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

   1 /*******************************************************************************
   2
   3   Intel 82599 Virtual Function driver
   4   Copyright(c) 1999 - 2015 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, see <http://www.gnu.org/licenses/>.
  17
  18   The full GNU General Public License is included in this distribution in
  19   the file called "COPYING".
  20
  21   Contact Information:
  22   e1000-devel Mailing List <e1000-devel@lists.sourceforge.net>
  23   Intel Corporation, 5200 N.E. Elam Young Parkway, Hillsboro, OR 97124-6497
  24
  25 *******************************************************************************/
  26
  27 /******************************************************************************
  28  Copyright (c)2006 - 2007 Myricom, Inc. for some LRO specific code
  29 ******************************************************************************/
  30
  31 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
  32
  33 #include <linux/types.h>
  34 #include <linux/bitops.h>
  35 #include <linux/module.h>
  36 #include <linux/pci.h>
  37 #include <linux/netdevice.h>
  38 #include <linux/vmalloc.h>
  39 #include <linux/string.h>
  40 #include <linux/in.h>
  41 #include <linux/ip.h>
  42 #include <linux/tcp.h>
  43 #include <linux/sctp.h>
  44 #include <linux/ipv6.h>
  45 #include <linux/slab.h>
  46 #include <net/checksum.h>
  47 #include <net/ip6_checksum.h>
  48 #include <linux/ethtool.h>
  49 #include <linux/if.h>
  50 #include <linux/if_vlan.h>
  51 #include <linux/prefetch.h>
  52
  53 #include "ixgbevf.h"
  54
  55 const char ixgbevf_driver_name[] = "ixgbevf";
  56 static const char ixgbevf_driver_string[] =
  57         "Intel(R) 10 Gigabit PCI Express Virtual Function Network Driver";
  58
  59 #define DRV_VERSION "2.12.1-k"
  60 const char ixgbevf_driver_version[] = DRV_VERSION;
  61 static char ixgbevf_copyright[] =
  62         "Copyright (c) 2009 - 2015 Intel Corporation.";
  63
  64 static const struct ixgbevf_info *ixgbevf_info_tbl[] = {
  65         [board_82599_vf]        = &ixgbevf_82599_vf_info,
  66         [board_82599_vf_hv]     = &ixgbevf_82599_vf_hv_info,
  67         [board_X540_vf]         = &ixgbevf_X540_vf_info,
  68         [board_X540_vf_hv]      = &ixgbevf_X540_vf_hv_info,
  69         [board_X550_vf]         = &ixgbevf_X550_vf_info,
  70         [board_X550_vf_hv]      = &ixgbevf_X550_vf_hv_info,
  71         [board_X550EM_x_vf]     = &ixgbevf_X550EM_x_vf_info,
  72         [board_X550EM_x_vf_hv]  = &ixgbevf_X550EM_x_vf_hv_info,
  73 };
  74
  75 /* ixgbevf_pci_tbl - PCI Device ID Table
  76  *
  77  * Wildcard entries (PCI_ANY_ID) should come last
  78  * Last entry must be all 0s
  79  *
  80  * { Vendor ID, Device ID, SubVendor ID, SubDevice ID,
  81  *   Class, Class Mask, private data (not used) }
  82  */
  83 static const struct pci_device_id ixgbevf_pci_tbl[] = {
  84         {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82599_VF), board_82599_vf },
  85         {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82599_VF_HV), board_82599_vf_hv },
  86         {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_X540_VF), board_X540_vf },
  87         {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_X540_VF_HV), board_X540_vf_hv },
  88         {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_X550_VF), board_X550_vf },
  89         {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_X550_VF_HV), board_X550_vf_hv },
  90         {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_X550EM_X_VF), board_X550EM_x_vf },
  91         {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_X550EM_X_VF_HV), board_X550EM_x_vf_hv},
  92         /* required last entry */
  93         {0, }
  94 };
  95 MODULE_DEVICE_TABLE(pci, ixgbevf_pci_tbl);
  96
  97 MODULE_AUTHOR("Intel Corporation, <linux.nics@intel.com>");
  98 MODULE_DESCRIPTION("Intel(R) 10 Gigabit Virtual Function Network Driver");
  99 MODULE_LICENSE("GPL");
 100 MODULE_VERSION(DRV_VERSION);
 101
 102 #define DEFAULT_MSG_ENABLE (NETIF_MSG_DRV|NETIF_MSG_PROBE|NETIF_MSG_LINK)
 103 static int debug = -1;
 104 module_param(debug, int, 0);
 105 MODULE_PARM_DESC(debug, "Debug level (0=none,...,16=all)");
 106
 107 static struct workqueue_struct *ixgbevf_wq;
 108
 109 static void ixgbevf_service_event_schedule(struct ixgbevf_adapter *adapter)
 110 {
 111         if (!test_bit(__IXGBEVF_DOWN, &adapter->state) &&
 112             !test_bit(__IXGBEVF_REMOVING, &adapter->state) &&
 113             !test_and_set_bit(__IXGBEVF_SERVICE_SCHED, &adapter->state))
 114                 queue_work(ixgbevf_wq, &adapter->service_task);
 115 }
 116
 117 static void ixgbevf_service_event_complete(struct ixgbevf_adapter *adapter)
 118 {
 119         BUG_ON(!test_bit(__IXGBEVF_SERVICE_SCHED, &adapter->state));
 120
 121         /* flush memory to make sure state is correct before next watchdog */
 122         smp_mb__before_atomic();
 123         clear_bit(__IXGBEVF_SERVICE_SCHED, &adapter->state);
 124 }
 125
 126 /* forward decls */
 127 static void ixgbevf_queue_reset_subtask(struct ixgbevf_adapter *adapter);
 128 static void ixgbevf_set_itr(struct ixgbevf_q_vector *q_vector);
 129 static void ixgbevf_free_all_rx_resources(struct ixgbevf_adapter *adapter);
 130
 131 static void ixgbevf_remove_adapter(struct ixgbe_hw *hw)
 132 {
 133         struct ixgbevf_adapter *adapter = hw->back;
 134
 135         if (!hw->hw_addr)
 136                 return;
 137         hw->hw_addr = NULL;
 138         dev_err(&adapter->pdev->dev, "Adapter removed\n");
 139         if (test_bit(__IXGBEVF_SERVICE_INITED, &adapter->state))
 140                 ixgbevf_service_event_schedule(adapter);
 141 }
 142
 143 static void ixgbevf_check_remove(struct ixgbe_hw *hw, u32 reg)
 144 {
 145         u32 value;
 146
 147         /* The following check not only optimizes a bit by not
 148          * performing a read on the status register when the
 149          * register just read was a status register read that
 150          * returned IXGBE_FAILED_READ_REG. It also blocks any
 151          * potential recursion.
 152          */
 153         if (reg == IXGBE_VFSTATUS) {
 154                 ixgbevf_remove_adapter(hw);
 155                 return;
 156         }
 157         value = ixgbevf_read_reg(hw, IXGBE_VFSTATUS);
 158         if (value == IXGBE_FAILED_READ_REG)
 159                 ixgbevf_remove_adapter(hw);
 160 }
 161
 162 u32 ixgbevf_read_reg(struct ixgbe_hw *hw, u32 reg)
 163 {
 164         u8 __iomem *reg_addr = ACCESS_ONCE(hw->hw_addr);
 165         u32 value;
 166
 167         if (IXGBE_REMOVED(reg_addr))
 168                 return IXGBE_FAILED_READ_REG;
 169         value = readl(reg_addr + reg);
 170         if (unlikely(value == IXGBE_FAILED_READ_REG))
 171                 ixgbevf_check_remove(hw, reg);
 172         return value;
 173 }
 174
 175 /**
 176  * ixgbevf_set_ivar - set IVAR registers - maps interrupt causes to vectors
 177  * @adapter: pointer to adapter struct
 178  * @direction: 0 for Rx, 1 for Tx, -1 for other causes
 179  * @queue: queue to map the corresponding interrupt to
 180  * @msix_vector: the vector to map to the corresponding queue
 181  **/
 182 static void ixgbevf_set_ivar(struct ixgbevf_adapter *adapter, s8 direction,
 183                              u8 queue, u8 msix_vector)
 184 {
 185         u32 ivar, index;
 186         struct ixgbe_hw *hw = &adapter->hw;
 187
 188         if (direction == -1) {
 189                 /* other causes */
 190                 msix_vector |= IXGBE_IVAR_ALLOC_VAL;
 191                 ivar = IXGBE_READ_REG(hw, IXGBE_VTIVAR_MISC);
 192                 ivar &= ~0xFF;
 193                 ivar |= msix_vector;
 194                 IXGBE_WRITE_REG(hw, IXGBE_VTIVAR_MISC, ivar);
 195         } else {
 196                 /* Tx or Rx causes */
 197                 msix_vector |= IXGBE_IVAR_ALLOC_VAL;
 198                 index = ((16 * (queue & 1)) + (8 * direction));
 199                 ivar = IXGBE_READ_REG(hw, IXGBE_VTIVAR(queue >> 1));
 200                 ivar &= ~(0xFF << index);
 201                 ivar |= (msix_vector << index);
 202                 IXGBE_WRITE_REG(hw, IXGBE_VTIVAR(queue >> 1), ivar);
 203         }
 204 }
 205
 206 static void ixgbevf_unmap_and_free_tx_resource(struct ixgbevf_ring *tx_ring,
 207                                         struct ixgbevf_tx_buffer *tx_buffer)
 208 {
 209         if (tx_buffer->skb) {
 210                 dev_kfree_skb_any(tx_buffer->skb);
 211                 if (dma_unmap_len(tx_buffer, len))
 212                         dma_unmap_single(tx_ring->dev,
 213                                          dma_unmap_addr(tx_buffer, dma),
 214                                          dma_unmap_len(tx_buffer, len),
 215                                          DMA_TO_DEVICE);
 216         } else if (dma_unmap_len(tx_buffer, len)) {
 217                 dma_unmap_page(tx_ring->dev,
 218                                dma_unmap_addr(tx_buffer, dma),
 219                                dma_unmap_len(tx_buffer, len),
 220                                DMA_TO_DEVICE);
 221         }
 222         tx_buffer->next_to_watch = NULL;
 223         tx_buffer->skb = NULL;
 224         dma_unmap_len_set(tx_buffer, len, 0);
 225         /* tx_buffer must be completely set up in the transmit path */
 226 }
 227
 228 static u64 ixgbevf_get_tx_completed(struct ixgbevf_ring *ring)
 229 {
 230         return ring->stats.packets;
 231 }
 232
 233 static u32 ixgbevf_get_tx_pending(struct ixgbevf_ring *ring)
 234 {
 235         struct ixgbevf_adapter *adapter = netdev_priv(ring->netdev);
 236         struct ixgbe_hw *hw = &adapter->hw;
 237
 238         u32 head = IXGBE_READ_REG(hw, IXGBE_VFTDH(ring->reg_idx));
 239         u32 tail = IXGBE_READ_REG(hw, IXGBE_VFTDT(ring->reg_idx));
 240
 241         if (head != tail)
 242                 return (head < tail) ?
 243                         tail - head : (tail + ring->count - head);
 244
 245         return 0;
 246 }
 247
 248 static inline bool ixgbevf_check_tx_hang(struct ixgbevf_ring *tx_ring)
 249 {
 250         u32 tx_done = ixgbevf_get_tx_completed(tx_ring);
 251         u32 tx_done_old = tx_ring->tx_stats.tx_done_old;
 252         u32 tx_pending = ixgbevf_get_tx_pending(tx_ring);
 253
 254         clear_check_for_tx_hang(tx_ring);
 255
 256         /* Check for a hung queue, but be thorough. This verifies
 257          * that a transmit has been completed since the previous
 258          * check AND there is at least one packet pending. The
 259          * ARMED bit is set to indicate a potential hang.
 260          */
 261         if ((tx_done_old == tx_done) && tx_pending) {
 262                 /* make sure it is true for two checks in a row */
 263                 return test_and_set_bit(__IXGBEVF_HANG_CHECK_ARMED,
 264                                         &tx_ring->state);
 265         }
 266         /* reset the countdown */
 267         clear_bit(__IXGBEVF_HANG_CHECK_ARMED, &tx_ring->state);
 268
 269         /* update completed stats and continue */
 270         tx_ring->tx_stats.tx_done_old = tx_done;
 271
 272         return false;
 273 }
 274
 275 static void ixgbevf_tx_timeout_reset(struct ixgbevf_adapter *adapter)
 276 {
 277         /* Do the reset outside of interrupt context */
 278         if (!test_bit(__IXGBEVF_DOWN, &adapter->state)) {
 279                 set_bit(__IXGBEVF_RESET_REQUESTED, &adapter->state);
 280                 ixgbevf_service_event_schedule(adapter);
 281         }
 282 }
 283
 284 /**
 285  * ixgbevf_tx_timeout - Respond to a Tx Hang
 286  * @netdev: network interface device structure
 287  **/
 288 static void ixgbevf_tx_timeout(struct net_device *netdev)
 289 {
 290         struct ixgbevf_adapter *adapter = netdev_priv(netdev);
 291
 292         ixgbevf_tx_timeout_reset(adapter);
 293 }
 294
 295 /**
 296  * ixgbevf_clean_tx_irq - Reclaim resources after transmit completes
 297  * @q_vector: board private structure
 298  * @tx_ring: tx ring to clean
 299  * @napi_budget: Used to determine if we are in netpoll
 300  **/
 301 static bool ixgbevf_clean_tx_irq(struct ixgbevf_q_vector *q_vector,
 302                                  struct ixgbevf_ring *tx_ring, int napi_budget)
 303 {
 304         struct ixgbevf_adapter *adapter = q_vector->adapter;
 305         struct ixgbevf_tx_buffer *tx_buffer;
 306         union ixgbe_adv_tx_desc *tx_desc;
 307         unsigned int total_bytes = 0, total_packets = 0;
 308         unsigned int budget = tx_ring->count / 2;
 309         unsigned int i = tx_ring->next_to_clean;
 310
 311         if (test_bit(__IXGBEVF_DOWN, &adapter->state))
 312                 return true;
 313
 314         tx_buffer = &tx_ring->tx_buffer_info[i];
 315         tx_desc = IXGBEVF_TX_DESC(tx_ring, i);
 316         i -= tx_ring->count;
 317
 318         do {
 319                 union ixgbe_adv_tx_desc *eop_desc = tx_buffer->next_to_watch;
 320
 321                 /* if next_to_watch is not set then there is no work pending */
 322                 if (!eop_desc)
 323                         break;
 324
 325                 /* prevent any other reads prior to eop_desc */
 326                 read_barrier_depends();
 327
 328                 /* if DD is not set pending work has not been completed */
 329                 if (!(eop_desc->wb.status & cpu_to_le32(IXGBE_TXD_STAT_DD)))
 330                         break;
 331
 332                 /* clear next_to_watch to prevent false hangs */
 333                 tx_buffer->next_to_watch = NULL;
 334
 335                 /* update the statistics for this packet */
 336                 total_bytes += tx_buffer->bytecount;
 337                 total_packets += tx_buffer->gso_segs;
 338
 339                 /* free the skb */
 340                 napi_consume_skb(tx_buffer->skb, napi_budget);
 341
 342                 /* unmap skb header data */
 343                 dma_unmap_single(tx_ring->dev,
 344                                  dma_unmap_addr(tx_buffer, dma),
 345                                  dma_unmap_len(tx_buffer, len),
 346                                  DMA_TO_DEVICE);
 347
 348                 /* clear tx_buffer data */
 349                 tx_buffer->skb = NULL;
 350                 dma_unmap_len_set(tx_buffer, len, 0);
 351
 352                 /* unmap remaining buffers */
 353                 while (tx_desc != eop_desc) {
 354                         tx_buffer++;
 355                         tx_desc++;
 356                         i++;
 357                         if (unlikely(!i)) {
 358                                 i -= tx_ring->count;
 359                                 tx_buffer = tx_ring->tx_buffer_info;
 360                                 tx_desc = IXGBEVF_TX_DESC(tx_ring, 0);
 361                         }
 362
 363                         /* unmap any remaining paged data */
 364                         if (dma_unmap_len(tx_buffer, len)) {
 365                                 dma_unmap_page(tx_ring->dev,
 366                                                dma_unmap_addr(tx_buffer, dma),
 367                                                dma_unmap_len(tx_buffer, len),
 368                                                DMA_TO_DEVICE);
 369                                 dma_unmap_len_set(tx_buffer, len, 0);
 370                         }
 371                 }
 372
 373                 /* move us one more past the eop_desc for start of next pkt */
 374                 tx_buffer++;
 375                 tx_desc++;
 376                 i++;
 377                 if (unlikely(!i)) {
 378                         i -= tx_ring->count;
 379                         tx_buffer = tx_ring->tx_buffer_info;
 380                         tx_desc = IXGBEVF_TX_DESC(tx_ring, 0);
 381                 }
 382
 383                 /* issue prefetch for next Tx descriptor */
 384                 prefetch(tx_desc);
 385
 386                 /* update budget accounting */
 387                 budget--;
 388         } while (likely(budget));
 389
 390         i += tx_ring->count;
 391         tx_ring->next_to_clean = i;
 392         u64_stats_update_begin(&tx_ring->syncp);
 393         tx_ring->stats.bytes += total_bytes;
 394         tx_ring->stats.packets += total_packets;
 395         u64_stats_update_end(&tx_ring->syncp);
 396         q_vector->tx.total_bytes += total_bytes;
 397         q_vector->tx.total_packets += total_packets;
 398
 399         if (check_for_tx_hang(tx_ring) && ixgbevf_check_tx_hang(tx_ring)) {
 400                 struct ixgbe_hw *hw = &adapter->hw;
 401                 union ixgbe_adv_tx_desc *eop_desc;
 402
 403                 eop_desc = tx_ring->tx_buffer_info[i].next_to_watch;
 404
 405                 pr_err("Detected Tx Unit Hang\n"
 406                        "  Tx Queue             <%d>\n"
 407                        "  TDH, TDT             <%x>, <%x>\n"
 408                        "  next_to_use          <%x>\n"
 409                        "  next_to_clean        <%x>\n"
 410                        "tx_buffer_info[next_to_clean]\n"
 411                        "  next_to_watch        <%p>\n"
 412                        "  eop_desc->wb.status  <%x>\n"
 413                        "  time_stamp           <%lx>\n"
 414                        "  jiffies              <%lx>\n",
 415                        tx_ring->queue_index,
 416                        IXGBE_READ_REG(hw, IXGBE_VFTDH(tx_ring->reg_idx)),
 417                        IXGBE_READ_REG(hw, IXGBE_VFTDT(tx_ring->reg_idx)),
 418                        tx_ring->next_to_use, i,
 419                        eop_desc, (eop_desc ? eop_desc->wb.status : 0),
 420                        tx_ring->tx_buffer_info[i].time_stamp, jiffies);
 421
 422                 netif_stop_subqueue(tx_ring->netdev, tx_ring->queue_index);
 423
 424                 /* schedule immediate reset if we believe we hung */
 425                 ixgbevf_tx_timeout_reset(adapter);
 426
 427                 return true;
 428         }
 429
 430 #define TX_WAKE_THRESHOLD (DESC_NEEDED * 2)
 431         if (unlikely(total_packets && netif_carrier_ok(tx_ring->netdev) &&
 432                      (ixgbevf_desc_unused(tx_ring) >= TX_WAKE_THRESHOLD))) {
 433                 /* Make sure that anybody stopping the queue after this
 434                  * sees the new next_to_clean.
 435                  */
 436                 smp_mb();
 437
 438                 if (__netif_subqueue_stopped(tx_ring->netdev,
 439                                              tx_ring->queue_index) &&
 440                     !test_bit(__IXGBEVF_DOWN, &adapter->state)) {
 441                         netif_wake_subqueue(tx_ring->netdev,
 442                                             tx_ring->queue_index);
 443                         ++tx_ring->tx_stats.restart_queue;
 444                 }
 445         }
 446
 447         return !!budget;
 448 }
 449
 450 /**
 451  * ixgbevf_rx_skb - Helper function to determine proper Rx method
 452  * @q_vector: structure containing interrupt and ring information
 453  * @skb: packet to send up
 454  **/
 455 static void ixgbevf_rx_skb(struct ixgbevf_q_vector *q_vector,
 456                            struct sk_buff *skb)
 457 {
 458 #ifdef CONFIG_NET_RX_BUSY_POLL
 459         skb_mark_napi_id(skb, &q_vector->napi);
 460
 461         if (ixgbevf_qv_busy_polling(q_vector)) {
 462                 netif_receive_skb(skb);
 463                 /* exit early if we busy polled */
 464                 return;
 465         }
 466 #endif /* CONFIG_NET_RX_BUSY_POLL */
 467
 468         napi_gro_receive(&q_vector->napi, skb);
 469 }
 470
 471 #define IXGBE_RSS_L4_TYPES_MASK \
 472         ((1ul << IXGBE_RXDADV_RSSTYPE_IPV4_TCP) | \
 473          (1ul << IXGBE_RXDADV_RSSTYPE_IPV4_UDP) | \
 474          (1ul << IXGBE_RXDADV_RSSTYPE_IPV6_TCP) | \
 475          (1ul << IXGBE_RXDADV_RSSTYPE_IPV6_UDP))
 476
 477 static inline void ixgbevf_rx_hash(struct ixgbevf_ring *ring,
 478                                    union ixgbe_adv_rx_desc *rx_desc,
 479                                    struct sk_buff *skb)
 480 {
 481         u16 rss_type;
 482
 483         if (!(ring->netdev->features & NETIF_F_RXHASH))
 484                 return;
 485
 486         rss_type = le16_to_cpu(rx_desc->wb.lower.lo_dword.hs_rss.pkt_info) &
 487                    IXGBE_RXDADV_RSSTYPE_MASK;
 488
 489         if (!rss_type)
 490                 return;
 491
 492         skb_set_hash(skb, le32_to_cpu(rx_desc->wb.lower.hi_dword.rss),
 493                      (IXGBE_RSS_L4_TYPES_MASK & (1ul << rss_type)) ?
