drivers/net/igbvf/netdev.c

   1 /*******************************************************************************
   2
   3   Intel(R) 82576 Virtual Function Linux driver
   4   Copyright(c) 2009 Intel Corporation.
   5
   6   This program is free software; you can redistribute it and/or modify it
   7   under the terms and conditions of the GNU General Public License,
   8   version 2, as published by the Free Software Foundation.
   9
  10   This program is distributed in the hope it will be useful, but WITHOUT
  11   ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
  12   FITNESS FOR A PARTICULAR PURPOSE.  See the GNU General Public License for
  13   more details.
  14
  15   You should have received a copy of the GNU General Public License along with
  16   this program; if not, write to the Free Software Foundation, Inc.,
  17   51 Franklin St - Fifth Floor, Boston, MA 02110-1301 USA.
  18
  19   The full GNU General Public License is included in this distribution in
  20   the file called "COPYING".
  21
  22   Contact Information:
  23   e1000-devel Mailing List <e1000-devel@lists.sourceforge.net>
  24   Intel Corporation, 5200 N.E. Elam Young Parkway, Hillsboro, OR 97124-6497
  25
  26 *******************************************************************************/
  27
  28 #include <linux/module.h>
  29 #include <linux/types.h>
  30 #include <linux/init.h>
  31 #include <linux/pci.h>
  32 #include <linux/vmalloc.h>
  33 #include <linux/pagemap.h>
  34 #include <linux/delay.h>
  35 #include <linux/netdevice.h>
  36 #include <linux/tcp.h>
  37 #include <linux/ipv6.h>
  38 #include <linux/slab.h>
  39 #include <net/checksum.h>
  40 #include <net/ip6_checksum.h>
  41 #include <linux/mii.h>
  42 #include <linux/ethtool.h>
  43 #include <linux/if_vlan.h>
  44 #include <linux/pm_qos_params.h>
  45
  46 #include "igbvf.h"
  47
  48 #define DRV_VERSION "1.0.0-k0"
  49 char igbvf_driver_name[] = "igbvf";
  50 const char igbvf_driver_version[] = DRV_VERSION;
  51 struct pm_qos_request_list *igbvf_driver_pm_qos_req;
  52 static const char igbvf_driver_string[] =
  53                                 "Intel(R) Virtual Function Network Driver";
  54 static const char igbvf_copyright[] = "Copyright (c) 2009 Intel Corporation.";
  55
  56 static int igbvf_poll(struct napi_struct *napi, int budget);
  57 static void igbvf_reset(struct igbvf_adapter *);
  58 static void igbvf_set_interrupt_capability(struct igbvf_adapter *);
  59 static void igbvf_reset_interrupt_capability(struct igbvf_adapter *);
  60
  61 static struct igbvf_info igbvf_vf_info = {
  62         .mac                    = e1000_vfadapt,
  63         .flags                  = 0,
  64         .pba                    = 10,
  65         .init_ops               = e1000_init_function_pointers_vf,
  66 };
  67
  68 static const struct igbvf_info *igbvf_info_tbl[] = {
  69         [board_vf]              = &igbvf_vf_info,
  70 };
  71
  72 /**
  73  * igbvf_desc_unused - calculate if we have unused descriptors
  74  **/
  75 static int igbvf_desc_unused(struct igbvf_ring *ring)
  76 {
  77         if (ring->next_to_clean > ring->next_to_use)
  78                 return ring->next_to_clean - ring->next_to_use - 1;
  79
  80         return ring->count + ring->next_to_clean - ring->next_to_use - 1;
  81 }
  82
  83 /**
  84  * igbvf_receive_skb - helper function to handle Rx indications
  85  * @adapter: board private structure
  86  * @status: descriptor status field as written by hardware
  87  * @vlan: descriptor vlan field as written by hardware (no le/be conversion)
  88  * @skb: pointer to sk_buff to be indicated to stack
  89  **/
  90 static void igbvf_receive_skb(struct igbvf_adapter *adapter,
  91                               struct net_device *netdev,
  92                               struct sk_buff *skb,
  93                               u32 status, u16 vlan)
  94 {
  95         if (adapter->vlgrp && (status & E1000_RXD_STAT_VP))
  96                 vlan_hwaccel_receive_skb(skb, adapter->vlgrp,
  97                                          le16_to_cpu(vlan) &
  98                                          E1000_RXD_SPC_VLAN_MASK);
  99         else
 100                 netif_receive_skb(skb);
 101 }
 102
 103 static inline void igbvf_rx_checksum_adv(struct igbvf_adapter *adapter,
 104                                          u32 status_err, struct sk_buff *skb)
 105 {
 106         skb->ip_summed = CHECKSUM_NONE;
 107
 108         /* Ignore Checksum bit is set or checksum is disabled through ethtool */
 109         if ((status_err & E1000_RXD_STAT_IXSM) ||
 110             (adapter->flags & IGBVF_FLAG_RX_CSUM_DISABLED))
 111                 return;
 112
 113         /* TCP/UDP checksum error bit is set */
 114         if (status_err &
 115             (E1000_RXDEXT_STATERR_TCPE | E1000_RXDEXT_STATERR_IPE)) {
 116                 /* let the stack verify checksum errors */
 117                 adapter->hw_csum_err++;
 118                 return;
 119         }
 120
 121         /* It must be a TCP or UDP packet with a valid checksum */
 122         if (status_err & (E1000_RXD_STAT_TCPCS | E1000_RXD_STAT_UDPCS))
 123                 skb->ip_summed = CHECKSUM_UNNECESSARY;
 124
 125         adapter->hw_csum_good++;
 126 }
 127
 128 /**
 129  * igbvf_alloc_rx_buffers - Replace used receive buffers; packet split
 130  * @rx_ring: address of ring structure to repopulate
 131  * @cleaned_count: number of buffers to repopulate
 132  **/
 133 static void igbvf_alloc_rx_buffers(struct igbvf_ring *rx_ring,
 134                                    int cleaned_count)
 135 {
 136         struct igbvf_adapter *adapter = rx_ring->adapter;
 137         struct net_device *netdev = adapter->netdev;
 138         struct pci_dev *pdev = adapter->pdev;
 139         union e1000_adv_rx_desc *rx_desc;
 140         struct igbvf_buffer *buffer_info;
 141         struct sk_buff *skb;
 142         unsigned int i;
 143         int bufsz;
 144
 145         i = rx_ring->next_to_use;
 146         buffer_info = &rx_ring->buffer_info[i];
 147
 148         if (adapter->rx_ps_hdr_size)
 149                 bufsz = adapter->rx_ps_hdr_size;
 150         else
 151                 bufsz = adapter->rx_buffer_len;
 152
 153         while (cleaned_count--) {
 154                 rx_desc = IGBVF_RX_DESC_ADV(*rx_ring, i);
 155
 156                 if (adapter->rx_ps_hdr_size && !buffer_info->page_dma) {
 157                         if (!buffer_info->page) {
 158                                 buffer_info->page = alloc_page(GFP_ATOMIC);
 159                                 if (!buffer_info->page) {
 160                                         adapter->alloc_rx_buff_failed++;
 161                                         goto no_buffers;
 162                                 }
 163                                 buffer_info->page_offset = 0;
 164                         } else {
 165                                 buffer_info->page_offset ^= PAGE_SIZE / 2;
 166                         }
 167                         buffer_info->page_dma =
 168                                 dma_map_page(&pdev->dev, buffer_info->page,
 169                                              buffer_info->page_offset,
 170                                              PAGE_SIZE / 2,
 171                                              DMA_FROM_DEVICE);
 172                 }
 173
 174                 if (!buffer_info->skb) {
 175                         skb = netdev_alloc_skb_ip_align(netdev, bufsz);
 176                         if (!skb) {
 177                                 adapter->alloc_rx_buff_failed++;
 178                                 goto no_buffers;
 179                         }
 180
 181                         buffer_info->skb = skb;
 182                         buffer_info->dma = dma_map_single(&pdev->dev, skb->data,
 183                                                           bufsz,
 184                                                           DMA_FROM_DEVICE);
 185                 }
 186                 /* Refresh the desc even if buffer_addrs didn't change because
 187                  * each write-back erases this info. */
 188                 if (adapter->rx_ps_hdr_size) {
 189                         rx_desc->read.pkt_addr =
 190                              cpu_to_le64(buffer_info->page_dma);
 191                         rx_desc->read.hdr_addr = cpu_to_le64(buffer_info->dma);
 192                 } else {
 193                         rx_desc->read.pkt_addr =
 194                              cpu_to_le64(buffer_info->dma);
 195                         rx_desc->read.hdr_addr = 0;
 196                 }
 197
 198                 i++;
 199                 if (i == rx_ring->count)
 200                         i = 0;
 201                 buffer_info = &rx_ring->buffer_info[i];
 202         }
 203
 204 no_buffers:
 205         if (rx_ring->next_to_use != i) {
 206                 rx_ring->next_to_use = i;
 207                 if (i == 0)
 208                         i = (rx_ring->count - 1);
 209                 else
 210                         i--;
 211
 212                 /* Force memory writes to complete before letting h/w
 213                  * know there are new descriptors to fetch.  (Only
 214                  * applicable for weak-ordered memory model archs,
 215                  * such as IA-64). */
 216                 wmb();
 217                 writel(i, adapter->hw.hw_addr + rx_ring->tail);
 218         }
 219 }
 220
 221 /**
 222  * igbvf_clean_rx_irq - Send received data up the network stack; legacy
 223  * @adapter: board private structure
 224  *
 225  * the return value indicates whether actual cleaning was done, there
 226  * is no guarantee that everything was cleaned
 227  **/
 228 static bool igbvf_clean_rx_irq(struct igbvf_adapter *adapter,
 229                                int *work_done, int work_to_do)
 230 {
 231         struct igbvf_ring *rx_ring = adapter->rx_ring;
 232         struct net_device *netdev = adapter->netdev;
 233         struct pci_dev *pdev = adapter->pdev;
 234         union e1000_adv_rx_desc *rx_desc, *next_rxd;
 235         struct igbvf_buffer *buffer_info, *next_buffer;
 236         struct sk_buff *skb;
 237         bool cleaned = false;
 238         int cleaned_count = 0;
 239         unsigned int total_bytes = 0, total_packets = 0;
 240         unsigned int i;
 241         u32 length, hlen, staterr;
 242
 243         i = rx_ring->next_to_clean;
 244         rx_desc = IGBVF_RX_DESC_ADV(*rx_ring, i);
 245         staterr = le32_to_cpu(rx_desc->wb.upper.status_error);
 246
 247         while (staterr & E1000_RXD_STAT_DD) {
 248                 if (*work_done >= work_to_do)
 249                         break;
 250                 (*work_done)++;
 251
 252                 buffer_info = &rx_ring->buffer_info[i];
 253
 254                 /* HW will not DMA in data larger than the given buffer, even
 255                  * if it parses the (NFS, of course) header to be larger.  In
 256                  * that case, it fills the header buffer and spills the rest
 257                  * into the page.
