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

   1 // SPDX-License-Identifier: GPL-2.0
   2 /* Copyright(c) 1999 - 2018 Intel Corporation. */
   3
   4 /******************************************************************************
   5  Copyright (c)2006 - 2007 Myricom, Inc. for some LRO specific code
   6 ******************************************************************************/
   7
   8 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
   9
  10 #include <linux/types.h>
  11 #include <linux/bitops.h>
  12 #include <linux/module.h>
  13 #include <linux/pci.h>
  14 #include <linux/netdevice.h>
  15 #include <linux/vmalloc.h>
  16 #include <linux/string.h>
  17 #include <linux/in.h>
  18 #include <linux/ip.h>
  19 #include <linux/tcp.h>
  20 #include <linux/sctp.h>
  21 #include <linux/ipv6.h>
  22 #include <linux/slab.h>
  23 #include <net/checksum.h>
  24 #include <net/ip6_checksum.h>
  25 #include <linux/ethtool.h>
  26 #include <linux/if.h>
  27 #include <linux/if_vlan.h>
  28 #include <linux/prefetch.h>
  29 #include <net/mpls.h>
  30 #include <linux/bpf.h>
  31 #include <linux/bpf_trace.h>
  32 #include <linux/atomic.h>
  33 #include <net/xfrm.h>
  34
  35 #include "ixgbevf.h"
  36
  37 const char ixgbevf_driver_name[] = "ixgbevf";
  38 static const char ixgbevf_driver_string[] =
  39         "Intel(R) 10 Gigabit PCI Express Virtual Function Network Driver";
  40
  41 static char ixgbevf_copyright[] =
  42         "Copyright (c) 2009 - 2018 Intel Corporation.";
  43
  44 static const struct ixgbevf_info *ixgbevf_info_tbl[] = {
  45         [board_82599_vf]        = &ixgbevf_82599_vf_info,
  46         [board_82599_vf_hv]     = &ixgbevf_82599_vf_hv_info,
  47         [board_X540_vf]         = &ixgbevf_X540_vf_info,
  48         [board_X540_vf_hv]      = &ixgbevf_X540_vf_hv_info,
  49         [board_X550_vf]         = &ixgbevf_X550_vf_info,
  50         [board_X550_vf_hv]      = &ixgbevf_X550_vf_hv_info,
  51         [board_X550EM_x_vf]     = &ixgbevf_X550EM_x_vf_info,
  52         [board_X550EM_x_vf_hv]  = &ixgbevf_X550EM_x_vf_hv_info,
  53         [board_x550em_a_vf]     = &ixgbevf_x550em_a_vf_info,
  54 };
  55
  56 /* ixgbevf_pci_tbl - PCI Device ID Table
  57  *
  58  * Wildcard entries (PCI_ANY_ID) should come last
  59  * Last entry must be all 0s
  60  *
  61  * { Vendor ID, Device ID, SubVendor ID, SubDevice ID,
  62  *   Class, Class Mask, private data (not used) }
  63  */
  64 static const struct pci_device_id ixgbevf_pci_tbl[] = {
  65         {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82599_VF), board_82599_vf },
  66         {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82599_VF_HV), board_82599_vf_hv },
  67         {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_X540_VF), board_X540_vf },
  68         {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_X540_VF_HV), board_X540_vf_hv },
  69         {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_X550_VF), board_X550_vf },
  70         {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_X550_VF_HV), board_X550_vf_hv },
  71         {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_X550EM_X_VF), board_X550EM_x_vf },
  72         {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_X550EM_X_VF_HV), board_X550EM_x_vf_hv},
  73         {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_X550EM_A_VF), board_x550em_a_vf },
  74         /* required last entry */
  75         {0, }
  76 };
  77 MODULE_DEVICE_TABLE(pci, ixgbevf_pci_tbl);
  78
  79 MODULE_DESCRIPTION("Intel(R) 10 Gigabit Virtual Function Network Driver");
  80 MODULE_LICENSE("GPL v2");
  81
  82 #define DEFAULT_MSG_ENABLE (NETIF_MSG_DRV|NETIF_MSG_PROBE|NETIF_MSG_LINK)
  83 static int debug = -1;
  84 module_param(debug, int, 0);
  85 MODULE_PARM_DESC(debug, "Debug level (0=none,...,16=all)");
  86
  87 static struct workqueue_struct *ixgbevf_wq;
  88
  89 static void ixgbevf_service_event_schedule(struct ixgbevf_adapter *adapter)
  90 {
  91         if (!test_bit(__IXGBEVF_DOWN, &adapter->state) &&
  92             !test_bit(__IXGBEVF_REMOVING, &adapter->state) &&
  93             !test_and_set_bit(__IXGBEVF_SERVICE_SCHED, &adapter->state))
  94                 queue_work(ixgbevf_wq, &adapter->service_task);
  95 }
  96
  97 static void ixgbevf_service_event_complete(struct ixgbevf_adapter *adapter)
  98 {
  99         BUG_ON(!test_bit(__IXGBEVF_SERVICE_SCHED, &adapter->state));
 100
 101         /* flush memory to make sure state is correct before next watchdog */
 102         smp_mb__before_atomic();
 103         clear_bit(__IXGBEVF_SERVICE_SCHED, &adapter->state);
 104 }
 105
 106 /* forward decls */
 107 static void ixgbevf_queue_reset_subtask(struct ixgbevf_adapter *adapter);
 108 static void ixgbevf_set_itr(struct ixgbevf_q_vector *q_vector);
 109 static void ixgbevf_free_all_rx_resources(struct ixgbevf_adapter *adapter);
 110 static bool ixgbevf_can_reuse_rx_page(struct ixgbevf_rx_buffer *rx_buffer);
 111 static void ixgbevf_reuse_rx_page(struct ixgbevf_ring *rx_ring,
 112                                   struct ixgbevf_rx_buffer *old_buff);
 113
 114 static void ixgbevf_remove_adapter(struct ixgbe_hw *hw)
 115 {
 116         struct ixgbevf_adapter *adapter = hw->back;
 117
 118         if (!hw->hw_addr)
 119                 return;
 120         hw->hw_addr = NULL;
 121         dev_err(&adapter->pdev->dev, "Adapter removed\n");
 122         if (test_bit(__IXGBEVF_SERVICE_INITED, &adapter->state))
 123                 ixgbevf_service_event_schedule(adapter);
 124 }
 125
 126 static void ixgbevf_check_remove(struct ixgbe_hw *hw, u32 reg)
 127 {
 128         u32 value;
 129
 130         /* The following check not only optimizes a bit by not
 131          * performing a read on the status register when the
 132          * register just read was a status register read that
 133          * returned IXGBE_FAILED_READ_REG. It also blocks any
 134          * potential recursion.
 135          */
 136         if (reg == IXGBE_VFSTATUS) {
 137                 ixgbevf_remove_adapter(hw);
 138                 return;
 139         }
 140         value = ixgbevf_read_reg(hw, IXGBE_VFSTATUS);
 141         if (value == IXGBE_FAILED_READ_REG)
 142                 ixgbevf_remove_adapter(hw);
 143 }
 144
 145 u32 ixgbevf_read_reg(struct ixgbe_hw *hw, u32 reg)
 146 {
 147         u8 __iomem *reg_addr = READ_ONCE(hw->hw_addr);
 148         u32 value;
 149
 150         if (IXGBE_REMOVED(reg_addr))
 151                 return IXGBE_FAILED_READ_REG;
 152         value = readl(reg_addr + reg);
 153         if (unlikely(value == IXGBE_FAILED_READ_REG))
 154                 ixgbevf_check_remove(hw, reg);
 155         return value;
 156 }
 157
 158 /**
 159  * ixgbevf_set_ivar - set IVAR registers - maps interrupt causes to vectors
 160  * @adapter: pointer to adapter struct
 161  * @direction: 0 for Rx, 1 for Tx, -1 for other causes
 162  * @queue: queue to map the corresponding interrupt to
 163  * @msix_vector: the vector to map to the corresponding queue
 164  **/
 165 static void ixgbevf_set_ivar(struct ixgbevf_adapter *adapter, s8 direction,
 166                              u8 queue, u8 msix_vector)
 167 {
 168         u32 ivar, index;
 169         struct ixgbe_hw *hw = &adapter->hw;
 170
 171         if (direction == -1) {
 172                 /* other causes */
 173                 msix_vector |= IXGBE_IVAR_ALLOC_VAL;
 174                 ivar = IXGBE_READ_REG(hw, IXGBE_VTIVAR_MISC);
 175                 ivar &= ~0xFF;
 176                 ivar |= msix_vector;
 177                 IXGBE_WRITE_REG(hw, IXGBE_VTIVAR_MISC, ivar);
 178         } else {
 179                 /* Tx or Rx causes */
 180                 msix_vector |= IXGBE_IVAR_ALLOC_VAL;
 181                 index = ((16 * (queue & 1)) + (8 * direction));
 182                 ivar = IXGBE_READ_REG(hw, IXGBE_VTIVAR(queue >> 1));
 183                 ivar &= ~(0xFF << index);
 184                 ivar |= (msix_vector << index);
 185                 IXGBE_WRITE_REG(hw, IXGBE_VTIVAR(queue >> 1), ivar);
 186         }
 187 }
 188
 189 static u64 ixgbevf_get_tx_completed(struct ixgbevf_ring *ring)
 190 {
 191         return ring->stats.packets;
 192 }
 193
 194 static u32 ixgbevf_get_tx_pending(struct ixgbevf_ring *ring)
 195 {
 196         struct ixgbevf_adapter *adapter = netdev_priv(ring->netdev);
 197         struct ixgbe_hw *hw = &adapter->hw;
 198
 199         u32 head = IXGBE_READ_REG(hw, IXGBE_VFTDH(ring->reg_idx));
 200         u32 tail = IXGBE_READ_REG(hw, IXGBE_VFTDT(ring->reg_idx));
 201
 202         if (head != tail)
 203                 return (head < tail) ?
 204                         tail - head : (tail + ring->count - head);
 205
 206         return 0;
 207 }
 208
 209 static inline bool ixgbevf_check_tx_hang(struct ixgbevf_ring *tx_ring)
 210 {
 211         u32 tx_done = ixgbevf_get_tx_completed(tx_ring);
 212         u32 tx_done_old = tx_ring->tx_stats.tx_done_old;
 213         u32 tx_pending = ixgbevf_get_tx_pending(tx_ring);
 214
 215         clear_check_for_tx_hang(tx_ring);
 216
 217         /* Check for a hung queue, but be thorough. This verifies
 218          * that a transmit has been completed since the previous
 219          * check AND there is at least one packet pending. The
 220          * ARMED bit is set to indicate a potential hang.
 221          */
 222         if ((tx_done_old == tx_done) && tx_pending) {
 223                 /* make sure it is true for two checks in a row */
 224                 return test_and_set_bit(__IXGBEVF_HANG_CHECK_ARMED,
 225                                         &tx_ring->state);
 226         }
 227         /* reset the countdown */
 228         clear_bit(__IXGBEVF_HANG_CHECK_ARMED, &tx_ring->state);
 229
 230         /* update completed stats and continue */
 231         tx_ring->tx_stats.tx_done_old = tx_done;
 232
 233         return false;
 234 }
 235
 236 static void ixgbevf_tx_timeout_reset(struct ixgbevf_adapter *adapter)
 237 {
 238         /* Do the reset outside of interrupt context */
 239         if (!test_bit(__IXGBEVF_DOWN, &adapter->state)) {
 240                 set_bit(__IXGBEVF_RESET_REQUESTED, &adapter->state);
 241                 ixgbevf_service_event_schedule(adapter);
 242         }
 243 }
 244
 245 /**
 246  * ixgbevf_tx_timeout - Respond to a Tx Hang
 247  * @netdev: network interface device structure
 248  * @txqueue: transmit queue hanging (unused)
 249  **/
 250 static void ixgbevf_tx_timeout(struct net_device *netdev, unsigned int __always_unused txqueue)
 251 {
 252         struct ixgbevf_adapter *adapter = netdev_priv(netdev);
 253
 254         ixgbevf_tx_timeout_reset(adapter);
 255 }
 256
 257 /**
 258  * ixgbevf_clean_tx_irq - Reclaim resources after transmit completes
 259  * @q_vector: board private structure
 260  * @tx_ring: tx ring to clean
 261  * @napi_budget: Used to determine if we are in netpoll
 262  **/
 263 static bool ixgbevf_clean_tx_irq(struct ixgbevf_q_vector *q_vector,
 264                                  struct ixgbevf_ring *tx_ring, int napi_budget)
 265 {
 266         struct ixgbevf_adapter *adapter = q_vector->adapter;
 267         struct ixgbevf_tx_buffer *tx_buffer;
 268         union ixgbe_adv_tx_desc *tx_desc;
 269         unsigned int total_bytes = 0, total_packets = 0, total_ipsec = 0;
 270         unsigned int budget = tx_ring->count / 2;
 271         unsigned int i = tx_ring->next_to_clean;
 272
 273         if (test_bit(__IXGBEVF_DOWN, &adapter->state))
 274                 return true;
 275
 276         tx_buffer = &tx_ring->tx_buffer_info[i];
 277         tx_desc = IXGBEVF_TX_DESC(tx_ring, i);
 278         i -= tx_ring->count;
 279
 280         do {
 281                 union ixgbe_adv_tx_desc *eop_desc = tx_buffer->next_to_watch;
 282
 283                 /* if next_to_watch is not set then there is no work pending */
 284                 if (!eop_desc)
 285                         break;
 286
 287                 /* prevent any other reads prior to eop_desc */
 288                 smp_rmb();
 289
 290                 /* if DD is not set pending work has not been completed */
 291                 if (!(eop_desc->wb.status & cpu_to_le32(IXGBE_TXD_STAT_DD)))
 292                         break;
 293
 294                 /* clear next_to_watch to prevent false hangs */
 295                 tx_buffer->next_to_watch = NULL;
 296
 297                 /* update the statistics for this packet */
 298                 total_bytes += tx_buffer->bytecount;
 299                 total_packets += tx_buffer->gso_segs;
 300                 if (tx_buffer->tx_flags & IXGBE_TX_FLAGS_IPSEC)
 301                         total_ipsec++;
 302
 303                 /* free the skb */
 304                 if (ring_is_xdp(tx_ring))
 305                         page_frag_free(tx_buffer->data);
 306                 else
 307                         napi_consume_skb(tx_buffer->skb, napi_budget);
 308
 309                 /* unmap skb header data */
 310                 dma_unmap_single(tx_ring->dev,
 311                                  dma_unmap_addr(tx_buffer, dma),
 312                                  dma_unmap_len(tx_buffer, len),
 313                                  DMA_TO_DEVICE);
 314
 315                 /* clear tx_buffer data */
 316                 dma_unmap_len_set(tx_buffer, len, 0);
 317
 318                 /* unmap remaining buffers */
 319                 while (tx_desc != eop_desc) {
 320                         tx_buffer++;
 321                         tx_desc++;
 322                         i++;
 323                         if (unlikely(!i)) {
 324                                 i -= tx_ring->count;
 325                                 tx_buffer = tx_ring->tx_buffer_info;
 326                                 tx_desc = IXGBEVF_TX_DESC(tx_ring, 0);
 327                         }
 328
 329                         /* unmap any remaining paged data */
 330                         if (dma_unmap_len(tx_buffer, len)) {
 331                                 dma_unmap_page(tx_ring->dev,
 332                                                dma_unmap_addr(tx_buffer, dma),
 333                                                dma_unmap_len(tx_buffer, len),
 334                                                DMA_TO_DEVICE);
 335                                 dma_unmap_len_set(tx_buffer, len, 0);
 336                         }
 337                 }
 338
 339                 /* move us one more past the eop_desc for start of next pkt */
 340                 tx_buffer++;
 341                 tx_desc++;
 342                 i++;
 343                 if (unlikely(!i)) {
 344                         i -= tx_ring->count;
 345                         tx_buffer = tx_ring->tx_buffer_info;
 346                         tx_desc = IXGBEVF_TX_DESC(tx_ring, 0);
 347                 }
 348
 349                 /* issue prefetch for next Tx descriptor */
 350                 prefetch(tx_desc);
 351
 352                 /* update budget accounting */
 353                 budget--;
 354         } while (likely(budget));
 355
 356         i += tx_ring->count;
 357         tx_ring->next_to_clean = i;
 358         u64_stats_update_begin(&tx_ring->syncp);
 359         tx_ring->stats.bytes += total_bytes;
 360         tx_ring->stats.packets += total_packets;
 361         u64_stats_update_end(&tx_ring->syncp);
 362         q_vector->tx.total_bytes += total_bytes;
 363         q_vector->tx.total_packets += total_packets;
 364         adapter->tx_ipsec += total_ipsec;
 365
 366         if (check_for_tx_hang(tx_ring) && ixgbevf_check_tx_hang(tx_ring)) {
 367                 struct ixgbe_hw *hw = &adapter->hw;
 368                 union ixgbe_adv_tx_desc *eop_desc;
 369
 370                 eop_desc = tx_ring->tx_buffer_info[i].next_to_watch;
 371
 372                 pr_err("Detected Tx Unit Hang%s\n"
 373                        "  Tx Queue             <%d>\n"
 374                        "  TDH, TDT             <%x>, <%x>\n"
 375                        "  next_to_use          <%x>\n"
 376                        "  next_to_clean        <%x>\n"
 377                        "tx_buffer_info[next_to_clean]\n"
 378                        "  next_to_watch        <%p>\n"
 379                        "  eop_desc->wb.status  <%x>\n"
 380                        "  time_stamp           <%lx>\n"
 381                        "  jiffies              <%lx>\n",
 382                        ring_is_xdp(tx_ring) ? " XDP" : "",
 383                        tx_ring->queue_index,
 384                        IXGBE_READ_REG(hw, IXGBE_VFTDH(tx_ring->reg_idx)),
 385                        IXGBE_READ_REG(hw, IXGBE_VFTDT(tx_ring->reg_idx)),
 386                        tx_ring->next_to_use, i,
 387                        eop_desc, (eop_desc ? eop_desc->wb.status : 0),
 388                        tx_ring->tx_buffer_info[i].time_stamp, jiffies);
 389
 390                 if (!ring_is_xdp(tx_ring))
 391                         netif_stop_subqueue(tx_ring->netdev,
 392                                             tx_ring->queue_index);
 393
 394                 /* schedule immediate reset if we believe we hung */
 395                 ixgbevf_tx_timeout_reset(adapter);
 396
 397                 return true;
 398         }
 399
 400         if (ring_is_xdp(tx_ring))
 401                 return !!budget;
 402
 403 #define TX_WAKE_THRESHOLD (DESC_NEEDED * 2)
 404         if (unlikely(total_packets && netif_carrier_ok(tx_ring->netdev) &&
 405                      (ixgbevf_desc_unused(tx_ring) >= TX_WAKE_THRESHOLD))) {
 406                 /* Make sure that anybody stopping the queue after this
 407                  * sees the new next_to_clean.
 408                  */
 409                 smp_mb();
 410
 411                 if (__netif_subqueue_stopped(tx_ring->netdev,
 412                                              tx_ring->queue_index) &&
 413                     !test_bit(__IXGBEVF_DOWN, &adapter->state)) {
 414                         netif_wake_subqueue(tx_ring->netdev,
 415                                             tx_ring->queue_index);
 416                         ++tx_ring->tx_stats.restart_queue;
 417                 }
 418         }
 419
 420         return !!budget;
 421 }
 422
 423 /**
 424  * ixgbevf_rx_skb - Helper function to determine proper Rx method
 425  * @q_vector: structure containing interrupt and ring information
 426  * @skb: packet to send up
 427  **/
 428 static void ixgbevf_rx_skb(struct ixgbevf_q_vector *q_vector,
 429                            struct sk_buff *skb)
 430 {
 431         napi_gro_receive(&q_vector->napi, skb);
 432 }
 433
 434 #define IXGBE_RSS_L4_TYPES_MASK \
 435         ((1ul << IXGBE_RXDADV_RSSTYPE_IPV4_TCP) | \
 436          (1ul << IXGBE_RXDADV_RSSTYPE_IPV4_UDP) | \
 437          (1ul << IXGBE_RXDADV_RSSTYPE_IPV6_TCP) | \
 438          (1ul << IXGBE_RXDADV_RSSTYPE_IPV6_UDP))
 439
 440 static inline void ixgbevf_rx_hash(struct ixgbevf_ring *ring,
 441                                    union ixgbe_adv_rx_desc *rx_desc,
 442                                    struct sk_buff *skb)
 443 {
 444         u16 rss_type;
 445
 446         if (!(ring->netdev->features & NETIF_F_RXHASH))
 447                 return;
 448
 449         rss_type = le16_to_cpu(rx_desc->wb.lower.lo_dword.hs_rss.pkt_info) &
 450                    IXGBE_RXDADV_RSSTYPE_MASK;
 451
 452         if (!rss_type)
 453                 return;
 454
 455         skb_set_hash(skb, le32_to_cpu(rx_desc->wb.lower.hi_dword.rss),
 456                      (IXGBE_RSS_L4_TYPES_MASK & (1ul << rss_type)) ?
 457                      PKT_HASH_TYPE_L4 : PKT_HASH_TYPE_L3);
 458 }
 459
 460 /**
 461  * ixgbevf_rx_checksum - indicate in skb if hw indicated a good cksum
 462  * @ring: structure containig ring specific data
 463  * @rx_desc: current Rx descriptor being processed
 464  * @skb: skb currently being received and modified
 465  **/
 466 static inline void ixgbevf_rx_checksum(struct ixgbevf_ring *ring,
 467                                        union ixgbe_adv_rx_desc *rx_desc,
 468                                        struct sk_buff *skb)
 469 {
 470         skb_checksum_none_assert(skb);
 471
 472         /* Rx csum disabled */
 473         if (!(ring->netdev->features & NETIF_F_RXCSUM))
 474                 return;
 475
 476         /* if IP and error */
 477         if (ixgbevf_test_staterr(rx_desc, IXGBE_RXD_STAT_IPCS) &&
 478             ixgbevf_test_staterr(rx_desc, IXGBE_RXDADV_ERR_IPE)) {
 479                 ring->rx_stats.csum_err++;
 480                 return;
 481         }
 482
 483         if (!ixgbevf_test_staterr(rx_desc, IXGBE_RXD_STAT_L4CS))
 484                 return;
 485
 486         if (ixgbevf_test_staterr(rx_desc, IXGBE_RXDADV_ERR_TCPE)) {
 487                 ring->rx_stats.csum_err++;
 488                 return;
 489         }
 490
 491         /* It must be a TCP or UDP packet with a valid checksum */
 492         skb->ip_summed = CHECKSUM_UNNECESSARY;
 493 }
 494
 495 /**
 496  * ixgbevf_process_skb_fields - Populate skb header fields from Rx descriptor
 497  * @rx_ring: rx descriptor ring packet is being transacted on
 498  * @rx_desc: pointer to the EOP Rx descriptor
 499  * @skb: pointer to current skb being populated
 500  *
 501  * This function checks the ring, descriptor, and packet information in
 502  * order to populate the checksum, VLAN, protocol, and other fields within
 503  * the skb.
 504  **/
 505 static void ixgbevf_process_skb_fields(struct ixgbevf_ring *rx_ring,
 506                                        union ixgbe_adv_rx_desc *rx_desc,
 507                                        struct sk_buff *skb)
 508 {
 509         ixgbevf_rx_hash(rx_ring, rx_desc, skb);
 510         ixgbevf_rx_checksum(rx_ring, rx_desc, skb);
 511
 512         if (ixgbevf_test_staterr(rx_desc, IXGBE_RXD_STAT_VP)) {
 513                 u16 vid = le16_to_cpu(rx_desc->wb.upper.vlan);
 514                 unsigned long *active_vlans = netdev_priv(rx_ring->netdev);
 515
 516                 if (test_bit(vid & VLAN_VID_MASK, active_vlans))
 517                         __vlan_hwaccel_put_tag(skb, htons(ETH_P_8021Q), vid);
 518         }
 519
 520         if (ixgbevf_test_staterr(rx_desc, IXGBE_RXDADV_STAT_SECP))
 521                 ixgbevf_ipsec_rx(rx_ring, rx_desc, skb);
 522
 523         skb->protocol = eth_type_trans(skb, rx_ring->netdev);
 524 }
 525
 526 static
 527 struct ixgbevf_rx_buffer *ixgbevf_get_rx_buffer(struct ixgbevf_ring *rx_ring,
 528                                                 const unsigned int size)
 529 {
 530         struct ixgbevf_rx_buffer *rx_buffer;
 531
 532         rx_buffer = &rx_ring->rx_buffer_info[rx_ring->next_to_clean];
 533         prefetchw(rx_buffer->page);
 534
 535         /* we are reusing so sync this buffer for CPU use */
 536         dma_sync_single_range_for_cpu(rx_ring->dev,
 537                                       rx_buffer->dma,
 538                                       rx_buffer->page_offset,
 539                                       size,
 540                                       DMA_FROM_DEVICE);
 541
 542         rx_buffer->pagecnt_bias--;
 543
 544         return rx_buffer;
 545 }
 546
 547 static void ixgbevf_put_rx_buffer(struct ixgbevf_ring *rx_ring,
 548                                   struct ixgbevf_rx_buffer *rx_buffer,
 549                                   struct sk_buff *skb)
 550 {
 551         if (ixgbevf_can_reuse_rx_page(rx_buffer)) {
 552                 /* hand second half of page back to the ring */
 553                 ixgbevf_reuse_rx_page(rx_ring, rx_buffer);
 554         } else {
 555                 if (IS_ERR(skb))
 556                         /* We are not reusing the buffer so unmap it and free
 557                          * any references we are holding to it
 558                          */
 559                         dma_unmap_page_attrs(rx_ring->dev, rx_buffer->dma,
 560                                              ixgbevf_rx_pg_size(rx_ring),
 561                                              DMA_FROM_DEVICE,
 562                                              IXGBEVF_RX_DMA_ATTR);
 563                 __page_frag_cache_drain(rx_buffer->page,
 564                                         rx_buffer->pagecnt_bias);
 565         }
 566
 567         /* clear contents of rx_buffer */
 568         rx_buffer->page = NULL;
 569 }
 570
 571 /**
 572  * ixgbevf_is_non_eop - process handling of non-EOP buffers
 573  * @rx_ring: Rx ring being processed
 574  * @rx_desc: Rx descriptor for current buffer
 575  *
 576  * This function updates next to clean.  If the buffer is an EOP buffer
 577  * this function exits returning false, otherwise it will place the
 578  * sk_buff in the next buffer to be chained and return true indicating
 579  * that this is in fact a non-EOP buffer.