 494                      PKT_HASH_TYPE_L4 : PKT_HASH_TYPE_L3);
 495 }
 496
 497 /**
 498  * ixgbevf_rx_checksum - indicate in skb if hw indicated a good cksum
 499  * @ring: structure containig ring specific data
 500  * @rx_desc: current Rx descriptor being processed
 501  * @skb: skb currently being received and modified
 502  **/
 503 static inline void ixgbevf_rx_checksum(struct ixgbevf_ring *ring,
 504                                        union ixgbe_adv_rx_desc *rx_desc,
 505                                        struct sk_buff *skb)
 506 {
 507         skb_checksum_none_assert(skb);
 508
 509         /* Rx csum disabled */
 510         if (!(ring->netdev->features & NETIF_F_RXCSUM))
 511                 return;
 512
 513         /* if IP and error */
 514         if (ixgbevf_test_staterr(rx_desc, IXGBE_RXD_STAT_IPCS) &&
 515             ixgbevf_test_staterr(rx_desc, IXGBE_RXDADV_ERR_IPE)) {
 516                 ring->rx_stats.csum_err++;
 517                 return;
 518         }
 519
 520         if (!ixgbevf_test_staterr(rx_desc, IXGBE_RXD_STAT_L4CS))
 521                 return;
 522
 523         if (ixgbevf_test_staterr(rx_desc, IXGBE_RXDADV_ERR_TCPE)) {
 524                 ring->rx_stats.csum_err++;
 525                 return;
 526         }
 527
 528         /* It must be a TCP or UDP packet with a valid checksum */
 529         skb->ip_summed = CHECKSUM_UNNECESSARY;
 530 }
 531
 532 /**
 533  * ixgbevf_process_skb_fields - Populate skb header fields from Rx descriptor
 534  * @rx_ring: rx descriptor ring packet is being transacted on
 535  * @rx_desc: pointer to the EOP Rx descriptor
 536  * @skb: pointer to current skb being populated
 537  *
 538  * This function checks the ring, descriptor, and packet information in
 539  * order to populate the checksum, VLAN, protocol, and other fields within
 540  * the skb.
 541  **/
 542 static void ixgbevf_process_skb_fields(struct ixgbevf_ring *rx_ring,
 543                                        union ixgbe_adv_rx_desc *rx_desc,
 544                                        struct sk_buff *skb)
 545 {
 546         ixgbevf_rx_hash(rx_ring, rx_desc, skb);
 547         ixgbevf_rx_checksum(rx_ring, rx_desc, skb);
 548
 549         if (ixgbevf_test_staterr(rx_desc, IXGBE_RXD_STAT_VP)) {
 550                 u16 vid = le16_to_cpu(rx_desc->wb.upper.vlan);
 551                 unsigned long *active_vlans = netdev_priv(rx_ring->netdev);
 552
 553                 if (test_bit(vid & VLAN_VID_MASK, active_vlans))
 554                         __vlan_hwaccel_put_tag(skb, htons(ETH_P_8021Q), vid);
 555         }
 556
 557         skb->protocol = eth_type_trans(skb, rx_ring->netdev);
 558 }
 559
 560 /**
 561  * ixgbevf_is_non_eop - process handling of non-EOP buffers
 562  * @rx_ring: Rx ring being processed
 563  * @rx_desc: Rx descriptor for current buffer
 564  * @skb: current socket buffer containing buffer in progress
 565  *
 566  * This function updates next to clean.  If the buffer is an EOP buffer
 567  * this function exits returning false, otherwise it will place the
 568  * sk_buff in the next buffer to be chained and return true indicating
 569  * that this is in fact a non-EOP buffer.
 570  **/
 571 static bool ixgbevf_is_non_eop(struct ixgbevf_ring *rx_ring,
 572                                union ixgbe_adv_rx_desc *rx_desc)
 573 {
 574         u32 ntc = rx_ring->next_to_clean + 1;
 575
 576         /* fetch, update, and store next to clean */
 577         ntc = (ntc < rx_ring->count) ? ntc : 0;
 578         rx_ring->next_to_clean = ntc;
 579
 580         prefetch(IXGBEVF_RX_DESC(rx_ring, ntc));
 581
 582         if (likely(ixgbevf_test_staterr(rx_desc, IXGBE_RXD_STAT_EOP)))
 583                 return false;
 584
 585         return true;
 586 }
 587
 588 static bool ixgbevf_alloc_mapped_page(struct ixgbevf_ring *rx_ring,
 589                                       struct ixgbevf_rx_buffer *bi)
 590 {
 591         struct page *page = bi->page;
 592         dma_addr_t dma = bi->dma;
 593
 594         /* since we are recycling buffers we should seldom need to alloc */
 595         if (likely(page))
 596                 return true;
 597
 598         /* alloc new page for storage */
 599         page = dev_alloc_page();
 600         if (unlikely(!page)) {
 601                 rx_ring->rx_stats.alloc_rx_page_failed++;
 602                 return false;
 603         }
 604
 605         /* map page for use */
 606         dma = dma_map_page(rx_ring->dev, page, 0,
 607                            PAGE_SIZE, DMA_FROM_DEVICE);
 608
 609         /* if mapping failed free memory back to system since
 610          * there isn't much point in holding memory we can't use
 611          */
 612         if (dma_mapping_error(rx_ring->dev, dma)) {
 613                 __free_page(page);
 614
 615                 rx_ring->rx_stats.alloc_rx_buff_failed++;
 616                 return false;
 617         }
 618
 619         bi->dma = dma;
 620         bi->page = page;
 621         bi->page_offset = 0;
 622
 623         return true;
 624 }
 625
 626 /**
 627  * ixgbevf_alloc_rx_buffers - Replace used receive buffers; packet split
 628  * @rx_ring: rx descriptor ring (for a specific queue) to setup buffers on
 629  * @cleaned_count: number of buffers to replace
 630  **/
 631 static void ixgbevf_alloc_rx_buffers(struct ixgbevf_ring *rx_ring,
 632                                      u16 cleaned_count)
 633 {
 634         union ixgbe_adv_rx_desc *rx_desc;
 635         struct ixgbevf_rx_buffer *bi;
 636         unsigned int i = rx_ring->next_to_use;
 637
 638         /* nothing to do or no valid netdev defined */
 639         if (!cleaned_count || !rx_ring->netdev)
 640                 return;
 641
 642         rx_desc = IXGBEVF_RX_DESC(rx_ring, i);
 643         bi = &rx_ring->rx_buffer_info[i];
 644         i -= rx_ring->count;
 645
 646         do {
 647                 if (!ixgbevf_alloc_mapped_page(rx_ring, bi))
 648                         break;
 649
 650                 /* Refresh the desc even if pkt_addr didn't change
 651                  * because each write-back erases this info.
 652                  */
 653                 rx_desc->read.pkt_addr = cpu_to_le64(bi->dma + bi->page_offset);
 654
 655                 rx_desc++;
 656                 bi++;
 657                 i++;
 658                 if (unlikely(!i)) {
 659                         rx_desc = IXGBEVF_RX_DESC(rx_ring, 0);
 660                         bi = rx_ring->rx_buffer_info;
 661                         i -= rx_ring->count;
 662                 }
 663
 664                 /* clear the hdr_addr for the next_to_use descriptor */
 665                 rx_desc->read.hdr_addr = 0;
 666
 667                 cleaned_count--;
 668         } while (cleaned_count);
 669
 670         i += rx_ring->count;
 671
 672         if (rx_ring->next_to_use != i) {
 673                 /* record the next descriptor to use */
 674                 rx_ring->next_to_use = i;
 675
 676                 /* update next to alloc since we have filled the ring */
 677                 rx_ring->next_to_alloc = i;
 678
 679                 /* Force memory writes to complete before letting h/w
 680                  * know there are new descriptors to fetch.  (Only
 681                  * applicable for weak-ordered memory model archs,
 682                  * such as IA-64).
 683                  */
 684                 wmb();
 685                 ixgbevf_write_tail(rx_ring, i);
 686         }
 687 }
 688
 689 /**
 690  * ixgbevf_cleanup_headers - Correct corrupted or empty headers
 691  * @rx_ring: rx descriptor ring packet is being transacted on
 692  * @rx_desc: pointer to the EOP Rx descriptor
 693  * @skb: pointer to current skb being fixed
 694  *
 695  * Check for corrupted packet headers caused by senders on the local L2
 696  * embedded NIC switch not setting up their Tx Descriptors right.  These
 697  * should be very rare.
 698  *
 699  * Also address the case where we are pulling data in on pages only
 700  * and as such no data is present in the skb header.
 701  *
 702  * In addition if skb is not at least 60 bytes we need to pad it so that
 703  * it is large enough to qualify as a valid Ethernet frame.
 704  *
 705  * Returns true if an error was encountered and skb was freed.
 706  **/
 707 static bool ixgbevf_cleanup_headers(struct ixgbevf_ring *rx_ring,
 708                                     union ixgbe_adv_rx_desc *rx_desc,
 709                                     struct sk_buff *skb)
 710 {
 711         /* verify that the packet does not have any known errors */
 712         if (unlikely(ixgbevf_test_staterr(rx_desc,
 713                                           IXGBE_RXDADV_ERR_FRAME_ERR_MASK))) {
 714                 struct net_device *netdev = rx_ring->netdev;
 715
 716                 if (!(netdev->features & NETIF_F_RXALL)) {
 717                         dev_kfree_skb_any(skb);
 718                         return true;
 719                 }
 720         }
 721
 722         /* if eth_skb_pad returns an error the skb was freed */
 723         if (eth_skb_pad(skb))
 724                 return true;
 725
 726         return false;
 727 }
 728
 729 /**
 730  * ixgbevf_reuse_rx_page - page flip buffer and store it back on the ring
 731  * @rx_ring: rx descriptor ring to store buffers on
 732  * @old_buff: donor buffer to have page reused
 733  *
 734  * Synchronizes page for reuse by the adapter
 735  **/
 736 static void ixgbevf_reuse_rx_page(struct ixgbevf_ring *rx_ring,
 737                                   struct ixgbevf_rx_buffer *old_buff)
 738 {
 739         struct ixgbevf_rx_buffer *new_buff;
 740         u16 nta = rx_ring->next_to_alloc;
 741
 742         new_buff = &rx_ring->rx_buffer_info[nta];
 743
 744         /* update, and store next to alloc */
 745         nta++;
 746         rx_ring->next_to_alloc = (nta < rx_ring->count) ? nta : 0;
 747
 748         /* transfer page from old buffer to new buffer */
 749         new_buff->page = old_buff->page;
 750         new_buff->dma = old_buff->dma;
 751         new_buff->page_offset = old_buff->page_offset;
 752
 753         /* sync the buffer for use by the device */
 754         dma_sync_single_range_for_device(rx_ring->dev, new_buff->dma,
 755                                          new_buff->page_offset,
 756                                          IXGBEVF_RX_BUFSZ,
 757                                          DMA_FROM_DEVICE);
 758 }
 759
 760 static inline bool ixgbevf_page_is_reserved(struct page *page)
 761 {
 762         return (page_to_nid(page) != numa_mem_id()) || page_is_pfmemalloc(page);
 763 }
 764
 765 /**
 766  * ixgbevf_add_rx_frag - Add contents of Rx buffer to sk_buff
 767  * @rx_ring: rx descriptor ring to transact packets on
 768  * @rx_buffer: buffer containing page to add
 769  * @rx_desc: descriptor containing length of buffer written by hardware
 770  * @skb: sk_buff to place the data into
 771  *
 772  * This function will add the data contained in rx_buffer->page to the skb.
 773  * This is done either through a direct copy if the data in the buffer is
 774  * less than the skb header size, otherwise it will just attach the page as
 775  * a frag to the skb.
 776  *
 777  * The function will then update the page offset if necessary and return
 778  * true if the buffer can be reused by the adapter.
 779  **/
 780 static bool ixgbevf_add_rx_frag(struct ixgbevf_ring *rx_ring,
 781                                 struct ixgbevf_rx_buffer *rx_buffer,
 782                                 union ixgbe_adv_rx_desc *rx_desc,
 783                                 struct sk_buff *skb)
 784 {
 785         struct page *page = rx_buffer->page;
 786         unsigned char *va = page_address(page) + rx_buffer->page_offset;
 787         unsigned int size = le16_to_cpu(rx_desc->wb.upper.length);
 788 #if (PAGE_SIZE < 8192)
 789         unsigned int truesize = IXGBEVF_RX_BUFSZ;
 790 #else
 791         unsigned int truesize = ALIGN(size, L1_CACHE_BYTES);
 792 #endif
 793         unsigned int pull_len;
 794
 795         if (unlikely(skb_is_nonlinear(skb)))
 796                 goto add_tail_frag;
 797
 798         if (likely(size <= IXGBEVF_RX_HDR_SIZE)) {
 799                 memcpy(__skb_put(skb, size), va, ALIGN(size, sizeof(long)));
 800
 801                 /* page is not reserved, we can reuse buffer as is */
 802                 if (likely(!ixgbevf_page_is_reserved(page)))
 803                         return true;
 804
 805                 /* this page cannot be reused so discard it */
 806                 put_page(page);
 807                 return false;
 808         }
 809
 810         /* we need the header to contain the greater of either ETH_HLEN or
 811          * 60 bytes if the skb->len is less than 60 for skb_pad.
 812          */
 813         pull_len = eth_get_headlen(va, IXGBEVF_RX_HDR_SIZE);
 814
 815         /* align pull length to size of long to optimize memcpy performance */
 816         memcpy(__skb_put(skb, pull_len), va, ALIGN(pull_len, sizeof(long)));
 817
 818         /* update all of the pointers */
 819         va += pull_len;
 820         size -= pull_len;
 821
 822 add_tail_frag:
 823         skb_add_rx_frag(skb, skb_shinfo(skb)->nr_frags, page,
 824                         (unsigned long)va & ~PAGE_MASK, size, truesize);
 825
 826         /* avoid re-using remote pages */
 827         if (unlikely(ixgbevf_page_is_reserved(page)))
 828                 return false;
 829
 830 #if (PAGE_SIZE < 8192)
 831         /* if we are only owner of page we can reuse it */
 832         if (unlikely(page_count(page) != 1))
 833                 return false;
 834
 835         /* flip page offset to other buffer */
 836         rx_buffer->page_offset ^= IXGBEVF_RX_BUFSZ;
 837
 838 #else
 839         /* move offset up to the next cache line */
 840         rx_buffer->page_offset += truesize;
 841
 842         if (rx_buffer->page_offset > (PAGE_SIZE - IXGBEVF_RX_BUFSZ))
 843                 return false;
 844
 845 #endif
 846         /* Even if we own the page, we are not allowed to use atomic_set()
 847          * This would break get_page_unless_zero() users.
 848          */
 849         page_ref_inc(page);
 850
 851         return true;
 852 }
 853
 854 static struct sk_buff *ixgbevf_fetch_rx_buffer(struct ixgbevf_ring *rx_ring,
 855                                                union ixgbe_adv_rx_desc *rx_desc,
 856                                                struct sk_buff *skb)
 857 {
 858         struct ixgbevf_rx_buffer *rx_buffer;
 859         struct page *page;
 860
 861         rx_buffer = &rx_ring->rx_buffer_info[rx_ring->next_to_clean];
 862         page = rx_buffer->page;
 863         prefetchw(page);
 864
 865         if (likely(!skb)) {
 866                 void *page_addr = page_address(page) +
 867                                   rx_buffer->page_offset;
 868
 869                 /* prefetch first cache line of first page */
 870                 prefetch(page_addr);
 871 #if L1_CACHE_BYTES < 128
 872                 prefetch(page_addr + L1_CACHE_BYTES);
 873 #endif
 874
 875                 /* allocate a skb to store the frags */
 876                 skb = netdev_alloc_skb_ip_align(rx_ring->netdev,
 877                                                 IXGBEVF_RX_HDR_SIZE);
 878                 if (unlikely(!skb)) {
 879                         rx_ring->rx_stats.alloc_rx_buff_failed++;
 880                         return NULL;
 881                 }
 882
 883                 /* we will be copying header into skb->data in
 884                  * pskb_may_pull so it is in our interest to prefetch
 885                  * it now to avoid a possible cache miss
 886                  */
 887                 prefetchw(skb->data);
 888         }
 889
 890         /* we are reusing so sync this buffer for CPU use */
 891         dma_sync_single_range_for_cpu(rx_ring->dev,
 892                                       rx_buffer->dma,
 893                                       rx_buffer->page_offset,
 894                                       IXGBEVF_RX_BUFSZ,
 895                                       DMA_FROM_DEVICE);
 896
 897         /* pull page into skb */
 898         if (ixgbevf_add_rx_frag(rx_ring, rx_buffer, rx_desc, skb)) {
 899                 /* hand second half of page back to the ring */
 900                 ixgbevf_reuse_rx_page(rx_ring, rx_buffer);
 901         } else {
 902                 /* we are not reusing the buffer so unmap it */
 903                 dma_unmap_page(rx_ring->dev, rx_buffer->dma,
 904                                PAGE_SIZE, DMA_FROM_DEVICE);
 905         }
 906
 907         /* clear contents of buffer_info */
 908         rx_buffer->dma = 0;
 909         rx_buffer->page = NULL;
 910
 911         return skb;
 912 }
 913
 914 static inline void ixgbevf_irq_enable_queues(struct ixgbevf_adapter *adapter,
 915                                              u32 qmask)
 916 {
 917         struct ixgbe_hw *hw = &adapter->hw;
 918
 919         IXGBE_WRITE_REG(hw, IXGBE_VTEIMS, qmask);
 920 }
 921
 922 static int ixgbevf_clean_rx_irq(struct ixgbevf_q_vector *q_vector,
 923                                 struct ixgbevf_ring *rx_ring,
 924                                 int budget)
 925 {
 926         unsigned int total_rx_bytes = 0, total_rx_packets = 0;
 927         u16 cleaned_count = ixgbevf_desc_unused(rx_ring);
 928         struct sk_buff *skb = rx_ring->skb;
 929
 930         while (likely(total_rx_packets < budget)) {
 931                 union ixgbe_adv_rx_desc *rx_desc;
 932
 933                 /* return some buffers to hardware, one at a time is too slow */
 934                 if (cleaned_count >= IXGBEVF_RX_BUFFER_WRITE) {
 935                         ixgbevf_alloc_rx_buffers(rx_ring, cleaned_count);
 936                         cleaned_count = 0;
 937                 }
 938
 939                 rx_desc = IXGBEVF_RX_DESC(rx_ring, rx_ring->next_to_clean);
 940
 941                 if (!ixgbevf_test_staterr(rx_desc, IXGBE_RXD_STAT_DD))
 942                         break;
 943
 944                 /* This memory barrier is needed to keep us from reading
 945                  * any other fields out of the rx_desc until we know the
 946                  * RXD_STAT_DD bit is set
 947                  */
 948                 rmb();
 949
 950                 /* retrieve a buffer from the ring */
 951                 skb = ixgbevf_fetch_rx_buffer(rx_ring, rx_desc, skb);
 952
 953                 /* exit if we failed to retrieve a buffer */
 954                 if (!skb)
 955                         break;
 956
 957                 cleaned_count++;
 958
 959                 /* fetch next buffer in frame if non-eop */
 960                 if (ixgbevf_is_non_eop(rx_ring, rx_desc))
 961                         continue;
 962
 963                 /* verify the packet layout is correct */
 964                 if (ixgbevf_cleanup_headers(rx_ring, rx_desc, skb)) {
 965                         skb = NULL;
 966                         continue;
 967                 }
 968
 969                 /* probably a little skewed due to removing CRC */
 970                 total_rx_bytes += skb->len;
 971
 972                 /* Workaround hardware that can't do proper VEPA multicast
 973                  * source pruning.
 974                  */
 975                 if ((skb->pkt_type == PACKET_BROADCAST ||
 976                      skb->pkt_type == PACKET_MULTICAST) &&
 977                     ether_addr_equal(rx_ring->netdev->dev_addr,
 978                                      eth_hdr(skb)->h_source)) {
 979                         dev_kfree_skb_irq(skb);
 980                         continue;
 981                 }
 982
 983                 /* populate checksum, VLAN, and protocol */
 984                 ixgbevf_process_skb_fields(rx_ring, rx_desc, skb);
 985
 986                 ixgbevf_rx_skb(q_vector, skb);
 987
 988                 /* reset skb pointer */
 989                 skb = NULL;
 990
 991                 /* update budget accounting */
 992                 total_rx_packets++;
 993         }
 994
 995         /* place incomplete frames back on ring for completion */
 996         rx_ring->skb = skb;
 997
 998         u64_stats_update_begin(&rx_ring->syncp);
 999         rx_ring->stats.packets += total_rx_packets;
1000         rx_ring->stats.bytes += total_rx_bytes;
1001         u64_stats_update_end(&rx_ring->syncp);
1002         q_vector->rx.total_packets += total_rx_packets;
1003         q_vector->rx.total_bytes += total_rx_bytes;
1004
1005         return total_rx_packets;
1006 }
1007
1008 /**
1009  * ixgbevf_poll - NAPI polling calback
1010  * @napi: napi struct with our devices info in it
1011  * @budget: amount of work driver is allowed to do this pass, in packets
1012  *
1013  * This function will clean more than one or more rings associated with a
1014  * q_vector.
