drivers/net/ethernet/intel/ixgbe/ixgbe_main.c

   1 // SPDX-License-Identifier: GPL-2.0
   2 /* Copyright(c) 1999 - 2018 Intel Corporation. */
   3
   4 #include <linux/types.h>
   5 #include <linux/module.h>
   6 #include <linux/pci.h>
   7 #include <linux/netdevice.h>
   8 #include <linux/vmalloc.h>
   9 #include <linux/string.h>
  10 #include <linux/in.h>
  11 #include <linux/interrupt.h>
  12 #include <linux/ip.h>
  13 #include <linux/tcp.h>
  14 #include <linux/sctp.h>
  15 #include <linux/pkt_sched.h>
  16 #include <linux/ipv6.h>
  17 #include <linux/slab.h>
  18 #include <net/checksum.h>
  19 #include <net/ip6_checksum.h>
  20 #include <linux/etherdevice.h>
  21 #include <linux/ethtool.h>
  22 #include <linux/if.h>
  23 #include <linux/if_vlan.h>
  24 #include <linux/if_macvlan.h>
  25 #include <linux/if_bridge.h>
  26 #include <linux/prefetch.h>
  27 #include <linux/bpf.h>
  28 #include <linux/bpf_trace.h>
  29 #include <linux/atomic.h>
  30 #include <linux/numa.h>
  31 #include <generated/utsrelease.h>
  32 #include <scsi/fc/fc_fcoe.h>
  33 #include <net/udp_tunnel.h>
  34 #include <net/pkt_cls.h>
  35 #include <net/tc_act/tc_gact.h>
  36 #include <net/tc_act/tc_mirred.h>
  37 #include <net/vxlan.h>
  38 #include <net/mpls.h>
  39 #include <net/xdp_sock_drv.h>
  40 #include <net/xfrm.h>
  41
  42 #include "ixgbe.h"
  43 #include "ixgbe_common.h"
  44 #include "ixgbe_dcb_82599.h"
  45 #include "ixgbe_phy.h"
  46 #include "ixgbe_sriov.h"
  47 #include "ixgbe_model.h"
  48 #include "ixgbe_txrx_common.h"
  49
  50 char ixgbe_driver_name[] = "ixgbe";
  51 static const char ixgbe_driver_string[] =
  52                               "Intel(R) 10 Gigabit PCI Express Network Driver";
  53 #ifdef IXGBE_FCOE
  54 char ixgbe_default_device_descr[] =
  55                               "Intel(R) 10 Gigabit Network Connection";
  56 #else
  57 static char ixgbe_default_device_descr[] =
  58                               "Intel(R) 10 Gigabit Network Connection";
  59 #endif
  60 static const char ixgbe_copyright[] =
  61                                 "Copyright (c) 1999-2016 Intel Corporation.";
  62
  63 static const char ixgbe_overheat_msg[] = "Network adapter has been stopped because it has over heated. Restart the computer. If the problem persists, power off the system and replace the adapter";
  64
  65 static const struct ixgbe_info *ixgbe_info_tbl[] = {
  66         [board_82598]           = &ixgbe_82598_info,
  67         [board_82599]           = &ixgbe_82599_info,
  68         [board_X540]            = &ixgbe_X540_info,
  69         [board_X550]            = &ixgbe_X550_info,
  70         [board_X550EM_x]        = &ixgbe_X550EM_x_info,
  71         [board_x550em_x_fw]     = &ixgbe_x550em_x_fw_info,
  72         [board_x550em_a]        = &ixgbe_x550em_a_info,
  73         [board_x550em_a_fw]     = &ixgbe_x550em_a_fw_info,
  74 };
  75
  76 /* ixgbe_pci_tbl - PCI Device ID Table
  77  *
  78  * Wildcard entries (PCI_ANY_ID) should come last
  79  * Last entry must be all 0s
  80  *
  81  * { Vendor ID, Device ID, SubVendor ID, SubDevice ID,
  82  *   Class, Class Mask, private data (not used) }
  83  */
  84 static const struct pci_device_id ixgbe_pci_tbl[] = {
  85         {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82598), board_82598 },
  86         {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82598AF_DUAL_PORT), board_82598 },
  87         {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82598AF_SINGLE_PORT), board_82598 },
  88         {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82598AT), board_82598 },
  89         {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82598AT2), board_82598 },
  90         {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82598EB_CX4), board_82598 },
  91         {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82598_CX4_DUAL_PORT), board_82598 },
  92         {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82598_DA_DUAL_PORT), board_82598 },
  93         {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82598_SR_DUAL_PORT_EM), board_82598 },
  94         {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82598EB_XF_LR), board_82598 },
  95         {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82598EB_SFP_LOM), board_82598 },
  96         {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82598_BX), board_82598 },
  97         {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82599_KX4), board_82599 },
  98         {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82599_XAUI_LOM), board_82599 },
  99         {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82599_KR), board_82599 },
 100         {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82599_SFP), board_82599 },
 101         {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82599_SFP_EM), board_82599 },
 102         {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82599_KX4_MEZZ), board_82599 },
 103         {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82599_CX4), board_82599 },
 104         {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82599_BACKPLANE_FCOE), board_82599 },
 105         {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82599_SFP_FCOE), board_82599 },
 106         {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82599_T3_LOM), board_82599 },
 107         {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82599_COMBO_BACKPLANE), board_82599 },
 108         {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_X540T), board_X540 },
 109         {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82599_SFP_SF2), board_82599 },
 110         {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82599_LS), board_82599 },
 111         {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82599_QSFP_SF_QP), board_82599 },
 112         {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82599EN_SFP), board_82599 },
 113         {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82599_SFP_SF_QP), board_82599 },
 114         {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_X540T1), board_X540 },
 115         {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_X550T), board_X550},
 116         {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_X550T1), board_X550},
 117         {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_X550EM_X_KX4), board_X550EM_x},
 118         {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_X550EM_X_XFI), board_X550EM_x},
 119         {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_X550EM_X_KR), board_X550EM_x},
 120         {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_X550EM_X_10G_T), board_X550EM_x},
 121         {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_X550EM_X_SFP), board_X550EM_x},
 122         {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_X550EM_X_1G_T), board_x550em_x_fw},
 123         {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_X550EM_A_KR), board_x550em_a },
 124         {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_X550EM_A_KR_L), board_x550em_a },
 125         {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_X550EM_A_SFP_N), board_x550em_a },
 126         {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_X550EM_A_SGMII), board_x550em_a },
 127         {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_X550EM_A_SGMII_L), board_x550em_a },
 128         {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_X550EM_A_10G_T), board_x550em_a},
 129         {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_X550EM_A_SFP), board_x550em_a },
 130         {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_X550EM_A_1G_T), board_x550em_a_fw },
 131         {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_X550EM_A_1G_T_L), board_x550em_a_fw },
 132         /* required last entry */
 133         {0, }
 134 };
 135 MODULE_DEVICE_TABLE(pci, ixgbe_pci_tbl);
 136
 137 #ifdef CONFIG_IXGBE_DCA
 138 static int ixgbe_notify_dca(struct notifier_block *, unsigned long event,
 139                             void *p);
 140 static struct notifier_block dca_notifier = {
 141         .notifier_call = ixgbe_notify_dca,
 142         .next          = NULL,
 143         .priority      = 0
 144 };
 145 #endif
 146
 147 #ifdef CONFIG_PCI_IOV
 148 static unsigned int max_vfs;
 149 module_param(max_vfs, uint, 0);
 150 MODULE_PARM_DESC(max_vfs,
 151                  "Maximum number of virtual functions to allocate per physical function - default is zero and maximum value is 63. (Deprecated)");
 152 #endif /* CONFIG_PCI_IOV */
 153
 154 static bool allow_unsupported_sfp;
 155 module_param(allow_unsupported_sfp, bool, 0);
 156 MODULE_PARM_DESC(allow_unsupported_sfp,
 157                  "Allow unsupported and untested SFP+ modules on 82599-based adapters");
 158
 159 #define DEFAULT_MSG_ENABLE (NETIF_MSG_DRV|NETIF_MSG_PROBE|NETIF_MSG_LINK)
 160 static int debug = -1;
 161 module_param(debug, int, 0);
 162 MODULE_PARM_DESC(debug, "Debug level (0=none,...,16=all)");
 163
 164 MODULE_AUTHOR("Intel Corporation, <linux.nics@intel.com>");
 165 MODULE_DESCRIPTION("Intel(R) 10 Gigabit PCI Express Network Driver");
 166 MODULE_LICENSE("GPL v2");
 167
 168 DEFINE_STATIC_KEY_FALSE(ixgbe_xdp_locking_key);
 169 EXPORT_SYMBOL(ixgbe_xdp_locking_key);
 170
 171 static struct workqueue_struct *ixgbe_wq;
 172
 173 static bool ixgbe_check_cfg_remove(struct ixgbe_hw *hw, struct pci_dev *pdev);
 174 static void ixgbe_watchdog_link_is_down(struct ixgbe_adapter *);
 175
 176 static const struct net_device_ops ixgbe_netdev_ops;
 177
 178 static bool netif_is_ixgbe(struct net_device *dev)
 179 {
 180         return dev && (dev->netdev_ops == &ixgbe_netdev_ops);
 181 }
 182
 183 static int ixgbe_read_pci_cfg_word_parent(struct ixgbe_adapter *adapter,
 184                                           u32 reg, u16 *value)
 185 {
 186         struct pci_dev *parent_dev;
 187         struct pci_bus *parent_bus;
 188
 189         parent_bus = adapter->pdev->bus->parent;
 190         if (!parent_bus)
 191                 return -1;
 192
 193         parent_dev = parent_bus->self;
 194         if (!parent_dev)
 195                 return -1;
 196
 197         if (!pci_is_pcie(parent_dev))
 198                 return -1;
 199
 200         pcie_capability_read_word(parent_dev, reg, value);
 201         if (*value == IXGBE_FAILED_READ_CFG_WORD &&
 202             ixgbe_check_cfg_remove(&adapter->hw, parent_dev))
 203                 return -1;
 204         return 0;
 205 }
 206
 207 static s32 ixgbe_get_parent_bus_info(struct ixgbe_adapter *adapter)
 208 {
 209         struct ixgbe_hw *hw = &adapter->hw;
 210         u16 link_status = 0;
 211         int err;
 212
 213         hw->bus.type = ixgbe_bus_type_pci_express;
 214
 215         /* Get the negotiated link width and speed from PCI config space of the
 216          * parent, as this device is behind a switch
 217          */
 218         err = ixgbe_read_pci_cfg_word_parent(adapter, 18, &link_status);
 219
 220         /* assume caller will handle error case */
 221         if (err)
 222                 return err;
 223
 224         hw->bus.width = ixgbe_convert_bus_width(link_status);
 225         hw->bus.speed = ixgbe_convert_bus_speed(link_status);
 226
 227         return 0;
 228 }
 229
 230 /**
 231  * ixgbe_pcie_from_parent - Determine whether PCIe info should come from parent
 232  * @hw: hw specific details
 233  *
 234  * This function is used by probe to determine whether a device's PCI-Express
 235  * bandwidth details should be gathered from the parent bus instead of from the
 236  * device. Used to ensure that various locations all have the correct device ID
 237  * checks.
 238  */
 239 static inline bool ixgbe_pcie_from_parent(struct ixgbe_hw *hw)
 240 {
 241         switch (hw->device_id) {
 242         case IXGBE_DEV_ID_82599_SFP_SF_QP:
 243         case IXGBE_DEV_ID_82599_QSFP_SF_QP:
 244                 return true;
 245         default:
 246                 return false;
 247         }
 248 }
 249
 250 static void ixgbe_check_minimum_link(struct ixgbe_adapter *adapter,
 251                                      int expected_gts)
 252 {
 253         struct ixgbe_hw *hw = &adapter->hw;
 254         struct pci_dev *pdev;
 255
 256         /* Some devices are not connected over PCIe and thus do not negotiate
 257          * speed. These devices do not have valid bus info, and thus any report
 258          * we generate may not be correct.
 259          */
 260         if (hw->bus.type == ixgbe_bus_type_internal)
 261                 return;
 262
 263         /* determine whether to use the parent device */
 264         if (ixgbe_pcie_from_parent(&adapter->hw))
 265                 pdev = adapter->pdev->bus->parent->self;
 266         else
 267                 pdev = adapter->pdev;
 268
 269         pcie_print_link_status(pdev);
 270 }
 271
 272 static void ixgbe_service_event_schedule(struct ixgbe_adapter *adapter)
 273 {
 274         if (!test_bit(__IXGBE_DOWN, &adapter->state) &&
 275             !test_bit(__IXGBE_REMOVING, &adapter->state) &&
 276             !test_and_set_bit(__IXGBE_SERVICE_SCHED, &adapter->state))
 277                 queue_work(ixgbe_wq, &adapter->service_task);
 278 }
 279
 280 static void ixgbe_remove_adapter(struct ixgbe_hw *hw)
 281 {
 282         struct ixgbe_adapter *adapter = hw->back;
 283
 284         if (!hw->hw_addr)
 285                 return;
 286         hw->hw_addr = NULL;
 287         e_dev_err("Adapter removed\n");
 288         if (test_bit(__IXGBE_SERVICE_INITED, &adapter->state))
 289                 ixgbe_service_event_schedule(adapter);
 290 }
 291
 292 static u32 ixgbe_check_remove(struct ixgbe_hw *hw, u32 reg)
 293 {
 294         u8 __iomem *reg_addr;
 295         u32 value;
 296         int i;
 297
 298         reg_addr = READ_ONCE(hw->hw_addr);
 299         if (ixgbe_removed(reg_addr))
 300                 return IXGBE_FAILED_READ_REG;
 301
 302         /* Register read of 0xFFFFFFF can indicate the adapter has been removed,
 303          * so perform several status register reads to determine if the adapter
 304          * has been removed.
 305          */
 306         for (i = 0; i < IXGBE_FAILED_READ_RETRIES; i++) {
 307                 value = readl(reg_addr + IXGBE_STATUS);
 308                 if (value != IXGBE_FAILED_READ_REG)
 309                         break;
 310                 mdelay(3);
 311         }
 312
 313         if (value == IXGBE_FAILED_READ_REG)
 314                 ixgbe_remove_adapter(hw);
 315         else
 316                 value = readl(reg_addr + reg);
 317         return value;
 318 }
 319
 320 /**
 321  * ixgbe_read_reg - Read from device register
 322  * @hw: hw specific details
 323  * @reg: offset of register to read
 324  *
 325  * Returns : value read or IXGBE_FAILED_READ_REG if removed
 326  *
 327  * This function is used to read device registers. It checks for device
 328  * removal by confirming any read that returns all ones by checking the
 329  * status register value for all ones. This function avoids reading from
 330  * the hardware if a removal was previously detected in which case it
 331  * returns IXGBE_FAILED_READ_REG (all ones).
 332  */
 333 u32 ixgbe_read_reg(struct ixgbe_hw *hw, u32 reg)
 334 {
 335         u8 __iomem *reg_addr = READ_ONCE(hw->hw_addr);
 336         u32 value;
 337
 338         if (ixgbe_removed(reg_addr))
 339                 return IXGBE_FAILED_READ_REG;
 340         if (unlikely(hw->phy.nw_mng_if_sel &
 341                      IXGBE_NW_MNG_IF_SEL_SGMII_ENABLE)) {
 342                 struct ixgbe_adapter *adapter;
 343                 int i;
 344
 345                 for (i = 0; i < 200; ++i) {
 346                         value = readl(reg_addr + IXGBE_MAC_SGMII_BUSY);
 347                         if (likely(!value))
 348                                 goto writes_completed;
 349                         if (value == IXGBE_FAILED_READ_REG) {
 350                                 ixgbe_remove_adapter(hw);
 351                                 return IXGBE_FAILED_READ_REG;
 352                         }
 353                         udelay(5);
 354                 }
 355
 356                 adapter = hw->back;
 357                 e_warn(hw, "register writes incomplete %08x\n", value);
 358         }
 359
 360 writes_completed:
 361         value = readl(reg_addr + reg);
 362         if (unlikely(value == IXGBE_FAILED_READ_REG))
 363                 value = ixgbe_check_remove(hw, reg);
 364         return value;
 365 }
 366
 367 static bool ixgbe_check_cfg_remove(struct ixgbe_hw *hw, struct pci_dev *pdev)
 368 {
 369         u16 value;
 370
 371         pci_read_config_word(pdev, PCI_VENDOR_ID, &value);
 372         if (value == IXGBE_FAILED_READ_CFG_WORD) {
 373                 ixgbe_remove_adapter(hw);
 374                 return true;
 375         }
 376         return false;
 377 }
 378
 379 u16 ixgbe_read_pci_cfg_word(struct ixgbe_hw *hw, u32 reg)
 380 {
 381         struct ixgbe_adapter *adapter = hw->back;
 382         u16 value;
 383
 384         if (ixgbe_removed(hw->hw_addr))
 385                 return IXGBE_FAILED_READ_CFG_WORD;
 386         pci_read_config_word(adapter->pdev, reg, &value);
 387         if (value == IXGBE_FAILED_READ_CFG_WORD &&
 388             ixgbe_check_cfg_remove(hw, adapter->pdev))
 389                 return IXGBE_FAILED_READ_CFG_WORD;
 390         return value;
 391 }
 392
 393 #ifdef CONFIG_PCI_IOV
 394 static u32 ixgbe_read_pci_cfg_dword(struct ixgbe_hw *hw, u32 reg)
 395 {
 396         struct ixgbe_adapter *adapter = hw->back;
 397         u32 value;
 398
 399         if (ixgbe_removed(hw->hw_addr))
 400                 return IXGBE_FAILED_READ_CFG_DWORD;
 401         pci_read_config_dword(adapter->pdev, reg, &value);
 402         if (value == IXGBE_FAILED_READ_CFG_DWORD &&
 403             ixgbe_check_cfg_remove(hw, adapter->pdev))
 404                 return IXGBE_FAILED_READ_CFG_DWORD;
 405         return value;
 406 }
 407 #endif /* CONFIG_PCI_IOV */
 408
 409 void ixgbe_write_pci_cfg_word(struct ixgbe_hw *hw, u32 reg, u16 value)
 410 {
 411         struct ixgbe_adapter *adapter = hw->back;
 412
 413         if (ixgbe_removed(hw->hw_addr))
 414                 return;
 415         pci_write_config_word(adapter->pdev, reg, value);
 416 }
 417
 418 static void ixgbe_service_event_complete(struct ixgbe_adapter *adapter)
 419 {
 420         BUG_ON(!test_bit(__IXGBE_SERVICE_SCHED, &adapter->state));
 421
 422         /* flush memory to make sure state is correct before next watchdog */
 423         smp_mb__before_atomic();
 424         clear_bit(__IXGBE_SERVICE_SCHED, &adapter->state);
 425 }
 426
 427 struct ixgbe_reg_info {
 428         u32 ofs;
 429         char *name;
 430 };
 431
 432 static const struct ixgbe_reg_info ixgbe_reg_info_tbl[] = {
 433
 434         /* General Registers */
 435         {IXGBE_CTRL, "CTRL"},
 436         {IXGBE_STATUS, "STATUS"},
 437         {IXGBE_CTRL_EXT, "CTRL_EXT"},
 438
 439         /* Interrupt Registers */
 440         {IXGBE_EICR, "EICR"},
 441
 442         /* RX Registers */
 443         {IXGBE_SRRCTL(0), "SRRCTL"},
 444         {IXGBE_DCA_RXCTRL(0), "DRXCTL"},
 445         {IXGBE_RDLEN(0), "RDLEN"},
 446         {IXGBE_RDH(0), "RDH"},
 447         {IXGBE_RDT(0), "RDT"},
 448         {IXGBE_RXDCTL(0), "RXDCTL"},
 449         {IXGBE_RDBAL(0), "RDBAL"},
 450         {IXGBE_RDBAH(0), "RDBAH"},
 451
 452         /* TX Registers */
 453         {IXGBE_TDBAL(0), "TDBAL"},
 454         {IXGBE_TDBAH(0), "TDBAH"},
 455         {IXGBE_TDLEN(0), "TDLEN"},
 456         {IXGBE_TDH(0), "TDH"},
 457         {IXGBE_TDT(0), "TDT"},
 458         {IXGBE_TXDCTL(0), "TXDCTL"},
 459
 460         /* List Terminator */
 461         { .name = NULL }
 462 };
 463
 464
 465 /*
 466  * ixgbe_regdump - register printout routine
 467  */
 468 static void ixgbe_regdump(struct ixgbe_hw *hw, struct ixgbe_reg_info *reginfo)
 469 {
 470         int i;
 471         char rname[16];
 472         u32 regs[64];
 473
 474         switch (reginfo->ofs) {
 475         case IXGBE_SRRCTL(0):
 476                 for (i = 0; i < 64; i++)
 477                         regs[i] = IXGBE_READ_REG(hw, IXGBE_SRRCTL(i));
 478                 break;
 479         case IXGBE_DCA_RXCTRL(0):
 480                 for (i = 0; i < 64; i++)
 481                         regs[i] = IXGBE_READ_REG(hw, IXGBE_DCA_RXCTRL(i));
 482                 break;
 483         case IXGBE_RDLEN(0):
 484                 for (i = 0; i < 64; i++)
 485                         regs[i] = IXGBE_READ_REG(hw, IXGBE_RDLEN(i));
 486                 break;
 487         case IXGBE_RDH(0):
 488                 for (i = 0; i < 64; i++)
 489                         regs[i] = IXGBE_READ_REG(hw, IXGBE_RDH(i));
 490                 break;
 491         case IXGBE_RDT(0):
 492                 for (i = 0; i < 64; i++)
 493                         regs[i] = IXGBE_READ_REG(hw, IXGBE_RDT(i));
 494                 break;
 495         case IXGBE_RXDCTL(0):
 496                 for (i = 0; i < 64; i++)
 497                         regs[i] = IXGBE_READ_REG(hw, IXGBE_RXDCTL(i));
 498                 break;
 499         case IXGBE_RDBAL(0):
 500                 for (i = 0; i < 64; i++)
 501                         regs[i] = IXGBE_READ_REG(hw, IXGBE_RDBAL(i));
 502                 break;
 503         case IXGBE_RDBAH(0):
 504                 for (i = 0; i < 64; i++)
 505                         regs[i] = IXGBE_READ_REG(hw, IXGBE_RDBAH(i));
 506                 break;
 507         case IXGBE_TDBAL(0):
 508                 for (i = 0; i < 64; i++)
 509                         regs[i] = IXGBE_READ_REG(hw, IXGBE_TDBAL(i));
 510                 break;
 511         case IXGBE_TDBAH(0):
 512                 for (i = 0; i < 64; i++)
 513                         regs[i] = IXGBE_READ_REG(hw, IXGBE_TDBAH(i));
 514                 break;
 515         case IXGBE_TDLEN(0):
 516                 for (i = 0; i < 64; i++)
 517                         regs[i] = IXGBE_READ_REG(hw, IXGBE_TDLEN(i));
 518                 break;
 519         case IXGBE_TDH(0):
 520                 for (i = 0; i < 64; i++)
 521                         regs[i] = IXGBE_READ_REG(hw, IXGBE_TDH(i));
 522                 break;
 523         case IXGBE_TDT(0):
 524                 for (i = 0; i < 64; i++)
 525                         regs[i] = IXGBE_READ_REG(hw, IXGBE_TDT(i));
 526                 break;
 527         case IXGBE_TXDCTL(0):
 528                 for (i = 0; i < 64; i++)
 529                         regs[i] = IXGBE_READ_REG(hw, IXGBE_TXDCTL(i));
 530                 break;
 531         default:
 532                 pr_info("%-15s %08x\n",
 533                         reginfo->name, IXGBE_READ_REG(hw, reginfo->ofs));
 534                 return;
 535         }
 536
 537         i = 0;
 538         while (i < 64) {
 539                 int j;
 540                 char buf[9 * 8 + 1];
 541                 char *p = buf;
 542
 543                 snprintf(rname, 16, "%s[%d-%d]", reginfo->name, i, i + 7);
 544                 for (j = 0; j < 8; j++)
 545                         p += sprintf(p, " %08x", regs[i++]);
 546                 pr_err("%-15s%s\n", rname, buf);
 547         }
 548
 549 }
 550
 551 static void ixgbe_print_buffer(struct ixgbe_ring *ring, int n)
 552 {
 553         struct ixgbe_tx_buffer *tx_buffer;
 554
 555         tx_buffer = &ring->tx_buffer_info[ring->next_to_clean];
 556         pr_info(" %5d %5X %5X %016llX %08X %p %016llX\n",
 557                 n, ring->next_to_use, ring->next_to_clean,
 558                 (u64)dma_unmap_addr(tx_buffer, dma),
 559                 dma_unmap_len(tx_buffer, len),
 560                 tx_buffer->next_to_watch,
 561                 (u64)tx_buffer->time_stamp);
 562 }
 563
 564 /*
 565  * ixgbe_dump - Print registers, tx-rings and rx-rings
 566  */
 567 static void ixgbe_dump(struct ixgbe_adapter *adapter)
 568 {
 569         struct net_device *netdev = adapter->netdev;
 570         struct ixgbe_hw *hw = &adapter->hw;
 571         struct ixgbe_reg_info *reginfo;
 572         int n = 0;
 573         struct ixgbe_ring *ring;
 574         struct ixgbe_tx_buffer *tx_buffer;
 575         union ixgbe_adv_tx_desc *tx_desc;
 576         struct my_u0 { u64 a; u64 b; } *u0;
 577         struct ixgbe_ring *rx_ring;
 578         union ixgbe_adv_rx_desc *rx_desc;
 579         struct ixgbe_rx_buffer *rx_buffer_info;
 580         int i = 0;
 581
 582         if (!netif_msg_hw(adapter))
 583                 return;
 584
 585         /* Print netdevice Info */
 586         if (netdev) {
 587                 dev_info(&adapter->pdev->dev, "Net device Info\n");
 588                 pr_info("Device Name     state            "
 589                         "trans_start\n");
 590                 pr_info("%-15s %016lX %016lX\n",
 591                         netdev->name,
 592                         netdev->state,
 593                         dev_trans_start(netdev));
 594         }
 595
 596         /* Print Registers */
 597         dev_info(&adapter->pdev->dev, "Register Dump\n");
 598         pr_info(" Register Name   Value\n");
 599         for (reginfo = (struct ixgbe_reg_info *)ixgbe_reg_info_tbl;
 600              reginfo->name; reginfo++) {
 601                 ixgbe_regdump(hw, reginfo);
 602         }
 603
 604         /* Print TX Ring Summary */
 605         if (!netdev || !netif_running(netdev))
 606                 return;
 607
 608         dev_info(&adapter->pdev->dev, "TX Rings Summary\n");
 609         pr_info(" %s     %s              %s        %s\n",
 610                 "Queue [NTU] [NTC] [bi(ntc)->dma  ]",
 611                 "leng", "ntw", "timestamp");
 612         for (n = 0; n < adapter->num_tx_queues; n++) {
 613                 ring = adapter->tx_ring[n];
 614                 ixgbe_print_buffer(ring, n);
 615         }
 616
 617         for (n = 0; n < adapter->num_xdp_queues; n++) {
 618                 ring = adapter->xdp_ring[n];
 619                 ixgbe_print_buffer(ring, n);
 620         }
 621
 622         /* Print TX Rings */
 623         if (!netif_msg_tx_done(adapter))
 624                 goto rx_ring_summary;
 625
 626         dev_info(&adapter->pdev->dev, "TX Rings Dump\n");
 627
 628         /* Transmit Descriptor Formats
 629          *
 630          * 82598 Advanced Transmit Descriptor
 631          *   +--------------------------------------------------------------+
 632          * 0 |         Buffer Address [63:0]                                |
 633          *   +--------------------------------------------------------------+
 634          * 8 |  PAYLEN  | POPTS  | IDX | STA | DCMD  |DTYP |  RSV |  DTALEN |
 635          *   +--------------------------------------------------------------+
 636          *   63       46 45    40 39 36 35 32 31   24 23 20 19              0
 637          *
 638          * 82598 Advanced Transmit Descriptor (Write-Back Format)
 639          *   +--------------------------------------------------------------+
 640          * 0 |                          RSV [63:0]                          |
 641          *   +--------------------------------------------------------------+
 642          * 8 |            RSV           |  STA  |          NXTSEQ           |
 643          *   +--------------------------------------------------------------+
 644          *   63                       36 35   32 31                         0
 645          *
 646          * 82599+ Advanced Transmit Descriptor
 647          *   +--------------------------------------------------------------+
 648          * 0 |         Buffer Address [63:0]                                |
 649          *   +--------------------------------------------------------------+
 650          * 8 |PAYLEN  |POPTS|CC|IDX  |STA  |DCMD  |DTYP |MAC  |RSV  |DTALEN |
 651          *   +--------------------------------------------------------------+
 652          *   63     46 45 40 39 38 36 35 32 31  24 23 20 19 18 17 16 15     0
 653          *
 654          * 82599+ Advanced Transmit Descriptor (Write-Back Format)
 655          *   +--------------------------------------------------------------+
 656          * 0 |                          RSV [63:0]                          |
 657          *   +--------------------------------------------------------------+
 658          * 8 |            RSV           |  STA  |           RSV             |
 659          *   +--------------------------------------------------------------+
 660          *   63                       36 35   32 31                         0
 661          */
 662
 663         for (n = 0; n < adapter->num_tx_queues; n++) {
 664                 ring = adapter->tx_ring[n];
 665                 pr_info("------------------------------------\n");
 666                 pr_info("TX QUEUE INDEX = %d\n", ring->queue_index);
 667                 pr_info("------------------------------------\n");
 668                 pr_info("%s%s    %s              %s        %s          %s\n",
 669                         "T [desc]     [address 63:0  ] ",
 670                         "[PlPOIdStDDt Ln] [bi->dma       ] ",
 671                         "leng", "ntw", "timestamp", "bi->skb");
 672
 673                 for (i = 0; ring->desc && (i < ring->count); i++) {
 674                         tx_desc = IXGBE_TX_DESC(ring, i);
 675                         tx_buffer = &ring->tx_buffer_info[i];
 676                         u0 = (struct my_u0 *)tx_desc;
 677                         if (dma_unmap_len(tx_buffer, len) > 0) {
 678                                 const char *ring_desc;
 679
 680                                 if (i == ring->next_to_use &&
 681                                     i == ring->next_to_clean)
 682                                         ring_desc = " NTC/U";
 683                                 else if (i == ring->next_to_use)
 684                                         ring_desc = " NTU";
 685                                 else if (i == ring->next_to_clean)
 686                                         ring_desc = " NTC";
 687                                 else
 688                                         ring_desc = "";
 689                                 pr_info("T [0x%03X]    %016llX %016llX %016llX %08X %p %016llX %p%s",
 690                                         i,
 691                                         le64_to_cpu((__force __le64)u0->a),
 692                                         le64_to_cpu((__force __le64)u0->b),
 693                                         (u64)dma_unmap_addr(tx_buffer, dma),
 694                                         dma_unmap_len(tx_buffer, len),
 695                                         tx_buffer->next_to_watch,
 696                                         (u64)tx_buffer->time_stamp,
 697                                         tx_buffer->skb,
 698                                         ring_desc);
 699
 700                                 if (netif_msg_pktdata(adapter) &&
 701                                     tx_buffer->skb)
 702                                         print_hex_dump(KERN_INFO, "",
 703                                                 DUMP_PREFIX_ADDRESS, 16, 1,
 704                                                 tx_buffer->skb->data,
 705                                                 dma_unmap_len(tx_buffer, len),
 706                                                 true);
 707                         }
 708                 }
 709         }
 710
 711         /* Print RX Rings Summary */
 712 rx_ring_summary:
 713         dev_info(&adapter->pdev->dev, "RX Rings Summary\n");
 714         pr_info("Queue [NTU] [NTC]\n");
 715         for (n = 0; n < adapter->num_rx_queues; n++) {
 716                 rx_ring = adapter->rx_ring[n];
 717                 pr_info("%5d %5X %5X\n",
 718                         n, rx_ring->next_to_use, rx_ring->next_to_clean);
 719         }
 720
 721         /* Print RX Rings */
 722         if (!netif_msg_rx_status(adapter))
 723                 return;
 724
 725         dev_info(&adapter->pdev->dev, "RX Rings Dump\n");
 726
 727         /* Receive Descriptor Formats
 728          *
 729          * 82598 Advanced Receive Descriptor (Read) Format
 730          *    63                                           1        0
 731          *    +-----------------------------------------------------+
 732          *  0 |       Packet Buffer Address [63:1]           |A0/NSE|
 733          *    +----------------------------------------------+------+
 734          *  8 |       Header Buffer Address [63:1]           |  DD  |
 735          *    +-----------------------------------------------------+
 736          *
 737          *
 738          * 82598 Advanced Receive Descriptor (Write-Back) Format
 739          *
 740          *   63       48 47    32 31  30      21 20 16 15   4 3     0
 741          *   +------------------------------------------------------+
 742          * 0 |       RSS Hash /  |SPH| HDR_LEN  | RSV |Packet|  RSS |
 743          *   | Packet   | IP     |   |          |     | Type | Type |
 744          *   | Checksum | Ident  |   |          |     |      |      |
 745          *   +------------------------------------------------------+
 746          * 8 | VLAN Tag | Length | Extended Error | Extended Status |
 747          *   +------------------------------------------------------+
 748          *   63       48 47    32 31            20 19               0
 749          *
 750          * 82599+ Advanced Receive Descriptor (Read) Format
 751          *    63                                           1        0
 752          *    +-----------------------------------------------------+
 753          *  0 |       Packet Buffer Address [63:1]           |A0/NSE|
 754          *    +----------------------------------------------+------+
 755          *  8 |       Header Buffer Address [63:1]           |  DD  |
 756          *    +-----------------------------------------------------+
 757          *
 758          *
 759          * 82599+ Advanced Receive Descriptor (Write-Back) Format
 760          *
 761          *   63       48 47    32 31  30      21 20 17 16   4 3     0
 762          *   +------------------------------------------------------+
 763          * 0 |RSS / Frag Checksum|SPH| HDR_LEN  |RSC- |Packet|  RSS |
 764          *   |/ RTT / PCoE_PARAM |   |          | CNT | Type | Type |
 765          *   |/ Flow Dir Flt ID  |   |          |     |      |      |
 766          *   +------------------------------------------------------+
 767          * 8 | VLAN Tag | Length |Extended Error| Xtnd Status/NEXTP |
 768          *   +------------------------------------------------------+
 769          *   63       48 47    32 31          20 19                 0
 770          */
 771
 772         for (n = 0; n < adapter->num_rx_queues; n++) {
 773                 rx_ring = adapter->rx_ring[n];
 774                 pr_info("------------------------------------\n");
 775                 pr_info("RX QUEUE INDEX = %d\n", rx_ring->queue_index);
 776                 pr_info("------------------------------------\n");
 777                 pr_info("%s%s%s\n",
 778                         "R  [desc]      [ PktBuf     A0] ",
 779                         "[  HeadBuf   DD] [bi->dma       ] [bi->skb       ] ",
 780                         "<-- Adv Rx Read format");
 781                 pr_info("%s%s%s\n",
 782                         "RWB[desc]      [PcsmIpSHl PtRs] ",
 783                         "[vl er S cks ln] ---------------- [bi->skb       ] ",
 784                         "<-- Adv Rx Write-Back format");
 785
 786                 for (i = 0; i < rx_ring->count; i++) {
 787                         const char *ring_desc;
 788
 789                         if (i == rx_ring->next_to_use)
 790                                 ring_desc = " NTU";
 791                         else if (i == rx_ring->next_to_clean)
 792                                 ring_desc = " NTC";
 793                         else
 794                                 ring_desc = "";
 795
 796                         rx_buffer_info = &rx_ring->rx_buffer_info[i];
 797                         rx_desc = IXGBE_RX_DESC(rx_ring, i);
 798                         u0 = (struct my_u0 *)rx_desc;
 799                         if (rx_desc->wb.upper.length) {
 800                                 /* Descriptor Done */
 801                                 pr_info("RWB[0x%03X]     %016llX %016llX ---------------- %p%s\n",
 802                                         i,
 803                                         le64_to_cpu((__force __le64)u0->a),
 804                                         le64_to_cpu((__force __le64)u0->b),
 805                                         rx_buffer_info->skb,
 806                                         ring_desc);
 807                         } else {
 808                                 pr_info("R  [0x%03X]     %016llX %016llX %016llX %p%s\n",
 809                                         i,
 810                                         le64_to_cpu((__force __le64)u0->a),
 811                                         le64_to_cpu((__force __le64)u0->b),
 812                                         (u64)rx_buffer_info->dma,
 813                                         rx_buffer_info->skb,
 814                                         ring_desc);
 815
 816                                 if (netif_msg_pktdata(adapter) &&
 817                                     rx_buffer_info->dma) {
 818                                         print_hex_dump(KERN_INFO, "",
 819                                            DUMP_PREFIX_ADDRESS, 16, 1,
 820                                            page_address(rx_buffer_info->page) +
 821                                                     rx_buffer_info->page_offset,
 822                                            ixgbe_rx_bufsz(rx_ring), true);
 823                                 }
 824                         }
 825                 }
 826         }
 827 }
 828
 829 static void ixgbe_release_hw_control(struct ixgbe_adapter *adapter)
 830 {
 831         u32 ctrl_ext;
 832
 833         /* Let firmware take over control of h/w */
 834         ctrl_ext = IXGBE_READ_REG(&adapter->hw, IXGBE_CTRL_EXT);
 835         IXGBE_WRITE_REG(&adapter->hw, IXGBE_CTRL_EXT,
 836                         ctrl_ext & ~IXGBE_CTRL_EXT_DRV_LOAD);
 837 }
 838
 839 static void ixgbe_get_hw_control(struct ixgbe_adapter *adapter)
 840 {
 841         u32 ctrl_ext;
 842
 843         /* Let firmware know the driver has taken over */
 844         ctrl_ext = IXGBE_READ_REG(&adapter->hw, IXGBE_CTRL_EXT);
 845         IXGBE_WRITE_REG(&adapter->hw, IXGBE_CTRL_EXT,
 846                         ctrl_ext | IXGBE_CTRL_EXT_DRV_LOAD);
 847 }
 848
 849 /**
 850  * ixgbe_set_ivar - set the IVAR registers, mapping interrupt causes to vectors
 851  * @adapter: pointer to adapter struct
 852  * @direction: 0 for Rx, 1 for Tx, -1 for other causes
 853  * @queue: queue to map the corresponding interrupt to
 854  * @msix_vector: the vector to map to the corresponding queue
 855  *
 856  */
 857 static void ixgbe_set_ivar(struct ixgbe_adapter *adapter, s8 direction,
 858                            u8 queue, u8 msix_vector)
 859 {
 860         u32 ivar, index;
 861         struct ixgbe_hw *hw = &adapter->hw;
 862         switch (hw->mac.type) {
 863         case ixgbe_mac_82598EB:
 864                 msix_vector |= IXGBE_IVAR_ALLOC_VAL;
 865                 if (direction == -1)
 866                         direction = 0;
 867                 index = (((direction * 64) + queue) >> 2) & 0x1F;
 868                 ivar = IXGBE_READ_REG(hw, IXGBE_IVAR(index));
 869                 ivar &= ~(0xFF << (8 * (queue & 0x3)));
 870                 ivar |= (msix_vector << (8 * (queue & 0x3)));
 871                 IXGBE_WRITE_REG(hw, IXGBE_IVAR(index), ivar);
 872                 break;
 873         case ixgbe_mac_82599EB:
 874         case ixgbe_mac_X540:
 875         case ixgbe_mac_X550:
 876         case ixgbe_mac_X550EM_x:
 877         case ixgbe_mac_x550em_a:
 878                 if (direction == -1) {
 879                         /* other causes */
 880                         msix_vector |= IXGBE_IVAR_ALLOC_VAL;
 881                         index = ((queue & 1) * 8);
 882                         ivar = IXGBE_READ_REG(&adapter->hw, IXGBE_IVAR_MISC);
 883                         ivar &= ~(0xFF << index);
 884                         ivar |= (msix_vector << index);
 885                         IXGBE_WRITE_REG(&adapter->hw, IXGBE_IVAR_MISC, ivar);
 886                         break;
 887                 } else {
 888                         /* tx or rx causes */
 889                         msix_vector |= IXGBE_IVAR_ALLOC_VAL;
 890                         index = ((16 * (queue & 1)) + (8 * direction));
 891                         ivar = IXGBE_READ_REG(hw, IXGBE_IVAR(queue >> 1));
 892                         ivar &= ~(0xFF << index);
 893                         ivar |= (msix_vector << index);
 894                         IXGBE_WRITE_REG(hw, IXGBE_IVAR(queue >> 1), ivar);
 895                         break;
 896                 }
 897         default:
 898                 break;
 899         }
 900 }
 901
 902 void ixgbe_irq_rearm_queues(struct ixgbe_adapter *adapter,
 903                             u64 qmask)
 904 {
 905         u32 mask;
 906
 907         switch (adapter->hw.mac.type) {
 908         case ixgbe_mac_82598EB:
 909                 mask = (IXGBE_EIMS_RTX_QUEUE & qmask);
 910                 IXGBE_WRITE_REG(&adapter->hw, IXGBE_EICS, mask);
 911                 break;
 912         case ixgbe_mac_82599EB:
 913         case ixgbe_mac_X540:
 914         case ixgbe_mac_X550:
 915         case ixgbe_mac_X550EM_x:
 916         case ixgbe_mac_x550em_a:
 917                 mask = (qmask & 0xFFFFFFFF);
 918                 IXGBE_WRITE_REG(&adapter->hw, IXGBE_EICS_EX(0), mask);
 919                 mask = (qmask >> 32);
 920                 IXGBE_WRITE_REG(&adapter->hw, IXGBE_EICS_EX(1), mask);
 921                 break;
 922         default:
 923                 break;
 924         }
 925 }
 926
 927 static void ixgbe_update_xoff_rx_lfc(struct ixgbe_adapter *adapter)
 928 {
 929         struct ixgbe_hw *hw = &adapter->hw;
 930         struct ixgbe_hw_stats *hwstats = &adapter->stats;
 931         int i;
 932         u32 data;
 933
 934         if ((hw->fc.current_mode != ixgbe_fc_full) &&
 935             (hw->fc.current_mode != ixgbe_fc_rx_pause))
 936                 return;
 937
 938         switch (hw->mac.type) {
 939         case ixgbe_mac_82598EB:
 940                 data = IXGBE_READ_REG(hw, IXGBE_LXOFFRXC);
 941                 break;
 942         default:
 943                 data = IXGBE_READ_REG(hw, IXGBE_LXOFFRXCNT);
 944         }
 945         hwstats->lxoffrxc += data;
 946
 947         /* refill credits (no tx hang) if we received xoff */
 948         if (!data)
 949                 return;
 950
 951         for (i = 0; i < adapter->num_tx_queues; i++)
 952                 clear_bit(__IXGBE_HANG_CHECK_ARMED,
 953                           &adapter->tx_ring[i]->state);
 954
 955         for (i = 0; i < adapter->num_xdp_queues; i++)
 956                 clear_bit(__IXGBE_HANG_CHECK_ARMED,
 957                           &adapter->xdp_ring[i]->state);
 958 }
 959
 960 static void ixgbe_update_xoff_received(struct ixgbe_adapter *adapter)
 961 {
 962         struct ixgbe_hw *hw = &adapter->hw;
 963         struct ixgbe_hw_stats *hwstats = &adapter->stats;
 964         u32 xoff[8] = {0};
 965         u8 tc;
 966         int i;
 967         bool pfc_en = adapter->dcb_cfg.pfc_mode_enable;
 968
 969         if (adapter->ixgbe_ieee_pfc)
 970                 pfc_en |= !!(adapter->ixgbe_ieee_pfc->pfc_en);
 971
 972         if (!(adapter->flags & IXGBE_FLAG_DCB_ENABLED) || !pfc_en) {
 973                 ixgbe_update_xoff_rx_lfc(adapter);
 974                 return;
 975         }
 976
 977         /* update stats for each tc, only valid with PFC enabled */
 978         for (i = 0; i < MAX_TX_PACKET_BUFFERS; i++) {
 979                 u32 pxoffrxc;
 980
 981                 switch (hw->mac.type) {
 982                 case ixgbe_mac_82598EB:
 983                         pxoffrxc = IXGBE_READ_REG(hw, IXGBE_PXOFFRXC(i));
 984                         break;
 985                 default:
 986                         pxoffrxc = IXGBE_READ_REG(hw, IXGBE_PXOFFRXCNT(i));
 987                 }
 988                 hwstats->pxoffrxc[i] += pxoffrxc;
 989                 /* Get the TC for given UP */
 990                 tc = netdev_get_prio_tc_map(adapter->netdev, i);
 991                 xoff[tc] += pxoffrxc;
 992         }
 993
 994         /* disarm tx queues that have received xoff frames */
 995         for (i = 0; i < adapter->num_tx_queues; i++) {
 996                 struct ixgbe_ring *tx_ring = adapter->tx_ring[i];
 997
 998                 tc = tx_ring->dcb_tc;
 999                 if (xoff[tc])
1000                         clear_bit(__IXGBE_HANG_CHECK_ARMED, &tx_ring->state);
1001         }
1002
1003         for (i = 0; i < adapter->num_xdp_queues; i++) {
1004                 struct ixgbe_ring *xdp_ring = adapter->xdp_ring[i];
1005
1006                 tc = xdp_ring->dcb_tc;
1007                 if (xoff[tc])
1008                         clear_bit(__IXGBE_HANG_CHECK_ARMED, &xdp_ring->state);
1009         }
1010 }
1011
1012 static u64 ixgbe_get_tx_completed(struct ixgbe_ring *ring)
1013 {
1014         return ring->stats.packets;
1015 }
1016
1017 static u64 ixgbe_get_tx_pending(struct ixgbe_ring *ring)
1018 {
1019         unsigned int head, tail;
1020
1021         head = ring->next_to_clean;
1022         tail = ring->next_to_use;
1023
1024         return ((head <= tail) ? tail : tail + ring->count) - head;
1025 }
1026
1027 static inline bool ixgbe_check_tx_hang(struct ixgbe_ring *tx_ring)
1028 {
1029         u32 tx_done = ixgbe_get_tx_completed(tx_ring);
1030         u32 tx_done_old = tx_ring->tx_stats.tx_done_old;
1031         u32 tx_pending = ixgbe_get_tx_pending(tx_ring);
1032
1033         clear_check_for_tx_hang(tx_ring);
1034
1035         /*
1036          * Check for a hung queue, but be thorough. This verifies
1037          * that a transmit has been completed since the previous
1038          * check AND there is at least one packet pending. The
1039          * ARMED bit is set to indicate a potential hang. The
1040          * bit is cleared if a pause frame is received to remove
1041          * false hang detection due to PFC or 802.3x frames. By
1042          * requiring this to fail twice we avoid races with
1043          * pfc clearing the ARMED bit and conditions where we
1044          * run the check_tx_hang logic with a transmit completion
1045          * pending but without time to complete it yet.
1046          */
1047         if (tx_done_old == tx_done && tx_pending)
1048                 /* make sure it is true for two checks in a row */
1049                 return test_and_set_bit(__IXGBE_HANG_CHECK_ARMED,
1050                                         &tx_ring->state);
1051         /* update completed stats and continue */
1052         tx_ring->tx_stats.tx_done_old = tx_done;
1053         /* reset the countdown */
1054         clear_bit(__IXGBE_HANG_CHECK_ARMED, &tx_ring->state);
1055
1056         return false;
1057 }
1058
1059 /**
1060  * ixgbe_tx_timeout_reset - initiate reset due to Tx timeout
1061  * @adapter: driver private struct
1062  **/
1063 static void ixgbe_tx_timeout_reset(struct ixgbe_adapter *adapter)
1064 {
1065
1066         /* Do the reset outside of interrupt context */
1067         if (!test_bit(__IXGBE_DOWN, &adapter->state)) {
1068                 set_bit(__IXGBE_RESET_REQUESTED, &adapter->state);
1069                 e_warn(drv, "initiating reset due to tx timeout\n");
1070                 ixgbe_service_event_schedule(adapter);
1071         }
1072 }
1073
1074 /**
1075  * ixgbe_tx_maxrate - callback to set the maximum per-queue bitrate
1076  * @netdev: network interface device structure
1077  * @queue_index: Tx queue to set
1078  * @maxrate: desired maximum transmit bitrate
1079  **/
1080 static int ixgbe_tx_maxrate(struct net_device *netdev,
1081                             int queue_index, u32 maxrate)
1082 {
1083         struct ixgbe_adapter *adapter = netdev_priv(netdev);
1084         struct ixgbe_hw *hw = &adapter->hw;
1085         u32 bcnrc_val = ixgbe_link_mbps(adapter);
1086
1087         if (!maxrate)
1088                 return 0;
1089
1090         /* Calculate the rate factor values to set */
1091         bcnrc_val <<= IXGBE_RTTBCNRC_RF_INT_SHIFT;
1092         bcnrc_val /= maxrate;
1093
1094         /* clear everything but the rate factor */
1095         bcnrc_val &= IXGBE_RTTBCNRC_RF_INT_MASK |
1096         IXGBE_RTTBCNRC_RF_DEC_MASK;
1097
1098         /* enable the rate scheduler */
1099         bcnrc_val |= IXGBE_RTTBCNRC_RS_ENA;
1100
1101         IXGBE_WRITE_REG(hw, IXGBE_RTTDQSEL, queue_index);
1102         IXGBE_WRITE_REG(hw, IXGBE_RTTBCNRC, bcnrc_val);
1103
1104         return 0;
1105 }
1106
1107 /**
1108  * ixgbe_clean_tx_irq - Reclaim resources after transmit completes
1109  * @q_vector: structure containing interrupt and ring information
1110  * @tx_ring: tx ring to clean
1111  * @napi_budget: Used to determine if we are in netpoll
1112  **/
1113 static bool ixgbe_clean_tx_irq(struct ixgbe_q_vector *q_vector,
1114                                struct ixgbe_ring *tx_ring, int napi_budget)
1115 {
1116         struct ixgbe_adapter *adapter = q_vector->adapter;
1117         struct ixgbe_tx_buffer *tx_buffer;
1118         union ixgbe_adv_tx_desc *tx_desc;
1119         unsigned int total_bytes = 0, total_packets = 0, total_ipsec = 0;
1120         unsigned int budget = q_vector->tx.work_limit;
1121         unsigned int i = tx_ring->next_to_clean;
1122
1123         if (test_bit(__IXGBE_DOWN, &adapter->state))
1124                 return true;
1125
1126         tx_buffer = &tx_ring->tx_buffer_info[i];
1127         tx_desc = IXGBE_TX_DESC(tx_ring, i);
1128         i -= tx_ring->count;
1129
1130         do {
1131                 union ixgbe_adv_tx_desc *eop_desc = tx_buffer->next_to_watch;
1132
1133                 /* if next_to_watch is not set then there is no work pending */
1134                 if (!eop_desc)
1135                         break;
1136
1137                 /* prevent any other reads prior to eop_desc */
1138                 smp_rmb();
1139
1140                 /* if DD is not set pending work has not been completed */
1141                 if (!(eop_desc->wb.status & cpu_to_le32(IXGBE_TXD_STAT_DD)))
1142                         break;
1143
1144                 /* clear next_to_watch to prevent false hangs */
1145                 tx_buffer->next_to_watch = NULL;
1146
1147                 /* update the statistics for this packet */
1148                 total_bytes += tx_buffer->bytecount;
1149                 total_packets += tx_buffer->gso_segs;
1150                 if (tx_buffer->tx_flags & IXGBE_TX_FLAGS_IPSEC)
1151                         total_ipsec++;
1152
1153                 /* free the skb */
1154                 if (ring_is_xdp(tx_ring))
1155                         xdp_return_frame(tx_buffer->xdpf);
1156                 else
1157                         napi_consume_skb(tx_buffer->skb, napi_budget);
1158
1159                 /* unmap skb header data */
1160                 dma_unmap_single(tx_ring->dev,
1161                                  dma_unmap_addr(tx_buffer, dma),
1162                                  dma_unmap_len(tx_buffer, len),
1163                                  DMA_TO_DEVICE);
1164
1165                 /* clear tx_buffer data */
1166                 dma_unmap_len_set(tx_buffer, len, 0);
1167
1168                 /* unmap remaining buffers */
1169                 while (tx_desc != eop_desc) {
1170                         tx_buffer++;
1171                         tx_desc++;
1172                         i++;
1173                         if (unlikely(!i)) {
1174                                 i -= tx_ring->count;
1175                                 tx_buffer = tx_ring->tx_buffer_info;
1176                                 tx_desc = IXGBE_TX_DESC(tx_ring, 0);
1177                         }
1178
1179                         /* unmap any remaining paged data */
1180                         if (dma_unmap_len(tx_buffer, len)) {
1181                                 dma_unmap_page(tx_ring->dev,
1182                                                dma_unmap_addr(tx_buffer, dma),
1183                                                dma_unmap_len(tx_buffer, len),
1184                                                DMA_TO_DEVICE);
1185                                 dma_unmap_len_set(tx_buffer, len, 0);
1186                         }
1187                 }
1188
1189                 /* move us one more past the eop_desc for start of next pkt */
1190                 tx_buffer++;
1191                 tx_desc++;
1192                 i++;
1193                 if (unlikely(!i)) {
1194                         i -= tx_ring->count;
1195                         tx_buffer = tx_ring->tx_buffer_info;
1196                         tx_desc = IXGBE_TX_DESC(tx_ring, 0);
1197                 }
1198
1199                 /* issue prefetch for next Tx descriptor */
1200                 prefetch(tx_desc);
1201
1202                 /* update budget accounting */
1203                 budget--;
1204         } while (likely(budget));
1205
1206         i += tx_ring->count;
1207         tx_ring->next_to_clean = i;
1208         u64_stats_update_begin(&tx_ring->syncp);
1209         tx_ring->stats.bytes += total_bytes;
1210         tx_ring->stats.packets += total_packets;
1211         u64_stats_update_end(&tx_ring->syncp);
1212         q_vector->tx.total_bytes += total_bytes;
1213         q_vector->tx.total_packets += total_packets;
1214         adapter->tx_ipsec += total_ipsec;
1215
1216         if (check_for_tx_hang(tx_ring) && ixgbe_check_tx_hang(tx_ring)) {
1217                 /* schedule immediate reset if we believe we hung */
1218                 struct ixgbe_hw *hw = &adapter->hw;
1219                 e_err(drv, "Detected Tx Unit Hang %s\n"
1220                         "  Tx Queue             <%d>\n"
1221                         "  TDH, TDT             <%x>, <%x>\n"
1222                         "  next_to_use          <%x>\n"
1223                         "  next_to_clean        <%x>\n"
1224                         "tx_buffer_info[next_to_clean]\n"
1225                         "  time_stamp           <%lx>\n"
1226                         "  jiffies              <%lx>\n",
1227                         ring_is_xdp(tx_ring) ? "(XDP)" : "",
1228                         tx_ring->queue_index,
1229                         IXGBE_READ_REG(hw, IXGBE_TDH(tx_ring->reg_idx)),
1230                         IXGBE_READ_REG(hw, IXGBE_TDT(tx_ring->reg_idx)),
1231                         tx_ring->next_to_use, i,
1232                         tx_ring->tx_buffer_info[i].time_stamp, jiffies);
1233
1234                 if (!ring_is_xdp(tx_ring))
1235                         netif_stop_subqueue(tx_ring->netdev,
1236                                             tx_ring->queue_index);
1237
1238                 e_info(probe,
1239                        "tx hang %d detected on queue %d, resetting adapter\n",
1240                         adapter->tx_timeout_count + 1, tx_ring->queue_index);
1241
1242                 /* schedule immediate reset if we believe we hung */
1243                 ixgbe_tx_timeout_reset(adapter);
1244
1245                 /* the adapter is about to reset, no point in enabling stuff */
1246                 return true;
1247         }
1248
1249         if (ring_is_xdp(tx_ring))
1250                 return !!budget;
1251
1252         netdev_tx_completed_queue(txring_txq(tx_ring),
1253                                   total_packets, total_bytes);
1254
1255 #define TX_WAKE_THRESHOLD (DESC_NEEDED * 2)
1256         if (unlikely(total_packets && netif_carrier_ok(tx_ring->netdev) &&
1257                      (ixgbe_desc_unused(tx_ring) >= TX_WAKE_THRESHOLD))) {
1258                 /* Make sure that anybody stopping the queue after this
1259                  * sees the new next_to_clean.
1260                  */
1261                 smp_mb();
1262                 if (__netif_subqueue_stopped(tx_ring->netdev,
1263                                              tx_ring->queue_index)
1264                     && !test_bit(__IXGBE_DOWN, &adapter->state)) {
1265                         netif_wake_subqueue(tx_ring->netdev,
1266                                             tx_ring->queue_index);
1267                         ++tx_ring->tx_stats.restart_queue;
1268                 }
1269         }
1270
1271         return !!budget;
1272 }
1273
1274 #ifdef CONFIG_IXGBE_DCA
1275 static void ixgbe_update_tx_dca(struct ixgbe_adapter *adapter,
1276                                 struct ixgbe_ring *tx_ring,
1277                                 int cpu)
1278 {
1279         struct ixgbe_hw *hw = &adapter->hw;
1280         u32 txctrl = 0;
1281         u16 reg_offset;
1282
1283         if (adapter->flags & IXGBE_FLAG_DCA_ENABLED)
1284                 txctrl = dca3_get_tag(tx_ring->dev, cpu);
1285
1286         switch (hw->mac.type) {
1287         case ixgbe_mac_82598EB:
1288                 reg_offset = IXGBE_DCA_TXCTRL(tx_ring->reg_idx);
1289                 break;
1290         case ixgbe_mac_82599EB:
1291         case ixgbe_mac_X540:
1292                 reg_offset = IXGBE_DCA_TXCTRL_82599(tx_ring->reg_idx);
1293                 txctrl <<= IXGBE_DCA_TXCTRL_CPUID_SHIFT_82599;
1294                 break;
1295         default:
1296                 /* for unknown hardware do not write register */
1297                 return;
1298         }
1299
1300         /*
1301          * We can enable relaxed ordering for reads, but not writes when
1302          * DCA is enabled.  This is due to a known issue in some chipsets
1303          * which will cause the DCA tag to be cleared.
1304          */
1305         txctrl |= IXGBE_DCA_TXCTRL_DESC_RRO_EN |
1306                   IXGBE_DCA_TXCTRL_DATA_RRO_EN |
1307                   IXGBE_DCA_TXCTRL_DESC_DCA_EN;
1308
1309         IXGBE_WRITE_REG(hw, reg_offset, txctrl);
1310 }
1311
1312 static void ixgbe_update_rx_dca(struct ixgbe_adapter *adapter,
1313                                 struct ixgbe_ring *rx_ring,
1314                                 int cpu)
1315 {
1316         struct ixgbe_hw *hw = &adapter->hw;
1317         u32 rxctrl = 0;
1318         u8 reg_idx = rx_ring->reg_idx;
1319
1320         if (adapter->flags & IXGBE_FLAG_DCA_ENABLED)
1321                 rxctrl = dca3_get_tag(rx_ring->dev, cpu);
1322
1323         switch (hw->mac.type) {
1324         case ixgbe_mac_82599EB:
1325         case ixgbe_mac_X540:
1326                 rxctrl <<= IXGBE_DCA_RXCTRL_CPUID_SHIFT_82599;
1327                 break;
1328         default:
1329                 break;
1330         }
1331
1332         /*
1333          * We can enable relaxed ordering for reads, but not writes when
1334          * DCA is enabled.  This is due to a known issue in some chipsets
1335          * which will cause the DCA tag to be cleared.
1336          */
1337         rxctrl |= IXGBE_DCA_RXCTRL_DESC_RRO_EN |
1338                   IXGBE_DCA_RXCTRL_DATA_DCA_EN |
1339                   IXGBE_DCA_RXCTRL_DESC_DCA_EN;
1340
1341         IXGBE_WRITE_REG(hw, IXGBE_DCA_RXCTRL(reg_idx), rxctrl);
1342 }
1343
1344 static void ixgbe_update_dca(struct ixgbe_q_vector *q_vector)
1345 {
1346         struct ixgbe_adapter *adapter = q_vector->adapter;
1347         struct ixgbe_ring *ring;
1348         int cpu = get_cpu();
1349
1350         if (q_vector->cpu == cpu)
1351                 goto out_no_update;
1352
1353         ixgbe_for_each_ring(ring, q_vector->tx)
1354                 ixgbe_update_tx_dca(adapter, ring, cpu);
1355
1356         ixgbe_for_each_ring(ring, q_vector->rx)
1357                 ixgbe_update_rx_dca(adapter, ring, cpu);
1358
1359         q_vector->cpu = cpu;
1360 out_no_update:
1361         put_cpu();
1362 }
1363
1364 static void ixgbe_setup_dca(struct ixgbe_adapter *adapter)
1365 {
1366         int i;
1367
1368         /* always use CB2 mode, difference is masked in the CB driver */
1369         if (adapter->flags & IXGBE_FLAG_DCA_ENABLED)
1370                 IXGBE_WRITE_REG(&adapter->hw, IXGBE_DCA_CTRL,
1371                                 IXGBE_DCA_CTRL_DCA_MODE_CB2);
1372         else
1373                 IXGBE_WRITE_REG(&adapter->hw, IXGBE_DCA_CTRL,
1374                                 IXGBE_DCA_CTRL_DCA_DISABLE);
1375
1376         for (i = 0; i < adapter->num_q_vectors; i++) {
1377                 adapter->q_vector[i]->cpu = -1;
1378                 ixgbe_update_dca(adapter->q_vector[i]);
1379         }
1380 }
1381
1382 static int __ixgbe_notify_dca(struct device *dev, void *data)
1383 {
1384         struct ixgbe_adapter *adapter = dev_get_drvdata(dev);
1385         unsigned long event = *(unsigned long *)data;
1386
1387         if (!(adapter->flags & IXGBE_FLAG_DCA_CAPABLE))
1388                 return 0;
1389
1390         switch (event) {
1391         case DCA_PROVIDER_ADD:
1392                 /* if we're already enabled, don't do it again */
1393                 if (adapter->flags & IXGBE_FLAG_DCA_ENABLED)
1394                         break;
1395                 if (dca_add_requester(dev) == 0) {
1396                         adapter->flags |= IXGBE_FLAG_DCA_ENABLED;
1397                         IXGBE_WRITE_REG(&adapter->hw, IXGBE_DCA_CTRL,
1398                                         IXGBE_DCA_CTRL_DCA_MODE_CB2);
1399                         break;
1400                 }
1401                 fallthrough; /* DCA is disabled. */
1402         case DCA_PROVIDER_REMOVE:
1403                 if (adapter->flags & IXGBE_FLAG_DCA_ENABLED) {
1404                         dca_remove_requester(dev);
1405                         adapter->flags &= ~IXGBE_FLAG_DCA_ENABLED;
1406                         IXGBE_WRITE_REG(&adapter->hw, IXGBE_DCA_CTRL,
1407                                         IXGBE_DCA_CTRL_DCA_DISABLE);
1408                 }
1409                 break;
1410         }
1411
1412         return 0;
1413 }
1414
1415 #endif /* CONFIG_IXGBE_DCA */
1416
1417 #define IXGBE_RSS_L4_TYPES_MASK \
1418         ((1ul << IXGBE_RXDADV_RSSTYPE_IPV4_TCP) | \
1419          (1ul << IXGBE_RXDADV_RSSTYPE_IPV4_UDP) | \
1420          (1ul << IXGBE_RXDADV_RSSTYPE_IPV6_TCP) | \
1421          (1ul << IXGBE_RXDADV_RSSTYPE_IPV6_UDP))
1422
1423 static inline void ixgbe_rx_hash(struct ixgbe_ring *ring,
1424                                  union ixgbe_adv_rx_desc *rx_desc,
1425                                  struct sk_buff *skb)
1426 {
1427         u16 rss_type;
1428
1429         if (!(ring->netdev->features & NETIF_F_RXHASH))
1430                 return;
1431
1432         rss_type = le16_to_cpu(rx_desc->wb.lower.lo_dword.hs_rss.pkt_info) &
1433                    IXGBE_RXDADV_RSSTYPE_MASK;
1434
1435         if (!rss_type)
1436                 return;
1437
1438         skb_set_hash(skb, le32_to_cpu(rx_desc->wb.lower.hi_dword.rss),
1439                      (IXGBE_RSS_L4_TYPES_MASK & (1ul << rss_type)) ?
