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

   1 /*******************************************************************************
   2
   3   Intel 82599 Virtual Function driver
   4   Copyright(c) 1999 - 2015 Intel Corporation.
   5
   6   This program is free software; you can redistribute it and/or modify it
   7   under the terms and conditions of the GNU General Public License,
   8   version 2, as published by the Free Software Foundation.
   9
  10   This program is distributed in the hope it will be useful, but WITHOUT
  11   ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
  12   FITNESS FOR A PARTICULAR PURPOSE.  See the GNU General Public License for
  13   more details.
  14
  15   You should have received a copy of the GNU General Public License along with
  16   this program; if not, see <http://www.gnu.org/licenses/>.
  17
  18   The full GNU General Public License is included in this distribution in
  19   the file called "COPYING".
  20
  21   Contact Information:
  22   e1000-devel Mailing List <e1000-devel@lists.sourceforge.net>
  23   Intel Corporation, 5200 N.E. Elam Young Parkway, Hillsboro, OR 97124-6497
  24
  25 *******************************************************************************/
  26
  27 /******************************************************************************
  28  Copyright (c)2006 - 2007 Myricom, Inc. for some LRO specific code
  29 ******************************************************************************/
  30
  31 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
  32
  33 #include <linux/types.h>
  34 #include <linux/bitops.h>
  35 #include <linux/module.h>
  36 #include <linux/pci.h>
  37 #include <linux/netdevice.h>
  38 #include <linux/vmalloc.h>
  39 #include <linux/string.h>
  40 #include <linux/in.h>
  41 #include <linux/ip.h>
  42 #include <linux/tcp.h>
  43 #include <linux/sctp.h>
  44 #include <linux/ipv6.h>
  45 #include <linux/slab.h>
  46 #include <net/checksum.h>
  47 #include <net/ip6_checksum.h>
  48 #include <linux/ethtool.h>
  49 #include <linux/if.h>
  50 #include <linux/if_vlan.h>
  51 #include <linux/prefetch.h>
  52
  53 #include "ixgbevf.h"
  54
  55 const char ixgbevf_driver_name[] = "ixgbevf";
  56 static const char ixgbevf_driver_string[] =
  57         "Intel(R) 10 Gigabit PCI Express Virtual Function Network Driver";
  58
  59 #define DRV_VERSION "2.12.1-k"
  60 const char ixgbevf_driver_version[] = DRV_VERSION;
  61 static char ixgbevf_copyright[] =
  62         "Copyright (c) 2009 - 2012 Intel Corporation.";
  63
  64 static const struct ixgbevf_info *ixgbevf_info_tbl[] = {
  65         [board_82599_vf] = &ixgbevf_82599_vf_info,
  66         [board_X540_vf]  = &ixgbevf_X540_vf_info,
  67         [board_X550_vf]  = &ixgbevf_X550_vf_info,
  68         [board_X550EM_x_vf] = &ixgbevf_X550EM_x_vf_info,
  69 };
  70
  71 /* ixgbevf_pci_tbl - PCI Device ID Table
  72  *
  73  * Wildcard entries (PCI_ANY_ID) should come last
  74  * Last entry must be all 0s
  75  *
  76  * { Vendor ID, Device ID, SubVendor ID, SubDevice ID,
  77  *   Class, Class Mask, private data (not used) }
  78  */
  79 static const struct pci_device_id ixgbevf_pci_tbl[] = {
  80         {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82599_VF), board_82599_vf },
  81         {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_X540_VF), board_X540_vf },
  82         {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_X550_VF), board_X550_vf },
  83         {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_X550EM_X_VF), board_X550EM_x_vf },
  84         /* required last entry */
  85         {0, }
  86 };
  87 MODULE_DEVICE_TABLE(pci, ixgbevf_pci_tbl);
  88
  89 MODULE_AUTHOR("Intel Corporation, <linux.nics@intel.com>");
  90 MODULE_DESCRIPTION("Intel(R) 10 Gigabit Virtual Function Network Driver");
  91 MODULE_LICENSE("GPL");
  92 MODULE_VERSION(DRV_VERSION);
  93
  94 #define DEFAULT_MSG_ENABLE (NETIF_MSG_DRV|NETIF_MSG_PROBE|NETIF_MSG_LINK)
  95 static int debug = -1;
  96 module_param(debug, int, 0);
  97 MODULE_PARM_DESC(debug, "Debug level (0=none,...,16=all)");
  98
  99 static void ixgbevf_service_event_schedule(struct ixgbevf_adapter *adapter)
 100 {
 101         if (!test_bit(__IXGBEVF_DOWN, &adapter->state) &&
 102             !test_bit(__IXGBEVF_REMOVING, &adapter->state) &&
 103             !test_and_set_bit(__IXGBEVF_SERVICE_SCHED, &adapter->state))
 104                 schedule_work(&adapter->service_task);
 105 }
 106
 107 static void ixgbevf_service_event_complete(struct ixgbevf_adapter *adapter)
 108 {
 109         BUG_ON(!test_bit(__IXGBEVF_SERVICE_SCHED, &adapter->state));
 110
 111         /* flush memory to make sure state is correct before next watchdog */
 112         smp_mb__before_atomic();
 113         clear_bit(__IXGBEVF_SERVICE_SCHED, &adapter->state);
 114 }
 115
 116 /* forward decls */
 117 static void ixgbevf_queue_reset_subtask(struct ixgbevf_adapter *adapter);
 118 static void ixgbevf_set_itr(struct ixgbevf_q_vector *q_vector);
 119 static void ixgbevf_free_all_rx_resources(struct ixgbevf_adapter *adapter);
 120
 121 static void ixgbevf_remove_adapter(struct ixgbe_hw *hw)
 122 {
 123         struct ixgbevf_adapter *adapter = hw->back;
 124
 125         if (!hw->hw_addr)
 126                 return;
 127         hw->hw_addr = NULL;
 128         dev_err(&adapter->pdev->dev, "Adapter removed\n");
 129         if (test_bit(__IXGBEVF_SERVICE_INITED, &adapter->state))
 130                 ixgbevf_service_event_schedule(adapter);
 131 }
 132
 133 static void ixgbevf_check_remove(struct ixgbe_hw *hw, u32 reg)
 134 {
 135         u32 value;
 136
 137         /* The following check not only optimizes a bit by not
 138          * performing a read on the status register when the
 139          * register just read was a status register read that
 140          * returned IXGBE_FAILED_READ_REG. It also blocks any
 141          * potential recursion.
 142          */
 143         if (reg == IXGBE_VFSTATUS) {
 144                 ixgbevf_remove_adapter(hw);
 145                 return;
 146         }
 147         value = ixgbevf_read_reg(hw, IXGBE_VFSTATUS);
 148         if (value == IXGBE_FAILED_READ_REG)
 149                 ixgbevf_remove_adapter(hw);
 150 }
 151
 152 u32 ixgbevf_read_reg(struct ixgbe_hw *hw, u32 reg)
 153 {
 154         u8 __iomem *reg_addr = ACCESS_ONCE(hw->hw_addr);
 155         u32 value;
 156
 157         if (IXGBE_REMOVED(reg_addr))
 158                 return IXGBE_FAILED_READ_REG;
 159         value = readl(reg_addr + reg);
 160         if (unlikely(value == IXGBE_FAILED_READ_REG))
 161                 ixgbevf_check_remove(hw, reg);
 162         return value;
 163 }
 164
 165 /**
 166  * ixgbevf_set_ivar - set IVAR registers - maps interrupt causes to vectors
 167  * @adapter: pointer to adapter struct
 168  * @direction: 0 for Rx, 1 for Tx, -1 for other causes
 169  * @queue: queue to map the corresponding interrupt to
 170  * @msix_vector: the vector to map to the corresponding queue
 171  **/
 172 static void ixgbevf_set_ivar(struct ixgbevf_adapter *adapter, s8 direction,
 173                              u8 queue, u8 msix_vector)
 174 {
 175         u32 ivar, index;
 176         struct ixgbe_hw *hw = &adapter->hw;
 177
 178         if (direction == -1) {
 179                 /* other causes */
 180                 msix_vector |= IXGBE_IVAR_ALLOC_VAL;
 181                 ivar = IXGBE_READ_REG(hw, IXGBE_VTIVAR_MISC);
 182                 ivar &= ~0xFF;
 183                 ivar |= msix_vector;
 184                 IXGBE_WRITE_REG(hw, IXGBE_VTIVAR_MISC, ivar);
 185         } else {
 186                 /* Tx or Rx causes */
 187                 msix_vector |= IXGBE_IVAR_ALLOC_VAL;
 188                 index = ((16 * (queue & 1)) + (8 * direction));
 189                 ivar = IXGBE_READ_REG(hw, IXGBE_VTIVAR(queue >> 1));
 190                 ivar &= ~(0xFF << index);
 191                 ivar |= (msix_vector << index);
 192                 IXGBE_WRITE_REG(hw, IXGBE_VTIVAR(queue >> 1), ivar);
 193         }
 194 }
 195
 196 static void ixgbevf_unmap_and_free_tx_resource(struct ixgbevf_ring *tx_ring,
 197                                         struct ixgbevf_tx_buffer *tx_buffer)
 198 {
 199         if (tx_buffer->skb) {
 200                 dev_kfree_skb_any(tx_buffer->skb);
 201                 if (dma_unmap_len(tx_buffer, len))
 202                         dma_unmap_single(tx_ring->dev,
 203                                          dma_unmap_addr(tx_buffer, dma),
 204                                          dma_unmap_len(tx_buffer, len),
 205                                          DMA_TO_DEVICE);
 206         } else if (dma_unmap_len(tx_buffer, len)) {
 207                 dma_unmap_page(tx_ring->dev,
 208                                dma_unmap_addr(tx_buffer, dma),
 209                                dma_unmap_len(tx_buffer, len),
 210                                DMA_TO_DEVICE);
 211         }
 212         tx_buffer->next_to_watch = NULL;
 213         tx_buffer->skb = NULL;
 214         dma_unmap_len_set(tx_buffer, len, 0);
 215         /* tx_buffer must be completely set up in the transmit path */
 216 }
 217
 218 static u64 ixgbevf_get_tx_completed(struct ixgbevf_ring *ring)
 219 {
 220         return ring->stats.packets;
 221 }
 222
 223 static u32 ixgbevf_get_tx_pending(struct ixgbevf_ring *ring)
 224 {
 225         struct ixgbevf_adapter *adapter = netdev_priv(ring->netdev);
 226         struct ixgbe_hw *hw = &adapter->hw;
 227
 228         u32 head = IXGBE_READ_REG(hw, IXGBE_VFTDH(ring->reg_idx));
 229         u32 tail = IXGBE_READ_REG(hw, IXGBE_VFTDT(ring->reg_idx));
 230
 231         if (head != tail)
 232                 return (head < tail) ?
 233                         tail - head : (tail + ring->count - head);
 234
 235         return 0;
 236 }
 237
 238 static inline bool ixgbevf_check_tx_hang(struct ixgbevf_ring *tx_ring)
 239 {
 240         u32 tx_done = ixgbevf_get_tx_completed(tx_ring);
 241         u32 tx_done_old = tx_ring->tx_stats.tx_done_old;
 242         u32 tx_pending = ixgbevf_get_tx_pending(tx_ring);
 243
 244         clear_check_for_tx_hang(tx_ring);
 245
 246         /* Check for a hung queue, but be thorough. This verifies
 247          * that a transmit has been completed since the previous
 248          * check AND there is at least one packet pending. The
 249          * ARMED bit is set to indicate a potential hang.
 250          */
 251         if ((tx_done_old == tx_done) && tx_pending) {
 252                 /* make sure it is true for two checks in a row */
 253                 return test_and_set_bit(__IXGBEVF_HANG_CHECK_ARMED,
 254                                         &tx_ring->state);
 255         }
 256         /* reset the countdown */
 257         clear_bit(__IXGBEVF_HANG_CHECK_ARMED, &tx_ring->state);
 258
 259         /* update completed stats and continue */
 260         tx_ring->tx_stats.tx_done_old = tx_done;
 261
 262         return false;
 263 }
 264
 265 static void ixgbevf_tx_timeout_reset(struct ixgbevf_adapter *adapter)
 266 {
 267         /* Do the reset outside of interrupt context */
 268         if (!test_bit(__IXGBEVF_DOWN, &adapter->state)) {
 269                 adapter->flags |= IXGBEVF_FLAG_RESET_REQUESTED;
 270                 ixgbevf_service_event_schedule(adapter);
 271         }
 272 }
 273
 274 /**
 275  * ixgbevf_tx_timeout - Respond to a Tx Hang
 276  * @netdev: network interface device structure
 277  **/
 278 static void ixgbevf_tx_timeout(struct net_device *netdev)
 279 {
 280         struct ixgbevf_adapter *adapter = netdev_priv(netdev);
 281
 282         ixgbevf_tx_timeout_reset(adapter);
 283 }
 284
 285 /**
 286  * ixgbevf_clean_tx_irq - Reclaim resources after transmit completes
 287  * @q_vector: board private structure
 288  * @tx_ring: tx ring to clean
 289  **/
 290 static bool ixgbevf_clean_tx_irq(struct ixgbevf_q_vector *q_vector,
 291                                  struct ixgbevf_ring *tx_ring)
 292 {
 293         struct ixgbevf_adapter *adapter = q_vector->adapter;
 294         struct ixgbevf_tx_buffer *tx_buffer;
 295         union ixgbe_adv_tx_desc *tx_desc;
 296         unsigned int total_bytes = 0, total_packets = 0;
 297         unsigned int budget = tx_ring->count / 2;
 298         unsigned int i = tx_ring->next_to_clean;
 299
 300         if (test_bit(__IXGBEVF_DOWN, &adapter->state))
 301                 return true;
 302
 303         tx_buffer = &tx_ring->tx_buffer_info[i];
 304         tx_desc = IXGBEVF_TX_DESC(tx_ring, i);
 305         i -= tx_ring->count;
 306
 307         do {
 308                 union ixgbe_adv_tx_desc *eop_desc = tx_buffer->next_to_watch;
 309
 310                 /* if next_to_watch is not set then there is no work pending */
 311                 if (!eop_desc)
 312                         break;
 313
 314                 /* prevent any other reads prior to eop_desc */
 315                 read_barrier_depends();
 316
 317                 /* if DD is not set pending work has not been completed */
 318                 if (!(eop_desc->wb.status & cpu_to_le32(IXGBE_TXD_STAT_DD)))
 319                         break;
 320
 321                 /* clear next_to_watch to prevent false hangs */
 322                 tx_buffer->next_to_watch = NULL;
 323
 324                 /* update the statistics for this packet */
 325                 total_bytes += tx_buffer->bytecount;
 326                 total_packets += tx_buffer->gso_segs;
 327
 328                 /* free the skb */
 329                 dev_kfree_skb_any(tx_buffer->skb);
 330
 331                 /* unmap skb header data */
 332                 dma_unmap_single(tx_ring->dev,
 333                                  dma_unmap_addr(tx_buffer, dma),
 334                                  dma_unmap_len(tx_buffer, len),
 335                                  DMA_TO_DEVICE);
 336
 337                 /* clear tx_buffer data */
 338                 tx_buffer->skb = NULL;
 339                 dma_unmap_len_set(tx_buffer, len, 0);
 340
 341                 /* unmap remaining buffers */
 342                 while (tx_desc != eop_desc) {
 343                         tx_buffer++;
 344                         tx_desc++;
 345                         i++;
 346                         if (unlikely(!i)) {
 347                                 i -= tx_ring->count;
 348                                 tx_buffer = tx_ring->tx_buffer_info;
 349                                 tx_desc = IXGBEVF_TX_DESC(tx_ring, 0);
 350                         }
 351
 352                         /* unmap any remaining paged data */
 353                         if (dma_unmap_len(tx_buffer, len)) {
 354                                 dma_unmap_page(tx_ring->dev,
 355                                                dma_unmap_addr(tx_buffer, dma),
 356                                                dma_unmap_len(tx_buffer, len),
 357                                                DMA_TO_DEVICE);
 358                                 dma_unmap_len_set(tx_buffer, len, 0);
 359                         }
 360                 }
 361
 362                 /* move us one more past the eop_desc for start of next pkt */
 363                 tx_buffer++;
 364                 tx_desc++;
 365                 i++;
 366                 if (unlikely(!i)) {
 367                         i -= tx_ring->count;
 368                         tx_buffer = tx_ring->tx_buffer_info;
 369                         tx_desc = IXGBEVF_TX_DESC(tx_ring, 0);
 370                 }
 371
 372                 /* issue prefetch for next Tx descriptor */
 373                 prefetch(tx_desc);
 374
 375                 /* update budget accounting */
 376                 budget--;
 377         } while (likely(budget));
 378
 379         i += tx_ring->count;
 380         tx_ring->next_to_clean = i;
 381         u64_stats_update_begin(&tx_ring->syncp);
 382         tx_ring->stats.bytes += total_bytes;
 383         tx_ring->stats.packets += total_packets;
 384         u64_stats_update_end(&tx_ring->syncp);
 385         q_vector->tx.total_bytes += total_bytes;
 386         q_vector->tx.total_packets += total_packets;
 387
 388         if (check_for_tx_hang(tx_ring) && ixgbevf_check_tx_hang(tx_ring)) {
 389                 struct ixgbe_hw *hw = &adapter->hw;
 390                 union ixgbe_adv_tx_desc *eop_desc;
 391
 392                 eop_desc = tx_ring->tx_buffer_info[i].next_to_watch;
 393
 394                 pr_err("Detected Tx Unit Hang\n"
 395                        "  Tx Queue             <%d>\n"
 396                        "  TDH, TDT             <%x>, <%x>\n"
 397                        "  next_to_use          <%x>\n"
 398                        "  next_to_clean        <%x>\n"
 399                        "tx_buffer_info[next_to_clean]\n"
 400                        "  next_to_watch        <%p>\n"
 401                        "  eop_desc->wb.status  <%x>\n"
 402                        "  time_stamp           <%lx>\n"
 403                        "  jiffies              <%lx>\n",
 404                        tx_ring->queue_index,
 405                        IXGBE_READ_REG(hw, IXGBE_VFTDH(tx_ring->reg_idx)),
 406                        IXGBE_READ_REG(hw, IXGBE_VFTDT(tx_ring->reg_idx)),
 407                        tx_ring->next_to_use, i,
 408                        eop_desc, (eop_desc ? eop_desc->wb.status : 0),
 409                        tx_ring->tx_buffer_info[i].time_stamp, jiffies);
 410
 411                 netif_stop_subqueue(tx_ring->netdev, tx_ring->queue_index);
 412
 413                 /* schedule immediate reset if we believe we hung */
 414                 ixgbevf_tx_timeout_reset(adapter);
 415
 416                 return true;
 417         }
 418
 419 #define TX_WAKE_THRESHOLD (DESC_NEEDED * 2)
 420         if (unlikely(total_packets && netif_carrier_ok(tx_ring->netdev) &&
 421                      (ixgbevf_desc_unused(tx_ring) >= TX_WAKE_THRESHOLD))) {
 422                 /* Make sure that anybody stopping the queue after this
 423                  * sees the new next_to_clean.
 424                  */
 425                 smp_mb();
 426
 427                 if (__netif_subqueue_stopped(tx_ring->netdev,
 428                                              tx_ring->queue_index) &&
 429                     !test_bit(__IXGBEVF_DOWN, &adapter->state)) {
 430                         netif_wake_subqueue(tx_ring->netdev,
 431                                             tx_ring->queue_index);
 432                         ++tx_ring->tx_stats.restart_queue;
 433                 }
 434         }
 435
 436         return !!budget;
 437 }
 438
 439 /**
 440  * ixgbevf_rx_skb - Helper function to determine proper Rx method
 441  * @q_vector: structure containing interrupt and ring information
 442  * @skb: packet to send up
 443  **/
 444 static void ixgbevf_rx_skb(struct ixgbevf_q_vector *q_vector,
 445                            struct sk_buff *skb)
 446 {
 447 #ifdef CONFIG_NET_RX_BUSY_POLL
 448         skb_mark_napi_id(skb, &q_vector->napi);
 449
 450         if (ixgbevf_qv_busy_polling(q_vector)) {
 451                 netif_receive_skb(skb);
 452                 /* exit early if we busy polled */
 453                 return;
 454         }
 455 #endif /* CONFIG_NET_RX_BUSY_POLL */
 456
 457         napi_gro_receive(&q_vector->napi, skb);
 458 }
 459
 460 #define IXGBE_RSS_L4_TYPES_MASK \
 461         ((1ul << IXGBE_RXDADV_RSSTYPE_IPV4_TCP) | \
 462          (1ul << IXGBE_RXDADV_RSSTYPE_IPV4_UDP) | \
 463          (1ul << IXGBE_RXDADV_RSSTYPE_IPV6_TCP) | \
 464          (1ul << IXGBE_RXDADV_RSSTYPE_IPV6_UDP))
 465
 466 static inline void ixgbevf_rx_hash(struct ixgbevf_ring *ring,
 467                                    union ixgbe_adv_rx_desc *rx_desc,
 468                                    struct sk_buff *skb)
 469 {
 470         u16 rss_type;
 471
 472         if (!(ring->netdev->features & NETIF_F_RXHASH))
 473                 return;
 474
 475         rss_type = le16_to_cpu(rx_desc->wb.lower.lo_dword.hs_rss.pkt_info) &
 476                    IXGBE_RXDADV_RSSTYPE_MASK;
 477
 478         if (!rss_type)
 479                 return;
 480
 481         skb_set_hash(skb, le32_to_cpu(rx_desc->wb.lower.hi_dword.rss),
 482                      (IXGBE_RSS_L4_TYPES_MASK & (1ul << rss_type)) ?
