drivers/net/ethernet/intel/ixgbevf/ixgbevf_main.c

   1 /*******************************************************************************
   2
   3   Intel 82599 Virtual Function driver
   4   Copyright(c) 1999 - 2014 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
  29 /******************************************************************************
  30  Copyright (c)2006 - 2007 Myricom, Inc. for some LRO specific code
  31 ******************************************************************************/
  32
  33 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
  34
  35 #include <linux/types.h>
  36 #include <linux/bitops.h>
  37 #include <linux/module.h>
  38 #include <linux/pci.h>
  39 #include <linux/netdevice.h>
  40 #include <linux/vmalloc.h>
  41 #include <linux/string.h>
  42 #include <linux/in.h>
  43 #include <linux/ip.h>
  44 #include <linux/tcp.h>
  45 #include <linux/sctp.h>
  46 #include <linux/ipv6.h>
  47 #include <linux/slab.h>
  48 #include <net/checksum.h>
  49 #include <net/ip6_checksum.h>
  50 #include <linux/ethtool.h>
  51 #include <linux/if.h>
  52 #include <linux/if_vlan.h>
  53 #include <linux/prefetch.h>
  54
  55 #include "ixgbevf.h"
  56
  57 const char ixgbevf_driver_name[] = "ixgbevf";
  58 static const char ixgbevf_driver_string[] =
  59         "Intel(R) 10 Gigabit PCI Express Virtual Function Network Driver";
  60
  61 #define DRV_VERSION "2.12.1-k"
  62 const char ixgbevf_driver_version[] = DRV_VERSION;
  63 static char ixgbevf_copyright[] =
  64         "Copyright (c) 2009 - 2012 Intel Corporation.";
  65
  66 static const struct ixgbevf_info *ixgbevf_info_tbl[] = {
  67         [board_82599_vf] = &ixgbevf_82599_vf_info,
  68         [board_X540_vf]  = &ixgbevf_X540_vf_info,
  69 };
  70
  71 /* ixgbevf_pci_tbl - PCI Device ID Table
  72  *
  73  * Wildcard entries (PCI_ANY_ID) should come last
  74  * Last entry must be all 0s
  75  *
  76  * { Vendor ID, Device ID, SubVendor ID, SubDevice ID,
  77  *   Class, Class Mask, private data (not used) }
  78  */
  79 static DEFINE_PCI_DEVICE_TABLE(ixgbevf_pci_tbl) = {
  80         {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82599_VF), board_82599_vf },
  81         {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_X540_VF), board_X540_vf },
  82         /* required last entry */
  83         {0, }
  84 };
  85 MODULE_DEVICE_TABLE(pci, ixgbevf_pci_tbl);
  86
  87 MODULE_AUTHOR("Intel Corporation, <linux.nics@intel.com>");
  88 MODULE_DESCRIPTION("Intel(R) 82599 Virtual Function Driver");
  89 MODULE_LICENSE("GPL");
  90 MODULE_VERSION(DRV_VERSION);
  91
  92 #define DEFAULT_MSG_ENABLE (NETIF_MSG_DRV|NETIF_MSG_PROBE|NETIF_MSG_LINK)
  93 static int debug = -1;
  94 module_param(debug, int, 0);
  95 MODULE_PARM_DESC(debug, "Debug level (0=none,...,16=all)");
  96
  97 /* forward decls */
  98 static void ixgbevf_queue_reset_subtask(struct ixgbevf_adapter *adapter);
  99 static void ixgbevf_set_itr(struct ixgbevf_q_vector *q_vector);
 100 static void ixgbevf_free_all_rx_resources(struct ixgbevf_adapter *adapter);
 101
 102 static void ixgbevf_remove_adapter(struct ixgbe_hw *hw)
 103 {
 104         struct ixgbevf_adapter *adapter = hw->back;
 105
 106         if (!hw->hw_addr)
 107                 return;
 108         hw->hw_addr = NULL;
 109         dev_err(&adapter->pdev->dev, "Adapter removed\n");
 110         if (test_bit(__IXGBEVF_WORK_INIT, &adapter->state))
 111                 schedule_work(&adapter->watchdog_task);
 112 }
 113
 114 static void ixgbevf_check_remove(struct ixgbe_hw *hw, u32 reg)
 115 {
 116         u32 value;
 117
 118         /* The following check not only optimizes a bit by not
 119          * performing a read on the status register when the
 120          * register just read was a status register read that
 121          * returned IXGBE_FAILED_READ_REG. It also blocks any
 122          * potential recursion.
 123          */
 124         if (reg == IXGBE_VFSTATUS) {
 125                 ixgbevf_remove_adapter(hw);
 126                 return;
 127         }
 128         value = ixgbevf_read_reg(hw, IXGBE_VFSTATUS);
 129         if (value == IXGBE_FAILED_READ_REG)
 130                 ixgbevf_remove_adapter(hw);
 131 }
 132
 133 u32 ixgbevf_read_reg(struct ixgbe_hw *hw, u32 reg)
 134 {
 135         u8 __iomem *reg_addr = ACCESS_ONCE(hw->hw_addr);
 136         u32 value;
 137
 138         if (IXGBE_REMOVED(reg_addr))
 139                 return IXGBE_FAILED_READ_REG;
 140         value = readl(reg_addr + reg);
 141         if (unlikely(value == IXGBE_FAILED_READ_REG))
 142                 ixgbevf_check_remove(hw, reg);
 143         return value;
 144 }
 145
 146 static inline void ixgbevf_release_rx_desc(struct ixgbevf_ring *rx_ring,
 147                                            u32 val)
 148 {
 149         rx_ring->next_to_use = val;
 150
 151         /*
 152          * Force memory writes to complete before letting h/w
 153          * know there are new descriptors to fetch.  (Only
 154          * applicable for weak-ordered memory model archs,
 155          * such as IA-64).
 156          */
 157         wmb();
 158         ixgbevf_write_tail(rx_ring, val);
 159 }
 160
 161 /**
 162  * ixgbevf_set_ivar - set IVAR registers - maps interrupt causes to vectors
 163  * @adapter: pointer to adapter struct
 164  * @direction: 0 for Rx, 1 for Tx, -1 for other causes
 165  * @queue: queue to map the corresponding interrupt to
 166  * @msix_vector: the vector to map to the corresponding queue
 167  */
 168 static void ixgbevf_set_ivar(struct ixgbevf_adapter *adapter, s8 direction,
 169                              u8 queue, u8 msix_vector)
 170 {
 171         u32 ivar, index;
 172         struct ixgbe_hw *hw = &adapter->hw;
 173         if (direction == -1) {
 174                 /* other causes */
 175                 msix_vector |= IXGBE_IVAR_ALLOC_VAL;
 176                 ivar = IXGBE_READ_REG(hw, IXGBE_VTIVAR_MISC);
 177                 ivar &= ~0xFF;
 178                 ivar |= msix_vector;
 179                 IXGBE_WRITE_REG(hw, IXGBE_VTIVAR_MISC, ivar);
 180         } else {
 181                 /* tx or rx causes */
 182                 msix_vector |= IXGBE_IVAR_ALLOC_VAL;
 183                 index = ((16 * (queue & 1)) + (8 * direction));
 184                 ivar = IXGBE_READ_REG(hw, IXGBE_VTIVAR(queue >> 1));
 185                 ivar &= ~(0xFF << index);
 186                 ivar |= (msix_vector << index);
 187                 IXGBE_WRITE_REG(hw, IXGBE_VTIVAR(queue >> 1), ivar);
 188         }
 189 }
 190
 191 static void ixgbevf_unmap_and_free_tx_resource(struct ixgbevf_ring *tx_ring,
 192                                         struct ixgbevf_tx_buffer *tx_buffer)
 193 {
 194         if (tx_buffer->skb) {
 195                 dev_kfree_skb_any(tx_buffer->skb);
 196                 if (dma_unmap_len(tx_buffer, len))
 197                         dma_unmap_single(tx_ring->dev,
 198                                          dma_unmap_addr(tx_buffer, dma),
 199                                          dma_unmap_len(tx_buffer, len),
 200                                          DMA_TO_DEVICE);
 201         } else if (dma_unmap_len(tx_buffer, len)) {
 202                 dma_unmap_page(tx_ring->dev,
 203                                dma_unmap_addr(tx_buffer, dma),
 204                                dma_unmap_len(tx_buffer, len),
 205                                DMA_TO_DEVICE);
 206         }
 207         tx_buffer->next_to_watch = NULL;
 208         tx_buffer->skb = NULL;
 209         dma_unmap_len_set(tx_buffer, len, 0);
 210         /* tx_buffer must be completely set up in the transmit path */
 211 }
 212
 213 #define IXGBE_MAX_TXD_PWR       14
 214 #define IXGBE_MAX_DATA_PER_TXD  (1 << IXGBE_MAX_TXD_PWR)
 215
 216 /* Tx Descriptors needed, worst case */
 217 #define TXD_USE_COUNT(S) DIV_ROUND_UP((S), IXGBE_MAX_DATA_PER_TXD)
 218 #define DESC_NEEDED (MAX_SKB_FRAGS + 4)
 219
 220 static void ixgbevf_tx_timeout(struct net_device *netdev);
 221
 222 /**
 223  * ixgbevf_clean_tx_irq - Reclaim resources after transmit completes
 224  * @q_vector: board private structure
 225  * @tx_ring: tx ring to clean
 226  **/
 227 static bool ixgbevf_clean_tx_irq(struct ixgbevf_q_vector *q_vector,
 228                                  struct ixgbevf_ring *tx_ring)
 229 {
 230         struct ixgbevf_adapter *adapter = q_vector->adapter;
 231         struct ixgbevf_tx_buffer *tx_buffer;
 232         union ixgbe_adv_tx_desc *tx_desc;
 233         unsigned int total_bytes = 0, total_packets = 0;
 234         unsigned int budget = tx_ring->count / 2;
 235         unsigned int i = tx_ring->next_to_clean;
 236
 237         if (test_bit(__IXGBEVF_DOWN, &adapter->state))
 238                 return true;
 239
 240         tx_buffer = &tx_ring->tx_buffer_info[i];
 241         tx_desc = IXGBEVF_TX_DESC(tx_ring, i);
 242         i -= tx_ring->count;
 243
 244         do {
 245                 union ixgbe_adv_tx_desc *eop_desc = tx_buffer->next_to_watch;
 246
 247                 /* if next_to_watch is not set then there is no work pending */
 248                 if (!eop_desc)
 249                         break;
 250
 251                 /* prevent any other reads prior to eop_desc */
 252                 read_barrier_depends();
 253
 254                 /* if DD is not set pending work has not been completed */
 255                 if (!(eop_desc->wb.status & cpu_to_le32(IXGBE_TXD_STAT_DD)))
 256                         break;
 257
 258                 /* clear next_to_watch to prevent false hangs */
 259                 tx_buffer->next_to_watch = NULL;
 260
 261                 /* update the statistics for this packet */
 262                 total_bytes += tx_buffer->bytecount;
 263                 total_packets += tx_buffer->gso_segs;
 264
 265                 /* free the skb */
 266                 dev_kfree_skb_any(tx_buffer->skb);
 267
 268                 /* unmap skb header data */
 269                 dma_unmap_single(tx_ring->dev,
 270                                  dma_unmap_addr(tx_buffer, dma),
 271                                  dma_unmap_len(tx_buffer, len),
 272                                  DMA_TO_DEVICE);
 273
 274                 /* clear tx_buffer data */
 275                 tx_buffer->skb = NULL;
 276                 dma_unmap_len_set(tx_buffer, len, 0);
 277
 278                 /* unmap remaining buffers */
 279                 while (tx_desc != eop_desc) {
 280                         tx_buffer++;
 281                         tx_desc++;
 282                         i++;
 283                         if (unlikely(!i)) {
 284                                 i -= tx_ring->count;
 285                                 tx_buffer = tx_ring->tx_buffer_info;
 286                                 tx_desc = IXGBEVF_TX_DESC(tx_ring, 0);
 287                         }
 288
 289                         /* unmap any remaining paged data */
 290                         if (dma_unmap_len(tx_buffer, len)) {
 291                                 dma_unmap_page(tx_ring->dev,
 292                                                dma_unmap_addr(tx_buffer, dma),
 293                                                dma_unmap_len(tx_buffer, len),
 294                                                DMA_TO_DEVICE);
 295                                 dma_unmap_len_set(tx_buffer, len, 0);
 296                         }
 297                 }
 298
 299                 /* move us one more past the eop_desc for start of next pkt */
 300                 tx_buffer++;
 301                 tx_desc++;
 302                 i++;
 303                 if (unlikely(!i)) {
 304                         i -= tx_ring->count;
 305                         tx_buffer = tx_ring->tx_buffer_info;
 306                         tx_desc = IXGBEVF_TX_DESC(tx_ring, 0);
 307                 }
 308
 309                 /* issue prefetch for next Tx descriptor */
 310                 prefetch(tx_desc);
 311
 312                 /* update budget accounting */
 313                 budget--;
 314         } while (likely(budget));
 315
 316         i += tx_ring->count;
 317         tx_ring->next_to_clean = i;
 318         u64_stats_update_begin(&tx_ring->syncp);
 319         tx_ring->stats.bytes += total_bytes;
 320         tx_ring->stats.packets += total_packets;
 321         u64_stats_update_end(&tx_ring->syncp);
 322         q_vector->tx.total_bytes += total_bytes;
 323         q_vector->tx.total_packets += total_packets;
 324
 325 #define TX_WAKE_THRESHOLD (DESC_NEEDED * 2)
 326         if (unlikely(total_packets && netif_carrier_ok(tx_ring->netdev) &&
 327                      (ixgbevf_desc_unused(tx_ring) >= TX_WAKE_THRESHOLD))) {
 328                 /* Make sure that anybody stopping the queue after this
 329                  * sees the new next_to_clean.
 330                  */
 331                 smp_mb();
 332
 333                 if (__netif_subqueue_stopped(tx_ring->netdev,
 334                                              tx_ring->queue_index) &&
 335                     !test_bit(__IXGBEVF_DOWN, &adapter->state)) {
 336                         netif_wake_subqueue(tx_ring->netdev,
 337                                             tx_ring->queue_index);
 338                         ++tx_ring->tx_stats.restart_queue;
 339                 }
 340         }
 341
 342         return !!budget;
 343 }
 344
 345 /**
 346  * ixgbevf_receive_skb - Send a completed packet up the stack
 347  * @q_vector: structure containing interrupt and ring information
 348  * @skb: packet to send up
 349  * @status: hardware indication of status of receive
 350  * @rx_desc: rx descriptor
 351  **/
 352 static void ixgbevf_receive_skb(struct ixgbevf_q_vector *q_vector,
 353                                 struct sk_buff *skb, u8 status,
 354                                 union ixgbe_adv_rx_desc *rx_desc)
 355 {
 356         struct ixgbevf_adapter *adapter = q_vector->adapter;
 357         bool is_vlan = (status & IXGBE_RXD_STAT_VP);
 358         u16 tag = le16_to_cpu(rx_desc->wb.upper.vlan);
 359
 360         if (is_vlan && test_bit(tag & VLAN_VID_MASK, adapter->active_vlans))
 361                 __vlan_hwaccel_put_tag(skb, htons(ETH_P_8021Q), tag);
 362
 363         if (!(adapter->flags & IXGBE_FLAG_IN_NETPOLL))
 364                 napi_gro_receive(&q_vector->napi, skb);
 365         else
 366                 netif_rx(skb);
 367 }
 368
 369 /**
 370  * ixgbevf_rx_skb - Helper function to determine proper Rx method
 371  * @q_vector: structure containing interrupt and ring information
 372  * @skb: packet to send up
 373  * @status: hardware indication of status of receive
 374  * @rx_desc: rx descriptor
 375  **/
 376 static void ixgbevf_rx_skb(struct ixgbevf_q_vector *q_vector,
 377                            struct sk_buff *skb, u8 status,
 378                            union ixgbe_adv_rx_desc *rx_desc)
 379 {
 380 #ifdef CONFIG_NET_RX_BUSY_POLL
 381         skb_mark_napi_id(skb, &q_vector->napi);
 382
 383         if (ixgbevf_qv_busy_polling(q_vector)) {
 384                 netif_receive_skb(skb);
 385                 /* exit early if we busy polled */
 386                 return;
 387         }
 388 #endif /* CONFIG_NET_RX_BUSY_POLL */
 389
 390         ixgbevf_receive_skb(q_vector, skb, status, rx_desc);
 391 }
 392
 393 /**
 394  * ixgbevf_rx_checksum - indicate in skb if hw indicated a good cksum
 395  * @ring: pointer to Rx descriptor ring structure
 396  * @status_err: hardware indication of status of receive
 397  * @skb: skb currently being received and modified
 398  **/
 399 static inline void ixgbevf_rx_checksum(struct ixgbevf_ring *ring,
 400                                        u32 status_err, struct sk_buff *skb)
 401 {
 402         skb_checksum_none_assert(skb);
 403
 404         /* Rx csum disabled */
 405         if (!(ring->netdev->features & NETIF_F_RXCSUM))
 406                 return;
 407
 408         /* if IP and error */
 409         if ((status_err & IXGBE_RXD_STAT_IPCS) &&
 410             (status_err & IXGBE_RXDADV_ERR_IPE)) {
 411                 ring->rx_stats.csum_err++;
 412                 return;
 413         }
 414
 415         if (!(status_err & IXGBE_RXD_STAT_L4CS))
 416                 return;
 417
 418         if (status_err & IXGBE_RXDADV_ERR_TCPE) {
 419                 ring->rx_stats.csum_err++;
 420                 return;
 421         }
 422
 423         /* It must be a TCP or UDP packet with a valid checksum */
 424         skb->ip_summed = CHECKSUM_UNNECESSARY;
 425 }
 426
 427 /**
 428  * ixgbevf_alloc_rx_buffers - Replace used receive buffers; packet split
 429  * @rx_ring: rx descriptor ring (for a specific queue) to setup buffers on
 430  **/
 431 static void ixgbevf_alloc_rx_buffers(struct ixgbevf_ring *rx_ring,
 432                                      int cleaned_count)
 433 {
 434         union ixgbe_adv_rx_desc *rx_desc;
 435         struct ixgbevf_rx_buffer *bi;
 436         unsigned int i = rx_ring->next_to_use;
 437
 438         while (cleaned_count--) {
 439                 rx_desc = IXGBEVF_RX_DESC(rx_ring, i);
 440                 bi = &rx_ring->rx_buffer_info[i];
 441
 442                 if (!bi->skb) {
 443                         struct sk_buff *skb;
 444
 445                         skb = netdev_alloc_skb_ip_align(rx_ring->netdev,
 446                                                         rx_ring->rx_buf_len);
 447                         if (!skb)
 448                                 goto no_buffers;
 449
 450                         bi->skb = skb;
 451
 452                         bi->dma = dma_map_single(rx_ring->dev, skb->data,
 453                                                  rx_ring->rx_buf_len,
 454                                                  DMA_FROM_DEVICE);
 455                         if (dma_mapping_error(rx_ring->dev, bi->dma)) {
 456                                 dev_kfree_skb(skb);
 457                                 bi->skb = NULL;
 458                                 dev_err(rx_ring->dev, "Rx DMA map failed\n");
 459                                 break;
 460                         }
 461                 }
 462                 rx_desc->read.pkt_addr = cpu_to_le64(bi->dma);
 463
 464                 i++;
 465                 if (i == rx_ring->count)
 466                         i = 0;
 467         }
 468
 469 no_buffers:
 470         rx_ring->rx_stats.alloc_rx_buff_failed++;
 471         if (rx_ring->next_to_use != i)
 472                 ixgbevf_release_rx_desc(rx_ring, i);
 473 }
 474
 475 static inline void ixgbevf_irq_enable_queues(struct ixgbevf_adapter *adapter,
 476                                              u32 qmask)
 477 {
 478         struct ixgbe_hw *hw = &adapter->hw;
 479
 480         IXGBE_WRITE_REG(hw, IXGBE_VTEIMS, qmask);
 481 }
 482
 483 static int ixgbevf_clean_rx_irq(struct ixgbevf_q_vector *q_vector,
 484                                 struct ixgbevf_ring *rx_ring,
 485                                 int budget)
 486 {
 487         union ixgbe_adv_rx_desc *rx_desc, *next_rxd;
 488         struct ixgbevf_rx_buffer *rx_buffer_info, *next_buffer;
 489         struct sk_buff *skb;
 490         unsigned int i;
 491         u32 len, staterr;
 492         int cleaned_count = 0;
 493         unsigned int total_rx_bytes = 0, total_rx_packets = 0;
 494
 495         i = rx_ring->next_to_clean;
 496         rx_desc = IXGBEVF_RX_DESC(rx_ring, i);
 497         staterr = le32_to_cpu(rx_desc->wb.upper.status_error);
 498         rx_buffer_info = &rx_ring->rx_buffer_info[i];
 499
 500         while (staterr & IXGBE_RXD_STAT_DD) {
 501                 if (!budget)
 502                         break;
 503                 budget--;
 504
 505                 rmb(); /* read descriptor and rx_buffer_info after status DD */
 506                 len = le16_to_cpu(rx_desc->wb.upper.length);
 507                 skb = rx_buffer_info->skb;
 508                 prefetch(skb->data - NET_IP_ALIGN);
 509                 rx_buffer_info->skb = NULL;
 510
 511                 if (rx_buffer_info->dma) {
 512                         dma_unmap_single(rx_ring->dev, rx_buffer_info->dma,
 513                                          rx_ring->rx_buf_len,
 514                                          DMA_FROM_DEVICE);
 515                         rx_buffer_info->dma = 0;
 516                         skb_put(skb, len);
 517                 }
 518
 519                 i++;
 520                 if (i == rx_ring->count)
 521                         i = 0;
 522
 523                 next_rxd = IXGBEVF_RX_DESC(rx_ring, i);
 524                 prefetch(next_rxd);
 525                 cleaned_count++;
 526
 527                 next_buffer = &rx_ring->rx_buffer_info[i];
 528
 529                 if (!(staterr & IXGBE_RXD_STAT_EOP)) {
 530                         skb->next = next_buffer->skb;
 531                         IXGBE_CB(skb->next)->prev = skb;
 532                         rx_ring->rx_stats.non_eop_descs++;
 533                         goto next_desc;
 534                 }
 535
 536                 /* we should not be chaining buffers, if we did drop the skb */
 537                 if (IXGBE_CB(skb)->prev) {
 538                         do {
 539                                 struct sk_buff *this = skb;
 540                                 skb = IXGBE_CB(skb)->prev;
 541                                 dev_kfree_skb(this);
 542                         } while (skb);
 543                         goto next_desc;
 544                 }
 545
 546                 /* ERR_MASK will only have valid bits if EOP set */
 547                 if (unlikely(staterr & IXGBE_RXDADV_ERR_FRAME_ERR_MASK)) {
 548                         dev_kfree_skb_irq(skb);
 549                         goto next_desc;
 550                 }
 551
 552                 ixgbevf_rx_checksum(rx_ring, staterr, skb);
 553
 554                 /* probably a little skewed due to removing CRC */
 555                 total_rx_bytes += skb->len;
 556                 total_rx_packets++;
 557
 558                 skb->protocol = eth_type_trans(skb, rx_ring->netdev);
 559
 560                 /* Workaround hardware that can't do proper VEPA multicast
 561                  * source pruning.