1440                      PKT_HASH_TYPE_L4 : PKT_HASH_TYPE_L3);
1441 }
1442
1443 #ifdef IXGBE_FCOE
1444 /**
1445  * ixgbe_rx_is_fcoe - check the rx desc for incoming pkt type
1446  * @ring: structure containing ring specific data
1447  * @rx_desc: advanced rx descriptor
1448  *
1449  * Returns : true if it is FCoE pkt
1450  */
1451 static inline bool ixgbe_rx_is_fcoe(struct ixgbe_ring *ring,
1452                                     union ixgbe_adv_rx_desc *rx_desc)
1453 {
1454         __le16 pkt_info = rx_desc->wb.lower.lo_dword.hs_rss.pkt_info;
1455
1456         return test_bit(__IXGBE_RX_FCOE, &ring->state) &&
1457                ((pkt_info & cpu_to_le16(IXGBE_RXDADV_PKTTYPE_ETQF_MASK)) ==
1458                 (cpu_to_le16(IXGBE_ETQF_FILTER_FCOE <<
1459                              IXGBE_RXDADV_PKTTYPE_ETQF_SHIFT)));
1460 }
1461
1462 #endif /* IXGBE_FCOE */
1463 /**
1464  * ixgbe_rx_checksum - indicate in skb if hw indicated a good cksum
1465  * @ring: structure containing ring specific data
1466  * @rx_desc: current Rx descriptor being processed
1467  * @skb: skb currently being received and modified
1468  **/
1469 static inline void ixgbe_rx_checksum(struct ixgbe_ring *ring,
1470                                      union ixgbe_adv_rx_desc *rx_desc,
1471                                      struct sk_buff *skb)
1472 {
1473         __le16 pkt_info = rx_desc->wb.lower.lo_dword.hs_rss.pkt_info;
1474         bool encap_pkt = false;
1475
1476         skb_checksum_none_assert(skb);
1477
1478         /* Rx csum disabled */
1479         if (!(ring->netdev->features & NETIF_F_RXCSUM))
1480                 return;
1481
1482         /* check for VXLAN and Geneve packets */
1483         if (pkt_info & cpu_to_le16(IXGBE_RXDADV_PKTTYPE_VXLAN)) {
1484                 encap_pkt = true;
1485                 skb->encapsulation = 1;
1486         }
1487
1488         /* if IP and error */
1489         if (ixgbe_test_staterr(rx_desc, IXGBE_RXD_STAT_IPCS) &&
1490             ixgbe_test_staterr(rx_desc, IXGBE_RXDADV_ERR_IPE)) {
1491                 ring->rx_stats.csum_err++;
1492                 return;
1493         }
1494
1495         if (!ixgbe_test_staterr(rx_desc, IXGBE_RXD_STAT_L4CS))
1496                 return;
1497
1498         if (ixgbe_test_staterr(rx_desc, IXGBE_RXDADV_ERR_TCPE)) {
1499                 /*
1500                  * 82599 errata, UDP frames with a 0 checksum can be marked as
1501                  * checksum errors.
1502                  */
1503                 if ((pkt_info & cpu_to_le16(IXGBE_RXDADV_PKTTYPE_UDP)) &&
1504                     test_bit(__IXGBE_RX_CSUM_UDP_ZERO_ERR, &ring->state))
1505                         return;
1506
1507                 ring->rx_stats.csum_err++;
1508                 return;
1509         }
1510
1511         /* It must be a TCP or UDP packet with a valid checksum */
1512         skb->ip_summed = CHECKSUM_UNNECESSARY;
1513         if (encap_pkt) {
1514                 if (!ixgbe_test_staterr(rx_desc, IXGBE_RXD_STAT_OUTERIPCS))
1515                         return;
1516
1517                 if (ixgbe_test_staterr(rx_desc, IXGBE_RXDADV_ERR_OUTERIPER)) {
1518                         skb->ip_summed = CHECKSUM_NONE;
1519                         return;
1520                 }
1521                 /* If we checked the outer header let the stack know */
1522                 skb->csum_level = 1;
1523         }
1524 }
1525
1526 static unsigned int ixgbe_rx_offset(struct ixgbe_ring *rx_ring)
1527 {
1528         return ring_uses_build_skb(rx_ring) ? IXGBE_SKB_PAD : 0;
1529 }
1530
1531 static bool ixgbe_alloc_mapped_page(struct ixgbe_ring *rx_ring,
1532                                     struct ixgbe_rx_buffer *bi)
1533 {
1534         struct page *page = bi->page;
1535         dma_addr_t dma;
1536
1537         /* since we are recycling buffers we should seldom need to alloc */
1538         if (likely(page))
1539                 return true;
1540
1541         /* alloc new page for storage */
1542         page = dev_alloc_pages(ixgbe_rx_pg_order(rx_ring));
1543         if (unlikely(!page)) {
1544                 rx_ring->rx_stats.alloc_rx_page_failed++;
1545                 return false;
1546         }
1547
1548         /* map page for use */
1549         dma = dma_map_page_attrs(rx_ring->dev, page, 0,
1550                                  ixgbe_rx_pg_size(rx_ring),
1551                                  DMA_FROM_DEVICE,
1552                                  IXGBE_RX_DMA_ATTR);
1553
1554         /*
1555          * if mapping failed free memory back to system since
1556          * there isn't much point in holding memory we can't use
1557          */
1558         if (dma_mapping_error(rx_ring->dev, dma)) {
1559                 __free_pages(page, ixgbe_rx_pg_order(rx_ring));
1560
1561                 rx_ring->rx_stats.alloc_rx_page_failed++;
1562                 return false;
1563         }
1564
1565         bi->dma = dma;
1566         bi->page = page;
1567         bi->page_offset = rx_ring->rx_offset;
1568         page_ref_add(page, USHRT_MAX - 1);
1569         bi->pagecnt_bias = USHRT_MAX;
1570         rx_ring->rx_stats.alloc_rx_page++;
1571
1572         return true;
1573 }
1574
1575 /**
1576  * ixgbe_alloc_rx_buffers - Replace used receive buffers
1577  * @rx_ring: ring to place buffers on
1578  * @cleaned_count: number of buffers to replace
1579  **/
1580 void ixgbe_alloc_rx_buffers(struct ixgbe_ring *rx_ring, u16 cleaned_count)
1581 {
1582         union ixgbe_adv_rx_desc *rx_desc;
1583         struct ixgbe_rx_buffer *bi;
1584         u16 i = rx_ring->next_to_use;
1585         u16 bufsz;
1586
1587         /* nothing to do */
1588         if (!cleaned_count)
1589                 return;
1590
1591         rx_desc = IXGBE_RX_DESC(rx_ring, i);
1592         bi = &rx_ring->rx_buffer_info[i];
1593         i -= rx_ring->count;
1594
1595         bufsz = ixgbe_rx_bufsz(rx_ring);
1596
1597         do {
1598                 if (!ixgbe_alloc_mapped_page(rx_ring, bi))
1599                         break;
1600
1601                 /* sync the buffer for use by the device */
1602                 dma_sync_single_range_for_device(rx_ring->dev, bi->dma,
1603                                                  bi->page_offset, bufsz,
1604                                                  DMA_FROM_DEVICE);
1605
1606                 /*
1607                  * Refresh the desc even if buffer_addrs didn't change
1608                  * because each write-back erases this info.
1609                  */
1610                 rx_desc->read.pkt_addr = cpu_to_le64(bi->dma + bi->page_offset);
1611
1612                 rx_desc++;
1613                 bi++;
1614                 i++;
1615                 if (unlikely(!i)) {
1616                         rx_desc = IXGBE_RX_DESC(rx_ring, 0);
1617                         bi = rx_ring->rx_buffer_info;
1618                         i -= rx_ring->count;
1619                 }
1620
1621                 /* clear the length for the next_to_use descriptor */
1622                 rx_desc->wb.upper.length = 0;
1623
1624                 cleaned_count--;
1625         } while (cleaned_count);
1626
1627         i += rx_ring->count;
1628
1629         if (rx_ring->next_to_use != i) {
1630                 rx_ring->next_to_use = i;
1631
1632                 /* update next to alloc since we have filled the ring */
1633                 rx_ring->next_to_alloc = i;
1634
1635                 /* Force memory writes to complete before letting h/w
1636                  * know there are new descriptors to fetch.  (Only
1637                  * applicable for weak-ordered memory model archs,
1638                  * such as IA-64).
1639                  */
1640                 wmb();
1641                 writel(i, rx_ring->tail);
1642         }
1643 }
1644
1645 static void ixgbe_set_rsc_gso_size(struct ixgbe_ring *ring,
1646                                    struct sk_buff *skb)
1647 {
1648         u16 hdr_len = skb_headlen(skb);
1649
1650         /* set gso_size to avoid messing up TCP MSS */
1651         skb_shinfo(skb)->gso_size = DIV_ROUND_UP((skb->len - hdr_len),
1652                                                  IXGBE_CB(skb)->append_cnt);
1653         skb_shinfo(skb)->gso_type = SKB_GSO_TCPV4;
1654 }
1655
1656 static void ixgbe_update_rsc_stats(struct ixgbe_ring *rx_ring,
1657                                    struct sk_buff *skb)
1658 {
1659         /* if append_cnt is 0 then frame is not RSC */
1660         if (!IXGBE_CB(skb)->append_cnt)
1661                 return;
1662
1663         rx_ring->rx_stats.rsc_count += IXGBE_CB(skb)->append_cnt;
1664         rx_ring->rx_stats.rsc_flush++;
1665
1666         ixgbe_set_rsc_gso_size(rx_ring, skb);
1667
1668         /* gso_size is computed using append_cnt so always clear it last */
1669         IXGBE_CB(skb)->append_cnt = 0;
1670 }
1671
1672 /**
1673  * ixgbe_process_skb_fields - Populate skb header fields from Rx descriptor
1674  * @rx_ring: rx descriptor ring packet is being transacted on
1675  * @rx_desc: pointer to the EOP Rx descriptor
1676  * @skb: pointer to current skb being populated
1677  *
1678  * This function checks the ring, descriptor, and packet information in
1679  * order to populate the hash, checksum, VLAN, timestamp, protocol, and
1680  * other fields within the skb.
1681  **/
1682 void ixgbe_process_skb_fields(struct ixgbe_ring *rx_ring,
1683                               union ixgbe_adv_rx_desc *rx_desc,
1684                               struct sk_buff *skb)
1685 {
1686         struct net_device *dev = rx_ring->netdev;
1687         u32 flags = rx_ring->q_vector->adapter->flags;
1688
1689         ixgbe_update_rsc_stats(rx_ring, skb);
1690
1691         ixgbe_rx_hash(rx_ring, rx_desc, skb);
1692
1693         ixgbe_rx_checksum(rx_ring, rx_desc, skb);
1694
1695         if (unlikely(flags & IXGBE_FLAG_RX_HWTSTAMP_ENABLED))
1696                 ixgbe_ptp_rx_hwtstamp(rx_ring, rx_desc, skb);
1697
1698         if ((dev->features & NETIF_F_HW_VLAN_CTAG_RX) &&
1699             ixgbe_test_staterr(rx_desc, IXGBE_RXD_STAT_VP)) {
1700                 u16 vid = le16_to_cpu(rx_desc->wb.upper.vlan);
1701                 __vlan_hwaccel_put_tag(skb, htons(ETH_P_8021Q), vid);
1702         }
1703
1704         if (ixgbe_test_staterr(rx_desc, IXGBE_RXDADV_STAT_SECP))
1705                 ixgbe_ipsec_rx(rx_ring, rx_desc, skb);
1706
1707         /* record Rx queue, or update MACVLAN statistics */
1708         if (netif_is_ixgbe(dev))
1709                 skb_record_rx_queue(skb, rx_ring->queue_index);
1710         else
1711                 macvlan_count_rx(netdev_priv(dev), skb->len + ETH_HLEN, true,
1712                                  false);
1713
1714         skb->protocol = eth_type_trans(skb, dev);
1715 }
1716
1717 void ixgbe_rx_skb(struct ixgbe_q_vector *q_vector,
1718                   struct sk_buff *skb)
1719 {
1720         napi_gro_receive(&q_vector->napi, skb);
1721 }
1722
1723 /**
1724  * ixgbe_is_non_eop - process handling of non-EOP buffers
1725  * @rx_ring: Rx ring being processed
1726  * @rx_desc: Rx descriptor for current buffer
1727  * @skb: Current socket buffer containing buffer in progress
1728  *
1729  * This function updates next to clean.  If the buffer is an EOP buffer
1730  * this function exits returning false, otherwise it will place the
1731  * sk_buff in the next buffer to be chained and return true indicating
1732  * that this is in fact a non-EOP buffer.
1733  **/
1734 static bool ixgbe_is_non_eop(struct ixgbe_ring *rx_ring,
1735                              union ixgbe_adv_rx_desc *rx_desc,
1736                              struct sk_buff *skb)
1737 {
1738         u32 ntc = rx_ring->next_to_clean + 1;
1739
1740         /* fetch, update, and store next to clean */
1741         ntc = (ntc < rx_ring->count) ? ntc : 0;
1742         rx_ring->next_to_clean = ntc;
1743
1744         prefetch(IXGBE_RX_DESC(rx_ring, ntc));
1745
1746         /* update RSC append count if present */
1747         if (ring_is_rsc_enabled(rx_ring)) {
1748                 __le32 rsc_enabled = rx_desc->wb.lower.lo_dword.data &
1749                                      cpu_to_le32(IXGBE_RXDADV_RSCCNT_MASK);
1750
1751                 if (unlikely(rsc_enabled)) {
1752                         u32 rsc_cnt = le32_to_cpu(rsc_enabled);
1753
1754                         rsc_cnt >>= IXGBE_RXDADV_RSCCNT_SHIFT;
1755                         IXGBE_CB(skb)->append_cnt += rsc_cnt - 1;
1756
1757                         /* update ntc based on RSC value */
1758                         ntc = le32_to_cpu(rx_desc->wb.upper.status_error);
1759                         ntc &= IXGBE_RXDADV_NEXTP_MASK;
1760                         ntc >>= IXGBE_RXDADV_NEXTP_SHIFT;
1761                 }
1762         }
1763
1764         /* if we are the last buffer then there is nothing else to do */
1765         if (likely(ixgbe_test_staterr(rx_desc, IXGBE_RXD_STAT_EOP)))
1766                 return false;
1767
1768         /* place skb in next buffer to be received */
1769         rx_ring->rx_buffer_info[ntc].skb = skb;
1770         rx_ring->rx_stats.non_eop_descs++;
1771
1772         return true;
1773 }
1774
1775 /**
1776  * ixgbe_pull_tail - ixgbe specific version of skb_pull_tail
1777  * @rx_ring: rx descriptor ring packet is being transacted on
1778  * @skb: pointer to current skb being adjusted
1779  *
1780  * This function is an ixgbe specific version of __pskb_pull_tail.  The
1781  * main difference between this version and the original function is that
1782  * this function can make several assumptions about the state of things
1783  * that allow for significant optimizations versus the standard function.
1784  * As a result we can do things like drop a frag and maintain an accurate
1785  * truesize for the skb.
1786  */
1787 static void ixgbe_pull_tail(struct ixgbe_ring *rx_ring,
1788                             struct sk_buff *skb)
1789 {
1790         skb_frag_t *frag = &skb_shinfo(skb)->frags[0];
1791         unsigned char *va;
1792         unsigned int pull_len;
1793
1794         /*
1795          * it is valid to use page_address instead of kmap since we are
1796          * working with pages allocated out of the lomem pool per
1797          * alloc_page(GFP_ATOMIC)
1798          */
1799         va = skb_frag_address(frag);
1800
1801         /*
1802          * we need the header to contain the greater of either ETH_HLEN or
1803          * 60 bytes if the skb->len is less than 60 for skb_pad.
1804          */
1805         pull_len = eth_get_headlen(skb->dev, va, IXGBE_RX_HDR_SIZE);
1806
1807         /* align pull length to size of long to optimize memcpy performance */
1808         skb_copy_to_linear_data(skb, va, ALIGN(pull_len, sizeof(long)));
1809
1810         /* update all of the pointers */
1811         skb_frag_size_sub(frag, pull_len);
1812         skb_frag_off_add(frag, pull_len);
1813         skb->data_len -= pull_len;
1814         skb->tail += pull_len;
1815 }
1816
1817 /**
1818  * ixgbe_dma_sync_frag - perform DMA sync for first frag of SKB
1819  * @rx_ring: rx descriptor ring packet is being transacted on
1820  * @skb: pointer to current skb being updated
1821  *
1822  * This function provides a basic DMA sync up for the first fragment of an
1823  * skb.  The reason for doing this is that the first fragment cannot be
1824  * unmapped until we have reached the end of packet descriptor for a buffer
1825  * chain.
1826  */
1827 static void ixgbe_dma_sync_frag(struct ixgbe_ring *rx_ring,
1828                                 struct sk_buff *skb)
1829 {
1830         if (ring_uses_build_skb(rx_ring)) {
1831                 unsigned long mask = (unsigned long)ixgbe_rx_pg_size(rx_ring) - 1;
1832                 unsigned long offset = (unsigned long)(skb->data) & mask;
1833
1834                 dma_sync_single_range_for_cpu(rx_ring->dev,
1835                                               IXGBE_CB(skb)->dma,
1836                                               offset,
1837                                               skb_headlen(skb),
1838                                               DMA_FROM_DEVICE);
1839         } else {
1840                 skb_frag_t *frag = &skb_shinfo(skb)->frags[0];
1841
1842                 dma_sync_single_range_for_cpu(rx_ring->dev,
1843                                               IXGBE_CB(skb)->dma,
1844                                               skb_frag_off(frag),
1845                                               skb_frag_size(frag),
1846                                               DMA_FROM_DEVICE);
1847         }
1848
1849         /* If the page was released, just unmap it. */
1850         if (unlikely(IXGBE_CB(skb)->page_released)) {
1851                 dma_unmap_page_attrs(rx_ring->dev, IXGBE_CB(skb)->dma,
1852                                      ixgbe_rx_pg_size(rx_ring),
1853                                      DMA_FROM_DEVICE,
1854                                      IXGBE_RX_DMA_ATTR);
1855         }
1856 }
1857
1858 /**
1859  * ixgbe_cleanup_headers - Correct corrupted or empty headers
1860  * @rx_ring: rx descriptor ring packet is being transacted on
1861  * @rx_desc: pointer to the EOP Rx descriptor
1862  * @skb: pointer to current skb being fixed
1863  *
1864  * Check if the skb is valid in the XDP case it will be an error pointer.
1865  * Return true in this case to abort processing and advance to next
1866  * descriptor.
1867  *
1868  * Check for corrupted packet headers caused by senders on the local L2
1869  * embedded NIC switch not setting up their Tx Descriptors right.  These
1870  * should be very rare.
1871  *
1872  * Also address the case where we are pulling data in on pages only
1873  * and as such no data is present in the skb header.
1874  *
1875  * In addition if skb is not at least 60 bytes we need to pad it so that
1876  * it is large enough to qualify as a valid Ethernet frame.
1877  *
1878  * Returns true if an error was encountered and skb was freed.
1879  **/
1880 bool ixgbe_cleanup_headers(struct ixgbe_ring *rx_ring,
1881                            union ixgbe_adv_rx_desc *rx_desc,
1882                            struct sk_buff *skb)
1883 {
1884         struct net_device *netdev = rx_ring->netdev;
1885
1886         /* XDP packets use error pointer so abort at this point */
1887         if (IS_ERR(skb))
1888                 return true;
1889
1890         /* Verify netdev is present, and that packet does not have any
1891          * errors that would be unacceptable to the netdev.
1892          */
1893         if (!netdev ||
1894             (unlikely(ixgbe_test_staterr(rx_desc,
1895                                          IXGBE_RXDADV_ERR_FRAME_ERR_MASK) &&
1896              !(netdev->features & NETIF_F_RXALL)))) {
1897                 dev_kfree_skb_any(skb);
1898                 return true;
1899         }
1900
1901         /* place header in linear portion of buffer */
1902         if (!skb_headlen(skb))
1903                 ixgbe_pull_tail(rx_ring, skb);
1904
1905 #ifdef IXGBE_FCOE
1906         /* do not attempt to pad FCoE Frames as this will disrupt DDP */
1907         if (ixgbe_rx_is_fcoe(rx_ring, rx_desc))
1908                 return false;
1909
1910 #endif
1911         /* if eth_skb_pad returns an error the skb was freed */
1912         if (eth_skb_pad(skb))
1913                 return true;
1914
1915         return false;
1916 }
1917
1918 /**
1919  * ixgbe_reuse_rx_page - page flip buffer and store it back on the ring
1920  * @rx_ring: rx descriptor ring to store buffers on
1921  * @old_buff: donor buffer to have page reused
1922  *
1923  * Synchronizes page for reuse by the adapter
1924  **/
1925 static void ixgbe_reuse_rx_page(struct ixgbe_ring *rx_ring,
1926                                 struct ixgbe_rx_buffer *old_buff)
1927 {
1928         struct ixgbe_rx_buffer *new_buff;
1929         u16 nta = rx_ring->next_to_alloc;
1930
1931         new_buff = &rx_ring->rx_buffer_info[nta];
1932
1933         /* update, and store next to alloc */
1934         nta++;
1935         rx_ring->next_to_alloc = (nta < rx_ring->count) ? nta : 0;
1936
1937         /* Transfer page from old buffer to new buffer.
1938          * Move each member individually to avoid possible store
1939          * forwarding stalls and unnecessary copy of skb.
1940          */
1941         new_buff->dma           = old_buff->dma;
1942         new_buff->page          = old_buff->page;
1943         new_buff->page_offset   = old_buff->page_offset;
1944         new_buff->pagecnt_bias  = old_buff->pagecnt_bias;
1945 }
1946
1947 static bool ixgbe_can_reuse_rx_page(struct ixgbe_rx_buffer *rx_buffer,
1948                                     int rx_buffer_pgcnt)
1949 {
1950         unsigned int pagecnt_bias = rx_buffer->pagecnt_bias;
1951         struct page *page = rx_buffer->page;
1952
1953         /* avoid re-using remote and pfmemalloc pages */
1954         if (!dev_page_is_reusable(page))
1955                 return false;
1956
1957 #if (PAGE_SIZE < 8192)
1958         /* if we are only owner of page we can reuse it */
1959         if (unlikely((rx_buffer_pgcnt - pagecnt_bias) > 1))
1960                 return false;
1961 #else
1962         /* The last offset is a bit aggressive in that we assume the
1963          * worst case of FCoE being enabled and using a 3K buffer.
1964          * However this should have minimal impact as the 1K extra is
1965          * still less than one buffer in size.
1966          */
1967 #define IXGBE_LAST_OFFSET \
1968         (SKB_WITH_OVERHEAD(PAGE_SIZE) - IXGBE_RXBUFFER_3K)
1969         if (rx_buffer->page_offset > IXGBE_LAST_OFFSET)
1970                 return false;
1971 #endif
1972
1973         /* If we have drained the page fragment pool we need to update
1974          * the pagecnt_bias and page count so that we fully restock the
1975          * number of references the driver holds.
1976          */
1977         if (unlikely(pagecnt_bias == 1)) {
1978                 page_ref_add(page, USHRT_MAX - 1);
1979                 rx_buffer->pagecnt_bias = USHRT_MAX;
1980         }
1981
1982         return true;
1983 }
1984
1985 /**
1986  * ixgbe_add_rx_frag - Add contents of Rx buffer to sk_buff
1987  * @rx_ring: rx descriptor ring to transact packets on
1988  * @rx_buffer: buffer containing page to add
1989  * @skb: sk_buff to place the data into
1990  * @size: size of data in rx_buffer
1991  *
1992  * This function will add the data contained in rx_buffer->page to the skb.
1993  * This is done either through a direct copy if the data in the buffer is
1994  * less than the skb header size, otherwise it will just attach the page as
1995  * a frag to the skb.
1996  *
1997  * The function will then update the page offset if necessary and return
1998  * true if the buffer can be reused by the adapter.
1999  **/
2000 static void ixgbe_add_rx_frag(struct ixgbe_ring *rx_ring,
2001                               struct ixgbe_rx_buffer *rx_buffer,
2002                               struct sk_buff *skb,
2003                               unsigned int size)
2004 {
2005 #if (PAGE_SIZE < 8192)
2006         unsigned int truesize = ixgbe_rx_pg_size(rx_ring) / 2;
2007 #else
2008         unsigned int truesize = rx_ring->rx_offset ?
2009                                 SKB_DATA_ALIGN(rx_ring->rx_offset + size) :
2010                                 SKB_DATA_ALIGN(size);
2011 #endif
2012         skb_add_rx_frag(skb, skb_shinfo(skb)->nr_frags, rx_buffer->page,
2013                         rx_buffer->page_offset, size, truesize);
2014 #if (PAGE_SIZE < 8192)
2015         rx_buffer->page_offset ^= truesize;
2016 #else
2017         rx_buffer->page_offset += truesize;
2018 #endif
2019 }
2020
2021 static struct ixgbe_rx_buffer *ixgbe_get_rx_buffer(struct ixgbe_ring *rx_ring,
2022                                                    union ixgbe_adv_rx_desc *rx_desc,
2023                                                    struct sk_buff **skb,
2024                                                    const unsigned int size,
2025                                                    int *rx_buffer_pgcnt)
2026 {
2027         struct ixgbe_rx_buffer *rx_buffer;
2028
2029         rx_buffer = &rx_ring->rx_buffer_info[rx_ring->next_to_clean];
2030         *rx_buffer_pgcnt =
2031 #if (PAGE_SIZE < 8192)
2032                 page_count(rx_buffer->page);
2033 #else
2034                 0;
2035 #endif
2036         prefetchw(rx_buffer->page);
2037         *skb = rx_buffer->skb;
2038
2039         /* Delay unmapping of the first packet. It carries the header
2040          * information, HW may still access the header after the writeback.
2041          * Only unmap it when EOP is reached
2042          */
2043         if (!ixgbe_test_staterr(rx_desc, IXGBE_RXD_STAT_EOP)) {
2044                 if (!*skb)
2045                         goto skip_sync;
2046         } else {
2047                 if (*skb)
2048                         ixgbe_dma_sync_frag(rx_ring, *skb);
2049         }
2050
2051         /* we are reusing so sync this buffer for CPU use */
2052         dma_sync_single_range_for_cpu(rx_ring->dev,
2053                                       rx_buffer->dma,
2054                                       rx_buffer->page_offset,
2055                                       size,
2056                                       DMA_FROM_DEVICE);
2057 skip_sync:
2058         rx_buffer->pagecnt_bias--;
2059
2060         return rx_buffer;
2061 }
2062
2063 static void ixgbe_put_rx_buffer(struct ixgbe_ring *rx_ring,
2064                                 struct ixgbe_rx_buffer *rx_buffer,
2065                                 struct sk_buff *skb,
2066                                 int rx_buffer_pgcnt)
2067 {
2068         if (ixgbe_can_reuse_rx_page(rx_buffer, rx_buffer_pgcnt)) {
2069                 /* hand second half of page back to the ring */
2070                 ixgbe_reuse_rx_page(rx_ring, rx_buffer);
2071         } else {
2072                 if (!IS_ERR(skb) && IXGBE_CB(skb)->dma == rx_buffer->dma) {
2073                         /* the page has been released from the ring */
2074                         IXGBE_CB(skb)->page_released = true;
2075                 } else {
2076                         /* we are not reusing the buffer so unmap it */
2077                         dma_unmap_page_attrs(rx_ring->dev, rx_buffer->dma,
2078                                              ixgbe_rx_pg_size(rx_ring),
2079                                              DMA_FROM_DEVICE,
2080                                              IXGBE_RX_DMA_ATTR);
2081                 }
2082                 __page_frag_cache_drain(rx_buffer->page,
2083                                         rx_buffer->pagecnt_bias);
2084         }
2085
2086         /* clear contents of rx_buffer */
2087         rx_buffer->page = NULL;
2088         rx_buffer->skb = NULL;
2089 }
2090
2091 static struct sk_buff *ixgbe_construct_skb(struct ixgbe_ring *rx_ring,
2092                                            struct ixgbe_rx_buffer *rx_buffer,
2093                                            struct xdp_buff *xdp,
2094                                            union ixgbe_adv_rx_desc *rx_desc)
2095 {
2096         unsigned int size = xdp->data_end - xdp->data;
2097 #if (PAGE_SIZE < 8192)
2098         unsigned int truesize = ixgbe_rx_pg_size(rx_ring) / 2;
2099 #else
2100         unsigned int truesize = SKB_DATA_ALIGN(xdp->data_end -
2101                                                xdp->data_hard_start);
2102 #endif
2103         struct sk_buff *skb;
2104
2105         /* prefetch first cache line of first page */
2106         net_prefetch(xdp->data);
2107
2108         /* Note, we get here by enabling legacy-rx via:
2109          *
2110          *    ethtool --set-priv-flags <dev> legacy-rx on
2111          *
2112          * In this mode, we currently get 0 extra XDP headroom as
2113          * opposed to having legacy-rx off, where we process XDP
2114          * packets going to stack via ixgbe_build_skb(). The latter
2115          * provides us currently with 192 bytes of headroom.
2116          *
2117          * For ixgbe_construct_skb() mode it means that the
2118          * xdp->data_meta will always point to xdp->data, since
2119          * the helper cannot expand the head. Should this ever
2120          * change in future for legacy-rx mode on, then lets also
2121          * add xdp->data_meta handling here.
2122          */
2123
2124         /* allocate a skb to store the frags */
2125         skb = napi_alloc_skb(&rx_ring->q_vector->napi, IXGBE_RX_HDR_SIZE);
2126         if (unlikely(!skb))
2127                 return NULL;
2128
2129         if (size > IXGBE_RX_HDR_SIZE) {
2130                 if (!ixgbe_test_staterr(rx_desc, IXGBE_RXD_STAT_EOP))
2131                         IXGBE_CB(skb)->dma = rx_buffer->dma;
2132
2133                 skb_add_rx_frag(skb, 0, rx_buffer->page,
2134                                 xdp->data - page_address(rx_buffer->page),
2135                                 size, truesize);
2136 #if (PAGE_SIZE < 8192)
2137                 rx_buffer->page_offset ^= truesize;
2138 #else
2139                 rx_buffer->page_offset += truesize;
2140 #endif
2141         } else {
2142                 memcpy(__skb_put(skb, size),
2143                        xdp->data, ALIGN(size, sizeof(long)));
2144                 rx_buffer->pagecnt_bias++;
2145         }
2146
2147         return skb;
2148 }
2149
2150 static struct sk_buff *ixgbe_build_skb(struct ixgbe_ring *rx_ring,
2151                                        struct ixgbe_rx_buffer *rx_buffer,
2152                                        struct xdp_buff *xdp,
2153                                        union ixgbe_adv_rx_desc *rx_desc)
2154 {
2155         unsigned int metasize = xdp->data - xdp->data_meta;
2156 #if (PAGE_SIZE < 8192)
2157         unsigned int truesize = ixgbe_rx_pg_size(rx_ring) / 2;
2158 #else
2159         unsigned int truesize = SKB_DATA_ALIGN(sizeof(struct skb_shared_info)) +
2160                                 SKB_DATA_ALIGN(xdp->data_end -
2161                                                xdp->data_hard_start);
2162 #endif
2163         struct sk_buff *skb;
2164
2165         /* Prefetch first cache line of first page. If xdp->data_meta
2166          * is unused, this points extactly as xdp->data, otherwise we
2167          * likely have a consumer accessing first few bytes of meta
2168          * data, and then actual data.
2169          */
2170         net_prefetch(xdp->data_meta);
2171
2172         /* build an skb to around the page buffer */
2173         skb = napi_build_skb(xdp->data_hard_start, truesize);
2174         if (unlikely(!skb))
2175                 return NULL;
2176
2177         /* update pointers within the skb to store the data */
2178         skb_reserve(skb, xdp->data - xdp->data_hard_start);
2179         __skb_put(skb, xdp->data_end - xdp->data);
2180         if (metasize)
2181                 skb_metadata_set(skb, metasize);
2182
2183         /* record DMA address if this is the start of a chain of buffers */
2184         if (!ixgbe_test_staterr(rx_desc, IXGBE_RXD_STAT_EOP))
2185                 IXGBE_CB(skb)->dma = rx_buffer->dma;
2186
2187         /* update buffer offset */
2188 #if (PAGE_SIZE < 8192)
2189         rx_buffer->page_offset ^= truesize;
2190 #else
2191         rx_buffer->page_offset += truesize;
2192 #endif
2193
2194         return skb;
2195 }
2196
2197 static struct sk_buff *ixgbe_run_xdp(struct ixgbe_adapter *adapter,
2198                                      struct ixgbe_ring *rx_ring,
2199                                      struct xdp_buff *xdp)
2200 {
2201         int err, result = IXGBE_XDP_PASS;
2202         struct bpf_prog *xdp_prog;
2203         struct ixgbe_ring *ring;
2204         struct xdp_frame *xdpf;
2205         u32 act;
2206
2207         xdp_prog = READ_ONCE(rx_ring->xdp_prog);
2208
2209         if (!xdp_prog)
2210                 goto xdp_out;
2211
2212         prefetchw(xdp->data_hard_start); /* xdp_frame write */
2213
2214         act = bpf_prog_run_xdp(xdp_prog, xdp);
2215         switch (act) {
2216         case XDP_PASS:
2217                 break;
2218         case XDP_TX:
2219                 xdpf = xdp_convert_buff_to_frame(xdp);
2220                 if (unlikely(!xdpf))
2221                         goto out_failure;
2222                 ring = ixgbe_determine_xdp_ring(adapter);
2223                 if (static_branch_unlikely(&ixgbe_xdp_locking_key))
2224                         spin_lock(&ring->tx_lock);
2225                 result = ixgbe_xmit_xdp_ring(ring, xdpf);
2226                 if (static_branch_unlikely(&ixgbe_xdp_locking_key))
2227                         spin_unlock(&ring->tx_lock);
2228                 if (result == IXGBE_XDP_CONSUMED)
2229                         goto out_failure;
2230                 break;
2231         case XDP_REDIRECT:
2232                 err = xdp_do_redirect(adapter->netdev, xdp, xdp_prog);
2233                 if (err)
2234                         goto out_failure;
2235                 result = IXGBE_XDP_REDIR;
2236                 break;
2237         default:
2238                 bpf_warn_invalid_xdp_action(rx_ring->netdev, xdp_prog, act);
2239                 fallthrough;
2240         case XDP_ABORTED:
2241 out_failure:
2242                 trace_xdp_exception(rx_ring->netdev, xdp_prog, act);
2243                 fallthrough; /* handle aborts by dropping packet */
2244         case XDP_DROP:
2245                 result = IXGBE_XDP_CONSUMED;
2246                 break;
2247         }
2248 xdp_out:
2249         return ERR_PTR(-result);
2250 }
2251
2252 static unsigned int ixgbe_rx_frame_truesize(struct ixgbe_ring *rx_ring,
2253                                             unsigned int size)
2254 {
2255         unsigned int truesize;
2256
2257 #if (PAGE_SIZE < 8192)
2258         truesize = ixgbe_rx_pg_size(rx_ring) / 2; /* Must be power-of-2 */
2259 #else
2260         truesize = rx_ring->rx_offset ?
2261                 SKB_DATA_ALIGN(rx_ring->rx_offset + size) +
2262                 SKB_DATA_ALIGN(sizeof(struct skb_shared_info)) :
2263                 SKB_DATA_ALIGN(size);
2264 #endif
2265         return truesize;
2266 }
2267
2268 static void ixgbe_rx_buffer_flip(struct ixgbe_ring *rx_ring,
2269                                  struct ixgbe_rx_buffer *rx_buffer,
2270                                  unsigned int size)
2271 {
2272         unsigned int truesize = ixgbe_rx_frame_truesize(rx_ring, size);
2273 #if (PAGE_SIZE < 8192)
2274         rx_buffer->page_offset ^= truesize;
2275 #else
2276         rx_buffer->page_offset += truesize;
2277 #endif
2278 }
2279
2280 /**
2281  * ixgbe_clean_rx_irq - Clean completed descriptors from Rx ring - bounce buf
2282  * @q_vector: structure containing interrupt and ring information
2283  * @rx_ring: rx descriptor ring to transact packets on
2284  * @budget: Total limit on number of packets to process
2285  *
2286  * This function provides a "bounce buffer" approach to Rx interrupt
2287  * processing.  The advantage to this is that on systems that have
2288  * expensive overhead for IOMMU access this provides a means of avoiding
2289  * it by maintaining the mapping of the page to the syste.
2290  *
2291  * Returns amount of work completed
2292  **/
2293 static int ixgbe_clean_rx_irq(struct ixgbe_q_vector *q_vector,
2294                                struct ixgbe_ring *rx_ring,
2295                                const int budget)
2296 {
2297         unsigned int total_rx_bytes = 0, total_rx_packets = 0, frame_sz = 0;
2298         struct ixgbe_adapter *adapter = q_vector->adapter;
2299 #ifdef IXGBE_FCOE
2300         int ddp_bytes;
2301         unsigned int mss = 0;
2302 #endif /* IXGBE_FCOE */
2303         u16 cleaned_count = ixgbe_desc_unused(rx_ring);
2304         unsigned int offset = rx_ring->rx_offset;
2305         unsigned int xdp_xmit = 0;
2306         struct xdp_buff xdp;
2307
2308         /* Frame size depend on rx_ring setup when PAGE_SIZE=4K */
2309 #if (PAGE_SIZE < 8192)
2310         frame_sz = ixgbe_rx_frame_truesize(rx_ring, 0);
2311 #endif
2312         xdp_init_buff(&xdp, frame_sz, &rx_ring->xdp_rxq);
2313
2314         while (likely(total_rx_packets < budget)) {
2315                 union ixgbe_adv_rx_desc *rx_desc;
2316                 struct ixgbe_rx_buffer *rx_buffer;
2317                 struct sk_buff *skb;
2318                 int rx_buffer_pgcnt;
2319                 unsigned int size;
2320
2321                 /* return some buffers to hardware, one at a time is too slow */
2322                 if (cleaned_count >= IXGBE_RX_BUFFER_WRITE) {
2323                         ixgbe_alloc_rx_buffers(rx_ring, cleaned_count);
2324                         cleaned_count = 0;
2325                 }
2326
2327                 rx_desc = IXGBE_RX_DESC(rx_ring, rx_ring->next_to_clean);
2328                 size = le16_to_cpu(rx_desc->wb.upper.length);
2329                 if (!size)
2330                         break;
2331
2332                 /* This memory barrier is needed to keep us from reading
2333                  * any other fields out of the rx_desc until we know the
2334                  * descriptor has been written back
2335                  */
2336                 dma_rmb();
2337
2338                 rx_buffer = ixgbe_get_rx_buffer(rx_ring, rx_desc, &skb, size, &rx_buffer_pgcnt);
2339
2340                 /* retrieve a buffer from the ring */
2341                 if (!skb) {
2342                         unsigned char *hard_start;
2343
2344                         hard_start = page_address(rx_buffer->page) +
2345                                      rx_buffer->page_offset - offset;
2346                         xdp_prepare_buff(&xdp, hard_start, offset, size, true);
2347                         xdp_buff_clear_frags_flag(&xdp);
2348 #if (PAGE_SIZE > 4096)
2349                         /* At larger PAGE_SIZE, frame_sz depend on len size */
2350                         xdp.frame_sz = ixgbe_rx_frame_truesize(rx_ring, size);
2351 #endif
2352                         skb = ixgbe_run_xdp(adapter, rx_ring, &xdp);
2353                 }
2354
2355                 if (IS_ERR(skb)) {
2356                         unsigned int xdp_res = -PTR_ERR(skb);
2357
2358                         if (xdp_res & (IXGBE_XDP_TX | IXGBE_XDP_REDIR)) {
2359                                 xdp_xmit |= xdp_res;
2360                                 ixgbe_rx_buffer_flip(rx_ring, rx_buffer, size);
2361                         } else {
2362                                 rx_buffer->pagecnt_bias++;
2363                         }
2364                         total_rx_packets++;
2365                         total_rx_bytes += size;
2366                 } else if (skb) {
2367                         ixgbe_add_rx_frag(rx_ring, rx_buffer, skb, size);
2368                 } else if (ring_uses_build_skb(rx_ring)) {
2369                         skb = ixgbe_build_skb(rx_ring, rx_buffer,
2370                                               &xdp, rx_desc);
2371                 } else {
2372                         skb = ixgbe_construct_skb(rx_ring, rx_buffer,
2373                                                   &xdp, rx_desc);
2374                 }
2375
2376                 /* exit if we failed to retrieve a buffer */
2377                 if (!skb) {
2378                         rx_ring->rx_stats.alloc_rx_buff_failed++;
2379                         rx_buffer->pagecnt_bias++;
2380                         break;
2381                 }
2382
2383                 ixgbe_put_rx_buffer(rx_ring, rx_buffer, skb, rx_buffer_pgcnt);
2384                 cleaned_count++;
2385
2386                 /* place incomplete frames back on ring for completion */
2387                 if (ixgbe_is_non_eop(rx_ring, rx_desc, skb))
2388                         continue;
2389
2390                 /* verify the packet layout is correct */
2391                 if (ixgbe_cleanup_headers(rx_ring, rx_desc, skb))
2392                         continue;
2393
2394                 /* probably a little skewed due to removing CRC */
2395                 total_rx_bytes += skb->len;
2396
2397                 /* populate checksum, timestamp, VLAN, and protocol */
2398                 ixgbe_process_skb_fields(rx_ring, rx_desc, skb);
2399
2400 #ifdef IXGBE_FCOE
2401                 /* if ddp, not passing to ULD unless for FCP_RSP or error */
2402                 if (ixgbe_rx_is_fcoe(rx_ring, rx_desc)) {
2403                         ddp_bytes = ixgbe_fcoe_ddp(adapter, rx_desc, skb);
2404                         /* include DDPed FCoE data */
2405                         if (ddp_bytes > 0) {
2406                                 if (!mss) {
2407                                         mss = rx_ring->netdev->mtu -
2408                                                 sizeof(struct fcoe_hdr) -
2409                                                 sizeof(struct fc_frame_header) -
2410                                                 sizeof(struct fcoe_crc_eof);
2411                                         if (mss > 512)
2412                                                 mss &= ~511;
2413                                 }
2414                                 total_rx_bytes += ddp_bytes;
2415                                 total_rx_packets += DIV_ROUND_UP(ddp_bytes,
2416                                                                  mss);
2417                         }
2418                         if (!ddp_bytes) {
2419                                 dev_kfree_skb_any(skb);
2420                                 continue;
2421                         }
2422                 }
2423
2424 #endif /* IXGBE_FCOE */
2425                 ixgbe_rx_skb(q_vector, skb);
2426
2427                 /* update budget accounting */
2428                 total_rx_packets++;
2429         }
2430
2431         if (xdp_xmit & IXGBE_XDP_REDIR)
2432                 xdp_do_flush_map();
2433
2434         if (xdp_xmit & IXGBE_XDP_TX) {
2435                 struct ixgbe_ring *ring = ixgbe_determine_xdp_ring(adapter);
2436
2437                 ixgbe_xdp_ring_update_tail_locked(ring);
2438         }
2439
2440         u64_stats_update_begin(&rx_ring->syncp);
2441         rx_ring->stats.packets += total_rx_packets;
2442         rx_ring->stats.bytes += total_rx_bytes;
2443         u64_stats_update_end(&rx_ring->syncp);
2444         q_vector->rx.total_packets += total_rx_packets;
2445         q_vector->rx.total_bytes += total_rx_bytes;
2446
2447         return total_rx_packets;
2448 }
2449
2450 /**
2451  * ixgbe_configure_msix - Configure MSI-X hardware
2452  * @adapter: board private structure
2453  *
2454  * ixgbe_configure_msix sets up the hardware to properly generate MSI-X
2455  * interrupts.
2456  **/
2457 static void ixgbe_configure_msix(struct ixgbe_adapter *adapter)
2458 {
2459         struct ixgbe_q_vector *q_vector;
2460         int v_idx;
2461         u32 mask;
2462
2463         /* Populate MSIX to EITR Select */
2464         if (adapter->num_vfs > 32) {
2465                 u32 eitrsel = BIT(adapter->num_vfs - 32) - 1;
2466                 IXGBE_WRITE_REG(&adapter->hw, IXGBE_EITRSEL, eitrsel);
2467         }
2468
2469         /*
2470          * Populate the IVAR table and set the ITR values to the
2471          * corresponding register.
2472          */
2473         for (v_idx = 0; v_idx < adapter->num_q_vectors; v_idx++) {
2474                 struct ixgbe_ring *ring;
2475                 q_vector = adapter->q_vector[v_idx];
2476
2477                 ixgbe_for_each_ring(ring, q_vector->rx)
2478                         ixgbe_set_ivar(adapter, 0, ring->reg_idx, v_idx);
2479
2480                 ixgbe_for_each_ring(ring, q_vector->tx)
2481                         ixgbe_set_ivar(adapter, 1, ring->reg_idx, v_idx);
2482
2483                 ixgbe_write_eitr(q_vector);
2484         }
2485
2486         switch (adapter->hw.mac.type) {
2487         case ixgbe_mac_82598EB:
2488                 ixgbe_set_ivar(adapter, -1, IXGBE_IVAR_OTHER_CAUSES_INDEX,
2489                                v_idx);
2490                 break;
2491         case ixgbe_mac_82599EB:
2492         case ixgbe_mac_X540:
2493         case ixgbe_mac_X550:
2494         case ixgbe_mac_X550EM_x:
2495         case ixgbe_mac_x550em_a:
2496                 ixgbe_set_ivar(adapter, -1, 1, v_idx);
2497                 break;
2498         default:
2499                 break;
2500         }
2501         IXGBE_WRITE_REG(&adapter->hw, IXGBE_EITR(v_idx), 1950);
2502
2503         /* set up to autoclear timer, and the vectors */
2504         mask = IXGBE_EIMS_ENABLE_MASK;
2505         mask &= ~(IXGBE_EIMS_OTHER |
2506                   IXGBE_EIMS_MAILBOX |
2507                   IXGBE_EIMS_LSC);
2508
2509         IXGBE_WRITE_REG(&adapter->hw, IXGBE_EIAC, mask);
2510 }
2511
2512 /**
2513  * ixgbe_update_itr - update the dynamic ITR value based on statistics
2514  * @q_vector: structure containing interrupt and ring information
2515  * @ring_container: structure containing ring performance data
2516  *
2517  *      Stores a new ITR value based on packets and byte
2518  *      counts during the last interrupt.  The advantage of per interrupt
2519  *      computation is faster updates and more accurate ITR for the current
2520  *      traffic pattern.  Constants in this function were computed
2521  *      based on theoretical maximum wire speed and thresholds were set based
2522  *      on testing data as well as attempting to minimize response time
2523  *      while increasing bulk throughput.
2524  **/
2525 static void ixgbe_update_itr(struct ixgbe_q_vector *q_vector,
2526                              struct ixgbe_ring_container *ring_container)
2527 {
2528         unsigned int itr = IXGBE_ITR_ADAPTIVE_MIN_USECS |
2529                            IXGBE_ITR_ADAPTIVE_LATENCY;
2530         unsigned int avg_wire_size, packets, bytes;
2531         unsigned long next_update = jiffies;
2532
2533         /* If we don't have any rings just leave ourselves set for maximum
2534          * possible latency so we take ourselves out of the equation.
2535          */
2536         if (!ring_container->ring)
2537                 return;
2538
2539         /* If we didn't update within up to 1 - 2 jiffies we can assume
2540          * that either packets are coming in so slow there hasn't been
2541          * any work, or that there is so much work that NAPI is dealing
2542          * with interrupt moderation and we don't need to do anything.
2543          */
2544         if (time_after(next_update, ring_container->next_update))
2545                 goto clear_counts;
2546
2547         packets = ring_container->total_packets;
2548
2549         /* We have no packets to actually measure against. This means
2550          * either one of the other queues on this vector is active or
2551          * we are a Tx queue doing TSO with too high of an interrupt rate.
2552          *
2553          * When this occurs just tick up our delay by the minimum value
2554          * and hope that this extra delay will prevent us from being called
2555          * without any work on our queue.
2556          */
2557         if (!packets) {
2558                 itr = (q_vector->itr >> 2) + IXGBE_ITR_ADAPTIVE_MIN_INC;
2559                 if (itr > IXGBE_ITR_ADAPTIVE_MAX_USECS)
2560                         itr = IXGBE_ITR_ADAPTIVE_MAX_USECS;
2561                 itr += ring_container->itr & IXGBE_ITR_ADAPTIVE_LATENCY;
2562                 goto clear_counts;
2563         }
2564
2565         bytes = ring_container->total_bytes;
2566
2567         /* If packets are less than 4 or bytes are less than 9000 assume
2568          * insufficient data to use bulk rate limiting approach. We are
2569          * likely latency driven.
2570          */
2571         if (packets < 4 && bytes < 9000) {
2572                 itr = IXGBE_ITR_ADAPTIVE_LATENCY;
2573                 goto adjust_by_size;
2574         }
2575
2576         /* Between 4 and 48 we can assume that our current interrupt delay
2577          * is only slightly too low. As such we should increase it by a small
2578          * fixed amount.
2579          */
2580         if (packets < 48) {
2581                 itr = (q_vector->itr >> 2) + IXGBE_ITR_ADAPTIVE_MIN_INC;
2582                 if (itr > IXGBE_ITR_ADAPTIVE_MAX_USECS)
2583                         itr = IXGBE_ITR_ADAPTIVE_MAX_USECS;
2584                 goto clear_counts;
2585         }
2586
2587         /* Between 48 and 96 is our "goldilocks" zone where we are working
2588          * out "just right". Just report that our current ITR is good for us.
2589          */
2590         if (packets < 96) {
2591                 itr = q_vector->itr >> 2;
2592                 goto clear_counts;
2593         }
2594
2595         /* If packet count is 96 or greater we are likely looking at a slight
2596          * overrun of the delay we want. Try halving our delay to see if that
2597          * will cut the number of packets in half per interrupt.
2598          */
2599         if (packets < 256) {
2600                 itr = q_vector->itr >> 3;
2601                 if (itr < IXGBE_ITR_ADAPTIVE_MIN_USECS)
2602                         itr = IXGBE_ITR_ADAPTIVE_MIN_USECS;
2603                 goto clear_counts;
2604         }
2605
2606         /* The paths below assume we are dealing with a bulk ITR since number
2607          * of packets is 256 or greater. We are just going to have to compute
2608          * a value and try to bring the count under control, though for smaller
2609          * packet sizes there isn't much we can do as NAPI polling will likely
2610          * be kicking in sooner rather than later.
2611          */
2612         itr = IXGBE_ITR_ADAPTIVE_BULK;
2613
2614 adjust_by_size:
2615         /* If packet counts are 256 or greater we can assume we have a gross
2616          * overestimation of what the rate should be. Instead of trying to fine
2617          * tune it just use the formula below to try and dial in an exact value
2618          * give the current packet size of the frame.
2619          */
2620         avg_wire_size = bytes / packets;
2621
2622         /* The following is a crude approximation of:
2623          *  wmem_default / (size + overhead) = desired_pkts_per_int
2624          *  rate / bits_per_byte / (size + ethernet overhead) = pkt_rate
2625          *  (desired_pkt_rate / pkt_rate) * usecs_per_sec = ITR value
2626          *
2627          * Assuming wmem_default is 212992 and overhead is 640 bytes per
2628          * packet, (256 skb, 64 headroom, 320 shared info), we can reduce the
2629          * formula down to
2630          *
2631          *  (170 * (size + 24)) / (size + 640) = ITR
2632          *
2633          * We first do some math on the packet size and then finally bitshift
2634          * by 8 after rounding up. We also have to account for PCIe link speed
2635          * difference as ITR scales based on this.
2636          */
2637         if (avg_wire_size <= 60) {
2638                 /* Start at 50k ints/sec */
2639                 avg_wire_size = 5120;
2640         } else if (avg_wire_size <= 316) {
2641                 /* 50K ints/sec to 16K ints/sec */
2642                 avg_wire_size *= 40;
2643                 avg_wire_size += 2720;
2644         } else if (avg_wire_size <= 1084) {
2645                 /* 16K ints/sec to 9.2K ints/sec */
2646                 avg_wire_size *= 15;
2647                 avg_wire_size += 11452;
2648         } else if (avg_wire_size < 1968) {
2649                 /* 9.2K ints/sec to 8K ints/sec */
2650                 avg_wire_size *= 5;
2651                 avg_wire_size += 22420;
2652         } else {
2653                 /* plateau at a limit of 8K ints/sec */
2654                 avg_wire_size = 32256;
2655         }
2656
2657         /* If we are in low latency mode half our delay which doubles the rate
2658          * to somewhere between 100K to 16K ints/sec
2659          */
2660         if (itr & IXGBE_ITR_ADAPTIVE_LATENCY)
2661                 avg_wire_size >>= 1;
2662
2663         /* Resultant value is 256 times larger than it needs to be. This
2664          * gives us room to adjust the value as needed to either increase
2665          * or decrease the value based on link speeds of 10G, 2.5G, 1G, etc.
2666          *
2667          * Use addition as we have already recorded the new latency flag
2668          * for the ITR value.
2669          */
2670         switch (q_vector->adapter->link_speed) {
2671         case IXGBE_LINK_SPEED_10GB_FULL:
2672         case IXGBE_LINK_SPEED_100_FULL:
2673         default:
2674                 itr += DIV_ROUND_UP(avg_wire_size,
2675                                     IXGBE_ITR_ADAPTIVE_MIN_INC * 256) *
2676                        IXGBE_ITR_ADAPTIVE_MIN_INC;
2677                 break;
2678         case IXGBE_LINK_SPEED_2_5GB_FULL:
2679         case IXGBE_LINK_SPEED_1GB_FULL:
2680         case IXGBE_LINK_SPEED_10_FULL:
2681                 if (avg_wire_size > 8064)
2682                         avg_wire_size = 8064;
2683                 itr += DIV_ROUND_UP(avg_wire_size,
2684                                     IXGBE_ITR_ADAPTIVE_MIN_INC * 64) *
2685                        IXGBE_ITR_ADAPTIVE_MIN_INC;
2686                 break;
2687         }
2688
2689 clear_counts:
2690         /* write back value */
2691         ring_container->itr = itr;
2692
2693         /* next update should occur within next jiffy */
2694         ring_container->next_update = next_update + 1;
2695
2696         ring_container->total_bytes = 0;
2697         ring_container->total_packets = 0;
2698 }
2699
2700 /**
2701  * ixgbe_write_eitr - write EITR register in hardware specific way
2702  * @q_vector: structure containing interrupt and ring information
2703  *
2704  * This function is made to be called by ethtool and by the driver
2705  * when it needs to update EITR registers at runtime.  Hardware
2706  * specific quirks/differences are taken care of here.
2707  */
2708 void ixgbe_write_eitr(struct ixgbe_q_vector *q_vector)
2709 {
2710         struct ixgbe_adapter *adapter = q_vector->adapter;
2711         struct ixgbe_hw *hw = &adapter->hw;
2712         int v_idx = q_vector->v_idx;
2713         u32 itr_reg = q_vector->itr & IXGBE_MAX_EITR;
2714
2715         switch (adapter->hw.mac.type) {
2716         case ixgbe_mac_82598EB:
2717                 /* must write high and low 16 bits to reset counter */
2718                 itr_reg |= (itr_reg << 16);
2719                 break;
2720         case ixgbe_mac_82599EB:
2721         case ixgbe_mac_X540:
2722         case ixgbe_mac_X550:
2723         case ixgbe_mac_X550EM_x:
2724         case ixgbe_mac_x550em_a:
2725                 /*
2726                  * set the WDIS bit to not clear the timer bits and cause an
2727                  * immediate assertion of the interrupt
2728                  */
2729                 itr_reg |= IXGBE_EITR_CNT_WDIS;
2730                 break;
2731         default:
2732                 break;
2733         }
2734         IXGBE_WRITE_REG(hw, IXGBE_EITR(v_idx), itr_reg);
2735 }
2736
2737 static void ixgbe_set_itr(struct ixgbe_q_vector *q_vector)
2738 {
2739         u32 new_itr;
2740
2741         ixgbe_update_itr(q_vector, &q_vector->tx);
2742         ixgbe_update_itr(q_vector, &q_vector->rx);
2743
2744         /* use the smallest value of new ITR delay calculations */
2745         new_itr = min(q_vector->rx.itr, q_vector->tx.itr);
2746
2747         /* Clear latency flag if set, shift into correct position */
2748         new_itr &= ~IXGBE_ITR_ADAPTIVE_LATENCY;
2749         new_itr <<= 2;
2750
2751         if (new_itr != q_vector->itr) {
2752                 /* save the algorithm value here */
2753                 q_vector->itr = new_itr;
2754
2755                 ixgbe_write_eitr(q_vector);
2756         }
2757 }
2758
2759 /**
2760  * ixgbe_check_overtemp_subtask - check for over temperature
2761  * @adapter: pointer to adapter
2762  **/
2763 static void ixgbe_check_overtemp_subtask(struct ixgbe_adapter *adapter)
2764 {
2765         struct ixgbe_hw *hw = &adapter->hw;
2766         u32 eicr = adapter->interrupt_event;
2767         s32 rc;
2768
2769         if (test_bit(__IXGBE_DOWN, &adapter->state))
2770                 return;
2771
2772         if (!(adapter->flags2 & IXGBE_FLAG2_TEMP_SENSOR_EVENT))
2773                 return;
2774
2775         adapter->flags2 &= ~IXGBE_FLAG2_TEMP_SENSOR_EVENT;
2776
2777         switch (hw->device_id) {
2778         case IXGBE_DEV_ID_82599_T3_LOM:
2779                 /*
2780                  * Since the warning interrupt is for both ports
2781                  * we don't have to check if:
2782                  *  - This interrupt wasn't for our port.
