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 "3.2.2-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         [board_x550em_a_vf]     = &ixgbevf_x550em_a_vf_info,
  74 };
  75
  76 /* ixgbevf_pci_tbl - PCI Device ID Table
  77  *
  78  * Wildcard entries (PCI_ANY_ID) should come last
  79  * Last entry must be all 0s
  80  *
  81  * { Vendor ID, Device ID, SubVendor ID, SubDevice ID,
  82  *   Class, Class Mask, private data (not used) }
  83  */
  84 static const struct pci_device_id ixgbevf_pci_tbl[] = {
  85         {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82599_VF), board_82599_vf },
  86         {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82599_VF_HV), board_82599_vf_hv },
  87         {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_X540_VF), board_X540_vf },
  88         {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_X540_VF_HV), board_X540_vf_hv },
  89         {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_X550_VF), board_X550_vf },
  90         {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_X550_VF_HV), board_X550_vf_hv },
  91         {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_X550EM_X_VF), board_X550EM_x_vf },
  92         {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_X550EM_X_VF_HV), board_X550EM_x_vf_hv},
  93         {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_X550EM_A_VF), board_x550em_a_vf },
  94         /* required last entry */
  95         {0, }
  96 };
  97 MODULE_DEVICE_TABLE(pci, ixgbevf_pci_tbl);
  98
  99 MODULE_AUTHOR("Intel Corporation, <linux.nics@intel.com>");
 100 MODULE_DESCRIPTION("Intel(R) 10 Gigabit Virtual Function Network Driver");
 101 MODULE_LICENSE("GPL");
 102 MODULE_VERSION(DRV_VERSION);
 103
 104 #define DEFAULT_MSG_ENABLE (NETIF_MSG_DRV|NETIF_MSG_PROBE|NETIF_MSG_LINK)
 105 static int debug = -1;
 106 module_param(debug, int, 0);
 107 MODULE_PARM_DESC(debug, "Debug level (0=none,...,16=all)");
 108
 109 static struct workqueue_struct *ixgbevf_wq;
 110
 111 static void ixgbevf_service_event_schedule(struct ixgbevf_adapter *adapter)
 112 {
 113         if (!test_bit(__IXGBEVF_DOWN, &adapter->state) &&
 114             !test_bit(__IXGBEVF_REMOVING, &adapter->state) &&
 115             !test_and_set_bit(__IXGBEVF_SERVICE_SCHED, &adapter->state))
 116                 queue_work(ixgbevf_wq, &adapter->service_task);
 117 }
 118
 119 static void ixgbevf_service_event_complete(struct ixgbevf_adapter *adapter)
 120 {
 121         BUG_ON(!test_bit(__IXGBEVF_SERVICE_SCHED, &adapter->state));
 122
 123         /* flush memory to make sure state is correct before next watchdog */
 124         smp_mb__before_atomic();
 125         clear_bit(__IXGBEVF_SERVICE_SCHED, &adapter->state);
 126 }
 127
 128 /* forward decls */
 129 static void ixgbevf_queue_reset_subtask(struct ixgbevf_adapter *adapter);
 130 static void ixgbevf_set_itr(struct ixgbevf_q_vector *q_vector);
 131 static void ixgbevf_free_all_rx_resources(struct ixgbevf_adapter *adapter);
 132
 133 static void ixgbevf_remove_adapter(struct ixgbe_hw *hw)
 134 {
 135         struct ixgbevf_adapter *adapter = hw->back;
 136
 137         if (!hw->hw_addr)
 138                 return;
 139         hw->hw_addr = NULL;
 140         dev_err(&adapter->pdev->dev, "Adapter removed\n");
 141         if (test_bit(__IXGBEVF_SERVICE_INITED, &adapter->state))
 142                 ixgbevf_service_event_schedule(adapter);
 143 }
 144
 145 static void ixgbevf_check_remove(struct ixgbe_hw *hw, u32 reg)
 146 {
 147         u32 value;
 148
 149         /* The following check not only optimizes a bit by not
 150          * performing a read on the status register when the
 151          * register just read was a status register read that
 152          * returned IXGBE_FAILED_READ_REG. It also blocks any
 153          * potential recursion.
 154          */
 155         if (reg == IXGBE_VFSTATUS) {
 156                 ixgbevf_remove_adapter(hw);
 157                 return;
 158         }
 159         value = ixgbevf_read_reg(hw, IXGBE_VFSTATUS);
 160         if (value == IXGBE_FAILED_READ_REG)
 161                 ixgbevf_remove_adapter(hw);
 162 }
 163
 164 u32 ixgbevf_read_reg(struct ixgbe_hw *hw, u32 reg)
 165 {
 166         u8 __iomem *reg_addr = ACCESS_ONCE(hw->hw_addr);
 167         u32 value;
 168
 169         if (IXGBE_REMOVED(reg_addr))
 170                 return IXGBE_FAILED_READ_REG;
 171         value = readl(reg_addr + reg);
 172         if (unlikely(value == IXGBE_FAILED_READ_REG))
 173                 ixgbevf_check_remove(hw, reg);
 174         return value;
 175 }
 176
 177 /**
 178  * ixgbevf_set_ivar - set IVAR registers - maps interrupt causes to vectors
 179  * @adapter: pointer to adapter struct
 180  * @direction: 0 for Rx, 1 for Tx, -1 for other causes
 181  * @queue: queue to map the corresponding interrupt to
 182  * @msix_vector: the vector to map to the corresponding queue
 183  **/
 184 static void ixgbevf_set_ivar(struct ixgbevf_adapter *adapter, s8 direction,
 185                              u8 queue, u8 msix_vector)
 186 {
 187         u32 ivar, index;
 188         struct ixgbe_hw *hw = &adapter->hw;
 189
 190         if (direction == -1) {
 191                 /* other causes */
 192                 msix_vector |= IXGBE_IVAR_ALLOC_VAL;
 193                 ivar = IXGBE_READ_REG(hw, IXGBE_VTIVAR_MISC);
 194                 ivar &= ~0xFF;
 195                 ivar |= msix_vector;
 196                 IXGBE_WRITE_REG(hw, IXGBE_VTIVAR_MISC, ivar);
 197         } else {
 198                 /* Tx or Rx causes */
 199                 msix_vector |= IXGBE_IVAR_ALLOC_VAL;
 200                 index = ((16 * (queue & 1)) + (8 * direction));
 201                 ivar = IXGBE_READ_REG(hw, IXGBE_VTIVAR(queue >> 1));
 202                 ivar &= ~(0xFF << index);
 203                 ivar |= (msix_vector << index);
 204                 IXGBE_WRITE_REG(hw, IXGBE_VTIVAR(queue >> 1), ivar);
 205         }
 206 }
 207
 208 static void ixgbevf_unmap_and_free_tx_resource(struct ixgbevf_ring *tx_ring,
 209                                         struct ixgbevf_tx_buffer *tx_buffer)
 210 {
 211         if (tx_buffer->skb) {
 212                 dev_kfree_skb_any(tx_buffer->skb);
 213                 if (dma_unmap_len(tx_buffer, len))
 214                         dma_unmap_single(tx_ring->dev,
 215                                          dma_unmap_addr(tx_buffer, dma),
 216                                          dma_unmap_len(tx_buffer, len),
 217                                          DMA_TO_DEVICE);
 218         } else if (dma_unmap_len(tx_buffer, len)) {
 219                 dma_unmap_page(tx_ring->dev,
 220                                dma_unmap_addr(tx_buffer, dma),
 221                                dma_unmap_len(tx_buffer, len),
 222                                DMA_TO_DEVICE);
 223         }
 224         tx_buffer->next_to_watch = NULL;
 225         tx_buffer->skb = NULL;
 226         dma_unmap_len_set(tx_buffer, len, 0);
 227         /* tx_buffer must be completely set up in the transmit path */
 228 }
 229
 230 static u64 ixgbevf_get_tx_completed(struct ixgbevf_ring *ring)
 231 {
 232         return ring->stats.packets;
 233 }
 234
 235 static u32 ixgbevf_get_tx_pending(struct ixgbevf_ring *ring)
 236 {
 237         struct ixgbevf_adapter *adapter = netdev_priv(ring->netdev);
 238         struct ixgbe_hw *hw = &adapter->hw;
 239
 240         u32 head = IXGBE_READ_REG(hw, IXGBE_VFTDH(ring->reg_idx));
 241         u32 tail = IXGBE_READ_REG(hw, IXGBE_VFTDT(ring->reg_idx));
 242
 243         if (head != tail)
 244                 return (head < tail) ?
 245                         tail - head : (tail + ring->count - head);
 246
 247         return 0;
 248 }
 249
 250 static inline bool ixgbevf_check_tx_hang(struct ixgbevf_ring *tx_ring)
 251 {
 252         u32 tx_done = ixgbevf_get_tx_completed(tx_ring);
 253         u32 tx_done_old = tx_ring->tx_stats.tx_done_old;
 254         u32 tx_pending = ixgbevf_get_tx_pending(tx_ring);
 255
 256         clear_check_for_tx_hang(tx_ring);
 257
 258         /* Check for a hung queue, but be thorough. This verifies
 259          * that a transmit has been completed since the previous
 260          * check AND there is at least one packet pending. The
 261          * ARMED bit is set to indicate a potential hang.
 262          */
 263         if ((tx_done_old == tx_done) && tx_pending) {
 264                 /* make sure it is true for two checks in a row */
 265                 return test_and_set_bit(__IXGBEVF_HANG_CHECK_ARMED,
 266                                         &tx_ring->state);
 267         }
 268         /* reset the countdown */
 269         clear_bit(__IXGBEVF_HANG_CHECK_ARMED, &tx_ring->state);
 270
 271         /* update completed stats and continue */
 272         tx_ring->tx_stats.tx_done_old = tx_done;
 273
 274         return false;
 275 }
 276
 277 static void ixgbevf_tx_timeout_reset(struct ixgbevf_adapter *adapter)
 278 {
 279         /* Do the reset outside of interrupt context */
 280         if (!test_bit(__IXGBEVF_DOWN, &adapter->state)) {
 281                 set_bit(__IXGBEVF_RESET_REQUESTED, &adapter->state);
 282                 ixgbevf_service_event_schedule(adapter);
 283         }
 284 }
 285
 286 /**
 287  * ixgbevf_tx_timeout - Respond to a Tx Hang
 288  * @netdev: network interface device structure
 289  **/
 290 static void ixgbevf_tx_timeout(struct net_device *netdev)
 291 {
 292         struct ixgbevf_adapter *adapter = netdev_priv(netdev);
 293
 294         ixgbevf_tx_timeout_reset(adapter);
 295 }
 296
 297 /**
 298  * ixgbevf_clean_tx_irq - Reclaim resources after transmit completes
 299  * @q_vector: board private structure
 300  * @tx_ring: tx ring to clean
 301  * @napi_budget: Used to determine if we are in netpoll
 302  **/
 303 static bool ixgbevf_clean_tx_irq(struct ixgbevf_q_vector *q_vector,
 304                                  struct ixgbevf_ring *tx_ring, int napi_budget)
 305 {
 306         struct ixgbevf_adapter *adapter = q_vector->adapter;
 307         struct ixgbevf_tx_buffer *tx_buffer;
 308         union ixgbe_adv_tx_desc *tx_desc;
 309         unsigned int total_bytes = 0, total_packets = 0;
 310         unsigned int budget = tx_ring->count / 2;
 311         unsigned int i = tx_ring->next_to_clean;
 312
 313         if (test_bit(__IXGBEVF_DOWN, &adapter->state))
 314                 return true;
 315
 316         tx_buffer = &tx_ring->tx_buffer_info[i];
 317         tx_desc = IXGBEVF_TX_DESC(tx_ring, i);
 318         i -= tx_ring->count;
 319
 320         do {
 321                 union ixgbe_adv_tx_desc *eop_desc = tx_buffer->next_to_watch;
 322
 323                 /* if next_to_watch is not set then there is no work pending */
 324                 if (!eop_desc)
 325                         break;
 326
 327                 /* prevent any other reads prior to eop_desc */
 328                 read_barrier_depends();
 329
 330                 /* if DD is not set pending work has not been completed */
 331                 if (!(eop_desc->wb.status & cpu_to_le32(IXGBE_TXD_STAT_DD)))
 332                         break;
 333
 334                 /* clear next_to_watch to prevent false hangs */
 335                 tx_buffer->next_to_watch = NULL;
 336
 337                 /* update the statistics for this packet */
 338                 total_bytes += tx_buffer->bytecount;
 339                 total_packets += tx_buffer->gso_segs;
 340
 341                 /* free the skb */
 342                 napi_consume_skb(tx_buffer->skb, napi_budget);
 343
 344                 /* unmap skb header data */
 345                 dma_unmap_single(tx_ring->dev,
 346                                  dma_unmap_addr(tx_buffer, dma),
 347                                  dma_unmap_len(tx_buffer, len),
 348                                  DMA_TO_DEVICE);
 349
 350                 /* clear tx_buffer data */
 351                 tx_buffer->skb = NULL;
 352                 dma_unmap_len_set(tx_buffer, len, 0);
 353
 354                 /* unmap remaining buffers */
 355                 while (tx_desc != eop_desc) {
 356                         tx_buffer++;
 357                         tx_desc++;
 358                         i++;
 359                         if (unlikely(!i)) {
 360                                 i -= tx_ring->count;
 361                                 tx_buffer = tx_ring->tx_buffer_info;
 362                                 tx_desc = IXGBEVF_TX_DESC(tx_ring, 0);
 363                         }
 364
 365                         /* unmap any remaining paged data */
 366                         if (dma_unmap_len(tx_buffer, len)) {
 367                                 dma_unmap_page(tx_ring->dev,
 368                                                dma_unmap_addr(tx_buffer, dma),
 369                                                dma_unmap_len(tx_buffer, len),
 370                                                DMA_TO_DEVICE);
 371                                 dma_unmap_len_set(tx_buffer, len, 0);
 372                         }
 373                 }
 374
 375                 /* move us one more past the eop_desc for start of next pkt */
 376                 tx_buffer++;
 377                 tx_desc++;
 378                 i++;
 379                 if (unlikely(!i)) {
 380                         i -= tx_ring->count;
 381                         tx_buffer = tx_ring->tx_buffer_info;
 382                         tx_desc = IXGBEVF_TX_DESC(tx_ring, 0);
 383                 }
 384
 385                 /* issue prefetch for next Tx descriptor */
 386                 prefetch(tx_desc);
 387
 388                 /* update budget accounting */
 389                 budget--;
 390         } while (likely(budget));
 391
 392         i += tx_ring->count;
 393         tx_ring->next_to_clean = i;
 394         u64_stats_update_begin(&tx_ring->syncp);
 395         tx_ring->stats.bytes += total_bytes;
 396         tx_ring->stats.packets += total_packets;
 397         u64_stats_update_end(&tx_ring->syncp);
 398         q_vector->tx.total_bytes += total_bytes;
 399         q_vector->tx.total_packets += total_packets;
 400
 401         if (check_for_tx_hang(tx_ring) && ixgbevf_check_tx_hang(tx_ring)) {
 402                 struct ixgbe_hw *hw = &adapter->hw;
 403                 union ixgbe_adv_tx_desc *eop_desc;
 404
 405                 eop_desc = tx_ring->tx_buffer_info[i].next_to_watch;
 406
 407                 pr_err("Detected Tx Unit Hang\n"
 408                        "  Tx Queue             <%d>\n"
 409                        "  TDH, TDT             <%x>, <%x>\n"
 410                        "  next_to_use          <%x>\n"
 411                        "  next_to_clean        <%x>\n"
 412                        "tx_buffer_info[next_to_clean]\n"
 413                        "  next_to_watch        <%p>\n"
 414                        "  eop_desc->wb.status  <%x>\n"
 415                        "  time_stamp           <%lx>\n"
 416                        "  jiffies              <%lx>\n",
 417                        tx_ring->queue_index,
 418                        IXGBE_READ_REG(hw, IXGBE_VFTDH(tx_ring->reg_idx)),
 419                        IXGBE_READ_REG(hw, IXGBE_VFTDT(tx_ring->reg_idx)),
 420                        tx_ring->next_to_use, i,
 421                        eop_desc, (eop_desc ? eop_desc->wb.status : 0),
 422                        tx_ring->tx_buffer_info[i].time_stamp, jiffies);
 423
 424                 netif_stop_subqueue(tx_ring->netdev, tx_ring->queue_index);
 425
 426                 /* schedule immediate reset if we believe we hung */
 427                 ixgbevf_tx_timeout_reset(adapter);
 428
 429                 return true;
 430         }
 431
 432 #define TX_WAKE_THRESHOLD (DESC_NEEDED * 2)
 433         if (unlikely(total_packets && netif_carrier_ok(tx_ring->netdev) &&
 434                      (ixgbevf_desc_unused(tx_ring) >= TX_WAKE_THRESHOLD))) {
 435                 /* Make sure that anybody stopping the queue after this
 436                  * sees the new next_to_clean.
 437                  */
 438                 smp_mb();
 439
 440                 if (__netif_subqueue_stopped(tx_ring->netdev,
 441                                              tx_ring->queue_index) &&
 442                     !test_bit(__IXGBEVF_DOWN, &adapter->state)) {
 443                         netif_wake_subqueue(tx_ring->netdev,
 444                                             tx_ring->queue_index);
 445                         ++tx_ring->tx_stats.restart_queue;
 446                 }
 447         }
 448
 449         return !!budget;
 450 }
 451
 452 /**
 453  * ixgbevf_rx_skb - Helper function to determine proper Rx method
 454  * @q_vector: structure containing interrupt and ring information
 455  * @skb: packet to send up
 456  **/
 457 static void ixgbevf_rx_skb(struct ixgbevf_q_vector *q_vector,
 458                            struct sk_buff *skb)
 459 {
 460 #ifdef CONFIG_NET_RX_BUSY_POLL
 461         skb_mark_napi_id(skb, &q_vector->napi);
 462
 463         if (ixgbevf_qv_busy_polling(q_vector)) {
 464                 netif_receive_skb(skb);
 465                 /* exit early if we busy polled */
 466                 return;
 467         }
 468 #endif /* CONFIG_NET_RX_BUSY_POLL */
 469
 470         napi_gro_receive(&q_vector->napi, skb);
 471 }
 472
 473 #define IXGBE_RSS_L4_TYPES_MASK \
 474         ((1ul << IXGBE_RXDADV_RSSTYPE_IPV4_TCP) | \
 475          (1ul << IXGBE_RXDADV_RSSTYPE_IPV4_UDP) | \
 476          (1ul << IXGBE_RXDADV_RSSTYPE_IPV6_TCP) | \
 477          (1ul << IXGBE_RXDADV_RSSTYPE_IPV6_UDP))
 478
 479 static inline void ixgbevf_rx_hash(struct ixgbevf_ring *ring,
 480                                    union ixgbe_adv_rx_desc *rx_desc,
 481                                    struct sk_buff *skb)
 482 {
 483         u16 rss_type;
 484
 485         if (!(ring->netdev->features & NETIF_F_RXHASH))
 486                 return;
 487
 488         rss_type = le16_to_cpu(rx_desc->wb.lower.lo_dword.hs_rss.pkt_info) &
 489                    IXGBE_RXDADV_RSSTYPE_MASK;
 490
 491         if (!rss_type)
 492                 return;
 493
 494         skb_set_hash(skb, le32_to_cpu(rx_desc->wb.lower.hi_dword.rss),
 495                      (IXGBE_RSS_L4_TYPES_MASK & (1ul << rss_type)) ?
 496                      PKT_HASH_TYPE_L4 : PKT_HASH_TYPE_L3);
 497 }
 498
 499 /**
 500  * ixgbevf_rx_checksum - indicate in skb if hw indicated a good cksum
 501  * @ring: structure containig ring specific data
 502  * @rx_desc: current Rx descriptor being processed
 503  * @skb: skb currently being received and modified
 504  **/
 505 static inline void ixgbevf_rx_checksum(struct ixgbevf_ring *ring,
 506                                        union ixgbe_adv_rx_desc *rx_desc,
 507                                        struct sk_buff *skb)
 508 {
 509         skb_checksum_none_assert(skb);
 510
 511         /* Rx csum disabled */
 512         if (!(ring->netdev->features & NETIF_F_RXCSUM))
 513                 return;
 514
 515         /* if IP and error */
 516         if (ixgbevf_test_staterr(rx_desc, IXGBE_RXD_STAT_IPCS) &&
 517             ixgbevf_test_staterr(rx_desc, IXGBE_RXDADV_ERR_IPE)) {
 518                 ring->rx_stats.csum_err++;
 519                 return;
 520         }
 521
 522         if (!ixgbevf_test_staterr(rx_desc, IXGBE_RXD_STAT_L4CS))
 523                 return;
 524
 525         if (ixgbevf_test_staterr(rx_desc, IXGBE_RXDADV_ERR_TCPE)) {
 526                 ring->rx_stats.csum_err++;
 527                 return;
 528         }
 529
 530         /* It must be a TCP or UDP packet with a valid checksum */
 531         skb->ip_summed = CHECKSUM_UNNECESSARY;
 532 }
 533
 534 /**
 535  * ixgbevf_process_skb_fields - Populate skb header fields from Rx descriptor
 536  * @rx_ring: rx descriptor ring packet is being transacted on
 537  * @rx_desc: pointer to the EOP Rx descriptor
 538  * @skb: pointer to current skb being populated
 539  *
 540  * This function checks the ring, descriptor, and packet information in
 541  * order to populate the checksum, VLAN, protocol, and other fields within
 542  * the skb.
 543  **/
 544 static void ixgbevf_process_skb_fields(struct ixgbevf_ring *rx_ring,
 545                                        union ixgbe_adv_rx_desc *rx_desc,
 546                                        struct sk_buff *skb)
 547 {
 548         ixgbevf_rx_hash(rx_ring, rx_desc, skb);
 549         ixgbevf_rx_checksum(rx_ring, rx_desc, skb);
 550
 551         if (ixgbevf_test_staterr(rx_desc, IXGBE_RXD_STAT_VP)) {
 552                 u16 vid = le16_to_cpu(rx_desc->wb.upper.vlan);
 553                 unsigned long *active_vlans = netdev_priv(rx_ring->netdev);
 554
 555                 if (test_bit(vid & VLAN_VID_MASK, active_vlans))
 556                         __vlan_hwaccel_put_tag(skb, htons(ETH_P_8021Q), vid);
 557         }
 558
 559         skb->protocol = eth_type_trans(skb, rx_ring->netdev);
 560 }
 561
 562 /**
 563  * ixgbevf_is_non_eop - process handling of non-EOP buffers
 564  * @rx_ring: Rx ring being processed
 565  * @rx_desc: Rx descriptor for current buffer
 566  * @skb: current socket buffer containing buffer in progress
 567  *
 568  * This function updates next to clean.  If the buffer is an EOP buffer
 569  * this function exits returning false, otherwise it will place the
 570  * sk_buff in the next buffer to be chained and return true indicating
 571  * that this is in fact a non-EOP buffer.
