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 #include <net/mpls.h>
  53
  54 #include "ixgbevf.h"
  55
  56 const char ixgbevf_driver_name[] = "ixgbevf";
  57 static const char ixgbevf_driver_string[] =
  58         "Intel(R) 10 Gigabit PCI Express Virtual Function Network Driver";
  59
  60 #define DRV_VERSION "4.1.0-k"
  61 const char ixgbevf_driver_version[] = DRV_VERSION;
  62 static char ixgbevf_copyright[] =
  63         "Copyright (c) 2009 - 2015 Intel Corporation.";
  64
  65 static const struct ixgbevf_info *ixgbevf_info_tbl[] = {
  66         [board_82599_vf]        = &ixgbevf_82599_vf_info,
  67         [board_82599_vf_hv]     = &ixgbevf_82599_vf_hv_info,
  68         [board_X540_vf]         = &ixgbevf_X540_vf_info,
  69         [board_X540_vf_hv]      = &ixgbevf_X540_vf_hv_info,
  70         [board_X550_vf]         = &ixgbevf_X550_vf_info,
  71         [board_X550_vf_hv]      = &ixgbevf_X550_vf_hv_info,
  72         [board_X550EM_x_vf]     = &ixgbevf_X550EM_x_vf_info,
  73         [board_X550EM_x_vf_hv]  = &ixgbevf_X550EM_x_vf_hv_info,
  74         [board_x550em_a_vf]     = &ixgbevf_x550em_a_vf_info,
  75 };
  76
  77 /* ixgbevf_pci_tbl - PCI Device ID Table
  78  *
  79  * Wildcard entries (PCI_ANY_ID) should come last
  80  * Last entry must be all 0s
  81  *
  82  * { Vendor ID, Device ID, SubVendor ID, SubDevice ID,
  83  *   Class, Class Mask, private data (not used) }
  84  */
  85 static const struct pci_device_id ixgbevf_pci_tbl[] = {
  86         {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82599_VF), board_82599_vf },
  87         {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82599_VF_HV), board_82599_vf_hv },
  88         {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_X540_VF), board_X540_vf },
  89         {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_X540_VF_HV), board_X540_vf_hv },
  90         {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_X550_VF), board_X550_vf },
  91         {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_X550_VF_HV), board_X550_vf_hv },
  92         {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_X550EM_X_VF), board_X550EM_x_vf },
  93         {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_X550EM_X_VF_HV), board_X550EM_x_vf_hv},
  94         {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_X550EM_A_VF), board_x550em_a_vf },
  95         /* required last entry */
  96         {0, }
  97 };
  98 MODULE_DEVICE_TABLE(pci, ixgbevf_pci_tbl);
  99
 100 MODULE_AUTHOR("Intel Corporation, <linux.nics@intel.com>");
 101 MODULE_DESCRIPTION("Intel(R) 10 Gigabit Virtual Function Network Driver");
 102 MODULE_LICENSE("GPL");
 103 MODULE_VERSION(DRV_VERSION);
 104
 105 #define DEFAULT_MSG_ENABLE (NETIF_MSG_DRV|NETIF_MSG_PROBE|NETIF_MSG_LINK)
 106 static int debug = -1;
 107 module_param(debug, int, 0);
 108 MODULE_PARM_DESC(debug, "Debug level (0=none,...,16=all)");
 109
 110 static struct workqueue_struct *ixgbevf_wq;
 111
 112 static void ixgbevf_service_event_schedule(struct ixgbevf_adapter *adapter)
 113 {
 114         if (!test_bit(__IXGBEVF_DOWN, &adapter->state) &&
 115             !test_bit(__IXGBEVF_REMOVING, &adapter->state) &&
 116             !test_and_set_bit(__IXGBEVF_SERVICE_SCHED, &adapter->state))
 117                 queue_work(ixgbevf_wq, &adapter->service_task);
 118 }
 119
 120 static void ixgbevf_service_event_complete(struct ixgbevf_adapter *adapter)
 121 {
 122         BUG_ON(!test_bit(__IXGBEVF_SERVICE_SCHED, &adapter->state));
 123
 124         /* flush memory to make sure state is correct before next watchdog */
 125         smp_mb__before_atomic();
 126         clear_bit(__IXGBEVF_SERVICE_SCHED, &adapter->state);
 127 }
 128
 129 /* forward decls */
 130 static void ixgbevf_queue_reset_subtask(struct ixgbevf_adapter *adapter);
 131 static void ixgbevf_set_itr(struct ixgbevf_q_vector *q_vector);
 132 static void ixgbevf_free_all_rx_resources(struct ixgbevf_adapter *adapter);
 133
 134 static void ixgbevf_remove_adapter(struct ixgbe_hw *hw)
 135 {
 136         struct ixgbevf_adapter *adapter = hw->back;
 137
 138         if (!hw->hw_addr)
 139                 return;
 140         hw->hw_addr = NULL;
 141         dev_err(&adapter->pdev->dev, "Adapter removed\n");
 142         if (test_bit(__IXGBEVF_SERVICE_INITED, &adapter->state))
 143                 ixgbevf_service_event_schedule(adapter);
 144 }
 145
 146 static void ixgbevf_check_remove(struct ixgbe_hw *hw, u32 reg)
 147 {
 148         u32 value;
 149
 150         /* The following check not only optimizes a bit by not
 151          * performing a read on the status register when the
 152          * register just read was a status register read that
 153          * returned IXGBE_FAILED_READ_REG. It also blocks any
 154          * potential recursion.
 155          */
 156         if (reg == IXGBE_VFSTATUS) {
 157                 ixgbevf_remove_adapter(hw);
 158                 return;
 159         }
 160         value = ixgbevf_read_reg(hw, IXGBE_VFSTATUS);
 161         if (value == IXGBE_FAILED_READ_REG)
 162                 ixgbevf_remove_adapter(hw);
 163 }
 164
 165 u32 ixgbevf_read_reg(struct ixgbe_hw *hw, u32 reg)
 166 {
 167         u8 __iomem *reg_addr = ACCESS_ONCE(hw->hw_addr);
 168         u32 value;
 169
 170         if (IXGBE_REMOVED(reg_addr))
 171                 return IXGBE_FAILED_READ_REG;
 172         value = readl(reg_addr + reg);
 173         if (unlikely(value == IXGBE_FAILED_READ_REG))
 174                 ixgbevf_check_remove(hw, reg);
 175         return value;
 176 }
 177
 178 /**
 179  * ixgbevf_set_ivar - set IVAR registers - maps interrupt causes to vectors
 180  * @adapter: pointer to adapter struct
 181  * @direction: 0 for Rx, 1 for Tx, -1 for other causes
 182  * @queue: queue to map the corresponding interrupt to
 183  * @msix_vector: the vector to map to the corresponding queue
 184  **/
 185 static void ixgbevf_set_ivar(struct ixgbevf_adapter *adapter, s8 direction,
 186                              u8 queue, u8 msix_vector)
 187 {
 188         u32 ivar, index;
 189         struct ixgbe_hw *hw = &adapter->hw;
 190
 191         if (direction == -1) {
 192                 /* other causes */
 193                 msix_vector |= IXGBE_IVAR_ALLOC_VAL;
 194                 ivar = IXGBE_READ_REG(hw, IXGBE_VTIVAR_MISC);
 195                 ivar &= ~0xFF;
 196                 ivar |= msix_vector;
 197                 IXGBE_WRITE_REG(hw, IXGBE_VTIVAR_MISC, ivar);
 198         } else {
 199                 /* Tx or Rx causes */
 200                 msix_vector |= IXGBE_IVAR_ALLOC_VAL;
 201                 index = ((16 * (queue & 1)) + (8 * direction));
 202                 ivar = IXGBE_READ_REG(hw, IXGBE_VTIVAR(queue >> 1));
 203                 ivar &= ~(0xFF << index);
 204                 ivar |= (msix_vector << index);
 205                 IXGBE_WRITE_REG(hw, IXGBE_VTIVAR(queue >> 1), ivar);
 206         }
 207 }
 208
 209 static void ixgbevf_unmap_and_free_tx_resource(struct ixgbevf_ring *tx_ring,
 210                                         struct ixgbevf_tx_buffer *tx_buffer)
 211 {
 212         if (tx_buffer->skb) {
 213                 dev_kfree_skb_any(tx_buffer->skb);
 214                 if (dma_unmap_len(tx_buffer, len))
 215                         dma_unmap_single(tx_ring->dev,
 216                                          dma_unmap_addr(tx_buffer, dma),
 217                                          dma_unmap_len(tx_buffer, len),
 218                                          DMA_TO_DEVICE);
 219         } else if (dma_unmap_len(tx_buffer, len)) {
 220                 dma_unmap_page(tx_ring->dev,
 221                                dma_unmap_addr(tx_buffer, dma),
 222                                dma_unmap_len(tx_buffer, len),
 223                                DMA_TO_DEVICE);
 224         }
 225         tx_buffer->next_to_watch = NULL;
 226         tx_buffer->skb = NULL;
 227         dma_unmap_len_set(tx_buffer, len, 0);
 228         /* tx_buffer must be completely set up in the transmit path */
 229 }
 230
 231 static u64 ixgbevf_get_tx_completed(struct ixgbevf_ring *ring)
 232 {
 233         return ring->stats.packets;
 234 }
 235
 236 static u32 ixgbevf_get_tx_pending(struct ixgbevf_ring *ring)
 237 {
 238         struct ixgbevf_adapter *adapter = netdev_priv(ring->netdev);
 239         struct ixgbe_hw *hw = &adapter->hw;
 240
 241         u32 head = IXGBE_READ_REG(hw, IXGBE_VFTDH(ring->reg_idx));
 242         u32 tail = IXGBE_READ_REG(hw, IXGBE_VFTDT(ring->reg_idx));
 243
 244         if (head != tail)
 245                 return (head < tail) ?
 246                         tail - head : (tail + ring->count - head);
 247
 248         return 0;
 249 }
 250
 251 static inline bool ixgbevf_check_tx_hang(struct ixgbevf_ring *tx_ring)
 252 {
 253         u32 tx_done = ixgbevf_get_tx_completed(tx_ring);
 254         u32 tx_done_old = tx_ring->tx_stats.tx_done_old;
 255         u32 tx_pending = ixgbevf_get_tx_pending(tx_ring);
 256
 257         clear_check_for_tx_hang(tx_ring);
 258
 259         /* Check for a hung queue, but be thorough. This verifies
 260          * that a transmit has been completed since the previous
 261          * check AND there is at least one packet pending. The
 262          * ARMED bit is set to indicate a potential hang.
 263          */
 264         if ((tx_done_old == tx_done) && tx_pending) {
 265                 /* make sure it is true for two checks in a row */
 266                 return test_and_set_bit(__IXGBEVF_HANG_CHECK_ARMED,
 267                                         &tx_ring->state);
 268         }
 269         /* reset the countdown */
 270         clear_bit(__IXGBEVF_HANG_CHECK_ARMED, &tx_ring->state);
 271
 272         /* update completed stats and continue */
 273         tx_ring->tx_stats.tx_done_old = tx_done;
 274
 275         return false;
 276 }
 277
 278 static void ixgbevf_tx_timeout_reset(struct ixgbevf_adapter *adapter)
 279 {
 280         /* Do the reset outside of interrupt context */
 281         if (!test_bit(__IXGBEVF_DOWN, &adapter->state)) {
 282                 set_bit(__IXGBEVF_RESET_REQUESTED, &adapter->state);
 283                 ixgbevf_service_event_schedule(adapter);
 284         }
 285 }
 286
 287 /**
 288  * ixgbevf_tx_timeout - Respond to a Tx Hang
 289  * @netdev: network interface device structure
 290  **/
 291 static void ixgbevf_tx_timeout(struct net_device *netdev)
 292 {
 293         struct ixgbevf_adapter *adapter = netdev_priv(netdev);
 294
 295         ixgbevf_tx_timeout_reset(adapter);
 296 }
 297
 298 /**
 299  * ixgbevf_clean_tx_irq - Reclaim resources after transmit completes
 300  * @q_vector: board private structure
 301  * @tx_ring: tx ring to clean
 302  * @napi_budget: Used to determine if we are in netpoll
 303  **/
 304 static bool ixgbevf_clean_tx_irq(struct ixgbevf_q_vector *q_vector,
 305                                  struct ixgbevf_ring *tx_ring, int napi_budget)
 306 {
 307         struct ixgbevf_adapter *adapter = q_vector->adapter;
 308         struct ixgbevf_tx_buffer *tx_buffer;
 309         union ixgbe_adv_tx_desc *tx_desc;
 310         unsigned int total_bytes = 0, total_packets = 0;
 311         unsigned int budget = tx_ring->count / 2;
 312         unsigned int i = tx_ring->next_to_clean;
 313
 314         if (test_bit(__IXGBEVF_DOWN, &adapter->state))
 315                 return true;
 316
 317         tx_buffer = &tx_ring->tx_buffer_info[i];
 318         tx_desc = IXGBEVF_TX_DESC(tx_ring, i);
 319         i -= tx_ring->count;
 320
 321         do {
 322                 union ixgbe_adv_tx_desc *eop_desc = tx_buffer->next_to_watch;
 323
 324                 /* if next_to_watch is not set then there is no work pending */
 325                 if (!eop_desc)
 326                         break;
 327
 328                 /* prevent any other reads prior to eop_desc */
 329                 smp_rmb();
 330
 331                 /* if DD is not set pending work has not been completed */
 332                 if (!(eop_desc->wb.status & cpu_to_le32(IXGBE_TXD_STAT_DD)))
 333                         break;
 334
 335                 /* clear next_to_watch to prevent false hangs */
 336                 tx_buffer->next_to_watch = NULL;
 337
 338                 /* update the statistics for this packet */
 339                 total_bytes += tx_buffer->bytecount;
 340                 total_packets += tx_buffer->gso_segs;
 341
 342                 /* free the skb */
 343                 napi_consume_skb(tx_buffer->skb, napi_budget);
 344
 345                 /* unmap skb header data */
 346                 dma_unmap_single(tx_ring->dev,
 347                                  dma_unmap_addr(tx_buffer, dma),
 348                                  dma_unmap_len(tx_buffer, len),
 349                                  DMA_TO_DEVICE);
 350
 351                 /* clear tx_buffer data */
 352                 tx_buffer->skb = NULL;
 353                 dma_unmap_len_set(tx_buffer, len, 0);
 354
 355                 /* unmap remaining buffers */
 356                 while (tx_desc != eop_desc) {
 357                         tx_buffer++;
 358                         tx_desc++;
 359                         i++;
 360                         if (unlikely(!i)) {
 361                                 i -= tx_ring->count;
 362                                 tx_buffer = tx_ring->tx_buffer_info;
 363                                 tx_desc = IXGBEVF_TX_DESC(tx_ring, 0);
 364                         }
 365
 366                         /* unmap any remaining paged data */
 367                         if (dma_unmap_len(tx_buffer, len)) {
 368                                 dma_unmap_page(tx_ring->dev,
 369                                                dma_unmap_addr(tx_buffer, dma),
 370                                                dma_unmap_len(tx_buffer, len),
 371                                                DMA_TO_DEVICE);
 372                                 dma_unmap_len_set(tx_buffer, len, 0);
 373                         }
 374                 }
 375
 376                 /* move us one more past the eop_desc for start of next pkt */
 377                 tx_buffer++;
 378                 tx_desc++;
 379                 i++;
 380                 if (unlikely(!i)) {
 381                         i -= tx_ring->count;
 382                         tx_buffer = tx_ring->tx_buffer_info;
 383                         tx_desc = IXGBEVF_TX_DESC(tx_ring, 0);
 384                 }
 385
 386                 /* issue prefetch for next Tx descriptor */
 387                 prefetch(tx_desc);
 388
 389                 /* update budget accounting */
 390                 budget--;
 391         } while (likely(budget));
 392
 393         i += tx_ring->count;
 394         tx_ring->next_to_clean = i;
 395         u64_stats_update_begin(&tx_ring->syncp);
 396         tx_ring->stats.bytes += total_bytes;
 397         tx_ring->stats.packets += total_packets;
 398         u64_stats_update_end(&tx_ring->syncp);
 399         q_vector->tx.total_bytes += total_bytes;
 400         q_vector->tx.total_packets += total_packets;
 401
 402         if (check_for_tx_hang(tx_ring) && ixgbevf_check_tx_hang(tx_ring)) {
 403                 struct ixgbe_hw *hw = &adapter->hw;
 404                 union ixgbe_adv_tx_desc *eop_desc;
 405
 406                 eop_desc = tx_ring->tx_buffer_info[i].next_to_watch;
 407
 408                 pr_err("Detected Tx Unit Hang\n"
 409                        "  Tx Queue             <%d>\n"
 410                        "  TDH, TDT             <%x>, <%x>\n"
 411                        "  next_to_use          <%x>\n"
 412                        "  next_to_clean        <%x>\n"
 413                        "tx_buffer_info[next_to_clean]\n"
 414                        "  next_to_watch        <%p>\n"
 415                        "  eop_desc->wb.status  <%x>\n"
 416                        "  time_stamp           <%lx>\n"
 417                        "  jiffies              <%lx>\n",
 418                        tx_ring->queue_index,
 419                        IXGBE_READ_REG(hw, IXGBE_VFTDH(tx_ring->reg_idx)),
 420                        IXGBE_READ_REG(hw, IXGBE_VFTDT(tx_ring->reg_idx)),
 421                        tx_ring->next_to_use, i,
 422                        eop_desc, (eop_desc ? eop_desc->wb.status : 0),
 423                        tx_ring->tx_buffer_info[i].time_stamp, jiffies);
 424
 425                 netif_stop_subqueue(tx_ring->netdev, tx_ring->queue_index);
 426
 427                 /* schedule immediate reset if we believe we hung */
 428                 ixgbevf_tx_timeout_reset(adapter);
 429
 430                 return true;
 431         }
 432
 433 #define TX_WAKE_THRESHOLD (DESC_NEEDED * 2)
 434         if (unlikely(total_packets && netif_carrier_ok(tx_ring->netdev) &&
 435                      (ixgbevf_desc_unused(tx_ring) >= TX_WAKE_THRESHOLD))) {
 436                 /* Make sure that anybody stopping the queue after this
 437                  * sees the new next_to_clean.
 438                  */
 439                 smp_mb();
 440
 441                 if (__netif_subqueue_stopped(tx_ring->netdev,
 442                                              tx_ring->queue_index) &&
 443                     !test_bit(__IXGBEVF_DOWN, &adapter->state)) {
 444                         netif_wake_subqueue(tx_ring->netdev,
 445                                             tx_ring->queue_index);
 446                         ++tx_ring->tx_stats.restart_queue;
 447                 }
 448         }
 449
 450         return !!budget;
 451 }
 452
 453 /**
 454  * ixgbevf_rx_skb - Helper function to determine proper Rx method
 455  * @q_vector: structure containing interrupt and ring information
 456  * @skb: packet to send up
 457  **/
 458 static void ixgbevf_rx_skb(struct ixgbevf_q_vector *q_vector,
 459                            struct sk_buff *skb)
 460 {
 461         napi_gro_receive(&q_vector->napi, skb);
 462 }
 463
 464 #define IXGBE_RSS_L4_TYPES_MASK \
 465         ((1ul << IXGBE_RXDADV_RSSTYPE_IPV4_TCP) | \
 466          (1ul << IXGBE_RXDADV_RSSTYPE_IPV4_UDP) | \
 467          (1ul << IXGBE_RXDADV_RSSTYPE_IPV6_TCP) | \
 468          (1ul << IXGBE_RXDADV_RSSTYPE_IPV6_UDP))
 469
 470 static inline void ixgbevf_rx_hash(struct ixgbevf_ring *ring,
 471                                    union ixgbe_adv_rx_desc *rx_desc,
 472                                    struct sk_buff *skb)
 473 {
 474         u16 rss_type;
 475
 476         if (!(ring->netdev->features & NETIF_F_RXHASH))
 477                 return;
 478
 479         rss_type = le16_to_cpu(rx_desc->wb.lower.lo_dword.hs_rss.pkt_info) &
 480                    IXGBE_RXDADV_RSSTYPE_MASK;
 481
 482         if (!rss_type)
 483                 return;
 484
 485         skb_set_hash(skb, le32_to_cpu(rx_desc->wb.lower.hi_dword.rss),
 486                      (IXGBE_RSS_L4_TYPES_MASK & (1ul << rss_type)) ?
 487                      PKT_HASH_TYPE_L4 : PKT_HASH_TYPE_L3);
 488 }
 489
 490 /**
 491  * ixgbevf_rx_checksum - indicate in skb if hw indicated a good cksum
 492  * @ring: structure containig ring specific data
 493  * @rx_desc: current Rx descriptor being processed
 494  * @skb: skb currently being received and modified
 495  **/
 496 static inline void ixgbevf_rx_checksum(struct ixgbevf_ring *ring,
 497                                        union ixgbe_adv_rx_desc *rx_desc,
 498                                        struct sk_buff *skb)
 499 {
 500         skb_checksum_none_assert(skb);
 501
 502         /* Rx csum disabled */
 503         if (!(ring->netdev->features & NETIF_F_RXCSUM))
 504                 return;
 505
 506         /* if IP and error */
 507         if (ixgbevf_test_staterr(rx_desc, IXGBE_RXD_STAT_IPCS) &&
 508             ixgbevf_test_staterr(rx_desc, IXGBE_RXDADV_ERR_IPE)) {
 509                 ring->rx_stats.csum_err++;
 510                 return;
 511         }
 512
 513         if (!ixgbevf_test_staterr(rx_desc, IXGBE_RXD_STAT_L4CS))
 514                 return;
 515
 516         if (ixgbevf_test_staterr(rx_desc, IXGBE_RXDADV_ERR_TCPE)) {
 517                 ring->rx_stats.csum_err++;
 518                 return;
 519         }
 520
 521         /* It must be a TCP or UDP packet with a valid checksum */
 522         skb->ip_summed = CHECKSUM_UNNECESSARY;
 523 }
 524
 525 /**
 526  * ixgbevf_process_skb_fields - Populate skb header fields from Rx descriptor
 527  * @rx_ring: rx descriptor ring packet is being transacted on
 528  * @rx_desc: pointer to the EOP Rx descriptor
 529  * @skb: pointer to current skb being populated
 530  *
 531  * This function checks the ring, descriptor, and packet information in
 532  * order to populate the checksum, VLAN, protocol, and other fields within
 533  * the skb.
