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

   1 /*******************************************************************************
   2
   3   Intel 82599 Virtual Function driver
   4   Copyright(c) 1999 - 2015 Intel Corporation.
   5
   6   This program is free software; you can redistribute it and/or modify it
   7   under the terms and conditions of the GNU General Public License,
   8   version 2, as published by the Free Software Foundation.
   9
  10   This program is distributed in the hope it will be useful, but WITHOUT
  11   ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
  12   FITNESS FOR A PARTICULAR PURPOSE.  See the GNU General Public License for
  13   more details.
  14
  15   You should have received a copy of the GNU General Public License along with
  16   this program; if not, see <http://www.gnu.org/licenses/>.
  17
  18   The full GNU General Public License is included in this distribution in
  19   the file called "COPYING".
  20
  21   Contact Information:
  22   e1000-devel Mailing List <e1000-devel@lists.sourceforge.net>
  23   Intel Corporation, 5200 N.E. Elam Young Parkway, Hillsboro, OR 97124-6497
  24
  25 *******************************************************************************/
  26
  27 /******************************************************************************
  28  Copyright (c)2006 - 2007 Myricom, Inc. for some LRO specific code
  29 ******************************************************************************/
  30
  31 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
  32
  33 #include <linux/types.h>
  34 #include <linux/bitops.h>
  35 #include <linux/module.h>
  36 #include <linux/pci.h>
  37 #include <linux/netdevice.h>
  38 #include <linux/vmalloc.h>
  39 #include <linux/string.h>
  40 #include <linux/in.h>
  41 #include <linux/ip.h>
  42 #include <linux/tcp.h>
  43 #include <linux/sctp.h>
  44 #include <linux/ipv6.h>
  45 #include <linux/slab.h>
  46 #include <net/checksum.h>
  47 #include <net/ip6_checksum.h>
  48 #include <linux/ethtool.h>
  49 #include <linux/if.h>
  50 #include <linux/if_vlan.h>
  51 #include <linux/prefetch.h>
  52
  53 #include "ixgbevf.h"
  54
  55 const char ixgbevf_driver_name[] = "ixgbevf";
  56 static const char ixgbevf_driver_string[] =
  57         "Intel(R) 10 Gigabit PCI Express Virtual Function Network Driver";
  58
  59 #define DRV_VERSION "2.12.1-k"
  60 const char ixgbevf_driver_version[] = DRV_VERSION;
  61 static char ixgbevf_copyright[] =
  62         "Copyright (c) 2009 - 2015 Intel Corporation.";
  63
  64 static const struct ixgbevf_info *ixgbevf_info_tbl[] = {
  65         [board_82599_vf] = &ixgbevf_82599_vf_info,
  66         [board_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 struct workqueue_struct *ixgbevf_wq;
 100
 101 static void ixgbevf_service_event_schedule(struct ixgbevf_adapter *adapter)
 102 {
 103         if (!test_bit(__IXGBEVF_DOWN, &adapter->state) &&
 104             !test_bit(__IXGBEVF_REMOVING, &adapter->state) &&
 105             !test_and_set_bit(__IXGBEVF_SERVICE_SCHED, &adapter->state))
 106                 queue_work(ixgbevf_wq, &adapter->service_task);
 107 }
 108
 109 static void ixgbevf_service_event_complete(struct ixgbevf_adapter *adapter)
 110 {
 111         BUG_ON(!test_bit(__IXGBEVF_SERVICE_SCHED, &adapter->state));
 112
 113         /* flush memory to make sure state is correct before next watchdog */
 114         smp_mb__before_atomic();
 115         clear_bit(__IXGBEVF_SERVICE_SCHED, &adapter->state);
 116 }
 117
 118 /* forward decls */
 119 static void ixgbevf_queue_reset_subtask(struct ixgbevf_adapter *adapter);
 120 static void ixgbevf_set_itr(struct ixgbevf_q_vector *q_vector);
 121 static void ixgbevf_free_all_rx_resources(struct ixgbevf_adapter *adapter);
 122
 123 static void ixgbevf_remove_adapter(struct ixgbe_hw *hw)
 124 {
 125         struct ixgbevf_adapter *adapter = hw->back;
 126
 127         if (!hw->hw_addr)
 128                 return;
 129         hw->hw_addr = NULL;
 130         dev_err(&adapter->pdev->dev, "Adapter removed\n");
 131         if (test_bit(__IXGBEVF_SERVICE_INITED, &adapter->state))
 132                 ixgbevf_service_event_schedule(adapter);
 133 }
 134
 135 static void ixgbevf_check_remove(struct ixgbe_hw *hw, u32 reg)
 136 {
 137         u32 value;
 138
 139         /* The following check not only optimizes a bit by not
 140          * performing a read on the status register when the
 141          * register just read was a status register read that
 142          * returned IXGBE_FAILED_READ_REG. It also blocks any
 143          * potential recursion.
 144          */
 145         if (reg == IXGBE_VFSTATUS) {
 146                 ixgbevf_remove_adapter(hw);
 147                 return;
 148         }
 149         value = ixgbevf_read_reg(hw, IXGBE_VFSTATUS);
 150         if (value == IXGBE_FAILED_READ_REG)
 151                 ixgbevf_remove_adapter(hw);
 152 }
 153
 154 u32 ixgbevf_read_reg(struct ixgbe_hw *hw, u32 reg)
 155 {
 156         u8 __iomem *reg_addr = ACCESS_ONCE(hw->hw_addr);
 157         u32 value;
 158
 159         if (IXGBE_REMOVED(reg_addr))
 160                 return IXGBE_FAILED_READ_REG;
 161         value = readl(reg_addr + reg);
 162         if (unlikely(value == IXGBE_FAILED_READ_REG))
 163                 ixgbevf_check_remove(hw, reg);
 164         return value;
 165 }
 166
 167 /**
 168  * ixgbevf_set_ivar - set IVAR registers - maps interrupt causes to vectors
 169  * @adapter: pointer to adapter struct
 170  * @direction: 0 for Rx, 1 for Tx, -1 for other causes
 171  * @queue: queue to map the corresponding interrupt to
 172  * @msix_vector: the vector to map to the corresponding queue
 173  **/
 174 static void ixgbevf_set_ivar(struct ixgbevf_adapter *adapter, s8 direction,
 175                              u8 queue, u8 msix_vector)
 176 {
 177         u32 ivar, index;
 178         struct ixgbe_hw *hw = &adapter->hw;
 179
 180         if (direction == -1) {
 181                 /* other causes */
 182                 msix_vector |= IXGBE_IVAR_ALLOC_VAL;
 183                 ivar = IXGBE_READ_REG(hw, IXGBE_VTIVAR_MISC);
 184                 ivar &= ~0xFF;
 185                 ivar |= msix_vector;
 186                 IXGBE_WRITE_REG(hw, IXGBE_VTIVAR_MISC, ivar);
 187         } else {
 188                 /* Tx or Rx causes */
 189                 msix_vector |= IXGBE_IVAR_ALLOC_VAL;
 190                 index = ((16 * (queue & 1)) + (8 * direction));
 191                 ivar = IXGBE_READ_REG(hw, IXGBE_VTIVAR(queue >> 1));
 192                 ivar &= ~(0xFF << index);
 193                 ivar |= (msix_vector << index);
 194                 IXGBE_WRITE_REG(hw, IXGBE_VTIVAR(queue >> 1), ivar);
 195         }
 196 }
 197
 198 static void ixgbevf_unmap_and_free_tx_resource(struct ixgbevf_ring *tx_ring,
 199                                         struct ixgbevf_tx_buffer *tx_buffer)
 200 {
 201         if (tx_buffer->skb) {
 202                 dev_kfree_skb_any(tx_buffer->skb);
 203                 if (dma_unmap_len(tx_buffer, len))
 204                         dma_unmap_single(tx_ring->dev,
 205                                          dma_unmap_addr(tx_buffer, dma),
 206                                          dma_unmap_len(tx_buffer, len),
 207                                          DMA_TO_DEVICE);
 208         } else if (dma_unmap_len(tx_buffer, len)) {
 209                 dma_unmap_page(tx_ring->dev,
 210                                dma_unmap_addr(tx_buffer, dma),
 211                                dma_unmap_len(tx_buffer, len),
 212                                DMA_TO_DEVICE);
 213         }
 214         tx_buffer->next_to_watch = NULL;
 215         tx_buffer->skb = NULL;
 216         dma_unmap_len_set(tx_buffer, len, 0);
 217         /* tx_buffer must be completely set up in the transmit path */
 218 }
 219
 220 static u64 ixgbevf_get_tx_completed(struct ixgbevf_ring *ring)
 221 {
 222         return ring->stats.packets;
 223 }
 224
 225 static u32 ixgbevf_get_tx_pending(struct ixgbevf_ring *ring)
 226 {
 227         struct ixgbevf_adapter *adapter = netdev_priv(ring->netdev);
 228         struct ixgbe_hw *hw = &adapter->hw;
 229
 230         u32 head = IXGBE_READ_REG(hw, IXGBE_VFTDH(ring->reg_idx));
 231         u32 tail = IXGBE_READ_REG(hw, IXGBE_VFTDT(ring->reg_idx));
 232
 233         if (head != tail)
 234                 return (head < tail) ?
 235                         tail - head : (tail + ring->count - head);
 236
 237         return 0;
 238 }
 239
 240 static inline bool ixgbevf_check_tx_hang(struct ixgbevf_ring *tx_ring)
 241 {
 242         u32 tx_done = ixgbevf_get_tx_completed(tx_ring);
 243         u32 tx_done_old = tx_ring->tx_stats.tx_done_old;
 244         u32 tx_pending = ixgbevf_get_tx_pending(tx_ring);
 245
 246         clear_check_for_tx_hang(tx_ring);
 247
 248         /* Check for a hung queue, but be thorough. This verifies
 249          * that a transmit has been completed since the previous
 250          * check AND there is at least one packet pending. The
 251          * ARMED bit is set to indicate a potential hang.
 252          */
 253         if ((tx_done_old == tx_done) && tx_pending) {
 254                 /* make sure it is true for two checks in a row */
 255                 return test_and_set_bit(__IXGBEVF_HANG_CHECK_ARMED,
 256                                         &tx_ring->state);
 257         }
 258         /* reset the countdown */
 259         clear_bit(__IXGBEVF_HANG_CHECK_ARMED, &tx_ring->state);
 260
 261         /* update completed stats and continue */
 262         tx_ring->tx_stats.tx_done_old = tx_done;
 263
 264         return false;
 265 }
 266
 267 static void ixgbevf_tx_timeout_reset(struct ixgbevf_adapter *adapter)
 268 {
 269         /* Do the reset outside of interrupt context */
 270         if (!test_bit(__IXGBEVF_DOWN, &adapter->state)) {
 271                 adapter->flags |= IXGBEVF_FLAG_RESET_REQUESTED;
 272                 ixgbevf_service_event_schedule(adapter);
 273         }
 274 }
 275
 276 /**
 277  * ixgbevf_tx_timeout - Respond to a Tx Hang
 278  * @netdev: network interface device structure
 279  **/
 280 static void ixgbevf_tx_timeout(struct net_device *netdev)
 281 {
 282         struct ixgbevf_adapter *adapter = netdev_priv(netdev);
 283
 284         ixgbevf_tx_timeout_reset(adapter);
 285 }
 286
 287 /**
 288  * ixgbevf_clean_tx_irq - Reclaim resources after transmit completes
 289  * @q_vector: board private structure
 290  * @tx_ring: tx ring to clean
 291  **/
 292 static bool ixgbevf_clean_tx_irq(struct ixgbevf_q_vector *q_vector,
 293                                  struct ixgbevf_ring *tx_ring)
 294 {
 295         struct ixgbevf_adapter *adapter = q_vector->adapter;
 296         struct ixgbevf_tx_buffer *tx_buffer;
 297         union ixgbe_adv_tx_desc *tx_desc;
 298         unsigned int total_bytes = 0, total_packets = 0;
 299         unsigned int budget = tx_ring->count / 2;
 300         unsigned int i = tx_ring->next_to_clean;
 301
 302         if (test_bit(__IXGBEVF_DOWN, &adapter->state))
 303                 return true;
 304
 305         tx_buffer = &tx_ring->tx_buffer_info[i];
 306         tx_desc = IXGBEVF_TX_DESC(tx_ring, i);
 307         i -= tx_ring->count;
 308
 309         do {
 310                 union ixgbe_adv_tx_desc *eop_desc = tx_buffer->next_to_watch;
 311
 312                 /* if next_to_watch is not set then there is no work pending */
 313                 if (!eop_desc)
 314                         break;
 315
 316                 /* prevent any other reads prior to eop_desc */
 317                 read_barrier_depends();
 318
 319                 /* if DD is not set pending work has not been completed */
 320                 if (!(eop_desc->wb.status & cpu_to_le32(IXGBE_TXD_STAT_DD)))
 321                         break;
 322
 323                 /* clear next_to_watch to prevent false hangs */
 324                 tx_buffer->next_to_watch = NULL;
 325
 326                 /* update the statistics for this packet */
 327                 total_bytes += tx_buffer->bytecount;
 328                 total_packets += tx_buffer->gso_segs;
 329
 330                 /* free the skb */
 331                 dev_kfree_skb_any(tx_buffer->skb);
 332
 333                 /* unmap skb header data */
 334                 dma_unmap_single(tx_ring->dev,
 335                                  dma_unmap_addr(tx_buffer, dma),
 336                                  dma_unmap_len(tx_buffer, len),
 337                                  DMA_TO_DEVICE);
 338
 339                 /* clear tx_buffer data */
 340                 tx_buffer->skb = NULL;
 341                 dma_unmap_len_set(tx_buffer, len, 0);
 342
 343                 /* unmap remaining buffers */
 344                 while (tx_desc != eop_desc) {
 345                         tx_buffer++;
 346                         tx_desc++;
 347                         i++;
 348                         if (unlikely(!i)) {
 349                                 i -= tx_ring->count;
 350                                 tx_buffer = tx_ring->tx_buffer_info;
 351                                 tx_desc = IXGBEVF_TX_DESC(tx_ring, 0);
 352                         }
 353
 354                         /* unmap any remaining paged data */
 355                         if (dma_unmap_len(tx_buffer, len)) {
 356                                 dma_unmap_page(tx_ring->dev,
 357                                                dma_unmap_addr(tx_buffer, dma),
 358                                                dma_unmap_len(tx_buffer, len),
 359                                                DMA_TO_DEVICE);
 360                                 dma_unmap_len_set(tx_buffer, len, 0);
 361                         }
 362                 }
 363
 364                 /* move us one more past the eop_desc for start of next pkt */
 365                 tx_buffer++;
 366                 tx_desc++;
 367                 i++;
 368                 if (unlikely(!i)) {
 369                         i -= tx_ring->count;
 370                         tx_buffer = tx_ring->tx_buffer_info;
 371                         tx_desc = IXGBEVF_TX_DESC(tx_ring, 0);
 372                 }
 373
 374                 /* issue prefetch for next Tx descriptor */
 375                 prefetch(tx_desc);
 376
 377                 /* update budget accounting */
 378                 budget--;
 379         } while (likely(budget));
 380
 381         i += tx_ring->count;
 382         tx_ring->next_to_clean = i;
 383         u64_stats_update_begin(&tx_ring->syncp);
 384         tx_ring->stats.bytes += total_bytes;
 385         tx_ring->stats.packets += total_packets;
 386         u64_stats_update_end(&tx_ring->syncp);
 387         q_vector->tx.total_bytes += total_bytes;
 388         q_vector->tx.total_packets += total_packets;
 389
 390         if (check_for_tx_hang(tx_ring) && ixgbevf_check_tx_hang(tx_ring)) {
 391                 struct ixgbe_hw *hw = &adapter->hw;
 392                 union ixgbe_adv_tx_desc *eop_desc;
 393
 394                 eop_desc = tx_ring->tx_buffer_info[i].next_to_watch;
 395
 396                 pr_err("Detected Tx Unit Hang\n"
 397                        "  Tx Queue             <%d>\n"
 398                        "  TDH, TDT             <%x>, <%x>\n"
 399                        "  next_to_use          <%x>\n"
 400                        "  next_to_clean        <%x>\n"
 401                        "tx_buffer_info[next_to_clean]\n"
 402                        "  next_to_watch        <%p>\n"
 403                        "  eop_desc->wb.status  <%x>\n"
 404                        "  time_stamp           <%lx>\n"
 405                        "  jiffies              <%lx>\n",
 406                        tx_ring->queue_index,
 407                        IXGBE_READ_REG(hw, IXGBE_VFTDH(tx_ring->reg_idx)),
 408                        IXGBE_READ_REG(hw, IXGBE_VFTDT(tx_ring->reg_idx)),
 409                        tx_ring->next_to_use, i,
 410                        eop_desc, (eop_desc ? eop_desc->wb.status : 0),
 411                        tx_ring->tx_buffer_info[i].time_stamp, jiffies);
 412
 413                 netif_stop_subqueue(tx_ring->netdev, tx_ring->queue_index);
 414
 415                 /* schedule immediate reset if we believe we hung */
 416                 ixgbevf_tx_timeout_reset(adapter);
 417
 418                 return true;
 419         }
 420
 421 #define TX_WAKE_THRESHOLD (DESC_NEEDED * 2)
 422         if (unlikely(total_packets && netif_carrier_ok(tx_ring->netdev) &&
 423                      (ixgbevf_desc_unused(tx_ring) >= TX_WAKE_THRESHOLD))) {
 424                 /* Make sure that anybody stopping the queue after this
 425                  * sees the new next_to_clean.
 426                  */
 427                 smp_mb();
 428
 429                 if (__netif_subqueue_stopped(tx_ring->netdev,
 430                                              tx_ring->queue_index) &&
 431                     !test_bit(__IXGBEVF_DOWN, &adapter->state)) {
 432                         netif_wake_subqueue(tx_ring->netdev,
 433                                             tx_ring->queue_index);
 434                         ++tx_ring->tx_stats.restart_queue;
 435                 }
 436         }
 437
 438         return !!budget;
 439 }
 440
 441 /**
 442  * ixgbevf_rx_skb - Helper function to determine proper Rx method
 443  * @q_vector: structure containing interrupt and ring information
 444  * @skb: packet to send up
 445  **/
 446 static void ixgbevf_rx_skb(struct ixgbevf_q_vector *q_vector,
 447                            struct sk_buff *skb)
 448 {
 449 #ifdef CONFIG_NET_RX_BUSY_POLL
 450         skb_mark_napi_id(skb, &q_vector->napi);
 451
 452         if (ixgbevf_qv_busy_polling(q_vector)) {
 453                 netif_receive_skb(skb);
 454                 /* exit early if we busy polled */
 455                 return;
 456         }
 457 #endif /* CONFIG_NET_RX_BUSY_POLL */
 458
 459         napi_gro_receive(&q_vector->napi, skb);
 460 }
 461
 462 #define IXGBE_RSS_L4_TYPES_MASK \
 463         ((1ul << IXGBE_RXDADV_RSSTYPE_IPV4_TCP) | \
 464          (1ul << IXGBE_RXDADV_RSSTYPE_IPV4_UDP) | \
 465          (1ul << IXGBE_RXDADV_RSSTYPE_IPV6_TCP) | \
 466          (1ul << IXGBE_RXDADV_RSSTYPE_IPV6_UDP))
 467
 468 static inline void ixgbevf_rx_hash(struct ixgbevf_ring *ring,
 469                                    union ixgbe_adv_rx_desc *rx_desc,
 470                                    struct sk_buff *skb)
 471 {
 472         u16 rss_type;
 473
 474         if (!(ring->netdev->features & NETIF_F_RXHASH))
 475                 return;
 476
 477         rss_type = le16_to_cpu(rx_desc->wb.lower.lo_dword.hs_rss.pkt_info) &
 478                    IXGBE_RXDADV_RSSTYPE_MASK;
 479
 480         if (!rss_type)
 481                 return;
 482
 483         skb_set_hash(skb, le32_to_cpu(rx_desc->wb.lower.hi_dword.rss),
 484                      (IXGBE_RSS_L4_TYPES_MASK & (1ul << rss_type)) ?
 485                      PKT_HASH_TYPE_L4 : PKT_HASH_TYPE_L3);
 486 }
 487
 488 /**
 489  * ixgbevf_rx_checksum - indicate in skb if hw indicated a good cksum
 490  * @ring: structure containig ring specific data
 491  * @rx_desc: current Rx descriptor being processed
 492  * @skb: skb currently being received and modified
 493  **/
 494 static inline void ixgbevf_rx_checksum(struct ixgbevf_ring *ring,
 495                                        union ixgbe_adv_rx_desc *rx_desc,
 496                                        struct sk_buff *skb)
 497 {
 498         skb_checksum_none_assert(skb);
 499
 500         /* Rx csum disabled */
 501         if (!(ring->netdev->features & NETIF_F_RXCSUM))
 502                 return;
 503
 504         /* if IP and error */
 505         if (ixgbevf_test_staterr(rx_desc, IXGBE_RXD_STAT_IPCS) &&
 506             ixgbevf_test_staterr(rx_desc, IXGBE_RXDADV_ERR_IPE)) {
 507                 ring->rx_stats.csum_err++;
 508                 return;
 509         }
 510
 511         if (!ixgbevf_test_staterr(rx_desc, IXGBE_RXD_STAT_L4CS))
 512                 return;
 513
 514         if (ixgbevf_test_staterr(rx_desc, IXGBE_RXDADV_ERR_TCPE)) {
 515                 ring->rx_stats.csum_err++;
 516                 return;
 517         }
 518
 519         /* It must be a TCP or UDP packet with a valid checksum */
 520         skb->ip_summed = CHECKSUM_UNNECESSARY;
 521 }
 522
 523 /**
 524  * ixgbevf_process_skb_fields - Populate skb header fields from Rx descriptor
 525  * @rx_ring: rx descriptor ring packet is being transacted on
 526  * @rx_desc: pointer to the EOP Rx descriptor
 527  * @skb: pointer to current skb being populated
 528  *
 529  * This function checks the ring, descriptor, and packet information in
 530  * order to populate the checksum, VLAN, protocol, and other fields within
 531  * the skb.
