmalloc.h>
  33 #include <linux/pagemap.h>
  34 #include <linux/delay.h>
  35 #include <linux/netdevice.h>
  36 #include <linux/tcp.h>
  37 #include <linux/ipv6.h>
  38 #include <linux/slab.h>
  39 #include <net/checksum.h>
  40 #include <net/ip6_checksum.h>
  41 #include <linux/mii.h>
  42 #include <linux/ethtool.h>
  43 #include <linux/if_vlan.h>
  44
  45 #include "igbvf.h"
  46
  47 #define DRV_VERSION "1.0.8-k0"
  48 char igbvf_driver_name[] = "igbvf";
  49 const char igbvf_driver_version[] = DRV_VERSION;
  50 static const char igbvf_driver_string[] =
  51                                 "Intel(R) Virtual Function Network Driver";
  52 static const char igbvf_copyright[] =
  53                                 "Copyright (c) 2009 - 2010 Intel Corporation.";
  54
  55 static int igbvf_poll(struct napi_struct *napi, int budget);
  56 static void igbvf_reset(struct igbvf_adapter *);
  57 static void igbvf_set_interrupt_capability(struct igbvf_adapter *);
  58 static void igbvf_reset_interrupt_capability(struct igbvf_adapter *);
  59
  60 static struct igbvf_info igbvf_vf_info = {
  61         .mac                    = e1000_vfadapt,
  62         .flags                  = 0,
  63         .pba                    = 10,
  64         .init_ops               = e1000_init_function_pointers_vf,
  65 };
  66
  67 static struct igbvf_info igbvf_i350_vf_info = {
  68         .mac                    = e1000_vfadapt_i350,
  69         .flags                  = 0,
  70         .pba                    = 10,
  71         .init_ops               = e1000_init_function_pointers_vf,
  72 };
  73
  74 static const struct igbvf_info *igbvf_info_tbl[] = {
  75         [board_vf]              = &igbvf_vf_info,
  76         [board_i350_vf]         = &igbvf_i350_vf_info,
  77 };
  78
  79 /**
  80  * igbvf_desc_unused - calculate if we have unused descriptors
  81  **/
  82 static int igbvf_desc_unused(struct igbvf_ring *ring)
  83 {
  84         if (ring->next_to_clean > ring->next_to_use)
  85                 return ring->next_to_clean - ring->next_to_use - 1;
  86
  87         return ring->count + ring->next_to_clean - ring->next_to_use - 1;
  88 }
  89
  90 /**
  91  * igbvf_receive_skb - helper function to handle Rx indications
  92  * @adapter: board private structure
  93  * @status: descriptor status field as written by hardware
  94  * @vlan: descriptor vlan field as written by hardware (no le/be conversion)
  95  * @skb: pointer to sk_buff to be indicated to stack
  96  **/
  97 static void igbvf_receive_skb(struct igbvf_adapter *adapter,
  98                               struct net_device *netdev,
  99                               struct sk_buff *skb,
 100                               u32 status, u16 vlan)
 101 {
 102         if (adapter->vlgrp && (status & E1000_RXD_STAT_VP))
 103                 vlan_hwaccel_receive_skb(skb, adapter->vlgrp,
 104                                          le16_to_cpu(vlan) &
 105                                          E1000_RXD_SPC_VLAN_MASK);
 106         else
 107                 netif_receive_skb(skb);
 108 }
 109
 110 static inline void igbvf_rx_checksum_adv(struct igbvf_adapter *adapter,
 111                                          u32 status_err, struct sk_buff *skb)
 112 {
 113         skb_checksum_none_assert(skb);
 114
 115         /* Ignore Checksum bit is set or checksum is disabled through ethtool */
 116         if ((status_err & E1000_RXD_STAT_IXSM) ||
 117             (adapter->flags & IGBVF_FLAG_RX_CSUM_DISABLED))
 118                 return;
 119
 120         /* TCP/UDP checksum error bit is set */
 121         if (status_err &
 122             (E1000_RXDEXT_STATERR_TCPE | E1000_RXDEXT_STATERR_IPE)) {
 123                 /* let the stack verify checksum errors */
 124                 adapter->hw_csum_err++;
 125                 return;
 126         }
 127
 128         /* It must be a TCP or UDP packet with a valid checksum */
 129         if (status_err & (E1000_RXD_STAT_TCPCS | E1000_RXD_STAT_UDPCS))
 130                 skb->ip_summed = CHECKSUM_UNNECESSARY;
 131
 132         adapter->hw_csum_good++;
 133 }
 134
 135 /**
 136  * igbvf_alloc_rx_buffers - Replace used receive buffers; packet split
 137  * @rx_ring: address of ring structure to repopulate
 138  * @cleaned_count: number of buffers to repopulate
 139  **/
 140 static void igbvf_alloc_rx_buffers(struct igbvf_ring *rx_ring,
 141                                    int cleaned_count)
 142 {
 143         struct igbvf_adapter *adapter = rx_ring->adapter;
 144         struct net_device *netdev = adapter->netdev;
 145         struct pci_dev *pdev = adapter->pdev;
 146         union e1000_adv_rx_desc *rx_desc;
 147         struct igbvf_buffer *buffer_info;
 148         struct sk_buff *skb;
 149         unsigned int i;
 150         int bufsz;
 151
 152         i = rx_ring->next_to_use;
 153         buffer_info = &rx_ring->buffer_info[i];
 154
 155         if (adapter->rx_ps_hdr_size)
 156                 bufsz = adapter->rx_ps_hdr_size;
 157         else
 158                 bufsz = adapter->rx_buffer_len;
 159
 160         while (cleaned_count--) {
 161                 rx_desc = IGBVF_RX_DESC_ADV(*rx_ring, i);
 162
 163                 if (adapter->rx_ps_hdr_size && !buffer_info->page_dma) {
 164                         if (!buffer_info->page) {
 165                                 buffer_info->page = alloc_page(GFP_ATOMIC);
 166                                 if (!buffer_info->page) {
 167                                         adapter->alloc_rx_buff_failed++;
 168                                         goto no_buffers;
 169                                 }
 170                                 buffer_info->page_offset = 0;
 171                         } else {
 172                                 buffer_info->page_offset ^= PAGE_SIZE / 2;
 173                         }
 174                         buffer_info->page_dma =
 175                                 dma_map_page(&pdev->dev, buffer_info->page,
 176                                              buffer_info->page_offset,
 177                                              PAGE_SIZE / 2,
 178                                              DMA_FROM_DEVICE);
 179                 }
 180
 181                 if (!buffer_info->skb) {
 182                         skb = netdev_alloc_skb_ip_align(netdev, bufsz);
 183                         if (!skb) {
 184                                 adapter->alloc_rx_buff_failed++;
 185                                 goto no_buffers;
 186                         }
 187
 188                         buffer_info->skb = skb;
 189                         buffer_info->dma = dma_map_single(&pdev->dev, skb->data,
 190                                                           bufsz,
 191                                                           DMA_FROM_DEVICE);
 192                 }
 193                 /* Refresh the desc even if buffer_addrs didn't change because
 194                  * each write-back erases this info. */
 195                 if (adapter->rx_ps_hdr_size) {
 196                         rx_desc->read.pkt_addr =
 197                              cpu_to_le64(buffer_info->page_dma);
 198                         rx_desc->read.hdr_addr = cpu_to_le64(buffer_info->dma);
 199                 } else {
 200                         rx_desc->read.pkt_addr =
 201                              cpu_to_le64(buffer_info->dma);
 202                         rx_desc->read.hdr_addr = 0;
 203                 }
 204
 205                 i++;
 206                 if (i == rx_ring->count)
 207                         i = 0;
 208                 buffer_info = &rx_ring->buffer_info[i];
 209         }
 210
 211 no_buffers:
 212         if (rx_ring->next_to_use != i) {
 213                 rx_ring->next_to_use = i;
 214                 if (i == 0)
 215                         i = (rx_ring->count - 1);
 216                 else
 217                         i--;
 218
 219                 /* Force memory writes to complete before letting h/w
 220                  * know there are new descriptors to fetch.  (Only
 221                  * applicable for weak-ordered memory model archs,
 222                  * such as IA-64). */
 223                 wmb();
 224                 writel(i, adapter->hw.hw_addr + rx_ring->tail);
 225         }
 226 }
 227
 228 /**
 229  * igbvf_clean_rx_irq - Send received data up the network stack; legacy
 230  * @adapter: board private structure
 231  *
 232  * the return value indicates whether actual cleaning was done, there
 233  * is no guarantee that everything was cleaned
 234  **/
 235 static bool igbvf_clean_rx_irq(struct igbvf_adapter *adapter,
 236                                int *work_done, int work_to_do)
 237 {
 238         struct igbvf_ring *rx_ring = adapter->rx_ring;
 239         struct net_device *netdev = adapter->netdev;
 240         struct pci_dev *pdev = adapter->pdev;
 241         union e1000_adv_rx_desc *rx_desc, *next_rxd;
 242         struct igbvf_buffer *buffer_info, *next_buffer;
 243         struct sk_buff *skb;
 244         bool cleaned = false;
 245         int cleaned_count = 0;
 246         unsigned int total_bytes = 0, total_packets = 0;
 247         unsigned int i;
 248         u32 length, hlen, staterr;
 249
 250         i = rx_ring->next_to_clean;
 251         rx_desc = IGBVF_RX_DESC_ADV(*rx_ring, i);
 252         staterr = le32_to_cpu(rx_desc->wb.upper.status_error);
 253
 254         while (staterr & E1000_RXD_STAT_DD) {
 255                 if (*work_done >= work_to_do)
 256                         break;
 257                 (*work_done)++;
 258                 rmb(); /* read descriptor and rx_buffer_info after status DD */
 259
 260                 buffer_info = &rx_ring->buffer_info[i];
 261
 262                 /* HW will not DMA in data larger than the given buffer, even
 263                  * if it parses the (NFS, of course) header to be larger.  In
 264                  * that case, it fills the header buffer and spills the rest
 265                  * into the page.
 266                  */
 267                 hlen = (le16_to_cpu(rx_desc->wb.lower.lo_dword.hs_rss.hdr_info) &
 268                   E1000_RXDADV_HDRBUFLEN_MASK) >> E1000_RXDADV_HDRBUFLEN_SHIFT;
 269                 if (hlen > adapter->rx_ps_hdr_size)
 270                         hlen = adapter->rx_ps_hdr_size;
 271
 272                 length = le16_to_cpu(rx_desc->wb.upper.length);
 273                 cleaned = true;
 274                 cleaned_count++;
 275
 276                 skb = buffer_info->skb;
 277                 prefetch(skb->data - NET_IP_ALIGN);
 278                 buffer_info->skb = NULL;
 279                 if (!adapter->rx_ps_hdr_size) {
 280                         dma_unmap_single(&pdev->dev, buffer_info->dma,
 281                                          adapter->rx_buffer_len,
 282                                          DMA_FROM_DEVICE);
 283                         buffer_info->dma = 0;
 284                         skb_put(skb, length);
 285                         goto send_up;
 286                 }
 287
 288                 if (!skb_shinfo(skb)->nr_frags) {
 289                         dma_unmap_single(&pdev->dev, buffer_info->dma,
 290                                          adapter->rx_ps_hdr_size,
 291                                          DMA_FROM_DEVICE);
 292                         skb_put(skb, hlen);
 293                 }
 294
 295                 if (length) {
 296                         dma_unmap_page(&pdev->dev, buffer_info->page_dma,
 297                                        PAGE_SIZE / 2,
 298                                        DMA_FROM_DEVICE);
 299                         buffer_info->page_dma = 0;
 300
 301                         skb_fill_page_desc(skb, skb_shinfo(skb)->nr_frags,
 302                                            buffer_info->page,
 303                                            buffer_info->page_offset,
 304                                            length);
 305
 306                         if ((adapter->rx_buffer_len > (PAGE_SIZE / 2)) ||
 307                             (page_count(buffer_info->page) != 1))
 308                                 buffer_info->page = NULL;
 309                         else
 310                                 get_page(buffer_info->page);
 311
 312                         skb->len += length;
 313                         skb->data_len += length;
 314                         skb->truesize += length;
 315                 }
 316 send_up:
 317                 i++;
 318                 if (i == rx_ring->count)
 319                         i = 0;
 320                 next_rxd = IGBVF_RX_DESC_ADV(*rx_ring, i);
 321                 prefetch(next_rxd);
 322                 next_buffer = &rx_ring->buffer_info[i];
 323
 324                 if (!(staterr & E1000_RXD_STAT_EOP)) {
 325                         buffer_info->skb = next_buffer->skb;
 326                         buffer_info->dma = next_buffer->dma;
 327                         next_buffer->skb = skb;
 328                         next_buffer->dma = 0;
 329                         goto next_desc;
 330                 }
 331
 332                 if (staterr & E1000_RXDEXT_ERR_FRAME_ERR_MASK) {
 333                         dev_kfree_skb_irq(skb);
 334                         goto next_desc;
 335                 }
 336
 337                 total_bytes += skb->len;
 338                 total_packets++;
 339
 340                 igbvf_rx_checksum_adv(adapter, staterr, skb);
 341
 342                 skb->protocol = eth_type_trans(skb, netdev);
 343
 344                 igbvf_receive_skb(adapter, netdev, skb, staterr,
 345                                   rx_desc->wb.upper.vlan);
 346
 347 next_desc:
 348                 rx_desc->wb.upper.status_error = 0;
 349
 350                 /* return some buffers to hardware, one at a time is too slow */
 351                 if (cleaned_count >= IGBVF_RX_BUFFER_WRITE) {
 352                         igbvf_alloc_rx_buffers(rx_ring, cleaned_count);
 353                         cleaned_count = 0;
 354                 }
 355
 356                 /* use prefetched values */
 357                 rx_desc = next_rxd;
 358                 buffer_info = next_buffer;
 359
