drivers/net/ethernet/intel/ixgbevf/ixgbevf

/* [<][>][^][v][top][bottom][index][help] */
This source file includes following definitions.
ixgbevf_service_event_schedule
ixgbevf_service_event_complete
ixgbevf_remove_adapter
ixgbevf_check_remove
ixgbevf_read_reg
ixgbevf_set_ivar
ixgbevf_get_tx_completed
ixgbevf_get_tx_pending
ixgbevf_check_tx_hang
ixgbevf_tx_timeout_reset
ixgbevf_tx_timeout
ixgbevf_clean_tx_irq
ixgbevf_rx_skb
ixgbevf_rx_hash
ixgbevf_rx_checksum
ixgbevf_process_skb_fields
ixgbevf_get_rx_buffer
ixgbevf_put_rx_buffer
ixgbevf_is_non_eop
ixgbevf_rx_offset
ixgbevf_alloc_mapped_page
ixgbevf_alloc_rx_buffers
ixgbevf_cleanup_headers
ixgbevf_reuse_rx_page
ixgbevf_page_is_reserved
ixgbevf_can_reuse_rx_page
ixgbevf_add_rx_frag
ixgbevf_construct_skb
ixgbevf_irq_enable_queues
ixgbevf_build_skb
ixgbevf_xmit_xdp_ring
ixgbevf_run_xdp
ixgbevf_rx_buffer_flip
ixgbevf_clean_rx_irq
ixgbevf_poll
ixgbevf_write_eitr
ixgbevf_configure_msix
ixgbevf_update_itr
ixgbevf_set_itr
ixgbevf_msix_other
ixgbevf_msix_clean_rings
ixgbevf_request_msix_irqs
ixgbevf_request_irq
ixgbevf_free_irq
ixgbevf_irq_disable
ixgbevf_irq_enable
ixgbevf_configure_tx_ring
ixgbevf_configure_tx
ixgbevf_configure_srrctl
ixgbevf_setup_psrtype
ixgbevf_disable_rx_queue
ixgbevf_rx_desc_queue_enable
ixgbevf_init_rss_key
ixgbevf_setup_vfmrqc
ixgbevf_configure_rx_ring
ixgbevf_set_rx_buffer_len
ixgbevf_configure_rx
ixgbevf_vlan_rx_add_vid
ixgbevf_vlan_rx_kill_vid
ixgbevf_restore_vlan
ixgbevf_write_uc_addr_list
ixgbevf_set_rx_mode
ixgbevf_napi_enable_all
ixgbevf_napi_disable_all
ixgbevf_configure_dcb
ixgbevf_configure
ixgbevf_save_reset_stats
ixgbevf_init_last_counter_stats
ixgbevf_negotiate_api
ixgbevf_up_complete
ixgbevf_up
ixgbevf_clean_rx_ring
ixgbevf_clean_tx_ring
ixgbevf_clean_all_rx_rings
ixgbevf_clean_all_tx_rings
ixgbevf_down
ixgbevf_reinit_locked
ixgbevf_reset
ixgbevf_acquire_msix_vectors
ixgbevf_set_num_queues
ixgbevf_set_interrupt_capability
ixgbevf_add_ring
ixgbevf_alloc_q_vector
ixgbevf_free_q_vector
ixgbevf_alloc_q_vectors
ixgbevf_free_q_vectors
ixgbevf_reset_interrupt_capability
ixgbevf_init_interrupt_scheme
ixgbevf_clear_interrupt_scheme
ixgbevf_sw_init
ixgbevf_update_stats
ixgbevf_service_timer
ixgbevf_reset_subtask
ixgbevf_check_hang_subtask
ixgbevf_watchdog_update_link
ixgbevf_watchdog_link_is_up
ixgbevf_watchdog_link_is_down
ixgbevf_watchdog_subtask
ixgbevf_service_task
ixgbevf_free_tx_resources
ixgbevf_free_all_tx_resources
ixgbevf_setup_tx_resources
ixgbevf_setup_all_tx_resources
ixgbevf_setup_rx_resources
ixgbevf_setup_all_rx_resources
ixgbevf_free_rx_resources
ixgbevf_free_all_rx_resources
ixgbevf_open
ixgbevf_close_suspend
ixgbevf_close
ixgbevf_queue_reset_subtask
ixgbevf_tx_ctxtdesc
ixgbevf_tso
ixgbevf_ipv6_csum_is_sctp
ixgbevf_tx_csum
ixgbevf_tx_cmd_type
ixgbevf_tx_olinfo_status
ixgbevf_tx_map
__ixgbevf_maybe_stop_tx
ixgbevf_maybe_stop_tx
ixgbevf_xmit_frame_ring
ixgbevf_xmit_frame
ixgbevf_set_mac
ixgbevf_change_mtu
ixgbevf_suspend
ixgbevf_resume
ixgbevf_shutdown
ixgbevf_get_tx_ring_stats
ixgbevf_get_stats
ixgbevf_features_check
ixgbevf_xdp_setup
ixgbevf_xdp
ixgbevf_assign_netdev_ops
ixgbevf_probe
ixgbevf_remove
ixgbevf_io_error_detected
ixgbevf_io_slot_reset
ixgbevf_io_resume
ixgbevf_init_module
ixgbevf_exit_module
ixgbevf_get_hw_dev_name
   1 // SPDX-License-Identifier: GPL-2.0
   2 /* Copyright(c) 1999 - 2018 Intel Corporation. */
   3 
   4 /******************************************************************************
   5  Copyright (c)2006 - 2007 Myricom, Inc. for some LRO specific code
   6 ******************************************************************************/
   7 
   8 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
   9 
  10 #include <linux/types.h>
  11 #include <linux/bitops.h>
  12 #include <linux/module.h>
  13 #include <linux/pci.h>
  14 #include <linux/netdevice.h>
  15 #include <linux/vmalloc.h>
  16 #include <linux/string.h>
  17 #include <linux/in.h>
  18 #include <linux/ip.h>
  19 #include <linux/tcp.h>
  20 #include <linux/sctp.h>
  21 #include <linux/ipv6.h>
  22 #include <linux/slab.h>
  23 #include <net/checksum.h>
  24 #include <net/ip6_checksum.h>
  25 #include <linux/ethtool.h>
  26 #include <linux/if.h>
  27 #include <linux/if_vlan.h>
  28 #include <linux/prefetch.h>
  29 #include <net/mpls.h>
  30 #include <linux/bpf.h>
  31 #include <linux/bpf_trace.h>
  32 #include <linux/atomic.h>
  33 #include <net/xfrm.h>
  34 
  35 #include "ixgbevf.h"
  36 
  37 const char ixgbevf_driver_name[] = "ixgbevf";
  38 static const char ixgbevf_driver_string[] =
  39         "Intel(R) 10 Gigabit PCI Express Virtual Function Network Driver";
  40 
  41 #define DRV_VERSION "4.1.0-k"
  42 const char ixgbevf_driver_version[] = DRV_VERSION;
  43 static char ixgbevf_copyright[] =
  44         "Copyright (c) 2009 - 2018 Intel Corporation.";
  45 
  46 static const struct ixgbevf_info *ixgbevf_info_tbl[] = {
  47         [board_82599_vf]        = &ixgbevf_82599_vf_info,
  48         [board_82599_vf_hv]     = &ixgbevf_82599_vf_hv_info,
  49         [board_X540_vf]         = &ixgbevf_X540_vf_info,
  50         [board_X540_vf_hv]      = &ixgbevf_X540_vf_hv_info,
  51         [board_X550_vf]         = &ixgbevf_X550_vf_info,
  52         [board_X550_vf_hv]      = &ixgbevf_X550_vf_hv_info,
  53         [board_X550EM_x_vf]     = &ixgbevf_X550EM_x_vf_info,
  54         [board_X550EM_x_vf_hv]  = &ixgbevf_X550EM_x_vf_hv_info,
  55         [board_x550em_a_vf]     = &ixgbevf_x550em_a_vf_info,
  56 };
  57 
  58 /* ixgbevf_pci_tbl - PCI Device ID Table
  59  *
  60  * Wildcard entries (PCI_ANY_ID) should come last
  61  * Last entry must be all 0s
  62  *
  63  * { Vendor ID, Device ID, SubVendor ID, SubDevice ID,
  64  *   Class, Class Mask, private data (not used) }
  65  */
  66 static const struct pci_device_id ixgbevf_pci_tbl[] = {
  67         {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82599_VF), board_82599_vf },
  68         {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82599_VF_HV), board_82599_vf_hv },
  69         {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_X540_VF), board_X540_vf },
  70         {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_X540_VF_HV), board_X540_vf_hv },
  71         {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_X550_VF), board_X550_vf },
  72         {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_X550_VF_HV), board_X550_vf_hv },
  73         {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_X550EM_X_VF), board_X550EM_x_vf },
  74         {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_X550EM_X_VF_HV), board_X550EM_x_vf_hv},
  75         {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_X550EM_A_VF), board_x550em_a_vf },
  76         /* required last entry */
  77         {0, }
  78 };
  79 MODULE_DEVICE_TABLE(pci, ixgbevf_pci_tbl);
  80 
  81 MODULE_AUTHOR("Intel Corporation, <linux.nics@intel.com>");
  82 MODULE_DESCRIPTION("Intel(R) 10 Gigabit Virtual Function Network Driver");
  83 MODULE_LICENSE("GPL v2");
  84 MODULE_VERSION(DRV_VERSION);
  85 
  86 #define DEFAULT_MSG_ENABLE (NETIF_MSG_DRV|NETIF_MSG_PROBE|NETIF_MSG_LINK)
  87 static int debug = -1;
  88 module_param(debug, int, 0);
  89 MODULE_PARM_DESC(debug, "Debug level (0=none,...,16=all)");
  90 
  91 static struct workqueue_struct *ixgbevf_wq;
  92 
  93 static void ixgbevf_service_event_schedule(struct ixgbevf_adapter *adapter)
  94 {
  95         if (!test_bit(__IXGBEVF_DOWN, &adapter->state) &&
  96             !test_bit(__IXGBEVF_REMOVING, &adapter->state) &&
  97             !test_and_set_bit(__IXGBEVF_SERVICE_SCHED, &adapter->state))
  98                 queue_work(ixgbevf_wq, &adapter->service_task);
  99 }
 100 
 101 static void ixgbevf_service_event_complete(struct ixgbevf_adapter *adapter)
 102 {
 103         BUG_ON(!test_bit(__IXGBEVF_SERVICE_SCHED, &adapter->state));
 104 
 105         /* flush memory to make sure state is correct before next watchdog */
 106         smp_mb__before_atomic();
 107         clear_bit(__IXGBEVF_SERVICE_SCHED, &adapter->state);
 108 }
 109 
 110 /* forward decls */
 111 static void ixgbevf_queue_reset_subtask(struct ixgbevf_adapter *adapter);
 112 static void ixgbevf_set_itr(struct ixgbevf_q_vector *q_vector);
 113 static void ixgbevf_free_all_rx_resources(struct ixgbevf_adapter *adapter);
 114 static bool ixgbevf_can_reuse_rx_page(struct ixgbevf_rx_buffer *rx_buffer);
 115 static void ixgbevf_reuse_rx_page(struct ixgbevf_ring *rx_ring,
 116                                   struct ixgbevf_rx_buffer *old_buff);
 117 
 118 static void ixgbevf_remove_adapter(struct ixgbe_hw *hw)
 119 {
 120         struct ixgbevf_adapter *adapter = hw->back;
 121 
 122         if (!hw->hw_addr)
 123                 return;
 124         hw->hw_addr = NULL;
 125         dev_err(&adapter->pdev->dev, "Adapter removed\n");
 126         if (test_bit(__IXGBEVF_SERVICE_INITED, &adapter->state))
 127                 ixgbevf_service_event_schedule(adapter);
 128 }
 129 
 130 static void ixgbevf_check_remove(struct ixgbe_hw *hw, u32 reg)
 131 {
 132         u32 value;
 133 
 134         /* The following check not only optimizes a bit by not
 135          * performing a read on the status register when the
 136          * register just read was a status register read that
 137          * returned IXGBE_FAILED_READ_REG. It also blocks any
 138          * potential recursion.
 139          */
 140         if (reg == IXGBE_VFSTATUS) {
 141                 ixgbevf_remove_adapter(hw);
 142                 return;
 143         }
 144         value = ixgbevf_read_reg(hw, IXGBE_VFSTATUS);
 145         if (value == IXGBE_FAILED_READ_REG)
 146                 ixgbevf_remove_adapter(hw);
 147 }
 148 
 149 u32 ixgbevf_read_reg(struct ixgbe_hw *hw, u32 reg)
 150 {
 151         u8 __iomem *reg_addr = READ_ONCE(hw->hw_addr);
 152         u32 value;
 153 
 154         if (IXGBE_REMOVED(reg_addr))
 155                 return IXGBE_FAILED_READ_REG;
 156         value = readl(reg_addr + reg);
 157         if (unlikely(value == IXGBE_FAILED_READ_REG))
 158                 ixgbevf_check_remove(hw, reg);
 159         return value;
 160 }
 161 
 162 /**
 163  * ixgbevf_set_ivar - set IVAR registers - maps interrupt causes to vectors
 164  * @adapter: pointer to adapter struct
 165  * @direction: 0 for Rx, 1 for Tx, -1 for other causes
 166  * @queue: queue to map the corresponding interrupt to
 167  * @msix_vector: the vector to map to the corresponding queue
 168  **/
 169 static void ixgbevf_set_ivar(struct ixgbevf_adapter *adapter, s8 direction,
 170                              u8 queue, u8 msix_vector)
 171 {
 172         u32 ivar, index;
 173         struct ixgbe_hw *hw = &adapter->hw;
 174 
 175         if (direction == -1) {
 176                 /* other causes */
 177                 msix_vector |= IXGBE_IVAR_ALLOC_VAL;
 178                 ivar = IXGBE_READ_REG(hw, IXGBE_VTIVAR_MISC);
 179                 ivar &= ~0xFF;
 180                 ivar |= msix_vector;
 181                 IXGBE_WRITE_REG(hw, IXGBE_VTIVAR_MISC, ivar);
 182         } else {
 183                 /* Tx or Rx causes */
 184                 msix_vector |= IXGBE_IVAR_ALLOC_VAL;
 185                 index = ((16 * (queue & 1)) + (8 * direction));
 186                 ivar = IXGBE_READ_REG(hw, IXGBE_VTIVAR(queue >> 1));
 187                 ivar &= ~(0xFF << index);
 188                 ivar |= (msix_vector << index);
 189                 IXGBE_WRITE_REG(hw, IXGBE_VTIVAR(queue >> 1), ivar);
 190         }
 191 }
 192 
 193 static u64 ixgbevf_get_tx_completed(struct ixgbevf_ring *ring)
 194 {
 195         return ring->stats.packets;
 196 }
 197 
 198 static u32 ixgbevf_get_tx_pending(struct ixgbevf_ring *ring)
 199 {
 200         struct ixgbevf_adapter *adapter = netdev_priv(ring->netdev);
 201         struct ixgbe_hw *hw = &adapter->hw;
 202 
 203         u32 head = IXGBE_READ_REG(hw, IXGBE_VFTDH(ring->reg_idx));
 204         u32 tail = IXGBE_READ_REG(hw, IXGBE_VFTDT(ring->reg_idx));
 205 
 206         if (head != tail)
 207                 return (head < tail) ?
 208                         tail - head : (tail + ring->count - head);
 209 
 210         return 0;
 211 }
 212 
 213 static inline bool ixgbevf_check_tx_hang(struct ixgbevf_ring *tx_ring)
 214 {
 215         u32 tx_done = ixgbevf_get_tx_completed(tx_ring);
 216         u32 tx_done_old = tx_ring->tx_stats.tx_done_old;
 217         u32 tx_pending = ixgbevf_get_tx_pending(tx_ring);
 218 
 219         clear_check_for_tx_hang(tx_ring);
 220 
 221         /* Check for a hung queue, but be thorough. This verifies
 222          * that a transmit has been completed since the previous
 223          * check AND there is at least one packet pending. The
 224          * ARMED bit is set to indicate a potential hang.
 225          */
 226         if ((tx_done_old == tx_done) && tx_pending) {
 227                 /* make sure it is true for two checks in a row */
 228                 return test_and_set_bit(__IXGBEVF_HANG_CHECK_ARMED,
 229                                         &tx_ring->state);
 230         }
 231         /* reset the countdown */
 232         clear_bit(__IXGBEVF_HANG_CHECK_ARMED, &tx_ring->state);
 233 
 234         /* update completed stats and continue */
 235         tx_ring->tx_stats.tx_done_old = tx_done;
 236 
 237         return false;
 238 }
 239 
 240 static void ixgbevf_tx_timeout_reset(struct ixgbevf_adapter *adapter)
 241 {
 242         /* Do the reset outside of interrupt context */
 243         if (!test_bit(__IXGBEVF_DOWN, &adapter->state)) {
 244                 set_bit(__IXGBEVF_RESET_REQUESTED, &adapter->state);
 245                 ixgbevf_service_event_schedule(adapter);
 246         }
 247 }
 248 
 249 /**
 250  * ixgbevf_tx_timeout - Respond to a Tx Hang
 251  * @netdev: network interface device structure
 252  **/
 253 static void ixgbevf_tx_timeout(struct net_device *netdev)
 254 {
 255         struct ixgbevf_adapter *adapter = netdev_priv(netdev);
 256 
 257         ixgbevf_tx_timeout_reset(adapter);
 258 }
 259 
 260 /**
 261  * ixgbevf_clean_tx_irq - Reclaim resources after transmit completes
 262  * @q_vector: board private structure
 263  * @tx_ring: tx ring to clean
 264  * @napi_budget: Used to determine if we are in netpoll
 265  **/
 266 static bool ixgbevf_clean_tx_irq(struct ixgbevf_q_vector *q_vector,
 267                                  struct ixgbevf_ring *tx_ring, int napi_budget)
 268 {
 269         struct ixgbevf_adapter *adapter = q_vector->adapter;
 270         struct ixgbevf_tx_buffer *tx_buffer;
 271         union ixgbe_adv_tx_desc *tx_desc;
 272         unsigned int total_bytes = 0, total_packets = 0, total_ipsec = 0;
 273         unsigned int budget = tx_ring->count / 2;
 274         unsigned int i = tx_ring->next_to_clean;
 275 
 276         if (test_bit(__IXGBEVF_DOWN, &adapter->state))
 277                 return true;
 278 
 279         tx_buffer = &tx_ring->tx_buffer_info[i];
 280         tx_desc = IXGBEVF_TX_DESC(tx_ring, i);
 281         i -= tx_ring->count;
 282 
 283         do {
 284                 union ixgbe_adv_tx_desc *eop_desc = tx_buffer->next_to_watch;
 285 
 286                 /* if next_to_watch is not set then there is no work pending */
 287                 if (!eop_desc)
 288                         break;
 289 
 290                 /* prevent any other reads prior to eop_desc */
 291                 smp_rmb();
 292 
 293                 /* if DD is not set pending work has not been completed */
 294                 if (!(eop_desc->wb.status & cpu_to_le32(IXGBE_TXD_STAT_DD)))
 295                         break;
 296 
 297                 /* clear next_to_watch to prevent false hangs */
 298                 tx_buffer->next_to_watch = NULL;
 299 
 300                 /* update the statistics for this packet */
 301                 total_bytes += tx_buffer->bytecount;
 302                 total_packets += tx_buffer->gso_segs;
 303                 if (tx_buffer->tx_flags & IXGBE_TX_FLAGS_IPSEC)
 304                         total_ipsec++;
 305 
 306                 /* free the skb */
 307                 if (ring_is_xdp(tx_ring))
 308                         page_frag_free(tx_buffer->data);
 309                 else
 310                         napi_consume_skb(tx_buffer->skb, napi_budget);
 311 
 312                 /* unmap skb header data */
 313                 dma_unmap_single(tx_ring->dev,
 314                                  dma_unmap_addr(tx_buffer, dma),
 315                                  dma_unmap_len(tx_buffer, len),
 316                                  DMA_TO_DEVICE);
 317 
 318                 /* clear tx_buffer data */
 319                 dma_unmap_len_set(tx_buffer, len, 0);
 320 
 321                 /* unmap remaining buffers */
 322                 while (tx_desc != eop_desc) {
 323                         tx_buffer++;
 324                         tx_desc++;
 325                         i++;
 326                         if (unlikely(!i)) {
 327                                 i -= tx_ring->count;
 328                                 tx_buffer = tx_ring->tx_buffer_info;
 329                                 tx_desc = IXGBEVF_TX_DESC(tx_ring, 0);
 330                         }
 331 
 332                         /* unmap any remaining paged data */
 333                         if (dma_unmap_len(tx_buffer, len)) {
 334                                 dma_unmap_page(tx_ring->dev,
 335                                                dma_unmap_addr(tx_buffer, dma),
 336                                                dma_unmap_len(tx_buffer, len),
 337                                                DMA_TO_DEVICE);
 338                                 dma_unmap_len_set(tx_buffer, len, 0);
 339                         }
 340                 }
 341 
 342                 /* move us one more past the eop_desc for start of next pkt */
 343                 tx_buffer++;
 344                 tx_desc++;
 345                 i++;
 346                 if (unlikely(!i)) {
 347                         i -= tx_ring->count;
 348                         tx_buffer = tx_ring->tx_buffer_info;
 349                         tx_desc = IXGBEVF_TX_DESC(tx_ring, 0);
 350                 }
 351 
 352                 /* issue prefetch for next Tx descriptor */
 353                 prefetch(tx_desc);
 354 
 355                 /* update budget accounting */
 356                 budget--;
 357         } while (likely(budget));
 358 
 359         i += tx_ring->count;
 360         tx_ring->next_to_clean = i;
 361         u64_stats_update_begin(&tx_ring->syncp);
 362         tx_ring->stats.bytes += total_bytes;
 363         tx_ring->stats.packets += total_packets;
 364         u64_stats_update_end(&tx_ring->syncp);
 365         q_vector->tx.total_bytes += total_bytes;
 366         q_vector->tx.total_packets += total_packets;
 367         adapter->tx_ipsec += total_ipsec;
 368 
 369         if (check_for_tx_hang(tx_ring) && ixgbevf_check_tx_hang(tx_ring)) {
 370                 struct ixgbe_hw *hw = &adapter->hw;
 371                 union ixgbe_adv_tx_desc *eop_desc;
 372 
 373                 eop_desc = tx_ring->tx_buffer_info[i].next_to_watch;
 374 
 375                 pr_err("Detected Tx Unit Hang%s\n"
 376                        "  Tx Queue             <%d>\n"
 377                        "  TDH, TDT             <%x>, <%x>\n"
 378                        "  next_to_use          <%x>\n"
 379                        "  next_to_clean        <%x>\n"
 380                        "tx_buffer_info[next_to_clean]\n"
 381                        "  next_to_watch        <%p>\n"
 382                        "  eop_desc->wb.status  <%x>\n"
 383                        "  time_stamp           <%lx>\n"
 384                        "  jiffies              <%lx>\n",
 385                        ring_is_xdp(tx_ring) ? " XDP" : "",
 386                        tx_ring->queue_index,
 387                        IXGBE_READ_REG(hw, IXGBE_VFTDH(tx_ring->reg_idx)),
 388                        IXGBE_READ_REG(hw, IXGBE_VFTDT(tx_ring->reg_idx)),
 389                        tx_ring->next_to_use, i,
 390                        eop_desc, (eop_desc ? eop_desc->wb.status : 0),
 391                        tx_ring->tx_buffer_info[i].time_stamp, jiffies);
 392 
 393                 if (!ring_is_xdp(tx_ring))
 394                         netif_stop_subqueue(tx_ring->netdev,
 395                                             tx_ring->queue_index);
 396 
 397                 /* schedule immediate reset if we believe we hung */
 398                 ixgbevf_tx_timeout_reset(adapter);
 399 
 400                 return true;
 401         }
 402 
 403         if (ring_is_xdp(tx_ring))
 404                 return !!budget;
 405 
 406 #define TX_WAKE_THRESHOLD (DESC_NEEDED * 2)
 407         if (unlikely(total_packets && netif_carrier_ok(tx_ring->netdev) &&
 408                      (ixgbevf_desc_unused(tx_ring) >= TX_WAKE_THRESHOLD))) {
 409                 /* Make sure that anybody stopping the queue after this
 410                  * sees the new next_to_clean.
 411                  */
 412                 smp_mb();
 413 
 414                 if (__netif_subqueue_stopped(tx_ring->netdev,
 415                                              tx_ring->queue_index) &&
 416                     !test_bit(__IXGBEVF_DOWN, &adapter->state)) {
 417                         netif_wake_subqueue(tx_ring->netdev,
 418                                             tx_ring->queue_index);
 419                         ++tx_ring->tx_stats.restart_queue;
 420                 }
 421         }
 422 
 423         return !!budget;
 424 }
 425 
 426 /**
 427  * ixgbevf_rx_skb - Helper function to determine proper Rx method
 428  * @q_vector: structure containing interrupt and ring information
 429  * @skb: packet to send up
 430  **/
 431 static void ixgbevf_rx_skb(struct ixgbevf_q_vector *q_vector,
 432                            struct sk_buff *skb)
 433 {
 434         napi_gro_receive(&q_vector->napi, skb);
 435 }
 436 
 437 #define IXGBE_RSS_L4_TYPES_MASK \
 438         ((1ul << IXGBE_RXDADV_RSSTYPE_IPV4_TCP) | \
 439          (1ul << IXGBE_RXDADV_RSSTYPE_IPV4_UDP) | \
 440          (1ul << IXGBE_RXDADV_RSSTYPE_IPV6_TCP) | \
 441          (1ul << IXGBE_RXDADV_RSSTYPE_IPV6_UDP))
 442 
 443 static inline void ixgbevf_rx_hash(struct ixgbevf_ring *ring,
 444                                    union ixgbe_adv_rx_desc *rx_desc,
 445                                    struct sk_buff *skb)
 446 {
 447         u16 rss_type;
 448 
 449         if (!(ring->netdev->features & NETIF_F_RXHASH))
 450                 return;
 451 
 452         rss_type = le16_to_cpu(rx_desc->wb.lower.lo_dword.hs_rss.pkt_info) &
 453                    IXGBE_RXDADV_RSSTYPE_MASK;
 454 
 455         if (!rss_type)
 456                 return;
 457 
 458         skb_set_hash(skb, le32_to_cpu(rx_desc->wb.lower.hi_dword.rss),
 459                      (IXGBE_RSS_L4_TYPES_MASK & (1ul << rss_type)) ?