 258                  */
 259                 hlen = (le16_to_cpu(rx_desc->wb.lower.lo_dword.hs_rss.hdr_info) &
 260                   E1000_RXDADV_HDRBUFLEN_MASK) >> E1000_RXDADV_HDRBUFLEN_SHIFT;
 261                 if (hlen > adapter->rx_ps_hdr_size)
 262                         hlen = adapter->rx_ps_hdr_size;
 263
 264                 length = le16_to_cpu(rx_desc->wb.upper.length);
 265                 cleaned = true;
 266                 cleaned_count++;
 267
 268                 skb = buffer_info->skb;
 269                 prefetch(skb->data - NET_IP_ALIGN);
 270                 buffer_info->skb = NULL;
 271                 if (!adapter->rx_ps_hdr_size) {
 272                         dma_unmap_single(&pdev->dev, buffer_info->dma,
 273                                          adapter->rx_buffer_len,
 274                                          DMA_FROM_DEVICE);
 275                         buffer_info->dma = 0;
 276                         skb_put(skb, length);
 277                         goto send_up;
 278                 }
 279
 280                 if (!skb_shinfo(skb)->nr_frags) {
 281                         dma_unmap_single(&pdev->dev, buffer_info->dma,
 282                                          adapter->rx_ps_hdr_size,
 283                                          DMA_FROM_DEVICE);
 284                         skb_put(skb, hlen);
 285                 }
 286
 287                 if (length) {
 288                         dma_unmap_page(&pdev->dev, buffer_info->page_dma,
 289                                        PAGE_SIZE / 2,
 290                                        DMA_FROM_DEVICE);
 291                         buffer_info->page_dma = 0;
 292
 293                         skb_fill_page_desc(skb, skb_shinfo(skb)->nr_frags,
 294                                            buffer_info->page,
 295                                            buffer_info->page_offset,
 296                                            length);
 297
 298                         if ((adapter->rx_buffer_len > (PAGE_SIZE / 2)) ||
 299                             (page_count(buffer_info->page) != 1))
 300                                 buffer_info->page = NULL;
 301                         else
 302                                 get_page(buffer_info->page);
 303
 304                         skb->len += length;
 305                         skb->data_len += length;
 306                         skb->truesize += length;
 307                 }
 308 send_up:
 309                 i++;
 310                 if (i == rx_ring->count)
 311                         i = 0;
 312                 next_rxd = IGBVF_RX_DESC_ADV(*rx_ring, i);
 313                 prefetch(next_rxd);
 314                 next_buffer = &rx_ring->buffer_info[i];
 315
 316                 if (!(staterr & E1000_RXD_STAT_EOP)) {
 317                         buffer_info->skb = next_buffer->skb;
 318                         buffer_info->dma = next_buffer->dma;
 319                         next_buffer->skb = skb;
 320                         next_buffer->dma = 0;
 321                         goto next_desc;
 322                 }
 323
 324                 if (staterr & E1000_RXDEXT_ERR_FRAME_ERR_MASK) {
 325                         dev_kfree_skb_irq(skb);
 326                         goto next_desc;
 327                 }
 328
 329                 total_bytes += skb->len;
 330                 total_packets++;
 331
 332                 igbvf_rx_checksum_adv(adapter, staterr, skb);
 333
 334                 skb->protocol = eth_type_trans(skb, netdev);
 335
 336                 igbvf_receive_skb(adapter, netdev, skb, staterr,
 337                                   rx_desc->wb.upper.vlan);
 338
 339 next_desc:
 340                 rx_desc->wb.upper.status_error = 0;
 341
 342                 /* return some buffers to hardware, one at a time is too slow */
 343                 if (cleaned_count >= IGBVF_RX_BUFFER_WRITE) {
 344                         igbvf_alloc_rx_buffers(rx_ring, cleaned_count);
 345                         cleaned_count = 0;
 346                 }
 347
 348                 /* use prefetched values */
 349                 rx_desc = next_rxd;
 350                 buffer_info = next_buffer;
 351
 352                 staterr = le32_to_cpu(rx_desc->wb.upper.status_error);
 353         }
 354
 355         rx_ring->next_to_clean = i;
 356         cleaned_count = igbvf_desc_unused(rx_ring);
 357
 358         if (cleaned_count)
 359                 igbvf_alloc_rx_buffers(rx_ring, cleaned_count);
 360
 361         adapter->total_rx_packets += total_packets;
 362         adapter->total_rx_bytes += total_bytes;
 363         adapter->net_stats.rx_bytes += total_bytes;
 364         adapter->net_stats.rx_packets += total_packets;
 365         return cleaned;
 366 }
 367
 368 static void igbvf_put_txbuf(struct igbvf_adapter *adapter,
 369                             struct igbvf_buffer *buffer_info)
 370 {
 371         if (buffer_info->dma) {
 372                 if (buffer_info->mapped_as_page)
 373                         dma_unmap_page(&adapter->pdev->dev,
 374                                        buffer_info->dma,
 375                                        buffer_info->length,
 376                                        DMA_TO_DEVICE);
 377                 else
 378                         dma_unmap_single(&adapter->pdev->dev,
 379                                          buffer_info->dma,
 380                                          buffer_info->length,
 381                                          DMA_TO_DEVICE);
 382                 buffer_info->dma = 0;
 383         }
 384         if (buffer_info->skb) {
 385                 dev_kfree_skb_any(buffer_info->skb);
 386                 buffer_info->skb = NULL;
 387         }
 388         buffer_info->time_stamp = 0;
 389 }
 390
 391 static void igbvf_print_tx_hang(struct igbvf_adapter *adapter)
 392 {
 393         struct igbvf_ring *tx_ring = adapter->tx_ring;
 394         unsigned int i = tx_ring->next_to_clean;
 395         unsigned int eop = tx_ring->buffer_info[i].next_to_watch;
 396         union e1000_adv_tx_desc *eop_desc = IGBVF_TX_DESC_ADV(*tx_ring, eop);
 397
 398         /* detected Tx unit hang */
 399         dev_err(&adapter->pdev->dev,
 400                 "Detected Tx Unit Hang:\n"
 401                 "  TDH                  <%x>\n"
 402                 "  TDT                  <%x>\n"
 403                 "  next_to_use          <%x>\n"
 404                 "  next_to_clean        <%x>\n"
 405                 "buffer_info[next_to_clean]:\n"
 406                 "  time_stamp           <%lx>\n"
 407                 "  next_to_watch        <%x>\n"
 408                 "  jiffies              <%lx>\n"
 409                 "  next_to_watch.status <%x>\n",
 410                 readl(adapter->hw.hw_addr + tx_ring->head),
 411                 readl(adapter->hw.hw_addr + tx_ring->tail),
 412                 tx_ring->next_to_use,
 413                 tx_ring->next_to_clean,
 414                 tx_ring->buffer_info[eop].time_stamp,
 415                 eop,
 416                 jiffies,
 417                 eop_desc->wb.status);
 418 }
 419
 420 /**
 421  * igbvf_setup_tx_resources - allocate Tx resources (Descriptors)
 422  * @adapter: board private structure
 423  *
 424  * Return 0 on success, negative on failure
 425  **/
 426 int igbvf_setup_tx_resources(struct igbvf_adapter *adapter,
 427                              struct igbvf_ring *tx_ring)
 428 {
 429         struct pci_dev *pdev = adapter->pdev;
 430         int size;
 431
 432         size = sizeof(struct igbvf_buffer) * tx_ring->count;
 433         tx_ring->buffer_info = vmalloc(size);
 434         if (!tx_ring->buffer_info)
 435                 goto err;
 436         memset(tx_ring->buffer_info, 0, size);
 437
 438         /* round up to nearest 4K */
 439         tx_ring->size = tx_ring->count * sizeof(union e1000_adv_tx_desc);
 440         tx_ring->size = ALIGN(tx_ring->size, 4096);
 441
 442         tx_ring->desc = dma_alloc_coherent(&pdev->dev, tx_ring->size,
 443                                            &tx_ring->dma, GFP_KERNEL);
 444
 445         if (!tx_ring->desc)
 446                 goto err;
 447
 448         tx_ring->adapter = adapter;
 449         tx_ring->next_to_use = 0;
 450         tx_ring->next_to_clean = 0;
 451
 452         return 0;
 453 err:
 454         vfree(tx_ring->buffer_info);
 455         dev_err(&adapter->pdev->dev,
 456                 "Unable to allocate memory for the transmit descriptor ring\n");
 457         return -ENOMEM;
 458 }
 459
 460 /**
 461  * igbvf_setup_rx_resources - allocate Rx resources (Descriptors)
 462  * @adapter: board private structure
 463  *
 464  * Returns 0 on success, negative on failure
 465  **/
 466 int igbvf_setup_rx_resources(struct igbvf_adapter *adapter,
 467                              struct igbvf_ring *rx_ring)
 468 {
 469         struct pci_dev *pdev = adapter->pdev;
 470         int size, desc_len;
 471
 472         size = sizeof(struct igbvf_buffer) * rx_ring->count;
 473         rx_ring->buffer_info = vmalloc(size);
 474         if (!rx_ring->buffer_info)
 475                 goto err;
 476         memset(rx_ring->buffer_info, 0, size);
 477
 478         desc_len = sizeof(union e1000_adv_rx_desc);
 479
 480         /* Round up to nearest 4K */
 481         rx_ring->size = rx_ring->count * desc_len;
 482         rx_ring->size = ALIGN(rx_ring->size, 4096);
 483
 484         rx_ring->desc = dma_alloc_coherent(&pdev->dev, rx_ring->size,
 485                                            &rx_ring->dma, GFP_KERNEL);
 486
 487         if (!rx_ring->desc)
 488                 goto err;
 489
 490         rx_ring->next_to_clean = 0;
 491         rx_ring->next_to_use = 0;
 492
 493         rx_ring->adapter = adapter;
 494
 495         return 0;
 496
 497 err:
 498         vfree(rx_ring->buffer_info);
 499         rx_ring->buffer_info = NULL;
 500         dev_err(&adapter->pdev->dev,
 501                 "Unable to allocate memory for the receive descriptor ring\n");
 502         return -ENOMEM;
 503 }
 504
 505 /**
 506  * igbvf_clean_tx_ring - Free Tx Buffers
 507  * @tx_ring: ring to be cleaned
 508  **/
 509 static void igbvf_clean_tx_ring(struct igbvf_ring *tx_ring)
 510 {
 511         struct igbvf_adapter *adapter = tx_ring->adapter;
 512         struct igbvf_buffer *buffer_info;
 513         unsigned long size;
 514         unsigned int i;
 515
 516         if (!tx_ring->buffer_info)
 517                 return;
 518
 519         /* Free all the Tx ring sk_buffs */
 520         for (i = 0; i < tx_ring->count; i++) {
 521                 buffer_info = &tx_ring->buffer_info[i];
 522                 igbvf_put_txbuf(adapter, buffer_info);
 523         }
 524
 525         size = sizeof(struct igbvf_buffer) * tx_ring->count;
 526         memset(tx_ring->buffer_info, 0, size);
 527
 528         /* Zero out the descriptor ring */
 529         memset(tx_ring->desc, 0, tx_ring->size);
 530
 531         tx_ring->next_to_use = 0;
 532         tx_ring->next_to_clean = 0;
 533
 534         writel(0, adapter->hw.hw_addr + tx_ring->head);
 535         writel(0, adapter->hw.hw_addr + tx_ring->tail);
 536 }
 537
 538 /**
 539  * igbvf_free_tx_resources - Free Tx Resources per Queue
 540  * @tx_ring: ring to free resources from
 541  *
 542  * Free all transmit software resources
 543  **/
 544 void igbvf_free_tx_resources(struct igbvf_ring *tx_ring)
 545 {
 546         struct pci_dev *pdev = tx_ring->adapter->pdev;
 547
 548         igbvf_clean_tx_ring(tx_ring);
 549
 550         vfree(tx_ring->buffer_info);
 551         tx_ring->buffer_info = NULL;
 552
 553         dma_free_coherent(&pdev->dev, tx_ring->size, tx_ring->desc,
 554                           tx_ring->dma);
 555
 556         tx_ring->desc = NULL;
 557 }
 558
 559 /**
 560  * igbvf_clean_rx_ring - Free Rx Buffers per Queue
 561  * @adapter: board private structure
 562  **/
 563 static void igbvf_clean_rx_ring(struct igbvf_ring *rx_ring)
 564 {
 565         struct igbvf_adapter *adapter = rx_ring->adapter;
 566         struct igbvf_buffer *buffer_info;
 567         struct pci_dev *pdev = adapter->pdev;
 568         unsigned long size;
 569         unsigned int i;
 570
 571         if (!rx_ring->buffer_info)
 572                 return;
 573
 574         /* Free all the Rx ring sk_buffs */
 575         for (i = 0; i < rx_ring->count; i++) {
 576                 buffer_info = &rx_ring->buffer_info[i];
 577                 if (buffer_info->dma) {
 578                         if (adapter->rx_ps_hdr_size){
 579                                 dma_unmap_single(&pdev->dev, buffer_info->dma,
 580                                                  adapter->rx_ps_hdr_size,
 581                                                  DMA_FROM_DEVICE);
 582                         } else {
 583                                 dma_unmap_single(&pdev->dev, buffer_info->dma,
 584                                                  adapter->rx_buffer_len,
 585                                                  DMA_FROM_DEVICE);
 586                         }
 587                         buffer_info->dma = 0;
 588                 }
 589
 590                 if (buffer_info->skb) {
 591                         dev_kfree_skb(buffer_info->skb);
 592                         buffer_info->skb = NULL;
 593                 }
 594
 595                 if (buffer_info->page) {
 596                         if (buffer_info->page_dma)
 597                                 dma_unmap_page(&pdev->dev,
 598                                                buffer_info->page_dma,
 599                                                PAGE_SIZE / 2,
 600                                                DMA_FROM_DEVICE);
 601                         put_page(buffer_info->page);
 602                         buffer_info->page = NULL;
 603                         buffer_info->page_dma = 0;
 604                         buffer_info->page_offset = 0;
 605                 }
 606         }
 607
 608         size = sizeof(struct igbvf_buffer) * rx_ring->count;
 609         memset(rx_ring->buffer_info, 0, size);
 610
 611         /* Zero out the descriptor ring */
 612         memset(rx_ring->desc, 0, rx_ring->size);
 613
 614         rx_ring->next_to_clean = 0;
 615         rx_ring->next_to_use = 0;
 616
 617         writel(0, adapter->hw.hw_addr + rx_ring->head);
 618         writel(0, adapter->hw.hw_addr + rx_ring->tail);
 619 }
 620
 621 /**
 622  * igbvf_free_rx_resources - Free Rx Resources
 623  * @rx_ring: ring to clean the resources from
 624  *
 625  * Free all receive software resources
 626  **/
 627
 628 void igbvf_free_rx_resources(struct igbvf_ring *rx_ring)
 629 {
 630         struct pci_dev *pdev = rx_ring->adapter->pdev;
 631
 632         igbvf_clean_rx_ring(rx_ring);
 633
 634         vfree(rx_ring->buffer_info);
 635         rx_ring->buffer_info = NULL;
 636
 637         dma_free_coherent(&pdev->dev, rx_ring->size, rx_ring->desc,
 638                           rx_ring->dma);
 639         rx_ring->desc = NULL;
 640 }
 641
 642 /**
 643  * igbvf_update_itr - update the dynamic ITR value based on statistics
 644  * @adapter: pointer to adapter
 645  * @itr_setting: current adapter->itr
 646  * @packets: the number of packets during this measurement interval
 647  * @bytes: the number of bytes during this measurement interval
 648  *
 649  *      Stores a new ITR value based on packets and byte
 650  *      counts during the last interrupt.  The advantage of per interrupt
 651  *      computation is faster updates and more accurate ITR for the current
 652  *      traffic pattern.  Constants in this function were computed
 653  *      based on theoretical maximum wire speed and thresholds were set based
 654  *      on testing data as well as attempting to minimize response time
 655  *      while increasing bulk throughput.  This functionality is controlled
 656  *      by the InterruptThrottleRate module parameter.
 657  **/
 658 static unsigned int igbvf_update_itr(struct igbvf_adapter *adapter,
 659                                      u16 itr_setting, int packets,
 660                                      int bytes)
 661 {
 662         unsigned int retval = itr_setting;
 663
 664         if (packets == 0)
 665                 goto update_itr_done;
 666
 667         switch (itr_setting) {
 668         case lowest_latency:
 669                 /* handle TSO and jumbo frames */
 670                 if (bytes/packets > 8000)
 671                         retval = bulk_latency;
 672                 else if ((packets < 5) && (bytes > 512))
 673                         retval = low_latency;
 674                 break;
 675         case low_latency:  /* 50 usec aka 20000 ints/s */
 676                 if (bytes > 10000) {
 677                         /* this if handles the TSO accounting */
 678                         if (bytes/packets > 8000)
 679                                 retval = bulk_latency;
 680                         else if ((packets < 10) || ((bytes/packets) > 1200))
 681                                 retval = bulk_latency;
 682                         else if ((packets > 35))
 683                                 retval = lowest_latency;
 684                 } else if (bytes/packets > 2000) {
 685                         retval = bulk_latency;
 686                 } else if (packets <= 2 && bytes < 512) {
 687                         retval = lowest_latency;
 688                 }
 689                 break;
 690         case bulk_latency: /* 250 usec aka 4000 ints/s */
 691                 if (bytes > 25000) {
 692                         if (packets > 35)
 693                                 retval = low_latency;
 694                 } else if (bytes < 6000) {
 695                         retval = low_latency;
 696                 }
 697                 break;
 698         }
 699
 700 update_itr_done:
 701         return retval;
 702 }
 703
 704 static void igbvf_set_itr(struct igbvf_adapter *adapter)
 705 {
 706         struct e1000_hw *hw = &adapter->hw;
 707         u16 current_itr;
 708         u32 new_itr = adapter->itr;
 709
 710         adapter->tx_itr = igbvf_update_itr(adapter, adapter->tx_itr,
 711                                            adapter->total_tx_packets,
 712                                            adapter->total_tx_bytes);
 713         /* conservative mode (itr 3) eliminates the lowest_latency setting */
 714         if (adapter->itr_setting == 3 && adapter->tx_itr == lowest_latency)
 715                 adapter->tx_itr = low_latency;
 716
 717         adapter->rx_itr = igbvf_update_itr(adapter, adapter->rx_itr,
 718                                            adapter->total_rx_packets,
 719                                            adapter->total_rx_bytes);
 720         /* conservative mode (itr 3) eliminates the lowest_latency setting */
 721         if (adapter->itr_setting == 3 && adapter->rx_itr == lowest_latency)
 722                 adapter->rx_itr = low_latency;
 723
 724         current_itr = max(adapter->rx_itr, adapter->tx_itr);
 725
 726         switch (current_itr) {
 727         /* counts and packets in update_itr are dependent on these numbers */
 728         case lowest_latency:
 729                 new_itr = 70000;
 730                 break;
 731         case low_latency:
 732                 new_itr = 20000; /* aka hwitr = ~200 */
 733                 break;
 734         case bulk_latency:
 735                 new_itr = 4000;
 736                 break;
 737         default:
 738                 break;
 739         }
 740
 741         if (new_itr != adapter->itr) {
 742                 /*
 743                  * this attempts to bias the interrupt rate towards Bulk
 744                  * by adding intermediate steps when interrupt rate is
 745                  * increasing
 746                  */
 747                 new_itr = new_itr > adapter->itr ?
