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
  34 #include "ixgbevf.h"
  35
  36 const char ixgbevf_driver_name[] = "ixgbevf";
  37 static const char ixgbevf_driver_string[] =
  38         "Intel(R) 10 Gigabit PCI Express Virtual Function Network Driver";
  39
  40 #define DRV_VERSION "4.1.0-k"
  41 const char ixgbevf_driver_version[] = DRV_VERSION;
  42 static char ixgbevf_copyright[] =
  43         "Copyright (c) 2009 - 2015 Intel Corporation.";
  44
  45 static const struct ixgbevf_info *ixgbevf_info_tbl[] = {
  46         [board_82599_vf]        = &ixgbevf_82599_vf_info,
  47         [board_82599_vf_hv]     = &ixgbevf_82599_vf_hv_info,
  48         [board_X540_vf]         = &ixgbevf_X540_vf_info,
  49         [board_X540_vf_hv]      = &ixgbevf_X540_vf_hv_info,
  50         [board_X550_vf]         = &ixgbevf_X550_vf_info,
  51         [board_X550_vf_hv]      = &ixgbevf_X550_vf_hv_info,
  52         [board_X550EM_x_vf]     = &ixgbevf_X550EM_x_vf_info,
  53         [board_X550EM_x_vf_hv]  = &ixgbevf_X550EM_x_vf_hv_info,
  54         [board_x550em_a_vf]     = &ixgbevf_x550em_a_vf_info,
  55 };
  56
  57 /* ixgbevf_pci_tbl - PCI Device ID Table
  58  *
  59  * Wildcard entries (PCI_ANY_ID) should come last
  60  * Last entry must be all 0s
  61  *
  62  * { Vendor ID, Device ID, SubVendor ID, SubDevice ID,
  63  *   Class, Class Mask, private data (not used) }
  64  */
  65 static const struct pci_device_id ixgbevf_pci_tbl[] = {
  66         {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82599_VF), board_82599_vf },
  67         {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82599_VF_HV), board_82599_vf_hv },
  68         {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_X540_VF), board_X540_vf },
  69         {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_X540_VF_HV), board_X540_vf_hv },
  70         {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_X550_VF), board_X550_vf },
  71         {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_X550_VF_HV), board_X550_vf_hv },
  72         {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_X550EM_X_VF), board_X550EM_x_vf },
  73         {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_X550EM_X_VF_HV), board_X550EM_x_vf_hv},
  74         {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_X550EM_A_VF), board_x550em_a_vf },
  75         /* required last entry */
  76         {0, }
  77 };
  78 MODULE_DEVICE_TABLE(pci, ixgbevf_pci_tbl);
  79
  80 MODULE_AUTHOR("Intel Corporation, <linux.nics@intel.com>");
  81 MODULE_DESCRIPTION("Intel(R) 10 Gigabit Virtual Function Network Driver");
  82 MODULE_LICENSE("GPL");
  83 MODULE_VERSION(DRV_VERSION);
  84
  85 #define DEFAULT_MSG_ENABLE (NETIF_MSG_DRV|NETIF_MSG_PROBE|NETIF_MSG_LINK)
  86 static int debug = -1;
  87 module_param(debug, int, 0);
  88 MODULE_PARM_DESC(debug, "Debug level (0=none,...,16=all)");
  89
  90 static struct workqueue_struct *ixgbevf_wq;
  91
  92 static void ixgbevf_service_event_schedule(struct ixgbevf_adapter *adapter)
  93 {
  94         if (!test_bit(__IXGBEVF_DOWN, &adapter->state) &&
  95             !test_bit(__IXGBEVF_REMOVING, &adapter->state) &&
  96             !test_and_set_bit(__IXGBEVF_SERVICE_SCHED, &adapter->state))
  97                 queue_work(ixgbevf_wq, &adapter->service_task);
  98 }
  99
 100 static void ixgbevf_service_event_complete(struct ixgbevf_adapter *adapter)
 101 {
 102         BUG_ON(!test_bit(__IXGBEVF_SERVICE_SCHED, &adapter->state));
 103
 104         /* flush memory to make sure state is correct before next watchdog */
 105         smp_mb__before_atomic();
 106         clear_bit(__IXGBEVF_SERVICE_SCHED, &adapter->state);
 107 }
 108
 109 /* forward decls */
 110 static void ixgbevf_queue_reset_subtask(struct ixgbevf_adapter *adapter);
 111 static void ixgbevf_set_itr(struct ixgbevf_q_vector *q_vector);
 112 static void ixgbevf_free_all_rx_resources(struct ixgbevf_adapter *adapter);
 113 static bool ixgbevf_can_reuse_rx_page(struct ixgbevf_rx_buffer *rx_buffer);
 114 static void ixgbevf_reuse_rx_page(struct ixgbevf_ring *rx_ring,
 115                                   struct ixgbevf_rx_buffer *old_buff);
 116
 117 static void ixgbevf_remove_adapter(struct ixgbe_hw *hw)
 118 {
 119         struct ixgbevf_adapter *adapter = hw->back;
 120
 121         if (!hw->hw_addr)
 122                 return;
 123         hw->hw_addr = NULL;
 124         dev_err(&adapter->pdev->dev, "Adapter removed\n");
 125         if (test_bit(__IXGBEVF_SERVICE_INITED, &adapter->state))
 126                 ixgbevf_service_event_schedule(adapter);
 127 }
 128
 129 static void ixgbevf_check_remove(struct ixgbe_hw *hw, u32 reg)
 130 {
 131         u32 value;
 132
 133         /* The following check not only optimizes a bit by not
 134          * performing a read on the status register when the
 135          * register just read was a status register read that
 136          * returned IXGBE_FAILED_READ_REG. It also blocks any
 137          * potential recursion.
 138          */
 139         if (reg == IXGBE_VFSTATUS) {
 140                 ixgbevf_remove_adapter(hw);
 141                 return;
 142         }
 143         value = ixgbevf_read_reg(hw, IXGBE_VFSTATUS);
 144         if (value == IXGBE_FAILED_READ_REG)
 145                 ixgbevf_remove_adapter(hw);
 146 }
 147
 148 u32 ixgbevf_read_reg(struct ixgbe_hw *hw, u32 reg)
 149 {
 150         u8 __iomem *reg_addr = READ_ONCE(hw->hw_addr);
 151         u32 value;
 152
 153         if (IXGBE_REMOVED(reg_addr))
 154                 return IXGBE_FAILED_READ_REG;
 155         value = readl(reg_addr + reg);
 156         if (unlikely(value == IXGBE_FAILED_READ_REG))
 157                 ixgbevf_check_remove(hw, reg);
 158         return value;
 159 }
 160
 161 /**
 162  * ixgbevf_set_ivar - set IVAR registers - maps interrupt causes to vectors
 163  * @adapter: pointer to adapter struct
 164  * @direction: 0 for Rx, 1 for Tx, -1 for other causes
 165  * @queue: queue to map the corresponding interrupt to
 166  * @msix_vector: the vector to map to the corresponding queue
 167  **/
 168 static void ixgbevf_set_ivar(struct ixgbevf_adapter *adapter, s8 direction,
 169                              u8 queue, u8 msix_vector)
 170 {
 171         u32 ivar, index;
 172         struct ixgbe_hw *hw = &adapter->hw;
 173
 174         if (direction == -1) {
 175                 /* other causes */
 176                 msix_vector |= IXGBE_IVAR_ALLOC_VAL;
 177                 ivar = IXGBE_READ_REG(hw, IXGBE_VTIVAR_MISC);
 178                 ivar &= ~0xFF;
 179                 ivar |= msix_vector;
 180                 IXGBE_WRITE_REG(hw, IXGBE_VTIVAR_MISC, ivar);
 181         } else {
 182                 /* Tx or Rx causes */
 183                 msix_vector |= IXGBE_IVAR_ALLOC_VAL;
 184                 index = ((16 * (queue & 1)) + (8 * direction));
 185                 ivar = IXGBE_READ_REG(hw, IXGBE_VTIVAR(queue >> 1));
 186                 ivar &= ~(0xFF << index);
 187                 ivar |= (msix_vector << index);
 188                 IXGBE_WRITE_REG(hw, IXGBE_VTIVAR(queue >> 1), ivar);
 189         }
 190 }
 191
 192 static u64 ixgbevf_get_tx_completed(struct ixgbevf_ring *ring)
 193 {
 194         return ring->stats.packets;
 195 }
 196
 197 static u32 ixgbevf_get_tx_pending(struct ixgbevf_ring *ring)
 198 {
 199         struct ixgbevf_adapter *adapter = netdev_priv(ring->netdev);
 200         struct ixgbe_hw *hw = &adapter->hw;
 201
 202         u32 head = IXGBE_READ_REG(hw, IXGBE_VFTDH(ring->reg_idx));
 203         u32 tail = IXGBE_READ_REG(hw, IXGBE_VFTDT(ring->reg_idx));
 204
 205         if (head != tail)
 206                 return (head < tail) ?
 207                         tail - head : (tail + ring->count - head);
 208
 209         return 0;
 210 }
 211
 212 static inline bool ixgbevf_check_tx_hang(struct ixgbevf_ring *tx_ring)
 213 {
 214         u32 tx_done = ixgbevf_get_tx_completed(tx_ring);
 215         u32 tx_done_old = tx_ring->tx_stats.tx_done_old;
 216         u32 tx_pending = ixgbevf_get_tx_pending(tx_ring);
 217
 218         clear_check_for_tx_hang(tx_ring);
 219
 220         /* Check for a hung queue, but be thorough. This verifies
 221          * that a transmit has been completed since the previous
 222          * check AND there is at least one packet pending. The
 223          * ARMED bit is set to indicate a potential hang.
 224          */
 225         if ((tx_done_old == tx_done) && tx_pending) {
 226                 /* make sure it is true for two checks in a row */
 227                 return test_and_set_bit(__IXGBEVF_HANG_CHECK_ARMED,
 228                                         &tx_ring->state);
 229         }
 230         /* reset the countdown */
 231         clear_bit(__IXGBEVF_HANG_CHECK_ARMED, &tx_ring->state);
 232
 233         /* update completed stats and continue */
 234         tx_ring->tx_stats.tx_done_old = tx_done;
 235
 236         return false;
 237 }
 238
 239 static void ixgbevf_tx_timeout_reset(struct ixgbevf_adapter *adapter)
 240 {
 241         /* Do the reset outside of interrupt context */
 242         if (!test_bit(__IXGBEVF_DOWN, &adapter->state)) {
 243                 set_bit(__IXGBEVF_RESET_REQUESTED, &adapter->state);
 244                 ixgbevf_service_event_schedule(adapter);
 245         }
 246 }
 247
 248 /**
 249  * ixgbevf_tx_timeout - Respond to a Tx Hang
 250  * @netdev: network interface device structure
 251  **/
 252 static void ixgbevf_tx_timeout(struct net_device *netdev)
 253 {
 254         struct ixgbevf_adapter *adapter = netdev_priv(netdev);
 255
 256         ixgbevf_tx_timeout_reset(adapter);
 257 }
 258
 259 /**
 260  * ixgbevf_clean_tx_irq - Reclaim resources after transmit completes
 261  * @q_vector: board private structure
 262  * @tx_ring: tx ring to clean
 263  * @napi_budget: Used to determine if we are in netpoll
 264  **/
 265 static bool ixgbevf_clean_tx_irq(struct ixgbevf_q_vector *q_vector,
 266                                  struct ixgbevf_ring *tx_ring, int napi_budget)
 267 {
 268         struct ixgbevf_adapter *adapter = q_vector->adapter;
 269         struct ixgbevf_tx_buffer *tx_buffer;
 270         union ixgbe_adv_tx_desc *tx_desc;
 271         unsigned int total_bytes = 0, total_packets = 0;
 272         unsigned int budget = tx_ring->count / 2;
 273         unsigned int i = tx_ring->next_to_clean;
 274
 275         if (test_bit(__IXGBEVF_DOWN, &adapter->state))
 276                 return true;
 277
 278         tx_buffer = &tx_ring->tx_buffer_info[i];
 279         tx_desc = IXGBEVF_TX_DESC(tx_ring, i);
 280         i -= tx_ring->count;
 281
 282         do {
 283                 union ixgbe_adv_tx_desc *eop_desc = tx_buffer->next_to_watch;
 284
 285                 /* if next_to_watch is not set then there is no work pending */
 286                 if (!eop_desc)
 287                         break;
 288
 289                 /* prevent any other reads prior to eop_desc */
 290                 smp_rmb();
 291
 292                 /* if DD is not set pending work has not been completed */
 293                 if (!(eop_desc->wb.status & cpu_to_le32(IXGBE_TXD_STAT_DD)))
 294                         break;
 295
 296                 /* clear next_to_watch to prevent false hangs */
 297                 tx_buffer->next_to_watch = NULL;
 298
 299                 /* update the statistics for this packet */
 300                 total_bytes += tx_buffer->bytecount;
 301                 total_packets += tx_buffer->gso_segs;
 302
 303                 /* free the skb */
 304                 if (ring_is_xdp(tx_ring))
 305                         page_frag_free(tx_buffer->data);
 306                 else
 307                         napi_consume_skb(tx_buffer->skb, napi_budget);
 308
 309                 /* unmap skb header data */
 310                 dma_unmap_single(tx_ring->dev,
 311                                  dma_unmap_addr(tx_buffer, dma),
 312                                  dma_unmap_len(tx_buffer, len),
 313                                  DMA_TO_DEVICE);
 314
 315                 /* clear tx_buffer data */
 316                 dma_unmap_len_set(tx_buffer, len, 0);
 317
 318                 /* unmap remaining buffers */
 319                 while (tx_desc != eop_desc) {
 320                         tx_buffer++;
 321                         tx_desc++;
 322                         i++;
 323                         if (unlikely(!i)) {
 324                                 i -= tx_ring->count;
 325                                 tx_buffer = tx_ring->tx_buffer_info;
 326                                 tx_desc = IXGBEVF_TX_DESC(tx_ring, 0);
 327                         }
 328
 329                         /* unmap any remaining paged data */
 330                         if (dma_unmap_len(tx_buffer, len)) {
 331                                 dma_unmap_page(tx_ring->dev,
 332                                                dma_unmap_addr(tx_buffer, dma),
 333                                                dma_unmap_len(tx_buffer, len),
 334                                                DMA_TO_DEVICE);
 335                                 dma_unmap_len_set(tx_buffer, len, 0);
 336                         }
 337                 }
 338
 339                 /* move us one more past the eop_desc for start of next pkt */
 340                 tx_buffer++;
 341                 tx_desc++;
 342                 i++;
 343                 if (unlikely(!i)) {
 344                         i -= tx_ring->count;
 345                         tx_buffer = tx_ring->tx_buffer_info;
 346                         tx_desc = IXGBEVF_TX_DESC(tx_ring, 0);
 347                 }
 348
 349                 /* issue prefetch for next Tx descriptor */
 350                 prefetch(tx_desc);
 351
 352                 /* update budget accounting */
 353                 budget--;
 354         } while (likely(budget));
 355
 356         i += tx_ring->count;
 357         tx_ring->next_to_clean = i;
 358         u64_stats_update_begin(&tx_ring->syncp);
 359         tx_ring->stats.bytes += total_bytes;
 360         tx_ring->stats.packets += total_packets;
 361         u64_stats_update_end(&tx_ring->syncp);
 362         q_vector->tx.total_bytes += total_bytes;
 363         q_vector->tx.total_packets += total_packets;
 364
 365         if (check_for_tx_hang(tx_ring) && ixgbevf_check_tx_hang(tx_ring)) {
 366                 struct ixgbe_hw *hw = &adapter->hw;
 367                 union ixgbe_adv_tx_desc *eop_desc;
 368
 369                 eop_desc = tx_ring->tx_buffer_info[i].next_to_watch;
 370
 371                 pr_err("Detected Tx Unit Hang%s\n"
 372                        "  Tx Queue             <%d>\n"
 373                        "  TDH, TDT             <%x>, <%x>\n"
 374                        "  next_to_use          <%x>\n"
 375                        "  next_to_clean        <%x>\n"
 376                        "tx_buffer_info[next_to_clean]\n"
 377                        "  next_to_watch        <%p>\n"
 378                        "  eop_desc->wb.status  <%x>\n"
 379                        "  time_stamp           <%lx>\n"
 380                        "  jiffies              <%lx>\n",
 381                        ring_is_xdp(tx_ring) ? " XDP" : "",
 382                        tx_ring->queue_index,
 383                        IXGBE_READ_REG(hw, IXGBE_VFTDH(tx_ring->reg_idx)),
 384                        IXGBE_READ_REG(hw, IXGBE_VFTDT(tx_ring->reg_idx)),
 385                        tx_ring->next_to_use, i,
 386                        eop_desc, (eop_desc ? eop_desc->wb.status : 0),
 387                        tx_ring->tx_buffer_info[i].time_stamp, jiffies);
 388
 389                 if (!ring_is_xdp(tx_ring))
 390                         netif_stop_subqueue(tx_ring->netdev,
 391                                             tx_ring->queue_index);
 392
 393                 /* schedule immediate reset if we believe we hung */
 394                 ixgbevf_tx_timeout_reset(adapter);
 395
 396                 return true;
 397         }
 398
 399         if (ring_is_xdp(tx_ring))
 400                 return !!budget;
 401
 402 #define TX_WAKE_THRESHOLD (DESC_NEEDED * 2)
 403         if (unlikely(total_packets && netif_carrier_ok(tx_ring->netdev) &&
 404                      (ixgbevf_desc_unused(tx_ring) >= TX_WAKE_THRESHOLD))) {
 405                 /* Make sure that anybody stopping the queue after this
 406                  * sees the new next_to_clean.
 407                  */
 408                 smp_mb();
 409
 410                 if (__netif_subqueue_stopped(tx_ring->netdev,
 411                                              tx_ring->queue_index) &&
 412                     !test_bit(__IXGBEVF_DOWN, &adapter->state)) {
 413                         netif_wake_subqueue(tx_ring->netdev,
 414                                             tx_ring->queue_index);
 415                         ++tx_ring->tx_stats.restart_queue;
 416                 }
 417         }
 418
 419         return !!budget;
 420 }
 421
 422 /**
 423  * ixgbevf_rx_skb - Helper function to determine proper Rx method
 424  * @q_vector: structure containing interrupt and ring information
 425  * @skb: packet to send up
 426  **/
 427 static void ixgbevf_rx_skb(struct ixgbevf_q_vector *q_vector,
 428                            struct sk_buff *skb)
 429 {
 430         napi_gro_receive(&q_vector->napi, skb);
 431 }
 432
 433 #define IXGBE_RSS_L4_TYPES_MASK \
 434         ((1ul << IXGBE_RXDADV_RSSTYPE_IPV4_TCP) | \
 435          (1ul << IXGBE_RXDADV_RSSTYPE_IPV4_UDP) | \
 436          (1ul << IXGBE_RXDADV_RSSTYPE_IPV6_TCP) | \
 437          (1ul << IXGBE_RXDADV_RSSTYPE_IPV6_UDP))
 438
 439 static inline void ixgbevf_rx_hash(struct ixgbevf_ring *ring,
 440                                    union ixgbe_adv_rx_desc *rx_desc,
 441                                    struct sk_buff *skb)
 442 {
 443         u16 rss_type;
 444
 445         if (!(ring->netdev->features & NETIF_F_RXHASH))
 446                 return;
 447
 448         rss_type = le16_to_cpu(rx_desc->wb.lower.lo_dword.hs_rss.pkt_info) &
 449                    IXGBE_RXDADV_RSSTYPE_MASK;
 450
 451         if (!rss_type)
 452                 return;
 453
 454         skb_set_hash(skb, le32_to_cpu(rx_desc->wb.lower.hi_dword.rss),
 455                      (IXGBE_RSS_L4_TYPES_MASK & (1ul << rss_type)) ?
 456                      PKT_HASH_TYPE_L4 : PKT_HASH_TYPE_L3);
 457 }
 458
 459 /**
 460  * ixgbevf_rx_checksum - indicate in skb if hw indicated a good cksum
 461  * @ring: structure containig ring specific data
 462  * @rx_desc: current Rx descriptor being processed
 463  * @skb: skb currently being received and modified
 464  **/
 465 static inline void ixgbevf_rx_checksum(struct ixgbevf_ring *ring,
 466                                        union ixgbe_adv_rx_desc *rx_desc,
 467                                        struct sk_buff *skb)
 468 {
 469         skb_checksum_none_assert(skb);
 470
 471         /* Rx csum disabled */
 472         if (!(ring->netdev->features & NETIF_F_RXCSUM))
 473                 return;
 474
 475         /* if IP and error */
 476         if (ixgbevf_test_staterr(rx_desc, IXGBE_RXD_STAT_IPCS) &&
 477             ixgbevf_test_staterr(rx_desc, IXGBE_RXDADV_ERR_IPE)) {
 478                 ring->rx_stats.csum_err++;
 479                 return;
 480         }
 481
 482         if (!ixgbevf_test_staterr(rx_desc, IXGBE_RXD_STAT_L4CS))
 483                 return;
 484
 485         if (ixgbevf_test_staterr(rx_desc, IXGBE_RXDADV_ERR_TCPE)) {
 486                 ring->rx_stats.csum_err++;
 487                 return;
 488         }
 489
 490         /* It must be a TCP or UDP packet with a valid checksum */
 491         skb->ip_summed = CHECKSUM_UNNECESSARY;
 492 }
 493
 494 /**
 495  * ixgbevf_process_skb_fields - Populate skb header fields from Rx descriptor
 496  * @rx_ring: rx descriptor ring packet is being transacted on
 497  * @rx_desc: pointer to the EOP Rx descriptor
 498  * @skb: pointer to current skb being populated
 499  *
 500  * This function checks the ring, descriptor, and packet information in
 501  * order to populate the checksum, VLAN, protocol, and other fields within
 502  * the skb.
 503  **/
 504 static void ixgbevf_process_skb_fields(struct ixgbevf_ring *rx_ring,
 505                                        union ixgbe_adv_rx_desc *rx_desc,
 506                                        struct sk_buff *skb)
 507 {
 508         ixgbevf_rx_hash(rx_ring, rx_desc, skb);
 509         ixgbevf_rx_checksum(rx_ring, rx_desc, skb);
 510
 511         if (ixgbevf_test_staterr(rx_desc, IXGBE_RXD_STAT_VP)) {
 512                 u16 vid = le16_to_cpu(rx_desc->wb.upper.vlan);
 513                 unsigned long *active_vlans = netdev_priv(rx_ring->netdev);
 514
 515                 if (test_bit(vid & VLAN_VID_MASK, active_vlans))
 516                         __vlan_hwaccel_put_tag(skb, htons(ETH_P_8021Q), vid);
 517         }
 518
 519         skb->protocol = eth_type_trans(skb, rx_ring->netdev);
 520 }
 521
 522 static
 523 struct ixgbevf_rx_buffer *ixgbevf_get_rx_buffer(struct ixgbevf_ring *rx_ring,
 524                                                 const unsigned int size)
 525 {
 526         struct ixgbevf_rx_buffer *rx_buffer;
 527
 528         rx_buffer = &rx_ring->rx_buffer_info[rx_ring->next_to_clean];
 529         prefetchw(rx_buffer->page);
 530
 531         /* we are reusing so sync this buffer for CPU use */
 532         dma_sync_single_range_for_cpu(rx_ring->dev,
 533                                       rx_buffer->dma,
 534                                       rx_buffer->page_offset,
 535                                       size,
 536                                       DMA_FROM_DEVICE);
 537
 538         rx_buffer->pagecnt_bias--;
 539
 540         return rx_buffer;
 541 }
 542
 543 static void ixgbevf_put_rx_buffer(struct ixgbevf_ring *rx_ring,
 544                                   struct ixgbevf_rx_buffer *rx_buffer,
 545                                   struct sk_buff *skb)
 546 {
 547         if (ixgbevf_can_reuse_rx_page(rx_buffer)) {
 548                 /* hand second half of page back to the ring */
 549                 ixgbevf_reuse_rx_page(rx_ring, rx_buffer);
 550         } else {
 551                 if (IS_ERR(skb))
 552                         /* We are not reusing the buffer so unmap it and free
 553                          * any references we are holding to it
 554                          */
 555                         dma_unmap_page_attrs(rx_ring->dev, rx_buffer->dma,
 556                                              ixgbevf_rx_pg_size(rx_ring),
 557                                              DMA_FROM_DEVICE,
 558                                              IXGBEVF_RX_DMA_ATTR);
 559                 __page_frag_cache_drain(rx_buffer->page,
 560                                         rx_buffer->pagecnt_bias);
 561         }
 562
 563         /* clear contents of rx_buffer */
 564         rx_buffer->page = NULL;
 565 }
 566
 567 /**
 568  * ixgbevf_is_non_eop - process handling of non-EOP buffers
 569  * @rx_ring: Rx ring being processed
 570  * @rx_desc: Rx descriptor for current buffer
 571  *
 572  * This function updates next to clean.  If the buffer is an EOP buffer
 573  * this function exits returning false, otherwise it will place the
 574  * sk_buff in the next buffer to be chained and return true indicating
 575  * that this is in fact a non-EOP buffer.