 483                      PKT_HASH_TYPE_L4 : PKT_HASH_TYPE_L3);
 484 }
 485
 486 /**
 487  * ixgbevf_rx_checksum - indicate in skb if hw indicated a good cksum
 488  * @ring: structure containig ring specific data
 489  * @rx_desc: current Rx descriptor being processed
 490  * @skb: skb currently being received and modified
 491  **/
 492 static inline void ixgbevf_rx_checksum(struct ixgbevf_ring *ring,
 493                                        union ixgbe_adv_rx_desc *rx_desc,
 494                                        struct sk_buff *skb)
 495 {
 496         skb_checksum_none_assert(skb);
 497
 498         /* Rx csum disabled */
 499         if (!(ring->netdev->features & NETIF_F_RXCSUM))
 500                 return;
 501
 502         /* if IP and error */
 503         if (ixgbevf_test_staterr(rx_desc, IXGBE_RXD_STAT_IPCS) &&
 504             ixgbevf_test_staterr(rx_desc, IXGBE_RXDADV_ERR_IPE)) {
 505                 ring->rx_stats.csum_err++;
 506                 return;
 507         }
 508
 509         if (!ixgbevf_test_staterr(rx_desc, IXGBE_RXD_STAT_L4CS))
 510                 return;
 511
 512         if (ixgbevf_test_staterr(rx_desc, IXGBE_RXDADV_ERR_TCPE)) {
 513                 ring->rx_stats.csum_err++;
 514                 return;
 515         }
 516
 517         /* It must be a TCP or UDP packet with a valid checksum */
 518         skb->ip_summed = CHECKSUM_UNNECESSARY;
 519 }
 520
 521 /**
 522  * ixgbevf_process_skb_fields - Populate skb header fields from Rx descriptor
 523  * @rx_ring: rx descriptor ring packet is being transacted on
 524  * @rx_desc: pointer to the EOP Rx descriptor
 525  * @skb: pointer to current skb being populated
 526  *
 527  * This function checks the ring, descriptor, and packet information in
 528  * order to populate the checksum, VLAN, protocol, and other fields within
 529  * the skb.
 530  **/
 531 static void ixgbevf_process_skb_fields(struct ixgbevf_ring *rx_ring,
 532                                        union ixgbe_adv_rx_desc *rx_desc,
 533                                        struct sk_buff *skb)
 534 {
 535         ixgbevf_rx_hash(rx_ring, rx_desc, skb);
 536         ixgbevf_rx_checksum(rx_ring, rx_desc, skb);
 537
 538         if (ixgbevf_test_staterr(rx_desc, IXGBE_RXD_STAT_VP)) {
 539                 u16 vid = le16_to_cpu(rx_desc->wb.upper.vlan);
 540                 unsigned long *active_vlans = netdev_priv(rx_ring->netdev);
 541
 542                 if (test_bit(vid & VLAN_VID_MASK, active_vlans))
 543                         __vlan_hwaccel_put_tag(skb, htons(ETH_P_8021Q), vid);
 544         }
 545
 546         skb->protocol = eth_type_trans(skb, rx_ring->netdev);
 547 }
 548
 549 /**
 550  * ixgbevf_is_non_eop - process handling of non-EOP buffers
 551  * @rx_ring: Rx ring being processed
 552  * @rx_desc: Rx descriptor for current buffer
 553  * @skb: current socket buffer containing buffer in progress
 554  *
 555  * This function updates next to clean.  If the buffer is an EOP buffer
 556  * this function exits returning false, otherwise it will place the
 557  * sk_buff in the next buffer to be chained and return true indicating
 558  * that this is in fact a non-EOP buffer.
 559  **/
 560 static bool ixgbevf_is_non_eop(struct ixgbevf_ring *rx_ring,
 561                                union ixgbe_adv_rx_desc *rx_desc)
 562 {
 563         u32 ntc = rx_ring->next_to_clean + 1;
 564
 565         /* fetch, update, and store next to clean */
 566         ntc = (ntc < rx_ring->count) ? ntc : 0;
 567         rx_ring->next_to_clean = ntc;
 568
 569         prefetch(IXGBEVF_RX_DESC(rx_ring, ntc));
 570
 571         if (likely(ixgbevf_test_staterr(rx_desc, IXGBE_RXD_STAT_EOP)))
 572                 return false;
 573
 574         return true;
 575 }
 576
 577 static bool ixgbevf_alloc_mapped_page(struct ixgbevf_ring *rx_ring,
 578                                       struct ixgbevf_rx_buffer *bi)
 579 {
 580         struct page *page = bi->page;
 581         dma_addr_t dma = bi->dma;
 582
 583         /* since we are recycling buffers we should seldom need to alloc */
 584         if (likely(page))
 585                 return true;
 586
 587         /* alloc new page for storage */
 588         page = dev_alloc_page();
 589         if (unlikely(!page)) {
 590                 rx_ring->rx_stats.alloc_rx_page_failed++;
 591                 return false;
 592         }
 593
 594         /* map page for use */
 595         dma = dma_map_page(rx_ring->dev, page, 0,
 596                            PAGE_SIZE, DMA_FROM_DEVICE);
 597
 598         /* if mapping failed free memory back to system since
 599          * there isn't much point in holding memory we can't use
 600          */
 601         if (dma_mapping_error(rx_ring->dev, dma)) {
 602                 __free_page(page);
 603
 604                 rx_ring->rx_stats.alloc_rx_buff_failed++;
 605                 return false;
 606         }
 607
 608         bi->dma = dma;
 609         bi->page = page;
 610         bi->page_offset = 0;
 611
 612         return true;
 613 }
 614
 615 /**
 616  * ixgbevf_alloc_rx_buffers - Replace used receive buffers; packet split
 617  * @rx_ring: rx descriptor ring (for a specific queue) to setup buffers on
 618  * @cleaned_count: number of buffers to replace
 619  **/
 620 static void ixgbevf_alloc_rx_buffers(struct ixgbevf_ring *rx_ring,
 621                                      u16 cleaned_count)
 622 {
 623         union ixgbe_adv_rx_desc *rx_desc;
 624         struct ixgbevf_rx_buffer *bi;
 625         unsigned int i = rx_ring->next_to_use;
 626
 627         /* nothing to do or no valid netdev defined */
 628         if (!cleaned_count || !rx_ring->netdev)
 629                 return;
 630
 631         rx_desc = IXGBEVF_RX_DESC(rx_ring, i);
 632         bi = &rx_ring->rx_buffer_info[i];
 633         i -= rx_ring->count;
 634
 635         do {
 636                 if (!ixgbevf_alloc_mapped_page(rx_ring, bi))
 637                         break;
 638
 639                 /* Refresh the desc even if pkt_addr didn't change
 640                  * because each write-back erases this info.
 641                  */
 642                 rx_desc->read.pkt_addr = cpu_to_le64(bi->dma + bi->page_offset);
 643
 644                 rx_desc++;
 645                 bi++;
 646                 i++;
 647                 if (unlikely(!i)) {
 648                         rx_desc = IXGBEVF_RX_DESC(rx_ring, 0);
 649                         bi = rx_ring->rx_buffer_info;
 650                         i -= rx_ring->count;
 651                 }
 652
 653                 /* clear the hdr_addr for the next_to_use descriptor */
 654                 rx_desc->read.hdr_addr = 0;
 655
 656                 cleaned_count--;
 657         } while (cleaned_count);
 658
 659         i += rx_ring->count;
 660
 661         if (rx_ring->next_to_use != i) {
 662                 /* record the next descriptor to use */
 663                 rx_ring->next_to_use = i;
 664
 665                 /* update next to alloc since we have filled the ring */
 666                 rx_ring->next_to_alloc = i;
 667
 668                 /* Force memory writes to complete before letting h/w
 669                  * know there are new descriptors to fetch.  (Only
 670                  * applicable for weak-ordered memory model archs,
 671                  * such as IA-64).
 672                  */
 673                 wmb();
 674                 ixgbevf_write_tail(rx_ring, i);
 675         }
 676 }
 677
 678 /**
 679  * ixgbevf_cleanup_headers - Correct corrupted or empty headers
 680  * @rx_ring: rx descriptor ring packet is being transacted on
 681  * @rx_desc: pointer to the EOP Rx descriptor
 682  * @skb: pointer to current skb being fixed
 683  *
 684  * Check for corrupted packet headers caused by senders on the local L2
 685  * embedded NIC switch not setting up their Tx Descriptors right.  These
 686  * should be very rare.
 687  *
 688  * Also address the case where we are pulling data in on pages only
 689  * and as such no data is present in the skb header.
 690  *
 691  * In addition if skb is not at least 60 bytes we need to pad it so that
 692  * it is large enough to qualify as a valid Ethernet frame.
 693  *
 694  * Returns true if an error was encountered and skb was freed.
 695  **/
 696 static bool ixgbevf_cleanup_headers(struct ixgbevf_ring *rx_ring,
 697                                     union ixgbe_adv_rx_desc *rx_desc,
 698                                     struct sk_buff *skb)
 699 {
 700         /* verify that the packet does not have any known errors */
 701         if (unlikely(ixgbevf_test_staterr(rx_desc,
 702                                           IXGBE_RXDADV_ERR_FRAME_ERR_MASK))) {
 703                 struct net_device *netdev = rx_ring->netdev;
 704
 705                 if (!(netdev->features & NETIF_F_RXALL)) {
 706                         dev_kfree_skb_any(skb);
 707                         return true;
 708                 }
 709         }
 710
 711         /* if eth_skb_pad returns an error the skb was freed */
 712         if (eth_skb_pad(skb))
 713                 return true;
 714
 715         return false;
 716 }
 717
 718 /**
 719  * ixgbevf_reuse_rx_page - page flip buffer and store it back on the ring
 720  * @rx_ring: rx descriptor ring to store buffers on
 721  * @old_buff: donor buffer to have page reused
 722  *
 723  * Synchronizes page for reuse by the adapter
 724  **/
 725 static void ixgbevf_reuse_rx_page(struct ixgbevf_ring *rx_ring,
 726                                   struct ixgbevf_rx_buffer *old_buff)
 727 {
 728         struct ixgbevf_rx_buffer *new_buff;
 729         u16 nta = rx_ring->next_to_alloc;
 730
 731         new_buff = &rx_ring->rx_buffer_info[nta];
 732
 733         /* update, and store next to alloc */
 734         nta++;
 735         rx_ring->next_to_alloc = (nta < rx_ring->count) ? nta : 0;
 736
 737         /* transfer page from old buffer to new buffer */
 738         new_buff->page = old_buff->page;
 739         new_buff->dma = old_buff->dma;
 740         new_buff->page_offset = old_buff->page_offset;
 741
 742         /* sync the buffer for use by the device */
 743         dma_sync_single_range_for_device(rx_ring->dev, new_buff->dma,
 744                                          new_buff->page_offset,
 745                                          IXGBEVF_RX_BUFSZ,
 746                                          DMA_FROM_DEVICE);
 747 }
 748
 749 static inline bool ixgbevf_page_is_reserved(struct page *page)
 750 {
 751         return (page_to_nid(page) != numa_mem_id()) || page_is_pfmemalloc(page);
 752 }
 753
 754 /**
 755  * ixgbevf_add_rx_frag - Add contents of Rx buffer to sk_buff
 756  * @rx_ring: rx descriptor ring to transact packets on
 757  * @rx_buffer: buffer containing page to add
 758  * @rx_desc: descriptor containing length of buffer written by hardware
 759  * @skb: sk_buff to place the data into
 760  *
 761  * This function will add the data contained in rx_buffer->page to the skb.
 762  * This is done either through a direct copy if the data in the buffer is
 763  * less than the skb header size, otherwise it will just attach the page as
 764  * a frag to the skb.
 765  *
 766  * The function will then update the page offset if necessary and return
 767  * true if the buffer can be reused by the adapter.
 768  **/
 769 static bool ixgbevf_add_rx_frag(struct ixgbevf_ring *rx_ring,
 770                                 struct ixgbevf_rx_buffer *rx_buffer,
 771                                 union ixgbe_adv_rx_desc *rx_desc,
 772                                 struct sk_buff *skb)
 773 {
 774         struct page *page = rx_buffer->page;
 775         unsigned char *va = page_address(page) + rx_buffer->page_offset;
 776         unsigned int size = le16_to_cpu(rx_desc->wb.upper.length);
 777 #if (PAGE_SIZE < 8192)
 778         unsigned int truesize = IXGBEVF_RX_BUFSZ;
 779 #else
 780         unsigned int truesize = ALIGN(size, L1_CACHE_BYTES);
 781 #endif
 782         unsigned int pull_len;
 783
 784         if (unlikely(skb_is_nonlinear(skb)))
 785                 goto add_tail_frag;
 786
 787         if (likely(size <= IXGBEVF_RX_HDR_SIZE)) {
 788                 memcpy(__skb_put(skb, size), va, ALIGN(size, sizeof(long)));
 789
 790                 /* page is not reserved, we can reuse buffer as is */
 791                 if (likely(!ixgbevf_page_is_reserved(page)))
 792                         return true;
 793
 794                 /* this page cannot be reused so discard it */
 795                 put_page(page);
 796                 return false;
 797         }
 798
 799         /* we need the header to contain the greater of either ETH_HLEN or
 800          * 60 bytes if the skb->len is less than 60 for skb_pad.
 801          */
 802         pull_len = eth_get_headlen(va, IXGBEVF_RX_HDR_SIZE);
 803
 804         /* align pull length to size of long to optimize memcpy performance */
 805         memcpy(__skb_put(skb, pull_len), va, ALIGN(pull_len, sizeof(long)));
 806
 807         /* update all of the pointers */
 808         va += pull_len;
 809         size -= pull_len;
 810
 811 add_tail_frag:
 812         skb_add_rx_frag(skb, skb_shinfo(skb)->nr_frags, page,
 813                         (unsigned long)va & ~PAGE_MASK, size, truesize);
 814
 815         /* avoid re-using remote pages */
 816         if (unlikely(ixgbevf_page_is_reserved(page)))
 817                 return false;
 818
 819 #if (PAGE_SIZE < 8192)
 820         /* if we are only owner of page we can reuse it */
 821         if (unlikely(page_count(page) != 1))
 822                 return false;
 823
 824         /* flip page offset to other buffer */
 825         rx_buffer->page_offset ^= IXGBEVF_RX_BUFSZ;
 826
 827 #else
 828         /* move offset up to the next cache line */
 829         rx_buffer->page_offset += truesize;
 830
 831         if (rx_buffer->page_offset > (PAGE_SIZE - IXGBEVF_RX_BUFSZ))
 832                 return false;
 833
 834 #endif
 835         /* Even if we own the page, we are not allowed to use atomic_set()
 836          * This would break get_page_unless_zero() users.
 837          */
 838         atomic_inc(&page->_count);
 839
 840         return true;
 841 }
 842
 843 static struct sk_buff *ixgbevf_fetch_rx_buffer(struct ixgbevf_ring *rx_ring,
 844                                                union ixgbe_adv_rx_desc *rx_desc,
 845                                                struct sk_buff *skb)
 846 {
 847         struct ixgbevf_rx_buffer *rx_buffer;
 848         struct page *page;
 849
 850         rx_buffer = &rx_ring->rx_buffer_info[rx_ring->next_to_clean];
 851         page = rx_buffer->page;
 852         prefetchw(page);
 853
 854         if (likely(!skb)) {
 855                 void *page_addr = page_address(page) +
 856                                   rx_buffer->page_offset;
 857
 858                 /* prefetch first cache line of first page */
 859                 prefetch(page_addr);
 860 #if L1_CACHE_BYTES < 128
 861                 prefetch(page_addr + L1_CACHE_BYTES);
 862 #endif
 863
 864                 /* allocate a skb to store the frags */
 865                 skb = netdev_alloc_skb_ip_align(rx_ring->netdev,
 866                                                 IXGBEVF_RX_HDR_SIZE);
 867                 if (unlikely(!skb)) {
 868                         rx_ring->rx_stats.alloc_rx_buff_failed++;
 869                         return NULL;
 870                 }
 871
 872                 /* we will be copying header into skb->data in
 873                  * pskb_may_pull so it is in our interest to prefetch
 874                  * it now to avoid a possible cache miss
 875                  */
 876                 prefetchw(skb->data);
 877         }
 878
 879         /* we are reusing so sync this buffer for CPU use */
 880         dma_sync_single_range_for_cpu(rx_ring->dev,
 881                                       rx_buffer->dma,
 882                                       rx_buffer->page_offset,
 883                                       IXGBEVF_RX_BUFSZ,
 884                                       DMA_FROM_DEVICE);
 885
 886         /* pull page into skb */
 887         if (ixgbevf_add_rx_frag(rx_ring, rx_buffer, rx_desc, skb)) {
 888                 /* hand second half of page back to the ring */
 889                 ixgbevf_reuse_rx_page(rx_ring, rx_buffer);
 890         } else {
 891                 /* we are not reusing the buffer so unmap it */
 892                 dma_unmap_page(rx_ring->dev, rx_buffer->dma,
 893                                PAGE_SIZE, DMA_FROM_DEVICE);
 894         }
 895
 896         /* clear contents of buffer_info */
 897         rx_buffer->dma = 0;
 898         rx_buffer->page = NULL;
 899
 900         return skb;
 901 }
 902
 903 static inline void ixgbevf_irq_enable_queues(struct ixgbevf_adapter *adapter,
 904                                              u32 qmask)
 905 {
 906         struct ixgbe_hw *hw = &adapter->hw;
 907
 908         IXGBE_WRITE_REG(hw, IXGBE_VTEIMS, qmask);
 909 }
 910
 911 static int ixgbevf_clean_rx_irq(struct ixgbevf_q_vector *q_vector,
 912                                 struct ixgbevf_ring *rx_ring,
 913                                 int budget)
 914 {
 915         unsigned int total_rx_bytes = 0, total_rx_packets = 0;
 916         u16 cleaned_count = ixgbevf_desc_unused(rx_ring);
 917         struct sk_buff *skb = rx_ring->skb;
 918
 919         while (likely(total_rx_packets < budget)) {
 920                 union ixgbe_adv_rx_desc *rx_desc;
 921
 922                 /* return some buffers to hardware, one at a time is too slow */
 923                 if (cleaned_count >= IXGBEVF_RX_BUFFER_WRITE) {
 924                         ixgbevf_alloc_rx_buffers(rx_ring, cleaned_count);
 925                         cleaned_count = 0;
 926                 }
 927
 928                 rx_desc = IXGBEVF_RX_DESC(rx_ring, rx_ring->next_to_clean);
 929
 930                 if (!ixgbevf_test_staterr(rx_desc, IXGBE_RXD_STAT_DD))
 931                         break;
 932
 933                 /* This memory barrier is needed to keep us from reading
 934                  * any other fields out of the rx_desc until we know the
 935                  * RXD_STAT_DD bit is set
 936                  */
 937                 rmb();
 938
 939                 /* retrieve a buffer from the ring */
 940                 skb = ixgbevf_fetch_rx_buffer(rx_ring, rx_desc, skb);
 941
 942                 /* exit if we failed to retrieve a buffer */
 943                 if (!skb)
 944                         break;
 945
 946                 cleaned_count++;
 947
 948                 /* fetch next buffer in frame if non-eop */
 949                 if (ixgbevf_is_non_eop(rx_ring, rx_desc))
 950                         continue;
 951
 952                 /* verify the packet layout is correct */
 953                 if (ixgbevf_cleanup_headers(rx_ring, rx_desc, skb)) {
 954                         skb = NULL;
 955                         continue;
 956                 }
 957
 958                 /* probably a little skewed due to removing CRC */
 959                 total_rx_bytes += skb->len;
 960
 961                 /* Workaround hardware that can't do proper VEPA multicast
 962                  * source pruning.
 963                  */
 964                 if ((skb->pkt_type == PACKET_BROADCAST ||
 965                      skb->pkt_type == PACKET_MULTICAST) &&
 966                     ether_addr_equal(rx_ring->netdev->dev_addr,
 967                                      eth_hdr(skb)->h_source)) {
 968                         dev_kfree_skb_irq(skb);
 969                         continue;
 970                 }
 971
 972                 /* populate checksum, VLAN, and protocol */
 973                 ixgbevf_process_skb_fields(rx_ring, rx_desc, skb);
 974
 975                 ixgbevf_rx_skb(q_vector, skb);
 976
 977                 /* reset skb pointer */
 978                 skb = NULL;
 979
 980                 /* update budget accounting */
 981                 total_rx_packets++;
 982         }
 983
 984         /* place incomplete frames back on ring for completion */
 985         rx_ring->skb = skb;
 986
 987         u64_stats_update_begin(&rx_ring->syncp);
 988         rx_ring->stats.packets += total_rx_packets;
 989         rx_ring->stats.bytes += total_rx_bytes;
 990         u64_stats_update_end(&rx_ring->syncp);
 991         q_vector->rx.total_packets += total_rx_packets;
 992         q_vector->rx.total_bytes += total_rx_bytes;
 993
 994         return total_rx_packets;
 995 }
 996
 997 /**
 998  * ixgbevf_poll - NAPI polling calback
 999  * @napi: napi struct with our devices info in it
1000  * @budget: amount of work driver is allowed to do this pass, in packets
1001  *
1002  * This function will clean more than one or more rings associated with a
1003  * q_vector.