 532  **/
 533 static void ixgbevf_process_skb_fields(struct ixgbevf_ring *rx_ring,
 534                                        union ixgbe_adv_rx_desc *rx_desc,
 535                                        struct sk_buff *skb)
 536 {
 537         ixgbevf_rx_hash(rx_ring, rx_desc, skb);
 538         ixgbevf_rx_checksum(rx_ring, rx_desc, skb);
 539
 540         if (ixgbevf_test_staterr(rx_desc, IXGBE_RXD_STAT_VP)) {
 541                 u16 vid = le16_to_cpu(rx_desc->wb.upper.vlan);
 542                 unsigned long *active_vlans = netdev_priv(rx_ring->netdev);
 543
 544                 if (test_bit(vid & VLAN_VID_MASK, active_vlans))
 545                         __vlan_hwaccel_put_tag(skb, htons(ETH_P_8021Q), vid);
 546         }
 547
 548         skb->protocol = eth_type_trans(skb, rx_ring->netdev);
 549 }
 550
 551 /**
 552  * ixgbevf_is_non_eop - process handling of non-EOP buffers
 553  * @rx_ring: Rx ring being processed
 554  * @rx_desc: Rx descriptor for current buffer
 555  * @skb: current socket buffer containing buffer in progress
 556  *
 557  * This function updates next to clean.  If the buffer is an EOP buffer
 558  * this function exits returning false, otherwise it will place the
 559  * sk_buff in the next buffer to be chained and return true indicating
 560  * that this is in fact a non-EOP buffer.
 561  **/
 562 static bool ixgbevf_is_non_eop(struct ixgbevf_ring *rx_ring,
 563                                union ixgbe_adv_rx_desc *rx_desc)
 564 {
 565         u32 ntc = rx_ring->next_to_clean + 1;
 566
 567         /* fetch, update, and store next to clean */
 568         ntc = (ntc < rx_ring->count) ? ntc : 0;
 569         rx_ring->next_to_clean = ntc;
 570
 571         prefetch(IXGBEVF_RX_DESC(rx_ring, ntc));
 572
 573         if (likely(ixgbevf_test_staterr(rx_desc, IXGBE_RXD_STAT_EOP)))
 574                 return false;
 575
 576         return true;
 577 }
 578
 579 static bool ixgbevf_alloc_mapped_page(struct ixgbevf_ring *rx_ring,
 580                                       struct ixgbevf_rx_buffer *bi)
 581 {
 582         struct page *page = bi->page;
 583         dma_addr_t dma = bi->dma;
 584
 585         /* since we are recycling buffers we should seldom need to alloc */
 586         if (likely(page))
 587                 return true;
 588
 589         /* alloc new page for storage */
 590         page = dev_alloc_page();
 591         if (unlikely(!page)) {
 592                 rx_ring->rx_stats.alloc_rx_page_failed++;
 593                 return false;
 594         }
 595
 596         /* map page for use */
 597         dma = dma_map_page(rx_ring->dev, page, 0,
 598                            PAGE_SIZE, DMA_FROM_DEVICE);
 599
 600         /* if mapping failed free memory back to system since
 601          * there isn't much point in holding memory we can't use
 602          */
 603         if (dma_mapping_error(rx_ring->dev, dma)) {
 604                 __free_page(page);
 605
 606                 rx_ring->rx_stats.alloc_rx_buff_failed++;
 607                 return false;
 608         }
 609
 610         bi->dma = dma;
 611         bi->page = page;
 612         bi->page_offset = 0;
 613
 614         return true;
 615 }
 616
 617 /**
 618  * ixgbevf_alloc_rx_buffers - Replace used receive buffers; packet split
 619  * @rx_ring: rx descriptor ring (for a specific queue) to setup buffers on
 620  * @cleaned_count: number of buffers to replace
 621  **/
 622 static void ixgbevf_alloc_rx_buffers(struct ixgbevf_ring *rx_ring,
 623                                      u16 cleaned_count)
 624 {
 625         union ixgbe_adv_rx_desc *rx_desc;
 626         struct ixgbevf_rx_buffer *bi;
 627         unsigned int i = rx_ring->next_to_use;
 628
 629         /* nothing to do or no valid netdev defined */
 630         if (!cleaned_count || !rx_ring->netdev)
 631                 return;
 632
 633         rx_desc = IXGBEVF_RX_DESC(rx_ring, i);
 634         bi = &rx_ring->rx_buffer_info[i];
 635         i -= rx_ring->count;
 636
 637         do {
 638                 if (!ixgbevf_alloc_mapped_page(rx_ring, bi))
 639                         break;
 640
 641                 /* Refresh the desc even if pkt_addr didn't change
 642                  * because each write-back erases this info.
 643                  */
 644                 rx_desc->read.pkt_addr = cpu_to_le64(bi->dma + bi->page_offset);
 645
 646                 rx_desc++;
 647                 bi++;
 648                 i++;
 649                 if (unlikely(!i)) {
 650                         rx_desc = IXGBEVF_RX_DESC(rx_ring, 0);
 651                         bi = rx_ring->rx_buffer_info;
 652                         i -= rx_ring->count;
 653                 }
 654
 655                 /* clear the hdr_addr for the next_to_use descriptor */
 656                 rx_desc->read.hdr_addr = 0;
 657
 658                 cleaned_count--;
 659         } while (cleaned_count);
 660
 661         i += rx_ring->count;
 662
 663         if (rx_ring->next_to_use != i) {
 664                 /* record the next descriptor to use */
 665                 rx_ring->next_to_use = i;
 666
 667                 /* update next to alloc since we have filled the ring */
 668                 rx_ring->next_to_alloc = i;
 669
 670                 /* Force memory writes to complete before letting h/w
 671                  * know there are new descriptors to fetch.  (Only
 672                  * applicable for weak-ordered memory model archs,
 673                  * such as IA-64).
 674                  */
 675                 wmb();
 676                 ixgbevf_write_tail(rx_ring, i);
 677         }
 678 }
 679
 680 /**
 681  * ixgbevf_cleanup_headers - Correct corrupted or empty headers
 682  * @rx_ring: rx descriptor ring packet is being transacted on
 683  * @rx_desc: pointer to the EOP Rx descriptor
 684  * @skb: pointer to current skb being fixed
 685  *
 686  * Check for corrupted packet headers caused by senders on the local L2
 687  * embedded NIC switch not setting up their Tx Descriptors right.  These
 688  * should be very rare.
 689  *
 690  * Also address the case where we are pulling data in on pages only
 691  * and as such no data is present in the skb header.
 692  *
 693  * In addition if skb is not at least 60 bytes we need to pad it so that
 694  * it is large enough to qualify as a valid Ethernet frame.
 695  *
 696  * Returns true if an error was encountered and skb was freed.
 697  **/
 698 static bool ixgbevf_cleanup_headers(struct ixgbevf_ring *rx_ring,
 699                                     union ixgbe_adv_rx_desc *rx_desc,
 700                                     struct sk_buff *skb)
 701 {
 702         /* verify that the packet does not have any known errors */
 703         if (unlikely(ixgbevf_test_staterr(rx_desc,
 704                                           IXGBE_RXDADV_ERR_FRAME_ERR_MASK))) {
 705                 struct net_device *netdev = rx_ring->netdev;
 706
 707                 if (!(netdev->features & NETIF_F_RXALL)) {
 708                         dev_kfree_skb_any(skb);
 709                         return true;
 710                 }
 711         }
 712
 713         /* if eth_skb_pad returns an error the skb was freed */
 714         if (eth_skb_pad(skb))
 715                 return true;
 716
 717         return false;
 718 }
 719
 720 /**
 721  * ixgbevf_reuse_rx_page - page flip buffer and store it back on the ring
 722  * @rx_ring: rx descriptor ring to store buffers on
 723  * @old_buff: donor buffer to have page reused
 724  *
 725  * Synchronizes page for reuse by the adapter
 726  **/
 727 static void ixgbevf_reuse_rx_page(struct ixgbevf_ring *rx_ring,
 728                                   struct ixgbevf_rx_buffer *old_buff)
 729 {
 730         struct ixgbevf_rx_buffer *new_buff;
 731         u16 nta = rx_ring->next_to_alloc;
 732
 733         new_buff = &rx_ring->rx_buffer_info[nta];
 734
 735         /* update, and store next to alloc */
 736         nta++;
 737         rx_ring->next_to_alloc = (nta < rx_ring->count) ? nta : 0;
 738
 739         /* transfer page from old buffer to new buffer */
 740         new_buff->page = old_buff->page;
 741         new_buff->dma = old_buff->dma;
 742         new_buff->page_offset = old_buff->page_offset;
 743
 744         /* sync the buffer for use by the device */
 745         dma_sync_single_range_for_device(rx_ring->dev, new_buff->dma,
 746                                          new_buff->page_offset,
 747                                          IXGBEVF_RX_BUFSZ,
 748                                          DMA_FROM_DEVICE);
 749 }
 750
 751 static inline bool ixgbevf_page_is_reserved(struct page *page)
 752 {
 753         return (page_to_nid(page) != numa_mem_id()) || page_is_pfmemalloc(page);
 754 }
 755
 756 /**
 757  * ixgbevf_add_rx_frag - Add contents of Rx buffer to sk_buff
 758  * @rx_ring: rx descriptor ring to transact packets on
 759  * @rx_buffer: buffer containing page to add
 760  * @rx_desc: descriptor containing length of buffer written by hardware
 761  * @skb: sk_buff to place the data into
 762  *
 763  * This function will add the data contained in rx_buffer->page to the skb.
 764  * This is done either through a direct copy if the data in the buffer is
 765  * less than the skb header size, otherwise it will just attach the page as
 766  * a frag to the skb.
 767  *
 768  * The function will then update the page offset if necessary and return
 769  * true if the buffer can be reused by the adapter.
 770  **/
 771 static bool ixgbevf_add_rx_frag(struct ixgbevf_ring *rx_ring,
 772                                 struct ixgbevf_rx_buffer *rx_buffer,
 773                                 union ixgbe_adv_rx_desc *rx_desc,
 774                                 struct sk_buff *skb)
 775 {
 776         struct page *page = rx_buffer->page;
 777         unsigned char *va = page_address(page) + rx_buffer->page_offset;
 778         unsigned int size = le16_to_cpu(rx_desc->wb.upper.length);
 779 #if (PAGE_SIZE < 8192)
 780         unsigned int truesize = IXGBEVF_RX_BUFSZ;
 781 #else
 782         unsigned int truesize = ALIGN(size, L1_CACHE_BYTES);
 783 #endif
 784         unsigned int pull_len;
 785
 786         if (unlikely(skb_is_nonlinear(skb)))
 787                 goto add_tail_frag;
 788
 789         if (likely(size <= IXGBEVF_RX_HDR_SIZE)) {
 790                 memcpy(__skb_put(skb, size), va, ALIGN(size, sizeof(long)));
 791
 792                 /* page is not reserved, we can reuse buffer as is */
 793                 if (likely(!ixgbevf_page_is_reserved(page)))
 794                         return true;
 795
 796                 /* this page cannot be reused so discard it */
 797                 put_page(page);
 798                 return false;
 799         }
 800
 801         /* we need the header to contain the greater of either ETH_HLEN or
 802          * 60 bytes if the skb->len is less than 60 for skb_pad.
 803          */
 804         pull_len = eth_get_headlen(va, IXGBEVF_RX_HDR_SIZE);
 805
 806         /* align pull length to size of long to optimize memcpy performance */
 807         memcpy(__skb_put(skb, pull_len), va, ALIGN(pull_len, sizeof(long)));
 808
 809         /* update all of the pointers */
 810         va += pull_len;
 811         size -= pull_len;
 812
 813 add_tail_frag:
 814         skb_add_rx_frag(skb, skb_shinfo(skb)->nr_frags, page,
 815                         (unsigned long)va & ~PAGE_MASK, size, truesize);
 816
 817         /* avoid re-using remote pages */
 818         if (unlikely(ixgbevf_page_is_reserved(page)))
 819                 return false;
 820
 821 #if (PAGE_SIZE < 8192)
 822         /* if we are only owner of page we can reuse it */
 823         if (unlikely(page_count(page) != 1))
 824                 return false;
 825
 826         /* flip page offset to other buffer */
 827         rx_buffer->page_offset ^= IXGBEVF_RX_BUFSZ;
 828
 829 #else
 830         /* move offset up to the next cache line */
 831         rx_buffer->page_offset += truesize;
 832
 833         if (rx_buffer->page_offset > (PAGE_SIZE - IXGBEVF_RX_BUFSZ))
 834                 return false;
 835
 836 #endif
 837         /* Even if we own the page, we are not allowed to use atomic_set()
 838          * This would break get_page_unless_zero() users.
 839          */
 840         atomic_inc(&page->_count);
 841
 842         return true;
 843 }
 844
 845 static struct sk_buff *ixgbevf_fetch_rx_buffer(struct ixgbevf_ring *rx_ring,
 846                                                union ixgbe_adv_rx_desc *rx_desc,
 847                                                struct sk_buff *skb)
 848 {
 849         struct ixgbevf_rx_buffer *rx_buffer;
 850         struct page *page;
 851
 852         rx_buffer = &rx_ring->rx_buffer_info[rx_ring->next_to_clean];
 853         page = rx_buffer->page;
 854         prefetchw(page);
 855
 856         if (likely(!skb)) {
 857                 void *page_addr = page_address(page) +
 858                                   rx_buffer->page_offset;
 859
 860                 /* prefetch first cache line of first page */
 861                 prefetch(page_addr);
 862 #if L1_CACHE_BYTES < 128
 863                 prefetch(page_addr + L1_CACHE_BYTES);
 864 #endif
 865
 866                 /* allocate a skb to store the frags */
 867                 skb = netdev_alloc_skb_ip_align(rx_ring->netdev,
 868                                                 IXGBEVF_RX_HDR_SIZE);
 869                 if (unlikely(!skb)) {
 870                         rx_ring->rx_stats.alloc_rx_buff_failed++;
 871                         return NULL;
 872                 }
 873
 874                 /* we will be copying header into skb->data in
 875                  * pskb_may_pull so it is in our interest to prefetch
 876                  * it now to avoid a possible cache miss
 877                  */
 878                 prefetchw(skb->data);
 879         }
 880
 881         /* we are reusing so sync this buffer for CPU use */
 882         dma_sync_single_range_for_cpu(rx_ring->dev,
 883                                       rx_buffer->dma,
 884                                       rx_buffer->page_offset,
 885                                       IXGBEVF_RX_BUFSZ,
 886                                       DMA_FROM_DEVICE);
 887
 888         /* pull page into skb */
 889         if (ixgbevf_add_rx_frag(rx_ring, rx_buffer, rx_desc, skb)) {
 890                 /* hand second half of page back to the ring */
 891                 ixgbevf_reuse_rx_page(rx_ring, rx_buffer);
 892         } else {
 893                 /* we are not reusing the buffer so unmap it */
 894                 dma_unmap_page(rx_ring->dev, rx_buffer->dma,
 895                                PAGE_SIZE, DMA_FROM_DEVICE);
 896         }
 897
 898         /* clear contents of buffer_info */
 899         rx_buffer->dma = 0;
 900         rx_buffer->page = NULL;
 901
 902         return skb;
 903 }
 904
 905 static inline void ixgbevf_irq_enable_queues(struct ixgbevf_adapter *adapter,
 906                                              u32 qmask)
 907 {
 908         struct ixgbe_hw *hw = &adapter->hw;
 909
 910         IXGBE_WRITE_REG(hw, IXGBE_VTEIMS, qmask);
 911 }
 912
 913 static int ixgbevf_clean_rx_irq(struct ixgbevf_q_vector *q_vector,
 914                                 struct ixgbevf_ring *rx_ring,
 915                                 int budget)
 916 {
 917         unsigned int total_rx_bytes = 0, total_rx_packets = 0;
 918         u16 cleaned_count = ixgbevf_desc_unused(rx_ring);
 919         struct sk_buff *skb = rx_ring->skb;
 920
 921         while (likely(total_rx_packets < budget)) {
 922                 union ixgbe_adv_rx_desc *rx_desc;
 923
 924                 /* return some buffers to hardware, one at a time is too slow */
 925                 if (cleaned_count >= IXGBEVF_RX_BUFFER_WRITE) {
 926                         ixgbevf_alloc_rx_buffers(rx_ring, cleaned_count);
 927                         cleaned_count = 0;
 928                 }
 929
 930                 rx_desc = IXGBEVF_RX_DESC(rx_ring, rx_ring->next_to_clean);
 931
 932                 if (!ixgbevf_test_staterr(rx_desc, IXGBE_RXD_STAT_DD))
 933                         break;
 934
 935                 /* This memory barrier is needed to keep us from reading
 936                  * any other fields out of the rx_desc until we know the
 937                  * RXD_STAT_DD bit is set
 938                  */
 939                 rmb();
 940
 941                 /* retrieve a buffer from the ring */
 942                 skb = ixgbevf_fetch_rx_buffer(rx_ring, rx_desc, skb);
 943
 944                 /* exit if we failed to retrieve a buffer */
 945                 if (!skb)
 946                         break;
 947
 948                 cleaned_count++;
 949
 950                 /* fetch next buffer in frame if non-eop */
 951                 if (ixgbevf_is_non_eop(rx_ring, rx_desc))
 952                         continue;
 953
 954                 /* verify the packet layout is correct */
 955                 if (ixgbevf_cleanup_headers(rx_ring, rx_desc, skb)) {
 956                         skb = NULL;
 957                         continue;
 958                 }
 959
 960                 /* probably a little skewed due to removing CRC */
 961                 total_rx_bytes += skb->len;
 962
 963                 /* Workaround hardware that can't do proper VEPA multicast
 964                  * source pruning.
 965                  */
 966                 if ((skb->pkt_type == PACKET_BROADCAST ||
 967                      skb->pkt_type == PACKET_MULTICAST) &&
 968                     ether_addr_equal(rx_ring->netdev->dev_addr,
 969                                      eth_hdr(skb)->h_source)) {
 970                         dev_kfree_skb_irq(skb);
 971                         continue;
 972                 }
 973
 974                 /* populate checksum, VLAN, and protocol */
 975                 ixgbevf_process_skb_fields(rx_ring, rx_desc, skb);
 976
 977                 ixgbevf_rx_skb(q_vector, skb);
 978
 979                 /* reset skb pointer */
 980                 skb = NULL;
 981
 982                 /* update budget accounting */
 983                 total_rx_packets++;
 984         }
 985
 986         /* place incomplete frames back on ring for completion */
 987         rx_ring->skb = skb;
 988
 989         u64_stats_update_begin(&rx_ring->syncp);
 990         rx_ring->stats.packets += total_rx_packets;
 991         rx_ring->stats.bytes += total_rx_bytes;
 992         u64_stats_update_end(&rx_ring->syncp);
 993         q_vector->rx.total_packets += total_rx_packets;
 994         q_vector->rx.total_bytes += total_rx_bytes;
 995
 996         return total_rx_packets;
 997 }
 998
 999 /**
1000  * ixgbevf_poll - NAPI polling calback
1001  * @napi: napi struct with our devices info in it
1002  * @budget: amount of work driver is allowed to do this pass, in packets
1003  *
1004  * This function will clean more than one or more rings associated with a
1005  * q_vector.
1006  **/
1007 static int ixgbevf_poll(struct napi_struct *napi, int budget)
1008 {
1009         struct ixgbevf_q_vector *q_vector =
1010                 container_of(napi, struct ixgbevf_q_vector, napi);
1011         struct ixgbevf_adapter *adapter = q_vector->adapter;
1012         struct ixgbevf_ring *ring;
1013         int per_ring_budget, work_done = 0;
1014         bool clean_complete = true;
1015
1016         ixgbevf_for_each_ring(ring, q_vector->tx)
1017                 clean_complete &= ixgbevf_clean_tx_irq(q_vector, ring);
1018
1019         if (budget <= 0)
1020                 return budget;
1021 #ifdef CONFIG_NET_RX_BUSY_POLL
1022         if (!ixgbevf_qv_lock_napi(q_vector))
1023                 return budget;
1024 #endif
1025
1026         /* attempt to distribute budget to each queue fairly, but don't allow
1027          * the budget to go below 1 because we'll exit polling
1028          */
1029         if (q_vector->rx.count > 1)
1030                 per_ring_budget = max(budget/q_vector->rx.count, 1);
1031         else
1032                 per_ring_budget = budget;
1033
1034         ixgbevf_for_each_ring(ring, q_vector->rx) {
1035                 int cleaned = ixgbevf_clean_rx_irq(q_vector, ring,
1036                                                    per_ring_budget);
1037                 work_done += cleaned;
1038                 clean_complete &= (cleaned < per_ring_budget);
1039         }
1040
1041 #ifdef CONFIG_NET_RX_BUSY_POLL
1042         ixgbevf_qv_unlock_napi(q_vector);
1043 #endif
1044
1045         /* If all work not completed, return budget and keep polling */
1046         if (!clean_complete)
1047                 return budget;
1048         /* all work done, exit the polling mode */
1049         napi_complete_done(napi, work_done);
1050         if (adapter->rx_itr_setting == 1)
1051                 ixgbevf_set_itr(q_vector);
1052         if (!test_bit(__IXGBEVF_DOWN, &adapter->state) &&
1053             !test_bit(__IXGBEVF_REMOVING, &adapter->state))
1054                 ixgbevf_irq_enable_queues(adapter,
1055                                           1 << q_vector->v_idx);
1056
1057         return 0;
1058 }
1059
1060 /**
1061  * ixgbevf_write_eitr - write VTEITR register in hardware specific way
1062  * @q_vector: structure containing interrupt and ring information
1063  **/
1064 void ixgbevf_write_eitr(struct ixgbevf_q_vector *q_vector)
1065 {
1066         struct ixgbevf_adapter *adapter = q_vector->adapter;
1067         struct ixgbe_hw *hw = &adapter->hw;
1068         int v_idx = q_vector->v_idx;
1069         u32 itr_reg = q_vector->itr & IXGBE_MAX_EITR;
1070
1071         /* set the WDIS bit to not clear the timer bits and cause an
1072          * immediate assertion of the interrupt
1073          */
1074         itr_reg |= IXGBE_EITR_CNT_WDIS;
1075
1076         IXGBE_WRITE_REG(hw, IXGBE_VTEITR(v_idx), itr_reg);
1077 }
1078
1079 #ifdef CONFIG_NET_RX_BUSY_POLL
1080 /* must be called with local_bh_disable()d */
1081 static int ixgbevf_busy_poll_recv(struct napi_struct *napi)
1082 {
1083         struct ixgbevf_q_vector *q_vector =
1084                         container_of(napi, struct ixgbevf_q_vector, napi);
1085         struct ixgbevf_adapter *adapter = q_vector->adapter;
1086         struct ixgbevf_ring  *ring;
1087         int found = 0;
1088
1089         if (test_bit(__IXGBEVF_DOWN, &adapter->state))
1090                 return LL_FLUSH_FAILED;
1091
1092         if (!ixgbevf_qv_lock_poll(q_vector))
1093                 return LL_FLUSH_BUSY;
1094
1095         ixgbevf_for_each_ring(ring, q_vector->rx) {
1096                 found = ixgbevf_clean_rx_irq(q_vector, ring, 4);
1097 #ifdef BP_EXTENDED_STATS
1098                 if (found)
1099                         ring->stats.cleaned += found;
1100                 else
1101                         ring->stats.misses++;
1102 #endif
1103                 if (found)
1104                         break;
1105         }
1106
1107         ixgbevf_qv_unlock_poll(q_vector);
1108
1109         return found;
1110 }
1111 #endif /* CONFIG_NET_RX_BUSY_POLL */
1112
1113 /**
1114  * ixgbevf_configure_msix - Configure MSI-X hardware
1115  * @adapter: board private structure
1116  *
1117  * ixgbevf_configure_msix sets up the hardware to properly generate MSI-X
1118  * interrupts.