 748                              min(adapter->itr + (new_itr >> 2), new_itr) :
 749                              new_itr;
 750                 adapter->itr = new_itr;
 751                 adapter->rx_ring->itr_val = 1952;
 752
 753                 if (adapter->msix_entries)
 754                         adapter->rx_ring->set_itr = 1;
 755                 else
 756                         ew32(ITR, 1952);
 757         }
 758 }
 759
 760 /**
 761  * igbvf_clean_tx_irq - Reclaim resources after transmit completes
 762  * @adapter: board private structure
 763  * returns true if ring is completely cleaned
 764  **/
 765 static bool igbvf_clean_tx_irq(struct igbvf_ring *tx_ring)
 766 {
 767         struct igbvf_adapter *adapter = tx_ring->adapter;
 768         struct e1000_hw *hw = &adapter->hw;
 769         struct net_device *netdev = adapter->netdev;
 770         struct igbvf_buffer *buffer_info;
 771         struct sk_buff *skb;
 772         union e1000_adv_tx_desc *tx_desc, *eop_desc;
 773         unsigned int total_bytes = 0, total_packets = 0;
 774         unsigned int i, eop, count = 0;
 775         bool cleaned = false;
 776
 777         i = tx_ring->next_to_clean;
 778         eop = tx_ring->buffer_info[i].next_to_watch;
 779         eop_desc = IGBVF_TX_DESC_ADV(*tx_ring, eop);
 780
 781         while ((eop_desc->wb.status & cpu_to_le32(E1000_TXD_STAT_DD)) &&
 782                (count < tx_ring->count)) {
 783                 for (cleaned = false; !cleaned; count++) {
 784                         tx_desc = IGBVF_TX_DESC_ADV(*tx_ring, i);
 785                         buffer_info = &tx_ring->buffer_info[i];
 786                         cleaned = (i == eop);
 787                         skb = buffer_info->skb;
 788
 789                         if (skb) {
 790                                 unsigned int segs, bytecount;
 791
 792                                 /* gso_segs is currently only valid for tcp */
 793                                 segs = skb_shinfo(skb)->gso_segs ?: 1;
 794                                 /* multiply data chunks by size of headers */
 795                                 bytecount = ((segs - 1) * skb_headlen(skb)) +
 796                                             skb->len;
 797                                 total_packets += segs;
 798                                 total_bytes += bytecount;
 799                         }
 800
 801                         igbvf_put_txbuf(adapter, buffer_info);
 802                         tx_desc->wb.status = 0;
 803
 804                         i++;
 805                         if (i == tx_ring->count)
 806                                 i = 0;
 807                 }
 808                 eop = tx_ring->buffer_info[i].next_to_watch;
 809                 eop_desc = IGBVF_TX_DESC_ADV(*tx_ring, eop);
 810         }
 811
 812         tx_ring->next_to_clean = i;
 813
 814         if (unlikely(count &&
 815                      netif_carrier_ok(netdev) &&
 816                      igbvf_desc_unused(tx_ring) >= IGBVF_TX_QUEUE_WAKE)) {
 817                 /* Make sure that anybody stopping the queue after this
 818                  * sees the new next_to_clean.
 819                  */
 820                 smp_mb();
 821                 if (netif_queue_stopped(netdev) &&
 822                     !(test_bit(__IGBVF_DOWN, &adapter->state))) {
 823                         netif_wake_queue(netdev);
 824                         ++adapter->restart_queue;
 825                 }
 826         }
 827
 828         if (adapter->detect_tx_hung) {
 829                 /* Detect a transmit hang in hardware, this serializes the
 830                  * check with the clearing of time_stamp and movement of i */
 831                 adapter->detect_tx_hung = false;
 832                 if (tx_ring->buffer_info[i].time_stamp &&
 833                     time_after(jiffies, tx_ring->buffer_info[i].time_stamp +
 834                                (adapter->tx_timeout_factor * HZ)) &&
 835                     !(er32(STATUS) & E1000_STATUS_TXOFF)) {
 836
 837                         tx_desc = IGBVF_TX_DESC_ADV(*tx_ring, i);
 838                         /* detected Tx unit hang */
 839                         igbvf_print_tx_hang(adapter);
 840
 841                         netif_stop_queue(netdev);
 842                 }
 843         }
 844         adapter->net_stats.tx_bytes += total_bytes;
 845         adapter->net_stats.tx_packets += total_packets;
 846         return (count < tx_ring->count);
 847 }
 848
 849 static irqreturn_t igbvf_msix_other(int irq, void *data)
 850 {
 851         struct net_device *netdev = data;
 852         struct igbvf_adapter *adapter = netdev_priv(netdev);
 853         struct e1000_hw *hw = &adapter->hw;
 854
 855         adapter->int_counter1++;
 856
 857         netif_carrier_off(netdev);
 858         hw->mac.get_link_status = 1;
 859         if (!test_bit(__IGBVF_DOWN, &adapter->state))
 860                 mod_timer(&adapter->watchdog_timer, jiffies + 1);
 861
 862         ew32(EIMS, adapter->eims_other);
 863
 864         return IRQ_HANDLED;
 865 }
 866
 867 static irqreturn_t igbvf_intr_msix_tx(int irq, void *data)
 868 {
 869         struct net_device *netdev = data;
 870         struct igbvf_adapter *adapter = netdev_priv(netdev);
 871         struct e1000_hw *hw = &adapter->hw;
 872         struct igbvf_ring *tx_ring = adapter->tx_ring;
 873
 874
 875         adapter->total_tx_bytes = 0;
 876         adapter->total_tx_packets = 0;
 877
 878         /* auto mask will automatically reenable the interrupt when we write
 879          * EICS */
 880         if (!igbvf_clean_tx_irq(tx_ring))
 881                 /* Ring was not completely cleaned, so fire another interrupt */
 882                 ew32(EICS, tx_ring->eims_value);
 883         else
 884                 ew32(EIMS, tx_ring->eims_value);
 885
 886         return IRQ_HANDLED;
 887 }
 888
 889 static irqreturn_t igbvf_intr_msix_rx(int irq, void *data)
 890 {
 891         struct net_device *netdev = data;
 892         struct igbvf_adapter *adapter = netdev_priv(netdev);
 893
 894         adapter->int_counter0++;
 895
 896         /* Write the ITR value calculated at the end of the
 897          * previous interrupt.
 898          */
 899         if (adapter->rx_ring->set_itr) {
 900                 writel(adapter->rx_ring->itr_val,
 901                        adapter->hw.hw_addr + adapter->rx_ring->itr_register);
 902                 adapter->rx_ring->set_itr = 0;
 903         }
 904
 905         if (napi_schedule_prep(&adapter->rx_ring->napi)) {
 906                 adapter->total_rx_bytes = 0;
 907                 adapter->total_rx_packets = 0;
 908                 __napi_schedule(&adapter->rx_ring->napi);
 909         }
 910
 911         return IRQ_HANDLED;
 912 }
 913
 914 #define IGBVF_NO_QUEUE -1
 915
 916 static void igbvf_assign_vector(struct igbvf_adapter *adapter, int rx_queue,
 917                                 int tx_queue, int msix_vector)
 918 {
 919         struct e1000_hw *hw = &adapter->hw;
 920         u32 ivar, index;
 921
 922         /* 82576 uses a table-based method for assigning vectors.
 923            Each queue has a single entry in the table to which we write
 924            a vector number along with a "valid" bit.  Sadly, the layout
 925            of the table is somewhat counterintuitive. */
 926         if (rx_queue > IGBVF_NO_QUEUE) {
 927                 index = (rx_queue >> 1);
 928                 ivar = array_er32(IVAR0, index);
 929                 if (rx_queue & 0x1) {
 930                         /* vector goes into third byte of register */
 931                         ivar = ivar & 0xFF00FFFF;
 932                         ivar |= (msix_vector | E1000_IVAR_VALID) << 16;
 933                 } else {
 934                         /* vector goes into low byte of register */
 935                         ivar = ivar & 0xFFFFFF00;
 936                         ivar |= msix_vector | E1000_IVAR_VALID;
 937                 }
 938                 adapter->rx_ring[rx_queue].eims_value = 1 << msix_vector;
 939                 array_ew32(IVAR0, index, ivar);
 940         }
 941         if (tx_queue > IGBVF_NO_QUEUE) {
 942                 index = (tx_queue >> 1);
 943                 ivar = array_er32(IVAR0, index);
 944                 if (tx_queue & 0x1) {
 945                         /* vector goes into high byte of register */
 946                         ivar = ivar & 0x00FFFFFF;
 947                         ivar |= (msix_vector | E1000_IVAR_VALID) << 24;
 948                 } else {
 949                         /* vector goes into second byte of register */
 950                         ivar = ivar & 0xFFFF00FF;
 951                         ivar |= (msix_vector | E1000_IVAR_VALID) << 8;
 952                 }
 953                 adapter->tx_ring[tx_queue].eims_value = 1 << msix_vector;
 954                 array_ew32(IVAR0, index, ivar);
 955         }
 956 }
 957
 958 /**
 959  * igbvf_configure_msix - Configure MSI-X hardware
 960  *
 961  * igbvf_configure_msix sets up the hardware to properly
 962  * generate MSI-X interrupts.
 963  **/
 964 static void igbvf_configure_msix(struct igbvf_adapter *adapter)
 965 {
 966         u32 tmp;
 967         struct e1000_hw *hw = &adapter->hw;
 968         struct igbvf_ring *tx_ring = adapter->tx_ring;
 969         struct igbvf_ring *rx_ring = adapter->rx_ring;
 970         int vector = 0;
 971
 972         adapter->eims_enable_mask = 0;
 973
 974         igbvf_assign_vector(adapter, IGBVF_NO_QUEUE, 0, vector++);
 975         adapter->eims_enable_mask |= tx_ring->eims_value;
 976         if (tx_ring->itr_val)
 977                 writel(tx_ring->itr_val,
 978                        hw->hw_addr + tx_ring->itr_register);
 979         else
 980                 writel(1952, hw->hw_addr + tx_ring->itr_register);
 981
 982         igbvf_assign_vector(adapter, 0, IGBVF_NO_QUEUE, vector++);
 983         adapter->eims_enable_mask |= rx_ring->eims_value;
 984         if (rx_ring->itr_val)
 985                 writel(rx_ring->itr_val,
 986                        hw->hw_addr + rx_ring->itr_register);
 987         else
 988                 writel(1952, hw->hw_addr + rx_ring->itr_register);
 989
 990         /* set vector for other causes, i.e. link changes */
 991
 992         tmp = (vector++ | E1000_IVAR_VALID);
 993
 994         ew32(IVAR_MISC, tmp);
 995
 996         adapter->eims_enable_mask = (1 << (vector)) - 1;
 997         adapter->eims_other = 1 << (vector - 1);
 998         e1e_flush();
 999 }
1000
1001 static void igbvf_reset_interrupt_capability(struct igbvf_adapter *adapter)
1002 {
1003         if (adapter->msix_entries) {
1004                 pci_disable_msix(adapter->pdev);
1005                 kfree(adapter->msix_entries);
1006                 adapter->msix_entries = NULL;
1007         }
1008 }
1009
1010 /**
1011  * igbvf_set_interrupt_capability - set MSI or MSI-X if supported
1012  *
1013  * Attempt to configure interrupts using the best available
1014  * capabilities of the hardware and kernel.
1015  **/
1016 static void igbvf_set_interrupt_capability(struct igbvf_adapter *adapter)
1017 {
1018         int err = -ENOMEM;
1019         int i;
1020
1021         /* we allocate 3 vectors, 1 for tx, 1 for rx, one for pf messages */
1022         adapter->msix_entries = kcalloc(3, sizeof(struct msix_entry),
1023                                         GFP_KERNEL);
1024         if (adapter->msix_entries) {
1025                 for (i = 0; i < 3; i++)
1026                         adapter->msix_entries[i].entry = i;
1027
1028                 err = pci_enable_msix(adapter->pdev,
1029                                       adapter->msix_entries, 3);
1030         }
1031
1032         if (err) {
1033                 /* MSI-X failed */
1034                 dev_err(&adapter->pdev->dev,
1035                         "Failed to initialize MSI-X interrupts.\n");
1036                 igbvf_reset_interrupt_capability(adapter);
1037         }
1038 }
1039
1040 /**
1041  * igbvf_request_msix - Initialize MSI-X interrupts
1042  *
1043  * igbvf_request_msix allocates MSI-X vectors and requests interrupts from the
1044  * kernel.
1045  **/
1046 static int igbvf_request_msix(struct igbvf_adapter *adapter)
1047 {
1048         struct net_device *netdev = adapter->netdev;
1049         int err = 0, vector = 0;
1050
1051         if (strlen(netdev->name) < (IFNAMSIZ - 5)) {
1052                 sprintf(adapter->tx_ring->name, "%s-tx-0", netdev->name);
1053                 sprintf(adapter->rx_ring->name, "%s-rx-0", netdev->name);
1054         } else {
1055                 memcpy(adapter->tx_ring->name, netdev->name, IFNAMSIZ);
1056                 memcpy(adapter->rx_ring->name, netdev->name, IFNAMSIZ);
1057         }
1058
1059         err = request_irq(adapter->msix_entries[vector].vector,
1060                           igbvf_intr_msix_tx, 0, adapter->tx_ring->name,
1061                           netdev);
1062         if (err)
1063                 goto out;
1064
1065         adapter->tx_ring->itr_register = E1000_EITR(vector);
1066         adapter->tx_ring->itr_val = 1952;
1067         vector++;
1068
1069         err = request_irq(adapter->msix_entries[vector].vector,
1070                           igbvf_intr_msix_rx, 0, adapter->rx_ring->name,
1071                           netdev);
1072         if (err)
1073                 goto out;
1074
1075         adapter->rx_ring->itr_register = E1000_EITR(vector);
1076         adapter->rx_ring->itr_val = 1952;
1077         vector++;
1078
1079         err = request_irq(adapter->msix_entries[vector].vector,
1080                           igbvf_msix_other, 0, netdev->name, netdev);
1081         if (err)
1082                 goto out;
1083
1084         igbvf_configure_msix(adapter);
1085         return 0;
1086 out:
1087         return err;
1088 }
1089
1090 /**
1091  * igbvf_alloc_queues - Allocate memory for all rings
1092  * @adapter: board private structure to initialize
1093  **/
1094 static int __devinit igbvf_alloc_queues(struct igbvf_adapter *adapter)
1095 {
1096         struct net_device *netdev = adapter->netdev;
1097
1098         adapter->tx_ring = kzalloc(sizeof(struct igbvf_ring), GFP_KERNEL);
1099         if (!adapter->tx_ring)
1100                 return -ENOMEM;
1101
1102         adapter->rx_ring = kzalloc(sizeof(struct igbvf_ring), GFP_KERNEL);
1103         if (!adapter->rx_ring) {
1104                 kfree(adapter->tx_ring);
1105                 return -ENOMEM;
1106         }
1107
1108         netif_napi_add(netdev, &adapter->rx_ring->napi, igbvf_poll, 64);
1109
1110         return 0;
1111 }
1112
1113 /**
1114  * igbvf_request_irq - initialize interrupts
1115  *
1116  * Attempts to configure interrupts using the best available
1117  * capabilities of the hardware and kernel.