2783                  *  - We may have missed the interrupt so always have to
2784                  *    check if we  got a LSC
2785                  */
2786                 if (!(eicr & IXGBE_EICR_GPI_SDP0_8259X) &&
2787                     !(eicr & IXGBE_EICR_LSC))
2788                         return;
2789
2790                 if (!(eicr & IXGBE_EICR_LSC) && hw->mac.ops.check_link) {
2791                         u32 speed;
2792                         bool link_up = false;
2793
2794                         hw->mac.ops.check_link(hw, &speed, &link_up, false);
2795
2796                         if (link_up)
2797                                 return;
2798                 }
2799
2800                 /* Check if this is not due to overtemp */
2801                 if (hw->phy.ops.check_overtemp(hw) != IXGBE_ERR_OVERTEMP)
2802                         return;
2803
2804                 break;
2805         case IXGBE_DEV_ID_X550EM_A_1G_T:
2806         case IXGBE_DEV_ID_X550EM_A_1G_T_L:
2807                 rc = hw->phy.ops.check_overtemp(hw);
2808                 if (rc != IXGBE_ERR_OVERTEMP)
2809                         return;
2810                 break;
2811         default:
2812                 if (adapter->hw.mac.type >= ixgbe_mac_X540)
2813                         return;
2814                 if (!(eicr & IXGBE_EICR_GPI_SDP0(hw)))
2815                         return;
2816                 break;
2817         }
2818         e_crit(drv, "%s\n", ixgbe_overheat_msg);
2819
2820         adapter->interrupt_event = 0;
2821 }
2822
2823 static void ixgbe_check_fan_failure(struct ixgbe_adapter *adapter, u32 eicr)
2824 {
2825         struct ixgbe_hw *hw = &adapter->hw;
2826
2827         if ((adapter->flags & IXGBE_FLAG_FAN_FAIL_CAPABLE) &&
2828             (eicr & IXGBE_EICR_GPI_SDP1(hw))) {
2829                 e_crit(probe, "Fan has stopped, replace the adapter\n");
2830                 /* write to clear the interrupt */
2831                 IXGBE_WRITE_REG(hw, IXGBE_EICR, IXGBE_EICR_GPI_SDP1(hw));
2832         }
2833 }
2834
2835 static void ixgbe_check_overtemp_event(struct ixgbe_adapter *adapter, u32 eicr)
2836 {
2837         struct ixgbe_hw *hw = &adapter->hw;
2838
2839         if (!(adapter->flags2 & IXGBE_FLAG2_TEMP_SENSOR_CAPABLE))
2840                 return;
2841
2842         switch (adapter->hw.mac.type) {
2843         case ixgbe_mac_82599EB:
2844                 /*
2845                  * Need to check link state so complete overtemp check
2846                  * on service task
2847                  */
2848                 if (((eicr & IXGBE_EICR_GPI_SDP0(hw)) ||
2849                      (eicr & IXGBE_EICR_LSC)) &&
2850                     (!test_bit(__IXGBE_DOWN, &adapter->state))) {
2851                         adapter->interrupt_event = eicr;
2852                         adapter->flags2 |= IXGBE_FLAG2_TEMP_SENSOR_EVENT;
2853                         ixgbe_service_event_schedule(adapter);
2854                         return;
2855                 }
2856                 return;
2857         case ixgbe_mac_x550em_a:
2858                 if (eicr & IXGBE_EICR_GPI_SDP0_X550EM_a) {
2859                         adapter->interrupt_event = eicr;
2860                         adapter->flags2 |= IXGBE_FLAG2_TEMP_SENSOR_EVENT;
2861                         ixgbe_service_event_schedule(adapter);
2862                         IXGBE_WRITE_REG(&adapter->hw, IXGBE_EIMC,
2863                                         IXGBE_EICR_GPI_SDP0_X550EM_a);
2864                         IXGBE_WRITE_REG(&adapter->hw, IXGBE_EICR,
2865                                         IXGBE_EICR_GPI_SDP0_X550EM_a);
2866                 }
2867                 return;
2868         case ixgbe_mac_X550:
2869         case ixgbe_mac_X540:
2870                 if (!(eicr & IXGBE_EICR_TS))
2871                         return;
2872                 break;
2873         default:
2874                 return;
2875         }
2876
2877         e_crit(drv, "%s\n", ixgbe_overheat_msg);
2878 }
2879
2880 static inline bool ixgbe_is_sfp(struct ixgbe_hw *hw)
2881 {
2882         switch (hw->mac.type) {
2883         case ixgbe_mac_82598EB:
2884                 if (hw->phy.type == ixgbe_phy_nl)
2885                         return true;
2886                 return false;
2887         case ixgbe_mac_82599EB:
2888         case ixgbe_mac_X550EM_x:
2889         case ixgbe_mac_x550em_a:
2890                 switch (hw->mac.ops.get_media_type(hw)) {
2891                 case ixgbe_media_type_fiber:
2892                 case ixgbe_media_type_fiber_qsfp:
2893                         return true;
2894                 default:
2895                         return false;
2896                 }
2897         default:
2898                 return false;
2899         }
2900 }
2901
2902 static void ixgbe_check_sfp_event(struct ixgbe_adapter *adapter, u32 eicr)
2903 {
2904         struct ixgbe_hw *hw = &adapter->hw;
2905         u32 eicr_mask = IXGBE_EICR_GPI_SDP2(hw);
2906
2907         if (!ixgbe_is_sfp(hw))
2908                 return;
2909
2910         /* Later MAC's use different SDP */
2911         if (hw->mac.type >= ixgbe_mac_X540)
2912                 eicr_mask = IXGBE_EICR_GPI_SDP0_X540;
2913
2914         if (eicr & eicr_mask) {
2915                 /* Clear the interrupt */
2916                 IXGBE_WRITE_REG(hw, IXGBE_EICR, eicr_mask);
2917                 if (!test_bit(__IXGBE_DOWN, &adapter->state)) {
2918                         adapter->flags2 |= IXGBE_FLAG2_SFP_NEEDS_RESET;
2919                         adapter->sfp_poll_time = 0;
2920                         ixgbe_service_event_schedule(adapter);
2921                 }
2922         }
2923
2924         if (adapter->hw.mac.type == ixgbe_mac_82599EB &&
2925             (eicr & IXGBE_EICR_GPI_SDP1(hw))) {
2926                 /* Clear the interrupt */
2927                 IXGBE_WRITE_REG(hw, IXGBE_EICR, IXGBE_EICR_GPI_SDP1(hw));
2928                 if (!test_bit(__IXGBE_DOWN, &adapter->state)) {
2929                         adapter->flags |= IXGBE_FLAG_NEED_LINK_CONFIG;
2930                         ixgbe_service_event_schedule(adapter);
2931                 }
2932         }
2933 }
2934
2935 static void ixgbe_check_lsc(struct ixgbe_adapter *adapter)
2936 {
2937         struct ixgbe_hw *hw = &adapter->hw;
2938
2939         adapter->lsc_int++;
2940         adapter->flags |= IXGBE_FLAG_NEED_LINK_UPDATE;
2941         adapter->link_check_timeout = jiffies;
2942         if (!test_bit(__IXGBE_DOWN, &adapter->state)) {
2943                 IXGBE_WRITE_REG(hw, IXGBE_EIMC, IXGBE_EIMC_LSC);
2944                 IXGBE_WRITE_FLUSH(hw);
2945                 ixgbe_service_event_schedule(adapter);
2946         }
2947 }
2948
2949 static inline void ixgbe_irq_enable_queues(struct ixgbe_adapter *adapter,
2950                                            u64 qmask)
2951 {
2952         u32 mask;
2953         struct ixgbe_hw *hw = &adapter->hw;
2954
2955         switch (hw->mac.type) {
2956         case ixgbe_mac_82598EB:
2957                 mask = (IXGBE_EIMS_RTX_QUEUE & qmask);
2958                 IXGBE_WRITE_REG(hw, IXGBE_EIMS, mask);
2959                 break;
2960         case ixgbe_mac_82599EB:
2961         case ixgbe_mac_X540:
2962         case ixgbe_mac_X550:
2963         case ixgbe_mac_X550EM_x:
2964         case ixgbe_mac_x550em_a:
2965                 mask = (qmask & 0xFFFFFFFF);
2966                 if (mask)
2967                         IXGBE_WRITE_REG(hw, IXGBE_EIMS_EX(0), mask);
2968                 mask = (qmask >> 32);
2969                 if (mask)
2970                         IXGBE_WRITE_REG(hw, IXGBE_EIMS_EX(1), mask);
2971                 break;
2972         default:
2973                 break;
2974         }
2975         /* skip the flush */
2976 }
2977
2978 /**
2979  * ixgbe_irq_enable - Enable default interrupt generation settings
2980  * @adapter: board private structure
2981  * @queues: enable irqs for queues
2982  * @flush: flush register write
2983  **/
2984 static inline void ixgbe_irq_enable(struct ixgbe_adapter *adapter, bool queues,
2985                                     bool flush)
2986 {
2987         struct ixgbe_hw *hw = &adapter->hw;
2988         u32 mask = (IXGBE_EIMS_ENABLE_MASK & ~IXGBE_EIMS_RTX_QUEUE);
2989
2990         /* don't reenable LSC while waiting for link */
2991         if (adapter->flags & IXGBE_FLAG_NEED_LINK_UPDATE)
2992                 mask &= ~IXGBE_EIMS_LSC;
2993
2994         if (adapter->flags2 & IXGBE_FLAG2_TEMP_SENSOR_CAPABLE)
2995                 switch (adapter->hw.mac.type) {
2996                 case ixgbe_mac_82599EB:
2997                         mask |= IXGBE_EIMS_GPI_SDP0(hw);
2998                         break;
2999                 case ixgbe_mac_X540:
3000                 case ixgbe_mac_X550:
3001                 case ixgbe_mac_X550EM_x:
3002                 case ixgbe_mac_x550em_a:
3003                         mask |= IXGBE_EIMS_TS;
3004                         break;
3005                 default:
3006                         break;
3007                 }
3008         if (adapter->flags & IXGBE_FLAG_FAN_FAIL_CAPABLE)
3009                 mask |= IXGBE_EIMS_GPI_SDP1(hw);
3010         switch (adapter->hw.mac.type) {
3011         case ixgbe_mac_82599EB:
3012                 mask |= IXGBE_EIMS_GPI_SDP1(hw);
3013                 mask |= IXGBE_EIMS_GPI_SDP2(hw);
3014                 fallthrough;
3015         case ixgbe_mac_X540:
3016         case ixgbe_mac_X550:
3017         case ixgbe_mac_X550EM_x:
3018         case ixgbe_mac_x550em_a:
3019                 if (adapter->hw.device_id == IXGBE_DEV_ID_X550EM_X_SFP ||
3020                     adapter->hw.device_id == IXGBE_DEV_ID_X550EM_A_SFP ||
3021                     adapter->hw.device_id == IXGBE_DEV_ID_X550EM_A_SFP_N)
3022                         mask |= IXGBE_EIMS_GPI_SDP0(&adapter->hw);
3023                 if (adapter->hw.phy.type == ixgbe_phy_x550em_ext_t)
3024                         mask |= IXGBE_EICR_GPI_SDP0_X540;
3025                 mask |= IXGBE_EIMS_ECC;
3026                 mask |= IXGBE_EIMS_MAILBOX;
3027                 break;
3028         default:
3029                 break;
3030         }
3031
3032         if ((adapter->flags & IXGBE_FLAG_FDIR_HASH_CAPABLE) &&
3033             !(adapter->flags2 & IXGBE_FLAG2_FDIR_REQUIRES_REINIT))
3034                 mask |= IXGBE_EIMS_FLOW_DIR;
3035
3036         IXGBE_WRITE_REG(&adapter->hw, IXGBE_EIMS, mask);
3037         if (queues)
3038                 ixgbe_irq_enable_queues(adapter, ~0);
3039         if (flush)
3040                 IXGBE_WRITE_FLUSH(&adapter->hw);
3041 }
3042
3043 static irqreturn_t ixgbe_msix_other(int irq, void *data)
3044 {
3045         struct ixgbe_adapter *adapter = data;
3046         struct ixgbe_hw *hw = &adapter->hw;
3047         u32 eicr;
3048
3049         /*
3050          * Workaround for Silicon errata.  Use clear-by-write instead
3051          * of clear-by-read.  Reading with EICS will return the
3052          * interrupt causes without clearing, which later be done
3053          * with the write to EICR.
3054          */
3055         eicr = IXGBE_READ_REG(hw, IXGBE_EICS);
3056
3057         /* The lower 16bits of the EICR register are for the queue interrupts
3058          * which should be masked here in order to not accidentally clear them if
3059          * the bits are high when ixgbe_msix_other is called. There is a race
3060          * condition otherwise which results in possible performance loss
3061          * especially if the ixgbe_msix_other interrupt is triggering
3062          * consistently (as it would when PPS is turned on for the X540 device)
3063          */
3064         eicr &= 0xFFFF0000;
3065
3066         IXGBE_WRITE_REG(hw, IXGBE_EICR, eicr);
3067
3068         if (eicr & IXGBE_EICR_LSC)
3069                 ixgbe_check_lsc(adapter);
3070
3071         if (eicr & IXGBE_EICR_MAILBOX)
3072                 ixgbe_msg_task(adapter);
3073
3074         switch (hw->mac.type) {
3075         case ixgbe_mac_82599EB:
3076         case ixgbe_mac_X540:
3077         case ixgbe_mac_X550:
3078         case ixgbe_mac_X550EM_x:
3079         case ixgbe_mac_x550em_a:
3080                 if (hw->phy.type == ixgbe_phy_x550em_ext_t &&
3081                     (eicr & IXGBE_EICR_GPI_SDP0_X540)) {
3082                         adapter->flags2 |= IXGBE_FLAG2_PHY_INTERRUPT;
3083                         ixgbe_service_event_schedule(adapter);
3084                         IXGBE_WRITE_REG(hw, IXGBE_EICR,
3085                                         IXGBE_EICR_GPI_SDP0_X540);
3086                 }
3087                 if (eicr & IXGBE_EICR_ECC) {
3088                         e_info(link, "Received ECC Err, initiating reset\n");
3089                         set_bit(__IXGBE_RESET_REQUESTED, &adapter->state);
3090                         ixgbe_service_event_schedule(adapter);
3091                         IXGBE_WRITE_REG(hw, IXGBE_EICR, IXGBE_EICR_ECC);
3092                 }
3093                 /* Handle Flow Director Full threshold interrupt */
3094                 if (eicr & IXGBE_EICR_FLOW_DIR) {
3095                         int reinit_count = 0;
3096                         int i;
3097                         for (i = 0; i < adapter->num_tx_queues; i++) {
3098                                 struct ixgbe_ring *ring = adapter->tx_ring[i];
3099                                 if (test_and_clear_bit(__IXGBE_TX_FDIR_INIT_DONE,
3100                                                        &ring->state))
3101                                         reinit_count++;
3102                         }
3103                         if (reinit_count) {
3104                                 /* no more flow director interrupts until after init */
3105                                 IXGBE_WRITE_REG(hw, IXGBE_EIMC, IXGBE_EIMC_FLOW_DIR);
3106                                 adapter->flags2 |= IXGBE_FLAG2_FDIR_REQUIRES_REINIT;
3107                                 ixgbe_service_event_schedule(adapter);
3108                         }
3109                 }
3110                 ixgbe_check_sfp_event(adapter, eicr);
3111                 ixgbe_check_overtemp_event(adapter, eicr);
3112                 break;
3113         default:
3114                 break;
3115         }
3116
3117         ixgbe_check_fan_failure(adapter, eicr);
3118
3119         if (unlikely(eicr & IXGBE_EICR_TIMESYNC))
3120                 ixgbe_ptp_check_pps_event(adapter);
3121
3122         /* re-enable the original interrupt state, no lsc, no queues */
3123         if (!test_bit(__IXGBE_DOWN, &adapter->state))
3124                 ixgbe_irq_enable(adapter, false, false);
3125
3126         return IRQ_HANDLED;
3127 }
3128
3129 static irqreturn_t ixgbe_msix_clean_rings(int irq, void *data)
3130 {
3131         struct ixgbe_q_vector *q_vector = data;
3132
3133         /* EIAM disabled interrupts (on this vector) for us */
3134
3135         if (q_vector->rx.ring || q_vector->tx.ring)
3136                 napi_schedule_irqoff(&q_vector->napi);
3137
3138         return IRQ_HANDLED;
3139 }
3140
3141 /**
3142  * ixgbe_poll - NAPI Rx polling callback
3143  * @napi: structure for representing this polling device
3144  * @budget: how many packets driver is allowed to clean
3145  *
3146  * This function is used for legacy and MSI, NAPI mode
3147  **/
3148 int ixgbe_poll(struct napi_struct *napi, int budget)
3149 {
3150         struct ixgbe_q_vector *q_vector =
3151                                 container_of(napi, struct ixgbe_q_vector, napi);
3152         struct ixgbe_adapter *adapter = q_vector->adapter;
3153         struct ixgbe_ring *ring;
3154         int per_ring_budget, work_done = 0;
3155         bool clean_complete = true;
3156
3157 #ifdef CONFIG_IXGBE_DCA
3158         if (adapter->flags & IXGBE_FLAG_DCA_ENABLED)
3159                 ixgbe_update_dca(q_vector);
3160 #endif
3161
3162         ixgbe_for_each_ring(ring, q_vector->tx) {
3163                 bool wd = ring->xsk_pool ?
3164                           ixgbe_clean_xdp_tx_irq(q_vector, ring, budget) :
3165                           ixgbe_clean_tx_irq(q_vector, ring, budget);
3166
3167                 if (!wd)
3168                         clean_complete = false;
3169         }
3170
3171         /* Exit if we are called by netpoll */
3172         if (budget <= 0)
3173                 return budget;
3174
3175         /* attempt to distribute budget to each queue fairly, but don't allow
3176          * the budget to go below 1 because we'll exit polling */
3177         if (q_vector->rx.count > 1)
3178                 per_ring_budget = max(budget/q_vector->rx.count, 1);
3179         else
3180                 per_ring_budget = budget;
3181
3182         ixgbe_for_each_ring(ring, q_vector->rx) {
3183                 int cleaned = ring->xsk_pool ?
3184                               ixgbe_clean_rx_irq_zc(q_vector, ring,
3185                                                     per_ring_budget) :
3186                               ixgbe_clean_rx_irq(q_vector, ring,
3187                                                  per_ring_budget);
3188
3189                 work_done += cleaned;
3190                 if (cleaned >= per_ring_budget)
3191                         clean_complete = false;
3192         }
3193
3194         /* If all work not completed, return budget and keep polling */
3195         if (!clean_complete)
3196                 return budget;
3197
3198         /* all work done, exit the polling mode */
3199         if (likely(napi_complete_done(napi, work_done))) {
3200                 if (adapter->rx_itr_setting & 1)
3201                         ixgbe_set_itr(q_vector);
3202                 if (!test_bit(__IXGBE_DOWN, &adapter->state))
3203                         ixgbe_irq_enable_queues(adapter,
3204                                                 BIT_ULL(q_vector->v_idx));
3205         }
3206
3207         return min(work_done, budget - 1);
3208 }
3209
3210 /**
3211  * ixgbe_request_msix_irqs - Initialize MSI-X interrupts
3212  * @adapter: board private structure
3213  *
3214  * ixgbe_request_msix_irqs allocates MSI-X vectors and requests
3215  * interrupts from the kernel.
3216  **/
3217 static int ixgbe_request_msix_irqs(struct ixgbe_adapter *adapter)
3218 {
3219         struct net_device *netdev = adapter->netdev;
3220         unsigned int ri = 0, ti = 0;
3221         int vector, err;
3222
3223         for (vector = 0; vector < adapter->num_q_vectors; vector++) {
3224                 struct ixgbe_q_vector *q_vector = adapter->q_vector[vector];
3225                 struct msix_entry *entry = &adapter->msix_entries[vector];
3226
3227                 if (q_vector->tx.ring && q_vector->rx.ring) {
3228                         snprintf(q_vector->name, sizeof(q_vector->name),
3229                                  "%s-TxRx-%u", netdev->name, ri++);
3230                         ti++;
3231                 } else if (q_vector->rx.ring) {
3232                         snprintf(q_vector->name, sizeof(q_vector->name),
3233                                  "%s-rx-%u", netdev->name, ri++);
3234                 } else if (q_vector->tx.ring) {
3235                         snprintf(q_vector->name, sizeof(q_vector->name),
3236                                  "%s-tx-%u", netdev->name, ti++);
3237                 } else {
3238                         /* skip this unused q_vector */
3239                         continue;
3240                 }
3241                 err = request_irq(entry->vector, &ixgbe_msix_clean_rings, 0,
3242                                   q_vector->name, q_vector);
3243                 if (err) {
3244                         e_err(probe, "request_irq failed for MSIX interrupt "
3245                               "Error: %d\n", err);
3246                         goto free_queue_irqs;
3247                 }
3248                 /* If Flow Director is enabled, set interrupt affinity */
3249                 if (adapter->flags & IXGBE_FLAG_FDIR_HASH_CAPABLE) {
3250                         /* assign the mask for this irq */
3251                         irq_update_affinity_hint(entry->vector,
3252                                                  &q_vector->affinity_mask);
3253                 }
3254         }
3255
3256         err = request_irq(adapter->msix_entries[vector].vector,
3257                           ixgbe_msix_other, 0, netdev->name, adapter);
3258         if (err) {
3259                 e_err(probe, "request_irq for msix_other failed: %d\n", err);
3260                 goto free_queue_irqs;
3261         }
3262
3263         return 0;
3264
3265 free_queue_irqs:
3266         while (vector) {
3267                 vector--;
3268                 irq_update_affinity_hint(adapter->msix_entries[vector].vector,
3269                                          NULL);
3270                 free_irq(adapter->msix_entries[vector].vector,
3271                          adapter->q_vector[vector]);
3272         }
3273         adapter->flags &= ~IXGBE_FLAG_MSIX_ENABLED;
3274         pci_disable_msix(adapter->pdev);
3275         kfree(adapter->msix_entries);
3276         adapter->msix_entries = NULL;
3277         return err;
3278 }
3279
3280 /**
3281  * ixgbe_intr - legacy mode Interrupt Handler
3282  * @irq: interrupt number
3283  * @data: pointer to a network interface device structure
3284  **/
3285 static irqreturn_t ixgbe_intr(int irq, void *data)
3286 {
3287         struct ixgbe_adapter *adapter = data;
3288         struct ixgbe_hw *hw = &adapter->hw;
3289         struct ixgbe_q_vector *q_vector = adapter->q_vector[0];
3290         u32 eicr;
3291
3292         /*
3293          * Workaround for silicon errata #26 on 82598.  Mask the interrupt
3294          * before the read of EICR.
3295          */
3296         IXGBE_WRITE_REG(hw, IXGBE_EIMC, IXGBE_IRQ_CLEAR_MASK);
3297
3298         /* for NAPI, using EIAM to auto-mask tx/rx interrupt bits on read
3299          * therefore no explicit interrupt disable is necessary */
3300         eicr = IXGBE_READ_REG(hw, IXGBE_EICR);
3301         if (!eicr) {
3302                 /*
3303                  * shared interrupt alert!
3304                  * make sure interrupts are enabled because the read will
3305                  * have disabled interrupts due to EIAM
3306                  * finish the workaround of silicon errata on 82598.  Unmask
3307                  * the interrupt that we masked before the EICR read.
3308                  */
3309                 if (!test_bit(__IXGBE_DOWN, &adapter->state))
3310                         ixgbe_irq_enable(adapter, true, true);
3311                 return IRQ_NONE;        /* Not our interrupt */
3312         }
3313
3314         if (eicr & IXGBE_EICR_LSC)
3315                 ixgbe_check_lsc(adapter);
3316
3317         switch (hw->mac.type) {
3318         case ixgbe_mac_82599EB:
3319                 ixgbe_check_sfp_event(adapter, eicr);
3320                 fallthrough;
3321         case ixgbe_mac_X540:
3322         case ixgbe_mac_X550:
3323         case ixgbe_mac_X550EM_x:
3324         case ixgbe_mac_x550em_a:
3325                 if (eicr & IXGBE_EICR_ECC) {
3326                         e_info(link, "Received ECC Err, initiating reset\n");
3327                         set_bit(__IXGBE_RESET_REQUESTED, &adapter->state);
3328                         ixgbe_service_event_schedule(adapter);
3329                         IXGBE_WRITE_REG(hw, IXGBE_EICR, IXGBE_EICR_ECC);
3330                 }
3331                 ixgbe_check_overtemp_event(adapter, eicr);
3332                 break;
3333         default:
3334                 break;
3335         }
3336
3337         ixgbe_check_fan_failure(adapter, eicr);
3338         if (unlikely(eicr & IXGBE_EICR_TIMESYNC))
3339                 ixgbe_ptp_check_pps_event(adapter);
3340
3341         /* would disable interrupts here but EIAM disabled it */
3342         napi_schedule_irqoff(&q_vector->napi);
3343
3344         /*
3345          * re-enable link(maybe) and non-queue interrupts, no flush.
3346          * ixgbe_poll will re-enable the queue interrupts
3347          */
3348         if (!test_bit(__IXGBE_DOWN, &adapter->state))
3349                 ixgbe_irq_enable(adapter, false, false);
3350
3351         return IRQ_HANDLED;
3352 }
3353
3354 /**
3355  * ixgbe_request_irq - initialize interrupts
3356  * @adapter: board private structure
3357  *
3358  * Attempts to configure interrupts using the best available
3359  * capabilities of the hardware and kernel.
3360  **/
3361 static int ixgbe_request_irq(struct ixgbe_adapter *adapter)
3362 {
3363         struct net_device *netdev = adapter->netdev;
3364         int err;
3365
3366         if (adapter->flags & IXGBE_FLAG_MSIX_ENABLED)
3367                 err = ixgbe_request_msix_irqs(adapter);
3368         else if (adapter->flags & IXGBE_FLAG_MSI_ENABLED)
3369                 err = request_irq(adapter->pdev->irq, ixgbe_intr, 0,
3370                                   netdev->name, adapter);
3371         else
3372                 err = request_irq(adapter->pdev->irq, ixgbe_intr, IRQF_SHARED,
3373                                   netdev->name, adapter);
3374
3375         if (err)
3376                 e_err(probe, "request_irq failed, Error %d\n", err);
3377
3378         return err;
3379 }
3380
3381 static void ixgbe_free_irq(struct ixgbe_adapter *adapter)
3382 {
3383         int vector;
3384
3385         if (!(adapter->flags & IXGBE_FLAG_MSIX_ENABLED)) {
3386                 free_irq(adapter->pdev->irq, adapter);
3387                 return;
3388         }
3389
3390         if (!adapter->msix_entries)
3391                 return;
3392
3393         for (vector = 0; vector < adapter->num_q_vectors; vector++) {
3394                 struct ixgbe_q_vector *q_vector = adapter->q_vector[vector];
3395                 struct msix_entry *entry = &adapter->msix_entries[vector];
3396
3397                 /* free only the irqs that were actually requested */
3398                 if (!q_vector->rx.ring && !q_vector->tx.ring)
3399                         continue;
3400
3401                 /* clear the affinity_mask in the IRQ descriptor */
3402                 irq_update_affinity_hint(entry->vector, NULL);
3403
3404                 free_irq(entry->vector, q_vector);
3405         }
3406
3407         free_irq(adapter->msix_entries[vector].vector, adapter);
3408 }
3409
3410 /**
3411  * ixgbe_irq_disable - Mask off interrupt generation on the NIC
3412  * @adapter: board private structure
3413  **/
3414 static inline void ixgbe_irq_disable(struct ixgbe_adapter *adapter)
3415 {
3416         switch (adapter->hw.mac.type) {
3417         case ixgbe_mac_82598EB:
3418                 IXGBE_WRITE_REG(&adapter->hw, IXGBE_EIMC, ~0);
3419                 break;
3420         case ixgbe_mac_82599EB:
3421         case ixgbe_mac_X540:
3422         case ixgbe_mac_X550:
3423         case ixgbe_mac_X550EM_x:
3424         case ixgbe_mac_x550em_a:
3425                 IXGBE_WRITE_REG(&adapter->hw, IXGBE_EIMC, 0xFFFF0000);
3426                 IXGBE_WRITE_REG(&adapter->hw, IXGBE_EIMC_EX(0), ~0);
3427                 IXGBE_WRITE_REG(&adapter->hw, IXGBE_EIMC_EX(1), ~0);
3428                 break;
3429         default:
3430                 break;
3431         }
3432         IXGBE_WRITE_FLUSH(&adapter->hw);
3433         if (adapter->flags & IXGBE_FLAG_MSIX_ENABLED) {
3434                 int vector;
3435
3436                 for (vector = 0; vector < adapter->num_q_vectors; vector++)
3437                         synchronize_irq(adapter->msix_entries[vector].vector);
3438
3439                 synchronize_irq(adapter->msix_entries[vector++].vector);
3440         } else {
3441                 synchronize_irq(adapter->pdev->irq);
3442         }
3443 }
3444
3445 /**
3446  * ixgbe_configure_msi_and_legacy - Initialize PIN (INTA...) and MSI interrupts
3447  * @adapter: board private structure
3448  *
3449  **/
3450 static void ixgbe_configure_msi_and_legacy(struct ixgbe_adapter *adapter)
3451 {
3452         struct ixgbe_q_vector *q_vector = adapter->q_vector[0];
3453
3454         ixgbe_write_eitr(q_vector);
3455
3456         ixgbe_set_ivar(adapter, 0, 0, 0);
3457         ixgbe_set_ivar(adapter, 1, 0, 0);
3458
3459         e_info(hw, "Legacy interrupt IVAR setup done\n");
3460 }
3461
3462 /**
3463  * ixgbe_configure_tx_ring - Configure 8259x Tx ring after Reset
3464  * @adapter: board private structure
3465  * @ring: structure containing ring specific data
3466  *
3467  * Configure the Tx descriptor ring after a reset.
3468  **/
3469 void ixgbe_configure_tx_ring(struct ixgbe_adapter *adapter,
3470                              struct ixgbe_ring *ring)
3471 {
3472         struct ixgbe_hw *hw = &adapter->hw;
3473         u64 tdba = ring->dma;
3474         int wait_loop = 10;
3475         u32 txdctl = IXGBE_TXDCTL_ENABLE;
3476         u8 reg_idx = ring->reg_idx;
3477
3478         ring->xsk_pool = NULL;
3479         if (ring_is_xdp(ring))
3480                 ring->xsk_pool = ixgbe_xsk_pool(adapter, ring);
3481
3482         /* disable queue to avoid issues while updating state */
3483         IXGBE_WRITE_REG(hw, IXGBE_TXDCTL(reg_idx), 0);
3484         IXGBE_WRITE_FLUSH(hw);
3485
3486         IXGBE_WRITE_REG(hw, IXGBE_TDBAL(reg_idx),
3487                         (tdba & DMA_BIT_MASK(32)));
3488         IXGBE_WRITE_REG(hw, IXGBE_TDBAH(reg_idx), (tdba >> 32));
3489         IXGBE_WRITE_REG(hw, IXGBE_TDLEN(reg_idx),
3490                         ring->count * sizeof(union ixgbe_adv_tx_desc));
3491         IXGBE_WRITE_REG(hw, IXGBE_TDH(reg_idx), 0);
3492         IXGBE_WRITE_REG(hw, IXGBE_TDT(reg_idx), 0);
3493         ring->tail = adapter->io_addr + IXGBE_TDT(reg_idx);
3494
3495         /*
3496          * set WTHRESH to encourage burst writeback, it should not be set
3497          * higher than 1 when:
3498          * - ITR is 0 as it could cause false TX hangs
3499          * - ITR is set to > 100k int/sec and BQL is enabled
3500          *
3501          * In order to avoid issues WTHRESH + PTHRESH should always be equal
3502          * to or less than the number of on chip descriptors, which is
3503          * currently 40.
3504          */
3505         if (!ring->q_vector || (ring->q_vector->itr < IXGBE_100K_ITR))
3506                 txdctl |= 1u << 16;     /* WTHRESH = 1 */
3507         else
3508                 txdctl |= 8u << 16;     /* WTHRESH = 8 */
3509
3510         /*
3511          * Setting PTHRESH to 32 both improves performance
3512          * and avoids a TX hang with DFP enabled
3513          */
3514         txdctl |= (1u << 8) |   /* HTHRESH = 1 */
3515                    32;          /* PTHRESH = 32 */
3516
3517         /* reinitialize flowdirector state */
3518         if (adapter->flags & IXGBE_FLAG_FDIR_HASH_CAPABLE) {
3519                 ring->atr_sample_rate = adapter->atr_sample_rate;
3520                 ring->atr_count = 0;
3521                 set_bit(__IXGBE_TX_FDIR_INIT_DONE, &ring->state);
3522         } else {
3523                 ring->atr_sample_rate = 0;
3524         }
3525
3526         /* initialize XPS */
3527         if (!test_and_set_bit(__IXGBE_TX_XPS_INIT_DONE, &ring->state)) {
3528                 struct ixgbe_q_vector *q_vector = ring->q_vector;
3529
3530                 if (q_vector)
3531                         netif_set_xps_queue(ring->netdev,
3532                                             &q_vector->affinity_mask,
3533                                             ring->queue_index);
3534         }
3535
3536         clear_bit(__IXGBE_HANG_CHECK_ARMED, &ring->state);
3537
3538         /* reinitialize tx_buffer_info */
3539         memset(ring->tx_buffer_info, 0,
3540                sizeof(struct ixgbe_tx_buffer) * ring->count);
3541
3542         /* enable queue */
3543         IXGBE_WRITE_REG(hw, IXGBE_TXDCTL(reg_idx), txdctl);
3544
3545         /* TXDCTL.EN will return 0 on 82598 if link is down, so skip it */
3546         if (hw->mac.type == ixgbe_mac_82598EB &&
3547             !(IXGBE_READ_REG(hw, IXGBE_LINKS) & IXGBE_LINKS_UP))
3548                 return;
3549
3550         /* poll to verify queue is enabled */
3551         do {
3552                 usleep_range(1000, 2000);
3553                 txdctl = IXGBE_READ_REG(hw, IXGBE_TXDCTL(reg_idx));
3554         } while (--wait_loop && !(txdctl & IXGBE_TXDCTL_ENABLE));
3555         if (!wait_loop)
3556                 hw_dbg(hw, "Could not enable Tx Queue %d\n", reg_idx);
3557 }
3558
3559 static void ixgbe_setup_mtqc(struct ixgbe_adapter *adapter)
3560 {
3561         struct ixgbe_hw *hw = &adapter->hw;
3562         u32 rttdcs, mtqc;
3563         u8 tcs = adapter->hw_tcs;
3564
3565         if (hw->mac.type == ixgbe_mac_82598EB)
3566                 return;
3567
3568         /* disable the arbiter while setting MTQC */
3569         rttdcs = IXGBE_READ_REG(hw, IXGBE_RTTDCS);
3570         rttdcs |= IXGBE_RTTDCS_ARBDIS;
3571         IXGBE_WRITE_REG(hw, IXGBE_RTTDCS, rttdcs);
3572
3573         /* set transmit pool layout */
3574         if (adapter->flags & IXGBE_FLAG_SRIOV_ENABLED) {
3575                 mtqc = IXGBE_MTQC_VT_ENA;
3576                 if (tcs > 4)
3577                         mtqc |= IXGBE_MTQC_RT_ENA | IXGBE_MTQC_8TC_8TQ;
3578                 else if (tcs > 1)
3579                         mtqc |= IXGBE_MTQC_RT_ENA | IXGBE_MTQC_4TC_4TQ;
3580                 else if (adapter->ring_feature[RING_F_VMDQ].mask ==
3581                          IXGBE_82599_VMDQ_4Q_MASK)
3582                         mtqc |= IXGBE_MTQC_32VF;
3583                 else
3584                         mtqc |= IXGBE_MTQC_64VF;
3585         } else {
3586                 if (tcs > 4) {
3587                         mtqc = IXGBE_MTQC_RT_ENA | IXGBE_MTQC_8TC_8TQ;
3588                 } else if (tcs > 1) {
3589                         mtqc = IXGBE_MTQC_RT_ENA | IXGBE_MTQC_4TC_4TQ;
3590                 } else {
3591                         u8 max_txq = adapter->num_tx_queues +
3592                                 adapter->num_xdp_queues;
3593                         if (max_txq > 63)
3594                                 mtqc = IXGBE_MTQC_RT_ENA | IXGBE_MTQC_4TC_4TQ;
3595                         else
3596                                 mtqc = IXGBE_MTQC_64Q_1PB;
3597                 }
3598         }
3599
3600         IXGBE_WRITE_REG(hw, IXGBE_MTQC, mtqc);
3601
3602         /* Enable Security TX Buffer IFG for multiple pb */
3603         if (tcs) {
3604                 u32 sectx = IXGBE_READ_REG(hw, IXGBE_SECTXMINIFG);
3605                 sectx |= IXGBE_SECTX_DCB;
3606                 IXGBE_WRITE_REG(hw, IXGBE_SECTXMINIFG, sectx);
3607         }
3608
3609         /* re-enable the arbiter */
3610         rttdcs &= ~IXGBE_RTTDCS_ARBDIS;
3611         IXGBE_WRITE_REG(hw, IXGBE_RTTDCS, rttdcs);
3612 }
3613
3614 /**
3615  * ixgbe_configure_tx - Configure 8259x Transmit Unit after Reset
3616  * @adapter: board private structure
3617  *
3618  * Configure the Tx unit of the MAC after a reset.
3619  **/
3620 static void ixgbe_configure_tx(struct ixgbe_adapter *adapter)
3621 {
3622         struct ixgbe_hw *hw = &adapter->hw;
3623         u32 dmatxctl;
3624         u32 i;
3625
3626         ixgbe_setup_mtqc(adapter);
3627
3628         if (hw->mac.type != ixgbe_mac_82598EB) {
3629                 /* DMATXCTL.EN must be before Tx queues are enabled */
3630                 dmatxctl = IXGBE_READ_REG(hw, IXGBE_DMATXCTL);
3631                 dmatxctl |= IXGBE_DMATXCTL_TE;
3632                 IXGBE_WRITE_REG(hw, IXGBE_DMATXCTL, dmatxctl);
3633         }
3634
3635         /* Setup the HW Tx Head and Tail descriptor pointers */
3636         for (i = 0; i < adapter->num_tx_queues; i++)
3637                 ixgbe_configure_tx_ring(adapter, adapter->tx_ring[i]);
3638         for (i = 0; i < adapter->num_xdp_queues; i++)
3639                 ixgbe_configure_tx_ring(adapter, adapter->xdp_ring[i]);
3640 }
3641
3642 static void ixgbe_enable_rx_drop(struct ixgbe_adapter *adapter,
3643                                  struct ixgbe_ring *ring)
3644 {
3645         struct ixgbe_hw *hw = &adapter->hw;
3646         u8 reg_idx = ring->reg_idx;
3647         u32 srrctl = IXGBE_READ_REG(hw, IXGBE_SRRCTL(reg_idx));
3648
3649         srrctl |= IXGBE_SRRCTL_DROP_EN;
3650
3651         IXGBE_WRITE_REG(hw, IXGBE_SRRCTL(reg_idx), srrctl);
3652 }
3653
3654 static void ixgbe_disable_rx_drop(struct ixgbe_adapter *adapter,
3655                                   struct ixgbe_ring *ring)
3656 {
3657         struct ixgbe_hw *hw = &adapter->hw;
3658         u8 reg_idx = ring->reg_idx;
3659         u32 srrctl = IXGBE_READ_REG(hw, IXGBE_SRRCTL(reg_idx));
3660
3661         srrctl &= ~IXGBE_SRRCTL_DROP_EN;
3662
3663         IXGBE_WRITE_REG(hw, IXGBE_SRRCTL(reg_idx), srrctl);
3664 }
3665
3666 #ifdef CONFIG_IXGBE_DCB
3667 void ixgbe_set_rx_drop_en(struct ixgbe_adapter *adapter)
3668 #else
3669 static void ixgbe_set_rx_drop_en(struct ixgbe_adapter *adapter)
3670 #endif
3671 {
3672         int i;
3673         bool pfc_en = adapter->dcb_cfg.pfc_mode_enable;
3674
3675         if (adapter->ixgbe_ieee_pfc)
3676                 pfc_en |= !!(adapter->ixgbe_ieee_pfc->pfc_en);
3677
3678         /*
3679          * We should set the drop enable bit if:
3680          *  SR-IOV is enabled
3681          *   or
3682          *  Number of Rx queues > 1 and flow control is disabled
3683          *
3684          *  This allows us to avoid head of line blocking for security
3685          *  and performance reasons.
3686          */
3687         if (adapter->num_vfs || (adapter->num_rx_queues > 1 &&
3688             !(adapter->hw.fc.current_mode & ixgbe_fc_tx_pause) && !pfc_en)) {
3689                 for (i = 0; i < adapter->num_rx_queues; i++)
3690                         ixgbe_enable_rx_drop(adapter, adapter->rx_ring[i]);
3691         } else {
3692                 for (i = 0; i < adapter->num_rx_queues; i++)
3693                         ixgbe_disable_rx_drop(adapter, adapter->rx_ring[i]);
3694         }
3695 }
3696
3697 #define IXGBE_SRRCTL_BSIZEHDRSIZE_SHIFT 2
3698
3699 static void ixgbe_configure_srrctl(struct ixgbe_adapter *adapter,
3700                                    struct ixgbe_ring *rx_ring)
3701 {
3702         struct ixgbe_hw *hw = &adapter->hw;
3703         u32 srrctl;
3704         u8 reg_idx = rx_ring->reg_idx;
3705
3706         if (hw->mac.type == ixgbe_mac_82598EB) {
3707                 u16 mask = adapter->ring_feature[RING_F_RSS].mask;
3708
3709                 /*
3710                  * if VMDq is not active we must program one srrctl register
3711                  * per RSS queue since we have enabled RDRXCTL.MVMEN
3712                  */
3713                 reg_idx &= mask;
3714         }
3715
3716         /* configure header buffer length, needed for RSC */
3717         srrctl = IXGBE_RX_HDR_SIZE << IXGBE_SRRCTL_BSIZEHDRSIZE_SHIFT;
3718
3719         /* configure the packet buffer length */
3720         if (rx_ring->xsk_pool) {
3721                 u32 xsk_buf_len = xsk_pool_get_rx_frame_size(rx_ring->xsk_pool);
3722
3723                 /* If the MAC support setting RXDCTL.RLPML, the
3724                  * SRRCTL[n].BSIZEPKT is set to PAGE_SIZE and
3725                  * RXDCTL.RLPML is set to the actual UMEM buffer
3726                  * size. If not, then we are stuck with a 1k buffer
3727                  * size resolution. In this case frames larger than
3728                  * the UMEM buffer size viewed in a 1k resolution will
3729                  * be dropped.
3730                  */
3731                 if (hw->mac.type != ixgbe_mac_82599EB)
3732                         srrctl |= PAGE_SIZE >> IXGBE_SRRCTL_BSIZEPKT_SHIFT;
3733                 else
3734                         srrctl |= xsk_buf_len >> IXGBE_SRRCTL_BSIZEPKT_SHIFT;
3735         } else if (test_bit(__IXGBE_RX_3K_BUFFER, &rx_ring->state)) {
3736                 srrctl |= IXGBE_RXBUFFER_3K >> IXGBE_SRRCTL_BSIZEPKT_SHIFT;
3737         } else {
3738                 srrctl |= IXGBE_RXBUFFER_2K >> IXGBE_SRRCTL_BSIZEPKT_SHIFT;
3739         }
3740
3741         /* configure descriptor type */
3742         srrctl |= IXGBE_SRRCTL_DESCTYPE_ADV_ONEBUF;
3743
3744         IXGBE_WRITE_REG(hw, IXGBE_SRRCTL(reg_idx), srrctl);
3745 }
3746
3747 /**
3748  * ixgbe_rss_indir_tbl_entries - Return RSS indirection table entries
3749  * @adapter: device handle
3750  *
3751  *  - 82598/82599/X540:     128
3752  *  - X550(non-SRIOV mode): 512
3753  *  - X550(SRIOV mode):     64
3754  */
3755 u32 ixgbe_rss_indir_tbl_entries(struct ixgbe_adapter *adapter)
3756 {
3757         if (adapter->hw.mac.type < ixgbe_mac_X550)
3758                 return 128;
3759         else if (adapter->flags & IXGBE_FLAG_SRIOV_ENABLED)
3760                 return 64;
3761         else
3762                 return 512;
3763 }
3764
3765 /**
3766  * ixgbe_store_key - Write the RSS key to HW
3767  * @adapter: device handle
3768  *
3769  * Write the RSS key stored in adapter.rss_key to HW.
3770  */
3771 void ixgbe_store_key(struct ixgbe_adapter *adapter)
3772 {
3773         struct ixgbe_hw *hw = &adapter->hw;
3774         int i;
3775
3776         for (i = 0; i < 10; i++)
3777                 IXGBE_WRITE_REG(hw, IXGBE_RSSRK(i), adapter->rss_key[i]);
3778 }
3779
3780 /**
3781  * ixgbe_init_rss_key - Initialize adapter RSS key
3782  * @adapter: device handle
3783  *
3784  * Allocates and initializes the RSS key if it is not allocated.
3785  **/
3786 static inline int ixgbe_init_rss_key(struct ixgbe_adapter *adapter)
3787 {
3788         u32 *rss_key;
3789
3790         if (!adapter->rss_key) {
3791                 rss_key = kzalloc(IXGBE_RSS_KEY_SIZE, GFP_KERNEL);
3792                 if (unlikely(!rss_key))
3793                         return -ENOMEM;
3794
3795                 netdev_rss_key_fill(rss_key, IXGBE_RSS_KEY_SIZE);
3796                 adapter->rss_key = rss_key;
3797         }
3798
3799         return 0;
3800 }
3801
3802 /**
3803  * ixgbe_store_reta - Write the RETA table to HW
3804  * @adapter: device handle
3805  *
3806  * Write the RSS redirection table stored in adapter.rss_indir_tbl[] to HW.
3807  */
3808 void ixgbe_store_reta(struct ixgbe_adapter *adapter)
3809 {
3810         u32 i, reta_entries = ixgbe_rss_indir_tbl_entries(adapter);
3811         struct ixgbe_hw *hw = &adapter->hw;
3812         u32 reta = 0;
3813         u32 indices_multi;
3814         u8 *indir_tbl = adapter->rss_indir_tbl;
3815
3816         /* Fill out the redirection table as follows:
3817          *  - 82598:      8 bit wide entries containing pair of 4 bit RSS
3818          *    indices.
3819          *  - 82599/X540: 8 bit wide entries containing 4 bit RSS index
3820          *  - X550:       8 bit wide entries containing 6 bit RSS index
3821          */
3822         if (adapter->hw.mac.type == ixgbe_mac_82598EB)
3823                 indices_multi = 0x11;
3824         else
3825                 indices_multi = 0x1;
3826
3827         /* Write redirection table to HW */
3828         for (i = 0; i < reta_entries; i++) {
3829                 reta |= indices_multi * indir_tbl[i] << (i & 0x3) * 8;
3830                 if ((i & 3) == 3) {
3831                         if (i < 128)
3832                                 IXGBE_WRITE_REG(hw, IXGBE_RETA(i >> 2), reta);
3833                         else
3834                                 IXGBE_WRITE_REG(hw, IXGBE_ERETA((i >> 2) - 32),
3835                                                 reta);
3836                         reta = 0;
3837                 }
3838         }
3839 }
3840
3841 /**
3842  * ixgbe_store_vfreta - Write the RETA table to HW (x550 devices in SRIOV mode)
3843  * @adapter: device handle
3844  *
3845  * Write the RSS redirection table stored in adapter.rss_indir_tbl[] to HW.
3846  */
3847 static void ixgbe_store_vfreta(struct ixgbe_adapter *adapter)
3848 {
3849         u32 i, reta_entries = ixgbe_rss_indir_tbl_entries(adapter);
3850         struct ixgbe_hw *hw = &adapter->hw;
3851         u32 vfreta = 0;
3852
3853         /* Write redirection table to HW */
3854         for (i = 0; i < reta_entries; i++) {
3855                 u16 pool = adapter->num_rx_pools;
3856
3857                 vfreta |= (u32)adapter->rss_indir_tbl[i] << (i & 0x3) * 8;
3858                 if ((i & 3) != 3)
3859                         continue;
3860
3861                 while (pool--)
3862                         IXGBE_WRITE_REG(hw,
3863                                         IXGBE_PFVFRETA(i >> 2, VMDQ_P(pool)),
3864                                         vfreta);
3865                 vfreta = 0;
3866         }
3867 }
3868
3869 static void ixgbe_setup_reta(struct ixgbe_adapter *adapter)
3870 {
3871         u32 i, j;
3872         u32 reta_entries = ixgbe_rss_indir_tbl_entries(adapter);
3873         u16 rss_i = adapter->ring_feature[RING_F_RSS].indices;
3874
3875         /* Program table for at least 4 queues w/ SR-IOV so that VFs can
3876          * make full use of any rings they may have.  We will use the
3877          * PSRTYPE register to control how many rings we use within the PF.
3878          */
3879         if ((adapter->flags & IXGBE_FLAG_SRIOV_ENABLED) && (rss_i < 4))
3880                 rss_i = 4;
3881
3882         /* Fill out hash function seeds */
3883         ixgbe_store_key(adapter);
3884
3885         /* Fill out redirection table */
3886         memset(adapter->rss_indir_tbl, 0, sizeof(adapter->rss_indir_tbl));
3887
3888         for (i = 0, j = 0; i < reta_entries; i++, j++) {
3889                 if (j == rss_i)
3890                         j = 0;
3891
3892                 adapter->rss_indir_tbl[i] = j;
3893         }
3894
3895         ixgbe_store_reta(adapter);
3896 }
3897
3898 static void ixgbe_setup_vfreta(struct ixgbe_adapter *adapter)
3899 {
3900         struct ixgbe_hw *hw = &adapter->hw;
3901         u16 rss_i = adapter->ring_feature[RING_F_RSS].indices;
3902         int i, j;
3903
3904         /* Fill out hash function seeds */
3905         for (i = 0; i < 10; i++) {
3906                 u16 pool = adapter->num_rx_pools;
3907
3908                 while (pool--)
3909                         IXGBE_WRITE_REG(hw,
3910                                         IXGBE_PFVFRSSRK(i, VMDQ_P(pool)),
3911                                         *(adapter->rss_key + i));
3912         }
3913
3914         /* Fill out the redirection table */
3915         for (i = 0, j = 0; i < 64; i++, j++) {
3916                 if (j == rss_i)
3917                         j = 0;
3918
3919                 adapter->rss_indir_tbl[i] = j;
3920         }
3921
3922         ixgbe_store_vfreta(adapter);
3923 }
3924
3925 static void ixgbe_setup_mrqc(struct ixgbe_adapter *adapter)
3926 {
3927         struct ixgbe_hw *hw = &adapter->hw;
3928         u32 mrqc = 0, rss_field = 0, vfmrqc = 0;
3929         u32 rxcsum;
3930
3931         /* Disable indicating checksum in descriptor, enables RSS hash */
3932         rxcsum = IXGBE_READ_REG(hw, IXGBE_RXCSUM);
3933         rxcsum |= IXGBE_RXCSUM_PCSD;
3934         IXGBE_WRITE_REG(hw, IXGBE_RXCSUM, rxcsum);
3935
3936         if (adapter->hw.mac.type == ixgbe_mac_82598EB) {
3937                 if (adapter->ring_feature[RING_F_RSS].mask)
3938                         mrqc = IXGBE_MRQC_RSSEN;
3939         } else {
3940                 u8 tcs = adapter->hw_tcs;
3941
3942                 if (adapter->flags & IXGBE_FLAG_SRIOV_ENABLED) {
3943                         if (tcs > 4)
3944                                 mrqc = IXGBE_MRQC_VMDQRT8TCEN;  /* 8 TCs */
3945                         else if (tcs > 1)
3946                                 mrqc = IXGBE_MRQC_VMDQRT4TCEN;  /* 4 TCs */
3947                         else if (adapter->ring_feature[RING_F_VMDQ].mask ==
3948                                  IXGBE_82599_VMDQ_4Q_MASK)
3949                                 mrqc = IXGBE_MRQC_VMDQRSS32EN;
3950                         else
3951                                 mrqc = IXGBE_MRQC_VMDQRSS64EN;
3952
3953                         /* Enable L3/L4 for Tx Switched packets only for X550,
3954                          * older devices do not support this feature
3955                          */
3956                         if (hw->mac.type >= ixgbe_mac_X550)
3957                                 mrqc |= IXGBE_MRQC_L3L4TXSWEN;
3958                 } else {
3959                         if (tcs > 4)
3960                                 mrqc = IXGBE_MRQC_RTRSS8TCEN;
3961                         else if (tcs > 1)
3962                                 mrqc = IXGBE_MRQC_RTRSS4TCEN;
3963                         else
3964                                 mrqc = IXGBE_MRQC_RSSEN;
3965                 }
3966         }
3967
3968         /* Perform hash on these packet types */
3969         rss_field |= IXGBE_MRQC_RSS_FIELD_IPV4 |
3970                      IXGBE_MRQC_RSS_FIELD_IPV4_TCP |
3971                      IXGBE_MRQC_RSS_FIELD_IPV6 |
3972                      IXGBE_MRQC_RSS_FIELD_IPV6_TCP;
3973
3974         if (adapter->flags2 & IXGBE_FLAG2_RSS_FIELD_IPV4_UDP)
3975                 rss_field |= IXGBE_MRQC_RSS_FIELD_IPV4_UDP;
3976         if (adapter->flags2 & IXGBE_FLAG2_RSS_FIELD_IPV6_UDP)
3977                 rss_field |= IXGBE_MRQC_RSS_FIELD_IPV6_UDP;
3978
3979         if ((hw->mac.type >= ixgbe_mac_X550) &&
3980             (adapter->flags & IXGBE_FLAG_SRIOV_ENABLED)) {
3981                 u16 pool = adapter->num_rx_pools;
3982
3983                 /* Enable VF RSS mode */
3984                 mrqc |= IXGBE_MRQC_MULTIPLE_RSS;
3985                 IXGBE_WRITE_REG(hw, IXGBE_MRQC, mrqc);
3986
3987                 /* Setup RSS through the VF registers */
3988                 ixgbe_setup_vfreta(adapter);
3989                 vfmrqc = IXGBE_MRQC_RSSEN;
3990                 vfmrqc |= rss_field;
3991
3992                 while (pool--)
3993                         IXGBE_WRITE_REG(hw,
3994                                         IXGBE_PFVFMRQC(VMDQ_P(pool)),
3995                                         vfmrqc);
3996         } else {
3997                 ixgbe_setup_reta(adapter);
3998                 mrqc |= rss_field;
3999                 IXGBE_WRITE_REG(hw, IXGBE_MRQC, mrqc);
4000         }
4001 }
4002
4003 /**
4004  * ixgbe_configure_rscctl - enable RSC for the indicated ring
4005  * @adapter: address of board private structure
4006  * @ring: structure containing ring specific data
4007  **/
4008 static void ixgbe_configure_rscctl(struct ixgbe_adapter *adapter,
4009                                    struct ixgbe_ring *ring)
4010 {
4011         struct ixgbe_hw *hw = &adapter->hw;
4012         u32 rscctrl;
4013         u8 reg_idx = ring->reg_idx;
4014
4015         if (!ring_is_rsc_enabled(ring))
4016                 return;
4017
4018         rscctrl = IXGBE_READ_REG(hw, IXGBE_RSCCTL(reg_idx));
4019         rscctrl |= IXGBE_RSCCTL_RSCEN;
4020         /*
4021          * we must limit the number of descriptors so that the
4022          * total size of max desc * buf_len is not greater
4023          * than 65536
4024          */
4025         rscctrl |= IXGBE_RSCCTL_MAXDESC_16;
4026         IXGBE_WRITE_REG(hw, IXGBE_RSCCTL(reg_idx), rscctrl);
4027 }
4028
4029 #define IXGBE_MAX_RX_DESC_POLL 10
4030 static void ixgbe_rx_desc_queue_enable(struct ixgbe_adapter *adapter,
4031                                        struct ixgbe_ring *ring)
4032 {
4033         struct ixgbe_hw *hw = &adapter->hw;
4034         int wait_loop = IXGBE_MAX_RX_DESC_POLL;
4035         u32 rxdctl;
4036         u8 reg_idx = ring->reg_idx;
4037
4038         if (ixgbe_removed(hw->hw_addr))
4039                 return;
4040         /* RXDCTL.EN will return 0 on 82598 if link is down, so skip it */
4041         if (hw->mac.type == ixgbe_mac_82598EB &&
4042             !(IXGBE_READ_REG(hw, IXGBE_LINKS) & IXGBE_LINKS_UP))
4043                 return;
4044
4045         do {
4046                 usleep_range(1000, 2000);
4047                 rxdctl = IXGBE_READ_REG(hw, IXGBE_RXDCTL(reg_idx));
4048         } while (--wait_loop && !(rxdctl & IXGBE_RXDCTL_ENABLE));
4049
4050         if (!wait_loop) {
4051                 e_err(drv, "RXDCTL.ENABLE on Rx queue %d not set within "
4052                       "the polling period\n", reg_idx);
4053         }
4054 }
4055
4056 void ixgbe_configure_rx_ring(struct ixgbe_adapter *adapter,
4057                              struct ixgbe_ring *ring)
4058 {
4059         struct ixgbe_hw *hw = &adapter->hw;
4060         union ixgbe_adv_rx_desc *rx_desc;
4061         u64 rdba = ring->dma;
4062         u32 rxdctl;
4063         u8 reg_idx = ring->reg_idx;
4064
4065         xdp_rxq_info_unreg_mem_model(&ring->xdp_rxq);
4066         ring->xsk_pool = ixgbe_xsk_pool(adapter, ring);
4067         if (ring->xsk_pool) {
4068                 WARN_ON(xdp_rxq_info_reg_mem_model(&ring->xdp_rxq,
4069                                                    MEM_TYPE_XSK_BUFF_POOL,
4070                                                    NULL));
4071                 xsk_pool_set_rxq_info(ring->xsk_pool, &ring->xdp_rxq);
4072         } else {
4073                 WARN_ON(xdp_rxq_info_reg_mem_model(&ring->xdp_rxq,
4074                                                    MEM_TYPE_PAGE_SHARED, NULL));
4075         }
4076
4077         /* disable queue to avoid use of these values while updating state */
4078         rxdctl = IXGBE_READ_REG(hw, IXGBE_RXDCTL(reg_idx));
4079         rxdctl &= ~IXGBE_RXDCTL_ENABLE;
4080
4081         /* write value back with RXDCTL.ENABLE bit cleared */
4082         IXGBE_WRITE_REG(hw, IXGBE_RXDCTL(reg_idx), rxdctl);
4083         IXGBE_WRITE_FLUSH(hw);
4084
4085         IXGBE_WRITE_REG(hw, IXGBE_RDBAL(reg_idx), (rdba & DMA_BIT_MASK(32)));
4086         IXGBE_WRITE_REG(hw, IXGBE_RDBAH(reg_idx), (rdba >> 32));
4087         IXGBE_WRITE_REG(hw, IXGBE_RDLEN(reg_idx),
4088                         ring->count * sizeof(union ixgbe_adv_rx_desc));
4089         /* Force flushing of IXGBE_RDLEN to prevent MDD */
4090         IXGBE_WRITE_FLUSH(hw);
4091
4092         IXGBE_WRITE_REG(hw, IXGBE_RDH(reg_idx), 0);
4093         IXGBE_WRITE_REG(hw, IXGBE_RDT(reg_idx), 0);
4094         ring->tail = adapter->io_addr + IXGBE_RDT(reg_idx);
4095
4096         ixgbe_configure_srrctl(adapter, ring);
4097         ixgbe_configure_rscctl(adapter, ring);
4098
4099         if (hw->mac.type == ixgbe_mac_82598EB) {
4100                 /*
4101                  * enable cache line friendly hardware writes:
4102                  * PTHRESH=32 descriptors (half the internal cache),
4103                  * this also removes ugly rx_no_buffer_count increment
4104                  * HTHRESH=4 descriptors (to minimize latency on fetch)
4105                  * WTHRESH=8 burst writeback up to two cache lines
4106                  */
4107                 rxdctl &= ~0x3FFFFF;
4108                 rxdctl |=  0x080420;
4109 #if (PAGE_SIZE < 8192)
4110         /* RXDCTL.RLPML does not work on 82599 */
4111         } else if (hw->mac.type != ixgbe_mac_82599EB) {
4112                 rxdctl &= ~(IXGBE_RXDCTL_RLPMLMASK |
4113                             IXGBE_RXDCTL_RLPML_EN);
4114
4115                 /* Limit the maximum frame size so we don't overrun the skb.
4116                  * This can happen in SRIOV mode when the MTU of the VF is
4117                  * higher than the MTU of the PF.
4118                  */
4119                 if (ring_uses_build_skb(ring) &&
4120                     !test_bit(__IXGBE_RX_3K_BUFFER, &ring->state))
4121                         rxdctl |= IXGBE_MAX_2K_FRAME_BUILD_SKB |
4122                                   IXGBE_RXDCTL_RLPML_EN;
4123 #endif
4124         }
4125
4126         ring->rx_offset = ixgbe_rx_offset(ring);
4127
4128         if (ring->xsk_pool && hw->mac.type != ixgbe_mac_82599EB) {
4129                 u32 xsk_buf_len = xsk_pool_get_rx_frame_size(ring->xsk_pool);
4130
4131                 rxdctl &= ~(IXGBE_RXDCTL_RLPMLMASK |
4132                             IXGBE_RXDCTL_RLPML_EN);
4133                 rxdctl |= xsk_buf_len | IXGBE_RXDCTL_RLPML_EN;
4134
4135                 ring->rx_buf_len = xsk_buf_len;
4136         }
4137
4138         /* initialize rx_buffer_info */
4139         memset(ring->rx_buffer_info, 0,
4140                sizeof(struct ixgbe_rx_buffer) * ring->count);
4141
4142         /* initialize Rx descriptor 0 */
4143         rx_desc = IXGBE_RX_DESC(ring, 0);
4144         rx_desc->wb.upper.length = 0;
4145
4146         /* enable receive descriptor ring */
4147         rxdctl |= IXGBE_RXDCTL_ENABLE;
4148         IXGBE_WRITE_REG(hw, IXGBE_RXDCTL(reg_idx), rxdctl);
4149
4150         ixgbe_rx_desc_queue_enable(adapter, ring);
4151         if (ring->xsk_pool)
4152                 ixgbe_alloc_rx_buffers_zc(ring, ixgbe_desc_unused(ring));
4153         else
4154                 ixgbe_alloc_rx_buffers(ring, ixgbe_desc_unused(ring));
4155 }
4156
4157 static void ixgbe_setup_psrtype(struct ixgbe_adapter *adapter)
4158 {
4159         struct ixgbe_hw *hw = &adapter->hw;
4160         int rss_i = adapter->ring_feature[RING_F_RSS].indices;
4161         u16 pool = adapter->num_rx_pools;
4162
4163         /* PSRTYPE must be initialized in non 82598 adapters */
4164         u32 psrtype = IXGBE_PSRTYPE_TCPHDR |
4165                       IXGBE_PSRTYPE_UDPHDR |
4166                       IXGBE_PSRTYPE_IPV4HDR |
4167                       IXGBE_PSRTYPE_L2HDR |
4168                       IXGBE_PSRTYPE_IPV6HDR;
4169
4170         if (hw->mac.type == ixgbe_mac_82598EB)
4171                 return;
4172
4173         if (rss_i > 3)
4174                 psrtype |= 2u << 29;
4175         else if (rss_i > 1)
4176                 psrtype |= 1u << 29;
4177
4178         while (pool--)
4179                 IXGBE_WRITE_REG(hw, IXGBE_PSRTYPE(VMDQ_P(pool)), psrtype);
4180 }
4181
4182 static void ixgbe_configure_virtualization(struct ixgbe_adapter *adapter)
4183 {
4184         struct ixgbe_hw *hw = &adapter->hw;
4185         u16 pool = adapter->num_rx_pools;
4186         u32 reg_offset, vf_shift, vmolr;
4187         u32 gcr_ext, vmdctl;
4188         int i;
4189
4190         if (!(adapter->flags & IXGBE_FLAG_SRIOV_ENABLED))
4191                 return;
4192
4193         vmdctl = IXGBE_READ_REG(hw, IXGBE_VT_CTL);
4194         vmdctl |= IXGBE_VMD_CTL_VMDQ_EN;
4195         vmdctl &= ~IXGBE_VT_CTL_POOL_MASK;
4196         vmdctl |= VMDQ_P(0) << IXGBE_VT_CTL_POOL_SHIFT;
4197         vmdctl |= IXGBE_VT_CTL_REPLEN;
4198         IXGBE_WRITE_REG(hw, IXGBE_VT_CTL, vmdctl);
4199
4200         /* accept untagged packets until a vlan tag is
4201          * specifically set for the VMDQ queue/pool
4202          */
4203         vmolr = IXGBE_VMOLR_AUPE;
4204         while (pool--)
4205                 IXGBE_WRITE_REG(hw, IXGBE_VMOLR(VMDQ_P(pool)), vmolr);
4206
4207         vf_shift = VMDQ_P(0) % 32;
4208         reg_offset = (VMDQ_P(0) >= 32) ? 1 : 0;
4209
4210         /* Enable only the PF's pool for Tx/Rx */
4211         IXGBE_WRITE_REG(hw, IXGBE_VFRE(reg_offset), GENMASK(31, vf_shift));
4212         IXGBE_WRITE_REG(hw, IXGBE_VFRE(reg_offset ^ 1), reg_offset - 1);
4213         IXGBE_WRITE_REG(hw, IXGBE_VFTE(reg_offset), GENMASK(31, vf_shift));
4214         IXGBE_WRITE_REG(hw, IXGBE_VFTE(reg_offset ^ 1), reg_offset - 1);
4215         if (adapter->bridge_mode == BRIDGE_MODE_VEB)
4216                 IXGBE_WRITE_REG(hw, IXGBE_PFDTXGSWC, IXGBE_PFDTXGSWC_VT_LBEN);
4217
4218         /* Map PF MAC address in RAR Entry 0 to first pool following VFs */
4219         hw->mac.ops.set_vmdq(hw, 0, VMDQ_P(0));
4220
4221         /* clear VLAN promisc flag so VFTA will be updated if necessary */
4222         adapter->flags2 &= ~IXGBE_FLAG2_VLAN_PROMISC;
4223
4224         /*
4225          * Set up VF register offsets for selected VT Mode,
4226          * i.e. 32 or 64 VFs for SR-IOV
4227          */
4228         switch (adapter->ring_feature[RING_F_VMDQ].mask) {
4229         case IXGBE_82599_VMDQ_8Q_MASK:
4230                 gcr_ext = IXGBE_GCR_EXT_VT_MODE_16;
4231                 break;
4232         case IXGBE_82599_VMDQ_4Q_MASK:
4233                 gcr_ext = IXGBE_GCR_EXT_VT_MODE_32;
4234                 break;
4235         default:
4236                 gcr_ext = IXGBE_GCR_EXT_VT_MODE_64;
4237                 break;
4238         }
4239
4240         IXGBE_WRITE_REG(hw, IXGBE_GCR_EXT, gcr_ext);
4241
4242         for (i = 0; i < adapter->num_vfs; i++) {
4243                 /* configure spoof checking */
4244                 ixgbe_ndo_set_vf_spoofchk(adapter->netdev, i,
4245                                           adapter->vfinfo[i].spoofchk_enabled);
4246
4247                 /* Enable/Disable RSS query feature  */
4248                 ixgbe_ndo_set_vf_rss_query_en(adapter->netdev, i,
4249                                           adapter->vfinfo[i].rss_query_enabled);
4250         }
4251 }
4252
4253 static void ixgbe_set_rx_buffer_len(struct ixgbe_adapter *adapter)
4254 {
4255         struct ixgbe_hw *hw = &adapter->hw;
4256         struct net_device *netdev = adapter->netdev;
4257         int max_frame = netdev->mtu + ETH_HLEN + ETH_FCS_LEN;
4258         struct ixgbe_ring *rx_ring;
4259         int i;
4260         u32 mhadd, hlreg0;
4261
4262 #ifdef IXGBE_FCOE
4263         /* adjust max frame to be able to do baby jumbo for FCoE */
4264         if ((adapter->flags & IXGBE_FLAG_FCOE_ENABLED) &&
4265             (max_frame < IXGBE_FCOE_JUMBO_FRAME_SIZE))
4266                 max_frame = IXGBE_FCOE_JUMBO_FRAME_SIZE;
4267
4268 #endif /* IXGBE_FCOE */
4269
4270         /* adjust max frame to be at least the size of a standard frame */
4271         if (max_frame < (ETH_FRAME_LEN + ETH_FCS_LEN))
4272                 max_frame = (ETH_FRAME_LEN + ETH_FCS_LEN);
4273
4274         mhadd = IXGBE_READ_REG(hw, IXGBE_MHADD);
4275         if (max_frame != (mhadd >> IXGBE_MHADD_MFS_SHIFT)) {
4276                 mhadd &= ~IXGBE_MHADD_MFS_MASK;
4277                 mhadd |= max_frame << IXGBE_MHADD_MFS_SHIFT;
4278
4279                 IXGBE_WRITE_REG(hw, IXGBE_MHADD, mhadd);
4280         }
4281
4282         hlreg0 = IXGBE_READ_REG(hw, IXGBE_HLREG0);
4283         /* set jumbo enable since MHADD.MFS is keeping size locked at max_frame */
4284         hlreg0 |= IXGBE_HLREG0_JUMBOEN;
4285         IXGBE_WRITE_REG(hw, IXGBE_HLREG0, hlreg0);
4286
4287         /*
4288          * Setup the HW Rx Head and Tail Descriptor Pointers and
4289          * the Base and Length of the Rx Descriptor Ring
4290          */
4291         for (i = 0; i < adapter->num_rx_queues; i++) {
4292                 rx_ring = adapter->rx_ring[i];
4293
4294                 clear_ring_rsc_enabled(rx_ring);
4295                 clear_bit(__IXGBE_RX_3K_BUFFER, &rx_ring->state);
4296                 clear_bit(__IXGBE_RX_BUILD_SKB_ENABLED, &rx_ring->state);
4297
4298                 if (adapter->flags2 & IXGBE_FLAG2_RSC_ENABLED)
4299                         set_ring_rsc_enabled(rx_ring);
4300
4301                 if (test_bit(__IXGBE_RX_FCOE, &rx_ring->state))
4302                         set_bit(__IXGBE_RX_3K_BUFFER, &rx_ring->state);
4303
4304                 if (adapter->flags2 & IXGBE_FLAG2_RX_LEGACY)
4305                         continue;
4306
4307                 set_bit(__IXGBE_RX_BUILD_SKB_ENABLED, &rx_ring->state);
4308
4309 #if (PAGE_SIZE < 8192)
4310                 if (adapter->flags2 & IXGBE_FLAG2_RSC_ENABLED)
4311                         set_bit(__IXGBE_RX_3K_BUFFER, &rx_ring->state);
4312
4313                 if (IXGBE_2K_TOO_SMALL_WITH_PADDING ||
4314                     (max_frame > (ETH_FRAME_LEN + ETH_FCS_LEN)))
4315                         set_bit(__IXGBE_RX_3K_BUFFER, &rx_ring->state);
4316 #endif
4317         }
4318 }
4319
4320 static void ixgbe_setup_rdrxctl(struct ixgbe_adapter *adapter)
4321 {
4322         struct ixgbe_hw *hw = &adapter->hw;
4323         u32 rdrxctl = IXGBE_READ_REG(hw, IXGBE_RDRXCTL);
4324
4325         switch (hw->mac.type) {
4326         case ixgbe_mac_82598EB:
4327                 /*
4328                  * For VMDq support of different descriptor types or
4329                  * buffer sizes through the use of multiple SRRCTL
4330                  * registers, RDRXCTL.MVMEN must be set to 1
4331                  *
4332                  * also, the manual doesn't mention it clearly but DCA hints
4333                  * will only use queue 0's tags unless this bit is set.  Side
4334                  * effects of setting this bit are only that SRRCTL must be
4335                  * fully programmed [0..15]
4336                  */
4337                 rdrxctl |= IXGBE_RDRXCTL_MVMEN;
4338                 break;
4339         case ixgbe_mac_X550:
4340         case ixgbe_mac_X550EM_x:
4341         case ixgbe_mac_x550em_a:
4342                 if (adapter->num_vfs)
4343                         rdrxctl |= IXGBE_RDRXCTL_PSP;
4344                 fallthrough;
4345         case ixgbe_mac_82599EB:
4346         case ixgbe_mac_X540:
4347                 /* Disable RSC for ACK packets */
4348                 IXGBE_WRITE_REG(hw, IXGBE_RSCDBU,
4349                    (IXGBE_RSCDBU_RSCACKDIS | IXGBE_READ_REG(hw, IXGBE_RSCDBU)));
4350                 rdrxctl &= ~IXGBE_RDRXCTL_RSCFRSTSIZE;
4351                 /* hardware requires some bits to be set by default */
4352                 rdrxctl |= (IXGBE_RDRXCTL_RSCACKC | IXGBE_RDRXCTL_FCOE_WRFIX);
4353                 rdrxctl |= IXGBE_RDRXCTL_CRCSTRIP;
4354                 break;
4355         default:
4356                 /* We should do nothing since we don't know this hardware */
4357                 return;
4358         }
4359
4360         IXGBE_WRITE_REG(hw, IXGBE_RDRXCTL, rdrxctl);
4361 }
4362
4363 /**
4364  * ixgbe_configure_rx - Configure 8259x Receive Unit after Reset
4365  * @adapter: board private structure
4366  *
4367  * Configure the Rx unit of the MAC after a reset.
