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_AUTHOR("Intel Corporation, <linux.nics@intel.com>");
  80 MODULE_DESCRIPTION("Intel(R) 10 Gigabit Virtual Function Network Driver");
  81 MODULE_LICENSE("GPL v2");
  82
  83 #define DEFAULT_MSG_ENABLE (NETIF_MSG_DRV|NETIF_MSG_PROBE|NETIF_MSG_LINK)
  84 static int debug = -1;
  85 module_param(debug, int, 0);
  86 MODULE_PARM_DESC(debug, "Debug level (0=none,...,16=all)");
  87
  88 static struct workqueue_struct *ixgbevf_wq;
  89
  90 static void ixgbevf_service_event_schedule(struct ixgbevf_adapter *adapter)
  91 {
  92         if (!test_bit(__IXGBEVF_DOWN, &adapter->state) &&
  93             !test_bit(__IXGBEVF_REMOVING, &adapter->state) &&
  94             !test_and_set_bit(__IXGBEVF_SERVICE_SCHED, &adapter->state))
  95                 queue_work(ixgbevf_wq, &adapter->service_task);
  96 }
  97
  98 static void ixgbevf_service_event_complete(struct ixgbevf_adapter *adapter)
  99 {
 100         BUG_ON(!test_bit(__IXGBEVF_SERVICE_SCHED, &adapter->state));
 101
 102         /* flush memory to make sure state is correct before next watchdog */
 103         smp_mb__before_atomic();
 104         clear_bit(__IXGBEVF_SERVICE_SCHED, &adapter->state);
 105 }
 106
 107 /* forward decls */
 108 static void ixgbevf_queue_reset_subtask(struct ixgbevf_adapter *adapter);
 109 static void ixgbevf_set_itr(struct ixgbevf_q_vector *q_vector);
 110 static void ixgbevf_free_all_rx_resources(struct ixgbevf_adapter *adapter);
 111 static bool ixgbevf_can_reuse_rx_page(struct ixgbevf_rx_buffer *rx_buffer);
 112 static void ixgbevf_reuse_rx_page(struct ixgbevf_ring *rx_ring,
 113                                   struct ixgbevf_rx_buffer *old_buff);
 114
 115 static void ixgbevf_remove_adapter(struct ixgbe_hw *hw)
 116 {
 117         struct ixgbevf_adapter *adapter = hw->back;
 118
 119         if (!hw->hw_addr)
 120                 return;
 121         hw->hw_addr = NULL;
 122         dev_err(&adapter->pdev->dev, "Adapter removed\n");
 123         if (test_bit(__IXGBEVF_SERVICE_INITED, &adapter->state))
 124                 ixgbevf_service_event_schedule(adapter);
 125 }
 126
 127 static void ixgbevf_check_remove(struct ixgbe_hw *hw, u32 reg)
 128 {
 129         u32 value;
 130
 131         /* The following check not only optimizes a bit by not
 132          * performing a read on the status register when the
 133          * register just read was a status register read that
 134          * returned IXGBE_FAILED_READ_REG. It also blocks any
 135          * potential recursion.
 136          */
 137         if (reg == IXGBE_VFSTATUS) {
 138                 ixgbevf_remove_adapter(hw);
 139                 return;
 140         }
 141         value = ixgbevf_read_reg(hw, IXGBE_VFSTATUS);
 142         if (value == IXGBE_FAILED_READ_REG)
 143                 ixgbevf_remove_adapter(hw);
 144 }
 145
 146 u32 ixgbevf_read_reg(struct ixgbe_hw *hw, u32 reg)
 147 {
 148         u8 __iomem *reg_addr = READ_ONCE(hw->hw_addr);
 149         u32 value;
 150
 151         if (IXGBE_REMOVED(reg_addr))
 152                 return IXGBE_FAILED_READ_REG;
 153         value = readl(reg_addr + reg);
 154         if (unlikely(value == IXGBE_FAILED_READ_REG))
 155                 ixgbevf_check_remove(hw, reg);
 156         return value;
 157 }
 158
 159 /**
 160  * ixgbevf_set_ivar - set IVAR registers - maps interrupt causes to vectors
 161  * @adapter: pointer to adapter struct
 162  * @direction: 0 for Rx, 1 for Tx, -1 for other causes
 163  * @queue: queue to map the corresponding interrupt to
 164  * @msix_vector: the vector to map to the corresponding queue
 165  **/
 166 static void ixgbevf_set_ivar(struct ixgbevf_adapter *adapter, s8 direction,
 167                              u8 queue, u8 msix_vector)
 168 {
 169         u32 ivar, index;
 170         struct ixgbe_hw *hw = &adapter->hw;
 171
 172         if (direction == -1) {
 173                 /* other causes */
 174                 msix_vector |= IXGBE_IVAR_ALLOC_VAL;
 175                 ivar = IXGBE_READ_REG(hw, IXGBE_VTIVAR_MISC);
 176                 ivar &= ~0xFF;
 177                 ivar |= msix_vector;
 178                 IXGBE_WRITE_REG(hw, IXGBE_VTIVAR_MISC, ivar);
 179         } else {
 180                 /* Tx or Rx causes */
 181                 msix_vector |= IXGBE_IVAR_ALLOC_VAL;
 182                 index = ((16 * (queue & 1)) + (8 * direction));
 183                 ivar = IXGBE_READ_REG(hw, IXGBE_VTIVAR(queue >> 1));
 184                 ivar &= ~(0xFF << index);
 185                 ivar |= (msix_vector << index);
 186                 IXGBE_WRITE_REG(hw, IXGBE_VTIVAR(queue >> 1), ivar);
 187         }
 188 }
 189
 190 static u64 ixgbevf_get_tx_completed(struct ixgbevf_ring *ring)
 191 {
 192         return ring->stats.packets;
 193 }
 194
 195 static u32 ixgbevf_get_tx_pending(struct ixgbevf_ring *ring)
 196 {
 197         struct ixgbevf_adapter *adapter = netdev_priv(ring->netdev);
 198         struct ixgbe_hw *hw = &adapter->hw;
 199
 200         u32 head = IXGBE_READ_REG(hw, IXGBE_VFTDH(ring->reg_idx));
 201         u32 tail = IXGBE_READ_REG(hw, IXGBE_VFTDT(ring->reg_idx));
 202
 203         if (head != tail)
 204                 return (head < tail) ?
 205                         tail - head : (tail + ring->count - head);
 206
 207         return 0;
 208 }
 209
 210 static inline bool ixgbevf_check_tx_hang(struct ixgbevf_ring *tx_ring)
 211 {
 212         u32 tx_done = ixgbevf_get_tx_completed(tx_ring);
 213         u32 tx_done_old = tx_ring->tx_stats.tx_done_old;
 214         u32 tx_pending = ixgbevf_get_tx_pending(tx_ring);
 215
 216         clear_check_for_tx_hang(tx_ring);
 217
 218         /* Check for a hung queue, but be thorough. This verifies
 219          * that a transmit has been completed since the previous
 220          * check AND there is at least one packet pending. The
 221          * ARMED bit is set to indicate a potential hang.
 222          */
 223         if ((tx_done_old == tx_done) && tx_pending) {
 224                 /* make sure it is true for two checks in a row */
 225                 return test_and_set_bit(__IXGBEVF_HANG_CHECK_ARMED,
 226                                         &tx_ring->state);
 227         }
 228         /* reset the countdown */
 229         clear_bit(__IXGBEVF_HANG_CHECK_ARMED, &tx_ring->state);
 230
 231         /* update completed stats and continue */
 232         tx_ring->tx_stats.tx_done_old = tx_done;
 233
 234         return false;
 235 }
 236
 237 static void ixgbevf_tx_timeout_reset(struct ixgbevf_adapter *adapter)
 238 {
 239         /* Do the reset outside of interrupt context */
 240         if (!test_bit(__IXGBEVF_DOWN, &adapter->state)) {
 241                 set_bit(__IXGBEVF_RESET_REQUESTED, &adapter->state);
 242                 ixgbevf_service_event_schedule(adapter);
 243         }
 244 }
 245
 246 /**
 247  * ixgbevf_tx_timeout - Respond to a Tx Hang
 248  * @netdev: network interface device structure
 249  * @txqueue: transmit queue hanging (unused)
 250  **/
 251 static void ixgbevf_tx_timeout(struct net_device *netdev, unsigned int __always_unused txqueue)
 252 {
 253         struct ixgbevf_adapter *adapter = netdev_priv(netdev);
 254
 255         ixgbevf_tx_timeout_reset(adapter);
 256 }
 257
 258 /**
 259  * ixgbevf_clean_tx_irq - Reclaim resources after transmit completes
 260  * @q_vector: board private structure
 261  * @tx_ring: tx ring to clean
 262  * @napi_budget: Used to determine if we are in netpoll
 263  **/
 264 static bool ixgbevf_clean_tx_irq(struct ixgbevf_q_vector *q_vector,
 265                                  struct ixgbevf_ring *tx_ring, int napi_budget)
 266 {
 267         struct ixgbevf_adapter *adapter = q_vector->adapter;
 268         struct ixgbevf_tx_buffer *tx_buffer;
 269         union ixgbe_adv_tx_desc *tx_desc;
 270         unsigned int total_bytes = 0, total_packets = 0, total_ipsec = 0;
 271         unsigned int budget = tx_ring->count / 2;
 272         unsigned int i = tx_ring->next_to_clean;
 273
 274         if (test_bit(__IXGBEVF_DOWN, &adapter->state))
 275                 return true;
 276
 277         tx_buffer = &tx_ring->tx_buffer_info[i];
 278         tx_desc = IXGBEVF_TX_DESC(tx_ring, i);
 279         i -= tx_ring->count;
 280
 281         do {
 282                 union ixgbe_adv_tx_desc *eop_desc = tx_buffer->next_to_watch;
 283
 284                 /* if next_to_watch is not set then there is no work pending */
 285                 if (!eop_desc)
 286                         break;
 287
 288                 /* prevent any other reads prior to eop_desc */
 289                 smp_rmb();
 290
 291                 /* if DD is not set pending work has not been completed */
 292                 if (!(eop_desc->wb.status & cpu_to_le32(IXGBE_TXD_STAT_DD)))
 293                         break;
 294
 295                 /* clear next_to_watch to prevent false hangs */
 296                 tx_buffer->next_to_watch = NULL;
 297
 298                 /* update the statistics for this packet */
 299                 total_bytes += tx_buffer->bytecount;
 300                 total_packets += tx_buffer->gso_segs;
 301                 if (tx_buffer->tx_flags & IXGBE_TX_FLAGS_IPSEC)
 302                         total_ipsec++;
 303
 304                 /* free the skb */
 305                 if (ring_is_xdp(tx_ring))
 306                         page_frag_free(tx_buffer->data);
 307                 else
 308                         napi_consume_skb(tx_buffer->skb, napi_budget);
 309
 310                 /* unmap skb header data */
 311                 dma_unmap_single(tx_ring->dev,
 312                                  dma_unmap_addr(tx_buffer, dma),
 313                                  dma_unmap_len(tx_buffer, len),
 314                                  DMA_TO_DEVICE);
 315
 316                 /* clear tx_buffer data */
 317                 dma_unmap_len_set(tx_buffer, len, 0);
 318
 319                 /* unmap remaining buffers */
 320                 while (tx_desc != eop_desc) {
 321                         tx_buffer++;
 322                         tx_desc++;
 323                         i++;
 324                         if (unlikely(!i)) {
 325                                 i -= tx_ring->count;
 326                                 tx_buffer = tx_ring->tx_buffer_info;
 327                                 tx_desc = IXGBEVF_TX_DESC(tx_ring, 0);
 328                         }
 329
 330                         /* unmap any remaining paged data */
 331                         if (dma_unmap_len(tx_buffer, len)) {
 332                                 dma_unmap_page(tx_ring->dev,
 333                                                dma_unmap_addr(tx_buffer, dma),
 334                                                dma_unmap_len(tx_buffer, len),
 335                                                DMA_TO_DEVICE);
 336                                 dma_unmap_len_set(tx_buffer, len, 0);
 337                         }
 338                 }
 339
 340                 /* move us one more past the eop_desc for start of next pkt */
 341                 tx_buffer++;
 342                 tx_desc++;
 343                 i++;
 344                 if (unlikely(!i)) {
 345                         i -= tx_ring->count;
 346                         tx_buffer = tx_ring->tx_buffer_info;
 347                         tx_desc = IXGBEVF_TX_DESC(tx_ring, 0);
 348                 }
 349
 350                 /* issue prefetch for next Tx descriptor */
 351                 prefetch(tx_desc);
 352
 353                 /* update budget accounting */
 354                 budget--;
 355         } while (likely(budget));
 356
 357         i += tx_ring->count;
 358         tx_ring->next_to_clean = i;
 359         u64_stats_update_begin(&tx_ring->syncp);
 360         tx_ring->stats.bytes += total_bytes;
 361         tx_ring->stats.packets += total_packets;
 362         u64_stats_update_end(&tx_ring->syncp);
 363         q_vector->tx.total_bytes += total_bytes;
 364         q_vector->tx.total_packets += total_packets;
 365         adapter->tx_ipsec += total_ipsec;
 366
 367         if (check_for_tx_hang(tx_ring) && ixgbevf_check_tx_hang(tx_ring)) {
 368                 struct ixgbe_hw *hw = &adapter->hw;
 369                 union ixgbe_adv_tx_desc *eop_desc;
 370
 371                 eop_desc = tx_ring->tx_buffer_info[i].next_to_watch;
 372
 373                 pr_err("Detected Tx Unit Hang%s\n"
 374                        "  Tx Queue             <%d>\n"
 375                        "  TDH, TDT             <%x>, <%x>\n"
 376                        "  next_to_use          <%x>\n"
 377                        "  next_to_clean        <%x>\n"
 378                        "tx_buffer_info[next_to_clean]\n"
 379                        "  next_to_watch        <%p>\n"
 380                        "  eop_desc->wb.status  <%x>\n"
 381                        "  time_stamp           <%lx>\n"
 382                        "  jiffies              <%lx>\n",
 383                        ring_is_xdp(tx_ring) ? " XDP" : "",
 384                        tx_ring->queue_index,
 385                        IXGBE_READ_REG(hw, IXGBE_VFTDH(tx_ring->reg_idx)),
 386                        IXGBE_READ_REG(hw, IXGBE_VFTDT(tx_ring->reg_idx)),
 387                        tx_ring->next_to_use, i,
 388                        eop_desc, (eop_desc ? eop_desc->wb.status : 0),
 389                        tx_ring->tx_buffer_info[i].time_stamp, jiffies);
 390
 391                 if (!ring_is_xdp(tx_ring))
 392                         netif_stop_subqueue(tx_ring->netdev,
 393                                             tx_ring->queue_index);
 394
 395                 /* schedule immediate reset if we believe we hung */
 396                 ixgbevf_tx_timeout_reset(adapter);
 397
 398                 return true;
 399         }
 400
 401         if (ring_is_xdp(tx_ring))
 402                 return !!budget;
 403
 404 #define TX_WAKE_THRESHOLD (DESC_NEEDED * 2)
 405         if (unlikely(total_packets && netif_carrier_ok(tx_ring->netdev) &&
 406                      (ixgbevf_desc_unused(tx_ring) >= TX_WAKE_THRESHOLD))) {
 407                 /* Make sure that anybody stopping the queue after this
 408                  * sees the new next_to_clean.
 409                  */
 410                 smp_mb();
 411
 412                 if (__netif_subqueue_stopped(tx_ring->netdev,
 413                                              tx_ring->queue_index) &&
 414                     !test_bit(__IXGBEVF_DOWN, &adapter->state)) {
 415                         netif_wake_subqueue(tx_ring->netdev,
 416                                             tx_ring->queue_index);
 417                         ++tx_ring->tx_stats.restart_queue;
 418                 }
 419         }
 420
 421         return !!budget;
 422 }
 423
 424 /**
 425  * ixgbevf_rx_skb - Helper function to determine proper Rx method
 426  * @q_vector: structure containing interrupt and ring information
 427  * @skb: packet to send up
 428  **/
 429 static void ixgbevf_rx_skb(struct ixgbevf_q_vector *q_vector,
 430                            struct sk_buff *skb)
 431 {
 432         napi_gro_receive(&q_vector->napi, skb);
 433 }
 434
 435 #define IXGBE_RSS_L4_TYPES_MASK \
 436         ((1ul << IXGBE_RXDADV_RSSTYPE_IPV4_TCP) | \
 437          (1ul << IXGBE_RXDADV_RSSTYPE_IPV4_UDP) | \
 438          (1ul << IXGBE_RXDADV_RSSTYPE_IPV6_TCP) | \
 439          (1ul << IXGBE_RXDADV_RSSTYPE_IPV6_UDP))
 440
 441 static inline void ixgbevf_rx_hash(struct ixgbevf_ring *ring,
 442                                    union ixgbe_adv_rx_desc *rx_desc,
 443                                    struct sk_buff *skb)
 444 {
 445         u16 rss_type;
 446
 447         if (!(ring->netdev->features & NETIF_F_RXHASH))
 448                 return;
 449
 450         rss_type = le16_to_cpu(rx_desc->wb.lower.lo_dword.hs_rss.pkt_info) &
 451                    IXGBE_RXDADV_RSSTYPE_MASK;
 452
 453         if (!rss_type)
 454                 return;
 455
 456         skb_set_hash(skb, le32_to_cpu(rx_desc->wb.lower.hi_dword.rss),
 457                      (IXGBE_RSS_L4_TYPES_MASK & (1ul << rss_type)) ?
 458                      PKT_HASH_TYPE_L4 : PKT_HASH_TYPE_L3);
 459 }
 460
 461 /**
 462  * ixgbevf_rx_checksum - indicate in skb if hw indicated a good cksum
 463  * @ring: structure containig ring specific data
 464  * @rx_desc: current Rx descriptor being processed
 465  * @skb: skb currently being received and modified
 466  **/
 467 static inline void ixgbevf_rx_checksum(struct ixgbevf_ring *ring,
 468                                        union ixgbe_adv_rx_desc *rx_desc,
 469                                        struct sk_buff *skb)
 470 {
 471         skb_checksum_none_assert(skb);
 472
 473         /* Rx csum disabled */
 474         if (!(ring->netdev->features & NETIF_F_RXCSUM))
 475                 return;
 476
 477         /* if IP and error */
 478         if (ixgbevf_test_staterr(rx_desc, IXGBE_RXD_STAT_IPCS) &&
 479             ixgbevf_test_staterr(rx_desc, IXGBE_RXDADV_ERR_IPE)) {
 480                 ring->rx_stats.csum_err++;
 481                 return;
 482         }
 483
 484         if (!ixgbevf_test_staterr(rx_desc, IXGBE_RXD_STAT_L4CS))
 485                 return;
 486
 487         if (ixgbevf_test_staterr(rx_desc, IXGBE_RXDADV_ERR_TCPE)) {
 488                 ring->rx_stats.csum_err++;
 489                 return;
 490         }
 491
 492         /* It must be a TCP or UDP packet with a valid checksum */
 493         skb->ip_summed = CHECKSUM_UNNECESSARY;
 494 }
 495
 496 /**
 497  * ixgbevf_process_skb_fields - Populate skb header fields from Rx descriptor
 498  * @rx_ring: rx descriptor ring packet is being transacted on
 499  * @rx_desc: pointer to the EOP Rx descriptor
 500  * @skb: pointer to current skb being populated
 501  *
 502  * This function checks the ring, descriptor, and packet information in
 503  * order to populate the checksum, VLAN, protocol, and other fields within
 504  * the skb.
 505  **/
 506 static void ixgbevf_process_skb_fields(struct ixgbevf_ring *rx_ring,
 507                                        union ixgbe_adv_rx_desc *rx_desc,
 508                                        struct sk_buff *skb)
 509 {
 510         ixgbevf_rx_hash(rx_ring, rx_desc, skb);
 511         ixgbevf_rx_checksum(rx_ring, rx_desc, skb);
 512
 513         if (ixgbevf_test_staterr(rx_desc, IXGBE_RXD_STAT_VP)) {
 514                 u16 vid = le16_to_cpu(rx_desc->wb.upper.vlan);
 515                 unsigned long *active_vlans = netdev_priv(rx_ring->netdev);
 516
 517                 if (test_bit(vid & VLAN_VID_MASK, active_vlans))
 518                         __vlan_hwaccel_put_tag(skb, htons(ETH_P_8021Q), vid);
 519         }
 520
 521         if (ixgbevf_test_staterr(rx_desc, IXGBE_RXDADV_STAT_SECP))
 522                 ixgbevf_ipsec_rx(rx_ring, rx_desc, skb);
 523
 524         skb->protocol = eth_type_trans(skb, rx_ring->netdev);
 525 }
 526
 527 static
 528 struct ixgbevf_rx_buffer *ixgbevf_get_rx_buffer(struct ixgbevf_ring *rx_ring,
 529                                                 const unsigned int size)
 530 {
 531         struct ixgbevf_rx_buffer *rx_buffer;
 532
 533         rx_buffer = &rx_ring->rx_buffer_info[rx_ring->next_to_clean];
 534         prefetchw(rx_buffer->page);
 535
 536         /* we are reusing so sync this buffer for CPU use */
 537         dma_sync_single_range_for_cpu(rx_ring->dev,
 538                                       rx_buffer->dma,
 539                                       rx_buffer->page_offset,
 540                                       size,
 541                                       DMA_FROM_DEVICE);
 542
 543         rx_buffer->pagecnt_bias--;
 544
 545         return rx_buffer;
 546 }
 547
 548 static void ixgbevf_put_rx_buffer(struct ixgbevf_ring *rx_ring,
 549                                   struct ixgbevf_rx_buffer *rx_buffer,
 550                                   struct sk_buff *skb)
 551 {
 552         if (ixgbevf_can_reuse_rx_page(rx_buffer)) {
 553                 /* hand second half of page back to the ring */
 554                 ixgbevf_reuse_rx_page(rx_ring, rx_buffer);
 555         } else {
 556                 if (IS_ERR(skb))
 557                         /* We are not reusing the buffer so unmap it and free
 558                          * any references we are holding to it
 559                          */
 560                         dma_unmap_page_attrs(rx_ring->dev, rx_buffer->dma,
 561                                              ixgbevf_rx_pg_size(rx_ring),
 562                                              DMA_FROM_DEVICE,
 563                                              IXGBEVF_RX_DMA_ATTR);
 564                 __page_frag_cache_drain(rx_buffer->page,
 565                                         rx_buffer->pagecnt_bias);
 566         }
 567
 568         /* clear contents of rx_buffer */
 569         rx_buffer->page = NULL;
 570 }
 571
 572 /**
 573  * ixgbevf_is_non_eop - process handling of non-EOP buffers
 574  * @rx_ring: Rx ring being processed
 575  * @rx_desc: Rx descriptor for current buffer
 576  *
 577  * This function updates next to clean.  If the buffer is an EOP buffer
 578  * this function exits returning false, otherwise it will place the
 579  * sk_buff in the next buffer to be chained and return true indicating
 580  * that this is in fact a non-EOP buffer.
 581  **/
 582 static bool ixgbevf_is_non_eop(struct ixgbevf_ring *rx_ring,
 583                                union ixgbe_adv_rx_desc *rx_desc)
 584 {
 585         u32 ntc = rx_ring->next_to_clean + 1;
 586
 587         /* fetch, update, and store next to clean */
 588         ntc = (ntc < rx_ring->count) ? ntc : 0;
 589         rx_ring->next_to_clean = ntc;
 590
 591         prefetch(IXGBEVF_RX_DESC(rx_ring, ntc));
 592
 593         if (likely(ixgbevf_test_staterr(rx_desc, IXGBE_RXD_STAT_EOP)))
 594                 return false;
 595
 596         return true;
 597 }
 598
 599 static inline unsigned int ixgbevf_rx_offset(struct ixgbevf_ring *rx_ring)
 600 {
 601         return ring_uses_build_skb(rx_ring) ? IXGBEVF_SKB_PAD : 0;
 602 }
 603
 604 static bool ixgbevf_alloc_mapped_page(struct ixgbevf_ring *rx_ring,
 605                                       struct ixgbevf_rx_buffer *bi)
 606 {
 607         struct page *page = bi->page;
 608         dma_addr_t dma;
 609
 610         /* since we are recycling buffers we should seldom need to alloc */
 611         if (likely(page))
 612                 return true;
 613
 614         /* alloc new page for storage */
 615         page = dev_alloc_pages(ixgbevf_rx_pg_order(rx_ring));
 616         if (unlikely(!page)) {
 617                 rx_ring->rx_stats.alloc_rx_page_failed++;
 618                 return false;
 619         }
 620
 621         /* map page for use */
 622         dma = dma_map_page_attrs(rx_ring->dev, page, 0,
 623                                  ixgbevf_rx_pg_size(rx_ring),
 624                                  DMA_FROM_DEVICE, IXGBEVF_RX_DMA_ATTR);
 625
 626         /* if mapping failed free memory back to system since
 627          * there isn't much point in holding memory we can't use
 628          */
 629         if (dma_mapping_error(rx_ring->dev, dma)) {
 630                 __free_pages(page, ixgbevf_rx_pg_order(rx_ring));
 631
 632                 rx_ring->rx_stats.alloc_rx_page_failed++;
 633                 return false;
 634         }
 635
 636         bi->dma = dma;
 637         bi->page = page;
 638         bi->page_offset = ixgbevf_rx_offset(rx_ring);
 639         bi->pagecnt_bias = 1;
 640         rx_ring->rx_stats.alloc_rx_page++;
 641
 642         return true;
 643 }
 644
 645 /**
 646  * ixgbevf_alloc_rx_buffers - Replace used receive buffers; packet split
 647  * @rx_ring: rx descriptor ring (for a specific queue) to setup buffers on
 648  * @cleaned_count: number of buffers to replace
 649  **/
 650 static void ixgbevf_alloc_rx_buffers(struct ixgbevf_ring *rx_ring,
 651                                      u16 cleaned_count)
 652 {
 653         union ixgbe_adv_rx_desc *rx_desc;
 654         struct ixgbevf_rx_buffer *bi;
 655         unsigned int i = rx_ring->next_to_use;
 656
 657         /* nothing to do or no valid netdev defined */
 658         if (!cleaned_count || !rx_ring->netdev)
 659                 return;
 660
 661         rx_desc = IXGBEVF_RX_DESC(rx_ring, i);
 662         bi = &rx_ring->rx_buffer_info[i];
 663         i -= rx_ring->count;
 664
 665         do {
 666                 if (!ixgbevf_alloc_mapped_page(rx_ring, bi))
 667                         break;
 668
 669                 /* sync the buffer for use by the device */
 670                 dma_sync_single_range_for_device(rx_ring->dev, bi->dma,
 671                                                  bi->page_offset,
 672                                                  ixgbevf_rx_bufsz(rx_ring),
 673                                                  DMA_FROM_DEVICE);
 674
 675                 /* Refresh the desc even if pkt_addr didn't change
 676                  * because each write-back erases this info.