 562                  */
 563                 if ((skb->pkt_type == PACKET_BROADCAST ||
 564                     skb->pkt_type == PACKET_MULTICAST) &&
 565                     ether_addr_equal(rx_ring->netdev->dev_addr,
 566                                      eth_hdr(skb)->h_source)) {
 567                         dev_kfree_skb_irq(skb);
 568                         goto next_desc;
 569                 }
 570
 571                 ixgbevf_rx_skb(q_vector, skb, staterr, rx_desc);
 572
 573 next_desc:
 574                 rx_desc->wb.upper.status_error = 0;
 575
 576                 /* return some buffers to hardware, one at a time is too slow */
 577                 if (cleaned_count >= IXGBEVF_RX_BUFFER_WRITE) {
 578                         ixgbevf_alloc_rx_buffers(rx_ring, cleaned_count);
 579                         cleaned_count = 0;
 580                 }
 581
 582                 /* use prefetched values */
 583                 rx_desc = next_rxd;
 584                 rx_buffer_info = &rx_ring->rx_buffer_info[i];
 585
 586                 staterr = le32_to_cpu(rx_desc->wb.upper.status_error);
 587         }
 588
 589         rx_ring->next_to_clean = i;
 590         cleaned_count = ixgbevf_desc_unused(rx_ring);
 591
 592         if (cleaned_count)
 593                 ixgbevf_alloc_rx_buffers(rx_ring, cleaned_count);
 594
 595         u64_stats_update_begin(&rx_ring->syncp);
 596         rx_ring->stats.packets += total_rx_packets;
 597         rx_ring->stats.bytes += total_rx_bytes;
 598         u64_stats_update_end(&rx_ring->syncp);
 599         q_vector->rx.total_packets += total_rx_packets;
 600         q_vector->rx.total_bytes += total_rx_bytes;
 601
 602         return total_rx_packets;
 603 }
 604
 605 /**
 606  * ixgbevf_poll - NAPI polling calback
 607  * @napi: napi struct with our devices info in it
 608  * @budget: amount of work driver is allowed to do this pass, in packets
 609  *
 610  * This function will clean more than one or more rings associated with a
 611  * q_vector.
 612  **/
 613 static int ixgbevf_poll(struct napi_struct *napi, int budget)
 614 {
 615         struct ixgbevf_q_vector *q_vector =
 616                 container_of(napi, struct ixgbevf_q_vector, napi);
 617         struct ixgbevf_adapter *adapter = q_vector->adapter;
 618         struct ixgbevf_ring *ring;
 619         int per_ring_budget;
 620         bool clean_complete = true;
 621
 622         ixgbevf_for_each_ring(ring, q_vector->tx)
 623                 clean_complete &= ixgbevf_clean_tx_irq(q_vector, ring);
 624
 625 #ifdef CONFIG_NET_RX_BUSY_POLL
 626         if (!ixgbevf_qv_lock_napi(q_vector))
 627                 return budget;
 628 #endif
 629
 630         /* attempt to distribute budget to each queue fairly, but don't allow
 631          * the budget to go below 1 because we'll exit polling */
 632         if (q_vector->rx.count > 1)
 633                 per_ring_budget = max(budget/q_vector->rx.count, 1);
 634         else
 635                 per_ring_budget = budget;
 636
 637         adapter->flags |= IXGBE_FLAG_IN_NETPOLL;
 638         ixgbevf_for_each_ring(ring, q_vector->rx)
 639                 clean_complete &= (ixgbevf_clean_rx_irq(q_vector, ring,
 640                                                         per_ring_budget)
 641                                    < per_ring_budget);
 642         adapter->flags &= ~IXGBE_FLAG_IN_NETPOLL;
 643
 644 #ifdef CONFIG_NET_RX_BUSY_POLL
 645         ixgbevf_qv_unlock_napi(q_vector);
 646 #endif
 647
 648         /* If all work not completed, return budget and keep polling */
 649         if (!clean_complete)
 650                 return budget;
 651         /* all work done, exit the polling mode */
 652         napi_complete(napi);
 653         if (adapter->rx_itr_setting & 1)
 654                 ixgbevf_set_itr(q_vector);
 655         if (!test_bit(__IXGBEVF_DOWN, &adapter->state) &&
 656             !test_bit(__IXGBEVF_REMOVING, &adapter->state))
 657                 ixgbevf_irq_enable_queues(adapter,
 658                                           1 << q_vector->v_idx);
 659
 660         return 0;
 661 }
 662
 663 /**
 664  * ixgbevf_write_eitr - write VTEITR register in hardware specific way
 665  * @q_vector: structure containing interrupt and ring information
 666  */
 667 void ixgbevf_write_eitr(struct ixgbevf_q_vector *q_vector)
 668 {
 669         struct ixgbevf_adapter *adapter = q_vector->adapter;
 670         struct ixgbe_hw *hw = &adapter->hw;
 671         int v_idx = q_vector->v_idx;
 672         u32 itr_reg = q_vector->itr & IXGBE_MAX_EITR;
 673
 674         /*
 675          * set the WDIS bit to not clear the timer bits and cause an
 676          * immediate assertion of the interrupt
 677          */
 678         itr_reg |= IXGBE_EITR_CNT_WDIS;
 679
 680         IXGBE_WRITE_REG(hw, IXGBE_VTEITR(v_idx), itr_reg);
 681 }
 682
 683 #ifdef CONFIG_NET_RX_BUSY_POLL
 684 /* must be called with local_bh_disable()d */
 685 static int ixgbevf_busy_poll_recv(struct napi_struct *napi)
 686 {
 687         struct ixgbevf_q_vector *q_vector =
 688                         container_of(napi, struct ixgbevf_q_vector, napi);
 689         struct ixgbevf_adapter *adapter = q_vector->adapter;
 690         struct ixgbevf_ring  *ring;
 691         int found = 0;
 692
 693         if (test_bit(__IXGBEVF_DOWN, &adapter->state))
 694                 return LL_FLUSH_FAILED;
 695
 696         if (!ixgbevf_qv_lock_poll(q_vector))
 697                 return LL_FLUSH_BUSY;
 698
 699         ixgbevf_for_each_ring(ring, q_vector->rx) {
 700                 found = ixgbevf_clean_rx_irq(q_vector, ring, 4);
 701 #ifdef BP_EXTENDED_STATS
 702                 if (found)
 703                         ring->stats.cleaned += found;
 704                 else
 705                         ring->stats.misses++;
 706 #endif
 707                 if (found)
 708                         break;
 709         }
 710
 711         ixgbevf_qv_unlock_poll(q_vector);
 712
 713         return found;
 714 }
 715 #endif /* CONFIG_NET_RX_BUSY_POLL */
 716
 717 /**
 718  * ixgbevf_configure_msix - Configure MSI-X hardware
 719  * @adapter: board private structure
 720  *
 721  * ixgbevf_configure_msix sets up the hardware to properly generate MSI-X
 722  * interrupts.
 723  **/
 724 static void ixgbevf_configure_msix(struct ixgbevf_adapter *adapter)
 725 {
 726         struct ixgbevf_q_vector *q_vector;
 727         int q_vectors, v_idx;
 728
 729         q_vectors = adapter->num_msix_vectors - NON_Q_VECTORS;
 730         adapter->eims_enable_mask = 0;
 731
 732         /*
 733          * Populate the IVAR table and set the ITR values to the
 734          * corresponding register.
 735          */
 736         for (v_idx = 0; v_idx < q_vectors; v_idx++) {
 737                 struct ixgbevf_ring *ring;
 738                 q_vector = adapter->q_vector[v_idx];
 739
 740                 ixgbevf_for_each_ring(ring, q_vector->rx)
 741                         ixgbevf_set_ivar(adapter, 0, ring->reg_idx, v_idx);
 742
 743                 ixgbevf_for_each_ring(ring, q_vector->tx)
 744                         ixgbevf_set_ivar(adapter, 1, ring->reg_idx, v_idx);
 745
 746                 if (q_vector->tx.ring && !q_vector->rx.ring) {
 747                         /* tx only vector */
 748                         if (adapter->tx_itr_setting == 1)
 749                                 q_vector->itr = IXGBE_10K_ITR;
 750                         else
 751                                 q_vector->itr = adapter->tx_itr_setting;
 752                 } else {
 753                         /* rx or rx/tx vector */
 754                         if (adapter->rx_itr_setting == 1)
 755                                 q_vector->itr = IXGBE_20K_ITR;
 756                         else
 757                                 q_vector->itr = adapter->rx_itr_setting;
 758                 }
 759
 760                 /* add q_vector eims value to global eims_enable_mask */
 761                 adapter->eims_enable_mask |= 1 << v_idx;
 762
 763                 ixgbevf_write_eitr(q_vector);
 764         }
 765
 766         ixgbevf_set_ivar(adapter, -1, 1, v_idx);
 767         /* setup eims_other and add value to global eims_enable_mask */
 768         adapter->eims_other = 1 << v_idx;
 769         adapter->eims_enable_mask |= adapter->eims_other;
 770 }
 771
 772 enum latency_range {
 773         lowest_latency = 0,
 774         low_latency = 1,
 775         bulk_latency = 2,
 776         latency_invalid = 255
 777 };
 778
 779 /**
 780  * ixgbevf_update_itr - update the dynamic ITR value based on statistics
 781  * @q_vector: structure containing interrupt and ring information
 782  * @ring_container: structure containing ring performance data
 783  *
 784  *      Stores a new ITR value based on packets and byte
 785  *      counts during the last interrupt.  The advantage of per interrupt
 786  *      computation is faster updates and more accurate ITR for the current
 787  *      traffic pattern.  Constants in this function were computed
 788  *      based on theoretical maximum wire speed and thresholds were set based
 789  *      on testing data as well as attempting to minimize response time
 790  *      while increasing bulk throughput.
 791  **/
 792 static void ixgbevf_update_itr(struct ixgbevf_q_vector *q_vector,
 793                                struct ixgbevf_ring_container *ring_container)
 794 {
 795         int bytes = ring_container->total_bytes;
 796         int packets = ring_container->total_packets;
 797         u32 timepassed_us;
 798         u64 bytes_perint;
 799         u8 itr_setting = ring_container->itr;
 800
 801         if (packets == 0)
 802                 return;
 803
 804         /* simple throttlerate management
 805          *    0-20MB/s lowest (100000 ints/s)
 806          *   20-100MB/s low   (20000 ints/s)
 807          *  100-1249MB/s bulk (8000 ints/s)
 808          */
 809         /* what was last interrupt timeslice? */
 810         timepassed_us = q_vector->itr >> 2;
 811         bytes_perint = bytes / timepassed_us; /* bytes/usec */
 812
 813         switch (itr_setting) {
 814         case lowest_latency:
 815                 if (bytes_perint > 10)
 816                         itr_setting = low_latency;
 817                 break;
 818         case low_latency:
 819                 if (bytes_perint > 20)
 820                         itr_setting = bulk_latency;
 821                 else if (bytes_perint <= 10)
 822                         itr_setting = lowest_latency;
 823                 break;
 824         case bulk_latency:
 825                 if (bytes_perint <= 20)
 826                         itr_setting = low_latency;
 827                 break;
 828         }
 829
 830         /* clear work counters since we have the values we need */
 831         ring_container->total_bytes = 0;
 832         ring_container->total_packets = 0;
 833
 834         /* write updated itr to ring container */
 835         ring_container->itr = itr_setting;
 836 }
 837
 838 static void ixgbevf_set_itr(struct ixgbevf_q_vector *q_vector)
 839 {
 840         u32 new_itr = q_vector->itr;
 841         u8 current_itr;
 842
 843         ixgbevf_update_itr(q_vector, &q_vector->tx);
 844         ixgbevf_update_itr(q_vector, &q_vector->rx);
 845
 846         current_itr = max(q_vector->rx.itr, q_vector->tx.itr);
 847
 848         switch (current_itr) {
 849         /* counts and packets in update_itr are dependent on these numbers */
 850         case lowest_latency:
 851                 new_itr = IXGBE_100K_ITR;
 852                 break;
 853         case low_latency:
 854                 new_itr = IXGBE_20K_ITR;
 855                 break;
 856         case bulk_latency:
 857         default:
 858                 new_itr = IXGBE_8K_ITR;
 859                 break;
 860         }
 861
 862         if (new_itr != q_vector->itr) {
 863                 /* do an exponential smoothing */
 864                 new_itr = (10 * new_itr * q_vector->itr) /
 865                           ((9 * new_itr) + q_vector->itr);
 866
 867                 /* save the algorithm value here */
 868                 q_vector->itr = new_itr;
 869
 870                 ixgbevf_write_eitr(q_vector);
 871         }
 872 }
 873
 874 static irqreturn_t ixgbevf_msix_other(int irq, void *data)
 875 {
 876         struct ixgbevf_adapter *adapter = data;
 877         struct ixgbe_hw *hw = &adapter->hw;
 878
 879         hw->mac.get_link_status = 1;
 880
 881         if (!test_bit(__IXGBEVF_DOWN, &adapter->state) &&
 882             !test_bit(__IXGBEVF_REMOVING, &adapter->state))
 883                 mod_timer(&adapter->watchdog_timer, jiffies);
 884
 885         IXGBE_WRITE_REG(hw, IXGBE_VTEIMS, adapter->eims_other);
 886
 887         return IRQ_HANDLED;
 888 }
 889
 890 /**
 891  * ixgbevf_msix_clean_rings - single unshared vector rx clean (all queues)
 892  * @irq: unused
 893  * @data: pointer to our q_vector struct for this interrupt vector
 894  **/
 895 static irqreturn_t ixgbevf_msix_clean_rings(int irq, void *data)
 896 {
 897         struct ixgbevf_q_vector *q_vector = data;
 898
 899         /* EIAM disabled interrupts (on this vector) for us */
 900         if (q_vector->rx.ring || q_vector->tx.ring)
 901                 napi_schedule(&q_vector->napi);
 902
 903         return IRQ_HANDLED;
 904 }
 905
 906 static inline void map_vector_to_rxq(struct ixgbevf_adapter *a, int v_idx,
 907                                      int r_idx)
 908 {
 909         struct ixgbevf_q_vector *q_vector = a->q_vector[v_idx];
 910
 911         a->rx_ring[r_idx]->next = q_vector->rx.ring;
 912         q_vector->rx.ring = a->rx_ring[r_idx];
 913         q_vector->rx.count++;
 914 }
 915
 916 static inline void map_vector_to_txq(struct ixgbevf_adapter *a, int v_idx,
 917                                      int t_idx)
 918 {
 919         struct ixgbevf_q_vector *q_vector = a->q_vector[v_idx];
 920
 921         a->tx_ring[t_idx]->next = q_vector->tx.ring;
 922         q_vector->tx.ring = a->tx_ring[t_idx];
 923         q_vector->tx.count++;
 924 }
 925
 926 /**
 927  * ixgbevf_map_rings_to_vectors - Maps descriptor rings to vectors
 928  * @adapter: board private structure to initialize
 929  *
 930  * This function maps descriptor rings to the queue-specific vectors
 931  * we were allotted through the MSI-X enabling code.  Ideally, we'd have
 932  * one vector per ring/queue, but on a constrained vector budget, we
 933  * group the rings as "efficiently" as possible.  You would add new
 934  * mapping configurations in here.
 935  **/
 936 static int ixgbevf_map_rings_to_vectors(struct ixgbevf_adapter *adapter)
 937 {
 938         int q_vectors;
 939         int v_start = 0;
 940         int rxr_idx = 0, txr_idx = 0;
 941         int rxr_remaining = adapter->num_rx_queues;
 942         int txr_remaining = adapter->num_tx_queues;
 943         int i, j;
 944         int rqpv, tqpv;
 945         int err = 0;
 946
 947         q_vectors = adapter->num_msix_vectors - NON_Q_VECTORS;
 948
 949         /*
 950          * The ideal configuration...
 951          * We have enough vectors to map one per queue.
 952          */
 953         if (q_vectors == adapter->num_rx_queues + adapter->num_tx_queues) {
 954                 for (; rxr_idx < rxr_remaining; v_start++, rxr_idx++)
 955                         map_vector_to_rxq(adapter, v_start, rxr_idx);
 956
 957                 for (; txr_idx < txr_remaining; v_start++, txr_idx++)
 958                         map_vector_to_txq(adapter, v_start, txr_idx);
 959                 goto out;
 960         }
 961
 962         /*
 963          * If we don't have enough vectors for a 1-to-1
 964          * mapping, we'll have to group them so there are
 965          * multiple queues per vector.