 360                 staterr = le32_to_cpu(rx_desc->wb.upper.status_error);
 361         }
 362
 363         rx_ring->next_to_clean = i;
 364         cleaned_count = igbvf_desc_unused(rx_ring);
 365
 366         if (cleaned_count)
 367                 igbvf_alloc_rx_buffers(rx_ring, cleaned_count);
 368
 369         adapter->total_rx_packets += total_packets;
 370         adapter->total_rx_bytes += total_bytes;
 371         adapter->net_stats.rx_bytes += total_bytes;
 372         adapter->net_stats.rx_packets += total_packets;
 373         return cleaned;
 374 }
 375
 376 static void igbvf_put_txbuf(struct igbvf_adapter *adapter,
 377                             struct igbvf_buffer *buffer_info)
 378 {
 379         if (buffer_info->dma) {
 380                 if (buffer_info->mapped_as_page)
 381                         dma_unmap_page(&adapter->pdev->dev,
 382                                        buffer_info->dma,
 383                                        buffer_info->length,
 384                                        DMA_TO_DEVICE);
 385                 else
 386                         dma_unmap_single(&adapter->pdev->dev,
 387                                          buffer_info->dma,
 388                                          buffer_info->length,
 389                                          DMA_TO_DEVICE);
 390                 buffer_info->dma = 0;
 391         }
 392         if (buffer_info->skb) {
 393                 dev_kfree_skb_any(buffer_info->skb);
 394                 buffer_info->skb = NULL;
 395         }
 396         buffer_info->time_stamp = 0;
 397 }
 398
 399 /**
 400  * igbvf_setup_tx_resources - allocate Tx resources (Descriptors)
 401  * @adapter: board private structure
 402  *
 403  * Return 0 on success, negative on failure
 404  **/
 405 int igbvf_setup_tx_resources(struct igbvf_adapter *adapter,
 406                              struct igbvf_ring *tx_ring)
 407 {
 408         struct pci_dev *pdev = adapter->pdev;
 409         int size;
 410
 411         size = sizeof(struct igbvf_buffer) * tx_ring->count;
 412         tx_ring->buffer_info = vzalloc(size);
 413         if (!tx_ring->buffer_info)
 414                 goto err;
 415
 416         /* round up to nearest 4K */
 417         tx_ring->size = tx_ring->count * sizeof(union e1000_adv_tx_desc);
 418         tx_ring->size = ALIGN(tx_ring->size, 4096);
 419
 420         tx_ring->desc = dma_alloc_coherent(&pdev->dev, tx_ring->size,
 421                                            &tx_ring->dma, GFP_KERNEL);
 422
 423         if (!tx_ring->desc)
 424                 goto err;
 425
 426         tx_ring->adapter = adapter;
 427         tx_ring->next_to_use = 0;
 428         tx_ring->next_to_clean = 0;
 429
 430         return 0;
 431 err:
 432         vfree(tx_ring->buffer_info);
 433         dev_err(&adapter->pdev->dev,
 434                 "Unable to allocate memory for the transmit descriptor ring\n");
 435         return -ENOMEM;
 436 }
 437
 438 /**
 439  * igbvf_setup_rx_resources - allocate Rx resources (Descriptors)
 440  * @adapter: board private structure
 441  *
 442  * Returns 0 on success, negative on failure
 443  **/
 444 int igbvf_setup_rx_resources(struct igbvf_adapter *adapter,
 445                              struct igbvf_ring *rx_ring)
 446 {
 447         struct pci_dev *pdev = adapter->pdev;
 448         int size, desc_len;
 449
 450         size = sizeof(struct igbvf_buffer) * rx_ring->count;
 451         rx_ring->buffer_info = vzalloc(size);
 452         if (!rx_ring->buffer_info)
 453                 goto err;
 454
 455         desc_len = sizeof(union e1000_adv_rx_desc);
 456
 457         /* Round up to nearest 4K */
 458         rx_ring->size = rx_ring->count * desc_len;
 459         rx_ring->size = ALIGN(rx_ring->size, 4096);
 460
 461         rx_ring->desc = dma_alloc_coherent(&pdev->dev, rx_ring->size,
 462                                            &rx_ring->dma, GFP_KERNEL);
 463
 464         if (!rx_ring->desc)
 465                 goto err;
 466
 467         rx_ring->next_to_clean = 0;
 468         rx_ring->next_to_use = 0;
 469
 470         rx_ring->adapter = adapter;
 471
 472         return 0;
 473
 474 err:
 475         vfree(rx_ring->buffer_info);
 476         rx_ring->buffer_info = NULL;
 477         dev_err(&adapter->pdev->dev,
 478                 "Unable to allocate memory for the receive descriptor ring\n");
 479         return -ENOMEM;
 480 }
 481
 482 /**
 483  * igbvf_clean_tx_ring - Free Tx Buffers
 484  * @tx_ring: ring to be cleaned
 485  **/
 486 static void igbvf_clean_tx_ring(struct igbvf_ring *tx_ring)
 487 {
 488         struct igbvf_adapter *adapter = tx_ring->adapter;
 489         struct igbvf_buffer *buffer_info;
 490         unsigned long size;
 491         unsigned int i;
 492
 493         if (!tx_ring->buffer_info)
 494                 return;
 495
 496         /* Free all the Tx ring sk_buffs */
 497         for (i = 0; i < tx_ring->count; i++) {
 498                 buffer_info = &tx_ring->buffer_info[i];
 499                 igbvf_put_txbuf(adapter, buffer_info);
 500         }
 501
 502         size = sizeof(struct igbvf_buffer) * tx_ring->count;
 503         memset(tx_ring->buffer_info, 0, size);
 504
 505         /* Zero out the descriptor ring */
 506         memset(tx_ring->desc, 0, tx_ring->size);
 507
 508         tx_ring->next_to_use = 0;
 509         tx_ring->next_to_clean = 0;
 510
 511         writel(0, adapter->hw.hw_addr + tx_ring->head);
 512         writel(0, adapter->hw.hw_addr + tx_ring->tail);
 513 }
 514
 515 /**
 516  * igbvf_free_tx_resources - Free Tx Resources per Queue
 517  * @tx_ring: ring to free resources from
 518  *
 519  * Free all transmit software resources
 520  **/
 521 void igbvf_free_tx_resources(struct igbvf_ring *tx_ring)
 522 {
 523         struct pci_dev *pdev = tx_ring->adapter->pdev;
 524
 525         igbvf_clean_tx_ring(tx_ring);
 526
 527         vfree(tx_ring->buffer_info);
 528         tx_ring->buffer_info = NULL;
 529
 530         dma_free_coherent(&pdev->dev, tx_ring->size, tx_ring->desc,
 531                           tx_ring->dma);
 532
 533         tx_ring->desc = NULL;
 534 }
 535
 536 /**
 537  * igbvf_clean_rx_ring - Free Rx Buffers per Queue
 538  * @adapter: board private structure
 539  **/
 540 static void igbvf_clean_rx_ring(struct igbvf_ring *rx_ring)
 541 {
 542         struct igbvf_adapter *adapter = rx_ring->adapter;
 543         struct igbvf_buffer *buffer_info;
 544         struct pci_dev *pdev = adapter->pdev;
 545         unsigned long size;
 546         unsigned int i;
 547
 548         if (!rx_ring->buffer_info)
 549                 return;
 550
 551         /* Free all the Rx ring sk_buffs */
 552         for (i = 0; i < rx_ring->count; i++) {
 553                 buffer_info = &rx_ring->buffer_info[i];
 554                 if (buffer_info->dma) {
 555                         if (adapter->rx_ps_hdr_size){
 556                                 dma_unmap_single(&pdev->dev, buffer_info->dma,
 557                                                  adapter->rx_ps_hdr_size,
 558                                                  DMA_FROM_DEVICE);
 559                         } else {
 560                                 dma_unmap_single(&pdev->dev, buffer_info->dma,
 561                                                  adapter->rx_buffer_len,
 562                                                  DMA_FROM_DEVICE);
 563                         }
 564                         buffer_info->dma = 0;
 565                 }
 566
 567                 if (buffer_info->skb) {
 568                         dev_kfree_skb(buffer_info->skb);
 569                         buffer_info->skb = NULL;
 570                 }
 571
 572                 if (buffer_info->page) {
 573                         if (buffer_info->page_dma)
 574                                 dma_unmap_page(&pdev->dev,
 575                                                buffer_info->page_dma,
 576                                                PAGE_SIZE / 2,
 577                                                DMA_FROM_DEVICE);
 578                         put_page(buffer_info->page);
 579                         buffer_info->page = NULL;
 580                         buffer_info->page_dma = 0;
 581                         buffer_info->page_offset = 0;
 582                 }
 583         }
 584
 585         size = sizeof(struct igbvf_buffer) * rx_ring->count;
 586         memset(rx_ring->buffer_info, 0, size);
 587
 588         /* Zero out the descriptor ring */
 589         memset(rx_ring->desc, 0, rx_ring->size);
 590
 591         rx_ring->next_to_clean = 0;
 592         rx_ring->next_to_use = 0;
 593
 594         writel(0, adapter->hw.hw_addr + rx_ring->head);
 595         writel(0, adapter->hw.hw_addr + rx_ring->tail);
 596 }
 597
 598 /**
 599  * igbvf_free_rx_resources - Free Rx Resources
 600  * @rx_ring: ring to clean the resources from
 601  *
 602  * Free all receive software resources
 603  **/
 604
 605 void igbvf_free_rx_resources(struct igbvf_ring *rx_ring)
 606 {
 607         struct pci_dev *pdev = rx_ring->adapter->pdev;
 608
 609         igbvf_clean_rx_ring(rx_ring);
 610
 611         vfree(rx_ring->buffer_info);
 612         rx_ring->buffer_info = NULL;
 613
 614         dma_free_coherent(&pdev->dev, rx_ring->size, rx_ring->desc,
 615                           rx_ring->dma);
 616         rx_ring->desc = NULL;
 617 }
 618
 619 /**
 620  * igbvf_update_itr - update the dynamic ITR value based on statistics
 621  * @adapter: pointer to adapter
 622  * @itr_setting: current adapter->itr
 623  * @packets: the number of packets during this measurement interval
 624  * @bytes: the number of bytes during this measurement interval
 625  *
 626  *      Stores a new ITR value based on packets and byte
 627  *      counts during the last interrupt.  The advantage of per interrupt
 628  *      computation is faster updates and more accurate ITR for the current
 629  *      traffic pattern.  Constants in this function were computed
 630  *      based on theoretical maximum wire speed and thresholds were set based
 631  *      on testing data as well as attempting to minimize response time
 632  *      while increasing bulk throughput.  This functionality is controlled
 633  *      by the InterruptThrottleRate module parameter.
 634  **/
 635 static unsigned int igbvf_update_itr(struct igbvf_adapter *adapter,
 636                                      u16 itr_setting, int packets,
 637                                      int bytes)
 638 {
 639         unsigned int retval = itr_setting;
 640
 641         if (packets == 0)
 642                 goto update_itr_done;
 643
 644         switch (itr_setting) {
 645         case lowest_latency:
 646                 /* handle TSO and jumbo frames */
 647                 if (bytes/packets > 8000)
 648                         retval = bulk_latency;
 649                 else if ((packets < 5) && (bytes > 512))
 650                         retval = low_latency;
 651                 break;
 652         case low_latency:  /* 50 usec aka 20000 ints/s */
 653                 if (bytes > 10000) {
 654                         /* this if handles the TSO accounting */
 655                         if (bytes/packets > 8000)
 656                                 retval = bulk_latency;
 657                         else if ((packets < 10) || ((bytes/packets) > 1200))
 658                                 retval = bulk_latency;
 659                         else if ((packets > 35))
 660                                 retval = lowest_latency;
 661                 } else if (bytes/packets > 2000) {
 662                         retval = bulk_latency;
 663                 } else if (packets <= 2 && bytes < 512) {
 664                         retval = lowest_latency;
 665                 }
 666                 break;
 667         case bulk_latency: /* 250 usec aka 4000 ints/s */
 668                 if (bytes > 25000) {
 669                         if (packets > 35)
 670                                 retval = low_latency;
 671                 } else if (bytes < 6000) {
 672                         retval = low_latency;
 673                 }
 674                 break;
 675         }
 676
 677 update_itr_done:
 678         return retval;
 679 }
 680
 681 static void igbvf_set_itr(struct igbvf_adapter *adapter)
 682 {
 683         struct e1000_hw *hw = &adapter->hw;
 684         u16 current_itr;
 685         u32 new_itr = adapter->itr;
 686
 687         adapter->tx_itr = igbvf_update_itr(adapter, adapter->tx_itr,
 688                                            adapter->total_tx_packets,
 689                                            adapter->total_tx_bytes);
 690         /* conservative mode (itr 3) eliminates the lowest_latency setting */
 691         if (adapter->itr_setting == 3 && adapter->tx_itr == lowest_latency)
 692                 adapter->tx_itr = low_latency;
 693
 694         adapter->rx_itr = igbvf_update_itr(adapter, adapter->rx_itr,
 695                                            adapter->total_rx_packets,
 696                                            adapter->total_rx_bytes);
 697         /* conservative mode (itr 3) eliminates the lowest_latency setting */
 698         if (adapter->itr_setting == 3 && adapter->rx_itr == lowest_latency)
 699                 adapter->rx_itr = low_latency;
 700
 701         current_itr = max(adapter->rx_itr, adapter->tx_itr);
 702
 703         switch (current_itr) {
 704         /* counts and packets in update_itr are dependent on these numbers */
 705         case lowest_latency:
 706                 new_itr = 70000;
 707                 break;
 708         case low_latency:
 709                 new_itr = 20000; /* aka hwitr = ~200 */
 710                 break;
 711         case bulk_latency:
 712                 new_itr = 4000;
 713                 break;
 714         default:
 715                 break;
 716         }
 717
 718         if (new_itr != adapter->itr) {
 719                 /*
 720                  * this attempts to bias the interrupt rate towards Bulk
 721                  * by adding intermediate steps when interrupt rate is
 722                  * increasing
 723                  */
 724                 new_itr = new_itr > adapter->itr ?