1118  **/
1119 static int igbvf_request_irq(struct igbvf_adapter *adapter)
1120 {
1121         int err = -1;
1122
1123         /* igbvf supports msi-x only */
1124         if (adapter->msix_entries)
1125                 err = igbvf_request_msix(adapter);
1126
1127         if (!err)
1128                 return err;
1129
1130         dev_err(&adapter->pdev->dev,
1131                 "Unable to allocate interrupt, Error: %d\n", err);
1132
1133         return err;
1134 }
1135
1136 static void igbvf_free_irq(struct igbvf_adapter *adapter)
1137 {
1138         struct net_device *netdev = adapter->netdev;
1139         int vector;
1140
1141         if (adapter->msix_entries) {
1142                 for (vector = 0; vector < 3; vector++)
1143                         free_irq(adapter->msix_entries[vector].vector, netdev);
1144         }
1145 }
1146
1147 /**
1148  * igbvf_irq_disable - Mask off interrupt generation on the NIC
1149  **/
1150 static void igbvf_irq_disable(struct igbvf_adapter *adapter)
1151 {
1152         struct e1000_hw *hw = &adapter->hw;
1153
1154         ew32(EIMC, ~0);
1155
1156         if (adapter->msix_entries)
1157                 ew32(EIAC, 0);
1158 }
1159
1160 /**
1161  * igbvf_irq_enable - Enable default interrupt generation settings
1162  **/
1163 static void igbvf_irq_enable(struct igbvf_adapter *adapter)
1164 {
1165         struct e1000_hw *hw = &adapter->hw;
1166
1167         ew32(EIAC, adapter->eims_enable_mask);
1168         ew32(EIAM, adapter->eims_enable_mask);
1169         ew32(EIMS, adapter->eims_enable_mask);
1170 }
1171
1172 /**
1173  * igbvf_poll - NAPI Rx polling callback
1174  * @napi: struct associated with this polling callback
1175  * @budget: amount of packets driver is allowed to process this poll
1176  **/
1177 static int igbvf_poll(struct napi_struct *napi, int budget)
1178 {
1179         struct igbvf_ring *rx_ring = container_of(napi, struct igbvf_ring, napi);
1180         struct igbvf_adapter *adapter = rx_ring->adapter;
1181         struct e1000_hw *hw = &adapter->hw;
1182         int work_done = 0;
1183
1184         igbvf_clean_rx_irq(adapter, &work_done, budget);
1185
1186         /* If not enough Rx work done, exit the polling mode */
1187         if (work_done < budget) {
1188                 napi_complete(napi);
1189
1190                 if (adapter->itr_setting & 3)
1191                         igbvf_set_itr(adapter);
1192
1193                 if (!test_bit(__IGBVF_DOWN, &adapter->state))
1194                         ew32(EIMS, adapter->rx_ring->eims_value);
1195         }
1196
1197         return work_done;
1198 }
1199
1200 /**
1201  * igbvf_set_rlpml - set receive large packet maximum length
1202  * @adapter: board private structure
1203  *
1204  * Configure the maximum size of packets that will be received
1205  */
1206 static void igbvf_set_rlpml(struct igbvf_adapter *adapter)
1207 {
1208         int max_frame_size = adapter->max_frame_size;
1209         struct e1000_hw *hw = &adapter->hw;
1210
1211         if (adapter->vlgrp)
1212                 max_frame_size += VLAN_TAG_SIZE;
1213
1214         e1000_rlpml_set_vf(hw, max_frame_size);
1215 }
1216
1217 static void igbvf_vlan_rx_add_vid(struct net_device *netdev, u16 vid)
1218 {
1219         struct igbvf_adapter *adapter = netdev_priv(netdev);
1220         struct e1000_hw *hw = &adapter->hw;
1221
1222         if (hw->mac.ops.set_vfta(hw, vid, true))
1223                 dev_err(&adapter->pdev->dev, "Failed to add vlan id %d\n", vid);
1224 }
1225
1226 static void igbvf_vlan_rx_kill_vid(struct net_device *netdev, u16 vid)
1227 {
1228         struct igbvf_adapter *adapter = netdev_priv(netdev);
1229         struct e1000_hw *hw = &adapter->hw;
1230
1231         igbvf_irq_disable(adapter);
1232         vlan_group_set_device(adapter->vlgrp, vid, NULL);
1233
1234         if (!test_bit(__IGBVF_DOWN, &adapter->state))
1235                 igbvf_irq_enable(adapter);
1236
1237         if (hw->mac.ops.set_vfta(hw, vid, false))
1238                 dev_err(&adapter->pdev->dev,
1239                         "Failed to remove vlan id %d\n", vid);
1240 }
1241
1242 static void igbvf_vlan_rx_register(struct net_device *netdev,
1243                                    struct vlan_group *grp)
1244 {
1245         struct igbvf_adapter *adapter = netdev_priv(netdev);
1246
1247         adapter->vlgrp = grp;
1248 }
1249
1250 static void igbvf_restore_vlan(struct igbvf_adapter *adapter)
1251 {
1252         u16 vid;
1253
1254         if (!adapter->vlgrp)
1255                 return;
1256
1257         for (vid = 0; vid < VLAN_GROUP_ARRAY_LEN; vid++) {
1258                 if (!vlan_group_get_device(adapter->vlgrp, vid))
1259                         continue;
1260                 igbvf_vlan_rx_add_vid(adapter->netdev, vid);
1261         }
1262
1263         igbvf_set_rlpml(adapter);
1264 }
1265
1266 /**
1267  * igbvf_configure_tx - Configure Transmit Unit after Reset
1268  * @adapter: board private structure
1269  *
1270  * Configure the Tx unit of the MAC after a reset.
1271  **/
1272 static void igbvf_configure_tx(struct igbvf_adapter *adapter)
1273 {
1274         struct e1000_hw *hw = &adapter->hw;
1275         struct igbvf_ring *tx_ring = adapter->tx_ring;
1276         u64 tdba;
1277         u32 txdctl, dca_txctrl;
1278
1279         /* disable transmits */
1280         txdctl = er32(TXDCTL(0));
1281         ew32(TXDCTL(0), txdctl & ~E1000_TXDCTL_QUEUE_ENABLE);
1282         msleep(10);
1283
1284         /* Setup the HW Tx Head and Tail descriptor pointers */
1285         ew32(TDLEN(0), tx_ring->count * sizeof(union e1000_adv_tx_desc));
1286         tdba = tx_ring->dma;
1287         ew32(TDBAL(0), (tdba & DMA_BIT_MASK(32)));
1288         ew32(TDBAH(0), (tdba >> 32));
1289         ew32(TDH(0), 0);
1290         ew32(TDT(0), 0);
1291         tx_ring->head = E1000_TDH(0);
1292         tx_ring->tail = E1000_TDT(0);
1293
1294         /* Turn off Relaxed Ordering on head write-backs.  The writebacks
1295          * MUST be delivered in order or it will completely screw up
1296          * our bookeeping.
1297          */
1298         dca_txctrl = er32(DCA_TXCTRL(0));
1299         dca_txctrl &= ~E1000_DCA_TXCTRL_TX_WB_RO_EN;
1300         ew32(DCA_TXCTRL(0), dca_txctrl);
1301
1302         /* enable transmits */
1303         txdctl |= E1000_TXDCTL_QUEUE_ENABLE;
1304         ew32(TXDCTL(0), txdctl);
1305
1306         /* Setup Transmit Descriptor Settings for eop descriptor */
1307         adapter->txd_cmd = E1000_ADVTXD_DCMD_EOP | E1000_ADVTXD_DCMD_IFCS;
1308
1309         /* enable Report Status bit */
1310         adapter->txd_cmd |= E1000_ADVTXD_DCMD_RS;
1311 }
1312
1313 /**
1314  * igbvf_setup_srrctl - configure the receive control registers
1315  * @adapter: Board private structure
1316  **/
1317 static void igbvf_setup_srrctl(struct igbvf_adapter *adapter)
1318 {
1319         struct e1000_hw *hw = &adapter->hw;
1320         u32 srrctl = 0;
1321
1322         srrctl &= ~(E1000_SRRCTL_DESCTYPE_MASK |
1323                     E1000_SRRCTL_BSIZEHDR_MASK |
1324                     E1000_SRRCTL_BSIZEPKT_MASK);
1325
1326         /* Enable queue drop to avoid head of line blocking */
1327         srrctl |= E1000_SRRCTL_DROP_EN;
1328
1329         /* Setup buffer sizes */
1330         srrctl |= ALIGN(adapter->rx_buffer_len, 1024) >>
1331                   E1000_SRRCTL_BSIZEPKT_SHIFT;
1332
1333         if (adapter->rx_buffer_len < 2048) {
1334                 adapter->rx_ps_hdr_size = 0;
1335                 srrctl |= E1000_SRRCTL_DESCTYPE_ADV_ONEBUF;
1336         } else {
1337                 adapter->rx_ps_hdr_size = 128;
1338                 srrctl |= adapter->rx_ps_hdr_size <<
1339                           E1000_SRRCTL_BSIZEHDRSIZE_SHIFT;
1340                 srrctl |= E1000_SRRCTL_DESCTYPE_HDR_SPLIT_ALWAYS;
1341         }
1342
1343         ew32(SRRCTL(0), srrctl);
1344 }
1345
1346 /**
1347  * igbvf_configure_rx - Configure Receive Unit after Reset
1348  * @adapter: board private structure
1349  *
1350  * Configure the Rx unit of the MAC after a reset.
1351  **/
1352 static void igbvf_configure_rx(struct igbvf_adapter *adapter)
1353 {
1354         struct e1000_hw *hw = &adapter->hw;
1355         struct igbvf_ring *rx_ring = adapter->rx_ring;
1356         u64 rdba;
1357         u32 rdlen, rxdctl;
1358
1359         /* disable receives */
1360         rxdctl = er32(RXDCTL(0));
1361         ew32(RXDCTL(0), rxdctl & ~E1000_RXDCTL_QUEUE_ENABLE);
1362         msleep(10);
1363
1364         rdlen = rx_ring->count * sizeof(union e1000_adv_rx_desc);
1365
1366         /*
1367          * Setup the HW Rx Head and Tail Descriptor Pointers and
1368          * the Base and Length of the Rx Descriptor Ring
1369          */
1370         rdba = rx_ring->dma;
1371         ew32(RDBAL(0), (rdba & DMA_BIT_MASK(32)));
1372         ew32(RDBAH(0), (rdba >> 32));
1373         ew32(RDLEN(0), rx_ring->count * sizeof(union e1000_adv_rx_desc));
1374         rx_ring->head = E1000_RDH(0);
1375         rx_ring->tail = E1000_RDT(0);
1376         ew32(RDH(0), 0);
1377         ew32(RDT(0), 0);
1378
1379         rxdctl |= E1000_RXDCTL_QUEUE_ENABLE;
1380         rxdctl &= 0xFFF00000;
1381         rxdctl |= IGBVF_RX_PTHRESH;
1382         rxdctl |= IGBVF_RX_HTHRESH << 8;
1383         rxdctl |= IGBVF_RX_WTHRESH << 16;
1384
1385         igbvf_set_rlpml(adapter);
1386
1387         /* enable receives */
1388         ew32(RXDCTL(0), rxdctl);
1389 }
1390
1391 /**
1392  * igbvf_set_multi - Multicast and Promiscuous mode set
1393  * @netdev: network interface device structure
1394  *
1395  * The set_multi entry point is called whenever the multicast address
1396  * list or the network interface flags are updated.  This routine is
1397  * responsible for configuring the hardware for proper multicast,
1398  * promiscuous mode, and all-multi behavior.
1399  **/
1400 static void igbvf_set_multi(struct net_device *netdev)
1401 {
1402         struct igbvf_adapter *adapter = netdev_priv(netdev);
1403         struct e1000_hw *hw = &adapter->hw;
1404         struct netdev_hw_addr *ha;
1405         u8  *mta_list = NULL;
1406         int i;
1407
1408         if (!netdev_mc_empty(netdev)) {
1409                 mta_list = kmalloc(netdev_mc_count(netdev) * 6, GFP_ATOMIC);
1410                 if (!mta_list) {
1411                         dev_err(&adapter->pdev->dev,
1412                                 "failed to allocate multicast filter list\n");
1413                         return;
1414                 }
1415         }
1416
1417         /* prepare a packed array of only addresses. */
1418         i = 0;
1419         netdev_for_each_mc_addr(ha, netdev)
1420                 memcpy(mta_list + (i++ * ETH_ALEN), ha->addr, ETH_ALEN);
1421
1422         hw->mac.ops.update_mc_addr_list(hw, mta_list, i, 0, 0);
1423         kfree(mta_list);
1424 }
1425
1426 /**
1427  * igbvf_configure - configure the hardware for Rx and Tx
1428  * @adapter: private board structure
1429  **/
1430 static void igbvf_configure(struct igbvf_adapter *adapter)
1431 {
1432         igbvf_set_multi(adapter->netdev);
1433
1434         igbvf_restore_vlan(adapter);
1435
1436         igbvf_configure_tx(adapter);
1437         igbvf_setup_srrctl(adapter);
1438         igbvf_configure_rx(adapter);
1439         igbvf_alloc_rx_buffers(adapter->rx_ring,
1440                                igbvf_desc_unused(adapter->rx_ring));
1441 }
1442
1443 /* igbvf_reset - bring the hardware into a known good state
1444  *
1445  * This function boots the hardware and enables some settings that
1446  * require a configuration cycle of the hardware - those cannot be
1447  * set/changed during runtime. After reset the device needs to be
1448  * properly configured for Rx, Tx etc.