1004  **/
1005 static int ixgbevf_poll(struct napi_struct *napi, int budget)
1006 {
1007         struct ixgbevf_q_vector *q_vector =
1008                 container_of(napi, struct ixgbevf_q_vector, napi);
1009         struct ixgbevf_adapter *adapter = q_vector->adapter;
1010         struct ixgbevf_ring *ring;
1011         int per_ring_budget, work_done = 0;
1012         bool clean_complete = true;
1013
1014         ixgbevf_for_each_ring(ring, q_vector->tx)
1015                 clean_complete &= ixgbevf_clean_tx_irq(q_vector, ring);
1016
1017 #ifdef CONFIG_NET_RX_BUSY_POLL
1018         if (!ixgbevf_qv_lock_napi(q_vector))
1019                 return budget;
1020 #endif
1021
1022         /* attempt to distribute budget to each queue fairly, but don't allow
1023          * the budget to go below 1 because we'll exit polling
1024          */
1025         if (q_vector->rx.count > 1)
1026                 per_ring_budget = max(budget/q_vector->rx.count, 1);
1027         else
1028                 per_ring_budget = budget;
1029
1030         ixgbevf_for_each_ring(ring, q_vector->rx) {
1031                 int cleaned = ixgbevf_clean_rx_irq(q_vector, ring,
1032                                                    per_ring_budget);
1033                 work_done += cleaned;
1034                 clean_complete &= (cleaned < per_ring_budget);
1035         }
1036
1037 #ifdef CONFIG_NET_RX_BUSY_POLL
1038         ixgbevf_qv_unlock_napi(q_vector);
1039 #endif
1040
1041         /* If all work not completed, return budget and keep polling */
1042         if (!clean_complete)
1043                 return budget;
1044         /* all work done, exit the polling mode */
1045         napi_complete_done(napi, work_done);
1046         if (adapter->rx_itr_setting & 1)
1047                 ixgbevf_set_itr(q_vector);
1048         if (!test_bit(__IXGBEVF_DOWN, &adapter->state) &&
1049             !test_bit(__IXGBEVF_REMOVING, &adapter->state))
1050                 ixgbevf_irq_enable_queues(adapter,
1051                                           1 << q_vector->v_idx);
1052
1053         return 0;
1054 }
1055
1056 /**
1057  * ixgbevf_write_eitr - write VTEITR register in hardware specific way
1058  * @q_vector: structure containing interrupt and ring information
1059  **/
1060 void ixgbevf_write_eitr(struct ixgbevf_q_vector *q_vector)
1061 {
1062         struct ixgbevf_adapter *adapter = q_vector->adapter;
1063         struct ixgbe_hw *hw = &adapter->hw;
1064         int v_idx = q_vector->v_idx;
1065         u32 itr_reg = q_vector->itr & IXGBE_MAX_EITR;
1066
1067         /* set the WDIS bit to not clear the timer bits and cause an
1068          * immediate assertion of the interrupt
1069          */
1070         itr_reg |= IXGBE_EITR_CNT_WDIS;
1071
1072         IXGBE_WRITE_REG(hw, IXGBE_VTEITR(v_idx), itr_reg);
1073 }
1074
1075 #ifdef CONFIG_NET_RX_BUSY_POLL
1076 /* must be called with local_bh_disable()d */
1077 static int ixgbevf_busy_poll_recv(struct napi_struct *napi)
1078 {
1079         struct ixgbevf_q_vector *q_vector =
1080                         container_of(napi, struct ixgbevf_q_vector, napi);
1081         struct ixgbevf_adapter *adapter = q_vector->adapter;
1082         struct ixgbevf_ring  *ring;
1083         int found = 0;
1084
1085         if (test_bit(__IXGBEVF_DOWN, &adapter->state))
1086                 return LL_FLUSH_FAILED;
1087
1088         if (!ixgbevf_qv_lock_poll(q_vector))
1089                 return LL_FLUSH_BUSY;
1090
1091         ixgbevf_for_each_ring(ring, q_vector->rx) {
1092                 found = ixgbevf_clean_rx_irq(q_vector, ring, 4);
1093 #ifdef BP_EXTENDED_STATS
1094                 if (found)
1095                         ring->stats.cleaned += found;
1096                 else
1097                         ring->stats.misses++;
1098 #endif
1099                 if (found)
1100                         break;
1101         }
1102
1103         ixgbevf_qv_unlock_poll(q_vector);
1104
1105         return found;
1106 }
1107 #endif /* CONFIG_NET_RX_BUSY_POLL */
1108
1109 /**
1110  * ixgbevf_configure_msix - Configure MSI-X hardware
1111  * @adapter: board private structure
1112  *
1113  * ixgbevf_configure_msix sets up the hardware to properly generate MSI-X
1114  * interrupts.
1115  **/
1116 static void ixgbevf_configure_msix(struct ixgbevf_adapter *adapter)
1117 {
1118         struct ixgbevf_q_vector *q_vector;
1119         int q_vectors, v_idx;
1120
1121         q_vectors = adapter->num_msix_vectors - NON_Q_VECTORS;
1122         adapter->eims_enable_mask = 0;
1123
1124         /* Populate the IVAR table and set the ITR values to the
1125          * corresponding register.
1126          */
1127         for (v_idx = 0; v_idx < q_vectors; v_idx++) {
1128                 struct ixgbevf_ring *ring;
1129
1130                 q_vector = adapter->q_vector[v_idx];
1131
1132                 ixgbevf_for_each_ring(ring, q_vector->rx)
1133                         ixgbevf_set_ivar(adapter, 0, ring->reg_idx, v_idx);
1134
1135                 ixgbevf_for_each_ring(ring, q_vector->tx)
1136                         ixgbevf_set_ivar(adapter, 1, ring->reg_idx, v_idx);
1137
1138                 if (q_vector->tx.ring && !q_vector->rx.ring) {
1139                         /* Tx only vector */
1140                         if (adapter->tx_itr_setting == 1)
1141                                 q_vector->itr = IXGBE_10K_ITR;
1142                         else
1143                                 q_vector->itr = adapter->tx_itr_setting;
1144                 } else {
1145                         /* Rx or Rx/Tx vector */
1146                         if (adapter->rx_itr_setting == 1)
1147                                 q_vector->itr = IXGBE_20K_ITR;
1148                         else
1149                                 q_vector->itr = adapter->rx_itr_setting;
1150                 }
1151
1152                 /* add q_vector eims value to global eims_enable_mask */
1153                 adapter->eims_enable_mask |= 1 << v_idx;
1154
1155                 ixgbevf_write_eitr(q_vector);
1156         }
1157
1158         ixgbevf_set_ivar(adapter, -1, 1, v_idx);
1159         /* setup eims_other and add value to global eims_enable_mask */
1160         adapter->eims_other = 1 << v_idx;
1161         adapter->eims_enable_mask |= adapter->eims_other;
1162 }
1163
1164 enum latency_range {
1165         lowest_latency = 0,
1166         low_latency = 1,
1167         bulk_latency = 2,
1168         latency_invalid = 255
1169 };
1170
1171 /**
1172  * ixgbevf_update_itr - update the dynamic ITR value based on statistics
1173  * @q_vector: structure containing interrupt and ring information
1174  * @ring_container: structure containing ring performance data
1175  *
1176  * Stores a new ITR value based on packets and byte
1177  * counts during the last interrupt.  The advantage of per interrupt
1178  * computation is faster updates and more accurate ITR for the current
1179  * traffic pattern.  Constants in this function were computed
1180  * based on theoretical maximum wire speed and thresholds were set based
1181  * on testing data as well as attempting to minimize response time
1182  * while increasing bulk throughput.
1183  **/
1184 static void ixgbevf_update_itr(struct ixgbevf_q_vector *q_vector,
1185                                struct ixgbevf_ring_container *ring_container)
1186 {
1187         int bytes = ring_container->total_bytes;
1188         int packets = ring_container->total_packets;
1189         u32 timepassed_us;
1190         u64 bytes_perint;
1191         u8 itr_setting = ring_container->itr;
1192
1193         if (packets == 0)
1194                 return;
1195
1196         /* simple throttle rate management
1197          *    0-20MB/s lowest (100000 ints/s)
1198          *   20-100MB/s low   (20000 ints/s)
1199          *  100-1249MB/s bulk (8000 ints/s)
1200          */
1201         /* what was last interrupt timeslice? */
1202         timepassed_us = q_vector->itr >> 2;
1203         bytes_perint = bytes / timepassed_us; /* bytes/usec */
1204
1205         switch (itr_setting) {
1206         case lowest_latency:
1207                 if (bytes_perint > 10)
1208                         itr_setting = low_latency;
1209                 break;
1210         case low_latency:
1211                 if (bytes_perint > 20)
1212                         itr_setting = bulk_latency;
1213                 else if (bytes_perint <= 10)
1214                         itr_setting = lowest_latency;
1215                 break;
1216         case bulk_latency:
1217                 if (bytes_perint <= 20)
1218                         itr_setting = low_latency;
1219                 break;
1220         }
1221
1222         /* clear work counters since we have the values we need */
1223         ring_container->total_bytes = 0;
1224         ring_container->total_packets = 0;
1225
1226         /* write updated itr to ring container */
1227         ring_container->itr = itr_setting;
1228 }
1229
1230 static void ixgbevf_set_itr(struct ixgbevf_q_vector *q_vector)
1231 {
1232         u32 new_itr = q_vector->itr;
1233         u8 current_itr;
1234
1235         ixgbevf_update_itr(q_vector, &q_vector->tx);
1236         ixgbevf_update_itr(q_vector, &q_vector->rx);
1237
1238         current_itr = max(q_vector->rx.itr, q_vector->tx.itr);
1239
1240         switch (current_itr) {
1241         /* counts and packets in update_itr are dependent on these numbers */
1242         case lowest_latency:
1243                 new_itr = IXGBE_100K_ITR;
1244                 break;
1245         case low_latency:
1246                 new_itr = IXGBE_20K_ITR;
1247                 break;
1248         case bulk_latency:
1249         default:
1250                 new_itr = IXGBE_8K_ITR;
1251                 break;
1252         }
1253
1254         if (new_itr != q_vector->itr) {
1255                 /* do an exponential smoothing */
1256                 new_itr = (10 * new_itr * q_vector->itr) /
1257                           ((9 * new_itr) + q_vector->itr);
1258
1259                 /* save the algorithm value here */
1260                 q_vector->itr = new_itr;
1261
1262                 ixgbevf_write_eitr(q_vector);
1263         }
1264 }
1265
1266 static irqreturn_t ixgbevf_msix_other(int irq, void *data)
1267 {
1268         struct ixgbevf_adapter *adapter = data;
1269         struct ixgbe_hw *hw = &adapter->hw;
1270
1271         hw->mac.get_link_status = 1;
1272
1273         ixgbevf_service_event_schedule(adapter);
1274
1275         IXGBE_WRITE_REG(hw, IXGBE_VTEIMS, adapter->eims_other);
1276
1277         return IRQ_HANDLED;
1278 }
1279
1280 /**
1281  * ixgbevf_msix_clean_rings - single unshared vector rx clean (all queues)
1282  * @irq: unused
1283  * @data: pointer to our q_vector struct for this interrupt vector
1284  **/
1285 static irqreturn_t ixgbevf_msix_clean_rings(int irq, void *data)
1286 {
1287         struct ixgbevf_q_vector *q_vector = data;
1288
1289         /* EIAM disabled interrupts (on this vector) for us */
1290         if (q_vector->rx.ring || q_vector->tx.ring)
1291                 napi_schedule(&q_vector->napi);
1292
1293         return IRQ_HANDLED;
1294 }
1295
1296 static inline void map_vector_to_rxq(struct ixgbevf_adapter *a, int v_idx,
1297                                      int r_idx)
1298 {
1299         struct ixgbevf_q_vector *q_vector = a->q_vector[v_idx];
1300
1301         a->rx_ring[r_idx]->next = q_vector->rx.ring;
1302         q_vector->rx.ring = a->rx_ring[r_idx];
1303         q_vector->rx.count++;
1304 }
1305
1306 static inline void map_vector_to_txq(struct ixgbevf_adapter *a, int v_idx,
1307                                      int t_idx)
1308 {
1309         struct ixgbevf_q_vector *q_vector = a->q_vector[v_idx];
1310
1311         a->tx_ring[t_idx]->next = q_vector->tx.ring;
1312         q_vector->tx.ring = a->tx_ring[t_idx];
1313         q_vector->tx.count++;
1314 }
1315
1316 /**
1317  * ixgbevf_map_rings_to_vectors - Maps descriptor rings to vectors
1318  * @adapter: board private structure to initialize
1319  *
1320  * This function maps descriptor rings to the queue-specific vectors
1321  * we were allotted through the MSI-X enabling code.  Ideally, we'd have
1322  * one vector per ring/queue, but on a constrained vector budget, we
1323  * group the rings as "efficiently" as possible.  You would add new
1324  * mapping configurations in here.
1325  **/
1326 static int ixgbevf_map_rings_to_vectors(struct ixgbevf_adapter *adapter)
1327 {
1328         int q_vectors;
1329         int v_start = 0;
1330         int rxr_idx = 0, txr_idx = 0;
1331         int rxr_remaining = adapter->num_rx_queues;
1332         int txr_remaining = adapter->num_tx_queues;
1333         int i, j;
1334         int rqpv, tqpv;
1335         int err = 0;
1336
1337         q_vectors = adapter->num_msix_vectors - NON_Q_VECTORS;
1338
1339         /* The ideal configuration...
1340          * We have enough vectors to map one per queue.
1341          */
1342         if (q_vectors == adapter->num_rx_queues + adapter->num_tx_queues) {
1343                 for (; rxr_idx < rxr_remaining; v_start++, rxr_idx++)
1344                         map_vector_to_rxq(adapter, v_start, rxr_idx);
1345
1346                 for (; txr_idx < txr_remaining; v_start++, txr_idx++)
1347                         map_vector_to_txq(adapter, v_start, txr_idx);
1348                 goto out;
1349         }
1350
1351         /* If we don't have enough vectors for a 1-to-1
1352          * mapping, we'll have to group them so there are
1353          * multiple queues per vector.
1354          */
1355         /* Re-adjusting *qpv takes care of the remainder. */
1356         for (i = v_start; i < q_vectors; i++) {
1357                 rqpv = DIV_ROUND_UP(rxr_remaining, q_vectors - i);
1358                 for (j = 0; j < rqpv; j++) {
1359                         map_vector_to_rxq(adapter, i, rxr_idx);
1360                         rxr_idx++;
1361                         rxr_remaining--;
1362                 }
1363         }
1364         for (i = v_start; i < q_vectors; i++) {
1365                 tqpv = DIV_ROUND_UP(txr_remaining, q_vectors - i);
1366                 for (j = 0; j < tqpv; j++) {
1367                         map_vector_to_txq(adapter, i, txr_idx);
1368                         txr_idx++;
1369                         txr_remaining--;
1370                 }
1371         }
1372
1373 out:
1374         return err;
1375 }
1376
1377 /**
1378  * ixgbevf_request_msix_irqs - Initialize MSI-X interrupts
1379  * @adapter: board private structure
1380  *
1381  * ixgbevf_request_msix_irqs allocates MSI-X vectors and requests
1382  * interrupts from the kernel.
1383  **/
1384 static int ixgbevf_request_msix_irqs(struct ixgbevf_adapter *adapter)
1385 {
1386         struct net_device *netdev = adapter->netdev;
1387         int q_vectors = adapter->num_msix_vectors - NON_Q_VECTORS;
1388         int vector, err;
1389         int ri = 0, ti = 0;
1390
1391         for (vector = 0; vector < q_vectors; vector++) {
1392                 struct ixgbevf_q_vector *q_vector = adapter->q_vector[vector];
1393                 struct msix_entry *entry = &adapter->msix_entries[vector];
1394
1395                 if (q_vector->tx.ring && q_vector->rx.ring) {
1396                         snprintf(q_vector->name, sizeof(q_vector->name) - 1,
1397                                  "%s-%s-%d", netdev->name, "TxRx", ri++);
1398                         ti++;
1399                 } else if (q_vector->rx.ring) {
1400                         snprintf(q_vector->name, sizeof(q_vector->name) - 1,
1401                                  "%s-%s-%d", netdev->name, "rx", ri++);
1402                 } else if (q_vector->tx.ring) {
1403                         snprintf(q_vector->name, sizeof(q_vector->name) - 1,
1404                                  "%s-%s-%d", netdev->name, "tx", ti++);
1405                 } else {
1406                         /* skip this unused q_vector */
1407                         continue;
1408                 }
1409                 err = request_irq(entry->vector, &ixgbevf_msix_clean_rings, 0,
1410                                   q_vector->name, q_vector);
1411                 if (err) {
1412                         hw_dbg(&adapter->hw,
1413                                "request_irq failed for MSIX interrupt Error: %d\n",
1414                                err);
1415                         goto free_queue_irqs;
1416                 }
1417         }
1418
1419         err = request_irq(adapter->msix_entries[vector].vector,
1420                           &ixgbevf_msix_other, 0, netdev->name, adapter);
1421         if (err) {
1422                 hw_dbg(&adapter->hw, "request_irq for msix_other failed: %d\n",
1423                        err);
1424                 goto free_queue_irqs;
1425         }
1426
1427         return 0;
1428
1429 free_queue_irqs:
1430         while (vector) {
1431                 vector--;
1432                 free_irq(adapter->msix_entries[vector].vector,
1433                          adapter->q_vector[vector]);
1434         }
1435         /* This failure is non-recoverable - it indicates the system is
1436          * out of MSIX vector resources and the VF driver cannot run
1437          * without them.  Set the number of msix vectors to zero
1438          * indicating that not enough can be allocated.  The error
1439          * will be returned to the user indicating device open failed.
1440          * Any further attempts to force the driver to open will also
1441          * fail.  The only way to recover is to unload the driver and
1442          * reload it again.  If the system has recovered some MSIX
1443          * vectors then it may succeed.
1444          */
1445         adapter->num_msix_vectors = 0;
1446         return err;
1447 }
1448
1449 static inline void ixgbevf_reset_q_vectors(struct ixgbevf_adapter *adapter)
1450 {
1451         int i, q_vectors = adapter->num_msix_vectors - NON_Q_VECTORS;
1452
1453         for (i = 0; i < q_vectors; i++) {
1454                 struct ixgbevf_q_vector *q_vector = adapter->q_vector[i];
1455
1456                 q_vector->rx.ring = NULL;
1457                 q_vector->tx.ring = NULL;
1458                 q_vector->rx.count = 0;
1459                 q_vector->tx.count = 0;
1460         }
1461 }
1462
1463 /**
1464  * ixgbevf_request_irq - initialize interrupts
1465  * @adapter: board private structure
1466  *
1467  * Attempts to configure interrupts using the best available
1468  * capabilities of the hardware and kernel.
1469  **/
1470 static int ixgbevf_request_irq(struct ixgbevf_adapter *adapter)
1471 {
1472         int err = 0;
1473
1474         err = ixgbevf_request_msix_irqs(adapter);
1475
1476         if (err)
1477                 hw_dbg(&adapter->hw, "request_irq failed, Error %d\n", err);
1478
1479         return err;
1480 }
1481
1482 static void ixgbevf_free_irq(struct ixgbevf_adapter *adapter)
1483 {
1484         int i, q_vectors;
1485
1486         q_vectors = adapter->num_msix_vectors;
1487         i = q_vectors - 1;
1488
1489         free_irq(adapter->msix_entries[i].vector, adapter);
1490         i--;
1491
1492         for (; i >= 0; i--) {
1493                 /* free only the irqs that were actually requested */
1494                 if (!adapter->q_vector[i]->rx.ring &&
1495                     !adapter->q_vector[i]->tx.ring)
1496                         continue;
1497
1498                 free_irq(adapter->msix_entries[i].vector,
1499                          adapter->q_vector[i]);
1500         }
1501
1502         ixgbevf_reset_q_vectors(adapter);
1503 }
1504
1505 /**
1506  * ixgbevf_irq_disable - Mask off interrupt generation on the NIC
1507  * @adapter: board private structure
1508  **/
1509 static inline void ixgbevf_irq_disable(struct ixgbevf_adapter *adapter)
1510 {
1511         struct ixgbe_hw *hw = &adapter->hw;
1512         int i;
1513
1514         IXGBE_WRITE_REG(hw, IXGBE_VTEIAM, 0);
1515         IXGBE_WRITE_REG(hw, IXGBE_VTEIMC, ~0);
1516         IXGBE_WRITE_REG(hw, IXGBE_VTEIAC, 0);
1517
1518         IXGBE_WRITE_FLUSH(hw);
1519
1520         for (i = 0; i < adapter->num_msix_vectors; i++)
1521                 synchronize_irq(adapter->msix_entries[i].vector);
1522 }
1523
1524 /**
1525  * ixgbevf_irq_enable - Enable default interrupt generation settings
1526  * @adapter: board private structure
1527  **/
1528 static inline void ixgbevf_irq_enable(struct ixgbevf_adapter *adapter)
1529 {
1530         struct ixgbe_hw *hw = &adapter->hw;
1531
1532         IXGBE_WRITE_REG(hw, IXGBE_VTEIAM, adapter->eims_enable_mask);
1533         IXGBE_WRITE_REG(hw, IXGBE_VTEIAC, adapter->eims_enable_mask);
1534         IXGBE_WRITE_REG(hw, IXGBE_VTEIMS, adapter->eims_enable_mask);
1535 }
1536
1537 /**
1538  * ixgbevf_configure_tx_ring - Configure 82599 VF Tx ring after Reset
1539  * @adapter: board private structure
1540  * @ring: structure containing ring specific data
1541  *
1542  * Configure the Tx descriptor ring after a reset.
1543  **/
1544 static void ixgbevf_configure_tx_ring(struct ixgbevf_adapter *adapter,
1545                                       struct ixgbevf_ring *ring)
1546 {
1547         struct ixgbe_hw *hw = &adapter->hw;
1548         u64 tdba = ring->dma;
1549         int wait_loop = 10;
1550         u32 txdctl = IXGBE_TXDCTL_ENABLE;
1551         u8 reg_idx = ring->reg_idx;
1552
1553         /* disable queue to avoid issues while updating state */
1554         IXGBE_WRITE_REG(hw, IXGBE_VFTXDCTL(reg_idx), IXGBE_TXDCTL_SWFLSH);
1555         IXGBE_WRITE_FLUSH(hw);
1556
1557         IXGBE_WRITE_REG(hw, IXGBE_VFTDBAL(reg_idx), tdba & DMA_BIT_MASK(32));
1558         IXGBE_WRITE_REG(hw, IXGBE_VFTDBAH(reg_idx), tdba >> 32);
1559         IXGBE_WRITE_REG(hw, IXGBE_VFTDLEN(reg_idx),
1560                         ring->count * sizeof(union ixgbe_adv_tx_desc));
1561
1562         /* disable head writeback */
1563         IXGBE_WRITE_REG(hw, IXGBE_VFTDWBAH(reg_idx), 0);
1564         IXGBE_WRITE_REG(hw, IXGBE_VFTDWBAL(reg_idx), 0);
1565
1566         /* enable relaxed ordering */
1567         IXGBE_WRITE_REG(hw, IXGBE_VFDCA_TXCTRL(reg_idx),
1568                         (IXGBE_DCA_TXCTRL_DESC_RRO_EN |
1569                          IXGBE_DCA_TXCTRL_DATA_RRO_EN));
1570
1571         /* reset head and tail pointers */
1572         IXGBE_WRITE_REG(hw, IXGBE_VFTDH(reg_idx), 0);
1573         IXGBE_WRITE_REG(hw, IXGBE_VFTDT(reg_idx), 0);
1574         ring->tail = adapter->io_addr + IXGBE_VFTDT(reg_idx);
1575
1576         /* reset ntu and ntc to place SW in sync with hardwdare */
1577         ring->next_to_clean = 0;
1578         ring->next_to_use = 0;
1579
1580         /* In order to avoid issues WTHRESH + PTHRESH should always be equal
1581          * to or less than the number of on chip descriptors, which is
1582          * currently 40.