 460                      PKT_HASH_TYPE_L4 : PKT_HASH_TYPE_L3);
 461 }
 462 
 463 /**
 464  * ixgbevf_rx_checksum - indicate in skb if hw indicated a good cksum
 465  * @ring: structure containig ring specific data
 466  * @rx_desc: current Rx descriptor being processed
 467  * @skb: skb currently being received and modified
 468  **/
 469 static inline void ixgbevf_rx_checksum(struct ixgbevf_ring *ring,
 470                                        union ixgbe_adv_rx_desc *rx_desc,
 471                                        struct sk_buff *skb)
 472 {
 473         skb_checksum_none_assert(skb);
 474 
 475         /* Rx csum disabled */
 476         if (!(ring->netdev->features & NETIF_F_RXCSUM))
 477                 return;
 478 
 479         /* if IP and error */
 480         if (ixgbevf_test_staterr(rx_desc, IXGBE_RXD_STAT_IPCS) &&
 481             ixgbevf_test_staterr(rx_desc, IXGBE_RXDADV_ERR_IPE)) {
 482                 ring->rx_stats.csum_err++;
 483                 return;
 484         }
 485 
 486         if (!ixgbevf_test_staterr(rx_desc, IXGBE_RXD_STAT_L4CS))
 487                 return;
 488 
 489         if (ixgbevf_test_staterr(rx_desc, IXGBE_RXDADV_ERR_TCPE)) {
 490                 ring->rx_stats.csum_err++;
 491                 return;
 492         }
 493 
 494         /* It must be a TCP or UDP packet with a valid checksum */
 495         skb->ip_summed = CHECKSUM_UNNECESSARY;
 496 }
 497 
 498 /**
 499  * ixgbevf_process_skb_fields - Populate skb header fields from Rx descriptor
 500  * @rx_ring: rx descriptor ring packet is being transacted on
 501  * @rx_desc: pointer to the EOP Rx descriptor
 502  * @skb: pointer to current skb being populated
 503  *
 504  * This function checks the ring, descriptor, and packet information in
 505  * order to populate the checksum, VLAN, protocol, and other fields within
 506  * the skb.
 507  **/
 508 static void ixgbevf_process_skb_fields(struct ixgbevf_ring *rx_ring,
 509                                        union ixgbe_adv_rx_desc *rx_desc,
 510                                        struct sk_buff *skb)
 511 {
 512         ixgbevf_rx_hash(rx_ring, rx_desc, skb);
 513         ixgbevf_rx_checksum(rx_ring, rx_desc, skb);
 514 
 515         if (ixgbevf_test_staterr(rx_desc, IXGBE_RXD_STAT_VP)) {
 516                 u16 vid = le16_to_cpu(rx_desc->wb.upper.vlan);
 517                 unsigned long *active_vlans = netdev_priv(rx_ring->netdev);
 518 
 519                 if (test_bit(vid & VLAN_VID_MASK, active_vlans))
 520                         __vlan_hwaccel_put_tag(skb, htons(ETH_P_8021Q), vid);
 521         }
 522 
 523         if (ixgbevf_test_staterr(rx_desc, IXGBE_RXDADV_STAT_SECP))
 524                 ixgbevf_ipsec_rx(rx_ring, rx_desc, skb);
 525 
 526         skb->protocol = eth_type_trans(skb, rx_ring->netdev);
 527 }
 528 
 529 static
 530 struct ixgbevf_rx_buffer *ixgbevf_get_rx_buffer(struct ixgbevf_ring *rx_ring,
 531                                                 const unsigned int size)
 532 {
 533         struct ixgbevf_rx_buffer *rx_buffer;
 534 
 535         rx_buffer = &rx_ring->rx_buffer_info[rx_ring->next_to_clean];
 536         prefetchw(rx_buffer->page);
 537 
 538         /* we are reusing so sync this buffer for CPU use */
 539         dma_sync_single_range_for_cpu(rx_ring->dev,
 540                                       rx_buffer->dma,
 541                                       rx_buffer->page_offset,
 542                                       size,
 543                                       DMA_FROM_DEVICE);
 544 
 545         rx_buffer->pagecnt_bias--;
 546 
 547         return rx_buffer;
 548 }
 549 
 550 static void ixgbevf_put_rx_buffer(struct ixgbevf_ring *rx_ring,
 551                                   struct ixgbevf_rx_buffer *rx_buffer,
 552                                   struct sk_buff *skb)
 553 {
 554         if (ixgbevf_can_reuse_rx_page(rx_buffer)) {
 555                 /* hand second half of page back to the ring */
 556                 ixgbevf_reuse_rx_page(rx_ring, rx_buffer);
 557         } else {
 558                 if (IS_ERR(skb))
 559                         /* We are not reusing the buffer so unmap it and free
 560                          * any references we are holding to it
 561                          */
 562                         dma_unmap_page_attrs(rx_ring->dev, rx_buffer->dma,
 563                                              ixgbevf_rx_pg_size(rx_ring),
 564                                              DMA_FROM_DEVICE,
 565                                              IXGBEVF_RX_DMA_ATTR);
 566                 __page_frag_cache_drain(rx_buffer->page,
 567                                         rx_buffer->pagecnt_bias);
 568         }
 569 
 570         /* clear contents of rx_buffer */
 571         rx_buffer->page = NULL;
 572 }
 573 
 574 /**
 575  * ixgbevf_is_non_eop - process handling of non-EOP buffers
 576  * @rx_ring: Rx ring being processed
 577  * @rx_desc: Rx descriptor for current buffer
 578  *
 579  * This function updates next to clean.  If the buffer is an EOP buffer
 580  * this function exits returning false, otherwise it will place the
 581  * sk_buff in the next buffer to be chained and return true indicating
 582  * that this is in fact a non-EOP buffer.
 583  **/
 584 static bool ixgbevf_is_non_eop(struct ixgbevf_ring *rx_ring,
 585                                union ixgbe_adv_rx_desc *rx_desc)
 586 {
 587         u32 ntc = rx_ring->next_to_clean + 1;
 588 
 589         /* fetch, update, and store next to clean */
 590         ntc = (ntc < rx_ring->count) ? ntc : 0;
 591         rx_ring->next_to_clean = ntc;
 592 
 593         prefetch(IXGBEVF_RX_DESC(rx_ring, ntc));
 594 
 595         if (likely(ixgbevf_test_staterr(rx_desc, IXGBE_RXD_STAT_EOP)))
 596                 return false;
 597 
 598         return true;
 599 }
 600 
 601 static inline unsigned int ixgbevf_rx_offset(struct ixgbevf_ring *rx_ring)
 602 {
 603         return ring_uses_build_skb(rx_ring) ? IXGBEVF_SKB_PAD : 0;
 604 }
 605 
 606 static bool ixgbevf_alloc_mapped_page(struct ixgbevf_ring *rx_ring,
 607                                       struct ixgbevf_rx_buffer *bi)
 608 {
 609         struct page *page = bi->page;
 610         dma_addr_t dma;
 611 
 612         /* since we are recycling buffers we should seldom need to alloc */
 613         if (likely(page))
 614                 return true;
 615 
 616         /* alloc new page for storage */
 617         page = dev_alloc_pages(ixgbevf_rx_pg_order(rx_ring));
 618         if (unlikely(!page)) {
 619                 rx_ring->rx_stats.alloc_rx_page_failed++;
 620                 return false;
 621         }
 622 
 623         /* map page for use */
 624         dma = dma_map_page_attrs(rx_ring->dev, page, 0,
 625                                  ixgbevf_rx_pg_size(rx_ring),
 626                                  DMA_FROM_DEVICE, IXGBEVF_RX_DMA_ATTR);
 627 
 628         /* if mapping failed free memory back to system since
 629          * there isn't much point in holding memory we can't use
 630          */
 631         if (dma_mapping_error(rx_ring->dev, dma)) {
 632                 __free_pages(page, ixgbevf_rx_pg_order(rx_ring));
 633 
 634                 rx_ring->rx_stats.alloc_rx_page_failed++;
 635                 return false;
 636         }
 637 
 638         bi->dma = dma;
 639         bi->page = page;
 640         bi->page_offset = ixgbevf_rx_offset(rx_ring);
 641         bi->pagecnt_bias = 1;
 642         rx_ring->rx_stats.alloc_rx_page++;
 643 
 644         return true;
 645 }
 646 
 647 /**
 648  * ixgbevf_alloc_rx_buffers - Replace used receive buffers; packet split
 649  * @rx_ring: rx descriptor ring (for a specific queue) to setup buffers on
 650  * @cleaned_count: number of buffers to replace
 651  **/
 652 static void ixgbevf_alloc_rx_buffers(struct ixgbevf_ring *rx_ring,
 653                                      u16 cleaned_count)
 654 {
 655         union ixgbe_adv_rx_desc *rx_desc;
 656         struct ixgbevf_rx_buffer *bi;
 657         unsigned int i = rx_ring->next_to_use;
 658 
 659         /* nothing to do or no valid netdev defined */
 660         if (!cleaned_count || !rx_ring->netdev)
 661                 return;
 662 
 663         rx_desc = IXGBEVF_RX_DESC(rx_ring, i);
 664         bi = &rx_ring->rx_buffer_info[i];
 665         i -= rx_ring->count;
 666 
 667         do {
 668                 if (!ixgbevf_alloc_mapped_page(rx_ring, bi))
 669                         break;
 670 
 671                 /* sync the buffer for use by the device */
 672                 dma_sync_single_range_for_device(rx_ring->dev, bi->dma,
 673                                                  bi->page_offset,
 674                                                  ixgbevf_rx_bufsz(rx_ring),
 675                                                  DMA_FROM_DEVICE);
 676 
 677                 /* Refresh the desc even if pkt_addr didn't change
 678                  * because each write-back erases this info.
 679                  */
 680                 rx_desc->read.pkt_addr = cpu_to_le64(bi->dma + bi->page_offset);
 681 
 682                 rx_desc++;
 683                 bi++;
 684                 i++;
 685                 if (unlikely(!i)) {
 686                         rx_desc = IXGBEVF_RX_DESC(rx_ring, 0);
 687                         bi = rx_ring->rx_buffer_info;
 688                         i -= rx_ring->count;
 689                 }
 690 
 691                 /* clear the length for the next_to_use descriptor */
 692                 rx_desc->wb.upper.length = 0;
 693 
 694                 cleaned_count--;
 695         } while (cleaned_count);
 696 
 697         i += rx_ring->count;
 698 
 699         if (rx_ring->next_to_use != i) {
 700                 /* record the next descriptor to use */
 701                 rx_ring->next_to_use = i;
 702 
 703                 /* update next to alloc since we have filled the ring */
 704                 rx_ring->next_to_alloc = i;
 705 
 706                 /* Force memory writes to complete before letting h/w
 707                  * know there are new descriptors to fetch.  (Only
 708                  * applicable for weak-ordered memory model archs,
 709                  * such as IA-64).
 710                  */
 711                 wmb();
 712                 ixgbevf_write_tail(rx_ring, i);
 713         }
 714 }
 715 
 716 /**
 717  * ixgbevf_cleanup_headers - Correct corrupted or empty headers
 718  * @rx_ring: rx descriptor ring packet is being transacted on
 719  * @rx_desc: pointer to the EOP Rx descriptor
 720  * @skb: pointer to current skb being fixed
 721  *
 722  * Check for corrupted packet headers caused by senders on the local L2
 723  * embedded NIC switch not setting up their Tx Descriptors right.  These
 724  * should be very rare.
 725  *
 726  * Also address the case where we are pulling data in on pages only
 727  * and as such no data is present in the skb header.
 728  *
 729  * In addition if skb is not at least 60 bytes we need to pad it so that
 730  * it is large enough to qualify as a valid Ethernet frame.
 731  *
 732  * Returns true if an error was encountered and skb was freed.
 733  **/
 734 static bool ixgbevf_cleanup_headers(struct ixgbevf_ring *rx_ring,
 735                                     union ixgbe_adv_rx_desc *rx_desc,
 736                                     struct sk_buff *skb)
 737 {
 738         /* XDP packets use error pointer so abort at this point */
 739         if (IS_ERR(skb))
 740                 return true;
 741 
 742         /* verify that the packet does not have any known errors */
 743         if (unlikely(ixgbevf_test_staterr(rx_desc,
 744                                           IXGBE_RXDADV_ERR_FRAME_ERR_MASK))) {
 745                 struct net_device *netdev = rx_ring->netdev;
 746 
 747                 if (!(netdev->features & NETIF_F_RXALL)) {
 748                         dev_kfree_skb_any(skb);
 749                         return true;
 750                 }
 751         }
 752 
 753         /* if eth_skb_pad returns an error the skb was freed */
 754         if (eth_skb_pad(skb))
 755                 return true;
 756 
 757         return false;
 758 }
 759 
 760 /**
 761  * ixgbevf_reuse_rx_page - page flip buffer and store it back on the ring
 762  * @rx_ring: rx descriptor ring to store buffers on
 763  * @old_buff: donor buffer to have page reused
 764  *
 765  * Synchronizes page for reuse by the adapter
 766  **/
 767 static void ixgbevf_reuse_rx_page(struct ixgbevf_ring *rx_ring,
 768                                   struct ixgbevf_rx_buffer *old_buff)
 769 {
 770         struct ixgbevf_rx_buffer *new_buff;
 771         u16 nta = rx_ring->next_to_alloc;
 772 
 773         new_buff = &rx_ring->rx_buffer_info[nta];
 774 
 775         /* update, and store next to alloc */
 776         nta++;
 777         rx_ring->next_to_alloc = (nta < rx_ring->count) ? nta : 0;
 778 
 779         /* transfer page from old buffer to new buffer */
 780         new_buff->page = old_buff->page;
 781         new_buff->dma = old_buff->dma;
 782         new_buff->page_offset = old_buff->page_offset;
 783         new_buff->pagecnt_bias = old_buff->pagecnt_bias;
 784 }
 785 
 786 static inline bool ixgbevf_page_is_reserved(struct page *page)
 787 {
 788         return (page_to_nid(page) != numa_mem_id()) || page_is_pfmemalloc(page);
 789 }
 790 
 791 static bool ixgbevf_can_reuse_rx_page(struct ixgbevf_rx_buffer *rx_buffer)
 792 {
 793         unsigned int pagecnt_bias = rx_buffer->pagecnt_bias;
 794         struct page *page = rx_buffer->page;
 795 
 796         /* avoid re-using remote pages */
 797         if (unlikely(ixgbevf_page_is_reserved(page)))
 798                 return false;
 799 
 800 #if (PAGE_SIZE < 8192)
 801         /* if we are only owner of page we can reuse it */
 802         if (unlikely((page_ref_count(page) - pagecnt_bias) > 1))
 803                 return false;
 804 #else
 805 #define IXGBEVF_LAST_OFFSET \
 806         (SKB_WITH_OVERHEAD(PAGE_SIZE) - IXGBEVF_RXBUFFER_2048)
 807 
 808         if (rx_buffer->page_offset > IXGBEVF_LAST_OFFSET)
 809                 return false;
 810 
 811 #endif
 812 
 813         /* If we have drained the page fragment pool we need to update
 814          * the pagecnt_bias and page count so that we fully restock the
 815          * number of references the driver holds.
 816          */
 817         if (unlikely(!pagecnt_bias)) {
 818                 page_ref_add(page, USHRT_MAX);
 819                 rx_buffer->pagecnt_bias = USHRT_MAX;
 820         }
 821 
 822         return true;
 823 }
 824 
 825 /**
 826  * ixgbevf_add_rx_frag - Add contents of Rx buffer to sk_buff
 827  * @rx_ring: rx descriptor ring to transact packets on
 828  * @rx_buffer: buffer containing page to add
 829  * @skb: sk_buff to place the data into
 830  * @size: size of buffer to be added
 831  *
 832  * This function will add the data contained in rx_buffer->page to the skb.
 833  **/
 834 static void ixgbevf_add_rx_frag(struct ixgbevf_ring *rx_ring,
 835                                 struct ixgbevf_rx_buffer *rx_buffer,
 836                                 struct sk_buff *skb,
 837                                 unsigned int size)
 838 {
 839 #if (PAGE_SIZE < 8192)
 840         unsigned int truesize = ixgbevf_rx_pg_size(rx_ring) / 2;
 841 #else
 842         unsigned int truesize = ring_uses_build_skb(rx_ring) ?
 843                                 SKB_DATA_ALIGN(IXGBEVF_SKB_PAD + size) :
 844                                 SKB_DATA_ALIGN(size);
 845 #endif
 846         skb_add_rx_frag(skb, skb_shinfo(skb)->nr_frags, rx_buffer->page,
 847                         rx_buffer->page_offset, size, truesize);
 848 #if (PAGE_SIZE < 8192)
 849         rx_buffer->page_offset ^= truesize;
 850 #else
 851         rx_buffer->page_offset += truesize;
 852 #endif
 853 }
 854 
 855 static
 856 struct sk_buff *ixgbevf_construct_skb(struct ixgbevf_ring *rx_ring,
 857                                       struct ixgbevf_rx_buffer *rx_buffer,
 858                                       struct xdp_buff *xdp,
 859                                       union ixgbe_adv_rx_desc *rx_desc)
 860 {
 861         unsigned int size = xdp->data_end - xdp->data;
 862 #if (PAGE_SIZE < 8192)
 863         unsigned int truesize = ixgbevf_rx_pg_size(rx_ring) / 2;
 864 #else
 865         unsigned int truesize = SKB_DATA_ALIGN(xdp->data_end -
 866                                                xdp->data_hard_start);
 867 #endif
 868         unsigned int headlen;
 869         struct sk_buff *skb;
 870 
 871         /* prefetch first cache line of first page */
 872         prefetch(xdp->data);
 873 #if L1_CACHE_BYTES < 128
 874         prefetch(xdp->data + L1_CACHE_BYTES);
 875 #endif
 876         /* Note, we get here by enabling legacy-rx via:
 877          *
 878          *    ethtool --set-priv-flags <dev> legacy-rx on
 879          *
 880          * In this mode, we currently get 0 extra XDP headroom as
 881          * opposed to having legacy-rx off, where we process XDP
 882          * packets going to stack via ixgbevf_build_skb().
 883          *
 884          * For ixgbevf_construct_skb() mode it means that the
 885          * xdp->data_meta will always point to xdp->data, since
 886          * the helper cannot expand the head. Should this ever
 887          * changed in future for legacy-rx mode on, then lets also
 888          * add xdp->data_meta handling here.
 889          */
 890 
 891         /* allocate a skb to store the frags */
 892         skb = napi_alloc_skb(&rx_ring->q_vector->napi, IXGBEVF_RX_HDR_SIZE);
 893         if (unlikely(!skb))
 894                 return NULL;
 895 
 896         /* Determine available headroom for copy */
 897         headlen = size;
 898         if (headlen > IXGBEVF_RX_HDR_SIZE)
 899                 headlen = eth_get_headlen(skb->dev, xdp->data,
 900                                           IXGBEVF_RX_HDR_SIZE);
 901 
 902         /* align pull length to size of long to optimize memcpy performance */
 903         memcpy(__skb_put(skb, headlen), xdp->data,
 904                ALIGN(headlen, sizeof(long)));
 905 
 906         /* update all of the pointers */
 907         size -= headlen;
 908         if (size) {
 909                 skb_add_rx_frag(skb, 0, rx_buffer->page,
 910                                 (xdp->data + headlen) -
 911                                         page_address(rx_buffer->page),
 912                                 size, truesize);
 913 #if (PAGE_SIZE < 8192)
 914                 rx_buffer->page_offset ^= truesize;
 915 #else
 916                 rx_buffer->page_offset += truesize;
 917 #endif
 918         } else {
 919                 rx_buffer->pagecnt_bias++;
 920         }
 921 
 922         return skb;
 923 }
 924 
 925 static inline void ixgbevf_irq_enable_queues(struct ixgbevf_adapter *adapter,
 926                                              u32 qmask)
 927 {
 928         struct ixgbe_hw *hw = &adapter->hw;
 929 
 930         IXGBE_WRITE_REG(hw, IXGBE_VTEIMS, qmask);
 931 }
 932 
 933 static struct sk_buff *ixgbevf_build_skb(struct ixgbevf_ring *rx_ring,
 934                                          struct ixgbevf_rx_buffer *rx_buffer,
 935                                          struct xdp_buff *xdp,
 936                                          union ixgbe_adv_rx_desc *rx_desc)
 937 {
 938         unsigned int metasize = xdp->data - xdp->data_meta;
 939 #if (PAGE_SIZE < 8192)
 940         unsigned int truesize = ixgbevf_rx_pg_size(rx_ring) / 2;
 941 #else
 942         unsigned int truesize = SKB_DATA_ALIGN(sizeof(struct skb_shared_info)) +
 943                                 SKB_DATA_ALIGN(xdp->data_end -
 944                                                xdp->data_hard_start);
 945 #endif
 946         struct sk_buff *skb;
 947 
 948         /* Prefetch first cache line of first page. If xdp->data_meta
 949          * is unused, this points to xdp->data, otherwise, we likely
 950          * have a consumer accessing first few bytes of meta data,
 951          * and then actual data.
 952          */
 953         prefetch(xdp->data_meta);
 954 #if L1_CACHE_BYTES < 128
 955         prefetch(xdp->data_meta + L1_CACHE_BYTES);
 956 #endif
 957 
 958         /* build an skb around the page buffer */
 959         skb = build_skb(xdp->data_hard_start, truesize);
 960         if (unlikely(!skb))
 961                 return NULL;
 962 
 963         /* update pointers within the skb to store the data */
 964         skb_reserve(skb, xdp->data - xdp->data_hard_start);
 965         __skb_put(skb, xdp->data_end - xdp->data);
 966         if (metasize)
 967                 skb_metadata_set(skb, metasize);
 968 
 969         /* update buffer offset */
 970 #if (PAGE_SIZE < 8192)
 971         rx_buffer->page_offset ^= truesize;
 972 #else
 973         rx_buffer->page_offset += truesize;
 974 #endif
 975 
 976         return skb;
 977 }
 978 
 979 #define IXGBEVF_XDP_PASS 0
 980 #define IXGBEVF_XDP_CONSUMED 1
 981 #define IXGBEVF_XDP_TX 2
 982 
 983 static int ixgbevf_xmit_xdp_ring(struct ixgbevf_ring *ring,
 984                                  struct xdp_buff *xdp)
 985 {
 986         struct ixgbevf_tx_buffer *tx_buffer;
 987         union ixgbe_adv_tx_desc *tx_desc;
 988         u32 len, cmd_type;
 989         dma_addr_t dma;
 990         u16 i;
 991 
 992         len = xdp->data_end - xdp->data;
 993 
 994         if (unlikely(!ixgbevf_desc_unused(ring)))
 995                 return IXGBEVF_XDP_CONSUMED;
 996 
 997         dma = dma_map_single(ring->dev, xdp->data, len, DMA_TO_DEVICE);
 998         if (dma_mapping_error(ring->dev, dma))
 999                 return IXGBEVF_XDP_CONSUMED;
1000 
1001         /* record the location of the first descriptor for this packet */
1002         i = ring->next_to_use;
1003         tx_buffer = &ring->tx_buffer_info[i];
1004 
1005         dma_unmap_len_set(tx_buffer, len, len);
1006         dma_unmap_addr_set(tx_buffer, dma, dma);
1007         tx_buffer->data = xdp->data;
1008         tx_buffer->bytecount = len;
1009         tx_buffer->gso_segs = 1;
1010         tx_buffer->protocol = 0;
1011 
1012         /* Populate minimal context descriptor that will provide for the
1013          * fact that we are expected to process Ethernet frames.
1014          */
1015         if (!test_bit(__IXGBEVF_TX_XDP_RING_PRIMED, &ring->state)) {
1016                 struct ixgbe_adv_tx_context_desc *context_desc;
1017 
1018                 set_bit(__IXGBEVF_TX_XDP_RING_PRIMED, &ring->state);
1019 
1020                 context_desc = IXGBEVF_TX_CTXTDESC(ring, 0);
1021                 context_desc->vlan_macip_lens   =
1022                         cpu_to_le32(ETH_HLEN << IXGBE_ADVTXD_MACLEN_SHIFT);
1023                 context_desc->fceof_saidx       = 0;
1024                 context_desc->type_tucmd_mlhl   =
1025                         cpu_to_le32(IXGBE_TXD_CMD_DEXT |
1026                                     IXGBE_ADVTXD_DTYP_CTXT);
1027                 context_desc->mss_l4len_idx     = 0;
1028 
1029                 i = 1;
1030         }
1031 
1032         /* put descriptor type bits */
1033         cmd_type = IXGBE_ADVTXD_DTYP_DATA |
1034                    IXGBE_ADVTXD_DCMD_DEXT |
1035                    IXGBE_ADVTXD_DCMD_IFCS;
1036         cmd_type |= len | IXGBE_TXD_CMD;
1037 
1038         tx_desc = IXGBEVF_TX_DESC(ring, i);
1039         tx_desc->read.buffer_addr = cpu_to_le64(dma);
1040 
1041         tx_desc->read.cmd_type_len = cpu_to_le32(cmd_type);
1042         tx_desc->read.olinfo_status =
1043                         cpu_to_le32((len << IXGBE_ADVTXD_PAYLEN_SHIFT) |
1044                                     IXGBE_ADVTXD_CC);
1045 
1046         /* Avoid any potential race with cleanup */
1047         smp_wmb();
1048 
1049         /* set next_to_watch value indicating a packet is present */
1050         i++;
1051         if (i == ring->count)
1052                 i = 0;
1053 
1054         tx_buffer->next_to_watch = tx_desc;
1055         ring->next_to_use = i;
1056 
1057         return IXGBEVF_XDP_TX;
1058 }
1059 
1060 static struct sk_buff *ixgbevf_run_xdp(struct ixgbevf_adapter *adapter,
1061                                        struct ixgbevf_ring  *rx_ring,
1062                                        struct xdp_buff *xdp)
1063 {
1064         int result = IXGBEVF_XDP_PASS;
1065         struct ixgbevf_ring *xdp_ring;
1066         struct bpf_prog *xdp_prog;
1067         u32 act;
1068 
1069         rcu_read_lock();
1070         xdp_prog = READ_ONCE(rx_ring->xdp_prog);
1071 
1072         if (!xdp_prog)
1073                 goto xdp_out;
1074 
1075         act = bpf_prog_run_xdp(xdp_prog, xdp);
1076         switch (act) {
1077         case XDP_PASS:
1078                 break;
1079         case XDP_TX:
1080                 xdp_ring = adapter->xdp_ring[rx_ring->queue_index];
1081                 result = ixgbevf_xmit_xdp_ring(xdp_ring, xdp);
1082                 break;
1083         default:
1084                 bpf_warn_invalid_xdp_action(act);
1085                 /* fallthrough */
1086         case XDP_ABORTED:
1087                 trace_xdp_exception(rx_ring->netdev, xdp_prog, act);
1088                 /* fallthrough -- handle aborts by dropping packet */
1089         case XDP_DROP:
1090                 result = IXGBEVF_XDP_CONSUMED;
1091                 break;
1092         }
1093 xdp_out:
1094         rcu_read_unlock();
1095         return ERR_PTR(-result);
1096 }
1097 
1098 static void ixgbevf_rx_buffer_flip(struct ixgbevf_ring *rx_ring,
1099                                    struct ixgbevf_rx_buffer *rx_buffer,
1100                                    unsigned int size)
1101 {
1102 #if (PAGE_SIZE < 8192)
1103         unsigned int truesize = ixgbevf_rx_pg_size(rx_ring) / 2;
1104 
1105         rx_buffer->page_offset ^= truesize;
1106 #else
1107         unsigned int truesize = ring_uses_build_skb(rx_ring) ?