 580  **/
 581 static bool ixgbevf_is_non_eop(struct ixgbevf_ring *rx_ring,
 582                                union ixgbe_adv_rx_desc *rx_desc)
 583 {
 584         u32 ntc = rx_ring->next_to_clean + 1;
 585
 586         /* fetch, update, and store next to clean */
 587         ntc = (ntc < rx_ring->count) ? ntc : 0;
 588         rx_ring->next_to_clean = ntc;
 589
 590         prefetch(IXGBEVF_RX_DESC(rx_ring, ntc));
 591
 592         if (likely(ixgbevf_test_staterr(rx_desc, IXGBE_RXD_STAT_EOP)))
 593                 return false;
 594
 595         return true;
 596 }
 597
 598 static inline unsigned int ixgbevf_rx_offset(struct ixgbevf_ring *rx_ring)
 599 {
 600         return ring_uses_build_skb(rx_ring) ? IXGBEVF_SKB_PAD : 0;
 601 }
 602
 603 static bool ixgbevf_alloc_mapped_page(struct ixgbevf_ring *rx_ring,
 604                                       struct ixgbevf_rx_buffer *bi)
 605 {
 606         struct page *page = bi->page;
 607         dma_addr_t dma;
 608
 609         /* since we are recycling buffers we should seldom need to alloc */
 610         if (likely(page))
 611                 return true;
 612
 613         /* alloc new page for storage */
 614         page = dev_alloc_pages(ixgbevf_rx_pg_order(rx_ring));
 615         if (unlikely(!page)) {
 616                 rx_ring->rx_stats.alloc_rx_page_failed++;
 617                 return false;
 618         }
 619
 620         /* map page for use */
 621         dma = dma_map_page_attrs(rx_ring->dev, page, 0,
 622                                  ixgbevf_rx_pg_size(rx_ring),
 623                                  DMA_FROM_DEVICE, IXGBEVF_RX_DMA_ATTR);
 624
 625         /* if mapping failed free memory back to system since
 626          * there isn't much point in holding memory we can't use
 627          */
 628         if (dma_mapping_error(rx_ring->dev, dma)) {
 629                 __free_pages(page, ixgbevf_rx_pg_order(rx_ring));
 630
 631                 rx_ring->rx_stats.alloc_rx_page_failed++;
 632                 return false;
 633         }
 634
 635         bi->dma = dma;
 636         bi->page = page;
 637         bi->page_offset = ixgbevf_rx_offset(rx_ring);
 638         bi->pagecnt_bias = 1;
 639         rx_ring->rx_stats.alloc_rx_page++;
 640
 641         return true;
 642 }
 643
 644 /**
 645  * ixgbevf_alloc_rx_buffers - Replace used receive buffers; packet split
 646  * @rx_ring: rx descriptor ring (for a specific queue) to setup buffers on
 647  * @cleaned_count: number of buffers to replace
 648  **/
 649 static void ixgbevf_alloc_rx_buffers(struct ixgbevf_ring *rx_ring,
 650                                      u16 cleaned_count)
 651 {
 652         union ixgbe_adv_rx_desc *rx_desc;
 653         struct ixgbevf_rx_buffer *bi;
 654         unsigned int i = rx_ring->next_to_use;
 655
 656         /* nothing to do or no valid netdev defined */
 657         if (!cleaned_count || !rx_ring->netdev)
 658                 return;
 659
 660         rx_desc = IXGBEVF_RX_DESC(rx_ring, i);
 661         bi = &rx_ring->rx_buffer_info[i];
 662         i -= rx_ring->count;
 663
 664         do {
 665                 if (!ixgbevf_alloc_mapped_page(rx_ring, bi))
 666                         break;
 667
 668                 /* sync the buffer for use by the device */
 669                 dma_sync_single_range_for_device(rx_ring->dev, bi->dma,
 670                                                  bi->page_offset,
 671                                                  ixgbevf_rx_bufsz(rx_ring),
 672                                                  DMA_FROM_DEVICE);
 673
 674                 /* Refresh the desc even if pkt_addr didn't change
 675                  * because each write-back erases this info.
 676                  */
 677                 rx_desc->read.pkt_addr = cpu_to_le64(bi->dma + bi->page_offset);
 678
 679                 rx_desc++;
 680                 bi++;
 681                 i++;
 682                 if (unlikely(!i)) {
 683                         rx_desc = IXGBEVF_RX_DESC(rx_ring, 0);
 684                         bi = rx_ring->rx_buffer_info;
 685                         i -= rx_ring->count;
 686                 }
 687
 688                 /* clear the length for the next_to_use descriptor */
 689                 rx_desc->wb.upper.length = 0;
 690
 691                 cleaned_count--;
 692         } while (cleaned_count);
 693
 694         i += rx_ring->count;
 695
 696         if (rx_ring->next_to_use != i) {
 697                 /* record the next descriptor to use */
 698                 rx_ring->next_to_use = i;
 699
 700                 /* update next to alloc since we have filled the ring */
 701                 rx_ring->next_to_alloc = i;
 702
 703                 /* Force memory writes to complete before letting h/w
 704                  * know there are new descriptors to fetch.  (Only
 705                  * applicable for weak-ordered memory model archs,
 706                  * such as IA-64).
 707                  */
 708                 wmb();
 709                 ixgbevf_write_tail(rx_ring, i);
 710         }
 711 }
 712
 713 /**
 714  * ixgbevf_cleanup_headers - Correct corrupted or empty headers
 715  * @rx_ring: rx descriptor ring packet is being transacted on
 716  * @rx_desc: pointer to the EOP Rx descriptor
 717  * @skb: pointer to current skb being fixed
 718  *
 719  * Check for corrupted packet headers caused by senders on the local L2
 720  * embedded NIC switch not setting up their Tx Descriptors right.  These
 721  * should be very rare.
 722  *
 723  * Also address the case where we are pulling data in on pages only
 724  * and as such no data is present in the skb header.
 725  *
 726  * In addition if skb is not at least 60 bytes we need to pad it so that
 727  * it is large enough to qualify as a valid Ethernet frame.
 728  *
 729  * Returns true if an error was encountered and skb was freed.
 730  **/
 731 static bool ixgbevf_cleanup_headers(struct ixgbevf_ring *rx_ring,
 732                                     union ixgbe_adv_rx_desc *rx_desc,
 733                                     struct sk_buff *skb)
 734 {
 735         /* XDP packets use error pointer so abort at this point */
 736         if (IS_ERR(skb))
 737                 return true;
 738
 739         /* verify that the packet does not have any known errors */
 740         if (unlikely(ixgbevf_test_staterr(rx_desc,
 741                                           IXGBE_RXDADV_ERR_FRAME_ERR_MASK))) {
 742                 struct net_device *netdev = rx_ring->netdev;
 743
 744                 if (!(netdev->features & NETIF_F_RXALL)) {
 745                         dev_kfree_skb_any(skb);
 746                         return true;
 747                 }
 748         }
 749
 750         /* if eth_skb_pad returns an error the skb was freed */
 751         if (eth_skb_pad(skb))
 752                 return true;
 753
 754         return false;
 755 }
 756
 757 /**
 758  * ixgbevf_reuse_rx_page - page flip buffer and store it back on the ring
 759  * @rx_ring: rx descriptor ring to store buffers on
 760  * @old_buff: donor buffer to have page reused
 761  *
 762  * Synchronizes page for reuse by the adapter
 763  **/
 764 static void ixgbevf_reuse_rx_page(struct ixgbevf_ring *rx_ring,
 765                                   struct ixgbevf_rx_buffer *old_buff)
 766 {
 767         struct ixgbevf_rx_buffer *new_buff;
 768         u16 nta = rx_ring->next_to_alloc;
 769
 770         new_buff = &rx_ring->rx_buffer_info[nta];
 771
 772         /* update, and store next to alloc */
 773         nta++;
 774         rx_ring->next_to_alloc = (nta < rx_ring->count) ? nta : 0;
 775
 776         /* transfer page from old buffer to new buffer */
 777         new_buff->page = old_buff->page;
 778         new_buff->dma = old_buff->dma;
 779         new_buff->page_offset = old_buff->page_offset;
 780         new_buff->pagecnt_bias = old_buff->pagecnt_bias;
 781 }
 782
 783 static bool ixgbevf_can_reuse_rx_page(struct ixgbevf_rx_buffer *rx_buffer)
 784 {
 785         unsigned int pagecnt_bias = rx_buffer->pagecnt_bias;
 786         struct page *page = rx_buffer->page;
 787
 788         /* avoid re-using remote and pfmemalloc pages */
 789         if (!dev_page_is_reusable(page))
 790                 return false;
 791
 792 #if (PAGE_SIZE < 8192)
 793         /* if we are only owner of page we can reuse it */
 794         if (unlikely((page_ref_count(page) - pagecnt_bias) > 1))
 795                 return false;
 796 #else
 797 #define IXGBEVF_LAST_OFFSET \
 798         (SKB_WITH_OVERHEAD(PAGE_SIZE) - IXGBEVF_RXBUFFER_2048)
 799
 800         if (rx_buffer->page_offset > IXGBEVF_LAST_OFFSET)
 801                 return false;
 802
 803 #endif
 804
 805         /* If we have drained the page fragment pool we need to update
 806          * the pagecnt_bias and page count so that we fully restock the
 807          * number of references the driver holds.
 808          */
 809         if (unlikely(!pagecnt_bias)) {
 810                 page_ref_add(page, USHRT_MAX);
 811                 rx_buffer->pagecnt_bias = USHRT_MAX;
 812         }
 813
 814         return true;
 815 }
 816
 817 /**
 818  * ixgbevf_add_rx_frag - Add contents of Rx buffer to sk_buff
 819  * @rx_ring: rx descriptor ring to transact packets on
 820  * @rx_buffer: buffer containing page to add
 821  * @skb: sk_buff to place the data into
 822  * @size: size of buffer to be added
 823  *
 824  * This function will add the data contained in rx_buffer->page to the skb.
 825  **/
 826 static void ixgbevf_add_rx_frag(struct ixgbevf_ring *rx_ring,
 827                                 struct ixgbevf_rx_buffer *rx_buffer,
 828                                 struct sk_buff *skb,
 829                                 unsigned int size)
 830 {
 831 #if (PAGE_SIZE < 8192)
 832         unsigned int truesize = ixgbevf_rx_pg_size(rx_ring) / 2;
 833 #else
 834         unsigned int truesize = ring_uses_build_skb(rx_ring) ?
 835                                 SKB_DATA_ALIGN(IXGBEVF_SKB_PAD + size) :
 836                                 SKB_DATA_ALIGN(size);
 837 #endif
 838         skb_add_rx_frag(skb, skb_shinfo(skb)->nr_frags, rx_buffer->page,
 839                         rx_buffer->page_offset, size, truesize);
 840 #if (PAGE_SIZE < 8192)
 841         rx_buffer->page_offset ^= truesize;
 842 #else
 843         rx_buffer->page_offset += truesize;
 844 #endif
 845 }
 846
 847 static
 848 struct sk_buff *ixgbevf_construct_skb(struct ixgbevf_ring *rx_ring,
 849                                       struct ixgbevf_rx_buffer *rx_buffer,
 850                                       struct xdp_buff *xdp,
 851                                       union ixgbe_adv_rx_desc *rx_desc)
 852 {
 853         unsigned int size = xdp->data_end - xdp->data;
 854 #if (PAGE_SIZE < 8192)
 855         unsigned int truesize = ixgbevf_rx_pg_size(rx_ring) / 2;
 856 #else
 857         unsigned int truesize = SKB_DATA_ALIGN(xdp->data_end -
 858                                                xdp->data_hard_start);
 859 #endif
 860         unsigned int headlen;
 861         struct sk_buff *skb;
 862
 863         /* prefetch first cache line of first page */
 864         net_prefetch(xdp->data);
 865
 866         /* Note, we get here by enabling legacy-rx via:
 867          *
 868          *    ethtool --set-priv-flags <dev> legacy-rx on
 869          *
 870          * In this mode, we currently get 0 extra XDP headroom as
 871          * opposed to having legacy-rx off, where we process XDP
 872          * packets going to stack via ixgbevf_build_skb().
 873          *
 874          * For ixgbevf_construct_skb() mode it means that the
 875          * xdp->data_meta will always point to xdp->data, since
 876          * the helper cannot expand the head. Should this ever
 877          * changed in future for legacy-rx mode on, then lets also
 878          * add xdp->data_meta handling here.
 879          */
 880
 881         /* allocate a skb to store the frags */
 882         skb = napi_alloc_skb(&rx_ring->q_vector->napi, IXGBEVF_RX_HDR_SIZE);
 883         if (unlikely(!skb))
 884                 return NULL;
 885
 886         /* Determine available headroom for copy */
 887         headlen = size;
 888         if (headlen > IXGBEVF_RX_HDR_SIZE)
 889                 headlen = eth_get_headlen(skb->dev, xdp->data,
 890                                           IXGBEVF_RX_HDR_SIZE);
 891
 892         /* align pull length to size of long to optimize memcpy performance */
 893         memcpy(__skb_put(skb, headlen), xdp->data,
 894                ALIGN(headlen, sizeof(long)));
 895
 896         /* update all of the pointers */
 897         size -= headlen;
 898         if (size) {
 899                 skb_add_rx_frag(skb, 0, rx_buffer->page,
 900                                 (xdp->data + headlen) -
 901                                         page_address(rx_buffer->page),
 902                                 size, truesize);
 903 #if (PAGE_SIZE < 8192)
 904                 rx_buffer->page_offset ^= truesize;
 905 #else
 906                 rx_buffer->page_offset += truesize;
 907 #endif
 908         } else {
 909                 rx_buffer->pagecnt_bias++;
 910         }
 911
 912         return skb;
 913 }
 914
 915 static inline void ixgbevf_irq_enable_queues(struct ixgbevf_adapter *adapter,
 916                                              u32 qmask)
 917 {
 918         struct ixgbe_hw *hw = &adapter->hw;
 919
 920         IXGBE_WRITE_REG(hw, IXGBE_VTEIMS, qmask);
 921 }
 922
 923 static struct sk_buff *ixgbevf_build_skb(struct ixgbevf_ring *rx_ring,
 924                                          struct ixgbevf_rx_buffer *rx_buffer,
 925                                          struct xdp_buff *xdp,
 926                                          union ixgbe_adv_rx_desc *rx_desc)
 927 {
 928         unsigned int metasize = xdp->data - xdp->data_meta;
 929 #if (PAGE_SIZE < 8192)
 930         unsigned int truesize = ixgbevf_rx_pg_size(rx_ring) / 2;
 931 #else
 932         unsigned int truesize = SKB_DATA_ALIGN(sizeof(struct skb_shared_info)) +
 933                                 SKB_DATA_ALIGN(xdp->data_end -
 934                                                xdp->data_hard_start);
 935 #endif
 936         struct sk_buff *skb;
 937
 938         /* Prefetch first cache line of first page. If xdp->data_meta
 939          * is unused, this points to xdp->data, otherwise, we likely
 940          * have a consumer accessing first few bytes of meta data,
 941          * and then actual data.
 942          */
 943         net_prefetch(xdp->data_meta);
 944
 945         /* build an skb around the page buffer */
 946         skb = napi_build_skb(xdp->data_hard_start, truesize);
 947         if (unlikely(!skb))
 948                 return NULL;
 949
 950         /* update pointers within the skb to store the data */
 951         skb_reserve(skb, xdp->data - xdp->data_hard_start);
 952         __skb_put(skb, xdp->data_end - xdp->data);
 953         if (metasize)
 954                 skb_metadata_set(skb, metasize);
 955
 956         /* update buffer offset */
 957 #if (PAGE_SIZE < 8192)
 958         rx_buffer->page_offset ^= truesize;
 959 #else
 960         rx_buffer->page_offset += truesize;
 961 #endif
 962
 963         return skb;
 964 }
 965
 966 #define IXGBEVF_XDP_PASS 0
 967 #define IXGBEVF_XDP_CONSUMED 1
 968 #define IXGBEVF_XDP_TX 2
 969
 970 static int ixgbevf_xmit_xdp_ring(struct ixgbevf_ring *ring,
 971                                  struct xdp_buff *xdp)
 972 {
 973         struct ixgbevf_tx_buffer *tx_buffer;
 974         union ixgbe_adv_tx_desc *tx_desc;
 975         u32 len, cmd_type;
 976         dma_addr_t dma;
 977         u16 i;
 978
 979         len = xdp->data_end - xdp->data;
 980
 981         if (unlikely(!ixgbevf_desc_unused(ring)))
 982                 return IXGBEVF_XDP_CONSUMED;
 983
 984         dma = dma_map_single(ring->dev, xdp->data, len, DMA_TO_DEVICE);
 985         if (dma_mapping_error(ring->dev, dma))
 986                 return IXGBEVF_XDP_CONSUMED;
 987
 988         /* record the location of the first descriptor for this packet */
 989         i = ring->next_to_use;
 990         tx_buffer = &ring->tx_buffer_info[i];
 991
 992         dma_unmap_len_set(tx_buffer, len, len);
 993         dma_unmap_addr_set(tx_buffer, dma, dma);
 994         tx_buffer->data = xdp->data;
 995         tx_buffer->bytecount = len;
 996         tx_buffer->gso_segs = 1;
 997         tx_buffer->protocol = 0;
 998
 999         /* Populate minimal context descriptor that will provide for the
1000          * fact that we are expected to process Ethernet frames.
1001          */
1002         if (!test_bit(__IXGBEVF_TX_XDP_RING_PRIMED, &ring->state)) {
1003                 struct ixgbe_adv_tx_context_desc *context_desc;
1004
1005                 set_bit(__IXGBEVF_TX_XDP_RING_PRIMED, &ring->state);
1006
1007                 context_desc = IXGBEVF_TX_CTXTDESC(ring, 0);
1008                 context_desc->vlan_macip_lens   =
1009                         cpu_to_le32(ETH_HLEN << IXGBE_ADVTXD_MACLEN_SHIFT);
1010                 context_desc->fceof_saidx       = 0;
1011                 context_desc->type_tucmd_mlhl   =
1012                         cpu_to_le32(IXGBE_TXD_CMD_DEXT |
1013                                     IXGBE_ADVTXD_DTYP_CTXT);
1014                 context_desc->mss_l4len_idx     = 0;
1015
1016                 i = 1;
1017         }
1018
1019         /* put descriptor type bits */
1020         cmd_type = IXGBE_ADVTXD_DTYP_DATA |
1021                    IXGBE_ADVTXD_DCMD_DEXT |
1022                    IXGBE_ADVTXD_DCMD_IFCS;
1023         cmd_type |= len | IXGBE_TXD_CMD;
1024
1025         tx_desc = IXGBEVF_TX_DESC(ring, i);
1026         tx_desc->read.buffer_addr = cpu_to_le64(dma);
1027
1028         tx_desc->read.cmd_type_len = cpu_to_le32(cmd_type);
1029         tx_desc->read.olinfo_status =
1030                         cpu_to_le32((len << IXGBE_ADVTXD_PAYLEN_SHIFT) |
1031                                     IXGBE_ADVTXD_CC);
1032
1033         /* Avoid any potential race with cleanup */
1034         smp_wmb();
1035
1036         /* set next_to_watch value indicating a packet is present */
1037         i++;
1038         if (i == ring->count)
1039                 i = 0;
1040
1041         tx_buffer->next_to_watch = tx_desc;
1042         ring->next_to_use = i;
1043
1044         return IXGBEVF_XDP_TX;
1045 }
1046
1047 static struct sk_buff *ixgbevf_run_xdp(struct ixgbevf_adapter *adapter,
1048                                        struct ixgbevf_ring  *rx_ring,
1049                                        struct xdp_buff *xdp)
1050 {
1051         int result = IXGBEVF_XDP_PASS;
1052         struct ixgbevf_ring *xdp_ring;
1053         struct bpf_prog *xdp_prog;
1054         u32 act;
1055
1056         xdp_prog = READ_ONCE(rx_ring->xdp_prog);
1057
1058         if (!xdp_prog)
1059                 goto xdp_out;
1060
1061         act = bpf_prog_run_xdp(xdp_prog, xdp);
1062         switch (act) {
1063         case XDP_PASS:
1064                 break;
1065         case XDP_TX:
1066                 xdp_ring = adapter->xdp_ring[rx_ring->queue_index];
1067                 result = ixgbevf_xmit_xdp_ring(xdp_ring, xdp);
1068                 if (result == IXGBEVF_XDP_CONSUMED)
1069                         goto out_failure;
1070                 break;
1071         default:
1072                 bpf_warn_invalid_xdp_action(rx_ring->netdev, xdp_prog, act);
1073                 fallthrough;
1074         case XDP_ABORTED:
1075 out_failure:
1076                 trace_xdp_exception(rx_ring->netdev, xdp_prog, act);
1077                 fallthrough; /* handle aborts by dropping packet */
1078         case XDP_DROP:
1079                 result = IXGBEVF_XDP_CONSUMED;
1080                 break;
1081         }
1082 xdp_out:
1083         return ERR_PTR(-result);
1084 }
1085
1086 static unsigned int ixgbevf_rx_frame_truesize(struct ixgbevf_ring *rx_ring,
1087                                               unsigned int size)
1088 {
1089         unsigned int truesize;
1090
1091 #if (PAGE_SIZE < 8192)
1092         truesize = ixgbevf_rx_pg_size(rx_ring) / 2; /* Must be power-of-2 */
1093 #else
1094         truesize = ring_uses_build_skb(rx_ring) ?
1095                 SKB_DATA_ALIGN(IXGBEVF_SKB_PAD + size) +
1096                 SKB_DATA_ALIGN(sizeof(struct skb_shared_info)) :
1097                 SKB_DATA_ALIGN(size);
1098 #endif
1099         return truesize;
1100 }
1101
1102 static void ixgbevf_rx_buffer_flip(struct ixgbevf_ring *rx_ring,
1103                                    struct ixgbevf_rx_buffer *rx_buffer,
1104                                    unsigned int size)
1105 {
1106         unsigned int truesize = ixgbevf_rx_frame_truesize(rx_ring, size);
1107
1108 #if (PAGE_SIZE < 8192)
1109         rx_buffer->page_offset ^= truesize;
1110 #else
1111         rx_buffer->page_offset += truesize;
1112 #endif
1113 }
1114
1115 static int ixgbevf_clean_rx_irq(struct ixgbevf_q_vector *q_vector,
1116                                 struct ixgbevf_ring *rx_ring,
1117                                 int budget)
1118 {
1119         unsigned int total_rx_bytes = 0, total_rx_packets = 0, frame_sz = 0;
1120         struct ixgbevf_adapter *adapter = q_vector->adapter;
1121         u16 cleaned_count = ixgbevf_desc_unused(rx_ring);
1122         struct sk_buff *skb = rx_ring->skb;
1123         bool xdp_xmit = false;
1124         struct xdp_buff xdp;
1125
1126         /* Frame size depend on rx_ring setup when PAGE_SIZE=4K */
1127 #if (PAGE_SIZE < 8192)
1128         frame_sz = ixgbevf_rx_frame_truesize(rx_ring, 0);
1129 #endif
1130         xdp_init_buff(&xdp, frame_sz, &rx_ring->xdp_rxq);
1131
1132         while (likely(total_rx_packets < budget)) {
1133                 struct ixgbevf_rx_buffer *rx_buffer;
1134                 union ixgbe_adv_rx_desc *rx_desc;
1135                 unsigned int size;
1136
1137                 /* return some buffers to hardware, one at a time is too slow */
1138                 if (cleaned_count >= IXGBEVF_RX_BUFFER_WRITE) {
1139                         ixgbevf_alloc_rx_buffers(rx_ring, cleaned_count);
1140                         cleaned_count = 0;
1141                 }
1142
1143                 rx_desc = IXGBEVF_RX_DESC(rx_ring, rx_ring->next_to_clean);
1144                 size = le16_to_cpu(rx_desc->wb.upper.length);
1145                 if (!size)
1146                         break;
1147
1148                 /* This memory barrier is needed to keep us from reading
1149                  * any other fields out of the rx_desc until we know the
1150                  * RXD_STAT_DD bit is set
1151                  */
1152                 rmb();
1153
1154                 rx_buffer = ixgbevf_get_rx_buffer(rx_ring, size);
1155
1156                 /* retrieve a buffer from the ring */
1157                 if (!skb) {
1158                         unsigned int offset = ixgbevf_rx_offset(rx_ring);
1159                         unsigned char *hard_start;
1160
1161                         hard_start = page_address(rx_buffer->page) +
1162                                      rx_buffer->page_offset - offset;
1163                         xdp_prepare_buff(&xdp, hard_start, offset, size, true);
1164 #if (PAGE_SIZE > 4096)
1165                         /* At larger PAGE_SIZE, frame_sz depend on len size */
1166                         xdp.frame_sz = ixgbevf_rx_frame_truesize(rx_ring, size);
1167 #endif
1168                         skb = ixgbevf_run_xdp(adapter, rx_ring, &xdp);
1169                 }
1170
1171                 if (IS_ERR(skb)) {
1172                         if (PTR_ERR(skb) == -IXGBEVF_XDP_TX) {
1173                                 xdp_xmit = true;
1174                                 ixgbevf_rx_buffer_flip(rx_ring, rx_buffer,
1175                                                        size);
1176                         } else {
1177                                 rx_buffer->pagecnt_bias++;
1178                         }
1179                         total_rx_packets++;
1180                         total_rx_bytes += size;
1181                 } else if (skb) {
1182                         ixgbevf_add_rx_frag(rx_ring, rx_buffer, skb, size);
1183                 } else if (ring_uses_build_skb(rx_ring)) {
1184                         skb = ixgbevf_build_skb(rx_ring, rx_buffer,
1185                                                 &xdp, rx_desc);
1186                 } else {
1187                         skb = ixgbevf_construct_skb(rx_ring, rx_buffer,
1188                                                     &xdp, rx_desc);
1189                 }
1190
1191                 /* exit if we failed to retrieve a buffer */
1192                 if (!skb) {
1193                         rx_ring->rx_stats.alloc_rx_buff_failed++;
1194                         rx_buffer->pagecnt_bias++;
1195                         break;
1196                 }
1197
1198                 ixgbevf_put_rx_buffer(rx_ring, rx_buffer, skb);
1199                 cleaned_count++;
1200
1201                 /* fetch next buffer in frame if non-eop */
1202                 if (ixgbevf_is_non_eop(rx_ring, rx_desc))
1203                         continue;
1204
1205                 /* verify the packet layout is correct */
1206                 if (ixgbevf_cleanup_headers(rx_ring, rx_desc, skb)) {
1207                         skb = NULL;
1208                         continue;
1209                 }
1210
1211                 /* probably a little skewed due to removing CRC */
1212                 total_rx_bytes += skb->len;
1213
1214                 /* Workaround hardware that can't do proper VEPA multicast
1215                  * source pruning.