 966          */
 967         /* Re-adjusting *qpv takes care of the remainder. */
 968         for (i = v_start; i < q_vectors; i++) {
 969                 rqpv = DIV_ROUND_UP(rxr_remaining, q_vectors - i);
 970                 for (j = 0; j < rqpv; j++) {
 971                         map_vector_to_rxq(adapter, i, rxr_idx);
 972                         rxr_idx++;
 973                         rxr_remaining--;
 974                 }
 975         }
 976         for (i = v_start; i < q_vectors; i++) {
 977                 tqpv = DIV_ROUND_UP(txr_remaining, q_vectors - i);
 978                 for (j = 0; j < tqpv; j++) {
 979                         map_vector_to_txq(adapter, i, txr_idx);
 980                         txr_idx++;
 981                         txr_remaining--;
 982                 }
 983         }
 984
 985 out:
 986         return err;
 987 }
 988
 989 /**
 990  * ixgbevf_request_msix_irqs - Initialize MSI-X interrupts
 991  * @adapter: board private structure
 992  *
 993  * ixgbevf_request_msix_irqs allocates MSI-X vectors and requests
 994  * interrupts from the kernel.
 995  **/
 996 static int ixgbevf_request_msix_irqs(struct ixgbevf_adapter *adapter)
 997 {
 998         struct net_device *netdev = adapter->netdev;
 999         int q_vectors = adapter->num_msix_vectors - NON_Q_VECTORS;
1000         int vector, err;
1001         int ri = 0, ti = 0;
1002
1003         for (vector = 0; vector < q_vectors; vector++) {
1004                 struct ixgbevf_q_vector *q_vector = adapter->q_vector[vector];
1005                 struct msix_entry *entry = &adapter->msix_entries[vector];
1006
1007                 if (q_vector->tx.ring && q_vector->rx.ring) {
1008                         snprintf(q_vector->name, sizeof(q_vector->name) - 1,
1009                                  "%s-%s-%d", netdev->name, "TxRx", ri++);
1010                         ti++;
1011                 } else if (q_vector->rx.ring) {
1012                         snprintf(q_vector->name, sizeof(q_vector->name) - 1,
1013                                  "%s-%s-%d", netdev->name, "rx", ri++);
1014                 } else if (q_vector->tx.ring) {
1015                         snprintf(q_vector->name, sizeof(q_vector->name) - 1,
1016                                  "%s-%s-%d", netdev->name, "tx", ti++);
1017                 } else {
1018                         /* skip this unused q_vector */
1019                         continue;
1020                 }
1021                 err = request_irq(entry->vector, &ixgbevf_msix_clean_rings, 0,
1022                                   q_vector->name, q_vector);
1023                 if (err) {
1024                         hw_dbg(&adapter->hw,
1025                                "request_irq failed for MSIX interrupt "
1026                                "Error: %d\n", err);
1027                         goto free_queue_irqs;
1028                 }
1029         }
1030
1031         err = request_irq(adapter->msix_entries[vector].vector,
1032                           &ixgbevf_msix_other, 0, netdev->name, adapter);
1033         if (err) {
1034                 hw_dbg(&adapter->hw,
1035                        "request_irq for msix_other failed: %d\n", err);
1036                 goto free_queue_irqs;
1037         }
1038
1039         return 0;
1040
1041 free_queue_irqs:
1042         while (vector) {
1043                 vector--;
1044                 free_irq(adapter->msix_entries[vector].vector,
1045                          adapter->q_vector[vector]);
1046         }
1047         /* This failure is non-recoverable - it indicates the system is
1048          * out of MSIX vector resources and the VF driver cannot run
1049          * without them.  Set the number of msix vectors to zero
1050          * indicating that not enough can be allocated.  The error
1051          * will be returned to the user indicating device open failed.
1052          * Any further attempts to force the driver to open will also
1053          * fail.  The only way to recover is to unload the driver and
1054          * reload it again.  If the system has recovered some MSIX
1055          * vectors then it may succeed.
1056          */
1057         adapter->num_msix_vectors = 0;
1058         return err;
1059 }
1060
1061 static inline void ixgbevf_reset_q_vectors(struct ixgbevf_adapter *adapter)
1062 {
1063         int i, q_vectors = adapter->num_msix_vectors - NON_Q_VECTORS;
1064
1065         for (i = 0; i < q_vectors; i++) {
1066                 struct ixgbevf_q_vector *q_vector = adapter->q_vector[i];
1067                 q_vector->rx.ring = NULL;
1068                 q_vector->tx.ring = NULL;
1069                 q_vector->rx.count = 0;
1070                 q_vector->tx.count = 0;
1071         }
1072 }
1073
1074 /**
1075  * ixgbevf_request_irq - initialize interrupts
1076  * @adapter: board private structure
1077  *
1078  * Attempts to configure interrupts using the best available
1079  * capabilities of the hardware and kernel.
1080  **/
1081 static int ixgbevf_request_irq(struct ixgbevf_adapter *adapter)
1082 {
1083         int err = 0;
1084
1085         err = ixgbevf_request_msix_irqs(adapter);
1086
1087         if (err)
1088                 hw_dbg(&adapter->hw,
1089                        "request_irq failed, Error %d\n", err);
1090
1091         return err;
1092 }
1093
1094 static void ixgbevf_free_irq(struct ixgbevf_adapter *adapter)
1095 {
1096         int i, q_vectors;
1097
1098         q_vectors = adapter->num_msix_vectors;
1099         i = q_vectors - 1;
1100
1101         free_irq(adapter->msix_entries[i].vector, adapter);
1102         i--;
1103
1104         for (; i >= 0; i--) {
1105                 /* free only the irqs that were actually requested */
1106                 if (!adapter->q_vector[i]->rx.ring &&
1107                     !adapter->q_vector[i]->tx.ring)
1108                         continue;
1109
1110                 free_irq(adapter->msix_entries[i].vector,
1111                          adapter->q_vector[i]);
1112         }
1113
1114         ixgbevf_reset_q_vectors(adapter);
1115 }
1116
1117 /**
1118  * ixgbevf_irq_disable - Mask off interrupt generation on the NIC
1119  * @adapter: board private structure
1120  **/
1121 static inline void ixgbevf_irq_disable(struct ixgbevf_adapter *adapter)
1122 {
1123         struct ixgbe_hw *hw = &adapter->hw;
1124         int i;
1125
1126         IXGBE_WRITE_REG(hw, IXGBE_VTEIAM, 0);
1127         IXGBE_WRITE_REG(hw, IXGBE_VTEIMC, ~0);
1128         IXGBE_WRITE_REG(hw, IXGBE_VTEIAC, 0);
1129
1130         IXGBE_WRITE_FLUSH(hw);
1131
1132         for (i = 0; i < adapter->num_msix_vectors; i++)
1133                 synchronize_irq(adapter->msix_entries[i].vector);
1134 }
1135
1136 /**
1137  * ixgbevf_irq_enable - Enable default interrupt generation settings
1138  * @adapter: board private structure
1139  **/
1140 static inline void ixgbevf_irq_enable(struct ixgbevf_adapter *adapter)
1141 {
1142         struct ixgbe_hw *hw = &adapter->hw;
1143
1144         IXGBE_WRITE_REG(hw, IXGBE_VTEIAM, adapter->eims_enable_mask);
1145         IXGBE_WRITE_REG(hw, IXGBE_VTEIAC, adapter->eims_enable_mask);
1146         IXGBE_WRITE_REG(hw, IXGBE_VTEIMS, adapter->eims_enable_mask);
1147 }
1148
1149 /**
1150  * ixgbevf_configure_tx_ring - Configure 82599 VF Tx ring after Reset
1151  * @adapter: board private structure
1152  * @ring: structure containing ring specific data
1153  *
1154  * Configure the Tx descriptor ring after a reset.
1155  **/
1156 static void ixgbevf_configure_tx_ring(struct ixgbevf_adapter *adapter,
1157                                       struct ixgbevf_ring *ring)
1158 {
1159         struct ixgbe_hw *hw = &adapter->hw;
1160         u64 tdba = ring->dma;
1161         int wait_loop = 10;
1162         u32 txdctl = IXGBE_TXDCTL_ENABLE;
1163         u8 reg_idx = ring->reg_idx;
1164
1165         /* disable queue to avoid issues while updating state */
1166         IXGBE_WRITE_REG(hw, IXGBE_VFTXDCTL(reg_idx), IXGBE_TXDCTL_SWFLSH);
1167         IXGBE_WRITE_FLUSH(hw);
1168
1169         IXGBE_WRITE_REG(hw, IXGBE_VFTDBAL(reg_idx), tdba & DMA_BIT_MASK(32));
1170         IXGBE_WRITE_REG(hw, IXGBE_VFTDBAH(reg_idx), tdba >> 32);
1171         IXGBE_WRITE_REG(hw, IXGBE_VFTDLEN(reg_idx),
1172                         ring->count * sizeof(union ixgbe_adv_tx_desc));
1173
1174         /* disable head writeback */
1175         IXGBE_WRITE_REG(hw, IXGBE_VFTDWBAH(reg_idx), 0);
1176         IXGBE_WRITE_REG(hw, IXGBE_VFTDWBAL(reg_idx), 0);
1177
1178         /* enable relaxed ordering */
1179         IXGBE_WRITE_REG(hw, IXGBE_VFDCA_TXCTRL(reg_idx),
1180                         (IXGBE_DCA_TXCTRL_DESC_RRO_EN |
1181                          IXGBE_DCA_TXCTRL_DATA_RRO_EN));
1182
1183         /* reset head and tail pointers */
1184         IXGBE_WRITE_REG(hw, IXGBE_VFTDH(reg_idx), 0);
1185         IXGBE_WRITE_REG(hw, IXGBE_VFTDT(reg_idx), 0);
1186         ring->tail = adapter->io_addr + IXGBE_VFTDT(reg_idx);
1187
1188         /* reset ntu and ntc to place SW in sync with hardwdare */
1189         ring->next_to_clean = 0;
1190         ring->next_to_use = 0;
1191
1192         /* In order to avoid issues WTHRESH + PTHRESH should always be equal
1193          * to or less than the number of on chip descriptors, which is
1194          * currently 40.
1195          */
1196         txdctl |= (8 << 16);    /* WTHRESH = 8 */
1197
1198         /* Setting PTHRESH to 32 both improves performance */
1199         txdctl |= (1 << 8) |    /* HTHRESH = 1 */
1200                   32;          /* PTHRESH = 32 */
1201
1202         IXGBE_WRITE_REG(hw, IXGBE_VFTXDCTL(reg_idx), txdctl);
1203
1204         /* poll to verify queue is enabled */
1205         do {
1206                 usleep_range(1000, 2000);
1207                 txdctl = IXGBE_READ_REG(hw, IXGBE_VFTXDCTL(reg_idx));
1208         }  while (--wait_loop && !(txdctl & IXGBE_TXDCTL_ENABLE));
1209         if (!wait_loop)
1210                 pr_err("Could not enable Tx Queue %d\n", reg_idx);
1211 }
1212
1213 /**
1214  * ixgbevf_configure_tx - Configure 82599 VF Transmit Unit after Reset
1215  * @adapter: board private structure
1216  *
1217  * Configure the Tx unit of the MAC after a reset.
1218  **/
1219 static void ixgbevf_configure_tx(struct ixgbevf_adapter *adapter)
1220 {
1221         u32 i;
1222
1223         /* Setup the HW Tx Head and Tail descriptor pointers */
1224         for (i = 0; i < adapter->num_tx_queues; i++)
1225                 ixgbevf_configure_tx_ring(adapter, adapter->tx_ring[i]);
1226 }
1227
1228 #define IXGBE_SRRCTL_BSIZEHDRSIZE_SHIFT 2
1229
1230 static void ixgbevf_configure_srrctl(struct ixgbevf_adapter *adapter, int index)
1231 {
1232         struct ixgbevf_ring *rx_ring;
1233         struct ixgbe_hw *hw = &adapter->hw;
1234         u32 srrctl;
1235
1236         rx_ring = adapter->rx_ring[index];
1237
1238         srrctl = IXGBE_SRRCTL_DROP_EN;
1239
1240         srrctl |= IXGBE_SRRCTL_DESCTYPE_ADV_ONEBUF;
1241
1242         srrctl |= ALIGN(rx_ring->rx_buf_len, 1024) >>
1243                   IXGBE_SRRCTL_BSIZEPKT_SHIFT;
1244
1245         IXGBE_WRITE_REG(hw, IXGBE_VFSRRCTL(index), srrctl);
1246 }
1247
1248 static void ixgbevf_setup_psrtype(struct ixgbevf_adapter *adapter)
1249 {
1250         struct ixgbe_hw *hw = &adapter->hw;
1251
1252         /* PSRTYPE must be initialized in 82599 */
1253         u32 psrtype = IXGBE_PSRTYPE_TCPHDR | IXGBE_PSRTYPE_UDPHDR |
1254                       IXGBE_PSRTYPE_IPV4HDR | IXGBE_PSRTYPE_IPV6HDR |
1255                       IXGBE_PSRTYPE_L2HDR;
1256
1257         if (adapter->num_rx_queues > 1)
1258                 psrtype |= 1 << 29;
1259
1260         IXGBE_WRITE_REG(hw, IXGBE_VFPSRTYPE, psrtype);
1261 }
1262
1263 static void ixgbevf_set_rx_buffer_len(struct ixgbevf_adapter *adapter)
1264 {
1265         struct ixgbe_hw *hw = &adapter->hw;
1266         struct net_device *netdev = adapter->netdev;
1267         int max_frame = netdev->mtu + ETH_HLEN + ETH_FCS_LEN;
1268         int i;
1269         u16 rx_buf_len;
1270
1271         /* notify the PF of our intent to use this size of frame */
1272         ixgbevf_rlpml_set_vf(hw, max_frame);
1273
1274         /* PF will allow an extra 4 bytes past for vlan tagged frames */
1275         max_frame += VLAN_HLEN;
1276
1277         /*
1278          * Allocate buffer sizes that fit well into 32K and
1279          * take into account max frame size of 9.5K
1280          */
1281         if ((hw->mac.type == ixgbe_mac_X540_vf) &&
1282             (max_frame <= MAXIMUM_ETHERNET_VLAN_SIZE))
1283                 rx_buf_len = MAXIMUM_ETHERNET_VLAN_SIZE;
1284         else if (max_frame <= IXGBEVF_RXBUFFER_2K)
1285                 rx_buf_len = IXGBEVF_RXBUFFER_2K;
1286         else if (max_frame <= IXGBEVF_RXBUFFER_4K)
1287                 rx_buf_len = IXGBEVF_RXBUFFER_4K;
1288         else if (max_frame <= IXGBEVF_RXBUFFER_8K)
1289                 rx_buf_len = IXGBEVF_RXBUFFER_8K;
1290         else
1291                 rx_buf_len = IXGBEVF_RXBUFFER_10K;
1292
1293         for (i = 0; i < adapter->num_rx_queues; i++)
1294                 adapter->rx_ring[i]->rx_buf_len = rx_buf_len;
1295 }
1296
1297 #define IXGBEVF_MAX_RX_DESC_POLL 10
1298 static void ixgbevf_disable_rx_queue(struct ixgbevf_adapter *adapter,
1299                                      struct ixgbevf_ring *ring)
1300 {
1301         struct ixgbe_hw *hw = &adapter->hw;
1302         int wait_loop = IXGBEVF_MAX_RX_DESC_POLL;
1303         u32 rxdctl;
1304         u8 reg_idx = ring->reg_idx;
1305
1306         if (IXGBE_REMOVED(hw->hw_addr))
1307                 return;
1308         rxdctl = IXGBE_READ_REG(hw, IXGBE_VFRXDCTL(reg_idx));
1309         rxdctl &= ~IXGBE_RXDCTL_ENABLE;
1310
1311         /* write value back with RXDCTL.ENABLE bit cleared */
1312         IXGBE_WRITE_REG(hw, IXGBE_VFRXDCTL(reg_idx), rxdctl);
1313
1314         /* the hardware may take up to 100us to really disable the rx queue */
1315         do {
1316                 udelay(10);
1317                 rxdctl = IXGBE_READ_REG(hw, IXGBE_VFRXDCTL(reg_idx));
1318         } while (--wait_loop && (rxdctl & IXGBE_RXDCTL_ENABLE));
1319
1320         if (!wait_loop)
1321                 pr_err("RXDCTL.ENABLE queue %d not cleared while polling\n",
1322                        reg_idx);
1323 }
1324
1325 static void ixgbevf_rx_desc_queue_enable(struct ixgbevf_adapter *adapter,
1326                                          struct ixgbevf_ring *ring)
1327 {
1328         struct ixgbe_hw *hw = &adapter->hw;
1329         int wait_loop = IXGBEVF_MAX_RX_DESC_POLL;
1330         u32 rxdctl;
1331         u8 reg_idx = ring->reg_idx;
1332
1333         if (IXGBE_REMOVED(hw->hw_addr))
1334                 return;
1335         do {
1336                 usleep_range(1000, 2000);
1337                 rxdctl = IXGBE_READ_REG(hw, IXGBE_VFRXDCTL(reg_idx));
1338         } while (--wait_loop && !(rxdctl & IXGBE_RXDCTL_ENABLE));
1339
1340         if (!wait_loop)
1341                 pr_err("RXDCTL.ENABLE queue %d not set while polling\n",
1342                        reg_idx);
1343 }
1344
1345 static void ixgbevf_configure_rx_ring(struct ixgbevf_adapter *adapter,
1346                                       struct ixgbevf_ring *ring)
1347 {
1348         struct ixgbe_hw *hw = &adapter->hw;
1349         u64 rdba = ring->dma;
1350         u32 rxdctl;
1351         u8 reg_idx = ring->reg_idx;
1352
1353         /* disable queue to avoid issues while updating state */
1354         rxdctl = IXGBE_READ_REG(hw, IXGBE_VFRXDCTL(reg_idx));
1355         ixgbevf_disable_rx_queue(adapter, ring);
1356
1357         IXGBE_WRITE_REG(hw, IXGBE_VFRDBAL(reg_idx), rdba & DMA_BIT_MASK(32));
1358         IXGBE_WRITE_REG(hw, IXGBE_VFRDBAH(reg_idx), rdba >> 32);
1359         IXGBE_WRITE_REG(hw, IXGBE_VFRDLEN(reg_idx),
1360                         ring->count * sizeof(union ixgbe_adv_rx_desc));
1361
1362         /* enable relaxed ordering */
1363         IXGBE_WRITE_REG(hw, IXGBE_VFDCA_RXCTRL(reg_idx),
1364                         IXGBE_DCA_RXCTRL_DESC_RRO_EN);
1365
1366         /* reset head and tail pointers */
1367         IXGBE_WRITE_REG(hw, IXGBE_VFRDH(reg_idx), 0);
1368         IXGBE_WRITE_REG(hw, IXGBE_VFRDT(reg_idx), 0);
1369         ring->tail = adapter->io_addr + IXGBE_VFRDT(reg_idx);
1370
1371         /* reset ntu and ntc to place SW in sync with hardwdare */
1372         ring->next_to_clean = 0;
1373         ring->next_to_use = 0;
1374
1375         ixgbevf_configure_srrctl(adapter, reg_idx);
1376
1377         /* prevent DMA from exceeding buffer space available */
1378         rxdctl &= ~IXGBE_RXDCTL_RLPMLMASK;
1379         rxdctl |= ring->rx_buf_len | IXGBE_RXDCTL_RLPML_EN;
1380         rxdctl |= IXGBE_RXDCTL_ENABLE | IXGBE_RXDCTL_VME;
1381         IXGBE_WRITE_REG(hw, IXGBE_VFRXDCTL(reg_idx), rxdctl);
1382
1383         ixgbevf_rx_desc_queue_enable(adapter, ring);
1384         ixgbevf_alloc_rx_buffers(ring, ixgbevf_desc_unused(ring));
1385 }
1386
1387 /**
1388  * ixgbevf_configure_rx - Configure 82599 VF Receive Unit after Reset
1389  * @adapter: board private structure
1390  *
1391  * Configure the Rx unit of the MAC after a reset.