 677                  */
 678                 rx_desc->read.pkt_addr = cpu_to_le64(bi->dma + bi->page_offset);
 679
 680                 rx_desc++;
 681                 bi++;
 682                 i++;
 683                 if (unlikely(!i)) {
 684                         rx_desc = IXGBEVF_RX_DESC(rx_ring, 0);
 685                         bi = rx_ring->rx_buffer_info;
 686                         i -= rx_ring->count;
 687                 }
 688
 689                 /* clear the length for the next_to_use descriptor */
 690                 rx_desc->wb.upper.length = 0;
 691
 692                 cleaned_count--;
 693         } while (cleaned_count);
 694
 695         i += rx_ring->count;
 696
 697         if (rx_ring->next_to_use != i) {
 698                 /* record the next descriptor to use */
 699                 rx_ring->next_to_use = i;
 700
 701                 /* update next to alloc since we have filled the ring */
 702                 rx_ring->next_to_alloc = i;
 703
 704                 /* Force memory writes to complete before letting h/w
 705                  * know there are new descriptors to fetch.  (Only
 706                  * applicable for weak-ordered memory model archs,
 707                  * such as IA-64).
 708                  */
 709                 wmb();
 710                 ixgbevf_write_tail(rx_ring, i);
 711         }
 712 }
 713
 714 /**
 715  * ixgbevf_cleanup_headers - Correct corrupted or empty headers
 716  * @rx_ring: rx descriptor ring packet is being transacted on
 717  * @rx_desc: pointer to the EOP Rx descriptor
 718  * @skb: pointer to current skb being fixed
 719  *
 720  * Check for corrupted packet headers caused by senders on the local L2
 721  * embedded NIC switch not setting up their Tx Descriptors right.  These
 722  * should be very rare.
 723  *
 724  * Also address the case where we are pulling data in on pages only
 725  * and as such no data is present in the skb header.
 726  *
 727  * In addition if skb is not at least 60 bytes we need to pad it so that
 728  * it is large enough to qualify as a valid Ethernet frame.
 729  *
 730  * Returns true if an error was encountered and skb was freed.
 731  **/
 732 static bool ixgbevf_cleanup_headers(struct ixgbevf_ring *rx_ring,
 733                                     union ixgbe_adv_rx_desc *rx_desc,
 734                                     struct sk_buff *skb)
 735 {
 736         /* XDP packets use error pointer so abort at this point */
 737         if (IS_ERR(skb))
 738                 return true;
 739
 740         /* verify that the packet does not have any known errors */
 741         if (unlikely(ixgbevf_test_staterr(rx_desc,
 742                                           IXGBE_RXDADV_ERR_FRAME_ERR_MASK))) {
 743                 struct net_device *netdev = rx_ring->netdev;
 744
 745                 if (!(netdev->features & NETIF_F_RXALL)) {
 746                         dev_kfree_skb_any(skb);
 747                         return true;
 748                 }
 749         }
 750
 751         /* if eth_skb_pad returns an error the skb was freed */
 752         if (eth_skb_pad(skb))
 753                 return true;
 754
 755         return false;
 756 }
 757
 758 /**
 759  * ixgbevf_reuse_rx_page - page flip buffer and store it back on the ring
 760  * @rx_ring: rx descriptor ring to store buffers on
 761  * @old_buff: donor buffer to have page reused
 762  *
 763  * Synchronizes page for reuse by the adapter
 764  **/
 765 static void ixgbevf_reuse_rx_page(struct ixgbevf_ring *rx_ring,
 766                                   struct ixgbevf_rx_buffer *old_buff)
 767 {
 768         struct ixgbevf_rx_buffer *new_buff;
 769         u16 nta = rx_ring->next_to_alloc;
 770
 771         new_buff = &rx_ring->rx_buffer_info[nta];
 772
 773         /* update, and store next to alloc */
 774         nta++;
 775         rx_ring->next_to_alloc = (nta < rx_ring->count) ? nta : 0;
 776
 777         /* transfer page from old buffer to new buffer */
 778         new_buff->page = old_buff->page;
 779         new_buff->dma = old_buff->dma;
 780         new_buff->page_offset = old_buff->page_offset;
 781         new_buff->pagecnt_bias = old_buff->pagecnt_bias;
 782 }
 783
 784 static bool ixgbevf_can_reuse_rx_page(struct ixgbevf_rx_buffer *rx_buffer)
 785 {
 786         unsigned int pagecnt_bias = rx_buffer->pagecnt_bias;
 787         struct page *page = rx_buffer->page;
 788
 789         /* avoid re-using remote and pfmemalloc pages */
 790         if (!dev_page_is_reusable(page))
 791                 return false;
 792
 793 #if (PAGE_SIZE < 8192)
 794         /* if we are only owner of page we can reuse it */
 795         if (unlikely((page_ref_count(page) - pagecnt_bias) > 1))
 796                 return false;
 797 #else
 798 #define IXGBEVF_LAST_OFFSET \
 799         (SKB_WITH_OVERHEAD(PAGE_SIZE) - IXGBEVF_RXBUFFER_2048)
 800
 801         if (rx_buffer->page_offset > IXGBEVF_LAST_OFFSET)
 802                 return false;
 803
 804 #endif
 805
 806         /* If we have drained the page fragment pool we need to update
 807          * the pagecnt_bias and page count so that we fully restock the
 808          * number of references the driver holds.
 809          */
 810         if (unlikely(!pagecnt_bias)) {
 811                 page_ref_add(page, USHRT_MAX);
 812                 rx_buffer->pagecnt_bias = USHRT_MAX;
 813         }
 814
 815         return true;
 816 }
 817
 818 /**
 819  * ixgbevf_add_rx_frag - Add contents of Rx buffer to sk_buff
 820  * @rx_ring: rx descriptor ring to transact packets on
 821  * @rx_buffer: buffer containing page to add
 822  * @skb: sk_buff to place the data into
 823  * @size: size of buffer to be added
 824  *
 825  * This function will add the data contained in rx_buffer->page to the skb.
 826  **/
 827 static void ixgbevf_add_rx_frag(struct ixgbevf_ring *rx_ring,
 828                                 struct ixgbevf_rx_buffer *rx_buffer,
 829                                 struct sk_buff *skb,
 830                                 unsigned int size)
 831 {
 832 #if (PAGE_SIZE < 8192)
 833         unsigned int truesize = ixgbevf_rx_pg_size(rx_ring) / 2;
 834 #else
 835         unsigned int truesize = ring_uses_build_skb(rx_ring) ?
 836                                 SKB_DATA_ALIGN(IXGBEVF_SKB_PAD + size) :
 837                                 SKB_DATA_ALIGN(size);
 838 #endif
 839         skb_add_rx_frag(skb, skb_shinfo(skb)->nr_frags, rx_buffer->page,
 840                         rx_buffer->page_offset, size, truesize);
 841 #if (PAGE_SIZE < 8192)
 842         rx_buffer->page_offset ^= truesize;
 843 #else
 844         rx_buffer->page_offset += truesize;
 845 #endif
 846 }
 847
 848 static
 849 struct sk_buff *ixgbevf_construct_skb(struct ixgbevf_ring *rx_ring,
 850                                       struct ixgbevf_rx_buffer *rx_buffer,
 851                                       struct xdp_buff *xdp,
 852                                       union ixgbe_adv_rx_desc *rx_desc)
 853 {
 854         unsigned int size = xdp->data_end - xdp->data;
 855 #if (PAGE_SIZE < 8192)
 856         unsigned int truesize = ixgbevf_rx_pg_size(rx_ring) / 2;
 857 #else
 858         unsigned int truesize = SKB_DATA_ALIGN(xdp->data_end -
 859                                                xdp->data_hard_start);
 860 #endif
 861         unsigned int headlen;
 862         struct sk_buff *skb;
 863
 864         /* prefetch first cache line of first page */
 865         net_prefetch(xdp->data);
 866
 867         /* Note, we get here by enabling legacy-rx via:
 868          *
 869          *    ethtool --set-priv-flags <dev> legacy-rx on
 870          *
 871          * In this mode, we currently get 0 extra XDP headroom as
 872          * opposed to having legacy-rx off, where we process XDP
 873          * packets going to stack via ixgbevf_build_skb().
 874          *
 875          * For ixgbevf_construct_skb() mode it means that the
 876          * xdp->data_meta will always point to xdp->data, since
 877          * the helper cannot expand the head. Should this ever
 878          * changed in future for legacy-rx mode on, then lets also
 879          * add xdp->data_meta handling here.
 880          */
 881
 882         /* allocate a skb to store the frags */
 883         skb = napi_alloc_skb(&rx_ring->q_vector->napi, IXGBEVF_RX_HDR_SIZE);
 884         if (unlikely(!skb))
 885                 return NULL;
 886
 887         /* Determine available headroom for copy */
 888         headlen = size;
 889         if (headlen > IXGBEVF_RX_HDR_SIZE)
 890                 headlen = eth_get_headlen(skb->dev, xdp->data,
 891                                           IXGBEVF_RX_HDR_SIZE);
 892
 893         /* align pull length to size of long to optimize memcpy performance */
 894         memcpy(__skb_put(skb, headlen), xdp->data,
 895                ALIGN(headlen, sizeof(long)));
 896
 897         /* update all of the pointers */
 898         size -= headlen;
 899         if (size) {
 900                 skb_add_rx_frag(skb, 0, rx_buffer->page,
 901                                 (xdp->data + headlen) -
 902                                         page_address(rx_buffer->page),
 903                                 size, truesize);
 904 #if (PAGE_SIZE < 8192)
 905                 rx_buffer->page_offset ^= truesize;
 906 #else
 907                 rx_buffer->page_offset += truesize;
 908 #endif
 909         } else {
 910                 rx_buffer->pagecnt_bias++;
 911         }
 912
 913         return skb;
 914 }
 915
 916 static inline void ixgbevf_irq_enable_queues(struct ixgbevf_adapter *adapter,
 917                                              u32 qmask)
 918 {
 919         struct ixgbe_hw *hw = &adapter->hw;
 920
 921         IXGBE_WRITE_REG(hw, IXGBE_VTEIMS, qmask);
 922 }
 923
 924 static struct sk_buff *ixgbevf_build_skb(struct ixgbevf_ring *rx_ring,
 925                                          struct ixgbevf_rx_buffer *rx_buffer,
 926                                          struct xdp_buff *xdp,
 927                                          union ixgbe_adv_rx_desc *rx_desc)
 928 {
 929         unsigned int metasize = xdp->data - xdp->data_meta;
 930 #if (PAGE_SIZE < 8192)
 931         unsigned int truesize = ixgbevf_rx_pg_size(rx_ring) / 2;
 932 #else
 933         unsigned int truesize = SKB_DATA_ALIGN(sizeof(struct skb_shared_info)) +
 934                                 SKB_DATA_ALIGN(xdp->data_end -
 935                                                xdp->data_hard_start);
 936 #endif
 937         struct sk_buff *skb;
 938
 939         /* Prefetch first cache line of first page. If xdp->data_meta
 940          * is unused, this points to xdp->data, otherwise, we likely
 941          * have a consumer accessing first few bytes of meta data,
 942          * and then actual data.
 943          */
 944         net_prefetch(xdp->data_meta);
 945
 946         /* build an skb around the page buffer */
 947         skb = napi_build_skb(xdp->data_hard_start, truesize);
 948         if (unlikely(!skb))
 949                 return NULL;
 950
 951         /* update pointers within the skb to store the data */
 952         skb_reserve(skb, xdp->data - xdp->data_hard_start);
 953         __skb_put(skb, xdp->data_end - xdp->data);
 954         if (metasize)
 955                 skb_metadata_set(skb, metasize);
 956
 957         /* update buffer offset */
 958 #if (PAGE_SIZE < 8192)
 959         rx_buffer->page_offset ^= truesize;
 960 #else
 961         rx_buffer->page_offset += truesize;
 962 #endif
 963
 964         return skb;
 965 }
 966
 967 #define IXGBEVF_XDP_PASS 0
 968 #define IXGBEVF_XDP_CONSUMED 1
 969 #define IXGBEVF_XDP_TX 2
 970
 971 static int ixgbevf_xmit_xdp_ring(struct ixgbevf_ring *ring,
 972                                  struct xdp_buff *xdp)
 973 {
 974         struct ixgbevf_tx_buffer *tx_buffer;
 975         union ixgbe_adv_tx_desc *tx_desc;
 976         u32 len, cmd_type;
 977         dma_addr_t dma;
 978         u16 i;
 979
 980         len = xdp->data_end - xdp->data;
 981
 982         if (unlikely(!ixgbevf_desc_unused(ring)))
 983                 return IXGBEVF_XDP_CONSUMED;
 984
 985         dma = dma_map_single(ring->dev, xdp->data, len, DMA_TO_DEVICE);
 986         if (dma_mapping_error(ring->dev, dma))
 987                 return IXGBEVF_XDP_CONSUMED;
 988
 989         /* record the location of the first descriptor for this packet */
 990         i = ring->next_to_use;
 991         tx_buffer = &ring->tx_buffer_info[i];
 992
 993         dma_unmap_len_set(tx_buffer, len, len);
 994         dma_unmap_addr_set(tx_buffer, dma, dma);
 995         tx_buffer->data = xdp->data;
 996         tx_buffer->bytecount = len;
 997         tx_buffer->gso_segs = 1;
 998         tx_buffer->protocol = 0;
 999
1000         /* Populate minimal context descriptor that will provide for the
1001          * fact that we are expected to process Ethernet frames.
1002          */
1003         if (!test_bit(__IXGBEVF_TX_XDP_RING_PRIMED, &ring->state)) {
1004                 struct ixgbe_adv_tx_context_desc *context_desc;
1005
1006                 set_bit(__IXGBEVF_TX_XDP_RING_PRIMED, &ring->state);
1007
1008                 context_desc = IXGBEVF_TX_CTXTDESC(ring, 0);
1009                 context_desc->vlan_macip_lens   =
1010                         cpu_to_le32(ETH_HLEN << IXGBE_ADVTXD_MACLEN_SHIFT);
1011                 context_desc->fceof_saidx       = 0;
1012                 context_desc->type_tucmd_mlhl   =
1013                         cpu_to_le32(IXGBE_TXD_CMD_DEXT |
1014                                     IXGBE_ADVTXD_DTYP_CTXT);
1015                 context_desc->mss_l4len_idx     = 0;
1016
1017                 i = 1;
1018         }
1019
1020         /* put descriptor type bits */
1021         cmd_type = IXGBE_ADVTXD_DTYP_DATA |
1022                    IXGBE_ADVTXD_DCMD_DEXT |
1023                    IXGBE_ADVTXD_DCMD_IFCS;
1024         cmd_type |= len | IXGBE_TXD_CMD;
1025
1026         tx_desc = IXGBEVF_TX_DESC(ring, i);
1027         tx_desc->read.buffer_addr = cpu_to_le64(dma);
1028
1029         tx_desc->read.cmd_type_len = cpu_to_le32(cmd_type);
1030         tx_desc->read.olinfo_status =
1031                         cpu_to_le32((len << IXGBE_ADVTXD_PAYLEN_SHIFT) |
1032                                     IXGBE_ADVTXD_CC);
1033
1034         /* Avoid any potential race with cleanup */
1035         smp_wmb();
1036
1037         /* set next_to_watch value indicating a packet is present */
1038         i++;
1039         if (i == ring->count)
1040                 i = 0;
1041
1042         tx_buffer->next_to_watch = tx_desc;
1043         ring->next_to_use = i;
1044
1045         return IXGBEVF_XDP_TX;
1046 }
1047
1048 static struct sk_buff *ixgbevf_run_xdp(struct ixgbevf_adapter *adapter,
1049                                        struct ixgbevf_ring  *rx_ring,
1050                                        struct xdp_buff *xdp)
1051 {
1052         int result = IXGBEVF_XDP_PASS;
1053         struct ixgbevf_ring *xdp_ring;
1054         struct bpf_prog *xdp_prog;
1055         u32 act;
1056
1057         xdp_prog = READ_ONCE(rx_ring->xdp_prog);
1058
1059         if (!xdp_prog)
1060                 goto xdp_out;
1061
1062         act = bpf_prog_run_xdp(xdp_prog, xdp);
1063         switch (act) {
1064         case XDP_PASS:
1065                 break;
1066         case XDP_TX:
1067                 xdp_ring = adapter->xdp_ring[rx_ring->queue_index];
1068                 result = ixgbevf_xmit_xdp_ring(xdp_ring, xdp);
1069                 if (result == IXGBEVF_XDP_CONSUMED)
1070                         goto out_failure;
1071                 break;
1072         default:
1073                 bpf_warn_invalid_xdp_action(rx_ring->netdev, xdp_prog, act);
1074                 fallthrough;
1075         case XDP_ABORTED:
1076 out_failure:
1077                 trace_xdp_exception(rx_ring->netdev, xdp_prog, act);
1078                 fallthrough; /* handle aborts by dropping packet */
1079         case XDP_DROP:
1080                 result = IXGBEVF_XDP_CONSUMED;
1081                 break;
1082         }
1083 xdp_out:
1084         return ERR_PTR(-result);
1085 }
1086
1087 static unsigned int ixgbevf_rx_frame_truesize(struct ixgbevf_ring *rx_ring,
1088                                               unsigned int size)
1089 {
1090         unsigned int truesize;
1091
1092 #if (PAGE_SIZE < 8192)
1093         truesize = ixgbevf_rx_pg_size(rx_ring) / 2; /* Must be power-of-2 */
1094 #else
1095         truesize = ring_uses_build_skb(rx_ring) ?
1096                 SKB_DATA_ALIGN(IXGBEVF_SKB_PAD + size) +
1097                 SKB_DATA_ALIGN(sizeof(struct skb_shared_info)) :
1098                 SKB_DATA_ALIGN(size);
1099 #endif
1100         return truesize;
1101 }
1102
1103 static void ixgbevf_rx_buffer_flip(struct ixgbevf_ring *rx_ring,
1104                                    struct ixgbevf_rx_buffer *rx_buffer,
1105                                    unsigned int size)
1106 {
1107         unsigned int truesize = ixgbevf_rx_frame_truesize(rx_ring, size);
1108
1109 #if (PAGE_SIZE < 8192)
1110         rx_buffer->page_offset ^= truesize;
1111 #else
1112         rx_buffer->page_offset += truesize;
1113 #endif
1114 }
1115
1116 static int ixgbevf_clean_rx_irq(struct ixgbevf_q_vector *q_vector,
1117                                 struct ixgbevf_ring *rx_ring,
1118                                 int budget)
1119 {
1120         unsigned int total_rx_bytes = 0, total_rx_packets = 0, frame_sz = 0;
1121         struct ixgbevf_adapter *adapter = q_vector->adapter;
1122         u16 cleaned_count = ixgbevf_desc_unused(rx_ring);
1123         struct sk_buff *skb = rx_ring->skb;
1124         bool xdp_xmit = false;
1125         struct xdp_buff xdp;
1126
1127         /* Frame size depend on rx_ring setup when PAGE_SIZE=4K */
1128 #if (PAGE_SIZE < 8192)
1129         frame_sz = ixgbevf_rx_frame_truesize(rx_ring, 0);
1130 #endif
1131         xdp_init_buff(&xdp, frame_sz, &rx_ring->xdp_rxq);
1132
1133         while (likely(total_rx_packets < budget)) {
1134                 struct ixgbevf_rx_buffer *rx_buffer;
1135                 union ixgbe_adv_rx_desc *rx_desc;
1136                 unsigned int size;
1137
1138                 /* return some buffers to hardware, one at a time is too slow */
1139                 if (cleaned_count >= IXGBEVF_RX_BUFFER_WRITE) {
1140                         ixgbevf_alloc_rx_buffers(rx_ring, cleaned_count);
1141                         cleaned_count = 0;
1142                 }
1143
1144                 rx_desc = IXGBEVF_RX_DESC(rx_ring, rx_ring->next_to_clean);
1145                 size = le16_to_cpu(rx_desc->wb.upper.length);
1146                 if (!size)
1147                         break;
1148
1149                 /* This memory barrier is needed to keep us from reading
1150                  * any other fields out of the rx_desc until we know the
1151                  * RXD_STAT_DD bit is set
1152                  */
1153                 rmb();
1154
1155                 rx_buffer = ixgbevf_get_rx_buffer(rx_ring, size);
1156
1157                 /* retrieve a buffer from the ring */
1158                 if (!skb) {
1159                         unsigned int offset = ixgbevf_rx_offset(rx_ring);
1160                         unsigned char *hard_start;
1161
1162                         hard_start = page_address(rx_buffer->page) +
1163                                      rx_buffer->page_offset - offset;
1164                         xdp_prepare_buff(&xdp, hard_start, offset, size, true);
1165 #if (PAGE_SIZE > 4096)
1166                         /* At larger PAGE_SIZE, frame_sz depend on len size */
1167                         xdp.frame_sz = ixgbevf_rx_frame_truesize(rx_ring, size);
1168 #endif
1169                         skb = ixgbevf_run_xdp(adapter, rx_ring, &xdp);
1170                 }
1171
1172                 if (IS_ERR(skb)) {
1173                         if (PTR_ERR(skb) == -IXGBEVF_XDP_TX) {
1174                                 xdp_xmit = true;
1175                                 ixgbevf_rx_buffer_flip(rx_ring, rx_buffer,
1176                                                        size);
1177                         } else {
1178                                 rx_buffer->pagecnt_bias++;
1179                         }
1180                         total_rx_packets++;
1181                         total_rx_bytes += size;
1182                 } else if (skb) {
1183                         ixgbevf_add_rx_frag(rx_ring, rx_buffer, skb, size);
1184                 } else if (ring_uses_build_skb(rx_ring)) {
1185                         skb = ixgbevf_build_skb(rx_ring, rx_buffer,
1186                                                 &xdp, rx_desc);
1187                 } else {
1188                         skb = ixgbevf_construct_skb(rx_ring, rx_buffer,
1189                                                     &xdp, rx_desc);
1190                 }
1191
1192                 /* exit if we failed to retrieve a buffer */
1193                 if (!skb) {
1194                         rx_ring->rx_stats.alloc_rx_buff_failed++;
1195                         rx_buffer->pagecnt_bias++;
1196                         break;
1197                 }
1198
1199                 ixgbevf_put_rx_buffer(rx_ring, rx_buffer, skb);
1200                 cleaned_count++;
1201
1202                 /* fetch next buffer in frame if non-eop */
1203                 if (ixgbevf_is_non_eop(rx_ring, rx_desc))
1204                         continue;
1205
1206                 /* verify the packet layout is correct */
1207                 if (ixgbevf_cleanup_headers(rx_ring, rx_desc, skb)) {
1208                         skb = NULL;
1209                         continue;
1210                 }
1211
1212                 /* probably a little skewed due to removing CRC */
1213                 total_rx_bytes += skb->len;
1214
1215                 /* Workaround hardware that can't do proper VEPA multicast
1216                  * source pruning.
1217                  */
1218                 if ((skb->pkt_type == PACKET_BROADCAST ||
1219                      skb->pkt_type == PACKET_MULTICAST) &&
1220                     ether_addr_equal(rx_ring->netdev->dev_addr,
1221                                      eth_hdr(skb)->h_source)) {
1222                         dev_kfree_skb_irq(skb);
1223                         continue;
1224                 }
1225
1226                 /* populate checksum, VLAN, and protocol */
1227                 ixgbevf_process_skb_fields(rx_ring, rx_desc, skb);
1228
1229                 ixgbevf_rx_skb(q_vector, skb);
1230
1231                 /* reset skb pointer */
1232                 skb = NULL;
1233
1234                 /* update budget accounting */
1235                 total_rx_packets++;
1236         }
1237
1238         /* place incomplete frames back on ring for completion */
1239         rx_ring->skb = skb;
1240
1241         if (xdp_xmit) {
1242                 struct ixgbevf_ring *xdp_ring =
1243                         adapter->xdp_ring[rx_ring->queue_index];
1244
1245                 /* Force memory writes to complete before letting h/w
1246                  * know there are new descriptors to fetch.