4368  **/
4369 static void ixgbe_configure_rx(struct ixgbe_adapter *adapter)
4370 {
4371         struct ixgbe_hw *hw = &adapter->hw;
4372         int i;
4373         u32 rxctrl, rfctl;
4374
4375         /* disable receives while setting up the descriptors */
4376         hw->mac.ops.disable_rx(hw);
4377
4378         ixgbe_setup_psrtype(adapter);
4379         ixgbe_setup_rdrxctl(adapter);
4380
4381         /* RSC Setup */
4382         rfctl = IXGBE_READ_REG(hw, IXGBE_RFCTL);
4383         rfctl &= ~IXGBE_RFCTL_RSC_DIS;
4384         if (!(adapter->flags2 & IXGBE_FLAG2_RSC_ENABLED))
4385                 rfctl |= IXGBE_RFCTL_RSC_DIS;
4386
4387         /* disable NFS filtering */
4388         rfctl |= (IXGBE_RFCTL_NFSW_DIS | IXGBE_RFCTL_NFSR_DIS);
4389         IXGBE_WRITE_REG(hw, IXGBE_RFCTL, rfctl);
4390
4391         /* Program registers for the distribution of queues */
4392         ixgbe_setup_mrqc(adapter);
4393
4394         /* set_rx_buffer_len must be called before ring initialization */
4395         ixgbe_set_rx_buffer_len(adapter);
4396
4397         /*
4398          * Setup the HW Rx Head and Tail Descriptor Pointers and
4399          * the Base and Length of the Rx Descriptor Ring
4400          */
4401         for (i = 0; i < adapter->num_rx_queues; i++)
4402                 ixgbe_configure_rx_ring(adapter, adapter->rx_ring[i]);
4403
4404         rxctrl = IXGBE_READ_REG(hw, IXGBE_RXCTRL);
4405         /* disable drop enable for 82598 parts */
4406         if (hw->mac.type == ixgbe_mac_82598EB)
4407                 rxctrl |= IXGBE_RXCTRL_DMBYPS;
4408
4409         /* enable all receives */
4410         rxctrl |= IXGBE_RXCTRL_RXEN;
4411         hw->mac.ops.enable_rx_dma(hw, rxctrl);
4412 }
4413
4414 static int ixgbe_vlan_rx_add_vid(struct net_device *netdev,
4415                                  __be16 proto, u16 vid)
4416 {
4417         struct ixgbe_adapter *adapter = netdev_priv(netdev);
4418         struct ixgbe_hw *hw = &adapter->hw;
4419
4420         /* add VID to filter table */
4421         if (!vid || !(adapter->flags2 & IXGBE_FLAG2_VLAN_PROMISC))
4422                 hw->mac.ops.set_vfta(&adapter->hw, vid, VMDQ_P(0), true, !!vid);
4423
4424         set_bit(vid, adapter->active_vlans);
4425
4426         return 0;
4427 }
4428
4429 static int ixgbe_find_vlvf_entry(struct ixgbe_hw *hw, u32 vlan)
4430 {
4431         u32 vlvf;
4432         int idx;
4433
4434         /* short cut the special case */
4435         if (vlan == 0)
4436                 return 0;
4437
4438         /* Search for the vlan id in the VLVF entries */
4439         for (idx = IXGBE_VLVF_ENTRIES; --idx;) {
4440                 vlvf = IXGBE_READ_REG(hw, IXGBE_VLVF(idx));
4441                 if ((vlvf & VLAN_VID_MASK) == vlan)
4442                         break;
4443         }
4444
4445         return idx;
4446 }
4447
4448 void ixgbe_update_pf_promisc_vlvf(struct ixgbe_adapter *adapter, u32 vid)
4449 {
4450         struct ixgbe_hw *hw = &adapter->hw;
4451         u32 bits, word;
4452         int idx;
4453
4454         idx = ixgbe_find_vlvf_entry(hw, vid);
4455         if (!idx)
4456                 return;
4457
4458         /* See if any other pools are set for this VLAN filter
4459          * entry other than the PF.
4460          */
4461         word = idx * 2 + (VMDQ_P(0) / 32);
4462         bits = ~BIT(VMDQ_P(0) % 32);
4463         bits &= IXGBE_READ_REG(hw, IXGBE_VLVFB(word));
4464
4465         /* Disable the filter so this falls into the default pool. */
4466         if (!bits && !IXGBE_READ_REG(hw, IXGBE_VLVFB(word ^ 1))) {
4467                 if (!(adapter->flags2 & IXGBE_FLAG2_VLAN_PROMISC))
4468                         IXGBE_WRITE_REG(hw, IXGBE_VLVFB(word), 0);
4469                 IXGBE_WRITE_REG(hw, IXGBE_VLVF(idx), 0);
4470         }
4471 }
4472
4473 static int ixgbe_vlan_rx_kill_vid(struct net_device *netdev,
4474                                   __be16 proto, u16 vid)
4475 {
4476         struct ixgbe_adapter *adapter = netdev_priv(netdev);
4477         struct ixgbe_hw *hw = &adapter->hw;
4478
4479         /* remove VID from filter table */
4480         if (vid && !(adapter->flags2 & IXGBE_FLAG2_VLAN_PROMISC))
4481                 hw->mac.ops.set_vfta(hw, vid, VMDQ_P(0), false, true);
4482
4483         clear_bit(vid, adapter->active_vlans);
4484
4485         return 0;
4486 }
4487
4488 /**
4489  * ixgbe_vlan_strip_disable - helper to disable hw vlan stripping
4490  * @adapter: driver data
4491  */
4492 static void ixgbe_vlan_strip_disable(struct ixgbe_adapter *adapter)
4493 {
4494         struct ixgbe_hw *hw = &adapter->hw;
4495         u32 vlnctrl;
4496         int i, j;
4497
4498         switch (hw->mac.type) {
4499         case ixgbe_mac_82598EB:
4500                 vlnctrl = IXGBE_READ_REG(hw, IXGBE_VLNCTRL);
4501                 vlnctrl &= ~IXGBE_VLNCTRL_VME;
4502                 IXGBE_WRITE_REG(hw, IXGBE_VLNCTRL, vlnctrl);
4503                 break;
4504         case ixgbe_mac_82599EB:
4505         case ixgbe_mac_X540:
4506         case ixgbe_mac_X550:
4507         case ixgbe_mac_X550EM_x:
4508         case ixgbe_mac_x550em_a:
4509                 for (i = 0; i < adapter->num_rx_queues; i++) {
4510                         struct ixgbe_ring *ring = adapter->rx_ring[i];
4511
4512                         if (!netif_is_ixgbe(ring->netdev))
4513                                 continue;
4514
4515                         j = ring->reg_idx;
4516                         vlnctrl = IXGBE_READ_REG(hw, IXGBE_RXDCTL(j));
4517                         vlnctrl &= ~IXGBE_RXDCTL_VME;
4518                         IXGBE_WRITE_REG(hw, IXGBE_RXDCTL(j), vlnctrl);
4519                 }
4520                 break;
4521         default:
4522                 break;
4523         }
4524 }
4525
4526 /**
4527  * ixgbe_vlan_strip_enable - helper to enable hw vlan stripping
4528  * @adapter: driver data
4529  */
4530 static void ixgbe_vlan_strip_enable(struct ixgbe_adapter *adapter)
4531 {
4532         struct ixgbe_hw *hw = &adapter->hw;
4533         u32 vlnctrl;
4534         int i, j;
4535
4536         switch (hw->mac.type) {
4537         case ixgbe_mac_82598EB:
4538                 vlnctrl = IXGBE_READ_REG(hw, IXGBE_VLNCTRL);
4539                 vlnctrl |= IXGBE_VLNCTRL_VME;
4540                 IXGBE_WRITE_REG(hw, IXGBE_VLNCTRL, vlnctrl);
4541                 break;
4542         case ixgbe_mac_82599EB:
4543         case ixgbe_mac_X540:
4544         case ixgbe_mac_X550:
4545         case ixgbe_mac_X550EM_x:
4546         case ixgbe_mac_x550em_a:
4547                 for (i = 0; i < adapter->num_rx_queues; i++) {
4548                         struct ixgbe_ring *ring = adapter->rx_ring[i];
4549
4550                         if (!netif_is_ixgbe(ring->netdev))
4551                                 continue;
4552
4553                         j = ring->reg_idx;
4554                         vlnctrl = IXGBE_READ_REG(hw, IXGBE_RXDCTL(j));
4555                         vlnctrl |= IXGBE_RXDCTL_VME;
4556                         IXGBE_WRITE_REG(hw, IXGBE_RXDCTL(j), vlnctrl);
4557                 }
4558                 break;
4559         default:
4560                 break;
4561         }
4562 }
4563
4564 static void ixgbe_vlan_promisc_enable(struct ixgbe_adapter *adapter)
4565 {
4566         struct ixgbe_hw *hw = &adapter->hw;
4567         u32 vlnctrl, i;
4568
4569         vlnctrl = IXGBE_READ_REG(hw, IXGBE_VLNCTRL);
4570
4571         if (adapter->flags & IXGBE_FLAG_VMDQ_ENABLED) {
4572         /* For VMDq and SR-IOV we must leave VLAN filtering enabled */
4573                 vlnctrl |= IXGBE_VLNCTRL_VFE;
4574                 IXGBE_WRITE_REG(hw, IXGBE_VLNCTRL, vlnctrl);
4575         } else {
4576                 vlnctrl &= ~IXGBE_VLNCTRL_VFE;
4577                 IXGBE_WRITE_REG(hw, IXGBE_VLNCTRL, vlnctrl);
4578                 return;
4579         }
4580
4581         /* Nothing to do for 82598 */
4582         if (hw->mac.type == ixgbe_mac_82598EB)
4583                 return;
4584
4585         /* We are already in VLAN promisc, nothing to do */
4586         if (adapter->flags2 & IXGBE_FLAG2_VLAN_PROMISC)
4587                 return;
4588
4589         /* Set flag so we don't redo unnecessary work */
4590         adapter->flags2 |= IXGBE_FLAG2_VLAN_PROMISC;
4591
4592         /* Add PF to all active pools */
4593         for (i = IXGBE_VLVF_ENTRIES; --i;) {
4594                 u32 reg_offset = IXGBE_VLVFB(i * 2 + VMDQ_P(0) / 32);
4595                 u32 vlvfb = IXGBE_READ_REG(hw, reg_offset);
4596
4597                 vlvfb |= BIT(VMDQ_P(0) % 32);
4598                 IXGBE_WRITE_REG(hw, reg_offset, vlvfb);
4599         }
4600
4601         /* Set all bits in the VLAN filter table array */
4602         for (i = hw->mac.vft_size; i--;)
4603                 IXGBE_WRITE_REG(hw, IXGBE_VFTA(i), ~0U);
4604 }
4605
4606 #define VFTA_BLOCK_SIZE 8
4607 static void ixgbe_scrub_vfta(struct ixgbe_adapter *adapter, u32 vfta_offset)
4608 {
4609         struct ixgbe_hw *hw = &adapter->hw;
4610         u32 vfta[VFTA_BLOCK_SIZE] = { 0 };
4611         u32 vid_start = vfta_offset * 32;
4612         u32 vid_end = vid_start + (VFTA_BLOCK_SIZE * 32);
4613         u32 i, vid, word, bits;
4614
4615         for (i = IXGBE_VLVF_ENTRIES; --i;) {
4616                 u32 vlvf = IXGBE_READ_REG(hw, IXGBE_VLVF(i));
4617
4618                 /* pull VLAN ID from VLVF */
4619                 vid = vlvf & VLAN_VID_MASK;
4620
4621                 /* only concern outselves with a certain range */
4622                 if (vid < vid_start || vid >= vid_end)
4623                         continue;
4624
4625                 if (vlvf) {
4626                         /* record VLAN ID in VFTA */
4627                         vfta[(vid - vid_start) / 32] |= BIT(vid % 32);
4628
4629                         /* if PF is part of this then continue */
4630                         if (test_bit(vid, adapter->active_vlans))
4631                                 continue;
4632                 }
4633
4634                 /* remove PF from the pool */
4635                 word = i * 2 + VMDQ_P(0) / 32;
4636                 bits = ~BIT(VMDQ_P(0) % 32);
4637                 bits &= IXGBE_READ_REG(hw, IXGBE_VLVFB(word));
4638                 IXGBE_WRITE_REG(hw, IXGBE_VLVFB(word), bits);
4639         }
4640
4641         /* extract values from active_vlans and write back to VFTA */
4642         for (i = VFTA_BLOCK_SIZE; i--;) {
4643                 vid = (vfta_offset + i) * 32;
4644                 word = vid / BITS_PER_LONG;
4645                 bits = vid % BITS_PER_LONG;
4646
4647                 vfta[i] |= adapter->active_vlans[word] >> bits;
4648
4649                 IXGBE_WRITE_REG(hw, IXGBE_VFTA(vfta_offset + i), vfta[i]);
4650         }
4651 }
4652
4653 static void ixgbe_vlan_promisc_disable(struct ixgbe_adapter *adapter)
4654 {
4655         struct ixgbe_hw *hw = &adapter->hw;
4656         u32 vlnctrl, i;
4657
4658         /* Set VLAN filtering to enabled */
4659         vlnctrl = IXGBE_READ_REG(hw, IXGBE_VLNCTRL);
4660         vlnctrl |= IXGBE_VLNCTRL_VFE;
4661         IXGBE_WRITE_REG(hw, IXGBE_VLNCTRL, vlnctrl);
4662
4663         if (!(adapter->flags & IXGBE_FLAG_VMDQ_ENABLED) ||
4664             hw->mac.type == ixgbe_mac_82598EB)
4665                 return;
4666
4667         /* We are not in VLAN promisc, nothing to do */
4668         if (!(adapter->flags2 & IXGBE_FLAG2_VLAN_PROMISC))
4669                 return;
4670
4671         /* Set flag so we don't redo unnecessary work */
4672         adapter->flags2 &= ~IXGBE_FLAG2_VLAN_PROMISC;
4673
4674         for (i = 0; i < hw->mac.vft_size; i += VFTA_BLOCK_SIZE)
4675                 ixgbe_scrub_vfta(adapter, i);
4676 }
4677
4678 static void ixgbe_restore_vlan(struct ixgbe_adapter *adapter)
4679 {
4680         u16 vid = 1;
4681
4682         ixgbe_vlan_rx_add_vid(adapter->netdev, htons(ETH_P_8021Q), 0);
4683
4684         for_each_set_bit_from(vid, adapter->active_vlans, VLAN_N_VID)
4685                 ixgbe_vlan_rx_add_vid(adapter->netdev, htons(ETH_P_8021Q), vid);
4686 }
4687
4688 /**
4689  * ixgbe_write_mc_addr_list - write multicast addresses to MTA
4690  * @netdev: network interface device structure
4691  *
4692  * Writes multicast address list to the MTA hash table.
4693  * Returns: -ENOMEM on failure
4694  *                0 on no addresses written
4695  *                X on writing X addresses to MTA
4696  **/
4697 static int ixgbe_write_mc_addr_list(struct net_device *netdev)
4698 {
4699         struct ixgbe_adapter *adapter = netdev_priv(netdev);
4700         struct ixgbe_hw *hw = &adapter->hw;
4701
4702         if (!netif_running(netdev))
4703                 return 0;
4704
4705         if (hw->mac.ops.update_mc_addr_list)
4706                 hw->mac.ops.update_mc_addr_list(hw, netdev);
4707         else
4708                 return -ENOMEM;
4709
4710 #ifdef CONFIG_PCI_IOV
4711         ixgbe_restore_vf_multicasts(adapter);
4712 #endif
4713
4714         return netdev_mc_count(netdev);
4715 }
4716
4717 #ifdef CONFIG_PCI_IOV
4718 void ixgbe_full_sync_mac_table(struct ixgbe_adapter *adapter)
4719 {
4720         struct ixgbe_mac_addr *mac_table = &adapter->mac_table[0];
4721         struct ixgbe_hw *hw = &adapter->hw;
4722         int i;
4723
4724         for (i = 0; i < hw->mac.num_rar_entries; i++, mac_table++) {
4725                 mac_table->state &= ~IXGBE_MAC_STATE_MODIFIED;
4726
4727                 if (mac_table->state & IXGBE_MAC_STATE_IN_USE)
4728                         hw->mac.ops.set_rar(hw, i,
4729                                             mac_table->addr,
4730                                             mac_table->pool,
4731                                             IXGBE_RAH_AV);
4732                 else
4733                         hw->mac.ops.clear_rar(hw, i);
4734         }
4735 }
4736
4737 #endif
4738 static void ixgbe_sync_mac_table(struct ixgbe_adapter *adapter)
4739 {
4740         struct ixgbe_mac_addr *mac_table = &adapter->mac_table[0];
4741         struct ixgbe_hw *hw = &adapter->hw;
4742         int i;
4743
4744         for (i = 0; i < hw->mac.num_rar_entries; i++, mac_table++) {
4745                 if (!(mac_table->state & IXGBE_MAC_STATE_MODIFIED))
4746                         continue;
4747
4748                 mac_table->state &= ~IXGBE_MAC_STATE_MODIFIED;
4749
4750                 if (mac_table->state & IXGBE_MAC_STATE_IN_USE)
4751                         hw->mac.ops.set_rar(hw, i,
4752                                             mac_table->addr,
4753                                             mac_table->pool,
4754                                             IXGBE_RAH_AV);
4755                 else
4756                         hw->mac.ops.clear_rar(hw, i);
4757         }
4758 }
4759
4760 static void ixgbe_flush_sw_mac_table(struct ixgbe_adapter *adapter)
4761 {
4762         struct ixgbe_mac_addr *mac_table = &adapter->mac_table[0];
4763         struct ixgbe_hw *hw = &adapter->hw;
4764         int i;
4765
4766         for (i = 0; i < hw->mac.num_rar_entries; i++, mac_table++) {
4767                 mac_table->state |= IXGBE_MAC_STATE_MODIFIED;
4768                 mac_table->state &= ~IXGBE_MAC_STATE_IN_USE;
4769         }
4770
4771         ixgbe_sync_mac_table(adapter);
4772 }
4773
4774 static int ixgbe_available_rars(struct ixgbe_adapter *adapter, u16 pool)
4775 {
4776         struct ixgbe_mac_addr *mac_table = &adapter->mac_table[0];
4777         struct ixgbe_hw *hw = &adapter->hw;
4778         int i, count = 0;
4779
4780         for (i = 0; i < hw->mac.num_rar_entries; i++, mac_table++) {
4781                 /* do not count default RAR as available */
4782                 if (mac_table->state & IXGBE_MAC_STATE_DEFAULT)
4783                         continue;
4784
4785                 /* only count unused and addresses that belong to us */
4786                 if (mac_table->state & IXGBE_MAC_STATE_IN_USE) {
4787                         if (mac_table->pool != pool)
4788                                 continue;
4789                 }
4790
4791                 count++;
4792         }
4793
4794         return count;
4795 }
4796
4797 /* this function destroys the first RAR entry */
4798 static void ixgbe_mac_set_default_filter(struct ixgbe_adapter *adapter)
4799 {
4800         struct ixgbe_mac_addr *mac_table = &adapter->mac_table[0];
4801         struct ixgbe_hw *hw = &adapter->hw;
4802
4803         memcpy(&mac_table->addr, hw->mac.addr, ETH_ALEN);
4804         mac_table->pool = VMDQ_P(0);
4805
4806         mac_table->state = IXGBE_MAC_STATE_DEFAULT | IXGBE_MAC_STATE_IN_USE;
4807
4808         hw->mac.ops.set_rar(hw, 0, mac_table->addr, mac_table->pool,
4809                             IXGBE_RAH_AV);
4810 }
4811
4812 int ixgbe_add_mac_filter(struct ixgbe_adapter *adapter,
4813                          const u8 *addr, u16 pool)
4814 {
4815         struct ixgbe_mac_addr *mac_table = &adapter->mac_table[0];
4816         struct ixgbe_hw *hw = &adapter->hw;
4817         int i;
4818
4819         if (is_zero_ether_addr(addr))
4820                 return -EINVAL;
4821
4822         for (i = 0; i < hw->mac.num_rar_entries; i++, mac_table++) {
4823                 if (mac_table->state & IXGBE_MAC_STATE_IN_USE)
4824                         continue;
4825
4826                 ether_addr_copy(mac_table->addr, addr);
4827                 mac_table->pool = pool;
4828
4829                 mac_table->state |= IXGBE_MAC_STATE_MODIFIED |
4830                                     IXGBE_MAC_STATE_IN_USE;
4831
4832                 ixgbe_sync_mac_table(adapter);
4833
4834                 return i;
4835         }
4836
4837         return -ENOMEM;
4838 }
4839
4840 int ixgbe_del_mac_filter(struct ixgbe_adapter *adapter,
4841                          const u8 *addr, u16 pool)
4842 {
4843         struct ixgbe_mac_addr *mac_table = &adapter->mac_table[0];
4844         struct ixgbe_hw *hw = &adapter->hw;
4845         int i;
4846
4847         if (is_zero_ether_addr(addr))
4848                 return -EINVAL;
4849
4850         /* search table for addr, if found clear IN_USE flag and sync */
4851         for (i = 0; i < hw->mac.num_rar_entries; i++, mac_table++) {
4852                 /* we can only delete an entry if it is in use */
4853                 if (!(mac_table->state & IXGBE_MAC_STATE_IN_USE))
4854                         continue;
4855                 /* we only care about entries that belong to the given pool */
4856                 if (mac_table->pool != pool)
4857                         continue;
4858                 /* we only care about a specific MAC address */
4859                 if (!ether_addr_equal(addr, mac_table->addr))
4860                         continue;
4861
4862                 mac_table->state |= IXGBE_MAC_STATE_MODIFIED;
4863                 mac_table->state &= ~IXGBE_MAC_STATE_IN_USE;
4864
4865                 ixgbe_sync_mac_table(adapter);
4866
4867                 return 0;
4868         }
4869
4870         return -ENOMEM;
4871 }
4872
4873 static int ixgbe_uc_sync(struct net_device *netdev, const unsigned char *addr)
4874 {
4875         struct ixgbe_adapter *adapter = netdev_priv(netdev);
4876         int ret;
4877
4878         ret = ixgbe_add_mac_filter(adapter, addr, VMDQ_P(0));
4879
4880         return min_t(int, ret, 0);
4881 }
4882
4883 static int ixgbe_uc_unsync(struct net_device *netdev, const unsigned char *addr)
4884 {
4885         struct ixgbe_adapter *adapter = netdev_priv(netdev);
4886
4887         ixgbe_del_mac_filter(adapter, addr, VMDQ_P(0));
4888
4889         return 0;
4890 }
4891
4892 /**
4893  * ixgbe_set_rx_mode - Unicast, Multicast and Promiscuous mode set
4894  * @netdev: network interface device structure
4895  *
4896  * The set_rx_method entry point is called whenever the unicast/multicast
4897  * address list or the network interface flags are updated.  This routine is
4898  * responsible for configuring the hardware for proper unicast, multicast and
4899  * promiscuous mode.
4900  **/
4901 void ixgbe_set_rx_mode(struct net_device *netdev)
4902 {
4903         struct ixgbe_adapter *adapter = netdev_priv(netdev);
4904         struct ixgbe_hw *hw = &adapter->hw;
4905         u32 fctrl, vmolr = IXGBE_VMOLR_BAM | IXGBE_VMOLR_AUPE;
4906         netdev_features_t features = netdev->features;
4907         int count;
4908
4909         /* Check for Promiscuous and All Multicast modes */
4910         fctrl = IXGBE_READ_REG(hw, IXGBE_FCTRL);
4911
4912         /* set all bits that we expect to always be set */
4913         fctrl &= ~IXGBE_FCTRL_SBP; /* disable store-bad-packets */
4914         fctrl |= IXGBE_FCTRL_BAM;
4915         fctrl |= IXGBE_FCTRL_DPF; /* discard pause frames when FC enabled */
4916         fctrl |= IXGBE_FCTRL_PMCF;
4917
4918         /* clear the bits we are changing the status of */
4919         fctrl &= ~(IXGBE_FCTRL_UPE | IXGBE_FCTRL_MPE);
4920         if (netdev->flags & IFF_PROMISC) {
4921                 hw->addr_ctrl.user_set_promisc = true;
4922                 fctrl |= (IXGBE_FCTRL_UPE | IXGBE_FCTRL_MPE);
4923                 vmolr |= IXGBE_VMOLR_MPE;
4924                 features &= ~NETIF_F_HW_VLAN_CTAG_FILTER;
4925         } else {
4926                 if (netdev->flags & IFF_ALLMULTI) {
4927                         fctrl |= IXGBE_FCTRL_MPE;
4928                         vmolr |= IXGBE_VMOLR_MPE;
4929                 }
4930                 hw->addr_ctrl.user_set_promisc = false;
4931         }
4932
4933         /*
4934          * Write addresses to available RAR registers, if there is not
4935          * sufficient space to store all the addresses then enable
4936          * unicast promiscuous mode
4937          */
4938         if (__dev_uc_sync(netdev, ixgbe_uc_sync, ixgbe_uc_unsync)) {
4939                 fctrl |= IXGBE_FCTRL_UPE;
4940                 vmolr |= IXGBE_VMOLR_ROPE;
4941         }
4942
4943         /* Write addresses to the MTA, if the attempt fails
4944          * then we should just turn on promiscuous mode so
4945          * that we can at least receive multicast traffic
4946          */
4947         count = ixgbe_write_mc_addr_list(netdev);
4948         if (count < 0) {
4949                 fctrl |= IXGBE_FCTRL_MPE;
4950                 vmolr |= IXGBE_VMOLR_MPE;
4951         } else if (count) {
4952                 vmolr |= IXGBE_VMOLR_ROMPE;
4953         }
4954
4955         if (hw->mac.type != ixgbe_mac_82598EB) {
4956                 vmolr |= IXGBE_READ_REG(hw, IXGBE_VMOLR(VMDQ_P(0))) &
4957                          ~(IXGBE_VMOLR_MPE | IXGBE_VMOLR_ROMPE |
4958                            IXGBE_VMOLR_ROPE);
4959                 IXGBE_WRITE_REG(hw, IXGBE_VMOLR(VMDQ_P(0)), vmolr);
4960         }
4961
4962         /* This is useful for sniffing bad packets. */
4963         if (features & NETIF_F_RXALL) {
4964                 /* UPE and MPE will be handled by normal PROMISC logic
4965                  * in e1000e_set_rx_mode */
4966                 fctrl |= (IXGBE_FCTRL_SBP | /* Receive bad packets */
4967                           IXGBE_FCTRL_BAM | /* RX All Bcast Pkts */
4968                           IXGBE_FCTRL_PMCF); /* RX All MAC Ctrl Pkts */
4969
4970                 fctrl &= ~(IXGBE_FCTRL_DPF);
4971                 /* NOTE:  VLAN filtering is disabled by setting PROMISC */
4972         }
4973
4974         IXGBE_WRITE_REG(hw, IXGBE_FCTRL, fctrl);
4975
4976         if (features & NETIF_F_HW_VLAN_CTAG_RX)
4977                 ixgbe_vlan_strip_enable(adapter);
4978         else
4979                 ixgbe_vlan_strip_disable(adapter);
4980
4981         if (features & NETIF_F_HW_VLAN_CTAG_FILTER)
4982                 ixgbe_vlan_promisc_disable(adapter);
4983         else
4984                 ixgbe_vlan_promisc_enable(adapter);
4985 }
4986
4987 static void ixgbe_napi_enable_all(struct ixgbe_adapter *adapter)
4988 {
4989         int q_idx;
4990
4991         for (q_idx = 0; q_idx < adapter->num_q_vectors; q_idx++)
4992                 napi_enable(&adapter->q_vector[q_idx]->napi);
4993 }
4994
4995 static void ixgbe_napi_disable_all(struct ixgbe_adapter *adapter)
4996 {
4997         int q_idx;
4998
4999         for (q_idx = 0; q_idx < adapter->num_q_vectors; q_idx++)
5000                 napi_disable(&adapter->q_vector[q_idx]->napi);
5001 }
5002
5003 static int ixgbe_udp_tunnel_sync(struct net_device *dev, unsigned int table)
5004 {
5005         struct ixgbe_adapter *adapter = netdev_priv(dev);
5006         struct ixgbe_hw *hw = &adapter->hw;
5007         struct udp_tunnel_info ti;
5008
5009         udp_tunnel_nic_get_port(dev, table, 0, &ti);
5010         if (ti.type == UDP_TUNNEL_TYPE_VXLAN)
5011                 adapter->vxlan_port = ti.port;
5012         else
5013                 adapter->geneve_port = ti.port;
5014
5015         IXGBE_WRITE_REG(hw, IXGBE_VXLANCTRL,
5016                         ntohs(adapter->vxlan_port) |
5017                         ntohs(adapter->geneve_port) <<
5018                                 IXGBE_VXLANCTRL_GENEVE_UDPPORT_SHIFT);
5019         return 0;
5020 }
5021
5022 static const struct udp_tunnel_nic_info ixgbe_udp_tunnels_x550 = {
5023         .sync_table     = ixgbe_udp_tunnel_sync,
5024         .flags          = UDP_TUNNEL_NIC_INFO_IPV4_ONLY,
5025         .tables         = {
5026                 { .n_entries = 1, .tunnel_types = UDP_TUNNEL_TYPE_VXLAN,  },
5027         },
5028 };
5029
5030 static const struct udp_tunnel_nic_info ixgbe_udp_tunnels_x550em_a = {
5031         .sync_table     = ixgbe_udp_tunnel_sync,
5032         .flags          = UDP_TUNNEL_NIC_INFO_IPV4_ONLY,
5033         .tables         = {
5034                 { .n_entries = 1, .tunnel_types = UDP_TUNNEL_TYPE_VXLAN,  },
5035                 { .n_entries = 1, .tunnel_types = UDP_TUNNEL_TYPE_GENEVE, },
5036         },
5037 };
5038
5039 #ifdef CONFIG_IXGBE_DCB
5040 /**
5041  * ixgbe_configure_dcb - Configure DCB hardware
5042  * @adapter: ixgbe adapter struct
5043  *
5044  * This is called by the driver on open to configure the DCB hardware.
5045  * This is also called by the gennetlink interface when reconfiguring
5046  * the DCB state.
5047  */
5048 static void ixgbe_configure_dcb(struct ixgbe_adapter *adapter)
5049 {
5050         struct ixgbe_hw *hw = &adapter->hw;
5051         int max_frame = adapter->netdev->mtu + ETH_HLEN + ETH_FCS_LEN;
5052
5053         if (!(adapter->flags & IXGBE_FLAG_DCB_ENABLED)) {
5054                 if (hw->mac.type == ixgbe_mac_82598EB)
5055                         netif_set_tso_max_size(adapter->netdev, 65536);
5056                 return;
5057         }
5058
5059         if (hw->mac.type == ixgbe_mac_82598EB)
5060                 netif_set_tso_max_size(adapter->netdev, 32768);
5061
5062 #ifdef IXGBE_FCOE
5063         if (adapter->netdev->features & NETIF_F_FCOE_MTU)
5064                 max_frame = max(max_frame, IXGBE_FCOE_JUMBO_FRAME_SIZE);
5065 #endif
5066
5067         /* reconfigure the hardware */
5068         if (adapter->dcbx_cap & DCB_CAP_DCBX_VER_CEE) {
5069                 ixgbe_dcb_calculate_tc_credits(hw, &adapter->dcb_cfg, max_frame,
5070                                                 DCB_TX_CONFIG);
5071                 ixgbe_dcb_calculate_tc_credits(hw, &adapter->dcb_cfg, max_frame,
5072                                                 DCB_RX_CONFIG);
5073                 ixgbe_dcb_hw_config(hw, &adapter->dcb_cfg);
5074         } else if (adapter->ixgbe_ieee_ets && adapter->ixgbe_ieee_pfc) {
5075                 ixgbe_dcb_hw_ets(&adapter->hw,
5076                                  adapter->ixgbe_ieee_ets,
5077                                  max_frame);
5078                 ixgbe_dcb_hw_pfc_config(&adapter->hw,
5079                                         adapter->ixgbe_ieee_pfc->pfc_en,
5080                                         adapter->ixgbe_ieee_ets->prio_tc);
5081         }
5082
5083         /* Enable RSS Hash per TC */
5084         if (hw->mac.type != ixgbe_mac_82598EB) {
5085                 u32 msb = 0;
5086                 u16 rss_i = adapter->ring_feature[RING_F_RSS].indices - 1;
5087
5088                 while (rss_i) {
5089                         msb++;
5090                         rss_i >>= 1;
5091                 }
5092
5093                 /* write msb to all 8 TCs in one write */
5094                 IXGBE_WRITE_REG(hw, IXGBE_RQTC, msb * 0x11111111);
5095         }
5096 }
5097 #endif
5098
5099 /* Additional bittime to account for IXGBE framing */
5100 #define IXGBE_ETH_FRAMING 20
5101
5102 /**
5103  * ixgbe_hpbthresh - calculate high water mark for flow control
5104  *
5105  * @adapter: board private structure to calculate for
5106  * @pb: packet buffer to calculate
5107  */
5108 static int ixgbe_hpbthresh(struct ixgbe_adapter *adapter, int pb)
5109 {
5110         struct ixgbe_hw *hw = &adapter->hw;
5111         struct net_device *dev = adapter->netdev;
5112         int link, tc, kb, marker;
5113         u32 dv_id, rx_pba;
5114
5115         /* Calculate max LAN frame size */
5116         tc = link = dev->mtu + ETH_HLEN + ETH_FCS_LEN + IXGBE_ETH_FRAMING;
5117
5118 #ifdef IXGBE_FCOE
5119         /* FCoE traffic class uses FCOE jumbo frames */
5120         if ((dev->features & NETIF_F_FCOE_MTU) &&
5121             (tc < IXGBE_FCOE_JUMBO_FRAME_SIZE) &&
5122             (pb == ixgbe_fcoe_get_tc(adapter)))
5123                 tc = IXGBE_FCOE_JUMBO_FRAME_SIZE;
5124 #endif
5125
5126         /* Calculate delay value for device */
5127         switch (hw->mac.type) {
5128         case ixgbe_mac_X540:
5129         case ixgbe_mac_X550:
5130         case ixgbe_mac_X550EM_x:
5131         case ixgbe_mac_x550em_a:
5132                 dv_id = IXGBE_DV_X540(link, tc);
5133                 break;
5134         default:
5135                 dv_id = IXGBE_DV(link, tc);
5136                 break;
5137         }
5138
5139         /* Loopback switch introduces additional latency */
5140         if (adapter->flags & IXGBE_FLAG_SRIOV_ENABLED)
5141                 dv_id += IXGBE_B2BT(tc);
5142
5143         /* Delay value is calculated in bit times convert to KB */
5144         kb = IXGBE_BT2KB(dv_id);
5145         rx_pba = IXGBE_READ_REG(hw, IXGBE_RXPBSIZE(pb)) >> 10;
5146
5147         marker = rx_pba - kb;
5148
5149         /* It is possible that the packet buffer is not large enough
5150          * to provide required headroom. In this case throw an error
5151          * to user and a do the best we can.
5152          */
5153         if (marker < 0) {
5154                 e_warn(drv, "Packet Buffer(%i) can not provide enough"
5155                             "headroom to support flow control."
5156                             "Decrease MTU or number of traffic classes\n", pb);
5157                 marker = tc + 1;
5158         }
5159
5160         return marker;
5161 }
5162
5163 /**
5164  * ixgbe_lpbthresh - calculate low water mark for for flow control
5165  *
5166  * @adapter: board private structure to calculate for
5167  * @pb: packet buffer to calculate
5168  */
5169 static int ixgbe_lpbthresh(struct ixgbe_adapter *adapter, int pb)
5170 {
5171         struct ixgbe_hw *hw = &adapter->hw;
5172         struct net_device *dev = adapter->netdev;
5173         int tc;
5174         u32 dv_id;
5175
5176         /* Calculate max LAN frame size */
5177         tc = dev->mtu + ETH_HLEN + ETH_FCS_LEN;
5178
5179 #ifdef IXGBE_FCOE
5180         /* FCoE traffic class uses FCOE jumbo frames */
5181         if ((dev->features & NETIF_F_FCOE_MTU) &&
5182             (tc < IXGBE_FCOE_JUMBO_FRAME_SIZE) &&
5183             (pb == netdev_get_prio_tc_map(dev, adapter->fcoe.up)))
5184                 tc = IXGBE_FCOE_JUMBO_FRAME_SIZE;
5185 #endif
5186
5187         /* Calculate delay value for device */
5188         switch (hw->mac.type) {
5189         case ixgbe_mac_X540:
5190         case ixgbe_mac_X550:
5191         case ixgbe_mac_X550EM_x:
5192         case ixgbe_mac_x550em_a:
5193                 dv_id = IXGBE_LOW_DV_X540(tc);
5194                 break;
5195         default:
5196                 dv_id = IXGBE_LOW_DV(tc);
5197                 break;
5198         }
5199
5200         /* Delay value is calculated in bit times convert to KB */
5201         return IXGBE_BT2KB(dv_id);
5202 }
5203
5204 /*
5205  * ixgbe_pbthresh_setup - calculate and setup high low water marks
5206  */
5207 static void ixgbe_pbthresh_setup(struct ixgbe_adapter *adapter)
5208 {
5209         struct ixgbe_hw *hw = &adapter->hw;
5210         int num_tc = adapter->hw_tcs;
5211         int i;
5212
5213         if (!num_tc)
5214                 num_tc = 1;
5215
5216         for (i = 0; i < num_tc; i++) {
5217                 hw->fc.high_water[i] = ixgbe_hpbthresh(adapter, i);
5218                 hw->fc.low_water[i] = ixgbe_lpbthresh(adapter, i);
5219
5220                 /* Low water marks must not be larger than high water marks */
5221                 if (hw->fc.low_water[i] > hw->fc.high_water[i])
5222                         hw->fc.low_water[i] = 0;
5223         }
5224
5225         for (; i < MAX_TRAFFIC_CLASS; i++)
5226                 hw->fc.high_water[i] = 0;
5227 }
5228
5229 static void ixgbe_configure_pb(struct ixgbe_adapter *adapter)
5230 {
5231         struct ixgbe_hw *hw = &adapter->hw;
5232         int hdrm;
5233         u8 tc = adapter->hw_tcs;
5234
5235         if (adapter->flags & IXGBE_FLAG_FDIR_HASH_CAPABLE ||
5236             adapter->flags & IXGBE_FLAG_FDIR_PERFECT_CAPABLE)
5237                 hdrm = 32 << adapter->fdir_pballoc;
5238         else
5239                 hdrm = 0;
5240
5241         hw->mac.ops.set_rxpba(hw, tc, hdrm, PBA_STRATEGY_EQUAL);
5242         ixgbe_pbthresh_setup(adapter);
5243 }
5244
5245 static void ixgbe_fdir_filter_restore(struct ixgbe_adapter *adapter)
5246 {
5247         struct ixgbe_hw *hw = &adapter->hw;
5248         struct hlist_node *node2;
5249         struct ixgbe_fdir_filter *filter;
5250         u8 queue;
5251
5252         spin_lock(&adapter->fdir_perfect_lock);
5253
5254         if (!hlist_empty(&adapter->fdir_filter_list))
5255                 ixgbe_fdir_set_input_mask_82599(hw, &adapter->fdir_mask);
5256
5257         hlist_for_each_entry_safe(filter, node2,
5258                                   &adapter->fdir_filter_list, fdir_node) {
5259                 if (filter->action == IXGBE_FDIR_DROP_QUEUE) {
5260                         queue = IXGBE_FDIR_DROP_QUEUE;
5261                 } else {
5262                         u32 ring = ethtool_get_flow_spec_ring(filter->action);
5263                         u8 vf = ethtool_get_flow_spec_ring_vf(filter->action);
5264
5265                         if (!vf && (ring >= adapter->num_rx_queues)) {
5266                                 e_err(drv, "FDIR restore failed without VF, ring: %u\n",
5267                                       ring);
5268                                 continue;
5269                         } else if (vf &&
5270                                    ((vf > adapter->num_vfs) ||
5271                                      ring >= adapter->num_rx_queues_per_pool)) {
5272                                 e_err(drv, "FDIR restore failed with VF, vf: %hhu, ring: %u\n",
5273                                       vf, ring);
5274                                 continue;
5275                         }
5276
5277                         /* Map the ring onto the absolute queue index */
5278                         if (!vf)
5279                                 queue = adapter->rx_ring[ring]->reg_idx;
5280                         else
5281                                 queue = ((vf - 1) *
5282                                         adapter->num_rx_queues_per_pool) + ring;
5283                 }
5284
5285                 ixgbe_fdir_write_perfect_filter_82599(hw,
5286                                 &filter->filter, filter->sw_idx, queue);
5287         }
5288
5289         spin_unlock(&adapter->fdir_perfect_lock);
5290 }
5291
5292 /**
5293  * ixgbe_clean_rx_ring - Free Rx Buffers per Queue
5294  * @rx_ring: ring to free buffers from
5295  **/
5296 static void ixgbe_clean_rx_ring(struct ixgbe_ring *rx_ring)
5297 {
5298         u16 i = rx_ring->next_to_clean;
5299         struct ixgbe_rx_buffer *rx_buffer = &rx_ring->rx_buffer_info[i];
5300
5301         if (rx_ring->xsk_pool) {
5302                 ixgbe_xsk_clean_rx_ring(rx_ring);
5303                 goto skip_free;
5304         }
5305
5306         /* Free all the Rx ring sk_buffs */
5307         while (i != rx_ring->next_to_alloc) {
5308                 if (rx_buffer->skb) {
5309                         struct sk_buff *skb = rx_buffer->skb;
5310                         if (IXGBE_CB(skb)->page_released)
5311                                 dma_unmap_page_attrs(rx_ring->dev,
5312                                                      IXGBE_CB(skb)->dma,
5313                                                      ixgbe_rx_pg_size(rx_ring),
5314                                                      DMA_FROM_DEVICE,
5315                                                      IXGBE_RX_DMA_ATTR);
5316                         dev_kfree_skb(skb);
5317                 }
5318
5319                 /* Invalidate cache lines that may have been written to by
5320                  * device so that we avoid corrupting memory.
5321                  */
5322                 dma_sync_single_range_for_cpu(rx_ring->dev,
5323                                               rx_buffer->dma,
5324                                               rx_buffer->page_offset,
5325                                               ixgbe_rx_bufsz(rx_ring),
5326                                               DMA_FROM_DEVICE);
5327
5328                 /* free resources associated with mapping */
5329                 dma_unmap_page_attrs(rx_ring->dev, rx_buffer->dma,
5330                                      ixgbe_rx_pg_size(rx_ring),
5331                                      DMA_FROM_DEVICE,
5332                                      IXGBE_RX_DMA_ATTR);
5333                 __page_frag_cache_drain(rx_buffer->page,
5334                                         rx_buffer->pagecnt_bias);
5335
5336                 i++;
5337                 rx_buffer++;
5338                 if (i == rx_ring->count) {
5339                         i = 0;
5340                         rx_buffer = rx_ring->rx_buffer_info;
5341                 }
5342         }
5343
5344 skip_free:
5345         rx_ring->next_to_alloc = 0;
5346         rx_ring->next_to_clean = 0;
5347         rx_ring->next_to_use = 0;
5348 }
5349
5350 static int ixgbe_fwd_ring_up(struct ixgbe_adapter *adapter,
5351                              struct ixgbe_fwd_adapter *accel)
5352 {
5353         u16 rss_i = adapter->ring_feature[RING_F_RSS].indices;
5354         int num_tc = netdev_get_num_tc(adapter->netdev);
5355         struct net_device *vdev = accel->netdev;
5356         int i, baseq, err;
5357
5358         baseq = accel->pool * adapter->num_rx_queues_per_pool;
5359         netdev_dbg(vdev, "pool %i:%i queues %i:%i\n",
5360                    accel->pool, adapter->num_rx_pools,
5361                    baseq, baseq + adapter->num_rx_queues_per_pool);
5362
5363         accel->rx_base_queue = baseq;
5364         accel->tx_base_queue = baseq;
5365
5366         /* record configuration for macvlan interface in vdev */
5367         for (i = 0; i < num_tc; i++)
5368                 netdev_bind_sb_channel_queue(adapter->netdev, vdev,
5369                                              i, rss_i, baseq + (rss_i * i));
5370
5371         for (i = 0; i < adapter->num_rx_queues_per_pool; i++)
5372                 adapter->rx_ring[baseq + i]->netdev = vdev;
5373
5374         /* Guarantee all rings are updated before we update the
5375          * MAC address filter.
5376          */
5377         wmb();
5378
5379         /* ixgbe_add_mac_filter will return an index if it succeeds, so we
5380          * need to only treat it as an error value if it is negative.
5381          */
5382         err = ixgbe_add_mac_filter(adapter, vdev->dev_addr,
5383                                    VMDQ_P(accel->pool));
5384         if (err >= 0)
5385                 return 0;
5386
5387         /* if we cannot add the MAC rule then disable the offload */
5388         macvlan_release_l2fw_offload(vdev);
5389
5390         for (i = 0; i < adapter->num_rx_queues_per_pool; i++)
5391                 adapter->rx_ring[baseq + i]->netdev = NULL;
5392
5393         netdev_err(vdev, "L2FW offload disabled due to L2 filter error\n");
5394
5395         /* unbind the queues and drop the subordinate channel config */
5396         netdev_unbind_sb_channel(adapter->netdev, vdev);
5397         netdev_set_sb_channel(vdev, 0);
5398
5399         clear_bit(accel->pool, adapter->fwd_bitmask);
5400         kfree(accel);
5401
5402         return err;
5403 }
5404
5405 static int ixgbe_macvlan_up(struct net_device *vdev,
5406                             struct netdev_nested_priv *priv)
5407 {
5408         struct ixgbe_adapter *adapter = (struct ixgbe_adapter *)priv->data;
5409         struct ixgbe_fwd_adapter *accel;
5410
5411         if (!netif_is_macvlan(vdev))
5412                 return 0;
5413
5414         accel = macvlan_accel_priv(vdev);
5415         if (!accel)
5416                 return 0;
5417
5418         ixgbe_fwd_ring_up(adapter, accel);
5419
5420         return 0;
5421 }
5422
5423 static void ixgbe_configure_dfwd(struct ixgbe_adapter *adapter)
5424 {
5425         struct netdev_nested_priv priv = {
5426                 .data = (void *)adapter,
5427         };
5428
5429         netdev_walk_all_upper_dev_rcu(adapter->netdev,
5430                                       ixgbe_macvlan_up, &priv);
5431 }
5432
5433 static void ixgbe_configure(struct ixgbe_adapter *adapter)
5434 {
5435         struct ixgbe_hw *hw = &adapter->hw;
5436
5437         ixgbe_configure_pb(adapter);
5438 #ifdef CONFIG_IXGBE_DCB
5439         ixgbe_configure_dcb(adapter);
5440 #endif
5441         /*
5442          * We must restore virtualization before VLANs or else
5443          * the VLVF registers will not be populated
5444          */
5445         ixgbe_configure_virtualization(adapter);
5446
5447         ixgbe_set_rx_mode(adapter->netdev);
5448         ixgbe_restore_vlan(adapter);
5449         ixgbe_ipsec_restore(adapter);
5450
5451         switch (hw->mac.type) {
5452         case ixgbe_mac_82599EB:
5453         case ixgbe_mac_X540:
5454                 hw->mac.ops.disable_rx_buff(hw);
5455                 break;
5456         default:
5457                 break;
5458         }
5459
5460         if (adapter->flags & IXGBE_FLAG_FDIR_HASH_CAPABLE) {
5461                 ixgbe_init_fdir_signature_82599(&adapter->hw,
5462                                                 adapter->fdir_pballoc);
5463         } else if (adapter->flags & IXGBE_FLAG_FDIR_PERFECT_CAPABLE) {
5464                 ixgbe_init_fdir_perfect_82599(&adapter->hw,
5465                                               adapter->fdir_pballoc);
5466                 ixgbe_fdir_filter_restore(adapter);
5467         }
5468
5469         switch (hw->mac.type) {
5470         case ixgbe_mac_82599EB:
5471         case ixgbe_mac_X540:
5472                 hw->mac.ops.enable_rx_buff(hw);
5473                 break;
5474         default:
5475                 break;
5476         }
5477
5478 #ifdef CONFIG_IXGBE_DCA
5479         /* configure DCA */
5480         if (adapter->flags & IXGBE_FLAG_DCA_CAPABLE)
5481                 ixgbe_setup_dca(adapter);
5482 #endif /* CONFIG_IXGBE_DCA */
5483
5484 #ifdef IXGBE_FCOE
5485         /* configure FCoE L2 filters, redirection table, and Rx control */
5486         ixgbe_configure_fcoe(adapter);
5487
5488 #endif /* IXGBE_FCOE */
5489         ixgbe_configure_tx(adapter);
5490         ixgbe_configure_rx(adapter);
5491         ixgbe_configure_dfwd(adapter);
5492 }
5493
5494 /**
5495  * ixgbe_sfp_link_config - set up SFP+ link
5496  * @adapter: pointer to private adapter struct
5497  **/
5498 static void ixgbe_sfp_link_config(struct ixgbe_adapter *adapter)
5499 {
5500         /*
5501          * We are assuming the worst case scenario here, and that
5502          * is that an SFP was inserted/removed after the reset
5503          * but before SFP detection was enabled.  As such the best
5504          * solution is to just start searching as soon as we start
5505          */
5506         if (adapter->hw.mac.type == ixgbe_mac_82598EB)
5507                 adapter->flags2 |= IXGBE_FLAG2_SEARCH_FOR_SFP;
5508
5509         adapter->flags2 |= IXGBE_FLAG2_SFP_NEEDS_RESET;
5510         adapter->sfp_poll_time = 0;
5511 }
5512
5513 /**
5514  * ixgbe_non_sfp_link_config - set up non-SFP+ link
5515  * @hw: pointer to private hardware struct
5516  *
5517  * Returns 0 on success, negative on failure
5518  **/
5519 static int ixgbe_non_sfp_link_config(struct ixgbe_hw *hw)
5520 {
5521         u32 speed;
5522         bool autoneg, link_up = false;
5523         int ret = IXGBE_ERR_LINK_SETUP;
5524
5525         if (hw->mac.ops.check_link)
5526                 ret = hw->mac.ops.check_link(hw, &speed, &link_up, false);
5527
5528         if (ret)
5529                 return ret;
5530
5531         speed = hw->phy.autoneg_advertised;
5532         if (!speed && hw->mac.ops.get_link_capabilities) {
5533                 ret = hw->mac.ops.get_link_capabilities(hw, &speed,
5534                                                         &autoneg);
5535                 /* remove NBASE-T speeds from default autonegotiation
5536                  * to accommodate broken network switches in the field
5537                  * which cannot cope with advertised NBASE-T speeds
5538                  */
5539                 speed &= ~(IXGBE_LINK_SPEED_5GB_FULL |
5540                            IXGBE_LINK_SPEED_2_5GB_FULL);
5541         }
5542
5543         if (ret)
5544                 return ret;
5545
5546         if (hw->mac.ops.setup_link)
5547                 ret = hw->mac.ops.setup_link(hw, speed, link_up);
5548
5549         return ret;
5550 }
5551
5552 static void ixgbe_setup_gpie(struct ixgbe_adapter *adapter)
5553 {
5554         struct ixgbe_hw *hw = &adapter->hw;
5555         u32 gpie = 0;
5556
5557         if (adapter->flags & IXGBE_FLAG_MSIX_ENABLED) {
5558                 gpie = IXGBE_GPIE_MSIX_MODE | IXGBE_GPIE_PBA_SUPPORT |
5559                        IXGBE_GPIE_OCD;
5560                 gpie |= IXGBE_GPIE_EIAME;
5561                 /*
5562                  * use EIAM to auto-mask when MSI-X interrupt is asserted
5563                  * this saves a register write for every interrupt
5564                  */
5565                 switch (hw->mac.type) {
5566                 case ixgbe_mac_82598EB:
5567                         IXGBE_WRITE_REG(hw, IXGBE_EIAM, IXGBE_EICS_RTX_QUEUE);
5568                         break;
5569                 case ixgbe_mac_82599EB:
5570                 case ixgbe_mac_X540:
5571                 case ixgbe_mac_X550:
5572                 case ixgbe_mac_X550EM_x:
5573                 case ixgbe_mac_x550em_a:
5574                 default:
5575                         IXGBE_WRITE_REG(hw, IXGBE_EIAM_EX(0), 0xFFFFFFFF);
5576                         IXGBE_WRITE_REG(hw, IXGBE_EIAM_EX(1), 0xFFFFFFFF);
5577                         break;
5578                 }
5579         } else {
5580                 /* legacy interrupts, use EIAM to auto-mask when reading EICR,
5581                  * specifically only auto mask tx and rx interrupts */
5582                 IXGBE_WRITE_REG(hw, IXGBE_EIAM, IXGBE_EICS_RTX_QUEUE);
5583         }
5584
5585         /* XXX: to interrupt immediately for EICS writes, enable this */
5586         /* gpie |= IXGBE_GPIE_EIMEN; */
5587
5588         if (adapter->flags & IXGBE_FLAG_SRIOV_ENABLED) {
5589                 gpie &= ~IXGBE_GPIE_VTMODE_MASK;
5590
5591                 switch (adapter->ring_feature[RING_F_VMDQ].mask) {
5592                 case IXGBE_82599_VMDQ_8Q_MASK:
5593                         gpie |= IXGBE_GPIE_VTMODE_16;
5594                         break;
5595                 case IXGBE_82599_VMDQ_4Q_MASK:
5596                         gpie |= IXGBE_GPIE_VTMODE_32;
5597                         break;
5598                 default:
5599                         gpie |= IXGBE_GPIE_VTMODE_64;
5600                         break;
5601                 }
5602         }
5603
5604         /* Enable Thermal over heat sensor interrupt */
5605         if (adapter->flags2 & IXGBE_FLAG2_TEMP_SENSOR_CAPABLE) {
5606                 switch (adapter->hw.mac.type) {
5607                 case ixgbe_mac_82599EB:
5608                         gpie |= IXGBE_SDP0_GPIEN_8259X;
5609                         break;
5610                 default:
5611                         break;
5612                 }
5613         }
5614
5615         /* Enable fan failure interrupt */
5616         if (adapter->flags & IXGBE_FLAG_FAN_FAIL_CAPABLE)
5617                 gpie |= IXGBE_SDP1_GPIEN(hw);
5618
5619         switch (hw->mac.type) {
5620         case ixgbe_mac_82599EB:
5621                 gpie |= IXGBE_SDP1_GPIEN_8259X | IXGBE_SDP2_GPIEN_8259X;
5622                 break;
5623         case ixgbe_mac_X550EM_x:
5624         case ixgbe_mac_x550em_a:
5625                 gpie |= IXGBE_SDP0_GPIEN_X540;
5626                 break;
5627         default:
5628                 break;
5629         }
5630
5631         IXGBE_WRITE_REG(hw, IXGBE_GPIE, gpie);
5632 }
5633
5634 static void ixgbe_up_complete(struct ixgbe_adapter *adapter)
5635 {
5636         struct ixgbe_hw *hw = &adapter->hw;
5637         int err;
5638         u32 ctrl_ext;
5639
5640         ixgbe_get_hw_control(adapter);
5641         ixgbe_setup_gpie(adapter);
5642
5643         if (adapter->flags & IXGBE_FLAG_MSIX_ENABLED)
5644                 ixgbe_configure_msix(adapter);
5645         else
5646                 ixgbe_configure_msi_and_legacy(adapter);
5647
5648         /* enable the optics for 82599 SFP+ fiber */
5649         if (hw->mac.ops.enable_tx_laser)
5650                 hw->mac.ops.enable_tx_laser(hw);
5651
5652         if (hw->phy.ops.set_phy_power)
5653                 hw->phy.ops.set_phy_power(hw, true);
5654
5655         smp_mb__before_atomic();
5656         clear_bit(__IXGBE_DOWN, &adapter->state);
5657         ixgbe_napi_enable_all(adapter);
5658
5659         if (ixgbe_is_sfp(hw)) {
5660                 ixgbe_sfp_link_config(adapter);
5661         } else {
5662                 err = ixgbe_non_sfp_link_config(hw);
5663                 if (err)
5664                         e_err(probe, "link_config FAILED %d\n", err);
5665         }
5666
5667         /* clear any pending interrupts, may auto mask */
5668         IXGBE_READ_REG(hw, IXGBE_EICR);
5669         ixgbe_irq_enable(adapter, true, true);
5670
5671         /*
5672          * If this adapter has a fan, check to see if we had a failure
5673          * before we enabled the interrupt.
5674          */
5675         if (adapter->flags & IXGBE_FLAG_FAN_FAIL_CAPABLE) {
5676                 u32 esdp = IXGBE_READ_REG(hw, IXGBE_ESDP);
5677                 if (esdp & IXGBE_ESDP_SDP1)
5678                         e_crit(drv, "Fan has stopped, replace the adapter\n");
5679         }
5680
5681         /* bring the link up in the watchdog, this could race with our first
5682          * link up interrupt but shouldn't be a problem */
5683         adapter->flags |= IXGBE_FLAG_NEED_LINK_UPDATE;
5684         adapter->link_check_timeout = jiffies;
5685         mod_timer(&adapter->service_timer, jiffies);
5686
5687         /* Set PF Reset Done bit so PF/VF Mail Ops can work */
5688         ctrl_ext = IXGBE_READ_REG(hw, IXGBE_CTRL_EXT);
5689         ctrl_ext |= IXGBE_CTRL_EXT_PFRSTD;
5690         IXGBE_WRITE_REG(hw, IXGBE_CTRL_EXT, ctrl_ext);
5691
5692         /* update setting rx tx for all active vfs */
5693         ixgbe_set_all_vfs(adapter);
5694 }
5695
5696 void ixgbe_reinit_locked(struct ixgbe_adapter *adapter)
5697 {
5698         /* put off any impending NetWatchDogTimeout */
5699         netif_trans_update(adapter->netdev);
5700
5701         while (test_and_set_bit(__IXGBE_RESETTING, &adapter->state))
5702                 usleep_range(1000, 2000);
5703         if (adapter->hw.phy.type == ixgbe_phy_fw)
5704                 ixgbe_watchdog_link_is_down(adapter);
5705         ixgbe_down(adapter);
5706         /*
5707          * If SR-IOV enabled then wait a bit before bringing the adapter
5708          * back up to give the VFs time to respond to the reset.  The
5709          * two second wait is based upon the watchdog timer cycle in
5710          * the VF driver.