1108                                 SKB_DATA_ALIGN(IXGBEVF_SKB_PAD + size) :
1109                                 SKB_DATA_ALIGN(size);
1110 
1111         rx_buffer->page_offset += truesize;
1112 #endif
1113 }
1114 
1115 static int ixgbevf_clean_rx_irq(struct ixgbevf_q_vector *q_vector,
1116                                 struct ixgbevf_ring *rx_ring,
1117                                 int budget)
1118 {
1119         unsigned int total_rx_bytes = 0, total_rx_packets = 0;
1120         struct ixgbevf_adapter *adapter = q_vector->adapter;
1121         u16 cleaned_count = ixgbevf_desc_unused(rx_ring);
1122         struct sk_buff *skb = rx_ring->skb;
1123         bool xdp_xmit = false;
1124         struct xdp_buff xdp;
1125 
1126         xdp.rxq = &rx_ring->xdp_rxq;
1127 
1128         while (likely(total_rx_packets < budget)) {
1129                 struct ixgbevf_rx_buffer *rx_buffer;
1130                 union ixgbe_adv_rx_desc *rx_desc;
1131                 unsigned int size;
1132 
1133                 /* return some buffers to hardware, one at a time is too slow */
1134                 if (cleaned_count >= IXGBEVF_RX_BUFFER_WRITE) {
1135                         ixgbevf_alloc_rx_buffers(rx_ring, cleaned_count);
1136                         cleaned_count = 0;
1137                 }
1138 
1139                 rx_desc = IXGBEVF_RX_DESC(rx_ring, rx_ring->next_to_clean);
1140                 size = le16_to_cpu(rx_desc->wb.upper.length);
1141                 if (!size)
1142                         break;
1143 
1144                 /* This memory barrier is needed to keep us from reading
1145                  * any other fields out of the rx_desc until we know the
1146                  * RXD_STAT_DD bit is set
1147                  */
1148                 rmb();
1149 
1150                 rx_buffer = ixgbevf_get_rx_buffer(rx_ring, size);
1151 
1152                 /* retrieve a buffer from the ring */
1153                 if (!skb) {
1154                         xdp.data = page_address(rx_buffer->page) +
1155                                    rx_buffer->page_offset;
1156                         xdp.data_meta = xdp.data;
1157                         xdp.data_hard_start = xdp.data -
1158                                               ixgbevf_rx_offset(rx_ring);
1159                         xdp.data_end = xdp.data + size;
1160 
1161                         skb = ixgbevf_run_xdp(adapter, rx_ring, &xdp);
1162                 }
1163 
1164                 if (IS_ERR(skb)) {
1165                         if (PTR_ERR(skb) == -IXGBEVF_XDP_TX) {
1166                                 xdp_xmit = true;
1167                                 ixgbevf_rx_buffer_flip(rx_ring, rx_buffer,
1168                                                        size);
1169                         } else {
1170                                 rx_buffer->pagecnt_bias++;
1171                         }
1172                         total_rx_packets++;
1173                         total_rx_bytes += size;
1174                 } else if (skb) {
1175                         ixgbevf_add_rx_frag(rx_ring, rx_buffer, skb, size);
1176                 } else if (ring_uses_build_skb(rx_ring)) {
1177                         skb = ixgbevf_build_skb(rx_ring, rx_buffer,
1178                                                 &xdp, rx_desc);
1179                 } else {
1180                         skb = ixgbevf_construct_skb(rx_ring, rx_buffer,
1181                                                     &xdp, rx_desc);
1182                 }
1183 
1184                 /* exit if we failed to retrieve a buffer */
1185                 if (!skb) {
1186                         rx_ring->rx_stats.alloc_rx_buff_failed++;
1187                         rx_buffer->pagecnt_bias++;
1188                         break;
1189                 }
1190 
1191                 ixgbevf_put_rx_buffer(rx_ring, rx_buffer, skb);
1192                 cleaned_count++;
1193 
1194                 /* fetch next buffer in frame if non-eop */
1195                 if (ixgbevf_is_non_eop(rx_ring, rx_desc))
1196                         continue;
1197 
1198                 /* verify the packet layout is correct */
1199                 if (ixgbevf_cleanup_headers(rx_ring, rx_desc, skb)) {
1200                         skb = NULL;
1201                         continue;
1202                 }
1203 
1204                 /* probably a little skewed due to removing CRC */
1205                 total_rx_bytes += skb->len;
1206 
1207                 /* Workaround hardware that can't do proper VEPA multicast
1208                  * source pruning.
1209                  */
1210                 if ((skb->pkt_type == PACKET_BROADCAST ||
1211                      skb->pkt_type == PACKET_MULTICAST) &&
1212                     ether_addr_equal(rx_ring->netdev->dev_addr,
1213                                      eth_hdr(skb)->h_source)) {
1214                         dev_kfree_skb_irq(skb);
1215                         continue;
1216                 }
1217 
1218                 /* populate checksum, VLAN, and protocol */
1219                 ixgbevf_process_skb_fields(rx_ring, rx_desc, skb);
1220 
1221                 ixgbevf_rx_skb(q_vector, skb);
1222 
1223                 /* reset skb pointer */
1224                 skb = NULL;
1225 
1226                 /* update budget accounting */
1227                 total_rx_packets++;
1228         }
1229 
1230         /* place incomplete frames back on ring for completion */
1231         rx_ring->skb = skb;
1232 
1233         if (xdp_xmit) {
1234                 struct ixgbevf_ring *xdp_ring =
1235                         adapter->xdp_ring[rx_ring->queue_index];
1236 
1237                 /* Force memory writes to complete before letting h/w
1238                  * know there are new descriptors to fetch.
1239                  */
1240                 wmb();
1241                 ixgbevf_write_tail(xdp_ring, xdp_ring->next_to_use);
1242         }
1243 
1244         u64_stats_update_begin(&rx_ring->syncp);
1245         rx_ring->stats.packets += total_rx_packets;
1246         rx_ring->stats.bytes += total_rx_bytes;
1247         u64_stats_update_end(&rx_ring->syncp);
1248         q_vector->rx.total_packets += total_rx_packets;
1249         q_vector->rx.total_bytes += total_rx_bytes;
1250 
1251         return total_rx_packets;
1252 }
1253 
1254 /**
1255  * ixgbevf_poll - NAPI polling calback
1256  * @napi: napi struct with our devices info in it
1257  * @budget: amount of work driver is allowed to do this pass, in packets
1258  *
1259  * This function will clean more than one or more rings associated with a
1260  * q_vector.
1261  **/
1262 static int ixgbevf_poll(struct napi_struct *napi, int budget)
1263 {
1264         struct ixgbevf_q_vector *q_vector =
1265                 container_of(napi, struct ixgbevf_q_vector, napi);
1266         struct ixgbevf_adapter *adapter = q_vector->adapter;
1267         struct ixgbevf_ring *ring;
1268         int per_ring_budget, work_done = 0;
1269         bool clean_complete = true;
1270 
1271         ixgbevf_for_each_ring(ring, q_vector->tx) {
1272                 if (!ixgbevf_clean_tx_irq(q_vector, ring, budget))
1273                         clean_complete = false;
1274         }
1275 
1276         if (budget <= 0)
1277                 return budget;
1278 
1279         /* attempt to distribute budget to each queue fairly, but don't allow
1280          * the budget to go below 1 because we'll exit polling
1281          */
1282         if (q_vector->rx.count > 1)
1283                 per_ring_budget = max(budget/q_vector->rx.count, 1);
1284         else
1285                 per_ring_budget = budget;
1286 
1287         ixgbevf_for_each_ring(ring, q_vector->rx) {
1288                 int cleaned = ixgbevf_clean_rx_irq(q_vector, ring,
1289                                                    per_ring_budget);
1290                 work_done += cleaned;
1291                 if (cleaned >= per_ring_budget)
1292                         clean_complete = false;
1293         }
1294 
1295         /* If all work not completed, return budget and keep polling */
1296         if (!clean_complete)
1297                 return budget;
1298 
1299         /* Exit the polling mode, but don't re-enable interrupts if stack might
1300          * poll us due to busy-polling
1301          */
1302         if (likely(napi_complete_done(napi, work_done))) {
1303                 if (adapter->rx_itr_setting == 1)
1304                         ixgbevf_set_itr(q_vector);
1305                 if (!test_bit(__IXGBEVF_DOWN, &adapter->state) &&
1306                     !test_bit(__IXGBEVF_REMOVING, &adapter->state))
1307                         ixgbevf_irq_enable_queues(adapter,
1308                                                   BIT(q_vector->v_idx));
1309         }
1310 
1311         return min(work_done, budget - 1);
1312 }
1313 
1314 /**
1315  * ixgbevf_write_eitr - write VTEITR register in hardware specific way
1316  * @q_vector: structure containing interrupt and ring information
1317  **/
1318 void ixgbevf_write_eitr(struct ixgbevf_q_vector *q_vector)
1319 {
1320         struct ixgbevf_adapter *adapter = q_vector->adapter;
1321         struct ixgbe_hw *hw = &adapter->hw;
1322         int v_idx = q_vector->v_idx;
1323         u32 itr_reg = q_vector->itr & IXGBE_MAX_EITR;
1324 
1325         /* set the WDIS bit to not clear the timer bits and cause an
1326          * immediate assertion of the interrupt
1327          */
1328         itr_reg |= IXGBE_EITR_CNT_WDIS;
1329 
1330         IXGBE_WRITE_REG(hw, IXGBE_VTEITR(v_idx), itr_reg);
1331 }
1332 
1333 /**
1334  * ixgbevf_configure_msix - Configure MSI-X hardware
1335  * @adapter: board private structure
1336  *
1337  * ixgbevf_configure_msix sets up the hardware to properly generate MSI-X
1338  * interrupts.
1339  **/
1340 static void ixgbevf_configure_msix(struct ixgbevf_adapter *adapter)
1341 {
1342         struct ixgbevf_q_vector *q_vector;
1343         int q_vectors, v_idx;
1344 
1345         q_vectors = adapter->num_msix_vectors - NON_Q_VECTORS;
1346         adapter->eims_enable_mask = 0;
1347 
1348         /* Populate the IVAR table and set the ITR values to the
1349          * corresponding register.
1350          */
1351         for (v_idx = 0; v_idx < q_vectors; v_idx++) {
1352                 struct ixgbevf_ring *ring;
1353 
1354                 q_vector = adapter->q_vector[v_idx];
1355 
1356                 ixgbevf_for_each_ring(ring, q_vector->rx)
1357                         ixgbevf_set_ivar(adapter, 0, ring->reg_idx, v_idx);
1358 
1359                 ixgbevf_for_each_ring(ring, q_vector->tx)
1360                         ixgbevf_set_ivar(adapter, 1, ring->reg_idx, v_idx);
1361 
1362                 if (q_vector->tx.ring && !q_vector->rx.ring) {
1363                         /* Tx only vector */
1364                         if (adapter->tx_itr_setting == 1)
1365                                 q_vector->itr = IXGBE_12K_ITR;
1366                         else
1367                                 q_vector->itr = adapter->tx_itr_setting;
1368                 } else {
1369                         /* Rx or Rx/Tx vector */
1370                         if (adapter->rx_itr_setting == 1)
1371                                 q_vector->itr = IXGBE_20K_ITR;
1372                         else
1373                                 q_vector->itr = adapter->rx_itr_setting;
1374                 }
1375 
1376                 /* add q_vector eims value to global eims_enable_mask */
1377                 adapter->eims_enable_mask |= BIT(v_idx);
1378 
1379                 ixgbevf_write_eitr(q_vector);
1380         }
1381 
1382         ixgbevf_set_ivar(adapter, -1, 1, v_idx);
1383         /* setup eims_other and add value to global eims_enable_mask */
1384         adapter->eims_other = BIT(v_idx);
1385         adapter->eims_enable_mask |= adapter->eims_other;
1386 }
1387 
1388 enum latency_range {
1389         lowest_latency = 0,
1390         low_latency = 1,
1391         bulk_latency = 2,
1392         latency_invalid = 255
1393 };
1394 
1395 /**
1396  * ixgbevf_update_itr - update the dynamic ITR value based on statistics
1397  * @q_vector: structure containing interrupt and ring information
1398  * @ring_container: structure containing ring performance data
1399  *
1400  * Stores a new ITR value based on packets and byte
1401  * counts during the last interrupt.  The advantage of per interrupt
1402  * computation is faster updates and more accurate ITR for the current
1403  * traffic pattern.  Constants in this function were computed
1404  * based on theoretical maximum wire speed and thresholds were set based
1405  * on testing data as well as attempting to minimize response time
1406  * while increasing bulk throughput.
1407  **/
1408 static void ixgbevf_update_itr(struct ixgbevf_q_vector *q_vector,
1409                                struct ixgbevf_ring_container *ring_container)
1410 {
1411         int bytes = ring_container->total_bytes;
1412         int packets = ring_container->total_packets;
1413         u32 timepassed_us;
1414         u64 bytes_perint;
1415         u8 itr_setting = ring_container->itr;
1416 
1417         if (packets == 0)
1418                 return;
1419 
1420         /* simple throttle rate management
1421          *    0-20MB/s lowest (100000 ints/s)
1422          *   20-100MB/s low   (20000 ints/s)
1423          *  100-1249MB/s bulk (12000 ints/s)
1424          */
1425         /* what was last interrupt timeslice? */
1426         timepassed_us = q_vector->itr >> 2;
1427         if (timepassed_us == 0)
1428                 return;
1429 
1430         bytes_perint = bytes / timepassed_us; /* bytes/usec */
1431 
1432         switch (itr_setting) {
1433         case lowest_latency:
1434                 if (bytes_perint > 10)
1435                         itr_setting = low_latency;
1436                 break;
1437         case low_latency:
1438                 if (bytes_perint > 20)
1439                         itr_setting = bulk_latency;
1440                 else if (bytes_perint <= 10)
1441                         itr_setting = lowest_latency;
1442                 break;
1443         case bulk_latency:
1444                 if (bytes_perint <= 20)
1445                         itr_setting = low_latency;
1446                 break;
1447         }
1448 
1449         /* clear work counters since we have the values we need */
1450         ring_container->total_bytes = 0;
1451         ring_container->total_packets = 0;
1452 
1453         /* write updated itr to ring container */
1454         ring_container->itr = itr_setting;
1455 }
1456 
1457 static void ixgbevf_set_itr(struct ixgbevf_q_vector *q_vector)
1458 {
1459         u32 new_itr = q_vector->itr;
1460         u8 current_itr;
1461 
1462         ixgbevf_update_itr(q_vector, &q_vector->tx);
1463         ixgbevf_update_itr(q_vector, &q_vector->rx);
1464 
1465         current_itr = max(q_vector->rx.itr, q_vector->tx.itr);
1466 
1467         switch (current_itr) {
1468         /* counts and packets in update_itr are dependent on these numbers */
1469         case lowest_latency:
1470                 new_itr = IXGBE_100K_ITR;
1471                 break;
1472         case low_latency:
1473                 new_itr = IXGBE_20K_ITR;
1474                 break;
1475         case bulk_latency:
1476                 new_itr = IXGBE_12K_ITR;
1477                 break;
1478         default:
1479                 break;
1480         }
1481 
1482         if (new_itr != q_vector->itr) {
1483                 /* do an exponential smoothing */
1484                 new_itr = (10 * new_itr * q_vector->itr) /
1485                           ((9 * new_itr) + q_vector->itr);
1486 
1487                 /* save the algorithm value here */
1488                 q_vector->itr = new_itr;
1489 
1490                 ixgbevf_write_eitr(q_vector);
1491         }
1492 }
1493 
1494 static irqreturn_t ixgbevf_msix_other(int irq, void *data)
1495 {
1496         struct ixgbevf_adapter *adapter = data;
1497         struct ixgbe_hw *hw = &adapter->hw;
1498 
1499         hw->mac.get_link_status = 1;
1500 
1501         ixgbevf_service_event_schedule(adapter);
1502 
1503         IXGBE_WRITE_REG(hw, IXGBE_VTEIMS, adapter->eims_other);
1504 
1505         return IRQ_HANDLED;
1506 }
1507 
1508 /**
1509  * ixgbevf_msix_clean_rings - single unshared vector rx clean (all queues)
1510  * @irq: unused
1511  * @data: pointer to our q_vector struct for this interrupt vector
1512  **/
1513 static irqreturn_t ixgbevf_msix_clean_rings(int irq, void *data)
1514 {
1515         struct ixgbevf_q_vector *q_vector = data;
1516 
1517         /* EIAM disabled interrupts (on this vector) for us */
1518         if (q_vector->rx.ring || q_vector->tx.ring)
1519                 napi_schedule_irqoff(&q_vector->napi);
1520 
1521         return IRQ_HANDLED;
1522 }
1523 
1524 /**
1525  * ixgbevf_request_msix_irqs - Initialize MSI-X interrupts
1526  * @adapter: board private structure
1527  *
1528  * ixgbevf_request_msix_irqs allocates MSI-X vectors and requests
1529  * interrupts from the kernel.
1530  **/
1531 static int ixgbevf_request_msix_irqs(struct ixgbevf_adapter *adapter)
1532 {
1533         struct net_device *netdev = adapter->netdev;
1534         int q_vectors = adapter->num_msix_vectors - NON_Q_VECTORS;
1535         unsigned int ri = 0, ti = 0;
1536         int vector, err;
1537 
1538         for (vector = 0; vector < q_vectors; vector++) {
1539                 struct ixgbevf_q_vector *q_vector = adapter->q_vector[vector];
1540                 struct msix_entry *entry = &adapter->msix_entries[vector];
1541 
1542                 if (q_vector->tx.ring && q_vector->rx.ring) {
1543                         snprintf(q_vector->name, sizeof(q_vector->name),
1544                                  "%s-TxRx-%u", netdev->name, ri++);
1545                         ti++;
1546                 } else if (q_vector->rx.ring) {
1547                         snprintf(q_vector->name, sizeof(q_vector->name),
1548                                  "%s-rx-%u", netdev->name, ri++);
1549                 } else if (q_vector->tx.ring) {
1550                         snprintf(q_vector->name, sizeof(q_vector->name),
1551                                  "%s-tx-%u", netdev->name, ti++);
1552                 } else {
1553                         /* skip this unused q_vector */
1554                         continue;
1555                 }
1556                 err = request_irq(entry->vector, &ixgbevf_msix_clean_rings, 0,
1557                                   q_vector->name, q_vector);
1558                 if (err) {
1559                         hw_dbg(&adapter->hw,
1560                                "request_irq failed for MSIX interrupt Error: %d\n",
1561                                err);
1562                         goto free_queue_irqs;
1563                 }
1564         }
1565 
1566         err = request_irq(adapter->msix_entries[vector].vector,
1567                           &ixgbevf_msix_other, 0, netdev->name, adapter);
1568         if (err) {
1569                 hw_dbg(&adapter->hw, "request_irq for msix_other failed: %d\n",
1570                        err);
1571                 goto free_queue_irqs;
1572         }
1573 
1574         return 0;
1575 
1576 free_queue_irqs:
1577         while (vector) {
1578                 vector--;
1579                 free_irq(adapter->msix_entries[vector].vector,
1580                          adapter->q_vector[vector]);
1581         }
1582         /* This failure is non-recoverable - it indicates the system is
1583          * out of MSIX vector resources and the VF driver cannot run
1584          * without them.  Set the number of msix vectors to zero
1585          * indicating that not enough can be allocated.  The error
1586          * will be returned to the user indicating device open failed.
1587          * Any further attempts to force the driver to open will also
1588          * fail.  The only way to recover is to unload the driver and
1589          * reload it again.  If the system has recovered some MSIX
1590          * vectors then it may succeed.
1591          */
1592         adapter->num_msix_vectors = 0;
1593         return err;
1594 }
1595 
1596 /**
1597  * ixgbevf_request_irq - initialize interrupts
1598  * @adapter: board private structure
1599  *
1600  * Attempts to configure interrupts using the best available
1601  * capabilities of the hardware and kernel.
1602  **/
1603 static int ixgbevf_request_irq(struct ixgbevf_adapter *adapter)
1604 {
1605         int err = ixgbevf_request_msix_irqs(adapter);
1606 
1607         if (err)
1608                 hw_dbg(&adapter->hw, "request_irq failed, Error %d\n", err);
1609 
1610         return err;
1611 }
1612 
1613 static void ixgbevf_free_irq(struct ixgbevf_adapter *adapter)
1614 {
1615         int i, q_vectors;
1616 
1617         if (!adapter->msix_entries)
1618                 return;
1619 
1620         q_vectors = adapter->num_msix_vectors;
1621         i = q_vectors - 1;
1622 
1623         free_irq(adapter->msix_entries[i].vector, adapter);
1624         i--;
1625 
1626         for (; i >= 0; i--) {
1627                 /* free only the irqs that were actually requested */
1628                 if (!adapter->q_vector[i]->rx.ring &&
1629                     !adapter->q_vector[i]->tx.ring)
1630                         continue;
1631 
1632                 free_irq(adapter->msix_entries[i].vector,
1633                          adapter->q_vector[i]);
1634         }
1635 }
1636 
1637 /**
1638  * ixgbevf_irq_disable - Mask off interrupt generation on the NIC
1639  * @adapter: board private structure
1640  **/
1641 static inline void ixgbevf_irq_disable(struct ixgbevf_adapter *adapter)
1642 {
1643         struct ixgbe_hw *hw = &adapter->hw;
1644         int i;
1645 
1646         IXGBE_WRITE_REG(hw, IXGBE_VTEIAM, 0);
1647         IXGBE_WRITE_REG(hw, IXGBE_VTEIMC, ~0);
1648         IXGBE_WRITE_REG(hw, IXGBE_VTEIAC, 0);
1649 
1650         IXGBE_WRITE_FLUSH(hw);
1651 
1652         for (i = 0; i < adapter->num_msix_vectors; i++)
1653                 synchronize_irq(adapter->msix_entries[i].vector);
1654 }
1655 
1656 /**
1657  * ixgbevf_irq_enable - Enable default interrupt generation settings
1658  * @adapter: board private structure
1659  **/
1660 static inline void ixgbevf_irq_enable(struct ixgbevf_adapter *adapter)
1661 {
1662         struct ixgbe_hw *hw = &adapter->hw;
1663 
1664         IXGBE_WRITE_REG(hw, IXGBE_VTEIAM, adapter->eims_enable_mask);
1665         IXGBE_WRITE_REG(hw, IXGBE_VTEIAC, adapter->eims_enable_mask);
1666         IXGBE_WRITE_REG(hw, IXGBE_VTEIMS, adapter->eims_enable_mask);
1667 }
1668 
1669 /**
1670  * ixgbevf_configure_tx_ring - Configure 82599 VF Tx ring after Reset
1671  * @adapter: board private structure
1672  * @ring: structure containing ring specific data
1673  *
1674  * Configure the Tx descriptor ring after a reset.
1675  **/
1676 static void ixgbevf_configure_tx_ring(struct ixgbevf_adapter *adapter,
1677                                       struct ixgbevf_ring *ring)
1678 {
1679         struct ixgbe_hw *hw = &adapter->hw;
1680         u64 tdba = ring->dma;
1681         int wait_loop = 10;
1682         u32 txdctl = IXGBE_TXDCTL_ENABLE;
1683         u8 reg_idx = ring->reg_idx;
1684 
1685         /* disable queue to avoid issues while updating state */
1686         IXGBE_WRITE_REG(hw, IXGBE_VFTXDCTL(reg_idx), IXGBE_TXDCTL_SWFLSH);
1687         IXGBE_WRITE_FLUSH(hw);
1688 
1689         IXGBE_WRITE_REG(hw, IXGBE_VFTDBAL(reg_idx), tdba & DMA_BIT_MASK(32));
1690         IXGBE_WRITE_REG(hw, IXGBE_VFTDBAH(reg_idx), tdba >> 32);
1691         IXGBE_WRITE_REG(hw, IXGBE_VFTDLEN(reg_idx),
1692                         ring->count * sizeof(union ixgbe_adv_tx_desc));
1693 
1694         /* disable head writeback */
1695         IXGBE_WRITE_REG(hw, IXGBE_VFTDWBAH(reg_idx), 0);
1696         IXGBE_WRITE_REG(hw, IXGBE_VFTDWBAL(reg_idx), 0);
1697 
1698         /* enable relaxed ordering */
1699         IXGBE_WRITE_REG(hw, IXGBE_VFDCA_TXCTRL(reg_idx),
1700                         (IXGBE_DCA_TXCTRL_DESC_RRO_EN |
1701                          IXGBE_DCA_TXCTRL_DATA_RRO_EN));
1702 
1703         /* reset head and tail pointers */
1704         IXGBE_WRITE_REG(hw, IXGBE_VFTDH(reg_idx), 0);
1705         IXGBE_WRITE_REG(hw, IXGBE_VFTDT(reg_idx), 0);
1706         ring->tail = adapter->io_addr + IXGBE_VFTDT(reg_idx);
1707 
1708         /* reset ntu and ntc to place SW in sync with hardwdare */
1709         ring->next_to_clean = 0;
1710         ring->next_to_use = 0;
1711 
1712         /* In order to avoid issues WTHRESH + PTHRESH should always be equal
1713          * to or less than the number of on chip descriptors, which is
1714          * currently 40.
1715          */
1716         txdctl |= (8 << 16);    /* WTHRESH = 8 */
1717 
1718         /* Setting PTHRESH to 32 both improves performance */
1719         txdctl |= (1u << 8) |    /* HTHRESH = 1 */
1720                    32;           /* PTHRESH = 32 */
1721 
1722         /* reinitialize tx_buffer_info */
1723         memset(ring->tx_buffer_info, 0,
1724                sizeof(struct ixgbevf_tx_buffer) * ring->count);
1725 
1726         clear_bit(__IXGBEVF_HANG_CHECK_ARMED, &ring->state);
1727         clear_bit(__IXGBEVF_TX_XDP_RING_PRIMED, &ring->state);
1728 
1729         IXGBE_WRITE_REG(hw, IXGBE_VFTXDCTL(reg_idx), txdctl);
1730 
1731         /* poll to verify queue is enabled */
1732         do {
1733                 usleep_range(1000, 2000);
1734                 txdctl = IXGBE_READ_REG(hw, IXGBE_VFTXDCTL(reg_idx));
1735         }  while (--wait_loop && !(txdctl & IXGBE_TXDCTL_ENABLE));
1736         if (!wait_loop)
1737                 hw_dbg(hw, "Could not enable Tx Queue %d\n", reg_idx);
1738 }
1739 
1740 /**
1741  * ixgbevf_configure_tx - Configure 82599 VF Transmit Unit after Reset
1742  * @adapter: board private structure
1743  *
1744  * Configure the Tx unit of the MAC after a reset.