 572  **/
 573 static bool ixgbevf_is_non_eop(struct ixgbevf_ring *rx_ring,
 574                                union ixgbe_adv_rx_desc *rx_desc)
 575 {
 576         u32 ntc = rx_ring->next_to_clean + 1;
 577
 578         /* fetch, update, and store next to clean */
 579         ntc = (ntc < rx_ring->count) ? ntc : 0;
 580         rx_ring->next_to_clean = ntc;
 581
 582         prefetch(IXGBEVF_RX_DESC(rx_ring, ntc));
 583
 584         if (likely(ixgbevf_test_staterr(rx_desc, IXGBE_RXD_STAT_EOP)))
 585                 return false;
 586
 587         return true;
 588 }
 589
 590 static bool ixgbevf_alloc_mapped_page(struct ixgbevf_ring *rx_ring,
 591                                       struct ixgbevf_rx_buffer *bi)
 592 {
 593         struct page *page = bi->page;
 594         dma_addr_t dma = bi->dma;
 595
 596         /* since we are recycling buffers we should seldom need to alloc */
 597         if (likely(page))
 598                 return true;
 599
 600         /* alloc new page for storage */
 601         page = dev_alloc_page();
 602         if (unlikely(!page)) {
 603                 rx_ring->rx_stats.alloc_rx_page_failed++;
 604                 return false;
 605         }
 606
 607         /* map page for use */
 608         dma = dma_map_page(rx_ring->dev, page, 0,
 609                            PAGE_SIZE, DMA_FROM_DEVICE);
 610
 611         /* if mapping failed free memory back to system since
 612          * there isn't much point in holding memory we can't use
 613          */
 614         if (dma_mapping_error(rx_ring->dev, dma)) {
 615                 __free_page(page);
 616
 617                 rx_ring->rx_stats.alloc_rx_buff_failed++;
 618                 return false;
 619         }
 620
 621         bi->dma = dma;
 622         bi->page = page;
 623         bi->page_offset = 0;
 624
 625         return true;
 626 }
 627
 628 /**
 629  * ixgbevf_alloc_rx_buffers - Replace used receive buffers; packet split
 630  * @rx_ring: rx descriptor ring (for a specific queue) to setup buffers on
 631  * @cleaned_count: number of buffers to replace
 632  **/
 633 static void ixgbevf_alloc_rx_buffers(struct ixgbevf_ring *rx_ring,
 634                                      u16 cleaned_count)
 635 {
 636         union ixgbe_adv_rx_desc *rx_desc;
 637         struct ixgbevf_rx_buffer *bi;
 638         unsigned int i = rx_ring->next_to_use;
 639
 640         /* nothing to do or no valid netdev defined */
 641         if (!cleaned_count || !rx_ring->netdev)
 642                 return;
 643
 644         rx_desc = IXGBEVF_RX_DESC(rx_ring, i);
 645         bi = &rx_ring->rx_buffer_info[i];
 646         i -= rx_ring->count;
 647
 648         do {
 649                 if (!ixgbevf_alloc_mapped_page(rx_ring, bi))
 650                         break;
 651
 652                 /* Refresh the desc even if pkt_addr didn't change
 653                  * because each write-back erases this info.
 654                  */
 655                 rx_desc->read.pkt_addr = cpu_to_le64(bi->dma + bi->page_offset);
 656
 657                 rx_desc++;
 658                 bi++;
 659                 i++;
 660                 if (unlikely(!i)) {
 661                         rx_desc = IXGBEVF_RX_DESC(rx_ring, 0);
 662                         bi = rx_ring->rx_buffer_info;
 663                         i -= rx_ring->count;
 664                 }
 665
 666                 /* clear the hdr_addr for the next_to_use descriptor */
 667                 rx_desc->read.hdr_addr = 0;
 668
 669                 cleaned_count--;
 670         } while (cleaned_count);
 671
 672         i += rx_ring->count;
 673
 674         if (rx_ring->next_to_use != i) {
 675                 /* record the next descriptor to use */
 676                 rx_ring->next_to_use = i;
 677
 678                 /* update next to alloc since we have filled the ring */
 679                 rx_ring->next_to_alloc = i;
 680
 681                 /* Force memory writes to complete before letting h/w
 682                  * know there are new descriptors to fetch.  (Only
 683                  * applicable for weak-ordered memory model archs,
 684                  * such as IA-64).
 685                  */
 686                 wmb();
 687                 ixgbevf_write_tail(rx_ring, i);
 688         }
 689 }
 690
 691 /**
 692  * ixgbevf_cleanup_headers - Correct corrupted or empty headers
 693  * @rx_ring: rx descriptor ring packet is being transacted on
 694  * @rx_desc: pointer to the EOP Rx descriptor
 695  * @skb: pointer to current skb being fixed
 696  *
 697  * Check for corrupted packet headers caused by senders on the local L2
 698  * embedded NIC switch not setting up their Tx Descriptors right.  These
 699  * should be very rare.
 700  *
 701  * Also address the case where we are pulling data in on pages only
 702  * and as such no data is present in the skb header.
 703  *
 704  * In addition if skb is not at least 60 bytes we need to pad it so that
 705  * it is large enough to qualify as a valid Ethernet frame.
 706  *
 707  * Returns true if an error was encountered and skb was freed.
 708  **/
 709 static bool ixgbevf_cleanup_headers(struct ixgbevf_ring *rx_ring,
 710                                     union ixgbe_adv_rx_desc *rx_desc,
 711                                     struct sk_buff *skb)
 712 {
 713         /* verify that the packet does not have any known errors */
 714         if (unlikely(ixgbevf_test_staterr(rx_desc,
 715                                           IXGBE_RXDADV_ERR_FRAME_ERR_MASK))) {
 716                 struct net_device *netdev = rx_ring->netdev;
 717
 718                 if (!(netdev->features & NETIF_F_RXALL)) {
 719                         dev_kfree_skb_any(skb);
 720                         return true;
 721                 }
 722         }
 723
 724         /* if eth_skb_pad returns an error the skb was freed */
 725         if (eth_skb_pad(skb))
 726                 return true;
 727
 728         return false;
 729 }
 730
 731 /**
 732  * ixgbevf_reuse_rx_page - page flip buffer and store it back on the ring
 733  * @rx_ring: rx descriptor ring to store buffers on
 734  * @old_buff: donor buffer to have page reused
 735  *
 736  * Synchronizes page for reuse by the adapter
 737  **/
 738 static void ixgbevf_reuse_rx_page(struct ixgbevf_ring *rx_ring,
 739                                   struct ixgbevf_rx_buffer *old_buff)
 740 {
 741         struct ixgbevf_rx_buffer *new_buff;
 742         u16 nta = rx_ring->next_to_alloc;
 743
 744         new_buff = &rx_ring->rx_buffer_info[nta];
 745
 746         /* update, and store next to alloc */
 747         nta++;
 748         rx_ring->next_to_alloc = (nta < rx_ring->count) ? nta : 0;
 749
 750         /* transfer page from old buffer to new buffer */
 751         new_buff->page = old_buff->page;
 752         new_buff->dma = old_buff->dma;
 753         new_buff->page_offset = old_buff->page_offset;
 754
 755         /* sync the buffer for use by the device */
 756         dma_sync_single_range_for_device(rx_ring->dev, new_buff->dma,
 757                                          new_buff->page_offset,
 758                                          IXGBEVF_RX_BUFSZ,
 759                                          DMA_FROM_DEVICE);
 760 }
 761
 762 static inline bool ixgbevf_page_is_reserved(struct page *page)
 763 {
 764         return (page_to_nid(page) != numa_mem_id()) || page_is_pfmemalloc(page);
 765 }
 766
 767 /**
 768  * ixgbevf_add_rx_frag - Add contents of Rx buffer to sk_buff
 769  * @rx_ring: rx descriptor ring to transact packets on
 770  * @rx_buffer: buffer containing page to add
 771  * @rx_desc: descriptor containing length of buffer written by hardware
 772  * @skb: sk_buff to place the data into
 773  *
 774  * This function will add the data contained in rx_buffer->page to the skb.
 775  * This is done either through a direct copy if the data in the buffer is
 776  * less than the skb header size, otherwise it will just attach the page as
 777  * a frag to the skb.
 778  *
 779  * The function will then update the page offset if necessary and return
 780  * true if the buffer can be reused by the adapter.
 781  **/
 782 static bool ixgbevf_add_rx_frag(struct ixgbevf_ring *rx_ring,
 783                                 struct ixgbevf_rx_buffer *rx_buffer,
 784                                 union ixgbe_adv_rx_desc *rx_desc,
 785                                 struct sk_buff *skb)
 786 {
 787         struct page *page = rx_buffer->page;
 788         unsigned char *va = page_address(page) + rx_buffer->page_offset;
 789         unsigned int size = le16_to_cpu(rx_desc->wb.upper.length);
 790 #if (PAGE_SIZE < 8192)
 791         unsigned int truesize = IXGBEVF_RX_BUFSZ;
 792 #else
 793         unsigned int truesize = ALIGN(size, L1_CACHE_BYTES);
 794 #endif
 795         unsigned int pull_len;
 796
 797         if (unlikely(skb_is_nonlinear(skb)))
 798                 goto add_tail_frag;
 799
 800         if (likely(size <= IXGBEVF_RX_HDR_SIZE)) {
 801                 memcpy(__skb_put(skb, size), va, ALIGN(size, sizeof(long)));
 802
 803                 /* page is not reserved, we can reuse buffer as is */
 804                 if (likely(!ixgbevf_page_is_reserved(page)))
 805                         return true;
 806
 807                 /* this page cannot be reused so discard it */
 808                 put_page(page);
 809                 return false;
 810         }
 811
 812         /* we need the header to contain the greater of either ETH_HLEN or
 813          * 60 bytes if the skb->len is less than 60 for skb_pad.
 814          */
 815         pull_len = eth_get_headlen(va, IXGBEVF_RX_HDR_SIZE);
 816
 817         /* align pull length to size of long to optimize memcpy performance */
 818         memcpy(__skb_put(skb, pull_len), va, ALIGN(pull_len, sizeof(long)));
 819
 820         /* update all of the pointers */
 821         va += pull_len;
 822         size -= pull_len;
 823
 824 add_tail_frag:
 825         skb_add_rx_frag(skb, skb_shinfo(skb)->nr_frags, page,
 826                         (unsigned long)va & ~PAGE_MASK, size, truesize);
 827
 828         /* avoid re-using remote pages */
 829         if (unlikely(ixgbevf_page_is_reserved(page)))
 830                 return false;
 831
 832 #if (PAGE_SIZE < 8192)
 833         /* if we are only owner of page we can reuse it */
 834         if (unlikely(page_count(page) != 1))
 835                 return false;
 836
 837         /* flip page offset to other buffer */
 838         rx_buffer->page_offset ^= IXGBEVF_RX_BUFSZ;
 839
 840 #else
 841         /* move offset up to the next cache line */
 842         rx_buffer->page_offset += truesize;
 843
 844         if (rx_buffer->page_offset > (PAGE_SIZE - IXGBEVF_RX_BUFSZ))
 845                 return false;
 846
 847 #endif
 848         /* Even if we own the page, we are not allowed to use atomic_set()
 849          * This would break get_page_unless_zero() users.
 850          */
 851         page_ref_inc(page);
 852
 853         return true;
 854 }
 855
 856 static struct sk_buff *ixgbevf_fetch_rx_buffer(struct ixgbevf_ring *rx_ring,
 857                                                union ixgbe_adv_rx_desc *rx_desc,
 858                                                struct sk_buff *skb)
 859 {
 860         struct ixgbevf_rx_buffer *rx_buffer;
 861         struct page *page;
 862
 863         rx_buffer = &rx_ring->rx_buffer_info[rx_ring->next_to_clean];
 864         page = rx_buffer->page;
 865         prefetchw(page);
 866
 867         if (likely(!skb)) {
 868                 void *page_addr = page_address(page) +
 869                                   rx_buffer->page_offset;
 870
 871                 /* prefetch first cache line of first page */
 872                 prefetch(page_addr);
 873 #if L1_CACHE_BYTES < 128
 874                 prefetch(page_addr + L1_CACHE_BYTES);
 875 #endif
 876
 877                 /* allocate a skb to store the frags */
 878                 skb = netdev_alloc_skb_ip_align(rx_ring->netdev,
 879                                                 IXGBEVF_RX_HDR_SIZE);
 880                 if (unlikely(!skb)) {
 881                         rx_ring->rx_stats.alloc_rx_buff_failed++;
 882                         return NULL;
 883                 }
 884
 885                 /* we will be copying header into skb->data in
 886                  * pskb_may_pull so it is in our interest to prefetch
 887                  * it now to avoid a possible cache miss
 888                  */
 889                 prefetchw(skb->data);
 890         }
 891
 892         /* we are reusing so sync this buffer for CPU use */
 893         dma_sync_single_range_for_cpu(rx_ring->dev,
 894                                       rx_buffer->dma,
 895                                       rx_buffer->page_offset,
 896                                       IXGBEVF_RX_BUFSZ,
 897                                       DMA_FROM_DEVICE);
 898
 899         /* pull page into skb */
 900         if (ixgbevf_add_rx_frag(rx_ring, rx_buffer, rx_desc, skb)) {
 901                 /* hand second half of page back to the ring */
 902                 ixgbevf_reuse_rx_page(rx_ring, rx_buffer);
 903         } else {
 904                 /* we are not reusing the buffer so unmap it */
 905                 dma_unmap_page(rx_ring->dev, rx_buffer->dma,
 906                                PAGE_SIZE, DMA_FROM_DEVICE);
 907         }
 908
 909         /* clear contents of buffer_info */
 910         rx_buffer->dma = 0;
 911         rx_buffer->page = NULL;
 912
 913         return skb;
 914 }
 915
 916 static inline void ixgbevf_irq_enable_queues(struct ixgbevf_adapter *adapter,
 917                                              u32 qmask)
 918 {
 919         struct ixgbe_hw *hw = &adapter->hw;
 920
 921         IXGBE_WRITE_REG(hw, IXGBE_VTEIMS, qmask);
 922 }
 923
 924 static int ixgbevf_clean_rx_irq(struct ixgbevf_q_vector *q_vector,
 925                                 struct ixgbevf_ring *rx_ring,
 926                                 int budget)
 927 {
 928         unsigned int total_rx_bytes = 0, total_rx_packets = 0;
 929         u16 cleaned_count = ixgbevf_desc_unused(rx_ring);
 930         struct sk_buff *skb = rx_ring->skb;
 931
 932         while (likely(total_rx_packets < budget)) {
 933                 union ixgbe_adv_rx_desc *rx_desc;
 934
 935                 /* return some buffers to hardware, one at a time is too slow */
 936                 if (cleaned_count >= IXGBEVF_RX_BUFFER_WRITE) {
 937                         ixgbevf_alloc_rx_buffers(rx_ring, cleaned_count);
 938                         cleaned_count = 0;
 939                 }
 940
 941                 rx_desc = IXGBEVF_RX_DESC(rx_ring, rx_ring->next_to_clean);
 942
 943                 if (!ixgbevf_test_staterr(rx_desc, IXGBE_RXD_STAT_DD))
 944                         break;
 945
 946                 /* This memory barrier is needed to keep us from reading
 947                  * any other fields out of the rx_desc until we know the
 948                  * RXD_STAT_DD bit is set
 949                  */
 950                 rmb();
 951
 952                 /* retrieve a buffer from the ring */
 953                 skb = ixgbevf_fetch_rx_buffer(rx_ring, rx_desc, skb);
 954
 955                 /* exit if we failed to retrieve a buffer */
 956                 if (!skb)
 957                         break;
 958
 959                 cleaned_count++;
 960
 961                 /* fetch next buffer in frame if non-eop */
 962                 if (ixgbevf_is_non_eop(rx_ring, rx_desc))
 963                         continue;
 964
 965                 /* verify the packet layout is correct */
 966                 if (ixgbevf_cleanup_headers(rx_ring, rx_desc, skb)) {
 967                         skb = NULL;
 968                         continue;
 969                 }
 970
 971                 /* probably a little skewed due to removing CRC */
 972                 total_rx_bytes += skb->len;
 973
 974                 /* Workaround hardware that can't do proper VEPA multicast
 975                  * source pruning.
 976                  */
 977                 if ((skb->pkt_type == PACKET_BROADCAST ||
 978                      skb->pkt_type == PACKET_MULTICAST) &&
 979                     ether_addr_equal(rx_ring->netdev->dev_addr,
 980                                      eth_hdr(skb)->h_source)) {
 981                         dev_kfree_skb_irq(skb);
 982                         continue;
 983                 }
 984
 985                 /* populate checksum, VLAN, and protocol */
 986                 ixgbevf_process_skb_fields(rx_ring, rx_desc, skb);
 987
 988                 ixgbevf_rx_skb(q_vector, skb);
 989
 990                 /* reset skb pointer */
 991                 skb = NULL;
 992
 993                 /* update budget accounting */
 994                 total_rx_packets++;
 995         }
 996
 997         /* place incomplete frames back on ring for completion */
 998         rx_ring->skb = skb;
 999
1000         u64_stats_update_begin(&rx_ring->syncp);
1001         rx_ring->stats.packets += total_rx_packets;
1002         rx_ring->stats.bytes += total_rx_bytes;
1003         u64_stats_update_end(&rx_ring->syncp);
1004         q_vector->rx.total_packets += total_rx_packets;
1005         q_vector->rx.total_bytes += total_rx_bytes;
1006
1007         return total_rx_packets;
1008 }
1009
1010 /**
1011  * ixgbevf_poll - NAPI polling calback
1012  * @napi: napi struct with our devices info in it
1013  * @budget: amount of work driver is allowed to do this pass, in packets
1014  *
1015  * This function will clean more than one or more rings associated with a
1016  * q_vector.
1017  **/
1018 static int ixgbevf_poll(struct napi_struct *napi, int budget)
1019 {
1020         struct ixgbevf_q_vector *q_vector =
1021                 container_of(napi, struct ixgbevf_q_vector, napi);
1022         struct ixgbevf_adapter *adapter = q_vector->adapter;
1023         struct ixgbevf_ring *ring;
1024         int per_ring_budget, work_done = 0;
1025         bool clean_complete = true;
1026
1027         ixgbevf_for_each_ring(ring, q_vector->tx) {
1028                 if (!ixgbevf_clean_tx_irq(q_vector, ring, budget))
1029                         clean_complete = false;
1030         }
1031
1032         if (budget <= 0)
1033                 return budget;
1034 #ifdef CONFIG_NET_RX_BUSY_POLL
1035         if (!ixgbevf_qv_lock_napi(q_vector))
1036                 return budget;
1037 #endif
1038
1039         /* attempt to distribute budget to each queue fairly, but don't allow
1040          * the budget to go below 1 because we'll exit polling
1041          */
1042         if (q_vector->rx.count > 1)
1043                 per_ring_budget = max(budget/q_vector->rx.count, 1);
1044         else
1045                 per_ring_budget = budget;
1046
1047         ixgbevf_for_each_ring(ring, q_vector->rx) {
1048                 int cleaned = ixgbevf_clean_rx_irq(q_vector, ring,
1049                                                    per_ring_budget);
1050                 work_done += cleaned;
1051                 if (cleaned >= per_ring_budget)
1052                         clean_complete = false;
1053         }
1054
1055 #ifdef CONFIG_NET_RX_BUSY_POLL
1056         ixgbevf_qv_unlock_napi(q_vector);
1057 #endif
1058
1059         /* If all work not completed, return budget and keep polling */
1060         if (!clean_complete)
1061                 return budget;
1062         /* all work done, exit the polling mode */
1063         napi_complete_done(napi, work_done);
1064         if (adapter->rx_itr_setting == 1)
1065                 ixgbevf_set_itr(q_vector);
1066         if (!test_bit(__IXGBEVF_DOWN, &adapter->state) &&
1067             !test_bit(__IXGBEVF_REMOVING, &adapter->state))
1068                 ixgbevf_irq_enable_queues(adapter,
1069                                           BIT(q_vector->v_idx));
1070
1071         return 0;
1072 }
1073
1074 /**
1075  * ixgbevf_write_eitr - write VTEITR register in hardware specific way
1076  * @q_vector: structure containing interrupt and ring information
1077  **/
1078 void ixgbevf_write_eitr(struct ixgbevf_q_vector *q_vector)
1079 {
1080         struct ixgbevf_adapter *adapter = q_vector->adapter;
1081         struct ixgbe_hw *hw = &adapter->hw;
1082         int v_idx = q_vector->v_idx;
1083         u32 itr_reg = q_vector->itr & IXGBE_MAX_EITR;
1084
1085         /* set the WDIS bit to not clear the timer bits and cause an
1086          * immediate assertion of the interrupt
1087          */
1088         itr_reg |= IXGBE_EITR_CNT_WDIS;
1089
1090         IXGBE_WRITE_REG(hw, IXGBE_VTEITR(v_idx), itr_reg);
1091 }
1092
1093 #ifdef CONFIG_NET_RX_BUSY_POLL
1094 /* must be called with local_bh_disable()d */
1095 static int ixgbevf_busy_poll_recv(struct napi_struct *napi)
1096 {
1097         struct ixgbevf_q_vector *q_vector =
1098                         container_of(napi, struct ixgbevf_q_vector, napi);
1099         struct ixgbevf_adapter *adapter = q_vector->adapter;
1100         struct ixgbevf_ring  *ring;
1101         int found = 0;
1102
1103         if (test_bit(__IXGBEVF_DOWN, &adapter->state))
1104                 return LL_FLUSH_FAILED;
1105
1106         if (!ixgbevf_qv_lock_poll(q_vector))
1107                 return LL_FLUSH_BUSY;
1108
1109         ixgbevf_for_each_ring(ring, q_vector->rx) {
1110                 found = ixgbevf_clean_rx_irq(q_vector, ring, 4);
1111 #ifdef BP_EXTENDED_STATS
1112                 if (found)
1113                         ring->stats.cleaned += found;
1114                 else
1115                         ring->stats.misses++;
1116 #endif
1117                 if (found)
1118                         break;
1119         }
1120
1121         ixgbevf_qv_unlock_poll(q_vector);
1122
1123         return found;
1124 }
1125 #endif /* CONFIG_NET_RX_BUSY_POLL */
1126
1127 /**
1128  * ixgbevf_configure_msix - Configure MSI-X hardware
1129  * @adapter: board private structure
1130  *
1131  * ixgbevf_configure_msix sets up the hardware to properly generate MSI-X
1132  * interrupts.