1015  **/
1016 static int ixgbevf_poll(struct napi_struct *napi, int budget)
1017 {
1018         struct ixgbevf_q_vector *q_vector =
1019                 container_of(napi, struct ixgbevf_q_vector, napi);
1020         struct ixgbevf_adapter *adapter = q_vector->adapter;
1021         struct ixgbevf_ring *ring;
1022         int per_ring_budget, work_done = 0;
1023         bool clean_complete = true;
1024
1025         ixgbevf_for_each_ring(ring, q_vector->tx) {
1026                 if (!ixgbevf_clean_tx_irq(q_vector, ring, budget))
1027                         clean_complete = false;
1028         }
1029
1030         if (budget <= 0)
1031                 return budget;
1032 #ifdef CONFIG_NET_RX_BUSY_POLL
1033         if (!ixgbevf_qv_lock_napi(q_vector))
1034                 return budget;
1035 #endif
1036
1037         /* attempt to distribute budget to each queue fairly, but don't allow
1038          * the budget to go below 1 because we'll exit polling
1039          */
1040         if (q_vector->rx.count > 1)
1041                 per_ring_budget = max(budget/q_vector->rx.count, 1);
1042         else
1043                 per_ring_budget = budget;
1044
1045         ixgbevf_for_each_ring(ring, q_vector->rx) {
1046                 int cleaned = ixgbevf_clean_rx_irq(q_vector, ring,
1047                                                    per_ring_budget);
1048                 work_done += cleaned;
1049                 if (cleaned >= per_ring_budget)
1050                         clean_complete = false;
1051         }
1052
1053 #ifdef CONFIG_NET_RX_BUSY_POLL
1054         ixgbevf_qv_unlock_napi(q_vector);
1055 #endif
1056
1057         /* If all work not completed, return budget and keep polling */
1058         if (!clean_complete)
1059                 return budget;
1060         /* all work done, exit the polling mode */
1061         napi_complete_done(napi, work_done);
1062         if (adapter->rx_itr_setting == 1)
1063                 ixgbevf_set_itr(q_vector);
1064         if (!test_bit(__IXGBEVF_DOWN, &adapter->state) &&
1065             !test_bit(__IXGBEVF_REMOVING, &adapter->state))
1066                 ixgbevf_irq_enable_queues(adapter,
1067                                           BIT(q_vector->v_idx));
1068
1069         return 0;
1070 }
1071
1072 /**
1073  * ixgbevf_write_eitr - write VTEITR register in hardware specific way
1074  * @q_vector: structure containing interrupt and ring information
1075  **/
1076 void ixgbevf_write_eitr(struct ixgbevf_q_vector *q_vector)
1077 {
1078         struct ixgbevf_adapter *adapter = q_vector->adapter;
1079         struct ixgbe_hw *hw = &adapter->hw;
1080         int v_idx = q_vector->v_idx;
1081         u32 itr_reg = q_vector->itr & IXGBE_MAX_EITR;
1082
1083         /* set the WDIS bit to not clear the timer bits and cause an
1084          * immediate assertion of the interrupt
1085          */
1086         itr_reg |= IXGBE_EITR_CNT_WDIS;
1087
1088         IXGBE_WRITE_REG(hw, IXGBE_VTEITR(v_idx), itr_reg);
1089 }
1090
1091 #ifdef CONFIG_NET_RX_BUSY_POLL
1092 /* must be called with local_bh_disable()d */
1093 static int ixgbevf_busy_poll_recv(struct napi_struct *napi)
1094 {
1095         struct ixgbevf_q_vector *q_vector =
1096                         container_of(napi, struct ixgbevf_q_vector, napi);
1097         struct ixgbevf_adapter *adapter = q_vector->adapter;
1098         struct ixgbevf_ring  *ring;
1099         int found = 0;
1100
1101         if (test_bit(__IXGBEVF_DOWN, &adapter->state))
1102                 return LL_FLUSH_FAILED;
1103
1104         if (!ixgbevf_qv_lock_poll(q_vector))
1105                 return LL_FLUSH_BUSY;
1106
1107         ixgbevf_for_each_ring(ring, q_vector->rx) {
1108                 found = ixgbevf_clean_rx_irq(q_vector, ring, 4);
1109 #ifdef BP_EXTENDED_STATS
1110                 if (found)
1111                         ring->stats.cleaned += found;
1112                 else
1113                         ring->stats.misses++;
1114 #endif
1115                 if (found)
1116                         break;
1117         }
1118
1119         ixgbevf_qv_unlock_poll(q_vector);
1120
1121         return found;
1122 }
1123 #endif /* CONFIG_NET_RX_BUSY_POLL */
1124
1125 /**
1126  * ixgbevf_configure_msix - Configure MSI-X hardware
1127  * @adapter: board private structure
1128  *
1129  * ixgbevf_configure_msix sets up the hardware to properly generate MSI-X
1130  * interrupts.
1131  **/
1132 static void ixgbevf_configure_msix(struct ixgbevf_adapter *adapter)
1133 {
1134         struct ixgbevf_q_vector *q_vector;
1135         int q_vectors, v_idx;
1136
1137         q_vectors = adapter->num_msix_vectors - NON_Q_VECTORS;
1138         adapter->eims_enable_mask = 0;
1139
1140         /* Populate the IVAR table and set the ITR values to the
1141          * corresponding register.
1142          */
1143         for (v_idx = 0; v_idx < q_vectors; v_idx++) {
1144                 struct ixgbevf_ring *ring;
1145
1146                 q_vector = adapter->q_vector[v_idx];
1147
1148                 ixgbevf_for_each_ring(ring, q_vector->rx)
1149                         ixgbevf_set_ivar(adapter, 0, ring->reg_idx, v_idx);
1150
1151                 ixgbevf_for_each_ring(ring, q_vector->tx)
1152                         ixgbevf_set_ivar(adapter, 1, ring->reg_idx, v_idx);
1153
1154                 if (q_vector->tx.ring && !q_vector->rx.ring) {
1155                         /* Tx only vector */
1156                         if (adapter->tx_itr_setting == 1)
1157                                 q_vector->itr = IXGBE_12K_ITR;
1158                         else
1159                                 q_vector->itr = adapter->tx_itr_setting;
1160                 } else {
1161                         /* Rx or Rx/Tx vector */
1162                         if (adapter->rx_itr_setting == 1)
1163                                 q_vector->itr = IXGBE_20K_ITR;
1164                         else
1165                                 q_vector->itr = adapter->rx_itr_setting;
1166                 }
1167
1168                 /* add q_vector eims value to global eims_enable_mask */
1169                 adapter->eims_enable_mask |= BIT(v_idx);
1170
1171                 ixgbevf_write_eitr(q_vector);
1172         }
1173
1174         ixgbevf_set_ivar(adapter, -1, 1, v_idx);
1175         /* setup eims_other and add value to global eims_enable_mask */
1176         adapter->eims_other = BIT(v_idx);
1177         adapter->eims_enable_mask |= adapter->eims_other;
1178 }
1179
1180 enum latency_range {
1181         lowest_latency = 0,
1182         low_latency = 1,
1183         bulk_latency = 2,
1184         latency_invalid = 255
1185 };
1186
1187 /**
1188  * ixgbevf_update_itr - update the dynamic ITR value based on statistics
1189  * @q_vector: structure containing interrupt and ring information
1190  * @ring_container: structure containing ring performance data
1191  *
1192  * Stores a new ITR value based on packets and byte
1193  * counts during the last interrupt.  The advantage of per interrupt
1194  * computation is faster updates and more accurate ITR for the current
1195  * traffic pattern.  Constants in this function were computed
1196  * based on theoretical maximum wire speed and thresholds were set based
1197  * on testing data as well as attempting to minimize response time
1198  * while increasing bulk throughput.
1199  **/
1200 static void ixgbevf_update_itr(struct ixgbevf_q_vector *q_vector,
1201                                struct ixgbevf_ring_container *ring_container)
1202 {
1203         int bytes = ring_container->total_bytes;
1204         int packets = ring_container->total_packets;
1205         u32 timepassed_us;
1206         u64 bytes_perint;
1207         u8 itr_setting = ring_container->itr;
1208
1209         if (packets == 0)
1210                 return;
1211
1212         /* simple throttle rate management
1213          *    0-20MB/s lowest (100000 ints/s)
1214          *   20-100MB/s low   (20000 ints/s)
1215          *  100-1249MB/s bulk (12000 ints/s)
1216          */
1217         /* what was last interrupt timeslice? */
1218         timepassed_us = q_vector->itr >> 2;
1219         bytes_perint = bytes / timepassed_us; /* bytes/usec */
1220
1221         switch (itr_setting) {
1222         case lowest_latency:
1223                 if (bytes_perint > 10)
1224                         itr_setting = low_latency;
1225                 break;
1226         case low_latency:
1227                 if (bytes_perint > 20)
1228                         itr_setting = bulk_latency;
1229                 else if (bytes_perint <= 10)
1230                         itr_setting = lowest_latency;
1231                 break;
1232         case bulk_latency:
1233                 if (bytes_perint <= 20)
1234                         itr_setting = low_latency;
1235                 break;
1236         }
1237
1238         /* clear work counters since we have the values we need */
1239         ring_container->total_bytes = 0;
1240         ring_container->total_packets = 0;
1241
1242         /* write updated itr to ring container */
1243         ring_container->itr = itr_setting;
1244 }
1245
1246 static void ixgbevf_set_itr(struct ixgbevf_q_vector *q_vector)
1247 {
1248         u32 new_itr = q_vector->itr;
1249         u8 current_itr;
1250
1251         ixgbevf_update_itr(q_vector, &q_vector->tx);
1252         ixgbevf_update_itr(q_vector, &q_vector->rx);
1253
1254         current_itr = max(q_vector->rx.itr, q_vector->tx.itr);
1255
1256         switch (current_itr) {
1257         /* counts and packets in update_itr are dependent on these numbers */
1258         case lowest_latency:
1259                 new_itr = IXGBE_100K_ITR;
1260                 break;
1261         case low_latency:
1262                 new_itr = IXGBE_20K_ITR;
1263                 break;
1264         case bulk_latency:
1265                 new_itr = IXGBE_12K_ITR;
1266                 break;
1267         default:
1268                 break;
1269         }
1270
1271         if (new_itr != q_vector->itr) {
1272                 /* do an exponential smoothing */
1273                 new_itr = (10 * new_itr * q_vector->itr) /
1274                           ((9 * new_itr) + q_vector->itr);
1275
1276                 /* save the algorithm value here */
1277                 q_vector->itr = new_itr;
1278
1279                 ixgbevf_write_eitr(q_vector);
1280         }
1281 }
1282
1283 static irqreturn_t ixgbevf_msix_other(int irq, void *data)
1284 {
1285         struct ixgbevf_adapter *adapter = data;
1286         struct ixgbe_hw *hw = &adapter->hw;
1287
1288         hw->mac.get_link_status = 1;
1289
1290         ixgbevf_service_event_schedule(adapter);
1291
1292         IXGBE_WRITE_REG(hw, IXGBE_VTEIMS, adapter->eims_other);
1293
1294         return IRQ_HANDLED;
1295 }
1296
1297 /**
1298  * ixgbevf_msix_clean_rings - single unshared vector rx clean (all queues)
1299  * @irq: unused
1300  * @data: pointer to our q_vector struct for this interrupt vector
1301  **/
1302 static irqreturn_t ixgbevf_msix_clean_rings(int irq, void *data)
1303 {
1304         struct ixgbevf_q_vector *q_vector = data;
1305
1306         /* EIAM disabled interrupts (on this vector) for us */
1307         if (q_vector->rx.ring || q_vector->tx.ring)
1308                 napi_schedule_irqoff(&q_vector->napi);
1309
1310         return IRQ_HANDLED;
1311 }
1312
1313 static inline void map_vector_to_rxq(struct ixgbevf_adapter *a, int v_idx,
1314                                      int r_idx)
1315 {
1316         struct ixgbevf_q_vector *q_vector = a->q_vector[v_idx];
1317
1318         a->rx_ring[r_idx]->next = q_vector->rx.ring;
1319         q_vector->rx.ring = a->rx_ring[r_idx];
1320         q_vector->rx.count++;
1321 }
1322
1323 static inline void map_vector_to_txq(struct ixgbevf_adapter *a, int v_idx,
1324                                      int t_idx)
1325 {
1326         struct ixgbevf_q_vector *q_vector = a->q_vector[v_idx];
1327
1328         a->tx_ring[t_idx]->next = q_vector->tx.ring;
1329         q_vector->tx.ring = a->tx_ring[t_idx];
1330         q_vector->tx.count++;
1331 }
1332
1333 /**
1334  * ixgbevf_map_rings_to_vectors - Maps descriptor rings to vectors
1335  * @adapter: board private structure to initialize
1336  *
1337  * This function maps descriptor rings to the queue-specific vectors
1338  * we were allotted through the MSI-X enabling code.  Ideally, we'd have
1339  * one vector per ring/queue, but on a constrained vector budget, we
1340  * group the rings as "efficiently" as possible.  You would add new
1341  * mapping configurations in here.
1342  **/
1343 static int ixgbevf_map_rings_to_vectors(struct ixgbevf_adapter *adapter)
1344 {
1345         int q_vectors;
1346         int v_start = 0;
1347         int rxr_idx = 0, txr_idx = 0;
1348         int rxr_remaining = adapter->num_rx_queues;
1349         int txr_remaining = adapter->num_tx_queues;
1350         int i, j;
1351         int rqpv, tqpv;
1352
1353         q_vectors = adapter->num_msix_vectors - NON_Q_VECTORS;
1354
1355         /* The ideal configuration...
1356          * We have enough vectors to map one per queue.
1357          */
1358         if (q_vectors == adapter->num_rx_queues + adapter->num_tx_queues) {
1359                 for (; rxr_idx < rxr_remaining; v_start++, rxr_idx++)
1360                         map_vector_to_rxq(adapter, v_start, rxr_idx);
1361
1362                 for (; txr_idx < txr_remaining; v_start++, txr_idx++)
1363                         map_vector_to_txq(adapter, v_start, txr_idx);
1364                 return 0;
1365         }
1366
1367         /* If we don't have enough vectors for a 1-to-1
1368          * mapping, we'll have to group them so there are
1369          * multiple queues per vector.
1370          */
1371         /* Re-adjusting *qpv takes care of the remainder. */
1372         for (i = v_start; i < q_vectors; i++) {
1373                 rqpv = DIV_ROUND_UP(rxr_remaining, q_vectors - i);
1374                 for (j = 0; j < rqpv; j++) {
1375                         map_vector_to_rxq(adapter, i, rxr_idx);
1376                         rxr_idx++;
1377                         rxr_remaining--;
1378                 }
1379         }
1380         for (i = v_start; i < q_vectors; i++) {
1381                 tqpv = DIV_ROUND_UP(txr_remaining, q_vectors - i);
1382                 for (j = 0; j < tqpv; j++) {
1383                         map_vector_to_txq(adapter, i, txr_idx);
1384                         txr_idx++;
1385                         txr_remaining--;
1386                 }
1387         }
1388
1389         return 0;
1390 }
1391
1392 /**
1393  * ixgbevf_request_msix_irqs - Initialize MSI-X interrupts
1394  * @adapter: board private structure
1395  *
1396  * ixgbevf_request_msix_irqs allocates MSI-X vectors and requests
1397  * interrupts from the kernel.
1398  **/
1399 static int ixgbevf_request_msix_irqs(struct ixgbevf_adapter *adapter)
1400 {
1401         struct net_device *netdev = adapter->netdev;
1402         int q_vectors = adapter->num_msix_vectors - NON_Q_VECTORS;
1403         int vector, err;
1404         int ri = 0, ti = 0;
1405
1406         for (vector = 0; vector < q_vectors; vector++) {
1407                 struct ixgbevf_q_vector *q_vector = adapter->q_vector[vector];
1408                 struct msix_entry *entry = &adapter->msix_entries[vector];
1409
1410                 if (q_vector->tx.ring && q_vector->rx.ring) {
1411                         snprintf(q_vector->name, sizeof(q_vector->name) - 1,
1412                                  "%s-%s-%d", netdev->name, "TxRx", ri++);
1413                         ti++;
1414                 } else if (q_vector->rx.ring) {
1415                         snprintf(q_vector->name, sizeof(q_vector->name) - 1,
1416                                  "%s-%s-%d", netdev->name, "rx", ri++);
1417                 } else if (q_vector->tx.ring) {
1418                         snprintf(q_vector->name, sizeof(q_vector->name) - 1,
1419                                  "%s-%s-%d", netdev->name, "tx", ti++);
1420                 } else {
1421                         /* skip this unused q_vector */
1422                         continue;
1423                 }
1424                 err = request_irq(entry->vector, &ixgbevf_msix_clean_rings, 0,
1425                                   q_vector->name, q_vector);
1426                 if (err) {
1427                         hw_dbg(&adapter->hw,
1428                                "request_irq failed for MSIX interrupt Error: %d\n",
1429                                err);
1430                         goto free_queue_irqs;
1431                 }
1432         }
1433
1434         err = request_irq(adapter->msix_entries[vector].vector,
1435                           &ixgbevf_msix_other, 0, netdev->name, adapter);
1436         if (err) {
1437                 hw_dbg(&adapter->hw, "request_irq for msix_other failed: %d\n",
1438                        err);
1439                 goto free_queue_irqs;
1440         }
1441
1442         return 0;
1443
1444 free_queue_irqs:
1445         while (vector) {
1446                 vector--;
1447                 free_irq(adapter->msix_entries[vector].vector,
1448                          adapter->q_vector[vector]);
1449         }
1450         /* This failure is non-recoverable - it indicates the system is
1451          * out of MSIX vector resources and the VF driver cannot run
1452          * without them.  Set the number of msix vectors to zero
1453          * indicating that not enough can be allocated.  The error
1454          * will be returned to the user indicating device open failed.
1455          * Any further attempts to force the driver to open will also
1456          * fail.  The only way to recover is to unload the driver and
1457          * reload it again.  If the system has recovered some MSIX
1458          * vectors then it may succeed.
1459          */
1460         adapter->num_msix_vectors = 0;
1461         return err;
1462 }
1463
1464 static inline void ixgbevf_reset_q_vectors(struct ixgbevf_adapter *adapter)
1465 {
1466         int i, q_vectors = adapter->num_msix_vectors - NON_Q_VECTORS;
1467
1468         for (i = 0; i < q_vectors; i++) {
1469                 struct ixgbevf_q_vector *q_vector = adapter->q_vector[i];
1470
1471                 q_vector->rx.ring = NULL;
1472                 q_vector->tx.ring = NULL;
1473                 q_vector->rx.count = 0;
1474                 q_vector->tx.count = 0;
1475         }
1476 }
1477
1478 /**
1479  * ixgbevf_request_irq - initialize interrupts
1480  * @adapter: board private structure
1481  *
1482  * Attempts to configure interrupts using the best available
1483  * capabilities of the hardware and kernel.
1484  **/
1485 static int ixgbevf_request_irq(struct ixgbevf_adapter *adapter)
1486 {
1487         int err = ixgbevf_request_msix_irqs(adapter);
1488
1489         if (err)
1490                 hw_dbg(&adapter->hw, "request_irq failed, Error %d\n", err);
1491
1492         return err;
1493 }
1494
1495 static void ixgbevf_free_irq(struct ixgbevf_adapter *adapter)
1496 {
1497         int i, q_vectors;
1498
1499         q_vectors = adapter->num_msix_vectors;
1500         i = q_vectors - 1;
1501
1502         free_irq(adapter->msix_entries[i].vector, adapter);
1503         i--;
1504
1505         for (; i >= 0; i--) {
1506                 /* free only the irqs that were actually requested */
1507                 if (!adapter->q_vector[i]->rx.ring &&
1508                     !adapter->q_vector[i]->tx.ring)
1509                         continue;
1510
1511                 free_irq(adapter->msix_entries[i].vector,
1512                          adapter->q_vector[i]);
1513         }
1514
1515         ixgbevf_reset_q_vectors(adapter);
1516 }
1517
1518 /**
1519  * ixgbevf_irq_disable - Mask off interrupt generation on the NIC
1520  * @adapter: board private structure
1521  **/
1522 static inline void ixgbevf_irq_disable(struct ixgbevf_adapter *adapter)
1523 {
1524         struct ixgbe_hw *hw = &adapter->hw;
1525         int i;
1526
1527         IXGBE_WRITE_REG(hw, IXGBE_VTEIAM, 0);
1528         IXGBE_WRITE_REG(hw, IXGBE_VTEIMC, ~0);
1529         IXGBE_WRITE_REG(hw, IXGBE_VTEIAC, 0);
1530
1531         IXGBE_WRITE_FLUSH(hw);
1532
1533         for (i = 0; i < adapter->num_msix_vectors; i++)
1534                 synchronize_irq(adapter->msix_entries[i].vector);
1535 }
1536
1537 /**
1538  * ixgbevf_irq_enable - Enable default interrupt generation settings
1539  * @adapter: board private structure
1540  **/
1541 static inline void ixgbevf_irq_enable(struct ixgbevf_adapter *adapter)
1542 {
1543         struct ixgbe_hw *hw = &adapter->hw;
1544
1545         IXGBE_WRITE_REG(hw, IXGBE_VTEIAM, adapter->eims_enable_mask);
1546         IXGBE_WRITE_REG(hw, IXGBE_VTEIAC, adapter->eims_enable_mask);
1547         IXGBE_WRITE_REG(hw, IXGBE_VTEIMS, adapter->eims_enable_mask);
1548 }
1549
1550 /**
1551  * ixgbevf_configure_tx_ring - Configure 82599 VF Tx ring after Reset
1552  * @adapter: board private structure
1553  * @ring: structure containing ring specific data
1554  *
1555  * Configure the Tx descriptor ring after a reset.
1556  **/
1557 static void ixgbevf_configure_tx_ring(struct ixgbevf_adapter *adapter,
1558                                       struct ixgbevf_ring *ring)
1559 {
1560         struct ixgbe_hw *hw = &adapter->hw;
1561         u64 tdba = ring->dma;
1562         int wait_loop = 10;
1563         u32 txdctl = IXGBE_TXDCTL_ENABLE;
1564         u8 reg_idx = ring->reg_idx;
1565
1566         /* disable queue to avoid issues while updating state */
1567         IXGBE_WRITE_REG(hw, IXGBE_VFTXDCTL(reg_idx), IXGBE_TXDCTL_SWFLSH);
1568         IXGBE_WRITE_FLUSH(hw);
1569
1570         IXGBE_WRITE_REG(hw, IXGBE_VFTDBAL(reg_idx), tdba & DMA_BIT_MASK(32));
1571         IXGBE_WRITE_REG(hw, IXGBE_VFTDBAH(reg_idx), tdba >> 32);
1572         IXGBE_WRITE_REG(hw, IXGBE_VFTDLEN(reg_idx),
1573                         ring->count * sizeof(union ixgbe_adv_tx_desc));
1574
1575         /* disable head writeback */
1576         IXGBE_WRITE_REG(hw, IXGBE_VFTDWBAH(reg_idx), 0);
1577         IXGBE_WRITE_REG(hw, IXGBE_VFTDWBAL(reg_idx), 0);
1578
1579         /* enable relaxed ordering */
1580         IXGBE_WRITE_REG(hw, IXGBE_VFDCA_TXCTRL(reg_idx),
1581                         (IXGBE_DCA_TXCTRL_DESC_RRO_EN |
1582                          IXGBE_DCA_TXCTRL_DATA_RRO_EN));
1583
1584         /* reset head and tail pointers */
1585         IXGBE_WRITE_REG(hw, IXGBE_VFTDH(reg_idx), 0);
1586         IXGBE_WRITE_REG(hw, IXGBE_VFTDT(reg_idx), 0);
1587         ring->tail = adapter->io_addr + IXGBE_VFTDT(reg_idx);
1588
1589         /* reset ntu and ntc to place SW in sync with hardwdare */
1590         ring->next_to_clean = 0;
1591         ring->next_to_use = 0;
1592
1593         /* In order to avoid issues WTHRESH + PTHRESH should always be equal
1594          * to or less than the number of on chip descriptors, which is
1595          * currently 40.
1596          */
1597         txdctl |= (8 << 16);    /* WTHRESH = 8 */
1598
1599         /* Setting PTHRESH to 32 both improves performance */
1600         txdctl |= (1u << 8) |    /* HTHRESH = 1 */
1601                    32;           /* PTHRESH = 32 */
1602
1603         clear_bit(__IXGBEVF_HANG_CHECK_ARMED, &ring->state);
1604
1605         IXGBE_WRITE_REG(hw, IXGBE_VFTXDCTL(reg_idx), txdctl);
1606
1607         /* poll to verify queue is enabled */
1608         do {
1609                 usleep_range(1000, 2000);
1610                 txdctl = IXGBE_READ_REG(hw, IXGBE_VFTXDCTL(reg_idx));
1611         }  while (--wait_loop && !(txdctl & IXGBE_TXDCTL_ENABLE));
1612         if (!wait_loop)
1613                 pr_err("Could not enable Tx Queue %d\n", reg_idx);
1614 }
1615
1616 /**
1617  * ixgbevf_configure_tx - Configure 82599 VF Transmit Unit after Reset
1618  * @adapter: board private structure
1619  *
1620  * Configure the Tx unit of the MAC after a reset.