1392  **/
1393 static void ixgbevf_configure_rx(struct ixgbevf_adapter *adapter)
1394 {
1395         int i;
1396
1397         ixgbevf_setup_psrtype(adapter);
1398
1399         /* set_rx_buffer_len must be called before ring initialization */
1400         ixgbevf_set_rx_buffer_len(adapter);
1401
1402         /* Setup the HW Rx Head and Tail Descriptor Pointers and
1403          * the Base and Length of the Rx Descriptor Ring */
1404         for (i = 0; i < adapter->num_rx_queues; i++)
1405                 ixgbevf_configure_rx_ring(adapter, adapter->rx_ring[i]);
1406 }
1407
1408 static int ixgbevf_vlan_rx_add_vid(struct net_device *netdev,
1409                                    __be16 proto, u16 vid)
1410 {
1411         struct ixgbevf_adapter *adapter = netdev_priv(netdev);
1412         struct ixgbe_hw *hw = &adapter->hw;
1413         int err;
1414
1415         spin_lock_bh(&adapter->mbx_lock);
1416
1417         /* add VID to filter table */
1418         err = hw->mac.ops.set_vfta(hw, vid, 0, true);
1419
1420         spin_unlock_bh(&adapter->mbx_lock);
1421
1422         /* translate error return types so error makes sense */
1423         if (err == IXGBE_ERR_MBX)
1424                 return -EIO;
1425
1426         if (err == IXGBE_ERR_INVALID_ARGUMENT)
1427                 return -EACCES;
1428
1429         set_bit(vid, adapter->active_vlans);
1430
1431         return err;
1432 }
1433
1434 static int ixgbevf_vlan_rx_kill_vid(struct net_device *netdev,
1435                                     __be16 proto, u16 vid)
1436 {
1437         struct ixgbevf_adapter *adapter = netdev_priv(netdev);
1438         struct ixgbe_hw *hw = &adapter->hw;
1439         int err = -EOPNOTSUPP;
1440
1441         spin_lock_bh(&adapter->mbx_lock);
1442
1443         /* remove VID from filter table */
1444         err = hw->mac.ops.set_vfta(hw, vid, 0, false);
1445
1446         spin_unlock_bh(&adapter->mbx_lock);
1447
1448         clear_bit(vid, adapter->active_vlans);
1449
1450         return err;
1451 }
1452
1453 static void ixgbevf_restore_vlan(struct ixgbevf_adapter *adapter)
1454 {
1455         u16 vid;
1456
1457         for_each_set_bit(vid, adapter->active_vlans, VLAN_N_VID)
1458                 ixgbevf_vlan_rx_add_vid(adapter->netdev,
1459                                         htons(ETH_P_8021Q), vid);
1460 }
1461
1462 static int ixgbevf_write_uc_addr_list(struct net_device *netdev)
1463 {
1464         struct ixgbevf_adapter *adapter = netdev_priv(netdev);
1465         struct ixgbe_hw *hw = &adapter->hw;
1466         int count = 0;
1467
1468         if ((netdev_uc_count(netdev)) > 10) {
1469                 pr_err("Too many unicast filters - No Space\n");
1470                 return -ENOSPC;
1471         }
1472
1473         if (!netdev_uc_empty(netdev)) {
1474                 struct netdev_hw_addr *ha;
1475                 netdev_for_each_uc_addr(ha, netdev) {
1476                         hw->mac.ops.set_uc_addr(hw, ++count, ha->addr);
1477                         udelay(200);
1478                 }
1479         } else {
1480                 /*
1481                  * If the list is empty then send message to PF driver to
1482                  * clear all macvlans on this VF.
1483                  */
1484                 hw->mac.ops.set_uc_addr(hw, 0, NULL);
1485         }
1486
1487         return count;
1488 }
1489
1490 /**
1491  * ixgbevf_set_rx_mode - Multicast and unicast set
1492  * @netdev: network interface device structure
1493  *
1494  * The set_rx_method entry point is called whenever the multicast address
1495  * list, unicast address list or the network interface flags are updated.
1496  * This routine is responsible for configuring the hardware for proper
1497  * multicast mode and configuring requested unicast filters.
1498  **/
1499 static void ixgbevf_set_rx_mode(struct net_device *netdev)
1500 {
1501         struct ixgbevf_adapter *adapter = netdev_priv(netdev);
1502         struct ixgbe_hw *hw = &adapter->hw;
1503
1504         spin_lock_bh(&adapter->mbx_lock);
1505
1506         /* reprogram multicast list */
1507         hw->mac.ops.update_mc_addr_list(hw, netdev);
1508
1509         ixgbevf_write_uc_addr_list(netdev);
1510
1511         spin_unlock_bh(&adapter->mbx_lock);
1512 }
1513
1514 static void ixgbevf_napi_enable_all(struct ixgbevf_adapter *adapter)
1515 {
1516         int q_idx;
1517         struct ixgbevf_q_vector *q_vector;
1518         int q_vectors = adapter->num_msix_vectors - NON_Q_VECTORS;
1519
1520         for (q_idx = 0; q_idx < q_vectors; q_idx++) {
1521                 q_vector = adapter->q_vector[q_idx];
1522 #ifdef CONFIG_NET_RX_BUSY_POLL
1523                 ixgbevf_qv_init_lock(adapter->q_vector[q_idx]);
1524 #endif
1525                 napi_enable(&q_vector->napi);
1526         }
1527 }
1528
1529 static void ixgbevf_napi_disable_all(struct ixgbevf_adapter *adapter)
1530 {
1531         int q_idx;
1532         struct ixgbevf_q_vector *q_vector;
1533         int q_vectors = adapter->num_msix_vectors - NON_Q_VECTORS;
1534
1535         for (q_idx = 0; q_idx < q_vectors; q_idx++) {
1536                 q_vector = adapter->q_vector[q_idx];
1537                 napi_disable(&q_vector->napi);
1538 #ifdef CONFIG_NET_RX_BUSY_POLL
1539                 while (!ixgbevf_qv_disable(adapter->q_vector[q_idx])) {
1540                         pr_info("QV %d locked\n", q_idx);
1541                         usleep_range(1000, 20000);
1542                 }
1543 #endif /* CONFIG_NET_RX_BUSY_POLL */
1544         }
1545 }
1546
1547 static int ixgbevf_configure_dcb(struct ixgbevf_adapter *adapter)
1548 {
1549         struct ixgbe_hw *hw = &adapter->hw;
1550         unsigned int def_q = 0;
1551         unsigned int num_tcs = 0;
1552         unsigned int num_rx_queues = 1;
1553         int err;
1554
1555         spin_lock_bh(&adapter->mbx_lock);
1556
1557         /* fetch queue configuration from the PF */
1558         err = ixgbevf_get_queues(hw, &num_tcs, &def_q);
1559
1560         spin_unlock_bh(&adapter->mbx_lock);
1561
1562         if (err)
1563                 return err;
1564
1565         if (num_tcs > 1) {
1566                 /* update default Tx ring register index */
1567                 adapter->tx_ring[0]->reg_idx = def_q;
1568
1569                 /* we need as many queues as traffic classes */
1570                 num_rx_queues = num_tcs;
1571         }
1572
1573         /* if we have a bad config abort request queue reset */
1574         if (adapter->num_rx_queues != num_rx_queues) {
1575                 /* force mailbox timeout to prevent further messages */
1576                 hw->mbx.timeout = 0;
1577
1578                 /* wait for watchdog to come around and bail us out */
1579                 adapter->flags |= IXGBEVF_FLAG_QUEUE_RESET_REQUESTED;
1580         }
1581
1582         return 0;
1583 }
1584
1585 static void ixgbevf_configure(struct ixgbevf_adapter *adapter)
1586 {
1587         ixgbevf_configure_dcb(adapter);
1588
1589         ixgbevf_set_rx_mode(adapter->netdev);
1590
1591         ixgbevf_restore_vlan(adapter);
1592
1593         ixgbevf_configure_tx(adapter);
1594         ixgbevf_configure_rx(adapter);
1595 }
1596
1597 static void ixgbevf_save_reset_stats(struct ixgbevf_adapter *adapter)
1598 {
1599         /* Only save pre-reset stats if there are some */
1600         if (adapter->stats.vfgprc || adapter->stats.vfgptc) {
1601                 adapter->stats.saved_reset_vfgprc += adapter->stats.vfgprc -
1602                         adapter->stats.base_vfgprc;
1603                 adapter->stats.saved_reset_vfgptc += adapter->stats.vfgptc -
1604                         adapter->stats.base_vfgptc;
1605                 adapter->stats.saved_reset_vfgorc += adapter->stats.vfgorc -
1606                         adapter->stats.base_vfgorc;
1607                 adapter->stats.saved_reset_vfgotc += adapter->stats.vfgotc -
1608                         adapter->stats.base_vfgotc;
1609                 adapter->stats.saved_reset_vfmprc += adapter->stats.vfmprc -
1610                         adapter->stats.base_vfmprc;
1611         }
1612 }
1613
1614 static void ixgbevf_init_last_counter_stats(struct ixgbevf_adapter *adapter)
1615 {
1616         struct ixgbe_hw *hw = &adapter->hw;
1617
1618         adapter->stats.last_vfgprc = IXGBE_READ_REG(hw, IXGBE_VFGPRC);
1619         adapter->stats.last_vfgorc = IXGBE_READ_REG(hw, IXGBE_VFGORC_LSB);
1620         adapter->stats.last_vfgorc |=
1621                 (((u64)(IXGBE_READ_REG(hw, IXGBE_VFGORC_MSB))) << 32);
1622         adapter->stats.last_vfgptc = IXGBE_READ_REG(hw, IXGBE_VFGPTC);
1623         adapter->stats.last_vfgotc = IXGBE_READ_REG(hw, IXGBE_VFGOTC_LSB);
1624         adapter->stats.last_vfgotc |=
1625                 (((u64)(IXGBE_READ_REG(hw, IXGBE_VFGOTC_MSB))) << 32);
1626         adapter->stats.last_vfmprc = IXGBE_READ_REG(hw, IXGBE_VFMPRC);
1627
1628         adapter->stats.base_vfgprc = adapter->stats.last_vfgprc;
1629         adapter->stats.base_vfgorc = adapter->stats.last_vfgorc;
1630         adapter->stats.base_vfgptc = adapter->stats.last_vfgptc;
1631         adapter->stats.base_vfgotc = adapter->stats.last_vfgotc;
1632         adapter->stats.base_vfmprc = adapter->stats.last_vfmprc;
1633 }
1634
1635 static void ixgbevf_negotiate_api(struct ixgbevf_adapter *adapter)
1636 {
1637         struct ixgbe_hw *hw = &adapter->hw;
1638         int api[] = { ixgbe_mbox_api_11,
1639                       ixgbe_mbox_api_10,
1640                       ixgbe_mbox_api_unknown };
1641         int err = 0, idx = 0;
1642
1643         spin_lock_bh(&adapter->mbx_lock);
1644
1645         while (api[idx] != ixgbe_mbox_api_unknown) {
1646                 err = ixgbevf_negotiate_api_version(hw, api[idx]);
1647                 if (!err)
1648                         break;
1649                 idx++;
1650         }
1651
1652         spin_unlock_bh(&adapter->mbx_lock);
1653 }
1654
1655 static void ixgbevf_up_complete(struct ixgbevf_adapter *adapter)
1656 {
1657         struct net_device *netdev = adapter->netdev;
1658         struct ixgbe_hw *hw = &adapter->hw;
1659
1660         ixgbevf_configure_msix(adapter);
1661
1662         spin_lock_bh(&adapter->mbx_lock);
1663
1664         if (is_valid_ether_addr(hw->mac.addr))
1665                 hw->mac.ops.set_rar(hw, 0, hw->mac.addr, 0);
1666         else
1667                 hw->mac.ops.set_rar(hw, 0, hw->mac.perm_addr, 0);
1668
1669         spin_unlock_bh(&adapter->mbx_lock);
1670
1671         smp_mb__before_atomic();
1672         clear_bit(__IXGBEVF_DOWN, &adapter->state);
1673         ixgbevf_napi_enable_all(adapter);
1674
1675         /* enable transmits */
1676         netif_tx_start_all_queues(netdev);
1677
1678         ixgbevf_save_reset_stats(adapter);
1679         ixgbevf_init_last_counter_stats(adapter);
1680
1681         hw->mac.get_link_status = 1;
1682         mod_timer(&adapter->watchdog_timer, jiffies);
1683 }
1684
1685 void ixgbevf_up(struct ixgbevf_adapter *adapter)
1686 {
1687         struct ixgbe_hw *hw = &adapter->hw;
1688
1689         ixgbevf_configure(adapter);
1690
1691         ixgbevf_up_complete(adapter);
1692
1693         /* clear any pending interrupts, may auto mask */
1694         IXGBE_READ_REG(hw, IXGBE_VTEICR);
1695
1696         ixgbevf_irq_enable(adapter);
1697 }
1698
1699 /**
1700  * ixgbevf_clean_rx_ring - Free Rx Buffers per Queue
1701  * @rx_ring: ring to free buffers from
1702  **/
1703 static void ixgbevf_clean_rx_ring(struct ixgbevf_ring *rx_ring)
1704 {
1705         unsigned long size;
1706         unsigned int i;
1707
1708         if (!rx_ring->rx_buffer_info)
1709                 return;
1710
1711         /* Free all the Rx ring sk_buffs */
1712         for (i = 0; i < rx_ring->count; i++) {
1713                 struct ixgbevf_rx_buffer *rx_buffer_info;
1714
1715                 rx_buffer_info = &rx_ring->rx_buffer_info[i];
1716                 if (rx_buffer_info->dma) {
1717                         dma_unmap_single(rx_ring->dev, rx_buffer_info->dma,
1718                                          rx_ring->rx_buf_len,
1719                                          DMA_FROM_DEVICE);
1720                         rx_buffer_info->dma = 0;
1721                 }
1722                 if (rx_buffer_info->skb) {
1723                         struct sk_buff *skb = rx_buffer_info->skb;
1724                         rx_buffer_info->skb = NULL;
1725                         do {
1726                                 struct sk_buff *this = skb;
1727                                 skb = IXGBE_CB(skb)->prev;
1728                                 dev_kfree_skb(this);
1729                         } while (skb);
1730                 }
1731         }
1732
1733         size = sizeof(struct ixgbevf_rx_buffer) * rx_ring->count;
1734         memset(rx_ring->rx_buffer_info, 0, size);
1735
1736         /* Zero out the descriptor ring */
1737         memset(rx_ring->desc, 0, rx_ring->size);
1738 }
1739
1740 /**
1741  * ixgbevf_clean_tx_ring - Free Tx Buffers
1742  * @tx_ring: ring to be cleaned
1743  **/
1744 static void ixgbevf_clean_tx_ring(struct ixgbevf_ring *tx_ring)
1745 {
1746         struct ixgbevf_tx_buffer *tx_buffer_info;
1747         unsigned long size;
1748         unsigned int i;
1749
1750         if (!tx_ring->tx_buffer_info)
1751                 return;
1752
1753         /* Free all the Tx ring sk_buffs */
1754         for (i = 0; i < tx_ring->count; i++) {
1755                 tx_buffer_info = &tx_ring->tx_buffer_info[i];
1756                 ixgbevf_unmap_and_free_tx_resource(tx_ring, tx_buffer_info);
1757         }
1758
1759         size = sizeof(struct ixgbevf_tx_buffer) * tx_ring->count;
1760         memset(tx_ring->tx_buffer_info, 0, size);
1761
1762         memset(tx_ring->desc, 0, tx_ring->size);
1763 }
1764
1765 /**
1766  * ixgbevf_clean_all_rx_rings - Free Rx Buffers for all queues
1767  * @adapter: board private structure
1768  **/
1769 static void ixgbevf_clean_all_rx_rings(struct ixgbevf_adapter *adapter)
1770 {
1771         int i;
1772
1773         for (i = 0; i < adapter->num_rx_queues; i++)
1774                 ixgbevf_clean_rx_ring(adapter->rx_ring[i]);
1775 }
1776
1777 /**
1778  * ixgbevf_clean_all_tx_rings - Free Tx Buffers for all queues
1779  * @adapter: board private structure
1780  **/
1781 static void ixgbevf_clean_all_tx_rings(struct ixgbevf_adapter *adapter)
1782 {
1783         int i;
1784
1785         for (i = 0; i < adapter->num_tx_queues; i++)
1786                 ixgbevf_clean_tx_ring(adapter->tx_ring[i]);
1787 }
1788
1789 void ixgbevf_down(struct ixgbevf_adapter *adapter)
1790 {
1791         struct net_device *netdev = adapter->netdev;
1792         struct ixgbe_hw *hw = &adapter->hw;
1793         int i;
1794
1795         /* signal that we are down to the interrupt handler */
1796         if (test_and_set_bit(__IXGBEVF_DOWN, &adapter->state))
1797                 return; /* do nothing if already down */
1798
1799         /* disable all enabled rx queues */
1800         for (i = 0; i < adapter->num_rx_queues; i++)
1801                 ixgbevf_disable_rx_queue(adapter, adapter->rx_ring[i]);
1802
1803         netif_tx_disable(netdev);
1804
1805         msleep(10);
1806
1807         netif_tx_stop_all_queues(netdev);
1808
1809         ixgbevf_irq_disable(adapter);
1810
1811         ixgbevf_napi_disable_all(adapter);
1812
1813         del_timer_sync(&adapter->watchdog_timer);
1814         /* can't call flush scheduled work here because it can deadlock
1815          * if linkwatch_event tries to acquire the rtnl_lock which we are
1816          * holding */
1817         while (adapter->flags & IXGBE_FLAG_IN_WATCHDOG_TASK)
1818                 msleep(1);
1819
1820         /* disable transmits in the hardware now that interrupts are off */
1821         for (i = 0; i < adapter->num_tx_queues; i++) {
1822                 u8 reg_idx = adapter->tx_ring[i]->reg_idx;
1823
1824                 IXGBE_WRITE_REG(hw, IXGBE_VFTXDCTL(reg_idx),
1825                                 IXGBE_TXDCTL_SWFLSH);
1826         }
1827
1828         netif_carrier_off(netdev);
1829
1830         if (!pci_channel_offline(adapter->pdev))
1831                 ixgbevf_reset(adapter);
1832
1833         ixgbevf_clean_all_tx_rings(adapter);
1834         ixgbevf_clean_all_rx_rings(adapter);
1835 }
1836
1837 void ixgbevf_reinit_locked(struct ixgbevf_adapter *adapter)
1838 {
1839         WARN_ON(in_interrupt());
1840
1841         while (test_and_set_bit(__IXGBEVF_RESETTING, &adapter->state))
1842                 msleep(1);
1843
1844         ixgbevf_down(adapter);
1845         ixgbevf_up(adapter);
1846
1847         clear_bit(__IXGBEVF_RESETTING, &adapter->state);
1848 }
1849
1850 void ixgbevf_reset(struct ixgbevf_adapter *adapter)
1851 {
1852         struct ixgbe_hw *hw = &adapter->hw;
1853         struct net_device *netdev = adapter->netdev;
1854
1855         if (hw->mac.ops.reset_hw(hw)) {
1856                 hw_dbg(hw, "PF still resetting\n");
1857         } else {
1858                 hw->mac.ops.init_hw(hw);
1859                 ixgbevf_negotiate_api(adapter);
1860         }
1861
1862         if (is_valid_ether_addr(adapter->hw.mac.addr)) {
1863                 memcpy(netdev->dev_addr, adapter->hw.mac.addr,
1864                        netdev->addr_len);
1865                 memcpy(netdev->perm_addr, adapter->hw.mac.addr,
1866                        netdev->addr_len);
1867         }
1868 }
1869
1870 static int ixgbevf_acquire_msix_vectors(struct ixgbevf_adapter *adapter,
1871                                         int vectors)
1872 {
1873         int vector_threshold;
1874
1875         /* We'll want at least 2 (vector_threshold):
1876          * 1) TxQ[0] + RxQ[0] handler
1877          * 2) Other (Link Status Change, etc.)