1745  **/
1746 static void ixgbevf_configure_tx(struct ixgbevf_adapter *adapter)
1747 {
1748         u32 i;
1749 
1750         /* Setup the HW Tx Head and Tail descriptor pointers */
1751         for (i = 0; i < adapter->num_tx_queues; i++)
1752                 ixgbevf_configure_tx_ring(adapter, adapter->tx_ring[i]);
1753         for (i = 0; i < adapter->num_xdp_queues; i++)
1754                 ixgbevf_configure_tx_ring(adapter, adapter->xdp_ring[i]);
1755 }
1756 
1757 #define IXGBE_SRRCTL_BSIZEHDRSIZE_SHIFT 2
1758 
1759 static void ixgbevf_configure_srrctl(struct ixgbevf_adapter *adapter,
1760                                      struct ixgbevf_ring *ring, int index)
1761 {
1762         struct ixgbe_hw *hw = &adapter->hw;
1763         u32 srrctl;
1764 
1765         srrctl = IXGBE_SRRCTL_DROP_EN;
1766 
1767         srrctl |= IXGBEVF_RX_HDR_SIZE << IXGBE_SRRCTL_BSIZEHDRSIZE_SHIFT;
1768         if (ring_uses_large_buffer(ring))
1769                 srrctl |= IXGBEVF_RXBUFFER_3072 >> IXGBE_SRRCTL_BSIZEPKT_SHIFT;
1770         else
1771                 srrctl |= IXGBEVF_RXBUFFER_2048 >> IXGBE_SRRCTL_BSIZEPKT_SHIFT;
1772         srrctl |= IXGBE_SRRCTL_DESCTYPE_ADV_ONEBUF;
1773 
1774         IXGBE_WRITE_REG(hw, IXGBE_VFSRRCTL(index), srrctl);
1775 }
1776 
1777 static void ixgbevf_setup_psrtype(struct ixgbevf_adapter *adapter)
1778 {
1779         struct ixgbe_hw *hw = &adapter->hw;
1780 
1781         /* PSRTYPE must be initialized in 82599 */
1782         u32 psrtype = IXGBE_PSRTYPE_TCPHDR | IXGBE_PSRTYPE_UDPHDR |
1783                       IXGBE_PSRTYPE_IPV4HDR | IXGBE_PSRTYPE_IPV6HDR |
1784                       IXGBE_PSRTYPE_L2HDR;
1785 
1786         if (adapter->num_rx_queues > 1)
1787                 psrtype |= BIT(29);
1788 
1789         IXGBE_WRITE_REG(hw, IXGBE_VFPSRTYPE, psrtype);
1790 }
1791 
1792 #define IXGBEVF_MAX_RX_DESC_POLL 10
1793 static void ixgbevf_disable_rx_queue(struct ixgbevf_adapter *adapter,
1794                                      struct ixgbevf_ring *ring)
1795 {
1796         struct ixgbe_hw *hw = &adapter->hw;
1797         int wait_loop = IXGBEVF_MAX_RX_DESC_POLL;
1798         u32 rxdctl;
1799         u8 reg_idx = ring->reg_idx;
1800 
1801         if (IXGBE_REMOVED(hw->hw_addr))
1802                 return;
1803         rxdctl = IXGBE_READ_REG(hw, IXGBE_VFRXDCTL(reg_idx));
1804         rxdctl &= ~IXGBE_RXDCTL_ENABLE;
1805 
1806         /* write value back with RXDCTL.ENABLE bit cleared */
1807         IXGBE_WRITE_REG(hw, IXGBE_VFRXDCTL(reg_idx), rxdctl);
1808 
1809         /* the hardware may take up to 100us to really disable the Rx queue */
1810         do {
1811                 udelay(10);
1812                 rxdctl = IXGBE_READ_REG(hw, IXGBE_VFRXDCTL(reg_idx));
1813         } while (--wait_loop && (rxdctl & IXGBE_RXDCTL_ENABLE));
1814 
1815         if (!wait_loop)
1816                 pr_err("RXDCTL.ENABLE queue %d not cleared while polling\n",
1817                        reg_idx);
1818 }
1819 
1820 static void ixgbevf_rx_desc_queue_enable(struct ixgbevf_adapter *adapter,
1821                                          struct ixgbevf_ring *ring)
1822 {
1823         struct ixgbe_hw *hw = &adapter->hw;
1824         int wait_loop = IXGBEVF_MAX_RX_DESC_POLL;
1825         u32 rxdctl;
1826         u8 reg_idx = ring->reg_idx;
1827 
1828         if (IXGBE_REMOVED(hw->hw_addr))
1829                 return;
1830         do {
1831                 usleep_range(1000, 2000);
1832                 rxdctl = IXGBE_READ_REG(hw, IXGBE_VFRXDCTL(reg_idx));
1833         } while (--wait_loop && !(rxdctl & IXGBE_RXDCTL_ENABLE));
1834 
1835         if (!wait_loop)
1836                 pr_err("RXDCTL.ENABLE queue %d not set while polling\n",
1837                        reg_idx);
1838 }
1839 
1840 /**
1841  * ixgbevf_init_rss_key - Initialize adapter RSS key
1842  * @adapter: device handle
1843  *
1844  * Allocates and initializes the RSS key if it is not allocated.
1845  **/
1846 static inline int ixgbevf_init_rss_key(struct ixgbevf_adapter *adapter)
1847 {
1848         u32 *rss_key;
1849 
1850         if (!adapter->rss_key) {
1851                 rss_key = kzalloc(IXGBEVF_RSS_HASH_KEY_SIZE, GFP_KERNEL);
1852                 if (unlikely(!rss_key))
1853                         return -ENOMEM;
1854 
1855                 netdev_rss_key_fill(rss_key, IXGBEVF_RSS_HASH_KEY_SIZE);
1856                 adapter->rss_key = rss_key;
1857         }
1858 
1859         return 0;
1860 }
1861 
1862 static void ixgbevf_setup_vfmrqc(struct ixgbevf_adapter *adapter)
1863 {
1864         struct ixgbe_hw *hw = &adapter->hw;
1865         u32 vfmrqc = 0, vfreta = 0;
1866         u16 rss_i = adapter->num_rx_queues;
1867         u8 i, j;
1868 
1869         /* Fill out hash function seeds */
1870         for (i = 0; i < IXGBEVF_VFRSSRK_REGS; i++)
1871                 IXGBE_WRITE_REG(hw, IXGBE_VFRSSRK(i), *(adapter->rss_key + i));
1872 
1873         for (i = 0, j = 0; i < IXGBEVF_X550_VFRETA_SIZE; i++, j++) {
1874                 if (j == rss_i)
1875                         j = 0;
1876 
1877                 adapter->rss_indir_tbl[i] = j;
1878 
1879                 vfreta |= j << (i & 0x3) * 8;
1880                 if ((i & 3) == 3) {
1881                         IXGBE_WRITE_REG(hw, IXGBE_VFRETA(i >> 2), vfreta);
1882                         vfreta = 0;
1883                 }
1884         }
1885 
1886         /* Perform hash on these packet types */
1887         vfmrqc |= IXGBE_VFMRQC_RSS_FIELD_IPV4 |
1888                 IXGBE_VFMRQC_RSS_FIELD_IPV4_TCP |
1889                 IXGBE_VFMRQC_RSS_FIELD_IPV6 |
1890                 IXGBE_VFMRQC_RSS_FIELD_IPV6_TCP;
1891 
1892         vfmrqc |= IXGBE_VFMRQC_RSSEN;
1893 
1894         IXGBE_WRITE_REG(hw, IXGBE_VFMRQC, vfmrqc);
1895 }
1896 
1897 static void ixgbevf_configure_rx_ring(struct ixgbevf_adapter *adapter,
1898                                       struct ixgbevf_ring *ring)
1899 {
1900         struct ixgbe_hw *hw = &adapter->hw;
1901         union ixgbe_adv_rx_desc *rx_desc;
1902         u64 rdba = ring->dma;
1903         u32 rxdctl;
1904         u8 reg_idx = ring->reg_idx;
1905 
1906         /* disable queue to avoid issues while updating state */
1907         rxdctl = IXGBE_READ_REG(hw, IXGBE_VFRXDCTL(reg_idx));
1908         ixgbevf_disable_rx_queue(adapter, ring);
1909 
1910         IXGBE_WRITE_REG(hw, IXGBE_VFRDBAL(reg_idx), rdba & DMA_BIT_MASK(32));
1911         IXGBE_WRITE_REG(hw, IXGBE_VFRDBAH(reg_idx), rdba >> 32);
1912         IXGBE_WRITE_REG(hw, IXGBE_VFRDLEN(reg_idx),
1913                         ring->count * sizeof(union ixgbe_adv_rx_desc));
1914 
1915 #ifndef CONFIG_SPARC
1916         /* enable relaxed ordering */
1917         IXGBE_WRITE_REG(hw, IXGBE_VFDCA_RXCTRL(reg_idx),
1918                         IXGBE_DCA_RXCTRL_DESC_RRO_EN);
1919 #else
1920         IXGBE_WRITE_REG(hw, IXGBE_VFDCA_RXCTRL(reg_idx),
1921                         IXGBE_DCA_RXCTRL_DESC_RRO_EN |
1922                         IXGBE_DCA_RXCTRL_DATA_WRO_EN);
1923 #endif
1924 
1925         /* reset head and tail pointers */
1926         IXGBE_WRITE_REG(hw, IXGBE_VFRDH(reg_idx), 0);
1927         IXGBE_WRITE_REG(hw, IXGBE_VFRDT(reg_idx), 0);
1928         ring->tail = adapter->io_addr + IXGBE_VFRDT(reg_idx);
1929 
1930         /* initialize rx_buffer_info */
1931         memset(ring->rx_buffer_info, 0,
1932                sizeof(struct ixgbevf_rx_buffer) * ring->count);
1933 
1934         /* initialize Rx descriptor 0 */
1935         rx_desc = IXGBEVF_RX_DESC(ring, 0);
1936         rx_desc->wb.upper.length = 0;
1937 
1938         /* reset ntu and ntc to place SW in sync with hardwdare */
1939         ring->next_to_clean = 0;
1940         ring->next_to_use = 0;
1941         ring->next_to_alloc = 0;
1942 
1943         ixgbevf_configure_srrctl(adapter, ring, reg_idx);
1944 
1945         /* RXDCTL.RLPML does not work on 82599 */
1946         if (adapter->hw.mac.type != ixgbe_mac_82599_vf) {
1947                 rxdctl &= ~(IXGBE_RXDCTL_RLPMLMASK |
1948                             IXGBE_RXDCTL_RLPML_EN);
1949 
1950 #if (PAGE_SIZE < 8192)
1951                 /* Limit the maximum frame size so we don't overrun the skb */
1952                 if (ring_uses_build_skb(ring) &&
1953                     !ring_uses_large_buffer(ring))
1954                         rxdctl |= IXGBEVF_MAX_FRAME_BUILD_SKB |
1955                                   IXGBE_RXDCTL_RLPML_EN;
1956 #endif
1957         }
1958 
1959         rxdctl |= IXGBE_RXDCTL_ENABLE | IXGBE_RXDCTL_VME;
1960         IXGBE_WRITE_REG(hw, IXGBE_VFRXDCTL(reg_idx), rxdctl);
1961 
1962         ixgbevf_rx_desc_queue_enable(adapter, ring);
1963         ixgbevf_alloc_rx_buffers(ring, ixgbevf_desc_unused(ring));
1964 }
1965 
1966 static void ixgbevf_set_rx_buffer_len(struct ixgbevf_adapter *adapter,
1967                                       struct ixgbevf_ring *rx_ring)
1968 {
1969         struct net_device *netdev = adapter->netdev;
1970         unsigned int max_frame = netdev->mtu + ETH_HLEN + ETH_FCS_LEN;
1971 
1972         /* set build_skb and buffer size flags */
1973         clear_ring_build_skb_enabled(rx_ring);
1974         clear_ring_uses_large_buffer(rx_ring);
1975 
1976         if (adapter->flags & IXGBEVF_FLAGS_LEGACY_RX)
1977                 return;
1978 
1979         set_ring_build_skb_enabled(rx_ring);
1980 
1981         if (PAGE_SIZE < 8192) {
1982                 if (max_frame <= IXGBEVF_MAX_FRAME_BUILD_SKB)
1983                         return;
1984 
1985                 set_ring_uses_large_buffer(rx_ring);
1986         }
1987 }
1988 
1989 /**
1990  * ixgbevf_configure_rx - Configure 82599 VF Receive Unit after Reset
1991  * @adapter: board private structure
1992  *
1993  * Configure the Rx unit of the MAC after a reset.
1994  **/
1995 static void ixgbevf_configure_rx(struct ixgbevf_adapter *adapter)
1996 {
1997         struct ixgbe_hw *hw = &adapter->hw;
1998         struct net_device *netdev = adapter->netdev;
1999         int i, ret;
2000 
2001         ixgbevf_setup_psrtype(adapter);
2002         if (hw->mac.type >= ixgbe_mac_X550_vf)
2003                 ixgbevf_setup_vfmrqc(adapter);
2004 
2005         spin_lock_bh(&adapter->mbx_lock);
2006         /* notify the PF of our intent to use this size of frame */
2007         ret = hw->mac.ops.set_rlpml(hw, netdev->mtu + ETH_HLEN + ETH_FCS_LEN);
2008         spin_unlock_bh(&adapter->mbx_lock);
2009         if (ret)
2010                 dev_err(&adapter->pdev->dev,
2011                         "Failed to set MTU at %d\n", netdev->mtu);
2012 
2013         /* Setup the HW Rx Head and Tail Descriptor Pointers and
2014          * the Base and Length of the Rx Descriptor Ring
2015          */
2016         for (i = 0; i < adapter->num_rx_queues; i++) {
2017                 struct ixgbevf_ring *rx_ring = adapter->rx_ring[i];
2018 
2019                 ixgbevf_set_rx_buffer_len(adapter, rx_ring);
2020                 ixgbevf_configure_rx_ring(adapter, rx_ring);
2021         }
2022 }
2023 
2024 static int ixgbevf_vlan_rx_add_vid(struct net_device *netdev,
2025                                    __be16 proto, u16 vid)
2026 {
2027         struct ixgbevf_adapter *adapter = netdev_priv(netdev);
2028         struct ixgbe_hw *hw = &adapter->hw;
2029         int err;
2030 
2031         spin_lock_bh(&adapter->mbx_lock);
2032 
2033         /* add VID to filter table */
2034         err = hw->mac.ops.set_vfta(hw, vid, 0, true);
2035 
2036         spin_unlock_bh(&adapter->mbx_lock);
2037 
2038         /* translate error return types so error makes sense */
2039         if (err == IXGBE_ERR_MBX)
2040                 return -EIO;
2041 
2042         if (err == IXGBE_ERR_INVALID_ARGUMENT)
2043                 return -EACCES;
2044 
2045         set_bit(vid, adapter->active_vlans);
2046 
2047         return err;
2048 }
2049 
2050 static int ixgbevf_vlan_rx_kill_vid(struct net_device *netdev,
2051                                     __be16 proto, u16 vid)
2052 {
2053         struct ixgbevf_adapter *adapter = netdev_priv(netdev);
2054         struct ixgbe_hw *hw = &adapter->hw;
2055         int err;
2056 
2057         spin_lock_bh(&adapter->mbx_lock);
2058 
2059         /* remove VID from filter table */
2060         err = hw->mac.ops.set_vfta(hw, vid, 0, false);
2061 
2062         spin_unlock_bh(&adapter->mbx_lock);
2063 
2064         clear_bit(vid, adapter->active_vlans);
2065 
2066         return err;
2067 }
2068 
2069 static void ixgbevf_restore_vlan(struct ixgbevf_adapter *adapter)
2070 {
2071         u16 vid;
2072 
2073         for_each_set_bit(vid, adapter->active_vlans, VLAN_N_VID)
2074                 ixgbevf_vlan_rx_add_vid(adapter->netdev,
2075                                         htons(ETH_P_8021Q), vid);
2076 }
2077 
2078 static int ixgbevf_write_uc_addr_list(struct net_device *netdev)
2079 {
2080         struct ixgbevf_adapter *adapter = netdev_priv(netdev);
2081         struct ixgbe_hw *hw = &adapter->hw;
2082         int count = 0;
2083 
2084         if (!netdev_uc_empty(netdev)) {
2085                 struct netdev_hw_addr *ha;
2086 
2087                 netdev_for_each_uc_addr(ha, netdev) {
2088                         hw->mac.ops.set_uc_addr(hw, ++count, ha->addr);
2089                         udelay(200);
2090                 }
2091         } else {
2092                 /* If the list is empty then send message to PF driver to
2093                  * clear all MAC VLANs on this VF.
2094                  */
2095                 hw->mac.ops.set_uc_addr(hw, 0, NULL);
2096         }
2097 
2098         return count;
2099 }
2100 
2101 /**
2102  * ixgbevf_set_rx_mode - Multicast and unicast set
2103  * @netdev: network interface device structure
2104  *
2105  * The set_rx_method entry point is called whenever the multicast address
2106  * list, unicast address list or the network interface flags are updated.
2107  * This routine is responsible for configuring the hardware for proper
2108  * multicast mode and configuring requested unicast filters.
2109  **/
2110 static void ixgbevf_set_rx_mode(struct net_device *netdev)
2111 {
2112         struct ixgbevf_adapter *adapter = netdev_priv(netdev);
2113         struct ixgbe_hw *hw = &adapter->hw;
2114         unsigned int flags = netdev->flags;
2115         int xcast_mode;
2116 
2117         /* request the most inclusive mode we need */
2118         if (flags & IFF_PROMISC)
2119                 xcast_mode = IXGBEVF_XCAST_MODE_PROMISC;
2120         else if (flags & IFF_ALLMULTI)
2121                 xcast_mode = IXGBEVF_XCAST_MODE_ALLMULTI;
2122         else if (flags & (IFF_BROADCAST | IFF_MULTICAST))
2123                 xcast_mode = IXGBEVF_XCAST_MODE_MULTI;
2124         else
2125                 xcast_mode = IXGBEVF_XCAST_MODE_NONE;
2126 
2127         spin_lock_bh(&adapter->mbx_lock);
2128 
2129         hw->mac.ops.update_xcast_mode(hw, xcast_mode);
2130 
2131         /* reprogram multicast list */
2132         hw->mac.ops.update_mc_addr_list(hw, netdev);
2133 
2134         ixgbevf_write_uc_addr_list(netdev);
2135 
2136         spin_unlock_bh(&adapter->mbx_lock);
2137 }
2138 
2139 static void ixgbevf_napi_enable_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_enable(&q_vector->napi);
2148         }
2149 }
2150 
2151 static void ixgbevf_napi_disable_all(struct ixgbevf_adapter *adapter)
2152 {
2153         int q_idx;
2154         struct ixgbevf_q_vector *q_vector;
2155         int q_vectors = adapter->num_msix_vectors - NON_Q_VECTORS;
2156 
2157         for (q_idx = 0; q_idx < q_vectors; q_idx++) {
2158                 q_vector = adapter->q_vector[q_idx];
2159                 napi_disable(&q_vector->napi);
2160         }
2161 }
2162 
2163 static int ixgbevf_configure_dcb(struct ixgbevf_adapter *adapter)
2164 {
2165         struct ixgbe_hw *hw = &adapter->hw;
2166         unsigned int def_q = 0;
2167         unsigned int num_tcs = 0;
2168         unsigned int num_rx_queues = adapter->num_rx_queues;
2169         unsigned int num_tx_queues = adapter->num_tx_queues;
2170         int err;
2171 
2172         spin_lock_bh(&adapter->mbx_lock);
2173 
2174         /* fetch queue configuration from the PF */
2175         err = ixgbevf_get_queues(hw, &num_tcs, &def_q);
2176 
2177         spin_unlock_bh(&adapter->mbx_lock);
2178 
2179         if (err)
2180                 return err;
2181 
2182         if (num_tcs > 1) {
2183                 /* we need only one Tx queue */
2184                 num_tx_queues = 1;
2185 
2186                 /* update default Tx ring register index */
2187                 adapter->tx_ring[0]->reg_idx = def_q;
2188 
2189                 /* we need as many queues as traffic classes */
2190                 num_rx_queues = num_tcs;
2191         }
2192 
2193         /* if we have a bad config abort request queue reset */
2194         if ((adapter->num_rx_queues != num_rx_queues) ||
2195             (adapter->num_tx_queues != num_tx_queues)) {
2196                 /* force mailbox timeout to prevent further messages */
2197                 hw->mbx.timeout = 0;
2198 
2199                 /* wait for watchdog to come around and bail us out */
2200                 set_bit(__IXGBEVF_QUEUE_RESET_REQUESTED, &adapter->state);
2201         }
2202 
2203         return 0;
2204 }
2205 
2206 static void ixgbevf_configure(struct ixgbevf_adapter *adapter)
2207 {
2208         ixgbevf_configure_dcb(adapter);
2209 
2210         ixgbevf_set_rx_mode(adapter->netdev);
2211 
2212         ixgbevf_restore_vlan(adapter);
2213         ixgbevf_ipsec_restore(adapter);
2214 
2215         ixgbevf_configure_tx(adapter);
2216         ixgbevf_configure_rx(adapter);
2217 }
2218 
2219 static void ixgbevf_save_reset_stats(struct ixgbevf_adapter *adapter)
2220 {
2221         /* Only save pre-reset stats if there are some */
2222         if (adapter->stats.vfgprc || adapter->stats.vfgptc) {
2223                 adapter->stats.saved_reset_vfgprc += adapter->stats.vfgprc -
2224                         adapter->stats.base_vfgprc;
2225                 adapter->stats.saved_reset_vfgptc += adapter->stats.vfgptc -
2226                         adapter->stats.base_vfgptc;
2227                 adapter->stats.saved_reset_vfgorc += adapter->stats.vfgorc -
2228                         adapter->stats.base_vfgorc;
2229                 adapter->stats.saved_reset_vfgotc += adapter->stats.vfgotc -
2230                         adapter->stats.base_vfgotc;
2231                 adapter->stats.saved_reset_vfmprc += adapter->stats.vfmprc -
2232                         adapter->stats.base_vfmprc;
2233         }
2234 }
2235 
2236 static void ixgbevf_init_last_counter_stats(struct ixgbevf_adapter *adapter)
2237 {
2238         struct ixgbe_hw *hw = &adapter->hw;
2239 
2240         adapter->stats.last_vfgprc = IXGBE_READ_REG(hw, IXGBE_VFGPRC);
2241         adapter->stats.last_vfgorc = IXGBE_READ_REG(hw, IXGBE_VFGORC_LSB);
2242         adapter->stats.last_vfgorc |=
2243                 (((u64)(IXGBE_READ_REG(hw, IXGBE_VFGORC_MSB))) << 32);
2244         adapter->stats.last_vfgptc = IXGBE_READ_REG(hw, IXGBE_VFGPTC);
2245         adapter->stats.last_vfgotc = IXGBE_READ_REG(hw, IXGBE_VFGOTC_LSB);
2246         adapter->stats.last_vfgotc |=
2247                 (((u64)(IXGBE_READ_REG(hw, IXGBE_VFGOTC_MSB))) << 32);
2248         adapter->stats.last_vfmprc = IXGBE_READ_REG(hw, IXGBE_VFMPRC);
2249 
2250         adapter->stats.base_vfgprc = adapter->stats.last_vfgprc;
2251         adapter->stats.base_vfgorc = adapter->stats.last_vfgorc;
2252         adapter->stats.base_vfgptc = adapter->stats.last_vfgptc;
2253         adapter->stats.base_vfgotc = adapter->stats.last_vfgotc;
2254         adapter->stats.base_vfmprc = adapter->stats.last_vfmprc;
2255 }
2256 
2257 static void ixgbevf_negotiate_api(struct ixgbevf_adapter *adapter)
2258 {
2259         struct ixgbe_hw *hw = &adapter->hw;
2260         static const int api[] = {
2261                 ixgbe_mbox_api_14,
2262                 ixgbe_mbox_api_13,
2263                 ixgbe_mbox_api_12,
2264                 ixgbe_mbox_api_11,
2265                 ixgbe_mbox_api_10,
2266                 ixgbe_mbox_api_unknown
2267         };
2268         int err, idx = 0;
2269 
2270         spin_lock_bh(&adapter->mbx_lock);
2271 
2272         while (api[idx] != ixgbe_mbox_api_unknown) {
2273                 err = hw->mac.ops.negotiate_api_version(hw, api[idx]);
2274                 if (!err)
2275                         break;
2276                 idx++;
2277         }
2278 
2279         spin_unlock_bh(&adapter->mbx_lock);
2280 }
2281 
2282 static void ixgbevf_up_complete(struct ixgbevf_adapter *adapter)
2283 {
2284         struct net_device *netdev = adapter->netdev;
2285         struct ixgbe_hw *hw = &adapter->hw;
2286 
2287         ixgbevf_configure_msix(adapter);
2288 
2289         spin_lock_bh(&adapter->mbx_lock);
2290 
2291         if (is_valid_ether_addr(hw->mac.addr))
2292                 hw->mac.ops.set_rar(hw, 0, hw->mac.addr, 0);
2293         else
2294                 hw->mac.ops.set_rar(hw, 0, hw->mac.perm_addr, 0);
2295 
2296         spin_unlock_bh(&adapter->mbx_lock);
2297 
2298         smp_mb__before_atomic();
2299         clear_bit(__IXGBEVF_DOWN, &adapter->state);
2300         ixgbevf_napi_enable_all(adapter);
2301 
2302         /* clear any pending interrupts, may auto mask */
2303         IXGBE_READ_REG(hw, IXGBE_VTEICR);
2304         ixgbevf_irq_enable(adapter);
2305 
2306         /* enable transmits */
2307         netif_tx_start_all_queues(netdev);
2308 
2309         ixgbevf_save_reset_stats(adapter);
2310         ixgbevf_init_last_counter_stats(adapter);
2311 
2312         hw->mac.get_link_status = 1;
2313         mod_timer(&adapter->service_timer, jiffies);
2314 }
2315 
2316 void ixgbevf_up(struct ixgbevf_adapter *adapter)
2317 {
2318         ixgbevf_configure(adapter);
2319 
2320         ixgbevf_up_complete(adapter);
2321 }
2322 
2323 /**
2324  * ixgbevf_clean_rx_ring - Free Rx Buffers per Queue
2325  * @rx_ring: ring to free buffers from
2326  **/
2327 static void ixgbevf_clean_rx_ring(struct ixgbevf_ring *rx_ring)
2328 {
2329         u16 i = rx_ring->next_to_clean;
2330 
2331         /* Free Rx ring sk_buff */
2332         if (rx_ring->skb) {
2333                 dev_kfree_skb(rx_ring->skb);
2334                 rx_ring->skb = NULL;
2335         }
2336 
2337         /* Free all the Rx ring pages */
2338         while (i != rx_ring->next_to_alloc) {
2339                 struct ixgbevf_rx_buffer *rx_buffer;
2340 
2341                 rx_buffer = &rx_ring->rx_buffer_info[i];
2342 
2343                 /* Invalidate cache lines that may have been written to by
2344                  * device so that we avoid corrupting memory.