1133  **/
1134 static void ixgbevf_configure_msix(struct ixgbevf_adapter *adapter)
1135 {
1136         struct ixgbevf_q_vector *q_vector;
1137         int q_vectors, v_idx;
1138
1139         q_vectors = adapter->num_msix_vectors - NON_Q_VECTORS;
1140         adapter->eims_enable_mask = 0;
1141
1142         /* Populate the IVAR table and set the ITR values to the
1143          * corresponding register.
1144          */
1145         for (v_idx = 0; v_idx < q_vectors; v_idx++) {
1146                 struct ixgbevf_ring *ring;
1147
1148                 q_vector = adapter->q_vector[v_idx];
1149
1150                 ixgbevf_for_each_ring(ring, q_vector->rx)
1151                         ixgbevf_set_ivar(adapter, 0, ring->reg_idx, v_idx);
1152
1153                 ixgbevf_for_each_ring(ring, q_vector->tx)
1154                         ixgbevf_set_ivar(adapter, 1, ring->reg_idx, v_idx);
1155
1156                 if (q_vector->tx.ring && !q_vector->rx.ring) {
1157                         /* Tx only vector */
1158                         if (adapter->tx_itr_setting == 1)
1159                                 q_vector->itr = IXGBE_12K_ITR;
1160                         else
1161                                 q_vector->itr = adapter->tx_itr_setting;
1162                 } else {
1163                         /* Rx or Rx/Tx vector */
1164                         if (adapter->rx_itr_setting == 1)
1165                                 q_vector->itr = IXGBE_20K_ITR;
1166                         else
1167                                 q_vector->itr = adapter->rx_itr_setting;
1168                 }
1169
1170                 /* add q_vector eims value to global eims_enable_mask */
1171                 adapter->eims_enable_mask |= BIT(v_idx);
1172
1173                 ixgbevf_write_eitr(q_vector);
1174         }
1175
1176         ixgbevf_set_ivar(adapter, -1, 1, v_idx);
1177         /* setup eims_other and add value to global eims_enable_mask */
1178         adapter->eims_other = BIT(v_idx);
1179         adapter->eims_enable_mask |= adapter->eims_other;
1180 }
1181
1182 enum latency_range {
1183         lowest_latency = 0,
1184         low_latency = 1,
1185         bulk_latency = 2,
1186         latency_invalid = 255
1187 };
1188
1189 /**
1190  * ixgbevf_update_itr - update the dynamic ITR value based on statistics
1191  * @q_vector: structure containing interrupt and ring information
1192  * @ring_container: structure containing ring performance data
1193  *
1194  * Stores a new ITR value based on packets and byte
1195  * counts during the last interrupt.  The advantage of per interrupt
1196  * computation is faster updates and more accurate ITR for the current
1197  * traffic pattern.  Constants in this function were computed
1198  * based on theoretical maximum wire speed and thresholds were set based
1199  * on testing data as well as attempting to minimize response time
1200  * while increasing bulk throughput.
1201  **/
1202 static void ixgbevf_update_itr(struct ixgbevf_q_vector *q_vector,
1203                                struct ixgbevf_ring_container *ring_container)
1204 {
1205         int bytes = ring_container->total_bytes;
1206         int packets = ring_container->total_packets;
1207         u32 timepassed_us;
1208         u64 bytes_perint;
1209         u8 itr_setting = ring_container->itr;
1210
1211         if (packets == 0)
1212                 return;
1213
1214         /* simple throttle rate management
1215          *    0-20MB/s lowest (100000 ints/s)
1216          *   20-100MB/s low   (20000 ints/s)
1217          *  100-1249MB/s bulk (12000 ints/s)
1218          */
1219         /* what was last interrupt timeslice? */
1220         timepassed_us = q_vector->itr >> 2;
1221         bytes_perint = bytes / timepassed_us; /* bytes/usec */
1222
1223         switch (itr_setting) {
1224         case lowest_latency:
1225                 if (bytes_perint > 10)
1226                         itr_setting = low_latency;
1227                 break;
1228         case low_latency:
1229                 if (bytes_perint > 20)
1230                         itr_setting = bulk_latency;
1231                 else if (bytes_perint <= 10)
1232                         itr_setting = lowest_latency;
1233                 break;
1234         case bulk_latency:
1235                 if (bytes_perint <= 20)
1236                         itr_setting = low_latency;
1237                 break;
1238         }
1239
1240         /* clear work counters since we have the values we need */
1241         ring_container->total_bytes = 0;
1242         ring_container->total_packets = 0;
1243
1244         /* write updated itr to ring container */
1245         ring_container->itr = itr_setting;
1246 }
1247
1248 static void ixgbevf_set_itr(struct ixgbevf_q_vector *q_vector)
1249 {
1250         u32 new_itr = q_vector->itr;
1251         u8 current_itr;
1252
1253         ixgbevf_update_itr(q_vector, &q_vector->tx);
1254         ixgbevf_update_itr(q_vector, &q_vector->rx);
1255
1256         current_itr = max(q_vector->rx.itr, q_vector->tx.itr);
1257
1258         switch (current_itr) {
1259         /* counts and packets in update_itr are dependent on these numbers */
1260         case lowest_latency:
1261                 new_itr = IXGBE_100K_ITR;
1262                 break;
1263         case low_latency:
1264                 new_itr = IXGBE_20K_ITR;
1265                 break;
1266         case bulk_latency:
1267                 new_itr = IXGBE_12K_ITR;
1268                 break;
1269         default:
1270                 break;
1271         }
1272
1273         if (new_itr != q_vector->itr) {
1274                 /* do an exponential smoothing */
1275                 new_itr = (10 * new_itr * q_vector->itr) /
1276                           ((9 * new_itr) + q_vector->itr);
1277
1278                 /* save the algorithm value here */
1279                 q_vector->itr = new_itr;
1280
1281                 ixgbevf_write_eitr(q_vector);
1282         }
1283 }
1284
1285 static irqreturn_t ixgbevf_msix_other(int irq, void *data)
1286 {
1287         struct ixgbevf_adapter *adapter = data;
1288         struct ixgbe_hw *hw = &adapter->hw;
1289
1290         hw->mac.get_link_status = 1;
1291
1292         ixgbevf_service_event_schedule(adapter);
1293
1294         IXGBE_WRITE_REG(hw, IXGBE_VTEIMS, adapter->eims_other);
1295
1296         return IRQ_HANDLED;
1297 }
1298
1299 /**
1300  * ixgbevf_msix_clean_rings - single unshared vector rx clean (all queues)
1301  * @irq: unused
1302  * @data: pointer to our q_vector struct for this interrupt vector
1303  **/
1304 static irqreturn_t ixgbevf_msix_clean_rings(int irq, void *data)
1305 {
1306         struct ixgbevf_q_vector *q_vector = data;
1307
1308         /* EIAM disabled interrupts (on this vector) for us */
1309         if (q_vector->rx.ring || q_vector->tx.ring)
1310                 napi_schedule_irqoff(&q_vector->napi);
1311
1312         return IRQ_HANDLED;
1313 }
1314
1315 static inline void map_vector_to_rxq(struct ixgbevf_adapter *a, int v_idx,
1316                                      int r_idx)
1317 {
1318         struct ixgbevf_q_vector *q_vector = a->q_vector[v_idx];
1319
1320         a->rx_ring[r_idx]->next = q_vector->rx.ring;
1321         q_vector->rx.ring = a->rx_ring[r_idx];
1322         q_vector->rx.count++;
1323 }
1324
1325 static inline void map_vector_to_txq(struct ixgbevf_adapter *a, int v_idx,
1326                                      int t_idx)
1327 {
1328         struct ixgbevf_q_vector *q_vector = a->q_vector[v_idx];
1329
1330         a->tx_ring[t_idx]->next = q_vector->tx.ring;
1331         q_vector->tx.ring = a->tx_ring[t_idx];
1332         q_vector->tx.count++;
1333 }
1334
1335 /**
1336  * ixgbevf_map_rings_to_vectors - Maps descriptor rings to vectors
1337  * @adapter: board private structure to initialize
1338  *
1339  * This function maps descriptor rings to the queue-specific vectors
1340  * we were allotted through the MSI-X enabling code.  Ideally, we'd have
1341  * one vector per ring/queue, but on a constrained vector budget, we
1342  * group the rings as "efficiently" as possible.  You would add new
1343  * mapping configurations in here.
1344  **/
1345 static int ixgbevf_map_rings_to_vectors(struct ixgbevf_adapter *adapter)
1346 {
1347         int q_vectors;
1348         int v_start = 0;
1349         int rxr_idx = 0, txr_idx = 0;
1350         int rxr_remaining = adapter->num_rx_queues;
1351         int txr_remaining = adapter->num_tx_queues;
1352         int i, j;
1353         int rqpv, tqpv;
1354
1355         q_vectors = adapter->num_msix_vectors - NON_Q_VECTORS;
1356
1357         /* The ideal configuration...
1358          * We have enough vectors to map one per queue.
1359          */
1360         if (q_vectors == adapter->num_rx_queues + adapter->num_tx_queues) {
1361                 for (; rxr_idx < rxr_remaining; v_start++, rxr_idx++)
1362                         map_vector_to_rxq(adapter, v_start, rxr_idx);
1363
1364                 for (; txr_idx < txr_remaining; v_start++, txr_idx++)
1365                         map_vector_to_txq(adapter, v_start, txr_idx);
1366                 return 0;
1367         }
1368
1369         /* If we don't have enough vectors for a 1-to-1
1370          * mapping, we'll have to group them so there are
1371          * multiple queues per vector.
1372          */
1373         /* Re-adjusting *qpv takes care of the remainder. */
1374         for (i = v_start; i < q_vectors; i++) {
1375                 rqpv = DIV_ROUND_UP(rxr_remaining, q_vectors - i);
1376                 for (j = 0; j < rqpv; j++) {
1377                         map_vector_to_rxq(adapter, i, rxr_idx);
1378                         rxr_idx++;
1379                         rxr_remaining--;
1380                 }
1381         }
1382         for (i = v_start; i < q_vectors; i++) {
1383                 tqpv = DIV_ROUND_UP(txr_remaining, q_vectors - i);
1384                 for (j = 0; j < tqpv; j++) {
1385                         map_vector_to_txq(adapter, i, txr_idx);
1386                         txr_idx++;
1387                         txr_remaining--;
1388                 }
1389         }
1390
1391         return 0;
1392 }
1393
1394 /**
1395  * ixgbevf_request_msix_irqs - Initialize MSI-X interrupts
1396  * @adapter: board private structure
1397  *
1398  * ixgbevf_request_msix_irqs allocates MSI-X vectors and requests
1399  * interrupts from the kernel.
1400  **/
1401 static int ixgbevf_request_msix_irqs(struct ixgbevf_adapter *adapter)
1402 {
1403         struct net_device *netdev = adapter->netdev;
1404         int q_vectors = adapter->num_msix_vectors - NON_Q_VECTORS;
1405         int vector, err;
1406         int ri = 0, ti = 0;
1407
1408         for (vector = 0; vector < q_vectors; vector++) {
1409                 struct ixgbevf_q_vector *q_vector = adapter->q_vector[vector];
1410                 struct msix_entry *entry = &adapter->msix_entries[vector];
1411
1412                 if (q_vector->tx.ring && q_vector->rx.ring) {
1413                         snprintf(q_vector->name, sizeof(q_vector->name) - 1,
1414                                  "%s-%s-%d", netdev->name, "TxRx", ri++);
1415                         ti++;
1416                 } else if (q_vector->rx.ring) {
1417                         snprintf(q_vector->name, sizeof(q_vector->name) - 1,
1418                                  "%s-%s-%d", netdev->name, "rx", ri++);
1419                 } else if (q_vector->tx.ring) {
1420                         snprintf(q_vector->name, sizeof(q_vector->name) - 1,
1421                                  "%s-%s-%d", netdev->name, "tx", ti++);
1422                 } else {
1423                         /* skip this unused q_vector */
1424                         continue;
1425                 }
1426                 err = request_irq(entry->vector, &ixgbevf_msix_clean_rings, 0,
1427                                   q_vector->name, q_vector);
1428                 if (err) {
1429                         hw_dbg(&adapter->hw,
1430                                "request_irq failed for MSIX interrupt Error: %d\n",
1431                                err);
1432                         goto free_queue_irqs;
1433                 }
1434         }
1435
1436         err = request_irq(adapter->msix_entries[vector].vector,
1437                           &ixgbevf_msix_other, 0, netdev->name, adapter);
1438         if (err) {
1439                 hw_dbg(&adapter->hw, "request_irq for msix_other failed: %d\n",
1440                        err);
1441                 goto free_queue_irqs;
1442         }
1443
1444         return 0;
1445
1446 free_queue_irqs:
1447         while (vector) {
1448                 vector--;
1449                 free_irq(adapter->msix_entries[vector].vector,
1450                          adapter->q_vector[vector]);
1451         }
1452         /* This failure is non-recoverable - it indicates the system is
1453          * out of MSIX vector resources and the VF driver cannot run
1454          * without them.  Set the number of msix vectors to zero
1455          * indicating that not enough can be allocated.  The error
1456          * will be returned to the user indicating device open failed.
1457          * Any further attempts to force the driver to open will also
1458          * fail.  The only way to recover is to unload the driver and
1459          * reload it again.  If the system has recovered some MSIX
1460          * vectors then it may succeed.
1461          */
1462         adapter->num_msix_vectors = 0;
1463         return err;
1464 }
1465
1466 static inline void ixgbevf_reset_q_vectors(struct ixgbevf_adapter *adapter)
1467 {
1468         int i, q_vectors = adapter->num_msix_vectors - NON_Q_VECTORS;
1469
1470         for (i = 0; i < q_vectors; i++) {
1471                 struct ixgbevf_q_vector *q_vector = adapter->q_vector[i];
1472
1473                 q_vector->rx.ring = NULL;
1474                 q_vector->tx.ring = NULL;
1475                 q_vector->rx.count = 0;
1476                 q_vector->tx.count = 0;
1477         }
1478 }
1479
1480 /**
1481  * ixgbevf_request_irq - initialize interrupts
1482  * @adapter: board private structure
1483  *
1484  * Attempts to configure interrupts using the best available
1485  * capabilities of the hardware and kernel.
1486  **/
1487 static int ixgbevf_request_irq(struct ixgbevf_adapter *adapter)
1488 {
1489         int err = ixgbevf_request_msix_irqs(adapter);
1490
1491         if (err)
1492                 hw_dbg(&adapter->hw, "request_irq failed, Error %d\n", err);
1493
1494         return err;
1495 }
1496
1497 static void ixgbevf_free_irq(struct ixgbevf_adapter *adapter)
1498 {
1499         int i, q_vectors;
1500
1501         q_vectors = adapter->num_msix_vectors;
1502         i = q_vectors - 1;
1503
1504         free_irq(adapter->msix_entries[i].vector, adapter);
1505         i--;
1506
1507         for (; i >= 0; i--) {
1508                 /* free only the irqs that were actually requested */
1509                 if (!adapter->q_vector[i]->rx.ring &&
1510                     !adapter->q_vector[i]->tx.ring)
1511                         continue;
1512
1513                 free_irq(adapter->msix_entries[i].vector,
1514                          adapter->q_vector[i]);
1515         }
1516
1517         ixgbevf_reset_q_vectors(adapter);
1518 }
1519
1520 /**
1521  * ixgbevf_irq_disable - Mask off interrupt generation on the NIC
1522  * @adapter: board private structure
1523  **/
1524 static inline void ixgbevf_irq_disable(struct ixgbevf_adapter *adapter)
1525 {
1526         struct ixgbe_hw *hw = &adapter->hw;
1527         int i;
1528
1529         IXGBE_WRITE_REG(hw, IXGBE_VTEIAM, 0);
1530         IXGBE_WRITE_REG(hw, IXGBE_VTEIMC, ~0);
1531         IXGBE_WRITE_REG(hw, IXGBE_VTEIAC, 0);
1532
1533         IXGBE_WRITE_FLUSH(hw);
1534
1535         for (i = 0; i < adapter->num_msix_vectors; i++)
1536                 synchronize_irq(adapter->msix_entries[i].vector);
1537 }
1538
1539 /**
1540  * ixgbevf_irq_enable - Enable default interrupt generation settings
1541  * @adapter: board private structure
1542  **/
1543 static inline void ixgbevf_irq_enable(struct ixgbevf_adapter *adapter)
1544 {
1545         struct ixgbe_hw *hw = &adapter->hw;
1546
1547         IXGBE_WRITE_REG(hw, IXGBE_VTEIAM, adapter->eims_enable_mask);
1548         IXGBE_WRITE_REG(hw, IXGBE_VTEIAC, adapter->eims_enable_mask);
1549         IXGBE_WRITE_REG(hw, IXGBE_VTEIMS, adapter->eims_enable_mask);
1550 }
1551
1552 /**
1553  * ixgbevf_configure_tx_ring - Configure 82599 VF Tx ring after Reset
1554  * @adapter: board private structure
1555  * @ring: structure containing ring specific data
1556  *
1557  * Configure the Tx descriptor ring after a reset.
1558  **/
1559 static void ixgbevf_configure_tx_ring(struct ixgbevf_adapter *adapter,
1560                                       struct ixgbevf_ring *ring)
1561 {
1562         struct ixgbe_hw *hw = &adapter->hw;
1563         u64 tdba = ring->dma;
1564         int wait_loop = 10;
1565         u32 txdctl = IXGBE_TXDCTL_ENABLE;
1566         u8 reg_idx = ring->reg_idx;
1567
1568         /* disable queue to avoid issues while updating state */
1569         IXGBE_WRITE_REG(hw, IXGBE_VFTXDCTL(reg_idx), IXGBE_TXDCTL_SWFLSH);
1570         IXGBE_WRITE_FLUSH(hw);
1571
1572         IXGBE_WRITE_REG(hw, IXGBE_VFTDBAL(reg_idx), tdba & DMA_BIT_MASK(32));
1573         IXGBE_WRITE_REG(hw, IXGBE_VFTDBAH(reg_idx), tdba >> 32);
1574         IXGBE_WRITE_REG(hw, IXGBE_VFTDLEN(reg_idx),
1575                         ring->count * sizeof(union ixgbe_adv_tx_desc));
1576
1577         /* disable head writeback */
1578         IXGBE_WRITE_REG(hw, IXGBE_VFTDWBAH(reg_idx), 0);
1579         IXGBE_WRITE_REG(hw, IXGBE_VFTDWBAL(reg_idx), 0);
1580
1581         /* enable relaxed ordering */
1582         IXGBE_WRITE_REG(hw, IXGBE_VFDCA_TXCTRL(reg_idx),
1583                         (IXGBE_DCA_TXCTRL_DESC_RRO_EN |
1584                          IXGBE_DCA_TXCTRL_DATA_RRO_EN));
1585
1586         /* reset head and tail pointers */
1587         IXGBE_WRITE_REG(hw, IXGBE_VFTDH(reg_idx), 0);
1588         IXGBE_WRITE_REG(hw, IXGBE_VFTDT(reg_idx), 0);
1589         ring->tail = adapter->io_addr + IXGBE_VFTDT(reg_idx);
1590
1591         /* reset ntu and ntc to place SW in sync with hardwdare */
1592         ring->next_to_clean = 0;
1593         ring->next_to_use = 0;
1594
1595         /* In order to avoid issues WTHRESH + PTHRESH should always be equal
1596          * to or less than the number of on chip descriptors, which is
1597          * currently 40.
1598          */
1599         txdctl |= (8 << 16);    /* WTHRESH = 8 */
1600
1601         /* Setting PTHRESH to 32 both improves performance */
1602         txdctl |= (1u << 8) |    /* HTHRESH = 1 */
1603                    32;           /* PTHRESH = 32 */
1604
1605         clear_bit(__IXGBEVF_HANG_CHECK_ARMED, &ring->state);
1606
1607         IXGBE_WRITE_REG(hw, IXGBE_VFTXDCTL(reg_idx), txdctl);
1608
1609         /* poll to verify queue is enabled */
1610         do {
1611                 usleep_range(1000, 2000);
1612                 txdctl = IXGBE_READ_REG(hw, IXGBE_VFTXDCTL(reg_idx));
1613         }  while (--wait_loop && !(txdctl & IXGBE_TXDCTL_ENABLE));
1614         if (!wait_loop)
1615                 hw_dbg(hw, "Could not enable Tx Queue %d\n", reg_idx);
1616 }
1617
1618 /**
1619  * ixgbevf_configure_tx - Configure 82599 VF Transmit Unit after Reset
1620  * @adapter: board private structure
1621  *
1622  * Configure the Tx unit of the MAC after a reset.