1449  */
1450 static void igbvf_reset(struct igbvf_adapter *adapter)
1451 {
1452         struct e1000_mac_info *mac = &adapter->hw.mac;
1453         struct net_device *netdev = adapter->netdev;
1454         struct e1000_hw *hw = &adapter->hw;
1455
1456         /* Allow time for pending master requests to run */
1457         if (mac->ops.reset_hw(hw))
1458                 dev_err(&adapter->pdev->dev, "PF still resetting\n");
1459
1460         mac->ops.init_hw(hw);
1461
1462         if (is_valid_ether_addr(adapter->hw.mac.addr)) {
1463                 memcpy(netdev->dev_addr, adapter->hw.mac.addr,
1464                        netdev->addr_len);
1465                 memcpy(netdev->perm_addr, adapter->hw.mac.addr,
1466                        netdev->addr_len);
1467         }
1468
1469         adapter->last_reset = jiffies;
1470 }
1471
1472 int igbvf_up(struct igbvf_adapter *adapter)
1473 {
1474         struct e1000_hw *hw = &adapter->hw;
1475
1476         /* hardware has been reset, we need to reload some things */
1477         igbvf_configure(adapter);
1478
1479         clear_bit(__IGBVF_DOWN, &adapter->state);
1480
1481         napi_enable(&adapter->rx_ring->napi);
1482         if (adapter->msix_entries)
1483                 igbvf_configure_msix(adapter);
1484
1485         /* Clear any pending interrupts. */
1486         er32(EICR);
1487         igbvf_irq_enable(adapter);
1488
1489         /* start the watchdog */
1490         hw->mac.get_link_status = 1;
1491         mod_timer(&adapter->watchdog_timer, jiffies + 1);
1492
1493
1494         return 0;
1495 }
1496
1497 void igbvf_down(struct igbvf_adapter *adapter)
1498 {
1499         struct net_device *netdev = adapter->netdev;
1500         struct e1000_hw *hw = &adapter->hw;
1501         u32 rxdctl, txdctl;
1502
1503         /*
1504          * signal that we're down so the interrupt handler does not
1505          * reschedule our watchdog timer
1506          */
1507         set_bit(__IGBVF_DOWN, &adapter->state);
1508
1509         /* disable receives in the hardware */
1510         rxdctl = er32(RXDCTL(0));
1511         ew32(RXDCTL(0), rxdctl & ~E1000_RXDCTL_QUEUE_ENABLE);
1512
1513         netif_stop_queue(netdev);
1514
1515         /* disable transmits in the hardware */
1516         txdctl = er32(TXDCTL(0));
1517         ew32(TXDCTL(0), txdctl & ~E1000_TXDCTL_QUEUE_ENABLE);
1518
1519         /* flush both disables and wait for them to finish */
1520         e1e_flush();
1521         msleep(10);
1522
1523         napi_disable(&adapter->rx_ring->napi);
1524
1525         igbvf_irq_disable(adapter);
1526
1527         del_timer_sync(&adapter->watchdog_timer);
1528
1529         netif_carrier_off(netdev);
1530
1531         /* record the stats before reset*/
1532         igbvf_update_stats(adapter);
1533
1534         adapter->link_speed = 0;
1535         adapter->link_duplex = 0;
1536
1537         igbvf_reset(adapter);
1538         igbvf_clean_tx_ring(adapter->tx_ring);
1539         igbvf_clean_rx_ring(adapter->rx_ring);
1540 }
1541
1542 void igbvf_reinit_locked(struct igbvf_adapter *adapter)
1543 {
1544         might_sleep();
1545         while (test_and_set_bit(__IGBVF_RESETTING, &adapter->state))
1546                 msleep(1);
1547         igbvf_down(adapter);
1548         igbvf_up(adapter);
1549         clear_bit(__IGBVF_RESETTING, &adapter->state);
1550 }
1551
1552 /**
1553  * igbvf_sw_init - Initialize general software structures (struct igbvf_adapter)
1554  * @adapter: board private structure to initialize
1555  *
1556  * igbvf_sw_init initializes the Adapter private data structure.
1557  * Fields are initialized based on PCI device information and
1558  * OS network device settings (MTU size).
1559  **/
1560 static int __devinit igbvf_sw_init(struct igbvf_adapter *adapter)
1561 {
1562         struct net_device *netdev = adapter->netdev;
1563         s32 rc;
1564
1565         adapter->rx_buffer_len = ETH_FRAME_LEN + VLAN_HLEN + ETH_FCS_LEN;
1566         adapter->rx_ps_hdr_size = 0;
1567         adapter->max_frame_size = netdev->mtu + ETH_HLEN + ETH_FCS_LEN;
1568         adapter->min_frame_size = ETH_ZLEN + ETH_FCS_LEN;
1569
1570         adapter->tx_int_delay = 8;
1571         adapter->tx_abs_int_delay = 32;
1572         adapter->rx_int_delay = 0;
1573         adapter->rx_abs_int_delay = 8;
1574         adapter->itr_setting = 3;
1575         adapter->itr = 20000;
1576
1577         /* Set various function pointers */
1578         adapter->ei->init_ops(&adapter->hw);
1579
1580         rc = adapter->hw.mac.ops.init_params(&adapter->hw);
1581         if (rc)
1582                 return rc;
1583
1584         rc = adapter->hw.mbx.ops.init_params(&adapter->hw);
1585         if (rc)
1586                 return rc;
1587
1588         igbvf_set_interrupt_capability(adapter);
1589
1590         if (igbvf_alloc_queues(adapter))
1591                 return -ENOMEM;
1592
1593         spin_lock_init(&adapter->tx_queue_lock);
1594
1595         /* Explicitly disable IRQ since the NIC can be in any state. */
1596         igbvf_irq_disable(adapter);
1597
1598         spin_lock_init(&adapter->stats_lock);
1599
1600         set_bit(__IGBVF_DOWN, &adapter->state);
1601         return 0;
1602 }
1603
1604 static void igbvf_initialize_last_counter_stats(struct igbvf_adapter *adapter)
1605 {
1606         struct e1000_hw *hw = &adapter->hw;
1607
1608         adapter->stats.last_gprc = er32(VFGPRC);
1609         adapter->stats.last_gorc = er32(VFGORC);
1610         adapter->stats.last_gptc = er32(VFGPTC);
1611         adapter->stats.last_gotc = er32(VFGOTC);
1612         adapter->stats.last_mprc = er32(VFMPRC);
1613         adapter->stats.last_gotlbc = er32(VFGOTLBC);
1614         adapter->stats.last_gptlbc = er32(VFGPTLBC);
1615         adapter->stats.last_gorlbc = er32(VFGORLBC);
1616         adapter->stats.last_gprlbc = er32(VFGPRLBC);
1617
1618         adapter->stats.base_gprc = er32(VFGPRC);
1619         adapter->stats.base_gorc = er32(VFGORC);
1620         adapter->stats.base_gptc = er32(VFGPTC);
1621         adapter->stats.base_gotc = er32(VFGOTC);
1622         adapter->stats.base_mprc = er32(VFMPRC);
1623         adapter->stats.base_gotlbc = er32(VFGOTLBC);
1624         adapter->stats.base_gptlbc = er32(VFGPTLBC);
1625         adapter->stats.base_gorlbc = er32(VFGORLBC);
1626         adapter->stats.base_gprlbc = er32(VFGPRLBC);
1627 }
1628
1629 /**
1630  * igbvf_open - Called when a network interface is made active
1631  * @netdev: network interface device structure
1632  *
1633  * Returns 0 on success, negative value on failure
1634  *
1635  * The open entry point is called when a network interface is made
1636  * active by the system (IFF_UP).  At this point all resources needed
1637  * for transmit and receive operations are allocated, the interrupt
1638  * handler is registered with the OS, the watchdog timer is started,
1639  * and the stack is notified that the interface is ready.
1640  **/
1641 static int igbvf_open(struct net_device *netdev)
1642 {
1643         struct igbvf_adapter *adapter = netdev_priv(netdev);
1644         struct e1000_hw *hw = &adapter->hw;
1645         int err;
1646
1647         /* disallow open during test */
1648         if (test_bit(__IGBVF_TESTING, &adapter->state))
1649                 return -EBUSY;
1650
1651         /* allocate transmit descriptors */
1652         err = igbvf_setup_tx_resources(adapter, adapter->tx_ring);
1653         if (err)
1654                 goto err_setup_tx;
1655
1656         /* allocate receive descriptors */
1657         err = igbvf_setup_rx_resources(adapter, adapter->rx_ring);
1658         if (err)
1659                 goto err_setup_rx;
1660
1661         /*
1662          * before we allocate an interrupt, we must be ready to handle it.
1663          * Setting DEBUG_SHIRQ in the kernel makes it fire an interrupt
1664          * as soon as we call pci_request_irq, so we have to setup our
1665          * clean_rx handler before we do so.
1666          */
1667         igbvf_configure(adapter);
1668
1669         err = igbvf_request_irq(adapter);
1670         if (err)
1671                 goto err_req_irq;
1672
1673         /* From here on the code is the same as igbvf_up() */
1674         clear_bit(__IGBVF_DOWN, &adapter->state);
1675
1676         napi_enable(&adapter->rx_ring->napi);
1677
1678         /* clear any pending interrupts */
1679         er32(EICR);
1680
1681         igbvf_irq_enable(adapter);
1682
1683         /* start the watchdog */
1684         hw->mac.get_link_status = 1;
1685         mod_timer(&adapter->watchdog_timer, jiffies + 1);
1686
1687         return 0;
1688
1689 err_req_irq:
1690         igbvf_free_rx_resources(adapter->rx_ring);
1691 err_setup_rx:
1692         igbvf_free_tx_resources(adapter->tx_ring);
1693 err_setup_tx:
1694         igbvf_reset(adapter);
1695
1696         return err;
1697 }
1698
1699 /**
1700  * igbvf_close - Disables a network interface
1701  * @netdev: network interface device structure
1702  *
1703  * Returns 0, this is not allowed to fail
1704  *
1705  * The close entry point is called when an interface is de-activated
1706  * by the OS.  The hardware is still under the drivers control, but
1707  * needs to be disabled.  A global MAC reset is issued to stop the
1708  * hardware, and all transmit and receive resources are freed.
1709  **/
1710 static int igbvf_close(struct net_device *netdev)
1711 {
1712         struct igbvf_adapter *adapter = netdev_priv(netdev);
1713
1714         WARN_ON(test_bit(__IGBVF_RESETTING, &adapter->state));
1715         igbvf_down(adapter);
1716
1717         igbvf_free_irq(adapter);
1718
1719         igbvf_free_tx_resources(adapter->tx_ring);
1720         igbvf_free_rx_resources(adapter->rx_ring);
1721
1722         return 0;
1723 }
1724 /**
1725  * igbvf_set_mac - Change the Ethernet Address of the NIC
1726  * @netdev: network interface device structure
1727  * @p: pointer to an address structure
1728  *
1729  * Returns 0 on success, negative on failure
1730  **/
1731 static int igbvf_set_mac(struct net_device *netdev, void *p)
1732 {
1733         struct igbvf_adapter *adapter = netdev_priv(netdev);
1734         struct e1000_hw *hw = &adapter->hw;
1735         struct sockaddr *addr = p;
1736
1737         if (!is_valid_ether_addr(addr->sa_data))
1738                 return -EADDRNOTAVAIL;
1739
1740         memcpy(hw->mac.addr, addr->sa_data, netdev->addr_len);
1741
1742         hw->mac.ops.rar_set(hw, hw->mac.addr, 0);
1743
1744         if (memcmp(addr->sa_data, hw->mac.addr, 6))
1745                 return -EADDRNOTAVAIL;
1746
1747         memcpy(netdev->dev_addr, addr->sa_data, netdev->addr_len);
1748
1749         return 0;
1750 }
1751
1752 #define UPDATE_VF_COUNTER(reg, name)                                    \
1753         {                                                               \
1754                 u32 current_counter = er32(reg);                        \
1755                 if (current_counter < adapter->stats.last_##name)       \
1756                         adapter->stats.name += 0x100000000LL;           \
1757                 adapter->stats.last_##name = current_counter;           \
1758                 adapter->stats.name &= 0xFFFFFFFF00000000LL;            \
1759                 adapter->stats.name |= current_counter;                 \
1760         }
1761
1762 /**
1763  * igbvf_update_stats - Update the board statistics counters
1764  * @adapter: board private structure
1765 **/
1766 void igbvf_update_stats(struct igbvf_adapter *adapter)
1767 {
1768         struct e1000_hw *hw = &adapter->hw;
1769         struct pci_dev *pdev = adapter->pdev;
1770
1771         /*
1772          * Prevent stats update while adapter is being reset, link is down
1773          * or if the pci connection is down.
1774          */
1775         if (adapter->link_speed == 0)
1776                 return;
1777
1778         if (test_bit(__IGBVF_RESETTING, &adapter->state))
1779                 return;
1780
1781         if (pci_channel_offline(pdev))
1782                 return;
1783
1784         UPDATE_VF_COUNTER(VFGPRC, gprc);
1785         UPDATE_VF_COUNTER(VFGORC, gorc);
1786         UPDATE_VF_COUNTER(VFGPTC, gptc);
1787         UPDATE_VF_COUNTER(VFGOTC, gotc);
1788         UPDATE_VF_COUNTER(VFMPRC, mprc);
1789         UPDATE_VF_COUNTER(VFGOTLBC, gotlbc);
1790         UPDATE_VF_COUNTER(VFGPTLBC, gptlbc);
1791         UPDATE_VF_COUNTER(VFGORLBC, gorlbc);
1792         UPDATE_VF_COUNTER(VFGPRLBC, gprlbc);
1793
1794         /* Fill out the OS statistics structure */
1795         adapter->net_stats.multicast = adapter->stats.mprc;
1796 }
1797
1798 static void igbvf_print_link_info(struct igbvf_adapter *adapter)
1799 {
1800         dev_info(&adapter->pdev->dev, "Link is Up %d Mbps %s\n",
1801                  adapter->link_speed,
1802                  ((adapter->link_duplex == FULL_DUPLEX) ?