2345                  */
2346                 dma_sync_single_range_for_cpu(rx_ring->dev,
2347                                               rx_buffer->dma,
2348                                               rx_buffer->page_offset,
2349                                               ixgbevf_rx_bufsz(rx_ring),
2350                                               DMA_FROM_DEVICE);
2351 
2352                 /* free resources associated with mapping */
2353                 dma_unmap_page_attrs(rx_ring->dev,
2354                                      rx_buffer->dma,
2355                                      ixgbevf_rx_pg_size(rx_ring),
2356                                      DMA_FROM_DEVICE,
2357                                      IXGBEVF_RX_DMA_ATTR);
2358 
2359                 __page_frag_cache_drain(rx_buffer->page,
2360                                         rx_buffer->pagecnt_bias);
2361 
2362                 i++;
2363                 if (i == rx_ring->count)
2364                         i = 0;
2365         }
2366 
2367         rx_ring->next_to_alloc = 0;
2368         rx_ring->next_to_clean = 0;
2369         rx_ring->next_to_use = 0;
2370 }
2371 
2372 /**
2373  * ixgbevf_clean_tx_ring - Free Tx Buffers
2374  * @tx_ring: ring to be cleaned
2375  **/
2376 static void ixgbevf_clean_tx_ring(struct ixgbevf_ring *tx_ring)
2377 {
2378         u16 i = tx_ring->next_to_clean;
2379         struct ixgbevf_tx_buffer *tx_buffer = &tx_ring->tx_buffer_info[i];
2380 
2381         while (i != tx_ring->next_to_use) {
2382                 union ixgbe_adv_tx_desc *eop_desc, *tx_desc;
2383 
2384                 /* Free all the Tx ring sk_buffs */
2385                 if (ring_is_xdp(tx_ring))
2386                         page_frag_free(tx_buffer->data);
2387                 else
2388                         dev_kfree_skb_any(tx_buffer->skb);
2389 
2390                 /* unmap skb header data */
2391                 dma_unmap_single(tx_ring->dev,
2392                                  dma_unmap_addr(tx_buffer, dma),
2393                                  dma_unmap_len(tx_buffer, len),
2394                                  DMA_TO_DEVICE);
2395 
2396                 /* check for eop_desc to determine the end of the packet */
2397                 eop_desc = tx_buffer->next_to_watch;
2398                 tx_desc = IXGBEVF_TX_DESC(tx_ring, i);
2399 
2400                 /* unmap remaining buffers */
2401                 while (tx_desc != eop_desc) {
2402                         tx_buffer++;
2403                         tx_desc++;
2404                         i++;
2405                         if (unlikely(i == tx_ring->count)) {
2406                                 i = 0;
2407                                 tx_buffer = tx_ring->tx_buffer_info;
2408                                 tx_desc = IXGBEVF_TX_DESC(tx_ring, 0);
2409                         }
2410 
2411                         /* unmap any remaining paged data */
2412                         if (dma_unmap_len(tx_buffer, len))
2413                                 dma_unmap_page(tx_ring->dev,
2414                                                dma_unmap_addr(tx_buffer, dma),
2415                                                dma_unmap_len(tx_buffer, len),
2416                                                DMA_TO_DEVICE);
2417                 }
2418 
2419                 /* move us one more past the eop_desc for start of next pkt */
2420                 tx_buffer++;
2421                 i++;
2422                 if (unlikely(i == tx_ring->count)) {
2423                         i = 0;
2424                         tx_buffer = tx_ring->tx_buffer_info;
2425                 }
2426         }
2427 
2428         /* reset next_to_use and next_to_clean */
2429         tx_ring->next_to_use = 0;
2430         tx_ring->next_to_clean = 0;
2431 
2432 }
2433 
2434 /**
2435  * ixgbevf_clean_all_rx_rings - Free Rx Buffers for all queues
2436  * @adapter: board private structure
2437  **/
2438 static void ixgbevf_clean_all_rx_rings(struct ixgbevf_adapter *adapter)
2439 {
2440         int i;
2441 
2442         for (i = 0; i < adapter->num_rx_queues; i++)
2443                 ixgbevf_clean_rx_ring(adapter->rx_ring[i]);
2444 }
2445 
2446 /**
2447  * ixgbevf_clean_all_tx_rings - Free Tx Buffers for all queues
2448  * @adapter: board private structure
2449  **/
2450 static void ixgbevf_clean_all_tx_rings(struct ixgbevf_adapter *adapter)
2451 {
2452         int i;
2453 
2454         for (i = 0; i < adapter->num_tx_queues; i++)
2455                 ixgbevf_clean_tx_ring(adapter->tx_ring[i]);
2456         for (i = 0; i < adapter->num_xdp_queues; i++)
2457                 ixgbevf_clean_tx_ring(adapter->xdp_ring[i]);
2458 }
2459 
2460 void ixgbevf_down(struct ixgbevf_adapter *adapter)
2461 {
2462         struct net_device *netdev = adapter->netdev;
2463         struct ixgbe_hw *hw = &adapter->hw;
2464         int i;
2465 
2466         /* signal that we are down to the interrupt handler */
2467         if (test_and_set_bit(__IXGBEVF_DOWN, &adapter->state))
2468                 return; /* do nothing if already down */
2469 
2470         /* disable all enabled Rx queues */
2471         for (i = 0; i < adapter->num_rx_queues; i++)
2472                 ixgbevf_disable_rx_queue(adapter, adapter->rx_ring[i]);
2473 
2474         usleep_range(10000, 20000);
2475 
2476         netif_tx_stop_all_queues(netdev);
2477 
2478         /* call carrier off first to avoid false dev_watchdog timeouts */
2479         netif_carrier_off(netdev);
2480         netif_tx_disable(netdev);
2481 
2482         ixgbevf_irq_disable(adapter);
2483 
2484         ixgbevf_napi_disable_all(adapter);
2485 
2486         del_timer_sync(&adapter->service_timer);
2487 
2488         /* disable transmits in the hardware now that interrupts are off */
2489         for (i = 0; i < adapter->num_tx_queues; i++) {
2490                 u8 reg_idx = adapter->tx_ring[i]->reg_idx;
2491 
2492                 IXGBE_WRITE_REG(hw, IXGBE_VFTXDCTL(reg_idx),
2493                                 IXGBE_TXDCTL_SWFLSH);
2494         }
2495 
2496         for (i = 0; i < adapter->num_xdp_queues; i++) {
2497                 u8 reg_idx = adapter->xdp_ring[i]->reg_idx;
2498 
2499                 IXGBE_WRITE_REG(hw, IXGBE_VFTXDCTL(reg_idx),
2500                                 IXGBE_TXDCTL_SWFLSH);
2501         }
2502 
2503         if (!pci_channel_offline(adapter->pdev))
2504                 ixgbevf_reset(adapter);
2505 
2506         ixgbevf_clean_all_tx_rings(adapter);
2507         ixgbevf_clean_all_rx_rings(adapter);
2508 }
2509 
2510 void ixgbevf_reinit_locked(struct ixgbevf_adapter *adapter)
2511 {
2512         WARN_ON(in_interrupt());
2513 
2514         while (test_and_set_bit(__IXGBEVF_RESETTING, &adapter->state))
2515                 msleep(1);
2516 
2517         ixgbevf_down(adapter);
2518         pci_set_master(adapter->pdev);
2519         ixgbevf_up(adapter);
2520 
2521         clear_bit(__IXGBEVF_RESETTING, &adapter->state);
2522 }
2523 
2524 void ixgbevf_reset(struct ixgbevf_adapter *adapter)
2525 {
2526         struct ixgbe_hw *hw = &adapter->hw;
2527         struct net_device *netdev = adapter->netdev;
2528 
2529         if (hw->mac.ops.reset_hw(hw)) {
2530                 hw_dbg(hw, "PF still resetting\n");
2531         } else {
2532                 hw->mac.ops.init_hw(hw);
2533                 ixgbevf_negotiate_api(adapter);
2534         }
2535 
2536         if (is_valid_ether_addr(adapter->hw.mac.addr)) {
2537                 ether_addr_copy(netdev->dev_addr, adapter->hw.mac.addr);
2538                 ether_addr_copy(netdev->perm_addr, adapter->hw.mac.addr);
2539         }
2540 
2541         adapter->last_reset = jiffies;
2542 }
2543 
2544 static int ixgbevf_acquire_msix_vectors(struct ixgbevf_adapter *adapter,
2545                                         int vectors)
2546 {
2547         int vector_threshold;
2548 
2549         /* We'll want at least 2 (vector_threshold):
2550          * 1) TxQ[0] + RxQ[0] handler
2551          * 2) Other (Link Status Change, etc.)
2552          */
2553         vector_threshold = MIN_MSIX_COUNT;
2554 
2555         /* The more we get, the more we will assign to Tx/Rx Cleanup
2556          * for the separate queues...where Rx Cleanup >= Tx Cleanup.
2557          * Right now, we simply care about how many we'll get; we'll
2558          * set them up later while requesting irq's.
2559          */
2560         vectors = pci_enable_msix_range(adapter->pdev, adapter->msix_entries,
2561                                         vector_threshold, vectors);
2562 
2563         if (vectors < 0) {
2564                 dev_err(&adapter->pdev->dev,
2565                         "Unable to allocate MSI-X interrupts\n");
2566                 kfree(adapter->msix_entries);
2567                 adapter->msix_entries = NULL;
2568                 return vectors;
2569         }
2570 
2571         /* Adjust for only the vectors we'll use, which is minimum
2572          * of max_msix_q_vectors + NON_Q_VECTORS, or the number of
2573          * vectors we were allocated.
2574          */
2575         adapter->num_msix_vectors = vectors;
2576 
2577         return 0;
2578 }
2579 
2580 /**
2581  * ixgbevf_set_num_queues - Allocate queues for device, feature dependent
2582  * @adapter: board private structure to initialize
2583  *
2584  * This is the top level queue allocation routine.  The order here is very
2585  * important, starting with the "most" number of features turned on at once,
2586  * and ending with the smallest set of features.  This way large combinations
2587  * can be allocated if they're turned on, and smaller combinations are the
2588  * fallthrough conditions.
2589  *
2590  **/
2591 static void ixgbevf_set_num_queues(struct ixgbevf_adapter *adapter)
2592 {
2593         struct ixgbe_hw *hw = &adapter->hw;
2594         unsigned int def_q = 0;
2595         unsigned int num_tcs = 0;
2596         int err;
2597 
2598         /* Start with base case */
2599         adapter->num_rx_queues = 1;
2600         adapter->num_tx_queues = 1;
2601         adapter->num_xdp_queues = 0;
2602 
2603         spin_lock_bh(&adapter->mbx_lock);
2604 
2605         /* fetch queue configuration from the PF */
2606         err = ixgbevf_get_queues(hw, &num_tcs, &def_q);
2607 
2608         spin_unlock_bh(&adapter->mbx_lock);
2609 
2610         if (err)
2611                 return;
2612 
2613         /* we need as many queues as traffic classes */
2614         if (num_tcs > 1) {
2615                 adapter->num_rx_queues = num_tcs;
2616         } else {
2617                 u16 rss = min_t(u16, num_online_cpus(), IXGBEVF_MAX_RSS_QUEUES);
2618 
2619                 switch (hw->api_version) {
2620                 case ixgbe_mbox_api_11:
2621                 case ixgbe_mbox_api_12:
2622                 case ixgbe_mbox_api_13:
2623                 case ixgbe_mbox_api_14:
2624                         if (adapter->xdp_prog &&
2625                             hw->mac.max_tx_queues == rss)
2626                                 rss = rss > 3 ? 2 : 1;
2627 
2628                         adapter->num_rx_queues = rss;
2629                         adapter->num_tx_queues = rss;
2630                         adapter->num_xdp_queues = adapter->xdp_prog ? rss : 0;
2631                 default:
2632                         break;
2633                 }
2634         }
2635 }
2636 
2637 /**
2638  * ixgbevf_set_interrupt_capability - set MSI-X or FAIL if not supported
2639  * @adapter: board private structure to initialize
2640  *
2641  * Attempt to configure the interrupts using the best available
2642  * capabilities of the hardware and the kernel.
2643  **/
2644 static int ixgbevf_set_interrupt_capability(struct ixgbevf_adapter *adapter)
2645 {
2646         int vector, v_budget;
2647 
2648         /* It's easy to be greedy for MSI-X vectors, but it really
2649          * doesn't do us much good if we have a lot more vectors
2650          * than CPU's.  So let's be conservative and only ask for
2651          * (roughly) the same number of vectors as there are CPU's.
2652          * The default is to use pairs of vectors.
2653          */
2654         v_budget = max(adapter->num_rx_queues, adapter->num_tx_queues);
2655         v_budget = min_t(int, v_budget, num_online_cpus());
2656         v_budget += NON_Q_VECTORS;
2657 
2658         adapter->msix_entries = kcalloc(v_budget,
2659                                         sizeof(struct msix_entry), GFP_KERNEL);
2660         if (!adapter->msix_entries)
2661                 return -ENOMEM;
2662 
2663         for (vector = 0; vector < v_budget; vector++)
2664                 adapter->msix_entries[vector].entry = vector;
2665 
2666         /* A failure in MSI-X entry allocation isn't fatal, but the VF driver
2667          * does not support any other modes, so we will simply fail here. Note
2668          * that we clean up the msix_entries pointer else-where.
2669          */
2670         return ixgbevf_acquire_msix_vectors(adapter, v_budget);
2671 }
2672 
2673 static void ixgbevf_add_ring(struct ixgbevf_ring *ring,
2674                              struct ixgbevf_ring_container *head)
2675 {
2676         ring->next = head->ring;
2677         head->ring = ring;
2678         head->count++;
2679 }
2680 
2681 /**
2682  * ixgbevf_alloc_q_vector - Allocate memory for a single interrupt vector
2683  * @adapter: board private structure to initialize
2684  * @v_idx: index of vector in adapter struct
2685  * @txr_count: number of Tx rings for q vector
2686  * @txr_idx: index of first Tx ring to assign
2687  * @xdp_count: total number of XDP rings to allocate
2688  * @xdp_idx: index of first XDP ring to allocate
2689  * @rxr_count: number of Rx rings for q vector
2690  * @rxr_idx: index of first Rx ring to assign
2691  *
2692  * We allocate one q_vector.  If allocation fails we return -ENOMEM.
2693  **/
2694 static int ixgbevf_alloc_q_vector(struct ixgbevf_adapter *adapter, int v_idx,
2695                                   int txr_count, int txr_idx,
2696                                   int xdp_count, int xdp_idx,
2697                                   int rxr_count, int rxr_idx)
2698 {
2699         struct ixgbevf_q_vector *q_vector;
2700         int reg_idx = txr_idx + xdp_idx;
2701         struct ixgbevf_ring *ring;
2702         int ring_count, size;
2703 
2704         ring_count = txr_count + xdp_count + rxr_count;
2705         size = sizeof(*q_vector) + (sizeof(*ring) * ring_count);
2706 
2707         /* allocate q_vector and rings */
2708         q_vector = kzalloc(size, GFP_KERNEL);
2709         if (!q_vector)
2710                 return -ENOMEM;
2711 
2712         /* initialize NAPI */
2713         netif_napi_add(adapter->netdev, &q_vector->napi, ixgbevf_poll, 64);
2714 
2715         /* tie q_vector and adapter together */
2716         adapter->q_vector[v_idx] = q_vector;
2717         q_vector->adapter = adapter;
2718         q_vector->v_idx = v_idx;
2719 
2720         /* initialize pointer to rings */
2721         ring = q_vector->ring;
2722 
2723         while (txr_count) {
2724                 /* assign generic ring traits */
2725                 ring->dev = &adapter->pdev->dev;
2726                 ring->netdev = adapter->netdev;
2727 
2728                 /* configure backlink on ring */
2729                 ring->q_vector = q_vector;
2730 
2731                 /* update q_vector Tx values */
2732                 ixgbevf_add_ring(ring, &q_vector->tx);
2733 
2734                 /* apply Tx specific ring traits */
2735                 ring->count = adapter->tx_ring_count;
2736                 ring->queue_index = txr_idx;
2737                 ring->reg_idx = reg_idx;
2738 
2739                 /* assign ring to adapter */
2740                  adapter->tx_ring[txr_idx] = ring;
2741 
2742                 /* update count and index */
2743                 txr_count--;
2744                 txr_idx++;
2745                 reg_idx++;
2746 
2747                 /* push pointer to next ring */
2748                 ring++;
2749         }
2750 
2751         while (xdp_count) {
2752                 /* assign generic ring traits */
2753                 ring->dev = &adapter->pdev->dev;
2754                 ring->netdev = adapter->netdev;
2755 
2756                 /* configure backlink on ring */
2757                 ring->q_vector = q_vector;
2758 
2759                 /* update q_vector Tx values */
2760                 ixgbevf_add_ring(ring, &q_vector->tx);
2761 
2762                 /* apply Tx specific ring traits */
2763                 ring->count = adapter->tx_ring_count;
2764                 ring->queue_index = xdp_idx;
2765                 ring->reg_idx = reg_idx;
2766                 set_ring_xdp(ring);
2767 
2768                 /* assign ring to adapter */
2769                 adapter->xdp_ring[xdp_idx] = ring;
2770 
2771                 /* update count and index */
2772                 xdp_count--;
2773                 xdp_idx++;
2774                 reg_idx++;
2775 
2776                 /* push pointer to next ring */
2777                 ring++;
2778         }
2779 
2780         while (rxr_count) {
2781                 /* assign generic ring traits */
2782                 ring->dev = &adapter->pdev->dev;
2783                 ring->netdev = adapter->netdev;
2784 
2785                 /* configure backlink on ring */
2786                 ring->q_vector = q_vector;
2787 
2788                 /* update q_vector Rx values */
2789                 ixgbevf_add_ring(ring, &q_vector->rx);
2790 
2791                 /* apply Rx specific ring traits */
2792                 ring->count = adapter->rx_ring_count;
2793                 ring->queue_index = rxr_idx;
2794                 ring->reg_idx = rxr_idx;
2795 
2796                 /* assign ring to adapter */
2797                 adapter->rx_ring[rxr_idx] = ring;
2798 
2799                 /* update count and index */
2800                 rxr_count--;
2801                 rxr_idx++;
2802 
2803                 /* push pointer to next ring */
2804                 ring++;
2805         }
2806 
2807         return 0;
2808 }
2809 
2810 /**
2811  * ixgbevf_free_q_vector - Free memory allocated for specific interrupt vector
2812  * @adapter: board private structure to initialize
2813  * @v_idx: index of vector in adapter struct
2814  *
2815  * This function frees the memory allocated to the q_vector.  In addition if
2816  * NAPI is enabled it will delete any references to the NAPI struct prior
2817  * to freeing the q_vector.
2818  **/
2819 static void ixgbevf_free_q_vector(struct ixgbevf_adapter *adapter, int v_idx)
2820 {
2821         struct ixgbevf_q_vector *q_vector = adapter->q_vector[v_idx];
2822         struct ixgbevf_ring *ring;
2823 
2824         ixgbevf_for_each_ring(ring, q_vector->tx) {
2825                 if (ring_is_xdp(ring))
2826                         adapter->xdp_ring[ring->queue_index] = NULL;
2827                 else
2828                         adapter->tx_ring[ring->queue_index] = NULL;
2829         }
2830 
2831         ixgbevf_for_each_ring(ring, q_vector->rx)
2832                 adapter->rx_ring[ring->queue_index] = NULL;
2833 
2834         adapter->q_vector[v_idx] = NULL;
2835         netif_napi_del(&q_vector->napi);
2836 
2837         /* ixgbevf_get_stats() might access the rings on this vector,
2838          * we must wait a grace period before freeing it.
2839          */
2840         kfree_rcu(q_vector, rcu);
2841 }
2842 
2843 /**
2844  * ixgbevf_alloc_q_vectors - Allocate memory for interrupt vectors
2845  * @adapter: board private structure to initialize
2846  *
2847  * We allocate one q_vector per queue interrupt.  If allocation fails we
2848  * return -ENOMEM.
2849  **/
2850 static int ixgbevf_alloc_q_vectors(struct ixgbevf_adapter *adapter)
2851 {
2852         int q_vectors = adapter->num_msix_vectors - NON_Q_VECTORS;
2853         int rxr_remaining = adapter->num_rx_queues;
2854         int txr_remaining = adapter->num_tx_queues;
2855         int xdp_remaining = adapter->num_xdp_queues;
2856         int rxr_idx = 0, txr_idx = 0, xdp_idx = 0, v_idx = 0;
2857         int err;
2858 
2859         if (q_vectors >= (rxr_remaining + txr_remaining + xdp_remaining)) {
2860                 for (; rxr_remaining; v_idx++, q_vectors--) {
2861                         int rqpv = DIV_ROUND_UP(rxr_remaining, q_vectors);
2862 
2863                         err = ixgbevf_alloc_q_vector(adapter, v_idx,
2864                                                      0, 0, 0, 0, rqpv, rxr_idx);
2865                         if (err)
2866                                 goto err_out;
2867 
2868                         /* update counts and index */
2869                         rxr_remaining -= rqpv;
2870                         rxr_idx += rqpv;
2871                 }
2872         }
2873 
2874         for (; q_vectors; v_idx++, q_vectors--) {
2875                 int rqpv = DIV_ROUND_UP(rxr_remaining, q_vectors);
2876                 int tqpv = DIV_ROUND_UP(txr_remaining, q_vectors);
2877                 int xqpv = DIV_ROUND_UP(xdp_remaining, q_vectors);
2878 
2879                 err = ixgbevf_alloc_q_vector(adapter, v_idx,
2880                                              tqpv, txr_idx,
2881                                              xqpv, xdp_idx,
2882                                              rqpv, rxr_idx);
2883 
2884                 if (err)
2885                         goto err_out;
2886 
2887                 /* update counts and index */
2888                 rxr_remaining -= rqpv;
2889                 rxr_idx += rqpv;
2890                 txr_remaining -= tqpv;
2891                 txr_idx += tqpv;
2892                 xdp_remaining -= xqpv;
2893                 xdp_idx += xqpv;
2894         }
2895 
2896         return 0;
2897 
2898 err_out:
2899         while (v_idx) {
2900                 v_idx--;
2901                 ixgbevf_free_q_vector(adapter, v_idx);
2902         }
2903 
2904         return -ENOMEM;
2905 }
2906 
2907 /**
2908  * ixgbevf_free_q_vectors - Free memory allocated for interrupt vectors
2909  * @adapter: board private structure to initialize
2910  *
2911  * This function frees the memory allocated to the q_vectors.  In addition if
2912  * NAPI is enabled it will delete any references to the NAPI struct prior
2913  * to freeing the q_vector.
2914  **/
2915 static void ixgbevf_free_q_vectors(struct ixgbevf_adapter *adapter)
2916 {
2917         int q_vectors = adapter->num_msix_vectors - NON_Q_VECTORS;
2918 
2919         while (q_vectors) {
2920                 q_vectors--;
2921                 ixgbevf_free_q_vector(adapter, q_vectors);
2922         }
2923 }
2924 
2925 /**
2926  * ixgbevf_reset_interrupt_capability - Reset MSIX setup
2927  * @adapter: board private structure
2928  *
2929  **/
2930 static void ixgbevf_reset_interrupt_capability(struct ixgbevf_adapter *adapter)
2931 {
2932         if (!adapter->msix_entries)
2933                 return;
2934 
2935         pci_disable_msix(adapter->pdev);
2936         kfree(adapter->msix_entries);
2937         adapter->msix_entries = NULL;
2938 }
2939 
2940 /**
2941  * ixgbevf_init_interrupt_scheme - Determine if MSIX is supported and init
2942  * @adapter: board private structure to initialize
2943  *
2944  **/
2945 static int ixgbevf_init_interrupt_scheme(struct ixgbevf_adapter *adapter)
2946 {
2947         int err;
2948 
2949         /* Number of supported queues */
2950         ixgbevf_set_num_queues(adapter);
2951 
2952         err = ixgbevf_set_interrupt_capability(adapter);
2953         if (err) {
2954                 hw_dbg(&adapter->hw,
2955                        "Unable to setup interrupt capabilities\n");
2956                 goto err_set_interrupt;
2957         }
2958 
2959         err = ixgbevf_alloc_q_vectors(adapter);
2960         if (err) {
2961                 hw_dbg(&adapter->hw, "Unable to allocate memory for queue vectors\n");
2962                 goto err_alloc_q_vectors;
2963         }
2964 
2965         hw_dbg(&adapter->hw, "Multiqueue %s: Rx Queue count = %u, Tx Queue count = %u XDP Queue count %u\n",
2966                (adapter->num_rx_queues > 1) ? "Enabled" : "Disabled",
2967                adapter->num_rx_queues, adapter->num_tx_queues,
2968                adapter->num_xdp_queues);
2969 
2970         set_bit(__IXGBEVF_DOWN, &adapter->state);
2971 
2972         return 0;
2973 err_alloc_q_vectors:
2974         ixgbevf_reset_interrupt_capability(adapter);
2975 err_set_interrupt:
2976         return err;
2977 }
2978 
2979 /**
2980  * ixgbevf_clear_interrupt_scheme - Clear the current interrupt scheme settings
2981  * @adapter: board private structure to clear interrupt scheme on
2982  *
2983  * We go through and clear interrupt specific resources and reset the structure
2984  * to pre-load conditions
2985  **/
2986 static void ixgbevf_clear_interrupt_scheme(struct ixgbevf_adapter *adapter)
2987 {
2988         adapter->num_tx_queues = 0;
2989         adapter->num_xdp_queues = 0;
2990         adapter->num_rx_queues = 0;
2991 
2992         ixgbevf_free_q_vectors(adapter);
2993         ixgbevf_reset_interrupt_capability(adapter);
2994 }
2995 
2996 /**
2997  * ixgbevf_sw_init - Initialize general software structures
2998  * @adapter: board private structure to initialize
2999  *
3000  * ixgbevf_sw_init initializes the Adapter private data structure.
3001  * Fields are initialized based on PCI device information and
3002  * OS network device settings (MTU size).