1623  **/
1624 static void ixgbevf_configure_tx(struct ixgbevf_adapter *adapter)
1625 {
1626         u32 i;
1627
1628         /* Setup the HW Tx Head and Tail descriptor pointers */
1629         for (i = 0; i < adapter->num_tx_queues; i++)
1630                 ixgbevf_configure_tx_ring(adapter, adapter->tx_ring[i]);
1631 }
1632
1633 #define IXGBE_SRRCTL_BSIZEHDRSIZE_SHIFT 2
1634
1635 static void ixgbevf_configure_srrctl(struct ixgbevf_adapter *adapter, int index)
1636 {
1637         struct ixgbe_hw *hw = &adapter->hw;
1638         u32 srrctl;
1639
1640         srrctl = IXGBE_SRRCTL_DROP_EN;
1641
1642         srrctl |= IXGBEVF_RX_HDR_SIZE << IXGBE_SRRCTL_BSIZEHDRSIZE_SHIFT;
1643         srrctl |= IXGBEVF_RX_BUFSZ >> IXGBE_SRRCTL_BSIZEPKT_SHIFT;
1644         srrctl |= IXGBE_SRRCTL_DESCTYPE_ADV_ONEBUF;
1645
1646         IXGBE_WRITE_REG(hw, IXGBE_VFSRRCTL(index), srrctl);
1647 }
1648
1649 static void ixgbevf_setup_psrtype(struct ixgbevf_adapter *adapter)
1650 {
1651         struct ixgbe_hw *hw = &adapter->hw;
1652
1653         /* PSRTYPE must be initialized in 82599 */
1654         u32 psrtype = IXGBE_PSRTYPE_TCPHDR | IXGBE_PSRTYPE_UDPHDR |
1655                       IXGBE_PSRTYPE_IPV4HDR | IXGBE_PSRTYPE_IPV6HDR |
1656                       IXGBE_PSRTYPE_L2HDR;
1657
1658         if (adapter->num_rx_queues > 1)
1659                 psrtype |= BIT(29);
1660
1661         IXGBE_WRITE_REG(hw, IXGBE_VFPSRTYPE, psrtype);
1662 }
1663
1664 #define IXGBEVF_MAX_RX_DESC_POLL 10
1665 static void ixgbevf_disable_rx_queue(struct ixgbevf_adapter *adapter,
1666                                      struct ixgbevf_ring *ring)
1667 {
1668         struct ixgbe_hw *hw = &adapter->hw;
1669         int wait_loop = IXGBEVF_MAX_RX_DESC_POLL;
1670         u32 rxdctl;
1671         u8 reg_idx = ring->reg_idx;
1672
1673         if (IXGBE_REMOVED(hw->hw_addr))
1674                 return;
1675         rxdctl = IXGBE_READ_REG(hw, IXGBE_VFRXDCTL(reg_idx));
1676         rxdctl &= ~IXGBE_RXDCTL_ENABLE;
1677
1678         /* write value back with RXDCTL.ENABLE bit cleared */
1679         IXGBE_WRITE_REG(hw, IXGBE_VFRXDCTL(reg_idx), rxdctl);
1680
1681         /* the hardware may take up to 100us to really disable the Rx queue */
1682         do {
1683                 udelay(10);
1684                 rxdctl = IXGBE_READ_REG(hw, IXGBE_VFRXDCTL(reg_idx));
1685         } while (--wait_loop && (rxdctl & IXGBE_RXDCTL_ENABLE));
1686
1687         if (!wait_loop)
1688                 pr_err("RXDCTL.ENABLE queue %d not cleared while polling\n",
1689                        reg_idx);
1690 }
1691
1692 static void ixgbevf_rx_desc_queue_enable(struct ixgbevf_adapter *adapter,
1693                                          struct ixgbevf_ring *ring)
1694 {
1695         struct ixgbe_hw *hw = &adapter->hw;
1696         int wait_loop = IXGBEVF_MAX_RX_DESC_POLL;
1697         u32 rxdctl;
1698         u8 reg_idx = ring->reg_idx;
1699
1700         if (IXGBE_REMOVED(hw->hw_addr))
1701                 return;
1702         do {
1703                 usleep_range(1000, 2000);
1704                 rxdctl = IXGBE_READ_REG(hw, IXGBE_VFRXDCTL(reg_idx));
1705         } while (--wait_loop && !(rxdctl & IXGBE_RXDCTL_ENABLE));
1706
1707         if (!wait_loop)
1708                 pr_err("RXDCTL.ENABLE queue %d not set while polling\n",
1709                        reg_idx);
1710 }
1711
1712 static void ixgbevf_setup_vfmrqc(struct ixgbevf_adapter *adapter)
1713 {
1714         struct ixgbe_hw *hw = &adapter->hw;
1715         u32 vfmrqc = 0, vfreta = 0;
1716         u16 rss_i = adapter->num_rx_queues;
1717         u8 i, j;
1718
1719         /* Fill out hash function seeds */
1720         netdev_rss_key_fill(adapter->rss_key, sizeof(adapter->rss_key));
1721         for (i = 0; i < IXGBEVF_VFRSSRK_REGS; i++)
1722                 IXGBE_WRITE_REG(hw, IXGBE_VFRSSRK(i), adapter->rss_key[i]);
1723
1724         for (i = 0, j = 0; i < IXGBEVF_X550_VFRETA_SIZE; i++, j++) {
1725                 if (j == rss_i)
1726                         j = 0;
1727
1728                 adapter->rss_indir_tbl[i] = j;
1729
1730                 vfreta |= j << (i & 0x3) * 8;
1731                 if ((i & 3) == 3) {
1732                         IXGBE_WRITE_REG(hw, IXGBE_VFRETA(i >> 2), vfreta);
1733                         vfreta = 0;
1734                 }
1735         }
1736
1737         /* Perform hash on these packet types */
1738         vfmrqc |= IXGBE_VFMRQC_RSS_FIELD_IPV4 |
1739                 IXGBE_VFMRQC_RSS_FIELD_IPV4_TCP |
1740                 IXGBE_VFMRQC_RSS_FIELD_IPV6 |
1741                 IXGBE_VFMRQC_RSS_FIELD_IPV6_TCP;
1742
1743         vfmrqc |= IXGBE_VFMRQC_RSSEN;
1744
1745         IXGBE_WRITE_REG(hw, IXGBE_VFMRQC, vfmrqc);
1746 }
1747
1748 static void ixgbevf_configure_rx_ring(struct ixgbevf_adapter *adapter,
1749                                       struct ixgbevf_ring *ring)
1750 {
1751         struct ixgbe_hw *hw = &adapter->hw;
1752         u64 rdba = ring->dma;
1753         u32 rxdctl;
1754         u8 reg_idx = ring->reg_idx;
1755
1756         /* disable queue to avoid issues while updating state */
1757         rxdctl = IXGBE_READ_REG(hw, IXGBE_VFRXDCTL(reg_idx));
1758         ixgbevf_disable_rx_queue(adapter, ring);
1759
1760         IXGBE_WRITE_REG(hw, IXGBE_VFRDBAL(reg_idx), rdba & DMA_BIT_MASK(32));
1761         IXGBE_WRITE_REG(hw, IXGBE_VFRDBAH(reg_idx), rdba >> 32);
1762         IXGBE_WRITE_REG(hw, IXGBE_VFRDLEN(reg_idx),
1763                         ring->count * sizeof(union ixgbe_adv_rx_desc));
1764
1765 #ifndef CONFIG_SPARC
1766         /* enable relaxed ordering */
1767         IXGBE_WRITE_REG(hw, IXGBE_VFDCA_RXCTRL(reg_idx),
1768                         IXGBE_DCA_RXCTRL_DESC_RRO_EN);
1769 #else
1770         IXGBE_WRITE_REG(hw, IXGBE_VFDCA_RXCTRL(reg_idx),
1771                         IXGBE_DCA_RXCTRL_DESC_RRO_EN |
1772                         IXGBE_DCA_RXCTRL_DATA_WRO_EN);
1773 #endif
1774
1775         /* reset head and tail pointers */
1776         IXGBE_WRITE_REG(hw, IXGBE_VFRDH(reg_idx), 0);
1777         IXGBE_WRITE_REG(hw, IXGBE_VFRDT(reg_idx), 0);
1778         ring->tail = adapter->io_addr + IXGBE_VFRDT(reg_idx);
1779
1780         /* reset ntu and ntc to place SW in sync with hardwdare */
1781         ring->next_to_clean = 0;
1782         ring->next_to_use = 0;
1783         ring->next_to_alloc = 0;
1784
1785         ixgbevf_configure_srrctl(adapter, reg_idx);
1786
1787         /* allow any size packet since we can handle overflow */
1788         rxdctl &= ~IXGBE_RXDCTL_RLPML_EN;
1789
1790         rxdctl |= IXGBE_RXDCTL_ENABLE | IXGBE_RXDCTL_VME;
1791         IXGBE_WRITE_REG(hw, IXGBE_VFRXDCTL(reg_idx), rxdctl);
1792
1793         ixgbevf_rx_desc_queue_enable(adapter, ring);
1794         ixgbevf_alloc_rx_buffers(ring, ixgbevf_desc_unused(ring));
1795 }
1796
1797 /**
1798  * ixgbevf_configure_rx - Configure 82599 VF Receive Unit after Reset
1799  * @adapter: board private structure
1800  *
1801  * Configure the Rx unit of the MAC after a reset.
1802  **/
1803 static void ixgbevf_configure_rx(struct ixgbevf_adapter *adapter)
1804 {
1805         struct ixgbe_hw *hw = &adapter->hw;
1806         struct net_device *netdev = adapter->netdev;
1807         int i, ret;
1808
1809         ixgbevf_setup_psrtype(adapter);
1810         if (hw->mac.type >= ixgbe_mac_X550_vf)
1811                 ixgbevf_setup_vfmrqc(adapter);
1812
1813         /* notify the PF of our intent to use this size of frame */
1814         ret = hw->mac.ops.set_rlpml(hw, netdev->mtu + ETH_HLEN + ETH_FCS_LEN);
1815         if (ret)
1816                 dev_err(&adapter->pdev->dev,
1817                         "Failed to set MTU at %d\n", netdev->mtu);
1818
1819         /* Setup the HW Rx Head and Tail Descriptor Pointers and
1820          * the Base and Length of the Rx Descriptor Ring
1821          */
1822         for (i = 0; i < adapter->num_rx_queues; i++)
1823                 ixgbevf_configure_rx_ring(adapter, adapter->rx_ring[i]);
1824 }
1825
1826 static int ixgbevf_vlan_rx_add_vid(struct net_device *netdev,
1827                                    __be16 proto, u16 vid)
1828 {
1829         struct ixgbevf_adapter *adapter = netdev_priv(netdev);
1830         struct ixgbe_hw *hw = &adapter->hw;
1831         int err;
1832
1833         spin_lock_bh(&adapter->mbx_lock);
1834
1835         /* add VID to filter table */
1836         err = hw->mac.ops.set_vfta(hw, vid, 0, true);
1837
1838         spin_unlock_bh(&adapter->mbx_lock);
1839
1840         /* translate error return types so error makes sense */
1841         if (err == IXGBE_ERR_MBX)
1842                 return -EIO;
1843
1844         if (err == IXGBE_ERR_INVALID_ARGUMENT)
1845                 return -EACCES;
1846
1847         set_bit(vid, adapter->active_vlans);
1848
1849         return err;
1850 }
1851
1852 static int ixgbevf_vlan_rx_kill_vid(struct net_device *netdev,
1853                                     __be16 proto, u16 vid)
1854 {
1855         struct ixgbevf_adapter *adapter = netdev_priv(netdev);
1856         struct ixgbe_hw *hw = &adapter->hw;
1857         int err;
1858
1859         spin_lock_bh(&adapter->mbx_lock);
1860
1861         /* remove VID from filter table */
1862         err = hw->mac.ops.set_vfta(hw, vid, 0, false);
1863
1864         spin_unlock_bh(&adapter->mbx_lock);
1865
1866         clear_bit(vid, adapter->active_vlans);
1867
1868         return err;
1869 }
1870
1871 static void ixgbevf_restore_vlan(struct ixgbevf_adapter *adapter)
1872 {
1873         u16 vid;
1874
1875         for_each_set_bit(vid, adapter->active_vlans, VLAN_N_VID)
1876                 ixgbevf_vlan_rx_add_vid(adapter->netdev,
1877                                         htons(ETH_P_8021Q), vid);
1878 }
1879
1880 static int ixgbevf_write_uc_addr_list(struct net_device *netdev)
1881 {
1882         struct ixgbevf_adapter *adapter = netdev_priv(netdev);
1883         struct ixgbe_hw *hw = &adapter->hw;
1884         int count = 0;
1885
1886         if ((netdev_uc_count(netdev)) > 10) {
1887                 pr_err("Too many unicast filters - No Space\n");
1888                 return -ENOSPC;
1889         }
1890
1891         if (!netdev_uc_empty(netdev)) {
1892                 struct netdev_hw_addr *ha;
1893
1894                 netdev_for_each_uc_addr(ha, netdev) {
1895                         hw->mac.ops.set_uc_addr(hw, ++count, ha->addr);
1896                         udelay(200);
1897                 }
1898         } else {
1899                 /* If the list is empty then send message to PF driver to
1900                  * clear all MAC VLANs on this VF.
1901                  */
1902                 hw->mac.ops.set_uc_addr(hw, 0, NULL);
1903         }
1904
1905         return count;
1906 }
1907
1908 /**
1909  * ixgbevf_set_rx_mode - Multicast and unicast set
1910  * @netdev: network interface device structure
1911  *
1912  * The set_rx_method entry point is called whenever the multicast address
1913  * list, unicast address list or the network interface flags are updated.
1914  * This routine is responsible for configuring the hardware for proper
1915  * multicast mode and configuring requested unicast filters.
1916  **/
1917 static void ixgbevf_set_rx_mode(struct net_device *netdev)
1918 {
1919         struct ixgbevf_adapter *adapter = netdev_priv(netdev);
1920         struct ixgbe_hw *hw = &adapter->hw;
1921         unsigned int flags = netdev->flags;
1922         int xcast_mode;
1923
1924         xcast_mode = (flags & IFF_ALLMULTI) ? IXGBEVF_XCAST_MODE_ALLMULTI :
1925                      (flags & (IFF_BROADCAST | IFF_MULTICAST)) ?
1926                      IXGBEVF_XCAST_MODE_MULTI : IXGBEVF_XCAST_MODE_NONE;
1927
1928         spin_lock_bh(&adapter->mbx_lock);
1929
1930         hw->mac.ops.update_xcast_mode(hw, xcast_mode);
1931
1932         /* reprogram multicast list */
1933         hw->mac.ops.update_mc_addr_list(hw, netdev);
1934
1935         ixgbevf_write_uc_addr_list(netdev);
1936
1937         spin_unlock_bh(&adapter->mbx_lock);
1938 }
1939
1940 static void ixgbevf_napi_enable_all(struct ixgbevf_adapter *adapter)
1941 {
1942         int q_idx;
1943         struct ixgbevf_q_vector *q_vector;
1944         int q_vectors = adapter->num_msix_vectors - NON_Q_VECTORS;
1945
1946         for (q_idx = 0; q_idx < q_vectors; q_idx++) {
1947                 q_vector = adapter->q_vector[q_idx];
1948 #ifdef CONFIG_NET_RX_BUSY_POLL
1949                 ixgbevf_qv_init_lock(adapter->q_vector[q_idx]);
1950 #endif
1951                 napi_enable(&q_vector->napi);
1952         }
1953 }
1954
1955 static void ixgbevf_napi_disable_all(struct ixgbevf_adapter *adapter)
1956 {
1957         int q_idx;
1958         struct ixgbevf_q_vector *q_vector;
1959         int q_vectors = adapter->num_msix_vectors - NON_Q_VECTORS;
1960
1961         for (q_idx = 0; q_idx < q_vectors; q_idx++) {
1962                 q_vector = adapter->q_vector[q_idx];
1963                 napi_disable(&q_vector->napi);
1964 #ifdef CONFIG_NET_RX_BUSY_POLL
1965                 while (!ixgbevf_qv_disable(adapter->q_vector[q_idx])) {
1966                         pr_info("QV %d locked\n", q_idx);
1967                         usleep_range(1000, 20000);
1968                 }
1969 #endif /* CONFIG_NET_RX_BUSY_POLL */
1970         }
1971 }
1972
1973 static int ixgbevf_configure_dcb(struct ixgbevf_adapter *adapter)
1974 {
1975         struct ixgbe_hw *hw = &adapter->hw;
1976         unsigned int def_q = 0;
1977         unsigned int num_tcs = 0;
1978         unsigned int num_rx_queues = adapter->num_rx_queues;
1979         unsigned int num_tx_queues = adapter->num_tx_queues;
1980         int err;
1981
1982         spin_lock_bh(&adapter->mbx_lock);
1983
1984         /* fetch queue configuration from the PF */
1985         err = ixgbevf_get_queues(hw, &num_tcs, &def_q);
1986
1987         spin_unlock_bh(&adapter->mbx_lock);
1988
1989         if (err)
1990                 return err;
1991
1992         if (num_tcs > 1) {
1993                 /* we need only one Tx queue */
1994                 num_tx_queues = 1;
1995
1996                 /* update default Tx ring register index */
1997                 adapter->tx_ring[0]->reg_idx = def_q;
1998
1999                 /* we need as many queues as traffic classes */
2000                 num_rx_queues = num_tcs;
2001         }
2002
2003         /* if we have a bad config abort request queue reset */
2004         if ((adapter->num_rx_queues != num_rx_queues) ||
2005             (adapter->num_tx_queues != num_tx_queues)) {
2006                 /* force mailbox timeout to prevent further messages */
2007                 hw->mbx.timeout = 0;
2008
2009                 /* wait for watchdog to come around and bail us out */
2010                 set_bit(__IXGBEVF_QUEUE_RESET_REQUESTED, &adapter->state);
2011         }
2012
2013         return 0;
2014 }
2015
2016 static void ixgbevf_configure(struct ixgbevf_adapter *adapter)
2017 {
2018         ixgbevf_configure_dcb(adapter);
2019
2020         ixgbevf_set_rx_mode(adapter->netdev);
2021
2022         ixgbevf_restore_vlan(adapter);
2023
2024         ixgbevf_configure_tx(adapter);
2025         ixgbevf_configure_rx(adapter);
2026 }
2027
2028 static void ixgbevf_save_reset_stats(struct ixgbevf_adapter *adapter)
2029 {
2030         /* Only save pre-reset stats if there are some */
2031         if (adapter->stats.vfgprc || adapter->stats.vfgptc) {
2032                 adapter->stats.saved_reset_vfgprc += adapter->stats.vfgprc -
2033                         adapter->stats.base_vfgprc;
2034                 adapter->stats.saved_reset_vfgptc += adapter->stats.vfgptc -
2035                         adapter->stats.base_vfgptc;
2036                 adapter->stats.saved_reset_vfgorc += adapter->stats.vfgorc -
2037                         adapter->stats.base_vfgorc;
2038                 adapter->stats.saved_reset_vfgotc += adapter->stats.vfgotc -
2039                         adapter->stats.base_vfgotc;
2040                 adapter->stats.saved_reset_vfmprc += adapter->stats.vfmprc -
2041                         adapter->stats.base_vfmprc;
2042         }
2043 }
2044
2045 static void ixgbevf_init_last_counter_stats(struct ixgbevf_adapter *adapter)
2046 {
2047         struct ixgbe_hw *hw = &adapter->hw;
2048
2049         adapter->stats.last_vfgprc = IXGBE_READ_REG(hw, IXGBE_VFGPRC);
2050         adapter->stats.last_vfgorc = IXGBE_READ_REG(hw, IXGBE_VFGORC_LSB);
2051         adapter->stats.last_vfgorc |=
2052                 (((u64)(IXGBE_READ_REG(hw, IXGBE_VFGORC_MSB))) << 32);
2053         adapter->stats.last_vfgptc = IXGBE_READ_REG(hw, IXGBE_VFGPTC);
2054         adapter->stats.last_vfgotc = IXGBE_READ_REG(hw, IXGBE_VFGOTC_LSB);
2055         adapter->stats.last_vfgotc |=
2056                 (((u64)(IXGBE_READ_REG(hw, IXGBE_VFGOTC_MSB))) << 32);
2057         adapter->stats.last_vfmprc = IXGBE_READ_REG(hw, IXGBE_VFMPRC);
2058
2059         adapter->stats.base_vfgprc = adapter->stats.last_vfgprc;
2060         adapter->stats.base_vfgorc = adapter->stats.last_vfgorc;
2061         adapter->stats.base_vfgptc = adapter->stats.last_vfgptc;
2062         adapter->stats.base_vfgotc = adapter->stats.last_vfgotc;
2063         adapter->stats.base_vfmprc = adapter->stats.last_vfmprc;
2064 }
2065
2066 static void ixgbevf_negotiate_api(struct ixgbevf_adapter *adapter)
2067 {
2068         struct ixgbe_hw *hw = &adapter->hw;
2069         int api[] = { ixgbe_mbox_api_12,
2070                       ixgbe_mbox_api_11,
2071                       ixgbe_mbox_api_10,
2072                       ixgbe_mbox_api_unknown };
2073         int err, idx = 0;
2074
2075         spin_lock_bh(&adapter->mbx_lock);
2076
2077         while (api[idx] != ixgbe_mbox_api_unknown) {
2078                 err = hw->mac.ops.negotiate_api_version(hw, api[idx]);
2079                 if (!err)
2080                         break;
2081                 idx++;
2082         }
2083
2084         spin_unlock_bh(&adapter->mbx_lock);
2085 }
2086
2087 static void ixgbevf_up_complete(struct ixgbevf_adapter *adapter)
2088 {
2089         struct net_device *netdev = adapter->netdev;
2090         struct ixgbe_hw *hw = &adapter->hw;
2091
2092         ixgbevf_configure_msix(adapter);
2093
2094         spin_lock_bh(&adapter->mbx_lock);
2095
2096         if (is_valid_ether_addr(hw->mac.addr))
2097                 hw->mac.ops.set_rar(hw, 0, hw->mac.addr, 0);
2098         else
2099                 hw->mac.ops.set_rar(hw, 0, hw->mac.perm_addr, 0);
2100
2101         spin_unlock_bh(&adapter->mbx_lock);
2102
2103         smp_mb__before_atomic();
2104         clear_bit(__IXGBEVF_DOWN, &adapter->state);
2105         ixgbevf_napi_enable_all(adapter);
2106
2107         /* clear any pending interrupts, may auto mask */
2108         IXGBE_READ_REG(hw, IXGBE_VTEICR);
2109         ixgbevf_irq_enable(adapter);
2110
2111         /* enable transmits */
2112         netif_tx_start_all_queues(netdev);
2113
2114         ixgbevf_save_reset_stats(adapter);
2115         ixgbevf_init_last_counter_stats(adapter);
2116
2117         hw->mac.get_link_status = 1;
2118         mod_timer(&adapter->service_timer, jiffies);
2119 }
2120
2121 void ixgbevf_up(struct ixgbevf_adapter *adapter)
2122 {
2123         ixgbevf_configure(adapter);
2124
2125         ixgbevf_up_complete(adapter);
2126 }
2127
2128 /**
2129  * ixgbevf_clean_rx_ring - Free Rx Buffers per Queue
2130  * @rx_ring: ring to free buffers from
2131  **/
2132 static void ixgbevf_clean_rx_ring(struct ixgbevf_ring *rx_ring)
2133 {
2134         struct device *dev = rx_ring->dev;
2135         unsigned long size;
2136         unsigned int i;
2137
2138         /* Free Rx ring sk_buff */
2139         if (rx_ring->skb) {
2140                 dev_kfree_skb(rx_ring->skb);
2141                 rx_ring->skb = NULL;
2142         }
2143
2144         /* ring already cleared, nothing to do */
2145         if (!rx_ring->rx_buffer_info)
2146                 return;
2147
2148         /* Free all the Rx ring pages */
2149         for (i = 0; i < rx_ring->count; i++) {
2150                 struct ixgbevf_rx_buffer *rx_buffer;
2151
2152                 rx_buffer = &rx_ring->rx_buffer_info[i];
2153                 if (rx_buffer->dma)
2154                         dma_unmap_page(dev, rx_buffer->dma,
2155                                        PAGE_SIZE, DMA_FROM_DEVICE);
2156                 rx_buffer->dma = 0;
2157                 if (rx_buffer->page)
2158                         __free_page(rx_buffer->page);
2159                 rx_buffer->page = NULL;
2160         }
2161
2162         size = sizeof(struct ixgbevf_rx_buffer) * rx_ring->count;
2163         memset(rx_ring->rx_buffer_info, 0, size);
2164
2165         /* Zero out the descriptor ring */
2166         memset(rx_ring->desc, 0, rx_ring->size);
2167 }
2168
2169 /**
2170  * ixgbevf_clean_tx_ring - Free Tx Buffers
2171  * @tx_ring: ring to be cleaned
2172  **/
2173 static void ixgbevf_clean_tx_ring(struct ixgbevf_ring *tx_ring)
2174 {
2175         struct ixgbevf_tx_buffer *tx_buffer_info;
2176         unsigned long size;
2177         unsigned int i;
2178
2179         if (!tx_ring->tx_buffer_info)
2180                 return;
2181
2182         /* Free all the Tx ring sk_buffs */
2183         for (i = 0; i < tx_ring->count; i++) {
2184                 tx_buffer_info = &tx_ring->tx_buffer_info[i];
2185                 ixgbevf_unmap_and_free_tx_resource(tx_ring, tx_buffer_info);
2186         }
2187
2188         size = sizeof(struct ixgbevf_tx_buffer) * tx_ring->count;
2189         memset(tx_ring->tx_buffer_info, 0, size);
2190
2191         memset(tx_ring->desc, 0, tx_ring->size);
2192 }
2193
2194 /**
2195  * ixgbevf_clean_all_rx_rings - Free Rx Buffers for all queues
2196  * @adapter: board private structure
2197  **/
2198 static void ixgbevf_clean_all_rx_rings(struct ixgbevf_adapter *adapter)
2199 {
2200         int i;
2201
2202         for (i = 0; i < adapter->num_rx_queues; i++)
2203                 ixgbevf_clean_rx_ring(adapter->rx_ring[i]);
2204 }
2205
2206 /**
2207  * ixgbevf_clean_all_tx_rings - Free Tx Buffers for all queues
2208  * @adapter: board private structure
2209  **/
2210 static void ixgbevf_clean_all_tx_rings(struct ixgbevf_adapter *adapter)
2211 {
2212         int i;
2213
2214         for (i = 0; i < adapter->num_tx_queues; i++)
2215                 ixgbevf_clean_tx_ring(adapter->tx_ring[i]);
2216 }
2217
2218 void ixgbevf_down(struct ixgbevf_adapter *adapter)
2219 {
2220         struct net_device *netdev = adapter->netdev;
2221         struct ixgbe_hw *hw = &adapter->hw;
2222         int i;
2223
2224         /* signal that we are down to the interrupt handler */
2225         if (test_and_set_bit(__IXGBEVF_DOWN, &adapter->state))
2226                 return; /* do nothing if already down */
2227
2228         /* disable all enabled Rx queues */
2229         for (i = 0; i < adapter->num_rx_queues; i++)
2230                 ixgbevf_disable_rx_queue(adapter, adapter->rx_ring[i]);
2231
2232         usleep_range(10000, 20000);
2233
2234         netif_tx_stop_all_queues(netdev);
2235
2236         /* call carrier off first to avoid false dev_watchdog timeouts */
2237         netif_carrier_off(netdev);
2238         netif_tx_disable(netdev);
2239
2240         ixgbevf_irq_disable(adapter);
2241
2242         ixgbevf_napi_disable_all(adapter);
2243
2244         del_timer_sync(&adapter->service_timer);
2245
2246         /* disable transmits in the hardware now that interrupts are off */
2247         for (i = 0; i < adapter->num_tx_queues; i++) {
2248                 u8 reg_idx = adapter->tx_ring[i]->reg_idx;
2249
2250                 IXGBE_WRITE_REG(hw, IXGBE_VFTXDCTL(reg_idx),
2251                                 IXGBE_TXDCTL_SWFLSH);
2252         }
2253
2254         if (!pci_channel_offline(adapter->pdev))
2255                 ixgbevf_reset(adapter);
2256
2257         ixgbevf_clean_all_tx_rings(adapter);
2258         ixgbevf_clean_all_rx_rings(adapter);
2259 }
2260
2261 void ixgbevf_reinit_locked(struct ixgbevf_adapter *adapter)
2262 {
2263         WARN_ON(in_interrupt());
2264
2265         while (test_and_set_bit(__IXGBEVF_RESETTING, &adapter->state))
2266                 msleep(1);
2267
2268         ixgbevf_down(adapter);
2269         ixgbevf_up(adapter);
2270
2271         clear_bit(__IXGBEVF_RESETTING, &adapter->state);
2272 }
2273
2274 void ixgbevf_reset(struct ixgbevf_adapter *adapter)
2275 {
2276         struct ixgbe_hw *hw = &adapter->hw;
2277         struct net_device *netdev = adapter->netdev;
2278
2279         if (hw->mac.ops.reset_hw(hw)) {
2280                 hw_dbg(hw, "PF still resetting\n");
2281         } else {
2282                 hw->mac.ops.init_hw(hw);
2283                 ixgbevf_negotiate_api(adapter);
2284         }
2285
2286         if (is_valid_ether_addr(adapter->hw.mac.addr)) {
2287                 ether_addr_copy(netdev->dev_addr, adapter->hw.mac.addr);
2288                 ether_addr_copy(netdev->perm_addr, adapter->hw.mac.addr);
2289         }
2290
2291         adapter->last_reset = jiffies;
2292 }
2293
2294 static int ixgbevf_acquire_msix_vectors(struct ixgbevf_adapter *adapter,
2295                                         int vectors)
2296 {
2297         int vector_threshold;
2298
2299         /* We'll want at least 2 (vector_threshold):
2300          * 1) TxQ[0] + RxQ[0] handler
2301          * 2) Other (Link Status Change, etc.)