1803                   "Full Duplex" : "Half Duplex"));
1804 }
1805
1806 static bool igbvf_has_link(struct igbvf_adapter *adapter)
1807 {
1808         struct e1000_hw *hw = &adapter->hw;
1809         s32 ret_val = E1000_SUCCESS;
1810         bool link_active;
1811
1812         /* If interface is down, stay link down */
1813         if (test_bit(__IGBVF_DOWN, &adapter->state))
1814                 return false;
1815
1816         ret_val = hw->mac.ops.check_for_link(hw);
1817         link_active = !hw->mac.get_link_status;
1818
1819         /* if check for link returns error we will need to reset */
1820         if (ret_val && time_after(jiffies, adapter->last_reset + (10 * HZ)))
1821                 schedule_work(&adapter->reset_task);
1822
1823         return link_active;
1824 }
1825
1826 /**
1827  * igbvf_watchdog - Timer Call-back
1828  * @data: pointer to adapter cast into an unsigned long
1829  **/
1830 static void igbvf_watchdog(unsigned long data)
1831 {
1832         struct igbvf_adapter *adapter = (struct igbvf_adapter *) data;
1833
1834         /* Do the rest outside of interrupt context */
1835         schedule_work(&adapter->watchdog_task);
1836 }
1837
1838 static void igbvf_watchdog_task(struct work_struct *work)
1839 {
1840         struct igbvf_adapter *adapter = container_of(work,
1841                                                      struct igbvf_adapter,
1842                                                      watchdog_task);
1843         struct net_device *netdev = adapter->netdev;
1844         struct e1000_mac_info *mac = &adapter->hw.mac;
1845         struct igbvf_ring *tx_ring = adapter->tx_ring;
1846         struct e1000_hw *hw = &adapter->hw;
1847         u32 link;
1848         int tx_pending = 0;
1849
1850         link = igbvf_has_link(adapter);
1851
1852         if (link) {
1853                 if (!netif_carrier_ok(netdev)) {
1854                         bool txb2b = 1;
1855
1856                         mac->ops.get_link_up_info(&adapter->hw,
1857                                                   &adapter->link_speed,
1858                                                   &adapter->link_duplex);
1859                         igbvf_print_link_info(adapter);
1860
1861                         /* adjust timeout factor according to speed/duplex */
1862                         adapter->tx_timeout_factor = 1;
1863                         switch (adapter->link_speed) {
1864                         case SPEED_10:
1865                                 txb2b = 0;
1866                                 adapter->tx_timeout_factor = 16;
1867                                 break;
1868                         case SPEED_100:
1869                                 txb2b = 0;
1870                                 /* maybe add some timeout factor ? */
1871                                 break;
1872                         }
1873
1874                         netif_carrier_on(netdev);
1875                         netif_wake_queue(netdev);
1876                 }
1877         } else {
1878                 if (netif_carrier_ok(netdev)) {
1879                         adapter->link_speed = 0;
1880                         adapter->link_duplex = 0;
1881                         dev_info(&adapter->pdev->dev, "Link is Down\n");
1882                         netif_carrier_off(netdev);
1883                         netif_stop_queue(netdev);
1884                 }
1885         }
1886
1887         if (netif_carrier_ok(netdev)) {
1888                 igbvf_update_stats(adapter);
1889         } else {
1890                 tx_pending = (igbvf_desc_unused(tx_ring) + 1 <
1891                               tx_ring->count);
1892                 if (tx_pending) {
1893                         /*
1894                          * We've lost link, so the controller stops DMA,
1895                          * but we've got queued Tx work that's never going
1896                          * to get done, so reset controller to flush Tx.
1897                          * (Do the reset outside of interrupt context).
1898                          */
1899                         adapter->tx_timeout_count++;
1900                         schedule_work(&adapter->reset_task);
1901                 }
1902         }
1903
1904         /* Cause software interrupt to ensure Rx ring is cleaned */
1905         ew32(EICS, adapter->rx_ring->eims_value);
1906
1907         /* Force detection of hung controller every watchdog period */
1908         adapter->detect_tx_hung = 1;
1909
1910         /* Reset the timer */
1911         if (!test_bit(__IGBVF_DOWN, &adapter->state))
1912                 mod_timer(&adapter->watchdog_timer,
1913                           round_jiffies(jiffies + (2 * HZ)));
1914 }
1915
1916 #define IGBVF_TX_FLAGS_CSUM             0x00000001
1917 #define IGBVF_TX_FLAGS_VLAN             0x00000002
1918 #define IGBVF_TX_FLAGS_TSO              0x00000004
1919 #define IGBVF_TX_FLAGS_IPV4             0x00000008
1920 #define IGBVF_TX_FLAGS_VLAN_MASK        0xffff0000
1921 #define IGBVF_TX_FLAGS_VLAN_SHIFT       16
1922
1923 static int igbvf_tso(struct igbvf_adapter *adapter,
1924                      struct igbvf_ring *tx_ring,
1925                      struct sk_buff *skb, u32 tx_flags, u8 *hdr_len)
1926 {
1927         struct e1000_adv_tx_context_desc *context_desc;
1928         unsigned int i;
1929         int err;
1930         struct igbvf_buffer *buffer_info;
1931         u32 info = 0, tu_cmd = 0;
1932         u32 mss_l4len_idx, l4len;
1933         *hdr_len = 0;
1934
1935         if (skb_header_cloned(skb)) {
1936                 err = pskb_expand_head(skb, 0, 0, GFP_ATOMIC);
1937                 if (err) {
1938                         dev_err(&adapter->pdev->dev,
1939                                 "igbvf_tso returning an error\n");
1940                         return err;
1941                 }
1942         }
1943
1944         l4len = tcp_hdrlen(skb);
1945         *hdr_len += l4len;
1946
1947         if (skb->protocol == htons(ETH_P_IP)) {
1948                 struct iphdr *iph = ip_hdr(skb);
1949                 iph->tot_len = 0;
1950                 iph->check = 0;
1951                 tcp_hdr(skb)->check = ~csum_tcpudp_magic(iph->saddr,
1952                                                          iph->daddr, 0,
1953                                                          IPPROTO_TCP,
1954                                                          0);
1955         } else if (skb_is_gso_v6(skb)) {
1956                 ipv6_hdr(skb)->payload_len = 0;
1957                 tcp_hdr(skb)->check = ~csum_ipv6_magic(&ipv6_hdr(skb)->saddr,
1958                                                        &ipv6_hdr(skb)->daddr,
1959                                                        0, IPPROTO_TCP, 0);
1960         }
1961
1962         i = tx_ring->next_to_use;
1963
1964         buffer_info = &tx_ring->buffer_info[i];
1965         context_desc = IGBVF_TX_CTXTDESC_ADV(*tx_ring, i);
1966         /* VLAN MACLEN IPLEN */
1967         if (tx_flags & IGBVF_TX_FLAGS_VLAN)
1968                 info |= (tx_flags & IGBVF_TX_FLAGS_VLAN_MASK);
1969         info |= (skb_network_offset(skb) << E1000_ADVTXD_MACLEN_SHIFT);
1970         *hdr_len += skb_network_offset(skb);
1971         info |= (skb_transport_header(skb) - skb_network_header(skb));
1972         *hdr_len += (skb_transport_header(skb) - skb_network_header(skb));
1973         context_desc->vlan_macip_lens = cpu_to_le32(info);
1974
1975         /* ADV DTYP TUCMD MKRLOC/ISCSIHEDLEN */
1976         tu_cmd |= (E1000_TXD_CMD_DEXT | E1000_ADVTXD_DTYP_CTXT);
1977
1978         if (skb->protocol == htons(ETH_P_IP))
1979                 tu_cmd |= E1000_ADVTXD_TUCMD_IPV4;
1980         tu_cmd |= E1000_ADVTXD_TUCMD_L4T_TCP;
1981
1982         context_desc->type_tucmd_mlhl = cpu_to_le32(tu_cmd);
1983
1984         /* MSS L4LEN IDX */
1985         mss_l4len_idx = (skb_shinfo(skb)->gso_size << E1000_ADVTXD_MSS_SHIFT);
1986         mss_l4len_idx |= (l4len << E1000_ADVTXD_L4LEN_SHIFT);
1987
1988         context_desc->mss_l4len_idx = cpu_to_le32(mss_l4len_idx);
1989         context_desc->seqnum_seed = 0;
1990
1991         buffer_info->time_stamp = jiffies;
1992         buffer_info->next_to_watch = i;
1993         buffer_info->dma = 0;
1994         i++;
1995         if (i == tx_ring->count)
1996                 i = 0;
1997
1998         tx_ring->next_to_use = i;
1999
2000         return true;
2001 }
2002
2003 static inline bool igbvf_tx_csum(struct igbvf_adapter *adapter,
2004                                  struct igbvf_ring *tx_ring,
2005                                  struct sk_buff *skb, u32 tx_flags)
2006 {
2007         struct e1000_adv_tx_context_desc *context_desc;
2008         unsigned int i;
2009         struct igbvf_buffer *buffer_info;
2010         u32 info = 0, tu_cmd = 0;
2011
2012         if ((skb->ip_summed == CHECKSUM_PARTIAL) ||
2013             (tx_flags & IGBVF_TX_FLAGS_VLAN)) {
2014                 i = tx_ring->next_to_use;
2015                 buffer_info = &tx_ring->buffer_info[i];
2016                 context_desc = IGBVF_TX_CTXTDESC_ADV(*tx_ring, i);
2017
2018                 if (tx_flags & IGBVF_TX_FLAGS_VLAN)
2019                         info |= (tx_flags & IGBVF_TX_FLAGS_VLAN_MASK);
2020
2021                 info |= (skb_network_offset(skb) << E1000_ADVTXD_MACLEN_SHIFT);
2022                 if (skb->ip_summed == CHECKSUM_PARTIAL)
2023                         info |= (skb_transport_header(skb) -
2024                                  skb_network_header(skb));
2025
2026
2027                 context_desc->vlan_macip_lens = cpu_to_le32(info);
2028
2029                 tu_cmd |= (E1000_TXD_CMD_DEXT | E1000_ADVTXD_DTYP_CTXT);
2030
2031                 if (skb->ip_summed == CHECKSUM_PARTIAL) {
2032                         switch (skb->protocol) {
2033                         case __constant_htons(ETH_P_IP):
2034                                 tu_cmd |= E1000_ADVTXD_TUCMD_IPV4;
2035                                 if (ip_hdr(skb)->protocol == IPPROTO_TCP)
2036                                         tu_cmd |= E1000_ADVTXD_TUCMD_L4T_TCP;
2037                                 break;
2038                         case __constant_htons(ETH_P_IPV6):
2039                                 if (ipv6_hdr(skb)->nexthdr == IPPROTO_TCP)
2040                                         tu_cmd |= E1000_ADVTXD_TUCMD_L4T_TCP;
2041                                 break;
2042                         default:
2043                                 break;
2044                         }
2045                 }
2046
2047                 context_desc->type_tucmd_mlhl = cpu_to_le32(tu_cmd);
2048                 context_desc->seqnum_seed = 0;
2049                 context_desc->mss_l4len_idx = 0;
2050
2051                 buffer_info->time_stamp = jiffies;
2052                 buffer_info->next_to_watch = i;
2053                 buffer_info->dma = 0;
2054                 i++;
2055                 if (i == tx_ring->count)
2056                         i = 0;
2057                 tx_ring->next_to_use = i;
2058
2059                 return true;
2060         }
2061
2062         return false;
2063 }
2064
2065 static int igbvf_maybe_stop_tx(struct net_device *netdev, int size)
2066 {
2067         struct igbvf_adapter *adapter = netdev_priv(netdev);
2068
2069         /* there is enough descriptors then we don't need to worry  */
2070         if (igbvf_desc_unused(adapter->tx_ring) >= size)
2071                 return 0;
2072
2073         netif_stop_queue(netdev);
2074
2075         smp_mb();
2076
2077         /* We need to check again just in case room has been made available */
2078         if (igbvf_desc_unused(adapter->tx_ring) < size)
2079                 return -EBUSY;
2080
2081         netif_wake_queue(netdev);
2082
2083         ++adapter->restart_queue;
2084         return 0;
2085 }
2086
2087 #define IGBVF_MAX_TXD_PWR       16
2088 #define IGBVF_MAX_DATA_PER_TXD  (1 << IGBVF_MAX_TXD_PWR)
2089
2090 static inline int igbvf_tx_map_adv(struct igbvf_adapter *adapter,
2091                                    struct igbvf_ring *tx_ring,
2092                                    struct sk_buff *skb,
2093                                    unsigned int first)
2094 {
2095         struct igbvf_buffer *buffer_info;
2096         struct pci_dev *pdev = adapter->pdev;
2097         unsigned int len = skb_headlen(skb);
2098         unsigned int count = 0, i;
2099         unsigned int f;
2100
2101         i = tx_ring->next_to_use;
2102
2103         buffer_info = &tx_ring->buffer_info[i];
2104         BUG_ON(len >= IGBVF_MAX_DATA_PER_TXD);
2105         buffer_info->length = len;
2106         /* set time_stamp *before* dma to help avoid a possible race */
2107         buffer_info->time_stamp = jiffies;
2108         buffer_info->next_to_watch = i;
2109         buffer_info->mapped_as_page = false;
2110         buffer_info->dma = dma_map_single(&pdev->dev, skb->data, len,
2111                                           DMA_TO_DEVICE);
2112         if (dma_mapping_error(&pdev->dev, buffer_info->dma))
2113                 goto dma_error;
2114
2115
2116         for (f = 0; f < skb_shinfo(skb)->nr_frags; f++) {
2117                 struct skb_frag_struct *frag;
2118
2119                 count++;
2120                 i++;
2121                 if (i == tx_ring->count)
2122                         i = 0;
2123
2124                 frag = &skb_shinfo(skb)->frags[f];
2125                 len = frag->size;
2126
2127                 buffer_info = &tx_ring->buffer_info[i];
2128                 BUG_ON(len >= IGBVF_MAX_DATA_PER_TXD);
2129                 buffer_info->length = len;
2130                 buffer_info->time_stamp = jiffies;
2131                 buffer_info->next_to_watch = i;
2132                 buffer_info->mapped_as_page = true;
2133                 buffer_info->dma = dma_map_page(&pdev->dev,
2134                                                 frag->page,
2135                                                 frag->page_offset,
2136                                                 len,
2137                                                 DMA_TO_DEVICE);
2138                 if (dma_mapping_error(&pdev->dev, buffer_info->dma))
2139                         goto dma_error;
2140         }
2141
2142         tx_ring->buffer_info[i].skb = skb;
2143         tx_ring->buffer_info[first].next_to_watch = i;
2144
2145         return ++count;
2146
2147 dma_error:
2148         dev_err(&pdev->dev, "TX DMA map failed\n");
2149
2150         /* clear timestamp and dma mappings for failed buffer_info mapping */
2151         buffer_info->dma = 0;
2152         buffer_info->time_stamp = 0;
2153         buffer_info->length = 0;
2154         buffer_info->next_to_watch = 0;
2155         buffer_info->mapped_as_page = false;
2156         if (count)
2157                 count--;
2158
2159         /* clear timestamp and dma mappings for remaining portion of packet */
2160         while (count--) {
2161                 if (i==0)
2162                         i += tx_ring->count;
2163                 i--;
2164                 buffer_info = &tx_ring->buffer_info[i];
2165                 igbvf_put_txbuf(adapter, buffer_info);
2166         }
2167
2168         return 0;
2169 }
2170
2171 static inline void igbvf_tx_queue_adv(struct igbvf_adapter *adapter,
2172                                       struct igbvf_ring *tx_ring,
2173                                       int tx_flags, int count, u32 paylen,
2174                                       u8 hdr_len)
2175 {
2176         union e1000_adv_tx_desc *tx_desc = NULL;
2177         struct igbvf_buffer *buffer_info;
2178         u32 olinfo_status = 0, cmd_type_len;
2179         unsigned int i;
2180
2181         cmd_type_len = (E1000_ADVTXD_DTYP_DATA | E1000_ADVTXD_DCMD_IFCS |
2182                         E1000_ADVTXD_DCMD_DEXT);
2183
2184         if (tx_flags & IGBVF_TX_FLAGS_VLAN)
2185                 cmd_type_len |= E1000_ADVTXD_DCMD_VLE;
2186
2187         if (tx_flags & IGBVF_TX_FLAGS_TSO) {
2188                 cmd_type_len |= E1000_ADVTXD_DCMD_TSE;
2189
2190                 /* insert tcp checksum */
2191                 olinfo_status |= E1000_TXD_POPTS_TXSM << 8;
2192
2193                 /* insert ip checksum */
2194                 if (tx_flags & IGBVF_TX_FLAGS_IPV4)
2195                         olinfo_status |= E1000_TXD_POPTS_IXSM << 8;
2196
2197         } else if (tx_flags & IGBVF_TX_FLAGS_CSUM) {
2198                 olinfo_status |= E1000_TXD_POPTS_TXSM << 8;
2199         }
2200
2201         olinfo_status |= ((paylen - hdr_len) << E1000_ADVTXD_PAYLEN_SHIFT);
2202
2203         i = tx_ring->next_to_use;
2204         while (count--) {
2205                 buffer_info = &tx_ring->buffer_info[i];
2206                 tx_desc = IGBVF_TX_DESC_ADV(*tx_ring, i);
2207                 tx_desc->read.buffer_addr = cpu_to_le64(buffer_info->dma);
2208                 tx_desc->read.cmd_type_len =
2209                          cpu_to_le32(cmd_type_len | buffer_info->length);
2210                 tx_desc->read.olinfo_status = cpu_to_le32(olinfo_status);
2211                 i++;
2212                 if (i == tx_ring->count)
2213                         i = 0;
2214         }
2215
2216         tx_desc->read.cmd_type_len |= cpu_to_le32(adapter->txd_cmd);
2217         /* Force memory writes to complete before letting h/w
2218          * know there are new descriptors to fetch.  (Only
2219          * applicable for weak-ordered memory model archs,
2220          * such as IA-64). */
2221         wmb();
2222
2223         tx_ring->next_to_use = i;
2224         writel(i, adapter->hw.hw_addr + tx_ring->tail);
2225         /* we need this if more than one processor can write to our tail
2226          * at a time, it syncronizes IO on IA64/Altix systems */
2227         mmiowb();
2228 }
2229
2230 static netdev_tx_t igbvf_xmit_frame_ring_adv(struct sk_buff *skb,
2231                                              struct net_device *netdev,
2232                                              struct igbvf_ring *tx_ring)
2233 {
2234         struct igbvf_adapter *adapter = netdev_priv(netdev);
2235         unsigned int first, tx_flags = 0;
2236         u8 hdr_len = 0;
2237         int count = 0;
2238         int tso = 0;
2239
2240         if (test_bit(__IGBVF_DOWN, &adapter->state)) {
2241                 dev_kfree_skb_any(skb);
2242                 return NETDEV_TX_OK;
2243         }
2244
2245         if (skb->len <= 0) {
2246                 dev_kfree_skb_any(skb);
2247                 return NETDEV_TX_OK;
2248         }
2249
2250         /*
2251          * need: count + 4 desc gap to keep tail from touching
2252          *       + 2 desc gap to keep tail from touching head,
2253          *       + 1 desc for skb->data,
2254          *       + 1 desc for context descriptor,
2255          * head, otherwise try next time
2256          */
2257         if (igbvf_maybe_stop_tx(netdev, skb_shinfo(skb)->nr_frags + 4)) {
2258                 /* this is a hard error */
2259                 return NETDEV_TX_BUSY;
2260         }
2261
2262         if (adapter->vlgrp && vlan_tx_tag_present(skb)) {
2263                 tx_flags |= IGBVF_TX_FLAGS_VLAN;
2264                 tx_flags |= (vlan_tx_tag_get(skb) << IGBVF_TX_FLAGS_VLAN_SHIFT);
2265         }
2266
2267         if (skb->protocol == htons(ETH_P_IP))
2268                 tx_flags |= IGBVF_TX_FLAGS_IPV4;
2269
2270         first = tx_ring->next_to_use;
2271
2272         tso = skb_is_gso(skb) ?