1583          */
1584         txdctl |= (8 << 16);    /* WTHRESH = 8 */
1585
1586         /* Setting PTHRESH to 32 both improves performance */
1587         txdctl |= (1 << 8) |    /* HTHRESH = 1 */
1588                   32;          /* PTHRESH = 32 */
1589
1590         clear_bit(__IXGBEVF_HANG_CHECK_ARMED, &ring->state);
1591
1592         IXGBE_WRITE_REG(hw, IXGBE_VFTXDCTL(reg_idx), txdctl);
1593
1594         /* poll to verify queue is enabled */
1595         do {
1596                 usleep_range(1000, 2000);
1597                 txdctl = IXGBE_READ_REG(hw, IXGBE_VFTXDCTL(reg_idx));
1598         }  while (--wait_loop && !(txdctl & IXGBE_TXDCTL_ENABLE));
1599         if (!wait_loop)
1600                 pr_err("Could not enable Tx Queue %d\n", reg_idx);
1601 }
1602
1603 /**
1604  * ixgbevf_configure_tx - Configure 82599 VF Transmit Unit after Reset
1605  * @adapter: board private structure
1606  *
1607  * Configure the Tx unit of the MAC after a reset.
1608  **/
1609 static void ixgbevf_configure_tx(struct ixgbevf_adapter *adapter)
1610 {
1611         u32 i;
1612
1613         /* Setup the HW Tx Head and Tail descriptor pointers */
1614         for (i = 0; i < adapter->num_tx_queues; i++)
1615                 ixgbevf_configure_tx_ring(adapter, adapter->tx_ring[i]);
1616 }
1617
1618 #define IXGBE_SRRCTL_BSIZEHDRSIZE_SHIFT 2
1619
1620 static void ixgbevf_configure_srrctl(struct ixgbevf_adapter *adapter, int index)
1621 {
1622         struct ixgbe_hw *hw = &adapter->hw;
1623         u32 srrctl;
1624
1625         srrctl = IXGBE_SRRCTL_DROP_EN;
1626
1627         srrctl |= IXGBEVF_RX_HDR_SIZE << IXGBE_SRRCTL_BSIZEHDRSIZE_SHIFT;
1628         srrctl |= IXGBEVF_RX_BUFSZ >> IXGBE_SRRCTL_BSIZEPKT_SHIFT;
1629         srrctl |= IXGBE_SRRCTL_DESCTYPE_ADV_ONEBUF;
1630
1631         IXGBE_WRITE_REG(hw, IXGBE_VFSRRCTL(index), srrctl);
1632 }
1633
1634 static void ixgbevf_setup_psrtype(struct ixgbevf_adapter *adapter)
1635 {
1636         struct ixgbe_hw *hw = &adapter->hw;
1637
1638         /* PSRTYPE must be initialized in 82599 */
1639         u32 psrtype = IXGBE_PSRTYPE_TCPHDR | IXGBE_PSRTYPE_UDPHDR |
1640                       IXGBE_PSRTYPE_IPV4HDR | IXGBE_PSRTYPE_IPV6HDR |
1641                       IXGBE_PSRTYPE_L2HDR;
1642
1643         if (adapter->num_rx_queues > 1)
1644                 psrtype |= 1 << 29;
1645
1646         IXGBE_WRITE_REG(hw, IXGBE_VFPSRTYPE, psrtype);
1647 }
1648
1649 #define IXGBEVF_MAX_RX_DESC_POLL 10
1650 static void ixgbevf_disable_rx_queue(struct ixgbevf_adapter *adapter,
1651                                      struct ixgbevf_ring *ring)
1652 {
1653         struct ixgbe_hw *hw = &adapter->hw;
1654         int wait_loop = IXGBEVF_MAX_RX_DESC_POLL;
1655         u32 rxdctl;
1656         u8 reg_idx = ring->reg_idx;
1657
1658         if (IXGBE_REMOVED(hw->hw_addr))
1659                 return;
1660         rxdctl = IXGBE_READ_REG(hw, IXGBE_VFRXDCTL(reg_idx));
1661         rxdctl &= ~IXGBE_RXDCTL_ENABLE;
1662
1663         /* write value back with RXDCTL.ENABLE bit cleared */
1664         IXGBE_WRITE_REG(hw, IXGBE_VFRXDCTL(reg_idx), rxdctl);
1665
1666         /* the hardware may take up to 100us to really disable the Rx queue */
1667         do {
1668                 udelay(10);
1669                 rxdctl = IXGBE_READ_REG(hw, IXGBE_VFRXDCTL(reg_idx));
1670         } while (--wait_loop && (rxdctl & IXGBE_RXDCTL_ENABLE));
1671
1672         if (!wait_loop)
1673                 pr_err("RXDCTL.ENABLE queue %d not cleared while polling\n",
1674                        reg_idx);
1675 }
1676
1677 static void ixgbevf_rx_desc_queue_enable(struct ixgbevf_adapter *adapter,
1678                                          struct ixgbevf_ring *ring)
1679 {
1680         struct ixgbe_hw *hw = &adapter->hw;
1681         int wait_loop = IXGBEVF_MAX_RX_DESC_POLL;
1682         u32 rxdctl;
1683         u8 reg_idx = ring->reg_idx;
1684
1685         if (IXGBE_REMOVED(hw->hw_addr))
1686                 return;
1687         do {
1688                 usleep_range(1000, 2000);
1689                 rxdctl = IXGBE_READ_REG(hw, IXGBE_VFRXDCTL(reg_idx));
1690         } while (--wait_loop && !(rxdctl & IXGBE_RXDCTL_ENABLE));
1691
1692         if (!wait_loop)
1693                 pr_err("RXDCTL.ENABLE queue %d not set while polling\n",
1694                        reg_idx);
1695 }
1696
1697 static void ixgbevf_setup_vfmrqc(struct ixgbevf_adapter *adapter)
1698 {
1699         struct ixgbe_hw *hw = &adapter->hw;
1700         u32 vfmrqc = 0, vfreta = 0;
1701         u16 rss_i = adapter->num_rx_queues;
1702         u8 i, j;
1703
1704         /* Fill out hash function seeds */
1705         netdev_rss_key_fill(adapter->rss_key, sizeof(adapter->rss_key));
1706         for (i = 0; i < IXGBEVF_VFRSSRK_REGS; i++)
1707                 IXGBE_WRITE_REG(hw, IXGBE_VFRSSRK(i), adapter->rss_key[i]);
1708
1709         for (i = 0, j = 0; i < IXGBEVF_X550_VFRETA_SIZE; i++, j++) {
1710                 if (j == rss_i)
1711                         j = 0;
1712
1713                 adapter->rss_indir_tbl[i] = j;
1714
1715                 vfreta |= j << (i & 0x3) * 8;
1716                 if ((i & 3) == 3) {
1717                         IXGBE_WRITE_REG(hw, IXGBE_VFRETA(i >> 2), vfreta);
1718                         vfreta = 0;
1719                 }
1720         }
1721
1722         /* Perform hash on these packet types */
1723         vfmrqc |= IXGBE_VFMRQC_RSS_FIELD_IPV4 |
1724                 IXGBE_VFMRQC_RSS_FIELD_IPV4_TCP |
1725                 IXGBE_VFMRQC_RSS_FIELD_IPV6 |
1726                 IXGBE_VFMRQC_RSS_FIELD_IPV6_TCP;
1727
1728         vfmrqc |= IXGBE_VFMRQC_RSSEN;
1729
1730         IXGBE_WRITE_REG(hw, IXGBE_VFMRQC, vfmrqc);
1731 }
1732
1733 static void ixgbevf_configure_rx_ring(struct ixgbevf_adapter *adapter,
1734                                       struct ixgbevf_ring *ring)
1735 {
1736         struct ixgbe_hw *hw = &adapter->hw;
1737         u64 rdba = ring->dma;
1738         u32 rxdctl;
1739         u8 reg_idx = ring->reg_idx;
1740
1741         /* disable queue to avoid issues while updating state */
1742         rxdctl = IXGBE_READ_REG(hw, IXGBE_VFRXDCTL(reg_idx));
1743         ixgbevf_disable_rx_queue(adapter, ring);
1744
1745         IXGBE_WRITE_REG(hw, IXGBE_VFRDBAL(reg_idx), rdba & DMA_BIT_MASK(32));
1746         IXGBE_WRITE_REG(hw, IXGBE_VFRDBAH(reg_idx), rdba >> 32);
1747         IXGBE_WRITE_REG(hw, IXGBE_VFRDLEN(reg_idx),
1748                         ring->count * sizeof(union ixgbe_adv_rx_desc));
1749
1750         /* enable relaxed ordering */
1751         IXGBE_WRITE_REG(hw, IXGBE_VFDCA_RXCTRL(reg_idx),
1752                         IXGBE_DCA_RXCTRL_DESC_RRO_EN);
1753
1754         /* reset head and tail pointers */
1755         IXGBE_WRITE_REG(hw, IXGBE_VFRDH(reg_idx), 0);
1756         IXGBE_WRITE_REG(hw, IXGBE_VFRDT(reg_idx), 0);
1757         ring->tail = adapter->io_addr + IXGBE_VFRDT(reg_idx);
1758
1759         /* reset ntu and ntc to place SW in sync with hardwdare */
1760         ring->next_to_clean = 0;
1761         ring->next_to_use = 0;
1762         ring->next_to_alloc = 0;
1763
1764         ixgbevf_configure_srrctl(adapter, reg_idx);
1765
1766         /* allow any size packet since we can handle overflow */
1767         rxdctl &= ~IXGBE_RXDCTL_RLPML_EN;
1768
1769         rxdctl |= IXGBE_RXDCTL_ENABLE | IXGBE_RXDCTL_VME;
1770         IXGBE_WRITE_REG(hw, IXGBE_VFRXDCTL(reg_idx), rxdctl);
1771
1772         ixgbevf_rx_desc_queue_enable(adapter, ring);
1773         ixgbevf_alloc_rx_buffers(ring, ixgbevf_desc_unused(ring));
1774 }
1775
1776 /**
1777  * ixgbevf_configure_rx - Configure 82599 VF Receive Unit after Reset
1778  * @adapter: board private structure
1779  *
1780  * Configure the Rx unit of the MAC after a reset.
1781  **/
1782 static void ixgbevf_configure_rx(struct ixgbevf_adapter *adapter)
1783 {
1784         int i;
1785         struct ixgbe_hw *hw = &adapter->hw;
1786         struct net_device *netdev = adapter->netdev;
1787
1788         ixgbevf_setup_psrtype(adapter);
1789         if (hw->mac.type >= ixgbe_mac_X550_vf)
1790                 ixgbevf_setup_vfmrqc(adapter);
1791
1792         /* notify the PF of our intent to use this size of frame */
1793         ixgbevf_rlpml_set_vf(hw, netdev->mtu + ETH_HLEN + ETH_FCS_LEN);
1794
1795         /* Setup the HW Rx Head and Tail Descriptor Pointers and
1796          * the Base and Length of the Rx Descriptor Ring
1797          */
1798         for (i = 0; i < adapter->num_rx_queues; i++)
1799                 ixgbevf_configure_rx_ring(adapter, adapter->rx_ring[i]);
1800 }
1801
1802 static int ixgbevf_vlan_rx_add_vid(struct net_device *netdev,
1803                                    __be16 proto, u16 vid)
1804 {
1805         struct ixgbevf_adapter *adapter = netdev_priv(netdev);
1806         struct ixgbe_hw *hw = &adapter->hw;
1807         int err;
1808
1809         spin_lock_bh(&adapter->mbx_lock);
1810
1811         /* add VID to filter table */
1812         err = hw->mac.ops.set_vfta(hw, vid, 0, true);
1813
1814         spin_unlock_bh(&adapter->mbx_lock);
1815
1816         /* translate error return types so error makes sense */
1817         if (err == IXGBE_ERR_MBX)
1818                 return -EIO;
1819
1820         if (err == IXGBE_ERR_INVALID_ARGUMENT)
1821                 return -EACCES;
1822
1823         set_bit(vid, adapter->active_vlans);
1824
1825         return err;
1826 }
1827
1828 static int ixgbevf_vlan_rx_kill_vid(struct net_device *netdev,
1829                                     __be16 proto, u16 vid)
1830 {
1831         struct ixgbevf_adapter *adapter = netdev_priv(netdev);
1832         struct ixgbe_hw *hw = &adapter->hw;
1833         int err = -EOPNOTSUPP;
1834
1835         spin_lock_bh(&adapter->mbx_lock);
1836
1837         /* remove VID from filter table */
1838         err = hw->mac.ops.set_vfta(hw, vid, 0, false);
1839
1840         spin_unlock_bh(&adapter->mbx_lock);
1841
1842         clear_bit(vid, adapter->active_vlans);
1843
1844         return err;
1845 }
1846
1847 static void ixgbevf_restore_vlan(struct ixgbevf_adapter *adapter)
1848 {
1849         u16 vid;
1850
1851         for_each_set_bit(vid, adapter->active_vlans, VLAN_N_VID)
1852                 ixgbevf_vlan_rx_add_vid(adapter->netdev,
1853                                         htons(ETH_P_8021Q), vid);
1854 }
1855
1856 static int ixgbevf_write_uc_addr_list(struct net_device *netdev)
1857 {
1858         struct ixgbevf_adapter *adapter = netdev_priv(netdev);
1859         struct ixgbe_hw *hw = &adapter->hw;
1860         int count = 0;
1861
1862         if ((netdev_uc_count(netdev)) > 10) {
1863                 pr_err("Too many unicast filters - No Space\n");
1864                 return -ENOSPC;
1865         }
1866
1867         if (!netdev_uc_empty(netdev)) {
1868                 struct netdev_hw_addr *ha;
1869
1870                 netdev_for_each_uc_addr(ha, netdev) {
1871                         hw->mac.ops.set_uc_addr(hw, ++count, ha->addr);
1872                         udelay(200);
1873                 }
1874         } else {
1875                 /* If the list is empty then send message to PF driver to
1876                  * clear all MAC VLANs on this VF.
1877                  */
1878                 hw->mac.ops.set_uc_addr(hw, 0, NULL);
1879         }
1880
1881         return count;
1882 }
1883
1884 /**
1885  * ixgbevf_set_rx_mode - Multicast and unicast set
1886  * @netdev: network interface device structure
1887  *
1888  * The set_rx_method entry point is called whenever the multicast address
1889  * list, unicast address list or the network interface flags are updated.
1890  * This routine is responsible for configuring the hardware for proper
1891  * multicast mode and configuring requested unicast filters.
1892  **/
1893 static void ixgbevf_set_rx_mode(struct net_device *netdev)
1894 {
1895         struct ixgbevf_adapter *adapter = netdev_priv(netdev);
1896         struct ixgbe_hw *hw = &adapter->hw;
1897         unsigned int flags = netdev->flags;
1898         int xcast_mode;
1899
1900         xcast_mode = (flags & IFF_ALLMULTI) ? IXGBEVF_XCAST_MODE_ALLMULTI :
1901                      (flags & (IFF_BROADCAST | IFF_MULTICAST)) ?
1902                      IXGBEVF_XCAST_MODE_MULTI : IXGBEVF_XCAST_MODE_NONE;
1903
1904         spin_lock_bh(&adapter->mbx_lock);
1905
1906         hw->mac.ops.update_xcast_mode(hw, netdev, xcast_mode);
1907
1908         /* reprogram multicast list */
1909         hw->mac.ops.update_mc_addr_list(hw, netdev);
1910
1911         ixgbevf_write_uc_addr_list(netdev);
1912
1913         spin_unlock_bh(&adapter->mbx_lock);
1914 }
1915
1916 static void ixgbevf_napi_enable_all(struct ixgbevf_adapter *adapter)
1917 {
1918         int q_idx;
1919         struct ixgbevf_q_vector *q_vector;
1920         int q_vectors = adapter->num_msix_vectors - NON_Q_VECTORS;
1921
1922         for (q_idx = 0; q_idx < q_vectors; q_idx++) {
1923                 q_vector = adapter->q_vector[q_idx];
1924 #ifdef CONFIG_NET_RX_BUSY_POLL
1925                 ixgbevf_qv_init_lock(adapter->q_vector[q_idx]);
1926 #endif
1927                 napi_enable(&q_vector->napi);
1928         }
1929 }
1930
1931 static void ixgbevf_napi_disable_all(struct ixgbevf_adapter *adapter)
1932 {
1933         int q_idx;
1934         struct ixgbevf_q_vector *q_vector;
1935         int q_vectors = adapter->num_msix_vectors - NON_Q_VECTORS;
1936
1937         for (q_idx = 0; q_idx < q_vectors; q_idx++) {
1938                 q_vector = adapter->q_vector[q_idx];
1939                 napi_disable(&q_vector->napi);
1940 #ifdef CONFIG_NET_RX_BUSY_POLL
1941                 while (!ixgbevf_qv_disable(adapter->q_vector[q_idx])) {
1942                         pr_info("QV %d locked\n", q_idx);
1943                         usleep_range(1000, 20000);
1944                 }
1945 #endif /* CONFIG_NET_RX_BUSY_POLL */
1946         }
1947 }
1948
1949 static int ixgbevf_configure_dcb(struct ixgbevf_adapter *adapter)
1950 {
1951         struct ixgbe_hw *hw = &adapter->hw;
1952         unsigned int def_q = 0;
1953         unsigned int num_tcs = 0;
1954         unsigned int num_rx_queues = adapter->num_rx_queues;
1955         unsigned int num_tx_queues = adapter->num_tx_queues;
1956         int err;
1957
1958         spin_lock_bh(&adapter->mbx_lock);
1959
1960         /* fetch queue configuration from the PF */
1961         err = ixgbevf_get_queues(hw, &num_tcs, &def_q);
1962
1963         spin_unlock_bh(&adapter->mbx_lock);
1964
1965         if (err)
1966                 return err;
1967
1968         if (num_tcs > 1) {
1969                 /* we need only one Tx queue */
1970                 num_tx_queues = 1;
1971
1972                 /* update default Tx ring register index */
1973                 adapter->tx_ring[0]->reg_idx = def_q;
1974
1975                 /* we need as many queues as traffic classes */
1976                 num_rx_queues = num_tcs;
1977         }
1978
1979         /* if we have a bad config abort request queue reset */
1980         if ((adapter->num_rx_queues != num_rx_queues) ||
1981             (adapter->num_tx_queues != num_tx_queues)) {
1982                 /* force mailbox timeout to prevent further messages */
1983                 hw->mbx.timeout = 0;
1984
1985                 /* wait for watchdog to come around and bail us out */
1986                 adapter->flags |= IXGBEVF_FLAG_QUEUE_RESET_REQUESTED;
1987         }
1988
1989         return 0;
1990 }
1991
1992 static void ixgbevf_configure(struct ixgbevf_adapter *adapter)
1993 {
1994         ixgbevf_configure_dcb(adapter);
1995
1996         ixgbevf_set_rx_mode(adapter->netdev);
1997
1998         ixgbevf_restore_vlan(adapter);
1999
2000         ixgbevf_configure_tx(adapter);
2001         ixgbevf_configure_rx(adapter);
2002 }
2003
2004 static void ixgbevf_save_reset_stats(struct ixgbevf_adapter *adapter)
2005 {
2006         /* Only save pre-reset stats if there are some */
2007         if (adapter->stats.vfgprc || adapter->stats.vfgptc) {
2008                 adapter->stats.saved_reset_vfgprc += adapter->stats.vfgprc -
2009                         adapter->stats.base_vfgprc;
2010                 adapter->stats.saved_reset_vfgptc += adapter->stats.vfgptc -
2011                         adapter->stats.base_vfgptc;
2012                 adapter->stats.saved_reset_vfgorc += adapter->stats.vfgorc -
2013                         adapter->stats.base_vfgorc;
2014                 adapter->stats.saved_reset_vfgotc += adapter->stats.vfgotc -
2015                         adapter->stats.base_vfgotc;
2016                 adapter->stats.saved_reset_vfmprc += adapter->stats.vfmprc -
2017                         adapter->stats.base_vfmprc;
2018         }
2019 }
2020
2021 static void ixgbevf_init_last_counter_stats(struct ixgbevf_adapter *adapter)
2022 {
2023         struct ixgbe_hw *hw = &adapter->hw;
2024
2025         adapter->stats.last_vfgprc = IXGBE_READ_REG(hw, IXGBE_VFGPRC);
2026         adapter->stats.last_vfgorc = IXGBE_READ_REG(hw, IXGBE_VFGORC_LSB);
2027         adapter->stats.last_vfgorc |=
2028                 (((u64)(IXGBE_READ_REG(hw, IXGBE_VFGORC_MSB))) << 32);
2029         adapter->stats.last_vfgptc = IXGBE_READ_REG(hw, IXGBE_VFGPTC);
2030         adapter->stats.last_vfgotc = IXGBE_READ_REG(hw, IXGBE_VFGOTC_LSB);
2031         adapter->stats.last_vfgotc |=
2032                 (((u64)(IXGBE_READ_REG(hw, IXGBE_VFGOTC_MSB))) << 32);
2033         adapter->stats.last_vfmprc = IXGBE_READ_REG(hw, IXGBE_VFMPRC);
2034
2035         adapter->stats.base_vfgprc = adapter->stats.last_vfgprc;
2036         adapter->stats.base_vfgorc = adapter->stats.last_vfgorc;
2037         adapter->stats.base_vfgptc = adapter->stats.last_vfgptc;
2038         adapter->stats.base_vfgotc = adapter->stats.last_vfgotc;
2039         adapter->stats.base_vfmprc = adapter->stats.last_vfmprc;
2040 }
2041
2042 static void ixgbevf_negotiate_api(struct ixgbevf_adapter *adapter)
2043 {
2044         struct ixgbe_hw *hw = &adapter->hw;
2045         int api[] = { ixgbe_mbox_api_12,
2046                       ixgbe_mbox_api_11,
2047                       ixgbe_mbox_api_10,
2048                       ixgbe_mbox_api_unknown };
2049         int err = 0, idx = 0;
2050
2051         spin_lock_bh(&adapter->mbx_lock);
2052
2053         while (api[idx] != ixgbe_mbox_api_unknown) {
2054                 err = ixgbevf_negotiate_api_version(hw, api[idx]);
2055                 if (!err)
2056                         break;
2057                 idx++;
2058         }
2059
2060         spin_unlock_bh(&adapter->mbx_lock);
2061 }
2062
2063 static void ixgbevf_up_complete(struct ixgbevf_adapter *adapter)
2064 {
2065         struct net_device *netdev = adapter->netdev;
2066         struct ixgbe_hw *hw = &adapter->hw;
2067
2068         ixgbevf_configure_msix(adapter);
2069
2070         spin_lock_bh(&adapter->mbx_lock);
2071
2072         if (is_valid_ether_addr(hw->mac.addr))
2073                 hw->mac.ops.set_rar(hw, 0, hw->mac.addr, 0);
2074         else
2075                 hw->mac.ops.set_rar(hw, 0, hw->mac.perm_addr, 0);
2076
2077         spin_unlock_bh(&adapter->mbx_lock);
2078
2079         smp_mb__before_atomic();
2080         clear_bit(__IXGBEVF_DOWN, &adapter->state);
2081         ixgbevf_napi_enable_all(adapter);
2082
2083         /* clear any pending interrupts, may auto mask */
2084         IXGBE_READ_REG(hw, IXGBE_VTEICR);