1621  **/
1622 static void ixgbevf_configure_tx(struct ixgbevf_adapter *adapter)
1623 {
1624         u32 i;
1625
1626         /* Setup the HW Tx Head and Tail descriptor pointers */
1627         for (i = 0; i < adapter->num_tx_queues; i++)
1628                 ixgbevf_configure_tx_ring(adapter, adapter->tx_ring[i]);
1629 }
1630
1631 #define IXGBE_SRRCTL_BSIZEHDRSIZE_SHIFT 2
1632
1633 static void ixgbevf_configure_srrctl(struct ixgbevf_adapter *adapter, int index)
1634 {
1635         struct ixgbe_hw *hw = &adapter->hw;
1636         u32 srrctl;
1637
1638         srrctl = IXGBE_SRRCTL_DROP_EN;
1639
1640         srrctl |= IXGBEVF_RX_HDR_SIZE << IXGBE_SRRCTL_BSIZEHDRSIZE_SHIFT;
1641         srrctl |= IXGBEVF_RX_BUFSZ >> IXGBE_SRRCTL_BSIZEPKT_SHIFT;
1642         srrctl |= IXGBE_SRRCTL_DESCTYPE_ADV_ONEBUF;
1643
1644         IXGBE_WRITE_REG(hw, IXGBE_VFSRRCTL(index), srrctl);
1645 }
1646
1647 static void ixgbevf_setup_psrtype(struct ixgbevf_adapter *adapter)
1648 {
1649         struct ixgbe_hw *hw = &adapter->hw;
1650
1651         /* PSRTYPE must be initialized in 82599 */
1652         u32 psrtype = IXGBE_PSRTYPE_TCPHDR | IXGBE_PSRTYPE_UDPHDR |
1653                       IXGBE_PSRTYPE_IPV4HDR | IXGBE_PSRTYPE_IPV6HDR |
1654                       IXGBE_PSRTYPE_L2HDR;
1655
1656         if (adapter->num_rx_queues > 1)
1657                 psrtype |= BIT(29);
1658
1659         IXGBE_WRITE_REG(hw, IXGBE_VFPSRTYPE, psrtype);
1660 }
1661
1662 #define IXGBEVF_MAX_RX_DESC_POLL 10
1663 static void ixgbevf_disable_rx_queue(struct ixgbevf_adapter *adapter,
1664                                      struct ixgbevf_ring *ring)
1665 {
1666         struct ixgbe_hw *hw = &adapter->hw;
1667         int wait_loop = IXGBEVF_MAX_RX_DESC_POLL;
1668         u32 rxdctl;
1669         u8 reg_idx = ring->reg_idx;
1670
1671         if (IXGBE_REMOVED(hw->hw_addr))
1672                 return;
1673         rxdctl = IXGBE_READ_REG(hw, IXGBE_VFRXDCTL(reg_idx));
1674         rxdctl &= ~IXGBE_RXDCTL_ENABLE;
1675
1676         /* write value back with RXDCTL.ENABLE bit cleared */
1677         IXGBE_WRITE_REG(hw, IXGBE_VFRXDCTL(reg_idx), rxdctl);
1678
1679         /* the hardware may take up to 100us to really disable the Rx queue */
1680         do {
1681                 udelay(10);
1682                 rxdctl = IXGBE_READ_REG(hw, IXGBE_VFRXDCTL(reg_idx));
1683         } while (--wait_loop && (rxdctl & IXGBE_RXDCTL_ENABLE));
1684
1685         if (!wait_loop)
1686                 pr_err("RXDCTL.ENABLE queue %d not cleared while polling\n",
1687                        reg_idx);
1688 }
1689
1690 static void ixgbevf_rx_desc_queue_enable(struct ixgbevf_adapter *adapter,
1691                                          struct ixgbevf_ring *ring)
1692 {
1693         struct ixgbe_hw *hw = &adapter->hw;
1694         int wait_loop = IXGBEVF_MAX_RX_DESC_POLL;
1695         u32 rxdctl;
1696         u8 reg_idx = ring->reg_idx;
1697
1698         if (IXGBE_REMOVED(hw->hw_addr))
1699                 return;
1700         do {
1701                 usleep_range(1000, 2000);
1702                 rxdctl = IXGBE_READ_REG(hw, IXGBE_VFRXDCTL(reg_idx));
1703         } while (--wait_loop && !(rxdctl & IXGBE_RXDCTL_ENABLE));
1704
1705         if (!wait_loop)
1706                 pr_err("RXDCTL.ENABLE queue %d not set while polling\n",
1707                        reg_idx);
1708 }
1709
1710 static void ixgbevf_setup_vfmrqc(struct ixgbevf_adapter *adapter)
1711 {
1712         struct ixgbe_hw *hw = &adapter->hw;
1713         u32 vfmrqc = 0, vfreta = 0;
1714         u16 rss_i = adapter->num_rx_queues;
1715         u8 i, j;
1716
1717         /* Fill out hash function seeds */
1718         netdev_rss_key_fill(adapter->rss_key, sizeof(adapter->rss_key));
1719         for (i = 0; i < IXGBEVF_VFRSSRK_REGS; i++)
1720                 IXGBE_WRITE_REG(hw, IXGBE_VFRSSRK(i), adapter->rss_key[i]);
1721
1722         for (i = 0, j = 0; i < IXGBEVF_X550_VFRETA_SIZE; i++, j++) {
1723                 if (j == rss_i)
1724                         j = 0;
1725
1726                 adapter->rss_indir_tbl[i] = j;
1727
1728                 vfreta |= j << (i & 0x3) * 8;
1729                 if ((i & 3) == 3) {
1730                         IXGBE_WRITE_REG(hw, IXGBE_VFRETA(i >> 2), vfreta);
1731                         vfreta = 0;
1732                 }
1733         }
1734
1735         /* Perform hash on these packet types */
1736         vfmrqc |= IXGBE_VFMRQC_RSS_FIELD_IPV4 |
1737                 IXGBE_VFMRQC_RSS_FIELD_IPV4_TCP |
1738                 IXGBE_VFMRQC_RSS_FIELD_IPV6 |
1739                 IXGBE_VFMRQC_RSS_FIELD_IPV6_TCP;
1740
1741         vfmrqc |= IXGBE_VFMRQC_RSSEN;
1742
1743         IXGBE_WRITE_REG(hw, IXGBE_VFMRQC, vfmrqc);
1744 }
1745
1746 static void ixgbevf_configure_rx_ring(struct ixgbevf_adapter *adapter,
1747                                       struct ixgbevf_ring *ring)
1748 {
1749         struct ixgbe_hw *hw = &adapter->hw;
1750         u64 rdba = ring->dma;
1751         u32 rxdctl;
1752         u8 reg_idx = ring->reg_idx;
1753
1754         /* disable queue to avoid issues while updating state */
1755         rxdctl = IXGBE_READ_REG(hw, IXGBE_VFRXDCTL(reg_idx));
1756         ixgbevf_disable_rx_queue(adapter, ring);
1757
1758         IXGBE_WRITE_REG(hw, IXGBE_VFRDBAL(reg_idx), rdba & DMA_BIT_MASK(32));
1759         IXGBE_WRITE_REG(hw, IXGBE_VFRDBAH(reg_idx), rdba >> 32);
1760         IXGBE_WRITE_REG(hw, IXGBE_VFRDLEN(reg_idx),
1761                         ring->count * sizeof(union ixgbe_adv_rx_desc));
1762
1763 #ifndef CONFIG_SPARC
1764         /* enable relaxed ordering */
1765         IXGBE_WRITE_REG(hw, IXGBE_VFDCA_RXCTRL(reg_idx),
1766                         IXGBE_DCA_RXCTRL_DESC_RRO_EN);
1767 #else
1768         IXGBE_WRITE_REG(hw, IXGBE_VFDCA_RXCTRL(reg_idx),
1769                         IXGBE_DCA_RXCTRL_DESC_RRO_EN |
1770                         IXGBE_DCA_RXCTRL_DATA_WRO_EN);
1771 #endif
1772
1773         /* reset head and tail pointers */
1774         IXGBE_WRITE_REG(hw, IXGBE_VFRDH(reg_idx), 0);
1775         IXGBE_WRITE_REG(hw, IXGBE_VFRDT(reg_idx), 0);
1776         ring->tail = adapter->io_addr + IXGBE_VFRDT(reg_idx);
1777
1778         /* reset ntu and ntc to place SW in sync with hardwdare */
1779         ring->next_to_clean = 0;
1780         ring->next_to_use = 0;
1781         ring->next_to_alloc = 0;
1782
1783         ixgbevf_configure_srrctl(adapter, reg_idx);
1784
1785         /* allow any size packet since we can handle overflow */
1786         rxdctl &= ~IXGBE_RXDCTL_RLPML_EN;
1787
1788         rxdctl |= IXGBE_RXDCTL_ENABLE | IXGBE_RXDCTL_VME;
1789         IXGBE_WRITE_REG(hw, IXGBE_VFRXDCTL(reg_idx), rxdctl);
1790
1791         ixgbevf_rx_desc_queue_enable(adapter, ring);
1792         ixgbevf_alloc_rx_buffers(ring, ixgbevf_desc_unused(ring));
1793 }
1794
1795 /**
1796  * ixgbevf_configure_rx - Configure 82599 VF Receive Unit after Reset
1797  * @adapter: board private structure
1798  *
1799  * Configure the Rx unit of the MAC after a reset.
1800  **/
1801 static void ixgbevf_configure_rx(struct ixgbevf_adapter *adapter)
1802 {
1803         int i;
1804         struct ixgbe_hw *hw = &adapter->hw;
1805         struct net_device *netdev = adapter->netdev;
1806
1807         ixgbevf_setup_psrtype(adapter);
1808         if (hw->mac.type >= ixgbe_mac_X550_vf)
1809                 ixgbevf_setup_vfmrqc(adapter);
1810
1811         /* notify the PF of our intent to use this size of frame */
1812         hw->mac.ops.set_rlpml(hw, netdev->mtu + ETH_HLEN + ETH_FCS_LEN);
1813
1814         /* Setup the HW Rx Head and Tail Descriptor Pointers and
1815          * the Base and Length of the Rx Descriptor Ring
1816          */
1817         for (i = 0; i < adapter->num_rx_queues; i++)
1818                 ixgbevf_configure_rx_ring(adapter, adapter->rx_ring[i]);
1819 }
1820
1821 static int ixgbevf_vlan_rx_add_vid(struct net_device *netdev,
1822                                    __be16 proto, u16 vid)
1823 {
1824         struct ixgbevf_adapter *adapter = netdev_priv(netdev);
1825         struct ixgbe_hw *hw = &adapter->hw;
1826         int err;
1827
1828         spin_lock_bh(&adapter->mbx_lock);
1829
1830         /* add VID to filter table */
1831         err = hw->mac.ops.set_vfta(hw, vid, 0, true);
1832
1833         spin_unlock_bh(&adapter->mbx_lock);
1834
1835         /* translate error return types so error makes sense */
1836         if (err == IXGBE_ERR_MBX)
1837                 return -EIO;
1838
1839         if (err == IXGBE_ERR_INVALID_ARGUMENT)
1840                 return -EACCES;
1841
1842         set_bit(vid, adapter->active_vlans);
1843
1844         return err;
1845 }
1846
1847 static int ixgbevf_vlan_rx_kill_vid(struct net_device *netdev,
1848                                     __be16 proto, u16 vid)
1849 {
1850         struct ixgbevf_adapter *adapter = netdev_priv(netdev);
1851         struct ixgbe_hw *hw = &adapter->hw;
1852         int err;
1853
1854         spin_lock_bh(&adapter->mbx_lock);
1855
1856         /* remove VID from filter table */
1857         err = hw->mac.ops.set_vfta(hw, vid, 0, false);
1858
1859         spin_unlock_bh(&adapter->mbx_lock);
1860
1861         clear_bit(vid, adapter->active_vlans);
1862
1863         return err;
1864 }
1865
1866 static void ixgbevf_restore_vlan(struct ixgbevf_adapter *adapter)
1867 {
1868         u16 vid;
1869
1870         for_each_set_bit(vid, adapter->active_vlans, VLAN_N_VID)
1871                 ixgbevf_vlan_rx_add_vid(adapter->netdev,
1872                                         htons(ETH_P_8021Q), vid);
1873 }
1874
1875 static int ixgbevf_write_uc_addr_list(struct net_device *netdev)
1876 {
1877         struct ixgbevf_adapter *adapter = netdev_priv(netdev);
1878         struct ixgbe_hw *hw = &adapter->hw;
1879         int count = 0;
1880
1881         if ((netdev_uc_count(netdev)) > 10) {
1882                 pr_err("Too many unicast filters - No Space\n");
1883                 return -ENOSPC;
1884         }
1885
1886         if (!netdev_uc_empty(netdev)) {
1887                 struct netdev_hw_addr *ha;
1888
1889                 netdev_for_each_uc_addr(ha, netdev) {
1890                         hw->mac.ops.set_uc_addr(hw, ++count, ha->addr);
1891                         udelay(200);
1892                 }
1893         } else {
1894                 /* If the list is empty then send message to PF driver to
1895                  * clear all MAC VLANs on this VF.
1896                  */
1897                 hw->mac.ops.set_uc_addr(hw, 0, NULL);
1898         }
1899
1900         return count;
1901 }
1902
1903 /**
1904  * ixgbevf_set_rx_mode - Multicast and unicast set
1905  * @netdev: network interface device structure
1906  *
1907  * The set_rx_method entry point is called whenever the multicast address
1908  * list, unicast address list or the network interface flags are updated.
1909  * This routine is responsible for configuring the hardware for proper
1910  * multicast mode and configuring requested unicast filters.
1911  **/
1912 static void ixgbevf_set_rx_mode(struct net_device *netdev)
1913 {
1914         struct ixgbevf_adapter *adapter = netdev_priv(netdev);
1915         struct ixgbe_hw *hw = &adapter->hw;
1916         unsigned int flags = netdev->flags;
1917         int xcast_mode;
1918
1919         xcast_mode = (flags & IFF_ALLMULTI) ? IXGBEVF_XCAST_MODE_ALLMULTI :
1920                      (flags & (IFF_BROADCAST | IFF_MULTICAST)) ?
1921                      IXGBEVF_XCAST_MODE_MULTI : IXGBEVF_XCAST_MODE_NONE;
1922
1923         spin_lock_bh(&adapter->mbx_lock);
1924
1925         hw->mac.ops.update_xcast_mode(hw, xcast_mode);
1926
1927         /* reprogram multicast list */
1928         hw->mac.ops.update_mc_addr_list(hw, netdev);
1929
1930         ixgbevf_write_uc_addr_list(netdev);
1931
1932         spin_unlock_bh(&adapter->mbx_lock);
1933 }
1934
1935 static void ixgbevf_napi_enable_all(struct ixgbevf_adapter *adapter)
1936 {
1937         int q_idx;
1938         struct ixgbevf_q_vector *q_vector;
1939         int q_vectors = adapter->num_msix_vectors - NON_Q_VECTORS;
1940
1941         for (q_idx = 0; q_idx < q_vectors; q_idx++) {
1942                 q_vector = adapter->q_vector[q_idx];
1943 #ifdef CONFIG_NET_RX_BUSY_POLL
1944                 ixgbevf_qv_init_lock(adapter->q_vector[q_idx]);
1945 #endif
1946                 napi_enable(&q_vector->napi);
1947         }
1948 }
1949
1950 static void ixgbevf_napi_disable_all(struct ixgbevf_adapter *adapter)
1951 {
1952         int q_idx;
1953         struct ixgbevf_q_vector *q_vector;
1954         int q_vectors = adapter->num_msix_vectors - NON_Q_VECTORS;
1955
1956         for (q_idx = 0; q_idx < q_vectors; q_idx++) {
1957                 q_vector = adapter->q_vector[q_idx];
1958                 napi_disable(&q_vector->napi);
1959 #ifdef CONFIG_NET_RX_BUSY_POLL
1960                 while (!ixgbevf_qv_disable(adapter->q_vector[q_idx])) {
1961                         pr_info("QV %d locked\n", q_idx);
1962                         usleep_range(1000, 20000);
1963                 }
1964 #endif /* CONFIG_NET_RX_BUSY_POLL */
1965         }
1966 }
1967
1968 static int ixgbevf_configure_dcb(struct ixgbevf_adapter *adapter)
1969 {
1970         struct ixgbe_hw *hw = &adapter->hw;
1971         unsigned int def_q = 0;
1972         unsigned int num_tcs = 0;
1973         unsigned int num_rx_queues = adapter->num_rx_queues;
1974         unsigned int num_tx_queues = adapter->num_tx_queues;
1975         int err;
1976
1977         spin_lock_bh(&adapter->mbx_lock);
1978
1979         /* fetch queue configuration from the PF */
1980         err = ixgbevf_get_queues(hw, &num_tcs, &def_q);
1981
1982         spin_unlock_bh(&adapter->mbx_lock);
1983
1984         if (err)
1985                 return err;
1986
1987         if (num_tcs > 1) {
1988                 /* we need only one Tx queue */
1989                 num_tx_queues = 1;
1990
1991                 /* update default Tx ring register index */
1992                 adapter->tx_ring[0]->reg_idx = def_q;
1993
1994                 /* we need as many queues as traffic classes */
1995                 num_rx_queues = num_tcs;
1996         }
1997
1998         /* if we have a bad config abort request queue reset */
1999         if ((adapter->num_rx_queues != num_rx_queues) ||
2000             (adapter->num_tx_queues != num_tx_queues)) {
2001                 /* force mailbox timeout to prevent further messages */
2002                 hw->mbx.timeout = 0;
2003
2004                 /* wait for watchdog to come around and bail us out */
2005                 set_bit(__IXGBEVF_QUEUE_RESET_REQUESTED, &adapter->state);
2006         }
2007
2008         return 0;
2009 }
2010
2011 static void ixgbevf_configure(struct ixgbevf_adapter *adapter)
2012 {
2013         ixgbevf_configure_dcb(adapter);
2014
2015         ixgbevf_set_rx_mode(adapter->netdev);
2016
2017         ixgbevf_restore_vlan(adapter);
2018
2019         ixgbevf_configure_tx(adapter);
2020         ixgbevf_configure_rx(adapter);
2021 }
2022
2023 static void ixgbevf_save_reset_stats(struct ixgbevf_adapter *adapter)
2024 {
2025         /* Only save pre-reset stats if there are some */
2026         if (adapter->stats.vfgprc || adapter->stats.vfgptc) {
2027                 adapter->stats.saved_reset_vfgprc += adapter->stats.vfgprc -
2028                         adapter->stats.base_vfgprc;
2029                 adapter->stats.saved_reset_vfgptc += adapter->stats.vfgptc -
2030                         adapter->stats.base_vfgptc;
2031                 adapter->stats.saved_reset_vfgorc += adapter->stats.vfgorc -
2032                         adapter->stats.base_vfgorc;
2033                 adapter->stats.saved_reset_vfgotc += adapter->stats.vfgotc -
2034                         adapter->stats.base_vfgotc;
2035                 adapter->stats.saved_reset_vfmprc += adapter->stats.vfmprc -
2036                         adapter->stats.base_vfmprc;
2037         }
2038 }
2039
2040 static void ixgbevf_init_last_counter_stats(struct ixgbevf_adapter *adapter)
2041 {
2042         struct ixgbe_hw *hw = &adapter->hw;
2043
2044         adapter->stats.last_vfgprc = IXGBE_READ_REG(hw, IXGBE_VFGPRC);
2045         adapter->stats.last_vfgorc = IXGBE_READ_REG(hw, IXGBE_VFGORC_LSB);
2046         adapter->stats.last_vfgorc |=
2047                 (((u64)(IXGBE_READ_REG(hw, IXGBE_VFGORC_MSB))) << 32);
2048         adapter->stats.last_vfgptc = IXGBE_READ_REG(hw, IXGBE_VFGPTC);
2049         adapter->stats.last_vfgotc = IXGBE_READ_REG(hw, IXGBE_VFGOTC_LSB);
2050         adapter->stats.last_vfgotc |=
2051                 (((u64)(IXGBE_READ_REG(hw, IXGBE_VFGOTC_MSB))) << 32);
2052         adapter->stats.last_vfmprc = IXGBE_READ_REG(hw, IXGBE_VFMPRC);
2053
2054         adapter->stats.base_vfgprc = adapter->stats.last_vfgprc;
2055         adapter->stats.base_vfgorc = adapter->stats.last_vfgorc;
2056         adapter->stats.base_vfgptc = adapter->stats.last_vfgptc;
2057         adapter->stats.base_vfgotc = adapter->stats.last_vfgotc;
2058         adapter->stats.base_vfmprc = adapter->stats.last_vfmprc;
2059 }
2060
2061 static void ixgbevf_negotiate_api(struct ixgbevf_adapter *adapter)
2062 {
2063         struct ixgbe_hw *hw = &adapter->hw;
2064         int api[] = { ixgbe_mbox_api_12,
2065                       ixgbe_mbox_api_11,
2066                       ixgbe_mbox_api_10,
2067                       ixgbe_mbox_api_unknown };
2068         int err, idx = 0;
2069
2070         spin_lock_bh(&adapter->mbx_lock);
2071
2072         while (api[idx] != ixgbe_mbox_api_unknown) {
2073                 err = hw->mac.ops.negotiate_api_version(hw, api[idx]);
2074                 if (!err)
2075                         break;
2076                 idx++;
2077         }
2078
2079         spin_unlock_bh(&adapter->mbx_lock);
2080 }
2081
2082 static void ixgbevf_up_complete(struct ixgbevf_adapter *adapter)
2083 {
2084         struct net_device *netdev = adapter->netdev;
2085         struct ixgbe_hw *hw = &adapter->hw;
2086
2087         ixgbevf_configure_msix(adapter);
2088
2089         spin_lock_bh(&adapter->mbx_lock);
2090
2091         if (is_valid_ether_addr(hw->mac.addr))
2092                 hw->mac.ops.set_rar(hw, 0, hw->mac.addr, 0);
2093         else
2094                 hw->mac.ops.set_rar(hw, 0, hw->mac.perm_addr, 0);
2095
2096         spin_unlock_bh(&adapter->mbx_lock);
2097
2098         smp_mb__before_atomic();
2099         clear_bit(__IXGBEVF_DOWN, &adapter->state);
2100         ixgbevf_napi_enable_all(adapter);
2101
2102         /* clear any pending interrupts, may auto mask */
2103         IXGBE_READ_REG(hw, IXGBE_VTEICR);
2104         ixgbevf_irq_enable(adapter);
2105
2106         /* enable transmits */
2107         netif_tx_start_all_queues(netdev);
2108
2109         ixgbevf_save_reset_stats(adapter);
2110         ixgbevf_init_last_counter_stats(adapter);
2111
2112         hw->mac.get_link_status = 1;
2113         mod_timer(&adapter->service_timer, jiffies);
2114 }
2115
2116 void ixgbevf_up(struct ixgbevf_adapter *adapter)
2117 {
2118         ixgbevf_configure(adapter);
2119
2120         ixgbevf_up_complete(adapter);
2121 }
2122
2123 /**
2124  * ixgbevf_clean_rx_ring - Free Rx Buffers per Queue
2125  * @rx_ring: ring to free buffers from
2126  **/
2127 static void ixgbevf_clean_rx_ring(struct ixgbevf_ring *rx_ring)
2128 {
2129         struct device *dev = rx_ring->dev;
2130         unsigned long size;
2131         unsigned int i;
2132
2133         /* Free Rx ring sk_buff */
2134         if (rx_ring->skb) {
2135                 dev_kfree_skb(rx_ring->skb);
2136                 rx_ring->skb = NULL;
2137         }
2138
2139         /* ring already cleared, nothing to do */
2140         if (!rx_ring->rx_buffer_info)
2141                 return;
2142
2143         /* Free all the Rx ring pages */
2144         for (i = 0; i < rx_ring->count; i++) {
2145                 struct ixgbevf_rx_buffer *rx_buffer;
2146
2147                 rx_buffer = &rx_ring->rx_buffer_info[i];
2148                 if (rx_buffer->dma)
2149                         dma_unmap_page(dev, rx_buffer->dma,
2150                                        PAGE_SIZE, DMA_FROM_DEVICE);
2151                 rx_buffer->dma = 0;
2152                 if (rx_buffer->page)
2153                         __free_page(rx_buffer->page);
2154                 rx_buffer->page = NULL;
2155         }
2156
2157         size = sizeof(struct ixgbevf_rx_buffer) * rx_ring->count;
2158         memset(rx_ring->rx_buffer_info, 0, size);
2159
2160         /* Zero out the descriptor ring */
2161         memset(rx_ring->desc, 0, rx_ring->size);
2162 }
2163
2164 /**
2165  * ixgbevf_clean_tx_ring - Free Tx Buffers
2166  * @tx_ring: ring to be cleaned
2167  **/
2168 static void ixgbevf_clean_tx_ring(struct ixgbevf_ring *tx_ring)
2169 {
2170         struct ixgbevf_tx_buffer *tx_buffer_info;
2171         unsigned long size;
2172         unsigned int i;
2173
2174         if (!tx_ring->tx_buffer_info)
2175                 return;