2273                 igbvf_tso(adapter, tx_ring, skb, tx_flags, &hdr_len) : 0;
2274         if (unlikely(tso < 0)) {
2275                 dev_kfree_skb_any(skb);
2276                 return NETDEV_TX_OK;
2277         }
2278
2279         if (tso)
2280                 tx_flags |= IGBVF_TX_FLAGS_TSO;
2281         else if (igbvf_tx_csum(adapter, tx_ring, skb, tx_flags) &&
2282                  (skb->ip_summed == CHECKSUM_PARTIAL))
2283                 tx_flags |= IGBVF_TX_FLAGS_CSUM;
2284
2285         /*
2286          * count reflects descriptors mapped, if 0 then mapping error
2287          * has occured and we need to rewind the descriptor queue
2288          */
2289         count = igbvf_tx_map_adv(adapter, tx_ring, skb, first);
2290
2291         if (count) {
2292                 igbvf_tx_queue_adv(adapter, tx_ring, tx_flags, count,
2293                                    skb->len, hdr_len);
2294                 /* Make sure there is space in the ring for the next send. */
2295                 igbvf_maybe_stop_tx(netdev, MAX_SKB_FRAGS + 4);
2296         } else {
2297                 dev_kfree_skb_any(skb);
2298                 tx_ring->buffer_info[first].time_stamp = 0;
2299                 tx_ring->next_to_use = first;
2300         }
2301
2302         return NETDEV_TX_OK;
2303 }
2304
2305 static netdev_tx_t igbvf_xmit_frame(struct sk_buff *skb,
2306                                     struct net_device *netdev)
2307 {
2308         struct igbvf_adapter *adapter = netdev_priv(netdev);
2309         struct igbvf_ring *tx_ring;
2310
2311         if (test_bit(__IGBVF_DOWN, &adapter->state)) {
2312                 dev_kfree_skb_any(skb);
2313                 return NETDEV_TX_OK;
2314         }
2315
2316         tx_ring = &adapter->tx_ring[0];
2317
2318         return igbvf_xmit_frame_ring_adv(skb, netdev, tx_ring);
2319 }
2320
2321 /**
2322  * igbvf_tx_timeout - Respond to a Tx Hang
2323  * @netdev: network interface device structure
2324  **/
2325 static void igbvf_tx_timeout(struct net_device *netdev)
2326 {
2327         struct igbvf_adapter *adapter = netdev_priv(netdev);
2328
2329         /* Do the reset outside of interrupt context */
2330         adapter->tx_timeout_count++;
2331         schedule_work(&adapter->reset_task);
2332 }
2333
2334 static void igbvf_reset_task(struct work_struct *work)
2335 {
2336         struct igbvf_adapter *adapter;
2337         adapter = container_of(work, struct igbvf_adapter, reset_task);
2338
2339         igbvf_reinit_locked(adapter);
2340 }
2341
2342 /**
2343  * igbvf_get_stats - Get System Network Statistics
2344  * @netdev: network interface device structure
2345  *
2346  * Returns the address of the device statistics structure.
2347  * The statistics are actually updated from the timer callback.
2348  **/
2349 static struct net_device_stats *igbvf_get_stats(struct net_device *netdev)
2350 {
2351         struct igbvf_adapter *adapter = netdev_priv(netdev);
2352
2353         /* only return the current stats */
2354         return &adapter->net_stats;
2355 }
2356
2357 /**
2358  * igbvf_change_mtu - Change the Maximum Transfer Unit
2359  * @netdev: network interface device structure
2360  * @new_mtu: new value for maximum frame size
2361  *
2362  * Returns 0 on success, negative on failure
2363  **/
2364 static int igbvf_change_mtu(struct net_device *netdev, int new_mtu)
2365 {
2366         struct igbvf_adapter *adapter = netdev_priv(netdev);
2367         int max_frame = new_mtu + ETH_HLEN + ETH_FCS_LEN;
2368
2369         if ((new_mtu < 68) || (max_frame > MAX_JUMBO_FRAME_SIZE)) {
2370                 dev_err(&adapter->pdev->dev, "Invalid MTU setting\n");
2371                 return -EINVAL;
2372         }
2373
2374 #define MAX_STD_JUMBO_FRAME_SIZE 9234
2375         if (max_frame > MAX_STD_JUMBO_FRAME_SIZE) {
2376                 dev_err(&adapter->pdev->dev, "MTU > 9216 not supported.\n");
2377                 return -EINVAL;
2378         }
2379
2380         while (test_and_set_bit(__IGBVF_RESETTING, &adapter->state))
2381                 msleep(1);
2382         /* igbvf_down has a dependency on max_frame_size */
2383         adapter->max_frame_size = max_frame;
2384         if (netif_running(netdev))
2385                 igbvf_down(adapter);
2386
2387         /*
2388          * NOTE: netdev_alloc_skb reserves 16 bytes, and typically NET_IP_ALIGN
2389          * means we reserve 2 more, this pushes us to allocate from the next
2390          * larger slab size.
2391          * i.e. RXBUFFER_2048 --> size-4096 slab
2392          * However with the new *_jumbo_rx* routines, jumbo receives will use
2393          * fragmented skbs
2394          */
2395
2396         if (max_frame <= 1024)
2397                 adapter->rx_buffer_len = 1024;
2398         else if (max_frame <= 2048)
2399                 adapter->rx_buffer_len = 2048;
2400         else
2401 #if (PAGE_SIZE / 2) > 16384
2402                 adapter->rx_buffer_len = 16384;
2403 #else
2404                 adapter->rx_buffer_len = PAGE_SIZE / 2;
2405 #endif
2406
2407
2408         /* adjust allocation if LPE protects us, and we aren't using SBP */
2409         if ((max_frame == ETH_FRAME_LEN + ETH_FCS_LEN) ||
2410              (max_frame == ETH_FRAME_LEN + VLAN_HLEN + ETH_FCS_LEN))
2411                 adapter->rx_buffer_len = ETH_FRAME_LEN + VLAN_HLEN +
2412                                          ETH_FCS_LEN;
2413
2414         dev_info(&adapter->pdev->dev, "changing MTU from %d to %d\n",
2415                  netdev->mtu, new_mtu);
2416         netdev->mtu = new_mtu;
2417
2418         if (netif_running(netdev))
2419                 igbvf_up(adapter);
2420         else
2421                 igbvf_reset(adapter);
2422
2423         clear_bit(__IGBVF_RESETTING, &adapter->state);
2424
2425         return 0;
2426 }
2427
2428 static int igbvf_ioctl(struct net_device *netdev, struct ifreq *ifr, int cmd)
2429 {
2430         switch (cmd) {
2431         default:
2432                 return -EOPNOTSUPP;
2433         }
2434 }
2435
2436 static int igbvf_suspend(struct pci_dev *pdev, pm_message_t state)
2437 {
2438         struct net_device *netdev = pci_get_drvdata(pdev);
2439         struct igbvf_adapter *adapter = netdev_priv(netdev);
2440 #ifdef CONFIG_PM
2441         int retval = 0;
2442 #endif
2443
2444         netif_device_detach(netdev);
2445
2446         if (netif_running(netdev)) {
2447                 WARN_ON(test_bit(__IGBVF_RESETTING, &adapter->state));
2448                 igbvf_down(adapter);
2449                 igbvf_free_irq(adapter);
2450         }
2451
2452 #ifdef CONFIG_PM
2453         retval = pci_save_state(pdev);
2454         if (retval)
2455                 return retval;
2456 #endif
2457
2458         pci_disable_device(pdev);
2459
2460         return 0;
2461 }
2462
2463 #ifdef CONFIG_PM
2464 static int igbvf_resume(struct pci_dev *pdev)
2465 {
2466         struct net_device *netdev = pci_get_drvdata(pdev);
2467         struct igbvf_adapter *adapter = netdev_priv(netdev);
2468         u32 err;
2469
2470         pci_restore_state(pdev);
2471         err = pci_enable_device_mem(pdev);
2472         if (err) {
2473                 dev_err(&pdev->dev, "Cannot enable PCI device from suspend\n");
2474                 return err;
2475         }
2476
2477         pci_set_master(pdev);
2478
2479         if (netif_running(netdev)) {
2480                 err = igbvf_request_irq(adapter);
2481                 if (err)
2482                         return err;
2483         }
2484
2485         igbvf_reset(adapter);
2486
2487         if (netif_running(netdev))
2488                 igbvf_up(adapter);
2489
2490         netif_device_attach(netdev);
2491
2492         return 0;
2493 }
2494 #endif
2495
2496 static void igbvf_shutdown(struct pci_dev *pdev)
2497 {
2498         igbvf_suspend(pdev, PMSG_SUSPEND);
2499 }
2500
2501 #ifdef CONFIG_NET_POLL_CONTROLLER
2502 /*
2503  * Polling 'interrupt' - used by things like netconsole to send skbs
2504  * without having to re-enable interrupts. It's not called while
2505  * the interrupt routine is executing.
2506  */
2507 static void igbvf_netpoll(struct net_device *netdev)
2508 {
2509         struct igbvf_adapter *adapter = netdev_priv(netdev);
2510
2511         disable_irq(adapter->pdev->irq);
2512
2513         igbvf_clean_tx_irq(adapter->tx_ring);
2514
2515         enable_irq(adapter->pdev->irq);
2516 }
2517 #endif
2518
2519 /**
2520  * igbvf_io_error_detected - called when PCI error is detected
2521  * @pdev: Pointer to PCI device
2522  * @state: The current pci connection state
2523  *
2524  * This function is called after a PCI bus error affecting
2525  * this device has been detected.
2526  */
2527 static pci_ers_result_t igbvf_io_error_detected(struct pci_dev *pdev,
2528                                                 pci_channel_state_t state)
2529 {
2530         struct net_device *netdev = pci_get_drvdata(pdev);
2531         struct igbvf_adapter *adapter = netdev_priv(netdev);
2532
2533         netif_device_detach(netdev);
2534
2535         if (state == pci_channel_io_perm_failure)
2536                 return PCI_ERS_RESULT_DISCONNECT;
2537
2538         if (netif_running(netdev))
2539                 igbvf_down(adapter);
2540         pci_disable_device(pdev);
2541
2542         /* Request a slot slot reset. */
2543         return PCI_ERS_RESULT_NEED_RESET;
2544 }
2545
2546 /**
2547  * igbvf_io_slot_reset - called after the pci bus has been reset.
2548  * @pdev: Pointer to PCI device
2549  *
2550  * Restart the card from scratch, as if from a cold-boot. Implementation
2551  * resembles the first-half of the igbvf_resume routine.