2302          */
2303         vector_threshold = MIN_MSIX_COUNT;
2304
2305         /* The more we get, the more we will assign to Tx/Rx Cleanup
2306          * for the separate queues...where Rx Cleanup >= Tx Cleanup.
2307          * Right now, we simply care about how many we'll get; we'll
2308          * set them up later while requesting irq's.
2309          */
2310         vectors = pci_enable_msix_range(adapter->pdev, adapter->msix_entries,
2311                                         vector_threshold, vectors);
2312
2313         if (vectors < 0) {
2314                 dev_err(&adapter->pdev->dev,
2315                         "Unable to allocate MSI-X interrupts\n");
2316                 kfree(adapter->msix_entries);
2317                 adapter->msix_entries = NULL;
2318                 return vectors;
2319         }
2320
2321         /* Adjust for only the vectors we'll use, which is minimum
2322          * of max_msix_q_vectors + NON_Q_VECTORS, or the number of
2323          * vectors we were allocated.
2324          */
2325         adapter->num_msix_vectors = vectors;
2326
2327         return 0;
2328 }
2329
2330 /**
2331  * ixgbevf_set_num_queues - Allocate queues for device, feature dependent
2332  * @adapter: board private structure to initialize
2333  *
2334  * This is the top level queue allocation routine.  The order here is very
2335  * important, starting with the "most" number of features turned on at once,
2336  * and ending with the smallest set of features.  This way large combinations
2337  * can be allocated if they're turned on, and smaller combinations are the
2338  * fallthrough conditions.
2339  *
2340  **/
2341 static void ixgbevf_set_num_queues(struct ixgbevf_adapter *adapter)
2342 {
2343         struct ixgbe_hw *hw = &adapter->hw;
2344         unsigned int def_q = 0;
2345         unsigned int num_tcs = 0;
2346         int err;
2347
2348         /* Start with base case */
2349         adapter->num_rx_queues = 1;
2350         adapter->num_tx_queues = 1;
2351
2352         spin_lock_bh(&adapter->mbx_lock);
2353
2354         /* fetch queue configuration from the PF */
2355         err = ixgbevf_get_queues(hw, &num_tcs, &def_q);
2356
2357         spin_unlock_bh(&adapter->mbx_lock);
2358
2359         if (err)
2360                 return;
2361
2362         /* we need as many queues as traffic classes */
2363         if (num_tcs > 1) {
2364                 adapter->num_rx_queues = num_tcs;
2365         } else {
2366                 u16 rss = min_t(u16, num_online_cpus(), IXGBEVF_MAX_RSS_QUEUES);
2367
2368                 switch (hw->api_version) {
2369                 case ixgbe_mbox_api_11:
2370                 case ixgbe_mbox_api_12:
2371                         adapter->num_rx_queues = rss;
2372                         adapter->num_tx_queues = rss;
2373                 default:
2374                         break;
2375                 }
2376         }
2377 }
2378
2379 /**
2380  * ixgbevf_alloc_queues - Allocate memory for all rings
2381  * @adapter: board private structure to initialize
2382  *
2383  * We allocate one ring per queue at run-time since we don't know the
2384  * number of queues at compile-time.  The polling_netdev array is
2385  * intended for Multiqueue, but should work fine with a single queue.
2386  **/
2387 static int ixgbevf_alloc_queues(struct ixgbevf_adapter *adapter)
2388 {
2389         struct ixgbevf_ring *ring;
2390         int rx = 0, tx = 0;
2391
2392         for (; tx < adapter->num_tx_queues; tx++) {
2393                 ring = kzalloc(sizeof(*ring), GFP_KERNEL);
2394                 if (!ring)
2395                         goto err_allocation;
2396
2397                 ring->dev = &adapter->pdev->dev;
2398                 ring->netdev = adapter->netdev;
2399                 ring->count = adapter->tx_ring_count;
2400                 ring->queue_index = tx;
2401                 ring->reg_idx = tx;
2402
2403                 adapter->tx_ring[tx] = ring;
2404         }
2405
2406         for (; rx < adapter->num_rx_queues; rx++) {
2407                 ring = kzalloc(sizeof(*ring), GFP_KERNEL);
2408                 if (!ring)
2409                         goto err_allocation;
2410
2411                 ring->dev = &adapter->pdev->dev;
2412                 ring->netdev = adapter->netdev;
2413
2414                 ring->count = adapter->rx_ring_count;
2415                 ring->queue_index = rx;
2416                 ring->reg_idx = rx;
2417
2418                 adapter->rx_ring[rx] = ring;
2419         }
2420
2421         return 0;
2422
2423 err_allocation:
2424         while (tx) {
2425                 kfree(adapter->tx_ring[--tx]);
2426                 adapter->tx_ring[tx] = NULL;
2427         }
2428
2429         while (rx) {
2430                 kfree(adapter->rx_ring[--rx]);
2431                 adapter->rx_ring[rx] = NULL;
2432         }
2433         return -ENOMEM;
2434 }
2435
2436 /**
2437  * ixgbevf_set_interrupt_capability - set MSI-X or FAIL if not supported
2438  * @adapter: board private structure to initialize
2439  *
2440  * Attempt to configure the interrupts using the best available
2441  * capabilities of the hardware and the kernel.
2442  **/
2443 static int ixgbevf_set_interrupt_capability(struct ixgbevf_adapter *adapter)
2444 {
2445         struct net_device *netdev = adapter->netdev;
2446         int err;
2447         int vector, v_budget;
2448
2449         /* It's easy to be greedy for MSI-X vectors, but it really
2450          * doesn't do us much good if we have a lot more vectors
2451          * than CPU's.  So let's be conservative and only ask for
2452          * (roughly) the same number of vectors as there are CPU's.
2453          * The default is to use pairs of vectors.
2454          */
2455         v_budget = max(adapter->num_rx_queues, adapter->num_tx_queues);
2456         v_budget = min_t(int, v_budget, num_online_cpus());
2457         v_budget += NON_Q_VECTORS;
2458
2459         /* A failure in MSI-X entry allocation isn't fatal, but it does
2460          * mean we disable MSI-X capabilities of the adapter.
2461          */
2462         adapter->msix_entries = kcalloc(v_budget,
2463                                         sizeof(struct msix_entry), GFP_KERNEL);
2464         if (!adapter->msix_entries)
2465                 return -ENOMEM;
2466
2467         for (vector = 0; vector < v_budget; vector++)
2468                 adapter->msix_entries[vector].entry = vector;
2469
2470         err = ixgbevf_acquire_msix_vectors(adapter, v_budget);
2471         if (err)
2472                 return err;
2473
2474         err = netif_set_real_num_tx_queues(netdev, adapter->num_tx_queues);
2475         if (err)
2476                 return err;
2477
2478         return netif_set_real_num_rx_queues(netdev, adapter->num_rx_queues);
2479 }
2480
2481 /**
2482  * ixgbevf_alloc_q_vectors - Allocate memory for interrupt vectors
2483  * @adapter: board private structure to initialize
2484  *
2485  * We allocate one q_vector per queue interrupt.  If allocation fails we
2486  * return -ENOMEM.
2487  **/
2488 static int ixgbevf_alloc_q_vectors(struct ixgbevf_adapter *adapter)
2489 {
2490         int q_idx, num_q_vectors;
2491         struct ixgbevf_q_vector *q_vector;
2492
2493         num_q_vectors = adapter->num_msix_vectors - NON_Q_VECTORS;
2494
2495         for (q_idx = 0; q_idx < num_q_vectors; q_idx++) {
2496                 q_vector = kzalloc(sizeof(struct ixgbevf_q_vector), GFP_KERNEL);
2497                 if (!q_vector)
2498                         goto err_out;
2499                 q_vector->adapter = adapter;
2500                 q_vector->v_idx = q_idx;
2501                 netif_napi_add(adapter->netdev, &q_vector->napi,
2502                                ixgbevf_poll, 64);
2503                 adapter->q_vector[q_idx] = q_vector;
2504         }
2505
2506         return 0;
2507
2508 err_out:
2509         while (q_idx) {
2510                 q_idx--;
2511                 q_vector = adapter->q_vector[q_idx];
2512 #ifdef CONFIG_NET_RX_BUSY_POLL
2513                 napi_hash_del(&q_vector->napi);
2514 #endif
2515                 netif_napi_del(&q_vector->napi);
2516                 kfree(q_vector);
2517                 adapter->q_vector[q_idx] = NULL;
2518         }
2519         return -ENOMEM;
2520 }
2521
2522 /**
2523  * ixgbevf_free_q_vectors - Free memory allocated for interrupt vectors
2524  * @adapter: board private structure to initialize
2525  *
2526  * This function frees the memory allocated to the q_vectors.  In addition if
2527  * NAPI is enabled it will delete any references to the NAPI struct prior
2528  * to freeing the q_vector.
2529  **/
2530 static void ixgbevf_free_q_vectors(struct ixgbevf_adapter *adapter)
2531 {
2532         int q_idx, num_q_vectors = adapter->num_msix_vectors - NON_Q_VECTORS;
2533
2534         for (q_idx = 0; q_idx < num_q_vectors; q_idx++) {
2535                 struct ixgbevf_q_vector *q_vector = adapter->q_vector[q_idx];
2536
2537                 adapter->q_vector[q_idx] = NULL;
2538 #ifdef CONFIG_NET_RX_BUSY_POLL
2539                 napi_hash_del(&q_vector->napi);
2540 #endif
2541                 netif_napi_del(&q_vector->napi);
2542                 kfree(q_vector);
2543         }
2544 }
2545
2546 /**
2547  * ixgbevf_reset_interrupt_capability - Reset MSIX setup
2548  * @adapter: board private structure
2549  *
2550  **/
2551 static void ixgbevf_reset_interrupt_capability(struct ixgbevf_adapter *adapter)
2552 {
2553         pci_disable_msix(adapter->pdev);
2554         kfree(adapter->msix_entries);
2555         adapter->msix_entries = NULL;
2556 }
2557
2558 /**
2559  * ixgbevf_init_interrupt_scheme - Determine if MSIX is supported and init
2560  * @adapter: board private structure to initialize
2561  *
2562  **/
2563 static int ixgbevf_init_interrupt_scheme(struct ixgbevf_adapter *adapter)
2564 {
2565         int err;
2566
2567         /* Number of supported queues */
2568         ixgbevf_set_num_queues(adapter);
2569
2570         err = ixgbevf_set_interrupt_capability(adapter);
2571         if (err) {
2572                 hw_dbg(&adapter->hw,
2573                        "Unable to setup interrupt capabilities\n");
2574                 goto err_set_interrupt;
2575         }
2576
2577         err = ixgbevf_alloc_q_vectors(adapter);
2578         if (err) {
2579                 hw_dbg(&adapter->hw, "Unable to allocate memory for queue vectors\n");
2580                 goto err_alloc_q_vectors;
2581         }
2582
2583         err = ixgbevf_alloc_queues(adapter);
2584         if (err) {
2585                 pr_err("Unable to allocate memory for queues\n");
2586                 goto err_alloc_queues;
2587         }
2588
2589         hw_dbg(&adapter->hw, "Multiqueue %s: Rx Queue count = %u, Tx Queue count = %u\n",
2590                (adapter->num_rx_queues > 1) ? "Enabled" :
2591                "Disabled", adapter->num_rx_queues, adapter->num_tx_queues);
2592
2593         set_bit(__IXGBEVF_DOWN, &adapter->state);
2594
2595         return 0;
2596 err_alloc_queues:
2597         ixgbevf_free_q_vectors(adapter);
2598 err_alloc_q_vectors:
2599         ixgbevf_reset_interrupt_capability(adapter);
2600 err_set_interrupt:
2601         return err;
2602 }
2603
2604 /**
2605  * ixgbevf_clear_interrupt_scheme - Clear the current interrupt scheme settings
2606  * @adapter: board private structure to clear interrupt scheme on
2607  *
2608  * We go through and clear interrupt specific resources and reset the structure
2609  * to pre-load conditions
2610  **/
2611 static void ixgbevf_clear_interrupt_scheme(struct ixgbevf_adapter *adapter)
2612 {
2613         int i;
2614
2615         for (i = 0; i < adapter->num_tx_queues; i++) {
2616                 kfree(adapter->tx_ring[i]);
2617                 adapter->tx_ring[i] = NULL;
2618         }
2619         for (i = 0; i < adapter->num_rx_queues; i++) {
2620                 kfree(adapter->rx_ring[i]);
2621                 adapter->rx_ring[i] = NULL;
2622         }
2623
2624         adapter->num_tx_queues = 0;
2625         adapter->num_rx_queues = 0;
2626
2627         ixgbevf_free_q_vectors(adapter);
2628         ixgbevf_reset_interrupt_capability(adapter);
2629 }
2630
2631 /**
2632  * ixgbevf_sw_init - Initialize general software structures
2633  * @adapter: board private structure to initialize
2634  *
2635  * ixgbevf_sw_init initializes the Adapter private data structure.
2636  * Fields are initialized based on PCI device information and
2637  * OS network device settings (MTU size).
2638  **/
2639 static int ixgbevf_sw_init(struct ixgbevf_adapter *adapter)
2640 {
2641         struct ixgbe_hw *hw = &adapter->hw;
2642         struct pci_dev *pdev = adapter->pdev;
2643         struct net_device *netdev = adapter->netdev;
2644         int err;
2645
2646         /* PCI config space info */
2647         hw->vendor_id = pdev->vendor;
2648         hw->device_id = pdev->device;
2649         hw->revision_id = pdev->revision;
2650         hw->subsystem_vendor_id = pdev->subsystem_vendor;
2651         hw->subsystem_device_id = pdev->subsystem_device;
2652
2653         hw->mbx.ops.init_params(hw);
2654
2655         /* assume legacy case in which PF would only give VF 2 queues */
2656         hw->mac.max_tx_queues = 2;
2657         hw->mac.max_rx_queues = 2;
2658
2659         /* lock to protect mailbox accesses */
2660         spin_lock_init(&adapter->mbx_lock);
2661
2662         err = hw->mac.ops.reset_hw(hw);
2663         if (err) {
2664                 dev_info(&pdev->dev,
2665                          "PF still in reset state.  Is the PF interface up?\n");
2666         } else {
2667                 err = hw->mac.ops.init_hw(hw);
2668                 if (err) {
2669                         pr_err("init_shared_code failed: %d\n", err);
2670                         goto out;
2671                 }
2672                 ixgbevf_negotiate_api(adapter);
2673                 err = hw->mac.ops.get_mac_addr(hw, hw->mac.addr);
2674                 if (err)
2675                         dev_info(&pdev->dev, "Error reading MAC address\n");
2676                 else if (is_zero_ether_addr(adapter->hw.mac.addr))
2677                         dev_info(&pdev->dev,
2678                                  "MAC address not assigned by administrator.\n");
2679                 ether_addr_copy(netdev->dev_addr, hw->mac.addr);
2680         }
2681
2682         if (!is_valid_ether_addr(netdev->dev_addr)) {
2683                 dev_info(&pdev->dev, "Assigning random MAC address\n");
2684                 eth_hw_addr_random(netdev);
2685                 ether_addr_copy(hw->mac.addr, netdev->dev_addr);
2686                 ether_addr_copy(hw->mac.perm_addr, netdev->dev_addr);
2687         }
2688
2689         /* Enable dynamic interrupt throttling rates */
2690         adapter->rx_itr_setting = 1;
2691         adapter->tx_itr_setting = 1;
2692
2693         /* set default ring sizes */
2694         adapter->tx_ring_count = IXGBEVF_DEFAULT_TXD;
2695         adapter->rx_ring_count = IXGBEVF_DEFAULT_RXD;
2696
2697         set_bit(__IXGBEVF_DOWN, &adapter->state);
2698         return 0;
2699
2700 out:
2701         return err;
2702 }
2703
2704 #define UPDATE_VF_COUNTER_32bit(reg, last_counter, counter)     \
2705         {                                                       \
2706                 u32 current_counter = IXGBE_READ_REG(hw, reg);  \
2707                 if (current_counter < last_counter)             \
2708                         counter += 0x100000000LL;               \
2709                 last_counter = current_counter;                 \
2710                 counter &= 0xFFFFFFFF00000000LL;                \
2711                 counter |= current_counter;                     \
2712         }
2713
2714 #define UPDATE_VF_COUNTER_36bit(reg_lsb, reg_msb, last_counter, counter) \
2715         {                                                                \
2716                 u64 current_counter_lsb = IXGBE_READ_REG(hw, reg_lsb);   \
2717                 u64 current_counter_msb = IXGBE_READ_REG(hw, reg_msb);   \
2718                 u64 current_counter = (current_counter_msb << 32) |      \
2719                         current_counter_lsb;                             \
2720                 if (current_counter < last_counter)                      \
2721                         counter += 0x1000000000LL;                       \
2722                 last_counter = current_counter;                          \
2723                 counter &= 0xFFFFFFF000000000LL;                         \
2724                 counter |= current_counter;                              \
2725         }
2726 /**
2727  * ixgbevf_update_stats - Update the board statistics counters.