1216                  */
1217                 if ((skb->pkt_type == PACKET_BROADCAST ||
1218                      skb->pkt_type == PACKET_MULTICAST) &&
1219                     ether_addr_equal(rx_ring->netdev->dev_addr,
1220                                      eth_hdr(skb)->h_source)) {
1221                         dev_kfree_skb_irq(skb);
1222                         continue;
1223                 }
1224
1225                 /* populate checksum, VLAN, and protocol */
1226                 ixgbevf_process_skb_fields(rx_ring, rx_desc, skb);
1227
1228                 ixgbevf_rx_skb(q_vector, skb);
1229
1230                 /* reset skb pointer */
1231                 skb = NULL;
1232
1233                 /* update budget accounting */
1234                 total_rx_packets++;
1235         }
1236
1237         /* place incomplete frames back on ring for completion */
1238         rx_ring->skb = skb;
1239
1240         if (xdp_xmit) {
1241                 struct ixgbevf_ring *xdp_ring =
1242                         adapter->xdp_ring[rx_ring->queue_index];
1243
1244                 /* Force memory writes to complete before letting h/w
1245                  * know there are new descriptors to fetch.
1246                  */
1247                 wmb();
1248                 ixgbevf_write_tail(xdp_ring, xdp_ring->next_to_use);
1249         }
1250
1251         u64_stats_update_begin(&rx_ring->syncp);
1252         rx_ring->stats.packets += total_rx_packets;
1253         rx_ring->stats.bytes += total_rx_bytes;
1254         u64_stats_update_end(&rx_ring->syncp);
1255         q_vector->rx.total_packets += total_rx_packets;
1256         q_vector->rx.total_bytes += total_rx_bytes;
1257
1258         return total_rx_packets;
1259 }
1260
1261 /**
1262  * ixgbevf_poll - NAPI polling calback
1263  * @napi: napi struct with our devices info in it
1264  * @budget: amount of work driver is allowed to do this pass, in packets
1265  *
1266  * This function will clean more than one or more rings associated with a
1267  * q_vector.
1268  **/
1269 static int ixgbevf_poll(struct napi_struct *napi, int budget)
1270 {
1271         struct ixgbevf_q_vector *q_vector =
1272                 container_of(napi, struct ixgbevf_q_vector, napi);
1273         struct ixgbevf_adapter *adapter = q_vector->adapter;
1274         struct ixgbevf_ring *ring;
1275         int per_ring_budget, work_done = 0;
1276         bool clean_complete = true;
1277
1278         ixgbevf_for_each_ring(ring, q_vector->tx) {
1279                 if (!ixgbevf_clean_tx_irq(q_vector, ring, budget))
1280                         clean_complete = false;
1281         }
1282
1283         if (budget <= 0)
1284                 return budget;
1285
1286         /* attempt to distribute budget to each queue fairly, but don't allow
1287          * the budget to go below 1 because we'll exit polling
1288          */
1289         if (q_vector->rx.count > 1)
1290                 per_ring_budget = max(budget/q_vector->rx.count, 1);
1291         else
1292                 per_ring_budget = budget;
1293
1294         ixgbevf_for_each_ring(ring, q_vector->rx) {
1295                 int cleaned = ixgbevf_clean_rx_irq(q_vector, ring,
1296                                                    per_ring_budget);
1297                 work_done += cleaned;
1298                 if (cleaned >= per_ring_budget)
1299                         clean_complete = false;
1300         }
1301
1302         /* If all work not completed, return budget and keep polling */
1303         if (!clean_complete)
1304                 return budget;
1305
1306         /* Exit the polling mode, but don't re-enable interrupts if stack might
1307          * poll us due to busy-polling
1308          */
1309         if (likely(napi_complete_done(napi, work_done))) {
1310                 if (adapter->rx_itr_setting == 1)
1311                         ixgbevf_set_itr(q_vector);
1312                 if (!test_bit(__IXGBEVF_DOWN, &adapter->state) &&
1313                     !test_bit(__IXGBEVF_REMOVING, &adapter->state))
1314                         ixgbevf_irq_enable_queues(adapter,
1315                                                   BIT(q_vector->v_idx));
1316         }
1317
1318         return min(work_done, budget - 1);
1319 }
1320
1321 /**
1322  * ixgbevf_write_eitr - write VTEITR register in hardware specific way
1323  * @q_vector: structure containing interrupt and ring information
1324  **/
1325 void ixgbevf_write_eitr(struct ixgbevf_q_vector *q_vector)
1326 {
1327         struct ixgbevf_adapter *adapter = q_vector->adapter;
1328         struct ixgbe_hw *hw = &adapter->hw;
1329         int v_idx = q_vector->v_idx;
1330         u32 itr_reg = q_vector->itr & IXGBE_MAX_EITR;
1331
1332         /* set the WDIS bit to not clear the timer bits and cause an
1333          * immediate assertion of the interrupt
1334          */
1335         itr_reg |= IXGBE_EITR_CNT_WDIS;
1336
1337         IXGBE_WRITE_REG(hw, IXGBE_VTEITR(v_idx), itr_reg);
1338 }
1339
1340 /**
1341  * ixgbevf_configure_msix - Configure MSI-X hardware
1342  * @adapter: board private structure
1343  *
1344  * ixgbevf_configure_msix sets up the hardware to properly generate MSI-X
1345  * interrupts.
1346  **/
1347 static void ixgbevf_configure_msix(struct ixgbevf_adapter *adapter)
1348 {
1349         struct ixgbevf_q_vector *q_vector;
1350         int q_vectors, v_idx;
1351
1352         q_vectors = adapter->num_msix_vectors - NON_Q_VECTORS;
1353         adapter->eims_enable_mask = 0;
1354
1355         /* Populate the IVAR table and set the ITR values to the
1356          * corresponding register.
1357          */
1358         for (v_idx = 0; v_idx < q_vectors; v_idx++) {
1359                 struct ixgbevf_ring *ring;
1360
1361                 q_vector = adapter->q_vector[v_idx];
1362
1363                 ixgbevf_for_each_ring(ring, q_vector->rx)
1364                         ixgbevf_set_ivar(adapter, 0, ring->reg_idx, v_idx);
1365
1366                 ixgbevf_for_each_ring(ring, q_vector->tx)
1367                         ixgbevf_set_ivar(adapter, 1, ring->reg_idx, v_idx);
1368
1369                 if (q_vector->tx.ring && !q_vector->rx.ring) {
1370                         /* Tx only vector */
1371                         if (adapter->tx_itr_setting == 1)
1372                                 q_vector->itr = IXGBE_12K_ITR;
1373                         else
1374                                 q_vector->itr = adapter->tx_itr_setting;
1375                 } else {
1376                         /* Rx or Rx/Tx vector */
1377                         if (adapter->rx_itr_setting == 1)
1378                                 q_vector->itr = IXGBE_20K_ITR;
1379                         else
1380                                 q_vector->itr = adapter->rx_itr_setting;
1381                 }
1382
1383                 /* add q_vector eims value to global eims_enable_mask */
1384                 adapter->eims_enable_mask |= BIT(v_idx);
1385
1386                 ixgbevf_write_eitr(q_vector);
1387         }
1388
1389         ixgbevf_set_ivar(adapter, -1, 1, v_idx);
1390         /* setup eims_other and add value to global eims_enable_mask */
1391         adapter->eims_other = BIT(v_idx);
1392         adapter->eims_enable_mask |= adapter->eims_other;
1393 }
1394
1395 enum latency_range {
1396         lowest_latency = 0,
1397         low_latency = 1,
1398         bulk_latency = 2,
1399         latency_invalid = 255
1400 };
1401
1402 /**
1403  * ixgbevf_update_itr - update the dynamic ITR value based on statistics
1404  * @q_vector: structure containing interrupt and ring information
1405  * @ring_container: structure containing ring performance data
1406  *
1407  * Stores a new ITR value based on packets and byte
1408  * counts during the last interrupt.  The advantage of per interrupt
1409  * computation is faster updates and more accurate ITR for the current
1410  * traffic pattern.  Constants in this function were computed
1411  * based on theoretical maximum wire speed and thresholds were set based
1412  * on testing data as well as attempting to minimize response time
1413  * while increasing bulk throughput.
1414  **/
1415 static void ixgbevf_update_itr(struct ixgbevf_q_vector *q_vector,
1416                                struct ixgbevf_ring_container *ring_container)
1417 {
1418         int bytes = ring_container->total_bytes;
1419         int packets = ring_container->total_packets;
1420         u32 timepassed_us;
1421         u64 bytes_perint;
1422         u8 itr_setting = ring_container->itr;
1423
1424         if (packets == 0)
1425                 return;
1426
1427         /* simple throttle rate management
1428          *    0-20MB/s lowest (100000 ints/s)
1429          *   20-100MB/s low   (20000 ints/s)
1430          *  100-1249MB/s bulk (12000 ints/s)
1431          */
1432         /* what was last interrupt timeslice? */
1433         timepassed_us = q_vector->itr >> 2;
1434         if (timepassed_us == 0)
1435                 return;
1436
1437         bytes_perint = bytes / timepassed_us; /* bytes/usec */
1438
1439         switch (itr_setting) {
1440         case lowest_latency:
1441                 if (bytes_perint > 10)
1442                         itr_setting = low_latency;
1443                 break;
1444         case low_latency:
1445                 if (bytes_perint > 20)
1446                         itr_setting = bulk_latency;
1447                 else if (bytes_perint <= 10)
1448                         itr_setting = lowest_latency;
1449                 break;
1450         case bulk_latency:
1451                 if (bytes_perint <= 20)
1452                         itr_setting = low_latency;
1453                 break;
1454         }
1455
1456         /* clear work counters since we have the values we need */
1457         ring_container->total_bytes = 0;
1458         ring_container->total_packets = 0;
1459
1460         /* write updated itr to ring container */
1461         ring_container->itr = itr_setting;
1462 }
1463
1464 static void ixgbevf_set_itr(struct ixgbevf_q_vector *q_vector)
1465 {
1466         u32 new_itr = q_vector->itr;
1467         u8 current_itr;
1468
1469         ixgbevf_update_itr(q_vector, &q_vector->tx);
1470         ixgbevf_update_itr(q_vector, &q_vector->rx);
1471
1472         current_itr = max(q_vector->rx.itr, q_vector->tx.itr);
1473
1474         switch (current_itr) {
1475         /* counts and packets in update_itr are dependent on these numbers */
1476         case lowest_latency:
1477                 new_itr = IXGBE_100K_ITR;
1478                 break;
1479         case low_latency:
1480                 new_itr = IXGBE_20K_ITR;
1481                 break;
1482         case bulk_latency:
1483                 new_itr = IXGBE_12K_ITR;
1484                 break;
1485         default:
1486                 break;
1487         }
1488
1489         if (new_itr != q_vector->itr) {
1490                 /* do an exponential smoothing */
1491                 new_itr = (10 * new_itr * q_vector->itr) /
1492                           ((9 * new_itr) + q_vector->itr);
1493
1494                 /* save the algorithm value here */
1495                 q_vector->itr = new_itr;
1496
1497                 ixgbevf_write_eitr(q_vector);
1498         }
1499 }
1500
1501 static irqreturn_t ixgbevf_msix_other(int irq, void *data)
1502 {
1503         struct ixgbevf_adapter *adapter = data;
1504         struct ixgbe_hw *hw = &adapter->hw;
1505
1506         hw->mac.get_link_status = 1;
1507
1508         ixgbevf_service_event_schedule(adapter);
1509
1510         IXGBE_WRITE_REG(hw, IXGBE_VTEIMS, adapter->eims_other);
1511
1512         return IRQ_HANDLED;
1513 }
1514
1515 /**
1516  * ixgbevf_msix_clean_rings - single unshared vector rx clean (all queues)
1517  * @irq: unused
1518  * @data: pointer to our q_vector struct for this interrupt vector
1519  **/
1520 static irqreturn_t ixgbevf_msix_clean_rings(int irq, void *data)
1521 {
1522         struct ixgbevf_q_vector *q_vector = data;
1523
1524         /* EIAM disabled interrupts (on this vector) for us */
1525         if (q_vector->rx.ring || q_vector->tx.ring)
1526                 napi_schedule_irqoff(&q_vector->napi);
1527
1528         return IRQ_HANDLED;
1529 }
1530
1531 /**
1532  * ixgbevf_request_msix_irqs - Initialize MSI-X interrupts
1533  * @adapter: board private structure
1534  *
1535  * ixgbevf_request_msix_irqs allocates MSI-X vectors and requests
1536  * interrupts from the kernel.
1537  **/
1538 static int ixgbevf_request_msix_irqs(struct ixgbevf_adapter *adapter)
1539 {
1540         struct net_device *netdev = adapter->netdev;
1541         int q_vectors = adapter->num_msix_vectors - NON_Q_VECTORS;
1542         unsigned int ri = 0, ti = 0;
1543         int vector, err;
1544
1545         for (vector = 0; vector < q_vectors; vector++) {
1546                 struct ixgbevf_q_vector *q_vector = adapter->q_vector[vector];
1547                 struct msix_entry *entry = &adapter->msix_entries[vector];
1548
1549                 if (q_vector->tx.ring && q_vector->rx.ring) {
1550                         snprintf(q_vector->name, sizeof(q_vector->name),
1551                                  "%s-TxRx-%u", netdev->name, ri++);
1552                         ti++;
1553                 } else if (q_vector->rx.ring) {
1554                         snprintf(q_vector->name, sizeof(q_vector->name),
1555                                  "%s-rx-%u", netdev->name, ri++);
1556                 } else if (q_vector->tx.ring) {
1557                         snprintf(q_vector->name, sizeof(q_vector->name),
1558                                  "%s-tx-%u", netdev->name, ti++);
1559                 } else {
1560                         /* skip this unused q_vector */
1561                         continue;
1562                 }
1563                 err = request_irq(entry->vector, &ixgbevf_msix_clean_rings, 0,
1564                                   q_vector->name, q_vector);
1565                 if (err) {
1566                         hw_dbg(&adapter->hw,
1567                                "request_irq failed for MSIX interrupt Error: %d\n",
1568                                err);
1569                         goto free_queue_irqs;
1570                 }
1571         }
1572
1573         err = request_irq(adapter->msix_entries[vector].vector,
1574                           &ixgbevf_msix_other, 0, netdev->name, adapter);
1575         if (err) {
1576                 hw_dbg(&adapter->hw, "request_irq for msix_other failed: %d\n",
1577                        err);
1578                 goto free_queue_irqs;
1579         }
1580
1581         return 0;
1582
1583 free_queue_irqs:
1584         while (vector) {
1585                 vector--;
1586                 free_irq(adapter->msix_entries[vector].vector,
1587                          adapter->q_vector[vector]);
1588         }
1589         /* This failure is non-recoverable - it indicates the system is
1590          * out of MSIX vector resources and the VF driver cannot run
1591          * without them.  Set the number of msix vectors to zero
1592          * indicating that not enough can be allocated.  The error
1593          * will be returned to the user indicating device open failed.
1594          * Any further attempts to force the driver to open will also
1595          * fail.  The only way to recover is to unload the driver and
1596          * reload it again.  If the system has recovered some MSIX
1597          * vectors then it may succeed.
1598          */
1599         adapter->num_msix_vectors = 0;
1600         return err;
1601 }
1602
1603 /**
1604  * ixgbevf_request_irq - initialize interrupts
1605  * @adapter: board private structure
1606  *
1607  * Attempts to configure interrupts using the best available
1608  * capabilities of the hardware and kernel.
1609  **/
1610 static int ixgbevf_request_irq(struct ixgbevf_adapter *adapter)
1611 {
1612         int err = ixgbevf_request_msix_irqs(adapter);
1613
1614         if (err)
1615                 hw_dbg(&adapter->hw, "request_irq failed, Error %d\n", err);
1616
1617         return err;
1618 }
1619
1620 static void ixgbevf_free_irq(struct ixgbevf_adapter *adapter)
1621 {
1622         int i, q_vectors;
1623
1624         if (!adapter->msix_entries)
1625                 return;
1626
1627         q_vectors = adapter->num_msix_vectors;
1628         i = q_vectors - 1;
1629
1630         free_irq(adapter->msix_entries[i].vector, adapter);
1631         i--;
1632
1633         for (; i >= 0; i--) {
1634                 /* free only the irqs that were actually requested */
1635                 if (!adapter->q_vector[i]->rx.ring &&
1636                     !adapter->q_vector[i]->tx.ring)
1637                         continue;
1638
1639                 free_irq(adapter->msix_entries[i].vector,
1640                          adapter->q_vector[i]);
1641         }
1642 }
1643
1644 /**
1645  * ixgbevf_irq_disable - Mask off interrupt generation on the NIC
1646  * @adapter: board private structure
1647  **/
1648 static inline void ixgbevf_irq_disable(struct ixgbevf_adapter *adapter)
1649 {
1650         struct ixgbe_hw *hw = &adapter->hw;
1651         int i;
1652
1653         IXGBE_WRITE_REG(hw, IXGBE_VTEIAM, 0);
1654         IXGBE_WRITE_REG(hw, IXGBE_VTEIMC, ~0);
1655         IXGBE_WRITE_REG(hw, IXGBE_VTEIAC, 0);
1656
1657         IXGBE_WRITE_FLUSH(hw);
1658
1659         for (i = 0; i < adapter->num_msix_vectors; i++)
1660                 synchronize_irq(adapter->msix_entries[i].vector);
1661 }
1662
1663 /**
1664  * ixgbevf_irq_enable - Enable default interrupt generation settings
1665  * @adapter: board private structure
1666  **/
1667 static inline void ixgbevf_irq_enable(struct ixgbevf_adapter *adapter)
1668 {
1669         struct ixgbe_hw *hw = &adapter->hw;
1670
1671         IXGBE_WRITE_REG(hw, IXGBE_VTEIAM, adapter->eims_enable_mask);
1672         IXGBE_WRITE_REG(hw, IXGBE_VTEIAC, adapter->eims_enable_mask);
1673         IXGBE_WRITE_REG(hw, IXGBE_VTEIMS, adapter->eims_enable_mask);
1674 }
1675
1676 /**
1677  * ixgbevf_configure_tx_ring - Configure 82599 VF Tx ring after Reset
1678  * @adapter: board private structure
1679  * @ring: structure containing ring specific data
1680  *
1681  * Configure the Tx descriptor ring after a reset.
1682  **/
1683 static void ixgbevf_configure_tx_ring(struct ixgbevf_adapter *adapter,
1684                                       struct ixgbevf_ring *ring)
1685 {
1686         struct ixgbe_hw *hw = &adapter->hw;
1687         u64 tdba = ring->dma;
1688         int wait_loop = 10;
1689         u32 txdctl = IXGBE_TXDCTL_ENABLE;
1690         u8 reg_idx = ring->reg_idx;
1691
1692         /* disable queue to avoid issues while updating state */
1693         IXGBE_WRITE_REG(hw, IXGBE_VFTXDCTL(reg_idx), IXGBE_TXDCTL_SWFLSH);
1694         IXGBE_WRITE_FLUSH(hw);
1695
1696         IXGBE_WRITE_REG(hw, IXGBE_VFTDBAL(reg_idx), tdba & DMA_BIT_MASK(32));
1697         IXGBE_WRITE_REG(hw, IXGBE_VFTDBAH(reg_idx), tdba >> 32);
1698         IXGBE_WRITE_REG(hw, IXGBE_VFTDLEN(reg_idx),
1699                         ring->count * sizeof(union ixgbe_adv_tx_desc));
1700
1701         /* disable head writeback */
1702         IXGBE_WRITE_REG(hw, IXGBE_VFTDWBAH(reg_idx), 0);
1703         IXGBE_WRITE_REG(hw, IXGBE_VFTDWBAL(reg_idx), 0);
1704
1705         /* enable relaxed ordering */
1706         IXGBE_WRITE_REG(hw, IXGBE_VFDCA_TXCTRL(reg_idx),
1707                         (IXGBE_DCA_TXCTRL_DESC_RRO_EN |
1708                          IXGBE_DCA_TXCTRL_DATA_RRO_EN));
1709
1710         /* reset head and tail pointers */
1711         IXGBE_WRITE_REG(hw, IXGBE_VFTDH(reg_idx), 0);
1712         IXGBE_WRITE_REG(hw, IXGBE_VFTDT(reg_idx), 0);
1713         ring->tail = adapter->io_addr + IXGBE_VFTDT(reg_idx);
1714
1715         /* reset ntu and ntc to place SW in sync with hardwdare */
1716         ring->next_to_clean = 0;
1717         ring->next_to_use = 0;
1718
1719         /* In order to avoid issues WTHRESH + PTHRESH should always be equal
1720          * to or less than the number of on chip descriptors, which is
1721          * currently 40.
1722          */
1723         txdctl |= (8 << 16);    /* WTHRESH = 8 */
1724
1725         /* Setting PTHRESH to 32 both improves performance */
1726         txdctl |= (1u << 8) |    /* HTHRESH = 1 */
1727                    32;           /* PTHRESH = 32 */
1728
1729         /* reinitialize tx_buffer_info */
1730         memset(ring->tx_buffer_info, 0,
1731                sizeof(struct ixgbevf_tx_buffer) * ring->count);
1732
1733         clear_bit(__IXGBEVF_HANG_CHECK_ARMED, &ring->state);
1734         clear_bit(__IXGBEVF_TX_XDP_RING_PRIMED, &ring->state);
1735
1736         IXGBE_WRITE_REG(hw, IXGBE_VFTXDCTL(reg_idx), txdctl);
1737
1738         /* poll to verify queue is enabled */
1739         do {
1740                 usleep_range(1000, 2000);
1741                 txdctl = IXGBE_READ_REG(hw, IXGBE_VFTXDCTL(reg_idx));
1742         }  while (--wait_loop && !(txdctl & IXGBE_TXDCTL_ENABLE));
1743         if (!wait_loop)
1744                 hw_dbg(hw, "Could not enable Tx Queue %d\n", reg_idx);
1745 }
1746
1747 /**
1748  * ixgbevf_configure_tx - Configure 82599 VF Transmit Unit after Reset
1749  * @adapter: board private structure
1750  *
1751  * Configure the Tx unit of the MAC after a reset.
1752  **/
1753 static void ixgbevf_configure_tx(struct ixgbevf_adapter *adapter)
1754 {
1755         u32 i;
1756
1757         /* Setup the HW Tx Head and Tail descriptor pointers */
1758         for (i = 0; i < adapter->num_tx_queues; i++)
1759                 ixgbevf_configure_tx_ring(adapter, adapter->tx_ring[i]);
1760         for (i = 0; i < adapter->num_xdp_queues; i++)
1761                 ixgbevf_configure_tx_ring(adapter, adapter->xdp_ring[i]);
1762 }
1763
1764 #define IXGBE_SRRCTL_BSIZEHDRSIZE_SHIFT 2
1765
1766 static void ixgbevf_configure_srrctl(struct ixgbevf_adapter *adapter,
1767                                      struct ixgbevf_ring *ring, int index)
1768 {
1769         struct ixgbe_hw *hw = &adapter->hw;
1770         u32 srrctl;
1771
1772         srrctl = IXGBE_SRRCTL_DROP_EN;
1773
1774         srrctl |= IXGBEVF_RX_HDR_SIZE << IXGBE_SRRCTL_BSIZEHDRSIZE_SHIFT;
1775         if (ring_uses_large_buffer(ring))
1776                 srrctl |= IXGBEVF_RXBUFFER_3072 >> IXGBE_SRRCTL_BSIZEPKT_SHIFT;
1777         else
1778                 srrctl |= IXGBEVF_RXBUFFER_2048 >> IXGBE_SRRCTL_BSIZEPKT_SHIFT;
1779         srrctl |= IXGBE_SRRCTL_DESCTYPE_ADV_ONEBUF;
1780
1781         IXGBE_WRITE_REG(hw, IXGBE_VFSRRCTL(index), srrctl);
1782 }
1783
1784 static void ixgbevf_setup_psrtype(struct ixgbevf_adapter *adapter)
1785 {
1786         struct ixgbe_hw *hw = &adapter->hw;
1787
1788         /* PSRTYPE must be initialized in 82599 */
1789         u32 psrtype = IXGBE_PSRTYPE_TCPHDR | IXGBE_PSRTYPE_UDPHDR |
1790                       IXGBE_PSRTYPE_IPV4HDR | IXGBE_PSRTYPE_IPV6HDR |
1791                       IXGBE_PSRTYPE_L2HDR;
1792
1793         if (adapter->num_rx_queues > 1)
1794                 psrtype |= BIT(29);
1795
1796         IXGBE_WRITE_REG(hw, IXGBE_VFPSRTYPE, psrtype);
1797 }
1798
1799 #define IXGBEVF_MAX_RX_DESC_POLL 10
1800 static void ixgbevf_disable_rx_queue(struct ixgbevf_adapter *adapter,
1801                                      struct ixgbevf_ring *ring)
1802 {
1803         struct ixgbe_hw *hw = &adapter->hw;
1804         int wait_loop = IXGBEVF_MAX_RX_DESC_POLL;
1805         u32 rxdctl;
1806         u8 reg_idx = ring->reg_idx;
1807
1808         if (IXGBE_REMOVED(hw->hw_addr))
1809                 return;
1810         rxdctl = IXGBE_READ_REG(hw, IXGBE_VFRXDCTL(reg_idx));
1811         rxdctl &= ~IXGBE_RXDCTL_ENABLE;
1812
1813         /* write value back with RXDCTL.ENABLE bit cleared */
1814         IXGBE_WRITE_REG(hw, IXGBE_VFRXDCTL(reg_idx), rxdctl);
1815
1816         /* the hardware may take up to 100us to really disable the Rx queue */
1817         do {
1818                 udelay(10);
1819                 rxdctl = IXGBE_READ_REG(hw, IXGBE_VFRXDCTL(reg_idx));
1820         } while (--wait_loop && (rxdctl & IXGBE_RXDCTL_ENABLE));
1821
1822         if (!wait_loop)
1823                 pr_err("RXDCTL.ENABLE queue %d not cleared while polling\n",
1824                        reg_idx);
1825 }
1826
1827 static void ixgbevf_rx_desc_queue_enable(struct ixgbevf_adapter *adapter,
1828                                          struct ixgbevf_ring *ring)
1829 {
1830         struct ixgbe_hw *hw = &adapter->hw;
1831         int wait_loop = IXGBEVF_MAX_RX_DESC_POLL;
1832         u32 rxdctl;
1833         u8 reg_idx = ring->reg_idx;
1834
1835         if (IXGBE_REMOVED(hw->hw_addr))
1836                 return;
1837         do {
1838                 usleep_range(1000, 2000);
1839                 rxdctl = IXGBE_READ_REG(hw, IXGBE_VFRXDCTL(reg_idx));
1840         } while (--wait_loop && !(rxdctl & IXGBE_RXDCTL_ENABLE));
1841
1842         if (!wait_loop)
1843                 pr_err("RXDCTL.ENABLE queue %d not set while polling\n",
1844                        reg_idx);
1845 }
1846
1847 /**
1848  * ixgbevf_init_rss_key - Initialize adapter RSS key
1849  * @adapter: device handle
1850  *
1851  * Allocates and initializes the RSS key if it is not allocated.