5711          */
5712         if (adapter->flags & IXGBE_FLAG_SRIOV_ENABLED)
5713                 msleep(2000);
5714         ixgbe_up(adapter);
5715         clear_bit(__IXGBE_RESETTING, &adapter->state);
5716 }
5717
5718 void ixgbe_up(struct ixgbe_adapter *adapter)
5719 {
5720         /* hardware has been reset, we need to reload some things */
5721         ixgbe_configure(adapter);
5722
5723         ixgbe_up_complete(adapter);
5724 }
5725
5726 static unsigned long ixgbe_get_completion_timeout(struct ixgbe_adapter *adapter)
5727 {
5728         u16 devctl2;
5729
5730         pcie_capability_read_word(adapter->pdev, PCI_EXP_DEVCTL2, &devctl2);
5731
5732         switch (devctl2 & IXGBE_PCIDEVCTRL2_TIMEO_MASK) {
5733         case IXGBE_PCIDEVCTRL2_17_34s:
5734         case IXGBE_PCIDEVCTRL2_4_8s:
5735                 /* For now we cap the upper limit on delay to 2 seconds
5736                  * as we end up going up to 34 seconds of delay in worst
5737                  * case timeout value.
5738                  */
5739         case IXGBE_PCIDEVCTRL2_1_2s:
5740                 return 2000000ul;       /* 2.0 s */
5741         case IXGBE_PCIDEVCTRL2_260_520ms:
5742                 return 520000ul;        /* 520 ms */
5743         case IXGBE_PCIDEVCTRL2_65_130ms:
5744                 return 130000ul;        /* 130 ms */
5745         case IXGBE_PCIDEVCTRL2_16_32ms:
5746                 return 32000ul;         /* 32 ms */
5747         case IXGBE_PCIDEVCTRL2_1_2ms:
5748                 return 2000ul;          /* 2 ms */
5749         case IXGBE_PCIDEVCTRL2_50_100us:
5750                 return 100ul;           /* 100 us */
5751         case IXGBE_PCIDEVCTRL2_16_32ms_def:
5752                 return 32000ul;         /* 32 ms */
5753         default:
5754                 break;
5755         }
5756
5757         /* We shouldn't need to hit this path, but just in case default as
5758          * though completion timeout is not supported and support 32ms.
5759          */
5760         return 32000ul;
5761 }
5762
5763 void ixgbe_disable_rx(struct ixgbe_adapter *adapter)
5764 {
5765         unsigned long wait_delay, delay_interval;
5766         struct ixgbe_hw *hw = &adapter->hw;
5767         int i, wait_loop;
5768         u32 rxdctl;
5769
5770         /* disable receives */
5771         hw->mac.ops.disable_rx(hw);
5772
5773         if (ixgbe_removed(hw->hw_addr))
5774                 return;
5775
5776         /* disable all enabled Rx queues */
5777         for (i = 0; i < adapter->num_rx_queues; i++) {
5778                 struct ixgbe_ring *ring = adapter->rx_ring[i];
5779                 u8 reg_idx = ring->reg_idx;
5780
5781                 rxdctl = IXGBE_READ_REG(hw, IXGBE_RXDCTL(reg_idx));
5782                 rxdctl &= ~IXGBE_RXDCTL_ENABLE;
5783                 rxdctl |= IXGBE_RXDCTL_SWFLSH;
5784
5785                 /* write value back with RXDCTL.ENABLE bit cleared */
5786                 IXGBE_WRITE_REG(hw, IXGBE_RXDCTL(reg_idx), rxdctl);
5787         }
5788
5789         /* RXDCTL.EN may not change on 82598 if link is down, so skip it */
5790         if (hw->mac.type == ixgbe_mac_82598EB &&
5791             !(IXGBE_READ_REG(hw, IXGBE_LINKS) & IXGBE_LINKS_UP))
5792                 return;
5793
5794         /* Determine our minimum delay interval. We will increase this value
5795          * with each subsequent test. This way if the device returns quickly
5796          * we should spend as little time as possible waiting, however as
5797          * the time increases we will wait for larger periods of time.
5798          *
5799          * The trick here is that we increase the interval using the
5800          * following pattern: 1x 3x 5x 7x 9x 11x 13x 15x 17x 19x. The result
5801          * of that wait is that it totals up to 100x whatever interval we
5802          * choose. Since our minimum wait is 100us we can just divide the
5803          * total timeout by 100 to get our minimum delay interval.
5804          */
5805         delay_interval = ixgbe_get_completion_timeout(adapter) / 100;
5806
5807         wait_loop = IXGBE_MAX_RX_DESC_POLL;
5808         wait_delay = delay_interval;
5809
5810         while (wait_loop--) {
5811                 usleep_range(wait_delay, wait_delay + 10);
5812                 wait_delay += delay_interval * 2;
5813                 rxdctl = 0;
5814
5815                 /* OR together the reading of all the active RXDCTL registers,
5816                  * and then test the result. We need the disable to complete
5817                  * before we start freeing the memory and invalidating the
5818                  * DMA mappings.
5819                  */
5820                 for (i = 0; i < adapter->num_rx_queues; i++) {
5821                         struct ixgbe_ring *ring = adapter->rx_ring[i];
5822                         u8 reg_idx = ring->reg_idx;
5823
5824                         rxdctl |= IXGBE_READ_REG(hw, IXGBE_RXDCTL(reg_idx));
5825                 }
5826
5827                 if (!(rxdctl & IXGBE_RXDCTL_ENABLE))
5828                         return;
5829         }
5830
5831         e_err(drv,
5832               "RXDCTL.ENABLE for one or more queues not cleared within the polling period\n");
5833 }
5834
5835 void ixgbe_disable_tx(struct ixgbe_adapter *adapter)
5836 {
5837         unsigned long wait_delay, delay_interval;
5838         struct ixgbe_hw *hw = &adapter->hw;
5839         int i, wait_loop;
5840         u32 txdctl;
5841
5842         if (ixgbe_removed(hw->hw_addr))
5843                 return;
5844
5845         /* disable all enabled Tx queues */
5846         for (i = 0; i < adapter->num_tx_queues; i++) {
5847                 struct ixgbe_ring *ring = adapter->tx_ring[i];
5848                 u8 reg_idx = ring->reg_idx;
5849
5850                 IXGBE_WRITE_REG(hw, IXGBE_TXDCTL(reg_idx), IXGBE_TXDCTL_SWFLSH);
5851         }
5852
5853         /* disable all enabled XDP Tx queues */
5854         for (i = 0; i < adapter->num_xdp_queues; i++) {
5855                 struct ixgbe_ring *ring = adapter->xdp_ring[i];
5856                 u8 reg_idx = ring->reg_idx;
5857
5858                 IXGBE_WRITE_REG(hw, IXGBE_TXDCTL(reg_idx), IXGBE_TXDCTL_SWFLSH);
5859         }
5860
5861         /* If the link is not up there shouldn't be much in the way of
5862          * pending transactions. Those that are left will be flushed out
5863          * when the reset logic goes through the flush sequence to clean out
5864          * the pending Tx transactions.
5865          */
5866         if (!(IXGBE_READ_REG(hw, IXGBE_LINKS) & IXGBE_LINKS_UP))
5867                 goto dma_engine_disable;
5868
5869         /* Determine our minimum delay interval. We will increase this value
5870          * with each subsequent test. This way if the device returns quickly
5871          * we should spend as little time as possible waiting, however as
5872          * the time increases we will wait for larger periods of time.
5873          *
5874          * The trick here is that we increase the interval using the
5875          * following pattern: 1x 3x 5x 7x 9x 11x 13x 15x 17x 19x. The result
5876          * of that wait is that it totals up to 100x whatever interval we
5877          * choose. Since our minimum wait is 100us we can just divide the
5878          * total timeout by 100 to get our minimum delay interval.
5879          */
5880         delay_interval = ixgbe_get_completion_timeout(adapter) / 100;
5881
5882         wait_loop = IXGBE_MAX_RX_DESC_POLL;
5883         wait_delay = delay_interval;
5884
5885         while (wait_loop--) {
5886                 usleep_range(wait_delay, wait_delay + 10);
5887                 wait_delay += delay_interval * 2;
5888                 txdctl = 0;
5889
5890                 /* OR together the reading of all the active TXDCTL registers,
5891                  * and then test the result. We need the disable to complete
5892                  * before we start freeing the memory and invalidating the
5893                  * DMA mappings.
5894                  */
5895                 for (i = 0; i < adapter->num_tx_queues; i++) {
5896                         struct ixgbe_ring *ring = adapter->tx_ring[i];
5897                         u8 reg_idx = ring->reg_idx;
5898
5899                         txdctl |= IXGBE_READ_REG(hw, IXGBE_TXDCTL(reg_idx));
5900                 }
5901                 for (i = 0; i < adapter->num_xdp_queues; i++) {
5902                         struct ixgbe_ring *ring = adapter->xdp_ring[i];
5903                         u8 reg_idx = ring->reg_idx;
5904
5905                         txdctl |= IXGBE_READ_REG(hw, IXGBE_TXDCTL(reg_idx));
5906                 }
5907
5908                 if (!(txdctl & IXGBE_TXDCTL_ENABLE))
5909                         goto dma_engine_disable;
5910         }
5911
5912         e_err(drv,
5913               "TXDCTL.ENABLE for one or more queues not cleared within the polling period\n");
5914
5915 dma_engine_disable:
5916         /* Disable the Tx DMA engine on 82599 and later MAC */
5917         switch (hw->mac.type) {
5918         case ixgbe_mac_82599EB:
5919         case ixgbe_mac_X540:
5920         case ixgbe_mac_X550:
5921         case ixgbe_mac_X550EM_x:
5922         case ixgbe_mac_x550em_a:
5923                 IXGBE_WRITE_REG(hw, IXGBE_DMATXCTL,
5924                                 (IXGBE_READ_REG(hw, IXGBE_DMATXCTL) &
5925                                  ~IXGBE_DMATXCTL_TE));
5926                 fallthrough;
5927         default:
5928                 break;
5929         }
5930 }
5931
5932 void ixgbe_reset(struct ixgbe_adapter *adapter)
5933 {
5934         struct ixgbe_hw *hw = &adapter->hw;
5935         struct net_device *netdev = adapter->netdev;
5936         int err;
5937
5938         if (ixgbe_removed(hw->hw_addr))
5939                 return;
5940         /* lock SFP init bit to prevent race conditions with the watchdog */
5941         while (test_and_set_bit(__IXGBE_IN_SFP_INIT, &adapter->state))
5942                 usleep_range(1000, 2000);
5943
5944         /* clear all SFP and link config related flags while holding SFP_INIT */
5945         adapter->flags2 &= ~(IXGBE_FLAG2_SEARCH_FOR_SFP |
5946                              IXGBE_FLAG2_SFP_NEEDS_RESET);
5947         adapter->flags &= ~IXGBE_FLAG_NEED_LINK_CONFIG;
5948
5949         err = hw->mac.ops.init_hw(hw);
5950         switch (err) {
5951         case 0:
5952         case IXGBE_ERR_SFP_NOT_PRESENT:
5953         case IXGBE_ERR_SFP_NOT_SUPPORTED:
5954                 break;
5955         case IXGBE_ERR_PRIMARY_REQUESTS_PENDING:
5956                 e_dev_err("primary disable timed out\n");
5957                 break;
5958         case IXGBE_ERR_EEPROM_VERSION:
5959                 /* We are running on a pre-production device, log a warning */
5960                 e_dev_warn("This device is a pre-production adapter/LOM. "
5961                            "Please be aware there may be issues associated with "
5962                            "your hardware.  If you are experiencing problems "
5963                            "please contact your Intel or hardware "
5964                            "representative who provided you with this "
5965                            "hardware.\n");
5966                 break;
5967         default:
5968                 e_dev_err("Hardware Error: %d\n", err);
5969         }
5970
5971         clear_bit(__IXGBE_IN_SFP_INIT, &adapter->state);
5972
5973         /* flush entries out of MAC table */
5974         ixgbe_flush_sw_mac_table(adapter);
5975         __dev_uc_unsync(netdev, NULL);
5976
5977         /* do not flush user set addresses */
5978         ixgbe_mac_set_default_filter(adapter);
5979
5980         /* update SAN MAC vmdq pool selection */
5981         if (hw->mac.san_mac_rar_index)
5982                 hw->mac.ops.set_vmdq_san_mac(hw, VMDQ_P(0));
5983
5984         if (test_bit(__IXGBE_PTP_RUNNING, &adapter->state))
5985                 ixgbe_ptp_reset(adapter);
5986
5987         if (hw->phy.ops.set_phy_power) {
5988                 if (!netif_running(adapter->netdev) && !adapter->wol)
5989                         hw->phy.ops.set_phy_power(hw, false);
5990                 else
5991                         hw->phy.ops.set_phy_power(hw, true);
5992         }
5993 }
5994
5995 /**
5996  * ixgbe_clean_tx_ring - Free Tx Buffers
5997  * @tx_ring: ring to be cleaned
5998  **/
5999 static void ixgbe_clean_tx_ring(struct ixgbe_ring *tx_ring)
6000 {
6001         u16 i = tx_ring->next_to_clean;
6002         struct ixgbe_tx_buffer *tx_buffer = &tx_ring->tx_buffer_info[i];
6003
6004         if (tx_ring->xsk_pool) {
6005                 ixgbe_xsk_clean_tx_ring(tx_ring);
6006                 goto out;
6007         }
6008
6009         while (i != tx_ring->next_to_use) {
6010                 union ixgbe_adv_tx_desc *eop_desc, *tx_desc;
6011
6012                 /* Free all the Tx ring sk_buffs */
6013                 if (ring_is_xdp(tx_ring))
6014                         xdp_return_frame(tx_buffer->xdpf);
6015                 else
6016                         dev_kfree_skb_any(tx_buffer->skb);
6017
6018                 /* unmap skb header data */
6019                 dma_unmap_single(tx_ring->dev,
6020                                  dma_unmap_addr(tx_buffer, dma),
6021                                  dma_unmap_len(tx_buffer, len),
6022                                  DMA_TO_DEVICE);
6023
6024                 /* check for eop_desc to determine the end of the packet */
6025                 eop_desc = tx_buffer->next_to_watch;
6026                 tx_desc = IXGBE_TX_DESC(tx_ring, i);
6027
6028                 /* unmap remaining buffers */
6029                 while (tx_desc != eop_desc) {
6030                         tx_buffer++;
6031                         tx_desc++;
6032                         i++;
6033                         if (unlikely(i == tx_ring->count)) {
6034                                 i = 0;
6035                                 tx_buffer = tx_ring->tx_buffer_info;
6036                                 tx_desc = IXGBE_TX_DESC(tx_ring, 0);
6037                         }
6038
6039                         /* unmap any remaining paged data */
6040                         if (dma_unmap_len(tx_buffer, len))
6041                                 dma_unmap_page(tx_ring->dev,
6042                                                dma_unmap_addr(tx_buffer, dma),
6043                                                dma_unmap_len(tx_buffer, len),
6044                                                DMA_TO_DEVICE);
6045                 }
6046
6047                 /* move us one more past the eop_desc for start of next pkt */
6048                 tx_buffer++;
6049                 i++;
6050                 if (unlikely(i == tx_ring->count)) {
6051                         i = 0;
6052                         tx_buffer = tx_ring->tx_buffer_info;
6053                 }
6054         }
6055
6056         /* reset BQL for queue */
6057         if (!ring_is_xdp(tx_ring))
6058                 netdev_tx_reset_queue(txring_txq(tx_ring));
6059
6060 out:
6061         /* reset next_to_use and next_to_clean */
6062         tx_ring->next_to_use = 0;
6063         tx_ring->next_to_clean = 0;
6064 }
6065
6066 /**
6067  * ixgbe_clean_all_rx_rings - Free Rx Buffers for all queues
6068  * @adapter: board private structure
6069  **/
6070 static void ixgbe_clean_all_rx_rings(struct ixgbe_adapter *adapter)
6071 {
6072         int i;
6073
6074         for (i = 0; i < adapter->num_rx_queues; i++)
6075                 ixgbe_clean_rx_ring(adapter->rx_ring[i]);
6076 }
6077
6078 /**
6079  * ixgbe_clean_all_tx_rings - Free Tx Buffers for all queues
6080  * @adapter: board private structure
6081  **/
6082 static void ixgbe_clean_all_tx_rings(struct ixgbe_adapter *adapter)
6083 {
6084         int i;
6085
6086         for (i = 0; i < adapter->num_tx_queues; i++)
6087                 ixgbe_clean_tx_ring(adapter->tx_ring[i]);
6088         for (i = 0; i < adapter->num_xdp_queues; i++)
6089                 ixgbe_clean_tx_ring(adapter->xdp_ring[i]);
6090 }
6091
6092 static void ixgbe_fdir_filter_exit(struct ixgbe_adapter *adapter)
6093 {
6094         struct hlist_node *node2;
6095         struct ixgbe_fdir_filter *filter;
6096
6097         spin_lock(&adapter->fdir_perfect_lock);
6098
6099         hlist_for_each_entry_safe(filter, node2,
6100                                   &adapter->fdir_filter_list, fdir_node) {
6101                 hlist_del(&filter->fdir_node);
6102                 kfree(filter);
6103         }
6104         adapter->fdir_filter_count = 0;
6105
6106         spin_unlock(&adapter->fdir_perfect_lock);
6107 }
6108
6109 void ixgbe_down(struct ixgbe_adapter *adapter)
6110 {
6111         struct net_device *netdev = adapter->netdev;
6112         struct ixgbe_hw *hw = &adapter->hw;
6113         int i;
6114
6115         /* signal that we are down to the interrupt handler */
6116         if (test_and_set_bit(__IXGBE_DOWN, &adapter->state))
6117                 return; /* do nothing if already down */
6118
6119         /* Shut off incoming Tx traffic */
6120         netif_tx_stop_all_queues(netdev);
6121
6122         /* call carrier off first to avoid false dev_watchdog timeouts */
6123         netif_carrier_off(netdev);
6124         netif_tx_disable(netdev);
6125
6126         /* Disable Rx */
6127         ixgbe_disable_rx(adapter);
6128
6129         /* synchronize_rcu() needed for pending XDP buffers to drain */
6130         if (adapter->xdp_ring[0])
6131                 synchronize_rcu();
6132
6133         ixgbe_irq_disable(adapter);
6134
6135         ixgbe_napi_disable_all(adapter);
6136
6137         clear_bit(__IXGBE_RESET_REQUESTED, &adapter->state);
6138         adapter->flags2 &= ~IXGBE_FLAG2_FDIR_REQUIRES_REINIT;
6139         adapter->flags &= ~IXGBE_FLAG_NEED_LINK_UPDATE;
6140
6141         del_timer_sync(&adapter->service_timer);
6142
6143         if (adapter->num_vfs) {
6144                 /* Clear EITR Select mapping */
6145                 IXGBE_WRITE_REG(&adapter->hw, IXGBE_EITRSEL, 0);
6146
6147                 /* Mark all the VFs as inactive */
6148                 for (i = 0 ; i < adapter->num_vfs; i++)
6149                         adapter->vfinfo[i].clear_to_send = false;
6150
6151                 /* update setting rx tx for all active vfs */
6152                 ixgbe_set_all_vfs(adapter);
6153         }
6154
6155         /* disable transmits in the hardware now that interrupts are off */
6156         ixgbe_disable_tx(adapter);
6157
6158         if (!pci_channel_offline(adapter->pdev))
6159                 ixgbe_reset(adapter);
6160
6161         /* power down the optics for 82599 SFP+ fiber */
6162         if (hw->mac.ops.disable_tx_laser)
6163                 hw->mac.ops.disable_tx_laser(hw);
6164
6165         ixgbe_clean_all_tx_rings(adapter);
6166         ixgbe_clean_all_rx_rings(adapter);
6167 }
6168
6169 /**
6170  * ixgbe_set_eee_capable - helper function to determine EEE support on X550
6171  * @adapter: board private structure
6172  */
6173 static void ixgbe_set_eee_capable(struct ixgbe_adapter *adapter)
6174 {
6175         struct ixgbe_hw *hw = &adapter->hw;
6176
6177         switch (hw->device_id) {
6178         case IXGBE_DEV_ID_X550EM_A_1G_T:
6179         case IXGBE_DEV_ID_X550EM_A_1G_T_L:
6180                 if (!hw->phy.eee_speeds_supported)
6181                         break;
6182                 adapter->flags2 |= IXGBE_FLAG2_EEE_CAPABLE;
6183                 if (!hw->phy.eee_speeds_advertised)
6184                         break;
6185                 adapter->flags2 |= IXGBE_FLAG2_EEE_ENABLED;
6186                 break;
6187         default:
6188                 adapter->flags2 &= ~IXGBE_FLAG2_EEE_CAPABLE;
6189                 adapter->flags2 &= ~IXGBE_FLAG2_EEE_ENABLED;
6190                 break;
6191         }
6192 }
6193
6194 /**
6195  * ixgbe_tx_timeout - Respond to a Tx Hang
6196  * @netdev: network interface device structure
6197  * @txqueue: queue number that timed out
6198  **/
6199 static void ixgbe_tx_timeout(struct net_device *netdev, unsigned int __always_unused txqueue)
6200 {
6201         struct ixgbe_adapter *adapter = netdev_priv(netdev);
6202
6203         /* Do the reset outside of interrupt context */
6204         ixgbe_tx_timeout_reset(adapter);
6205 }
6206
6207 #ifdef CONFIG_IXGBE_DCB
6208 static void ixgbe_init_dcb(struct ixgbe_adapter *adapter)
6209 {
6210         struct ixgbe_hw *hw = &adapter->hw;
6211         struct tc_configuration *tc;
6212         int j;
6213
6214         switch (hw->mac.type) {
6215         case ixgbe_mac_82598EB:
6216         case ixgbe_mac_82599EB:
6217                 adapter->dcb_cfg.num_tcs.pg_tcs = MAX_TRAFFIC_CLASS;
6218                 adapter->dcb_cfg.num_tcs.pfc_tcs = MAX_TRAFFIC_CLASS;
6219                 break;
6220         case ixgbe_mac_X540:
6221         case ixgbe_mac_X550:
6222                 adapter->dcb_cfg.num_tcs.pg_tcs = X540_TRAFFIC_CLASS;
6223                 adapter->dcb_cfg.num_tcs.pfc_tcs = X540_TRAFFIC_CLASS;
6224                 break;
6225         case ixgbe_mac_X550EM_x:
6226         case ixgbe_mac_x550em_a:
6227         default:
6228                 adapter->dcb_cfg.num_tcs.pg_tcs = DEF_TRAFFIC_CLASS;
6229                 adapter->dcb_cfg.num_tcs.pfc_tcs = DEF_TRAFFIC_CLASS;
6230                 break;
6231         }
6232
6233         /* Configure DCB traffic classes */
6234         for (j = 0; j < MAX_TRAFFIC_CLASS; j++) {
6235                 tc = &adapter->dcb_cfg.tc_config[j];
6236                 tc->path[DCB_TX_CONFIG].bwg_id = 0;
6237                 tc->path[DCB_TX_CONFIG].bwg_percent = 12 + (j & 1);
6238                 tc->path[DCB_RX_CONFIG].bwg_id = 0;
6239                 tc->path[DCB_RX_CONFIG].bwg_percent = 12 + (j & 1);
6240                 tc->dcb_pfc = pfc_disabled;
6241         }
6242
6243         /* Initialize default user to priority mapping, UPx->TC0 */
6244         tc = &adapter->dcb_cfg.tc_config[0];
6245         tc->path[DCB_TX_CONFIG].up_to_tc_bitmap = 0xFF;
6246         tc->path[DCB_RX_CONFIG].up_to_tc_bitmap = 0xFF;
6247
6248         adapter->dcb_cfg.bw_percentage[DCB_TX_CONFIG][0] = 100;
6249         adapter->dcb_cfg.bw_percentage[DCB_RX_CONFIG][0] = 100;
6250         adapter->dcb_cfg.pfc_mode_enable = false;
6251         adapter->dcb_set_bitmap = 0x00;
6252         if (adapter->flags & IXGBE_FLAG_DCB_CAPABLE)
6253                 adapter->dcbx_cap = DCB_CAP_DCBX_HOST | DCB_CAP_DCBX_VER_CEE;
6254         memcpy(&adapter->temp_dcb_cfg, &adapter->dcb_cfg,
6255                sizeof(adapter->temp_dcb_cfg));
6256 }
6257 #endif
6258
6259 /**
6260  * ixgbe_sw_init - Initialize general software structures (struct ixgbe_adapter)
6261  * @adapter: board private structure to initialize
6262  * @ii: pointer to ixgbe_info for device
6263  *
6264  * ixgbe_sw_init initializes the Adapter private data structure.
6265  * Fields are initialized based on PCI device information and
6266  * OS network device settings (MTU size).
6267  **/
6268 static int ixgbe_sw_init(struct ixgbe_adapter *adapter,
6269                          const struct ixgbe_info *ii)
6270 {
6271         struct ixgbe_hw *hw = &adapter->hw;
6272         struct pci_dev *pdev = adapter->pdev;
6273         unsigned int rss, fdir;
6274         u32 fwsm;
6275         int i;
6276
6277         /* PCI config space info */
6278
6279         hw->vendor_id = pdev->vendor;
6280         hw->device_id = pdev->device;
6281         hw->revision_id = pdev->revision;
6282         hw->subsystem_vendor_id = pdev->subsystem_vendor;
6283         hw->subsystem_device_id = pdev->subsystem_device;
6284
6285         /* get_invariants needs the device IDs */
6286         ii->get_invariants(hw);
6287
6288         /* Set common capability flags and settings */
6289         rss = min_t(int, ixgbe_max_rss_indices(adapter), num_online_cpus());
6290         adapter->ring_feature[RING_F_RSS].limit = rss;
6291         adapter->flags2 |= IXGBE_FLAG2_RSC_CAPABLE;
6292         adapter->max_q_vectors = MAX_Q_VECTORS_82599;
6293         adapter->atr_sample_rate = 20;
6294         fdir = min_t(int, IXGBE_MAX_FDIR_INDICES, num_online_cpus());
6295         adapter->ring_feature[RING_F_FDIR].limit = fdir;
6296         adapter->fdir_pballoc = IXGBE_FDIR_PBALLOC_64K;
6297         adapter->ring_feature[RING_F_VMDQ].limit = 1;
6298 #ifdef CONFIG_IXGBE_DCA
6299         adapter->flags |= IXGBE_FLAG_DCA_CAPABLE;
6300 #endif
6301 #ifdef CONFIG_IXGBE_DCB
6302         adapter->flags |= IXGBE_FLAG_DCB_CAPABLE;
6303         adapter->flags &= ~IXGBE_FLAG_DCB_ENABLED;
6304 #endif
6305 #ifdef IXGBE_FCOE
6306         adapter->flags |= IXGBE_FLAG_FCOE_CAPABLE;
6307         adapter->flags &= ~IXGBE_FLAG_FCOE_ENABLED;
6308 #ifdef CONFIG_IXGBE_DCB
6309         /* Default traffic class to use for FCoE */
6310         adapter->fcoe.up = IXGBE_FCOE_DEFTC;
6311 #endif /* CONFIG_IXGBE_DCB */
6312 #endif /* IXGBE_FCOE */
6313
6314         /* initialize static ixgbe jump table entries */
6315         adapter->jump_tables[0] = kzalloc(sizeof(*adapter->jump_tables[0]),
6316                                           GFP_KERNEL);
6317         if (!adapter->jump_tables[0])
6318                 return -ENOMEM;
6319         adapter->jump_tables[0]->mat = ixgbe_ipv4_fields;
6320
6321         for (i = 1; i < IXGBE_MAX_LINK_HANDLE; i++)
6322                 adapter->jump_tables[i] = NULL;
6323
6324         adapter->mac_table = kcalloc(hw->mac.num_rar_entries,
6325                                      sizeof(struct ixgbe_mac_addr),
6326                                      GFP_KERNEL);
6327         if (!adapter->mac_table)
6328                 return -ENOMEM;
6329
6330         if (ixgbe_init_rss_key(adapter))
6331                 return -ENOMEM;
6332
6333         adapter->af_xdp_zc_qps = bitmap_zalloc(IXGBE_MAX_XDP_QS, GFP_KERNEL);
6334         if (!adapter->af_xdp_zc_qps)
6335                 return -ENOMEM;
6336
6337         /* Set MAC specific capability flags and exceptions */
6338         switch (hw->mac.type) {
6339         case ixgbe_mac_82598EB:
6340                 adapter->flags2 &= ~IXGBE_FLAG2_RSC_CAPABLE;
6341
6342                 if (hw->device_id == IXGBE_DEV_ID_82598AT)
6343                         adapter->flags |= IXGBE_FLAG_FAN_FAIL_CAPABLE;
6344
6345                 adapter->max_q_vectors = MAX_Q_VECTORS_82598;
6346                 adapter->ring_feature[RING_F_FDIR].limit = 0;
6347                 adapter->atr_sample_rate = 0;
6348                 adapter->fdir_pballoc = 0;
6349 #ifdef IXGBE_FCOE
6350                 adapter->flags &= ~IXGBE_FLAG_FCOE_CAPABLE;
6351                 adapter->flags &= ~IXGBE_FLAG_FCOE_ENABLED;
6352 #ifdef CONFIG_IXGBE_DCB
6353                 adapter->fcoe.up = 0;
6354 #endif /* IXGBE_DCB */
6355 #endif /* IXGBE_FCOE */
6356                 break;
6357         case ixgbe_mac_82599EB:
6358                 if (hw->device_id == IXGBE_DEV_ID_82599_T3_LOM)
6359                         adapter->flags2 |= IXGBE_FLAG2_TEMP_SENSOR_CAPABLE;
6360                 break;
6361         case ixgbe_mac_X540:
6362                 fwsm = IXGBE_READ_REG(hw, IXGBE_FWSM(hw));
6363                 if (fwsm & IXGBE_FWSM_TS_ENABLED)
6364                         adapter->flags2 |= IXGBE_FLAG2_TEMP_SENSOR_CAPABLE;
6365                 break;
6366         case ixgbe_mac_x550em_a:
6367                 switch (hw->device_id) {
6368                 case IXGBE_DEV_ID_X550EM_A_1G_T:
6369                 case IXGBE_DEV_ID_X550EM_A_1G_T_L:
6370                         adapter->flags2 |= IXGBE_FLAG2_TEMP_SENSOR_CAPABLE;
6371                         break;
6372                 default:
6373                         break;
6374                 }
6375                 fallthrough;
6376         case ixgbe_mac_X550EM_x:
6377 #ifdef CONFIG_IXGBE_DCB
6378                 adapter->flags &= ~IXGBE_FLAG_DCB_CAPABLE;
6379 #endif
6380 #ifdef IXGBE_FCOE
6381                 adapter->flags &= ~IXGBE_FLAG_FCOE_CAPABLE;
6382 #ifdef CONFIG_IXGBE_DCB
6383                 adapter->fcoe.up = 0;
6384 #endif /* IXGBE_DCB */
6385 #endif /* IXGBE_FCOE */
6386                 fallthrough;
6387         case ixgbe_mac_X550:
6388                 if (hw->mac.type == ixgbe_mac_X550)
6389                         adapter->flags2 |= IXGBE_FLAG2_TEMP_SENSOR_CAPABLE;
6390 #ifdef CONFIG_IXGBE_DCA
6391                 adapter->flags &= ~IXGBE_FLAG_DCA_CAPABLE;
6392 #endif
6393                 break;
6394         default:
6395                 break;
6396         }
6397
6398 #ifdef IXGBE_FCOE
6399         /* FCoE support exists, always init the FCoE lock */
6400         spin_lock_init(&adapter->fcoe.lock);
6401
6402 #endif
6403         /* n-tuple support exists, always init our spinlock */
6404         spin_lock_init(&adapter->fdir_perfect_lock);
6405
6406 #ifdef CONFIG_IXGBE_DCB
6407         ixgbe_init_dcb(adapter);
6408 #endif
6409         ixgbe_init_ipsec_offload(adapter);
6410
6411         /* default flow control settings */
6412         hw->fc.requested_mode = ixgbe_fc_full;
6413         hw->fc.current_mode = ixgbe_fc_full;    /* init for ethtool output */
6414         ixgbe_pbthresh_setup(adapter);
6415         hw->fc.pause_time = IXGBE_DEFAULT_FCPAUSE;
6416         hw->fc.send_xon = true;
6417         hw->fc.disable_fc_autoneg = ixgbe_device_supports_autoneg_fc(hw);
6418
6419 #ifdef CONFIG_PCI_IOV
6420         if (max_vfs > 0)
6421                 e_dev_warn("Enabling SR-IOV VFs using the max_vfs module parameter is deprecated - please use the pci sysfs interface instead.\n");
6422
6423         /* assign number of SR-IOV VFs */
6424         if (hw->mac.type != ixgbe_mac_82598EB) {
6425                 if (max_vfs > IXGBE_MAX_VFS_DRV_LIMIT) {
6426                         max_vfs = 0;
6427                         e_dev_warn("max_vfs parameter out of range. Not assigning any SR-IOV VFs\n");
6428                 }
6429         }
6430 #endif /* CONFIG_PCI_IOV */
6431
6432         /* enable itr by default in dynamic mode */
6433         adapter->rx_itr_setting = 1;
6434         adapter->tx_itr_setting = 1;
6435
6436         /* set default ring sizes */
6437         adapter->tx_ring_count = IXGBE_DEFAULT_TXD;
6438         adapter->rx_ring_count = IXGBE_DEFAULT_RXD;
6439
6440         /* set default work limits */
6441         adapter->tx_work_limit = IXGBE_DEFAULT_TX_WORK;
6442
6443         /* initialize eeprom parameters */
6444         if (ixgbe_init_eeprom_params_generic(hw)) {
6445                 e_dev_err("EEPROM initialization failed\n");
6446                 return -EIO;
6447         }
6448
6449         /* PF holds first pool slot */
6450         set_bit(0, adapter->fwd_bitmask);
6451         set_bit(__IXGBE_DOWN, &adapter->state);
6452
6453         return 0;
6454 }
6455
6456 /**
6457  * ixgbe_setup_tx_resources - allocate Tx resources (Descriptors)
6458  * @tx_ring:    tx descriptor ring (for a specific queue) to setup
6459  *
6460  * Return 0 on success, negative on failure
6461  **/
6462 int ixgbe_setup_tx_resources(struct ixgbe_ring *tx_ring)
6463 {
6464         struct device *dev = tx_ring->dev;
6465         int orig_node = dev_to_node(dev);
6466         int ring_node = NUMA_NO_NODE;
6467         int size;
6468
6469         size = sizeof(struct ixgbe_tx_buffer) * tx_ring->count;
6470
6471         if (tx_ring->q_vector)
6472                 ring_node = tx_ring->q_vector->numa_node;
6473
6474         tx_ring->tx_buffer_info = vmalloc_node(size, ring_node);
6475         if (!tx_ring->tx_buffer_info)
6476                 tx_ring->tx_buffer_info = vmalloc(size);
6477         if (!tx_ring->tx_buffer_info)
6478                 goto err;
6479
6480         /* round up to nearest 4K */
6481         tx_ring->size = tx_ring->count * sizeof(union ixgbe_adv_tx_desc);
6482         tx_ring->size = ALIGN(tx_ring->size, 4096);
6483
6484         set_dev_node(dev, ring_node);
6485         tx_ring->desc = dma_alloc_coherent(dev,
6486                                            tx_ring->size,
6487                                            &tx_ring->dma,
6488                                            GFP_KERNEL);
6489         set_dev_node(dev, orig_node);
6490         if (!tx_ring->desc)
6491                 tx_ring->desc = dma_alloc_coherent(dev, tx_ring->size,
6492                                                    &tx_ring->dma, GFP_KERNEL);
6493         if (!tx_ring->desc)
6494                 goto err;
6495
6496         tx_ring->next_to_use = 0;
6497         tx_ring->next_to_clean = 0;
6498         return 0;
6499
6500 err:
6501         vfree(tx_ring->tx_buffer_info);
6502         tx_ring->tx_buffer_info = NULL;
6503         dev_err(dev, "Unable to allocate memory for the Tx descriptor ring\n");
6504         return -ENOMEM;
6505 }
6506
6507 /**
6508  * ixgbe_setup_all_tx_resources - allocate all queues Tx resources
6509  * @adapter: board private structure
6510  *
6511  * If this function returns with an error, then it's possible one or
6512  * more of the rings is populated (while the rest are not).  It is the
6513  * callers duty to clean those orphaned rings.
6514  *
6515  * Return 0 on success, negative on failure
6516  **/
6517 static int ixgbe_setup_all_tx_resources(struct ixgbe_adapter *adapter)
6518 {
6519         int i, j = 0, err = 0;
6520
6521         for (i = 0; i < adapter->num_tx_queues; i++) {
6522                 err = ixgbe_setup_tx_resources(adapter->tx_ring[i]);
6523                 if (!err)
6524                         continue;
6525
6526                 e_err(probe, "Allocation for Tx Queue %u failed\n", i);
6527                 goto err_setup_tx;
6528         }
6529         for (j = 0; j < adapter->num_xdp_queues; j++) {
6530                 err = ixgbe_setup_tx_resources(adapter->xdp_ring[j]);
6531                 if (!err)
6532                         continue;
6533
6534                 e_err(probe, "Allocation for Tx Queue %u failed\n", j);
6535                 goto err_setup_tx;
6536         }
6537
6538         return 0;
6539 err_setup_tx:
6540         /* rewind the index freeing the rings as we go */
6541         while (j--)
6542                 ixgbe_free_tx_resources(adapter->xdp_ring[j]);
6543         while (i--)
6544                 ixgbe_free_tx_resources(adapter->tx_ring[i]);
6545         return err;
6546 }
6547
6548 static int ixgbe_rx_napi_id(struct ixgbe_ring *rx_ring)
6549 {
6550         struct ixgbe_q_vector *q_vector = rx_ring->q_vector;
6551
6552         return q_vector ? q_vector->napi.napi_id : 0;
6553 }
6554
6555 /**
6556  * ixgbe_setup_rx_resources - allocate Rx resources (Descriptors)
6557  * @adapter: pointer to ixgbe_adapter
6558  * @rx_ring:    rx descriptor ring (for a specific queue) to setup
6559  *
6560  * Returns 0 on success, negative on failure
6561  **/
6562 int ixgbe_setup_rx_resources(struct ixgbe_adapter *adapter,
6563                              struct ixgbe_ring *rx_ring)
6564 {
6565         struct device *dev = rx_ring->dev;
6566         int orig_node = dev_to_node(dev);
6567         int ring_node = NUMA_NO_NODE;
6568         int size;
6569
6570         size = sizeof(struct ixgbe_rx_buffer) * rx_ring->count;
6571
6572         if (rx_ring->q_vector)
6573                 ring_node = rx_ring->q_vector->numa_node;
6574
6575         rx_ring->rx_buffer_info = vmalloc_node(size, ring_node);
6576         if (!rx_ring->rx_buffer_info)
6577                 rx_ring->rx_buffer_info = vmalloc(size);
6578         if (!rx_ring->rx_buffer_info)
6579                 goto err;
6580
6581         /* Round up to nearest 4K */
6582         rx_ring->size = rx_ring->count * sizeof(union ixgbe_adv_rx_desc);
6583         rx_ring->size = ALIGN(rx_ring->size, 4096);
6584
6585         set_dev_node(dev, ring_node);
6586         rx_ring->desc = dma_alloc_coherent(dev,
6587                                            rx_ring->size,
6588                                            &rx_ring->dma,
6589                                            GFP_KERNEL);
6590         set_dev_node(dev, orig_node);
6591         if (!rx_ring->desc)
6592                 rx_ring->desc = dma_alloc_coherent(dev, rx_ring->size,
6593                                                    &rx_ring->dma, GFP_KERNEL);
6594         if (!rx_ring->desc)
6595                 goto err;
6596
6597         rx_ring->next_to_clean = 0;
6598         rx_ring->next_to_use = 0;
6599
6600         /* XDP RX-queue info */
6601         if (xdp_rxq_info_reg(&rx_ring->xdp_rxq, adapter->netdev,
6602                              rx_ring->queue_index, ixgbe_rx_napi_id(rx_ring)) < 0)
6603                 goto err;
6604
6605         rx_ring->xdp_prog = adapter->xdp_prog;
6606
6607         return 0;
6608 err:
6609         vfree(rx_ring->rx_buffer_info);
6610         rx_ring->rx_buffer_info = NULL;
6611         dev_err(dev, "Unable to allocate memory for the Rx descriptor ring\n");
6612         return -ENOMEM;
6613 }
6614
6615 /**
6616  * ixgbe_setup_all_rx_resources - allocate all queues Rx resources
6617  * @adapter: board private structure
6618  *
6619  * If this function returns with an error, then it's possible one or
6620  * more of the rings is populated (while the rest are not).  It is the
6621  * callers duty to clean those orphaned rings.
6622  *
6623  * Return 0 on success, negative on failure
6624  **/
6625 static int ixgbe_setup_all_rx_resources(struct ixgbe_adapter *adapter)
6626 {
6627         int i, err = 0;
6628
6629         for (i = 0; i < adapter->num_rx_queues; i++) {
6630                 err = ixgbe_setup_rx_resources(adapter, adapter->rx_ring[i]);
6631                 if (!err)
6632                         continue;
6633
6634                 e_err(probe, "Allocation for Rx Queue %u failed\n", i);
6635                 goto err_setup_rx;
6636         }
6637
6638 #ifdef IXGBE_FCOE
6639         err = ixgbe_setup_fcoe_ddp_resources(adapter);
6640         if (!err)
6641 #endif
6642                 return 0;
6643 err_setup_rx:
6644         /* rewind the index freeing the rings as we go */
6645         while (i--)
6646                 ixgbe_free_rx_resources(adapter->rx_ring[i]);
6647         return err;
6648 }
6649
6650 /**
6651  * ixgbe_free_tx_resources - Free Tx Resources per Queue
6652  * @tx_ring: Tx descriptor ring for a specific queue
6653  *
6654  * Free all transmit software resources
6655  **/
6656 void ixgbe_free_tx_resources(struct ixgbe_ring *tx_ring)
6657 {
6658         ixgbe_clean_tx_ring(tx_ring);
6659
6660         vfree(tx_ring->tx_buffer_info);
6661         tx_ring->tx_buffer_info = NULL;
6662
6663         /* if not set, then don't free */
6664         if (!tx_ring->desc)
6665                 return;
6666
6667         dma_free_coherent(tx_ring->dev, tx_ring->size,
6668                           tx_ring->desc, tx_ring->dma);
6669
6670         tx_ring->desc = NULL;
6671 }
6672
6673 /**
6674  * ixgbe_free_all_tx_resources - Free Tx Resources for All Queues
6675  * @adapter: board private structure
6676  *
6677  * Free all transmit software resources
6678  **/
6679 static void ixgbe_free_all_tx_resources(struct ixgbe_adapter *adapter)
6680 {
6681         int i;
6682
6683         for (i = 0; i < adapter->num_tx_queues; i++)
6684                 if (adapter->tx_ring[i]->desc)
6685                         ixgbe_free_tx_resources(adapter->tx_ring[i]);
6686         for (i = 0; i < adapter->num_xdp_queues; i++)
6687                 if (adapter->xdp_ring[i]->desc)
6688                         ixgbe_free_tx_resources(adapter->xdp_ring[i]);
6689 }
6690
6691 /**
6692  * ixgbe_free_rx_resources - Free Rx Resources
6693  * @rx_ring: ring to clean the resources from
6694  *
6695  * Free all receive software resources
6696  **/
6697 void ixgbe_free_rx_resources(struct ixgbe_ring *rx_ring)
6698 {
6699         ixgbe_clean_rx_ring(rx_ring);
6700
6701         rx_ring->xdp_prog = NULL;
6702         xdp_rxq_info_unreg(&rx_ring->xdp_rxq);
6703         vfree(rx_ring->rx_buffer_info);
6704         rx_ring->rx_buffer_info = NULL;
6705
6706         /* if not set, then don't free */
6707         if (!rx_ring->desc)
6708                 return;
6709
6710         dma_free_coherent(rx_ring->dev, rx_ring->size,
6711                           rx_ring->desc, rx_ring->dma);
6712
6713         rx_ring->desc = NULL;
6714 }
6715
6716 /**
6717  * ixgbe_free_all_rx_resources - Free Rx Resources for All Queues
6718  * @adapter: board private structure
6719  *
6720  * Free all receive software resources
6721  **/
6722 static void ixgbe_free_all_rx_resources(struct ixgbe_adapter *adapter)
6723 {
6724         int i;
6725
6726 #ifdef IXGBE_FCOE
6727         ixgbe_free_fcoe_ddp_resources(adapter);
6728
6729 #endif
6730         for (i = 0; i < adapter->num_rx_queues; i++)
6731                 if (adapter->rx_ring[i]->desc)
6732                         ixgbe_free_rx_resources(adapter->rx_ring[i]);
6733 }
6734
6735 /**
6736  * ixgbe_change_mtu - Change the Maximum Transfer Unit
6737  * @netdev: network interface device structure
6738  * @new_mtu: new value for maximum frame size
6739  *
6740  * Returns 0 on success, negative on failure
6741  **/
6742 static int ixgbe_change_mtu(struct net_device *netdev, int new_mtu)
6743 {
6744         struct ixgbe_adapter *adapter = netdev_priv(netdev);
6745
6746         if (adapter->xdp_prog) {
6747                 int new_frame_size = new_mtu + ETH_HLEN + ETH_FCS_LEN +
6748                                      VLAN_HLEN;
6749                 int i;
6750
6751                 for (i = 0; i < adapter->num_rx_queues; i++) {
6752                         struct ixgbe_ring *ring = adapter->rx_ring[i];
6753
6754                         if (new_frame_size > ixgbe_rx_bufsz(ring)) {
6755                                 e_warn(probe, "Requested MTU size is not supported with XDP\n");
6756                                 return -EINVAL;
6757                         }
6758                 }
6759         }
6760
6761         /*
6762          * For 82599EB we cannot allow legacy VFs to enable their receive
6763          * paths when MTU greater than 1500 is configured.  So display a
6764          * warning that legacy VFs will be disabled.
6765          */
6766         if ((adapter->flags & IXGBE_FLAG_SRIOV_ENABLED) &&
6767             (adapter->hw.mac.type == ixgbe_mac_82599EB) &&
6768             (new_mtu > ETH_DATA_LEN))
6769                 e_warn(probe, "Setting MTU > 1500 will disable legacy VFs\n");
6770
6771         netdev_dbg(netdev, "changing MTU from %d to %d\n",
6772                    netdev->mtu, new_mtu);
6773
6774         /* must set new MTU before calling down or up */
6775         netdev->mtu = new_mtu;
6776
6777         if (netif_running(netdev))
6778                 ixgbe_reinit_locked(adapter);
6779
6780         return 0;
6781 }
6782
6783 /**
6784  * ixgbe_open - Called when a network interface is made active
6785  * @netdev: network interface device structure
6786  *
6787  * Returns 0 on success, negative value on failure
6788  *
6789  * The open entry point is called when a network interface is made
6790  * active by the system (IFF_UP).  At this point all resources needed
6791  * for transmit and receive operations are allocated, the interrupt
6792  * handler is registered with the OS, the watchdog timer is started,
6793  * and the stack is notified that the interface is ready.
6794  **/
6795 int ixgbe_open(struct net_device *netdev)
6796 {
6797         struct ixgbe_adapter *adapter = netdev_priv(netdev);
6798         struct ixgbe_hw *hw = &adapter->hw;
6799         int err, queues;
6800
6801         /* disallow open during test */
6802         if (test_bit(__IXGBE_TESTING, &adapter->state))
6803                 return -EBUSY;
6804
6805         netif_carrier_off(netdev);
6806
6807         /* allocate transmit descriptors */
6808         err = ixgbe_setup_all_tx_resources(adapter);
6809         if (err)
6810                 goto err_setup_tx;
6811
6812         /* allocate receive descriptors */
6813         err = ixgbe_setup_all_rx_resources(adapter);
6814         if (err)
6815                 goto err_setup_rx;
6816
6817         ixgbe_configure(adapter);
6818
6819         err = ixgbe_request_irq(adapter);
6820         if (err)
6821                 goto err_req_irq;
6822
6823         /* Notify the stack of the actual queue counts. */
6824         queues = adapter->num_tx_queues;
6825         err = netif_set_real_num_tx_queues(netdev, queues);
6826         if (err)
6827                 goto err_set_queues;
6828
6829         queues = adapter->num_rx_queues;
6830         err = netif_set_real_num_rx_queues(netdev, queues);
6831         if (err)
6832                 goto err_set_queues;
6833
6834         ixgbe_ptp_init(adapter);
6835
6836         ixgbe_up_complete(adapter);
6837
6838         udp_tunnel_nic_reset_ntf(netdev);
6839
6840         return 0;
6841
6842 err_set_queues:
6843         ixgbe_free_irq(adapter);
6844 err_req_irq:
6845         ixgbe_free_all_rx_resources(adapter);
6846         if (hw->phy.ops.set_phy_power && !adapter->wol)
6847                 hw->phy.ops.set_phy_power(&adapter->hw, false);
6848 err_setup_rx:
6849         ixgbe_free_all_tx_resources(adapter);
6850 err_setup_tx:
6851         ixgbe_reset(adapter);
6852
6853         return err;
6854 }
6855
6856 static void ixgbe_close_suspend(struct ixgbe_adapter *adapter)
6857 {
6858         ixgbe_ptp_suspend(adapter);
6859
6860         if (adapter->hw.phy.ops.enter_lplu) {
6861                 adapter->hw.phy.reset_disable = true;
6862                 ixgbe_down(adapter);
6863                 adapter->hw.phy.ops.enter_lplu(&adapter->hw);
6864                 adapter->hw.phy.reset_disable = false;
6865         } else {
6866                 ixgbe_down(adapter);
6867         }
6868
6869         ixgbe_free_irq(adapter);
6870
6871         ixgbe_free_all_tx_resources(adapter);
6872         ixgbe_free_all_rx_resources(adapter);
6873 }
6874
6875 /**
6876  * ixgbe_close - Disables a network interface
6877  * @netdev: network interface device structure
6878  *
6879  * Returns 0, this is not allowed to fail
6880  *
6881  * The close entry point is called when an interface is de-activated
6882  * by the OS.  The hardware is still under the drivers control, but
6883  * needs to be disabled.  A global MAC reset is issued to stop the
6884  * hardware, and all transmit and receive resources are freed.
6885  **/
6886 int ixgbe_close(struct net_device *netdev)
6887 {
6888         struct ixgbe_adapter *adapter = netdev_priv(netdev);
6889
6890         ixgbe_ptp_stop(adapter);
6891
6892         if (netif_device_present(netdev))
6893                 ixgbe_close_suspend(adapter);
6894
6895         ixgbe_fdir_filter_exit(adapter);
6896
6897         ixgbe_release_hw_control(adapter);
6898
6899         return 0;
6900 }
6901
6902 static int __maybe_unused ixgbe_resume(struct device *dev_d)
6903 {
6904         struct pci_dev *pdev = to_pci_dev(dev_d);
6905         struct ixgbe_adapter *adapter = pci_get_drvdata(pdev);
6906         struct net_device *netdev = adapter->netdev;
6907         u32 err;
6908
6909         adapter->hw.hw_addr = adapter->io_addr;
6910
6911         err = pci_enable_device_mem(pdev);
6912         if (err) {
6913                 e_dev_err("Cannot enable PCI device from suspend\n");
6914                 return err;
6915         }
6916         smp_mb__before_atomic();
6917         clear_bit(__IXGBE_DISABLED, &adapter->state);
6918         pci_set_master(pdev);
6919
6920         device_wakeup_disable(dev_d);
6921
6922         ixgbe_reset(adapter);
6923
6924         IXGBE_WRITE_REG(&adapter->hw, IXGBE_WUS, ~0);
6925
6926         rtnl_lock();
6927         err = ixgbe_init_interrupt_scheme(adapter);
6928         if (!err && netif_running(netdev))
6929                 err = ixgbe_open(netdev);
6930
6931
6932         if (!err)
6933                 netif_device_attach(netdev);
6934         rtnl_unlock();
6935
6936         return err;
6937 }
6938
6939 static int __ixgbe_shutdown(struct pci_dev *pdev, bool *enable_wake)
6940 {
6941         struct ixgbe_adapter *adapter = pci_get_drvdata(pdev);
6942         struct net_device *netdev = adapter->netdev;
6943         struct ixgbe_hw *hw = &adapter->hw;
6944         u32 ctrl;
6945         u32 wufc = adapter->wol;
6946
6947         rtnl_lock();
6948         netif_device_detach(netdev);
6949
6950         if (netif_running(netdev))
6951                 ixgbe_close_suspend(adapter);
6952
6953         ixgbe_clear_interrupt_scheme(adapter);
6954         rtnl_unlock();
6955
6956         if (hw->mac.ops.stop_link_on_d3)
6957                 hw->mac.ops.stop_link_on_d3(hw);
6958
6959         if (wufc) {
6960                 u32 fctrl;
6961
6962                 ixgbe_set_rx_mode(netdev);
6963
6964                 /* enable the optics for 82599 SFP+ fiber as we can WoL */
6965                 if (hw->mac.ops.enable_tx_laser)
6966                         hw->mac.ops.enable_tx_laser(hw);
6967
6968                 /* enable the reception of multicast packets */
6969                 fctrl = IXGBE_READ_REG(hw, IXGBE_FCTRL);
6970                 fctrl |= IXGBE_FCTRL_MPE;
6971                 IXGBE_WRITE_REG(hw, IXGBE_FCTRL, fctrl);
6972
6973                 ctrl = IXGBE_READ_REG(hw, IXGBE_CTRL);
6974                 ctrl |= IXGBE_CTRL_GIO_DIS;
6975                 IXGBE_WRITE_REG(hw, IXGBE_CTRL, ctrl);
6976
6977                 IXGBE_WRITE_REG(hw, IXGBE_WUFC, wufc);
6978         } else {
6979                 IXGBE_WRITE_REG(hw, IXGBE_WUC, 0);
6980                 IXGBE_WRITE_REG(hw, IXGBE_WUFC, 0);
6981         }
6982
6983         switch (hw->mac.type) {
6984         case ixgbe_mac_82598EB:
6985                 pci_wake_from_d3(pdev, false);
6986                 break;
6987         case ixgbe_mac_82599EB:
6988         case ixgbe_mac_X540:
6989         case ixgbe_mac_X550:
6990         case ixgbe_mac_X550EM_x:
6991         case ixgbe_mac_x550em_a:
6992                 pci_wake_from_d3(pdev, !!wufc);
6993                 break;
6994         default:
6995                 break;
6996         }
6997
6998         *enable_wake = !!wufc;
6999         if (hw->phy.ops.set_phy_power && !*enable_wake)
7000                 hw->phy.ops.set_phy_power(hw, false);
7001
7002         ixgbe_release_hw_control(adapter);
7003
7004         if (!test_and_set_bit(__IXGBE_DISABLED, &adapter->state))
7005                 pci_disable_device(pdev);
7006
7007         return 0;
7008 }
7009
7010 static int __maybe_unused ixgbe_suspend(struct device *dev_d)
7011 {
7012         struct pci_dev *pdev = to_pci_dev(dev_d);
7013         int retval;
7014         bool wake;
7015
7016         retval = __ixgbe_shutdown(pdev, &wake);
7017
7018         device_set_wakeup_enable(dev_d, wake);
7019
7020         return retval;
7021 }
7022
7023 static void ixgbe_shutdown(struct pci_dev *pdev)
7024 {
7025         bool wake;
7026
7027         __ixgbe_shutdown(pdev, &wake);
7028
7029         if (system_state == SYSTEM_POWER_OFF) {
7030                 pci_wake_from_d3(pdev, wake);
7031                 pci_set_power_state(pdev, PCI_D3hot);
7032         }
7033 }
7034
7035 /**
7036  * ixgbe_update_stats - Update the board statistics counters.