 725                              min(adapter->itr + (new_itr >> 2), new_itr) :
 726                              new_itr;
 727                 adapter->itr = new_itr;
 728                 adapter->rx_ring->itr_val = 1952;
 729
 730                 if (adapter->msix_entries)
 731                         adapter->rx_ring->set_itr = 1;
 732                 else
 733                         ew32(ITR, 1952);
 734         }
 735 }
 736
 737 /**
 738  * igbvf_clean_tx_irq - Reclaim resources after transmit completes
 739  * @adapter: board private structure
 740  * returns true if ring is completely cleaned
 741  **/
 742 static bool igbvf_clean_tx_irq(struct igbvf_ring *tx_ring)
 743 {
 744         struct igbvf_adapter *adapter = tx_ring->adapter;
 745         struct net_device *netdev = adapter->netdev;
 746         struct igbvf_buffer *buffer_info;
 747         struct sk_buff *skb;
 748         union e1000_adv_tx_desc *tx_desc, *eop_desc;
 749         unsigned int total_bytes = 0, total_packets = 0;
 750         unsigned int i, eop, count = 0;
 751         bool cleaned = false;
 752
 753         i = tx_ring->next_to_clean;
 754         eop = tx_ring->buffer_info[i].next_to_watch;
 755         eop_desc = IGBVF_TX_DESC_ADV(*tx_ring, eop);
 756
 757         while ((eop_desc->wb.status & cpu_to_le32(E1000_TXD_STAT_DD)) &&
 758                (count < tx_ring->count)) {
 759                 rmb();  /* read buffer_info after eop_desc status */
 760                 for (cleaned = false; !cleaned; count++) {
 761                         tx_desc = IGBVF_TX_DESC_ADV(*tx_ring, i);
 762                         buffer_info = &tx_ring->buffer_info[i];
 763                         cleaned = (i == eop);
 764                         skb = buffer_info->skb;
 765
 766                         if (skb) {
 767                                 unsigned int segs, bytecount;
 768
 769                                 /* gso_segs is currently only valid for tcp */
 770                                 segs = skb_shinfo(skb)->gso_segs ?: 1;
 771                                 /* multiply data chunks by size of headers */
 772                                 bytecount = ((segs - 1) * skb_headlen(skb)) +
 773                                             skb->len;
 774                                 total_packets += segs;
 775                                 total_bytes += bytecount;
 776                         }
 777
 778                         igbvf_put_txbuf(adapter, buffer_info);
 779                         tx_desc->wb.status = 0;
 780
 781                         i++;
 782                         if (i == tx_ring->count)
 783                                 i = 0;
 784                 }
 785                 eop = tx_ring->buffer_info[i].next_to_watch;
 786                 eop_desc = IGBVF_TX_DESC_ADV(*tx_ring, eop);
 787         }
 788
 789         tx_ring->next_to_clean = i;
 790
 791         if (unlikely(count &&
 792                      netif_carrier_ok(netdev) &&
 793                      igbvf_desc_unused(tx_ring) >= IGBVF_TX_QUEUE_WAKE)) {
 794                 /* Make sure that anybody stopping the queue after this
 795                  * sees the new next_to_clean.
 796                  */
 797                 smp_mb();
 798                 if (netif_queue_stopped(netdev) &&
 799                     !(test_bit(__IGBVF_DOWN, &adapter->state))) {
 800                         netif_wake_queue(netdev);
 801                         ++adapter->restart_queue;
 802                 }
 803         }
 804
 805         adapter->net_stats.tx_bytes += total_bytes;
 806         adapter->net_stats.tx_packets += total_packets;
 807         return count < tx_ring->count;
 808 }
 809
 810 static irqreturn_t igbvf_msix_other(int irq, void *data)
 811 {
 812         struct net_device *netdev = data;
 813         struct igbvf_adapter *adapter = netdev_priv(netdev);
 814         struct e1000_hw *hw = &adapter->hw;
 815
 816         adapter->int_counter1++;
 817
 818         netif_carrier_off(netdev);
 819         hw->mac.get_link_status = 1;
 820         if (!test_bit(__IGBVF_DOWN, &adapter->state))
 821                 mod_timer(&adapter->watchdog_timer, jiffies + 1);
 822
 823         ew32(EIMS, adapter->eims_other);
 824
 825         return IRQ_HANDLED;
 826 }
 827
 828 static irqreturn_t igbvf_intr_msix_tx(int irq, void *data)
 829 {
 830         struct net_device *netdev = data;
 831         struct igbvf_adapter *adapter = netdev_priv(netdev);
 832         struct e1000_hw *hw = &adapter->hw;
 833         struct igbvf_ring *tx_ring = adapter->tx_ring;
 834
 835
 836         adapter->total_tx_bytes = 0;
 837         adapter->total_tx_packets = 0;
 838
 839         /* auto mask will automatically reenable the interrupt when we write
 840          * EICS */
 841         if (!igbvf_clean_tx_irq(tx_ring))
 842                 /* Ring was not completely cleaned, so fire another interrupt */
 843                 ew32(EICS, tx_ring->eims_value);
 844         else
 845                 ew32(EIMS, tx_ring->eims_value);
 846
 847         return IRQ_HANDLED;
 848 }
 849
 850 static irqreturn_t igbvf_intr_msix_rx(int irq, void *data)
 851 {
 852         struct net_device *netdev = data;
 853         struct igbvf_adapter *adapter = netdev_priv(netdev);
 854
 855         adapter->int_counter0++;
 856
 857         /* Write the ITR value calculated at the end of the
 858          * previous interrupt.
 859          */
 860         if (adapter->rx_ring->set_itr) {
 861                 writel(adapter->rx_ring->itr_val,
 862                        adapter->hw.hw_addr + adapter->rx_ring->itr_register);
 863                 adapter->rx_ring->set_itr = 0;
 864         }
 865
 866         if (napi_schedule_prep(&adapter->rx_ring->napi)) {
 867                 adapter->total_rx_bytes = 0;
 868                 adapter->total_rx_packets = 0;
 869                 __napi_schedule(&adapter->rx_ring->napi);
 870         }
 871
 872         return IRQ_HANDLED;
 873 }
 874
 875 #define IGBVF_NO_QUEUE -1
 876
 877 static void igbvf_assign_vector(struct igbvf_adapter *adapter, int rx_queue,
 878                                 int tx_queue, int msix_vector)
 879 {
 880         struct e1000_hw *hw = &adapter->hw;
 881         u32 ivar, index;
 882
 883         /* 82576 uses a table-based method for assigning vectors.
 884            Each queue has a single entry in the table to which we write
 885            a vector number along with a "valid" bit.  Sadly, the layout
 886            of the table is somewhat counterintuitive. */
 887         if (rx_queue > IGBVF_NO_QUEUE) {
 888                 index = (rx_queue >> 1);
 889                 ivar = array_er32(IVAR0, index);
 890                 if (rx_queue & 0x1) {
 891                         /* vector goes into third byte of register */
 892                         ivar = ivar & 0xFF00FFFF;
 893                         ivar |= (msix_vector | E1000_IVAR_VALID) << 16;
 894                 } else {
 895                         /* vector goes into low byte of register */
 896                         ivar = ivar & 0xFFFFFF00;
 897                         ivar |= msix_vector | E1000_IVAR_VALID;
 898                 }
 899                 adapter->rx_ring[rx_queue].eims_value = 1 << msix_vector;
 900                 array_ew32(IVAR0, index, ivar);
 901         }
 902         if (tx_queue > IGBVF_NO_QUEUE) {
 903                 index = (tx_queue >> 1);
 904                 ivar = array_er32(IVAR0, index);
 905                 if (tx_queue & 0x1) {
 906                         /* vector goes into high byte of register */
 907                         ivar = ivar & 0x00FFFFFF;
 908                         ivar |= (msix_vector | E1000_IVAR_VALID) << 24;
 909                 } else {
 910                         /* vector goes into second byte of register */
 911                         ivar = ivar & 0xFFFF00FF;
 912                         ivar |= (msix_vector | E1000_IVAR_VALID) << 8;
 913                 }
 914                 adapter->tx_ring[tx_queue].eims_value = 1 << msix_vector;
 915                 array_ew32(IVAR0, index, ivar);
 916         }
 917 }
 918
 919 /**
 920  * igbvf_configure_msix - Configure MSI-X hardware
 921  *
 922  * igbvf_configure_msix sets up the hardware to properly
 923  * generate MSI-X interrupts.
 924  **/
 925 static void igbvf_configure_msix(struct igbvf_adapter *adapter)
 926 {
 927         u32 tmp;
 928         struct e1000_hw *hw = &adapter->hw;
 929         struct igbvf_ring *tx_ring = adapter->tx_ring;
 930         struct igbvf_ring *rx_ring = adapter->rx_ring;
 931         int vector = 0;
 932
 933         adapter->eims_enable_mask = 0;
 934
 935         igbvf_assign_vector(adapter, IGBVF_NO_QUEUE, 0, vector++);
 936         adapter->eims_enable_mask |= tx_ring->eims_value;
 937         if (tx_ring->itr_val)
 938                 writel(tx_ring->itr_val,
 939                        hw->hw_addr + tx_ring->itr_register);
 940         else
 941                 writel(1952, hw->hw_addr + tx_ring->itr_register);
 942
 943         igbvf_assign_vector(adapter, 0, IGBVF_NO_QUEUE, vector++);
 944         adapter->eims_enable_mask |= rx_ring->eims_value;
 945         if (rx_ring->itr_val)
 946                 writel(rx_ring->itr_val,
 947                        hw->hw_addr + rx_ring->itr_register);
 948         else
 949                 writel(1952, hw->hw_addr + rx_ring->itr_register);
 950
 951         /* set vector for other causes, i.e. link changes */
 952
 953         tmp = (vector++ | E1000_IVAR_VALID);
 954
 955         ew32(IVAR_MISC, tmp);
 956
 957         adapter->eims_enable_mask = (1 << (vector)) - 1;
 958         adapter->eims_other = 1 << (vector - 1);
 959         e1e_flush();
 960 }
 961
 962 static void igbvf_reset_interrupt_capability(struct igbvf_adapter *adapter)
 963 {
 964         if (adapter->msix_entries) {
 965                 pci_disable_msix(adapter->pdev);
 966                 kfree(adapter->msix_entries);
 967                 adapter->msix_entries = NULL;
 968         }
 969 }
 970
 971 /**
 972  * igbvf_set_interrupt_capability - set MSI or MSI-X if supported
 973  *
 974  * Attempt to configure interrupts using the best available
 975  * capabilities of the hardware and kernel.
 976  **/
 977 static void igbvf_set_interrupt_capability(struct igbvf_adapter *adapter)
 978 {
 979         int err = -ENOMEM;
 980         int i;
 981
 982         /* we allocate 3 vectors, 1 for tx, 1 for rx, one for pf messages */
 983         adapter->msix_entries = kcalloc(3, sizeof(struct msix_entry),
 984                                         GFP_KERNEL);
 985         if (adapter->msix_entries) {
 986                 for (i = 0; i < 3; i++)
 987                         adapter->msix_entries[i].entry = i;
 988
 989                 err = pci_enable_msix(adapter->pdev,
 990                                       adapter->msix_entries, 3);
 991         }
 992
 993         if (err) {
 994                 /* MSI-X failed */
 995                 dev_err(&adapter->pdev->dev,
 996                         "Failed to initialize MSI-X interrupts.\n");
 997                 igbvf_reset_interrupt_capability(adapter);
 998         }
 999 }
1000
1001 /**
1002  * igbvf_request_msix - Initialize MSI-X interrupts
1003  *
1004  * igbvf_request_msix allocates MSI-X vectors and requests interrupts from the
1005  * kernel.
1006  **/
1007 static int igbvf_request_msix(struct igbvf_adapter *adapter)
1008 {
1009         struct net_device *netdev = adapter->netdev;
1010         int err = 0, vector = 0;
1011
1012         if (strlen(netdev->name) < (IFNAMSIZ - 5)) {
1013                 sprintf(adapter->tx_ring->name, "%s-tx-0", netdev->name);
1014                 sprintf(adapter->rx_ring->name, "%s-rx-0", netdev->name);
1015         } else {
1016                 memcpy(adapter->tx_ring->name, netdev->name, IFNAMSIZ);
1017                 memcpy(adapter->rx_ring->name, netdev->name, IFNAMSIZ);
1018         }
1019
1020         err = request_irq(adapter->msix_entries[vector].vector,
1021                           igbvf_intr_msix_tx, 0, adapter->tx_ring->name,
1022                           netdev);
1023         if (err)
1024                 goto out;
1025
1026         adapter->tx_ring->itr_register = E1000_EITR(vector);
1027         adapter->tx_ring->itr_val = 1952;
1028         vector++;
1029
1030         err = request_irq(adapter->msix_entries[vector].vector,
1031                           igbvf_intr_msix_rx, 0, adapter->rx_ring->name,
1032                           netdev);
1033         if (err)
1034                 goto out;
1035
1036         adapter->rx_ring->itr_register = E1000_EITR(vector);
1037         adapter->rx_ring->itr_val = 1952;
1038         vector++;
1039
1040         err = request_irq(adapter->msix_entries[vector].vector,
1041                           igbvf_msix_other, 0, netdev->name, netdev);
1042         if (err)
1043                 goto out;
1044
1045         igbvf_configure_msix(adapter);
1046         return 0;
1047 out:
1048         return err;
1049 }
1050
1051 /**
1052  * igbvf_alloc_queues - Allocate memory for all rings
1053  * @adapter: board private structure to initialize
1054  **/
1055 static int __devinit igbvf_alloc_queues(struct igbvf_adapter *adapter)
1056 {
1057         struct net_device *netdev = adapter->netdev;
1058
1059         adapter->tx_ring = kzalloc(sizeof(struct igbvf_ring), GFP_KERNEL);
1060         if (!adapter->tx_ring)
1061                 return -ENOMEM;
1062
1063         adapter->rx_ring = kzalloc(sizeof(struct igbvf_ring), GFP_KERNEL);
1064         if (!adapter->rx_ring) {
1065                 kfree(adapter->tx_ring);
1066                 return -ENOMEM;
1067         }
1068
1069         netif_napi_add(netdev, &adapter->rx_ring->napi, igbvf_poll, 64);
1070
1071         return 0;
1072 }
1073
1074 /**
1075  * igbvf_request_irq - initialize interrupts
1076  *
1077  * Attempts to configure interrupts using the best available
1078  * capabilities of the hardware and kernel.