2176
2177         /* Free all the Tx ring sk_buffs */
2178         for (i = 0; i < tx_ring->count; i++) {
2179                 tx_buffer_info = &tx_ring->tx_buffer_info[i];
2180                 ixgbevf_unmap_and_free_tx_resource(tx_ring, tx_buffer_info);
2181         }
2182
2183         size = sizeof(struct ixgbevf_tx_buffer) * tx_ring->count;
2184         memset(tx_ring->tx_buffer_info, 0, size);
2185
2186         memset(tx_ring->desc, 0, tx_ring->size);
2187 }
2188
2189 /**
2190  * ixgbevf_clean_all_rx_rings - Free Rx Buffers for all queues
2191  * @adapter: board private structure
2192  **/
2193 static void ixgbevf_clean_all_rx_rings(struct ixgbevf_adapter *adapter)
2194 {
2195         int i;
2196
2197         for (i = 0; i < adapter->num_rx_queues; i++)
2198                 ixgbevf_clean_rx_ring(adapter->rx_ring[i]);
2199 }
2200
2201 /**
2202  * ixgbevf_clean_all_tx_rings - Free Tx Buffers for all queues
2203  * @adapter: board private structure
2204  **/
2205 static void ixgbevf_clean_all_tx_rings(struct ixgbevf_adapter *adapter)
2206 {
2207         int i;
2208
2209         for (i = 0; i < adapter->num_tx_queues; i++)
2210                 ixgbevf_clean_tx_ring(adapter->tx_ring[i]);
2211 }
2212
2213 void ixgbevf_down(struct ixgbevf_adapter *adapter)
2214 {
2215         struct net_device *netdev = adapter->netdev;
2216         struct ixgbe_hw *hw = &adapter->hw;
2217         int i;
2218
2219         /* signal that we are down to the interrupt handler */
2220         if (test_and_set_bit(__IXGBEVF_DOWN, &adapter->state))
2221                 return; /* do nothing if already down */
2222
2223         /* disable all enabled Rx queues */
2224         for (i = 0; i < adapter->num_rx_queues; i++)
2225                 ixgbevf_disable_rx_queue(adapter, adapter->rx_ring[i]);
2226
2227         usleep_range(10000, 20000);
2228
2229         netif_tx_stop_all_queues(netdev);
2230
2231         /* call carrier off first to avoid false dev_watchdog timeouts */
2232         netif_carrier_off(netdev);
2233         netif_tx_disable(netdev);
2234
2235         ixgbevf_irq_disable(adapter);
2236
2237         ixgbevf_napi_disable_all(adapter);
2238
2239         del_timer_sync(&adapter->service_timer);
2240
2241         /* disable transmits in the hardware now that interrupts are off */
2242         for (i = 0; i < adapter->num_tx_queues; i++) {
2243                 u8 reg_idx = adapter->tx_ring[i]->reg_idx;
2244
2245                 IXGBE_WRITE_REG(hw, IXGBE_VFTXDCTL(reg_idx),
2246                                 IXGBE_TXDCTL_SWFLSH);
2247         }
2248
2249         if (!pci_channel_offline(adapter->pdev))
2250                 ixgbevf_reset(adapter);
2251
2252         ixgbevf_clean_all_tx_rings(adapter);
2253         ixgbevf_clean_all_rx_rings(adapter);
2254 }
2255
2256 void ixgbevf_reinit_locked(struct ixgbevf_adapter *adapter)
2257 {
2258         WARN_ON(in_interrupt());
2259
2260         while (test_and_set_bit(__IXGBEVF_RESETTING, &adapter->state))
2261                 msleep(1);
2262
2263         ixgbevf_down(adapter);
2264         ixgbevf_up(adapter);
2265
2266         clear_bit(__IXGBEVF_RESETTING, &adapter->state);
2267 }
2268
2269 void ixgbevf_reset(struct ixgbevf_adapter *adapter)
2270 {
2271         struct ixgbe_hw *hw = &adapter->hw;
2272         struct net_device *netdev = adapter->netdev;
2273
2274         if (hw->mac.ops.reset_hw(hw)) {
2275                 hw_dbg(hw, "PF still resetting\n");
2276         } else {
2277                 hw->mac.ops.init_hw(hw);
2278                 ixgbevf_negotiate_api(adapter);
2279         }
2280
2281         if (is_valid_ether_addr(adapter->hw.mac.addr)) {
2282                 ether_addr_copy(netdev->dev_addr, adapter->hw.mac.addr);
2283                 ether_addr_copy(netdev->perm_addr, adapter->hw.mac.addr);
2284         }
2285
2286         adapter->last_reset = jiffies;
2287 }
2288
2289 static int ixgbevf_acquire_msix_vectors(struct ixgbevf_adapter *adapter,
2290                                         int vectors)
2291 {
2292         int vector_threshold;
2293
2294         /* We'll want at least 2 (vector_threshold):
2295          * 1) TxQ[0] + RxQ[0] handler
2296          * 2) Other (Link Status Change, etc.)
2297          */
2298         vector_threshold = MIN_MSIX_COUNT;
2299
2300         /* The more we get, the more we will assign to Tx/Rx Cleanup
2301          * for the separate queues...where Rx Cleanup >= Tx Cleanup.
2302          * Right now, we simply care about how many we'll get; we'll
2303          * set them up later while requesting irq's.
2304          */
2305         vectors = pci_enable_msix_range(adapter->pdev, adapter->msix_entries,
2306                                         vector_threshold, vectors);
2307
2308         if (vectors < 0) {
2309                 dev_err(&adapter->pdev->dev,
2310                         "Unable to allocate MSI-X interrupts\n");
2311                 kfree(adapter->msix_entries);
2312                 adapter->msix_entries = NULL;
2313                 return vectors;
2314         }
2315
2316         /* Adjust for only the vectors we'll use, which is minimum
2317          * of max_msix_q_vectors + NON_Q_VECTORS, or the number of
2318          * vectors we were allocated.
2319          */
2320         adapter->num_msix_vectors = vectors;
2321
2322         return 0;
2323 }
2324
2325 /**
2326  * ixgbevf_set_num_queues - Allocate queues for device, feature dependent
2327  * @adapter: board private structure to initialize
2328  *
2329  * This is the top level queue allocation routine.  The order here is very
2330  * important, starting with the "most" number of features turned on at once,
2331  * and ending with the smallest set of features.  This way large combinations
2332  * can be allocated if they're turned on, and smaller combinations are the
2333  * fallthrough conditions.
2334  *
2335  **/
2336 static void ixgbevf_set_num_queues(struct ixgbevf_adapter *adapter)
2337 {
2338         struct ixgbe_hw *hw = &adapter->hw;
2339         unsigned int def_q = 0;
2340         unsigned int num_tcs = 0;
2341         int err;
2342
2343         /* Start with base case */
2344         adapter->num_rx_queues = 1;
2345         adapter->num_tx_queues = 1;
2346
2347         spin_lock_bh(&adapter->mbx_lock);
2348
2349         /* fetch queue configuration from the PF */
2350         err = ixgbevf_get_queues(hw, &num_tcs, &def_q);
2351
2352         spin_unlock_bh(&adapter->mbx_lock);
2353
2354         if (err)
2355                 return;
2356
2357         /* we need as many queues as traffic classes */
2358         if (num_tcs > 1) {
2359                 adapter->num_rx_queues = num_tcs;
2360         } else {
2361                 u16 rss = min_t(u16, num_online_cpus(), IXGBEVF_MAX_RSS_QUEUES);
2362
2363                 switch (hw->api_version) {
2364                 case ixgbe_mbox_api_11:
2365                 case ixgbe_mbox_api_12:
2366                         adapter->num_rx_queues = rss;
2367                         adapter->num_tx_queues = rss;
2368                 default:
2369                         break;
2370                 }
2371         }
2372 }
2373
2374 /**
2375  * ixgbevf_alloc_queues - Allocate memory for all rings
2376  * @adapter: board private structure to initialize
2377  *
2378  * We allocate one ring per queue at run-time since we don't know the
2379  * number of queues at compile-time.  The polling_netdev array is
2380  * intended for Multiqueue, but should work fine with a single queue.
2381  **/
2382 static int ixgbevf_alloc_queues(struct ixgbevf_adapter *adapter)
2383 {
2384         struct ixgbevf_ring *ring;
2385         int rx = 0, tx = 0;
2386
2387         for (; tx < adapter->num_tx_queues; tx++) {
2388                 ring = kzalloc(sizeof(*ring), GFP_KERNEL);
2389                 if (!ring)
2390                         goto err_allocation;
2391
2392                 ring->dev = &adapter->pdev->dev;
2393                 ring->netdev = adapter->netdev;
2394                 ring->count = adapter->tx_ring_count;
2395                 ring->queue_index = tx;
2396                 ring->reg_idx = tx;
2397
2398                 adapter->tx_ring[tx] = ring;
2399         }
2400
2401         for (; rx < adapter->num_rx_queues; rx++) {
2402                 ring = kzalloc(sizeof(*ring), GFP_KERNEL);
2403                 if (!ring)
2404                         goto err_allocation;
2405
2406                 ring->dev = &adapter->pdev->dev;
2407                 ring->netdev = adapter->netdev;
2408
2409                 ring->count = adapter->rx_ring_count;
2410                 ring->queue_index = rx;
2411                 ring->reg_idx = rx;
2412
2413                 adapter->rx_ring[rx] = ring;
2414         }
2415
2416         return 0;
2417
2418 err_allocation:
2419         while (tx) {
2420                 kfree(adapter->tx_ring[--tx]);
2421                 adapter->tx_ring[tx] = NULL;
2422         }
2423
2424         while (rx) {
2425                 kfree(adapter->rx_ring[--rx]);
2426                 adapter->rx_ring[rx] = NULL;
2427         }
2428         return -ENOMEM;
2429 }
2430
2431 /**
2432  * ixgbevf_set_interrupt_capability - set MSI-X or FAIL if not supported
2433  * @adapter: board private structure to initialize
2434  *
2435  * Attempt to configure the interrupts using the best available
2436  * capabilities of the hardware and the kernel.
2437  **/
2438 static int ixgbevf_set_interrupt_capability(struct ixgbevf_adapter *adapter)
2439 {
2440         struct net_device *netdev = adapter->netdev;
2441         int err;
2442         int vector, v_budget;
2443
2444         /* It's easy to be greedy for MSI-X vectors, but it really
2445          * doesn't do us much good if we have a lot more vectors
2446          * than CPU's.  So let's be conservative and only ask for
2447          * (roughly) the same number of vectors as there are CPU's.
2448          * The default is to use pairs of vectors.
2449          */
2450         v_budget = max(adapter->num_rx_queues, adapter->num_tx_queues);
2451         v_budget = min_t(int, v_budget, num_online_cpus());
2452         v_budget += NON_Q_VECTORS;
2453
2454         /* A failure in MSI-X entry allocation isn't fatal, but it does
2455          * mean we disable MSI-X capabilities of the adapter.
2456          */
2457         adapter->msix_entries = kcalloc(v_budget,
2458                                         sizeof(struct msix_entry), GFP_KERNEL);
2459         if (!adapter->msix_entries)
2460                 return -ENOMEM;
2461
2462         for (vector = 0; vector < v_budget; vector++)
2463                 adapter->msix_entries[vector].entry = vector;
2464
2465         err = ixgbevf_acquire_msix_vectors(adapter, v_budget);
2466         if (err)
2467                 return err;
2468
2469         err = netif_set_real_num_tx_queues(netdev, adapter->num_tx_queues);
2470         if (err)
2471                 return err;
2472
2473         return netif_set_real_num_rx_queues(netdev, adapter->num_rx_queues);
2474 }
2475
2476 /**
2477  * ixgbevf_alloc_q_vectors - Allocate memory for interrupt vectors
2478  * @adapter: board private structure to initialize
2479  *
2480  * We allocate one q_vector per queue interrupt.  If allocation fails we
2481  * return -ENOMEM.
2482  **/
2483 static int ixgbevf_alloc_q_vectors(struct ixgbevf_adapter *adapter)
2484 {
2485         int q_idx, num_q_vectors;
2486         struct ixgbevf_q_vector *q_vector;
2487
2488         num_q_vectors = adapter->num_msix_vectors - NON_Q_VECTORS;
2489
2490         for (q_idx = 0; q_idx < num_q_vectors; q_idx++) {
2491                 q_vector = kzalloc(sizeof(struct ixgbevf_q_vector), GFP_KERNEL);
2492                 if (!q_vector)
2493                         goto err_out;
2494                 q_vector->adapter = adapter;
2495                 q_vector->v_idx = q_idx;
2496                 netif_napi_add(adapter->netdev, &q_vector->napi,
2497                                ixgbevf_poll, 64);
2498                 adapter->q_vector[q_idx] = q_vector;
2499         }
2500
2501         return 0;
2502
2503 err_out:
2504         while (q_idx) {
2505                 q_idx--;
2506                 q_vector = adapter->q_vector[q_idx];
2507 #ifdef CONFIG_NET_RX_BUSY_POLL
2508                 napi_hash_del(&q_vector->napi);
2509 #endif
2510                 netif_napi_del(&q_vector->napi);
2511                 kfree(q_vector);
2512                 adapter->q_vector[q_idx] = NULL;
2513         }
2514         return -ENOMEM;
2515 }
2516
2517 /**
2518  * ixgbevf_free_q_vectors - Free memory allocated for interrupt vectors
2519  * @adapter: board private structure to initialize
2520  *
2521  * This function frees the memory allocated to the q_vectors.  In addition if
2522  * NAPI is enabled it will delete any references to the NAPI struct prior
2523  * to freeing the q_vector.
2524  **/
2525 static void ixgbevf_free_q_vectors(struct ixgbevf_adapter *adapter)
2526 {
2527         int q_idx, num_q_vectors = adapter->num_msix_vectors - NON_Q_VECTORS;
2528
2529         for (q_idx = 0; q_idx < num_q_vectors; q_idx++) {
2530                 struct ixgbevf_q_vector *q_vector = adapter->q_vector[q_idx];
2531
2532                 adapter->q_vector[q_idx] = NULL;
2533 #ifdef CONFIG_NET_RX_BUSY_POLL
2534                 napi_hash_del(&q_vector->napi);
2535 #endif
2536                 netif_napi_del(&q_vector->napi);
2537                 kfree(q_vector);
2538         }
2539 }
2540
2541 /**
2542  * ixgbevf_reset_interrupt_capability - Reset MSIX setup
2543  * @adapter: board private structure
2544  *
2545  **/
2546 static void ixgbevf_reset_interrupt_capability(struct ixgbevf_adapter *adapter)
2547 {
2548         pci_disable_msix(adapter->pdev);
2549         kfree(adapter->msix_entries);
2550         adapter->msix_entries = NULL;
2551 }
2552
2553 /**
2554  * ixgbevf_init_interrupt_scheme - Determine if MSIX is supported and init
2555  * @adapter: board private structure to initialize
2556  *
2557  **/
2558 static int ixgbevf_init_interrupt_scheme(struct ixgbevf_adapter *adapter)
2559 {
2560         int err;
2561
2562         /* Number of supported queues */
2563         ixgbevf_set_num_queues(adapter);
2564
2565         err = ixgbevf_set_interrupt_capability(adapter);
2566         if (err) {
2567                 hw_dbg(&adapter->hw,
2568                        "Unable to setup interrupt capabilities\n");
2569                 goto err_set_interrupt;
2570         }
2571
2572         err = ixgbevf_alloc_q_vectors(adapter);
2573         if (err) {
2574                 hw_dbg(&adapter->hw, "Unable to allocate memory for queue vectors\n");
2575                 goto err_alloc_q_vectors;
2576         }
2577
2578         err = ixgbevf_alloc_queues(adapter);
2579         if (err) {
2580                 pr_err("Unable to allocate memory for queues\n");
2581                 goto err_alloc_queues;
2582         }
2583
2584         hw_dbg(&adapter->hw, "Multiqueue %s: Rx Queue count = %u, Tx Queue count = %u\n",
2585                (adapter->num_rx_queues > 1) ? "Enabled" :
2586                "Disabled", adapter->num_rx_queues, adapter->num_tx_queues);
2587
2588         set_bit(__IXGBEVF_DOWN, &adapter->state);
2589
2590         return 0;
2591 err_alloc_queues:
2592         ixgbevf_free_q_vectors(adapter);
2593 err_alloc_q_vectors:
2594         ixgbevf_reset_interrupt_capability(adapter);
2595 err_set_interrupt:
2596         return err;
2597 }
2598
2599 /**
2600  * ixgbevf_clear_interrupt_scheme - Clear the current interrupt scheme settings
2601  * @adapter: board private structure to clear interrupt scheme on
2602  *
2603  * We go through and clear interrupt specific resources and reset the structure
2604  * to pre-load conditions
2605  **/
2606 static void ixgbevf_clear_interrupt_scheme(struct ixgbevf_adapter *adapter)
2607 {
2608         int i;
2609
2610         for (i = 0; i < adapter->num_tx_queues; i++) {
2611                 kfree(adapter->tx_ring[i]);
2612                 adapter->tx_ring[i] = NULL;
2613         }
2614         for (i = 0; i < adapter->num_rx_queues; i++) {
2615                 kfree(adapter->rx_ring[i]);
2616                 adapter->rx_ring[i] = NULL;
2617         }
2618
2619         adapter->num_tx_queues = 0;
2620         adapter->num_rx_queues = 0;
2621
2622         ixgbevf_free_q_vectors(adapter);
2623         ixgbevf_reset_interrupt_capability(adapter);
2624 }
2625
2626 /**
2627  * ixgbevf_sw_init - Initialize general software structures
2628  * @adapter: board private structure to initialize
2629  *
2630  * ixgbevf_sw_init initializes the Adapter private data structure.
2631  * Fields are initialized based on PCI device information and
2632  * OS network device settings (MTU size).
2633  **/
2634 static int ixgbevf_sw_init(struct ixgbevf_adapter *adapter)
2635 {
2636         struct ixgbe_hw *hw = &adapter->hw;
2637         struct pci_dev *pdev = adapter->pdev;
2638         struct net_device *netdev = adapter->netdev;
2639         int err;
2640
2641         /* PCI config space info */
2642         hw->vendor_id = pdev->vendor;
2643         hw->device_id = pdev->device;
2644         hw->revision_id = pdev->revision;
2645         hw->subsystem_vendor_id = pdev->subsystem_vendor;
2646         hw->subsystem_device_id = pdev->subsystem_device;
2647
2648         hw->mbx.ops.init_params(hw);
2649
2650         /* assume legacy case in which PF would only give VF 2 queues */
2651         hw->mac.max_tx_queues = 2;
2652         hw->mac.max_rx_queues = 2;
2653
2654         /* lock to protect mailbox accesses */
2655         spin_lock_init(&adapter->mbx_lock);
2656
2657         err = hw->mac.ops.reset_hw(hw);
2658         if (err) {
2659                 dev_info(&pdev->dev,
2660                          "PF still in reset state.  Is the PF interface up?\n");
2661         } else {
2662                 err = hw->mac.ops.init_hw(hw);
2663                 if (err) {
2664                         pr_err("init_shared_code failed: %d\n", err);
2665                         goto out;
2666                 }
2667                 ixgbevf_negotiate_api(adapter);
2668                 err = hw->mac.ops.get_mac_addr(hw, hw->mac.addr);
2669                 if (err)
2670                         dev_info(&pdev->dev, "Error reading MAC address\n");
2671                 else if (is_zero_ether_addr(adapter->hw.mac.addr))
2672                         dev_info(&pdev->dev,
2673                                  "MAC address not assigned by administrator.\n");
2674                 ether_addr_copy(netdev->dev_addr, hw->mac.addr);
2675         }
2676
2677         if (!is_valid_ether_addr(netdev->dev_addr)) {
2678                 dev_info(&pdev->dev, "Assigning random MAC address\n");
2679                 eth_hw_addr_random(netdev);
2680                 ether_addr_copy(hw->mac.addr, netdev->dev_addr);
2681                 ether_addr_copy(hw->mac.perm_addr, netdev->dev_addr);
2682         }
2683
2684         /* Enable dynamic interrupt throttling rates */
2685         adapter->rx_itr_setting = 1;
2686         adapter->tx_itr_setting = 1;
2687
2688         /* set default ring sizes */
2689         adapter->tx_ring_count = IXGBEVF_DEFAULT_TXD;
2690         adapter->rx_ring_count = IXGBEVF_DEFAULT_RXD;
2691
2692         set_bit(__IXGBEVF_DOWN, &adapter->state);
2693         return 0;
2694
2695 out:
2696         return err;
2697 }
2698
2699 #define UPDATE_VF_COUNTER_32bit(reg, last_counter, counter)     \
2700         {                                                       \
2701                 u32 current_counter = IXGBE_READ_REG(hw, reg);  \
2702                 if (current_counter < last_counter)             \
2703                         counter += 0x100000000LL;               \
2704                 last_counter = current_counter;                 \
2705                 counter &= 0xFFFFFFFF00000000LL;                \
2706                 counter |= current_counter;                     \
2707         }
2708
2709 #define UPDATE_VF_COUNTER_36bit(reg_lsb, reg_msb, last_counter, counter) \
2710         {                                                                \
2711                 u64 current_counter_lsb = IXGBE_READ_REG(hw, reg_lsb);   \
2712                 u64 current_counter_msb = IXGBE_READ_REG(hw, reg_msb);   \
2713                 u64 current_counter = (current_counter_msb << 32) |      \
2714                         current_counter_lsb;                             \
2715                 if (current_counter < last_counter)                      \
2716                         counter += 0x1000000000LL;                       \
2717                 last_counter = current_counter;                          \
2718                 counter &= 0xFFFFFFF000000000LL;                         \
2719                 counter |= current_counter;                              \
2720         }
2721 /**
2722  * ixgbevf_update_stats - Update the board statistics counters.