1878          */
1879         vector_threshold = MIN_MSIX_COUNT;
1880
1881         /* The more we get, the more we will assign to Tx/Rx Cleanup
1882          * for the separate queues...where Rx Cleanup >= Tx Cleanup.
1883          * Right now, we simply care about how many we'll get; we'll
1884          * set them up later while requesting irq's.
1885          */
1886         vectors = pci_enable_msix_range(adapter->pdev, adapter->msix_entries,
1887                                         vector_threshold, vectors);
1888
1889         if (vectors < 0) {
1890                 dev_err(&adapter->pdev->dev,
1891                         "Unable to allocate MSI-X interrupts\n");
1892                 kfree(adapter->msix_entries);
1893                 adapter->msix_entries = NULL;
1894                 return vectors;
1895         }
1896
1897         /* Adjust for only the vectors we'll use, which is minimum
1898          * of max_msix_q_vectors + NON_Q_VECTORS, or the number of
1899          * vectors we were allocated.
1900          */
1901         adapter->num_msix_vectors = vectors;
1902
1903         return 0;
1904 }
1905
1906 /**
1907  * ixgbevf_set_num_queues - Allocate queues for device, feature dependent
1908  * @adapter: board private structure to initialize
1909  *
1910  * This is the top level queue allocation routine.  The order here is very
1911  * important, starting with the "most" number of features turned on at once,
1912  * and ending with the smallest set of features.  This way large combinations
1913  * can be allocated if they're turned on, and smaller combinations are the
1914  * fallthrough conditions.
1915  *
1916  **/
1917 static void ixgbevf_set_num_queues(struct ixgbevf_adapter *adapter)
1918 {
1919         struct ixgbe_hw *hw = &adapter->hw;
1920         unsigned int def_q = 0;
1921         unsigned int num_tcs = 0;
1922         int err;
1923
1924         /* Start with base case */
1925         adapter->num_rx_queues = 1;
1926         adapter->num_tx_queues = 1;
1927
1928         spin_lock_bh(&adapter->mbx_lock);
1929
1930         /* fetch queue configuration from the PF */
1931         err = ixgbevf_get_queues(hw, &num_tcs, &def_q);
1932
1933         spin_unlock_bh(&adapter->mbx_lock);
1934
1935         if (err)
1936                 return;
1937
1938         /* we need as many queues as traffic classes */
1939         if (num_tcs > 1)
1940                 adapter->num_rx_queues = num_tcs;
1941 }
1942
1943 /**
1944  * ixgbevf_alloc_queues - Allocate memory for all rings
1945  * @adapter: board private structure to initialize
1946  *
1947  * We allocate one ring per queue at run-time since we don't know the
1948  * number of queues at compile-time.  The polling_netdev array is
1949  * intended for Multiqueue, but should work fine with a single queue.
1950  **/
1951 static int ixgbevf_alloc_queues(struct ixgbevf_adapter *adapter)
1952 {
1953         struct ixgbevf_ring *ring;
1954         int rx = 0, tx = 0;
1955
1956         for (; tx < adapter->num_tx_queues; tx++) {
1957                 ring = kzalloc(sizeof(*ring), GFP_KERNEL);
1958                 if (!ring)
1959                         goto err_allocation;
1960
1961                 ring->dev = &adapter->pdev->dev;
1962                 ring->netdev = adapter->netdev;
1963                 ring->count = adapter->tx_ring_count;
1964                 ring->queue_index = tx;
1965                 ring->reg_idx = tx;
1966
1967                 adapter->tx_ring[tx] = ring;
1968         }
1969
1970         for (; rx < adapter->num_rx_queues; rx++) {
1971                 ring = kzalloc(sizeof(*ring), GFP_KERNEL);
1972                 if (!ring)
1973                         goto err_allocation;
1974
1975                 ring->dev = &adapter->pdev->dev;
1976                 ring->netdev = adapter->netdev;
1977
1978                 ring->count = adapter->rx_ring_count;
1979                 ring->queue_index = rx;
1980                 ring->reg_idx = rx;
1981
1982                 adapter->rx_ring[rx] = ring;
1983         }
1984
1985         return 0;
1986
1987 err_allocation:
1988         while (tx) {
1989                 kfree(adapter->tx_ring[--tx]);
1990                 adapter->tx_ring[tx] = NULL;
1991         }
1992
1993         while (rx) {
1994                 kfree(adapter->rx_ring[--rx]);
1995                 adapter->rx_ring[rx] = NULL;
1996         }
1997         return -ENOMEM;
1998 }
1999
2000 /**
2001  * ixgbevf_set_interrupt_capability - set MSI-X or FAIL if not supported
2002  * @adapter: board private structure to initialize
2003  *
2004  * Attempt to configure the interrupts using the best available
2005  * capabilities of the hardware and the kernel.
2006  **/
2007 static int ixgbevf_set_interrupt_capability(struct ixgbevf_adapter *adapter)
2008 {
2009         struct net_device *netdev = adapter->netdev;
2010         int err = 0;
2011         int vector, v_budget;
2012
2013         /*
2014          * It's easy to be greedy for MSI-X vectors, but it really
2015          * doesn't do us much good if we have a lot more vectors
2016          * than CPU's.  So let's be conservative and only ask for
2017          * (roughly) the same number of vectors as there are CPU's.
2018          * The default is to use pairs of vectors.
2019          */
2020         v_budget = max(adapter->num_rx_queues, adapter->num_tx_queues);
2021         v_budget = min_t(int, v_budget, num_online_cpus());
2022         v_budget += NON_Q_VECTORS;
2023
2024         /* A failure in MSI-X entry allocation isn't fatal, but it does
2025          * mean we disable MSI-X capabilities of the adapter. */
2026         adapter->msix_entries = kcalloc(v_budget,
2027                                         sizeof(struct msix_entry), GFP_KERNEL);
2028         if (!adapter->msix_entries) {
2029                 err = -ENOMEM;
2030                 goto out;
2031         }
2032
2033         for (vector = 0; vector < v_budget; vector++)
2034                 adapter->msix_entries[vector].entry = vector;
2035
2036         err = ixgbevf_acquire_msix_vectors(adapter, v_budget);
2037         if (err)
2038                 goto out;
2039
2040         err = netif_set_real_num_tx_queues(netdev, adapter->num_tx_queues);
2041         if (err)
2042                 goto out;
2043
2044         err = netif_set_real_num_rx_queues(netdev, adapter->num_rx_queues);
2045
2046 out:
2047         return err;
2048 }
2049
2050 /**
2051  * ixgbevf_alloc_q_vectors - Allocate memory for interrupt vectors
2052  * @adapter: board private structure to initialize
2053  *
2054  * We allocate one q_vector per queue interrupt.  If allocation fails we
2055  * return -ENOMEM.
2056  **/
2057 static int ixgbevf_alloc_q_vectors(struct ixgbevf_adapter *adapter)
2058 {
2059         int q_idx, num_q_vectors;
2060         struct ixgbevf_q_vector *q_vector;
2061
2062         num_q_vectors = adapter->num_msix_vectors - NON_Q_VECTORS;
2063
2064         for (q_idx = 0; q_idx < num_q_vectors; q_idx++) {
2065                 q_vector = kzalloc(sizeof(struct ixgbevf_q_vector), GFP_KERNEL);
2066                 if (!q_vector)
2067                         goto err_out;
2068                 q_vector->adapter = adapter;
2069                 q_vector->v_idx = q_idx;
2070                 netif_napi_add(adapter->netdev, &q_vector->napi,
2071                                ixgbevf_poll, 64);
2072 #ifdef CONFIG_NET_RX_BUSY_POLL
2073                 napi_hash_add(&q_vector->napi);
2074 #endif
2075                 adapter->q_vector[q_idx] = q_vector;
2076         }
2077
2078         return 0;
2079
2080 err_out:
2081         while (q_idx) {
2082                 q_idx--;
2083                 q_vector = adapter->q_vector[q_idx];
2084 #ifdef CONFIG_NET_RX_BUSY_POLL
2085                 napi_hash_del(&q_vector->napi);
2086 #endif
2087                 netif_napi_del(&q_vector->napi);
2088                 kfree(q_vector);
2089                 adapter->q_vector[q_idx] = NULL;
2090         }
2091         return -ENOMEM;
2092 }
2093
2094 /**
2095  * ixgbevf_free_q_vectors - Free memory allocated for interrupt vectors
2096  * @adapter: board private structure to initialize
2097  *
2098  * This function frees the memory allocated to the q_vectors.  In addition if
2099  * NAPI is enabled it will delete any references to the NAPI struct prior
2100  * to freeing the q_vector.
2101  **/
2102 static void ixgbevf_free_q_vectors(struct ixgbevf_adapter *adapter)
2103 {
2104         int q_idx, num_q_vectors = adapter->num_msix_vectors - NON_Q_VECTORS;
2105
2106         for (q_idx = 0; q_idx < num_q_vectors; q_idx++) {
2107                 struct ixgbevf_q_vector *q_vector = adapter->q_vector[q_idx];
2108
2109                 adapter->q_vector[q_idx] = NULL;
2110 #ifdef CONFIG_NET_RX_BUSY_POLL
2111                 napi_hash_del(&q_vector->napi);
2112 #endif
2113                 netif_napi_del(&q_vector->napi);
2114                 kfree(q_vector);
2115         }
2116 }
2117
2118 /**
2119  * ixgbevf_reset_interrupt_capability - Reset MSIX setup
2120  * @adapter: board private structure
2121  *
2122  **/
2123 static void ixgbevf_reset_interrupt_capability(struct ixgbevf_adapter *adapter)
2124 {
2125         pci_disable_msix(adapter->pdev);
2126         kfree(adapter->msix_entries);
2127         adapter->msix_entries = NULL;
2128 }
2129
2130 /**
2131  * ixgbevf_init_interrupt_scheme - Determine if MSIX is supported and init
2132  * @adapter: board private structure to initialize
2133  *
2134  **/
2135 static int ixgbevf_init_interrupt_scheme(struct ixgbevf_adapter *adapter)
2136 {
2137         int err;
2138
2139         /* Number of supported queues */
2140         ixgbevf_set_num_queues(adapter);
2141
2142         err = ixgbevf_set_interrupt_capability(adapter);
2143         if (err) {
2144                 hw_dbg(&adapter->hw,
2145                        "Unable to setup interrupt capabilities\n");
2146                 goto err_set_interrupt;
2147         }
2148
2149         err = ixgbevf_alloc_q_vectors(adapter);
2150         if (err) {
2151                 hw_dbg(&adapter->hw, "Unable to allocate memory for queue "
2152                        "vectors\n");
2153                 goto err_alloc_q_vectors;
2154         }
2155
2156         err = ixgbevf_alloc_queues(adapter);
2157         if (err) {
2158                 pr_err("Unable to allocate memory for queues\n");
2159                 goto err_alloc_queues;
2160         }
2161
2162         hw_dbg(&adapter->hw, "Multiqueue %s: Rx Queue count = %u, "
2163                "Tx Queue count = %u\n",
2164                (adapter->num_rx_queues > 1) ? "Enabled" :
2165                "Disabled", adapter->num_rx_queues, adapter->num_tx_queues);
2166
2167         set_bit(__IXGBEVF_DOWN, &adapter->state);
2168
2169         return 0;
2170 err_alloc_queues:
2171         ixgbevf_free_q_vectors(adapter);
2172 err_alloc_q_vectors:
2173         ixgbevf_reset_interrupt_capability(adapter);
2174 err_set_interrupt:
2175         return err;
2176 }
2177
2178 /**
2179  * ixgbevf_clear_interrupt_scheme - Clear the current interrupt scheme settings
2180  * @adapter: board private structure to clear interrupt scheme on
2181  *
2182  * We go through and clear interrupt specific resources and reset the structure
2183  * to pre-load conditions
2184  **/
2185 static void ixgbevf_clear_interrupt_scheme(struct ixgbevf_adapter *adapter)
2186 {
2187         int i;
2188
2189         for (i = 0; i < adapter->num_tx_queues; i++) {
2190                 kfree(adapter->tx_ring[i]);
2191                 adapter->tx_ring[i] = NULL;
2192         }
2193         for (i = 0; i < adapter->num_rx_queues; i++) {
2194                 kfree(adapter->rx_ring[i]);
2195                 adapter->rx_ring[i] = NULL;
2196         }
2197
2198         adapter->num_tx_queues = 0;
2199         adapter->num_rx_queues = 0;
2200
2201         ixgbevf_free_q_vectors(adapter);
2202         ixgbevf_reset_interrupt_capability(adapter);
2203 }
2204
2205 /**
2206  * ixgbevf_sw_init - Initialize general software structures
2207  * (struct ixgbevf_adapter)
2208  * @adapter: board private structure to initialize
2209  *
2210  * ixgbevf_sw_init initializes the Adapter private data structure.
2211  * Fields are initialized based on PCI device information and
2212  * OS network device settings (MTU size).
2213  **/
2214 static int ixgbevf_sw_init(struct ixgbevf_adapter *adapter)
2215 {
2216         struct ixgbe_hw *hw = &adapter->hw;
2217         struct pci_dev *pdev = adapter->pdev;
2218         struct net_device *netdev = adapter->netdev;
2219         int err;
2220
2221         /* PCI config space info */
2222
2223         hw->vendor_id = pdev->vendor;
2224         hw->device_id = pdev->device;
2225         hw->revision_id = pdev->revision;
2226         hw->subsystem_vendor_id = pdev->subsystem_vendor;
2227         hw->subsystem_device_id = pdev->subsystem_device;
2228
2229         hw->mbx.ops.init_params(hw);
2230
2231         /* assume legacy case in which PF would only give VF 2 queues */
2232         hw->mac.max_tx_queues = 2;
2233         hw->mac.max_rx_queues = 2;
2234
2235         /* lock to protect mailbox accesses */
2236         spin_lock_init(&adapter->mbx_lock);
2237
2238         err = hw->mac.ops.reset_hw(hw);
2239         if (err) {
2240                 dev_info(&pdev->dev,
2241                          "PF still in reset state.  Is the PF interface up?\n");
2242         } else {
2243                 err = hw->mac.ops.init_hw(hw);
2244                 if (err) {
2245                         pr_err("init_shared_code failed: %d\n", err);
2246                         goto out;
2247                 }
2248                 ixgbevf_negotiate_api(adapter);
2249                 err = hw->mac.ops.get_mac_addr(hw, hw->mac.addr);
2250                 if (err)
2251                         dev_info(&pdev->dev, "Error reading MAC address\n");
2252                 else if (is_zero_ether_addr(adapter->hw.mac.addr))
2253                         dev_info(&pdev->dev,
2254                                  "MAC address not assigned by administrator.\n");
2255                 memcpy(netdev->dev_addr, hw->mac.addr, netdev->addr_len);
2256         }
2257
2258         if (!is_valid_ether_addr(netdev->dev_addr)) {
2259                 dev_info(&pdev->dev, "Assigning random MAC address\n");
2260                 eth_hw_addr_random(netdev);
2261                 memcpy(hw->mac.addr, netdev->dev_addr, netdev->addr_len);
2262         }
2263
2264         /* Enable dynamic interrupt throttling rates */
2265         adapter->rx_itr_setting = 1;
2266         adapter->tx_itr_setting = 1;
2267
2268         /* set default ring sizes */
2269         adapter->tx_ring_count = IXGBEVF_DEFAULT_TXD;
2270         adapter->rx_ring_count = IXGBEVF_DEFAULT_RXD;
2271
2272         set_bit(__IXGBEVF_DOWN, &adapter->state);
2273         return 0;
2274
2275 out:
2276         return err;
2277 }
2278
2279 #define UPDATE_VF_COUNTER_32bit(reg, last_counter, counter)     \
2280         {                                                       \
2281                 u32 current_counter = IXGBE_READ_REG(hw, reg);  \
2282                 if (current_counter < last_counter)             \
2283                         counter += 0x100000000LL;               \
2284                 last_counter = current_counter;                 \
2285                 counter &= 0xFFFFFFFF00000000LL;                \
2286                 counter |= current_counter;                     \
2287         }
2288
2289 #define UPDATE_VF_COUNTER_36bit(reg_lsb, reg_msb, last_counter, counter) \
2290         {                                                                \
2291                 u64 current_counter_lsb = IXGBE_READ_REG(hw, reg_lsb);   \
2292                 u64 current_counter_msb = IXGBE_READ_REG(hw, reg_msb);   \
2293                 u64 current_counter = (current_counter_msb << 32) |      \
2294                         current_counter_lsb;                             \
2295                 if (current_counter < last_counter)                      \
2296                         counter += 0x1000000000LL;                       \
2297                 last_counter = current_counter;                          \
2298                 counter &= 0xFFFFFFF000000000LL;                         \
2299                 counter |= current_counter;                              \
2300         }
2301 /**
2302  * ixgbevf_update_stats - Update the board statistics counters.