1079  **/
1080 static int igbvf_request_irq(struct igbvf_adapter *adapter)
1081 {
1082         int err = -1;
1083
1084         /* igbvf supports msi-x only */
1085         if (adapter->msix_entries)
1086                 err = igbvf_request_msix(adapter);
1087
1088         if (!err)
1089                 return err;
1090
1091         dev_err(&adapter->pdev->dev,
1092                 "Unable to allocate interrupt, Error: %d\n", err);
1093
1094         return err;
1095 }
1096
1097 static void igbvf_free_irq(struct igbvf_adapter *adapter)
1098 {
1099         struct net_device *netdev = adapter->netdev;
1100         int vector;
1101
1102         if (adapter->msix_entries) {
1103                 for (vector = 0; vector < 3; vector++)
1104                         free_irq(adapter->msix_entries[vector].vector, netdev);
1105         }
1106 }
1107
1108 /**
1109  * igbvf_irq_disable - Mask off interrupt generation on the NIC
1110  **/
1111 static void igbvf_irq_disable(struct igbvf_adapter *adapter)
1112 {
1113         struct e1000_hw *hw = &adapter->hw;
1114
1115         ew32(EIMC, ~0);
1116
1117         if (adapter->msix_entries)
1118                 ew32(EIAC, 0);
1119 }
1120
1121 /**
1122  * igbvf_irq_enable - Enable default interrupt generation settings
1123  **/
1124 static void igbvf_irq_enable(struct igbvf_adapter *adapter)
1125 {
1126         struct e1000_hw *hw = &adapter->hw;
1127
1128         ew32(EIAC, adapter->eims_enable_mask);
1129         ew32(EIAM, adapter->eims_enable_mask);
1130         ew32(EIMS, adapter->eims_enable_mask);
1131 }
1132
1133 /**
1134  * igbvf_poll - NAPI Rx polling callback
1135  * @napi: struct associated with this polling callback
1136  * @budget: amount of packets driver is allowed to process this poll
1137  **/
1138 static int igbvf_poll(struct napi_struct *napi, int budget)
1139 {
1140         struct igbvf_ring *rx_ring = container_of(napi, struct igbvf_ring, napi);
1141         struct igbvf_adapter *adapter = rx_ring->adapter;
1142         struct e1000_hw *hw = &adapter->hw;
1143         int work_done = 0;
1144
1145         igbvf_clean_rx_irq(adapter, &work_done, budget);
1146
1147         /* If not enough Rx work done, exit the polling mode */
1148         if (work_done < budget) {
1149                 napi_complete(napi);
1150
1151                 if (adapter->itr_setting & 3)
1152                         igbvf_set_itr(adapter);
1153
1154                 if (!test_bit(__IGBVF_DOWN, &adapter->state))
1155                         ew32(EIMS, adapter->rx_ring->eims_value);
1156         }
1157
1158         return work_done;
1159 }
1160
1161 /**
1162  * igbvf_set_rlpml - set receive large packet maximum length
1163  * @adapter: board private structure
1164  *
1165  * Configure the maximum size of packets that will be received
1166  */
1167 static void igbvf_set_rlpml(struct igbvf_adapter *adapter)
1168 {
1169         int max_frame_size = adapter->max_frame_size;
1170         struct e1000_hw *hw = &adapter->hw;
1171
1172         if (adapter->vlgrp)
1173                 max_frame_size += VLAN_TAG_SIZE;
1174
1175         e1000_rlpml_set_vf(hw, max_frame_size);
1176 }
1177
1178 static void igbvf_vlan_rx_add_vid(struct net_device *netdev, u16 vid)
1179 {
1180         struct igbvf_adapter *adapter = netdev_priv(netdev);
1181         struct e1000_hw *hw = &adapter->hw;
1182
1183         if (hw->mac.ops.set_vfta(hw, vid, true))
1184                 dev_err(&adapter->pdev->dev, "Failed to add vlan id %d\n", vid);
1185 }
1186
1187 static void igbvf_vlan_rx_kill_vid(struct net_device *netdev, u16 vid)
1188 {
1189         struct igbvf_adapter *adapter = netdev_priv(netdev);
1190         struct e1000_hw *hw = &adapter->hw;
1191
1192         igbvf_irq_disable(adapter);
1193         vlan_group_set_device(adapter->vlgrp, vid, NULL);
1194
1195         if (!test_bit(__IGBVF_DOWN, &adapter->state))
1196                 igbvf_irq_enable(adapter);
1197
1198         if (hw->mac.ops.set_vfta(hw, vid, false))
1199                 dev_err(&adapter->pdev->dev,
1200                         "Failed to remove vlan id %d\n", vid);
1201 }
1202
1203 static void igbvf_vlan_rx_register(struct net_device *netdev,
1204                                    struct vlan_group *grp)
1205 {
1206         struct igbvf_adapter *adapter = netdev_priv(netdev);
1207
1208         adapter->vlgrp = grp;
1209 }
1210
1211 static void igbvf_restore_vlan(struct igbvf_adapter *adapter)
1212 {
1213         u16 vid;
1214
1215         if (!adapter->vlgrp)
1216                 return;
1217
1218         for (vid = 0; vid < VLAN_N_VID; vid++) {
1219                 if (!vlan_group_get_device(adapter->vlgrp, vid))
1220                         continue;
1221                 igbvf_vlan_rx_add_vid(adapter->netdev, vid);
1222         }
1223
1224         igbvf_set_rlpml(adapter);
1225 }
1226
1227 /**
1228  * igbvf_configure_tx - Configure Transmit Unit after Reset
1229  * @adapter: board private structure
1230  *
1231  * Configure the Tx unit of the MAC after a reset.
1232  **/
1233 static void igbvf_configure_tx(struct igbvf_adapter *adapter)
1234 {
1235         struct e1000_hw *hw = &adapter->hw;
1236         struct igbvf_ring *tx_ring = adapter->tx_ring;
1237         u64 tdba;
1238         u32 txdctl, dca_txctrl;
1239
1240         /* disable transmits */
1241         txdctl = er32(TXDCTL(0));
1242         ew32(TXDCTL(0), txdctl & ~E1000_TXDCTL_QUEUE_ENABLE);
1243         msleep(10);
1244
1245         /* Setup the HW Tx Head and Tail descriptor pointers */
1246         ew32(TDLEN(0), tx_ring->count * sizeof(union e1000_adv_tx_desc));
1247         tdba = tx_ring->dma;
1248         ew32(TDBAL(0), (tdba & DMA_BIT_MASK(32)));
1249         ew32(TDBAH(0), (tdba >> 32));
1250         ew32(TDH(0), 0);
1251         ew32(TDT(0), 0);
1252         tx_ring->head = E1000_TDH(0);
1253         tx_ring->tail = E1000_TDT(0);
1254
1255         /* Turn off Relaxed Ordering on head write-backs.  The writebacks
1256          * MUST be delivered in order or it will completely screw up
1257          * our bookeeping.
1258          */
1259         dca_txctrl = er32(DCA_TXCTRL(0));
1260         dca_txctrl &= ~E1000_DCA_TXCTRL_TX_WB_RO_EN;
1261         ew32(DCA_TXCTRL(0), dca_txctrl);
1262
1263         /* enable transmits */
1264         txdctl |= E1000_TXDCTL_QUEUE_ENABLE;
1265         ew32(TXDCTL(0), txdctl);
1266
1267         /* Setup Transmit Descriptor Settings for eop descriptor */
1268         adapter->txd_cmd = E1000_ADVTXD_DCMD_EOP | E1000_ADVTXD_DCMD_IFCS;
1269
1270         /* enable Report Status bit */
1271         adapter->txd_cmd |= E1000_ADVTXD_DCMD_RS;
1272 }
1273
1274 /**
1275  * igbvf_setup_srrctl - configure the receive control registers
1276  * @adapter: Board private structure
1277  **/
1278 static void igbvf_setup_srrctl(struct igbvf_adapter *adapter)
1279 {
1280         struct e1000_hw *hw = &adapter->hw;
1281         u32 srrctl = 0;
1282
1283         srrctl &= ~(E1000_SRRCTL_DESCTYPE_MASK |
1284                     E1000_SRRCTL_BSIZEHDR_MASK |
1285                     E1000_SRRCTL_BSIZEPKT_MASK);
1286
1287         /* Enable queue drop to avoid head of line blocking */
1288         srrctl |= E1000_SRRCTL_DROP_EN;
1289
1290         /* Setup buffer sizes */
1291         srrctl |= ALIGN(adapter->rx_buffer_len, 1024) >>
1292                   E1000_SRRCTL_BSIZEPKT_SHIFT;
1293
1294         if (adapter->rx_buffer_len < 2048) {
1295                 adapter->rx_ps_hdr_size = 0;
1296                 srrctl |= E1000_SRRCTL_DESCTYPE_ADV_ONEBUF;
1297         } else {
1298                 adapter->rx_ps_hdr_size = 128;
1299                 srrctl |= adapter->rx_ps_hdr_size <<
1300                           E1000_SRRCTL_BSIZEHDRSIZE_SHIFT;
1301                 srrctl |= E1000_SRRCTL_DESCTYPE_HDR_SPLIT_ALWAYS;
1302         }
1303
1304         ew32(SRRCTL(0), srrctl);
1305 }
1306
1307 /**
1308  * igbvf_configure_rx - Configure Receive Unit after Reset
1309  * @adapter: board private structure
1310  *
1311  * Configure the Rx unit of the MAC after a reset.
1312  **/
1313 static void igbvf_configure_rx(struct igbvf_adapter *adapter)
1314 {
1315         struct e1000_hw *hw = &adapter->hw;
1316         struct igbvf_ring *rx_ring = adapter->rx_ring;
1317         u64 rdba;
1318         u32 rdlen, rxdctl;
1319
1320         /* disable receives */
1321         rxdctl = er32(RXDCTL(0));
1322         ew32(RXDCTL(0), rxdctl & ~E1000_RXDCTL_QUEUE_ENABLE);
1323         msleep(10);
1324
1325         rdlen = rx_ring->count * sizeof(union e1000_adv_rx_desc);
1326
1327         /*
1328          * Setup the HW Rx Head and Tail Descriptor Pointers and
1329          * the Base and Length of the Rx Descriptor Ring
1330          */
1331         rdba = rx_ring->dma;
1332         ew32(RDBAL(0), (rdba & DMA_BIT_MASK(32)));
1333         ew32(RDBAH(0), (rdba >> 32));
1334         ew32(RDLEN(0), rx_ring->count * sizeof(union e1000_adv_rx_desc));
1335         rx_ring->head = E1000_RDH(0);
1336         rx_ring->tail = E1000_RDT(0);
1337         ew32(RDH(0), 0);
1338         ew32(RDT(0), 0);
1339
1340         rxdctl |= E1000_RXDCTL_QUEUE_ENABLE;
1341         rxdctl &= 0xFFF00000;
1342         rxdctl |= IGBVF_RX_PTHRESH;
1343         rxdctl |= IGBVF_RX_HTHRESH << 8;
1344         rxdctl |= IGBVF_RX_WTHRESH << 16;
1345
1346         igbvf_set_rlpml(adapter);
1347
1348         /* enable receives */
1349         ew32(RXDCTL(0), rxdctl);
1350 }
1351
1352 /**
1353  * igbvf_set_multi - Multicast and Promiscuous mode set
1354  * @netdev: network interface device structure
1355  *
1356  * The set_multi entry point is called whenever the multicast address
1357  * list or the network interface flags are updated.  This routine is
1358  * responsible for configuring the hardware for proper multicast,
1359  * promiscuous mode, and all-multi behavior.
1360  **/
1361 static void igbvf_set_multi(struct net_device *netdev)
1362 {
1363         struct igbvf_adapter *adapter = netdev_priv(netdev);
1364         struct e1000_hw *hw = &adapter->hw;
1365         struct netdev_hw_addr *ha;
1366         u8  *mta_list = NULL;
1367         int i;
1368
1369         if (!netdev_mc_empty(netdev)) {
1370                 mta_list = kmalloc(netdev_mc_count(netdev) * 6, GFP_ATOMIC);
1371                 if (!mta_list) {
1372                         dev_err(&adapter->pdev->dev,
1373                                 "failed to allocate multicast filter list\n");
1374                         return;
1375                 }
1376         }
1377
1378         /* prepare a packed array of only addresses. */
1379         i = 0;
1380         netdev_for_each_mc_addr(ha, netdev)
1381                 memcpy(mta_list + (i++ * ETH_ALEN), ha->addr, ETH_ALEN);
1382
1383         hw->mac.ops.update_mc_addr_list(hw, mta_list, i, 0, 0);
1384         kfree(mta_list);
1385 }
1386
1387 /**
1388  * igbvf_configure - configure the hardware for Rx and Tx
1389  * @adapter: private board structure
1390  **/
1391 static void igbvf_configure(struct igbvf_adapter *adapter)
1392 {
1393         igbvf_set_multi(adapter->netdev);
1394
1395         igbvf_restore_vlan(adapter);
1396
1397         igbvf_configure_tx(adapter);
1398         igbvf_setup_srrctl(adapter);
1399         igbvf_configure_rx(adapter);
1400         igbvf_alloc_rx_buffers(adapter->rx_ring,
1401                                igbvf_desc_unused(adapter->rx_ring));
1402 }
1403
1404 /* igbvf_reset - bring the hardware into a known good state
1405  *
1406  * This function boots the hardware and enables some settings that
1407  * require a configuration cycle of the hardware - those cannot be
1408  * set/changed during runtime. After reset the device needs to be
1409  * properly configured for Rx, Tx etc.