1119  **/
1120 static void ixgbevf_configure_msix(struct ixgbevf_adapter *adapter)
1121 {
1122         struct ixgbevf_q_vector *q_vector;
1123         int q_vectors, v_idx;
1124
1125         q_vectors = adapter->num_msix_vectors - NON_Q_VECTORS;
1126         adapter->eims_enable_mask = 0;
1127
1128         /* Populate the IVAR table and set the ITR values to the
1129          * corresponding register.
1130          */
1131         for (v_idx = 0; v_idx < q_vectors; v_idx++) {
1132                 struct ixgbevf_ring *ring;
1133
1134                 q_vector = adapter->q_vector[v_idx];
1135
1136                 ixgbevf_for_each_ring(ring, q_vector->rx)
1137                         ixgbevf_set_ivar(adapter, 0, ring->reg_idx, v_idx);
1138
1139                 ixgbevf_for_each_ring(ring, q_vector->tx)
1140                         ixgbevf_set_ivar(adapter, 1, ring->reg_idx, v_idx);
1141
1142                 if (q_vector->tx.ring && !q_vector->rx.ring) {
1143                         /* Tx only vector */
1144                         if (adapter->tx_itr_setting == 1)
1145                                 q_vector->itr = IXGBE_12K_ITR;
1146                         else
1147                                 q_vector->itr = adapter->tx_itr_setting;
1148                 } else {
1149                         /* Rx or Rx/Tx vector */
1150                         if (adapter->rx_itr_setting == 1)
1151                                 q_vector->itr = IXGBE_20K_ITR;
1152                         else
1153                                 q_vector->itr = adapter->rx_itr_setting;
1154                 }
1155
1156                 /* add q_vector eims value to global eims_enable_mask */
1157                 adapter->eims_enable_mask |= 1 << v_idx;
1158
1159                 ixgbevf_write_eitr(q_vector);
1160         }
1161
1162         ixgbevf_set_ivar(adapter, -1, 1, v_idx);
1163         /* setup eims_other and add value to global eims_enable_mask */
1164         adapter->eims_other = 1 << v_idx;
1165         adapter->eims_enable_mask |= adapter->eims_other;
1166 }
1167
1168 enum latency_range {
1169         lowest_latency = 0,
1170         low_latency = 1,
1171         bulk_latency = 2,
1172         latency_invalid = 255
1173 };
1174
1175 /**
1176  * ixgbevf_update_itr - update the dynamic ITR value based on statistics
1177  * @q_vector: structure containing interrupt and ring information
1178  * @ring_container: structure containing ring performance data
1179  *
1180  * Stores a new ITR value based on packets and byte
1181  * counts during the last interrupt.  The advantage of per interrupt
1182  * computation is faster updates and more accurate ITR for the current
1183  * traffic pattern.  Constants in this function were computed
1184  * based on theoretical maximum wire speed and thresholds were set based
1185  * on testing data as well as attempting to minimize response time
1186  * while increasing bulk throughput.
1187  **/
1188 static void ixgbevf_update_itr(struct ixgbevf_q_vector *q_vector,
1189                                struct ixgbevf_ring_container *ring_container)
1190 {
1191         int bytes = ring_container->total_bytes;
1192         int packets = ring_container->total_packets;
1193         u32 timepassed_us;
1194         u64 bytes_perint;
1195         u8 itr_setting = ring_container->itr;
1196
1197         if (packets == 0)
1198                 return;
1199
1200         /* simple throttle rate management
1201          *    0-20MB/s lowest (100000 ints/s)
1202          *   20-100MB/s low   (20000 ints/s)
1203          *  100-1249MB/s bulk (12000 ints/s)
1204          */
1205         /* what was last interrupt timeslice? */
1206         timepassed_us = q_vector->itr >> 2;
1207         bytes_perint = bytes / timepassed_us; /* bytes/usec */
1208
1209         switch (itr_setting) {
1210         case lowest_latency:
1211                 if (bytes_perint > 10)
1212                         itr_setting = low_latency;
1213                 break;
1214         case low_latency:
1215                 if (bytes_perint > 20)
1216                         itr_setting = bulk_latency;
1217                 else if (bytes_perint <= 10)
1218                         itr_setting = lowest_latency;
1219                 break;
1220         case bulk_latency:
1221                 if (bytes_perint <= 20)
1222                         itr_setting = low_latency;
1223                 break;
1224         }
1225
1226         /* clear work counters since we have the values we need */
1227         ring_container->total_bytes = 0;
1228         ring_container->total_packets = 0;
1229
1230         /* write updated itr to ring container */
1231         ring_container->itr = itr_setting;
1232 }
1233
1234 static void ixgbevf_set_itr(struct ixgbevf_q_vector *q_vector)
1235 {
1236         u32 new_itr = q_vector->itr;
1237         u8 current_itr;
1238
1239         ixgbevf_update_itr(q_vector, &q_vector->tx);
1240         ixgbevf_update_itr(q_vector, &q_vector->rx);
1241
1242         current_itr = max(q_vector->rx.itr, q_vector->tx.itr);
1243
1244         switch (current_itr) {
1245         /* counts and packets in update_itr are dependent on these numbers */
1246         case lowest_latency:
1247                 new_itr = IXGBE_100K_ITR;
1248                 break;
1249         case low_latency:
1250                 new_itr = IXGBE_20K_ITR;
1251                 break;
1252         case bulk_latency:
1253                 new_itr = IXGBE_12K_ITR;
1254                 break;
1255         default:
1256                 break;
1257         }
1258
1259         if (new_itr != q_vector->itr) {
1260                 /* do an exponential smoothing */
1261                 new_itr = (10 * new_itr * q_vector->itr) /
1262                           ((9 * new_itr) + q_vector->itr);
1263
1264                 /* save the algorithm value here */
1265                 q_vector->itr = new_itr;
1266
1267                 ixgbevf_write_eitr(q_vector);
1268         }
1269 }
1270
1271 static irqreturn_t ixgbevf_msix_other(int irq, void *data)
1272 {
1273         struct ixgbevf_adapter *adapter = data;
1274         struct ixgbe_hw *hw = &adapter->hw;
1275
1276         hw->mac.get_link_status = 1;
1277
1278         ixgbevf_service_event_schedule(adapter);
1279
1280         IXGBE_WRITE_REG(hw, IXGBE_VTEIMS, adapter->eims_other);
1281
1282         return IRQ_HANDLED;
1283 }
1284
1285 /**
1286  * ixgbevf_msix_clean_rings - single unshared vector rx clean (all queues)
1287  * @irq: unused
1288  * @data: pointer to our q_vector struct for this interrupt vector
1289  **/
1290 static irqreturn_t ixgbevf_msix_clean_rings(int irq, void *data)
1291 {
1292         struct ixgbevf_q_vector *q_vector = data;
1293
1294         /* EIAM disabled interrupts (on this vector) for us */
1295         if (q_vector->rx.ring || q_vector->tx.ring)
1296                 napi_schedule_irqoff(&q_vector->napi);
1297
1298         return IRQ_HANDLED;
1299 }
1300
1301 static inline void map_vector_to_rxq(struct ixgbevf_adapter *a, int v_idx,
1302                                      int r_idx)
1303 {
1304         struct ixgbevf_q_vector *q_vector = a->q_vector[v_idx];
1305
1306         a->rx_ring[r_idx]->next = q_vector->rx.ring;
1307         q_vector->rx.ring = a->rx_ring[r_idx];
1308         q_vector->rx.count++;
1309 }
1310
1311 static inline void map_vector_to_txq(struct ixgbevf_adapter *a, int v_idx,
1312                                      int t_idx)
1313 {
1314         struct ixgbevf_q_vector *q_vector = a->q_vector[v_idx];
1315
1316         a->tx_ring[t_idx]->next = q_vector->tx.ring;
1317         q_vector->tx.ring = a->tx_ring[t_idx];
1318         q_vector->tx.count++;
1319 }
1320
1321 /**
1322  * ixgbevf_map_rings_to_vectors - Maps descriptor rings to vectors
1323  * @adapter: board private structure to initialize
1324  *
1325  * This function maps descriptor rings to the queue-specific vectors
1326  * we were allotted through the MSI-X enabling code.  Ideally, we'd have
1327  * one vector per ring/queue, but on a constrained vector budget, we
1328  * group the rings as "efficiently" as possible.  You would add new
1329  * mapping configurations in here.
1330  **/
1331 static int ixgbevf_map_rings_to_vectors(struct ixgbevf_adapter *adapter)
1332 {
1333         int q_vectors;
1334         int v_start = 0;
1335         int rxr_idx = 0, txr_idx = 0;
1336         int rxr_remaining = adapter->num_rx_queues;
1337         int txr_remaining = adapter->num_tx_queues;
1338         int i, j;
1339         int rqpv, tqpv;
1340
1341         q_vectors = adapter->num_msix_vectors - NON_Q_VECTORS;
1342
1343         /* The ideal configuration...
1344          * We have enough vectors to map one per queue.
1345          */
1346         if (q_vectors == adapter->num_rx_queues + adapter->num_tx_queues) {
1347                 for (; rxr_idx < rxr_remaining; v_start++, rxr_idx++)
1348                         map_vector_to_rxq(adapter, v_start, rxr_idx);
1349
1350                 for (; txr_idx < txr_remaining; v_start++, txr_idx++)
1351                         map_vector_to_txq(adapter, v_start, txr_idx);
1352                 return 0;
1353         }
1354
1355         /* If we don't have enough vectors for a 1-to-1
1356          * mapping, we'll have to group them so there are
1357          * multiple queues per vector.
1358          */
1359         /* Re-adjusting *qpv takes care of the remainder. */
1360         for (i = v_start; i < q_vectors; i++) {
1361                 rqpv = DIV_ROUND_UP(rxr_remaining, q_vectors - i);
1362                 for (j = 0; j < rqpv; j++) {
1363                         map_vector_to_rxq(adapter, i, rxr_idx);
1364                         rxr_idx++;
1365                         rxr_remaining--;
1366                 }
1367         }
1368         for (i = v_start; i < q_vectors; i++) {
1369                 tqpv = DIV_ROUND_UP(txr_remaining, q_vectors - i);
1370                 for (j = 0; j < tqpv; j++) {
1371                         map_vector_to_txq(adapter, i, txr_idx);
1372                         txr_idx++;
1373                         txr_remaining--;
1374                 }
1375         }
1376
1377         return 0;
1378 }
1379
1380 /**
1381  * ixgbevf_request_msix_irqs - Initialize MSI-X interrupts
1382  * @adapter: board private structure
1383  *
1384  * ixgbevf_request_msix_irqs allocates MSI-X vectors and requests
1385  * interrupts from the kernel.
1386  **/
1387 static int ixgbevf_request_msix_irqs(struct ixgbevf_adapter *adapter)
1388 {
1389         struct net_device *netdev = adapter->netdev;
1390         int q_vectors = adapter->num_msix_vectors - NON_Q_VECTORS;
1391         int vector, err;
1392         int ri = 0, ti = 0;
1393
1394         for (vector = 0; vector < q_vectors; vector++) {
1395                 struct ixgbevf_q_vector *q_vector = adapter->q_vector[vector];
1396                 struct msix_entry *entry = &adapter->msix_entries[vector];
1397
1398                 if (q_vector->tx.ring && q_vector->rx.ring) {
1399                         snprintf(q_vector->name, sizeof(q_vector->name) - 1,
1400                                  "%s-%s-%d", netdev->name, "TxRx", ri++);
1401                         ti++;
1402                 } else if (q_vector->rx.ring) {
1403                         snprintf(q_vector->name, sizeof(q_vector->name) - 1,
1404                                  "%s-%s-%d", netdev->name, "rx", ri++);
1405                 } else if (q_vector->tx.ring) {
1406                         snprintf(q_vector->name, sizeof(q_vector->name) - 1,
1407                                  "%s-%s-%d", netdev->name, "tx", ti++);
1408                 } else {
1409                         /* skip this unused q_vector */
1410                         continue;
1411                 }
1412                 err = request_irq(entry->vector, &ixgbevf_msix_clean_rings, 0,
1413                                   q_vector->name, q_vector);
1414                 if (err) {
1415                         hw_dbg(&adapter->hw,
1416                                "request_irq failed for MSIX interrupt Error: %d\n",
1417                                err);
1418                         goto free_queue_irqs;
1419                 }
1420         }
1421
1422         err = request_irq(adapter->msix_entries[vector].vector,
1423                           &ixgbevf_msix_other, 0, netdev->name, adapter);
1424         if (err) {
1425                 hw_dbg(&adapter->hw, "request_irq for msix_other failed: %d\n",
1426                        err);
1427                 goto free_queue_irqs;
1428         }
1429
1430         return 0;
1431
1432 free_queue_irqs:
1433         while (vector) {
1434                 vector--;
1435                 free_irq(adapter->msix_entries[vector].vector,
1436                          adapter->q_vector[vector]);
1437         }
1438         /* This failure is non-recoverable - it indicates the system is
1439          * out of MSIX vector resources and the VF driver cannot run
1440          * without them.  Set the number of msix vectors to zero
1441          * indicating that not enough can be allocated.  The error
1442          * will be returned to the user indicating device open failed.
1443          * Any further attempts to force the driver to open will also
1444          * fail.  The only way to recover is to unload the driver and
1445          * reload it again.  If the system has recovered some MSIX
1446          * vectors then it may succeed.
1447          */
1448         adapter->num_msix_vectors = 0;
1449         return err;
1450 }
1451
1452 static inline void ixgbevf_reset_q_vectors(struct ixgbevf_adapter *adapter)
1453 {
1454         int i, q_vectors = adapter->num_msix_vectors - NON_Q_VECTORS;
1455
1456         for (i = 0; i < q_vectors; i++) {
1457                 struct ixgbevf_q_vector *q_vector = adapter->q_vector[i];
1458
1459                 q_vector->rx.ring = NULL;
1460                 q_vector->tx.ring = NULL;
1461                 q_vector->rx.count = 0;
1462                 q_vector->tx.count = 0;
1463         }
1464 }
1465
1466 /**
1467  * ixgbevf_request_irq - initialize interrupts
1468  * @adapter: board private structure
1469  *
1470  * Attempts to configure interrupts using the best available
1471  * capabilities of the hardware and kernel.
1472  **/
1473 static int ixgbevf_request_irq(struct ixgbevf_adapter *adapter)
1474 {
1475         int err = ixgbevf_request_msix_irqs(adapter);
1476
1477         if (err)
1478                 hw_dbg(&adapter->hw, "request_irq failed, Error %d\n", err);
1479
1480         return err;
1481 }
1482
1483 static void ixgbevf_free_irq(struct ixgbevf_adapter *adapter)
1484 {
1485         int i, q_vectors;
1486
1487         q_vectors = adapter->num_msix_vectors;
1488         i = q_vectors - 1;
1489
1490         free_irq(adapter->msix_entries[i].vector, adapter);
1491         i--;
1492
1493         for (; i >= 0; i--) {
1494                 /* free only the irqs that were actually requested */
1495                 if (!adapter->q_vector[i]->rx.ring &&
1496                     !adapter->q_vector[i]->tx.ring)
1497                         continue;
1498
1499                 free_irq(adapter->msix_entries[i].vector,
1500                          adapter->q_vector[i]);
1501         }
1502
1503         ixgbevf_reset_q_vectors(adapter);
1504 }
1505
1506 /**
1507  * ixgbevf_irq_disable - Mask off interrupt generation on the NIC
1508  * @adapter: board private structure
1509  **/
1510 static inline void ixgbevf_irq_disable(struct ixgbevf_adapter *adapter)
1511 {
1512         struct ixgbe_hw *hw = &adapter->hw;
1513         int i;
1514
1515         IXGBE_WRITE_REG(hw, IXGBE_VTEIAM, 0);
1516         IXGBE_WRITE_REG(hw, IXGBE_VTEIMC, ~0);
1517         IXGBE_WRITE_REG(hw, IXGBE_VTEIAC, 0);
1518
1519         IXGBE_WRITE_FLUSH(hw);
1520
1521         for (i = 0; i < adapter->num_msix_vectors; i++)
1522                 synchronize_irq(adapter->msix_entries[i].vector);
1523 }
1524
1525 /**
1526  * ixgbevf_irq_enable - Enable default interrupt generation settings
1527  * @adapter: board private structure
1528  **/
1529 static inline void ixgbevf_irq_enable(struct ixgbevf_adapter *adapter)
1530 {
1531         struct ixgbe_hw *hw = &adapter->hw;
1532
1533         IXGBE_WRITE_REG(hw, IXGBE_VTEIAM, adapter->eims_enable_mask);
1534         IXGBE_WRITE_REG(hw, IXGBE_VTEIAC, adapter->eims_enable_mask);
1535         IXGBE_WRITE_REG(hw, IXGBE_VTEIMS, adapter->eims_enable_mask);
1536 }
1537
1538 /**
1539  * ixgbevf_configure_tx_ring - Configure 82599 VF Tx ring after Reset
1540  * @adapter: board private structure
1541  * @ring: structure containing ring specific data
1542  *
1543  * Configure the Tx descriptor ring after a reset.
1544  **/
1545 static void ixgbevf_configure_tx_ring(struct ixgbevf_adapter *adapter,
1546                                       struct ixgbevf_ring *ring)
1547 {
1548         struct ixgbe_hw *hw = &adapter->hw;
1549         u64 tdba = ring->dma;
1550         int wait_loop = 10;
1551         u32 txdctl = IXGBE_TXDCTL_ENABLE;
1552         u8 reg_idx = ring->reg_idx;
1553
1554         /* disable queue to avoid issues while updating state */
1555         IXGBE_WRITE_REG(hw, IXGBE_VFTXDCTL(reg_idx), IXGBE_TXDCTL_SWFLSH);
1556         IXGBE_WRITE_FLUSH(hw);
1557
1558         IXGBE_WRITE_REG(hw, IXGBE_VFTDBAL(reg_idx), tdba & DMA_BIT_MASK(32));
1559         IXGBE_WRITE_REG(hw, IXGBE_VFTDBAH(reg_idx), tdba >> 32);
1560         IXGBE_WRITE_REG(hw, IXGBE_VFTDLEN(reg_idx),
1561                         ring->count * sizeof(union ixgbe_adv_tx_desc));
1562
1563         /* disable head writeback */
1564         IXGBE_WRITE_REG(hw, IXGBE_VFTDWBAH(reg_idx), 0);
1565         IXGBE_WRITE_REG(hw, IXGBE_VFTDWBAL(reg_idx), 0);
1566
1567         /* enable relaxed ordering */
1568         IXGBE_WRITE_REG(hw, IXGBE_VFDCA_TXCTRL(reg_idx),
1569                         (IXGBE_DCA_TXCTRL_DESC_RRO_EN |
1570                          IXGBE_DCA_TXCTRL_DATA_RRO_EN));
1571
1572         /* reset head and tail pointers */
1573         IXGBE_WRITE_REG(hw, IXGBE_VFTDH(reg_idx), 0);
1574         IXGBE_WRITE_REG(hw, IXGBE_VFTDT(reg_idx), 0);
1575         ring->tail = adapter->io_addr + IXGBE_VFTDT(reg_idx);
1576
1577         /* reset ntu and ntc to place SW in sync with hardwdare */
1578         ring->next_to_clean = 0;
1579         ring->next_to_use = 0;
1580
1581         /* In order to avoid issues WTHRESH + PTHRESH should always be equal
1582          * to or less than the number of on chip descriptors, which is
1583          * currently 40.
1584          */
1585         txdctl |= (8 << 16);    /* WTHRESH = 8 */
1586
1587         /* Setting PTHRESH to 32 both improves performance */
1588         txdctl |= (1 << 8) |    /* HTHRESH = 1 */
1589                   32;          /* PTHRESH = 32 */
1590
1591         clear_bit(__IXGBEVF_HANG_CHECK_ARMED, &ring->state);
1592
1593         IXGBE_WRITE_REG(hw, IXGBE_VFTXDCTL(reg_idx), txdctl);
1594
1595         /* poll to verify queue is enabled */
1596         do {
1597                 usleep_range(1000, 2000);
1598                 txdctl = IXGBE_READ_REG(hw, IXGBE_VFTXDCTL(reg_idx));
1599         }  while (--wait_loop && !(txdctl & IXGBE_TXDCTL_ENABLE));
1600         if (!wait_loop)
1601                 pr_err("Could not enable Tx Queue %d\n", reg_idx);
1602 }
1603
1604 /**
1605  * ixgbevf_configure_tx - Configure 82599 VF Transmit Unit after Reset
1606  * @adapter: board private structure
1607  *
1608  * Configure the Tx unit of the MAC after a reset.