 534  **/
 535 static void ixgbevf_process_skb_fields(struct ixgbevf_ring *rx_ring,
 536                                        union ixgbe_adv_rx_desc *rx_desc,
 537                                        struct sk_buff *skb)
 538 {
 539         ixgbevf_rx_hash(rx_ring, rx_desc, skb);
 540         ixgbevf_rx_checksum(rx_ring, rx_desc, skb);
 541
 542         if (ixgbevf_test_staterr(rx_desc, IXGBE_RXD_STAT_VP)) {
 543                 u16 vid = le16_to_cpu(rx_desc->wb.upper.vlan);
 544                 unsigned long *active_vlans = netdev_priv(rx_ring->netdev);
 545
 546                 if (test_bit(vid & VLAN_VID_MASK, active_vlans))
 547                         __vlan_hwaccel_put_tag(skb, htons(ETH_P_8021Q), vid);
 548         }
 549
 550         skb->protocol = eth_type_trans(skb, rx_ring->netdev);
 551 }
 552
 553 /**
 554  * ixgbevf_is_non_eop - process handling of non-EOP buffers
 555  * @rx_ring: Rx ring being processed
 556  * @rx_desc: Rx descriptor for current buffer
 557  * @skb: current socket buffer containing buffer in progress
 558  *
 559  * This function updates next to clean.  If the buffer is an EOP buffer
 560  * this function exits returning false, otherwise it will place the
 561  * sk_buff in the next buffer to be chained and return true indicating
 562  * that this is in fact a non-EOP buffer.
 563  **/
 564 static bool ixgbevf_is_non_eop(struct ixgbevf_ring *rx_ring,
 565                                union ixgbe_adv_rx_desc *rx_desc)
 566 {
 567         u32 ntc = rx_ring->next_to_clean + 1;
 568
 569         /* fetch, update, and store next to clean */
 570         ntc = (ntc < rx_ring->count) ? ntc : 0;
 571         rx_ring->next_to_clean = ntc;
 572
 573         prefetch(IXGBEVF_RX_DESC(rx_ring, ntc));
 574
 575         if (likely(ixgbevf_test_staterr(rx_desc, IXGBE_RXD_STAT_EOP)))
 576                 return false;
 577
 578         return true;
 579 }
 580
 581 static bool ixgbevf_alloc_mapped_page(struct ixgbevf_ring *rx_ring,
 582                                       struct ixgbevf_rx_buffer *bi)
 583 {
 584         struct page *page = bi->page;
 585         dma_addr_t dma = bi->dma;
 586
 587         /* since we are recycling buffers we should seldom need to alloc */
 588         if (likely(page))
 589                 return true;
 590
 591         /* alloc new page for storage */
 592         page = dev_alloc_page();
 593         if (unlikely(!page)) {
 594                 rx_ring->rx_stats.alloc_rx_page_failed++;
 595                 return false;
 596         }
 597
 598         /* map page for use */
 599         dma = dma_map_page(rx_ring->dev, page, 0,
 600                            PAGE_SIZE, DMA_FROM_DEVICE);
 601
 602         /* if mapping failed free memory back to system since
 603          * there isn't much point in holding memory we can't use
 604          */
 605         if (dma_mapping_error(rx_ring->dev, dma)) {
 606                 __free_page(page);
 607
 608                 rx_ring->rx_stats.alloc_rx_buff_failed++;
 609                 return false;
 610         }
 611
 612         bi->dma = dma;
 613         bi->page = page;
 614         bi->page_offset = 0;
 615
 616         return true;
 617 }
 618
 619 /**
 620  * ixgbevf_alloc_rx_buffers - Replace used receive buffers; packet split
 621  * @rx_ring: rx descriptor ring (for a specific queue) to setup buffers on
 622  * @cleaned_count: number of buffers to replace
 623  **/
 624 static void ixgbevf_alloc_rx_buffers(struct ixgbevf_ring *rx_ring,
 625                                      u16 cleaned_count)
 626 {
 627         union ixgbe_adv_rx_desc *rx_desc;
 628         struct ixgbevf_rx_buffer *bi;
 629         unsigned int i = rx_ring->next_to_use;
 630
 631         /* nothing to do or no valid netdev defined */
 632         if (!cleaned_count || !rx_ring->netdev)
 633                 return;
 634
 635         rx_desc = IXGBEVF_RX_DESC(rx_ring, i);
 636         bi = &rx_ring->rx_buffer_info[i];
 637         i -= rx_ring->count;
 638
 639         do {
 640                 if (!ixgbevf_alloc_mapped_page(rx_ring, bi))
 641                         break;
 642
 643                 /* Refresh the desc even if pkt_addr didn't change
 644                  * because each write-back erases this info.
 645                  */
 646                 rx_desc->read.pkt_addr = cpu_to_le64(bi->dma + bi->page_offset);
 647
 648                 rx_desc++;
 649                 bi++;
 650                 i++;
 651                 if (unlikely(!i)) {
 652                         rx_desc = IXGBEVF_RX_DESC(rx_ring, 0);
 653                         bi = rx_ring->rx_buffer_info;
 654                         i -= rx_ring->count;
 655                 }
 656
 657                 /* clear the hdr_addr for the next_to_use descriptor */
 658                 rx_desc->read.hdr_addr = 0;
 659
 660                 cleaned_count--;
 661         } while (cleaned_count);
 662
 663         i += rx_ring->count;
 664
 665         if (rx_ring->next_to_use != i) {
 666                 /* record the next descriptor to use */
 667                 rx_ring->next_to_use = i;
 668
 669                 /* update next to alloc since we have filled the ring */
 670                 rx_ring->next_to_alloc = i;
 671
 672                 /* Force memory writes to complete before letting h/w
 673                  * know there are new descriptors to fetch.  (Only
 674                  * applicable for weak-ordered memory model archs,
 675                  * such as IA-64).
 676                  */
 677                 wmb();
 678                 ixgbevf_write_tail(rx_ring, i);
 679         }
 680 }
 681
 682 /**
 683  * ixgbevf_cleanup_headers - Correct corrupted or empty headers
 684  * @rx_ring: rx descriptor ring packet is being transacted on
 685  * @rx_desc: pointer to the EOP Rx descriptor
 686  * @skb: pointer to current skb being fixed
 687  *
 688  * Check for corrupted packet headers caused by senders on the local L2
 689  * embedded NIC switch not setting up their Tx Descriptors right.  These
 690  * should be very rare.
 691  *
 692  * Also address the case where we are pulling data in on pages only
 693  * and as such no data is present in the skb header.
 694  *
 695  * In addition if skb is not at least 60 bytes we need to pad it so that
 696  * it is large enough to qualify as a valid Ethernet frame.
 697  *
 698  * Returns true if an error was encountered and skb was freed.
 699  **/
 700 static bool ixgbevf_cleanup_headers(struct ixgbevf_ring *rx_ring,
 701                                     union ixgbe_adv_rx_desc *rx_desc,
 702                                     struct sk_buff *skb)
 703 {
 704         /* verify that the packet does not have any known errors */
 705         if (unlikely(ixgbevf_test_staterr(rx_desc,
 706                                           IXGBE_RXDADV_ERR_FRAME_ERR_MASK))) {
 707                 struct net_device *netdev = rx_ring->netdev;
 708
 709                 if (!(netdev->features & NETIF_F_RXALL)) {
 710                         dev_kfree_skb_any(skb);
 711                         return true;
 712                 }
 713         }
 714
 715         /* if eth_skb_pad returns an error the skb was freed */
 716         if (eth_skb_pad(skb))
 717                 return true;
 718
 719         return false;
 720 }
 721
 722 /**
 723  * ixgbevf_reuse_rx_page - page flip buffer and store it back on the ring
 724  * @rx_ring: rx descriptor ring to store buffers on
 725  * @old_buff: donor buffer to have page reused
 726  *
 727  * Synchronizes page for reuse by the adapter
 728  **/
 729 static void ixgbevf_reuse_rx_page(struct ixgbevf_ring *rx_ring,
 730                                   struct ixgbevf_rx_buffer *old_buff)
 731 {
 732         struct ixgbevf_rx_buffer *new_buff;
 733         u16 nta = rx_ring->next_to_alloc;
 734
 735         new_buff = &rx_ring->rx_buffer_info[nta];
 736
 737         /* update, and store next to alloc */
 738         nta++;
 739         rx_ring->next_to_alloc = (nta < rx_ring->count) ? nta : 0;
 740
 741         /* transfer page from old buffer to new buffer */
 742         new_buff->page = old_buff->page;
 743         new_buff->dma = old_buff->dma;
 744         new_buff->page_offset = old_buff->page_offset;
 745
 746         /* sync the buffer for use by the device */
 747         dma_sync_single_range_for_device(rx_ring->dev, new_buff->dma,
 748                                          new_buff->page_offset,
 749                                          IXGBEVF_RX_BUFSZ,
 750                                          DMA_FROM_DEVICE);
 751 }
 752
 753 static inline bool ixgbevf_page_is_reserved(struct page *page)
 754 {
 755         return (page_to_nid(page) != numa_mem_id()) || page_is_pfmemalloc(page);
 756 }
 757
 758 /**
 759  * ixgbevf_add_rx_frag - Add contents of Rx buffer to sk_buff
 760  * @rx_ring: rx descriptor ring to transact packets on
 761  * @rx_buffer: buffer containing page to add
 762  * @rx_desc: descriptor containing length of buffer written by hardware
 763  * @skb: sk_buff to place the data into
 764  *
 765  * This function will add the data contained in rx_buffer->page to the skb.
 766  * This is done either through a direct copy if the data in the buffer is
 767  * less than the skb header size, otherwise it will just attach the page as
 768  * a frag to the skb.
 769  *
 770  * The function will then update the page offset if necessary and return
 771  * true if the buffer can be reused by the adapter.
 772  **/
 773 static bool ixgbevf_add_rx_frag(struct ixgbevf_ring *rx_ring,
 774                                 struct ixgbevf_rx_buffer *rx_buffer,
 775                                 union ixgbe_adv_rx_desc *rx_desc,
 776                                 struct sk_buff *skb)
 777 {
 778         struct page *page = rx_buffer->page;
 779         unsigned char *va = page_address(page) + rx_buffer->page_offset;
 780         unsigned int size = le16_to_cpu(rx_desc->wb.upper.length);
 781 #if (PAGE_SIZE < 8192)
 782         unsigned int truesize = IXGBEVF_RX_BUFSZ;
 783 #else
 784         unsigned int truesize = ALIGN(size, L1_CACHE_BYTES);
 785 #endif
 786         unsigned int pull_len;
 787
 788         if (unlikely(skb_is_nonlinear(skb)))
 789                 goto add_tail_frag;
 790
 791         if (likely(size <= IXGBEVF_RX_HDR_SIZE)) {
 792                 memcpy(__skb_put(skb, size), va, ALIGN(size, sizeof(long)));
 793
 794                 /* page is not reserved, we can reuse buffer as is */
 795                 if (likely(!ixgbevf_page_is_reserved(page)))
 796                         return true;
 797
 798                 /* this page cannot be reused so discard it */
 799                 put_page(page);
 800                 return false;
 801         }
 802
 803         /* we need the header to contain the greater of either ETH_HLEN or
 804          * 60 bytes if the skb->len is less than 60 for skb_pad.
 805          */
 806         pull_len = eth_get_headlen(va, IXGBEVF_RX_HDR_SIZE);
 807
 808         /* align pull length to size of long to optimize memcpy performance */
 809         memcpy(__skb_put(skb, pull_len), va, ALIGN(pull_len, sizeof(long)));
 810
 811         /* update all of the pointers */
 812         va += pull_len;
 813         size -= pull_len;
 814
 815 add_tail_frag:
 816         skb_add_rx_frag(skb, skb_shinfo(skb)->nr_frags, page,
 817                         (unsigned long)va & ~PAGE_MASK, size, truesize);
 818
 819         /* avoid re-using remote pages */
 820         if (unlikely(ixgbevf_page_is_reserved(page)))
 821                 return false;
 822
 823 #if (PAGE_SIZE < 8192)
 824         /* if we are only owner of page we can reuse it */
 825         if (unlikely(page_count(page) != 1))
 826                 return false;
 827
 828         /* flip page offset to other buffer */
 829         rx_buffer->page_offset ^= IXGBEVF_RX_BUFSZ;
 830
 831 #else
 832         /* move offset up to the next cache line */
 833         rx_buffer->page_offset += truesize;
 834
 835         if (rx_buffer->page_offset > (PAGE_SIZE - IXGBEVF_RX_BUFSZ))
 836                 return false;
 837
 838 #endif
 839         /* Even if we own the page, we are not allowed to use atomic_set()
 840          * This would break get_page_unless_zero() users.
 841          */
 842         page_ref_inc(page);
 843
 844         return true;
 845 }
 846
 847 static struct sk_buff *ixgbevf_fetch_rx_buffer(struct ixgbevf_ring *rx_ring,
 848                                                union ixgbe_adv_rx_desc *rx_desc,
 849                                                struct sk_buff *skb)
 850 {
 851         struct ixgbevf_rx_buffer *rx_buffer;
 852         struct page *page;
 853
 854         rx_buffer = &rx_ring->rx_buffer_info[rx_ring->next_to_clean];
 855         page = rx_buffer->page;
 856         prefetchw(page);
 857
 858         if (likely(!skb)) {
 859                 void *page_addr = page_address(page) +
 860                                   rx_buffer->page_offset;
 861
 862                 /* prefetch first cache line of first page */
 863                 prefetch(page_addr);
 864 #if L1_CACHE_BYTES < 128
 865                 prefetch(page_addr + L1_CACHE_BYTES);
 866 #endif
 867
 868                 /* allocate a skb to store the frags */
 869                 skb = netdev_alloc_skb_ip_align(rx_ring->netdev,
 870                                                 IXGBEVF_RX_HDR_SIZE);
 871                 if (unlikely(!skb)) {
 872                         rx_ring->rx_stats.alloc_rx_buff_failed++;
 873                         return NULL;
 874                 }
 875
 876                 /* we will be copying header into skb->data in
 877                  * pskb_may_pull so it is in our interest to prefetch
 878                  * it now to avoid a possible cache miss
 879                  */
 880                 prefetchw(skb->data);
 881         }
 882
 883         /* we are reusing so sync this buffer for CPU use */
 884         dma_sync_single_range_for_cpu(rx_ring->dev,
 885                                       rx_buffer->dma,
 886                                       rx_buffer->page_offset,
 887                                       IXGBEVF_RX_BUFSZ,
 888                                       DMA_FROM_DEVICE);
 889
 890         /* pull page into skb */
 891         if (ixgbevf_add_rx_frag(rx_ring, rx_buffer, rx_desc, skb)) {
 892                 /* hand second half of page back to the ring */
 893                 ixgbevf_reuse_rx_page(rx_ring, rx_buffer);
 894         } else {
 895                 /* we are not reusing the buffer so unmap it */
 896                 dma_unmap_page(rx_ring->dev, rx_buffer->dma,
 897                                PAGE_SIZE, DMA_FROM_DEVICE);
 898         }
 899
 900         /* clear contents of buffer_info */
 901         rx_buffer->dma = 0;
 902         rx_buffer->page = NULL;
 903
 904         return skb;
 905 }
 906
 907 static inline void ixgbevf_irq_enable_queues(struct ixgbevf_adapter *adapter,
 908                                              u32 qmask)
 909 {
 910         struct ixgbe_hw *hw = &adapter->hw;
 911
 912         IXGBE_WRITE_REG(hw, IXGBE_VTEIMS, qmask);
 913 }
 914
 915 static int ixgbevf_clean_rx_irq(struct ixgbevf_q_vector *q_vector,
 916                                 struct ixgbevf_ring *rx_ring,
 917                                 int budget)
 918 {
 919         unsigned int total_rx_bytes = 0, total_rx_packets = 0;
 920         u16 cleaned_count = ixgbevf_desc_unused(rx_ring);
 921         struct sk_buff *skb = rx_ring->skb;
 922
 923         while (likely(total_rx_packets < budget)) {
 924                 union ixgbe_adv_rx_desc *rx_desc;
 925
 926                 /* return some buffers to hardware, one at a time is too slow */
 927                 if (cleaned_count >= IXGBEVF_RX_BUFFER_WRITE) {
 928                         ixgbevf_alloc_rx_buffers(rx_ring, cleaned_count);
 929                         cleaned_count = 0;
 930                 }
 931
 932                 rx_desc = IXGBEVF_RX_DESC(rx_ring, rx_ring->next_to_clean);
 933
 934                 if (!ixgbevf_test_staterr(rx_desc, IXGBE_RXD_STAT_DD))
 935                         break;
 936
 937                 /* This memory barrier is needed to keep us from reading
 938                  * any other fields out of the rx_desc until we know the
 939                  * RXD_STAT_DD bit is set
 940                  */
 941                 rmb();
 942
 943                 /* retrieve a buffer from the ring */
 944                 skb = ixgbevf_fetch_rx_buffer(rx_ring, rx_desc, skb);
 945
 946                 /* exit if we failed to retrieve a buffer */
 947                 if (!skb)
 948                         break;
 949
 950                 cleaned_count++;
 951
 952                 /* fetch next buffer in frame if non-eop */
 953                 if (ixgbevf_is_non_eop(rx_ring, rx_desc))
 954                         continue;
 955
 956                 /* verify the packet layout is correct */
 957                 if (ixgbevf_cleanup_headers(rx_ring, rx_desc, skb)) {
 958                         skb = NULL;
 959                         continue;
 960                 }
 961
 962                 /* probably a little skewed due to removing CRC */
 963                 total_rx_bytes += skb->len;
 964
 965                 /* Workaround hardware that can't do proper VEPA multicast
 966                  * source pruning.
 967                  */
 968                 if ((skb->pkt_type == PACKET_BROADCAST ||
 969                      skb->pkt_type == PACKET_MULTICAST) &&
 970                     ether_addr_equal(rx_ring->netdev->dev_addr,
 971                                      eth_hdr(skb)->h_source)) {
 972                         dev_kfree_skb_irq(skb);
 973                         continue;
 974                 }
 975
 976                 /* populate checksum, VLAN, and protocol */
 977                 ixgbevf_process_skb_fields(rx_ring, rx_desc, skb);
 978
 979                 ixgbevf_rx_skb(q_vector, skb);
 980
 981                 /* reset skb pointer */
 982                 skb = NULL;
 983
 984                 /* update budget accounting */
 985                 total_rx_packets++;
 986         }
 987
 988         /* place incomplete frames back on ring for completion */
 989         rx_ring->skb = skb;
 990
 991         u64_stats_update_begin(&rx_ring->syncp);
 992         rx_ring->stats.packets += total_rx_packets;
 993         rx_ring->stats.bytes += total_rx_bytes;
 994         u64_stats_update_end(&rx_ring->syncp);
 995         q_vector->rx.total_packets += total_rx_packets;
 996         q_vector->rx.total_bytes += total_rx_bytes;
 997
 998         return total_rx_packets;
 999 }
1000
1001 /**
1002  * ixgbevf_poll - NAPI polling calback
1003  * @napi: napi struct with our devices info in it
1004  * @budget: amount of work driver is allowed to do this pass, in packets
1005  *
1006  * This function will clean more than one or more rings associated with a
1007  * q_vector.
1008  **/
1009 static int ixgbevf_poll(struct napi_struct *napi, int budget)
1010 {
1011         struct ixgbevf_q_vector *q_vector =
1012                 container_of(napi, struct ixgbevf_q_vector, napi);
1013         struct ixgbevf_adapter *adapter = q_vector->adapter;
1014         struct ixgbevf_ring *ring;
1015         int per_ring_budget, work_done = 0;
1016         bool clean_complete = true;
1017
1018         ixgbevf_for_each_ring(ring, q_vector->tx) {
1019                 if (!ixgbevf_clean_tx_irq(q_vector, ring, budget))
1020                         clean_complete = false;
1021         }
1022
1023         if (budget <= 0)
1024                 return budget;
1025
1026         /* attempt to distribute budget to each queue fairly, but don't allow
1027          * the budget to go below 1 because we'll exit polling
1028          */
1029         if (q_vector->rx.count > 1)
1030                 per_ring_budget = max(budget/q_vector->rx.count, 1);
1031         else
1032                 per_ring_budget = budget;
1033
1034         ixgbevf_for_each_ring(ring, q_vector->rx) {
1035                 int cleaned = ixgbevf_clean_rx_irq(q_vector, ring,
1036                                                    per_ring_budget);
1037                 work_done += cleaned;
1038                 if (cleaned >= per_ring_budget)
1039                         clean_complete = false;
1040         }
1041
1042         /* If all work not completed, return budget and keep polling */
1043         if (!clean_complete)
1044                 return budget;
1045         /* all work done, exit the polling mode */
1046         napi_complete_done(napi, work_done);
1047         if (adapter->rx_itr_setting == 1)
1048                 ixgbevf_set_itr(q_vector);
1049         if (!test_bit(__IXGBEVF_DOWN, &adapter->state) &&
1050             !test_bit(__IXGBEVF_REMOVING, &adapter->state))
1051                 ixgbevf_irq_enable_queues(adapter,
1052                                           BIT(q_vector->v_idx));
1053
1054         return 0;
1055 }
1056
1057 /**
1058  * ixgbevf_write_eitr - write VTEITR register in hardware specific way
1059  * @q_vector: structure containing interrupt and ring information
1060  **/
1061 void ixgbevf_write_eitr(struct ixgbevf_q_vector *q_vector)
1062 {
1063         struct ixgbevf_adapter *adapter = q_vector->adapter;
1064         struct ixgbe_hw *hw = &adapter->hw;
1065         int v_idx = q_vector->v_idx;
1066         u32 itr_reg = q_vector->itr & IXGBE_MAX_EITR;
1067
1068         /* set the WDIS bit to not clear the timer bits and cause an
1069          * immediate assertion of the interrupt
1070          */
1071         itr_reg |= IXGBE_EITR_CNT_WDIS;
1072
1073         IXGBE_WRITE_REG(hw, IXGBE_VTEITR(v_idx), itr_reg);
1074 }
1075
1076 /**
1077  * ixgbevf_configure_msix - Configure MSI-X hardware
1078  * @adapter: board private structure
1079  *
1080  * ixgbevf_configure_msix sets up the hardware to properly generate MSI-X
1081  * interrupts.
1082  **/
1083 static void ixgbevf_configure_msix(struct ixgbevf_adapter *adapter)
1084 {
1085         struct ixgbevf_q_vector *q_vector;
1086         int q_vectors, v_idx;
1087
1088         q_vectors = adapter->num_msix_vectors - NON_Q_VECTORS;
1089         adapter->eims_enable_mask = 0;
1090
1091         /* Populate the IVAR table and set the ITR values to the
1092          * corresponding register.