3003  **/
3004 static int ixgbevf_sw_init(struct ixgbevf_adapter *adapter)
3005 {
3006         struct ixgbe_hw *hw = &adapter->hw;
3007         struct pci_dev *pdev = adapter->pdev;
3008         struct net_device *netdev = adapter->netdev;
3009         int err;
3010 
3011         /* PCI config space info */
3012         hw->vendor_id = pdev->vendor;
3013         hw->device_id = pdev->device;
3014         hw->revision_id = pdev->revision;
3015         hw->subsystem_vendor_id = pdev->subsystem_vendor;
3016         hw->subsystem_device_id = pdev->subsystem_device;
3017 
3018         hw->mbx.ops.init_params(hw);
3019 
3020         if (hw->mac.type >= ixgbe_mac_X550_vf) {
3021                 err = ixgbevf_init_rss_key(adapter);
3022                 if (err)
3023                         goto out;
3024         }
3025 
3026         /* assume legacy case in which PF would only give VF 2 queues */
3027         hw->mac.max_tx_queues = 2;
3028         hw->mac.max_rx_queues = 2;
3029 
3030         /* lock to protect mailbox accesses */
3031         spin_lock_init(&adapter->mbx_lock);
3032 
3033         err = hw->mac.ops.reset_hw(hw);
3034         if (err) {
3035                 dev_info(&pdev->dev,
3036                          "PF still in reset state.  Is the PF interface up?\n");
3037         } else {
3038                 err = hw->mac.ops.init_hw(hw);
3039                 if (err) {
3040                         pr_err("init_shared_code failed: %d\n", err);
3041                         goto out;
3042                 }
3043                 ixgbevf_negotiate_api(adapter);
3044                 err = hw->mac.ops.get_mac_addr(hw, hw->mac.addr);
3045                 if (err)
3046                         dev_info(&pdev->dev, "Error reading MAC address\n");
3047                 else if (is_zero_ether_addr(adapter->hw.mac.addr))
3048                         dev_info(&pdev->dev,
3049                                  "MAC address not assigned by administrator.\n");
3050                 ether_addr_copy(netdev->dev_addr, hw->mac.addr);
3051         }
3052 
3053         if (!is_valid_ether_addr(netdev->dev_addr)) {
3054                 dev_info(&pdev->dev, "Assigning random MAC address\n");
3055                 eth_hw_addr_random(netdev);
3056                 ether_addr_copy(hw->mac.addr, netdev->dev_addr);
3057                 ether_addr_copy(hw->mac.perm_addr, netdev->dev_addr);
3058         }
3059 
3060         /* Enable dynamic interrupt throttling rates */
3061         adapter->rx_itr_setting = 1;
3062         adapter->tx_itr_setting = 1;
3063 
3064         /* set default ring sizes */
3065         adapter->tx_ring_count = IXGBEVF_DEFAULT_TXD;
3066         adapter->rx_ring_count = IXGBEVF_DEFAULT_RXD;
3067 
3068         set_bit(__IXGBEVF_DOWN, &adapter->state);
3069         return 0;
3070 
3071 out:
3072         return err;
3073 }
3074 
3075 #define UPDATE_VF_COUNTER_32bit(reg, last_counter, counter)     \
3076         {                                                       \
3077                 u32 current_counter = IXGBE_READ_REG(hw, reg);  \
3078                 if (current_counter < last_counter)             \
3079                         counter += 0x100000000LL;               \
3080                 last_counter = current_counter;                 \
3081                 counter &= 0xFFFFFFFF00000000LL;                \
3082                 counter |= current_counter;                     \
3083         }
3084 
3085 #define UPDATE_VF_COUNTER_36bit(reg_lsb, reg_msb, last_counter, counter) \
3086         {                                                                \
3087                 u64 current_counter_lsb = IXGBE_READ_REG(hw, reg_lsb);   \
3088                 u64 current_counter_msb = IXGBE_READ_REG(hw, reg_msb);   \
3089                 u64 current_counter = (current_counter_msb << 32) |      \
3090                         current_counter_lsb;                             \
3091                 if (current_counter < last_counter)                      \
3092                         counter += 0x1000000000LL;                       \
3093                 last_counter = current_counter;                          \
3094                 counter &= 0xFFFFFFF000000000LL;                         \
3095                 counter |= current_counter;                              \
3096         }
3097 /**
3098  * ixgbevf_update_stats - Update the board statistics counters.
3099  * @adapter: board private structure
3100  **/
3101 void ixgbevf_update_stats(struct ixgbevf_adapter *adapter)
3102 {
3103         struct ixgbe_hw *hw = &adapter->hw;
3104         u64 alloc_rx_page_failed = 0, alloc_rx_buff_failed = 0;
3105         u64 alloc_rx_page = 0, hw_csum_rx_error = 0;
3106         int i;
3107 
3108         if (test_bit(__IXGBEVF_DOWN, &adapter->state) ||
3109             test_bit(__IXGBEVF_RESETTING, &adapter->state))
3110                 return;
3111 
3112         UPDATE_VF_COUNTER_32bit(IXGBE_VFGPRC, adapter->stats.last_vfgprc,
3113                                 adapter->stats.vfgprc);
3114         UPDATE_VF_COUNTER_32bit(IXGBE_VFGPTC, adapter->stats.last_vfgptc,
3115                                 adapter->stats.vfgptc);
3116         UPDATE_VF_COUNTER_36bit(IXGBE_VFGORC_LSB, IXGBE_VFGORC_MSB,
3117                                 adapter->stats.last_vfgorc,
3118                                 adapter->stats.vfgorc);
3119         UPDATE_VF_COUNTER_36bit(IXGBE_VFGOTC_LSB, IXGBE_VFGOTC_MSB,
3120                                 adapter->stats.last_vfgotc,
3121                                 adapter->stats.vfgotc);
3122         UPDATE_VF_COUNTER_32bit(IXGBE_VFMPRC, adapter->stats.last_vfmprc,
3123                                 adapter->stats.vfmprc);
3124 
3125         for (i = 0;  i  < adapter->num_rx_queues;  i++) {
3126                 struct ixgbevf_ring *rx_ring = adapter->rx_ring[i];
3127 
3128                 hw_csum_rx_error += rx_ring->rx_stats.csum_err;
3129                 alloc_rx_page_failed += rx_ring->rx_stats.alloc_rx_page_failed;
3130                 alloc_rx_buff_failed += rx_ring->rx_stats.alloc_rx_buff_failed;
3131                 alloc_rx_page += rx_ring->rx_stats.alloc_rx_page;
3132         }
3133 
3134         adapter->hw_csum_rx_error = hw_csum_rx_error;
3135         adapter->alloc_rx_page_failed = alloc_rx_page_failed;
3136         adapter->alloc_rx_buff_failed = alloc_rx_buff_failed;
3137         adapter->alloc_rx_page = alloc_rx_page;
3138 }
3139 
3140 /**
3141  * ixgbevf_service_timer - Timer Call-back
3142  * @t: pointer to timer_list struct
3143  **/
3144 static void ixgbevf_service_timer(struct timer_list *t)
3145 {
3146         struct ixgbevf_adapter *adapter = from_timer(adapter, t,
3147                                                      service_timer);
3148 
3149         /* Reset the timer */
3150         mod_timer(&adapter->service_timer, (HZ * 2) + jiffies);
3151 
3152         ixgbevf_service_event_schedule(adapter);
3153 }
3154 
3155 static void ixgbevf_reset_subtask(struct ixgbevf_adapter *adapter)
3156 {
3157         if (!test_and_clear_bit(__IXGBEVF_RESET_REQUESTED, &adapter->state))
3158                 return;
3159 
3160         rtnl_lock();
3161         /* If we're already down or resetting, just bail */
3162         if (test_bit(__IXGBEVF_DOWN, &adapter->state) ||
3163             test_bit(__IXGBEVF_REMOVING, &adapter->state) ||
3164             test_bit(__IXGBEVF_RESETTING, &adapter->state)) {
3165                 rtnl_unlock();
3166                 return;
3167         }
3168 
3169         adapter->tx_timeout_count++;
3170 
3171         ixgbevf_reinit_locked(adapter);
3172         rtnl_unlock();
3173 }
3174 
3175 /**
3176  * ixgbevf_check_hang_subtask - check for hung queues and dropped interrupts
3177  * @adapter: pointer to the device adapter structure
3178  *
3179  * This function serves two purposes.  First it strobes the interrupt lines
3180  * in order to make certain interrupts are occurring.  Secondly it sets the
3181  * bits needed to check for TX hangs.  As a result we should immediately
3182  * determine if a hang has occurred.
3183  **/
3184 static void ixgbevf_check_hang_subtask(struct ixgbevf_adapter *adapter)
3185 {
3186         struct ixgbe_hw *hw = &adapter->hw;
3187         u32 eics = 0;
3188         int i;
3189 
3190         /* If we're down or resetting, just bail */
3191         if (test_bit(__IXGBEVF_DOWN, &adapter->state) ||
3192             test_bit(__IXGBEVF_RESETTING, &adapter->state))
3193                 return;
3194 
3195         /* Force detection of hung controller */
3196         if (netif_carrier_ok(adapter->netdev)) {
3197                 for (i = 0; i < adapter->num_tx_queues; i++)
3198                         set_check_for_tx_hang(adapter->tx_ring[i]);
3199                 for (i = 0; i < adapter->num_xdp_queues; i++)
3200                         set_check_for_tx_hang(adapter->xdp_ring[i]);
3201         }
3202 
3203         /* get one bit for every active Tx/Rx interrupt vector */
3204         for (i = 0; i < adapter->num_msix_vectors - NON_Q_VECTORS; i++) {
3205                 struct ixgbevf_q_vector *qv = adapter->q_vector[i];
3206 
3207                 if (qv->rx.ring || qv->tx.ring)
3208                         eics |= BIT(i);
3209         }
3210 
3211         /* Cause software interrupt to ensure rings are cleaned */
3212         IXGBE_WRITE_REG(hw, IXGBE_VTEICS, eics);
3213 }
3214 
3215 /**
3216  * ixgbevf_watchdog_update_link - update the link status
3217  * @adapter: pointer to the device adapter structure
3218  **/
3219 static void ixgbevf_watchdog_update_link(struct ixgbevf_adapter *adapter)
3220 {
3221         struct ixgbe_hw *hw = &adapter->hw;
3222         u32 link_speed = adapter->link_speed;
3223         bool link_up = adapter->link_up;
3224         s32 err;
3225 
3226         spin_lock_bh(&adapter->mbx_lock);
3227 
3228         err = hw->mac.ops.check_link(hw, &link_speed, &link_up, false);
3229 
3230         spin_unlock_bh(&adapter->mbx_lock);
3231 
3232         /* if check for link returns error we will need to reset */
3233         if (err && time_after(jiffies, adapter->last_reset + (10 * HZ))) {
3234                 set_bit(__IXGBEVF_RESET_REQUESTED, &adapter->state);
3235                 link_up = false;
3236         }
3237 
3238         adapter->link_up = link_up;
3239         adapter->link_speed = link_speed;
3240 }
3241 
3242 /**
3243  * ixgbevf_watchdog_link_is_up - update netif_carrier status and
3244  *                               print link up message
3245  * @adapter: pointer to the device adapter structure
3246  **/
3247 static void ixgbevf_watchdog_link_is_up(struct ixgbevf_adapter *adapter)
3248 {
3249         struct net_device *netdev = adapter->netdev;
3250 
3251         /* only continue if link was previously down */
3252         if (netif_carrier_ok(netdev))
3253                 return;
3254 
3255         dev_info(&adapter->pdev->dev, "NIC Link is Up %s\n",
3256                  (adapter->link_speed == IXGBE_LINK_SPEED_10GB_FULL) ?
3257                  "10 Gbps" :
3258                  (adapter->link_speed == IXGBE_LINK_SPEED_1GB_FULL) ?
3259                  "1 Gbps" :
3260                  (adapter->link_speed == IXGBE_LINK_SPEED_100_FULL) ?
3261                  "100 Mbps" :
3262                  "unknown speed");
3263 
3264         netif_carrier_on(netdev);
3265 }
3266 
3267 /**
3268  * ixgbevf_watchdog_link_is_down - update netif_carrier status and
3269  *                                 print link down message
3270  * @adapter: pointer to the adapter structure
3271  **/
3272 static void ixgbevf_watchdog_link_is_down(struct ixgbevf_adapter *adapter)
3273 {
3274         struct net_device *netdev = adapter->netdev;
3275 
3276         adapter->link_speed = 0;
3277 
3278         /* only continue if link was up previously */
3279         if (!netif_carrier_ok(netdev))
3280                 return;
3281 
3282         dev_info(&adapter->pdev->dev, "NIC Link is Down\n");
3283 
3284         netif_carrier_off(netdev);
3285 }
3286 
3287 /**
3288  * ixgbevf_watchdog_subtask - worker thread to bring link up
3289  * @adapter: board private structure
3290  **/
3291 static void ixgbevf_watchdog_subtask(struct ixgbevf_adapter *adapter)
3292 {
3293         /* if interface is down do nothing */
3294         if (test_bit(__IXGBEVF_DOWN, &adapter->state) ||
3295             test_bit(__IXGBEVF_RESETTING, &adapter->state))
3296                 return;
3297 
3298         ixgbevf_watchdog_update_link(adapter);
3299 
3300         if (adapter->link_up)
3301                 ixgbevf_watchdog_link_is_up(adapter);
3302         else
3303                 ixgbevf_watchdog_link_is_down(adapter);
3304 
3305         ixgbevf_update_stats(adapter);
3306 }
3307 
3308 /**
3309  * ixgbevf_service_task - manages and runs subtasks
3310  * @work: pointer to work_struct containing our data
3311  **/
3312 static void ixgbevf_service_task(struct work_struct *work)
3313 {
3314         struct ixgbevf_adapter *adapter = container_of(work,
3315                                                        struct ixgbevf_adapter,
3316                                                        service_task);
3317         struct ixgbe_hw *hw = &adapter->hw;
3318 
3319         if (IXGBE_REMOVED(hw->hw_addr)) {
3320                 if (!test_bit(__IXGBEVF_DOWN, &adapter->state)) {
3321                         rtnl_lock();
3322                         ixgbevf_down(adapter);
3323                         rtnl_unlock();
3324                 }
3325                 return;
3326         }
3327 
3328         ixgbevf_queue_reset_subtask(adapter);
3329         ixgbevf_reset_subtask(adapter);
3330         ixgbevf_watchdog_subtask(adapter);
3331         ixgbevf_check_hang_subtask(adapter);
3332 
3333         ixgbevf_service_event_complete(adapter);
3334 }
3335 
3336 /**
3337  * ixgbevf_free_tx_resources - Free Tx Resources per Queue
3338  * @tx_ring: Tx descriptor ring for a specific queue
3339  *
3340  * Free all transmit software resources
3341  **/
3342 void ixgbevf_free_tx_resources(struct ixgbevf_ring *tx_ring)
3343 {
3344         ixgbevf_clean_tx_ring(tx_ring);
3345 
3346         vfree(tx_ring->tx_buffer_info);
3347         tx_ring->tx_buffer_info = NULL;
3348 
3349         /* if not set, then don't free */
3350         if (!tx_ring->desc)
3351                 return;
3352 
3353         dma_free_coherent(tx_ring->dev, tx_ring->size, tx_ring->desc,
3354                           tx_ring->dma);
3355 
3356         tx_ring->desc = NULL;
3357 }
3358 
3359 /**
3360  * ixgbevf_free_all_tx_resources - Free Tx Resources for All Queues
3361  * @adapter: board private structure
3362  *
3363  * Free all transmit software resources
3364  **/
3365 static void ixgbevf_free_all_tx_resources(struct ixgbevf_adapter *adapter)
3366 {
3367         int i;
3368 
3369         for (i = 0; i < adapter->num_tx_queues; i++)
3370                 if (adapter->tx_ring[i]->desc)
3371                         ixgbevf_free_tx_resources(adapter->tx_ring[i]);
3372         for (i = 0; i < adapter->num_xdp_queues; i++)
3373                 if (adapter->xdp_ring[i]->desc)
3374                         ixgbevf_free_tx_resources(adapter->xdp_ring[i]);
3375 }
3376 
3377 /**
3378  * ixgbevf_setup_tx_resources - allocate Tx resources (Descriptors)
3379  * @tx_ring: Tx descriptor ring (for a specific queue) to setup
3380  *
3381  * Return 0 on success, negative on failure
3382  **/
3383 int ixgbevf_setup_tx_resources(struct ixgbevf_ring *tx_ring)
3384 {
3385         struct ixgbevf_adapter *adapter = netdev_priv(tx_ring->netdev);
3386         int size;
3387 
3388         size = sizeof(struct ixgbevf_tx_buffer) * tx_ring->count;
3389         tx_ring->tx_buffer_info = vmalloc(size);
3390         if (!tx_ring->tx_buffer_info)
3391                 goto err;
3392 
3393         u64_stats_init(&tx_ring->syncp);
3394 
3395         /* round up to nearest 4K */
3396         tx_ring->size = tx_ring->count * sizeof(union ixgbe_adv_tx_desc);
3397         tx_ring->size = ALIGN(tx_ring->size, 4096);
3398 
3399         tx_ring->desc = dma_alloc_coherent(tx_ring->dev, tx_ring->size,
3400                                            &tx_ring->dma, GFP_KERNEL);
3401         if (!tx_ring->desc)
3402                 goto err;
3403 
3404         return 0;
3405 
3406 err:
3407         vfree(tx_ring->tx_buffer_info);
3408         tx_ring->tx_buffer_info = NULL;
3409         hw_dbg(&adapter->hw, "Unable to allocate memory for the transmit descriptor ring\n");
3410         return -ENOMEM;
3411 }
3412 
3413 /**
3414  * ixgbevf_setup_all_tx_resources - allocate all queues Tx resources
3415  * @adapter: board private structure
3416  *
3417  * If this function returns with an error, then it's possible one or
3418  * more of the rings is populated (while the rest are not).  It is the
3419  * callers duty to clean those orphaned rings.
3420  *
3421  * Return 0 on success, negative on failure
3422  **/
3423 static int ixgbevf_setup_all_tx_resources(struct ixgbevf_adapter *adapter)
3424 {
3425         int i, j = 0, err = 0;
3426 
3427         for (i = 0; i < adapter->num_tx_queues; i++) {
3428                 err = ixgbevf_setup_tx_resources(adapter->tx_ring[i]);
3429                 if (!err)
3430                         continue;
3431                 hw_dbg(&adapter->hw, "Allocation for Tx Queue %u failed\n", i);
3432                 goto err_setup_tx;
3433         }
3434 
3435         for (j = 0; j < adapter->num_xdp_queues; j++) {
3436                 err = ixgbevf_setup_tx_resources(adapter->xdp_ring[j]);
3437                 if (!err)
3438                         continue;
3439                 hw_dbg(&adapter->hw, "Allocation for XDP Queue %u failed\n", j);
3440                 goto err_setup_tx;
3441         }
3442 
3443         return 0;
3444 err_setup_tx:
3445         /* rewind the index freeing the rings as we go */
3446         while (j--)
3447                 ixgbevf_free_tx_resources(adapter->xdp_ring[j]);
3448         while (i--)
3449                 ixgbevf_free_tx_resources(adapter->tx_ring[i]);
3450 
3451         return err;
3452 }
3453 
3454 /**
3455  * ixgbevf_setup_rx_resources - allocate Rx resources (Descriptors)
3456  * @adapter: board private structure
3457  * @rx_ring: Rx descriptor ring (for a specific queue) to setup
3458  *
3459  * Returns 0 on success, negative on failure
3460  **/
3461 int ixgbevf_setup_rx_resources(struct ixgbevf_adapter *adapter,
3462                                struct ixgbevf_ring *rx_ring)
3463 {
3464         int size;
3465 
3466         size = sizeof(struct ixgbevf_rx_buffer) * rx_ring->count;
3467         rx_ring->rx_buffer_info = vmalloc(size);
3468         if (!rx_ring->rx_buffer_info)
3469                 goto err;
3470 
3471         u64_stats_init(&rx_ring->syncp);
3472 
3473         /* Round up to nearest 4K */
3474         rx_ring->size = rx_ring->count * sizeof(union ixgbe_adv_rx_desc);
3475         rx_ring->size = ALIGN(rx_ring->size, 4096);
3476 
3477         rx_ring->desc = dma_alloc_coherent(rx_ring->dev, rx_ring->size,
3478                                            &rx_ring->dma, GFP_KERNEL);
3479 
3480         if (!rx_ring->desc)
3481                 goto err;
3482 
3483         /* XDP RX-queue info */
3484         if (xdp_rxq_info_reg(&rx_ring->xdp_rxq, adapter->netdev,
3485                              rx_ring->queue_index) < 0)
3486                 goto err;
3487 
3488         rx_ring->xdp_prog = adapter->xdp_prog;
3489 
3490         return 0;
3491 err:
3492         vfree(rx_ring->rx_buffer_info);
3493         rx_ring->rx_buffer_info = NULL;
3494         dev_err(rx_ring->dev, "Unable to allocate memory for the Rx descriptor ring\n");
3495         return -ENOMEM;
3496 }
3497 
3498 /**
3499  * ixgbevf_setup_all_rx_resources - allocate all queues Rx resources
3500  * @adapter: board private structure
3501  *
3502  * If this function returns with an error, then it's possible one or
3503  * more of the rings is populated (while the rest are not).  It is the
3504  * callers duty to clean those orphaned rings.
3505  *
3506  * Return 0 on success, negative on failure
3507  **/
3508 static int ixgbevf_setup_all_rx_resources(struct ixgbevf_adapter *adapter)
3509 {
3510         int i, err = 0;
3511 
3512         for (i = 0; i < adapter->num_rx_queues; i++) {
3513                 err = ixgbevf_setup_rx_resources(adapter, adapter->rx_ring[i]);
3514                 if (!err)
3515                         continue;
3516                 hw_dbg(&adapter->hw, "Allocation for Rx Queue %u failed\n", i);
3517                 goto err_setup_rx;
3518         }
3519 
3520         return 0;
3521 err_setup_rx:
3522         /* rewind the index freeing the rings as we go */
3523         while (i--)
3524                 ixgbevf_free_rx_resources(adapter->rx_ring[i]);
3525         return err;
3526 }
3527 
3528 /**
3529  * ixgbevf_free_rx_resources - Free Rx Resources
3530  * @rx_ring: ring to clean the resources from
3531  *
3532  * Free all receive software resources
3533  **/
3534 void ixgbevf_free_rx_resources(struct ixgbevf_ring *rx_ring)
3535 {
3536         ixgbevf_clean_rx_ring(rx_ring);
3537 
3538         rx_ring->xdp_prog = NULL;
3539         xdp_rxq_info_unreg(&rx_ring->xdp_rxq);
3540         vfree(rx_ring->rx_buffer_info);
3541         rx_ring->rx_buffer_info = NULL;
3542 
3543         dma_free_coherent(rx_ring->dev, rx_ring->size, rx_ring->desc,
3544                           rx_ring->dma);
3545 
3546         rx_ring->desc = NULL;
3547 }
3548 
3549 /**
3550  * ixgbevf_free_all_rx_resources - Free Rx Resources for All Queues
3551  * @adapter: board private structure
3552  *
3553  * Free all receive software resources
3554  **/
3555 static void ixgbevf_free_all_rx_resources(struct ixgbevf_adapter *adapter)
3556 {
3557         int i;
3558 
3559         for (i = 0; i < adapter->num_rx_queues; i++)
3560                 if (adapter->rx_ring[i]->desc)
3561                         ixgbevf_free_rx_resources(adapter->rx_ring[i]);
3562 }
3563 
3564 /**
3565  * ixgbevf_open - Called when a network interface is made active
3566  * @netdev: network interface device structure
3567  *
3568  * Returns 0 on success, negative value on failure
3569  *
3570  * The open entry point is called when a network interface is made
3571  * active by the system (IFF_UP).  At this point all resources needed
3572  * for transmit and receive operations are allocated, the interrupt
3573  * handler is registered with the OS, the watchdog timer is started,
3574  * and the stack is notified that the interface is ready.
3575  **/
3576 int ixgbevf_open(struct net_device *netdev)
3577 {
3578         struct ixgbevf_adapter *adapter = netdev_priv(netdev);
3579         struct ixgbe_hw *hw = &adapter->hw;
3580         int err;
3581 
3582         /* A previous failure to open the device because of a lack of
3583          * available MSIX vector resources may have reset the number
3584          * of msix vectors variable to zero.  The only way to recover
3585          * is to unload/reload the driver and hope that the system has
3586          * been able to recover some MSIX vector resources.
3587          */
3588         if (!adapter->num_msix_vectors)
3589                 return -ENOMEM;
3590 
3591         if (hw->adapter_stopped) {
3592                 ixgbevf_reset(adapter);
3593                 /* if adapter is still stopped then PF isn't up and
3594                  * the VF can't start.
3595                  */
3596                 if (hw->adapter_stopped) {
3597                         err = IXGBE_ERR_MBX;
3598                         pr_err("Unable to start - perhaps the PF Driver isn't up yet\n");
3599                         goto err_setup_reset;
3600                 }
3601         }
3602 
3603         /* disallow open during test */
3604         if (test_bit(__IXGBEVF_TESTING, &adapter->state))
3605                 return -EBUSY;
3606 
3607         netif_carrier_off(netdev);
3608 
3609         /* allocate transmit descriptors */
3610         err = ixgbevf_setup_all_tx_resources(adapter);
3611         if (err)
3612                 goto err_setup_tx;
3613 
3614         /* allocate receive descriptors */
3615         err = ixgbevf_setup_all_rx_resources(adapter);
3616         if (err)
3617                 goto err_setup_rx;
3618 
3619         ixgbevf_configure(adapter);
3620 
3621         err = ixgbevf_request_irq(adapter);
3622         if (err)
3623                 goto err_req_irq;
3624 
3625         /* Notify the stack of the actual queue counts. */
3626         err = netif_set_real_num_tx_queues(netdev, adapter->num_tx_queues);
3627         if (err)
3628                 goto err_set_queues;
3629 
3630         err = netif_set_real_num_rx_queues(netdev, adapter->num_rx_queues);
3631         if (err)
3632                 goto err_set_queues;
3633 
3634         ixgbevf_up_complete(adapter);
3635 
3636         return 0;
3637 
3638 err_set_queues:
3639         ixgbevf_free_irq(adapter);
3640 err_req_irq:
3641         ixgbevf_free_all_rx_resources(adapter);
3642 err_setup_rx:
3643         ixgbevf_free_all_tx_resources(adapter);
3644 err_setup_tx:
3645         ixgbevf_reset(adapter);
3646 err_setup_reset:
3647 
3648         return err;
3649 }
3650 
3651 /**
3652  * ixgbevf_close_suspend - actions necessary to both suspend and close flows
3653  * @adapter: the private adapter struct
3654  *
3655  * This function should contain the necessary work common to both suspending
3656  * and closing of the device.
3657  */
3658 static void ixgbevf_close_suspend(struct ixgbevf_adapter *adapter)
3659 {
3660         ixgbevf_down(adapter);
3661         ixgbevf_free_irq(adapter);
3662         ixgbevf_free_all_tx_resources(adapter);
3663         ixgbevf_free_all_rx_resources(adapter);
3664 }
3665 
3666 /**
3667  * ixgbevf_close - Disables a network interface
3668  * @netdev: network interface device structure
3669  *
3670  * Returns 0, this is not allowed to fail
3671  *
3672  * The close entry point is called when an interface is de-activated
3673  * by the OS.  The hardware is still under the drivers control, but
3674  * needs to be disabled.  A global MAC reset is issued to stop the
3675  * hardware, and all transmit and receive resources are freed.