1609  **/
1610 static void ixgbevf_configure_tx(struct ixgbevf_adapter *adapter)
1611 {
1612         u32 i;
1613
1614         /* Setup the HW Tx Head and Tail descriptor pointers */
1615         for (i = 0; i < adapter->num_tx_queues; i++)
1616                 ixgbevf_configure_tx_ring(adapter, adapter->tx_ring[i]);
1617 }
1618
1619 #define IXGBE_SRRCTL_BSIZEHDRSIZE_SHIFT 2
1620
1621 static void ixgbevf_configure_srrctl(struct ixgbevf_adapter *adapter, int index)
1622 {
1623         struct ixgbe_hw *hw = &adapter->hw;
1624         u32 srrctl;
1625
1626         srrctl = IXGBE_SRRCTL_DROP_EN;
1627
1628         srrctl |= IXGBEVF_RX_HDR_SIZE << IXGBE_SRRCTL_BSIZEHDRSIZE_SHIFT;
1629         srrctl |= IXGBEVF_RX_BUFSZ >> IXGBE_SRRCTL_BSIZEPKT_SHIFT;
1630         srrctl |= IXGBE_SRRCTL_DESCTYPE_ADV_ONEBUF;
1631
1632         IXGBE_WRITE_REG(hw, IXGBE_VFSRRCTL(index), srrctl);
1633 }
1634
1635 static void ixgbevf_setup_psrtype(struct ixgbevf_adapter *adapter)
1636 {
1637         struct ixgbe_hw *hw = &adapter->hw;
1638
1639         /* PSRTYPE must be initialized in 82599 */
1640         u32 psrtype = IXGBE_PSRTYPE_TCPHDR | IXGBE_PSRTYPE_UDPHDR |
1641                       IXGBE_PSRTYPE_IPV4HDR | IXGBE_PSRTYPE_IPV6HDR |
1642                       IXGBE_PSRTYPE_L2HDR;
1643
1644         if (adapter->num_rx_queues > 1)
1645                 psrtype |= 1 << 29;
1646
1647         IXGBE_WRITE_REG(hw, IXGBE_VFPSRTYPE, psrtype);
1648 }
1649
1650 #define IXGBEVF_MAX_RX_DESC_POLL 10
1651 static void ixgbevf_disable_rx_queue(struct ixgbevf_adapter *adapter,
1652                                      struct ixgbevf_ring *ring)
1653 {
1654         struct ixgbe_hw *hw = &adapter->hw;
1655         int wait_loop = IXGBEVF_MAX_RX_DESC_POLL;
1656         u32 rxdctl;
1657         u8 reg_idx = ring->reg_idx;
1658
1659         if (IXGBE_REMOVED(hw->hw_addr))
1660                 return;
1661         rxdctl = IXGBE_READ_REG(hw, IXGBE_VFRXDCTL(reg_idx));
1662         rxdctl &= ~IXGBE_RXDCTL_ENABLE;
1663
1664         /* write value back with RXDCTL.ENABLE bit cleared */
1665         IXGBE_WRITE_REG(hw, IXGBE_VFRXDCTL(reg_idx), rxdctl);
1666
1667         /* the hardware may take up to 100us to really disable the Rx queue */
1668         do {
1669                 udelay(10);
1670                 rxdctl = IXGBE_READ_REG(hw, IXGBE_VFRXDCTL(reg_idx));
1671         } while (--wait_loop && (rxdctl & IXGBE_RXDCTL_ENABLE));
1672
1673         if (!wait_loop)
1674                 pr_err("RXDCTL.ENABLE queue %d not cleared while polling\n",
1675                        reg_idx);
1676 }
1677
1678 static void ixgbevf_rx_desc_queue_enable(struct ixgbevf_adapter *adapter,
1679                                          struct ixgbevf_ring *ring)
1680 {
1681         struct ixgbe_hw *hw = &adapter->hw;
1682         int wait_loop = IXGBEVF_MAX_RX_DESC_POLL;
1683         u32 rxdctl;
1684         u8 reg_idx = ring->reg_idx;
1685
1686         if (IXGBE_REMOVED(hw->hw_addr))
1687                 return;
1688         do {
1689                 usleep_range(1000, 2000);
1690                 rxdctl = IXGBE_READ_REG(hw, IXGBE_VFRXDCTL(reg_idx));
1691         } while (--wait_loop && !(rxdctl & IXGBE_RXDCTL_ENABLE));
1692
1693         if (!wait_loop)
1694                 pr_err("RXDCTL.ENABLE queue %d not set while polling\n",
1695                        reg_idx);
1696 }
1697
1698 static void ixgbevf_setup_vfmrqc(struct ixgbevf_adapter *adapter)
1699 {
1700         struct ixgbe_hw *hw = &adapter->hw;
1701         u32 vfmrqc = 0, vfreta = 0;
1702         u16 rss_i = adapter->num_rx_queues;
1703         u8 i, j;
1704
1705         /* Fill out hash function seeds */
1706         netdev_rss_key_fill(adapter->rss_key, sizeof(adapter->rss_key));
1707         for (i = 0; i < IXGBEVF_VFRSSRK_REGS; i++)
1708                 IXGBE_WRITE_REG(hw, IXGBE_VFRSSRK(i), adapter->rss_key[i]);
1709
1710         for (i = 0, j = 0; i < IXGBEVF_X550_VFRETA_SIZE; i++, j++) {
1711                 if (j == rss_i)
1712                         j = 0;
1713
1714                 adapter->rss_indir_tbl[i] = j;
1715
1716                 vfreta |= j << (i & 0x3) * 8;
1717                 if ((i & 3) == 3) {
1718                         IXGBE_WRITE_REG(hw, IXGBE_VFRETA(i >> 2), vfreta);
1719                         vfreta = 0;
1720                 }
1721         }
1722
1723         /* Perform hash on these packet types */
1724         vfmrqc |= IXGBE_VFMRQC_RSS_FIELD_IPV4 |
1725                 IXGBE_VFMRQC_RSS_FIELD_IPV4_TCP |
1726                 IXGBE_VFMRQC_RSS_FIELD_IPV6 |
1727                 IXGBE_VFMRQC_RSS_FIELD_IPV6_TCP;
1728
1729         vfmrqc |= IXGBE_VFMRQC_RSSEN;
1730
1731         IXGBE_WRITE_REG(hw, IXGBE_VFMRQC, vfmrqc);
1732 }
1733
1734 static void ixgbevf_configure_rx_ring(struct ixgbevf_adapter *adapter,
1735                                       struct ixgbevf_ring *ring)
1736 {
1737         struct ixgbe_hw *hw = &adapter->hw;
1738         u64 rdba = ring->dma;
1739         u32 rxdctl;
1740         u8 reg_idx = ring->reg_idx;
1741
1742         /* disable queue to avoid issues while updating state */
1743         rxdctl = IXGBE_READ_REG(hw, IXGBE_VFRXDCTL(reg_idx));
1744         ixgbevf_disable_rx_queue(adapter, ring);
1745
1746         IXGBE_WRITE_REG(hw, IXGBE_VFRDBAL(reg_idx), rdba & DMA_BIT_MASK(32));
1747         IXGBE_WRITE_REG(hw, IXGBE_VFRDBAH(reg_idx), rdba >> 32);
1748         IXGBE_WRITE_REG(hw, IXGBE_VFRDLEN(reg_idx),
1749                         ring->count * sizeof(union ixgbe_adv_rx_desc));
1750
1751         /* enable relaxed ordering */
1752         IXGBE_WRITE_REG(hw, IXGBE_VFDCA_RXCTRL(reg_idx),
1753                         IXGBE_DCA_RXCTRL_DESC_RRO_EN);
1754
1755         /* reset head and tail pointers */
1756         IXGBE_WRITE_REG(hw, IXGBE_VFRDH(reg_idx), 0);
1757         IXGBE_WRITE_REG(hw, IXGBE_VFRDT(reg_idx), 0);
1758         ring->tail = adapter->io_addr + IXGBE_VFRDT(reg_idx);
1759
1760         /* reset ntu and ntc to place SW in sync with hardwdare */
1761         ring->next_to_clean = 0;
1762         ring->next_to_use = 0;
1763         ring->next_to_alloc = 0;
1764
1765         ixgbevf_configure_srrctl(adapter, reg_idx);
1766
1767         /* allow any size packet since we can handle overflow */
1768         rxdctl &= ~IXGBE_RXDCTL_RLPML_EN;
1769
1770         rxdctl |= IXGBE_RXDCTL_ENABLE | IXGBE_RXDCTL_VME;
1771         IXGBE_WRITE_REG(hw, IXGBE_VFRXDCTL(reg_idx), rxdctl);
1772
1773         ixgbevf_rx_desc_queue_enable(adapter, ring);
1774         ixgbevf_alloc_rx_buffers(ring, ixgbevf_desc_unused(ring));
1775 }
1776
1777 /**
1778  * ixgbevf_configure_rx - Configure 82599 VF Receive Unit after Reset
1779  * @adapter: board private structure
1780  *
1781  * Configure the Rx unit of the MAC after a reset.
1782  **/
1783 static void ixgbevf_configure_rx(struct ixgbevf_adapter *adapter)
1784 {
1785         int i;
1786         struct ixgbe_hw *hw = &adapter->hw;
1787         struct net_device *netdev = adapter->netdev;
1788
1789         ixgbevf_setup_psrtype(adapter);
1790         if (hw->mac.type >= ixgbe_mac_X550_vf)
1791                 ixgbevf_setup_vfmrqc(adapter);
1792
1793         /* notify the PF of our intent to use this size of frame */
1794         ixgbevf_rlpml_set_vf(hw, netdev->mtu + ETH_HLEN + ETH_FCS_LEN);
1795
1796         /* Setup the HW Rx Head and Tail Descriptor Pointers and
1797          * the Base and Length of the Rx Descriptor Ring
1798          */
1799         for (i = 0; i < adapter->num_rx_queues; i++)
1800                 ixgbevf_configure_rx_ring(adapter, adapter->rx_ring[i]);
1801 }
1802
1803 static int ixgbevf_vlan_rx_add_vid(struct net_device *netdev,
1804                                    __be16 proto, u16 vid)
1805 {
1806         struct ixgbevf_adapter *adapter = netdev_priv(netdev);
1807         struct ixgbe_hw *hw = &adapter->hw;
1808         int err;
1809
1810         spin_lock_bh(&adapter->mbx_lock);
1811
1812         /* add VID to filter table */
1813         err = hw->mac.ops.set_vfta(hw, vid, 0, true);
1814
1815         spin_unlock_bh(&adapter->mbx_lock);
1816
1817         /* translate error return types so error makes sense */
1818         if (err == IXGBE_ERR_MBX)
1819                 return -EIO;
1820
1821         if (err == IXGBE_ERR_INVALID_ARGUMENT)
1822                 return -EACCES;
1823
1824         set_bit(vid, adapter->active_vlans);
1825
1826         return err;
1827 }
1828
1829 static int ixgbevf_vlan_rx_kill_vid(struct net_device *netdev,
1830                                     __be16 proto, u16 vid)
1831 {
1832         struct ixgbevf_adapter *adapter = netdev_priv(netdev);
1833         struct ixgbe_hw *hw = &adapter->hw;
1834         int err;
1835
1836         spin_lock_bh(&adapter->mbx_lock);
1837
1838         /* remove VID from filter table */
1839         err = hw->mac.ops.set_vfta(hw, vid, 0, false);
1840
1841         spin_unlock_bh(&adapter->mbx_lock);
1842
1843         clear_bit(vid, adapter->active_vlans);
1844
1845         return err;
1846 }
1847
1848 static void ixgbevf_restore_vlan(struct ixgbevf_adapter *adapter)
1849 {
1850         u16 vid;
1851
1852         for_each_set_bit(vid, adapter->active_vlans, VLAN_N_VID)
1853                 ixgbevf_vlan_rx_add_vid(adapter->netdev,
1854                                         htons(ETH_P_8021Q), vid);
1855 }
1856
1857 static int ixgbevf_write_uc_addr_list(struct net_device *netdev)
1858 {
1859         struct ixgbevf_adapter *adapter = netdev_priv(netdev);
1860         struct ixgbe_hw *hw = &adapter->hw;
1861         int count = 0;
1862
1863         if ((netdev_uc_count(netdev)) > 10) {
1864                 pr_err("Too many unicast filters - No Space\n");
1865                 return -ENOSPC;
1866         }
1867
1868         if (!netdev_uc_empty(netdev)) {
1869                 struct netdev_hw_addr *ha;
1870
1871                 netdev_for_each_uc_addr(ha, netdev) {
1872                         hw->mac.ops.set_uc_addr(hw, ++count, ha->addr);
1873                         udelay(200);
1874                 }
1875         } else {
1876                 /* If the list is empty then send message to PF driver to
1877                  * clear all MAC VLANs on this VF.
1878                  */
1879                 hw->mac.ops.set_uc_addr(hw, 0, NULL);
1880         }
1881
1882         return count;
1883 }
1884
1885 /**
1886  * ixgbevf_set_rx_mode - Multicast and unicast set
1887  * @netdev: network interface device structure
1888  *
1889  * The set_rx_method entry point is called whenever the multicast address
1890  * list, unicast address list or the network interface flags are updated.
1891  * This routine is responsible for configuring the hardware for proper
1892  * multicast mode and configuring requested unicast filters.
1893  **/
1894 static void ixgbevf_set_rx_mode(struct net_device *netdev)
1895 {
1896         struct ixgbevf_adapter *adapter = netdev_priv(netdev);
1897         struct ixgbe_hw *hw = &adapter->hw;
1898         unsigned int flags = netdev->flags;
1899         int xcast_mode;
1900
1901         xcast_mode = (flags & IFF_ALLMULTI) ? IXGBEVF_XCAST_MODE_ALLMULTI :
1902                      (flags & (IFF_BROADCAST | IFF_MULTICAST)) ?
1903                      IXGBEVF_XCAST_MODE_MULTI : IXGBEVF_XCAST_MODE_NONE;
1904
1905         spin_lock_bh(&adapter->mbx_lock);
1906
1907         hw->mac.ops.update_xcast_mode(hw, netdev, xcast_mode);
1908
1909         /* reprogram multicast list */
1910         hw->mac.ops.update_mc_addr_list(hw, netdev);
1911
1912         ixgbevf_write_uc_addr_list(netdev);
1913
1914         spin_unlock_bh(&adapter->mbx_lock);
1915 }
1916
1917 static void ixgbevf_napi_enable_all(struct ixgbevf_adapter *adapter)
1918 {
1919         int q_idx;
1920         struct ixgbevf_q_vector *q_vector;
1921         int q_vectors = adapter->num_msix_vectors - NON_Q_VECTORS;
1922
1923         for (q_idx = 0; q_idx < q_vectors; q_idx++) {
1924                 q_vector = adapter->q_vector[q_idx];
1925 #ifdef CONFIG_NET_RX_BUSY_POLL
1926                 ixgbevf_qv_init_lock(adapter->q_vector[q_idx]);
1927 #endif
1928                 napi_enable(&q_vector->napi);
1929         }
1930 }
1931
1932 static void ixgbevf_napi_disable_all(struct ixgbevf_adapter *adapter)
1933 {
1934         int q_idx;
1935         struct ixgbevf_q_vector *q_vector;
1936         int q_vectors = adapter->num_msix_vectors - NON_Q_VECTORS;
1937
1938         for (q_idx = 0; q_idx < q_vectors; q_idx++) {
1939                 q_vector = adapter->q_vector[q_idx];
1940                 napi_disable(&q_vector->napi);
1941 #ifdef CONFIG_NET_RX_BUSY_POLL
1942                 while (!ixgbevf_qv_disable(adapter->q_vector[q_idx])) {
1943                         pr_info("QV %d locked\n", q_idx);
1944                         usleep_range(1000, 20000);
1945                 }
1946 #endif /* CONFIG_NET_RX_BUSY_POLL */
1947         }
1948 }
1949
1950 static int ixgbevf_configure_dcb(struct ixgbevf_adapter *adapter)
1951 {
1952         struct ixgbe_hw *hw = &adapter->hw;
1953         unsigned int def_q = 0;
1954         unsigned int num_tcs = 0;
1955         unsigned int num_rx_queues = adapter->num_rx_queues;
1956         unsigned int num_tx_queues = adapter->num_tx_queues;
1957         int err;
1958
1959         spin_lock_bh(&adapter->mbx_lock);
1960
1961         /* fetch queue configuration from the PF */
1962         err = ixgbevf_get_queues(hw, &num_tcs, &def_q);
1963
1964         spin_unlock_bh(&adapter->mbx_lock);
1965
1966         if (err)
1967                 return err;
1968
1969         if (num_tcs > 1) {
1970                 /* we need only one Tx queue */
1971                 num_tx_queues = 1;
1972
1973                 /* update default Tx ring register index */
1974                 adapter->tx_ring[0]->reg_idx = def_q;
1975
1976                 /* we need as many queues as traffic classes */
1977                 num_rx_queues = num_tcs;
1978         }
1979
1980         /* if we have a bad config abort request queue reset */
1981         if ((adapter->num_rx_queues != num_rx_queues) ||
1982             (adapter->num_tx_queues != num_tx_queues)) {
1983                 /* force mailbox timeout to prevent further messages */
1984                 hw->mbx.timeout = 0;
1985
1986                 /* wait for watchdog to come around and bail us out */
1987                 adapter->flags |= IXGBEVF_FLAG_QUEUE_RESET_REQUESTED;
1988         }
1989
1990         return 0;
1991 }
1992
1993 static void ixgbevf_configure(struct ixgbevf_adapter *adapter)
1994 {
1995         ixgbevf_configure_dcb(adapter);
1996
1997         ixgbevf_set_rx_mode(adapter->netdev);
1998
1999         ixgbevf_restore_vlan(adapter);
2000
2001         ixgbevf_configure_tx(adapter);
2002         ixgbevf_configure_rx(adapter);
2003 }
2004
2005 static void ixgbevf_save_reset_stats(struct ixgbevf_adapter *adapter)
2006 {
2007         /* Only save pre-reset stats if there are some */
2008         if (adapter->stats.vfgprc || adapter->stats.vfgptc) {
2009                 adapter->stats.saved_reset_vfgprc += adapter->stats.vfgprc -
2010                         adapter->stats.base_vfgprc;
2011                 adapter->stats.saved_reset_vfgptc += adapter->stats.vfgptc -
2012                         adapter->stats.base_vfgptc;
2013                 adapter->stats.saved_reset_vfgorc += adapter->stats.vfgorc -
2014                         adapter->stats.base_vfgorc;
2015                 adapter->stats.saved_reset_vfgotc += adapter->stats.vfgotc -
2016                         adapter->stats.base_vfgotc;
2017                 adapter->stats.saved_reset_vfmprc += adapter->stats.vfmprc -
2018                         adapter->stats.base_vfmprc;
2019         }
2020 }
2021
2022 static void ixgbevf_init_last_counter_stats(struct ixgbevf_adapter *adapter)
2023 {
2024         struct ixgbe_hw *hw = &adapter->hw;
2025
2026         adapter->stats.last_vfgprc = IXGBE_READ_REG(hw, IXGBE_VFGPRC);
2027         adapter->stats.last_vfgorc = IXGBE_READ_REG(hw, IXGBE_VFGORC_LSB);
2028         adapter->stats.last_vfgorc |=
2029                 (((u64)(IXGBE_READ_REG(hw, IXGBE_VFGORC_MSB))) << 32);
2030         adapter->stats.last_vfgptc = IXGBE_READ_REG(hw, IXGBE_VFGPTC);
2031         adapter->stats.last_vfgotc = IXGBE_READ_REG(hw, IXGBE_VFGOTC_LSB);
2032         adapter->stats.last_vfgotc |=
2033                 (((u64)(IXGBE_READ_REG(hw, IXGBE_VFGOTC_MSB))) << 32);
2034         adapter->stats.last_vfmprc = IXGBE_READ_REG(hw, IXGBE_VFMPRC);
2035
2036         adapter->stats.base_vfgprc = adapter->stats.last_vfgprc;
2037         adapter->stats.base_vfgorc = adapter->stats.last_vfgorc;
2038         adapter->stats.base_vfgptc = adapter->stats.last_vfgptc;
2039         adapter->stats.base_vfgotc = adapter->stats.last_vfgotc;
2040         adapter->stats.base_vfmprc = adapter->stats.last_vfmprc;
2041 }
2042
2043 static void ixgbevf_negotiate_api(struct ixgbevf_adapter *adapter)
2044 {
2045         struct ixgbe_hw *hw = &adapter->hw;
2046         int api[] = { ixgbe_mbox_api_12,
2047                       ixgbe_mbox_api_11,
2048                       ixgbe_mbox_api_10,
2049                       ixgbe_mbox_api_unknown };
2050         int err, idx = 0;
2051
2052         spin_lock_bh(&adapter->mbx_lock);
2053
2054         while (api[idx] != ixgbe_mbox_api_unknown) {
2055                 err = ixgbevf_negotiate_api_version(hw, api[idx]);
2056                 if (!err)
2057                         break;
2058                 idx++;
2059         }
2060
2061         spin_unlock_bh(&adapter->mbx_lock);
2062 }
2063
2064 static void ixgbevf_up_complete(struct ixgbevf_adapter *adapter)
2065 {
2066         struct net_device *netdev = adapter->netdev;
2067         struct ixgbe_hw *hw = &adapter->hw;
2068
2069         ixgbevf_configure_msix(adapter);
2070
2071         spin_lock_bh(&adapter->mbx_lock);
2072
2073         if (is_valid_ether_addr(hw->mac.addr))
2074                 hw->mac.ops.set_rar(hw, 0, hw->mac.addr, 0);
2075         else
2076                 hw->mac.ops.set_rar(hw, 0, hw->mac.perm_addr, 0);
2077
2078         spin_unlock_bh(&adapter->mbx_lock);
2079
2080         smp_mb__before_atomic();
2081         clear_bit(__IXGBEVF_DOWN, &adapter->state);
2082         ixgbevf_napi_enable_all(adapter);
2083
2084         /* clear any pending interrupts, may auto mask */
2085         IXGBE_READ_REG(hw, IXGBE_VTEICR);
2086         ixgbevf_irq_enable(adapter);
2087
2088         /* enable transmits */
2089         netif_tx_start_all_queues(netdev);
2090
2091         ixgbevf_save_reset_stats(adapter);
2092         ixgbevf_init_last_counter_stats(adapter);
2093
2094         hw->mac.get_link_status = 1;
2095         mod_timer(&adapter->service_timer, jiffies);
2096 }
2097