 576  **/
 577 static bool ixgbevf_is_non_eop(struct ixgbevf_ring *rx_ring,
 578                                union ixgbe_adv_rx_desc *rx_desc)
 579 {
 580         u32 ntc = rx_ring->next_to_clean + 1;
 581
 582         /* fetch, update, and store next to clean */
 583         ntc = (ntc < rx_ring->count) ? ntc : 0;
 584         rx_ring->next_to_clean = ntc;
 585
 586         prefetch(IXGBEVF_RX_DESC(rx_ring, ntc));
 587
 588         if (likely(ixgbevf_test_staterr(rx_desc, IXGBE_RXD_STAT_EOP)))
 589                 return false;
 590
 591         return true;
 592 }
 593
 594 static inline unsigned int ixgbevf_rx_offset(struct ixgbevf_ring *rx_ring)
 595 {
 596         return ring_uses_build_skb(rx_ring) ? IXGBEVF_SKB_PAD : 0;
 597 }
 598
 599 static bool ixgbevf_alloc_mapped_page(struct ixgbevf_ring *rx_ring,
 600                                       struct ixgbevf_rx_buffer *bi)
 601 {
 602         struct page *page = bi->page;
 603         dma_addr_t dma;
 604
 605         /* since we are recycling buffers we should seldom need to alloc */
 606         if (likely(page))
 607                 return true;
 608
 609         /* alloc new page for storage */
 610         page = dev_alloc_pages(ixgbevf_rx_pg_order(rx_ring));
 611         if (unlikely(!page)) {
 612                 rx_ring->rx_stats.alloc_rx_page_failed++;
 613                 return false;
 614         }
 615
 616         /* map page for use */
 617         dma = dma_map_page_attrs(rx_ring->dev, page, 0,
 618                                  ixgbevf_rx_pg_size(rx_ring),
 619                                  DMA_FROM_DEVICE, IXGBEVF_RX_DMA_ATTR);
 620
 621         /* if mapping failed free memory back to system since
 622          * there isn't much point in holding memory we can't use
 623          */
 624         if (dma_mapping_error(rx_ring->dev, dma)) {
 625                 __free_pages(page, ixgbevf_rx_pg_order(rx_ring));
 626
 627                 rx_ring->rx_stats.alloc_rx_page_failed++;
 628                 return false;
 629         }
 630
 631         bi->dma = dma;
 632         bi->page = page;
 633         bi->page_offset = ixgbevf_rx_offset(rx_ring);
 634         bi->pagecnt_bias = 1;
 635         rx_ring->rx_stats.alloc_rx_page++;
 636
 637         return true;
 638 }
 639
 640 /**
 641  * ixgbevf_alloc_rx_buffers - Replace used receive buffers; packet split
 642  * @rx_ring: rx descriptor ring (for a specific queue) to setup buffers on
 643  * @cleaned_count: number of buffers to replace
 644  **/
 645 static void ixgbevf_alloc_rx_buffers(struct ixgbevf_ring *rx_ring,
 646                                      u16 cleaned_count)
 647 {
 648         union ixgbe_adv_rx_desc *rx_desc;
 649         struct ixgbevf_rx_buffer *bi;
 650         unsigned int i = rx_ring->next_to_use;
 651
 652         /* nothing to do or no valid netdev defined */
 653         if (!cleaned_count || !rx_ring->netdev)
 654                 return;
 655
 656         rx_desc = IXGBEVF_RX_DESC(rx_ring, i);
 657         bi = &rx_ring->rx_buffer_info[i];
 658         i -= rx_ring->count;
 659
 660         do {
 661                 if (!ixgbevf_alloc_mapped_page(rx_ring, bi))
 662                         break;
 663
 664                 /* sync the buffer for use by the device */
 665                 dma_sync_single_range_for_device(rx_ring->dev, bi->dma,
 666                                                  bi->page_offset,
 667                                                  ixgbevf_rx_bufsz(rx_ring),
 668                                                  DMA_FROM_DEVICE);
 669
 670                 /* Refresh the desc even if pkt_addr didn't change
 671                  * because each write-back erases this info.
 672                  */
 673                 rx_desc->read.pkt_addr = cpu_to_le64(bi->dma + bi->page_offset);
 674
 675                 rx_desc++;
 676                 bi++;
 677                 i++;
 678                 if (unlikely(!i)) {
 679                         rx_desc = IXGBEVF_RX_DESC(rx_ring, 0);
 680                         bi = rx_ring->rx_buffer_info;
 681                         i -= rx_ring->count;
 682                 }
 683
 684                 /* clear the length for the next_to_use descriptor */
 685                 rx_desc->wb.upper.length = 0;
 686
 687                 cleaned_count--;
 688         } while (cleaned_count);
 689
 690         i += rx_ring->count;
 691
 692         if (rx_ring->next_to_use != i) {
 693                 /* record the next descriptor to use */
 694                 rx_ring->next_to_use = i;
 695
 696                 /* update next to alloc since we have filled the ring */
 697                 rx_ring->next_to_alloc = i;
 698
 699                 /* Force memory writes to complete before letting h/w
 700                  * know there are new descriptors to fetch.  (Only
 701                  * applicable for weak-ordered memory model archs,
 702                  * such as IA-64).
 703                  */
 704                 wmb();
 705                 ixgbevf_write_tail(rx_ring, i);
 706         }
 707 }
 708
 709 /**
 710  * ixgbevf_cleanup_headers - Correct corrupted or empty headers
 711  * @rx_ring: rx descriptor ring packet is being transacted on
 712  * @rx_desc: pointer to the EOP Rx descriptor
 713  * @skb: pointer to current skb being fixed
 714  *
 715  * Check for corrupted packet headers caused by senders on the local L2
 716  * embedded NIC switch not setting up their Tx Descriptors right.  These
 717  * should be very rare.
 718  *
 719  * Also address the case where we are pulling data in on pages only
 720  * and as such no data is present in the skb header.
 721  *
 722  * In addition if skb is not at least 60 bytes we need to pad it so that
 723  * it is large enough to qualify as a valid Ethernet frame.
 724  *
 725  * Returns true if an error was encountered and skb was freed.
 726  **/
 727 static bool ixgbevf_cleanup_headers(struct ixgbevf_ring *rx_ring,
 728                                     union ixgbe_adv_rx_desc *rx_desc,
 729                                     struct sk_buff *skb)
 730 {
 731         /* XDP packets use error pointer so abort at this point */
 732         if (IS_ERR(skb))
 733                 return true;
 734
 735         /* verify that the packet does not have any known errors */
 736         if (unlikely(ixgbevf_test_staterr(rx_desc,
 737                                           IXGBE_RXDADV_ERR_FRAME_ERR_MASK))) {
 738                 struct net_device *netdev = rx_ring->netdev;
 739
 740                 if (!(netdev->features & NETIF_F_RXALL)) {
 741                         dev_kfree_skb_any(skb);
 742                         return true;
 743                 }
 744         }
 745
 746         /* if eth_skb_pad returns an error the skb was freed */
 747         if (eth_skb_pad(skb))
 748                 return true;
 749
 750         return false;
 751 }
 752
 753 /**
 754  * ixgbevf_reuse_rx_page - page flip buffer and store it back on the ring
 755  * @rx_ring: rx descriptor ring to store buffers on
 756  * @old_buff: donor buffer to have page reused
 757  *
 758  * Synchronizes page for reuse by the adapter
 759  **/
 760 static void ixgbevf_reuse_rx_page(struct ixgbevf_ring *rx_ring,
 761                                   struct ixgbevf_rx_buffer *old_buff)
 762 {
 763         struct ixgbevf_rx_buffer *new_buff;
 764         u16 nta = rx_ring->next_to_alloc;
 765
 766         new_buff = &rx_ring->rx_buffer_info[nta];
 767
 768         /* update, and store next to alloc */
 769         nta++;
 770         rx_ring->next_to_alloc = (nta < rx_ring->count) ? nta : 0;
 771
 772         /* transfer page from old buffer to new buffer */
 773         new_buff->page = old_buff->page;
 774         new_buff->dma = old_buff->dma;
 775         new_buff->page_offset = old_buff->page_offset;
 776         new_buff->pagecnt_bias = old_buff->pagecnt_bias;
 777 }
 778
 779 static inline bool ixgbevf_page_is_reserved(struct page *page)
 780 {
 781         return (page_to_nid(page) != numa_mem_id()) || page_is_pfmemalloc(page);
 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 pages */
 790         if (unlikely(ixgbevf_page_is_reserved(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         prefetch(xdp->data);
 866 #if L1_CACHE_BYTES < 128
 867         prefetch(xdp->data + L1_CACHE_BYTES);
 868 #endif
 869         /* Note, we get here by enabling legacy-rx via:
 870          *
 871          *    ethtool --set-priv-flags <dev> legacy-rx on
 872          *
 873          * In this mode, we currently get 0 extra XDP headroom as
 874          * opposed to having legacy-rx off, where we process XDP
 875          * packets going to stack via ixgbevf_build_skb().
 876          *
 877          * For ixgbevf_construct_skb() mode it means that the
 878          * xdp->data_meta will always point to xdp->data, since
 879          * the helper cannot expand the head. Should this ever
 880          * changed in future for legacy-rx mode on, then lets also
 881          * add xdp->data_meta handling here.
 882          */
 883
 884         /* allocate a skb to store the frags */
 885         skb = napi_alloc_skb(&rx_ring->q_vector->napi, IXGBEVF_RX_HDR_SIZE);
 886         if (unlikely(!skb))
 887                 return NULL;
 888
 889         /* Determine available headroom for copy */
 890         headlen = size;
 891         if (headlen > IXGBEVF_RX_HDR_SIZE)
 892                 headlen = eth_get_headlen(xdp->data, IXGBEVF_RX_HDR_SIZE);
 893
 894         /* align pull length to size of long to optimize memcpy performance */
 895         memcpy(__skb_put(skb, headlen), xdp->data,
 896                ALIGN(headlen, sizeof(long)));
 897
 898         /* update all of the pointers */
 899         size -= headlen;
 900         if (size) {
 901                 skb_add_rx_frag(skb, 0, rx_buffer->page,
 902                                 (xdp->data + headlen) -
 903                                         page_address(rx_buffer->page),
 904                                 size, truesize);
 905 #if (PAGE_SIZE < 8192)
 906                 rx_buffer->page_offset ^= truesize;
 907 #else
 908                 rx_buffer->page_offset += truesize;
 909 #endif
 910         } else {
 911                 rx_buffer->pagecnt_bias++;
 912         }
 913
 914         return skb;
 915 }
 916
 917 static inline void ixgbevf_irq_enable_queues(struct ixgbevf_adapter *adapter,
 918                                              u32 qmask)
 919 {
 920         struct ixgbe_hw *hw = &adapter->hw;
 921
 922         IXGBE_WRITE_REG(hw, IXGBE_VTEIMS, qmask);
 923 }
 924
 925 static struct sk_buff *ixgbevf_build_skb(struct ixgbevf_ring *rx_ring,
 926                                          struct ixgbevf_rx_buffer *rx_buffer,
 927                                          struct xdp_buff *xdp,
 928                                          union ixgbe_adv_rx_desc *rx_desc)
 929 {
 930         unsigned int metasize = xdp->data - xdp->data_meta;
 931 #if (PAGE_SIZE < 8192)
 932         unsigned int truesize = ixgbevf_rx_pg_size(rx_ring) / 2;
 933 #else
 934         unsigned int truesize = SKB_DATA_ALIGN(sizeof(struct skb_shared_info)) +
 935                                 SKB_DATA_ALIGN(xdp->data_end -
 936                                                xdp->data_hard_start);
 937 #endif
 938         struct sk_buff *skb;
 939
 940         /* Prefetch first cache line of first page. If xdp->data_meta
 941          * is unused, this points to xdp->data, otherwise, we likely
 942          * have a consumer accessing first few bytes of meta data,
 943          * and then actual data.
 944          */
 945         prefetch(xdp->data_meta);
 946 #if L1_CACHE_BYTES < 128
 947         prefetch(xdp->data_meta + L1_CACHE_BYTES);
 948 #endif
 949
 950         /* build an skb around the page buffer */
 951         skb = build_skb(xdp->data_hard_start, truesize);
 952         if (unlikely(!skb))
 953                 return NULL;
 954
 955         /* update pointers within the skb to store the data */
 956         skb_reserve(skb, xdp->data - xdp->data_hard_start);
 957         __skb_put(skb, xdp->data_end - xdp->data);
 958         if (metasize)
 959                 skb_metadata_set(skb, metasize);
 960
 961         /* update buffer offset */
 962 #if (PAGE_SIZE < 8192)
 963         rx_buffer->page_offset ^= truesize;
 964 #else
 965         rx_buffer->page_offset += truesize;
 966 #endif
 967
 968         return skb;
 969 }
 970
 971 #define IXGBEVF_XDP_PASS 0
 972 #define IXGBEVF_XDP_CONSUMED 1
 973 #define IXGBEVF_XDP_TX 2
 974
 975 static int ixgbevf_xmit_xdp_ring(struct ixgbevf_ring *ring,
 976                                  struct xdp_buff *xdp)
 977 {
 978         struct ixgbevf_tx_buffer *tx_buffer;
 979         union ixgbe_adv_tx_desc *tx_desc;
 980         u32 len, cmd_type;
 981         dma_addr_t dma;
 982         u16 i;
 983
 984         len = xdp->data_end - xdp->data;
 985
 986         if (unlikely(!ixgbevf_desc_unused(ring)))
 987                 return IXGBEVF_XDP_CONSUMED;
 988
 989         dma = dma_map_single(ring->dev, xdp->data, len, DMA_TO_DEVICE);
 990         if (dma_mapping_error(ring->dev, dma))
 991                 return IXGBEVF_XDP_CONSUMED;
 992
 993         /* record the location of the first descriptor for this packet */
 994         i = ring->next_to_use;
 995         tx_buffer = &ring->tx_buffer_info[i];
 996
 997         dma_unmap_len_set(tx_buffer, len, len);
 998         dma_unmap_addr_set(tx_buffer, dma, dma);
 999         tx_buffer->data = xdp->data;
1000         tx_buffer->bytecount = len;
1001         tx_buffer->gso_segs = 1;
1002         tx_buffer->protocol = 0;
1003
1004         /* Populate minimal context descriptor that will provide for the
1005          * fact that we are expected to process Ethernet frames.
1006          */
1007         if (!test_bit(__IXGBEVF_TX_XDP_RING_PRIMED, &ring->state)) {
1008                 struct ixgbe_adv_tx_context_desc *context_desc;
1009
1010                 set_bit(__IXGBEVF_TX_XDP_RING_PRIMED, &ring->state);
1011
1012                 context_desc = IXGBEVF_TX_CTXTDESC(ring, 0);
1013                 context_desc->vlan_macip_lens   =
1014                         cpu_to_le32(ETH_HLEN << IXGBE_ADVTXD_MACLEN_SHIFT);
1015                 context_desc->seqnum_seed       = 0;
1016                 context_desc->type_tucmd_mlhl   =
1017                         cpu_to_le32(IXGBE_TXD_CMD_DEXT |
1018                                     IXGBE_ADVTXD_DTYP_CTXT);
1019                 context_desc->mss_l4len_idx     = 0;
1020
1021                 i = 1;
1022         }
1023
1024         /* put descriptor type bits */
1025         cmd_type = IXGBE_ADVTXD_DTYP_DATA |
1026                    IXGBE_ADVTXD_DCMD_DEXT |
1027                    IXGBE_ADVTXD_DCMD_IFCS;
1028         cmd_type |= len | IXGBE_TXD_CMD;
1029
1030         tx_desc = IXGBEVF_TX_DESC(ring, i);
1031         tx_desc->read.buffer_addr = cpu_to_le64(dma);
1032
1033         tx_desc->read.cmd_type_len = cpu_to_le32(cmd_type);
1034         tx_desc->read.olinfo_status =
1035                         cpu_to_le32((len << IXGBE_ADVTXD_PAYLEN_SHIFT) |
1036                                     IXGBE_ADVTXD_CC);
1037
1038         /* Avoid any potential race with cleanup */
1039         smp_wmb();
1040
1041         /* set next_to_watch value indicating a packet is present */
1042         i++;
1043         if (i == ring->count)
1044                 i = 0;
1045
1046         tx_buffer->next_to_watch = tx_desc;
1047         ring->next_to_use = i;
1048
1049         return IXGBEVF_XDP_TX;
1050 }
1051
1052 static struct sk_buff *ixgbevf_run_xdp(struct ixgbevf_adapter *adapter,
1053                                        struct ixgbevf_ring  *rx_ring,
1054                                        struct xdp_buff *xdp)
1055 {
1056         int result = IXGBEVF_XDP_PASS;
1057         struct ixgbevf_ring *xdp_ring;
1058         struct bpf_prog *xdp_prog;
1059         u32 act;
1060
1061         rcu_read_lock();
1062         xdp_prog = READ_ONCE(rx_ring->xdp_prog);
1063
1064         if (!xdp_prog)
1065                 goto xdp_out;
1066
1067         act = bpf_prog_run_xdp(xdp_prog, xdp);
1068         switch (act) {
1069         case XDP_PASS:
1070                 break;
1071         case XDP_TX:
1072                 xdp_ring = adapter->xdp_ring[rx_ring->queue_index];
1073                 result = ixgbevf_xmit_xdp_ring(xdp_ring, xdp);
1074                 if (result == IXGBEVF_XDP_CONSUMED)
1075                         goto out_failure;
1076                 break;
1077         default:
1078                 bpf_warn_invalid_xdp_action(act);
1079                 /* fallthrough */
1080         case XDP_ABORTED:
1081 out_failure:
1082                 trace_xdp_exception(rx_ring->netdev, xdp_prog, act);
1083                 /* fallthrough -- handle aborts by dropping packet */
1084         case XDP_DROP:
1085                 result = IXGBEVF_XDP_CONSUMED;
1086                 break;
1087         }
1088 xdp_out:
1089         rcu_read_unlock();
1090         return ERR_PTR(-result);
1091 }
1092
1093 static void ixgbevf_rx_buffer_flip(struct ixgbevf_ring *rx_ring,
1094                                    struct ixgbevf_rx_buffer *rx_buffer,
1095                                    unsigned int size)
1096 {
1097 #if (PAGE_SIZE < 8192)
1098         unsigned int truesize = ixgbevf_rx_pg_size(rx_ring) / 2;
1099
1100         rx_buffer->page_offset ^= truesize;
1101 #else
1102         unsigned int truesize = ring_uses_build_skb(rx_ring) ?
1103                                 SKB_DATA_ALIGN(IXGBEVF_SKB_PAD + size) :
1104                                 SKB_DATA_ALIGN(size);
1105
1106         rx_buffer->page_offset += truesize;
1107 #endif
1108 }
1109
1110 static int ixgbevf_clean_rx_irq(struct ixgbevf_q_vector *q_vector,
1111                                 struct ixgbevf_ring *rx_ring,
1112                                 int budget)
1113 {
1114         unsigned int total_rx_bytes = 0, total_rx_packets = 0;
1115         struct ixgbevf_adapter *adapter = q_vector->adapter;
1116         u16 cleaned_count = ixgbevf_desc_unused(rx_ring);
1117         struct sk_buff *skb = rx_ring->skb;
1118         bool xdp_xmit = false;
1119         struct xdp_buff xdp;
1120
1121         xdp.rxq = &rx_ring->xdp_rxq;
1122
1123         while (likely(total_rx_packets < budget)) {
1124                 struct ixgbevf_rx_buffer *rx_buffer;
1125                 union ixgbe_adv_rx_desc *rx_desc;
1126                 unsigned int size;
1127
1128                 /* return some buffers to hardware, one at a time is too slow */
1129                 if (cleaned_count >= IXGBEVF_RX_BUFFER_WRITE) {
1130                         ixgbevf_alloc_rx_buffers(rx_ring, cleaned_count);
1131                         cleaned_count = 0;
1132                 }
1133
1134                 rx_desc = IXGBEVF_RX_DESC(rx_ring, rx_ring->next_to_clean);
1135                 size = le16_to_cpu(rx_desc->wb.upper.length);
1136                 if (!size)
1137                         break;
1138
1139                 /* This memory barrier is needed to keep us from reading
1140                  * any other fields out of the rx_desc until we know the
1141                  * RXD_STAT_DD bit is set
1142                  */
1143                 rmb();
1144
1145                 rx_buffer = ixgbevf_get_rx_buffer(rx_ring, size);
1146
1147                 /* retrieve a buffer from the ring */
1148                 if (!skb) {
1149                         xdp.data = page_address(rx_buffer->page) +
1150                                    rx_buffer->page_offset;
1151                         xdp.data_meta = xdp.data;
1152                         xdp.data_hard_start = xdp.data -
1153                                               ixgbevf_rx_offset(rx_ring);
1154                         xdp.data_end = xdp.data + size;
1155
1156                         skb = ixgbevf_run_xdp(adapter, rx_ring, &xdp);
1157                 }
1158
1159                 if (IS_ERR(skb)) {
1160                         if (PTR_ERR(skb) == -IXGBEVF_XDP_TX) {
1161                                 xdp_xmit = true;
1162                                 ixgbevf_rx_buffer_flip(rx_ring, rx_buffer,
1163                                                        size);
1164                         } else {
1165                                 rx_buffer->pagecnt_bias++;
1166                         }
1167                         total_rx_packets++;
1168                         total_rx_bytes += size;
1169                 } else if (skb) {
1170                         ixgbevf_add_rx_frag(rx_ring, rx_buffer, skb, size);
1171                 } else if (ring_uses_build_skb(rx_ring)) {
1172                         skb = ixgbevf_build_skb(rx_ring, rx_buffer,
1173                                                 &xdp, rx_desc);
1174                 } else {
1175                         skb = ixgbevf_construct_skb(rx_ring, rx_buffer,
1176                                                     &xdp, rx_desc);
1177                 }
1178
1179                 /* exit if we failed to retrieve a buffer */
1180                 if (!skb) {
1181                         rx_ring->rx_stats.alloc_rx_buff_failed++;
1182                         rx_buffer->pagecnt_bias++;
1183                         break;
1184                 }
1185
1186                 ixgbevf_put_rx_buffer(rx_ring, rx_buffer, skb);
1187                 cleaned_count++;
1188
1189                 /* fetch next buffer in frame if non-eop */
1190                 if (ixgbevf_is_non_eop(rx_ring, rx_desc))
1191                         continue;
1192
1193                 /* verify the packet layout is correct */
1194                 if (ixgbevf_cleanup_headers(rx_ring, rx_desc, skb)) {
1195                         skb = NULL;
1196                         continue;
1197                 }
1198
1199                 /* probably a little skewed due to removing CRC */
1200                 total_rx_bytes += skb->len;
1201
1202                 /* Workaround hardware that can't do proper VEPA multicast
1203                  * source pruning.
1204                  */
1205                 if ((skb->pkt_type == PACKET_BROADCAST ||
1206                      skb->pkt_type == PACKET_MULTICAST) &&
1207                     ether_addr_equal(rx_ring->netdev->dev_addr,
1208                                      eth_hdr(skb)->h_source)) {
1209                         dev_kfree_skb_irq(skb);
1210                         continue;
1211                 }
1212
1213                 /* populate checksum, VLAN, and protocol */
1214                 ixgbevf_process_skb_fields(rx_ring, rx_desc, skb);
1215
1216                 ixgbevf_rx_skb(q_vector, skb);
1217
1218                 /* reset skb pointer */
1219                 skb = NULL;
1220
1221                 /* update budget accounting */
1222                 total_rx_packets++;
1223         }
1224
1225         /* place incomplete frames back on ring for completion */
1226         rx_ring->skb = skb;
1227
1228         if (xdp_xmit) {
1229                 struct ixgbevf_ring *xdp_ring =
1230                         adapter->xdp_ring[rx_ring->queue_index];
1231
1232                 /* Force memory writes to complete before letting h/w
1233                  * know there are new descriptors to fetch.
1234                  */
1235                 wmb();
1236                 ixgbevf_write_tail(xdp_ring, xdp_ring->next_to_use);
1237         }
1238
1239         u64_stats_update_begin(&rx_ring->syncp);
1240         rx_ring->stats.packets += total_rx_packets;
1241         rx_ring->stats.bytes += total_rx_bytes;
1242         u64_stats_update_end(&rx_ring->syncp);
1243         q_vector->rx.total_packets += total_rx_packets;
1244         q_vector->rx.total_bytes += total_rx_bytes;
1245
1246         return total_rx_packets;
1247 }
1248
1249 /**
1250  * ixgbevf_poll - NAPI polling calback
1251  * @napi: napi struct with our devices info in it
1252  * @budget: amount of work driver is allowed to do this pass, in packets
1253  *
1254  * This function will clean more than one or more rings associated with a
1255  * q_vector.