1852  **/
1853 static inline int ixgbevf_init_rss_key(struct ixgbevf_adapter *adapter)
1854 {
1855         u32 *rss_key;
1856
1857         if (!adapter->rss_key) {
1858                 rss_key = kzalloc(IXGBEVF_RSS_HASH_KEY_SIZE, GFP_KERNEL);
1859                 if (unlikely(!rss_key))
1860                         return -ENOMEM;
1861
1862                 netdev_rss_key_fill(rss_key, IXGBEVF_RSS_HASH_KEY_SIZE);
1863                 adapter->rss_key = rss_key;
1864         }
1865
1866         return 0;
1867 }
1868
1869 static void ixgbevf_setup_vfmrqc(struct ixgbevf_adapter *adapter)
1870 {
1871         struct ixgbe_hw *hw = &adapter->hw;
1872         u32 vfmrqc = 0, vfreta = 0;
1873         u16 rss_i = adapter->num_rx_queues;
1874         u8 i, j;
1875
1876         /* Fill out hash function seeds */
1877         for (i = 0; i < IXGBEVF_VFRSSRK_REGS; i++)
1878                 IXGBE_WRITE_REG(hw, IXGBE_VFRSSRK(i), *(adapter->rss_key + i));
1879
1880         for (i = 0, j = 0; i < IXGBEVF_X550_VFRETA_SIZE; i++, j++) {
1881                 if (j == rss_i)
1882                         j = 0;
1883
1884                 adapter->rss_indir_tbl[i] = j;
1885
1886                 vfreta |= j << (i & 0x3) * 8;
1887                 if ((i & 3) == 3) {
1888                         IXGBE_WRITE_REG(hw, IXGBE_VFRETA(i >> 2), vfreta);
1889                         vfreta = 0;
1890                 }
1891         }
1892
1893         /* Perform hash on these packet types */
1894         vfmrqc |= IXGBE_VFMRQC_RSS_FIELD_IPV4 |
1895                 IXGBE_VFMRQC_RSS_FIELD_IPV4_TCP |
1896                 IXGBE_VFMRQC_RSS_FIELD_IPV6 |
1897                 IXGBE_VFMRQC_RSS_FIELD_IPV6_TCP;
1898
1899         vfmrqc |= IXGBE_VFMRQC_RSSEN;
1900
1901         IXGBE_WRITE_REG(hw, IXGBE_VFMRQC, vfmrqc);
1902 }
1903
1904 static void ixgbevf_configure_rx_ring(struct ixgbevf_adapter *adapter,
1905                                       struct ixgbevf_ring *ring)
1906 {
1907         struct ixgbe_hw *hw = &adapter->hw;
1908         union ixgbe_adv_rx_desc *rx_desc;
1909         u64 rdba = ring->dma;
1910         u32 rxdctl;
1911         u8 reg_idx = ring->reg_idx;
1912
1913         /* disable queue to avoid issues while updating state */
1914         rxdctl = IXGBE_READ_REG(hw, IXGBE_VFRXDCTL(reg_idx));
1915         ixgbevf_disable_rx_queue(adapter, ring);
1916
1917         IXGBE_WRITE_REG(hw, IXGBE_VFRDBAL(reg_idx), rdba & DMA_BIT_MASK(32));
1918         IXGBE_WRITE_REG(hw, IXGBE_VFRDBAH(reg_idx), rdba >> 32);
1919         IXGBE_WRITE_REG(hw, IXGBE_VFRDLEN(reg_idx),
1920                         ring->count * sizeof(union ixgbe_adv_rx_desc));
1921
1922 #ifndef CONFIG_SPARC
1923         /* enable relaxed ordering */
1924         IXGBE_WRITE_REG(hw, IXGBE_VFDCA_RXCTRL(reg_idx),
1925                         IXGBE_DCA_RXCTRL_DESC_RRO_EN);
1926 #else
1927         IXGBE_WRITE_REG(hw, IXGBE_VFDCA_RXCTRL(reg_idx),
1928                         IXGBE_DCA_RXCTRL_DESC_RRO_EN |
1929                         IXGBE_DCA_RXCTRL_DATA_WRO_EN);
1930 #endif
1931
1932         /* reset head and tail pointers */
1933         IXGBE_WRITE_REG(hw, IXGBE_VFRDH(reg_idx), 0);
1934         IXGBE_WRITE_REG(hw, IXGBE_VFRDT(reg_idx), 0);
1935         ring->tail = adapter->io_addr + IXGBE_VFRDT(reg_idx);
1936
1937         /* initialize rx_buffer_info */
1938         memset(ring->rx_buffer_info, 0,
1939                sizeof(struct ixgbevf_rx_buffer) * ring->count);
1940
1941         /* initialize Rx descriptor 0 */
1942         rx_desc = IXGBEVF_RX_DESC(ring, 0);
1943         rx_desc->wb.upper.length = 0;
1944
1945         /* reset ntu and ntc to place SW in sync with hardwdare */
1946         ring->next_to_clean = 0;
1947         ring->next_to_use = 0;
1948         ring->next_to_alloc = 0;
1949
1950         ixgbevf_configure_srrctl(adapter, ring, reg_idx);
1951
1952         /* RXDCTL.RLPML does not work on 82599 */
1953         if (adapter->hw.mac.type != ixgbe_mac_82599_vf) {
1954                 rxdctl &= ~(IXGBE_RXDCTL_RLPMLMASK |
1955                             IXGBE_RXDCTL_RLPML_EN);
1956
1957 #if (PAGE_SIZE < 8192)
1958                 /* Limit the maximum frame size so we don't overrun the skb */
1959                 if (ring_uses_build_skb(ring) &&
1960                     !ring_uses_large_buffer(ring))
1961                         rxdctl |= IXGBEVF_MAX_FRAME_BUILD_SKB |
1962                                   IXGBE_RXDCTL_RLPML_EN;
1963 #endif
1964         }
1965
1966         rxdctl |= IXGBE_RXDCTL_ENABLE | IXGBE_RXDCTL_VME;
1967         IXGBE_WRITE_REG(hw, IXGBE_VFRXDCTL(reg_idx), rxdctl);
1968
1969         ixgbevf_rx_desc_queue_enable(adapter, ring);
1970         ixgbevf_alloc_rx_buffers(ring, ixgbevf_desc_unused(ring));
1971 }
1972
1973 static void ixgbevf_set_rx_buffer_len(struct ixgbevf_adapter *adapter,
1974                                       struct ixgbevf_ring *rx_ring)
1975 {
1976         struct net_device *netdev = adapter->netdev;
1977         unsigned int max_frame = netdev->mtu + ETH_HLEN + ETH_FCS_LEN;
1978
1979         /* set build_skb and buffer size flags */
1980         clear_ring_build_skb_enabled(rx_ring);
1981         clear_ring_uses_large_buffer(rx_ring);
1982
1983         if (adapter->flags & IXGBEVF_FLAGS_LEGACY_RX)
1984                 return;
1985
1986         if (PAGE_SIZE < 8192)
1987                 if (max_frame > IXGBEVF_MAX_FRAME_BUILD_SKB)
1988                         set_ring_uses_large_buffer(rx_ring);
1989
1990         /* 82599 can't rely on RXDCTL.RLPML to restrict the size of the frame */
1991         if (adapter->hw.mac.type == ixgbe_mac_82599_vf && !ring_uses_large_buffer(rx_ring))
1992                 return;
1993
1994         set_ring_build_skb_enabled(rx_ring);
1995 }
1996
1997 /**
1998  * ixgbevf_configure_rx - Configure 82599 VF Receive Unit after Reset
1999  * @adapter: board private structure
2000  *
2001  * Configure the Rx unit of the MAC after a reset.
2002  **/
2003 static void ixgbevf_configure_rx(struct ixgbevf_adapter *adapter)
2004 {
2005         struct ixgbe_hw *hw = &adapter->hw;
2006         struct net_device *netdev = adapter->netdev;
2007         int i, ret;
2008
2009         ixgbevf_setup_psrtype(adapter);
2010         if (hw->mac.type >= ixgbe_mac_X550_vf)
2011                 ixgbevf_setup_vfmrqc(adapter);
2012
2013         spin_lock_bh(&adapter->mbx_lock);
2014         /* notify the PF of our intent to use this size of frame */
2015         ret = hw->mac.ops.set_rlpml(hw, netdev->mtu + ETH_HLEN + ETH_FCS_LEN);
2016         spin_unlock_bh(&adapter->mbx_lock);
2017         if (ret)
2018                 dev_err(&adapter->pdev->dev,
2019                         "Failed to set MTU at %d\n", netdev->mtu);
2020
2021         /* Setup the HW Rx Head and Tail Descriptor Pointers and
2022          * the Base and Length of the Rx Descriptor Ring
2023          */
2024         for (i = 0; i < adapter->num_rx_queues; i++) {
2025                 struct ixgbevf_ring *rx_ring = adapter->rx_ring[i];
2026
2027                 ixgbevf_set_rx_buffer_len(adapter, rx_ring);
2028                 ixgbevf_configure_rx_ring(adapter, rx_ring);
2029         }
2030 }
2031
2032 static int ixgbevf_vlan_rx_add_vid(struct net_device *netdev,
2033                                    __be16 proto, u16 vid)
2034 {
2035         struct ixgbevf_adapter *adapter = netdev_priv(netdev);
2036         struct ixgbe_hw *hw = &adapter->hw;
2037         int err;
2038
2039         spin_lock_bh(&adapter->mbx_lock);
2040
2041         /* add VID to filter table */
2042         err = hw->mac.ops.set_vfta(hw, vid, 0, true);
2043
2044         spin_unlock_bh(&adapter->mbx_lock);
2045
2046         if (err) {
2047                 netdev_err(netdev, "VF could not set VLAN %d\n", vid);
2048
2049                 /* translate error return types so error makes sense */
2050                 if (err == IXGBE_ERR_MBX)
2051                         return -EIO;
2052
2053                 if (err == IXGBE_ERR_INVALID_ARGUMENT)
2054                         return -EACCES;
2055         }
2056
2057         set_bit(vid, adapter->active_vlans);
2058
2059         return err;
2060 }
2061
2062 static int ixgbevf_vlan_rx_kill_vid(struct net_device *netdev,
2063                                     __be16 proto, u16 vid)
2064 {
2065         struct ixgbevf_adapter *adapter = netdev_priv(netdev);
2066         struct ixgbe_hw *hw = &adapter->hw;
2067         int err;
2068
2069         spin_lock_bh(&adapter->mbx_lock);
2070
2071         /* remove VID from filter table */
2072         err = hw->mac.ops.set_vfta(hw, vid, 0, false);
2073
2074         spin_unlock_bh(&adapter->mbx_lock);
2075
2076         if (err)
2077                 netdev_err(netdev, "Could not remove VLAN %d\n", vid);
2078
2079         clear_bit(vid, adapter->active_vlans);
2080
2081         return err;
2082 }
2083
2084 static void ixgbevf_restore_vlan(struct ixgbevf_adapter *adapter)
2085 {
2086         u16 vid;
2087
2088         for_each_set_bit(vid, adapter->active_vlans, VLAN_N_VID)
2089                 ixgbevf_vlan_rx_add_vid(adapter->netdev,
2090                                         htons(ETH_P_8021Q), vid);
2091 }
2092
2093 static int ixgbevf_write_uc_addr_list(struct net_device *netdev)
2094 {
2095         struct ixgbevf_adapter *adapter = netdev_priv(netdev);
2096         struct ixgbe_hw *hw = &adapter->hw;
2097         int count = 0;
2098
2099         if (!netdev_uc_empty(netdev)) {
2100                 struct netdev_hw_addr *ha;
2101
2102                 netdev_for_each_uc_addr(ha, netdev) {
2103                         hw->mac.ops.set_uc_addr(hw, ++count, ha->addr);
2104                         udelay(200);
2105                 }
2106         } else {
2107                 /* If the list is empty then send message to PF driver to
2108                  * clear all MAC VLANs on this VF.
2109                  */
2110                 hw->mac.ops.set_uc_addr(hw, 0, NULL);
2111         }
2112
2113         return count;
2114 }
2115
2116 /**
2117  * ixgbevf_set_rx_mode - Multicast and unicast set
2118  * @netdev: network interface device structure
2119  *
2120  * The set_rx_method entry point is called whenever the multicast address
2121  * list, unicast address list or the network interface flags are updated.
2122  * This routine is responsible for configuring the hardware for proper
2123  * multicast mode and configuring requested unicast filters.
2124  **/
2125 static void ixgbevf_set_rx_mode(struct net_device *netdev)
2126 {
2127         struct ixgbevf_adapter *adapter = netdev_priv(netdev);
2128         struct ixgbe_hw *hw = &adapter->hw;
2129         unsigned int flags = netdev->flags;
2130         int xcast_mode;
2131
2132         /* request the most inclusive mode we need */
2133         if (flags & IFF_PROMISC)
2134                 xcast_mode = IXGBEVF_XCAST_MODE_PROMISC;
2135         else if (flags & IFF_ALLMULTI)
2136                 xcast_mode = IXGBEVF_XCAST_MODE_ALLMULTI;
2137         else if (flags & (IFF_BROADCAST | IFF_MULTICAST))
2138                 xcast_mode = IXGBEVF_XCAST_MODE_MULTI;
2139         else
2140                 xcast_mode = IXGBEVF_XCAST_MODE_NONE;
2141
2142         spin_lock_bh(&adapter->mbx_lock);
2143
2144         hw->mac.ops.update_xcast_mode(hw, xcast_mode);
2145
2146         /* reprogram multicast list */
2147         hw->mac.ops.update_mc_addr_list(hw, netdev);
2148
2149         ixgbevf_write_uc_addr_list(netdev);
2150
2151         spin_unlock_bh(&adapter->mbx_lock);
2152 }
2153
2154 static void ixgbevf_napi_enable_all(struct ixgbevf_adapter *adapter)
2155 {
2156         int q_idx;
2157         struct ixgbevf_q_vector *q_vector;
2158         int q_vectors = adapter->num_msix_vectors - NON_Q_VECTORS;
2159
2160         for (q_idx = 0; q_idx < q_vectors; q_idx++) {
2161                 q_vector = adapter->q_vector[q_idx];
2162                 napi_enable(&q_vector->napi);
2163         }
2164 }
2165
2166 static void ixgbevf_napi_disable_all(struct ixgbevf_adapter *adapter)
2167 {
2168         int q_idx;
2169         struct ixgbevf_q_vector *q_vector;
2170         int q_vectors = adapter->num_msix_vectors - NON_Q_VECTORS;
2171
2172         for (q_idx = 0; q_idx < q_vectors; q_idx++) {
2173                 q_vector = adapter->q_vector[q_idx];
2174                 napi_disable(&q_vector->napi);
2175         }
2176 }
2177
2178 static int ixgbevf_configure_dcb(struct ixgbevf_adapter *adapter)
2179 {
2180         struct ixgbe_hw *hw = &adapter->hw;
2181         unsigned int def_q = 0;
2182         unsigned int num_tcs = 0;
2183         unsigned int num_rx_queues = adapter->num_rx_queues;
2184         unsigned int num_tx_queues = adapter->num_tx_queues;
2185         int err;
2186
2187         spin_lock_bh(&adapter->mbx_lock);
2188
2189         /* fetch queue configuration from the PF */
2190         err = ixgbevf_get_queues(hw, &num_tcs, &def_q);
2191
2192         spin_unlock_bh(&adapter->mbx_lock);
2193
2194         if (err)
2195                 return err;
2196
2197         if (num_tcs > 1) {
2198                 /* we need only one Tx queue */
2199                 num_tx_queues = 1;
2200
2201                 /* update default Tx ring register index */
2202                 adapter->tx_ring[0]->reg_idx = def_q;
2203
2204                 /* we need as many queues as traffic classes */
2205                 num_rx_queues = num_tcs;
2206         }
2207
2208         /* if we have a bad config abort request queue reset */
2209         if ((adapter->num_rx_queues != num_rx_queues) ||
2210             (adapter->num_tx_queues != num_tx_queues)) {
2211                 /* force mailbox timeout to prevent further messages */
2212                 hw->mbx.timeout = 0;
2213
2214                 /* wait for watchdog to come around and bail us out */
2215                 set_bit(__IXGBEVF_QUEUE_RESET_REQUESTED, &adapter->state);
2216         }
2217
2218         return 0;
2219 }
2220
2221 static void ixgbevf_configure(struct ixgbevf_adapter *adapter)
2222 {
2223         ixgbevf_configure_dcb(adapter);
2224
2225         ixgbevf_set_rx_mode(adapter->netdev);
2226
2227         ixgbevf_restore_vlan(adapter);
2228         ixgbevf_ipsec_restore(adapter);
2229
2230         ixgbevf_configure_tx(adapter);
2231         ixgbevf_configure_rx(adapter);
2232 }
2233
2234 static void ixgbevf_save_reset_stats(struct ixgbevf_adapter *adapter)
2235 {
2236         /* Only save pre-reset stats if there are some */
2237         if (adapter->stats.vfgprc || adapter->stats.vfgptc) {
2238                 adapter->stats.saved_reset_vfgprc += adapter->stats.vfgprc -
2239                         adapter->stats.base_vfgprc;
2240                 adapter->stats.saved_reset_vfgptc += adapter->stats.vfgptc -
2241                         adapter->stats.base_vfgptc;
2242                 adapter->stats.saved_reset_vfgorc += adapter->stats.vfgorc -
2243                         adapter->stats.base_vfgorc;
2244                 adapter->stats.saved_reset_vfgotc += adapter->stats.vfgotc -
2245                         adapter->stats.base_vfgotc;
2246                 adapter->stats.saved_reset_vfmprc += adapter->stats.vfmprc -
2247                         adapter->stats.base_vfmprc;
2248         }
2249 }
2250
2251 static void ixgbevf_init_last_counter_stats(struct ixgbevf_adapter *adapter)
2252 {
2253         struct ixgbe_hw *hw = &adapter->hw;
2254
2255         adapter->stats.last_vfgprc = IXGBE_READ_REG(hw, IXGBE_VFGPRC);
2256         adapter->stats.last_vfgorc = IXGBE_READ_REG(hw, IXGBE_VFGORC_LSB);
2257         adapter->stats.last_vfgorc |=
2258                 (((u64)(IXGBE_READ_REG(hw, IXGBE_VFGORC_MSB))) << 32);
2259         adapter->stats.last_vfgptc = IXGBE_READ_REG(hw, IXGBE_VFGPTC);
2260         adapter->stats.last_vfgotc = IXGBE_READ_REG(hw, IXGBE_VFGOTC_LSB);
2261         adapter->stats.last_vfgotc |=
2262                 (((u64)(IXGBE_READ_REG(hw, IXGBE_VFGOTC_MSB))) << 32);
2263         adapter->stats.last_vfmprc = IXGBE_READ_REG(hw, IXGBE_VFMPRC);
2264
2265         adapter->stats.base_vfgprc = adapter->stats.last_vfgprc;
2266         adapter->stats.base_vfgorc = adapter->stats.last_vfgorc;
2267         adapter->stats.base_vfgptc = adapter->stats.last_vfgptc;
2268         adapter->stats.base_vfgotc = adapter->stats.last_vfgotc;
2269         adapter->stats.base_vfmprc = adapter->stats.last_vfmprc;
2270 }
2271
2272 static void ixgbevf_negotiate_api(struct ixgbevf_adapter *adapter)
2273 {
2274         struct ixgbe_hw *hw = &adapter->hw;
2275         static const int api[] = {
2276                 ixgbe_mbox_api_15,
2277                 ixgbe_mbox_api_14,
2278                 ixgbe_mbox_api_13,
2279                 ixgbe_mbox_api_12,
2280                 ixgbe_mbox_api_11,
2281                 ixgbe_mbox_api_10,
2282                 ixgbe_mbox_api_unknown
2283         };
2284         int err, idx = 0;
2285
2286         spin_lock_bh(&adapter->mbx_lock);
2287
2288         while (api[idx] != ixgbe_mbox_api_unknown) {
2289                 err = hw->mac.ops.negotiate_api_version(hw, api[idx]);
2290                 if (!err)
2291                         break;
2292                 idx++;
2293         }
2294
2295         if (hw->api_version >= ixgbe_mbox_api_15) {
2296                 hw->mbx.ops.init_params(hw);
2297                 memcpy(&hw->mbx.ops, &ixgbevf_mbx_ops,
2298                        sizeof(struct ixgbe_mbx_operations));
2299         }
2300
2301         spin_unlock_bh(&adapter->mbx_lock);
2302 }
2303
2304 static void ixgbevf_up_complete(struct ixgbevf_adapter *adapter)
2305 {
2306         struct net_device *netdev = adapter->netdev;
2307         struct pci_dev *pdev = adapter->pdev;
2308         struct ixgbe_hw *hw = &adapter->hw;
2309         bool state;
2310
2311         ixgbevf_configure_msix(adapter);
2312
2313         spin_lock_bh(&adapter->mbx_lock);
2314
2315         if (is_valid_ether_addr(hw->mac.addr))
2316                 hw->mac.ops.set_rar(hw, 0, hw->mac.addr, 0);
2317         else
2318                 hw->mac.ops.set_rar(hw, 0, hw->mac.perm_addr, 0);
2319
2320         spin_unlock_bh(&adapter->mbx_lock);
2321
2322         state = adapter->link_state;
2323         hw->mac.ops.get_link_state(hw, &adapter->link_state);
2324         if (state && state != adapter->link_state)
2325                 dev_info(&pdev->dev, "VF is administratively disabled\n");
2326
2327         smp_mb__before_atomic();
2328         clear_bit(__IXGBEVF_DOWN, &adapter->state);
2329         ixgbevf_napi_enable_all(adapter);
2330
2331         /* clear any pending interrupts, may auto mask */
2332         IXGBE_READ_REG(hw, IXGBE_VTEICR);
2333         ixgbevf_irq_enable(adapter);
2334
2335         /* enable transmits */
2336         netif_tx_start_all_queues(netdev);
2337
2338         ixgbevf_save_reset_stats(adapter);
2339         ixgbevf_init_last_counter_stats(adapter);
2340
2341         hw->mac.get_link_status = 1;
2342         mod_timer(&adapter->service_timer, jiffies);
2343 }
2344
2345 void ixgbevf_up(struct ixgbevf_adapter *adapter)
2346 {
2347         ixgbevf_configure(adapter);
2348
2349         ixgbevf_up_complete(adapter);
2350 }
2351
2352 /**
2353  * ixgbevf_clean_rx_ring - Free Rx Buffers per Queue
2354  * @rx_ring: ring to free buffers from
2355  **/
2356 static void ixgbevf_clean_rx_ring(struct ixgbevf_ring *rx_ring)
2357 {
2358         u16 i = rx_ring->next_to_clean;
2359
2360         /* Free Rx ring sk_buff */
2361         if (rx_ring->skb) {
2362                 dev_kfree_skb(rx_ring->skb);
2363                 rx_ring->skb = NULL;
2364         }
2365
2366         /* Free all the Rx ring pages */
2367         while (i != rx_ring->next_to_alloc) {
2368                 struct ixgbevf_rx_buffer *rx_buffer;
2369
2370                 rx_buffer = &rx_ring->rx_buffer_info[i];
2371
2372                 /* Invalidate cache lines that may have been written to by
2373                  * device so that we avoid corrupting memory.