7037  * @adapter: board private structure
7038  **/
7039 void ixgbe_update_stats(struct ixgbe_adapter *adapter)
7040 {
7041         struct net_device *netdev = adapter->netdev;
7042         struct ixgbe_hw *hw = &adapter->hw;
7043         struct ixgbe_hw_stats *hwstats = &adapter->stats;
7044         u64 total_mpc = 0;
7045         u32 i, missed_rx = 0, mpc, bprc, lxon, lxoff, xon_off_tot;
7046         u64 non_eop_descs = 0, restart_queue = 0, tx_busy = 0;
7047         u64 alloc_rx_page_failed = 0, alloc_rx_buff_failed = 0;
7048         u64 alloc_rx_page = 0;
7049         u64 bytes = 0, packets = 0, hw_csum_rx_error = 0;
7050
7051         if (test_bit(__IXGBE_DOWN, &adapter->state) ||
7052             test_bit(__IXGBE_RESETTING, &adapter->state))
7053                 return;
7054
7055         if (adapter->flags2 & IXGBE_FLAG2_RSC_ENABLED) {
7056                 u64 rsc_count = 0;
7057                 u64 rsc_flush = 0;
7058                 for (i = 0; i < adapter->num_rx_queues; i++) {
7059                         rsc_count += adapter->rx_ring[i]->rx_stats.rsc_count;
7060                         rsc_flush += adapter->rx_ring[i]->rx_stats.rsc_flush;
7061                 }
7062                 adapter->rsc_total_count = rsc_count;
7063                 adapter->rsc_total_flush = rsc_flush;
7064         }
7065
7066         for (i = 0; i < adapter->num_rx_queues; i++) {
7067                 struct ixgbe_ring *rx_ring = READ_ONCE(adapter->rx_ring[i]);
7068
7069                 if (!rx_ring)
7070                         continue;
7071                 non_eop_descs += rx_ring->rx_stats.non_eop_descs;
7072                 alloc_rx_page += rx_ring->rx_stats.alloc_rx_page;
7073                 alloc_rx_page_failed += rx_ring->rx_stats.alloc_rx_page_failed;
7074                 alloc_rx_buff_failed += rx_ring->rx_stats.alloc_rx_buff_failed;
7075                 hw_csum_rx_error += rx_ring->rx_stats.csum_err;
7076                 bytes += rx_ring->stats.bytes;
7077                 packets += rx_ring->stats.packets;
7078         }
7079         adapter->non_eop_descs = non_eop_descs;
7080         adapter->alloc_rx_page = alloc_rx_page;
7081         adapter->alloc_rx_page_failed = alloc_rx_page_failed;
7082         adapter->alloc_rx_buff_failed = alloc_rx_buff_failed;
7083         adapter->hw_csum_rx_error = hw_csum_rx_error;
7084         netdev->stats.rx_bytes = bytes;
7085         netdev->stats.rx_packets = packets;
7086
7087         bytes = 0;
7088         packets = 0;
7089         /* gather some stats to the adapter struct that are per queue */
7090         for (i = 0; i < adapter->num_tx_queues; i++) {
7091                 struct ixgbe_ring *tx_ring = READ_ONCE(adapter->tx_ring[i]);
7092
7093                 if (!tx_ring)
7094                         continue;
7095                 restart_queue += tx_ring->tx_stats.restart_queue;
7096                 tx_busy += tx_ring->tx_stats.tx_busy;
7097                 bytes += tx_ring->stats.bytes;
7098                 packets += tx_ring->stats.packets;
7099         }
7100         for (i = 0; i < adapter->num_xdp_queues; i++) {
7101                 struct ixgbe_ring *xdp_ring = READ_ONCE(adapter->xdp_ring[i]);
7102
7103                 if (!xdp_ring)
7104                         continue;
7105                 restart_queue += xdp_ring->tx_stats.restart_queue;
7106                 tx_busy += xdp_ring->tx_stats.tx_busy;
7107                 bytes += xdp_ring->stats.bytes;
7108                 packets += xdp_ring->stats.packets;
7109         }
7110         adapter->restart_queue = restart_queue;
7111         adapter->tx_busy = tx_busy;
7112         netdev->stats.tx_bytes = bytes;
7113         netdev->stats.tx_packets = packets;
7114
7115         hwstats->crcerrs += IXGBE_READ_REG(hw, IXGBE_CRCERRS);
7116
7117         /* 8 register reads */
7118         for (i = 0; i < 8; i++) {
7119                 /* for packet buffers not used, the register should read 0 */
7120                 mpc = IXGBE_READ_REG(hw, IXGBE_MPC(i));
7121                 missed_rx += mpc;
7122                 hwstats->mpc[i] += mpc;
7123                 total_mpc += hwstats->mpc[i];
7124                 hwstats->pxontxc[i] += IXGBE_READ_REG(hw, IXGBE_PXONTXC(i));
7125                 hwstats->pxofftxc[i] += IXGBE_READ_REG(hw, IXGBE_PXOFFTXC(i));
7126                 switch (hw->mac.type) {
7127                 case ixgbe_mac_82598EB:
7128                         hwstats->rnbc[i] += IXGBE_READ_REG(hw, IXGBE_RNBC(i));
7129                         hwstats->qbtc[i] += IXGBE_READ_REG(hw, IXGBE_QBTC(i));
7130                         hwstats->qbrc[i] += IXGBE_READ_REG(hw, IXGBE_QBRC(i));
7131                         hwstats->pxonrxc[i] +=
7132                                 IXGBE_READ_REG(hw, IXGBE_PXONRXC(i));
7133                         break;
7134                 case ixgbe_mac_82599EB:
7135                 case ixgbe_mac_X540:
7136                 case ixgbe_mac_X550:
7137                 case ixgbe_mac_X550EM_x:
7138                 case ixgbe_mac_x550em_a:
7139                         hwstats->pxonrxc[i] +=
7140                                 IXGBE_READ_REG(hw, IXGBE_PXONRXCNT(i));
7141                         break;
7142                 default:
7143                         break;
7144                 }
7145         }
7146
7147         /*16 register reads */
7148         for (i = 0; i < 16; i++) {
7149                 hwstats->qptc[i] += IXGBE_READ_REG(hw, IXGBE_QPTC(i));
7150                 hwstats->qprc[i] += IXGBE_READ_REG(hw, IXGBE_QPRC(i));
7151                 if ((hw->mac.type == ixgbe_mac_82599EB) ||
7152                     (hw->mac.type == ixgbe_mac_X540) ||
7153                     (hw->mac.type == ixgbe_mac_X550) ||
7154                     (hw->mac.type == ixgbe_mac_X550EM_x) ||
7155                     (hw->mac.type == ixgbe_mac_x550em_a)) {
7156                         hwstats->qbtc[i] += IXGBE_READ_REG(hw, IXGBE_QBTC_L(i));
7157                         IXGBE_READ_REG(hw, IXGBE_QBTC_H(i)); /* to clear */
7158                         hwstats->qbrc[i] += IXGBE_READ_REG(hw, IXGBE_QBRC_L(i));
7159                         IXGBE_READ_REG(hw, IXGBE_QBRC_H(i)); /* to clear */
7160                 }
7161         }
7162
7163         hwstats->gprc += IXGBE_READ_REG(hw, IXGBE_GPRC);
7164         /* work around hardware counting issue */
7165         hwstats->gprc -= missed_rx;
7166
7167         ixgbe_update_xoff_received(adapter);
7168
7169         /* 82598 hardware only has a 32 bit counter in the high register */
7170         switch (hw->mac.type) {
7171         case ixgbe_mac_82598EB:
7172                 hwstats->lxonrxc += IXGBE_READ_REG(hw, IXGBE_LXONRXC);
7173                 hwstats->gorc += IXGBE_READ_REG(hw, IXGBE_GORCH);
7174                 hwstats->gotc += IXGBE_READ_REG(hw, IXGBE_GOTCH);
7175                 hwstats->tor += IXGBE_READ_REG(hw, IXGBE_TORH);
7176                 break;
7177         case ixgbe_mac_X540:
7178         case ixgbe_mac_X550:
7179         case ixgbe_mac_X550EM_x:
7180         case ixgbe_mac_x550em_a:
7181                 /* OS2BMC stats are X540 and later */
7182                 hwstats->o2bgptc += IXGBE_READ_REG(hw, IXGBE_O2BGPTC);
7183                 hwstats->o2bspc += IXGBE_READ_REG(hw, IXGBE_O2BSPC);
7184                 hwstats->b2ospc += IXGBE_READ_REG(hw, IXGBE_B2OSPC);
7185                 hwstats->b2ogprc += IXGBE_READ_REG(hw, IXGBE_B2OGPRC);
7186                 fallthrough;
7187         case ixgbe_mac_82599EB:
7188                 for (i = 0; i < 16; i++)
7189                         adapter->hw_rx_no_dma_resources +=
7190                                              IXGBE_READ_REG(hw, IXGBE_QPRDC(i));
7191                 hwstats->gorc += IXGBE_READ_REG(hw, IXGBE_GORCL);
7192                 IXGBE_READ_REG(hw, IXGBE_GORCH); /* to clear */
7193                 hwstats->gotc += IXGBE_READ_REG(hw, IXGBE_GOTCL);
7194                 IXGBE_READ_REG(hw, IXGBE_GOTCH); /* to clear */
7195                 hwstats->tor += IXGBE_READ_REG(hw, IXGBE_TORL);
7196                 IXGBE_READ_REG(hw, IXGBE_TORH); /* to clear */
7197                 hwstats->lxonrxc += IXGBE_READ_REG(hw, IXGBE_LXONRXCNT);
7198                 hwstats->fdirmatch += IXGBE_READ_REG(hw, IXGBE_FDIRMATCH);
7199                 hwstats->fdirmiss += IXGBE_READ_REG(hw, IXGBE_FDIRMISS);
7200 #ifdef IXGBE_FCOE
7201                 hwstats->fccrc += IXGBE_READ_REG(hw, IXGBE_FCCRC);
7202                 hwstats->fcoerpdc += IXGBE_READ_REG(hw, IXGBE_FCOERPDC);
7203                 hwstats->fcoeprc += IXGBE_READ_REG(hw, IXGBE_FCOEPRC);
7204                 hwstats->fcoeptc += IXGBE_READ_REG(hw, IXGBE_FCOEPTC);
7205                 hwstats->fcoedwrc += IXGBE_READ_REG(hw, IXGBE_FCOEDWRC);
7206                 hwstats->fcoedwtc += IXGBE_READ_REG(hw, IXGBE_FCOEDWTC);
7207                 /* Add up per cpu counters for total ddp aloc fail */
7208                 if (adapter->fcoe.ddp_pool) {
7209                         struct ixgbe_fcoe *fcoe = &adapter->fcoe;
7210                         struct ixgbe_fcoe_ddp_pool *ddp_pool;
7211                         unsigned int cpu;
7212                         u64 noddp = 0, noddp_ext_buff = 0;
7213                         for_each_possible_cpu(cpu) {
7214                                 ddp_pool = per_cpu_ptr(fcoe->ddp_pool, cpu);
7215                                 noddp += ddp_pool->noddp;
7216                                 noddp_ext_buff += ddp_pool->noddp_ext_buff;
7217                         }
7218                         hwstats->fcoe_noddp = noddp;
7219                         hwstats->fcoe_noddp_ext_buff = noddp_ext_buff;
7220                 }
7221 #endif /* IXGBE_FCOE */
7222                 break;
7223         default:
7224                 break;
7225         }
7226         bprc = IXGBE_READ_REG(hw, IXGBE_BPRC);
7227         hwstats->bprc += bprc;
7228         hwstats->mprc += IXGBE_READ_REG(hw, IXGBE_MPRC);
7229         if (hw->mac.type == ixgbe_mac_82598EB)
7230                 hwstats->mprc -= bprc;
7231         hwstats->roc += IXGBE_READ_REG(hw, IXGBE_ROC);
7232         hwstats->prc64 += IXGBE_READ_REG(hw, IXGBE_PRC64);
7233         hwstats->prc127 += IXGBE_READ_REG(hw, IXGBE_PRC127);
7234         hwstats->prc255 += IXGBE_READ_REG(hw, IXGBE_PRC255);
7235         hwstats->prc511 += IXGBE_READ_REG(hw, IXGBE_PRC511);
7236         hwstats->prc1023 += IXGBE_READ_REG(hw, IXGBE_PRC1023);
7237         hwstats->prc1522 += IXGBE_READ_REG(hw, IXGBE_PRC1522);
7238         hwstats->rlec += IXGBE_READ_REG(hw, IXGBE_RLEC);
7239         lxon = IXGBE_READ_REG(hw, IXGBE_LXONTXC);
7240         hwstats->lxontxc += lxon;
7241         lxoff = IXGBE_READ_REG(hw, IXGBE_LXOFFTXC);
7242         hwstats->lxofftxc += lxoff;
7243         hwstats->gptc += IXGBE_READ_REG(hw, IXGBE_GPTC);
7244         hwstats->mptc += IXGBE_READ_REG(hw, IXGBE_MPTC);
7245         /*
7246          * 82598 errata - tx of flow control packets is included in tx counters
7247          */
7248         xon_off_tot = lxon + lxoff;
7249         hwstats->gptc -= xon_off_tot;
7250         hwstats->mptc -= xon_off_tot;
7251         hwstats->gotc -= (xon_off_tot * (ETH_ZLEN + ETH_FCS_LEN));
7252         hwstats->ruc += IXGBE_READ_REG(hw, IXGBE_RUC);
7253         hwstats->rfc += IXGBE_READ_REG(hw, IXGBE_RFC);
7254         hwstats->rjc += IXGBE_READ_REG(hw, IXGBE_RJC);
7255         hwstats->tpr += IXGBE_READ_REG(hw, IXGBE_TPR);
7256         hwstats->ptc64 += IXGBE_READ_REG(hw, IXGBE_PTC64);
7257         hwstats->ptc64 -= xon_off_tot;
7258         hwstats->ptc127 += IXGBE_READ_REG(hw, IXGBE_PTC127);
7259         hwstats->ptc255 += IXGBE_READ_REG(hw, IXGBE_PTC255);
7260         hwstats->ptc511 += IXGBE_READ_REG(hw, IXGBE_PTC511);
7261         hwstats->ptc1023 += IXGBE_READ_REG(hw, IXGBE_PTC1023);
7262         hwstats->ptc1522 += IXGBE_READ_REG(hw, IXGBE_PTC1522);
7263         hwstats->bptc += IXGBE_READ_REG(hw, IXGBE_BPTC);
7264
7265         /* Fill out the OS statistics structure */
7266         netdev->stats.multicast = hwstats->mprc;
7267
7268         /* Rx Errors */
7269         netdev->stats.rx_errors = hwstats->crcerrs + hwstats->rlec;
7270         netdev->stats.rx_dropped = 0;
7271         netdev->stats.rx_length_errors = hwstats->rlec;
7272         netdev->stats.rx_crc_errors = hwstats->crcerrs;
7273         netdev->stats.rx_missed_errors = total_mpc;
7274 }
7275
7276 /**
7277  * ixgbe_fdir_reinit_subtask - worker thread to reinit FDIR filter table
7278  * @adapter: pointer to the device adapter structure
7279  **/
7280 static void ixgbe_fdir_reinit_subtask(struct ixgbe_adapter *adapter)
7281 {
7282         struct ixgbe_hw *hw = &adapter->hw;
7283         int i;
7284
7285         if (!(adapter->flags2 & IXGBE_FLAG2_FDIR_REQUIRES_REINIT))
7286                 return;
7287
7288         adapter->flags2 &= ~IXGBE_FLAG2_FDIR_REQUIRES_REINIT;
7289
7290         /* if interface is down do nothing */
7291         if (test_bit(__IXGBE_DOWN, &adapter->state))
7292                 return;
7293
7294         /* do nothing if we are not using signature filters */
7295         if (!(adapter->flags & IXGBE_FLAG_FDIR_HASH_CAPABLE))
7296                 return;
7297
7298         adapter->fdir_overflow++;
7299
7300         if (ixgbe_reinit_fdir_tables_82599(hw) == 0) {
7301                 for (i = 0; i < adapter->num_tx_queues; i++)
7302                         set_bit(__IXGBE_TX_FDIR_INIT_DONE,
7303                                 &(adapter->tx_ring[i]->state));
7304                 for (i = 0; i < adapter->num_xdp_queues; i++)
7305                         set_bit(__IXGBE_TX_FDIR_INIT_DONE,
7306                                 &adapter->xdp_ring[i]->state);
7307                 /* re-enable flow director interrupts */
7308                 IXGBE_WRITE_REG(hw, IXGBE_EIMS, IXGBE_EIMS_FLOW_DIR);
7309         } else {
7310                 e_err(probe, "failed to finish FDIR re-initialization, "
7311                       "ignored adding FDIR ATR filters\n");
7312         }
7313 }
7314
7315 /**
7316  * ixgbe_check_hang_subtask - check for hung queues and dropped interrupts
7317  * @adapter: pointer to the device adapter structure
7318  *
7319  * This function serves two purposes.  First it strobes the interrupt lines
7320  * in order to make certain interrupts are occurring.  Secondly it sets the
7321  * bits needed to check for TX hangs.  As a result we should immediately
7322  * determine if a hang has occurred.
7323  */
7324 static void ixgbe_check_hang_subtask(struct ixgbe_adapter *adapter)
7325 {
7326         struct ixgbe_hw *hw = &adapter->hw;
7327         u64 eics = 0;
7328         int i;
7329
7330         /* If we're down, removing or resetting, just bail */
7331         if (test_bit(__IXGBE_DOWN, &adapter->state) ||
7332             test_bit(__IXGBE_REMOVING, &adapter->state) ||
7333             test_bit(__IXGBE_RESETTING, &adapter->state))
7334                 return;
7335
7336         /* Force detection of hung controller */
7337         if (netif_carrier_ok(adapter->netdev)) {
7338                 for (i = 0; i < adapter->num_tx_queues; i++)
7339                         set_check_for_tx_hang(adapter->tx_ring[i]);
7340                 for (i = 0; i < adapter->num_xdp_queues; i++)
7341                         set_check_for_tx_hang(adapter->xdp_ring[i]);
7342         }
7343
7344         if (!(adapter->flags & IXGBE_FLAG_MSIX_ENABLED)) {
7345                 /*
7346                  * for legacy and MSI interrupts don't set any bits
7347                  * that are enabled for EIAM, because this operation
7348                  * would set *both* EIMS and EICS for any bit in EIAM
7349                  */
7350                 IXGBE_WRITE_REG(hw, IXGBE_EICS,
7351                         (IXGBE_EICS_TCP_TIMER | IXGBE_EICS_OTHER));
7352         } else {
7353                 /* get one bit for every active tx/rx interrupt vector */
7354                 for (i = 0; i < adapter->num_q_vectors; i++) {
7355                         struct ixgbe_q_vector *qv = adapter->q_vector[i];
7356                         if (qv->rx.ring || qv->tx.ring)
7357                                 eics |= BIT_ULL(i);
7358                 }
7359         }
7360
7361         /* Cause software interrupt to ensure rings are cleaned */
7362         ixgbe_irq_rearm_queues(adapter, eics);
7363 }
7364
7365 /**
7366  * ixgbe_watchdog_update_link - update the link status
7367  * @adapter: pointer to the device adapter structure
7368  **/
7369 static void ixgbe_watchdog_update_link(struct ixgbe_adapter *adapter)
7370 {
7371         struct ixgbe_hw *hw = &adapter->hw;
7372         u32 link_speed = adapter->link_speed;
7373         bool link_up = adapter->link_up;
7374         bool pfc_en = adapter->dcb_cfg.pfc_mode_enable;
7375
7376         if (!(adapter->flags & IXGBE_FLAG_NEED_LINK_UPDATE))
7377                 return;
7378
7379         if (hw->mac.ops.check_link) {
7380                 hw->mac.ops.check_link(hw, &link_speed, &link_up, false);
7381         } else {
7382                 /* always assume link is up, if no check link function */
7383                 link_speed = IXGBE_LINK_SPEED_10GB_FULL;
7384                 link_up = true;
7385         }
7386
7387         if (adapter->ixgbe_ieee_pfc)
7388                 pfc_en |= !!(adapter->ixgbe_ieee_pfc->pfc_en);
7389
7390         if (link_up && !((adapter->flags & IXGBE_FLAG_DCB_ENABLED) && pfc_en)) {
7391                 hw->mac.ops.fc_enable(hw);
7392                 ixgbe_set_rx_drop_en(adapter);
7393         }
7394
7395         if (link_up ||
7396             time_after(jiffies, (adapter->link_check_timeout +
7397                                  IXGBE_TRY_LINK_TIMEOUT))) {
7398                 adapter->flags &= ~IXGBE_FLAG_NEED_LINK_UPDATE;
7399                 IXGBE_WRITE_REG(hw, IXGBE_EIMS, IXGBE_EIMC_LSC);
7400                 IXGBE_WRITE_FLUSH(hw);
7401         }
7402
7403         adapter->link_up = link_up;
7404         adapter->link_speed = link_speed;
7405 }
7406
7407 static void ixgbe_update_default_up(struct ixgbe_adapter *adapter)
7408 {
7409 #ifdef CONFIG_IXGBE_DCB
7410         struct net_device *netdev = adapter->netdev;
7411         struct dcb_app app = {
7412                               .selector = IEEE_8021QAZ_APP_SEL_ETHERTYPE,
7413                               .protocol = 0,
7414                              };
7415         u8 up = 0;
7416
7417         if (adapter->dcbx_cap & DCB_CAP_DCBX_VER_IEEE)
7418                 up = dcb_ieee_getapp_mask(netdev, &app);
7419
7420         adapter->default_up = (up > 1) ? (ffs(up) - 1) : 0;
7421 #endif
7422 }
7423
7424 /**
7425  * ixgbe_watchdog_link_is_up - update netif_carrier status and
7426  *                             print link up message
7427  * @adapter: pointer to the device adapter structure
7428  **/
7429 static void ixgbe_watchdog_link_is_up(struct ixgbe_adapter *adapter)
7430 {
7431         struct net_device *netdev = adapter->netdev;
7432         struct ixgbe_hw *hw = &adapter->hw;
7433         u32 link_speed = adapter->link_speed;
7434         const char *speed_str;
7435         bool flow_rx, flow_tx;
7436
7437         /* only continue if link was previously down */
7438         if (netif_carrier_ok(netdev))
7439                 return;
7440
7441         adapter->flags2 &= ~IXGBE_FLAG2_SEARCH_FOR_SFP;
7442
7443         switch (hw->mac.type) {
7444         case ixgbe_mac_82598EB: {
7445                 u32 frctl = IXGBE_READ_REG(hw, IXGBE_FCTRL);
7446                 u32 rmcs = IXGBE_READ_REG(hw, IXGBE_RMCS);
7447                 flow_rx = !!(frctl & IXGBE_FCTRL_RFCE);
7448                 flow_tx = !!(rmcs & IXGBE_RMCS_TFCE_802_3X);
7449         }
7450                 break;
7451         case ixgbe_mac_X540:
7452         case ixgbe_mac_X550:
7453         case ixgbe_mac_X550EM_x:
7454         case ixgbe_mac_x550em_a:
7455         case ixgbe_mac_82599EB: {
7456                 u32 mflcn = IXGBE_READ_REG(hw, IXGBE_MFLCN);
7457                 u32 fccfg = IXGBE_READ_REG(hw, IXGBE_FCCFG);
7458                 flow_rx = !!(mflcn & IXGBE_MFLCN_RFCE);
7459                 flow_tx = !!(fccfg & IXGBE_FCCFG_TFCE_802_3X);
7460         }
7461                 break;
7462         default:
7463                 flow_tx = false;
7464                 flow_rx = false;
7465                 break;
7466         }
7467
7468         adapter->last_rx_ptp_check = jiffies;
7469
7470         if (test_bit(__IXGBE_PTP_RUNNING, &adapter->state))
7471                 ixgbe_ptp_start_cyclecounter(adapter);
7472
7473         switch (link_speed) {
7474         case IXGBE_LINK_SPEED_10GB_FULL:
7475                 speed_str = "10 Gbps";
7476                 break;
7477         case IXGBE_LINK_SPEED_5GB_FULL:
7478                 speed_str = "5 Gbps";
7479                 break;
7480         case IXGBE_LINK_SPEED_2_5GB_FULL:
7481                 speed_str = "2.5 Gbps";
7482                 break;
7483         case IXGBE_LINK_SPEED_1GB_FULL:
7484                 speed_str = "1 Gbps";
7485                 break;
7486         case IXGBE_LINK_SPEED_100_FULL:
7487                 speed_str = "100 Mbps";
7488                 break;
7489         case IXGBE_LINK_SPEED_10_FULL:
7490                 speed_str = "10 Mbps";
7491                 break;
7492         default:
7493                 speed_str = "unknown speed";
7494                 break;
7495         }
7496         e_info(drv, "NIC Link is Up %s, Flow Control: %s\n", speed_str,
7497                ((flow_rx && flow_tx) ? "RX/TX" :
7498                (flow_rx ? "RX" :
7499                (flow_tx ? "TX" : "None"))));
7500
7501         netif_carrier_on(netdev);
7502         ixgbe_check_vf_rate_limit(adapter);
7503
7504         /* enable transmits */
7505         netif_tx_wake_all_queues(adapter->netdev);
7506
7507         /* update the default user priority for VFs */
7508         ixgbe_update_default_up(adapter);
7509
7510         /* ping all the active vfs to let them know link has changed */
7511         ixgbe_ping_all_vfs(adapter);
7512 }
7513
7514 /**
7515  * ixgbe_watchdog_link_is_down - update netif_carrier status and
7516  *                               print link down message
7517  * @adapter: pointer to the adapter structure
7518  **/
7519 static void ixgbe_watchdog_link_is_down(struct ixgbe_adapter *adapter)
7520 {
7521         struct net_device *netdev = adapter->netdev;
7522         struct ixgbe_hw *hw = &adapter->hw;
7523
7524         adapter->link_up = false;
7525         adapter->link_speed = 0;
7526
7527         /* only continue if link was up previously */
7528         if (!netif_carrier_ok(netdev))
7529                 return;
7530
7531         /* poll for SFP+ cable when link is down */
7532         if (ixgbe_is_sfp(hw) && hw->mac.type == ixgbe_mac_82598EB)
7533                 adapter->flags2 |= IXGBE_FLAG2_SEARCH_FOR_SFP;
7534
7535         if (test_bit(__IXGBE_PTP_RUNNING, &adapter->state))
7536                 ixgbe_ptp_start_cyclecounter(adapter);
7537
7538         e_info(drv, "NIC Link is Down\n");
7539         netif_carrier_off(netdev);
7540
7541         /* ping all the active vfs to let them know link has changed */
7542         ixgbe_ping_all_vfs(adapter);
7543 }
7544
7545 static bool ixgbe_ring_tx_pending(struct ixgbe_adapter *adapter)
7546 {
7547         int i;
7548
7549         for (i = 0; i < adapter->num_tx_queues; i++) {
7550                 struct ixgbe_ring *tx_ring = adapter->tx_ring[i];
7551
7552                 if (tx_ring->next_to_use != tx_ring->next_to_clean)
7553                         return true;
7554         }
7555
7556         for (i = 0; i < adapter->num_xdp_queues; i++) {
7557                 struct ixgbe_ring *ring = adapter->xdp_ring[i];
7558
7559                 if (ring->next_to_use != ring->next_to_clean)
7560                         return true;
7561         }
7562
7563         return false;
7564 }
7565
7566 static bool ixgbe_vf_tx_pending(struct ixgbe_adapter *adapter)
7567 {
7568         struct ixgbe_hw *hw = &adapter->hw;
7569         struct ixgbe_ring_feature *vmdq = &adapter->ring_feature[RING_F_VMDQ];
7570         u32 q_per_pool = __ALIGN_MASK(1, ~vmdq->mask);
7571
7572         int i, j;
7573
7574         if (!adapter->num_vfs)
7575                 return false;
7576
7577         /* resetting the PF is only needed for MAC before X550 */
7578         if (hw->mac.type >= ixgbe_mac_X550)
7579                 return false;
7580
7581         for (i = 0; i < adapter->num_vfs; i++) {
7582                 for (j = 0; j < q_per_pool; j++) {
7583                         u32 h, t;
7584
7585                         h = IXGBE_READ_REG(hw, IXGBE_PVFTDHN(q_per_pool, i, j));
7586                         t = IXGBE_READ_REG(hw, IXGBE_PVFTDTN(q_per_pool, i, j));
7587
7588                         if (h != t)
7589                                 return true;
7590                 }
7591         }
7592
7593         return false;
7594 }
7595
7596 /**
7597  * ixgbe_watchdog_flush_tx - flush queues on link down
7598  * @adapter: pointer to the device adapter structure
7599  **/
7600 static void ixgbe_watchdog_flush_tx(struct ixgbe_adapter *adapter)
7601 {
7602         if (!netif_carrier_ok(adapter->netdev)) {
7603                 if (ixgbe_ring_tx_pending(adapter) ||
7604                     ixgbe_vf_tx_pending(adapter)) {
7605                         /* We've lost link, so the controller stops DMA,
7606                          * but we've got queued Tx work that's never going
7607                          * to get done, so reset controller to flush Tx.
7608                          * (Do the reset outside of interrupt context).
7609                          */
7610                         e_warn(drv, "initiating reset to clear Tx work after link loss\n");
7611                         set_bit(__IXGBE_RESET_REQUESTED, &adapter->state);
7612                 }
7613         }
7614 }
7615
7616 #ifdef CONFIG_PCI_IOV
7617 static void ixgbe_bad_vf_abort(struct ixgbe_adapter *adapter, u32 vf)
7618 {
7619         struct ixgbe_hw *hw = &adapter->hw;
7620
7621         if (adapter->hw.mac.type == ixgbe_mac_82599EB &&
7622             adapter->flags2 & IXGBE_FLAG2_AUTO_DISABLE_VF) {
7623                 adapter->vfinfo[vf].primary_abort_count++;
7624                 if (adapter->vfinfo[vf].primary_abort_count ==
7625                     IXGBE_PRIMARY_ABORT_LIMIT) {
7626                         ixgbe_set_vf_link_state(adapter, vf,
7627                                                 IFLA_VF_LINK_STATE_DISABLE);
7628                         adapter->vfinfo[vf].primary_abort_count = 0;
7629
7630                         e_info(drv,
7631                                "Malicious Driver Detection event detected on PF %d VF %d MAC: %pM mdd-disable-vf=on",
7632                                hw->bus.func, vf,
7633                                adapter->vfinfo[vf].vf_mac_addresses);
7634                 }
7635         }
7636 }
7637
7638 static void ixgbe_check_for_bad_vf(struct ixgbe_adapter *adapter)
7639 {
7640         struct ixgbe_hw *hw = &adapter->hw;
7641         struct pci_dev *pdev = adapter->pdev;
7642         unsigned int vf;
7643         u32 gpc;
7644
7645         if (!(netif_carrier_ok(adapter->netdev)))
7646                 return;
7647
7648         gpc = IXGBE_READ_REG(hw, IXGBE_TXDGPC);
7649         if (gpc) /* If incrementing then no need for the check below */
7650                 return;
7651         /* Check to see if a bad DMA write target from an errant or
7652          * malicious VF has caused a PCIe error.  If so then we can
7653          * issue a VFLR to the offending VF(s) and then resume without
7654          * requesting a full slot reset.
7655          */
7656
7657         if (!pdev)
7658                 return;
7659
7660         /* check status reg for all VFs owned by this PF */
7661         for (vf = 0; vf < adapter->num_vfs; ++vf) {
7662                 struct pci_dev *vfdev = adapter->vfinfo[vf].vfdev;
7663                 u16 status_reg;
7664
7665                 if (!vfdev)
7666                         continue;
7667                 pci_read_config_word(vfdev, PCI_STATUS, &status_reg);
7668                 if (status_reg != IXGBE_FAILED_READ_CFG_WORD &&
7669                     status_reg & PCI_STATUS_REC_MASTER_ABORT) {
7670                         ixgbe_bad_vf_abort(adapter, vf);
7671                         pcie_flr(vfdev);
7672                 }
7673         }
7674 }
7675
7676 static void ixgbe_spoof_check(struct ixgbe_adapter *adapter)
7677 {
7678         u32 ssvpc;
7679
7680         /* Do not perform spoof check for 82598 or if not in IOV mode */
7681         if (adapter->hw.mac.type == ixgbe_mac_82598EB ||
7682             adapter->num_vfs == 0)
7683                 return;
7684
7685         ssvpc = IXGBE_READ_REG(&adapter->hw, IXGBE_SSVPC);
7686
7687         /*
7688          * ssvpc register is cleared on read, if zero then no
7689          * spoofed packets in the last interval.
7690          */
7691         if (!ssvpc)
7692                 return;
7693
7694         e_warn(drv, "%u Spoofed packets detected\n", ssvpc);
7695 }
7696 #else
7697 static void ixgbe_spoof_check(struct ixgbe_adapter __always_unused *adapter)
7698 {
7699 }
7700
7701 static void
7702 ixgbe_check_for_bad_vf(struct ixgbe_adapter __always_unused *adapter)
7703 {
7704 }
7705 #endif /* CONFIG_PCI_IOV */
7706
7707
7708 /**
7709  * ixgbe_watchdog_subtask - check and bring link up
7710  * @adapter: pointer to the device adapter structure
7711  **/
7712 static void ixgbe_watchdog_subtask(struct ixgbe_adapter *adapter)
7713 {
7714         /* if interface is down, removing or resetting, do nothing */
7715         if (test_bit(__IXGBE_DOWN, &adapter->state) ||
7716             test_bit(__IXGBE_REMOVING, &adapter->state) ||
7717             test_bit(__IXGBE_RESETTING, &adapter->state))
7718                 return;
7719
7720         ixgbe_watchdog_update_link(adapter);
7721
7722         if (adapter->link_up)
7723                 ixgbe_watchdog_link_is_up(adapter);
7724         else
7725                 ixgbe_watchdog_link_is_down(adapter);
7726
7727         ixgbe_check_for_bad_vf(adapter);
7728         ixgbe_spoof_check(adapter);
7729         ixgbe_update_stats(adapter);
7730
7731         ixgbe_watchdog_flush_tx(adapter);
7732 }
7733
7734 /**
7735  * ixgbe_sfp_detection_subtask - poll for SFP+ cable
7736  * @adapter: the ixgbe adapter structure
7737  **/
7738 static void ixgbe_sfp_detection_subtask(struct ixgbe_adapter *adapter)
7739 {
7740         struct ixgbe_hw *hw = &adapter->hw;
7741         s32 err;
7742
7743         /* not searching for SFP so there is nothing to do here */
7744         if (!(adapter->flags2 & IXGBE_FLAG2_SEARCH_FOR_SFP) &&
7745             !(adapter->flags2 & IXGBE_FLAG2_SFP_NEEDS_RESET))
7746                 return;
7747
7748         if (adapter->sfp_poll_time &&
7749             time_after(adapter->sfp_poll_time, jiffies))
7750                 return; /* If not yet time to poll for SFP */
7751
7752         /* someone else is in init, wait until next service event */
7753         if (test_and_set_bit(__IXGBE_IN_SFP_INIT, &adapter->state))
7754                 return;
7755
7756         adapter->sfp_poll_time = jiffies + IXGBE_SFP_POLL_JIFFIES - 1;
7757
7758         err = hw->phy.ops.identify_sfp(hw);
7759         if (err == IXGBE_ERR_SFP_NOT_SUPPORTED)
7760                 goto sfp_out;
7761
7762         if (err == IXGBE_ERR_SFP_NOT_PRESENT) {
7763                 /* If no cable is present, then we need to reset
7764                  * the next time we find a good cable. */
7765                 adapter->flags2 |= IXGBE_FLAG2_SFP_NEEDS_RESET;
7766         }
7767
7768         /* exit on error */
7769         if (err)
7770                 goto sfp_out;
7771
7772         /* exit if reset not needed */
7773         if (!(adapter->flags2 & IXGBE_FLAG2_SFP_NEEDS_RESET))
7774                 goto sfp_out;
7775
7776         adapter->flags2 &= ~IXGBE_FLAG2_SFP_NEEDS_RESET;
7777
7778         /*
7779          * A module may be identified correctly, but the EEPROM may not have
7780          * support for that module.  setup_sfp() will fail in that case, so
7781          * we should not allow that module to load.
7782          */
7783         if (hw->mac.type == ixgbe_mac_82598EB)
7784                 err = hw->phy.ops.reset(hw);
7785         else
7786                 err = hw->mac.ops.setup_sfp(hw);
7787
7788         if (err == IXGBE_ERR_SFP_NOT_SUPPORTED)
7789                 goto sfp_out;
7790
7791         adapter->flags |= IXGBE_FLAG_NEED_LINK_CONFIG;
7792         e_info(probe, "detected SFP+: %d\n", hw->phy.sfp_type);
7793
7794 sfp_out:
7795         clear_bit(__IXGBE_IN_SFP_INIT, &adapter->state);
7796
7797         if ((err == IXGBE_ERR_SFP_NOT_SUPPORTED) &&
7798             (adapter->netdev->reg_state == NETREG_REGISTERED)) {
7799                 e_dev_err("failed to initialize because an unsupported "
7800                           "SFP+ module type was detected.\n");
7801                 e_dev_err("Reload the driver after installing a "
7802                           "supported module.\n");
7803                 unregister_netdev(adapter->netdev);
7804         }
7805 }
7806
7807 /**
7808  * ixgbe_sfp_link_config_subtask - set up link SFP after module install
7809  * @adapter: the ixgbe adapter structure
7810  **/
7811 static void ixgbe_sfp_link_config_subtask(struct ixgbe_adapter *adapter)
7812 {
7813         struct ixgbe_hw *hw = &adapter->hw;
7814         u32 cap_speed;
7815         u32 speed;
7816         bool autoneg = false;
7817
7818         if (!(adapter->flags & IXGBE_FLAG_NEED_LINK_CONFIG))
7819                 return;
7820
7821         /* someone else is in init, wait until next service event */
7822         if (test_and_set_bit(__IXGBE_IN_SFP_INIT, &adapter->state))
7823                 return;
7824
7825         adapter->flags &= ~IXGBE_FLAG_NEED_LINK_CONFIG;
7826
7827         hw->mac.ops.get_link_capabilities(hw, &cap_speed, &autoneg);
7828
7829         /* advertise highest capable link speed */
7830         if (!autoneg && (cap_speed & IXGBE_LINK_SPEED_10GB_FULL))
7831                 speed = IXGBE_LINK_SPEED_10GB_FULL;
7832         else
7833                 speed = cap_speed & (IXGBE_LINK_SPEED_10GB_FULL |
7834                                      IXGBE_LINK_SPEED_1GB_FULL);
7835
7836         if (hw->mac.ops.setup_link)
7837                 hw->mac.ops.setup_link(hw, speed, true);
7838
7839         adapter->flags |= IXGBE_FLAG_NEED_LINK_UPDATE;
7840         adapter->link_check_timeout = jiffies;
7841         clear_bit(__IXGBE_IN_SFP_INIT, &adapter->state);
7842 }
7843
7844 /**
7845  * ixgbe_service_timer - Timer Call-back
7846  * @t: pointer to timer_list structure
7847  **/
7848 static void ixgbe_service_timer(struct timer_list *t)
7849 {
7850         struct ixgbe_adapter *adapter = from_timer(adapter, t, service_timer);
7851         unsigned long next_event_offset;
7852
7853         /* poll faster when waiting for link */
7854         if (adapter->flags & IXGBE_FLAG_NEED_LINK_UPDATE)
7855                 next_event_offset = HZ / 10;
7856         else
7857                 next_event_offset = HZ * 2;
7858
7859         /* Reset the timer */
7860         mod_timer(&adapter->service_timer, next_event_offset + jiffies);
7861
7862         ixgbe_service_event_schedule(adapter);
7863 }
7864
7865 static void ixgbe_phy_interrupt_subtask(struct ixgbe_adapter *adapter)
7866 {
7867         struct ixgbe_hw *hw = &adapter->hw;
7868         u32 status;
7869
7870         if (!(adapter->flags2 & IXGBE_FLAG2_PHY_INTERRUPT))
7871                 return;
7872
7873         adapter->flags2 &= ~IXGBE_FLAG2_PHY_INTERRUPT;
7874
7875         if (!hw->phy.ops.handle_lasi)
7876                 return;
7877
7878         status = hw->phy.ops.handle_lasi(&adapter->hw);
7879         if (status != IXGBE_ERR_OVERTEMP)
7880                 return;
7881
7882         e_crit(drv, "%s\n", ixgbe_overheat_msg);
7883 }
7884
7885 static void ixgbe_reset_subtask(struct ixgbe_adapter *adapter)
7886 {
7887         if (!test_and_clear_bit(__IXGBE_RESET_REQUESTED, &adapter->state))
7888                 return;
7889
7890         rtnl_lock();
7891         /* If we're already down, removing or resetting, just bail */
7892         if (test_bit(__IXGBE_DOWN, &adapter->state) ||
7893             test_bit(__IXGBE_REMOVING, &adapter->state) ||
7894             test_bit(__IXGBE_RESETTING, &adapter->state)) {
7895                 rtnl_unlock();
7896                 return;
7897         }
7898
7899         ixgbe_dump(adapter);
7900         netdev_err(adapter->netdev, "Reset adapter\n");
7901         adapter->tx_timeout_count++;
7902
7903         ixgbe_reinit_locked(adapter);
7904         rtnl_unlock();
7905 }
7906
7907 /**
7908  * ixgbe_check_fw_error - Check firmware for errors
7909  * @adapter: the adapter private structure
7910  *
7911  * Check firmware errors in register FWSM
7912  */
7913 static bool ixgbe_check_fw_error(struct ixgbe_adapter *adapter)
7914 {
7915         struct ixgbe_hw *hw = &adapter->hw;
7916         u32 fwsm;
7917
7918         /* read fwsm.ext_err_ind register and log errors */
7919         fwsm = IXGBE_READ_REG(hw, IXGBE_FWSM(hw));
7920
7921         if (fwsm & IXGBE_FWSM_EXT_ERR_IND_MASK ||
7922             !(fwsm & IXGBE_FWSM_FW_VAL_BIT))
7923                 e_dev_warn("Warning firmware error detected FWSM: 0x%08X\n",
7924                            fwsm);
7925
7926         if (hw->mac.ops.fw_recovery_mode && hw->mac.ops.fw_recovery_mode(hw)) {
7927                 e_dev_err("Firmware recovery mode detected. Limiting functionality. Refer to the Intel(R) Ethernet Adapters and Devices User Guide for details on firmware recovery mode.\n");
7928                 return true;
7929         }
7930
7931         return false;
7932 }
7933
7934 /**
7935  * ixgbe_service_task - manages and runs subtasks
7936  * @work: pointer to work_struct containing our data
7937  **/
7938 static void ixgbe_service_task(struct work_struct *work)
7939 {
7940         struct ixgbe_adapter *adapter = container_of(work,
7941                                                      struct ixgbe_adapter,
7942                                                      service_task);
7943         if (ixgbe_removed(adapter->hw.hw_addr)) {
7944                 if (!test_bit(__IXGBE_DOWN, &adapter->state)) {
7945                         rtnl_lock();
7946                         ixgbe_down(adapter);
7947                         rtnl_unlock();
7948                 }
7949                 ixgbe_service_event_complete(adapter);
7950                 return;
7951         }
7952         if (ixgbe_check_fw_error(adapter)) {
7953                 if (!test_bit(__IXGBE_DOWN, &adapter->state))
7954                         unregister_netdev(adapter->netdev);
7955                 ixgbe_service_event_complete(adapter);
7956                 return;
7957         }
7958         ixgbe_reset_subtask(adapter);
7959         ixgbe_phy_interrupt_subtask(adapter);
7960         ixgbe_sfp_detection_subtask(adapter);
7961         ixgbe_sfp_link_config_subtask(adapter);
7962         ixgbe_check_overtemp_subtask(adapter);
7963         ixgbe_watchdog_subtask(adapter);
7964         ixgbe_fdir_reinit_subtask(adapter);
7965         ixgbe_check_hang_subtask(adapter);
7966
7967         if (test_bit(__IXGBE_PTP_RUNNING, &adapter->state)) {
7968                 ixgbe_ptp_overflow_check(adapter);
7969                 if (adapter->flags & IXGBE_FLAG_RX_HWTSTAMP_IN_REGISTER)
7970                         ixgbe_ptp_rx_hang(adapter);
7971                 ixgbe_ptp_tx_hang(adapter);
7972         }
7973
7974         ixgbe_service_event_complete(adapter);
7975 }
7976
7977 static int ixgbe_tso(struct ixgbe_ring *tx_ring,
7978                      struct ixgbe_tx_buffer *first,
7979                      u8 *hdr_len,
7980                      struct ixgbe_ipsec_tx_data *itd)
7981 {
7982         u32 vlan_macip_lens, type_tucmd, mss_l4len_idx;
7983         struct sk_buff *skb = first->skb;
7984         union {
7985                 struct iphdr *v4;
7986                 struct ipv6hdr *v6;
7987                 unsigned char *hdr;
7988         } ip;
7989         union {
7990                 struct tcphdr *tcp;
7991                 struct udphdr *udp;
7992                 unsigned char *hdr;
7993         } l4;
7994         u32 paylen, l4_offset;
7995         u32 fceof_saidx = 0;
7996         int err;
7997
7998         if (skb->ip_summed != CHECKSUM_PARTIAL)
7999                 return 0;
8000
8001         if (!skb_is_gso(skb))
8002                 return 0;
8003
8004         err = skb_cow_head(skb, 0);
8005         if (err < 0)
8006                 return err;
8007
8008         if (eth_p_mpls(first->protocol))
8009                 ip.hdr = skb_inner_network_header(skb);
8010         else
8011                 ip.hdr = skb_network_header(skb);
8012         l4.hdr = skb_checksum_start(skb);
8013
8014         /* ADV DTYP TUCMD MKRLOC/ISCSIHEDLEN */
8015         type_tucmd = (skb_shinfo(skb)->gso_type & SKB_GSO_UDP_L4) ?
8016                       IXGBE_ADVTXD_TUCMD_L4T_UDP : IXGBE_ADVTXD_TUCMD_L4T_TCP;
8017
8018         /* initialize outer IP header fields */
8019         if (ip.v4->version == 4) {
8020                 unsigned char *csum_start = skb_checksum_start(skb);
8021                 unsigned char *trans_start = ip.hdr + (ip.v4->ihl * 4);
8022                 int len = csum_start - trans_start;
8023
8024                 /* IP header will have to cancel out any data that
8025                  * is not a part of the outer IP header, so set to
8026                  * a reverse csum if needed, else init check to 0.
8027                  */
8028                 ip.v4->check = (skb_shinfo(skb)->gso_type & SKB_GSO_PARTIAL) ?
8029                                            csum_fold(csum_partial(trans_start,
8030                                                                   len, 0)) : 0;
8031                 type_tucmd |= IXGBE_ADVTXD_TUCMD_IPV4;
8032
8033                 ip.v4->tot_len = 0;
8034                 first->tx_flags |= IXGBE_TX_FLAGS_TSO |
8035                                    IXGBE_TX_FLAGS_CSUM |
8036                                    IXGBE_TX_FLAGS_IPV4;
8037         } else {
8038                 ip.v6->payload_len = 0;
8039                 first->tx_flags |= IXGBE_TX_FLAGS_TSO |
8040                                    IXGBE_TX_FLAGS_CSUM;
8041         }
8042
8043         /* determine offset of inner transport header */
8044         l4_offset = l4.hdr - skb->data;
8045
8046         /* remove payload length from inner checksum */
8047         paylen = skb->len - l4_offset;
8048
8049         if (type_tucmd & IXGBE_ADVTXD_TUCMD_L4T_TCP) {
8050                 /* compute length of segmentation header */
8051                 *hdr_len = (l4.tcp->doff * 4) + l4_offset;
8052                 csum_replace_by_diff(&l4.tcp->check,
8053                                      (__force __wsum)htonl(paylen));
8054         } else {
8055                 /* compute length of segmentation header */
8056                 *hdr_len = sizeof(*l4.udp) + l4_offset;
8057                 csum_replace_by_diff(&l4.udp->check,
8058                                      (__force __wsum)htonl(paylen));
8059         }
8060
8061         /* update gso size and bytecount with header size */
8062         first->gso_segs = skb_shinfo(skb)->gso_segs;
8063         first->bytecount += (first->gso_segs - 1) * *hdr_len;
8064
8065         /* mss_l4len_id: use 0 as index for TSO */
8066         mss_l4len_idx = (*hdr_len - l4_offset) << IXGBE_ADVTXD_L4LEN_SHIFT;
8067         mss_l4len_idx |= skb_shinfo(skb)->gso_size << IXGBE_ADVTXD_MSS_SHIFT;
8068
8069         fceof_saidx |= itd->sa_idx;
8070         type_tucmd |= itd->flags | itd->trailer_len;
8071
8072         /* vlan_macip_lens: HEADLEN, MACLEN, VLAN tag */
8073         vlan_macip_lens = l4.hdr - ip.hdr;
8074         vlan_macip_lens |= (ip.hdr - skb->data) << IXGBE_ADVTXD_MACLEN_SHIFT;
8075         vlan_macip_lens |= first->tx_flags & IXGBE_TX_FLAGS_VLAN_MASK;
8076
8077         ixgbe_tx_ctxtdesc(tx_ring, vlan_macip_lens, fceof_saidx, type_tucmd,
8078                           mss_l4len_idx);
8079
8080         return 1;
8081 }
8082
8083 static void ixgbe_tx_csum(struct ixgbe_ring *tx_ring,
8084                           struct ixgbe_tx_buffer *first,
8085                           struct ixgbe_ipsec_tx_data *itd)
8086 {
8087         struct sk_buff *skb = first->skb;
8088         u32 vlan_macip_lens = 0;
8089         u32 fceof_saidx = 0;
8090         u32 type_tucmd = 0;
8091
8092         if (skb->ip_summed != CHECKSUM_PARTIAL) {
8093 csum_failed:
8094                 if (!(first->tx_flags & (IXGBE_TX_FLAGS_HW_VLAN |
8095                                          IXGBE_TX_FLAGS_CC)))
8096                         return;
8097                 goto no_csum;
8098         }
8099
8100         switch (skb->csum_offset) {
8101         case offsetof(struct tcphdr, check):
8102                 type_tucmd = IXGBE_ADVTXD_TUCMD_L4T_TCP;
8103                 fallthrough;
8104         case offsetof(struct udphdr, check):
8105                 break;
8106         case offsetof(struct sctphdr, checksum):
8107                 /* validate that this is actually an SCTP request */
8108                 if (skb_csum_is_sctp(skb)) {
8109                         type_tucmd = IXGBE_ADVTXD_TUCMD_L4T_SCTP;
8110                         break;
8111                 }
8112                 fallthrough;
8113         default:
8114                 skb_checksum_help(skb);
8115                 goto csum_failed;
8116         }
8117
8118         /* update TX checksum flag */
8119         first->tx_flags |= IXGBE_TX_FLAGS_CSUM;
8120         vlan_macip_lens = skb_checksum_start_offset(skb) -
8121                           skb_network_offset(skb);
8122 no_csum:
8123         /* vlan_macip_lens: MACLEN, VLAN tag */
8124         vlan_macip_lens |= skb_network_offset(skb) << IXGBE_ADVTXD_MACLEN_SHIFT;
8125         vlan_macip_lens |= first->tx_flags & IXGBE_TX_FLAGS_VLAN_MASK;
8126
8127         fceof_saidx |= itd->sa_idx;
8128         type_tucmd |= itd->flags | itd->trailer_len;
8129
8130         ixgbe_tx_ctxtdesc(tx_ring, vlan_macip_lens, fceof_saidx, type_tucmd, 0);
8131 }
8132
8133 #define IXGBE_SET_FLAG(_input, _flag, _result) \
8134         ((_flag <= _result) ? \
8135          ((u32)(_input & _flag) * (_result / _flag)) : \
8136          ((u32)(_input & _flag) / (_flag / _result)))
8137
8138 static u32 ixgbe_tx_cmd_type(struct sk_buff *skb, u32 tx_flags)
8139 {
8140         /* set type for advanced descriptor with frame checksum insertion */
8141         u32 cmd_type = IXGBE_ADVTXD_DTYP_DATA |
8142                        IXGBE_ADVTXD_DCMD_DEXT |
8143                        IXGBE_ADVTXD_DCMD_IFCS;
8144
8145         /* set HW vlan bit if vlan is present */
8146         cmd_type |= IXGBE_SET_FLAG(tx_flags, IXGBE_TX_FLAGS_HW_VLAN,
8147                                    IXGBE_ADVTXD_DCMD_VLE);
8148
8149         /* set segmentation enable bits for TSO/FSO */
8150         cmd_type |= IXGBE_SET_FLAG(tx_flags, IXGBE_TX_FLAGS_TSO,
8151                                    IXGBE_ADVTXD_DCMD_TSE);
8152
8153         /* set timestamp bit if present */
8154         cmd_type |= IXGBE_SET_FLAG(tx_flags, IXGBE_TX_FLAGS_TSTAMP,
8155                                    IXGBE_ADVTXD_MAC_TSTAMP);
8156
8157         /* insert frame checksum */
8158         cmd_type ^= IXGBE_SET_FLAG(skb->no_fcs, 1, IXGBE_ADVTXD_DCMD_IFCS);
8159
8160         return cmd_type;
8161 }
8162
8163 static void ixgbe_tx_olinfo_status(union ixgbe_adv_tx_desc *tx_desc,
8164                                    u32 tx_flags, unsigned int paylen)
8165 {
8166         u32 olinfo_status = paylen << IXGBE_ADVTXD_PAYLEN_SHIFT;
8167
8168         /* enable L4 checksum for TSO and TX checksum offload */
8169         olinfo_status |= IXGBE_SET_FLAG(tx_flags,
8170                                         IXGBE_TX_FLAGS_CSUM,
8171                                         IXGBE_ADVTXD_POPTS_TXSM);
8172
8173         /* enable IPv4 checksum for TSO */
8174         olinfo_status |= IXGBE_SET_FLAG(tx_flags,
8175                                         IXGBE_TX_FLAGS_IPV4,
8176                                         IXGBE_ADVTXD_POPTS_IXSM);
8177
8178         /* enable IPsec */
8179         olinfo_status |= IXGBE_SET_FLAG(tx_flags,
8180                                         IXGBE_TX_FLAGS_IPSEC,
8181                                         IXGBE_ADVTXD_POPTS_IPSEC);
8182
8183         /*
8184          * Check Context must be set if Tx switch is enabled, which it
8185          * always is for case where virtual functions are running
8186          */
8187         olinfo_status |= IXGBE_SET_FLAG(tx_flags,
8188                                         IXGBE_TX_FLAGS_CC,
8189                                         IXGBE_ADVTXD_CC);
8190
8191         tx_desc->read.olinfo_status = cpu_to_le32(olinfo_status);
8192 }
8193
8194 static int __ixgbe_maybe_stop_tx(struct ixgbe_ring *tx_ring, u16 size)
8195 {
8196         netif_stop_subqueue(tx_ring->netdev, tx_ring->queue_index);
8197
8198         /* Herbert's original patch had:
8199          *  smp_mb__after_netif_stop_queue();
8200          * but since that doesn't exist yet, just open code it.
8201          */
8202         smp_mb();
8203
8204         /* We need to check again in a case another CPU has just
8205          * made room available.
8206          */
8207         if (likely(ixgbe_desc_unused(tx_ring) < size))
8208                 return -EBUSY;
8209
8210         /* A reprieve! - use start_queue because it doesn't call schedule */
8211         netif_start_subqueue(tx_ring->netdev, tx_ring->queue_index);
8212         ++tx_ring->tx_stats.restart_queue;
8213         return 0;
8214 }
8215
8216 static inline int ixgbe_maybe_stop_tx(struct ixgbe_ring *tx_ring, u16 size)
8217 {
8218         if (likely(ixgbe_desc_unused(tx_ring) >= size))
8219                 return 0;
8220
8221         return __ixgbe_maybe_stop_tx(tx_ring, size);
8222 }
8223
8224 static int ixgbe_tx_map(struct ixgbe_ring *tx_ring,
8225                         struct ixgbe_tx_buffer *first,
8226                         const u8 hdr_len)
8227 {
8228         struct sk_buff *skb = first->skb;
8229         struct ixgbe_tx_buffer *tx_buffer;
8230         union ixgbe_adv_tx_desc *tx_desc;
8231         skb_frag_t *frag;
8232         dma_addr_t dma;
8233         unsigned int data_len, size;
8234         u32 tx_flags = first->tx_flags;
8235         u32 cmd_type = ixgbe_tx_cmd_type(skb, tx_flags);
8236         u16 i = tx_ring->next_to_use;
8237
8238         tx_desc = IXGBE_TX_DESC(tx_ring, i);
8239
8240         ixgbe_tx_olinfo_status(tx_desc, tx_flags, skb->len - hdr_len);
8241
8242         size = skb_headlen(skb);
8243         data_len = skb->data_len;
8244
8245 #ifdef IXGBE_FCOE
8246         if (tx_flags & IXGBE_TX_FLAGS_FCOE) {
8247                 if (data_len < sizeof(struct fcoe_crc_eof)) {
8248                         size -= sizeof(struct fcoe_crc_eof) - data_len;
8249                         data_len = 0;
8250                 } else {
8251                         data_len -= sizeof(struct fcoe_crc_eof);
8252                 }
8253         }
8254
8255 #endif
8256         dma = dma_map_single(tx_ring->dev, skb->data, size, DMA_TO_DEVICE);
8257
8258         tx_buffer = first;
8259
8260         for (frag = &skb_shinfo(skb)->frags[0];; frag++) {
8261                 if (dma_mapping_error(tx_ring->dev, dma))
8262                         goto dma_error;
8263
8264                 /* record length, and DMA address */
8265                 dma_unmap_len_set(tx_buffer, len, size);
8266                 dma_unmap_addr_set(tx_buffer, dma, dma);
8267
8268                 tx_desc->read.buffer_addr = cpu_to_le64(dma);
8269
8270                 while (unlikely(size > IXGBE_MAX_DATA_PER_TXD)) {
8271                         tx_desc->read.cmd_type_len =
8272                                 cpu_to_le32(cmd_type ^ IXGBE_MAX_DATA_PER_TXD);
8273
8274                         i++;
8275                         tx_desc++;
8276                         if (i == tx_ring->count) {
8277                                 tx_desc = IXGBE_TX_DESC(tx_ring, 0);
8278                                 i = 0;
8279                         }
8280                         tx_desc->read.olinfo_status = 0;
8281
8282                         dma += IXGBE_MAX_DATA_PER_TXD;
8283                         size -= IXGBE_MAX_DATA_PER_TXD;
8284
8285                         tx_desc->read.buffer_addr = cpu_to_le64(dma);
8286                 }
8287
8288                 if (likely(!data_len))
8289                         break;
8290
8291                 tx_desc->read.cmd_type_len = cpu_to_le32(cmd_type ^ size);
8292
8293                 i++;
8294                 tx_desc++;
8295                 if (i == tx_ring->count) {
8296                         tx_desc = IXGBE_TX_DESC(tx_ring, 0);
8297                         i = 0;
8298                 }
8299                 tx_desc->read.olinfo_status = 0;
8300
8301 #ifdef IXGBE_FCOE
8302                 size = min_t(unsigned int, data_len, skb_frag_size(frag));
8303 #else
8304                 size = skb_frag_size(frag);
8305 #endif
8306                 data_len -= size;
8307
8308                 dma = skb_frag_dma_map(tx_ring->dev, frag, 0, size,
8309                                        DMA_TO_DEVICE);
8310
8311                 tx_buffer = &tx_ring->tx_buffer_info[i];
8312         }
8313
8314         /* write last descriptor with RS and EOP bits */
8315         cmd_type |= size | IXGBE_TXD_CMD;
8316         tx_desc->read.cmd_type_len = cpu_to_le32(cmd_type);
8317
8318         netdev_tx_sent_queue(txring_txq(tx_ring), first->bytecount);
8319
8320         /* set the timestamp */
8321         first->time_stamp = jiffies;
8322
8323         skb_tx_timestamp(skb);
8324
8325         /*
8326          * Force memory writes to complete before letting h/w know there
8327          * are new descriptors to fetch.  (Only applicable for weak-ordered
8328          * memory model archs, such as IA-64).
8329          *
8330          * We also need this memory barrier to make certain all of the
8331          * status bits have been updated before next_to_watch is written.
8332          */
8333         wmb();
8334
8335         /* set next_to_watch value indicating a packet is present */
8336         first->next_to_watch = tx_desc;
8337
8338         i++;
8339         if (i == tx_ring->count)
8340                 i = 0;
8341
8342         tx_ring->next_to_use = i;
8343
8344         ixgbe_maybe_stop_tx(tx_ring, DESC_NEEDED);
8345
8346         if (netif_xmit_stopped(txring_txq(tx_ring)) || !netdev_xmit_more()) {
8347                 writel(i, tx_ring->tail);
8348         }
8349
8350         return 0;
8351 dma_error:
8352         dev_err(tx_ring->dev, "TX DMA map failed\n");
8353
8354         /* clear dma mappings for failed tx_buffer_info map */
8355         for (;;) {
8356                 tx_buffer = &tx_ring->tx_buffer_info[i];
8357                 if (dma_unmap_len(tx_buffer, len))
8358                         dma_unmap_page(tx_ring->dev,
8359                                        dma_unmap_addr(tx_buffer, dma),
8360                                        dma_unmap_len(tx_buffer, len),
8361                                        DMA_TO_DEVICE);
8362                 dma_unmap_len_set(tx_buffer, len, 0);
8363                 if (tx_buffer == first)
8364                         break;
8365                 if (i == 0)
8366                         i += tx_ring->count;
8367                 i--;
8368         }
8369
8370         dev_kfree_skb_any(first->skb);
8371         first->skb = NULL;
8372
8373         tx_ring->next_to_use = i;
8374
8375         return -1;
8376 }
8377
8378 static void ixgbe_atr(struct ixgbe_ring *ring,
8379                       struct ixgbe_tx_buffer *first)
8380 {
8381         struct ixgbe_q_vector *q_vector = ring->q_vector;
8382         union ixgbe_atr_hash_dword input = { .dword = 0 };
8383         union ixgbe_atr_hash_dword common = { .dword = 0 };
8384         union {
8385                 unsigned char *network;
8386                 struct iphdr *ipv4;
8387                 struct ipv6hdr *ipv6;
8388         } hdr;
8389         struct tcphdr *th;
8390         unsigned int hlen;
8391         struct sk_buff *skb;
8392         __be16 vlan_id;
8393         int l4_proto;
8394
8395         /* if ring doesn't have a interrupt vector, cannot perform ATR */
8396         if (!q_vector)
8397                 return;
8398
8399         /* do nothing if sampling is disabled */
8400         if (!ring->atr_sample_rate)
8401                 return;
8402
8403         ring->atr_count++;
8404
8405         /* currently only IPv4/IPv6 with TCP is supported */
8406         if ((first->protocol != htons(ETH_P_IP)) &&
8407             (first->protocol != htons(ETH_P_IPV6)))
8408                 return;
8409
8410         /* snag network header to get L4 type and address */
8411         skb = first->skb;
8412         hdr.network = skb_network_header(skb);
8413         if (unlikely(hdr.network <= skb->data))
8414                 return;
8415         if (skb->encapsulation &&
8416             first->protocol == htons(ETH_P_IP) &&
8417             hdr.ipv4->protocol == IPPROTO_UDP) {
8418                 struct ixgbe_adapter *adapter = q_vector->adapter;
8419
8420                 if (unlikely(skb_tail_pointer(skb) < hdr.network +
8421                              VXLAN_HEADROOM))
8422                         return;
8423
8424                 /* verify the port is recognized as VXLAN */
8425                 if (adapter->vxlan_port &&
8426                     udp_hdr(skb)->dest == adapter->vxlan_port)
8427                         hdr.network = skb_inner_network_header(skb);
8428
8429                 if (adapter->geneve_port &&
8430                     udp_hdr(skb)->dest == adapter->geneve_port)
8431                         hdr.network = skb_inner_network_header(skb);
8432         }
8433
8434         /* Make sure we have at least [minimum IPv4 header + TCP]
8435          * or [IPv6 header] bytes
8436          */
8437         if (unlikely(skb_tail_pointer(skb) < hdr.network + 40))
8438                 return;
8439
8440         /* Currently only IPv4/IPv6 with TCP is supported */
8441         switch (hdr.ipv4->version) {
8442         case IPVERSION:
8443                 /* access ihl as u8 to avoid unaligned access on ia64 */
8444                 hlen = (hdr.network[0] & 0x0F) << 2;
8445                 l4_proto = hdr.ipv4->protocol;
8446                 break;
8447         case 6:
8448                 hlen = hdr.network - skb->data;
8449                 l4_proto = ipv6_find_hdr(skb, &hlen, IPPROTO_TCP, NULL, NULL);
8450                 hlen -= hdr.network - skb->data;
8451                 break;
8452         default:
8453                 return;
8454         }
8455
8456         if (l4_proto != IPPROTO_TCP)
8457                 return;
8458
8459         if (unlikely(skb_tail_pointer(skb) < hdr.network +
8460                      hlen + sizeof(struct tcphdr)))
8461                 return;
8462
8463         th = (struct tcphdr *)(hdr.network + hlen);
8464
8465         /* skip this packet since the socket is closing */
8466         if (th->fin)
8467                 return;
8468
8469         /* sample on all syn packets or once every atr sample count */
8470         if (!th->syn && (ring->atr_count < ring->atr_sample_rate))
8471                 return;
8472
8473         /* reset sample count */
8474         ring->atr_count = 0;
8475
8476         vlan_id = htons(first->tx_flags >> IXGBE_TX_FLAGS_VLAN_SHIFT);
8477
8478         /*
8479          * src and dst are inverted, think how the receiver sees them
8480          *
8481          * The input is broken into two sections, a non-compressed section
8482          * containing vm_pool, vlan_id, and flow_type.  The rest of the data
8483          * is XORed together and stored in the compressed dword.