1256  **/
1257 static int ixgbevf_poll(struct napi_struct *napi, int budget)
1258 {
1259         struct ixgbevf_q_vector *q_vector =
1260                 container_of(napi, struct ixgbevf_q_vector, napi);
1261         struct ixgbevf_adapter *adapter = q_vector->adapter;
1262         struct ixgbevf_ring *ring;
1263         int per_ring_budget, work_done = 0;
1264         bool clean_complete = true;
1265
1266         ixgbevf_for_each_ring(ring, q_vector->tx) {
1267                 if (!ixgbevf_clean_tx_irq(q_vector, ring, budget))
1268                         clean_complete = false;
1269         }
1270
1271         if (budget <= 0)
1272                 return budget;
1273
1274         /* attempt to distribute budget to each queue fairly, but don't allow
1275          * the budget to go below 1 because we'll exit polling
1276          */
1277         if (q_vector->rx.count > 1)
1278                 per_ring_budget = max(budget/q_vector->rx.count, 1);
1279         else
1280                 per_ring_budget = budget;
1281
1282         ixgbevf_for_each_ring(ring, q_vector->rx) {
1283                 int cleaned = ixgbevf_clean_rx_irq(q_vector, ring,
1284                                                    per_ring_budget);
1285                 work_done += cleaned;
1286                 if (cleaned >= per_ring_budget)
1287                         clean_complete = false;
1288         }
1289
1290         /* If all work not completed, return budget and keep polling */
1291         if (!clean_complete)
1292                 return budget;
1293         /* all work done, exit the polling mode */
1294         napi_complete_done(napi, work_done);
1295         if (adapter->rx_itr_setting == 1)
1296                 ixgbevf_set_itr(q_vector);
1297         if (!test_bit(__IXGBEVF_DOWN, &adapter->state) &&
1298             !test_bit(__IXGBEVF_REMOVING, &adapter->state))
1299                 ixgbevf_irq_enable_queues(adapter,
1300                                           BIT(q_vector->v_idx));
1301
1302         return 0;
1303 }
1304
1305 /**
1306  * ixgbevf_write_eitr - write VTEITR register in hardware specific way
1307  * @q_vector: structure containing interrupt and ring information
1308  **/
1309 void ixgbevf_write_eitr(struct ixgbevf_q_vector *q_vector)
1310 {
1311         struct ixgbevf_adapter *adapter = q_vector->adapter;
1312         struct ixgbe_hw *hw = &adapter->hw;
1313         int v_idx = q_vector->v_idx;
1314         u32 itr_reg = q_vector->itr & IXGBE_MAX_EITR;
1315
1316         /* set the WDIS bit to not clear the timer bits and cause an
1317          * immediate assertion of the interrupt
1318          */
1319         itr_reg |= IXGBE_EITR_CNT_WDIS;
1320
1321         IXGBE_WRITE_REG(hw, IXGBE_VTEITR(v_idx), itr_reg);
1322 }
1323
1324 /**
1325  * ixgbevf_configure_msix - Configure MSI-X hardware
1326  * @adapter: board private structure
1327  *
1328  * ixgbevf_configure_msix sets up the hardware to properly generate MSI-X
1329  * interrupts.
1330  **/
1331 static void ixgbevf_configure_msix(struct ixgbevf_adapter *adapter)
1332 {
1333         struct ixgbevf_q_vector *q_vector;
1334         int q_vectors, v_idx;
1335
1336         q_vectors = adapter->num_msix_vectors - NON_Q_VECTORS;
1337         adapter->eims_enable_mask = 0;
1338
1339         /* Populate the IVAR table and set the ITR values to the
1340          * corresponding register.
1341          */
1342         for (v_idx = 0; v_idx < q_vectors; v_idx++) {
1343                 struct ixgbevf_ring *ring;
1344
1345                 q_vector = adapter->q_vector[v_idx];
1346
1347                 ixgbevf_for_each_ring(ring, q_vector->rx)
1348                         ixgbevf_set_ivar(adapter, 0, ring->reg_idx, v_idx);
1349
1350                 ixgbevf_for_each_ring(ring, q_vector->tx)
1351                         ixgbevf_set_ivar(adapter, 1, ring->reg_idx, v_idx);
1352
1353                 if (q_vector->tx.ring && !q_vector->rx.ring) {
1354                         /* Tx only vector */
1355                         if (adapter->tx_itr_setting == 1)
1356                                 q_vector->itr = IXGBE_12K_ITR;
1357                         else
1358                                 q_vector->itr = adapter->tx_itr_setting;
1359                 } else {
1360                         /* Rx or Rx/Tx vector */
1361                         if (adapter->rx_itr_setting == 1)
1362                                 q_vector->itr = IXGBE_20K_ITR;
1363                         else
1364                                 q_vector->itr = adapter->rx_itr_setting;
1365                 }
1366
1367                 /* add q_vector eims value to global eims_enable_mask */
1368                 adapter->eims_enable_mask |= BIT(v_idx);
1369
1370                 ixgbevf_write_eitr(q_vector);
1371         }
1372
1373         ixgbevf_set_ivar(adapter, -1, 1, v_idx);
1374         /* setup eims_other and add value to global eims_enable_mask */
1375         adapter->eims_other = BIT(v_idx);
1376         adapter->eims_enable_mask |= adapter->eims_other;
1377 }
1378
1379 enum latency_range {
1380         lowest_latency = 0,
1381         low_latency = 1,
1382         bulk_latency = 2,
1383         latency_invalid = 255
1384 };
1385
1386 /**
1387  * ixgbevf_update_itr - update the dynamic ITR value based on statistics
1388  * @q_vector: structure containing interrupt and ring information
1389  * @ring_container: structure containing ring performance data
1390  *
1391  * Stores a new ITR value based on packets and byte
1392  * counts during the last interrupt.  The advantage of per interrupt
1393  * computation is faster updates and more accurate ITR for the current
1394  * traffic pattern.  Constants in this function were computed
1395  * based on theoretical maximum wire speed and thresholds were set based
1396  * on testing data as well as attempting to minimize response time
1397  * while increasing bulk throughput.
1398  **/
1399 static void ixgbevf_update_itr(struct ixgbevf_q_vector *q_vector,
1400                                struct ixgbevf_ring_container *ring_container)
1401 {
1402         int bytes = ring_container->total_bytes;
1403         int packets = ring_container->total_packets;
1404         u32 timepassed_us;
1405         u64 bytes_perint;
1406         u8 itr_setting = ring_container->itr;
1407
1408         if (packets == 0)
1409                 return;
1410
1411         /* simple throttle rate management
1412          *    0-20MB/s lowest (100000 ints/s)
1413          *   20-100MB/s low   (20000 ints/s)
1414          *  100-1249MB/s bulk (12000 ints/s)
1415          */
1416         /* what was last interrupt timeslice? */
1417         timepassed_us = q_vector->itr >> 2;
1418         bytes_perint = bytes / timepassed_us; /* bytes/usec */
1419
1420         switch (itr_setting) {
1421         case lowest_latency:
1422                 if (bytes_perint > 10)
1423                         itr_setting = low_latency;
1424                 break;
1425         case low_latency:
1426                 if (bytes_perint > 20)
1427                         itr_setting = bulk_latency;
1428                 else if (bytes_perint <= 10)
1429                         itr_setting = lowest_latency;
1430                 break;
1431         case bulk_latency:
1432                 if (bytes_perint <= 20)
1433                         itr_setting = low_latency;
1434                 break;
1435         }
1436
1437         /* clear work counters since we have the values we need */
1438         ring_container->total_bytes = 0;
1439         ring_container->total_packets = 0;
1440
1441         /* write updated itr to ring container */
1442         ring_container->itr = itr_setting;
1443 }
1444
1445 static void ixgbevf_set_itr(struct ixgbevf_q_vector *q_vector)
1446 {
1447         u32 new_itr = q_vector->itr;
1448         u8 current_itr;
1449
1450         ixgbevf_update_itr(q_vector, &q_vector->tx);
1451         ixgbevf_update_itr(q_vector, &q_vector->rx);
1452
1453         current_itr = max(q_vector->rx.itr, q_vector->tx.itr);
1454
1455         switch (current_itr) {
1456         /* counts and packets in update_itr are dependent on these numbers */
1457         case lowest_latency:
1458                 new_itr = IXGBE_100K_ITR;
1459                 break;
1460         case low_latency:
1461                 new_itr = IXGBE_20K_ITR;
1462                 break;
1463         case bulk_latency:
1464                 new_itr = IXGBE_12K_ITR;
1465                 break;
1466         default:
1467                 break;
1468         }
1469
1470         if (new_itr != q_vector->itr) {
1471                 /* do an exponential smoothing */
1472                 new_itr = (10 * new_itr * q_vector->itr) /
1473                           ((9 * new_itr) + q_vector->itr);
1474
1475                 /* save the algorithm value here */
1476                 q_vector->itr = new_itr;
1477
1478                 ixgbevf_write_eitr(q_vector);
1479         }
1480 }
1481
1482 static irqreturn_t ixgbevf_msix_other(int irq, void *data)
1483 {
1484         struct ixgbevf_adapter *adapter = data;
1485         struct ixgbe_hw *hw = &adapter->hw;
1486
1487         hw->mac.get_link_status = 1;
1488
1489         ixgbevf_service_event_schedule(adapter);
1490
1491         IXGBE_WRITE_REG(hw, IXGBE_VTEIMS, adapter->eims_other);
1492
1493         return IRQ_HANDLED;
1494 }
1495
1496 /**
1497  * ixgbevf_msix_clean_rings - single unshared vector rx clean (all queues)
1498  * @irq: unused
1499  * @data: pointer to our q_vector struct for this interrupt vector
1500  **/
1501 static irqreturn_t ixgbevf_msix_clean_rings(int irq, void *data)
1502 {
1503         struct ixgbevf_q_vector *q_vector = data;
1504
1505         /* EIAM disabled interrupts (on this vector) for us */
1506         if (q_vector->rx.ring || q_vector->tx.ring)
1507                 napi_schedule_irqoff(&q_vector->napi);
1508
1509         return IRQ_HANDLED;
1510 }
1511
1512 /**
1513  * ixgbevf_request_msix_irqs - Initialize MSI-X interrupts
1514  * @adapter: board private structure
1515  *
1516  * ixgbevf_request_msix_irqs allocates MSI-X vectors and requests
1517  * interrupts from the kernel.
1518  **/
1519 static int ixgbevf_request_msix_irqs(struct ixgbevf_adapter *adapter)
1520 {
1521         struct net_device *netdev = adapter->netdev;
1522         int q_vectors = adapter->num_msix_vectors - NON_Q_VECTORS;
1523         unsigned int ri = 0, ti = 0;
1524         int vector, err;
1525
1526         for (vector = 0; vector < q_vectors; vector++) {
1527                 struct ixgbevf_q_vector *q_vector = adapter->q_vector[vector];
1528                 struct msix_entry *entry = &adapter->msix_entries[vector];
1529
1530                 if (q_vector->tx.ring && q_vector->rx.ring) {
1531                         snprintf(q_vector->name, sizeof(q_vector->name),
1532                                  "%s-TxRx-%u", netdev->name, ri++);
1533                         ti++;
1534                 } else if (q_vector->rx.ring) {
1535                         snprintf(q_vector->name, sizeof(q_vector->name),
1536                                  "%s-rx-%u", netdev->name, ri++);
1537                 } else if (q_vector->tx.ring) {
1538                         snprintf(q_vector->name, sizeof(q_vector->name),
1539                                  "%s-tx-%u", netdev->name, ti++);
1540                 } else {
1541                         /* skip this unused q_vector */
1542                         continue;
1543                 }
1544                 err = request_irq(entry->vector, &ixgbevf_msix_clean_rings, 0,
1545                                   q_vector->name, q_vector);
1546                 if (err) {
1547                         hw_dbg(&adapter->hw,
1548                                "request_irq failed for MSIX interrupt Error: %d\n",
1549                                err);
1550                         goto free_queue_irqs;
1551                 }
1552         }
1553
1554         err = request_irq(adapter->msix_entries[vector].vector,
1555                           &ixgbevf_msix_other, 0, netdev->name, adapter);
1556         if (err) {
1557                 hw_dbg(&adapter->hw, "request_irq for msix_other failed: %d\n",
1558                        err);
1559                 goto free_queue_irqs;
1560         }
1561
1562         return 0;
1563
1564 free_queue_irqs:
1565         while (vector) {
1566                 vector--;
1567                 free_irq(adapter->msix_entries[vector].vector,
1568                          adapter->q_vector[vector]);
1569         }
1570         /* This failure is non-recoverable - it indicates the system is
1571          * out of MSIX vector resources and the VF driver cannot run
1572          * without them.  Set the number of msix vectors to zero
1573          * indicating that not enough can be allocated.  The error
1574          * will be returned to the user indicating device open failed.
1575          * Any further attempts to force the driver to open will also
1576          * fail.  The only way to recover is to unload the driver and
1577          * reload it again.  If the system has recovered some MSIX
1578          * vectors then it may succeed.
1579          */
1580         adapter->num_msix_vectors = 0;
1581         return err;
1582 }
1583
1584 /**
1585  * ixgbevf_request_irq - initialize interrupts
1586  * @adapter: board private structure
1587  *
1588  * Attempts to configure interrupts using the best available
1589  * capabilities of the hardware and kernel.
1590  **/
1591 static int ixgbevf_request_irq(struct ixgbevf_adapter *adapter)
1592 {
1593         int err = ixgbevf_request_msix_irqs(adapter);
1594
1595         if (err)
1596                 hw_dbg(&adapter->hw, "request_irq failed, Error %d\n", err);
1597
1598         return err;
1599 }
1600
1601 static void ixgbevf_free_irq(struct ixgbevf_adapter *adapter)
1602 {
1603         int i, q_vectors;
1604
1605         if (!adapter->msix_entries)
1606                 return;
1607
1608         q_vectors = adapter->num_msix_vectors;
1609         i = q_vectors - 1;
1610
1611         free_irq(adapter->msix_entries[i].vector, adapter);
1612         i--;
1613
1614         for (; i >= 0; i--) {
1615                 /* free only the irqs that were actually requested */
1616                 if (!adapter->q_vector[i]->rx.ring &&
1617                     !adapter->q_vector[i]->tx.ring)
1618                         continue;
1619
1620                 free_irq(adapter->msix_entries[i].vector,
1621                          adapter->q_vector[i]);
1622         }
1623 }
1624
1625 /**
1626  * ixgbevf_irq_disable - Mask off interrupt generation on the NIC
1627  * @adapter: board private structure
1628  **/
1629 static inline void ixgbevf_irq_disable(struct ixgbevf_adapter *adapter)
1630 {
1631         struct ixgbe_hw *hw = &adapter->hw;
1632         int i;
1633
1634         IXGBE_WRITE_REG(hw, IXGBE_VTEIAM, 0);
1635         IXGBE_WRITE_REG(hw, IXGBE_VTEIMC, ~0);
1636         IXGBE_WRITE_REG(hw, IXGBE_VTEIAC, 0);
1637
1638         IXGBE_WRITE_FLUSH(hw);
1639
1640         for (i = 0; i < adapter->num_msix_vectors; i++)
1641                 synchronize_irq(adapter->msix_entries[i].vector);
1642 }
1643
1644 /**
1645  * ixgbevf_irq_enable - Enable default interrupt generation settings
1646  * @adapter: board private structure
1647  **/
1648 static inline void ixgbevf_irq_enable(struct ixgbevf_adapter *adapter)
1649 {
1650         struct ixgbe_hw *hw = &adapter->hw;
1651
1652         IXGBE_WRITE_REG(hw, IXGBE_VTEIAM, adapter->eims_enable_mask);
1653         IXGBE_WRITE_REG(hw, IXGBE_VTEIAC, adapter->eims_enable_mask);
1654         IXGBE_WRITE_REG(hw, IXGBE_VTEIMS, adapter->eims_enable_mask);
1655 }
1656
1657 /**
1658  * ixgbevf_configure_tx_ring - Configure 82599 VF Tx ring after Reset
1659  * @adapter: board private structure
1660  * @ring: structure containing ring specific data
1661  *
1662  * Configure the Tx descriptor ring after a reset.
1663  **/
1664 static void ixgbevf_configure_tx_ring(struct ixgbevf_adapter *adapter,
1665                                       struct ixgbevf_ring *ring)
1666 {
1667         struct ixgbe_hw *hw = &adapter->hw;
1668         u64 tdba = ring->dma;
1669         int wait_loop = 10;
1670         u32 txdctl = IXGBE_TXDCTL_ENABLE;
1671         u8 reg_idx = ring->reg_idx;
1672
1673         /* disable queue to avoid issues while updating state */
1674         IXGBE_WRITE_REG(hw, IXGBE_VFTXDCTL(reg_idx), IXGBE_TXDCTL_SWFLSH);
1675         IXGBE_WRITE_FLUSH(hw);
1676
1677         IXGBE_WRITE_REG(hw, IXGBE_VFTDBAL(reg_idx), tdba & DMA_BIT_MASK(32));
1678         IXGBE_WRITE_REG(hw, IXGBE_VFTDBAH(reg_idx), tdba >> 32);
1679         IXGBE_WRITE_REG(hw, IXGBE_VFTDLEN(reg_idx),
1680                         ring->count * sizeof(union ixgbe_adv_tx_desc));
1681
1682         /* disable head writeback */
1683         IXGBE_WRITE_REG(hw, IXGBE_VFTDWBAH(reg_idx), 0);
1684         IXGBE_WRITE_REG(hw, IXGBE_VFTDWBAL(reg_idx), 0);
1685
1686         /* enable relaxed ordering */
1687         IXGBE_WRITE_REG(hw, IXGBE_VFDCA_TXCTRL(reg_idx),
1688                         (IXGBE_DCA_TXCTRL_DESC_RRO_EN |
1689                          IXGBE_DCA_TXCTRL_DATA_RRO_EN));
1690
1691         /* reset head and tail pointers */
1692         IXGBE_WRITE_REG(hw, IXGBE_VFTDH(reg_idx), 0);
1693         IXGBE_WRITE_REG(hw, IXGBE_VFTDT(reg_idx), 0);
1694         ring->tail = adapter->io_addr + IXGBE_VFTDT(reg_idx);
1695
1696         /* reset ntu and ntc to place SW in sync with hardwdare */
1697         ring->next_to_clean = 0;
1698         ring->next_to_use = 0;
1699
1700         /* In order to avoid issues WTHRESH + PTHRESH should always be equal
1701          * to or less than the number of on chip descriptors, which is
1702          * currently 40.
1703          */
1704         txdctl |= (8 << 16);    /* WTHRESH = 8 */
1705
1706         /* Setting PTHRESH to 32 both improves performance */
1707         txdctl |= (1u << 8) |    /* HTHRESH = 1 */
1708                    32;           /* PTHRESH = 32 */
1709
1710         /* reinitialize tx_buffer_info */
1711         memset(ring->tx_buffer_info, 0,
1712                sizeof(struct ixgbevf_tx_buffer) * ring->count);
1713
1714         clear_bit(__IXGBEVF_HANG_CHECK_ARMED, &ring->state);
1715         clear_bit(__IXGBEVF_TX_XDP_RING_PRIMED, &ring->state);
1716
1717         IXGBE_WRITE_REG(hw, IXGBE_VFTXDCTL(reg_idx), txdctl);
1718
1719         /* poll to verify queue is enabled */
1720         do {
1721                 usleep_range(1000, 2000);
1722                 txdctl = IXGBE_READ_REG(hw, IXGBE_VFTXDCTL(reg_idx));
1723         }  while (--wait_loop && !(txdctl & IXGBE_TXDCTL_ENABLE));
1724         if (!wait_loop)
1725                 hw_dbg(hw, "Could not enable Tx Queue %d\n", reg_idx);
1726 }
1727
1728 /**
1729  * ixgbevf_configure_tx - Configure 82599 VF Transmit Unit after Reset
1730  * @adapter: board private structure
1731  *
1732  * Configure the Tx unit of the MAC after a reset.
1733  **/
1734 static void ixgbevf_configure_tx(struct ixgbevf_adapter *adapter)
1735 {
1736         u32 i;
1737
1738         /* Setup the HW Tx Head and Tail descriptor pointers */
1739         for (i = 0; i < adapter->num_tx_queues; i++)
1740                 ixgbevf_configure_tx_ring(adapter, adapter->tx_ring[i]);
1741         for (i = 0; i < adapter->num_xdp_queues; i++)
1742                 ixgbevf_configure_tx_ring(adapter, adapter->xdp_ring[i]);
1743 }
1744
1745 #define IXGBE_SRRCTL_BSIZEHDRSIZE_SHIFT 2
1746
1747 static void ixgbevf_configure_srrctl(struct ixgbevf_adapter *adapter,
1748                                      struct ixgbevf_ring *ring, int index)
1749 {
1750         struct ixgbe_hw *hw = &adapter->hw;
1751         u32 srrctl;
1752
1753         srrctl = IXGBE_SRRCTL_DROP_EN;
1754
1755         srrctl |= IXGBEVF_RX_HDR_SIZE << IXGBE_SRRCTL_BSIZEHDRSIZE_SHIFT;
1756         if (ring_uses_large_buffer(ring))
1757                 srrctl |= IXGBEVF_RXBUFFER_3072 >> IXGBE_SRRCTL_BSIZEPKT_SHIFT;
1758         else
1759                 srrctl |= IXGBEVF_RXBUFFER_2048 >> IXGBE_SRRCTL_BSIZEPKT_SHIFT;
1760         srrctl |= IXGBE_SRRCTL_DESCTYPE_ADV_ONEBUF;
1761
1762         IXGBE_WRITE_REG(hw, IXGBE_VFSRRCTL(index), srrctl);
1763 }
1764
1765 static void ixgbevf_setup_psrtype(struct ixgbevf_adapter *adapter)
1766 {
1767         struct ixgbe_hw *hw = &adapter->hw;
1768
1769         /* PSRTYPE must be initialized in 82599 */
1770         u32 psrtype = IXGBE_PSRTYPE_TCPHDR | IXGBE_PSRTYPE_UDPHDR |
1771                       IXGBE_PSRTYPE_IPV4HDR | IXGBE_PSRTYPE_IPV6HDR |
1772                       IXGBE_PSRTYPE_L2HDR;
1773
1774         if (adapter->num_rx_queues > 1)
1775                 psrtype |= BIT(29);
1776
1777         IXGBE_WRITE_REG(hw, IXGBE_VFPSRTYPE, psrtype);
1778 }
1779
1780 #define IXGBEVF_MAX_RX_DESC_POLL 10
1781 static void ixgbevf_disable_rx_queue(struct ixgbevf_adapter *adapter,
1782                                      struct ixgbevf_ring *ring)
1783 {
1784         struct ixgbe_hw *hw = &adapter->hw;
1785         int wait_loop = IXGBEVF_MAX_RX_DESC_POLL;
1786         u32 rxdctl;
1787         u8 reg_idx = ring->reg_idx;
1788
1789         if (IXGBE_REMOVED(hw->hw_addr))
1790                 return;
1791         rxdctl = IXGBE_READ_REG(hw, IXGBE_VFRXDCTL(reg_idx));
1792         rxdctl &= ~IXGBE_RXDCTL_ENABLE;
1793
1794         /* write value back with RXDCTL.ENABLE bit cleared */
1795         IXGBE_WRITE_REG(hw, IXGBE_VFRXDCTL(reg_idx), rxdctl);
1796
1797         /* the hardware may take up to 100us to really disable the Rx queue */
1798         do {
1799                 udelay(10);
1800                 rxdctl = IXGBE_READ_REG(hw, IXGBE_VFRXDCTL(reg_idx));
1801         } while (--wait_loop && (rxdctl & IXGBE_RXDCTL_ENABLE));
1802
1803         if (!wait_loop)
1804                 pr_err("RXDCTL.ENABLE queue %d not cleared while polling\n",
1805                        reg_idx);
1806 }
1807
1808 static void ixgbevf_rx_desc_queue_enable(struct ixgbevf_adapter *adapter,
1809                                          struct ixgbevf_ring *ring)
1810 {
1811         struct ixgbe_hw *hw = &adapter->hw;
1812         int wait_loop = IXGBEVF_MAX_RX_DESC_POLL;
1813         u32 rxdctl;
1814         u8 reg_idx = ring->reg_idx;
1815
1816         if (IXGBE_REMOVED(hw->hw_addr))
1817                 return;
1818         do {
1819                 usleep_range(1000, 2000);
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 set while polling\n",
1825                        reg_idx);
1826 }
1827
1828 /**
1829  * ixgbevf_init_rss_key - Initialize adapter RSS key
1830  * @adapter: device handle
1831  *
1832  * Allocates and initializes the RSS key if it is not allocated.