1247                  */
1248                 wmb();
1249                 ixgbevf_write_tail(xdp_ring, xdp_ring->next_to_use);
1250         }
1251
1252         u64_stats_update_begin(&rx_ring->syncp);
1253         rx_ring->stats.packets += total_rx_packets;
1254         rx_ring->stats.bytes += total_rx_bytes;
1255         u64_stats_update_end(&rx_ring->syncp);
1256         q_vector->rx.total_packets += total_rx_packets;
1257         q_vector->rx.total_bytes += total_rx_bytes;
1258
1259         return total_rx_packets;
1260 }
1261
1262 /**
1263  * ixgbevf_poll - NAPI polling calback
1264  * @napi: napi struct with our devices info in it
1265  * @budget: amount of work driver is allowed to do this pass, in packets
1266  *
1267  * This function will clean more than one or more rings associated with a
1268  * q_vector.
1269  **/
1270 static int ixgbevf_poll(struct napi_struct *napi, int budget)
1271 {
1272         struct ixgbevf_q_vector *q_vector =
1273                 container_of(napi, struct ixgbevf_q_vector, napi);
1274         struct ixgbevf_adapter *adapter = q_vector->adapter;
1275         struct ixgbevf_ring *ring;
1276         int per_ring_budget, work_done = 0;
1277         bool clean_complete = true;
1278
1279         ixgbevf_for_each_ring(ring, q_vector->tx) {
1280                 if (!ixgbevf_clean_tx_irq(q_vector, ring, budget))
1281                         clean_complete = false;
1282         }
1283
1284         if (budget <= 0)
1285                 return budget;
1286
1287         /* attempt to distribute budget to each queue fairly, but don't allow
1288          * the budget to go below 1 because we'll exit polling
1289          */
1290         if (q_vector->rx.count > 1)
1291                 per_ring_budget = max(budget/q_vector->rx.count, 1);
1292         else
1293                 per_ring_budget = budget;
1294
1295         ixgbevf_for_each_ring(ring, q_vector->rx) {
1296                 int cleaned = ixgbevf_clean_rx_irq(q_vector, ring,
1297                                                    per_ring_budget);
1298                 work_done += cleaned;
1299                 if (cleaned >= per_ring_budget)
1300                         clean_complete = false;
1301         }
1302
1303         /* If all work not completed, return budget and keep polling */
1304         if (!clean_complete)
1305                 return budget;
1306
1307         /* Exit the polling mode, but don't re-enable interrupts if stack might
1308          * poll us due to busy-polling
1309          */
1310         if (likely(napi_complete_done(napi, work_done))) {
1311                 if (adapter->rx_itr_setting == 1)
1312                         ixgbevf_set_itr(q_vector);
1313                 if (!test_bit(__IXGBEVF_DOWN, &adapter->state) &&
1314                     !test_bit(__IXGBEVF_REMOVING, &adapter->state))
1315                         ixgbevf_irq_enable_queues(adapter,
1316                                                   BIT(q_vector->v_idx));
1317         }
1318
1319         return min(work_done, budget - 1);
1320 }
1321
1322 /**
1323  * ixgbevf_write_eitr - write VTEITR register in hardware specific way
1324  * @q_vector: structure containing interrupt and ring information
1325  **/
1326 void ixgbevf_write_eitr(struct ixgbevf_q_vector *q_vector)
1327 {
1328         struct ixgbevf_adapter *adapter = q_vector->adapter;
1329         struct ixgbe_hw *hw = &adapter->hw;
1330         int v_idx = q_vector->v_idx;
1331         u32 itr_reg = q_vector->itr & IXGBE_MAX_EITR;
1332
1333         /* set the WDIS bit to not clear the timer bits and cause an
1334          * immediate assertion of the interrupt
1335          */
1336         itr_reg |= IXGBE_EITR_CNT_WDIS;
1337
1338         IXGBE_WRITE_REG(hw, IXGBE_VTEITR(v_idx), itr_reg);
1339 }
1340
1341 /**
1342  * ixgbevf_configure_msix - Configure MSI-X hardware
1343  * @adapter: board private structure
1344  *
1345  * ixgbevf_configure_msix sets up the hardware to properly generate MSI-X
1346  * interrupts.
1347  **/
1348 static void ixgbevf_configure_msix(struct ixgbevf_adapter *adapter)
1349 {
1350         struct ixgbevf_q_vector *q_vector;
1351         int q_vectors, v_idx;
1352
1353         q_vectors = adapter->num_msix_vectors - NON_Q_VECTORS;
1354         adapter->eims_enable_mask = 0;
1355
1356         /* Populate the IVAR table and set the ITR values to the
1357          * corresponding register.
1358          */
1359         for (v_idx = 0; v_idx < q_vectors; v_idx++) {
1360                 struct ixgbevf_ring *ring;
1361
1362                 q_vector = adapter->q_vector[v_idx];
1363
1364                 ixgbevf_for_each_ring(ring, q_vector->rx)
1365                         ixgbevf_set_ivar(adapter, 0, ring->reg_idx, v_idx);
1366
1367                 ixgbevf_for_each_ring(ring, q_vector->tx)
1368                         ixgbevf_set_ivar(adapter, 1, ring->reg_idx, v_idx);
1369
1370                 if (q_vector->tx.ring && !q_vector->rx.ring) {
1371                         /* Tx only vector */
1372                         if (adapter->tx_itr_setting == 1)
1373                                 q_vector->itr = IXGBE_12K_ITR;
1374                         else
1375                                 q_vector->itr = adapter->tx_itr_setting;
1376                 } else {
1377                         /* Rx or Rx/Tx vector */
1378                         if (adapter->rx_itr_setting == 1)
1379                                 q_vector->itr = IXGBE_20K_ITR;
1380                         else
1381                                 q_vector->itr = adapter->rx_itr_setting;
1382                 }
1383
1384                 /* add q_vector eims value to global eims_enable_mask */
1385                 adapter->eims_enable_mask |= BIT(v_idx);
1386
1387                 ixgbevf_write_eitr(q_vector);
1388         }
1389
1390         ixgbevf_set_ivar(adapter, -1, 1, v_idx);
1391         /* setup eims_other and add value to global eims_enable_mask */
1392         adapter->eims_other = BIT(v_idx);
1393         adapter->eims_enable_mask |= adapter->eims_other;
1394 }
1395
1396 enum latency_range {
1397         lowest_latency = 0,
1398         low_latency = 1,
1399         bulk_latency = 2,
1400         latency_invalid = 255
1401 };
1402
1403 /**
1404  * ixgbevf_update_itr - update the dynamic ITR value based on statistics
1405  * @q_vector: structure containing interrupt and ring information
1406  * @ring_container: structure containing ring performance data
1407  *
1408  * Stores a new ITR value based on packets and byte
1409  * counts during the last interrupt.  The advantage of per interrupt
1410  * computation is faster updates and more accurate ITR for the current
1411  * traffic pattern.  Constants in this function were computed
1412  * based on theoretical maximum wire speed and thresholds were set based
1413  * on testing data as well as attempting to minimize response time
1414  * while increasing bulk throughput.
1415  **/
1416 static void ixgbevf_update_itr(struct ixgbevf_q_vector *q_vector,
1417                                struct ixgbevf_ring_container *ring_container)
1418 {
1419         int bytes = ring_container->total_bytes;
1420         int packets = ring_container->total_packets;
1421         u32 timepassed_us;
1422         u64 bytes_perint;
1423         u8 itr_setting = ring_container->itr;
1424
1425         if (packets == 0)
1426                 return;
1427
1428         /* simple throttle rate management
1429          *    0-20MB/s lowest (100000 ints/s)
1430          *   20-100MB/s low   (20000 ints/s)
1431          *  100-1249MB/s bulk (12000 ints/s)
1432          */
1433         /* what was last interrupt timeslice? */
1434         timepassed_us = q_vector->itr >> 2;
1435         if (timepassed_us == 0)
1436                 return;
1437
1438         bytes_perint = bytes / timepassed_us; /* bytes/usec */
1439
1440         switch (itr_setting) {
1441         case lowest_latency:
1442                 if (bytes_perint > 10)
1443                         itr_setting = low_latency;
1444                 break;
1445         case low_latency:
1446                 if (bytes_perint > 20)
1447                         itr_setting = bulk_latency;
1448                 else if (bytes_perint <= 10)
1449                         itr_setting = lowest_latency;
1450                 break;
1451         case bulk_latency:
1452                 if (bytes_perint <= 20)
1453                         itr_setting = low_latency;
1454                 break;
1455         }
1456
1457         /* clear work counters since we have the values we need */
1458         ring_container->total_bytes = 0;
1459         ring_container->total_packets = 0;
1460
1461         /* write updated itr to ring container */
1462         ring_container->itr = itr_setting;
1463 }
1464
1465 static void ixgbevf_set_itr(struct ixgbevf_q_vector *q_vector)
1466 {
1467         u32 new_itr = q_vector->itr;
1468         u8 current_itr;
1469
1470         ixgbevf_update_itr(q_vector, &q_vector->tx);
1471         ixgbevf_update_itr(q_vector, &q_vector->rx);
1472
1473         current_itr = max(q_vector->rx.itr, q_vector->tx.itr);
1474
1475         switch (current_itr) {
1476         /* counts and packets in update_itr are dependent on these numbers */
1477         case lowest_latency:
1478                 new_itr = IXGBE_100K_ITR;
1479                 break;
1480         case low_latency:
1481                 new_itr = IXGBE_20K_ITR;
1482                 break;
1483         case bulk_latency:
1484                 new_itr = IXGBE_12K_ITR;
1485                 break;
1486         default:
1487                 break;
1488         }
1489
1490         if (new_itr != q_vector->itr) {
1491                 /* do an exponential smoothing */
1492                 new_itr = (10 * new_itr * q_vector->itr) /
1493                           ((9 * new_itr) + q_vector->itr);
1494
1495                 /* save the algorithm value here */
1496                 q_vector->itr = new_itr;
1497
1498                 ixgbevf_write_eitr(q_vector);
1499         }
1500 }
1501
1502 static irqreturn_t ixgbevf_msix_other(int irq, void *data)
1503 {
1504         struct ixgbevf_adapter *adapter = data;
1505         struct ixgbe_hw *hw = &adapter->hw;
1506
1507         hw->mac.get_link_status = 1;
1508
1509         ixgbevf_service_event_schedule(adapter);
1510
1511         IXGBE_WRITE_REG(hw, IXGBE_VTEIMS, adapter->eims_other);
1512
1513         return IRQ_HANDLED;
1514 }
1515
1516 /**
1517  * ixgbevf_msix_clean_rings - single unshared vector rx clean (all queues)
1518  * @irq: unused
1519  * @data: pointer to our q_vector struct for this interrupt vector
1520  **/
1521 static irqreturn_t ixgbevf_msix_clean_rings(int irq, void *data)
1522 {
1523         struct ixgbevf_q_vector *q_vector = data;
1524
1525         /* EIAM disabled interrupts (on this vector) for us */
1526         if (q_vector->rx.ring || q_vector->tx.ring)
1527                 napi_schedule_irqoff(&q_vector->napi);
1528
1529         return IRQ_HANDLED;
1530 }
1531
1532 /**
1533  * ixgbevf_request_msix_irqs - Initialize MSI-X interrupts
1534  * @adapter: board private structure
1535  *
1536  * ixgbevf_request_msix_irqs allocates MSI-X vectors and requests
1537  * interrupts from the kernel.
1538  **/
1539 static int ixgbevf_request_msix_irqs(struct ixgbevf_adapter *adapter)
1540 {
1541         struct net_device *netdev = adapter->netdev;
1542         int q_vectors = adapter->num_msix_vectors - NON_Q_VECTORS;
1543         unsigned int ri = 0, ti = 0;
1544         int vector, err;
1545
1546         for (vector = 0; vector < q_vectors; vector++) {
1547                 struct ixgbevf_q_vector *q_vector = adapter->q_vector[vector];
1548                 struct msix_entry *entry = &adapter->msix_entries[vector];
1549
1550                 if (q_vector->tx.ring && q_vector->rx.ring) {
1551                         snprintf(q_vector->name, sizeof(q_vector->name),
1552                                  "%s-TxRx-%u", netdev->name, ri++);
1553                         ti++;
1554                 } else if (q_vector->rx.ring) {
1555                         snprintf(q_vector->name, sizeof(q_vector->name),
1556                                  "%s-rx-%u", netdev->name, ri++);
1557                 } else if (q_vector->tx.ring) {
1558                         snprintf(q_vector->name, sizeof(q_vector->name),
1559                                  "%s-tx-%u", netdev->name, ti++);
1560                 } else {
1561                         /* skip this unused q_vector */
1562                         continue;
1563                 }
1564                 err = request_irq(entry->vector, &ixgbevf_msix_clean_rings, 0,
1565                                   q_vector->name, q_vector);
1566                 if (err) {
1567                         hw_dbg(&adapter->hw,
1568                                "request_irq failed for MSIX interrupt Error: %d\n",
1569                                err);
1570                         goto free_queue_irqs;
1571                 }
1572         }
1573
1574         err = request_irq(adapter->msix_entries[vector].vector,
1575                           &ixgbevf_msix_other, 0, netdev->name, adapter);
1576         if (err) {
1577                 hw_dbg(&adapter->hw, "request_irq for msix_other failed: %d\n",
1578                        err);
1579                 goto free_queue_irqs;
1580         }
1581
1582         return 0;
1583
1584 free_queue_irqs:
1585         while (vector) {
1586                 vector--;
1587                 free_irq(adapter->msix_entries[vector].vector,
1588                          adapter->q_vector[vector]);
1589         }
1590         /* This failure is non-recoverable - it indicates the system is
1591          * out of MSIX vector resources and the VF driver cannot run
1592          * without them.  Set the number of msix vectors to zero
1593          * indicating that not enough can be allocated.  The error
1594          * will be returned to the user indicating device open failed.
1595          * Any further attempts to force the driver to open will also
1596          * fail.  The only way to recover is to unload the driver and
1597          * reload it again.  If the system has recovered some MSIX
1598          * vectors then it may succeed.
1599          */
1600         adapter->num_msix_vectors = 0;
1601         return err;
1602 }
1603
1604 /**
1605  * ixgbevf_request_irq - initialize interrupts
1606  * @adapter: board private structure
1607  *
1608  * Attempts to configure interrupts using the best available
1609  * capabilities of the hardware and kernel.
1610  **/
1611 static int ixgbevf_request_irq(struct ixgbevf_adapter *adapter)
1612 {
1613         int err = ixgbevf_request_msix_irqs(adapter);
1614
1615         if (err)
1616                 hw_dbg(&adapter->hw, "request_irq failed, Error %d\n", err);
1617
1618         return err;
1619 }
1620
1621 static void ixgbevf_free_irq(struct ixgbevf_adapter *adapter)
1622 {
1623         int i, q_vectors;
1624
1625         if (!adapter->msix_entries)
1626                 return;
1627
1628         q_vectors = adapter->num_msix_vectors;
1629         i = q_vectors - 1;
1630
1631         free_irq(adapter->msix_entries[i].vector, adapter);
1632         i--;
1633
1634         for (; i >= 0; i--) {
1635                 /* free only the irqs that were actually requested */
1636                 if (!adapter->q_vector[i]->rx.ring &&
1637                     !adapter->q_vector[i]->tx.ring)
1638                         continue;
1639
1640                 free_irq(adapter->msix_entries[i].vector,
1641                          adapter->q_vector[i]);
1642         }
1643 }
1644
1645 /**
1646  * ixgbevf_irq_disable - Mask off interrupt generation on the NIC
1647  * @adapter: board private structure
1648  **/
1649 static inline void ixgbevf_irq_disable(struct ixgbevf_adapter *adapter)
1650 {
1651         struct ixgbe_hw *hw = &adapter->hw;
1652         int i;
1653
1654         IXGBE_WRITE_REG(hw, IXGBE_VTEIAM, 0);
1655         IXGBE_WRITE_REG(hw, IXGBE_VTEIMC, ~0);
1656         IXGBE_WRITE_REG(hw, IXGBE_VTEIAC, 0);
1657
1658         IXGBE_WRITE_FLUSH(hw);
1659
1660         for (i = 0; i < adapter->num_msix_vectors; i++)
1661                 synchronize_irq(adapter->msix_entries[i].vector);
1662 }
1663
1664 /**
1665  * ixgbevf_irq_enable - Enable default interrupt generation settings
1666  * @adapter: board private structure
1667  **/
1668 static inline void ixgbevf_irq_enable(struct ixgbevf_adapter *adapter)
1669 {
1670         struct ixgbe_hw *hw = &adapter->hw;
1671
1672         IXGBE_WRITE_REG(hw, IXGBE_VTEIAM, adapter->eims_enable_mask);
1673         IXGBE_WRITE_REG(hw, IXGBE_VTEIAC, adapter->eims_enable_mask);
1674         IXGBE_WRITE_REG(hw, IXGBE_VTEIMS, adapter->eims_enable_mask);
1675 }
1676
1677 /**
1678  * ixgbevf_configure_tx_ring - Configure 82599 VF Tx ring after Reset
1679  * @adapter: board private structure
1680  * @ring: structure containing ring specific data
1681  *
1682  * Configure the Tx descriptor ring after a reset.
1683  **/
1684 static void ixgbevf_configure_tx_ring(struct ixgbevf_adapter *adapter,
1685                                       struct ixgbevf_ring *ring)
1686 {
1687         struct ixgbe_hw *hw = &adapter->hw;
1688         u64 tdba = ring->dma;
1689         int wait_loop = 10;
1690         u32 txdctl = IXGBE_TXDCTL_ENABLE;
1691         u8 reg_idx = ring->reg_idx;
1692
1693         /* disable queue to avoid issues while updating state */
1694         IXGBE_WRITE_REG(hw, IXGBE_VFTXDCTL(reg_idx), IXGBE_TXDCTL_SWFLSH);
1695         IXGBE_WRITE_FLUSH(hw);
1696
1697         IXGBE_WRITE_REG(hw, IXGBE_VFTDBAL(reg_idx), tdba & DMA_BIT_MASK(32));
1698         IXGBE_WRITE_REG(hw, IXGBE_VFTDBAH(reg_idx), tdba >> 32);
1699         IXGBE_WRITE_REG(hw, IXGBE_VFTDLEN(reg_idx),
1700                         ring->count * sizeof(union ixgbe_adv_tx_desc));
1701
1702         /* disable head writeback */
1703         IXGBE_WRITE_REG(hw, IXGBE_VFTDWBAH(reg_idx), 0);
1704         IXGBE_WRITE_REG(hw, IXGBE_VFTDWBAL(reg_idx), 0);
1705
1706         /* enable relaxed ordering */
1707         IXGBE_WRITE_REG(hw, IXGBE_VFDCA_TXCTRL(reg_idx),
1708                         (IXGBE_DCA_TXCTRL_DESC_RRO_EN |
1709                          IXGBE_DCA_TXCTRL_DATA_RRO_EN));
1710
1711         /* reset head and tail pointers */
1712         IXGBE_WRITE_REG(hw, IXGBE_VFTDH(reg_idx), 0);
1713         IXGBE_WRITE_REG(hw, IXGBE_VFTDT(reg_idx), 0);
1714         ring->tail = adapter->io_addr + IXGBE_VFTDT(reg_idx);
1715
1716         /* reset ntu and ntc to place SW in sync with hardwdare */
1717         ring->next_to_clean = 0;
1718         ring->next_to_use = 0;
1719
1720         /* In order to avoid issues WTHRESH + PTHRESH should always be equal
1721          * to or less than the number of on chip descriptors, which is
1722          * currently 40.
1723          */
1724         txdctl |= (8 << 16);    /* WTHRESH = 8 */
1725
1726         /* Setting PTHRESH to 32 both improves performance */
1727         txdctl |= (1u << 8) |    /* HTHRESH = 1 */
1728                    32;           /* PTHRESH = 32 */
1729
1730         /* reinitialize tx_buffer_info */
1731         memset(ring->tx_buffer_info, 0,
1732                sizeof(struct ixgbevf_tx_buffer) * ring->count);
1733
1734         clear_bit(__IXGBEVF_HANG_CHECK_ARMED, &ring->state);
1735         clear_bit(__IXGBEVF_TX_XDP_RING_PRIMED, &ring->state);
1736
1737         IXGBE_WRITE_REG(hw, IXGBE_VFTXDCTL(reg_idx), txdctl);
1738
1739         /* poll to verify queue is enabled */
1740         do {
1741                 usleep_range(1000, 2000);
1742                 txdctl = IXGBE_READ_REG(hw, IXGBE_VFTXDCTL(reg_idx));
1743         }  while (--wait_loop && !(txdctl & IXGBE_TXDCTL_ENABLE));
1744         if (!wait_loop)
1745                 hw_dbg(hw, "Could not enable Tx Queue %d\n", reg_idx);
1746 }
1747
1748 /**
1749  * ixgbevf_configure_tx - Configure 82599 VF Transmit Unit after Reset
1750  * @adapter: board private structure
1751  *
1752  * Configure the Tx unit of the MAC after a reset.
1753  **/
1754 static void ixgbevf_configure_tx(struct ixgbevf_adapter *adapter)
1755 {
1756         u32 i;
1757
1758         /* Setup the HW Tx Head and Tail descriptor pointers */
1759         for (i = 0; i < adapter->num_tx_queues; i++)
1760                 ixgbevf_configure_tx_ring(adapter, adapter->tx_ring[i]);
1761         for (i = 0; i < adapter->num_xdp_queues; i++)
1762                 ixgbevf_configure_tx_ring(adapter, adapter->xdp_ring[i]);
1763 }
1764
1765 #define IXGBE_SRRCTL_BSIZEHDRSIZE_SHIFT 2
1766
1767 static void ixgbevf_configure_srrctl(struct ixgbevf_adapter *adapter,
1768                                      struct ixgbevf_ring *ring, int index)
1769 {
1770         struct ixgbe_hw *hw = &adapter->hw;
1771         u32 srrctl;
1772
1773         srrctl = IXGBE_SRRCTL_DROP_EN;
1774
1775         srrctl |= IXGBEVF_RX_HDR_SIZE << IXGBE_SRRCTL_BSIZEHDRSIZE_SHIFT;
1776         if (ring_uses_large_buffer(ring))
1777                 srrctl |= IXGBEVF_RXBUFFER_3072 >> IXGBE_SRRCTL_BSIZEPKT_SHIFT;
1778         else
1779                 srrctl |= IXGBEVF_RXBUFFER_2048 >> IXGBE_SRRCTL_BSIZEPKT_SHIFT;
1780         srrctl |= IXGBE_SRRCTL_DESCTYPE_ADV_ONEBUF;
1781
1782         IXGBE_WRITE_REG(hw, IXGBE_VFSRRCTL(index), srrctl);
1783 }
1784
1785 static void ixgbevf_setup_psrtype(struct ixgbevf_adapter *adapter)
1786 {
1787         struct ixgbe_hw *hw = &adapter->hw;
1788
1789         /* PSRTYPE must be initialized in 82599 */
1790         u32 psrtype = IXGBE_PSRTYPE_TCPHDR | IXGBE_PSRTYPE_UDPHDR |
1791                       IXGBE_PSRTYPE_IPV4HDR | IXGBE_PSRTYPE_IPV6HDR |
1792                       IXGBE_PSRTYPE_L2HDR;
1793
1794         if (adapter->num_rx_queues > 1)
1795                 psrtype |= BIT(29);
1796
1797         IXGBE_WRITE_REG(hw, IXGBE_VFPSRTYPE, psrtype);
1798 }
1799
1800 #define IXGBEVF_MAX_RX_DESC_POLL 10
1801 static void ixgbevf_disable_rx_queue(struct ixgbevf_adapter *adapter,
1802                                      struct ixgbevf_ring *ring)
1803 {
1804         struct ixgbe_hw *hw = &adapter->hw;
1805         int wait_loop = IXGBEVF_MAX_RX_DESC_POLL;
1806         u32 rxdctl;
1807         u8 reg_idx = ring->reg_idx;
1808
1809         if (IXGBE_REMOVED(hw->hw_addr))
1810                 return;
1811         rxdctl = IXGBE_READ_REG(hw, IXGBE_VFRXDCTL(reg_idx));
1812         rxdctl &= ~IXGBE_RXDCTL_ENABLE;
1813
1814         /* write value back with RXDCTL.ENABLE bit cleared */
1815         IXGBE_WRITE_REG(hw, IXGBE_VFRXDCTL(reg_idx), rxdctl);
1816
1817         /* the hardware may take up to 100us to really disable the Rx queue */
1818         do {
1819                 udelay(10);
1820                 rxdctl = IXGBE_READ_REG(hw, IXGBE_VFRXDCTL(reg_idx));
1821         } while (--wait_loop && (rxdctl & IXGBE_RXDCTL_ENABLE));
1822
1823         if (!wait_loop)
1824                 pr_err("RXDCTL.ENABLE queue %d not cleared while polling\n",
1825                        reg_idx);
1826 }
1827
1828 static void ixgbevf_rx_desc_queue_enable(struct ixgbevf_adapter *adapter,
1829                                          struct ixgbevf_ring *ring)
1830 {
1831         struct ixgbe_hw *hw = &adapter->hw;
1832         int wait_loop = IXGBEVF_MAX_RX_DESC_POLL;
1833         u32 rxdctl;
1834         u8 reg_idx = ring->reg_idx;
1835
1836         if (IXGBE_REMOVED(hw->hw_addr))
1837                 return;
1838         do {
1839                 usleep_range(1000, 2000);
1840                 rxdctl = IXGBE_READ_REG(hw, IXGBE_VFRXDCTL(reg_idx));
1841         } while (--wait_loop && !(rxdctl & IXGBE_RXDCTL_ENABLE));
1842
1843         if (!wait_loop)
1844                 pr_err("RXDCTL.ENABLE queue %d not set while polling\n",
1845                        reg_idx);
1846 }
1847
1848 /**
1849  * ixgbevf_init_rss_key - Initialize adapter RSS key
1850  * @adapter: device handle
1851  *
1852  * Allocates and initializes the RSS key if it is not allocated.