2098 void ixgbevf_up(struct ixgbevf_adapter *adapter)
2099 {
2100         ixgbevf_configure(adapter);
2101
2102         ixgbevf_up_complete(adapter);
2103 }
2104
2105 /**
2106  * ixgbevf_clean_rx_ring - Free Rx Buffers per Queue
2107  * @rx_ring: ring to free buffers from
2108  **/
2109 static void ixgbevf_clean_rx_ring(struct ixgbevf_ring *rx_ring)
2110 {
2111         struct device *dev = rx_ring->dev;
2112         unsigned long size;
2113         unsigned int i;
2114
2115         /* Free Rx ring sk_buff */
2116         if (rx_ring->skb) {
2117                 dev_kfree_skb(rx_ring->skb);
2118                 rx_ring->skb = NULL;
2119         }
2120
2121         /* ring already cleared, nothing to do */
2122         if (!rx_ring->rx_buffer_info)
2123                 return;
2124
2125         /* Free all the Rx ring pages */
2126         for (i = 0; i < rx_ring->count; i++) {
2127                 struct ixgbevf_rx_buffer *rx_buffer;
2128
2129                 rx_buffer = &rx_ring->rx_buffer_info[i];
2130                 if (rx_buffer->dma)
2131                         dma_unmap_page(dev, rx_buffer->dma,
2132                                        PAGE_SIZE, DMA_FROM_DEVICE);
2133                 rx_buffer->dma = 0;
2134                 if (rx_buffer->page)
2135                         __free_page(rx_buffer->page);
2136                 rx_buffer->page = NULL;
2137         }
2138
2139         size = sizeof(struct ixgbevf_rx_buffer) * rx_ring->count;
2140         memset(rx_ring->rx_buffer_info, 0, size);
2141
2142         /* Zero out the descriptor ring */
2143         memset(rx_ring->desc, 0, rx_ring->size);
2144 }
2145
2146 /**
2147  * ixgbevf_clean_tx_ring - Free Tx Buffers
2148  * @tx_ring: ring to be cleaned
2149  **/
2150 static void ixgbevf_clean_tx_ring(struct ixgbevf_ring *tx_ring)
2151 {
2152         struct ixgbevf_tx_buffer *tx_buffer_info;
2153         unsigned long size;
2154         unsigned int i;
2155
2156         if (!tx_ring->tx_buffer_info)
2157                 return;
2158
2159         /* Free all the Tx ring sk_buffs */
2160         for (i = 0; i < tx_ring->count; i++) {
2161                 tx_buffer_info = &tx_ring->tx_buffer_info[i];
2162                 ixgbevf_unmap_and_free_tx_resource(tx_ring, tx_buffer_info);
2163         }
2164
2165         size = sizeof(struct ixgbevf_tx_buffer) * tx_ring->count;
2166         memset(tx_ring->tx_buffer_info, 0, size);
2167
2168         memset(tx_ring->desc, 0, tx_ring->size);
2169 }
2170
2171 /**
2172  * ixgbevf_clean_all_rx_rings - Free Rx Buffers for all queues
2173  * @adapter: board private structure
2174  **/
2175 static void ixgbevf_clean_all_rx_rings(struct ixgbevf_adapter *adapter)
2176 {
2177         int i;
2178
2179         for (i = 0; i < adapter->num_rx_queues; i++)
2180                 ixgbevf_clean_rx_ring(adapter->rx_ring[i]);
2181 }
2182
2183 /**
2184  * ixgbevf_clean_all_tx_rings - Free Tx Buffers for all queues
2185  * @adapter: board private structure
2186  **/
2187 static void ixgbevf_clean_all_tx_rings(struct ixgbevf_adapter *adapter)
2188 {
2189         int i;
2190
2191         for (i = 0; i < adapter->num_tx_queues; i++)
2192                 ixgbevf_clean_tx_ring(adapter->tx_ring[i]);
2193 }
2194
2195 void ixgbevf_down(struct ixgbevf_adapter *adapter)
2196 {
2197         struct net_device *netdev = adapter->netdev;
2198         struct ixgbe_hw *hw = &adapter->hw;
2199         int i;
2200
2201         /* signal that we are down to the interrupt handler */
2202         if (test_and_set_bit(__IXGBEVF_DOWN, &adapter->state))
2203                 return; /* do nothing if already down */
2204
2205         /* disable all enabled Rx queues */
2206         for (i = 0; i < adapter->num_rx_queues; i++)
2207                 ixgbevf_disable_rx_queue(adapter, adapter->rx_ring[i]);
2208
2209         usleep_range(10000, 20000);
2210
2211         netif_tx_stop_all_queues(netdev);
2212
2213         /* call carrier off first to avoid false dev_watchdog timeouts */
2214         netif_carrier_off(netdev);
2215         netif_tx_disable(netdev);
2216
2217         ixgbevf_irq_disable(adapter);
2218
2219         ixgbevf_napi_disable_all(adapter);
2220
2221         del_timer_sync(&adapter->service_timer);
2222
2223         /* disable transmits in the hardware now that interrupts are off */
2224         for (i = 0; i < adapter->num_tx_queues; i++) {
2225                 u8 reg_idx = adapter->tx_ring[i]->reg_idx;
2226
2227                 IXGBE_WRITE_REG(hw, IXGBE_VFTXDCTL(reg_idx),
2228                                 IXGBE_TXDCTL_SWFLSH);
2229         }
2230
2231         if (!pci_channel_offline(adapter->pdev))
2232                 ixgbevf_reset(adapter);
2233
2234         ixgbevf_clean_all_tx_rings(adapter);
2235         ixgbevf_clean_all_rx_rings(adapter);
2236 }
2237
2238 void ixgbevf_reinit_locked(struct ixgbevf_adapter *adapter)
2239 {
2240         WARN_ON(in_interrupt());
2241
2242         while (test_and_set_bit(__IXGBEVF_RESETTING, &adapter->state))
2243                 msleep(1);
2244
2245         ixgbevf_down(adapter);
2246         ixgbevf_up(adapter);
2247
2248         clear_bit(__IXGBEVF_RESETTING, &adapter->state);
2249 }
2250
2251 void ixgbevf_reset(struct ixgbevf_adapter *adapter)
2252 {
2253         struct ixgbe_hw *hw = &adapter->hw;
2254         struct net_device *netdev = adapter->netdev;
2255
2256         if (hw->mac.ops.reset_hw(hw)) {
2257                 hw_dbg(hw, "PF still resetting\n");
2258         } else {
2259                 hw->mac.ops.init_hw(hw);
2260                 ixgbevf_negotiate_api(adapter);
2261         }
2262
2263         if (is_valid_ether_addr(adapter->hw.mac.addr)) {
2264                 ether_addr_copy(netdev->dev_addr, adapter->hw.mac.addr);
2265                 ether_addr_copy(netdev->perm_addr, adapter->hw.mac.addr);
2266         }
2267
2268         adapter->last_reset = jiffies;
2269 }
2270
2271 static int ixgbevf_acquire_msix_vectors(struct ixgbevf_adapter *adapter,
2272                                         int vectors)
2273 {
2274         int vector_threshold;
2275
2276         /* We'll want at least 2 (vector_threshold):
2277          * 1) TxQ[0] + RxQ[0] handler
2278          * 2) Other (Link Status Change, etc.)
2279          */
2280         vector_threshold = MIN_MSIX_COUNT;
2281
2282         /* The more we get, the more we will assign to Tx/Rx Cleanup
2283          * for the separate queues...where Rx Cleanup >= Tx Cleanup.
2284          * Right now, we simply care about how many we'll get; we'll
2285          * set them up later while requesting irq's.
2286          */
2287         vectors = pci_enable_msix_range(adapter->pdev, adapter->msix_entries,
2288                                         vector_threshold, vectors);
2289
2290         if (vectors < 0) {
2291                 dev_err(&adapter->pdev->dev,
2292                         "Unable to allocate MSI-X interrupts\n");
2293                 kfree(adapter->msix_entries);
2294                 adapter->msix_entries = NULL;
2295                 return vectors;
2296         }
2297
2298         /* Adjust for only the vectors we'll use, which is minimum
2299          * of max_msix_q_vectors + NON_Q_VECTORS, or the number of
2300          * vectors we were allocated.
2301          */
2302         adapter->num_msix_vectors = vectors;
2303
2304         return 0;
2305 }
2306
2307 /**
2308  * ixgbevf_set_num_queues - Allocate queues for device, feature dependent
2309  * @adapter: board private structure to initialize
2310  *
2311  * This is the top level queue allocation routine.  The order here is very
2312  * important, starting with the "most" number of features turned on at once,
2313  * and ending with the smallest set of features.  This way large combinations
2314  * can be allocated if they're turned on, and smaller combinations are the
2315  * fallthrough conditions.
2316  *
2317  **/
2318 static void ixgbevf_set_num_queues(struct ixgbevf_adapter *adapter)
2319 {
2320         struct ixgbe_hw *hw = &adapter->hw;
2321         unsigned int def_q = 0;
2322         unsigned int num_tcs = 0;
2323         int err;
2324
2325         /* Start with base case */
2326         adapter->num_rx_queues = 1;
2327         adapter->num_tx_queues = 1;
2328
2329         spin_lock_bh(&adapter->mbx_lock);
2330
2331         /* fetch queue configuration from the PF */
2332         err = ixgbevf_get_queues(hw, &num_tcs, &def_q);
2333
2334         spin_unlock_bh(&adapter->mbx_lock);
2335
2336         if (err)
2337                 return;
2338
2339         /* we need as many queues as traffic classes */
2340         if (num_tcs > 1) {
2341                 adapter->num_rx_queues = num_tcs;
2342         } else {
2343                 u16 rss = min_t(u16, num_online_cpus(), IXGBEVF_MAX_RSS_QUEUES);
2344
2345                 switch (hw->api_version) {
2346                 case ixgbe_mbox_api_11:
2347                 case ixgbe_mbox_api_12:
2348                         adapter->num_rx_queues = rss;
2349                         adapter->num_tx_queues = rss;
2350                 default:
2351                         break;
2352                 }
2353         }
2354 }
2355
2356 /**
2357  * ixgbevf_alloc_queues - Allocate memory for all rings
2358  * @adapter: board private structure to initialize
2359  *
2360  * We allocate one ring per queue at run-time since we don't know the
2361  * number of queues at compile-time.  The polling_netdev array is
2362  * intended for Multiqueue, but should work fine with a single queue.
2363  **/
2364 static int ixgbevf_alloc_queues(struct ixgbevf_adapter *adapter)
2365 {
2366         struct ixgbevf_ring *ring;
2367         int rx = 0, tx = 0;
2368
2369         for (; tx < adapter->num_tx_queues; tx++) {
2370                 ring = kzalloc(sizeof(*ring), GFP_KERNEL);
2371                 if (!ring)
2372                         goto err_allocation;
2373
2374                 ring->dev = &adapter->pdev->dev;
2375                 ring->netdev = adapter->netdev;
2376                 ring->count = adapter->tx_ring_count;
2377                 ring->queue_index = tx;
2378                 ring->reg_idx = tx;
2379
2380                 adapter->tx_ring[tx] = ring;
2381         }
2382
2383         for (; rx < adapter->num_rx_queues; rx++) {
2384                 ring = kzalloc(sizeof(*ring), GFP_KERNEL);
2385                 if (!ring)
2386                         goto err_allocation;
2387
2388                 ring->dev = &adapter->pdev->dev;
2389                 ring->netdev = adapter->netdev;
2390
2391                 ring->count = adapter->rx_ring_count;
2392                 ring->queue_index = rx;
2393                 ring->reg_idx = rx;
2394
2395                 adapter->rx_ring[rx] = ring;
2396         }
2397
2398         return 0;
2399
2400 err_allocation:
2401         while (tx) {
2402                 kfree(adapter->tx_ring[--tx]);
2403                 adapter->tx_ring[tx] = NULL;
2404         }
2405
2406         while (rx) {
2407                 kfree(adapter->rx_ring[--rx]);
2408                 adapter->rx_ring[rx] = NULL;
2409         }
2410         return -ENOMEM;
2411 }
2412
2413 /**
2414  * ixgbevf_set_interrupt_capability - set MSI-X or FAIL if not supported
2415  * @adapter: board private structure to initialize
2416  *
2417  * Attempt to configure the interrupts using the best available
2418  * capabilities of the hardware and the kernel.
2419  **/
2420 static int ixgbevf_set_interrupt_capability(struct ixgbevf_adapter *adapter)
2421 {
2422         struct net_device *netdev = adapter->netdev;
2423         int err;
2424         int vector, v_budget;
2425
2426         /* It's easy to be greedy for MSI-X vectors, but it really
2427          * doesn't do us much good if we have a lot more vectors
2428          * than CPU's.  So let's be conservative and only ask for
2429          * (roughly) the same number of vectors as there are CPU's.
2430          * The default is to use pairs of vectors.
2431          */
2432         v_budget = max(adapter->num_rx_queues, adapter->num_tx_queues);
2433         v_budget = min_t(int, v_budget, num_online_cpus());
2434         v_budget += NON_Q_VECTORS;
2435
2436         /* A failure in MSI-X entry allocation isn't fatal, but it does
2437          * mean we disable MSI-X capabilities of the adapter.
2438          */
2439         adapter->msix_entries = kcalloc(v_budget,
2440                                         sizeof(struct msix_entry), GFP_KERNEL);
2441         if (!adapter->msix_entries)
2442                 return -ENOMEM;
2443
2444         for (vector = 0; vector < v_budget; vector++)
2445                 adapter->msix_entries[vector].entry = vector;
2446
2447         err = ixgbevf_acquire_msix_vectors(adapter, v_budget);
2448         if (err)
2449                 return err;
2450
2451         err = netif_set_real_num_tx_queues(netdev, adapter->num_tx_queues);
2452         if (err)
2453                 return err;
2454
2455         return netif_set_real_num_rx_queues(netdev, adapter->num_rx_queues);
2456 }
2457
2458 /**
2459  * ixgbevf_alloc_q_vectors - Allocate memory for interrupt vectors
2460  * @adapter: board private structure to initialize
2461  *
2462  * We allocate one q_vector per queue interrupt.  If allocation fails we
2463  * return -ENOMEM.
2464  **/
2465 static int ixgbevf_alloc_q_vectors(struct ixgbevf_adapter *adapter)
2466 {
2467         int q_idx, num_q_vectors;
2468         struct ixgbevf_q_vector *q_vector;
2469
2470         num_q_vectors = adapter->num_msix_vectors - NON_Q_VECTORS;
2471
2472         for (q_idx = 0; q_idx < num_q_vectors; q_idx++) {
2473                 q_vector = kzalloc(sizeof(struct ixgbevf_q_vector), GFP_KERNEL);
2474                 if (!q_vector)
2475                         goto err_out;
2476                 q_vector->adapter = adapter;
2477                 q_vector->v_idx = q_idx;
2478                 netif_napi_add(adapter->netdev, &q_vector->napi,
2479                                ixgbevf_poll, 64);
2480                 adapter->q_vector[q_idx] = q_vector;
2481         }
2482
2483         return 0;
2484
2485 err_out:
2486         while (q_idx) {
2487                 q_idx--;
2488                 q_vector = adapter->q_vector[q_idx];
2489 #ifdef CONFIG_NET_RX_BUSY_POLL
2490                 napi_hash_del(&q_vector->napi);
2491 #endif
2492                 netif_napi_del(&q_vector->napi);
2493                 kfree(q_vector);
2494                 adapter->q_vector[q_idx] = NULL;
2495         }
2496         return -ENOMEM;
2497 }
2498
2499 /**
2500  * ixgbevf_free_q_vectors - Free memory allocated for interrupt vectors
2501  * @adapter: board private structure to initialize
2502  *
2503  * This function frees the memory allocated to the q_vectors.  In addition if
2504  * NAPI is enabled it will delete any references to the NAPI struct prior
2505  * to freeing the q_vector.
2506  **/
2507 static void ixgbevf_free_q_vectors(struct ixgbevf_adapter *adapter)
2508 {
2509         int q_idx, num_q_vectors = adapter->num_msix_vectors - NON_Q_VECTORS;
2510
2511         for (q_idx = 0; q_idx < num_q_vectors; q_idx++) {
2512                 struct ixgbevf_q_vector *q_vector = adapter->q_vector[q_idx];
2513
2514                 adapter->q_vector[q_idx] = NULL;
2515 #ifdef CONFIG_NET_RX_BUSY_POLL
2516                 napi_hash_del(&q_vector->napi);
2517 #endif
2518                 netif_napi_del(&q_vector->napi);
2519                 kfree(q_vector);
2520         }
2521 }
2522
2523 /**
2524  * ixgbevf_reset_interrupt_capability - Reset MSIX setup
2525  * @adapter: board private structure
2526  *
2527  **/
2528 static void ixgbevf_reset_interrupt_capability(struct ixgbevf_adapter *adapter)
2529 {
2530         pci_disable_msix(adapter->pdev);
2531         kfree(adapter->msix_entries);
2532         adapter->msix_entries = NULL;
2533 }
2534
2535 /**
2536  * ixgbevf_init_interrupt_scheme - Determine if MSIX is supported and init
2537  * @adapter: board private structure to initialize
2538  *
2539  **/
2540 static int ixgbevf_init_interrupt_scheme(struct ixgbevf_adapter *adapter)
2541 {
2542         int err;
2543
2544         /* Number of supported queues */
2545         ixgbevf_set_num_queues(adapter);
2546
2547         err = ixgbevf_set_interrupt_capability(adapter);
2548         if (err) {
2549                 hw_dbg(&adapter->hw,
2550                        "Unable to setup interrupt capabilities\n");
2551                 goto err_set_interrupt;
2552         }
2553
2554         err = ixgbevf_alloc_q_vectors(adapter);
2555         if (err) {
2556                 hw_dbg(&adapter->hw, "Unable to allocate memory for queue vectors\n");
2557                 goto err_alloc_q_vectors;
2558         }
2559
2560         err = ixgbevf_alloc_queues(adapter);
2561         if (err) {
2562                 pr_err("Unable to allocate memory for queues\n");
2563                 goto err_alloc_queues;
2564         }
2565
2566         hw_dbg(&adapter->hw, "Multiqueue %s: Rx Queue count = %u, Tx Queue count = %u\n",
2567                (adapter->num_rx_queues > 1) ? "Enabled" :
2568                "Disabled", adapter->num_rx_queues, adapter->num_tx_queues);
2569
2570         set_bit(__IXGBEVF_DOWN, &adapter->state);
2571
2572         return 0;
2573 err_alloc_queues:
2574         ixgbevf_free_q_vectors(adapter);
2575 err_alloc_q_vectors:
2576         ixgbevf_reset_interrupt_capability(adapter);
2577 err_set_interrupt:
2578         return err;
2579 }
2580
2581 /**
2582  * ixgbevf_clear_interrupt_scheme - Clear the current interrupt scheme settings
2583  * @adapter: board private structure to clear interrupt scheme on
2584  *
2585  * We go through and clear interrupt specific resources and reset the structure
2586  * to pre-load conditions
2587  **/
2588 static void ixgbevf_clear_interrupt_scheme(struct ixgbevf_adapter *adapter)
2589 {
2590         int i;
2591
2592         for (i = 0; i < adapter->num_tx_queues; i++) {
2593                 kfree(adapter->tx_ring[i]);
2594                 adapter->tx_ring[i] = NULL;
2595         }
2596         for (i = 0; i < adapter->num_rx_queues; i++) {
2597                 kfree(adapter->rx_ring[i]);
2598                 adapter->rx_ring[i] = NULL;
2599         }
2600
2601         adapter->num_tx_queues = 0;
2602         adapter->num_rx_queues = 0;
2603
2604         ixgbevf_free_q_vectors(adapter);
2605         ixgbevf_reset_interrupt_capability(adapter);
2606 }
2607
2608 /**
2609  * ixgbevf_sw_init - Initialize general software structures
2610  * @adapter: board private structure to initialize
2611  *
2612  * ixgbevf_sw_init initializes the Adapter private data structure.
2613  * Fields are initialized based on PCI device information and
2614  * OS network device settings (MTU size).