2723  * @adapter: board private structure
2724  **/
2725 void ixgbevf_update_stats(struct ixgbevf_adapter *adapter)
2726 {
2727         struct ixgbe_hw *hw = &adapter->hw;
2728         int i;
2729
2730         if (test_bit(__IXGBEVF_DOWN, &adapter->state) ||
2731             test_bit(__IXGBEVF_RESETTING, &adapter->state))
2732                 return;
2733
2734         UPDATE_VF_COUNTER_32bit(IXGBE_VFGPRC, adapter->stats.last_vfgprc,
2735                                 adapter->stats.vfgprc);
2736         UPDATE_VF_COUNTER_32bit(IXGBE_VFGPTC, adapter->stats.last_vfgptc,
2737                                 adapter->stats.vfgptc);
2738         UPDATE_VF_COUNTER_36bit(IXGBE_VFGORC_LSB, IXGBE_VFGORC_MSB,
2739                                 adapter->stats.last_vfgorc,
2740                                 adapter->stats.vfgorc);
2741         UPDATE_VF_COUNTER_36bit(IXGBE_VFGOTC_LSB, IXGBE_VFGOTC_MSB,
2742                                 adapter->stats.last_vfgotc,
2743                                 adapter->stats.vfgotc);
2744         UPDATE_VF_COUNTER_32bit(IXGBE_VFMPRC, adapter->stats.last_vfmprc,
2745                                 adapter->stats.vfmprc);
2746
2747         for (i = 0;  i  < adapter->num_rx_queues;  i++) {
2748                 adapter->hw_csum_rx_error +=
2749                         adapter->rx_ring[i]->hw_csum_rx_error;
2750                 adapter->rx_ring[i]->hw_csum_rx_error = 0;
2751         }
2752 }
2753
2754 /**
2755  * ixgbevf_service_timer - Timer Call-back
2756  * @data: pointer to adapter cast into an unsigned long
2757  **/
2758 static void ixgbevf_service_timer(unsigned long data)
2759 {
2760         struct ixgbevf_adapter *adapter = (struct ixgbevf_adapter *)data;
2761
2762         /* Reset the timer */
2763         mod_timer(&adapter->service_timer, (HZ * 2) + jiffies);
2764
2765         ixgbevf_service_event_schedule(adapter);
2766 }
2767
2768 static void ixgbevf_reset_subtask(struct ixgbevf_adapter *adapter)
2769 {
2770         if (!test_and_clear_bit(__IXGBEVF_RESET_REQUESTED, &adapter->state))
2771                 return;
2772
2773         /* If we're already down or resetting, just bail */
2774         if (test_bit(__IXGBEVF_DOWN, &adapter->state) ||
2775             test_bit(__IXGBEVF_RESETTING, &adapter->state))
2776                 return;
2777
2778         adapter->tx_timeout_count++;
2779
2780         ixgbevf_reinit_locked(adapter);
2781 }
2782
2783 /**
2784  * ixgbevf_check_hang_subtask - check for hung queues and dropped interrupts
2785  * @adapter: pointer to the device adapter structure
2786  *
2787  * This function serves two purposes.  First it strobes the interrupt lines
2788  * in order to make certain interrupts are occurring.  Secondly it sets the
2789  * bits needed to check for TX hangs.  As a result we should immediately
2790  * determine if a hang has occurred.
2791  **/
2792 static void ixgbevf_check_hang_subtask(struct ixgbevf_adapter *adapter)
2793 {
2794         struct ixgbe_hw *hw = &adapter->hw;
2795         u32 eics = 0;
2796         int i;
2797
2798         /* If we're down or resetting, just bail */
2799         if (test_bit(__IXGBEVF_DOWN, &adapter->state) ||
2800             test_bit(__IXGBEVF_RESETTING, &adapter->state))
2801                 return;
2802
2803         /* Force detection of hung controller */
2804         if (netif_carrier_ok(adapter->netdev)) {
2805                 for (i = 0; i < adapter->num_tx_queues; i++)
2806                         set_check_for_tx_hang(adapter->tx_ring[i]);
2807         }
2808
2809         /* get one bit for every active Tx/Rx interrupt vector */
2810         for (i = 0; i < adapter->num_msix_vectors - NON_Q_VECTORS; i++) {
2811                 struct ixgbevf_q_vector *qv = adapter->q_vector[i];
2812
2813                 if (qv->rx.ring || qv->tx.ring)
2814                         eics |= BIT(i);
2815         }
2816
2817         /* Cause software interrupt to ensure rings are cleaned */
2818         IXGBE_WRITE_REG(hw, IXGBE_VTEICS, eics);
2819 }
2820
2821 /**
2822  * ixgbevf_watchdog_update_link - update the link status
2823  * @adapter: pointer to the device adapter structure
2824  **/
2825 static void ixgbevf_watchdog_update_link(struct ixgbevf_adapter *adapter)
2826 {
2827         struct ixgbe_hw *hw = &adapter->hw;
2828         u32 link_speed = adapter->link_speed;
2829         bool link_up = adapter->link_up;
2830         s32 err;
2831
2832         spin_lock_bh(&adapter->mbx_lock);
2833
2834         err = hw->mac.ops.check_link(hw, &link_speed, &link_up, false);
2835
2836         spin_unlock_bh(&adapter->mbx_lock);
2837
2838         /* if check for link returns error we will need to reset */
2839         if (err && time_after(jiffies, adapter->last_reset + (10 * HZ))) {
2840                 set_bit(__IXGBEVF_RESET_REQUESTED, &adapter->state);
2841                 link_up = false;
2842         }
2843
2844         adapter->link_up = link_up;
2845         adapter->link_speed = link_speed;
2846 }
2847
2848 /**
2849  * ixgbevf_watchdog_link_is_up - update netif_carrier status and
2850  *                               print link up message
2851  * @adapter: pointer to the device adapter structure
2852  **/
2853 static void ixgbevf_watchdog_link_is_up(struct ixgbevf_adapter *adapter)
2854 {
2855         struct net_device *netdev = adapter->netdev;
2856
2857         /* only continue if link was previously down */
2858         if (netif_carrier_ok(netdev))
2859                 return;
2860
2861         dev_info(&adapter->pdev->dev, "NIC Link is Up %s\n",
2862                  (adapter->link_speed == IXGBE_LINK_SPEED_10GB_FULL) ?
2863                  "10 Gbps" :
2864                  (adapter->link_speed == IXGBE_LINK_SPEED_1GB_FULL) ?
2865                  "1 Gbps" :
2866                  (adapter->link_speed == IXGBE_LINK_SPEED_100_FULL) ?
2867                  "100 Mbps" :
2868                  "unknown speed");
2869
2870         netif_carrier_on(netdev);
2871 }
2872
2873 /**
2874  * ixgbevf_watchdog_link_is_down - update netif_carrier status and
2875  *                                 print link down message
2876  * @adapter: pointer to the adapter structure
2877  **/
2878 static void ixgbevf_watchdog_link_is_down(struct ixgbevf_adapter *adapter)
2879 {
2880         struct net_device *netdev = adapter->netdev;
2881
2882         adapter->link_speed = 0;
2883
2884         /* only continue if link was up previously */
2885         if (!netif_carrier_ok(netdev))
2886                 return;
2887
2888         dev_info(&adapter->pdev->dev, "NIC Link is Down\n");
2889
2890         netif_carrier_off(netdev);
2891 }
2892
2893 /**
2894  * ixgbevf_watchdog_subtask - worker thread to bring link up
2895  * @work: pointer to work_struct containing our data
2896  **/
2897 static void ixgbevf_watchdog_subtask(struct ixgbevf_adapter *adapter)
2898 {
2899         /* if interface is down do nothing */
2900         if (test_bit(__IXGBEVF_DOWN, &adapter->state) ||
2901             test_bit(__IXGBEVF_RESETTING, &adapter->state))
2902                 return;
2903
2904         ixgbevf_watchdog_update_link(adapter);
2905
2906         if (adapter->link_up)
2907                 ixgbevf_watchdog_link_is_up(adapter);
2908         else
2909                 ixgbevf_watchdog_link_is_down(adapter);
2910
2911         ixgbevf_update_stats(adapter);
2912 }
2913
2914 /**
2915  * ixgbevf_service_task - manages and runs subtasks
2916  * @work: pointer to work_struct containing our data
2917  **/
2918 static void ixgbevf_service_task(struct work_struct *work)
2919 {
2920         struct ixgbevf_adapter *adapter = container_of(work,
2921                                                        struct ixgbevf_adapter,
2922                                                        service_task);
2923         struct ixgbe_hw *hw = &adapter->hw;
2924
2925         if (IXGBE_REMOVED(hw->hw_addr)) {
2926                 if (!test_bit(__IXGBEVF_DOWN, &adapter->state)) {
2927                         rtnl_lock();
2928                         ixgbevf_down(adapter);
2929                         rtnl_unlock();
2930                 }
2931                 return;
2932         }
2933
2934         ixgbevf_queue_reset_subtask(adapter);
2935         ixgbevf_reset_subtask(adapter);
2936         ixgbevf_watchdog_subtask(adapter);
2937         ixgbevf_check_hang_subtask(adapter);
2938
2939         ixgbevf_service_event_complete(adapter);
2940 }
2941
2942 /**
2943  * ixgbevf_free_tx_resources - Free Tx Resources per Queue
2944  * @tx_ring: Tx descriptor ring for a specific queue
2945  *
2946  * Free all transmit software resources
2947  **/
2948 void ixgbevf_free_tx_resources(struct ixgbevf_ring *tx_ring)
2949 {
2950         ixgbevf_clean_tx_ring(tx_ring);
2951
2952         vfree(tx_ring->tx_buffer_info);
2953         tx_ring->tx_buffer_info = NULL;
2954
2955         /* if not set, then don't free */
2956         if (!tx_ring->desc)
2957                 return;
2958
2959         dma_free_coherent(tx_ring->dev, tx_ring->size, tx_ring->desc,
2960                           tx_ring->dma);
2961
2962         tx_ring->desc = NULL;
2963 }
2964
2965 /**
2966  * ixgbevf_free_all_tx_resources - Free Tx Resources for All Queues
2967  * @adapter: board private structure
2968  *
2969  * Free all transmit software resources
2970  **/
2971 static void ixgbevf_free_all_tx_resources(struct ixgbevf_adapter *adapter)
2972 {
2973         int i;
2974
2975         for (i = 0; i < adapter->num_tx_queues; i++)
2976                 if (adapter->tx_ring[i]->desc)
2977                         ixgbevf_free_tx_resources(adapter->tx_ring[i]);
2978 }
2979
2980 /**
2981  * ixgbevf_setup_tx_resources - allocate Tx resources (Descriptors)
2982  * @tx_ring: Tx descriptor ring (for a specific queue) to setup
2983  *
2984  * Return 0 on success, negative on failure
2985  **/
2986 int ixgbevf_setup_tx_resources(struct ixgbevf_ring *tx_ring)
2987 {
2988         int size;
2989
2990         size = sizeof(struct ixgbevf_tx_buffer) * tx_ring->count;
2991         tx_ring->tx_buffer_info = vzalloc(size);
2992         if (!tx_ring->tx_buffer_info)
2993                 goto err;
2994
2995         /* round up to nearest 4K */
2996         tx_ring->size = tx_ring->count * sizeof(union ixgbe_adv_tx_desc);
2997         tx_ring->size = ALIGN(tx_ring->size, 4096);
2998
2999         tx_ring->desc = dma_alloc_coherent(tx_ring->dev, tx_ring->size,
3000                                            &tx_ring->dma, GFP_KERNEL);
3001         if (!tx_ring->desc)
3002                 goto err;
3003
3004         return 0;
3005
3006 err:
3007         vfree(tx_ring->tx_buffer_info);
3008         tx_ring->tx_buffer_info = NULL;
3009         hw_dbg(&adapter->hw, "Unable to allocate memory for the transmit descriptor ring\n");
3010         return -ENOMEM;
3011 }
3012
3013 /**
3014  * ixgbevf_setup_all_tx_resources - allocate all queues Tx resources
3015  * @adapter: board private structure
3016  *
3017  * If this function returns with an error, then it's possible one or
3018  * more of the rings is populated (while the rest are not).  It is the
3019  * callers duty to clean those orphaned rings.
3020  *
3021  * Return 0 on success, negative on failure
3022  **/
3023 static int ixgbevf_setup_all_tx_resources(struct ixgbevf_adapter *adapter)
3024 {
3025         int i, err = 0;
3026
3027         for (i = 0; i < adapter->num_tx_queues; i++) {
3028                 err = ixgbevf_setup_tx_resources(adapter->tx_ring[i]);
3029                 if (!err)
3030                         continue;
3031                 hw_dbg(&adapter->hw, "Allocation for Tx Queue %u failed\n", i);
3032                 break;
3033         }
3034
3035         return err;
3036 }
3037
3038 /**
3039  * ixgbevf_setup_rx_resources - allocate Rx resources (Descriptors)
3040  * @rx_ring: Rx descriptor ring (for a specific queue) to setup
3041  *
3042  * Returns 0 on success, negative on failure
3043  **/
3044 int ixgbevf_setup_rx_resources(struct ixgbevf_ring *rx_ring)
3045 {
3046         int size;
3047
3048         size = sizeof(struct ixgbevf_rx_buffer) * rx_ring->count;
3049         rx_ring->rx_buffer_info = vzalloc(size);
3050         if (!rx_ring->rx_buffer_info)
3051                 goto err;
3052
3053         /* Round up to nearest 4K */
3054         rx_ring->size = rx_ring->count * sizeof(union ixgbe_adv_rx_desc);
3055         rx_ring->size = ALIGN(rx_ring->size, 4096);
3056
3057         rx_ring->desc = dma_alloc_coherent(rx_ring->dev, rx_ring->size,
3058                                            &rx_ring->dma, GFP_KERNEL);
3059
3060         if (!rx_ring->desc)
3061                 goto err;
3062
3063         return 0;
3064 err:
3065         vfree(rx_ring->rx_buffer_info);
3066         rx_ring->rx_buffer_info = NULL;
3067         dev_err(rx_ring->dev, "Unable to allocate memory for the Rx descriptor ring\n");
3068         return -ENOMEM;
3069 }
3070
3071 /**
3072  * ixgbevf_setup_all_rx_resources - allocate all queues Rx resources
3073  * @adapter: board private structure
3074  *
3075  * If this function returns with an error, then it's possible one or
3076  * more of the rings is populated (while the rest are not).  It is the
3077  * callers duty to clean those orphaned rings.
3078  *
3079  * Return 0 on success, negative on failure
3080  **/
3081 static int ixgbevf_setup_all_rx_resources(struct ixgbevf_adapter *adapter)
3082 {
3083         int i, err = 0;
3084
3085         for (i = 0; i < adapter->num_rx_queues; i++) {
3086                 err = ixgbevf_setup_rx_resources(adapter->rx_ring[i]);
3087                 if (!err)
3088                         continue;
3089                 hw_dbg(&adapter->hw, "Allocation for Rx Queue %u failed\n", i);
3090                 break;
3091         }
3092         return err;
3093 }
3094
3095 /**
3096  * ixgbevf_free_rx_resources - Free Rx Resources
3097  * @rx_ring: ring to clean the resources from
3098  *
3099  * Free all receive software resources
3100  **/
3101 void ixgbevf_free_rx_resources(struct ixgbevf_ring *rx_ring)
3102 {
3103         ixgbevf_clean_rx_ring(rx_ring);
3104
3105         vfree(rx_ring->rx_buffer_info);
3106         rx_ring->rx_buffer_info = NULL;
3107
3108         dma_free_coherent(rx_ring->dev, rx_ring->size, rx_ring->desc,
3109                           rx_ring->dma);
3110
3111         rx_ring->desc = NULL;
3112 }
3113
3114 /**
3115  * ixgbevf_free_all_rx_resources - Free Rx Resources for All Queues
3116  * @adapter: board private structure
3117  *
3118  * Free all receive software resources
3119  **/
3120 static void ixgbevf_free_all_rx_resources(struct ixgbevf_adapter *adapter)
3121 {
3122         int i;
3123
3124         for (i = 0; i < adapter->num_rx_queues; i++)
3125                 if (adapter->rx_ring[i]->desc)
3126                         ixgbevf_free_rx_resources(adapter->rx_ring[i]);
3127 }
3128
3129 /**
3130  * ixgbevf_open - Called when a network interface is made active
3131  * @netdev: network interface device structure
3132  *
3133  * Returns 0 on success, negative value on failure
3134  *
3135  * The open entry point is called when a network interface is made
3136  * active by the system (IFF_UP).  At this point all resources needed
3137  * for transmit and receive operations are allocated, the interrupt
3138  * handler is registered with the OS, the watchdog timer is started,
3139  * and the stack is notified that the interface is ready.
3140  **/
3141 int ixgbevf_open(struct net_device *netdev)
3142 {
3143         struct ixgbevf_adapter *adapter = netdev_priv(netdev);
3144         struct ixgbe_hw *hw = &adapter->hw;
3145         int err;
3146
3147         /* A previous failure to open the device because of a lack of
3148          * available MSIX vector resources may have reset the number
3149          * of msix vectors variable to zero.  The only way to recover
3150          * is to unload/reload the driver and hope that the system has
3151          * been able to recover some MSIX vector resources.
3152          */
3153         if (!adapter->num_msix_vectors)
3154                 return -ENOMEM;
3155
3156         if (hw->adapter_stopped) {
3157                 ixgbevf_reset(adapter);
3158                 /* if adapter is still stopped then PF isn't up and
3159                  * the VF can't start.
3160                  */
3161                 if (hw->adapter_stopped) {
3162                         err = IXGBE_ERR_MBX;
3163                         pr_err("Unable to start - perhaps the PF Driver isn't up yet\n");
3164                         goto err_setup_reset;
3165                 }
3166         }
3167
3168         /* disallow open during test */
3169         if (test_bit(__IXGBEVF_TESTING, &adapter->state))
3170                 return -EBUSY;
3171
3172         netif_carrier_off(netdev);
3173
3174         /* allocate transmit descriptors */
3175         err = ixgbevf_setup_all_tx_resources(adapter);
3176         if (err)
3177                 goto err_setup_tx;
3178
3179         /* allocate receive descriptors */
3180         err = ixgbevf_setup_all_rx_resources(adapter);
3181         if (err)
3182                 goto err_setup_rx;
3183
3184         ixgbevf_configure(adapter);
3185
3186         /* Map the Tx/Rx rings to the vectors we were allotted.
3187          * if request_irq will be called in this function map_rings
3188          * must be called *before* up_complete
3189          */
3190         ixgbevf_map_rings_to_vectors(adapter);
3191
3192         err = ixgbevf_request_irq(adapter);
3193         if (err)
3194                 goto err_req_irq;
3195
3196         ixgbevf_up_complete(adapter);
3197
3198         return 0;
3199
3200 err_req_irq:
3201         ixgbevf_down(adapter);
3202 err_setup_rx:
3203         ixgbevf_free_all_rx_resources(adapter);
3204 err_setup_tx:
3205         ixgbevf_free_all_tx_resources(adapter);
3206         ixgbevf_reset(adapter);
3207
3208 err_setup_reset:
3209
3210         return err;
3211 }
3212
3213 /**
3214  * ixgbevf_close - Disables a network interface
3215  * @netdev: network interface device structure
3216  *
3217  * Returns 0, this is not allowed to fail
3218  *
3219  * The close entry point is called when an interface is de-activated
3220  * by the OS.  The hardware is still under the drivers control, but
3221  * needs to be disabled.  A global MAC reset is issued to stop the
3222  * hardware, and all transmit and receive resources are freed.
3223  **/
3224 int ixgbevf_close(struct net_device *netdev)
3225 {
3226         struct ixgbevf_adapter *adapter = netdev_priv(netdev);
3227
3228         ixgbevf_down(adapter);
3229         ixgbevf_free_irq(adapter);
3230
3231         ixgbevf_free_all_tx_resources(adapter);
3232         ixgbevf_free_all_rx_resources(adapter);
3233
3234         return 0;
3235 }
3236
3237 static void ixgbevf_queue_reset_subtask(struct ixgbevf_adapter *adapter)
3238 {
3239         struct net_device *dev = adapter->netdev;
3240
3241         if (!test_and_clear_bit(__IXGBEVF_QUEUE_RESET_REQUESTED,
3242                                 &adapter->state))
3243                 return;
3244
3245         /* if interface is down do nothing */
3246         if (test_bit(__IXGBEVF_DOWN, &adapter->state) ||
3247             test_bit(__IXGBEVF_RESETTING, &adapter->state))
3248                 return;
3249
3250         /* Hardware has to reinitialize queues and interrupts to
3251          * match packet buffer alignment. Unfortunately, the
3252          * hardware is not flexible enough to do this dynamically.