1853  **/
1854 static inline int ixgbevf_init_rss_key(struct ixgbevf_adapter *adapter)
1855 {
1856         u32 *rss_key;
1857
1858         if (!adapter->rss_key) {
1859                 rss_key = kzalloc(IXGBEVF_RSS_HASH_KEY_SIZE, GFP_KERNEL);
1860                 if (unlikely(!rss_key))
1861                         return -ENOMEM;
1862
1863                 netdev_rss_key_fill(rss_key, IXGBEVF_RSS_HASH_KEY_SIZE);
1864                 adapter->rss_key = rss_key;
1865         }
1866
1867         return 0;
1868 }
1869
1870 static void ixgbevf_setup_vfmrqc(struct ixgbevf_adapter *adapter)
1871 {
1872         struct ixgbe_hw *hw = &adapter->hw;
1873         u32 vfmrqc = 0, vfreta = 0;
1874         u16 rss_i = adapter->num_rx_queues;
1875         u8 i, j;
1876
1877         /* Fill out hash function seeds */
1878         for (i = 0; i < IXGBEVF_VFRSSRK_REGS; i++)
1879                 IXGBE_WRITE_REG(hw, IXGBE_VFRSSRK(i), *(adapter->rss_key + i));
1880
1881         for (i = 0, j = 0; i < IXGBEVF_X550_VFRETA_SIZE; i++, j++) {
1882                 if (j == rss_i)
1883                         j = 0;
1884
1885                 adapter->rss_indir_tbl[i] = j;
1886
1887                 vfreta |= j << (i & 0x3) * 8;
1888                 if ((i & 3) == 3) {
1889                         IXGBE_WRITE_REG(hw, IXGBE_VFRETA(i >> 2), vfreta);
1890                         vfreta = 0;
1891                 }
1892         }
1893
1894         /* Perform hash on these packet types */
1895         vfmrqc |= IXGBE_VFMRQC_RSS_FIELD_IPV4 |
1896                 IXGBE_VFMRQC_RSS_FIELD_IPV4_TCP |
1897                 IXGBE_VFMRQC_RSS_FIELD_IPV6 |
1898                 IXGBE_VFMRQC_RSS_FIELD_IPV6_TCP;
1899
1900         vfmrqc |= IXGBE_VFMRQC_RSSEN;
1901
1902         IXGBE_WRITE_REG(hw, IXGBE_VFMRQC, vfmrqc);
1903 }
1904
1905 static void ixgbevf_configure_rx_ring(struct ixgbevf_adapter *adapter,
1906                                       struct ixgbevf_ring *ring)
1907 {
1908         struct ixgbe_hw *hw = &adapter->hw;
1909         union ixgbe_adv_rx_desc *rx_desc;
1910         u64 rdba = ring->dma;
1911         u32 rxdctl;
1912         u8 reg_idx = ring->reg_idx;
1913
1914         /* disable queue to avoid issues while updating state */
1915         rxdctl = IXGBE_READ_REG(hw, IXGBE_VFRXDCTL(reg_idx));
1916         ixgbevf_disable_rx_queue(adapter, ring);
1917
1918         IXGBE_WRITE_REG(hw, IXGBE_VFRDBAL(reg_idx), rdba & DMA_BIT_MASK(32));
1919         IXGBE_WRITE_REG(hw, IXGBE_VFRDBAH(reg_idx), rdba >> 32);
1920         IXGBE_WRITE_REG(hw, IXGBE_VFRDLEN(reg_idx),
1921                         ring->count * sizeof(union ixgbe_adv_rx_desc));
1922
1923 #ifndef CONFIG_SPARC
1924         /* enable relaxed ordering */
1925         IXGBE_WRITE_REG(hw, IXGBE_VFDCA_RXCTRL(reg_idx),
1926                         IXGBE_DCA_RXCTRL_DESC_RRO_EN);
1927 #else
1928         IXGBE_WRITE_REG(hw, IXGBE_VFDCA_RXCTRL(reg_idx),
1929                         IXGBE_DCA_RXCTRL_DESC_RRO_EN |
1930                         IXGBE_DCA_RXCTRL_DATA_WRO_EN);
1931 #endif
1932
1933         /* reset head and tail pointers */
1934         IXGBE_WRITE_REG(hw, IXGBE_VFRDH(reg_idx), 0);
1935         IXGBE_WRITE_REG(hw, IXGBE_VFRDT(reg_idx), 0);
1936         ring->tail = adapter->io_addr + IXGBE_VFRDT(reg_idx);
1937
1938         /* initialize rx_buffer_info */
1939         memset(ring->rx_buffer_info, 0,
1940                sizeof(struct ixgbevf_rx_buffer) * ring->count);
1941
1942         /* initialize Rx descriptor 0 */
1943         rx_desc = IXGBEVF_RX_DESC(ring, 0);
1944         rx_desc->wb.upper.length = 0;
1945
1946         /* reset ntu and ntc to place SW in sync with hardwdare */
1947         ring->next_to_clean = 0;
1948         ring->next_to_use = 0;
1949         ring->next_to_alloc = 0;
1950
1951         ixgbevf_configure_srrctl(adapter, ring, reg_idx);
1952
1953         /* RXDCTL.RLPML does not work on 82599 */
1954         if (adapter->hw.mac.type != ixgbe_mac_82599_vf) {
1955                 rxdctl &= ~(IXGBE_RXDCTL_RLPMLMASK |
1956                             IXGBE_RXDCTL_RLPML_EN);
1957
1958 #if (PAGE_SIZE < 8192)
1959                 /* Limit the maximum frame size so we don't overrun the skb */
1960                 if (ring_uses_build_skb(ring) &&
1961                     !ring_uses_large_buffer(ring))
1962                         rxdctl |= IXGBEVF_MAX_FRAME_BUILD_SKB |
1963                                   IXGBE_RXDCTL_RLPML_EN;
1964 #endif
1965         }
1966
1967         rxdctl |= IXGBE_RXDCTL_ENABLE | IXGBE_RXDCTL_VME;
1968         IXGBE_WRITE_REG(hw, IXGBE_VFRXDCTL(reg_idx), rxdctl);
1969
1970         ixgbevf_rx_desc_queue_enable(adapter, ring);
1971         ixgbevf_alloc_rx_buffers(ring, ixgbevf_desc_unused(ring));
1972 }
1973
1974 static void ixgbevf_set_rx_buffer_len(struct ixgbevf_adapter *adapter,
1975                                       struct ixgbevf_ring *rx_ring)
1976 {
1977         struct net_device *netdev = adapter->netdev;
1978         unsigned int max_frame = netdev->mtu + ETH_HLEN + ETH_FCS_LEN;
1979
1980         /* set build_skb and buffer size flags */
1981         clear_ring_build_skb_enabled(rx_ring);
1982         clear_ring_uses_large_buffer(rx_ring);
1983
1984         if (adapter->flags & IXGBEVF_FLAGS_LEGACY_RX)
1985                 return;
1986
1987         if (PAGE_SIZE < 8192)
1988                 if (max_frame > IXGBEVF_MAX_FRAME_BUILD_SKB)
1989                         set_ring_uses_large_buffer(rx_ring);
1990
1991         /* 82599 can't rely on RXDCTL.RLPML to restrict the size of the frame */
1992         if (adapter->hw.mac.type == ixgbe_mac_82599_vf && !ring_uses_large_buffer(rx_ring))
1993                 return;
1994
1995         set_ring_build_skb_enabled(rx_ring);
1996 }
1997
1998 /**
1999  * ixgbevf_configure_rx - Configure 82599 VF Receive Unit after Reset
2000  * @adapter: board private structure
2001  *
2002  * Configure the Rx unit of the MAC after a reset.
2003  **/
2004 static void ixgbevf_configure_rx(struct ixgbevf_adapter *adapter)
2005 {
2006         struct ixgbe_hw *hw = &adapter->hw;
2007         struct net_device *netdev = adapter->netdev;
2008         int i, ret;
2009
2010         ixgbevf_setup_psrtype(adapter);
2011         if (hw->mac.type >= ixgbe_mac_X550_vf)
2012                 ixgbevf_setup_vfmrqc(adapter);
2013
2014         spin_lock_bh(&adapter->mbx_lock);
2015         /* notify the PF of our intent to use this size of frame */
2016         ret = hw->mac.ops.set_rlpml(hw, netdev->mtu + ETH_HLEN + ETH_FCS_LEN);
2017         spin_unlock_bh(&adapter->mbx_lock);
2018         if (ret)
2019                 dev_err(&adapter->pdev->dev,
2020                         "Failed to set MTU at %d\n", netdev->mtu);
2021
2022         /* Setup the HW Rx Head and Tail Descriptor Pointers and
2023          * the Base and Length of the Rx Descriptor Ring
2024          */
2025         for (i = 0; i < adapter->num_rx_queues; i++) {
2026                 struct ixgbevf_ring *rx_ring = adapter->rx_ring[i];
2027
2028                 ixgbevf_set_rx_buffer_len(adapter, rx_ring);
2029                 ixgbevf_configure_rx_ring(adapter, rx_ring);
2030         }
2031 }
2032
2033 static int ixgbevf_vlan_rx_add_vid(struct net_device *netdev,
2034                                    __be16 proto, u16 vid)
2035 {
2036         struct ixgbevf_adapter *adapter = netdev_priv(netdev);
2037         struct ixgbe_hw *hw = &adapter->hw;
2038         int err;
2039
2040         spin_lock_bh(&adapter->mbx_lock);
2041
2042         /* add VID to filter table */
2043         err = hw->mac.ops.set_vfta(hw, vid, 0, true);
2044
2045         spin_unlock_bh(&adapter->mbx_lock);
2046
2047         /* translate error return types so error makes sense */
2048         if (err == IXGBE_ERR_MBX)
2049                 return -EIO;
2050
2051         if (err == IXGBE_ERR_INVALID_ARGUMENT)
2052                 return -EACCES;
2053
2054         set_bit(vid, adapter->active_vlans);
2055
2056         return err;
2057 }
2058
2059 static int ixgbevf_vlan_rx_kill_vid(struct net_device *netdev,
2060                                     __be16 proto, u16 vid)
2061 {
2062         struct ixgbevf_adapter *adapter = netdev_priv(netdev);
2063         struct ixgbe_hw *hw = &adapter->hw;
2064         int err;
2065
2066         spin_lock_bh(&adapter->mbx_lock);
2067
2068         /* remove VID from filter table */
2069         err = hw->mac.ops.set_vfta(hw, vid, 0, false);
2070
2071         spin_unlock_bh(&adapter->mbx_lock);
2072
2073         clear_bit(vid, adapter->active_vlans);
2074
2075         return err;
2076 }
2077
2078 static void ixgbevf_restore_vlan(struct ixgbevf_adapter *adapter)
2079 {
2080         u16 vid;
2081
2082         for_each_set_bit(vid, adapter->active_vlans, VLAN_N_VID)
2083                 ixgbevf_vlan_rx_add_vid(adapter->netdev,
2084                                         htons(ETH_P_8021Q), vid);
2085 }
2086
2087 static int ixgbevf_write_uc_addr_list(struct net_device *netdev)
2088 {
2089         struct ixgbevf_adapter *adapter = netdev_priv(netdev);
2090         struct ixgbe_hw *hw = &adapter->hw;
2091         int count = 0;
2092
2093         if (!netdev_uc_empty(netdev)) {
2094                 struct netdev_hw_addr *ha;
2095
2096                 netdev_for_each_uc_addr(ha, netdev) {
2097                         hw->mac.ops.set_uc_addr(hw, ++count, ha->addr);
2098                         udelay(200);
2099                 }
2100         } else {
2101                 /* If the list is empty then send message to PF driver to
2102                  * clear all MAC VLANs on this VF.
2103                  */
2104                 hw->mac.ops.set_uc_addr(hw, 0, NULL);
2105         }
2106
2107         return count;
2108 }
2109
2110 /**
2111  * ixgbevf_set_rx_mode - Multicast and unicast set
2112  * @netdev: network interface device structure
2113  *
2114  * The set_rx_method entry point is called whenever the multicast address
2115  * list, unicast address list or the network interface flags are updated.
2116  * This routine is responsible for configuring the hardware for proper
2117  * multicast mode and configuring requested unicast filters.
2118  **/
2119 static void ixgbevf_set_rx_mode(struct net_device *netdev)
2120 {
2121         struct ixgbevf_adapter *adapter = netdev_priv(netdev);
2122         struct ixgbe_hw *hw = &adapter->hw;
2123         unsigned int flags = netdev->flags;
2124         int xcast_mode;
2125
2126         /* request the most inclusive mode we need */
2127         if (flags & IFF_PROMISC)
2128                 xcast_mode = IXGBEVF_XCAST_MODE_PROMISC;
2129         else if (flags & IFF_ALLMULTI)
2130                 xcast_mode = IXGBEVF_XCAST_MODE_ALLMULTI;
2131         else if (flags & (IFF_BROADCAST | IFF_MULTICAST))
2132                 xcast_mode = IXGBEVF_XCAST_MODE_MULTI;
2133         else
2134                 xcast_mode = IXGBEVF_XCAST_MODE_NONE;
2135
2136         spin_lock_bh(&adapter->mbx_lock);
2137
2138         hw->mac.ops.update_xcast_mode(hw, xcast_mode);
2139
2140         /* reprogram multicast list */
2141         hw->mac.ops.update_mc_addr_list(hw, netdev);
2142
2143         ixgbevf_write_uc_addr_list(netdev);
2144
2145         spin_unlock_bh(&adapter->mbx_lock);
2146 }
2147
2148 static void ixgbevf_napi_enable_all(struct ixgbevf_adapter *adapter)
2149 {
2150         int q_idx;
2151         struct ixgbevf_q_vector *q_vector;
2152         int q_vectors = adapter->num_msix_vectors - NON_Q_VECTORS;
2153
2154         for (q_idx = 0; q_idx < q_vectors; q_idx++) {
2155                 q_vector = adapter->q_vector[q_idx];
2156                 napi_enable(&q_vector->napi);
2157         }
2158 }
2159
2160 static void ixgbevf_napi_disable_all(struct ixgbevf_adapter *adapter)
2161 {
2162         int q_idx;
2163         struct ixgbevf_q_vector *q_vector;
2164         int q_vectors = adapter->num_msix_vectors - NON_Q_VECTORS;
2165
2166         for (q_idx = 0; q_idx < q_vectors; q_idx++) {
2167                 q_vector = adapter->q_vector[q_idx];
2168                 napi_disable(&q_vector->napi);
2169         }
2170 }
2171
2172 static int ixgbevf_configure_dcb(struct ixgbevf_adapter *adapter)
2173 {
2174         struct ixgbe_hw *hw = &adapter->hw;
2175         unsigned int def_q = 0;
2176         unsigned int num_tcs = 0;
2177         unsigned int num_rx_queues = adapter->num_rx_queues;
2178         unsigned int num_tx_queues = adapter->num_tx_queues;
2179         int err;
2180
2181         spin_lock_bh(&adapter->mbx_lock);
2182
2183         /* fetch queue configuration from the PF */
2184         err = ixgbevf_get_queues(hw, &num_tcs, &def_q);
2185
2186         spin_unlock_bh(&adapter->mbx_lock);
2187
2188         if (err)
2189                 return err;
2190
2191         if (num_tcs > 1) {
2192                 /* we need only one Tx queue */
2193                 num_tx_queues = 1;
2194
2195                 /* update default Tx ring register index */
2196                 adapter->tx_ring[0]->reg_idx = def_q;
2197
2198                 /* we need as many queues as traffic classes */
2199                 num_rx_queues = num_tcs;
2200         }
2201
2202         /* if we have a bad config abort request queue reset */
2203         if ((adapter->num_rx_queues != num_rx_queues) ||
2204             (adapter->num_tx_queues != num_tx_queues)) {
2205                 /* force mailbox timeout to prevent further messages */
2206                 hw->mbx.timeout = 0;
2207
2208                 /* wait for watchdog to come around and bail us out */
2209                 set_bit(__IXGBEVF_QUEUE_RESET_REQUESTED, &adapter->state);
2210         }
2211
2212         return 0;
2213 }
2214
2215 static void ixgbevf_configure(struct ixgbevf_adapter *adapter)
2216 {
2217         ixgbevf_configure_dcb(adapter);
2218
2219         ixgbevf_set_rx_mode(adapter->netdev);
2220
2221         ixgbevf_restore_vlan(adapter);
2222         ixgbevf_ipsec_restore(adapter);
2223
2224         ixgbevf_configure_tx(adapter);
2225         ixgbevf_configure_rx(adapter);
2226 }
2227
2228 static void ixgbevf_save_reset_stats(struct ixgbevf_adapter *adapter)
2229 {
2230         /* Only save pre-reset stats if there are some */
2231         if (adapter->stats.vfgprc || adapter->stats.vfgptc) {
2232                 adapter->stats.saved_reset_vfgprc += adapter->stats.vfgprc -
2233                         adapter->stats.base_vfgprc;
2234                 adapter->stats.saved_reset_vfgptc += adapter->stats.vfgptc -
2235                         adapter->stats.base_vfgptc;
2236                 adapter->stats.saved_reset_vfgorc += adapter->stats.vfgorc -
2237                         adapter->stats.base_vfgorc;
2238                 adapter->stats.saved_reset_vfgotc += adapter->stats.vfgotc -
2239                         adapter->stats.base_vfgotc;
2240                 adapter->stats.saved_reset_vfmprc += adapter->stats.vfmprc -
2241                         adapter->stats.base_vfmprc;
2242         }
2243 }
2244
2245 static void ixgbevf_init_last_counter_stats(struct ixgbevf_adapter *adapter)
2246 {
2247         struct ixgbe_hw *hw = &adapter->hw;
2248
2249         adapter->stats.last_vfgprc = IXGBE_READ_REG(hw, IXGBE_VFGPRC);
2250         adapter->stats.last_vfgorc = IXGBE_READ_REG(hw, IXGBE_VFGORC_LSB);
2251         adapter->stats.last_vfgorc |=
2252                 (((u64)(IXGBE_READ_REG(hw, IXGBE_VFGORC_MSB))) << 32);
2253         adapter->stats.last_vfgptc = IXGBE_READ_REG(hw, IXGBE_VFGPTC);
2254         adapter->stats.last_vfgotc = IXGBE_READ_REG(hw, IXGBE_VFGOTC_LSB);
2255         adapter->stats.last_vfgotc |=
2256                 (((u64)(IXGBE_READ_REG(hw, IXGBE_VFGOTC_MSB))) << 32);
2257         adapter->stats.last_vfmprc = IXGBE_READ_REG(hw, IXGBE_VFMPRC);
2258
2259         adapter->stats.base_vfgprc = adapter->stats.last_vfgprc;
2260         adapter->stats.base_vfgorc = adapter->stats.last_vfgorc;
2261         adapter->stats.base_vfgptc = adapter->stats.last_vfgptc;
2262         adapter->stats.base_vfgotc = adapter->stats.last_vfgotc;
2263         adapter->stats.base_vfmprc = adapter->stats.last_vfmprc;
2264 }
2265
2266 static void ixgbevf_negotiate_api(struct ixgbevf_adapter *adapter)
2267 {
2268         struct ixgbe_hw *hw = &adapter->hw;
2269         static const int api[] = {
2270                 ixgbe_mbox_api_15,
2271                 ixgbe_mbox_api_14,
2272                 ixgbe_mbox_api_13,
2273                 ixgbe_mbox_api_12,
2274                 ixgbe_mbox_api_11,
2275                 ixgbe_mbox_api_10,
2276                 ixgbe_mbox_api_unknown
2277         };
2278         int err, idx = 0;
2279
2280         spin_lock_bh(&adapter->mbx_lock);
2281
2282         while (api[idx] != ixgbe_mbox_api_unknown) {
2283                 err = hw->mac.ops.negotiate_api_version(hw, api[idx]);
2284                 if (!err)
2285                         break;
2286                 idx++;
2287         }
2288
2289         if (hw->api_version >= ixgbe_mbox_api_15) {
2290                 hw->mbx.ops.init_params(hw);
2291                 memcpy(&hw->mbx.ops, &ixgbevf_mbx_ops,
2292                        sizeof(struct ixgbe_mbx_operations));
2293         }
2294
2295         spin_unlock_bh(&adapter->mbx_lock);
2296 }
2297
2298 static void ixgbevf_up_complete(struct ixgbevf_adapter *adapter)
2299 {
2300         struct net_device *netdev = adapter->netdev;
2301         struct ixgbe_hw *hw = &adapter->hw;
2302
2303         ixgbevf_configure_msix(adapter);
2304
2305         spin_lock_bh(&adapter->mbx_lock);
2306
2307         if (is_valid_ether_addr(hw->mac.addr))
2308                 hw->mac.ops.set_rar(hw, 0, hw->mac.addr, 0);
2309         else
2310                 hw->mac.ops.set_rar(hw, 0, hw->mac.perm_addr, 0);
2311
2312         spin_unlock_bh(&adapter->mbx_lock);
2313
2314         smp_mb__before_atomic();
2315         clear_bit(__IXGBEVF_DOWN, &adapter->state);
2316         ixgbevf_napi_enable_all(adapter);
2317
2318         /* clear any pending interrupts, may auto mask */
2319         IXGBE_READ_REG(hw, IXGBE_VTEICR);
2320         ixgbevf_irq_enable(adapter);
2321
2322         /* enable transmits */
2323         netif_tx_start_all_queues(netdev);
2324
2325         ixgbevf_save_reset_stats(adapter);
2326         ixgbevf_init_last_counter_stats(adapter);
2327
2328         hw->mac.get_link_status = 1;
2329         mod_timer(&adapter->service_timer, jiffies);
2330 }
2331
2332 void ixgbevf_up(struct ixgbevf_adapter *adapter)
2333 {
2334         ixgbevf_configure(adapter);
2335
2336         ixgbevf_up_complete(adapter);
2337 }
2338
2339 /**
2340  * ixgbevf_clean_rx_ring - Free Rx Buffers per Queue
2341  * @rx_ring: ring to free buffers from
2342  **/
2343 static void ixgbevf_clean_rx_ring(struct ixgbevf_ring *rx_ring)
2344 {
2345         u16 i = rx_ring->next_to_clean;
2346
2347         /* Free Rx ring sk_buff */
2348         if (rx_ring->skb) {
2349                 dev_kfree_skb(rx_ring->skb);
2350                 rx_ring->skb = NULL;
2351         }
2352
2353         /* Free all the Rx ring pages */
2354         while (i != rx_ring->next_to_alloc) {
2355                 struct ixgbevf_rx_buffer *rx_buffer;
2356
2357                 rx_buffer = &rx_ring->rx_buffer_info[i];
2358
2359                 /* Invalidate cache lines that may have been written to by
2360                  * device so that we avoid corrupting memory.