1093          */
1094         for (v_idx = 0; v_idx < q_vectors; v_idx++) {
1095                 struct ixgbevf_ring *ring;
1096
1097                 q_vector = adapter->q_vector[v_idx];
1098
1099                 ixgbevf_for_each_ring(ring, q_vector->rx)
1100                         ixgbevf_set_ivar(adapter, 0, ring->reg_idx, v_idx);
1101
1102                 ixgbevf_for_each_ring(ring, q_vector->tx)
1103                         ixgbevf_set_ivar(adapter, 1, ring->reg_idx, v_idx);
1104
1105                 if (q_vector->tx.ring && !q_vector->rx.ring) {
1106                         /* Tx only vector */
1107                         if (adapter->tx_itr_setting == 1)
1108                                 q_vector->itr = IXGBE_12K_ITR;
1109                         else
1110                                 q_vector->itr = adapter->tx_itr_setting;
1111                 } else {
1112                         /* Rx or Rx/Tx vector */
1113                         if (adapter->rx_itr_setting == 1)
1114                                 q_vector->itr = IXGBE_20K_ITR;
1115                         else
1116                                 q_vector->itr = adapter->rx_itr_setting;
1117                 }
1118
1119                 /* add q_vector eims value to global eims_enable_mask */
1120                 adapter->eims_enable_mask |= BIT(v_idx);
1121
1122                 ixgbevf_write_eitr(q_vector);
1123         }
1124
1125         ixgbevf_set_ivar(adapter, -1, 1, v_idx);
1126         /* setup eims_other and add value to global eims_enable_mask */
1127         adapter->eims_other = BIT(v_idx);
1128         adapter->eims_enable_mask |= adapter->eims_other;
1129 }
1130
1131 enum latency_range {
1132         lowest_latency = 0,
1133         low_latency = 1,
1134         bulk_latency = 2,
1135         latency_invalid = 255
1136 };
1137
1138 /**
1139  * ixgbevf_update_itr - update the dynamic ITR value based on statistics
1140  * @q_vector: structure containing interrupt and ring information
1141  * @ring_container: structure containing ring performance data
1142  *
1143  * Stores a new ITR value based on packets and byte
1144  * counts during the last interrupt.  The advantage of per interrupt
1145  * computation is faster updates and more accurate ITR for the current
1146  * traffic pattern.  Constants in this function were computed
1147  * based on theoretical maximum wire speed and thresholds were set based
1148  * on testing data as well as attempting to minimize response time
1149  * while increasing bulk throughput.
1150  **/
1151 static void ixgbevf_update_itr(struct ixgbevf_q_vector *q_vector,
1152                                struct ixgbevf_ring_container *ring_container)
1153 {
1154         int bytes = ring_container->total_bytes;
1155         int packets = ring_container->total_packets;
1156         u32 timepassed_us;
1157         u64 bytes_perint;
1158         u8 itr_setting = ring_container->itr;
1159
1160         if (packets == 0)
1161                 return;
1162
1163         /* simple throttle rate management
1164          *    0-20MB/s lowest (100000 ints/s)
1165          *   20-100MB/s low   (20000 ints/s)
1166          *  100-1249MB/s bulk (12000 ints/s)
1167          */
1168         /* what was last interrupt timeslice? */
1169         timepassed_us = q_vector->itr >> 2;
1170         bytes_perint = bytes / timepassed_us; /* bytes/usec */
1171
1172         switch (itr_setting) {
1173         case lowest_latency:
1174                 if (bytes_perint > 10)
1175                         itr_setting = low_latency;
1176                 break;
1177         case low_latency:
1178                 if (bytes_perint > 20)
1179                         itr_setting = bulk_latency;
1180                 else if (bytes_perint <= 10)
1181                         itr_setting = lowest_latency;
1182                 break;
1183         case bulk_latency:
1184                 if (bytes_perint <= 20)
1185                         itr_setting = low_latency;
1186                 break;
1187         }
1188
1189         /* clear work counters since we have the values we need */
1190         ring_container->total_bytes = 0;
1191         ring_container->total_packets = 0;
1192
1193         /* write updated itr to ring container */
1194         ring_container->itr = itr_setting;
1195 }
1196
1197 static void ixgbevf_set_itr(struct ixgbevf_q_vector *q_vector)
1198 {
1199         u32 new_itr = q_vector->itr;
1200         u8 current_itr;
1201
1202         ixgbevf_update_itr(q_vector, &q_vector->tx);
1203         ixgbevf_update_itr(q_vector, &q_vector->rx);
1204
1205         current_itr = max(q_vector->rx.itr, q_vector->tx.itr);
1206
1207         switch (current_itr) {
1208         /* counts and packets in update_itr are dependent on these numbers */
1209         case lowest_latency:
1210                 new_itr = IXGBE_100K_ITR;
1211                 break;
1212         case low_latency:
1213                 new_itr = IXGBE_20K_ITR;
1214                 break;
1215         case bulk_latency:
1216                 new_itr = IXGBE_12K_ITR;
1217                 break;
1218         default:
1219                 break;
1220         }
1221
1222         if (new_itr != q_vector->itr) {
1223                 /* do an exponential smoothing */
1224                 new_itr = (10 * new_itr * q_vector->itr) /
1225                           ((9 * new_itr) + q_vector->itr);
1226
1227                 /* save the algorithm value here */
1228                 q_vector->itr = new_itr;
1229
1230                 ixgbevf_write_eitr(q_vector);
1231         }
1232 }
1233
1234 static irqreturn_t ixgbevf_msix_other(int irq, void *data)
1235 {
1236         struct ixgbevf_adapter *adapter = data;
1237         struct ixgbe_hw *hw = &adapter->hw;
1238
1239         hw->mac.get_link_status = 1;
1240
1241         ixgbevf_service_event_schedule(adapter);
1242
1243         IXGBE_WRITE_REG(hw, IXGBE_VTEIMS, adapter->eims_other);
1244
1245         return IRQ_HANDLED;
1246 }
1247
1248 /**
1249  * ixgbevf_msix_clean_rings - single unshared vector rx clean (all queues)
1250  * @irq: unused
1251  * @data: pointer to our q_vector struct for this interrupt vector
1252  **/
1253 static irqreturn_t ixgbevf_msix_clean_rings(int irq, void *data)
1254 {
1255         struct ixgbevf_q_vector *q_vector = data;
1256
1257         /* EIAM disabled interrupts (on this vector) for us */
1258         if (q_vector->rx.ring || q_vector->tx.ring)
1259                 napi_schedule_irqoff(&q_vector->napi);
1260
1261         return IRQ_HANDLED;
1262 }
1263
1264 static inline void map_vector_to_rxq(struct ixgbevf_adapter *a, int v_idx,
1265                                      int r_idx)
1266 {
1267         struct ixgbevf_q_vector *q_vector = a->q_vector[v_idx];
1268
1269         a->rx_ring[r_idx]->next = q_vector->rx.ring;
1270         q_vector->rx.ring = a->rx_ring[r_idx];
1271         q_vector->rx.count++;
1272 }
1273
1274 static inline void map_vector_to_txq(struct ixgbevf_adapter *a, int v_idx,
1275                                      int t_idx)
1276 {
1277         struct ixgbevf_q_vector *q_vector = a->q_vector[v_idx];
1278
1279         a->tx_ring[t_idx]->next = q_vector->tx.ring;
1280         q_vector->tx.ring = a->tx_ring[t_idx];
1281         q_vector->tx.count++;
1282 }
1283
1284 /**
1285  * ixgbevf_map_rings_to_vectors - Maps descriptor rings to vectors
1286  * @adapter: board private structure to initialize
1287  *
1288  * This function maps descriptor rings to the queue-specific vectors
1289  * we were allotted through the MSI-X enabling code.  Ideally, we'd have
1290  * one vector per ring/queue, but on a constrained vector budget, we
1291  * group the rings as "efficiently" as possible.  You would add new
1292  * mapping configurations in here.
1293  **/
1294 static int ixgbevf_map_rings_to_vectors(struct ixgbevf_adapter *adapter)
1295 {
1296         int q_vectors;
1297         int v_start = 0;
1298         int rxr_idx = 0, txr_idx = 0;
1299         int rxr_remaining = adapter->num_rx_queues;
1300         int txr_remaining = adapter->num_tx_queues;
1301         int i, j;
1302         int rqpv, tqpv;
1303
1304         q_vectors = adapter->num_msix_vectors - NON_Q_VECTORS;
1305
1306         /* The ideal configuration...
1307          * We have enough vectors to map one per queue.
1308          */
1309         if (q_vectors == adapter->num_rx_queues + adapter->num_tx_queues) {
1310                 for (; rxr_idx < rxr_remaining; v_start++, rxr_idx++)
1311                         map_vector_to_rxq(adapter, v_start, rxr_idx);
1312
1313                 for (; txr_idx < txr_remaining; v_start++, txr_idx++)
1314                         map_vector_to_txq(adapter, v_start, txr_idx);
1315                 return 0;
1316         }
1317
1318         /* If we don't have enough vectors for a 1-to-1
1319          * mapping, we'll have to group them so there are
1320          * multiple queues per vector.
1321          */
1322         /* Re-adjusting *qpv takes care of the remainder. */
1323         for (i = v_start; i < q_vectors; i++) {
1324                 rqpv = DIV_ROUND_UP(rxr_remaining, q_vectors - i);
1325                 for (j = 0; j < rqpv; j++) {
1326                         map_vector_to_rxq(adapter, i, rxr_idx);
1327                         rxr_idx++;
1328                         rxr_remaining--;
1329                 }
1330         }
1331         for (i = v_start; i < q_vectors; i++) {
1332                 tqpv = DIV_ROUND_UP(txr_remaining, q_vectors - i);
1333                 for (j = 0; j < tqpv; j++) {
1334                         map_vector_to_txq(adapter, i, txr_idx);
1335                         txr_idx++;
1336                         txr_remaining--;
1337                 }
1338         }
1339
1340         return 0;
1341 }
1342
1343 /**
1344  * ixgbevf_request_msix_irqs - Initialize MSI-X interrupts
1345  * @adapter: board private structure
1346  *
1347  * ixgbevf_request_msix_irqs allocates MSI-X vectors and requests
1348  * interrupts from the kernel.
1349  **/
1350 static int ixgbevf_request_msix_irqs(struct ixgbevf_adapter *adapter)
1351 {
1352         struct net_device *netdev = adapter->netdev;
1353         int q_vectors = adapter->num_msix_vectors - NON_Q_VECTORS;
1354         unsigned int ri = 0, ti = 0;
1355         int vector, err;
1356
1357         for (vector = 0; vector < q_vectors; vector++) {
1358                 struct ixgbevf_q_vector *q_vector = adapter->q_vector[vector];
1359                 struct msix_entry *entry = &adapter->msix_entries[vector];
1360
1361                 if (q_vector->tx.ring && q_vector->rx.ring) {
1362                         snprintf(q_vector->name, sizeof(q_vector->name),
1363                                  "%s-TxRx-%u", netdev->name, ri++);
1364                         ti++;
1365                 } else if (q_vector->rx.ring) {
1366                         snprintf(q_vector->name, sizeof(q_vector->name),
1367                                  "%s-rx-%u", netdev->name, ri++);
1368                 } else if (q_vector->tx.ring) {
1369                         snprintf(q_vector->name, sizeof(q_vector->name),
1370                                  "%s-tx-%u", netdev->name, ti++);
1371                 } else {
1372                         /* skip this unused q_vector */
1373                         continue;
1374                 }
1375                 err = request_irq(entry->vector, &ixgbevf_msix_clean_rings, 0,
1376                                   q_vector->name, q_vector);
1377                 if (err) {
1378                         hw_dbg(&adapter->hw,
1379                                "request_irq failed for MSIX interrupt Error: %d\n",
1380                                err);
1381                         goto free_queue_irqs;
1382                 }
1383         }
1384
1385         err = request_irq(adapter->msix_entries[vector].vector,
1386                           &ixgbevf_msix_other, 0, netdev->name, adapter);
1387         if (err) {
1388                 hw_dbg(&adapter->hw, "request_irq for msix_other failed: %d\n",
1389                        err);
1390                 goto free_queue_irqs;
1391         }
1392
1393         return 0;
1394
1395 free_queue_irqs:
1396         while (vector) {
1397                 vector--;
1398                 free_irq(adapter->msix_entries[vector].vector,
1399                          adapter->q_vector[vector]);
1400         }
1401         /* This failure is non-recoverable - it indicates the system is
1402          * out of MSIX vector resources and the VF driver cannot run
1403          * without them.  Set the number of msix vectors to zero
1404          * indicating that not enough can be allocated.  The error
1405          * will be returned to the user indicating device open failed.
1406          * Any further attempts to force the driver to open will also
1407          * fail.  The only way to recover is to unload the driver and
1408          * reload it again.  If the system has recovered some MSIX
1409          * vectors then it may succeed.
1410          */
1411         adapter->num_msix_vectors = 0;
1412         return err;
1413 }
1414
1415 static inline void ixgbevf_reset_q_vectors(struct ixgbevf_adapter *adapter)
1416 {
1417         int i, q_vectors = adapter->num_msix_vectors - NON_Q_VECTORS;
1418
1419         for (i = 0; i < q_vectors; i++) {
1420                 struct ixgbevf_q_vector *q_vector = adapter->q_vector[i];
1421
1422                 q_vector->rx.ring = NULL;
1423                 q_vector->tx.ring = NULL;
1424                 q_vector->rx.count = 0;
1425                 q_vector->tx.count = 0;
1426         }
1427 }
1428
1429 /**
1430  * ixgbevf_request_irq - initialize interrupts
1431  * @adapter: board private structure
1432  *
1433  * Attempts to configure interrupts using the best available
1434  * capabilities of the hardware and kernel.
1435  **/
1436 static int ixgbevf_request_irq(struct ixgbevf_adapter *adapter)
1437 {
1438         int err = ixgbevf_request_msix_irqs(adapter);
1439
1440         if (err)
1441                 hw_dbg(&adapter->hw, "request_irq failed, Error %d\n", err);
1442
1443         return err;
1444 }
1445
1446 static void ixgbevf_free_irq(struct ixgbevf_adapter *adapter)
1447 {
1448         int i, q_vectors;
1449
1450         if (!adapter->msix_entries)
1451                 return;
1452
1453         q_vectors = adapter->num_msix_vectors;
1454         i = q_vectors - 1;
1455
1456         free_irq(adapter->msix_entries[i].vector, adapter);
1457         i--;
1458
1459         for (; i >= 0; i--) {
1460                 /* free only the irqs that were actually requested */
1461                 if (!adapter->q_vector[i]->rx.ring &&
1462                     !adapter->q_vector[i]->tx.ring)
1463                         continue;
1464
1465                 free_irq(adapter->msix_entries[i].vector,
1466                          adapter->q_vector[i]);
1467         }
1468
1469         ixgbevf_reset_q_vectors(adapter);
1470 }
1471
1472 /**
1473  * ixgbevf_irq_disable - Mask off interrupt generation on the NIC
1474  * @adapter: board private structure
1475  **/
1476 static inline void ixgbevf_irq_disable(struct ixgbevf_adapter *adapter)
1477 {
1478         struct ixgbe_hw *hw = &adapter->hw;
1479         int i;
1480
1481         IXGBE_WRITE_REG(hw, IXGBE_VTEIAM, 0);
1482         IXGBE_WRITE_REG(hw, IXGBE_VTEIMC, ~0);
1483         IXGBE_WRITE_REG(hw, IXGBE_VTEIAC, 0);
1484
1485         IXGBE_WRITE_FLUSH(hw);
1486
1487         for (i = 0; i < adapter->num_msix_vectors; i++)
1488                 synchronize_irq(adapter->msix_entries[i].vector);
1489 }
1490
1491 /**
1492  * ixgbevf_irq_enable - Enable default interrupt generation settings
1493  * @adapter: board private structure
1494  **/
1495 static inline void ixgbevf_irq_enable(struct ixgbevf_adapter *adapter)
1496 {
1497         struct ixgbe_hw *hw = &adapter->hw;
1498
1499         IXGBE_WRITE_REG(hw, IXGBE_VTEIAM, adapter->eims_enable_mask);
1500         IXGBE_WRITE_REG(hw, IXGBE_VTEIAC, adapter->eims_enable_mask);
1501         IXGBE_WRITE_REG(hw, IXGBE_VTEIMS, adapter->eims_enable_mask);
1502 }
1503
1504 /**
1505  * ixgbevf_configure_tx_ring - Configure 82599 VF Tx ring after Reset
1506  * @adapter: board private structure
1507  * @ring: structure containing ring specific data
1508  *
1509  * Configure the Tx descriptor ring after a reset.
1510  **/
1511 static void ixgbevf_configure_tx_ring(struct ixgbevf_adapter *adapter,
1512                                       struct ixgbevf_ring *ring)
1513 {
1514         struct ixgbe_hw *hw = &adapter->hw;
1515         u64 tdba = ring->dma;
1516         int wait_loop = 10;
1517         u32 txdctl = IXGBE_TXDCTL_ENABLE;
1518         u8 reg_idx = ring->reg_idx;
1519
1520         /* disable queue to avoid issues while updating state */
1521         IXGBE_WRITE_REG(hw, IXGBE_VFTXDCTL(reg_idx), IXGBE_TXDCTL_SWFLSH);
1522         IXGBE_WRITE_FLUSH(hw);
1523
1524         IXGBE_WRITE_REG(hw, IXGBE_VFTDBAL(reg_idx), tdba & DMA_BIT_MASK(32));
1525         IXGBE_WRITE_REG(hw, IXGBE_VFTDBAH(reg_idx), tdba >> 32);
1526         IXGBE_WRITE_REG(hw, IXGBE_VFTDLEN(reg_idx),
1527                         ring->count * sizeof(union ixgbe_adv_tx_desc));
1528
1529         /* disable head writeback */
1530         IXGBE_WRITE_REG(hw, IXGBE_VFTDWBAH(reg_idx), 0);
1531         IXGBE_WRITE_REG(hw, IXGBE_VFTDWBAL(reg_idx), 0);
1532
1533         /* enable relaxed ordering */
1534         IXGBE_WRITE_REG(hw, IXGBE_VFDCA_TXCTRL(reg_idx),
1535                         (IXGBE_DCA_TXCTRL_DESC_RRO_EN |
1536                          IXGBE_DCA_TXCTRL_DATA_RRO_EN));
1537
1538         /* reset head and tail pointers */
1539         IXGBE_WRITE_REG(hw, IXGBE_VFTDH(reg_idx), 0);
1540         IXGBE_WRITE_REG(hw, IXGBE_VFTDT(reg_idx), 0);
1541         ring->tail = adapter->io_addr + IXGBE_VFTDT(reg_idx);
1542
1543         /* reset ntu and ntc to place SW in sync with hardwdare */
1544         ring->next_to_clean = 0;
1545         ring->next_to_use = 0;
1546
1547         /* In order to avoid issues WTHRESH + PTHRESH should always be equal
1548          * to or less than the number of on chip descriptors, which is
1549          * currently 40.
1550          */
1551         txdctl |= (8 << 16);    /* WTHRESH = 8 */
1552
1553         /* Setting PTHRESH to 32 both improves performance */
1554         txdctl |= (1u << 8) |    /* HTHRESH = 1 */
1555                    32;           /* PTHRESH = 32 */
1556
1557         clear_bit(__IXGBEVF_HANG_CHECK_ARMED, &ring->state);
1558
1559         IXGBE_WRITE_REG(hw, IXGBE_VFTXDCTL(reg_idx), txdctl);
1560
1561         /* poll to verify queue is enabled */
1562         do {
1563                 usleep_range(1000, 2000);
1564                 txdctl = IXGBE_READ_REG(hw, IXGBE_VFTXDCTL(reg_idx));
1565         }  while (--wait_loop && !(txdctl & IXGBE_TXDCTL_ENABLE));
1566         if (!wait_loop)
1567                 hw_dbg(hw, "Could not enable Tx Queue %d\n", reg_idx);
1568 }
1569
1570 /**
1571  * ixgbevf_configure_tx - Configure 82599 VF Transmit Unit after Reset
1572  * @adapter: board private structure
1573  *
1574  * Configure the Tx unit of the MAC after a reset.