3253          */
3254         if (netif_running(dev))
3255                 ixgbevf_close(dev);
3256
3257         ixgbevf_clear_interrupt_scheme(adapter);
3258         ixgbevf_init_interrupt_scheme(adapter);
3259
3260         if (netif_running(dev))
3261                 ixgbevf_open(dev);
3262 }
3263
3264 static void ixgbevf_tx_ctxtdesc(struct ixgbevf_ring *tx_ring,
3265                                 u32 vlan_macip_lens, u32 type_tucmd,
3266                                 u32 mss_l4len_idx)
3267 {
3268         struct ixgbe_adv_tx_context_desc *context_desc;
3269         u16 i = tx_ring->next_to_use;
3270
3271         context_desc = IXGBEVF_TX_CTXTDESC(tx_ring, i);
3272
3273         i++;
3274         tx_ring->next_to_use = (i < tx_ring->count) ? i : 0;
3275
3276         /* set bits to identify this as an advanced context descriptor */
3277         type_tucmd |= IXGBE_TXD_CMD_DEXT | IXGBE_ADVTXD_DTYP_CTXT;
3278
3279         context_desc->vlan_macip_lens   = cpu_to_le32(vlan_macip_lens);
3280         context_desc->seqnum_seed       = 0;
3281         context_desc->type_tucmd_mlhl   = cpu_to_le32(type_tucmd);
3282         context_desc->mss_l4len_idx     = cpu_to_le32(mss_l4len_idx);
3283 }
3284
3285 static int ixgbevf_tso(struct ixgbevf_ring *tx_ring,
3286                        struct ixgbevf_tx_buffer *first,
3287                        u8 *hdr_len)
3288 {
3289         u32 vlan_macip_lens, type_tucmd, mss_l4len_idx;
3290         struct sk_buff *skb = first->skb;
3291         union {
3292                 struct iphdr *v4;
3293                 struct ipv6hdr *v6;
3294                 unsigned char *hdr;
3295         } ip;
3296         union {
3297                 struct tcphdr *tcp;
3298                 unsigned char *hdr;
3299         } l4;
3300         u32 paylen, l4_offset;
3301         int err;
3302
3303         if (skb->ip_summed != CHECKSUM_PARTIAL)
3304                 return 0;
3305
3306         if (!skb_is_gso(skb))
3307                 return 0;
3308
3309         err = skb_cow_head(skb, 0);
3310         if (err < 0)
3311                 return err;
3312
3313         ip.hdr = skb_network_header(skb);
3314         l4.hdr = skb_checksum_start(skb);
3315
3316         /* ADV DTYP TUCMD MKRLOC/ISCSIHEDLEN */
3317         type_tucmd = IXGBE_ADVTXD_TUCMD_L4T_TCP;
3318
3319         /* initialize outer IP header fields */
3320         if (ip.v4->version == 4) {
3321                 /* IP header will have to cancel out any data that
3322                  * is not a part of the outer IP header
3323                  */
3324                 ip.v4->check = csum_fold(csum_add(lco_csum(skb),
3325                                                   csum_unfold(l4.tcp->check)));
3326                 type_tucmd |= IXGBE_ADVTXD_TUCMD_IPV4;
3327
3328                 ip.v4->tot_len = 0;
3329                 first->tx_flags |= IXGBE_TX_FLAGS_TSO |
3330                                    IXGBE_TX_FLAGS_CSUM |
3331                                    IXGBE_TX_FLAGS_IPV4;
3332         } else {
3333                 ip.v6->payload_len = 0;
3334                 first->tx_flags |= IXGBE_TX_FLAGS_TSO |
3335                                    IXGBE_TX_FLAGS_CSUM;
3336         }
3337
3338         /* determine offset of inner transport header */
3339         l4_offset = l4.hdr - skb->data;
3340
3341         /* compute length of segmentation header */
3342         *hdr_len = (l4.tcp->doff * 4) + l4_offset;
3343
3344         /* remove payload length from inner checksum */
3345         paylen = skb->len - l4_offset;
3346         csum_replace_by_diff(&l4.tcp->check, htonl(paylen));
3347
3348         /* update gso size and bytecount with header size */
3349         first->gso_segs = skb_shinfo(skb)->gso_segs;
3350         first->bytecount += (first->gso_segs - 1) * *hdr_len;
3351
3352         /* mss_l4len_id: use 1 as index for TSO */
3353         mss_l4len_idx = (*hdr_len - l4_offset) << IXGBE_ADVTXD_L4LEN_SHIFT;
3354         mss_l4len_idx |= skb_shinfo(skb)->gso_size << IXGBE_ADVTXD_MSS_SHIFT;
3355         mss_l4len_idx |= (1u << IXGBE_ADVTXD_IDX_SHIFT);
3356
3357         /* vlan_macip_lens: HEADLEN, MACLEN, VLAN tag */
3358         vlan_macip_lens = l4.hdr - ip.hdr;
3359         vlan_macip_lens |= (ip.hdr - skb->data) << IXGBE_ADVTXD_MACLEN_SHIFT;
3360         vlan_macip_lens |= first->tx_flags & IXGBE_TX_FLAGS_VLAN_MASK;
3361
3362         ixgbevf_tx_ctxtdesc(tx_ring, vlan_macip_lens,
3363                             type_tucmd, mss_l4len_idx);
3364
3365         return 1;
3366 }
3367
3368 static inline bool ixgbevf_ipv6_csum_is_sctp(struct sk_buff *skb)
3369 {
3370         unsigned int offset = 0;
3371
3372         ipv6_find_hdr(skb, &offset, IPPROTO_SCTP, NULL, NULL);
3373
3374         return offset == skb_checksum_start_offset(skb);
3375 }
3376
3377 static void ixgbevf_tx_csum(struct ixgbevf_ring *tx_ring,
3378                             struct ixgbevf_tx_buffer *first)
3379 {
3380         struct sk_buff *skb = first->skb;
3381         u32 vlan_macip_lens = 0;
3382         u32 type_tucmd = 0;
3383
3384         if (skb->ip_summed != CHECKSUM_PARTIAL)
3385                 goto no_csum;
3386
3387         switch (skb->csum_offset) {
3388         case offsetof(struct tcphdr, check):
3389                 type_tucmd = IXGBE_ADVTXD_TUCMD_L4T_TCP;
3390                 /* fall through */
3391         case offsetof(struct udphdr, check):
3392                 break;
3393         case offsetof(struct sctphdr, checksum):
3394                 /* validate that this is actually an SCTP request */
3395                 if (((first->protocol == htons(ETH_P_IP)) &&
3396                      (ip_hdr(skb)->protocol == IPPROTO_SCTP)) ||
3397                     ((first->protocol == htons(ETH_P_IPV6)) &&
3398                      ixgbevf_ipv6_csum_is_sctp(skb))) {
3399                         type_tucmd = IXGBE_ADVTXD_TUCMD_L4T_SCTP;
3400                         break;
3401                 }
3402                 /* fall through */
3403         default:
3404                 skb_checksum_help(skb);
3405                 goto no_csum;
3406         }
3407         /* update TX checksum flag */
3408         first->tx_flags |= IXGBE_TX_FLAGS_CSUM;
3409         vlan_macip_lens = skb_checksum_start_offset(skb) -
3410                           skb_network_offset(skb);
3411 no_csum:
3412         /* vlan_macip_lens: MACLEN, VLAN tag */
3413         vlan_macip_lens |= skb_network_offset(skb) << IXGBE_ADVTXD_MACLEN_SHIFT;
3414         vlan_macip_lens |= first->tx_flags & IXGBE_TX_FLAGS_VLAN_MASK;
3415
3416         ixgbevf_tx_ctxtdesc(tx_ring, vlan_macip_lens, type_tucmd, 0);
3417 }
3418
3419 static __le32 ixgbevf_tx_cmd_type(u32 tx_flags)
3420 {
3421         /* set type for advanced descriptor with frame checksum insertion */
3422         __le32 cmd_type = cpu_to_le32(IXGBE_ADVTXD_DTYP_DATA |
3423                                       IXGBE_ADVTXD_DCMD_IFCS |
3424                                       IXGBE_ADVTXD_DCMD_DEXT);
3425
3426         /* set HW VLAN bit if VLAN is present */
3427         if (tx_flags & IXGBE_TX_FLAGS_VLAN)
3428                 cmd_type |= cpu_to_le32(IXGBE_ADVTXD_DCMD_VLE);
3429
3430         /* set segmentation enable bits for TSO/FSO */
3431         if (tx_flags & IXGBE_TX_FLAGS_TSO)
3432                 cmd_type |= cpu_to_le32(IXGBE_ADVTXD_DCMD_TSE);
3433
3434         return cmd_type;
3435 }
3436
3437 static void ixgbevf_tx_olinfo_status(union ixgbe_adv_tx_desc *tx_desc,
3438                                      u32 tx_flags, unsigned int paylen)
3439 {
3440         __le32 olinfo_status = cpu_to_le32(paylen << IXGBE_ADVTXD_PAYLEN_SHIFT);
3441
3442         /* enable L4 checksum for TSO and TX checksum offload */
3443         if (tx_flags & IXGBE_TX_FLAGS_CSUM)
3444                 olinfo_status |= cpu_to_le32(IXGBE_ADVTXD_POPTS_TXSM);
3445
3446         /* enble IPv4 checksum for TSO */
3447         if (tx_flags & IXGBE_TX_FLAGS_IPV4)
3448                 olinfo_status |= cpu_to_le32(IXGBE_ADVTXD_POPTS_IXSM);
3449
3450         /* use index 1 context for TSO/FSO/FCOE */
3451         if (tx_flags & IXGBE_TX_FLAGS_TSO)
3452                 olinfo_status |= cpu_to_le32(1u << IXGBE_ADVTXD_IDX_SHIFT);
3453
3454         /* Check Context must be set if Tx switch is enabled, which it
3455          * always is for case where virtual functions are running
3456          */
3457         olinfo_status |= cpu_to_le32(IXGBE_ADVTXD_CC);
3458
3459         tx_desc->read.olinfo_status = olinfo_status;
3460 }
3461
3462 static void ixgbevf_tx_map(struct ixgbevf_ring *tx_ring,
3463                            struct ixgbevf_tx_buffer *first,
3464                            const u8 hdr_len)
3465 {
3466         dma_addr_t dma;
3467         struct sk_buff *skb = first->skb;
3468         struct ixgbevf_tx_buffer *tx_buffer;
3469         union ixgbe_adv_tx_desc *tx_desc;
3470         struct skb_frag_struct *frag = &skb_shinfo(skb)->frags[0];
3471         unsigned int data_len = skb->data_len;
3472         unsigned int size = skb_headlen(skb);
3473         unsigned int paylen = skb->len - hdr_len;
3474         u32 tx_flags = first->tx_flags;
3475         __le32 cmd_type;
3476         u16 i = tx_ring->next_to_use;
3477
3478         tx_desc = IXGBEVF_TX_DESC(tx_ring, i);
3479
3480         ixgbevf_tx_olinfo_status(tx_desc, tx_flags, paylen);
3481         cmd_type = ixgbevf_tx_cmd_type(tx_flags);
3482
3483         dma = dma_map_single(tx_ring->dev, skb->data, size, DMA_TO_DEVICE);
3484         if (dma_mapping_error(tx_ring->dev, dma))
3485                 goto dma_error;
3486
3487         /* record length, and DMA address */
3488         dma_unmap_len_set(first, len, size);
3489         dma_unmap_addr_set(first, dma, dma);
3490
3491         tx_desc->read.buffer_addr = cpu_to_le64(dma);
3492
3493         for (;;) {
3494                 while (unlikely(size > IXGBE_MAX_DATA_PER_TXD)) {
3495                         tx_desc->read.cmd_type_len =
3496                                 cmd_type | cpu_to_le32(IXGBE_MAX_DATA_PER_TXD);
3497
3498                         i++;
3499                         tx_desc++;
3500                         if (i == tx_ring->count) {
3501                                 tx_desc = IXGBEVF_TX_DESC(tx_ring, 0);
3502                                 i = 0;
3503                         }
3504
3505                         dma += IXGBE_MAX_DATA_PER_TXD;
3506                         size -= IXGBE_MAX_DATA_PER_TXD;
3507
3508                         tx_desc->read.buffer_addr = cpu_to_le64(dma);
3509                         tx_desc->read.olinfo_status = 0;
3510                 }
3511
3512                 if (likely(!data_len))
3513                         break;
3514
3515                 tx_desc->read.cmd_type_len = cmd_type | cpu_to_le32(size);
3516
3517                 i++;
3518                 tx_desc++;
3519                 if (i == tx_ring->count) {
3520                         tx_desc = IXGBEVF_TX_DESC(tx_ring, 0);
3521                         i = 0;
3522                 }
3523
3524                 size = skb_frag_size(frag);
3525                 data_len -= size;
3526
3527                 dma = skb_frag_dma_map(tx_ring->dev, frag, 0, size,
3528                                        DMA_TO_DEVICE);
3529                 if (dma_mapping_error(tx_ring->dev, dma))
3530                         goto dma_error;
3531
3532                 tx_buffer = &tx_ring->tx_buffer_info[i];
3533                 dma_unmap_len_set(tx_buffer, len, size);
3534                 dma_unmap_addr_set(tx_buffer, dma, dma);
3535
3536                 tx_desc->read.buffer_addr = cpu_to_le64(dma);
3537                 tx_desc->read.olinfo_status = 0;
3538
3539                 frag++;
3540         }
3541
3542         /* write last descriptor with RS and EOP bits */
3543         cmd_type |= cpu_to_le32(size) | cpu_to_le32(IXGBE_TXD_CMD);
3544         tx_desc->read.cmd_type_len = cmd_type;
3545
3546         /* set the timestamp */
3547         first->time_stamp = jiffies;
3548
3549         /* Force memory writes to complete before letting h/w know there
3550          * are new descriptors to fetch.  (Only applicable for weak-ordered
3551          * memory model archs, such as IA-64).
3552          *
3553          * We also need this memory barrier (wmb) to make certain all of the
3554          * status bits have been updated before next_to_watch is written.
3555          */
3556         wmb();
3557
3558         /* set next_to_watch value indicating a packet is present */
3559         first->next_to_watch = tx_desc;
3560
3561         i++;
3562         if (i == tx_ring->count)
3563                 i = 0;
3564
3565         tx_ring->next_to_use = i;
3566
3567         /* notify HW of packet */
3568         ixgbevf_write_tail(tx_ring, i);
3569
3570         return;
3571 dma_error:
3572         dev_err(tx_ring->dev, "TX DMA map failed\n");
3573
3574         /* clear dma mappings for failed tx_buffer_info map */
3575         for (;;) {
3576                 tx_buffer = &tx_ring->tx_buffer_info[i];
3577                 ixgbevf_unmap_and_free_tx_resource(tx_ring, tx_buffer);
3578                 if (tx_buffer == first)
3579                         break;
3580                 if (i == 0)
3581                         i = tx_ring->count;
3582                 i--;
3583         }
3584
3585         tx_ring->next_to_use = i;
3586 }
3587
3588 static int __ixgbevf_maybe_stop_tx(struct ixgbevf_ring *tx_ring, int size)
3589 {
3590         netif_stop_subqueue(tx_ring->netdev, tx_ring->queue_index);
3591         /* Herbert's original patch had:
3592          *  smp_mb__after_netif_stop_queue();
3593          * but since that doesn't exist yet, just open code it.
3594          */
3595         smp_mb();
3596
3597         /* We need to check again in a case another CPU has just
3598          * made room available.
3599          */
3600         if (likely(ixgbevf_desc_unused(tx_ring) < size))
3601                 return -EBUSY;
3602
3603         /* A reprieve! - use start_queue because it doesn't call schedule */
3604         netif_start_subqueue(tx_ring->netdev, tx_ring->queue_index);
3605         ++tx_ring->tx_stats.restart_queue;
3606
3607         return 0;
3608 }
3609
3610 static int ixgbevf_maybe_stop_tx(struct ixgbevf_ring *tx_ring, int size)
3611 {
3612         if (likely(ixgbevf_desc_unused(tx_ring) >= size))
3613                 return 0;
3614         return __ixgbevf_maybe_stop_tx(tx_ring, size);
3615 }
3616
3617 static int ixgbevf_xmit_frame(struct sk_buff *skb, struct net_device *netdev)
3618 {
3619         struct ixgbevf_adapter *adapter = netdev_priv(netdev);
3620         struct ixgbevf_tx_buffer *first;
3621         struct ixgbevf_ring *tx_ring;
3622         int tso;
3623         u32 tx_flags = 0;
3624         u16 count = TXD_USE_COUNT(skb_headlen(skb));
3625 #if PAGE_SIZE > IXGBE_MAX_DATA_PER_TXD
3626         unsigned short f;
3627 #endif
3628         u8 hdr_len = 0;
3629         u8 *dst_mac = skb_header_pointer(skb, 0, 0, NULL);
3630
3631         if (!dst_mac || is_link_local_ether_addr(dst_mac)) {
3632                 dev_kfree_skb_any(skb);
3633                 return NETDEV_TX_OK;
3634         }
3635
3636         tx_ring = adapter->tx_ring[skb->queue_mapping];
3637
3638         /* need: 1 descriptor per page * PAGE_SIZE/IXGBE_MAX_DATA_PER_TXD,
3639          *       + 1 desc for skb_headlen/IXGBE_MAX_DATA_PER_TXD,
3640          *       + 2 desc gap to keep tail from touching head,
3641          *       + 1 desc for context descriptor,
3642          * otherwise try next time
3643          */
3644 #if PAGE_SIZE > IXGBE_MAX_DATA_PER_TXD
3645         for (f = 0; f < skb_shinfo(skb)->nr_frags; f++)
3646                 count += TXD_USE_COUNT(skb_shinfo(skb)->frags[f].size);
3647 #else
3648         count += skb_shinfo(skb)->nr_frags;
3649 #endif
3650         if (ixgbevf_maybe_stop_tx(tx_ring, count + 3)) {
3651                 tx_ring->tx_stats.tx_busy++;
3652                 return NETDEV_TX_BUSY;
3653         }
3654
3655         /* record the location of the first descriptor for this packet */
3656         first = &tx_ring->tx_buffer_info[tx_ring->next_to_use];
3657         first->skb = skb;
3658         first->bytecount = skb->len;
3659         first->gso_segs = 1;
3660
3661         if (skb_vlan_tag_present(skb)) {
3662                 tx_flags |= skb_vlan_tag_get(skb);
3663                 tx_flags <<= IXGBE_TX_FLAGS_VLAN_SHIFT;
3664                 tx_flags |= IXGBE_TX_FLAGS_VLAN;
3665         }
3666
3667         /* record initial flags and protocol */
3668         first->tx_flags = tx_flags;
3669         first->protocol = vlan_get_protocol(skb);
3670
3671         tso = ixgbevf_tso(tx_ring, first, &hdr_len);
3672         if (tso < 0)
3673                 goto out_drop;
3674         else if (!tso)
3675                 ixgbevf_tx_csum(tx_ring, first);
3676
3677         ixgbevf_tx_map(tx_ring, first, hdr_len);
3678
3679         ixgbevf_maybe_stop_tx(tx_ring, DESC_NEEDED);
3680
3681         return NETDEV_TX_OK;
3682
3683 out_drop:
3684         dev_kfree_skb_any(first->skb);
3685         first->skb = NULL;
3686
3687         return NETDEV_TX_OK;
3688 }
3689
3690 /**
3691  * ixgbevf_set_mac - Change the Ethernet Address of the NIC
3692  * @netdev: network interface device structure
3693  * @p: pointer to an address structure
3694  *
3695  * Returns 0 on success, negative on failure
3696  **/
3697 static int ixgbevf_set_mac(struct net_device *netdev, void *p)
3698 {
3699         struct ixgbevf_adapter *adapter = netdev_priv(netdev);
3700         struct ixgbe_hw *hw = &adapter->hw;
3701         struct sockaddr *addr = p;
3702         int err;
3703
3704         if (!is_valid_ether_addr(addr->sa_data))
3705                 return -EADDRNOTAVAIL;
3706
3707         spin_lock_bh(&adapter->mbx_lock);
3708
3709         err = hw->mac.ops.set_rar(hw, 0, addr->sa_data, 0);
3710
3711         spin_unlock_bh(&adapter->mbx_lock);
3712
3713         if (err)
3714                 return -EPERM;
3715
3716         ether_addr_copy(hw->mac.addr, addr->sa_data);
3717         ether_addr_copy(netdev->dev_addr, addr->sa_data);
3718
3719         return 0;
3720 }
3721
3722 /**
3723  * ixgbevf_change_mtu - Change the Maximum Transfer Unit
3724  * @netdev: network interface device structure
3725  * @new_mtu: new value for maximum frame size
3726  *
3727  * Returns 0 on success, negative on failure
3728  **/
3729 static int ixgbevf_change_mtu(struct net_device *netdev, int new_mtu)
3730 {
3731         struct ixgbevf_adapter *adapter = netdev_priv(netdev);
3732         struct ixgbe_hw *hw = &adapter->hw;
3733         int max_frame = new_mtu + ETH_HLEN + ETH_FCS_LEN;
3734         int max_possible_frame = MAXIMUM_ETHERNET_VLAN_SIZE;
3735
3736         switch (adapter->hw.api_version) {
3737         case ixgbe_mbox_api_11:
3738         case ixgbe_mbox_api_12:
3739                 max_possible_frame = IXGBE_MAX_JUMBO_FRAME_SIZE;
3740                 break;
3741         default:
3742                 if (adapter->hw.mac.type != ixgbe_mac_82599_vf)
3743                         max_possible_frame = IXGBE_MAX_JUMBO_FRAME_SIZE;
3744                 break;
3745         }
3746
3747         /* MTU < 68 is an error and causes problems on some kernels */
3748         if ((new_mtu < 68) || (max_frame > max_possible_frame))
3749                 return -EINVAL;
3750
3751         hw_dbg(hw, "changing MTU from %d to %d\n",
3752                netdev->mtu, new_mtu);
3753         /* must set new MTU before calling down or up */
3754         netdev->mtu = new_mtu;
3755
3756         /* notify the PF of our intent to use this size of frame */
3757         hw->mac.ops.set_rlpml(hw, max_frame);
3758
3759         return 0;
3760 }
3761
3762 #ifdef CONFIG_NET_POLL_CONTROLLER
3763 /* Polling 'interrupt' - used by things like netconsole to send skbs
3764  * without having to re-enable interrupts. It's not called while
3765  * the interrupt routine is executing.