2615  **/
2616 static int ixgbevf_sw_init(struct ixgbevf_adapter *adapter)
2617 {
2618         struct ixgbe_hw *hw = &adapter->hw;
2619         struct pci_dev *pdev = adapter->pdev;
2620         struct net_device *netdev = adapter->netdev;
2621         int err;
2622
2623         /* PCI config space info */
2624         hw->vendor_id = pdev->vendor;
2625         hw->device_id = pdev->device;
2626         hw->revision_id = pdev->revision;
2627         hw->subsystem_vendor_id = pdev->subsystem_vendor;
2628         hw->subsystem_device_id = pdev->subsystem_device;
2629
2630         hw->mbx.ops.init_params(hw);
2631
2632         /* assume legacy case in which PF would only give VF 2 queues */
2633         hw->mac.max_tx_queues = 2;
2634         hw->mac.max_rx_queues = 2;
2635
2636         /* lock to protect mailbox accesses */
2637         spin_lock_init(&adapter->mbx_lock);
2638
2639         err = hw->mac.ops.reset_hw(hw);
2640         if (err) {
2641                 dev_info(&pdev->dev,
2642                          "PF still in reset state.  Is the PF interface up?\n");
2643         } else {
2644                 err = hw->mac.ops.init_hw(hw);
2645                 if (err) {
2646                         pr_err("init_shared_code failed: %d\n", err);
2647                         goto out;
2648                 }
2649                 ixgbevf_negotiate_api(adapter);
2650                 err = hw->mac.ops.get_mac_addr(hw, hw->mac.addr);
2651                 if (err)
2652                         dev_info(&pdev->dev, "Error reading MAC address\n");
2653                 else if (is_zero_ether_addr(adapter->hw.mac.addr))
2654                         dev_info(&pdev->dev,
2655                                  "MAC address not assigned by administrator.\n");
2656                 ether_addr_copy(netdev->dev_addr, hw->mac.addr);
2657         }
2658
2659         if (!is_valid_ether_addr(netdev->dev_addr)) {
2660                 dev_info(&pdev->dev, "Assigning random MAC address\n");
2661                 eth_hw_addr_random(netdev);
2662                 ether_addr_copy(hw->mac.addr, netdev->dev_addr);
2663                 ether_addr_copy(hw->mac.perm_addr, netdev->dev_addr);
2664         }
2665
2666         /* Enable dynamic interrupt throttling rates */
2667         adapter->rx_itr_setting = 1;
2668         adapter->tx_itr_setting = 1;
2669
2670         /* set default ring sizes */
2671         adapter->tx_ring_count = IXGBEVF_DEFAULT_TXD;
2672         adapter->rx_ring_count = IXGBEVF_DEFAULT_RXD;
2673
2674         set_bit(__IXGBEVF_DOWN, &adapter->state);
2675         return 0;
2676
2677 out:
2678         return err;
2679 }
2680
2681 #define UPDATE_VF_COUNTER_32bit(reg, last_counter, counter)     \
2682         {                                                       \
2683                 u32 current_counter = IXGBE_READ_REG(hw, reg);  \
2684                 if (current_counter < last_counter)             \
2685                         counter += 0x100000000LL;               \
2686                 last_counter = current_counter;                 \
2687                 counter &= 0xFFFFFFFF00000000LL;                \
2688                 counter |= current_counter;                     \
2689         }
2690
2691 #define UPDATE_VF_COUNTER_36bit(reg_lsb, reg_msb, last_counter, counter) \
2692         {                                                                \
2693                 u64 current_counter_lsb = IXGBE_READ_REG(hw, reg_lsb);   \
2694                 u64 current_counter_msb = IXGBE_READ_REG(hw, reg_msb);   \
2695                 u64 current_counter = (current_counter_msb << 32) |      \
2696                         current_counter_lsb;                             \
2697                 if (current_counter < last_counter)                      \
2698                         counter += 0x1000000000LL;                       \
2699                 last_counter = current_counter;                          \
2700                 counter &= 0xFFFFFFF000000000LL;                         \
2701                 counter |= current_counter;                              \
2702         }
2703 /**
2704  * ixgbevf_update_stats - Update the board statistics counters.
2705  * @adapter: board private structure
2706  **/
2707 void ixgbevf_update_stats(struct ixgbevf_adapter *adapter)
2708 {
2709         struct ixgbe_hw *hw = &adapter->hw;
2710         int i;
2711
2712         if (test_bit(__IXGBEVF_DOWN, &adapter->state) ||
2713             test_bit(__IXGBEVF_RESETTING, &adapter->state))
2714                 return;
2715
2716         UPDATE_VF_COUNTER_32bit(IXGBE_VFGPRC, adapter->stats.last_vfgprc,
2717                                 adapter->stats.vfgprc);
2718         UPDATE_VF_COUNTER_32bit(IXGBE_VFGPTC, adapter->stats.last_vfgptc,
2719                                 adapter->stats.vfgptc);
2720         UPDATE_VF_COUNTER_36bit(IXGBE_VFGORC_LSB, IXGBE_VFGORC_MSB,
2721                                 adapter->stats.last_vfgorc,
2722                                 adapter->stats.vfgorc);
2723         UPDATE_VF_COUNTER_36bit(IXGBE_VFGOTC_LSB, IXGBE_VFGOTC_MSB,
2724                                 adapter->stats.last_vfgotc,
2725                                 adapter->stats.vfgotc);
2726         UPDATE_VF_COUNTER_32bit(IXGBE_VFMPRC, adapter->stats.last_vfmprc,
2727                                 adapter->stats.vfmprc);
2728
2729         for (i = 0;  i  < adapter->num_rx_queues;  i++) {
2730                 adapter->hw_csum_rx_error +=
2731                         adapter->rx_ring[i]->hw_csum_rx_error;
2732                 adapter->rx_ring[i]->hw_csum_rx_error = 0;
2733         }
2734 }
2735
2736 /**
2737  * ixgbevf_service_timer - Timer Call-back
2738  * @data: pointer to adapter cast into an unsigned long
2739  **/
2740 static void ixgbevf_service_timer(unsigned long data)
2741 {
2742         struct ixgbevf_adapter *adapter = (struct ixgbevf_adapter *)data;
2743
2744         /* Reset the timer */
2745         mod_timer(&adapter->service_timer, (HZ * 2) + jiffies);
2746
2747         ixgbevf_service_event_schedule(adapter);
2748 }
2749
2750 static void ixgbevf_reset_subtask(struct ixgbevf_adapter *adapter)
2751 {
2752         if (!(adapter->flags & IXGBEVF_FLAG_RESET_REQUESTED))
2753                 return;
2754
2755         adapter->flags &= ~IXGBEVF_FLAG_RESET_REQUESTED;
2756
2757         /* If we're already down or resetting, just bail */
2758         if (test_bit(__IXGBEVF_DOWN, &adapter->state) ||
2759             test_bit(__IXGBEVF_RESETTING, &adapter->state))
2760                 return;
2761
2762         adapter->tx_timeout_count++;
2763
2764         ixgbevf_reinit_locked(adapter);
2765 }
2766
2767 /**
2768  * ixgbevf_check_hang_subtask - check for hung queues and dropped interrupts
2769  * @adapter: pointer to the device adapter structure
2770  *
2771  * This function serves two purposes.  First it strobes the interrupt lines
2772  * in order to make certain interrupts are occurring.  Secondly it sets the
2773  * bits needed to check for TX hangs.  As a result we should immediately
2774  * determine if a hang has occurred.
2775  **/
2776 static void ixgbevf_check_hang_subtask(struct ixgbevf_adapter *adapter)
2777 {
2778         struct ixgbe_hw *hw = &adapter->hw;
2779         u32 eics = 0;
2780         int i;
2781
2782         /* If we're down or resetting, just bail */
2783         if (test_bit(__IXGBEVF_DOWN, &adapter->state) ||
2784             test_bit(__IXGBEVF_RESETTING, &adapter->state))
2785                 return;
2786
2787         /* Force detection of hung controller */
2788         if (netif_carrier_ok(adapter->netdev)) {
2789                 for (i = 0; i < adapter->num_tx_queues; i++)
2790                         set_check_for_tx_hang(adapter->tx_ring[i]);
2791         }
2792
2793         /* get one bit for every active Tx/Rx interrupt vector */
2794         for (i = 0; i < adapter->num_msix_vectors - NON_Q_VECTORS; i++) {
2795                 struct ixgbevf_q_vector *qv = adapter->q_vector[i];
2796
2797                 if (qv->rx.ring || qv->tx.ring)
2798                         eics |= 1 << i;
2799         }
2800
2801         /* Cause software interrupt to ensure rings are cleaned */
2802         IXGBE_WRITE_REG(hw, IXGBE_VTEICS, eics);
2803 }
2804
2805 /**
2806  * ixgbevf_watchdog_update_link - update the link status
2807  * @adapter: pointer to the device adapter structure
2808  **/
2809 static void ixgbevf_watchdog_update_link(struct ixgbevf_adapter *adapter)
2810 {
2811         struct ixgbe_hw *hw = &adapter->hw;
2812         u32 link_speed = adapter->link_speed;
2813         bool link_up = adapter->link_up;
2814         s32 err;
2815
2816         spin_lock_bh(&adapter->mbx_lock);
2817
2818         err = hw->mac.ops.check_link(hw, &link_speed, &link_up, false);
2819
2820         spin_unlock_bh(&adapter->mbx_lock);
2821
2822         /* if check for link returns error we will need to reset */
2823         if (err && time_after(jiffies, adapter->last_reset + (10 * HZ))) {
2824                 adapter->flags |= IXGBEVF_FLAG_RESET_REQUESTED;
2825                 link_up = false;
2826         }
2827
2828         adapter->link_up = link_up;
2829         adapter->link_speed = link_speed;
2830 }
2831
2832 /**
2833  * ixgbevf_watchdog_link_is_up - update netif_carrier status and
2834  *                               print link up message
2835  * @adapter: pointer to the device adapter structure
2836  **/
2837 static void ixgbevf_watchdog_link_is_up(struct ixgbevf_adapter *adapter)
2838 {
2839         struct net_device *netdev = adapter->netdev;
2840
2841         /* only continue if link was previously down */
2842         if (netif_carrier_ok(netdev))
2843                 return;
2844
2845         dev_info(&adapter->pdev->dev, "NIC Link is Up %s\n",
2846                  (adapter->link_speed == IXGBE_LINK_SPEED_10GB_FULL) ?
2847                  "10 Gbps" :
2848                  (adapter->link_speed == IXGBE_LINK_SPEED_1GB_FULL) ?
2849                  "1 Gbps" :
2850                  (adapter->link_speed == IXGBE_LINK_SPEED_100_FULL) ?
2851                  "100 Mbps" :
2852                  "unknown speed");
2853
2854         netif_carrier_on(netdev);
2855 }
2856
2857 /**
2858  * ixgbevf_watchdog_link_is_down - update netif_carrier status and
2859  *                                 print link down message
2860  * @adapter: pointer to the adapter structure
2861  **/
2862 static void ixgbevf_watchdog_link_is_down(struct ixgbevf_adapter *adapter)
2863 {
2864         struct net_device *netdev = adapter->netdev;
2865
2866         adapter->link_speed = 0;
2867
2868         /* only continue if link was up previously */
2869         if (!netif_carrier_ok(netdev))
2870                 return;
2871
2872         dev_info(&adapter->pdev->dev, "NIC Link is Down\n");
2873
2874         netif_carrier_off(netdev);
2875 }
2876
2877 /**
2878  * ixgbevf_watchdog_subtask - worker thread to bring link up
2879  * @work: pointer to work_struct containing our data
2880  **/
2881 static void ixgbevf_watchdog_subtask(struct ixgbevf_adapter *adapter)
2882 {
2883         /* if interface is down do nothing */
2884         if (test_bit(__IXGBEVF_DOWN, &adapter->state) ||
2885             test_bit(__IXGBEVF_RESETTING, &adapter->state))
2886                 return;
2887
2888         ixgbevf_watchdog_update_link(adapter);
2889
2890         if (adapter->link_up)
2891                 ixgbevf_watchdog_link_is_up(adapter);
2892         else
2893                 ixgbevf_watchdog_link_is_down(adapter);
2894
2895         ixgbevf_update_stats(adapter);
2896 }
2897
2898 /**
2899  * ixgbevf_service_task - manages and runs subtasks
2900  * @work: pointer to work_struct containing our data
2901  **/
2902 static void ixgbevf_service_task(struct work_struct *work)
2903 {
2904         struct ixgbevf_adapter *adapter = container_of(work,
2905                                                        struct ixgbevf_adapter,
2906                                                        service_task);
2907         struct ixgbe_hw *hw = &adapter->hw;
2908
2909         if (IXGBE_REMOVED(hw->hw_addr)) {
2910                 if (!test_bit(__IXGBEVF_DOWN, &adapter->state)) {
2911                         rtnl_lock();
2912                         ixgbevf_down(adapter);
2913                         rtnl_unlock();
2914                 }
2915                 return;
2916         }
2917
2918         ixgbevf_queue_reset_subtask(adapter);
2919         ixgbevf_reset_subtask(adapter);
2920         ixgbevf_watchdog_subtask(adapter);
2921         ixgbevf_check_hang_subtask(adapter);
2922
2923         ixgbevf_service_event_complete(adapter);
2924 }
2925
2926 /**
2927  * ixgbevf_free_tx_resources - Free Tx Resources per Queue
2928  * @tx_ring: Tx descriptor ring for a specific queue
2929  *
2930  * Free all transmit software resources
2931  **/
2932 void ixgbevf_free_tx_resources(struct ixgbevf_ring *tx_ring)
2933 {
2934         ixgbevf_clean_tx_ring(tx_ring);
2935
2936         vfree(tx_ring->tx_buffer_info);
2937         tx_ring->tx_buffer_info = NULL;
2938
2939         /* if not set, then don't free */
2940         if (!tx_ring->desc)
2941                 return;
2942
2943         dma_free_coherent(tx_ring->dev, tx_ring->size, tx_ring->desc,
2944                           tx_ring->dma);
2945
2946         tx_ring->desc = NULL;
2947 }
2948
2949 /**
2950  * ixgbevf_free_all_tx_resources - Free Tx Resources for All Queues
2951  * @adapter: board private structure
2952  *
2953  * Free all transmit software resources
2954  **/
2955 static void ixgbevf_free_all_tx_resources(struct ixgbevf_adapter *adapter)
2956 {
2957         int i;
2958
2959         for (i = 0; i < adapter->num_tx_queues; i++)
2960                 if (adapter->tx_ring[i]->desc)
2961                         ixgbevf_free_tx_resources(adapter->tx_ring[i]);
2962 }
2963
2964 /**
2965  * ixgbevf_setup_tx_resources - allocate Tx resources (Descriptors)
2966  * @tx_ring: Tx descriptor ring (for a specific queue) to setup
2967  *
2968  * Return 0 on success, negative on failure
2969  **/
2970 int ixgbevf_setup_tx_resources(struct ixgbevf_ring *tx_ring)
2971 {
2972         int size;
2973
2974         size = sizeof(struct ixgbevf_tx_buffer) * tx_ring->count;
2975         tx_ring->tx_buffer_info = vzalloc(size);
2976         if (!tx_ring->tx_buffer_info)
2977                 goto err;
2978
2979         /* round up to nearest 4K */
2980         tx_ring->size = tx_ring->count * sizeof(union ixgbe_adv_tx_desc);
2981         tx_ring->size = ALIGN(tx_ring->size, 4096);
2982
2983         tx_ring->desc = dma_alloc_coherent(tx_ring->dev, tx_ring->size,
2984                                            &tx_ring->dma, GFP_KERNEL);
2985         if (!tx_ring->desc)
2986                 goto err;
2987
2988         return 0;
2989
2990 err:
2991         vfree(tx_ring->tx_buffer_info);
2992         tx_ring->tx_buffer_info = NULL;
2993         hw_dbg(&adapter->hw, "Unable to allocate memory for the transmit descriptor ring\n");
2994         return -ENOMEM;
2995 }
2996
2997 /**
2998  * ixgbevf_setup_all_tx_resources - allocate all queues Tx resources
2999  * @adapter: board private structure
3000  *
3001  * If this function returns with an error, then it's possible one or
3002  * more of the rings is populated (while the rest are not).  It is the
3003  * callers duty to clean those orphaned rings.
3004  *
3005  * Return 0 on success, negative on failure
3006  **/
3007 static int ixgbevf_setup_all_tx_resources(struct ixgbevf_adapter *adapter)
3008 {
3009         int i, err = 0;
3010
3011         for (i = 0; i < adapter->num_tx_queues; i++) {
3012                 err = ixgbevf_setup_tx_resources(adapter->tx_ring[i]);
3013                 if (!err)
3014                         continue;
3015                 hw_dbg(&adapter->hw, "Allocation for Tx Queue %u failed\n", i);
3016                 break;
3017         }
3018
3019         return err;
3020 }
3021
3022 /**
3023  * ixgbevf_setup_rx_resources - allocate Rx resources (Descriptors)
3024  * @rx_ring: Rx descriptor ring (for a specific queue) to setup
3025  *
3026  * Returns 0 on success, negative on failure
3027  **/
3028 int ixgbevf_setup_rx_resources(struct ixgbevf_ring *rx_ring)
3029 {
3030         int size;
3031
3032         size = sizeof(struct ixgbevf_rx_buffer) * rx_ring->count;
3033         rx_ring->rx_buffer_info = vzalloc(size);
3034         if (!rx_ring->rx_buffer_info)
3035                 goto err;
3036
3037         /* Round up to nearest 4K */
3038         rx_ring->size = rx_ring->count * sizeof(union ixgbe_adv_rx_desc);
3039         rx_ring->size = ALIGN(rx_ring->size, 4096);
3040
3041         rx_ring->desc = dma_alloc_coherent(rx_ring->dev, rx_ring->size,
3042                                            &rx_ring->dma, GFP_KERNEL);
3043
3044         if (!rx_ring->desc)
3045                 goto err;
3046
3047         return 0;
3048 err:
3049         vfree(rx_ring->rx_buffer_info);
3050         rx_ring->rx_buffer_info = NULL;
3051         dev_err(rx_ring->dev, "Unable to allocate memory for the Rx descriptor ring\n");
3052         return -ENOMEM;
3053 }
3054
3055 /**
3056  * ixgbevf_setup_all_rx_resources - allocate all queues Rx resources
3057  * @adapter: board private structure
3058  *
3059  * If this function returns with an error, then it's possible one or
3060  * more of the rings is populated (while the rest are not).  It is the
3061  * callers duty to clean those orphaned rings.
3062  *
3063  * Return 0 on success, negative on failure
3064  **/
3065 static int ixgbevf_setup_all_rx_resources(struct ixgbevf_adapter *adapter)
3066 {
3067         int i, err = 0;
3068
3069         for (i = 0; i < adapter->num_rx_queues; i++) {
3070                 err = ixgbevf_setup_rx_resources(adapter->rx_ring[i]);
3071                 if (!err)
3072                         continue;
3073                 hw_dbg(&adapter->hw, "Allocation for Rx Queue %u failed\n", i);
3074                 break;
3075         }
3076         return err;
3077 }
3078
3079 /**
3080  * ixgbevf_free_rx_resources - Free Rx Resources
3081  * @rx_ring: ring to clean the resources from
3082  *
3083  * Free all receive software resources
3084  **/
3085 void ixgbevf_free_rx_resources(struct ixgbevf_ring *rx_ring)
3086 {
3087         ixgbevf_clean_rx_ring(rx_ring);
3088
3089         vfree(rx_ring->rx_buffer_info);
3090         rx_ring->rx_buffer_info = NULL;
3091
3092         dma_free_coherent(rx_ring->dev, rx_ring->size, rx_ring->desc,
3093                           rx_ring->dma);
3094
3095         rx_ring->desc = NULL;
3096 }
3097
3098 /**
3099  * ixgbevf_free_all_rx_resources - Free Rx Resources for All Queues
3100  * @adapter: board private structure
3101  *
3102  * Free all receive software resources
3103  **/
3104 static void ixgbevf_free_all_rx_resources(struct ixgbevf_adapter *adapter)
3105 {
3106         int i;
3107
3108         for (i = 0; i < adapter->num_rx_queues; i++)
3109                 if (adapter->rx_ring[i]->desc)
3110                         ixgbevf_free_rx_resources(adapter->rx_ring[i]);
3111 }
3112
3113 /**
3114  * ixgbevf_open - Called when a network interface is made active
3115  * @netdev: network interface device structure
3116  *
3117  * Returns 0 on success, negative value on failure
3118  *
3119  * The open entry point is called when a network interface is made
3120  * active by the system (IFF_UP).  At this point all resources needed
3121  * for transmit and receive operations are allocated, the interrupt
3122  * handler is registered with the OS, the watchdog timer is started,
3123  * and the stack is notified that the interface is ready.
3124  **/
3125 static int ixgbevf_open(struct net_device *netdev)
3126 {
3127         struct ixgbevf_adapter *adapter = netdev_priv(netdev);
3128         struct ixgbe_hw *hw = &adapter->hw;
3129         int err;
3130
3131         /* A previous failure to open the device because of a lack of
3132          * available MSIX vector resources may have reset the number
3133          * of msix vectors variable to zero.  The only way to recover
3134          * is to unload/reload the driver and hope that the system has
3135          * been able to recover some MSIX vector resources.
3136          */
3137         if (!adapter->num_msix_vectors)
3138                 return -ENOMEM;
3139
3140         if (hw->adapter_stopped) {
3141                 ixgbevf_reset(adapter);
3142                 /* if adapter is still stopped then PF isn't up and
3143                  * the VF can't start.
3144                  */
3145                 if (hw->adapter_stopped) {
3146                         err = IXGBE_ERR_MBX;
3147                         pr_err("Unable to start - perhaps the PF Driver isn't up yet\n");
3148                         goto err_setup_reset;
3149                 }
3150         }
3151
3152         /* disallow open during test */
3153         if (test_bit(__IXGBEVF_TESTING, &adapter->state))
3154                 return -EBUSY;
3155
3156         netif_carrier_off(netdev);
3157
3158         /* allocate transmit descriptors */
3159         err = ixgbevf_setup_all_tx_resources(adapter);
3160         if (err)
3161                 goto err_setup_tx;
3162
3163         /* allocate receive descriptors */
3164         err = ixgbevf_setup_all_rx_resources(adapter);
3165         if (err)
3166                 goto err_setup_rx;
3167
3168         ixgbevf_configure(adapter);
3169
3170         /* Map the Tx/Rx rings to the vectors we were allotted.
3171          * if request_irq will be called in this function map_rings
3172          * must be called *before* up_complete
3173          */
3174         ixgbevf_map_rings_to_vectors(adapter);
3175
3176         err = ixgbevf_request_irq(adapter);
3177         if (err)
3178                 goto err_req_irq;
3179
3180         ixgbevf_up_complete(adapter);
3181
3182         return 0;
3183
3184 err_req_irq:
3185         ixgbevf_down(adapter);
3186 err_setup_rx:
3187         ixgbevf_free_all_rx_resources(adapter);
3188 err_setup_tx:
3189         ixgbevf_free_all_tx_resources(adapter);
3190         ixgbevf_reset(adapter);
3191
3192 err_setup_reset:
3193
3194         return err;
3195 }
3196
3197 /**
3198  * ixgbevf_close - Disables a network interface
3199  * @netdev: network interface device structure
3200  *
3201  * Returns 0, this is not allowed to fail
3202  *
3203  * The close entry point is called when an interface is de-activated
3204  * by the OS.  The hardware is still under the drivers control, but
3205  * needs to be disabled.  A global MAC reset is issued to stop the
3206  * hardware, and all transmit and receive resources are freed.