3676  **/
3677 int ixgbevf_close(struct net_device *netdev)
3678 {
3679         struct ixgbevf_adapter *adapter = netdev_priv(netdev);
3680 
3681         if (netif_device_present(netdev))
3682                 ixgbevf_close_suspend(adapter);
3683 
3684         return 0;
3685 }
3686 
3687 static void ixgbevf_queue_reset_subtask(struct ixgbevf_adapter *adapter)
3688 {
3689         struct net_device *dev = adapter->netdev;
3690 
3691         if (!test_and_clear_bit(__IXGBEVF_QUEUE_RESET_REQUESTED,
3692                                 &adapter->state))
3693                 return;
3694 
3695         /* if interface is down do nothing */
3696         if (test_bit(__IXGBEVF_DOWN, &adapter->state) ||
3697             test_bit(__IXGBEVF_RESETTING, &adapter->state))
3698                 return;
3699 
3700         /* Hardware has to reinitialize queues and interrupts to
3701          * match packet buffer alignment. Unfortunately, the
3702          * hardware is not flexible enough to do this dynamically.
3703          */
3704         rtnl_lock();
3705 
3706         if (netif_running(dev))
3707                 ixgbevf_close(dev);
3708 
3709         ixgbevf_clear_interrupt_scheme(adapter);
3710         ixgbevf_init_interrupt_scheme(adapter);
3711 
3712         if (netif_running(dev))
3713                 ixgbevf_open(dev);
3714 
3715         rtnl_unlock();
3716 }
3717 
3718 static void ixgbevf_tx_ctxtdesc(struct ixgbevf_ring *tx_ring,
3719                                 u32 vlan_macip_lens, u32 fceof_saidx,
3720                                 u32 type_tucmd, u32 mss_l4len_idx)
3721 {
3722         struct ixgbe_adv_tx_context_desc *context_desc;
3723         u16 i = tx_ring->next_to_use;
3724 
3725         context_desc = IXGBEVF_TX_CTXTDESC(tx_ring, i);
3726 
3727         i++;
3728         tx_ring->next_to_use = (i < tx_ring->count) ? i : 0;
3729 
3730         /* set bits to identify this as an advanced context descriptor */
3731         type_tucmd |= IXGBE_TXD_CMD_DEXT | IXGBE_ADVTXD_DTYP_CTXT;
3732 
3733         context_desc->vlan_macip_lens   = cpu_to_le32(vlan_macip_lens);
3734         context_desc->fceof_saidx       = cpu_to_le32(fceof_saidx);
3735         context_desc->type_tucmd_mlhl   = cpu_to_le32(type_tucmd);
3736         context_desc->mss_l4len_idx     = cpu_to_le32(mss_l4len_idx);
3737 }
3738 
3739 static int ixgbevf_tso(struct ixgbevf_ring *tx_ring,
3740                        struct ixgbevf_tx_buffer *first,
3741                        u8 *hdr_len,
3742                        struct ixgbevf_ipsec_tx_data *itd)
3743 {
3744         u32 vlan_macip_lens, type_tucmd, mss_l4len_idx;
3745         struct sk_buff *skb = first->skb;
3746         union {
3747                 struct iphdr *v4;
3748                 struct ipv6hdr *v6;
3749                 unsigned char *hdr;
3750         } ip;
3751         union {
3752                 struct tcphdr *tcp;
3753                 unsigned char *hdr;
3754         } l4;
3755         u32 paylen, l4_offset;
3756         u32 fceof_saidx = 0;
3757         int err;
3758 
3759         if (skb->ip_summed != CHECKSUM_PARTIAL)
3760                 return 0;
3761 
3762         if (!skb_is_gso(skb))
3763                 return 0;
3764 
3765         err = skb_cow_head(skb, 0);
3766         if (err < 0)
3767                 return err;
3768 
3769         if (eth_p_mpls(first->protocol))
3770                 ip.hdr = skb_inner_network_header(skb);
3771         else
3772                 ip.hdr = skb_network_header(skb);
3773         l4.hdr = skb_checksum_start(skb);
3774 
3775         /* ADV DTYP TUCMD MKRLOC/ISCSIHEDLEN */
3776         type_tucmd = IXGBE_ADVTXD_TUCMD_L4T_TCP;
3777 
3778         /* initialize outer IP header fields */
3779         if (ip.v4->version == 4) {
3780                 unsigned char *csum_start = skb_checksum_start(skb);
3781                 unsigned char *trans_start = ip.hdr + (ip.v4->ihl * 4);
3782                 int len = csum_start - trans_start;
3783 
3784                 /* IP header will have to cancel out any data that
3785                  * is not a part of the outer IP header, so set to
3786                  * a reverse csum if needed, else init check to 0.
3787                  */
3788                 ip.v4->check = (skb_shinfo(skb)->gso_type & SKB_GSO_PARTIAL) ?
3789                                            csum_fold(csum_partial(trans_start,
3790                                                                   len, 0)) : 0;
3791                 type_tucmd |= IXGBE_ADVTXD_TUCMD_IPV4;
3792 
3793                 ip.v4->tot_len = 0;
3794                 first->tx_flags |= IXGBE_TX_FLAGS_TSO |
3795                                    IXGBE_TX_FLAGS_CSUM |
3796                                    IXGBE_TX_FLAGS_IPV4;
3797         } else {
3798                 ip.v6->payload_len = 0;
3799                 first->tx_flags |= IXGBE_TX_FLAGS_TSO |
3800                                    IXGBE_TX_FLAGS_CSUM;
3801         }
3802 
3803         /* determine offset of inner transport header */
3804         l4_offset = l4.hdr - skb->data;
3805 
3806         /* compute length of segmentation header */
3807         *hdr_len = (l4.tcp->doff * 4) + l4_offset;
3808 
3809         /* remove payload length from inner checksum */
3810         paylen = skb->len - l4_offset;
3811         csum_replace_by_diff(&l4.tcp->check, htonl(paylen));
3812 
3813         /* update gso size and bytecount with header size */
3814         first->gso_segs = skb_shinfo(skb)->gso_segs;
3815         first->bytecount += (first->gso_segs - 1) * *hdr_len;
3816 
3817         /* mss_l4len_id: use 1 as index for TSO */
3818         mss_l4len_idx = (*hdr_len - l4_offset) << IXGBE_ADVTXD_L4LEN_SHIFT;
3819         mss_l4len_idx |= skb_shinfo(skb)->gso_size << IXGBE_ADVTXD_MSS_SHIFT;
3820         mss_l4len_idx |= (1u << IXGBE_ADVTXD_IDX_SHIFT);
3821 
3822         fceof_saidx |= itd->pfsa;
3823         type_tucmd |= itd->flags | itd->trailer_len;
3824 
3825         /* vlan_macip_lens: HEADLEN, MACLEN, VLAN tag */
3826         vlan_macip_lens = l4.hdr - ip.hdr;
3827         vlan_macip_lens |= (ip.hdr - skb->data) << IXGBE_ADVTXD_MACLEN_SHIFT;
3828         vlan_macip_lens |= first->tx_flags & IXGBE_TX_FLAGS_VLAN_MASK;
3829 
3830         ixgbevf_tx_ctxtdesc(tx_ring, vlan_macip_lens, fceof_saidx, type_tucmd,
3831                             mss_l4len_idx);
3832 
3833         return 1;
3834 }
3835 
3836 static inline bool ixgbevf_ipv6_csum_is_sctp(struct sk_buff *skb)
3837 {
3838         unsigned int offset = 0;
3839 
3840         ipv6_find_hdr(skb, &offset, IPPROTO_SCTP, NULL, NULL);
3841 
3842         return offset == skb_checksum_start_offset(skb);
3843 }
3844 
3845 static void ixgbevf_tx_csum(struct ixgbevf_ring *tx_ring,
3846                             struct ixgbevf_tx_buffer *first,
3847                             struct ixgbevf_ipsec_tx_data *itd)
3848 {
3849         struct sk_buff *skb = first->skb;
3850         u32 vlan_macip_lens = 0;
3851         u32 fceof_saidx = 0;
3852         u32 type_tucmd = 0;
3853 
3854         if (skb->ip_summed != CHECKSUM_PARTIAL)
3855                 goto no_csum;
3856 
3857         switch (skb->csum_offset) {
3858         case offsetof(struct tcphdr, check):
3859                 type_tucmd = IXGBE_ADVTXD_TUCMD_L4T_TCP;
3860                 /* fall through */
3861         case offsetof(struct udphdr, check):
3862                 break;
3863         case offsetof(struct sctphdr, checksum):
3864                 /* validate that this is actually an SCTP request */
3865                 if (((first->protocol == htons(ETH_P_IP)) &&
3866                      (ip_hdr(skb)->protocol == IPPROTO_SCTP)) ||
3867                     ((first->protocol == htons(ETH_P_IPV6)) &&
3868                      ixgbevf_ipv6_csum_is_sctp(skb))) {
3869                         type_tucmd = IXGBE_ADVTXD_TUCMD_L4T_SCTP;
3870                         break;
3871                 }
3872                 /* fall through */
3873         default:
3874                 skb_checksum_help(skb);
3875                 goto no_csum;
3876         }
3877 
3878         if (first->protocol == htons(ETH_P_IP))
3879                 type_tucmd |= IXGBE_ADVTXD_TUCMD_IPV4;
3880 
3881         /* update TX checksum flag */
3882         first->tx_flags |= IXGBE_TX_FLAGS_CSUM;
3883         vlan_macip_lens = skb_checksum_start_offset(skb) -
3884                           skb_network_offset(skb);
3885 no_csum:
3886         /* vlan_macip_lens: MACLEN, VLAN tag */
3887         vlan_macip_lens |= skb_network_offset(skb) << IXGBE_ADVTXD_MACLEN_SHIFT;
3888         vlan_macip_lens |= first->tx_flags & IXGBE_TX_FLAGS_VLAN_MASK;
3889 
3890         fceof_saidx |= itd->pfsa;
3891         type_tucmd |= itd->flags | itd->trailer_len;
3892 
3893         ixgbevf_tx_ctxtdesc(tx_ring, vlan_macip_lens,
3894                             fceof_saidx, type_tucmd, 0);
3895 }
3896 
3897 static __le32 ixgbevf_tx_cmd_type(u32 tx_flags)
3898 {
3899         /* set type for advanced descriptor with frame checksum insertion */
3900         __le32 cmd_type = cpu_to_le32(IXGBE_ADVTXD_DTYP_DATA |
3901                                       IXGBE_ADVTXD_DCMD_IFCS |
3902                                       IXGBE_ADVTXD_DCMD_DEXT);
3903 
3904         /* set HW VLAN bit if VLAN is present */
3905         if (tx_flags & IXGBE_TX_FLAGS_VLAN)
3906                 cmd_type |= cpu_to_le32(IXGBE_ADVTXD_DCMD_VLE);
3907 
3908         /* set segmentation enable bits for TSO/FSO */
3909         if (tx_flags & IXGBE_TX_FLAGS_TSO)
3910                 cmd_type |= cpu_to_le32(IXGBE_ADVTXD_DCMD_TSE);
3911 
3912         return cmd_type;
3913 }
3914 
3915 static void ixgbevf_tx_olinfo_status(union ixgbe_adv_tx_desc *tx_desc,
3916                                      u32 tx_flags, unsigned int paylen)
3917 {
3918         __le32 olinfo_status = cpu_to_le32(paylen << IXGBE_ADVTXD_PAYLEN_SHIFT);
3919 
3920         /* enable L4 checksum for TSO and TX checksum offload */
3921         if (tx_flags & IXGBE_TX_FLAGS_CSUM)
3922                 olinfo_status |= cpu_to_le32(IXGBE_ADVTXD_POPTS_TXSM);
3923 
3924         /* enble IPv4 checksum for TSO */
3925         if (tx_flags & IXGBE_TX_FLAGS_IPV4)
3926                 olinfo_status |= cpu_to_le32(IXGBE_ADVTXD_POPTS_IXSM);
3927 
3928         /* enable IPsec */
3929         if (tx_flags & IXGBE_TX_FLAGS_IPSEC)
3930                 olinfo_status |= cpu_to_le32(IXGBE_ADVTXD_POPTS_IPSEC);
3931 
3932         /* use index 1 context for TSO/FSO/FCOE/IPSEC */
3933         if (tx_flags & (IXGBE_TX_FLAGS_TSO | IXGBE_TX_FLAGS_IPSEC))
3934                 olinfo_status |= cpu_to_le32(1u << IXGBE_ADVTXD_IDX_SHIFT);
3935 
3936         /* Check Context must be set if Tx switch is enabled, which it
3937          * always is for case where virtual functions are running
3938          */
3939         olinfo_status |= cpu_to_le32(IXGBE_ADVTXD_CC);
3940 
3941         tx_desc->read.olinfo_status = olinfo_status;
3942 }
3943 
3944 static void ixgbevf_tx_map(struct ixgbevf_ring *tx_ring,
3945                            struct ixgbevf_tx_buffer *first,
3946                            const u8 hdr_len)
3947 {
3948         struct sk_buff *skb = first->skb;
3949         struct ixgbevf_tx_buffer *tx_buffer;
3950         union ixgbe_adv_tx_desc *tx_desc;
3951         skb_frag_t *frag;
3952         dma_addr_t dma;
3953         unsigned int data_len, size;
3954         u32 tx_flags = first->tx_flags;
3955         __le32 cmd_type = ixgbevf_tx_cmd_type(tx_flags);
3956         u16 i = tx_ring->next_to_use;
3957 
3958         tx_desc = IXGBEVF_TX_DESC(tx_ring, i);
3959 
3960         ixgbevf_tx_olinfo_status(tx_desc, tx_flags, skb->len - hdr_len);
3961 
3962         size = skb_headlen(skb);
3963         data_len = skb->data_len;
3964 
3965         dma = dma_map_single(tx_ring->dev, skb->data, size, DMA_TO_DEVICE);
3966 
3967         tx_buffer = first;
3968 
3969         for (frag = &skb_shinfo(skb)->frags[0];; frag++) {
3970                 if (dma_mapping_error(tx_ring->dev, dma))
3971                         goto dma_error;
3972 
3973                 /* record length, and DMA address */
3974                 dma_unmap_len_set(tx_buffer, len, size);
3975                 dma_unmap_addr_set(tx_buffer, dma, dma);
3976 
3977                 tx_desc->read.buffer_addr = cpu_to_le64(dma);
3978 
3979                 while (unlikely(size > IXGBE_MAX_DATA_PER_TXD)) {
3980                         tx_desc->read.cmd_type_len =
3981                                 cmd_type | cpu_to_le32(IXGBE_MAX_DATA_PER_TXD);
3982 
3983                         i++;
3984                         tx_desc++;
3985                         if (i == tx_ring->count) {
3986                                 tx_desc = IXGBEVF_TX_DESC(tx_ring, 0);
3987                                 i = 0;
3988                         }
3989                         tx_desc->read.olinfo_status = 0;
3990 
3991                         dma += IXGBE_MAX_DATA_PER_TXD;
3992                         size -= IXGBE_MAX_DATA_PER_TXD;
3993 
3994                         tx_desc->read.buffer_addr = cpu_to_le64(dma);
3995                 }
3996 
3997                 if (likely(!data_len))
3998                         break;
3999 
4000                 tx_desc->read.cmd_type_len = cmd_type | cpu_to_le32(size);
4001 
4002                 i++;
4003                 tx_desc++;
4004                 if (i == tx_ring->count) {
4005                         tx_desc = IXGBEVF_TX_DESC(tx_ring, 0);
4006                         i = 0;
4007                 }
4008                 tx_desc->read.olinfo_status = 0;
4009 
4010                 size = skb_frag_size(frag);
4011                 data_len -= size;
4012 
4013                 dma = skb_frag_dma_map(tx_ring->dev, frag, 0, size,
4014                                        DMA_TO_DEVICE);
4015 
4016                 tx_buffer = &tx_ring->tx_buffer_info[i];
4017         }
4018 
4019         /* write last descriptor with RS and EOP bits */
4020         cmd_type |= cpu_to_le32(size) | cpu_to_le32(IXGBE_TXD_CMD);
4021         tx_desc->read.cmd_type_len = cmd_type;
4022 
4023         /* set the timestamp */
4024         first->time_stamp = jiffies;
4025 
4026         skb_tx_timestamp(skb);
4027 
4028         /* Force memory writes to complete before letting h/w know there
4029          * are new descriptors to fetch.  (Only applicable for weak-ordered
4030          * memory model archs, such as IA-64).
4031          *
4032          * We also need this memory barrier (wmb) to make certain all of the
4033          * status bits have been updated before next_to_watch is written.
4034          */
4035         wmb();
4036 
4037         /* set next_to_watch value indicating a packet is present */
4038         first->next_to_watch = tx_desc;
4039 
4040         i++;
4041         if (i == tx_ring->count)
4042                 i = 0;
4043 
4044         tx_ring->next_to_use = i;
4045 
4046         /* notify HW of packet */
4047         ixgbevf_write_tail(tx_ring, i);
4048 
4049         return;
4050 dma_error:
4051         dev_err(tx_ring->dev, "TX DMA map failed\n");
4052         tx_buffer = &tx_ring->tx_buffer_info[i];
4053 
4054         /* clear dma mappings for failed tx_buffer_info map */
4055         while (tx_buffer != first) {
4056                 if (dma_unmap_len(tx_buffer, len))
4057                         dma_unmap_page(tx_ring->dev,
4058                                        dma_unmap_addr(tx_buffer, dma),
4059                                        dma_unmap_len(tx_buffer, len),
4060                                        DMA_TO_DEVICE);
4061                 dma_unmap_len_set(tx_buffer, len, 0);
4062 
4063                 if (i-- == 0)
4064                         i += tx_ring->count;
4065                 tx_buffer = &tx_ring->tx_buffer_info[i];
4066         }
4067 
4068         if (dma_unmap_len(tx_buffer, len))
4069                 dma_unmap_single(tx_ring->dev,
4070                                  dma_unmap_addr(tx_buffer, dma),
4071                                  dma_unmap_len(tx_buffer, len),
4072                                  DMA_TO_DEVICE);
4073         dma_unmap_len_set(tx_buffer, len, 0);
4074 
4075         dev_kfree_skb_any(tx_buffer->skb);
4076         tx_buffer->skb = NULL;
4077 
4078         tx_ring->next_to_use = i;
4079 }
4080 
4081 static int __ixgbevf_maybe_stop_tx(struct ixgbevf_ring *tx_ring, int size)
4082 {
4083         netif_stop_subqueue(tx_ring->netdev, tx_ring->queue_index);
4084         /* Herbert's original patch had:
4085          *  smp_mb__after_netif_stop_queue();
4086          * but since that doesn't exist yet, just open code it.
4087          */
4088         smp_mb();
4089 
4090         /* We need to check again in a case another CPU has just
4091          * made room available.
4092          */
4093         if (likely(ixgbevf_desc_unused(tx_ring) < size))
4094                 return -EBUSY;
4095 
4096         /* A reprieve! - use start_queue because it doesn't call schedule */
4097         netif_start_subqueue(tx_ring->netdev, tx_ring->queue_index);
4098         ++tx_ring->tx_stats.restart_queue;
4099 
4100         return 0;
4101 }
4102 
4103 static int ixgbevf_maybe_stop_tx(struct ixgbevf_ring *tx_ring, int size)
4104 {
4105         if (likely(ixgbevf_desc_unused(tx_ring) >= size))
4106                 return 0;
4107         return __ixgbevf_maybe_stop_tx(tx_ring, size);
4108 }
4109 
4110 static int ixgbevf_xmit_frame_ring(struct sk_buff *skb,
4111                                    struct ixgbevf_ring *tx_ring)
4112 {
4113         struct ixgbevf_tx_buffer *first;
4114         int tso;
4115         u32 tx_flags = 0;
4116         u16 count = TXD_USE_COUNT(skb_headlen(skb));
4117         struct ixgbevf_ipsec_tx_data ipsec_tx = { 0 };
4118 #if PAGE_SIZE > IXGBE_MAX_DATA_PER_TXD
4119         unsigned short f;
4120 #endif
4121         u8 hdr_len = 0;
4122         u8 *dst_mac = skb_header_pointer(skb, 0, 0, NULL);
4123 
4124         if (!dst_mac || is_link_local_ether_addr(dst_mac)) {
4125                 dev_kfree_skb_any(skb);
4126                 return NETDEV_TX_OK;
4127         }
4128 
4129         /* need: 1 descriptor per page * PAGE_SIZE/IXGBE_MAX_DATA_PER_TXD,
4130          *       + 1 desc for skb_headlen/IXGBE_MAX_DATA_PER_TXD,
4131          *       + 2 desc gap to keep tail from touching head,
4132          *       + 1 desc for context descriptor,
4133          * otherwise try next time
4134          */
4135 #if PAGE_SIZE > IXGBE_MAX_DATA_PER_TXD
4136         for (f = 0; f < skb_shinfo(skb)->nr_frags; f++) {
4137                 skb_frag_t *frag = &skb_shinfo(skb)->frags[f];
4138 
4139                 count += TXD_USE_COUNT(skb_frag_size(frag));
4140         }
4141 #else
4142         count += skb_shinfo(skb)->nr_frags;
4143 #endif
4144         if (ixgbevf_maybe_stop_tx(tx_ring, count + 3)) {
4145                 tx_ring->tx_stats.tx_busy++;
4146                 return NETDEV_TX_BUSY;
4147         }
4148 
4149         /* record the location of the first descriptor for this packet */
4150         first = &tx_ring->tx_buffer_info[tx_ring->next_to_use];
4151         first->skb = skb;
4152         first->bytecount = skb->len;
4153         first->gso_segs = 1;
4154 
4155         if (skb_vlan_tag_present(skb)) {
4156                 tx_flags |= skb_vlan_tag_get(skb);
4157                 tx_flags <<= IXGBE_TX_FLAGS_VLAN_SHIFT;
4158                 tx_flags |= IXGBE_TX_FLAGS_VLAN;
4159         }
4160 
4161         /* record initial flags and protocol */
4162         first->tx_flags = tx_flags;
4163         first->protocol = vlan_get_protocol(skb);
4164 
4165 #ifdef CONFIG_IXGBEVF_IPSEC
4166         if (xfrm_offload(skb) && !ixgbevf_ipsec_tx(tx_ring, first, &ipsec_tx))
4167                 goto out_drop;
4168 #endif
4169         tso = ixgbevf_tso(tx_ring, first, &hdr_len, &ipsec_tx);
4170         if (tso < 0)
4171                 goto out_drop;
4172         else if (!tso)
4173                 ixgbevf_tx_csum(tx_ring, first, &ipsec_tx);
4174 
4175         ixgbevf_tx_map(tx_ring, first, hdr_len);
4176 
4177         ixgbevf_maybe_stop_tx(tx_ring, DESC_NEEDED);
4178 
4179         return NETDEV_TX_OK;
4180 
4181 out_drop:
4182         dev_kfree_skb_any(first->skb);
4183         first->skb = NULL;
4184 
4185         return NETDEV_TX_OK;
4186 }
4187 
4188 static netdev_tx_t ixgbevf_xmit_frame(struct sk_buff *skb, struct net_device *netdev)
4189 {
4190         struct ixgbevf_adapter *adapter = netdev_priv(netdev);
4191         struct ixgbevf_ring *tx_ring;
4192 
4193         if (skb->len <= 0) {
4194                 dev_kfree_skb_any(skb);
4195                 return NETDEV_TX_OK;
4196         }
4197 
4198         /* The minimum packet size for olinfo paylen is 17 so pad the skb
4199          * in order to meet this minimum size requirement.
4200          */
4201         if (skb->len < 17) {
4202                 if (skb_padto(skb, 17))
4203                         return NETDEV_TX_OK;
4204                 skb->len = 17;
4205         }
4206 
4207         tx_ring = adapter->tx_ring[skb->queue_mapping];
4208         return ixgbevf_xmit_frame_ring(skb, tx_ring);
4209 }
4210 
4211 /**
4212  * ixgbevf_set_mac - Change the Ethernet Address of the NIC
4213  * @netdev: network interface device structure
4214  * @p: pointer to an address structure
4215  *
4216  * Returns 0 on success, negative on failure
4217  **/
4218 static int ixgbevf_set_mac(struct net_device *netdev, void *p)
4219 {
4220         struct ixgbevf_adapter *adapter = netdev_priv(netdev);
4221         struct ixgbe_hw *hw = &adapter->hw;
4222         struct sockaddr *addr = p;
4223         int err;
4224 
4225         if (!is_valid_ether_addr(addr->sa_data))
4226                 return -EADDRNOTAVAIL;
4227 
4228         spin_lock_bh(&adapter->mbx_lock);
4229 
4230         err = hw->mac.ops.set_rar(hw, 0, addr->sa_data, 0);
4231 
4232         spin_unlock_bh(&adapter->mbx_lock);
4233 
4234         if (err)
4235                 return -EPERM;
4236 
4237         ether_addr_copy(hw->mac.addr, addr->sa_data);
4238         ether_addr_copy(hw->mac.perm_addr, addr->sa_data);
4239         ether_addr_copy(netdev->dev_addr, addr->sa_data);
4240 
4241         return 0;
4242 }
4243 
4244 /**
4245  * ixgbevf_change_mtu - Change the Maximum Transfer Unit
4246  * @netdev: network interface device structure
4247  * @new_mtu: new value for maximum frame size
4248  *
4249  * Returns 0 on success, negative on failure
4250  **/
4251 static int ixgbevf_change_mtu(struct net_device *netdev, int new_mtu)
4252 {
4253         struct ixgbevf_adapter *adapter = netdev_priv(netdev);
4254         struct ixgbe_hw *hw = &adapter->hw;
4255         int max_frame = new_mtu + ETH_HLEN + ETH_FCS_LEN;
4256         int ret;
4257 
4258         /* prevent MTU being changed to a size unsupported by XDP */
4259         if (adapter->xdp_prog) {
4260                 dev_warn(&adapter->pdev->dev, "MTU cannot be changed while XDP program is loaded\n");
4261                 return -EPERM;
4262         }
4263 
4264         spin_lock_bh(&adapter->mbx_lock);
4265         /* notify the PF of our intent to use this size of frame */
4266         ret = hw->mac.ops.set_rlpml(hw, max_frame);
4267         spin_unlock_bh(&adapter->mbx_lock);
4268         if (ret)
4269                 return -EINVAL;
4270 
4271         hw_dbg(hw, "changing MTU from %d to %d\n",
4272                netdev->mtu, new_mtu);
4273 
4274         /* must set new MTU before calling down or up */
4275         netdev->mtu = new_mtu;
4276 
4277         if (netif_running(netdev))
4278                 ixgbevf_reinit_locked(adapter);
4279 
4280         return 0;
4281 }
4282 
4283 static int ixgbevf_suspend(struct pci_dev *pdev, pm_message_t state)
4284 {
4285         struct net_device *netdev = pci_get_drvdata(pdev);
4286         struct ixgbevf_adapter *adapter = netdev_priv(netdev);
4287 #ifdef CONFIG_PM
4288         int retval = 0;
4289 #endif
4290 
4291         rtnl_lock();
4292         netif_device_detach(netdev);
4293 
4294         if (netif_running(netdev))
4295                 ixgbevf_close_suspend(adapter);
4296 
4297         ixgbevf_clear_interrupt_scheme(adapter);
4298         rtnl_unlock();
4299 
4300 #ifdef CONFIG_PM
4301         retval = pci_save_state(pdev);
4302         if (retval)
4303                 return retval;
4304 
4305 #endif
4306         if (!test_and_set_bit(__IXGBEVF_DISABLED, &adapter->state))
4307                 pci_disable_device(pdev);
4308 
4309         return 0;
4310 }
4311 
4312 #ifdef CONFIG_PM
4313 static int ixgbevf_resume(struct pci_dev *pdev)
4314 {
4315         struct net_device *netdev = pci_get_drvdata(pdev);
4316         struct ixgbevf_adapter *adapter = netdev_priv(netdev);
4317         u32 err;
4318 
4319         pci_restore_state(pdev);
4320         /* pci_restore_state clears dev->state_saved so call
4321          * pci_save_state to restore it.