2374                  */
2375                 dma_sync_single_range_for_cpu(rx_ring->dev,
2376                                               rx_buffer->dma,
2377                                               rx_buffer->page_offset,
2378                                               ixgbevf_rx_bufsz(rx_ring),
2379                                               DMA_FROM_DEVICE);
2380
2381                 /* free resources associated with mapping */
2382                 dma_unmap_page_attrs(rx_ring->dev,
2383                                      rx_buffer->dma,
2384                                      ixgbevf_rx_pg_size(rx_ring),
2385                                      DMA_FROM_DEVICE,
2386                                      IXGBEVF_RX_DMA_ATTR);
2387
2388                 __page_frag_cache_drain(rx_buffer->page,
2389                                         rx_buffer->pagecnt_bias);
2390
2391                 i++;
2392                 if (i == rx_ring->count)
2393                         i = 0;
2394         }
2395
2396         rx_ring->next_to_alloc = 0;
2397         rx_ring->next_to_clean = 0;
2398         rx_ring->next_to_use = 0;
2399 }
2400
2401 /**
2402  * ixgbevf_clean_tx_ring - Free Tx Buffers
2403  * @tx_ring: ring to be cleaned
2404  **/
2405 static void ixgbevf_clean_tx_ring(struct ixgbevf_ring *tx_ring)
2406 {
2407         u16 i = tx_ring->next_to_clean;
2408         struct ixgbevf_tx_buffer *tx_buffer = &tx_ring->tx_buffer_info[i];
2409
2410         while (i != tx_ring->next_to_use) {
2411                 union ixgbe_adv_tx_desc *eop_desc, *tx_desc;
2412
2413                 /* Free all the Tx ring sk_buffs */
2414                 if (ring_is_xdp(tx_ring))
2415                         page_frag_free(tx_buffer->data);
2416                 else
2417                         dev_kfree_skb_any(tx_buffer->skb);
2418
2419                 /* unmap skb header data */
2420                 dma_unmap_single(tx_ring->dev,
2421                                  dma_unmap_addr(tx_buffer, dma),
2422                                  dma_unmap_len(tx_buffer, len),
2423                                  DMA_TO_DEVICE);
2424
2425                 /* check for eop_desc to determine the end of the packet */
2426                 eop_desc = tx_buffer->next_to_watch;
2427                 tx_desc = IXGBEVF_TX_DESC(tx_ring, i);
2428
2429                 /* unmap remaining buffers */
2430                 while (tx_desc != eop_desc) {
2431                         tx_buffer++;
2432                         tx_desc++;
2433                         i++;
2434                         if (unlikely(i == tx_ring->count)) {
2435                                 i = 0;
2436                                 tx_buffer = tx_ring->tx_buffer_info;
2437                                 tx_desc = IXGBEVF_TX_DESC(tx_ring, 0);
2438                         }
2439
2440                         /* unmap any remaining paged data */
2441                         if (dma_unmap_len(tx_buffer, len))
2442                                 dma_unmap_page(tx_ring->dev,
2443                                                dma_unmap_addr(tx_buffer, dma),
2444                                                dma_unmap_len(tx_buffer, len),
2445                                                DMA_TO_DEVICE);
2446                 }
2447
2448                 /* move us one more past the eop_desc for start of next pkt */
2449                 tx_buffer++;
2450                 i++;
2451                 if (unlikely(i == tx_ring->count)) {
2452                         i = 0;
2453                         tx_buffer = tx_ring->tx_buffer_info;
2454                 }
2455         }
2456
2457         /* reset next_to_use and next_to_clean */
2458         tx_ring->next_to_use = 0;
2459         tx_ring->next_to_clean = 0;
2460
2461 }
2462
2463 /**
2464  * ixgbevf_clean_all_rx_rings - Free Rx Buffers for all queues
2465  * @adapter: board private structure
2466  **/
2467 static void ixgbevf_clean_all_rx_rings(struct ixgbevf_adapter *adapter)
2468 {
2469         int i;
2470
2471         for (i = 0; i < adapter->num_rx_queues; i++)
2472                 ixgbevf_clean_rx_ring(adapter->rx_ring[i]);
2473 }
2474
2475 /**
2476  * ixgbevf_clean_all_tx_rings - Free Tx Buffers for all queues
2477  * @adapter: board private structure
2478  **/
2479 static void ixgbevf_clean_all_tx_rings(struct ixgbevf_adapter *adapter)
2480 {
2481         int i;
2482
2483         for (i = 0; i < adapter->num_tx_queues; i++)
2484                 ixgbevf_clean_tx_ring(adapter->tx_ring[i]);
2485         for (i = 0; i < adapter->num_xdp_queues; i++)
2486                 ixgbevf_clean_tx_ring(adapter->xdp_ring[i]);
2487 }
2488
2489 void ixgbevf_down(struct ixgbevf_adapter *adapter)
2490 {
2491         struct net_device *netdev = adapter->netdev;
2492         struct ixgbe_hw *hw = &adapter->hw;
2493         int i;
2494
2495         /* signal that we are down to the interrupt handler */
2496         if (test_and_set_bit(__IXGBEVF_DOWN, &adapter->state))
2497                 return; /* do nothing if already down */
2498
2499         /* disable all enabled Rx queues */
2500         for (i = 0; i < adapter->num_rx_queues; i++)
2501                 ixgbevf_disable_rx_queue(adapter, adapter->rx_ring[i]);
2502
2503         usleep_range(10000, 20000);
2504
2505         netif_tx_stop_all_queues(netdev);
2506
2507         /* call carrier off first to avoid false dev_watchdog timeouts */
2508         netif_carrier_off(netdev);
2509         netif_tx_disable(netdev);
2510
2511         ixgbevf_irq_disable(adapter);
2512
2513         ixgbevf_napi_disable_all(adapter);
2514
2515         del_timer_sync(&adapter->service_timer);
2516
2517         /* disable transmits in the hardware now that interrupts are off */
2518         for (i = 0; i < adapter->num_tx_queues; i++) {
2519                 u8 reg_idx = adapter->tx_ring[i]->reg_idx;
2520
2521                 IXGBE_WRITE_REG(hw, IXGBE_VFTXDCTL(reg_idx),
2522                                 IXGBE_TXDCTL_SWFLSH);
2523         }
2524
2525         for (i = 0; i < adapter->num_xdp_queues; i++) {
2526                 u8 reg_idx = adapter->xdp_ring[i]->reg_idx;
2527
2528                 IXGBE_WRITE_REG(hw, IXGBE_VFTXDCTL(reg_idx),
2529                                 IXGBE_TXDCTL_SWFLSH);
2530         }
2531
2532         if (!pci_channel_offline(adapter->pdev))
2533                 ixgbevf_reset(adapter);
2534
2535         ixgbevf_clean_all_tx_rings(adapter);
2536         ixgbevf_clean_all_rx_rings(adapter);
2537 }
2538
2539 void ixgbevf_reinit_locked(struct ixgbevf_adapter *adapter)
2540 {
2541         while (test_and_set_bit(__IXGBEVF_RESETTING, &adapter->state))
2542                 msleep(1);
2543
2544         ixgbevf_down(adapter);
2545         pci_set_master(adapter->pdev);
2546         ixgbevf_up(adapter);
2547
2548         clear_bit(__IXGBEVF_RESETTING, &adapter->state);
2549 }
2550
2551 void ixgbevf_reset(struct ixgbevf_adapter *adapter)
2552 {
2553         struct ixgbe_hw *hw = &adapter->hw;
2554         struct net_device *netdev = adapter->netdev;
2555
2556         if (hw->mac.ops.reset_hw(hw)) {
2557                 hw_dbg(hw, "PF still resetting\n");
2558         } else {
2559                 hw->mac.ops.init_hw(hw);
2560                 ixgbevf_negotiate_api(adapter);
2561         }
2562
2563         if (is_valid_ether_addr(adapter->hw.mac.addr)) {
2564                 eth_hw_addr_set(netdev, adapter->hw.mac.addr);
2565                 ether_addr_copy(netdev->perm_addr, adapter->hw.mac.addr);
2566         }
2567
2568         adapter->last_reset = jiffies;
2569 }
2570
2571 static int ixgbevf_acquire_msix_vectors(struct ixgbevf_adapter *adapter,
2572                                         int vectors)
2573 {
2574         int vector_threshold;
2575
2576         /* We'll want at least 2 (vector_threshold):
2577          * 1) TxQ[0] + RxQ[0] handler
2578          * 2) Other (Link Status Change, etc.)
2579          */
2580         vector_threshold = MIN_MSIX_COUNT;
2581
2582         /* The more we get, the more we will assign to Tx/Rx Cleanup
2583          * for the separate queues...where Rx Cleanup >= Tx Cleanup.
2584          * Right now, we simply care about how many we'll get; we'll
2585          * set them up later while requesting irq's.
2586          */
2587         vectors = pci_enable_msix_range(adapter->pdev, adapter->msix_entries,
2588                                         vector_threshold, vectors);
2589
2590         if (vectors < 0) {
2591                 dev_err(&adapter->pdev->dev,
2592                         "Unable to allocate MSI-X interrupts\n");
2593                 kfree(adapter->msix_entries);
2594                 adapter->msix_entries = NULL;
2595                 return vectors;
2596         }
2597
2598         /* Adjust for only the vectors we'll use, which is minimum
2599          * of max_msix_q_vectors + NON_Q_VECTORS, or the number of
2600          * vectors we were allocated.
2601          */
2602         adapter->num_msix_vectors = vectors;
2603
2604         return 0;
2605 }
2606
2607 /**
2608  * ixgbevf_set_num_queues - Allocate queues for device, feature dependent
2609  * @adapter: board private structure to initialize
2610  *
2611  * This is the top level queue allocation routine.  The order here is very
2612  * important, starting with the "most" number of features turned on at once,
2613  * and ending with the smallest set of features.  This way large combinations
2614  * can be allocated if they're turned on, and smaller combinations are the
2615  * fall through conditions.
2616  *
2617  **/
2618 static void ixgbevf_set_num_queues(struct ixgbevf_adapter *adapter)
2619 {
2620         struct ixgbe_hw *hw = &adapter->hw;
2621         unsigned int def_q = 0;
2622         unsigned int num_tcs = 0;
2623         int err;
2624
2625         /* Start with base case */
2626         adapter->num_rx_queues = 1;
2627         adapter->num_tx_queues = 1;
2628         adapter->num_xdp_queues = 0;
2629
2630         spin_lock_bh(&adapter->mbx_lock);
2631
2632         /* fetch queue configuration from the PF */
2633         err = ixgbevf_get_queues(hw, &num_tcs, &def_q);
2634
2635         spin_unlock_bh(&adapter->mbx_lock);
2636
2637         if (err)
2638                 return;
2639
2640         /* we need as many queues as traffic classes */
2641         if (num_tcs > 1) {
2642                 adapter->num_rx_queues = num_tcs;
2643         } else {
2644                 u16 rss = min_t(u16, num_online_cpus(), IXGBEVF_MAX_RSS_QUEUES);
2645
2646                 switch (hw->api_version) {
2647                 case ixgbe_mbox_api_11:
2648                 case ixgbe_mbox_api_12:
2649                 case ixgbe_mbox_api_13:
2650                 case ixgbe_mbox_api_14:
2651                 case ixgbe_mbox_api_15:
2652                         if (adapter->xdp_prog &&
2653                             hw->mac.max_tx_queues == rss)
2654                                 rss = rss > 3 ? 2 : 1;
2655
2656                         adapter->num_rx_queues = rss;
2657                         adapter->num_tx_queues = rss;
2658                         adapter->num_xdp_queues = adapter->xdp_prog ? rss : 0;
2659                         break;
2660                 default:
2661                         break;
2662                 }
2663         }
2664 }
2665
2666 /**
2667  * ixgbevf_set_interrupt_capability - set MSI-X or FAIL if not supported
2668  * @adapter: board private structure to initialize
2669  *
2670  * Attempt to configure the interrupts using the best available
2671  * capabilities of the hardware and the kernel.
2672  **/
2673 static int ixgbevf_set_interrupt_capability(struct ixgbevf_adapter *adapter)
2674 {
2675         int vector, v_budget;
2676
2677         /* It's easy to be greedy for MSI-X vectors, but it really
2678          * doesn't do us much good if we have a lot more vectors
2679          * than CPU's.  So let's be conservative and only ask for
2680          * (roughly) the same number of vectors as there are CPU's.
2681          * The default is to use pairs of vectors.
2682          */
2683         v_budget = max(adapter->num_rx_queues, adapter->num_tx_queues);
2684         v_budget = min_t(int, v_budget, num_online_cpus());
2685         v_budget += NON_Q_VECTORS;
2686
2687         adapter->msix_entries = kcalloc(v_budget,
2688                                         sizeof(struct msix_entry), GFP_KERNEL);
2689         if (!adapter->msix_entries)
2690                 return -ENOMEM;
2691
2692         for (vector = 0; vector < v_budget; vector++)
2693                 adapter->msix_entries[vector].entry = vector;
2694
2695         /* A failure in MSI-X entry allocation isn't fatal, but the VF driver
2696          * does not support any other modes, so we will simply fail here. Note
2697          * that we clean up the msix_entries pointer else-where.
2698          */
2699         return ixgbevf_acquire_msix_vectors(adapter, v_budget);
2700 }
2701
2702 static void ixgbevf_add_ring(struct ixgbevf_ring *ring,
2703                              struct ixgbevf_ring_container *head)
2704 {
2705         ring->next = head->ring;
2706         head->ring = ring;
2707         head->count++;
2708 }
2709
2710 /**
2711  * ixgbevf_alloc_q_vector - Allocate memory for a single interrupt vector
2712  * @adapter: board private structure to initialize
2713  * @v_idx: index of vector in adapter struct
2714  * @txr_count: number of Tx rings for q vector
2715  * @txr_idx: index of first Tx ring to assign
2716  * @xdp_count: total number of XDP rings to allocate
2717  * @xdp_idx: index of first XDP ring to allocate
2718  * @rxr_count: number of Rx rings for q vector
2719  * @rxr_idx: index of first Rx ring to assign
2720  *
2721  * We allocate one q_vector.  If allocation fails we return -ENOMEM.
2722  **/
2723 static int ixgbevf_alloc_q_vector(struct ixgbevf_adapter *adapter, int v_idx,
2724                                   int txr_count, int txr_idx,
2725                                   int xdp_count, int xdp_idx,
2726                                   int rxr_count, int rxr_idx)
2727 {
2728         struct ixgbevf_q_vector *q_vector;
2729         int reg_idx = txr_idx + xdp_idx;
2730         struct ixgbevf_ring *ring;
2731         int ring_count, size;
2732
2733         ring_count = txr_count + xdp_count + rxr_count;
2734         size = sizeof(*q_vector) + (sizeof(*ring) * ring_count);
2735
2736         /* allocate q_vector and rings */
2737         q_vector = kzalloc(size, GFP_KERNEL);
2738         if (!q_vector)
2739                 return -ENOMEM;
2740
2741         /* initialize NAPI */
2742         netif_napi_add(adapter->netdev, &q_vector->napi, ixgbevf_poll);
2743
2744         /* tie q_vector and adapter together */
2745         adapter->q_vector[v_idx] = q_vector;
2746         q_vector->adapter = adapter;
2747         q_vector->v_idx = v_idx;
2748
2749         /* initialize pointer to rings */
2750         ring = q_vector->ring;
2751
2752         while (txr_count) {
2753                 /* assign generic ring traits */
2754                 ring->dev = &adapter->pdev->dev;
2755                 ring->netdev = adapter->netdev;
2756
2757                 /* configure backlink on ring */
2758                 ring->q_vector = q_vector;
2759
2760                 /* update q_vector Tx values */
2761                 ixgbevf_add_ring(ring, &q_vector->tx);
2762
2763                 /* apply Tx specific ring traits */
2764                 ring->count = adapter->tx_ring_count;
2765                 ring->queue_index = txr_idx;
2766                 ring->reg_idx = reg_idx;
2767
2768                 /* assign ring to adapter */
2769                 adapter->tx_ring[txr_idx] = ring;
2770
2771                 /* update count and index */
2772                 txr_count--;
2773                 txr_idx++;
2774                 reg_idx++;
2775
2776                 /* push pointer to next ring */
2777                 ring++;
2778         }
2779
2780         while (xdp_count) {
2781                 /* assign generic ring traits */
2782                 ring->dev = &adapter->pdev->dev;
2783                 ring->netdev = adapter->netdev;
2784
2785                 /* configure backlink on ring */
2786                 ring->q_vector = q_vector;
2787
2788                 /* update q_vector Tx values */
2789                 ixgbevf_add_ring(ring, &q_vector->tx);
2790
2791                 /* apply Tx specific ring traits */
2792                 ring->count = adapter->tx_ring_count;
2793                 ring->queue_index = xdp_idx;
2794                 ring->reg_idx = reg_idx;
2795                 set_ring_xdp(ring);
2796
2797                 /* assign ring to adapter */
2798                 adapter->xdp_ring[xdp_idx] = ring;
2799
2800                 /* update count and index */
2801                 xdp_count--;
2802                 xdp_idx++;
2803                 reg_idx++;
2804
2805                 /* push pointer to next ring */
2806                 ring++;
2807         }
2808
2809         while (rxr_count) {
2810                 /* assign generic ring traits */
2811                 ring->dev = &adapter->pdev->dev;
2812                 ring->netdev = adapter->netdev;
2813
2814                 /* configure backlink on ring */
2815                 ring->q_vector = q_vector;
2816
2817                 /* update q_vector Rx values */
2818                 ixgbevf_add_ring(ring, &q_vector->rx);
2819
2820                 /* apply Rx specific ring traits */
2821                 ring->count = adapter->rx_ring_count;
2822                 ring->queue_index = rxr_idx;
2823                 ring->reg_idx = rxr_idx;
2824
2825                 /* assign ring to adapter */
2826                 adapter->rx_ring[rxr_idx] = ring;
2827
2828                 /* update count and index */
2829                 rxr_count--;
2830                 rxr_idx++;
2831
2832                 /* push pointer to next ring */
2833                 ring++;
2834         }
2835
2836         return 0;
2837 }
2838
2839 /**
2840  * ixgbevf_free_q_vector - Free memory allocated for specific interrupt vector
2841  * @adapter: board private structure to initialize
2842  * @v_idx: index of vector in adapter struct
2843  *
2844  * This function frees the memory allocated to the q_vector.  In addition if
2845  * NAPI is enabled it will delete any references to the NAPI struct prior
2846  * to freeing the q_vector.
2847  **/
2848 static void ixgbevf_free_q_vector(struct ixgbevf_adapter *adapter, int v_idx)
2849 {
2850         struct ixgbevf_q_vector *q_vector = adapter->q_vector[v_idx];
2851         struct ixgbevf_ring *ring;
2852
2853         ixgbevf_for_each_ring(ring, q_vector->tx) {
2854                 if (ring_is_xdp(ring))
2855                         adapter->xdp_ring[ring->queue_index] = NULL;
2856                 else
2857                         adapter->tx_ring[ring->queue_index] = NULL;
2858         }
2859
2860         ixgbevf_for_each_ring(ring, q_vector->rx)
2861                 adapter->rx_ring[ring->queue_index] = NULL;
2862
2863         adapter->q_vector[v_idx] = NULL;
2864         netif_napi_del(&q_vector->napi);
2865
2866         /* ixgbevf_get_stats() might access the rings on this vector,
2867          * we must wait a grace period before freeing it.
2868          */
2869         kfree_rcu(q_vector, rcu);
2870 }
2871
2872 /**
2873  * ixgbevf_alloc_q_vectors - Allocate memory for interrupt vectors
2874  * @adapter: board private structure to initialize
2875  *
2876  * We allocate one q_vector per queue interrupt.  If allocation fails we
2877  * return -ENOMEM.
2878  **/
2879 static int ixgbevf_alloc_q_vectors(struct ixgbevf_adapter *adapter)
2880 {
2881         int q_vectors = adapter->num_msix_vectors - NON_Q_VECTORS;
2882         int rxr_remaining = adapter->num_rx_queues;
2883         int txr_remaining = adapter->num_tx_queues;
2884         int xdp_remaining = adapter->num_xdp_queues;
2885         int rxr_idx = 0, txr_idx = 0, xdp_idx = 0, v_idx = 0;
2886         int err;
2887
2888         if (q_vectors >= (rxr_remaining + txr_remaining + xdp_remaining)) {
2889                 for (; rxr_remaining; v_idx++, q_vectors--) {
2890                         int rqpv = DIV_ROUND_UP(rxr_remaining, q_vectors);
2891
2892                         err = ixgbevf_alloc_q_vector(adapter, v_idx,
2893                                                      0, 0, 0, 0, rqpv, rxr_idx);
2894                         if (err)
2895                                 goto err_out;
2896
2897                         /* update counts and index */
2898                         rxr_remaining -= rqpv;
2899                         rxr_idx += rqpv;
2900                 }
2901         }
2902
2903         for (; q_vectors; v_idx++, q_vectors--) {
2904                 int rqpv = DIV_ROUND_UP(rxr_remaining, q_vectors);
2905                 int tqpv = DIV_ROUND_UP(txr_remaining, q_vectors);
2906                 int xqpv = DIV_ROUND_UP(xdp_remaining, q_vectors);
2907
2908                 err = ixgbevf_alloc_q_vector(adapter, v_idx,
2909                                              tqpv, txr_idx,
2910                                              xqpv, xdp_idx,
2911                                              rqpv, rxr_idx);
2912
2913                 if (err)
2914                         goto err_out;
2915
2916                 /* update counts and index */
2917                 rxr_remaining -= rqpv;
2918                 rxr_idx += rqpv;
2919                 txr_remaining -= tqpv;
2920                 txr_idx += tqpv;
2921                 xdp_remaining -= xqpv;
2922                 xdp_idx += xqpv;
2923         }
2924
2925         return 0;
2926
2927 err_out:
2928         while (v_idx) {
2929                 v_idx--;
2930                 ixgbevf_free_q_vector(adapter, v_idx);
2931         }
2932
2933         return -ENOMEM;
2934 }
2935
2936 /**
2937  * ixgbevf_free_q_vectors - Free memory allocated for interrupt vectors
2938  * @adapter: board private structure to initialize
2939  *
2940  * This function frees the memory allocated to the q_vectors.  In addition if
2941  * NAPI is enabled it will delete any references to the NAPI struct prior
2942  * to freeing the q_vector.
2943  **/
2944 static void ixgbevf_free_q_vectors(struct ixgbevf_adapter *adapter)
2945 {
2946         int q_vectors = adapter->num_msix_vectors - NON_Q_VECTORS;
2947
2948         while (q_vectors) {
2949                 q_vectors--;
2950                 ixgbevf_free_q_vector(adapter, q_vectors);
2951         }
2952 }
2953
2954 /**
2955  * ixgbevf_reset_interrupt_capability - Reset MSIX setup
2956  * @adapter: board private structure
2957  *
2958  **/
2959 static void ixgbevf_reset_interrupt_capability(struct ixgbevf_adapter *adapter)
2960 {
2961         if (!adapter->msix_entries)
2962                 return;
2963
2964         pci_disable_msix(adapter->pdev);
2965         kfree(adapter->msix_entries);
2966         adapter->msix_entries = NULL;
2967 }
2968
2969 /**
2970  * ixgbevf_init_interrupt_scheme - Determine if MSIX is supported and init
2971  * @adapter: board private structure to initialize
2972  *
2973  **/
2974 static int ixgbevf_init_interrupt_scheme(struct ixgbevf_adapter *adapter)
2975 {
2976         int err;
2977
2978         /* Number of supported queues */
2979         ixgbevf_set_num_queues(adapter);
2980
2981         err = ixgbevf_set_interrupt_capability(adapter);
2982         if (err) {
2983                 hw_dbg(&adapter->hw,
2984                        "Unable to setup interrupt capabilities\n");
2985                 goto err_set_interrupt;
2986         }
2987
2988         err = ixgbevf_alloc_q_vectors(adapter);
2989         if (err) {
2990                 hw_dbg(&adapter->hw, "Unable to allocate memory for queue vectors\n");
2991                 goto err_alloc_q_vectors;
2992         }
2993
2994         hw_dbg(&adapter->hw, "Multiqueue %s: Rx Queue count = %u, Tx Queue count = %u XDP Queue count %u\n",
2995                (adapter->num_rx_queues > 1) ? "Enabled" : "Disabled",
2996                adapter->num_rx_queues, adapter->num_tx_queues,
2997                adapter->num_xdp_queues);
2998
2999         set_bit(__IXGBEVF_DOWN, &adapter->state);
3000
3001         return 0;
3002 err_alloc_q_vectors:
3003         ixgbevf_reset_interrupt_capability(adapter);
3004 err_set_interrupt:
3005         return err;
3006 }
3007
3008 /**
3009  * ixgbevf_clear_interrupt_scheme - Clear the current interrupt scheme settings
3010  * @adapter: board private structure to clear interrupt scheme on
3011  *
3012  * We go through and clear interrupt specific resources and reset the structure
3013  * to pre-load conditions
3014  **/
3015 static void ixgbevf_clear_interrupt_scheme(struct ixgbevf_adapter *adapter)
3016 {
3017         adapter->num_tx_queues = 0;
3018         adapter->num_xdp_queues = 0;
3019         adapter->num_rx_queues = 0;
3020
3021         ixgbevf_free_q_vectors(adapter);
3022         ixgbevf_reset_interrupt_capability(adapter);
3023 }
3024
3025 /**
3026  * ixgbevf_sw_init - Initialize general software structures
3027  * @adapter: board private structure to initialize
3028  *
3029  * ixgbevf_sw_init initializes the Adapter private data structure.
3030  * Fields are initialized based on PCI device information and
3031  * OS network device settings (MTU size).