8484          */
8485         input.formatted.vlan_id = vlan_id;
8486
8487         /*
8488          * since src port and flex bytes occupy the same word XOR them together
8489          * and write the value to source port portion of compressed dword
8490          */
8491         if (first->tx_flags & (IXGBE_TX_FLAGS_SW_VLAN | IXGBE_TX_FLAGS_HW_VLAN))
8492                 common.port.src ^= th->dest ^ htons(ETH_P_8021Q);
8493         else
8494                 common.port.src ^= th->dest ^ first->protocol;
8495         common.port.dst ^= th->source;
8496
8497         switch (hdr.ipv4->version) {
8498         case IPVERSION:
8499                 input.formatted.flow_type = IXGBE_ATR_FLOW_TYPE_TCPV4;
8500                 common.ip ^= hdr.ipv4->saddr ^ hdr.ipv4->daddr;
8501                 break;
8502         case 6:
8503                 input.formatted.flow_type = IXGBE_ATR_FLOW_TYPE_TCPV6;
8504                 common.ip ^= hdr.ipv6->saddr.s6_addr32[0] ^
8505                              hdr.ipv6->saddr.s6_addr32[1] ^
8506                              hdr.ipv6->saddr.s6_addr32[2] ^
8507                              hdr.ipv6->saddr.s6_addr32[3] ^
8508                              hdr.ipv6->daddr.s6_addr32[0] ^
8509                              hdr.ipv6->daddr.s6_addr32[1] ^
8510                              hdr.ipv6->daddr.s6_addr32[2] ^
8511                              hdr.ipv6->daddr.s6_addr32[3];
8512                 break;
8513         default:
8514                 break;
8515         }
8516
8517         if (hdr.network != skb_network_header(skb))
8518                 input.formatted.flow_type |= IXGBE_ATR_L4TYPE_TUNNEL_MASK;
8519
8520         /* This assumes the Rx queue and Tx queue are bound to the same CPU */
8521         ixgbe_fdir_add_signature_filter_82599(&q_vector->adapter->hw,
8522                                               input, common, ring->queue_index);
8523 }
8524
8525 #ifdef IXGBE_FCOE
8526 static u16 ixgbe_select_queue(struct net_device *dev, struct sk_buff *skb,
8527                               struct net_device *sb_dev)
8528 {
8529         struct ixgbe_adapter *adapter;
8530         struct ixgbe_ring_feature *f;
8531         int txq;
8532
8533         if (sb_dev) {
8534                 u8 tc = netdev_get_prio_tc_map(dev, skb->priority);
8535                 struct net_device *vdev = sb_dev;
8536
8537                 txq = vdev->tc_to_txq[tc].offset;
8538                 txq += reciprocal_scale(skb_get_hash(skb),
8539                                         vdev->tc_to_txq[tc].count);
8540
8541                 return txq;
8542         }
8543
8544         /*
8545          * only execute the code below if protocol is FCoE
8546          * or FIP and we have FCoE enabled on the adapter
8547          */
8548         switch (vlan_get_protocol(skb)) {
8549         case htons(ETH_P_FCOE):
8550         case htons(ETH_P_FIP):
8551                 adapter = netdev_priv(dev);
8552
8553                 if (!sb_dev && (adapter->flags & IXGBE_FLAG_FCOE_ENABLED))
8554                         break;
8555                 fallthrough;
8556         default:
8557                 return netdev_pick_tx(dev, skb, sb_dev);
8558         }
8559
8560         f = &adapter->ring_feature[RING_F_FCOE];
8561
8562         txq = skb_rx_queue_recorded(skb) ? skb_get_rx_queue(skb) :
8563                                            smp_processor_id();
8564
8565         while (txq >= f->indices)
8566                 txq -= f->indices;
8567
8568         return txq + f->offset;
8569 }
8570
8571 #endif
8572 int ixgbe_xmit_xdp_ring(struct ixgbe_ring *ring,
8573                         struct xdp_frame *xdpf)
8574 {
8575         struct skb_shared_info *sinfo = xdp_get_shared_info_from_frame(xdpf);
8576         u8 nr_frags = unlikely(xdp_frame_has_frags(xdpf)) ? sinfo->nr_frags : 0;
8577         u16 i = 0, index = ring->next_to_use;
8578         struct ixgbe_tx_buffer *tx_head = &ring->tx_buffer_info[index];
8579         struct ixgbe_tx_buffer *tx_buff = tx_head;
8580         union ixgbe_adv_tx_desc *tx_desc = IXGBE_TX_DESC(ring, index);
8581         u32 cmd_type, len = xdpf->len;
8582         void *data = xdpf->data;
8583
8584         if (unlikely(ixgbe_desc_unused(ring) < 1 + nr_frags))
8585                 return IXGBE_XDP_CONSUMED;
8586
8587         tx_head->bytecount = xdp_get_frame_len(xdpf);
8588         tx_head->gso_segs = 1;
8589         tx_head->xdpf = xdpf;
8590
8591         tx_desc->read.olinfo_status =
8592                 cpu_to_le32(tx_head->bytecount << IXGBE_ADVTXD_PAYLEN_SHIFT);
8593
8594         for (;;) {
8595                 dma_addr_t dma;
8596
8597                 dma = dma_map_single(ring->dev, data, len, DMA_TO_DEVICE);
8598                 if (dma_mapping_error(ring->dev, dma))
8599                         goto unmap;
8600
8601                 dma_unmap_len_set(tx_buff, len, len);
8602                 dma_unmap_addr_set(tx_buff, dma, dma);
8603
8604                 cmd_type = IXGBE_ADVTXD_DTYP_DATA | IXGBE_ADVTXD_DCMD_DEXT |
8605                            IXGBE_ADVTXD_DCMD_IFCS | len;
8606                 tx_desc->read.cmd_type_len = cpu_to_le32(cmd_type);
8607                 tx_desc->read.buffer_addr = cpu_to_le64(dma);
8608                 tx_buff->protocol = 0;
8609
8610                 if (++index == ring->count)
8611                         index = 0;
8612
8613                 if (i == nr_frags)
8614                         break;
8615
8616                 tx_buff = &ring->tx_buffer_info[index];
8617                 tx_desc = IXGBE_TX_DESC(ring, index);
8618                 tx_desc->read.olinfo_status = 0;
8619
8620                 data = skb_frag_address(&sinfo->frags[i]);
8621                 len = skb_frag_size(&sinfo->frags[i]);
8622                 i++;
8623         }
8624         /* put descriptor type bits */
8625         tx_desc->read.cmd_type_len |= cpu_to_le32(IXGBE_TXD_CMD);
8626
8627         /* Avoid any potential race with xdp_xmit and cleanup */
8628         smp_wmb();
8629
8630         tx_head->next_to_watch = tx_desc;
8631         ring->next_to_use = index;
8632
8633         return IXGBE_XDP_TX;
8634
8635 unmap:
8636         for (;;) {
8637                 tx_buff = &ring->tx_buffer_info[index];
8638                 if (dma_unmap_len(tx_buff, len))
8639                         dma_unmap_page(ring->dev, dma_unmap_addr(tx_buff, dma),
8640                                        dma_unmap_len(tx_buff, len),
8641                                        DMA_TO_DEVICE);
8642                 dma_unmap_len_set(tx_buff, len, 0);
8643                 if (tx_buff == tx_head)
8644                         break;
8645
8646                 if (!index)
8647                         index += ring->count;
8648                 index--;
8649         }
8650
8651         return IXGBE_XDP_CONSUMED;
8652 }
8653
8654 netdev_tx_t ixgbe_xmit_frame_ring(struct sk_buff *skb,
8655                           struct ixgbe_adapter *adapter,
8656                           struct ixgbe_ring *tx_ring)
8657 {
8658         struct ixgbe_tx_buffer *first;
8659         int tso;
8660         u32 tx_flags = 0;
8661         unsigned short f;
8662         u16 count = TXD_USE_COUNT(skb_headlen(skb));
8663         struct ixgbe_ipsec_tx_data ipsec_tx = { 0 };
8664         __be16 protocol = skb->protocol;
8665         u8 hdr_len = 0;
8666
8667         /*
8668          * need: 1 descriptor per page * PAGE_SIZE/IXGBE_MAX_DATA_PER_TXD,
8669          *       + 1 desc for skb_headlen/IXGBE_MAX_DATA_PER_TXD,
8670          *       + 2 desc gap to keep tail from touching head,
8671          *       + 1 desc for context descriptor,
8672          * otherwise try next time
8673          */
8674         for (f = 0; f < skb_shinfo(skb)->nr_frags; f++)
8675                 count += TXD_USE_COUNT(skb_frag_size(
8676                                                 &skb_shinfo(skb)->frags[f]));
8677
8678         if (ixgbe_maybe_stop_tx(tx_ring, count + 3)) {
8679                 tx_ring->tx_stats.tx_busy++;
8680                 return NETDEV_TX_BUSY;
8681         }
8682
8683         /* record the location of the first descriptor for this packet */
8684         first = &tx_ring->tx_buffer_info[tx_ring->next_to_use];
8685         first->skb = skb;
8686         first->bytecount = skb->len;
8687         first->gso_segs = 1;
8688
8689         /* if we have a HW VLAN tag being added default to the HW one */
8690         if (skb_vlan_tag_present(skb)) {
8691                 tx_flags |= skb_vlan_tag_get(skb) << IXGBE_TX_FLAGS_VLAN_SHIFT;
8692                 tx_flags |= IXGBE_TX_FLAGS_HW_VLAN;
8693         /* else if it is a SW VLAN check the next protocol and store the tag */
8694         } else if (protocol == htons(ETH_P_8021Q)) {
8695                 struct vlan_hdr *vhdr, _vhdr;
8696                 vhdr = skb_header_pointer(skb, ETH_HLEN, sizeof(_vhdr), &_vhdr);
8697                 if (!vhdr)
8698                         goto out_drop;
8699
8700                 tx_flags |= ntohs(vhdr->h_vlan_TCI) <<
8701                                   IXGBE_TX_FLAGS_VLAN_SHIFT;
8702                 tx_flags |= IXGBE_TX_FLAGS_SW_VLAN;
8703         }
8704         protocol = vlan_get_protocol(skb);
8705
8706         if (unlikely(skb_shinfo(skb)->tx_flags & SKBTX_HW_TSTAMP) &&
8707             adapter->ptp_clock) {
8708                 if (adapter->tstamp_config.tx_type == HWTSTAMP_TX_ON &&
8709                     !test_and_set_bit_lock(__IXGBE_PTP_TX_IN_PROGRESS,
8710                                            &adapter->state)) {
8711                         skb_shinfo(skb)->tx_flags |= SKBTX_IN_PROGRESS;
8712                         tx_flags |= IXGBE_TX_FLAGS_TSTAMP;
8713
8714                         /* schedule check for Tx timestamp */
8715                         adapter->ptp_tx_skb = skb_get(skb);
8716                         adapter->ptp_tx_start = jiffies;
8717                         schedule_work(&adapter->ptp_tx_work);
8718                 } else {
8719                         adapter->tx_hwtstamp_skipped++;
8720                 }
8721         }
8722
8723 #ifdef CONFIG_PCI_IOV
8724         /*
8725          * Use the l2switch_enable flag - would be false if the DMA
8726          * Tx switch had been disabled.
8727          */
8728         if (adapter->flags & IXGBE_FLAG_SRIOV_ENABLED)
8729                 tx_flags |= IXGBE_TX_FLAGS_CC;
8730
8731 #endif
8732         /* DCB maps skb priorities 0-7 onto 3 bit PCP of VLAN tag. */
8733         if ((adapter->flags & IXGBE_FLAG_DCB_ENABLED) &&
8734             ((tx_flags & (IXGBE_TX_FLAGS_HW_VLAN | IXGBE_TX_FLAGS_SW_VLAN)) ||
8735              (skb->priority != TC_PRIO_CONTROL))) {
8736                 tx_flags &= ~IXGBE_TX_FLAGS_VLAN_PRIO_MASK;
8737                 tx_flags |= (skb->priority & 0x7) <<
8738                                         IXGBE_TX_FLAGS_VLAN_PRIO_SHIFT;
8739                 if (tx_flags & IXGBE_TX_FLAGS_SW_VLAN) {
8740                         struct vlan_ethhdr *vhdr;
8741
8742                         if (skb_cow_head(skb, 0))
8743                                 goto out_drop;
8744                         vhdr = (struct vlan_ethhdr *)skb->data;
8745                         vhdr->h_vlan_TCI = htons(tx_flags >>
8746                                                  IXGBE_TX_FLAGS_VLAN_SHIFT);
8747                 } else {
8748                         tx_flags |= IXGBE_TX_FLAGS_HW_VLAN;
8749                 }
8750         }
8751
8752         /* record initial flags and protocol */
8753         first->tx_flags = tx_flags;
8754         first->protocol = protocol;
8755
8756 #ifdef IXGBE_FCOE
8757         /* setup tx offload for FCoE */
8758         if ((protocol == htons(ETH_P_FCOE)) &&
8759             (tx_ring->netdev->features & (NETIF_F_FSO | NETIF_F_FCOE_CRC))) {
8760                 tso = ixgbe_fso(tx_ring, first, &hdr_len);
8761                 if (tso < 0)
8762                         goto out_drop;
8763
8764                 goto xmit_fcoe;
8765         }
8766
8767 #endif /* IXGBE_FCOE */
8768
8769 #ifdef CONFIG_IXGBE_IPSEC
8770         if (xfrm_offload(skb) &&
8771             !ixgbe_ipsec_tx(tx_ring, first, &ipsec_tx))
8772                 goto out_drop;
8773 #endif
8774         tso = ixgbe_tso(tx_ring, first, &hdr_len, &ipsec_tx);
8775         if (tso < 0)
8776                 goto out_drop;
8777         else if (!tso)
8778                 ixgbe_tx_csum(tx_ring, first, &ipsec_tx);
8779
8780         /* add the ATR filter if ATR is on */
8781         if (test_bit(__IXGBE_TX_FDIR_INIT_DONE, &tx_ring->state))
8782                 ixgbe_atr(tx_ring, first);
8783
8784 #ifdef IXGBE_FCOE
8785 xmit_fcoe:
8786 #endif /* IXGBE_FCOE */
8787         if (ixgbe_tx_map(tx_ring, first, hdr_len))
8788                 goto cleanup_tx_timestamp;
8789
8790         return NETDEV_TX_OK;
8791
8792 out_drop:
8793         dev_kfree_skb_any(first->skb);
8794         first->skb = NULL;
8795 cleanup_tx_timestamp:
8796         if (unlikely(tx_flags & IXGBE_TX_FLAGS_TSTAMP)) {
8797                 dev_kfree_skb_any(adapter->ptp_tx_skb);
8798                 adapter->ptp_tx_skb = NULL;
8799                 cancel_work_sync(&adapter->ptp_tx_work);
8800                 clear_bit_unlock(__IXGBE_PTP_TX_IN_PROGRESS, &adapter->state);
8801         }
8802
8803         return NETDEV_TX_OK;
8804 }
8805
8806 static netdev_tx_t __ixgbe_xmit_frame(struct sk_buff *skb,
8807                                       struct net_device *netdev,
8808                                       struct ixgbe_ring *ring)
8809 {
8810         struct ixgbe_adapter *adapter = netdev_priv(netdev);
8811         struct ixgbe_ring *tx_ring;
8812
8813         /*
8814          * The minimum packet size for olinfo paylen is 17 so pad the skb
8815          * in order to meet this minimum size requirement.
8816          */
8817         if (skb_put_padto(skb, 17))
8818                 return NETDEV_TX_OK;
8819
8820         tx_ring = ring ? ring : adapter->tx_ring[skb_get_queue_mapping(skb)];
8821         if (unlikely(test_bit(__IXGBE_TX_DISABLED, &tx_ring->state)))
8822                 return NETDEV_TX_BUSY;
8823
8824         return ixgbe_xmit_frame_ring(skb, adapter, tx_ring);
8825 }
8826
8827 static netdev_tx_t ixgbe_xmit_frame(struct sk_buff *skb,
8828                                     struct net_device *netdev)
8829 {
8830         return __ixgbe_xmit_frame(skb, netdev, NULL);
8831 }
8832
8833 /**
8834  * ixgbe_set_mac - Change the Ethernet Address of the NIC
8835  * @netdev: network interface device structure
8836  * @p: pointer to an address structure
8837  *
8838  * Returns 0 on success, negative on failure
8839  **/
8840 static int ixgbe_set_mac(struct net_device *netdev, void *p)
8841 {
8842         struct ixgbe_adapter *adapter = netdev_priv(netdev);
8843         struct ixgbe_hw *hw = &adapter->hw;
8844         struct sockaddr *addr = p;
8845
8846         if (!is_valid_ether_addr(addr->sa_data))
8847                 return -EADDRNOTAVAIL;
8848
8849         eth_hw_addr_set(netdev, addr->sa_data);
8850         memcpy(hw->mac.addr, addr->sa_data, netdev->addr_len);
8851
8852         ixgbe_mac_set_default_filter(adapter);
8853
8854         return 0;
8855 }
8856
8857 static int
8858 ixgbe_mdio_read(struct net_device *netdev, int prtad, int devad, u16 addr)
8859 {
8860         struct ixgbe_adapter *adapter = netdev_priv(netdev);
8861         struct ixgbe_hw *hw = &adapter->hw;
8862         u16 value;
8863         int rc;
8864
8865         if (adapter->mii_bus) {
8866                 int regnum = addr;
8867
8868                 if (devad != MDIO_DEVAD_NONE)
8869                         regnum |= (devad << 16) | MII_ADDR_C45;
8870
8871                 return mdiobus_read(adapter->mii_bus, prtad, regnum);
8872         }
8873
8874         if (prtad != hw->phy.mdio.prtad)
8875                 return -EINVAL;
8876         rc = hw->phy.ops.read_reg(hw, addr, devad, &value);
8877         if (!rc)
8878                 rc = value;
8879         return rc;
8880 }
8881
8882 static int ixgbe_mdio_write(struct net_device *netdev, int prtad, int devad,
8883                             u16 addr, u16 value)
8884 {
8885         struct ixgbe_adapter *adapter = netdev_priv(netdev);
8886         struct ixgbe_hw *hw = &adapter->hw;
8887
8888         if (adapter->mii_bus) {
8889                 int regnum = addr;
8890
8891                 if (devad != MDIO_DEVAD_NONE)
8892                         regnum |= (devad << 16) | MII_ADDR_C45;
8893
8894                 return mdiobus_write(adapter->mii_bus, prtad, regnum, value);
8895         }
8896
8897         if (prtad != hw->phy.mdio.prtad)
8898                 return -EINVAL;
8899         return hw->phy.ops.write_reg(hw, addr, devad, value);
8900 }
8901
8902 static int ixgbe_ioctl(struct net_device *netdev, struct ifreq *req, int cmd)
8903 {
8904         struct ixgbe_adapter *adapter = netdev_priv(netdev);
8905
8906         switch (cmd) {
8907         case SIOCSHWTSTAMP:
8908                 return ixgbe_ptp_set_ts_config(adapter, req);
8909         case SIOCGHWTSTAMP:
8910                 return ixgbe_ptp_get_ts_config(adapter, req);
8911         case SIOCGMIIPHY:
8912                 if (!adapter->hw.phy.ops.read_reg)
8913                         return -EOPNOTSUPP;
8914                 fallthrough;
8915         default:
8916                 return mdio_mii_ioctl(&adapter->hw.phy.mdio, if_mii(req), cmd);
8917         }
8918 }
8919
8920 /**
8921  * ixgbe_add_sanmac_netdev - Add the SAN MAC address to the corresponding
8922  * netdev->dev_addrs
8923  * @dev: network interface device structure
8924  *
8925  * Returns non-zero on failure
8926  **/
8927 static int ixgbe_add_sanmac_netdev(struct net_device *dev)
8928 {
8929         int err = 0;
8930         struct ixgbe_adapter *adapter = netdev_priv(dev);
8931         struct ixgbe_hw *hw = &adapter->hw;
8932
8933         if (is_valid_ether_addr(hw->mac.san_addr)) {
8934                 rtnl_lock();
8935                 err = dev_addr_add(dev, hw->mac.san_addr, NETDEV_HW_ADDR_T_SAN);
8936                 rtnl_unlock();
8937
8938                 /* update SAN MAC vmdq pool selection */
8939                 hw->mac.ops.set_vmdq_san_mac(hw, VMDQ_P(0));
8940         }
8941         return err;
8942 }
8943
8944 /**
8945  * ixgbe_del_sanmac_netdev - Removes the SAN MAC address to the corresponding
8946  * netdev->dev_addrs
8947  * @dev: network interface device structure
8948  *
8949  * Returns non-zero on failure
8950  **/
8951 static int ixgbe_del_sanmac_netdev(struct net_device *dev)
8952 {
8953         int err = 0;
8954         struct ixgbe_adapter *adapter = netdev_priv(dev);
8955         struct ixgbe_mac_info *mac = &adapter->hw.mac;
8956
8957         if (is_valid_ether_addr(mac->san_addr)) {
8958                 rtnl_lock();
8959                 err = dev_addr_del(dev, mac->san_addr, NETDEV_HW_ADDR_T_SAN);
8960                 rtnl_unlock();
8961         }
8962         return err;
8963 }
8964
8965 static void ixgbe_get_ring_stats64(struct rtnl_link_stats64 *stats,
8966                                    struct ixgbe_ring *ring)
8967 {
8968         u64 bytes, packets;
8969         unsigned int start;
8970
8971         if (ring) {
8972                 do {
8973                         start = u64_stats_fetch_begin_irq(&ring->syncp);
8974                         packets = ring->stats.packets;
8975                         bytes   = ring->stats.bytes;
8976                 } while (u64_stats_fetch_retry_irq(&ring->syncp, start));
8977                 stats->tx_packets += packets;
8978                 stats->tx_bytes   += bytes;
8979         }
8980 }
8981
8982 static void ixgbe_get_stats64(struct net_device *netdev,
8983                               struct rtnl_link_stats64 *stats)
8984 {
8985         struct ixgbe_adapter *adapter = netdev_priv(netdev);
8986         int i;
8987
8988         rcu_read_lock();
8989         for (i = 0; i < adapter->num_rx_queues; i++) {
8990                 struct ixgbe_ring *ring = READ_ONCE(adapter->rx_ring[i]);
8991                 u64 bytes, packets;
8992                 unsigned int start;
8993
8994                 if (ring) {
8995                         do {
8996                                 start = u64_stats_fetch_begin_irq(&ring->syncp);
8997                                 packets = ring->stats.packets;
8998                                 bytes   = ring->stats.bytes;
8999                         } while (u64_stats_fetch_retry_irq(&ring->syncp, start));
9000                         stats->rx_packets += packets;
9001                         stats->rx_bytes   += bytes;
9002                 }
9003         }
9004
9005         for (i = 0; i < adapter->num_tx_queues; i++) {
9006                 struct ixgbe_ring *ring = READ_ONCE(adapter->tx_ring[i]);
9007
9008                 ixgbe_get_ring_stats64(stats, ring);
9009         }
9010         for (i = 0; i < adapter->num_xdp_queues; i++) {
9011                 struct ixgbe_ring *ring = READ_ONCE(adapter->xdp_ring[i]);
9012
9013                 ixgbe_get_ring_stats64(stats, ring);
9014         }
9015         rcu_read_unlock();
9016
9017         /* following stats updated by ixgbe_watchdog_task() */
9018         stats->multicast        = netdev->stats.multicast;
9019         stats->rx_errors        = netdev->stats.rx_errors;
9020         stats->rx_length_errors = netdev->stats.rx_length_errors;
9021         stats->rx_crc_errors    = netdev->stats.rx_crc_errors;
9022         stats->rx_missed_errors = netdev->stats.rx_missed_errors;
9023 }
9024
9025 #ifdef CONFIG_IXGBE_DCB
9026 /**
9027  * ixgbe_validate_rtr - verify 802.1Qp to Rx packet buffer mapping is valid.
9028  * @adapter: pointer to ixgbe_adapter
9029  * @tc: number of traffic classes currently enabled
9030  *
9031  * Configure a valid 802.1Qp to Rx packet buffer mapping ie confirm
9032  * 802.1Q priority maps to a packet buffer that exists.
9033  */
9034 static void ixgbe_validate_rtr(struct ixgbe_adapter *adapter, u8 tc)
9035 {
9036         struct ixgbe_hw *hw = &adapter->hw;
9037         u32 reg, rsave;
9038         int i;
9039
9040         /* 82598 have a static priority to TC mapping that can not
9041          * be changed so no validation is needed.
9042          */
9043         if (hw->mac.type == ixgbe_mac_82598EB)
9044                 return;
9045
9046         reg = IXGBE_READ_REG(hw, IXGBE_RTRUP2TC);
9047         rsave = reg;
9048
9049         for (i = 0; i < MAX_TRAFFIC_CLASS; i++) {
9050                 u8 up2tc = reg >> (i * IXGBE_RTRUP2TC_UP_SHIFT);
9051
9052                 /* If up2tc is out of bounds default to zero */
9053                 if (up2tc > tc)
9054                         reg &= ~(0x7 << IXGBE_RTRUP2TC_UP_SHIFT);
9055         }
9056
9057         if (reg != rsave)
9058                 IXGBE_WRITE_REG(hw, IXGBE_RTRUP2TC, reg);
9059
9060         return;
9061 }
9062
9063 /**
9064  * ixgbe_set_prio_tc_map - Configure netdev prio tc map
9065  * @adapter: Pointer to adapter struct
9066  *
9067  * Populate the netdev user priority to tc map
9068  */
9069 static void ixgbe_set_prio_tc_map(struct ixgbe_adapter *adapter)
9070 {
9071         struct net_device *dev = adapter->netdev;
9072         struct ixgbe_dcb_config *dcb_cfg = &adapter->dcb_cfg;
9073         struct ieee_ets *ets = adapter->ixgbe_ieee_ets;
9074         u8 prio;
9075
9076         for (prio = 0; prio < MAX_USER_PRIORITY; prio++) {
9077                 u8 tc = 0;
9078
9079                 if (adapter->dcbx_cap & DCB_CAP_DCBX_VER_CEE)
9080                         tc = ixgbe_dcb_get_tc_from_up(dcb_cfg, 0, prio);
9081                 else if (ets)
9082                         tc = ets->prio_tc[prio];
9083
9084                 netdev_set_prio_tc_map(dev, prio, tc);
9085         }
9086 }
9087
9088 #endif /* CONFIG_IXGBE_DCB */
9089 static int ixgbe_reassign_macvlan_pool(struct net_device *vdev,
9090                                        struct netdev_nested_priv *priv)
9091 {
9092         struct ixgbe_adapter *adapter = (struct ixgbe_adapter *)priv->data;
9093         struct ixgbe_fwd_adapter *accel;
9094         int pool;
9095
9096         /* we only care about macvlans... */
9097         if (!netif_is_macvlan(vdev))
9098                 return 0;
9099
9100         /* that have hardware offload enabled... */
9101         accel = macvlan_accel_priv(vdev);
9102         if (!accel)
9103                 return 0;
9104
9105         /* If we can relocate to a different bit do so */
9106         pool = find_first_zero_bit(adapter->fwd_bitmask, adapter->num_rx_pools);
9107         if (pool < adapter->num_rx_pools) {
9108                 set_bit(pool, adapter->fwd_bitmask);
9109                 accel->pool = pool;
9110                 return 0;
9111         }
9112
9113         /* if we cannot find a free pool then disable the offload */
9114         netdev_err(vdev, "L2FW offload disabled due to lack of queue resources\n");
9115         macvlan_release_l2fw_offload(vdev);
9116
9117         /* unbind the queues and drop the subordinate channel config */
9118         netdev_unbind_sb_channel(adapter->netdev, vdev);
9119         netdev_set_sb_channel(vdev, 0);
9120
9121         kfree(accel);
9122
9123         return 0;
9124 }
9125
9126 static void ixgbe_defrag_macvlan_pools(struct net_device *dev)
9127 {
9128         struct ixgbe_adapter *adapter = netdev_priv(dev);
9129         struct netdev_nested_priv priv = {
9130                 .data = (void *)adapter,
9131         };
9132
9133         /* flush any stale bits out of the fwd bitmask */
9134         bitmap_clear(adapter->fwd_bitmask, 1, 63);
9135
9136         /* walk through upper devices reassigning pools */
9137         netdev_walk_all_upper_dev_rcu(dev, ixgbe_reassign_macvlan_pool,
9138                                       &priv);
9139 }
9140
9141 /**
9142  * ixgbe_setup_tc - configure net_device for multiple traffic classes
9143  *
9144  * @dev: net device to configure
9145  * @tc: number of traffic classes to enable
9146  */
9147 int ixgbe_setup_tc(struct net_device *dev, u8 tc)
9148 {
9149         struct ixgbe_adapter *adapter = netdev_priv(dev);
9150         struct ixgbe_hw *hw = &adapter->hw;
9151
9152         /* Hardware supports up to 8 traffic classes */
9153         if (tc > adapter->dcb_cfg.num_tcs.pg_tcs)
9154                 return -EINVAL;
9155
9156         if (hw->mac.type == ixgbe_mac_82598EB && tc && tc < MAX_TRAFFIC_CLASS)
9157                 return -EINVAL;
9158
9159         /* Hardware has to reinitialize queues and interrupts to
9160          * match packet buffer alignment. Unfortunately, the
9161          * hardware is not flexible enough to do this dynamically.
9162          */
9163         if (netif_running(dev))
9164                 ixgbe_close(dev);
9165         else
9166                 ixgbe_reset(adapter);
9167
9168         ixgbe_clear_interrupt_scheme(adapter);
9169
9170 #ifdef CONFIG_IXGBE_DCB
9171         if (tc) {
9172                 if (adapter->xdp_prog) {
9173                         e_warn(probe, "DCB is not supported with XDP\n");
9174
9175                         ixgbe_init_interrupt_scheme(adapter);
9176                         if (netif_running(dev))
9177                                 ixgbe_open(dev);
9178                         return -EINVAL;
9179                 }
9180
9181                 netdev_set_num_tc(dev, tc);
9182                 ixgbe_set_prio_tc_map(adapter);
9183
9184                 adapter->hw_tcs = tc;
9185                 adapter->flags |= IXGBE_FLAG_DCB_ENABLED;
9186
9187                 if (adapter->hw.mac.type == ixgbe_mac_82598EB) {
9188                         adapter->last_lfc_mode = adapter->hw.fc.requested_mode;
9189                         adapter->hw.fc.requested_mode = ixgbe_fc_none;
9190                 }
9191         } else {
9192                 netdev_reset_tc(dev);
9193
9194                 if (adapter->hw.mac.type == ixgbe_mac_82598EB)
9195                         adapter->hw.fc.requested_mode = adapter->last_lfc_mode;
9196
9197                 adapter->flags &= ~IXGBE_FLAG_DCB_ENABLED;
9198                 adapter->hw_tcs = tc;
9199
9200                 adapter->temp_dcb_cfg.pfc_mode_enable = false;
9201                 adapter->dcb_cfg.pfc_mode_enable = false;
9202         }
9203
9204         ixgbe_validate_rtr(adapter, tc);
9205
9206 #endif /* CONFIG_IXGBE_DCB */
9207         ixgbe_init_interrupt_scheme(adapter);
9208
9209         ixgbe_defrag_macvlan_pools(dev);
9210
9211         if (netif_running(dev))
9212                 return ixgbe_open(dev);
9213
9214         return 0;
9215 }
9216
9217 static int ixgbe_delete_clsu32(struct ixgbe_adapter *adapter,
9218                                struct tc_cls_u32_offload *cls)
9219 {
9220         u32 hdl = cls->knode.handle;
9221         u32 uhtid = TC_U32_USERHTID(cls->knode.handle);
9222         u32 loc = cls->knode.handle & 0xfffff;
9223         int err = 0, i, j;
9224         struct ixgbe_jump_table *jump = NULL;
9225
9226         if (loc > IXGBE_MAX_HW_ENTRIES)
9227                 return -EINVAL;
9228
9229         if ((uhtid != 0x800) && (uhtid >= IXGBE_MAX_LINK_HANDLE))
9230                 return -EINVAL;
9231
9232         /* Clear this filter in the link data it is associated with */
9233         if (uhtid != 0x800) {
9234                 jump = adapter->jump_tables[uhtid];
9235                 if (!jump)
9236                         return -EINVAL;
9237                 if (!test_bit(loc - 1, jump->child_loc_map))
9238                         return -EINVAL;
9239                 clear_bit(loc - 1, jump->child_loc_map);
9240         }
9241
9242         /* Check if the filter being deleted is a link */
9243         for (i = 1; i < IXGBE_MAX_LINK_HANDLE; i++) {
9244                 jump = adapter->jump_tables[i];
9245                 if (jump && jump->link_hdl == hdl) {
9246                         /* Delete filters in the hardware in the child hash
9247                          * table associated with this link
9248                          */
9249                         for (j = 0; j < IXGBE_MAX_HW_ENTRIES; j++) {
9250                                 if (!test_bit(j, jump->child_loc_map))
9251                                         continue;
9252                                 spin_lock(&adapter->fdir_perfect_lock);
9253                                 err = ixgbe_update_ethtool_fdir_entry(adapter,
9254                                                                       NULL,
9255                                                                       j + 1);
9256                                 spin_unlock(&adapter->fdir_perfect_lock);
9257                                 clear_bit(j, jump->child_loc_map);
9258                         }
9259                         /* Remove resources for this link */
9260                         kfree(jump->input);
9261                         kfree(jump->mask);
9262                         kfree(jump);
9263                         adapter->jump_tables[i] = NULL;
9264                         return err;
9265                 }
9266         }
9267
9268         spin_lock(&adapter->fdir_perfect_lock);
9269         err = ixgbe_update_ethtool_fdir_entry(adapter, NULL, loc);
9270         spin_unlock(&adapter->fdir_perfect_lock);
9271         return err;
9272 }
9273
9274 static int ixgbe_configure_clsu32_add_hnode(struct ixgbe_adapter *adapter,
9275                                             struct tc_cls_u32_offload *cls)
9276 {
9277         u32 uhtid = TC_U32_USERHTID(cls->hnode.handle);
9278
9279         if (uhtid >= IXGBE_MAX_LINK_HANDLE)
9280                 return -EINVAL;
9281
9282         /* This ixgbe devices do not support hash tables at the moment
9283          * so abort when given hash tables.
9284          */
9285         if (cls->hnode.divisor > 0)
9286                 return -EINVAL;
9287
9288         set_bit(uhtid - 1, &adapter->tables);
9289         return 0;
9290 }
9291
9292 static int ixgbe_configure_clsu32_del_hnode(struct ixgbe_adapter *adapter,
9293                                             struct tc_cls_u32_offload *cls)
9294 {
9295         u32 uhtid = TC_U32_USERHTID(cls->hnode.handle);
9296
9297         if (uhtid >= IXGBE_MAX_LINK_HANDLE)
9298                 return -EINVAL;
9299
9300         clear_bit(uhtid - 1, &adapter->tables);
9301         return 0;
9302 }
9303
9304 #ifdef CONFIG_NET_CLS_ACT
9305 struct upper_walk_data {
9306         struct ixgbe_adapter *adapter;
9307         u64 action;
9308         int ifindex;
9309         u8 queue;
9310 };
9311
9312 static int get_macvlan_queue(struct net_device *upper,
9313                              struct netdev_nested_priv *priv)
9314 {
9315         if (netif_is_macvlan(upper)) {
9316                 struct ixgbe_fwd_adapter *vadapter = macvlan_accel_priv(upper);
9317                 struct ixgbe_adapter *adapter;
9318                 struct upper_walk_data *data;
9319                 int ifindex;
9320
9321                 data = (struct upper_walk_data *)priv->data;
9322                 ifindex = data->ifindex;
9323                 adapter = data->adapter;
9324                 if (vadapter && upper->ifindex == ifindex) {
9325                         data->queue = adapter->rx_ring[vadapter->rx_base_queue]->reg_idx;
9326                         data->action = data->queue;
9327                         return 1;
9328                 }
9329         }
9330
9331         return 0;
9332 }
9333
9334 static int handle_redirect_action(struct ixgbe_adapter *adapter, int ifindex,
9335                                   u8 *queue, u64 *action)
9336 {
9337         struct ixgbe_ring_feature *vmdq = &adapter->ring_feature[RING_F_VMDQ];
9338         unsigned int num_vfs = adapter->num_vfs, vf;
9339         struct netdev_nested_priv priv;
9340         struct upper_walk_data data;
9341         struct net_device *upper;
9342
9343         /* redirect to a SRIOV VF */
9344         for (vf = 0; vf < num_vfs; ++vf) {
9345                 upper = pci_get_drvdata(adapter->vfinfo[vf].vfdev);
9346                 if (upper->ifindex == ifindex) {
9347                         *queue = vf * __ALIGN_MASK(1, ~vmdq->mask);
9348                         *action = vf + 1;
9349                         *action <<= ETHTOOL_RX_FLOW_SPEC_RING_VF_OFF;
9350                         return 0;
9351                 }
9352         }
9353
9354         /* redirect to a offloaded macvlan netdev */
9355         data.adapter = adapter;
9356         data.ifindex = ifindex;
9357         data.action = 0;
9358         data.queue = 0;
9359         priv.data = (void *)&data;
9360         if (netdev_walk_all_upper_dev_rcu(adapter->netdev,
9361                                           get_macvlan_queue, &priv)) {
9362                 *action = data.action;
9363                 *queue = data.queue;
9364
9365                 return 0;
9366         }
9367
9368         return -EINVAL;
9369 }
9370
9371 static int parse_tc_actions(struct ixgbe_adapter *adapter,
9372                             struct tcf_exts *exts, u64 *action, u8 *queue)
9373 {
9374         const struct tc_action *a;
9375         int i;
9376
9377         if (!tcf_exts_has_actions(exts))
9378                 return -EINVAL;
9379
9380         tcf_exts_for_each_action(i, a, exts) {
9381                 /* Drop action */
9382                 if (is_tcf_gact_shot(a)) {
9383                         *action = IXGBE_FDIR_DROP_QUEUE;
9384                         *queue = IXGBE_FDIR_DROP_QUEUE;
9385                         return 0;
9386                 }
9387
9388                 /* Redirect to a VF or a offloaded macvlan */
9389                 if (is_tcf_mirred_egress_redirect(a)) {
9390                         struct net_device *dev = tcf_mirred_dev(a);
9391
9392                         if (!dev)
9393                                 return -EINVAL;
9394                         return handle_redirect_action(adapter, dev->ifindex,
9395                                                       queue, action);
9396                 }
9397
9398                 return -EINVAL;
9399         }
9400
9401         return -EINVAL;
9402 }
9403 #else
9404 static int parse_tc_actions(struct ixgbe_adapter *adapter,
9405                             struct tcf_exts *exts, u64 *action, u8 *queue)
9406 {
9407         return -EINVAL;
9408 }
9409 #endif /* CONFIG_NET_CLS_ACT */
9410
9411 static int ixgbe_clsu32_build_input(struct ixgbe_fdir_filter *input,
9412                                     union ixgbe_atr_input *mask,
9413                                     struct tc_cls_u32_offload *cls,
9414                                     struct ixgbe_mat_field *field_ptr,
9415                                     struct ixgbe_nexthdr *nexthdr)
9416 {
9417         int i, j, off;
9418         __be32 val, m;
9419         bool found_entry = false, found_jump_field = false;
9420
9421         for (i = 0; i < cls->knode.sel->nkeys; i++) {
9422                 off = cls->knode.sel->keys[i].off;
9423                 val = cls->knode.sel->keys[i].val;
9424                 m = cls->knode.sel->keys[i].mask;
9425
9426                 for (j = 0; field_ptr[j].val; j++) {
9427                         if (field_ptr[j].off == off) {
9428                                 field_ptr[j].val(input, mask, (__force u32)val,
9429                                                  (__force u32)m);
9430                                 input->filter.formatted.flow_type |=
9431                                         field_ptr[j].type;
9432                                 found_entry = true;
9433                                 break;
9434                         }
9435                 }
9436                 if (nexthdr) {
9437                         if (nexthdr->off == cls->knode.sel->keys[i].off &&
9438                             nexthdr->val ==
9439                             (__force u32)cls->knode.sel->keys[i].val &&
9440                             nexthdr->mask ==
9441                             (__force u32)cls->knode.sel->keys[i].mask)
9442                                 found_jump_field = true;
9443                         else
9444                                 continue;
9445                 }
9446         }
9447
9448         if (nexthdr && !found_jump_field)
9449                 return -EINVAL;
9450
9451         if (!found_entry)
9452                 return 0;
9453
9454         mask->formatted.flow_type = IXGBE_ATR_L4TYPE_IPV6_MASK |
9455                                     IXGBE_ATR_L4TYPE_MASK;
9456
9457         if (input->filter.formatted.flow_type == IXGBE_ATR_FLOW_TYPE_IPV4)
9458                 mask->formatted.flow_type &= IXGBE_ATR_L4TYPE_IPV6_MASK;
9459
9460         return 0;
9461 }
9462
9463 static int ixgbe_configure_clsu32(struct ixgbe_adapter *adapter,
9464                                   struct tc_cls_u32_offload *cls)
9465 {
9466         __be16 protocol = cls->common.protocol;
9467         u32 loc = cls->knode.handle & 0xfffff;
9468         struct ixgbe_hw *hw = &adapter->hw;
9469         struct ixgbe_mat_field *field_ptr;
9470         struct ixgbe_fdir_filter *input = NULL;
9471         union ixgbe_atr_input *mask = NULL;
9472         struct ixgbe_jump_table *jump = NULL;
9473         int i, err = -EINVAL;
9474         u8 queue;
9475         u32 uhtid, link_uhtid;
9476
9477         uhtid = TC_U32_USERHTID(cls->knode.handle);
9478         link_uhtid = TC_U32_USERHTID(cls->knode.link_handle);
9479
9480         /* At the moment cls_u32 jumps to network layer and skips past
9481          * L2 headers. The canonical method to match L2 frames is to use
9482          * negative values. However this is error prone at best but really
9483          * just broken because there is no way to "know" what sort of hdr
9484          * is in front of the network layer. Fix cls_u32 to support L2
9485          * headers when needed.
9486          */
9487         if (protocol != htons(ETH_P_IP))
9488                 return err;
9489
9490         if (loc >= ((1024 << adapter->fdir_pballoc) - 2)) {
9491                 e_err(drv, "Location out of range\n");
9492                 return err;
9493         }
9494
9495         /* cls u32 is a graph starting at root node 0x800. The driver tracks
9496          * links and also the fields used to advance the parser across each
9497          * link (e.g. nexthdr/eat parameters from 'tc'). This way we can map
9498          * the u32 graph onto the hardware parse graph denoted in ixgbe_model.h
9499          * To add support for new nodes update ixgbe_model.h parse structures
9500          * this function _should_ be generic try not to hardcode values here.
9501          */
9502         if (uhtid == 0x800) {
9503                 field_ptr = (adapter->jump_tables[0])->mat;
9504         } else {
9505                 if (uhtid >= IXGBE_MAX_LINK_HANDLE)
9506                         return err;
9507                 if (!adapter->jump_tables[uhtid])
9508                         return err;
9509                 field_ptr = (adapter->jump_tables[uhtid])->mat;
9510         }
9511
9512         if (!field_ptr)
9513                 return err;
9514
9515         /* At this point we know the field_ptr is valid and need to either
9516          * build cls_u32 link or attach filter. Because adding a link to
9517          * a handle that does not exist is invalid and the same for adding
9518          * rules to handles that don't exist.
9519          */
9520
9521         if (link_uhtid) {
9522                 struct ixgbe_nexthdr *nexthdr = ixgbe_ipv4_jumps;
9523
9524                 if (link_uhtid >= IXGBE_MAX_LINK_HANDLE)
9525                         return err;
9526
9527                 if (!test_bit(link_uhtid - 1, &adapter->tables))
9528                         return err;
9529
9530                 /* Multiple filters as links to the same hash table are not
9531                  * supported. To add a new filter with the same next header
9532                  * but different match/jump conditions, create a new hash table
9533                  * and link to it.
9534                  */
9535                 if (adapter->jump_tables[link_uhtid] &&
9536                     (adapter->jump_tables[link_uhtid])->link_hdl) {
9537                         e_err(drv, "Link filter exists for link: %x\n",
9538                               link_uhtid);
9539                         return err;
9540                 }
9541
9542                 for (i = 0; nexthdr[i].jump; i++) {
9543                         if (nexthdr[i].o != cls->knode.sel->offoff ||
9544                             nexthdr[i].s != cls->knode.sel->offshift ||
9545                             nexthdr[i].m !=
9546                             (__force u32)cls->knode.sel->offmask)
9547                                 return err;
9548
9549                         jump = kzalloc(sizeof(*jump), GFP_KERNEL);
9550                         if (!jump)
9551                                 return -ENOMEM;
9552                         input = kzalloc(sizeof(*input), GFP_KERNEL);
9553                         if (!input) {
9554                                 err = -ENOMEM;
9555                                 goto free_jump;
9556                         }
9557                         mask = kzalloc(sizeof(*mask), GFP_KERNEL);
9558                         if (!mask) {
9559                                 err = -ENOMEM;
9560                                 goto free_input;
9561                         }
9562                         jump->input = input;
9563                         jump->mask = mask;
9564                         jump->link_hdl = cls->knode.handle;
9565
9566                         err = ixgbe_clsu32_build_input(input, mask, cls,
9567                                                        field_ptr, &nexthdr[i]);
9568                         if (!err) {
9569                                 jump->mat = nexthdr[i].jump;
9570                                 adapter->jump_tables[link_uhtid] = jump;
9571                                 break;
9572                         } else {
9573                                 kfree(mask);
9574                                 kfree(input);
9575                                 kfree(jump);
9576                         }
9577                 }
9578                 return 0;
9579         }
9580
9581         input = kzalloc(sizeof(*input), GFP_KERNEL);
9582         if (!input)
9583                 return -ENOMEM;
9584         mask = kzalloc(sizeof(*mask), GFP_KERNEL);
9585         if (!mask) {
9586                 err = -ENOMEM;
9587                 goto free_input;
9588         }
9589
9590         if ((uhtid != 0x800) && (adapter->jump_tables[uhtid])) {
9591                 if ((adapter->jump_tables[uhtid])->input)
9592                         memcpy(input, (adapter->jump_tables[uhtid])->input,
9593                                sizeof(*input));
9594                 if ((adapter->jump_tables[uhtid])->mask)
9595                         memcpy(mask, (adapter->jump_tables[uhtid])->mask,
9596                                sizeof(*mask));
9597
9598                 /* Lookup in all child hash tables if this location is already
9599                  * filled with a filter
9600                  */
9601                 for (i = 1; i < IXGBE_MAX_LINK_HANDLE; i++) {
9602                         struct ixgbe_jump_table *link = adapter->jump_tables[i];
9603
9604                         if (link && (test_bit(loc - 1, link->child_loc_map))) {
9605                                 e_err(drv, "Filter exists in location: %x\n",
9606                                       loc);
9607                                 err = -EINVAL;
9608                                 goto err_out;
9609                         }
9610                 }
9611         }
9612         err = ixgbe_clsu32_build_input(input, mask, cls, field_ptr, NULL);
9613         if (err)
9614                 goto err_out;
9615
9616         err = parse_tc_actions(adapter, cls->knode.exts, &input->action,
9617                                &queue);
9618         if (err < 0)
9619                 goto err_out;
9620
9621         input->sw_idx = loc;
9622
9623         spin_lock(&adapter->fdir_perfect_lock);
9624
9625         if (hlist_empty(&adapter->fdir_filter_list)) {
9626                 memcpy(&adapter->fdir_mask, mask, sizeof(*mask));
9627                 err = ixgbe_fdir_set_input_mask_82599(hw, mask);
9628                 if (err)
9629                         goto err_out_w_lock;
9630         } else if (memcmp(&adapter->fdir_mask, mask, sizeof(*mask))) {
9631                 err = -EINVAL;
9632                 goto err_out_w_lock;
9633         }
9634
9635         ixgbe_atr_compute_perfect_hash_82599(&input->filter, mask);
9636         err = ixgbe_fdir_write_perfect_filter_82599(hw, &input->filter,
9637                                                     input->sw_idx, queue);
9638         if (err)
9639                 goto err_out_w_lock;
9640
9641         ixgbe_update_ethtool_fdir_entry(adapter, input, input->sw_idx);
9642         spin_unlock(&adapter->fdir_perfect_lock);
9643
9644         if ((uhtid != 0x800) && (adapter->jump_tables[uhtid]))
9645                 set_bit(loc - 1, (adapter->jump_tables[uhtid])->child_loc_map);
9646
9647         kfree(mask);
9648         return err;
9649 err_out_w_lock:
9650         spin_unlock(&adapter->fdir_perfect_lock);
9651 err_out:
9652         kfree(mask);
9653 free_input:
9654         kfree(input);
9655 free_jump:
9656         kfree(jump);
9657         return err;
9658 }
9659
9660 static int ixgbe_setup_tc_cls_u32(struct ixgbe_adapter *adapter,
9661                                   struct tc_cls_u32_offload *cls_u32)
9662 {
9663         switch (cls_u32->command) {
9664         case TC_CLSU32_NEW_KNODE:
9665         case TC_CLSU32_REPLACE_KNODE:
9666                 return ixgbe_configure_clsu32(adapter, cls_u32);
9667         case TC_CLSU32_DELETE_KNODE:
9668                 return ixgbe_delete_clsu32(adapter, cls_u32);
9669         case TC_CLSU32_NEW_HNODE:
9670         case TC_CLSU32_REPLACE_HNODE:
9671                 return ixgbe_configure_clsu32_add_hnode(adapter, cls_u32);
9672         case TC_CLSU32_DELETE_HNODE:
9673                 return ixgbe_configure_clsu32_del_hnode(adapter, cls_u32);
9674         default:
9675                 return -EOPNOTSUPP;
9676         }
9677 }
9678
9679 static int ixgbe_setup_tc_block_cb(enum tc_setup_type type, void *type_data,
9680                                    void *cb_priv)
9681 {
9682         struct ixgbe_adapter *adapter = cb_priv;
9683
9684         if (!tc_cls_can_offload_and_chain0(adapter->netdev, type_data))
9685                 return -EOPNOTSUPP;
9686
9687         switch (type) {
9688         case TC_SETUP_CLSU32:
9689                 return ixgbe_setup_tc_cls_u32(adapter, type_data);
9690         default:
9691                 return -EOPNOTSUPP;
9692         }
9693 }
9694
9695 static int ixgbe_setup_tc_mqprio(struct net_device *dev,
9696                                  struct tc_mqprio_qopt *mqprio)
9697 {
9698         mqprio->hw = TC_MQPRIO_HW_OFFLOAD_TCS;
9699         return ixgbe_setup_tc(dev, mqprio->num_tc);
9700 }
9701
9702 static LIST_HEAD(ixgbe_block_cb_list);
9703
9704 static int __ixgbe_setup_tc(struct net_device *dev, enum tc_setup_type type,
9705                             void *type_data)
9706 {
9707         struct ixgbe_adapter *adapter = netdev_priv(dev);
9708
9709         switch (type) {
9710         case TC_SETUP_BLOCK:
9711                 return flow_block_cb_setup_simple(type_data,
9712                                                   &ixgbe_block_cb_list,
9713                                                   ixgbe_setup_tc_block_cb,
9714                                                   adapter, adapter, true);
9715         case TC_SETUP_QDISC_MQPRIO:
9716                 return ixgbe_setup_tc_mqprio(dev, type_data);
9717         default:
9718                 return -EOPNOTSUPP;
9719         }
9720 }
9721
9722 #ifdef CONFIG_PCI_IOV
9723 void ixgbe_sriov_reinit(struct ixgbe_adapter *adapter)
9724 {
9725         struct net_device *netdev = adapter->netdev;
9726
9727         rtnl_lock();
9728         ixgbe_setup_tc(netdev, adapter->hw_tcs);
9729         rtnl_unlock();
9730 }
9731
9732 #endif
9733 void ixgbe_do_reset(struct net_device *netdev)
9734 {
9735         struct ixgbe_adapter *adapter = netdev_priv(netdev);
9736
9737         if (netif_running(netdev))
9738                 ixgbe_reinit_locked(adapter);
9739         else
9740                 ixgbe_reset(adapter);
9741 }
9742
9743 static netdev_features_t ixgbe_fix_features(struct net_device *netdev,
9744                                             netdev_features_t features)
9745 {
9746         struct ixgbe_adapter *adapter = netdev_priv(netdev);
9747
9748         /* If Rx checksum is disabled, then RSC/LRO should also be disabled */
9749         if (!(features & NETIF_F_RXCSUM))
9750                 features &= ~NETIF_F_LRO;
9751
9752         /* Turn off LRO if not RSC capable */
9753         if (!(adapter->flags2 & IXGBE_FLAG2_RSC_CAPABLE))
9754                 features &= ~NETIF_F_LRO;
9755
9756         if (adapter->xdp_prog && (features & NETIF_F_LRO)) {
9757                 e_dev_err("LRO is not supported with XDP\n");
9758                 features &= ~NETIF_F_LRO;
9759         }
9760
9761         return features;
9762 }
9763
9764 static void ixgbe_reset_l2fw_offload(struct ixgbe_adapter *adapter)
9765 {
9766         int rss = min_t(int, ixgbe_max_rss_indices(adapter),
9767                         num_online_cpus());
9768
9769         /* go back to full RSS if we're not running SR-IOV */
9770         if (!adapter->ring_feature[RING_F_VMDQ].offset)
9771                 adapter->flags &= ~(IXGBE_FLAG_VMDQ_ENABLED |
9772                                     IXGBE_FLAG_SRIOV_ENABLED);
9773
9774         adapter->ring_feature[RING_F_RSS].limit = rss;
9775         adapter->ring_feature[RING_F_VMDQ].limit = 1;
9776
9777         ixgbe_setup_tc(adapter->netdev, adapter->hw_tcs);
9778 }
9779
9780 static int ixgbe_set_features(struct net_device *netdev,
9781                               netdev_features_t features)
9782 {
9783         struct ixgbe_adapter *adapter = netdev_priv(netdev);
9784         netdev_features_t changed = netdev->features ^ features;
9785         bool need_reset = false;
9786
9787         /* Make sure RSC matches LRO, reset if change */
9788         if (!(features & NETIF_F_LRO)) {
9789                 if (adapter->flags2 & IXGBE_FLAG2_RSC_ENABLED)
9790                         need_reset = true;
9791                 adapter->flags2 &= ~IXGBE_FLAG2_RSC_ENABLED;
9792         } else if ((adapter->flags2 & IXGBE_FLAG2_RSC_CAPABLE) &&
9793                    !(adapter->flags2 & IXGBE_FLAG2_RSC_ENABLED)) {
9794                 if (adapter->rx_itr_setting == 1 ||
9795                     adapter->rx_itr_setting > IXGBE_MIN_RSC_ITR) {
9796                         adapter->flags2 |= IXGBE_FLAG2_RSC_ENABLED;
9797                         need_reset = true;
9798                 } else if ((changed ^ features) & NETIF_F_LRO) {
9799                         e_info(probe, "rx-usecs set too low, "
9800                                "disabling RSC\n");
9801                 }
9802         }
9803
9804         /*
9805          * Check if Flow Director n-tuple support or hw_tc support was
9806          * enabled or disabled.  If the state changed, we need to reset.
9807          */
9808         if ((features & NETIF_F_NTUPLE) || (features & NETIF_F_HW_TC)) {
9809                 /* turn off ATR, enable perfect filters and reset */
9810                 if (!(adapter->flags & IXGBE_FLAG_FDIR_PERFECT_CAPABLE))
9811                         need_reset = true;
9812
9813                 adapter->flags &= ~IXGBE_FLAG_FDIR_HASH_CAPABLE;
9814                 adapter->flags |= IXGBE_FLAG_FDIR_PERFECT_CAPABLE;
9815         } else {
9816                 /* turn off perfect filters, enable ATR and reset */
9817                 if (adapter->flags & IXGBE_FLAG_FDIR_PERFECT_CAPABLE)
9818                         need_reset = true;
9819
9820                 adapter->flags &= ~IXGBE_FLAG_FDIR_PERFECT_CAPABLE;
9821
9822                 /* We cannot enable ATR if SR-IOV is enabled */
9823                 if (adapter->flags & IXGBE_FLAG_SRIOV_ENABLED ||
9824                     /* We cannot enable ATR if we have 2 or more tcs */
9825                     (adapter->hw_tcs > 1) ||
9826                     /* We cannot enable ATR if RSS is disabled */
9827                     (adapter->ring_feature[RING_F_RSS].limit <= 1) ||
9828                     /* A sample rate of 0 indicates ATR disabled */
9829                     (!adapter->atr_sample_rate))
9830                         ; /* do nothing not supported */
9831                 else /* otherwise supported and set the flag */
9832                         adapter->flags |= IXGBE_FLAG_FDIR_HASH_CAPABLE;
9833         }
9834
9835         if (changed & NETIF_F_RXALL)
9836                 need_reset = true;
9837
9838         netdev->features = features;
9839
9840         if ((changed & NETIF_F_HW_L2FW_DOFFLOAD) && adapter->num_rx_pools > 1)
9841                 ixgbe_reset_l2fw_offload(adapter);
9842         else if (need_reset)
9843                 ixgbe_do_reset(netdev);
9844         else if (changed & (NETIF_F_HW_VLAN_CTAG_RX |
9845                             NETIF_F_HW_VLAN_CTAG_FILTER))
9846                 ixgbe_set_rx_mode(netdev);
9847
9848         return 1;
9849 }
9850
9851 static int ixgbe_ndo_fdb_add(struct ndmsg *ndm, struct nlattr *tb[],
9852                              struct net_device *dev,
9853                              const unsigned char *addr, u16 vid,
9854                              u16 flags,
9855                              struct netlink_ext_ack *extack)
9856 {
9857         /* guarantee we can provide a unique filter for the unicast address */
9858         if (is_unicast_ether_addr(addr) || is_link_local_ether_addr(addr)) {
9859                 struct ixgbe_adapter *adapter = netdev_priv(dev);
9860                 u16 pool = VMDQ_P(0);
9861
9862                 if (netdev_uc_count(dev) >= ixgbe_available_rars(adapter, pool))
9863                         return -ENOMEM;
9864         }
9865
9866         return ndo_dflt_fdb_add(ndm, tb, dev, addr, vid, flags);
9867 }
9868
9869 /**
9870  * ixgbe_configure_bridge_mode - set various bridge modes
9871  * @adapter: the private structure
9872  * @mode: requested bridge mode
9873  *
9874  * Configure some settings require for various bridge modes.
9875  **/
9876 static int ixgbe_configure_bridge_mode(struct ixgbe_adapter *adapter,
9877                                        __u16 mode)
9878 {
9879         struct ixgbe_hw *hw = &adapter->hw;
9880         unsigned int p, num_pools;
9881         u32 vmdctl;
9882
9883         switch (mode) {
9884         case BRIDGE_MODE_VEPA:
9885                 /* disable Tx loopback, rely on switch hairpin mode */
9886                 IXGBE_WRITE_REG(&adapter->hw, IXGBE_PFDTXGSWC, 0);
9887
9888                 /* must enable Rx switching replication to allow multicast
9889                  * packet reception on all VFs, and to enable source address
9890                  * pruning.
9891                  */
9892                 vmdctl = IXGBE_READ_REG(hw, IXGBE_VMD_CTL);
9893                 vmdctl |= IXGBE_VT_CTL_REPLEN;
9894                 IXGBE_WRITE_REG(hw, IXGBE_VMD_CTL, vmdctl);
9895
9896                 /* enable Rx source address pruning. Note, this requires
9897                  * replication to be enabled or else it does nothing.
9898                  */
9899                 num_pools = adapter->num_vfs + adapter->num_rx_pools;
9900                 for (p = 0; p < num_pools; p++) {
9901                         if (hw->mac.ops.set_source_address_pruning)
9902                                 hw->mac.ops.set_source_address_pruning(hw,
9903                                                                        true,
9904                                                                        p);
9905                 }
9906                 break;
9907         case BRIDGE_MODE_VEB:
9908                 /* enable Tx loopback for internal VF/PF communication */
9909                 IXGBE_WRITE_REG(&adapter->hw, IXGBE_PFDTXGSWC,
9910                                 IXGBE_PFDTXGSWC_VT_LBEN);
9911
9912                 /* disable Rx switching replication unless we have SR-IOV
9913                  * virtual functions
9914                  */
9915                 vmdctl = IXGBE_READ_REG(hw, IXGBE_VMD_CTL);
9916                 if (!adapter->num_vfs)
9917                         vmdctl &= ~IXGBE_VT_CTL_REPLEN;
9918                 IXGBE_WRITE_REG(hw, IXGBE_VMD_CTL, vmdctl);
9919
9920                 /* disable Rx source address pruning, since we don't expect to
9921                  * be receiving external loopback of our transmitted frames.
9922                  */
9923                 num_pools = adapter->num_vfs + adapter->num_rx_pools;
9924                 for (p = 0; p < num_pools; p++) {
9925                         if (hw->mac.ops.set_source_address_pruning)
9926                                 hw->mac.ops.set_source_address_pruning(hw,
9927                                                                        false,
9928                                                                        p);
9929                 }
9930                 break;
9931         default:
9932                 return -EINVAL;
9933         }
9934
9935         adapter->bridge_mode = mode;
9936
9937         e_info(drv, "enabling bridge mode: %s\n",
9938                mode == BRIDGE_MODE_VEPA ? "VEPA" : "VEB");
9939
9940         return 0;
9941 }
9942
9943 static int ixgbe_ndo_bridge_setlink(struct net_device *dev,
9944                                     struct nlmsghdr *nlh, u16 flags,
9945                                     struct netlink_ext_ack *extack)
9946 {
9947         struct ixgbe_adapter *adapter = netdev_priv(dev);
9948         struct nlattr *attr, *br_spec;
9949         int rem;
9950
9951         if (!(adapter->flags & IXGBE_FLAG_SRIOV_ENABLED))
9952                 return -EOPNOTSUPP;
9953
9954         br_spec = nlmsg_find_attr(nlh, sizeof(struct ifinfomsg), IFLA_AF_SPEC);
9955         if (!br_spec)
9956                 return -EINVAL;
9957
9958         nla_for_each_nested(attr, br_spec, rem) {
9959                 int status;
9960                 __u16 mode;
9961
9962                 if (nla_type(attr) != IFLA_BRIDGE_MODE)
9963                         continue;
9964
9965                 if (nla_len(attr) < sizeof(mode))
9966                         return -EINVAL;
9967
9968                 mode = nla_get_u16(attr);
9969                 status = ixgbe_configure_bridge_mode(adapter, mode);
9970                 if (status)
9971                         return status;
9972
9973                 break;
9974         }
9975
9976         return 0;
9977 }
9978
9979 static int ixgbe_ndo_bridge_getlink(struct sk_buff *skb, u32 pid, u32 seq,
9980                                     struct net_device *dev,
9981                                     u32 filter_mask, int nlflags)
9982 {
9983         struct ixgbe_adapter *adapter = netdev_priv(dev);
9984
9985         if (!(adapter->flags & IXGBE_FLAG_SRIOV_ENABLED))
9986                 return 0;
9987
9988         return ndo_dflt_bridge_getlink(skb, pid, seq, dev,
9989                                        adapter->bridge_mode, 0, 0, nlflags,
9990                                        filter_mask, NULL);
9991 }
9992
9993 static void *ixgbe_fwd_add(struct net_device *pdev, struct net_device *vdev)
9994 {
9995         struct ixgbe_adapter *adapter = netdev_priv(pdev);
9996         struct ixgbe_fwd_adapter *accel;
9997         int tcs = adapter->hw_tcs ? : 1;
9998         int pool, err;
9999
10000         if (adapter->xdp_prog) {
10001                 e_warn(probe, "L2FW offload is not supported with XDP\n");
10002                 return ERR_PTR(-EINVAL);
10003         }
10004
10005         /* The hardware supported by ixgbe only filters on the destination MAC
10006          * address. In order to avoid issues we only support offloading modes
10007          * where the hardware can actually provide the functionality.
10008          */
10009         if (!macvlan_supports_dest_filter(vdev))
10010                 return ERR_PTR(-EMEDIUMTYPE);
10011
10012         /* We need to lock down the macvlan to be a single queue device so that
10013          * we can reuse the tc_to_txq field in the macvlan netdev to represent
10014          * the queue mapping to our netdev.
10015          */
10016         if (netif_is_multiqueue(vdev))
10017                 return ERR_PTR(-ERANGE);
10018
10019         pool = find_first_zero_bit(adapter->fwd_bitmask, adapter->num_rx_pools);
10020         if (pool == adapter->num_rx_pools) {
10021                 u16 used_pools = adapter->num_vfs + adapter->num_rx_pools;
10022                 u16 reserved_pools;
10023
10024                 if (((adapter->flags & IXGBE_FLAG_DCB_ENABLED) &&
10025                      adapter->num_rx_pools >= (MAX_TX_QUEUES / tcs)) ||
10026                     adapter->num_rx_pools > IXGBE_MAX_MACVLANS)
10027                         return ERR_PTR(-EBUSY);
10028
10029                 /* Hardware has a limited number of available pools. Each VF,
10030                  * and the PF require a pool. Check to ensure we don't
10031                  * attempt to use more then the available number of pools.
10032                  */
10033                 if (used_pools >= IXGBE_MAX_VF_FUNCTIONS)
10034                         return ERR_PTR(-EBUSY);
10035
10036                 /* Enable VMDq flag so device will be set in VM mode */
10037                 adapter->flags |= IXGBE_FLAG_VMDQ_ENABLED |
10038                                   IXGBE_FLAG_SRIOV_ENABLED;
10039
10040                 /* Try to reserve as many queues per pool as possible,
10041                  * we start with the configurations that support 4 queues
10042                  * per pools, followed by 2, and then by just 1 per pool.
10043                  */
10044                 if (used_pools < 32 && adapter->num_rx_pools < 16)
10045                         reserved_pools = min_t(u16,
10046                                                32 - used_pools,
10047                                                16 - adapter->num_rx_pools);
10048                 else if (adapter->num_rx_pools < 32)
10049                         reserved_pools = min_t(u16,
10050                                                64 - used_pools,
10051                                                32 - adapter->num_rx_pools);
10052                 else
10053                         reserved_pools = 64 - used_pools;
10054
10055
10056                 if (!reserved_pools)
10057                         return ERR_PTR(-EBUSY);
10058
10059                 adapter->ring_feature[RING_F_VMDQ].limit += reserved_pools;
10060
10061                 /* Force reinit of ring allocation with VMDQ enabled */
10062                 err = ixgbe_setup_tc(pdev, adapter->hw_tcs);
10063                 if (err)
10064                         return ERR_PTR(err);
10065
10066                 if (pool >= adapter->num_rx_pools)
10067                         return ERR_PTR(-ENOMEM);
10068         }
10069
10070         accel = kzalloc(sizeof(*accel), GFP_KERNEL);
10071         if (!accel)
10072                 return ERR_PTR(-ENOMEM);
10073
10074         set_bit(pool, adapter->fwd_bitmask);
10075         netdev_set_sb_channel(vdev, pool);
10076         accel->pool = pool;
10077         accel->netdev = vdev;
10078
10079         if (!netif_running(pdev))
10080                 return accel;
10081
10082         err = ixgbe_fwd_ring_up(adapter, accel);
10083         if (err)
10084                 return ERR_PTR(err);
10085
10086         return accel;
10087 }
10088
10089 static void ixgbe_fwd_del(struct net_device *pdev, void *priv)
10090 {
10091         struct ixgbe_fwd_adapter *accel = priv;
10092         struct ixgbe_adapter *adapter = netdev_priv(pdev);
10093         unsigned int rxbase = accel->rx_base_queue;
10094         unsigned int i;
10095
10096         /* delete unicast filter associated with offloaded interface */
10097         ixgbe_del_mac_filter(adapter, accel->netdev->dev_addr,
10098                              VMDQ_P(accel->pool));
10099
10100         /* Allow remaining Rx packets to get flushed out of the
10101          * Rx FIFO before we drop the netdev for the ring.