2728  * @adapter: board private structure
2729  **/
2730 void ixgbevf_update_stats(struct ixgbevf_adapter *adapter)
2731 {
2732         struct ixgbe_hw *hw = &adapter->hw;
2733         int i;
2734
2735         if (test_bit(__IXGBEVF_DOWN, &adapter->state) ||
2736             test_bit(__IXGBEVF_RESETTING, &adapter->state))
2737                 return;
2738
2739         UPDATE_VF_COUNTER_32bit(IXGBE_VFGPRC, adapter->stats.last_vfgprc,
2740                                 adapter->stats.vfgprc);
2741         UPDATE_VF_COUNTER_32bit(IXGBE_VFGPTC, adapter->stats.last_vfgptc,
2742                                 adapter->stats.vfgptc);
2743         UPDATE_VF_COUNTER_36bit(IXGBE_VFGORC_LSB, IXGBE_VFGORC_MSB,
2744                                 adapter->stats.last_vfgorc,
2745                                 adapter->stats.vfgorc);
2746         UPDATE_VF_COUNTER_36bit(IXGBE_VFGOTC_LSB, IXGBE_VFGOTC_MSB,
2747                                 adapter->stats.last_vfgotc,
2748                                 adapter->stats.vfgotc);
2749         UPDATE_VF_COUNTER_32bit(IXGBE_VFMPRC, adapter->stats.last_vfmprc,
2750                                 adapter->stats.vfmprc);
2751
2752         for (i = 0;  i  < adapter->num_rx_queues;  i++) {
2753                 adapter->hw_csum_rx_error +=
2754                         adapter->rx_ring[i]->hw_csum_rx_error;
2755                 adapter->rx_ring[i]->hw_csum_rx_error = 0;
2756         }
2757 }
2758
2759 /**
2760  * ixgbevf_service_timer - Timer Call-back
2761  * @data: pointer to adapter cast into an unsigned long
2762  **/
2763 static void ixgbevf_service_timer(unsigned long data)
2764 {
2765         struct ixgbevf_adapter *adapter = (struct ixgbevf_adapter *)data;
2766
2767         /* Reset the timer */
2768         mod_timer(&adapter->service_timer, (HZ * 2) + jiffies);
2769
2770         ixgbevf_service_event_schedule(adapter);
2771 }
2772
2773 static void ixgbevf_reset_subtask(struct ixgbevf_adapter *adapter)
2774 {
2775         if (!test_and_clear_bit(__IXGBEVF_RESET_REQUESTED, &adapter->state))
2776                 return;
2777
2778         /* If we're already down or resetting, just bail */
2779         if (test_bit(__IXGBEVF_DOWN, &adapter->state) ||
2780             test_bit(__IXGBEVF_REMOVING, &adapter->state) ||
2781             test_bit(__IXGBEVF_RESETTING, &adapter->state))
2782                 return;
2783
2784         adapter->tx_timeout_count++;
2785
2786         rtnl_lock();
2787         ixgbevf_reinit_locked(adapter);
2788         rtnl_unlock();
2789 }
2790
2791 /**
2792  * ixgbevf_check_hang_subtask - check for hung queues and dropped interrupts
2793  * @adapter: pointer to the device adapter structure
2794  *
2795  * This function serves two purposes.  First it strobes the interrupt lines
2796  * in order to make certain interrupts are occurring.  Secondly it sets the
2797  * bits needed to check for TX hangs.  As a result we should immediately
2798  * determine if a hang has occurred.
2799  **/
2800 static void ixgbevf_check_hang_subtask(struct ixgbevf_adapter *adapter)
2801 {
2802         struct ixgbe_hw *hw = &adapter->hw;
2803         u32 eics = 0;
2804         int i;
2805
2806         /* If we're down or resetting, just bail */
2807         if (test_bit(__IXGBEVF_DOWN, &adapter->state) ||
2808             test_bit(__IXGBEVF_RESETTING, &adapter->state))
2809                 return;
2810
2811         /* Force detection of hung controller */
2812         if (netif_carrier_ok(adapter->netdev)) {
2813                 for (i = 0; i < adapter->num_tx_queues; i++)
2814                         set_check_for_tx_hang(adapter->tx_ring[i]);
2815         }
2816
2817         /* get one bit for every active Tx/Rx interrupt vector */
2818         for (i = 0; i < adapter->num_msix_vectors - NON_Q_VECTORS; i++) {
2819                 struct ixgbevf_q_vector *qv = adapter->q_vector[i];
2820
2821                 if (qv->rx.ring || qv->tx.ring)
2822                         eics |= BIT(i);
2823         }
2824
2825         /* Cause software interrupt to ensure rings are cleaned */
2826         IXGBE_WRITE_REG(hw, IXGBE_VTEICS, eics);
2827 }
2828
2829 /**
2830  * ixgbevf_watchdog_update_link - update the link status
2831  * @adapter: pointer to the device adapter structure
2832  **/
2833 static void ixgbevf_watchdog_update_link(struct ixgbevf_adapter *adapter)
2834 {
2835         struct ixgbe_hw *hw = &adapter->hw;
2836         u32 link_speed = adapter->link_speed;
2837         bool link_up = adapter->link_up;
2838         s32 err;
2839
2840         spin_lock_bh(&adapter->mbx_lock);
2841
2842         err = hw->mac.ops.check_link(hw, &link_speed, &link_up, false);
2843
2844         spin_unlock_bh(&adapter->mbx_lock);
2845
2846         /* if check for link returns error we will need to reset */
2847         if (err && time_after(jiffies, adapter->last_reset + (10 * HZ))) {
2848                 set_bit(__IXGBEVF_RESET_REQUESTED, &adapter->state);
2849                 link_up = false;
2850         }
2851
2852         adapter->link_up = link_up;
2853         adapter->link_speed = link_speed;
2854 }
2855
2856 /**
2857  * ixgbevf_watchdog_link_is_up - update netif_carrier status and
2858  *                               print link up message
2859  * @adapter: pointer to the device adapter structure
2860  **/
2861 static void ixgbevf_watchdog_link_is_up(struct ixgbevf_adapter *adapter)
2862 {
2863         struct net_device *netdev = adapter->netdev;
2864
2865         /* only continue if link was previously down */
2866         if (netif_carrier_ok(netdev))
2867                 return;
2868
2869         dev_info(&adapter->pdev->dev, "NIC Link is Up %s\n",
2870                  (adapter->link_speed == IXGBE_LINK_SPEED_10GB_FULL) ?
2871                  "10 Gbps" :
2872                  (adapter->link_speed == IXGBE_LINK_SPEED_1GB_FULL) ?
2873                  "1 Gbps" :
2874                  (adapter->link_speed == IXGBE_LINK_SPEED_100_FULL) ?
2875                  "100 Mbps" :
2876                  "unknown speed");
2877
2878         netif_carrier_on(netdev);
2879 }
2880
2881 /**
2882  * ixgbevf_watchdog_link_is_down - update netif_carrier status and
2883  *                                 print link down message
2884  * @adapter: pointer to the adapter structure
2885  **/
2886 static void ixgbevf_watchdog_link_is_down(struct ixgbevf_adapter *adapter)
2887 {
2888         struct net_device *netdev = adapter->netdev;
2889
2890         adapter->link_speed = 0;
2891
2892         /* only continue if link was up previously */
2893         if (!netif_carrier_ok(netdev))
2894                 return;
2895
2896         dev_info(&adapter->pdev->dev, "NIC Link is Down\n");
2897
2898         netif_carrier_off(netdev);
2899 }
2900
2901 /**
2902  * ixgbevf_watchdog_subtask - worker thread to bring link up
2903  * @work: pointer to work_struct containing our data
2904  **/
2905 static void ixgbevf_watchdog_subtask(struct ixgbevf_adapter *adapter)
2906 {
2907         /* if interface is down do nothing */
2908         if (test_bit(__IXGBEVF_DOWN, &adapter->state) ||
2909             test_bit(__IXGBEVF_RESETTING, &adapter->state))
2910                 return;
2911
2912         ixgbevf_watchdog_update_link(adapter);
2913
2914         if (adapter->link_up)
2915                 ixgbevf_watchdog_link_is_up(adapter);
2916         else
2917                 ixgbevf_watchdog_link_is_down(adapter);
2918
2919         ixgbevf_update_stats(adapter);
2920 }
2921
2922 /**
2923  * ixgbevf_service_task - manages and runs subtasks
2924  * @work: pointer to work_struct containing our data
2925  **/
2926 static void ixgbevf_service_task(struct work_struct *work)
2927 {
2928         struct ixgbevf_adapter *adapter = container_of(work,
2929                                                        struct ixgbevf_adapter,
2930                                                        service_task);
2931         struct ixgbe_hw *hw = &adapter->hw;
2932
2933         if (IXGBE_REMOVED(hw->hw_addr)) {
2934                 if (!test_bit(__IXGBEVF_DOWN, &adapter->state)) {
2935                         rtnl_lock();
2936                         ixgbevf_down(adapter);
2937                         rtnl_unlock();
2938                 }
2939                 return;
2940         }
2941
2942         ixgbevf_queue_reset_subtask(adapter);
2943         ixgbevf_reset_subtask(adapter);
2944         ixgbevf_watchdog_subtask(adapter);
2945         ixgbevf_check_hang_subtask(adapter);
2946
2947         ixgbevf_service_event_complete(adapter);
2948 }
2949
2950 /**
2951  * ixgbevf_free_tx_resources - Free Tx Resources per Queue
2952  * @tx_ring: Tx descriptor ring for a specific queue
2953  *
2954  * Free all transmit software resources
2955  **/
2956 void ixgbevf_free_tx_resources(struct ixgbevf_ring *tx_ring)
2957 {
2958         ixgbevf_clean_tx_ring(tx_ring);
2959
2960         vfree(tx_ring->tx_buffer_info);
2961         tx_ring->tx_buffer_info = NULL;
2962
2963         /* if not set, then don't free */
2964         if (!tx_ring->desc)
2965                 return;
2966
2967         dma_free_coherent(tx_ring->dev, tx_ring->size, tx_ring->desc,
2968                           tx_ring->dma);
2969
2970         tx_ring->desc = NULL;
2971 }
2972
2973 /**
2974  * ixgbevf_free_all_tx_resources - Free Tx Resources for All Queues
2975  * @adapter: board private structure
2976  *
2977  * Free all transmit software resources
2978  **/
2979 static void ixgbevf_free_all_tx_resources(struct ixgbevf_adapter *adapter)
2980 {
2981         int i;
2982
2983         for (i = 0; i < adapter->num_tx_queues; i++)
2984                 if (adapter->tx_ring[i]->desc)
2985                         ixgbevf_free_tx_resources(adapter->tx_ring[i]);
2986 }
2987
2988 /**
2989  * ixgbevf_setup_tx_resources - allocate Tx resources (Descriptors)
2990  * @tx_ring: Tx descriptor ring (for a specific queue) to setup
2991  *
2992  * Return 0 on success, negative on failure
2993  **/
2994 int ixgbevf_setup_tx_resources(struct ixgbevf_ring *tx_ring)
2995 {
2996         struct ixgbevf_adapter *adapter = netdev_priv(tx_ring->netdev);
2997         int size;
2998
2999         size = sizeof(struct ixgbevf_tx_buffer) * tx_ring->count;
3000         tx_ring->tx_buffer_info = vzalloc(size);
3001         if (!tx_ring->tx_buffer_info)
3002                 goto err;
3003
3004         /* round up to nearest 4K */
3005         tx_ring->size = tx_ring->count * sizeof(union ixgbe_adv_tx_desc);
3006         tx_ring->size = ALIGN(tx_ring->size, 4096);
3007
3008         tx_ring->desc = dma_alloc_coherent(tx_ring->dev, tx_ring->size,
3009                                            &tx_ring->dma, GFP_KERNEL);
3010         if (!tx_ring->desc)
3011                 goto err;
3012
3013         return 0;
3014
3015 err:
3016         vfree(tx_ring->tx_buffer_info);
3017         tx_ring->tx_buffer_info = NULL;
3018         hw_dbg(&adapter->hw, "Unable to allocate memory for the transmit descriptor ring\n");
3019         return -ENOMEM;
3020 }
3021
3022 /**
3023  * ixgbevf_setup_all_tx_resources - allocate all queues Tx resources
3024  * @adapter: board private structure
3025  *
3026  * If this function returns with an error, then it's possible one or
3027  * more of the rings is populated (while the rest are not).  It is the
3028  * callers duty to clean those orphaned rings.
3029  *
3030  * Return 0 on success, negative on failure
3031  **/
3032 static int ixgbevf_setup_all_tx_resources(struct ixgbevf_adapter *adapter)
3033 {
3034         int i, err = 0;
3035
3036         for (i = 0; i < adapter->num_tx_queues; i++) {
3037                 err = ixgbevf_setup_tx_resources(adapter->tx_ring[i]);
3038                 if (!err)
3039                         continue;
3040                 hw_dbg(&adapter->hw, "Allocation for Tx Queue %u failed\n", i);
3041                 break;
3042         }
3043
3044         return err;
3045 }
3046
3047 /**
3048  * ixgbevf_setup_rx_resources - allocate Rx resources (Descriptors)
3049  * @rx_ring: Rx descriptor ring (for a specific queue) to setup
3050  *
3051  * Returns 0 on success, negative on failure
3052  **/
3053 int ixgbevf_setup_rx_resources(struct ixgbevf_ring *rx_ring)
3054 {
3055         int size;
3056
3057         size = sizeof(struct ixgbevf_rx_buffer) * rx_ring->count;
3058         rx_ring->rx_buffer_info = vzalloc(size);
3059         if (!rx_ring->rx_buffer_info)
3060                 goto err;
3061
3062         /* Round up to nearest 4K */
3063         rx_ring->size = rx_ring->count * sizeof(union ixgbe_adv_rx_desc);
3064         rx_ring->size = ALIGN(rx_ring->size, 4096);
3065
3066         rx_ring->desc = dma_alloc_coherent(rx_ring->dev, rx_ring->size,
3067                                            &rx_ring->dma, GFP_KERNEL);
3068
3069         if (!rx_ring->desc)
3070                 goto err;
3071
3072         return 0;
3073 err:
3074         vfree(rx_ring->rx_buffer_info);
3075         rx_ring->rx_buffer_info = NULL;
3076         dev_err(rx_ring->dev, "Unable to allocate memory for the Rx descriptor ring\n");
3077         return -ENOMEM;
3078 }
3079
3080 /**
3081  * ixgbevf_setup_all_rx_resources - allocate all queues Rx resources
3082  * @adapter: board private structure
3083  *
3084  * If this function returns with an error, then it's possible one or
3085  * more of the rings is populated (while the rest are not).  It is the
3086  * callers duty to clean those orphaned rings.
3087  *
3088  * Return 0 on success, negative on failure
3089  **/
3090 static int ixgbevf_setup_all_rx_resources(struct ixgbevf_adapter *adapter)
3091 {
3092         int i, err = 0;
3093
3094         for (i = 0; i < adapter->num_rx_queues; i++) {
3095                 err = ixgbevf_setup_rx_resources(adapter->rx_ring[i]);
3096                 if (!err)
3097                         continue;
3098                 hw_dbg(&adapter->hw, "Allocation for Rx Queue %u failed\n", i);
3099                 break;
3100         }
3101         return err;
3102 }
3103
3104 /**
3105  * ixgbevf_free_rx_resources - Free Rx Resources
3106  * @rx_ring: ring to clean the resources from
3107  *
3108  * Free all receive software resources
3109  **/
3110 void ixgbevf_free_rx_resources(struct ixgbevf_ring *rx_ring)
3111 {
3112         ixgbevf_clean_rx_ring(rx_ring);
3113
3114         vfree(rx_ring->rx_buffer_info);
3115         rx_ring->rx_buffer_info = NULL;
3116
3117         dma_free_coherent(rx_ring->dev, rx_ring->size, rx_ring->desc,
3118                           rx_ring->dma);
3119
3120         rx_ring->desc = NULL;
3121 }
3122
3123 /**
3124  * ixgbevf_free_all_rx_resources - Free Rx Resources for All Queues
3125  * @adapter: board private structure
3126  *
3127  * Free all receive software resources
3128  **/
3129 static void ixgbevf_free_all_rx_resources(struct ixgbevf_adapter *adapter)
3130 {
3131         int i;
3132
3133         for (i = 0; i < adapter->num_rx_queues; i++)
3134                 if (adapter->rx_ring[i]->desc)
3135                         ixgbevf_free_rx_resources(adapter->rx_ring[i]);
3136 }
3137
3138 /**
3139  * ixgbevf_open - Called when a network interface is made active
3140  * @netdev: network interface device structure
3141  *
3142  * Returns 0 on success, negative value on failure
3143  *
3144  * The open entry point is called when a network interface is made
3145  * active by the system (IFF_UP).  At this point all resources needed
3146  * for transmit and receive operations are allocated, the interrupt
3147  * handler is registered with the OS, the watchdog timer is started,
3148  * and the stack is notified that the interface is ready.
3149  **/
3150 int ixgbevf_open(struct net_device *netdev)
3151 {
3152         struct ixgbevf_adapter *adapter = netdev_priv(netdev);
3153         struct ixgbe_hw *hw = &adapter->hw;
3154         int err;
3155
3156         /* A previous failure to open the device because of a lack of
3157          * available MSIX vector resources may have reset the number
3158          * of msix vectors variable to zero.  The only way to recover
3159          * is to unload/reload the driver and hope that the system has
3160          * been able to recover some MSIX vector resources.
3161          */
3162         if (!adapter->num_msix_vectors)
3163                 return -ENOMEM;
3164
3165         if (hw->adapter_stopped) {
3166                 ixgbevf_reset(adapter);
3167                 /* if adapter is still stopped then PF isn't up and
3168                  * the VF can't start.
3169                  */
3170                 if (hw->adapter_stopped) {
3171                         err = IXGBE_ERR_MBX;
3172                         pr_err("Unable to start - perhaps the PF Driver isn't up yet\n");
3173                         goto err_setup_reset;
3174                 }
3175         }
3176
3177         /* disallow open during test */
3178         if (test_bit(__IXGBEVF_TESTING, &adapter->state))
3179                 return -EBUSY;
3180
3181         netif_carrier_off(netdev);
3182
3183         /* allocate transmit descriptors */
3184         err = ixgbevf_setup_all_tx_resources(adapter);
3185         if (err)
3186                 goto err_setup_tx;
3187
3188         /* allocate receive descriptors */
3189         err = ixgbevf_setup_all_rx_resources(adapter);
3190         if (err)
3191                 goto err_setup_rx;
3192
3193         ixgbevf_configure(adapter);
3194
3195         /* Map the Tx/Rx rings to the vectors we were allotted.
3196          * if request_irq will be called in this function map_rings
3197          * must be called *before* up_complete
3198          */
3199         ixgbevf_map_rings_to_vectors(adapter);
3200
3201         err = ixgbevf_request_irq(adapter);
3202         if (err)
3203                 goto err_req_irq;
3204
3205         ixgbevf_up_complete(adapter);
3206
3207         return 0;
3208
3209 err_req_irq:
3210         ixgbevf_down(adapter);
3211 err_setup_rx:
3212         ixgbevf_free_all_rx_resources(adapter);
3213 err_setup_tx:
3214         ixgbevf_free_all_tx_resources(adapter);
3215         ixgbevf_reset(adapter);
3216
3217 err_setup_reset:
3218
3219         return err;
3220 }
3221
3222 /**
3223  * ixgbevf_close - Disables a network interface
3224  * @netdev: network interface device structure
3225  *
3226  * Returns 0, this is not allowed to fail
3227  *
3228  * The close entry point is called when an interface is de-activated
3229  * by the OS.  The hardware is still under the drivers control, but
3230  * needs to be disabled.  A global MAC reset is issued to stop the
3231  * hardware, and all transmit and receive resources are freed.
3232  **/
3233 int ixgbevf_close(struct net_device *netdev)
3234 {
3235         struct ixgbevf_adapter *adapter = netdev_priv(netdev);
3236
3237         ixgbevf_down(adapter);
3238         ixgbevf_free_irq(adapter);
3239
3240         ixgbevf_free_all_tx_resources(adapter);
3241         ixgbevf_free_all_rx_resources(adapter);
3242
3243         return 0;
3244 }
3245
3246 static void ixgbevf_queue_reset_subtask(struct ixgbevf_adapter *adapter)
3247 {
3248         struct net_device *dev = adapter->netdev;
3249
3250         if (!test_and_clear_bit(__IXGBEVF_QUEUE_RESET_REQUESTED,
3251                                 &adapter->state))
3252                 return;
3253
3254         /* if interface is down do nothing */
3255         if (test_bit(__IXGBEVF_DOWN, &adapter->state) ||
3256             test_bit(__IXGBEVF_RESETTING, &adapter->state))
3257                 return;
3258
3259         /* Hardware has to reinitialize queues and interrupts to
3260          * match packet buffer alignment. Unfortunately, the
3261          * hardware is not flexible enough to do this dynamically.