3032  **/
3033 static int ixgbevf_sw_init(struct ixgbevf_adapter *adapter)
3034 {
3035         struct ixgbe_hw *hw = &adapter->hw;
3036         struct pci_dev *pdev = adapter->pdev;
3037         struct net_device *netdev = adapter->netdev;
3038         int err;
3039
3040         /* PCI config space info */
3041         hw->vendor_id = pdev->vendor;
3042         hw->device_id = pdev->device;
3043         hw->revision_id = pdev->revision;
3044         hw->subsystem_vendor_id = pdev->subsystem_vendor;
3045         hw->subsystem_device_id = pdev->subsystem_device;
3046
3047         hw->mbx.ops.init_params(hw);
3048
3049         if (hw->mac.type >= ixgbe_mac_X550_vf) {
3050                 err = ixgbevf_init_rss_key(adapter);
3051                 if (err)
3052                         goto out;
3053         }
3054
3055         /* assume legacy case in which PF would only give VF 2 queues */
3056         hw->mac.max_tx_queues = 2;
3057         hw->mac.max_rx_queues = 2;
3058
3059         /* lock to protect mailbox accesses */
3060         spin_lock_init(&adapter->mbx_lock);
3061
3062         err = hw->mac.ops.reset_hw(hw);
3063         if (err) {
3064                 dev_info(&pdev->dev,
3065                          "PF still in reset state.  Is the PF interface up?\n");
3066         } else {
3067                 err = hw->mac.ops.init_hw(hw);
3068                 if (err) {
3069                         pr_err("init_shared_code failed: %d\n", err);
3070                         goto out;
3071                 }
3072                 ixgbevf_negotiate_api(adapter);
3073                 err = hw->mac.ops.get_mac_addr(hw, hw->mac.addr);
3074                 if (err)
3075                         dev_info(&pdev->dev, "Error reading MAC address\n");
3076                 else if (is_zero_ether_addr(adapter->hw.mac.addr))
3077                         dev_info(&pdev->dev,
3078                                  "MAC address not assigned by administrator.\n");
3079                 eth_hw_addr_set(netdev, hw->mac.addr);
3080         }
3081
3082         if (!is_valid_ether_addr(netdev->dev_addr)) {
3083                 dev_info(&pdev->dev, "Assigning random MAC address\n");
3084                 eth_hw_addr_random(netdev);
3085                 ether_addr_copy(hw->mac.addr, netdev->dev_addr);
3086                 ether_addr_copy(hw->mac.perm_addr, netdev->dev_addr);
3087         }
3088
3089         /* Enable dynamic interrupt throttling rates */
3090         adapter->rx_itr_setting = 1;
3091         adapter->tx_itr_setting = 1;
3092
3093         /* set default ring sizes */
3094         adapter->tx_ring_count = IXGBEVF_DEFAULT_TXD;
3095         adapter->rx_ring_count = IXGBEVF_DEFAULT_RXD;
3096
3097         adapter->link_state = true;
3098
3099         set_bit(__IXGBEVF_DOWN, &adapter->state);
3100         return 0;
3101
3102 out:
3103         return err;
3104 }
3105
3106 #define UPDATE_VF_COUNTER_32bit(reg, last_counter, counter)     \
3107         {                                                       \
3108                 u32 current_counter = IXGBE_READ_REG(hw, reg);  \
3109                 if (current_counter < last_counter)             \
3110                         counter += 0x100000000LL;               \
3111                 last_counter = current_counter;                 \
3112                 counter &= 0xFFFFFFFF00000000LL;                \
3113                 counter |= current_counter;                     \
3114         }
3115
3116 #define UPDATE_VF_COUNTER_36bit(reg_lsb, reg_msb, last_counter, counter) \
3117         {                                                                \
3118                 u64 current_counter_lsb = IXGBE_READ_REG(hw, reg_lsb);   \
3119                 u64 current_counter_msb = IXGBE_READ_REG(hw, reg_msb);   \
3120                 u64 current_counter = (current_counter_msb << 32) |      \
3121                         current_counter_lsb;                             \
3122                 if (current_counter < last_counter)                      \
3123                         counter += 0x1000000000LL;                       \
3124                 last_counter = current_counter;                          \
3125                 counter &= 0xFFFFFFF000000000LL;                         \
3126                 counter |= current_counter;                              \
3127         }
3128 /**
3129  * ixgbevf_update_stats - Update the board statistics counters.
3130  * @adapter: board private structure
3131  **/
3132 void ixgbevf_update_stats(struct ixgbevf_adapter *adapter)
3133 {
3134         struct ixgbe_hw *hw = &adapter->hw;
3135         u64 alloc_rx_page_failed = 0, alloc_rx_buff_failed = 0;
3136         u64 alloc_rx_page = 0, hw_csum_rx_error = 0;
3137         int i;
3138
3139         if (test_bit(__IXGBEVF_DOWN, &adapter->state) ||
3140             test_bit(__IXGBEVF_RESETTING, &adapter->state))
3141                 return;
3142
3143         UPDATE_VF_COUNTER_32bit(IXGBE_VFGPRC, adapter->stats.last_vfgprc,
3144                                 adapter->stats.vfgprc);
3145         UPDATE_VF_COUNTER_32bit(IXGBE_VFGPTC, adapter->stats.last_vfgptc,
3146                                 adapter->stats.vfgptc);
3147         UPDATE_VF_COUNTER_36bit(IXGBE_VFGORC_LSB, IXGBE_VFGORC_MSB,
3148                                 adapter->stats.last_vfgorc,
3149                                 adapter->stats.vfgorc);
3150         UPDATE_VF_COUNTER_36bit(IXGBE_VFGOTC_LSB, IXGBE_VFGOTC_MSB,
3151                                 adapter->stats.last_vfgotc,
3152                                 adapter->stats.vfgotc);
3153         UPDATE_VF_COUNTER_32bit(IXGBE_VFMPRC, adapter->stats.last_vfmprc,
3154                                 adapter->stats.vfmprc);
3155
3156         for (i = 0;  i  < adapter->num_rx_queues;  i++) {
3157                 struct ixgbevf_ring *rx_ring = adapter->rx_ring[i];
3158
3159                 hw_csum_rx_error += rx_ring->rx_stats.csum_err;
3160                 alloc_rx_page_failed += rx_ring->rx_stats.alloc_rx_page_failed;
3161                 alloc_rx_buff_failed += rx_ring->rx_stats.alloc_rx_buff_failed;
3162                 alloc_rx_page += rx_ring->rx_stats.alloc_rx_page;
3163         }
3164
3165         adapter->hw_csum_rx_error = hw_csum_rx_error;
3166         adapter->alloc_rx_page_failed = alloc_rx_page_failed;
3167         adapter->alloc_rx_buff_failed = alloc_rx_buff_failed;
3168         adapter->alloc_rx_page = alloc_rx_page;
3169 }
3170
3171 /**
3172  * ixgbevf_service_timer - Timer Call-back
3173  * @t: pointer to timer_list struct
3174  **/
3175 static void ixgbevf_service_timer(struct timer_list *t)
3176 {
3177         struct ixgbevf_adapter *adapter = from_timer(adapter, t,
3178                                                      service_timer);
3179
3180         /* Reset the timer */
3181         mod_timer(&adapter->service_timer, (HZ * 2) + jiffies);
3182
3183         ixgbevf_service_event_schedule(adapter);
3184 }
3185
3186 static void ixgbevf_reset_subtask(struct ixgbevf_adapter *adapter)
3187 {
3188         if (!test_and_clear_bit(__IXGBEVF_RESET_REQUESTED, &adapter->state))
3189                 return;
3190
3191         rtnl_lock();
3192         /* If we're already down or resetting, just bail */
3193         if (test_bit(__IXGBEVF_DOWN, &adapter->state) ||
3194             test_bit(__IXGBEVF_REMOVING, &adapter->state) ||
3195             test_bit(__IXGBEVF_RESETTING, &adapter->state)) {
3196                 rtnl_unlock();
3197                 return;
3198         }
3199
3200         adapter->tx_timeout_count++;
3201
3202         ixgbevf_reinit_locked(adapter);
3203         rtnl_unlock();
3204 }
3205
3206 /**
3207  * ixgbevf_check_hang_subtask - check for hung queues and dropped interrupts
3208  * @adapter: pointer to the device adapter structure
3209  *
3210  * This function serves two purposes.  First it strobes the interrupt lines
3211  * in order to make certain interrupts are occurring.  Secondly it sets the
3212  * bits needed to check for TX hangs.  As a result we should immediately
3213  * determine if a hang has occurred.
3214  **/
3215 static void ixgbevf_check_hang_subtask(struct ixgbevf_adapter *adapter)
3216 {
3217         struct ixgbe_hw *hw = &adapter->hw;
3218         u32 eics = 0;
3219         int i;
3220
3221         /* If we're down or resetting, just bail */
3222         if (test_bit(__IXGBEVF_DOWN, &adapter->state) ||
3223             test_bit(__IXGBEVF_RESETTING, &adapter->state))
3224                 return;
3225
3226         /* Force detection of hung controller */
3227         if (netif_carrier_ok(adapter->netdev)) {
3228                 for (i = 0; i < adapter->num_tx_queues; i++)
3229                         set_check_for_tx_hang(adapter->tx_ring[i]);
3230                 for (i = 0; i < adapter->num_xdp_queues; i++)
3231                         set_check_for_tx_hang(adapter->xdp_ring[i]);
3232         }
3233
3234         /* get one bit for every active Tx/Rx interrupt vector */
3235         for (i = 0; i < adapter->num_msix_vectors - NON_Q_VECTORS; i++) {
3236                 struct ixgbevf_q_vector *qv = adapter->q_vector[i];
3237
3238                 if (qv->rx.ring || qv->tx.ring)
3239                         eics |= BIT(i);
3240         }
3241
3242         /* Cause software interrupt to ensure rings are cleaned */
3243         IXGBE_WRITE_REG(hw, IXGBE_VTEICS, eics);
3244 }
3245
3246 /**
3247  * ixgbevf_watchdog_update_link - update the link status
3248  * @adapter: pointer to the device adapter structure
3249  **/
3250 static void ixgbevf_watchdog_update_link(struct ixgbevf_adapter *adapter)
3251 {
3252         struct ixgbe_hw *hw = &adapter->hw;
3253         u32 link_speed = adapter->link_speed;
3254         bool link_up = adapter->link_up;
3255         s32 err;
3256
3257         spin_lock_bh(&adapter->mbx_lock);
3258
3259         err = hw->mac.ops.check_link(hw, &link_speed, &link_up, false);
3260
3261         spin_unlock_bh(&adapter->mbx_lock);
3262
3263         /* if check for link returns error we will need to reset */
3264         if (err && time_after(jiffies, adapter->last_reset + (10 * HZ))) {
3265                 set_bit(__IXGBEVF_RESET_REQUESTED, &adapter->state);
3266                 link_up = false;
3267         }
3268
3269         adapter->link_up = link_up;
3270         adapter->link_speed = link_speed;
3271 }
3272
3273 /**
3274  * ixgbevf_watchdog_link_is_up - update netif_carrier status and
3275  *                               print link up message
3276  * @adapter: pointer to the device adapter structure
3277  **/
3278 static void ixgbevf_watchdog_link_is_up(struct ixgbevf_adapter *adapter)
3279 {
3280         struct net_device *netdev = adapter->netdev;
3281
3282         /* only continue if link was previously down */
3283         if (netif_carrier_ok(netdev))
3284                 return;
3285
3286         dev_info(&adapter->pdev->dev, "NIC Link is Up %s\n",
3287                  (adapter->link_speed == IXGBE_LINK_SPEED_10GB_FULL) ?
3288                  "10 Gbps" :
3289                  (adapter->link_speed == IXGBE_LINK_SPEED_1GB_FULL) ?
3290                  "1 Gbps" :
3291                  (adapter->link_speed == IXGBE_LINK_SPEED_100_FULL) ?
3292                  "100 Mbps" :
3293                  "unknown speed");
3294
3295         netif_carrier_on(netdev);
3296 }
3297
3298 /**
3299  * ixgbevf_watchdog_link_is_down - update netif_carrier status and
3300  *                                 print link down message
3301  * @adapter: pointer to the adapter structure
3302  **/
3303 static void ixgbevf_watchdog_link_is_down(struct ixgbevf_adapter *adapter)
3304 {
3305         struct net_device *netdev = adapter->netdev;
3306
3307         adapter->link_speed = 0;
3308
3309         /* only continue if link was up previously */
3310         if (!netif_carrier_ok(netdev))
3311                 return;
3312
3313         dev_info(&adapter->pdev->dev, "NIC Link is Down\n");
3314
3315         netif_carrier_off(netdev);
3316 }
3317
3318 /**
3319  * ixgbevf_watchdog_subtask - worker thread to bring link up
3320  * @adapter: board private structure
3321  **/
3322 static void ixgbevf_watchdog_subtask(struct ixgbevf_adapter *adapter)
3323 {
3324         /* if interface is down do nothing */
3325         if (test_bit(__IXGBEVF_DOWN, &adapter->state) ||
3326             test_bit(__IXGBEVF_RESETTING, &adapter->state))
3327                 return;
3328
3329         ixgbevf_watchdog_update_link(adapter);
3330
3331         if (adapter->link_up && adapter->link_state)
3332                 ixgbevf_watchdog_link_is_up(adapter);
3333         else
3334                 ixgbevf_watchdog_link_is_down(adapter);
3335
3336         ixgbevf_update_stats(adapter);
3337 }
3338
3339 /**
3340  * ixgbevf_service_task - manages and runs subtasks
3341  * @work: pointer to work_struct containing our data
3342  **/
3343 static void ixgbevf_service_task(struct work_struct *work)
3344 {
3345         struct ixgbevf_adapter *adapter = container_of(work,
3346                                                        struct ixgbevf_adapter,
3347                                                        service_task);
3348         struct ixgbe_hw *hw = &adapter->hw;
3349
3350         if (IXGBE_REMOVED(hw->hw_addr)) {
3351                 if (!test_bit(__IXGBEVF_DOWN, &adapter->state)) {
3352                         rtnl_lock();
3353                         ixgbevf_down(adapter);
3354                         rtnl_unlock();
3355                 }
3356                 return;
3357         }
3358
3359         ixgbevf_queue_reset_subtask(adapter);
3360         ixgbevf_reset_subtask(adapter);
3361         ixgbevf_watchdog_subtask(adapter);
3362         ixgbevf_check_hang_subtask(adapter);
3363
3364         ixgbevf_service_event_complete(adapter);
3365 }
3366
3367 /**
3368  * ixgbevf_free_tx_resources - Free Tx Resources per Queue
3369  * @tx_ring: Tx descriptor ring for a specific queue
3370  *
3371  * Free all transmit software resources
3372  **/
3373 void ixgbevf_free_tx_resources(struct ixgbevf_ring *tx_ring)
3374 {
3375         ixgbevf_clean_tx_ring(tx_ring);
3376
3377         vfree(tx_ring->tx_buffer_info);
3378         tx_ring->tx_buffer_info = NULL;
3379
3380         /* if not set, then don't free */
3381         if (!tx_ring->desc)
3382                 return;
3383
3384         dma_free_coherent(tx_ring->dev, tx_ring->size, tx_ring->desc,
3385                           tx_ring->dma);
3386
3387         tx_ring->desc = NULL;
3388 }
3389
3390 /**
3391  * ixgbevf_free_all_tx_resources - Free Tx Resources for All Queues
3392  * @adapter: board private structure
3393  *
3394  * Free all transmit software resources
3395  **/
3396 static void ixgbevf_free_all_tx_resources(struct ixgbevf_adapter *adapter)
3397 {
3398         int i;
3399
3400         for (i = 0; i < adapter->num_tx_queues; i++)
3401                 if (adapter->tx_ring[i]->desc)
3402                         ixgbevf_free_tx_resources(adapter->tx_ring[i]);
3403         for (i = 0; i < adapter->num_xdp_queues; i++)
3404                 if (adapter->xdp_ring[i]->desc)
3405                         ixgbevf_free_tx_resources(adapter->xdp_ring[i]);
3406 }
3407
3408 /**
3409  * ixgbevf_setup_tx_resources - allocate Tx resources (Descriptors)
3410  * @tx_ring: Tx descriptor ring (for a specific queue) to setup
3411  *
3412  * Return 0 on success, negative on failure
3413  **/
3414 int ixgbevf_setup_tx_resources(struct ixgbevf_ring *tx_ring)
3415 {
3416         struct ixgbevf_adapter *adapter = netdev_priv(tx_ring->netdev);
3417         int size;
3418
3419         size = sizeof(struct ixgbevf_tx_buffer) * tx_ring->count;
3420         tx_ring->tx_buffer_info = vmalloc(size);
3421         if (!tx_ring->tx_buffer_info)
3422                 goto err;
3423
3424         u64_stats_init(&tx_ring->syncp);
3425
3426         /* round up to nearest 4K */
3427         tx_ring->size = tx_ring->count * sizeof(union ixgbe_adv_tx_desc);
3428         tx_ring->size = ALIGN(tx_ring->size, 4096);
3429
3430         tx_ring->desc = dma_alloc_coherent(tx_ring->dev, tx_ring->size,
3431                                            &tx_ring->dma, GFP_KERNEL);
3432         if (!tx_ring->desc)
3433                 goto err;
3434
3435         return 0;
3436
3437 err:
3438         vfree(tx_ring->tx_buffer_info);
3439         tx_ring->tx_buffer_info = NULL;
3440         hw_dbg(&adapter->hw, "Unable to allocate memory for the transmit descriptor ring\n");
3441         return -ENOMEM;
3442 }
3443
3444 /**
3445  * ixgbevf_setup_all_tx_resources - allocate all queues Tx resources
3446  * @adapter: board private structure
3447  *
3448  * If this function returns with an error, then it's possible one or
3449  * more of the rings is populated (while the rest are not).  It is the
3450  * callers duty to clean those orphaned rings.
3451  *
3452  * Return 0 on success, negative on failure
3453  **/
3454 static int ixgbevf_setup_all_tx_resources(struct ixgbevf_adapter *adapter)
3455 {
3456         int i, j = 0, err = 0;
3457
3458         for (i = 0; i < adapter->num_tx_queues; i++) {
3459                 err = ixgbevf_setup_tx_resources(adapter->tx_ring[i]);
3460                 if (!err)
3461                         continue;
3462                 hw_dbg(&adapter->hw, "Allocation for Tx Queue %u failed\n", i);
3463                 goto err_setup_tx;
3464         }
3465
3466         for (j = 0; j < adapter->num_xdp_queues; j++) {
3467                 err = ixgbevf_setup_tx_resources(adapter->xdp_ring[j]);
3468                 if (!err)
3469                         continue;
3470                 hw_dbg(&adapter->hw, "Allocation for XDP Queue %u failed\n", j);
3471                 goto err_setup_tx;
3472         }
3473
3474         return 0;
3475 err_setup_tx:
3476         /* rewind the index freeing the rings as we go */
3477         while (j--)
3478                 ixgbevf_free_tx_resources(adapter->xdp_ring[j]);
3479         while (i--)
3480                 ixgbevf_free_tx_resources(adapter->tx_ring[i]);
3481
3482         return err;
3483 }
3484
3485 /**
3486  * ixgbevf_setup_rx_resources - allocate Rx resources (Descriptors)
3487  * @adapter: board private structure
3488  * @rx_ring: Rx descriptor ring (for a specific queue) to setup
3489  *
3490  * Returns 0 on success, negative on failure
3491  **/
3492 int ixgbevf_setup_rx_resources(struct ixgbevf_adapter *adapter,
3493                                struct ixgbevf_ring *rx_ring)
3494 {
3495         int size;
3496
3497         size = sizeof(struct ixgbevf_rx_buffer) * rx_ring->count;
3498         rx_ring->rx_buffer_info = vmalloc(size);
3499         if (!rx_ring->rx_buffer_info)
3500                 goto err;
3501
3502         u64_stats_init(&rx_ring->syncp);
3503
3504         /* Round up to nearest 4K */
3505         rx_ring->size = rx_ring->count * sizeof(union ixgbe_adv_rx_desc);
3506         rx_ring->size = ALIGN(rx_ring->size, 4096);
3507
3508         rx_ring->desc = dma_alloc_coherent(rx_ring->dev, rx_ring->size,
3509                                            &rx_ring->dma, GFP_KERNEL);
3510
3511         if (!rx_ring->desc)
3512                 goto err;
3513
3514         /* XDP RX-queue info */
3515         if (xdp_rxq_info_reg(&rx_ring->xdp_rxq, adapter->netdev,
3516                              rx_ring->queue_index, 0) < 0)
3517                 goto err;
3518
3519         rx_ring->xdp_prog = adapter->xdp_prog;
3520
3521         return 0;
3522 err:
3523         vfree(rx_ring->rx_buffer_info);
3524         rx_ring->rx_buffer_info = NULL;
3525         dev_err(rx_ring->dev, "Unable to allocate memory for the Rx descriptor ring\n");
3526         return -ENOMEM;
3527 }
3528
3529 /**
3530  * ixgbevf_setup_all_rx_resources - allocate all queues Rx resources
3531  * @adapter: board private structure
3532  *
3533  * If this function returns with an error, then it's possible one or
3534  * more of the rings is populated (while the rest are not).  It is the
3535  * callers duty to clean those orphaned rings.
3536  *
3537  * Return 0 on success, negative on failure
3538  **/
3539 static int ixgbevf_setup_all_rx_resources(struct ixgbevf_adapter *adapter)
3540 {
3541         int i, err = 0;
3542
3543         for (i = 0; i < adapter->num_rx_queues; i++) {
3544                 err = ixgbevf_setup_rx_resources(adapter, adapter->rx_ring[i]);
3545                 if (!err)
3546                         continue;
3547                 hw_dbg(&adapter->hw, "Allocation for Rx Queue %u failed\n", i);
3548                 goto err_setup_rx;
3549         }
3550
3551         return 0;
3552 err_setup_rx:
3553         /* rewind the index freeing the rings as we go */
3554         while (i--)
3555                 ixgbevf_free_rx_resources(adapter->rx_ring[i]);
3556         return err;
3557 }
3558
3559 /**
3560  * ixgbevf_free_rx_resources - Free Rx Resources
3561  * @rx_ring: ring to clean the resources from
3562  *
3563  * Free all receive software resources
3564  **/
3565 void ixgbevf_free_rx_resources(struct ixgbevf_ring *rx_ring)
3566 {
3567         ixgbevf_clean_rx_ring(rx_ring);
3568
3569         rx_ring->xdp_prog = NULL;
3570         xdp_rxq_info_unreg(&rx_ring->xdp_rxq);
3571         vfree(rx_ring->rx_buffer_info);
3572         rx_ring->rx_buffer_info = NULL;
3573
3574         dma_free_coherent(rx_ring->dev, rx_ring->size, rx_ring->desc,
3575                           rx_ring->dma);
3576
3577         rx_ring->desc = NULL;
3578 }
3579
3580 /**
3581  * ixgbevf_free_all_rx_resources - Free Rx Resources for All Queues
3582  * @adapter: board private structure
3583  *
3584  * Free all receive software resources
3585  **/
3586 static void ixgbevf_free_all_rx_resources(struct ixgbevf_adapter *adapter)
3587 {
3588         int i;
3589
3590         for (i = 0; i < adapter->num_rx_queues; i++)
3591                 if (adapter->rx_ring[i]->desc)
3592                         ixgbevf_free_rx_resources(adapter->rx_ring[i]);
3593 }
3594
3595 /**
3596  * ixgbevf_open - Called when a network interface is made active
3597  * @netdev: network interface device structure
3598  *
3599  * Returns 0 on success, negative value on failure
3600  *
3601  * The open entry point is called when a network interface is made
3602  * active by the system (IFF_UP).  At this point all resources needed
3603  * for transmit and receive operations are allocated, the interrupt
3604  * handler is registered with the OS, the watchdog timer is started,
3605  * and the stack is notified that the interface is ready.
3606  **/
3607 int ixgbevf_open(struct net_device *netdev)
3608 {
3609         struct ixgbevf_adapter *adapter = netdev_priv(netdev);
3610         struct ixgbe_hw *hw = &adapter->hw;
3611         int err;
3612
3613         /* A previous failure to open the device because of a lack of
3614          * available MSIX vector resources may have reset the number
3615          * of msix vectors variable to zero.  The only way to recover
3616          * is to unload/reload the driver and hope that the system has
3617          * been able to recover some MSIX vector resources.
3618          */
3619         if (!adapter->num_msix_vectors)
3620                 return -ENOMEM;
3621
3622         if (hw->adapter_stopped) {
3623                 ixgbevf_reset(adapter);
3624                 /* if adapter is still stopped then PF isn't up and
3625                  * the VF can't start.
3626                  */
3627                 if (hw->adapter_stopped) {
3628                         err = IXGBE_ERR_MBX;
3629                         pr_err("Unable to start - perhaps the PF Driver isn't up yet\n");
3630                         goto err_setup_reset;
3631                 }
3632         }
3633
3634         /* disallow open during test */
3635         if (test_bit(__IXGBEVF_TESTING, &adapter->state))
3636                 return -EBUSY;
3637
3638         netif_carrier_off(netdev);
3639
3640         /* allocate transmit descriptors */
3641         err = ixgbevf_setup_all_tx_resources(adapter);
3642         if (err)
3643                 goto err_setup_tx;
3644
3645         /* allocate receive descriptors */
3646         err = ixgbevf_setup_all_rx_resources(adapter);
3647         if (err)
3648                 goto err_setup_rx;
3649
3650         ixgbevf_configure(adapter);
3651
3652         err = ixgbevf_request_irq(adapter);
3653         if (err)
3654                 goto err_req_irq;
3655
3656         /* Notify the stack of the actual queue counts. */
3657         err = netif_set_real_num_tx_queues(netdev, adapter->num_tx_queues);
3658         if (err)
3659                 goto err_set_queues;
3660
3661         err = netif_set_real_num_rx_queues(netdev, adapter->num_rx_queues);
3662         if (err)
3663                 goto err_set_queues;
3664
3665         ixgbevf_up_complete(adapter);
3666
3667         return 0;
3668
3669 err_set_queues:
3670         ixgbevf_free_irq(adapter);
3671 err_req_irq:
3672         ixgbevf_free_all_rx_resources(adapter);
3673 err_setup_rx:
3674         ixgbevf_free_all_tx_resources(adapter);
3675 err_setup_tx:
3676         ixgbevf_reset(adapter);
3677 err_setup_reset:
3678
3679         return err;
3680 }
3681
3682 /**
3683  * ixgbevf_close_suspend - actions necessary to both suspend and close flows
3684  * @adapter: the private adapter struct
3685  *
3686  * This function should contain the necessary work common to both suspending
3687  * and closing of the device.