2085         ixgbevf_irq_enable(adapter);
2086
2087         /* enable transmits */
2088         netif_tx_start_all_queues(netdev);
2089
2090         ixgbevf_save_reset_stats(adapter);
2091         ixgbevf_init_last_counter_stats(adapter);
2092
2093         hw->mac.get_link_status = 1;
2094         mod_timer(&adapter->service_timer, jiffies);
2095 }
2096
2097 void ixgbevf_up(struct ixgbevf_adapter *adapter)
2098 {
2099         ixgbevf_configure(adapter);
2100
2101         ixgbevf_up_complete(adapter);
2102 }
2103
2104 /**
2105  * ixgbevf_clean_rx_ring - Free Rx Buffers per Queue
2106  * @rx_ring: ring to free buffers from
2107  **/
2108 static void ixgbevf_clean_rx_ring(struct ixgbevf_ring *rx_ring)
2109 {
2110         struct device *dev = rx_ring->dev;
2111         unsigned long size;
2112         unsigned int i;
2113
2114         /* Free Rx ring sk_buff */
2115         if (rx_ring->skb) {
2116                 dev_kfree_skb(rx_ring->skb);
2117                 rx_ring->skb = NULL;
2118         }
2119
2120         /* ring already cleared, nothing to do */
2121         if (!rx_ring->rx_buffer_info)
2122                 return;
2123
2124         /* Free all the Rx ring pages */
2125         for (i = 0; i < rx_ring->count; i++) {
2126                 struct ixgbevf_rx_buffer *rx_buffer;
2127
2128                 rx_buffer = &rx_ring->rx_buffer_info[i];
2129                 if (rx_buffer->dma)
2130                         dma_unmap_page(dev, rx_buffer->dma,
2131                                        PAGE_SIZE, DMA_FROM_DEVICE);
2132                 rx_buffer->dma = 0;
2133                 if (rx_buffer->page)
2134                         __free_page(rx_buffer->page);
2135                 rx_buffer->page = NULL;
2136         }
2137
2138         size = sizeof(struct ixgbevf_rx_buffer) * rx_ring->count;
2139         memset(rx_ring->rx_buffer_info, 0, size);
2140
2141         /* Zero out the descriptor ring */
2142         memset(rx_ring->desc, 0, rx_ring->size);
2143 }
2144
2145 /**
2146  * ixgbevf_clean_tx_ring - Free Tx Buffers
2147  * @tx_ring: ring to be cleaned
2148  **/
2149 static void ixgbevf_clean_tx_ring(struct ixgbevf_ring *tx_ring)
2150 {
2151         struct ixgbevf_tx_buffer *tx_buffer_info;
2152         unsigned long size;
2153         unsigned int i;
2154
2155         if (!tx_ring->tx_buffer_info)
2156                 return;
2157
2158         /* Free all the Tx ring sk_buffs */
2159         for (i = 0; i < tx_ring->count; i++) {
2160                 tx_buffer_info = &tx_ring->tx_buffer_info[i];
2161                 ixgbevf_unmap_and_free_tx_resource(tx_ring, tx_buffer_info);
2162         }
2163
2164         size = sizeof(struct ixgbevf_tx_buffer) * tx_ring->count;
2165         memset(tx_ring->tx_buffer_info, 0, size);
2166
2167         memset(tx_ring->desc, 0, tx_ring->size);
2168 }
2169
2170 /**
2171  * ixgbevf_clean_all_rx_rings - Free Rx Buffers for all queues
2172  * @adapter: board private structure
2173  **/
2174 static void ixgbevf_clean_all_rx_rings(struct ixgbevf_adapter *adapter)
2175 {
2176         int i;
2177
2178         for (i = 0; i < adapter->num_rx_queues; i++)
2179                 ixgbevf_clean_rx_ring(adapter->rx_ring[i]);
2180 }
2181
2182 /**
2183  * ixgbevf_clean_all_tx_rings - Free Tx Buffers for all queues
2184  * @adapter: board private structure
2185  **/
2186 static void ixgbevf_clean_all_tx_rings(struct ixgbevf_adapter *adapter)
2187 {
2188         int i;
2189
2190         for (i = 0; i < adapter->num_tx_queues; i++)
2191                 ixgbevf_clean_tx_ring(adapter->tx_ring[i]);
2192 }
2193
2194 void ixgbevf_down(struct ixgbevf_adapter *adapter)
2195 {
2196         struct net_device *netdev = adapter->netdev;
2197         struct ixgbe_hw *hw = &adapter->hw;
2198         int i;
2199
2200         /* signal that we are down to the interrupt handler */
2201         if (test_and_set_bit(__IXGBEVF_DOWN, &adapter->state))
2202                 return; /* do nothing if already down */
2203
2204         /* disable all enabled Rx queues */
2205         for (i = 0; i < adapter->num_rx_queues; i++)
2206                 ixgbevf_disable_rx_queue(adapter, adapter->rx_ring[i]);
2207
2208         usleep_range(10000, 20000);
2209
2210         netif_tx_stop_all_queues(netdev);
2211
2212         /* call carrier off first to avoid false dev_watchdog timeouts */
2213         netif_carrier_off(netdev);
2214         netif_tx_disable(netdev);
2215
2216         ixgbevf_irq_disable(adapter);
2217
2218         ixgbevf_napi_disable_all(adapter);
2219
2220         del_timer_sync(&adapter->service_timer);
2221
2222         /* disable transmits in the hardware now that interrupts are off */
2223         for (i = 0; i < adapter->num_tx_queues; i++) {
2224                 u8 reg_idx = adapter->tx_ring[i]->reg_idx;
2225
2226                 IXGBE_WRITE_REG(hw, IXGBE_VFTXDCTL(reg_idx),
2227                                 IXGBE_TXDCTL_SWFLSH);
2228         }
2229
2230         if (!pci_channel_offline(adapter->pdev))
2231                 ixgbevf_reset(adapter);
2232
2233         ixgbevf_clean_all_tx_rings(adapter);
2234         ixgbevf_clean_all_rx_rings(adapter);
2235 }
2236
2237 void ixgbevf_reinit_locked(struct ixgbevf_adapter *adapter)
2238 {
2239         WARN_ON(in_interrupt());
2240
2241         while (test_and_set_bit(__IXGBEVF_RESETTING, &adapter->state))
2242                 msleep(1);
2243
2244         ixgbevf_down(adapter);
2245         ixgbevf_up(adapter);
2246
2247         clear_bit(__IXGBEVF_RESETTING, &adapter->state);
2248 }
2249
2250 void ixgbevf_reset(struct ixgbevf_adapter *adapter)
2251 {
2252         struct ixgbe_hw *hw = &adapter->hw;
2253         struct net_device *netdev = adapter->netdev;
2254
2255         if (hw->mac.ops.reset_hw(hw)) {
2256                 hw_dbg(hw, "PF still resetting\n");
2257         } else {
2258                 hw->mac.ops.init_hw(hw);
2259                 ixgbevf_negotiate_api(adapter);
2260         }
2261
2262         if (is_valid_ether_addr(adapter->hw.mac.addr)) {
2263                 memcpy(netdev->dev_addr, adapter->hw.mac.addr,
2264                        netdev->addr_len);
2265                 memcpy(netdev->perm_addr, adapter->hw.mac.addr,
2266                        netdev->addr_len);
2267         }
2268
2269         adapter->last_reset = jiffies;
2270 }
2271
2272 static int ixgbevf_acquire_msix_vectors(struct ixgbevf_adapter *adapter,
2273                                         int vectors)
2274 {
2275         int vector_threshold;
2276
2277         /* We'll want at least 2 (vector_threshold):
2278          * 1) TxQ[0] + RxQ[0] handler
2279          * 2) Other (Link Status Change, etc.)
2280          */
2281         vector_threshold = MIN_MSIX_COUNT;
2282
2283         /* The more we get, the more we will assign to Tx/Rx Cleanup
2284          * for the separate queues...where Rx Cleanup >= Tx Cleanup.
2285          * Right now, we simply care about how many we'll get; we'll
2286          * set them up later while requesting irq's.
2287          */
2288         vectors = pci_enable_msix_range(adapter->pdev, adapter->msix_entries,
2289                                         vector_threshold, vectors);
2290
2291         if (vectors < 0) {
2292                 dev_err(&adapter->pdev->dev,
2293                         "Unable to allocate MSI-X interrupts\n");
2294                 kfree(adapter->msix_entries);
2295                 adapter->msix_entries = NULL;
2296                 return vectors;
2297         }
2298
2299         /* Adjust for only the vectors we'll use, which is minimum
2300          * of max_msix_q_vectors + NON_Q_VECTORS, or the number of
2301          * vectors we were allocated.
2302          */
2303         adapter->num_msix_vectors = vectors;
2304
2305         return 0;
2306 }
2307
2308 /**
2309  * ixgbevf_set_num_queues - Allocate queues for device, feature dependent
2310  * @adapter: board private structure to initialize
2311  *
2312  * This is the top level queue allocation routine.  The order here is very
2313  * important, starting with the "most" number of features turned on at once,
2314  * and ending with the smallest set of features.  This way large combinations
2315  * can be allocated if they're turned on, and smaller combinations are the
2316  * fallthrough conditions.
2317  *
2318  **/
2319 static void ixgbevf_set_num_queues(struct ixgbevf_adapter *adapter)
2320 {
2321         struct ixgbe_hw *hw = &adapter->hw;
2322         unsigned int def_q = 0;
2323         unsigned int num_tcs = 0;
2324         int err;
2325
2326         /* Start with base case */
2327         adapter->num_rx_queues = 1;
2328         adapter->num_tx_queues = 1;
2329
2330         spin_lock_bh(&adapter->mbx_lock);
2331
2332         /* fetch queue configuration from the PF */
2333         err = ixgbevf_get_queues(hw, &num_tcs, &def_q);
2334
2335         spin_unlock_bh(&adapter->mbx_lock);
2336
2337         if (err)
2338                 return;
2339
2340         /* we need as many queues as traffic classes */
2341         if (num_tcs > 1) {
2342                 adapter->num_rx_queues = num_tcs;
2343         } else {
2344                 u16 rss = min_t(u16, num_online_cpus(), IXGBEVF_MAX_RSS_QUEUES);
2345
2346                 switch (hw->api_version) {
2347                 case ixgbe_mbox_api_11:
2348                 case ixgbe_mbox_api_12:
2349                         adapter->num_rx_queues = rss;
2350                         adapter->num_tx_queues = rss;
2351                 default:
2352                         break;
2353                 }
2354         }
2355 }
2356
2357 /**
2358  * ixgbevf_alloc_queues - Allocate memory for all rings
2359  * @adapter: board private structure to initialize
2360  *
2361  * We allocate one ring per queue at run-time since we don't know the
2362  * number of queues at compile-time.  The polling_netdev array is
2363  * intended for Multiqueue, but should work fine with a single queue.
2364  **/
2365 static int ixgbevf_alloc_queues(struct ixgbevf_adapter *adapter)
2366 {
2367         struct ixgbevf_ring *ring;
2368         int rx = 0, tx = 0;
2369
2370         for (; tx < adapter->num_tx_queues; tx++) {
2371                 ring = kzalloc(sizeof(*ring), GFP_KERNEL);
2372                 if (!ring)
2373                         goto err_allocation;
2374
2375                 ring->dev = &adapter->pdev->dev;
2376                 ring->netdev = adapter->netdev;
2377                 ring->count = adapter->tx_ring_count;
2378                 ring->queue_index = tx;
2379                 ring->reg_idx = tx;
2380
2381                 adapter->tx_ring[tx] = ring;
2382         }
2383
2384         for (; rx < adapter->num_rx_queues; rx++) {
2385                 ring = kzalloc(sizeof(*ring), GFP_KERNEL);
2386                 if (!ring)
2387                         goto err_allocation;
2388
2389                 ring->dev = &adapter->pdev->dev;
2390                 ring->netdev = adapter->netdev;
2391
2392                 ring->count = adapter->rx_ring_count;
2393                 ring->queue_index = rx;
2394                 ring->reg_idx = rx;
2395
2396                 adapter->rx_ring[rx] = ring;
2397         }
2398
2399         return 0;
2400
2401 err_allocation:
2402         while (tx) {
2403                 kfree(adapter->tx_ring[--tx]);
2404                 adapter->tx_ring[tx] = NULL;
2405         }
2406
2407         while (rx) {
2408                 kfree(adapter->rx_ring[--rx]);
2409                 adapter->rx_ring[rx] = NULL;
2410         }
2411         return -ENOMEM;
2412 }
2413
2414 /**
2415  * ixgbevf_set_interrupt_capability - set MSI-X or FAIL if not supported
2416  * @adapter: board private structure to initialize
2417  *
2418  * Attempt to configure the interrupts using the best available
2419  * capabilities of the hardware and the kernel.
2420  **/
2421 static int ixgbevf_set_interrupt_capability(struct ixgbevf_adapter *adapter)
2422 {
2423         struct net_device *netdev = adapter->netdev;
2424         int err = 0;
2425         int vector, v_budget;
2426
2427         /* It's easy to be greedy for MSI-X vectors, but it really
2428          * doesn't do us much good if we have a lot more vectors
2429          * than CPU's.  So let's be conservative and only ask for
2430          * (roughly) the same number of vectors as there are CPU's.
2431          * The default is to use pairs of vectors.
2432          */
2433         v_budget = max(adapter->num_rx_queues, adapter->num_tx_queues);
2434         v_budget = min_t(int, v_budget, num_online_cpus());
2435         v_budget += NON_Q_VECTORS;
2436
2437         /* A failure in MSI-X entry allocation isn't fatal, but it does
2438          * mean we disable MSI-X capabilities of the adapter.
2439          */
2440         adapter->msix_entries = kcalloc(v_budget,
2441                                         sizeof(struct msix_entry), GFP_KERNEL);
2442         if (!adapter->msix_entries) {
2443                 err = -ENOMEM;
2444                 goto out;
2445         }
2446
2447         for (vector = 0; vector < v_budget; vector++)
2448                 adapter->msix_entries[vector].entry = vector;
2449
2450         err = ixgbevf_acquire_msix_vectors(adapter, v_budget);
2451         if (err)
2452                 goto out;
2453
2454         err = netif_set_real_num_tx_queues(netdev, adapter->num_tx_queues);
2455         if (err)
2456                 goto out;
2457
2458         err = netif_set_real_num_rx_queues(netdev, adapter->num_rx_queues);
2459
2460 out:
2461         return err;
2462 }
2463
2464 /**
2465  * ixgbevf_alloc_q_vectors - Allocate memory for interrupt vectors
2466  * @adapter: board private structure to initialize
2467  *
2468  * We allocate one q_vector per queue interrupt.  If allocation fails we
2469  * return -ENOMEM.
2470  **/
2471 static int ixgbevf_alloc_q_vectors(struct ixgbevf_adapter *adapter)
2472 {
2473         int q_idx, num_q_vectors;
2474         struct ixgbevf_q_vector *q_vector;
2475
2476         num_q_vectors = adapter->num_msix_vectors - NON_Q_VECTORS;
2477
2478         for (q_idx = 0; q_idx < num_q_vectors; q_idx++) {
2479                 q_vector = kzalloc(sizeof(struct ixgbevf_q_vector), GFP_KERNEL);
2480                 if (!q_vector)
2481                         goto err_out;
2482                 q_vector->adapter = adapter;
2483                 q_vector->v_idx = q_idx;
2484                 netif_napi_add(adapter->netdev, &q_vector->napi,
2485                                ixgbevf_poll, 64);
2486 #ifdef CONFIG_NET_RX_BUSY_POLL
2487                 napi_hash_add(&q_vector->napi);
2488 #endif
2489                 adapter->q_vector[q_idx] = q_vector;
2490         }
2491
2492         return 0;
2493
2494 err_out:
2495         while (q_idx) {
2496                 q_idx--;
2497                 q_vector = adapter->q_vector[q_idx];
2498 #ifdef CONFIG_NET_RX_BUSY_POLL
2499                 napi_hash_del(&q_vector->napi);
2500 #endif
2501                 netif_napi_del(&q_vector->napi);
2502                 kfree(q_vector);
2503                 adapter->q_vector[q_idx] = NULL;
2504         }
2505         return -ENOMEM;
2506 }
2507
2508 /**
2509  * ixgbevf_free_q_vectors - Free memory allocated for interrupt vectors
2510  * @adapter: board private structure to initialize
2511  *
2512  * This function frees the memory allocated to the q_vectors.  In addition if
2513  * NAPI is enabled it will delete any references to the NAPI struct prior
2514  * to freeing the q_vector.
2515  **/
2516 static void ixgbevf_free_q_vectors(struct ixgbevf_adapter *adapter)
2517 {
2518         int q_idx, num_q_vectors = adapter->num_msix_vectors - NON_Q_VECTORS;
2519
2520         for (q_idx = 0; q_idx < num_q_vectors; q_idx++) {
2521                 struct ixgbevf_q_vector *q_vector = adapter->q_vector[q_idx];
2522
2523                 adapter->q_vector[q_idx] = NULL;
2524 #ifdef CONFIG_NET_RX_BUSY_POLL
2525                 napi_hash_del(&q_vector->napi);
2526 #endif
2527                 netif_napi_del(&q_vector->napi);
2528                 kfree(q_vector);
2529         }
2530 }
2531
2532 /**
2533  * ixgbevf_reset_interrupt_capability - Reset MSIX setup
2534  * @adapter: board private structure
2535  *
2536  **/
2537 static void ixgbevf_reset_interrupt_capability(struct ixgbevf_adapter *adapter)
2538 {
2539         pci_disable_msix(adapter->pdev);
2540         kfree(adapter->msix_entries);
2541         adapter->msix_entries = NULL;
2542 }
2543
2544 /**
2545  * ixgbevf_init_interrupt_scheme - Determine if MSIX is supported and init
2546  * @adapter: board private structure to initialize
2547  *
2548  **/
2549 static int ixgbevf_init_interrupt_scheme(struct ixgbevf_adapter *adapter)
2550 {
2551         int err;
2552
2553         /* Number of supported queues */
2554         ixgbevf_set_num_queues(adapter);
2555
2556         err = ixgbevf_set_interrupt_capability(adapter);
2557         if (err) {
2558                 hw_dbg(&adapter->hw,
2559                        "Unable to setup interrupt capabilities\n");
2560                 goto err_set_interrupt;
2561         }
2562
2563         err = ixgbevf_alloc_q_vectors(adapter);
2564         if (err) {
2565                 hw_dbg(&adapter->hw, "Unable to allocate memory for queue vectors\n");
2566                 goto err_alloc_q_vectors;
2567         }
2568
2569         err = ixgbevf_alloc_queues(adapter);
2570         if (err) {
2571                 pr_err("Unable to allocate memory for queues\n");
2572                 goto err_alloc_queues;
2573         }
2574
2575         hw_dbg(&adapter->hw, "Multiqueue %s: Rx Queue count = %u, Tx Queue count = %u\n",
2576                (adapter->num_rx_queues > 1) ? "Enabled" :
2577                "Disabled", adapter->num_rx_queues, adapter->num_tx_queues);
2578
2579         set_bit(__IXGBEVF_DOWN, &adapter->state);
2580
2581         return 0;
2582 err_alloc_queues:
2583         ixgbevf_free_q_vectors(adapter);
2584 err_alloc_q_vectors:
2585         ixgbevf_reset_interrupt_capability(adapter);
2586 err_set_interrupt:
2587         return err;
2588 }
2589
2590 /**
2591  * ixgbevf_clear_interrupt_scheme - Clear the current interrupt scheme settings
2592  * @adapter: board private structure to clear interrupt scheme on
2593  *
2594  * We go through and clear interrupt specific resources and reset the structure
2595  * to pre-load conditions
2596  **/
2597 static void ixgbevf_clear_interrupt_scheme(struct ixgbevf_adapter *adapter)
2598 {
2599         int i;
2600
2601         for (i = 0; i < adapter->num_tx_queues; i++) {
2602                 kfree(adapter->tx_ring[i]);
2603                 adapter->tx_ring[i] = NULL;
2604         }
2605         for (i = 0; i < adapter->num_rx_queues; i++) {
2606                 kfree(adapter->rx_ring[i]);
2607                 adapter->rx_ring[i] = NULL;
2608         }
2609
2610         adapter->num_tx_queues = 0;
2611         adapter->num_rx_queues = 0;
2612
2613         ixgbevf_free_q_vectors(adapter);
2614         ixgbevf_reset_interrupt_capability(adapter);
2615 }
2616
2617 /**
2618  * ixgbevf_sw_init - Initialize general software structures
2619  * @adapter: board private structure to initialize
2620  *
2621  * ixgbevf_sw_init initializes the Adapter private data structure.
2622  * Fields are initialized based on PCI device information and
2623  * OS network device settings (MTU size).