10102          */
10103         usleep_range(10000, 20000);
10104
10105         for (i = 0; i < adapter->num_rx_queues_per_pool; i++) {
10106                 struct ixgbe_ring *ring = adapter->rx_ring[rxbase + i];
10107                 struct ixgbe_q_vector *qv = ring->q_vector;
10108
10109                 /* Make sure we aren't processing any packets and clear
10110                  * netdev to shut down the ring.
10111                  */
10112                 if (netif_running(adapter->netdev))
10113                         napi_synchronize(&qv->napi);
10114                 ring->netdev = NULL;
10115         }
10116
10117         /* unbind the queues and drop the subordinate channel config */
10118         netdev_unbind_sb_channel(pdev, accel->netdev);
10119         netdev_set_sb_channel(accel->netdev, 0);
10120
10121         clear_bit(accel->pool, adapter->fwd_bitmask);
10122         kfree(accel);
10123 }
10124
10125 #define IXGBE_MAX_MAC_HDR_LEN           127
10126 #define IXGBE_MAX_NETWORK_HDR_LEN       511
10127
10128 static netdev_features_t
10129 ixgbe_features_check(struct sk_buff *skb, struct net_device *dev,
10130                      netdev_features_t features)
10131 {
10132         unsigned int network_hdr_len, mac_hdr_len;
10133
10134         /* Make certain the headers can be described by a context descriptor */
10135         mac_hdr_len = skb_network_header(skb) - skb->data;
10136         if (unlikely(mac_hdr_len > IXGBE_MAX_MAC_HDR_LEN))
10137                 return features & ~(NETIF_F_HW_CSUM |
10138                                     NETIF_F_SCTP_CRC |
10139                                     NETIF_F_GSO_UDP_L4 |
10140                                     NETIF_F_HW_VLAN_CTAG_TX |
10141                                     NETIF_F_TSO |
10142                                     NETIF_F_TSO6);
10143
10144         network_hdr_len = skb_checksum_start(skb) - skb_network_header(skb);
10145         if (unlikely(network_hdr_len >  IXGBE_MAX_NETWORK_HDR_LEN))
10146                 return features & ~(NETIF_F_HW_CSUM |
10147                                     NETIF_F_SCTP_CRC |
10148                                     NETIF_F_GSO_UDP_L4 |
10149                                     NETIF_F_TSO |
10150                                     NETIF_F_TSO6);
10151
10152         /* We can only support IPV4 TSO in tunnels if we can mangle the
10153          * inner IP ID field, so strip TSO if MANGLEID is not supported.
10154          * IPsec offoad sets skb->encapsulation but still can handle
10155          * the TSO, so it's the exception.
10156          */
10157         if (skb->encapsulation && !(features & NETIF_F_TSO_MANGLEID)) {
10158 #ifdef CONFIG_IXGBE_IPSEC
10159                 if (!secpath_exists(skb))
10160 #endif
10161                         features &= ~NETIF_F_TSO;
10162         }
10163
10164         return features;
10165 }
10166
10167 static int ixgbe_xdp_setup(struct net_device *dev, struct bpf_prog *prog)
10168 {
10169         int i, frame_size = dev->mtu + ETH_HLEN + ETH_FCS_LEN + VLAN_HLEN;
10170         struct ixgbe_adapter *adapter = netdev_priv(dev);
10171         struct bpf_prog *old_prog;
10172         bool need_reset;
10173         int num_queues;
10174
10175         if (adapter->flags & IXGBE_FLAG_SRIOV_ENABLED)
10176                 return -EINVAL;
10177
10178         if (adapter->flags & IXGBE_FLAG_DCB_ENABLED)
10179                 return -EINVAL;
10180
10181         /* verify ixgbe ring attributes are sufficient for XDP */
10182         for (i = 0; i < adapter->num_rx_queues; i++) {
10183                 struct ixgbe_ring *ring = adapter->rx_ring[i];
10184
10185                 if (ring_is_rsc_enabled(ring))
10186                         return -EINVAL;
10187
10188                 if (frame_size > ixgbe_rx_bufsz(ring))
10189                         return -EINVAL;
10190         }
10191
10192         /* if the number of cpus is much larger than the maximum of queues,
10193          * we should stop it and then return with ENOMEM like before.
10194          */
10195         if (nr_cpu_ids > IXGBE_MAX_XDP_QS * 2)
10196                 return -ENOMEM;
10197         else if (nr_cpu_ids > IXGBE_MAX_XDP_QS)
10198                 static_branch_inc(&ixgbe_xdp_locking_key);
10199
10200         old_prog = xchg(&adapter->xdp_prog, prog);
10201         need_reset = (!!prog != !!old_prog);
10202
10203         /* If transitioning XDP modes reconfigure rings */
10204         if (need_reset) {
10205                 int err;
10206
10207                 if (!prog)
10208                         /* Wait until ndo_xsk_wakeup completes. */
10209                         synchronize_rcu();
10210                 err = ixgbe_setup_tc(dev, adapter->hw_tcs);
10211
10212                 if (err) {
10213                         rcu_assign_pointer(adapter->xdp_prog, old_prog);
10214                         return -EINVAL;
10215                 }
10216         } else {
10217                 for (i = 0; i < adapter->num_rx_queues; i++)
10218                         (void)xchg(&adapter->rx_ring[i]->xdp_prog,
10219                             adapter->xdp_prog);
10220         }
10221
10222         if (old_prog)
10223                 bpf_prog_put(old_prog);
10224
10225         /* Kick start the NAPI context if there is an AF_XDP socket open
10226          * on that queue id. This so that receiving will start.
10227          */
10228         if (need_reset && prog) {
10229                 num_queues = min_t(int, adapter->num_rx_queues,
10230                                    adapter->num_xdp_queues);
10231                 for (i = 0; i < num_queues; i++)
10232                         if (adapter->xdp_ring[i]->xsk_pool)
10233                                 (void)ixgbe_xsk_wakeup(adapter->netdev, i,
10234                                                        XDP_WAKEUP_RX);
10235         }
10236
10237         return 0;
10238 }
10239
10240 static int ixgbe_xdp(struct net_device *dev, struct netdev_bpf *xdp)
10241 {
10242         struct ixgbe_adapter *adapter = netdev_priv(dev);
10243
10244         switch (xdp->command) {
10245         case XDP_SETUP_PROG:
10246                 return ixgbe_xdp_setup(dev, xdp->prog);
10247         case XDP_SETUP_XSK_POOL:
10248                 return ixgbe_xsk_pool_setup(adapter, xdp->xsk.pool,
10249                                             xdp->xsk.queue_id);
10250
10251         default:
10252                 return -EINVAL;
10253         }
10254 }
10255
10256 void ixgbe_xdp_ring_update_tail(struct ixgbe_ring *ring)
10257 {
10258         /* Force memory writes to complete before letting h/w know there
10259          * are new descriptors to fetch.
10260          */
10261         wmb();
10262         writel(ring->next_to_use, ring->tail);
10263 }
10264
10265 void ixgbe_xdp_ring_update_tail_locked(struct ixgbe_ring *ring)
10266 {
10267         if (static_branch_unlikely(&ixgbe_xdp_locking_key))
10268                 spin_lock(&ring->tx_lock);
10269         ixgbe_xdp_ring_update_tail(ring);
10270         if (static_branch_unlikely(&ixgbe_xdp_locking_key))
10271                 spin_unlock(&ring->tx_lock);
10272 }
10273
10274 static int ixgbe_xdp_xmit(struct net_device *dev, int n,
10275                           struct xdp_frame **frames, u32 flags)
10276 {
10277         struct ixgbe_adapter *adapter = netdev_priv(dev);
10278         struct ixgbe_ring *ring;
10279         int nxmit = 0;
10280         int i;
10281
10282         if (unlikely(test_bit(__IXGBE_DOWN, &adapter->state)))
10283                 return -ENETDOWN;
10284
10285         if (unlikely(flags & ~XDP_XMIT_FLAGS_MASK))
10286                 return -EINVAL;
10287
10288         /* During program transitions its possible adapter->xdp_prog is assigned
10289          * but ring has not been configured yet. In this case simply abort xmit.
10290          */
10291         ring = adapter->xdp_prog ? ixgbe_determine_xdp_ring(adapter) : NULL;
10292         if (unlikely(!ring))
10293                 return -ENXIO;
10294
10295         if (unlikely(test_bit(__IXGBE_TX_DISABLED, &ring->state)))
10296                 return -ENXIO;
10297
10298         if (static_branch_unlikely(&ixgbe_xdp_locking_key))
10299                 spin_lock(&ring->tx_lock);
10300
10301         for (i = 0; i < n; i++) {
10302                 struct xdp_frame *xdpf = frames[i];
10303                 int err;
10304
10305                 err = ixgbe_xmit_xdp_ring(ring, xdpf);
10306                 if (err != IXGBE_XDP_TX)
10307                         break;
10308                 nxmit++;
10309         }
10310
10311         if (unlikely(flags & XDP_XMIT_FLUSH))
10312                 ixgbe_xdp_ring_update_tail(ring);
10313
10314         if (static_branch_unlikely(&ixgbe_xdp_locking_key))
10315                 spin_unlock(&ring->tx_lock);
10316
10317         return nxmit;
10318 }
10319
10320 static const struct net_device_ops ixgbe_netdev_ops = {
10321         .ndo_open               = ixgbe_open,
10322         .ndo_stop               = ixgbe_close,
10323         .ndo_start_xmit         = ixgbe_xmit_frame,
10324         .ndo_set_rx_mode        = ixgbe_set_rx_mode,
10325         .ndo_validate_addr      = eth_validate_addr,
10326         .ndo_set_mac_address    = ixgbe_set_mac,
10327         .ndo_change_mtu         = ixgbe_change_mtu,
10328         .ndo_tx_timeout         = ixgbe_tx_timeout,
10329         .ndo_set_tx_maxrate     = ixgbe_tx_maxrate,
10330         .ndo_vlan_rx_add_vid    = ixgbe_vlan_rx_add_vid,
10331         .ndo_vlan_rx_kill_vid   = ixgbe_vlan_rx_kill_vid,
10332         .ndo_eth_ioctl          = ixgbe_ioctl,
10333         .ndo_set_vf_mac         = ixgbe_ndo_set_vf_mac,
10334         .ndo_set_vf_vlan        = ixgbe_ndo_set_vf_vlan,
10335         .ndo_set_vf_rate        = ixgbe_ndo_set_vf_bw,
10336         .ndo_set_vf_spoofchk    = ixgbe_ndo_set_vf_spoofchk,
10337         .ndo_set_vf_link_state  = ixgbe_ndo_set_vf_link_state,
10338         .ndo_set_vf_rss_query_en = ixgbe_ndo_set_vf_rss_query_en,
10339         .ndo_set_vf_trust       = ixgbe_ndo_set_vf_trust,
10340         .ndo_get_vf_config      = ixgbe_ndo_get_vf_config,
10341         .ndo_get_stats64        = ixgbe_get_stats64,
10342         .ndo_setup_tc           = __ixgbe_setup_tc,
10343 #ifdef IXGBE_FCOE
10344         .ndo_select_queue       = ixgbe_select_queue,
10345         .ndo_fcoe_ddp_setup = ixgbe_fcoe_ddp_get,
10346         .ndo_fcoe_ddp_target = ixgbe_fcoe_ddp_target,
10347         .ndo_fcoe_ddp_done = ixgbe_fcoe_ddp_put,
10348         .ndo_fcoe_enable = ixgbe_fcoe_enable,
10349         .ndo_fcoe_disable = ixgbe_fcoe_disable,
10350         .ndo_fcoe_get_wwn = ixgbe_fcoe_get_wwn,
10351         .ndo_fcoe_get_hbainfo = ixgbe_fcoe_get_hbainfo,
10352 #endif /* IXGBE_FCOE */
10353         .ndo_set_features = ixgbe_set_features,
10354         .ndo_fix_features = ixgbe_fix_features,
10355         .ndo_fdb_add            = ixgbe_ndo_fdb_add,
10356         .ndo_bridge_setlink     = ixgbe_ndo_bridge_setlink,
10357         .ndo_bridge_getlink     = ixgbe_ndo_bridge_getlink,
10358         .ndo_dfwd_add_station   = ixgbe_fwd_add,
10359         .ndo_dfwd_del_station   = ixgbe_fwd_del,
10360         .ndo_features_check     = ixgbe_features_check,
10361         .ndo_bpf                = ixgbe_xdp,
10362         .ndo_xdp_xmit           = ixgbe_xdp_xmit,
10363         .ndo_xsk_wakeup         = ixgbe_xsk_wakeup,
10364 };
10365
10366 static void ixgbe_disable_txr_hw(struct ixgbe_adapter *adapter,
10367                                  struct ixgbe_ring *tx_ring)
10368 {
10369         unsigned long wait_delay, delay_interval;
10370         struct ixgbe_hw *hw = &adapter->hw;
10371         u8 reg_idx = tx_ring->reg_idx;
10372         int wait_loop;
10373         u32 txdctl;
10374
10375         IXGBE_WRITE_REG(hw, IXGBE_TXDCTL(reg_idx), IXGBE_TXDCTL_SWFLSH);
10376
10377         /* delay mechanism from ixgbe_disable_tx */
10378         delay_interval = ixgbe_get_completion_timeout(adapter) / 100;
10379
10380         wait_loop = IXGBE_MAX_RX_DESC_POLL;
10381         wait_delay = delay_interval;
10382
10383         while (wait_loop--) {
10384                 usleep_range(wait_delay, wait_delay + 10);
10385                 wait_delay += delay_interval * 2;
10386                 txdctl = IXGBE_READ_REG(hw, IXGBE_TXDCTL(reg_idx));
10387
10388                 if (!(txdctl & IXGBE_TXDCTL_ENABLE))
10389                         return;
10390         }
10391
10392         e_err(drv, "TXDCTL.ENABLE not cleared within the polling period\n");
10393 }
10394
10395 static void ixgbe_disable_txr(struct ixgbe_adapter *adapter,
10396                               struct ixgbe_ring *tx_ring)
10397 {
10398         set_bit(__IXGBE_TX_DISABLED, &tx_ring->state);
10399         ixgbe_disable_txr_hw(adapter, tx_ring);
10400 }
10401
10402 static void ixgbe_disable_rxr_hw(struct ixgbe_adapter *adapter,
10403                                  struct ixgbe_ring *rx_ring)
10404 {
10405         unsigned long wait_delay, delay_interval;
10406         struct ixgbe_hw *hw = &adapter->hw;
10407         u8 reg_idx = rx_ring->reg_idx;
10408         int wait_loop;
10409         u32 rxdctl;
10410
10411         rxdctl = IXGBE_READ_REG(hw, IXGBE_RXDCTL(reg_idx));
10412         rxdctl &= ~IXGBE_RXDCTL_ENABLE;
10413         rxdctl |= IXGBE_RXDCTL_SWFLSH;
10414
10415         /* write value back with RXDCTL.ENABLE bit cleared */
10416         IXGBE_WRITE_REG(hw, IXGBE_RXDCTL(reg_idx), rxdctl);
10417
10418         /* RXDCTL.EN may not change on 82598 if link is down, so skip it */
10419         if (hw->mac.type == ixgbe_mac_82598EB &&
10420             !(IXGBE_READ_REG(hw, IXGBE_LINKS) & IXGBE_LINKS_UP))
10421                 return;
10422
10423         /* delay mechanism from ixgbe_disable_rx */
10424         delay_interval = ixgbe_get_completion_timeout(adapter) / 100;
10425
10426         wait_loop = IXGBE_MAX_RX_DESC_POLL;
10427         wait_delay = delay_interval;
10428
10429         while (wait_loop--) {
10430                 usleep_range(wait_delay, wait_delay + 10);
10431                 wait_delay += delay_interval * 2;
10432                 rxdctl = IXGBE_READ_REG(hw, IXGBE_RXDCTL(reg_idx));
10433
10434                 if (!(rxdctl & IXGBE_RXDCTL_ENABLE))
10435                         return;
10436         }
10437
10438         e_err(drv, "RXDCTL.ENABLE not cleared within the polling period\n");
10439 }
10440
10441 static void ixgbe_reset_txr_stats(struct ixgbe_ring *tx_ring)
10442 {
10443         memset(&tx_ring->stats, 0, sizeof(tx_ring->stats));
10444         memset(&tx_ring->tx_stats, 0, sizeof(tx_ring->tx_stats));
10445 }
10446
10447 static void ixgbe_reset_rxr_stats(struct ixgbe_ring *rx_ring)
10448 {
10449         memset(&rx_ring->stats, 0, sizeof(rx_ring->stats));
10450         memset(&rx_ring->rx_stats, 0, sizeof(rx_ring->rx_stats));
10451 }
10452
10453 /**
10454  * ixgbe_txrx_ring_disable - Disable Rx/Tx/XDP Tx rings
10455  * @adapter: adapter structure
10456  * @ring: ring index
10457  *
10458  * This function disables a certain Rx/Tx/XDP Tx ring. The function
10459  * assumes that the netdev is running.
10460  **/
10461 void ixgbe_txrx_ring_disable(struct ixgbe_adapter *adapter, int ring)
10462 {
10463         struct ixgbe_ring *rx_ring, *tx_ring, *xdp_ring;
10464
10465         rx_ring = adapter->rx_ring[ring];
10466         tx_ring = adapter->tx_ring[ring];
10467         xdp_ring = adapter->xdp_ring[ring];
10468
10469         ixgbe_disable_txr(adapter, tx_ring);
10470         if (xdp_ring)
10471                 ixgbe_disable_txr(adapter, xdp_ring);
10472         ixgbe_disable_rxr_hw(adapter, rx_ring);
10473
10474         if (xdp_ring)
10475                 synchronize_rcu();
10476
10477         /* Rx/Tx/XDP Tx share the same napi context. */
10478         napi_disable(&rx_ring->q_vector->napi);
10479
10480         ixgbe_clean_tx_ring(tx_ring);
10481         if (xdp_ring)
10482                 ixgbe_clean_tx_ring(xdp_ring);
10483         ixgbe_clean_rx_ring(rx_ring);
10484
10485         ixgbe_reset_txr_stats(tx_ring);
10486         if (xdp_ring)
10487                 ixgbe_reset_txr_stats(xdp_ring);
10488         ixgbe_reset_rxr_stats(rx_ring);
10489 }
10490
10491 /**
10492  * ixgbe_txrx_ring_enable - Enable Rx/Tx/XDP Tx rings
10493  * @adapter: adapter structure
10494  * @ring: ring index
10495  *
10496  * This function enables a certain Rx/Tx/XDP Tx ring. The function
10497  * assumes that the netdev is running.
10498  **/
10499 void ixgbe_txrx_ring_enable(struct ixgbe_adapter *adapter, int ring)
10500 {
10501         struct ixgbe_ring *rx_ring, *tx_ring, *xdp_ring;
10502
10503         rx_ring = adapter->rx_ring[ring];
10504         tx_ring = adapter->tx_ring[ring];
10505         xdp_ring = adapter->xdp_ring[ring];
10506
10507         /* Rx/Tx/XDP Tx share the same napi context. */
10508         napi_enable(&rx_ring->q_vector->napi);
10509
10510         ixgbe_configure_tx_ring(adapter, tx_ring);
10511         if (xdp_ring)
10512                 ixgbe_configure_tx_ring(adapter, xdp_ring);
10513         ixgbe_configure_rx_ring(adapter, rx_ring);
10514
10515         clear_bit(__IXGBE_TX_DISABLED, &tx_ring->state);
10516         if (xdp_ring)
10517                 clear_bit(__IXGBE_TX_DISABLED, &xdp_ring->state);
10518 }
10519
10520 /**
10521  * ixgbe_enumerate_functions - Get the number of ports this device has
10522  * @adapter: adapter structure
10523  *
10524  * This function enumerates the phsyical functions co-located on a single slot,
10525  * in order to determine how many ports a device has. This is most useful in
10526  * determining the required GT/s of PCIe bandwidth necessary for optimal
10527  * performance.
10528  **/
10529 static inline int ixgbe_enumerate_functions(struct ixgbe_adapter *adapter)
10530 {
10531         struct pci_dev *entry, *pdev = adapter->pdev;
10532         int physfns = 0;
10533
10534         /* Some cards can not use the generic count PCIe functions method,
10535          * because they are behind a parent switch, so we hardcode these with
10536          * the correct number of functions.
10537          */
10538         if (ixgbe_pcie_from_parent(&adapter->hw))
10539                 physfns = 4;
10540
10541         list_for_each_entry(entry, &adapter->pdev->bus->devices, bus_list) {
10542                 /* don't count virtual functions */
10543                 if (entry->is_virtfn)
10544                         continue;
10545
10546                 /* When the devices on the bus don't all match our device ID,
10547                  * we can't reliably determine the correct number of
10548                  * functions. This can occur if a function has been direct
10549                  * attached to a virtual machine using VT-d, for example. In
10550                  * this case, simply return -1 to indicate this.
10551                  */
10552                 if ((entry->vendor != pdev->vendor) ||
10553                     (entry->device != pdev->device))
10554                         return -1;
10555
10556                 physfns++;
10557         }
10558
10559         return physfns;
10560 }
10561
10562 /**
10563  * ixgbe_wol_supported - Check whether device supports WoL
10564  * @adapter: the adapter private structure
10565  * @device_id: the device ID
10566  * @subdevice_id: the subsystem device ID
10567  *
10568  * This function is used by probe and ethtool to determine
10569  * which devices have WoL support
10570  *
10571  **/
10572 bool ixgbe_wol_supported(struct ixgbe_adapter *adapter, u16 device_id,
10573                          u16 subdevice_id)
10574 {
10575         struct ixgbe_hw *hw = &adapter->hw;
10576         u16 wol_cap = adapter->eeprom_cap & IXGBE_DEVICE_CAPS_WOL_MASK;
10577
10578         /* WOL not supported on 82598 */
10579         if (hw->mac.type == ixgbe_mac_82598EB)
10580                 return false;
10581
10582         /* check eeprom to see if WOL is enabled for X540 and newer */
10583         if (hw->mac.type >= ixgbe_mac_X540) {
10584                 if ((wol_cap == IXGBE_DEVICE_CAPS_WOL_PORT0_1) ||
10585                     ((wol_cap == IXGBE_DEVICE_CAPS_WOL_PORT0) &&
10586                      (hw->bus.func == 0)))
10587                         return true;
10588         }
10589
10590         /* WOL is determined based on device IDs for 82599 MACs */
10591         switch (device_id) {
10592         case IXGBE_DEV_ID_82599_SFP:
10593                 /* Only these subdevices could supports WOL */
10594                 switch (subdevice_id) {
10595                 case IXGBE_SUBDEV_ID_82599_560FLR:
10596                 case IXGBE_SUBDEV_ID_82599_LOM_SNAP6:
10597                 case IXGBE_SUBDEV_ID_82599_SFP_WOL0:
10598                 case IXGBE_SUBDEV_ID_82599_SFP_2OCP:
10599                         /* only support first port */
10600                         if (hw->bus.func != 0)
10601                                 break;
10602                         fallthrough;
10603                 case IXGBE_SUBDEV_ID_82599_SP_560FLR:
10604                 case IXGBE_SUBDEV_ID_82599_SFP:
10605                 case IXGBE_SUBDEV_ID_82599_RNDC:
10606                 case IXGBE_SUBDEV_ID_82599_ECNA_DP:
10607                 case IXGBE_SUBDEV_ID_82599_SFP_1OCP:
10608                 case IXGBE_SUBDEV_ID_82599_SFP_LOM_OEM1:
10609                 case IXGBE_SUBDEV_ID_82599_SFP_LOM_OEM2:
10610                         return true;
10611                 }
10612                 break;
10613         case IXGBE_DEV_ID_82599EN_SFP:
10614                 /* Only these subdevices support WOL */
10615                 switch (subdevice_id) {
10616                 case IXGBE_SUBDEV_ID_82599EN_SFP_OCP1:
10617                         return true;
10618                 }
10619                 break;
10620         case IXGBE_DEV_ID_82599_COMBO_BACKPLANE:
10621                 /* All except this subdevice support WOL */
10622                 if (subdevice_id != IXGBE_SUBDEV_ID_82599_KX4_KR_MEZZ)
10623                         return true;
10624                 break;
10625         case IXGBE_DEV_ID_82599_KX4:
10626                 return  true;
10627         default:
10628                 break;
10629         }
10630
10631         return false;
10632 }
10633
10634 /**
10635  * ixgbe_set_fw_version - Set FW version
10636  * @adapter: the adapter private structure
10637  *
10638  * This function is used by probe and ethtool to determine the FW version to
10639  * format to display. The FW version is taken from the EEPROM/NVM.
10640  */
10641 static void ixgbe_set_fw_version(struct ixgbe_adapter *adapter)
10642 {
10643         struct ixgbe_hw *hw = &adapter->hw;
10644         struct ixgbe_nvm_version nvm_ver;
10645
10646         ixgbe_get_oem_prod_version(hw, &nvm_ver);
10647         if (nvm_ver.oem_valid) {
10648                 snprintf(adapter->eeprom_id, sizeof(adapter->eeprom_id),
10649                          "%x.%x.%x", nvm_ver.oem_major, nvm_ver.oem_minor,
10650                          nvm_ver.oem_release);
10651                 return;
10652         }
10653
10654         ixgbe_get_etk_id(hw, &nvm_ver);
10655         ixgbe_get_orom_version(hw, &nvm_ver);
10656
10657         if (nvm_ver.or_valid) {
10658                 snprintf(adapter->eeprom_id, sizeof(adapter->eeprom_id),
10659                          "0x%08x, %d.%d.%d", nvm_ver.etk_id, nvm_ver.or_major,
10660                          nvm_ver.or_build, nvm_ver.or_patch);
10661                 return;
10662         }
10663
10664         /* Set ETrack ID format */
10665         snprintf(adapter->eeprom_id, sizeof(adapter->eeprom_id),
10666                  "0x%08x", nvm_ver.etk_id);
10667 }
10668
10669 /**
10670  * ixgbe_probe - Device Initialization Routine
10671  * @pdev: PCI device information struct
10672  * @ent: entry in ixgbe_pci_tbl
10673  *
10674  * Returns 0 on success, negative on failure
10675  *
10676  * ixgbe_probe initializes an adapter identified by a pci_dev structure.
10677  * The OS initialization, configuring of the adapter private structure,
10678  * and a hardware reset occur.
10679  **/
10680 static int ixgbe_probe(struct pci_dev *pdev, const struct pci_device_id *ent)
10681 {
10682         struct net_device *netdev;
10683         struct ixgbe_adapter *adapter = NULL;
10684         struct ixgbe_hw *hw;
10685         const struct ixgbe_info *ii = ixgbe_info_tbl[ent->driver_data];
10686         unsigned int indices = MAX_TX_QUEUES;
10687         u8 part_str[IXGBE_PBANUM_LENGTH];
10688         int i, err, expected_gts;
10689         bool disable_dev = false;
10690 #ifdef IXGBE_FCOE
10691         u16 device_caps;
10692 #endif
10693         u32 eec;
10694
10695         /* Catch broken hardware that put the wrong VF device ID in
10696          * the PCIe SR-IOV capability.
10697          */
10698         if (pdev->is_virtfn) {
10699                 WARN(1, KERN_ERR "%s (%hx:%hx) should not be a VF!\n",
10700                      pci_name(pdev), pdev->vendor, pdev->device);
10701                 return -EINVAL;
10702         }
10703
10704         err = pci_enable_device_mem(pdev);
10705         if (err)
10706                 return err;
10707
10708         err = dma_set_mask_and_coherent(&pdev->dev, DMA_BIT_MASK(64));
10709         if (err) {
10710                 dev_err(&pdev->dev,
10711                         "No usable DMA configuration, aborting\n");
10712                 goto err_dma;
10713         }
10714
10715         err = pci_request_mem_regions(pdev, ixgbe_driver_name);
10716         if (err) {
10717                 dev_err(&pdev->dev,
10718                         "pci_request_selected_regions failed 0x%x\n", err);
10719                 goto err_pci_reg;
10720         }
10721
10722         pci_enable_pcie_error_reporting(pdev);
10723
10724         pci_set_master(pdev);
10725         pci_save_state(pdev);
10726
10727         if (ii->mac == ixgbe_mac_82598EB) {
10728 #ifdef CONFIG_IXGBE_DCB
10729                 /* 8 TC w/ 4 queues per TC */
10730                 indices = 4 * MAX_TRAFFIC_CLASS;
10731 #else
10732                 indices = IXGBE_MAX_RSS_INDICES;
10733 #endif
10734         }
10735
10736         netdev = alloc_etherdev_mq(sizeof(struct ixgbe_adapter), indices);
10737         if (!netdev) {
10738                 err = -ENOMEM;
10739                 goto err_alloc_etherdev;
10740         }
10741
10742         SET_NETDEV_DEV(netdev, &pdev->dev);
10743
10744         adapter = netdev_priv(netdev);
10745
10746         adapter->netdev = netdev;
10747         adapter->pdev = pdev;
10748         hw = &adapter->hw;
10749         hw->back = adapter;
10750         adapter->msg_enable = netif_msg_init(debug, DEFAULT_MSG_ENABLE);
10751
10752         hw->hw_addr = ioremap(pci_resource_start(pdev, 0),
10753                               pci_resource_len(pdev, 0));
10754         adapter->io_addr = hw->hw_addr;
10755         if (!hw->hw_addr) {
10756                 err = -EIO;
10757                 goto err_ioremap;
10758         }
10759
10760         netdev->netdev_ops = &ixgbe_netdev_ops;
10761         ixgbe_set_ethtool_ops(netdev);
10762         netdev->watchdog_timeo = 5 * HZ;
10763         strlcpy(netdev->name, pci_name(pdev), sizeof(netdev->name));
10764
10765         /* Setup hw api */
10766         hw->mac.ops   = *ii->mac_ops;
10767         hw->mac.type  = ii->mac;
10768         hw->mvals     = ii->mvals;
10769         if (ii->link_ops)
10770                 hw->link.ops  = *ii->link_ops;
10771
10772         /* EEPROM */
10773         hw->eeprom.ops = *ii->eeprom_ops;
10774         eec = IXGBE_READ_REG(hw, IXGBE_EEC(hw));
10775         if (ixgbe_removed(hw->hw_addr)) {
10776                 err = -EIO;
10777                 goto err_ioremap;
10778         }
10779         /* If EEPROM is valid (bit 8 = 1), use default otherwise use bit bang */
10780         if (!(eec & BIT(8)))
10781                 hw->eeprom.ops.read = &ixgbe_read_eeprom_bit_bang_generic;
10782
10783         /* PHY */
10784         hw->phy.ops = *ii->phy_ops;
10785         hw->phy.sfp_type = ixgbe_sfp_type_unknown;
10786         /* ixgbe_identify_phy_generic will set prtad and mmds properly */
10787         hw->phy.mdio.prtad = MDIO_PRTAD_NONE;
10788         hw->phy.mdio.mmds = 0;
10789         hw->phy.mdio.mode_support = MDIO_SUPPORTS_C45 | MDIO_EMULATE_C22;
10790         hw->phy.mdio.dev = netdev;
10791         hw->phy.mdio.mdio_read = ixgbe_mdio_read;
10792         hw->phy.mdio.mdio_write = ixgbe_mdio_write;
10793
10794         /* setup the private structure */
10795         err = ixgbe_sw_init(adapter, ii);
10796         if (err)
10797                 goto err_sw_init;
10798
10799         if (adapter->hw.mac.type == ixgbe_mac_82599EB)
10800                 adapter->flags2 |= IXGBE_FLAG2_AUTO_DISABLE_VF;
10801
10802         switch (adapter->hw.mac.type) {
10803         case ixgbe_mac_X550:
10804         case ixgbe_mac_X550EM_x:
10805                 netdev->udp_tunnel_nic_info = &ixgbe_udp_tunnels_x550;
10806                 break;
10807         case ixgbe_mac_x550em_a:
10808                 netdev->udp_tunnel_nic_info = &ixgbe_udp_tunnels_x550em_a;
10809                 break;
10810         default:
10811                 break;
10812         }
10813
10814         /* Make sure the SWFW semaphore is in a valid state */
10815         if (hw->mac.ops.init_swfw_sync)
10816                 hw->mac.ops.init_swfw_sync(hw);
10817
10818         /* Make it possible the adapter to be woken up via WOL */
10819         switch (adapter->hw.mac.type) {
10820         case ixgbe_mac_82599EB:
10821         case ixgbe_mac_X540:
10822         case ixgbe_mac_X550:
10823         case ixgbe_mac_X550EM_x:
10824         case ixgbe_mac_x550em_a:
10825                 IXGBE_WRITE_REG(&adapter->hw, IXGBE_WUS, ~0);
10826                 break;
10827         default:
10828                 break;
10829         }
10830
10831         /*
10832          * If there is a fan on this device and it has failed log the
10833          * failure.
10834          */
10835         if (adapter->flags & IXGBE_FLAG_FAN_FAIL_CAPABLE) {
10836                 u32 esdp = IXGBE_READ_REG(hw, IXGBE_ESDP);
10837                 if (esdp & IXGBE_ESDP_SDP1)
10838                         e_crit(probe, "Fan has stopped, replace the adapter\n");
10839         }
10840
10841         if (allow_unsupported_sfp)
10842                 hw->allow_unsupported_sfp = allow_unsupported_sfp;
10843
10844         /* reset_hw fills in the perm_addr as well */
10845         hw->phy.reset_if_overtemp = true;
10846         err = hw->mac.ops.reset_hw(hw);
10847         hw->phy.reset_if_overtemp = false;
10848         ixgbe_set_eee_capable(adapter);
10849         if (err == IXGBE_ERR_SFP_NOT_PRESENT) {
10850                 err = 0;
10851         } else if (err == IXGBE_ERR_SFP_NOT_SUPPORTED) {
10852                 e_dev_err("failed to load because an unsupported SFP+ or QSFP module type was detected.\n");
10853                 e_dev_err("Reload the driver after installing a supported module.\n");
10854                 goto err_sw_init;
10855         } else if (err) {
10856                 e_dev_err("HW Init failed: %d\n", err);
10857                 goto err_sw_init;
10858         }
10859
10860 #ifdef CONFIG_PCI_IOV
10861         /* SR-IOV not supported on the 82598 */
10862         if (adapter->hw.mac.type == ixgbe_mac_82598EB)
10863                 goto skip_sriov;
10864         /* Mailbox */
10865         ixgbe_init_mbx_params_pf(hw);
10866         hw->mbx.ops = ii->mbx_ops;
10867         pci_sriov_set_totalvfs(pdev, IXGBE_MAX_VFS_DRV_LIMIT);
10868         ixgbe_enable_sriov(adapter, max_vfs);
10869 skip_sriov:
10870
10871 #endif
10872         netdev->features = NETIF_F_SG |
10873                            NETIF_F_TSO |
10874                            NETIF_F_TSO6 |
10875                            NETIF_F_RXHASH |
10876                            NETIF_F_RXCSUM |
10877                            NETIF_F_HW_CSUM;
10878
10879 #define IXGBE_GSO_PARTIAL_FEATURES (NETIF_F_GSO_GRE | \
10880                                     NETIF_F_GSO_GRE_CSUM | \
10881                                     NETIF_F_GSO_IPXIP4 | \
10882                                     NETIF_F_GSO_IPXIP6 | \
10883                                     NETIF_F_GSO_UDP_TUNNEL | \
10884                                     NETIF_F_GSO_UDP_TUNNEL_CSUM)
10885
10886         netdev->gso_partial_features = IXGBE_GSO_PARTIAL_FEATURES;
10887         netdev->features |= NETIF_F_GSO_PARTIAL |
10888                             IXGBE_GSO_PARTIAL_FEATURES;
10889
10890         if (hw->mac.type >= ixgbe_mac_82599EB)
10891                 netdev->features |= NETIF_F_SCTP_CRC | NETIF_F_GSO_UDP_L4;
10892
10893 #ifdef CONFIG_IXGBE_IPSEC
10894 #define IXGBE_ESP_FEATURES      (NETIF_F_HW_ESP | \
10895                                  NETIF_F_HW_ESP_TX_CSUM | \
10896                                  NETIF_F_GSO_ESP)
10897
10898         if (adapter->ipsec)
10899                 netdev->features |= IXGBE_ESP_FEATURES;
10900 #endif
10901         /* copy netdev features into list of user selectable features */
10902         netdev->hw_features |= netdev->features |
10903                                NETIF_F_HW_VLAN_CTAG_FILTER |
10904                                NETIF_F_HW_VLAN_CTAG_RX |
10905                                NETIF_F_HW_VLAN_CTAG_TX |
10906                                NETIF_F_RXALL |
10907                                NETIF_F_HW_L2FW_DOFFLOAD;
10908
10909         if (hw->mac.type >= ixgbe_mac_82599EB)
10910                 netdev->hw_features |= NETIF_F_NTUPLE |
10911                                        NETIF_F_HW_TC;
10912
10913         netdev->features |= NETIF_F_HIGHDMA;
10914
10915         netdev->vlan_features |= netdev->features | NETIF_F_TSO_MANGLEID;
10916         netdev->hw_enc_features |= netdev->vlan_features;
10917         netdev->mpls_features |= NETIF_F_SG |
10918                                  NETIF_F_TSO |
10919                                  NETIF_F_TSO6 |
10920                                  NETIF_F_HW_CSUM;
10921         netdev->mpls_features |= IXGBE_GSO_PARTIAL_FEATURES;
10922
10923         /* set this bit last since it cannot be part of vlan_features */
10924         netdev->features |= NETIF_F_HW_VLAN_CTAG_FILTER |
10925                             NETIF_F_HW_VLAN_CTAG_RX |
10926                             NETIF_F_HW_VLAN_CTAG_TX;
10927
10928         netdev->priv_flags |= IFF_UNICAST_FLT;
10929         netdev->priv_flags |= IFF_SUPP_NOFCS;
10930
10931         /* MTU range: 68 - 9710 */
10932         netdev->min_mtu = ETH_MIN_MTU;
10933         netdev->max_mtu = IXGBE_MAX_JUMBO_FRAME_SIZE - (ETH_HLEN + ETH_FCS_LEN);
10934
10935 #ifdef CONFIG_IXGBE_DCB
10936         if (adapter->flags & IXGBE_FLAG_DCB_CAPABLE)
10937                 netdev->dcbnl_ops = &ixgbe_dcbnl_ops;
10938 #endif
10939
10940 #ifdef IXGBE_FCOE
10941         if (adapter->flags & IXGBE_FLAG_FCOE_CAPABLE) {
10942                 unsigned int fcoe_l;
10943
10944                 if (hw->mac.ops.get_device_caps) {
10945                         hw->mac.ops.get_device_caps(hw, &device_caps);
10946                         if (device_caps & IXGBE_DEVICE_CAPS_FCOE_OFFLOADS)
10947                                 adapter->flags &= ~IXGBE_FLAG_FCOE_CAPABLE;
10948                 }
10949
10950
10951                 fcoe_l = min_t(int, IXGBE_FCRETA_SIZE, num_online_cpus());
10952                 adapter->ring_feature[RING_F_FCOE].limit = fcoe_l;
10953
10954                 netdev->features |= NETIF_F_FSO |
10955                                     NETIF_F_FCOE_CRC;
10956
10957                 netdev->vlan_features |= NETIF_F_FSO |
10958                                          NETIF_F_FCOE_CRC |
10959                                          NETIF_F_FCOE_MTU;
10960         }
10961 #endif /* IXGBE_FCOE */
10962         if (adapter->flags2 & IXGBE_FLAG2_RSC_CAPABLE)
10963                 netdev->hw_features |= NETIF_F_LRO;
10964         if (adapter->flags2 & IXGBE_FLAG2_RSC_ENABLED)
10965                 netdev->features |= NETIF_F_LRO;
10966
10967         if (ixgbe_check_fw_error(adapter)) {
10968                 err = -EIO;
10969                 goto err_sw_init;
10970         }
10971
10972         /* make sure the EEPROM is good */
10973         if (hw->eeprom.ops.validate_checksum(hw, NULL) < 0) {
10974                 e_dev_err("The EEPROM Checksum Is Not Valid\n");
10975                 err = -EIO;
10976                 goto err_sw_init;
10977         }
10978
10979         eth_platform_get_mac_address(&adapter->pdev->dev,
10980                                      adapter->hw.mac.perm_addr);
10981
10982         eth_hw_addr_set(netdev, hw->mac.perm_addr);
10983
10984         if (!is_valid_ether_addr(netdev->dev_addr)) {
10985                 e_dev_err("invalid MAC address\n");
10986                 err = -EIO;
10987                 goto err_sw_init;
10988         }
10989
10990         /* Set hw->mac.addr to permanent MAC address */
10991         ether_addr_copy(hw->mac.addr, hw->mac.perm_addr);
10992         ixgbe_mac_set_default_filter(adapter);
10993
10994         timer_setup(&adapter->service_timer, ixgbe_service_timer, 0);
10995
10996         if (ixgbe_removed(hw->hw_addr)) {
10997                 err = -EIO;
10998                 goto err_sw_init;
10999         }
11000         INIT_WORK(&adapter->service_task, ixgbe_service_task);
11001         set_bit(__IXGBE_SERVICE_INITED, &adapter->state);
11002         clear_bit(__IXGBE_SERVICE_SCHED, &adapter->state);
11003
11004         err = ixgbe_init_interrupt_scheme(adapter);
11005         if (err)
11006                 goto err_sw_init;
11007
11008         for (i = 0; i < adapter->num_rx_queues; i++)
11009                 u64_stats_init(&adapter->rx_ring[i]->syncp);
11010         for (i = 0; i < adapter->num_tx_queues; i++)
11011                 u64_stats_init(&adapter->tx_ring[i]->syncp);
11012         for (i = 0; i < adapter->num_xdp_queues; i++)
11013                 u64_stats_init(&adapter->xdp_ring[i]->syncp);
11014
11015         /* WOL not supported for all devices */
11016         adapter->wol = 0;
11017         hw->eeprom.ops.read(hw, 0x2c, &adapter->eeprom_cap);
11018         hw->wol_enabled = ixgbe_wol_supported(adapter, pdev->device,
11019                                                 pdev->subsystem_device);
11020         if (hw->wol_enabled)
11021                 adapter->wol = IXGBE_WUFC_MAG;
11022
11023         device_set_wakeup_enable(&adapter->pdev->dev, adapter->wol);
11024
11025         /* save off EEPROM version number */
11026         ixgbe_set_fw_version(adapter);
11027
11028         /* pick up the PCI bus settings for reporting later */
11029         if (ixgbe_pcie_from_parent(hw))
11030                 ixgbe_get_parent_bus_info(adapter);
11031         else
11032                  hw->mac.ops.get_bus_info(hw);
11033
11034         /* calculate the expected PCIe bandwidth required for optimal
11035          * performance. Note that some older parts will never have enough
11036          * bandwidth due to being older generation PCIe parts. We clamp these
11037          * parts to ensure no warning is displayed if it can't be fixed.
11038          */
11039         switch (hw->mac.type) {
11040         case ixgbe_mac_82598EB:
11041                 expected_gts = min(ixgbe_enumerate_functions(adapter) * 10, 16);
11042                 break;
11043         default:
11044                 expected_gts = ixgbe_enumerate_functions(adapter) * 10;
11045                 break;
11046         }
11047
11048         /* don't check link if we failed to enumerate functions */
11049         if (expected_gts > 0)
11050                 ixgbe_check_minimum_link(adapter, expected_gts);
11051
11052         err = ixgbe_read_pba_string_generic(hw, part_str, sizeof(part_str));
11053         if (err)
11054                 strlcpy(part_str, "Unknown", sizeof(part_str));
11055         if (ixgbe_is_sfp(hw) && hw->phy.sfp_type != ixgbe_sfp_type_not_present)
11056                 e_dev_info("MAC: %d, PHY: %d, SFP+: %d, PBA No: %s\n",
11057                            hw->mac.type, hw->phy.type, hw->phy.sfp_type,
11058                            part_str);
11059         else
11060                 e_dev_info("MAC: %d, PHY: %d, PBA No: %s\n",
11061                            hw->mac.type, hw->phy.type, part_str);
11062
11063         e_dev_info("%pM\n", netdev->dev_addr);
11064
11065         /* reset the hardware with the new settings */
11066         err = hw->mac.ops.start_hw(hw);
11067         if (err == IXGBE_ERR_EEPROM_VERSION) {
11068                 /* We are running on a pre-production device, log a warning */
11069                 e_dev_warn("This device is a pre-production adapter/LOM. "
11070                            "Please be aware there may be issues associated "
11071                            "with your hardware.  If you are experiencing "
11072                            "problems please contact your Intel or hardware "
11073                            "representative who provided you with this "
11074                            "hardware.\n");
11075         }
11076         strcpy(netdev->name, "eth%d");
11077         pci_set_drvdata(pdev, adapter);
11078         err = register_netdev(netdev);
11079         if (err)
11080                 goto err_register;
11081
11082
11083         /* power down the optics for 82599 SFP+ fiber */
11084         if (hw->mac.ops.disable_tx_laser)
11085                 hw->mac.ops.disable_tx_laser(hw);
11086
11087         /* carrier off reporting is important to ethtool even BEFORE open */
11088         netif_carrier_off(netdev);
11089
11090 #ifdef CONFIG_IXGBE_DCA
11091         if (dca_add_requester(&pdev->dev) == 0) {
11092                 adapter->flags |= IXGBE_FLAG_DCA_ENABLED;
11093                 ixgbe_setup_dca(adapter);
11094         }
11095 #endif
11096         if (adapter->flags & IXGBE_FLAG_SRIOV_ENABLED) {
11097                 e_info(probe, "IOV is enabled with %d VFs\n", adapter->num_vfs);
11098                 for (i = 0; i < adapter->num_vfs; i++)
11099                         ixgbe_vf_configuration(pdev, (i | 0x10000000));
11100         }
11101
11102         /* firmware requires driver version to be 0xFFFFFFFF
11103          * since os does not support feature
11104          */
11105         if (hw->mac.ops.set_fw_drv_ver)
11106                 hw->mac.ops.set_fw_drv_ver(hw, 0xFF, 0xFF, 0xFF, 0xFF,
11107                                            sizeof(UTS_RELEASE) - 1,
11108                                            UTS_RELEASE);
11109
11110         /* add san mac addr to netdev */
11111         ixgbe_add_sanmac_netdev(netdev);
11112
11113         e_dev_info("%s\n", ixgbe_default_device_descr);
11114
11115 #ifdef CONFIG_IXGBE_HWMON
11116         if (ixgbe_sysfs_init(adapter))
11117                 e_err(probe, "failed to allocate sysfs resources\n");
11118 #endif /* CONFIG_IXGBE_HWMON */
11119
11120         ixgbe_dbg_adapter_init(adapter);
11121
11122         /* setup link for SFP devices with MNG FW, else wait for IXGBE_UP */
11123         if (ixgbe_mng_enabled(hw) && ixgbe_is_sfp(hw) && hw->mac.ops.setup_link)
11124                 hw->mac.ops.setup_link(hw,
11125                         IXGBE_LINK_SPEED_10GB_FULL | IXGBE_LINK_SPEED_1GB_FULL,
11126                         true);
11127
11128         err = ixgbe_mii_bus_init(hw);
11129         if (err)
11130                 goto err_netdev;
11131
11132         return 0;
11133
11134 err_netdev:
11135         unregister_netdev(netdev);
11136 err_register:
11137         ixgbe_release_hw_control(adapter);
11138         ixgbe_clear_interrupt_scheme(adapter);
11139 err_sw_init:
11140         ixgbe_disable_sriov(adapter);
11141         adapter->flags2 &= ~IXGBE_FLAG2_SEARCH_FOR_SFP;
11142         iounmap(adapter->io_addr);
11143         kfree(adapter->jump_tables[0]);
11144         kfree(adapter->mac_table);
11145         kfree(adapter->rss_key);
11146         bitmap_free(adapter->af_xdp_zc_qps);
11147 err_ioremap:
11148         disable_dev = !test_and_set_bit(__IXGBE_DISABLED, &adapter->state);
11149         free_netdev(netdev);
11150 err_alloc_etherdev:
11151         pci_disable_pcie_error_reporting(pdev);
11152         pci_release_mem_regions(pdev);
11153 err_pci_reg:
11154 err_dma:
11155         if (!adapter || disable_dev)
11156                 pci_disable_device(pdev);
11157         return err;
11158 }
11159
11160 /**
11161  * ixgbe_remove - Device Removal Routine
11162  * @pdev: PCI device information struct
11163  *
11164  * ixgbe_remove is called by the PCI subsystem to alert the driver
11165  * that it should release a PCI device.  The could be caused by a
11166  * Hot-Plug event, or because the driver is going to be removed from
11167  * memory.
11168  **/
11169 static void ixgbe_remove(struct pci_dev *pdev)
11170 {
11171         struct ixgbe_adapter *adapter = pci_get_drvdata(pdev);
11172         struct net_device *netdev;
11173         bool disable_dev;
11174         int i;
11175
11176         /* if !adapter then we already cleaned up in probe */
11177         if (!adapter)
11178                 return;
11179
11180         netdev  = adapter->netdev;
11181         ixgbe_dbg_adapter_exit(adapter);
11182
11183         set_bit(__IXGBE_REMOVING, &adapter->state);
11184         cancel_work_sync(&adapter->service_task);
11185
11186         if (adapter->mii_bus)
11187                 mdiobus_unregister(adapter->mii_bus);
11188
11189 #ifdef CONFIG_IXGBE_DCA
11190         if (adapter->flags & IXGBE_FLAG_DCA_ENABLED) {
11191                 adapter->flags &= ~IXGBE_FLAG_DCA_ENABLED;
11192                 dca_remove_requester(&pdev->dev);
11193                 IXGBE_WRITE_REG(&adapter->hw, IXGBE_DCA_CTRL,
11194                                 IXGBE_DCA_CTRL_DCA_DISABLE);
11195         }
11196
11197 #endif
11198 #ifdef CONFIG_IXGBE_HWMON
11199         ixgbe_sysfs_exit(adapter);
11200 #endif /* CONFIG_IXGBE_HWMON */
11201
11202         /* remove the added san mac */
11203         ixgbe_del_sanmac_netdev(netdev);
11204
11205 #ifdef CONFIG_PCI_IOV
11206         ixgbe_disable_sriov(adapter);
11207 #endif
11208         if (netdev->reg_state == NETREG_REGISTERED)
11209                 unregister_netdev(netdev);
11210
11211         ixgbe_stop_ipsec_offload(adapter);
11212         ixgbe_clear_interrupt_scheme(adapter);
11213
11214         ixgbe_release_hw_control(adapter);
11215
11216 #ifdef CONFIG_DCB
11217         kfree(adapter->ixgbe_ieee_pfc);
11218         kfree(adapter->ixgbe_ieee_ets);
11219
11220 #endif
11221         iounmap(adapter->io_addr);
11222         pci_release_mem_regions(pdev);
11223
11224         e_dev_info("complete\n");
11225
11226         for (i = 0; i < IXGBE_MAX_LINK_HANDLE; i++) {
11227                 if (adapter->jump_tables[i]) {
11228                         kfree(adapter->jump_tables[i]->input);
11229                         kfree(adapter->jump_tables[i]->mask);
11230                 }
11231                 kfree(adapter->jump_tables[i]);
11232         }
11233
11234         kfree(adapter->mac_table);
11235         kfree(adapter->rss_key);
11236         bitmap_free(adapter->af_xdp_zc_qps);
11237         disable_dev = !test_and_set_bit(__IXGBE_DISABLED, &adapter->state);
11238         free_netdev(netdev);
11239
11240         pci_disable_pcie_error_reporting(pdev);
11241
11242         if (disable_dev)
11243                 pci_disable_device(pdev);
11244 }
11245
11246 /**
11247  * ixgbe_io_error_detected - called when PCI error is detected
11248  * @pdev: Pointer to PCI device
11249  * @state: The current pci connection state
11250  *
11251  * This function is called after a PCI bus error affecting
11252  * this device has been detected.
11253  */
11254 static pci_ers_result_t ixgbe_io_error_detected(struct pci_dev *pdev,
11255                                                 pci_channel_state_t state)
11256 {
11257         struct ixgbe_adapter *adapter = pci_get_drvdata(pdev);
11258         struct net_device *netdev = adapter->netdev;
11259
11260 #ifdef CONFIG_PCI_IOV
11261         struct ixgbe_hw *hw = &adapter->hw;
11262         struct pci_dev *bdev, *vfdev;
11263         u32 dw0, dw1, dw2, dw3;
11264         int vf, pos;
11265         u16 req_id, pf_func;
11266
11267         if (adapter->hw.mac.type == ixgbe_mac_82598EB ||
11268             adapter->num_vfs == 0)
11269                 goto skip_bad_vf_detection;
11270
11271         bdev = pdev->bus->self;
11272         while (bdev && (pci_pcie_type(bdev) != PCI_EXP_TYPE_ROOT_PORT))
11273                 bdev = bdev->bus->self;
11274
11275         if (!bdev)
11276                 goto skip_bad_vf_detection;
11277
11278         pos = pci_find_ext_capability(bdev, PCI_EXT_CAP_ID_ERR);
11279         if (!pos)
11280                 goto skip_bad_vf_detection;
11281
11282         dw0 = ixgbe_read_pci_cfg_dword(hw, pos + PCI_ERR_HEADER_LOG);
11283         dw1 = ixgbe_read_pci_cfg_dword(hw, pos + PCI_ERR_HEADER_LOG + 4);
11284         dw2 = ixgbe_read_pci_cfg_dword(hw, pos + PCI_ERR_HEADER_LOG + 8);
11285         dw3 = ixgbe_read_pci_cfg_dword(hw, pos + PCI_ERR_HEADER_LOG + 12);
11286         if (ixgbe_removed(hw->hw_addr))
11287                 goto skip_bad_vf_detection;
11288
11289         req_id = dw1 >> 16;
11290         /* On the 82599 if bit 7 of the requestor ID is set then it's a VF */
11291         if (!(req_id & 0x0080))
11292                 goto skip_bad_vf_detection;
11293
11294         pf_func = req_id & 0x01;
11295         if ((pf_func & 1) == (pdev->devfn & 1)) {
11296                 unsigned int device_id;
11297
11298                 vf = (req_id & 0x7F) >> 1;
11299                 e_dev_err("VF %d has caused a PCIe error\n", vf);
11300                 e_dev_err("TLP: dw0: %8.8x\tdw1: %8.8x\tdw2: "
11301                                 "%8.8x\tdw3: %8.8x\n",
11302                 dw0, dw1, dw2, dw3);
11303                 switch (adapter->hw.mac.type) {
11304                 case ixgbe_mac_82599EB:
11305                         device_id = IXGBE_82599_VF_DEVICE_ID;
11306                         break;
11307                 case ixgbe_mac_X540:
11308                         device_id = IXGBE_X540_VF_DEVICE_ID;
11309                         break;
11310                 case ixgbe_mac_X550:
11311                         device_id = IXGBE_DEV_ID_X550_VF;
11312                         break;
11313                 case ixgbe_mac_X550EM_x:
11314                         device_id = IXGBE_DEV_ID_X550EM_X_VF;
11315                         break;
11316                 case ixgbe_mac_x550em_a:
11317                         device_id = IXGBE_DEV_ID_X550EM_A_VF;
11318                         break;
11319                 default:
11320                         device_id = 0;
11321                         break;
11322                 }
11323
11324                 /* Find the pci device of the offending VF */
11325                 vfdev = pci_get_device(PCI_VENDOR_ID_INTEL, device_id, NULL);
11326                 while (vfdev) {
11327                         if (vfdev->devfn == (req_id & 0xFF))
11328                                 break;
11329                         vfdev = pci_get_device(PCI_VENDOR_ID_INTEL,
11330                                                device_id, vfdev);
11331                 }
11332                 /*
11333                  * There's a slim chance the VF could have been hot plugged,
11334                  * so if it is no longer present we don't need to issue the
11335                  * VFLR.  Just clean up the AER in that case.
11336                  */
11337                 if (vfdev) {
11338                         pcie_flr(vfdev);
11339                         /* Free device reference count */
11340                         pci_dev_put(vfdev);
11341                 }
11342         }
11343
11344         /*
11345          * Even though the error may have occurred on the other port
11346          * we still need to increment the vf error reference count for
11347          * both ports because the I/O resume function will be called
11348          * for both of them.
11349          */
11350         adapter->vferr_refcount++;
11351
11352         return PCI_ERS_RESULT_RECOVERED;
11353
11354 skip_bad_vf_detection:
11355 #endif /* CONFIG_PCI_IOV */
11356         if (!test_bit(__IXGBE_SERVICE_INITED, &adapter->state))
11357                 return PCI_ERS_RESULT_DISCONNECT;
11358
11359         if (!netif_device_present(netdev))
11360                 return PCI_ERS_RESULT_DISCONNECT;
11361
11362         rtnl_lock();
11363         netif_device_detach(netdev);
11364
11365         if (netif_running(netdev))
11366                 ixgbe_close_suspend(adapter);
11367
11368         if (state == pci_channel_io_perm_failure) {
11369                 rtnl_unlock();
11370                 return PCI_ERS_RESULT_DISCONNECT;
11371         }
11372
11373         if (!test_and_set_bit(__IXGBE_DISABLED, &adapter->state))
11374                 pci_disable_device(pdev);
11375         rtnl_unlock();
11376
11377         /* Request a slot reset. */
11378         return PCI_ERS_RESULT_NEED_RESET;
11379 }
11380
11381 /**
11382  * ixgbe_io_slot_reset - called after the pci bus has been reset.
11383  * @pdev: Pointer to PCI device
11384  *
11385  * Restart the card from scratch, as if from a cold-boot.
11386  */
11387 static pci_ers_result_t ixgbe_io_slot_reset(struct pci_dev *pdev)
11388 {
11389         struct ixgbe_adapter *adapter = pci_get_drvdata(pdev);
11390         pci_ers_result_t result;
11391
11392         if (pci_enable_device_mem(pdev)) {
11393                 e_err(probe, "Cannot re-enable PCI device after reset.\n");
11394                 result = PCI_ERS_RESULT_DISCONNECT;
11395         } else {
11396                 smp_mb__before_atomic();
11397                 clear_bit(__IXGBE_DISABLED, &adapter->state);
11398                 adapter->hw.hw_addr = adapter->io_addr;
11399                 pci_set_master(pdev);
11400                 pci_restore_state(pdev);
11401                 pci_save_state(pdev);
11402
11403                 pci_wake_from_d3(pdev, false);
11404
11405                 ixgbe_reset(adapter);
11406                 IXGBE_WRITE_REG(&adapter->hw, IXGBE_WUS, ~0);
11407                 result = PCI_ERS_RESULT_RECOVERED;
11408         }
11409
11410         return result;
11411 }
11412
11413 /**
11414  * ixgbe_io_resume - called when traffic can start flowing again.
11415  * @pdev: Pointer to PCI device
11416  *
11417  * This callback is called when the error recovery driver tells us that
11418  * its OK to resume normal operation.
11419  */
11420 static void ixgbe_io_resume(struct pci_dev *pdev)
11421 {
11422         struct ixgbe_adapter *adapter = pci_get_drvdata(pdev);
11423         struct net_device *netdev = adapter->netdev;
11424
11425 #ifdef CONFIG_PCI_IOV
11426         if (adapter->vferr_refcount) {
11427                 e_info(drv, "Resuming after VF err\n");
11428                 adapter->vferr_refcount--;
11429                 return;
11430         }
11431
11432 #endif
11433         rtnl_lock();
11434         if (netif_running(netdev))
11435                 ixgbe_open(netdev);
11436
11437         netif_device_attach(netdev);
11438         rtnl_unlock();
11439 }
11440
11441 static const struct pci_error_handlers ixgbe_err_handler = {
11442         .error_detected = ixgbe_io_error_detected,
11443         .slot_reset = ixgbe_io_slot_reset,
11444         .resume = ixgbe_io_resume,
11445 };
11446
11447 static SIMPLE_DEV_PM_OPS(ixgbe_pm_ops, ixgbe_suspend, ixgbe_resume);
11448
11449 static struct pci_driver ixgbe_driver = {
11450         .name      = ixgbe_driver_name,
11451         .id_table  = ixgbe_pci_tbl,
11452         .probe     = ixgbe_probe,
11453         .remove    = ixgbe_remove,
11454         .driver.pm = &ixgbe_pm_ops,
11455         .shutdown  = ixgbe_shutdown,
11456         .sriov_configure = ixgbe_pci_sriov_configure,
11457         .err_handler = &ixgbe_err_handler
11458 };
11459
11460 /**
11461  * ixgbe_init_module - Driver Registration Routine
11462  *
11463  * ixgbe_init_module is the first routine called when the driver is
11464  * loaded. All it does is register with the PCI subsystem.
11465  **/
11466 static int __init ixgbe_init_module(void)
11467 {
11468         int ret;
11469         pr_info("%s\n", ixgbe_driver_string);
11470         pr_info("%s\n", ixgbe_copyright);
11471
11472         ixgbe_wq = create_singlethread_workqueue(ixgbe_driver_name);
11473         if (!ixgbe_wq) {
11474                 pr_err("%s: Failed to create workqueue\n", ixgbe_driver_name);
11475                 return -ENOMEM;
11476         }
11477
11478         ixgbe_dbg_init();
11479
11480         ret = pci_register_driver(&ixgbe_driver);
11481         if (ret) {
11482                 destroy_workqueue(ixgbe_wq);
11483                 ixgbe_dbg_exit();
11484                 return ret;
11485         }
11486
11487 #ifdef CONFIG_IXGBE_DCA
11488         dca_register_notify(&dca_notifier);
11489 #endif
11490
11491         return 0;
11492 }
11493
11494 module_init(ixgbe_init_module);
11495
11496 /**
11497  * ixgbe_exit_module - Driver Exit Cleanup Routine
11498  *
11499  * ixgbe_exit_module is called just before the driver is removed
11500  * from memory.
11501  **/
11502 static void __exit ixgbe_exit_module(void)
11503 {
11504 #ifdef CONFIG_IXGBE_DCA
11505         dca_unregister_notify(&dca_notifier);
11506 #endif
11507         pci_unregister_driver(&ixgbe_driver);
11508
11509         ixgbe_dbg_exit();
11510         if (ixgbe_wq) {
11511                 destroy_workqueue(ixgbe_wq);
11512                 ixgbe_wq = NULL;
11513         }
11514 }
11515
11516 #ifdef CONFIG_IXGBE_DCA
11517 static int ixgbe_notify_dca(struct notifier_block *nb, unsigned long event,
11518                             void *p)
11519 {
11520         int ret_val;
11521
11522         ret_val = driver_for_each_device(&ixgbe_driver.driver, NULL, &event,
11523                                          __ixgbe_notify_dca);
11524
11525         return ret_val ? NOTIFY_BAD : NOTIFY_DONE;
11526 }
11527
11528 #endif /* CONFIG_IXGBE_DCA */
11529
11530 module_exit(ixgbe_exit_module);
11531
11532 /* ixgbe_main.c */