1575  **/
1576 static void ixgbevf_configure_tx(struct ixgbevf_adapter *adapter)
1577 {
1578         u32 i;
1579
1580         /* Setup the HW Tx Head and Tail descriptor pointers */
1581         for (i = 0; i < adapter->num_tx_queues; i++)
1582                 ixgbevf_configure_tx_ring(adapter, adapter->tx_ring[i]);
1583 }
1584
1585 #define IXGBE_SRRCTL_BSIZEHDRSIZE_SHIFT 2
1586
1587 static void ixgbevf_configure_srrctl(struct ixgbevf_adapter *adapter, int index)
1588 {
1589         struct ixgbe_hw *hw = &adapter->hw;
1590         u32 srrctl;
1591
1592         srrctl = IXGBE_SRRCTL_DROP_EN;
1593
1594         srrctl |= IXGBEVF_RX_HDR_SIZE << IXGBE_SRRCTL_BSIZEHDRSIZE_SHIFT;
1595         srrctl |= IXGBEVF_RX_BUFSZ >> IXGBE_SRRCTL_BSIZEPKT_SHIFT;
1596         srrctl |= IXGBE_SRRCTL_DESCTYPE_ADV_ONEBUF;
1597
1598         IXGBE_WRITE_REG(hw, IXGBE_VFSRRCTL(index), srrctl);
1599 }
1600
1601 static void ixgbevf_setup_psrtype(struct ixgbevf_adapter *adapter)
1602 {
1603         struct ixgbe_hw *hw = &adapter->hw;
1604
1605         /* PSRTYPE must be initialized in 82599 */
1606         u32 psrtype = IXGBE_PSRTYPE_TCPHDR | IXGBE_PSRTYPE_UDPHDR |
1607                       IXGBE_PSRTYPE_IPV4HDR | IXGBE_PSRTYPE_IPV6HDR |
1608                       IXGBE_PSRTYPE_L2HDR;
1609
1610         if (adapter->num_rx_queues > 1)
1611                 psrtype |= BIT(29);
1612
1613         IXGBE_WRITE_REG(hw, IXGBE_VFPSRTYPE, psrtype);
1614 }
1615
1616 #define IXGBEVF_MAX_RX_DESC_POLL 10
1617 static void ixgbevf_disable_rx_queue(struct ixgbevf_adapter *adapter,
1618                                      struct ixgbevf_ring *ring)
1619 {
1620         struct ixgbe_hw *hw = &adapter->hw;
1621         int wait_loop = IXGBEVF_MAX_RX_DESC_POLL;
1622         u32 rxdctl;
1623         u8 reg_idx = ring->reg_idx;
1624
1625         if (IXGBE_REMOVED(hw->hw_addr))
1626                 return;
1627         rxdctl = IXGBE_READ_REG(hw, IXGBE_VFRXDCTL(reg_idx));
1628         rxdctl &= ~IXGBE_RXDCTL_ENABLE;
1629
1630         /* write value back with RXDCTL.ENABLE bit cleared */
1631         IXGBE_WRITE_REG(hw, IXGBE_VFRXDCTL(reg_idx), rxdctl);
1632
1633         /* the hardware may take up to 100us to really disable the Rx queue */
1634         do {
1635                 udelay(10);
1636                 rxdctl = IXGBE_READ_REG(hw, IXGBE_VFRXDCTL(reg_idx));
1637         } while (--wait_loop && (rxdctl & IXGBE_RXDCTL_ENABLE));
1638
1639         if (!wait_loop)
1640                 pr_err("RXDCTL.ENABLE queue %d not cleared while polling\n",
1641                        reg_idx);
1642 }
1643
1644 static void ixgbevf_rx_desc_queue_enable(struct ixgbevf_adapter *adapter,
1645                                          struct ixgbevf_ring *ring)
1646 {
1647         struct ixgbe_hw *hw = &adapter->hw;
1648         int wait_loop = IXGBEVF_MAX_RX_DESC_POLL;
1649         u32 rxdctl;
1650         u8 reg_idx = ring->reg_idx;
1651
1652         if (IXGBE_REMOVED(hw->hw_addr))
1653                 return;
1654         do {
1655                 usleep_range(1000, 2000);
1656                 rxdctl = IXGBE_READ_REG(hw, IXGBE_VFRXDCTL(reg_idx));
1657         } while (--wait_loop && !(rxdctl & IXGBE_RXDCTL_ENABLE));
1658
1659         if (!wait_loop)
1660                 pr_err("RXDCTL.ENABLE queue %d not set while polling\n",
1661                        reg_idx);
1662 }
1663
1664 /**
1665  * ixgbevf_init_rss_key - Initialize adapter RSS key
1666  * @adapter: device handle
1667  *
1668  * Allocates and initializes the RSS key if it is not allocated.
1669  **/
1670 static inline int ixgbevf_init_rss_key(struct ixgbevf_adapter *adapter)
1671 {
1672         u32 *rss_key;
1673
1674         if (!adapter->rss_key) {
1675                 rss_key = kzalloc(IXGBEVF_RSS_HASH_KEY_SIZE, GFP_KERNEL);
1676                 if (unlikely(!rss_key))
1677                         return -ENOMEM;
1678
1679                 netdev_rss_key_fill(rss_key, IXGBEVF_RSS_HASH_KEY_SIZE);
1680                 adapter->rss_key = rss_key;
1681         }
1682
1683         return 0;
1684 }
1685
1686 static void ixgbevf_setup_vfmrqc(struct ixgbevf_adapter *adapter)
1687 {
1688         struct ixgbe_hw *hw = &adapter->hw;
1689         u32 vfmrqc = 0, vfreta = 0;
1690         u16 rss_i = adapter->num_rx_queues;
1691         u8 i, j;
1692
1693         /* Fill out hash function seeds */
1694         for (i = 0; i < IXGBEVF_VFRSSRK_REGS; i++)
1695                 IXGBE_WRITE_REG(hw, IXGBE_VFRSSRK(i), *(adapter->rss_key + i));
1696
1697         for (i = 0, j = 0; i < IXGBEVF_X550_VFRETA_SIZE; i++, j++) {
1698                 if (j == rss_i)
1699                         j = 0;
1700
1701                 adapter->rss_indir_tbl[i] = j;
1702
1703                 vfreta |= j << (i & 0x3) * 8;
1704                 if ((i & 3) == 3) {
1705                         IXGBE_WRITE_REG(hw, IXGBE_VFRETA(i >> 2), vfreta);
1706                         vfreta = 0;
1707                 }
1708         }
1709
1710         /* Perform hash on these packet types */
1711         vfmrqc |= IXGBE_VFMRQC_RSS_FIELD_IPV4 |
1712                 IXGBE_VFMRQC_RSS_FIELD_IPV4_TCP |
1713                 IXGBE_VFMRQC_RSS_FIELD_IPV6 |
1714                 IXGBE_VFMRQC_RSS_FIELD_IPV6_TCP;
1715
1716         vfmrqc |= IXGBE_VFMRQC_RSSEN;
1717
1718         IXGBE_WRITE_REG(hw, IXGBE_VFMRQC, vfmrqc);
1719 }
1720
1721 static void ixgbevf_configure_rx_ring(struct ixgbevf_adapter *adapter,
1722                                       struct ixgbevf_ring *ring)
1723 {
1724         struct ixgbe_hw *hw = &adapter->hw;
1725         u64 rdba = ring->dma;
1726         u32 rxdctl;
1727         u8 reg_idx = ring->reg_idx;
1728
1729         /* disable queue to avoid issues while updating state */
1730         rxdctl = IXGBE_READ_REG(hw, IXGBE_VFRXDCTL(reg_idx));
1731         ixgbevf_disable_rx_queue(adapter, ring);
1732
1733         IXGBE_WRITE_REG(hw, IXGBE_VFRDBAL(reg_idx), rdba & DMA_BIT_MASK(32));
1734         IXGBE_WRITE_REG(hw, IXGBE_VFRDBAH(reg_idx), rdba >> 32);
1735         IXGBE_WRITE_REG(hw, IXGBE_VFRDLEN(reg_idx),
1736                         ring->count * sizeof(union ixgbe_adv_rx_desc));
1737
1738 #ifndef CONFIG_SPARC
1739         /* enable relaxed ordering */
1740         IXGBE_WRITE_REG(hw, IXGBE_VFDCA_RXCTRL(reg_idx),
1741                         IXGBE_DCA_RXCTRL_DESC_RRO_EN);
1742 #else
1743         IXGBE_WRITE_REG(hw, IXGBE_VFDCA_RXCTRL(reg_idx),
1744                         IXGBE_DCA_RXCTRL_DESC_RRO_EN |
1745                         IXGBE_DCA_RXCTRL_DATA_WRO_EN);
1746 #endif
1747
1748         /* reset head and tail pointers */
1749         IXGBE_WRITE_REG(hw, IXGBE_VFRDH(reg_idx), 0);
1750         IXGBE_WRITE_REG(hw, IXGBE_VFRDT(reg_idx), 0);
1751         ring->tail = adapter->io_addr + IXGBE_VFRDT(reg_idx);
1752
1753         /* reset ntu and ntc to place SW in sync with hardwdare */
1754         ring->next_to_clean = 0;
1755         ring->next_to_use = 0;
1756         ring->next_to_alloc = 0;
1757
1758         ixgbevf_configure_srrctl(adapter, reg_idx);
1759
1760         /* allow any size packet since we can handle overflow */
1761         rxdctl &= ~IXGBE_RXDCTL_RLPML_EN;
1762
1763         rxdctl |= IXGBE_RXDCTL_ENABLE | IXGBE_RXDCTL_VME;
1764         IXGBE_WRITE_REG(hw, IXGBE_VFRXDCTL(reg_idx), rxdctl);
1765
1766         ixgbevf_rx_desc_queue_enable(adapter, ring);
1767         ixgbevf_alloc_rx_buffers(ring, ixgbevf_desc_unused(ring));
1768 }
1769
1770 /**
1771  * ixgbevf_configure_rx - Configure 82599 VF Receive Unit after Reset
1772  * @adapter: board private structure
1773  *
1774  * Configure the Rx unit of the MAC after a reset.
1775  **/
1776 static void ixgbevf_configure_rx(struct ixgbevf_adapter *adapter)
1777 {
1778         struct ixgbe_hw *hw = &adapter->hw;
1779         struct net_device *netdev = adapter->netdev;
1780         int i, ret;
1781
1782         ixgbevf_setup_psrtype(adapter);
1783         if (hw->mac.type >= ixgbe_mac_X550_vf)
1784                 ixgbevf_setup_vfmrqc(adapter);
1785
1786         spin_lock_bh(&adapter->mbx_lock);
1787         /* notify the PF of our intent to use this size of frame */
1788         ret = hw->mac.ops.set_rlpml(hw, netdev->mtu + ETH_HLEN + ETH_FCS_LEN);
1789         spin_unlock_bh(&adapter->mbx_lock);
1790         if (ret)
1791                 dev_err(&adapter->pdev->dev,
1792                         "Failed to set MTU at %d\n", netdev->mtu);
1793
1794         /* Setup the HW Rx Head and Tail Descriptor Pointers and
1795          * the Base and Length of the Rx Descriptor Ring
1796          */
1797         for (i = 0; i < adapter->num_rx_queues; i++)
1798                 ixgbevf_configure_rx_ring(adapter, adapter->rx_ring[i]);
1799 }
1800
1801 static int ixgbevf_vlan_rx_add_vid(struct net_device *netdev,
1802                                    __be16 proto, u16 vid)
1803 {
1804         struct ixgbevf_adapter *adapter = netdev_priv(netdev);
1805         struct ixgbe_hw *hw = &adapter->hw;
1806         int err;
1807
1808         spin_lock_bh(&adapter->mbx_lock);
1809
1810         /* add VID to filter table */
1811         err = hw->mac.ops.set_vfta(hw, vid, 0, true);
1812
1813         spin_unlock_bh(&adapter->mbx_lock);
1814
1815         /* translate error return types so error makes sense */
1816         if (err == IXGBE_ERR_MBX)
1817                 return -EIO;
1818
1819         if (err == IXGBE_ERR_INVALID_ARGUMENT)
1820                 return -EACCES;
1821
1822         set_bit(vid, adapter->active_vlans);
1823
1824         return err;
1825 }
1826
1827 static int ixgbevf_vlan_rx_kill_vid(struct net_device *netdev,
1828                                     __be16 proto, u16 vid)
1829 {
1830         struct ixgbevf_adapter *adapter = netdev_priv(netdev);
1831         struct ixgbe_hw *hw = &adapter->hw;
1832         int err;
1833
1834         spin_lock_bh(&adapter->mbx_lock);
1835
1836         /* remove VID from filter table */
1837         err = hw->mac.ops.set_vfta(hw, vid, 0, false);
1838
1839         spin_unlock_bh(&adapter->mbx_lock);
1840
1841         clear_bit(vid, adapter->active_vlans);
1842
1843         return err;
1844 }
1845
1846 static void ixgbevf_restore_vlan(struct ixgbevf_adapter *adapter)
1847 {
1848         u16 vid;
1849
1850         for_each_set_bit(vid, adapter->active_vlans, VLAN_N_VID)
1851                 ixgbevf_vlan_rx_add_vid(adapter->netdev,
1852                                         htons(ETH_P_8021Q), vid);
1853 }
1854
1855 static int ixgbevf_write_uc_addr_list(struct net_device *netdev)
1856 {
1857         struct ixgbevf_adapter *adapter = netdev_priv(netdev);
1858         struct ixgbe_hw *hw = &adapter->hw;
1859         int count = 0;
1860
1861         if (!netdev_uc_empty(netdev)) {
1862                 struct netdev_hw_addr *ha;
1863
1864                 netdev_for_each_uc_addr(ha, netdev) {
1865                         hw->mac.ops.set_uc_addr(hw, ++count, ha->addr);
1866                         udelay(200);
1867                 }
1868         } else {
1869                 /* If the list is empty then send message to PF driver to
1870                  * clear all MAC VLANs on this VF.
1871                  */
1872                 hw->mac.ops.set_uc_addr(hw, 0, NULL);
1873         }
1874
1875         return count;
1876 }
1877
1878 /**
1879  * ixgbevf_set_rx_mode - Multicast and unicast set
1880  * @netdev: network interface device structure
1881  *
1882  * The set_rx_method entry point is called whenever the multicast address
1883  * list, unicast address list or the network interface flags are updated.
1884  * This routine is responsible for configuring the hardware for proper
1885  * multicast mode and configuring requested unicast filters.
1886  **/
1887 static void ixgbevf_set_rx_mode(struct net_device *netdev)
1888 {
1889         struct ixgbevf_adapter *adapter = netdev_priv(netdev);
1890         struct ixgbe_hw *hw = &adapter->hw;
1891         unsigned int flags = netdev->flags;
1892         int xcast_mode;
1893
1894         xcast_mode = (flags & IFF_ALLMULTI) ? IXGBEVF_XCAST_MODE_ALLMULTI :
1895                      (flags & (IFF_BROADCAST | IFF_MULTICAST)) ?
1896                      IXGBEVF_XCAST_MODE_MULTI : IXGBEVF_XCAST_MODE_NONE;
1897
1898         /* request the most inclusive mode we need */
1899         if (flags & IFF_PROMISC)
1900                 xcast_mode = IXGBEVF_XCAST_MODE_PROMISC;
1901         else if (flags & IFF_ALLMULTI)
1902                 xcast_mode = IXGBEVF_XCAST_MODE_ALLMULTI;
1903         else if (flags & (IFF_BROADCAST | IFF_MULTICAST))
1904                 xcast_mode = IXGBEVF_XCAST_MODE_MULTI;
1905         else
1906                 xcast_mode = IXGBEVF_XCAST_MODE_NONE;
1907
1908         spin_lock_bh(&adapter->mbx_lock);
1909
1910         hw->mac.ops.update_xcast_mode(hw, xcast_mode);
1911
1912         /* reprogram multicast list */
1913         hw->mac.ops.update_mc_addr_list(hw, netdev);
1914
1915         ixgbevf_write_uc_addr_list(netdev);
1916
1917         spin_unlock_bh(&adapter->mbx_lock);
1918 }
1919
1920 static void ixgbevf_napi_enable_all(struct ixgbevf_adapter *adapter)
1921 {
1922         int q_idx;
1923         struct ixgbevf_q_vector *q_vector;
1924         int q_vectors = adapter->num_msix_vectors - NON_Q_VECTORS;
1925
1926         for (q_idx = 0; q_idx < q_vectors; q_idx++) {
1927                 q_vector = adapter->q_vector[q_idx];
1928                 napi_enable(&q_vector->napi);
1929         }
1930 }
1931
1932 static void ixgbevf_napi_disable_all(struct ixgbevf_adapter *adapter)
1933 {
1934         int q_idx;
1935         struct ixgbevf_q_vector *q_vector;
1936         int q_vectors = adapter->num_msix_vectors - NON_Q_VECTORS;
1937
1938         for (q_idx = 0; q_idx < q_vectors; q_idx++) {
1939                 q_vector = adapter->q_vector[q_idx];
1940                 napi_disable(&q_vector->napi);
1941         }
1942 }
1943
1944 static int ixgbevf_configure_dcb(struct ixgbevf_adapter *adapter)
1945 {
1946         struct ixgbe_hw *hw = &adapter->hw;
1947         unsigned int def_q = 0;
1948         unsigned int num_tcs = 0;
1949         unsigned int num_rx_queues = adapter->num_rx_queues;
1950         unsigned int num_tx_queues = adapter->num_tx_queues;
1951         int err;
1952
1953         spin_lock_bh(&adapter->mbx_lock);
1954
1955         /* fetch queue configuration from the PF */
1956         err = ixgbevf_get_queues(hw, &num_tcs, &def_q);
1957
1958         spin_unlock_bh(&adapter->mbx_lock);
1959
1960         if (err)
1961                 return err;
1962
1963         if (num_tcs > 1) {
1964                 /* we need only one Tx queue */
1965                 num_tx_queues = 1;
1966
1967                 /* update default Tx ring register index */
1968                 adapter->tx_ring[0]->reg_idx = def_q;
1969
1970                 /* we need as many queues as traffic classes */
1971                 num_rx_queues = num_tcs;
1972         }
1973
1974         /* if we have a bad config abort request queue reset */
1975         if ((adapter->num_rx_queues != num_rx_queues) ||
1976             (adapter->num_tx_queues != num_tx_queues)) {
1977                 /* force mailbox timeout to prevent further messages */
1978                 hw->mbx.timeout = 0;
1979
1980                 /* wait for watchdog to come around and bail us out */
1981                 set_bit(__IXGBEVF_QUEUE_RESET_REQUESTED, &adapter->state);
1982         }
1983
1984         return 0;
1985 }
1986
1987 static void ixgbevf_configure(struct ixgbevf_adapter *adapter)
1988 {
1989         ixgbevf_configure_dcb(adapter);
1990
1991         ixgbevf_set_rx_mode(adapter->netdev);
1992
1993         ixgbevf_restore_vlan(adapter);
1994
1995         ixgbevf_configure_tx(adapter);
1996         ixgbevf_configure_rx(adapter);
1997 }
1998
1999 static void ixgbevf_save_reset_stats(struct ixgbevf_adapter *adapter)
2000 {
2001         /* Only save pre-reset stats if there are some */
2002         if (adapter->stats.vfgprc || adapter->stats.vfgptc) {
2003                 adapter->stats.saved_reset_vfgprc += adapter->stats.vfgprc -
2004                         adapter->stats.base_vfgprc;
2005                 adapter->stats.saved_reset_vfgptc += adapter->stats.vfgptc -
2006                         adapter->stats.base_vfgptc;
2007                 adapter->stats.saved_reset_vfgorc += adapter->stats.vfgorc -
2008                         adapter->stats.base_vfgorc;
2009                 adapter->stats.saved_reset_vfgotc += adapter->stats.vfgotc -
2010                         adapter->stats.base_vfgotc;
2011                 adapter->stats.saved_reset_vfmprc += adapter->stats.vfmprc -
2012                         adapter->stats.base_vfmprc;
2013         }
2014 }
2015
2016 static void ixgbevf_init_last_counter_stats(struct ixgbevf_adapter *adapter)
2017 {
2018         struct ixgbe_hw *hw = &adapter->hw;
2019
2020         adapter->stats.last_vfgprc = IXGBE_READ_REG(hw, IXGBE_VFGPRC);
2021         adapter->stats.last_vfgorc = IXGBE_READ_REG(hw, IXGBE_VFGORC_LSB);
2022         adapter->stats.last_vfgorc |=
2023                 (((u64)(IXGBE_READ_REG(hw, IXGBE_VFGORC_MSB))) << 32);
2024         adapter->stats.last_vfgptc = IXGBE_READ_REG(hw, IXGBE_VFGPTC);
2025         adapter->stats.last_vfgotc = IXGBE_READ_REG(hw, IXGBE_VFGOTC_LSB);
2026         adapter->stats.last_vfgotc |=
2027                 (((u64)(IXGBE_READ_REG(hw, IXGBE_VFGOTC_MSB))) << 32);
2028         adapter->stats.last_vfmprc = IXGBE_READ_REG(hw, IXGBE_VFMPRC);
2029
2030         adapter->stats.base_vfgprc = adapter->stats.last_vfgprc;
2031         adapter->stats.base_vfgorc = adapter->stats.last_vfgorc;
2032         adapter->stats.base_vfgptc = adapter->stats.last_vfgptc;
2033         adapter->stats.base_vfgotc = adapter->stats.last_vfgotc;
2034         adapter->stats.base_vfmprc = adapter->stats.last_vfmprc;
2035 }
2036
2037 static void ixgbevf_negotiate_api(struct ixgbevf_adapter *adapter)
2038 {
2039         struct ixgbe_hw *hw = &adapter->hw;
2040         int api[] = { ixgbe_mbox_api_13,
2041                       ixgbe_mbox_api_12,
2042                       ixgbe_mbox_api_11,
2043                       ixgbe_mbox_api_10,
2044                       ixgbe_mbox_api_unknown };
2045         int err, idx = 0;
2046
2047         spin_lock_bh(&adapter->mbx_lock);
2048
2049         while (api[idx] != ixgbe_mbox_api_unknown) {
2050                 err = hw->mac.ops.negotiate_api_version(hw, api[idx]);
2051                 if (!err)
2052                         break;
2053                 idx++;
2054         }
2055
2056         spin_unlock_bh(&adapter->mbx_lock);
2057 }
2058
2059 static void ixgbevf_up_complete(struct ixgbevf_adapter *adapter)
2060 {
2061         struct net_device *netdev = adapter->netdev;
2062         struct ixgbe_hw *hw = &adapter->hw;
2063
2064         ixgbevf_configure_msix(adapter);
2065
2066         spin_lock_bh(&adapter->mbx_lock);
2067
2068         if (is_valid_ether_addr(hw->mac.addr))
2069                 hw->mac.ops.set_rar(hw, 0, hw->mac.addr, 0);
2070         else
2071                 hw->mac.ops.set_rar(hw, 0, hw->mac.perm_addr, 0);
2072
2073         spin_unlock_bh(&adapter->mbx_lock);
2074
2075         smp_mb__before_atomic();
2076         clear_bit(__IXGBEVF_DOWN, &adapter->state);
2077         ixgbevf_napi_enable_all(adapter);
2078
2079         /* clear any pending interrupts, may auto mask */
2080         IXGBE_READ_REG(hw, IXGBE_VTEICR);
2081         ixgbevf_irq_enable(adapter);
2082
2083         /* enable transmits */
2084         netif_tx_start_all_queues(netdev);
2085
2086         ixgbevf_save_reset_stats(adapter);
2087         ixgbevf_init_last_counter_stats(adapter);
2088
2089         hw->mac.get_link_status = 1;
2090         mod_timer(&adapter->service_timer, jiffies);
2091 }
2092
2093 void ixgbevf_up(struct ixgbevf_adapter *adapter)
2094 {
2095         ixgbevf_configure(adapter);
2096
2097         ixgbevf_up_complete(adapter);
2098 }
2099
2100 /**
2101  * ixgbevf_clean_rx_ring - Free Rx Buffers per Queue
2102  * @rx_ring: ring to free buffers from
2103  **/
2104 static void ixgbevf_clean_rx_ring(struct ixgbevf_ring *rx_ring)
2105 {
2106         struct device *dev = rx_ring->dev;
2107         unsigned long size;
2108         unsigned int i;
2109
2110         /* Free Rx ring sk_buff */
2111         if (rx_ring->skb) {
2112                 dev_kfree_skb(rx_ring->skb);
2113                 rx_ring->skb = NULL;
2114         }
2115
2116         /* ring already cleared, nothing to do */
2117         if (!rx_ring->rx_buffer_info)
2118                 return;
2119
2120         /* Free all the Rx ring pages */
2121         for (i = 0; i < rx_ring->count; i++) {
2122                 struct ixgbevf_rx_buffer *rx_buffer;
2123
2124                 rx_buffer = &rx_ring->rx_buffer_info[i];
2125                 if (rx_buffer->dma)
2126                         dma_unmap_page(dev, rx_buffer->dma,
2127                                        PAGE_SIZE, DMA_FROM_DEVICE);
2128                 rx_buffer->dma = 0;
2129                 if (rx_buffer->page)
2130                         __free_page(rx_buffer->page);
2131                 rx_buffer->page = NULL;
2132         }
2133
2134         size = sizeof(struct ixgbevf_rx_buffer) * rx_ring->count;
2135         memset(rx_ring->rx_buffer_info, 0, size);
2136
2137         /* Zero out the descriptor ring */
2138         memset(rx_ring->desc, 0, rx_ring->size);
2139 }
2140
2141 /**
2142  * ixgbevf_clean_tx_ring - Free Tx Buffers
2143  * @tx_ring: ring to be cleaned
2144  **/
2145 static void ixgbevf_clean_tx_ring(struct ixgbevf_ring *tx_ring)
2146 {
2147         struct ixgbevf_tx_buffer *tx_buffer_info;
2148         unsigned long size;
2149         unsigned int i;
2150
2151         if (!tx_ring->tx_buffer_info)
2152                 return;
2153
2154         /* Free all the Tx ring sk_buffs */
2155         for (i = 0; i < tx_ring->count; i++) {
2156                 tx_buffer_info = &tx_ring->tx_buffer_info[i];
2157                 ixgbevf_unmap_and_free_tx_resource(tx_ring, tx_buffer_info);
2158         }
2159
2160         size = sizeof(struct ixgbevf_tx_buffer) * tx_ring->count;
2161         memset(tx_ring->tx_buffer_info, 0, size);
2162
2163         memset(tx_ring->desc, 0, tx_ring->size);
2164 }
2165
2166 /**
2167  * ixgbevf_clean_all_rx_rings - Free Rx Buffers for all queues
2168  * @adapter: board private structure
2169  **/
2170 static void ixgbevf_clean_all_rx_rings(struct ixgbevf_adapter *adapter)
2171 {
2172         int i;
2173
2174         for (i = 0; i < adapter->num_rx_queues; i++)
2175                 ixgbevf_clean_rx_ring(adapter->rx_ring[i]);
2176 }
2177
2178 /**
2179  * ixgbevf_clean_all_tx_rings - Free Tx Buffers for all queues
2180  * @adapter: board private structure
2181  **/
2182 static void ixgbevf_clean_all_tx_rings(struct ixgbevf_adapter *adapter)
2183 {
2184         int i;
2185
2186         for (i = 0; i < adapter->num_tx_queues; i++)
2187                 ixgbevf_clean_tx_ring(adapter->tx_ring[i]);
2188 }
2189
2190 void ixgbevf_down(struct ixgbevf_adapter *adapter)
2191 {
2192         struct net_device *netdev = adapter->netdev;
2193         struct ixgbe_hw *hw = &adapter->hw;
2194         int i;
2195
2196         /* signal that we are down to the interrupt handler */
2197         if (test_and_set_bit(__IXGBEVF_DOWN, &adapter->state))
2198                 return; /* do nothing if already down */
2199
2200         /* disable all enabled Rx queues */
2201         for (i = 0; i < adapter->num_rx_queues; i++)
2202                 ixgbevf_disable_rx_queue(adapter, adapter->rx_ring[i]);
2203
2204         usleep_range(10000, 20000);
2205
2206         netif_tx_stop_all_queues(netdev);
2207
2208         /* call carrier off first to avoid false dev_watchdog timeouts */
2209         netif_carrier_off(netdev);
2210         netif_tx_disable(netdev);
2211
2212         ixgbevf_irq_disable(adapter);
2213
2214         ixgbevf_napi_disable_all(adapter);
2215
2216         del_timer_sync(&adapter->service_timer);
2217
2218         /* disable transmits in the hardware now that interrupts are off */
2219         for (i = 0; i < adapter->num_tx_queues; i++) {
2220                 u8 reg_idx = adapter->tx_ring[i]->reg_idx;
2221
2222                 IXGBE_WRITE_REG(hw, IXGBE_VFTXDCTL(reg_idx),
2223                                 IXGBE_TXDCTL_SWFLSH);
2224         }
2225
2226         if (!pci_channel_offline(adapter->pdev))
2227                 ixgbevf_reset(adapter);
2228
2229         ixgbevf_clean_all_tx_rings(adapter);
2230         ixgbevf_clean_all_rx_rings(adapter);
2231 }
2232
2233 void ixgbevf_reinit_locked(struct ixgbevf_adapter *adapter)
2234 {
2235         WARN_ON(in_interrupt());
2236
2237         while (test_and_set_bit(__IXGBEVF_RESETTING, &adapter->state))
2238                 msleep(1);
2239
2240         ixgbevf_down(adapter);
2241         ixgbevf_up(adapter);
2242
2243         clear_bit(__IXGBEVF_RESETTING, &adapter->state);
2244 }
2245
2246 void ixgbevf_reset(struct ixgbevf_adapter *adapter)
2247 {
2248         struct ixgbe_hw *hw = &adapter->hw;
2249         struct net_device *netdev = adapter->netdev;
2250
2251         if (hw->mac.ops.reset_hw(hw)) {
2252                 hw_dbg(hw, "PF still resetting\n");
2253         } else {
2254                 hw->mac.ops.init_hw(hw);
2255                 ixgbevf_negotiate_api(adapter);
2256         }
2257
2258         if (is_valid_ether_addr(adapter->hw.mac.addr)) {
2259                 ether_addr_copy(netdev->dev_addr, adapter->hw.mac.addr);
2260                 ether_addr_copy(netdev->perm_addr, adapter->hw.mac.addr);
2261         }
2262
2263         adapter->last_reset = jiffies;
2264 }
2265
2266 static int ixgbevf_acquire_msix_vectors(struct ixgbevf_adapter *adapter,
2267                                         int vectors)
2268 {
2269         int vector_threshold;
2270
2271         /* We'll want at least 2 (vector_threshold):
2272          * 1) TxQ[0] + RxQ[0] handler
2273          * 2) Other (Link Status Change, etc.)