3688  */
3689 static void ixgbevf_close_suspend(struct ixgbevf_adapter *adapter)
3690 {
3691         ixgbevf_down(adapter);
3692         ixgbevf_free_irq(adapter);
3693         ixgbevf_free_all_tx_resources(adapter);
3694         ixgbevf_free_all_rx_resources(adapter);
3695 }
3696
3697 /**
3698  * ixgbevf_close - Disables a network interface
3699  * @netdev: network interface device structure
3700  *
3701  * Returns 0, this is not allowed to fail
3702  *
3703  * The close entry point is called when an interface is de-activated
3704  * by the OS.  The hardware is still under the drivers control, but
3705  * needs to be disabled.  A global MAC reset is issued to stop the
3706  * hardware, and all transmit and receive resources are freed.
3707  **/
3708 int ixgbevf_close(struct net_device *netdev)
3709 {
3710         struct ixgbevf_adapter *adapter = netdev_priv(netdev);
3711
3712         if (netif_device_present(netdev))
3713                 ixgbevf_close_suspend(adapter);
3714
3715         return 0;
3716 }
3717
3718 static void ixgbevf_queue_reset_subtask(struct ixgbevf_adapter *adapter)
3719 {
3720         struct net_device *dev = adapter->netdev;
3721
3722         if (!test_and_clear_bit(__IXGBEVF_QUEUE_RESET_REQUESTED,
3723                                 &adapter->state))
3724                 return;
3725
3726         /* if interface is down do nothing */
3727         if (test_bit(__IXGBEVF_DOWN, &adapter->state) ||
3728             test_bit(__IXGBEVF_RESETTING, &adapter->state))
3729                 return;
3730
3731         /* Hardware has to reinitialize queues and interrupts to
3732          * match packet buffer alignment. Unfortunately, the
3733          * hardware is not flexible enough to do this dynamically.
3734          */
3735         rtnl_lock();
3736
3737         if (netif_running(dev))
3738                 ixgbevf_close(dev);
3739
3740         ixgbevf_clear_interrupt_scheme(adapter);
3741         ixgbevf_init_interrupt_scheme(adapter);
3742
3743         if (netif_running(dev))
3744                 ixgbevf_open(dev);
3745
3746         rtnl_unlock();
3747 }
3748
3749 static void ixgbevf_tx_ctxtdesc(struct ixgbevf_ring *tx_ring,
3750                                 u32 vlan_macip_lens, u32 fceof_saidx,
3751                                 u32 type_tucmd, u32 mss_l4len_idx)
3752 {
3753         struct ixgbe_adv_tx_context_desc *context_desc;
3754         u16 i = tx_ring->next_to_use;
3755
3756         context_desc = IXGBEVF_TX_CTXTDESC(tx_ring, i);
3757
3758         i++;
3759         tx_ring->next_to_use = (i < tx_ring->count) ? i : 0;
3760
3761         /* set bits to identify this as an advanced context descriptor */
3762         type_tucmd |= IXGBE_TXD_CMD_DEXT | IXGBE_ADVTXD_DTYP_CTXT;
3763
3764         context_desc->vlan_macip_lens   = cpu_to_le32(vlan_macip_lens);
3765         context_desc->fceof_saidx       = cpu_to_le32(fceof_saidx);
3766         context_desc->type_tucmd_mlhl   = cpu_to_le32(type_tucmd);
3767         context_desc->mss_l4len_idx     = cpu_to_le32(mss_l4len_idx);
3768 }
3769
3770 static int ixgbevf_tso(struct ixgbevf_ring *tx_ring,
3771                        struct ixgbevf_tx_buffer *first,
3772                        u8 *hdr_len,
3773                        struct ixgbevf_ipsec_tx_data *itd)
3774 {
3775         u32 vlan_macip_lens, type_tucmd, mss_l4len_idx;
3776         struct sk_buff *skb = first->skb;
3777         union {
3778                 struct iphdr *v4;
3779                 struct ipv6hdr *v6;
3780                 unsigned char *hdr;
3781         } ip;
3782         union {
3783                 struct tcphdr *tcp;
3784                 unsigned char *hdr;
3785         } l4;
3786         u32 paylen, l4_offset;
3787         u32 fceof_saidx = 0;
3788         int err;
3789
3790         if (skb->ip_summed != CHECKSUM_PARTIAL)
3791                 return 0;
3792
3793         if (!skb_is_gso(skb))
3794                 return 0;
3795
3796         err = skb_cow_head(skb, 0);
3797         if (err < 0)
3798                 return err;
3799
3800         if (eth_p_mpls(first->protocol))
3801                 ip.hdr = skb_inner_network_header(skb);
3802         else
3803                 ip.hdr = skb_network_header(skb);
3804         l4.hdr = skb_checksum_start(skb);
3805
3806         /* ADV DTYP TUCMD MKRLOC/ISCSIHEDLEN */
3807         type_tucmd = IXGBE_ADVTXD_TUCMD_L4T_TCP;
3808
3809         /* initialize outer IP header fields */
3810         if (ip.v4->version == 4) {
3811                 unsigned char *csum_start = skb_checksum_start(skb);
3812                 unsigned char *trans_start = ip.hdr + (ip.v4->ihl * 4);
3813                 int len = csum_start - trans_start;
3814
3815                 /* IP header will have to cancel out any data that
3816                  * is not a part of the outer IP header, so set to
3817                  * a reverse csum if needed, else init check to 0.
3818                  */
3819                 ip.v4->check = (skb_shinfo(skb)->gso_type & SKB_GSO_PARTIAL) ?
3820                                            csum_fold(csum_partial(trans_start,
3821                                                                   len, 0)) : 0;
3822                 type_tucmd |= IXGBE_ADVTXD_TUCMD_IPV4;
3823
3824                 ip.v4->tot_len = 0;
3825                 first->tx_flags |= IXGBE_TX_FLAGS_TSO |
3826                                    IXGBE_TX_FLAGS_CSUM |
3827                                    IXGBE_TX_FLAGS_IPV4;
3828         } else {
3829                 ip.v6->payload_len = 0;
3830                 first->tx_flags |= IXGBE_TX_FLAGS_TSO |
3831                                    IXGBE_TX_FLAGS_CSUM;
3832         }
3833
3834         /* determine offset of inner transport header */
3835         l4_offset = l4.hdr - skb->data;
3836
3837         /* compute length of segmentation header */
3838         *hdr_len = (l4.tcp->doff * 4) + l4_offset;
3839
3840         /* remove payload length from inner checksum */
3841         paylen = skb->len - l4_offset;
3842         csum_replace_by_diff(&l4.tcp->check, (__force __wsum)htonl(paylen));
3843
3844         /* update gso size and bytecount with header size */
3845         first->gso_segs = skb_shinfo(skb)->gso_segs;
3846         first->bytecount += (first->gso_segs - 1) * *hdr_len;
3847
3848         /* mss_l4len_id: use 1 as index for TSO */
3849         mss_l4len_idx = (*hdr_len - l4_offset) << IXGBE_ADVTXD_L4LEN_SHIFT;
3850         mss_l4len_idx |= skb_shinfo(skb)->gso_size << IXGBE_ADVTXD_MSS_SHIFT;
3851         mss_l4len_idx |= (1u << IXGBE_ADVTXD_IDX_SHIFT);
3852
3853         fceof_saidx |= itd->pfsa;
3854         type_tucmd |= itd->flags | itd->trailer_len;
3855
3856         /* vlan_macip_lens: HEADLEN, MACLEN, VLAN tag */
3857         vlan_macip_lens = l4.hdr - ip.hdr;
3858         vlan_macip_lens |= (ip.hdr - skb->data) << IXGBE_ADVTXD_MACLEN_SHIFT;
3859         vlan_macip_lens |= first->tx_flags & IXGBE_TX_FLAGS_VLAN_MASK;
3860
3861         ixgbevf_tx_ctxtdesc(tx_ring, vlan_macip_lens, fceof_saidx, type_tucmd,
3862                             mss_l4len_idx);
3863
3864         return 1;
3865 }
3866
3867 static void ixgbevf_tx_csum(struct ixgbevf_ring *tx_ring,
3868                             struct ixgbevf_tx_buffer *first,
3869                             struct ixgbevf_ipsec_tx_data *itd)
3870 {
3871         struct sk_buff *skb = first->skb;
3872         u32 vlan_macip_lens = 0;
3873         u32 fceof_saidx = 0;
3874         u32 type_tucmd = 0;
3875
3876         if (skb->ip_summed != CHECKSUM_PARTIAL)
3877                 goto no_csum;
3878
3879         switch (skb->csum_offset) {
3880         case offsetof(struct tcphdr, check):
3881                 type_tucmd = IXGBE_ADVTXD_TUCMD_L4T_TCP;
3882                 fallthrough;
3883         case offsetof(struct udphdr, check):
3884                 break;
3885         case offsetof(struct sctphdr, checksum):
3886                 /* validate that this is actually an SCTP request */
3887                 if (skb_csum_is_sctp(skb)) {
3888                         type_tucmd = IXGBE_ADVTXD_TUCMD_L4T_SCTP;
3889                         break;
3890                 }
3891                 fallthrough;
3892         default:
3893                 skb_checksum_help(skb);
3894                 goto no_csum;
3895         }
3896
3897         if (first->protocol == htons(ETH_P_IP))
3898                 type_tucmd |= IXGBE_ADVTXD_TUCMD_IPV4;
3899
3900         /* update TX checksum flag */
3901         first->tx_flags |= IXGBE_TX_FLAGS_CSUM;
3902         vlan_macip_lens = skb_checksum_start_offset(skb) -
3903                           skb_network_offset(skb);
3904 no_csum:
3905         /* vlan_macip_lens: MACLEN, VLAN tag */
3906         vlan_macip_lens |= skb_network_offset(skb) << IXGBE_ADVTXD_MACLEN_SHIFT;
3907         vlan_macip_lens |= first->tx_flags & IXGBE_TX_FLAGS_VLAN_MASK;
3908
3909         fceof_saidx |= itd->pfsa;
3910         type_tucmd |= itd->flags | itd->trailer_len;
3911
3912         ixgbevf_tx_ctxtdesc(tx_ring, vlan_macip_lens,
3913                             fceof_saidx, type_tucmd, 0);
3914 }
3915
3916 static __le32 ixgbevf_tx_cmd_type(u32 tx_flags)
3917 {
3918         /* set type for advanced descriptor with frame checksum insertion */
3919         __le32 cmd_type = cpu_to_le32(IXGBE_ADVTXD_DTYP_DATA |
3920                                       IXGBE_ADVTXD_DCMD_IFCS |
3921                                       IXGBE_ADVTXD_DCMD_DEXT);
3922
3923         /* set HW VLAN bit if VLAN is present */
3924         if (tx_flags & IXGBE_TX_FLAGS_VLAN)
3925                 cmd_type |= cpu_to_le32(IXGBE_ADVTXD_DCMD_VLE);
3926
3927         /* set segmentation enable bits for TSO/FSO */
3928         if (tx_flags & IXGBE_TX_FLAGS_TSO)
3929                 cmd_type |= cpu_to_le32(IXGBE_ADVTXD_DCMD_TSE);
3930
3931         return cmd_type;
3932 }
3933
3934 static void ixgbevf_tx_olinfo_status(union ixgbe_adv_tx_desc *tx_desc,
3935                                      u32 tx_flags, unsigned int paylen)
3936 {
3937         __le32 olinfo_status = cpu_to_le32(paylen << IXGBE_ADVTXD_PAYLEN_SHIFT);
3938
3939         /* enable L4 checksum for TSO and TX checksum offload */
3940         if (tx_flags & IXGBE_TX_FLAGS_CSUM)
3941                 olinfo_status |= cpu_to_le32(IXGBE_ADVTXD_POPTS_TXSM);
3942
3943         /* enble IPv4 checksum for TSO */
3944         if (tx_flags & IXGBE_TX_FLAGS_IPV4)
3945                 olinfo_status |= cpu_to_le32(IXGBE_ADVTXD_POPTS_IXSM);
3946
3947         /* enable IPsec */
3948         if (tx_flags & IXGBE_TX_FLAGS_IPSEC)
3949                 olinfo_status |= cpu_to_le32(IXGBE_ADVTXD_POPTS_IPSEC);
3950
3951         /* use index 1 context for TSO/FSO/FCOE/IPSEC */
3952         if (tx_flags & (IXGBE_TX_FLAGS_TSO | IXGBE_TX_FLAGS_IPSEC))
3953                 olinfo_status |= cpu_to_le32(1u << IXGBE_ADVTXD_IDX_SHIFT);
3954
3955         /* Check Context must be set if Tx switch is enabled, which it
3956          * always is for case where virtual functions are running
3957          */
3958         olinfo_status |= cpu_to_le32(IXGBE_ADVTXD_CC);
3959
3960         tx_desc->read.olinfo_status = olinfo_status;
3961 }
3962
3963 static void ixgbevf_tx_map(struct ixgbevf_ring *tx_ring,
3964                            struct ixgbevf_tx_buffer *first,
3965                            const u8 hdr_len)
3966 {
3967         struct sk_buff *skb = first->skb;
3968         struct ixgbevf_tx_buffer *tx_buffer;
3969         union ixgbe_adv_tx_desc *tx_desc;
3970         skb_frag_t *frag;
3971         dma_addr_t dma;
3972         unsigned int data_len, size;
3973         u32 tx_flags = first->tx_flags;
3974         __le32 cmd_type = ixgbevf_tx_cmd_type(tx_flags);
3975         u16 i = tx_ring->next_to_use;
3976
3977         tx_desc = IXGBEVF_TX_DESC(tx_ring, i);
3978
3979         ixgbevf_tx_olinfo_status(tx_desc, tx_flags, skb->len - hdr_len);
3980
3981         size = skb_headlen(skb);
3982         data_len = skb->data_len;
3983
3984         dma = dma_map_single(tx_ring->dev, skb->data, size, DMA_TO_DEVICE);
3985
3986         tx_buffer = first;
3987
3988         for (frag = &skb_shinfo(skb)->frags[0];; frag++) {
3989                 if (dma_mapping_error(tx_ring->dev, dma))
3990                         goto dma_error;
3991
3992                 /* record length, and DMA address */
3993                 dma_unmap_len_set(tx_buffer, len, size);
3994                 dma_unmap_addr_set(tx_buffer, dma, dma);
3995
3996                 tx_desc->read.buffer_addr = cpu_to_le64(dma);
3997
3998                 while (unlikely(size > IXGBE_MAX_DATA_PER_TXD)) {
3999                         tx_desc->read.cmd_type_len =
4000                                 cmd_type | cpu_to_le32(IXGBE_MAX_DATA_PER_TXD);
4001
4002                         i++;
4003                         tx_desc++;
4004                         if (i == tx_ring->count) {
4005                                 tx_desc = IXGBEVF_TX_DESC(tx_ring, 0);
4006                                 i = 0;
4007                         }
4008                         tx_desc->read.olinfo_status = 0;
4009
4010                         dma += IXGBE_MAX_DATA_PER_TXD;
4011                         size -= IXGBE_MAX_DATA_PER_TXD;
4012
4013                         tx_desc->read.buffer_addr = cpu_to_le64(dma);
4014                 }
4015
4016                 if (likely(!data_len))
4017                         break;
4018
4019                 tx_desc->read.cmd_type_len = cmd_type | cpu_to_le32(size);
4020
4021                 i++;
4022                 tx_desc++;
4023                 if (i == tx_ring->count) {
4024                         tx_desc = IXGBEVF_TX_DESC(tx_ring, 0);
4025                         i = 0;
4026                 }
4027                 tx_desc->read.olinfo_status = 0;
4028
4029                 size = skb_frag_size(frag);
4030                 data_len -= size;
4031
4032                 dma = skb_frag_dma_map(tx_ring->dev, frag, 0, size,
4033                                        DMA_TO_DEVICE);
4034
4035                 tx_buffer = &tx_ring->tx_buffer_info[i];
4036         }
4037
4038         /* write last descriptor with RS and EOP bits */
4039         cmd_type |= cpu_to_le32(size) | cpu_to_le32(IXGBE_TXD_CMD);
4040         tx_desc->read.cmd_type_len = cmd_type;
4041
4042         /* set the timestamp */
4043         first->time_stamp = jiffies;
4044
4045         skb_tx_timestamp(skb);
4046
4047         /* Force memory writes to complete before letting h/w know there
4048          * are new descriptors to fetch.  (Only applicable for weak-ordered
4049          * memory model archs, such as IA-64).
4050          *
4051          * We also need this memory barrier (wmb) to make certain all of the
4052          * status bits have been updated before next_to_watch is written.
4053          */
4054         wmb();
4055
4056         /* set next_to_watch value indicating a packet is present */
4057         first->next_to_watch = tx_desc;
4058
4059         i++;
4060         if (i == tx_ring->count)
4061                 i = 0;
4062
4063         tx_ring->next_to_use = i;
4064
4065         /* notify HW of packet */
4066         ixgbevf_write_tail(tx_ring, i);
4067
4068         return;
4069 dma_error:
4070         dev_err(tx_ring->dev, "TX DMA map failed\n");
4071         tx_buffer = &tx_ring->tx_buffer_info[i];
4072
4073         /* clear dma mappings for failed tx_buffer_info map */
4074         while (tx_buffer != first) {
4075                 if (dma_unmap_len(tx_buffer, len))
4076                         dma_unmap_page(tx_ring->dev,
4077                                        dma_unmap_addr(tx_buffer, dma),
4078                                        dma_unmap_len(tx_buffer, len),
4079                                        DMA_TO_DEVICE);
4080                 dma_unmap_len_set(tx_buffer, len, 0);
4081
4082                 if (i-- == 0)
4083                         i += tx_ring->count;
4084                 tx_buffer = &tx_ring->tx_buffer_info[i];
4085         }
4086
4087         if (dma_unmap_len(tx_buffer, len))
4088                 dma_unmap_single(tx_ring->dev,
4089                                  dma_unmap_addr(tx_buffer, dma),
4090                                  dma_unmap_len(tx_buffer, len),
4091                                  DMA_TO_DEVICE);
4092         dma_unmap_len_set(tx_buffer, len, 0);
4093
4094         dev_kfree_skb_any(tx_buffer->skb);
4095         tx_buffer->skb = NULL;
4096
4097         tx_ring->next_to_use = i;
4098 }
4099
4100 static int __ixgbevf_maybe_stop_tx(struct ixgbevf_ring *tx_ring, int size)
4101 {
4102         netif_stop_subqueue(tx_ring->netdev, tx_ring->queue_index);
4103         /* Herbert's original patch had:
4104          *  smp_mb__after_netif_stop_queue();
4105          * but since that doesn't exist yet, just open code it.
4106          */
4107         smp_mb();
4108
4109         /* We need to check again in a case another CPU has just
4110          * made room available.
4111          */
4112         if (likely(ixgbevf_desc_unused(tx_ring) < size))
4113                 return -EBUSY;
4114
4115         /* A reprieve! - use start_queue because it doesn't call schedule */
4116         netif_start_subqueue(tx_ring->netdev, tx_ring->queue_index);
4117         ++tx_ring->tx_stats.restart_queue;
4118
4119         return 0;
4120 }
4121
4122 static int ixgbevf_maybe_stop_tx(struct ixgbevf_ring *tx_ring, int size)
4123 {
4124         if (likely(ixgbevf_desc_unused(tx_ring) >= size))
4125                 return 0;
4126         return __ixgbevf_maybe_stop_tx(tx_ring, size);
4127 }
4128
4129 static int ixgbevf_xmit_frame_ring(struct sk_buff *skb,
4130                                    struct ixgbevf_ring *tx_ring)
4131 {
4132         struct ixgbevf_tx_buffer *first;
4133         int tso;
4134         u32 tx_flags = 0;
4135         u16 count = TXD_USE_COUNT(skb_headlen(skb));
4136         struct ixgbevf_ipsec_tx_data ipsec_tx = { 0 };
4137 #if PAGE_SIZE > IXGBE_MAX_DATA_PER_TXD
4138         unsigned short f;
4139 #endif
4140         u8 hdr_len = 0;
4141         u8 *dst_mac = skb_header_pointer(skb, 0, 0, NULL);
4142
4143         if (!dst_mac || is_link_local_ether_addr(dst_mac)) {
4144                 dev_kfree_skb_any(skb);
4145                 return NETDEV_TX_OK;
4146         }
4147
4148         /* need: 1 descriptor per page * PAGE_SIZE/IXGBE_MAX_DATA_PER_TXD,
4149          *       + 1 desc for skb_headlen/IXGBE_MAX_DATA_PER_TXD,
4150          *       + 2 desc gap to keep tail from touching head,
4151          *       + 1 desc for context descriptor,
4152          * otherwise try next time
4153          */
4154 #if PAGE_SIZE > IXGBE_MAX_DATA_PER_TXD
4155         for (f = 0; f < skb_shinfo(skb)->nr_frags; f++) {
4156                 skb_frag_t *frag = &skb_shinfo(skb)->frags[f];
4157
4158                 count += TXD_USE_COUNT(skb_frag_size(frag));
4159         }
4160 #else
4161         count += skb_shinfo(skb)->nr_frags;
4162 #endif
4163         if (ixgbevf_maybe_stop_tx(tx_ring, count + 3)) {
4164                 tx_ring->tx_stats.tx_busy++;
4165                 return NETDEV_TX_BUSY;
4166         }
4167
4168         /* record the location of the first descriptor for this packet */
4169         first = &tx_ring->tx_buffer_info[tx_ring->next_to_use];
4170         first->skb = skb;
4171         first->bytecount = skb->len;
4172         first->gso_segs = 1;
4173
4174         if (skb_vlan_tag_present(skb)) {
4175                 tx_flags |= skb_vlan_tag_get(skb);
4176                 tx_flags <<= IXGBE_TX_FLAGS_VLAN_SHIFT;
4177                 tx_flags |= IXGBE_TX_FLAGS_VLAN;
4178         }
4179
4180         /* record initial flags and protocol */
4181         first->tx_flags = tx_flags;
4182         first->protocol = vlan_get_protocol(skb);
4183
4184 #ifdef CONFIG_IXGBEVF_IPSEC
4185         if (xfrm_offload(skb) && !ixgbevf_ipsec_tx(tx_ring, first, &ipsec_tx))
4186                 goto out_drop;
4187 #endif
4188         tso = ixgbevf_tso(tx_ring, first, &hdr_len, &ipsec_tx);
4189         if (tso < 0)
4190                 goto out_drop;
4191         else if (!tso)
4192                 ixgbevf_tx_csum(tx_ring, first, &ipsec_tx);
4193
4194         ixgbevf_tx_map(tx_ring, first, hdr_len);
4195
4196         ixgbevf_maybe_stop_tx(tx_ring, DESC_NEEDED);
4197
4198         return NETDEV_TX_OK;
4199
4200 out_drop:
4201         dev_kfree_skb_any(first->skb);
4202         first->skb = NULL;
4203
4204         return NETDEV_TX_OK;
4205 }
4206
4207 static netdev_tx_t ixgbevf_xmit_frame(struct sk_buff *skb, struct net_device *netdev)
4208 {
4209         struct ixgbevf_adapter *adapter = netdev_priv(netdev);
4210         struct ixgbevf_ring *tx_ring;
4211
4212         if (skb->len <= 0) {
4213                 dev_kfree_skb_any(skb);
4214                 return NETDEV_TX_OK;
4215         }
4216
4217         /* The minimum packet size for olinfo paylen is 17 so pad the skb
4218          * in order to meet this minimum size requirement.