2624  **/
2625 static int ixgbevf_sw_init(struct ixgbevf_adapter *adapter)
2626 {
2627         struct ixgbe_hw *hw = &adapter->hw;
2628         struct pci_dev *pdev = adapter->pdev;
2629         struct net_device *netdev = adapter->netdev;
2630         int err;
2631
2632         /* PCI config space info */
2633         hw->vendor_id = pdev->vendor;
2634         hw->device_id = pdev->device;
2635         hw->revision_id = pdev->revision;
2636         hw->subsystem_vendor_id = pdev->subsystem_vendor;
2637         hw->subsystem_device_id = pdev->subsystem_device;
2638
2639         hw->mbx.ops.init_params(hw);
2640
2641         /* assume legacy case in which PF would only give VF 2 queues */
2642         hw->mac.max_tx_queues = 2;
2643         hw->mac.max_rx_queues = 2;
2644
2645         /* lock to protect mailbox accesses */
2646         spin_lock_init(&adapter->mbx_lock);
2647
2648         err = hw->mac.ops.reset_hw(hw);
2649         if (err) {
2650                 dev_info(&pdev->dev,
2651                          "PF still in reset state.  Is the PF interface up?\n");
2652         } else {
2653                 err = hw->mac.ops.init_hw(hw);
2654                 if (err) {
2655                         pr_err("init_shared_code failed: %d\n", err);
2656                         goto out;
2657                 }
2658                 ixgbevf_negotiate_api(adapter);
2659                 err = hw->mac.ops.get_mac_addr(hw, hw->mac.addr);
2660                 if (err)
2661                         dev_info(&pdev->dev, "Error reading MAC address\n");
2662                 else if (is_zero_ether_addr(adapter->hw.mac.addr))
2663                         dev_info(&pdev->dev,
2664                                  "MAC address not assigned by administrator.\n");
2665                 memcpy(netdev->dev_addr, hw->mac.addr, netdev->addr_len);
2666         }
2667
2668         if (!is_valid_ether_addr(netdev->dev_addr)) {
2669                 dev_info(&pdev->dev, "Assigning random MAC address\n");
2670                 eth_hw_addr_random(netdev);
2671                 memcpy(hw->mac.addr, netdev->dev_addr, netdev->addr_len);
2672         }
2673
2674         /* Enable dynamic interrupt throttling rates */
2675         adapter->rx_itr_setting = 1;
2676         adapter->tx_itr_setting = 1;
2677
2678         /* set default ring sizes */
2679         adapter->tx_ring_count = IXGBEVF_DEFAULT_TXD;
2680         adapter->rx_ring_count = IXGBEVF_DEFAULT_RXD;
2681
2682         set_bit(__IXGBEVF_DOWN, &adapter->state);
2683         return 0;
2684
2685 out:
2686         return err;
2687 }
2688
2689 #define UPDATE_VF_COUNTER_32bit(reg, last_counter, counter)     \
2690         {                                                       \
2691                 u32 current_counter = IXGBE_READ_REG(hw, reg);  \
2692                 if (current_counter < last_counter)             \
2693                         counter += 0x100000000LL;               \
2694                 last_counter = current_counter;                 \
2695                 counter &= 0xFFFFFFFF00000000LL;                \
2696                 counter |= current_counter;                     \
2697         }
2698
2699 #define UPDATE_VF_COUNTER_36bit(reg_lsb, reg_msb, last_counter, counter) \
2700         {                                                                \
2701                 u64 current_counter_lsb = IXGBE_READ_REG(hw, reg_lsb);   \
2702                 u64 current_counter_msb = IXGBE_READ_REG(hw, reg_msb);   \
2703                 u64 current_counter = (current_counter_msb << 32) |      \
2704                         current_counter_lsb;                             \
2705                 if (current_counter < last_counter)                      \
2706                         counter += 0x1000000000LL;                       \
2707                 last_counter = current_counter;                          \
2708                 counter &= 0xFFFFFFF000000000LL;                         \
2709                 counter |= current_counter;                              \
2710         }
2711 /**
2712  * ixgbevf_update_stats - Update the board statistics counters.
2713  * @adapter: board private structure
2714  **/
2715 void ixgbevf_update_stats(struct ixgbevf_adapter *adapter)
2716 {
2717         struct ixgbe_hw *hw = &adapter->hw;
2718         int i;
2719
2720         if (test_bit(__IXGBEVF_DOWN, &adapter->state) ||
2721             test_bit(__IXGBEVF_RESETTING, &adapter->state))
2722                 return;
2723
2724         UPDATE_VF_COUNTER_32bit(IXGBE_VFGPRC, adapter->stats.last_vfgprc,
2725                                 adapter->stats.vfgprc);
2726         UPDATE_VF_COUNTER_32bit(IXGBE_VFGPTC, adapter->stats.last_vfgptc,
2727                                 adapter->stats.vfgptc);
2728         UPDATE_VF_COUNTER_36bit(IXGBE_VFGORC_LSB, IXGBE_VFGORC_MSB,
2729                                 adapter->stats.last_vfgorc,
2730                                 adapter->stats.vfgorc);
2731         UPDATE_VF_COUNTER_36bit(IXGBE_VFGOTC_LSB, IXGBE_VFGOTC_MSB,
2732                                 adapter->stats.last_vfgotc,
2733                                 adapter->stats.vfgotc);
2734         UPDATE_VF_COUNTER_32bit(IXGBE_VFMPRC, adapter->stats.last_vfmprc,
2735                                 adapter->stats.vfmprc);
2736
2737         for (i = 0;  i  < adapter->num_rx_queues;  i++) {
2738                 adapter->hw_csum_rx_error +=
2739                         adapter->rx_ring[i]->hw_csum_rx_error;
2740                 adapter->rx_ring[i]->hw_csum_rx_error = 0;
2741         }
2742 }
2743
2744 /**
2745  * ixgbevf_service_timer - Timer Call-back
2746  * @data: pointer to adapter cast into an unsigned long
2747  **/
2748 static void ixgbevf_service_timer(unsigned long data)
2749 {
2750         struct ixgbevf_adapter *adapter = (struct ixgbevf_adapter *)data;
2751
2752         /* Reset the timer */
2753         mod_timer(&adapter->service_timer, (HZ * 2) + jiffies);
2754
2755         ixgbevf_service_event_schedule(adapter);
2756 }
2757
2758 static void ixgbevf_reset_subtask(struct ixgbevf_adapter *adapter)
2759 {
2760         if (!(adapter->flags & IXGBEVF_FLAG_RESET_REQUESTED))
2761                 return;
2762
2763         adapter->flags &= ~IXGBEVF_FLAG_RESET_REQUESTED;
2764
2765         /* If we're already down or resetting, just bail */
2766         if (test_bit(__IXGBEVF_DOWN, &adapter->state) ||
2767             test_bit(__IXGBEVF_RESETTING, &adapter->state))
2768                 return;
2769
2770         adapter->tx_timeout_count++;
2771
2772         ixgbevf_reinit_locked(adapter);
2773 }
2774
2775 /**
2776  * ixgbevf_check_hang_subtask - check for hung queues and dropped interrupts
2777  * @adapter: pointer to the device adapter structure
2778  *
2779  * This function serves two purposes.  First it strobes the interrupt lines
2780  * in order to make certain interrupts are occurring.  Secondly it sets the
2781  * bits needed to check for TX hangs.  As a result we should immediately
2782  * determine if a hang has occurred.
2783  **/
2784 static void ixgbevf_check_hang_subtask(struct ixgbevf_adapter *adapter)
2785 {
2786         struct ixgbe_hw *hw = &adapter->hw;
2787         u32 eics = 0;
2788         int i;
2789
2790         /* If we're down or resetting, just bail */
2791         if (test_bit(__IXGBEVF_DOWN, &adapter->state) ||
2792             test_bit(__IXGBEVF_RESETTING, &adapter->state))
2793                 return;
2794
2795         /* Force detection of hung controller */
2796         if (netif_carrier_ok(adapter->netdev)) {
2797                 for (i = 0; i < adapter->num_tx_queues; i++)
2798                         set_check_for_tx_hang(adapter->tx_ring[i]);
2799         }
2800
2801         /* get one bit for every active Tx/Rx interrupt vector */
2802         for (i = 0; i < adapter->num_msix_vectors - NON_Q_VECTORS; i++) {
2803                 struct ixgbevf_q_vector *qv = adapter->q_vector[i];
2804
2805                 if (qv->rx.ring || qv->tx.ring)
2806                         eics |= 1 << i;
2807         }
2808
2809         /* Cause software interrupt to ensure rings are cleaned */
2810         IXGBE_WRITE_REG(hw, IXGBE_VTEICS, eics);
2811 }
2812
2813 /**
2814  * ixgbevf_watchdog_update_link - update the link status
2815  * @adapter: pointer to the device adapter structure
2816  **/
2817 static void ixgbevf_watchdog_update_link(struct ixgbevf_adapter *adapter)
2818 {
2819         struct ixgbe_hw *hw = &adapter->hw;
2820         u32 link_speed = adapter->link_speed;
2821         bool link_up = adapter->link_up;
2822         s32 err;
2823
2824         spin_lock_bh(&adapter->mbx_lock);
2825
2826         err = hw->mac.ops.check_link(hw, &link_speed, &link_up, false);
2827
2828         spin_unlock_bh(&adapter->mbx_lock);
2829
2830         /* if check for link returns error we will need to reset */
2831         if (err && time_after(jiffies, adapter->last_reset + (10 * HZ))) {
2832                 adapter->flags |= IXGBEVF_FLAG_RESET_REQUESTED;
2833                 link_up = false;
2834         }
2835
2836         adapter->link_up = link_up;
2837         adapter->link_speed = link_speed;
2838 }
2839
2840 /**
2841  * ixgbevf_watchdog_link_is_up - update netif_carrier status and
2842  *                               print link up message
2843  * @adapter: pointer to the device adapter structure
2844  **/
2845 static void ixgbevf_watchdog_link_is_up(struct ixgbevf_adapter *adapter)
2846 {
2847         struct net_device *netdev = adapter->netdev;
2848
2849         /* only continue if link was previously down */
2850         if (netif_carrier_ok(netdev))
2851                 return;
2852
2853         dev_info(&adapter->pdev->dev, "NIC Link is Up %s\n",
2854                  (adapter->link_speed == IXGBE_LINK_SPEED_10GB_FULL) ?
2855                  "10 Gbps" :
2856                  (adapter->link_speed == IXGBE_LINK_SPEED_1GB_FULL) ?
2857                  "1 Gbps" :
2858                  (adapter->link_speed == IXGBE_LINK_SPEED_100_FULL) ?
2859                  "100 Mbps" :
2860                  "unknown speed");
2861
2862         netif_carrier_on(netdev);
2863 }
2864
2865 /**
2866  * ixgbevf_watchdog_link_is_down - update netif_carrier status and
2867  *                                 print link down message
2868  * @adapter: pointer to the adapter structure
2869  **/
2870 static void ixgbevf_watchdog_link_is_down(struct ixgbevf_adapter *adapter)
2871 {
2872         struct net_device *netdev = adapter->netdev;
2873
2874         adapter->link_speed = 0;
2875
2876         /* only continue if link was up previously */
2877         if (!netif_carrier_ok(netdev))
2878                 return;
2879
2880         dev_info(&adapter->pdev->dev, "NIC Link is Down\n");
2881
2882         netif_carrier_off(netdev);
2883 }
2884
2885 /**
2886  * ixgbevf_watchdog_subtask - worker thread to bring link up
2887  * @work: pointer to work_struct containing our data
2888  **/
2889 static void ixgbevf_watchdog_subtask(struct ixgbevf_adapter *adapter)
2890 {
2891         /* if interface is down do nothing */
2892         if (test_bit(__IXGBEVF_DOWN, &adapter->state) ||
2893             test_bit(__IXGBEVF_RESETTING, &adapter->state))
2894                 return;
2895
2896         ixgbevf_watchdog_update_link(adapter);
2897
2898         if (adapter->link_up)
2899                 ixgbevf_watchdog_link_is_up(adapter);
2900         else
2901                 ixgbevf_watchdog_link_is_down(adapter);
2902
2903         ixgbevf_update_stats(adapter);
2904 }
2905
2906 /**
2907  * ixgbevf_service_task - manages and runs subtasks
2908  * @work: pointer to work_struct containing our data
2909  **/
2910 static void ixgbevf_service_task(struct work_struct *work)
2911 {
2912         struct ixgbevf_adapter *adapter = container_of(work,
2913                                                        struct ixgbevf_adapter,
2914                                                        service_task);
2915         struct ixgbe_hw *hw = &adapter->hw;
2916
2917         if (IXGBE_REMOVED(hw->hw_addr)) {
2918                 if (!test_bit(__IXGBEVF_DOWN, &adapter->state)) {
2919                         rtnl_lock();
2920                         ixgbevf_down(adapter);
2921                         rtnl_unlock();
2922                 }
2923                 return;
2924         }
2925
2926         ixgbevf_queue_reset_subtask(adapter);
2927         ixgbevf_reset_subtask(adapter);
2928         ixgbevf_watchdog_subtask(adapter);
2929         ixgbevf_check_hang_subtask(adapter);
2930
2931         ixgbevf_service_event_complete(adapter);
2932 }
2933
2934 /**
2935  * ixgbevf_free_tx_resources - Free Tx Resources per Queue
2936  * @tx_ring: Tx descriptor ring for a specific queue
2937  *
2938  * Free all transmit software resources
2939  **/
2940 void ixgbevf_free_tx_resources(struct ixgbevf_ring *tx_ring)
2941 {
2942         ixgbevf_clean_tx_ring(tx_ring);
2943
2944         vfree(tx_ring->tx_buffer_info);
2945         tx_ring->tx_buffer_info = NULL;
2946
2947         /* if not set, then don't free */
2948         if (!tx_ring->desc)
2949                 return;
2950
2951         dma_free_coherent(tx_ring->dev, tx_ring->size, tx_ring->desc,
2952                           tx_ring->dma);
2953
2954         tx_ring->desc = NULL;
2955 }
2956
2957 /**
2958  * ixgbevf_free_all_tx_resources - Free Tx Resources for All Queues
2959  * @adapter: board private structure
2960  *
2961  * Free all transmit software resources
2962  **/
2963 static void ixgbevf_free_all_tx_resources(struct ixgbevf_adapter *adapter)
2964 {
2965         int i;
2966
2967         for (i = 0; i < adapter->num_tx_queues; i++)
2968                 if (adapter->tx_ring[i]->desc)
2969                         ixgbevf_free_tx_resources(adapter->tx_ring[i]);
2970 }
2971
2972 /**
2973  * ixgbevf_setup_tx_resources - allocate Tx resources (Descriptors)
2974  * @tx_ring: Tx descriptor ring (for a specific queue) to setup
2975  *
2976  * Return 0 on success, negative on failure
2977  **/
2978 int ixgbevf_setup_tx_resources(struct ixgbevf_ring *tx_ring)
2979 {
2980         int size;
2981
2982         size = sizeof(struct ixgbevf_tx_buffer) * tx_ring->count;
2983         tx_ring->tx_buffer_info = vzalloc(size);
2984         if (!tx_ring->tx_buffer_info)
2985                 goto err;
2986
2987         /* round up to nearest 4K */
2988         tx_ring->size = tx_ring->count * sizeof(union ixgbe_adv_tx_desc);
2989         tx_ring->size = ALIGN(tx_ring->size, 4096);
2990
2991         tx_ring->desc = dma_alloc_coherent(tx_ring->dev, tx_ring->size,
2992                                            &tx_ring->dma, GFP_KERNEL);
2993         if (!tx_ring->desc)
2994                 goto err;
2995
2996         return 0;
2997
2998 err:
2999         vfree(tx_ring->tx_buffer_info);
3000         tx_ring->tx_buffer_info = NULL;
3001         hw_dbg(&adapter->hw, "Unable to allocate memory for the transmit descriptor ring\n");
3002         return -ENOMEM;
3003 }
3004
3005 /**
3006  * ixgbevf_setup_all_tx_resources - allocate all queues Tx resources
3007  * @adapter: board private structure
3008  *
3009  * If this function returns with an error, then it's possible one or
3010  * more of the rings is populated (while the rest are not).  It is the
3011  * callers duty to clean those orphaned rings.
3012  *
3013  * Return 0 on success, negative on failure
3014  **/
3015 static int ixgbevf_setup_all_tx_resources(struct ixgbevf_adapter *adapter)
3016 {
3017         int i, err = 0;
3018
3019         for (i = 0; i < adapter->num_tx_queues; i++) {
3020                 err = ixgbevf_setup_tx_resources(adapter->tx_ring[i]);
3021                 if (!err)
3022                         continue;
3023                 hw_dbg(&adapter->hw, "Allocation for Tx Queue %u failed\n", i);
3024                 break;
3025         }
3026
3027         return err;
3028 }
3029
3030 /**
3031  * ixgbevf_setup_rx_resources - allocate Rx resources (Descriptors)
3032  * @rx_ring: Rx descriptor ring (for a specific queue) to setup
3033  *
3034  * Returns 0 on success, negative on failure
3035  **/
3036 int ixgbevf_setup_rx_resources(struct ixgbevf_ring *rx_ring)
3037 {
3038         int size;
3039
3040         size = sizeof(struct ixgbevf_rx_buffer) * rx_ring->count;
3041         rx_ring->rx_buffer_info = vzalloc(size);
3042         if (!rx_ring->rx_buffer_info)
3043                 goto err;
3044
3045         /* Round up to nearest 4K */
3046         rx_ring->size = rx_ring->count * sizeof(union ixgbe_adv_rx_desc);
3047         rx_ring->size = ALIGN(rx_ring->size, 4096);
3048
3049         rx_ring->desc = dma_alloc_coherent(rx_ring->dev, rx_ring->size,
3050                                            &rx_ring->dma, GFP_KERNEL);
3051
3052         if (!rx_ring->desc)
3053                 goto err;
3054
3055         return 0;
3056 err:
3057         vfree(rx_ring->rx_buffer_info);
3058         rx_ring->rx_buffer_info = NULL;
3059         dev_err(rx_ring->dev, "Unable to allocate memory for the Rx descriptor ring\n");
3060         return -ENOMEM;
3061 }
3062
3063 /**
3064  * ixgbevf_setup_all_rx_resources - allocate all queues Rx resources
3065  * @adapter: board private structure
3066  *
3067  * If this function returns with an error, then it's possible one or
3068  * more of the rings is populated (while the rest are not).  It is the
3069  * callers duty to clean those orphaned rings.
3070  *
3071  * Return 0 on success, negative on failure
3072  **/
3073 static int ixgbevf_setup_all_rx_resources(struct ixgbevf_adapter *adapter)
3074 {
3075         int i, err = 0;
3076
3077         for (i = 0; i < adapter->num_rx_queues; i++) {
3078                 err = ixgbevf_setup_rx_resources(adapter->rx_ring[i]);
3079                 if (!err)
3080                         continue;
3081                 hw_dbg(&adapter->hw, "Allocation for Rx Queue %u failed\n", i);
3082                 break;
3083         }
3084         return err;
3085 }
3086
3087 /**
3088  * ixgbevf_free_rx_resources - Free Rx Resources
3089  * @rx_ring: ring to clean the resources from
3090  *
3091  * Free all receive software resources
3092  **/
3093 void ixgbevf_free_rx_resources(struct ixgbevf_ring *rx_ring)
3094 {
3095         ixgbevf_clean_rx_ring(rx_ring);
3096
3097         vfree(rx_ring->rx_buffer_info);
3098         rx_ring->rx_buffer_info = NULL;
3099
3100         dma_free_coherent(rx_ring->dev, rx_ring->size, rx_ring->desc,
3101                           rx_ring->dma);
3102
3103         rx_ring->desc = NULL;
3104 }
3105
3106 /**
3107  * ixgbevf_free_all_rx_resources - Free Rx Resources for All Queues
3108  * @adapter: board private structure
3109  *
3110  * Free all receive software resources
3111  **/
3112 static void ixgbevf_free_all_rx_resources(struct ixgbevf_adapter *adapter)
3113 {
3114         int i;
3115
3116         for (i = 0; i < adapter->num_rx_queues; i++)
3117                 if (adapter->rx_ring[i]->desc)
3118                         ixgbevf_free_rx_resources(adapter->rx_ring[i]);
3119 }
3120
3121 /**
3122  * ixgbevf_open - Called when a network interface is made active
3123  * @netdev: network interface device structure
3124  *
3125  * Returns 0 on success, negative value on failure
3126  *
3127  * The open entry point is called when a network interface is made
3128  * active by the system (IFF_UP).  At this point all resources needed
3129  * for transmit and receive operations are allocated, the interrupt
3130  * handler is registered with the OS, the watchdog timer is started,
3131  * and the stack is notified that the interface is ready.
3132  **/
3133 static int ixgbevf_open(struct net_device *netdev)
3134 {
3135         struct ixgbevf_adapter *adapter = netdev_priv(netdev);
3136         struct ixgbe_hw *hw = &adapter->hw;
3137         int err;
3138
3139         /* A previous failure to open the device because of a lack of
3140          * available MSIX vector resources may have reset the number
3141          * of msix vectors variable to zero.  The only way to recover
3142          * is to unload/reload the driver and hope that the system has
3143          * been able to recover some MSIX vector resources.
3144          */
3145         if (!adapter->num_msix_vectors)
3146                 return -ENOMEM;
3147
3148         if (hw->adapter_stopped) {
3149                 ixgbevf_reset(adapter);
3150                 /* if adapter is still stopped then PF isn't up and
3151                  * the VF can't start.
3152                  */
3153                 if (hw->adapter_stopped) {
3154                         err = IXGBE_ERR_MBX;
3155                         pr_err("Unable to start - perhaps the PF Driver isn't up yet\n");
3156                         goto err_setup_reset;
3157                 }
3158         }
3159
3160         /* disallow open during test */
3161         if (test_bit(__IXGBEVF_TESTING, &adapter->state))
3162                 return -EBUSY;
3163
3164         netif_carrier_off(netdev);
3165
3166         /* allocate transmit descriptors */
3167         err = ixgbevf_setup_all_tx_resources(adapter);
3168         if (err)
3169                 goto err_setup_tx;
3170
3171         /* allocate receive descriptors */
3172         err = ixgbevf_setup_all_rx_resources(adapter);
3173         if (err)
3174                 goto err_setup_rx;
3175
3176         ixgbevf_configure(adapter);
3177
3178         /* Map the Tx/Rx rings to the vectors we were allotted.
3179          * if request_irq will be called in this function map_rings
3180          * must be called *before* up_complete
3181          */
3182         ixgbevf_map_rings_to_vectors(adapter);
3183
3184         err = ixgbevf_request_irq(adapter);
3185         if (err)
3186                 goto err_req_irq;
3187
3188         ixgbevf_up_complete(adapter);
3189
3190         return 0;
3191
3192 err_req_irq:
3193         ixgbevf_down(adapter);
3194 err_setup_rx:
3195         ixgbevf_free_all_rx_resources(adapter);
3196 err_setup_tx:
3197         ixgbevf_free_all_tx_resources(adapter);
3198         ixgbevf_reset(adapter);
3199
3200 err_setup_reset:
3201
3202         return err;
3203 }
3204
3205 /**
3206  * ixgbevf_close - Disables a network interface
3207  * @netdev: network interface device structure
3208  *
3209  * Returns 0, this is not allowed to fail
3210  *
3211  * The close entry point is called when an interface is de-activated
3212  * by the OS.  The hardware is still under the drivers control, but
3213  * needs to be disabled.  A global MAC reset is issued to stop the
3214  * hardware, and all transmit and receive resources are freed.