2274          */
2275         vector_threshold = MIN_MSIX_COUNT;
2276
2277         /* The more we get, the more we will assign to Tx/Rx Cleanup
2278          * for the separate queues...where Rx Cleanup >= Tx Cleanup.
2279          * Right now, we simply care about how many we'll get; we'll
2280          * set them up later while requesting irq's.
2281          */
2282         vectors = pci_enable_msix_range(adapter->pdev, adapter->msix_entries,
2283                                         vector_threshold, vectors);
2284
2285         if (vectors < 0) {
2286                 dev_err(&adapter->pdev->dev,
2287                         "Unable to allocate MSI-X interrupts\n");
2288                 kfree(adapter->msix_entries);
2289                 adapter->msix_entries = NULL;
2290                 return vectors;
2291         }
2292
2293         /* Adjust for only the vectors we'll use, which is minimum
2294          * of max_msix_q_vectors + NON_Q_VECTORS, or the number of
2295          * vectors we were allocated.
2296          */
2297         adapter->num_msix_vectors = vectors;
2298
2299         return 0;
2300 }
2301
2302 /**
2303  * ixgbevf_set_num_queues - Allocate queues for device, feature dependent
2304  * @adapter: board private structure to initialize
2305  *
2306  * This is the top level queue allocation routine.  The order here is very
2307  * important, starting with the "most" number of features turned on at once,
2308  * and ending with the smallest set of features.  This way large combinations
2309  * can be allocated if they're turned on, and smaller combinations are the
2310  * fallthrough conditions.
2311  *
2312  **/
2313 static void ixgbevf_set_num_queues(struct ixgbevf_adapter *adapter)
2314 {
2315         struct ixgbe_hw *hw = &adapter->hw;
2316         unsigned int def_q = 0;
2317         unsigned int num_tcs = 0;
2318         int err;
2319
2320         /* Start with base case */
2321         adapter->num_rx_queues = 1;
2322         adapter->num_tx_queues = 1;
2323
2324         spin_lock_bh(&adapter->mbx_lock);
2325
2326         /* fetch queue configuration from the PF */
2327         err = ixgbevf_get_queues(hw, &num_tcs, &def_q);
2328
2329         spin_unlock_bh(&adapter->mbx_lock);
2330
2331         if (err)
2332                 return;
2333
2334         /* we need as many queues as traffic classes */
2335         if (num_tcs > 1) {
2336                 adapter->num_rx_queues = num_tcs;
2337         } else {
2338                 u16 rss = min_t(u16, num_online_cpus(), IXGBEVF_MAX_RSS_QUEUES);
2339
2340                 switch (hw->api_version) {
2341                 case ixgbe_mbox_api_11:
2342                 case ixgbe_mbox_api_12:
2343                 case ixgbe_mbox_api_13:
2344                         adapter->num_rx_queues = rss;
2345                         adapter->num_tx_queues = rss;
2346                 default:
2347                         break;
2348                 }
2349         }
2350 }
2351
2352 /**
2353  * ixgbevf_alloc_queues - Allocate memory for all rings
2354  * @adapter: board private structure to initialize
2355  *
2356  * We allocate one ring per queue at run-time since we don't know the
2357  * number of queues at compile-time.  The polling_netdev array is
2358  * intended for Multiqueue, but should work fine with a single queue.
2359  **/
2360 static int ixgbevf_alloc_queues(struct ixgbevf_adapter *adapter)
2361 {
2362         struct ixgbevf_ring *ring;
2363         int rx = 0, tx = 0;
2364
2365         for (; tx < adapter->num_tx_queues; tx++) {
2366                 ring = kzalloc(sizeof(*ring), GFP_KERNEL);
2367                 if (!ring)
2368                         goto err_allocation;
2369
2370                 ring->dev = &adapter->pdev->dev;
2371                 ring->netdev = adapter->netdev;
2372                 ring->count = adapter->tx_ring_count;
2373                 ring->queue_index = tx;
2374                 ring->reg_idx = tx;
2375
2376                 adapter->tx_ring[tx] = ring;
2377         }
2378
2379         for (; rx < adapter->num_rx_queues; rx++) {
2380                 ring = kzalloc(sizeof(*ring), GFP_KERNEL);
2381                 if (!ring)
2382                         goto err_allocation;
2383
2384                 ring->dev = &adapter->pdev->dev;
2385                 ring->netdev = adapter->netdev;
2386
2387                 ring->count = adapter->rx_ring_count;
2388                 ring->queue_index = rx;
2389                 ring->reg_idx = rx;
2390
2391                 adapter->rx_ring[rx] = ring;
2392         }
2393
2394         return 0;
2395
2396 err_allocation:
2397         while (tx) {
2398                 kfree(adapter->tx_ring[--tx]);
2399                 adapter->tx_ring[tx] = NULL;
2400         }
2401
2402         while (rx) {
2403                 kfree(adapter->rx_ring[--rx]);
2404                 adapter->rx_ring[rx] = NULL;
2405         }
2406         return -ENOMEM;
2407 }
2408
2409 /**
2410  * ixgbevf_set_interrupt_capability - set MSI-X or FAIL if not supported
2411  * @adapter: board private structure to initialize
2412  *
2413  * Attempt to configure the interrupts using the best available
2414  * capabilities of the hardware and the kernel.
2415  **/
2416 static int ixgbevf_set_interrupt_capability(struct ixgbevf_adapter *adapter)
2417 {
2418         struct net_device *netdev = adapter->netdev;
2419         int err;
2420         int vector, v_budget;
2421
2422         /* It's easy to be greedy for MSI-X vectors, but it really
2423          * doesn't do us much good if we have a lot more vectors
2424          * than CPU's.  So let's be conservative and only ask for
2425          * (roughly) the same number of vectors as there are CPU's.
2426          * The default is to use pairs of vectors.
2427          */
2428         v_budget = max(adapter->num_rx_queues, adapter->num_tx_queues);
2429         v_budget = min_t(int, v_budget, num_online_cpus());
2430         v_budget += NON_Q_VECTORS;
2431
2432         /* A failure in MSI-X entry allocation isn't fatal, but it does
2433          * mean we disable MSI-X capabilities of the adapter.
2434          */
2435         adapter->msix_entries = kcalloc(v_budget,
2436                                         sizeof(struct msix_entry), GFP_KERNEL);
2437         if (!adapter->msix_entries)
2438                 return -ENOMEM;
2439
2440         for (vector = 0; vector < v_budget; vector++)
2441                 adapter->msix_entries[vector].entry = vector;
2442
2443         err = ixgbevf_acquire_msix_vectors(adapter, v_budget);
2444         if (err)
2445                 return err;
2446
2447         err = netif_set_real_num_tx_queues(netdev, adapter->num_tx_queues);
2448         if (err)
2449                 return err;
2450
2451         return netif_set_real_num_rx_queues(netdev, adapter->num_rx_queues);
2452 }
2453
2454 /**
2455  * ixgbevf_alloc_q_vectors - Allocate memory for interrupt vectors
2456  * @adapter: board private structure to initialize
2457  *
2458  * We allocate one q_vector per queue interrupt.  If allocation fails we
2459  * return -ENOMEM.
2460  **/
2461 static int ixgbevf_alloc_q_vectors(struct ixgbevf_adapter *adapter)
2462 {
2463         int q_idx, num_q_vectors;
2464         struct ixgbevf_q_vector *q_vector;
2465
2466         num_q_vectors = adapter->num_msix_vectors - NON_Q_VECTORS;
2467
2468         for (q_idx = 0; q_idx < num_q_vectors; q_idx++) {
2469                 q_vector = kzalloc(sizeof(struct ixgbevf_q_vector), GFP_KERNEL);
2470                 if (!q_vector)
2471                         goto err_out;
2472                 q_vector->adapter = adapter;
2473                 q_vector->v_idx = q_idx;
2474                 netif_napi_add(adapter->netdev, &q_vector->napi,
2475                                ixgbevf_poll, 64);
2476                 adapter->q_vector[q_idx] = q_vector;
2477         }
2478
2479         return 0;
2480
2481 err_out:
2482         while (q_idx) {
2483                 q_idx--;
2484                 q_vector = adapter->q_vector[q_idx];
2485 #ifdef CONFIG_NET_RX_BUSY_POLL
2486                 napi_hash_del(&q_vector->napi);
2487 #endif
2488                 netif_napi_del(&q_vector->napi);
2489                 kfree(q_vector);
2490                 adapter->q_vector[q_idx] = NULL;
2491         }
2492         return -ENOMEM;
2493 }
2494
2495 /**
2496  * ixgbevf_free_q_vectors - Free memory allocated for interrupt vectors
2497  * @adapter: board private structure to initialize
2498  *
2499  * This function frees the memory allocated to the q_vectors.  In addition if
2500  * NAPI is enabled it will delete any references to the NAPI struct prior
2501  * to freeing the q_vector.
2502  **/
2503 static void ixgbevf_free_q_vectors(struct ixgbevf_adapter *adapter)
2504 {
2505         int q_idx, num_q_vectors = adapter->num_msix_vectors - NON_Q_VECTORS;
2506
2507         for (q_idx = 0; q_idx < num_q_vectors; q_idx++) {
2508                 struct ixgbevf_q_vector *q_vector = adapter->q_vector[q_idx];
2509
2510                 adapter->q_vector[q_idx] = NULL;
2511 #ifdef CONFIG_NET_RX_BUSY_POLL
2512                 napi_hash_del(&q_vector->napi);
2513 #endif
2514                 netif_napi_del(&q_vector->napi);
2515                 kfree(q_vector);
2516         }
2517 }
2518
2519 /**
2520  * ixgbevf_reset_interrupt_capability - Reset MSIX setup
2521  * @adapter: board private structure
2522  *
2523  **/
2524 static void ixgbevf_reset_interrupt_capability(struct ixgbevf_adapter *adapter)
2525 {
2526         if (!adapter->msix_entries)
2527                 return;
2528
2529         pci_disable_msix(adapter->pdev);
2530         kfree(adapter->msix_entries);
2531         adapter->msix_entries = NULL;
2532 }
2533
2534 /**
2535  * ixgbevf_init_interrupt_scheme - Determine if MSIX is supported and init
2536  * @adapter: board private structure to initialize
2537  *
2538  **/
2539 static int ixgbevf_init_interrupt_scheme(struct ixgbevf_adapter *adapter)
2540 {
2541         int err;
2542
2543         /* Number of supported queues */
2544         ixgbevf_set_num_queues(adapter);
2545
2546         err = ixgbevf_set_interrupt_capability(adapter);
2547         if (err) {
2548                 hw_dbg(&adapter->hw,
2549                        "Unable to setup interrupt capabilities\n");
2550                 goto err_set_interrupt;
2551         }
2552
2553         err = ixgbevf_alloc_q_vectors(adapter);
2554         if (err) {
2555                 hw_dbg(&adapter->hw, "Unable to allocate memory for queue vectors\n");
2556                 goto err_alloc_q_vectors;
2557         }
2558
2559         err = ixgbevf_alloc_queues(adapter);
2560         if (err) {
2561                 pr_err("Unable to allocate memory for queues\n");
2562                 goto err_alloc_queues;
2563         }
2564
2565         hw_dbg(&adapter->hw, "Multiqueue %s: Rx Queue count = %u, Tx Queue count = %u\n",
2566                (adapter->num_rx_queues > 1) ? "Enabled" :
2567                "Disabled", adapter->num_rx_queues, adapter->num_tx_queues);
2568
2569         set_bit(__IXGBEVF_DOWN, &adapter->state);
2570
2571         return 0;
2572 err_alloc_queues:
2573         ixgbevf_free_q_vectors(adapter);
2574 err_alloc_q_vectors:
2575         ixgbevf_reset_interrupt_capability(adapter);
2576 err_set_interrupt:
2577         return err;
2578 }
2579
2580 /**
2581  * ixgbevf_clear_interrupt_scheme - Clear the current interrupt scheme settings
2582  * @adapter: board private structure to clear interrupt scheme on
2583  *
2584  * We go through and clear interrupt specific resources and reset the structure
2585  * to pre-load conditions
2586  **/
2587 static void ixgbevf_clear_interrupt_scheme(struct ixgbevf_adapter *adapter)
2588 {
2589         int i;
2590
2591         for (i = 0; i < adapter->num_tx_queues; i++) {
2592                 kfree(adapter->tx_ring[i]);
2593                 adapter->tx_ring[i] = NULL;
2594         }
2595         for (i = 0; i < adapter->num_rx_queues; i++) {
2596                 kfree(adapter->rx_ring[i]);
2597                 adapter->rx_ring[i] = NULL;
2598         }
2599
2600         adapter->num_tx_queues = 0;
2601         adapter->num_rx_queues = 0;
2602
2603         ixgbevf_free_q_vectors(adapter);
2604         ixgbevf_reset_interrupt_capability(adapter);
2605 }
2606
2607 /**
2608  * ixgbevf_sw_init - Initialize general software structures
2609  * @adapter: board private structure to initialize
2610  *
2611  * ixgbevf_sw_init initializes the Adapter private data structure.
2612  * Fields are initialized based on PCI device information and
2613  * OS network device settings (MTU size).
2614  **/
2615 static int ixgbevf_sw_init(struct ixgbevf_adapter *adapter)
2616 {
2617         struct ixgbe_hw *hw = &adapter->hw;
2618         struct pci_dev *pdev = adapter->pdev;
2619         struct net_device *netdev = adapter->netdev;
2620         int err;
2621
2622         /* PCI config space info */
2623         hw->vendor_id = pdev->vendor;
2624         hw->device_id = pdev->device;
2625         hw->revision_id = pdev->revision;
2626         hw->subsystem_vendor_id = pdev->subsystem_vendor;
2627         hw->subsystem_device_id = pdev->subsystem_device;
2628
2629         hw->mbx.ops.init_params(hw);
2630
2631         if (hw->mac.type >= ixgbe_mac_X550_vf) {
2632                 err = ixgbevf_init_rss_key(adapter);
2633                 if (err)
2634                         goto out;
2635         }
2636
2637         /* assume legacy case in which PF would only give VF 2 queues */
2638         hw->mac.max_tx_queues = 2;
2639         hw->mac.max_rx_queues = 2;
2640
2641         /* lock to protect mailbox accesses */
2642         spin_lock_init(&adapter->mbx_lock);
2643
2644         err = hw->mac.ops.reset_hw(hw);
2645         if (err) {
2646                 dev_info(&pdev->dev,
2647                          "PF still in reset state.  Is the PF interface up?\n");
2648         } else {
2649                 err = hw->mac.ops.init_hw(hw);
2650                 if (err) {
2651                         pr_err("init_shared_code failed: %d\n", err);
2652                         goto out;
2653                 }
2654                 ixgbevf_negotiate_api(adapter);
2655                 err = hw->mac.ops.get_mac_addr(hw, hw->mac.addr);
2656                 if (err)
2657                         dev_info(&pdev->dev, "Error reading MAC address\n");
2658                 else if (is_zero_ether_addr(adapter->hw.mac.addr))
2659                         dev_info(&pdev->dev,
2660                                  "MAC address not assigned by administrator.\n");
2661                 ether_addr_copy(netdev->dev_addr, hw->mac.addr);
2662         }
2663
2664         if (!is_valid_ether_addr(netdev->dev_addr)) {
2665                 dev_info(&pdev->dev, "Assigning random MAC address\n");
2666                 eth_hw_addr_random(netdev);
2667                 ether_addr_copy(hw->mac.addr, netdev->dev_addr);
2668                 ether_addr_copy(hw->mac.perm_addr, netdev->dev_addr);
2669         }
2670
2671         /* Enable dynamic interrupt throttling rates */
2672         adapter->rx_itr_setting = 1;
2673         adapter->tx_itr_setting = 1;
2674
2675         /* set default ring sizes */
2676         adapter->tx_ring_count = IXGBEVF_DEFAULT_TXD;
2677         adapter->rx_ring_count = IXGBEVF_DEFAULT_RXD;
2678
2679         set_bit(__IXGBEVF_DOWN, &adapter->state);
2680         return 0;
2681
2682 out:
2683         return err;
2684 }
2685
2686 #define UPDATE_VF_COUNTER_32bit(reg, last_counter, counter)     \
2687         {                                                       \
2688                 u32 current_counter = IXGBE_READ_REG(hw, reg);  \
2689                 if (current_counter < last_counter)             \
2690                         counter += 0x100000000LL;               \
2691                 last_counter = current_counter;                 \
2692                 counter &= 0xFFFFFFFF00000000LL;                \
2693                 counter |= current_counter;                     \
2694         }
2695
2696 #define UPDATE_VF_COUNTER_36bit(reg_lsb, reg_msb, last_counter, counter) \
2697         {                                                                \
2698                 u64 current_counter_lsb = IXGBE_READ_REG(hw, reg_lsb);   \
2699                 u64 current_counter_msb = IXGBE_READ_REG(hw, reg_msb);   \
2700                 u64 current_counter = (current_counter_msb << 32) |      \
2701                         current_counter_lsb;                             \
2702                 if (current_counter < last_counter)                      \
2703                         counter += 0x1000000000LL;                       \
2704                 last_counter = current_counter;                          \
2705                 counter &= 0xFFFFFFF000000000LL;                         \
2706                 counter |= current_counter;                              \
2707         }
2708 /**
2709  * ixgbevf_update_stats - Update the board statistics counters.