2303  * @adapter: board private structure
2304  **/
2305 void ixgbevf_update_stats(struct ixgbevf_adapter *adapter)
2306 {
2307         struct ixgbe_hw *hw = &adapter->hw;
2308         int i;
2309
2310         if (!adapter->link_up)
2311                 return;
2312
2313         UPDATE_VF_COUNTER_32bit(IXGBE_VFGPRC, adapter->stats.last_vfgprc,
2314                                 adapter->stats.vfgprc);
2315         UPDATE_VF_COUNTER_32bit(IXGBE_VFGPTC, adapter->stats.last_vfgptc,
2316                                 adapter->stats.vfgptc);
2317         UPDATE_VF_COUNTER_36bit(IXGBE_VFGORC_LSB, IXGBE_VFGORC_MSB,
2318                                 adapter->stats.last_vfgorc,
2319                                 adapter->stats.vfgorc);
2320         UPDATE_VF_COUNTER_36bit(IXGBE_VFGOTC_LSB, IXGBE_VFGOTC_MSB,
2321                                 adapter->stats.last_vfgotc,
2322                                 adapter->stats.vfgotc);
2323         UPDATE_VF_COUNTER_32bit(IXGBE_VFMPRC, adapter->stats.last_vfmprc,
2324                                 adapter->stats.vfmprc);
2325
2326         for (i = 0;  i  < adapter->num_rx_queues;  i++) {
2327                 adapter->hw_csum_rx_error +=
2328                         adapter->rx_ring[i]->hw_csum_rx_error;
2329                 adapter->rx_ring[i]->hw_csum_rx_error = 0;
2330         }
2331 }
2332
2333 /**
2334  * ixgbevf_watchdog - Timer Call-back
2335  * @data: pointer to adapter cast into an unsigned long
2336  **/
2337 static void ixgbevf_watchdog(unsigned long data)
2338 {
2339         struct ixgbevf_adapter *adapter = (struct ixgbevf_adapter *)data;
2340         struct ixgbe_hw *hw = &adapter->hw;
2341         u32 eics = 0;
2342         int i;
2343
2344         /*
2345          * Do the watchdog outside of interrupt context due to the lovely
2346          * delays that some of the newer hardware requires
2347          */
2348
2349         if (test_bit(__IXGBEVF_DOWN, &adapter->state))
2350                 goto watchdog_short_circuit;
2351
2352         /* get one bit for every active tx/rx interrupt vector */
2353         for (i = 0; i < adapter->num_msix_vectors - NON_Q_VECTORS; i++) {
2354                 struct ixgbevf_q_vector *qv = adapter->q_vector[i];
2355                 if (qv->rx.ring || qv->tx.ring)
2356                         eics |= 1 << i;
2357         }
2358
2359         IXGBE_WRITE_REG(hw, IXGBE_VTEICS, eics);
2360
2361 watchdog_short_circuit:
2362         schedule_work(&adapter->watchdog_task);
2363 }
2364
2365 /**
2366  * ixgbevf_tx_timeout - Respond to a Tx Hang
2367  * @netdev: network interface device structure
2368  **/
2369 static void ixgbevf_tx_timeout(struct net_device *netdev)
2370 {
2371         struct ixgbevf_adapter *adapter = netdev_priv(netdev);
2372
2373         /* Do the reset outside of interrupt context */
2374         schedule_work(&adapter->reset_task);
2375 }
2376
2377 static void ixgbevf_reset_task(struct work_struct *work)
2378 {
2379         struct ixgbevf_adapter *adapter;
2380         adapter = container_of(work, struct ixgbevf_adapter, reset_task);
2381
2382         /* If we're already down or resetting, just bail */
2383         if (test_bit(__IXGBEVF_DOWN, &adapter->state) ||
2384             test_bit(__IXGBEVF_REMOVING, &adapter->state) ||
2385             test_bit(__IXGBEVF_RESETTING, &adapter->state))
2386                 return;
2387
2388         adapter->tx_timeout_count++;
2389
2390         ixgbevf_reinit_locked(adapter);
2391 }
2392
2393 /**
2394  * ixgbevf_watchdog_task - worker thread to bring link up
2395  * @work: pointer to work_struct containing our data
2396  **/
2397 static void ixgbevf_watchdog_task(struct work_struct *work)
2398 {
2399         struct ixgbevf_adapter *adapter = container_of(work,
2400                                                        struct ixgbevf_adapter,
2401                                                        watchdog_task);
2402         struct net_device *netdev = adapter->netdev;
2403         struct ixgbe_hw *hw = &adapter->hw;
2404         u32 link_speed = adapter->link_speed;
2405         bool link_up = adapter->link_up;
2406         s32 need_reset;
2407
2408         if (IXGBE_REMOVED(hw->hw_addr)) {
2409                 if (!test_bit(__IXGBEVF_DOWN, &adapter->state)) {
2410                         rtnl_lock();
2411                         ixgbevf_down(adapter);
2412                         rtnl_unlock();
2413                 }
2414                 return;
2415         }
2416         ixgbevf_queue_reset_subtask(adapter);
2417
2418         adapter->flags |= IXGBE_FLAG_IN_WATCHDOG_TASK;
2419
2420         /*
2421          * Always check the link on the watchdog because we have
2422          * no LSC interrupt
2423          */
2424         spin_lock_bh(&adapter->mbx_lock);
2425
2426         need_reset = hw->mac.ops.check_link(hw, &link_speed, &link_up, false);
2427
2428         spin_unlock_bh(&adapter->mbx_lock);
2429
2430         if (need_reset) {
2431                 adapter->link_up = link_up;
2432                 adapter->link_speed = link_speed;
2433                 netif_carrier_off(netdev);
2434                 netif_tx_stop_all_queues(netdev);
2435                 schedule_work(&adapter->reset_task);
2436                 goto pf_has_reset;
2437         }
2438         adapter->link_up = link_up;
2439         adapter->link_speed = link_speed;
2440
2441         if (link_up) {
2442                 if (!netif_carrier_ok(netdev)) {
2443                         char *link_speed_string;
2444                         switch (link_speed) {
2445                         case IXGBE_LINK_SPEED_10GB_FULL:
2446                                 link_speed_string = "10 Gbps";
2447                                 break;
2448                         case IXGBE_LINK_SPEED_1GB_FULL:
2449                                 link_speed_string = "1 Gbps";
2450                                 break;
2451                         case IXGBE_LINK_SPEED_100_FULL:
2452                                 link_speed_string = "100 Mbps";
2453                                 break;
2454                         default:
2455                                 link_speed_string = "unknown speed";
2456                                 break;
2457                         }
2458                         dev_info(&adapter->pdev->dev,
2459                                 "NIC Link is Up, %s\n", link_speed_string);
2460                         netif_carrier_on(netdev);
2461                         netif_tx_wake_all_queues(netdev);
2462                 }
2463         } else {
2464                 adapter->link_up = false;
2465                 adapter->link_speed = 0;
2466                 if (netif_carrier_ok(netdev)) {
2467                         dev_info(&adapter->pdev->dev, "NIC Link is Down\n");
2468                         netif_carrier_off(netdev);
2469                         netif_tx_stop_all_queues(netdev);
2470                 }
2471         }
2472
2473         ixgbevf_update_stats(adapter);
2474
2475 pf_has_reset:
2476         /* Reset the timer */
2477         if (!test_bit(__IXGBEVF_DOWN, &adapter->state) &&
2478             !test_bit(__IXGBEVF_REMOVING, &adapter->state))
2479                 mod_timer(&adapter->watchdog_timer,
2480                           round_jiffies(jiffies + (2 * HZ)));
2481
2482         adapter->flags &= ~IXGBE_FLAG_IN_WATCHDOG_TASK;
2483 }
2484
2485 /**
2486  * ixgbevf_free_tx_resources - Free Tx Resources per Queue
2487  * @tx_ring: Tx descriptor ring for a specific queue
2488  *
2489  * Free all transmit software resources
2490  **/
2491 void ixgbevf_free_tx_resources(struct ixgbevf_ring *tx_ring)
2492 {
2493         ixgbevf_clean_tx_ring(tx_ring);
2494
2495         vfree(tx_ring->tx_buffer_info);
2496         tx_ring->tx_buffer_info = NULL;
2497
2498         /* if not set, then don't free */
2499         if (!tx_ring->desc)
2500                 return;
2501
2502         dma_free_coherent(tx_ring->dev, tx_ring->size, tx_ring->desc,
2503                           tx_ring->dma);
2504
2505         tx_ring->desc = NULL;
2506 }
2507
2508 /**
2509  * ixgbevf_free_all_tx_resources - Free Tx Resources for All Queues
2510  * @adapter: board private structure
2511  *
2512  * Free all transmit software resources
2513  **/
2514 static void ixgbevf_free_all_tx_resources(struct ixgbevf_adapter *adapter)
2515 {
2516         int i;
2517
2518         for (i = 0; i < adapter->num_tx_queues; i++)
2519                 if (adapter->tx_ring[i]->desc)
2520                         ixgbevf_free_tx_resources(adapter->tx_ring[i]);
2521 }
2522
2523 /**
2524  * ixgbevf_setup_tx_resources - allocate Tx resources (Descriptors)
2525  * @tx_ring:    tx descriptor ring (for a specific queue) to setup
2526  *
2527  * Return 0 on success, negative on failure
2528  **/
2529 int ixgbevf_setup_tx_resources(struct ixgbevf_ring *tx_ring)
2530 {
2531         int size;
2532
2533         size = sizeof(struct ixgbevf_tx_buffer) * tx_ring->count;
2534         tx_ring->tx_buffer_info = vzalloc(size);
2535         if (!tx_ring->tx_buffer_info)
2536                 goto err;
2537
2538         /* round up to nearest 4K */
2539         tx_ring->size = tx_ring->count * sizeof(union ixgbe_adv_tx_desc);
2540         tx_ring->size = ALIGN(tx_ring->size, 4096);
2541
2542         tx_ring->desc = dma_alloc_coherent(tx_ring->dev, tx_ring->size,
2543                                            &tx_ring->dma, GFP_KERNEL);
2544         if (!tx_ring->desc)
2545                 goto err;
2546
2547         return 0;
2548
2549 err:
2550         vfree(tx_ring->tx_buffer_info);
2551         tx_ring->tx_buffer_info = NULL;
2552         hw_dbg(&adapter->hw, "Unable to allocate memory for the transmit "
2553                "descriptor ring\n");
2554         return -ENOMEM;
2555 }
2556
2557 /**
2558  * ixgbevf_setup_all_tx_resources - allocate all queues Tx resources
2559  * @adapter: board private structure
2560  *
2561  * If this function returns with an error, then it's possible one or
2562  * more of the rings is populated (while the rest are not).  It is the
2563  * callers duty to clean those orphaned rings.
2564  *
2565  * Return 0 on success, negative on failure
2566  **/
2567 static int ixgbevf_setup_all_tx_resources(struct ixgbevf_adapter *adapter)
2568 {
2569         int i, err = 0;
2570
2571         for (i = 0; i < adapter->num_tx_queues; i++) {
2572                 err = ixgbevf_setup_tx_resources(adapter->tx_ring[i]);
2573                 if (!err)
2574                         continue;
2575                 hw_dbg(&adapter->hw,
2576                        "Allocation for Tx Queue %u failed\n", i);
2577                 break;
2578         }
2579
2580         return err;
2581 }
2582
2583 /**
2584  * ixgbevf_setup_rx_resources - allocate Rx resources (Descriptors)
2585  * @rx_ring:    rx descriptor ring (for a specific queue) to setup
2586  *
2587  * Returns 0 on success, negative on failure
2588  **/
2589 int ixgbevf_setup_rx_resources(struct ixgbevf_ring *rx_ring)
2590 {
2591         int size;
2592
2593         size = sizeof(struct ixgbevf_rx_buffer) * rx_ring->count;
2594         rx_ring->rx_buffer_info = vzalloc(size);
2595         if (!rx_ring->rx_buffer_info)
2596                 goto err;
2597
2598         /* Round up to nearest 4K */
2599         rx_ring->size = rx_ring->count * sizeof(union ixgbe_adv_rx_desc);
2600         rx_ring->size = ALIGN(rx_ring->size, 4096);
2601
2602         rx_ring->desc = dma_alloc_coherent(rx_ring->dev, rx_ring->size,
2603                                            &rx_ring->dma, GFP_KERNEL);
2604
2605         if (!rx_ring->desc)
2606                 goto err;
2607
2608         return 0;
2609 err:
2610         vfree(rx_ring->rx_buffer_info);
2611         rx_ring->rx_buffer_info = NULL;
2612         dev_err(rx_ring->dev, "Unable to allocate memory for the Rx descriptor ring\n");
2613         return -ENOMEM;
2614 }
2615
2616 /**
2617  * ixgbevf_setup_all_rx_resources - allocate all queues Rx resources
2618  * @adapter: board private structure
2619  *
2620  * If this function returns with an error, then it's possible one or
2621  * more of the rings is populated (while the rest are not).  It is the
2622  * callers duty to clean those orphaned rings.
2623  *
2624  * Return 0 on success, negative on failure
2625  **/
2626 static int ixgbevf_setup_all_rx_resources(struct ixgbevf_adapter *adapter)
2627 {
2628         int i, err = 0;
2629
2630         for (i = 0; i < adapter->num_rx_queues; i++) {
2631                 err = ixgbevf_setup_rx_resources(adapter->rx_ring[i]);
2632                 if (!err)
2633                         continue;
2634                 hw_dbg(&adapter->hw,
2635                        "Allocation for Rx Queue %u failed\n", i);
2636                 break;
2637         }
2638         return err;
2639 }
2640
2641 /**
2642  * ixgbevf_free_rx_resources - Free Rx Resources
2643  * @rx_ring: ring to clean the resources from
2644  *
2645  * Free all receive software resources
2646  **/
2647 void ixgbevf_free_rx_resources(struct ixgbevf_ring *rx_ring)
2648 {
2649         ixgbevf_clean_rx_ring(rx_ring);
2650
2651         vfree(rx_ring->rx_buffer_info);
2652         rx_ring->rx_buffer_info = NULL;
2653
2654         dma_free_coherent(rx_ring->dev, rx_ring->size, rx_ring->desc,
2655                           rx_ring->dma);
2656
2657         rx_ring->desc = NULL;
2658 }
2659
2660 /**
2661  * ixgbevf_free_all_rx_resources - Free Rx Resources for All Queues
2662  * @adapter: board private structure
2663  *
2664  * Free all receive software resources
2665  **/
2666 static void ixgbevf_free_all_rx_resources(struct ixgbevf_adapter *adapter)
2667 {
2668         int i;
2669
2670         for (i = 0; i < adapter->num_rx_queues; i++)
2671                 if (adapter->rx_ring[i]->desc)
2672                         ixgbevf_free_rx_resources(adapter->rx_ring[i]);
2673 }
2674
2675 /**
2676  * ixgbevf_open - Called when a network interface is made active
2677  * @netdev: network interface device structure
2678  *
2679  * Returns 0 on success, negative value on failure
2680  *
2681  * The open entry point is called when a network interface is made
2682  * active by the system (IFF_UP).  At this point all resources needed
2683  * for transmit and receive operations are allocated, the interrupt
2684  * handler is registered with the OS, the watchdog timer is started,
2685  * and the stack is notified that the interface is ready.
2686  **/
2687 static int ixgbevf_open(struct net_device *netdev)
2688 {
2689         struct ixgbevf_adapter *adapter = netdev_priv(netdev);
2690         struct ixgbe_hw *hw = &adapter->hw;
2691         int err;
2692
2693         /* A previous failure to open the device because of a lack of
2694          * available MSIX vector resources may have reset the number
2695          * of msix vectors variable to zero.  The only way to recover
2696          * is to unload/reload the driver and hope that the system has
2697          * been able to recover some MSIX vector resources.
2698          */
2699         if (!adapter->num_msix_vectors)
2700                 return -ENOMEM;
2701
2702         /* disallow open during test */
2703         if (test_bit(__IXGBEVF_TESTING, &adapter->state))
2704                 return -EBUSY;
2705
2706         if (hw->adapter_stopped) {
2707                 ixgbevf_reset(adapter);
2708                 /* if adapter is still stopped then PF isn't up and
2709                  * the vf can't start. */
2710                 if (hw->adapter_stopped) {
2711                         err = IXGBE_ERR_MBX;
2712                         pr_err("Unable to start - perhaps the PF Driver isn't "
2713                                "up yet\n");
2714                         goto err_setup_reset;
2715                 }
2716         }
2717
2718         /* allocate transmit descriptors */
2719         err = ixgbevf_setup_all_tx_resources(adapter);
2720         if (err)
2721                 goto err_setup_tx;
2722
2723         /* allocate receive descriptors */
2724         err = ixgbevf_setup_all_rx_resources(adapter);
2725         if (err)
2726                 goto err_setup_rx;
2727
2728         ixgbevf_configure(adapter);
2729
2730         /*
2731          * Map the Tx/Rx rings to the vectors we were allotted.
2732          * if request_irq will be called in this function map_rings
2733          * must be called *before* up_complete
2734          */
2735         ixgbevf_map_rings_to_vectors(adapter);
2736
2737         ixgbevf_up_complete(adapter);
2738
2739         /* clear any pending interrupts, may auto mask */
2740         IXGBE_READ_REG(hw, IXGBE_VTEICR);
2741         err = ixgbevf_request_irq(adapter);
2742         if (err)
2743                 goto err_req_irq;
2744
2745         ixgbevf_irq_enable(adapter);
2746
2747         return 0;
2748
2749 err_req_irq:
2750         ixgbevf_down(adapter);
2751 err_setup_rx:
2752         ixgbevf_free_all_rx_resources(adapter);
2753 err_setup_tx:
2754         ixgbevf_free_all_tx_resources(adapter);
2755         ixgbevf_reset(adapter);
2756
2757 err_setup_reset:
2758
2759         return err;
2760 }
2761
2762 /**
2763  * ixgbevf_close - Disables a network interface
2764  * @netdev: network interface device structure
2765  *
2766  * Returns 0, this is not allowed to fail
2767  *
2768  * The close entry point is called when an interface is de-activated
2769  * by the OS.  The hardware is still under the drivers control, but
2770  * needs to be disabled.  A global MAC reset is issued to stop the
2771  * hardware, and all transmit and receive resources are freed.
2772  **/
2773 static int ixgbevf_close(struct net_device *netdev)
2774 {
2775         struct ixgbevf_adapter *adapter = netdev_priv(netdev);
2776
2777         ixgbevf_down(adapter);
2778         ixgbevf_free_irq(adapter);
2779
2780         ixgbevf_free_all_tx_resources(adapter);
2781         ixgbevf_free_all_rx_resources(adapter);
2782
2783         return 0;
2784 }
2785
2786 static void ixgbevf_queue_reset_subtask(struct ixgbevf_adapter *adapter)
2787 {
2788         struct net_device *dev = adapter->netdev;
2789
2790         if (!(adapter->flags & IXGBEVF_FLAG_QUEUE_RESET_REQUESTED))
2791                 return;
2792
2793         adapter->flags &= ~IXGBEVF_FLAG_QUEUE_RESET_REQUESTED;
2794
2795         /* if interface is down do nothing */
2796         if (test_bit(__IXGBEVF_DOWN, &adapter->state) ||
2797             test_bit(__IXGBEVF_RESETTING, &adapter->state))
2798                 return;
2799
2800         /* Hardware has to reinitialize queues and interrupts to
2801          * match packet buffer alignment. Unfortunately, the
2802          * hardware is not flexible enough to do this dynamically.