1833  **/
1834 static inline int ixgbevf_init_rss_key(struct ixgbevf_adapter *adapter)
1835 {
1836         u32 *rss_key;
1837
1838         if (!adapter->rss_key) {
1839                 rss_key = kzalloc(IXGBEVF_RSS_HASH_KEY_SIZE, GFP_KERNEL);
1840                 if (unlikely(!rss_key))
1841                         return -ENOMEM;
1842
1843                 netdev_rss_key_fill(rss_key, IXGBEVF_RSS_HASH_KEY_SIZE);
1844                 adapter->rss_key = rss_key;
1845         }
1846
1847         return 0;
1848 }
1849
1850 static void ixgbevf_setup_vfmrqc(struct ixgbevf_adapter *adapter)
1851 {
1852         struct ixgbe_hw *hw = &adapter->hw;
1853         u32 vfmrqc = 0, vfreta = 0;
1854         u16 rss_i = adapter->num_rx_queues;
1855         u8 i, j;
1856
1857         /* Fill out hash function seeds */
1858         for (i = 0; i < IXGBEVF_VFRSSRK_REGS; i++)
1859                 IXGBE_WRITE_REG(hw, IXGBE_VFRSSRK(i), *(adapter->rss_key + i));
1860
1861         for (i = 0, j = 0; i < IXGBEVF_X550_VFRETA_SIZE; i++, j++) {
1862                 if (j == rss_i)
1863                         j = 0;
1864
1865                 adapter->rss_indir_tbl[i] = j;
1866
1867                 vfreta |= j << (i & 0x3) * 8;
1868                 if ((i & 3) == 3) {
1869                         IXGBE_WRITE_REG(hw, IXGBE_VFRETA(i >> 2), vfreta);
1870                         vfreta = 0;
1871                 }
1872         }
1873
1874         /* Perform hash on these packet types */
1875         vfmrqc |= IXGBE_VFMRQC_RSS_FIELD_IPV4 |
1876                 IXGBE_VFMRQC_RSS_FIELD_IPV4_TCP |
1877                 IXGBE_VFMRQC_RSS_FIELD_IPV6 |
1878                 IXGBE_VFMRQC_RSS_FIELD_IPV6_TCP;
1879
1880         vfmrqc |= IXGBE_VFMRQC_RSSEN;
1881
1882         IXGBE_WRITE_REG(hw, IXGBE_VFMRQC, vfmrqc);
1883 }
1884
1885 static void ixgbevf_configure_rx_ring(struct ixgbevf_adapter *adapter,
1886                                       struct ixgbevf_ring *ring)
1887 {
1888         struct ixgbe_hw *hw = &adapter->hw;
1889         union ixgbe_adv_rx_desc *rx_desc;
1890         u64 rdba = ring->dma;
1891         u32 rxdctl;
1892         u8 reg_idx = ring->reg_idx;
1893
1894         /* disable queue to avoid issues while updating state */
1895         rxdctl = IXGBE_READ_REG(hw, IXGBE_VFRXDCTL(reg_idx));
1896         ixgbevf_disable_rx_queue(adapter, ring);
1897
1898         IXGBE_WRITE_REG(hw, IXGBE_VFRDBAL(reg_idx), rdba & DMA_BIT_MASK(32));
1899         IXGBE_WRITE_REG(hw, IXGBE_VFRDBAH(reg_idx), rdba >> 32);
1900         IXGBE_WRITE_REG(hw, IXGBE_VFRDLEN(reg_idx),
1901                         ring->count * sizeof(union ixgbe_adv_rx_desc));
1902
1903 #ifndef CONFIG_SPARC
1904         /* enable relaxed ordering */
1905         IXGBE_WRITE_REG(hw, IXGBE_VFDCA_RXCTRL(reg_idx),
1906                         IXGBE_DCA_RXCTRL_DESC_RRO_EN);
1907 #else
1908         IXGBE_WRITE_REG(hw, IXGBE_VFDCA_RXCTRL(reg_idx),
1909                         IXGBE_DCA_RXCTRL_DESC_RRO_EN |
1910                         IXGBE_DCA_RXCTRL_DATA_WRO_EN);
1911 #endif
1912
1913         /* reset head and tail pointers */
1914         IXGBE_WRITE_REG(hw, IXGBE_VFRDH(reg_idx), 0);
1915         IXGBE_WRITE_REG(hw, IXGBE_VFRDT(reg_idx), 0);
1916         ring->tail = adapter->io_addr + IXGBE_VFRDT(reg_idx);
1917
1918         /* initialize rx_buffer_info */
1919         memset(ring->rx_buffer_info, 0,
1920                sizeof(struct ixgbevf_rx_buffer) * ring->count);
1921
1922         /* initialize Rx descriptor 0 */
1923         rx_desc = IXGBEVF_RX_DESC(ring, 0);
1924         rx_desc->wb.upper.length = 0;
1925
1926         /* reset ntu and ntc to place SW in sync with hardwdare */
1927         ring->next_to_clean = 0;
1928         ring->next_to_use = 0;
1929         ring->next_to_alloc = 0;
1930
1931         ixgbevf_configure_srrctl(adapter, ring, reg_idx);
1932
1933         /* RXDCTL.RLPML does not work on 82599 */
1934         if (adapter->hw.mac.type != ixgbe_mac_82599_vf) {
1935                 rxdctl &= ~(IXGBE_RXDCTL_RLPMLMASK |
1936                             IXGBE_RXDCTL_RLPML_EN);
1937
1938 #if (PAGE_SIZE < 8192)
1939                 /* Limit the maximum frame size so we don't overrun the skb */
1940                 if (ring_uses_build_skb(ring) &&
1941                     !ring_uses_large_buffer(ring))
1942                         rxdctl |= IXGBEVF_MAX_FRAME_BUILD_SKB |
1943                                   IXGBE_RXDCTL_RLPML_EN;
1944 #endif
1945         }
1946
1947         rxdctl |= IXGBE_RXDCTL_ENABLE | IXGBE_RXDCTL_VME;
1948         IXGBE_WRITE_REG(hw, IXGBE_VFRXDCTL(reg_idx), rxdctl);
1949
1950         ixgbevf_rx_desc_queue_enable(adapter, ring);
1951         ixgbevf_alloc_rx_buffers(ring, ixgbevf_desc_unused(ring));
1952 }
1953
1954 static void ixgbevf_set_rx_buffer_len(struct ixgbevf_adapter *adapter,
1955                                       struct ixgbevf_ring *rx_ring)
1956 {
1957         struct net_device *netdev = adapter->netdev;
1958         unsigned int max_frame = netdev->mtu + ETH_HLEN + ETH_FCS_LEN;
1959
1960         /* set build_skb and buffer size flags */
1961         clear_ring_build_skb_enabled(rx_ring);
1962         clear_ring_uses_large_buffer(rx_ring);
1963
1964         if (adapter->flags & IXGBEVF_FLAGS_LEGACY_RX)
1965                 return;
1966
1967         set_ring_build_skb_enabled(rx_ring);
1968
1969         if (PAGE_SIZE < 8192) {
1970                 if (max_frame <= IXGBEVF_MAX_FRAME_BUILD_SKB)
1971                         return;
1972
1973                 set_ring_uses_large_buffer(rx_ring);
1974         }
1975 }
1976
1977 /**
1978  * ixgbevf_configure_rx - Configure 82599 VF Receive Unit after Reset
1979  * @adapter: board private structure
1980  *
1981  * Configure the Rx unit of the MAC after a reset.
1982  **/
1983 static void ixgbevf_configure_rx(struct ixgbevf_adapter *adapter)
1984 {
1985         struct ixgbe_hw *hw = &adapter->hw;
1986         struct net_device *netdev = adapter->netdev;
1987         int i, ret;
1988
1989         ixgbevf_setup_psrtype(adapter);
1990         if (hw->mac.type >= ixgbe_mac_X550_vf)
1991                 ixgbevf_setup_vfmrqc(adapter);
1992
1993         spin_lock_bh(&adapter->mbx_lock);
1994         /* notify the PF of our intent to use this size of frame */
1995         ret = hw->mac.ops.set_rlpml(hw, netdev->mtu + ETH_HLEN + ETH_FCS_LEN);
1996         spin_unlock_bh(&adapter->mbx_lock);
1997         if (ret)
1998                 dev_err(&adapter->pdev->dev,
1999                         "Failed to set MTU at %d\n", netdev->mtu);
2000
2001         /* Setup the HW Rx Head and Tail Descriptor Pointers and
2002          * the Base and Length of the Rx Descriptor Ring
2003          */
2004         for (i = 0; i < adapter->num_rx_queues; i++) {
2005                 struct ixgbevf_ring *rx_ring = adapter->rx_ring[i];
2006
2007                 ixgbevf_set_rx_buffer_len(adapter, rx_ring);
2008                 ixgbevf_configure_rx_ring(adapter, rx_ring);
2009         }
2010 }
2011
2012 static int ixgbevf_vlan_rx_add_vid(struct net_device *netdev,
2013                                    __be16 proto, u16 vid)
2014 {
2015         struct ixgbevf_adapter *adapter = netdev_priv(netdev);
2016         struct ixgbe_hw *hw = &adapter->hw;
2017         int err;
2018
2019         spin_lock_bh(&adapter->mbx_lock);
2020
2021         /* add VID to filter table */
2022         err = hw->mac.ops.set_vfta(hw, vid, 0, true);
2023
2024         spin_unlock_bh(&adapter->mbx_lock);
2025
2026         /* translate error return types so error makes sense */
2027         if (err == IXGBE_ERR_MBX)
2028                 return -EIO;
2029
2030         if (err == IXGBE_ERR_INVALID_ARGUMENT)
2031                 return -EACCES;
2032
2033         set_bit(vid, adapter->active_vlans);
2034
2035         return err;
2036 }
2037
2038 static int ixgbevf_vlan_rx_kill_vid(struct net_device *netdev,
2039                                     __be16 proto, u16 vid)
2040 {
2041         struct ixgbevf_adapter *adapter = netdev_priv(netdev);
2042         struct ixgbe_hw *hw = &adapter->hw;
2043         int err;
2044
2045         spin_lock_bh(&adapter->mbx_lock);
2046
2047         /* remove VID from filter table */
2048         err = hw->mac.ops.set_vfta(hw, vid, 0, false);
2049
2050         spin_unlock_bh(&adapter->mbx_lock);
2051
2052         clear_bit(vid, adapter->active_vlans);
2053
2054         return err;
2055 }
2056
2057 static void ixgbevf_restore_vlan(struct ixgbevf_adapter *adapter)
2058 {
2059         u16 vid;
2060
2061         for_each_set_bit(vid, adapter->active_vlans, VLAN_N_VID)
2062                 ixgbevf_vlan_rx_add_vid(adapter->netdev,
2063                                         htons(ETH_P_8021Q), vid);
2064 }
2065
2066 static int ixgbevf_write_uc_addr_list(struct net_device *netdev)
2067 {
2068         struct ixgbevf_adapter *adapter = netdev_priv(netdev);
2069         struct ixgbe_hw *hw = &adapter->hw;
2070         int count = 0;
2071
2072         if (!netdev_uc_empty(netdev)) {
2073                 struct netdev_hw_addr *ha;
2074
2075                 netdev_for_each_uc_addr(ha, netdev) {
2076                         hw->mac.ops.set_uc_addr(hw, ++count, ha->addr);
2077                         udelay(200);
2078                 }
2079         } else {
2080                 /* If the list is empty then send message to PF driver to
2081                  * clear all MAC VLANs on this VF.
2082                  */
2083                 hw->mac.ops.set_uc_addr(hw, 0, NULL);
2084         }
2085
2086         return count;
2087 }
2088
2089 /**
2090  * ixgbevf_set_rx_mode - Multicast and unicast set
2091  * @netdev: network interface device structure
2092  *
2093  * The set_rx_method entry point is called whenever the multicast address
2094  * list, unicast address list or the network interface flags are updated.
2095  * This routine is responsible for configuring the hardware for proper
2096  * multicast mode and configuring requested unicast filters.
2097  **/
2098 static void ixgbevf_set_rx_mode(struct net_device *netdev)
2099 {
2100         struct ixgbevf_adapter *adapter = netdev_priv(netdev);
2101         struct ixgbe_hw *hw = &adapter->hw;
2102         unsigned int flags = netdev->flags;
2103         int xcast_mode;
2104
2105         /* request the most inclusive mode we need */
2106         if (flags & IFF_PROMISC)
2107                 xcast_mode = IXGBEVF_XCAST_MODE_PROMISC;
2108         else if (flags & IFF_ALLMULTI)
2109                 xcast_mode = IXGBEVF_XCAST_MODE_ALLMULTI;
2110         else if (flags & (IFF_BROADCAST | IFF_MULTICAST))
2111                 xcast_mode = IXGBEVF_XCAST_MODE_MULTI;
2112         else
2113                 xcast_mode = IXGBEVF_XCAST_MODE_NONE;
2114
2115         spin_lock_bh(&adapter->mbx_lock);
2116
2117         hw->mac.ops.update_xcast_mode(hw, xcast_mode);
2118
2119         /* reprogram multicast list */
2120         hw->mac.ops.update_mc_addr_list(hw, netdev);
2121
2122         ixgbevf_write_uc_addr_list(netdev);
2123
2124         spin_unlock_bh(&adapter->mbx_lock);
2125 }
2126
2127 static void ixgbevf_napi_enable_all(struct ixgbevf_adapter *adapter)
2128 {
2129         int q_idx;
2130         struct ixgbevf_q_vector *q_vector;
2131         int q_vectors = adapter->num_msix_vectors - NON_Q_VECTORS;
2132
2133         for (q_idx = 0; q_idx < q_vectors; q_idx++) {
2134                 q_vector = adapter->q_vector[q_idx];
2135                 napi_enable(&q_vector->napi);
2136         }
2137 }
2138
2139 static void ixgbevf_napi_disable_all(struct ixgbevf_adapter *adapter)
2140 {
2141         int q_idx;
2142         struct ixgbevf_q_vector *q_vector;
2143         int q_vectors = adapter->num_msix_vectors - NON_Q_VECTORS;
2144
2145         for (q_idx = 0; q_idx < q_vectors; q_idx++) {
2146                 q_vector = adapter->q_vector[q_idx];
2147                 napi_disable(&q_vector->napi);
2148         }
2149 }
2150
2151 static int ixgbevf_configure_dcb(struct ixgbevf_adapter *adapter)
2152 {
2153         struct ixgbe_hw *hw = &adapter->hw;
2154         unsigned int def_q = 0;
2155         unsigned int num_tcs = 0;
2156         unsigned int num_rx_queues = adapter->num_rx_queues;
2157         unsigned int num_tx_queues = adapter->num_tx_queues;
2158         int err;
2159
2160         spin_lock_bh(&adapter->mbx_lock);
2161
2162         /* fetch queue configuration from the PF */
2163         err = ixgbevf_get_queues(hw, &num_tcs, &def_q);
2164
2165         spin_unlock_bh(&adapter->mbx_lock);
2166
2167         if (err)
2168                 return err;
2169
2170         if (num_tcs > 1) {
2171                 /* we need only one Tx queue */
2172                 num_tx_queues = 1;
2173
2174                 /* update default Tx ring register index */
2175                 adapter->tx_ring[0]->reg_idx = def_q;
2176
2177                 /* we need as many queues as traffic classes */
2178                 num_rx_queues = num_tcs;
2179         }
2180
2181         /* if we have a bad config abort request queue reset */
2182         if ((adapter->num_rx_queues != num_rx_queues) ||
2183             (adapter->num_tx_queues != num_tx_queues)) {
2184                 /* force mailbox timeout to prevent further messages */
2185                 hw->mbx.timeout = 0;
2186
2187                 /* wait for watchdog to come around and bail us out */
2188                 set_bit(__IXGBEVF_QUEUE_RESET_REQUESTED, &adapter->state);
2189         }
2190
2191         return 0;
2192 }
2193
2194 static void ixgbevf_configure(struct ixgbevf_adapter *adapter)
2195 {
2196         ixgbevf_configure_dcb(adapter);
2197
2198         ixgbevf_set_rx_mode(adapter->netdev);
2199
2200         ixgbevf_restore_vlan(adapter);
2201
2202         ixgbevf_configure_tx(adapter);
2203         ixgbevf_configure_rx(adapter);
2204 }
2205
2206 static void ixgbevf_save_reset_stats(struct ixgbevf_adapter *adapter)
2207 {
2208         /* Only save pre-reset stats if there are some */
2209         if (adapter->stats.vfgprc || adapter->stats.vfgptc) {
2210                 adapter->stats.saved_reset_vfgprc += adapter->stats.vfgprc -
2211                         adapter->stats.base_vfgprc;
2212                 adapter->stats.saved_reset_vfgptc += adapter->stats.vfgptc -
2213                         adapter->stats.base_vfgptc;
2214                 adapter->stats.saved_reset_vfgorc += adapter->stats.vfgorc -
2215                         adapter->stats.base_vfgorc;
2216                 adapter->stats.saved_reset_vfgotc += adapter->stats.vfgotc -
2217                         adapter->stats.base_vfgotc;
2218                 adapter->stats.saved_reset_vfmprc += adapter->stats.vfmprc -
2219                         adapter->stats.base_vfmprc;
2220         }
2221 }
2222
2223 static void ixgbevf_init_last_counter_stats(struct ixgbevf_adapter *adapter)
2224 {
2225         struct ixgbe_hw *hw = &adapter->hw;
2226
2227         adapter->stats.last_vfgprc = IXGBE_READ_REG(hw, IXGBE_VFGPRC);
2228         adapter->stats.last_vfgorc = IXGBE_READ_REG(hw, IXGBE_VFGORC_LSB);
2229         adapter->stats.last_vfgorc |=
2230                 (((u64)(IXGBE_READ_REG(hw, IXGBE_VFGORC_MSB))) << 32);
2231         adapter->stats.last_vfgptc = IXGBE_READ_REG(hw, IXGBE_VFGPTC);
2232         adapter->stats.last_vfgotc = IXGBE_READ_REG(hw, IXGBE_VFGOTC_LSB);
2233         adapter->stats.last_vfgotc |=
2234                 (((u64)(IXGBE_READ_REG(hw, IXGBE_VFGOTC_MSB))) << 32);
2235         adapter->stats.last_vfmprc = IXGBE_READ_REG(hw, IXGBE_VFMPRC);
2236
2237         adapter->stats.base_vfgprc = adapter->stats.last_vfgprc;
2238         adapter->stats.base_vfgorc = adapter->stats.last_vfgorc;
2239         adapter->stats.base_vfgptc = adapter->stats.last_vfgptc;
2240         adapter->stats.base_vfgotc = adapter->stats.last_vfgotc;
2241         adapter->stats.base_vfmprc = adapter->stats.last_vfmprc;
2242 }
2243
2244 static void ixgbevf_negotiate_api(struct ixgbevf_adapter *adapter)
2245 {
2246         struct ixgbe_hw *hw = &adapter->hw;
2247         int api[] = { ixgbe_mbox_api_13,
2248                       ixgbe_mbox_api_12,
2249                       ixgbe_mbox_api_11,
2250                       ixgbe_mbox_api_10,
2251                       ixgbe_mbox_api_unknown };
2252         int err, idx = 0;
2253
2254         spin_lock_bh(&adapter->mbx_lock);
2255
2256         while (api[idx] != ixgbe_mbox_api_unknown) {
2257                 err = hw->mac.ops.negotiate_api_version(hw, api[idx]);
2258                 if (!err)
2259                         break;
2260                 idx++;
2261         }
2262
2263         spin_unlock_bh(&adapter->mbx_lock);
2264 }
2265
2266 static void ixgbevf_up_complete(struct ixgbevf_adapter *adapter)
2267 {
2268         struct net_device *netdev = adapter->netdev;
2269         struct ixgbe_hw *hw = &adapter->hw;
2270
2271         ixgbevf_configure_msix(adapter);
2272
2273         spin_lock_bh(&adapter->mbx_lock);
2274
2275         if (is_valid_ether_addr(hw->mac.addr))
2276                 hw->mac.ops.set_rar(hw, 0, hw->mac.addr, 0);
2277         else
2278                 hw->mac.ops.set_rar(hw, 0, hw->mac.perm_addr, 0);
2279
2280         spin_unlock_bh(&adapter->mbx_lock);
2281
2282         smp_mb__before_atomic();
2283         clear_bit(__IXGBEVF_DOWN, &adapter->state);
2284         ixgbevf_napi_enable_all(adapter);
2285
2286         /* clear any pending interrupts, may auto mask */
2287         IXGBE_READ_REG(hw, IXGBE_VTEICR);
2288         ixgbevf_irq_enable(adapter);
2289
2290         /* enable transmits */
2291         netif_tx_start_all_queues(netdev);
2292
2293         ixgbevf_save_reset_stats(adapter);
2294         ixgbevf_init_last_counter_stats(adapter);
2295
2296         hw->mac.get_link_status = 1;
2297         mod_timer(&adapter->service_timer, jiffies);
2298 }
2299
2300 void ixgbevf_up(struct ixgbevf_adapter *adapter)
2301 {
2302         ixgbevf_configure(adapter);
2303
2304         ixgbevf_up_complete(adapter);
2305 }
2306
2307 /**
2308  * ixgbevf_clean_rx_ring - Free Rx Buffers per Queue
2309  * @rx_ring: ring to free buffers from
2310  **/
2311 static void ixgbevf_clean_rx_ring(struct ixgbevf_ring *rx_ring)
2312 {
2313         u16 i = rx_ring->next_to_clean;
2314
2315         /* Free Rx ring sk_buff */
2316         if (rx_ring->skb) {
2317                 dev_kfree_skb(rx_ring->skb);
2318                 rx_ring->skb = NULL;
2319         }
2320
2321         /* Free all the Rx ring pages */
2322         while (i != rx_ring->next_to_alloc) {
2323                 struct ixgbevf_rx_buffer *rx_buffer;
2324
2325                 rx_buffer = &rx_ring->rx_buffer_info[i];
2326
2327                 /* Invalidate cache lines that may have been written to by
2328                  * device so that we avoid corrupting memory.