4219          */
4220         if (skb->len < 17) {
4221                 if (skb_padto(skb, 17))
4222                         return NETDEV_TX_OK;
4223                 skb->len = 17;
4224         }
4225
4226         tx_ring = adapter->tx_ring[skb->queue_mapping];
4227         return ixgbevf_xmit_frame_ring(skb, tx_ring);
4228 }
4229
4230 /**
4231  * ixgbevf_set_mac - Change the Ethernet Address of the NIC
4232  * @netdev: network interface device structure
4233  * @p: pointer to an address structure
4234  *
4235  * Returns 0 on success, negative on failure
4236  **/
4237 static int ixgbevf_set_mac(struct net_device *netdev, void *p)
4238 {
4239         struct ixgbevf_adapter *adapter = netdev_priv(netdev);
4240         struct ixgbe_hw *hw = &adapter->hw;
4241         struct sockaddr *addr = p;
4242         int err;
4243
4244         if (!is_valid_ether_addr(addr->sa_data))
4245                 return -EADDRNOTAVAIL;
4246
4247         spin_lock_bh(&adapter->mbx_lock);
4248
4249         err = hw->mac.ops.set_rar(hw, 0, addr->sa_data, 0);
4250
4251         spin_unlock_bh(&adapter->mbx_lock);
4252
4253         if (err)
4254                 return -EPERM;
4255
4256         ether_addr_copy(hw->mac.addr, addr->sa_data);
4257         ether_addr_copy(hw->mac.perm_addr, addr->sa_data);
4258         eth_hw_addr_set(netdev, addr->sa_data);
4259
4260         return 0;
4261 }
4262
4263 /**
4264  * ixgbevf_change_mtu - Change the Maximum Transfer Unit
4265  * @netdev: network interface device structure
4266  * @new_mtu: new value for maximum frame size
4267  *
4268  * Returns 0 on success, negative on failure
4269  **/
4270 static int ixgbevf_change_mtu(struct net_device *netdev, int new_mtu)
4271 {
4272         struct ixgbevf_adapter *adapter = netdev_priv(netdev);
4273         struct ixgbe_hw *hw = &adapter->hw;
4274         int max_frame = new_mtu + ETH_HLEN + ETH_FCS_LEN;
4275         int ret;
4276
4277         /* prevent MTU being changed to a size unsupported by XDP */
4278         if (adapter->xdp_prog) {
4279                 dev_warn(&adapter->pdev->dev, "MTU cannot be changed while XDP program is loaded\n");
4280                 return -EPERM;
4281         }
4282
4283         spin_lock_bh(&adapter->mbx_lock);
4284         /* notify the PF of our intent to use this size of frame */
4285         ret = hw->mac.ops.set_rlpml(hw, max_frame);
4286         spin_unlock_bh(&adapter->mbx_lock);
4287         if (ret)
4288                 return -EINVAL;
4289
4290         hw_dbg(hw, "changing MTU from %d to %d\n",
4291                netdev->mtu, new_mtu);
4292
4293         /* must set new MTU before calling down or up */
4294         WRITE_ONCE(netdev->mtu, new_mtu);
4295
4296         if (netif_running(netdev))
4297                 ixgbevf_reinit_locked(adapter);
4298
4299         return 0;
4300 }
4301
4302 static int ixgbevf_suspend(struct device *dev_d)
4303 {
4304         struct net_device *netdev = dev_get_drvdata(dev_d);
4305         struct ixgbevf_adapter *adapter = netdev_priv(netdev);
4306
4307         rtnl_lock();
4308         netif_device_detach(netdev);
4309
4310         if (netif_running(netdev))
4311                 ixgbevf_close_suspend(adapter);
4312
4313         ixgbevf_clear_interrupt_scheme(adapter);
4314         rtnl_unlock();
4315
4316         return 0;
4317 }
4318
4319 static int ixgbevf_resume(struct device *dev_d)
4320 {
4321         struct pci_dev *pdev = to_pci_dev(dev_d);
4322         struct net_device *netdev = pci_get_drvdata(pdev);
4323         struct ixgbevf_adapter *adapter = netdev_priv(netdev);
4324         u32 err;
4325
4326         adapter->hw.hw_addr = adapter->io_addr;
4327         smp_mb__before_atomic();
4328         clear_bit(__IXGBEVF_DISABLED, &adapter->state);
4329         pci_set_master(pdev);
4330
4331         ixgbevf_reset(adapter);
4332
4333         rtnl_lock();
4334         err = ixgbevf_init_interrupt_scheme(adapter);
4335         if (!err && netif_running(netdev))
4336                 err = ixgbevf_open(netdev);
4337         rtnl_unlock();
4338         if (err)
4339                 return err;
4340
4341         netif_device_attach(netdev);
4342
4343         return err;
4344 }
4345
4346 static void ixgbevf_shutdown(struct pci_dev *pdev)
4347 {
4348         ixgbevf_suspend(&pdev->dev);
4349 }
4350
4351 static void ixgbevf_get_tx_ring_stats(struct rtnl_link_stats64 *stats,
4352                                       const struct ixgbevf_ring *ring)
4353 {
4354         u64 bytes, packets;
4355         unsigned int start;
4356
4357         if (ring) {
4358                 do {
4359                         start = u64_stats_fetch_begin(&ring->syncp);
4360                         bytes = ring->stats.bytes;
4361                         packets = ring->stats.packets;
4362                 } while (u64_stats_fetch_retry(&ring->syncp, start));
4363                 stats->tx_bytes += bytes;
4364                 stats->tx_packets += packets;
4365         }
4366 }
4367
4368 static void ixgbevf_get_stats(struct net_device *netdev,
4369                               struct rtnl_link_stats64 *stats)
4370 {
4371         struct ixgbevf_adapter *adapter = netdev_priv(netdev);
4372         unsigned int start;
4373         u64 bytes, packets;
4374         const struct ixgbevf_ring *ring;
4375         int i;
4376
4377         ixgbevf_update_stats(adapter);
4378
4379         stats->multicast = adapter->stats.vfmprc - adapter->stats.base_vfmprc;
4380
4381         rcu_read_lock();
4382         for (i = 0; i < adapter->num_rx_queues; i++) {
4383                 ring = adapter->rx_ring[i];
4384                 do {
4385                         start = u64_stats_fetch_begin(&ring->syncp);
4386                         bytes = ring->stats.bytes;
4387                         packets = ring->stats.packets;
4388                 } while (u64_stats_fetch_retry(&ring->syncp, start));
4389                 stats->rx_bytes += bytes;
4390                 stats->rx_packets += packets;
4391         }
4392
4393         for (i = 0; i < adapter->num_tx_queues; i++) {
4394                 ring = adapter->tx_ring[i];
4395                 ixgbevf_get_tx_ring_stats(stats, ring);
4396         }
4397
4398         for (i = 0; i < adapter->num_xdp_queues; i++) {
4399                 ring = adapter->xdp_ring[i];
4400                 ixgbevf_get_tx_ring_stats(stats, ring);
4401         }
4402         rcu_read_unlock();
4403 }
4404
4405 #define IXGBEVF_MAX_MAC_HDR_LEN         127
4406 #define IXGBEVF_MAX_NETWORK_HDR_LEN     511
4407
4408 static netdev_features_t
4409 ixgbevf_features_check(struct sk_buff *skb, struct net_device *dev,
4410                        netdev_features_t features)
4411 {
4412         unsigned int network_hdr_len, mac_hdr_len;
4413
4414         /* Make certain the headers can be described by a context descriptor */
4415         mac_hdr_len = skb_network_offset(skb);
4416         if (unlikely(mac_hdr_len > IXGBEVF_MAX_MAC_HDR_LEN))
4417                 return features & ~(NETIF_F_HW_CSUM |
4418                                     NETIF_F_SCTP_CRC |
4419                                     NETIF_F_HW_VLAN_CTAG_TX |
4420                                     NETIF_F_TSO |
4421                                     NETIF_F_TSO6);
4422
4423         network_hdr_len = skb_checksum_start(skb) - skb_network_header(skb);
4424         if (unlikely(network_hdr_len >  IXGBEVF_MAX_NETWORK_HDR_LEN))
4425                 return features & ~(NETIF_F_HW_CSUM |
4426                                     NETIF_F_SCTP_CRC |
4427                                     NETIF_F_TSO |
4428                                     NETIF_F_TSO6);
4429
4430         /* We can only support IPV4 TSO in tunnels if we can mangle the
4431          * inner IP ID field, so strip TSO if MANGLEID is not supported.
4432          */
4433         if (skb->encapsulation && !(features & NETIF_F_TSO_MANGLEID))
4434                 features &= ~NETIF_F_TSO;
4435
4436         return features;
4437 }
4438
4439 static int ixgbevf_xdp_setup(struct net_device *dev, struct bpf_prog *prog)
4440 {
4441         int i, frame_size = dev->mtu + ETH_HLEN + ETH_FCS_LEN + VLAN_HLEN;
4442         struct ixgbevf_adapter *adapter = netdev_priv(dev);
4443         struct bpf_prog *old_prog;
4444
4445         /* verify ixgbevf ring attributes are sufficient for XDP */
4446         for (i = 0; i < adapter->num_rx_queues; i++) {
4447                 struct ixgbevf_ring *ring = adapter->rx_ring[i];
4448
4449                 if (frame_size > ixgbevf_rx_bufsz(ring))
4450                         return -EINVAL;
4451         }
4452
4453         old_prog = xchg(&adapter->xdp_prog, prog);
4454
4455         /* If transitioning XDP modes reconfigure rings */
4456         if (!!prog != !!old_prog) {
4457                 /* Hardware has to reinitialize queues and interrupts to
4458                  * match packet buffer alignment. Unfortunately, the
4459                  * hardware is not flexible enough to do this dynamically.
4460                  */
4461                 if (netif_running(dev))
4462                         ixgbevf_close(dev);
4463
4464                 ixgbevf_clear_interrupt_scheme(adapter);
4465                 ixgbevf_init_interrupt_scheme(adapter);
4466
4467                 if (netif_running(dev))
4468                         ixgbevf_open(dev);
4469         } else {
4470                 for (i = 0; i < adapter->num_rx_queues; i++)
4471                         xchg(&adapter->rx_ring[i]->xdp_prog, adapter->xdp_prog);
4472         }
4473
4474         if (old_prog)
4475                 bpf_prog_put(old_prog);
4476
4477         return 0;
4478 }
4479
4480 static int ixgbevf_xdp(struct net_device *dev, struct netdev_bpf *xdp)
4481 {
4482         switch (xdp->command) {
4483         case XDP_SETUP_PROG:
4484                 return ixgbevf_xdp_setup(dev, xdp->prog);
4485         default:
4486                 return -EINVAL;
4487         }
4488 }
4489
4490 static const struct net_device_ops ixgbevf_netdev_ops = {
4491         .ndo_open               = ixgbevf_open,
4492         .ndo_stop               = ixgbevf_close,
4493         .ndo_start_xmit         = ixgbevf_xmit_frame,
4494         .ndo_set_rx_mode        = ixgbevf_set_rx_mode,
4495         .ndo_get_stats64        = ixgbevf_get_stats,
4496         .ndo_validate_addr      = eth_validate_addr,
4497         .ndo_set_mac_address    = ixgbevf_set_mac,
4498         .ndo_change_mtu         = ixgbevf_change_mtu,
4499         .ndo_tx_timeout         = ixgbevf_tx_timeout,
4500         .ndo_vlan_rx_add_vid    = ixgbevf_vlan_rx_add_vid,
4501         .ndo_vlan_rx_kill_vid   = ixgbevf_vlan_rx_kill_vid,
4502         .ndo_features_check     = ixgbevf_features_check,
4503         .ndo_bpf                = ixgbevf_xdp,
4504 };
4505
4506 static void ixgbevf_assign_netdev_ops(struct net_device *dev)
4507 {
4508         dev->netdev_ops = &ixgbevf_netdev_ops;
4509         ixgbevf_set_ethtool_ops(dev);
4510         dev->watchdog_timeo = 5 * HZ;
4511 }
4512
4513 /**
4514  * ixgbevf_probe - Device Initialization Routine
4515  * @pdev: PCI device information struct
4516  * @ent: entry in ixgbevf_pci_tbl
4517  *
4518  * Returns 0 on success, negative on failure
4519  *
4520  * ixgbevf_probe initializes an adapter identified by a pci_dev structure.
4521  * The OS initialization, configuring of the adapter private structure,
4522  * and a hardware reset occur.
4523  **/
4524 static int ixgbevf_probe(struct pci_dev *pdev, const struct pci_device_id *ent)
4525 {
4526         struct net_device *netdev;
4527         struct ixgbevf_adapter *adapter = NULL;
4528         struct ixgbe_hw *hw = NULL;
4529         const struct ixgbevf_info *ii = ixgbevf_info_tbl[ent->driver_data];
4530         bool disable_dev = false;
4531         int err;
4532
4533         err = pci_enable_device(pdev);
4534         if (err)
4535                 return err;
4536
4537         err = dma_set_mask_and_coherent(&pdev->dev, DMA_BIT_MASK(64));
4538         if (err) {
4539                 dev_err(&pdev->dev, "No usable DMA configuration, aborting\n");
4540                 goto err_dma;
4541         }
4542
4543         err = pci_request_regions(pdev, ixgbevf_driver_name);
4544         if (err) {
4545                 dev_err(&pdev->dev, "pci_request_regions failed 0x%x\n", err);
4546                 goto err_pci_reg;
4547         }
4548
4549         pci_set_master(pdev);
4550
4551         netdev = alloc_etherdev_mq(sizeof(struct ixgbevf_adapter),
4552                                    MAX_TX_QUEUES);
4553         if (!netdev) {
4554                 err = -ENOMEM;
4555                 goto err_alloc_etherdev;
4556         }
4557
4558         SET_NETDEV_DEV(netdev, &pdev->dev);
4559
4560         adapter = netdev_priv(netdev);
4561
4562         adapter->netdev = netdev;
4563         adapter->pdev = pdev;
4564         hw = &adapter->hw;
4565         hw->back = adapter;
4566         adapter->msg_enable = netif_msg_init(debug, DEFAULT_MSG_ENABLE);
4567
4568         /* call save state here in standalone driver because it relies on
4569          * adapter struct to exist, and needs to call netdev_priv
4570          */
4571         pci_save_state(pdev);
4572
4573         hw->hw_addr = ioremap(pci_resource_start(pdev, 0),
4574                               pci_resource_len(pdev, 0));
4575         adapter->io_addr = hw->hw_addr;
4576         if (!hw->hw_addr) {
4577                 err = -EIO;
4578                 goto err_ioremap;
4579         }
4580
4581         ixgbevf_assign_netdev_ops(netdev);
4582
4583         /* Setup HW API */
4584         memcpy(&hw->mac.ops, ii->mac_ops, sizeof(hw->mac.ops));
4585         hw->mac.type  = ii->mac;
4586
4587         memcpy(&hw->mbx.ops, &ixgbevf_mbx_ops_legacy,
4588                sizeof(struct ixgbe_mbx_operations));
4589
4590         /* setup the private structure */
4591         err = ixgbevf_sw_init(adapter);
4592         if (err)
4593                 goto err_sw_init;
4594
4595         /* The HW MAC address was set and/or determined in sw_init */
4596         if (!is_valid_ether_addr(netdev->dev_addr)) {
4597                 pr_err("invalid MAC address\n");
4598                 err = -EIO;
4599                 goto err_sw_init;
4600         }
4601
4602         netdev->hw_features = NETIF_F_SG |
4603                               NETIF_F_TSO |
4604                               NETIF_F_TSO6 |
4605                               NETIF_F_RXCSUM |
4606                               NETIF_F_HW_CSUM |
4607                               NETIF_F_SCTP_CRC;
4608
4609 #define IXGBEVF_GSO_PARTIAL_FEATURES (NETIF_F_GSO_GRE | \
4610                                       NETIF_F_GSO_GRE_CSUM | \
4611                                       NETIF_F_GSO_IPXIP4 | \
4612                                       NETIF_F_GSO_IPXIP6 | \
4613                                       NETIF_F_GSO_UDP_TUNNEL | \
4614                                       NETIF_F_GSO_UDP_TUNNEL_CSUM)
4615
4616         netdev->gso_partial_features = IXGBEVF_GSO_PARTIAL_FEATURES;
4617         netdev->hw_features |= NETIF_F_GSO_PARTIAL |
4618                                IXGBEVF_GSO_PARTIAL_FEATURES;
4619
4620         netdev->features = netdev->hw_features | NETIF_F_HIGHDMA;
4621
4622         netdev->vlan_features |= netdev->features | NETIF_F_TSO_MANGLEID;
4623         netdev->mpls_features |= NETIF_F_SG |
4624                                  NETIF_F_TSO |
4625                                  NETIF_F_TSO6 |
4626                                  NETIF_F_HW_CSUM;
4627         netdev->mpls_features |= IXGBEVF_GSO_PARTIAL_FEATURES;
4628         netdev->hw_enc_features |= netdev->vlan_features;
4629
4630         /* set this bit last since it cannot be part of vlan_features */
4631         netdev->features |= NETIF_F_HW_VLAN_CTAG_FILTER |
4632                             NETIF_F_HW_VLAN_CTAG_RX |
4633                             NETIF_F_HW_VLAN_CTAG_TX;
4634
4635         netdev->priv_flags |= IFF_UNICAST_FLT;
4636         netdev->xdp_features = NETDEV_XDP_ACT_BASIC;
4637
4638         /* MTU range: 68 - 1504 or 9710 */
4639         netdev->min_mtu = ETH_MIN_MTU;
4640         switch (adapter->hw.api_version) {
4641         case ixgbe_mbox_api_11:
4642         case ixgbe_mbox_api_12:
4643         case ixgbe_mbox_api_13:
4644         case ixgbe_mbox_api_14:
4645         case ixgbe_mbox_api_15:
4646                 netdev->max_mtu = IXGBE_MAX_JUMBO_FRAME_SIZE -
4647                                   (ETH_HLEN + ETH_FCS_LEN);
4648                 break;
4649         default:
4650                 if (adapter->hw.mac.type != ixgbe_mac_82599_vf)
4651                         netdev->max_mtu = IXGBE_MAX_JUMBO_FRAME_SIZE -
4652                                           (ETH_HLEN + ETH_FCS_LEN);
4653                 else
4654                         netdev->max_mtu = ETH_DATA_LEN + ETH_FCS_LEN;
4655                 break;
4656         }
4657
4658         if (IXGBE_REMOVED(hw->hw_addr)) {
4659                 err = -EIO;
4660                 goto err_sw_init;
4661         }
4662
4663         timer_setup(&adapter->service_timer, ixgbevf_service_timer, 0);
4664
4665         INIT_WORK(&adapter->service_task, ixgbevf_service_task);
4666         set_bit(__IXGBEVF_SERVICE_INITED, &adapter->state);
4667         clear_bit(__IXGBEVF_SERVICE_SCHED, &adapter->state);
4668
4669         err = ixgbevf_init_interrupt_scheme(adapter);
4670         if (err)
4671                 goto err_sw_init;
4672
4673         strcpy(netdev->name, "eth%d");
4674
4675         err = register_netdev(netdev);
4676         if (err)
4677                 goto err_register;
4678
4679         pci_set_drvdata(pdev, netdev);
4680         netif_carrier_off(netdev);
4681         ixgbevf_init_ipsec_offload(adapter);
4682
4683         ixgbevf_init_last_counter_stats(adapter);
4684
4685         /* print the VF info */
4686         dev_info(&pdev->dev, "%pM\n", netdev->dev_addr);
4687         dev_info(&pdev->dev, "MAC: %d\n", hw->mac.type);
4688
4689         switch (hw->mac.type) {
4690         case ixgbe_mac_X550_vf:
4691                 dev_info(&pdev->dev, "Intel(R) X550 Virtual Function\n");
4692                 break;
4693         case ixgbe_mac_X540_vf:
4694                 dev_info(&pdev->dev, "Intel(R) X540 Virtual Function\n");
4695                 break;
4696         case ixgbe_mac_82599_vf:
4697         default:
4698                 dev_info(&pdev->dev, "Intel(R) 82599 Virtual Function\n");
4699                 break;
4700         }
4701
4702         return 0;
4703
4704 err_register:
4705         ixgbevf_clear_interrupt_scheme(adapter);
4706 err_sw_init:
4707         ixgbevf_reset_interrupt_capability(adapter);
4708         iounmap(adapter->io_addr);
4709         kfree(adapter->rss_key);
4710 err_ioremap:
4711         disable_dev = !test_and_set_bit(__IXGBEVF_DISABLED, &adapter->state);
4712         free_netdev(netdev);
4713 err_alloc_etherdev:
4714         pci_release_regions(pdev);
4715 err_pci_reg:
4716 err_dma:
4717         if (!adapter || disable_dev)
4718                 pci_disable_device(pdev);
4719         return err;
4720 }
4721
4722 /**
4723  * ixgbevf_remove - Device Removal Routine
4724  * @pdev: PCI device information struct
4725  *
4726  * ixgbevf_remove is called by the PCI subsystem to alert the driver
4727  * that it should release a PCI device.  The could be caused by a
4728  * Hot-Plug event, or because the driver is going to be removed from
4729  * memory.
4730  **/
4731 static void ixgbevf_remove(struct pci_dev *pdev)
4732 {
4733         struct net_device *netdev = pci_get_drvdata(pdev);
4734         struct ixgbevf_adapter *adapter;
4735         bool disable_dev;
4736
4737         if (!netdev)
4738                 return;
4739
4740         adapter = netdev_priv(netdev);
4741
4742         set_bit(__IXGBEVF_REMOVING, &adapter->state);
4743         cancel_work_sync(&adapter->service_task);
4744
4745         if (netdev->reg_state == NETREG_REGISTERED)
4746                 unregister_netdev(netdev);
4747
4748         ixgbevf_stop_ipsec_offload(adapter);
4749         ixgbevf_clear_interrupt_scheme(adapter);
4750         ixgbevf_reset_interrupt_capability(adapter);
4751
4752         iounmap(adapter->io_addr);
4753         pci_release_regions(pdev);
4754
4755         hw_dbg(&adapter->hw, "Remove complete\n");
4756
4757         kfree(adapter->rss_key);
4758         disable_dev = !test_and_set_bit(__IXGBEVF_DISABLED, &adapter->state);
4759         free_netdev(netdev);
4760
4761         if (disable_dev)
4762                 pci_disable_device(pdev);
4763 }
4764
4765 /**
4766  * ixgbevf_io_error_detected - called when PCI error is detected
4767  * @pdev: Pointer to PCI device
4768  * @state: The current pci connection state
4769  *
4770  * This function is called after a PCI bus error affecting
4771  * this device has been detected.
4772  **/
4773 static pci_ers_result_t ixgbevf_io_error_detected(struct pci_dev *pdev,
4774                                                   pci_channel_state_t state)
4775 {
4776         struct net_device *netdev = pci_get_drvdata(pdev);
4777         struct ixgbevf_adapter *adapter = netdev_priv(netdev);
4778
4779         if (!test_bit(__IXGBEVF_SERVICE_INITED, &adapter->state))
4780                 return PCI_ERS_RESULT_DISCONNECT;
4781
4782         rtnl_lock();
4783         netif_device_detach(netdev);
4784
4785         if (netif_running(netdev))
4786                 ixgbevf_close_suspend(adapter);
4787
4788         if (state == pci_channel_io_perm_failure) {
4789                 rtnl_unlock();
4790                 return PCI_ERS_RESULT_DISCONNECT;
4791         }
4792
4793         if (!test_and_set_bit(__IXGBEVF_DISABLED, &adapter->state))
4794                 pci_disable_device(pdev);
4795         rtnl_unlock();
4796
4797         /* Request a slot reset. */
4798         return PCI_ERS_RESULT_NEED_RESET;
4799 }
4800
4801 /**
4802  * ixgbevf_io_slot_reset - called after the pci bus has been reset.
4803  * @pdev: Pointer to PCI device
4804  *
4805  * Restart the card from scratch, as if from a cold-boot. Implementation
4806  * resembles the first-half of the ixgbevf_resume routine.
4807  **/
4808 static pci_ers_result_t ixgbevf_io_slot_reset(struct pci_dev *pdev)
4809 {
4810         struct net_device *netdev = pci_get_drvdata(pdev);
4811         struct ixgbevf_adapter *adapter = netdev_priv(netdev);
4812
4813         if (pci_enable_device_mem(pdev)) {
4814                 dev_err(&pdev->dev,
4815                         "Cannot re-enable PCI device after reset.\n");
4816                 return PCI_ERS_RESULT_DISCONNECT;
4817         }
4818
4819         adapter->hw.hw_addr = adapter->io_addr;
4820         smp_mb__before_atomic();
4821         clear_bit(__IXGBEVF_DISABLED, &adapter->state);
4822         pci_set_master(pdev);
4823
4824         ixgbevf_reset(adapter);
4825
4826         return PCI_ERS_RESULT_RECOVERED;
4827 }
4828
4829 /**
4830  * ixgbevf_io_resume - called when traffic can start flowing again.
4831  * @pdev: Pointer to PCI device
4832  *
4833  * This callback is called when the error recovery driver tells us that
4834  * its OK to resume normal operation. Implementation resembles the
4835  * second-half of the ixgbevf_resume routine.
4836  **/
4837 static void ixgbevf_io_resume(struct pci_dev *pdev)
4838 {
4839         struct net_device *netdev = pci_get_drvdata(pdev);
4840
4841         rtnl_lock();
4842         if (netif_running(netdev))
4843                 ixgbevf_open(netdev);
4844
4845         netif_device_attach(netdev);
4846         rtnl_unlock();
4847 }
4848
4849 /* PCI Error Recovery (ERS) */
4850 static const struct pci_error_handlers ixgbevf_err_handler = {
4851         .error_detected = ixgbevf_io_error_detected,
4852         .slot_reset = ixgbevf_io_slot_reset,
4853         .resume = ixgbevf_io_resume,
4854 };
4855
4856 static DEFINE_SIMPLE_DEV_PM_OPS(ixgbevf_pm_ops, ixgbevf_suspend, ixgbevf_resume);
4857
4858 static struct pci_driver ixgbevf_driver = {
4859         .name           = ixgbevf_driver_name,
4860         .id_table       = ixgbevf_pci_tbl,
4861         .probe          = ixgbevf_probe,
4862         .remove         = ixgbevf_remove,
4863
4864         /* Power Management Hooks */
4865         .driver.pm      = pm_sleep_ptr(&ixgbevf_pm_ops),
4866
4867         .shutdown       = ixgbevf_shutdown,
4868         .err_handler    = &ixgbevf_err_handler
4869 };
4870
4871 /**
4872  * ixgbevf_init_module - Driver Registration Routine
4873  *
4874  * ixgbevf_init_module is the first routine called when the driver is
4875  * loaded. All it does is register with the PCI subsystem.
4876  **/
4877 static int __init ixgbevf_init_module(void)
4878 {
4879         int err;
4880
4881         pr_info("%s\n", ixgbevf_driver_string);
4882         pr_info("%s\n", ixgbevf_copyright);
4883         ixgbevf_wq = create_singlethread_workqueue(ixgbevf_driver_name);
4884         if (!ixgbevf_wq) {
4885                 pr_err("%s: Failed to create workqueue\n", ixgbevf_driver_name);
4886                 return -ENOMEM;
4887         }
4888
4889         err = pci_register_driver(&ixgbevf_driver);
4890         if (err) {
4891                 destroy_workqueue(ixgbevf_wq);
4892                 return err;
4893         }
4894
4895         return 0;
4896 }
4897
4898 module_init(ixgbevf_init_module);
4899
4900 /**
4901  * ixgbevf_exit_module - Driver Exit Cleanup Routine
4902  *
4903  * ixgbevf_exit_module is called just before the driver is removed
4904  * from memory.
4905  **/
4906 static void __exit ixgbevf_exit_module(void)
4907 {
4908         pci_unregister_driver(&ixgbevf_driver);
4909         if (ixgbevf_wq) {
4910                 destroy_workqueue(ixgbevf_wq);
4911                 ixgbevf_wq = NULL;
4912         }
4913 }
4914
4915 #ifdef DEBUG
4916 /**
4917  * ixgbevf_get_hw_dev_name - return device name string
4918  * used by hardware layer to print debugging information
4919  * @hw: pointer to private hardware struct
4920  **/
4921 char *ixgbevf_get_hw_dev_name(struct ixgbe_hw *hw)
4922 {
4923         struct ixgbevf_adapter *adapter = hw->back;
4924
4925         return adapter->netdev->name;
4926 }
4927
4928 #endif
4929 module_exit(ixgbevf_exit_module);
4930
4931 /* ixgbevf_main.c */