2803          */
2804         if (netif_running(dev))
2805                 ixgbevf_close(dev);
2806
2807         ixgbevf_clear_interrupt_scheme(adapter);
2808         ixgbevf_init_interrupt_scheme(adapter);
2809
2810         if (netif_running(dev))
2811                 ixgbevf_open(dev);
2812 }
2813
2814 static void ixgbevf_tx_ctxtdesc(struct ixgbevf_ring *tx_ring,
2815                                 u32 vlan_macip_lens, u32 type_tucmd,
2816                                 u32 mss_l4len_idx)
2817 {
2818         struct ixgbe_adv_tx_context_desc *context_desc;
2819         u16 i = tx_ring->next_to_use;
2820
2821         context_desc = IXGBEVF_TX_CTXTDESC(tx_ring, i);
2822
2823         i++;
2824         tx_ring->next_to_use = (i < tx_ring->count) ? i : 0;
2825
2826         /* set bits to identify this as an advanced context descriptor */
2827         type_tucmd |= IXGBE_TXD_CMD_DEXT | IXGBE_ADVTXD_DTYP_CTXT;
2828
2829         context_desc->vlan_macip_lens   = cpu_to_le32(vlan_macip_lens);
2830         context_desc->seqnum_seed       = 0;
2831         context_desc->type_tucmd_mlhl   = cpu_to_le32(type_tucmd);
2832         context_desc->mss_l4len_idx     = cpu_to_le32(mss_l4len_idx);
2833 }
2834
2835 static int ixgbevf_tso(struct ixgbevf_ring *tx_ring,
2836                        struct ixgbevf_tx_buffer *first,
2837                        u8 *hdr_len)
2838 {
2839         struct sk_buff *skb = first->skb;
2840         u32 vlan_macip_lens, type_tucmd;
2841         u32 mss_l4len_idx, l4len;
2842         int err;
2843
2844         if (skb->ip_summed != CHECKSUM_PARTIAL)
2845                 return 0;
2846
2847         if (!skb_is_gso(skb))
2848                 return 0;
2849
2850         err = skb_cow_head(skb, 0);
2851         if (err < 0)
2852                 return err;
2853
2854         /* ADV DTYP TUCMD MKRLOC/ISCSIHEDLEN */
2855         type_tucmd = IXGBE_ADVTXD_TUCMD_L4T_TCP;
2856
2857         if (skb->protocol == htons(ETH_P_IP)) {
2858                 struct iphdr *iph = ip_hdr(skb);
2859                 iph->tot_len = 0;
2860                 iph->check = 0;
2861                 tcp_hdr(skb)->check = ~csum_tcpudp_magic(iph->saddr,
2862                                                          iph->daddr, 0,
2863                                                          IPPROTO_TCP,
2864                                                          0);
2865                 type_tucmd |= IXGBE_ADVTXD_TUCMD_IPV4;
2866                 first->tx_flags |= IXGBE_TX_FLAGS_TSO |
2867                                    IXGBE_TX_FLAGS_CSUM |
2868                                    IXGBE_TX_FLAGS_IPV4;
2869         } else if (skb_is_gso_v6(skb)) {
2870                 ipv6_hdr(skb)->payload_len = 0;
2871                 tcp_hdr(skb)->check =
2872                     ~csum_ipv6_magic(&ipv6_hdr(skb)->saddr,
2873                                      &ipv6_hdr(skb)->daddr,
2874                                      0, IPPROTO_TCP, 0);
2875                 first->tx_flags |= IXGBE_TX_FLAGS_TSO |
2876                                    IXGBE_TX_FLAGS_CSUM;
2877         }
2878
2879         /* compute header lengths */
2880         l4len = tcp_hdrlen(skb);
2881         *hdr_len += l4len;
2882         *hdr_len = skb_transport_offset(skb) + l4len;
2883
2884         /* update gso size and bytecount with header size */
2885         first->gso_segs = skb_shinfo(skb)->gso_segs;
2886         first->bytecount += (first->gso_segs - 1) * *hdr_len;
2887
2888         /* mss_l4len_id: use 1 as index for TSO */
2889         mss_l4len_idx = l4len << IXGBE_ADVTXD_L4LEN_SHIFT;
2890         mss_l4len_idx |= skb_shinfo(skb)->gso_size << IXGBE_ADVTXD_MSS_SHIFT;
2891         mss_l4len_idx |= 1 << IXGBE_ADVTXD_IDX_SHIFT;
2892
2893         /* vlan_macip_lens: HEADLEN, MACLEN, VLAN tag */
2894         vlan_macip_lens = skb_network_header_len(skb);
2895         vlan_macip_lens |= skb_network_offset(skb) << IXGBE_ADVTXD_MACLEN_SHIFT;
2896         vlan_macip_lens |= first->tx_flags & IXGBE_TX_FLAGS_VLAN_MASK;
2897
2898         ixgbevf_tx_ctxtdesc(tx_ring, vlan_macip_lens,
2899                             type_tucmd, mss_l4len_idx);
2900
2901         return 1;
2902 }
2903
2904 static void ixgbevf_tx_csum(struct ixgbevf_ring *tx_ring,
2905                             struct ixgbevf_tx_buffer *first)
2906 {
2907         struct sk_buff *skb = first->skb;
2908         u32 vlan_macip_lens = 0;
2909         u32 mss_l4len_idx = 0;
2910         u32 type_tucmd = 0;
2911
2912         if (skb->ip_summed == CHECKSUM_PARTIAL) {
2913                 u8 l4_hdr = 0;
2914                 switch (skb->protocol) {
2915                 case htons(ETH_P_IP):
2916                         vlan_macip_lens |= skb_network_header_len(skb);
2917                         type_tucmd |= IXGBE_ADVTXD_TUCMD_IPV4;
2918                         l4_hdr = ip_hdr(skb)->protocol;
2919                         break;
2920                 case htons(ETH_P_IPV6):
2921                         vlan_macip_lens |= skb_network_header_len(skb);
2922                         l4_hdr = ipv6_hdr(skb)->nexthdr;
2923                         break;
2924                 default:
2925                         if (unlikely(net_ratelimit())) {
2926                                 dev_warn(tx_ring->dev,
2927                                  "partial checksum but proto=%x!\n",
2928                                  first->protocol);
2929                         }
2930                         break;
2931                 }
2932
2933                 switch (l4_hdr) {
2934                 case IPPROTO_TCP:
2935                         type_tucmd |= IXGBE_ADVTXD_TUCMD_L4T_TCP;
2936                         mss_l4len_idx = tcp_hdrlen(skb) <<
2937                                         IXGBE_ADVTXD_L4LEN_SHIFT;
2938                         break;
2939                 case IPPROTO_SCTP:
2940                         type_tucmd |= IXGBE_ADVTXD_TUCMD_L4T_SCTP;
2941                         mss_l4len_idx = sizeof(struct sctphdr) <<
2942                                         IXGBE_ADVTXD_L4LEN_SHIFT;
2943                         break;
2944                 case IPPROTO_UDP:
2945                         mss_l4len_idx = sizeof(struct udphdr) <<
2946                                         IXGBE_ADVTXD_L4LEN_SHIFT;
2947                         break;
2948                 default:
2949                         if (unlikely(net_ratelimit())) {
2950                                 dev_warn(tx_ring->dev,
2951                                  "partial checksum but l4 proto=%x!\n",
2952                                  l4_hdr);
2953                         }
2954                         break;
2955                 }
2956
2957                 /* update TX checksum flag */
2958                 first->tx_flags |= IXGBE_TX_FLAGS_CSUM;
2959         }
2960
2961         /* vlan_macip_lens: MACLEN, VLAN tag */
2962         vlan_macip_lens |= skb_network_offset(skb) << IXGBE_ADVTXD_MACLEN_SHIFT;
2963         vlan_macip_lens |= first->tx_flags & IXGBE_TX_FLAGS_VLAN_MASK;
2964
2965         ixgbevf_tx_ctxtdesc(tx_ring, vlan_macip_lens,
2966                             type_tucmd, mss_l4len_idx);
2967 }
2968
2969 static __le32 ixgbevf_tx_cmd_type(u32 tx_flags)
2970 {
2971         /* set type for advanced descriptor with frame checksum insertion */
2972         __le32 cmd_type = cpu_to_le32(IXGBE_ADVTXD_DTYP_DATA |
2973                                       IXGBE_ADVTXD_DCMD_IFCS |
2974                                       IXGBE_ADVTXD_DCMD_DEXT);
2975
2976         /* set HW vlan bit if vlan is present */
2977         if (tx_flags & IXGBE_TX_FLAGS_VLAN)
2978                 cmd_type |= cpu_to_le32(IXGBE_ADVTXD_DCMD_VLE);
2979
2980         /* set segmentation enable bits for TSO/FSO */
2981         if (tx_flags & IXGBE_TX_FLAGS_TSO)
2982                 cmd_type |= cpu_to_le32(IXGBE_ADVTXD_DCMD_TSE);
2983
2984         return cmd_type;
2985 }
2986
2987 static void ixgbevf_tx_olinfo_status(union ixgbe_adv_tx_desc *tx_desc,
2988                                      u32 tx_flags, unsigned int paylen)
2989 {
2990         __le32 olinfo_status = cpu_to_le32(paylen << IXGBE_ADVTXD_PAYLEN_SHIFT);
2991
2992         /* enable L4 checksum for TSO and TX checksum offload */
2993         if (tx_flags & IXGBE_TX_FLAGS_CSUM)
2994                 olinfo_status |= cpu_to_le32(IXGBE_ADVTXD_POPTS_TXSM);
2995
2996         /* enble IPv4 checksum for TSO */
2997         if (tx_flags & IXGBE_TX_FLAGS_IPV4)
2998                 olinfo_status |= cpu_to_le32(IXGBE_ADVTXD_POPTS_IXSM);
2999
3000         /* use index 1 context for TSO/FSO/FCOE */
3001         if (tx_flags & IXGBE_TX_FLAGS_TSO)
3002                 olinfo_status |= cpu_to_le32(1 << IXGBE_ADVTXD_IDX_SHIFT);
3003
3004         /* Check Context must be set if Tx switch is enabled, which it
3005          * always is for case where virtual functions are running
3006          */
3007         olinfo_status |= cpu_to_le32(IXGBE_ADVTXD_CC);
3008
3009         tx_desc->read.olinfo_status = olinfo_status;
3010 }
3011
3012 static void ixgbevf_tx_map(struct ixgbevf_ring *tx_ring,
3013                            struct ixgbevf_tx_buffer *first,
3014                            const u8 hdr_len)
3015 {
3016         dma_addr_t dma;
3017         struct sk_buff *skb = first->skb;
3018         struct ixgbevf_tx_buffer *tx_buffer;
3019         union ixgbe_adv_tx_desc *tx_desc;
3020         struct skb_frag_struct *frag = &skb_shinfo(skb)->frags[0];
3021         unsigned int data_len = skb->data_len;
3022         unsigned int size = skb_headlen(skb);
3023         unsigned int paylen = skb->len - hdr_len;
3024         u32 tx_flags = first->tx_flags;
3025         __le32 cmd_type;
3026         u16 i = tx_ring->next_to_use;
3027
3028         tx_desc = IXGBEVF_TX_DESC(tx_ring, i);
3029
3030         ixgbevf_tx_olinfo_status(tx_desc, tx_flags, paylen);
3031         cmd_type = ixgbevf_tx_cmd_type(tx_flags);
3032
3033         dma = dma_map_single(tx_ring->dev, skb->data, size, DMA_TO_DEVICE);
3034         if (dma_mapping_error(tx_ring->dev, dma))
3035                 goto dma_error;
3036
3037         /* record length, and DMA address */
3038         dma_unmap_len_set(first, len, size);
3039         dma_unmap_addr_set(first, dma, dma);
3040
3041         tx_desc->read.buffer_addr = cpu_to_le64(dma);
3042
3043         for (;;) {
3044                 while (unlikely(size > IXGBE_MAX_DATA_PER_TXD)) {
3045                         tx_desc->read.cmd_type_len =
3046                                 cmd_type | cpu_to_le32(IXGBE_MAX_DATA_PER_TXD);
3047
3048                         i++;
3049                         tx_desc++;
3050                         if (i == tx_ring->count) {
3051                                 tx_desc = IXGBEVF_TX_DESC(tx_ring, 0);
3052                                 i = 0;
3053                         }
3054
3055                         dma += IXGBE_MAX_DATA_PER_TXD;
3056                         size -= IXGBE_MAX_DATA_PER_TXD;
3057
3058                         tx_desc->read.buffer_addr = cpu_to_le64(dma);
3059                         tx_desc->read.olinfo_status = 0;
3060                 }
3061
3062                 if (likely(!data_len))
3063                         break;
3064
3065                 tx_desc->read.cmd_type_len = cmd_type | cpu_to_le32(size);
3066
3067                 i++;
3068                 tx_desc++;
3069                 if (i == tx_ring->count) {
3070                         tx_desc = IXGBEVF_TX_DESC(tx_ring, 0);
3071                         i = 0;
3072                 }
3073
3074                 size = skb_frag_size(frag);
3075                 data_len -= size;
3076
3077                 dma = skb_frag_dma_map(tx_ring->dev, frag, 0, size,
3078                                        DMA_TO_DEVICE);
3079                 if (dma_mapping_error(tx_ring->dev, dma))
3080                         goto dma_error;
3081
3082                 tx_buffer = &tx_ring->tx_buffer_info[i];
3083                 dma_unmap_len_set(tx_buffer, len, size);
3084                 dma_unmap_addr_set(tx_buffer, dma, dma);
3085
3086                 tx_desc->read.buffer_addr = cpu_to_le64(dma);
3087                 tx_desc->read.olinfo_status = 0;
3088
3089                 frag++;
3090         }
3091
3092         /* write last descriptor with RS and EOP bits */
3093         cmd_type |= cpu_to_le32(size) | cpu_to_le32(IXGBE_TXD_CMD);
3094         tx_desc->read.cmd_type_len = cmd_type;
3095
3096         /* set the timestamp */
3097         first->time_stamp = jiffies;
3098
3099         /* Force memory writes to complete before letting h/w know there
3100          * are new descriptors to fetch.  (Only applicable for weak-ordered
3101          * memory model archs, such as IA-64).
3102          *
3103          * We also need this memory barrier (wmb) to make certain all of the
3104          * status bits have been updated before next_to_watch is written.
3105          */
3106         wmb();
3107
3108         /* set next_to_watch value indicating a packet is present */
3109         first->next_to_watch = tx_desc;
3110
3111         i++;
3112         if (i == tx_ring->count)
3113                 i = 0;
3114
3115         tx_ring->next_to_use = i;
3116
3117         /* notify HW of packet */
3118         ixgbevf_write_tail(tx_ring, i);
3119
3120         return;
3121 dma_error:
3122         dev_err(tx_ring->dev, "TX DMA map failed\n");
3123
3124         /* clear dma mappings for failed tx_buffer_info map */
3125         for (;;) {
3126                 tx_buffer = &tx_ring->tx_buffer_info[i];
3127                 ixgbevf_unmap_and_free_tx_resource(tx_ring, tx_buffer);
3128                 if (tx_buffer == first)
3129                         break;
3130                 if (i == 0)
3131                         i = tx_ring->count;
3132                 i--;
3133         }
3134
3135         tx_ring->next_to_use = i;
3136 }
3137
3138 static int __ixgbevf_maybe_stop_tx(struct ixgbevf_ring *tx_ring, int size)
3139 {
3140         netif_stop_subqueue(tx_ring->netdev, tx_ring->queue_index);
3141         /* Herbert's original patch had:
3142          *  smp_mb__after_netif_stop_queue();
3143          * but since that doesn't exist yet, just open code it. */
3144         smp_mb();
3145
3146         /* We need to check again in a case another CPU has just
3147          * made room available. */
3148         if (likely(ixgbevf_desc_unused(tx_ring) < size))
3149                 return -EBUSY;
3150
3151         /* A reprieve! - use start_queue because it doesn't call schedule */
3152         netif_start_subqueue(tx_ring->netdev, tx_ring->queue_index);
3153         ++tx_ring->tx_stats.restart_queue;
3154
3155         return 0;
3156 }
3157
3158 static int ixgbevf_maybe_stop_tx(struct ixgbevf_ring *tx_ring, int size)
3159 {
3160         if (likely(ixgbevf_desc_unused(tx_ring) >= size))
3161                 return 0;
3162         return __ixgbevf_maybe_stop_tx(tx_ring, size);
3163 }
3164
3165 static int ixgbevf_xmit_frame(struct sk_buff *skb, struct net_device *netdev)
3166 {
3167         struct ixgbevf_adapter *adapter = netdev_priv(netdev);
3168         struct ixgbevf_tx_buffer *first;
3169         struct ixgbevf_ring *tx_ring;
3170         int tso;
3171         u32 tx_flags = 0;
3172         u16 count = TXD_USE_COUNT(skb_headlen(skb));
3173 #if PAGE_SIZE > IXGBE_MAX_DATA_PER_TXD
3174         unsigned short f;
3175 #endif
3176         u8 hdr_len = 0;
3177         u8 *dst_mac = skb_header_pointer(skb, 0, 0, NULL);
3178
3179         if (!dst_mac || is_link_local_ether_addr(dst_mac)) {
3180                 dev_kfree_skb(skb);
3181                 return NETDEV_TX_OK;
3182         }
3183
3184         tx_ring = adapter->tx_ring[skb->queue_mapping];
3185
3186         /*
3187          * need: 1 descriptor per page * PAGE_SIZE/IXGBE_MAX_DATA_PER_TXD,
3188          *       + 1 desc for skb_headlen/IXGBE_MAX_DATA_PER_TXD,
3189          *       + 2 desc gap to keep tail from touching head,
3190          *       + 1 desc for context descriptor,
3191          * otherwise try next time
3192          */
3193 #if PAGE_SIZE > IXGBE_MAX_DATA_PER_TXD
3194         for (f = 0; f < skb_shinfo(skb)->nr_frags; f++)
3195                 count += TXD_USE_COUNT(skb_shinfo(skb)->frags[f].size);
3196 #else
3197         count += skb_shinfo(skb)->nr_frags;
3198 #endif
3199         if (ixgbevf_maybe_stop_tx(tx_ring, count + 3)) {
3200                 tx_ring->tx_stats.tx_busy++;
3201                 return NETDEV_TX_BUSY;
3202         }
3203
3204         /* record the location of the first descriptor for this packet */
3205         first = &tx_ring->tx_buffer_info[tx_ring->next_to_use];
3206         first->skb = skb;
3207         first->bytecount = skb->len;
3208         first->gso_segs = 1;
3209
3210         if (vlan_tx_tag_present(skb)) {
3211                 tx_flags |= vlan_tx_tag_get(skb);
3212                 tx_flags <<= IXGBE_TX_FLAGS_VLAN_SHIFT;
3213                 tx_flags |= IXGBE_TX_FLAGS_VLAN;
3214         }
3215
3216         /* record initial flags and protocol */
3217         first->tx_flags = tx_flags;
3218         first->protocol = vlan_get_protocol(skb);
3219
3220         tso = ixgbevf_tso(tx_ring, first, &hdr_len);
3221         if (tso < 0)
3222                 goto out_drop;
3223         else if (!tso)
3224                 ixgbevf_tx_csum(tx_ring, first);
3225
3226         ixgbevf_tx_map(tx_ring, first, hdr_len);
3227
3228         ixgbevf_maybe_stop_tx(tx_ring, DESC_NEEDED);
3229
3230         return NETDEV_TX_OK;
3231
3232 out_drop:
3233         dev_kfree_skb_any(first->skb);
3234         first->skb = NULL;
3235
3236         return NETDEV_TX_OK;
3237 }
3238
3239 /**
3240  * ixgbevf_set_mac - Change the Ethernet Address of the NIC
3241  * @netdev: network interface device structure
3242  * @p: pointer to an address structure
3243  *
3244  * Returns 0 on success, negative on failure
3245  **/
3246 static int ixgbevf_set_mac(struct net_device *netdev, void *p)
3247 {
3248         struct ixgbevf_adapter *adapter = netdev_priv(netdev);
3249         struct ixgbe_hw *hw = &adapter->hw;
3250         struct sockaddr *addr = p;
3251
3252         if (!is_valid_ether_addr(addr->sa_data))
3253                 return -EADDRNOTAVAIL;
3254
3255         memcpy(netdev->dev_addr, addr->sa_data, netdev->addr_len);
3256         memcpy(hw->mac.addr, addr->sa_data, netdev->addr_len);
3257
3258         spin_lock_bh(&adapter->mbx_lock);
3259
3260         hw->mac.ops.set_rar(hw, 0, hw->mac.addr, 0);
3261
3262         spin_unlock_bh(&adapter->mbx_lock);
3263
3264         return 0;
3265 }
3266
3267 /**
3268  * ixgbevf_change_mtu - Change the Maximum Transfer Unit
3269  * @netdev: network interface device structure
3270  * @new_mtu: new value for maximum frame size
3271  *
3272  * Returns 0 on success, negative on failure
3273  **/
3274 static int ixgbevf_change_mtu(struct net_device *netdev, int new_mtu)
3275 {
3276         struct ixgbevf_adapter *adapter = netdev_priv(netdev);
3277         int max_frame = new_mtu + ETH_HLEN + ETH_FCS_LEN;
3278         int max_possible_frame = MAXIMUM_ETHERNET_VLAN_SIZE;
3279
3280         switch (adapter->hw.api_version) {
3281         case ixgbe_mbox_api_11:
3282                 max_possible_frame = IXGBE_MAX_JUMBO_FRAME_SIZE;
3283                 break;
3284         default:
3285                 if (adapter->hw.mac.type == ixgbe_mac_X540_vf)
3286                         max_possible_frame = IXGBE_MAX_JUMBO_FRAME_SIZE;
3287                 break;
3288         }
3289
3290         /* MTU < 68 is an error and causes problems on some kernels */
3291         if ((new_mtu < 68) || (max_frame > max_possible_frame))
3292                 return -EINVAL;
3293
3294         hw_dbg(&adapter->hw, "changing MTU from %d to %d\n",
3295                netdev->mtu, new_mtu);
3296         /* must set new MTU before calling down or up */
3297         netdev->mtu = new_mtu;
3298
3299         if (netif_running(netdev))
3300                 ixgbevf_reinit_locked(adapter);
3301
3302         return 0;
3303 }
3304
3305 static int ixgbevf_suspend(struct pci_dev *pdev, pm_message_t state)
3306 {
3307         struct net_device *netdev = pci_get_drvdata(pdev);
3308         struct ixgbevf_adapter *adapter = netdev_priv(netdev);
3309 #ifdef CONFIG_PM
3310         int retval = 0;
3311 #endif
3312
3313         netif_device_detach(netdev);
3314
3315         if (netif_running(netdev)) {
3316                 rtnl_lock();
3317                 ixgbevf_down(adapter);
3318                 ixgbevf_free_irq(adapter);
3319                 ixgbevf_free_all_tx_resources(adapter);
3320                 ixgbevf_free_all_rx_resources(adapter);
3321                 rtnl_unlock();
3322         }
3323
3324         ixgbevf_clear_interrupt_scheme(adapter);
3325
3326 #ifdef CONFIG_PM
3327         retval = pci_save_state(pdev);
3328         if (retval)
3329                 return retval;
3330
3331 #endif
3332         if (!test_and_set_bit(__IXGBEVF_DISABLED, &adapter->state))
3333                 pci_disable_device(pdev);
3334
3335         return 0;
3336 }
3337
3338 #ifdef CONFIG_PM
3339 static int ixgbevf_resume(struct pci_dev *pdev)
3340 {
3341         struct net_device *netdev = pci_get_drvdata(pdev);
3342         struct ixgbevf_adapter *adapter = netdev_priv(netdev);
3343         u32 err;
3344
3345         pci_restore_state(pdev);
3346         /*
3347          * pci_restore_state clears dev->state_saved so call
3348          * pci_save_state to restore it.