3262          */
3263         if (netif_running(dev))
3264                 ixgbevf_close(dev);
3265
3266         ixgbevf_clear_interrupt_scheme(adapter);
3267         ixgbevf_init_interrupt_scheme(adapter);
3268
3269         if (netif_running(dev))
3270                 ixgbevf_open(dev);
3271 }
3272
3273 static void ixgbevf_tx_ctxtdesc(struct ixgbevf_ring *tx_ring,
3274                                 u32 vlan_macip_lens, u32 type_tucmd,
3275                                 u32 mss_l4len_idx)
3276 {
3277         struct ixgbe_adv_tx_context_desc *context_desc;
3278         u16 i = tx_ring->next_to_use;
3279
3280         context_desc = IXGBEVF_TX_CTXTDESC(tx_ring, i);
3281
3282         i++;
3283         tx_ring->next_to_use = (i < tx_ring->count) ? i : 0;
3284
3285         /* set bits to identify this as an advanced context descriptor */
3286         type_tucmd |= IXGBE_TXD_CMD_DEXT | IXGBE_ADVTXD_DTYP_CTXT;
3287
3288         context_desc->vlan_macip_lens   = cpu_to_le32(vlan_macip_lens);
3289         context_desc->seqnum_seed       = 0;
3290         context_desc->type_tucmd_mlhl   = cpu_to_le32(type_tucmd);
3291         context_desc->mss_l4len_idx     = cpu_to_le32(mss_l4len_idx);
3292 }
3293
3294 static int ixgbevf_tso(struct ixgbevf_ring *tx_ring,
3295                        struct ixgbevf_tx_buffer *first,
3296                        u8 *hdr_len)
3297 {
3298         u32 vlan_macip_lens, type_tucmd, mss_l4len_idx;
3299         struct sk_buff *skb = first->skb;
3300         union {
3301                 struct iphdr *v4;
3302                 struct ipv6hdr *v6;
3303                 unsigned char *hdr;
3304         } ip;
3305         union {
3306                 struct tcphdr *tcp;
3307                 unsigned char *hdr;
3308         } l4;
3309         u32 paylen, l4_offset;
3310         int err;
3311
3312         if (skb->ip_summed != CHECKSUM_PARTIAL)
3313                 return 0;
3314
3315         if (!skb_is_gso(skb))
3316                 return 0;
3317
3318         err = skb_cow_head(skb, 0);
3319         if (err < 0)
3320                 return err;
3321
3322         ip.hdr = skb_network_header(skb);
3323         l4.hdr = skb_checksum_start(skb);
3324
3325         /* ADV DTYP TUCMD MKRLOC/ISCSIHEDLEN */
3326         type_tucmd = IXGBE_ADVTXD_TUCMD_L4T_TCP;
3327
3328         /* initialize outer IP header fields */
3329         if (ip.v4->version == 4) {
3330                 /* IP header will have to cancel out any data that
3331                  * is not a part of the outer IP header
3332                  */
3333                 ip.v4->check = csum_fold(csum_add(lco_csum(skb),
3334                                                   csum_unfold(l4.tcp->check)));
3335                 type_tucmd |= IXGBE_ADVTXD_TUCMD_IPV4;
3336
3337                 ip.v4->tot_len = 0;
3338                 first->tx_flags |= IXGBE_TX_FLAGS_TSO |
3339                                    IXGBE_TX_FLAGS_CSUM |
3340                                    IXGBE_TX_FLAGS_IPV4;
3341         } else {
3342                 ip.v6->payload_len = 0;
3343                 first->tx_flags |= IXGBE_TX_FLAGS_TSO |
3344                                    IXGBE_TX_FLAGS_CSUM;
3345         }
3346
3347         /* determine offset of inner transport header */
3348         l4_offset = l4.hdr - skb->data;
3349
3350         /* compute length of segmentation header */
3351         *hdr_len = (l4.tcp->doff * 4) + l4_offset;
3352
3353         /* remove payload length from inner checksum */
3354         paylen = skb->len - l4_offset;
3355         csum_replace_by_diff(&l4.tcp->check, htonl(paylen));
3356
3357         /* update gso size and bytecount with header size */
3358         first->gso_segs = skb_shinfo(skb)->gso_segs;
3359         first->bytecount += (first->gso_segs - 1) * *hdr_len;
3360
3361         /* mss_l4len_id: use 1 as index for TSO */
3362         mss_l4len_idx = (*hdr_len - l4_offset) << IXGBE_ADVTXD_L4LEN_SHIFT;
3363         mss_l4len_idx |= skb_shinfo(skb)->gso_size << IXGBE_ADVTXD_MSS_SHIFT;
3364         mss_l4len_idx |= (1u << IXGBE_ADVTXD_IDX_SHIFT);
3365
3366         /* vlan_macip_lens: HEADLEN, MACLEN, VLAN tag */
3367         vlan_macip_lens = l4.hdr - ip.hdr;
3368         vlan_macip_lens |= (ip.hdr - skb->data) << IXGBE_ADVTXD_MACLEN_SHIFT;
3369         vlan_macip_lens |= first->tx_flags & IXGBE_TX_FLAGS_VLAN_MASK;
3370
3371         ixgbevf_tx_ctxtdesc(tx_ring, vlan_macip_lens,
3372                             type_tucmd, mss_l4len_idx);
3373
3374         return 1;
3375 }
3376
3377 static inline bool ixgbevf_ipv6_csum_is_sctp(struct sk_buff *skb)
3378 {
3379         unsigned int offset = 0;
3380
3381         ipv6_find_hdr(skb, &offset, IPPROTO_SCTP, NULL, NULL);
3382
3383         return offset == skb_checksum_start_offset(skb);
3384 }
3385
3386 static void ixgbevf_tx_csum(struct ixgbevf_ring *tx_ring,
3387                             struct ixgbevf_tx_buffer *first)
3388 {
3389         struct sk_buff *skb = first->skb;
3390         u32 vlan_macip_lens = 0;
3391         u32 type_tucmd = 0;
3392
3393         if (skb->ip_summed != CHECKSUM_PARTIAL)
3394                 goto no_csum;
3395
3396         switch (skb->csum_offset) {
3397         case offsetof(struct tcphdr, check):
3398                 type_tucmd = IXGBE_ADVTXD_TUCMD_L4T_TCP;
3399                 /* fall through */
3400         case offsetof(struct udphdr, check):
3401                 break;
3402         case offsetof(struct sctphdr, checksum):
3403                 /* validate that this is actually an SCTP request */
3404                 if (((first->protocol == htons(ETH_P_IP)) &&
3405                      (ip_hdr(skb)->protocol == IPPROTO_SCTP)) ||
3406                     ((first->protocol == htons(ETH_P_IPV6)) &&
3407                      ixgbevf_ipv6_csum_is_sctp(skb))) {
3408                         type_tucmd = IXGBE_ADVTXD_TUCMD_L4T_SCTP;
3409                         break;
3410                 }
3411                 /* fall through */
3412         default:
3413                 skb_checksum_help(skb);
3414                 goto no_csum;
3415         }
3416         /* update TX checksum flag */
3417         first->tx_flags |= IXGBE_TX_FLAGS_CSUM;
3418         vlan_macip_lens = skb_checksum_start_offset(skb) -
3419                           skb_network_offset(skb);
3420 no_csum:
3421         /* vlan_macip_lens: MACLEN, VLAN tag */
3422         vlan_macip_lens |= skb_network_offset(skb) << IXGBE_ADVTXD_MACLEN_SHIFT;
3423         vlan_macip_lens |= first->tx_flags & IXGBE_TX_FLAGS_VLAN_MASK;
3424
3425         ixgbevf_tx_ctxtdesc(tx_ring, vlan_macip_lens, type_tucmd, 0);
3426 }
3427
3428 static __le32 ixgbevf_tx_cmd_type(u32 tx_flags)
3429 {
3430         /* set type for advanced descriptor with frame checksum insertion */
3431         __le32 cmd_type = cpu_to_le32(IXGBE_ADVTXD_DTYP_DATA |
3432                                       IXGBE_ADVTXD_DCMD_IFCS |
3433                                       IXGBE_ADVTXD_DCMD_DEXT);
3434
3435         /* set HW VLAN bit if VLAN is present */
3436         if (tx_flags & IXGBE_TX_FLAGS_VLAN)
3437                 cmd_type |= cpu_to_le32(IXGBE_ADVTXD_DCMD_VLE);
3438
3439         /* set segmentation enable bits for TSO/FSO */
3440         if (tx_flags & IXGBE_TX_FLAGS_TSO)
3441                 cmd_type |= cpu_to_le32(IXGBE_ADVTXD_DCMD_TSE);
3442
3443         return cmd_type;
3444 }
3445
3446 static void ixgbevf_tx_olinfo_status(union ixgbe_adv_tx_desc *tx_desc,
3447                                      u32 tx_flags, unsigned int paylen)
3448 {
3449         __le32 olinfo_status = cpu_to_le32(paylen << IXGBE_ADVTXD_PAYLEN_SHIFT);
3450
3451         /* enable L4 checksum for TSO and TX checksum offload */
3452         if (tx_flags & IXGBE_TX_FLAGS_CSUM)
3453                 olinfo_status |= cpu_to_le32(IXGBE_ADVTXD_POPTS_TXSM);
3454
3455         /* enble IPv4 checksum for TSO */
3456         if (tx_flags & IXGBE_TX_FLAGS_IPV4)
3457                 olinfo_status |= cpu_to_le32(IXGBE_ADVTXD_POPTS_IXSM);
3458
3459         /* use index 1 context for TSO/FSO/FCOE */
3460         if (tx_flags & IXGBE_TX_FLAGS_TSO)
3461                 olinfo_status |= cpu_to_le32(1u << IXGBE_ADVTXD_IDX_SHIFT);
3462
3463         /* Check Context must be set if Tx switch is enabled, which it
3464          * always is for case where virtual functions are running
3465          */
3466         olinfo_status |= cpu_to_le32(IXGBE_ADVTXD_CC);
3467
3468         tx_desc->read.olinfo_status = olinfo_status;
3469 }
3470
3471 static void ixgbevf_tx_map(struct ixgbevf_ring *tx_ring,
3472                            struct ixgbevf_tx_buffer *first,
3473                            const u8 hdr_len)
3474 {
3475         dma_addr_t dma;
3476         struct sk_buff *skb = first->skb;
3477         struct ixgbevf_tx_buffer *tx_buffer;
3478         union ixgbe_adv_tx_desc *tx_desc;
3479         struct skb_frag_struct *frag = &skb_shinfo(skb)->frags[0];
3480         unsigned int data_len = skb->data_len;
3481         unsigned int size = skb_headlen(skb);
3482         unsigned int paylen = skb->len - hdr_len;
3483         u32 tx_flags = first->tx_flags;
3484         __le32 cmd_type;
3485         u16 i = tx_ring->next_to_use;
3486
3487         tx_desc = IXGBEVF_TX_DESC(tx_ring, i);
3488
3489         ixgbevf_tx_olinfo_status(tx_desc, tx_flags, paylen);
3490         cmd_type = ixgbevf_tx_cmd_type(tx_flags);
3491
3492         dma = dma_map_single(tx_ring->dev, skb->data, size, DMA_TO_DEVICE);
3493         if (dma_mapping_error(tx_ring->dev, dma))
3494                 goto dma_error;
3495
3496         /* record length, and DMA address */
3497         dma_unmap_len_set(first, len, size);
3498         dma_unmap_addr_set(first, dma, dma);
3499
3500         tx_desc->read.buffer_addr = cpu_to_le64(dma);
3501
3502         for (;;) {
3503                 while (unlikely(size > IXGBE_MAX_DATA_PER_TXD)) {
3504                         tx_desc->read.cmd_type_len =
3505                                 cmd_type | cpu_to_le32(IXGBE_MAX_DATA_PER_TXD);
3506
3507                         i++;
3508                         tx_desc++;
3509                         if (i == tx_ring->count) {
3510                                 tx_desc = IXGBEVF_TX_DESC(tx_ring, 0);
3511                                 i = 0;
3512                         }
3513
3514                         dma += IXGBE_MAX_DATA_PER_TXD;
3515                         size -= IXGBE_MAX_DATA_PER_TXD;
3516
3517                         tx_desc->read.buffer_addr = cpu_to_le64(dma);
3518                         tx_desc->read.olinfo_status = 0;
3519                 }
3520
3521                 if (likely(!data_len))
3522                         break;
3523
3524                 tx_desc->read.cmd_type_len = cmd_type | cpu_to_le32(size);
3525
3526                 i++;
3527                 tx_desc++;
3528                 if (i == tx_ring->count) {
3529                         tx_desc = IXGBEVF_TX_DESC(tx_ring, 0);
3530                         i = 0;
3531                 }
3532
3533                 size = skb_frag_size(frag);
3534                 data_len -= size;
3535
3536                 dma = skb_frag_dma_map(tx_ring->dev, frag, 0, size,
3537                                        DMA_TO_DEVICE);
3538                 if (dma_mapping_error(tx_ring->dev, dma))
3539                         goto dma_error;
3540
3541                 tx_buffer = &tx_ring->tx_buffer_info[i];
3542                 dma_unmap_len_set(tx_buffer, len, size);
3543                 dma_unmap_addr_set(tx_buffer, dma, dma);
3544
3545                 tx_desc->read.buffer_addr = cpu_to_le64(dma);
3546                 tx_desc->read.olinfo_status = 0;
3547
3548                 frag++;
3549         }
3550
3551         /* write last descriptor with RS and EOP bits */
3552         cmd_type |= cpu_to_le32(size) | cpu_to_le32(IXGBE_TXD_CMD);
3553         tx_desc->read.cmd_type_len = cmd_type;
3554
3555         /* set the timestamp */
3556         first->time_stamp = jiffies;
3557
3558         /* Force memory writes to complete before letting h/w know there
3559          * are new descriptors to fetch.  (Only applicable for weak-ordered
3560          * memory model archs, such as IA-64).
3561          *
3562          * We also need this memory barrier (wmb) to make certain all of the
3563          * status bits have been updated before next_to_watch is written.
3564          */
3565         wmb();
3566
3567         /* set next_to_watch value indicating a packet is present */
3568         first->next_to_watch = tx_desc;
3569
3570         i++;
3571         if (i == tx_ring->count)
3572                 i = 0;
3573
3574         tx_ring->next_to_use = i;
3575
3576         /* notify HW of packet */
3577         ixgbevf_write_tail(tx_ring, i);
3578
3579         return;
3580 dma_error:
3581         dev_err(tx_ring->dev, "TX DMA map failed\n");
3582
3583         /* clear dma mappings for failed tx_buffer_info map */
3584         for (;;) {
3585                 tx_buffer = &tx_ring->tx_buffer_info[i];
3586                 ixgbevf_unmap_and_free_tx_resource(tx_ring, tx_buffer);
3587                 if (tx_buffer == first)
3588                         break;
3589                 if (i == 0)
3590                         i = tx_ring->count;
3591                 i--;
3592         }
3593
3594         tx_ring->next_to_use = i;
3595 }
3596
3597 static int __ixgbevf_maybe_stop_tx(struct ixgbevf_ring *tx_ring, int size)
3598 {
3599         netif_stop_subqueue(tx_ring->netdev, tx_ring->queue_index);
3600         /* Herbert's original patch had:
3601          *  smp_mb__after_netif_stop_queue();
3602          * but since that doesn't exist yet, just open code it.
3603          */
3604         smp_mb();
3605
3606         /* We need to check again in a case another CPU has just
3607          * made room available.
3608          */
3609         if (likely(ixgbevf_desc_unused(tx_ring) < size))
3610                 return -EBUSY;
3611
3612         /* A reprieve! - use start_queue because it doesn't call schedule */
3613         netif_start_subqueue(tx_ring->netdev, tx_ring->queue_index);
3614         ++tx_ring->tx_stats.restart_queue;
3615
3616         return 0;
3617 }
3618
3619 static int ixgbevf_maybe_stop_tx(struct ixgbevf_ring *tx_ring, int size)
3620 {
3621         if (likely(ixgbevf_desc_unused(tx_ring) >= size))
3622                 return 0;
3623         return __ixgbevf_maybe_stop_tx(tx_ring, size);
3624 }
3625
3626 static int ixgbevf_xmit_frame(struct sk_buff *skb, struct net_device *netdev)
3627 {
3628         struct ixgbevf_adapter *adapter = netdev_priv(netdev);
3629         struct ixgbevf_tx_buffer *first;
3630         struct ixgbevf_ring *tx_ring;
3631         int tso;
3632         u32 tx_flags = 0;
3633         u16 count = TXD_USE_COUNT(skb_headlen(skb));
3634 #if PAGE_SIZE > IXGBE_MAX_DATA_PER_TXD
3635         unsigned short f;
3636 #endif
3637         u8 hdr_len = 0;
3638         u8 *dst_mac = skb_header_pointer(skb, 0, 0, NULL);
3639
3640         if (!dst_mac || is_link_local_ether_addr(dst_mac)) {
3641                 dev_kfree_skb_any(skb);
3642                 return NETDEV_TX_OK;
3643         }
3644
3645         tx_ring = adapter->tx_ring[skb->queue_mapping];
3646
3647         /* need: 1 descriptor per page * PAGE_SIZE/IXGBE_MAX_DATA_PER_TXD,
3648          *       + 1 desc for skb_headlen/IXGBE_MAX_DATA_PER_TXD,
3649          *       + 2 desc gap to keep tail from touching head,
3650          *       + 1 desc for context descriptor,
3651          * otherwise try next time
3652          */
3653 #if PAGE_SIZE > IXGBE_MAX_DATA_PER_TXD
3654         for (f = 0; f < skb_shinfo(skb)->nr_frags; f++)
3655                 count += TXD_USE_COUNT(skb_shinfo(skb)->frags[f].size);
3656 #else
3657         count += skb_shinfo(skb)->nr_frags;
3658 #endif
3659         if (ixgbevf_maybe_stop_tx(tx_ring, count + 3)) {
3660                 tx_ring->tx_stats.tx_busy++;
3661                 return NETDEV_TX_BUSY;
3662         }
3663
3664         /* record the location of the first descriptor for this packet */
3665         first = &tx_ring->tx_buffer_info[tx_ring->next_to_use];
3666         first->skb = skb;
3667         first->bytecount = skb->len;
3668         first->gso_segs = 1;
3669
3670         if (skb_vlan_tag_present(skb)) {
3671                 tx_flags |= skb_vlan_tag_get(skb);
3672                 tx_flags <<= IXGBE_TX_FLAGS_VLAN_SHIFT;
3673                 tx_flags |= IXGBE_TX_FLAGS_VLAN;
3674         }
3675
3676         /* record initial flags and protocol */
3677         first->tx_flags = tx_flags;
3678         first->protocol = vlan_get_protocol(skb);
3679
3680         tso = ixgbevf_tso(tx_ring, first, &hdr_len);
3681         if (tso < 0)
3682                 goto out_drop;
3683         else if (!tso)
3684                 ixgbevf_tx_csum(tx_ring, first);
3685
3686         ixgbevf_tx_map(tx_ring, first, hdr_len);
3687
3688         ixgbevf_maybe_stop_tx(tx_ring, DESC_NEEDED);
3689
3690         return NETDEV_TX_OK;
3691
3692 out_drop:
3693         dev_kfree_skb_any(first->skb);
3694         first->skb = NULL;
3695
3696         return NETDEV_TX_OK;
3697 }
3698
3699 /**
3700  * ixgbevf_set_mac - Change the Ethernet Address of the NIC
3701  * @netdev: network interface device structure
3702  * @p: pointer to an address structure
3703  *
3704  * Returns 0 on success, negative on failure
3705  **/
3706 static int ixgbevf_set_mac(struct net_device *netdev, void *p)
3707 {
3708         struct ixgbevf_adapter *adapter = netdev_priv(netdev);
3709         struct ixgbe_hw *hw = &adapter->hw;
3710         struct sockaddr *addr = p;
3711         int err;
3712
3713         if (!is_valid_ether_addr(addr->sa_data))
3714                 return -EADDRNOTAVAIL;
3715
3716         spin_lock_bh(&adapter->mbx_lock);
3717
3718         err = hw->mac.ops.set_rar(hw, 0, addr->sa_data, 0);
3719
3720         spin_unlock_bh(&adapter->mbx_lock);
3721
3722         if (err)
3723                 return -EPERM;
3724
3725         ether_addr_copy(hw->mac.addr, addr->sa_data);
3726         ether_addr_copy(netdev->dev_addr, addr->sa_data);
3727
3728         return 0;
3729 }
3730
3731 /**
3732  * ixgbevf_change_mtu - Change the Maximum Transfer Unit
3733  * @netdev: network interface device structure
3734  * @new_mtu: new value for maximum frame size
3735  *
3736  * Returns 0 on success, negative on failure
3737  **/
3738 static int ixgbevf_change_mtu(struct net_device *netdev, int new_mtu)
3739 {
3740         struct ixgbevf_adapter *adapter = netdev_priv(netdev);
3741         struct ixgbe_hw *hw = &adapter->hw;
3742         int max_frame = new_mtu + ETH_HLEN + ETH_FCS_LEN;
3743         int max_possible_frame = MAXIMUM_ETHERNET_VLAN_SIZE;
3744         int ret;
3745
3746         switch (adapter->hw.api_version) {
3747         case ixgbe_mbox_api_11:
3748         case ixgbe_mbox_api_12:
3749                 max_possible_frame = IXGBE_MAX_JUMBO_FRAME_SIZE;
3750                 break;
3751         default:
3752                 if (adapter->hw.mac.type != ixgbe_mac_82599_vf)
3753                         max_possible_frame = IXGBE_MAX_JUMBO_FRAME_SIZE;
3754                 break;
3755         }
3756
3757         /* MTU < 68 is an error and causes problems on some kernels */
3758         if ((new_mtu < 68) || (max_frame > max_possible_frame))
3759                 return -EINVAL;
3760
3761         /* notify the PF of our intent to use this size of frame */
3762         ret = hw->mac.ops.set_rlpml(hw, max_frame);
3763         if (ret)
3764                 return -EINVAL;
3765
3766         hw_dbg(hw, "changing MTU from %d to %d\n",
3767                netdev->mtu, new_mtu);
3768
3769         /* must set new MTU before calling down or up */
3770         netdev->mtu = new_mtu;
3771
3772         return 0;
3773 }
3774
3775 #ifdef CONFIG_NET_POLL_CONTROLLER
3776 /* Polling 'interrupt' - used by things like netconsole to send skbs
3777  * without having to re-enable interrupts. It's not called while
3778  * the interrupt routine is executing.