2361                  */
2362                 dma_sync_single_range_for_cpu(rx_ring->dev,
2363                                               rx_buffer->dma,
2364                                               rx_buffer->page_offset,
2365                                               ixgbevf_rx_bufsz(rx_ring),
2366                                               DMA_FROM_DEVICE);
2367
2368                 /* free resources associated with mapping */
2369                 dma_unmap_page_attrs(rx_ring->dev,
2370                                      rx_buffer->dma,
2371                                      ixgbevf_rx_pg_size(rx_ring),
2372                                      DMA_FROM_DEVICE,
2373                                      IXGBEVF_RX_DMA_ATTR);
2374
2375                 __page_frag_cache_drain(rx_buffer->page,
2376                                         rx_buffer->pagecnt_bias);
2377
2378                 i++;
2379                 if (i == rx_ring->count)
2380                         i = 0;
2381         }
2382
2383         rx_ring->next_to_alloc = 0;
2384         rx_ring->next_to_clean = 0;
2385         rx_ring->next_to_use = 0;
2386 }
2387
2388 /**
2389  * ixgbevf_clean_tx_ring - Free Tx Buffers
2390  * @tx_ring: ring to be cleaned
2391  **/
2392 static void ixgbevf_clean_tx_ring(struct ixgbevf_ring *tx_ring)
2393 {
2394         u16 i = tx_ring->next_to_clean;
2395         struct ixgbevf_tx_buffer *tx_buffer = &tx_ring->tx_buffer_info[i];
2396
2397         while (i != tx_ring->next_to_use) {
2398                 union ixgbe_adv_tx_desc *eop_desc, *tx_desc;
2399
2400                 /* Free all the Tx ring sk_buffs */
2401                 if (ring_is_xdp(tx_ring))
2402                         page_frag_free(tx_buffer->data);
2403                 else
2404                         dev_kfree_skb_any(tx_buffer->skb);
2405
2406                 /* unmap skb header data */
2407                 dma_unmap_single(tx_ring->dev,
2408                                  dma_unmap_addr(tx_buffer, dma),
2409                                  dma_unmap_len(tx_buffer, len),
2410                                  DMA_TO_DEVICE);
2411
2412                 /* check for eop_desc to determine the end of the packet */
2413                 eop_desc = tx_buffer->next_to_watch;
2414                 tx_desc = IXGBEVF_TX_DESC(tx_ring, i);
2415
2416                 /* unmap remaining buffers */
2417                 while (tx_desc != eop_desc) {
2418                         tx_buffer++;
2419                         tx_desc++;
2420                         i++;
2421                         if (unlikely(i == tx_ring->count)) {
2422                                 i = 0;
2423                                 tx_buffer = tx_ring->tx_buffer_info;
2424                                 tx_desc = IXGBEVF_TX_DESC(tx_ring, 0);
2425                         }
2426
2427                         /* unmap any remaining paged data */
2428                         if (dma_unmap_len(tx_buffer, len))
2429                                 dma_unmap_page(tx_ring->dev,
2430                                                dma_unmap_addr(tx_buffer, dma),
2431                                                dma_unmap_len(tx_buffer, len),
2432                                                DMA_TO_DEVICE);
2433                 }
2434
2435                 /* move us one more past the eop_desc for start of next pkt */
2436                 tx_buffer++;
2437                 i++;
2438                 if (unlikely(i == tx_ring->count)) {
2439                         i = 0;
2440                         tx_buffer = tx_ring->tx_buffer_info;
2441                 }
2442         }
2443
2444         /* reset next_to_use and next_to_clean */
2445         tx_ring->next_to_use = 0;
2446         tx_ring->next_to_clean = 0;
2447
2448 }
2449
2450 /**
2451  * ixgbevf_clean_all_rx_rings - Free Rx Buffers for all queues
2452  * @adapter: board private structure
2453  **/
2454 static void ixgbevf_clean_all_rx_rings(struct ixgbevf_adapter *adapter)
2455 {
2456         int i;
2457
2458         for (i = 0; i < adapter->num_rx_queues; i++)
2459                 ixgbevf_clean_rx_ring(adapter->rx_ring[i]);
2460 }
2461
2462 /**
2463  * ixgbevf_clean_all_tx_rings - Free Tx Buffers for all queues
2464  * @adapter: board private structure
2465  **/
2466 static void ixgbevf_clean_all_tx_rings(struct ixgbevf_adapter *adapter)
2467 {
2468         int i;
2469
2470         for (i = 0; i < adapter->num_tx_queues; i++)
2471                 ixgbevf_clean_tx_ring(adapter->tx_ring[i]);
2472         for (i = 0; i < adapter->num_xdp_queues; i++)
2473                 ixgbevf_clean_tx_ring(adapter->xdp_ring[i]);
2474 }
2475
2476 void ixgbevf_down(struct ixgbevf_adapter *adapter)
2477 {
2478         struct net_device *netdev = adapter->netdev;
2479         struct ixgbe_hw *hw = &adapter->hw;
2480         int i;
2481
2482         /* signal that we are down to the interrupt handler */
2483         if (test_and_set_bit(__IXGBEVF_DOWN, &adapter->state))
2484                 return; /* do nothing if already down */
2485
2486         /* disable all enabled Rx queues */
2487         for (i = 0; i < adapter->num_rx_queues; i++)
2488                 ixgbevf_disable_rx_queue(adapter, adapter->rx_ring[i]);
2489
2490         usleep_range(10000, 20000);
2491
2492         netif_tx_stop_all_queues(netdev);
2493
2494         /* call carrier off first to avoid false dev_watchdog timeouts */
2495         netif_carrier_off(netdev);
2496         netif_tx_disable(netdev);
2497
2498         ixgbevf_irq_disable(adapter);
2499
2500         ixgbevf_napi_disable_all(adapter);
2501
2502         del_timer_sync(&adapter->service_timer);
2503
2504         /* disable transmits in the hardware now that interrupts are off */
2505         for (i = 0; i < adapter->num_tx_queues; i++) {
2506                 u8 reg_idx = adapter->tx_ring[i]->reg_idx;
2507
2508                 IXGBE_WRITE_REG(hw, IXGBE_VFTXDCTL(reg_idx),
2509                                 IXGBE_TXDCTL_SWFLSH);
2510         }
2511
2512         for (i = 0; i < adapter->num_xdp_queues; i++) {
2513                 u8 reg_idx = adapter->xdp_ring[i]->reg_idx;
2514
2515                 IXGBE_WRITE_REG(hw, IXGBE_VFTXDCTL(reg_idx),
2516                                 IXGBE_TXDCTL_SWFLSH);
2517         }
2518
2519         if (!pci_channel_offline(adapter->pdev))
2520                 ixgbevf_reset(adapter);
2521
2522         ixgbevf_clean_all_tx_rings(adapter);
2523         ixgbevf_clean_all_rx_rings(adapter);
2524 }
2525
2526 void ixgbevf_reinit_locked(struct ixgbevf_adapter *adapter)
2527 {
2528         while (test_and_set_bit(__IXGBEVF_RESETTING, &adapter->state))
2529                 msleep(1);
2530
2531         ixgbevf_down(adapter);
2532         pci_set_master(adapter->pdev);
2533         ixgbevf_up(adapter);
2534
2535         clear_bit(__IXGBEVF_RESETTING, &adapter->state);
2536 }
2537
2538 void ixgbevf_reset(struct ixgbevf_adapter *adapter)
2539 {
2540         struct ixgbe_hw *hw = &adapter->hw;
2541         struct net_device *netdev = adapter->netdev;
2542
2543         if (hw->mac.ops.reset_hw(hw)) {
2544                 hw_dbg(hw, "PF still resetting\n");
2545         } else {
2546                 hw->mac.ops.init_hw(hw);
2547                 ixgbevf_negotiate_api(adapter);
2548         }
2549
2550         if (is_valid_ether_addr(adapter->hw.mac.addr)) {
2551                 eth_hw_addr_set(netdev, adapter->hw.mac.addr);
2552                 ether_addr_copy(netdev->perm_addr, adapter->hw.mac.addr);
2553         }
2554
2555         adapter->last_reset = jiffies;
2556 }
2557
2558 static int ixgbevf_acquire_msix_vectors(struct ixgbevf_adapter *adapter,
2559                                         int vectors)
2560 {
2561         int vector_threshold;
2562
2563         /* We'll want at least 2 (vector_threshold):
2564          * 1) TxQ[0] + RxQ[0] handler
2565          * 2) Other (Link Status Change, etc.)
2566          */
2567         vector_threshold = MIN_MSIX_COUNT;
2568
2569         /* The more we get, the more we will assign to Tx/Rx Cleanup
2570          * for the separate queues...where Rx Cleanup >= Tx Cleanup.
2571          * Right now, we simply care about how many we'll get; we'll
2572          * set them up later while requesting irq's.
2573          */
2574         vectors = pci_enable_msix_range(adapter->pdev, adapter->msix_entries,
2575                                         vector_threshold, vectors);
2576
2577         if (vectors < 0) {
2578                 dev_err(&adapter->pdev->dev,
2579                         "Unable to allocate MSI-X interrupts\n");
2580                 kfree(adapter->msix_entries);
2581                 adapter->msix_entries = NULL;
2582                 return vectors;
2583         }
2584
2585         /* Adjust for only the vectors we'll use, which is minimum
2586          * of max_msix_q_vectors + NON_Q_VECTORS, or the number of
2587          * vectors we were allocated.
2588          */
2589         adapter->num_msix_vectors = vectors;
2590
2591         return 0;
2592 }
2593
2594 /**
2595  * ixgbevf_set_num_queues - Allocate queues for device, feature dependent
2596  * @adapter: board private structure to initialize
2597  *
2598  * This is the top level queue allocation routine.  The order here is very
2599  * important, starting with the "most" number of features turned on at once,
2600  * and ending with the smallest set of features.  This way large combinations
2601  * can be allocated if they're turned on, and smaller combinations are the
2602  * fall through conditions.
2603  *
2604  **/
2605 static void ixgbevf_set_num_queues(struct ixgbevf_adapter *adapter)
2606 {
2607         struct ixgbe_hw *hw = &adapter->hw;
2608         unsigned int def_q = 0;
2609         unsigned int num_tcs = 0;
2610         int err;
2611
2612         /* Start with base case */
2613         adapter->num_rx_queues = 1;
2614         adapter->num_tx_queues = 1;
2615         adapter->num_xdp_queues = 0;
2616
2617         spin_lock_bh(&adapter->mbx_lock);
2618
2619         /* fetch queue configuration from the PF */
2620         err = ixgbevf_get_queues(hw, &num_tcs, &def_q);
2621
2622         spin_unlock_bh(&adapter->mbx_lock);
2623
2624         if (err)
2625                 return;
2626
2627         /* we need as many queues as traffic classes */
2628         if (num_tcs > 1) {
2629                 adapter->num_rx_queues = num_tcs;
2630         } else {
2631                 u16 rss = min_t(u16, num_online_cpus(), IXGBEVF_MAX_RSS_QUEUES);
2632
2633                 switch (hw->api_version) {
2634                 case ixgbe_mbox_api_11:
2635                 case ixgbe_mbox_api_12:
2636                 case ixgbe_mbox_api_13:
2637                 case ixgbe_mbox_api_14:
2638                 case ixgbe_mbox_api_15:
2639                         if (adapter->xdp_prog &&
2640                             hw->mac.max_tx_queues == rss)
2641                                 rss = rss > 3 ? 2 : 1;
2642
2643                         adapter->num_rx_queues = rss;
2644                         adapter->num_tx_queues = rss;
2645                         adapter->num_xdp_queues = adapter->xdp_prog ? rss : 0;
2646                         break;
2647                 default:
2648                         break;
2649                 }
2650         }
2651 }
2652
2653 /**
2654  * ixgbevf_set_interrupt_capability - set MSI-X or FAIL if not supported
2655  * @adapter: board private structure to initialize
2656  *
2657  * Attempt to configure the interrupts using the best available
2658  * capabilities of the hardware and the kernel.
2659  **/
2660 static int ixgbevf_set_interrupt_capability(struct ixgbevf_adapter *adapter)
2661 {
2662         int vector, v_budget;
2663
2664         /* It's easy to be greedy for MSI-X vectors, but it really
2665          * doesn't do us much good if we have a lot more vectors
2666          * than CPU's.  So let's be conservative and only ask for
2667          * (roughly) the same number of vectors as there are CPU's.
2668          * The default is to use pairs of vectors.
2669          */
2670         v_budget = max(adapter->num_rx_queues, adapter->num_tx_queues);
2671         v_budget = min_t(int, v_budget, num_online_cpus());
2672         v_budget += NON_Q_VECTORS;
2673
2674         adapter->msix_entries = kcalloc(v_budget,
2675                                         sizeof(struct msix_entry), GFP_KERNEL);
2676         if (!adapter->msix_entries)
2677                 return -ENOMEM;
2678
2679         for (vector = 0; vector < v_budget; vector++)
2680                 adapter->msix_entries[vector].entry = vector;
2681
2682         /* A failure in MSI-X entry allocation isn't fatal, but the VF driver
2683          * does not support any other modes, so we will simply fail here. Note
2684          * that we clean up the msix_entries pointer else-where.
2685          */
2686         return ixgbevf_acquire_msix_vectors(adapter, v_budget);
2687 }
2688
2689 static void ixgbevf_add_ring(struct ixgbevf_ring *ring,
2690                              struct ixgbevf_ring_container *head)
2691 {
2692         ring->next = head->ring;
2693         head->ring = ring;
2694         head->count++;
2695 }
2696
2697 /**
2698  * ixgbevf_alloc_q_vector - Allocate memory for a single interrupt vector
2699  * @adapter: board private structure to initialize
2700  * @v_idx: index of vector in adapter struct
2701  * @txr_count: number of Tx rings for q vector
2702  * @txr_idx: index of first Tx ring to assign
2703  * @xdp_count: total number of XDP rings to allocate
2704  * @xdp_idx: index of first XDP ring to allocate
2705  * @rxr_count: number of Rx rings for q vector
2706  * @rxr_idx: index of first Rx ring to assign
2707  *
2708  * We allocate one q_vector.  If allocation fails we return -ENOMEM.
2709  **/
2710 static int ixgbevf_alloc_q_vector(struct ixgbevf_adapter *adapter, int v_idx,
2711                                   int txr_count, int txr_idx,
2712                                   int xdp_count, int xdp_idx,
2713                                   int rxr_count, int rxr_idx)
2714 {
2715         struct ixgbevf_q_vector *q_vector;
2716         int reg_idx = txr_idx + xdp_idx;
2717         struct ixgbevf_ring *ring;
2718         int ring_count, size;
2719
2720         ring_count = txr_count + xdp_count + rxr_count;
2721         size = sizeof(*q_vector) + (sizeof(*ring) * ring_count);
2722
2723         /* allocate q_vector and rings */
2724         q_vector = kzalloc(size, GFP_KERNEL);
2725         if (!q_vector)
2726                 return -ENOMEM;
2727
2728         /* initialize NAPI */
2729         netif_napi_add(adapter->netdev, &q_vector->napi, ixgbevf_poll, 64);
2730
2731         /* tie q_vector and adapter together */
2732         adapter->q_vector[v_idx] = q_vector;
2733         q_vector->adapter = adapter;
2734         q_vector->v_idx = v_idx;
2735
2736         /* initialize pointer to rings */
2737         ring = q_vector->ring;
2738
2739         while (txr_count) {
2740                 /* assign generic ring traits */
2741                 ring->dev = &adapter->pdev->dev;
2742                 ring->netdev = adapter->netdev;
2743
2744                 /* configure backlink on ring */
2745                 ring->q_vector = q_vector;
2746
2747                 /* update q_vector Tx values */
2748                 ixgbevf_add_ring(ring, &q_vector->tx);
2749
2750                 /* apply Tx specific ring traits */
2751                 ring->count = adapter->tx_ring_count;
2752                 ring->queue_index = txr_idx;
2753                 ring->reg_idx = reg_idx;
2754
2755                 /* assign ring to adapter */
2756                 adapter->tx_ring[txr_idx] = ring;
2757
2758                 /* update count and index */
2759                 txr_count--;
2760                 txr_idx++;
2761                 reg_idx++;
2762
2763                 /* push pointer to next ring */
2764                 ring++;
2765         }
2766
2767         while (xdp_count) {
2768                 /* assign generic ring traits */
2769                 ring->dev = &adapter->pdev->dev;
2770                 ring->netdev = adapter->netdev;
2771
2772                 /* configure backlink on ring */
2773                 ring->q_vector = q_vector;
2774
2775                 /* update q_vector Tx values */
2776                 ixgbevf_add_ring(ring, &q_vector->tx);
2777
2778                 /* apply Tx specific ring traits */
2779                 ring->count = adapter->tx_ring_count;
2780                 ring->queue_index = xdp_idx;
2781                 ring->reg_idx = reg_idx;
2782                 set_ring_xdp(ring);
2783
2784                 /* assign ring to adapter */
2785                 adapter->xdp_ring[xdp_idx] = ring;
2786
2787                 /* update count and index */
2788                 xdp_count--;
2789                 xdp_idx++;
2790                 reg_idx++;
2791
2792                 /* push pointer to next ring */
2793                 ring++;
2794         }
2795
2796         while (rxr_count) {
2797                 /* assign generic ring traits */
2798                 ring->dev = &adapter->pdev->dev;
2799                 ring->netdev = adapter->netdev;
2800
2801                 /* configure backlink on ring */
2802                 ring->q_vector = q_vector;
2803
2804                 /* update q_vector Rx values */
2805                 ixgbevf_add_ring(ring, &q_vector->rx);
2806
2807                 /* apply Rx specific ring traits */
2808                 ring->count = adapter->rx_ring_count;
2809                 ring->queue_index = rxr_idx;
2810                 ring->reg_idx = rxr_idx;
2811
2812                 /* assign ring to adapter */
2813                 adapter->rx_ring[rxr_idx] = ring;
2814
2815                 /* update count and index */
2816                 rxr_count--;
2817                 rxr_idx++;
2818
2819                 /* push pointer to next ring */
2820                 ring++;
2821         }
2822
2823         return 0;
2824 }
2825
2826 /**
2827  * ixgbevf_free_q_vector - Free memory allocated for specific interrupt vector
2828  * @adapter: board private structure to initialize
2829  * @v_idx: index of vector in adapter struct
2830  *
2831  * This function frees the memory allocated to the q_vector.  In addition if
2832  * NAPI is enabled it will delete any references to the NAPI struct prior
2833  * to freeing the q_vector.
2834  **/
2835 static void ixgbevf_free_q_vector(struct ixgbevf_adapter *adapter, int v_idx)
2836 {
2837         struct ixgbevf_q_vector *q_vector = adapter->q_vector[v_idx];
2838         struct ixgbevf_ring *ring;
2839
2840         ixgbevf_for_each_ring(ring, q_vector->tx) {
2841                 if (ring_is_xdp(ring))
2842                         adapter->xdp_ring[ring->queue_index] = NULL;
2843                 else
2844                         adapter->tx_ring[ring->queue_index] = NULL;
2845         }
2846
2847         ixgbevf_for_each_ring(ring, q_vector->rx)
2848                 adapter->rx_ring[ring->queue_index] = NULL;
2849
2850         adapter->q_vector[v_idx] = NULL;
2851         netif_napi_del(&q_vector->napi);
2852
2853         /* ixgbevf_get_stats() might access the rings on this vector,
2854          * we must wait a grace period before freeing it.
2855          */
2856         kfree_rcu(q_vector, rcu);
2857 }
2858
2859 /**
2860  * ixgbevf_alloc_q_vectors - Allocate memory for interrupt vectors
2861  * @adapter: board private structure to initialize
2862  *
2863  * We allocate one q_vector per queue interrupt.  If allocation fails we
2864  * return -ENOMEM.
2865  **/
2866 static int ixgbevf_alloc_q_vectors(struct ixgbevf_adapter *adapter)
2867 {
2868         int q_vectors = adapter->num_msix_vectors - NON_Q_VECTORS;
2869         int rxr_remaining = adapter->num_rx_queues;
2870         int txr_remaining = adapter->num_tx_queues;
2871         int xdp_remaining = adapter->num_xdp_queues;
2872         int rxr_idx = 0, txr_idx = 0, xdp_idx = 0, v_idx = 0;
2873         int err;
2874
2875         if (q_vectors >= (rxr_remaining + txr_remaining + xdp_remaining)) {
2876                 for (; rxr_remaining; v_idx++, q_vectors--) {
2877                         int rqpv = DIV_ROUND_UP(rxr_remaining, q_vectors);
2878
2879                         err = ixgbevf_alloc_q_vector(adapter, v_idx,
2880                                                      0, 0, 0, 0, rqpv, rxr_idx);
2881                         if (err)
2882                                 goto err_out;
2883
2884                         /* update counts and index */
2885                         rxr_remaining -= rqpv;
2886                         rxr_idx += rqpv;
2887                 }
2888         }
2889
2890         for (; q_vectors; v_idx++, q_vectors--) {
2891                 int rqpv = DIV_ROUND_UP(rxr_remaining, q_vectors);
2892                 int tqpv = DIV_ROUND_UP(txr_remaining, q_vectors);
2893                 int xqpv = DIV_ROUND_UP(xdp_remaining, q_vectors);
2894
2895                 err = ixgbevf_alloc_q_vector(adapter, v_idx,
2896                                              tqpv, txr_idx,
2897                                              xqpv, xdp_idx,
2898                                              rqpv, rxr_idx);
2899
2900                 if (err)
2901                         goto err_out;
2902
2903                 /* update counts and index */
2904                 rxr_remaining -= rqpv;
2905                 rxr_idx += rqpv;
2906                 txr_remaining -= tqpv;
2907                 txr_idx += tqpv;
2908                 xdp_remaining -= xqpv;
2909                 xdp_idx += xqpv;
2910         }
2911
2912         return 0;
2913
2914 err_out:
2915         while (v_idx) {
2916                 v_idx--;
2917                 ixgbevf_free_q_vector(adapter, v_idx);
2918         }
2919
2920         return -ENOMEM;
2921 }
2922
2923 /**
2924  * ixgbevf_free_q_vectors - Free memory allocated for interrupt vectors
2925  * @adapter: board private structure to initialize
2926  *
2927  * This function frees the memory allocated to the q_vectors.  In addition if
2928  * NAPI is enabled it will delete any references to the NAPI struct prior
2929  * to freeing the q_vector.
2930  **/
2931 static void ixgbevf_free_q_vectors(struct ixgbevf_adapter *adapter)
2932 {
2933         int q_vectors = adapter->num_msix_vectors - NON_Q_VECTORS;
2934
2935         while (q_vectors) {
2936                 q_vectors--;
2937                 ixgbevf_free_q_vector(adapter, q_vectors);
2938         }
2939 }
2940
2941 /**
2942  * ixgbevf_reset_interrupt_capability - Reset MSIX setup
2943  * @adapter: board private structure
2944  *
2945  **/
2946 static void ixgbevf_reset_interrupt_capability(struct ixgbevf_adapter *adapter)
2947 {
2948         if (!adapter->msix_entries)
2949                 return;
2950
2951         pci_disable_msix(adapter->pdev);
2952         kfree(adapter->msix_entries);
2953         adapter->msix_entries = NULL;
2954 }
2955
2956 /**
2957  * ixgbevf_init_interrupt_scheme - Determine if MSIX is supported and init
2958  * @adapter: board private structure to initialize
2959  *
2960  **/
2961 static int ixgbevf_init_interrupt_scheme(struct ixgbevf_adapter *adapter)
2962 {
2963         int err;
2964
2965         /* Number of supported queues */
2966         ixgbevf_set_num_queues(adapter);
2967
2968         err = ixgbevf_set_interrupt_capability(adapter);
2969         if (err) {
2970                 hw_dbg(&adapter->hw,
2971                        "Unable to setup interrupt capabilities\n");
2972                 goto err_set_interrupt;
2973         }
2974
2975         err = ixgbevf_alloc_q_vectors(adapter);
2976         if (err) {
2977                 hw_dbg(&adapter->hw, "Unable to allocate memory for queue vectors\n");
2978                 goto err_alloc_q_vectors;
2979         }
2980
2981         hw_dbg(&adapter->hw, "Multiqueue %s: Rx Queue count = %u, Tx Queue count = %u XDP Queue count %u\n",
2982                (adapter->num_rx_queues > 1) ? "Enabled" : "Disabled",
2983                adapter->num_rx_queues, adapter->num_tx_queues,
2984                adapter->num_xdp_queues);
2985
2986         set_bit(__IXGBEVF_DOWN, &adapter->state);
2987
2988         return 0;
2989 err_alloc_q_vectors:
2990         ixgbevf_reset_interrupt_capability(adapter);
2991 err_set_interrupt:
2992         return err;
2993 }
2994
2995 /**
2996  * ixgbevf_clear_interrupt_scheme - Clear the current interrupt scheme settings
2997  * @adapter: board private structure to clear interrupt scheme on
2998  *
2999  * We go through and clear interrupt specific resources and reset the structure
3000  * to pre-load conditions
3001  **/
3002 static void ixgbevf_clear_interrupt_scheme(struct ixgbevf_adapter *adapter)
3003 {
3004         adapter->num_tx_queues = 0;
3005         adapter->num_xdp_queues = 0;
3006         adapter->num_rx_queues = 0;
3007
3008         ixgbevf_free_q_vectors(adapter);
3009         ixgbevf_reset_interrupt_capability(adapter);
3010 }
3011
3012 /**
3013  * ixgbevf_sw_init - Initialize general software structures
3014  * @adapter: board private structure to initialize
3015  *
3016  * ixgbevf_sw_init initializes the Adapter private data structure.
3017  * Fields are initialized based on PCI device information and
3018  * OS network device settings (MTU size).