4322          */
4323         pci_save_state(pdev);
4324 
4325         err = pci_enable_device_mem(pdev);
4326         if (err) {
4327                 dev_err(&pdev->dev, "Cannot enable PCI device from suspend\n");
4328                 return err;
4329         }
4330 
4331         adapter->hw.hw_addr = adapter->io_addr;
4332         smp_mb__before_atomic();
4333         clear_bit(__IXGBEVF_DISABLED, &adapter->state);
4334         pci_set_master(pdev);
4335 
4336         ixgbevf_reset(adapter);
4337 
4338         rtnl_lock();
4339         err = ixgbevf_init_interrupt_scheme(adapter);
4340         if (!err && netif_running(netdev))
4341                 err = ixgbevf_open(netdev);
4342         rtnl_unlock();
4343         if (err)
4344                 return err;
4345 
4346         netif_device_attach(netdev);
4347 
4348         return err;
4349 }
4350 
4351 #endif /* CONFIG_PM */
4352 static void ixgbevf_shutdown(struct pci_dev *pdev)
4353 {
4354         ixgbevf_suspend(pdev, PMSG_SUSPEND);
4355 }
4356 
4357 static void ixgbevf_get_tx_ring_stats(struct rtnl_link_stats64 *stats,
4358                                       const struct ixgbevf_ring *ring)
4359 {
4360         u64 bytes, packets;
4361         unsigned int start;
4362 
4363         if (ring) {
4364                 do {
4365                         start = u64_stats_fetch_begin_irq(&ring->syncp);
4366                         bytes = ring->stats.bytes;
4367                         packets = ring->stats.packets;
4368                 } while (u64_stats_fetch_retry_irq(&ring->syncp, start));
4369                 stats->tx_bytes += bytes;
4370                 stats->tx_packets += packets;
4371         }
4372 }
4373 
4374 static void ixgbevf_get_stats(struct net_device *netdev,
4375                               struct rtnl_link_stats64 *stats)
4376 {
4377         struct ixgbevf_adapter *adapter = netdev_priv(netdev);
4378         unsigned int start;
4379         u64 bytes, packets;
4380         const struct ixgbevf_ring *ring;
4381         int i;
4382 
4383         ixgbevf_update_stats(adapter);
4384 
4385         stats->multicast = adapter->stats.vfmprc - adapter->stats.base_vfmprc;
4386 
4387         rcu_read_lock();
4388         for (i = 0; i < adapter->num_rx_queues; i++) {
4389                 ring = adapter->rx_ring[i];
4390                 do {
4391                         start = u64_stats_fetch_begin_irq(&ring->syncp);
4392                         bytes = ring->stats.bytes;
4393                         packets = ring->stats.packets;
4394                 } while (u64_stats_fetch_retry_irq(&ring->syncp, start));
4395                 stats->rx_bytes += bytes;
4396                 stats->rx_packets += packets;
4397         }
4398 
4399         for (i = 0; i < adapter->num_tx_queues; i++) {
4400                 ring = adapter->tx_ring[i];
4401                 ixgbevf_get_tx_ring_stats(stats, ring);
4402         }
4403 
4404         for (i = 0; i < adapter->num_xdp_queues; i++) {
4405                 ring = adapter->xdp_ring[i];
4406                 ixgbevf_get_tx_ring_stats(stats, ring);
4407         }
4408         rcu_read_unlock();
4409 }
4410 
4411 #define IXGBEVF_MAX_MAC_HDR_LEN         127
4412 #define IXGBEVF_MAX_NETWORK_HDR_LEN     511
4413 
4414 static netdev_features_t
4415 ixgbevf_features_check(struct sk_buff *skb, struct net_device *dev,
4416                        netdev_features_t features)
4417 {
4418         unsigned int network_hdr_len, mac_hdr_len;
4419 
4420         /* Make certain the headers can be described by a context descriptor */
4421         mac_hdr_len = skb_network_header(skb) - skb->data;
4422         if (unlikely(mac_hdr_len > IXGBEVF_MAX_MAC_HDR_LEN))
4423                 return features & ~(NETIF_F_HW_CSUM |
4424                                     NETIF_F_SCTP_CRC |
4425                                     NETIF_F_HW_VLAN_CTAG_TX |
4426                                     NETIF_F_TSO |
4427                                     NETIF_F_TSO6);
4428 
4429         network_hdr_len = skb_checksum_start(skb) - skb_network_header(skb);
4430         if (unlikely(network_hdr_len >  IXGBEVF_MAX_NETWORK_HDR_LEN))
4431                 return features & ~(NETIF_F_HW_CSUM |
4432                                     NETIF_F_SCTP_CRC |
4433                                     NETIF_F_TSO |
4434                                     NETIF_F_TSO6);
4435 
4436         /* We can only support IPV4 TSO in tunnels if we can mangle the
4437          * inner IP ID field, so strip TSO if MANGLEID is not supported.
4438          */
4439         if (skb->encapsulation && !(features & NETIF_F_TSO_MANGLEID))
4440                 features &= ~NETIF_F_TSO;
4441 
4442         return features;
4443 }
4444 
4445 static int ixgbevf_xdp_setup(struct net_device *dev, struct bpf_prog *prog)
4446 {
4447         int i, frame_size = dev->mtu + ETH_HLEN + ETH_FCS_LEN + VLAN_HLEN;
4448         struct ixgbevf_adapter *adapter = netdev_priv(dev);
4449         struct bpf_prog *old_prog;
4450 
4451         /* verify ixgbevf ring attributes are sufficient for XDP */
4452         for (i = 0; i < adapter->num_rx_queues; i++) {
4453                 struct ixgbevf_ring *ring = adapter->rx_ring[i];
4454 
4455                 if (frame_size > ixgbevf_rx_bufsz(ring))
4456                         return -EINVAL;
4457         }
4458 
4459         old_prog = xchg(&adapter->xdp_prog, prog);
4460 
4461         /* If transitioning XDP modes reconfigure rings */
4462         if (!!prog != !!old_prog) {
4463                 /* Hardware has to reinitialize queues and interrupts to
4464                  * match packet buffer alignment. Unfortunately, the
4465                  * hardware is not flexible enough to do this dynamically.
4466                  */
4467                 if (netif_running(dev))
4468                         ixgbevf_close(dev);
4469 
4470                 ixgbevf_clear_interrupt_scheme(adapter);
4471                 ixgbevf_init_interrupt_scheme(adapter);
4472 
4473                 if (netif_running(dev))
4474                         ixgbevf_open(dev);
4475         } else {
4476                 for (i = 0; i < adapter->num_rx_queues; i++)
4477                         xchg(&adapter->rx_ring[i]->xdp_prog, adapter->xdp_prog);
4478         }
4479 
4480         if (old_prog)
4481                 bpf_prog_put(old_prog);
4482 
4483         return 0;
4484 }
4485 
4486 static int ixgbevf_xdp(struct net_device *dev, struct netdev_bpf *xdp)
4487 {
4488         struct ixgbevf_adapter *adapter = netdev_priv(dev);
4489 
4490         switch (xdp->command) {
4491         case XDP_SETUP_PROG:
4492                 return ixgbevf_xdp_setup(dev, xdp->prog);
4493         case XDP_QUERY_PROG:
4494                 xdp->prog_id = adapter->xdp_prog ?
4495                                adapter->xdp_prog->aux->id : 0;
4496                 return 0;
4497         default:
4498                 return -EINVAL;
4499         }
4500 }
4501 
4502 static const struct net_device_ops ixgbevf_netdev_ops = {
4503         .ndo_open               = ixgbevf_open,
4504         .ndo_stop               = ixgbevf_close,
4505         .ndo_start_xmit         = ixgbevf_xmit_frame,
4506         .ndo_set_rx_mode        = ixgbevf_set_rx_mode,
4507         .ndo_get_stats64        = ixgbevf_get_stats,
4508         .ndo_validate_addr      = eth_validate_addr,
4509         .ndo_set_mac_address    = ixgbevf_set_mac,
4510         .ndo_change_mtu         = ixgbevf_change_mtu,
4511         .ndo_tx_timeout         = ixgbevf_tx_timeout,
4512         .ndo_vlan_rx_add_vid    = ixgbevf_vlan_rx_add_vid,
4513         .ndo_vlan_rx_kill_vid   = ixgbevf_vlan_rx_kill_vid,
4514         .ndo_features_check     = ixgbevf_features_check,
4515         .ndo_bpf                = ixgbevf_xdp,
4516 };
4517 
4518 static void ixgbevf_assign_netdev_ops(struct net_device *dev)
4519 {
4520         dev->netdev_ops = &ixgbevf_netdev_ops;
4521         ixgbevf_set_ethtool_ops(dev);
4522         dev->watchdog_timeo = 5 * HZ;
4523 }
4524 
4525 /**
4526  * ixgbevf_probe - Device Initialization Routine
4527  * @pdev: PCI device information struct
4528  * @ent: entry in ixgbevf_pci_tbl
4529  *
4530  * Returns 0 on success, negative on failure
4531  *
4532  * ixgbevf_probe initializes an adapter identified by a pci_dev structure.
4533  * The OS initialization, configuring of the adapter private structure,
4534  * and a hardware reset occur.
4535  **/
4536 static int ixgbevf_probe(struct pci_dev *pdev, const struct pci_device_id *ent)
4537 {
4538         struct net_device *netdev;
4539         struct ixgbevf_adapter *adapter = NULL;
4540         struct ixgbe_hw *hw = NULL;
4541         const struct ixgbevf_info *ii = ixgbevf_info_tbl[ent->driver_data];
4542         int err, pci_using_dac;
4543         bool disable_dev = false;
4544 
4545         err = pci_enable_device(pdev);
4546         if (err)
4547                 return err;
4548 
4549         if (!dma_set_mask_and_coherent(&pdev->dev, DMA_BIT_MASK(64))) {
4550                 pci_using_dac = 1;
4551         } else {
4552                 err = dma_set_mask_and_coherent(&pdev->dev, DMA_BIT_MASK(32));
4553                 if (err) {
4554                         dev_err(&pdev->dev, "No usable DMA configuration, aborting\n");
4555                         goto err_dma;
4556                 }
4557                 pci_using_dac = 0;
4558         }
4559 
4560         err = pci_request_regions(pdev, ixgbevf_driver_name);
4561         if (err) {
4562                 dev_err(&pdev->dev, "pci_request_regions failed 0x%x\n", err);
4563                 goto err_pci_reg;
4564         }
4565 
4566         pci_set_master(pdev);
4567 
4568         netdev = alloc_etherdev_mq(sizeof(struct ixgbevf_adapter),
4569                                    MAX_TX_QUEUES);
4570         if (!netdev) {
4571                 err = -ENOMEM;
4572                 goto err_alloc_etherdev;
4573         }
4574 
4575         SET_NETDEV_DEV(netdev, &pdev->dev);
4576 
4577         adapter = netdev_priv(netdev);
4578 
4579         adapter->netdev = netdev;
4580         adapter->pdev = pdev;
4581         hw = &adapter->hw;
4582         hw->back = adapter;
4583         adapter->msg_enable = netif_msg_init(debug, DEFAULT_MSG_ENABLE);
4584 
4585         /* call save state here in standalone driver because it relies on
4586          * adapter struct to exist, and needs to call netdev_priv
4587          */
4588         pci_save_state(pdev);
4589 
4590         hw->hw_addr = ioremap(pci_resource_start(pdev, 0),
4591                               pci_resource_len(pdev, 0));
4592         adapter->io_addr = hw->hw_addr;
4593         if (!hw->hw_addr) {
4594                 err = -EIO;
4595                 goto err_ioremap;
4596         }
4597 
4598         ixgbevf_assign_netdev_ops(netdev);
4599 
4600         /* Setup HW API */
4601         memcpy(&hw->mac.ops, ii->mac_ops, sizeof(hw->mac.ops));
4602         hw->mac.type  = ii->mac;
4603 
4604         memcpy(&hw->mbx.ops, &ixgbevf_mbx_ops,
4605                sizeof(struct ixgbe_mbx_operations));
4606 
4607         /* setup the private structure */
4608         err = ixgbevf_sw_init(adapter);
4609         if (err)
4610                 goto err_sw_init;
4611 
4612         /* The HW MAC address was set and/or determined in sw_init */
4613         if (!is_valid_ether_addr(netdev->dev_addr)) {
4614                 pr_err("invalid MAC address\n");
4615                 err = -EIO;
4616                 goto err_sw_init;
4617         }
4618 
4619         netdev->hw_features = NETIF_F_SG |
4620                               NETIF_F_TSO |
4621                               NETIF_F_TSO6 |
4622                               NETIF_F_RXCSUM |
4623                               NETIF_F_HW_CSUM |
4624                               NETIF_F_SCTP_CRC;
4625 
4626 #define IXGBEVF_GSO_PARTIAL_FEATURES (NETIF_F_GSO_GRE | \
4627                                       NETIF_F_GSO_GRE_CSUM | \
4628                                       NETIF_F_GSO_IPXIP4 | \
4629                                       NETIF_F_GSO_IPXIP6 | \
4630                                       NETIF_F_GSO_UDP_TUNNEL | \
4631                                       NETIF_F_GSO_UDP_TUNNEL_CSUM)
4632 
4633         netdev->gso_partial_features = IXGBEVF_GSO_PARTIAL_FEATURES;
4634         netdev->hw_features |= NETIF_F_GSO_PARTIAL |
4635                                IXGBEVF_GSO_PARTIAL_FEATURES;
4636 
4637         netdev->features = netdev->hw_features;
4638 
4639         if (pci_using_dac)
4640                 netdev->features |= NETIF_F_HIGHDMA;
4641 
4642         netdev->vlan_features |= netdev->features | NETIF_F_TSO_MANGLEID;
4643         netdev->mpls_features |= NETIF_F_SG |
4644                                  NETIF_F_TSO |
4645                                  NETIF_F_TSO6 |
4646                                  NETIF_F_HW_CSUM;
4647         netdev->mpls_features |= IXGBEVF_GSO_PARTIAL_FEATURES;
4648         netdev->hw_enc_features |= netdev->vlan_features;
4649 
4650         /* set this bit last since it cannot be part of vlan_features */
4651         netdev->features |= NETIF_F_HW_VLAN_CTAG_FILTER |
4652                             NETIF_F_HW_VLAN_CTAG_RX |
4653                             NETIF_F_HW_VLAN_CTAG_TX;
4654 
4655         netdev->priv_flags |= IFF_UNICAST_FLT;
4656 
4657         /* MTU range: 68 - 1504 or 9710 */
4658         netdev->min_mtu = ETH_MIN_MTU;
4659         switch (adapter->hw.api_version) {
4660         case ixgbe_mbox_api_11:
4661         case ixgbe_mbox_api_12:
4662         case ixgbe_mbox_api_13:
4663         case ixgbe_mbox_api_14:
4664                 netdev->max_mtu = IXGBE_MAX_JUMBO_FRAME_SIZE -
4665                                   (ETH_HLEN + ETH_FCS_LEN);
4666                 break;
4667         default:
4668                 if (adapter->hw.mac.type != ixgbe_mac_82599_vf)
4669                         netdev->max_mtu = IXGBE_MAX_JUMBO_FRAME_SIZE -
4670                                           (ETH_HLEN + ETH_FCS_LEN);
4671                 else
4672                         netdev->max_mtu = ETH_DATA_LEN + ETH_FCS_LEN;
4673                 break;
4674         }
4675 
4676         if (IXGBE_REMOVED(hw->hw_addr)) {
4677                 err = -EIO;
4678                 goto err_sw_init;
4679         }
4680 
4681         timer_setup(&adapter->service_timer, ixgbevf_service_timer, 0);
4682 
4683         INIT_WORK(&adapter->service_task, ixgbevf_service_task);
4684         set_bit(__IXGBEVF_SERVICE_INITED, &adapter->state);
4685         clear_bit(__IXGBEVF_SERVICE_SCHED, &adapter->state);
4686 
4687         err = ixgbevf_init_interrupt_scheme(adapter);
4688         if (err)
4689                 goto err_sw_init;
4690 
4691         strcpy(netdev->name, "eth%d");
4692 
4693         err = register_netdev(netdev);
4694         if (err)
4695                 goto err_register;
4696 
4697         pci_set_drvdata(pdev, netdev);
4698         netif_carrier_off(netdev);
4699         ixgbevf_init_ipsec_offload(adapter);
4700 
4701         ixgbevf_init_last_counter_stats(adapter);
4702 
4703         /* print the VF info */
4704         dev_info(&pdev->dev, "%pM\n", netdev->dev_addr);
4705         dev_info(&pdev->dev, "MAC: %d\n", hw->mac.type);
4706 
4707         switch (hw->mac.type) {
4708         case ixgbe_mac_X550_vf:
4709                 dev_info(&pdev->dev, "Intel(R) X550 Virtual Function\n");
4710                 break;
4711         case ixgbe_mac_X540_vf:
4712                 dev_info(&pdev->dev, "Intel(R) X540 Virtual Function\n");
4713                 break;
4714         case ixgbe_mac_82599_vf:
4715         default:
4716                 dev_info(&pdev->dev, "Intel(R) 82599 Virtual Function\n");
4717                 break;
4718         }
4719 
4720         return 0;
4721 
4722 err_register:
4723         ixgbevf_clear_interrupt_scheme(adapter);
4724 err_sw_init:
4725         ixgbevf_reset_interrupt_capability(adapter);
4726         iounmap(adapter->io_addr);
4727         kfree(adapter->rss_key);
4728 err_ioremap:
4729         disable_dev = !test_and_set_bit(__IXGBEVF_DISABLED, &adapter->state);
4730         free_netdev(netdev);
4731 err_alloc_etherdev:
4732         pci_release_regions(pdev);
4733 err_pci_reg:
4734 err_dma:
4735         if (!adapter || disable_dev)
4736                 pci_disable_device(pdev);
4737         return err;
4738 }
4739 
4740 /**
4741  * ixgbevf_remove - Device Removal Routine
4742  * @pdev: PCI device information struct
4743  *
4744  * ixgbevf_remove is called by the PCI subsystem to alert the driver
4745  * that it should release a PCI device.  The could be caused by a
4746  * Hot-Plug event, or because the driver is going to be removed from
4747  * memory.
4748  **/
4749 static void ixgbevf_remove(struct pci_dev *pdev)
4750 {
4751         struct net_device *netdev = pci_get_drvdata(pdev);
4752         struct ixgbevf_adapter *adapter;
4753         bool disable_dev;
4754 
4755         if (!netdev)
4756                 return;
4757 
4758         adapter = netdev_priv(netdev);
4759 
4760         set_bit(__IXGBEVF_REMOVING, &adapter->state);
4761         cancel_work_sync(&adapter->service_task);
4762 
4763         if (netdev->reg_state == NETREG_REGISTERED)
4764                 unregister_netdev(netdev);
4765 
4766         ixgbevf_stop_ipsec_offload(adapter);
4767         ixgbevf_clear_interrupt_scheme(adapter);
4768         ixgbevf_reset_interrupt_capability(adapter);
4769 
4770         iounmap(adapter->io_addr);
4771         pci_release_regions(pdev);
4772 
4773         hw_dbg(&adapter->hw, "Remove complete\n");
4774 
4775         kfree(adapter->rss_key);
4776         disable_dev = !test_and_set_bit(__IXGBEVF_DISABLED, &adapter->state);
4777         free_netdev(netdev);
4778 
4779         if (disable_dev)
4780                 pci_disable_device(pdev);
4781 }
4782 
4783 /**
4784  * ixgbevf_io_error_detected - called when PCI error is detected
4785  * @pdev: Pointer to PCI device
4786  * @state: The current pci connection state
4787  *
4788  * This function is called after a PCI bus error affecting
4789  * this device has been detected.
4790  **/
4791 static pci_ers_result_t ixgbevf_io_error_detected(struct pci_dev *pdev,
4792                                                   pci_channel_state_t state)
4793 {
4794         struct net_device *netdev = pci_get_drvdata(pdev);
4795         struct ixgbevf_adapter *adapter = netdev_priv(netdev);
4796 
4797         if (!test_bit(__IXGBEVF_SERVICE_INITED, &adapter->state))
4798                 return PCI_ERS_RESULT_DISCONNECT;
4799 
4800         rtnl_lock();
4801         netif_device_detach(netdev);
4802 
4803         if (netif_running(netdev))
4804                 ixgbevf_close_suspend(adapter);
4805 
4806         if (state == pci_channel_io_perm_failure) {
4807                 rtnl_unlock();
4808                 return PCI_ERS_RESULT_DISCONNECT;
4809         }
4810 
4811         if (!test_and_set_bit(__IXGBEVF_DISABLED, &adapter->state))
4812                 pci_disable_device(pdev);
4813         rtnl_unlock();
4814 
4815         /* Request a slot slot reset. */
4816         return PCI_ERS_RESULT_NEED_RESET;
4817 }
4818 
4819 /**
4820  * ixgbevf_io_slot_reset - called after the pci bus has been reset.
4821  * @pdev: Pointer to PCI device
4822  *
4823  * Restart the card from scratch, as if from a cold-boot. Implementation
4824  * resembles the first-half of the ixgbevf_resume routine.
4825  **/
4826 static pci_ers_result_t ixgbevf_io_slot_reset(struct pci_dev *pdev)
4827 {
4828         struct net_device *netdev = pci_get_drvdata(pdev);
4829         struct ixgbevf_adapter *adapter = netdev_priv(netdev);
4830 
4831         if (pci_enable_device_mem(pdev)) {
4832                 dev_err(&pdev->dev,
4833                         "Cannot re-enable PCI device after reset.\n");
4834                 return PCI_ERS_RESULT_DISCONNECT;
4835         }
4836 
4837         adapter->hw.hw_addr = adapter->io_addr;
4838         smp_mb__before_atomic();
4839         clear_bit(__IXGBEVF_DISABLED, &adapter->state);
4840         pci_set_master(pdev);
4841 
4842         ixgbevf_reset(adapter);
4843 
4844         return PCI_ERS_RESULT_RECOVERED;
4845 }
4846 
4847 /**
4848  * ixgbevf_io_resume - called when traffic can start flowing again.
4849  * @pdev: Pointer to PCI device
4850  *
4851  * This callback is called when the error recovery driver tells us that
4852  * its OK to resume normal operation. Implementation resembles the
4853  * second-half of the ixgbevf_resume routine.
4854  **/
4855 static void ixgbevf_io_resume(struct pci_dev *pdev)
4856 {
4857         struct net_device *netdev = pci_get_drvdata(pdev);
4858 
4859         rtnl_lock();
4860         if (netif_running(netdev))
4861                 ixgbevf_open(netdev);
4862 
4863         netif_device_attach(netdev);
4864         rtnl_unlock();
4865 }
4866 
4867 /* PCI Error Recovery (ERS) */
4868 static const struct pci_error_handlers ixgbevf_err_handler = {
4869         .error_detected = ixgbevf_io_error_detected,
4870         .slot_reset = ixgbevf_io_slot_reset,
4871         .resume = ixgbevf_io_resume,
4872 };
4873 
4874 static struct pci_driver ixgbevf_driver = {
4875         .name           = ixgbevf_driver_name,
4876         .id_table       = ixgbevf_pci_tbl,
4877         .probe          = ixgbevf_probe,
4878         .remove         = ixgbevf_remove,
4879 #ifdef CONFIG_PM
4880         /* Power Management Hooks */
4881         .suspend        = ixgbevf_suspend,
4882         .resume         = ixgbevf_resume,
4883 #endif
4884         .shutdown       = ixgbevf_shutdown,
4885         .err_handler    = &ixgbevf_err_handler
4886 };
4887 
4888 /**
4889  * ixgbevf_init_module - Driver Registration Routine
4890  *
4891  * ixgbevf_init_module is the first routine called when the driver is
4892  * loaded. All it does is register with the PCI subsystem.
4893  **/
4894 static int __init ixgbevf_init_module(void)
4895 {
4896         pr_info("%s - version %s\n", ixgbevf_driver_string,
4897                 ixgbevf_driver_version);
4898 
4899         pr_info("%s\n", ixgbevf_copyright);
4900         ixgbevf_wq = create_singlethread_workqueue(ixgbevf_driver_name);
4901         if (!ixgbevf_wq) {
4902                 pr_err("%s: Failed to create workqueue\n", ixgbevf_driver_name);
4903                 return -ENOMEM;
4904         }
4905 
4906         return pci_register_driver(&ixgbevf_driver);
4907 }
4908 
4909 module_init(ixgbevf_init_module);
4910 
4911 /**
4912  * ixgbevf_exit_module - Driver Exit Cleanup Routine
4913  *
4914  * ixgbevf_exit_module is called just before the driver is removed
4915  * from memory.
4916  **/
4917 static void __exit ixgbevf_exit_module(void)
4918 {
4919         pci_unregister_driver(&ixgbevf_driver);
4920         if (ixgbevf_wq) {
4921                 destroy_workqueue(ixgbevf_wq);
4922                 ixgbevf_wq = NULL;
4923         }
4924 }
4925 
4926 #ifdef DEBUG
4927 /**
4928  * ixgbevf_get_hw_dev_name - return device name string
4929  * used by hardware layer to print debugging information
4930  * @hw: pointer to private hardware struct
4931  **/
4932 char *ixgbevf_get_hw_dev_name(struct ixgbe_hw *hw)
4933 {
4934         struct ixgbevf_adapter *adapter = hw->back;
4935 
4936         return adapter->netdev->name;
4937 }
4938 
4939 #endif
4940 module_exit(ixgbevf_exit_module);
4941 
4942 /* ixgbevf_main.c */
/* [<][>][^][v][top][bottom][index][help] */
root/drivers/net/ethernet/intel/ixgbevf/ixgbevf_main.c

DEFINITIONS