3215  **/
3216 static int ixgbevf_close(struct net_device *netdev)
3217 {
3218         struct ixgbevf_adapter *adapter = netdev_priv(netdev);
3219
3220         ixgbevf_down(adapter);
3221         ixgbevf_free_irq(adapter);
3222
3223         ixgbevf_free_all_tx_resources(adapter);
3224         ixgbevf_free_all_rx_resources(adapter);
3225
3226         return 0;
3227 }
3228
3229 static void ixgbevf_queue_reset_subtask(struct ixgbevf_adapter *adapter)
3230 {
3231         struct net_device *dev = adapter->netdev;
3232
3233         if (!(adapter->flags & IXGBEVF_FLAG_QUEUE_RESET_REQUESTED))
3234                 return;
3235
3236         adapter->flags &= ~IXGBEVF_FLAG_QUEUE_RESET_REQUESTED;
3237
3238         /* if interface is down do nothing */
3239         if (test_bit(__IXGBEVF_DOWN, &adapter->state) ||
3240             test_bit(__IXGBEVF_RESETTING, &adapter->state))
3241                 return;
3242
3243         /* Hardware has to reinitialize queues and interrupts to
3244          * match packet buffer alignment. Unfortunately, the
3245          * hardware is not flexible enough to do this dynamically.
3246          */
3247         if (netif_running(dev))
3248                 ixgbevf_close(dev);
3249
3250         ixgbevf_clear_interrupt_scheme(adapter);
3251         ixgbevf_init_interrupt_scheme(adapter);
3252
3253         if (netif_running(dev))
3254                 ixgbevf_open(dev);
3255 }
3256
3257 static void ixgbevf_tx_ctxtdesc(struct ixgbevf_ring *tx_ring,
3258                                 u32 vlan_macip_lens, u32 type_tucmd,
3259                                 u32 mss_l4len_idx)
3260 {
3261         struct ixgbe_adv_tx_context_desc *context_desc;
3262         u16 i = tx_ring->next_to_use;
3263
3264         context_desc = IXGBEVF_TX_CTXTDESC(tx_ring, i);
3265
3266         i++;
3267         tx_ring->next_to_use = (i < tx_ring->count) ? i : 0;
3268
3269         /* set bits to identify this as an advanced context descriptor */
3270         type_tucmd |= IXGBE_TXD_CMD_DEXT | IXGBE_ADVTXD_DTYP_CTXT;
3271
3272         context_desc->vlan_macip_lens   = cpu_to_le32(vlan_macip_lens);
3273         context_desc->seqnum_seed       = 0;
3274         context_desc->type_tucmd_mlhl   = cpu_to_le32(type_tucmd);
3275         context_desc->mss_l4len_idx     = cpu_to_le32(mss_l4len_idx);
3276 }
3277
3278 static int ixgbevf_tso(struct ixgbevf_ring *tx_ring,
3279                        struct ixgbevf_tx_buffer *first,
3280                        u8 *hdr_len)
3281 {
3282         struct sk_buff *skb = first->skb;
3283         u32 vlan_macip_lens, type_tucmd;
3284         u32 mss_l4len_idx, l4len;
3285         int err;
3286
3287         if (skb->ip_summed != CHECKSUM_PARTIAL)
3288                 return 0;
3289
3290         if (!skb_is_gso(skb))
3291                 return 0;
3292
3293         err = skb_cow_head(skb, 0);
3294         if (err < 0)
3295                 return err;
3296
3297         /* ADV DTYP TUCMD MKRLOC/ISCSIHEDLEN */
3298         type_tucmd = IXGBE_ADVTXD_TUCMD_L4T_TCP;
3299
3300         if (first->protocol == htons(ETH_P_IP)) {
3301                 struct iphdr *iph = ip_hdr(skb);
3302
3303                 iph->tot_len = 0;
3304                 iph->check = 0;
3305                 tcp_hdr(skb)->check = ~csum_tcpudp_magic(iph->saddr,
3306                                                          iph->daddr, 0,
3307                                                          IPPROTO_TCP,
3308                                                          0);
3309                 type_tucmd |= IXGBE_ADVTXD_TUCMD_IPV4;
3310                 first->tx_flags |= IXGBE_TX_FLAGS_TSO |
3311                                    IXGBE_TX_FLAGS_CSUM |
3312                                    IXGBE_TX_FLAGS_IPV4;
3313         } else if (skb_is_gso_v6(skb)) {
3314                 ipv6_hdr(skb)->payload_len = 0;
3315                 tcp_hdr(skb)->check =
3316                     ~csum_ipv6_magic(&ipv6_hdr(skb)->saddr,
3317                                      &ipv6_hdr(skb)->daddr,
3318                                      0, IPPROTO_TCP, 0);
3319                 first->tx_flags |= IXGBE_TX_FLAGS_TSO |
3320                                    IXGBE_TX_FLAGS_CSUM;
3321         }
3322
3323         /* compute header lengths */
3324         l4len = tcp_hdrlen(skb);
3325         *hdr_len += l4len;
3326         *hdr_len = skb_transport_offset(skb) + l4len;
3327
3328         /* update GSO size and bytecount with header size */
3329         first->gso_segs = skb_shinfo(skb)->gso_segs;
3330         first->bytecount += (first->gso_segs - 1) * *hdr_len;
3331
3332         /* mss_l4len_id: use 1 as index for TSO */
3333         mss_l4len_idx = l4len << IXGBE_ADVTXD_L4LEN_SHIFT;
3334         mss_l4len_idx |= skb_shinfo(skb)->gso_size << IXGBE_ADVTXD_MSS_SHIFT;
3335         mss_l4len_idx |= 1 << IXGBE_ADVTXD_IDX_SHIFT;
3336
3337         /* vlan_macip_lens: HEADLEN, MACLEN, VLAN tag */
3338         vlan_macip_lens = skb_network_header_len(skb);
3339         vlan_macip_lens |= skb_network_offset(skb) << IXGBE_ADVTXD_MACLEN_SHIFT;
3340         vlan_macip_lens |= first->tx_flags & IXGBE_TX_FLAGS_VLAN_MASK;
3341
3342         ixgbevf_tx_ctxtdesc(tx_ring, vlan_macip_lens,
3343                             type_tucmd, mss_l4len_idx);
3344
3345         return 1;
3346 }
3347
3348 static void ixgbevf_tx_csum(struct ixgbevf_ring *tx_ring,
3349                             struct ixgbevf_tx_buffer *first)
3350 {
3351         struct sk_buff *skb = first->skb;
3352         u32 vlan_macip_lens = 0;
3353         u32 mss_l4len_idx = 0;
3354         u32 type_tucmd = 0;
3355
3356         if (skb->ip_summed == CHECKSUM_PARTIAL) {
3357                 u8 l4_hdr = 0;
3358
3359                 switch (first->protocol) {
3360                 case htons(ETH_P_IP):
3361                         vlan_macip_lens |= skb_network_header_len(skb);
3362                         type_tucmd |= IXGBE_ADVTXD_TUCMD_IPV4;
3363                         l4_hdr = ip_hdr(skb)->protocol;
3364                         break;
3365                 case htons(ETH_P_IPV6):
3366                         vlan_macip_lens |= skb_network_header_len(skb);
3367                         l4_hdr = ipv6_hdr(skb)->nexthdr;
3368                         break;
3369                 default:
3370                         if (unlikely(net_ratelimit())) {
3371                                 dev_warn(tx_ring->dev,
3372                                          "partial checksum but proto=%x!\n",
3373                                          first->protocol);
3374                         }
3375                         break;
3376                 }
3377
3378                 switch (l4_hdr) {
3379                 case IPPROTO_TCP:
3380                         type_tucmd |= IXGBE_ADVTXD_TUCMD_L4T_TCP;
3381                         mss_l4len_idx = tcp_hdrlen(skb) <<
3382                                         IXGBE_ADVTXD_L4LEN_SHIFT;
3383                         break;
3384                 case IPPROTO_SCTP:
3385                         type_tucmd |= IXGBE_ADVTXD_TUCMD_L4T_SCTP;
3386                         mss_l4len_idx = sizeof(struct sctphdr) <<
3387                                         IXGBE_ADVTXD_L4LEN_SHIFT;
3388                         break;
3389                 case IPPROTO_UDP:
3390                         mss_l4len_idx = sizeof(struct udphdr) <<
3391                                         IXGBE_ADVTXD_L4LEN_SHIFT;
3392                         break;
3393                 default:
3394                         if (unlikely(net_ratelimit())) {
3395                                 dev_warn(tx_ring->dev,
3396                                          "partial checksum but l4 proto=%x!\n",
3397                                          l4_hdr);
3398                         }
3399                         break;
3400                 }
3401
3402                 /* update TX checksum flag */
3403                 first->tx_flags |= IXGBE_TX_FLAGS_CSUM;
3404         }
3405
3406         /* vlan_macip_lens: MACLEN, VLAN tag */
3407         vlan_macip_lens |= skb_network_offset(skb) << IXGBE_ADVTXD_MACLEN_SHIFT;
3408         vlan_macip_lens |= first->tx_flags & IXGBE_TX_FLAGS_VLAN_MASK;
3409
3410         ixgbevf_tx_ctxtdesc(tx_ring, vlan_macip_lens,
3411                             type_tucmd, mss_l4len_idx);
3412 }
3413
3414 static __le32 ixgbevf_tx_cmd_type(u32 tx_flags)
3415 {
3416         /* set type for advanced descriptor with frame checksum insertion */
3417         __le32 cmd_type = cpu_to_le32(IXGBE_ADVTXD_DTYP_DATA |
3418                                       IXGBE_ADVTXD_DCMD_IFCS |
3419                                       IXGBE_ADVTXD_DCMD_DEXT);
3420
3421         /* set HW VLAN bit if VLAN is present */
3422         if (tx_flags & IXGBE_TX_FLAGS_VLAN)
3423                 cmd_type |= cpu_to_le32(IXGBE_ADVTXD_DCMD_VLE);
3424
3425         /* set segmentation enable bits for TSO/FSO */
3426         if (tx_flags & IXGBE_TX_FLAGS_TSO)
3427                 cmd_type |= cpu_to_le32(IXGBE_ADVTXD_DCMD_TSE);
3428
3429         return cmd_type;
3430 }
3431
3432 static void ixgbevf_tx_olinfo_status(union ixgbe_adv_tx_desc *tx_desc,
3433                                      u32 tx_flags, unsigned int paylen)
3434 {
3435         __le32 olinfo_status = cpu_to_le32(paylen << IXGBE_ADVTXD_PAYLEN_SHIFT);
3436
3437         /* enable L4 checksum for TSO and TX checksum offload */
3438         if (tx_flags & IXGBE_TX_FLAGS_CSUM)
3439                 olinfo_status |= cpu_to_le32(IXGBE_ADVTXD_POPTS_TXSM);
3440
3441         /* enble IPv4 checksum for TSO */
3442         if (tx_flags & IXGBE_TX_FLAGS_IPV4)
3443                 olinfo_status |= cpu_to_le32(IXGBE_ADVTXD_POPTS_IXSM);
3444
3445         /* use index 1 context for TSO/FSO/FCOE */
3446         if (tx_flags & IXGBE_TX_FLAGS_TSO)
3447                 olinfo_status |= cpu_to_le32(1 << IXGBE_ADVTXD_IDX_SHIFT);
3448
3449         /* Check Context must be set if Tx switch is enabled, which it
3450          * always is for case where virtual functions are running
3451          */
3452         olinfo_status |= cpu_to_le32(IXGBE_ADVTXD_CC);
3453
3454         tx_desc->read.olinfo_status = olinfo_status;
3455 }
3456
3457 static void ixgbevf_tx_map(struct ixgbevf_ring *tx_ring,
3458                            struct ixgbevf_tx_buffer *first,
3459                            const u8 hdr_len)
3460 {
3461         dma_addr_t dma;
3462         struct sk_buff *skb = first->skb;
3463         struct ixgbevf_tx_buffer *tx_buffer;
3464         union ixgbe_adv_tx_desc *tx_desc;
3465         struct skb_frag_struct *frag = &skb_shinfo(skb)->frags[0];
3466         unsigned int data_len = skb->data_len;
3467         unsigned int size = skb_headlen(skb);
3468         unsigned int paylen = skb->len - hdr_len;
3469         u32 tx_flags = first->tx_flags;
3470         __le32 cmd_type;
3471         u16 i = tx_ring->next_to_use;
3472
3473         tx_desc = IXGBEVF_TX_DESC(tx_ring, i);
3474
3475         ixgbevf_tx_olinfo_status(tx_desc, tx_flags, paylen);
3476         cmd_type = ixgbevf_tx_cmd_type(tx_flags);
3477
3478         dma = dma_map_single(tx_ring->dev, skb->data, size, DMA_TO_DEVICE);
3479         if (dma_mapping_error(tx_ring->dev, dma))
3480                 goto dma_error;
3481
3482         /* record length, and DMA address */
3483         dma_unmap_len_set(first, len, size);
3484         dma_unmap_addr_set(first, dma, dma);
3485
3486         tx_desc->read.buffer_addr = cpu_to_le64(dma);
3487
3488         for (;;) {
3489                 while (unlikely(size > IXGBE_MAX_DATA_PER_TXD)) {
3490                         tx_desc->read.cmd_type_len =
3491                                 cmd_type | cpu_to_le32(IXGBE_MAX_DATA_PER_TXD);
3492
3493                         i++;
3494                         tx_desc++;
3495                         if (i == tx_ring->count) {
3496                                 tx_desc = IXGBEVF_TX_DESC(tx_ring, 0);
3497                                 i = 0;
3498                         }
3499
3500                         dma += IXGBE_MAX_DATA_PER_TXD;
3501                         size -= IXGBE_MAX_DATA_PER_TXD;
3502
3503                         tx_desc->read.buffer_addr = cpu_to_le64(dma);
3504                         tx_desc->read.olinfo_status = 0;
3505                 }
3506
3507                 if (likely(!data_len))
3508                         break;
3509
3510                 tx_desc->read.cmd_type_len = cmd_type | cpu_to_le32(size);
3511
3512                 i++;
3513                 tx_desc++;
3514                 if (i == tx_ring->count) {
3515                         tx_desc = IXGBEVF_TX_DESC(tx_ring, 0);
3516                         i = 0;
3517                 }
3518
3519                 size = skb_frag_size(frag);
3520                 data_len -= size;
3521
3522                 dma = skb_frag_dma_map(tx_ring->dev, frag, 0, size,
3523                                        DMA_TO_DEVICE);
3524                 if (dma_mapping_error(tx_ring->dev, dma))
3525                         goto dma_error;
3526
3527                 tx_buffer = &tx_ring->tx_buffer_info[i];
3528                 dma_unmap_len_set(tx_buffer, len, size);
3529                 dma_unmap_addr_set(tx_buffer, dma, dma);
3530
3531                 tx_desc->read.buffer_addr = cpu_to_le64(dma);
3532                 tx_desc->read.olinfo_status = 0;
3533
3534                 frag++;
3535         }
3536
3537         /* write last descriptor with RS and EOP bits */
3538         cmd_type |= cpu_to_le32(size) | cpu_to_le32(IXGBE_TXD_CMD);
3539         tx_desc->read.cmd_type_len = cmd_type;
3540
3541         /* set the timestamp */
3542         first->time_stamp = jiffies;
3543
3544         /* Force memory writes to complete before letting h/w know there
3545          * are new descriptors to fetch.  (Only applicable for weak-ordered
3546          * memory model archs, such as IA-64).
3547          *
3548          * We also need this memory barrier (wmb) to make certain all of the
3549          * status bits have been updated before next_to_watch is written.
3550          */
3551         wmb();
3552
3553         /* set next_to_watch value indicating a packet is present */
3554         first->next_to_watch = tx_desc;
3555
3556         i++;
3557         if (i == tx_ring->count)
3558                 i = 0;
3559
3560         tx_ring->next_to_use = i;
3561
3562         /* notify HW of packet */
3563         ixgbevf_write_tail(tx_ring, i);
3564
3565         return;
3566 dma_error:
3567         dev_err(tx_ring->dev, "TX DMA map failed\n");
3568
3569         /* clear dma mappings for failed tx_buffer_info map */
3570         for (;;) {
3571                 tx_buffer = &tx_ring->tx_buffer_info[i];
3572                 ixgbevf_unmap_and_free_tx_resource(tx_ring, tx_buffer);
3573                 if (tx_buffer == first)
3574                         break;
3575                 if (i == 0)
3576                         i = tx_ring->count;
3577                 i--;
3578         }
3579
3580         tx_ring->next_to_use = i;
3581 }
3582
3583 static int __ixgbevf_maybe_stop_tx(struct ixgbevf_ring *tx_ring, int size)
3584 {
3585         netif_stop_subqueue(tx_ring->netdev, tx_ring->queue_index);
3586         /* Herbert's original patch had:
3587          *  smp_mb__after_netif_stop_queue();
3588          * but since that doesn't exist yet, just open code it.
3589          */
3590         smp_mb();
3591
3592         /* We need to check again in a case another CPU has just
3593          * made room available.
3594          */
3595         if (likely(ixgbevf_desc_unused(tx_ring) < size))
3596                 return -EBUSY;
3597
3598         /* A reprieve! - use start_queue because it doesn't call schedule */
3599         netif_start_subqueue(tx_ring->netdev, tx_ring->queue_index);
3600         ++tx_ring->tx_stats.restart_queue;
3601
3602         return 0;
3603 }
3604
3605 static int ixgbevf_maybe_stop_tx(struct ixgbevf_ring *tx_ring, int size)
3606 {
3607         if (likely(ixgbevf_desc_unused(tx_ring) >= size))
3608                 return 0;
3609         return __ixgbevf_maybe_stop_tx(tx_ring, size);
3610 }
3611
3612 static int ixgbevf_xmit_frame(struct sk_buff *skb, struct net_device *netdev)
3613 {
3614         struct ixgbevf_adapter *adapter = netdev_priv(netdev);
3615         struct ixgbevf_tx_buffer *first;
3616         struct ixgbevf_ring *tx_ring;
3617         int tso;
3618         u32 tx_flags = 0;
3619         u16 count = TXD_USE_COUNT(skb_headlen(skb));
3620 #if PAGE_SIZE > IXGBE_MAX_DATA_PER_TXD
3621         unsigned short f;
3622 #endif
3623         u8 hdr_len = 0;
3624         u8 *dst_mac = skb_header_pointer(skb, 0, 0, NULL);
3625
3626         if (!dst_mac || is_link_local_ether_addr(dst_mac)) {
3627                 dev_kfree_skb_any(skb);
3628                 return NETDEV_TX_OK;
3629         }
3630
3631         tx_ring = adapter->tx_ring[skb->queue_mapping];
3632
3633         /* need: 1 descriptor per page * PAGE_SIZE/IXGBE_MAX_DATA_PER_TXD,
3634          *       + 1 desc for skb_headlen/IXGBE_MAX_DATA_PER_TXD,
3635          *       + 2 desc gap to keep tail from touching head,
3636          *       + 1 desc for context descriptor,
3637          * otherwise try next time
3638          */
3639 #if PAGE_SIZE > IXGBE_MAX_DATA_PER_TXD
3640         for (f = 0; f < skb_shinfo(skb)->nr_frags; f++)
3641                 count += TXD_USE_COUNT(skb_shinfo(skb)->frags[f].size);
3642 #else
3643         count += skb_shinfo(skb)->nr_frags;
3644 #endif
3645         if (ixgbevf_maybe_stop_tx(tx_ring, count + 3)) {
3646                 tx_ring->tx_stats.tx_busy++;
3647                 return NETDEV_TX_BUSY;
3648         }
3649
3650         /* record the location of the first descriptor for this packet */
3651         first = &tx_ring->tx_buffer_info[tx_ring->next_to_use];
3652         first->skb = skb;
3653         first->bytecount = skb->len;
3654         first->gso_segs = 1;
3655
3656         if (skb_vlan_tag_present(skb)) {
3657                 tx_flags |= skb_vlan_tag_get(skb);
3658                 tx_flags <<= IXGBE_TX_FLAGS_VLAN_SHIFT;
3659                 tx_flags |= IXGBE_TX_FLAGS_VLAN;
3660         }
3661
3662         /* record initial flags and protocol */
3663         first->tx_flags = tx_flags;
3664         first->protocol = vlan_get_protocol(skb);
3665
3666         tso = ixgbevf_tso(tx_ring, first, &hdr_len);
3667         if (tso < 0)
3668                 goto out_drop;
3669         else if (!tso)
3670                 ixgbevf_tx_csum(tx_ring, first);
3671
3672         ixgbevf_tx_map(tx_ring, first, hdr_len);
3673
3674         ixgbevf_maybe_stop_tx(tx_ring, DESC_NEEDED);
3675
3676         return NETDEV_TX_OK;
3677
3678 out_drop:
3679         dev_kfree_skb_any(first->skb);
3680         first->skb = NULL;
3681
3682         return NETDEV_TX_OK;
3683 }
3684
3685 /**
3686  * ixgbevf_set_mac - Change the Ethernet Address of the NIC
3687  * @netdev: network interface device structure
3688  * @p: pointer to an address structure
3689  *
3690  * Returns 0 on success, negative on failure
3691  **/
3692 static int ixgbevf_set_mac(struct net_device *netdev, void *p)
3693 {
3694         struct ixgbevf_adapter *adapter = netdev_priv(netdev);
3695         struct ixgbe_hw *hw = &adapter->hw;
3696         struct sockaddr *addr = p;
3697
3698         if (!is_valid_ether_addr(addr->sa_data))
3699                 return -EADDRNOTAVAIL;
3700
3701         memcpy(netdev->dev_addr, addr->sa_data, netdev->addr_len);
3702         memcpy(hw->mac.addr, addr->sa_data, netdev->addr_len);
3703
3704         spin_lock_bh(&adapter->mbx_lock);
3705
3706         hw->mac.ops.set_rar(hw, 0, hw->mac.addr, 0);
3707
3708         spin_unlock_bh(&adapter->mbx_lock);
3709
3710         return 0;
3711 }
3712
3713 /**
3714  * ixgbevf_change_mtu - Change the Maximum Transfer Unit
3715  * @netdev: network interface device structure
3716  * @new_mtu: new value for maximum frame size
3717  *
3718  * Returns 0 on success, negative on failure
3719  **/
3720 static int ixgbevf_change_mtu(struct net_device *netdev, int new_mtu)
3721 {
3722         struct ixgbevf_adapter *adapter = netdev_priv(netdev);
3723         struct ixgbe_hw *hw = &adapter->hw;
3724         int max_frame = new_mtu + ETH_HLEN + ETH_FCS_LEN;
3725         int max_possible_frame = MAXIMUM_ETHERNET_VLAN_SIZE;
3726
3727         switch (adapter->hw.api_version) {
3728         case ixgbe_mbox_api_11:
3729         case ixgbe_mbox_api_12:
3730                 max_possible_frame = IXGBE_MAX_JUMBO_FRAME_SIZE;
3731                 break;
3732         default:
3733                 if (adapter->hw.mac.type != ixgbe_mac_82599_vf)
3734                         max_possible_frame = IXGBE_MAX_JUMBO_FRAME_SIZE;
3735                 break;
3736         }
3737
3738         /* MTU < 68 is an error and causes problems on some kernels */
3739         if ((new_mtu < 68) || (max_frame > max_possible_frame))
3740                 return -EINVAL;
3741
3742         hw_dbg(hw, "changing MTU from %d to %d\n",
3743                netdev->mtu, new_mtu);
3744         /* must set new MTU before calling down or up */
3745         netdev->mtu = new_mtu;
3746
3747         /* notify the PF of our intent to use this size of frame */
3748         ixgbevf_rlpml_set_vf(hw, max_frame);
3749
3750         return 0;
3751 }
3752
3753 #ifdef CONFIG_NET_POLL_CONTROLLER
3754 /* Polling 'interrupt' - used by things like netconsole to send skbs
3755  * without having to re-enable interrupts. It's not called while
3756  * the interrupt routine is executing.