2710  * @adapter: board private structure
2711  **/
2712 void ixgbevf_update_stats(struct ixgbevf_adapter *adapter)
2713 {
2714         struct ixgbe_hw *hw = &adapter->hw;
2715         int i;
2716
2717         if (test_bit(__IXGBEVF_DOWN, &adapter->state) ||
2718             test_bit(__IXGBEVF_RESETTING, &adapter->state))
2719                 return;
2720
2721         UPDATE_VF_COUNTER_32bit(IXGBE_VFGPRC, adapter->stats.last_vfgprc,
2722                                 adapter->stats.vfgprc);
2723         UPDATE_VF_COUNTER_32bit(IXGBE_VFGPTC, adapter->stats.last_vfgptc,
2724                                 adapter->stats.vfgptc);
2725         UPDATE_VF_COUNTER_36bit(IXGBE_VFGORC_LSB, IXGBE_VFGORC_MSB,
2726                                 adapter->stats.last_vfgorc,
2727                                 adapter->stats.vfgorc);
2728         UPDATE_VF_COUNTER_36bit(IXGBE_VFGOTC_LSB, IXGBE_VFGOTC_MSB,
2729                                 adapter->stats.last_vfgotc,
2730                                 adapter->stats.vfgotc);
2731         UPDATE_VF_COUNTER_32bit(IXGBE_VFMPRC, adapter->stats.last_vfmprc,
2732                                 adapter->stats.vfmprc);
2733
2734         for (i = 0;  i  < adapter->num_rx_queues;  i++) {
2735                 adapter->hw_csum_rx_error +=
2736                         adapter->rx_ring[i]->hw_csum_rx_error;
2737                 adapter->rx_ring[i]->hw_csum_rx_error = 0;
2738         }
2739 }
2740
2741 /**
2742  * ixgbevf_service_timer - Timer Call-back
2743  * @data: pointer to adapter cast into an unsigned long
2744  **/
2745 static void ixgbevf_service_timer(unsigned long data)
2746 {
2747         struct ixgbevf_adapter *adapter = (struct ixgbevf_adapter *)data;
2748
2749         /* Reset the timer */
2750         mod_timer(&adapter->service_timer, (HZ * 2) + jiffies);
2751
2752         ixgbevf_service_event_schedule(adapter);
2753 }
2754
2755 static void ixgbevf_reset_subtask(struct ixgbevf_adapter *adapter)
2756 {
2757         if (!test_and_clear_bit(__IXGBEVF_RESET_REQUESTED, &adapter->state))
2758                 return;
2759
2760         /* If we're already down or resetting, just bail */
2761         if (test_bit(__IXGBEVF_DOWN, &adapter->state) ||
2762             test_bit(__IXGBEVF_REMOVING, &adapter->state) ||
2763             test_bit(__IXGBEVF_RESETTING, &adapter->state))
2764                 return;
2765
2766         adapter->tx_timeout_count++;
2767
2768         rtnl_lock();
2769         ixgbevf_reinit_locked(adapter);
2770         rtnl_unlock();
2771 }
2772
2773 /**
2774  * ixgbevf_check_hang_subtask - check for hung queues and dropped interrupts
2775  * @adapter: pointer to the device adapter structure
2776  *
2777  * This function serves two purposes.  First it strobes the interrupt lines
2778  * in order to make certain interrupts are occurring.  Secondly it sets the
2779  * bits needed to check for TX hangs.  As a result we should immediately
2780  * determine if a hang has occurred.
2781  **/
2782 static void ixgbevf_check_hang_subtask(struct ixgbevf_adapter *adapter)
2783 {
2784         struct ixgbe_hw *hw = &adapter->hw;
2785         u32 eics = 0;
2786         int i;
2787
2788         /* If we're down or resetting, just bail */
2789         if (test_bit(__IXGBEVF_DOWN, &adapter->state) ||
2790             test_bit(__IXGBEVF_RESETTING, &adapter->state))
2791                 return;
2792
2793         /* Force detection of hung controller */
2794         if (netif_carrier_ok(adapter->netdev)) {
2795                 for (i = 0; i < adapter->num_tx_queues; i++)
2796                         set_check_for_tx_hang(adapter->tx_ring[i]);
2797         }
2798
2799         /* get one bit for every active Tx/Rx interrupt vector */
2800         for (i = 0; i < adapter->num_msix_vectors - NON_Q_VECTORS; i++) {
2801                 struct ixgbevf_q_vector *qv = adapter->q_vector[i];
2802
2803                 if (qv->rx.ring || qv->tx.ring)
2804                         eics |= BIT(i);
2805         }
2806
2807         /* Cause software interrupt to ensure rings are cleaned */
2808         IXGBE_WRITE_REG(hw, IXGBE_VTEICS, eics);
2809 }
2810
2811 /**
2812  * ixgbevf_watchdog_update_link - update the link status
2813  * @adapter: pointer to the device adapter structure
2814  **/
2815 static void ixgbevf_watchdog_update_link(struct ixgbevf_adapter *adapter)
2816 {
2817         struct ixgbe_hw *hw = &adapter->hw;
2818         u32 link_speed = adapter->link_speed;
2819         bool link_up = adapter->link_up;
2820         s32 err;
2821
2822         spin_lock_bh(&adapter->mbx_lock);
2823
2824         err = hw->mac.ops.check_link(hw, &link_speed, &link_up, false);
2825
2826         spin_unlock_bh(&adapter->mbx_lock);
2827
2828         /* if check for link returns error we will need to reset */
2829         if (err && time_after(jiffies, adapter->last_reset + (10 * HZ))) {
2830                 set_bit(__IXGBEVF_RESET_REQUESTED, &adapter->state);
2831                 link_up = false;
2832         }
2833
2834         adapter->link_up = link_up;
2835         adapter->link_speed = link_speed;
2836 }
2837
2838 /**
2839  * ixgbevf_watchdog_link_is_up - update netif_carrier status and
2840  *                               print link up message
2841  * @adapter: pointer to the device adapter structure
2842  **/
2843 static void ixgbevf_watchdog_link_is_up(struct ixgbevf_adapter *adapter)
2844 {
2845         struct net_device *netdev = adapter->netdev;
2846
2847         /* only continue if link was previously down */
2848         if (netif_carrier_ok(netdev))
2849                 return;
2850
2851         dev_info(&adapter->pdev->dev, "NIC Link is Up %s\n",
2852                  (adapter->link_speed == IXGBE_LINK_SPEED_10GB_FULL) ?
2853                  "10 Gbps" :
2854                  (adapter->link_speed == IXGBE_LINK_SPEED_1GB_FULL) ?
2855                  "1 Gbps" :
2856                  (adapter->link_speed == IXGBE_LINK_SPEED_100_FULL) ?
2857                  "100 Mbps" :
2858                  "unknown speed");
2859
2860         netif_carrier_on(netdev);
2861 }
2862
2863 /**
2864  * ixgbevf_watchdog_link_is_down - update netif_carrier status and
2865  *                                 print link down message
2866  * @adapter: pointer to the adapter structure
2867  **/
2868 static void ixgbevf_watchdog_link_is_down(struct ixgbevf_adapter *adapter)
2869 {
2870         struct net_device *netdev = adapter->netdev;
2871
2872         adapter->link_speed = 0;
2873
2874         /* only continue if link was up previously */
2875         if (!netif_carrier_ok(netdev))
2876                 return;
2877
2878         dev_info(&adapter->pdev->dev, "NIC Link is Down\n");
2879
2880         netif_carrier_off(netdev);
2881 }
2882
2883 /**
2884  * ixgbevf_watchdog_subtask - worker thread to bring link up
2885  * @work: pointer to work_struct containing our data
2886  **/
2887 static void ixgbevf_watchdog_subtask(struct ixgbevf_adapter *adapter)
2888 {
2889         /* if interface is down do nothing */
2890         if (test_bit(__IXGBEVF_DOWN, &adapter->state) ||
2891             test_bit(__IXGBEVF_RESETTING, &adapter->state))
2892                 return;
2893
2894         ixgbevf_watchdog_update_link(adapter);
2895
2896         if (adapter->link_up)
2897                 ixgbevf_watchdog_link_is_up(adapter);
2898         else
2899                 ixgbevf_watchdog_link_is_down(adapter);
2900
2901         ixgbevf_update_stats(adapter);
2902 }
2903
2904 /**
2905  * ixgbevf_service_task - manages and runs subtasks
2906  * @work: pointer to work_struct containing our data
2907  **/
2908 static void ixgbevf_service_task(struct work_struct *work)
2909 {
2910         struct ixgbevf_adapter *adapter = container_of(work,
2911                                                        struct ixgbevf_adapter,
2912                                                        service_task);
2913         struct ixgbe_hw *hw = &adapter->hw;
2914
2915         if (IXGBE_REMOVED(hw->hw_addr)) {
2916                 if (!test_bit(__IXGBEVF_DOWN, &adapter->state)) {
2917                         rtnl_lock();
2918                         ixgbevf_down(adapter);
2919                         rtnl_unlock();
2920                 }
2921                 return;
2922         }
2923
2924         ixgbevf_queue_reset_subtask(adapter);
2925         ixgbevf_reset_subtask(adapter);
2926         ixgbevf_watchdog_subtask(adapter);
2927         ixgbevf_check_hang_subtask(adapter);
2928
2929         ixgbevf_service_event_complete(adapter);
2930 }
2931
2932 /**
2933  * ixgbevf_free_tx_resources - Free Tx Resources per Queue
2934  * @tx_ring: Tx descriptor ring for a specific queue
2935  *
2936  * Free all transmit software resources
2937  **/
2938 void ixgbevf_free_tx_resources(struct ixgbevf_ring *tx_ring)
2939 {
2940         ixgbevf_clean_tx_ring(tx_ring);
2941
2942         vfree(tx_ring->tx_buffer_info);
2943         tx_ring->tx_buffer_info = NULL;
2944
2945         /* if not set, then don't free */
2946         if (!tx_ring->desc)
2947                 return;
2948
2949         dma_free_coherent(tx_ring->dev, tx_ring->size, tx_ring->desc,
2950                           tx_ring->dma);
2951
2952         tx_ring->desc = NULL;
2953 }
2954
2955 /**
2956  * ixgbevf_free_all_tx_resources - Free Tx Resources for All Queues
2957  * @adapter: board private structure
2958  *
2959  * Free all transmit software resources
2960  **/
2961 static void ixgbevf_free_all_tx_resources(struct ixgbevf_adapter *adapter)
2962 {
2963         int i;
2964
2965         for (i = 0; i < adapter->num_tx_queues; i++)
2966                 if (adapter->tx_ring[i]->desc)
2967                         ixgbevf_free_tx_resources(adapter->tx_ring[i]);
2968 }
2969
2970 /**
2971  * ixgbevf_setup_tx_resources - allocate Tx resources (Descriptors)
2972  * @tx_ring: Tx descriptor ring (for a specific queue) to setup
2973  *
2974  * Return 0 on success, negative on failure
2975  **/
2976 int ixgbevf_setup_tx_resources(struct ixgbevf_ring *tx_ring)
2977 {
2978         struct ixgbevf_adapter *adapter = netdev_priv(tx_ring->netdev);
2979         int size;
2980
2981         size = sizeof(struct ixgbevf_tx_buffer) * tx_ring->count;
2982         tx_ring->tx_buffer_info = vzalloc(size);
2983         if (!tx_ring->tx_buffer_info)
2984                 goto err;
2985
2986         u64_stats_init(&tx_ring->syncp);
2987
2988         /* round up to nearest 4K */
2989         tx_ring->size = tx_ring->count * sizeof(union ixgbe_adv_tx_desc);
2990         tx_ring->size = ALIGN(tx_ring->size, 4096);
2991
2992         tx_ring->desc = dma_alloc_coherent(tx_ring->dev, tx_ring->size,
2993                                            &tx_ring->dma, GFP_KERNEL);
2994         if (!tx_ring->desc)
2995                 goto err;
2996
2997         return 0;
2998
2999 err:
3000         vfree(tx_ring->tx_buffer_info);
3001         tx_ring->tx_buffer_info = NULL;
3002         hw_dbg(&adapter->hw, "Unable to allocate memory for the transmit descriptor ring\n");
3003         return -ENOMEM;
3004 }
3005
3006 /**
3007  * ixgbevf_setup_all_tx_resources - allocate all queues Tx resources
3008  * @adapter: board private structure
3009  *
3010  * If this function returns with an error, then it's possible one or
3011  * more of the rings is populated (while the rest are not).  It is the
3012  * callers duty to clean those orphaned rings.
3013  *
3014  * Return 0 on success, negative on failure
3015  **/
3016 static int ixgbevf_setup_all_tx_resources(struct ixgbevf_adapter *adapter)
3017 {
3018         int i, err = 0;
3019
3020         for (i = 0; i < adapter->num_tx_queues; i++) {
3021                 err = ixgbevf_setup_tx_resources(adapter->tx_ring[i]);
3022                 if (!err)
3023                         continue;
3024                 hw_dbg(&adapter->hw, "Allocation for Tx Queue %u failed\n", i);
3025                 break;
3026         }
3027
3028         return err;
3029 }
3030
3031 /**
3032  * ixgbevf_setup_rx_resources - allocate Rx resources (Descriptors)
3033  * @rx_ring: Rx descriptor ring (for a specific queue) to setup
3034  *
3035  * Returns 0 on success, negative on failure
3036  **/
3037 int ixgbevf_setup_rx_resources(struct ixgbevf_ring *rx_ring)
3038 {
3039         int size;
3040
3041         size = sizeof(struct ixgbevf_rx_buffer) * rx_ring->count;
3042         rx_ring->rx_buffer_info = vzalloc(size);
3043         if (!rx_ring->rx_buffer_info)
3044                 goto err;
3045
3046         u64_stats_init(&rx_ring->syncp);
3047
3048         /* Round up to nearest 4K */
3049         rx_ring->size = rx_ring->count * sizeof(union ixgbe_adv_rx_desc);
3050         rx_ring->size = ALIGN(rx_ring->size, 4096);
3051
3052         rx_ring->desc = dma_alloc_coherent(rx_ring->dev, rx_ring->size,
3053                                            &rx_ring->dma, GFP_KERNEL);
3054
3055         if (!rx_ring->desc)
3056                 goto err;
3057
3058         return 0;
3059 err:
3060         vfree(rx_ring->rx_buffer_info);
3061         rx_ring->rx_buffer_info = NULL;
3062         dev_err(rx_ring->dev, "Unable to allocate memory for the Rx descriptor ring\n");
3063         return -ENOMEM;
3064 }
3065
3066 /**
3067  * ixgbevf_setup_all_rx_resources - allocate all queues Rx resources
3068  * @adapter: board private structure
3069  *
3070  * If this function returns with an error, then it's possible one or
3071  * more of the rings is populated (while the rest are not).  It is the
3072  * callers duty to clean those orphaned rings.
3073  *
3074  * Return 0 on success, negative on failure
3075  **/
3076 static int ixgbevf_setup_all_rx_resources(struct ixgbevf_adapter *adapter)
3077 {
3078         int i, err = 0;
3079
3080         for (i = 0; i < adapter->num_rx_queues; i++) {
3081                 err = ixgbevf_setup_rx_resources(adapter->rx_ring[i]);
3082                 if (!err)
3083                         continue;
3084                 hw_dbg(&adapter->hw, "Allocation for Rx Queue %u failed\n", i);
3085                 break;
3086         }
3087         return err;
3088 }
3089
3090 /**
3091  * ixgbevf_free_rx_resources - Free Rx Resources
3092  * @rx_ring: ring to clean the resources from
3093  *
3094  * Free all receive software resources
3095  **/
3096 void ixgbevf_free_rx_resources(struct ixgbevf_ring *rx_ring)
3097 {
3098         ixgbevf_clean_rx_ring(rx_ring);
3099
3100         vfree(rx_ring->rx_buffer_info);
3101         rx_ring->rx_buffer_info = NULL;
3102
3103         dma_free_coherent(rx_ring->dev, rx_ring->size, rx_ring->desc,
3104                           rx_ring->dma);
3105
3106         rx_ring->desc = NULL;
3107 }
3108
3109 /**
3110  * ixgbevf_free_all_rx_resources - Free Rx Resources for All Queues
3111  * @adapter: board private structure
3112  *
3113  * Free all receive software resources
3114  **/
3115 static void ixgbevf_free_all_rx_resources(struct ixgbevf_adapter *adapter)
3116 {
3117         int i;
3118
3119         for (i = 0; i < adapter->num_rx_queues; i++)
3120                 if (adapter->rx_ring[i]->desc)
3121                         ixgbevf_free_rx_resources(adapter->rx_ring[i]);
3122 }
3123
3124 /**
3125  * ixgbevf_open - Called when a network interface is made active
3126  * @netdev: network interface device structure
3127  *
3128  * Returns 0 on success, negative value on failure
3129  *
3130  * The open entry point is called when a network interface is made
3131  * active by the system (IFF_UP).  At this point all resources needed
3132  * for transmit and receive operations are allocated, the interrupt
3133  * handler is registered with the OS, the watchdog timer is started,
3134  * and the stack is notified that the interface is ready.
3135  **/
3136 int ixgbevf_open(struct net_device *netdev)
3137 {
3138         struct ixgbevf_adapter *adapter = netdev_priv(netdev);
3139         struct ixgbe_hw *hw = &adapter->hw;
3140         int err;
3141
3142         /* A previous failure to open the device because of a lack of
3143          * available MSIX vector resources may have reset the number
3144          * of msix vectors variable to zero.  The only way to recover
3145          * is to unload/reload the driver and hope that the system has
3146          * been able to recover some MSIX vector resources.
3147          */
3148         if (!adapter->num_msix_vectors)
3149                 return -ENOMEM;
3150
3151         if (hw->adapter_stopped) {
3152                 ixgbevf_reset(adapter);
3153                 /* if adapter is still stopped then PF isn't up and
3154                  * the VF can't start.
3155                  */
3156                 if (hw->adapter_stopped) {
3157                         err = IXGBE_ERR_MBX;
3158                         pr_err("Unable to start - perhaps the PF Driver isn't up yet\n");
3159                         goto err_setup_reset;
3160                 }
3161         }
3162
3163         /* disallow open during test */
3164         if (test_bit(__IXGBEVF_TESTING, &adapter->state))
3165                 return -EBUSY;
3166
3167         netif_carrier_off(netdev);
3168
3169         /* allocate transmit descriptors */
3170         err = ixgbevf_setup_all_tx_resources(adapter);
3171         if (err)
3172                 goto err_setup_tx;
3173
3174         /* allocate receive descriptors */
3175         err = ixgbevf_setup_all_rx_resources(adapter);
3176         if (err)
3177                 goto err_setup_rx;
3178
3179         ixgbevf_configure(adapter);
3180
3181         /* Map the Tx/Rx rings to the vectors we were allotted.
3182          * if request_irq will be called in this function map_rings
3183          * must be called *before* up_complete
3184          */
3185         ixgbevf_map_rings_to_vectors(adapter);
3186
3187         err = ixgbevf_request_irq(adapter);
3188         if (err)
3189                 goto err_req_irq;
3190
3191         ixgbevf_up_complete(adapter);
3192
3193         return 0;
3194
3195 err_req_irq:
3196         ixgbevf_down(adapter);
3197 err_setup_rx:
3198         ixgbevf_free_all_rx_resources(adapter);
3199 err_setup_tx:
3200         ixgbevf_free_all_tx_resources(adapter);
3201         ixgbevf_reset(adapter);
3202
3203 err_setup_reset:
3204
3205         return err;
3206 }
3207
3208 /**
3209  * ixgbevf_close_suspend - actions necessary to both suspend and close flows
3210  * @adapter: the private adapter struct
3211  *
3212  * This function should contain the necessary work common to both suspending
3213  * and closing of the device.
3214  */
3215 static void ixgbevf_close_suspend(struct ixgbevf_adapter *adapter)
3216 {
3217         ixgbevf_down(adapter);
3218         ixgbevf_free_irq(adapter);
3219         ixgbevf_free_all_tx_resources(adapter);
3220         ixgbevf_free_all_rx_resources(adapter);
3221 }
3222
3223 /**
3224  * ixgbevf_close - Disables a network interface
3225  * @netdev: network interface device structure
3226  *
3227  * Returns 0, this is not allowed to fail
3228  *
3229  * The close entry point is called when an interface is de-activated
3230  * by the OS.  The hardware is still under the drivers control, but
3231  * needs to be disabled.  A global MAC reset is issued to stop the
3232  * hardware, and all transmit and receive resources are freed.
3233  **/
3234 int ixgbevf_close(struct net_device *netdev)
3235 {
3236         struct ixgbevf_adapter *adapter = netdev_priv(netdev);
3237
3238         if (netif_device_present(netdev))
3239                 ixgbevf_close_suspend(adapter);
3240
3241         return 0;
3242 }
3243
3244 static void ixgbevf_queue_reset_subtask(struct ixgbevf_adapter *adapter)
3245 {
3246         struct net_device *dev = adapter->netdev;
3247
3248         if (!test_and_clear_bit(__IXGBEVF_QUEUE_RESET_REQUESTED,
3249                                 &adapter->state))
3250                 return;
3251
3252         /* if interface is down do nothing */
3253         if (test_bit(__IXGBEVF_DOWN, &adapter->state) ||
3254             test_bit(__IXGBEVF_RESETTING, &adapter->state))
3255                 return;
3256
3257         /* Hardware has to reinitialize queues and interrupts to
3258          * match packet buffer alignment. Unfortunately, the
3259          * hardware is not flexible enough to do this dynamically.