2329                  */
2330                 dma_sync_single_range_for_cpu(rx_ring->dev,
2331                                               rx_buffer->dma,
2332                                               rx_buffer->page_offset,
2333                                               ixgbevf_rx_bufsz(rx_ring),
2334                                               DMA_FROM_DEVICE);
2335
2336                 /* free resources associated with mapping */
2337                 dma_unmap_page_attrs(rx_ring->dev,
2338                                      rx_buffer->dma,
2339                                      ixgbevf_rx_pg_size(rx_ring),
2340                                      DMA_FROM_DEVICE,
2341                                      IXGBEVF_RX_DMA_ATTR);
2342
2343                 __page_frag_cache_drain(rx_buffer->page,
2344                                         rx_buffer->pagecnt_bias);
2345
2346                 i++;
2347                 if (i == rx_ring->count)
2348                         i = 0;
2349         }
2350
2351         rx_ring->next_to_alloc = 0;
2352         rx_ring->next_to_clean = 0;
2353         rx_ring->next_to_use = 0;
2354 }
2355
2356 /**
2357  * ixgbevf_clean_tx_ring - Free Tx Buffers
2358  * @tx_ring: ring to be cleaned
2359  **/
2360 static void ixgbevf_clean_tx_ring(struct ixgbevf_ring *tx_ring)
2361 {
2362         u16 i = tx_ring->next_to_clean;
2363         struct ixgbevf_tx_buffer *tx_buffer = &tx_ring->tx_buffer_info[i];
2364
2365         while (i != tx_ring->next_to_use) {
2366                 union ixgbe_adv_tx_desc *eop_desc, *tx_desc;
2367
2368                 /* Free all the Tx ring sk_buffs */
2369                 if (ring_is_xdp(tx_ring))
2370                         page_frag_free(tx_buffer->data);
2371                 else
2372                         dev_kfree_skb_any(tx_buffer->skb);
2373
2374                 /* unmap skb header data */
2375                 dma_unmap_single(tx_ring->dev,
2376                                  dma_unmap_addr(tx_buffer, dma),
2377                                  dma_unmap_len(tx_buffer, len),
2378                                  DMA_TO_DEVICE);
2379
2380                 /* check for eop_desc to determine the end of the packet */
2381                 eop_desc = tx_buffer->next_to_watch;
2382                 tx_desc = IXGBEVF_TX_DESC(tx_ring, i);
2383
2384                 /* unmap remaining buffers */
2385                 while (tx_desc != eop_desc) {
2386                         tx_buffer++;
2387                         tx_desc++;
2388                         i++;
2389                         if (unlikely(i == tx_ring->count)) {
2390                                 i = 0;
2391                                 tx_buffer = tx_ring->tx_buffer_info;
2392                                 tx_desc = IXGBEVF_TX_DESC(tx_ring, 0);
2393                         }
2394
2395                         /* unmap any remaining paged data */
2396                         if (dma_unmap_len(tx_buffer, len))
2397                                 dma_unmap_page(tx_ring->dev,
2398                                                dma_unmap_addr(tx_buffer, dma),
2399                                                dma_unmap_len(tx_buffer, len),
2400                                                DMA_TO_DEVICE);
2401                 }
2402
2403                 /* move us one more past the eop_desc for start of next pkt */
2404                 tx_buffer++;
2405                 i++;
2406                 if (unlikely(i == tx_ring->count)) {
2407                         i = 0;
2408                         tx_buffer = tx_ring->tx_buffer_info;
2409                 }
2410         }
2411
2412         /* reset next_to_use and next_to_clean */
2413         tx_ring->next_to_use = 0;
2414         tx_ring->next_to_clean = 0;
2415
2416 }
2417
2418 /**
2419  * ixgbevf_clean_all_rx_rings - Free Rx Buffers for all queues
2420  * @adapter: board private structure
2421  **/
2422 static void ixgbevf_clean_all_rx_rings(struct ixgbevf_adapter *adapter)
2423 {
2424         int i;
2425
2426         for (i = 0; i < adapter->num_rx_queues; i++)
2427                 ixgbevf_clean_rx_ring(adapter->rx_ring[i]);
2428 }
2429
2430 /**
2431  * ixgbevf_clean_all_tx_rings - Free Tx Buffers for all queues
2432  * @adapter: board private structure
2433  **/
2434 static void ixgbevf_clean_all_tx_rings(struct ixgbevf_adapter *adapter)
2435 {
2436         int i;
2437
2438         for (i = 0; i < adapter->num_tx_queues; i++)
2439                 ixgbevf_clean_tx_ring(adapter->tx_ring[i]);
2440         for (i = 0; i < adapter->num_xdp_queues; i++)
2441                 ixgbevf_clean_tx_ring(adapter->xdp_ring[i]);
2442 }
2443
2444 void ixgbevf_down(struct ixgbevf_adapter *adapter)
2445 {
2446         struct net_device *netdev = adapter->netdev;
2447         struct ixgbe_hw *hw = &adapter->hw;
2448         int i;
2449
2450         /* signal that we are down to the interrupt handler */
2451         if (test_and_set_bit(__IXGBEVF_DOWN, &adapter->state))
2452                 return; /* do nothing if already down */
2453
2454         /* disable all enabled Rx queues */
2455         for (i = 0; i < adapter->num_rx_queues; i++)
2456                 ixgbevf_disable_rx_queue(adapter, adapter->rx_ring[i]);
2457
2458         usleep_range(10000, 20000);
2459
2460         netif_tx_stop_all_queues(netdev);
2461
2462         /* call carrier off first to avoid false dev_watchdog timeouts */
2463         netif_carrier_off(netdev);
2464         netif_tx_disable(netdev);
2465
2466         ixgbevf_irq_disable(adapter);
2467
2468         ixgbevf_napi_disable_all(adapter);
2469
2470         del_timer_sync(&adapter->service_timer);
2471
2472         /* disable transmits in the hardware now that interrupts are off */
2473         for (i = 0; i < adapter->num_tx_queues; i++) {
2474                 u8 reg_idx = adapter->tx_ring[i]->reg_idx;
2475
2476                 IXGBE_WRITE_REG(hw, IXGBE_VFTXDCTL(reg_idx),
2477                                 IXGBE_TXDCTL_SWFLSH);
2478         }
2479
2480         for (i = 0; i < adapter->num_xdp_queues; i++) {
2481                 u8 reg_idx = adapter->xdp_ring[i]->reg_idx;
2482
2483                 IXGBE_WRITE_REG(hw, IXGBE_VFTXDCTL(reg_idx),
2484                                 IXGBE_TXDCTL_SWFLSH);
2485         }
2486
2487         if (!pci_channel_offline(adapter->pdev))
2488                 ixgbevf_reset(adapter);
2489
2490         ixgbevf_clean_all_tx_rings(adapter);
2491         ixgbevf_clean_all_rx_rings(adapter);
2492 }
2493
2494 void ixgbevf_reinit_locked(struct ixgbevf_adapter *adapter)
2495 {
2496         WARN_ON(in_interrupt());
2497
2498         while (test_and_set_bit(__IXGBEVF_RESETTING, &adapter->state))
2499                 msleep(1);
2500
2501         ixgbevf_down(adapter);
2502         ixgbevf_up(adapter);
2503
2504         clear_bit(__IXGBEVF_RESETTING, &adapter->state);
2505 }
2506
2507 void ixgbevf_reset(struct ixgbevf_adapter *adapter)
2508 {
2509         struct ixgbe_hw *hw = &adapter->hw;
2510         struct net_device *netdev = adapter->netdev;
2511
2512         if (hw->mac.ops.reset_hw(hw)) {
2513                 hw_dbg(hw, "PF still resetting\n");
2514         } else {
2515                 hw->mac.ops.init_hw(hw);
2516                 ixgbevf_negotiate_api(adapter);
2517         }
2518
2519         if (is_valid_ether_addr(adapter->hw.mac.addr)) {
2520                 ether_addr_copy(netdev->dev_addr, adapter->hw.mac.addr);
2521                 ether_addr_copy(netdev->perm_addr, adapter->hw.mac.addr);
2522         }
2523
2524         adapter->last_reset = jiffies;
2525 }
2526
2527 static int ixgbevf_acquire_msix_vectors(struct ixgbevf_adapter *adapter,
2528                                         int vectors)
2529 {
2530         int vector_threshold;
2531
2532         /* We'll want at least 2 (vector_threshold):
2533          * 1) TxQ[0] + RxQ[0] handler
2534          * 2) Other (Link Status Change, etc.)
2535          */
2536         vector_threshold = MIN_MSIX_COUNT;
2537
2538         /* The more we get, the more we will assign to Tx/Rx Cleanup
2539          * for the separate queues...where Rx Cleanup >= Tx Cleanup.
2540          * Right now, we simply care about how many we'll get; we'll
2541          * set them up later while requesting irq's.
2542          */
2543         vectors = pci_enable_msix_range(adapter->pdev, adapter->msix_entries,
2544                                         vector_threshold, vectors);
2545
2546         if (vectors < 0) {
2547                 dev_err(&adapter->pdev->dev,
2548                         "Unable to allocate MSI-X interrupts\n");
2549                 kfree(adapter->msix_entries);
2550                 adapter->msix_entries = NULL;
2551                 return vectors;
2552         }
2553
2554         /* Adjust for only the vectors we'll use, which is minimum
2555          * of max_msix_q_vectors + NON_Q_VECTORS, or the number of
2556          * vectors we were allocated.
2557          */
2558         adapter->num_msix_vectors = vectors;
2559
2560         return 0;
2561 }
2562
2563 /**
2564  * ixgbevf_set_num_queues - Allocate queues for device, feature dependent
2565  * @adapter: board private structure to initialize
2566  *
2567  * This is the top level queue allocation routine.  The order here is very
2568  * important, starting with the "most" number of features turned on at once,
2569  * and ending with the smallest set of features.  This way large combinations
2570  * can be allocated if they're turned on, and smaller combinations are the
2571  * fallthrough conditions.
2572  *
2573  **/
2574 static void ixgbevf_set_num_queues(struct ixgbevf_adapter *adapter)
2575 {
2576         struct ixgbe_hw *hw = &adapter->hw;
2577         unsigned int def_q = 0;
2578         unsigned int num_tcs = 0;
2579         int err;
2580
2581         /* Start with base case */
2582         adapter->num_rx_queues = 1;
2583         adapter->num_tx_queues = 1;
2584         adapter->num_xdp_queues = 0;
2585
2586         spin_lock_bh(&adapter->mbx_lock);
2587
2588         /* fetch queue configuration from the PF */
2589         err = ixgbevf_get_queues(hw, &num_tcs, &def_q);
2590
2591         spin_unlock_bh(&adapter->mbx_lock);
2592
2593         if (err)
2594                 return;
2595
2596         /* we need as many queues as traffic classes */
2597         if (num_tcs > 1) {
2598                 adapter->num_rx_queues = num_tcs;
2599         } else {
2600                 u16 rss = min_t(u16, num_online_cpus(), IXGBEVF_MAX_RSS_QUEUES);
2601
2602                 switch (hw->api_version) {
2603                 case ixgbe_mbox_api_11:
2604                 case ixgbe_mbox_api_12:
2605                 case ixgbe_mbox_api_13:
2606                         if (adapter->xdp_prog &&
2607                             hw->mac.max_tx_queues == rss)
2608                                 rss = rss > 3 ? 2 : 1;
2609
2610                         adapter->num_rx_queues = rss;
2611                         adapter->num_tx_queues = rss;
2612                         adapter->num_xdp_queues = adapter->xdp_prog ? rss : 0;
2613                 default:
2614                         break;
2615                 }
2616         }
2617 }
2618
2619 /**
2620  * ixgbevf_set_interrupt_capability - set MSI-X or FAIL if not supported
2621  * @adapter: board private structure to initialize
2622  *
2623  * Attempt to configure the interrupts using the best available
2624  * capabilities of the hardware and the kernel.
2625  **/
2626 static int ixgbevf_set_interrupt_capability(struct ixgbevf_adapter *adapter)
2627 {
2628         int vector, v_budget;
2629
2630         /* It's easy to be greedy for MSI-X vectors, but it really
2631          * doesn't do us much good if we have a lot more vectors
2632          * than CPU's.  So let's be conservative and only ask for
2633          * (roughly) the same number of vectors as there are CPU's.
2634          * The default is to use pairs of vectors.
2635          */
2636         v_budget = max(adapter->num_rx_queues, adapter->num_tx_queues);
2637         v_budget = min_t(int, v_budget, num_online_cpus());
2638         v_budget += NON_Q_VECTORS;
2639
2640         adapter->msix_entries = kcalloc(v_budget,
2641                                         sizeof(struct msix_entry), GFP_KERNEL);
2642         if (!adapter->msix_entries)
2643                 return -ENOMEM;
2644
2645         for (vector = 0; vector < v_budget; vector++)
2646                 adapter->msix_entries[vector].entry = vector;
2647
2648         /* A failure in MSI-X entry allocation isn't fatal, but the VF driver
2649          * does not support any other modes, so we will simply fail here. Note
2650          * that we clean up the msix_entries pointer else-where.
2651          */
2652         return ixgbevf_acquire_msix_vectors(adapter, v_budget);
2653 }
2654
2655 static void ixgbevf_add_ring(struct ixgbevf_ring *ring,
2656                              struct ixgbevf_ring_container *head)
2657 {
2658         ring->next = head->ring;
2659         head->ring = ring;
2660         head->count++;
2661 }
2662
2663 /**
2664  * ixgbevf_alloc_q_vector - Allocate memory for a single interrupt vector
2665  * @adapter: board private structure to initialize
2666  * @v_idx: index of vector in adapter struct
2667  * @txr_count: number of Tx rings for q vector
2668  * @txr_idx: index of first Tx ring to assign
2669  * @xdp_count: total number of XDP rings to allocate
2670  * @xdp_idx: index of first XDP ring to allocate
2671  * @rxr_count: number of Rx rings for q vector
2672  * @rxr_idx: index of first Rx ring to assign
2673  *
2674  * We allocate one q_vector.  If allocation fails we return -ENOMEM.
2675  **/
2676 static int ixgbevf_alloc_q_vector(struct ixgbevf_adapter *adapter, int v_idx,
2677                                   int txr_count, int txr_idx,
2678                                   int xdp_count, int xdp_idx,
2679                                   int rxr_count, int rxr_idx)
2680 {
2681         struct ixgbevf_q_vector *q_vector;
2682         int reg_idx = txr_idx + xdp_idx;
2683         struct ixgbevf_ring *ring;
2684         int ring_count, size;
2685
2686         ring_count = txr_count + xdp_count + rxr_count;
2687         size = sizeof(*q_vector) + (sizeof(*ring) * ring_count);
2688
2689         /* allocate q_vector and rings */
2690         q_vector = kzalloc(size, GFP_KERNEL);
2691         if (!q_vector)
2692                 return -ENOMEM;
2693
2694         /* initialize NAPI */
2695         netif_napi_add(adapter->netdev, &q_vector->napi, ixgbevf_poll, 64);
2696
2697         /* tie q_vector and adapter together */
2698         adapter->q_vector[v_idx] = q_vector;
2699         q_vector->adapter = adapter;
2700         q_vector->v_idx = v_idx;
2701
2702         /* initialize pointer to rings */
2703         ring = q_vector->ring;
2704
2705         while (txr_count) {
2706                 /* assign generic ring traits */
2707                 ring->dev = &adapter->pdev->dev;
2708                 ring->netdev = adapter->netdev;
2709
2710                 /* configure backlink on ring */
2711                 ring->q_vector = q_vector;
2712
2713                 /* update q_vector Tx values */
2714                 ixgbevf_add_ring(ring, &q_vector->tx);
2715
2716                 /* apply Tx specific ring traits */
2717                 ring->count = adapter->tx_ring_count;
2718                 ring->queue_index = txr_idx;
2719                 ring->reg_idx = reg_idx;
2720
2721                 /* assign ring to adapter */
2722                  adapter->tx_ring[txr_idx] = ring;
2723
2724                 /* update count and index */
2725                 txr_count--;
2726                 txr_idx++;
2727                 reg_idx++;
2728
2729                 /* push pointer to next ring */
2730                 ring++;
2731         }
2732
2733         while (xdp_count) {
2734                 /* assign generic ring traits */
2735                 ring->dev = &adapter->pdev->dev;
2736                 ring->netdev = adapter->netdev;
2737
2738                 /* configure backlink on ring */
2739                 ring->q_vector = q_vector;
2740
2741                 /* update q_vector Tx values */
2742                 ixgbevf_add_ring(ring, &q_vector->tx);
2743
2744                 /* apply Tx specific ring traits */
2745                 ring->count = adapter->tx_ring_count;
2746                 ring->queue_index = xdp_idx;
2747                 ring->reg_idx = reg_idx;
2748                 set_ring_xdp(ring);
2749
2750                 /* assign ring to adapter */
2751                 adapter->xdp_ring[xdp_idx] = ring;
2752
2753                 /* update count and index */
2754                 xdp_count--;
2755                 xdp_idx++;
2756                 reg_idx++;
2757
2758                 /* push pointer to next ring */
2759                 ring++;
2760         }
2761
2762         while (rxr_count) {
2763                 /* assign generic ring traits */
2764                 ring->dev = &adapter->pdev->dev;
2765                 ring->netdev = adapter->netdev;
2766
2767                 /* configure backlink on ring */
2768                 ring->q_vector = q_vector;
2769
2770                 /* update q_vector Rx values */
2771                 ixgbevf_add_ring(ring, &q_vector->rx);
2772
2773                 /* apply Rx specific ring traits */
2774                 ring->count = adapter->rx_ring_count;
2775                 ring->queue_index = rxr_idx;
2776                 ring->reg_idx = rxr_idx;
2777
2778                 /* assign ring to adapter */
2779                 adapter->rx_ring[rxr_idx] = ring;
2780
2781                 /* update count and index */
2782                 rxr_count--;
2783                 rxr_idx++;
2784
2785                 /* push pointer to next ring */
2786                 ring++;
2787         }
2788
2789         return 0;
2790 }
2791
2792 /**
2793  * ixgbevf_free_q_vector - Free memory allocated for specific interrupt vector
2794  * @adapter: board private structure to initialize
2795  * @v_idx: index of vector in adapter struct
2796  *
2797  * This function frees the memory allocated to the q_vector.  In addition if
2798  * NAPI is enabled it will delete any references to the NAPI struct prior
2799  * to freeing the q_vector.
2800  **/
2801 static void ixgbevf_free_q_vector(struct ixgbevf_adapter *adapter, int v_idx)
2802 {
2803         struct ixgbevf_q_vector *q_vector = adapter->q_vector[v_idx];
2804         struct ixgbevf_ring *ring;
2805
2806         ixgbevf_for_each_ring(ring, q_vector->tx) {
2807                 if (ring_is_xdp(ring))
2808                         adapter->xdp_ring[ring->queue_index] = NULL;
2809                 else
2810                         adapter->tx_ring[ring->queue_index] = NULL;
2811         }
2812
2813         ixgbevf_for_each_ring(ring, q_vector->rx)
2814                 adapter->rx_ring[ring->queue_index] = NULL;
2815
2816         adapter->q_vector[v_idx] = NULL;
2817         netif_napi_del(&q_vector->napi);
2818
2819         /* ixgbevf_get_stats() might access the rings on this vector,
2820          * we must wait a grace period before freeing it.
2821          */
2822         kfree_rcu(q_vector, rcu);
2823 }
2824
2825 /**
2826  * ixgbevf_alloc_q_vectors - Allocate memory for interrupt vectors
2827  * @adapter: board private structure to initialize
2828  *
2829  * We allocate one q_vector per queue interrupt.  If allocation fails we
2830  * return -ENOMEM.
2831  **/
2832 static int ixgbevf_alloc_q_vectors(struct ixgbevf_adapter *adapter)
2833 {
2834         int q_vectors = adapter->num_msix_vectors - NON_Q_VECTORS;
2835         int rxr_remaining = adapter->num_rx_queues;
2836         int txr_remaining = adapter->num_tx_queues;
2837         int xdp_remaining = adapter->num_xdp_queues;
2838         int rxr_idx = 0, txr_idx = 0, xdp_idx = 0, v_idx = 0;
2839         int err;
2840
2841         if (q_vectors >= (rxr_remaining + txr_remaining + xdp_remaining)) {
2842                 for (; rxr_remaining; v_idx++, q_vectors--) {
2843                         int rqpv = DIV_ROUND_UP(rxr_remaining, q_vectors);
2844
2845                         err = ixgbevf_alloc_q_vector(adapter, v_idx,
2846                                                      0, 0, 0, 0, rqpv, rxr_idx);
2847                         if (err)
2848                                 goto err_out;
2849
2850                         /* update counts and index */
2851                         rxr_remaining -= rqpv;
2852                         rxr_idx += rqpv;
2853                 }
2854         }
2855
2856         for (; q_vectors; v_idx++, q_vectors--) {
2857                 int rqpv = DIV_ROUND_UP(rxr_remaining, q_vectors);
2858                 int tqpv = DIV_ROUND_UP(txr_remaining, q_vectors);
2859                 int xqpv = DIV_ROUND_UP(xdp_remaining, q_vectors);
2860
2861                 err = ixgbevf_alloc_q_vector(adapter, v_idx,
2862                                              tqpv, txr_idx,
2863                                              xqpv, xdp_idx,
2864                                              rqpv, rxr_idx);
2865
2866                 if (err)
2867                         goto err_out;
2868
2869                 /* update counts and index */
2870                 rxr_remaining -= rqpv;
2871                 rxr_idx += rqpv;
2872                 txr_remaining -= tqpv;
2873                 txr_idx += tqpv;
2874                 xdp_remaining -= xqpv;
2875                 xdp_idx += xqpv;
2876         }
2877
2878         return 0;
2879
2880 err_out:
2881         while (v_idx) {
2882                 v_idx--;
2883                 ixgbevf_free_q_vector(adapter, v_idx);
2884         }
2885
2886         return -ENOMEM;
2887 }
2888
2889 /**
2890  * ixgbevf_free_q_vectors - Free memory allocated for interrupt vectors
2891  * @adapter: board private structure to initialize
2892  *
2893  * This function frees the memory allocated to the q_vectors.  In addition if
2894  * NAPI is enabled it will delete any references to the NAPI struct prior
2895  * to freeing the q_vector.
2896  **/
2897 static void ixgbevf_free_q_vectors(struct ixgbevf_adapter *adapter)
2898 {
2899         int q_vectors = adapter->num_msix_vectors - NON_Q_VECTORS;
2900
2901         while (q_vectors) {
2902                 q_vectors--;
2903                 ixgbevf_free_q_vector(adapter, q_vectors);
2904         }
2905 }
2906
2907 /**
2908  * ixgbevf_reset_interrupt_capability - Reset MSIX setup
2909  * @adapter: board private structure
2910  *
2911  **/
2912 static void ixgbevf_reset_interrupt_capability(struct ixgbevf_adapter *adapter)
2913 {
2914         if (!adapter->msix_entries)
2915                 return;
2916
2917         pci_disable_msix(adapter->pdev);
2918         kfree(adapter->msix_entries);
2919         adapter->msix_entries = NULL;
2920 }
2921
2922 /**
2923  * ixgbevf_init_interrupt_scheme - Determine if MSIX is supported and init
2924  * @adapter: board private structure to initialize
2925  *
2926  **/
2927 static int ixgbevf_init_interrupt_scheme(struct ixgbevf_adapter *adapter)
2928 {
2929         int err;
2930
2931         /* Number of supported queues */
2932         ixgbevf_set_num_queues(adapter);
2933
2934         err = ixgbevf_set_interrupt_capability(adapter);
2935         if (err) {
2936                 hw_dbg(&adapter->hw,
2937                        "Unable to setup interrupt capabilities\n");
2938                 goto err_set_interrupt;
2939         }
2940
2941         err = ixgbevf_alloc_q_vectors(adapter);
2942         if (err) {
2943                 hw_dbg(&adapter->hw, "Unable to allocate memory for queue vectors\n");
2944                 goto err_alloc_q_vectors;
2945         }
2946
2947         hw_dbg(&adapter->hw, "Multiqueue %s: Rx Queue count = %u, Tx Queue count = %u XDP Queue count %u\n",
2948                (adapter->num_rx_queues > 1) ? "Enabled" : "Disabled",
2949                adapter->num_rx_queues, adapter->num_tx_queues,
2950                adapter->num_xdp_queues);
2951
2952         set_bit(__IXGBEVF_DOWN, &adapter->state);
2953
2954         return 0;
2955 err_alloc_q_vectors:
2956         ixgbevf_reset_interrupt_capability(adapter);
2957 err_set_interrupt:
2958         return err;
2959 }
2960
2961 /**
2962  * ixgbevf_clear_interrupt_scheme - Clear the current interrupt scheme settings
2963  * @adapter: board private structure to clear interrupt scheme on
2964  *
2965  * We go through and clear interrupt specific resources and reset the structure
2966  * to pre-load conditions
2967  **/
2968 static void ixgbevf_clear_interrupt_scheme(struct ixgbevf_adapter *adapter)
2969 {
2970         adapter->num_tx_queues = 0;
2971         adapter->num_xdp_queues = 0;
2972         adapter->num_rx_queues = 0;
2973
2974         ixgbevf_free_q_vectors(adapter);
2975         ixgbevf_reset_interrupt_capability(adapter);
2976 }
2977
2978 /**
2979  * ixgbevf_sw_init - Initialize general software structures
2980  * @adapter: board private structure to initialize
2981  *
2982  * ixgbevf_sw_init initializes the Adapter private data structure.
2983  * Fields are initialized based on PCI device information and
2984  * OS network device settings (MTU size).
2985  **/
2986 static int ixgbevf_sw_init(struct ixgbevf_adapter *adapter)
2987 {
2988         struct ixgbe_hw *hw = &adapter->hw;
2989         struct pci_dev *pdev = adapter->pdev;
2990         struct net_device *netdev = adapter->netdev;
2991         int err;
2992
2993         /* PCI config space info */
2994         hw->vendor_id = pdev->vendor;
2995         hw->device_id = pdev->device;
2996         hw->revision_id = pdev->revision;
2997         hw->subsystem_vendor_id = pdev->subsystem_vendor;
2998         hw->subsystem_device_id = pdev->subsystem_device;
2999
3000         hw->mbx.ops.init_params(hw);
3001
3002         if (hw->mac.type >= ixgbe_mac_X550_vf) {
3003                 err = ixgbevf_init_rss_key(adapter);
3004                 if (err)
3005                         goto out;
3006         }
3007
3008         /* assume legacy case in which PF would only give VF 2 queues */
3009         hw->mac.max_tx_queues = 2;
3010         hw->mac.max_rx_queues = 2;
3011
3012         /* lock to protect mailbox accesses */
3013         spin_lock_init(&adapter->mbx_lock);
3014
3015         err = hw->mac.ops.reset_hw(hw);
3016         if (err) {
3017                 dev_info(&pdev->dev,
3018                          "PF still in reset state.  Is the PF interface up?\n");
3019         } else {
3020                 err = hw->mac.ops.init_hw(hw);
3021                 if (err) {
3022                         pr_err("init_shared_code failed: %d\n", err);
3023                         goto out;
3024                 }
3025                 ixgbevf_negotiate_api(adapter);
3026                 err = hw->mac.ops.get_mac_addr(hw, hw->mac.addr);
3027                 if (err)
3028                         dev_info(&pdev->dev, "Error reading MAC address\n");
3029                 else if (is_zero_ether_addr(adapter->hw.mac.addr))
3030                         dev_info(&pdev->dev,
3031                                  "MAC address not assigned by administrator.\n");
3032                 ether_addr_copy(netdev->dev_addr, hw->mac.addr);
3033         }
3034
3035         if (!is_valid_ether_addr(netdev->dev_addr)) {
3036                 dev_info(&pdev->dev, "Assigning random MAC address\n");
3037                 eth_hw_addr_random(netdev);
3038                 ether_addr_copy(hw->mac.addr, netdev->dev_addr);
3039                 ether_addr_copy(hw->mac.perm_addr, netdev->dev_addr);
3040         }
3041
3042         /* Enable dynamic interrupt throttling rates */
3043         adapter->rx_itr_setting = 1;
3044         adapter->tx_itr_setting = 1;
3045
3046         /* set default ring sizes */
3047         adapter->tx_ring_count = IXGBEVF_DEFAULT_TXD;
3048         adapter->rx_ring_count = IXGBEVF_DEFAULT_RXD;
3049
3050         set_bit(__IXGBEVF_DOWN, &adapter->state);
3051         return 0;
3052
3053 out:
3054         return err;
3055 }
3056
3057 #define UPDATE_VF_COUNTER_32bit(reg, last_counter, counter)     \
3058         {                                                       \
3059                 u32 current_counter = IXGBE_READ_REG(hw, reg);  \
3060                 if (current_counter < last_counter)             \
3061                         counter += 0x100000000LL;               \
3062                 last_counter = current_counter;                 \
3063                 counter &= 0xFFFFFFFF00000000LL;                \
3064                 counter |= current_counter;                     \
3065         }
3066
3067 #define UPDATE_VF_COUNTER_36bit(reg_lsb, reg_msb, last_counter, counter) \
3068         {                                                                \
3069                 u64 current_counter_lsb = IXGBE_READ_REG(hw, reg_lsb);   \
3070                 u64 current_counter_msb = IXGBE_READ_REG(hw, reg_msb);   \
3071                 u64 current_counter = (current_counter_msb << 32) |      \
3072                         current_counter_lsb;                             \
3073                 if (current_counter < last_counter)                      \
3074                         counter += 0x1000000000LL;                       \
3075                 last_counter = current_counter;                          \
3076                 counter &= 0xFFFFFFF000000000LL;                         \
3077                 counter |= current_counter;                              \
3078         }
3079 /**
3080  * ixgbevf_update_stats - Update the board statistics counters.