3757  */
3758 static void ixgbevf_netpoll(struct net_device *netdev)
3759 {
3760         struct ixgbevf_adapter *adapter = netdev_priv(netdev);
3761         int i;
3762
3763         /* if interface is down do nothing */
3764         if (test_bit(__IXGBEVF_DOWN, &adapter->state))
3765                 return;
3766         for (i = 0; i < adapter->num_rx_queues; i++)
3767                 ixgbevf_msix_clean_rings(0, adapter->q_vector[i]);
3768 }
3769 #endif /* CONFIG_NET_POLL_CONTROLLER */
3770
3771 static int ixgbevf_suspend(struct pci_dev *pdev, pm_message_t state)
3772 {
3773         struct net_device *netdev = pci_get_drvdata(pdev);
3774         struct ixgbevf_adapter *adapter = netdev_priv(netdev);
3775 #ifdef CONFIG_PM
3776         int retval = 0;
3777 #endif
3778
3779         netif_device_detach(netdev);
3780
3781         if (netif_running(netdev)) {
3782                 rtnl_lock();
3783                 ixgbevf_down(adapter);
3784                 ixgbevf_free_irq(adapter);
3785                 ixgbevf_free_all_tx_resources(adapter);
3786                 ixgbevf_free_all_rx_resources(adapter);
3787                 rtnl_unlock();
3788         }
3789
3790         ixgbevf_clear_interrupt_scheme(adapter);
3791
3792 #ifdef CONFIG_PM
3793         retval = pci_save_state(pdev);
3794         if (retval)
3795                 return retval;
3796
3797 #endif
3798         if (!test_and_set_bit(__IXGBEVF_DISABLED, &adapter->state))
3799                 pci_disable_device(pdev);
3800
3801         return 0;
3802 }
3803
3804 #ifdef CONFIG_PM
3805 static int ixgbevf_resume(struct pci_dev *pdev)
3806 {
3807         struct net_device *netdev = pci_get_drvdata(pdev);
3808         struct ixgbevf_adapter *adapter = netdev_priv(netdev);
3809         u32 err;
3810
3811         pci_restore_state(pdev);
3812         /* pci_restore_state clears dev->state_saved so call
3813          * pci_save_state to restore it.
3814          */
3815         pci_save_state(pdev);
3816
3817         err = pci_enable_device_mem(pdev);
3818         if (err) {
3819                 dev_err(&pdev->dev, "Cannot enable PCI device from suspend\n");
3820                 return err;
3821         }
3822         smp_mb__before_atomic();
3823         clear_bit(__IXGBEVF_DISABLED, &adapter->state);
3824         pci_set_master(pdev);
3825
3826         ixgbevf_reset(adapter);
3827
3828         rtnl_lock();
3829         err = ixgbevf_init_interrupt_scheme(adapter);
3830         rtnl_unlock();
3831         if (err) {
3832                 dev_err(&pdev->dev, "Cannot initialize interrupts\n");
3833                 return err;
3834         }
3835
3836         if (netif_running(netdev)) {
3837                 err = ixgbevf_open(netdev);
3838                 if (err)
3839                         return err;
3840         }
3841
3842         netif_device_attach(netdev);
3843
3844         return err;
3845 }
3846
3847 #endif /* CONFIG_PM */
3848 static void ixgbevf_shutdown(struct pci_dev *pdev)
3849 {
3850         ixgbevf_suspend(pdev, PMSG_SUSPEND);
3851 }
3852
3853 static struct rtnl_link_stats64 *ixgbevf_get_stats(struct net_device *netdev,
3854                                                 struct rtnl_link_stats64 *stats)
3855 {
3856         struct ixgbevf_adapter *adapter = netdev_priv(netdev);
3857         unsigned int start;
3858         u64 bytes, packets;
3859         const struct ixgbevf_ring *ring;
3860         int i;
3861
3862         ixgbevf_update_stats(adapter);
3863
3864         stats->multicast = adapter->stats.vfmprc - adapter->stats.base_vfmprc;
3865
3866         for (i = 0; i < adapter->num_rx_queues; i++) {
3867                 ring = adapter->rx_ring[i];
3868                 do {
3869                         start = u64_stats_fetch_begin_irq(&ring->syncp);
3870                         bytes = ring->stats.bytes;
3871                         packets = ring->stats.packets;
3872                 } while (u64_stats_fetch_retry_irq(&ring->syncp, start));
3873                 stats->rx_bytes += bytes;
3874                 stats->rx_packets += packets;
3875         }
3876
3877         for (i = 0; i < adapter->num_tx_queues; i++) {
3878                 ring = adapter->tx_ring[i];
3879                 do {
3880                         start = u64_stats_fetch_begin_irq(&ring->syncp);
3881                         bytes = ring->stats.bytes;
3882                         packets = ring->stats.packets;
3883                 } while (u64_stats_fetch_retry_irq(&ring->syncp, start));
3884                 stats->tx_bytes += bytes;
3885                 stats->tx_packets += packets;
3886         }
3887
3888         return stats;
3889 }
3890
3891 static const struct net_device_ops ixgbevf_netdev_ops = {
3892         .ndo_open               = ixgbevf_open,
3893         .ndo_stop               = ixgbevf_close,
3894         .ndo_start_xmit         = ixgbevf_xmit_frame,
3895         .ndo_set_rx_mode        = ixgbevf_set_rx_mode,
3896         .ndo_get_stats64        = ixgbevf_get_stats,
3897         .ndo_validate_addr      = eth_validate_addr,
3898         .ndo_set_mac_address    = ixgbevf_set_mac,
3899         .ndo_change_mtu         = ixgbevf_change_mtu,
3900         .ndo_tx_timeout         = ixgbevf_tx_timeout,
3901         .ndo_vlan_rx_add_vid    = ixgbevf_vlan_rx_add_vid,
3902         .ndo_vlan_rx_kill_vid   = ixgbevf_vlan_rx_kill_vid,
3903 #ifdef CONFIG_NET_RX_BUSY_POLL
3904         .ndo_busy_poll          = ixgbevf_busy_poll_recv,
3905 #endif
3906 #ifdef CONFIG_NET_POLL_CONTROLLER
3907         .ndo_poll_controller    = ixgbevf_netpoll,
3908 #endif
3909         .ndo_features_check     = passthru_features_check,
3910 };
3911
3912 static void ixgbevf_assign_netdev_ops(struct net_device *dev)
3913 {
3914         dev->netdev_ops = &ixgbevf_netdev_ops;
3915         ixgbevf_set_ethtool_ops(dev);
3916         dev->watchdog_timeo = 5 * HZ;
3917 }
3918
3919 /**
3920  * ixgbevf_probe - Device Initialization Routine
3921  * @pdev: PCI device information struct
3922  * @ent: entry in ixgbevf_pci_tbl
3923  *
3924  * Returns 0 on success, negative on failure
3925  *
3926  * ixgbevf_probe initializes an adapter identified by a pci_dev structure.
3927  * The OS initialization, configuring of the adapter private structure,
3928  * and a hardware reset occur.
3929  **/
3930 static int ixgbevf_probe(struct pci_dev *pdev, const struct pci_device_id *ent)
3931 {
3932         struct net_device *netdev;
3933         struct ixgbevf_adapter *adapter = NULL;
3934         struct ixgbe_hw *hw = NULL;
3935         const struct ixgbevf_info *ii = ixgbevf_info_tbl[ent->driver_data];
3936         int err, pci_using_dac;
3937         bool disable_dev = false;
3938
3939         err = pci_enable_device(pdev);
3940         if (err)
3941                 return err;
3942
3943         if (!dma_set_mask_and_coherent(&pdev->dev, DMA_BIT_MASK(64))) {
3944                 pci_using_dac = 1;
3945         } else {
3946                 err = dma_set_mask_and_coherent(&pdev->dev, DMA_BIT_MASK(32));
3947                 if (err) {
3948                         dev_err(&pdev->dev, "No usable DMA configuration, aborting\n");
3949                         goto err_dma;
3950                 }
3951                 pci_using_dac = 0;
3952         }
3953
3954         err = pci_request_regions(pdev, ixgbevf_driver_name);
3955         if (err) {
3956                 dev_err(&pdev->dev, "pci_request_regions failed 0x%x\n", err);
3957                 goto err_pci_reg;
3958         }
3959
3960         pci_set_master(pdev);
3961
3962         netdev = alloc_etherdev_mq(sizeof(struct ixgbevf_adapter),
3963                                    MAX_TX_QUEUES);
3964         if (!netdev) {
3965                 err = -ENOMEM;
3966                 goto err_alloc_etherdev;
3967         }
3968
3969         SET_NETDEV_DEV(netdev, &pdev->dev);
3970
3971         adapter = netdev_priv(netdev);
3972
3973         adapter->netdev = netdev;
3974         adapter->pdev = pdev;
3975         hw = &adapter->hw;
3976         hw->back = adapter;
3977         adapter->msg_enable = netif_msg_init(debug, DEFAULT_MSG_ENABLE);
3978
3979         /* call save state here in standalone driver because it relies on
3980          * adapter struct to exist, and needs to call netdev_priv
3981          */
3982         pci_save_state(pdev);
3983
3984         hw->hw_addr = ioremap(pci_resource_start(pdev, 0),
3985                               pci_resource_len(pdev, 0));
3986         adapter->io_addr = hw->hw_addr;
3987         if (!hw->hw_addr) {
3988                 err = -EIO;
3989                 goto err_ioremap;
3990         }
3991
3992         ixgbevf_assign_netdev_ops(netdev);
3993
3994         /* Setup HW API */
3995         memcpy(&hw->mac.ops, ii->mac_ops, sizeof(hw->mac.ops));
3996         hw->mac.type  = ii->mac;
3997
3998         memcpy(&hw->mbx.ops, &ixgbevf_mbx_ops,
3999                sizeof(struct ixgbe_mbx_operations));
4000
4001         /* setup the private structure */
4002         err = ixgbevf_sw_init(adapter);
4003         if (err)
4004                 goto err_sw_init;
4005
4006         /* The HW MAC address was set and/or determined in sw_init */
4007         if (!is_valid_ether_addr(netdev->dev_addr)) {
4008                 pr_err("invalid MAC address\n");
4009                 err = -EIO;
4010                 goto err_sw_init;
4011         }
4012
4013         netdev->hw_features = NETIF_F_SG |
4014                               NETIF_F_IP_CSUM |
4015                               NETIF_F_IPV6_CSUM |
4016                               NETIF_F_TSO |
4017                               NETIF_F_TSO6 |
4018                               NETIF_F_RXCSUM;
4019
4020         netdev->features = netdev->hw_features |
4021                            NETIF_F_HW_VLAN_CTAG_TX |
4022                            NETIF_F_HW_VLAN_CTAG_RX |
4023                            NETIF_F_HW_VLAN_CTAG_FILTER;
4024
4025         netdev->vlan_features |= NETIF_F_TSO |
4026                                  NETIF_F_TSO6 |
4027                                  NETIF_F_IP_CSUM |
4028                                  NETIF_F_IPV6_CSUM |
4029                                  NETIF_F_SG;
4030
4031         if (pci_using_dac)
4032                 netdev->features |= NETIF_F_HIGHDMA;
4033
4034         netdev->priv_flags |= IFF_UNICAST_FLT;
4035
4036         if (IXGBE_REMOVED(hw->hw_addr)) {
4037                 err = -EIO;
4038                 goto err_sw_init;
4039         }
4040
4041         setup_timer(&adapter->service_timer, &ixgbevf_service_timer,
4042                     (unsigned long)adapter);
4043
4044         INIT_WORK(&adapter->service_task, ixgbevf_service_task);
4045         set_bit(__IXGBEVF_SERVICE_INITED, &adapter->state);
4046         clear_bit(__IXGBEVF_SERVICE_SCHED, &adapter->state);
4047
4048         err = ixgbevf_init_interrupt_scheme(adapter);
4049         if (err)
4050                 goto err_sw_init;
4051
4052         strcpy(netdev->name, "eth%d");
4053
4054         err = register_netdev(netdev);
4055         if (err)
4056                 goto err_register;
4057
4058         pci_set_drvdata(pdev, netdev);
4059         netif_carrier_off(netdev);
4060
4061         ixgbevf_init_last_counter_stats(adapter);
4062
4063         /* print the VF info */
4064         dev_info(&pdev->dev, "%pM\n", netdev->dev_addr);
4065         dev_info(&pdev->dev, "MAC: %d\n", hw->mac.type);
4066
4067         switch (hw->mac.type) {
4068         case ixgbe_mac_X550_vf:
4069                 dev_info(&pdev->dev, "Intel(R) X550 Virtual Function\n");
4070                 break;
4071         case ixgbe_mac_X540_vf:
4072                 dev_info(&pdev->dev, "Intel(R) X540 Virtual Function\n");
4073                 break;
4074         case ixgbe_mac_82599_vf:
4075         default:
4076                 dev_info(&pdev->dev, "Intel(R) 82599 Virtual Function\n");
4077                 break;
4078         }
4079
4080         return 0;
4081
4082 err_register:
4083         ixgbevf_clear_interrupt_scheme(adapter);
4084 err_sw_init:
4085         ixgbevf_reset_interrupt_capability(adapter);
4086         iounmap(adapter->io_addr);
4087 err_ioremap:
4088         disable_dev = !test_and_set_bit(__IXGBEVF_DISABLED, &adapter->state);
4089         free_netdev(netdev);
4090 err_alloc_etherdev:
4091         pci_release_regions(pdev);
4092 err_pci_reg:
4093 err_dma:
4094         if (!adapter || disable_dev)
4095                 pci_disable_device(pdev);
4096         return err;
4097 }
4098
4099 /**
4100  * ixgbevf_remove - Device Removal Routine
4101  * @pdev: PCI device information struct
4102  *
4103  * ixgbevf_remove is called by the PCI subsystem to alert the driver
4104  * that it should release a PCI device.  The could be caused by a
4105  * Hot-Plug event, or because the driver is going to be removed from
4106  * memory.
4107  **/
4108 static void ixgbevf_remove(struct pci_dev *pdev)
4109 {
4110         struct net_device *netdev = pci_get_drvdata(pdev);
4111         struct ixgbevf_adapter *adapter;
4112         bool disable_dev;
4113
4114         if (!netdev)
4115                 return;
4116
4117         adapter = netdev_priv(netdev);
4118
4119         set_bit(__IXGBEVF_REMOVING, &adapter->state);
4120         cancel_work_sync(&adapter->service_task);
4121
4122         if (netdev->reg_state == NETREG_REGISTERED)
4123                 unregister_netdev(netdev);
4124
4125         ixgbevf_clear_interrupt_scheme(adapter);
4126         ixgbevf_reset_interrupt_capability(adapter);
4127
4128         iounmap(adapter->io_addr);
4129         pci_release_regions(pdev);
4130
4131         hw_dbg(&adapter->hw, "Remove complete\n");
4132
4133         disable_dev = !test_and_set_bit(__IXGBEVF_DISABLED, &adapter->state);
4134         free_netdev(netdev);
4135
4136         if (disable_dev)
4137                 pci_disable_device(pdev);
4138 }
4139
4140 /**
4141  * ixgbevf_io_error_detected - called when PCI error is detected
4142  * @pdev: Pointer to PCI device
4143  * @state: The current pci connection state
4144  *
4145  * This function is called after a PCI bus error affecting
4146  * this device has been detected.
4147  **/
4148 static pci_ers_result_t ixgbevf_io_error_detected(struct pci_dev *pdev,
4149                                                   pci_channel_state_t state)
4150 {
4151         struct net_device *netdev = pci_get_drvdata(pdev);
4152         struct ixgbevf_adapter *adapter = netdev_priv(netdev);
4153
4154         if (!test_bit(__IXGBEVF_SERVICE_INITED, &adapter->state))
4155                 return PCI_ERS_RESULT_DISCONNECT;
4156
4157         rtnl_lock();
4158         netif_device_detach(netdev);
4159
4160         if (state == pci_channel_io_perm_failure) {
4161                 rtnl_unlock();
4162                 return PCI_ERS_RESULT_DISCONNECT;
4163         }
4164
4165         if (netif_running(netdev))
4166                 ixgbevf_down(adapter);
4167
4168         if (!test_and_set_bit(__IXGBEVF_DISABLED, &adapter->state))
4169                 pci_disable_device(pdev);
4170         rtnl_unlock();
4171
4172         /* Request a slot slot reset. */
4173         return PCI_ERS_RESULT_NEED_RESET;
4174 }
4175
4176 /**
4177  * ixgbevf_io_slot_reset - called after the pci bus has been reset.
4178  * @pdev: Pointer to PCI device
4179  *
4180  * Restart the card from scratch, as if from a cold-boot. Implementation
4181  * resembles the first-half of the ixgbevf_resume routine.
4182  **/
4183 static pci_ers_result_t ixgbevf_io_slot_reset(struct pci_dev *pdev)
4184 {
4185         struct net_device *netdev = pci_get_drvdata(pdev);
4186         struct ixgbevf_adapter *adapter = netdev_priv(netdev);
4187
4188         if (pci_enable_device_mem(pdev)) {
4189                 dev_err(&pdev->dev,
4190                         "Cannot re-enable PCI device after reset.\n");
4191                 return PCI_ERS_RESULT_DISCONNECT;
4192         }
4193
4194         smp_mb__before_atomic();
4195         clear_bit(__IXGBEVF_DISABLED, &adapter->state);
4196         pci_set_master(pdev);
4197
4198         ixgbevf_reset(adapter);
4199
4200         return PCI_ERS_RESULT_RECOVERED;
4201 }
4202
4203 /**
4204  * ixgbevf_io_resume - called when traffic can start flowing again.
4205  * @pdev: Pointer to PCI device
4206  *
4207  * This callback is called when the error recovery driver tells us that
4208  * its OK to resume normal operation. Implementation resembles the
4209  * second-half of the ixgbevf_resume routine.
4210  **/
4211 static void ixgbevf_io_resume(struct pci_dev *pdev)
4212 {
4213         struct net_device *netdev = pci_get_drvdata(pdev);
4214         struct ixgbevf_adapter *adapter = netdev_priv(netdev);
4215
4216         if (netif_running(netdev))
4217                 ixgbevf_up(adapter);
4218
4219         netif_device_attach(netdev);
4220 }
4221
4222 /* PCI Error Recovery (ERS) */
4223 static const struct pci_error_handlers ixgbevf_err_handler = {
4224         .error_detected = ixgbevf_io_error_detected,
4225         .slot_reset = ixgbevf_io_slot_reset,
4226         .resume = ixgbevf_io_resume,
4227 };
4228
4229 static struct pci_driver ixgbevf_driver = {
4230         .name           = ixgbevf_driver_name,
4231         .id_table       = ixgbevf_pci_tbl,
4232         .probe          = ixgbevf_probe,
4233         .remove         = ixgbevf_remove,
4234 #ifdef CONFIG_PM
4235         /* Power Management Hooks */
4236         .suspend        = ixgbevf_suspend,
4237         .resume         = ixgbevf_resume,
4238 #endif
4239         .shutdown       = ixgbevf_shutdown,
4240         .err_handler    = &ixgbevf_err_handler
4241 };
4242
4243 /**
4244  * ixgbevf_init_module - Driver Registration Routine
4245  *
4246  * ixgbevf_init_module is the first routine called when the driver is
4247  * loaded. All it does is register with the PCI subsystem.
4248  **/
4249 static int __init ixgbevf_init_module(void)
4250 {
4251         int ret;
4252
4253         pr_info("%s - version %s\n", ixgbevf_driver_string,
4254                 ixgbevf_driver_version);
4255
4256         pr_info("%s\n", ixgbevf_copyright);
4257
4258         ret = pci_register_driver(&ixgbevf_driver);
4259         return ret;
4260 }
4261
4262 module_init(ixgbevf_init_module);
4263
4264 /**
4265  * ixgbevf_exit_module - Driver Exit Cleanup Routine
4266  *
4267  * ixgbevf_exit_module is called just before the driver is removed
4268  * from memory.
4269  **/
4270 static void __exit ixgbevf_exit_module(void)
4271 {
4272         pci_unregister_driver(&ixgbevf_driver);
4273 }
4274
4275 #ifdef DEBUG
4276 /**
4277  * ixgbevf_get_hw_dev_name - return device name string
4278  * used by hardware layer to print debugging information
4279  **/
4280 char *ixgbevf_get_hw_dev_name(struct ixgbe_hw *hw)
4281 {
4282         struct ixgbevf_adapter *adapter = hw->back;
4283
4284         return adapter->netdev->name;
4285 }
4286
4287 #endif
4288 module_exit(ixgbevf_exit_module);
4289
4290 /* ixgbevf_main.c */