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