3260          */
3261         rtnl_lock();
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         rtnl_unlock();
3273 }
3274
3275 static void ixgbevf_tx_ctxtdesc(struct ixgbevf_ring *tx_ring,
3276                                 u32 vlan_macip_lens, u32 type_tucmd,
3277                                 u32 mss_l4len_idx)
3278 {
3279         struct ixgbe_adv_tx_context_desc *context_desc;
3280         u16 i = tx_ring->next_to_use;
3281
3282         context_desc = IXGBEVF_TX_CTXTDESC(tx_ring, i);
3283
3284         i++;
3285         tx_ring->next_to_use = (i < tx_ring->count) ? i : 0;
3286
3287         /* set bits to identify this as an advanced context descriptor */
3288         type_tucmd |= IXGBE_TXD_CMD_DEXT | IXGBE_ADVTXD_DTYP_CTXT;
3289
3290         context_desc->vlan_macip_lens   = cpu_to_le32(vlan_macip_lens);
3291         context_desc->seqnum_seed       = 0;
3292         context_desc->type_tucmd_mlhl   = cpu_to_le32(type_tucmd);
3293         context_desc->mss_l4len_idx     = cpu_to_le32(mss_l4len_idx);
3294 }
3295
3296 static int ixgbevf_tso(struct ixgbevf_ring *tx_ring,
3297                        struct ixgbevf_tx_buffer *first,
3298                        u8 *hdr_len)
3299 {
3300         u32 vlan_macip_lens, type_tucmd, mss_l4len_idx;
3301         struct sk_buff *skb = first->skb;
3302         union {
3303                 struct iphdr *v4;
3304                 struct ipv6hdr *v6;
3305                 unsigned char *hdr;
3306         } ip;
3307         union {
3308                 struct tcphdr *tcp;
3309                 unsigned char *hdr;
3310         } l4;
3311         u32 paylen, l4_offset;
3312         int err;
3313
3314         if (skb->ip_summed != CHECKSUM_PARTIAL)
3315                 return 0;
3316
3317         if (!skb_is_gso(skb))
3318                 return 0;
3319
3320         err = skb_cow_head(skb, 0);
3321         if (err < 0)
3322                 return err;
3323
3324         if (eth_p_mpls(first->protocol))
3325                 ip.hdr = skb_inner_network_header(skb);
3326         else
3327                 ip.hdr = skb_network_header(skb);
3328         l4.hdr = skb_checksum_start(skb);
3329
3330         /* ADV DTYP TUCMD MKRLOC/ISCSIHEDLEN */
3331         type_tucmd = IXGBE_ADVTXD_TUCMD_L4T_TCP;
3332
3333         /* initialize outer IP header fields */
3334         if (ip.v4->version == 4) {
3335                 unsigned char *csum_start = skb_checksum_start(skb);
3336                 unsigned char *trans_start = ip.hdr + (ip.v4->ihl * 4);
3337
3338                 /* IP header will have to cancel out any data that
3339                  * is not a part of the outer IP header
3340                  */
3341                 ip.v4->check = csum_fold(csum_partial(trans_start,
3342                                                       csum_start - trans_start,
3343                                                       0));
3344                 type_tucmd |= IXGBE_ADVTXD_TUCMD_IPV4;
3345
3346                 ip.v4->tot_len = 0;
3347                 first->tx_flags |= IXGBE_TX_FLAGS_TSO |
3348                                    IXGBE_TX_FLAGS_CSUM |
3349                                    IXGBE_TX_FLAGS_IPV4;
3350         } else {
3351                 ip.v6->payload_len = 0;
3352                 first->tx_flags |= IXGBE_TX_FLAGS_TSO |
3353                                    IXGBE_TX_FLAGS_CSUM;
3354         }
3355
3356         /* determine offset of inner transport header */
3357         l4_offset = l4.hdr - skb->data;
3358
3359         /* compute length of segmentation header */
3360         *hdr_len = (l4.tcp->doff * 4) + l4_offset;
3361
3362         /* remove payload length from inner checksum */
3363         paylen = skb->len - l4_offset;
3364         csum_replace_by_diff(&l4.tcp->check, htonl(paylen));
3365
3366         /* update gso size and bytecount with header size */
3367         first->gso_segs = skb_shinfo(skb)->gso_segs;
3368         first->bytecount += (first->gso_segs - 1) * *hdr_len;
3369
3370         /* mss_l4len_id: use 1 as index for TSO */
3371         mss_l4len_idx = (*hdr_len - l4_offset) << IXGBE_ADVTXD_L4LEN_SHIFT;
3372         mss_l4len_idx |= skb_shinfo(skb)->gso_size << IXGBE_ADVTXD_MSS_SHIFT;
3373         mss_l4len_idx |= (1u << IXGBE_ADVTXD_IDX_SHIFT);
3374
3375         /* vlan_macip_lens: HEADLEN, MACLEN, VLAN tag */
3376         vlan_macip_lens = l4.hdr - ip.hdr;
3377         vlan_macip_lens |= (ip.hdr - skb->data) << IXGBE_ADVTXD_MACLEN_SHIFT;
3378         vlan_macip_lens |= first->tx_flags & IXGBE_TX_FLAGS_VLAN_MASK;
3379
3380         ixgbevf_tx_ctxtdesc(tx_ring, vlan_macip_lens,
3381                             type_tucmd, mss_l4len_idx);
3382
3383         return 1;
3384 }
3385
3386 static inline bool ixgbevf_ipv6_csum_is_sctp(struct sk_buff *skb)
3387 {
3388         unsigned int offset = 0;
3389
3390         ipv6_find_hdr(skb, &offset, IPPROTO_SCTP, NULL, NULL);
3391
3392         return offset == skb_checksum_start_offset(skb);
3393 }
3394
3395 static void ixgbevf_tx_csum(struct ixgbevf_ring *tx_ring,
3396                             struct ixgbevf_tx_buffer *first)
3397 {
3398         struct sk_buff *skb = first->skb;
3399         u32 vlan_macip_lens = 0;
3400         u32 type_tucmd = 0;
3401
3402         if (skb->ip_summed != CHECKSUM_PARTIAL)
3403                 goto no_csum;
3404
3405         switch (skb->csum_offset) {
3406         case offsetof(struct tcphdr, check):
3407                 type_tucmd = IXGBE_ADVTXD_TUCMD_L4T_TCP;
3408                 /* fall through */
3409         case offsetof(struct udphdr, check):
3410                 break;
3411         case offsetof(struct sctphdr, checksum):
3412                 /* validate that this is actually an SCTP request */
3413                 if (((first->protocol == htons(ETH_P_IP)) &&
3414                      (ip_hdr(skb)->protocol == IPPROTO_SCTP)) ||
3415                     ((first->protocol == htons(ETH_P_IPV6)) &&
3416                      ixgbevf_ipv6_csum_is_sctp(skb))) {
3417                         type_tucmd = IXGBE_ADVTXD_TUCMD_L4T_SCTP;
3418                         break;
3419                 }
3420                 /* fall through */
3421         default:
3422                 skb_checksum_help(skb);
3423                 goto no_csum;
3424         }
3425
3426         if (first->protocol == htons(ETH_P_IP))
3427                 type_tucmd |= IXGBE_ADVTXD_TUCMD_IPV4;
3428
3429         /* update TX checksum flag */
3430         first->tx_flags |= IXGBE_TX_FLAGS_CSUM;
3431         vlan_macip_lens = skb_checksum_start_offset(skb) -
3432                           skb_network_offset(skb);
3433 no_csum:
3434         /* vlan_macip_lens: MACLEN, VLAN tag */
3435         vlan_macip_lens |= skb_network_offset(skb) << IXGBE_ADVTXD_MACLEN_SHIFT;
3436         vlan_macip_lens |= first->tx_flags & IXGBE_TX_FLAGS_VLAN_MASK;
3437
3438         ixgbevf_tx_ctxtdesc(tx_ring, vlan_macip_lens, type_tucmd, 0);
3439 }
3440
3441 static __le32 ixgbevf_tx_cmd_type(u32 tx_flags)
3442 {
3443         /* set type for advanced descriptor with frame checksum insertion */
3444         __le32 cmd_type = cpu_to_le32(IXGBE_ADVTXD_DTYP_DATA |
3445                                       IXGBE_ADVTXD_DCMD_IFCS |
3446                                       IXGBE_ADVTXD_DCMD_DEXT);
3447
3448         /* set HW VLAN bit if VLAN is present */
3449         if (tx_flags & IXGBE_TX_FLAGS_VLAN)
3450                 cmd_type |= cpu_to_le32(IXGBE_ADVTXD_DCMD_VLE);
3451
3452         /* set segmentation enable bits for TSO/FSO */
3453         if (tx_flags & IXGBE_TX_FLAGS_TSO)
3454                 cmd_type |= cpu_to_le32(IXGBE_ADVTXD_DCMD_TSE);
3455
3456         return cmd_type;
3457 }
3458
3459 static void ixgbevf_tx_olinfo_status(union ixgbe_adv_tx_desc *tx_desc,
3460                                      u32 tx_flags, unsigned int paylen)
3461 {
3462         __le32 olinfo_status = cpu_to_le32(paylen << IXGBE_ADVTXD_PAYLEN_SHIFT);
3463
3464         /* enable L4 checksum for TSO and TX checksum offload */
3465         if (tx_flags & IXGBE_TX_FLAGS_CSUM)
3466                 olinfo_status |= cpu_to_le32(IXGBE_ADVTXD_POPTS_TXSM);
3467
3468         /* enble IPv4 checksum for TSO */
3469         if (tx_flags & IXGBE_TX_FLAGS_IPV4)
3470                 olinfo_status |= cpu_to_le32(IXGBE_ADVTXD_POPTS_IXSM);
3471
3472         /* use index 1 context for TSO/FSO/FCOE */
3473         if (tx_flags & IXGBE_TX_FLAGS_TSO)
3474                 olinfo_status |= cpu_to_le32(1u << IXGBE_ADVTXD_IDX_SHIFT);
3475
3476         /* Check Context must be set if Tx switch is enabled, which it
3477          * always is for case where virtual functions are running
3478          */
3479         olinfo_status |= cpu_to_le32(IXGBE_ADVTXD_CC);
3480
3481         tx_desc->read.olinfo_status = olinfo_status;
3482 }
3483
3484 static void ixgbevf_tx_map(struct ixgbevf_ring *tx_ring,
3485                            struct ixgbevf_tx_buffer *first,
3486                            const u8 hdr_len)
3487 {
3488         dma_addr_t dma;
3489         struct sk_buff *skb = first->skb;
3490         struct ixgbevf_tx_buffer *tx_buffer;
3491         union ixgbe_adv_tx_desc *tx_desc;
3492         struct skb_frag_struct *frag = &skb_shinfo(skb)->frags[0];
3493         unsigned int data_len = skb->data_len;
3494         unsigned int size = skb_headlen(skb);
3495         unsigned int paylen = skb->len - hdr_len;
3496         u32 tx_flags = first->tx_flags;
3497         __le32 cmd_type;
3498         u16 i = tx_ring->next_to_use;
3499
3500         tx_desc = IXGBEVF_TX_DESC(tx_ring, i);
3501
3502         ixgbevf_tx_olinfo_status(tx_desc, tx_flags, paylen);
3503         cmd_type = ixgbevf_tx_cmd_type(tx_flags);
3504
3505         dma = dma_map_single(tx_ring->dev, skb->data, size, DMA_TO_DEVICE);
3506         if (dma_mapping_error(tx_ring->dev, dma))
3507                 goto dma_error;
3508
3509         /* record length, and DMA address */
3510         dma_unmap_len_set(first, len, size);
3511         dma_unmap_addr_set(first, dma, dma);
3512
3513         tx_desc->read.buffer_addr = cpu_to_le64(dma);
3514
3515         for (;;) {
3516                 while (unlikely(size > IXGBE_MAX_DATA_PER_TXD)) {
3517                         tx_desc->read.cmd_type_len =
3518                                 cmd_type | cpu_to_le32(IXGBE_MAX_DATA_PER_TXD);
3519
3520                         i++;
3521                         tx_desc++;
3522                         if (i == tx_ring->count) {
3523                                 tx_desc = IXGBEVF_TX_DESC(tx_ring, 0);
3524                                 i = 0;
3525                         }
3526
3527                         dma += IXGBE_MAX_DATA_PER_TXD;
3528                         size -= IXGBE_MAX_DATA_PER_TXD;
3529
3530                         tx_desc->read.buffer_addr = cpu_to_le64(dma);
3531                         tx_desc->read.olinfo_status = 0;
3532                 }
3533
3534                 if (likely(!data_len))
3535                         break;
3536
3537                 tx_desc->read.cmd_type_len = cmd_type | cpu_to_le32(size);
3538
3539                 i++;
3540                 tx_desc++;
3541                 if (i == tx_ring->count) {
3542                         tx_desc = IXGBEVF_TX_DESC(tx_ring, 0);
3543                         i = 0;
3544                 }
3545
3546                 size = skb_frag_size(frag);
3547                 data_len -= size;
3548
3549                 dma = skb_frag_dma_map(tx_ring->dev, frag, 0, size,
3550                                        DMA_TO_DEVICE);
3551                 if (dma_mapping_error(tx_ring->dev, dma))
3552                         goto dma_error;
3553
3554                 tx_buffer = &tx_ring->tx_buffer_info[i];
3555                 dma_unmap_len_set(tx_buffer, len, size);
3556                 dma_unmap_addr_set(tx_buffer, dma, dma);
3557
3558                 tx_desc->read.buffer_addr = cpu_to_le64(dma);
3559                 tx_desc->read.olinfo_status = 0;
3560
3561                 frag++;
3562         }
3563
3564         /* write last descriptor with RS and EOP bits */
3565         cmd_type |= cpu_to_le32(size) | cpu_to_le32(IXGBE_TXD_CMD);
3566         tx_desc->read.cmd_type_len = cmd_type;
3567
3568         /* set the timestamp */
3569         first->time_stamp = jiffies;
3570
3571         /* Force memory writes to complete before letting h/w know there
3572          * are new descriptors to fetch.  (Only applicable for weak-ordered
3573          * memory model archs, such as IA-64).
3574          *
3575          * We also need this memory barrier (wmb) to make certain all of the
3576          * status bits have been updated before next_to_watch is written.
3577          */
3578         wmb();
3579
3580         /* set next_to_watch value indicating a packet is present */
3581         first->next_to_watch = tx_desc;
3582
3583         i++;
3584         if (i == tx_ring->count)
3585                 i = 0;
3586
3587         tx_ring->next_to_use = i;
3588
3589         /* notify HW of packet */
3590         ixgbevf_write_tail(tx_ring, i);
3591
3592         return;
3593 dma_error:
3594         dev_err(tx_ring->dev, "TX DMA map failed\n");
3595
3596         /* clear dma mappings for failed tx_buffer_info map */
3597         for (;;) {
3598                 tx_buffer = &tx_ring->tx_buffer_info[i];
3599                 ixgbevf_unmap_and_free_tx_resource(tx_ring, tx_buffer);
3600                 if (tx_buffer == first)
3601                         break;
3602                 if (i == 0)
3603                         i = tx_ring->count;
3604                 i--;
3605         }
3606
3607         tx_ring->next_to_use = i;
3608 }
3609
3610 static int __ixgbevf_maybe_stop_tx(struct ixgbevf_ring *tx_ring, int size)
3611 {
3612         netif_stop_subqueue(tx_ring->netdev, tx_ring->queue_index);
3613         /* Herbert's original patch had:
3614          *  smp_mb__after_netif_stop_queue();
3615          * but since that doesn't exist yet, just open code it.
3616          */
3617         smp_mb();
3618
3619         /* We need to check again in a case another CPU has just
3620          * made room available.
3621          */
3622         if (likely(ixgbevf_desc_unused(tx_ring) < size))
3623                 return -EBUSY;
3624
3625         /* A reprieve! - use start_queue because it doesn't call schedule */
3626         netif_start_subqueue(tx_ring->netdev, tx_ring->queue_index);
3627         ++tx_ring->tx_stats.restart_queue;
3628
3629         return 0;
3630 }
3631
3632 static int ixgbevf_maybe_stop_tx(struct ixgbevf_ring *tx_ring, int size)
3633 {
3634         if (likely(ixgbevf_desc_unused(tx_ring) >= size))
3635                 return 0;
3636         return __ixgbevf_maybe_stop_tx(tx_ring, size);
3637 }
3638
3639 static int ixgbevf_xmit_frame(struct sk_buff *skb, struct net_device *netdev)
3640 {
3641         struct ixgbevf_adapter *adapter = netdev_priv(netdev);
3642         struct ixgbevf_tx_buffer *first;
3643         struct ixgbevf_ring *tx_ring;
3644         int tso;
3645         u32 tx_flags = 0;
3646         u16 count = TXD_USE_COUNT(skb_headlen(skb));
3647 #if PAGE_SIZE > IXGBE_MAX_DATA_PER_TXD
3648         unsigned short f;
3649 #endif
3650         u8 hdr_len = 0;
3651         u8 *dst_mac = skb_header_pointer(skb, 0, 0, NULL);
3652
3653         if (!dst_mac || is_link_local_ether_addr(dst_mac)) {
3654                 dev_kfree_skb_any(skb);
3655                 return NETDEV_TX_OK;
3656         }
3657
3658         tx_ring = adapter->tx_ring[skb->queue_mapping];
3659
3660         /* need: 1 descriptor per page * PAGE_SIZE/IXGBE_MAX_DATA_PER_TXD,
3661          *       + 1 desc for skb_headlen/IXGBE_MAX_DATA_PER_TXD,
3662          *       + 2 desc gap to keep tail from touching head,
3663          *       + 1 desc for context descriptor,
3664          * otherwise try next time
3665          */
3666 #if PAGE_SIZE > IXGBE_MAX_DATA_PER_TXD
3667         for (f = 0; f < skb_shinfo(skb)->nr_frags; f++)
3668                 count += TXD_USE_COUNT(skb_shinfo(skb)->frags[f].size);
3669 #else
3670         count += skb_shinfo(skb)->nr_frags;
3671 #endif
3672         if (ixgbevf_maybe_stop_tx(tx_ring, count + 3)) {
3673                 tx_ring->tx_stats.tx_busy++;
3674                 return NETDEV_TX_BUSY;
3675         }
3676
3677         /* record the location of the first descriptor for this packet */
3678         first = &tx_ring->tx_buffer_info[tx_ring->next_to_use];
3679         first->skb = skb;
3680         first->bytecount = skb->len;
3681         first->gso_segs = 1;
3682
3683         if (skb_vlan_tag_present(skb)) {
3684                 tx_flags |= skb_vlan_tag_get(skb);
3685                 tx_flags <<= IXGBE_TX_FLAGS_VLAN_SHIFT;
3686                 tx_flags |= IXGBE_TX_FLAGS_VLAN;
3687         }
3688
3689         /* record initial flags and protocol */
3690         first->tx_flags = tx_flags;
3691         first->protocol = vlan_get_protocol(skb);
3692
3693         tso = ixgbevf_tso(tx_ring, first, &hdr_len);
3694         if (tso < 0)
3695                 goto out_drop;
3696         else if (!tso)
3697                 ixgbevf_tx_csum(tx_ring, first);
3698
3699         ixgbevf_tx_map(tx_ring, first, hdr_len);
3700
3701         ixgbevf_maybe_stop_tx(tx_ring, DESC_NEEDED);
3702
3703         return NETDEV_TX_OK;
3704
3705 out_drop:
3706         dev_kfree_skb_any(first->skb);
3707         first->skb = NULL;
3708
3709         return NETDEV_TX_OK;
3710 }
3711
3712 /**
3713  * ixgbevf_set_mac - Change the Ethernet Address of the NIC
3714  * @netdev: network interface device structure
3715  * @p: pointer to an address structure
3716  *
3717  * Returns 0 on success, negative on failure
3718  **/
3719 static int ixgbevf_set_mac(struct net_device *netdev, void *p)
3720 {
3721         struct ixgbevf_adapter *adapter = netdev_priv(netdev);
3722         struct ixgbe_hw *hw = &adapter->hw;
3723         struct sockaddr *addr = p;
3724         int err;
3725
3726         if (!is_valid_ether_addr(addr->sa_data))
3727                 return -EADDRNOTAVAIL;
3728
3729         spin_lock_bh(&adapter->mbx_lock);
3730
3731         err = hw->mac.ops.set_rar(hw, 0, addr->sa_data, 0);
3732
3733         spin_unlock_bh(&adapter->mbx_lock);
3734
3735         if (err)
3736                 return -EPERM;
3737
3738         ether_addr_copy(hw->mac.addr, addr->sa_data);
3739         ether_addr_copy(hw->mac.perm_addr, addr->sa_data);
3740         ether_addr_copy(netdev->dev_addr, addr->sa_data);
3741
3742         return 0;
3743 }
3744
3745 /**
3746  * ixgbevf_change_mtu - Change the Maximum Transfer Unit
3747  * @netdev: network interface device structure
3748  * @new_mtu: new value for maximum frame size
3749  *
3750  * Returns 0 on success, negative on failure
3751  **/
3752 static int ixgbevf_change_mtu(struct net_device *netdev, int new_mtu)
3753 {
3754         struct ixgbevf_adapter *adapter = netdev_priv(netdev);
3755         struct ixgbe_hw *hw = &adapter->hw;
3756         int max_frame = new_mtu + ETH_HLEN + ETH_FCS_LEN;
3757         int ret;
3758
3759         spin_lock_bh(&adapter->mbx_lock);
3760         /* notify the PF of our intent to use this size of frame */
3761         ret = hw->mac.ops.set_rlpml(hw, max_frame);
3762         spin_unlock_bh(&adapter->mbx_lock);
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         rtnl_lock();
3802         netif_device_detach(netdev);
3803
3804         if (netif_running(netdev))
3805                 ixgbevf_close_suspend(adapter);
3806
3807         ixgbevf_clear_interrupt_scheme(adapter);
3808         rtnl_unlock();
3809
3810 #ifdef CONFIG_PM
3811         retval = pci_save_state(pdev);
3812         if (retval)
3813                 return retval;
3814
3815 #endif
3816         if (!test_and_set_bit(__IXGBEVF_DISABLED, &adapter->state))
3817                 pci_disable_device(pdev);
3818
3819         return 0;
3820 }
3821
3822 #ifdef CONFIG_PM
3823 static int ixgbevf_resume(struct pci_dev *pdev)
3824 {
3825         struct net_device *netdev = pci_get_drvdata(pdev);
3826         struct ixgbevf_adapter *adapter = netdev_priv(netdev);
3827         u32 err;
3828
3829         pci_restore_state(pdev);
3830         /* pci_restore_state clears dev->state_saved so call
3831          * pci_save_state to restore it.