3081  * @adapter: board private structure
3082  **/
3083 void ixgbevf_update_stats(struct ixgbevf_adapter *adapter)
3084 {
3085         struct ixgbe_hw *hw = &adapter->hw;
3086         u64 alloc_rx_page_failed = 0, alloc_rx_buff_failed = 0;
3087         u64 alloc_rx_page = 0, hw_csum_rx_error = 0;
3088         int i;
3089
3090         if (test_bit(__IXGBEVF_DOWN, &adapter->state) ||
3091             test_bit(__IXGBEVF_RESETTING, &adapter->state))
3092                 return;
3093
3094         UPDATE_VF_COUNTER_32bit(IXGBE_VFGPRC, adapter->stats.last_vfgprc,
3095                                 adapter->stats.vfgprc);
3096         UPDATE_VF_COUNTER_32bit(IXGBE_VFGPTC, adapter->stats.last_vfgptc,
3097                                 adapter->stats.vfgptc);
3098         UPDATE_VF_COUNTER_36bit(IXGBE_VFGORC_LSB, IXGBE_VFGORC_MSB,
3099                                 adapter->stats.last_vfgorc,
3100                                 adapter->stats.vfgorc);
3101         UPDATE_VF_COUNTER_36bit(IXGBE_VFGOTC_LSB, IXGBE_VFGOTC_MSB,
3102                                 adapter->stats.last_vfgotc,
3103                                 adapter->stats.vfgotc);
3104         UPDATE_VF_COUNTER_32bit(IXGBE_VFMPRC, adapter->stats.last_vfmprc,
3105                                 adapter->stats.vfmprc);
3106
3107         for (i = 0;  i  < adapter->num_rx_queues;  i++) {
3108                 struct ixgbevf_ring *rx_ring = adapter->rx_ring[i];
3109
3110                 hw_csum_rx_error += rx_ring->rx_stats.csum_err;
3111                 alloc_rx_page_failed += rx_ring->rx_stats.alloc_rx_page_failed;
3112                 alloc_rx_buff_failed += rx_ring->rx_stats.alloc_rx_buff_failed;
3113                 alloc_rx_page += rx_ring->rx_stats.alloc_rx_page;
3114         }
3115
3116         adapter->hw_csum_rx_error = hw_csum_rx_error;
3117         adapter->alloc_rx_page_failed = alloc_rx_page_failed;
3118         adapter->alloc_rx_buff_failed = alloc_rx_buff_failed;
3119         adapter->alloc_rx_page = alloc_rx_page;
3120 }
3121
3122 /**
3123  * ixgbevf_service_timer - Timer Call-back
3124  * @t: pointer to timer_list struct
3125  **/
3126 static void ixgbevf_service_timer(struct timer_list *t)
3127 {
3128         struct ixgbevf_adapter *adapter = from_timer(adapter, t,
3129                                                      service_timer);
3130
3131         /* Reset the timer */
3132         mod_timer(&adapter->service_timer, (HZ * 2) + jiffies);
3133
3134         ixgbevf_service_event_schedule(adapter);
3135 }
3136
3137 static void ixgbevf_reset_subtask(struct ixgbevf_adapter *adapter)
3138 {
3139         if (!test_and_clear_bit(__IXGBEVF_RESET_REQUESTED, &adapter->state))
3140                 return;
3141
3142         rtnl_lock();
3143         /* If we're already down or resetting, just bail */
3144         if (test_bit(__IXGBEVF_DOWN, &adapter->state) ||
3145             test_bit(__IXGBEVF_REMOVING, &adapter->state) ||
3146             test_bit(__IXGBEVF_RESETTING, &adapter->state)) {
3147                 rtnl_unlock();
3148                 return;
3149         }
3150
3151         adapter->tx_timeout_count++;
3152
3153         ixgbevf_reinit_locked(adapter);
3154         rtnl_unlock();
3155 }
3156
3157 /**
3158  * ixgbevf_check_hang_subtask - check for hung queues and dropped interrupts
3159  * @adapter: pointer to the device adapter structure
3160  *
3161  * This function serves two purposes.  First it strobes the interrupt lines
3162  * in order to make certain interrupts are occurring.  Secondly it sets the
3163  * bits needed to check for TX hangs.  As a result we should immediately
3164  * determine if a hang has occurred.
3165  **/
3166 static void ixgbevf_check_hang_subtask(struct ixgbevf_adapter *adapter)
3167 {
3168         struct ixgbe_hw *hw = &adapter->hw;
3169         u32 eics = 0;
3170         int i;
3171
3172         /* If we're down or resetting, just bail */
3173         if (test_bit(__IXGBEVF_DOWN, &adapter->state) ||
3174             test_bit(__IXGBEVF_RESETTING, &adapter->state))
3175                 return;
3176
3177         /* Force detection of hung controller */
3178         if (netif_carrier_ok(adapter->netdev)) {
3179                 for (i = 0; i < adapter->num_tx_queues; i++)
3180                         set_check_for_tx_hang(adapter->tx_ring[i]);
3181                 for (i = 0; i < adapter->num_xdp_queues; i++)
3182                         set_check_for_tx_hang(adapter->xdp_ring[i]);
3183         }
3184
3185         /* get one bit for every active Tx/Rx interrupt vector */
3186         for (i = 0; i < adapter->num_msix_vectors - NON_Q_VECTORS; i++) {
3187                 struct ixgbevf_q_vector *qv = adapter->q_vector[i];
3188
3189                 if (qv->rx.ring || qv->tx.ring)
3190                         eics |= BIT(i);
3191         }
3192
3193         /* Cause software interrupt to ensure rings are cleaned */
3194         IXGBE_WRITE_REG(hw, IXGBE_VTEICS, eics);
3195 }
3196
3197 /**
3198  * ixgbevf_watchdog_update_link - update the link status
3199  * @adapter: pointer to the device adapter structure
3200  **/
3201 static void ixgbevf_watchdog_update_link(struct ixgbevf_adapter *adapter)
3202 {
3203         struct ixgbe_hw *hw = &adapter->hw;
3204         u32 link_speed = adapter->link_speed;
3205         bool link_up = adapter->link_up;
3206         s32 err;
3207
3208         spin_lock_bh(&adapter->mbx_lock);
3209
3210         err = hw->mac.ops.check_link(hw, &link_speed, &link_up, false);
3211
3212         spin_unlock_bh(&adapter->mbx_lock);
3213
3214         /* if check for link returns error we will need to reset */
3215         if (err && time_after(jiffies, adapter->last_reset + (10 * HZ))) {
3216                 set_bit(__IXGBEVF_RESET_REQUESTED, &adapter->state);
3217                 link_up = false;
3218         }
3219
3220         adapter->link_up = link_up;
3221         adapter->link_speed = link_speed;
3222 }
3223
3224 /**
3225  * ixgbevf_watchdog_link_is_up - update netif_carrier status and
3226  *                               print link up message
3227  * @adapter: pointer to the device adapter structure
3228  **/
3229 static void ixgbevf_watchdog_link_is_up(struct ixgbevf_adapter *adapter)
3230 {
3231         struct net_device *netdev = adapter->netdev;
3232
3233         /* only continue if link was previously down */
3234         if (netif_carrier_ok(netdev))
3235                 return;
3236
3237         dev_info(&adapter->pdev->dev, "NIC Link is Up %s\n",
3238                  (adapter->link_speed == IXGBE_LINK_SPEED_10GB_FULL) ?
3239                  "10 Gbps" :
3240                  (adapter->link_speed == IXGBE_LINK_SPEED_1GB_FULL) ?
3241                  "1 Gbps" :
3242                  (adapter->link_speed == IXGBE_LINK_SPEED_100_FULL) ?
3243                  "100 Mbps" :
3244                  "unknown speed");
3245
3246         netif_carrier_on(netdev);
3247 }
3248
3249 /**
3250  * ixgbevf_watchdog_link_is_down - update netif_carrier status and
3251  *                                 print link down message
3252  * @adapter: pointer to the adapter structure
3253  **/
3254 static void ixgbevf_watchdog_link_is_down(struct ixgbevf_adapter *adapter)
3255 {
3256         struct net_device *netdev = adapter->netdev;
3257
3258         adapter->link_speed = 0;
3259
3260         /* only continue if link was up previously */
3261         if (!netif_carrier_ok(netdev))
3262                 return;
3263
3264         dev_info(&adapter->pdev->dev, "NIC Link is Down\n");
3265
3266         netif_carrier_off(netdev);
3267 }
3268
3269 /**
3270  * ixgbevf_watchdog_subtask - worker thread to bring link up
3271  * @adapter: board private structure
3272  **/
3273 static void ixgbevf_watchdog_subtask(struct ixgbevf_adapter *adapter)
3274 {
3275         /* if interface is down do nothing */
3276         if (test_bit(__IXGBEVF_DOWN, &adapter->state) ||
3277             test_bit(__IXGBEVF_RESETTING, &adapter->state))
3278                 return;
3279
3280         ixgbevf_watchdog_update_link(adapter);
3281
3282         if (adapter->link_up)
3283                 ixgbevf_watchdog_link_is_up(adapter);
3284         else
3285                 ixgbevf_watchdog_link_is_down(adapter);
3286
3287         ixgbevf_update_stats(adapter);
3288 }
3289
3290 /**
3291  * ixgbevf_service_task - manages and runs subtasks
3292  * @work: pointer to work_struct containing our data
3293  **/
3294 static void ixgbevf_service_task(struct work_struct *work)
3295 {
3296         struct ixgbevf_adapter *adapter = container_of(work,
3297                                                        struct ixgbevf_adapter,
3298                                                        service_task);
3299         struct ixgbe_hw *hw = &adapter->hw;
3300
3301         if (IXGBE_REMOVED(hw->hw_addr)) {
3302                 if (!test_bit(__IXGBEVF_DOWN, &adapter->state)) {
3303                         rtnl_lock();
3304                         ixgbevf_down(adapter);
3305                         rtnl_unlock();
3306                 }
3307                 return;
3308         }
3309
3310         ixgbevf_queue_reset_subtask(adapter);
3311         ixgbevf_reset_subtask(adapter);
3312         ixgbevf_watchdog_subtask(adapter);
3313         ixgbevf_check_hang_subtask(adapter);
3314
3315         ixgbevf_service_event_complete(adapter);
3316 }
3317
3318 /**
3319  * ixgbevf_free_tx_resources - Free Tx Resources per Queue
3320  * @tx_ring: Tx descriptor ring for a specific queue
3321  *
3322  * Free all transmit software resources
3323  **/
3324 void ixgbevf_free_tx_resources(struct ixgbevf_ring *tx_ring)
3325 {
3326         ixgbevf_clean_tx_ring(tx_ring);
3327
3328         vfree(tx_ring->tx_buffer_info);
3329         tx_ring->tx_buffer_info = NULL;
3330
3331         /* if not set, then don't free */
3332         if (!tx_ring->desc)
3333                 return;
3334
3335         dma_free_coherent(tx_ring->dev, tx_ring->size, tx_ring->desc,
3336                           tx_ring->dma);
3337
3338         tx_ring->desc = NULL;
3339 }
3340
3341 /**
3342  * ixgbevf_free_all_tx_resources - Free Tx Resources for All Queues
3343  * @adapter: board private structure
3344  *
3345  * Free all transmit software resources
3346  **/
3347 static void ixgbevf_free_all_tx_resources(struct ixgbevf_adapter *adapter)
3348 {
3349         int i;
3350
3351         for (i = 0; i < adapter->num_tx_queues; i++)
3352                 if (adapter->tx_ring[i]->desc)
3353                         ixgbevf_free_tx_resources(adapter->tx_ring[i]);
3354         for (i = 0; i < adapter->num_xdp_queues; i++)
3355                 if (adapter->xdp_ring[i]->desc)
3356                         ixgbevf_free_tx_resources(adapter->xdp_ring[i]);
3357 }
3358
3359 /**
3360  * ixgbevf_setup_tx_resources - allocate Tx resources (Descriptors)
3361  * @tx_ring: Tx descriptor ring (for a specific queue) to setup
3362  *
3363  * Return 0 on success, negative on failure
3364  **/
3365 int ixgbevf_setup_tx_resources(struct ixgbevf_ring *tx_ring)
3366 {
3367         struct ixgbevf_adapter *adapter = netdev_priv(tx_ring->netdev);
3368         int size;
3369
3370         size = sizeof(struct ixgbevf_tx_buffer) * tx_ring->count;
3371         tx_ring->tx_buffer_info = vmalloc(size);
3372         if (!tx_ring->tx_buffer_info)
3373                 goto err;
3374
3375         u64_stats_init(&tx_ring->syncp);
3376
3377         /* round up to nearest 4K */
3378         tx_ring->size = tx_ring->count * sizeof(union ixgbe_adv_tx_desc);
3379         tx_ring->size = ALIGN(tx_ring->size, 4096);
3380
3381         tx_ring->desc = dma_alloc_coherent(tx_ring->dev, tx_ring->size,
3382                                            &tx_ring->dma, GFP_KERNEL);
3383         if (!tx_ring->desc)
3384                 goto err;
3385
3386         return 0;
3387
3388 err:
3389         vfree(tx_ring->tx_buffer_info);
3390         tx_ring->tx_buffer_info = NULL;
3391         hw_dbg(&adapter->hw, "Unable to allocate memory for the transmit descriptor ring\n");
3392         return -ENOMEM;
3393 }
3394
3395 /**
3396  * ixgbevf_setup_all_tx_resources - allocate all queues Tx resources
3397  * @adapter: board private structure
3398  *
3399  * If this function returns with an error, then it's possible one or
3400  * more of the rings is populated (while the rest are not).  It is the
3401  * callers duty to clean those orphaned rings.
3402  *
3403  * Return 0 on success, negative on failure
3404  **/
3405 static int ixgbevf_setup_all_tx_resources(struct ixgbevf_adapter *adapter)
3406 {
3407         int i, j = 0, err = 0;
3408
3409         for (i = 0; i < adapter->num_tx_queues; i++) {
3410                 err = ixgbevf_setup_tx_resources(adapter->tx_ring[i]);
3411                 if (!err)
3412                         continue;
3413                 hw_dbg(&adapter->hw, "Allocation for Tx Queue %u failed\n", i);
3414                 goto err_setup_tx;
3415         }
3416
3417         for (j = 0; j < adapter->num_xdp_queues; j++) {
3418                 err = ixgbevf_setup_tx_resources(adapter->xdp_ring[j]);
3419                 if (!err)
3420                         continue;
3421                 hw_dbg(&adapter->hw, "Allocation for XDP Queue %u failed\n", j);
3422                 goto err_setup_tx;
3423         }
3424
3425         return 0;
3426 err_setup_tx:
3427         /* rewind the index freeing the rings as we go */
3428         while (j--)
3429                 ixgbevf_free_tx_resources(adapter->xdp_ring[j]);
3430         while (i--)
3431                 ixgbevf_free_tx_resources(adapter->tx_ring[i]);
3432
3433         return err;
3434 }
3435
3436 /**
3437  * ixgbevf_setup_rx_resources - allocate Rx resources (Descriptors)
3438  * @adapter: board private structure
3439  * @rx_ring: Rx descriptor ring (for a specific queue) to setup
3440  *
3441  * Returns 0 on success, negative on failure
3442  **/
3443 int ixgbevf_setup_rx_resources(struct ixgbevf_adapter *adapter,
3444                                struct ixgbevf_ring *rx_ring)
3445 {
3446         int size;
3447
3448         size = sizeof(struct ixgbevf_rx_buffer) * rx_ring->count;
3449         rx_ring->rx_buffer_info = vmalloc(size);
3450         if (!rx_ring->rx_buffer_info)
3451                 goto err;
3452
3453         u64_stats_init(&rx_ring->syncp);
3454
3455         /* Round up to nearest 4K */
3456         rx_ring->size = rx_ring->count * sizeof(union ixgbe_adv_rx_desc);
3457         rx_ring->size = ALIGN(rx_ring->size, 4096);
3458
3459         rx_ring->desc = dma_alloc_coherent(rx_ring->dev, rx_ring->size,
3460                                            &rx_ring->dma, GFP_KERNEL);
3461
3462         if (!rx_ring->desc)
3463                 goto err;
3464
3465         /* XDP RX-queue info */
3466         if (xdp_rxq_info_reg(&rx_ring->xdp_rxq, adapter->netdev,
3467                              rx_ring->queue_index) < 0)
3468                 goto err;
3469
3470         rx_ring->xdp_prog = adapter->xdp_prog;
3471
3472         return 0;
3473 err:
3474         vfree(rx_ring->rx_buffer_info);
3475         rx_ring->rx_buffer_info = NULL;
3476         dev_err(rx_ring->dev, "Unable to allocate memory for the Rx descriptor ring\n");
3477         return -ENOMEM;
3478 }
3479
3480 /**
3481  * ixgbevf_setup_all_rx_resources - allocate all queues Rx resources
3482  * @adapter: board private structure
3483  *
3484  * If this function returns with an error, then it's possible one or
3485  * more of the rings is populated (while the rest are not).  It is the
3486  * callers duty to clean those orphaned rings.
3487  *
3488  * Return 0 on success, negative on failure
3489  **/
3490 static int ixgbevf_setup_all_rx_resources(struct ixgbevf_adapter *adapter)
3491 {
3492         int i, err = 0;
3493
3494         for (i = 0; i < adapter->num_rx_queues; i++) {
3495                 err = ixgbevf_setup_rx_resources(adapter, adapter->rx_ring[i]);
3496                 if (!err)
3497                         continue;
3498                 hw_dbg(&adapter->hw, "Allocation for Rx Queue %u failed\n", i);
3499                 goto err_setup_rx;
3500         }
3501
3502         return 0;
3503 err_setup_rx:
3504         /* rewind the index freeing the rings as we go */
3505         while (i--)
3506                 ixgbevf_free_rx_resources(adapter->rx_ring[i]);
3507         return err;
3508 }
3509
3510 /**
3511  * ixgbevf_free_rx_resources - Free Rx Resources
3512  * @rx_ring: ring to clean the resources from
3513  *
3514  * Free all receive software resources
3515  **/
3516 void ixgbevf_free_rx_resources(struct ixgbevf_ring *rx_ring)
3517 {
3518         ixgbevf_clean_rx_ring(rx_ring);
3519
3520         rx_ring->xdp_prog = NULL;
3521         xdp_rxq_info_unreg(&rx_ring->xdp_rxq);
3522         vfree(rx_ring->rx_buffer_info);
3523         rx_ring->rx_buffer_info = NULL;
3524
3525         dma_free_coherent(rx_ring->dev, rx_ring->size, rx_ring->desc,
3526                           rx_ring->dma);
3527
3528         rx_ring->desc = NULL;
3529 }
3530
3531 /**
3532  * ixgbevf_free_all_rx_resources - Free Rx Resources for All Queues
3533  * @adapter: board private structure
3534  *
3535  * Free all receive software resources
3536  **/
3537 static void ixgbevf_free_all_rx_resources(struct ixgbevf_adapter *adapter)
3538 {
3539         int i;
3540
3541         for (i = 0; i < adapter->num_rx_queues; i++)
3542                 if (adapter->rx_ring[i]->desc)
3543                         ixgbevf_free_rx_resources(adapter->rx_ring[i]);
3544 }
3545
3546 /**
3547  * ixgbevf_open - Called when a network interface is made active
3548  * @netdev: network interface device structure
3549  *
3550  * Returns 0 on success, negative value on failure
3551  *
3552  * The open entry point is called when a network interface is made
3553  * active by the system (IFF_UP).  At this point all resources needed
3554  * for transmit and receive operations are allocated, the interrupt
3555  * handler is registered with the OS, the watchdog timer is started,
3556  * and the stack is notified that the interface is ready.
3557  **/
3558 int ixgbevf_open(struct net_device *netdev)
3559 {
3560         struct ixgbevf_adapter *adapter = netdev_priv(netdev);
3561         struct ixgbe_hw *hw = &adapter->hw;
3562         int err;
3563
3564         /* A previous failure to open the device because of a lack of
3565          * available MSIX vector resources may have reset the number
3566          * of msix vectors variable to zero.  The only way to recover
3567          * is to unload/reload the driver and hope that the system has
3568          * been able to recover some MSIX vector resources.
3569          */
3570         if (!adapter->num_msix_vectors)
3571                 return -ENOMEM;
3572
3573         if (hw->adapter_stopped) {
3574                 ixgbevf_reset(adapter);
3575                 /* if adapter is still stopped then PF isn't up and
3576                  * the VF can't start.
3577                  */
3578                 if (hw->adapter_stopped) {
3579                         err = IXGBE_ERR_MBX;
3580                         pr_err("Unable to start - perhaps the PF Driver isn't up yet\n");
3581                         goto err_setup_reset;
3582                 }
3583         }
3584
3585         /* disallow open during test */
3586         if (test_bit(__IXGBEVF_TESTING, &adapter->state))
3587                 return -EBUSY;
3588
3589         netif_carrier_off(netdev);
3590
3591         /* allocate transmit descriptors */
3592         err = ixgbevf_setup_all_tx_resources(adapter);
3593         if (err)
3594                 goto err_setup_tx;
3595
3596         /* allocate receive descriptors */
3597         err = ixgbevf_setup_all_rx_resources(adapter);
3598         if (err)
3599                 goto err_setup_rx;
3600
3601         ixgbevf_configure(adapter);
3602
3603         err = ixgbevf_request_irq(adapter);
3604         if (err)
3605                 goto err_req_irq;
3606
3607         /* Notify the stack of the actual queue counts. */
3608         err = netif_set_real_num_tx_queues(netdev, adapter->num_tx_queues);
3609         if (err)
3610                 goto err_set_queues;
3611
3612         err = netif_set_real_num_rx_queues(netdev, adapter->num_rx_queues);
3613         if (err)
3614                 goto err_set_queues;
3615
3616         ixgbevf_up_complete(adapter);
3617
3618         return 0;
3619
3620 err_set_queues:
3621         ixgbevf_free_irq(adapter);
3622 err_req_irq:
3623         ixgbevf_free_all_rx_resources(adapter);
3624 err_setup_rx:
3625         ixgbevf_free_all_tx_resources(adapter);
3626 err_setup_tx:
3627         ixgbevf_reset(adapter);
3628 err_setup_reset:
3629
3630         return err;
3631 }
3632
3633 /**
3634  * ixgbevf_close_suspend - actions necessary to both suspend and close flows
3635  * @adapter: the private adapter struct
3636  *
3637  * This function should contain the necessary work common to both suspending
3638  * and closing of the device.
3639  */
3640 static void ixgbevf_close_suspend(struct ixgbevf_adapter *adapter)
3641 {
3642         ixgbevf_down(adapter);
3643         ixgbevf_free_irq(adapter);
3644         ixgbevf_free_all_tx_resources(adapter);
3645         ixgbevf_free_all_rx_resources(adapter);
3646 }
3647
3648 /**
3649  * ixgbevf_close - Disables a network interface
3650  * @netdev: network interface device structure
3651  *
3652  * Returns 0, this is not allowed to fail
3653  *
3654  * The close entry point is called when an interface is de-activated
3655  * by the OS.  The hardware is still under the drivers control, but
3656  * needs to be disabled.  A global MAC reset is issued to stop the
3657  * hardware, and all transmit and receive resources are freed.
3658  **/
3659 int ixgbevf_close(struct net_device *netdev)
3660 {
3661         struct ixgbevf_adapter *adapter = netdev_priv(netdev);
3662
3663         if (netif_device_present(netdev))
3664                 ixgbevf_close_suspend(adapter);
3665
3666         return 0;
3667 }
3668
3669 static void ixgbevf_queue_reset_subtask(struct ixgbevf_adapter *adapter)
3670 {
3671         struct net_device *dev = adapter->netdev;
3672
3673         if (!test_and_clear_bit(__IXGBEVF_QUEUE_RESET_REQUESTED,
3674                                 &adapter->state))
3675                 return;
3676
3677         /* if interface is down do nothing */
3678         if (test_bit(__IXGBEVF_DOWN, &adapter->state) ||
3679             test_bit(__IXGBEVF_RESETTING, &adapter->state))
3680                 return;
3681
3682         /* Hardware has to reinitialize queues and interrupts to
3683          * match packet buffer alignment. Unfortunately, the
3684          * hardware is not flexible enough to do this dynamically.