3019  **/
3020 static int ixgbevf_sw_init(struct ixgbevf_adapter *adapter)
3021 {
3022         struct ixgbe_hw *hw = &adapter->hw;
3023         struct pci_dev *pdev = adapter->pdev;
3024         struct net_device *netdev = adapter->netdev;
3025         int err;
3026
3027         /* PCI config space info */
3028         hw->vendor_id = pdev->vendor;
3029         hw->device_id = pdev->device;
3030         hw->revision_id = pdev->revision;
3031         hw->subsystem_vendor_id = pdev->subsystem_vendor;
3032         hw->subsystem_device_id = pdev->subsystem_device;
3033
3034         hw->mbx.ops.init_params(hw);
3035
3036         if (hw->mac.type >= ixgbe_mac_X550_vf) {
3037                 err = ixgbevf_init_rss_key(adapter);
3038                 if (err)
3039                         goto out;
3040         }
3041
3042         /* assume legacy case in which PF would only give VF 2 queues */
3043         hw->mac.max_tx_queues = 2;
3044         hw->mac.max_rx_queues = 2;
3045
3046         /* lock to protect mailbox accesses */
3047         spin_lock_init(&adapter->mbx_lock);
3048
3049         err = hw->mac.ops.reset_hw(hw);
3050         if (err) {
3051                 dev_info(&pdev->dev,
3052                          "PF still in reset state.  Is the PF interface up?\n");
3053         } else {
3054                 err = hw->mac.ops.init_hw(hw);
3055                 if (err) {
3056                         pr_err("init_shared_code failed: %d\n", err);
3057                         goto out;
3058                 }
3059                 ixgbevf_negotiate_api(adapter);
3060                 err = hw->mac.ops.get_mac_addr(hw, hw->mac.addr);
3061                 if (err)
3062                         dev_info(&pdev->dev, "Error reading MAC address\n");
3063                 else if (is_zero_ether_addr(adapter->hw.mac.addr))
3064                         dev_info(&pdev->dev,
3065                                  "MAC address not assigned by administrator.\n");
3066                 eth_hw_addr_set(netdev, hw->mac.addr);
3067         }
3068
3069         if (!is_valid_ether_addr(netdev->dev_addr)) {
3070                 dev_info(&pdev->dev, "Assigning random MAC address\n");
3071                 eth_hw_addr_random(netdev);
3072                 ether_addr_copy(hw->mac.addr, netdev->dev_addr);
3073                 ether_addr_copy(hw->mac.perm_addr, netdev->dev_addr);
3074         }
3075
3076         /* Enable dynamic interrupt throttling rates */
3077         adapter->rx_itr_setting = 1;
3078         adapter->tx_itr_setting = 1;
3079
3080         /* set default ring sizes */
3081         adapter->tx_ring_count = IXGBEVF_DEFAULT_TXD;
3082         adapter->rx_ring_count = IXGBEVF_DEFAULT_RXD;
3083
3084         set_bit(__IXGBEVF_DOWN, &adapter->state);
3085         return 0;
3086
3087 out:
3088         return err;
3089 }
3090
3091 #define UPDATE_VF_COUNTER_32bit(reg, last_counter, counter)     \
3092         {                                                       \
3093                 u32 current_counter = IXGBE_READ_REG(hw, reg);  \
3094                 if (current_counter < last_counter)             \
3095                         counter += 0x100000000LL;               \
3096                 last_counter = current_counter;                 \
3097                 counter &= 0xFFFFFFFF00000000LL;                \
3098                 counter |= current_counter;                     \
3099         }
3100
3101 #define UPDATE_VF_COUNTER_36bit(reg_lsb, reg_msb, last_counter, counter) \
3102         {                                                                \
3103                 u64 current_counter_lsb = IXGBE_READ_REG(hw, reg_lsb);   \
3104                 u64 current_counter_msb = IXGBE_READ_REG(hw, reg_msb);   \
3105                 u64 current_counter = (current_counter_msb << 32) |      \
3106                         current_counter_lsb;                             \
3107                 if (current_counter < last_counter)                      \
3108                         counter += 0x1000000000LL;                       \
3109                 last_counter = current_counter;                          \
3110                 counter &= 0xFFFFFFF000000000LL;                         \
3111                 counter |= current_counter;                              \
3112         }
3113 /**
3114  * ixgbevf_update_stats - Update the board statistics counters.
3115  * @adapter: board private structure
3116  **/
3117 void ixgbevf_update_stats(struct ixgbevf_adapter *adapter)
3118 {
3119         struct ixgbe_hw *hw = &adapter->hw;
3120         u64 alloc_rx_page_failed = 0, alloc_rx_buff_failed = 0;
3121         u64 alloc_rx_page = 0, hw_csum_rx_error = 0;
3122         int i;
3123
3124         if (test_bit(__IXGBEVF_DOWN, &adapter->state) ||
3125             test_bit(__IXGBEVF_RESETTING, &adapter->state))
3126                 return;
3127
3128         UPDATE_VF_COUNTER_32bit(IXGBE_VFGPRC, adapter->stats.last_vfgprc,
3129                                 adapter->stats.vfgprc);
3130         UPDATE_VF_COUNTER_32bit(IXGBE_VFGPTC, adapter->stats.last_vfgptc,
3131                                 adapter->stats.vfgptc);
3132         UPDATE_VF_COUNTER_36bit(IXGBE_VFGORC_LSB, IXGBE_VFGORC_MSB,
3133                                 adapter->stats.last_vfgorc,
3134                                 adapter->stats.vfgorc);
3135         UPDATE_VF_COUNTER_36bit(IXGBE_VFGOTC_LSB, IXGBE_VFGOTC_MSB,
3136                                 adapter->stats.last_vfgotc,
3137                                 adapter->stats.vfgotc);
3138         UPDATE_VF_COUNTER_32bit(IXGBE_VFMPRC, adapter->stats.last_vfmprc,
3139                                 adapter->stats.vfmprc);
3140
3141         for (i = 0;  i  < adapter->num_rx_queues;  i++) {
3142                 struct ixgbevf_ring *rx_ring = adapter->rx_ring[i];
3143
3144                 hw_csum_rx_error += rx_ring->rx_stats.csum_err;
3145                 alloc_rx_page_failed += rx_ring->rx_stats.alloc_rx_page_failed;
3146                 alloc_rx_buff_failed += rx_ring->rx_stats.alloc_rx_buff_failed;
3147                 alloc_rx_page += rx_ring->rx_stats.alloc_rx_page;
3148         }
3149
3150         adapter->hw_csum_rx_error = hw_csum_rx_error;
3151         adapter->alloc_rx_page_failed = alloc_rx_page_failed;
3152         adapter->alloc_rx_buff_failed = alloc_rx_buff_failed;
3153         adapter->alloc_rx_page = alloc_rx_page;
3154 }
3155
3156 /**
3157  * ixgbevf_service_timer - Timer Call-back
3158  * @t: pointer to timer_list struct
3159  **/
3160 static void ixgbevf_service_timer(struct timer_list *t)
3161 {
3162         struct ixgbevf_adapter *adapter = from_timer(adapter, t,
3163                                                      service_timer);
3164
3165         /* Reset the timer */
3166         mod_timer(&adapter->service_timer, (HZ * 2) + jiffies);
3167
3168         ixgbevf_service_event_schedule(adapter);
3169 }
3170
3171 static void ixgbevf_reset_subtask(struct ixgbevf_adapter *adapter)
3172 {
3173         if (!test_and_clear_bit(__IXGBEVF_RESET_REQUESTED, &adapter->state))
3174                 return;
3175
3176         rtnl_lock();
3177         /* If we're already down or resetting, just bail */
3178         if (test_bit(__IXGBEVF_DOWN, &adapter->state) ||
3179             test_bit(__IXGBEVF_REMOVING, &adapter->state) ||
3180             test_bit(__IXGBEVF_RESETTING, &adapter->state)) {
3181                 rtnl_unlock();
3182                 return;
3183         }
3184
3185         adapter->tx_timeout_count++;
3186
3187         ixgbevf_reinit_locked(adapter);
3188         rtnl_unlock();
3189 }
3190
3191 /**
3192  * ixgbevf_check_hang_subtask - check for hung queues and dropped interrupts
3193  * @adapter: pointer to the device adapter structure
3194  *
3195  * This function serves two purposes.  First it strobes the interrupt lines
3196  * in order to make certain interrupts are occurring.  Secondly it sets the
3197  * bits needed to check for TX hangs.  As a result we should immediately
3198  * determine if a hang has occurred.
3199  **/
3200 static void ixgbevf_check_hang_subtask(struct ixgbevf_adapter *adapter)
3201 {
3202         struct ixgbe_hw *hw = &adapter->hw;
3203         u32 eics = 0;
3204         int i;
3205
3206         /* If we're down or resetting, just bail */
3207         if (test_bit(__IXGBEVF_DOWN, &adapter->state) ||
3208             test_bit(__IXGBEVF_RESETTING, &adapter->state))
3209                 return;
3210
3211         /* Force detection of hung controller */
3212         if (netif_carrier_ok(adapter->netdev)) {
3213                 for (i = 0; i < adapter->num_tx_queues; i++)
3214                         set_check_for_tx_hang(adapter->tx_ring[i]);
3215                 for (i = 0; i < adapter->num_xdp_queues; i++)
3216                         set_check_for_tx_hang(adapter->xdp_ring[i]);
3217         }
3218
3219         /* get one bit for every active Tx/Rx interrupt vector */
3220         for (i = 0; i < adapter->num_msix_vectors - NON_Q_VECTORS; i++) {
3221                 struct ixgbevf_q_vector *qv = adapter->q_vector[i];
3222
3223                 if (qv->rx.ring || qv->tx.ring)
3224                         eics |= BIT(i);
3225         }
3226
3227         /* Cause software interrupt to ensure rings are cleaned */
3228         IXGBE_WRITE_REG(hw, IXGBE_VTEICS, eics);
3229 }
3230
3231 /**
3232  * ixgbevf_watchdog_update_link - update the link status
3233  * @adapter: pointer to the device adapter structure
3234  **/
3235 static void ixgbevf_watchdog_update_link(struct ixgbevf_adapter *adapter)
3236 {
3237         struct ixgbe_hw *hw = &adapter->hw;
3238         u32 link_speed = adapter->link_speed;
3239         bool link_up = adapter->link_up;
3240         s32 err;
3241
3242         spin_lock_bh(&adapter->mbx_lock);
3243
3244         err = hw->mac.ops.check_link(hw, &link_speed, &link_up, false);
3245
3246         spin_unlock_bh(&adapter->mbx_lock);
3247
3248         /* if check for link returns error we will need to reset */
3249         if (err && time_after(jiffies, adapter->last_reset + (10 * HZ))) {
3250                 set_bit(__IXGBEVF_RESET_REQUESTED, &adapter->state);
3251                 link_up = false;
3252         }
3253
3254         adapter->link_up = link_up;
3255         adapter->link_speed = link_speed;
3256 }
3257
3258 /**
3259  * ixgbevf_watchdog_link_is_up - update netif_carrier status and
3260  *                               print link up message
3261  * @adapter: pointer to the device adapter structure
3262  **/
3263 static void ixgbevf_watchdog_link_is_up(struct ixgbevf_adapter *adapter)
3264 {
3265         struct net_device *netdev = adapter->netdev;
3266
3267         /* only continue if link was previously down */
3268         if (netif_carrier_ok(netdev))
3269                 return;
3270
3271         dev_info(&adapter->pdev->dev, "NIC Link is Up %s\n",
3272                  (adapter->link_speed == IXGBE_LINK_SPEED_10GB_FULL) ?
3273                  "10 Gbps" :
3274                  (adapter->link_speed == IXGBE_LINK_SPEED_1GB_FULL) ?
3275                  "1 Gbps" :
3276                  (adapter->link_speed == IXGBE_LINK_SPEED_100_FULL) ?
3277                  "100 Mbps" :
3278                  "unknown speed");
3279
3280         netif_carrier_on(netdev);
3281 }
3282
3283 /**
3284  * ixgbevf_watchdog_link_is_down - update netif_carrier status and
3285  *                                 print link down message
3286  * @adapter: pointer to the adapter structure
3287  **/
3288 static void ixgbevf_watchdog_link_is_down(struct ixgbevf_adapter *adapter)
3289 {
3290         struct net_device *netdev = adapter->netdev;
3291
3292         adapter->link_speed = 0;
3293
3294         /* only continue if link was up previously */
3295         if (!netif_carrier_ok(netdev))
3296                 return;
3297
3298         dev_info(&adapter->pdev->dev, "NIC Link is Down\n");
3299
3300         netif_carrier_off(netdev);
3301 }
3302
3303 /**
3304  * ixgbevf_watchdog_subtask - worker thread to bring link up
3305  * @adapter: board private structure
3306  **/
3307 static void ixgbevf_watchdog_subtask(struct ixgbevf_adapter *adapter)
3308 {
3309         /* if interface is down do nothing */
3310         if (test_bit(__IXGBEVF_DOWN, &adapter->state) ||
3311             test_bit(__IXGBEVF_RESETTING, &adapter->state))
3312                 return;
3313
3314         ixgbevf_watchdog_update_link(adapter);
3315
3316         if (adapter->link_up)
3317                 ixgbevf_watchdog_link_is_up(adapter);
3318         else
3319                 ixgbevf_watchdog_link_is_down(adapter);
3320
3321         ixgbevf_update_stats(adapter);
3322 }
3323
3324 /**
3325  * ixgbevf_service_task - manages and runs subtasks
3326  * @work: pointer to work_struct containing our data
3327  **/
3328 static void ixgbevf_service_task(struct work_struct *work)
3329 {
3330         struct ixgbevf_adapter *adapter = container_of(work,
3331                                                        struct ixgbevf_adapter,
3332                                                        service_task);
3333         struct ixgbe_hw *hw = &adapter->hw;
3334
3335         if (IXGBE_REMOVED(hw->hw_addr)) {
3336                 if (!test_bit(__IXGBEVF_DOWN, &adapter->state)) {
3337                         rtnl_lock();
3338                         ixgbevf_down(adapter);
3339                         rtnl_unlock();
3340                 }
3341                 return;
3342         }
3343
3344         ixgbevf_queue_reset_subtask(adapter);
3345         ixgbevf_reset_subtask(adapter);
3346         ixgbevf_watchdog_subtask(adapter);
3347         ixgbevf_check_hang_subtask(adapter);
3348
3349         ixgbevf_service_event_complete(adapter);
3350 }
3351
3352 /**
3353  * ixgbevf_free_tx_resources - Free Tx Resources per Queue
3354  * @tx_ring: Tx descriptor ring for a specific queue
3355  *
3356  * Free all transmit software resources
3357  **/
3358 void ixgbevf_free_tx_resources(struct ixgbevf_ring *tx_ring)
3359 {
3360         ixgbevf_clean_tx_ring(tx_ring);
3361
3362         vfree(tx_ring->tx_buffer_info);
3363         tx_ring->tx_buffer_info = NULL;
3364
3365         /* if not set, then don't free */
3366         if (!tx_ring->desc)
3367                 return;
3368
3369         dma_free_coherent(tx_ring->dev, tx_ring->size, tx_ring->desc,
3370                           tx_ring->dma);
3371
3372         tx_ring->desc = NULL;
3373 }
3374
3375 /**
3376  * ixgbevf_free_all_tx_resources - Free Tx Resources for All Queues
3377  * @adapter: board private structure
3378  *
3379  * Free all transmit software resources
3380  **/
3381 static void ixgbevf_free_all_tx_resources(struct ixgbevf_adapter *adapter)
3382 {
3383         int i;
3384
3385         for (i = 0; i < adapter->num_tx_queues; i++)
3386                 if (adapter->tx_ring[i]->desc)
3387                         ixgbevf_free_tx_resources(adapter->tx_ring[i]);
3388         for (i = 0; i < adapter->num_xdp_queues; i++)
3389                 if (adapter->xdp_ring[i]->desc)
3390                         ixgbevf_free_tx_resources(adapter->xdp_ring[i]);
3391 }
3392
3393 /**
3394  * ixgbevf_setup_tx_resources - allocate Tx resources (Descriptors)
3395  * @tx_ring: Tx descriptor ring (for a specific queue) to setup
3396  *
3397  * Return 0 on success, negative on failure
3398  **/
3399 int ixgbevf_setup_tx_resources(struct ixgbevf_ring *tx_ring)
3400 {
3401         struct ixgbevf_adapter *adapter = netdev_priv(tx_ring->netdev);
3402         int size;
3403
3404         size = sizeof(struct ixgbevf_tx_buffer) * tx_ring->count;
3405         tx_ring->tx_buffer_info = vmalloc(size);
3406         if (!tx_ring->tx_buffer_info)
3407                 goto err;
3408
3409         u64_stats_init(&tx_ring->syncp);
3410
3411         /* round up to nearest 4K */
3412         tx_ring->size = tx_ring->count * sizeof(union ixgbe_adv_tx_desc);
3413         tx_ring->size = ALIGN(tx_ring->size, 4096);
3414
3415         tx_ring->desc = dma_alloc_coherent(tx_ring->dev, tx_ring->size,
3416                                            &tx_ring->dma, GFP_KERNEL);
3417         if (!tx_ring->desc)
3418                 goto err;
3419
3420         return 0;
3421
3422 err:
3423         vfree(tx_ring->tx_buffer_info);
3424         tx_ring->tx_buffer_info = NULL;
3425         hw_dbg(&adapter->hw, "Unable to allocate memory for the transmit descriptor ring\n");
3426         return -ENOMEM;
3427 }
3428
3429 /**
3430  * ixgbevf_setup_all_tx_resources - allocate all queues Tx resources
3431  * @adapter: board private structure
3432  *
3433  * If this function returns with an error, then it's possible one or
3434  * more of the rings is populated (while the rest are not).  It is the
3435  * callers duty to clean those orphaned rings.
3436  *
3437  * Return 0 on success, negative on failure
3438  **/
3439 static int ixgbevf_setup_all_tx_resources(struct ixgbevf_adapter *adapter)
3440 {
3441         int i, j = 0, err = 0;
3442
3443         for (i = 0; i < adapter->num_tx_queues; i++) {
3444                 err = ixgbevf_setup_tx_resources(adapter->tx_ring[i]);
3445                 if (!err)
3446                         continue;
3447                 hw_dbg(&adapter->hw, "Allocation for Tx Queue %u failed\n", i);
3448                 goto err_setup_tx;
3449         }
3450
3451         for (j = 0; j < adapter->num_xdp_queues; j++) {
3452                 err = ixgbevf_setup_tx_resources(adapter->xdp_ring[j]);
3453                 if (!err)
3454                         continue;
3455                 hw_dbg(&adapter->hw, "Allocation for XDP Queue %u failed\n", j);
3456                 goto err_setup_tx;
3457         }
3458
3459         return 0;
3460 err_setup_tx:
3461         /* rewind the index freeing the rings as we go */
3462         while (j--)
3463                 ixgbevf_free_tx_resources(adapter->xdp_ring[j]);
3464         while (i--)
3465                 ixgbevf_free_tx_resources(adapter->tx_ring[i]);
3466
3467         return err;
3468 }
3469
3470 /**
3471  * ixgbevf_setup_rx_resources - allocate Rx resources (Descriptors)
3472  * @adapter: board private structure
3473  * @rx_ring: Rx descriptor ring (for a specific queue) to setup
3474  *
3475  * Returns 0 on success, negative on failure
3476  **/
3477 int ixgbevf_setup_rx_resources(struct ixgbevf_adapter *adapter,
3478                                struct ixgbevf_ring *rx_ring)
3479 {
3480         int size;
3481
3482         size = sizeof(struct ixgbevf_rx_buffer) * rx_ring->count;
3483         rx_ring->rx_buffer_info = vmalloc(size);
3484         if (!rx_ring->rx_buffer_info)
3485                 goto err;
3486
3487         u64_stats_init(&rx_ring->syncp);
3488
3489         /* Round up to nearest 4K */
3490         rx_ring->size = rx_ring->count * sizeof(union ixgbe_adv_rx_desc);
3491         rx_ring->size = ALIGN(rx_ring->size, 4096);
3492
3493         rx_ring->desc = dma_alloc_coherent(rx_ring->dev, rx_ring->size,
3494                                            &rx_ring->dma, GFP_KERNEL);
3495
3496         if (!rx_ring->desc)
3497                 goto err;
3498
3499         /* XDP RX-queue info */
3500         if (xdp_rxq_info_reg(&rx_ring->xdp_rxq, adapter->netdev,
3501                              rx_ring->queue_index, 0) < 0)
3502                 goto err;
3503
3504         rx_ring->xdp_prog = adapter->xdp_prog;
3505
3506         return 0;
3507 err:
3508         vfree(rx_ring->rx_buffer_info);
3509         rx_ring->rx_buffer_info = NULL;
3510         dev_err(rx_ring->dev, "Unable to allocate memory for the Rx descriptor ring\n");
3511         return -ENOMEM;
3512 }
3513
3514 /**
3515  * ixgbevf_setup_all_rx_resources - allocate all queues Rx resources
3516  * @adapter: board private structure
3517  *
3518  * If this function returns with an error, then it's possible one or
3519  * more of the rings is populated (while the rest are not).  It is the
3520  * callers duty to clean those orphaned rings.
3521  *
3522  * Return 0 on success, negative on failure
3523  **/
3524 static int ixgbevf_setup_all_rx_resources(struct ixgbevf_adapter *adapter)
3525 {
3526         int i, err = 0;
3527
3528         for (i = 0; i < adapter->num_rx_queues; i++) {
3529                 err = ixgbevf_setup_rx_resources(adapter, adapter->rx_ring[i]);
3530                 if (!err)
3531                         continue;
3532                 hw_dbg(&adapter->hw, "Allocation for Rx Queue %u failed\n", i);
3533                 goto err_setup_rx;
3534         }
3535
3536         return 0;
3537 err_setup_rx:
3538         /* rewind the index freeing the rings as we go */
3539         while (i--)
3540                 ixgbevf_free_rx_resources(adapter->rx_ring[i]);
3541         return err;
3542 }
3543
3544 /**
3545  * ixgbevf_free_rx_resources - Free Rx Resources
3546  * @rx_ring: ring to clean the resources from
3547  *
3548  * Free all receive software resources
3549  **/
3550 void ixgbevf_free_rx_resources(struct ixgbevf_ring *rx_ring)
3551 {
3552         ixgbevf_clean_rx_ring(rx_ring);
3553
3554         rx_ring->xdp_prog = NULL;
3555         xdp_rxq_info_unreg(&rx_ring->xdp_rxq);
3556         vfree(rx_ring->rx_buffer_info);
3557         rx_ring->rx_buffer_info = NULL;
3558
3559         dma_free_coherent(rx_ring->dev, rx_ring->size, rx_ring->desc,
3560                           rx_ring->dma);
3561
3562         rx_ring->desc = NULL;
3563 }
3564
3565 /**
3566  * ixgbevf_free_all_rx_resources - Free Rx Resources for All Queues
3567  * @adapter: board private structure
3568  *
3569  * Free all receive software resources
3570  **/
3571 static void ixgbevf_free_all_rx_resources(struct ixgbevf_adapter *adapter)
3572 {
3573         int i;
3574
3575         for (i = 0; i < adapter->num_rx_queues; i++)
3576                 if (adapter->rx_ring[i]->desc)
3577                         ixgbevf_free_rx_resources(adapter->rx_ring[i]);
3578 }
3579
3580 /**
3581  * ixgbevf_open - Called when a network interface is made active
3582  * @netdev: network interface device structure
3583  *
3584  * Returns 0 on success, negative value on failure
3585  *
3586  * The open entry point is called when a network interface is made
3587  * active by the system (IFF_UP).  At this point all resources needed
3588  * for transmit and receive operations are allocated, the interrupt
3589  * handler is registered with the OS, the watchdog timer is started,
3590  * and the stack is notified that the interface is ready.
3591  **/
3592 int ixgbevf_open(struct net_device *netdev)
3593 {
3594         struct ixgbevf_adapter *adapter = netdev_priv(netdev);
3595         struct ixgbe_hw *hw = &adapter->hw;
3596         int err;
3597
3598         /* A previous failure to open the device because of a lack of
3599          * available MSIX vector resources may have reset the number
3600          * of msix vectors variable to zero.  The only way to recover
3601          * is to unload/reload the driver and hope that the system has
3602          * been able to recover some MSIX vector resources.
3603          */
3604         if (!adapter->num_msix_vectors)
3605                 return -ENOMEM;
3606
3607         if (hw->adapter_stopped) {
3608                 ixgbevf_reset(adapter);
3609                 /* if adapter is still stopped then PF isn't up and
3610                  * the VF can't start.
3611                  */
3612                 if (hw->adapter_stopped) {
3613                         err = IXGBE_ERR_MBX;
3614                         pr_err("Unable to start - perhaps the PF Driver isn't up yet\n");
3615                         goto err_setup_reset;
3616                 }
3617         }
3618
3619         /* disallow open during test */
3620         if (test_bit(__IXGBEVF_TESTING, &adapter->state))
3621                 return -EBUSY;
3622
3623         netif_carrier_off(netdev);
3624
3625         /* allocate transmit descriptors */
3626         err = ixgbevf_setup_all_tx_resources(adapter);
3627         if (err)
3628                 goto err_setup_tx;
3629
3630         /* allocate receive descriptors */
3631         err = ixgbevf_setup_all_rx_resources(adapter);
3632         if (err)
3633                 goto err_setup_rx;
3634
3635         ixgbevf_configure(adapter);
3636
3637         err = ixgbevf_request_irq(adapter);
3638         if (err)
3639                 goto err_req_irq;
3640
3641         /* Notify the stack of the actual queue counts. */
3642         err = netif_set_real_num_tx_queues(netdev, adapter->num_tx_queues);
3643         if (err)
3644                 goto err_set_queues;
3645
3646         err = netif_set_real_num_rx_queues(netdev, adapter->num_rx_queues);
3647         if (err)
3648                 goto err_set_queues;
3649
3650         ixgbevf_up_complete(adapter);
3651
3652         return 0;
3653
3654 err_set_queues:
3655         ixgbevf_free_irq(adapter);
3656 err_req_irq:
3657         ixgbevf_free_all_rx_resources(adapter);
3658 err_setup_rx:
3659         ixgbevf_free_all_tx_resources(adapter);
3660 err_setup_tx:
3661         ixgbevf_reset(adapter);
3662 err_setup_reset:
3663
3664         return err;
3665 }
3666
3667 /**
3668  * ixgbevf_close_suspend - actions necessary to both suspend and close flows
3669  * @adapter: the private adapter struct
3670  *
3671  * This function should contain the necessary work common to both suspending
3672  * and closing of the device.
3673  */
3674 static void ixgbevf_close_suspend(struct ixgbevf_adapter *adapter)
3675 {
3676         ixgbevf_down(adapter);
3677         ixgbevf_free_irq(adapter);
3678         ixgbevf_free_all_tx_resources(adapter);
3679         ixgbevf_free_all_rx_resources(adapter);
3680 }
3681
3682 /**
3683  * ixgbevf_close - Disables a network interface
3684  * @netdev: network interface device structure
3685  *
3686  * Returns 0, this is not allowed to fail
3687  *
3688  * The close entry point is called when an interface is de-activated
3689  * by the OS.  The hardware is still under the drivers control, but
3690  * needs to be disabled.  A global MAC reset is issued to stop the
3691  * hardware, and all transmit and receive resources are freed.