1410  */
1411 static void igbvf_reset(struct igbvf_adapter *adapter)
1412 {
1413         struct e1000_mac_info *mac = &adapter->hw.mac;
1414         struct net_device *netdev = adapter->netdev;
1415         struct e1000_hw *hw = &adapter->hw;
1416
1417         /* Allow time for pending master requests to run */
1418         if (mac->ops.reset_hw(hw))
1419                 dev_err(&adapter->pdev->dev, "PF still resetting\n");
1420
1421         mac->ops.init_hw(hw);
1422
1423         if (is_valid_ether_addr(adapter->hw.mac.addr)) {
1424                 memcpy(netdev->dev_addr, adapter->hw.mac.addr,
1425                        netdev->addr_len);
1426                 memcpy(netdev->perm_addr, adapter->hw.mac.addr,
1427                        netdev->addr_len);
1428         }
1429
1430         adapter->last_reset = jiffies;
1431 }
1432
1433 int igbvf_up(struct igbvf_adapter *adapter)
1434 {
1435         struct e1000_hw *hw = &adapter->hw;
1436
1437         /* hardware has been reset, we need to reload some things */
1438         igbvf_configure(adapter);
1439
1440         clear_bit(__IGBVF_DOWN, &adapter->state);
1441
1442         napi_enable(&adapter->rx_ring->napi);
1443         if (adapter->msix_entries)
1444                 igbvf_configure_msix(adapter);
1445
1446         /* Clear any pending interrupts. */
1447         er32(EICR);
1448         igbvf_irq_enable(adapter);
1449
1450         /* start the watchdog */
1451         hw->mac.get_link_status = 1;
1452         mod_timer(&adapter->watchdog_timer, jiffies + 1);
1453
1454
1455         return 0;
1456 }
1457
1458 void igbvf_down(struct igbvf_adapter *adapter)
1459 {
1460         struct net_device *netdev = adapter->netdev;
1461         struct e1000_hw *hw = &adapter->hw;
1462         u32 rxdctl, txdctl;
1463
1464         /*
1465          * signal that we're down so the interrupt handler does not
1466          * reschedule our watchdog timer
1467          */
1468         set_bit(__IGBVF_DOWN, &adapter->state);
1469
1470         /* disable receives in the hardware */
1471         rxdctl = er32(RXDCTL(0));
1472         ew32(RXDCTL(0), rxdctl & ~E1000_RXDCTL_QUEUE_ENABLE);
1473
1474         netif_stop_queue(netdev);
1475
1476         /* disable transmits in the hardware */
1477         txdctl = er32(TXDCTL(0));
1478         ew32(TXDCTL(0), txdctl & ~E1000_TXDCTL_QUEUE_ENABLE);
1479
1480         /* flush both disables and wait for them to finish */
1481         e1e_flush();
1482         msleep(10);
1483
1484         napi_disable(&adapter->rx_ring->napi);
1485
1486         igbvf_irq_disable(adapter);
1487
1488         del_timer_sync(&adapter->watchdog_timer);
1489
1490         netif_carrier_off(netdev);
1491
1492         /* record the stats before reset*/
1493         igbvf_update_stats(adapter);
1494
1495         adapter->link_speed = 0;
1496         adapter->link_duplex = 0;
1497
1498         igbvf_reset(adapter);
1499         igbvf_clean_tx_ring(adapter->tx_ring);
1500         igbvf_clean_rx_ring(adapter->rx_ring);
1501 }
1502
1503 void igbvf_reinit_locked(struct igbvf_adapter *adapter)
1504 {
1505         might_sleep();
1506         while (test_and_set_bit(__IGBVF_RESETTING, &adapter->state))
1507                 msleep(1);
1508         igbvf_down(adapter);
1509         igbvf_up(adapter);
1510         clear_bit(__IGBVF_RESETTING, &adapter->state);
1511 }
1512
1513 /**
1514  * igbvf_sw_init - Initialize general software structures (struct igbvf_adapter)
1515  * @adapter: board private structure to initialize
1516  *
1517  * igbvf_sw_init initializes the Adapter private data structure.
1518  * Fields are initialized based on PCI device information and
1519  * OS network device settings (MTU size).
1520  **/
1521 static int __devinit igbvf_sw_init(struct igbvf_adapter *adapter)
1522 {
1523         struct net_device *netdev = adapter->netdev;
1524         s32 rc;
1525
1526         adapter->rx_buffer_len = ETH_FRAME_LEN + VLAN_HLEN + ETH_FCS_LEN;
1527         adapter->rx_ps_hdr_size = 0;
1528         adapter->max_frame_size = netdev->mtu + ETH_HLEN + ETH_FCS_LEN;
1529         adapter->min_frame_size = ETH_ZLEN + ETH_FCS_LEN;
1530
1531         adapter->tx_int_delay = 8;
1532         adapter->tx_abs_int_delay = 32;
1533         adapter->rx_int_delay = 0;
1534         adapter->rx_abs_int_delay = 8;
1535         adapter->itr_setting = 3;
1536         adapter->itr = 20000;
1537
1538         /* Set various function pointers */
1539         adapter->ei->init_ops(&adapter->hw);
1540
1541         rc = adapter->hw.mac.ops.init_params(&adapter->hw);
1542         if (rc)
1543                 return rc;
1544
1545         rc = adapter->hw.mbx.ops.init_params(&adapter->hw);
1546         if (rc)
1547                 return rc;
1548
1549         igbvf_set_interrupt_capability(adapter);
1550
1551         if (igbvf_alloc_queues(adapter))
1552                 return -ENOMEM;
1553
1554         spin_lock_init(&adapter->tx_queue_lock);
1555
1556         /* Explicitly disable IRQ since the NIC can be in any state. */
1557         igbvf_irq_disable(adapter);
1558
1559         spin_lock_init(&adapter->stats_lock);
1560
1561         set_bit(__IGBVF_DOWN, &adapter->state);
1562         return 0;
1563 }
1564
1565 static void igbvf_initialize_last_counter_stats(struct igbvf_adapter *adapter)
1566 {
1567         struct e1000_hw *hw = &adapter->hw;
1568
1569         adapter->stats.last_gprc = er32(VFGPRC);
1570         adapter->stats.last_gorc = er32(VFGORC);
1571         adapter->stats.last_gptc = er32(VFGPTC);
1572         adapter->stats.last_gotc = er32(VFGOTC);
1573         adapter->stats.last_mprc = er32(VFMPRC);
1574         adapter->stats.last_gotlbc = er32(VFGOTLBC);
1575         adapter->stats.last_gptlbc = er32(VFGPTLBC);
1576         adapter->stats.last_gorlbc = er32(VFGORLBC);
1577         adapter->stats.last_gprlbc = er32(VFGPRLBC);
1578
1579         adapter->stats.base_gprc = er32(VFGPRC);
1580         adapter->stats.base_gorc = er32(VFGORC);
1581         adapter->stats.base_gptc = er32(VFGPTC);
1582         adapter->stats.base_gotc = er32(VFGOTC);
1583         adapter->stats.base_mprc = er32(VFMPRC);
1584         adapter->stats.base_gotlbc = er32(VFGOTLBC);
1585         adapter->stats.base_gptlbc = er32(VFGPTLBC);
1586         adapter->stats.base_gorlbc = er32(VFGORLBC);
1587         adapter->stats.base_gprlbc = er32(VFGPRLBC);
1588 }
1589
1590 /**
1591  * igbvf_open - Called when a network interface is made active
1592  * @netdev: network interface device structure
1593  *
1594  * Returns 0 on success, negative value on failure
1595  *
1596  * The open entry point is called when a network interface is made
1597  * active by the system (IFF_UP).  At this point all resources needed
1598  * for transmit and receive operations are allocated, the interrupt
1599  * handler is registered with the OS, the watchdog timer is started,
1600  * and the stack is notified that the interface is ready.
1601  **/
1602 static int igbvf_open(struct net_device *netdev)
1603 {
1604         struct igbvf_adapter *adapter = netdev_priv(netdev);
1605         struct e1000_hw *hw = &adapter->hw;
1606         int err;
1607
1608         /* disallow open during test */
1609         if (test_bit(__IGBVF_TESTING, &adapter->state))
1610                 return -EBUSY;
1611
1612         /* allocate transmit descriptors */
1613         err = igbvf_setup_tx_resources(adapter, adapter->tx_ring);
1614         if (err)
1615                 goto err_setup_tx;
1616
1617         /* allocate receive descriptors */
1618         err = igbvf_setup_rx_resources(adapter, adapter->rx_ring);
1619         if (err)
1620                 goto err_setup_rx;
1621
1622         /*
1623          * before we allocate an interrupt, we must be ready to handle it.
1624          * Setting DEBUG_SHIRQ in the kernel makes it fire an interrupt
1625          * as soon as we call pci_request_irq, so we have to setup our
1626          * clean_rx handler before we do so.
1627          */
1628         igbvf_configure(adapter);
1629
1630         err = igbvf_request_irq(adapter);
1631         if (err)
1632                 goto err_req_irq;
1633
1634         /* From here on the code is the same as igbvf_up() */
1635         clear_bit(__IGBVF_DOWN, &adapter->state);
1636
1637         napi_enable(&adapter->rx_ring->napi);
1638
1639         /* clear any pending interrupts */
1640         er32(EICR);
1641
1642         igbvf_irq_enable(adapter);
1643
1644         /* start the watchdog */
1645         hw->mac.get_link_status = 1;
1646         mod_timer(&adapter->watchdog_timer, jiffies + 1);
1647
1648         return 0;
1649
1650 err_req_irq:
1651         igbvf_free_rx_resources(adapter->rx_ring);
1652 err_setup_rx:
1653         igbvf_free_tx_resources(adapter->tx_ring);
1654 err_setup_tx:
1655         igbvf_reset(adapter);
1656
1657         return err;
1658 }
1659
1660 /**
1661  * igbvf_close - Disables a network interface
1662  * @netdev: network interface device structure
1663  *
1664  * Returns 0, this is not allowed to fail
1665  *
1666  * The close entry point is called when an interface is de-activated
1667  * by the OS.  The hardware is still under the drivers control, but
1668  * needs to be disabled.  A global MAC reset is issued to stop the
1669  * hardware, and all transmit and receive resources are freed.
1670  **/
1671 static int igbvf_close(struct net_device *netdev)
1672 {
1673         struct igbvf_adapter *adapter = netdev_priv(netdev);
1674
1675         WARN_ON(test_bit(__IGBVF_RESETTING, &adapter->state));
1676         igbvf_down(adapter);
1677
1678         igbvf_free_irq(adapter);
1679
1680         igbvf_free_tx_resources(adapter->tx_ring);
1681         igbvf_free_rx_resources(adapter->rx_ring);
1682
1683         return 0;
1684 }
1685 /**
1686  * igbvf_set_mac - Change the Ethernet Address of the NIC
1687  * @netdev: network interface device structure
1688  * @p: pointer to an address structure
1689  *
1690  * Returns 0 on success, negative on failure
1691  **/
1692 static int igbvf_set_mac(struct net_device *netdev, void *p)
1693 {
1694         struct igbvf_adapter *adapter = netdev_priv(netdev);
1695         struct e1000_hw *hw = &adapter->hw;
1696         struct sockaddr *addr = p;
1697
1698         if (!is_valid_ether_addr(addr->sa_data))
1699                 return -EADDRNOTAVAIL;
1700
1701         memcpy(hw->mac.addr, addr->sa_data, netdev->addr_len);
1702
1703         hw->mac.ops.rar_set(hw, hw->mac.addr, 0);
1704
1705         if (memcmp(addr->sa_data, hw->mac.addr, 6))
1706                 return -EADDRNOTAVAIL;
1707
1708         memcpy(netdev->dev_addr, addr->sa_data, netdev->addr_len);
1709
1710         return 0;
1711 }
1712
1713 #define UPDATE_VF_COUNTER(reg, name)                                    \
1714         {                                                               \
1715                 u32 current_counter = er32(reg);                        \
1716                 if (current_counter < adapter->stats.last_##name)       \
1717                         adapter->stats.name += 0x100000000LL;           \
1718                 adapter->stats.last_##name = current_counter;           \
1719                 adapter->stats.name &= 0xFFFFFFFF00000000LL;            \
1720                 adapter->stats.name |= current_counter;                 \
1721         }
1722
1723 /**
1724  * igbvf_update_stats - Update the board statistics counters
1725  * @adapter: board private structure
1726 **/
1727 void igbvf_update_stats(struct igbvf_adapter *adapter)
1728 {
1729         struct e1000_hw *hw = &adapter->hw;
1730         struct pci_dev *pdev = adapter->pdev;
1731
1732         /*
1733          * Prevent stats update while adapter is being reset, link is down
1734          * or if the pci connection is down.
1735          */
1736         if (adapter->link_speed == 0)
1737                 return;
1738
1739         if (test_bit(__IGBVF_RESETTING, &adapter->state))
1740                 return;
1741
1742         if (pci_channel_offline(pdev))
1743                 return;
1744
1745         UPDATE_VF_COUNTER(VFGPRC, gprc);
1746         UPDATE_VF_COUNTER(VFGORC, gorc);
1747         UPDATE_VF_COUNTER(VFGPTC, gptc);
1748         UPDATE_VF_COUNTER(VFGOTC, gotc);
1749         UPDATE_VF_COUNTER(VFMPRC, mprc);
1750         UPDATE_VF_COUNTER(VFGOTLBC, gotlbc);
1751         UPDATE_VF_COUNTER(VFGPTLBC, gptlbc);
1752         UPDATE_VF_COUNTER(VFGORLBC, gorlbc);
1753         UPDATE_VF_COUNTER(VFGPRLBC, gprlbc);
1754
1755         /* Fill out the OS statistics structure */
1756         adapter->net_stats.multicast = adapter->stats.mprc;
1757 }
1758
1759 static void igbvf_print_link_info(struct igbvf_adapter *adapter)
1760 {
1761         dev_info(&adapter->pdev->dev, "Link is Up %d Mbps %s\n",
1762                  adapter->link_speed,
1763                  ((adapter->link_duplex == FULL_DUPLEX) ?
1764                   "Full Duplex" : "Half Duplex"));
1765 }
1766
1767 static bool igbvf_has_link(struct igbvf_adapter *adapter)
1768 {
1769         struct e1000_hw *hw = &adapter->hw;
1770         s32 ret_val = E1000_SUCCESS;
1771         bool link_active;
1772
1773         /* If interface is down, stay link down */
1774         if (test_bit(__IGBVF_DOWN, &adapter->state))
1775                 return false;
1776
1777         ret_val = hw->mac.ops.check_for_link(hw);
1778         link_active = !hw->mac.get_link_status;
1779
1780         /* if check for link returns error we will need to reset */
1781         if (ret_val && time_after(jiffies, adapter->last_reset + (10 * HZ)))
1782                 schedule_work(&adapter->reset_task);
1783
1784         return link_active;
1785 }
1786
1787 /**
1788  * igbvf_watchdog - Timer Call-back
1789  * @data: pointer to adapter cast into an unsigned long
1790  **/
1791 static void igbvf_watchdog(unsigned long data)
1792 {
1793         struct igbvf_adapter *adapter = (struct igbvf_adapter *) data;
1794
1795         /* Do the rest outside of interrupt context */
1796         schedule_work(&adapter->watchdog_task);
1797 }
1798
1799 static void igbvf_watchdog_task(struct work_struct *work)
1800 {
1801         struct igbvf_adapter *adapter = container_of(work,
1802                                                      struct igbvf_adapter,
1803                                                      watchdog_task);
1804         struct net_device *netdev = adapter->netdev;
1805         struct e1000_mac_info *mac = &adapter->hw.mac;
1806         struct igbvf_ring *tx_ring = adapter->tx_ring;
1807         struct e1000_hw *hw = &adapter->hw;
1808         u32 link;
1809         int tx_pending = 0;
1810
1811         link = igbvf_has_link(adapter);
1812
1813         if (link) {
1814                 if (!netif_carrier_ok(netdev)) {
1815                         mac->ops.get_link_up_info(&adapter->hw,
1816                                                   &adapter->link_speed,
1817                                                   &adapter->link_duplex);
1818                         igbvf_print_link_info(adapter);
1819
1820                         netif_carrier_on(netdev);
1821                         netif_wake_queue(netdev);
1822                 }
1823         } else {
1824                 if (netif_carrier_ok(netdev)) {
1825                         adapter->link_speed = 0;
1826                         adapter->link_duplex = 0;
1827                         dev_info(&adapter->pdev->dev, "Link is Down\n");
1828                         netif_carrier_off(netdev);
1829                         netif_stop_queue(netdev);
1830                 }
1831         }
1832
1833         if (netif_carrier_ok(netdev)) {
1834                 igbvf_update_stats(adapter);
1835         } else {
1836                 tx_pending = (igbvf_desc_unused(tx_ring) + 1 <
1837                               tx_ring->count);
1838                 if (tx_pending) {
1839                         /*
1840                          * We've lost link, so the controller stops DMA,
1841                          * but we've got queued Tx work that's never going
1842                          * to get done, so reset controller to flush Tx.