3349          */
3350         pci_save_state(pdev);
3351
3352         err = pci_enable_device_mem(pdev);
3353         if (err) {
3354                 dev_err(&pdev->dev, "Cannot enable PCI device from suspend\n");
3355                 return err;
3356         }
3357         smp_mb__before_atomic();
3358         clear_bit(__IXGBEVF_DISABLED, &adapter->state);
3359         pci_set_master(pdev);
3360
3361         ixgbevf_reset(adapter);
3362
3363         rtnl_lock();
3364         err = ixgbevf_init_interrupt_scheme(adapter);
3365         rtnl_unlock();
3366         if (err) {
3367                 dev_err(&pdev->dev, "Cannot initialize interrupts\n");
3368                 return err;
3369         }
3370
3371         if (netif_running(netdev)) {
3372                 err = ixgbevf_open(netdev);
3373                 if (err)
3374                         return err;
3375         }
3376
3377         netif_device_attach(netdev);
3378
3379         return err;
3380 }
3381
3382 #endif /* CONFIG_PM */
3383 static void ixgbevf_shutdown(struct pci_dev *pdev)
3384 {
3385         ixgbevf_suspend(pdev, PMSG_SUSPEND);
3386 }
3387
3388 static struct rtnl_link_stats64 *ixgbevf_get_stats(struct net_device *netdev,
3389                                                 struct rtnl_link_stats64 *stats)
3390 {
3391         struct ixgbevf_adapter *adapter = netdev_priv(netdev);
3392         unsigned int start;
3393         u64 bytes, packets;
3394         const struct ixgbevf_ring *ring;
3395         int i;
3396
3397         ixgbevf_update_stats(adapter);
3398
3399         stats->multicast = adapter->stats.vfmprc - adapter->stats.base_vfmprc;
3400
3401         for (i = 0; i < adapter->num_rx_queues; i++) {
3402                 ring = adapter->rx_ring[i];
3403                 do {
3404                         start = u64_stats_fetch_begin_irq(&ring->syncp);
3405                         bytes = ring->stats.bytes;
3406                         packets = ring->stats.packets;
3407                 } while (u64_stats_fetch_retry_irq(&ring->syncp, start));
3408                 stats->rx_bytes += bytes;
3409                 stats->rx_packets += packets;
3410         }
3411
3412         for (i = 0; i < adapter->num_tx_queues; i++) {
3413                 ring = adapter->tx_ring[i];
3414                 do {
3415                         start = u64_stats_fetch_begin_irq(&ring->syncp);
3416                         bytes = ring->stats.bytes;
3417                         packets = ring->stats.packets;
3418                 } while (u64_stats_fetch_retry_irq(&ring->syncp, start));
3419                 stats->tx_bytes += bytes;
3420                 stats->tx_packets += packets;
3421         }
3422
3423         return stats;
3424 }
3425
3426 static const struct net_device_ops ixgbevf_netdev_ops = {
3427         .ndo_open               = ixgbevf_open,
3428         .ndo_stop               = ixgbevf_close,
3429         .ndo_start_xmit         = ixgbevf_xmit_frame,
3430         .ndo_set_rx_mode        = ixgbevf_set_rx_mode,
3431         .ndo_get_stats64        = ixgbevf_get_stats,
3432         .ndo_validate_addr      = eth_validate_addr,
3433         .ndo_set_mac_address    = ixgbevf_set_mac,
3434         .ndo_change_mtu         = ixgbevf_change_mtu,
3435         .ndo_tx_timeout         = ixgbevf_tx_timeout,
3436         .ndo_vlan_rx_add_vid    = ixgbevf_vlan_rx_add_vid,
3437         .ndo_vlan_rx_kill_vid   = ixgbevf_vlan_rx_kill_vid,
3438 #ifdef CONFIG_NET_RX_BUSY_POLL
3439         .ndo_busy_poll          = ixgbevf_busy_poll_recv,
3440 #endif
3441 };
3442
3443 static void ixgbevf_assign_netdev_ops(struct net_device *dev)
3444 {
3445         dev->netdev_ops = &ixgbevf_netdev_ops;
3446         ixgbevf_set_ethtool_ops(dev);
3447         dev->watchdog_timeo = 5 * HZ;
3448 }
3449
3450 /**
3451  * ixgbevf_probe - Device Initialization Routine
3452  * @pdev: PCI device information struct
3453  * @ent: entry in ixgbevf_pci_tbl
3454  *
3455  * Returns 0 on success, negative on failure
3456  *
3457  * ixgbevf_probe initializes an adapter identified by a pci_dev structure.
3458  * The OS initialization, configuring of the adapter private structure,
3459  * and a hardware reset occur.
3460  **/
3461 static int ixgbevf_probe(struct pci_dev *pdev, const struct pci_device_id *ent)
3462 {
3463         struct net_device *netdev;
3464         struct ixgbevf_adapter *adapter = NULL;
3465         struct ixgbe_hw *hw = NULL;
3466         const struct ixgbevf_info *ii = ixgbevf_info_tbl[ent->driver_data];
3467         static int cards_found;
3468         int err, pci_using_dac;
3469
3470         err = pci_enable_device(pdev);
3471         if (err)
3472                 return err;
3473
3474         if (!dma_set_mask_and_coherent(&pdev->dev, DMA_BIT_MASK(64))) {
3475                 pci_using_dac = 1;
3476         } else {
3477                 err = dma_set_mask_and_coherent(&pdev->dev, DMA_BIT_MASK(32));
3478                 if (err) {
3479                         dev_err(&pdev->dev, "No usable DMA "
3480                                 "configuration, aborting\n");
3481                         goto err_dma;
3482                 }
3483                 pci_using_dac = 0;
3484         }
3485
3486         err = pci_request_regions(pdev, ixgbevf_driver_name);
3487         if (err) {
3488                 dev_err(&pdev->dev, "pci_request_regions failed 0x%x\n", err);
3489                 goto err_pci_reg;
3490         }
3491
3492         pci_set_master(pdev);
3493
3494         netdev = alloc_etherdev_mq(sizeof(struct ixgbevf_adapter),
3495                                    MAX_TX_QUEUES);
3496         if (!netdev) {
3497                 err = -ENOMEM;
3498                 goto err_alloc_etherdev;
3499         }
3500
3501         SET_NETDEV_DEV(netdev, &pdev->dev);
3502
3503         pci_set_drvdata(pdev, netdev);
3504         adapter = netdev_priv(netdev);
3505
3506         adapter->netdev = netdev;
3507         adapter->pdev = pdev;
3508         hw = &adapter->hw;
3509         hw->back = adapter;
3510         adapter->msg_enable = netif_msg_init(debug, DEFAULT_MSG_ENABLE);
3511
3512         /*
3513          * call save state here in standalone driver because it relies on
3514          * adapter struct to exist, and needs to call netdev_priv
3515          */
3516         pci_save_state(pdev);
3517
3518         hw->hw_addr = ioremap(pci_resource_start(pdev, 0),
3519                               pci_resource_len(pdev, 0));
3520         adapter->io_addr = hw->hw_addr;
3521         if (!hw->hw_addr) {
3522                 err = -EIO;
3523                 goto err_ioremap;
3524         }
3525
3526         ixgbevf_assign_netdev_ops(netdev);
3527
3528         adapter->bd_number = cards_found;
3529
3530         /* Setup hw api */
3531         memcpy(&hw->mac.ops, ii->mac_ops, sizeof(hw->mac.ops));
3532         hw->mac.type  = ii->mac;
3533
3534         memcpy(&hw->mbx.ops, &ixgbevf_mbx_ops,
3535                sizeof(struct ixgbe_mbx_operations));
3536
3537         /* setup the private structure */
3538         err = ixgbevf_sw_init(adapter);
3539         if (err)
3540                 goto err_sw_init;
3541
3542         /* The HW MAC address was set and/or determined in sw_init */
3543         if (!is_valid_ether_addr(netdev->dev_addr)) {
3544                 pr_err("invalid MAC address\n");
3545                 err = -EIO;
3546                 goto err_sw_init;
3547         }
3548
3549         netdev->hw_features = NETIF_F_SG |
3550                            NETIF_F_IP_CSUM |
3551                            NETIF_F_IPV6_CSUM |
3552                            NETIF_F_TSO |
3553                            NETIF_F_TSO6 |
3554                            NETIF_F_RXCSUM;
3555
3556         netdev->features = netdev->hw_features |
3557                            NETIF_F_HW_VLAN_CTAG_TX |
3558                            NETIF_F_HW_VLAN_CTAG_RX |
3559                            NETIF_F_HW_VLAN_CTAG_FILTER;
3560
3561         netdev->vlan_features |= NETIF_F_TSO;
3562         netdev->vlan_features |= NETIF_F_TSO6;
3563         netdev->vlan_features |= NETIF_F_IP_CSUM;
3564         netdev->vlan_features |= NETIF_F_IPV6_CSUM;
3565         netdev->vlan_features |= NETIF_F_SG;
3566
3567         if (pci_using_dac)
3568                 netdev->features |= NETIF_F_HIGHDMA;
3569
3570         netdev->priv_flags |= IFF_UNICAST_FLT;
3571
3572         init_timer(&adapter->watchdog_timer);
3573         adapter->watchdog_timer.function = ixgbevf_watchdog;
3574         adapter->watchdog_timer.data = (unsigned long)adapter;
3575
3576         if (IXGBE_REMOVED(hw->hw_addr)) {
3577                 err = -EIO;
3578                 goto err_sw_init;
3579         }
3580         INIT_WORK(&adapter->reset_task, ixgbevf_reset_task);
3581         INIT_WORK(&adapter->watchdog_task, ixgbevf_watchdog_task);
3582         set_bit(__IXGBEVF_WORK_INIT, &adapter->state);
3583
3584         err = ixgbevf_init_interrupt_scheme(adapter);
3585         if (err)
3586                 goto err_sw_init;
3587
3588         strcpy(netdev->name, "eth%d");
3589
3590         err = register_netdev(netdev);
3591         if (err)
3592                 goto err_register;
3593
3594         netif_carrier_off(netdev);
3595
3596         ixgbevf_init_last_counter_stats(adapter);
3597
3598         /* print the MAC address */
3599         hw_dbg(hw, "%pM\n", netdev->dev_addr);
3600
3601         hw_dbg(hw, "MAC: %d\n", hw->mac.type);
3602
3603         hw_dbg(hw, "Intel(R) 82599 Virtual Function\n");
3604         cards_found++;
3605         return 0;
3606
3607 err_register:
3608         ixgbevf_clear_interrupt_scheme(adapter);
3609 err_sw_init:
3610         ixgbevf_reset_interrupt_capability(adapter);
3611         iounmap(adapter->io_addr);
3612 err_ioremap:
3613         free_netdev(netdev);
3614 err_alloc_etherdev:
3615         pci_release_regions(pdev);
3616 err_pci_reg:
3617 err_dma:
3618         if (!test_and_set_bit(__IXGBEVF_DISABLED, &adapter->state))
3619                 pci_disable_device(pdev);
3620         return err;
3621 }
3622
3623 /**
3624  * ixgbevf_remove - Device Removal Routine
3625  * @pdev: PCI device information struct
3626  *
3627  * ixgbevf_remove is called by the PCI subsystem to alert the driver
3628  * that it should release a PCI device.  The could be caused by a
3629  * Hot-Plug event, or because the driver is going to be removed from
3630  * memory.
3631  **/
3632 static void ixgbevf_remove(struct pci_dev *pdev)
3633 {
3634         struct net_device *netdev = pci_get_drvdata(pdev);
3635         struct ixgbevf_adapter *adapter = netdev_priv(netdev);
3636
3637         set_bit(__IXGBEVF_REMOVING, &adapter->state);
3638
3639         del_timer_sync(&adapter->watchdog_timer);
3640
3641         cancel_work_sync(&adapter->reset_task);
3642         cancel_work_sync(&adapter->watchdog_task);
3643
3644         if (netdev->reg_state == NETREG_REGISTERED)
3645                 unregister_netdev(netdev);
3646
3647         ixgbevf_clear_interrupt_scheme(adapter);
3648         ixgbevf_reset_interrupt_capability(adapter);
3649
3650         iounmap(adapter->io_addr);
3651         pci_release_regions(pdev);
3652
3653         hw_dbg(&adapter->hw, "Remove complete\n");
3654
3655         free_netdev(netdev);
3656
3657         if (!test_and_set_bit(__IXGBEVF_DISABLED, &adapter->state))
3658                 pci_disable_device(pdev);
3659 }
3660
3661 /**
3662  * ixgbevf_io_error_detected - called when PCI error is detected
3663  * @pdev: Pointer to PCI device
3664  * @state: The current pci connection state
3665  *
3666  * This function is called after a PCI bus error affecting
3667  * this device has been detected.
3668  */
3669 static pci_ers_result_t ixgbevf_io_error_detected(struct pci_dev *pdev,
3670                                                   pci_channel_state_t state)
3671 {
3672         struct net_device *netdev = pci_get_drvdata(pdev);
3673         struct ixgbevf_adapter *adapter = netdev_priv(netdev);
3674
3675         if (!test_bit(__IXGBEVF_WORK_INIT, &adapter->state))
3676                 return PCI_ERS_RESULT_DISCONNECT;
3677
3678         rtnl_lock();
3679         netif_device_detach(netdev);
3680
3681         if (state == pci_channel_io_perm_failure) {
3682                 rtnl_unlock();
3683                 return PCI_ERS_RESULT_DISCONNECT;
3684         }
3685
3686         if (netif_running(netdev))
3687                 ixgbevf_down(adapter);
3688
3689         if (!test_and_set_bit(__IXGBEVF_DISABLED, &adapter->state))
3690                 pci_disable_device(pdev);
3691         rtnl_unlock();
3692
3693         /* Request a slot slot reset. */
3694         return PCI_ERS_RESULT_NEED_RESET;
3695 }
3696
3697 /**
3698  * ixgbevf_io_slot_reset - called after the pci bus has been reset.
3699  * @pdev: Pointer to PCI device
3700  *
3701  * Restart the card from scratch, as if from a cold-boot. Implementation
3702  * resembles the first-half of the ixgbevf_resume routine.
3703  */
3704 static pci_ers_result_t ixgbevf_io_slot_reset(struct pci_dev *pdev)
3705 {
3706         struct net_device *netdev = pci_get_drvdata(pdev);
3707         struct ixgbevf_adapter *adapter = netdev_priv(netdev);
3708
3709         if (pci_enable_device_mem(pdev)) {
3710                 dev_err(&pdev->dev,
3711                         "Cannot re-enable PCI device after reset.\n");
3712                 return PCI_ERS_RESULT_DISCONNECT;
3713         }
3714
3715         smp_mb__before_atomic();
3716         clear_bit(__IXGBEVF_DISABLED, &adapter->state);
3717         pci_set_master(pdev);
3718
3719         ixgbevf_reset(adapter);
3720
3721         return PCI_ERS_RESULT_RECOVERED;
3722 }
3723
3724 /**
3725  * ixgbevf_io_resume - called when traffic can start flowing again.
3726  * @pdev: Pointer to PCI device
3727  *
3728  * This callback is called when the error recovery driver tells us that
3729  * its OK to resume normal operation. Implementation resembles the
3730  * second-half of the ixgbevf_resume routine.
3731  */
3732 static void ixgbevf_io_resume(struct pci_dev *pdev)
3733 {
3734         struct net_device *netdev = pci_get_drvdata(pdev);
3735         struct ixgbevf_adapter *adapter = netdev_priv(netdev);
3736
3737         if (netif_running(netdev))
3738                 ixgbevf_up(adapter);
3739
3740         netif_device_attach(netdev);
3741 }
3742
3743 /* PCI Error Recovery (ERS) */
3744 static const struct pci_error_handlers ixgbevf_err_handler = {
3745         .error_detected = ixgbevf_io_error_detected,
3746         .slot_reset = ixgbevf_io_slot_reset,
3747         .resume = ixgbevf_io_resume,
3748 };
3749
3750 static struct pci_driver ixgbevf_driver = {
3751         .name     = ixgbevf_driver_name,
3752         .id_table = ixgbevf_pci_tbl,
3753         .probe    = ixgbevf_probe,
3754         .remove   = ixgbevf_remove,
3755 #ifdef CONFIG_PM
3756         /* Power Management Hooks */
3757         .suspend  = ixgbevf_suspend,
3758         .resume   = ixgbevf_resume,
3759 #endif
3760         .shutdown = ixgbevf_shutdown,
3761         .err_handler = &ixgbevf_err_handler
3762 };
3763
3764 /**
3765  * ixgbevf_init_module - Driver Registration Routine
3766  *
3767  * ixgbevf_init_module is the first routine called when the driver is
3768  * loaded. All it does is register with the PCI subsystem.
3769  **/
3770 static int __init ixgbevf_init_module(void)
3771 {
3772         int ret;
3773         pr_info("%s - version %s\n", ixgbevf_driver_string,
3774                 ixgbevf_driver_version);
3775
3776         pr_info("%s\n", ixgbevf_copyright);
3777
3778         ret = pci_register_driver(&ixgbevf_driver);
3779         return ret;
3780 }
3781
3782 module_init(ixgbevf_init_module);
3783
3784 /**
3785  * ixgbevf_exit_module - Driver Exit Cleanup Routine
3786  *
3787  * ixgbevf_exit_module is called just before the driver is removed
3788  * from memory.
3789  **/
3790 static void __exit ixgbevf_exit_module(void)
3791 {
3792         pci_unregister_driver(&ixgbevf_driver);
3793 }
3794
3795 #ifdef DEBUG
3796 /**
3797  * ixgbevf_get_hw_dev_name - return device name string
3798  * used by hardware layer to print debugging information
3799  **/
3800 char *ixgbevf_get_hw_dev_name(struct ixgbe_hw *hw)
3801 {
3802         struct ixgbevf_adapter *adapter = hw->back;
3803         return adapter->netdev->name;
3804 }
3805
3806 #endif
3807 module_exit(ixgbevf_exit_module);
3808
3809 /* ixgbevf_main.c */