3207  **/
3208 static int ixgbevf_close(struct net_device *netdev)
3209 {
3210         struct ixgbevf_adapter *adapter = netdev_priv(netdev);
3211
3212         ixgbevf_down(adapter);
3213         ixgbevf_free_irq(adapter);
3214
3215         ixgbevf_free_all_tx_resources(adapter);
3216         ixgbevf_free_all_rx_resources(adapter);
3217
3218         return 0;
3219 }
3220
3221 static void ixgbevf_queue_reset_subtask(struct ixgbevf_adapter *adapter)
3222 {
3223         struct net_device *dev = adapter->netdev;
3224
3225         if (!(adapter->flags & IXGBEVF_FLAG_QUEUE_RESET_REQUESTED))
3226                 return;
3227
3228         adapter->flags &= ~IXGBEVF_FLAG_QUEUE_RESET_REQUESTED;
3229
3230         /* if interface is down do nothing */
3231         if (test_bit(__IXGBEVF_DOWN, &adapter->state) ||
3232             test_bit(__IXGBEVF_RESETTING, &adapter->state))
3233                 return;
3234
3235         /* Hardware has to reinitialize queues and interrupts to
3236          * match packet buffer alignment. Unfortunately, the
3237          * hardware is not flexible enough to do this dynamically.
3238          */
3239         if (netif_running(dev))
3240                 ixgbevf_close(dev);
3241
3242         ixgbevf_clear_interrupt_scheme(adapter);
3243         ixgbevf_init_interrupt_scheme(adapter);
3244
3245         if (netif_running(dev))
3246                 ixgbevf_open(dev);
3247 }
3248
3249 static void ixgbevf_tx_ctxtdesc(struct ixgbevf_ring *tx_ring,
3250                                 u32 vlan_macip_lens, u32 type_tucmd,
3251                                 u32 mss_l4len_idx)
3252 {
3253         struct ixgbe_adv_tx_context_desc *context_desc;
3254         u16 i = tx_ring->next_to_use;
3255
3256         context_desc = IXGBEVF_TX_CTXTDESC(tx_ring, i);
3257
3258         i++;
3259         tx_ring->next_to_use = (i < tx_ring->count) ? i : 0;
3260
3261         /* set bits to identify this as an advanced context descriptor */
3262         type_tucmd |= IXGBE_TXD_CMD_DEXT | IXGBE_ADVTXD_DTYP_CTXT;
3263
3264         context_desc->vlan_macip_lens   = cpu_to_le32(vlan_macip_lens);
3265         context_desc->seqnum_seed       = 0;
3266         context_desc->type_tucmd_mlhl   = cpu_to_le32(type_tucmd);
3267         context_desc->mss_l4len_idx     = cpu_to_le32(mss_l4len_idx);
3268 }
3269
3270 static int ixgbevf_tso(struct ixgbevf_ring *tx_ring,
3271                        struct ixgbevf_tx_buffer *first,
3272                        u8 *hdr_len)
3273 {
3274         struct sk_buff *skb = first->skb;
3275         u32 vlan_macip_lens, type_tucmd;
3276         u32 mss_l4len_idx, l4len;
3277         int err;
3278
3279         if (skb->ip_summed != CHECKSUM_PARTIAL)
3280                 return 0;
3281
3282         if (!skb_is_gso(skb))
3283                 return 0;
3284
3285         err = skb_cow_head(skb, 0);
3286         if (err < 0)
3287                 return err;
3288
3289         /* ADV DTYP TUCMD MKRLOC/ISCSIHEDLEN */
3290         type_tucmd = IXGBE_ADVTXD_TUCMD_L4T_TCP;
3291
3292         if (first->protocol == htons(ETH_P_IP)) {
3293                 struct iphdr *iph = ip_hdr(skb);
3294
3295                 iph->tot_len = 0;
3296                 iph->check = 0;
3297                 tcp_hdr(skb)->check = ~csum_tcpudp_magic(iph->saddr,
3298                                                          iph->daddr, 0,
3299                                                          IPPROTO_TCP,
3300                                                          0);
3301                 type_tucmd |= IXGBE_ADVTXD_TUCMD_IPV4;
3302                 first->tx_flags |= IXGBE_TX_FLAGS_TSO |
3303                                    IXGBE_TX_FLAGS_CSUM |
3304                                    IXGBE_TX_FLAGS_IPV4;
3305         } else if (skb_is_gso_v6(skb)) {
3306                 ipv6_hdr(skb)->payload_len = 0;
3307                 tcp_hdr(skb)->check =
3308                     ~csum_ipv6_magic(&ipv6_hdr(skb)->saddr,
3309                                      &ipv6_hdr(skb)->daddr,
3310                                      0, IPPROTO_TCP, 0);
3311                 first->tx_flags |= IXGBE_TX_FLAGS_TSO |
3312                                    IXGBE_TX_FLAGS_CSUM;
3313         }
3314
3315         /* compute header lengths */
3316         l4len = tcp_hdrlen(skb);
3317         *hdr_len += l4len;
3318         *hdr_len = skb_transport_offset(skb) + l4len;
3319
3320         /* update GSO size and bytecount with header size */
3321         first->gso_segs = skb_shinfo(skb)->gso_segs;
3322         first->bytecount += (first->gso_segs - 1) * *hdr_len;
3323
3324         /* mss_l4len_id: use 1 as index for TSO */
3325         mss_l4len_idx = l4len << IXGBE_ADVTXD_L4LEN_SHIFT;
3326         mss_l4len_idx |= skb_shinfo(skb)->gso_size << IXGBE_ADVTXD_MSS_SHIFT;
3327         mss_l4len_idx |= 1 << IXGBE_ADVTXD_IDX_SHIFT;
3328
3329         /* vlan_macip_lens: HEADLEN, MACLEN, VLAN tag */
3330         vlan_macip_lens = skb_network_header_len(skb);
3331         vlan_macip_lens |= skb_network_offset(skb) << IXGBE_ADVTXD_MACLEN_SHIFT;
3332         vlan_macip_lens |= first->tx_flags & IXGBE_TX_FLAGS_VLAN_MASK;
3333
3334         ixgbevf_tx_ctxtdesc(tx_ring, vlan_macip_lens,
3335                             type_tucmd, mss_l4len_idx);
3336
3337         return 1;
3338 }
3339
3340 static void ixgbevf_tx_csum(struct ixgbevf_ring *tx_ring,
3341                             struct ixgbevf_tx_buffer *first)
3342 {
3343         struct sk_buff *skb = first->skb;
3344         u32 vlan_macip_lens = 0;
3345         u32 mss_l4len_idx = 0;
3346         u32 type_tucmd = 0;
3347
3348         if (skb->ip_summed == CHECKSUM_PARTIAL) {
3349                 u8 l4_hdr = 0;
3350                 __be16 frag_off;
3351
3352                 switch (first->protocol) {
3353                 case htons(ETH_P_IP):
3354                         vlan_macip_lens |= skb_network_header_len(skb);
3355                         type_tucmd |= IXGBE_ADVTXD_TUCMD_IPV4;
3356                         l4_hdr = ip_hdr(skb)->protocol;
3357                         break;
3358                 case htons(ETH_P_IPV6):
3359                         vlan_macip_lens |= skb_network_header_len(skb);
3360                         l4_hdr = ipv6_hdr(skb)->nexthdr;
3361                         if (likely(skb_network_header_len(skb) ==
3362                                    sizeof(struct ipv6hdr)))
3363                                 break;
3364                         ipv6_skip_exthdr(skb, skb_network_offset(skb) +
3365                                               sizeof(struct ipv6hdr),
3366                                          &l4_hdr, &frag_off);
3367                         if (unlikely(frag_off))
3368                                 l4_hdr = NEXTHDR_FRAGMENT;
3369                         break;
3370                 default:
3371                         break;
3372                 }
3373
3374                 switch (l4_hdr) {
3375                 case IPPROTO_TCP:
3376                         type_tucmd |= IXGBE_ADVTXD_TUCMD_L4T_TCP;
3377                         mss_l4len_idx = tcp_hdrlen(skb) <<
3378                                         IXGBE_ADVTXD_L4LEN_SHIFT;
3379                         break;
3380                 case IPPROTO_SCTP:
3381                         type_tucmd |= IXGBE_ADVTXD_TUCMD_L4T_SCTP;
3382                         mss_l4len_idx = sizeof(struct sctphdr) <<
3383                                         IXGBE_ADVTXD_L4LEN_SHIFT;
3384                         break;
3385                 case IPPROTO_UDP:
3386                         mss_l4len_idx = sizeof(struct udphdr) <<
3387                                         IXGBE_ADVTXD_L4LEN_SHIFT;
3388                         break;
3389                 default:
3390                         if (unlikely(net_ratelimit())) {
3391                                 dev_warn(tx_ring->dev,
3392                                          "partial checksum, l3 proto=%x, l4 proto=%x\n",
3393                                          first->protocol, l4_hdr);
3394                         }
3395                         skb_checksum_help(skb);
3396                         goto no_csum;
3397                 }
3398
3399                 /* update TX checksum flag */
3400                 first->tx_flags |= IXGBE_TX_FLAGS_CSUM;
3401         }
3402
3403 no_csum:
3404         /* vlan_macip_lens: MACLEN, VLAN tag */
3405         vlan_macip_lens |= skb_network_offset(skb) << IXGBE_ADVTXD_MACLEN_SHIFT;
3406         vlan_macip_lens |= first->tx_flags & IXGBE_TX_FLAGS_VLAN_MASK;
3407
3408         ixgbevf_tx_ctxtdesc(tx_ring, vlan_macip_lens,
3409                             type_tucmd, mss_l4len_idx);
3410 }
3411
3412 static __le32 ixgbevf_tx_cmd_type(u32 tx_flags)
3413 {
3414         /* set type for advanced descriptor with frame checksum insertion */
3415         __le32 cmd_type = cpu_to_le32(IXGBE_ADVTXD_DTYP_DATA |
3416                                       IXGBE_ADVTXD_DCMD_IFCS |
3417                                       IXGBE_ADVTXD_DCMD_DEXT);
3418
3419         /* set HW VLAN bit if VLAN is present */
3420         if (tx_flags & IXGBE_TX_FLAGS_VLAN)
3421                 cmd_type |= cpu_to_le32(IXGBE_ADVTXD_DCMD_VLE);
3422
3423         /* set segmentation enable bits for TSO/FSO */
3424         if (tx_flags & IXGBE_TX_FLAGS_TSO)
3425                 cmd_type |= cpu_to_le32(IXGBE_ADVTXD_DCMD_TSE);
3426
3427         return cmd_type;
3428 }
3429
3430 static void ixgbevf_tx_olinfo_status(union ixgbe_adv_tx_desc *tx_desc,
3431                                      u32 tx_flags, unsigned int paylen)
3432 {
3433         __le32 olinfo_status = cpu_to_le32(paylen << IXGBE_ADVTXD_PAYLEN_SHIFT);
3434
3435         /* enable L4 checksum for TSO and TX checksum offload */
3436         if (tx_flags & IXGBE_TX_FLAGS_CSUM)
3437                 olinfo_status |= cpu_to_le32(IXGBE_ADVTXD_POPTS_TXSM);
3438
3439         /* enble IPv4 checksum for TSO */
3440         if (tx_flags & IXGBE_TX_FLAGS_IPV4)
3441                 olinfo_status |= cpu_to_le32(IXGBE_ADVTXD_POPTS_IXSM);
3442
3443         /* use index 1 context for TSO/FSO/FCOE */
3444         if (tx_flags & IXGBE_TX_FLAGS_TSO)
3445                 olinfo_status |= cpu_to_le32(1 << IXGBE_ADVTXD_IDX_SHIFT);
3446
3447         /* Check Context must be set if Tx switch is enabled, which it
3448          * always is for case where virtual functions are running
3449          */
3450         olinfo_status |= cpu_to_le32(IXGBE_ADVTXD_CC);
3451
3452         tx_desc->read.olinfo_status = olinfo_status;
3453 }
3454
3455 static void ixgbevf_tx_map(struct ixgbevf_ring *tx_ring,
3456                            struct ixgbevf_tx_buffer *first,
3457                            const u8 hdr_len)
3458 {
3459         dma_addr_t dma;
3460         struct sk_buff *skb = first->skb;
3461         struct ixgbevf_tx_buffer *tx_buffer;
3462         union ixgbe_adv_tx_desc *tx_desc;
3463         struct skb_frag_struct *frag = &skb_shinfo(skb)->frags[0];
3464         unsigned int data_len = skb->data_len;
3465         unsigned int size = skb_headlen(skb);
3466         unsigned int paylen = skb->len - hdr_len;
3467         u32 tx_flags = first->tx_flags;
3468         __le32 cmd_type;
3469         u16 i = tx_ring->next_to_use;
3470
3471         tx_desc = IXGBEVF_TX_DESC(tx_ring, i);
3472
3473         ixgbevf_tx_olinfo_status(tx_desc, tx_flags, paylen);
3474         cmd_type = ixgbevf_tx_cmd_type(tx_flags);
3475
3476         dma = dma_map_single(tx_ring->dev, skb->data, size, DMA_TO_DEVICE);
3477         if (dma_mapping_error(tx_ring->dev, dma))
3478                 goto dma_error;
3479
3480         /* record length, and DMA address */
3481         dma_unmap_len_set(first, len, size);
3482         dma_unmap_addr_set(first, dma, dma);
3483
3484         tx_desc->read.buffer_addr = cpu_to_le64(dma);
3485
3486         for (;;) {
3487                 while (unlikely(size > IXGBE_MAX_DATA_PER_TXD)) {
3488                         tx_desc->read.cmd_type_len =
3489                                 cmd_type | cpu_to_le32(IXGBE_MAX_DATA_PER_TXD);
3490
3491                         i++;
3492                         tx_desc++;
3493                         if (i == tx_ring->count) {
3494                                 tx_desc = IXGBEVF_TX_DESC(tx_ring, 0);
3495                                 i = 0;
3496                         }
3497
3498                         dma += IXGBE_MAX_DATA_PER_TXD;
3499                         size -= IXGBE_MAX_DATA_PER_TXD;
3500
3501                         tx_desc->read.buffer_addr = cpu_to_le64(dma);
3502                         tx_desc->read.olinfo_status = 0;
3503                 }
3504
3505                 if (likely(!data_len))
3506                         break;
3507
3508                 tx_desc->read.cmd_type_len = cmd_type | cpu_to_le32(size);
3509
3510                 i++;
3511                 tx_desc++;
3512                 if (i == tx_ring->count) {
3513                         tx_desc = IXGBEVF_TX_DESC(tx_ring, 0);
3514                         i = 0;
3515                 }
3516
3517                 size = skb_frag_size(frag);
3518                 data_len -= size;
3519
3520                 dma = skb_frag_dma_map(tx_ring->dev, frag, 0, size,
3521                                        DMA_TO_DEVICE);
3522                 if (dma_mapping_error(tx_ring->dev, dma))
3523                         goto dma_error;
3524
3525                 tx_buffer = &tx_ring->tx_buffer_info[i];
3526                 dma_unmap_len_set(tx_buffer, len, size);
3527                 dma_unmap_addr_set(tx_buffer, dma, dma);
3528
3529                 tx_desc->read.buffer_addr = cpu_to_le64(dma);
3530                 tx_desc->read.olinfo_status = 0;
3531
3532                 frag++;
3533         }
3534
3535         /* write last descriptor with RS and EOP bits */
3536         cmd_type |= cpu_to_le32(size) | cpu_to_le32(IXGBE_TXD_CMD);
3537         tx_desc->read.cmd_type_len = cmd_type;
3538
3539         /* set the timestamp */
3540         first->time_stamp = jiffies;
3541
3542         /* Force memory writes to complete before letting h/w know there
3543          * are new descriptors to fetch.  (Only applicable for weak-ordered
3544          * memory model archs, such as IA-64).
3545          *
3546          * We also need this memory barrier (wmb) to make certain all of the
3547          * status bits have been updated before next_to_watch is written.
3548          */
3549         wmb();
3550
3551         /* set next_to_watch value indicating a packet is present */
3552         first->next_to_watch = tx_desc;
3553
3554         i++;
3555         if (i == tx_ring->count)
3556                 i = 0;
3557
3558         tx_ring->next_to_use = i;
3559
3560         /* notify HW of packet */
3561         ixgbevf_write_tail(tx_ring, i);
3562
3563         return;
3564 dma_error:
3565         dev_err(tx_ring->dev, "TX DMA map failed\n");
3566
3567         /* clear dma mappings for failed tx_buffer_info map */
3568         for (;;) {
3569                 tx_buffer = &tx_ring->tx_buffer_info[i];
3570                 ixgbevf_unmap_and_free_tx_resource(tx_ring, tx_buffer);
3571                 if (tx_buffer == first)
3572                         break;
3573                 if (i == 0)
3574                         i = tx_ring->count;
3575                 i--;
3576         }
3577
3578         tx_ring->next_to_use = i;
3579 }
3580
3581 static int __ixgbevf_maybe_stop_tx(struct ixgbevf_ring *tx_ring, int size)
3582 {
3583         netif_stop_subqueue(tx_ring->netdev, tx_ring->queue_index);
3584         /* Herbert's original patch had:
3585          *  smp_mb__after_netif_stop_queue();
3586          * but since that doesn't exist yet, just open code it.
3587          */
3588         smp_mb();
3589
3590         /* We need to check again in a case another CPU has just
3591          * made room available.
3592          */
3593         if (likely(ixgbevf_desc_unused(tx_ring) < size))
3594                 return -EBUSY;
3595
3596         /* A reprieve! - use start_queue because it doesn't call schedule */
3597         netif_start_subqueue(tx_ring->netdev, tx_ring->queue_index);
3598         ++tx_ring->tx_stats.restart_queue;
3599
3600         return 0;
3601 }
3602
3603 static int ixgbevf_maybe_stop_tx(struct ixgbevf_ring *tx_ring, int size)
3604 {
3605         if (likely(ixgbevf_desc_unused(tx_ring) >= size))
3606                 return 0;
3607         return __ixgbevf_maybe_stop_tx(tx_ring, size);
3608 }
3609
3610 static int ixgbevf_xmit_frame(struct sk_buff *skb, struct net_device *netdev)
3611 {
3612         struct ixgbevf_adapter *adapter = netdev_priv(netdev);
3613         struct ixgbevf_tx_buffer *first;
3614         struct ixgbevf_ring *tx_ring;
3615         int tso;
3616         u32 tx_flags = 0;
3617         u16 count = TXD_USE_COUNT(skb_headlen(skb));
3618 #if PAGE_SIZE > IXGBE_MAX_DATA_PER_TXD
3619         unsigned short f;
3620 #endif
3621         u8 hdr_len = 0;
3622         u8 *dst_mac = skb_header_pointer(skb, 0, 0, NULL);
3623
3624         if (!dst_mac || is_link_local_ether_addr(dst_mac)) {
3625                 dev_kfree_skb_any(skb);
3626                 return NETDEV_TX_OK;
3627         }
3628
3629         tx_ring = adapter->tx_ring[skb->queue_mapping];
3630
3631         /* need: 1 descriptor per page * PAGE_SIZE/IXGBE_MAX_DATA_PER_TXD,
3632          *       + 1 desc for skb_headlen/IXGBE_MAX_DATA_PER_TXD,
3633          *       + 2 desc gap to keep tail from touching head,
3634          *       + 1 desc for context descriptor,
3635          * otherwise try next time
3636          */
3637 #if PAGE_SIZE > IXGBE_MAX_DATA_PER_TXD
3638         for (f = 0; f < skb_shinfo(skb)->nr_frags; f++)
3639                 count += TXD_USE_COUNT(skb_shinfo(skb)->frags[f].size);
3640 #else
3641         count += skb_shinfo(skb)->nr_frags;
3642 #endif
3643         if (ixgbevf_maybe_stop_tx(tx_ring, count + 3)) {
3644                 tx_ring->tx_stats.tx_busy++;
3645                 return NETDEV_TX_BUSY;
3646         }
3647
3648         /* record the location of the first descriptor for this packet */
3649         first = &tx_ring->tx_buffer_info[tx_ring->next_to_use];
3650         first->skb = skb;
3651         first->bytecount = skb->len;
3652         first->gso_segs = 1;
3653
3654         if (skb_vlan_tag_present(skb)) {
3655                 tx_flags |= skb_vlan_tag_get(skb);
3656                 tx_flags <<= IXGBE_TX_FLAGS_VLAN_SHIFT;
3657                 tx_flags |= IXGBE_TX_FLAGS_VLAN;
3658         }
3659
3660         /* record initial flags and protocol */
3661         first->tx_flags = tx_flags;
3662         first->protocol = vlan_get_protocol(skb);
3663
3664         tso = ixgbevf_tso(tx_ring, first, &hdr_len);
3665         if (tso < 0)
3666                 goto out_drop;
3667         else if (!tso)
3668                 ixgbevf_tx_csum(tx_ring, first);
3669
3670         ixgbevf_tx_map(tx_ring, first, hdr_len);
3671
3672         ixgbevf_maybe_stop_tx(tx_ring, DESC_NEEDED);
3673
3674         return NETDEV_TX_OK;
3675
3676 out_drop:
3677         dev_kfree_skb_any(first->skb);
3678         first->skb = NULL;
3679
3680         return NETDEV_TX_OK;
3681 }
3682
3683 /**
3684  * ixgbevf_set_mac - Change the Ethernet Address of the NIC
3685  * @netdev: network interface device structure
3686  * @p: pointer to an address structure
3687  *
3688  * Returns 0 on success, negative on failure
3689  **/
3690 static int ixgbevf_set_mac(struct net_device *netdev, void *p)
3691 {
3692         struct ixgbevf_adapter *adapter = netdev_priv(netdev);
3693         struct ixgbe_hw *hw = &adapter->hw;
3694         struct sockaddr *addr = p;
3695
3696         if (!is_valid_ether_addr(addr->sa_data))
3697                 return -EADDRNOTAVAIL;
3698
3699         ether_addr_copy(netdev->dev_addr, addr->sa_data);
3700         ether_addr_copy(hw->mac.addr, addr->sa_data);
3701
3702         spin_lock_bh(&adapter->mbx_lock);
3703
3704         hw->mac.ops.set_rar(hw, 0, hw->mac.addr, 0);
3705
3706         spin_unlock_bh(&adapter->mbx_lock);
3707
3708         return 0;
3709 }
3710
3711 /**
3712  * ixgbevf_change_mtu - Change the Maximum Transfer Unit
3713  * @netdev: network interface device structure
3714  * @new_mtu: new value for maximum frame size
3715  *
3716  * Returns 0 on success, negative on failure
3717  **/
3718 static int ixgbevf_change_mtu(struct net_device *netdev, int new_mtu)
3719 {
3720         struct ixgbevf_adapter *adapter = netdev_priv(netdev);
3721         struct ixgbe_hw *hw = &adapter->hw;
3722         int max_frame = new_mtu + ETH_HLEN + ETH_FCS_LEN;
3723         int max_possible_frame = MAXIMUM_ETHERNET_VLAN_SIZE;
3724
3725         switch (adapter->hw.api_version) {
3726         case ixgbe_mbox_api_11:
3727         case ixgbe_mbox_api_12:
3728                 max_possible_frame = IXGBE_MAX_JUMBO_FRAME_SIZE;
3729                 break;
3730         default:
3731                 if (adapter->hw.mac.type != ixgbe_mac_82599_vf)
3732                         max_possible_frame = IXGBE_MAX_JUMBO_FRAME_SIZE;
3733                 break;
3734         }
3735
3736         /* MTU < 68 is an error and causes problems on some kernels */
3737         if ((new_mtu < 68) || (max_frame > max_possible_frame))
3738                 return -EINVAL;
3739
3740         hw_dbg(hw, "changing MTU from %d to %d\n",
3741                netdev->mtu, new_mtu);
3742         /* must set new MTU before calling down or up */
3743         netdev->mtu = new_mtu;
3744
3745         /* notify the PF of our intent to use this size of frame */
3746         ixgbevf_rlpml_set_vf(hw, max_frame);
3747
3748         return 0;
3749 }
3750
3751 #ifdef CONFIG_NET_POLL_CONTROLLER
3752 /* Polling 'interrupt' - used by things like netconsole to send skbs
3753  * without having to re-enable interrupts. It's not called while
3754  * the interrupt routine is executing.