1843                          * (Do the reset outside of interrupt context).
1844                          */
1845                         adapter->tx_timeout_count++;
1846                         schedule_work(&adapter->reset_task);
1847                 }
1848         }
1849
1850         /* Cause software interrupt to ensure Rx ring is cleaned */
1851         ew32(EICS, adapter->rx_ring->eims_value);
1852
1853         /* Reset the timer */
1854         if (!test_bit(__IGBVF_DOWN, &adapter->state))
1855                 mod_timer(&adapter->watchdog_timer,
1856                           round_jiffies(jiffies + (2 * HZ)));
1857 }
1858
1859 #define IGBVF_TX_FLAGS_CSUM             0x00000001
1860 #define IGBVF_TX_FLAGS_VLAN             0x00000002
1861 #define IGBVF_TX_FLAGS_TSO              0x00000004
1862 #define IGBVF_TX_FLAGS_IPV4             0x00000008
1863 #define IGBVF_TX_FLAGS_VLAN_MASK        0xffff0000
1864 #define IGBVF_TX_FLAGS_VLAN_SHIFT       16
1865
1866 static int igbvf_tso(struct igbvf_adapter *adapter,
1867                      struct igbvf_ring *tx_ring,
1868                      struct sk_buff *skb, u32 tx_flags, u8 *hdr_len)
1869 {
1870         struct e1000_adv_tx_context_desc *context_desc;
1871         unsigned int i;
1872         int err;
1873         struct igbvf_buffer *buffer_info;
1874         u32 info = 0, tu_cmd = 0;
1875         u32 mss_l4len_idx, l4len;
1876         *hdr_len = 0;
1877
1878         if (skb_header_cloned(skb)) {
1879                 err = pskb_expand_head(skb, 0, 0, GFP_ATOMIC);
1880                 if (err) {
1881                         dev_err(&adapter->pdev->dev,
1882                                 "igbvf_tso returning an error\n");
1883                         return err;
1884                 }
1885         }
1886
1887         l4len = tcp_hdrlen(skb);
1888         *hdr_len += l4len;
1889
1890         if (skb->protocol == htons(ETH_P_IP)) {
1891                 struct iphdr *iph = ip_hdr(skb);
1892                 iph->tot_len = 0;
1893                 iph->check = 0;
1894                 tcp_hdr(skb)->check = ~csum_tcpudp_magic(iph->saddr,
1895                                                          iph->daddr, 0,
1896                                                          IPPROTO_TCP,
1897                                                          0);
1898         } else if (skb_is_gso_v6(skb)) {
1899                 ipv6_hdr(skb)->payload_len = 0;
1900                 tcp_hdr(skb)->check = ~csum_ipv6_magic(&ipv6_hdr(skb)->saddr,
1901                                                        &ipv6_hdr(skb)->daddr,
1902                                                        0, IPPROTO_TCP, 0);
1903         }
1904
1905         i = tx_ring->next_to_use;
1906
1907         buffer_info = &tx_ring->buffer_info[i];
1908         context_desc = IGBVF_TX_CTXTDESC_ADV(*tx_ring, i);
1909         /* VLAN MACLEN IPLEN */
1910         if (tx_flags & IGBVF_TX_FLAGS_VLAN)
1911                 info |= (tx_flags & IGBVF_TX_FLAGS_VLAN_MASK);
1912         info |= (skb_network_offset(skb) << E1000_ADVTXD_MACLEN_SHIFT);
1913         *hdr_len += skb_network_offset(skb);
1914         info |= (skb_transport_header(skb) - skb_network_header(skb));
1915         *hdr_len += (skb_transport_header(skb) - skb_network_header(skb));
1916         context_desc->vlan_macip_lens = cpu_to_le32(info);
1917
1918         /* ADV DTYP TUCMD MKRLOC/ISCSIHEDLEN */
1919         tu_cmd |= (E1000_TXD_CMD_DEXT | E1000_ADVTXD_DTYP_CTXT);
1920
1921         if (skb->protocol == htons(ETH_P_IP))
1922                 tu_cmd |= E1000_ADVTXD_TUCMD_IPV4;
1923         tu_cmd |= E1000_ADVTXD_TUCMD_L4T_TCP;
1924
1925         context_desc->type_tucmd_mlhl = cpu_to_le32(tu_cmd);
1926
1927         /* MSS L4LEN IDX */
1928         mss_l4len_idx = (skb_shinfo(skb)->gso_size << E1000_ADVTXD_MSS_SHIFT);
1929         mss_l4len_idx |= (l4len << E1000_ADVTXD_L4LEN_SHIFT);
1930
1931         context_desc->mss_l4len_idx = cpu_to_le32(mss_l4len_idx);
1932         context_desc->seqnum_seed = 0;
1933
1934         buffer_info->time_stamp = jiffies;
1935         buffer_info->next_to_watch = i;
1936         buffer_info->dma = 0;
1937         i++;
1938         if (i == tx_ring->count)
1939                 i = 0;
1940
1941         tx_ring->next_to_use = i;
1942
1943         return true;
1944 }
1945
1946 static inline bool igbvf_tx_csum(struct igbvf_adapter *adapter,
1947                                  struct igbvf_ring *tx_ring,
1948                                  struct sk_buff *skb, u32 tx_flags)
1949 {
1950         struct e1000_adv_tx_context_desc *context_desc;
1951         unsigned int i;
1952         struct igbvf_buffer *buffer_info;
1953         u32 info = 0, tu_cmd = 0;
1954
1955         if ((skb->ip_summed == CHECKSUM_PARTIAL) ||
1956             (tx_flags & IGBVF_TX_FLAGS_VLAN)) {
1957                 i = tx_ring->next_to_use;
1958                 buffer_info = &tx_ring->buffer_info[i];
1959                 context_desc = IGBVF_TX_CTXTDESC_ADV(*tx_ring, i);
1960
1961                 if (tx_flags & IGBVF_TX_FLAGS_VLAN)
1962                         info |= (tx_flags & IGBVF_TX_FLAGS_VLAN_MASK);
1963
1964                 info |= (skb_network_offset(skb) << E1000_ADVTXD_MACLEN_SHIFT);
1965                 if (skb->ip_summed == CHECKSUM_PARTIAL)
1966                         info |= (skb_transport_header(skb) -
1967                                  skb_network_header(skb));
1968
1969
1970                 context_desc->vlan_macip_lens = cpu_to_le32(info);
1971
1972                 tu_cmd |= (E1000_TXD_CMD_DEXT | E1000_ADVTXD_DTYP_CTXT);
1973
1974                 if (skb->ip_summed == CHECKSUM_PARTIAL) {
1975                         switch (skb->protocol) {
1976                         case __constant_htons(ETH_P_IP):
1977                                 tu_cmd |= E1000_ADVTXD_TUCMD_IPV4;
1978                                 if (ip_hdr(skb)->protocol == IPPROTO_TCP)
1979                                         tu_cmd |= E1000_ADVTXD_TUCMD_L4T_TCP;
1980                                 break;
1981                         case __constant_htons(ETH_P_IPV6):
1982                                 if (ipv6_hdr(skb)->nexthdr == IPPROTO_TCP)
1983                                         tu_cmd |= E1000_ADVTXD_TUCMD_L4T_TCP;
1984                                 break;
1985                         default:
1986                                 break;
1987                         }
1988                 }
1989
1990                 context_desc->type_tucmd_mlhl = cpu_to_le32(tu_cmd);
1991                 context_desc->seqnum_seed = 0;
1992                 context_desc->mss_l4len_idx = 0;
1993
1994                 buffer_info->time_stamp = jiffies;
1995                 buffer_info->next_to_watch = i;
1996                 buffer_info->dma = 0;
1997                 i++;
1998                 if (i == tx_ring->count)
1999                         i = 0;
2000                 tx_ring->next_to_use = i;
2001
2002                 return true;
2003         }
2004
2005         return false;
2006 }
2007
2008 static int igbvf_maybe_stop_tx(struct net_device *netdev, int size)
2009 {
2010         struct igbvf_adapter *adapter = netdev_priv(netdev);
2011
2012         /* there is enough descriptors then we don't need to worry  */
2013         if (igbvf_desc_unused(adapter->tx_ring) >= size)
2014                 return 0;
2015
2016         netif_stop_queue(netdev);
2017
2018         smp_mb();
2019
2020         /* We need to check again just in case room has been made available */
2021         if (igbvf_desc_unused(adapter->tx_ring) < size)
2022                 return -EBUSY;
2023
2024         netif_wake_queue(netdev);
2025
2026         ++adapter->restart_queue;
2027         return 0;
2028 }
2029
2030 #define IGBVF_MAX_TXD_PWR       16
2031 #define IGBVF_MAX_DATA_PER_TXD  (1 << IGBVF_MAX_TXD_PWR)
2032
2033 static inline int igbvf_tx_map_adv(struct igbvf_adapter *adapter,
2034                                    struct igbvf_ring *tx_ring,
2035                                    struct sk_buff *skb,
2036                                    unsigned int first)
2037 {
2038         struct igbvf_buffer *buffer_info;
2039         struct pci_dev *pdev = adapter->pdev;
2040         unsigned int len = skb_headlen(skb);
2041         unsigned int count = 0, i;
2042         unsigned int f;
2043
2044         i = tx_ring->next_to_use;
2045
2046         buffer_info = &tx_ring->buffer_info[i];
2047         BUG_ON(len >= IGBVF_MAX_DATA_PER_TXD);
2048         buffer_info->length = len;
2049         /* set time_stamp *before* dma to help avoid a possible race */
2050         buffer_info->time_stamp = jiffies;
2051         buffer_info->next_to_watch = i;
2052         buffer_info->mapped_as_page = false;
2053         buffer_info->dma = dma_map_single(&pdev->dev, skb->data, len,
2054                                           DMA_TO_DEVICE);
2055         if (dma_mapping_error(&pdev->dev, buffer_info->dma))
2056                 goto dma_error;
2057
2058
2059         for (f = 0; f < skb_shinfo(skb)->nr_frags; f++) {
2060                 struct skb_frag_struct *frag;
2061
2062                 count++;
2063                 i++;
2064                 if (i == tx_ring->count)
2065                         i = 0;
2066
2067                 frag = &skb_shinfo(skb)->frags[f];
2068                 len = frag->size;
2069
2070                 buffer_info = &tx_ring->buffer_info[i];
2071                 BUG_ON(len >= IGBVF_MAX_DATA_PER_TXD);
2072                 buffer_info->length = len;
2073                 buffer_info->time_stamp = jiffies;
2074                 buffer_info->next_to_watch = i;
2075                 buffer_info->mapped_as_page = true;
2076                 buffer_info->dma = dma_map_page(&pdev->dev,
2077                                                 frag->page,
2078                                                 frag->page_offset,
2079                                                 len,
2080                                                 DMA_TO_DEVICE);
2081                 if (dma_mapping_error(&pdev->dev, buffer_info->dma))
2082                         goto dma_error;
2083         }
2084
2085         tx_ring->buffer_info[i].skb = skb;
2086         tx_ring->buffer_info[first].next_to_watch = i;
2087
2088         return ++count;
2089
2090 dma_error:
2091         dev_err(&pdev->dev, "TX DMA map failed\n");
2092
2093         /* clear timestamp and dma mappings for failed buffer_info mapping */
2094         buffer_info->dma = 0;
2095         buffer_info->time_stamp = 0;
2096         buffer_info->length = 0;
2097         buffer_info->next_to_watch = 0;
2098         buffer_info->mapped_as_page = false;
2099         if (count)
2100                 count--;
2101
2102         /* clear timestamp and dma mappings for remaining portion of packet */
2103         while (count--) {
2104                 if (i==0)
2105                         i += tx_ring->count;
2106                 i--;
2107                 buffer_info = &tx_ring->buffer_info[i];
2108                 igbvf_put_txbuf(adapter, buffer_info);
2109         }
2110
2111         return 0;
2112 }
2113
2114 static inline void igbvf_tx_queue_adv(struct igbvf_adapter *adapter,
2115                                       struct igbvf_ring *tx_ring,
2116                                       int tx_flags, int count, u32 paylen,
2117                                       u8 hdr_len)
2118 {
2119         union e1000_adv_tx_desc *tx_desc = NULL;
2120         struct igbvf_buffer *buffer_info;
2121         u32 olinfo_status = 0, cmd_type_len;
2122         unsigned int i;
2123
2124         cmd_type_len = (E1000_ADVTXD_DTYP_DATA | E1000_ADVTXD_DCMD_IFCS |
2125                         E1000_ADVTXD_DCMD_DEXT);
2126
2127         if (tx_flags & IGBVF_TX_FLAGS_VLAN)
2128                 cmd_type_len |= E1000_ADVTXD_DCMD_VLE;
2129
2130         if (tx_flags & IGBVF_TX_FLAGS_TSO) {
2131                 cmd_type_len |= E1000_ADVTXD_DCMD_TSE;
2132
2133                 /* insert tcp checksum */
2134                 olinfo_status |= E1000_TXD_POPTS_TXSM << 8;
2135
2136                 /* insert ip checksum */
2137                 if (tx_flags & IGBVF_TX_FLAGS_IPV4)
2138                         olinfo_status |= E1000_TXD_POPTS_IXSM << 8;
2139
2140         } else if (tx_flags & IGBVF_TX_FLAGS_CSUM) {
2141                 olinfo_status |= E1000_TXD_POPTS_TXSM << 8;
2142         }
2143
2144         olinfo_status |= ((paylen - hdr_len) << E1000_ADVTXD_PAYLEN_SHIFT);
2145
2146         i = tx_ring->next_to_use;
2147         while (count--) {
2148                 buffer_info = &tx_ring->buffer_info[i];
2149                 tx_desc = IGBVF_TX_DESC_ADV(*tx_ring, i);
2150                 tx_desc->read.buffer_addr = cpu_to_le64(buffer_info->dma);
2151                 tx_desc->read.cmd_type_len =
2152                          cpu_to_le32(cmd_type_len | buffer_info->length);
2153                 tx_desc->read.olinfo_status = cpu_to_le32(olinfo_status);
2154                 i++;
2155                 if (i == tx_ring->count)
2156                         i = 0;
2157         }
2158
2159         tx_desc->read.cmd_type_len |= cpu_to_le32(adapter->txd_cmd);
2160         /* Force memory writes to complete before letting h/w
2161          * know there are new descriptors to fetch.  (Only
2162          * applicable for weak-ordered memory model archs,
2163          * such as IA-64). */
2164         wmb();
2165
2166         tx_ring->next_to_use = i;
2167         writel(i, adapter->hw.hw_addr + tx_ring->tail);
2168         /* we need this if more than one processor can write to our tail
2169          * at a time, it syncronizes IO on IA64/Altix systems */
2170         mmiowb();
2171 }
2172
2173 static netdev_tx_t igbvf_xmit_frame_ring_adv(struct sk_buff *skb,
2174                                              struct net_device *netdev,
2175                                              struct igbvf_ring *tx_ring)
2176 {
2177         struct igbvf_adapter *adapter = netdev_priv(netdev);
2178         unsigned int first, tx_flags = 0;
2179         u8 hdr_len = 0;
2180         int count = 0;
2181         int tso = 0;
2182
2183         if (test_bit(__IGBVF_DOWN, &adapter->state)) {
2184                 dev_kfree_skb_any(skb);
2185                 return NETDEV_TX_OK;
2186         }
2187
2188         if (skb->len <= 0) {
2189                 dev_kfree_skb_any(skb);
2190                 return NETDEV_TX_OK;
2191         }
2192
2193         /*
2194          * need: count + 4 desc gap to keep tail from touching
2195          *       + 2 desc gap to keep tail from touching head,
2196          *       + 1 desc for skb->data,
2197          *       + 1 desc for context descriptor,
2198          * head, otherwise try next time
2199          */
2200         if (igbvf_maybe_stop_tx(netdev, skb_shinfo(skb)->nr_frags + 4)) {
2201                 /* this is a hard error */
2202                 return NETDEV_TX_BUSY;
2203         }
2204
2205         if (adapter->vlgrp && vlan_tx_tag_present(skb)) {
2206                 tx_flags |= IGBVF_TX_FLAGS_VLAN;
2207                 tx_flags |= (vlan_tx_tag_get(skb) << IGBVF_TX_FLAGS_VLAN_SHIFT);
2208         }
2209
2210         if (skb->protocol == htons(ETH_P_IP))
2211                 tx_flags |= IGBVF_TX_FLAGS_IPV4;
2212
2213         first = tx_ring->next_to_use;
2214
2215         tso = skb_is_gso(skb) ?