3692  **/
3693 int ixgbevf_close(struct net_device *netdev)
3694 {
3695         struct ixgbevf_adapter *adapter = netdev_priv(netdev);
3696
3697         if (netif_device_present(netdev))
3698                 ixgbevf_close_suspend(adapter);
3699
3700         return 0;
3701 }
3702
3703 static void ixgbevf_queue_reset_subtask(struct ixgbevf_adapter *adapter)
3704 {
3705         struct net_device *dev = adapter->netdev;
3706
3707         if (!test_and_clear_bit(__IXGBEVF_QUEUE_RESET_REQUESTED,
3708                                 &adapter->state))
3709                 return;
3710
3711         /* if interface is down do nothing */
3712         if (test_bit(__IXGBEVF_DOWN, &adapter->state) ||
3713             test_bit(__IXGBEVF_RESETTING, &adapter->state))
3714                 return;
3715
3716         /* Hardware has to reinitialize queues and interrupts to
3717          * match packet buffer alignment. Unfortunately, the
3718          * hardware is not flexible enough to do this dynamically.
3719          */
3720         rtnl_lock();
3721
3722         if (netif_running(dev))
3723                 ixgbevf_close(dev);
3724
3725         ixgbevf_clear_interrupt_scheme(adapter);
3726         ixgbevf_init_interrupt_scheme(adapter);
3727
3728         if (netif_running(dev))
3729                 ixgbevf_open(dev);
3730
3731         rtnl_unlock();
3732 }
3733
3734 static void ixgbevf_tx_ctxtdesc(struct ixgbevf_ring *tx_ring,
3735                                 u32 vlan_macip_lens, u32 fceof_saidx,
3736                                 u32 type_tucmd, u32 mss_l4len_idx)
3737 {
3738         struct ixgbe_adv_tx_context_desc *context_desc;
3739         u16 i = tx_ring->next_to_use;
3740
3741         context_desc = IXGBEVF_TX_CTXTDESC(tx_ring, i);
3742
3743         i++;
3744         tx_ring->next_to_use = (i < tx_ring->count) ? i : 0;
3745
3746         /* set bits to identify this as an advanced context descriptor */
3747         type_tucmd |= IXGBE_TXD_CMD_DEXT | IXGBE_ADVTXD_DTYP_CTXT;
3748
3749         context_desc->vlan_macip_lens   = cpu_to_le32(vlan_macip_lens);
3750         context_desc->fceof_saidx       = cpu_to_le32(fceof_saidx);
3751         context_desc->type_tucmd_mlhl   = cpu_to_le32(type_tucmd);
3752         context_desc->mss_l4len_idx     = cpu_to_le32(mss_l4len_idx);
3753 }
3754
3755 static int ixgbevf_tso(struct ixgbevf_ring *tx_ring,
3756                        struct ixgbevf_tx_buffer *first,
3757                        u8 *hdr_len,
3758                        struct ixgbevf_ipsec_tx_data *itd)
3759 {
3760         u32 vlan_macip_lens, type_tucmd, mss_l4len_idx;
3761         struct sk_buff *skb = first->skb;
3762         union {
3763                 struct iphdr *v4;
3764                 struct ipv6hdr *v6;
3765                 unsigned char *hdr;
3766         } ip;
3767         union {
3768                 struct tcphdr *tcp;
3769                 unsigned char *hdr;
3770         } l4;
3771         u32 paylen, l4_offset;
3772         u32 fceof_saidx = 0;
3773         int err;
3774
3775         if (skb->ip_summed != CHECKSUM_PARTIAL)
3776                 return 0;
3777
3778         if (!skb_is_gso(skb))
3779                 return 0;
3780
3781         err = skb_cow_head(skb, 0);
3782         if (err < 0)
3783                 return err;
3784
3785         if (eth_p_mpls(first->protocol))
3786                 ip.hdr = skb_inner_network_header(skb);
3787         else
3788                 ip.hdr = skb_network_header(skb);
3789         l4.hdr = skb_checksum_start(skb);
3790
3791         /* ADV DTYP TUCMD MKRLOC/ISCSIHEDLEN */
3792         type_tucmd = IXGBE_ADVTXD_TUCMD_L4T_TCP;
3793
3794         /* initialize outer IP header fields */
3795         if (ip.v4->version == 4) {
3796                 unsigned char *csum_start = skb_checksum_start(skb);
3797                 unsigned char *trans_start = ip.hdr + (ip.v4->ihl * 4);
3798                 int len = csum_start - trans_start;
3799
3800                 /* IP header will have to cancel out any data that
3801                  * is not a part of the outer IP header, so set to
3802                  * a reverse csum if needed, else init check to 0.
3803                  */
3804                 ip.v4->check = (skb_shinfo(skb)->gso_type & SKB_GSO_PARTIAL) ?
3805                                            csum_fold(csum_partial(trans_start,
3806                                                                   len, 0)) : 0;
3807                 type_tucmd |= IXGBE_ADVTXD_TUCMD_IPV4;
3808
3809                 ip.v4->tot_len = 0;
3810                 first->tx_flags |= IXGBE_TX_FLAGS_TSO |
3811                                    IXGBE_TX_FLAGS_CSUM |
3812                                    IXGBE_TX_FLAGS_IPV4;
3813         } else {
3814                 ip.v6->payload_len = 0;
3815                 first->tx_flags |= IXGBE_TX_FLAGS_TSO |
3816                                    IXGBE_TX_FLAGS_CSUM;
3817         }
3818
3819         /* determine offset of inner transport header */
3820         l4_offset = l4.hdr - skb->data;
3821
3822         /* compute length of segmentation header */
3823         *hdr_len = (l4.tcp->doff * 4) + l4_offset;
3824
3825         /* remove payload length from inner checksum */
3826         paylen = skb->len - l4_offset;
3827         csum_replace_by_diff(&l4.tcp->check, (__force __wsum)htonl(paylen));
3828
3829         /* update gso size and bytecount with header size */
3830         first->gso_segs = skb_shinfo(skb)->gso_segs;
3831         first->bytecount += (first->gso_segs - 1) * *hdr_len;
3832
3833         /* mss_l4len_id: use 1 as index for TSO */
3834         mss_l4len_idx = (*hdr_len - l4_offset) << IXGBE_ADVTXD_L4LEN_SHIFT;
3835         mss_l4len_idx |= skb_shinfo(skb)->gso_size << IXGBE_ADVTXD_MSS_SHIFT;
3836         mss_l4len_idx |= (1u << IXGBE_ADVTXD_IDX_SHIFT);
3837
3838         fceof_saidx |= itd->pfsa;
3839         type_tucmd |= itd->flags | itd->trailer_len;
3840
3841         /* vlan_macip_lens: HEADLEN, MACLEN, VLAN tag */
3842         vlan_macip_lens = l4.hdr - ip.hdr;
3843         vlan_macip_lens |= (ip.hdr - skb->data) << IXGBE_ADVTXD_MACLEN_SHIFT;
3844         vlan_macip_lens |= first->tx_flags & IXGBE_TX_FLAGS_VLAN_MASK;
3845
3846         ixgbevf_tx_ctxtdesc(tx_ring, vlan_macip_lens, fceof_saidx, type_tucmd,
3847                             mss_l4len_idx);
3848
3849         return 1;
3850 }
3851
3852 static void ixgbevf_tx_csum(struct ixgbevf_ring *tx_ring,
3853                             struct ixgbevf_tx_buffer *first,
3854                             struct ixgbevf_ipsec_tx_data *itd)
3855 {
3856         struct sk_buff *skb = first->skb;
3857         u32 vlan_macip_lens = 0;
3858         u32 fceof_saidx = 0;
3859         u32 type_tucmd = 0;
3860
3861         if (skb->ip_summed != CHECKSUM_PARTIAL)
3862                 goto no_csum;
3863
3864         switch (skb->csum_offset) {
3865         case offsetof(struct tcphdr, check):
3866                 type_tucmd = IXGBE_ADVTXD_TUCMD_L4T_TCP;
3867                 fallthrough;
3868         case offsetof(struct udphdr, check):
3869                 break;
3870         case offsetof(struct sctphdr, checksum):
3871                 /* validate that this is actually an SCTP request */
3872                 if (skb_csum_is_sctp(skb)) {
3873                         type_tucmd = IXGBE_ADVTXD_TUCMD_L4T_SCTP;
3874                         break;
3875                 }
3876                 fallthrough;
3877         default:
3878                 skb_checksum_help(skb);
3879                 goto no_csum;
3880         }
3881
3882         if (first->protocol == htons(ETH_P_IP))
3883                 type_tucmd |= IXGBE_ADVTXD_TUCMD_IPV4;
3884
3885         /* update TX checksum flag */
3886         first->tx_flags |= IXGBE_TX_FLAGS_CSUM;
3887         vlan_macip_lens = skb_checksum_start_offset(skb) -
3888                           skb_network_offset(skb);
3889 no_csum:
3890         /* vlan_macip_lens: MACLEN, VLAN tag */
3891         vlan_macip_lens |= skb_network_offset(skb) << IXGBE_ADVTXD_MACLEN_SHIFT;
3892         vlan_macip_lens |= first->tx_flags & IXGBE_TX_FLAGS_VLAN_MASK;
3893
3894         fceof_saidx |= itd->pfsa;
3895         type_tucmd |= itd->flags | itd->trailer_len;
3896
3897         ixgbevf_tx_ctxtdesc(tx_ring, vlan_macip_lens,
3898                             fceof_saidx, type_tucmd, 0);
3899 }
3900
3901 static __le32 ixgbevf_tx_cmd_type(u32 tx_flags)
3902 {
3903         /* set type for advanced descriptor with frame checksum insertion */
3904         __le32 cmd_type = cpu_to_le32(IXGBE_ADVTXD_DTYP_DATA |
3905                                       IXGBE_ADVTXD_DCMD_IFCS |
3906                                       IXGBE_ADVTXD_DCMD_DEXT);
3907
3908         /* set HW VLAN bit if VLAN is present */
3909         if (tx_flags & IXGBE_TX_FLAGS_VLAN)
3910                 cmd_type |= cpu_to_le32(IXGBE_ADVTXD_DCMD_VLE);
3911
3912         /* set segmentation enable bits for TSO/FSO */
3913         if (tx_flags & IXGBE_TX_FLAGS_TSO)
3914                 cmd_type |= cpu_to_le32(IXGBE_ADVTXD_DCMD_TSE);
3915
3916         return cmd_type;
3917 }
3918
3919 static void ixgbevf_tx_olinfo_status(union ixgbe_adv_tx_desc *tx_desc,
3920                                      u32 tx_flags, unsigned int paylen)
3921 {
3922         __le32 olinfo_status = cpu_to_le32(paylen << IXGBE_ADVTXD_PAYLEN_SHIFT);
3923
3924         /* enable L4 checksum for TSO and TX checksum offload */
3925         if (tx_flags & IXGBE_TX_FLAGS_CSUM)
3926                 olinfo_status |= cpu_to_le32(IXGBE_ADVTXD_POPTS_TXSM);
3927
3928         /* enble IPv4 checksum for TSO */
3929         if (tx_flags & IXGBE_TX_FLAGS_IPV4)
3930                 olinfo_status |= cpu_to_le32(IXGBE_ADVTXD_POPTS_IXSM);
3931
3932         /* enable IPsec */
3933         if (tx_flags & IXGBE_TX_FLAGS_IPSEC)
3934                 olinfo_status |= cpu_to_le32(IXGBE_ADVTXD_POPTS_IPSEC);
3935
3936         /* use index 1 context for TSO/FSO/FCOE/IPSEC */
3937         if (tx_flags & (IXGBE_TX_FLAGS_TSO | IXGBE_TX_FLAGS_IPSEC))
3938                 olinfo_status |= cpu_to_le32(1u << IXGBE_ADVTXD_IDX_SHIFT);
3939
3940         /* Check Context must be set if Tx switch is enabled, which it
3941          * always is for case where virtual functions are running
3942          */
3943         olinfo_status |= cpu_to_le32(IXGBE_ADVTXD_CC);
3944
3945         tx_desc->read.olinfo_status = olinfo_status;
3946 }
3947
3948 static void ixgbevf_tx_map(struct ixgbevf_ring *tx_ring,
3949                            struct ixgbevf_tx_buffer *first,
3950                            const u8 hdr_len)
3951 {
3952         struct sk_buff *skb = first->skb;
3953         struct ixgbevf_tx_buffer *tx_buffer;
3954         union ixgbe_adv_tx_desc *tx_desc;
3955         skb_frag_t *frag;
3956         dma_addr_t dma;
3957         unsigned int data_len, size;
3958         u32 tx_flags = first->tx_flags;
3959         __le32 cmd_type = ixgbevf_tx_cmd_type(tx_flags);
3960         u16 i = tx_ring->next_to_use;
3961
3962         tx_desc = IXGBEVF_TX_DESC(tx_ring, i);
3963
3964         ixgbevf_tx_olinfo_status(tx_desc, tx_flags, skb->len - hdr_len);
3965
3966         size = skb_headlen(skb);
3967         data_len = skb->data_len;
3968
3969         dma = dma_map_single(tx_ring->dev, skb->data, size, DMA_TO_DEVICE);
3970
3971         tx_buffer = first;
3972
3973         for (frag = &skb_shinfo(skb)->frags[0];; frag++) {
3974                 if (dma_mapping_error(tx_ring->dev, dma))
3975                         goto dma_error;
3976
3977                 /* record length, and DMA address */
3978                 dma_unmap_len_set(tx_buffer, len, size);
3979                 dma_unmap_addr_set(tx_buffer, dma, dma);
3980
3981                 tx_desc->read.buffer_addr = cpu_to_le64(dma);
3982
3983                 while (unlikely(size > IXGBE_MAX_DATA_PER_TXD)) {
3984                         tx_desc->read.cmd_type_len =
3985                                 cmd_type | cpu_to_le32(IXGBE_MAX_DATA_PER_TXD);
3986
3987                         i++;
3988                         tx_desc++;
3989                         if (i == tx_ring->count) {
3990                                 tx_desc = IXGBEVF_TX_DESC(tx_ring, 0);
3991                                 i = 0;
3992                         }
3993                         tx_desc->read.olinfo_status = 0;
3994
3995                         dma += IXGBE_MAX_DATA_PER_TXD;
3996                         size -= IXGBE_MAX_DATA_PER_TXD;
3997
3998                         tx_desc->read.buffer_addr = cpu_to_le64(dma);
3999                 }
4000
4001                 if (likely(!data_len))
4002                         break;
4003
4004                 tx_desc->read.cmd_type_len = cmd_type | cpu_to_le32(size);
4005
4006                 i++;
4007                 tx_desc++;
4008                 if (i == tx_ring->count) {
4009                         tx_desc = IXGBEVF_TX_DESC(tx_ring, 0);
4010                         i = 0;
4011                 }
4012                 tx_desc->read.olinfo_status = 0;
4013
4014                 size = skb_frag_size(frag);
4015                 data_len -= size;
4016
4017                 dma = skb_frag_dma_map(tx_ring->dev, frag, 0, size,
4018                                        DMA_TO_DEVICE);
4019
4020                 tx_buffer = &tx_ring->tx_buffer_info[i];
4021         }
4022
4023         /* write last descriptor with RS and EOP bits */
4024         cmd_type |= cpu_to_le32(size) | cpu_to_le32(IXGBE_TXD_CMD);
4025         tx_desc->read.cmd_type_len = cmd_type;
4026
4027         /* set the timestamp */
4028         first->time_stamp = jiffies;
4029
4030         skb_tx_timestamp(skb);
4031
4032         /* Force memory writes to complete before letting h/w know there
4033          * are new descriptors to fetch.  (Only applicable for weak-ordered
4034          * memory model archs, such as IA-64).
4035          *
4036          * We also need this memory barrier (wmb) to make certain all of the
4037          * status bits have been updated before next_to_watch is written.
4038          */
4039         wmb();
4040
4041         /* set next_to_watch value indicating a packet is present */
4042         first->next_to_watch = tx_desc;
4043
4044         i++;
4045         if (i == tx_ring->count)
4046                 i = 0;
4047
4048         tx_ring->next_to_use = i;
4049
4050         /* notify HW of packet */
4051         ixgbevf_write_tail(tx_ring, i);
4052
4053         return;
4054 dma_error:
4055         dev_err(tx_ring->dev, "TX DMA map failed\n");
4056         tx_buffer = &tx_ring->tx_buffer_info[i];
4057
4058         /* clear dma mappings for failed tx_buffer_info map */
4059         while (tx_buffer != first) {
4060                 if (dma_unmap_len(tx_buffer, len))
4061                         dma_unmap_page(tx_ring->dev,
4062                                        dma_unmap_addr(tx_buffer, dma),
4063                                        dma_unmap_len(tx_buffer, len),
4064                                        DMA_TO_DEVICE);
4065                 dma_unmap_len_set(tx_buffer, len, 0);
4066
4067                 if (i-- == 0)
4068                         i += tx_ring->count;
4069                 tx_buffer = &tx_ring->tx_buffer_info[i];
4070         }
4071
4072         if (dma_unmap_len(tx_buffer, len))
4073                 dma_unmap_single(tx_ring->dev,
4074                                  dma_unmap_addr(tx_buffer, dma),
4075                                  dma_unmap_len(tx_buffer, len),
4076                                  DMA_TO_DEVICE);
4077         dma_unmap_len_set(tx_buffer, len, 0);
4078
4079         dev_kfree_skb_any(tx_buffer->skb);
4080         tx_buffer->skb = NULL;
4081
4082         tx_ring->next_to_use = i;
4083 }
4084
4085 static int __ixgbevf_maybe_stop_tx(struct ixgbevf_ring *tx_ring, int size)
4086 {
4087         netif_stop_subqueue(tx_ring->netdev, tx_ring->queue_index);
4088         /* Herbert's original patch had:
4089          *  smp_mb__after_netif_stop_queue();
4090          * but since that doesn't exist yet, just open code it.
4091          */
4092         smp_mb();
4093
4094         /* We need to check again in a case another CPU has just
4095          * made room available.
4096          */
4097         if (likely(ixgbevf_desc_unused(tx_ring) < size))
4098                 return -EBUSY;
4099
4100         /* A reprieve! - use start_queue because it doesn't call schedule */
4101         netif_start_subqueue(tx_ring->netdev, tx_ring->queue_index);
4102         ++tx_ring->tx_stats.restart_queue;
4103
4104         return 0;
4105 }
4106
4107 static int ixgbevf_maybe_stop_tx(struct ixgbevf_ring *tx_ring, int size)
4108 {
4109         if (likely(ixgbevf_desc_unused(tx_ring) >= size))
4110                 return 0;
4111         return __ixgbevf_maybe_stop_tx(tx_ring, size);
4112 }
4113
4114 static int ixgbevf_xmit_frame_ring(struct sk_buff *skb,
4115                                    struct ixgbevf_ring *tx_ring)
4116 {
4117         struct ixgbevf_tx_buffer *first;
4118         int tso;
4119         u32 tx_flags = 0;
4120         u16 count = TXD_USE_COUNT(skb_headlen(skb));
4121         struct ixgbevf_ipsec_tx_data ipsec_tx = { 0 };
4122 #if PAGE_SIZE > IXGBE_MAX_DATA_PER_TXD
4123         unsigned short f;
4124 #endif
4125         u8 hdr_len = 0;
4126         u8 *dst_mac = skb_header_pointer(skb, 0, 0, NULL);
4127
4128         if (!dst_mac || is_link_local_ether_addr(dst_mac)) {
4129                 dev_kfree_skb_any(skb);
4130                 return NETDEV_TX_OK;
4131         }
4132
4133         /* need: 1 descriptor per page * PAGE_SIZE/IXGBE_MAX_DATA_PER_TXD,
4134          *       + 1 desc for skb_headlen/IXGBE_MAX_DATA_PER_TXD,
4135          *       + 2 desc gap to keep tail from touching head,
4136          *       + 1 desc for context descriptor,
4137          * otherwise try next time
4138          */
4139 #if PAGE_SIZE > IXGBE_MAX_DATA_PER_TXD
4140         for (f = 0; f < skb_shinfo(skb)->nr_frags; f++) {
4141                 skb_frag_t *frag = &skb_shinfo(skb)->frags[f];
4142
4143                 count += TXD_USE_COUNT(skb_frag_size(frag));
4144         }
4145 #else
4146         count += skb_shinfo(skb)->nr_frags;
4147 #endif
4148         if (ixgbevf_maybe_stop_tx(tx_ring, count + 3)) {
4149                 tx_ring->tx_stats.tx_busy++;
4150                 return NETDEV_TX_BUSY;
4151         }
4152
4153         /* record the location of the first descriptor for this packet */
4154         first = &tx_ring->tx_buffer_info[tx_ring->next_to_use];
4155         first->skb = skb;
4156         first->bytecount = skb->len;
4157         first->gso_segs = 1;
4158
4159         if (skb_vlan_tag_present(skb)) {
4160                 tx_flags |= skb_vlan_tag_get(skb);
4161                 tx_flags <<= IXGBE_TX_FLAGS_VLAN_SHIFT;
4162                 tx_flags |= IXGBE_TX_FLAGS_VLAN;
4163         }
4164
4165         /* record initial flags and protocol */
4166         first->tx_flags = tx_flags;
4167         first->protocol = vlan_get_protocol(skb);
4168
4169 #ifdef CONFIG_IXGBEVF_IPSEC
4170         if (xfrm_offload(skb) && !ixgbevf_ipsec_tx(tx_ring, first, &ipsec_tx))
4171                 goto out_drop;
4172 #endif
4173         tso = ixgbevf_tso(tx_ring, first, &hdr_len, &ipsec_tx);
4174         if (tso < 0)
4175                 goto out_drop;
4176         else if (!tso)
4177                 ixgbevf_tx_csum(tx_ring, first, &ipsec_tx);
4178
4179         ixgbevf_tx_map(tx_ring, first, hdr_len);
4180
4181         ixgbevf_maybe_stop_tx(tx_ring, DESC_NEEDED);
4182
4183         return NETDEV_TX_OK;
4184
4185 out_drop:
4186         dev_kfree_skb_any(first->skb);
4187         first->skb = NULL;
4188
4189         return NETDEV_TX_OK;
4190 }
4191
4192 static netdev_tx_t ixgbevf_xmit_frame(struct sk_buff *skb, struct net_device *netdev)
4193 {
4194         struct ixgbevf_adapter *adapter = netdev_priv(netdev);
4195         struct ixgbevf_ring *tx_ring;
4196
4197         if (skb->len <= 0) {
4198                 dev_kfree_skb_any(skb);
4199                 return NETDEV_TX_OK;
4200         }
4201
4202         /* The minimum packet size for olinfo paylen is 17 so pad the skb
4203          * in order to meet this minimum size requirement.