3755  */
3756 static void ixgbevf_netpoll(struct net_device *netdev)
3757 {
3758         struct ixgbevf_adapter *adapter = netdev_priv(netdev);
3759         int i;
3760
3761         /* if interface is down do nothing */
3762         if (test_bit(__IXGBEVF_DOWN, &adapter->state))
3763                 return;
3764         for (i = 0; i < adapter->num_rx_queues; i++)
3765                 ixgbevf_msix_clean_rings(0, adapter->q_vector[i]);
3766 }
3767 #endif /* CONFIG_NET_POLL_CONTROLLER */
3768
3769 static int ixgbevf_suspend(struct pci_dev *pdev, pm_message_t state)
3770 {
3771         struct net_device *netdev = pci_get_drvdata(pdev);
3772         struct ixgbevf_adapter *adapter = netdev_priv(netdev);
3773 #ifdef CONFIG_PM
3774         int retval = 0;
3775 #endif
3776
3777         netif_device_detach(netdev);
3778
3779         if (netif_running(netdev)) {
3780                 rtnl_lock();
3781                 ixgbevf_down(adapter);
3782                 ixgbevf_free_irq(adapter);
3783                 ixgbevf_free_all_tx_resources(adapter);
3784                 ixgbevf_free_all_rx_resources(adapter);
3785                 rtnl_unlock();
3786         }
3787
3788         ixgbevf_clear_interrupt_scheme(adapter);
3789
3790 #ifdef CONFIG_PM
3791         retval = pci_save_state(pdev);
3792         if (retval)
3793                 return retval;
3794
3795 #endif
3796         if (!test_and_set_bit(__IXGBEVF_DISABLED, &adapter->state))
3797                 pci_disable_device(pdev);
3798
3799         return 0;
3800 }
3801
3802 #ifdef CONFIG_PM
3803 static int ixgbevf_resume(struct pci_dev *pdev)
3804 {
3805         struct net_device *netdev = pci_get_drvdata(pdev);
3806         struct ixgbevf_adapter *adapter = netdev_priv(netdev);
3807         u32 err;
3808
3809         pci_restore_state(pdev);
3810         /* pci_restore_state clears dev->state_saved so call
3811          * pci_save_state to restore it.
3812          */
3813         pci_save_state(pdev);
3814
3815         err = pci_enable_device_mem(pdev);
3816         if (err) {
3817                 dev_err(&pdev->dev, "Cannot enable PCI device from suspend\n");
3818                 return err;
3819         }
3820         smp_mb__before_atomic();
3821         clear_bit(__IXGBEVF_DISABLED, &adapter->state);
3822         pci_set_master(pdev);
3823
3824         ixgbevf_reset(adapter);
3825
3826         rtnl_lock();
3827         err = ixgbevf_init_interrupt_scheme(adapter);
3828         rtnl_unlock();
3829         if (err) {
3830                 dev_err(&pdev->dev, "Cannot initialize interrupts\n");
3831                 return err;
3832         }
3833
3834         if (netif_running(netdev)) {
3835                 err = ixgbevf_open(netdev);
3836                 if (err)
3837                         return err;
3838         }
3839
3840         netif_device_attach(netdev);
3841
3842         return err;
3843 }
3844
3845 #endif /* CONFIG_PM */
3846 static void ixgbevf_shutdown(struct pci_dev *pdev)
3847 {
3848         ixgbevf_suspend(pdev, PMSG_SUSPEND);
3849 }
3850
3851 static struct rtnl_link_stats64 *ixgbevf_get_stats(struct net_device *netdev,
3852                                                 struct rtnl_link_stats64 *stats)
3853 {
3854         struct ixgbevf_adapter *adapter = netdev_priv(netdev);
3855         unsigned int start;
3856         u64 bytes, packets;
3857         const struct ixgbevf_ring *ring;
3858         int i;
3859
3860         ixgbevf_update_stats(adapter);
3861
3862         stats->multicast = adapter->stats.vfmprc - adapter->stats.base_vfmprc;
3863
3864         for (i = 0; i < adapter->num_rx_queues; i++) {
3865                 ring = adapter->rx_ring[i];
3866                 do {
3867                         start = u64_stats_fetch_begin_irq(&ring->syncp);
3868                         bytes = ring->stats.bytes;
3869                         packets = ring->stats.packets;
3870                 } while (u64_stats_fetch_retry_irq(&ring->syncp, start));
3871                 stats->rx_bytes += bytes;
3872                 stats->rx_packets += packets;
3873         }
3874
3875         for (i = 0; i < adapter->num_tx_queues; i++) {
3876                 ring = adapter->tx_ring[i];
3877                 do {
3878                         start = u64_stats_fetch_begin_irq(&ring->syncp);
3879                         bytes = ring->stats.bytes;
3880                         packets = ring->stats.packets;
3881                 } while (u64_stats_fetch_retry_irq(&ring->syncp, start));
3882                 stats->tx_bytes += bytes;
3883                 stats->tx_packets += packets;
3884         }
3885
3886         return stats;
3887 }
3888
3889 static const struct net_device_ops ixgbevf_netdev_ops = {
3890         .ndo_open               = ixgbevf_open,
3891         .ndo_stop               = ixgbevf_close,
3892         .ndo_start_xmit         = ixgbevf_xmit_frame,
3893         .ndo_set_rx_mode        = ixgbevf_set_rx_mode,
3894         .ndo_get_stats64        = ixgbevf_get_stats,
3895         .ndo_validate_addr      = eth_validate_addr,
3896         .ndo_set_mac_address    = ixgbevf_set_mac,
3897         .ndo_change_mtu         = ixgbevf_change_mtu,
3898         .ndo_tx_timeout         = ixgbevf_tx_timeout,
3899         .ndo_vlan_rx_add_vid    = ixgbevf_vlan_rx_add_vid,
3900         .ndo_vlan_rx_kill_vid   = ixgbevf_vlan_rx_kill_vid,
3901 #ifdef CONFIG_NET_RX_BUSY_POLL
3902         .ndo_busy_poll          = ixgbevf_busy_poll_recv,
3903 #endif
3904 #ifdef CONFIG_NET_POLL_CONTROLLER
3905         .ndo_poll_controller    = ixgbevf_netpoll,
3906 #endif
3907         .ndo_features_check     = passthru_features_check,
3908 };
3909
3910 static void ixgbevf_assign_netdev_ops(struct net_device *dev)
3911 {
3912         dev->netdev_ops = &ixgbevf_netdev_ops;
3913         ixgbevf_set_ethtool_ops(dev);
3914         dev->watchdog_timeo = 5 * HZ;
3915 }
3916
3917 /**
3918  * ixgbevf_probe - Device Initialization Routine
3919  * @pdev: PCI device information struct
3920  * @ent: entry in ixgbevf_pci_tbl
3921  *
3922  * Returns 0 on success, negative on failure
3923  *
3924  * ixgbevf_probe initializes an adapter identified by a pci_dev structure.
3925  * The OS initialization, configuring of the adapter private structure,
3926  * and a hardware reset occur.
3927  **/
3928 static int ixgbevf_probe(struct pci_dev *pdev, const struct pci_device_id *ent)
3929 {
3930         struct net_device *netdev;
3931         struct ixgbevf_adapter *adapter = NULL;
3932         struct ixgbe_hw *hw = NULL;
3933         const struct ixgbevf_info *ii = ixgbevf_info_tbl[ent->driver_data];
3934         int err, pci_using_dac;
3935         bool disable_dev = false;
3936
3937         err = pci_enable_device(pdev);
3938         if (err)
3939                 return err;
3940
3941         if (!dma_set_mask_and_coherent(&pdev->dev, DMA_BIT_MASK(64))) {
3942                 pci_using_dac = 1;
3943         } else {
3944                 err = dma_set_mask_and_coherent(&pdev->dev, DMA_BIT_MASK(32));
3945                 if (err) {
3946                         dev_err(&pdev->dev, "No usable DMA configuration, aborting\n");
3947                         goto err_dma;
3948                 }
3949                 pci_using_dac = 0;
3950         }
3951
3952         err = pci_request_regions(pdev, ixgbevf_driver_name);
3953         if (err) {
3954                 dev_err(&pdev->dev, "pci_request_regions failed 0x%x\n", err);
3955                 goto err_pci_reg;
3956         }
3957
3958         pci_set_master(pdev);
3959
3960         netdev = alloc_etherdev_mq(sizeof(struct ixgbevf_adapter),
3961                                    MAX_TX_QUEUES);
3962         if (!netdev) {
3963                 err = -ENOMEM;
3964                 goto err_alloc_etherdev;
3965         }
3966
3967         SET_NETDEV_DEV(netdev, &pdev->dev);
3968
3969         adapter = netdev_priv(netdev);
3970
3971         adapter->netdev = netdev;
3972         adapter->pdev = pdev;
3973         hw = &adapter->hw;
3974         hw->back = adapter;
3975         adapter->msg_enable = netif_msg_init(debug, DEFAULT_MSG_ENABLE);
3976
3977         /* call save state here in standalone driver because it relies on
3978          * adapter struct to exist, and needs to call netdev_priv
3979          */
3980         pci_save_state(pdev);
3981
3982         hw->hw_addr = ioremap(pci_resource_start(pdev, 0),
3983                               pci_resource_len(pdev, 0));
3984         adapter->io_addr = hw->hw_addr;
3985         if (!hw->hw_addr) {
3986                 err = -EIO;
3987                 goto err_ioremap;
3988         }
3989
3990         ixgbevf_assign_netdev_ops(netdev);
3991
3992         /* Setup HW API */
3993         memcpy(&hw->mac.ops, ii->mac_ops, sizeof(hw->mac.ops));
3994         hw->mac.type  = ii->mac;
3995
3996         memcpy(&hw->mbx.ops, &ixgbevf_mbx_ops,
3997                sizeof(struct ixgbe_mbx_operations));
3998
3999         /* setup the private structure */
4000         err = ixgbevf_sw_init(adapter);
4001         if (err)
4002                 goto err_sw_init;
4003
4004         /* The HW MAC address was set and/or determined in sw_init */
4005         if (!is_valid_ether_addr(netdev->dev_addr)) {
4006                 pr_err("invalid MAC address\n");
4007                 err = -EIO;
4008                 goto err_sw_init;
4009         }
4010
4011         netdev->hw_features = NETIF_F_SG |
4012                               NETIF_F_IP_CSUM |
4013                               NETIF_F_IPV6_CSUM |
4014                               NETIF_F_TSO |
4015                               NETIF_F_TSO6 |
4016                               NETIF_F_RXCSUM;
4017
4018         netdev->features = netdev->hw_features |
4019                            NETIF_F_HW_VLAN_CTAG_TX |
4020                            NETIF_F_HW_VLAN_CTAG_RX |
4021                            NETIF_F_HW_VLAN_CTAG_FILTER;
4022
4023         netdev->vlan_features |= NETIF_F_TSO |
4024                                  NETIF_F_TSO6 |
4025                                  NETIF_F_IP_CSUM |
4026                                  NETIF_F_IPV6_CSUM |
4027                                  NETIF_F_SG;
4028
4029         if (pci_using_dac)
4030                 netdev->features |= NETIF_F_HIGHDMA;
4031
4032         netdev->priv_flags |= IFF_UNICAST_FLT;
4033
4034         if (IXGBE_REMOVED(hw->hw_addr)) {
4035                 err = -EIO;
4036                 goto err_sw_init;
4037         }
4038
4039         setup_timer(&adapter->service_timer, &ixgbevf_service_timer,
4040                     (unsigned long)adapter);
4041
4042         INIT_WORK(&adapter->service_task, ixgbevf_service_task);
4043         set_bit(__IXGBEVF_SERVICE_INITED, &adapter->state);
4044         clear_bit(__IXGBEVF_SERVICE_SCHED, &adapter->state);
4045
4046         err = ixgbevf_init_interrupt_scheme(adapter);
4047         if (err)
4048                 goto err_sw_init;
4049
4050         strcpy(netdev->name, "eth%d");
4051
4052         err = register_netdev(netdev);
4053         if (err)
4054                 goto err_register;
4055
4056         pci_set_drvdata(pdev, netdev);
4057         netif_carrier_off(netdev);
4058
4059         ixgbevf_init_last_counter_stats(adapter);
4060
4061         /* print the VF info */
4062         dev_info(&pdev->dev, "%pM\n", netdev->dev_addr);
4063         dev_info(&pdev->dev, "MAC: %d\n", hw->mac.type);
4064
4065         switch (hw->mac.type) {
4066         case ixgbe_mac_X550_vf:
4067                 dev_info(&pdev->dev, "Intel(R) X550 Virtual Function\n");
4068                 break;
4069         case ixgbe_mac_X540_vf:
4070                 dev_info(&pdev->dev, "Intel(R) X540 Virtual Function\n");
4071                 break;
4072         case ixgbe_mac_82599_vf:
4073         default:
4074                 dev_info(&pdev->dev, "Intel(R) 82599 Virtual Function\n");
4075                 break;
4076         }
4077
4078         return 0;
4079
4080 err_register:
4081         ixgbevf_clear_interrupt_scheme(adapter);
4082 err_sw_init:
4083         ixgbevf_reset_interrupt_capability(adapter);
4084         iounmap(adapter->io_addr);
4085 err_ioremap:
4086         disable_dev = !test_and_set_bit(__IXGBEVF_DISABLED, &adapter->state);
4087         free_netdev(netdev);
4088 err_alloc_etherdev:
4089         pci_release_regions(pdev);
4090 err_pci_reg:
4091 err_dma:
4092         if (!adapter || disable_dev)
4093                 pci_disable_device(pdev);
4094         return err;
4095 }
4096
4097 /**
4098  * ixgbevf_remove - Device Removal Routine
4099  * @pdev: PCI device information struct
4100  *
4101  * ixgbevf_remove is called by the PCI subsystem to alert the driver
4102  * that it should release a PCI device.  The could be caused by a
4103  * Hot-Plug event, or because the driver is going to be removed from
4104  * memory.
4105  **/
4106 static void ixgbevf_remove(struct pci_dev *pdev)
4107 {
4108         struct net_device *netdev = pci_get_drvdata(pdev);
4109         struct ixgbevf_adapter *adapter;
4110         bool disable_dev;
4111
4112         if (!netdev)
4113                 return;
4114
4115         adapter = netdev_priv(netdev);
4116
4117         set_bit(__IXGBEVF_REMOVING, &adapter->state);
4118         cancel_work_sync(&adapter->service_task);
4119
4120         if (netdev->reg_state == NETREG_REGISTERED)
4121                 unregister_netdev(netdev);
4122
4123         ixgbevf_clear_interrupt_scheme(adapter);
4124         ixgbevf_reset_interrupt_capability(adapter);
4125
4126         iounmap(adapter->io_addr);
4127         pci_release_regions(pdev);
4128
4129         hw_dbg(&adapter->hw, "Remove complete\n");
4130
4131         disable_dev = !test_and_set_bit(__IXGBEVF_DISABLED, &adapter->state);
4132         free_netdev(netdev);
4133
4134         if (disable_dev)
4135                 pci_disable_device(pdev);
4136 }
4137
4138 /**
4139  * ixgbevf_io_error_detected - called when PCI error is detected
4140  * @pdev: Pointer to PCI device
4141  * @state: The current pci connection state
4142  *
4143  * This function is called after a PCI bus error affecting
4144  * this device has been detected.
4145  **/
4146 static pci_ers_result_t ixgbevf_io_error_detected(struct pci_dev *pdev,
4147                                                   pci_channel_state_t state)
4148 {
4149         struct net_device *netdev = pci_get_drvdata(pdev);
4150         struct ixgbevf_adapter *adapter = netdev_priv(netdev);
4151
4152         if (!test_bit(__IXGBEVF_SERVICE_INITED, &adapter->state))
4153                 return PCI_ERS_RESULT_DISCONNECT;
4154
4155         rtnl_lock();
4156         netif_device_detach(netdev);
4157
4158         if (state == pci_channel_io_perm_failure) {
4159                 rtnl_unlock();
4160                 return PCI_ERS_RESULT_DISCONNECT;
4161         }
4162
4163         if (netif_running(netdev))
4164                 ixgbevf_down(adapter);
4165
4166         if (!test_and_set_bit(__IXGBEVF_DISABLED, &adapter->state))
4167                 pci_disable_device(pdev);
4168         rtnl_unlock();
4169
4170         /* Request a slot slot reset. */
4171         return PCI_ERS_RESULT_NEED_RESET;
4172 }
4173
4174 /**
4175  * ixgbevf_io_slot_reset - called after the pci bus has been reset.
4176  * @pdev: Pointer to PCI device
4177  *
4178  * Restart the card from scratch, as if from a cold-boot. Implementation
4179  * resembles the first-half of the ixgbevf_resume routine.
4180  **/
4181 static pci_ers_result_t ixgbevf_io_slot_reset(struct pci_dev *pdev)
4182 {
4183         struct net_device *netdev = pci_get_drvdata(pdev);
4184         struct ixgbevf_adapter *adapter = netdev_priv(netdev);
4185
4186         if (pci_enable_device_mem(pdev)) {
4187                 dev_err(&pdev->dev,
4188                         "Cannot re-enable PCI device after reset.\n");
4189                 return PCI_ERS_RESULT_DISCONNECT;
4190         }
4191
4192         smp_mb__before_atomic();
4193         clear_bit(__IXGBEVF_DISABLED, &adapter->state);
4194         pci_set_master(pdev);
4195
4196         ixgbevf_reset(adapter);
4197
4198         return PCI_ERS_RESULT_RECOVERED;
4199 }
4200
4201 /**
4202  * ixgbevf_io_resume - called when traffic can start flowing again.
4203  * @pdev: Pointer to PCI device
4204  *
4205  * This callback is called when the error recovery driver tells us that
4206  * its OK to resume normal operation. Implementation resembles the
4207  * second-half of the ixgbevf_resume routine.
4208  **/
4209 static void ixgbevf_io_resume(struct pci_dev *pdev)
4210 {
4211         struct net_device *netdev = pci_get_drvdata(pdev);
4212         struct ixgbevf_adapter *adapter = netdev_priv(netdev);
4213
4214         if (netif_running(netdev))
4215                 ixgbevf_up(adapter);
4216
4217         netif_device_attach(netdev);
4218 }
4219
4220 /* PCI Error Recovery (ERS) */
4221 static const struct pci_error_handlers ixgbevf_err_handler = {
4222         .error_detected = ixgbevf_io_error_detected,
4223         .slot_reset = ixgbevf_io_slot_reset,
4224         .resume = ixgbevf_io_resume,
4225 };
4226
4227 static struct pci_driver ixgbevf_driver = {
4228         .name           = ixgbevf_driver_name,
4229         .id_table       = ixgbevf_pci_tbl,
4230         .probe          = ixgbevf_probe,
4231         .remove         = ixgbevf_remove,
4232 #ifdef CONFIG_PM
4233         /* Power Management Hooks */
4234         .suspend        = ixgbevf_suspend,
4235         .resume         = ixgbevf_resume,
4236 #endif
4237         .shutdown       = ixgbevf_shutdown,
4238         .err_handler    = &ixgbevf_err_handler
4239 };
4240
4241 /**
4242  * ixgbevf_init_module - Driver Registration Routine
4243  *
4244  * ixgbevf_init_module is the first routine called when the driver is
4245  * loaded. All it does is register with the PCI subsystem.
4246  **/
4247 static int __init ixgbevf_init_module(void)
4248 {
4249         pr_info("%s - version %s\n", ixgbevf_driver_string,
4250                 ixgbevf_driver_version);
4251
4252         pr_info("%s\n", ixgbevf_copyright);
4253         ixgbevf_wq = create_singlethread_workqueue(ixgbevf_driver_name);
4254         if (!ixgbevf_wq) {
4255                 pr_err("%s: Failed to create workqueue\n", ixgbevf_driver_name);
4256                 return -ENOMEM;
4257         }
4258
4259         return pci_register_driver(&ixgbevf_driver);
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         if (ixgbevf_wq) {
4274                 destroy_workqueue(ixgbevf_wq);
4275                 ixgbevf_wq = NULL;
4276         }
4277 }
4278
4279 #ifdef DEBUG
4280 /**
4281  * ixgbevf_get_hw_dev_name - return device name string
4282  * used by hardware layer to print debugging information
4283  **/
4284 char *ixgbevf_get_hw_dev_name(struct ixgbe_hw *hw)
4285 {
4286         struct ixgbevf_adapter *adapter = hw->back;
4287
4288         return adapter->netdev->name;
4289 }
4290
4291 #endif
4292 module_exit(ixgbevf_exit_module);
4293
4294 /* ixgbevf_main.c */