2216                 igbvf_tso(adapter, tx_ring, skb, tx_flags, &hdr_len) : 0;
2217         if (unlikely(tso < 0)) {
2218                 dev_kfree_skb_any(skb);
2219                 return NETDEV_TX_OK;
2220         }
2221
2222         if (tso)
2223                 tx_flags |= IGBVF_TX_FLAGS_TSO;
2224         else if (igbvf_tx_csum(adapter, tx_ring, skb, tx_flags) &&
2225                  (skb->ip_summed == CHECKSUM_PARTIAL))
2226                 tx_flags |= IGBVF_TX_FLAGS_CSUM;
2227
2228         /*
2229          * count reflects descriptors mapped, if 0 then mapping error
2230          * has occurred and we need to rewind the descriptor queue
2231          */
2232         count = igbvf_tx_map_adv(adapter, tx_ring, skb, first);
2233
2234         if (count) {
2235                 igbvf_tx_queue_adv(adapter, tx_ring, tx_flags, count,
2236                                    skb->len, hdr_len);
2237                 /* Make sure there is space in the ring for the next send. */
2238                 igbvf_maybe_stop_tx(netdev, MAX_SKB_FRAGS + 4);
2239         } else {
2240                 dev_kfree_skb_any(skb);
2241                 tx_ring->buffer_info[first].time_stamp = 0;
2242                 tx_ring->next_to_use = first;
2243         }
2244
2245         return NETDEV_TX_OK;
2246 }
2247
2248 static netdev_tx_t igbvf_xmit_frame(struct sk_buff *skb,
2249                                     struct net_device *netdev)
2250 {
2251         struct igbvf_adapter *adapter = netdev_priv(netdev);
2252         struct igbvf_ring *tx_ring;
2253
2254         if (test_bit(__IGBVF_DOWN, &adapter->state)) {
2255                 dev_kfree_skb_any(skb);
2256                 return NETDEV_TX_OK;
2257         }
2258
2259         tx_ring = &adapter->tx_ring[0];
2260
2261         return igbvf_xmit_frame_ring_adv(skb, netdev, tx_ring);
2262 }
2263
2264 /**
2265  * igbvf_tx_timeout - Respond to a Tx Hang
2266  * @netdev: network interface device structure
2267  **/
2268 static void igbvf_tx_timeout(struct net_device *netdev)
2269 {
2270         struct igbvf_adapter *adapter = netdev_priv(netdev);
2271
2272         /* Do the reset outside of interrupt context */
2273         adapter->tx_timeout_count++;
2274         schedule_work(&adapter->reset_task);
2275 }
2276
2277 static void igbvf_reset_task(struct work_struct *work)
2278 {
2279         struct igbvf_adapter *adapter;
2280         adapter = container_of(work, struct igbvf_adapter, reset_task);
2281
2282         igbvf_reinit_locked(adapter);
2283 }
2284
2285 /**
2286  * igbvf_get_stats - Get System Network Statistics
2287  * @netdev: network interface device structure
2288  *
2289  * Returns the address of the device statistics structure.
2290  * The statistics are actually updated from the timer callback.
2291  **/
2292 static struct net_device_stats *igbvf_get_stats(struct net_device *netdev)
2293 {
2294         struct igbvf_adapter *adapter = netdev_priv(netdev);
2295
2296         /* only return the current stats */
2297         return &adapter->net_stats;
2298 }
2299
2300 /**
2301  * igbvf_change_mtu - Change the Maximum Transfer Unit
2302  * @netdev: network interface device structure
2303  * @new_mtu: new value for maximum frame size
2304  *
2305  * Returns 0 on success, negative on failure
2306  **/
2307 static int igbvf_change_mtu(struct net_device *netdev, int new_mtu)
2308 {
2309         struct igbvf_adapter *adapter = netdev_priv(netdev);
2310         int max_frame = new_mtu + ETH_HLEN + ETH_FCS_LEN;
2311
2312         if ((new_mtu < 68) || (max_frame > MAX_JUMBO_FRAME_SIZE)) {
2313                 dev_err(&adapter->pdev->dev, "Invalid MTU setting\n");
2314                 return -EINVAL;
2315         }
2316
2317 #define MAX_STD_JUMBO_FRAME_SIZE 9234
2318         if (max_frame > MAX_STD_JUMBO_FRAME_SIZE) {
2319                 dev_err(&adapter->pdev->dev, "MTU > 9216 not supported.\n");
2320                 return -EINVAL;
2321         }
2322
2323         while (test_and_set_bit(__IGBVF_RESETTING, &adapter->state))
2324                 msleep(1);
2325         /* igbvf_down has a dependency on max_frame_size */
2326         adapter->max_frame_size = max_frame;
2327         if (netif_running(netdev))
2328                 igbvf_down(adapter);
2329
2330         /*
2331          * NOTE: netdev_alloc_skb reserves 16 bytes, and typically NET_IP_ALIGN
2332          * means we reserve 2 more, this pushes us to allocate from the next
2333          * larger slab size.
2334          * i.e. RXBUFFER_2048 --> size-4096 slab
2335          * However with the new *_jumbo_rx* routines, jumbo receives will use
2336          * fragmented skbs
2337          */
2338
2339         if (max_frame <= 1024)
2340                 adapter->rx_buffer_len = 1024;
2341         else if (max_frame <= 2048)
2342                 adapter->rx_buffer_len = 2048;
2343         else
2344 #if (PAGE_SIZE / 2) > 16384
2345                 adapter->rx_buffer_len = 16384;
2346 #else
2347                 adapter->rx_buffer_len = PAGE_SIZE / 2;
2348 #endif
2349
2350
2351         /* adjust allocation if LPE protects us, and we aren't using SBP */
2352         if ((max_frame == ETH_FRAME_LEN + ETH_FCS_LEN) ||
2353              (max_frame == ETH_FRAME_LEN + VLAN_HLEN + ETH_FCS_LEN))
2354                 adapter->rx_buffer_len = ETH_FRAME_LEN + VLAN_HLEN +
2355                                          ETH_FCS_LEN;
2356
2357         dev_info(&adapter->pdev->dev, "changing MTU from %d to %d\n",
2358                  netdev->mtu, new_mtu);
2359         netdev->mtu = new_mtu;
2360
2361         if (netif_running(netdev))
2362                 igbvf_up(adapter);
2363         else
2364                 igbvf_reset(adapter);
2365
2366         clear_bit(__IGBVF_RESETTING, &adapter->state);
2367
2368         return 0;
2369 }
2370
2371 static int igbvf_ioctl(struct net_device *netdev, struct ifreq *ifr, int cmd)
2372 {
2373         switch (cmd) {
2374         default:
2375                 return -EOPNOTSUPP;
2376         }
2377 }
2378
2379 static int igbvf_suspend(struct pci_dev *pdev, pm_message_t state)
2380 {
2381         struct net_device *netdev = pci_get_drvdata(pdev);
2382         struct igbvf_adapter *adapter = netdev_priv(netdev);
2383 #ifdef CONFIG_PM
2384         int retval = 0;
2385 #endif
2386
2387         netif_device_detach(netdev);
2388
2389         if (netif_running(netdev)) {
2390                 WARN_ON(test_bit(__IGBVF_RESETTING, &adapter->state));
2391                 igbvf_down(adapter);
2392                 igbvf_free_irq(adapter);
2393         }
2394
2395 #ifdef CONFIG_PM
2396         retval = pci_save_state(pdev);
2397         if (retval)
2398                 return retval;
2399 #endif
2400
2401         pci_disable_device(pdev);
2402
2403         return 0;
2404 }
2405
2406 #ifdef CONFIG_PM
2407 static int igbvf_resume(struct pci_dev *pdev)
2408 {
2409         struct net_device *netdev = pci_get_drvdata(pdev);
2410         struct igbvf_adapter *adapter = netdev_priv(netdev);
2411         u32 err;
2412
2413         pci_restore_state(pdev);
2414         err = pci_enable_device_mem(pdev);
2415         if (err) {
2416                 dev_err(&pdev->dev, "Cannot enable PCI device from suspend\n");
2417                 return err;
2418         }
2419
2420         pci_set_master(pdev);
2421
2422         if (netif_running(netdev)) {
2423                 err = igbvf_request_irq(adapter);
2424                 if (err)
2425                         return err;
2426         }
2427
2428         igbvf_reset(adapter);
2429
2430         if (netif_running(netdev))
2431                 igbvf_up(adapter);
2432
2433         netif_device_attach(netdev);
2434
2435         return 0;
2436 }
2437 #endif
2438
2439 static void igbvf_shutdown(struct pci_dev *pdev)
2440 {
2441         igbvf_suspend(pdev, PMSG_SUSPEND);
2442 }
2443
2444 #ifdef CONFIG_NET_POLL_CONTROLLER
2445 /*
2446  * Polling 'interrupt' - used by things like netconsole to send skbs
2447  * without having to re-enable interrupts. It's not called while
2448  * the interrupt routine is executing.
2449  */
2450 static void igbvf_netpoll(struct net_device *netdev)
2451 {
2452         struct igbvf_adapter *adapter = netdev_priv(netdev);
2453
2454         disable_irq(adapter->pdev->irq);
2455
2456         igbvf_clean_tx_irq(adapter->tx_ring);
2457
2458         enable_irq(adapter->pdev->irq);
2459 }
2460 #endif
2461
2462 /**
2463  * igbvf_io_error_detected - called when PCI error is detected
2464  * @pdev: Pointer to PCI device
2465  * @state: The current pci connection state
2466  *
2467  * This function is called after a PCI bus error affecting
2468  * this device has been detected.
2469  */
2470 static pci_ers_result_t igbvf_io_error_detected(struct pci_dev *pdev,
2471                                                 pci_channel_state_t state)
2472 {
2473         struct net_device *netdev = pci_get_drvdata(pdev);
2474         struct igbvf_adapter *adapter = netdev_priv(netdev);
2475
2476         netif_device_detach(netdev);
2477
2478         if (state == pci_channel_io_perm_failure)
2479                 return PCI_ERS_RESULT_DISCONNECT;
2480
2481         if (netif_running(netdev))
2482                 igbvf_down(adapter);
2483         pci_disable_device(pdev);
2484
2485         /* Request a slot slot reset. */
2486         return PCI_ERS_RESULT_NEED_RESET;
2487 }
2488
2489 /**
2490  * igbvf_io_slot_reset - called after the pci bus has been reset.
2491  * @pdev: Pointer to PCI device
2492  *
2493  * Restart the card from scratch, as if from a cold-boot. Implementation
2494  * resembles the first-half of the igbvf_resume routine.
2495  */
2496 static pci_ers_result_t igbvf_io_slot_reset(struct pci_dev *pdev)
2497 {
2498         struct net_device *netdev = pci_get_drvdata(pdev);
2499         struct igbvf_adapter *adapter = netdev_priv(netdev);
2500
2501         if (pci_enable_device_mem(pdev)) {
2502                 dev_err(&pdev->dev,
2503                         "Cannot re-enable PCI device after reset.\n");
2504                 return PCI_ERS_RESULT_DISCONNECT;
2505         }
2506         pci_set_master(pdev);
2507
2508         igbvf_reset(adapter);
2509
2510         return PCI_ERS_RESULT_RECOVERED;
2511 }
2512
2513 /**
2514  * igbvf_io_resume - called when traffic can start flowing again.