2552  */
2553 static pci_ers_result_t igbvf_io_slot_reset(struct pci_dev *pdev)
2554 {
2555         struct net_device *netdev = pci_get_drvdata(pdev);
2556         struct igbvf_adapter *adapter = netdev_priv(netdev);
2557
2558         if (pci_enable_device_mem(pdev)) {
2559                 dev_err(&pdev->dev,
2560                         "Cannot re-enable PCI device after reset.\n");
2561                 return PCI_ERS_RESULT_DISCONNECT;
2562         }
2563         pci_set_master(pdev);
2564
2565         igbvf_reset(adapter);
2566
2567         return PCI_ERS_RESULT_RECOVERED;
2568 }
2569
2570 /**
2571  * igbvf_io_resume - called when traffic can start flowing again.
2572  * @pdev: Pointer to PCI device
2573  *
2574  * This callback is called when the error recovery driver tells us that
2575  * its OK to resume normal operation. Implementation resembles the
2576  * second-half of the igbvf_resume routine.
2577  */
2578 static void igbvf_io_resume(struct pci_dev *pdev)
2579 {
2580         struct net_device *netdev = pci_get_drvdata(pdev);
2581         struct igbvf_adapter *adapter = netdev_priv(netdev);
2582
2583         if (netif_running(netdev)) {
2584                 if (igbvf_up(adapter)) {
2585                         dev_err(&pdev->dev,
2586                                 "can't bring device back up after reset\n");
2587                         return;
2588                 }
2589         }
2590
2591         netif_device_attach(netdev);
2592 }
2593
2594 static void igbvf_print_device_info(struct igbvf_adapter *adapter)
2595 {
2596         struct e1000_hw *hw = &adapter->hw;
2597         struct net_device *netdev = adapter->netdev;
2598         struct pci_dev *pdev = adapter->pdev;
2599
2600         dev_info(&pdev->dev, "Intel(R) 82576 Virtual Function\n");
2601         dev_info(&pdev->dev, "Address: %pM\n", netdev->dev_addr);
2602         dev_info(&pdev->dev, "MAC: %d\n", hw->mac.type);
2603 }
2604
2605 static const struct net_device_ops igbvf_netdev_ops = {
2606         .ndo_open                       = igbvf_open,
2607         .ndo_stop                       = igbvf_close,
2608         .ndo_start_xmit                 = igbvf_xmit_frame,
2609         .ndo_get_stats                  = igbvf_get_stats,
2610         .ndo_set_multicast_list         = igbvf_set_multi,
2611         .ndo_set_mac_address            = igbvf_set_mac,
2612         .ndo_change_mtu                 = igbvf_change_mtu,
2613         .ndo_do_ioctl                   = igbvf_ioctl,
2614         .ndo_tx_timeout                 = igbvf_tx_timeout,
2615         .ndo_vlan_rx_register           = igbvf_vlan_rx_register,
2616         .ndo_vlan_rx_add_vid            = igbvf_vlan_rx_add_vid,
2617         .ndo_vlan_rx_kill_vid           = igbvf_vlan_rx_kill_vid,
2618 #ifdef CONFIG_NET_POLL_CONTROLLER
2619         .ndo_poll_controller            = igbvf_netpoll,
2620 #endif
2621 };
2622
2623 /**
2624  * igbvf_probe - Device Initialization Routine
2625  * @pdev: PCI device information struct
2626  * @ent: entry in igbvf_pci_tbl
2627  *
2628  * Returns 0 on success, negative on failure
2629  *
2630  * igbvf_probe initializes an adapter identified by a pci_dev structure.
2631  * The OS initialization, configuring of the adapter private structure,
2632  * and a hardware reset occur.
2633  **/
2634 static int __devinit igbvf_probe(struct pci_dev *pdev,
2635                                  const struct pci_device_id *ent)
2636 {
2637         struct net_device *netdev;
2638         struct igbvf_adapter *adapter;
2639         struct e1000_hw *hw;
2640         const struct igbvf_info *ei = igbvf_info_tbl[ent->driver_data];
2641
2642         static int cards_found;
2643         int err, pci_using_dac;
2644
2645         err = pci_enable_device_mem(pdev);
2646         if (err)
2647                 return err;
2648
2649         pci_using_dac = 0;
2650         err = dma_set_mask(&pdev->dev, DMA_BIT_MASK(64));
2651         if (!err) {
2652                 err = dma_set_coherent_mask(&pdev->dev, DMA_BIT_MASK(64));
2653                 if (!err)
2654                         pci_using_dac = 1;
2655         } else {
2656                 err = dma_set_mask(&pdev->dev, DMA_BIT_MASK(32));
2657                 if (err) {
2658                         err = dma_set_coherent_mask(&pdev->dev,
2659                                                     DMA_BIT_MASK(32));
2660                         if (err) {
2661                                 dev_err(&pdev->dev, "No usable DMA "
2662                                         "configuration, aborting\n");
2663                                 goto err_dma;
2664                         }
2665                 }
2666         }
2667
2668         err = pci_request_regions(pdev, igbvf_driver_name);
2669         if (err)
2670                 goto err_pci_reg;
2671
2672         pci_set_master(pdev);
2673
2674         err = -ENOMEM;
2675         netdev = alloc_etherdev(sizeof(struct igbvf_adapter));
2676         if (!netdev)
2677                 goto err_alloc_etherdev;
2678
2679         SET_NETDEV_DEV(netdev, &pdev->dev);
2680
2681         pci_set_drvdata(pdev, netdev);
2682         adapter = netdev_priv(netdev);
2683         hw = &adapter->hw;
2684         adapter->netdev = netdev;
2685         adapter->pdev = pdev;
2686         adapter->ei = ei;
2687         adapter->pba = ei->pba;
2688         adapter->flags = ei->flags;
2689         adapter->hw.back = adapter;
2690         adapter->hw.mac.type = ei->mac;
2691         adapter->msg_enable = (1 << NETIF_MSG_DRV | NETIF_MSG_PROBE) - 1;
2692
2693         /* PCI config space info */
2694
2695         hw->vendor_id = pdev->vendor;
2696         hw->device_id = pdev->device;
2697         hw->subsystem_vendor_id = pdev->subsystem_vendor;
2698         hw->subsystem_device_id = pdev->subsystem_device;
2699
2700         pci_read_config_byte(pdev, PCI_REVISION_ID, &hw->revision_id);
2701
2702         err = -EIO;
2703         adapter->hw.hw_addr = ioremap(pci_resource_start(pdev, 0),
2704                                       pci_resource_len(pdev, 0));
2705
2706         if (!adapter->hw.hw_addr)
2707                 goto err_ioremap;
2708
2709         if (ei->get_variants) {
2710                 err = ei->get_variants(adapter);
2711                 if (err)
2712                         goto err_ioremap;
2713         }
2714
2715         /* setup adapter struct */
2716         err = igbvf_sw_init(adapter);
2717         if (err)
2718                 goto err_sw_init;
2719
2720         /* construct the net_device struct */
2721         netdev->netdev_ops = &igbvf_netdev_ops;
2722
2723         igbvf_set_ethtool_ops(netdev);
2724         netdev->watchdog_timeo = 5 * HZ;
2725         strncpy(netdev->name, pci_name(pdev), sizeof(netdev->name) - 1);
2726
2727         adapter->bd_number = cards_found++;
2728
2729         netdev->features = NETIF_F_SG |
2730                            NETIF_F_IP_CSUM |
2731                            NETIF_F_HW_VLAN_TX |
2732                            NETIF_F_HW_VLAN_RX |
2733                            NETIF_F_HW_VLAN_FILTER;
2734
2735         netdev->features |= NETIF_F_IPV6_CSUM;
2736         netdev->features |= NETIF_F_TSO;
2737         netdev->features |= NETIF_F_TSO6;
2738
2739         if (pci_using_dac)
2740                 netdev->features |= NETIF_F_HIGHDMA;
2741
2742         netdev->vlan_features |= NETIF_F_TSO;
2743         netdev->vlan_features |= NETIF_F_TSO6;
2744         netdev->vlan_features |= NETIF_F_IP_CSUM;
2745         netdev->vlan_features |= NETIF_F_IPV6_CSUM;
2746         netdev->vlan_features |= NETIF_F_SG;
2747
2748         /*reset the controller to put the device in a known good state */
2749         err = hw->mac.ops.reset_hw(hw);
2750         if (err) {
2751                 dev_info(&pdev->dev,
2752                          "PF still in reset state, assigning new address."
2753                          " Is the PF interface up?\n");
2754                 dev_hw_addr_random(adapter->netdev, hw->mac.addr);
2755         } else {
2756                 err = hw->mac.ops.read_mac_addr(hw);
2757                 if (err) {
2758                         dev_err(&pdev->dev, "Error reading MAC address\n");
2759                         goto err_hw_init;
2760                 }
2761         }
2762
2763         memcpy(netdev->dev_addr, adapter->hw.mac.addr, netdev->addr_len);
2764         memcpy(netdev->perm_addr, adapter->hw.mac.addr, netdev->addr_len);
2765
2766         if (!is_valid_ether_addr(netdev->perm_addr)) {
2767                 dev_err(&pdev->dev, "Invalid MAC Address: %pM\n",
2768                         netdev->dev_addr);
2769                 err = -EIO;
2770                 goto err_hw_init;
2771         }
2772
2773         setup_timer(&adapter->watchdog_timer, &igbvf_watchdog,
2774                     (unsigned long) adapter);
2775
2776         INIT_WORK(&adapter->reset_task, igbvf_reset_task);
2777         INIT_WORK(&adapter->watchdog_task, igbvf_watchdog_task);
2778
2779         /* ring size defaults */
2780         adapter->rx_ring->count = 1024;
2781         adapter->tx_ring->count = 1024;
2782
2783         /* reset the hardware with the new settings */
2784         igbvf_reset(adapter);
2785
2786         /* tell the stack to leave us alone until igbvf_open() is called */
2787         netif_carrier_off(netdev);
2788         netif_stop_queue(netdev);
2789
2790         strcpy(netdev->name, "eth%d");
2791         err = register_netdev(netdev);
2792         if (err)
2793                 goto err_hw_init;
2794
2795         igbvf_print_device_info(adapter);
2796
2797         igbvf_initialize_last_counter_stats(adapter);
2798
2799         return 0;
2800
2801 err_hw_init:
2802         kfree(adapter->tx_ring);
2803         kfree(adapter->rx_ring);
2804 err_sw_init:
2805         igbvf_reset_interrupt_capability(adapter);
2806         iounmap(adapter->hw.hw_addr);
2807 err_ioremap:
2808         free_netdev(netdev);
2809 err_alloc_etherdev:
2810         pci_release_regions(pdev);
2811 err_pci_reg:
2812 err_dma:
2813         pci_disable_device(pdev);
2814         return err;
2815 }
2816
2817 /**
2818  * igbvf_remove - Device Removal Routine
2819  * @pdev: PCI device information struct
2820  *
2821  * igbvf_remove is called by the PCI subsystem to alert the driver
2822  * that it should release a PCI device.  The could be caused by a
2823  * Hot-Plug event, or because the driver is going to be removed from
2824  * memory.
2825  **/
2826 static void __devexit igbvf_remove(struct pci_dev *pdev)
2827 {
2828         struct net_device *netdev = pci_get_drvdata(pdev);
2829         struct igbvf_adapter *adapter = netdev_priv(netdev);
2830         struct e1000_hw *hw = &adapter->hw;
2831
2832         /*
2833          * flush_scheduled work may reschedule our watchdog task, so
2834          * explicitly disable watchdog tasks from being rescheduled
2835          */
2836         set_bit(__IGBVF_DOWN, &adapter->state);
2837         del_timer_sync(&adapter->watchdog_timer);
2838
2839         flush_scheduled_work();
2840
2841         unregister_netdev(netdev);
2842
2843         igbvf_reset_interrupt_capability(adapter);
2844
2845         /*
2846          * it is important to delete the napi struct prior to freeing the
2847          * rx ring so that you do not end up with null pointer refs
2848          */
2849         netif_napi_del(&adapter->rx_ring->napi);
2850         kfree(adapter->tx_ring);
2851         kfree(adapter->rx_ring);
2852
2853         iounmap(hw->hw_addr);
2854         if (hw->flash_address)
2855                 iounmap(hw->flash_address);
2856         pci_release_regions(pdev);
2857
2858         free_netdev(netdev);
2859
2860         pci_disable_device(pdev);
2861 }
2862
2863 /* PCI Error Recovery (ERS) */
2864 static struct pci_error_handlers igbvf_err_handler = {
2865         .error_detected = igbvf_io_error_detected,
2866         .slot_reset = igbvf_io_slot_reset,
2867         .resume = igbvf_io_resume,
2868 };
2869
2870 static DEFINE_PCI_DEVICE_TABLE(igbvf_pci_tbl) = {
2871         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82576_VF), board_vf },
2872         { } /* terminate list */
2873 };
2874 MODULE_DEVICE_TABLE(pci, igbvf_pci_tbl);
2875
2876 /* PCI Device API Driver */
2877 static struct pci_driver igbvf_driver = {
2878         .name     = igbvf_driver_name,
2879         .id_table = igbvf_pci_tbl,
2880         .probe    = igbvf_probe,
2881         .remove   = __devexit_p(igbvf_remove),
2882 #ifdef CONFIG_PM
2883         /* Power Management Hooks */
2884         .suspend  = igbvf_suspend,
2885         .resume   = igbvf_resume,
2886 #endif
2887         .shutdown = igbvf_shutdown,
2888         .err_handler = &igbvf_err_handler
2889 };
2890
2891 /**
2892  * igbvf_init_module - Driver Registration Routine
2893  *
2894  * igbvf_init_module is the first routine called when the driver is
2895  * loaded. All it does is register with the PCI subsystem.
2896  **/
2897 static int __init igbvf_init_module(void)
2898 {
2899         int ret;
2900         printk(KERN_INFO "%s - version %s\n",
2901                igbvf_driver_string, igbvf_driver_version);
2902         printk(KERN_INFO "%s\n", igbvf_copyright);
2903
2904         ret = pci_register_driver(&igbvf_driver);
2905         igbvf_driver_pm_qos_req = pm_qos_add_request(PM_QOS_CPU_DMA_LATENCY,
2906                                PM_QOS_DEFAULT_VALUE);
2907
2908         return ret;
2909 }
2910 module_init(igbvf_init_module);
2911
2912 /**
2913  * igbvf_exit_module - Driver Exit Cleanup Routine
2914  *
2915  * igbvf_exit_module is called just before the driver is removed
2916  * from memory.
2917  **/
2918 static void __exit igbvf_exit_module(void)
2919 {
2920         pci_unregister_driver(&igbvf_driver);
2921         pm_qos_remove_request(igbvf_driver_pm_qos_req);
2922         igbvf_driver_pm_qos_req = NULL;
2923 }
2924 module_exit(igbvf_exit_module);
2925
2926
2927 MODULE_AUTHOR("Intel Corporation, <e1000-devel@lists.sourceforge.net>");
2928 MODULE_DESCRIPTION("Intel(R) 82576 Virtual Function Network Driver");
2929 MODULE_LICENSE("GPL");
2930 MODULE_VERSION(DRV_VERSION);
2931
2932 /* netdev.c */