3685          */
3686         rtnl_lock();
3687
3688         if (netif_running(dev))
3689                 ixgbevf_close(dev);
3690
3691         ixgbevf_clear_interrupt_scheme(adapter);
3692         ixgbevf_init_interrupt_scheme(adapter);
3693
3694         if (netif_running(dev))
3695                 ixgbevf_open(dev);
3696
3697         rtnl_unlock();
3698 }
3699
3700 static void ixgbevf_tx_ctxtdesc(struct ixgbevf_ring *tx_ring,
3701                                 u32 vlan_macip_lens, u32 type_tucmd,
3702                                 u32 mss_l4len_idx)
3703 {
3704         struct ixgbe_adv_tx_context_desc *context_desc;
3705         u16 i = tx_ring->next_to_use;
3706
3707         context_desc = IXGBEVF_TX_CTXTDESC(tx_ring, i);
3708
3709         i++;
3710         tx_ring->next_to_use = (i < tx_ring->count) ? i : 0;
3711
3712         /* set bits to identify this as an advanced context descriptor */
3713         type_tucmd |= IXGBE_TXD_CMD_DEXT | IXGBE_ADVTXD_DTYP_CTXT;
3714
3715         context_desc->vlan_macip_lens   = cpu_to_le32(vlan_macip_lens);
3716         context_desc->seqnum_seed       = 0;
3717         context_desc->type_tucmd_mlhl   = cpu_to_le32(type_tucmd);
3718         context_desc->mss_l4len_idx     = cpu_to_le32(mss_l4len_idx);
3719 }
3720
3721 static int ixgbevf_tso(struct ixgbevf_ring *tx_ring,
3722                        struct ixgbevf_tx_buffer *first,
3723                        u8 *hdr_len)
3724 {
3725         u32 vlan_macip_lens, type_tucmd, mss_l4len_idx;
3726         struct sk_buff *skb = first->skb;
3727         union {
3728                 struct iphdr *v4;
3729                 struct ipv6hdr *v6;
3730                 unsigned char *hdr;
3731         } ip;
3732         union {
3733                 struct tcphdr *tcp;
3734                 unsigned char *hdr;
3735         } l4;
3736         u32 paylen, l4_offset;
3737         int err;
3738
3739         if (skb->ip_summed != CHECKSUM_PARTIAL)
3740                 return 0;
3741
3742         if (!skb_is_gso(skb))
3743                 return 0;
3744
3745         err = skb_cow_head(skb, 0);
3746         if (err < 0)
3747                 return err;
3748
3749         if (eth_p_mpls(first->protocol))
3750                 ip.hdr = skb_inner_network_header(skb);
3751         else
3752                 ip.hdr = skb_network_header(skb);
3753         l4.hdr = skb_checksum_start(skb);
3754
3755         /* ADV DTYP TUCMD MKRLOC/ISCSIHEDLEN */
3756         type_tucmd = IXGBE_ADVTXD_TUCMD_L4T_TCP;
3757
3758         /* initialize outer IP header fields */
3759         if (ip.v4->version == 4) {
3760                 unsigned char *csum_start = skb_checksum_start(skb);
3761                 unsigned char *trans_start = ip.hdr + (ip.v4->ihl * 4);
3762
3763                 /* IP header will have to cancel out any data that
3764                  * is not a part of the outer IP header
3765                  */
3766                 ip.v4->check = csum_fold(csum_partial(trans_start,
3767                                                       csum_start - trans_start,
3768                                                       0));
3769                 type_tucmd |= IXGBE_ADVTXD_TUCMD_IPV4;
3770
3771                 ip.v4->tot_len = 0;
3772                 first->tx_flags |= IXGBE_TX_FLAGS_TSO |
3773                                    IXGBE_TX_FLAGS_CSUM |
3774                                    IXGBE_TX_FLAGS_IPV4;
3775         } else {
3776                 ip.v6->payload_len = 0;
3777                 first->tx_flags |= IXGBE_TX_FLAGS_TSO |
3778                                    IXGBE_TX_FLAGS_CSUM;
3779         }
3780
3781         /* determine offset of inner transport header */
3782         l4_offset = l4.hdr - skb->data;
3783
3784         /* compute length of segmentation header */
3785         *hdr_len = (l4.tcp->doff * 4) + l4_offset;
3786
3787         /* remove payload length from inner checksum */
3788         paylen = skb->len - l4_offset;
3789         csum_replace_by_diff(&l4.tcp->check, htonl(paylen));
3790
3791         /* update gso size and bytecount with header size */
3792         first->gso_segs = skb_shinfo(skb)->gso_segs;
3793         first->bytecount += (first->gso_segs - 1) * *hdr_len;
3794
3795         /* mss_l4len_id: use 1 as index for TSO */
3796         mss_l4len_idx = (*hdr_len - l4_offset) << IXGBE_ADVTXD_L4LEN_SHIFT;
3797         mss_l4len_idx |= skb_shinfo(skb)->gso_size << IXGBE_ADVTXD_MSS_SHIFT;
3798         mss_l4len_idx |= (1u << IXGBE_ADVTXD_IDX_SHIFT);
3799
3800         /* vlan_macip_lens: HEADLEN, MACLEN, VLAN tag */
3801         vlan_macip_lens = l4.hdr - ip.hdr;
3802         vlan_macip_lens |= (ip.hdr - skb->data) << IXGBE_ADVTXD_MACLEN_SHIFT;
3803         vlan_macip_lens |= first->tx_flags & IXGBE_TX_FLAGS_VLAN_MASK;
3804
3805         ixgbevf_tx_ctxtdesc(tx_ring, vlan_macip_lens,
3806                             type_tucmd, mss_l4len_idx);
3807
3808         return 1;
3809 }
3810
3811 static inline bool ixgbevf_ipv6_csum_is_sctp(struct sk_buff *skb)
3812 {
3813         unsigned int offset = 0;
3814
3815         ipv6_find_hdr(skb, &offset, IPPROTO_SCTP, NULL, NULL);
3816
3817         return offset == skb_checksum_start_offset(skb);
3818 }
3819
3820 static void ixgbevf_tx_csum(struct ixgbevf_ring *tx_ring,
3821                             struct ixgbevf_tx_buffer *first)
3822 {
3823         struct sk_buff *skb = first->skb;
3824         u32 vlan_macip_lens = 0;
3825         u32 type_tucmd = 0;
3826
3827         if (skb->ip_summed != CHECKSUM_PARTIAL)
3828                 goto no_csum;
3829
3830         switch (skb->csum_offset) {
3831         case offsetof(struct tcphdr, check):
3832                 type_tucmd = IXGBE_ADVTXD_TUCMD_L4T_TCP;
3833                 /* fall through */
3834         case offsetof(struct udphdr, check):
3835                 break;
3836         case offsetof(struct sctphdr, checksum):
3837                 /* validate that this is actually an SCTP request */
3838                 if (((first->protocol == htons(ETH_P_IP)) &&
3839                      (ip_hdr(skb)->protocol == IPPROTO_SCTP)) ||
3840                     ((first->protocol == htons(ETH_P_IPV6)) &&
3841                      ixgbevf_ipv6_csum_is_sctp(skb))) {
3842                         type_tucmd = IXGBE_ADVTXD_TUCMD_L4T_SCTP;
3843                         break;
3844                 }
3845                 /* fall through */
3846         default:
3847                 skb_checksum_help(skb);
3848                 goto no_csum;
3849         }
3850
3851         if (first->protocol == htons(ETH_P_IP))
3852                 type_tucmd |= IXGBE_ADVTXD_TUCMD_IPV4;
3853
3854         /* update TX checksum flag */
3855         first->tx_flags |= IXGBE_TX_FLAGS_CSUM;
3856         vlan_macip_lens = skb_checksum_start_offset(skb) -
3857                           skb_network_offset(skb);
3858 no_csum:
3859         /* vlan_macip_lens: MACLEN, VLAN tag */
3860         vlan_macip_lens |= skb_network_offset(skb) << IXGBE_ADVTXD_MACLEN_SHIFT;
3861         vlan_macip_lens |= first->tx_flags & IXGBE_TX_FLAGS_VLAN_MASK;
3862
3863         ixgbevf_tx_ctxtdesc(tx_ring, vlan_macip_lens, type_tucmd, 0);
3864 }
3865
3866 static __le32 ixgbevf_tx_cmd_type(u32 tx_flags)
3867 {
3868         /* set type for advanced descriptor with frame checksum insertion */
3869         __le32 cmd_type = cpu_to_le32(IXGBE_ADVTXD_DTYP_DATA |
3870                                       IXGBE_ADVTXD_DCMD_IFCS |
3871                                       IXGBE_ADVTXD_DCMD_DEXT);
3872
3873         /* set HW VLAN bit if VLAN is present */
3874         if (tx_flags & IXGBE_TX_FLAGS_VLAN)
3875                 cmd_type |= cpu_to_le32(IXGBE_ADVTXD_DCMD_VLE);
3876
3877         /* set segmentation enable bits for TSO/FSO */
3878         if (tx_flags & IXGBE_TX_FLAGS_TSO)
3879                 cmd_type |= cpu_to_le32(IXGBE_ADVTXD_DCMD_TSE);
3880
3881         return cmd_type;
3882 }
3883
3884 static void ixgbevf_tx_olinfo_status(union ixgbe_adv_tx_desc *tx_desc,
3885                                      u32 tx_flags, unsigned int paylen)
3886 {
3887         __le32 olinfo_status = cpu_to_le32(paylen << IXGBE_ADVTXD_PAYLEN_SHIFT);
3888
3889         /* enable L4 checksum for TSO and TX checksum offload */
3890         if (tx_flags & IXGBE_TX_FLAGS_CSUM)
3891                 olinfo_status |= cpu_to_le32(IXGBE_ADVTXD_POPTS_TXSM);
3892
3893         /* enble IPv4 checksum for TSO */
3894         if (tx_flags & IXGBE_TX_FLAGS_IPV4)
3895                 olinfo_status |= cpu_to_le32(IXGBE_ADVTXD_POPTS_IXSM);
3896
3897         /* use index 1 context for TSO/FSO/FCOE */
3898         if (tx_flags & IXGBE_TX_FLAGS_TSO)
3899                 olinfo_status |= cpu_to_le32(1u << IXGBE_ADVTXD_IDX_SHIFT);
3900
3901         /* Check Context must be set if Tx switch is enabled, which it
3902          * always is for case where virtual functions are running
3903          */
3904         olinfo_status |= cpu_to_le32(IXGBE_ADVTXD_CC);
3905
3906         tx_desc->read.olinfo_status = olinfo_status;
3907 }
3908
3909 static void ixgbevf_tx_map(struct ixgbevf_ring *tx_ring,
3910                            struct ixgbevf_tx_buffer *first,
3911                            const u8 hdr_len)
3912 {
3913         struct sk_buff *skb = first->skb;
3914         struct ixgbevf_tx_buffer *tx_buffer;
3915         union ixgbe_adv_tx_desc *tx_desc;
3916         struct skb_frag_struct *frag;
3917         dma_addr_t dma;
3918         unsigned int data_len, size;
3919         u32 tx_flags = first->tx_flags;
3920         __le32 cmd_type = ixgbevf_tx_cmd_type(tx_flags);
3921         u16 i = tx_ring->next_to_use;
3922
3923         tx_desc = IXGBEVF_TX_DESC(tx_ring, i);
3924
3925         ixgbevf_tx_olinfo_status(tx_desc, tx_flags, skb->len - hdr_len);
3926
3927         size = skb_headlen(skb);
3928         data_len = skb->data_len;
3929
3930         dma = dma_map_single(tx_ring->dev, skb->data, size, DMA_TO_DEVICE);
3931
3932         tx_buffer = first;
3933
3934         for (frag = &skb_shinfo(skb)->frags[0];; frag++) {
3935                 if (dma_mapping_error(tx_ring->dev, dma))
3936                         goto dma_error;
3937
3938                 /* record length, and DMA address */
3939                 dma_unmap_len_set(tx_buffer, len, size);
3940                 dma_unmap_addr_set(tx_buffer, dma, dma);
3941
3942                 tx_desc->read.buffer_addr = cpu_to_le64(dma);
3943
3944                 while (unlikely(size > IXGBE_MAX_DATA_PER_TXD)) {
3945                         tx_desc->read.cmd_type_len =
3946                                 cmd_type | cpu_to_le32(IXGBE_MAX_DATA_PER_TXD);
3947
3948                         i++;
3949                         tx_desc++;
3950                         if (i == tx_ring->count) {
3951                                 tx_desc = IXGBEVF_TX_DESC(tx_ring, 0);
3952                                 i = 0;
3953                         }
3954                         tx_desc->read.olinfo_status = 0;
3955
3956                         dma += IXGBE_MAX_DATA_PER_TXD;
3957                         size -= IXGBE_MAX_DATA_PER_TXD;
3958
3959                         tx_desc->read.buffer_addr = cpu_to_le64(dma);
3960                 }
3961
3962                 if (likely(!data_len))
3963                         break;
3964
3965                 tx_desc->read.cmd_type_len = cmd_type | cpu_to_le32(size);
3966
3967                 i++;
3968                 tx_desc++;
3969                 if (i == tx_ring->count) {
3970                         tx_desc = IXGBEVF_TX_DESC(tx_ring, 0);
3971                         i = 0;
3972                 }
3973                 tx_desc->read.olinfo_status = 0;
3974
3975                 size = skb_frag_size(frag);
3976                 data_len -= size;
3977
3978                 dma = skb_frag_dma_map(tx_ring->dev, frag, 0, size,
3979                                        DMA_TO_DEVICE);
3980
3981                 tx_buffer = &tx_ring->tx_buffer_info[i];
3982         }
3983
3984         /* write last descriptor with RS and EOP bits */
3985         cmd_type |= cpu_to_le32(size) | cpu_to_le32(IXGBE_TXD_CMD);
3986         tx_desc->read.cmd_type_len = cmd_type;
3987
3988         /* set the timestamp */
3989         first->time_stamp = jiffies;
3990
3991         /* Force memory writes to complete before letting h/w know there
3992          * are new descriptors to fetch.  (Only applicable for weak-ordered
3993          * memory model archs, such as IA-64).
3994          *
3995          * We also need this memory barrier (wmb) to make certain all of the
3996          * status bits have been updated before next_to_watch is written.
3997          */
3998         wmb();
3999
4000         /* set next_to_watch value indicating a packet is present */
4001         first->next_to_watch = tx_desc;
4002
4003         i++;
4004         if (i == tx_ring->count)
4005                 i = 0;
4006
4007         tx_ring->next_to_use = i;
4008
4009         /* notify HW of packet */
4010         ixgbevf_write_tail(tx_ring, i);
4011
4012         return;
4013 dma_error:
4014         dev_err(tx_ring->dev, "TX DMA map failed\n");
4015         tx_buffer = &tx_ring->tx_buffer_info[i];
4016
4017         /* clear dma mappings for failed tx_buffer_info map */
4018         while (tx_buffer != first) {
4019                 if (dma_unmap_len(tx_buffer, len))
4020                         dma_unmap_page(tx_ring->dev,
4021                                        dma_unmap_addr(tx_buffer, dma),
4022                                        dma_unmap_len(tx_buffer, len),
4023                                        DMA_TO_DEVICE);
4024                 dma_unmap_len_set(tx_buffer, len, 0);
4025
4026                 if (i-- == 0)
4027                         i += tx_ring->count;
4028                 tx_buffer = &tx_ring->tx_buffer_info[i];
4029         }
4030
4031         if (dma_unmap_len(tx_buffer, len))
4032                 dma_unmap_single(tx_ring->dev,
4033                                  dma_unmap_addr(tx_buffer, dma),
4034                                  dma_unmap_len(tx_buffer, len),
4035                                  DMA_TO_DEVICE);
4036         dma_unmap_len_set(tx_buffer, len, 0);
4037
4038         dev_kfree_skb_any(tx_buffer->skb);
4039         tx_buffer->skb = NULL;
4040
4041         tx_ring->next_to_use = i;
4042 }
4043
4044 static int __ixgbevf_maybe_stop_tx(struct ixgbevf_ring *tx_ring, int size)
4045 {
4046         netif_stop_subqueue(tx_ring->netdev, tx_ring->queue_index);
4047         /* Herbert's original patch had:
4048          *  smp_mb__after_netif_stop_queue();
4049          * but since that doesn't exist yet, just open code it.
4050          */
4051         smp_mb();
4052
4053         /* We need to check again in a case another CPU has just
4054          * made room available.
4055          */
4056         if (likely(ixgbevf_desc_unused(tx_ring) < size))
4057                 return -EBUSY;
4058
4059         /* A reprieve! - use start_queue because it doesn't call schedule */
4060         netif_start_subqueue(tx_ring->netdev, tx_ring->queue_index);
4061         ++tx_ring->tx_stats.restart_queue;
4062
4063         return 0;
4064 }
4065
4066 static int ixgbevf_maybe_stop_tx(struct ixgbevf_ring *tx_ring, int size)
4067 {
4068         if (likely(ixgbevf_desc_unused(tx_ring) >= size))
4069                 return 0;
4070         return __ixgbevf_maybe_stop_tx(tx_ring, size);
4071 }
4072
4073 static int ixgbevf_xmit_frame_ring(struct sk_buff *skb,
4074                                    struct ixgbevf_ring *tx_ring)
4075 {
4076         struct ixgbevf_tx_buffer *first;
4077         int tso;
4078         u32 tx_flags = 0;
4079         u16 count = TXD_USE_COUNT(skb_headlen(skb));
4080 #if PAGE_SIZE > IXGBE_MAX_DATA_PER_TXD
4081         unsigned short f;
4082 #endif
4083         u8 hdr_len = 0;
4084         u8 *dst_mac = skb_header_pointer(skb, 0, 0, NULL);
4085
4086         if (!dst_mac || is_link_local_ether_addr(dst_mac)) {
4087                 dev_kfree_skb_any(skb);
4088                 return NETDEV_TX_OK;
4089         }
4090
4091         /* need: 1 descriptor per page * PAGE_SIZE/IXGBE_MAX_DATA_PER_TXD,
4092          *       + 1 desc for skb_headlen/IXGBE_MAX_DATA_PER_TXD,
4093          *       + 2 desc gap to keep tail from touching head,
4094          *       + 1 desc for context descriptor,
4095          * otherwise try next time
4096          */
4097 #if PAGE_SIZE > IXGBE_MAX_DATA_PER_TXD
4098         for (f = 0; f < skb_shinfo(skb)->nr_frags; f++)
4099                 count += TXD_USE_COUNT(skb_shinfo(skb)->frags[f].size);
4100 #else
4101         count += skb_shinfo(skb)->nr_frags;
4102 #endif
4103         if (ixgbevf_maybe_stop_tx(tx_ring, count + 3)) {
4104                 tx_ring->tx_stats.tx_busy++;
4105                 return NETDEV_TX_BUSY;
4106         }
4107
4108         /* record the location of the first descriptor for this packet */
4109         first = &tx_ring->tx_buffer_info[tx_ring->next_to_use];
4110         first->skb = skb;
4111         first->bytecount = skb->len;
4112         first->gso_segs = 1;
4113
4114         if (skb_vlan_tag_present(skb)) {
4115                 tx_flags |= skb_vlan_tag_get(skb);
4116                 tx_flags <<= IXGBE_TX_FLAGS_VLAN_SHIFT;
4117                 tx_flags |= IXGBE_TX_FLAGS_VLAN;
4118         }
4119
4120         /* record initial flags and protocol */
4121         first->tx_flags = tx_flags;
4122         first->protocol = vlan_get_protocol(skb);
4123
4124         tso = ixgbevf_tso(tx_ring, first, &hdr_len);
4125         if (tso < 0)
4126                 goto out_drop;
4127         else if (!tso)
4128                 ixgbevf_tx_csum(tx_ring, first);
4129
4130         ixgbevf_tx_map(tx_ring, first, hdr_len);
4131
4132         ixgbevf_maybe_stop_tx(tx_ring, DESC_NEEDED);
4133
4134         return NETDEV_TX_OK;
4135
4136 out_drop:
4137         dev_kfree_skb_any(first->skb);
4138         first->skb = NULL;
4139
4140         return NETDEV_TX_OK;
4141 }
4142
4143 static netdev_tx_t ixgbevf_xmit_frame(struct sk_buff *skb, struct net_device *netdev)
4144 {
4145         struct ixgbevf_adapter *adapter = netdev_priv(netdev);
4146         struct ixgbevf_ring *tx_ring;
4147
4148         if (skb->len <= 0) {
4149                 dev_kfree_skb_any(skb);
4150                 return NETDEV_TX_OK;
4151         }
4152
4153         /* The minimum packet size for olinfo paylen is 17 so pad the skb
4154          * in order to meet this minimum size requirement.
4155          */
4156         if (skb->len < 17) {
4157                 if (skb_padto(skb, 17))
4158                         return NETDEV_TX_OK;
4159                 skb->len = 17;
4160         }
4161
4162         tx_ring = adapter->tx_ring[skb->queue_mapping];
4163         return ixgbevf_xmit_frame_ring(skb, tx_ring);
4164 }
4165
4166 /**
4167  * ixgbevf_set_mac - Change the Ethernet Address of the NIC
4168  * @netdev: network interface device structure
4169  * @p: pointer to an address structure
4170  *
4171  * Returns 0 on success, negative on failure
4172  **/
4173 static int ixgbevf_set_mac(struct net_device *netdev, void *p)
4174 {
4175         struct ixgbevf_adapter *adapter = netdev_priv(netdev);
4176         struct ixgbe_hw *hw = &adapter->hw;
4177         struct sockaddr *addr = p;
4178         int err;
4179
4180         if (!is_valid_ether_addr(addr->sa_data))
4181                 return -EADDRNOTAVAIL;
4182
4183         spin_lock_bh(&adapter->mbx_lock);
4184
4185         err = hw->mac.ops.set_rar(hw, 0, addr->sa_data, 0);
4186
4187         spin_unlock_bh(&adapter->mbx_lock);
4188
4189         if (err)
4190                 return -EPERM;
4191
4192         ether_addr_copy(hw->mac.addr, addr->sa_data);
4193         ether_addr_copy(hw->mac.perm_addr, addr->sa_data);
4194         ether_addr_copy(netdev->dev_addr, addr->sa_data);
4195
4196         return 0;
4197 }
4198
4199 /**
4200  * ixgbevf_change_mtu - Change the Maximum Transfer Unit
4201  * @netdev: network interface device structure
4202  * @new_mtu: new value for maximum frame size
4203  *
4204  * Returns 0 on success, negative on failure
4205  **/
4206 static int ixgbevf_change_mtu(struct net_device *netdev, int new_mtu)
4207 {
4208         struct ixgbevf_adapter *adapter = netdev_priv(netdev);
4209         struct ixgbe_hw *hw = &adapter->hw;
4210         int max_frame = new_mtu + ETH_HLEN + ETH_FCS_LEN;
4211         int ret;
4212
4213         /* prevent MTU being changed to a size unsupported by XDP */
4214         if (adapter->xdp_prog) {
4215                 dev_warn(&adapter->pdev->dev, "MTU cannot be changed while XDP program is loaded\n");
4216                 return -EPERM;
4217         }
4218
4219         spin_lock_bh(&adapter->mbx_lock);
4220         /* notify the PF of our intent to use this size of frame */
4221         ret = hw->mac.ops.set_rlpml(hw, max_frame);
4222         spin_unlock_bh(&adapter->mbx_lock);
4223         if (ret)
4224                 return -EINVAL;
4225
4226         hw_dbg(hw, "changing MTU from %d to %d\n",
4227                netdev->mtu, new_mtu);
4228
4229         /* must set new MTU before calling down or up */
4230         netdev->mtu = new_mtu;
4231
4232         if (netif_running(netdev))
4233                 ixgbevf_reinit_locked(adapter);
4234
4235         return 0;
4236 }
4237
4238 static int ixgbevf_suspend(struct pci_dev *pdev, pm_message_t state)
4239 {
4240         struct net_device *netdev = pci_get_drvdata(pdev);
4241         struct ixgbevf_adapter *adapter = netdev_priv(netdev);
4242 #ifdef CONFIG_PM
4243         int retval = 0;
4244 #endif
4245
4246         rtnl_lock();
4247         netif_device_detach(netdev);
4248
4249         if (netif_running(netdev))
4250                 ixgbevf_close_suspend(adapter);
4251
4252         ixgbevf_clear_interrupt_scheme(adapter);
4253         rtnl_unlock();
4254
4255 #ifdef CONFIG_PM
4256         retval = pci_save_state(pdev);
4257         if (retval)
4258                 return retval;
4259
4260 #endif
4261         if (!test_and_set_bit(__IXGBEVF_DISABLED, &adapter->state))
4262                 pci_disable_device(pdev);
4263
4264         return 0;
4265 }
4266
4267 #ifdef CONFIG_PM
4268 static int ixgbevf_resume(struct pci_dev *pdev)
4269 {
4270         struct net_device *netdev = pci_get_drvdata(pdev);
4271         struct ixgbevf_adapter *adapter = netdev_priv(netdev);
4272         u32 err;
4273
4274         pci_restore_state(pdev);
4275         /* pci_restore_state clears dev->state_saved so call
4276          * pci_save_state to restore it.