4204          */
4205         if (skb->len < 17) {
4206                 if (skb_padto(skb, 17))
4207                         return NETDEV_TX_OK;
4208                 skb->len = 17;
4209         }
4210
4211         tx_ring = adapter->tx_ring[skb->queue_mapping];
4212         return ixgbevf_xmit_frame_ring(skb, tx_ring);
4213 }
4214
4215 /**
4216  * ixgbevf_set_mac - Change the Ethernet Address of the NIC
4217  * @netdev: network interface device structure
4218  * @p: pointer to an address structure
4219  *
4220  * Returns 0 on success, negative on failure
4221  **/
4222 static int ixgbevf_set_mac(struct net_device *netdev, void *p)
4223 {
4224         struct ixgbevf_adapter *adapter = netdev_priv(netdev);
4225         struct ixgbe_hw *hw = &adapter->hw;
4226         struct sockaddr *addr = p;
4227         int err;
4228
4229         if (!is_valid_ether_addr(addr->sa_data))
4230                 return -EADDRNOTAVAIL;
4231
4232         spin_lock_bh(&adapter->mbx_lock);
4233
4234         err = hw->mac.ops.set_rar(hw, 0, addr->sa_data, 0);
4235
4236         spin_unlock_bh(&adapter->mbx_lock);
4237
4238         if (err)
4239                 return -EPERM;
4240
4241         ether_addr_copy(hw->mac.addr, addr->sa_data);
4242         ether_addr_copy(hw->mac.perm_addr, addr->sa_data);
4243         eth_hw_addr_set(netdev, addr->sa_data);
4244
4245         return 0;
4246 }
4247
4248 /**
4249  * ixgbevf_change_mtu - Change the Maximum Transfer Unit
4250  * @netdev: network interface device structure
4251  * @new_mtu: new value for maximum frame size
4252  *
4253  * Returns 0 on success, negative on failure
4254  **/
4255 static int ixgbevf_change_mtu(struct net_device *netdev, int new_mtu)
4256 {
4257         struct ixgbevf_adapter *adapter = netdev_priv(netdev);
4258         struct ixgbe_hw *hw = &adapter->hw;
4259         int max_frame = new_mtu + ETH_HLEN + ETH_FCS_LEN;
4260         int ret;
4261
4262         /* prevent MTU being changed to a size unsupported by XDP */
4263         if (adapter->xdp_prog) {
4264                 dev_warn(&adapter->pdev->dev, "MTU cannot be changed while XDP program is loaded\n");
4265                 return -EPERM;
4266         }
4267
4268         spin_lock_bh(&adapter->mbx_lock);
4269         /* notify the PF of our intent to use this size of frame */
4270         ret = hw->mac.ops.set_rlpml(hw, max_frame);
4271         spin_unlock_bh(&adapter->mbx_lock);
4272         if (ret)
4273                 return -EINVAL;
4274
4275         hw_dbg(hw, "changing MTU from %d to %d\n",
4276                netdev->mtu, new_mtu);
4277
4278         /* must set new MTU before calling down or up */
4279         netdev->mtu = new_mtu;
4280
4281         if (netif_running(netdev))
4282                 ixgbevf_reinit_locked(adapter);
4283
4284         return 0;
4285 }
4286
4287 static int __maybe_unused ixgbevf_suspend(struct device *dev_d)
4288 {
4289         struct net_device *netdev = dev_get_drvdata(dev_d);
4290         struct ixgbevf_adapter *adapter = netdev_priv(netdev);
4291
4292         rtnl_lock();
4293         netif_device_detach(netdev);
4294
4295         if (netif_running(netdev))
4296                 ixgbevf_close_suspend(adapter);
4297
4298         ixgbevf_clear_interrupt_scheme(adapter);
4299         rtnl_unlock();
4300
4301         return 0;
4302 }
4303
4304 static int __maybe_unused ixgbevf_resume(struct device *dev_d)
4305 {
4306         struct pci_dev *pdev = to_pci_dev(dev_d);
4307         struct net_device *netdev = pci_get_drvdata(pdev);
4308         struct ixgbevf_adapter *adapter = netdev_priv(netdev);
4309         u32 err;
4310
4311         adapter->hw.hw_addr = adapter->io_addr;
4312         smp_mb__before_atomic();
4313         clear_bit(__IXGBEVF_DISABLED, &adapter->state);
4314         pci_set_master(pdev);
4315
4316         ixgbevf_reset(adapter);
4317
4318         rtnl_lock();
4319         err = ixgbevf_init_interrupt_scheme(adapter);
4320         if (!err && netif_running(netdev))
4321                 err = ixgbevf_open(netdev);
4322         rtnl_unlock();
4323         if (err)
4324                 return err;
4325
4326         netif_device_attach(netdev);
4327
4328         return err;
4329 }
4330
4331 static void ixgbevf_shutdown(struct pci_dev *pdev)
4332 {
4333         ixgbevf_suspend(&pdev->dev);
4334 }
4335
4336 static void ixgbevf_get_tx_ring_stats(struct rtnl_link_stats64 *stats,
4337                                       const struct ixgbevf_ring *ring)
4338 {
4339         u64 bytes, packets;
4340         unsigned int start;
4341
4342         if (ring) {
4343                 do {
4344                         start = u64_stats_fetch_begin_irq(&ring->syncp);
4345                         bytes = ring->stats.bytes;
4346                         packets = ring->stats.packets;
4347                 } while (u64_stats_fetch_retry_irq(&ring->syncp, start));
4348                 stats->tx_bytes += bytes;
4349                 stats->tx_packets += packets;
4350         }
4351 }
4352
4353 static void ixgbevf_get_stats(struct net_device *netdev,
4354                               struct rtnl_link_stats64 *stats)
4355 {
4356         struct ixgbevf_adapter *adapter = netdev_priv(netdev);
4357         unsigned int start;
4358         u64 bytes, packets;
4359         const struct ixgbevf_ring *ring;
4360         int i;
4361
4362         ixgbevf_update_stats(adapter);
4363
4364         stats->multicast = adapter->stats.vfmprc - adapter->stats.base_vfmprc;
4365
4366         rcu_read_lock();
4367         for (i = 0; i < adapter->num_rx_queues; i++) {
4368                 ring = adapter->rx_ring[i];
4369                 do {
4370                         start = u64_stats_fetch_begin_irq(&ring->syncp);
4371                         bytes = ring->stats.bytes;
4372                         packets = ring->stats.packets;
4373                 } while (u64_stats_fetch_retry_irq(&ring->syncp, start));
4374                 stats->rx_bytes += bytes;
4375                 stats->rx_packets += packets;
4376         }
4377
4378         for (i = 0; i < adapter->num_tx_queues; i++) {
4379                 ring = adapter->tx_ring[i];
4380                 ixgbevf_get_tx_ring_stats(stats, ring);
4381         }
4382
4383         for (i = 0; i < adapter->num_xdp_queues; i++) {
4384                 ring = adapter->xdp_ring[i];
4385                 ixgbevf_get_tx_ring_stats(stats, ring);
4386         }
4387         rcu_read_unlock();
4388 }
4389
4390 #define IXGBEVF_MAX_MAC_HDR_LEN         127
4391 #define IXGBEVF_MAX_NETWORK_HDR_LEN     511
4392
4393 static netdev_features_t
4394 ixgbevf_features_check(struct sk_buff *skb, struct net_device *dev,
4395                        netdev_features_t features)
4396 {
4397         unsigned int network_hdr_len, mac_hdr_len;
4398
4399         /* Make certain the headers can be described by a context descriptor */
4400         mac_hdr_len = skb_network_header(skb) - skb->data;
4401         if (unlikely(mac_hdr_len > IXGBEVF_MAX_MAC_HDR_LEN))
4402                 return features & ~(NETIF_F_HW_CSUM |
4403                                     NETIF_F_SCTP_CRC |
4404                                     NETIF_F_HW_VLAN_CTAG_TX |
4405                                     NETIF_F_TSO |
4406                                     NETIF_F_TSO6);
4407
4408         network_hdr_len = skb_checksum_start(skb) - skb_network_header(skb);
4409         if (unlikely(network_hdr_len >  IXGBEVF_MAX_NETWORK_HDR_LEN))
4410                 return features & ~(NETIF_F_HW_CSUM |
4411                                     NETIF_F_SCTP_CRC |
4412                                     NETIF_F_TSO |
4413                                     NETIF_F_TSO6);
4414
4415         /* We can only support IPV4 TSO in tunnels if we can mangle the
4416          * inner IP ID field, so strip TSO if MANGLEID is not supported.
4417          */
4418         if (skb->encapsulation && !(features & NETIF_F_TSO_MANGLEID))
4419                 features &= ~NETIF_F_TSO;
4420
4421         return features;
4422 }
4423
4424 static int ixgbevf_xdp_setup(struct net_device *dev, struct bpf_prog *prog)
4425 {
4426         int i, frame_size = dev->mtu + ETH_HLEN + ETH_FCS_LEN + VLAN_HLEN;
4427         struct ixgbevf_adapter *adapter = netdev_priv(dev);
4428         struct bpf_prog *old_prog;
4429
4430         /* verify ixgbevf ring attributes are sufficient for XDP */
4431         for (i = 0; i < adapter->num_rx_queues; i++) {
4432                 struct ixgbevf_ring *ring = adapter->rx_ring[i];
4433
4434                 if (frame_size > ixgbevf_rx_bufsz(ring))
4435                         return -EINVAL;
4436         }
4437
4438         old_prog = xchg(&adapter->xdp_prog, prog);
4439
4440         /* If transitioning XDP modes reconfigure rings */
4441         if (!!prog != !!old_prog) {
4442                 /* Hardware has to reinitialize queues and interrupts to
4443                  * match packet buffer alignment. Unfortunately, the
4444                  * hardware is not flexible enough to do this dynamically.
4445                  */
4446                 if (netif_running(dev))
4447                         ixgbevf_close(dev);
4448
4449                 ixgbevf_clear_interrupt_scheme(adapter);
4450                 ixgbevf_init_interrupt_scheme(adapter);
4451
4452                 if (netif_running(dev))
4453                         ixgbevf_open(dev);
4454         } else {
4455                 for (i = 0; i < adapter->num_rx_queues; i++)
4456                         xchg(&adapter->rx_ring[i]->xdp_prog, adapter->xdp_prog);
4457         }
4458
4459         if (old_prog)
4460                 bpf_prog_put(old_prog);
4461
4462         return 0;
4463 }
4464
4465 static int ixgbevf_xdp(struct net_device *dev, struct netdev_bpf *xdp)
4466 {
4467         switch (xdp->command) {
4468         case XDP_SETUP_PROG:
4469                 return ixgbevf_xdp_setup(dev, xdp->prog);
4470         default:
4471                 return -EINVAL;
4472         }
4473 }
4474
4475 static const struct net_device_ops ixgbevf_netdev_ops = {
4476         .ndo_open               = ixgbevf_open,
4477         .ndo_stop               = ixgbevf_close,
4478         .ndo_start_xmit         = ixgbevf_xmit_frame,
4479         .ndo_set_rx_mode        = ixgbevf_set_rx_mode,
4480         .ndo_get_stats64        = ixgbevf_get_stats,
4481         .ndo_validate_addr      = eth_validate_addr,
4482         .ndo_set_mac_address    = ixgbevf_set_mac,
4483         .ndo_change_mtu         = ixgbevf_change_mtu,
4484         .ndo_tx_timeout         = ixgbevf_tx_timeout,
4485         .ndo_vlan_rx_add_vid    = ixgbevf_vlan_rx_add_vid,
4486         .ndo_vlan_rx_kill_vid   = ixgbevf_vlan_rx_kill_vid,
4487         .ndo_features_check     = ixgbevf_features_check,
4488         .ndo_bpf                = ixgbevf_xdp,
4489 };
4490
4491 static void ixgbevf_assign_netdev_ops(struct net_device *dev)
4492 {
4493         dev->netdev_ops = &ixgbevf_netdev_ops;
4494         ixgbevf_set_ethtool_ops(dev);
4495         dev->watchdog_timeo = 5 * HZ;
4496 }
4497
4498 /**
4499  * ixgbevf_probe - Device Initialization Routine
4500  * @pdev: PCI device information struct
4501  * @ent: entry in ixgbevf_pci_tbl
4502  *
4503  * Returns 0 on success, negative on failure
4504  *
4505  * ixgbevf_probe initializes an adapter identified by a pci_dev structure.
4506  * The OS initialization, configuring of the adapter private structure,
4507  * and a hardware reset occur.
4508  **/
4509 static int ixgbevf_probe(struct pci_dev *pdev, const struct pci_device_id *ent)
4510 {
4511         struct net_device *netdev;
4512         struct ixgbevf_adapter *adapter = NULL;
4513         struct ixgbe_hw *hw = NULL;
4514         const struct ixgbevf_info *ii = ixgbevf_info_tbl[ent->driver_data];
4515         bool disable_dev = false;
4516         int err;
4517
4518         err = pci_enable_device(pdev);
4519         if (err)
4520                 return err;
4521
4522         err = dma_set_mask_and_coherent(&pdev->dev, DMA_BIT_MASK(64));
4523         if (err) {
4524                 dev_err(&pdev->dev, "No usable DMA configuration, aborting\n");
4525                 goto err_dma;
4526         }
4527
4528         err = pci_request_regions(pdev, ixgbevf_driver_name);
4529         if (err) {
4530                 dev_err(&pdev->dev, "pci_request_regions failed 0x%x\n", err);
4531                 goto err_pci_reg;
4532         }
4533
4534         pci_set_master(pdev);
4535
4536         netdev = alloc_etherdev_mq(sizeof(struct ixgbevf_adapter),
4537                                    MAX_TX_QUEUES);
4538         if (!netdev) {
4539                 err = -ENOMEM;
4540                 goto err_alloc_etherdev;
4541         }
4542
4543         SET_NETDEV_DEV(netdev, &pdev->dev);
4544
4545         adapter = netdev_priv(netdev);
4546
4547         adapter->netdev = netdev;
4548         adapter->pdev = pdev;
4549         hw = &adapter->hw;
4550         hw->back = adapter;
4551         adapter->msg_enable = netif_msg_init(debug, DEFAULT_MSG_ENABLE);
4552
4553         /* call save state here in standalone driver because it relies on
4554          * adapter struct to exist, and needs to call netdev_priv
4555          */
4556         pci_save_state(pdev);
4557
4558         hw->hw_addr = ioremap(pci_resource_start(pdev, 0),
4559                               pci_resource_len(pdev, 0));
4560         adapter->io_addr = hw->hw_addr;
4561         if (!hw->hw_addr) {
4562                 err = -EIO;
4563                 goto err_ioremap;
4564         }
4565
4566         ixgbevf_assign_netdev_ops(netdev);
4567
4568         /* Setup HW API */
4569         memcpy(&hw->mac.ops, ii->mac_ops, sizeof(hw->mac.ops));
4570         hw->mac.type  = ii->mac;
4571
4572         memcpy(&hw->mbx.ops, &ixgbevf_mbx_ops_legacy,
4573                sizeof(struct ixgbe_mbx_operations));
4574
4575         /* setup the private structure */
4576         err = ixgbevf_sw_init(adapter);
4577         if (err)
4578                 goto err_sw_init;
4579
4580         /* The HW MAC address was set and/or determined in sw_init */
4581         if (!is_valid_ether_addr(netdev->dev_addr)) {
4582                 pr_err("invalid MAC address\n");
4583                 err = -EIO;
4584                 goto err_sw_init;
4585         }
4586
4587         netdev->hw_features = NETIF_F_SG |
4588                               NETIF_F_TSO |
4589                               NETIF_F_TSO6 |
4590                               NETIF_F_RXCSUM |
4591                               NETIF_F_HW_CSUM |
4592                               NETIF_F_SCTP_CRC;
4593
4594 #define IXGBEVF_GSO_PARTIAL_FEATURES (NETIF_F_GSO_GRE | \
4595                                       NETIF_F_GSO_GRE_CSUM | \
4596                                       NETIF_F_GSO_IPXIP4 | \
4597                                       NETIF_F_GSO_IPXIP6 | \
4598                                       NETIF_F_GSO_UDP_TUNNEL | \
4599                                       NETIF_F_GSO_UDP_TUNNEL_CSUM)
4600
4601         netdev->gso_partial_features = IXGBEVF_GSO_PARTIAL_FEATURES;
4602         netdev->hw_features |= NETIF_F_GSO_PARTIAL |
4603                                IXGBEVF_GSO_PARTIAL_FEATURES;
4604
4605         netdev->features = netdev->hw_features | NETIF_F_HIGHDMA;
4606
4607         netdev->vlan_features |= netdev->features | NETIF_F_TSO_MANGLEID;
4608         netdev->mpls_features |= NETIF_F_SG |
4609                                  NETIF_F_TSO |
4610                                  NETIF_F_TSO6 |
4611                                  NETIF_F_HW_CSUM;
4612         netdev->mpls_features |= IXGBEVF_GSO_PARTIAL_FEATURES;
4613         netdev->hw_enc_features |= netdev->vlan_features;
4614
4615         /* set this bit last since it cannot be part of vlan_features */
4616         netdev->features |= NETIF_F_HW_VLAN_CTAG_FILTER |
4617                             NETIF_F_HW_VLAN_CTAG_RX |
4618                             NETIF_F_HW_VLAN_CTAG_TX;
4619
4620         netdev->priv_flags |= IFF_UNICAST_FLT;
4621
4622         /* MTU range: 68 - 1504 or 9710 */
4623         netdev->min_mtu = ETH_MIN_MTU;
4624         switch (adapter->hw.api_version) {
4625         case ixgbe_mbox_api_11:
4626         case ixgbe_mbox_api_12:
4627         case ixgbe_mbox_api_13:
4628         case ixgbe_mbox_api_14:
4629         case ixgbe_mbox_api_15:
4630                 netdev->max_mtu = IXGBE_MAX_JUMBO_FRAME_SIZE -
4631                                   (ETH_HLEN + ETH_FCS_LEN);
4632                 break;
4633         default:
4634                 if (adapter->hw.mac.type != ixgbe_mac_82599_vf)
4635                         netdev->max_mtu = IXGBE_MAX_JUMBO_FRAME_SIZE -
4636                                           (ETH_HLEN + ETH_FCS_LEN);
4637                 else
4638                         netdev->max_mtu = ETH_DATA_LEN + ETH_FCS_LEN;
4639                 break;
4640         }
4641
4642         if (IXGBE_REMOVED(hw->hw_addr)) {
4643                 err = -EIO;
4644                 goto err_sw_init;
4645         }
4646
4647         timer_setup(&adapter->service_timer, ixgbevf_service_timer, 0);
4648
4649         INIT_WORK(&adapter->service_task, ixgbevf_service_task);
4650         set_bit(__IXGBEVF_SERVICE_INITED, &adapter->state);
4651         clear_bit(__IXGBEVF_SERVICE_SCHED, &adapter->state);
4652
4653         err = ixgbevf_init_interrupt_scheme(adapter);
4654         if (err)
4655                 goto err_sw_init;
4656
4657         strcpy(netdev->name, "eth%d");
4658
4659         err = register_netdev(netdev);
4660         if (err)
4661                 goto err_register;
4662
4663         pci_set_drvdata(pdev, netdev);
4664         netif_carrier_off(netdev);
4665         ixgbevf_init_ipsec_offload(adapter);
4666
4667         ixgbevf_init_last_counter_stats(adapter);
4668
4669         /* print the VF info */
4670         dev_info(&pdev->dev, "%pM\n", netdev->dev_addr);
4671         dev_info(&pdev->dev, "MAC: %d\n", hw->mac.type);
4672
4673         switch (hw->mac.type) {
4674         case ixgbe_mac_X550_vf:
4675                 dev_info(&pdev->dev, "Intel(R) X550 Virtual Function\n");
4676                 break;
4677         case ixgbe_mac_X540_vf:
4678                 dev_info(&pdev->dev, "Intel(R) X540 Virtual Function\n");
4679                 break;
4680         case ixgbe_mac_82599_vf:
4681         default:
4682                 dev_info(&pdev->dev, "Intel(R) 82599 Virtual Function\n");
4683                 break;
4684         }
4685
4686         return 0;
4687
4688 err_register:
4689         ixgbevf_clear_interrupt_scheme(adapter);
4690 err_sw_init:
4691         ixgbevf_reset_interrupt_capability(adapter);
4692         iounmap(adapter->io_addr);
4693         kfree(adapter->rss_key);
4694 err_ioremap:
4695         disable_dev = !test_and_set_bit(__IXGBEVF_DISABLED, &adapter->state);
4696         free_netdev(netdev);
4697 err_alloc_etherdev:
4698         pci_release_regions(pdev);
4699 err_pci_reg:
4700 err_dma:
4701         if (!adapter || disable_dev)
4702                 pci_disable_device(pdev);
4703         return err;
4704 }
4705
4706 /**
4707  * ixgbevf_remove - Device Removal Routine
4708  * @pdev: PCI device information struct
4709  *
4710  * ixgbevf_remove is called by the PCI subsystem to alert the driver
4711  * that it should release a PCI device.  The could be caused by a
4712  * Hot-Plug event, or because the driver is going to be removed from
4713  * memory.
4714  **/
4715 static void ixgbevf_remove(struct pci_dev *pdev)
4716 {
4717         struct net_device *netdev = pci_get_drvdata(pdev);
4718         struct ixgbevf_adapter *adapter;
4719         bool disable_dev;
4720
4721         if (!netdev)
4722                 return;
4723
4724         adapter = netdev_priv(netdev);
4725
4726         set_bit(__IXGBEVF_REMOVING, &adapter->state);
4727         cancel_work_sync(&adapter->service_task);
4728
4729         if (netdev->reg_state == NETREG_REGISTERED)
4730                 unregister_netdev(netdev);
4731
4732         ixgbevf_stop_ipsec_offload(adapter);
4733         ixgbevf_clear_interrupt_scheme(adapter);
4734         ixgbevf_reset_interrupt_capability(adapter);
4735
4736         iounmap(adapter->io_addr);
4737         pci_release_regions(pdev);
4738
4739         hw_dbg(&adapter->hw, "Remove complete\n");
4740
4741         kfree(adapter->rss_key);
4742         disable_dev = !test_and_set_bit(__IXGBEVF_DISABLED, &adapter->state);
4743         free_netdev(netdev);
4744
4745         if (disable_dev)
4746                 pci_disable_device(pdev);
4747 }
4748
4749 /**
4750  * ixgbevf_io_error_detected - called when PCI error is detected
4751  * @pdev: Pointer to PCI device
4752  * @state: The current pci connection state
4753  *
4754  * This function is called after a PCI bus error affecting
4755  * this device has been detected.
4756  **/
4757 static pci_ers_result_t ixgbevf_io_error_detected(struct pci_dev *pdev,
4758                                                   pci_channel_state_t state)
4759 {
4760         struct net_device *netdev = pci_get_drvdata(pdev);
4761         struct ixgbevf_adapter *adapter = netdev_priv(netdev);
4762
4763         if (!test_bit(__IXGBEVF_SERVICE_INITED, &adapter->state))
4764                 return PCI_ERS_RESULT_DISCONNECT;
4765
4766         rtnl_lock();
4767         netif_device_detach(netdev);
4768
4769         if (netif_running(netdev))
4770                 ixgbevf_close_suspend(adapter);
4771
4772         if (state == pci_channel_io_perm_failure) {
4773                 rtnl_unlock();
4774                 return PCI_ERS_RESULT_DISCONNECT;
4775         }
4776
4777         if (!test_and_set_bit(__IXGBEVF_DISABLED, &adapter->state))
4778                 pci_disable_device(pdev);
4779         rtnl_unlock();
4780
4781         /* Request a slot slot reset. */
4782         return PCI_ERS_RESULT_NEED_RESET;
4783 }
4784
4785 /**
4786  * ixgbevf_io_slot_reset - called after the pci bus has been reset.
4787  * @pdev: Pointer to PCI device
4788  *
4789  * Restart the card from scratch, as if from a cold-boot. Implementation
4790  * resembles the first-half of the ixgbevf_resume routine.
4791  **/
4792 static pci_ers_result_t ixgbevf_io_slot_reset(struct pci_dev *pdev)
4793 {
4794         struct net_device *netdev = pci_get_drvdata(pdev);
4795         struct ixgbevf_adapter *adapter = netdev_priv(netdev);
4796
4797         if (pci_enable_device_mem(pdev)) {
4798                 dev_err(&pdev->dev,
4799                         "Cannot re-enable PCI device after reset.\n");
4800                 return PCI_ERS_RESULT_DISCONNECT;
4801         }
4802
4803         adapter->hw.hw_addr = adapter->io_addr;
4804         smp_mb__before_atomic();
4805         clear_bit(__IXGBEVF_DISABLED, &adapter->state);
4806         pci_set_master(pdev);
4807
4808         ixgbevf_reset(adapter);
4809
4810         return PCI_ERS_RESULT_RECOVERED;
4811 }
4812
4813 /**
4814  * ixgbevf_io_resume - called when traffic can start flowing again.
4815  * @pdev: Pointer to PCI device
4816  *
4817  * This callback is called when the error recovery driver tells us that
4818  * its OK to resume normal operation. Implementation resembles the
4819  * second-half of the ixgbevf_resume routine.
4820  **/
4821 static void ixgbevf_io_resume(struct pci_dev *pdev)
4822 {
4823         struct net_device *netdev = pci_get_drvdata(pdev);
4824
4825         rtnl_lock();
4826         if (netif_running(netdev))
4827                 ixgbevf_open(netdev);
4828
4829         netif_device_attach(netdev);
4830         rtnl_unlock();
4831 }
4832
4833 /* PCI Error Recovery (ERS) */
4834 static const struct pci_error_handlers ixgbevf_err_handler = {
4835         .error_detected = ixgbevf_io_error_detected,
4836         .slot_reset = ixgbevf_io_slot_reset,
4837         .resume = ixgbevf_io_resume,
4838 };
4839
4840 static SIMPLE_DEV_PM_OPS(ixgbevf_pm_ops, ixgbevf_suspend, ixgbevf_resume);
4841
4842 static struct pci_driver ixgbevf_driver = {
4843         .name           = ixgbevf_driver_name,
4844         .id_table       = ixgbevf_pci_tbl,
4845         .probe          = ixgbevf_probe,
4846         .remove         = ixgbevf_remove,
4847
4848         /* Power Management Hooks */
4849         .driver.pm      = &ixgbevf_pm_ops,
4850
4851         .shutdown       = ixgbevf_shutdown,
4852         .err_handler    = &ixgbevf_err_handler
4853 };
4854
4855 /**
4856  * ixgbevf_init_module - Driver Registration Routine
4857  *
4858  * ixgbevf_init_module is the first routine called when the driver is
4859  * loaded. All it does is register with the PCI subsystem.
4860  **/
4861 static int __init ixgbevf_init_module(void)
4862 {
4863         pr_info("%s\n", ixgbevf_driver_string);
4864         pr_info("%s\n", ixgbevf_copyright);
4865         ixgbevf_wq = create_singlethread_workqueue(ixgbevf_driver_name);
4866         if (!ixgbevf_wq) {
4867                 pr_err("%s: Failed to create workqueue\n", ixgbevf_driver_name);
4868                 return -ENOMEM;
4869         }
4870
4871         return pci_register_driver(&ixgbevf_driver);
4872 }
4873
4874 module_init(ixgbevf_init_module);
4875
4876 /**
4877  * ixgbevf_exit_module - Driver Exit Cleanup Routine
4878  *
4879  * ixgbevf_exit_module is called just before the driver is removed
4880  * from memory.
4881  **/
4882 static void __exit ixgbevf_exit_module(void)
4883 {
4884         pci_unregister_driver(&ixgbevf_driver);
4885         if (ixgbevf_wq) {
4886                 destroy_workqueue(ixgbevf_wq);
4887                 ixgbevf_wq = NULL;
4888         }
4889 }
4890
4891 #ifdef DEBUG
4892 /**
4893  * ixgbevf_get_hw_dev_name - return device name string
4894  * used by hardware layer to print debugging information
4895  * @hw: pointer to private hardware struct
4896  **/
4897 char *ixgbevf_get_hw_dev_name(struct ixgbe_hw *hw)
4898 {
4899         struct ixgbevf_adapter *adapter = hw->back;
4900
4901         return adapter->netdev->name;
4902 }
4903
4904 #endif
4905 module_exit(ixgbevf_exit_module);
4906
4907 /* ixgbevf_main.c */