3766  */
3767 static void ixgbevf_netpoll(struct net_device *netdev)
3768 {
3769         struct ixgbevf_adapter *adapter = netdev_priv(netdev);
3770         int i;
3771
3772         /* if interface is down do nothing */
3773         if (test_bit(__IXGBEVF_DOWN, &adapter->state))
3774                 return;
3775         for (i = 0; i < adapter->num_rx_queues; i++)
3776                 ixgbevf_msix_clean_rings(0, adapter->q_vector[i]);
3777 }
3778 #endif /* CONFIG_NET_POLL_CONTROLLER */
3779
3780 static int ixgbevf_suspend(struct pci_dev *pdev, pm_message_t state)
3781 {
3782         struct net_device *netdev = pci_get_drvdata(pdev);
3783         struct ixgbevf_adapter *adapter = netdev_priv(netdev);
3784 #ifdef CONFIG_PM
3785         int retval = 0;
3786 #endif
3787
3788         netif_device_detach(netdev);
3789
3790         if (netif_running(netdev)) {
3791                 rtnl_lock();
3792                 ixgbevf_down(adapter);
3793                 ixgbevf_free_irq(adapter);
3794                 ixgbevf_free_all_tx_resources(adapter);
3795                 ixgbevf_free_all_rx_resources(adapter);
3796                 rtnl_unlock();
3797         }
3798
3799         ixgbevf_clear_interrupt_scheme(adapter);
3800
3801 #ifdef CONFIG_PM
3802         retval = pci_save_state(pdev);
3803         if (retval)
3804                 return retval;
3805
3806 #endif
3807         if (!test_and_set_bit(__IXGBEVF_DISABLED, &adapter->state))
3808                 pci_disable_device(pdev);
3809
3810         return 0;
3811 }
3812
3813 #ifdef CONFIG_PM
3814 static int ixgbevf_resume(struct pci_dev *pdev)
3815 {
3816         struct net_device *netdev = pci_get_drvdata(pdev);
3817         struct ixgbevf_adapter *adapter = netdev_priv(netdev);
3818         u32 err;
3819
3820         pci_restore_state(pdev);
3821         /* pci_restore_state clears dev->state_saved so call
3822          * pci_save_state to restore it.
3823          */
3824         pci_save_state(pdev);
3825
3826         err = pci_enable_device_mem(pdev);
3827         if (err) {
3828                 dev_err(&pdev->dev, "Cannot enable PCI device from suspend\n");
3829                 return err;
3830         }
3831         smp_mb__before_atomic();
3832         clear_bit(__IXGBEVF_DISABLED, &adapter->state);
3833         pci_set_master(pdev);
3834
3835         ixgbevf_reset(adapter);
3836
3837         rtnl_lock();
3838         err = ixgbevf_init_interrupt_scheme(adapter);
3839         rtnl_unlock();
3840         if (err) {
3841                 dev_err(&pdev->dev, "Cannot initialize interrupts\n");
3842                 return err;
3843         }
3844
3845         if (netif_running(netdev)) {
3846                 err = ixgbevf_open(netdev);
3847                 if (err)
3848                         return err;
3849         }
3850
3851         netif_device_attach(netdev);
3852
3853         return err;
3854 }
3855
3856 #endif /* CONFIG_PM */
3857 static void ixgbevf_shutdown(struct pci_dev *pdev)
3858 {
3859         ixgbevf_suspend(pdev, PMSG_SUSPEND);
3860 }
3861
3862 static struct rtnl_link_stats64 *ixgbevf_get_stats(struct net_device *netdev,
3863                                                 struct rtnl_link_stats64 *stats)
3864 {
3865         struct ixgbevf_adapter *adapter = netdev_priv(netdev);
3866         unsigned int start;
3867         u64 bytes, packets;
3868         const struct ixgbevf_ring *ring;
3869         int i;
3870
3871         ixgbevf_update_stats(adapter);
3872
3873         stats->multicast = adapter->stats.vfmprc - adapter->stats.base_vfmprc;
3874
3875         for (i = 0; i < adapter->num_rx_queues; i++) {
3876                 ring = adapter->rx_ring[i];
3877                 do {
3878                         start = u64_stats_fetch_begin_irq(&ring->syncp);
3879                         bytes = ring->stats.bytes;
3880                         packets = ring->stats.packets;
3881                 } while (u64_stats_fetch_retry_irq(&ring->syncp, start));
3882                 stats->rx_bytes += bytes;
3883                 stats->rx_packets += packets;
3884         }
3885
3886         for (i = 0; i < adapter->num_tx_queues; i++) {
3887                 ring = adapter->tx_ring[i];
3888                 do {
3889                         start = u64_stats_fetch_begin_irq(&ring->syncp);
3890                         bytes = ring->stats.bytes;
3891                         packets = ring->stats.packets;
3892                 } while (u64_stats_fetch_retry_irq(&ring->syncp, start));
3893                 stats->tx_bytes += bytes;
3894                 stats->tx_packets += packets;
3895         }
3896
3897         return stats;
3898 }
3899
3900 #define IXGBEVF_MAX_MAC_HDR_LEN         127
3901 #define IXGBEVF_MAX_NETWORK_HDR_LEN     511
3902
3903 static netdev_features_t
3904 ixgbevf_features_check(struct sk_buff *skb, struct net_device *dev,
3905                        netdev_features_t features)
3906 {
3907         unsigned int network_hdr_len, mac_hdr_len;
3908
3909         /* Make certain the headers can be described by a context descriptor */
3910         mac_hdr_len = skb_network_header(skb) - skb->data;
3911         if (unlikely(mac_hdr_len > IXGBEVF_MAX_MAC_HDR_LEN))
3912                 return features & ~(NETIF_F_HW_CSUM |
3913                                     NETIF_F_SCTP_CRC |
3914                                     NETIF_F_HW_VLAN_CTAG_TX |
3915                                     NETIF_F_TSO |
3916                                     NETIF_F_TSO6);
3917
3918         network_hdr_len = skb_checksum_start(skb) - skb_network_header(skb);
3919         if (unlikely(network_hdr_len >  IXGBEVF_MAX_NETWORK_HDR_LEN))
3920                 return features & ~(NETIF_F_HW_CSUM |
3921                                     NETIF_F_SCTP_CRC |
3922                                     NETIF_F_TSO |
3923                                     NETIF_F_TSO6);
3924
3925         /* We can only support IPV4 TSO in tunnels if we can mangle the
3926          * inner IP ID field, so strip TSO if MANGLEID is not supported.
3927          */
3928         if (skb->encapsulation && !(features & NETIF_F_TSO_MANGLEID))
3929                 features &= ~NETIF_F_TSO;
3930
3931         return features;
3932 }
3933
3934 static const struct net_device_ops ixgbevf_netdev_ops = {
3935         .ndo_open               = ixgbevf_open,
3936         .ndo_stop               = ixgbevf_close,
3937         .ndo_start_xmit         = ixgbevf_xmit_frame,
3938         .ndo_set_rx_mode        = ixgbevf_set_rx_mode,
3939         .ndo_get_stats64        = ixgbevf_get_stats,
3940         .ndo_validate_addr      = eth_validate_addr,
3941         .ndo_set_mac_address    = ixgbevf_set_mac,
3942         .ndo_change_mtu         = ixgbevf_change_mtu,
3943         .ndo_tx_timeout         = ixgbevf_tx_timeout,
3944         .ndo_vlan_rx_add_vid    = ixgbevf_vlan_rx_add_vid,
3945         .ndo_vlan_rx_kill_vid   = ixgbevf_vlan_rx_kill_vid,
3946 #ifdef CONFIG_NET_RX_BUSY_POLL
3947         .ndo_busy_poll          = ixgbevf_busy_poll_recv,
3948 #endif
3949 #ifdef CONFIG_NET_POLL_CONTROLLER
3950         .ndo_poll_controller    = ixgbevf_netpoll,
3951 #endif
3952         .ndo_features_check     = ixgbevf_features_check,
3953 };
3954
3955 static void ixgbevf_assign_netdev_ops(struct net_device *dev)
3956 {
3957         dev->netdev_ops = &ixgbevf_netdev_ops;
3958         ixgbevf_set_ethtool_ops(dev);
3959         dev->watchdog_timeo = 5 * HZ;
3960 }
3961
3962 /**
3963  * ixgbevf_probe - Device Initialization Routine
3964  * @pdev: PCI device information struct
3965  * @ent: entry in ixgbevf_pci_tbl
3966  *
3967  * Returns 0 on success, negative on failure
3968  *
3969  * ixgbevf_probe initializes an adapter identified by a pci_dev structure.
3970  * The OS initialization, configuring of the adapter private structure,
3971  * and a hardware reset occur.
3972  **/
3973 static int ixgbevf_probe(struct pci_dev *pdev, const struct pci_device_id *ent)
3974 {
3975         struct net_device *netdev;
3976         struct ixgbevf_adapter *adapter = NULL;
3977         struct ixgbe_hw *hw = NULL;
3978         const struct ixgbevf_info *ii = ixgbevf_info_tbl[ent->driver_data];
3979         int err, pci_using_dac;
3980         bool disable_dev = false;
3981
3982         err = pci_enable_device(pdev);
3983         if (err)
3984                 return err;
3985
3986         if (!dma_set_mask_and_coherent(&pdev->dev, DMA_BIT_MASK(64))) {
3987                 pci_using_dac = 1;
3988         } else {
3989                 err = dma_set_mask_and_coherent(&pdev->dev, DMA_BIT_MASK(32));
3990                 if (err) {
3991                         dev_err(&pdev->dev, "No usable DMA configuration, aborting\n");
3992                         goto err_dma;
3993                 }
3994                 pci_using_dac = 0;
3995         }
3996
3997         err = pci_request_regions(pdev, ixgbevf_driver_name);
3998         if (err) {
3999                 dev_err(&pdev->dev, "pci_request_regions failed 0x%x\n", err);
4000                 goto err_pci_reg;
4001         }
4002
4003         pci_set_master(pdev);
4004
4005         netdev = alloc_etherdev_mq(sizeof(struct ixgbevf_adapter),
4006                                    MAX_TX_QUEUES);
4007         if (!netdev) {
4008                 err = -ENOMEM;
4009                 goto err_alloc_etherdev;
4010         }
4011
4012         SET_NETDEV_DEV(netdev, &pdev->dev);
4013
4014         adapter = netdev_priv(netdev);
4015
4016         adapter->netdev = netdev;
4017         adapter->pdev = pdev;
4018         hw = &adapter->hw;
4019         hw->back = adapter;
4020         adapter->msg_enable = netif_msg_init(debug, DEFAULT_MSG_ENABLE);
4021
4022         /* call save state here in standalone driver because it relies on
4023          * adapter struct to exist, and needs to call netdev_priv
4024          */
4025         pci_save_state(pdev);
4026
4027         hw->hw_addr = ioremap(pci_resource_start(pdev, 0),
4028                               pci_resource_len(pdev, 0));
4029         adapter->io_addr = hw->hw_addr;
4030         if (!hw->hw_addr) {
4031                 err = -EIO;
4032                 goto err_ioremap;
4033         }
4034
4035         ixgbevf_assign_netdev_ops(netdev);
4036
4037         /* Setup HW API */
4038         memcpy(&hw->mac.ops, ii->mac_ops, sizeof(hw->mac.ops));
4039         hw->mac.type  = ii->mac;
4040
4041         memcpy(&hw->mbx.ops, &ixgbevf_mbx_ops,
4042                sizeof(struct ixgbe_mbx_operations));
4043
4044         /* setup the private structure */
4045         err = ixgbevf_sw_init(adapter);
4046         if (err)
4047                 goto err_sw_init;
4048
4049         /* The HW MAC address was set and/or determined in sw_init */
4050         if (!is_valid_ether_addr(netdev->dev_addr)) {
4051                 pr_err("invalid MAC address\n");
4052                 err = -EIO;
4053                 goto err_sw_init;
4054         }
4055
4056         netdev->hw_features = NETIF_F_SG |
4057                               NETIF_F_TSO |
4058                               NETIF_F_TSO6 |
4059                               NETIF_F_RXCSUM |
4060                               NETIF_F_HW_CSUM |
4061                               NETIF_F_SCTP_CRC;
4062
4063 #define IXGBEVF_GSO_PARTIAL_FEATURES (NETIF_F_GSO_GRE | \
4064                                       NETIF_F_GSO_GRE_CSUM | \
4065                                       NETIF_F_GSO_IPXIP4 | \
4066                                       NETIF_F_GSO_IPXIP6 | \
4067                                       NETIF_F_GSO_UDP_TUNNEL | \
4068                                       NETIF_F_GSO_UDP_TUNNEL_CSUM)
4069
4070         netdev->gso_partial_features = IXGBEVF_GSO_PARTIAL_FEATURES;
4071         netdev->hw_features |= NETIF_F_GSO_PARTIAL |
4072                                IXGBEVF_GSO_PARTIAL_FEATURES;
4073
4074         netdev->features = netdev->hw_features;
4075
4076         if (pci_using_dac)
4077                 netdev->features |= NETIF_F_HIGHDMA;
4078
4079         netdev->vlan_features |= netdev->features | NETIF_F_TSO_MANGLEID;
4080         netdev->mpls_features |= NETIF_F_HW_CSUM;
4081         netdev->hw_enc_features |= netdev->vlan_features;
4082
4083         /* set this bit last since it cannot be part of vlan_features */
4084         netdev->features |= NETIF_F_HW_VLAN_CTAG_FILTER |
4085                             NETIF_F_HW_VLAN_CTAG_RX |
4086                             NETIF_F_HW_VLAN_CTAG_TX;
4087
4088         netdev->priv_flags |= IFF_UNICAST_FLT;
4089
4090         if (IXGBE_REMOVED(hw->hw_addr)) {
4091                 err = -EIO;
4092                 goto err_sw_init;
4093         }
4094
4095         setup_timer(&adapter->service_timer, &ixgbevf_service_timer,
4096                     (unsigned long)adapter);
4097
4098         INIT_WORK(&adapter->service_task, ixgbevf_service_task);
4099         set_bit(__IXGBEVF_SERVICE_INITED, &adapter->state);
4100         clear_bit(__IXGBEVF_SERVICE_SCHED, &adapter->state);
4101
4102         err = ixgbevf_init_interrupt_scheme(adapter);
4103         if (err)
4104                 goto err_sw_init;
4105
4106         strcpy(netdev->name, "eth%d");
4107
4108         err = register_netdev(netdev);
4109         if (err)
4110                 goto err_register;
4111
4112         pci_set_drvdata(pdev, netdev);
4113         netif_carrier_off(netdev);
4114
4115         ixgbevf_init_last_counter_stats(adapter);
4116
4117         /* print the VF info */
4118         dev_info(&pdev->dev, "%pM\n", netdev->dev_addr);
4119         dev_info(&pdev->dev, "MAC: %d\n", hw->mac.type);
4120
4121         switch (hw->mac.type) {
4122         case ixgbe_mac_X550_vf:
4123                 dev_info(&pdev->dev, "Intel(R) X550 Virtual Function\n");
4124                 break;
4125         case ixgbe_mac_X540_vf:
4126                 dev_info(&pdev->dev, "Intel(R) X540 Virtual Function\n");
4127                 break;
4128         case ixgbe_mac_82599_vf:
4129         default:
4130                 dev_info(&pdev->dev, "Intel(R) 82599 Virtual Function\n");
4131                 break;
4132         }
4133
4134         return 0;
4135
4136 err_register:
4137         ixgbevf_clear_interrupt_scheme(adapter);
4138 err_sw_init:
4139         ixgbevf_reset_interrupt_capability(adapter);
4140         iounmap(adapter->io_addr);
4141 err_ioremap:
4142         disable_dev = !test_and_set_bit(__IXGBEVF_DISABLED, &adapter->state);
4143         free_netdev(netdev);
4144 err_alloc_etherdev:
4145         pci_release_regions(pdev);
4146 err_pci_reg:
4147 err_dma:
4148         if (!adapter || disable_dev)
4149                 pci_disable_device(pdev);
4150         return err;
4151 }
4152
4153 /**
4154  * ixgbevf_remove - Device Removal Routine
4155  * @pdev: PCI device information struct
4156  *
4157  * ixgbevf_remove is called by the PCI subsystem to alert the driver
4158  * that it should release a PCI device.  The could be caused by a
4159  * Hot-Plug event, or because the driver is going to be removed from
4160  * memory.
4161  **/
4162 static void ixgbevf_remove(struct pci_dev *pdev)
4163 {
4164         struct net_device *netdev = pci_get_drvdata(pdev);
4165         struct ixgbevf_adapter *adapter;
4166         bool disable_dev;
4167
4168         if (!netdev)
4169                 return;
4170
4171         adapter = netdev_priv(netdev);
4172
4173         set_bit(__IXGBEVF_REMOVING, &adapter->state);
4174         cancel_work_sync(&adapter->service_task);
4175
4176         if (netdev->reg_state == NETREG_REGISTERED)
4177                 unregister_netdev(netdev);
4178
4179         ixgbevf_clear_interrupt_scheme(adapter);
4180         ixgbevf_reset_interrupt_capability(adapter);
4181
4182         iounmap(adapter->io_addr);
4183         pci_release_regions(pdev);
4184
4185         hw_dbg(&adapter->hw, "Remove complete\n");
4186
4187         disable_dev = !test_and_set_bit(__IXGBEVF_DISABLED, &adapter->state);
4188         free_netdev(netdev);
4189
4190         if (disable_dev)
4191                 pci_disable_device(pdev);
4192 }
4193
4194 /**
4195  * ixgbevf_io_error_detected - called when PCI error is detected
4196  * @pdev: Pointer to PCI device
4197  * @state: The current pci connection state
4198  *
4199  * This function is called after a PCI bus error affecting
4200  * this device has been detected.
4201  **/
4202 static pci_ers_result_t ixgbevf_io_error_detected(struct pci_dev *pdev,
4203                                                   pci_channel_state_t state)
4204 {
4205         struct net_device *netdev = pci_get_drvdata(pdev);
4206         struct ixgbevf_adapter *adapter = netdev_priv(netdev);
4207
4208         if (!test_bit(__IXGBEVF_SERVICE_INITED, &adapter->state))
4209                 return PCI_ERS_RESULT_DISCONNECT;
4210
4211         rtnl_lock();
4212         netif_device_detach(netdev);
4213
4214         if (state == pci_channel_io_perm_failure) {
4215                 rtnl_unlock();
4216                 return PCI_ERS_RESULT_DISCONNECT;
4217         }
4218
4219         if (netif_running(netdev))
4220                 ixgbevf_down(adapter);
4221
4222         if (!test_and_set_bit(__IXGBEVF_DISABLED, &adapter->state))
4223                 pci_disable_device(pdev);
4224         rtnl_unlock();
4225
4226         /* Request a slot slot reset. */
4227         return PCI_ERS_RESULT_NEED_RESET;
4228 }
4229
4230 /**
4231  * ixgbevf_io_slot_reset - called after the pci bus has been reset.
4232  * @pdev: Pointer to PCI device
4233  *
4234  * Restart the card from scratch, as if from a cold-boot. Implementation
4235  * resembles the first-half of the ixgbevf_resume routine.
4236  **/
4237 static pci_ers_result_t ixgbevf_io_slot_reset(struct pci_dev *pdev)
4238 {
4239         struct net_device *netdev = pci_get_drvdata(pdev);
4240         struct ixgbevf_adapter *adapter = netdev_priv(netdev);
4241
4242         if (pci_enable_device_mem(pdev)) {
4243                 dev_err(&pdev->dev,
4244                         "Cannot re-enable PCI device after reset.\n");
4245                 return PCI_ERS_RESULT_DISCONNECT;
4246         }
4247
4248         smp_mb__before_atomic();
4249         clear_bit(__IXGBEVF_DISABLED, &adapter->state);
4250         pci_set_master(pdev);
4251
4252         ixgbevf_reset(adapter);
4253
4254         return PCI_ERS_RESULT_RECOVERED;
4255 }
4256
4257 /**
4258  * ixgbevf_io_resume - called when traffic can start flowing again.
4259  * @pdev: Pointer to PCI device
4260  *
4261  * This callback is called when the error recovery driver tells us that
4262  * its OK to resume normal operation. Implementation resembles the
4263  * second-half of the ixgbevf_resume routine.
4264  **/
4265 static void ixgbevf_io_resume(struct pci_dev *pdev)
4266 {
4267         struct net_device *netdev = pci_get_drvdata(pdev);
4268         struct ixgbevf_adapter *adapter = netdev_priv(netdev);
4269
4270         if (netif_running(netdev))
4271                 ixgbevf_up(adapter);
4272
4273         netif_device_attach(netdev);
4274 }
4275
4276 /* PCI Error Recovery (ERS) */
4277 static const struct pci_error_handlers ixgbevf_err_handler = {
4278         .error_detected = ixgbevf_io_error_detected,
4279         .slot_reset = ixgbevf_io_slot_reset,
4280         .resume = ixgbevf_io_resume,
4281 };
4282
4283 static struct pci_driver ixgbevf_driver = {
4284         .name           = ixgbevf_driver_name,
4285         .id_table       = ixgbevf_pci_tbl,
4286         .probe          = ixgbevf_probe,
4287         .remove         = ixgbevf_remove,
4288 #ifdef CONFIG_PM
4289         /* Power Management Hooks */
4290         .suspend        = ixgbevf_suspend,
4291         .resume         = ixgbevf_resume,
4292 #endif
4293         .shutdown       = ixgbevf_shutdown,
4294         .err_handler    = &ixgbevf_err_handler
4295 };
4296
4297 /**
4298  * ixgbevf_init_module - Driver Registration Routine
4299  *
4300  * ixgbevf_init_module is the first routine called when the driver is
4301  * loaded. All it does is register with the PCI subsystem.
4302  **/
4303 static int __init ixgbevf_init_module(void)
4304 {
4305         pr_info("%s - version %s\n", ixgbevf_driver_string,
4306                 ixgbevf_driver_version);
4307
4308         pr_info("%s\n", ixgbevf_copyright);
4309         ixgbevf_wq = create_singlethread_workqueue(ixgbevf_driver_name);
4310         if (!ixgbevf_wq) {
4311                 pr_err("%s: Failed to create workqueue\n", ixgbevf_driver_name);
4312                 return -ENOMEM;
4313         }
4314
4315         return pci_register_driver(&ixgbevf_driver);
4316 }
4317
4318 module_init(ixgbevf_init_module);
4319
4320 /**
4321  * ixgbevf_exit_module - Driver Exit Cleanup Routine
4322  *
4323  * ixgbevf_exit_module is called just before the driver is removed
4324  * from memory.
4325  **/
4326 static void __exit ixgbevf_exit_module(void)
4327 {
4328         pci_unregister_driver(&ixgbevf_driver);
4329         if (ixgbevf_wq) {
4330                 destroy_workqueue(ixgbevf_wq);
4331                 ixgbevf_wq = NULL;
4332         }
4333 }
4334
4335 #ifdef DEBUG
4336 /**
4337  * ixgbevf_get_hw_dev_name - return device name string
4338  * used by hardware layer to print debugging information
4339  **/
4340 char *ixgbevf_get_hw_dev_name(struct ixgbe_hw *hw)
4341 {
4342         struct ixgbevf_adapter *adapter = hw->back;
4343
4344         return adapter->netdev->name;
4345 }
4346
4347 #endif
4348 module_exit(ixgbevf_exit_module);
4349
4350 /* ixgbevf_main.c */