4277          */
4278         pci_save_state(pdev);
4279
4280         err = pci_enable_device_mem(pdev);
4281         if (err) {
4282                 dev_err(&pdev->dev, "Cannot enable PCI device from suspend\n");
4283                 return err;
4284         }
4285
4286         adapter->hw.hw_addr = adapter->io_addr;
4287         smp_mb__before_atomic();
4288         clear_bit(__IXGBEVF_DISABLED, &adapter->state);
4289         pci_set_master(pdev);
4290
4291         ixgbevf_reset(adapter);
4292
4293         rtnl_lock();
4294         err = ixgbevf_init_interrupt_scheme(adapter);
4295         if (!err && netif_running(netdev))
4296                 err = ixgbevf_open(netdev);
4297         rtnl_unlock();
4298         if (err)
4299                 return err;
4300
4301         netif_device_attach(netdev);
4302
4303         return err;
4304 }
4305
4306 #endif /* CONFIG_PM */
4307 static void ixgbevf_shutdown(struct pci_dev *pdev)
4308 {
4309         ixgbevf_suspend(pdev, PMSG_SUSPEND);
4310 }
4311
4312 static void ixgbevf_get_tx_ring_stats(struct rtnl_link_stats64 *stats,
4313                                       const struct ixgbevf_ring *ring)
4314 {
4315         u64 bytes, packets;
4316         unsigned int start;
4317
4318         if (ring) {
4319                 do {
4320                         start = u64_stats_fetch_begin_irq(&ring->syncp);
4321                         bytes = ring->stats.bytes;
4322                         packets = ring->stats.packets;
4323                 } while (u64_stats_fetch_retry_irq(&ring->syncp, start));
4324                 stats->tx_bytes += bytes;
4325                 stats->tx_packets += packets;
4326         }
4327 }
4328
4329 static void ixgbevf_get_stats(struct net_device *netdev,
4330                               struct rtnl_link_stats64 *stats)
4331 {
4332         struct ixgbevf_adapter *adapter = netdev_priv(netdev);
4333         unsigned int start;
4334         u64 bytes, packets;
4335         const struct ixgbevf_ring *ring;
4336         int i;
4337
4338         ixgbevf_update_stats(adapter);
4339
4340         stats->multicast = adapter->stats.vfmprc - adapter->stats.base_vfmprc;
4341
4342         rcu_read_lock();
4343         for (i = 0; i < adapter->num_rx_queues; i++) {
4344                 ring = adapter->rx_ring[i];
4345                 do {
4346                         start = u64_stats_fetch_begin_irq(&ring->syncp);
4347                         bytes = ring->stats.bytes;
4348                         packets = ring->stats.packets;
4349                 } while (u64_stats_fetch_retry_irq(&ring->syncp, start));
4350                 stats->rx_bytes += bytes;
4351                 stats->rx_packets += packets;
4352         }
4353
4354         for (i = 0; i < adapter->num_tx_queues; i++) {
4355                 ring = adapter->tx_ring[i];
4356                 ixgbevf_get_tx_ring_stats(stats, ring);
4357         }
4358
4359         for (i = 0; i < adapter->num_xdp_queues; i++) {
4360                 ring = adapter->xdp_ring[i];
4361                 ixgbevf_get_tx_ring_stats(stats, ring);
4362         }
4363         rcu_read_unlock();
4364 }
4365
4366 #define IXGBEVF_MAX_MAC_HDR_LEN         127
4367 #define IXGBEVF_MAX_NETWORK_HDR_LEN     511
4368
4369 static netdev_features_t
4370 ixgbevf_features_check(struct sk_buff *skb, struct net_device *dev,
4371                        netdev_features_t features)
4372 {
4373         unsigned int network_hdr_len, mac_hdr_len;
4374
4375         /* Make certain the headers can be described by a context descriptor */
4376         mac_hdr_len = skb_network_header(skb) - skb->data;
4377         if (unlikely(mac_hdr_len > IXGBEVF_MAX_MAC_HDR_LEN))
4378                 return features & ~(NETIF_F_HW_CSUM |
4379                                     NETIF_F_SCTP_CRC |
4380                                     NETIF_F_HW_VLAN_CTAG_TX |
4381                                     NETIF_F_TSO |
4382                                     NETIF_F_TSO6);
4383
4384         network_hdr_len = skb_checksum_start(skb) - skb_network_header(skb);
4385         if (unlikely(network_hdr_len >  IXGBEVF_MAX_NETWORK_HDR_LEN))
4386                 return features & ~(NETIF_F_HW_CSUM |
4387                                     NETIF_F_SCTP_CRC |
4388                                     NETIF_F_TSO |
4389                                     NETIF_F_TSO6);
4390
4391         /* We can only support IPV4 TSO in tunnels if we can mangle the
4392          * inner IP ID field, so strip TSO if MANGLEID is not supported.
4393          */
4394         if (skb->encapsulation && !(features & NETIF_F_TSO_MANGLEID))
4395                 features &= ~NETIF_F_TSO;
4396
4397         return features;
4398 }
4399
4400 static int ixgbevf_xdp_setup(struct net_device *dev, struct bpf_prog *prog)
4401 {
4402         int i, frame_size = dev->mtu + ETH_HLEN + ETH_FCS_LEN + VLAN_HLEN;
4403         struct ixgbevf_adapter *adapter = netdev_priv(dev);
4404         struct bpf_prog *old_prog;
4405
4406         /* verify ixgbevf ring attributes are sufficient for XDP */
4407         for (i = 0; i < adapter->num_rx_queues; i++) {
4408                 struct ixgbevf_ring *ring = adapter->rx_ring[i];
4409
4410                 if (frame_size > ixgbevf_rx_bufsz(ring))
4411                         return -EINVAL;
4412         }
4413
4414         old_prog = xchg(&adapter->xdp_prog, prog);
4415
4416         /* If transitioning XDP modes reconfigure rings */
4417         if (!!prog != !!old_prog) {
4418                 /* Hardware has to reinitialize queues and interrupts to
4419                  * match packet buffer alignment. Unfortunately, the
4420                  * hardware is not flexible enough to do this dynamically.
4421                  */
4422                 if (netif_running(dev))
4423                         ixgbevf_close(dev);
4424
4425                 ixgbevf_clear_interrupt_scheme(adapter);
4426                 ixgbevf_init_interrupt_scheme(adapter);
4427
4428                 if (netif_running(dev))
4429                         ixgbevf_open(dev);
4430         } else {
4431                 for (i = 0; i < adapter->num_rx_queues; i++)
4432                         xchg(&adapter->rx_ring[i]->xdp_prog, adapter->xdp_prog);
4433         }
4434
4435         if (old_prog)
4436                 bpf_prog_put(old_prog);
4437
4438         return 0;
4439 }
4440
4441 static int ixgbevf_xdp(struct net_device *dev, struct netdev_bpf *xdp)
4442 {
4443         struct ixgbevf_adapter *adapter = netdev_priv(dev);
4444
4445         switch (xdp->command) {
4446         case XDP_SETUP_PROG:
4447                 return ixgbevf_xdp_setup(dev, xdp->prog);
4448         case XDP_QUERY_PROG:
4449                 xdp->prog_id = adapter->xdp_prog ?
4450                                adapter->xdp_prog->aux->id : 0;
4451                 return 0;
4452         default:
4453                 return -EINVAL;
4454         }
4455 }
4456
4457 static const struct net_device_ops ixgbevf_netdev_ops = {
4458         .ndo_open               = ixgbevf_open,
4459         .ndo_stop               = ixgbevf_close,
4460         .ndo_start_xmit         = ixgbevf_xmit_frame,
4461         .ndo_set_rx_mode        = ixgbevf_set_rx_mode,
4462         .ndo_get_stats64        = ixgbevf_get_stats,
4463         .ndo_validate_addr      = eth_validate_addr,
4464         .ndo_set_mac_address    = ixgbevf_set_mac,
4465         .ndo_change_mtu         = ixgbevf_change_mtu,
4466         .ndo_tx_timeout         = ixgbevf_tx_timeout,
4467         .ndo_vlan_rx_add_vid    = ixgbevf_vlan_rx_add_vid,
4468         .ndo_vlan_rx_kill_vid   = ixgbevf_vlan_rx_kill_vid,
4469         .ndo_features_check     = ixgbevf_features_check,
4470         .ndo_bpf                = ixgbevf_xdp,
4471 };
4472
4473 static void ixgbevf_assign_netdev_ops(struct net_device *dev)
4474 {
4475         dev->netdev_ops = &ixgbevf_netdev_ops;
4476         ixgbevf_set_ethtool_ops(dev);
4477         dev->watchdog_timeo = 5 * HZ;
4478 }
4479
4480 /**
4481  * ixgbevf_probe - Device Initialization Routine
4482  * @pdev: PCI device information struct
4483  * @ent: entry in ixgbevf_pci_tbl
4484  *
4485  * Returns 0 on success, negative on failure
4486  *
4487  * ixgbevf_probe initializes an adapter identified by a pci_dev structure.
4488  * The OS initialization, configuring of the adapter private structure,
4489  * and a hardware reset occur.
4490  **/
4491 static int ixgbevf_probe(struct pci_dev *pdev, const struct pci_device_id *ent)
4492 {
4493         struct net_device *netdev;
4494         struct ixgbevf_adapter *adapter = NULL;
4495         struct ixgbe_hw *hw = NULL;
4496         const struct ixgbevf_info *ii = ixgbevf_info_tbl[ent->driver_data];
4497         int err, pci_using_dac;
4498         bool disable_dev = false;
4499
4500         err = pci_enable_device(pdev);
4501         if (err)
4502                 return err;
4503
4504         if (!dma_set_mask_and_coherent(&pdev->dev, DMA_BIT_MASK(64))) {
4505                 pci_using_dac = 1;
4506         } else {
4507                 err = dma_set_mask_and_coherent(&pdev->dev, DMA_BIT_MASK(32));
4508                 if (err) {
4509                         dev_err(&pdev->dev, "No usable DMA configuration, aborting\n");
4510                         goto err_dma;
4511                 }
4512                 pci_using_dac = 0;
4513         }
4514
4515         err = pci_request_regions(pdev, ixgbevf_driver_name);
4516         if (err) {
4517                 dev_err(&pdev->dev, "pci_request_regions failed 0x%x\n", err);
4518                 goto err_pci_reg;
4519         }
4520
4521         pci_set_master(pdev);
4522
4523         netdev = alloc_etherdev_mq(sizeof(struct ixgbevf_adapter),
4524                                    MAX_TX_QUEUES);
4525         if (!netdev) {
4526                 err = -ENOMEM;
4527                 goto err_alloc_etherdev;
4528         }
4529
4530         SET_NETDEV_DEV(netdev, &pdev->dev);
4531
4532         adapter = netdev_priv(netdev);
4533
4534         adapter->netdev = netdev;
4535         adapter->pdev = pdev;
4536         hw = &adapter->hw;
4537         hw->back = adapter;
4538         adapter->msg_enable = netif_msg_init(debug, DEFAULT_MSG_ENABLE);
4539
4540         /* call save state here in standalone driver because it relies on
4541          * adapter struct to exist, and needs to call netdev_priv
4542          */
4543         pci_save_state(pdev);
4544
4545         hw->hw_addr = ioremap(pci_resource_start(pdev, 0),
4546                               pci_resource_len(pdev, 0));
4547         adapter->io_addr = hw->hw_addr;
4548         if (!hw->hw_addr) {
4549                 err = -EIO;
4550                 goto err_ioremap;
4551         }
4552
4553         ixgbevf_assign_netdev_ops(netdev);
4554
4555         /* Setup HW API */
4556         memcpy(&hw->mac.ops, ii->mac_ops, sizeof(hw->mac.ops));
4557         hw->mac.type  = ii->mac;
4558
4559         memcpy(&hw->mbx.ops, &ixgbevf_mbx_ops,
4560                sizeof(struct ixgbe_mbx_operations));
4561
4562         /* setup the private structure */
4563         err = ixgbevf_sw_init(adapter);
4564         if (err)
4565                 goto err_sw_init;
4566
4567         /* The HW MAC address was set and/or determined in sw_init */
4568         if (!is_valid_ether_addr(netdev->dev_addr)) {
4569                 pr_err("invalid MAC address\n");
4570                 err = -EIO;
4571                 goto err_sw_init;
4572         }
4573
4574         netdev->hw_features = NETIF_F_SG |
4575                               NETIF_F_TSO |
4576                               NETIF_F_TSO6 |
4577                               NETIF_F_RXCSUM |
4578                               NETIF_F_HW_CSUM |
4579                               NETIF_F_SCTP_CRC;
4580
4581 #define IXGBEVF_GSO_PARTIAL_FEATURES (NETIF_F_GSO_GRE | \
4582                                       NETIF_F_GSO_GRE_CSUM | \
4583                                       NETIF_F_GSO_IPXIP4 | \
4584                                       NETIF_F_GSO_IPXIP6 | \
4585                                       NETIF_F_GSO_UDP_TUNNEL | \
4586                                       NETIF_F_GSO_UDP_TUNNEL_CSUM)
4587
4588         netdev->gso_partial_features = IXGBEVF_GSO_PARTIAL_FEATURES;
4589         netdev->hw_features |= NETIF_F_GSO_PARTIAL |
4590                                IXGBEVF_GSO_PARTIAL_FEATURES;
4591
4592         netdev->features = netdev->hw_features;
4593
4594         if (pci_using_dac)
4595                 netdev->features |= NETIF_F_HIGHDMA;
4596
4597         netdev->vlan_features |= netdev->features | NETIF_F_TSO_MANGLEID;
4598         netdev->mpls_features |= NETIF_F_SG |
4599                                  NETIF_F_TSO |
4600                                  NETIF_F_TSO6 |
4601                                  NETIF_F_HW_CSUM;
4602         netdev->mpls_features |= IXGBEVF_GSO_PARTIAL_FEATURES;
4603         netdev->hw_enc_features |= netdev->vlan_features;
4604
4605         /* set this bit last since it cannot be part of vlan_features */
4606         netdev->features |= NETIF_F_HW_VLAN_CTAG_FILTER |
4607                             NETIF_F_HW_VLAN_CTAG_RX |
4608                             NETIF_F_HW_VLAN_CTAG_TX;
4609
4610         netdev->priv_flags |= IFF_UNICAST_FLT;
4611
4612         /* MTU range: 68 - 1504 or 9710 */
4613         netdev->min_mtu = ETH_MIN_MTU;
4614         switch (adapter->hw.api_version) {
4615         case ixgbe_mbox_api_11:
4616         case ixgbe_mbox_api_12:
4617         case ixgbe_mbox_api_13:
4618                 netdev->max_mtu = IXGBE_MAX_JUMBO_FRAME_SIZE -
4619                                   (ETH_HLEN + ETH_FCS_LEN);
4620                 break;
4621         default:
4622                 if (adapter->hw.mac.type != ixgbe_mac_82599_vf)
4623                         netdev->max_mtu = IXGBE_MAX_JUMBO_FRAME_SIZE -
4624                                           (ETH_HLEN + ETH_FCS_LEN);
4625                 else
4626                         netdev->max_mtu = ETH_DATA_LEN + ETH_FCS_LEN;
4627                 break;
4628         }
4629
4630         if (IXGBE_REMOVED(hw->hw_addr)) {
4631                 err = -EIO;
4632                 goto err_sw_init;
4633         }
4634
4635         timer_setup(&adapter->service_timer, ixgbevf_service_timer, 0);
4636
4637         INIT_WORK(&adapter->service_task, ixgbevf_service_task);
4638         set_bit(__IXGBEVF_SERVICE_INITED, &adapter->state);
4639         clear_bit(__IXGBEVF_SERVICE_SCHED, &adapter->state);
4640
4641         err = ixgbevf_init_interrupt_scheme(adapter);
4642         if (err)
4643                 goto err_sw_init;
4644
4645         strcpy(netdev->name, "eth%d");
4646
4647         err = register_netdev(netdev);
4648         if (err)
4649                 goto err_register;
4650
4651         pci_set_drvdata(pdev, netdev);
4652         netif_carrier_off(netdev);
4653
4654         ixgbevf_init_last_counter_stats(adapter);
4655
4656         /* print the VF info */
4657         dev_info(&pdev->dev, "%pM\n", netdev->dev_addr);
4658         dev_info(&pdev->dev, "MAC: %d\n", hw->mac.type);
4659
4660         switch (hw->mac.type) {
4661         case ixgbe_mac_X550_vf:
4662                 dev_info(&pdev->dev, "Intel(R) X550 Virtual Function\n");
4663                 break;
4664         case ixgbe_mac_X540_vf:
4665                 dev_info(&pdev->dev, "Intel(R) X540 Virtual Function\n");
4666                 break;
4667         case ixgbe_mac_82599_vf:
4668         default:
4669                 dev_info(&pdev->dev, "Intel(R) 82599 Virtual Function\n");
4670                 break;
4671         }
4672
4673         return 0;
4674
4675 err_register:
4676         ixgbevf_clear_interrupt_scheme(adapter);
4677 err_sw_init:
4678         ixgbevf_reset_interrupt_capability(adapter);
4679         iounmap(adapter->io_addr);
4680         kfree(adapter->rss_key);
4681 err_ioremap:
4682         disable_dev = !test_and_set_bit(__IXGBEVF_DISABLED, &adapter->state);
4683         free_netdev(netdev);
4684 err_alloc_etherdev:
4685         pci_release_regions(pdev);
4686 err_pci_reg:
4687 err_dma:
4688         if (!adapter || disable_dev)
4689                 pci_disable_device(pdev);
4690         return err;
4691 }
4692
4693 /**
4694  * ixgbevf_remove - Device Removal Routine
4695  * @pdev: PCI device information struct
4696  *
4697  * ixgbevf_remove is called by the PCI subsystem to alert the driver
4698  * that it should release a PCI device.  The could be caused by a
4699  * Hot-Plug event, or because the driver is going to be removed from
4700  * memory.
4701  **/
4702 static void ixgbevf_remove(struct pci_dev *pdev)
4703 {
4704         struct net_device *netdev = pci_get_drvdata(pdev);
4705         struct ixgbevf_adapter *adapter;
4706         bool disable_dev;
4707
4708         if (!netdev)
4709                 return;
4710
4711         adapter = netdev_priv(netdev);
4712
4713         set_bit(__IXGBEVF_REMOVING, &adapter->state);
4714         cancel_work_sync(&adapter->service_task);
4715
4716         if (netdev->reg_state == NETREG_REGISTERED)
4717                 unregister_netdev(netdev);
4718
4719         ixgbevf_clear_interrupt_scheme(adapter);
4720         ixgbevf_reset_interrupt_capability(adapter);
4721
4722         iounmap(adapter->io_addr);
4723         pci_release_regions(pdev);
4724
4725         hw_dbg(&adapter->hw, "Remove complete\n");
4726
4727         kfree(adapter->rss_key);
4728         disable_dev = !test_and_set_bit(__IXGBEVF_DISABLED, &adapter->state);
4729         free_netdev(netdev);
4730
4731         if (disable_dev)
4732                 pci_disable_device(pdev);
4733 }
4734
4735 /**
4736  * ixgbevf_io_error_detected - called when PCI error is detected
4737  * @pdev: Pointer to PCI device
4738  * @state: The current pci connection state
4739  *
4740  * This function is called after a PCI bus error affecting
4741  * this device has been detected.
4742  **/
4743 static pci_ers_result_t ixgbevf_io_error_detected(struct pci_dev *pdev,
4744                                                   pci_channel_state_t state)
4745 {
4746         struct net_device *netdev = pci_get_drvdata(pdev);
4747         struct ixgbevf_adapter *adapter = netdev_priv(netdev);
4748
4749         if (!test_bit(__IXGBEVF_SERVICE_INITED, &adapter->state))
4750                 return PCI_ERS_RESULT_DISCONNECT;
4751
4752         rtnl_lock();
4753         netif_device_detach(netdev);
4754
4755         if (netif_running(netdev))
4756                 ixgbevf_close_suspend(adapter);
4757
4758         if (state == pci_channel_io_perm_failure) {
4759                 rtnl_unlock();
4760                 return PCI_ERS_RESULT_DISCONNECT;
4761         }
4762
4763         if (!test_and_set_bit(__IXGBEVF_DISABLED, &adapter->state))
4764                 pci_disable_device(pdev);
4765         rtnl_unlock();
4766
4767         /* Request a slot slot reset. */
4768         return PCI_ERS_RESULT_NEED_RESET;
4769 }
4770
4771 /**
4772  * ixgbevf_io_slot_reset - called after the pci bus has been reset.
4773  * @pdev: Pointer to PCI device
4774  *
4775  * Restart the card from scratch, as if from a cold-boot. Implementation
4776  * resembles the first-half of the ixgbevf_resume routine.
4777  **/
4778 static pci_ers_result_t ixgbevf_io_slot_reset(struct pci_dev *pdev)
4779 {
4780         struct net_device *netdev = pci_get_drvdata(pdev);
4781         struct ixgbevf_adapter *adapter = netdev_priv(netdev);
4782
4783         if (pci_enable_device_mem(pdev)) {
4784                 dev_err(&pdev->dev,
4785                         "Cannot re-enable PCI device after reset.\n");
4786                 return PCI_ERS_RESULT_DISCONNECT;
4787         }
4788
4789         adapter->hw.hw_addr = adapter->io_addr;
4790         smp_mb__before_atomic();
4791         clear_bit(__IXGBEVF_DISABLED, &adapter->state);
4792         pci_set_master(pdev);
4793
4794         ixgbevf_reset(adapter);
4795
4796         return PCI_ERS_RESULT_RECOVERED;
4797 }
4798
4799 /**
4800  * ixgbevf_io_resume - called when traffic can start flowing again.
4801  * @pdev: Pointer to PCI device
4802  *
4803  * This callback is called when the error recovery driver tells us that
4804  * its OK to resume normal operation. Implementation resembles the
4805  * second-half of the ixgbevf_resume routine.
4806  **/
4807 static void ixgbevf_io_resume(struct pci_dev *pdev)
4808 {
4809         struct net_device *netdev = pci_get_drvdata(pdev);
4810
4811         rtnl_lock();
4812         if (netif_running(netdev))
4813                 ixgbevf_open(netdev);
4814
4815         netif_device_attach(netdev);
4816         rtnl_unlock();
4817 }
4818
4819 /* PCI Error Recovery (ERS) */
4820 static const struct pci_error_handlers ixgbevf_err_handler = {
4821         .error_detected = ixgbevf_io_error_detected,
4822         .slot_reset = ixgbevf_io_slot_reset,
4823         .resume = ixgbevf_io_resume,
4824 };
4825
4826 static struct pci_driver ixgbevf_driver = {
4827         .name           = ixgbevf_driver_name,
4828         .id_table       = ixgbevf_pci_tbl,
4829         .probe          = ixgbevf_probe,
4830         .remove         = ixgbevf_remove,
4831 #ifdef CONFIG_PM
4832         /* Power Management Hooks */
4833         .suspend        = ixgbevf_suspend,
4834         .resume         = ixgbevf_resume,
4835 #endif
4836         .shutdown       = ixgbevf_shutdown,
4837         .err_handler    = &ixgbevf_err_handler
4838 };
4839
4840 /**
4841  * ixgbevf_init_module - Driver Registration Routine
4842  *
4843  * ixgbevf_init_module is the first routine called when the driver is
4844  * loaded. All it does is register with the PCI subsystem.
4845  **/
4846 static int __init ixgbevf_init_module(void)
4847 {
4848         pr_info("%s - version %s\n", ixgbevf_driver_string,
4849                 ixgbevf_driver_version);
4850
4851         pr_info("%s\n", ixgbevf_copyright);
4852         ixgbevf_wq = create_singlethread_workqueue(ixgbevf_driver_name);
4853         if (!ixgbevf_wq) {
4854                 pr_err("%s: Failed to create workqueue\n", ixgbevf_driver_name);
4855                 return -ENOMEM;
4856         }
4857
4858         return pci_register_driver(&ixgbevf_driver);
4859 }
4860
4861 module_init(ixgbevf_init_module);
4862
4863 /**
4864  * ixgbevf_exit_module - Driver Exit Cleanup Routine
4865  *
4866  * ixgbevf_exit_module is called just before the driver is removed
4867  * from memory.
4868  **/
4869 static void __exit ixgbevf_exit_module(void)
4870 {
4871         pci_unregister_driver(&ixgbevf_driver);
4872         if (ixgbevf_wq) {
4873                 destroy_workqueue(ixgbevf_wq);
4874                 ixgbevf_wq = NULL;
4875         }
4876 }
4877
4878 #ifdef DEBUG
4879 /**
4880  * ixgbevf_get_hw_dev_name - return device name string
4881  * used by hardware layer to print debugging information
4882  * @hw: pointer to private hardware struct
4883  **/
4884 char *ixgbevf_get_hw_dev_name(struct ixgbe_hw *hw)
4885 {
4886         struct ixgbevf_adapter *adapter = hw->back;
4887
4888         return adapter->netdev->name;
4889 }
4890
4891 #endif
4892 module_exit(ixgbevf_exit_module);
4893
4894 /* ixgbevf_main.c */