2515  * @pdev: Pointer to PCI device
2516  *
2517  * This callback is called when the error recovery driver tells us that
2518  * its OK to resume normal operation. Implementation resembles the
2519  * second-half of the igbvf_resume routine.
2520  */
2521 static void igbvf_io_resume(struct pci_dev *pdev)
2522 {
2523         struct net_device *netdev = pci_get_drvdata(pdev);
2524         struct igbvf_adapter *adapter = netdev_priv(netdev);
2525
2526         if (netif_running(netdev)) {
2527                 if (igbvf_up(adapter)) {
2528                         dev_err(&pdev->dev,
2529                                 "can't bring device back up after reset\n");
2530                         return;
2531                 }
2532         }
2533
2534         netif_device_attach(netdev);
2535 }
2536
2537 static void igbvf_print_device_info(struct igbvf_adapter *adapter)
2538 {
2539         struct e1000_hw *hw = &adapter->hw;
2540         struct net_device *netdev = adapter->netdev;
2541         struct pci_dev *pdev = adapter->pdev;
2542
2543         dev_info(&pdev->dev, "Intel(R) 82576 Virtual Function\n");
2544         dev_info(&pdev->dev, "Address: %pM\n", netdev->dev_addr);
2545         dev_info(&pdev->dev, "MAC: %d\n", hw->mac.type);
2546 }
2547
2548 static const struct net_device_ops igbvf_netdev_ops = {
2549         .ndo_open                       = igbvf_open,
2550         .ndo_stop                       = igbvf_close,
2551         .ndo_start_xmit                 = igbvf_xmit_frame,
2552         .ndo_get_stats                  = igbvf_get_stats,
2553         .ndo_set_multicast_list         = igbvf_set_multi,
2554         .ndo_set_mac_address            = igbvf_set_mac,
2555         .ndo_change_mtu                 = igbvf_change_mtu,
2556         .ndo_do_ioctl                   = igbvf_ioctl,
2557         .ndo_tx_timeout                 = igbvf_tx_timeout,
2558         .ndo_vlan_rx_register           = igbvf_vlan_rx_register,
2559         .ndo_vlan_rx_add_vid            = igbvf_vlan_rx_add_vid,
2560         .ndo_vlan_rx_kill_vid           = igbvf_vlan_rx_kill_vid,
2561 #ifdef CONFIG_NET_POLL_CONTROLLER
2562         .ndo_poll_controller            = igbvf_netpoll,
2563 #endif
2564 };
2565
2566 /**
2567  * igbvf_probe - Device Initialization Routine
2568  * @pdev: PCI device information struct
2569  * @ent: entry in igbvf_pci_tbl
2570  *
2571  * Returns 0 on success, negative on failure
2572  *
2573  * igbvf_probe initializes an adapter identified by a pci_dev structure.
2574  * The OS initialization, configuring of the adapter private structure,
2575  * and a hardware reset occur.
2576  **/
2577 static int __devinit igbvf_probe(struct pci_dev *pdev,
2578                                  const struct pci_device_id *ent)
2579 {
2580         struct net_device *netdev;
2581         struct igbvf_adapter *adapter;
2582         struct e1000_hw *hw;
2583         const struct igbvf_info *ei = igbvf_info_tbl[ent->driver_data];
2584
2585         static int cards_found;
2586         int err, pci_using_dac;
2587
2588         err = pci_enable_device_mem(pdev);
2589         if (err)
2590                 return err;
2591
2592         pci_using_dac = 0;
2593         err = dma_set_mask(&pdev->dev, DMA_BIT_MASK(64));
2594         if (!err) {
2595                 err = dma_set_coherent_mask(&pdev->dev, DMA_BIT_MASK(64));
2596                 if (!err)
2597                         pci_using_dac = 1;
2598         } else {
2599                 err = dma_set_mask(&pdev->dev, DMA_BIT_MASK(32));
2600                 if (err) {
2601                         err = dma_set_coherent_mask(&pdev->dev,
2602                                                     DMA_BIT_MASK(32));
2603                         if (err) {
2604                                 dev_err(&pdev->dev, "No usable DMA "
2605                                         "configuration, aborting\n");
2606                                 goto err_dma;
2607                         }
2608                 }
2609         }
2610
2611         err = pci_request_regions(pdev, igbvf_driver_name);
2612         if (err)
2613                 goto err_pci_reg;
2614
2615         pci_set_master(pdev);
2616
2617         err = -ENOMEM;
2618         netdev = alloc_etherdev(sizeof(struct igbvf_adapter));
2619         if (!netdev)
2620                 goto err_alloc_etherdev;
2621
2622         SET_NETDEV_DEV(netdev, &pdev->dev);
2623
2624         pci_set_drvdata(pdev, netdev);
2625         adapter = netdev_priv(netdev);
2626         hw = &adapter->hw;
2627         adapter->netdev = netdev;
2628         adapter->pdev = pdev;
2629         adapter->ei = ei;
2630         adapter->pba = ei->pba;
2631         adapter->flags = ei->flags;
2632         adapter->hw.back = adapter;
2633         adapter->hw.mac.type = ei->mac;
2634         adapter->msg_enable = (1 << NETIF_MSG_DRV | NETIF_MSG_PROBE) - 1;
2635
2636         /* PCI config space info */
2637
2638         hw->vendor_id = pdev->vendor;
2639         hw->device_id = pdev->device;
2640         hw->subsystem_vendor_id = pdev->subsystem_vendor;
2641         hw->subsystem_device_id = pdev->subsystem_device;
2642         hw->revision_id = pdev->revision;
2643
2644         err = -EIO;
2645         adapter->hw.hw_addr = ioremap(pci_resource_start(pdev, 0),
2646                                       pci_resource_len(pdev, 0));
2647
2648         if (!adapter->hw.hw_addr)
2649                 goto err_ioremap;
2650
2651         if (ei->get_variants) {
2652                 err = ei->get_variants(adapter);
2653                 if (err)
2654                         goto err_ioremap;
2655         }
2656
2657         /* setup adapter struct */
2658         err = igbvf_sw_init(adapter);
2659         if (err)
2660                 goto err_sw_init;
2661
2662         /* construct the net_device struct */
2663         netdev->netdev_ops = &igbvf_netdev_ops;
2664
2665         igbvf_set_ethtool_ops(netdev);
2666         netdev->watchdog_timeo = 5 * HZ;
2667         strncpy(netdev->name, pci_name(pdev), sizeof(netdev->name) - 1);
2668
2669         adapter->bd_number = cards_found++;
2670
2671         netdev->features = NETIF_F_SG |
2672                            NETIF_F_IP_CSUM |
2673                            NETIF_F_HW_VLAN_TX |
2674                            NETIF_F_HW_VLAN_RX |
2675                            NETIF_F_HW_VLAN_FILTER;
2676
2677         netdev->features |= NETIF_F_IPV6_CSUM;
2678         netdev->features |= NETIF_F_TSO;
2679         netdev->features |= NETIF_F_TSO6;
2680
2681         if (pci_using_dac)
2682                 netdev->features |= NETIF_F_HIGHDMA;
2683
2684         netdev->vlan_features |= NETIF_F_TSO;
2685         netdev->vlan_features |= NETIF_F_TSO6;
2686         netdev->vlan_features |= NETIF_F_IP_CSUM;
2687         netdev->vlan_features |= NETIF_F_IPV6_CSUM;
2688         netdev->vlan_features |= NETIF_F_SG;
2689
2690         /*reset the controller to put the device in a known good state */
2691         err = hw->mac.ops.reset_hw(hw);
2692         if (err) {
2693                 dev_info(&pdev->dev,
2694                          "PF still in reset state, assigning new address."
2695                          " Is the PF interface up?\n");
2696                 dev_hw_addr_random(adapter->netdev, hw->mac.addr);
2697         } else {
2698                 err = hw->mac.ops.read_mac_addr(hw);
2699                 if (err) {
2700                         dev_err(&pdev->dev, "Error reading MAC address\n");
2701                         goto err_hw_init;
2702                 }
2703         }
2704
2705         memcpy(netdev->dev_addr, adapter->hw.mac.addr, netdev->addr_len);
2706         memcpy(netdev->perm_addr, adapter->hw.mac.addr, netdev->addr_len);
2707
2708         if (!is_valid_ether_addr(netdev->perm_addr)) {
2709                 dev_err(&pdev->dev, "Invalid MAC Address: %pM\n",
2710                         netdev->dev_addr);
2711                 err = -EIO;
2712                 goto err_hw_init;
2713         }
2714
2715         setup_timer(&adapter->watchdog_timer, &igbvf_watchdog,
2716                     (unsigned long) adapter);
2717
2718         INIT_WORK(&adapter->reset_task, igbvf_reset_task);
2719         INIT_WORK(&adapter->watchdog_task, igbvf_watchdog_task);
2720
2721         /* ring size defaults */
2722         adapter->rx_ring->count = 1024;
2723         adapter->tx_ring->count = 1024;
2724
2725         /* reset the hardware with the new settings */
2726         igbvf_reset(adapter);
2727
2728         strcpy(netdev->name, "eth%d");
2729         err = register_netdev(netdev);
2730         if (err)
2731                 goto err_hw_init;
2732
2733         /* tell the stack to leave us alone until igbvf_open() is called */
2734         netif_carrier_off(netdev);
2735         netif_stop_queue(netdev);
2736
2737         igbvf_print_device_info(adapter);
2738
2739         igbvf_initialize_last_counter_stats(adapter);
2740
2741         return 0;
2742
2743 err_hw_init:
2744         kfree(adapter->tx_ring);
2745         kfree(adapter->rx_ring);
2746 err_sw_init:
2747         igbvf_reset_interrupt_capability(adapter);
2748         iounmap(adapter->hw.hw_addr);
2749 err_ioremap:
2750         free_netdev(netdev);
2751 err_alloc_etherdev:
2752         pci_release_regions(pdev);
2753 err_pci_reg:
2754 err_dma:
2755         pci_disable_device(pdev);
2756         return err;
2757 }
2758
2759 /**
2760  * igbvf_remove - Device Removal Routine
2761  * @pdev: PCI device information struct
2762  *
2763  * igbvf_remove is called by the PCI subsystem to alert the driver
2764  * that it should release a PCI device.  The could be caused by a
2765  * Hot-Plug event, or because the driver is going to be removed from
2766  * memory.
2767  **/
2768 static void __devexit igbvf_remove(struct pci_dev *pdev)
2769 {
2770         struct net_device *netdev = pci_get_drvdata(pdev);
2771         struct igbvf_adapter *adapter = netdev_priv(netdev);
2772         struct e1000_hw *hw = &adapter->hw;
2773
2774         /*
2775          * The watchdog timer may be rescheduled, so explicitly
2776          * disable it from being rescheduled.
2777          */
2778         set_bit(__IGBVF_DOWN, &adapter->state);
2779         del_timer_sync(&adapter->watchdog_timer);
2780
2781         cancel_work_sync(&adapter->reset_task);
2782         cancel_work_sync(&adapter->watchdog_task);
2783
2784         unregister_netdev(netdev);
2785
2786         igbvf_reset_interrupt_capability(adapter);
2787
2788         /*
2789          * it is important to delete the napi struct prior to freeing the
2790          * rx ring so that you do not end up with null pointer refs
2791          */
2792         netif_napi_del(&adapter->rx_ring->napi);
2793         kfree(adapter->tx_ring);
2794         kfree(adapter->rx_ring);
2795
2796         iounmap(hw->hw_addr);
2797         if (hw->flash_address)
2798                 iounmap(hw->flash_address);
2799         pci_release_regions(pdev);
2800
2801         free_netdev(netdev);
2802
2803         pci_disable_device(pdev);
2804 }
2805
2806 /* PCI Error Recovery (ERS) */
2807 static struct pci_error_handlers igbvf_err_handler = {
2808         .error_detected = igbvf_io_error_detected,
2809         .slot_reset = igbvf_io_slot_reset,
2810         .resume = igbvf_io_resume,
2811 };
2812
2813 static DEFINE_PCI_DEVICE_TABLE(igbvf_pci_tbl) = {
2814         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82576_VF), board_vf },
2815         { PCI_VDEVICE(INTEL, E1000_DEV_ID_I350_VF), board_i350_vf },
2816         { } /* terminate list */
2817 };
2818 MODULE_DEVICE_TABLE(pci, igbvf_pci_tbl);
2819
2820 /* PCI Device API Driver */
2821 static struct pci_driver igbvf_driver = {
2822         .name     = igbvf_driver_name,
2823         .id_table = igbvf_pci_tbl,
2824         .probe    = igbvf_probe,
2825         .remove   = __devexit_p(igbvf_remove),
2826 #ifdef CONFIG_PM
2827         /* Power Management Hooks */
2828         .suspend  = igbvf_suspend,
2829         .resume   = igbvf_resume,
2830 #endif
2831         .shutdown = igbvf_shutdown,
2832         .err_handler = &igbvf_err_handler
2833 };
2834
2835 /**
2836  * igbvf_init_module - Driver Registration Routine
2837  *
2838  * igbvf_init_module is the first routine called when the driver is
2839  * loaded. All it does is register with the PCI subsystem.
2840  **/
2841 static int __init igbvf_init_module(void)
2842 {
2843         int ret;
2844         printk(KERN_INFO "%s - version %s\n",
2845                igbvf_driver_string, igbvf_driver_version);
2846         printk(KERN_INFO "%s\n", igbvf_copyright);
2847
2848         ret = pci_register_driver(&igbvf_driver);
2849
2850         return ret;
2851 }
2852 module_init(igbvf_init_module);
2853
2854 /**
2855  * igbvf_exit_module - Driver Exit Cleanup Routine
2856  *
2857  * igbvf_exit_module is called just before the driver is removed
2858  * from memory.
2859  **/
2860 static void __exit igbvf_exit_module(void)
2861 {
2862         pci_unregister_driver(&igbvf_driver);
2863 }
2864 module_exit(igbvf_exit_module);
2865
2866
2867 MODULE_AUTHOR("Intel Corporation, <e1000-devel@lists.sourceforge.net>");
2868 MODULE_DESCRIPTION("Intel(R) 82576 Virtual Function Network Driver");
2869 MODULE_LICENSE("GPL");
2870 MODULE_VERSION(DRV_VERSION);
2871
2872 /* netdev.c */