3832          */
3833         pci_save_state(pdev);
3834
3835         err = pci_enable_device_mem(pdev);
3836         if (err) {
3837                 dev_err(&pdev->dev, "Cannot enable PCI device from suspend\n");
3838                 return err;
3839         }
3840
3841         adapter->hw.hw_addr = adapter->io_addr;
3842         smp_mb__before_atomic();
3843         clear_bit(__IXGBEVF_DISABLED, &adapter->state);
3844         pci_set_master(pdev);
3845
3846         ixgbevf_reset(adapter);
3847
3848         rtnl_lock();
3849         err = ixgbevf_init_interrupt_scheme(adapter);
3850         rtnl_unlock();
3851         if (err) {
3852                 dev_err(&pdev->dev, "Cannot initialize interrupts\n");
3853                 return err;
3854         }
3855
3856         if (netif_running(netdev)) {
3857                 err = ixgbevf_open(netdev);
3858                 if (err)
3859                         return err;
3860         }
3861
3862         netif_device_attach(netdev);
3863
3864         return err;
3865 }
3866
3867 #endif /* CONFIG_PM */
3868 static void ixgbevf_shutdown(struct pci_dev *pdev)
3869 {
3870         ixgbevf_suspend(pdev, PMSG_SUSPEND);
3871 }
3872
3873 static void ixgbevf_get_stats(struct net_device *netdev,
3874                               struct rtnl_link_stats64 *stats)
3875 {
3876         struct ixgbevf_adapter *adapter = netdev_priv(netdev);
3877         unsigned int start;
3878         u64 bytes, packets;
3879         const struct ixgbevf_ring *ring;
3880         int i;
3881
3882         ixgbevf_update_stats(adapter);
3883
3884         stats->multicast = adapter->stats.vfmprc - adapter->stats.base_vfmprc;
3885
3886         for (i = 0; i < adapter->num_rx_queues; i++) {
3887                 ring = adapter->rx_ring[i];
3888                 do {
3889                         start = u64_stats_fetch_begin_irq(&ring->syncp);
3890                         bytes = ring->stats.bytes;
3891                         packets = ring->stats.packets;
3892                 } while (u64_stats_fetch_retry_irq(&ring->syncp, start));
3893                 stats->rx_bytes += bytes;
3894                 stats->rx_packets += packets;
3895         }
3896
3897         for (i = 0; i < adapter->num_tx_queues; i++) {
3898                 ring = adapter->tx_ring[i];
3899                 do {
3900                         start = u64_stats_fetch_begin_irq(&ring->syncp);
3901                         bytes = ring->stats.bytes;
3902                         packets = ring->stats.packets;
3903                 } while (u64_stats_fetch_retry_irq(&ring->syncp, start));
3904                 stats->tx_bytes += bytes;
3905                 stats->tx_packets += packets;
3906         }
3907 }
3908
3909 #define IXGBEVF_MAX_MAC_HDR_LEN         127
3910 #define IXGBEVF_MAX_NETWORK_HDR_LEN     511
3911
3912 static netdev_features_t
3913 ixgbevf_features_check(struct sk_buff *skb, struct net_device *dev,
3914                        netdev_features_t features)
3915 {
3916         unsigned int network_hdr_len, mac_hdr_len;
3917
3918         /* Make certain the headers can be described by a context descriptor */
3919         mac_hdr_len = skb_network_header(skb) - skb->data;
3920         if (unlikely(mac_hdr_len > IXGBEVF_MAX_MAC_HDR_LEN))
3921                 return features & ~(NETIF_F_HW_CSUM |
3922                                     NETIF_F_SCTP_CRC |
3923                                     NETIF_F_HW_VLAN_CTAG_TX |
3924                                     NETIF_F_TSO |
3925                                     NETIF_F_TSO6);
3926
3927         network_hdr_len = skb_checksum_start(skb) - skb_network_header(skb);
3928         if (unlikely(network_hdr_len >  IXGBEVF_MAX_NETWORK_HDR_LEN))
3929                 return features & ~(NETIF_F_HW_CSUM |
3930                                     NETIF_F_SCTP_CRC |
3931                                     NETIF_F_TSO |
3932                                     NETIF_F_TSO6);
3933
3934         /* We can only support IPV4 TSO in tunnels if we can mangle the
3935          * inner IP ID field, so strip TSO if MANGLEID is not supported.
3936          */
3937         if (skb->encapsulation && !(features & NETIF_F_TSO_MANGLEID))
3938                 features &= ~NETIF_F_TSO;
3939
3940         return features;
3941 }
3942
3943 static const struct net_device_ops ixgbevf_netdev_ops = {
3944         .ndo_open               = ixgbevf_open,
3945         .ndo_stop               = ixgbevf_close,
3946         .ndo_start_xmit         = ixgbevf_xmit_frame,
3947         .ndo_set_rx_mode        = ixgbevf_set_rx_mode,
3948         .ndo_get_stats64        = ixgbevf_get_stats,
3949         .ndo_validate_addr      = eth_validate_addr,
3950         .ndo_set_mac_address    = ixgbevf_set_mac,
3951         .ndo_change_mtu         = ixgbevf_change_mtu,
3952         .ndo_tx_timeout         = ixgbevf_tx_timeout,
3953         .ndo_vlan_rx_add_vid    = ixgbevf_vlan_rx_add_vid,
3954         .ndo_vlan_rx_kill_vid   = ixgbevf_vlan_rx_kill_vid,
3955 #ifdef CONFIG_NET_POLL_CONTROLLER
3956         .ndo_poll_controller    = ixgbevf_netpoll,
3957 #endif
3958         .ndo_features_check     = ixgbevf_features_check,
3959 };
3960
3961 static void ixgbevf_assign_netdev_ops(struct net_device *dev)
3962 {
3963         dev->netdev_ops = &ixgbevf_netdev_ops;
3964         ixgbevf_set_ethtool_ops(dev);
3965         dev->watchdog_timeo = 5 * HZ;
3966 }
3967
3968 /**
3969  * ixgbevf_probe - Device Initialization Routine
3970  * @pdev: PCI device information struct
3971  * @ent: entry in ixgbevf_pci_tbl
3972  *
3973  * Returns 0 on success, negative on failure
3974  *
3975  * ixgbevf_probe initializes an adapter identified by a pci_dev structure.
3976  * The OS initialization, configuring of the adapter private structure,
3977  * and a hardware reset occur.
3978  **/
3979 static int ixgbevf_probe(struct pci_dev *pdev, const struct pci_device_id *ent)
3980 {
3981         struct net_device *netdev;
3982         struct ixgbevf_adapter *adapter = NULL;
3983         struct ixgbe_hw *hw = NULL;
3984         const struct ixgbevf_info *ii = ixgbevf_info_tbl[ent->driver_data];
3985         int err, pci_using_dac;
3986         bool disable_dev = false;
3987
3988         err = pci_enable_device(pdev);
3989         if (err)
3990                 return err;
3991
3992         if (!dma_set_mask_and_coherent(&pdev->dev, DMA_BIT_MASK(64))) {
3993                 pci_using_dac = 1;
3994         } else {
3995                 err = dma_set_mask_and_coherent(&pdev->dev, DMA_BIT_MASK(32));
3996                 if (err) {
3997                         dev_err(&pdev->dev, "No usable DMA configuration, aborting\n");
3998                         goto err_dma;
3999                 }
4000                 pci_using_dac = 0;
4001         }
4002
4003         err = pci_request_regions(pdev, ixgbevf_driver_name);
4004         if (err) {
4005                 dev_err(&pdev->dev, "pci_request_regions failed 0x%x\n", err);
4006                 goto err_pci_reg;
4007         }
4008
4009         pci_set_master(pdev);
4010
4011         netdev = alloc_etherdev_mq(sizeof(struct ixgbevf_adapter),
4012                                    MAX_TX_QUEUES);
4013         if (!netdev) {
4014                 err = -ENOMEM;
4015                 goto err_alloc_etherdev;
4016         }
4017
4018         SET_NETDEV_DEV(netdev, &pdev->dev);
4019
4020         adapter = netdev_priv(netdev);
4021
4022         adapter->netdev = netdev;
4023         adapter->pdev = pdev;
4024         hw = &adapter->hw;
4025         hw->back = adapter;
4026         adapter->msg_enable = netif_msg_init(debug, DEFAULT_MSG_ENABLE);
4027
4028         /* call save state here in standalone driver because it relies on
4029          * adapter struct to exist, and needs to call netdev_priv
4030          */
4031         pci_save_state(pdev);
4032
4033         hw->hw_addr = ioremap(pci_resource_start(pdev, 0),
4034                               pci_resource_len(pdev, 0));
4035         adapter->io_addr = hw->hw_addr;
4036         if (!hw->hw_addr) {
4037                 err = -EIO;
4038                 goto err_ioremap;
4039         }
4040
4041         ixgbevf_assign_netdev_ops(netdev);
4042
4043         /* Setup HW API */
4044         memcpy(&hw->mac.ops, ii->mac_ops, sizeof(hw->mac.ops));
4045         hw->mac.type  = ii->mac;
4046
4047         memcpy(&hw->mbx.ops, &ixgbevf_mbx_ops,
4048                sizeof(struct ixgbe_mbx_operations));
4049
4050         /* setup the private structure */
4051         err = ixgbevf_sw_init(adapter);
4052         if (err)
4053                 goto err_sw_init;
4054
4055         /* The HW MAC address was set and/or determined in sw_init */
4056         if (!is_valid_ether_addr(netdev->dev_addr)) {
4057                 pr_err("invalid MAC address\n");
4058                 err = -EIO;
4059                 goto err_sw_init;
4060         }
4061
4062         netdev->hw_features = NETIF_F_SG |
4063                               NETIF_F_TSO |
4064                               NETIF_F_TSO6 |
4065                               NETIF_F_RXCSUM |
4066                               NETIF_F_HW_CSUM |
4067                               NETIF_F_SCTP_CRC;
4068
4069 #define IXGBEVF_GSO_PARTIAL_FEATURES (NETIF_F_GSO_GRE | \
4070                                       NETIF_F_GSO_GRE_CSUM | \
4071                                       NETIF_F_GSO_IPXIP4 | \
4072                                       NETIF_F_GSO_IPXIP6 | \
4073                                       NETIF_F_GSO_UDP_TUNNEL | \
4074                                       NETIF_F_GSO_UDP_TUNNEL_CSUM)
4075
4076         netdev->gso_partial_features = IXGBEVF_GSO_PARTIAL_FEATURES;
4077         netdev->hw_features |= NETIF_F_GSO_PARTIAL |
4078                                IXGBEVF_GSO_PARTIAL_FEATURES;
4079
4080         netdev->features = netdev->hw_features;
4081
4082         if (pci_using_dac)
4083                 netdev->features |= NETIF_F_HIGHDMA;
4084
4085         netdev->vlan_features |= netdev->features | NETIF_F_TSO_MANGLEID;
4086         netdev->mpls_features |= NETIF_F_SG |
4087                                  NETIF_F_TSO |
4088                                  NETIF_F_TSO6 |
4089                                  NETIF_F_HW_CSUM;
4090         netdev->mpls_features |= IXGBEVF_GSO_PARTIAL_FEATURES;
4091         netdev->hw_enc_features |= netdev->vlan_features;
4092
4093         /* set this bit last since it cannot be part of vlan_features */
4094         netdev->features |= NETIF_F_HW_VLAN_CTAG_FILTER |
4095                             NETIF_F_HW_VLAN_CTAG_RX |
4096                             NETIF_F_HW_VLAN_CTAG_TX;
4097
4098         netdev->priv_flags |= IFF_UNICAST_FLT;
4099
4100         /* MTU range: 68 - 1504 or 9710 */
4101         netdev->min_mtu = ETH_MIN_MTU;
4102         switch (adapter->hw.api_version) {
4103         case ixgbe_mbox_api_11:
4104         case ixgbe_mbox_api_12:
4105         case ixgbe_mbox_api_13:
4106                 netdev->max_mtu = IXGBE_MAX_JUMBO_FRAME_SIZE -
4107                                   (ETH_HLEN + ETH_FCS_LEN);
4108                 break;
4109         default:
4110                 if (adapter->hw.mac.type != ixgbe_mac_82599_vf)
4111                         netdev->max_mtu = IXGBE_MAX_JUMBO_FRAME_SIZE -
4112                                           (ETH_HLEN + ETH_FCS_LEN);
4113                 else
4114                         netdev->max_mtu = ETH_DATA_LEN + ETH_FCS_LEN;
4115                 break;
4116         }
4117
4118         if (IXGBE_REMOVED(hw->hw_addr)) {
4119                 err = -EIO;
4120                 goto err_sw_init;
4121         }
4122
4123         setup_timer(&adapter->service_timer, &ixgbevf_service_timer,
4124                     (unsigned long)adapter);
4125
4126         INIT_WORK(&adapter->service_task, ixgbevf_service_task);
4127         set_bit(__IXGBEVF_SERVICE_INITED, &adapter->state);
4128         clear_bit(__IXGBEVF_SERVICE_SCHED, &adapter->state);
4129
4130         err = ixgbevf_init_interrupt_scheme(adapter);
4131         if (err)
4132                 goto err_sw_init;
4133
4134         strcpy(netdev->name, "eth%d");
4135
4136         err = register_netdev(netdev);
4137         if (err)
4138                 goto err_register;
4139
4140         pci_set_drvdata(pdev, netdev);
4141         netif_carrier_off(netdev);
4142
4143         ixgbevf_init_last_counter_stats(adapter);
4144
4145         /* print the VF info */
4146         dev_info(&pdev->dev, "%pM\n", netdev->dev_addr);
4147         dev_info(&pdev->dev, "MAC: %d\n", hw->mac.type);
4148
4149         switch (hw->mac.type) {
4150         case ixgbe_mac_X550_vf:
4151                 dev_info(&pdev->dev, "Intel(R) X550 Virtual Function\n");
4152                 break;
4153         case ixgbe_mac_X540_vf:
4154                 dev_info(&pdev->dev, "Intel(R) X540 Virtual Function\n");
4155                 break;
4156         case ixgbe_mac_82599_vf:
4157         default:
4158                 dev_info(&pdev->dev, "Intel(R) 82599 Virtual Function\n");
4159                 break;
4160         }
4161
4162         return 0;
4163
4164 err_register:
4165         ixgbevf_clear_interrupt_scheme(adapter);
4166 err_sw_init:
4167         ixgbevf_reset_interrupt_capability(adapter);
4168         iounmap(adapter->io_addr);
4169         kfree(adapter->rss_key);
4170 err_ioremap:
4171         disable_dev = !test_and_set_bit(__IXGBEVF_DISABLED, &adapter->state);
4172         free_netdev(netdev);
4173 err_alloc_etherdev:
4174         pci_release_regions(pdev);
4175 err_pci_reg:
4176 err_dma:
4177         if (!adapter || disable_dev)
4178                 pci_disable_device(pdev);
4179         return err;
4180 }
4181
4182 /**
4183  * ixgbevf_remove - Device Removal Routine
4184  * @pdev: PCI device information struct
4185  *
4186  * ixgbevf_remove is called by the PCI subsystem to alert the driver
4187  * that it should release a PCI device.  The could be caused by a
4188  * Hot-Plug event, or because the driver is going to be removed from
4189  * memory.
4190  **/
4191 static void ixgbevf_remove(struct pci_dev *pdev)
4192 {
4193         struct net_device *netdev = pci_get_drvdata(pdev);
4194         struct ixgbevf_adapter *adapter;
4195         bool disable_dev;
4196
4197         if (!netdev)
4198                 return;
4199
4200         adapter = netdev_priv(netdev);
4201
4202         set_bit(__IXGBEVF_REMOVING, &adapter->state);
4203         cancel_work_sync(&adapter->service_task);
4204
4205         if (netdev->reg_state == NETREG_REGISTERED)
4206                 unregister_netdev(netdev);
4207
4208         ixgbevf_clear_interrupt_scheme(adapter);
4209         ixgbevf_reset_interrupt_capability(adapter);
4210
4211         iounmap(adapter->io_addr);
4212         pci_release_regions(pdev);
4213
4214         hw_dbg(&adapter->hw, "Remove complete\n");
4215
4216         kfree(adapter->rss_key);
4217         disable_dev = !test_and_set_bit(__IXGBEVF_DISABLED, &adapter->state);
4218         free_netdev(netdev);
4219
4220         if (disable_dev)
4221                 pci_disable_device(pdev);
4222 }
4223
4224 /**
4225  * ixgbevf_io_error_detected - called when PCI error is detected
4226  * @pdev: Pointer to PCI device
4227  * @state: The current pci connection state
4228  *
4229  * This function is called after a PCI bus error affecting
4230  * this device has been detected.
4231  **/
4232 static pci_ers_result_t ixgbevf_io_error_detected(struct pci_dev *pdev,
4233                                                   pci_channel_state_t state)
4234 {
4235         struct net_device *netdev = pci_get_drvdata(pdev);
4236         struct ixgbevf_adapter *adapter = netdev_priv(netdev);
4237
4238         if (!test_bit(__IXGBEVF_SERVICE_INITED, &adapter->state))
4239                 return PCI_ERS_RESULT_DISCONNECT;
4240
4241         rtnl_lock();
4242         netif_device_detach(netdev);
4243
4244         if (state == pci_channel_io_perm_failure) {
4245                 rtnl_unlock();
4246                 return PCI_ERS_RESULT_DISCONNECT;
4247         }
4248
4249         if (netif_running(netdev))
4250                 ixgbevf_close_suspend(adapter);
4251
4252         if (!test_and_set_bit(__IXGBEVF_DISABLED, &adapter->state))
4253                 pci_disable_device(pdev);
4254         rtnl_unlock();
4255
4256         /* Request a slot slot reset. */
4257         return PCI_ERS_RESULT_NEED_RESET;
4258 }
4259
4260 /**
4261  * ixgbevf_io_slot_reset - called after the pci bus has been reset.
4262  * @pdev: Pointer to PCI device
4263  *
4264  * Restart the card from scratch, as if from a cold-boot. Implementation
4265  * resembles the first-half of the ixgbevf_resume routine.
4266  **/
4267 static pci_ers_result_t ixgbevf_io_slot_reset(struct pci_dev *pdev)
4268 {
4269         struct net_device *netdev = pci_get_drvdata(pdev);
4270         struct ixgbevf_adapter *adapter = netdev_priv(netdev);
4271
4272         if (pci_enable_device_mem(pdev)) {
4273                 dev_err(&pdev->dev,
4274                         "Cannot re-enable PCI device after reset.\n");
4275                 return PCI_ERS_RESULT_DISCONNECT;
4276         }
4277
4278         adapter->hw.hw_addr = adapter->io_addr;
4279         smp_mb__before_atomic();
4280         clear_bit(__IXGBEVF_DISABLED, &adapter->state);
4281         pci_set_master(pdev);
4282
4283         ixgbevf_reset(adapter);
4284
4285         return PCI_ERS_RESULT_RECOVERED;
4286 }
4287
4288 /**
4289  * ixgbevf_io_resume - called when traffic can start flowing again.
4290  * @pdev: Pointer to PCI device
4291  *
4292  * This callback is called when the error recovery driver tells us that
4293  * its OK to resume normal operation. Implementation resembles the
4294  * second-half of the ixgbevf_resume routine.
4295  **/
4296 static void ixgbevf_io_resume(struct pci_dev *pdev)
4297 {
4298         struct net_device *netdev = pci_get_drvdata(pdev);
4299
4300         rtnl_lock();
4301         if (netif_running(netdev))
4302                 ixgbevf_open(netdev);
4303
4304         netif_device_attach(netdev);
4305         rtnl_unlock();
4306 }
4307
4308 /* PCI Error Recovery (ERS) */
4309 static const struct pci_error_handlers ixgbevf_err_handler = {
4310         .error_detected = ixgbevf_io_error_detected,
4311         .slot_reset = ixgbevf_io_slot_reset,
4312         .resume = ixgbevf_io_resume,
4313 };
4314
4315 static struct pci_driver ixgbevf_driver = {
4316         .name           = ixgbevf_driver_name,
4317         .id_table       = ixgbevf_pci_tbl,
4318         .probe          = ixgbevf_probe,
4319         .remove         = ixgbevf_remove,
4320 #ifdef CONFIG_PM
4321         /* Power Management Hooks */
4322         .suspend        = ixgbevf_suspend,
4323         .resume         = ixgbevf_resume,
4324 #endif
4325         .shutdown       = ixgbevf_shutdown,
4326         .err_handler    = &ixgbevf_err_handler
4327 };
4328
4329 /**
4330  * ixgbevf_init_module - Driver Registration Routine
4331  *
4332  * ixgbevf_init_module is the first routine called when the driver is
4333  * loaded. All it does is register with the PCI subsystem.
4334  **/
4335 static int __init ixgbevf_init_module(void)
4336 {
4337         pr_info("%s - version %s\n", ixgbevf_driver_string,
4338                 ixgbevf_driver_version);
4339
4340         pr_info("%s\n", ixgbevf_copyright);
4341         ixgbevf_wq = create_singlethread_workqueue(ixgbevf_driver_name);
4342         if (!ixgbevf_wq) {
4343                 pr_err("%s: Failed to create workqueue\n", ixgbevf_driver_name);
4344                 return -ENOMEM;
4345         }
4346
4347         return pci_register_driver(&ixgbevf_driver);
4348 }
4349
4350 module_init(ixgbevf_init_module);
4351
4352 /**
4353  * ixgbevf_exit_module - Driver Exit Cleanup Routine
4354  *
4355  * ixgbevf_exit_module is called just before the driver is removed
4356  * from memory.
4357  **/
4358 static void __exit ixgbevf_exit_module(void)
4359 {
4360         pci_unregister_driver(&ixgbevf_driver);
4361         if (ixgbevf_wq) {
4362                 destroy_workqueue(ixgbevf_wq);
4363                 ixgbevf_wq = NULL;
4364         }
4365 }
4366
4367 #ifdef DEBUG
4368 /**
4369  * ixgbevf_get_hw_dev_name - return device name string
4370  * used by hardware layer to print debugging information
4371  **/
4372 char *ixgbevf_get_hw_dev_name(struct ixgbe_hw *hw)
4373 {
4374         struct ixgbevf_adapter *adapter = hw->back;
4375
4376         return adapter->netdev->name;
4377 }
4378
4379 #endif
4380 module_exit(ixgbevf_exit_module);
4381
4382 /* ixgbevf_main.c */