3779  */
3780 static void ixgbevf_netpoll(struct net_device *netdev)
3781 {
3782         struct ixgbevf_adapter *adapter = netdev_priv(netdev);
3783         int i;
3784
3785         /* if interface is down do nothing */
3786         if (test_bit(__IXGBEVF_DOWN, &adapter->state))
3787                 return;
3788         for (i = 0; i < adapter->num_rx_queues; i++)
3789                 ixgbevf_msix_clean_rings(0, adapter->q_vector[i]);
3790 }
3791 #endif /* CONFIG_NET_POLL_CONTROLLER */
3792
3793 static int ixgbevf_suspend(struct pci_dev *pdev, pm_message_t state)
3794 {
3795         struct net_device *netdev = pci_get_drvdata(pdev);
3796         struct ixgbevf_adapter *adapter = netdev_priv(netdev);
3797 #ifdef CONFIG_PM
3798         int retval = 0;
3799 #endif
3800
3801         netif_device_detach(netdev);
3802
3803         if (netif_running(netdev)) {
3804                 rtnl_lock();
3805                 ixgbevf_down(adapter);
3806                 ixgbevf_free_irq(adapter);
3807                 ixgbevf_free_all_tx_resources(adapter);
3808                 ixgbevf_free_all_rx_resources(adapter);
3809                 rtnl_unlock();
3810         }
3811
3812         ixgbevf_clear_interrupt_scheme(adapter);
3813
3814 #ifdef CONFIG_PM
3815         retval = pci_save_state(pdev);
3816         if (retval)
3817                 return retval;
3818
3819 #endif
3820         if (!test_and_set_bit(__IXGBEVF_DISABLED, &adapter->state))
3821                 pci_disable_device(pdev);
3822
3823         return 0;
3824 }
3825
3826 #ifdef CONFIG_PM
3827 static int ixgbevf_resume(struct pci_dev *pdev)
3828 {
3829         struct net_device *netdev = pci_get_drvdata(pdev);
3830         struct ixgbevf_adapter *adapter = netdev_priv(netdev);
3831         u32 err;
3832
3833         pci_restore_state(pdev);
3834         /* pci_restore_state clears dev->state_saved so call
3835          * pci_save_state to restore it.
3836          */
3837         pci_save_state(pdev);
3838
3839         err = pci_enable_device_mem(pdev);
3840         if (err) {
3841                 dev_err(&pdev->dev, "Cannot enable PCI device from suspend\n");
3842                 return err;
3843         }
3844         smp_mb__before_atomic();
3845         clear_bit(__IXGBEVF_DISABLED, &adapter->state);
3846         pci_set_master(pdev);
3847
3848         ixgbevf_reset(adapter);
3849
3850         rtnl_lock();
3851         err = ixgbevf_init_interrupt_scheme(adapter);
3852         rtnl_unlock();
3853         if (err) {
3854                 dev_err(&pdev->dev, "Cannot initialize interrupts\n");
3855                 return err;
3856         }
3857
3858         if (netif_running(netdev)) {
3859                 err = ixgbevf_open(netdev);
3860                 if (err)
3861                         return err;
3862         }
3863
3864         netif_device_attach(netdev);
3865
3866         return err;
3867 }
3868
3869 #endif /* CONFIG_PM */
3870 static void ixgbevf_shutdown(struct pci_dev *pdev)
3871 {
3872         ixgbevf_suspend(pdev, PMSG_SUSPEND);
3873 }
3874
3875 static struct rtnl_link_stats64 *ixgbevf_get_stats(struct net_device *netdev,
3876                                                 struct rtnl_link_stats64 *stats)
3877 {
3878         struct ixgbevf_adapter *adapter = netdev_priv(netdev);
3879         unsigned int start;
3880         u64 bytes, packets;
3881         const struct ixgbevf_ring *ring;
3882         int i;
3883
3884         ixgbevf_update_stats(adapter);
3885
3886         stats->multicast = adapter->stats.vfmprc - adapter->stats.base_vfmprc;
3887
3888         for (i = 0; i < adapter->num_rx_queues; i++) {
3889                 ring = adapter->rx_ring[i];
3890                 do {
3891                         start = u64_stats_fetch_begin_irq(&ring->syncp);
3892                         bytes = ring->stats.bytes;
3893                         packets = ring->stats.packets;
3894                 } while (u64_stats_fetch_retry_irq(&ring->syncp, start));
3895                 stats->rx_bytes += bytes;
3896                 stats->rx_packets += packets;
3897         }
3898
3899         for (i = 0; i < adapter->num_tx_queues; i++) {
3900                 ring = adapter->tx_ring[i];
3901                 do {
3902                         start = u64_stats_fetch_begin_irq(&ring->syncp);
3903                         bytes = ring->stats.bytes;
3904                         packets = ring->stats.packets;
3905                 } while (u64_stats_fetch_retry_irq(&ring->syncp, start));
3906                 stats->tx_bytes += bytes;
3907                 stats->tx_packets += packets;
3908         }
3909
3910         return stats;
3911 }
3912
3913 #define IXGBEVF_MAX_MAC_HDR_LEN         127
3914 #define IXGBEVF_MAX_NETWORK_HDR_LEN     511
3915
3916 static netdev_features_t
3917 ixgbevf_features_check(struct sk_buff *skb, struct net_device *dev,
3918                        netdev_features_t features)
3919 {
3920         unsigned int network_hdr_len, mac_hdr_len;
3921
3922         /* Make certain the headers can be described by a context descriptor */
3923         mac_hdr_len = skb_network_header(skb) - skb->data;
3924         if (unlikely(mac_hdr_len > IXGBEVF_MAX_MAC_HDR_LEN))
3925                 return features & ~(NETIF_F_HW_CSUM |
3926                                     NETIF_F_SCTP_CRC |
3927                                     NETIF_F_HW_VLAN_CTAG_TX |
3928                                     NETIF_F_TSO |
3929                                     NETIF_F_TSO6);
3930
3931         network_hdr_len = skb_checksum_start(skb) - skb_network_header(skb);
3932         if (unlikely(network_hdr_len >  IXGBEVF_MAX_NETWORK_HDR_LEN))
3933                 return features & ~(NETIF_F_HW_CSUM |
3934                                     NETIF_F_SCTP_CRC |
3935                                     NETIF_F_TSO |
3936                                     NETIF_F_TSO6);
3937
3938         /* We can only support IPV4 TSO in tunnels if we can mangle the
3939          * inner IP ID field, so strip TSO if MANGLEID is not supported.
3940          */
3941         if (skb->encapsulation && !(features & NETIF_F_TSO_MANGLEID))
3942                 features &= ~NETIF_F_TSO;
3943
3944         return features;
3945 }
3946
3947 static const struct net_device_ops ixgbevf_netdev_ops = {
3948         .ndo_open               = ixgbevf_open,
3949         .ndo_stop               = ixgbevf_close,
3950         .ndo_start_xmit         = ixgbevf_xmit_frame,
3951         .ndo_set_rx_mode        = ixgbevf_set_rx_mode,
3952         .ndo_get_stats64        = ixgbevf_get_stats,
3953         .ndo_validate_addr      = eth_validate_addr,
3954         .ndo_set_mac_address    = ixgbevf_set_mac,
3955         .ndo_change_mtu         = ixgbevf_change_mtu,
3956         .ndo_tx_timeout         = ixgbevf_tx_timeout,
3957         .ndo_vlan_rx_add_vid    = ixgbevf_vlan_rx_add_vid,
3958         .ndo_vlan_rx_kill_vid   = ixgbevf_vlan_rx_kill_vid,
3959 #ifdef CONFIG_NET_RX_BUSY_POLL
3960         .ndo_busy_poll          = ixgbevf_busy_poll_recv,
3961 #endif
3962 #ifdef CONFIG_NET_POLL_CONTROLLER
3963         .ndo_poll_controller    = ixgbevf_netpoll,
3964 #endif
3965         .ndo_features_check     = ixgbevf_features_check,
3966 };
3967
3968 static void ixgbevf_assign_netdev_ops(struct net_device *dev)
3969 {
3970         dev->netdev_ops = &ixgbevf_netdev_ops;
3971         ixgbevf_set_ethtool_ops(dev);
3972         dev->watchdog_timeo = 5 * HZ;
3973 }
3974
3975 /**
3976  * ixgbevf_probe - Device Initialization Routine
3977  * @pdev: PCI device information struct
3978  * @ent: entry in ixgbevf_pci_tbl
3979  *
3980  * Returns 0 on success, negative on failure
3981  *
3982  * ixgbevf_probe initializes an adapter identified by a pci_dev structure.
3983  * The OS initialization, configuring of the adapter private structure,
3984  * and a hardware reset occur.
3985  **/
3986 static int ixgbevf_probe(struct pci_dev *pdev, const struct pci_device_id *ent)
3987 {
3988         struct net_device *netdev;
3989         struct ixgbevf_adapter *adapter = NULL;
3990         struct ixgbe_hw *hw = NULL;
3991         const struct ixgbevf_info *ii = ixgbevf_info_tbl[ent->driver_data];
3992         int err, pci_using_dac;
3993         bool disable_dev = false;
3994
3995         err = pci_enable_device(pdev);
3996         if (err)
3997                 return err;
3998
3999         if (!dma_set_mask_and_coherent(&pdev->dev, DMA_BIT_MASK(64))) {
4000                 pci_using_dac = 1;
4001         } else {
4002                 err = dma_set_mask_and_coherent(&pdev->dev, DMA_BIT_MASK(32));
4003                 if (err) {
4004                         dev_err(&pdev->dev, "No usable DMA configuration, aborting\n");
4005                         goto err_dma;
4006                 }
4007                 pci_using_dac = 0;
4008         }
4009
4010         err = pci_request_regions(pdev, ixgbevf_driver_name);
4011         if (err) {
4012                 dev_err(&pdev->dev, "pci_request_regions failed 0x%x\n", err);
4013                 goto err_pci_reg;
4014         }
4015
4016         pci_set_master(pdev);
4017
4018         netdev = alloc_etherdev_mq(sizeof(struct ixgbevf_adapter),
4019                                    MAX_TX_QUEUES);
4020         if (!netdev) {
4021                 err = -ENOMEM;
4022                 goto err_alloc_etherdev;
4023         }
4024
4025         SET_NETDEV_DEV(netdev, &pdev->dev);
4026
4027         adapter = netdev_priv(netdev);
4028
4029         adapter->netdev = netdev;
4030         adapter->pdev = pdev;
4031         hw = &adapter->hw;
4032         hw->back = adapter;
4033         adapter->msg_enable = netif_msg_init(debug, DEFAULT_MSG_ENABLE);
4034
4035         /* call save state here in standalone driver because it relies on
4036          * adapter struct to exist, and needs to call netdev_priv
4037          */
4038         pci_save_state(pdev);
4039
4040         hw->hw_addr = ioremap(pci_resource_start(pdev, 0),
4041                               pci_resource_len(pdev, 0));
4042         adapter->io_addr = hw->hw_addr;
4043         if (!hw->hw_addr) {
4044                 err = -EIO;
4045                 goto err_ioremap;
4046         }
4047
4048         ixgbevf_assign_netdev_ops(netdev);
4049
4050         /* Setup HW API */
4051         memcpy(&hw->mac.ops, ii->mac_ops, sizeof(hw->mac.ops));
4052         hw->mac.type  = ii->mac;
4053
4054         memcpy(&hw->mbx.ops, &ixgbevf_mbx_ops,
4055                sizeof(struct ixgbe_mbx_operations));
4056
4057         /* setup the private structure */
4058         err = ixgbevf_sw_init(adapter);
4059         if (err)
4060                 goto err_sw_init;
4061
4062         /* The HW MAC address was set and/or determined in sw_init */
4063         if (!is_valid_ether_addr(netdev->dev_addr)) {
4064                 pr_err("invalid MAC address\n");
4065                 err = -EIO;
4066                 goto err_sw_init;
4067         }
4068
4069         netdev->hw_features = NETIF_F_SG |
4070                               NETIF_F_TSO |
4071                               NETIF_F_TSO6 |
4072                               NETIF_F_RXCSUM |
4073                               NETIF_F_HW_CSUM |
4074                               NETIF_F_SCTP_CRC;
4075
4076 #define IXGBEVF_GSO_PARTIAL_FEATURES (NETIF_F_GSO_GRE | \
4077                                       NETIF_F_GSO_GRE_CSUM | \
4078                                       NETIF_F_GSO_IPXIP4 | \
4079                                       NETIF_F_GSO_IPXIP6 | \
4080                                       NETIF_F_GSO_UDP_TUNNEL | \
4081                                       NETIF_F_GSO_UDP_TUNNEL_CSUM)
4082
4083         netdev->gso_partial_features = IXGBEVF_GSO_PARTIAL_FEATURES;
4084         netdev->hw_features |= NETIF_F_GSO_PARTIAL |
4085                                IXGBEVF_GSO_PARTIAL_FEATURES;
4086
4087         netdev->features = netdev->hw_features;
4088
4089         if (pci_using_dac)
4090                 netdev->features |= NETIF_F_HIGHDMA;
4091
4092         netdev->vlan_features |= netdev->features | NETIF_F_TSO_MANGLEID;
4093         netdev->mpls_features |= NETIF_F_HW_CSUM;
4094         netdev->hw_enc_features |= netdev->vlan_features;
4095
4096         /* set this bit last since it cannot be part of vlan_features */
4097         netdev->features |= NETIF_F_HW_VLAN_CTAG_FILTER |
4098                             NETIF_F_HW_VLAN_CTAG_RX |
4099                             NETIF_F_HW_VLAN_CTAG_TX;
4100
4101         netdev->priv_flags |= IFF_UNICAST_FLT;
4102
4103         if (IXGBE_REMOVED(hw->hw_addr)) {
4104                 err = -EIO;
4105                 goto err_sw_init;
4106         }
4107
4108         setup_timer(&adapter->service_timer, &ixgbevf_service_timer,
4109                     (unsigned long)adapter);
4110
4111         INIT_WORK(&adapter->service_task, ixgbevf_service_task);
4112         set_bit(__IXGBEVF_SERVICE_INITED, &adapter->state);
4113         clear_bit(__IXGBEVF_SERVICE_SCHED, &adapter->state);
4114
4115         err = ixgbevf_init_interrupt_scheme(adapter);
4116         if (err)
4117                 goto err_sw_init;
4118
4119         strcpy(netdev->name, "eth%d");
4120
4121         err = register_netdev(netdev);
4122         if (err)
4123                 goto err_register;
4124
4125         pci_set_drvdata(pdev, netdev);
4126         netif_carrier_off(netdev);
4127
4128         ixgbevf_init_last_counter_stats(adapter);
4129
4130         /* print the VF info */
4131         dev_info(&pdev->dev, "%pM\n", netdev->dev_addr);
4132         dev_info(&pdev->dev, "MAC: %d\n", hw->mac.type);
4133
4134         switch (hw->mac.type) {
4135         case ixgbe_mac_X550_vf:
4136                 dev_info(&pdev->dev, "Intel(R) X550 Virtual Function\n");
4137                 break;
4138         case ixgbe_mac_X540_vf:
4139                 dev_info(&pdev->dev, "Intel(R) X540 Virtual Function\n");
4140                 break;
4141         case ixgbe_mac_82599_vf:
4142         default:
4143                 dev_info(&pdev->dev, "Intel(R) 82599 Virtual Function\n");
4144                 break;
4145         }
4146
4147         return 0;
4148
4149 err_register:
4150         ixgbevf_clear_interrupt_scheme(adapter);
4151 err_sw_init:
4152         ixgbevf_reset_interrupt_capability(adapter);
4153         iounmap(adapter->io_addr);
4154 err_ioremap:
4155         disable_dev = !test_and_set_bit(__IXGBEVF_DISABLED, &adapter->state);
4156         free_netdev(netdev);
4157 err_alloc_etherdev:
4158         pci_release_regions(pdev);
4159 err_pci_reg:
4160 err_dma:
4161         if (!adapter || disable_dev)
4162                 pci_disable_device(pdev);
4163         return err;
4164 }
4165
4166 /**
4167  * ixgbevf_remove - Device Removal Routine
4168  * @pdev: PCI device information struct
4169  *
4170  * ixgbevf_remove is called by the PCI subsystem to alert the driver
4171  * that it should release a PCI device.  The could be caused by a
4172  * Hot-Plug event, or because the driver is going to be removed from
4173  * memory.
4174  **/
4175 static void ixgbevf_remove(struct pci_dev *pdev)
4176 {
4177         struct net_device *netdev = pci_get_drvdata(pdev);
4178         struct ixgbevf_adapter *adapter;
4179         bool disable_dev;
4180
4181         if (!netdev)
4182                 return;
4183
4184         adapter = netdev_priv(netdev);
4185
4186         set_bit(__IXGBEVF_REMOVING, &adapter->state);
4187         cancel_work_sync(&adapter->service_task);
4188
4189         if (netdev->reg_state == NETREG_REGISTERED)
4190                 unregister_netdev(netdev);
4191
4192         ixgbevf_clear_interrupt_scheme(adapter);
4193         ixgbevf_reset_interrupt_capability(adapter);
4194
4195         iounmap(adapter->io_addr);
4196         pci_release_regions(pdev);
4197
4198         hw_dbg(&adapter->hw, "Remove complete\n");
4199
4200         disable_dev = !test_and_set_bit(__IXGBEVF_DISABLED, &adapter->state);
4201         free_netdev(netdev);
4202
4203         if (disable_dev)
4204                 pci_disable_device(pdev);
4205 }
4206
4207 /**
4208  * ixgbevf_io_error_detected - called when PCI error is detected
4209  * @pdev: Pointer to PCI device
4210  * @state: The current pci connection state
4211  *
4212  * This function is called after a PCI bus error affecting
4213  * this device has been detected.
4214  **/
4215 static pci_ers_result_t ixgbevf_io_error_detected(struct pci_dev *pdev,
4216                                                   pci_channel_state_t state)
4217 {
4218         struct net_device *netdev = pci_get_drvdata(pdev);
4219         struct ixgbevf_adapter *adapter = netdev_priv(netdev);
4220
4221         if (!test_bit(__IXGBEVF_SERVICE_INITED, &adapter->state))
4222                 return PCI_ERS_RESULT_DISCONNECT;
4223
4224         rtnl_lock();
4225         netif_device_detach(netdev);
4226
4227         if (state == pci_channel_io_perm_failure) {
4228                 rtnl_unlock();
4229                 return PCI_ERS_RESULT_DISCONNECT;
4230         }
4231
4232         if (netif_running(netdev))
4233                 ixgbevf_down(adapter);
4234
4235         if (!test_and_set_bit(__IXGBEVF_DISABLED, &adapter->state))
4236                 pci_disable_device(pdev);
4237         rtnl_unlock();
4238
4239         /* Request a slot slot reset. */
4240         return PCI_ERS_RESULT_NEED_RESET;
4241 }
4242
4243 /**
4244  * ixgbevf_io_slot_reset - called after the pci bus has been reset.
4245  * @pdev: Pointer to PCI device
4246  *
4247  * Restart the card from scratch, as if from a cold-boot. Implementation
4248  * resembles the first-half of the ixgbevf_resume routine.
4249  **/
4250 static pci_ers_result_t ixgbevf_io_slot_reset(struct pci_dev *pdev)
4251 {
4252         struct net_device *netdev = pci_get_drvdata(pdev);
4253         struct ixgbevf_adapter *adapter = netdev_priv(netdev);
4254
4255         if (pci_enable_device_mem(pdev)) {
4256                 dev_err(&pdev->dev,
4257                         "Cannot re-enable PCI device after reset.\n");
4258                 return PCI_ERS_RESULT_DISCONNECT;
4259         }
4260
4261         smp_mb__before_atomic();
4262         clear_bit(__IXGBEVF_DISABLED, &adapter->state);
4263         pci_set_master(pdev);
4264
4265         ixgbevf_reset(adapter);
4266
4267         return PCI_ERS_RESULT_RECOVERED;
4268 }
4269
4270 /**
4271  * ixgbevf_io_resume - called when traffic can start flowing again.
4272  * @pdev: Pointer to PCI device
4273  *
4274  * This callback is called when the error recovery driver tells us that
4275  * its OK to resume normal operation. Implementation resembles the
4276  * second-half of the ixgbevf_resume routine.
4277  **/
4278 static void ixgbevf_io_resume(struct pci_dev *pdev)
4279 {
4280         struct net_device *netdev = pci_get_drvdata(pdev);
4281         struct ixgbevf_adapter *adapter = netdev_priv(netdev);
4282
4283         if (netif_running(netdev))
4284                 ixgbevf_up(adapter);
4285
4286         netif_device_attach(netdev);
4287 }
4288
4289 /* PCI Error Recovery (ERS) */
4290 static const struct pci_error_handlers ixgbevf_err_handler = {
4291         .error_detected = ixgbevf_io_error_detected,
4292         .slot_reset = ixgbevf_io_slot_reset,
4293         .resume = ixgbevf_io_resume,
4294 };
4295
4296 static struct pci_driver ixgbevf_driver = {
4297         .name           = ixgbevf_driver_name,
4298         .id_table       = ixgbevf_pci_tbl,
4299         .probe          = ixgbevf_probe,
4300         .remove         = ixgbevf_remove,
4301 #ifdef CONFIG_PM
4302         /* Power Management Hooks */
4303         .suspend        = ixgbevf_suspend,
4304         .resume         = ixgbevf_resume,
4305 #endif
4306         .shutdown       = ixgbevf_shutdown,
4307         .err_handler    = &ixgbevf_err_handler
4308 };
4309
4310 /**
4311  * ixgbevf_init_module - Driver Registration Routine
4312  *
4313  * ixgbevf_init_module is the first routine called when the driver is
4314  * loaded. All it does is register with the PCI subsystem.
4315  **/
4316 static int __init ixgbevf_init_module(void)
4317 {
4318         pr_info("%s - version %s\n", ixgbevf_driver_string,
4319                 ixgbevf_driver_version);
4320
4321         pr_info("%s\n", ixgbevf_copyright);
4322         ixgbevf_wq = create_singlethread_workqueue(ixgbevf_driver_name);
4323         if (!ixgbevf_wq) {
4324                 pr_err("%s: Failed to create workqueue\n", ixgbevf_driver_name);
4325                 return -ENOMEM;
4326         }
4327
4328         return pci_register_driver(&ixgbevf_driver);
4329 }
4330
4331 module_init(ixgbevf_init_module);
4332
4333 /**
4334  * ixgbevf_exit_module - Driver Exit Cleanup Routine
4335  *
4336  * ixgbevf_exit_module is called just before the driver is removed
4337  * from memory.
4338  **/
4339 static void __exit ixgbevf_exit_module(void)
4340 {
4341         pci_unregister_driver(&ixgbevf_driver);
4342         if (ixgbevf_wq) {
4343                 destroy_workqueue(ixgbevf_wq);
4344                 ixgbevf_wq = NULL;
4345         }
4346 }
4347
4348 #ifdef DEBUG
4349 /**
4350  * ixgbevf_get_hw_dev_name - return device name string
4351  * used by hardware layer to print debugging information
4352  **/
4353 char *ixgbevf_get_hw_dev_name(struct ixgbe_hw *hw)
4354 {
4355         struct ixgbevf_adapter *adapter = hw->back;
4356
4357         return adapter->netdev->name;
4358 }
4359
4360 #endif
4361 module_exit(ixgbevf_exit_module);
4362
4363 /* ixgbevf_main.c */